Datasets:

Hendrik-a

/

arxiv

like 0

in semiconductor spintronic devices , the semiconductor is usually lightly doped and nondegenerate , and moderate electric fields can dominate the carrier motion . we recently derived a drift - diffusion equation for spin polarization in the semiconductors by consistently taking into account electric - field effects and nondegenerate electron statistics and identified a high - field diffusive regime which has no analogue in metals . here spin injection from a ferromagnet ( fm ) into a nonmagnetic semiconductor ( ns ) is extensively studied by applying this spin drift - diffusion equation to several typical injection structures such as fm / ns , fm / ns / fm , and fm / ns / ns structures . we find that in the high - field regime spin injection from a ferromagnet into a semiconductor is enhanced by several orders of magnitude . for injection structures with interfacial barriers , the electric field further enhances spin injection considerably . in fm / ns / fm structures high electric fields destroy the symmetry between the two magnets at low fields , where both magnets are equally important for spin injection , and spin injection becomes locally determined by the magnet from which carriers flow into the semiconductor . the field - induced spin injection enhancement should also be insensitive to the presence of a highly doped nonmagnetic semiconductor ( ns ) at the fm interface , thus fm / ns/ns structures should also manifest efficient spin injection at high fields . furthermore , high fields substantially reduce the magnetoresistance observable in a recent experiment on spin injection from magnetic semiconductors .

cond-mat0206321

semiconductor devices based on the control and manipulation of electron spin ( semiconductor spintronics ) have recently attracted considerable attention since the discovery of long spin relaxation times and large spin transport distances in semiconductors and various device structures.@xcite in order to design and fabricate high - performance spintronic devices , a comprehensive understanding of spin transport and injection properties of semiconductors and heterostructures is needed . in theoretical studies of spin transport and injection in semiconductors @xcite the spin polarization is usually assumed to obey the same diffusion equation as in metals,@xcite @xmath1 where @xmath2 is the electrochemical potential of up - spin ( down - spin ) electrons . in this diffusion equation , the electric field does not play any role , and spin polarization decays away on a length scale of @xmath3 from an injection point . this is reasonable for metals because the electric field * e * is essentially screened . for semiconductor spintronic devices , however , the semiconductor often is lightly doped and nondegenerate , and a moderate electric field can dominate the carrier motion . in fact , experiments have shown that electric fields can affect spin diffusion in semiconductors dramatically.@xcite in ref . , we examined the role of electric field on spin transport in nondegenerate semiconductors and derived a drift - diffusion equation for spin polarization , @xmath4 where @xmath5 is the deviation of up - spin ( down - spin ) electron density from its equilibrium value , @xmath6 the boltzmann constant , and @xmath7 the temperature . this equation consistently takes into account electric - field effects and nondegenerate electron statistics . we identified a high - field diffusive regime which has no analogue in metals . this regime occurs for field as small as 1 v / cm at low temperatures . two distinct spin diffusion lengths now characterize spin motion , i.e. , up - stream ( @xmath8 ) and down - stream ( @xmath9 ) spin diffusion lengths . this is a further example of the analogy @xcite between up / down - spin electron in semiconductor spin transport and majority / minority carriers in semiconductor charge transport , where the presence of electric field also results in two charge diffusion lengths.@xcite we applied this spin drift - diffusion equation to study spin injection from a ferromagnet into a semiconductor , and showed that the electric - field effects on spin injection could be described in terms of the two field - induced spin diffusion lengths in the semiconductor . in this paper , first , we derive a more general drift - diffusion equation of spin polarization valid for both doped semiconductors and metals , and demonstrate the development from eq . ( [ nondeg ] ) to eq . ( [ deg ] ) as the system changes from nondegenerate to degenerate . we establish the relation between the electrochemical potential splitting and density imbalance of up - spin and down - spin electrons in nondegenerate systems , and clarify the two spin polarizations , i.e. , spin polarization of current and spin polarization of density . we then use the spin drift - diffusion equation for nondegenerate systems , eq . ( [ nondeg ] ) , to analyze several typical one - dimensional device geometries . we find that high fields also enhance spin injection from a ferromagnet to a semiconductor in structure with a spin - selective barrier . rashba,@xcite smith and silver,@xcite fert and jaffrs,@xcite and flatt , byers , and lau in ref . have considered such a barrier in the low - field regime . the field enhancement and the interface enhancement of spin injection may reinforce each other to achieve high injection efficiencies in different structures . next , we study spin injection in sandwiched fm / ns / fm structures with and without spin - selective interfacial barriers . at low fields the two magnets are equally important to determine spin injection into the semiconductor and the spin injection efficiency is sensitive to the relative orientation of the two magnets . we find that in the high - field regime , this symmetry is broken and spin injection is locally determined by the magnet from which carriers are injected into the semiconductor . the spin injection efficiency can be enhanced by orders of magnitude by increasing the electric field for both parallel and anti - parallel orientation of the two magnets . we further consider fm / ns@xmath0/ns structures , where a highly doped nonmagnetic semiconductor ( ns@xmath0 ) is placed near the magnet interface . such a configuration is common in structures designed to overcome the schottky barrier between a magnet and a semiconductor , and is intrinsic to fm / inas , where densely occupied surface states form at the interface . we find that spin injection at the strong - field limit in such a structure is controlled by the total electric current flowing into semiconductors and insensitive to the distinction between semiconductors . thus high fields can effectively enhance spin injection in such structures as well . finally we explore electric - field effects on magnetoresistance of a magnetic semiconductor ( ms)/ns / ms structure . a magnetic semiconductor can have extremely large spin polarization at low temperatures and the magnetization can be easily adjusted by applying an external magnetic field . a large positive magnetoresistance have been observed in ms / ns / ms structures.@xcite we find that this magnetoresistance collapses in the high - field regime , suggesting a sensitive test of the electric - field effects on spin transport in semiconductors . the paper is organized as follows . in sec . ii we review the general spin drift - diffusion equation in nondegenerate and degenerate systems and analyze the field - induced spin diffusion lengths . in sec . iii we investigate spin injection in fm / ns structures with an interfacial barrier . sections iv and v contain results on spin injection in fm / ns / fm structures and in fm / ns / ns structures , respectively . section vi is devoted to the electric - field effects on magnetoresistance in ms / ns / ms structures . in sec . vii , we summarize our conclusions . in this section we derive a more general drift - diffusion equation for spin polarization valid in both degenerate and nondegenerate systems and discuss the electric - field effects on spin transport by analyzing the structure of this equation . discussion of the drift - diffusion equations of carrier motion in semiconductors involving spin - dependent processes can be tracked back decades ago . pierce _ incorporated spin relaxation in a one - dimension diffusion model of up - spin and down - spin carrier densities at zero electric field.@xcite sogawa _ et al . _ investigated spin transport in wires with a set of complete drift - diffusion equations for minority carriers , which explicitly include spin - flip and recombination processes as well as the electric field.@xcite more recently zuti _ et al . _ displayed such equations for both minority and majority carriers.@xcite usually extensive numerical calculations are required to solve the set of drift - diffusion equations together with the poisson s equation self - consistently . the spin drift - diffusion equation we derive is a _ single _ equation instead of a set of equations and can be solved analytically in several interesting geometries . the role of this spin drift - diffusion equation in spin transport is similar to that of the ambipolar drift - diffusion equation in charge transport . the system we consider here is @xmath10-doped ( @xmath11-doped systems can be analyzed similarly ) , which can be ferromagnetic or nonmagnetic . the analysis presented in this section is valid not only in doped semiconductors but also in metals . we assume that there is no space charge and the material is homogeneous . the current for up - spin and down - spin can be written as @xmath12 which consists of the drift current and the diffusion one . here @xmath13 is the up - spin ( down - spin ) electron diffusion constant and @xmath14 the up - spin ( down - spin ) conductivity . the change of up - spin ( down - spin ) conductivity from its unperturbed value @xmath15 in the presence of spin polarization , @xmath16 , is assumed to be proportional to @xmath5 , the up - spin ( down - spin ) electron density _ deviation _ from its equilibrium value @xmath17 , @xmath18 here the mobility @xmath19 is independent of field and density over the range of density variation @xmath5 . the continuity equations for up - spin and down - spin electrons in systems including spin - flip scattering process are @xmath20 where @xmath21 ( @xmath22 ) is the rate with which up - spin ( down - spin ) electrons scatter to down - spin ( up - spin ) electrons . here the recombination process is neglected because the system we consider is doped ( unipolar ) . in steady state [ @xmath23 , we have @xmath24 @xmath25 where @xmath26 , and @xmath27 is the dielectric constant of the system . for a homogeneous system without space - charge , @xmath28 should be balanced by a local change of hole concentration . in doped systems , however , spin polarization can be created without changing electrons or hole densities,@xcite and therefore , @xmath29 care is required , however , to avoid setting @xmath30 directly in eqs . ( [ upspin ] ) and ( [ downspin]).@xcite instead we multiply eq . ( [ upspin ] ) by @xmath31 and eq . ( [ downspin ] ) by @xmath32 , and substract one from the other , eliminating the terms containing @xmath33 . only then do we set @xmath34 . now we have @xmath35 where the effective mobility @xmath36 and the effective diffusion constant @xmath37 for spin polarization are @xmath38 and @xmath39 is the intrinsic spin - diffusion length , where the spin relaxation time @xmath40 is defined via @xmath41 . equations ( [ mobility ] ) and ( [ diffusion ] ) indicate that the behavior of spin transport is controlled by the _ minority _ spin species . this is analogous to the ambipolar charge transport , where minority charge carriers determine the behavior of excess charge transport.@xcite for nonmagnetic systems , @xmath42 and @xmath43 . for ferromagnetic systems , @xmath36 and @xmath37 are approximately the mobility and the diffusion constant for the lower - conductivity spin species , usually the minority spins . we will assign the down - spin label to this species , so @xmath44 and @xmath45 . thus the coefficient of the second term in eq . ( [ general ] ) can be approximated from the single - band form of the einstein relation,@xcite @xmath46 where @xmath47 is the energy measured from the bottom edge of the conduction band for the minority spin , @xmath48 is the density of states for the minority spin and @xmath49 the distribution function . a more accurate evaluation of @xmath50 when @xmath51 could not be done without knowledge of , e.g. , @xmath52 . one special exception exists , however , for nondegenerate semiconductors , where @xmath49 has the boltzmann form , @xmath53 , and @xmath54 . thus we obtain eq . ( [ nondeg ] ) to describe the transport of @xmath55 , the natural measure of the spin polarization in semiconductors . we emphasize that eq . ( [ nondeg ] ) is also valid for highly spin - polarized ( including ferromagnetic ) nondegenerate semiconductors . for degenerate systems , @xmath49 in eq . ( [ einstein ] ) should have the fermi - dirac form . in a three - dimensional ( 3d ) system , @xmath56 , we have approximately @xmath57}{f_{1/2 } [ ( \varepsilon_f-\varepsilon^b_{\downarrow})/k_b t ] } , \label{3d}\ ] ] where @xmath58 is the fermi energy , @xmath59 the bottom edge of the conduction band for the minority spin , and @xmath60^{-1}$].@xcite in a two - dimensional ( 2d ) system , @xmath48 is a constant , @xmath61[1+e^{-(\varepsilon_f-\varepsilon^b_{\downarrow})/k_b t}]}. \label{2d}\ ] ] we can define a critical field @xmath62 such that when @xmath63 , the drift term will be more important than the diffusive term in eq . ( [ general ] ) , and neglecting the electric - field effects on spin transport in this regime can not be justified . in fig . 1 , we plot @xmath64 as a function of electron density for different temperatures in 2d and 3d @xmath10-doped gaas using a typical spin diffusion length @xmath65m.@xcite we can see that for electron densities ranging from @xmath66 to @xmath67 @xmath68 the critical field @xmath64 is not beyond realistic fields under which spintronic devices operate . in particular , at low temperatures @xmath64 can be as low as 1 v / cm in lightly and moderately doped semiconductors . even for 100% spin polarized _ n_-doped znmnse , @xmath69 @xmath68 , the drift term is relevant for @xmath70 v / cm at @xmath71 k. thus the electric field should be taken into account to properly interpret phenomena involving spin transport in both magnetic and nonmagnetic semiconductors . as a function of electron density for different temperatures in 3d systems . the inset is for 2d systems . the effective electron mass @xmath72 , where @xmath73 is the free electron mass.,width=264 ] for highly degenerate systems such as metals , in which @xmath74 ev @xmath75 , from eqs . ( [ 3d ] ) and ( [ 2d ] ) , @xmath76 for a 3d system and @xmath77 for a 2d system . for a typical spin diffusion length of metal , @xmath78 nm , the field has to exceed @xmath79 v / cm for the drift term to become comparable to the diffusion term in eq . ( [ general ] ) . thus in metals under realistic fields the drift term can be neglected . using the relation between the electrochemical potential , @xmath2 , and the nonequilibrium carrier density , @xmath5 , in a highly degenerate system @xmath80,\ ] ] where @xmath81 is the up - spin ( down - spin ) density of states at the fermi energy , and * e*@xmath82 , we find that eqs . ( [ neutral ] ) and ( [ general ] ) reduce to @xmath83 which is consistent with eq . ( 2.18 ) in ref . , the spin transport equation derived for metallic systems . it is straightforward to see that eq . ( [ deg ] ) is contained in eq . ( [ e_potential ] ) . in deriving eq . ( [ e_potential ] ) we have assumed that the spin - dependent conductivity is proportional to the spin - dependent density of states at the fermi level , @xmath84 . the effects of electron - electron interaction ( alterations of the spin stiffness , spin drag , etc . ) will modify the value of @xmath50 , although for the temperatures of greatest interest and moderate density the corrections are small.@xcite equation ( [ nondeg ] ) provides a framework to understand spin transport in nondegenerate semiconductors , and eqs . ( [ general]-[2d ] ) in moderately degenerate semiconductors . we see that the electric field , unlike that in metals , plays a central role on spin transport . comparing eq . ( [ nondeg ] ) with the drift - diffusion equation of minority carriers ( electrons ) in @xmath11-doped semiconductors,@xcite @xmath85 where @xmath86 is the intrinsic electron diffusion length , we find that eq . ( [ nondeg ] ) and eq . ( [ charge ] ) have the same structure and the electric field is expected to play a similar role in both situations . for minority carrier transport it is well - known that the electric field gives rise to two distinct charge diffusion lengths and considerably modifies minority charge injection . the spin drift - diffusion equation ( [ general ] ) , together with the local charge neutrality constraint eq . ( [ neutral ] ) , dramatically alters the spin transport behavior in semiconductors from that expected from eq . ( [ deg ] ) . the general solution to eq . ( [ deg ] ) ( restricting variation to the @xmath87-direction ) is @xmath88 where @xmath89 and @xmath90 are constants . in contrast the general form of solution to eq . ( [ nondeg ] ) is @xmath91 where @xmath92 and @xmath93 are the roots of the quadratic equation , @xmath94 to understand the physical consequence of the electric field on the spin diffusion , we suppose that a continuous spin imbalance is injected at @xmath95 , @xmath96 , and the electric field is along the @xmath97 direction . the spin polarization will gradually decay in size as the distance from the point of injection increases and eventually go to zero at @xmath98 . the distribution of the spin polarization then can be described by @xmath99 @xmath100 where we define two quantities @xmath9 and @xmath8 as the down - stream and up - stream spin diffusion lengths , respectively , @xmath101^{-1 } , \label{d - length_g}\ ] ] @xmath102^{-1 } , \label{u - length_g}\ ] ] and @xmath103 . here @xmath50 in 3d and 2d systems can be evaluated via eqs . ( [ 3d ] ) and ( [ 2d ] ) , respectively . for nondegenerate semiconductors , ( [ d - length_g ] ) and ( [ u - length_g ] ) reduce to @xmath104^{-1 } , \label{d - length}\ ] ] @xmath105^{-1}.\ ] ] equation ( [ d - length ] ) was reported in ref . . in the absence of the field , the down - stream and up - stream lengths are equal to the intrinsic diffusion length @xmath3 . with increasing field the down - stream diffusion length @xmath9 increases , whereas the up - stream diffusion length @xmath8 decreases . it was also shown in ref . that spin transport distance in semiconductors can be increased by an electric field ( down - stream diffusion length ) , but the up - stream diffusion length was not discussed . a high - field regime for spin transport in semiconductors can be defined by @xmath106 , where @xmath107 . in this regime , @xmath8 and @xmath9 deviate from @xmath3 considerably and the spin diffusion behavior is qualitatively different from that in low fields . we emphasize that since @xmath3 is large in semiconductors , this regime is not beyond realistic fields where most spintronic devices operate . for a typical spin diffusion length in semiconductors , @xmath108 @xmath109m,@xcite @xmath110 v / cm at @xmath111 k and @xmath112 v / cm at @xmath113 k. the physics of the field effects on the spin diffusion becomes clearer at the strong - field limit , where @xmath114 . in this limit , the electrons move with drift velocity @xmath115 and so does the spin polarization . @xmath9 is simply the distance over which the carriers move within the spin life time @xmath40,@xcite @xmath116 for the up - stream diffusion length @xmath8 at this limit , @xmath117 which simply corresponds to a boltzmann distribution of electrons in a retarding field.@xcite in the literature of spin transport in metals , the spin polarization is usually described by the splitting of electrochemical potentials for up - spin and down - spin electrons . for nondegenerate semiconductors , the density difference between up - spin and down - spin electrons is a natural way to characterize the spin polarization . it is therefore useful to establish the connection between these two quantities . the electrochemical potentials for up - spin and down - spin electrons in a semiconductor are related to their densities via @xmath118,\ ] ] where @xmath119 is the value that the electrochemical potential would have without spin polarization , @xmath120 where @xmath121 is the electrical conductivity of the semiconductor and * j * is the total electrical current . in a doped semiconductor with a homogeneous carrier concentration , the electrochemical potential @xmath119 at postion * x * is given by @xmath122 where @xmath123 is a constant . thus the electrochemical potentials for individual spins are @xmath124 the electrochemical potential splitting , @xmath125 , and the density difference , @xmath55 , between up - spin and down - spin electrons then are related via @xmath126.\ ] ] therefore it is advantageous to use @xmath55 instead of @xmath125 to describe spin transport in semiconductors , for the spin drift - diffusion equation is linear in terms of the former , but would be nonlinear in terms of the latter . when @xmath127 , @xmath128 and we have the drift - diffusion equation for the electrochemical potential splitting , @xmath129 in this linear differential equation for @xmath125 , the electric field still plays a central role and there are two distinct diffusion lengths , i.e. , the down - stream ( @xmath9 ) and the up - stream ( @xmath8 ) diffusion lengths , for @xmath125 . thus spin transport predicted by eq . ( [ nondeg_mu ] ) would be still qualitatively different from that expected from eq . ( [ deg ] ) . there exist two definitions in literature to characterize spin polarization in nonmagnetic semiconductors . one definition uses the density difference between up - spin and down - spin electrons , @xmath130 where @xmath131 is the total electron density . the other uses the current difference between up - spin and down - spin electrons , @xmath132 generally speaking , these two spin polarization are different , although they are related . to find the relationship between these two polarizations in a homogeneous nonmagnetic semiconductor , we note that @xmath133 by using the local charge neutrality condition eq . ( [ neutral ] ) , we obtain @xmath134 for a steady spin imbalance injected at @xmath95 , as discussed in sec . ii.b , according to the general solution of eqs . ( [ d - stream ] ) and ( [ u - stream ] ) , @xmath135 and the relation between the spin polarization of current @xmath136 and the spin polarization of density @xmath137 can be written as @xmath138 \label{ap - d_g}\ ] ] for @xmath139 , and @xmath140 \label{ap - u_g}\ ] ] for @xmath141 . @xmath50 is eqs . ( [ ap - d_g ] ) and ( [ ap - u_g ] ) can be calculated using eqs . ( [ 3d ] ) and ( [ 2d ] ) for 3d and 2d systems with different doping concentrations and temperatures . in the nondegenerate limit , ( [ ap - d_g ] ) reduces to @xmath142 which is equivalent to eq . ( 5 ) in ref . , where the authors studied magnetization ( @xmath143 ) in the presence of current with a _ given _ spin polarization ( @xmath144 ) in semiconductors . in the meantime eq . ( [ ap - u_g ] ) reduces to @xmath145 thus in semiconductors @xmath136 is proportional to @xmath137 , and the ratio between them depends on the electric field and its direction . we first consider a simple one - dimensional spin injection structure to elucidate the underlying physics of electric - field enhanced spin injection . this injection structure comprises a semi - infinite degenerate ferromagnet ( @xmath146 ) and a semi - infinite nonmagnetic nondegenerate semiconductor ( @xmath139 ) . electrons are injected from the magnet into the semiconductor , and therefore , the electric field is antiparallel to the @xmath87-axis . in the ferromagnet the electrochemical potentials for individual spins satisfy eq . ( [ e_potential ] ) , which has the following general solution : @xmath147 where @xmath148 is the up - spin ( down - spin ) electrical conductivity of the ferromagnet , and @xmath149 is the total electron current , which is a constant throughout the structure in steady state . we use @xmath150 and @xmath151 to denote the intrinsic spin diffusion length in the ferromagnet and in the semiconductor , respectively . in the semiconductor , up - spin and down - spin electron densties satisfy the spin drift - diffusion equation for nondegenerate systems , eq . ( [ nondeg ] ) , as well as the local charge neutrality condition , eq . ( [ neutral ] ) . the general solution can be written as @xmath152 and accordingly the electrochemical potentials for individual spins are @xmath153+e ex -b.\ ] ] here @xmath154 is total electron density . in general , a schottky barrier will form between the magnet and the semiconductor when the two materials are placed together . since the charge neutrality condition would be severely violated in the depletion region of a schottky barrier , a wide depletion region is undesirable for spin transport and spin coherence . specifically the presence of holes dramatically shortens the electron spin coherence time . we consider instead structures with a very thin interfacial barrier between the magnet and the semiconductor . a spin - selective interfacial barrier was examined in refs . as a way to circumvent the resistance mismatch obstacle for spin injection from a ferromagnetic metal into a semiconductor . if there is no spin - flip scattering at the interface , the current for an individual spin is continuous across the interface and is related to the spin - dependent electrochemical potential change across the interface via ohmic s law , giving rise to the following boundary conditions : @xmath155,\\ j_{\downarrow}(0 ^ -)&=&g_{\downarrow}[\mu_{\downarrow}(0^+ ) -\mu_{\downarrow}(0 ^ -)],\\ j_{\uparrow}(0 ^ -)&-&j_{\downarrow}(0 ^ - ) = j_{\uparrow}(0^+)-j_{\downarrow}(0^+),\end{aligned}\ ] ] where @xmath156 is the interfacial conductance for up - spin ( down - spin ) electrons and the current of individual spins @xmath157 can be calculated via @xmath158 . these three equations completely determine the three unknown coefficients @xmath159 , @xmath123 , @xmath160 in eqs . ( 3.1)-(3.3 ) . the solution of @xmath5 in eq . ( [ solution_n ] ) and the relation of eq . ( [ ap - d ] ) indicate that in the semiconductor @xmath161 , where @xmath162 is the spin injection efficiency at the interface . we obtain an equation for @xmath162 , @xmath163 = \frac{k_b t}{ee \sigma_s } \ln \frac{-k_bt / e e l_u + \alpha_0}{-k_bt / e e l_u -\alpha_0},\ ] ] where @xmath164 is the conductivity of the ferromagnet , @xmath165 the conductivity of the semiconductor , and @xmath166 the spin polarization in the ferromagnet . we solve this equation and plot the spin injection of current @xmath162 as a function of the electric field in fig . we see that the electric field can substantially enhance the spin injection efficiency in fm / ns structures . we note that spin injection enhancement from a spin - selective interfacial barrier between the ferromagnet and the semiconductor , which has been identified in the low - field regime,@xcite becomes more pronounced in the high - field regime . as a function of electric field . the interfacial conductance for up - spin electrons is twice larger than that for down - spin electrons , @xmath167 . solid , dashed , and dot - dashed lines correspond to @xmath168 ( transparent interface ) , 10@xmath169 , and 10@xmath170 @xmath171 , respectively . other parameters are @xmath172 , @xmath173 nm , @xmath174 @xmath109 m , and @xmath175 ( @xmath176 cm)@xmath177.,width=264 ] in the small spin polarization limit , @xmath178 , @xmath136 can be expressed in an explicit form , @xmath179^{-1}\nonumber\\ & \times&\bigg[\frac{p_f l^{(f)}}{(1-p^2_f)\sigma_f}+\frac{g_{\uparrow}-g_{\downarrow}}{4g_{\uparrow}g_{\downarrow}}\bigg ] e^{-x / l_d}. \label{ex_fn}\end{aligned}\ ] ] this expression clearly shows that the electric field and the spin - selective interfacial resistance both enhance spin injection , but in different ways . the electric - field effects on spin injection can be described in terms of the two field - induced diffusion lengths . both diffusion lengths affect spin injection favorably : the up - stream length @xmath8 controls the relevant resistance in the semiconductor , which determines the spin injection efficiency . with increasing field this effective resistance , @xmath180 , becomes smaller , and accordingly the spin injection efficiency is enhanced . the transport distance of the injected spin polarization in the semiconductor , however , is controlled by the down - stream length @xmath9 . as the field increases , this distance becomes longer . on the other hand , the spin - selective interfacial barrier provides another spin polarization source besides the spin - aligned electrons in the ferromagnet and acts as a spin filter which permits electrons with a particular spin to pass through the interface , and therefore enhances spin injection . moreover , the spin - selective barrier has no effect on the transport distance of the injected spin polarization in the semiconductor . we now contrast eq . ( [ ex_fn ] ) with that obtained by previous calculations @xcite based on eq . ( [ deg ] ) . the spin injection @xmath181^{-1}\nonumber\\ & \times&\big[\frac{p_f l^{(f)}}{(1-p^2_f)\sigma_f}+\frac{g_{\uparrow}-g_{\downarrow}}{4g_{\uparrow}g_{\downarrow}}\big]e^{-x / l^{(s ) } } \label{zero_e}\end{aligned}\ ] ] is given by the zero - field result of eq . ( [ ex_fn ] ) . for a transparent interface with @xmath182 , the effective resistance in the magnet , @xmath183 , is much less than its counterpart in the semiconductor , @xmath184 ( as @xmath185 and @xmath186 ) . ( [ zero_e ] ) suggests that this resistance mismatch makes it virtually impossible to realize an appreciable spin injection from a ferromagnetic metal to a semiconductor without spin - selective interfacial barrier . however , the more general description of the spin transport in semiconductors indicates that the effective semiconductor resistance to be compared with @xmath183 should be @xmath180 rather than @xmath184 . since @xmath8 can be smaller than @xmath151 by orders of magnitude in the high - field regime , this `` conductivity mismatch '' obstacle may be overcome with the help of strong electric fields , or equivalently , large injection currents . we note that although it has been realized that spin injection can be enhanced by increasing the total injection current,@xcite the treatment there used eq . ( [ deg ] ) to describe spin transport in nondegenerate semiconductors , where the electric - field effects were not taken into account . thus the physics of the field - dependent spin transport was not captured and the treatment was incomplete . in the presence of the interfacial barrier , the relative importance of the two mechanisms for the spin injection efficiency enhancement , electric field and spin - selective interfacial barrier , depends on the relative magnitude of @xmath187 , @xmath188 , and @xmath189 , the resistances of the ferromagnet , the semiconductor , and the interface , which may vary from system to system . for systems with @xmath190 , the spin - selective interfacial barrier dominates in enhancing the spin injection efficiency . on the other hand , for systems with @xmath191 or @xmath192 , the electric field can enhance spin injection efficiently . the transport distance of injected spin polarization in the semiconductor , however , can only be enhanced by the electric field , even in systems with @xmath193 . furthermore , the two mechanisms have different temperature and field dependences . the field - enhanced spin injection efficiency increases with the electric field ( current ) and decreases with the temperature . the interface - enhanced spin injection likely depends on neither the field nor the temperature . to quantitatively demonstrate the electric - field effects on spin injection , we choose the realistic parameters of a spin injection device as follows : @xmath172 and @xmath173 nm ( as in co),@xcite @xmath174 @xmath109 m ( as in gaas).@xcite first we examine the structures without an interfacial barrier . for a ferromagmetic metal / semiconductor structure , e.g. , co / gaas , @xmath194 , the spin injection efficiency increases from @xmath195 at zero field to 2% at @xmath196 nm@xmath177 . for a ferromagnetic semiconductor with @xmath197 and @xmath198 , the spin injection efficiency increases from 0.02% at zero field to 2% at @xmath199 nm@xmath177 , which corresponds to @xmath200 v / cm , or @xmath201 a/@xmath202 for a typical semiconductor conductivity @xmath203 ( @xmath176 cm)@xmath177 , at @xmath113 k. suppose there existed a spin - selective interfacial barrier between the magnet and the semiconductor with @xmath204 ( @xmath176 @xmath202)@xmath177 . for the co / gaas structure , the spin injection efficiency would be 0.1% at zero field ( much greater than @xmath195 in a similar injection structure without a spin - selective interfacial barrier ) , which can be further enhanced to 10% at @xmath205 nm@xmath177 , or @xmath206 kv / cm at @xmath111 k. for the spin injection structure from the ferromagnetic semiconductor , the spin injection efficiency is 0.1% at zero field ( also considerably greater than 0.02% in a similar structure without a spin - selective interfacial barrier ) , which can be further increased to 10% at @xmath199 nm@xmath177 . thus the combination of electric field and spin - selective interfacial barrier may help explain the large spin injection percentages from znmnse to znse,@xcite from gamnas to gaas,@xcite from fe to gaas,@xcite as well as the dramatic increase in spin injection with current in ref . . figure 3 shows spin polarization of current ( @xmath162 ) and spin polarization of density ( @xmath207 ) at the interface as a function of electric field . according to eq . ( [ ap - d ] ) , @xmath208 . we can see that in the low - field regime , the density polarization in the semiconductor is much smaller than the current polarization ; whereas in the high - field regime , the two polarizations become equal . thus the difference between high - field injection and low - field injection is that a strong electric field in high - field injection results in a macroscopic density difference between up - spin and down - spin electrons ; whereas in low - field injection the carrier densities of up - spin and down spin electrons remain the same . and spin polarization of density @xmath207 at the interface , respectively . the lower curves are for transparent interface structure and the upper ones are for structure with a spin selective interfacial barrier , @xmath209 @xmath171 . other parameters are the same as in fig . 2 . , width=264 ] another important quantity is the boundary resistance @xmath210 , which can be used to analyze how easy a current can convert its spin while flowing from the magnet into the semiconductor , @xmath211 which can be expressed in terms of the spin injection efficiency and the resistances of the magnet , the semiconductor , and the interface . @xmath212\nonumber\\ & + & \frac{g_{\uparrow}+g_{\downarrow}}{4g_{\uparrow}g_{\downarrow } } -p_f\frac{g_{\uparrow}-g_{\downarrow}}{4g_{\uparrow}g_{\downarrow}}.\end{aligned}\ ] ] we plot @xmath210 in fig . 4 as a function of electric field . with increasing field , the boundary resistance decreases , indicating that the field helps current conversion from a higher polarization ( in the magnet ) to a lower polarization ( in the semiconductor ) . for structures with a transparent interface , in the small polarization limit [ @xmath213 , @xmath214^{-1 } p^2_f\frac{l^{(f)}}{\sigma_f}\frac{l_u}{\sigma_s},\ ] ] which clearly shows that the boundary resistance is determined by the up - stream diffusion length @xmath8 rather than @xmath151 and should be a function of electric field . as a function of electric field . solid line is for structure with transparent interface . dashed line is for structure with a spin - selective interfacial barrier , @xmath209 @xmath171 . other parameters are the same as in fig . 2 . , width=264 ] in this section we consider a sandwiched structure that comprises a semi - infinite ferromagnet ( @xmath146 ) , a nonmagnetic semiconductor with width @xmath215 , and a semi - infinite ferromagnet ( @xmath216 ) . @xmath215 is assumed to be much shorter than the semiconductor intrinsic spin diffusion length @xmath151 , as in most spintronic devices . the two ferromagnets are otherwise identical with possible different orientations of magnetization ( parallel and anti - parallel ) . electrons are injected from the left magnet to the semiconductor and the electric field is antiparallel to the @xmath87-axis . the spin transport in the magnets is described by eq . ( [ e_potential ] ) with the following general solutions : @xmath217 in the left magnet ( @xmath141 ) , and @xmath218 in the right magnet ( @xmath216 ) . here we use superscript or subscript @xmath3 ( @xmath219 ) to label the quantities in the left ( right ) ferromagnet . the spin - dependent electron densities in the semiconductor satisfy eqs . ( [ nondeg ] ) and ( [ neutral ] ) , and the general solutions at @xmath220 can be written as @xmath221,\ ] ] and therefore the electrochemical potentials for individual spins are @xmath222\nonumber\\ & + & e ex -b_0.\end{aligned}\ ] ] the six unknowns in the above solutions , @xmath223 , @xmath89 , @xmath224 , @xmath225 , @xmath226 , and @xmath227 , are determined by the following six independent boundary conditions : @xmath228,\\ j_{\downarrow}(0 ^ -)&=&g_{\downarrow}[\mu_{\downarrow}(0^+)-\mu_{\downarrow}(0 ^ -)],\\ j_{\uparrow}(0 ^ -)&-&j_{\downarrow}(0 ^ - ) = j_{\uparrow}(0^+)-j_{\downarrow}(0^+),\\ j_{\uparrow}(x_0^+)&=&{\tilde g}_{\uparrow}[\mu_{\uparrow}(x_0^+)-\mu_{\downarrow}(x_0 ^ -)],\\ j_{\downarrow}(x_0^+)&=&{\tilde g}_{\downarrow}[\mu_{\downarrow}(x_0^+)-\mu_{\downarrow}(x_0 ^ -)],\\ j_{\uparrow}(x_0 ^ -)&-&j_{\downarrow}(x_0 ^ - ) = j_{\uparrow}(x_0^+)-j_{\downarrow}(x_0^+),\end{aligned}\ ] ] where @xmath156 and @xmath229 are the spin - dependent conductances at the left and the right interfaces , respectively . we find the two equations for the spin polarizations of current at interfaces , @xmath230 and @xmath231 , by matching the above boundary conditions , eqs . ( 4.5a - f ) , @xmath232\bigg[\frac{g_{\uparrow}+g_{\downarrow}}{2g_{\uparrow}g_{\downarrow}}+ \frac{2l^{(f)}}{(1-p^2_l)\sigma_f}\bigg ] = \frac{k_b t}{ee \sigma_s } \ln \frac{1+z_1}{1-z_1}}\\ { \displaystyle \frac{p_r({\tilde g}_{\uparrow}+{\tilde g}_{\downarrow})}{2{\tilde g}_{\uparrow}{\tilde g}_{\downarrow } } -\frac{{\tilde g}_{\uparrow}-{\tilde g}_{\downarrow}}{2{\tilde g}_{\uparrow}{\tilde g}_{\downarrow } } + \big[\alpha(x_0)-p_r\big]\bigg[\frac{{\tilde g}_{\uparrow}+{\tilde g}_{\downarrow } } { 2{\tilde g}_{\uparrow}{\tilde g}_{\downarrow}}+ \frac{2l^{(f)}}{(1-p^2_r)\sigma_f}\bigg ] = \frac{k_b t}{ee \sigma_s } \ln \frac{1-z_2}{1+z_2 } } \end{array}\right.,\ ] ] where @xmath233 and @xmath234 are the spin polarization in the left and right magnets , respectively , and @xmath235 when @xmath230 and @xmath231 are known , we can express the spin polarization of current in the semiconductor ( @xmath220 ) as @xmath236 and the spin polarization of density in the semiconductor ( @xmath220 ) as @xmath237 figure 5 depicts the spin injection efficiency at the left interface ( @xmath162 ) as a function of electric field for the parallel configuration @xmath238 and the anti - parallel configuration @xmath239 . for structures with transparent interfaces ( @xmath240 ) , we see that the spin injection in both configurations can be enhanced by orders of magnitude by increasing the field . in the low - field regime , spin injection can only be achieved in structures with the parallel configuration . in the high - field regime , spin injection from the left magnet into the semiconductor for both configurations are the same , indicating that only the magnet from which carriers are injected is important to the spin injection efficiency . for structures with transparent interfaces , spin polarization in the semiconductor @xmath136 can be expressed in compact forms in both the low- and high - field regimes . in fact , in the low - field regime , where @xmath241 , we reproduce eq . ( 7 ) in ref . , @xmath242^{-1 } \frac{(p_l+p_r)l^{(f ) } } { ( 1-p^2_f)\sigma_f } , \label{zero_e3}\ ] ] for @xmath220 . the above expression indicates that the spin injection strongly depends on the relative orientation of the two ferromagnetic metals : for the parallel configuration , @xmath136 is finite ; whereas for the anti - parallel configuration , spin injection is not possible . in this limit , the left magnet and the right one are equally important in determining spin injection . the magnitude of spin injection @xmath144 is determined by the ratio of the effective resistances in the magnet ( @xmath183 ) and in the semiconductor ( @xmath243 ) . for typical device parameters @xmath244 , and the spin injection efficiency would be extremely small due to this resistance mismatch . the spin polarization of density in this limit can be written as @xmath245 . at the zero - field limit , the density difference between up - spin and down - spin electrons vanishes . on the other hand , in the high - field regime , @xmath8 and @xmath9 can differ by orders of magnitude , and usually @xmath246 . we find that in this limit , for @xmath247 , @xmath248^{-1 } \frac{p_l l^{(f)}}{(1-p^2_l)\sigma_f } , \label{ex_fnf}\ ] ] and @xmath249 , which is close to @xmath144 in the strong - field limit . we see that in the high - field regime , spin injection is locally determined by the magnet from which carriers are injected , and the `` remote '' magnet that collects the current becomes irrelevant in determining the spin injection efficiency . moreover , the effective resistance of the semiconductor to be compared with that of the magnet ( @xmath183 ) is @xmath180 rather than @xmath243 . in this regime , the spin polarization of current and the spin polarization of density have similar amplitude . thus in the high - field regime , according to eq . ( [ ex_fnf ] ) , the spin injection behavior in the sandwiched fm / ns / fm structure would be the same as in a simpler fm / ns structure . the reason that the second magnet becomes unimportant for spin injection in the high - field regime is that the influence of the second magnet on spin transport in the semiconductor is localized within the up - stream length ( @xmath8 ) from the magnet . @xmath8 decreases with increasing field and would be much shorter than @xmath215 in the high - field regime . in fig . 5 , we also plot spin injection as a function of the electric field in structure with two identical spin - selective interfacial barriers between the magnet and the semiconductor . in the calculations it is assumed that @xmath250 for the parallel configuration , and @xmath251 for the anti - parallel configuration . we see that the electric field substantially enhances spin injection in structures with spin - selective interfacial barrier . in the high - field regime , as in the transparent case , spin injection from the left magnet into the semiconductor does not depend on the orientation of the right magnet . as a function of electric field . solid and dashed lines correspond to parallel configuration ( @xmath252 ) and anti - parallel configuration ( @xmath239 ) , respectively . the lower curves are for structure with transparent interface . the upper curves are for structure with a spin - selective interfacial barrier , @xmath209 @xmath171 . other parameters are @xmath172 , @xmath173 nm , @xmath253 m , @xmath254 m , and @xmath255 cm)@xmath177.,width=264 ] in semiconductor spin injection structures , a highly doped nonmagnetic semiconductor ( ns@xmath0 ) is often placed near the magnet interface to overcome the schottky barrier between a magnet and a semiconductor . this configuration is also intrinsic to fm / inas , where densely occupied surface states form at the interface . here we consider an injection structure that comprises a semi - infinite magnet ( @xmath146 ) , a finite nonmagnetic semiconductor with conductivity @xmath121 and carrier concentration @xmath154 ( @xmath220 ) , and a semi - infinite nonmagnetic semiconductor with conductivity @xmath256 and carrier concentration @xmath257 ( @xmath216 ) . the electrochemical potentials in the magnet satisfy eq . ( [ e_potential ] ) , and the electron densities in semiconductors satisfy eqs . ( [ nondeg ] ) and ( [ neutral ] ) . the general solution of the electrochemical potentials in the magnet and the semiconductors can be written as @xmath258 for @xmath141 , @xmath259\nonumber\\ & + & \frac{e jx}{\sigma_s } -b_0\end{aligned}\ ] ] for @xmath220 , and @xmath260+\frac{e jx}{\tilde{\sigma}_s } -b_1\ ] ] for @xmath216 . here we consider structures without interface resistance between two semiconductor regions , and the boundary conditions are given by eq . ( 4.5a - f ) with @xmath261 . these equations completely determine the six unknowns , @xmath262 ( @xmath263 ) , @xmath264 ( @xmath265 ) , and @xmath160 , in expressions of eqs . ( 5.1)-(5.3 ) . we obatin an equation for @xmath266 , the spin polarization of current at the interface between the two semiconductors , @xmath267 \bigg[\big(\frac{\tilde{l}_u\sigma_s}{l_u\tilde{\sigma}_s}a-\frac{\tilde{l}_u\sigma_s}{l_d\tilde{\sigma}_s}b\big)z - p_f\bigg],\end{aligned}\ ] ] where @xmath268 , and @xmath269^{-1 } \big(\frac{\tilde{\sigma}_s}{\tilde{l}_d}+\frac{\sigma_s}{l_u}\big)e^{x_0/l_d},\\ b&=&\bigg[\sigma_s\big(\frac{1}{l_d}+\frac{1}{l_u}\big)\bigg]^{-1 } \big(\frac{-\tilde{\sigma}_s}{\tilde{l}_d}+\frac{\sigma_s}{l_d}\big)e^{-x_0/l_u}.\end{aligned}\ ] ] figure 6 illustrates the spin injection efficiency @xmath231 as a function of the total electric current @xmath149 . we see that in the low - field regime , the conductivities of both semiconductors are important to determine spin injection . as the total effective resistance ( @xmath270 ) of semiconductors decreases , the spin injection efficiency increases . with increase of the current or the field , spin injection can be enhanced considerably . moreover , in the strong - field limit , the spin injection efficiency will be determined by the total current flowing into the semiconductors . in fact , at the zero - field limit , where @xmath271 , @xmath272 , and @xmath273 , @xmath231 can be written as a simpler form , @xmath274^{-1 } \bigg[\frac{l^{(f)}p_f}{(1-p^2_f)\sigma_f}+\frac{g_{\uparrow}-g_{\downarrow } } { 4g_{\uparrow}g_{\downarrow}}\bigg],\ ] ] which shows that the ratio between the total effective resistance of the two semiconductors , @xmath270 , and that of the magnet , @xmath183 , together with the spin - selective interfacial barrier between the ferromagnet and the semiconductor , determines the spin injection efficiency . on the other hand , at the strong - field limit , where @xmath275 , @xmath276 and @xmath246 , @xmath231 can be expressed in a even more compact form , @xmath277^{-1}\nonumber\\ & \times&\bigg[\frac{l^{(f)}p_f}{(1-p^2_f)\sigma_f}+\frac{g_{\uparrow}-g_{\downarrow } } { 4g_{\uparrow}g_{\downarrow}}\bigg],\end{aligned}\ ] ] which indicates that in fm / ns / ns structures spin injection is controlled by the total current flowing into the semiconductors and a distinction between the two semiconductors becomes unimportant . , ( 1,10 ) , and ( 10,10 ) @xmath278 cm)@xmath177 . the lower curves are for structure with transparent interface . the upper curves are for structure with a spin - selective interfacial barrier , @xmath209 @xmath171 . other parameters are @xmath172 , @xmath173 nm , @xmath279 m , and @xmath280 cm)@xmath177.,width=264 ] in ms / ns / ms structures a strong positive magnetoresistance effect has been observed at low temperatures.@xcite as the applied magnetic field is changed from 0 to @xmath281 the spin polarization in a magnetic semiconductor can change at low temperatures from 0 to @xmath282% . electric - field dependence of the magnetoresistance can be expected based on the different spin injection behaviors in the high - field regime and in the low - field regime discussed in sec . iv . in the low - field regime , the densities of up - spin and down - spin electrons in a nonmagnetic semiconductor remain the same even in the presence of a fully spin - polarized current . thus in the semiconductor only _ half _ of electrons contribute to the conductance if the current is 100% spin - polarized , and the resistance of the semiconductor should be twice of that for a unpolarized current . hence the semiconductor resistance strongly depends on the spin polarization of the current in the low - field regime . in the high - field regime , however , the spin polarization of density @xmath143 is close to the spin polarization of current @xmath144 . therefore if the current is 100% spin - polarized , the electron density would be also fully spin - polarized , and _ all _ electrons would contribute to the conductance . hence in the high - field regime , the semiconductor resistance is only weakly dependent on the spin polarization , and the magnetoresistance vanishes . here we calculate how the magnetoresistance depends on the electric field . the magnetic semiconductor we consider here is degenerate and its spin transport is described by eq . ( [ e_potential ] ) . thus all results obtained in sec . iv are also applicable to the ms / ns / ms structures . we also assume that the interfaces between the two materials are transparent , i.e. , zero interface resistance . the magnetizations of the two magentic semiconductors are identical and parallel , @xmath283 , which are zero in the absence of external magnetic field and finite for a given external magnetic field * h*. the resistance of the nonmagnetic semiconductor , @xmath219 , would depend on the spin polarization in the magnetic semiconductors because of spin accumulation at the heterostructure interfaces , and is therefore also a function of the external magnetic field . the resistance of the nonmagnetic semiconductor can be calculated via @xmath284 which certainly would be @xmath285 when attached to unpolarized magnetic semiconductors ( zero magnetic field ) . the resistance in the presence of spin - polarized magnetic semiconductors with @xmath273 can be expressed as @xmath286\sigma_m}\big[2p({\bf h})-\alpha(0)-\alpha(x_0)\big].\ ] ] here @xmath230 and @xmath231 are the spin polarization of current at the left and right interfaces , @xmath287 the spin diffusion length in the magnetic semiconductor , and @xmath288 the conductivity of the magnetic semiconductor . in the low - field regime , according to eq . ( [ zero_e3 ] ) , we find that the magnetoresistance @xmath289\sigma_m x_0}{2\sigma_s l^{(m)}}\bigg)^{-1}.\ ] ] this can be significant for an ms / ns / ms structure because of the large @xmath290 [ @xmath291 and small conductivity ( @xmath292 ) in magnetic semiconductors . we would like to point out , however , that for ferromagnetic metal / semiconductor / ferromagnetic metal structures , the magnetoresistance is usually too weak to detect due to the conductivity mismatch between metals and semiconductors ( @xmath293 ) . in the high - field regime , by using eq . ( [ ex_fnf ] ) we express the magnetoresistance as @xmath294\sigma_m } { l^{(m)}\sigma_s}\bigg)^{-1}.\ ] ] we see from the above expression that the effect of the electric field on the magnetoresistance can be described in terms of the field - induced up - stream spin diffusion length . increasing the electric field will decrease the magnetoresistance because @xmath8 decreases with the electric field . figure 7 illustrates the magnetoresistance as a function of electric field for an ms / nm / ms structure . we see that with increasing electric field the magnetoresistance diminishes . for example , if the parameters are chosen as follows : @xmath253 m , @xmath295 nm , @xmath254 m , @xmath296 , and @xmath297 , the magnetoresistance @xmath298 decreases from 83% at zero field to 5.3% at @xmath299 nm@xmath177 , or @xmath300 v / cm for @xmath301 cm)@xmath177 at @xmath113 k. as a function of electric field . solid , dashed , and dot - dashed lines correspond to @xmath297 , 0.9 , and 0.8 , respectively . other parameters are @xmath253 m , @xmath295 nm , @xmath254 m , and @xmath296.,width=264 ] in summary , we have derived a general drift - diffusion equation for spin polarization in both degenerate and nondegenerate systems by consistently taking into account electric - field effects . we have demonstrated that as a system changes from degenerate to nondegenerate , the electric field becomes more and more important in spin transport . we have identified a high - field diffusive regime in nondegenerate semiconductors which has no analogue in metals . in this regime , there are two distinct spin diffusion lengths , i.e. , up - stream and down - stream spin diffusion lengths . we have applied this more general drift - diffusion equation for spin polarization to several typical injection structures encountered in semiconductor spintronic devices to study the spin injection behavior in these structures and the effects of electric fields . the high - field description of the spin transport in semiconductors predicts that the electric field can effectively enhance spin injection from a ferromagnet into a semiconductor . for structures with a spin - selective interfacial barrier we find that the electric field further enhances spin injection substantially . the combination of the field enhancement and the interface enhancement of spin injections may help us to obtain a comprehensive understanding of observed large spin injection in a variety of fm / ns , fm / ns / fm , and fm / ns / ns structures and current - dependent spin injection . the consequences of spin injection into a semiconductor in the high - field regime are qualitatively different from those in the low - field regime . a high - field injection creates a notable density difference between up - spin and down - spin electrons and the spin polarization of density has a similar value as the spin polarization of current . in contrast a low - field injection does not create an appreciable density difference between up - spin and down - spin electrons , and the spin polarization of current is quite different from the spin polarization of density . one consequence of these different spin injection behaviors is that for sandwiched fm / ns / fm structures the high field destroys the symmetry between the two magnets at low fields , where both magnets are equally important to determine spin injection . the efficiency of spin injection into semiconductors in the high - field regime is `` locally '' determined by the magnet from which carriers are injected into the semiconductor and the magnet that collects the carriers becomes irrelevant . another consequence is that for fm / ns / ns structures spin injection efficiency in the high - field regime is only determined by the total injected electric current and the distinction between the semiconductors becomes unimportant . we have also examined the electric - field effect on magnetoresistance in ms / ns / ms structures . in the low - field regime , the magnetoresistance in an ms / ns / ms structure can be significant , as reported in ref . . with increasing electric field , the magnetoresistance diminishes quickly . the underlying physics is that in the high - field regime the spin polarization of density is similar to the spin polarization of current in the nonmagnetic semiconductor and _ all _ electrons contribute to the conductance in the presence of spin - polarized current in contrast to _ half _ of electrons in the low - field regime if the current is fully polarized . thus in the high - field regime , the resistance of the nonmagnetic semiconductor only weakly depends on the spin polarization of current and the magnetic field , giving rise to a diminishing magnetoresistance effect in the high - field regime . our calculations in this paper present a broad picture of electric field - dependent spin transport and spin injection phenomena into semiconductors . our theory also provides physical insight into the field - induced enhancement of spin injection as well as the electric field - dependent magnetoresistance and suggests high - field injection as a simple way to amplify spin injection into semiconductor spintronic devices . s. a. wolf , d. d. awschalom , r. a. buhrman , j. m. daughton , s. von molnr , m. l. roukes , a. y. chtchelkanova , and d. m. treger , science , * 294 * , 1488 ( 2001 ) , and references therein . , edited by d. d. awschalom , n. samarth , and d. loss ( springer verlag , berlin , 2002 ) g. schmidt , d. ferrand , l. w. molenkamp , a. t. filip , and b. j. van wees , phys . rev . b * 62 * , r4790 ( 2000 ) . e. i. rashba , phys . b * 62 * , r16267 ( 2000 ) . d. l. smith and r. n. silver , phys . b * 64 * , 045323 ( 2001 ) . a. fert and h. jaffrs , phys . b * 64 * , 184420 ( 2002 ) . p. c. van son , h. van kempen , and p. wyder , phys . * 58 * , 2271 ( 1987 ) . j. m. kikkawa and d. d. awschalom , nature ( london ) * 397 * , 139 ( 1999 ) . i. malajovich , j. j. berry , n. samarth , and d. d. awschalom , nature ( london ) * 411 * , 770 ( 2001 ) . z. g. yu and m. e. flatt , cond - mat/0201425 . m. e. flatt and j. m. byers , phys . * 84 * , 4220 ( 2000 ) . m. e. flatt and g. vignale , appl . . lett . * 78 * , 1273 ( 2001 ) . see , e.g. , r. a. smith , _ semiconductors _ ( cambridge university press , new york , 1978 ) . g. schmidt , g. richter , p. grabs , c. gould , d. ferrand , and l. w. molenkamp , phys . lett . * 87 * , 227203 ( 2001 ) . d. t. pierce , r. j. celotta , g .- c . wang , w. n. unertl , a. galejs , c. e. kuyatt , and s. r. mielczarek , rev . instrum . * 51 * , 478 ( 1980 ) . t. sogawa , h. ando , and s. ando , phys . b * 61 * , 5535 ( 2000 ) . i. zuti , j. fabian , s. das sarma , phys . b * 64 * , 121201 ( 2001 ) ; phys . * 88 * , 066603 ( 2002 ) . j. m. kikkawa and d. d. awschalom , phys . lett . * 80 * , 4313 ( 1998 ) ; j. m. kikkawa , i. p. smorchkova , n. samarth , and d. d. awschalom , science * 277 * , 1284 ( 1997 ) . s. hershfield and h. l. zhao , phys b * 56 * , 3296 ( 1997 ) . i. damico and g. vignale , europhys . lett . * 55 * , 566 ( 2001 ) . a. g. aronov and g. e. pikus , fiz . . poluprovodn . * 10 * , 1177 ( 1976 ) [ sov . . semicond . * 10 * , 698 ( 1976 ) ] . a. barthlmy , a. fert , and f. petroff , in _ handbook of magnetic materials _ , edited by k. h. j. buschow ( elsevier science , amsterdam , 1999 ) , vol . r. flederling , m. keim , g. reuscher , w. ossau , g. schmidt , a. waag , and l. w. molenkamp , nature ( london ) * 402 * , 787 ( 1999 ) . b. t. jonker , y. d. park , b. r. bennett , h. d. cheong , g. kioseoglou , and a. petrou , phys . rev . b * 62 * , 8180 ( 2000 ) . y. ohno , d. k. young , b. beschoten , f. matsukura , h. ohno , and d. d. awschalom , nature ( london ) * 402 * , 790 ( 1999 ) . h. j. zhu , m. ramsteiner , h. kostial , m. wassermeier , h. -p . schnherr , and k. h. ploog , phys . lett . * 87 * , 016601 ( 2001 ) . a. t. hanbicki , b. t. jonker , g. itskos , g. kioseoglou , and a. petrou , appl . . lett . * 60 * , 1240 ( 2002 ) . v. f. motsnyi , v. i. safarov , j. de boeck , j. das , w. van roy , e. goovaerts , and g. borghs , cond - mat/0110240 .

surface waves on a stationary flow of water are considered , in a linear model that includes the surface tension of the fluid . the resulting gravity - capillary waves experience a rich array of horizon effects when propagating against the flow . in some cases three horizons ( points where the group velocity of the wave reverses ) exist for waves with a single laboratory frequency . some of these effects are familiar in fluid mechanics under the name of wave blocking , but other aspects , in particular waves with negative co - moving frequency and the hawking effect , were overlooked until surface waves were investigated as examples of analogue gravity [ schtzhold r and unruh w g 2002 _ phys . rev . _ d * 66 * 044019 ] . a comprehensive presentation of the various horizon effects for gravity - capillary waves is given , with emphasis on the deep water / short wavelength case where many analytical results can be derived . a similarity of the state space of the waves to that of a thermodynamic system is pointed out .

1004.5546

the interest in black - hole analogues has been mainly driven by the intriguing possibility of observing hawking radiation in the laboratory @xcite . in addition to the experimental challenges , this pursuit has important theory implications because of the well - known weakness in the derivation of the hawking effect for real black holes @xcite . hawking s semi - classical calculation @xcite is based on a consideration of fields that are assumed to have no appreciable gravitational effect compared to the black hole , but the derivation contradicts this assumption because the fields attain arbitrarily high frequencies ( and therefore energies ) at the horizon . this so - called trans - planckian problem reveals the lack of any proper understanding of quantum - gravitational effects . unfortunately , the question of whether black holes really radiate does not seem to be amenable to experimental investigation because such radiation would be completely swamped by the cosmic microwave background . in black - hole analogues the trans - planckian problem is avoided by dispersion that limits the blue - shifting of waves at horizons @xcite . the hawking effect has been found to persist even in the presence of dispersion @xcite , although in general numerical simulations must be resorted to in order to verify this . the lack of a good understanding of the hawking effect with arbitrary dispersion is an important and presumably solvable problem , whereas a proper understanding of the hawking effect in real black holes requires unknown physics to deal with the infinite blue - shifting . analogue systems therefore provide the opportunity to understand the hawking effect using known physics and to verify it experimentally ; neither of these tasks can be achieved with real black holes . the introduction of dispersion into the hawking effect opens up a variety of new possibilities . indeed , one of the lessons of analogue systems is that the physics of horizons , including the hawking effect , has a breadth and richness that is not immediately apparent from the case of real black holes . one can consider white - hole and black - hole horizons @xcite , that both give rise to the hawking effect ; horizons allow the blue - shifting and red - shifting of probe waves @xcite ; depending on the dispersion , the probe waves can bounce back at horizons , or go straight through , or bounce back and forth a number of times ( see below ) ; two horizons can communicate leading to runaway quantum hawking radiation or a classical instability @xcite . perhaps the least exotic black - hole analogue that has been proposed is that of surface waves on a moving fluid . schtzhold and unruh @xcite showed that long - wavelength surface waves in a shallow moving fluid obey the klein - gordon equation in a curved space - time geometry . by varying the flow speed so that it exceeds the wave group velocity in some region , one produces a horizon . close to the horizon the long - wavelength assumption breaks down and the detailed behaviour is governed by the dispersion ; the wave evolution is no longer described by an effective space - time metric and there is therefore no trans - planckian problem of infinite blue - shifting . further results for surface waves from an analogue - gravity perspective were given in @xcite . recently , we performed experiments to investigate the behaviour of surface waves at a white - hole horizon , using water waves in a wave tank with a counter - flow @xcite . the speed of the counter - flow was varied along the tank by the presence of a sloping region on the bottom . figure [ horizoninfluid ] shows how waves propagating against the flow are blocked at the point where the counter - current reaches the group velocity of the wave ; this blocking line is the white - hole horizon . video of the incoming wave was used to produce the space - time diagram in figure [ blueshifting ] ( note that the directions of the flow and the wave are reversed in the plot of figure [ blueshifting ] compared to figure [ horizoninfluid ] ) . the linear features with positive slope in the space - time diagram are the evolution of the wave crests and troughs in time the world - lines of the crests and troughs . the inverse @xmath1 of the slope of the world - lines gives the speed of the crests and troughs , the phase velocity @xmath2 of the wave since the phase is @xmath3 ; the lines curve upward as the wave reaches the white - hole horizon , showing a decrease in the phase velocity . the wavelength in the laboratory is revealed by drawing a horizontal line in the diagram and measuring the distance between two crests ; because of the curving upward of the crest world - lines , the wavelength is seen to decrease as the horizon is approached . this is the characteristic wavelength - shortening ( blue - shifting " ) of waves at a white - hole horizon , also observed in an optical analogue @xcite . the inset in figure [ blueshifting ] shows the behaviour of rays at a white - hole horizon where there is no dispersion ; the rays stick at the horizon which corresponds to an infinite blue - shifting of the waves ( the trans - planckian problem ) . it is the dispersion of the surface waves that limits the amount of blue - shifting at the horizon ; in this paper we will discuss some of the rich set of possible behaviours . not surprisingly , the blocking of waves by a counter - flow that exceeds the wave group velocity is a phenomenon that is well - known in the fluid - mechanics community @xcite . the blue - shifting of waves at the blocking line ( white - hole horizon ) is also well - known and has been investigated experimentally @xcite . there is however another possible process in the interaction of a wave with a counter - flow , one that does not seem to have been considered by the fluid - mechanics community , even though it is a possibility clearly visible in the dispersion relation @xcite : this process is the hawking effect . in a stationary flow , where the flow speed at each point in the tank is constant in time but varies from point to point , the frequency in the laboratory frame of a surface wave on the flow is conserved . as we have seen , the wavelength in the laboratory is not conserved , so a white - hole horizon converts the wavelength of a wave propagating against the flow to a different wavelength while conserving its frequency in the laboratory . figure [ blueshifting ] is an example of such a wave ; it is right - moving against the flow ( ingoing against the flow ) with positive ( angular ) frequency @xmath4 and positive wave number @xmath5 and the horizon increases @xmath5 while keeping @xmath4 fixed . an important quantity is the frequency of the wave in a frame co - moving with the flow ; this co - moving frequency is not conserved , but the ingoing wave and the blue - shifted wave with higher @xmath5 both have positive co - moving frequencies @xcite . it turns out that there is often a solution of the dispersion relation for the fixed positive input frequency @xmath4 that has a _ negative _ @xmath5 and a _ negative _ co - moving frequency @xcite . the laboratory frequency @xmath4 of the input wave must be conserved in the wave evolution but when there is a wave with negative co - moving frequency at same value of @xmath4 , there exists the possibility that it could be generated in the interaction of the input wave with the counter - flow . these waves with negative co - moving frequency would be produced in addition to the blue - shifted waves with positive co - moving frequency ; this process is the hawking effect , which is at root a classical effect , although with extraordinary quantum implications @xcite . the strangest feature of the hawking effect is that it is an amplification of the ingoing wave , an extraction of energy from the flow ( or from whatever background creates the horizon ) . the famous quantum hawking radiation is a rather straightforward consequence of the classical hawking effect when the fields underlying the waves are quantized ( the role of input waves is then played by the quantum vacuum and the energy extraction in the hawking process allows the spontaneous creation of field quanta ) @xcite . in our experiments described above @xcite we observed indications of waves with negative co - moving frequency . numerical simulations indicated that the hawking effect would be unobservable in the regimes covered in the experiments @xcite , but it is not clear how well the wave evolution was described by the theoretical model used in the simulations . it appears that our paper was the first experimental search for the hawking effect with surface waves , and further experiments are needed . we hope through this paper to increase awareness of the hawking effect among researchers in fluid mechanics by showing how it has been overlooked in the existing literature on wave blocking . complementary to this , we hope to show those familiar with horizon physics some of the rich horizon effects that occur in models of wave blocking that are used in fluid mechanics . such surprising connections between apparently disparate areas of physics ( and engineering ) are often a source of inspiration to both sides . it is important to stress that the model of surface waves considered in this paper is a linear one : the dynamics is given by a linear wave equation , but with a nonlinear dispersion relation . as is well known , surface waves behave non - linearly if the wave amplitude is large enough , and this presents a potential problem since the shortening of the wavelength at a horizon necessarily leads to an increase in the wave amplitude . the limiting of blue - shifting by dispersion also limits the amplitude increase at blocking lines and , for the linear model to be applicable , it is important that the wave amplitude not get too large . the maximum amplitude attained during the interaction with a counter - flow can to some extent be controlled through the amplitude of the incident wave , and dissipation ( see next paragraph ) also helps to limit the amplitude . the interaction of nonlinear waves with a counter - flow is a highly complex process that is not discussed here ( see for example @xcite ) . also neglected in the model used below is the viscosity of the fluid , and this limits the accuracy of the results when wavelengths are blue - shifted down to very small ( capillary ) wavelengths since viscosity quickly damps such waves . but it is easy to make allowance for this latter drawback of the model by bearing in mind the limited propagation length of capillary waves . the subjects of black holes and water waves have a historical connection through the figure of pierre - simon de laplace . in his `` exposition du systme du monde '' in 1796 , laplace famously introduced the term _ toile sombre _ ( dark star ) to denote an object whose gravitational field is strong enough to prevent light from escaping ( the same concept had been described in 1783 by the reverend john michell in a letter to cavendish ) @xcite . laplace is also well known for having derived ( in 1775 ) a dispersion relation for surface waves on water @xcite . when the water has a background flow ( which does not vary with depth ) and the effect of surface tension is included , the dispersion relation takes the form widely used in fluid mechanics to describe waves on moving water @xcite : @xmath6 here @xmath4 is the ( angular ) frequency of the wave in the laboratory frame and @xmath5 the wave - number ; @xmath7 is the speed of the flow , @xmath8 the depth of the fluid , @xmath9 the gravitational constant , @xmath10 the fluid density and @xmath11 the surface tension . for water , @xmath12 and @xmath13 . equation ( [ disprel ] ) is a one - dimensional dispersion relation suitable for water - tank experiments . when the flow is stationary ( @xmath7 independent of time , but gradually varying in space ) , @xmath4 is a constant but @xmath5 varies with spatial position @xmath14 . waves described by ( [ disprel ] ) are a consequence of gravity ( @xmath9 ) and surface tension ( @xmath11 ) and are called gravity - capillary waves . for small @xmath5 the gravity term dominates , which we call the gravity regime , whereas for large @xmath5 surface tension dominates , giving the capillary regime . the pure - gravity case corresponds to @xmath15 . the quantity @xmath16 is the frequency in a frame co - moving with the fluid . hence the positive , respectively negative , square roots of ( [ disprel ] ) @xmath17 correspond to positive , respectively negative , co - moving frequencies . as described in the introduction , the hawking effect is the generation of a wave on the negative branch of ( [ disprel2 ] ) from a wave on the positive branch , through interaction with a counter - flow . it is a remarkable fact that in the extensive fluid - mechanics literature on the waves ( [ disprel ] ) , including the pure - gravity case @xmath15 , there seems never to have been any consideration of the possibility of conversion of waves from the positive to the negative branch of ( [ disprel2 ] ) through the blocking effect @xcite . if such a conversion process had been investigated , the hawking effect would presumably have been ( re-)discovered in fluid mechanics . solutions of the dispersion relation ( [ disprel ] ) are usually represented graphically . we obtain from ( [ disprel2 ] ) @xmath18 figure [ dispersion ] plots both branches of the right - hand side of ( [ disprel3 ] ) as functions of @xmath5 , for a fixed value of @xmath19 ; figure ( a ) shows the pure - gravity case ( @xmath15 ) and figure ( b ) shows the full dispersion relation with surface tension included . the positive branch of ( [ disprel3 ] ) , corresponding to positive co - moving frequency , is shown in green while the negative branch , corresponding to negative co - moving frequency , is shown in blue . the intersection of these curves with a given horizontal line , such as the red line in the figures , gives the possible waves for the frequency @xmath4 given by that line . in a stationary flow @xmath4 is conserved but the plots of the right - hand side of ( [ disprel3 ] ) change with spatial position as @xmath20 changes and one can trace the evolution of a given solution by following its intersection point with the horizontal line of fixed @xmath4 . as one traces the evolution of the intersection point , the changing group velocity @xmath21 is the slope of the tangent to the curve at the point of intersection with the horizontal line . ; ( a ) shows the pure gravity case ( @xmath15 ) and ( b ) shows the gravity - capillary case . the green curves are the positive branch of the right - hand side of ( [ disprel3 ] ) ( positive co - moving frequency ) and the blue curves are the negative branch ( negative co - moving frequency ) . intersections of these curves with the horizontal red line ( value of @xmath4 ) show possible waves at that @xmath4 . if @xmath7 becomes more negative the green and blue curves rotate clockwise about the origin . local maxima and minima of the green and blue curves show the possibility of blocking waves by means of appropriate velocity profiles @xmath20 ( see figure [ rays ] ) . a water depth @xmath22 is assumed . , title="fig:",width=453 ] ; ( a ) shows the pure gravity case ( @xmath15 ) and ( b ) shows the gravity - capillary case . the green curves are the positive branch of the right - hand side of ( [ disprel3 ] ) ( positive co - moving frequency ) and the blue curves are the negative branch ( negative co - moving frequency ) . intersections of these curves with the horizontal red line ( value of @xmath4 ) show possible waves at that @xmath4 . if @xmath7 becomes more negative the green and blue curves rotate clockwise about the origin . local maxima and minima of the green and blue curves show the possibility of blocking waves by means of appropriate velocity profiles @xmath20 ( see figure [ rays ] ) . a water depth @xmath22 is assumed . , title="fig:",width=453 ] in the pure - gravity case ( figure [ dispersion ] ( a ) ) there are at most four real roots of the dispersion relation ( @xmath23 , @xmath24 , @xmath25 and @xmath26 ) . the solution @xmath23 is a right - moving wave in the laboratory , having positive phase and group velocities , propagating against a left - moving counter - flow @xmath19 . if the flow speed @xmath27 increases as the wave moves to the right ( @xmath7 becomes more negative ) the green and blue curves tip over clockwise about the origin ( as in figure [ rays ] ) , so the root @xmath23 increases the wave is blue - shifted . when the wave reaches a point where the flow speed has increased to make the roots @xmath23 and @xmath24 coalesce at a local maximum of the green curve , the group velocity of the wave is zero it has been blocked at a white - hole horizon . the wave has been stopped by a negative group acceleration " that is still non - zero at the blocking point so the group velocity decreases to negative values ; the wave moves back to the left in the laboratory on the @xmath24 root of the dispersion relation , back into the region where the counter - flow is slower than the blocking speed and where the wave previously had wave number @xmath23 . the ingoing wave @xmath23 has thus been blue - shifted to @xmath24 by the white - hole horizon ; the blue - shifted wave @xmath24 has positive _ phase _ velocity , so its crests move to the right in the laboratory , but it has negative group velocity . the third real root of the dispersion relation with positive co - moving frequency ( green curve ) in figure [ dispersion ] ( a ) is @xmath25 ; this is simply a wave propagating in the same direction as the flow , to the left with negative phase and group velocities . the solution @xmath25 is rather trivial and is of no interest for horizon effects . the root @xmath26 in figure [ dispersion ] ( a ) has negative co - moving frequency ( blue curve ) and is of great interest for horizon effects ; the conversion of some of the input wave @xmath23 to @xmath26 is the hawking effect . since it has a negative wave number , the wave @xmath26 has a negative phase velocity in the laboratory ; its crests move backwards relative to the direction of the ingoing wave , in contrast to @xmath23 and @xmath24 , which both have positive phase velocity . it is essential to understand that the three waves @xmath23 , @xmath24 and @xmath26 are all propagating to the right _ relative to the fluid _ , even though in the laboratory @xmath26 has a phase velocity pointing left and both @xmath24 and @xmath26 have a group velocity pointing left ( @xmath25 is the root corresponding to a wave moving to the left relative to the fluid ) . unlike the blue - shifting of @xmath23 to @xmath24 , the existence of conversion from @xmath23 to @xmath26 can not be deduced from dispersion plots , which only reveal it as a possibility . the amount of conversion of @xmath23 to @xmath26 depends on the details of the dispersion and the velocity profile @xmath20 . in simple cases involving limited dispersion the hawking effect is determined by the slope @xmath28 of @xmath7 at the horizon ( this slope is the analogue of the surface gravity of a black hole , the acceleration due to gravity at the horizon ) , but for general dispersion and velocity profiles no analytical formula for the size of the effect has been found and one must resort to numerical simulations of the wave evolution . one aspect of the challenge to find a good intuitive understanding of the hawking effect is apparent from the description of the horizon given above : this was taken as the point where the flow speed matched the group velocity of the blue - shifting wave . but the phase velocity of this wave is greater than its group velocity , so one can have a group - velocity horizon but no phase - velocity horizon . on the other hand , one can have both a group- and a phase - velocity horizon , with in principle an arbitrary distance between these two horizons and a completely different value of @xmath28 at each horizon . the size of the hawking effect is influenced by these and other factors . there is also the further possibility of having the maximum flow speed close to but less than that required for a group - velocity horizon . in this case one would expect some wave tunneling into the blue - shifted root @xmath24 , and perhaps also into @xmath26 ( tunneling of surface waves has been studied in @xcite ) . numerical simulations indicate that this method of generating @xmath26 without a group - velocity horizon is mathematically possible for a steep enough velocity profile , but it should not be possible in practice @xcite . in the experiments reported in @xcite waves with negative phase velocity were observed even in the absence of a white - hole group - velocity horizon , but , as stated in the introduction , the origin of those waves is not clear . the conversion of @xmath23 to @xmath24 discussed above is well known in fluid mechanics , under the name of wave blocking @xcite . the superposition of the @xmath23 and @xmath24 waves has been shown to be describable by an airy interference pattern @xcite . in contrast , the root @xmath26 in figure [ dispersion](a ) has been largely neglected by the fluid - mechanics community . although the graphical representation of the dispersion relation is standard in fluid mechanics , very few authors @xcite plot the negative-@xmath5 part , in either the pure - gravity or gravity - capillary cases , and the conversion of @xmath23 to @xmath26 appears not to have been considered . are shown in the dispersion plots ( upper ) and the @xmath14-positions where the profile takes these four values are shown in the ray solutions ( lower ) . the values are ( in @xmath29 ) @xmath30 , @xmath31 , @xmath32 and @xmath33 . we use a hyperbolic tangent function @xmath34 to describe the variation of the velocity profile , with a typical length @xmath35 . the conserved frequency @xmath4 of the waves ( red horizontal line in the dispersion plots ) corresponds to a period @xmath36 . a water depth @xmath22 is assumed.,width=453 ] are shown in the dispersion plots ( upper ) and the @xmath14-positions where the profile takes these four values are shown in the ray solutions ( lower ) . the values are ( in @xmath29 ) @xmath30 , @xmath31 , @xmath32 and @xmath33 . we use a hyperbolic tangent function @xmath34 to describe the variation of the velocity profile , with a typical length @xmath35 . the conserved frequency @xmath4 of the waves ( red horizontal line in the dispersion plots ) corresponds to a period @xmath36 . a water depth @xmath22 is assumed.,width=377 ] turning to the full gravity - capillary case , figure [ dispersion ] ( b ) shows ( for fixed @xmath19 ) how the surface tension @xmath11 changes the dispersion relation at large wave numbers compared to the pure - gravity case in figure [ dispersion ] ( a ) . for the value of @xmath7 plotted , the positive co - moving frequency curve ( green ) has a local minimum as well as a local maximum , and this is also the case for the negative co - moving frequency curve ( blue ) , although the local minimum of the latter curve always occurs at negative laboratory frequency @xmath4 . if @xmath7 becomes more negative the curves tip over clockwise about the origin , so that for larger counter - flow speed there exist roots with negative co - moving frequencies at the @xmath4 shown by the red line . each local maximum or minimum of the green and blue curves reveals the possibility of reversing the group velocity of a wave with an appropriate velocity profile , as in the discussion of the local maximum in the pure - gravity case ( figure [ dispersion ] ( a ) ) . the local maximum of the green curve in figure [ dispersion ] ( b ) allows the wave blocking and blue shifting of an incident right - moving wave as in the pure - gravity case . we refer to this blocking line as the white horizon ( from white hole ) . but after the white horizon has reversed the group velocity of the incident wave so that it now moves to the left , this blue - shifting wave encounters another blocking line because of the local minimum of the green curve . we refer to this second blocking line for the blue - shifting wave as the blue horizon . at the blue horizon the group velocity reverses once more to become positive so the wave moves to the right again towards the white horizon . this time the ( still blue - shifting ) wave goes right through the white horizon , so overall the incident wave undergoes a double bounce . figure [ rays ] shows the graphical solution of the dispersion relation at four values @xmath37 of a velocity profile @xmath38 and also a numerical solution of the ray equations for the incident right - moving wave ( green curve in the lower figure ) . the @xmath14-positions where the velocity profile takes the four values used in the dispersion plots are shown in the ray plot . rays move at the group velocity and so the wave blocking is clear from the ray plots . also shown is the ray solution for the wave with negative co - moving frequency ( blue curve ) ; this wave is initially left - moving in the laboratory but its group velocity is reversed at a blocking line we refer to as the negative horizon ( this horizon does not exist in the pure - gravity case see @xcite ) . comparing the dispersion and ray plots in figure [ rays ] ( and ignoring the co - propagating wave @xmath25 discussed above ) one can see how the increasing counter - flow speed as @xmath14 increases gives , successively , one @xmath5 root in the profile where @xmath39 , three roots in the region where @xmath7 lies between @xmath40 and @xmath41 , five roots in the region where @xmath7 lies between @xmath41 and @xmath42 , and three roots in the region where @xmath43 . one can also see how these roots relate to the ray behaviour . ( see @xcite for ray plots in the pure - gravity case . ) figure [ rays2 ] shows an example where the frequency @xmath4 and counter - flow profile @xmath38 are such that only a wave with positive co - moving frequency exists . the wave again displays the double - bouncing behaviour . figure [ num ] shows a numerical solution for a wave packet centered on the ray in figure [ rays2 ] . this simulation was obtained by solving the scalar wave equation describing the surface wave on the counter - flow ; it was shown by schtzhold and unruh @xcite that this equation takes the form of the klein - gordon equation in a curved space - time with added higher - order dispersion , in this case the dispersion ( [ disprel ] ) of gravity - capillary waves : @xmath44 the method of numerically solving equations of the form ( [ waveeqn ] ) , for essentially arbitrary dispersion , is described in @xcite ; further examples of this kind of numerical solution for surface waves appear in @xcite and @xcite . in the wave - packet simulation in figure [ num ] , the continuous blue - shifting that accompanies the double bounce is apparent . because of the spread of frequencies in the wave - packet there is some leakage of the initial wave through the white horizon ( first bounce ) as well a spreading and separation of frequency components at the blue horizon ( second bounce ) . for extensive numerical simulations of gravity - capillary waves in the presence of a current , see @xcite . corresponds to a period @xmath45 . for this frequency there is no wave with negative co - moving frequency in the range @xmath46 of counter - flow speeds used . the three values of the flow velocity shown are ( in @xmath29 ) @xmath47 , @xmath48 , @xmath49 . a water depth @xmath22 is assumed.,width=377 ] corresponds to a period @xmath45 . for this frequency there is no wave with negative co - moving frequency in the range @xmath46 of counter - flow speeds used . the three values of the flow velocity shown are ( in @xmath29 ) @xmath47 , @xmath48 , @xmath49 . a water depth @xmath22 is assumed.,width=377 ] . , width=415 ] . , width=415 ] . , width=415 ] . , width=415 ] . , width=415 ] . , width=415 ] . , width=415 ] . , width=415 ] . , width=415 ] in practice the blue - shifting of incident gravity waves into the capillary regime , as in figure [ num ] , will be limited by viscosity , which is not included in the model we have been discussing . as a consequence , the highly blue - shifted waves produced at the blue horizon will dissipate rapidly . an experimental investigation of these effects was first performed by badulin _ et al . _ these authors observed the initial blocking of waves in the gravity regime ( white horizon ) and the subsequent conversion into waves in the capillary regime ( blue horizon ) which then propagated through the original blocking line and vanished through viscous damping @xcite . gravity - capillary waves on a counter - flow have also been studied experimentally by klinke and long @xcite , who produced space - time diagrams of the wave evolution . trulsen and mei @xcite give a theoretical treatment that includes numerical simulations . a recent theoretical survey is given by huang @xcite . we have so far discussed only certain features of the dispersion relation ( [ disprel ] ) . in the next section we classify in more detail how the presence or absence of the various horizons , and their positions , depend on the conserved frequency @xmath4 and velocity profile @xmath20 . the influence of the water depth @xmath8 in the dispersion relation ( [ disprel ] ) disappears when @xmath50 , so that @xmath51 ( depending on the sign of @xmath5 ) . this is the case of wavelengths short compared to the water depth and it gives the polynomial dispersion relation @xmath52 that is easier to handle analytically than the original ( [ disprel ] ) . the waves considered in figure [ dispersion ] to figure [ num ] are in fact very well described by the deep water / short wavelength dispersion relation ( [ dispshort ] ) , as were the waves studied in the experiments @xcite and @xcite . at the blocking lines or horizons discussed in the last section , the group velocity vanishes and the dispersion curve @xmath53 has a local extremum . three possible horizons for gravity - capillary waves were identified : the white , blue and negative horizons in figure [ dispersion](b ) . at each horizon two real roots of the dispersion relation coalesce into one double root and then disappear : in the terminology of dynamical systems it is a saddle - node or tangent bifurcation @xcite . the order parameter of the bifurcation is the wave number whereas the two control parameters are the velocity @xmath7 ( the external field " ) and the frequency @xmath4 ( the internal parameter " ) . following the approach in @xcite , we find the horizons by looking for a double root @xmath54 of the cubic dispersion relation ( [ dispshort ] ) : @xmath55 where @xmath56 is the remaining simple root . comparing coefficients of @xmath5 in ( [ dispshort ] ) and ( [ double ] ) we obtain expressions for @xmath54 and @xmath56 , as follows . the comparison of coefficients gives three equations for the two unknowns @xmath56 and @xmath54 ; two of these equations are solved for @xmath56 and @xmath54 , and the the third equation is then a constraint relating @xmath56 to @xmath54 . in the case of positive wave numbers ( plus sign in ( [ dispshort ] ) ) this procedure gives @xmath57 { \displaystyle k_1=\frac{\rho u^2}{3\gamma } \left ( 1\mp 2\sqrt{1-\frac{3\gamma}{\rho u^4}(g+2\omega u)}\right ) } \end{array } \quad \mbox{(positive wave numbers)}\ ] ] with the constraint @xmath58 the constraint ( [ k1k2 ] ) , for both sign possibilities in ( [ kspos ] ) , leads to @xmath59=0 \quad\left[\!\!\ ! \begin{array}{c } \mbox{positive wave } \\ \mbox{numbers } \end{array}\!\!\ ! \right]\ ] ] for negative wave numbers ( minus sign in ( [ dispshort ] ) ) @xmath54 and @xmath56 are @xmath60 { \displaystyle k_1=-\frac{\rho u^2}{3\gamma } \left ( 1\mp 2\sqrt{1-\frac{3\gamma}{\rho u^4}(g-2\omega u)}\right ) } \end{array } \quad \mbox{(negative wave numbers)}\ ] ] with the constraint @xmath61 that leads to @xmath62=0 \ \begin{array}{c } \mbox{negative wave } \\ \mbox{numbers } \end{array}\!\!\ ! \right]\ ] ] the significance of the constraints ( [ conpos ] ) and ( [ conneg ] ) is clear from figure [ dispersion](b ) . with a particular choice of @xmath7 , local extrema in the dispersion plot ( corresponding to double roots @xmath54 ) occur at values of @xmath4 determined by this choice of @xmath7 . for a given @xmath7 , the constraint ( [ conpos ] ) or ( [ conneg ] ) is a quartic in @xmath4 whose real roots give all the frequencies at which a blocking line ( horizon ) occurs for this @xmath7 . alternatively , upon fixing @xmath4 the constraint gives a quintic in @xmath7 whose real roots are all the flow velocities that give a horizon at this frequency . at which the white , blue and negative horizons occur , as a function of the wave period @xmath63 ( black curves ) . the curve lying on or close to the straight red line is the white horizon , the red line showing the pure - gravity white horizon . the curve approaching the asymptote @xmath64 from below is the blue horizon , and the curve approaching this asymptote from above is the negative horizon . , width=453 ] as in the previous section , we consider only positive @xmath4 and negative @xmath7 , whereas @xmath5 can be positive or negative ( this gives no restriction in the horizon effects ) . the white horizon in figure [ dispersion](b ) occurs at a value of @xmath5 low enough for the influence of the surface tension @xmath11 in ( [ dispshort ] ) to be very small . a very good approximation for the white horizon , which is exact in the pure - gravity case , is therefore obtained by putting @xmath15 . neglecting @xmath11 , the constraint ( [ conpos ] ) gives @xmath65 note that the approximation ( [ congrav ] ) is also obtained in the large @xmath7 limit , so for large @xmath7 it becomes the exact constraint at the white horizon for gravity - capillary waves . from ( [ congrav ] ) we recover the relation between @xmath7 and @xmath4 that gives the blocking of gravity waves at a white - hole horizon @xcite @xmath66 where @xmath63 is the period . the corresponding value of the double root @xmath54 is obtained from the @xmath67 limit of ( [ kspos ] ) ; the lower sign in the first of ( [ kspos ] ) gives the only finite expression : @xmath68 equations ( [ ugrav])([k2grav ] ) show the exact relationship between the frequency , wave number and counter - flow speed at the white horizon for pure - gravity waves . note from ( [ ugrav ] ) that the flow speed at the white horizon is proportional to the conserved period of the blocked wave . the straight line @xmath69 versus @xmath63 is shown in red in figure [ phase ] . superimposed on that red line is a black curve that shows the exact relationship between flow speed and period for gravity - capillary waves . in line with the comments above , the red line agrees very well with the gravity - capillary case except for small @xmath70 , and therefore small @xmath63 . the striking feature of the gravity - capillary curve is that it ends at the point labeled @xmath71 ; the white horizon thus does not exist for periods @xmath63 that are below a critical value @xmath72 , or for counter - flows that do not reach a critical speed @xmath73 . the existence of this threshold can be seen from the dispersion plots in figure [ rays ] : for flow velocity @xmath74 there is no local maximum of the green curve , so no frequency @xmath4 experiences a white horizon at this flow velocity ; in contrast , the other flow velocities plotted in the figure all give a local maximum of the green curve and therefore a white horizon for the frequency at this maximum . similarly , if the period @xmath63 is too small ( frequency @xmath4 too large ) , the horizontal red line in figure [ rays ] will not intersect a local maximum in the dispersion plot for any @xmath7 , so there can be no white horizon for such periods . . the wave with positive co - moving frequency occurs at a point of inflection when @xmath75 . this means that the corresponding ray ( green ) has a group velocity that slows to zero when @xmath70 increases to @xmath73 ; the group velocity does not reverse , however , and the ray resumes its propagation into regions of higher @xmath70 . , width=453 ] . the wave with positive co - moving frequency occurs at a point of inflection when @xmath75 . this means that the corresponding ray ( green ) has a group velocity that slows to zero when @xmath70 increases to @xmath73 ; the group velocity does not reverse , however , and the ray resumes its propagation into regions of higher @xmath70 . , width=377 ] let us look in more detail at the critical values @xmath71 . figure [ inflection ] shows graphically the occurrence of the threshold for the white horizon . we see that the disappearance of a local maximum in the green curve , as @xmath70 decreases , is accompanied by the disappearance of the local minimum , and at the critical value @xmath76 the two local extrema coalesce to form a point of inflection at frequency @xmath77 . this shows that @xmath71 is also the threshold for the occurrence of the blue horizon , which requires a local minimum in the green curve . at @xmath71 the two double roots @xmath54 in ( [ kspos ] ) ( white and blue horizons ) coincide , and in fact the same value is taken by the simple root @xmath56 , as can also be seen from figure [ inflection ] . the values @xmath71 can be obtained by solving for the point of inflection @xmath78 in the dispersion relation ( [ dispshort ] ) ( @xmath79 ) ; these two equations can be solved for @xmath76 and the critical wave number @xmath80 , and @xmath77 then follows from the dispersion realtion . alternatively , the point of inflection is found by demanding that the square - root expression in ( [ kspos ] ) vanishes so that all @xmath54 and @xmath56 coincide . the result is @xmath81 and @xmath82 with the wave number @xmath83 in using the method ( [ double ] ) of searching for the system parameters at horizons , we noted that they correspond to saddle - node or tangent bifurcations in dynamical - systems theory @xcite . the cusp ( @xmath84 ) in the @xmath70 vs @xmath63 diagram ( figure [ phase ] ) is the point where two saddle - node lines ( horizons ) intersect . in the terminology of dynamical systems this corresponds to a so - called pitchfork bifurcation @xcite . -axis ( @xmath85 , @xmath86 ) . this counter - flow velocity is given by ( [ ugamma ] ) and has the value @xmath87 . , width=453 ] since @xmath71 is also a threshold for the existence of the blue horizon , the curve in the @xmath70 versus @xmath63 diagram ( figure [ phase ] ) relating the period to the flow speed at the blue horizon must also end at @xmath71 ; this curve is also shown in figure [ phase ] and it is seen to approach an asymptotic value , labelled @xmath88 , as @xmath89 . this is because @xmath89 means @xmath90 and the dispersion plot in figure [ bluezero ] shows that the local minimum giving the blue horizon occurs at @xmath85 for a finite non - zero @xmath7 that we call @xmath88 . we find the flow velocity @xmath88 as follows . the constraint ( [ conpos ] ) relates the values of @xmath7 and @xmath4 at all horizons ( double roots of the dispersion relation ) for waves with positive @xmath5 . hence by taking @xmath90 in ( [ conpos ] ) we obtain @xmath88 ; this limit of ( [ conpos ] ) gives @xmath91 so @xmath92 the wave vector at the local minimum ( blue horizon at @xmath85 ) in figure [ bluezero ] is found by inserting the velocity @xmath88 into the expression for the double root @xmath54 in ( [ kspos ] ) and taking the upper sign : @xmath93 the other local extremum at @xmath79 in figure [ bluezero ] , the white horizon for a non - zero @xmath4 given by the horizontal red line in the figure , occurs at @xmath94 and the corresponding @xmath4 ( red line ) is @xmath95 where the exact but lengthy numerical coefficients have not been reproduced . the simple root @xmath56 for @xmath96 is zero for @xmath85 , while for @xmath4 given by ( [ omwhitezeroblue ] ) ( red line figure [ bluezero ] ) the simple root is ( intersection of red line with green curve at large @xmath79 in figure [ bluezero ] ) @xmath97 these last three results are obtained from the constraint ( [ conpos ] ) and the expressions ( [ kspos ] ) for the double and single roots , with @xmath96 . we see from figure [ bluezero ] that in the limit @xmath90 ( @xmath89 ) the negative horizon also occurs at the flow velocity @xmath88 , as well as the blue horizon , and the wave vector at the negative horizon is minus that at the blue horizon , @xmath98 . figure [ bluezero ] shows that , for waves with positive laboratory frequency @xmath4 , the counter - flow velocity @xmath88 is the threshold for the existence of waves with negative co - moving frequency ; the threshold flow velocity for such waves is the threshold for the negative horizon . it follows that the curve in the @xmath70 versus @xmath63 diagram ( figure [ phase ] ) relating the period to the flow speed at the negative horizon must lie above the line @xmath64 and asymptotically approach this line as @xmath89 . this negative - horizon curve is also plotted in figure [ phase ] ; it lies above the blue - horizon curve but shares with it the asymptote @xmath64 . unlike the white- and blue - horizon curves in figure [ phase ] , which both end at the cusp @xmath71 for small @xmath63 , the negative - horizon curve diverges to @xmath99 as @xmath100 this behaviour of the negative horizon is clear from the dispersion plots because as @xmath100 ( @xmath101 ) the flow speed @xmath70 must increase without limit in order for the local maximum in the negative-@xmath5 curve to reach the horizontal frequency line ( see for example figure [ rays ] ) . the flow velocity @xmath88 that appears as an asymptote in the @xmath70 versus @xmath63 diagram ( figure [ phase ] ) has additional significance in fluid mechanics . firstly , it is a well - known property of gravity - capillary waves on _ static _ water ( @xmath102 ) that the minimum phase velocity of the waves is given ( apart from the sign ) by the expression ( [ ugamma ] ) for @xmath88 . the velocity @xmath88 is also important in the case of _ shear flows _ , i.e. velocity profiles that change with the fluid depth . it was shown by caponi _ et al . _ @xcite that a sufficient condition for a shear flow to become spontaneously unstable is for the flow velocity on the surface of the fluid to exceed @xmath88 ; the instability leads to the generation of gravity - capillary waves on the fluid surface @xcite . another example in shear flows is the appearance of negative - energy waves at the interface of two fluid layers , which occurs when the relative velocity of the layers exceeds @xmath88 ; this is related to the famous kelvin - helmholtz instability , as discussed by fabrikant and stepanyants @xcite . . the roman numerals i and v refer to counter - flow speeds that lie in the regions labeled by these numerals in figure [ phase ] . the period is @xmath103 and the flow velocities are @xmath104 ( i ) and @xmath105 ( v).,width=453 ] . the roman numerals i and v refer to counter - flow speeds that lie in the regions labeled by these numerals in figure [ phase ] . the period is @xmath103 and the flow velocities are @xmath104 ( i ) and @xmath105 ( v).,width=377 ] figure [ phase ] allows the classification of the behaviour of gravity - capillary waves on a stationary counter - flow . the period @xmath63 is conserved in the wave evolution so by fixing a vertical line in the figure one can distinguish five qualitatively different possibilities for waves of a single frequency : \1 . as one moves into regions in the velocity profile @xmath38 with higher counter - flow speeds , one moves from region i in figure [ phase ] into region v. the dispersion plots and ray solutions for this case are shown in figure [ itov ] . notable is the fact that there is no white or blue horizon . @xmath107 . here the line of constant @xmath63 is the vertical green line in figure [ phase ] that separates region i from regions ii and iii and passes through the cusp point @xmath71 . the dispersion and ray plots are shown in figure [ inflection ] ; as already discussed , this case is the threshold for the appearance of the white and blue horizons . . the roman numerals refer to counter - flow speeds that lie in the regions labeled by these numerals in figure [ phase ] . the period is @xmath108 and the flow velocities are @xmath109 ( ii ) , @xmath110 ( iv ) , @xmath111 ( iii ) and @xmath112 ( vi).,width=453 ] . the roman numerals refer to counter - flow speeds that lie in the regions labeled by these numerals in figure [ phase ] . the period is @xmath108 and the flow velocities are @xmath109 ( ii ) , @xmath110 ( iv ) , @xmath111 ( iii ) and @xmath112 ( vi).,width=377 ] \3 . the line of constant @xmath63 lies between the green and brown vertical lines in figure [ phase ] and so passes through region iii . here increasing counter - flow speeds takes us from region ii to iv to iii to vi in figure [ phase ] . the dispersion plots and ray solutions for this case are shown in figure [ iitoivtoiiitovi ] . here there is a white and blue horizon , and the white horizon occurs at a lower counter - flow speed than the negative horizon . ( see figure [ phase ] ) . both the white and negative horizons occur at the same counter - flow velocity @xmath114 , the point @xmath115 being the intersection of the curves for these two horizons in figure [ phase ] . for water , @xmath116 , @xmath117.,width=453 ] ( see figure [ phase ] ) . both the white and negative horizons occur at the same counter - flow velocity @xmath114 , the point @xmath115 being the intersection of the curves for these two horizons in figure [ phase ] . for water , @xmath116 , @xmath117.,width=377 ] \4 . @xmath118 , defined by the brown vertical line in figure [ phase ] . the line @xmath118 passes through the point where regions iii , iv , vii and vi meet , at @xmath119 . the significance of the point @xmath120 is that it is the intersection of the white - horizon curve and the negative - horizon curve ; this means that for a wave with period @xmath121 the white horizon occurs at the same counter - flow speed as the negative horizon . figure [ dispb ] confirms this in the dispersion plots and ray solutions . the values @xmath122 must be found numerically and for water they are @xmath116 , @xmath117 . . the roman numerals refer to counter - flow speeds that lie in the regions labeled by these numerals in figure [ phase ] . the period is @xmath123 and the flow velocities are @xmath124 ( ii ) , @xmath125 ( iv ) , @xmath126 ( vii ) and @xmath127 ( vi).,width=453 ] . the roman numerals refer to counter - flow speeds that lie in the regions labeled by these numerals in figure [ phase ] . the period is @xmath123 and the flow velocities are @xmath124 ( ii ) , @xmath125 ( iv ) , @xmath126 ( vii ) and @xmath127 ( vi).,width=377 ] \5 . @xmath128 . here the line of constant @xmath63 passes through region vii in figure [ phase ] . increasing counter - flow speeds takes us from region ii to iv to vii to vi . the dispersion plots and ray solutions for this case are shown in figure [ iitoivtoviitovi ] . here there is a white and blue horizon , and the white horizon occurs at a higher counter - flow speed than the negative horizon ( compare carefully with case 3 above ) . trulsen @xcite inferred from his results the structure of the @xmath129-diagram of figure [ phase ] , but without the negative - horizon curve . he also derived the existence of the cusp @xmath71 as a triple - root solution of the dispersion relation @xcite . recently , we observed ( with continuous waves trains ) the regions vi , vii and iv of figure [ phase ] , as reported in our experimental @xmath70 versus @xmath63 diagram @xcite . the distinction between the regions ii and iv was unclear from our data . our wave - maker was limited to a minimum period of @xmath130 which is higher than @xmath131 . our focus was on the conversion into waves with negative co - moving frequency rather than conversion into capillary waves in the double - bouncing behaviour described above . conversion to capillary waves is difficult to observe experimentally because of the rapid dissipation of capillary waves ( we were unaware of the work by badulin _ et al . _ in addition , we used a rather high period ( far from @xmath72 ) to get long wavelengths of the ingoing waves since the waves with negative co - moving frequency should be produced with a drastic reduction of the wavelength according to the dispersion relation . we were surprised to find indications of waves with negative co - moving frequency even without wave blocking ( a white horizon ) . badulin _ et al . _ @xcite performed experiments with wave packets ( three to ten wave cycles centred on periods in the range @xmath132@xmath133 ) sent on a counter - flow with speed @xmath70 between @xmath134 and @xmath135 over a sloping bottom . double bouncing of the input waves was observed with a strong reduction in both wavelength ( from @xmath136 to @xmath137 ! ) and amplitude . these authors presented only one photograph , at @xmath138 , of the conversion phenomenon and one measurement of the amplitude of waves as a function of the position / velocity ( @xmath139 and @xmath140 ) , but beautiful measurements of the effect of the velocity on the wavelength . no results for periods less than @xmath141 were reported . in the previous section we summarized the behaviour of gravity - capillary waves ( with @xmath0 ) on a stationary counter - flow by means of a diagram in the @xmath129-parameter space ( figure [ phase ] ) . this diagram allows one to visualize the evolution of an incident wave of a single frequency ( which is conserved ) , as was illustrated in the five cases in figures [ itov ] , [ inflection ] , [ iitoivtoiiitovi],[dispb ] and [ iitoivtoviitovi ] . in this section we note a similarity between figure [ phase ] and a phase diagram in thermodynamics , where the horizon lines in figure [ phase ] are analogous to the lines separating different phases ( first - order phase transition ) . in particular , the cusp point @xmath71 in figure [ phase ] looks like a critical point ( second - order phase transition ) in a phase diagram . let us explore this thermodynamic analogy a little further . in thermodynamics a system is described by an equation of state of the form @xmath142 , where @xmath143 is the pressure , @xmath144 is the volume and @xmath145 is the temperature . for example , the equation of state of a van der waals gas can be written @xcite : @xmath146 where @xmath147 is the number of molecules divided by avogadro s number , @xmath148 is the gas constant , @xmath149 relates to the non - zero volume of the molecules and @xmath150 is a measure of the molecular interaction . a familiar property of the van der waals gas is the existence of a critical point in the @xmath151-phase diagram ; this is associated with a fold catastrophe in the surface ( [ vdw ] ) in @xmath152-space that constitutes the state - space of the gas @xcite . now the dispersion relation ( [ dispshort ] ) , written as @xmath153 describes the state of a wave on a counter - flow as a surface in @xmath154-space and this surface has the same kind of fold catastrophe as the van der waals gas see figure [ cata ] . the connection between wave blocking and catastrophe theory was inferred a long time ago by peregrine and smith @xcite and more recently by trulsen @xcite . the white and blue horizons appear as curves on the state surface in figure [ cata ] , which converge and join at the critical point " @xmath155 given by ( [ tc])([kc ] ) . only the positive-@xmath5 part of the state surface is shown in figure [ cata ] , since this contains the point analogous to the thermodynamic critical point . by projecting the surface in figure [ cata ] on to the @xmath156-plane one obtains the part of the phase diagram " figure [ phase ] containing the white and blue horizons , similar to the @xmath151-diagram of the van der waals gas @xcite . -space given by ( [ uktsurface ] ) . this surface has a fold catastrophe in the positive-@xmath5 sector , similar to the fold catastrophe of the van der waals gas @xcite . the red curve on the state surface is the white horizon and the blue curve is the blue horizon . the horizon curves meet at @xmath155 , which is analogous to the critical point of the van der waals gas . the white- and blue - horizon curves in figure [ phase ] are the projection of the curves on the state surface to the @xmath156-plane . , width=453 ] we can summarize the analogy between the van der waals gas and waves on a counter - flow by the following table , which we stress describes only a qualitative relationship : [ cols="<,^,>",options="header " , ] the perfect gas is seen to correspond to very large @xmath4 and @xmath7 ( pure advection @xmath157 of the surface waves ) . in addition , the perfect gas is obtained by setting the parameters @xmath150 and @xmath149 to zero ; similarly , pure advection of the surface waves corresponds to setting the parameters @xmath11 and @xmath9 to zero . as a function of @xmath5 for different periods @xmath63 : figure ( a ) is for @xmath158 , ( b ) for @xmath107 and ( c ) for @xmath159 . figure ( d ) shows a series of plots of @xmath160 for different @xmath63 ; the green curve is for @xmath100 ( pure advection ) . the blue curves in figures ( a)(c ) show the pure - gravity case . the black dotted line in all four figures shows , in the pure - gravity case , the locus of the local minimum of @xmath160 as @xmath63 changes . the horizontal dotted purple line is @xmath88 ; this is the value of @xmath7 at the local maximum of @xmath160 for @xmath159 ( figure ( c ) ) . the horizontal dotted green line is @xmath76 ; this is the value of @xmath7 at the point of inflection of @xmath160 for @xmath107 ( figure ( b ) ) . the horizontal dotted orange line is @xmath161 , which is proportional to @xmath63 ( see ( [ ugrav ] ) ) ; it gives the value of @xmath7 at the local minimum in the pure gravity case ( blue curves in figures ( a)(c ) ) . , width=566 ] the fold catastrophe in figure [ cata ] can also be projected to the @xmath162-plane ( figure [ uvsk ] ) . from the dispersion relation ( [ dispshort ] ) these projections are given by @xmath163 with a fixed value of @xmath4 ( @xmath63 ) . figure [ uvsk ] shows the curves @xmath160 for different periods @xmath63 , for both the gravity - capillary case ( red lines in ( a ) , ( b ) and ( c ) ) and the pure - gravity case ( blue lines in ( a ) , ( b ) and ( c ) ) . figure [ uvsk](d ) shows @xmath160 for a range of periods @xmath63 in the gravity - capillary case ; these isoperiod curves are analogous to the andrews isotherms for a real gas . we have seen from dispersion plots that wave - blocking corresponds to local extrema of the function @xmath164 . it follows from the implicit function theorem that the minimum ( maximum ) of @xmath160 corresponds to the maximum ( minimum ) of @xmath164 , and therefore blocking lines are given by the local extrema of @xmath160 . these local extrema @xmath165 are the analogues of the spinodal line @xmath166 of the van der waals gas . without surface tension ( blue lines in figure [ uvsk](a)(c ) ) , @xmath160 has a single minimum that lies on the dotted black line in these plots . with surface tension ( red lines in figure [ uvsk](a)(c ) ) , @xmath160 has in addition a local maximum . at @xmath107 ( figure [ uvsk](b ) ) the minimum and the maximum of @xmath160 merge and an inflection point appears ; this corresponds to the cusp in the @xmath156-plane ( figure [ phase ] ) . we can define a mechanical susceptibility @xmath167 , analogous to the isothermal compressibility coefficient @xmath168 , that diverges at the horizons , just as the compressibility of the gas diverges at the spinodal line . in the limit of infinite period @xmath169 ( @xmath170 ) we found that the wave is described by a flow speed ( [ ugamma ] ) and wave number ( [ k2zeroblue ] ) . we recover this result from the @xmath85 case of ( [ uk ] ) , which is @xmath171 this is plotted in figure [ uvsk](c ) and has a maximum @xmath172 at wave number @xmath173 where @xmath174 is the capillary length . this wave number in ( [ uktinf ] ) reproduces @xmath88 , given by ( [ ugamma ] ) . the principle of corresponding states implies that the properties of real gas are universal functions of the state variables scaled to the critical point . for the van der waals gas , it is well known that the equation of state can be written in a universal form around the critical point ( @xmath175 , @xmath176 and @xmath177 ) : @xmath178 where the subscript @xmath179 means reduced variable ( @xmath180 , @xmath181 and @xmath182 ) . similarly , using the scalings ( @xmath183 , @xmath184 and @xmath185 ) , we find the universal dispersion relation : @xmath186 whatever the fluid ( surface tension , density ) , its wave - like behavior will be the same close to the cusp . the dimensionless form of the constraint ( [ conpos ] ) becomes : @xmath187 one recovers @xmath188 by imposing @xmath189 . by introducing another scaling @xmath190 the constraint reads : @xmath191 one recovers @xmath192 by imposing @xmath189 . we have described the interaction of linear gravity - capillary waves with a counter - flow , with emphasis on the various horizon effects ( wave blocking in fluid - mechanics terminology ) . the case of waves with negative co - moving frequency has been included throughout ; these waves are crucial for the hawking effect and they have been neglected in the fluid - mechanics literature on wave blocking . the hawking effect is a remarkable process in which an incident wave generates a wave with negative co - moving frequency , with a resulting _ amplification _ of the incident wave ( this implies an extraction of energy from the flow ) . it has been shown that this process is robust in the presence of dispersion @xcite and the linear theory of surface waves falls into the class of systems that exhibit the hawking effect @xcite . experimental evidence of the generation of waves with negative co - moving frequency was reported in @xcite and further experiments are planned . we presented analytical results for the deep water / short wavelength case @xmath0 that are more comprehensive than those given elsewhere . a similarity of the state space of the waves to that of a thermodynamic system was pointed out ; the curves in the state space representing the horizon lines are analogous to curves separating thermodynamic phases and there is even an analogue of a thermodynamic critical point . schtzhold and unruh showed that the regime of gravity - wave propagation in an effective schwarzschild - like metric corresponds to the shallow water limit @xmath193 @xcite . the interaction with the white - hole horizon necessarily tunes a wave out of this regime ( @xmath194 ) into the @xmath0 regime in which dispersion causes the effective - metric description to break down . in other words , for pure gravity waves the white horizon is not dispersive when @xmath193 whereas it is dispersive when @xmath0 . on the other hand , waves in the @xmath0 regime considered in this paper stay in this regime in the interaction with the counter - flow . previous results on the shape of waves at a blocking line ( horizon ) can be developed further for gravity - capillary waves . the horizons for surface waves can be treated as examples of saddle - node lines , which also describe caustics in optics . it is well known that an airy function describes both the intensity of light close to an optical caustic of the fold type @xcite and the water shape for the amplitude of gravity waves close to a blocking line @xcite . it will be shown elsewhere that the airy function depends on a stopping length " @xmath195 ( roughly the width of the arch of the airy function ) which scales like @xmath196 where @xmath197 is the position of the white horizon , and @xmath198 is the surface gravity " at the horizon . an experimental measurement of the airy shape has been carried out by chawla and kirby @xcite . if surface tension is taken into account , we have seen that there is a critical point at the intersection of two saddle - node lines ( white and blue horizons ) . we anticipate that the wave at the critical point will be described by a pearcey catastrophe integral @xcite due to the superposition of two airy catastrophe integrals for the two saddle - node lines . we are indebted to yury stepanyants , erwann aubry and gil jannes for fruitful discussions . this research was supported by the rgion paca ( projet exploratoire hydro ) , the conseil gnral 06 , the scottish government , the royal society of edinburgh and the royal society of london .

we present a detailed study of a peculiar source detected in the cosmos survey at z=0.359 . source cxoc j100043.1 + 020637 , also known as cid-42 , presents two compact optical sources embedded in the same galaxy . the distance between the two , measured in the hst / acs image , is 0.495``.005 '' that , at the redshift of the source , corresponds to a projected separation of 2.46.02 kpc . a large ( km / s ) velocity offset between the narrow and broad components of h has been measured in three different optical spectra from the vlt / vimos and magellan / imacs instruments . cid-42 is also the only x - ray source in cosmos having in its x - ray spectra a strong _ redshifted _ broad absorption iron line , and an iron emission line , drawing an _ inverted _ p - cygni profile . the and data show that the absorption line is variable in energy bye=500 ev over 4 years and that the absorber has to be highly ionized , in order not to leave a signature in the soft x - ray spectrum . that these features , the morphology , the velocity offset and the _ inverted _ p - cygni profile , occur in the same source is unlikely to be a coincidence . we envisage two possible explanations , both exceptional , for this system : ( 1 ) a gravitational wave recoiling black hole ( bh ) , caught 1 - 10 myr after merging , ( 2 ) a type 1/ type 2 system in the same galaxy where the type 1 is recoiling due to slingshot effect produced by a triple bh system . the first possibility gives us a candidate gravitational waves recoiling bh with both spectroscopic and imaging signatures . in the second case , the x - ray absorption line can be explained as a bal - like outflow from the foreground nucleus ( a type 2 agn ) at the rearer one ( a type 1 agn ) , which illuminates the otherwise undetectable wind , giving us the first opportunity to show that fast winds are present in obscured agn , and possibly universal in agns .

1003.0020

double super - massive black holes ( smbhs ) within a single galaxy are predicted from the combination of hierarchical models of structure formation ( see colpi & dotti 2009 for a recent review ) , and the observed link between black hole ( bh ) mass and host galaxy bulge mass in the local universe ( e.g. , magorrian et al . 1998 , kormendy & gebhardt 2001 ) . kiloparsec scale double nuclei should be common at redshift z@xmath12 where the merging process is very efficient ( springel et al . 2005 ) but at this high z they are hard to resolve . at later epochs ( [email protected] ) they are easier to find , but the merger rate is lower ( hopkins et al . 2008 ) and binary smbhs should be relatively scarce . for these reasons , observational evidence for binary active smbhs at kpc separations remains sparse . a few low redshift examples have been found . the most famous double active smbhs are the highly obscured pair in the ultra - luminous infrared galaxy ( ulirg ) ngc 6240 ( komossa et al . 2003 ; z=0.0245 ) . imaging and spectroscopy of ngc 6240 show the presence of two active galactic nuclei ( agns ) separated by just 1.2 kpc . only few other agn pairs at kiloparsecs separations have been discovered so far , all by chance : arp 299 ( ballo et al . 2004 ; z=0.01 ) , mrk 463 ( bianchi et al . 2008 ; z=0.05 ) , 3c 75 ( hudson et al . 2006 ; z=0.02 ) , and 0402 + 379 ( rodriguez et al . 2006 ; z=0.05 ) . a candidate sub - parsec binary system at z=0.38 has been recently discovered in the sdss through the detection of two emission line systems in the optical spectrum ( boroson & lauer 2009 ) , but this interpretation is still under debate ( e.g. , chornock et al . 2010 ) . once the smbh binary tightens , its merger is characterized by the emission of gravitational waves ( gw ) and the merged bh under favorable conditions ( of galaxy mass ratio and bh spins ) can recoil with respect to the center of the galaxy ( peres 1962 , bekenstein 1973 ) at substantial velocities ( few@xmath510@xmath6 - 10@xmath7 km / s , campanelli et al . if not enough mass is driven to the smbh binary before the merger , it can stall until the arrival of a third galaxy induces the hardening of the binary which ejects the just arrived bh for gravitational slingshot effect ( saslaw et al . 1974 ) . recently , observational searches were initiated for single quasars which are displaced from their host galaxy spatially or spectroscopically due to gw recoil after the merger ( e.g. , bonning et al . only a few candidates have been reported so far ( komossa et al . 2008 , shields et al . 2009 ) , all of them discovered through the detection of multiple line systems in optical spectra , but more prosaic interpretations for these candidates are not yet ruled out . the quasar he 0450 - 2958 is candidate to be a bh ejected by the nucleus of a companion galaxy during a major merger ( magain et al . 2005 , hoffman & loeb 2006 ) , but questions remain about the nature of this source ( e.g. , merritt et al . 2006 , kim et al . 2007 , jahnke et al . 2009a ) . a survey for binary smbh systems or the discovery of many recoiling bh would be helpful to test evolutionary merger models and to put more robust constraints on our understanding of bh and galaxy co - evolution ( e.g. , hopkins et al . 2008 ) . the cosmic evolution survey ( cosmos ; scoville et al . 2007a ) is a deep multiwavelength survey of 2 deg@xmath6 , based on a 590-orbit acs program ( koekemoer et al . 2007 ) , the largest mosaic ever made by the hst . of the @xmath8 galaxies to @xmath9 detected in the full field , the @xmath10-band ( f814w ) acs data provide morphological information down to @xmath11 for 70,000 galaxies and enables detailed structural analysis out to @xmath12 for @xmath116,000 galaxies . the large number of spectra ( @xmath1320,000 , lilly et al . 2007 , 2009 , trump et al . 2007 ) available for galaxies and quasars allow to search for sources with more than one line system , although the typical resolution available ( @xmath14 ) only allows sources with high velocity shifts to be found . the rich multiwavelength cosmos database allows detailed study of the source seds at all frequencies from radio to x - ray . these properties make cosmos an excellent database in which to look for sources just before or after the smbhs binary merging . according to the prediction of volonteri & madau ( 2008 ) , the number of detectable off - center agn ( up to 30 kpc away from the center of its host ) in the hst cosmos area should be @xmath130 , for the best case of large kicks ( spinning holes ) , long decay timescales ( no bulge ) , and long active phase , while considering those x - ray emitting the expected number is @xmath11 . in contrast , the short lifetime of binary bh at small separations implies a much lower number of sub - parsec candidates in the whole area , even at low redshift ( @xmath41/deg@xmath6 , volonteri et al . 2009 ) . smbhs are easier to find when they are active , and x - ray surveys are the most efficient means for finding agns ( e.g. , brandt & hasinger 2005 ) . hence , we started our survey for pre- and post - binary candidates with the optical counterparts of x - ray sources within cosmos . a visual inspection of the optical counterparts of all 2600 x - ray sources , detected in the -cosmos ( c - cosmos , elvis et al . 2009 ) and xmm - cosmos ( hasinger et al . 2007 ) surveys , using the hst / acs f814w images ( koekemoer et al . 2007 ; leauthaud et al . 2007 ) , led to the discovery of just one candidate recoil smbh in the source cxoc j100043.1 + 020637 ( cid-42 , elvis 2009 ) at z=0.359 . at the same time , comerford et al . ( 2009 ) reported the discovery of dual agns in the same source , through the detection of two emission line systems in a keck / deimos spectrum . we report here a detailed study of cid-42 , including the sed analysis and optical and x - ray imaging and spectroscopy . we discuss several models for the nature of this system . in this paper , magnitudes are reported in the ab system ( oke & gunn 1983 ) and a wmap 5-year cosmology with h@xmath15 = 71 km s@xmath16 mpc@xmath16 , @xmath17 = 0.26 and @xmath18=0.74 is assumed . the peculiar source is the optical counterpart of the x - ray source cid-42 ( elvis et al . 2009 ; civano et al . 2010 in prep . ) , also known as xmmu j100043.1 + 020636 ( cappelluti et al . 2009 ; brusa et al . 2007 , 2010 submitted ) . figure [ 3col ] shows a 30@xmath530 two color image , created by overlaying the 0.5 - 8 kev ( in purple color scale ) and acs f814w ( in yellow / white color scale ) images , around cid-42 . two sources ( white ) are clearly visible in the hst image . the galaxy shows structures to the ne and sw suggestive of tidal tails formed during a first galaxy - galaxy encounter ( e.g. , cox et al . 2006 ) . cid-42 is in the central area of the cosmos field and , for this reason , has been observed in all the cosmos bands , multiple times in some cases , allowing the construction of the fir - uv spectral energy distribution ( sed ) . in fig . [ sed ] ( left panel ) , the total rest - frame sed ( using data from frayer et al . 2009 , le floch et al . 2009 , sanders et al . 2007 , capak et al . 2007 , mccracken et al . 2010 , taniguchi et al . 2007 , zamojski et al . 2007 ) is compared with the template of a seyfert 1.8 and of the double nucleus prototype ngc 6240 ( dashed and dotted line , respectively ; polletta et al . 2007 ) normalized to the k - band luminosity . the elvis et al . ( 1994 ) quasar sed , normalized as it will be explained in section 3.1 , is also shown ( continuous red line ) . the radio emission is [email protected] mjy at 20 cm ( schinnerer et al . 2007 , bondi et al . 2008 ) , and the source is classified as radio - quiet using the wilkes & elvis ( 1987 ) definition ( @xmath19=-0.55 ) . although the optical spectrum shows strong emission lines typical of an agn ( see section [ spec ] ) , the optical sed is dominated by the galaxy component . the infrared luminosity is lower than that of ngc 6240 ( normalized in the k band ) but consistent with that of a luminous infrared galaxy ( [email protected]@xmath510@xmath21 l@xmath22 ) which could suggest ongoing star - formation activity as expected in a merger phase ( mihos & hernquist 1996 ) . however , nuclear activity could be contributing strongly to the ir emission , as shown from the comparison with the seyfert 1.8 sed . the good fit to the uv with the last sed suggests that the obscuring gas is not compton thick as in ngc 6240 . the bolometric luminosity estimated from the sed is l@xmath23=2.86@xmath510@xmath24 l@xmath22 . the stellar mass and star formation rate ( sfr ) of the host galaxy , derived from the fit to the _ u _ band to 4.5@xmath25 photometry ( to avoid the wavelengths where the agn contribution can be relevant ) using a set of 11 models with smooth star formation histories ( bruzual and charlot 2003 ; imf from chabrier 2003 ) , are m@xmath26 m@xmath22 and sfr@xmath1100 m@xmath22/yr ( bolzonella et al . 2009 ) . our estimated m@xmath27 and sfr are meant to be upper limits because the nuclear component can still contribute to the above quantities . given this mass and sfr , cid-42 is actively star - forming , with a sfr consistent with the overall x - ray selected agn population ( silverman et al . 2009 ) . the 5 year optical and near - ir light curve , as computed by salvato et al . ( 2009 ) , is shown in fig . [ sed ] ( top - right ) . no significant variability has been detected , as expected given that the optical sed is dominated by the host galaxy light . the source was imaged in the x - rays for 74 ks by and for 165 ks by acis - i . the data were taken over 2.5 years ( 8 observations ) and the data over 1.5 years ( 6 observations ) , for a total of 14 observations over 4 years . the reported exposure is the sum of all the effective exposure times . the luminosities derived in the two co - added sets of observations are l@xmath28=4.8@xmath510@xmath29 erg s@xmath16 and 1.9@xmath510@xmath29 erg s@xmath16 for and , respectively . the factor 2.5 difference between the two luminosities is too large to be explained by cross - calibration problems between the two instruments and implies significant x - ray variability . the full 4 year x - ray light curve ( fig . [ sed ] , right - middle ) shows a factor of 3 variability in the full ( 0.5 - 10 kev ) band flux , while in the soft ( 0.5 - 2 kev ) and hard ( 2 - 10 kev ) band the variability reaches a factor 4 and 4.5 , respectively . no significant variability in the spectral slope has been measured ( fig . [ sed ] , right - bottom ) . figure [ hst_fit ] ( left ) shows a 9@xmath59 zoom around source cid-42 in the acs / hst image ( 0.03/pixel ; filter f814w ) . the distance measured between the two optical centers is 0.495``@xmath00.005 '' ( 16.5 pixels ) which corresponds to a projected separation of [email protected] kpc at the source s redshift . we modeled the optical surface brightness of cid-42 with galfit ( peng et al . 2002 , version 3.0 ) . we use an empirical point - spread - function ( psf ) created by averaging stars in the acs frames ( jahnke et al . 2004 , 2009b ) to represent a point source in the field and to use for convolution of the parametric galaxy models . we tested several combinations and numbers of point sources and sersic profiles ( sersic 1968 ) with a free sersic - parameter @xmath30 . we also made use of two new features in galfit v3.0 , a fourier mode modification of isophotes , restricted in this case to f1 ( lopsidedness ) , and a mode to model the ne tidal arm . the goals of this exercise were to ( 1 ) test whether the two nuclei in the system are point sources , ( 2 ) extract reliable point source fluxes , and ( 3 ) put constraints on the relative fluxes of the galaxy components around the two sources . the best description of the system consists of one point source and three sersic components . the model and the residuals are shown in figure [ hst_fit ] ( middle and right ) . we find that , above the general background , only the south - eastern nucleus ( se ) is well modeled with a point source , while the north - western nucleus ( nw ) is extended though compact . this finding is independent of the exact number and geometry of large scale galaxy components added to the model . by this model , we estimate the point - like se nucleus to have @xmath31 mag ( square in fig . [ sed ] , left ) with an uncertainty of @xmath32 mag . the first sersic component corresponds to the nw nucleus and is very compact , with a scale length of @xmath101 ( @xmath1 0.5 kpc ) , smaller than the typical dimension of host galaxies of x - ray selected agn ( e.g. , schade et al . 2000 ) , and @xmath33 mag ( triangle in fig . [ sed ] , left ) . the second sersic component is a large spheroid ( r@xmath34=3 , @xmath115 kpc ) and the third sersic component describes part of the overall light distribution plus the spiral arm . the latter two components combined have @xmath35 mag and together describe the host galaxy light distribution , likely extending the profile of the nw source . the overall system brightness obtained with the fit is @xmath36 mag ( pentagon in fig . [ sed ] , left ) , in agreement with the photometry used for the sed . in order to test the goodness of the fit , we ran the galfit analysis with a simple model ( 1 point - like source and 1 sersic component ) , as was reported by gabor et al . this fitting runs into calculation boundaries , indicating that a 2 component model is not adequate to represent this complex system . a minimum of 3 components ( 1 psf and 2 sersic ) is required . instead , the best fit is obtained using 4 components , as explained above . given the point - like unresolved profile of the se source , it could be considered as a unobscured bh ( i.e. type 1 agn ) and it is possible to compute its bolometric luminosity , by normalizing to its f814w magnitude a typical type 1 agn sed ( elvis et al . 1994 ; red line in figure [ sed ] , left ) . this gives a bolometric luminosity for the se source of l@xmath37=2.8@xmath510@xmath38 erg s@xmath16 , consistent with being a moderately luminous agn . the merging system cid-42 is not an isolated source , as is evident in fig . [ 3col ] . using combined and data ( with prior point source removal ) , extended emission has been detected around cid-42 . the x - ray detected group ( finoguenov et al . 2007 , 2010 in prep . ) is spectroscopically confirmed with 7 galaxy members in the zcosmos bright catalog ( lilly et al . 2007 , 2009 ) and 30 members in the photometric redshift catalog ( ilbert et al . 2009 ) , of which 10 are visible in fig . the extended x - ray luminosity in the 0.1 - 2.4 kev band is 1.4@xmath510@xmath39 erg s@xmath16 . the x - ray centroid , though uncertain ( 30 positional error ) , is consistent with the position of cid-42 . the detection of extended x - ray emission from hot gas of the group allows us to estimate the mass of the halo in which the merging galaxy resides . the total gravitational mass of the group is m@xmath40= [email protected] @xmath510@xmath41 m@xmath22 is the mass enclosed in r@xmath40 , which is the radius within which the matter density is 200 times the critical density of the universe . ] , based on the average lx - m@xmath40 calibration of leauthaud et al . the corresponding r@xmath40 is 106 ( @xmath10.5 mpc ) , so cid-42 is located in the core of the group . the source cid-42 has been identified also as the second most massive group galaxy ( mmgg , details in leauthaud et al . furthermore , the galaxy group hosting cid-42 is embedded in a large over - dense region at [email protected] covering the whole cosmos field ( scoville et al . 2007b , gilli et al . 2009 ) . cid-42 was imaged with in 6 acis - i pointings with different exposure times ( 14 to 46 ks ) and off - axis angles ( 3-7.5 ) spread over 4 months . given the acis - i resolution ( 0.5" / pixel ) , it is not possible to perform an accurate decomposition analysis as performed with hst / acs data , but it is possible to study the emission in different energy bands . we analyzed the x - ray emission in both soft ( 0.5 - 2 kev ) and hard ( 2 - 7 kev ) band images , using the observation in which the source has been observed at the smallest off - axis angle ( @xmath1 3 ) and with the longest exposure time ( 46 ks ) . in this observation , the source has 187 net counts in the full band in a region corresponding to 95% of the encircled energy fraction for a point source at that position . most of the counts ( 123 out of 187 ) are detected in the 0.5 - 2 kev band . comparing to the psf which applies for this energy band , off - axis angle , and azimuth , there is some evidence for extension along the axis joining the two optical nuclei . to characterize this better we did a detailed ray - trace simulation using chart and marx with the assumption of equal flux from the two optical nuclei . the results are illustrated in fig . [ ds9 ] where we show the soft band x - ray image ( top right ) and the simulation ( lower right ) . the observed and simulated images are qualitatively consistent , indicating that there is no evidence for galaxy - scale x - ray diffuse emission in this source . one would like to usefully constrain the flux ratio of the two point sources , but this is not possible here given the psf extent and astrometric uncertainty at 3 off - axis . for completeness , we also simulated the 2 - 7 kev hard band data as coming from the two nuclear point sources and also in this case we found that the observed image ( fig . [ ds9 ] , lower left ) is consistent with the simulated image . line plotted in the observed frame . the components used to reproduce the spectral features in the zcosmos spectrum are shown at the bottom of the frame in red , as well as the spectral fit . the flux is in arbitrary unit . the emission lines used in the fit are labelled . the chip gap in the imacs spectrum is evident . the thin dashed lines represent the observed wavelength of the zcosmos h@xmath2 lines ( black ) and the predicted wavelength where the broad h@xmath2 should peak in the imacs spectrum ( red ) . the strong residuals at 6825 are due to the presence of an absorption telluric line . _ bottom _ : cid-42 imacs spectrum around the h@xmath42 line plotted in the observed frame . the line colors are as above . , scaledwidth=45.0% ] optical spectra are available from sdss ( adelman - mccarthy et al . 2008 ) , imacs / magellan ( trump et al . 2007 , 2009a ) and vimos / vlt ( zcosmos , lilly et al . 2007 , 2009 ) . the source has been observed on february 2005 with imacs and then twice with vimos on may and june 2005 . given the 1 width of the slits ( oriented n - s ) both in the vimos and imacs observations , and the small e - w separation of the sources ( @xmath10.3 ) , the flux from both falls within the slit . the 2d spectra have been analyzed , but none of them spatially resolve the two nuclei , while in comerford et al . ( 2009 ) the two nuclei seem to be resolved in a 2d deimos spectrum ; the spatial resolution of the deimos spectrum is similar to the imacs one ( imacs : 1 pixel=0.11 , deimos 1 pixel=0.11 , vimos 1 pixel=0.205 ) but its spectral resolution ( r@xmath13000 ) is higher . the two zcosmos spectra ( 5500 - 9500 ) have an average signal - to - noise ratio of 6 ( 11 around the h@xmath2 line ; fig . [ optspec ] ) and spectral resolution of r@xmath1600 . the imacs spectrum ( 5600 - 9200 , r@xmath1700 ) has a detector chip gap at 6450 - 6560 , cutting off part of the h@xmath2 line . the h@xmath42 emission line is in a region where fringing becomes strong in vimos . although h@xmath42 is visible in the imacs spectrum , which uses the nod and shuffle technique to avoid fringing , the peak of the line is saturated . for the above reasons , we started the spectral analysis by fitting the blue part of the zcosmos spectrum around the h@xmath2 and the [ oiii ] doublet . the fitting was performed using the iraf routine specfit ( kriss 1994 ) . we modeled the spectrum with a linear component for the continuum plus 4 gaussian lines , 1 broad and 3 narrow : the [ oiii ] doublets , the broad and narrow h@xmath2 . the central wavelength , width and intensity of the gaussians were left free to vary . the redshift measured from the peak of the narrow emission lines ( [ oiii ] and h@xmath2 ) is z=0.359 , coincident with the redshift listed in lilly et al . ( 2007 , 2009 ) and comerford et al . this redshift is also consistent with the one measured from absorption lines ( e.g. , h@xmath43 ) . the fwhm of the narrow emission lines , corrected for instrumental broadening , is 560@xmath0120 km / s . the 3 optical spectra show a typical type 1.5 ( osterbrock & ferland 2006 ) agn feature : a broad h@xmath2 component plus a narrow core that has a line width similar to the forbidden [ oiii ] lines . unusually though , the broad and narrow h@xmath2 components of cid-42 have widely offset peaks , potentially implying that the emission comes from different regions , consistent with the presence of two unresolved sources . the same offset is measured in both the zcosmos spectra and so , in order to have a better measurement of this offset , we summed them . we measure an offset @xmath44 30@xmath07 ( 1360@xmath0320 km / s ) between the centroids of the broad and narrow h@xmath2 . small residuals at the peak of the broad line possibly suggest the presence of another narrow component at the same wavelength that , when fitted , unfortunately has a flux below the s / n ratio of the spectrum . if the offset between the two components is due to the narrow lines being produced by one of the two optical sources and the broad line by the other , then an instrumental shift due to the spatial separation and the orientation of the sources with respect to the e - w vimos dispersion axis has to be considered . this effect shifts the broad component towards the red by 4 , which should be subtracted from the measured offset . the net offset ( 26@xmath07 ) implies a difference in velocity of @xmath11180@xmath0320 km / s between the h@xmath2 narrow and broad components , under this assumption . the 2d imacs spectrum ( see fig . [ bidim ] ) shows a clear shift of the peak of the broad line with respect to the narrow line , but does not show the spatial separation as seen in comerford et al . ( 2009 ) , despite the same pixel scale of the two instruments ( imacs and deimos ) , most likely due to different seeing conditions . the presence of a detector chip gap close to the h@xmath2 region in the imacs spectrum prevents a good fitting of the line profile . however , a narrow line is present with a broad peak redshifted from the narrow core . using the shift measured in the zcosmos spectrum , we predicted the wavelength where the broad line in imacs should peak . the instrumental shift due to position and slit orientation in the imacs spectrum would move the broad component towards the blue ( i.e. in the opposite direction to the zcosmos spectrum shift ) because of the opposite dispersion direction of the two instruments , so that the two peaks have a different observed offset with respect to zcosmos . the predicted offset in imacs is @xmath120 , which matches with the observed line peak ( red dashed line in fig.[optspec ] , top ) . if the lines are not coming from two different sources , the predicted offset in imacs would not correspond with the observed line peak , although it could be consistent within the errors . we performed further checks to test the possible presence of a second system of lines with the same offset of the broad h@xmath2 . the first would be the detection of a second [ oiii ] doublet with the same shift measured in the h@xmath2 but , at the expected wavelength , a telluric feature ( @xmath16825 ) prevents the detection of the corresponding redshifted [ oiii ] lines . the second test would be the detection of a redshifted broad h@xmath42 line . as explained above , h@xmath42 lies in a region dominated by fringing in the zcosmos spectrum , while in imacs the line peak is saturated . the fit of the h@xmath42 and [ nii ] doublet in the imacs spectrum returns 3 narrow lines at the same redshift of the narrow [ oiii ] doublet ( fig.[optspec ] , bottom ) . the broad h@xmath42 component peak is slightly offset ( @xmath1 5 ) to the red from the narrow one but this shift is not enough to be consistent with the shift predicted from h@xmath2 , even when the instrumental shift is taken into account . although a proper fitting of the continuum has been performed , a 1:2 ratio for the [ nii ] doublet flux is observed instead of the 1:3 ratio required by atomic physics . on the contrary , the expected 1:3 ratio has been measured in the [ oiii ] doublet flux in all spectra . this suggests that , around h@xmath42 , the imacs spectrum is not well behaved and so the measurement of the broad line peak is still uncertain . we also performed a simple two - component fit to the imacs spectrum to determine the relative nuclear and host contributions in the 4000 - 9000 range . following trump et al . ( 2009b ) , we find the combination of quasar ( vanden berk et al . 2001 ) and galaxy ( eisenstein et al . 2001 ) which maximizes the bayesian probability function and best describes the spectrum . the estimated nuclear to host ratio is @xmath13:2 . systematic errors , arising from neglecting the contribution from other components , could affect this estimated ratio . however , this simple fit shows that the spectrum is not consistent with a simple blue quasar , but instead includes a significant host contribution ( consistent with the galfit decomposition of section 3.1 ) . the significant host contribution may also be part of the reason for which the fitting around the h@xmath42 region returns different results from those obtained for h@xmath2 and [ oiii ] . the x - ray data available ( @xmath1 800 counts in acis and @xmath1 2200 counts in the pn ) allow us to perform good quality x - ray spectral analysis when pn and acis spectra are fitted together . the results are reported in table [ xray ] . due to the modest number of counts , we used the cash statistic ( cash 1979 ) . a fit with an absorbed power law returns a spectral index @[email protected] , with no cold intrinsic absorption , in good agreement with the results of mainieri et al . ( 2007 , @[email protected] ) . residuals in the very soft energy band ( @xmath40.5 kev , mostly in the spectrum ) suggest the presence of a thermal component ( @[email protected] kev ) that accounts for 8%-9% of the 0.3 - 2 kev luminosity , which could be due to either a nuclear soft - excess or hot gas in the galaxy . the addition of the soft component to the fit produces a slightly flatter spectral slope with @xmath45=1.95@xmath47 . the most striking features of the x - ray spectrum are around 4.5 kev ( observed frame , fig . [ xspec ] , left ) . the residuals suggest the presence of both emission and absorption features in both spectra , forming a kind of _ inverted _ p - cygni profile , i.e. , the absorption component is _ redshifted _ with respect to the emission component ( fig . [ xspec ] , right ) . the emission feature , which is consistent with being a neutral iron line , has a constant flux and is more prominent in the spectrum ( ew=570@xmath0260 ev , in the rest frame ) , where the continuum is fainter , than in the one ( ew=142@xmath48 ev , in the rest frame ) , where the continuum is 2 times higher . the absorption feature instead shows strong variability between and data , both in flux and width : the line is well fitted with a broad ( @xmath49=0.22@xmath50 kev , 1@xmath49 error ) gaussian line in the spectrum , where it is more prominent , while it is unresolved in the data ( @xmath510.5 kev ) . assuming z=0.359 , the central rest - frame energy of the absorption line is around 6 kev and therefore it seems to be a _ redshifted _ absorption iron line . the shape and intensity of the background spectrum have been studied in order to discard the possibility for the absorption line to be a background artifact . the fact that the absorption line has been detected in spectra from 2 different instruments ( epic and acis ) , already rejects this hypothesis . in order to check if the _ inverted _ p - cygni feature can be mimicked by other spectral models , we also performed the spectral fitting using more complex models , a broken power - law and a partial covering model . the two models reproduce the continuum better than a single power - law , given the larger number of parameters , but none of them reproduces the absorption and emission features . in particular , the edge around 5 - 6 kev in the partial covering model does not reproduce the absorption feature . the @xmath3cash statistic obtained adding the two gaussian lines to the models is the same as that obtained with a single power - law , meaning that the absorption feature is real and independent from the model adopted to reproduce the continuum . we note that the partial covering model returns n@xmath5210@xmath53 @xmath54 and a covering fraction of 40% , which can be translated as the sum of two power - law of almost the same intensity , one obscured and the other not . a spectral fitting with a photo - ionized gas model to the total spectrum has also been performed . unfortunately , given the low ccd resolution , it is not possible to distinguish single absorption lines in the soft band . the equivalent width of the absorption feature with a photoionized - gas model implies a high column density of the absorber ( n@xmath525@xmath5 10@xmath53@xmath54 ) , and the modest absorption measured in the soft band ( [email protected]@xmath5 10@xmath56@xmath54 in the rest frame ) requires the absorber to be highly ionized ( log @xmath573 ) . from the centroid of the line , we measure a velocity of the outflow of 30000@xmath04000 km / s . in order to check the nature of the broadening of the absorption line , we analyzed the spectra individually . given that the data come from 8 observations taken over three years , variability may be responsible for the broadening . nine ( 5 and 4 ) out of the 14 available observations have enough counts ( @xmath13125 ) to perform an individual spectral fit . the absorption feature is present at 2 to 3 @xmath49 in all the spectra , apart from 1 observation where , although the number of counts is enough for the fitting ( 187 net counts ) , the continuum flux is faint and thus the absorption line is not detected . the intensity ( with 1 @xmath49 error ) , width and rest - frame energy centroid of the line measured in the 8 spectra are plotted in figure [ lc ] ( red= , black= ) . in contrast to the summed spectrum , the absorption features show a narrow profile , with only upper limits on the width in all the spectra ( @xmath51 250 ev ) and the ew is consistent with being constant , implying a column density of gas of the order of @xmath15@xmath510@xmath53@xmath54 ( bianchi et al . 2005 ) . the line energy centroid in the spectra changes in time over the range 5.9 - 6.3 kev , while in the spectra it is almost the same , within the errors . this variation explains the broadening measured in the total spectrum , as due to the superposition of narrow lines with different energy centroids . the variability of the energy centroid is of @xmath3e=+500 ev in the first 2 years and @xmath3e=-400 ev in the last year . these shifts are comparable to the 400 ev separation of the fei and the fexxvi k@xmath42 absorption lines , so a change in ionization could be in principle be responsible for the observed shifts . we analyzed the hardness ratio in all the spectra to check for column density variability to be related with variability of the ionization state , but , given the large error bars , we do not detect statistically relevant variability . instead , if this variation is converted in a pure velocity variation , it implies a range in velocity of 0.02 - 0.07@xmath58 for neutral iron ( fe i ) , with a deceleration of 10,000 km / s / yr in the first two years and a faster acceleration in the last year of observations . if the lines are due to completely ionized iron ( fe xxvi ) absorption , the velocity is higher ( 0.09 - 0.14@xmath58 ) and the implied acceleration is slightly lower . because of the degeneracy between velocity and ionization state , the same conclusions on the absorber parameters ( density , ionization state and velocity ) are obtained doing the single line analysis and the fitting with a photo - ionized model . .,scaledwidth=50.0% ] ccc + & & + + continuum & & + mekal kt ( kev ) & @xmath4 0.13 & @xmath4 0.13 + mekal f@xmath59 ( erg s@xmath60 ) & @xmath61 & 6.4@xmath62 @xmath63 + photon index @xmath45 & @xmath64 & @xmath64 + n@xmath65 ( @xmath66 ) & @xmath40.02 & @xmath40.02 + emission line & & + obs energy ( kev ) & 6.44@xmath67 & 6.60@xmath68 + line @xmath49 ( kev ) & @xmath40.57 & @xmath40.20 + flux ( erg s@xmath60 ) & 1.50@xmath62 @xmath69 & 1.35@xmath62 @xmath70 + ew ( ev ) & 570 @xmath0260 & 142@xmath48 + absorption line & & + obs energy ( kev ) & 5.86@xmath71 & 6.01@xmath72 + line @xmath49 ( kev ) & @xmath40.50 & 0.22@xmath73 + flux ( erg s@xmath60 ) & -5.68@xmath74 @xmath75 & -4.07@xmath62 @xmath76 + ew ( ev ) & -200 @xmath77 & -441 @xmath78 + flux and luminosity ( @xmath79 ) & & + total f@xmath80 & 4.49@xmath81 @xmath82 & 10.90@xmath81 @xmath83 + total l@xmath80 & 1.93@xmath84 & 4.85@xmath84 + total f@xmath85 & 1.89@xmath81 @xmath86 & 4.80@xmath81 @xmath87 + total l@xmath85 & 7.90@xmath88 & 2.18@xmath84 + total f@xmath89 & 2.60@xmath81 @xmath90 & 6.10@xmath81 @xmath91 + total l@xmath89 & 1.14@xmath84 & 2.66@xmath84 + we report in this paper on the properties of a peculiar and unique source at z=0.359 , cid-42 , in a survey of x - ray sources with acs counterparts in cosmos . cid-42 is a unique source in three specific ways . first , in the 2 deg@xmath6 cosmos area , there are @xmath14 x - ray sources out of @xmath1550 at [email protected] with optical morphologies indicative of ongoing major mergers , but only one of them ( cid-42 ) clearly shows two cores resolved in the hst image embedded in the same galaxy but unresolved in x - rays . the surface brightness decomposition applied to the acs image strongly suggests the presence of a point like unresolved source ( se source ) and a more extended , but still compact , source ( nw source ) . the tidal tails to the ne and sw , in the acs image , are evidence of merging ( toomre & toomre 1972 , dubinski et al . 1999 , springel & white 1999 ) . the morphology is consistent with either a early passage state , in a phase subsequent to the first passage , or with a later stage of the merging after the coalescence of the two bhs ( mihos & hernquist 1996 , cox et al . 2006 , hopkins et al . 2008 ) . second , in 3 optical spectra obtained with two different instruments ( vimos and imacs ) a velocity offset larger than 1000 km / s is measured between the broad and narrow component of h@xmath2 line . large velocity offsets between two different line systems can be observed if two bhs in a virialized system are rapidly rotating very close to each other ( sub - parsec separation ) , as suggested for the double line system of boroson & lauer ( 2009 ) . however , assuming that the broad and narrow lines are coming from the se and nw sources respectively , virialization is not appropriate for the kiloparsec scale separation of the two sources in cid-42 . a possibility to justify the large velocity offset is a recoil effect . large recoiling velocity have been predicted by theoretical models and simulations for both gw wave kicks ( campanelli et al . 2007a , b ) and slingshot effect kicks in triple bh systems ( hoffman & loeb 2007 ) . the unresolved morphology of the se source allows us to assume that it plays the role of the unobscured smbh ( a type 1 agn ) and produces the broad h@xmath2 line . therefore the se source is recoiling with respect to the system with a velocity of @xmath131000 km / s . as the bh recoils from the center of a galaxy or from another bh , the closest , most tightly bound , regions ( disk and blrs ) are carried with it , and the more distant regions ( nlrs ) are left behind ( loeb 2007 ) . given the distance between the recoiling bh and the center of the whole system ( measured on the plane of the sky ) and one component of the bh velocity ( the radial velocity ) with respect to the galaxy , it is possible to compute a lower limit on the time since the bh has been kicked of @xmath14@xmath510@xmath92 yr . the detection of a recoiling bh is related to its lifetime and it is extremely rare because it never gets large separation before fading . the lifetime of the active phase of the bh depends on the amount of accreting material it carries after the kick . in principle , higher velocities result in a smaller ejected disc . in cid-42 , the bh is still emitting enough radiation to ionize the blr , so it had to carry enough material in order to be still active . it is possible to compute the bh mass of se source using the relation of vestergaard & peterson ( 2006 ) between the line width , the bh mass and the continuum luminosity , by using the h@xmath2 line width ( 3500@xmath0233 km / s ) . as shown in figure [ sed ] , the total continuum luminosity is mostly dominated by the galaxy light . taking advantage of the surface brightness decomposition results , we can extrapolate the 5100 nuclear luminosity of @xmath14@xmath510@xmath29 erg s@xmath16by normalizing a standard quasar template ( red line in fig . [ sed ] , elvis et al . 1994 ) to the se source luminosity in the acs filter ( square in fig . [ sed ] ) . under this assumption , the bh mass of the se source is @xmath16.5@xmath510@xmath93 m@xmath22 . given the total mass of the galaxy and the estimated bh mass , the ratio m@xmath94/m@xmath95 is a factor 2 higher than the hring & rix ( 2004 ) m@xmath94-m@xmath95 relation for local spheroids . although the errors on the bh and galaxy mass are quite large , this is consistent with the fact that cid-42 is a peculiar source or that the local relation deviates at high z ( see for example treu et al . 2007 and merloni et al . 2010 ) . if we consider that the galaxy mass is an upper limit to the bulge mass , the offset from the relation could be even bigger . using the bolometric luminosity and bh mass of the se source , we find that the eddington ratio is l / [email protected] , consistent either with the nuclear activity of this source having just been triggered or already declining . using the correlation between the eddington ratio and the bolometric correction in the 2 - 10 kev energy range ( vasudevan & fabian 2009 , young et al . 2010 ) , it is possible to estimate the 2 - 10 kev luminosity emitted by the se source to be in the range 0.6 - 1@xmath510@xmath29 erg s@xmath16 , consistent with the 2 - 10 kev luminosity measured for cid-42 ( see table [ xray ] ) . the third feature which makes cid-42 unique is the detection , in the x - ray spectra , of an extraordinarily strong , _ redshifted _ absorption iron line and an emission iron line , forming an _ inverted _ p - cygni profile . out of the 100 x - ray sources with more than 1000 counts in the stacked x - ray spectrum only cid-42 shows an iron line in absorption ( mainieri et al . 2007 , lanzuisi et al . 2010 in prep . ) . the absorption line is intrinsically narrow . its energy centroid changes by 500 ev over the 4 years of observations . the broadening of the absorption line in the total spectrum is the result of the stacking of narrow lines redshifted by different velocities in the different observations . such _ redshifted _ absorption lines are rare . most known x - ray absorbers in agns are blueshifted ( see cappi 2006 for a review ) and 85% of them are at low redshift ( [email protected] ) . as blueshifted absorbers necessarily imply out - flowing winds , redshifted absorbers would seem to require high velocity inflows , which must therefore be situated very close to the central black hole . the few convincing cases of redshifted lines are all narrow absorption lines detected in seyfert 1 sources ( mrk 335 , longinotti et al . 2007 ; pg 1211 + 143 , reeves et al . 2005 ; mrk 509 , dadina et al . 2005 ) , and have suggested explanation as gas falling with relativistic velocity directly into the bh at a few tens of gravitational radii . the cluster of rare features seen in this source , more likely associated with a brief moment of the merging event , is a strong argument against the chance superposition hypothesis . any model for the system should preferably link these unique features . we describe two possible models below . the spatial offset of the se point - like source with respect to the nw source and the kinematic offset of the broad h@xmath2 line with respect to the narrow line system can be explained by a smbh ejected by gravitational radiation produced in the black hole merger . according to theoretical models , during the final coalescence of two bhs in a merger event , the gravitational waves produced can impart a large kick to the resultant bh , which recoils in the opposite direction with respect to the center of the galaxy ( peres 1962 ) . although some approximate methods indicated that the kick velocities are most likely small ( @xmath4few@xmath5100 km / s , blanchet et al . 2005 , damour & gopakumar 2006 ) , the results from full numerical relativity simulations show that the recoil speed can be up to 4000 km / s ( campanelli et al . 2007a , b ) . it is difficult to estimate how often significant recoil kicks occur , because the recoil kick distribution depends on the spin and mass ratio distributions of the progenitor bh binaries , which are not well known . several authors ( schnittman & buonanno 2007 , campanelli et al . 2007a , b and baker et al . 2008 ) estimated , using results from full numerical relativity simulations , that 12 - 36% of recoiling bhs will have velocities @xmath13 500 km / s and 3 - 13% @xmath97 1000 km / s , assuming a spin of 0.9 with random orientations and using different assumptions on how the kick velocity scales with bh progenitor mass ratio ( 0.1@xmath981 ) . these fractions might be smaller if bh spins are on average lower or if gas torques are efficient at aligning the spin axes in a way that is unfavorable for large kicks ( bogdanovic et al . 2007 , dotti et al . 2009 ) . to constrain the actual recoil kick distribution , better observational constraints on merging bhs or observations of individual recoiling bhs are required . a candidate recoiling smbh with a large kick velocity of 2650 km / s has been spectroscopically discovered by komossa et al . ( 2008 ) , although alternative models have been proposed to explain the two systems of lines in the optical spectrum . a second possible candidate has been also proposed by shields et al . ( 2009 ) . the high velocity offset ( v@xmath131000 km / s ) measured in the optical spectra of cid-42 is consistent with the numbers obtained in simulations , although the probability of finding such high velocity recoiling bh is low ( @xmath410% ) . the surface brightness decomposition of the cid-42 acs image is consistent with the ejected bh hypothesis too : the se point - like source , the ejected bh , is spatially separated from the compact , but not point - like , nw bright source , which can easily represent the galaxy core from which the bh has been ejected . the fact that the broad lines are redshifted with respect to the stellar absorption lines is consistent with the recoiling bh picture , because they are following the bh , as explained at the beginning of section 5 . the narrow lines in the spectra have the same redshift of the stellar absorption lines but , given that the nlrs , residing in the nw source , are not powered anymore by the bh and will fade long before the ejected bh reaches kpc distances , another way to produce the narrow lines is needed . the line flux ratios indicate that the emission is produced by nuclear activity and not by star formation ( comerford et al . the narrow lines could be produced by the ejected bh ionizing the local galaxy ism on its way out from the center . in this case , the narrow lines should have the same redshift of the galaxy , as observed , implying that the bh is still inside a region of the galaxy disk with high ism density or is passing by molecular clouds . in this case , the velocity offset measured in the optical spectrum , not corrected for instrumental shift ( see section 4.1 ) , is 1360@xmath0320 km / s , which represents the radial component of the total velocity . lousto et al . ( 2009 ) showed in a statistical studies of black - hole binaries during the inspiral and merger , that when the kick velocity is @xmath991000 km / s , the direction of the kick is most likely perpendicular to the bh orbital plane . they found a probability of 6.3% for large recoil velocities ( @xmath131000 km / s ) along the line of sight . if the accretion disk is aligned with the nearly face - on galaxy disk as a whole , then a large kick closely aligned with the line of sight is expected , and it is consistent with what observed in cid-42 , if the source is a recoiling bh . the time since the bh has been kicked ( @xmath14@xmath510@xmath92 yr ) , as computed at the beginning of section 5 , is in agreement with the prediction of the effective time for the persistence of off - center quasar activity ( 10@xmath100 yr , blecha & loeb 2008 ) . this number is a lower limit but it should be a good estimate , assuming that the total kick velocity is not much higher than the radial velocity , given that this is already in the extreme tail of the kick velocity distribution predicted by models and simulations . according to blecha & loeb ( 2008 ) model , for a kick velocity in the range 1100 - 1500 km / s and a bh mass of 10@xmath93 - 10@xmath101 m@xmath22 , the mass of the bh disc , in order to be still emitting , is @xmath12 - 3% of the bh mass . models predict that the effect on galaxy morphology of bh recoil is to produce significant asymmetry ( kornreich & lovelace 2008 ) . this is seen in the morphology of cid-42 where the galaxy core ( the nw source ) is slightly misplaced with respect to the geometric center of the galaxy disc . the variable fe - k absorption line can be alternatively explained , as already explained at the beginning of this section , by material inflowing into the bh , as it has been proposed in a few other sources ( cappi 2006 , see section 5 ) . in order to have such a high velocity , the inflow of material would have to be very close to the nucleus ( few tens of schwarzschild radii ) . one means of achieving high inflow velocity is with a photon bubble instability ( arons 1992 , gammie 1998 , begelman 2006 ) which produces a propagating pattern of low and high density regions through which the radiation has to go . the photon bubble instability is currently used to explain the _ inverted _ p - cygni profiles detected in the spectra of luminous blue variable ( lbv ) stars ( e.g. , van marle et al . although this would be a physical explanation , it requires super - eddington accretion , that is not observed in cid-42 . although the high velocity inflow possibility is very interesting , without a detailed modeling it remains mere hypotheses . in the following we list the evidences in favor of a picture in which cid-42 contains 2 agns , an unobscured type 1 agn in the se source and an obscured type 2 agn in the nw one . 1 . the surface brightness decomposition of the acs image finds a compact , but not point - like , bright source in the nw nucleus . 2 . the shift ( @xmath1021180 @xmath0360 km / s ) between the broad and narrow components of the h@xmath2 line in the optical spectrum also suggests the presence of two active nuclei , one responsible for the broad line emission ( i.e. a type 1 agn ) , the second responsible for the narrow emission lines only ( i.e. a type 2 agn ) . a possible explanation for the presence of only one narrow emission line system could be that the narrow line regions ( nlrs ) of the two sources are mixed up , and , for this reason , are at the same redshift measured for the galaxy from the absorption lines . however , we discard this hypothesis as nlrs , at the luminosity of cid-42 , are expected to extend 0.1 - 1 kpc from the nucleus ( schmitt et al . 2003 ) , much less than the projected separation measured between cid-42 nuclei . 3 . the x - ray flux variability , and the presence of a strong iron emission line ( ew@xmath1500 ev ) when the continuum is weak also suggest a combination of a variable type 1 agn and a constant obscured type 2 agn . usually , strong iron emission lines ( ew@xmath11kev ) are associated with heavily obscured ( compton thick ) sources and are less prominent in unobscured ones ( ew@xmath1150 ev ; e.g. , see figure 9 of guainazzi et al . 2005a ) . as already noted by guainazzi et al . ( 2005b ) , all the agn pairs observed so far are found to be heavily obscured or compton thick in the x - rays , in both components . the presence of at least one obscured agn in our system is consistent with this picture but the possibility that we have a type 2 plus a type 1 agn embedded in the same galaxy is new , although we believe cid-42 is not a binary . assuming that the [ oiii ] emission lines are produced by the type 2 agn only , it is possible to roughly estimate the contribution of the two sources to the total x - ray luminosity . the type 2 luminosity is estimated from the relation between the l@xmath103-l@xmath104}$ ] measured in the cosmos field ( silverman et al . 2009 ) and the type 1 agn luminosity is then found by subtraction . given the [ oiii ] luminosity ( @xmath13@xmath510@xmath105 erg s@xmath16 ) and the uncertainties on this relation , the luminosity of the nw type 2 nucleus would be @xmath19@xmath510@xmath39 erg s@xmath16 in the 2 - 10 kev band . the derived type 1 agn luminosity ( 2 - 10 kev ) is in the range 0.3 - 1.8@xmath510@xmath29 erg s@xmath16 , due to the strong measured variability ( fig . [ sed ] , right ) . the last number is consistent with the 2 - 10 kev luminosity obtained in the previous section with the eddington ratio . as explained at the beginning of section 5 , the large velocity shift between the two nuclei implied by the optical spectra would require a large mass of the system assuming it is virialized ( i.e. the two nuclei are gravitationally bound in a circular orbit ) . comerford et al . ( 2009 ) detected two sets of [ oiii ] and h@xmath42 lines in a keck / deimos spectrum , from which a @xmath3v=150 km / s is measured by detecting two sets of emission line . this lower velocity shift allows a more `` reasonable '' mass for a virialized system , however , virialization , which is a fair assumption for close pairs , is probably not appropriate for the kiloparsec scale separation of the two cid-42 sources . most likely , the velocity measured by comerford et al . ( 2009 ) is due to the rotation of the gas in the galaxy as a whole , as the system is in a violent state of gravitational interaction . the large velocity offset between the two source is more likely due to a nearly radial orbit , with the type 1 se agn is recoiling with respect to the type 2 agn . a velocity difference of this size may result from a slingshot ejection in a triple bh encounter ( saslaw et al . 1974 , hoffman & loeb 2006 , 2007 ) . if a bh binary stalls for a time of @xmath1 few gyrs before it coalesces , due to the depletion of stars in the loss cone in a gas poor merger ( merritt & milosavljevic 2005 ) , its host galaxy could merge with another galaxy and a third bh may spiral in and undergo a strong 3-body interaction with the binary . the just arrived system will merge with the original galaxy . the star formation rate ( @xmath1100 m@xmath22/yr ) measured from the sed fitting , supports the scenario of a merger in a early phase . the merger will drive mass and gas to the binary , refilling its loss cone . the configuration will become gravitationally unstable and this event will end with the further reduction of the separation or the coalescence of the binary and the ejection of the lightest body at a speed comparable to the binary s orbital speed ( heggie 1975 ) . the large projected distance ( 2.4 kpc ) between the two sources and the high velocity measured in in cid-42 are possible for this kind of ejection ( hoffman & loeb 2006 , 2007 ) , although the probability to have a certain velocity depends from the initial condition of the system and a proper modeling is still undergoing . on the basis of the above analysis , the se nucleus is the unobscured type 1 recoiling agn , while the nw is an obscured type 2 agn , resulting from the coalescence of a binary . the two nuclei lie close to the plane of the sky possibly moving on a radial orbit in a direction close to the line of sight , though a small angle has to be considered given the 2.4 kpc separation . relative to the mean velocity of the system , the se nucleus is moving away from the observer and the nw one towards the observer . the suggested geometry of the system is shown in figure [ model ] . the se type 1 nucleus then has : a strong unobscured optical continuum and blr , leading to the point - like hst / acs image , the broad h@xmath2 emission line and a strong , variable , x - ray continuum emission . the obscured nw type 2 nucleus has : a weak optical continuum , leading to a compact but extended hst / acs source dominated by the galaxy light , only narrow emission lines of h@xmath2 and [ oiii ] , and a weak x - ray continuum , with a strong fluorescent iron emission line ( ew@xmath106 570 ev ) . the proximity of the two sources allows a special interpretation of the fe - k absorption line as a `` backlit wind '' . in this scenario , a fast , highly ionized bal - like wind is emitted by the nw type 2 nucleus ( located in front ) . in almost all circumstances this wind would be undetectable . however , in the case of cid-42 , the _ receding _ wind crosses our line of sight to the se type 1 nucleus , leading to absorption of the type 1 continuum in redshifted iron . the blueshifted emission of the approaching wind remains invisible as there is no background source to illuminate it . the column density of the flow derived from the ew of the absorption line is consistent with what is seen in bal wind ( @xmath1310@xmath53 @xmath54 ; green et al . 2001 , gallagher et al . 2002 , chartas et al . the fact that we measure no x - ray photoelectric absorption at low energies implies that the wind is highly ionized , something observed also in other bal ( e.g. , telfer et al . 1998 , hamann 1998 , chartas et al . 2009 ) and x - ray absorbers ( e.g. , vignali et al . 2000 , reeves et al . 2009 ) . the variations of the iron absorption line energy centroid on years timescale can be explained either by velocity variations ( @xmath110,000 km / s per year ) in the flow , or by a change in the ionization state , and so in the density ( proga et al . 2000 ) or , possibly , to a precession of the flow , so that a change in inclination can produce a difference in velocity , or by a combination of all these effects . bals , where we look directly along the flow , are known to vary on year timescales ( e.g. , lundgren et al . 2007 , gibson et al . 2008 ) , so variability should be more likely in a section viewed across the flow . the variability implies that the bal - like flow must be highly structured , even 2.5 kpc away from the nucleus . this is not unusual as it has been shown by detection of radio jets still relativistic at mpc distance from their origin ( e.g. , pks 0637 - 752 , schwartz et al . 2000 ) . as we have only one example of _ redshifted _ iron absorber and only one recoiling bh , this scenario is only plausible if the bal - like wind has a large opening angle , making an interception of the rear recoiling type 1 nucleus likely . winds at a few 1000 km / s are normal in type 1 agns ( reynolds 1997 , piconcelli et al . 2005 ) and much faster ( 0.1 - 0.2@xmath58 ) bal winds , confined to modest solid angles , are not unusual ( weymann et al . 1991 , hamann et al . 1993 , ogle et al . fast winds at kpc - scale distances from type 1 agn have been found as well ( arav et al . 2001 ) and have been proposed to be universal in agns ( elvis 2000 ) . however , there has been no possibility to test this idea in type 2 agns . cid-42 provides this opportunity and gives us the first detection of a wind in a type 2 agn . we have reported the multiwavelength properties of cid-42 , which presents three unusual features : two close sources , embedded in the same galaxy , resolved in the optical acs image , but unresolved in the x - ray one , a high velocity offset , measured between the broad and narrow h@xmath2 lines in 3 different optical spectra and a _ redshifted _ absorption iron line in the x - ray spectrum . thanks to the rich database of the cosmos survey , the analysis of the properties of cid-42 has been performed at all wavelengths . the overall analysis suggests two possible explanations for the source : a gw recoiling bh , caught @xmath1few myr after the kick from the center of the galaxy or a type 1/type 2 agn system where the type 1 is recoiling due to a slingshot effect . analysis of the acs image shows the presence of a se point - like source , playing the role of the type 1 agn ( or smbh with disc and blrs ) in both scenarios , confirmed by the x - ray flux variability across the time and by the offset in velocity measured between the narrow and the broad components of the h@xmath2 line . the nw source , being less compact than the se one , could be either the naked core of the galaxy from which a bh has been kicked out or a type 2 agn . the presence of a _ redshifted _ absorption iron line changing its energy centroid in the different observations , allows to explain the geometry of the system by inflow of material into the bh in the gw recoil bh case or with the `` backlit wind '' model in the second scenario . monitoring observations in the x - ray would be suitable to study the moving x - ray absorption feature and its variability . the `` backlit wind '' implies that fast bal - like winds are present in type 2 agns , an otherwise untestable hypothesis . the variability of the iron absorption is , in this special case , a new tool to study bal flows . this model predicts corresponding high ionization uv absorption lines ( e.g. ovi ) , making cid-42 a suitable target for hst / cos observations . high - resolution observations with hrc could resolve the two sources to see if both are x - ray emitting or not , while optical and ir spectroscopy with higher spatial resolution , which currently is only possible with hst , could confirm the still uncertain velocity shift measured in h@xmath2 in other lines . if future observations allow us to confirm that the gw recoiling bh is the best explanation of the system , it will be the first time in which this phenomenon is observed via both spatially offset from the galactic core and a velocity offset in the optical spectrum . although models for this system have been sketched qualitatively in this paper , a numerical analysis will be the subject of a forthcoming paper . upcoming keck integral field unit ( ifu ) observations with osiris will help us to disentangle major competing models mapping the inner region of cid-42 . ifu observations on larger scales could map the velocity of gas and stars in the galaxy and in the tails , in order to constrain the theoretical models on the geometry of the merger and the inclination of the galaxies in the first encounter ( donghia et al . in prep . ) . calibrating the number of recoiling bh , produced either in close triple encounters and by gravitational waves ejection could be of great importance for the proposed laser interferometer space antenna ( lisa ) mission ( see discussion in loeb 2009 ) . cid-42 is a unique source with a cluster of rare features . whichever is the best explanation , in cid-42 we are witnessing a runaway bh fc thanks e. costantini , t.j . cox , m. dotti , l. hernquist and a. sesana for useful discussion . the authors thanks chien y. peng for useful discussions about galfit . the authors thank the anonymous referee , whose critical analysis helped to improve this paper , making it more interesting . this work was supported in part by nasa _ chandra _ grant number go7 - 8136a ( fc , me , af ) , nasa contract nas8 - 39073 ( chandra x - ray center ) . kj acknowledges support from the emmy noether programme of the german science foundation ( dfg ) through grant number ja 1114/3 - 1 . in italy this work is supported by asi / inaf contracts i/023/05/0 , i/024/05/0 and i/088/06 , by prin / mur grant 2006 - 02 - 5203 . in germany this project is supported by the bundesministerium fr bildung und forschung / deutsches zentrum fr luft und raumfahrt and the max planck society .

we quantify the expected observed statistical and systematic uncertainties of the escape velocity as a measure of the gravitational potential and total mass of galaxy clusters . we focus our attention on low redshift ( z ) clusters , where large and deep spectroscopic datasets currently exist . utilizing a suite of millennium simulation semi - analytic galaxy catalogs , we find that the dynamical mass , as traced by either the virial relation or the escape velocity , is robust to variations in how dynamical friction is applied to `` orphan '' galaxies in the mock catalogs ( i.e. , those galaxies whose dark matter halos have fallen below the resolution limit ) . we find that the caustic technique recovers the known halo masses ( ) with a third less scatter compared to the virial masses . the bias we measure increases quickly as the number of galaxies used decreases . for n , the scatter in the escape velocity mass is dominated by projections along the line - of - sight . algorithmic uncertainties from the determination of the projected escape velocity profile are negligible . we quantify how target selection based on magnitude , color , and projected radial separation can induce small additional biases into the escape velocity masses . using , the caustic technique has a per cluster scatter in of and bias for clusters with masses at .

1307.0017

upcoming galaxy cluster surveys from the millimeter to the x - ray wavelengths have the potential to identify thousands to hundreds of thousands of groups and clusters @xcite . these large cluster samples will be used to constrain the cosmological parameters which govern the growth of structure in our universe @xcite . regardless of the luminous tracer used to infer the underlying dark matter distribution ( e.g. , normal galaxies , emission - line galaxies , luminous red galaxies , quasi - stellar objects , galaxy clusters ) , our ability to constrain the cosmological parameters depends on the accuracy and precision of the mass . unlike most galaxies , the masses of galaxy clusters can be directly inferred via the observational signature of the gravitational potential ( or its derivative ) . in fact , clusters are the only object for which current technology enable three physically independent mass estimation techniques : via the dynamics of the member galaxies , via the hot gas in the intra - cluster medium , and via gravitational lensing . these three techniques provide a vital cross - check on the mass estimation techniques , assuming one can quantify the statistical precision and accuracy of the cluster mass estimates themselves . cosmological n - body and hydrodynamic simulations play a new and important role in characterizing the statistical and systematic uncertainties on cluster mass estimates @xcite . there has been excellent recent progress on the important step of utilizing realistic mock astronomical observations based on ideal simulations @xcite . our primary goal is to utilize a diverse suite of semi - analytic galaxy catalogs to study how well a realistic spectroscopic program can constrain the dynamical masses of galaxy clusters in the low redshift universe . we quantify our results based on the scatter and bias of the `` observed '' dynamical mass when compared to the halo masses m@xmath10 , which refers to the mass within a radius of @xmath11 . when inferring the dynamical masses of clusters , we require an accurate measurement of the galaxy peculiar velocities . with their total dispersion , their dispersion profiles , and/or their escape velocity profile , we can infer cluster masses based on the virial theorem , the jeans relation for a collision - less fluid , or the caustic technique . to first order , galaxies dynamically respond to the influence of the newtonian gravitational potential , regardless of their luminosities , shapes , colors , or star - formation histories . yet galaxies traveling within a cluster environment are likely to have had one or more localized and short - lived gravitational interaction over its lifetime . this `` dynamical friction '' alters the total cluster velocity distribution away from its simple newtonian expectations , an effect that needs to be captured by the semi - analytic galaxies . in this work , we use sub - halo catalogs as well as a suite of semi - analytic mock galaxy catalogs , to explore how sensitive dynamical masses are to different prescriptions of this dynamical friction . given the above context and the state of both cosmological simulations and semi - analytic galaxy formation , the question as to whether these mock galaxies capture the true velocity distribution of real galaxies in clusters is still unanswered . however , the current simulated data now contain a wide variety of options for how galaxies are pasted into the clusters . more - so , these semi - analytic techniques create mock galaxy clusters which very closely resemble the observed universe in many respects @xcite . while the semi - analytic galaxy catalogs are not fully simulated universes , they do provide us with an opportunity to assess how well our theories and algorithms could do under realistic observational conditions ( e.g. , non - ideal target selection ) . in this study , we focus on how different tracers in the n - body simulations ( e.g. sub - halos or semi - analytic galaxies ) allow us to observe the gravitational potential and measure the projected escape velocity to infer the cluster mass . this technique is analogous to applying the jeans equation , except that the cluster observable is the radial escape velocity as opposed to the velocity dispersion and that the escape velocity maps directly to the gravitational potential , whereas the jeans analysis maps to its derivative . these differences are subtle but important . regardless , both the jean s technique and the caustic technique posit that the radius / velocity _ phase - space _ does indeed map directly to the gravitational potential and through some simplifying assumptions , ultimately to the gravitational mass . our primary goal is to present the statistical characterization ( accuracy and precision ) of caustic inferred halos masses , as well as study the effects of survey strategy when planning spectroscopic follow - up . in section 2 we discuss the caustic technique in detail and apply this technique in section 3 on n - body simulations using the underlying particles , the sub - halos , as well as on the semi - analytic mock galaxy catalogs . using the galaxy catalogs , we incorporate realistic targeting scenarios and show the effects on the measured bias and scatter . under newtonian dynamics , the escape velocity is related to the gravitational potential of the system , @xmath12 if the dynamics of the system are controlled by the gravitational potential , tracers which have not escaped the potential well should exist in a well - defined region of @xmath13 phase space , where @xmath14 is the physical or projected radius from the center of the cluster and @xmath15 is the peculiar 3-dimensional velocity or projected 1-dimensional velocity respectively relative to the bulk cluster motion . the edge of this region in @xmath13 space within which bound tracers are allowed to exist defines the escape velocity , @xmath16 . in figure 1 , we show an example halo from the millennium simulation where we identify the actual gravitational potential of the dark matter @xmath17 ( red lines ) and the iso - density contour which traces the escape velocity profile of the halo ( blue lines ) . in the left panel , the velocities and radii are 3-dimensional and in spherical coordinates while in the right panel they are projected along one line - of - sight . the surface that defines the density edge in the @xmath13 phase space is an iso - density contour that follows @xmath16 and therefore @xmath18 . in observed data , we identify the projected @xmath16 surface by applying standard kernel density estimation techniques to the dynamical tracers in the @xmath13 phase - space . the observed tracers have inherent observational uncertainties in both the radial and velocity directions . in this work , we focus on low - redshift sdss - like observations with spectroscopic precision of @xmath19 or @xmath20 in normalized coordinates and astrometric precision of @xmath21 . therefore , our kernel must be non - symmetric to account for the factor of ten difference in the two dimensions of the phase - space . @xcite showed that such axis weighting does not have a large effect on the mass profile determination , something that we confirm in this work . we use a fixed multi - dimensional gaussian kernel with a width in the @xmath14 and @xmath15 directions that independently adapt to the sampling according to ( silverman 1998 ) @xmath22 where @xmath23 is the number of dynamical tracers in the total phase - space and @xmath24 is the dispersion in the radial and velocity dimensions . equation [ eq : kernel ] minimizes the mean integrated squared error of the density estimate , which is the sum of the square of the statistical bias and the variance , also known as the statistical risk @xcite ) . while @xcite adopt an adaptive kernel technique , we will show that a standard fixed kernel recovers the cluster mass estimates with low scatter and bias . @xcite assert that any realistic models of galaxy clusters exhibit escape velocity profiles that at no point exceed @xmath25 where @xmath26 . if an iso - density contour breaks this limit along its surface , the @xmath16 value is replaced with a new value that yields @xmath25 . here , we follow the prescription used in @xcite which invoke a looser constraint of @xmath27 rather than @xmath26 . this allows the algorithm to remove very drastic changes in @xmath28 , but does not overly restrict the iso - density contour values . once the iso - density contours are determined we must choose which surface corresponds to the escape velocity . however , here we follow the standard procedure of assuming that the data are for time - averaged , self - gravitating , isolated clusters in a steady state . as described in @xcite , virialization means that the average system kinetic energy is half of the average system potential energy ( ke = @xmath29pe and @xmath30 ) within the virial radius @xmath31 . in combination with equation [ eq : escape ] this leads to @xmath32 equation [ eq : vesc_sigmav ] is defined for three dimensional measurements of the velocities . @xcite test this virialization condition in the millennium simulation and find it to hold when the system average energies are calculated near the virial radius . in real data we only observe the projected line - of - sight component of both the velocity dispersion and the escape velocity . there could exist some level of anisotropy in the velocity vectors : for example the radial and non - radial components of the velocity are not equal : @xmath33 . this is parameterized by the anisotropy parameter @xmath34 . by assuming that the escape velocity profile we measure in projection @xmath35 , where @xmath36 is where the 3-dimensional radius @xmath14 equals the projected radius @xmath37 , @xcite show that @xmath38 and as in @xcite , we define the projection correction term : @xmath39 recalling that equation [ eq : vesc_sigmav ] is for the three - dimensional velocities , it can be re - written in terms of the observed projected quantities ( i.e. , the projected radii @xmath40 and the line - of - sight velocities of the galaxies ) : @xmath41 we calculate the velocity dispersion @xmath42 ( hereafter labeled @xmath43 ) using a robust median - weighted technique on galaxies within the projected radius that has @xmath44 ( i.e. , @xmath11 ) , which is @xmath45 . interlopers are removed via a shifting - gapper technique described in [ sec : interloper_removal ] . we then choose the iso - density contour that satisfies equation [ eq : vesc_sigmav_proj ] . there is some uncertainty in the determination of this surface , which we quantify in [ sec : results_los ] . we note that we always use the same tracers when calculating the dispersions and the phase - space density ideally , we could use the escape velocity profile , which we assume to be the potential profile through equation [ eq : escape ] , to estimate a mass by using the poisson equation @xmath46 or some variation to arrive at the mass profile . however , this ideal scenario involves the challenge of taking derivatives of a noisy estimate of @xmath47 . instead , @xcite introduce an alternative estimation through the partial mass differential equation @xmath48 . invoking equation [ eq : escape ] , we may rewrite this differential as : @xmath49 and integrate to arrive at : @xmath50 after identifying the iso - density contour that describes the projected @xmath51 , we now have an estimate for @xmath18 by using equation [ eq : escape ] and @xmath52 and our equation now becomes : @xmath53 where @xmath54 @xcite claim that @xmath55 is roughly constant as a function of radius from 1 - 3@xmath11 as calibrated against simulations . for instance , @xcite find @xmath56 and @xcite find @xmath57 . we use a constant value calibrated against the millennium simulations to be @xmath58 and discuss the implications of both the assumption of a constant as well as its calibration in section [ sec : discussion ] . for a variation of the caustic technique that does not require this calibration , see @xcite . @xcite show that the velocity dispersion of a dark matter halo obeys a very tight virial relationship when compared with the critical dark matter mass @xmath1 of the form : @xmath59 where @xmath60 is the 1d velocity dispersion of the dark matter , or more precisely the 3d velocity dispersion divided by the @xmath61 . for the millennium simulation , we use the normalization @xmath62 km / s and slope @xmath63 . @xcite find these to be consistent between different cosmological simulations . the scatter in @xmath64 at fixed @xmath1 from the n - body simulations for equation [ eq : evrard_rel ] is @xmath65 . with slope @xmath29 , this inherent scatter implies a mass scatter from the virial relation to be @xmath66 . however , this amount of scatter only applies when the 3d velocity dispersion is known , which is never the case in the observed universe . @xcite show that the real scatter in mass as measured from the line - of - sight velocity dispersion is closer to @xmath67 when 100 red - sequence galaxies are used . this huge increase in scatter stems mostly from the projection of the galaxies and their velocities along the line - of - sight . @xcite use the @xcite semi - analytic galaxy catalogs but they do not use the @xcite calibration . instead , they re - calibrate the slope and amplitude of equation [ eq : evrard_rel ] using the 3-dimensional information of the _ galaxies _ as opposed to the particles . @xcite also calibrate their measured scatter in velocity and mass against a specific set of 100 red - sequence galaxies within the 3d virial radius of each halo . in the former case , the slope and amplitude will differ from the dark matter if the galaxies are dynamically biased with respect to the dark matter . in the latter case , relationships for the scatter ( e.g. , as a function of the number of observed galaxies ) become meaningless for larger samples ( and it is quite common to have clusters with more than 100 observed galaxies in their @xmath13 phase space ) . in this work , we use the virial - mass relationship from @xcite . this is an important distinction , because the different semi - analytic galaxy catalogs can each have inherent dynamical biases between the dark matter and the galaxies that we would otherwise not detect . similarly , the scatter in the measured line - of - sight velocity dispersions are determined against the particle velocity dispersions . this is also important , in that the observed line - of - sight scatter in the velocity dispersions and masses can be compared between the different semi - analytic galaxy catalogs . we select 100 halos from the millennium simulation to perform this study . while the halos are not chosen on the basis of any specific physical characteristics , our sample aims for fairly even mass sampling over @xmath68 . the average mass @xmath69 and the average critical radius @xmath70mpc . we then use four semi - analytic catalogs @xcite created using the millennium simulation along with the identified subhalos within each halo to test the caustic technique in section [ sec : results ] . @xcite suggest that in its most basic form , a theory of galaxy formation is a set of rules motivated by physical processes which transforms a halo mass function into an observed galaxy luminosity function . in fact , most semi - analytic techniques ( which define the rule - set ) judge their success primarily by comparing to published galaxy luminosity functions . in this work , we are equally concerned with how well those galaxies trace the underlying radial velocity phase - space , which we use to measure the escape velocity of halos and their gravitational potentials . galaxies in the semi - analytic algorithms are first identified at the location of collapsed sub - structure within halos . these are nominally the sub - halos and algorithms like subfind @xcite which have been shown to work well to identify all of the sub - structure in n - body simulations ( see also @xcite ) . once identified , these sub - halos trace the positions and velocities of the galaxy population . rules are put in place to define when and how a burst of activity ( e.g. star - formation / nuclear ) occurs . these sub - halos can grow in total mass ( and luminosity ) by accreting gas ( or a model for gas ) , other particles , and other sub - halos through merging . it is the merging and other dynamical interactions which are responsible for altering the radial velocity phase - space and thus affecting our ability to use dynamical tracers for the halo mass . at any given redshift ( or snapshot output of an n - body simulation ) , the sub - halos which have survived a merger up to that point would not be good tracers of the halo radial velocity phase - space ( see also figure [ fig : mvsm ] ) . this is because at the resolution of the millennium simulation , the sub - halos in simulations are easier to destroy compared to galaxies in the real universe , which are much more compact and gravitationally bound . so while an interaction might destroy a sub - halo , there is no reason to think it would destroy a galaxy . in modern semi - analytic techniques , decisions ( or a rule set ) must be defined to decide what to do with a `` galaxy '' after its sub - halo is no longer identified in a snapshot . the most common approach is to identify the most - bound particle in the sub - halo before it was destroyed and follow it as a surviving galaxy through future snapshots ( i.e. , to lower redshifts ) . these are sometimes called `` orphans '' as they are semi - analytic galaxies that have lost their dark matter halos . the above rule does not happen in reality : the evolution of the position and velocity of a galaxy can not be determined from a single particle . so a technique is applied to define statistically how long a galaxy might survive before it is ripped apart in a merger ( i.e. , the merging time due to dynamical friction ) : @xmath71 where @xmath72 is the circular velocity for a mass in a gravitational potential defined by an isothermal sphere at radius @xmath73 of the sub - halo and the masses of the halo and the sub - halo are @xmath74 and @xmath75 respectively @xcite , and @xmath76 is a coefficient needed to reproduce observed luminosity functions at the luminous end @xcite . in this rule - set , a clock is started when a galaxy s sub - halo is destroyed . the surviving galaxy is merged with its nearest galaxy or sub - halo after this time - scale has expired . when the final merging happens , the galaxy is destroyed and its stars , gas , and dark matter are distributed in various ways ( i.e. , it is a rule - set in the semi - analytic algorithm ) . however , this statistical rule - set only defines _ when _ a dark matter orphaned galaxy is destroyed , not how . for instance , the particle which represents the galaxy is no longer actually merging or changing its orbit due to dynamical interactions outside the normal particle - particle interactions . this would require a new rule - set . the above description is complicated and there are numerous ways in which a semi - analytic techniques can implement dynamical events in the lifetime of a simulated galaxy . this is why we will investigate the effects on the caustic masses from different implementations of halo dynamics in the semi - analytic galaxy catalogs . in this analysis we start with all of the semi - analytic galaxies within a 60h@xmath77mpc - length 3-dimensional box around each halo center . we then sub - select the n brightest galaxies ( where n = 100,50,25 , etc ) within a projected radius in these volumes to create the halo radius / velocity phase - space diagrams along one or more lines - of - sight . these volumes place limits on the projected phase - space velocities that are @xmath78km / s relative to the halo velocity centroids . since the typical escape velocities are @xmath79km / s , these volumes are large enough to incorporate realistic projection effects ( see figure [ fig : caustics ] ) . with projection , interlopers ( non - member foreground / background galaxies ) can play a large role in affect both the measured phase - space density as well as the line - of - sight velocity dispersion . therefore , an effort must be made to systematically remove the interlopers from each line - of - sight projection . many methods have been devised to identify and remove interloper galaxies from a sample around a halo . these include color - based @xcite , @xmath80-clipping , gapper @xcite , and phase - space selection methods @xcite . in this study , we do not try and optimize any particular method , but rather choose one in existence that returns a minimal bias for velocity dispersion and mass at large sampling . we limit ourselves by testing only a few basic interloper removal techniques , and compare their relative biases and resulting scatter . in figure [ fig : interlop ] we compare three common techniques to measure velocity dispersions against the underlying dark matter . the first technique is to apply a simple upper / lower bound in velocity space ( @xmath81 km / s ) . the second is to iteratively remove outliers via sigma clipping ( @xmath82 ) . the third is called a shifting - gapper , where we work in the full phase - space and identify velocity gaps as a function of radius as indicators of interloping sub - structure . this shifting - gapper technique is similar to what is applied in @xcite . galaxies are sorted into bins as a function of radius while keeping the number in each bin constant at @xmath83 . when odd multiples of 25 are used , the last radial bin may have less than 25 galaxies . in each bin , the galaxies are sorted by their peculiar velocity and a f - pseudosigma " @xcite is calculated and used as the velocity gap to remove perceived interlopers . the calculation of f - pseudosigma is an iterative process that we stop if less than 5 galaxies remain in a bin or if f - pseudosigma @xmath84 km / s . in all cases , the final interloper - cleaned velocity dispersion is calculated using a robust bi - weighted estimator , which we confirm is always less biased than a simple standard deviation @xcite . when applying only a simple velocity boundary , the dispersions are biased high and the scatter is large ( for @xmath85 ) . however , both the sigma clipping and the shifting - gapper techniques do better and equally well in recovering unbiased velocity dispersions ( see also @xcite ) . the shifting - gapper does a better job at reducing the scatter at all values of n@xmath86 . this is because the technique utilizes the 2-dimensional phase - space data as opposed to the one - dimensional velocity distributions . as such , the shifting - gapper can identify sub - structure in the phase - space which sigma clipping can not . clusters with poor sampling @xmath87 , have dispersions that are biased low . throughout the rest of this work we apply the shifting - gapper technique to measure velocity dispersions . we measure the caustic mass @xmath88 inside @xmath11 and compare directly to the spherical halo mass @xmath1 . in figure [ fig : mvsm ] we show how @xmath88 inferred from the n@xmath86 brightest galaxies inside @xmath11 scales with @xmath1 . in each panel , we highlight the @xmath89 inferred masses using the guo semi - analytic galaxies ( dots ) . the error bars are detailed in [ sec : results_surface ] & [ sec : results_los ] . we note that in this figure , @xmath88 for each of the 100 halos is measured along a single 60h@xmath77mpc line - of - sight , thus representing a realistic observing scenario . we measure the percent scatter in @xmath90 at fixed @xmath91 as well as the bias determined from the error weighted mean difference between @xmath90 and @xmath91 . in order to compare to the virial masses , we also quantify the log scatter and bias of the projected line - of - sight velocity dispersion and the virial mass ( equation [ eq : evrard_rel ] ) . at small n@xmath92 , the caustic masses are biased low by @xmath66 on average , and the log - normal scatter about the fit to the data is @xmath93 . at n@xmath94 , the bias disappears and the scatter reduces to @xmath95 at n@xmath96 . we show a summary of the caustic mass bias and scatter in figures [ fig : scatt_vs_n ] and [ fig : bias_vs_n ] . these data are presented in tables 1 - 5 . at fixed n@xmath86 , the biases of the different semi - analytics agree to within their errors and so unless otherwise noted , we focus the rest of our analyses on the @xcite galaxies . typical observations of galaxy clusters can detect or easily measure the spectroscopic redshifts of a handful to several hundred galaxies within the virial radius of a system . fundamental to the caustic technique is the projection of observed galaxies into a radial and velocity phase space , within which we identify a critical iso - density transition or caustic . in [ methods ] , we describe how we use a multi - dimensional gaussian kernel to estimate the underlying phase space density based on an observed tracer population . density estimates always carry some degree of uncertainty due to limited sampling . we therefore expect that small or sparsely observed clusters will exhibit a large amount of uncertainty in the density estimation process . we quantify this uncertainty and it s effect on the caustic mass estimate by using the jack - knife technique to re - sample the phase - space . for each jack - knife re - sampling @xmath97 of a cluster with n@xmath86 , we re - calculate the velocity dispersion ( @xmath98 ) and re - calibrate the escape velocity iso - density contour according to equation [ eq : vesc_sigmav_proj ] to infer a jack - knife caustic mass ( @xmath99 ) according to equation [ eq : caustic_final ] . we show the distribution of @xmath100 for four randomly chosen halos in figure [ fig : frac_scat ] ( upper left ) . for each halo , we then calculate the standard deviations of these caustic masses over the @xmath97 jack - knife re - samplings . we plot a histogram ( red ) of these standard deviations ( for all 100 halos ) in the lower panels of figure [ fig : frac_scat ] for two different values of n@xmath86 ( 15 and 100 ) . the means of the red histograms in the lower panels of figure [ fig : frac_scat ] define the _ average caustic - surface induced scatter _ as a function of n@xmath86 . for n@xmath101 , the distribution of the averages of the mass uncertainties is sharply peaked around a few percent . therefore , the surface uncertainty plays a negligible role in the total caustic mass uncertainty at n@xmath101 . by n@xmath102 , the mean of the distribution shifts to 20% and the distribution becomes less sharply peaked . as a result of the lower density of sampling , the surface we estimate becomes less robust . observationally , we only have one line - of - sight ( hereafter abbreviated as `` l.o.s '' ) to a cluster in our universe . if clusters are not spherically symmetric in either physical or velocity space , the mass we measure will be dependent on our l.o.s to the cluster . this dependence introduces a l.o.s scatter into our observations which contributes to the scatter in mass relationships . in simulations , we are not bounded to one l.o.s and can make mass measurements along many l.o.s to a single cluster . we show the distribution of @xmath103 for four randomly chosen halos in figure [ fig : frac_scat ] ( upper right ) for 100 l.o.s . recall that the histograms in the upper left panel of [ fig : frac_scat ] are the 100 jack - knife re - samplings of a single line - of - sight . these l.o.s . distributions are much broader than the surface distributions . we again take the standard deviation ( @xmath104 ) as an estimate of the l.o.s scatter for each of our 100 halos . the distribution of these _ scatters _ is shown in the lower panels of figure [ fig : frac_scat ] in the blue histograms . the means of the blue histograms in the lower panels of figure [ fig : frac_scat ] define the _ average line - of - sight induced scatter _ as a function of n@xmath86 . when n@xmath101 , we can see that our clusters all exhibit l.o.s scatter between 20 - 40% with a mean of @xmath95 . by n@xmath102 , the l.o.s scatter can be anywhere from 40 - 90% for a given cluster with a mean of @xmath105 . the mean of the l.o.s caustic mass scatter distributions are shown as a function of n@xmath86 in the left panel of figure [ fig : scatt_vs_n ] ( dashed lines ) . additionally , we show the mean l.o.s scatter in virial mass calculated with equation [ eq : evrard_rel ] and the mean l.o.s scatter in velocity dispersion . the l.o.s scatters decrease as we increase the n@xmath86 . the scatter in l.o.s caustic mass approaches a minimum value of @xmath95 for large n@xmath86 and the virial mass reaches @xmath106 . the scatter in l.o.s velocity dispersion decreases to a minimum of @xmath66 at n@xmath96 which agrees closely with @xcite who find a similar value using all galaxies for a large sample of halos in the @xcite semi - analytic catalog . we also calculate the observed mass scatter in the caustic and virial masses directly by measuring it from figure [ fig : mvsm ] , which represents the scatter we would expect from a single realization of a line - of - sight to each halo . in this case , we take 100 different realizations to determine the expectation value of the scatter , which we plot as a function of @xmath107 in the left panel of figure [ fig : scatt_vs_n ] ( solid lines ) . this scatter is larger than we predict from the statistical distributions determined from figure [ fig : frac_scat ] and shown as the dotted - lines in [ fig : scatt_vs_n ] . we identify the unaccounted component as originating from an intrinsic scatter in the caustic and virial masses . for instance , @xcite calculate the intrinsic scatter in the 3d particle velocity dispersion at fixed halo mass to be @xmath65 . from equation [ eq : evrard_rel ] , this translates to a 15% intrinsic virial mass uncertainty . @xcite calculate the intrinsic scatter for the caustic masses in 3d to be @xmath108 , also using the particles . this intrinsic ( i.e. , 3d ) uncertainty is a systematic error which should be independent of the systematic uncertainty as a result of projection effects in each l.o.s . to confirm this fact , we perform a spearman correlation test on @xmath109 vs. @xmath110 , where @xmath111 are all l.o.s mass measurements of a cluster , and @xmath112 is the caustic mass measured from the 3-dimensional positions and velocities of the galaxies . we look at 100 l.o.s and find an average correlation coefficient of @xmath113 for the caustic mass and @xmath114 for the virial masses . this confirms that the two ( l.o.s . and 3d ) systematic uncertainties are uncorrelated . in the right panel of figure [ fig : scatt_vs_n ] we add the intrinsic scatter in quadrature to our statistical representations of the line - of - sight scatter ( dotted lines in figure [ fig : scatt_vs_n]-left ) . we then fully recover the observed total scatter to high precision and these are the values we present in tables 1 - 5 . equation [ eq : evrard_rel ] implies a relationship of @xmath115 between the log scatter in virial mass and log scatter in velocity dispersion ; however , the relationship between scatter in caustic mass and scatter in velocity dispersion is not immediately obvious as the two do nt have an exact analytic relationship . we do know that the escape velocity surface is calibrated by the velocity dispersion and the caustic mass is calculated by integrating the @xmath116 . a naive assumption is that the scatter in velocity dispersion dominates over all other systematic scatters in the caustic technique . if so , we should find that the log scatter in caustic mass is very nearly twice that of the log scatter in velocity dispersion . in figure [ fig : vdisp_vs_mass_scatt ] we show that the virial mass obeys the predicted log mass - velocity dispersion scatter relation ( blue line ) as expected . we also show that the sensitivity of the scatter in caustic mass is well predicted by twice the scatter in velocity dispersion ( red line ) ; however the absolute value is slightly higher due to other systematic sources of uncertainty . this implies that other forms of systematic uncertainty in the technique are indeed small , and that the line - of - sight scatter in velocity dispersion dominates the uncertainty in caustic mass , albeit to a lesser degree than it does in the virial mass . in addition to the mass scatter for a given n@xmath86 , we also show how the average mass bias depends on n@xmath86 . as before with the total scatter , we measure the average bias for the 100 halos as @xmath117 after choosing only one l.o.s to each . we repeat this measurement for 100 different l.o.s and report the average sample bias . the bias in caustic mass is shown in figure [ fig : bias_vs_n ] ( middle ) for all 4 semi - analytics and the subhalos . at small n@xmath86 the caustic mass can be biased very low compared with @xmath1 . however , above n@xmath118 , the log bias is unbiased and all the semi - analytics agree to within the @xmath119-errors . a similar trend is seen with the virial mass ( left ) , but when compared with the caustic mass , the virial mass exhibits a larger bias for small n@xmath86 . the virial mass bias , and to a lesser degree the caustic mass bias , are dependent on the velocity dispersion bias which is shown as a function of n@xmath86 in [ fig : bias_vs_n ] ( right ) . we find that the four different semi - analytics generally agree to within the errors on the means ( from the 100 l.o.s . ) . the @xcite galaxies are always lower than the others , but not significantly . the sub - halos are very biased . we discuss these trends in section [ sec : robust ] up to now , we have assumed that we have spectroscopic follow - up that is complete for the @xmath23 brightest galaxies within the projected @xmath11 of each halo . here , we drop that constraint and allow more realistic targeting algorithms . we include selection based on galaxy magnitude , membership within the red - sequence , and projected distance from the cluster center . in figure [ fig : bias_vs_bright ] we keep n@xmath86 and color constant ( i.e. , only the red sequence ) and show the bias ( top ) and scatter ( bottom ) as we decrease the fraction of original brightest ( in absolute magnitude ) galaxies . starting from the sorted 50 brightest galaxies , we have different options for how we replace galaxies . for instance , we could replace them randomly , or from the brightest to the dimmest , or from the dimmest to the brightest , etc . in figure [ fig : bias_vs_bright ] we show two different replacement techniques starting from the _ dimmest _ ( solid lines ) or _ brightest _ galaxies ( dashed lines ) . we always keep the 5 brightest galaxies in each case , and we replace galaxies with those starting from the @xmath120 brightest within r@xmath10 . we find that the bias does not depend on in how the replacement is done . as shown in figure [ fig : bias_vs_bright ] , the virial mass is more affected by target selection based on galaxy luminosity than the caustic mass . to minimize brightness - induced mass biases , the @xcite semi - analytics indicate that one should always strive to target the brightest galaxies . there is no change in the average scatter as dimmer galaxies are added to the sample . in figure [ fig : red_bias_vs_red ] , we keep n@xmath86 constant and show the bias ( top ) and scatter ( bottom ) as a function of red galaxy fraction . we start with the @xmath121 brightest red - sequence galaxies , and replace the dimmest fraction of those galaxies with the brightest blue non - red - sequence galaxies within the projected @xmath11 . we also conduct the test by replacing the galaxies randomly , with no noticeable difference in the results . color plays a bigger role in selection - induced biases and scatter when compared to brightness and again , the virial masses are more affected than the caustic masses . the velocity dispersion can be biased as much as @xmath122 higher than the baseline bias when only 25% of the sample are bright red sequence galaxies and the rest are bright bright blue galaxies . consequently , we see the average bias for both caustic mass and virial mass change from a slight negative bias for purely red - sequence galaxies of -5% , to a positive bias of @xmath123 and @xmath124 respectively when we use a small fraction of the original red sequence sample . this is the color - dependent velocity segregation effect noticed in real data @xcite and it is due to the inclusion of blue galaxies with higher infall velocities . decreasing the fraction of bright red galaxies to bright blue galaxies can also increase the expected scatter in mass by as much as @xmath125 in caustic mass and @xmath126 in virial mass as compared with the sample of all red sequence galaxies . this is due to the higher velocities of the typically infalling blue galaxy population that can depend more on l.o.s to a cluster . it is very important to target the bright red galaxies in order to avoid this color - selection induced bias . in figure [ fig : red_bias_vs_core ] ( left ) , we show the bias ( top ) and scatter ( bottom ) as a function of the core fraction . we start with the @xmath121 brightest galaxies , and replace the dimmest galaxies whose projected locations are within 200kpc of the center with galaxies whose projected radii are greater than 700kpc from the center . this simulates survey data that under - samples the core due to fiber collisions or slit overlap in the dense inner regions of clusters and can not extract spectra for all observable galaxies . on the other hand , it is often the case where nearby massive clusters are not well - sampled out to the virial radius and beyond . in figure [ fig : red_bias_vs_core ] ( right ) we show the bias ( top ) and scatter ( bottom ) as a function of the outskirt wing " fraction . starting with the @xmath121 brightest galaxies , we replace the dimmest galaxies whose projected radii are greater than 700kpc with galaxies that are within 200kpc from the center . in the first case , the more we under - sample the core relative to the outskirts , the more biased our estimates become of mass and velocity dispersion . this is true regardless of whether we use the virial relation or the caustic technique . no matter how much we under - sample the core , the caustic inferred masses are less biased than the virial estimates . this is due to the fact that the caustic technique does not depend as heavily on uncertainty in velocity dispersion ( see figure [ fig : vdisp_vs_mass_scatt ] ) , and under - sampling the projected core of a cluster will lower the measured velocity dispersion . we see no effect on the mass scatter when under - sampling the core . when we under - sample the outskirts or wings " relative to the core , we see dramatically different responses from the caustic and virial mass estimators . under - sampling the outskirts has the effect of raising the measured velocity dispersion . this is because the galaxies with large l.o.s velocity are preferentially projected near the cluster core while under - sampling the outskirts removes galaxies with preferentially smaller l.o.s velocities . this raises the virial mass estimate as it is directly proportional to velocity dispersion . however , the caustic inferred masses exhibit the opposite trend since we are reducing the phase - space density in the outer regions by removing these galaxies . the mass scatter is unaffected by under - sampling the outskirts . to minimize radial - selection induced biases , it is important to evenly sample the phase space . with the availability of large and complete spectroscopic surveys like the sloan digital sky survey ( * ? ? ? * sdss ) and gama ( galaxy mass and assembly gama ) @xcite , highly multiplexed multi - object optical spectroscopy on 8 m class telescopes ( imacs @xcite , vimos @xcite , deimos @xcite ) , and the latest generation of multi - object near - infrared spectrographs ( mmirs @xcite , flamingos 2 @xcite , mosfire @xcite ) , we can measure galaxy large numbers of galaxy velocities in clusters from @xmath127 to @xmath128 . and just as we can use three different observables to infer the masses of clusters ( see section [ sec : intro ] ) , we can use the galaxy velocities in three different ways to infer dynamical mass ( e.g. , via virial scaling , the jean s equation , and the escape velocity ) . the questions then are how well do these different dynamical techniques work under realistic observing conditions and how do they compare . here , we compare the virial relation via the velocity dispersion @xcite and the escape velocity mass inferred from the caustic technique @xcite . we examine four different semi - analytic mock galaxy catalogs , as well as the sub - halos and the particles in the millennium simulation . the measured scatter and bias in the inferred dynamical masses are quantified in tables 1 - 5 . the level of scatter and bias varies by only a few percent between most of the different semi - analytic tracers . the exception is for the @xcite galaxies , which sometimes differ by slightly more than the 1@xmath80 errors on the bias estimates . in one sense , this robustness is not surprising , since the majority of galaxies in the semi - analytics are attached to sub - halos , which are the same for each of the mock galaxy catalogs . however , there are a variety of rules which are applied to the `` orphan '' population ( see section [ sec : methods_data ] ) . for the satellite population , these orphan galaxies are quite common at low stellar masses , while the exact fraction depends on the semi - analytic rule - set . in the semi - analytic catalogs studied here , the fraction of @xmath129 brightest galaxies that are orphans ranges from 50% ( bower ) to 25% ( guo ) . consider some of the differences in the orphan population between the @xcite semi - analytics compared to @xcite , @xcite , and @xcite : the bower sample uses a merger tree that differed in how friend - of - friend groups were defined , how spuriously linked halos were handled , how independent halos were identified , and how descendants of halos were tracked through time ; the bower sample includes the sub - halo orbital energy and angular momentum when calculating the merging timescale of the orphans ; the @xcite semi - analytics go one step further by modeling the orbital decay of the orphans before they are destroyed . these algorithmic differences manifest as variations in the spatial distribution of the satellites within the halos . the de lucia semi - analytic galaxies have much flatter density profiles within the virial radius when compared to the @xcite galaxies @xcite . since the @xmath13 phase - space is different between the semi - analytics , one might expect that the caustic technique ( which is based on the phase - space density ) would also differ significantly . however , this is not the case and even when the orphan fraction and radial densities differ by factors of two or more , the caustic masses vary by @xmath130 . we interpret this lack of difference between the different semi - analytic techniques as a measure of the robustness of the technique to large systematic variations in the @xmath13 velocity distribution function . the one exception to this robustness is for the sub - halo population where we detect large biases in the velocity dispersion compared to the dark matter . since the caustic mass is calibrated using the velocity dispersion , this bias carries through into a bias in both the virial mass and the caustic mass . we attribute this to how poorly the sub - halos trace the @xmath13 phase - space within halos in simulations with resolution similar to the millennium simulation ( see also @xcite ) . one could ask whether the semi - analytic techniques have converged in their representation of the @xmath13 phase - space inside clusters , as represented by the fraction of orphan galaxies . @xcite have shown that at higher resolutions , galaxies are no longer orphaned from sub - halos . instead , a new `` crisis '' of sub - halo over - abundance presents itself . @xcite also show that this crisis is averted through rule - sets analogous to the ones used at lower resolution . in other words , instead of following the most - bound particle of a destroyed sub - halo , one attaches semi - analytic galaxies only to those sub - halos that previously achieved some minimal mass threshold before entering the cluster . the velocity dispersion of these semi - analytic galaxies is unbiased with respect to the dark matter , which is what we already find here . while none of these semi - analytic prescriptions is how nature places galaxies into halos , we have established that the virial relation and the caustic technique are robust to variations in their rule - sets . this baseline allows for a comparison of the absolute levels of the inferred mass biases and scatters between the virial and the caustic techniques . for a fixed number of tracers projected to lie within @xmath131 of the clusters , the scatter in the caustic mass @xmath90 at fixed @xmath91 is about 1/3 smaller than @xmath132 . in other words , at fixed n , the caustic technique is a more precise estimate of the true halo mass compared to the virial relation . the growth of the mass scatter as @xmath107 decreases is the same for the two techniques . as with the virial relation and the velocity dispersion , the dominant component of the scatter in the caustic mass is from the line - of - sight variations in the observed velocity dispersion and phase - space density ( see also @xcite ; @xcite ) . uncertainties in @xmath88 induced from the sampling of the phase space do not contribute unless n@xmath133 , below which they quickly grow to be a dominant component . in figure [ fig : bias_vs_n ] we show the strong dependence of the bias on the number of tracers used in the mass calculation . @xcite suggest that this was due to dynamical friction , since we ( as they ) explicitly restrict the sub - samples to the brightest galaxies . however at fixed n , large variations in the brightness distribution of the galaxies has only a small affect on the observed bias . the same can be said for color selection ( see figure [ fig : red_bias_vs_red ] ) . thus while velocity segregation from sub - sampling plays a role , it is minimal compared to the number of galaxies used in the mass determination . this holds for both the virial masses as well as the caustic masses . put simply , if a targeting algorithm requires a trade - off between number and color , number wins . regardless of how biases are introduced into the velocity distributions ( e.g. , through number , brightness , or color ) the caustic mass is less affected than the virial mass . the trends in the biases for the virial and caustics masses are similar . the exception is when radial selection constraints are applied . we find that the velocity dispersion is more susceptible to incompletenesses in the core sampling while the caustic technique is more susceptible to incompletenesses in the cluster outskirts . the former trend can be understood from a dynamically cold population of galaxies which dominates the dispersion measurement at low sampling . the latter is also a sampling issue , where the iso - density contour gets lost in the background for low sampling in the cluster outskirts . finally , we note that the implementation of the caustic technique utilizes a step which calibrates to simulations ( see @xmath134 in equation [ eq : caustic_final ] ) . this is no different than the virial calibration ( equation [ eq : evrard_rel ] ) . @xcite find @xmath56 and @xcite find @xmath57 , which corresponds to a systematic uncertainty of 30% in mass . this is larger than any biases we detect from sample selection . one explanation the difference in the value for @xmath134 is in how @xcite and @xcite define their mock galaxy catalogs : the former used only halos ( with ten particles or more ) while the latter used only particles to define galaxies . since sub - halos in n - body simulations are biased tracers of the dark matter velocity dispersions , the calibration factor and caustic masses will also be biased ( see figure [ fig : bias_vs_n ] middle and right panels ) . we test this hypothesis by re - calibrating @xmath134 using the millennium simulation sub - halos alone . the high halo velocity dispersions of the sub - halos require a lower calibration factor than from the dark matter . we find @xmath56 , identical to what is used in @xcite , who note that their own halo velocity dispersions are biased with respect to the dark matter in their simulations . @xcite explicitly used dark matter particle positions and velocities to define their galaxies and calibrate @xmath134 . the semi - analytics we use lie in between these two extremes , which explains why we find @xmath135 . @xcite notes that @xmath134 is not constant with radius in simulations and that the caustic mass profiles can over - estimate the mass within @xmath136 , where velocity anisotropies become smaller . likewise , @xcite compare caustic - derived and weak - lensing - derived mass profiles and find disagreement within @xmath137 , as expected from the anisotropy profiles in simulations . in this work , we are not measuring mass profiles but instead the integrated masses of the clusters out to @xmath11 . as shown in @xcite , unbiased halo masses require an appropriate measure of @xmath138 from equation [ eq : caustic_final ] , which is what we do to achieve unbiased caustic masses for the @xcite semi - analytic galaxies . this is without question a tuning step which carries with it a certain level of additional systematic uncertainty . it is very unlikely that the radius - velocity phase - space of galaxies in the universe is represented by the sub - halo population in n - body simulations ( regardless of resolution ) . current observational constraints on the bias between the galaxy and dark matter velocity dispersions are measured indirectly ( e.g. , @xcite ) and find it to be @xmath139 when a multitude of independent cosmological priors are leveraged . from the current simulation and observational work , it is not likely that @xmath56 , which corresponds to a galaxy / dark matter velocity bias of 10% in a @xmath140cdm simulation ( see table 2 sub - halos ) . if galaxies and dark matter have dispersions that are unbiased with respect to each other , the caustic masses ( integrated to @xmath11 ) would be low by @xmath14120% using @xmath56 @xcite . ideally , the simulation calibration factor can be dropped entirely , and in @xcite we present a revised derivation of the caustic technique that relies only on the observable parameters : the nfw density scale parameter , @xmath142 , and @xmath143 and their statistical and systematic uncertainties . in this work , our main conclusions are as follows . * we measure velocity dispersions , virial masses ( @xmath132 ) , and caustic masses ( @xmath88 ) for 100 halos in the millennium simulation with masses @xmath144 . these halos exist at low redshift ( @xmath145 ) and are populated with galaxies via four different semi - analytic prescriptions @xcite . * the resulting scatter and bias in @xmath89 relative to the halo mass ( @xmath1 ) is robust and largely independent of the semi - analytic prescription used to populate cluster - size dark matter halos with galaxies . the exception is when only subhalos are observed which do not trace the dark matter velocity field and measured velocity dispersions . the velocity dispersion for the subhalos can be biased high by @xmath146-@xmath125 resulting in a caustic mass bias of @xmath147-@xmath148 . * as with the @xmath132 , the dominant component of the scatter in @xmath89 is from the line - of - sight variations in the observed velocity dispersion . however , for a fixed number of tracers ( n@xmath86 ) projected to lie within @xmath131 of the clusters , the scatter in the caustic mass @xmath149 is @xmath95 at n@xmath94 which is @xmath150 smaller than @xmath132 . * the bias in @xmath89 relative to @xmath1 is strongly dependent on n@xmath86 . while color selection , radial completeness , and magnitude can play a role in inducing bias depending on how a cluster is observed , their effect is much less than n@xmath86 . given a choice , it is better to use a full , unrestricted sample to estimate caustic masses rather than a color selected ( such as red sequence ) sample to achieve larger numbers . * we find a caustic mass calibration factor @xmath58 . this differs from the calibrations based on either just the dark matter particles ( @xmath151 ) or just the subhalos ( @xmath152 ) . while galaxy / dm velocity bias affects the caustic mass less than the dynamical mass , it is an important component of the total systematic uncertainty . the authors made use of the flux high performance computing cluster at the university of michigan . the millennium simulation databases used in this paper and the web application providing online access to them were constructed as part of the activities of the german astrophysical virtual observatory ( gavo ) . the authors want to especially thank gerard lemson for his assistance and access to the particle data . we also wish to thank the referee , craig harrison , august evrard , heidi wu , risa wechsler , and michael busha for their helpful comments and discussion . this material is based upon work supported by the national science foundation graduate student research fellowship under grant no . dge 1256260 . , m. , et al . 2003 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 4834 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . p. guhathakurta , 161172 , s. , elston , r. , raines , s. n. , et al . 2006 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 6269 , society of photo - optical instrumentation engineers ( spie ) conference series , o. , et al . 2003 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 4841 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . m. iye & a. f. m. moorwood , 16701681 , i. s. , steidel , c. c. , matthews , k. , epps , h. , & adkins , s. m. 2008 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 7014 , society of photo - optical instrumentation engineers ( spie ) conference series , b. a. , fabricant , d. , geary , j. , martini , p. , nystrom , g. , elston , r. , eikenberry , s. s. , & epps , h. 2004 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 5492 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . a. f. m. moorwood & m. iye , 13061313 c | cc | cc | cc | c + + & & & & orphans + n & bias & scatter & bias & scatter & bias & scatter & fraction + & @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & + 150 & [email protected] & 0.15 & [email protected] & 0.45 & [email protected] & 0.31 & 0.40 + 100 & [email protected] & 0.17 & [email protected] & 0.50 & [email protected] & 0.33 & 0.34 + 75 & [email protected] & 0.18 & [email protected] & 0.54 & [email protected] & 0.36 & 0.30 + 50 & [email protected] & 0.21 & [email protected] & 0.62 & [email protected] & 0.41 & 0.26 + 35 & [email protected] & 0.23 & [email protected] & 0.68 & [email protected] & 0.46 & 0.25 + 25 & [email protected] & 0.26 & [email protected] & 0.78 & [email protected] & 0.53 & 0.23 + 15 & [email protected] & 0.35 & [email protected] & 1.04 & [email protected] & 0.73 & 0.21 + 10 & [email protected] & 0.40 & [email protected] & 1.28 & [email protected] & 0.94 & 0.21 + c | cc | cc | cc | c + + & & & & orphans + n & bias & scatter & bias & scatter & bias & scatter & fraction + & @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & + 150 & [email protected] & 0.15 & [email protected] & 0.45 & [email protected] & 0.31 & 0.36 + 100 & [email protected] & 0.16 & [email protected] & 0.49 & [email protected] & 0.33 & 0.32 + 75 & [email protected] & 0.18 & [email protected] & 0.54 & [email protected] & 0.36 & 0.30 + 50 & [email protected] & 0.21 & [email protected] & 0.62 & [email protected] & 0.42 & 0.28 + 35 & [email protected] & 0.23 & [email protected] & 0.68 & [email protected] & 0.46 & 0.28 + 25 & [email protected] & 0.26 & [email protected] & 0.80 & [email protected] & 0.54 & 0.26 + 15 & [email protected] & 0.35 & [email protected] & 1.03 & [email protected] & 0.73 & 0.25 + 10 & [email protected] & 0.43 & [email protected] & 1.29 & [email protected] & 0.96 & 0.23 + c | cc | cc | cc | c + + & & & & orphans + n & bias & scatter & bias & scatter & bias & scatter & fraction + & @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & + 150 & [email protected] & 0.15 & [email protected] & 0.44 & [email protected] & 0.29 & 0.39 + 100 & [email protected] & 0.16 & [email protected] & 0.48 & [email protected] & 0.32 & 0.34 + 75 & [email protected] & 0.18 & [email protected] & 0.54 & [email protected] & 0.36 & 0.31 + 50 & [email protected] & 0.21 & [email protected] & 0.62 & [email protected] & 0.41 & 0.27 + 35 & [email protected] & 0.23 & [email protected] & 0.68 & [email protected] & 0.46 & 0.26 + 25 & [email protected] & 0.27 & [email protected] & 0.80 & [email protected] & 0.54 & 0.25 + 15 & [email protected] & 0.34 & [email protected] & 1.02 & [email protected] & 0.72 & 0.25 + 10 & [email protected] & 0.42 & [email protected] & 1.27 & [email protected] & 0.95 & 0.25 + c | cc | cc | cc | c + + & & & & orphans + n & bias & scatter & bias & scatter & bias & scatter & fraction + & @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & + 150 & [email protected] & 0.15 & [email protected] & 0.48 & [email protected] & 0.33 & 0.55 + 100 & [email protected] & 0.16 & [email protected] & 0.51 & [email protected] & 0.36 & 0.52 + 75 & [email protected] & 0.18 & [email protected] & 0.55 & [email protected] & 0.39 & 0.51 + 50 & [email protected] & 0.21 & [email protected] & 0.63 & [email protected] & 0.43 & 0.49 + 35 & [email protected] & 0.22 & [email protected] & 0.67 & [email protected] & 0.49 & 0.48 + 25 & [email protected] & 0.26 & [email protected] & 0.78 & [email protected] & 0.55 & 0.46 + 15 & [email protected] & 0.32 & [email protected] & 0.96 & [email protected] & 0.71 & 0.43 + 10 & [email protected] & 0.41 & [email protected] & 1.21 & [email protected] & 0.93 & 0.39 + c | cc | cc | cc + + & & & + n & bias & scatter & bias & scatter & bias & scatter + & @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 + 150 & [email protected] & 0.21 & [email protected] & 0.63 & [email protected] & 0.41 + 100 & [email protected] & 0.21 & [email protected] & 0.64 & [email protected] & 0.39 + 75 & [email protected] & 0.21 & [email protected] & 0.63 & [email protected] & 0.40 + 50 & [email protected] & 0.21 & [email protected] & 0.65 & [email protected] & 0.43 + 35 & [email protected] & 0.23 & [email protected] & 0.69 & [email protected] & 0.47 + 25 & [email protected] & 0.27 & [email protected] & 0.81 & [email protected] & 0.54 + 15 & [email protected] & 0.33 & [email protected] & 1.01 & [email protected] & 0.69 + 10 & [email protected] & 0.41 & [email protected] & 1.23 & [email protected] & 0.83 +

we use conformal maps to study a free boundary problem for a two - fluid electromechanical system , where the interface between the fluids is determined by the combined effects of electrostatic forces , gravity and surface tension . the free boundary in our system develops sharp corners / singularities in certain parameter regimes , and this is an impediment to using existing single - scale " numerical conformal mapping methods . the difficulty is due to the phenomenon of crowding , i.e. the tendency of nodes in the preimage plane to concentrate near the sharp regions of the boundary , leaving the smooth regions of the boundary poorly resolved . a natural idea is to exploit the scale separation between the sharp regions and smooth regions to solve for each region separately , and then stitch the solutions together . however , this is not straightforward as conformal maps are rigid global " objects , and it is not obvious how one would patch two conformal maps together to obtain a new conformal map . we develop a multi - scale " ( i.e. adaptive ) conformal mapping method that allows us to carry out this program of stitching conformal maps on different scales together . we successfully apply our method to the electromechanical model problem and discuss how it generalizes to other situations .

1405.6965

conformal mapping techniques are powerful tools for analyzing two dimensional models , and have been applied to a variety of problems @xcite including aerodynamics @xcite , fluid flows @xcite , free boundary problems @xcite potential theory / electrostatics @xcite , elasticity @xcite and even computer vision @xcite . abstractly , a problem of interest in the physical plane " is transformed into a simpler problem in the pre - image plane " by an appropriately chosen complex analytic function , i.e. a conformal map @xcite relating the pre - image plane to the physical plane . in addition to generating analytic solutions , conformal mapping methods have also been applied in conjunction with numerical techniques to solve various problems of physical interest @xcite . an important issue in applying numerical ( discretized ) conformal maps to two dimensional problems where the domain has sharp regions / near singularities is the phenomenon of _ crowding_. crowding refers to the tendency for image nodes to accumulate in the physical plane near regions of high boundary curvature and leave low curvature regions extremely coarsely represented . crowding plagues all standard numerical conformal mapping methods when used to generate image domains with highly curved boundaries @xcite , and there are specialized algorithms to mitigate the effects of crowding for conformal mappings to _ known _ polygonal domains @xcite . in this paper , we consider a model electrostatic free boundary problem where the interface can develop regions of extremely high curvature . we develop a matched asymptotics method for conformal maps , that completely avoids the computational difficulties associated with crowding for _ a priori unknown domains_. an interesting interplay between interface geometry and induced stresses exists in electromechanical systems . when a conducting interface is deflected by an external field , the induced charge distribution /the local electrostatic pressure is higher in the parts of the sheet with high curvature . this in turn causes even higher deflections and sharper curvatures . this can produce extremely sharp and nearly singular shapes , as well as _ the pull - in instability _ , a runaway effect due to the positive feedback between curvature and electrostatic forces , whereby no equilibrium solutions exist beyond a critical forcing strength . for these reasons we choose an electromechanical model as a test case for applying our techniques for multi - scale conformal maps . this model also has connections to recent studies that have been primarily driven by the development of microelectromechanical systems ( mems ) , including micro - scale capacitors and actuators @xcite . the paper is organized as follows : in section [ sec : model ] we present the basic electromechanical model , a two fluid system where the interface is governed by a balance between electrostatic forces , gravity and surface tension . in section [ sec : conformal ] we present a reformulation of our model problem in terms of conformal mappings , and discuss the advantages of such a reformulation . in section [ sec : collocation ] we present the basic numerical method for constructing numerical conformal maps , and apply it to situations where the free boundary is not sharp " . the method breaks down as the interface develops sharp features , and in section [ sec : multi_scale ] we present our matching technique for conformal maps that applies in a regime where the free boundary is sharp . we present a concluding discussion in section [ sec : discussion ] . we consider a two fluid system an infinite conducting fluid initially in the half - space @xmath0 , and an infinite dielectric fluid in the half space @xmath1 . gravity acts in the @xmath2 direction , and the density of the upper fluid @xmath3 is less than the density of the lower fluid @xmath4 , so the interface is gravitationally stable . the interface is deflected by a line charge with strength @xmath5 per unit length , along the @xmath6 direction , placed in the upper fluid at a distance @xmath7 above the undeflected interface . since all the induced charge on the fluid will reside on its surface , we can equally well consider a model with an infinite conducting sheet on an elastic foundation . the conducting sheet is initially at @xmath8 and is subsequently pulled up by a line charge placed at @xmath9 . the restoring force from the foundation can be modelled as a linear function of the deflection from @xmath10 , and this agrees with the expression for the hydrostatic pressure on the interface in the two - fluid model , which is also a linear function of the deflection . the setup is depicted in fig . [ fig : generic_model ] , and by translation invariance in the @xmath11-direction , it suffices to consider a two dimensional problem in the @xmath12 plane with a one dimensional interface @xmath13 . the pressure balance along the deflected interface @xmath13 is @xmath14 where @xmath15 is the surface tension of the interface , @xmath16 is the curvature of the interface at @xmath17 , @xmath18 is the dielectric permittivity of the upper fluid , @xmath19 is the unit normal to the interface , and @xmath20 is the magnitude of the electric field on the conducting interface . when forced by a line charge , the system is closed by the equations @xmath21 the electric field @xmath22 vanishes as @xmath23 , so we get @xmath24 as @xmath23 , _ i.e _ gravity stabilizes the interface for large @xmath25 . in order to simplify the equations , we pick units to make @xmath26 . this is equivalent to the non - dimensionalization : @xmath27 for the lengths where @xmath28 is the capillary length , and @xmath29 we will henceforth work exclusively with the non - dimensionalized equations so there is no confusion in dropping the primes . this yields the nondimensional equations [ e : h_gov_pt ] @xmath30 the system has a reflection symmetry @xmath31 , and for simplicity we will restrict our attention to symmetric solutions satisfying @xmath32 . the system has two dimensionless parameters , a non - dimensional charge per unit length ( i.e forcing ) @xmath5 and a non - dimensional aspect ratio @xmath7 , the ratio of the vertical length scale ( the charge location ) and the horizontal length scale , the capillary length . the local electrostatic pressure @xmath33 at each point on the sheet depends on the entire deflection profile @xmath34 through the boundary conditions on @xmath35 . this non - local coupling prevents the direct calculation of closed form solutions of , and a numerical treatment is required instead . we will now reformulate the system in terms of conformal mappings , with the aim being to simplify " the non - local nature of the problem . @xmath36 will denote the complex coordinate in the preimage plane and @xmath37 the coordinate in the physical plane . assume that we have a smooth interface @xmath13 with @xmath24 as @xmath38 . then the upper fluid domain @xmath39 is a proper subset of @xmath40 . the riemann mapping theorem @xcite thus asserts the existence of a bijective conformal map @xmath41 such that @xmath42 , where @xmath43 is the ( open ) unit disk @xmath44 in the preimage @xmath45-plane . this mapping is unique if we additionally impose the normalization " @xmath46 and symmetry @xmath47 where the overbar denotes complex conjugation . the symmetry along with the unboundedness of @xmath48 implies the boundary condition " @xmath49 . the mapping @xmath50 is illustrated schematically in figure [ fig : mapping ] . for the undeflected interface @xmath51 , it is easy to check that the appropriate conformal mapping @xcite is @xmath52 for a deflected interface , assuming it is smooth and converges to the undeflected interface as @xmath23 , we show rigorously ( mathematical details in a subsequent paper ) that the conformal map @xmath50 is a bounded " modification of @xmath53 _ i.e. _ @xmath54 where @xmath55 is analytic on the _ closed _ unit disk @xmath56 , @xmath57 is real , and we have the requirements @xmath58 the representation shows that conformal map @xmath50 is continuous on @xmath59 , i.e the only singularity of the map on the closed unit disk @xmath56 is a pole at @xmath60 with a _ residue @xmath61 corresponding to the boundary conditions @xmath24 as @xmath23 . the interface is then defined in terms of the conformal map @xmath50 parametrically by taking limits from inside the disk , i.e. : @xmath62 this yields @xmath63 , a _ conformal parameterization _ of the the interface . the invariance of laplace s equation under conformal transformations now allows us to solve using the electrostatic potential in the preimage plane to obtain @xmath64 a direct calculation using the parameterization @xmath65 so that @xmath66 shows that @xmath67 \\ \kappa(x(\theta ) ) & = \frac{x'(\theta ) h''(\theta ) - h'(\theta ) x''(\theta)}{(x'(\theta)^2 + h'(\theta)^2)^{3/2 } } = \frac{\text{im}[f_{\theta\theta}(\theta)\overline{f_{\theta}}(\theta)]}{\vert f_{\theta}(\theta ) \vert^3}\end{aligned}\ ] ] the force balance thus reduces to a single equation for @xmath68 : @xmath69 + \frac{\text{im}[f_{\theta\theta}(\theta)\overline{f_{\theta}}(\theta)]}{\vert f_{\theta}(\theta ) \vert^3 } = 0 . \label{eq : new_ode}\ ] ] observe that this equation is _ local _ in @xmath70 unlike the original formulation where the force balance equation is _ nonlocal_. also observe that we have one real " equation for a complex " function @xmath68 , so we should expect additional condition(s ) that determine @xmath68 . for any function @xmath71 that is analytic on a domain containing the closed unit disk , the real and imaginary parts of @xmath72 are not independent . rather , they are related by the periodic hilbert transform @xcite @xmath73 & = -\mathcal{h}[\text{im}[\upsilon ] ] + \text{re}[\upsilon[0 ] ] \\ \text{im}[\upsilon ] & = \mathcal{h}[\text{re}[\upsilon ] ] + \text{im}[\upsilon[0]]\end{aligned}\ ] ] because they are boundary values of conjugate harmonic functions on the unit disk @xcite . the periodic hilbert transform @xmath74 of a periodic function @xmath75 is defined @xcite by @xmath76(\phi ) = \lim_{\epsilon \to 0 } \frac{1}{2\pi}\int_{\epsilon}^\pi \cot\left(\frac{\theta}{2}\right ) \left[\xi(\phi-\theta ) -\xi(\phi+\theta)\right ] d\theta\ ] ] we will rewrite this condition in a slightly different form that exploits the natural mapping between analytic functions on the closed unit disk and fourier series given by @xmath77 where the sums do indeed converge because the coefficients @xmath78 decay exponentially , and @xmath72 . since @xmath71 is analytic at 0 , it follows that @xmath79 these equations are _ local _ in the transform domain ( in terms of @xmath80 ) , but are _ nonlocal _ integral equations in terms of @xmath70 . from , it follows that @xmath81 is analytic on the closed unit disk , so choosing @xmath82 in yields @xmath83 the symmetry condition implies @xmath84 finally , we remark that the normalization is _ nonlocal _ in terms of @xmath68 since @xmath85 and the boundary condition can be recast as @xmath86 equations , together give a reformulation of in terms of a ( normalized ) conformal parameterization of the interface . the key advantage of this reformulation is that , every equation is local either on the unit circle in the preimage plane ( i.e. in terms of @xmath70 ) or in the fourier transform domain ( i.e. in terms of @xmath80 ) . since we can efficiently switch between functions on the unit circle and their fourier transforms using ffts @xcite , this reformulation , if discretized appropriately , will give an efficient numerical algorithm for solving . to solve for the interface numerically , we will discretize the representation for @xmath50 in . for clarity we use a tilde to denote discretized numerical approximations , _ e.g. _ @xmath87 is a numerical approximation to @xmath50 . since every analytic function on the unit disk has a convergent taylor series @xcite , we can define families of conformal maps with finitely many degrees of freedom by truncating the power series for @xmath55 centered at 0 : @xmath88 in our numerics we always take @xmath89 , a power of 2 . then @xmath90 given by restricting @xmath87 to the unit circle automatically satisfies . explicitly @xmath91 the symmetry requirement implies that @xmath92 are all real , so the family of conformal maps in satisfying and is parameterized by @xmath93 real quantities . the normalization yields @xmath94 and the boundary condition implies that @xmath95 we now turn to discretizing . let us first consider the problem of determining @xmath96 given @xmath5 and @xmath7 . in general , we can not expect the ( numerical ) residual @xmath97 \equiv \frac{q^2}{4 \pi^2 \vert \widetilde{f}_{\theta}(\theta ) \vert^2 } - \text{re}[\widetilde{f}(\theta ) ] + \frac{\text{im}[\widetilde{f}_{\theta\theta}(\theta)\overline{\widetilde{f}_{\theta}}(\theta)]}{\vert \widetilde{f}_{\theta}(\theta ) \vert^3 } } \label{eq : residual}\ ] ] to vanish for all @xmath98 for our numerical maps with finitely many degrees of freedom . the residuals at points @xmath70 and @xmath99 are _ not independent _ since @xmath100 so that @xmath101(\theta ) = r[\widetilde{f}](-\theta)$ ] . consequently , we get a possible discretization of for every choice of @xmath102 distinct points @xmath103 such that @xmath104 for any pair @xmath105 , without loss of generality , we can take these @xmath102 points in @xmath106 , i.e with non - negative angles . as we discussed in sec . [ sec : conformal ] , we will get efficient numerical algorithms if we can easily switch between values of the function @xmath107 and its fourier coefficients @xmath78 . this can be achieved using ffts if we pick the points @xmath108 uniformly spaced on the unit circle and @xmath89 . in particular , for the discretization in , we can apply the fft to compute the function values and derivatives at evenly spaced nodes @xmath109 , @xmath110 . the quantities @xmath57 and @xmath111 are now determined from , and the conditions @xmath112(\theta_m ) = 0 $ ] for @xmath113 . in implementing this method , we find that the numerical algorithm is better conditioned if we instead parameterize the maps by the @xmath93 quantities @xmath57 and @xmath114$].in this parameterization the boundary condition implies that @xmath115 , so we have @xmath116 independent real parameters . we also have @xmath116 real conditions from the vanishing of the residual at @xmath117 and the normalization . the coefficients @xmath118 and the height " parameters @xmath119 are of course related by a linear transformation which can be expressed in terms of the discrete fourier transform . the final point to consider is that we _ do not expect _ the system will have a unique global solution for all given @xmath5 and @xmath7 , since we expect the system to display multiple equilibrium solutions for some choices of @xmath5 and @xmath7 based on the analogy with the mems model @xcite . as in the analysis of the mems model , we use a continuation method @xcite with the tip height of the interface , @xmath120 and the charge location @xmath7 as the specified parameters . @xmath7 and @xmath120 are thus the independent quantities , and the equilibrium equations determine the charge @xmath5 and the other heights @xmath121 as functions of @xmath120 and @xmath7 . given @xmath7 and @xmath120 , we compute the residual @xmath101(\theta_m)$ ] as a function of @xmath122 through the following steps : 1 . compute the fourier coefficients @xmath118 corresponding to the height parameters @xmath123 where @xmath115 . 2 . compute @xmath124 and @xmath125 using the inverse fft and the representations @xmath126 and @xmath127 . 3 . compute the residual @xmath101 $ ] at the points @xmath128 using . we then compute a numerical approximation of the true deflection profile and corresponding charge strength @xmath5 by minimizing @xmath129(\theta_m))^2 $ ] using a nonlinear least squares method implemented in the matlab function lsqnonlin @xcite . note that every ( local ) minimum for the sum of the squares of the residual is _ not necessarily an approximate solution _ to . we in fact need that the _ residual vector be zero _ to within a prescribed tolerance . figure [ fig : orig_bifn_diags ] shows examples of bifurcation diagrams computed using this method for @xmath130 . for all these choices of @xmath7 , the @xmath131 vs. @xmath5 curves exhibit saddle - node bifurcations i.e. they turn around " at a critical value of @xmath132 . for values of @xmath133 , the system has _ no solution_. physically , if we perform the experiment of increasing @xmath5 quasi - statically from 0 past the critical value @xmath134 , the tip height @xmath135 will track the lower branch of the appropriate curve in figure [ fig : orig_bifn_diags ] , until @xmath136 at which point the conducting fluid will be drawn to the point charge grounding " it . this is the the analog in the two - fluid system of the pull - in instability at finite forcing strength @xmath5 as reported for small aspect ratio mems @xcite . deflections below and above the saddle - node bifurcation are stable and unstable respectively depending on the sign of @xmath137 . indeed , along the upper branch , a small perturbation of @xmath120 away from the equilibrium value @xmath135 will result in an interface that is subject to a higher forcing than necessary to support it ( if the deflection is toward the charge ) or a lower forcing than necessary ( if the deflection is away from the charge ) . in either case , the interface will evolve dynamically and either be drawn to the point charge or relax to the ( stable ) equilibrium on the lower branch , so the upper branch of equilibria is unstable to small perturbations . the solid dots on the figure represent the largest @xmath120 ( for each @xmath7 ) for which the method converges with @xmath138 . while this limiting @xmath120 increases with @xmath102 , it does so very slowly and we can not get much closer to the vertical axis for any value of @xmath139 ( the bound on @xmath102 reflects our limitations on memory size and processor speed ) . nonetheless , the figure is very suggestive that , for all @xmath7 , the bifurcation curve approaches the point @xmath140 ( the solid dots on the vertical axis ) . figure [ fig : orig_solutions ] shows the numerically computed solutions for different tip heights @xmath120 with @xmath7 fixed at 1 . the figure shows both stable ( @xmath141 ) and unstable profiles . the highest profile that we can compute before the collocation method breaks down with @xmath138 is @xmath142 . it is evident from the figure that the tips ( region near @xmath143 ) are getting very sharp on the scale of the capillary length ( nondimensionalized to 1 ) as @xmath120 approaches @xmath7 . this reflects _ non - uniformity / multiple scale behavior _ of the solutions of . in particular , the small scale structures are governed by a different balance ( surface tension vs electrostatic pressure ) from the large scale structure ( surface tension vs gravity ) . a natural approach that will accounts for this non - uniformity it to discretize the problem _ adaptively _ , i.e. in a manner which reflects the local length scale in the solution . one would like to have roughly equal number of points to resolve each of the regions with a different dominant balance . a weakness of the conformal mapping method , as described in section [ sec : collocation ] is that it is intrinsically _ non adaptive_. we do not get to pick the points @xmath108 which are the nodes " in our discretization of the continuous system . rather , they are required to be uniformly spaced on the preimage plane in order to use fft techniques . figure [ f : nodes ] shows the locations in the physical plane ( i.e. on the interface ) of uniformly space nodes on the unit circle for @xmath144 . contrary to our initial expectation , the breakdown of the collocation method is not due to poor resolution of the tip ; rather it is due to poor resolution of the decaying region of the interface , where the dominant large - scale " balance is between gravity and surface tension , and the relevant length scale for this region is the capillary length . this effect is easily understood by recognizing that the electric field on the interface is related to the inter - node separation in the physical plane since @xmath145 . since the electric field ( and the induced charge ) concentrates near sharp tips , the nodes in the physical plane will concentrate near sharp tips . the tendency for image nodes to accumulate near regions of relatively high interface curvature , known as _ crowding _ , is ubiquitous when using conformal maps defined on the unit disk @xmath43 to represent interfaces that feature very disparate length scales . it has been recognized as a substantial impediment to using numerical methods to find conformal maps for regions that have multiple length - scales / singularities in their boundary @xcite . this is not a peculiarity of the conformal mapping based method : alternative methods that we could have applied , including boundary element methods based on the boundary integral formulations , or direct minimization of the total system energy , must all be adapted to deal with the disparate length scales that characterize stiff problems . another approach to avoid crowding is to pick the nodes adaptively ( and possibly non - uniformly ) in the preimage plane . in this case , fft methods do not apply . the hilbert transform constraint relating the real and the imaginary parts of the interface parameterization @xmath68 yields a linear system with a matrix that has a poor condition number @xcite . this constraint has to be imposed along with the minimization of the residual , and wilkening @xcite has applied this idea with careful and sophisticated numerical methods to analyze the crests of large amplitude standing water waves , which were conjectured to have a corner singularity in the free interface @xcite . in section [ sec : matching ] below , we detail a different approach to this problem , one that borrows from the philosophy of matched asymptotics in multiple scale analysis @xcite . the collocation method breaks down when the tip of the interface @xmath120 approaches the charge location @xmath7 . this motivates the definition @xmath146 , as the relevant small parameter that defines this asymptotic regime . when @xmath147 is small , we expect that the tip is sharp and the total charge @xmath148 induced on the sheet accumulates ( and hence , image nodes crowd ) near the high - curvature sheet tip . as @xmath149 , all the charge concentrates on the tip , so to leading order in @xmath147 , we can model the charge on the interface as a concentrated charge @xmath148 at the tip @xmath150 . the interface is then effectively supported by an upward electrostatic point force of strength @xmath151 applied to the tip , and relaxes under the influences of gravity and elasticity everywhere else ( where the electric field / induced charge is zero in this leading order approximation ) . the resulting leading order deflection profiles are governed by a collapsed form of the equation , @xmath152 which can be integrated exactly . multiplying by @xmath153 and integrating in from infinity yields @xmath154 where the constant term on the right hand side has been determined by applying the asymptotic boundary condition . we can solve this equation , provided that @xmath155 , to get the implicit representation @xmath156 for @xmath157 , we see that the first , third and fourth terms in the above expression are bounded and @xmath158 . consequently , for large @xmath159 , we have @xmath160 integrating across @xmath143 , we see that @xmath161 is continuous but has a corner ( discontinuous slope ) at @xmath162 : @xmath163 where @xmath164 is the total force on the interface from the charge ( or equivalently , by newton s third law , the force on the charge pulling it towards the interface ) . this equation is a statement of the balance between the surface tension forces on a corner and the total electrostatic force on the interface . we will call the solutions in the leading order , large deflection solutions , because they are the appropriate limit solutions as @xmath149 ( tip curvature @xmath165 ) and they can not be obtained as a perturbation expansion in powers of @xmath5 about the @xmath166 undeformed interface . in particular , they are not smooth and have a corner at @xmath162 , unlike the perturbation series solutions in powers of @xmath5 which are necessarily smooth . we define a parameter @xmath167 such that @xmath168 denotes the outside angle of the leading order interface tip corner , so that @xmath169 ( see fig . [ fig : schematic_solns ] ) . now yields @xmath170 where @xmath171 takes values in @xmath172 $ ] . figure [ fig : schematic_solns ] shows the comparison between the solutions in and the numerically computed profile @xmath34 from the collocation method with @xmath173 . note that the solutions agree very well outside the tip region , but in the tip region , there are consistent deviations between the leading order , large deflection solution and the numerical solution of the interface . electrostatic forces are important near the tip , and our crude approximation of the electrostatic force is insufficient to resolve the tip region . an important point to note is that @xmath174 . this brings up the question of how to determine the appropriate value of @xmath175 that corresponds to a given @xmath120 with the charge location fixed at @xmath7 . we address this question in more detail below , but the basic idea is that the right " parameters to describe the sharp - tip asymptotic regime are @xmath175 and @xmath176 . in terms of these parameters , @xmath177 and @xmath7 are given by asymptotic expansions @xmath178 , @xmath179 where @xmath180 depends on @xmath175 but not on @xmath147 . leading order large deflection profiles ( @xmath181 , @xmath182 , dashed curves ) corresponding to computed interfaces from fig . [ fig : orig_solutions ] ( @xmath183 . @xmath184 , solid curves ) . the inset shows a zoomed view near the point @xmath185 . the width of the region in which the leading order large deflection approximation is inaccurate approaches zero as @xmath186 and the outer tip angle is @xmath187 . ] combining eqs . , and , we obtain @xmath188 for fixed @xmath7 . in order to compare this expression with the numerical result in fig . [ fig : orig_bifn_diags ] , we need to compute ( or estimate ) @xmath175 for a given @xmath120 and @xmath7 . as we argue above @xmath189 is @xmath190 , so we get @xmath191 this relationship fully describes the leading order bifurcation diagram behavior @xmath192 in the small-@xmath147 regime where numerical methods face the greatest challenges . figure [ f : pt_leading_h_vs_q_small_l ] compares a typical computed bifurcation diagram from fig . [ fig : orig_bifn_diags ] to the corresponding leading order large deflection diagram based on . the leading order approximation is seen to provide a consistent extension of the computed bifurcation diagram to the small @xmath147 regime and captures the overall bifurcation structure of the system even for @xmath158 values of @xmath147 . we now outline a new matching scheme based on numerical conformal mappings that combines high accuracy for all values of @xmath147 with efficient handling of the small @xmath147 regime . the scheme modifies leading order large deflection profiles with a more subtle description of the true tip shape . we adopt the usual notation of matched asymptotics from this point forward , labeling the conformal maps that describe leading order large deflection profiles ` outer ' solutions , and the conformal maps that represent the sheet tip profile ` inner ' solutions . we remark that rather than just the solution profile @xmath193 ( see ) we need an appropriate conformal parameterization @xmath194 for the outer region . the use of conformal maps to represent both inner and outer solutions is essential for maintaining the fundamental relationship between global sheet geometry and electric field that characterizes this problem , _ i.e. _ the localization of the force balance in . in the same complex - variable framework as in the original numerical scheme , outer solutions are governed by the equation @xmath195 + \frac{\text{im}[g_{\psi\psi}(\psi)\overline{g_{\psi}}(\psi)]}{\vert g_{\psi}(\psi ) \vert^3 } = 0 , \label{outer_ode}\ ] ] where @xmath196\rbrace$ ] is the corresponding deflection profile . this is obtained from by dropping the electrostatic pressure term , and it is indeed eq . away from @xmath162 . the relevant quantities in , namely the deflection of the interface and its curvature , are properties of the profile @xmath193 but are independent of the particular parameterization @xmath197 of the interface , with the restriction that @xmath198 is the tip and @xmath199 correspond to the boundary condition @xmath200 as @xmath23 . thus , if a conformal map @xmath201 on @xmath43 has boundary values @xmath197 which satisfies , then so will the composite map @xmath202 where @xmath203 is any conformal automorphism from the closed unit disc @xmath204 to itself ( * ? ? ? * chapter 8) that fixes the points @xmath205 and @xmath206 . since all conformal automorphisms defined on @xmath204 also map the unit circle @xmath207 to itself , these compositions alter the parametrization but not the shape of the original deflection profile . this symmetry of the equation is akin to gauge freedom " in other physical systems , where one is allowed to make certain transformations which keep all the physical " quantities ( here the deflection profile ) invariant . the collection of conformal automorphisms of @xmath204 fixing @xmath208 and @xmath209 forms a one parameter group of mappings @xmath210 with the group composition law @xmath211 we use a collocation method , as in sec [ sec : collocation ] to compute a numerical conformal map that ( approximately ) solve . an important consideration in developing a collocation method to solve for a conformal map , is to first construct a finite dimensional family of conformal maps , that is _ adapted _ to the particular problem we are trying to solve . in particular , the family of maps in are appropriate for interfaces which decay to zero and are otherwise smooth . this family is therefore _ not appropriate _ for seeking solutions of the outer equation . while the outer profile @xmath193 still decays as @xmath23 , the profile is no longer smooth it has a corner with an outer angle @xmath212 at @xmath143 . in order to incorporate this corner with a know angle , we consider a family of conformal maps given by the composition of the family of smooth maps in with a fixed analytic map which generates a corner . we define the corner map " @xmath213 by @xmath214 where @xmath215 is an arbitrary parameter . this map is analytic and single valued except on a branch cut which we can choose to be segment @xmath216 $ ] . also , @xmath217 , and the image of the imaginary axis under the map @xmath218 will have a corner at @xmath175 with the correct outer " tip angle @xmath168 . finally , we emphasize that the parameter @xmath215 is arbitrary and a useful numerical check of our method is to verify that our results are independent of the specific choice we make for @xmath219 . as we discuss in appendix [ apndx : outer ] , we get an appropriate family of candidate maps for approximating the outer solution by composing this corner map with the family of maps in . in particular , we seek ( approximate ) solutions of the outer equation within the family of conformal maps @xmath220 we numerically determine the constants @xmath118 and @xmath57 that give the outer solution by requiring that the function @xmath221 satisfies the appropriate normalization and boundary conditions , and at the _ collocation points _ @xmath222 . the details of our numerical implementation are presented in the appendix [ apndx : outer ] . the sharp corners of outer solutions are regularized by elastic tension close to the tip as shown in fig . [ fig : schematic_solns ] . electrostatic pressure and sheet curvature are both high near such geometry , so we anticipate the existence of a scaling in which the stabilizing effect of gravity is relatively small . as evident from fig . [ fig : schematic_solns ] and from the fact that the outer solutions give a corner at @xmath143 , the break down of the outer solution occurs on a length scale @xmath147 that is set by the separation between the charge location and the interface tip . thus we set @xmath223 ( equivalent to scaling all physical lengths by @xmath147 but recognizing that the tip is within @xmath190 of the prescribed @xmath175 ) and @xmath224 ( motivated by the scaling in ) in . dropping terms that are @xmath190 results in a balance between the electrostatic pressure and surface tension : @xmath225}{\vert \gamma_{\theta}(\theta ) \vert^3 } = 0 \label{inner_ode}\ ] ] where @xmath226 and @xmath227 are both @xmath158 , and @xmath228 is the corresponding rescaled deflection profile near the sheet tip . this equation , along with the condition @xmath229 has two one - parameter symmetries . we can translate the solutions by a real constant @xmath230 , because the equation only sees derivatives of @xmath226 , and we have not as yet imposed any boundary conditions on @xmath226 which break this translation invariance . the second symmetry is the scaling @xmath231 for all @xmath232 . this is related to physical observation that without gravity ( and hence without the capillary length scale ) , the equation should have a scale invariance , as it has no intrinsic length scale . as in the previous section , we therefore do not expect unique solutions to without specifying additional _ normalization(s ) _ of the solution . and with an appropriate normalization , we expect to find a one parameter family of solutions that are determined entirely by the specified value of @xmath233 or equivalently @xmath234 , as this was the only input in determining the normalized outer solution . we now discuss the appropriate normalization conditions as well as the discretization of in order to numerically compute the inner solutions which describe the tip region . an immediate observation from the form of , or with gravity neglected , is that the curvature is non - positive everywhere . in order to match the outer solutions ( which have a corner at 0 ) , it is necessary that as @xmath235 , the inner profile @xmath236 should approach a pair of straight line asymptotes , that make an ( outer ) angle @xmath212 . this situation is depicted schematically in fig . [ fig : inner_schematic ] . let @xmath237 denote the conformal map on the ( open ) unit disk @xmath43 whose boundary values on the unit circle determine @xmath226 . we normalize away the scale invariance for @xmath237 by setting the separation between the tip of the interface and the charge location at @xmath208 , and incorporate the value of @xmath238 by setting boundary conditions " on the behavior of @xmath239 as @xmath240 : [ e : pt_in_bcs_recast ] @xmath241 & \to 0 \text { as } \omega \to -1 \label{e : pt_in_bcs_recast_2}\end{aligned}\ ] ] where @xmath242 and @xmath243 are constants whose values are determined as part of the numerical solution . we refer to @xmath242 as the ` asymptotic offset ' of a computed solution . we still have the translation symmetry @xmath244 for any @xmath245 . for our numerical solutions , we need a family of conformal maps that is adapted to the boundary conditions for the inner solution . as we show in appendix [ apndx : inner ] , an appropriate family of conformal maps for the inner solutions is @xmath246^{\frac{1}{\eta } } \label{inner - conformal}\ ] ] that depends on @xmath247 real parameters @xmath248 and @xmath243 . to get a unique solution , we break the translation invariance of the solutions by specifying the additional requirement @xmath249 the constant @xmath250 is ( somewhat ) arbitrary and the restriction @xmath251 is explained in the appendix . the values @xmath118 and @xmath57 that give the outer solution by requiring that the function @xmath252 satisfies the appropriate normalisation and boundary conditions , and at the _ collocation points _ @xmath253 . ( details in appendix [ apndx : inner ] ) . for intermediate values of @xmath238 ( not too close to @xmath254 or equivalently @xmath175 not close to @xmath255 ) and @xmath250 moderately close to @xmath256 , the algorithm for computing inner solutions produces curves with all of the desired properties ( fig . [ f : pt_in_comp_prof ] ) . a check of our numerical method is that the physical results should be independent of the specific choice of @xmath250 ( as with the choice of the parameter @xmath219 in computing the outer solutions ) . for instance , the asymptotic offset @xmath242 can be computed in terms of the coefficients in the map @xmath257 as @xmath258 . the symmetries of the equation now imply that we should have the identity @xmath259 for some constant @xmath260 depending only on @xmath238 and independent of @xmath250 . this relation is not respected by computed solutions if @xmath250 becomes either very close to @xmath256 or larger than a ( _ a priori _ unknown ) threshold value ( fig . [ f : pt_in_offset_vs_t ] ) . for the value of @xmath238 used in these demonstrations , the expected linear behavior is only displayed for @xmath261 . in addition to giving a check on our numerical method , computing profiles for several @xmath250 values allows us to identify the range that produces consistent solutions . now that we can compute inner and outer solutions separately , we need to combine them to produce an approximate solution of the full system . this is not straightforward , as conformal maps are rigid " global objects . conformal maps on the disk are necessarily smooth ( infinitely differentiable ) , and the real and the imaginary parts must satisfy the nonlocal relation . consequently , we can not simply patch together two conformal maps on different portions of the disk , to produce a composite conformal map . in fact , if two conformal maps agree in a small neighborhood of a single point , then by analytic continuation they necessarily are identical , i.e. they have a common maximal domain of analyticity and the maps are identical on this domain @xcite . since the inner and the outer conformal maps are not identically equal ( for instance near the tip ) , they are different everywhere " , and any map that agrees with one can not agree with the other . at first sight one might conclude there _ there is no way to put two distinct conformal maps together_. of course , we do not require that the composite map agree _ exactly _ with either the inner or the outer conformal map . rather the combined map should approximate the inner and the outer conformal maps well " on different pieces of the unit disk . we will now turn this qualitative notion into a quantitative algorithm for combining inner and outer conformal maps to generate a composite map to solve with _ a uniformly small error_. as discussed above the outer equation and the inner equation have nontrivial symmetries . in order to compute ( unique ) numerical solutions of these equations , we break these symmetries by imposing additional normalization conditions . we now have to un - normalize " i.e. , pick the appropriate symmetry transformed versions of the computed inner and the outer solutions , so that , together they give a solution for the original system . this is achieved by the _ matching principle _ , which requires that there exist an overlap region where both the inner and outer solution describe the true physical solution , and thus agree with each other _ asymptotically _ @xcite . the full ( symmetry related ) family of conformal maps describing the inner solution is given by @xmath262 , where @xmath263 is the _ inner variable _ and @xmath264 are elements of the symmetry group , describing scaling and vertical translation of the inner solution . the outer limit " of this inner solution is the behavior of this map as @xmath240 and from our representation for the inner solution a direct calculation yields ( see also ) @xmath265 \\ & + \epsilon o((1+\omega)^{\frac{1}{\eta}}).\end{aligned}\ ] ] the full ( symmetry related ) family of conformal maps describing the outer solution is given by @xmath266 , where the reparameterization invariance of the outer solution implies that the _ outer variable _ @xmath45 has no intrinsic physical meaning unless it is @xmath208 ( corresponding to the tip ) or @xmath209 corresponding to the point at @xmath267 . rather if @xmath45 and @xmath268 are two variables related by a conformal automorphism that fixes @xmath269 , @xmath270 then both @xmath271 and @xmath272 are valid descriptions of the outer solution . in contrast , the inner variable @xmath273 does have intrinsic physical meaning ( @xmath274 at the charge location , for instance ) . thus , fixing the symmetry element in the reparametrization group is the same as determining @xmath275 such that the inner solution @xmath276 matches with the outer solution @xmath277 in an overlap region . we write this relation in shorthand as the " outer variable @xmath278 , meaning the outer variable is identified with that particular choice of @xmath45 for which the matched outer solution is given by the normalized map @xmath279 computed numerically as in section [ sec : outer ] . a direct calculation reveals that for @xmath278 @xmath280 so to match the inner and outer solutions , it is convenient to represent the outer limit of the inner solution and the inner limit of the outer solution in terms of @xmath281 and @xmath282 respectively . van dyke s matching rule @xcite requires that the outer limit of @xmath283 agree with the inner limit of @xmath284 . we can implement this rule by expanding @xmath283 in @xmath285 , @xmath284 in @xmath286 and using the relation @xmath278 between @xmath45 and @xmath273 . this yields the conditions @xmath287 the quantities @xmath288 and @xmath289 are defined in and . they are obtained by solving problems and that are _ independent _ of @xmath147 . consequently they are @xmath158 quantities which can only depend on @xmath175 and not on @xmath147 . matching thus gives the requirement @xmath290 where we have defined an @xmath158 constant @xmath291 . the overlap function is calculated using in @xmath283 to be @xmath292 from the success in applying van dyke s matching rule it follows that 1 . in the outer " region @xmath293 , the inner solution agrees very well with the overlap function , 2 . in the inner " region @xmath294 , the outer solution agrees very well with the overlap function , 3 . there is a non - trivial overlap region where @xmath294 and @xmath293 . in this region , all the three functions @xmath295 and @xmath296 agree with each other , where by agree well " we mean that the leading terms in an asymptotic expansion in @xmath149 are identical . this argument can also be recast in terms of _ matching through an intermediate variable _ @xcite . for any given @xmath297 and complex number @xmath298 with @xmath299 , there is a sufficiently small @xmath18 such that for all @xmath300 , @xmath301 , the open unit disk . from the relation between @xmath45 and @xmath273 , we get @xmath302 substituting these expressions in @xmath283 and @xmath284 respectively , we get @xmath303 so the first two terms in the asymptotic expansions of the inner and outer conformal maps agree as @xmath149 with @xmath298 fixed . using @xmath304 and @xmath305 as the boundaries " of the overlap region , we determine that the inner and the outer profiles agree in physical " space for @xmath306 . this makes good physical sense because the the overlap region should be all the length scales between the tip curvature scale where electrostatics balances curvature and the capillary length where gravity balances curvature . in this intermediate regime curvature dominates both gravity and electrostatics , so the solutions correspond to @xmath307 , i.e. straight lines . figure [ f : pt_map_overlap ] shows the overlap region , described by straight lines , where the inner and outer solutions agree . [ f : pt_mat_overlap ] we can now construct a composite numerical conformal map @xmath87 on @xmath43 which is a uniformly valid approximation @xcite ( i.e works both in the inner and the outer region ) by @xmath308 this map , as a linear combination of conformal maps , is manifestly a conformal map on the unit disk . for @xmath309 small , i.e. in the outer region @xmath310 and for @xmath311 small , i.e. in the inner region @xmath312 . this composite conformal map therefore approaches the right limits " asymptotically in the inner and the outer regions ( which overlap ) and is thus an approximate solution with a uniformly small error . the interface profile is obtained by taking the limit @xmath313 . our matching procedure is schematically illustrated in figure . [ fig : two - circles ] . since we construct the composite map by solving for the inner and the outer conformal maps separately , the underlying domain on which the numerical conformal map @xmath87 is defined is naturally interpreted as a _ pair of ( discretized ) unit circles_. the outer and the inner variables are related by @xmath278 and the matched profile in the ( physical ) @xmath11-plane is given by @xmath314 as in . note the non - uniformity in the relative distribution of the inner an the outer nodes . we emphasize that , for matching conformal maps , the overlap region , as described by the intermediate variable in is _ two dimensional_. in particular , our matching procedure requires that the inner and the outer solutions agree on a open set in the complex plane , and it does not suffice , in general , to only have agreement of the inner and the outer profiles , i.e the functions representing @xmath34 . on the unit disks in the @xmath273- and @xmath45-planes , the overlap region is given by the common region between @xmath315 ( the outer region ) and @xmath316 ( the inner region ) . we indicate this region in the @xmath273- and @xmath45-planes by hatching in figure [ fig : two - circles ] . we remark that the densities of the inner and outer nodes are comparable in the overlap region but not outside it . this is to be expected as the inner and the outer solutions give equally well resolved numerical solutions in the overlap region . for the schematic plot in figure [ fig : two - circles ] , the densities do not differ by more than a factor of 2 in the overlap region , so there are no more than 4 and no fewer than 1 circle for every 2 asterisks and vice - versa . in contrast , the nodes outside the overlap region are overwhelmingly either circles ( the inner region ) or asterisks ( the outer region ) , showing that in these regions either the inner or the outer solution is inaccurate / poorly resolved . because the inner and outer solutions have their own set of nodes , it suffices to have as few as 16 nodes in each region to get fully resolved matched solutions for the interface . the solution in figure [ fig : two - circles ] was generated with 32 inner and 32 outer nodes . figure[f : family_matched_solns ] show matched interface profiles created by combining a single pair of inner ( 32 node ) and outer ( 128 node ) profiles using different values of @xmath147 . generating profiles in this way ( with fixed @xmath175 ) is extremely efficient once the initial calculations required to determine the appropriate inner and outer solutions have been performed . note that the charge location @xmath7 is not equal for all these interfaces ; rather it varies with @xmath147 . this is a minor drawback when seeking solutions of the forward problem . in order to compute the solutions with a specifies @xmath7 and @xmath120 , we need to invert the ( nonlinear ! ) relation between these parameters , and the asymptotic parameters @xmath175 and @xmath147 . [ f : family_matched_solns ] . observe that the matched profile @xmath317 for @xmath318 ( a few times the capillary length ) in figure [ f : family_matched_solns ] . this disagrees with the physical intuition that the force from the line charge will only deflect the interface upwards . in fact , we can prove rigorously that the sheet can not deflect below the @xmath25-axis for the system . this dip below the axis in our matched solutions , albeit very small , reveals a possible flaw in our matched solution and therefore demands an explanation . in our outer solutions , the balance is between gravity and elasticity , and for large @xmath25 , the profile @xmath193 decays exponentially , as in . for any multipole source , the electric field on an asymptotically flat conductor decays algebraically . far away from the tip , the combination of the forcing charge and the induced charge on the tip is effectively a dipole of strength @xmath319 . consequently , the induced charge decays no faster than @xmath320 for large @xmath25 , and the electrostatic pressure decays as the square of the induced charge @xmath321 . comparing this expression with the hydrostatic pressure in @xmath193 we get a crossover when @xmath322 . consequently , no matter how small is @xmath323 , there always exists an outer - outer region where the dominant balance is no longer between gravity and elasticity . rather , it is between gravity and electrostatics . further , the boundary of this outer - outer region , @xmath324 is not very different from the capillary length scale @xmath325 unless @xmath147 is exceedingly small . a formal calculation keeping higher orders in our composite matched solution confirms the conclusions from this heuristic argument . the outer - outer region is given by the crossover @xmath326 so that for @xmath327 it is no longer true that the presumed dominant term @xmath328 is larger than the ignored higher order corrections in the presumed smaller term @xmath329 . in other words , the composite solution in no longer a valid asymptotic expansion in the outer - outer region , and one has to introduce a new layer " to resolve the non - uniformity in the expansion of the solution in powers of @xmath147 @xcite . this argument also illustrates one of the strengths of the method of matched asymptotics _ the composite solutions have in them , the information about when they are no longer valid _ @xcite . as we argued above , the scaling for this outer - outer region , @xmath330 differs from the outer - scale @xmath331 only by a logarithmic factor , so that for any realistic @xmath147 , there is not a clear separation between the outer and the outer - outer regions . this makes the matching problem difficult , as is the case with other matched asymptotics problems that have logarithmic terms @xcite . we will present a more extensive analysis of the outer - outer region and the associated matching problem in a future publication we also remark that the outer - outer region has a very small effect on the solution in the inner region near the tip , and on the computed bifurcation diagram using the matched solution shown in fig . [ f : pt_full_vs_lead_vs_mat_bifn ] . the figure confirms that the matching technique described in this paper produces uniformly valid , high accuracy solutions ( errors are @xmath190 in the inner _ and _ the outer regions ) for all values of @xmath147 while remaining computationally efficient ( especially ! ) when @xmath147 is small . in this paper , we develop an adaptive numerical conformal mapping method for efficiently computing multiple scale structures in 2 dimensional problems . the central result is that , with our method , we can use conformal mapping techniques to analyze structures that have widely disparate length scales . this is a significant improvement over existing numerical conformal mapping methods for free boundary problems , because they are essentially single - scale " and are plagued by the problem of crowding when applied to multiple - scale problems . as we discuss in sec . [ sec : matching ] , there are good reasons why one would not even consider matched asymptotics for conformal maps , and to our knowledge , this is the first work that successfully combines numerical conformal mappings with the method of matched asymptotics . we consider a model system is governed by three basic forces gravity , electrostatics and surface tension . the interface is governed by multiple dominant balances between different forcing / restoring mechanisms , so the free boundary does develop a multiple scale structure . the model problem is thus a good test bed for developing our method , although it is not ( to our knowledge ) directly relevant to applications . we reformulate the problem in terms of conformal mappings it converts a non - local problem into a problem that is governed by local equations , except some of the equations are in real " space an some are in reciprocal " space ( i.e in the fourier transform domain ) . since the fast fourier transform ( fft ) allows us to efficiently convert from real to reciprocal space and vice - versa , we can get efficient numerical methods for approximating solutions to our problem in terms of discretized conformal maps . this approach leads to the collocation method as described in section [ sec : collocation ] . the fft requires evenly spaced nodes ( in both real and reciprocal space ) , so we are no longer free to choose the node locations in our discretization . this is a particular concern in multiple - scale problems , since the nodes tend to concentrate in regions of high curvature , leaving the rest of the domain under - resolved . this phenomenon , called crowding , is a significant impediment to the using conformal mapping methods for multiple - scale problems . the underlying issue , namely the stiffness " of the governing equations , also shows up in other methods ( boundary integral / boundary element etc . ) , and is responsible for the challenges of numerically studying free boundary problems with widely disparate scales . in our model , we find that the numerical collocation method no longer converges to a solution once the tip curvature gets sufficiently large . the obvious " physical parameters of the model are the non - dimensional charge strength @xmath5 and location @xmath7 . however , the correct " asymptotic parameters in the sharp - tip regime are the uncorrected " tip height @xmath175 and the tip - charge separation @xmath147 . in section [ sec : concentrated ] we use physical arguments to determine the charge distribution on the interface and deduce the appropriate scaling for the the physical parameters in terms of the asymptotic parameters . we emphasize that the analysis in this section _ did not use conformal maps _ , and as such the ideas also extend to three dimensional versions the two - fluid or mems models . using scalings in terms of the asymptotic parameter @xmath147 , we reduce the governing equation to an _ outer _ equation that describes the balance between gravity and surface tension , and an _ inner _ equation that describes the balance between electrostatic forces and surface tension . asymptotic equations _ have non - trivial symmetries , and we determine _ discretized conformal map solutions _ of these equations with appropriate normalization using the collocation method as in section [ sec : collocation ] . it is important that we pick the discretization appropriately to reflect the known structure of the solutions ( a corner in the interface for the outer solution , asymptote to a pair of straight lines with a known angle for the inner solution ) . because each region has a single balance , we are not solving for a multiple - scale structure , and the collocation method works well ( sections [ sec : outer ] and [ sec : inner ] ) . in applying this method to other problems , it is crucial that we identify the correct " asymptotic parameters and scaling relations for the physical quantities , since this knowledge is required to derive the asymptotic limiting equations for the system . we then use the relevant symmetries of the asymptotic equations ( the method is powerful precisely because conformal maps have a rich group of symmetries ) to determine the appropriate matching conditions that allows us to combine distinct conformal maps on different scales to form a single composite multi - scale " conformal map . in section [ sec : matching ] , we discuss this issue in detail for our model system . the key insight is that the natural domain for discretized multi - scale conformal maps is not a single complex plane . rather , it is two ( or perhaps more for other problems ) sets of evenly spaced points on unit - circles , that are related by non - trivial mapping functions ( conformal automorphisms of the unit disk ) . each discretized unit circle resolves one scale ( i.e region with one dominant balance ) in the solution , and the requirement that the functions defined on the various circles agree ( asymptotically ) on their mutual overlaps allow us to patch them together to construct a composite multi - scale conformal map . since each unit circle is discretized evenly , we can still exploit the efficiency of the fft in solving the asymptotic equations . and the the morphing of the unit circle by conformal automorphisms allows us to ( in effect ) pick the nodes in our discretization adaptively and thus resolve structures with disparate length scales in the solution ( see figure [ fig : two - circles ] ) . this idea seems related to the cross - ratios of delaunay triangulation ( crdt ) algorithm for computing schwarz - christoffel transformations for nonconvex / multi - armed polygonal domains @xcite . the crdt algorithm is designed to combat crowding , and a key idea in is to use multiple maps into unit circles that are related by conformal automorphisms . these automorphisms are chosen to blow - up " portions of the unit circle to ensure that nodes are not crowded ( locally ) . however , unlike our method of asymptotic matching of conformal maps , the crdt algorithm exploits the invariance of the cross ratio @xcite under conformal automorphisms to ( implicitly ) construct a composite conformal map . further work is needed to explicate the precise connection between our multi - scale method for free boundary problems and the crdt algorithm for given multi - scale " polygonal domains . this work was supported by the nsf through dms grant 0807501 . we are also grateful to two anonymous referees whose comments on an earlier version helped significantly improve the presentation and discussion of our results . the outer solutions are entirely determined by a single parameter @xmath175 . given a value @xmath332 , we use to compute the appropriate corner angle @xmath212 . in terms of @xmath333 , we define the corner map " @xmath213 by @xmath334 with a branch cut on the segment @xmath216 $ ] . expanding about @xmath335 and @xmath336 , we get @xmath337 we define the candidate maps @xmath338 , where the ( generic ) maps @xmath87 are defined in . the maps @xmath279 have the right behaviour at @xmath339 , namely a corner with outer angle @xmath212 at the point @xmath340 . as before , the @xmath341 symmetry of the outer profile @xmath193 implies that the coefficients @xmath342 are all real . let @xmath343 and @xmath344 . using the definition of @xmath279 and the behavior of @xmath345 as @xmath346 , we see that as @xmath347 , @xmath348 \to h_{out}(0)+b(-1 ) -b(1 ) + \frac{t(\eta-1)}{\eta}\ ] ] the boundary condition that @xmath200 as @xmath349 yields @xmath350 finally we have one ( real ) degree of freedom from the allowed conformal reparameterization of the outer solutions . we fix this with a gauge condition " @xmath351 . defining @xmath222 , we have @xmath352 independent ( real ) equations , which are respectively the vanishing of the residual at @xmath353 equally spaced collocation points , the gauge fixing normalization and the boundary condition as @xmath23 . @xmath354 & \equiv - \text{re}[g(\psi_m ) ] + \frac{\text{im}[g_{\psi\psi}(\psi_m)\overline{g_{\psi}}(\psi_m)]}{\vert g_{\psi}(\psi_m ) \vert^3 } = 0 , \\ b(-1 ) & \equiv \sum_{j = 0}^{m_{out } } ( -1)^j \beta_j = 0 \\ b(1 ) & \equiv \sum_{j = 0}^{m_{out } } \beta_j = h_{out}(0 ) + \frac{t(\eta-1)}{\eta}\end{aligned}\ ] ] this matches up with the number of independent parameters in the candidate conformal parameterization of the interface @xmath355 motivated by , we consider @xmath356 a straightforward expansion in @xmath357 yields @xmath358 recalling that @xmath359 , we see that even after subtracting the pole at @xmath360 , @xmath361 is not analytic at @xmath360 in contrast to the maps in . consequently , in our numerical approximation @xmath257 , we need to incorporate this non - analytic behavior as @xmath240 into the discretization . we define a @xmath247 dimensional family of conformal maps by @xmath362 @xmath363 is the number of nodes we are choosing to represent the inner solution . @xmath364 is a map from the preimage @xmath273 plane to an intermediate plane . the map into the physical plane is then given by @xmath365^{\frac{1}{\eta}}$ ] . the map from the intermediate plane to the physical plane has branch points at @xmath256 and @xmath267 and we choose the branch cut along the negative real axis . in order that the image @xmath366 not intersect this branch cut , we need that @xmath367 . a conformal parametrization of the interface for the inner solution is given by @xmath252 . the parameters @xmath368 and @xmath369 for the solution are then determined by the following conditions @xmath370(\theta_m ) = \frac{q^2}{4 \pi^2 \vert \widetilde{\gamma}_{\theta}(\theta_m ) \vert^2 } + \frac{\text{im}[\widetilde{\gamma}_{\theta\theta}(\theta_m)\overline{\widetilde{\gamma}_{\theta}}(\theta_m)]}{\vert \widetilde{\gamma}_{\theta}(\theta_m ) \vert^3 } = 0,\ ] ] is the vanishing of the residual pressure at the collocation points @xmath371 ; @xmath372 is essentially the boundary condition as @xmath23 since it ensures that @xmath373 so that the boundary condition holds with @xmath374 ; and @xmath375 are normalizations for picking a unique element of the two parameter symmetry group of the inner equation . we thus have @xmath376 ( real ) equations to determine @xmath376 real parameters @xmath377 and @xmath378 and the constants are determined by minimizing the sum of the squares of the discrepancies in the above @xmath376 conditions . 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 _ _ ( , ) link:\doibase 10.1017/s0022112092001927 [ * * , ( ) ] link:\doibase 10.1016/j.euromechflu.2007.02.001 [ * * , ( ) ] @noop * * , ( ) @noop _ _ , vol . ( , ) @noop _ _ , vol . ( , ) @noop * * , ( ) @noop _ _ ( , ) pp . , @noop _ _ , vol . ( , ) in @noop _ _ ( ) pp . in @noop _ _ , vol . , ( , ) @noop * * , ( ) in @noop _ _ ( ) pp . link:\doibase 10.3390/s100606149 [ * * , ( ) ] @noop _ _ , vol . ( , ) @noop _ _ , vol . ( , ) @noop _ _ ( , ) link:\doibase 10.1137/s0036139900381079 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1103/physrevlett.107.184501 [ * * , ( ) ] @noop * * , ( ) @noop _ _ ( , ) @noop _ _ , vol . ( , ) @noop _ _ , ed . ( , ) pp . , @noop _ _ , vol .

we report a phase switch monte carlo ( psmc ) method study of the freezing line of the lennard - jones ( lj ) fluid . our work generalizes to soft potentials the original application of the method to hard sphere freezing , and builds on a previous psmc study of the lj system by errington ( j. chem . phys . * 120 * , 3130 ( 2004 ) ) . the latter work is extended by tracing a large section of the lennard - jones freezing curve , the results for which we compare to a previous gibbs - duhem integration study . additionally we provide new background regarding the statistical mechanical basis of the psmc method and extensive implementation details . pacs numbers : 64.70fx , 68.35.rh

cond-mat0511167

the freezing of a disordered fluid into an ordered crystalline solid undoubtedly represents one of the most spectacular examples of thermodynamics in action . but despite its ubiquity and familiarity , key aspects of the phenomenon remain poorly understood across a variety of systems @xcite . principal among the reasons for this is the difficulty of obtaining accurate simulation estimates for fluid - solid phase coexistence properties . in this section we outline the most commonly used contemporary approaches for tracing freezing boundaries , identify their key strengths and weaknesses , and highlight recent developments in the search for improvements . the staple method for locating fluid - solid coexistence is thermodynamic integration ( ti ) @xcite . here , the free energy of each phase ( fluid ( f ) and crystalline solid ( cs ) ) is computed for states covering a range of densities , using integration methods which connect their thermodynamic properties with those of reference states of known free energy . the two branches of the free - energy are then matched to determine the freezing parameters . while there is much to commend ti ( principally its simplicity ) it turns out to be less than ideal in a number of respects @xcite . these have been documented in detail elsewhere ( see eg . @xcite ) ; but briefly , the main difficulty is one of identifying a suitable reference state together with a reversible phase space pathway by which it can be reached ; a poor choice of integration path may encounter singularities both real and artificial @xcite which compromise the calculation . additionally , corrections may be required to allow for the fact that the path does not quite reach the idealized reference state @xcite . moreover , the method focuses on the absolute free energies of the phases @xcite , rather than the quantity of interest their free energy _ difference_. finally , there currently exists no reliable and comprehensive method for estimating errors on phase boundaries computed via ti . obtaining a whole f - cs phase boundary using ti is a computationally challenging task . however , knowledge of one point on the f - cs coexistence boundary ( obtained , for instance , by using ti ) can be used to bootstrap a gibbs - duhem integration ( gdi ) scheme to trace the entire curve without further calculation of free energies . the gdi method @xcite exploits the generalized clausius - clapeyron equation to express the slope of the phase boundary entirely in terms of single - phase averages . this is clearly a virtue since it avoids the need for two - phase sampling . however , _ without _ any ` reconnection ' of the two configuration - spaces at subsequent simulation state points , the gdi approach offers no feedback on integration errors . since there will generally exist a _ band _ of metastable states on each side of the phase boundary , it is possible for the integration to wander significantly from the true boundary with no indication that anything is wrong . as a result , the calculation of meaningful uncertainties is problematic . gdi has been used in a number of studies , most notably in the context of freezing of hard and soft spheres @xcite . more recently , attention has shifted to developing methods for tackling the problem of locating f - cs coexistence via _ direct _ measurements of free energy differences . to do so , one must construct a reversible sampling path between the phases . the obvious `` physical '' path , traversing the region where both phases coexist whilst practicable when the two phases share the same symmetry ( see eg . ref . @xcite ) turns out to be computationally problematic in the f - cs context . the main reasons for this are the large degree of metastability of the two phases , the extended timescale for the crystallization process , and the tendency of any crystal formed to exhibit vacancies , interstitials and other defects . accordingly , recent effort has focussed on trying to identify alternative inter - phase routes for the f - cs problem . one such alternative route , called constrained fluid-@xmath0 integration has been proposed by grochla@xcite and extended by elke et al @xcite . it involves the controlled transformation of the fluid phase to the solid phase in a series of stages . during the initial stage the fluid is gradually changed to a weakly interacting fluid by reducing the strength of interparticle interactions . at the next stage , a set of gaussian potential wells of prescribed width are switched on at the sites of a crystalline lattice of the appropriate symmetry . simultaneously , the volume of the system is gradually changed from a value typical of the fluid to one typical of the solid . the particles of the weakly interacting fluid are then allowed to diffuse to the lattice sites , where they are captured by the gaussian wells . the final stage involves gradually restoring the particle interactions to full strength , whilst simultaneously switching off the gaussian potential wells . integration of free energy derivatives is used to estimate the free energy difference along the path . the constrained fluid-@xmath0 integration method was tested by application to the lennard - jones fluid where it was used to determine two coexistence state points . this test led the author to conclude that the method `` can not be said to be computationally efficient '' @xcite , at least in its present form . this finding presumably reflects the rather long and fragmented nature of the inter - phase path . although each of the stages of the path was initially reported to be reversible @xcite , further application of the method @xcite found this to be contingent on the correct tuning ( via trial and error ) of four separate constants relating to the characteristics of the gaussian potential wells . indeed , it is not clear to us that one can generally expect _ a - priori _ that the stated path will be trouble - free ; for example , it seems hard to rule out that the transformation of the original fluid to a weakly interacting fluid might intersect a first order ( pseudo ) liquid - vapor transition . furthermore we speculate that for large systems the reliance on particle diffusion to translate particles to unoccupied lattice sites may become problematic . very recently , mastny and de pablo @xcite have proposed a method based on measurements of the density of states , which aims to directly estimate the free energy difference of f and cs phases . the rationale for the method is its authors assertion that : `` to connect the free energy of the solid and liquid phases , there must exist a portion of energy and volume space that can be simultaneously sampled by both solid and liquid phases '' . to exploit this premise , a series of simulations of the lennard - jones system were performed , each restricted to and sampling the configuration space of one of a set of overlapping `` windows '' in the space of the total energy @xmath1 and volume @xmath2 . successive windows were positioned appropriately such as to form a path linking the two well - separated regions of @xmath3 associated with the typical configurations of the respective phases . along this path , a central range of @xmath3 was indeed reported to be found for which both liquid and solid phase configurations could be sampled . a simple average @xcite of the solid phase and liquid phase density of states was accumulated in the central range and joined for continuity to the measured forms of the liquid and solid density of states on either side . in our opinion , the existence of a range of @xmath1 and @xmath2 that can be sampled by both liquid and solid phases is insufficient to connect the respective branches of the density of states because these are ( like their underlying sets of characteristic configurations ) , fundamentally _ distinct_. instead , for a proper connection , the phases must be linked via a continuous ( and reversible ) path through configuration space . in the context of the method of mastny and de pablo , this necessitates repeated ( back and forth ) _ transitions _ between the two pure phases . since no mention is made in ref . @xcite of any such transitions , the validity of the method would appear to be questionable @xcite . another very recent approach to the f - cs problem , which bears some resemblance to that of ref . @xcite , is the `` multi - nph '' method of escobedo @xcite . here a path is constructed in the _ enthalpy _ of the system , spanning the range of values between those typical of the respective equilibrium phases at some prescribed pressure . the system temperature is ascertained at each point along the path , and exhibits a discontinuity at some value of the enthalpy as the favored phase of the system changes . ti with respect to the temperature variations along the path yields the free energy difference . the problem reported with this method was that it was apparently necessary for equilibration purposes to initialize the system in the cs phase at all points along the enthalpy path . however , it was difficult to be sure whether , for a given enthalpy value , the system had relaxed to the phase of minimum free energy , and hence whether the ti result was unbiased . this seems to us to be a manifestation of the familiar drawback of ti , namely the difficulty of finding a reversible integration path . phase switch monte carlo ( psmc ) @xcite is a further relatively recent method designed to directly link f and cs phases via a reversible sampling path . the method builds on previous work @xcite in which it was demonstrated that the configuration spaces associated with two phases of a many - body system can both be visited in a single monte carlo simulation , by harnessing extended sampling methods to facilitate a direct switch from one phase to the other . the method , which is quite generally applicable , was initially deployed in a study of hard sphere freezing @xcite . subsequently , it was applied to soft potentials by errington @xcite , who used it to calculate two points on the freezing line of the lennard - jones system for a number of system sizes . the present paper complements and extends errington s psmc study of the lj system . it is organized as follows . in sec . [ sec : statmech ] we review some of the key results in the statistical mechanical formulation of f and cs phase configurational weights , and show how to construct , on this basis , a theoretical framework for computational estimation of free energy differences . sec [ sec : method ] describes how this framework is exploited in principle , and realized in practice , by the psmc method . an application of the method to the lennard - jones system is reported in sec . [ sec : results ] where we present estimates of large portions of the freezing curve for a number of system sizes . these we compare with the results of a previous gdi study , bootstrapped using ti at a single coexistence point . finally sec . [ sec : concs ] details our conclusions and the prospects for further applications . within the framework of statistical mechanics , the free energy of a given phase is expressible in terms of its _ configurational weight_. information regarding this weight can be accumulated via a simulation in the course of an exploration ( sampling ) of the microstates of the phase . while for fluid phases such an exploration encounters no great obstacles , a complication arises in the case of crystalline solids . specifically , it transpires that the relevant phase space of the solid is effectively fragmented into a number of mutually inaccessible regions . each fragment corresponds to a distinct permutation of the particles with respect to the lattice sites . the solid phase configurational weight can only be estimated for the single fragment in which it is initiated . since on symmetry grounds the weight contribution of every fragment is identical , the overall configurational weight of the solid is obtained by multiplying the measured weight of one fragment by the number of fragments . in this section we first show how to correctly count this fragment number . we then proceed to describe how the ratio of configurational weights of two phases is related to the ratio of their _ a priori _ probabilities ( a quantity directly accessible to simulation ) and thence to the free energy difference . consider a periodic system of @xmath4 classical particles confined to a volume @xmath2 , variable under a prescribed constant external pressure @xmath5 , and in equilibrium with a heat bath at prescribed constant temperature @xmath6 . within this constant-@xmath7 ensemble the configurational weight of a phase may be written as @xmath8 here , @xmath9 labels the phase , while @xmath10 with @xmath11 the configurational energy @xcite . the @xmath9-label on the integral stands for some _ configurational constraint _ that picks out configurations @xmath12 that ` belong ' to phase @xmath9 . in the present work we shall be concerned with phases which are either fluid ( @xmath13 ) or crystalline solid ( @xmath14 ) in character , and choose to formulate the constraint in a form which reflects the situation actually encountered in monte carlo simulations of single phases . specifically , let @xmath15 denote some representative configuration of phase @xmath9 . then the constraint may be regarded as picking out those configurations which can be reached from @xmath16 on the simulation timescale . the timescale is presumed to be sufficiently long to explore ` one phase ' but still short compared to ( unaided ) inter - phase traverses . it is convenient to use the sites defined by @xmath16 as the origins of the particle coordinates . thus we define a set of displacement vectors @xmath17 where @xmath18 and write @xmath19 . in the case of the fluid phase , particles have the run of the entire system and hence _ all _ contributing configurations are reachable from any one ; we may write simply @xmath20 where @xmath21 is some specific but arbitrary fluid configuration , which can be selected at random in the course of mc exploration of the fluid phase . for the crystalline phase it is natural to choose @xmath22 to define the sites of a lattice of the appropriate symmetry and scale . but here one must recognize that the complete cs configuration space actually comprises a number of distinct mutually inaccessible _ fragments _ @xcite corresponding essentially to the different permutations of particles amongst lattice sites . by symmetry each fragment should contribute equally to the configurational weight . but in general , the typical timescale for particle interchanges between lattice sites greatly exceeds the accessible simulation timescale . in these circumstances mc simulation will visit ( and thus count ) only the states within the fragment in which it is initiated . the total configurational weight of the cs phase must therefore be obtained by multiplying the measured contribution of one fragment by the total number of fragments . since global translation ( permitted by the boundary conditions ) ensures that _ one _ fragment includes _ all _ possible locations of any chosen particle , the number of fragments is the number of ways of assigning the other @xmath23 particles to @xmath23 wigner - seitz cells of some underlying notional fixed lattice @xcite . this number is @xmath24 . thus @xmath25 in order to make contact with the simulation methodology to be described below , it is expedient to define the total _ a priori _ probability of phase @xmath9 . this is given by the ratio of the configurational weight of phase @xmath9 to the total weight of the two phases @xcite : @xmath26 combining this last equation with eqs . [ eq : zcaldef ] , [ eq : zsimpf ] and [ eq : zsimpcs ] , allows the ratio of the configurational weights of the two phases to be expressed in terms of the ratio of their _ a priori _ probabilities @xcite : @xmath27 the link between the configurational weight of a phase and its gibbs free energy is forged by the definition : @xmath28 inserting this into eq . [ eq : zrat ] , the gibbs free - energy - density difference between the phases can be written in the ( strategically suggestive ) form : @xmath29 eqs . [ eq : zrat ] and [ eq : gibbs ] provide a foundation for mc studies of fluid - solid phase coexistence . their formal promise is as follows : given an inter - phase sampling scheme that visits both the f and cs phases , @xmath30 ( and hence @xmath31 ) may be obtained as the ratio of the _ a priori _ probabilities of the phases . the latter quantity can be estimated simply as the ratio of the number of visits to each phase recorded in the course of a simulation . in order to exploit this framework , however , one must first design a singularity - free phase space path linking the f and cs phases , and then formulate a sampling strategy to traverse it . phase switch monte carlo is one realization of such a scheme . the relative stability of the f and cs phases is determined by the ratio of the associated configurational weights @xmath32 ( eq . [ eq : zrat ] ) . to measure that ratio one needs a mc procedure which visits both solid and fluid regions of configuration space in the course of a single simulation run . the phase switch mc method is a general approach that facilitates such two - phase sampling . its principal feature is a phase space ` leap ' @xcite that directly maps a pure phase configuration of one phase onto a pure phase configuration of the other . the motivation for choosing such an inter - phase route for the f - cs problem is that it circumvents interfacial ( mixed - phase ) configurations and their attendant sampling difficulties ( see sec . [ sec : intro ] ) . in this section we describe how to apply the phase switch method in this context . in each phase ; the displacement vectors @xmath33 connect the centers of the distinguishable ( numbered ) particles to these sites . the switch operation shown swaps the representative sites of one phase for those of the other phase , whilst maintaining @xmath33 constant . more generally , any desired transformation of @xmath33 ( eg . a scaling see text ) can be incorporated in the switch @xcite . the particular configuration @xmath33 shown is a `` gateway '' state ( see text ) because the magnitude of the effective energy change under the switch is small.,width=340 ] the key to implementing the phase switch is the representation ( eq . [ eq : origin ] ) of particle coordinates in term of displacements with respect to a representative configuration @xmath34 . by construction , the system may be transformed between the cs and f phase representative configurations simply by switching the vectors ( @xmath35 ) . thus by continuity , _ any _ cs ( f ) configuration ` sufficiently close ' to the representative one will also transform to a f ( cs ) state under this operation @xcite . this phase switch can itself be realized as an mc step , so that the phase label @xmath9 becomes a stochastic variable . by ` sufficiently close ' here , we mean that the energy change associated with the phase switch is such as to afford a reasonable chance of acceptance . fig . [ fig : psschem ] provides an illustration of the mechanism . we term the set of configurations for which the mc phase switch will be _ accepted _ the ` gateway ' states . in general , however , these gateway states constitute only a small fraction of the total respective configuration spaces of the phases , and consequently will rarely ( if ever ) be visited on simulation timescales . to ensure effective two - phase sampling , the probabilities with which the gateway states are visited must be _ enhanced_. this can be achieved by appeal to extended ( biased ) sampling methods . [ fig : blobs ] shows a schematic representation of the procedure . cs phase space replica fragments ( see text).,title="fig:",width=302 ] + the bias necessary to promote sampling of the gateway states is administered with respect to an ` order parameter ' some suitably chosen macrovariable of the system . the freedom in selecting this order parameter is circumscribed by the requirement that the associated microstates form a contiguous path through phase space linking the large number of equilibrium ( typical ) microstates to the relatively small number of gateway states . in the next subsection , we describe one choice of order parameter which has proved itself successful in this regard . one can devise a variety of order parameters that serve to form a suitable path leading from the equilibrium states to the gateway states . a definition that we have found to be effective is a variant of one originally employed by one of us in ref . @xcite , and which closely resembles a recent proposal by errington @xcite . here the order parameter comes in two parts ( or modes ) : ` tether ' and ` energy ' . the tether mode serves to draw particles close to the representative sites to which they are nominally associated . then , once all particles are within a prescribed distance of these sites , tether mode switches off and an energy mode order parameter takes over to guide particle to the gateway states for which the phase switch energy cost is sufficiently small to be accepted . to elaborate , let @xmath36 denote the order parameter in mode @xmath37 and phase @xmath9 . then for tether mode we write @xmath38 and define an associated order parameter @xmath39 where @xmath40 is the distance of particle @xmath41 from its lattice site measured in units of the box length , and @xmath42 is a prescribed dimensionless threshold radius . only particles whose displacement @xmath43 exceeds this threshold contribute to @xmath44 . the tether mode is active iff @xmath45 for at least one particle @xmath41 , i.e. when @xmath46 . otherwise control hands over to the ` energy ' mode @xmath47 ; its associated order parameter is defined by @xmath48 where @xmath49 measures the change ( under the phase switch operation ) of the hamiltonian @xmath50 with respect to its value @xmath51 in the representative microstate @xmath34 , with the latter scaled to the instantaneous volume of phase @xmath9 @xcite . the presence of the logarithm in eq . [ eq : emode ] is designed to moderate the scale of the contribution of the energy cost to @xmath52 , which might otherwise become excessively large for particles with a strongly repulsive core to their interaction potential . given these definitions , and recalling eqs . [ eq : zsimpf ] and [ eq : zsimpcs ] , the entire region of accessible configuration space can be described by the ensemble @xmath53 where @xmath54 is the _ effective hamiltonian _ defined by @xmath55 here @xmath56 represents a set of _ multicanonical weights _ @xcite , associated with the macrostates of the macrovariable @xmath57 . the set can usefully be split into four ` branches ' , one for each combination of mode and phase . the role of the weights is to modify ( with respect to the boltzmann sampling ) the probability with which the various @xmath58 values are visited . more specifically , as discussed below , we tailor their form in order to ensure roughly _ uniform _ sampling over the relevant range of @xmath58 . simulations in the ensemble described by eq . [ eq : ztot ] allow one to measure the multicanonical probability distribution @xmath59 the form of which is accumulated in practice via a simple list ( see sec . [ sec : data ] ) . the bias introduced by the multicanonical weights can be unfolded to give the equilibrium distribution ( to within an unspecified normalization constant ) : @xmath60 the distribution of any single observable can be readily extracted from @xmath61 by integration . similarly the ratio of configurational weights is obtained as @xmath62 from which the gibbs free - energy - density difference follows directly via eq . [ eq : gibbs ] . the monte carlo procedure we have adopted utilizes four types of update , which we describe below . in implementing these updates , it is computationally expedient to hold one particle _ fixed _ at its representative site in each phase . this suppresses the global translation mode in the cs phase and eliminates the need for association swaps ( see below ) in this phase . choosing to make the assignment @xmath63 ( implying that @xmath64 ) , we use @xmath65 to represent sets @xmath33 of displacement coordinates which satisfy this condition . consequences for acceptance probabilities are examined in appendix . [ sec : append1 ] . the mc procedure for one - phase exploration has three types of update : ` particle translations ' , ` association swaps ' and ` volume moves ' . 1 . _ ` particle translations'_. this is a standard mc procedure : a site ( identified by one of the vectors in @xmath66 or @xmath67 ) is chosen at random and a candidate state is chosen by incrementing the position coordinate of the particle associated with that site by a random vector whose components are drawn from a zero - mean uniform distribution of prescribed width . this operation changes both @xmath68 and @xmath65 2 . _ ` association swaps'_. in this operation we choose two distinct sites at random ( @xmath41 and @xmath69 say ) and identify the corresponding displacement vectors @xmath70 and @xmath71 . the candidate state is defined by the replacements + @xmath72 + this process can be regarded as a change of association : the particle which was associated with site @xmath69 is now associated with site @xmath41 ( and _ vice versa _ ) . it changes the _ representation _ of the configuration ( the coordinates @xmath33 ) ; but it leaves the physical configuration invariant . + we apply it only to the fluid phase where particles can wander far from their representative sites and need to be reined back in order to reach the gateway states . one _ may _ implement association updates in the cs phase too : the simulation can then be thought of as exploring different cs - phase fragments ; the factor of @xmath24 in eq . [ eq : zsimpcs ] is no longer needed and it is no longer necessary to clamp one particle . ` volume moves'_. the volume is also updated in the conventional way , by a random walk of prescribed maximum step size , with particle position coordinates @xmath73 and representative sites @xmath74 rescaled . note that we maintain the ratio @xmath75 constant throughout , so the notional ` conjugate ' phase @xmath76 also undergoes a dilation by the same factor . in each of the above cases the transition to the candidate state is accepted ( see appendix [ sec : append1 ] ) with the probability @xmath77\right\}\end{aligned}\ ] ] with @xmath78 all three types of mc move may involve a change in the order parameter @xmath58 and hence a change @xmath79 in the multicanonical weights implicit in @xmath80 ( cf . . [ eq : effpot ] ) . note that particle translations or association swaps that hand over control from one order parameter mode to another , involve a weight change which is calculated from the respective branches of the weight function . for example , for a hand - over from tether mode to energy mode , in which the order parameter changes from @xmath81 to @xmath82 the appropriate weight change is calculated as @xmath83 1 . ` _ inter - phase switch _ ' . the final type of mc update is the phase switch , which entails replacing one set of the representative configuration vectors , @xmath34 say , by the other , @xmath84 . if the equilibrium densities of the two phases were close to one another it would also be possible to keep the volumes constant in the switch . but that is not the case in general . accordingly if the procedure is to be efficient the switch needs to incorporate an appropriate _ volume scaling _ of the system @xcite . it is natural to fix that scaling so that a ` typical ' volume @xmath85 of phase @xmath9 is matched to a ` typical ' volume @xmath86 of phase @xmath76 . in practice we take @xmath87 to be the mean volume of phase @xmath9 . the switch is accepted ( see appendix [ sec : append1 ] ) with the probability + @xmath88\right\}\end{aligned}\ ] ] + where + @xmath89 + the phase switch occurs only from states in which the energy mode order parameter is active ; the associated change in multicanonical weights is + @xmath90 + where @xmath91=@xmath92 are the new and old order parameter values respectively . this weight discontinuity is at our disposal ; we tune it so as to cancel the typical contribution to the acceptance probability associated with the switch . when discussing our results ( sec . [ sec : results ] ) we shall occasionally refer to a `` monte carlo sweep '' . this we take to comprise @xmath4 trial particle translations , @xmath4 trial association swaps , @xmath93 trial volume move and @xmath93 trial phase switch . in the course of the simulations , observations are stored as lists @xcite @xmath94 ( @xmath95 ) . estimates for the distribution of any observables were accumulated by appropriate binning and weighting of the members of the list @xmath96 with @xmath97 etc . in analyzing the data , extensive use was made of histogram extrapolation ( he ) . this method greatly enhances computational efficiency by facilitating the scanning of a region of pressure and temperature around each simulation state point , without recourse to further simulation . to achieve this , he _ reweights _ the data obtained at one state point @xmath98 to yield estimates appropriate to another not - too - distant state point @xmath99 . the reliable range of the extrapolation in @xmath5 and @xmath6 is dependent on the statistical quality of the original data . for further information on the he technique , we refer the interested reader to ref . its utility in the context of tracking a freezing line will be illustrated in sec . [ sec : results ] . prior to collection of phase switch data , it is necessary to determine a suitable set of multicanonical weights @xmath56 to enable the system to pass readily between the equilibrium and gateway states . in practice the requisite weights are defined with respect to the _ macrostates _ formed by discretising ( binning ) the order parameter macrovariable @xmath57 . for a given mode @xmath37 and phase @xmath9 , the bin width @xmath100 was chosen such that for all pairs of successive bins @xmath101 and @xmath102 , the corresponding weights satisfy @xmath103 . doing so ensures that the weight associated with any given macrostate is representative of the full range of the underlying macrovariable , thereby guaranteeing a reasonable mc acceptance rate . consideration of eqs . [ eq : effpot ] and eqs . [ eq : sampdist ] shows that the form of the weights that confers equal probability on every macrostate @xmath58 ( and hence ensures that all are well sampled ) is @xmath104 . for adequate sampling one therefore requires a prior estimate of @xmath105 . a variety of techniques exist for providing such an estimate , some of which are described in refs . the approach we have utilized in this work is the transition matrix monte carlo method ( tmmc ) @xcite . tmmc extracts information on macrostate probabilities by focusing on the statistics of _ transitions _ between these macrostates . the method proceeds by defining a collection matrix , @xmath106 , in which information regarding the macrostate transitions is accumulated . at every mc step , the proposed transition is recorded in @xmath106 ( irrespective of whether or not it is actually accepted ) according to : @xmath107 where @xmath108 is the acceptance probability of the move under the hamiltonian @xmath109 ( rather than the effective hamiltonian @xmath110 ) . the macrostate transition probabilities derive from the collection matrix via : @xmath111 where the denominator on the rhs sums over all possible values of the macrostates . knowledge of these transition probabilities permits , in turn , an estimate of the probabilities @xmath105 . although a number of approaches exist ( see eg . ref . @xcite ) , for extracting @xmath105 from t , a simple , yet unbiased and stable method , considers exclusively those transitions which occur between adjacent macrostates ( i.e. examines only the diagonal and first off - diagonal elements of @xmath112 ) . application of the ` balance ' condition for macrostate transitions @xcite : @xmath113 then permits assignment of the weight function via @xmath114 the weight function was updated in this manner at regular intervals ( typically every 1000 monte carlo sweeps ) . each update allowed the simulation to explore an ever wider range of order parameter values @xcite . we note , however , that weight function updates should not be performed too frequently during the weight evolution process because , strictly speaking , they violates detailed balance . empirically , however , at the stated update interval , the solution was found to be highly accurate and the procedure efficient . for the purposes of constructing the weight function , we found it beneficial to decompose the full range of @xmath58 into a number of overlapping ` windows ' . within each window a fragment of the overall weight function was determined and joined for continuity to those of neighboring windows . the weight differences across the three mode transitions ( tether to energy mode for each phase , as well as the phase switch itself ) were found using a simple root - finding algorithm in a simulation restricted to the appropriate extremes of the order parameter . both energy modes were calculated as single windows , while the solid and fluid tether modes were split into 3 and 10 windows respectively ( for the largest system size studied @xmath115 ) . we have employed the phase switch monte carlo method to study the freezing properties of particles whose interactions are described by the lennard - jones potential @xmath116\ : . \label{eq : lj}\ ] ] the system is assumed to be contained within a cubic box of volume @xmath2 , which fluctuates under the control of an applied external pressure @xmath5 and is in thermal equilibrium at temperature @xmath6 . in common with previous studies of freezing in the lj system @xcite , particle interactions were truncated at one - half the box length , and running tail corrections applied to the energy in the standard manner @xcite . as described in sec . [ sec : moves ] , the mc moves employed were particle translations , volume moves , association swaps and phase switches . of these , we note that volume moves and phase switches are computationally inexpensive to implement because the associated energy change is obtainable directly from the scaling with ( respect to coordinate dilations ) of the repulsive and attractive parts of the configurational energy @xcite . association swaps are similarly cheap because they leave the configurational energy unaltered . since the position of one particle was clamped at its representative site in each phase , cs phase association swaps were not required ( see sec . [ sec : moves ] and appendix [ sec : append1 ] ) . the phase switch mc method requires for its operation the specification of a representative configuration @xmath34 for both the cs and f phases . for @xmath66 we adopt a perfect lattice of the appropriate symmetry ( here fcc ) and scale ( corresponding to a rough estimate of the cs phase coexistence density ) . for @xmath67 , we randomly select a configuration in the course of a simulation of the fluid phase @xcite . with regard to the order parameter ( sec sec [ sec : op ] ) , the hand - over from tether to energy mode is controlled by the parameter @xmath42 in eq . [ eq : tether ] . this should be chosen such that at the hand - over particles are sufficiently close to their representative sites that the energy cost @xmath117 of the switch ( eq . [ eq : cost ] ) is not saturated at a large value . otherwise @xmath118 will be unable to guide the particles to positions favorable for the phase switch . we find that satisfactory results are obtained by choosing @xmath42 consistent with @xmath119 at the hand - over . the formalism described in secs . [ sec : statmech ] and [ sec : method ] assumes that the configurational constraint confines the cs phase to the phase space fragment in which it is initiated . while this appears to be strictly observed for f - cs coexistence in hard spheres , as studied in ref . @xcite , for the softer interactions of the lj system , interchange of neighboring particles between lattice sites was observed to occur , albeit very rarely . the simplest way of dealing with this effect is to suppress it ; to which end we introduced an upper bound on the particle displacements in the cs phase . we set @xmath120 , a choice which ( we have verified ) has a negligible effect on the equilibrium properties of the cs phase for the system sizes we have studied . the strategy for obtaining the f - cs coexistence line is as follows . initially one requires a rough estimate for some coexistence state point . this may derive either from any available literature estimates or , alternatively , coexistence can be bracketed by scanning ( backwards and forwards ) a range of pressures at some nominated temperature until the system spontaneously transforms between phases . next it is necessary to customize a suitable set of multicanonical weights which allow the system to sample both phases . this can be done by employing , for example , the transition matrix method described in sec . [ sec : weights ] . an example weight function is shown in fig . [ fig : exampleweight ] for a system of @xmath121 particles close to coexistence at @xmath122 . lj particles at @xmath123 , @xmath124 . the four branches of the weight function have been laid out with respect to the underlying discretisation into bins ( see sec . [ sec : weights]).,width=302 ] having determined a suitable weight function , a production run is then performed to obtain data of high statistical quality . from this data one can readily extract ( see sec . [ sec : data ] ) the ratio @xmath125 ( eq . [ eq : zratio ] ) and hence the free energy difference @xmath126 ( eq . [ eq : gibbs ] ) between the phases . histogram extrapolation ( he ) ( sec . [ sec : data ] ) then permits the location of the coexistence pressure at which this difference vanishes ( i.e. when @xmath127 ) . the corresponding coexistence number densities can then be simply read off from the peak positions in the coexistence form of @xmath128 , where @xmath129 . [ fig : density_hist ] shows a typical coexistence form of @xmath128 , obtained for @xmath121 at @xmath123 . for a system of @xmath121 lj particles at @xmath123 , @xmath130 . a selection of data points are shown . lines are guides to the eye.,width=302 ] in addition to simplifying the determination of the coexistence pressure at a given temperature , he also permits simultaneous extrapolation in the temperature . this facilitates the scanning of a segment of the coexistence line in the @xmath131 plane . although the range of near coexistence @xmath131 values for which a single simulation provides reliable information is necessarily limited , additional simulations can be performed near the vestiges of this range . the data from simulations performed at a number of different state points can then be synthesised self - consistently using multiple histogram reweighting @xcite . as well as guiding the choice of near coexistence values of @xmath5 and @xmath6 for subsequent simulations , he also provides an estimate of the appropriate order parameter distribution @xmath132 , which ( recall sec . [ sec : weights ] ) serves as a suitable weight function . thus by a repeated combination of simulation followed by he , the coexistence line can be straightforwardly tracked , without the need to ever re - determine a weight function _ ab initio _ @xcite . estimates of the freezing boundary have been obtained using the psmc method within the constant-@xmath7 ensemble for systems of sizes @xmath133 and @xmath134 particles . freezing boundary data has also been obtained for @xmath135 , but here the system is sufficiently small that transitions back and forth between f and cs phases occur _ spontaneously _ , over a range of pressures , and a density distribution ( sampling both phases ) can be determined and a coexistence pressure inferred without the need for psmc . spontaneous transitions were also found to occur ( albeit much more rarely ) in the @xmath136 system . here we observed spontaneous freezing as we tracked the coexistence curve using the methods described in sec . [ sec : model ] . we emphasize that this effect was also seen for this system size in a ` bare ' constant-@xmath7 simulation at coexistence and does not therefore appear to be caused ( or rendered more frequent ) by the multicanonical weighting of psmc . the larger system sizes ( @xmath121 and @xmath115 ) did not exhibit spontaneous transitions on the timescale of our simulations , presumably because of the increased free energy cost of forming an interface . the spontaneous freezing that occurs in the @xmath136 system undermines our strategy whereby the phase label @xmath9 keeps track of the current phase . nevertheless , we were able to suppress the problem by tracking the coexistence line along a path parallel to its true trajectory , but displaced slightly towards the fluid side . this was achieved by simulating at a pressure some @xmath137 less than the coexistence value predicted by the histogram extrapolation from the previous near - coexistence state point . the estimate of the coexistence pressure was easily corrected for the imposed shift by means of histogram extrapolation . separate simulation state points for the @xmath135 system size , @xmath138 state points for @xmath136 , @xmath139 points for @xmath121 and @xmath140 points for the @xmath115 system size . also included for comparison are the psmc estimates of errington @xcite for @xmath141 , together with the gdi results of agrawal and kofke @xcite for @xmath115 . * ( b ) * a closeup of the region around @xmath142 . the vertical error bars corresponding to errington s data points @xcite and the horizontal error bars to the gdi study . symbols are the estimates arising from the present study ( given explicitly in tables [ tab : ljpd108]-[tab : ljpd500 ] ) , with uncertainties smaller than the symbol size in each case . lines are interpolations between the data points , based on histogram extrapolation.,width=321 ] plane in the region of the triple point . shown are the estimated solid and fluid coexistence densities for @xmath115 and @xmath136 . also included for comparison are the gdi estimates of agrawal and kofke@xcite for @xmath115 . uncertainties are smaller than the symbol sizes ; lines are guides to the eye , width=321 ] the measured phase boundary in the @xmath131 plane , and a portion of the phase diagram in the @xmath143 plane , are shown in figs . [ fig : ljpd ] and [ fig : pd - pt ] respectively for the various system sizes studied . full data are also tabulated in tables [ tab : ljpd108]-[tab : ljpd500 ] together with their uncertainties . with regard to the latter , we note that analysis of statistical errors within the psmc framework is extremely transparent : at a given temperature , the uncertainty in the pressure is simply given by @xmath144=\frac{\sigma[{{\mathcal{r}}_{\mbox{\sc { f , cs}}}}]}{|\delta \bar{v}|}\ : , \label{eq : error}\ ] ] where @xmath145 , and @xmath146 $ ] is the associated uncertainty in @xmath125 , this being controlled ( at heart ) by the statistics of the inter - phase switch and measurable from a simple block error analysis . we note that owing to the extensive factor @xmath147 in the denominator of eq . [ eq : error ] , the _ minimum _ uncertainty in @xmath125 required to attain a prescribed error in the pressure _ grows _ like @xmath4 @xcite . as with any method , the error analysis becomes more involved if one chooses to combine a number of separate data sets via a multiple histogram analysis @xcite . in these circumstances we have found it useful to perform a bootstrap error analysis with @xmath148 re - samples , taking an estimate of the block size as the correlation length . the error estimate was found to depend only weakly on the block size . all error bars in tables [ tab : ljpd108]-[tab : ljpd500 ] are calculated in this way and correspond to a @xmath149 confidence interval . we stress that it is not strictly necessary to perform this more sophisticated error analysis , adequate ( albeit slightly overly conservative ) error bars can be assigned simply on the basis of block averages . r@lr@lr@lr@lr@lr@l + & & & & & + 0&.01250 & 3489&(8 ) & 2&.5682(18 ) & 2&.5531(10 ) & 1563&.4(6 ) & 1572&.15(31 ) + 0&.01550 & 2635&(9 ) & 2&.4141(11 ) & 2&.3277(13 ) & 1204&.7(3 ) & 1251&.84(54 ) + 0&.01900 & 2039&(3 ) & 2&.3045(4 ) & 2&.2197(2 ) & 1008&.5(1 ) & 1050&.32(8 ) + 0&.02350 & 1549&(2 ) & 2&.1905(2 ) & 2&.1091(1 ) & 813&.48(7 ) & 847&.951(32 ) + 0&.02800 & 1232&(1 ) & 2&.1014(3 ) & 2&.0212(5 ) & 632&.69(7 ) & 660&.221(73 ) + 0&.03500 & 922&(1 ) & 1&.9935(2 ) & 1&.9182(3 ) & 503&.94(5 ) & 526&.283(37 ) + 0&.04500 & 663&.1(9 ) & 1&.8803(2 ) & 1&.8078(3 ) & 389&.12(4 ) & 406&.995(38 ) + 0&.05500 & 507&.7(8 ) & 1&.7935(1 ) & 1&.7235(2 ) & 312&.12(2 ) & 326&.930(22 ) + 0&.06500 & 406&.5(6 ) & 1&.7260(2 ) & 1&.6576(1 ) & 257&.63(2 ) & 270&.276(10 ) + 0&.07500 & 334&.6(9 ) & 1&.6702(7 ) & 1&.6033(3 ) & 219&.03(6 ) & 230&.064(21 ) + 0&.09000 & 262&.0(7 ) & 1&.6026(7 ) & 1&.5383(5 ) & 169&.65(3 ) & 178&.562(29 ) + 0&.11000 & 197&.7(4 ) & 1&.5299(3 ) & 1&.4659(2 ) & 135&.32(2 ) & 142&.881(16 ) + 0&.13000 & 157&.8(3 ) & 1&.4773(3 ) & 1&.4142(2 ) & 110&.82(2 ) & 117&.339(8 ) + 0&.16000 & 117&.2(2 ) & 1&.4111(4 ) & 1&.3477(6 ) & 82&.416(15 ) & 87&.742(24 ) + 0&.17500 & 103&.0(3 ) & 1&.3827(3 ) & 1&.3213(5 ) & 72&.448(1 ) & 77&.294(24 ) + 0&.21000 & 78&.95(13 ) & 1&.3304(3 ) & 1&.2692(1 ) & 56&.230(9 ) & 60&.335(4 ) + 0&.24000 & 64&.68(8 ) & 1&.2940(1 ) & 1&.2327(1 ) & 47&.043(4 ) & 50&.738(5 ) + 0&.30000 & 45&.66(6 ) & 1&.2349(1 ) & 1&.1734(4 ) & 30&.560(4 ) & 33&.483(7 ) + 0&.37000 & 32&.43(5 ) & 1&.1835(1 ) & 1&.1214(1 ) & 22&.669(2 ) & 25&.210(3 ) + 0&.43000 & 25&.12(4 ) & 1&.1505(11 ) & 1&.0863(4 ) & 16&.44(50 ) & 18&.718(10 ) + 0&.50000 & 19&.13(3 ) & 1&.1180(11 ) & 1&.0525(3 ) & 11&.17(34 ) & 13&.163(6 ) + 0&.62000 & 12&.56(4 ) & 1&.0762(12 ) & 1&.0070(3 ) & 5&.21(16 ) & 6&.908(4 ) + 0&.69257 & 9&.840(17 ) & 1&.0557(11 ) & 0&.9844(2 ) & 2&.72(8 ) & 4&.271(3 ) + 0&.80000 & 6&.859(8 ) & 1&.0291(11 ) & 0&.9542(2 ) & -0&.23(1 ) & 1&.157(3 ) + 0&.95000 & 4&.105(10 ) & 1&.0023(11 ) & 0&.9212(1 ) & -2&.98(91 ) & -1&.735(2 ) + 1&.10000 & 2&.236(7 ) & 0&.9812(11 ) & 0&.8925(2 ) & -4&.85(15 ) & -3&.701(2 ) + 1&.20000 & 1&.293(4 ) & 0&.9695(11 ) & 0&.8741(1 ) & -5&.86(18 ) & -4&.752(1 ) + 1&.26282(42 ) & 0&.8173 & 0&.9637(11 ) & 0&.8645(1 ) & -6&.30(19 ) & -5&.207(1 ) + 1&.33510(54 ) & 0&.3150 & 0&.9570(11 ) & 0&.8519(2 ) & -6&.83(21 ) & -5&.755(1 ) + 1&.37193(39 ) & 0&.1003 & 0&.9543(11 ) & 0&.8464(3 ) & -7&.07(21 ) & -5&.994(2 ) + 1&.38077(56 ) & 0&.0466 & 0&.9537(11 ) & 0&.8447(2 ) & -7&.13(21 ) & -6&.051(2 ) + 1&.38564(46 ) & 0&.0250 & 0&.9537(11 ) & 0&.8447(2 ) & -7&.15(22 ) & -6&.080(1 ) + 1&.38674(47 ) & 0&.0150 & 0&.9533(11 ) & 0&.8438(2 ) & -7&.16(22 ) & -6&.087(2 ) + 1&.38963(91 ) & 0&.0010 & 0&.9534(11 ) & 0&.8439(1 ) & -7&.18(22 ) & -6&.105(1 ) + r@lr@lr@lr@lr@lr@l + & & & & & + 0&.39000 & 30&.400(47 ) & 1&.1771(3 ) & 1&.10965(3 ) & 20&.200(22 ) & 22&.8182(6 ) + 0&.43000 & 25&.390(180 ) & 1&.1525(5 ) & 1&.08401(12 ) & 15&.870(22 ) & 18&.2534(30 ) + 0&.47000 & 21&.942(37 ) & 1&.1355(2 ) & 1&.06611(1 ) & 12&.430(13 ) & 14&.6269(1 ) + 0&.54000 & 16&.928(34 ) & 1&.1065(2 ) & 1&.0355(2 ) & 8&.1409(89 ) & 10&.0970(21 ) + 0&.60000 & 13&.667(49 ) & 1&.0854(3 ) & 1&.0130(2 ) & 5&.4899(69 ) & 7&.2971(19 ) + 0&.68000 & 10&.470(23 ) & 1&.0616(2 ) & 0&.9861(5 ) & 3&.1561(39 ) & 4&.8445(52 ) + 0&.68000 & 10&.477(18 ) & 1&.0619(4 ) & 0&.9864(1 ) & 3&.1494(43 ) & 4&.8415(15 ) + 0&.80000 & 7&.068(12 ) & 1&.0332(3 ) & 0&.9532(1 ) & -0&.0465(6 ) & 1&.4633(7 ) + 0&.91000 & 4&.885(13 ) & 1&.0126(3 ) & 0&.9278(1 ) & -2&.1525(28 ) & -0&.7639(17 ) + 1&.00000 & 3&.546(9 ) & 0&.9990(3 ) & 0&.9098(1 ) & -3&.5496(44 ) & -2&.2385(8 ) + 1&.10000 & 2&.340(9 ) & 0&.9851(3 ) & 0&.8901(1 ) & -4&.7187(56 ) & -3&.4660(11 ) + 1&.20000 & 1&.418(6 ) & 0&.9745(3 ) & 0&.8734(1 ) & -5&.7189(62 ) & -4&.5181(11 ) + 1&.29969(94 ) & 0&.6456 & 0&.9651(3 ) & 0&.8560(3 ) & -6&.5252(70 ) & -5&.3518(26 ) + 1&.33481(74 ) & 0&.4263 & 0&.9628(3 ) & 0&.8508(1 ) & -6&.7631(74 ) & -5&.5961(9 ) + 1&.36274(71 ) & 0&.2500 & 0&.9604(3 ) & 0&.8463(2 ) & -6&.9512(74 ) & -5&.7908(18 ) + 1&.38853(64 ) & 0&.1000 & 0&.9587(3 ) & 0&.8422(2 ) & -7&.1148(76 ) & -5&.9582(11 ) + 1&.40025(60 ) & 0&.0300 & 0&.9578(3 ) & 0&.8405(1 ) & -7&.1896(77 ) & -6&.0372(8 ) + 1&.40527(68 ) & 0&.0010 & 0&.9574(3 ) & 0&.8395(1 ) & -7&.2208(77 ) & -6&.0681(6 ) + r@lr@lr@lr@lr@lr@l + & & & & & + 1&.336(2 ) & 0&.5500 & 0&.9655(1 ) & 0&.8566(2 ) & -6&.637(1 ) & -5&.486(2 ) + 1&.3613(3 ) & 0&.3836 & 0&.96337(4 ) & 0&.85274(3 ) & -6&.815(1 ) & -5&.6705(2 ) + 1&.393(1 ) & 0&.2091 & 0&.9617(1 ) & 0&.8490(1 ) & -7&.007(1 ) & -5&.871(1 ) + 1&.418(2 ) & 0&.0200 & 0&.9581(1 ) & 0&.843(1 ) & -7&.198(1 ) & -6&.08(1 ) + ideally one should like to extrapolate coexistence parameters obtained for a range of finite system sizes to the thermodynamic limit using a finite - size scaling ansatz . in the hard sphere case , a clear @xmath150 dependence of the coexistence pressure has been observed using the psmc method @xcite , and there seems no reason to expect a different scaling in the case of the lj potential . however , in the present study ( as well as in that of ref . @xcite ) , a good fit to this scaling form could not be obtained . as noted in ref . @xcite , the reason for this is presumably traceable to the truncation of the interparticle potential cutoff , which was set to be one - half the box length in all cases . this choice , while facilitating direct comparison with existing literature data ( principally ref . @xcite ) , necessarily introduces a degree of coupling between the potential truncation and the system size , notwithstanding the use of a mean - field based tail correction . for the rather limited range of system sizes studied in this work , the use of such a tail correction is unlikely to provide a good approximation to the full potential , particularly for the smaller system sizes . indeed , it is well known that many aspects of the phase behaviour of the lj system are acutely sensitive to the details of the potential cutoff @xcite . in summary , the phase switch mc method allows one to locate directly and transparently fluid - solid coexistence parameters and their associated uncertainties within the appropriate ( constant - pressure ) ensemble . to achieve this the method connects the phases via a direct inter - phase sampling path , thereby facilitating estimates of free energy differences from a single simulation . the course of this path encompasses configurations which are exclusively pure - phase in character . accordingly , one can be confident that it robust , i.e. free of both singularities and crystalline defects . we have applied the psmc method to calculate , for a number of system sizes , significant portions of the @xmath131 freezing curve of the lj fluid to high accuracy ( see the error bars quoted in tables [ tab : ljpd108]-[tab : ljpd500 ] ) . in so doing we have illustrated the strategic advantages of histogram extrapolation ( he ) as an efficient means of tracking the coexistence line . furthermore we have seen ( cf . @xcite ) that he provides reliable estimates of the requisite multicanonical weight function , thereby obviating the need to redetermine the weights from scratch at each successive near - coexistence state point . while it is probably fair to say that the psmc remains a somewhat computationally intensive strategy ( at least in the context of the f - cs problem ) , it is by no means prohibitively so on the scale of its competitors . given the strengths of the method we have identified , we feel it should represent an attractive option , especially when high precision estimates of freezing boundaries are required . we have compared our results with those of the previous psmc study of errington @xcite and the gdi study of agrawal and kofke @xcite for @xmath115 . not surprisingly , our results agree ( to within error ) with those of errington . however , ( and as also noted by errington ) , the results for @xmath115 do not overlap with the error bars quoted in the gdi study of the same system size ( and potential truncation ) by agrawal and kofke @xcite . this is possibly due to the inherent limitations of the gdi scheme as already noted in sec . [ sec : intro ] specifically the lack of any means to reconnect the phases beyond the initial ` starting ' coexistence state point from where the integration is launched . it may additionally reflect a poor estimate of the starting point itself , which was determined using ti by transforming the lj system to a hard sphere reference system via a route which takes in an intermediate system of soft spheres . in particular , the coexistence pressure taken for the hard sphere reference system is now believed to be an overestimate @xcite . as regards the prospects for further applications , it would certainly be worthwhile in the specific case of the lj system to attempt to decouple the system size scaling of the cutoff from intrinsic finite - size effects by repeating the present study for a constant interparticle potential cutoff . this should permit a reliable extrapolation of coexistence parameters to the thermodynamic limit , albeit for the particular choice of cutoff . more generally , it would be interesting to attempt to apply the psmc method to molecular systems . here , however , the order parameter would probably need to be augmented in order to not only draw molecules to their representative sites , but also to align them appropriately . the authors thank jeff errington for providing numerical data from ref . @xcite . here we derive the monte carlo acceptance probabilities used in the psmc method . as described in the sec . [ sec : method ] , we choose to fix one particle at its representative site . the effect of this is to reduce the configurational weight of each phase by a factor of the system volume @xmath2 . thus for the fluid phase , eq . [ eq : zsimpf ] becomes we note that the acceptance formula eq . [ eq : moveacc ] is actually no different to that pertaining to a fully unconstrained constant-@xmath7 ensemble @xcite . indeed a little thought reveals that the act of fixing one particle is simply equivalent to viewing the unconstrained system from the reference frame of that particle . such a change of reference frame has no consequences for acceptance criteria . consider next the inter - phase switch . this replaces one set of representative vectors , @xmath34 say , by the other , @xmath84 , while at the same time scaling the system volume and particle displacements by a constant factor chosen such that a ` typical ' volume @xmath85 of phase @xmath9 is matched to a ` typical ' volume @xmath86 of phase @xmath76 . for a volume update which _ scales _ the current value , one has @xmath168 , and referring back to the detailed balance relation eq . [ eq : db ] one finds eg the singular behaviour in the single - occupancy - cell ( soc ) model : h. ogura , m matsuda , t. ogawa , n. ogita and a. ueda , prog . * 58 * , 419 ( 1977 ) . for anything but hard spheres , in traversing the fluid phase there is also a liquid - vapor binodal to be circumnavigated . the actual implementation swaps replicas between the overlapping windows . these swaps are accepted with probability unity whenever two adjacent replicas are in the overlap region , and are rejected otherwise . in the central region , which has two exactly overlapping windows one for liquid and one for solid trail swaps are accepted with unit probability , yielding a ` density of states ' which is the simple average of those for the solid and liquid branches . at our starting point ( eq . [ eq : zsimpdef ] ) the indices @xmath177 on the coordinates @xmath178 are _ particle _ labels which confer distinguishability . any ` correction for indistinguishability ' is made _ post hoc_. with this perspective the ` fragments ' are ` distinct ' . essentially , the global translation mode can be regarded as allowing the system to access @xmath4 permutations of the particles with respect to the underlying notional _ fixed _ lattice . doing so reduces the actual number of permutations from @xmath179 to @xmath24 . this point ( first emphasized in ref . @xcite ) appears to have been neglected in previous studies of cs phase free energies . in this work , the scaling at the switch is applied to the box size , the representative configuration vectors and the displacement vectors . previous work @xcite chose to keep the displacement vectors constant at the switch . both methods are legitimate provided that the consequences for the switch acceptance probability ( see appendix . [ sec : append1 ] ) are accommodated . to ensure that no two sites of @xmath67 are atypically close to one another ( which might otherwise impede access to the gateway states ) , we ` anneal ' this configuration before adopting it . this entails quenching the system ( at constant volume ) , while suppressing crystallization by constraining particles to remain within a maximum displacement @xmath180 from their starting point . coexistence parameters are determined inherently more accurately by the psmc method than the number and energy densities of the respective phases ( cf . . [ eq : error ] ) . one could therefore reduce ( at relatively little cost ) the uncertainties associated with the densities simply by performing additional single phase simulations at the stated coexistence parameters .

planets are built from planetesimals : solids larger than a kilometer which grow by colliding pairwise . planetesimals themselves are unlikely to form by two - body collisions ; sub - km objects have gravitational fields individually too weak , and electrostatic attraction is too feeble for growth beyond a few cm . we review the possibility that planetesimals form when self - gravity brings together vast ensembles of small particles . even when self - gravity is weak , aerodynamic processes can accumulate solids relative to gas , paving the way for gravitational collapse . particles pile up as they drift radially inward . gas turbulence stirs particles , but can also seed collapse by clumping them . while the feedback of solids on gas triggers vertical shear instabilities that obstruct self - gravity , this same feedback triggers streaming instabilities that strongly concentrate particles . numerical simulations find that solids cm in size gravitationally collapse in turbulent disks . we outline areas for progress , including the possibility that still smaller objects self - gravitate . # 1#2#1#2 epsf.def psfig.sty planet formation , solar system , accretion , planets , circumstellar disks , fluid mechanics , turbulence = 0.5em

0909.2652

the nebular hypothesis that planets coalesce from disks of gas and dust orbiting young stars is confirmed today in broad outline . protoplanetary disks of ages 110 myr are now studied routinely ( figure [ fig : hh30 ] ; @xcite ) . they contain enough mass to spawn planets like those detected around hundreds of gyr - old stars @xcite . debris disks , observable via dust generated from collisions between larger parent bodies , bridge our understanding in the 10100 myr interval @xcite . these parent bodies may represent first - generation planetesimals , the building blocks of planets . our solar system may preserve a record of planet formation , in the size distributions of asteroids @xcite and kuiper belt objects @xcite . the remarkable journey that disk solids make in growing from microscopic dust to earth - mass ( @xmath2 ) planets divides into three legs . at the smallest sizes @xmath3 , chemical bonds and van der waals forces enable grains to stick to one another . at the largest sizes @xmath4 , gravity promotes growth . pairs of objects coagulate upon colliding because their gravity is strong enough to retain collision fragments . at intermediate sizes lies the domain of planetesimal formation . it presents the most challenging of terrains ; here grain surfaces are insufficiently sticky and the gravity of a single object too feeble . @xcite kindled the hope that where the individual fails , the collective might succeed : if solid particles have a large enough density en masse , their gravity can draw them together . the many calculations seeking to realize this vision are the subject of our review . an incompressible fluid body in hydrostatic equilibrium , orbiting a star of mass @xmath5 at distance @xmath6 , has sufficient self - gravity to resist disruption by the star s tidal gravitational field if its density exceeds the roche value @xmath7 @xcite . though we will see that the effects of self - gravity manifest at arbitrarily low densities ( [ sec : draggi ] ) , the roche density is a fine benchmark against which to measure progress towards forming planetesimals , and we will use it as such . it is a formidably large density , 23 orders of magnitude greater than typical densities at disk midplanes ( [ sec_disks ] ) . @xcite , and independently ( * ? ? ? * hereafter gw ) , proposed that densities approaching roche ( actually a factor of 20 smaller , see [ sec : dyncollapse ] ) could be achieved by having dust settle vertically and accumulate in a thin `` sublayer '' at the midplane . @xcite pointed out that large densities were not so easily achieved that turbulence generated by vertical shear across the sublayer would halt settling and prevent gravitational instability ( gi ) . gw appreciated that the sublayer would be turbulent , but overlooked the decisive impact of turbulence on the layer s thickness and density . this stalemate has been broken indeed the entire playing field redrawn in the last decade by abandoning three assumptions . first , the height - integrated surface density ratio of dust to gas need not be solar . enriching the disk in metals , say by photoevaporation of gas @xcite or by radial drifts of particles ( * ? ? ? * hereafter ys ) , can stop vertical shear turbulence from forestalling gi ( @xcite , ys , @xcite ) . second , turbulence , of whatever origin , is not always the enemy of gi . magneto - rotational tubulence is observed in numerical simulations to generate long - lived structures in gas that can trap particles @xcite . third , gas drags particles , but by newton s third law , particles also drag gas . properly accounting for the backreaction of particles on gas leads to powerful drag instabilities that can concentrate particles and seed gi @xcite . this review is organized as follows . section 2 establishes order - of - magnitude properties of protoplanetary disks using an up - to - date model for the minimum - mass solar nebula . attention is paid to primitive chondritic meteorites . section 3 gives a primer on the aerodynamics of grains , treating gas drag in the `` test particle '' limit where backreaction is neglected . we discuss how grain dynamics are affected by gas turbulence . section 4 describes the extent to which grains can grow by sticking . section 5 rehearses and comments on toomre s criterion for gi and how the criterion is removed when self - gravity is combined with gas drag . section 6 treats sublayer shearing instabilities . section 7 introduces secular drag instabilities that concentrate particles without recourse to self - gravity . we offer new insights into goodman & pindor s @xcite toy model for drag instabilities . section 8 outlines the nonlinear outcome of gi , describing breakthrough numerical simulations . finally , a summary is supplied in section 9 , where we also list a few forefront problems . our study is peppered throughout with calculations that readers are encouraged to reproduce and improve . our review complements others in the field of planet formation . @xcite do more justice than we have to the growth of particle aggregates by grain - grain sticking . @xcite provide a pedagogical and cutting - edge review of protoplanet accretion by gravitationally focussed , pairwise collisions . for a review of planetesimal formation that overlaps ours and covers some topics in greater detail , see the les houches lectures by @xcite . a starting point for calculations is the minimum - mass solar nebula ( mmsn ) , derived by adding enough h and he to solar system planets to restore them to solar composition , and spreading the augmented masses into abutting annuli centered on their orbits ( e.g. , * ? ? ? this exercise is uncertain , since estimates of the total `` metal '' ( non - h and non - he ) content of jupiter range from 10 to 42 @xmath2 ; for saturn the range is 15 to 30 @xmath2 @xcite . even if these uncertainties were reduced , it is unlikely our solar system formed strictly from the mmsn . the conversion of disk metals into planets can not be 100% efficient , and current planetary orbits may differ significantly from their original ones ( e.g. , * ? ? ? * ) . moreover a major theme of this review is that planetesimal formation may require the metal fraction to evolve to values greater than that of the bulk sun , at least in certain regions of the disk . thus the utility of the mmsn lies in providing a baseline for discussing real - world complications , and in establishing orders of magnitude . for numerical estimates in this review , we adopt a disk surface density : @xmath8 where subscripts @xmath9 and @xmath10 denote gas and particles ( condensed metals ) , respectively . our mmsn ( @xmath11 , @xmath12 ) , uses the updated condensate mass fraction for solar abundances of 0.015 @xcite . when methane ice sublimates above temperature @xmath13 , @xmath14 ; above @xmath0@xmath15 , water and all other ices are lost so that @xmath16 @xcite . the coefficient for eq . ( [ eq_sigmad ] ) is chosen to give @xmath17 of solids in an annulus centered on the earth s orbit when @xmath18 and @xmath16 . we invoke values of @xmath19 to account for various metal enrichment processes , depicted in figure [ fig : enrich ] and discussed throughout this review ( e.g. , @xcite , ys , @xcite ) . integrated to @xmath20 , eq . ( [ eq_sigmag ] ) yields a mass of @xmath21 . astronomical observations of disks orbiting myr - old , sun - like stars suggest they contain 0.0010.1 @xmath22100 @xmath23 of gas and dust @xcite . disk masses are derived from mm - wave radiation from dust , located @xmath0100 au from host stars . from the dust emission is calculated a dust mass , and from the dust mass a gas mass is extrapolated assuming a solar dust - to - gas ratio . the gas mass thus imputed is uncertain because the disk metallicity is not known , and because the modeled dust mass depends on an unknown grain size distribution . disk masses may be systematically underestimated because grains may have grown to sizes @xmath24 mm and would therefore be practically invisible at sub - cm wavelengths ( e.g. , * ? ? ? * ) . based on their near - ultraviolet excess emission @xcite , young sun - like stars accrete gas from disks at a typical rate @xmath25 ( e.g. , * ? ? accretion implies that disk gas can not everywhere be static ( `` passive '' ) . mass is transported inward by the outward transport of angular momentum , either by turbulence or by ordered flows @xcite . the transport mechanism or , as is commonly stated , the origin of disk viscosity remains obscure , with proposals ranging from turbulence driven by the magneto - rotational instability ( mri ; @xcite ) to vortices @xcite to gravitational torques @xcite . when transport is local , our ignorance is encapsulated in the parameter @xmath26 ( e.g. , * ? ? ? * ) , the ratio of the local shear stress to the total pressure . crude , disk - averaged values of @xmath27 are inferred from observations of @xmath28 vs. age @xcite . accretion may be restricted to certain radii and times , as in models that rely upon the mri not all of the disk may be sufficiently ionized to couple to magnetic fields ( e.g. , * ? ? ? * see also @xcite ) . magnetically inert regions are called `` dead zones '' @xcite . whether planetesimals form in active or passive regions of the disk is an outstanding question . this review will explore both cases . the fraction of stars with near - infrared excess emission attributable to optically thick , gaseous disks decreases from near unity at stellar ages @xmath29 myr , to @xmath30% at @xmath31 myr @xcite . whatever combination of accretion , photoevaporation , and planet formation is responsible for this observed decline , giant planets including ice giants , which contain more hydrogen than can be explained by accretion of hydrated solids @xcite must form within several myr before the gas dissipates . the same deadline characterizes planetesimal formation , since the ice giants , which by mass are 8090% rock and ice , coagulated from planetesimals . current models set the formation of solar system giant planets within a disk with more solids , and sometimes more gas , than the mmsn . @xcite identify neptune and uranus with `` isolation - mass '' oligarchs at @xmath3230 au , requiring @xmath33 . becomes `` isolated '' when it has accreted an annulus of disk material having a width about @xmath34 the radius of its hill sphere , @xmath35^{1/3}r$ ] ( e.g. , * ? ? ? the average density of the planet spread through its hill sphere is of order @xmath36 . ] @xcite prefer @xmath37 at @xmath38 , so that jupiter s core can accrete its gaseous envelope within 3 myr . at the midplane of a passive disk one heated solely by stellar radiation the gas temperature , scale height , and density are approximately @xmath39 these are adapted from @xcite , adjusted for a disk obeying ( [ eq_sigmag])([eq_sigmad ] ) , orbiting a young star of mass @xmath40 , radius @xmath41 , and temperature @xmath42 . thus in our mmsn , water ice starts to condense outside @xmath0@xmath43 au , and methane freezes outside @xmath012 au . more detailed models of passive disks are reviewed by @xcite . turbulent accretion can give higher midplane temperatures . the density and thermal structure of active disks depends on the assumed viscosity profile , @xmath44 . one model is that of @xcite , which assumes @xmath45 and is tailored to fit broadband spectra and reflected light images . for simplicity , estimates in this review employ ( [ eq_sigmag])([eq_rho ] ) for both active and passive disks . what do astronomical observations tell us about grain sizes ? in disk surface layers directly illuminated by optical light from host stars , grain sizes are 1@xmath46 , as deduced from mid - infrared silicate emission bands @xcite , and from scattered light images at similar wavelength @xcite . surface grains have settled vertically , residing at heights @xmath47 above the midplane of 1@xmath48 @xcite ; for a disk in which dust and gas are well mixed , @xmath495 . at disk midplanes , mm to cm - sized grains are routinely invoked to match mm to cm - wave spectra and images ( e.g. , * ? ? ? * ; * ? ? ? the most primitive meteorites , of nearly solar photospheric composition except in volatile elements , also offer data on particle sizes . confounding meteoriticists and astronomers alike is why up to @xmath090% of their volume is filled with chondrules : once molten , 0.1-mm to cm - sized spheres that solidified 4.57 gyr ago ( e.g. , * ? ? ? chondrule petrology is consistent with their having been heated just above liquidus for less than minutes , and having cooled for hours to days . the heating mechanism is not known ; nebular shocks are suspected , but the origin of such shocks is debated @xcite . chondrules might well have been the building blocks of the first - generation planetesimals , brought together by self - gravity . that @xmath010% of chondrules are binaries which collided and fused while still partially molten implies that when chondrules were suspended in space , they had large collective densities @xcite , possibly exceeding the roche value . extremely dusty environments are also indicated by the retention of volatiles in chondrules @xcite . gas drags particles . the degree of coupling is measured by the dimensionless stopping time @xmath50 , where @xmath51 is the keplerian angular velocity and @xmath52 measures how long it takes a particle of mass @xmath53 , radius @xmath54 , and internal density @xmath55 to have its speed @xmath56 relative to gas be reduced by order unity . the gas sound speed is @xmath57 . particles are well entrained in gas when @xmath58 . we have given the two cases for the drag force @xmath59 most relevant for planetesimal formation : epstein s @xcite law of free molecular drag , arising from the difference @xmath0@xmath60 $ ] in momentum fluxes received by the windward and leeward faces of the particle , and stokes drag for low reynolds number @xmath61 . given ( [ eq_rho ] ) , the mean free path for collisions between gas molecules is @xmath62 . , that reproduces the dynamical viscosity @xmath63 at 200 k. but constant @xmath64 assumes @xmath65 , which fails for @xmath66 k , the sutherland constant for h@xmath67 . at colder temperatures @xmath68 so that @xmath69 @xcite . for simplicity we neglect the lower values of viscosity and @xmath70 that this effect would produce in cold regions . [ foot_chapcow ] ] other cases for @xmath59 are given by @xcite and @xcite . for @xmath71 and @xmath72 , marginally coupled bodies ( @xmath73 ) have @xmath74 increasing from 35 to 120 cm in the stokes regime as @xmath6 runs from 1 to 7 au ; thereafter @xmath75 declines with @xmath6 in the epstein regime , with @xmath76 cm at 30 au . gas and dust move at different velocities because pressure gradients barely accelerate particles with @xmath77 . for the moment , let us neglect turbulence , and assume the collective particle density @xmath78 so we can ignore the backreaction of dust on gas . then the azimuthal gas velocity @xmath79 obeys @xmath80 pressure @xmath81 makes the gas rotate more slowly than the local keplerian velocity @xmath82 by @xmath83 using the thin disk approximation @xmath84 . while @xmath85 is independent of total disk mass and is nearly independent of @xmath6 , it increases in hotter disks , perhaps to @xmath86 . an individual particle seeks to orbit at the full @xmath87 . thus it normally experiences a headwind , except possibly where there are variations to the power - law descent of @xmath88 , as considered at the end of [ sec_pile ] . in a passive disk , a particle settles vertically to the midplane and , on longer timescales , drifts radially inward as the headwind saps its angular momentum . the following is drawn from ( * ? ? ? * see also @xcite ) . if we neglect backreaction , the cylindrical components of the gas velocity are @xmath89 , @xmath90 , and @xmath91 . in @xmath47 , the particle s equation of motion reads @xmath92 where the last term accounts for stellar gravity , for @xmath93 . for @xmath58 , the particle settles at terminal velocity @xmath94 . for @xmath95 , the particle behaves as a lightly damped harmonic oscillator whose amplitude decays as @xmath96 . a characteristic time for settling , valid for all @xmath97 , is @xmath98 in @xmath6 and @xmath99 , @xmath100 writing @xmath101 where @xmath102 , we approximate @xmath103 and drop @xmath104 . both approximations can be checked a posteriori . to first order in @xmath105 , @xmath106 the time for radial drift , @xmath107 , is @xmath0@xmath108 longer than @xmath109 . both times are minimized , and velocities relative to gas are maximized at @xmath0@xmath110 , for marginally coupled bodies . this is the well - known problem that boulders having sizes @xmath111 drift towards the star in @xmath112 , too quickly to form planets . as particles of a given size drift inward , they tend to `` pile up , '' increasing @xmath113 ( ys , @xcite ) ; see figure [ fig : enrich ] . their inward mass flux decreases with decreasing @xmath6 if epstein drag applies and the outer regions are not already depleted in solids . idealized pile - ups march inward on the drift timescale with increasing amplitude . the actual ability of particles to pile up coherently will be affected by ongoing evolution of the particle size distribution and by any disk turbulence . @xcite explore one prescription for turbulent viscosity predicated on the khi that amplifies pile - ups , but its basis is unproven ; see 5.2.3 of @xcite . @xcite show , by combining the drift and turbulent diffusion of particles ( see [ sec : turb ] ) with the viscous evolution of @xmath114 , that the solids - to - gas ratio in a disk evolves , and often increases , as it accretes onto the star . a special pile - up of dust could occur just inside an ice - line ( e.g. * ? ? ? * ) . when `` dirty snowballs '' drift towards and evaporate inside the ice - line , they unleash small dust grains that may accumulate there ( see * ? ? * who included this effect in a coagulation model ) . furthermore , the ice - line may act as a `` cold trap '' : vapor that diffuses radially outward and crosses the line will condense and accumulate @xcite . though water is the dominant volatile , the methane condensation front could also be significant ( [ sec_disks ] ) . we have shown that inward particle drift is a direct consequence of gas pressure decreasing radially outward . by the same physics , solids of all sizes drift into and collect within local pressure maxima @xcite , with @xmath115 particles accumulating fastest . if a pressure bump has a narrow radial width @xmath116 , even small amplitudes @xmath117 will produce local maxima . proposed sources of pressure bumps include gaseous spiral arms @xcite , anticyclonic vortices @xcite , the aforementioned ice - lines , and fluctuations in magneto - rotational turbulence ( * ? ? ? * see [ sec : turbcon ] ) , including spatial variations in the background flux @xcite or magnetic resistivity , as at the edges of dead zones @xcite . @xcite discusses pressure trapping in more detail . the response of solids to turbulent gas is multi - faceted ( see @xcite , especially their section 4.1 ) . over many eddy times , turbulence diffuses solids , vertically and radially . on shorter timescales , transient flow structures can have just the opposite effect : they can collect particles . many studies assume that turbulence in disks is kolmogorov in character , and in our example calculations below we also adopt this view . we associate @xmath118 , @xmath119 , and @xmath120 with the speed , turnover time , and length of the largest and fastest `` outer scale '' eddies . smaller eddies of length scale @xmath121 have slower speeds @xmath122 and shorter turnover times @xmath123 . the cascade of energy from larger to smaller scales terminates at the `` inner scale , '' characterized by @xmath124 , @xmath125 , and @xmath126 set by the molecular viscosity @xmath127 . for many estimates we assume that outer scale eddies have lifetimes limited by orbital shear , so that @xmath128 . by ignoring the difference between the turbulent diffusivity for mass , @xmath129 , with the turbulent diffusivity for angular momentum , @xmath130 , we extract the convenient prescriptions @xmath131 and @xmath132 ( @xcite , see also @xcite ) . however the assumption of homogeneous , isotropic turbulence characterized by cascades of energy down to smaller scales may be incorrect . simulations of mri turbulence , both global @xcite and in large shearing boxes @xcite , show that inverse cascades produce pressure perturbations that survive many tens of orbits and have in - plane sizes larger than @xmath133 . these coherent structures may accumulate particles by pressure trapping ( figure [ fig : passiveconc ] , [ sec_pile ] ) . in active regions of the disk , relative gas - particle and particle - particle velocities may be dominated by turbulence @xcite , rather than the systematic drifts considered in [ sec_lam ] . @xcite derive integral expressions for these relative velocites , evaluating them numerically for kolmogorov turbulence . consider identical grains with @xmath134 , i.e. stopping times between the inner and outer scales . collision speeds are acquired from eddies with a turnover time @xmath135 . if @xmath136 , coupling to the eddy is too weak to excite particle motion . if @xmath137 , grains are so tightly coupled to the incompressible gas flow that they do not collide . thus colliding grains have relative velocity @xmath138which is not to be confused with the random speed of an individual particle , @xmath139 . for @xmath140 , particles are loosely coupled to all eddies . during @xmath141 there are @xmath142 kicks from the strongest , outer scale eddies , so the random walk in particle velocity saturates at @xmath143 @xcite . since the random walks of different particles are uncorrelated , the random velocity is now also the typical relative speed . combining our expressions for collision speeds gives @xmath144 where the stokes number @xmath145 . this agrees roughly with @xcite . @xcite provides more precise fitting formulae further refined by @xcite that include important corrections for @xmath146 and unequal particle sizes . @xcite confirm that orbital dynamics introduces only modest corrections when @xmath147 . thus collision speeds in disks peak at @xmath118 for @xmath115 particles , since @xmath148 when @xmath149 . magneto - rotational turbulence with @xmath150 can induce collision speeds faster than drift speeds ( eqs . [ eq_rdot ] and [ eq_phidot ] ) . on timescales @xmath151 , turbulence diffuses particles spatially . @xcite compute , formally and by order - of - magnitude methods , the particle diffusivity @xmath152 as a function of the gas mass diffusivity @xmath153 . when @xmath97 is small , orbital dynamics are negligible . if further the stokes number @xmath154 , then particles are well coupled to gas and @xmath155 . if @xmath156 , it takes @xmath141 for the loosely coupled particle to have its random velocity @xmath157 ( derived in the paragraph just above eq . [ eq : vcoll ] ) changed by order unity . then @xmath158 . thus when @xmath159 , @xmath155 regardless of @xmath160 . in general , @xmath161 because orbital epicycles limit diffusion @xcite . the time for particles to diffuse radially @xmath162 is shorter than the drift time @xmath163 when @xmath164 . thus for @xmath165 , sub - cm - sized particles are well coupled to the disk accretion flow ; but see @xcite for possible height - dependent complications . by equating the vertical diffusion time @xmath166 with the gravitational settling time @xmath109 , we find a turbulent dust scale height @xmath167 for all @xmath97 . @xcite derive and numerically confirm ( [ eq_hp ] ) for @xmath95 . the @xmath58 limit is well - known @xcite , but see the end of [ sec : turbcon ] for a possible correction . on timescales @xmath168 , particles tend to be centrifugally flung out of high vorticity eddies , and funneled into their interstices of lower vorticity @xcite . since centrifugal support in eddy vortices makes them low pressure centers , turbulent concentration is consistent with the tendency of particles to seek high pressure ( [ sec_pile ] , @xcite ) . particles of given @xmath141 are concentrated preferentially by eddies that turn over on the same timescale ; in other words , particles are concentrated most strongly by eddies to which they are marginally coupled . in a kolmogorov cascade , smaller eddies are more effective than larger eddies at concentrating their respective , marginally coupled particles . this is because vorticity @xmath169 , down to the dissipation scale where @xmath170 and @xmath171 for @xmath71 , @xmath172 , and @xmath149 . turbulent concentration is special since it can collect @xmath58 particles . those particles that are optimally concentrated are marginally coupled to the smallest eddies : they have @xmath173 , which corresponds to @xmath174 at 1 au and @xmath175 at 30 au , with some uncertainty in the viscosity ( see footnote [ foot_chapcow ] ) . @xcite apply turbulent concentration to chondrules : objects having sizes 0.11 mm are optimally concentrated at @xmath176 au in their model disk , consistent with our estimate of @xmath177 above . the shape of the chondrule size distribution is well reproduced by turbulent concentration . this raises two interesting possibilities : ( 1 ) the chondrule size distribution was initially much wider , and has since been narrowed to the one observed today by turbulent concentration ; or less likely ( 2 ) chondrules happen to form with sizes that closely match optimal concentration scales at the location of the asteroid belt . the upper limit on concentration is set by mass loading , the backreaction of particles on the turbulence . in direct numerical simulations of @xmath177-sized particles in turbulent gas , @xcite find a maximum concentration factor @xmath178 at the grid , i.e. dissipation scale . this maximum value obtains for moderate reynolds numbers @xmath179 and a box - averaged @xmath180 . extrapolating these results using `` cascade multipliers '' @xcite , they show that @xmath181 still applies at higher @xmath182 . maximal clumping at the dissipation scale yields a mass @xmath183 equal to that contained in a compact 10-cm solid at @xmath176 au . however these clumps are unlikely to compactify because of their extremely short characteristic lifetimes , on the order of @xmath184 @xcite . for our disk parameters , @xmath185 minutes . * hereafter chs08 ) introduce the possibility that @xmath177-sized solids concentrate on much larger length scales @xmath186 , or @xmath187 at @xmath188 . the clump mass @xmath0@xmath189 would correspond to a compact @xmath020-km planetesimal , for @xmath190 concentration . large scale fluctuations might arise as an `` intermittent '' phenomenon : deviations from self - similarity that show up in high order structure functions @xcite . evidence exists for particle clumping at larger scales , i.e. in the inertial range of isotropic turbulence , but preferentially for particles larger than @xmath177 , with @xmath191 @xcite . this is to be expected since eddies in the inertial range have turnover times longer than @xmath184 . whether clumping of @xmath192-sized or larger particles happens at astrophysically interesting amplitudes and rates is an active area of research . chs08 find that an @xmath193-scale clump is massive enough to survive ram pressure stripping as it plows through gas . they propose that self - gravity draws solids to the clump center at the terminal velocity , a process that lasts @xmath0100 orbits . the turbulence that is invoked to create the clump could still destroy it , since the lifetime of an @xmath193-eddy at 2.5 au is @xmath194 orbit . streaming instabilities ( [ sec : si ] ) may aid in clump survival , as they show that mass loading can promote particle clumping in keplerian disks . the enticing possibility that chondrules comprise first - generation planetesimals warrants further investigation in this area . ( [ eq_hp ] ) for the turbulent dust scale height might need revision because turbulent concentration accelerates vertical settling @xcite . the correction has not been investigated , but would apply for @xmath195 . turbulent concentration is also neglected in studies of grain growth by sticking , a subject to which we now turn . particles stick upon colliding if they move slowly enough , dissipate enough energy during impact , and are small enough since surface area increases relative to mass for smaller bodies . we can estimate the relevant orders of magnitude by comparing the initial kinetic energy of two identical elastic spheres colliding at speed @xmath196 , with the surface binding energy at the moment of maximum deformation . from hertz s law of contact , the radius of the compressed cap is @xmath197 , where @xmath198 is young s modulus . the binding energy is @xmath0@xmath199 , where @xmath200 is the surface tension from unsaturated bonds , and exceeds the collisional energy for @xmath201 for ice @xcite . for @xmath202-sized spheres made of silicate , ( [ eq_stick ] ) yields @xmath203 , about an order of magnitude smaller than sticking velocities measured experimentally @xcite . in laminar disks , growth by sticking is expected to stall at @xmath0 cm sizes . from equation([eq_rdot ] ) , radial drift velocities exceed @xmath0@xmath204the sticking velocity for @xmath202-sized monomers once @xmath205 , corresponding to particle sizes @xmath206 , @xmath207 , and @xmath208 mm at 1 , 5 , and 30 au , respectively , for @xmath71 . thus super - cm particles would fail to accrete smaller grains that move with the gas . as judged by equation([eq_stick ] ) , comparably - sized , super - cm bodies would stick only at unrealistically low relative speeds , with extremely low surface binding energies per unit mass . indeed bodies @xmath209 mm in size have not been observed to stick experimentally @xcite . particle porosity abets sticking by allowing for greater dissipation of kinetic energy @xcite . moreover porosity lowers @xmath55 , which permits larger masses to coagulate when growth is limited to a fixed @xmath97 , as above . the degree of enhanced mass growth depends on the drag regime ; fixing the stopping time gives a particle mass @xmath210 for epstein drag , and @xmath211 for stokes drag . these advantages are limited , however , since collisions at speeds of @xmath212@xmath213 restructure and compactify aggregates composed of @xmath202-sized monomers @xcite . simulations of grain growth by @xcite and @xcite include vertical settling and neglect radial drifts . @xcite find that in laminar disks , porous aggregates achieve equivalent compact sizes of a few cm before settling to the midplane , while particles assumed to be always compact grow up to 1 cm ( see their figure 3 , models s2 , s5 , and s6 ) . @xcite calculate that in turbulent disks , porous bodies attain equivalent compact sizes of several cm , for @xmath214 at @xmath215 and 5 au . in both these studies , @xmath202-sized grains deplete , and the disk becomes optically thin , within @xmath0@xmath216 yr . this timescale is too short to be reconciled with astronomical observations of disk spectra . to maintain the population of small grains and the disk s optical depth over 110 myr , models may need to account for collisional fragmentation of aggregates , or condensation of dust from silicate vapor in hot , active disk regions @xcite . turbulent concentration of dust ( [ sec : turbcon ] ) could also be an important correction to coagulation models , not only because densities are enhanced , but because collision speeds are likely reduced in particle clumps , as measured in simulations of the streaming instability ( * ? ? ? * [ sec : si ] ) . we analyze the stability of a planar , self - gravitating , rotating sheet of dust , modeled as a fluid ( gw , @xcite ) . for the moment we ignore interactions with gas . unperturbed , the dust has surface density @xmath217 , angular speed @xmath218 , and barotropic pressure @xmath219 , where @xmath220 is the velocity dispersion . for axisymmetric perturbations , the linearized equations for continuity and momentum read ( e.g. , chapter 6 of * ? * ) @xmath221 where perturbations are primed , @xmath222 is the perturbation gravitational potential , and @xmath223 is the epicyclic frequency of radial oscillations in practice nearly equal to @xmath218 , its value for a kepler potential . for wkb waves , perturbations @xmath224 where @xmath225 , and @xmath226 @xcite . then ( [ eq_pert1])([eq_pert3 ] ) yield the well - known dispersion relation for axisymmetric waves : @xmath227 which informs us that pressure stabilizes short wavelengths , rotation stabilizes long ones , and self - gravity de - stabilizes intermediate wavelengths when @xmath228 @xcite . modes whose wavelengths exceed @xmath229 ( neutrally stable for @xmath230 ) are always stable . the fastest growing mode has @xmath231 . a standard expression for the vertical thickness of the dust layer is @xmath232 ( caveat emptor : ( [ eq_hp ] ) shows this is not valid when @xmath195 ) . then toomre s criterion ( [ eq_q ] ) translates into a thickness criterion @xmath233 or equivalently , a requirement on the midplane density @xmath234 which has the same scaling as the roche criterion ( [ sec_intro ] ) , but is smaller by a geometric factor of @xmath020 that arises because the gravity of an axisymmetric ring is greater than that of a tidally distorted ellipsoid . the conditions for dynamical collapse differ by orders of magnitude from those afforded by the gas disk : @xmath235 turbulence would have to be exceptionally weak to allow particles to sediment to such a thin layer : from ( [ eq_hp ] ) , @xmath26 would need to be @xmath236 . overcoming this obstacle and showing that there are alternatives to dynamical collapse from a disk of uniform solar abundance are the goals of this review , with some of the main ideas summarized in figures [ fig : enrich ] and [ fig : ladder ] . @xcite solves for the linear stability of a midplane layer composed of perfectly coupled dust and gas . the layer is confined by the pressure of overlying gas layers . as with toomre s calculation above , axisymmetry is assumed ; however the midplane layer s vertical thickness is free to change . dynamical instability occurs for midplane densities about 4 times greater than @xmath237 because gas pressure helps to stabilize the midplane layer . since the layer is incompressible , in - plane motions give rise to out - of - plane `` bulges . '' though not modeled , slow sedimentation of dust to the centers of these bulges could produce km - scale planetesimals . toomre s criterion ( [ eq_q ] ) is a good rule of thumb for deciding when self - gravity matters , but it misleads because it is derived for axisymmetric waves . in reality , gi proceeds non - axisymmetrically . non - axisymmetric waves amplify when swung from leading to trailing by the background radial shear @xcite . waves grow even if @xmath238 , but especially strongly as @xmath239 . @xcite gives a delightful tutorial , explaining swing amplification as a near - resonance between radial shear , epicyclic motion , and self - gravity . amplification is restricted to the interval , of duration @xmath0@xmath240 , when wave pitch angles slew from about -1 to 1 rad ; upon completion , @xmath241 and amplitudes have grown by factors @xmath242 for @xmath243 . modes most prone to growth have @xmath244 . for these reasons , gw describe the disk s initial fragments as having a characteristic length @xmath245 and mass @xmath246 , where @xmath247 is an order unity parameter that contains our uncertainty about the spectrum of seed perturbations and how close @xmath248 is to unity . at @xmath249 , @xmath250 equates to a compact rocky planetesimal having size @xmath251 ; this estimate figures prominently in the lore of `` kilometer - sized planetesimals . '' numerical n - body simulations of inelastically colliding particles verify that within 12 orbital periods , particles aggregate on the scale @xmath252 @xcite . the linear mass scale @xmath253 is not as prominent in either particle or gas simulations @xcite , partly because fragments accrete rapidly suggesting that @xmath253 be used cautiously , and probably as a lower limit . the stability properties discussed above pertain to dust treated as a single frictionless fluid . but gas - dust interactions can change this picture qualitatively . the simplest modification is to introduce drag terms to the momentum equations @xcite : @xmath254 the meaning of @xmath255 bears clarification . with drag , particles drift radially ( [ sec_lam ] ) . therefore perturbed quantities refer to a background with steady @xmath256 , and our wkb analysis is restricted to @xmath257 . backreaction on gas , whose properties are assumed fixed , is neglected . when @xmath58 , the drag - modified equations yield a new dispersion relation : @xmath258 which implies that modes for which @xmath259 are unstable even when @xmath238 ! the fastest growing mode has @xmath260 and growth rate @xmath261 . thus drag destabilizes long wavelength modes , on longer timescales , compared to the dissipationless case . these qualitative differences in particular the removal of toomre s @xmath248-criterion apply for a variety of drag regimes , and for cases that include turbulent diffusion of particles , accounted for by adding @xmath262 to the right - hand side of ( [ eq_pert1 ] ) ( youdin , in preparation ; karim shariff & jeffrey cuzzi , personal communication ) . * hereafter gp00 ) explain the instability simply . in the unperturbed state , dust has constant surface density and drifts radially inward at constant velocity . an axisymmetric overdense ring of width @xmath263 exerts a gravitational pull radially inward at its outer edge . provided @xmath264 , dust rotates faster there , while the gas velocity is unaltered . the increased drag causes dust to flow at a greater rate into the annulus . likewise there is a flow into the annulus at the inner edge , where dust experiences less drag . thus when combined with gas drag , self - gravity , no matter how weak , can draw particles into rings . the stability of non - axisymmetric modes with dissipation is a largely open field . @xcite make some exploratory integrations of the linearized perturbation equations . in n - body simulations that include @xmath265 drag and a fixed gas velocity field , particles cluster more readily because gas drag damps their velocity dispersion @xcite . particles may be prevented from settling to the midplane not only by turbulence intrinsic to the gas ( e.g. , sustained by the mri ) , but also by turbulence triggered by the particles themselves @xcite . we consider here instabilities caused by small @xmath58 solids that drive a vertical shear . in [ sec : di ] we show that larger particles , in addition to inducing a vertical shear , can drive turbulence of a different character via relative streaming motions . as dust settles , a particle - rich sublayer develops which orbits at nearly the full keplerian velocity . the dust - poor gas above and below shears by at speeds approaching the pressure - supported value @xmath266 ( eq . [ eq_etavk ] ) . such a stratified shear flow is subject to a kelvin - helmholtz - type instability ( khi ) . for a non - rotating flow which we later show turns out to be a fair analogue for actual vertically shearing flows in disks the khi can develop when the richardson number @xmath267 somewhere @xcite , where @xmath268 is the vertical gravitational acceleration and @xmath269 ( see * ? ? ? * for the applicability of @xmath270 to a dust - gas mixture ) . the square root of @xmath270 is the brunt - vaisala frequency for buoyant vertical oscillations , divided by the shear rate @xmath271 . this is a sensible criterion : when the buoyancy frequency is small , vertically displaced fluid elements fail to return to their equilibrium positions before being whisked away by the background shear . when we neglect self - gravity and approximate @xmath272 , and further take @xmath273 and @xmath274 , criterion ( [ eq_ri ] ) implies that the khi prevents the sublayer from having a thickness less than @xmath275 thus the midplane density falls short of @xmath237 by a factor of @xmath0@xmath276 , roughly consistent with findings by @xcite , who use mixing length models for particle - laden turbulence scaled to laboratory experiments . if we insist that the disk be of bulk solar composition , then achieving higher densities requires either that particles be larger ( @xmath277 ) so as to decouple from the turbulence , or that they clump locally ( [ sec : turbcon ] , [ sec : di ] ) . nominally we can estimate , using ( [ eq_hp ] ) , that turbulence intrinsic to gas dominates the khi when @xmath278 . whether mri - dead zones ( [ sec_disks ] ) are sufficiently passive for the khi to manifest remains an outstanding question . our @xmath26 scalings may not apply to the weak motions possibly excited by mri - active surface layers ( @xcite ; see also @xcite for detailed estimates of the active layer thickness ) . in one of the earliest works to treat the khi in disks realistically by accounting for orbital differential rotation , @xcite numerically integrate the linearized perturbation equations and find that maximum growth factors increase strongly with midplane dust - to - gas ratios so strongly that @xmath279 is likely to stall at values just above unity in disks having bulk solar metallicity . their conclusion is borne out by shearing box simulations @xcite , performed in the limit that dust is perfectly coupled to gas ( @xmath280 ) . these simulations demonstrate that despite the inclusion of orbital dynamics , richardson numbers characterizing instability remain between 0.1 and 1 ( figure [ fig : khi ] ) . in retrospect this is not surprising , since rotational and radial shearing frequencies @xmath0@xmath281 are only of the same order as both the brunt - vaisala frequency , and the vertical shearing frequency @xmath282 , when @xmath283 . nevertheless radial shear is critical for limiting the growth of perturbations that otherwise run away when only the coriolis force is included ( cf . as pointed out by j. goodman ( personal communication ; see @xcite ) , the khi may not be the only instability at play . dusty sublayers may also be baroclinically unstable : their isodensity surfaces , dominated by dust , do not align with their isobars , dictated by gas @xcite . axisymmetric baroclinic stability is assured by @xcite and by analogy with the solberg - hoiland criteria for rotating stars @xcite . specifically , displacements at constant entropy are almost perfectly radial when vertical stratification of the dust is strong . thus in a keplerian disk , outward adiabatic displacements move toward regions of higher specific angular momentum , and are rayleigh stable . aside from postulating large , aerodynamically decoupled particles , another way to prevent the shearing instability from forestalling gi is to substantially raise @xmath284 . @xcite discovers that for flows with constant @xmath285 , @xmath28630 can lead to @xmath287 in a midplane cusp . in fact , by further increasing @xmath284 slightly , and accounting properly for vertical self - gravity , the cusp becomes singular : @xmath288 . ys perform similar calculations , finding @xmath289 could be achieved for @xmath284 as low as 2 , depending on disk parameters . they show that the cusp develops when @xmath290 throughout the layer , which from ( [ eq_hpri ] ) requires @xmath291 , or @xmath292 . physically , as solids are added , the buoyancy frequency increases , while the shear saturates at @xmath293 . numerical simulations , which as yet do not include vertical self - gravity , suggest @xmath294 to attain densities approaching @xmath237 @xcite . mechanisms to achieve supersolar metallicities are shown in figure [ fig : enrich ] and described in [ sec_pile ] . whether or not they are gravitationally unstable , thin dust layers for which @xmath295 are also prone to the streaming instability ( [ sec : si ] ) , a mechanism that can further concentrate solids relative to gas when @xmath296 . the vertical shearing instability described in [ sec : thin ] arises in dense particle layers , even when dust is perfectly coupled to gas ( @xmath297 ) . when relative motion between gas and dust is allowed ( @xmath296 ) , new secular ( i.e. requiring dissipation ) instabilities arise . these drag instabilities can strongly clump particles relative to gas . as with vertical shear , drag instabilities require that @xmath279 be large enough that particles backreact significantly on gas . self - gravity is not required , in contradistinction to the drag - assisted gi of [ sec : draggi ] . we describe two different models of drag instabilities . the secular dust layer instability ( sdli ) of gp00 considers the drag arising from turbulent stresses on the stratified dust layer . the streaming instability ( si ) of ( * ? ? ? * hereafter yg05 ) accounts for the drag forces acting mutually between gas and particles in a laminar unstratified disk . both analytic models as well as numerical simulations that add extra physics to the si @xcite show that particles clump . the sdli uses a single - fluid , height - integrated , axisymmetric model of the dust - rich sublayer . dust - poor gas at the top and bottom of the sublayer shears by at relative speed @xmath110 , and is envisioned to be turbulent ( [ sec : thin ] ) . these turbulent boundary layers exert drag on the dust layer . turbulent surface stresses are assumed to be communicated throughout the entire layer ; the detailed vertical response is , by construction , ignored . to calculate the turbulent drag , gp00 adopt prescriptions from gw and @xcite for `` plate drag , '' drawing an analogy with ekman flow past a rigid plate . the sdli generates overdense rings on orbital timescales , with radial widths comparable to the sublayer vertical thickness . the use of the plate drag formula is not well founded , except perhaps if particles in the sublayer all had @xmath115 @xcite . but gp00 argue that their results are more general , transcending the specific drag prescription used to obtain them . the crucial assumption underlying the sdli is that drag is collective , i.e. it depends on the surface density of the sublayer . gp00 construct a toy model that demonstrates how any collective drag force leads to instability . the insights apply to drag instabilities generally , including the si . in the toy model , the sublayer is described by a 1d distribution of mass density @xmath217 and velocity @xmath298 : @xmath299 where @xmath300 accounts for drag , and @xmath301 serves as a proxy for gravity , pressure , and coriolis forces . the equilibrium state has uniform mass density @xmath302 and inward drift velocity @xmath303 . we might imagine @xmath304 increases with @xmath217insofar as the plate drag torque is independent of @xmath217 while the angular momentum of the sublayer is proportional to @xmath217 , so that the drift speed decreases with increasing @xmath217but the details are unimportant : instability arises as long as @xmath304 and @xmath268 depend on @xmath217 differently . with no loss of generality , all of the @xmath217-dependence is relegated to @xmath305 . we take linear perturbations , denoted by primes , to have a fourier dependence @xmath306 . the drag coefficient is expanded @xmath307 a non - zero drift @xmath308 is necessary for perturbations to @xmath304 to enter linearly . the dispersion relation reads @xmath309 where @xmath310 and @xmath311 . for any @xmath312 the positive root gives growth , since @xmath313 for all real @xmath314 . thus any collective drag produces instability . to understand the growth mechanism , we examine the eigenfunctions @xmath315 and @xmath255 , first taylor expanding the growing mode about @xmath316 , @xmath317 and similarly expanding @xmath318 . by substituting these into the linearized eqs . ( [ eq : toy1])([eq : toy2 ] ) and equating terms of the same order , we see that @xmath319 , @xmath320 , and @xmath321 . for @xmath322 , oscillations are stable with wave speed @xmath323 . this is a neutral mode where @xmath315 simply advects with the background flow . to first order in @xmath316 , @xmath324 is anti - phased with @xmath315 ; if @xmath325 , the drift speed decreases ( increases ) at density maxima ( minima ) due to the perturbed drag force ; the wave does not amplify but its speed shifts by @xmath326 . to second order in @xmath316 , the acceleration of @xmath324 is subject to drag , which induces a secondary flow @xmath327 . since @xmath327 is @xmath328 out of phase with @xmath315 , the mode grows , at rate @xmath329 . the phase shift enters because accelerations change sign at density extrema . figure [ fig : di ] depicts the eigenfunctions . the toy model shows that essential components of drag instabilities include the background drift @xmath308 , which allows collective effects to enter linearly , and time - dependent oscillations that drag forces overstabilize . since the streaming instability involves a minimal amount of physics keplerian orbital motion , gas pressure , and a drag acceleration that is linear in relative velocity it supports the robustness of drag instabilities . yg05 uncover the si by modeling gas and dust as two interacting fluids that obey @xmath330 where @xmath331 . vertical gravity is neglected , so @xmath6 is the cylindrical radius . * hereafter yj07 ) confirm that gas compressibility is justifiably neglected in ( [ eq : si4 ] ) , and that the fluid approximation for dust holds . , then the fluid approximation is expected to be valid on timescales @xmath332 . a more rigorous criterion would be useful since many nonlinear simulations ( e.g. * ? ? ? * ) use the two - fluid approach . ] the steady - state solutions to ( [ eq : si1])([eq : si4 ] ) give keplerian motion , plus radial and azimuthal drifts between dust and gas ( * ? ? ? * [ sec_lam ] ) . the relative streaming motion is linearly overstable ( yg05 ) , behaving similarly to the gp00 toy model . the main complication is that while unstable modes can be axisymmetric , they necessarily involve motions in all three directions . particle clumping occurs through the backreaction term in ( [ eq : si2 ] ) : @xmath333 . since @xmath334 and @xmath141 are constant by assumption in yg05 , and to excellent approximation for subsonic flows only perturbations to @xmath335 couple linearly to background streaming motions @xmath336 . thus particle clumping is required to extract energy from streaming motions , as in the gp00 toy model . the si is ultimately powered by the background gas pressure gradient , which does work on radially flowing gas : @xmath337 . since streaming motions transport angular momentum inward , no energy is extracted from the keplerian shear flow ( yj07 ) . the si is controlled by two parameters : @xmath97 and @xmath338 , the background dust - to - gas ratio . while the si always exists , growth is slow in the test - particle limit @xmath339 , and in the perfect coupling limit @xmath58 . as @xmath340 , background drift speeds peak , and modes can grow on orbital times . the nonlinear consequences of the si are best studied by computer simulations , which can selectively include other physics . treating particles as a continuous fluid simplifies analytic calculations , but causes difficulties in numerical simulations . the use of artificial viscosity to avoid density discontinuities in the pressureless fluid can underestimate particle concentrations and otherwise compromise results . alternatively , `` hybrid '' simulations model gas as a fluid on an eulerian grid and solids with lagrangian superparticles , each representing a swarm of actual particles . yj07 simulate the linear growth of si , and confirm that the hybrid and two - fluid approaches give convergent results . they also provide the eigenfunctions of growing fourier modes , useful for testing codes with two - way drag forces @xcite . @xcite perform nonlinear 3d simulations of the idealized si using superparticles . marginally coupled , @xmath341 solids concentrate by factors of several hundred for a range of @xmath342 . better coupled @xmath343 particles give particle overdensities of several tens , but only for @xmath344 . when @xmath345 , particle clumping is of order unity . sample simulations are displayed in figure [ fig:7sisters ] . realistically assessing how effectively the si can concentrate particles requires that vertical stratification be included ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * hereafter jym09 ) , since high dust - to - gas ratios also trigger the khi ( [ sec : thin ] ) . jym09 perform 3d hybrid simulations of @xmath343@xmath43 solids to study the combined effects of khi and si . they find a metallicity threshold for strong clumping : when @xmath284 increases from 2/3 to 4/3 , the si , unaided by self - gravity , triggers overdensities having @xmath346 . the @xmath284 threshold corresponds to @xmath347 at the midplane , similar to the analytic criterion for the saturation of the khi ( ys ) . it remains to be seen if this threshold will shift , or be manifested differently , for smaller particles . hybrid simulations with si that include self - gravity are discussed in [ sec_nonlinear ] . the fragmentation of the disk into marginally self - gravitating clumps is only the beginning of the story of planetesimal formation . about six orders of magnitude need to be traversed from the roche density to solid density . even if clumps survive buffeting by turbulence and ram pressure stripping @xcite , their further contraction is resisted by internal random motions and net angular momentum . gas drag slows the sedimentation of particles toward clump centers , but it also enables such settling by braking the clump s rotation and damping random motions . for @xmath58 , collapse proceeds on timescales longer than the free - fall time ; how much longer is unclear . at large enough densities , inelastic collisions between particles further reduce random motions and accelerate gravitational collapse . the popular notion that gi produces planetesimals having sizes on the order of a kilometer probably originates from gw , who recognize that a fragment having the full mass @xmath250 ( [ sec : criteria ] ) has too much angular momentum to collapse unimpeded to solid density . they propose instead that first - generation planetesimals form from overdense regions having horizontal sizes @xmath348 ; such clumps , if they conserve angular momentum , collapse into solid objects having sizes @xmath349 , spinning just below break - up ( see also 15.2 of * ? ? ? but this estimate is unjustified because it assumes that scales @xmath350 can collapse ; this would require @xmath351 , and indeed gw assume unrealistically that @xmath352 . while reliable analytic estimates of the sizes of first - generation planetesimals are lacking , numerical simulations have made great strides , providing a proof of principle that gravitational collapse can occur , even in turbulent flows . @xcite synthesize practically all of the physics described in this review by executing 3d , self - gravitating , vertically stratified , shearing box simulations of superparticles in mri - turbulent gas . particles all have @xmath3531 . three ways of concentrating particles manifest : trapping within transient maxima in gas pressure ( [ sec_pile ] ) ; the streaming instability ( [ sec : si ] ) ; and self - gravity . within only a few orbital periods , the largest bound cluster of superparticles attains a mass @xmath0@xmath354 that of the 450-km - radius asteroid ceres though collapse to solid densities is not explicitly followed because of finite grid resolution . surprisingly , magneto - rotational turbulence hastens gravitational collapse , through pressure trapping . the main uncertainties of these simulations are the assumption that sticking produces the largish seed particles , having sizes of 1560 cm at 5 au ; and the specific realization of mri turbulence , which depends on the unknown net magnetic flux and on microscopic dissipation parameters that are too small to simulate exactly @xcite . mri turbulence , though it appears to aid gravitational collapse , is not a necessary ingredient . figure [ fig:7sisters ] displays how bound clumps also form in an unmagnetized simulation ( jym09 ) . for a collection of grains within a protoplanetary disk to become gravitationally bound , their density @xmath335 must exceed the roche density , which for typical parameters is over 1000 times the ambient gas density @xmath334 . equivalently , the dust - to - gas ratio must be enhanced over the protosolar value by a factor of 60,000 . daunting as it may seem , the requirement may be met by several mechanisms , working in concert , that concentrate particles relative to gas . it certainly helps to have a disk or regions of the disk with an abundance of solids relative to gas greater than that given by solar composition . the dust - to - gas column density ratio @xmath113 may need to be supersolar by factors of 210 ( [ sec : thin ] ; see also @xcite ) for planetesimals to form . the sharp rise in exoplanet detection rates with host star metallicity @xcite reinforces the importance of initial dust abundance . disks can also become enriched with time , as pile - ups due to radial drifts and photoevaporation of gas increase @xmath113 ( figure [ fig : enrich ] ) . observations of gas giants are consistent with their origin in metal - enriched disks . bulk metallicities range from @xmath0@xmath355 solar ( jupiter ; [ sec_intro ] ) to @xmath0@xmath356 solar ( hd 141569 ; @xcite ) . two popular misconceptions deserve correction . first , self - gravity need not be comparable to stellar tides before its effects are felt . gas drag on particles , abetted by arbitrarily weak self - gravity , collects particles into overdense axisymmetric rings ( [ sec : criteria ] ) . second , gas turbulence does not necessarily obstruct gi . there is growing evidence that magneto - rotational turbulence creates long - lived pressure maxima that can trap particles ( [ sec_pile ] , [ sec : turb ] ) . the above concentration and enrichment mechanisms combined with vertical settling ( figure [ fig : enrich])can eventually lead to @xmath357 . this condition presents a much lower hurdle than the roche criterion , or even toomre s @xmath248-criterion . having as much inertia in particles as in gas paves the way for collective drag effects notably the streaming instability that can clump particles still further , all the way to roche density . figure [ fig : ladder ] illustrates the size ladder that dust must climb to form planets , identifying key processes . 1 . chondrules arguably the building blocks of first - generation planetesimals were flash - heated in an extremely dust - rich environment , with @xmath335 possibly exceeding @xmath358 ( [ sec_intro ] ) . that none of the concentration mechanisms discussed in this review ( see [ sec : turbcon ] ) explains how they were heated indicates our story may be incomplete . in shock models for chondrules , the isothermal post - shock gas density exceeds the pre - shock density by the square of the shock mach number , a factor of about 30just enough for @xmath359 . this opens the exciting possibility of a violent origin for both chondrules and planetesimals , though leaves unidentified the source of such strong shocks . 2 . to what extent does dust settle to the disk midplane ? the answer depends on stirring by turbulence , whether driven by the solids or externally imposed . common idealizations notably that the dust density profile in otherwise laminar disks has constant richardson number @xmath360hold only approximately at best , as shown by simulations performed in the limit that dust and gas are perfectly coupled @xcite . long - term evolution of the dust distribution requires relaxing the perfect coupling assumption , so that dust can sediment out of gas . novel numerical approaches such as implicit time evolution and multiple time stepping will need development . the wide separation between stopping , orbital , and sedimentation timescales could reveal new phenomena . there may not even be a steady state to which dust settles . in laminar disks , the dust distribution may cycle through long periods of settling interrupted by brief bursts of nonuniform turbulence wherever @xmath270 drops below unity . simulations that include the streaming instability ( e.g. , * ? ? ? * ) find that particles moderately coupled to gas having stopping times @xmath361clump strongly enough to undergo gravitational collapse . does the same conclusion apply to yet smaller solids , increasing the overlap between process 1growth by sticking and process 3 in figure [ fig : ladder ] ? pessimistically , smaller solids might not settle to a layer with @xmath362 where streaming instabilities are most effective . also clump lifetimes decrease with decreasing stopping time ( figure [ fig : si3d ] , [ sec : turbcon ] ) . optimistically , particle concentration is observed in current simulations to increase with finer grid resolution , with no convergence as yet . an adaptive mesh applied to a shearing box would be a powerful tool to track the densest particle clumps . + together points 2 and 3 emphasize that elucidating the respective roles of sticking and self - gravity requires bridging our understanding of particle dynamics from the aerodynamically well - coupled to the marginally decoupled . the post - roche evolution of particle clusters is largely unexplored . a cluster could fragment into multiple ones , much as interstellar gas clouds undergo hierarchical fragmentation as the jeans mass decreases with higher density . the number of fragments could be large if the initial overdensity is an azimuthally extended ringlet . a record of the initial mass function for planetesimals might even be preserved today , in the mass functions of asteroids and kuiper belt objects @xcite . extremely widely separated binaries @xcite may be gravitationally collapsed clusters that fissioned . we thank xylar asay - davis , joe barranco , jeff cuzzi , josh eisner , anders johansen , aaron lee , and paolo padoan for discussions . jrgen blum , anders johansen , alessandro morbidelli , gordon ogilvie , and jeff oishi provided helpful and encouraging remarks on draft versions of this manuscript . this review owes its origin to the lunchtime talk series organized through the berkeley center for integrative planetary sciences ( cips ) . this work was supported in part by national science foundation grant ast-0507805 . s , papaloizou j , lesur g , heinemann t. 2009 . . in _ astronomical society of the pacific conference series _ nv pogorelov , e audit , p colella , gp zank , vol . 406 of _ astronomical society of the pacific conference series _ 2000 . . in _ origin of the earth and moon , edited by r.m canup and k. righter and 69 collaborating authors . tucson : university of arizona press . , p.75 - 84 _ , eds . rm canup , k righter , et al . university of arizona press an . in _ star formation in the interstellar medium : in honor of david hollenbach _ , eds . d johnstone , fc adams , dnc lin , da neufeeld , ec ostriker , vol . 323 of _ astronomical society of the pacific conference series _

as a result of deep hard x - ray observations by _ chandra _ and _ xmm - newton _ a significant fraction of the cosmic x - ray background ( cxrb ) has been resolved into individual sources . these objects are almost all active galactic nuclei ( agn ) and optical followup observations find that they are mostly obscured type 2 agn , have seyfert - like x - ray luminosities ( i.e. , ergs s ) , and peak in redshift at . since this redshift is similar to the peak in the cosmic star - formation rate , this paper proposes that the obscuring material required for agn unification is regulated by star - formation within the host galaxy . we test this idea by computing cxrb synthesis models with a ratio of type 2/type 1 agn that is a function of both and 210 kev x - ray luminosity , . the evolutionary models are constrained by parameterizing the observed type 1 agn fractions from the recent work by barger et al . the parameterization which simultaneously best accounts for barger s data , the cxrb spectrum and the x - ray number counts has a local , low- type 2/type 1 ratio of 4 , and predicts a type 2 agn fraction which evolves as . models with no redshift evolution yielded much poorer fits to the barger type 1 agn fractions . this particular evolution predicts a type 2/type 1 ratio of 12 for , and thus the deep x - ray surveys are missing about half the obscured agn with these luminosities . these objects are likely to be compton thick . the parameterization also predicts a covering fraction approaching unity at and which would be very difficult to explain using a simple unevolving torus model . overall , these calculations show that the current data strongly supports a change to the agn unification scenario where the obscuration is connected with star formation in the host galaxy rather than a molecular torus alone . the evolution of the obscuration implies a close relationship between star formation and agn fueling , most likely due to minor mergers or interactions .

astro-ph0509356

in recent years , theories of the formation and evolution of active galactic nuclei ( agn ) have been dominated by a simple geometric unification model @xcite . it is envisaged that an optically thick obscuring torus lies between the broad - line and narrow - line regions at a distance of @xmath9 pc from the central black hole , and the orientation to the line - of - sight of this torus is the dominant parameter in determining the observational properties of an agn . the opening angle of the torus is such that it blocks the view of the optical / uv continuum and broad - line region for most observers , and therefore accounts for the observed dominance of narrow - line agn ( type 2s ) over broad - line agn ( type 1s ) in the local universe @xcite . the torus will also produce absorption in the x - ray band , and , indeed x - ray observations generally show that optical type 2 agn exhibit more soft x - ray absorption than type 1 objects @xcite . however , there are exceptions to this rule ( e.g. , * ? ? ? * ; * ? ? ? * ) and rapid absorption variability observed in some local agn @xcite indicate that refinements to the toroidal unification picture are required . the geometric agn unification model is also an important component in explaining the shape of the cosmic x - ray background ( cxrb ; @xcite ) at energies @xmath10 kev . between 1 and 20 kev the cxrb can be described by a power - law with a photon - index of @xmath11 @xcite , while most unobscured agn are observed to have @xmath12 ( after taking into account the effects of compton reflection ; @xcite ) . it was therefore postulated that the spectrum of the cxrb can be built up by summing the spectra of many obscured agn over a distribution in redshift , luminosity and absorbing column density @xmath13 ( @xcite ; see @xcite for a recent review ) . this picture has now been largely confirmed by very deep x - ray observations by _ chandra _ and _ xmm - newton _ that have resolved 5090% of the background below 10 kev into individual sources @xcite . interestingly , optical followup of these objects have shown that they mostly reside at redshifts @xmath14 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and have x - ray luminosities typical of seyfert galaxies ( @xmath15 erg s@xmath1 ) . this is in contrast with the predictions of the earlier synthesis models @xcite which were based on the local type 2/type 1 ratio , and thus expected a large number of type 2 quasars obscured high luminosity objects ( @xmath16 erg s@xmath1 ) at redshifts @xmath17 . indeed , x - ray luminosity functions measured by @xcite , @xcite , and @xcite show that at @xmath18 the space density of agn is dominated by moderate - luminosity seyfert galaxies . @xcite argue , however , that an unevolving unified model ( i.e. , a torus ) is still consistent with the data if the deep surveys are subject to significant selection effects ( see also @xcite ) . at the same time as the cxrb was being revealed , observations by the _ hubble space telescope _ showed a remarkable connection between the mass of the supermassive black hole in the centers of local , inactive galaxies and the mass of the galactic bulge @xcite . it now seems clear that the growth of both the central black hole and galaxy must be connected in a fundamental way , and that feedback from the accreting black hole and/or star forming - regions have a significant impact on the structure of the bulge ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? while the connection between star - formation and black hole accretion has long been suspected based on observations of local seyfert 2s ( e.g. , * ? ? ? * ; * ? ? ? * ) , further evidence has been provided by the redshift distribution of the _ chandra _ cxrb sources @xcite , which peaks at about the same redshift as the cosmic star formation rate @xcite . more recent cxrb synthesis models which allow type 2 objects to separately evolve as infrared - bright star - forming galaxies @xcite can reproduce the low-@xmath3 peak in the redshift distribution , but seem to overpredict the number of obscured agn at @xmath19 @xcite . recent observations are providing significant pressure on the basic assumption that an unevolving unified model can be applied for all agn at all redshifts . the comprehensive optical follow - up work of @xcite on x - ray deep fields provides compelling evidence that the fraction of type 1 objects increases significantly with luminosity even at redshifts as low as @xmath20@xmath21 . the large numbers of obscured seyferts at @xmath22 indicate galaxy formation and evolution is still ongoing at this era , but is evidently different than the earlier quasar epoch which seems to have a much smaller fraction of obscured agn @xcite . perhaps there are large numbers of type 2 quasars which have been missed in the deep surveys due to selection effects ( i.e. , buried quasars which release the majority of their bolometric luminosity in the infrared ; @xcite ) , a possibility that future _ spitzer _ surveys should elucidate . alternatively , the lack of type 2 quasars may be explained by invoking the receding torus model @xcite , where , because of the effects of dust sublimation , the inner radius of the torus is increased at higher luminosity . another possibility is that radiation pressure flattens the torus in high luminosity objects @xcite . both these explanations , since they assume no connection between the agn and the host galaxy , should be independent of redshift , but figure 19 in the paper by @xcite also shows evidence that the broad - line agn fraction decreases with redshift at a given luminosity . these authors find that at the redshift where the bulk of the cxrb is produced ( @xmath23 ) , the type 1 fraction at a luminosity of @xmath24 erg s@xmath1 is only @xmath2520% as compared to 7080% at @xmath26 . in addition , the low redshift ( @xmath27 ) optical type 1 and 2 luminosity functions constructed by @xcite from nearly 3000 agn in the sloan digital sky survey find a type 2/type 1 ratio of approximately unity at low seyfert luminosities , with type 1s dominating at higher luminosities ( see also @xcite ) . these results all imply that the obscuring material responsible for blocking the agn continuum in seyfert 2s evolves with redshift , as well as luminosity . moreover , if the sloan results are accurate the absorbing medium must evolve so that it increases its covering fraction by a factor of 34 from @xmath28 to @xmath7 . this evolution exactly tracks that of the star formation rate in the universe . the central idea proposed here is that the obscuration important for providing the correct shape of the cxrb is regulated by and controlled via active star - formation within the galaxy . the actual processes responsible for the obscuration are unknown , but it may be associated with star - formation on many - hundred pc scales , or be brought in closer to the nucleus by a starburst disk @xcite . thus , the type 2/type 1 ratio does evolve with redshift and peaks at @xmath7 , where both the star - formation rate is a maximum and the majority of the obscured agn discovered by the deep surveys are found . similar ideas have been discussed before @xcite , but here we use the latest observational constraints on the cxrb spectrum , source number counts and type 2/type 1 ratio ( @xmath29 ) to determine what kind of agn evolution is possible and if it is consistent with the above hypothesis . we begin by describing in the next section the calculation of the cxrb spectrum and number counts , including assumptions on the agn spectral model and evolution of @xmath30 . section [ sect : results ] presents the results of the computations and compares them against the data . we discuss the implications of the results for the unified model and models of galaxy formation / evolution in sect . [ sect : discuss ] , and then present our conclusions in sect . [ sect : concl ] . this paper assumes the standard _ wmap _ and _ boomerang _ @xmath31-dominated cosmology : @xmath32 km s@xmath1 mpc@xmath1 , @xmath33 , and @xmath34 @xcite . the calculation of the cxrb spectrum and number counts follows the standard method outlined in earlier works ( e.g. , * ? ? ? * ; * ? ? ? the spectral intensity of the cxrb ( in kev @xmath35 s@xmath1 kev@xmath1 str@xmath1 ) at energy @xmath36 is given by @xmath37 where @xmath38 is the x - ray luminosity function for agn ( in mpc@xmath39 ) , @xmath40 is the absorbed observed - frame spectrum of an agn ( in kev @xmath35 s@xmath1 kev@xmath1 ) with intrinsic luminosity @xmath4 at redshift @xmath3 , and @xmath41 is the luminosity distance to redshift @xmath3 : @xmath42 a similar expression is used to calculate the number of sources per square degree with flux ( defined in some energy band ) greater than @xmath43 , @xmath44 @xmath45 where @xmath46 is the conversion factor from str@xmath1 to degrees@xmath47 , @xmath48 is the rest - frame luminosity corresponding to observed - frame flux @xmath43 at redshift @xmath3 . to generate a cxrb model with the above equations , a number of key ingredients must be specified . for the normalization and evolution of the space density of agn , we use the hard x - ray luminosity function ( hxlf ) of @xcite . it is defined for the absorption corrected luminosity in the rest - frame 210 kev band . as it is absorption - corrected , it describes the density and evolution of all detected agn , both type 1 and 2 . @xcite found that the evolution of the luminosity function was best described using a luminosity - dependent density evolution ( ldde ) model such that the lower luminosity objects peaked in density at lower redshift than sources with higher luminosities . in contrast , @xcite found that pure luminosity evolution was the best fit to their hard x - ray luminosity function at @xmath19 . however , recent work by @xcite and @xcite have both found that ldde best describes agn evolution from @xmath28 to @xmath49 . in this paper , we assume the ldde model and parameters of @xcite . following @xcite and @xcite , we take @xmath50 , @xmath51 , @xmath52 , and @xmath53 , where from this point on the luminosity @xmath4 refers to 210 kev luminosity . consistent with using the @xcite hxlf where there is no distinction between the type 1 or type 2 objects , a single spectral model is assumed for all the agn in the calculation . @xcite have recently shown that there is little change in the x - ray spectral shape of agns over a large range of luminosity and redshift . the spectrum defined here consists of a power - law with photon - index @xmath12 and high - energy rollover at 375 kev , augmented by compton reflection . the pexrav model of @xcite is used to calculate the reflection spectrum with a reflection fraction of unity , solar abundances , and an inclination angle of 63 degrees . this model is very similar to ones assumed in other cxrb synthesis calculations ( e.g. , * ? ? ? the spectrum was calculated from 0.1 to 900 kev in 1000 logarithmically spaced steps . the following expression is used to convert from 210 kev flux , @xmath54 , to luminosity ( e.g. , * ? ? ? * ) : @xmath55 as has been standard over the last few years , type 2 objects are defined here to be those which are seen through columns @xmath56 , with type 1s being less absorbed . once the ratio of absorbed to unabsorbed objects ( @xmath30 ) is calculated for a given @xmath3 and @xmath4 ( see [ sub : ratio ] ) , the @xmath13 distribution for the obscured and unobscured agn must be defined . however , the distribution of absorbing columns for agn is unknown except for locally , where @xcite measured it for nearby luminous seyfert 2s . ten @xmath13 bins are defined in the calculation : @xmath57 . the implicit assumption is that objects which are severely compton - thick with @xmath58 will not significantly contribute to the observed cxrb . in lieu of any other information , we first make the simplest choice and assume a flat @xmath13distribution for both type 1 and type 2 sources . computations were also performed with the @xcite distribution where 75% of all type 2s have @xmath59 and half are compton thick . for @xmath60 , the solar abundance photoelectric cross - sections of @xcite are used to modify the spectrum . at higher columns , where compton scattering becomes important within the absorber , we extracted the results plotted by @xcite and divided them by their incident spectrum ( @xmath61 power - law with cutoff energy of 500 kev ) to obtain the absorption / transmission curve in our energy range . we then applied the appropriate curve to our model spectrum to account for compton - thick absorbing columns . the effects of correctly including the absorption and transmission properties of compton - thick columns are illustrated in figure [ fig : tu ] where we reproduce the results of @xcite . including the proper treatment of compton thick absorbers produced a poorer fit for this set of parameters . the central hypothesis in this work is that the type 2/type 1 ratio , @xmath30 , will increase with redshift until @xmath7 . in addition , @xmath30 will decrease with @xmath4 as seen in many recent observational campaigns . since the leading explanations for this effect are based on radiative processes associated with the central engine , we assume that there is no redshift dependence for this @xmath4-evolution . there is still , however , immense freedom in how the evolution of @xmath62 can be described , and as can be seen from previous work ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the cxrb is not very constraining on its own . we therefore make use of the results presented by @xcite in their figure 19 which shows the observed fraction of type 1 objects as a function of 28 kev luminosity in three different redshift bins : @xmath63 , @xmath64 , and @xmath65 . while the exact parameterization of @xmath62 remains unconstrained , barger s data provides landmarks to which the model can be pinned . ideally , we would be able to compute the type 1 fraction from a self - consistent model of star formation and black hole growth from redshift 1 to zero . however , such a complicated calculation is not necessary for this study . at this stage , it suffices to determine if redshift evolution of @xmath30 is needed to explain the observations , and , if so , to quantify the degree of the evolution . we are thus searching for the best description of @xmath62 that simultaneously accounts for barger s data , the cxrb spectrum , and the x - ray number count distributions ( in multiple bands ) . in practice , we do not compute @xmath62 , but the related quantities @xmath66 and @xmath67 , the fraction of type 2s and type 1s , respectively . these three parameters are related by @xmath68 and @xmath69 four different parameterizations were used to explore the evolution of the agn type 2/type 1 ratio from @xmath70 . in addition to the assumptions outlined above , these particular forms were chosen solely as good candidates for accounting for the observed data from @xcite , i.e. , to smoothly run from 01 over the observed range of @xmath4 . the parameterizations are : @xmath71 @xmath72 @xmath73 and @xmath74 aside from @xmath75 and @xmath76 the only other free parameter is the normalization @xmath77 . this is set by defining @xmath30 at @xmath78 and @xmath79 , denoted by @xmath80 . at redshifts @xmath81 the fractions @xmath82 or @xmath83 are fixed at their @xmath84 values , mimicking the roughly flat star - formation rate during this era @xcite . if the value of @xmath83 or @xmath82 ever became negative or @xmath10 then it was forced to zero or one , respectively . for every @xmath85 , we search for the best @xmath75 and @xmath76 in each of the four parameterizations by @xmath86 fitting to the barger data . equations [ eq : expmodel]-[eq : cosmodel ] are evaluated at the midpoints of the three redshift bins used by @xcite , namely , @xmath87 and @xmath88 . the values of @xmath4 are taken directly from the barger data , and we ignore the small difference between 28 kev and 210 kev luminosities ( 17% when @xmath12 ) . the errorbars on each datapoint are obtained from their plot . if only one errorbar is visible then we assume they are symmetric . the @xmath89 data point in the @xmath90 bin and the @xmath91 point in the @xmath92 bin are omitted in the @xmath86 calculation because they were an upper - limit or had very small errorbars . the @xmath86 calculation took into account the data from all three redshift bins , thus we are searching for the model that can best account for the observed redshift and luminosity evolution over the entire range in @xmath3 and @xmath4 . there are 13 data points and 2 free parameters , resulting in 11 degrees of freedom ( d.o.f . ) . the minimum @xmath86 is searched for by iterating over a large @xmath75 and @xmath76 parameter space . once the best fit @xmath75 and @xmath76 are known for each @xmath80 , the evolution equations [ eq : expmodel]-[eq : cosmodel ] are inserted into the cxrb machinery described in [ sub : form ] . predictions of the cxrb spectrum and number counts are then computed for the different evolutions and compared with the observations . as in previous work @xcite , the normalization of the predicted spectrum is multiplied by a constant in order to fit the observed data ( whose normalization is also uncertain ; @xcite ) . the best fit values of @xmath75 and @xmath76 for all four evolution models are shown in table [ table : constantalpha ] , along with the computed @xmath86 and the corresponding null - hypothesis probability @xmath93 . the power - law parametrization ( eq . [ eq : plawmodel ] ) seems to provide the best fit to the barger data for every assumed @xmath80 , with some very low values of @xmath86 obtained when @xmath94 . the cosine model ( eq . [ eq : cosmodel ] ) is the next best parameterization , with good fits obtained for every @xmath95 . the exponential model ( eq . [ eq : expmodel ] ) only finds very good fits at large values of @xmath80 , while the @xmath96 power - law model ( eq . [ eq : loglmodel ] ) results in @xmath97d.o.f.@xmath98 . in order to use these values of @xmath75 and @xmath76 to predict the cxrb properties , an @xmath13 distribution must be assumed . in [ subsub : simple ] we use a simple prescription where the columns of type ii agn are spread uniformly between @xmath99 and @xmath100 , and similarly for type i agn between @xmath101 and @xmath102 . results assuming the @xcite @xmath13distribution are presented in [ subsub : rms ] . while the normalization of the cxrb is still uncertain @xcite , the spectral slope is well constrained to be @xmath103 above 1 kev ( e.g. * ? ? ? * ) . to provide a quantitative estimate of how well our evolutionary models fit the cxrb spectrum , we computed @xmath104 between 1 and 10 kev and the results are listed in the penultimate column of table [ table : constantalpha ] . interestingly , the power - law evolutionary model , which gives the best fit to the barger data , fails badly in recovering the shape of the cxrb spectrum . this is illustrated in fig . [ fig:6to1 ] for the case of @xmath105 . we find that there are far too few absorbed sources in this model to successfully account for the shape of the observed cxrb . in fact , we find that , in general , the closer one fits the @xcite data , the poorer the fit to the cxrb spectrum . the best match to the observed cxrb was found with the @xmath96 power - law model ( eq . [ eq : loglmodel ] ) , the one that provided decent , but not spectacular , fits to barger s data ( @xmath106 ) . figure [ fig:1to1 ] plots the results from this evolutionary model assuming @xmath107 , the value pointed to by the recent sloan analysis @xcite . this model can successfully reproduce the properties of both the cxrb spectrum and the number counts , but produces a very poor fit to barger s type 1 fractions . indeed , the model , which requires very strong @xmath3 evolution to fit the cxrb , predicts that the x - ray surveys are missing many low-@xmath4 type 1 objects at @xmath108 as well as many high-@xmath4 absorbed objects at all @xmath3 , both by factors of 25 . as a consequence , type 2 quasars are predicted to be 45 times as numerous as type 1 quasars at @xmath109 , a result that does not seem to be borne out by recent observations @xcite . figure [ fig:4to1 ] represents a compromise solution . here , we plot the results of the @xmath96 power - law model with @xmath110 , similar to the classic ratio of @xcite . table [ table : constantalpha ] tells us that the fit to barger s data is adequate , and that the 110 kev slope is @xmath111 , close to the observed value . the predicted number counts seem consistent with the observations from the _ chandra _ deep fields . in this case , the prediction is that the x - ray surveys are missing @xmath112% of the absorbed high-@xmath4 objects . in figure [ fig : contour ] we plot contours of @xmath62 predicted by this evolution model between @xmath113 . the ratio of type 2 to type 1 quasars is predicted to be between 1 and 2 , more consistent with the recent observational limits . the model also predicts that low-@xmath4 agn at high @xmath3 will almost always be completely obscured , a situation that would be difficult to obtain for a standard unevolving circumnuclear agn torus . in a survey of local seyfert 2 galaxies , @xcite found that @xmath114% had @xmath59 , and about half were compton thick . we computed cxrb models using the same values of @xmath75 and @xmath76 as in the previous section , but now assuming the @xcite @xmath13 distribution for type 2 objects ( the @xmath13distribution for type 1 agn was unchanged ) . this results in 50% more compton thick objects than with the previous flat distribution . the photon - index between 1 and 10 kev for the predicted cxrb spectra for these models , denoted @xmath115 , is listed in the final column of table [ table : constantalpha ] . in general , there are only slight differences in the spectral slopes from the two @xmath13 distributions . for those cases when the barger data is fit very well and type 2 agns are not dominant then the spectral slope is slightly softer than before . this is because more absorbed agn are compton thick and thus have little effect on the spectral shape below 10 kev ( see fig . [ fig:6to1risaliti ] ) . however , if type 2 agns are the dominant population , as in cases which fit the barger data poorly , the additional compton thick sources do slightly harden the 110 kev spectral slope . while the effect on spectral shape below 10 kev may be minimal , the additional compton thick agn will significantly influence the predicted spectrum at higher energies , particularly when type 2 objects are dominant . this is illustrated in figures [ fig:1to1risaliti ] and [ fig:4to1risaliti ] where we plot the predicted cxrb spectrum and number count distributions for the @xmath96 power - law model when @xmath107 and @xmath116 , respectively . these plots are the equivalent of ones from figs . [ fig:1to1 ] and [ fig:4to1 ] , but with the @xcite @xmath13 distribution . one can clearly see that the large number of compton thick sources in the risaliti distribution provides a much better fit to the peak of the observed cxrb spectrum at 4050 kev than the simple @xmath13 model used previously . however , the additional absorption also impacts the number counts , and we find that in both the @xmath107 and @xmath116 models , the 28 kevnumber counts underpredict the observations at all fluxes . these results imply that additional compton thick sources above the simple @xmath13 model are necessary to best fit the cxrb spectrum , but that the @xcite distribution can not hold at all @xmath3 and @xmath4 . the above results show that we have managed to account for the @xcite data , the cxrb spectrum , and the x - ray number counts with a model that assumes @xmath3 evolution of @xmath30 . however , it is not clear if the data require such evolution ; indeed , successful fits of the cxrb spectrum and number counts have been obtained with no @xmath3 evolution of @xmath30 @xcite . the barger data provides an additional constraint , so to test if @xmath3 evolution is needed to fit these data , we removed the @xmath3-dependence from equations [ eq : expmodel]-[eq : cosmodel ] and performed new @xmath86 fitting for the luminosity parameter @xmath76 . the results are shown in table [ table : nozdependence ] and should be compared with the results in table [ table : constantalpha ] . in every case but one , the fit with no redshift dependence is worse , often by a large amount , than the corresponding one with @xmath3 dependence . in fact , there are only 18 models with no @xmath3-dependence that resulted in @xmath117 , as compared to 33 when redshift evolution was allowed . in addition , when adequate fits were obtained they required @xmath118 , which reduces their plausibility . our conclusion is that , to the extent that the @xcite data is an accurate depiction of @xmath30 from @xmath70 , then the data is best described by a model that includes redshift evolution of the type 2/type 1 agn ratio . moreover , if the sloan data are accurate and @xmath119 then strong redshift evolution is required to fit the cxrb . the ldde models of the x - ray agn luminosity function @xcite indicate that the most massive black holes , typically responsible for luminous quasars , form earlier than less massive black holes which power seyfert galaxies ( i.e. , anti - hierarchal formation ; @xcite ) . also , radiative and wind - driven feedback will be more important for luminous agns and will combat the tendency for increasing obscuration . thus , it may be possible that the redshift evolution of the type 2/type 1 agn ratio may also depend on luminosity . to test this , we allow the @xmath75 parameter in our evolutionary models ( eqs . [ eq : expmodel]-[eq : cosmodel ] ) to be dependent on @xmath96 via a simple linear relation : @xmath120 . the slope and intercept of this line was determined by fixing @xmath75 at @xmath121 , denoted @xmath122 , to the appropriate value in table [ table : constantalpha ] , and fit the barger data for @xmath123 , the value of @xmath75 at @xmath124 , along with @xmath76 . the results of this exercise are shown in table [ table : variablealpha ] . in all cases , the quality of the fits are equal to or better than the ones where @xmath75 was independent of @xmath4 , although the two models are statistically identical . in terms of the direction of the @xmath4-dependence , the results are inconclusive . the exponential evolutionary model prefers a positive slope ( i.e. , the covering factor increases faster with @xmath3 at higher @xmath4 ) when @xmath125 , and zero slope in all other cases . the power - law model gives a negative slope ( slower increase of the covering factor with @xmath3 at higher @xmath4 ) for @xmath126 and a positive one for @xmath127 . the cosine model seems to vary from zero slope to slightly positive , but the @xmath96 power - law model consistently produces a negative slope , often with @xmath128 , which implies _ negative _ evolution of the covering factor with @xmath3 for large @xmath4 agn . our conclusion is that the available data is not constraining enough to presently determine if there is a luminosity dependence to the redshift evolution of @xmath30 . there is tentative evidence that a negative slope is preferred , which , as mentioned above , indicates that obscuration increases more slowly with @xmath3 for more luminous agns . this can be understood if the star - formation rate in these more luminous agns does not increase very rapidly from @xmath70 . in the previous section , we found that an evolving type 2/type 1 agn ratio , @xmath62 , determined by fitting the type 1 ratios observed by @xcite can account for the cxrb spectrum and number counts . indeed , models with no @xmath3 dependence provide significantly worse fits to barger s data . here , we discuss the impact of this result on the observations , agn unification and evolution , as well as the connection to galaxy evolution . section [ sect : results ] presented moderately strong evidence for redshift evolution of the obscuring medium around an agn , but the strength of that evolution depends on the assumption on what the local type 2/type 1 ratio is at low @xmath4 . on one extreme , if we accept the sloan results that @xmath107 @xcite , then very strong @xmath3 evolution is needed ( @xmath129 ) in order to account for the cxrb spectrum ( figs . [ fig:1to1 ] and [ fig:1to1risaliti ] ) . however , this parameterization completely fails to fit the type 1 ratios observed by @xcite , requiring the x - ray surveys to miss many low - luminosity unobscured sources and high luminosity obscured agn . on the other hand , models which best fit the barger data ( e.g. , figs . [ fig:6to1 ] and [ fig:6to1risaliti ] ) can not reproduce the shape of the cxrb spectrum for any @xmath80 there are too many unobscured agn . we conclude that the best compromise solution is the @xmath110 model shown in figs . [ fig:4to1 ] and [ fig:4to1risaliti ] . this parameterization provides an adequate fit to the barger data and a good match to the observed shape of the cxrb spectrum and x - ray number counts . it predicts a redshift evolution of the type 2 agn fraction of @xmath130 . we prefer this solution over the @xmath107 sloan - inspired model because the value of @xmath80 is consistent with earlier work @xcite and it requires fewer missing obscured sources at higher @xmath3 . the implication is that the sloan work is missing local obscured sources . these could be agn whose narrow - line regions are obscured and reddened , and therefore only appear as agn in the x - ray ( i.e. , x - ray bright optically inactive galaxies or elusive agn ; @xcite ) . future observational work is required to accurate measure the local value of @xmath30 over a range in @xmath4 . this type of measurement is vital to the understanding of agn and galaxy evolution only when the local type 2/type 1 ratio is well known can the amount of evolution be determined . as described by @xcite , it is likely that the deep x - ray surveys are missing heavily obscured agn , especially ones that are compton thick . evidence for this can be seen in the hard x - ray number counts , which have yet to exhibit a turnover @xcite , and the stacking analysis of @xcite , who shows that the cxrb may only be @xmath131 % resolved at @xmath132 kev . we propose that the missing obscured agn predicted by the evolution model will mostly be compton thick , but that the type 2/type 1 ratio will only be @xmath133 at higher agn luminosities ( fig . [ fig : contour ] ) . comparing the results of the cxrb model between a simple uniform @xmath13distribution and the @xcite one , where half of the seyfert 2s are compton thick , showed that additional compton thick sources over the uniform model are needed to improve the fit to the cxrb spectrum at @xmath134 kev . however , the @xcite ratio is too large to hold in general , because the 28 kev number counts are significantly underpredicted . similar to @xmath30 , the fraction of compton - thick seyfert 2s is likely to be a function of @xmath3 and @xmath4 . thus , measurements of the @xmath13 distribution at different redshifts and luminosities are needed to accurately constrain the predictions of cxrb synthesis models . a focusing hard x - ray instrument , such as _ _ nustar _ _ , is necessary to discover and trace the evolution of the compton thick agn population . these three topics are discussed together because our results , when combined with recent observational and theoretical work , now show that they are connected and dependent processes . one can not discuss the formation and evolution of a galaxy without accounting for the growth of its black hole @xcite . similarly , we assert that the evolution of agn and their observational appearance is connected to the evolution of the host galaxy . the fact that the ratio of obscured to unobscured agn increases with redshift requires a change to the traditional unification paradigm . since the obscuring medium is changing with @xmath3 it must be influenced by forces and evolution on the extragalactic scale . the cosmic star formation rate , which peaks at a very similar redshift as @xmath30 , seems to be the most likely candidate . as star formation increases in a galaxy it signals an increase in the amount of absorbing gas and dust present that can act as an obscuring medium for any agn . the location of the absorbing material would depend on the location and size of the star forming regions and could exist close to the dust sublimation radius ( and thus mimic some properties of an absorbing torus ) , but could also be spread over a large extent in the inner part of the galaxy @xcite . the idea of an extended , more galactic - scale obscuring medium is consistent with recent infra - red ( ir ) observations which have pointed out the remarkable similarity in the mid - to - far - ir emission between seyfert 2s and 1s @xcite , in contrast to the predictions of the simple molecular torus model . of course , galactic scale obscuration of agn emission is commonly seen in the x - ray band when observing the massively - star - forming ultra - luminous infrared galaxies ( ulirgs ; e.g. , @xcite ) . if , as seems to be the case , galactic star formation and agn activity are related then the question of agn fueling arises ( see recent review by @xcite ) . at this point it is important to distinguish between high luminosity quasars and lower luminosity seyferts , which dominate the production of the cxrb . as is well known , quasars peak in density at high redshift ( @xmath135 ) , and here interactions between gas - rich disk galaxies are more common @xcite and would cause the ignition of a substantial starburst and accretion onto a black hole ( e.g. , * ? ? ? indeed , only these most luminous agns show observational evidence for interactions @xcite . at lower luminosities there is little observational evidence for seyfert galaxies being associated with mergers @xcite . likewise , there is no evidence that the decline in star formation from @xmath2 is due to a decrease in major mergers @xcite . rather , it is possible that at lower redshifts most interactions of a large disk galaxy are with smaller , gas poor satellites @xcite that may trigger an accretion phase onto the black hole , but with a much lower amount of obscuration regulated by the associated star formation . therefore , it is not valid to apply the results of major merger calculations to observations like the cxrb which are dominated by seyferts . depending on the exact nature of the interaction , a number of different obscuration geometries could be set up around the black hole , from a nearby starburst disk to an extended gas / dust lane ( cf . * ) . also , because fueling a black hole is difficult ( the angular momentum problem ; @xcite ) , a very long time may pass following the interaction before any gas makes it close enough to the center to be accreted rapidly onto the hole . this delay between the causal event and the resulting agn activity could mask the cause from observations , particularly if the interaction is weak . thus , the galaxy may not look like it was disrupted by a merger or interaction , but the resulting star formation and agn activity could still be ongoing . the power of the cxrb spectrum and the deep x - ray surveys is that they allow an investigation of the global properties of agn over cosmic time . we now know that most of the accretion in the universe is obscured , and that this obscuration evolves similar to the star - formation rate . these facts deepen the connection between star - formation and agn fueling as well as between black hole growth and galaxy evolution . to fully explore and understand these connections , the sources of the cxrb must be found and investigated in more detail over a range of redshift and luminosity . this paper explored the hypothesis that the obscuration around an agn evolves with both redshift and luminosity . the motivation behind this idea is that the obscuration is due to star - formation ongoing in the host galaxy . our conclusions are the following : * to the extent that the type 1 agn fractions observed by @xcite are accurate , redshift evolution of the type 2/type 1 agn ( @xmath30 ) to @xmath7 is required to best fit the data . * the evolutionary model that most successfully reproduced the barger data , the cxrb spectrum , and the x - ray number counts had @xmath136 , consistent with @xcite . the redshift evolution of the type 2 agn fraction is @xmath5 . * these models imply that the deep x - ray surveys are missing about 50% of obscured agn with @xmath6 at all @xmath3 . these are likely to be mostly compton thick . * additional compton - thick sources above the uniform @xmath13 distribution improve the fit to the cxrb spectrum . however , applying the @xcite distribution , where 50% of all type 2s are compton thick , over all @xmath4 and @xmath3 underpredicts the 28 kev number counts . the fraction of compton thick obscured objects should be less than 50% but will likely vary with @xmath3 and @xmath4 . * there are very tentative indications that the covering factor evolves more slowly with @xmath3 for more luminous agn . * a simple , non - evolving torus can not alone provide the agn obscuration over all cosmic time , and the shape of the cxrb spectrum is due to obscuration correlated with star - formation within the evolving host galaxy . * seyfert galaxies , which dominate the production of the cxrb , are likely fueled by minor mergers or interactions which trigger a star formation event somewhere in the nuclear region . unlike quasars , there may be a significant delay between the interaction and the subsequent ignition of the agn . antonucci , r. , 1993 , , 31 , 473 bahcall , j.n . , kirhakos , s. , saxe , d.h . & schneider , d.p . , 1997 , , 479 , 642 ballantyne , d.r . , weingartner , j.c . & murray , n. , 2003 , , 409 , 503 barcons , x. , carrera , f.j . & ceballos , m.t . , 2003 , , 339 , 757 barger , a.j . , cowie , l.l . , brandt , w.n . , capak , p. , garmire , g.p . , hornschemeier , a.e . , steffan , a.t . & wehner , e.h . , 2002 , , 124 , 1839 barger , a.j . , cowie , l.l . , capak , p. , alexander , d.m . , bauer , f.e . , fernandez , e. , brandt , w.n . , garmire , g.p . & hornschemeier , a.e . , 2003 , , 126 , 632 barger , a.j . , cowie , l.l . , mushotzky , r.f . , yang , y. , wang , w .- h . , steffen , a.t . & capak , p. , 2005 , , 129 , 578 bassani , l. , dadina , m. , maiolino , r. , salvati , m. , risaliti , g. , della ceca , r. , matt , g. & zamorani , g. , 1999 , , 121 , 473 bell , e.f . , et al . , 2005 , , 625 , 23 brandt , w.n . & hasinger , g. , 2005 , , 43 , 827 chary , r. & elbaz , d. , 2001 , , 556 , 562 cid fernandes , r. , jr . , heckman , t.m . , schmitt , h.r . , golzlez delgado , r.m . & storchi - bergmann , t. , 2001 , , 558 , 81 comastri , a. , setti , g. , zamorani , g. & hasinger , g. , 1995 , , 296 , 1 comastri , a. , et al . , 2002 , , 571 , 771 conselice , c.j . , bershady , m.a . , dickinson , m. & papovich , c. , 2003 , , 126 , 1183 de luca , a. & molendi , s. , 2004 , , 419 , 837 di matteo , t. , springel , v. & hernquist , l. , 2005 , , 533 , 604 disney , m.j . , et al . , 1995 , , 376 , 150 fabian , a.c . , barcons , x. , almaini , o. & iwasawa , k. , 1998 , , 297 , l11 fabian , a.c . , 1999 , , 308 , l39 ferrarese , l. & merritt , d. , 2000 , , 539 , l9 franceschini , a. , hasinger , g. , miyaji , t. & malquori , d. , 1999 , , 310 , l5 franceschini , a. , braito , v. & fadda , d. , 2002 , , 335 , l51 gandhi , p. & fabian , a.c . , 2003 , , 339 , 1095 gebhardt , k. , et al . , 2000 , , 539 , l13 georgantopolous , i. & georgakakis , a. , 2005 , , 358 , 131 giacconi , r. , gursky , h. , paolini , f.r . & rossi , b.b . , 1962 , , 9 , 439 gilli , r. , 2004 , advances in space research , 34 , 2470 gilli , r. , risaliti , g. & salvati , m. , 1999 , , 347 , 424 gilli , r. , salvati , m. & hasinger , g. , 2001 , , 366 , 407 grogin , et al . , 2005 , , 627 , l97 gruber , d.e . , matteson , j.l . , peterson , l.e . & jung , g.v . , 1999 , , 520 , 124 hao , l. , et al . , 2005 , , 129 , 1795 hasinger , g. , miyaji , t. & schmidt , m. , 2005 , , 441 , 417 heckman , t.m . , ptak , a. , hornschemeier , a. & kauffmann , g. , 2005 , , 634 , 161 hines , d.c . , schmidt , g.d . , smith , p.s . , cutri , r.m . & low , f.j . , 1995 , , 450 , l1 hogg , d.w . , baldry , i.k . , blanton , m.r . & eisenstein , d.j . , 2002 , preprint ( astro - ph/0210394 ) hopkins , p.f . , hernquist , l. , martini , p. , cox , t.j . , roberston , b. , di matteo , t. & springel , v. , 2005 , , 625 , l71 hutchings , j.b . , 1987 , , 320 , 122 iwasawa , k. , sanders , d.b . , evans , a.s . , trentham , n. , miniutti , g. & spoon , h.w.w . , 2005 , , 357 , 565 jogee , s. , 2006 , in physics of active galactic nuclei at all scales , ed . d. alloin ( new york : springer ) , in press ( astro - ph/0408383 ) joguet , b. , kunth , d. , melnick , j. , terlevich , r. & terlevich , e. , 2001 , , 380 , 19 kauffmann , g. & haehnelt , m. , 2000 , , 311 , 576 kirhakos , s. , bahcall , j.n . , schneider , d.p . & kristian , j. , 1999 , , 520 , 67 knigl , a. & kartje , j.f . , 1994 , , 434 , 446 kuraszkiewicz , j.k . , et al . , 2003 , , 500 , 128 kushino , a. , ishisaki , y. , morita , u. , yamasaki , n.y . , ishida , m. , ohashi , t. & ueda , y. , 2002 , , 54 , 327 la franca , f. , et al . , 2005 , , 635 , 864 laurikainen , e. & salo , h. , 1995 , , 293 , 683 lawrence , a. , 1991 , , 252 , 586 le floch , e. , et al . , 2005 , , 632 , 169 lehmer , b.d . , et al , 2005 , , 161 , 21 lumb , d.h . , warwick , r.s . , page , m. & de luca , a. , 2002 , , 389 , 93 lutz , d. , maiolino , r. , spoon , h.w.w . & moorwood , a.f.m . , 2004 , , 418 , 465 madau , p. , ghisellini , g. & fabian , a.c . , 1994 , , 267 , l17 mactavish , c.j . , et al . , 2005 , , submitted ( astro - ph/0507503 ) madau , p. , ferguson , h.c . , dickinson , m.e . , giavalisco , m. , steidel , c.c . & fruchter , a. , 1996 , , 283 , 1388 magdziarz p. & zdziarski , a.a . , 1995 , , 273 , 837 maggorian , j. et al . , 1998 , , 115 , 2285 maiolino , r. & rieke , g.h . , 1995 , , 454 , 94 maiolino , r. , et al . , 2003 , , 344 , l59 martnez - sansigre , a. , rawlings , s. , lacy , m. , fadda , d. , marleau , f.r . , simpson , c. , wilott , c.j . & jarvis , m.j . , 2005 , , 436 , 666 matt , g. , pompilio , f. & la franca , f. , 1999 , , 4 , 191 mcleod , k.k . & rieke , g.h . , 1995 , , 441 , 96 merloni , a. , 2004 , , 353 , 1035 mihos , j.c . & hernquist , l. , 1994 , , 425 , l13 morrison , r. & mccammon , d. , 1983 , , 270 , 119 murray , n. , quataert , e. & thompson , t.a . , 2005 , , 618 , 569 mushotzky , r.f . , done , c. & pounds , k.a . , 1993 , , 31 , 717 mushotzky , r.f . , cowie , l.l . , barger , a.j . & arnaud , k.a . , 2000 , , 404 , 459 panessa , f. & bassani , l. , 2002 , , 394 , 435 perola , g.c . , et al . , 2004 , , 421 , 491 pompilio , f. , la franca , f. & matt , g. , 2000 , , 353 , 440 rigby , j.r . , rieke , g.h . , prez - gonzlez , p.g . , donley , j.l . , alonso - herrero , a. , huang , j .- s . , barmby , p. & fazio , g.g . , 2005 , , 627 , 134 risaliti , g. , maiolino , r. & salvati , m. , 1999 , , 522 , 157 risaliti , g. , elvis , m. & nicastro , f. , 2002 , , 571 , 234 risaliti , g. , elvis , m. , fabbiano , g. , baldi , a. & zezas , a. , 2005 , , 623 , l93 rosati , p. , et al . , 2002 , , 566 , 667 schmitt , h.r . , 2001 , , 122 , 2243 setti , g. & woltjer , l. , 1989 , , 224 , l21 shemmer , o. , brandt , w.n . , vignali , c. , schneider , d.p . , fan , x. , richards , g.t . & strauss , m.a . , 2005 , , 630 , 729 silk , j. & rees , m.j . , 1998 , , 331 , l1 simpson , c. , 2005 , , 360 , 565 spergel , d.n . , et al . , 2003 , , 148 , 175 szokoly , g.p . , et al . , 2004 , , 155 , 271 tamura , n. , ohta , k. & ueda , y. , 2006 , , 365 , 134 thompson , t.a . , quataert , e. & murray , n. , 2005 , , 630 , 167 tozzi , p. et al . , 2001 , , 562 , 42 treister , e. & urry , c.m . , 2005 , , 630 , 115 treister , e. et al . , 2004 , , 616 , 123 tremaine , s. , et al . , 2002 , , 574 , 740 ueda , y. , akiyama , m. , ohta , k. & miyaji , t. , 2003 , , 598 , 886 waskett , t.j . , eales , s.a . , gear , w.k . , mccracken , h.j . , lilly , s & brodwin , m. , 2005 , , 363 , 801 wolf , c. , et al . , 2005 , , 630 , 771 worsley , m.a . , fabian , a.c . , bauer , f.e . , alexander , d.m . , hasinger , g. , mateos , s. , brunner , h. , brandt , w.n . & schneider , d.p . , 2005 , , 357 , 1281 wyithe , j.s.b . & loeb , a. , 2003 , , 595 , 614 zheng , w. et al . , 2004 , , 155 , 73 ccccccc 1 & 14.9 & 42.5 & 215 & @xmath137 & 1.574 & 1.578 + 2 & 12.2 & 42.5 & 70.9 & @xmath138 & 1.505 & 1.505 + 3 & 11.1 & 42.5 & 31.8 & @xmath139 & 1.463 & 1.460 + 4 & 5.4 & 43 & 18.4 & 0.072 & 1.501 & 1.504 + 5 & 5.1 & 43 & 11.4 & 0.41 & 1.487 & 1.489 + 6 & 4.9 & 43 & 9 & 0.62 & 1.474 & 1.476 + 7 & 4.8 & 43 & 8.8 & 0.64 & 1.461 & 1.462 + 8 & 5.1 & 42.9 & 9.5 & 0.58 & 1.453 & 1.452 + 9 & 5 & 42.9 & 10.5 & 0.49 & 1.446 & 1.444 + 10 & 4.9 & 42.9 & 11.7 & 0.39 & 1.440 & 1.438 + 1 & 3.5 & 0.25 & 15.3 & 0.17 & 1.477 & 1.473 + 2 & 3 & 0.3 & 11.6 & 0.39 & 1.489 & 1.488 + 3 & 2.6 & 0.33 & 9.6 & 0.57 & 1.50 & 1.501 + 4 & 2.3 & 0.36 & 8.3 & 0.69 & 1.516 & 1.519 + 5 & 2.1 & 0.38 & 7.5 & 0.76 & 1.52 & 1.525 + 6 & 1.9 & 0.39 & 7.0 & 0.8 & 1.52 & 1.525 + 7 & 1.7 & 0.39 & 6.7 & 0.82 & 1.513 & 1.519 + 8 & 1.5 & 0.4 & 6.5 & 0.84 & 1.521 & 1.528 + 9 & 1.4 & 0.41 & 6.3 & 0.85 & 1.521 & 1.528 + 10 & 1.3 & 0.42 & 6.3 & 0.85 & 1.523 & 1.530 + 1 & 0.9 & 1.3 & 28.4 & @xmath140 & 1.412 & 1.402 + 2 & 0.5 & 2.7 & 18.4 & 0.072 & 1.407 & 1.398 + 3 & 0.4 & 4.7 & 15.2 & 0.18 & 1.423 & 1.419 + 4 & 0.3 & 4.8 & 13.8 & 0.24 & 1.418 & 1.413 + 5 & 0.3 & 6.3 & 13.1 & 0.29 & 1.434 & 1.432 + 6 & 0.2 & 5.1 & 12.7 & 0.32 & 1.415 & 1.411 + 7 & 0.2 & 5.9 & 12.2 & 0.35 & 1.425 & 1.422 + 8 & 0.2 & 6.5 & 11.9 & 0.37 & 1.431 & 1.429 + 9 & 0.2 & 6.9 & 11.8 & 0.38 & 1.434 & 1.433 + 10 & 0.2 & 7.3 & 11.7 & 0.39 & 1.439 & 1.438 + 1 & 0.9 & 5.5 & 25.0 & @xmath141 & 1.467 & 1.465 + 2 & 0.5 & 4.4 & 13.8 & 0.24 & 1.474 & 1.475 + 3 & 0.3 & 4.4 & 11.2 & 0.43 & 1.466 & 1.468 + 4 & 0.2 & 4.3 & 10.1 & 0.52 & 1.465 & 1.467 + 5 & 0.2 & 3.8 & 8.9 & 0.63 & 1.480 & 1.484 + 6 & 0.1 & 4.2 & 9.1 & 0.61 & 1.461 & 1.464 + 7 & 0.1 & 3.9 & 8.3 & 0.69 & 1.472 & 1.476 + 8 & 0.1 & 3.7 & 8.1 & 0.70 & 1.481 & 1.486 + 9 & 0.1 & 3.6 & 8.2 & 0.69 & 1.484 & 1.490 + 10 & 0.1 & 3.5 & 8.5 & 0.67 & 1.489 & 1.495 + cccc 1 & 46.9 & 279 & @xmath142 + 2 & 44.8 & 96 & @xmath143 + 3 & 44.6 & 43.3 & @xmath144 + 4 & 44.6 & 24.8 & 0.016 + 5 & 44.5 & 17.8 & 0.12 + 6 & 44.5 & 15.8 & 0.20 + 7 & 44.4 & 15.8 & 0.20 + 8 & 44.4 & 16.4 & 0.18 + 9 & 44.4 & 17.5 & 0.13 + 10 & 44.4 & 19.0 & 0.088 + 1 & 0.0 & 279 & @xmath142 + 2 & 0.0 & 97 & @xmath145 + 3 & 0.0 & 46.4 & @xmath146 + 4 & 0.0 & 29.1 & @xmath147 + 5 & 0.01 & 23.0 & 0.028 + 6 & 0.08 & 19.6 & 0.076 + 7 & 0.14 & 16.7 & 0.16 + 8 & 0.18 & 14.4 & 0.27 + 9 & 0.21 & 12.6 & 0.40 + 10 & 0.24 & 11.2 & 0.51 + 1 & 0.0 & 279 & @xmath142 + 2 & 0.0 & 97 & @xmath145 + 3 & 0.0 & 46.4 & @xmath146 + 4 & 0.0 & 29.1 & @xmath147 + 5 & 0.1 & 23.0 & 0.028 + 6 & 1.0 & 20.2 & 0.063 + 7 & 1.7 & 18.4 & 0.10 + 8 & 2.2 & 17.2 & 0.14 + 9 & 2.7 & 16.2 & 0.18 + 10 & 3.0 & 15.5 & 0.21 + 1 & 39.9 & 280 & @xmath148 + 2 & 39.9 & 97 & @xmath145 + 3 & 39.9 & 46.5 & @xmath146 + 4 & 9.7 & 28.6 & @xmath149 + 5 & 6.3 & 20.1 & 0.065 + 6 & 5.3 & 15.5 & 0.22 + 7 & 4.8 & 12.7 & 0.39 + 8 & 4.5 & 11.0 & 0.53 + 9 & 4.3 & 9.9 & 0.62 + 10 & 4.1 & 9.2 & 0.69 + cccccc 1 & 14.9 & 17.9 & 42 & 120 & @xmath150 + 2 & 12.2 & 16.1 & 42 & 38.7 & @xmath151 + 3 & 11.1 & 15.2 & 42 & 19.5 & 0.052 + 4 & 5.4 & 14 & 42 & 14.5 & 0.21 + 5 & 5.1 & 5.1 & 43 & 11.4 & 0.41 + 6 & 4.9 & 4.9 & 43 & 9 & 0.62 + 7 & 4.8 & 4.8 & 43 & 8.8 & 0.64 + 8 & 5.1 & 5.1 & 42.9 & 9.5 & 0.58 + 9 & 5 & 5 & 42.9 & 10.5 & 0.49 + 10 & 4.9 & 4.9 & 42.9 & 11.7 & 0.39 + 1 & 3.5 & 1.0 & 0.0 & 10.6 & 0.48 + 2 & 3 & 1.0 & 0.09 & 9.0 & 0.63 + 3 & 2.6 & 1.2 & 0.18 & 8.1 & 0.70 + 4 & 2.3 & 1.3 & 0.24 & 7.6 & 0.75 + 5 & 2.1 & 1.5 & 0.31 & 7.1 & 0.79 + 6 & 1.9 & 1.5 & 0.34 & 6.8 & 0.81 + 7 & 1.7 & 1.5 & 0.37 & 6.6 & 0.83 + 8 & 1.5 & 1.6 & 0.41 & 6.5 & 0.84 + 9 & 1.4 & 1.5 & 0.42 & 6.3 & 0.85 + 10 & 1.3 & 1.6 & 0.45 & 6.2 & 0.86 + 1 & 0.9 & 0.8 & 0.0 & 2.5 & @xmath152 + 2 & 0.5 & 0.3 & 0.2 & 17.1 & 0.11 + 3 & 0.4 & 0.0 & 0.0 & 13.5 & 0.26 + 4 & 0.3 & 0.0 & 1.1 & 12.9 & 0.30 + 5 & 0.3 & 0.0 & 2.6 & 11.9 & 0.37 + 6 & 0.2 & 0.0 & 2.7 & 12.4 & 0.34 + 7 & 0.2 & 0.0 & 3.4 & 11.8 & 0.38 + 8 & 0.2 & 0.0 & 4.0 & 11.4 & 0.41 + 9 & 0.2 & 0.0 & 4.5 & 11.2 & 0.43 + 10 & 0.2 & 0.0 & 4.8 & 11.1 & 0.44 + 1 & 0.9 & 0.9 & 5.5 & 25.0 & @xmath141 + 2 & 0.5 & 0.5 & 4.4 & 13.8 & 0.24 + 3 & 0.3 & 0.4 & 3.9 & 10.5 & 0.49 + 4 & 0.2 & 0.4 & 3.4 & 9.1 & 0.61 + 5 & 0.2 & 0.2 & 3.8 & 8.9 & 0.63 + 6 & 0.1 & 0.3 & 3.3 & 8.0 & 0.71 + 7 & 0.1 & 0.2 & 3.5 & 8.0 & 0.71 + 8 & 0.1 & 0.1 & 3.7 & 8.1 & 0.70 + 9 & 0.1 & 0.1 & 3.6 & 8.2 & 0.69 + 10 & 0.1 & 0.0 & 3.9 & 8.3 & 0.69 +

we report the discovery of hat - p-26b , a transiting extrasolar planet orbiting the moderately bright v=11.744 k1 dwarf star gsc 0320 - 01027 , with a periodd , transit epoch ( bjd ) , and transit durationd . the host star has a mass of , radius of , effective temperature k , and metallicity}}= { \ifthenelse{\equal{{i}}{i}}{{\ensuremath{-0.04\pm0.08}}}{{\ensuremath{-0.04\pm0.08}}}}$ ] . the planetary companion has a mass of , and radius of yielding a mean density of . hat - p-26b is the fourth neptune - mass transiting planet discovered to date . it has a mass that is comparable to those of neptune and uranus , and slightly smaller than those of the other transiting super - neptunes , but a radius that is% larger than those of neptune and uranus , and also larger than those of the other transiting super - neptunes . hat - p-26b is consistent with theoretical models of an irradiated neptune - mass planet with a 10 heavy element core that comprises of its mass with the remainder contained in a significant hydrogen - helium envelope , though the exact composition is uncertain as there are significant differences between various theoretical models at the neptune - mass regime . the equatorial declination of the star makes it easily accessible to both northern and southern ground - based facilities for follow - up observations .

1010.1008

transiting exoplanets ( teps ) are tremendously useful objects for studying the properties of planets outside of the solar system because their photometric transits , combined with precise measurements of the radial velocity variations of their host stars , enable determinations of their masses and radii . of the @xmath17 confirmed teps discovered to date , all but five are saturn or jupiter - size gas giant planets with masses above @xmath18@xmath9 . the five teps below this limit , including the super - earths corot-7b ( @xmath19@xmath9 , @xmath20@xmath11 ; * ? ? ? * ; * ? ? ? * ) , and gj 1214b ( @xmath21@xmath9 , @xmath22@xmath11 ; * ? ? ? * ) , and the super - neptunes gj 436b ( @xmath23@xmath9 , @xmath24@xmath11 ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , hat - p-11b ( @xmath25@xmath9 , @xmath26@xmath11 ; * ? ? ? * ) , and kepler-4b ( @xmath27@xmath9 , @xmath28@xmath11 ; * ? ? ? * ; * ? ? ? * ) are likely composed primarily of elements heavier than hydrogren and helium , and are therefore assumed to be qualitatively different from the more massive gas giants . in addition to these five teps , the candidate tep kepler-9d has an estimated radius of @xmath29@xmath30 @xcite , and is most likely a low - mass planet @xcite , but currently lacks a mass determination . while the gas giant planets exhibit a wide range of radii at fixed mass ( and hence a great diversity in their physical structure at fixed mass ) , the low mass teps , together with the six solar system planets smaller than saturn , appear to follow a nearly monotonic relation between mass and radius . the two super - neptunes with precise radius measurements ( gj 436b and hat - p-11b ) have radii that are similar to one another ( to within 15% ) as well as to uranus ( @xmath31@xmath9 , @xmath32@xmath11 ) and neptune ( @xmath33@xmath9 , @xmath34@xmath11 ) . while the mass and radius of kepler-4b given in the discovery paper @xcite are similar to those of gj 436b and hat - p-11b , a reanalysis by @xcite finds that the radius may be @xmath35% larger , though with a @xmath36 uncertainty , it may still be similar to the other super - neptunes . the lack of significant scatter in the radii among uranus , neptune , and the super - neptunes is perhaps surprising given the vast range of radii permitted by theoretical structure models for planets in this mass range . for example , the theoretical models by @xcite predict that a @xmath37 nonirradiated neptune - mass planet may have a radius that ranges from 0.14@xmath11 ( pure iron composition ) to 0.29@xmath11 ( pure water ice composition ) or 0.86@xmath11 ( pure gas composition ) . these same models also predict that the radii of gas - dominated neptune - mass planets should be far more sensitive to stellar irradiation than those of jupiter - mass planets . for example , a @xmath38 pure hydrogen - helium neptune - mass planet at @xmath39au has a radius of @xmath40@xmath11 compared to @xmath41@xmath11 for a similarly irradiated jupiter - mass planet . in this paper we present the discovery of hat - p-26b , a tep orbiting the relatively bright star gsc 0320 - 01027 with a mass similar to that of neptune , but with a radius of @xmath42@xmath11 or 65% larger than that of neptune . this is the 26th tep discovered by the hungarian - made automated telescope network ( hatnet ; * ? ? ? * ) survey . in operation since 2003 , hatnet has now covered approximately 14% of the sky , searching for teps around bright stars ( @xmath43 ) . hatnet operates six wide - field instruments : four at the fred lawrence whipple observatory ( flwo ) in arizona , and two on the roof of the hangar servicing the smithsonian astrophysical observatory s submillimeter array , in hawaii . the layout of the paper is as follows . in we report the detection of the photometric signal and the follow - up spectroscopic and photometric observations of hat - p-26 . in we describe the analysis of the data , beginning with the determination of the stellar parameters , continuing with a discussion of the methods used to rule out nonplanetary , false positive scenarios which could mimic the photometric and spectroscopic observations , and finishing with a description of our global modeling of the photometry and radial velocities . our findings are discussed in . the transits of hat - p-26b were detected with the hat-5 and hat-6 telescopes in arizona , and with the hat-8 and hat-9 telescopes in hawaii . two regions around gsc 0320 - 01027 , corresponding to fields internally labeled as 376 and 377 , were both observed on a nightly basis between 2009 jan and 2009 aug , whenever weather conditions permitted . for field 376 we gathered 11,500 exposures of 5 minutes at a 5.5 minute cadence . approximately 1500 of these exposures were rejected by our photometric pipeline because they yielded poor photometry for a significant number of stars . each image contained approximately 17,000 stars down to sloan @xmath44 . for the brightest stars in the field , we achieved a per - image photometric precision of 4mmag . for field 377 we gathered 5200 exposures with the same exposure time and cadence ; we rejected aproximately 700 exposures . each image contained approximately 19,000 stars down to sloan @xmath44 . we achieved a similar photometric precision for the brightest stars in this field . the calibration of the hatnet frames was carried out using standard photometric procedures . the calibrated images were then subjected to star detection and astrometry , as described in @xcite . aperture photometry was performed on each image at the stellar centroids derived from the two micron all sky survey ( 2mass ; * ? ? ? * ) catalog and the individual astrometric solutions . the resulting light curves were decorrelated ( cleaned of trends ) using the external parameter decorrelation ( epd ; see * ? ? ? * ) technique in `` constant '' mode and the trend filtering algorithm ( tfa ; see * ? ? ? the light curves were searched for periodic box - shaped signals using the box least - squares ( bls ; see * ? ? ? * ) method . we detected a significant signal in the light curve of gsc 0320 - 01027 ( also known as 2mass 14123753 + 0403359 ; @xmath45 , @xmath46 ; j2000 ; v=11.744 @xcite ) , with an apparent depth of @xmath47mmag ( @xmath48mmag when using tfa in signal - reconstruction mode ) , and a period of @xmath49days ( see figure [ fig : hatnet ] ) . the drop in brightness had a first - to - last - contact duration , relative to the total period , of @xmath50 , corresponding to a total duration of @xmath51 hr . we removed the transits from the combined 376/377 light curve and searched for additional transiting objects using the bls method ; no significant signals were found in the data . we also searched the transit - cleaned light curve for periodic variations ( e.g. due to stellar rotation ) using the discrete fourier transform ( e.g. * ? ? ? * ) , and found no coherent variation with an amplitude greater than @xmath52mmag . as is routine in the hatnet project , all candidates are subjected to careful scrutiny before investing valuable time on large telescopes . this includes spectroscopic observations at relatively modest facilities to establish whether the transit - like feature in the light curve of a candidate might be due to astrophysical phenomena other than a planet transiting a star . many of these false positives are associated with large radial - velocity variations in the star ( tens of @xmath53 ) that are easily recognized . to carry out this reconnaissance spectroscopy , we made use of the tillinghast reflector echelle spectrograph ( tres ; * ? ? ? * ) on the 1.5 m tillinghast reflector at flwo . this instrument provides high - resolution spectra which , with even modest signal - to - noise ( s / n ) ratios , are suitable for deriving rvs with moderate precision ( @xmath54@xmath53 ) for slowly rotating stars . we also use these spectra to estimate effective temperatures , surface gravities , and projected rotational velocities of the target star . using the medium fiber on tres , we obtained two spectra of hat - p-26 on the nights of 2009 dec 26 and 2009 dec 27 . the spectra have a resolution of @xmath55 and a wavelength coverage of @xmath143900 - 8900 . the spectra were extracted and analyzed according to the procedure outlined by @xcite and @xcite . the individual velocity measurements of @xmath56@xmath53 and @xmath57@xmath53 were consistent with no detectable rv variation within the measurement precision . both spectra were single - lined , i.e. , there is no evidence for additional stars in the system . the atmospheric parameters we infer from these observations are the following : effective temperature @xmath58 , surface gravity @xmath59 ( log cgs ) , and projected rotational velocity @xmath60 . the effective temperature corresponds to an early k dwarf . the mean heliocentric rv of hat - p-26 is @xmath61@xmath53 . given the significant transit detection by hatnet , and the encouraging tres results that rule out obvious false positives , we proceeded with the follow - up of this candidate by obtaining high - resolution , high - s / n spectra to characterize the rv variations , and to refine the determination of the stellar parameters . for this we used the hires instrument @xcite on the keck i telescope located on mauna kea , hawaii , between 2009 dec and 2010 june . the width of the spectrometer slit was @xmath62 , resulting in a resolving power of @xmath63 , with a wavelength coverage of @xmath1438008000 . we obtained 12 exposures through an iodine gas absorption cell , which was used to superimpose a dense forest of @xmath64 lines on the stellar spectrum and establish an accurate wavelength fiducial ( see * ? ? ? an additional exposure was taken without the iodine cell , for use as a template in the reductions . relative rvs in the solar system barycentric frame were derived as described by @xcite , incorporating full modeling of the spatial and temporal variations of the instrumental profile . the rv measurements and their uncertainties are listed in table [ tab : rvs ] . the period - folded data , along with our best fit described below in , are displayed in figure [ fig : rvbis ] . [ ht ] [ ht ] in the same figure we show also the @xmath65 index , which is a measure of the chromospheric activity of the star derived from the flux in the cores of the h and k lines ( figure [ fig : cahk ] shows representative keck spectra including the h and k lines for hat - p-26 ) . this index was computed and calibrated to the scale of @xcite following the procedure described by @xcite . we find a median value of @xmath66 with a standard deviation of @xmath67 . assuming @xmath68 based on the effective temperature measured in , this corresponds to @xmath69 @xcite . from @xcite the lower tenth percentile @xmath65 value among california planet search ( cps ) targets with @xmath68 is @xmath70 . the measured @xmath65 value is only slightly higher than this , implying that hat - p-26 is a chromospherically quiet star . we do not detect any significant variation of the @xmath65 index correlated with orbital phase ; such a correlation might have indicated that the rv variations could be due to stellar activity , casting doubt on the planetary nature of the candidate . @xmath71 & @xmath72 & @xmath73 & @xmath74 & @xmath75 & @xmath76 & @xmath77 + @xmath78 & & & @xmath79 & @xmath80 & @xmath81 & @xmath82 + @xmath83 & @xmath84 & @xmath85 & @xmath86 & @xmath87 & @xmath88 & @xmath89 + @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 & @xmath95 & @xmath96 + @xmath97 & @xmath98 & @xmath99 & @xmath100 & @xmath101 & @xmath102 & @xmath103 + @xmath104 & @xmath105 & @xmath106 & @xmath107 & @xmath108 & @xmath109 & @xmath110 + @xmath111 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & @xmath76 & @xmath116 + @xmath117 & @xmath118 & @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath123 + @xmath124 & @xmath125 & @xmath126 & @xmath127 & @xmath128 & @xmath129 & @xmath130 + @xmath131 & @xmath132 & @xmath133 & @xmath134 & @xmath135 & @xmath129 & @xmath136 + @xmath137 & @xmath138 & @xmath139 & @xmath140 & @xmath141 & @xmath142 & @xmath143 + @xmath144 & @xmath145 & @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 + @xmath151 & @xmath152 & @xmath139 & @xmath153 & @xmath80 & @xmath154 & @xmath155 + [ -1.5ex ] in order to permit a more accurate modeling of the light curve , we conducted additional photometric observations with the keplercam ccd camera on the telescope . we observed five transit events of hat - p-26 on the nights of 2010 jan 05 , mar 31 , apr 04 , may 08 and may 25 ( figure [ fig : lc ] ) . these observations are summarized in table [ tab : phfusummary ] . keplercam/ & 2010 jan 05 & 191 & 40 & sloan @xmath156 band + keplercam/ & 2010 mar 31 & 161 & 59 & sloan @xmath156 band + keplercam/ & 2010 apr 04 & 291 & 64 & sloan @xmath156 band + keplercam/ & 2010 may 08 & 596 & 44 & sloan @xmath156 band + keplercam/ & 2010 may 25 & 298 & 59 & sloan @xmath156 band + [ -1.5ex ] the reduction of these images , including basic calibration , astrometry , and aperture photometry , was performed as described by @xcite . we performed epd and tfa to remove trends simultaneously with the light curve modeling ( for more details , see , and @xcite ) . the final time series are shown in the top portion of figure [ fig : lc ] , along with our best - fit transit light curve model described below ; the individual measurements are reported in table [ tab : phfu ] . the combined phase - folded follow - up light curve is displayed in figure [ fig : phfu ] . lrrrr @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 + @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath161 + @xmath166 & @xmath167 & @xmath164 & @xmath168 & @xmath161 + @xmath169 & @xmath170 & @xmath164 & @xmath171 & @xmath161 + @xmath172 & @xmath173 & @xmath159 & @xmath174 & @xmath161 + @xmath175 & @xmath176 & @xmath164 & @xmath177 & @xmath161 + @xmath178 & @xmath179 & @xmath164 & @xmath180 & @xmath161 + @xmath181 & @xmath182 & @xmath159 & @xmath183 & @xmath161 + @xmath184 & @xmath185 & @xmath186 & @xmath187 & @xmath161 + @xmath188 & @xmath189 & @xmath164 & @xmath190 & @xmath161 + [ -1.5ex ] fundamental parameters of the host star hat - p-26 such as the mass ( @xmath191 ) and radius ( @xmath192 ) , which are needed to infer the planetary properties , depend strongly on other stellar quantities that can be derived spectroscopically . for this we have relied on our template spectrum obtained with the keck / hires instrument , and the analysis package known as spectroscopy made easy ( sme ; * ? ? ? * ) , along with the atomic line database of @xcite . sme yielded the following values and uncertainties : effective temperature @xmath193k , stellar surface gravity @xmath194(cgs ) , metallicity @xmath7}}={\ensuremath{-0.04\pm0.08}}$]dex , and projected rotational velocity @xmath195 , in which the uncertainties for @xmath196 and @xmath7}}$ ] have been increased by a factor of two over their formal values to include our estimates of the systematic uncertainties . in principle the effective temperature and metallicity , along with the surface gravity taken as a luminosity indicator , could be used as constraints to infer the stellar mass and radius by comparison with stellar evolution models . however , the effect of @xmath197 on the spectral line shapes is quite subtle , and as a result it is typically difficult to determine accurately , so that it is a rather poor luminosity indicator in practice . for planetary transits a stronger constraint is often provided by @xmath198 the mean stellar density , which is closely related to the @xmath199 normalized semimajor axis . the quantity @xmath198 can be derived directly from the combination of the transit light curves @xcite and the rv data ( required for eccentric cases , see ) . this , in turn , allows us to improve on the determination of the spectroscopic parameters by supplying an indirect constraint on the weakly determined spectroscopic value of @xmath197 that removes degeneracies . we take this approach here , as described below . the validity of our assumption , namely that the adequate physical model describing our data is a planetary transit ( as opposed to a blend ) , is shown later in . our values of @xmath200 , @xmath197 , and @xmath201 $ ] were used to determine auxiliary quantities needed in the global modeling of the follow - up photometry and radial velocities ( specifically , the limb - darkening coefficients ) . this modeling , the details of which are described in , uses a monte carlo approach to deliver the numerical probability distribution of @xmath198 and other fitted variables . for further details we refer the reader to @xcite . when combining @xmath198 ( a luminosity proxy ) with assumed gaussian distributions for @xmath200 and @xmath201 $ ] based on the sme determinations , a comparison with stellar evolution models allows the probability distributions of other stellar properties to be inferred , including @xmath197 . here we use the stellar evolution calculations from the yonsei - yale ( yy ) series by . for the case of hat - p-26b , the eccentricity is poorly constrained by the rv data due to the low semiamplitude of the signal . this in turn leads to a significant uncertainty on @xmath198 . however , not all combinations of @xmath201 $ ] , @xmath200 and @xmath198 are realized by physical stellar models . in particular , if we conservatively adopt a maximum stellar age of @xmath202gyr , corresponding approximately to the age of the universe ( * ? ? * find @xmath203gyr ) , and a minimum age of @xmath204myr , corresponding roughly to the zero - age main - sequence , stars with @xmath205 and @xmath7}}= { \ensuremath{-0.04}}$ ] are not found to have densities in the range @xmath206 or surface gravities in the range @xmath207 ( here @xmath208 corresponds to an evolved star with @xmath209@xmath4 , while @xmath210 corresponds to a main sequence star with @xmath211@xmath4 ) . figure [ fig : iso ] shows the inferred location of the star in a diagram of @xmath198 versus @xmath200 , analoguous to the classical h - r diagram , for three cases : fixing the eccentricity of the orbit to zero , allowing the eccentricity to vary , and allowing the eccentricity to vary , but only accepting parameter combinations which match to a position in the yy isochrones with @xmath212 . the stellar properties and their approximate 1@xmath213 and 2@xmath213 confidence boundaries are displayed against the backdrop of isochrones for the measured metallicity of @xmath201 $ ] = @xmath214 , and a range of ages . for the zero eccentricity case the comparison against the model isochrones was carried out for each of 30,000 monte carlo trial sets ( see ) . we find good overlap between the trials and the model isochrones in 71% of the trials , the @xmath201 $ ] , @xmath200 and @xmath198 parameter combination matched to a physical location in the diagram that has an age that is within the aforementioned range . however , when the eccentricity is allowed to vary , the model for the light curves and rv data tends toward low values of @xmath215 which may only be fit by pre - main sequence stellar models , or stellar models older than the age of the universe . in this case only 15% out of 100,000 monte carlo trial sets match to physical locations in the diagram with ages within the allowed range . by requiring the star to have an age between 0.1gyr and 13.8gyr , we effectively impose a tighter constraint on the orbital eccentricity than is possible from the rv data alone ( we find an eccentricity of @xmath216 , as compared with @xmath217 when not requiring a match to the stellar models ; see also ) . adopting the parameters which result from allowing the eccentricity to vary while requiring the star to have an age between 0.1gyr and 13.8gyr yields a stellar surface gravity of @xmath218 , which is very close to the value from our sme analysis . the values for the atmospheric parameters of the star are collected in table [ tab : stellar ] , together with the adopted values for the macroturbulent and microturbulent velocities . the stellar evolution modeling provides color indices that may be compared against the measured values as a sanity check . the best available measurements are the near - infrared magnitudes from the 2mass catalogue @xcite , @xmath219 , @xmath220 and @xmath221 ; which we have converted to the photometric system of the models ( eso ) using the transformations by @xcite . the resulting measured color index is @xmath222 . this is within 1@xmath213 of the predicted value from the isochrones of @xmath223 . the distance to the object may be computed from the absolute @xmath224 magnitude from the models ( @xmath225 ) and the 2mass @xmath226 magnitude , which has the advantage of being less affected by extinction than optical magnitudes . the result is @xmath227pc , where the uncertainty excludes possible systematics in the model isochrones that are difficult to quantify . an additional check on our stellar model can be performed by comparing the isochrone - based age estimate to activity - based age estimates . using the activity - rotation and activity - age relations from ( * equations 5 and 3 ) we convert the value of @xmath228 determined in into a rossby number ( @xmath229 , where @xmath230 is the rotation period and @xmath231 is the convective turnover time - scale ) and an age . we find @xmath232 , and @xmath233 or @xmath234gyr , where we adopt the estimated uncertainties on @xmath235 and @xmath236 from @xcite . the @xmath235 value may be converted to a rotation period using the relation for @xmath231 given by @xcite . we find @xmath237d . the rotation period and color may also be used to obtain a separate age estimate from the gyrochronology relation given by ( * ? ? ? * equation 12 ) . this gives @xmath238gyr , where the scatter on this relation is less than the scatter on the age - activity relation because it includes a correction for stellar color . the age inferred from @xmath228 is consistent with the isochrone - based age of @xmath239gyr . the equatorial rotation period inferred from the spectroscopic determination of @xmath240 assuming @xmath241 is @xmath242d , which is shorter than the expected value based on @xmath228 , though the upper limit is poorly constrained . as discussed below in we measure a rv jitter of @xmath243@xmath244 for hat - p-26 . from @xcite the lower tenth percentile jitter among hires / keck observations for cps stars with @xmath245 and @xmath246 is @xmath247@xmath244 , implying that hat - p-26 has an exceptionally low jitter value only a handful of stars in this color range have measured jitter values less than that of hat - p-26 . we note that the jitter may be higher ( @xmath248@xmath244 ) if the orbit is circular , though this value is still quite low . lrl @xmath196 ( k ) & & sme + @xmath7}}$ ] & & sme + @xmath249 ( @xmath53 ) & & sme + @xmath250 ( @xmath53 ) & & sme + @xmath251 ( @xmath53 ) & & sme + @xmath252 ( @xmath53 ) & @xmath253 & tres + @xmath254 ( mag ) & 11.744 & tass + @xmath255 ( mag ) & @xmath256 & tass + @xmath257 ( mag ) & @xmath258 & 2mass + @xmath259 ( mag ) & @xmath260 & 2mass + @xmath226 ( mag ) & @xmath261 & 2mass + @xmath262 ( @xmath263 ) & @xmath264 & yy+@xmath199+sme + @xmath265 ( @xmath266 ) & @xmath267 & yy+@xmath199+sme + @xmath268 ( cgs ) & @xmath269 & yy+@xmath199+sme + @xmath270 ( @xmath271 ) & @xmath272 & yy+@xmath199+sme + @xmath273 ( mag ) & @xmath274 & yy+@xmath199+sme + @xmath275 ( mag , eso ) & @xmath276 & yy+@xmath199+sme + age ( gyr ) & @xmath277 & yy+@xmath199+sme + distance ( pc ) & @xmath278 & yy+@xmath199+sme + [ -1.5ex ] our initial spectroscopic analyses discussed in and rule out the most obvious astrophysical false positive scenarios . however , more subtle phenomena such as blends ( contamination by an unresolved eclipsing binary , whether in the background or associated with the target ) can still mimic both the photometric and spectroscopic signatures we see . in the following sections we consider and rule out the possibility that such scenarios may have caused the observed photometric and spectroscopic features . following @xcite , we explored the possibility that the measured radial velocities are not real , but are instead caused by distortions in the spectral line profiles due to contamination from a nearby unresolved eclipsing binary . a bisector analysis based on the keck spectra was done as described in 5 of @xcite . while the bisector spans show no evidence for variation in phase with the orbital period , the scatter on these values ( @xmath279@xmath244 rms ) is large relative to the rv semi - amplitude ( @xmath280@xmath244 ) , and thus the lack of variation does not provide a strong constraint on possible blend scenarios . we note that some of the scatter in the bisector spans may be due to contamination from the sky background ( predominately moonlight)correcting the bisector spans for sky contamination as discussed by @xcite reduces the rms to @xmath281@xmath244 , however the precision is still insufficient to rule out blend scenarios . following our earlier work @xcite we make use of the high proper motion of hat - p-26 to rule out the possibility that the observed transits and rv variation may be due to a background eclipsing binary that is aligned , by chance , with the foreground k1 dwarf hat - p-26 . to reproduce the observed @xmath282 deep transit , the background object can not be more than 5.6 mag fainter than hat - p-26 ( objects fainter than this would contribute less than 0.6% of the total combined light and so could not cause the transit even if they were to be completely eclipsed by an object that emits no light ) . because hat - p-26 has a high proper motion ( @xmath283@xmath284 ; * ? ? ? * ) it is possible to use the palomar observatory sky survey plates from 1950 ( poss - i , red and blue plates ) to view the sky at the current position of hat - p-26 . between 1950 and the follow - up observations in 2010 , hat - p-26 has moved @xmath285 . figure [ fig : ppm ] shows an image stamp from the poss - i plate compared with a recent observation from the . we can rule out a background object down to @xmath286mag within @xmath287 of the current position of hat - p-26 . any background object must be @xmath288 mag fainter than hat - p-26 and thus could not be responsible for the observed transit . following @xcite , @xcite , and @xcite we attempt to model the observations as a hierarchical triple system . we consider 4 possibilities : 1 . one star orbited by a planet , 2 . three stars , 2 fainter stars are eclipsing , 3 . two stars , 1 planet , planet orbits the fainter star , 4 . two stars , 1 planet , planet orbits the brighter star . here case 1 is the fiducial model to which we compare the various blend models . we model the observed follow - up and hatnet light curves ( including only points that are within one transit duration of the primary transit or secondary eclipse assuming zero eccentricity ) together with the 2mass and tass photometry . in each case we fix the mass of the brightest star to @xmath289@xmath4 ; this ensures that we reproduce the effective temperature , metallicity , and surface gravity determined from the sme analysis when using the padova isochrones ( see below ) . we have also attempted to perform the fits described below allowing the mass of the brightest star to vary . we find that in this case the mass of the brightest star is still constrained to be close to @xmath289@xmath4 by the broad - band photometry , even if the spectroscopic parameters are not included . we therefore conclude that fixing the mass of the brightest star is justified . in all cases we vary the distance to the system and two parameters allowing for dilution in the two hatnet light curves , and we include simultaneous epd and tfa in fitting the light curves ( see ) . in each case we draw the stellar radii and magnitudes from the 13.0gyr padova isochrone @xcite , extended below 0.15@xmath4 with the @xcite isochrones . we use these rather than the yy isochrones for this analysis because of the need to allow for stars with @xmath290@xmath4 , which is the lower limit available for the yy models . we use the jktebop program @xcite which is based on the eclipsing binary orbit program ( ebop ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) to generate the model light curves . we optimize the free parameters using the downhill simplex algorithm together with classical linear least squares for the epd and tfa parameters . we rescale the errors for each light curve such that @xmath291 per degree of freedom is 1.0 for the out of transit portion of the light curve . note that this is done prior to applying the epd / tfa corrections , as a result @xmath291 per degree of freedom is less than 1.0 for each of the best - fit models discussed below . if the rescaling is not performed , the difference in @xmath291 between the best - fit models is even more significant than what is given below , and the blend - models may be rejected with higher confidence . _ case 1 : 1 star , 1 planet _ : in addition to the parameters mentioned above , in this case we vary the radius of the planet and the impact parameter of the transit . the best - fit model has @xmath292 for 3567 degrees of freedom . the parameters that we obtain for the planet are comparable to those obtained from the global modelling described in . _ case 2 : 3 stars _ : for case 2 we vary the masses of the eclipsing components , and the impact parameter of the eclipse . we find no model of three stars which reproduces the observations . the transit depth and duration can not be fit when the three stars are constrained to fall on the same isochrone , and the brightest star has @xmath293@xmath4 . the best - fit case 2 model consists of equal masses for the brightest two stars , and @xmath294@xmath4 ( the lowest stellar mass in the @xcite isochrones ) for the transiting star . such a model is inconsistent with our spectroscopic observations ( it would have been easily identified as a double - lined binary at one of the quadrature phases ) , and as we will show , can be rejected from the light curves alone . the best - fit case 2 model yields @xmath295 for 3566 degrees of freedom and produces model transits that are too deep compared to the observed transit . the case 1 model achieves a lower @xmath291 with fewer parameters than the case 2 model , so the case 1 model is preferred over the case 2 model . assuming that the errors are uncorrelated and follow a gaussian distribution , the case 2 model can be rejected in favor of the case 1 model at the @xmath296 confidence level . alternatively , one might suppose that any apparent correlations in the residuals of the best - fit case 2 model are not due to errors in the model but instead are due to uncorrected systematic errors in the measurements ; large systematic errors in the measurements would increase the probability that case 1 might give a better fit to the data , by chance , than case 2 . to establish the statistical signficance with which we may reject case 2 while allowing for possible systematic errors in the measurements , we conduct 1000 monte carlo simulations in which we assume the best - fit case 2 model scenario is correct , shuffle the residuals from this fit in a manner that preserves the correlations ( this is done by taking the fourier transform of the residuals , randomly changing phases of the transform while preserving the amplitudes , and transforming back to the time domain ) , and fit both the case 2 and case 1 models to the simulated data . the median value of @xmath297 is @xmath298 with a standard deviation of @xmath299 . none of the 1000 trials have @xmath300 , the measured value . based on this analysis we reject case 2 , i.e. the hierarchical triple star system scenario , at the @xmath301 level . _ case 3 : 2 stars , planet orbits the fainter star _ : in this scenario hat - p-26b is a transiting planet , but it would have a radius that is larger than what we infer ( it may be a saturn- or jupiter - size planet rather than a neptune - size planet ) . for this case we vary the mass of the faint planet - hosting star , the radius of the planet , and the impact parameter of the transit . we assume the mass of the planet is negligible relative to the mass of its faint host star . figure [ fig : chi2blend ] shows @xmath291 as a function of the mass of the planet - hosting star for this scenario . the best - fit case 3 solution has @xmath302 for 3566 degrees of freedom , and corresponds to a system where the two stars are of equal mass and the planet has a radius of @xmath303@xmath11 . as the mass of the planet host is decreased the value of @xmath291 increases . repeating the procedure outlined above to establish the statistical significance at which we may reject case 3 we find that the @xmath304 limit on @xmath305 is @xmath306 , which results in a @xmath304 lower - limit of @xmath307@xmath4 on the mass of the planet hosting star . we may thereby place a @xmath304 lower limit of @xmath308 on the @xmath254-band luminosity ratio of the two stars . a second set of lines with a luminosity ratio of @xmath309 would have easily been detectable in both the keck and tres spectra unless the stars had very similar @xmath310 velocities ( the spectral lines are quite narrow with @xmath195 ) . the poor fit for this blend model relative to the fiducial model together with the tight constraints on the relative @xmath310 velocities and luminosity ratios of the stars in the blend models that may yet fit the data leads us to reject this blend scenario in favor of the simpler model of a single star hosting a transiting planet . _ case 4 : 2 stars , planet orbits the brighter star _ : as in the previous case , in this scenario hat - p-26b is a transiting planet , but the dilution from the blending star means that the true radius is larger than what we infer in case 1 . in this case we vary the radius of the planet , the mass of the faint star , and the impact parameter of the transit . again we assume the mass of the planet is negligible relative to the mass of its bright host star . figure [ fig : chi2blend ] shows @xmath291 as a function of the mass of the faint star . the smallest value of @xmath291 is achieved when the faint star contributes negligible light to the system , which is effectively equivalent to the fiducial scenario represented by case 1 . two effects cause @xmath291 to increase with stellar mass . first , the shape of the transit is subtly changed in a manner that gives a poorer fit to the observations . second , when the mass of the faint star is less than that of the transit host the model broad - band photometry for the blended system is redder than for the single star scenario , and is inconsistent with the observed photometry . this gives rise to the peak in @xmath311 at @xmath312@xmath4 . the case 4 model where the faint star has @xmath313@xmath4 can be rejected as in case 3 . for lower masses , we place a @xmath304 upper limit of 0.55@xmath4 on the mass of the faint star , yielding a @xmath304 upper limit on the luminosity ratio of @xmath18 . we conclude that at most the planet radius @xmath314 may be 8% larger than what we find in if there is an undetected faint secondary star in the system . here we summarize the procedure that we followed to model the hatnet photometry , the follow - up photometry , and the radial velocities simultaneously . this procedure is described in greater detail in @xcite . the follow - up light curves were modeled using analytic formulae based on @xcite , with quadratic limb darkening coefficients for the sloan @xmath156 band interpolated from the tables by @xcite for the spectroscopic parameters of the star as determined from the sme analysis ( ) . we modeled the hatnet data using an approximation to the @xcite formulae as described in @xcite . the rvs were fitted with an eccentric keplerian model . our physical model consisted of 8 main parameters , including : the time of the first transit center observed with hatnet ( taken to be event @xmath315 ) , @xmath316 , and that of the last transit center observed with the telescope , @xmath317 , the normalized planetary radius @xmath318 , the square of the impact parameter @xmath319 , the reciprocal of the half duration of the transit @xmath320 as given in @xcite , the rv semiamplitude @xmath224 , and the lagrangian elements @xmath321 and @xmath322 , where @xmath323 is the longitude of periastron . five additional parameters were included that have to do with the instrumental configuration . these are the hatnet blend factors @xmath324 , and @xmath325 , which account for possible dilution of the transit in the hatnet light curves from background stars due to the broad psf ( @xmath326 fwhm ) , the out - of - transit magnitudes for each hatnet field , @xmath327 and @xmath328 , and the relative zero - point @xmath329 of the keck rvs . the physical model was extended with an instrumental model for the follow - up light curves that describes brightness variations caused by systematic errors in the measurements . we adopted a `` local '' epd- and `` global '' tfa - model @xcite , using 20 template stars for the tfa procedure and six epd parameters for each follow - up light curve . in summary , the total number of fitted parameters was 13 ( physical model with 5 configuration - related parameters ) + 30 ( local epd ) + 20 ( global tfa ) = 63 , i.e. , much smaller than the number of data points ( 1450 , counting only rv measurements and follow - up photometry measurements ) . as described in @xcite , we use a combination of the downhill simplex method ( amoeba ; see * ? ? ? * ) , the classical linear least squares algorithm , and the markov chain monte - carlo method ( mcmc , see * ? ? ? * ) to obtain a best - fit model together with _ a posteriori _ distributions for the fitted parameters . these distributions were then used to obtain _ a posteriori _ distributions for other quantities of interest , such as @xmath215 . as described in , @xmath215 was used together with stellar evolution models to obtain _ a posteriori _ distributions for stellar parameters , such as @xmath262 and @xmath265 , which are needed to determine @xmath330 and @xmath331 . the resulting parameters pertaining to the light curves and velocity curves , together with derived physical parameters of the planet , are listed under the `` adopted value '' column heading of table [ tab : planetparam ] . included in this table is the rv `` jitter '' . this is a component of assumed astrophysical noise intrinsic to the star , possibly with a contribution from instrumental errors as well , that we added in quadrature to the internal errors for the rvs in order to achieve @xmath332 from the rv data for the global fit . auxiliary parameters not listed in the table are : @xmath333 ( bjd ) , @xmath334 ( bjd ) , the blending factors @xmath335 , and @xmath336 for the hatnet field 376 and 377 light curves , respectively , and @xmath337@xmath244 . the latter quantity represents an arbitrary offset for the keck rvs , but does not correspond to the true center - of - mass velocity of the system , which was listed earlier as @xmath252 in table [ tab : stellar ] . we find a mass for the planet of @xmath338 and a radius of @xmath339 , leading to a mean density @xmath340@xmath13 . we also find that the eccentricity of the orbit may be different from zero : @xmath216 , @xmath341 . however , as we show in , this is at best significant at only the 88% confidence level . we also carried out the analysis described above with the eccentricity fixed to zero . the resulting parameters are given in table [ tab : planetparam ] under the column heading `` @xmath342 '' . the results are discussed further in . finally , we conducted an independent model of the system based on @xcite . the primary differences between this model and the adopted model are differences in the choice of parameters to vary in the fit : we use @xmath343 as defined in @xcite rather than @xmath344 , @xmath345 rather than @xmath346 , and @xmath347 rather than @xmath348 . we also allowed for a linear drift in the radial velocities @xmath349 , and a time shift @xmath350 in the radial velocities due to possible additional bodies in the system on trojan orbits with hat - p-26b . we chose to include both @xmath351 and @xmath349 rather than fixing them to zero as the value of the bayseian information criterion ( bic ; e.g. * ? ? ? * ) was lower for the best - fit model including these parameters , than for models where one or both of these parameters were fixed to zero . the resulting parameters are given in table [ tab : planetparam ] under the column heading `` @xmath352 . '' the parameter values from this model are consistent with those from the adopted model , which gives confidence that our results are robust to changes in the choice of fitting parameters . @xmath353 ( days ) & @xmath354 & @xmath355 & @xmath356 + @xmath357 ( @xmath358 ) & @xmath359 & @xmath360 & @xmath361 + @xmath362 ( days ) & @xmath363 & @xmath364 & @xmath365 + @xmath366 ( days ) & @xmath367 & @xmath368 & @xmath369 + @xmath370 & @xmath371 & @xmath372 & @xmath373 + @xmath320 & @xmath374 & @xmath375 & @xmath376 + @xmath377 & @xmath378 & @xmath379 & @xmath380 + @xmath319 & @xmath381 & @xmath382 & @xmath383 + @xmath384 & @xmath385 & @xmath386 & @xmath387 + @xmath156 ( deg ) & @xmath388 & @xmath389 & @xmath390 + @xmath391 ( linear term ) & @xmath392 & @xmath392 & @xmath393 + @xmath394 ( quadratic term ) & @xmath395 & @xmath395 & @xmath396 + @xmath224 ( @xmath244 ) & @xmath397 & @xmath398 & @xmath399 + @xmath400 & @xmath401 & @xmath402 & @xmath403 + @xmath404 & @xmath405 & @xmath402 & @xmath406 + @xmath407 & @xmath408 & @xmath402 & @xmath409 + @xmath323 ( deg ) & @xmath410 & @xmath411 & @xmath412 + @xmath349 ( ms@xmath413d@xmath413 ) & @xmath414 & @xmath414 & @xmath415 + @xmath351 ( d ) & @xmath414 & @xmath414 & @xmath416 + rv jitter ( @xmath244 ) & @xmath417 & @xmath248 & @xmath418 + rv fit rms ( @xmath244 ) & @xmath248 & @xmath419 & @xmath418 + @xmath420 ( bjd ) & @xmath421 & @xmath422 & @xmath423 + @xmath424 & @xmath425 & @xmath364 & @xmath426 + @xmath427 & @xmath428 & @xmath368 & @xmath429 + @xmath330 ( @xmath430 ) & @xmath431 & @xmath432 & @xmath433 + @xmath331 ( @xmath434 ) & @xmath435 & @xmath436 & @xmath437 + @xmath438 & @xmath439 & @xmath440 & @xmath441 + @xmath442 ( @xmath13 ) & @xmath443 & @xmath444 & @xmath445 + @xmath446 ( cgs ) & @xmath447 & @xmath448 & @xmath449 + @xmath450 ( au ) & @xmath451 & @xmath452 & @xmath453 + @xmath454 ( k ) & @xmath455 & @xmath456 & @xmath457 + @xmath458 & @xmath459 & @xmath459 & @xmath460 + @xmath461 ( @xmath462@xmath463 ) & @xmath464 & @xmath465 & @xmath466 + @xmath467 ( @xmath462@xmath463 ) & @xmath468 & @xmath465 & @xmath469 + @xmath470 ( @xmath462@xmath463 ) & @xmath471 & @xmath465 & @xmath472 + [ -1.5ex ] we have presented the discovery of hat - p-26b , a transiting neptune - mass planet . below we discuss the physical properties of this planet , and compare them to the properties of similar planets ; we comment on the possibility that the planet has undergone significant evaporation , on the significance of its orbital eccentricity , and on the possible presence of additional bodies in the system ; and we discuss the prospects for detailed follow - up studies . figure [ fig : exomr ] compares hat - p-26b to the other known teps on a mass - radius diagram . with a density of @xmath12@xmath13 , hat - p-26b is significantly less dense than the four other neptune - size planets with well measured masses and radii ( uranus , neptune , gj 436b , hat - p-11b ) . for kepler-4b , @xcite find a large uncertainty on the radius which results from significant uncertainties on the eccentricity and the transit impact parameter . kepler-4b may be comparable in size to gj 436b and hat - p-11b , or it could be even less dense than hat - p-26b . from the theoretical models of @xcite , hat - p-26b has a radius that is well above the maximum radius of @xmath473@xmath11 for a @xmath474@xmath9 planet lacking a hydrogen - helium envelope ( i.e. a planet with a 100% water - ice composition ) . the best - fit mass and radius for hat - p-26b falls just below the 4gyr model with a 10@xmath15 rocky core and 8@xmath15 gas envelope , implying that a 4gyr model with a slightly higher core mass would provide a better match to the mass and radius . we note that the isochrone - based age ( @xmath277gyr ) and the activity - based age ( @xmath475gyr ) for the hat - p-26 system are somewhat older than 4gyr , so the inferred core - mass would therefore be somewhat smaller . we also compare hat - p-26b to the theoretical models of @xcite which predict more significant inflation due to irradiation for low mass planets than do the @xcite models . in this case the radius of hat - p-26b is intermediate between the @xmath476 and @xmath477 heavy - element enrichment models . observations of the transiting hot jupiters hd 209458b and hd 189733b in the lyman-@xmath478 line have indicated that both planets are evaporating at a rate of up to @xmath479gs@xmath413 ( e.g , * ? ? ? * ; * ? ? ? prompted by the observations for hd 209458b , several theoretical studies have indicated that atmospheric evaporation is likely to be important for close - in planets , particularly those with low surface gravities , such as hot neptunes ( see for example @xcite and the review by @xcite ) . it has even been suggested that some close - in neptune - mass and smaller planets may be the evaporated cores of planets which initially had masses comparable to saturn or jupiter ( e.g. * ? ? ? * ) . in the case of energy - limited escape , the evaporative mass - loss is given by ( see @xcite , and @xcite ; see also @xcite and @xcite for applications to corot-7b ) : @xmath480 where @xmath481 is the incident flux of extreme ultraviolet ( xuv ) stellar radiation , @xmath482 is the heating efficiency and is estimated to be @xmath483 for the case of hd 209458b @xcite , and @xmath484 is a factor that accounts for an enhancement of the evaporation rate in the presence of tides , and is given by : @xmath485 where @xmath486 is the ratio of the roche radius to the planet radius . @xcite find that for solar type stars the xuv flux at 1 au integrated over the wavelength range @xmath487 to @xmath488 is given by : @xmath489 where @xmath490 is the age in gyr . to our knowledge , a similar study has not been completed for k dwarfs , however long term x - ray observations of the 5 - 6gyr @xmath478 cen ab system reveal that on average the k1 dwarf star @xmath478 cen b has an x - ray luminosity in the 6 - 60 band that is approximately twice that of the sun , while the g2 dwarf @xmath478 cen a has a luminosity that is approximately half that of the sun @xcite . for simplicity we therefore assume that the total xuv luminosity of hat - p-26 is comparable to that of the sun ( 4.64ergss@xmath413@xmath491 at 1 au ; * ? ? ? * ) , which is likely correct to within an order of magnitude . assuming @xmath492 , we estimate that the expected present - day mass - loss rate for hat - p-26b is @xmath493 . to determine the total mass lost by hat - p-26b over its lifetime , we integrate equation [ eqn : mdotevap ] assuming an age of @xmath494gyr , @xmath495 for @xmath496gyr and @xmath497 for @xmath498gyr , neglecting tidal evolution of the orbit , and assuming that the radius is constant . we find that hat - p-26b may have lost a significant fraction its mass ( @xmath14 30% ) ; the exact value depends strongly on several poorly constrained parameters including @xmath481 and its dependence on age for a k1 dwarf , @xmath482 , and the age of the system . using the relation given by @xcite , the expected tidal circularization time - scale for hat - p-26b is @xmath499gyr which is much less than the age of the system . this time - scale is estimated assuming a large tidal quality factor of @xmath500 , and that there are no additional bodies in the system exciting the eccentricity . however , because at least two of the three hot neptunes have significant eccentricities ( gj 436b has @xmath501 , @xcite ; and hat - p-11b has @xmath502 , @xcite ; the eccentricity for kepler-4b is poorly constrained , @xcite ) , we can not conclude that the eccentricity must be zero on physical grounds , and therefore do not adopt a zero - eccentricity model for the parameter determination . as discussed in the eccentricity of hat - p-26b is poorly constrained by the rv observations , and is instead constrained by requiring that the star be younger than the age of the universe ( without the age constraint we get @xmath503 , whereas including the age constraint gives @xmath216 ) . to establish the significance of the eccentricity measurement , we also fit a model with the eccentricity fixed to zero . an f - test ( e.g. * ? ? ? * ) allows us to reject the null hypothesis of zero eccentricity with only 79% confidence . alternatively , the @xcite test for the significance of an eccentricity measurement gives a false alarm probability of @xmath504% for detecting @xmath505 with an error of @xmath506 , or 88% confidence that the orbit is eccentric . if the eccentricity is fixed to zero , the required jitter to achieve @xmath507 is 2.4@xmath244 , which is closer to the typical keck / hires jitter of other chromospherically quiet early k dwarfs than the jitter of @xmath243@xmath244 that is obtained with an eccentric orbit fit . we therefore are not able to claim a significant eccentricity for hat - p-26b , and instead may only place a @xmath508 confidence upper limit of @xmath509 . for the @xmath352 model discussed at the end of , the @xmath508 confidence upper limit is @xmath510 . further rv observations , or a photometric detection of the occultation of the planet by its host star , are needed to determine if the eccentricity is nonzero . the @xmath352 model discussed in with parameters given in table [ tab : planetparam ] includes a linear drift in the radial velocities , @xmath349 , as a free parameter . we find @xmath511ms@xmath413d@xmath413 . conducting an odds ratio test ( see * ? ? ? * ) , we conclude that the drift is real with 96.6% confidence , making this a @xmath512 detection . while the detection is not significant enough for us to be highly confident that there is at least one additional body in the system , this suggestive result implies that hat - p-26 warrants long - term rv monitoring . we also searched for a linear time - shift in the rvs due to potential trojans . we do not detect a significant shift , and may exclude @xmath513d with 95% confidence . this translates to an upper limit of @xmath514@xmath15 on the mass of a trojan companion , which is greater than the mass of the planet . with the present data we are thus not able to place a meaningful limit on the presence of trojan companions . hat - p-26 has a number of features that make it an attractive target for potential follow - up studies . at @xmath515 , it is bright enough that precision spectroscopic and photometric observations are feasible with moderate integration times . the equatorial declination of @xmath46 also means that hat - p-26 is accessible to both northern and southern ground - based facilities . the exceptionally low jitter will facilitate further rv observations , which might be used to confirm and refine the eccentricity determination , to measure the rossiter - mclaughlin effect ( r - m ; discussed in more detail below ) and to search for additional planets in the system . a detection of the occultation of hat - p-26b by hat - p-26 with irac/ _ spitzer _ would provide a strong constraint on @xmath516 , while the duration of the occultation would provide a constraint on @xmath322 . we note that the median value of the _ a posteriori _ distribution for the time of occultation that results from our global fit when the eccentricity is allowed to vary ( ) is @xmath517h after the expected time of occultation assuming a circular orbit . the expected depth of the occultation event is a challenging @xmath518 and @xmath519 at 3.6 and 4.5 respectively . scaling from tres-4 , a somewhat fainter star at these wavelengths , for which @xcite measured occultations at 3.6 and 4.5 using irac/_spitzer _ with precisions of @xmath520 and @xmath521 respectively , one may hope to achieve a @xmath522 and @xmath523 detection for hat - p-26b for one event at each bandpass . one would need to observe 5 and 4 occultations respectively to achieve a @xmath304 detection . recent measurements of the r - m effect for teps have revealed a substantial population of planets on orbits that are significantly misaligned with the spin axes of their host stars ( e.g. * ? ? ? @xcite note that misalignment appears to be more prevalent for planets orbiting stars with @xmath524k , and suggest that most close - in planets migrate by planet - planet or planet - star scattering mechanisms , or by the kozai effect , rather than disk migration , and that tidal dissipation in the convective surfaces of cooler stars realigns the stellar spin axis to the orbital axis of the close - in massive planet . @xcite also finds evidence that planets orbiting stars with @xmath525@xmath4 are more likely to be misaligned than planets orbiting cooler stars using a method that is independent of the r - m measurements . one prediction of the @xcite hypothesis is that lower mass planets orbiting cool stars should show a greater degree of misalignment than higher mass planets due to their reduced tidal influence . the detection of misalignment for hat - p-11b @xcite is consistent with this hypothesis . measuring the r - m effect for hat - p-26b would provide an additional test . using equation ( 40 ) from @xcite , the expected maximum amplitude of the r - m effect for hat - p-26b is @xmath526@xmath244 , which given the low jitter of hat - p-26 , should be detectable at @xmath527 . by measuring the primary transit depth as a function of wavelength it is possible to obtain a transmission spectrum of an exoplanet s atmosphere . such observations have been made for a handful of planets ( e.g. @xcite ; see also the review by @xcite ) . following @xcite , the expected difference in transit depth between two wavelengths is given approximately by : @xmath528 where @xmath529 is the scale height of the atmosphere , @xmath530 is the planet surface gravity , @xmath531 is the mean molecular weight of the atmosphere , and @xmath532 where @xmath533 and @xmath534 are the opacities per gram of material at wavelengths in a strong atomic or molecular line and in the nearby continuum respectively . assuming a pure h@xmath535 atmosphere , @xmath536 kg , we find for hat - p-26b @xmath537 km , and @xmath538 . if instead we assume that the atmosphere has the same composition as neptune ( e.g. * ? ? ? * ) , we have @xmath539 kg , @xmath540 km , and @xmath541 . for comparison , assuming a pure h@xmath535 atmosphere , the planet hd 209458b has @xmath542 , while gj 436b has @xmath543 , hat - p-11b has @xmath544 and kepler-4b has @xmath545 . due to its low surface gravity , hat - p-26b easily has the highest expected transmission spectrum signal among the known transiting neptune - mass planets . while it is relatively faint compared to the well studied planets hd 209458b and hd 189733b , we note that @xcite used the gran telescopio canarias ( gtc ) to detect a @xmath546 absorption level at 7582 due to potassium in the atmosphere of xo-2b , which orbits a @xmath547 early k star . scaling from this observation , it should be possible to detect components in the atmosphere of hat - p-26b with @xmath548 at the @xmath549 level using the gtc . in summary , hat - p-26b is a low - density neptune - mass planet . its low - density relative to the other known neptune - mass planets means that hat - p-26b likely has a more significant hydrogen - helium gas envelope than its counterparts . the existence of hat - p-26b provides empirical evidence that , like hot jupiters , hot neptunes also exhibit a wide range of densities . comparing to the @xcite models , we find that hat - p-26b is likely composed of a gas envelope and a heavy - element core that are approximately equal in mass , while the @xcite models prefer a higher heavy - element fraction . it is also likely that irradiation - driven mass - loss has played a significant role in the evolution of hat - p-26b we find that the planet may have lost @xmath550 of its present - day mass over the course of its history , though this conclusion depends strongly on a number of very poorly constrained parameters , particularly the xuv flux of hat - p-26 and its evolution in time . we place a 95% confidence upper limit on the eccentricity of @xmath509 . if further observations detect a nonzero eccentricity , it would mean that at least three of the four known neptune - mass teps have nonzero eccentricities , which may imply that the tidal quality factor is higher than expected for these planets . observations of the planetary occultation event for hat - p-26b with irac/_spitzer _ would greatly constrain the eccentricity , however the low expected depth is likely to make this a challenging observation . we find suggestive evidence for a linear drift in the rvs which is significant at the @xmath512 level . if confirmed , this would imply the existence of at least one additional body in the hat - p-26 system . with an expected r - m amplitude of @xmath551@xmath244 and a low stellar rv jitter , hat - p-26b is a good target to measure the r - m effect and thereby test the hypothesis that low - mass planets are more likely to be misaligned than high - mass planets . the low surface gravity also makes hat - p-26b a good target for transmission spectroscopy . hatnet operations have been funded by nasa grants nng04gn74 g , nnx08af23 g and sao ir&d grants . work of g..b . and j. johnson were supported by the postdoctoral fellowship of the nsf astronomy and astrophysics program ( ast-0702843 and ast-0702821 , respectively ) . gt acknowledges partial support from nasa grant nnx09af59 g . we acknowledge partial support also from the kepler mission under nasa cooperative agreement ncc2 - 1390 ( d.w.l . thanks the hungarian scientific research foundation ( otka ) for support through grant k-81373 . this research has made use of keck telescope time granted through nasa ( n018hr and n167hr ) .

we present the development of the collimated bipolar jets from the symbiotic prototype zand that appeared and disappeared during its 2006 outburst . we monitored the outburst with the optical high - resolution spectroscopy and multicolor photometry . in 2006 july zand reached its historical maximum at . aftermag decline in mid - august , it kept its brightness at a high level of up to 2007 january . during this period , rapid photometric variations withmag on the timescale of hours developed . simultaneously , high - velocity satellite components appeared on both sides of the h and h emission line profiles . their presence was transient , being detected to the end of 2006 . they were launched asymmetrically with the red / blue velocity ratio of 1.21.3 . from about mid - august onward they became symmetric atkms , reducing the velocity tokms at their disappearance . spectral properties of these satellite emissions indicated the ejection of bipolar jets collimated within an average opening angle of 6 . if the jets were expelled at the escape velocity then the mass of the accreting white dwarf is . we estimated the average outflow rate via jets to , during their august september maximum , which corresponds to the emitting mass in jets , . during their lifetime , the jets released the total mass of . evolution in the rapid photometric variability and asymmetric ejection of jets around the optical maximum can be explained by a disruption of the inner parts of the disk caused by radiation - induced warping of the disk .

0811.0158

collimated jets from astrophysical objects represent very exciting events . they have been detected from various kinds of objects . a common element in all jet - producing systems is the central accreting star , which implies that their accretion disks play an important role in the formation of this type of mass outflow ( e.g. * ? ? ? * ; * ? ? ? * and references therein ) . symbiotic stars are long - period interacting binaries ( orbital periods are in the range of years ) , in which a white dwarf ( wd ) accretes material from the wind of a cool giant . this process generates a very hot and luminous source of radiation ( @xmath17k , @xmath18@xmath19 ) and accumulates a certain amount of material onto the wd s surface that enlarges its effective radius to about 0.1@xmath20 during their quiescent phases . during active phases a large , geometrically and optically thick circumstellar disk with radius of a few @xmath20 develops around the accretor ( * ? ? ? * figs . 26 and 27 ) therefore , the symbiotic stars are good candidates for producing high - velocity outflows in the form of collimated jets , mainly during outbursts . however , signatures of collimated outflows were indicated for only 10 from roughly 200 known symbiotics @xcite : seven by resolving their spatial structure with the radio and x - ray imaging ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and three by the optical spectroscopy ( e.g. * ? ? ? * ; * ? ? ? nevertheless , jets in symbiotic binaries are investigated very intensively from x - ray to radio wavelengths to understand better the physical process driving the collimated high - velocity mass outflow from accreting white dwarfs ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? zand is considered a prototype symbiotic star . its recent activity started from 2000 autumn and reached the optical maxima in 2000 december , 2004 september and 2006 july ( fig . 1 ) . during the previous well observed large eruptions ( 1985 , 2000 ) , signatures of high - velocity outflows were detected through the broadening of emission lines , p - cygni type of profiles and the extended wings of hydrogen lines ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but never in the form of collimated bipolar jets . the only signature for a nonsymmetric outflow from zand was indicated by the 5ghz radio map from 2001 september , which showed a transient jet - like extension @xcite however , no counterpart in the optical spectrum could be recognized @xcite . recently , during the maximum of the 2006 outburst , @xcite discovered spectral signatures of bipolar jets from zand . their evidence in the optical spectra was confirmed by @xcite and @xcite . in this contribution we investigate evolution of the jet features from their first appearance in 2006 july to their disappearance at the end of 2006 , determine their physical parameters , and discuss the disk jet connection during the outburst . ccccc date & julian date & region & @xmath21 & obs . + ( dd / mm / yyyy ) & jd 24 ... & ( nm ) & ( min ) & + 15/12/2005 & 53720.127 & 406 - 776 & 3@xmath2210 & het + 04/06/2006 & 53890.528 & 462 - 472 & 30 & asiago + & 53890.551 & 482 - 492 & 30 & asiago + & 53890.575 & 652 - 666 & 30 & asiago + 25/07/2006 & 53942.809 & 642 - 671 & 5 & ddo + 01/08/2006 & 53949.837 & 462 - 492 & 15 & ddo + 30/08/2006 & 53978.688 & 462 - 492 & 20 & ddo + & 53978.706 & 642 - 671 & 10 & ddo + 10/09/2006 & 53989.602 & 462 - 472 & 30 & asiago + & 53989.626 & 482 - 492 & 30 & asiago + & 53989.650 & 652 - 666 & 30 & asiago + 18/10/2006 & 54027.387 & 641 - 692 & 33 & ondejov + 09/11/2006 & 54049.418 & 641 - 692 & 50 & ondejov + 23/11/2006 & 54063.428 & 641 - 692 & 50 & ondejov + 27/12/2006 & 54096.905 & 575 - 690 & 20 & oao + & 54096.919 & 575 - 690 & 20 & oao + 30/01/2007 & 54131.250 & 652 - 666 & 5 & asiago + & 54131.263 & 652 - 666 & 30 & asiago + & 54131.286 & 672 - 686 & 30 & asiago + 11/04/2007 & 54202.402 & 642 - 671 & 5.5 & ddo + 22/06/2007 & 54274.697 & 642 - 671 & 6.0 & ddo + 31/07/2007 & 54313.790 & 642 - 671 & 3.0 & ddo + 27/08/2007 & 54340.485 & 652 - 686 & 30 & asiago + our observations of zand during its 2006 outburst were carried out at different observatories . \(i ) at the mcdonald observatory by the 9.2-m hobby eberly telescope ( * ? ? ? * het in table 1 ) just prior to the rise of the major 2006 outburst . observations were performed using the echelle high dispersion spectrograph @xcite with the 316g5936 cross - disperser , 3@xmath23 fiber , and 2 ccds ( 4096@xmath224100 ) with 2@xmath221 binning . data were reduced using the standard iraf procedures , which involved bias , bad pixels , flatfield , scattered light , continuum , and cosmic rays corrections . wavelength calibration was done using th - ar lamp and has an error of about 0.1kms@xmath8 . observations were recorded in 71 orders which cover the spectral regions 40605880 , 60107760 with the resolution of about r = 60000 and s / n@xmath2200 in the continuum . for the purpose of this paper we use only regions containing h@xmath5 , h@xmath6 , heii4686 , hei4713 and the raman - scattered ovi6825 line . \(ii ) at the david dunlap observatory , university of toronto ( ddo ) the high - resolution spectroscopy was performed by the single dispersion slit spectrograph equipped with a jobin yovon horiba ccd detector ( 2048@xmath22512 pixels of 13.5@xmath24 m size ; thinned back illuminated chip ) mounted at the cassegrain focus of the 1.88-m telescope . the resolution power was 12000 and 8000 around the h@xmath5 and heii4686@xmath25h@xmath6 region , respectively . basic treatment of the spectra was done using the iraf - package software . during each night two very different exposures were applied to obtain a well defined h@xmath5 profile and the continuum . telluric sharp absorptions in the h@xmath5 region were eliminated with the aid of simultaneously observed standard star 50boo . \(iii ) at the asiago astrophysical observatory ( asiago ) . here the high - dispersion spectroscopy was secured by the reosc echelle spectrograph equipped with a aimo e2vccd47 - 10 back illuminated ccd detector ( 1100@xmath221100 pixels of 13@xmath24 m size ) mounted at the cassegrain focus of the 1.82-m telescope at mt . the resolution power with the 200@xmath24 m width slit was approximately 25000 and 17000 at the h@xmath5 and heii4686 - h@xmath6 region , respectively . the spectroscopic data were processed with the midas software package and a software developed at the astronomical observatory of capodimonte in napoli . ( iv ) at the ondejov observatory ( ondejov ) the high - resolution spectroscopy was performed by using the coude single - dispersion slit spectrograph of 2-m reflector and the bror ccd camera with the site-005 800@xmath222030pixels chip . the resolution power at the h@xmath5 region was 10000 . standard initial reduction of ccd spectra ( bias subtraction , flat - fielding and wavelength calibration ) was carried out using modified midas and iraf packages . final processing of the data was done with the aid of the spefo - package software developed at the ondejov observatory @xcite . \(v ) at the okayama astrophysical observatory ( oao ) the high - dispersion spectra were secured with the high - dispersion echelle spectrograph @xcite at the f/29 coud focus of the 1.88-m telescope . the dimension of the ccd ( eev 42 - 80 ) was 2048@xmath224096 pixels of 13.5@xmath26 . the red cross - disperser was used . the spectral resolving power was 54000 around the h@xmath5 region . the reduction and analysis was performed with the iraf - package software . two similar exposures were co - added to obtain a better signal - to - noise ratio ( s / n ) . \(vi ) at the skalnat pleso and star lesn ( pavilion g2 ) observatories , classical photoelectric @xmath0 measurements were carried out by single - channel photometers mounted in the cassegrain foci of 0.6-m reflectors ( see * ? ? ? * in detail ) . the star bd+474192 ( sao53150 , @xmath27 = 8.99 , @xmath28 = 0.41 , @xmath29 = 0.14 , @xmath30 = 0.10 ; c1 in fig . 3 ) was used as the standard star for both photoelectric and ccd observations . fast ccd photometry was performed at star lesn observatory using the 0.5-m telescope ( pavilion g1 ) . the sbig st10 mxe ccd camera with the chip 2184@xmath221472 pixels and the @xmath31 johnson - cousins filter set were mounted at the newtonian focus . the size of the pixel is 6.8@xmath24 m and the scale 0.56@xmath23/pixel , corresponding to the field of view ( fov ) of a ccd frame of about 24@xmath2216 arcmin . ( vii ) at the monteboo observatory of the masaryk university in brno ( czech republic ) , additional high - time - resolution optical photometry was obtained during two nights , 2006 september 6 and 2006 september 11 . the observational data were carried out using sbig st-8 dual chip ccd camera and kron - cousins @xmath32 filter set attached at the newtonian focus of the 602/2780 telescope . the ccd chip has 1534@xmath221020 pixels with the size of 9@xmath24 m and the scale 2.04@xmath23/pixel ( [email protected]@xmath2211.4arcmin ) . all ccd frames were dark - subtracted , flat - fielded and corrected for cosmic rays . heliocentric correction was applied for all data points . ccd photometric data were processed with the midas software package and a software c - munipack , developed at the masaryk university in brno . other details of the ccd photometric reduction were described by @xcite . to verify the reality of zand rapid variations during selected nights we measured additional comparison stars : c2 ( gsc 03645 - 01592 , @xmath27 = 11.17 , @xmath28 = 0.53 ) , c3 ( gsc2.2 n012002338 , @xmath33mag ) and c4 ( sao53133 , @xmath27 = 9.17 , @xmath28 = 1.36 , @xmath29 = 1.11 ) . the best results were achieved using the standard c1 , because it is very similar to zand during outbursts in both the brightness and the spectral type ( see figs . 2 and 3 ) . the errors of the zand c1 differences were mostly less than 0.01mag . the journal of spectroscopic observations is given in table 1 . a correction for heliocentric velocity was applied to all spectra . arbitrary flux units were converted to absolute fluxes with the aid of the simultaneous @xmath0 photometry corrected for emission lines . the method was described by @xcite . uncertainties of such a continuum calibration are of a few percent for the star s brightness around 9mag in the @xmath34 passbands ( see * ? ? ? * in detail ) . observations were dereddened with @xmath35 = 0.30 and resulting parameters were scaled to a distance of 1.5kpc ( e.g. * ? ? ? the top - right panel of fig . 1 shows the @xmath36 lcs of zand covering its recent 2006 active phase . prior to and after the peak july s magnitudes , the color index @xmath37 ( figs . 1 and 2 ) . in july , during the maximum , the index had became temporarily much bluer ( @xmath38 ) , which could be caused by a transient increase of the nebular emission . to understand better this behavior we modeled the @xmath39 flux points ( fig . 2 ) fluxes were estimated from the measured @xmath40 magnitudes , corrected for the h@xmath5 emission , using transformations of @xcite for the less red stars . ] using the method of disentangling the composite continuum as described by @xcite . our solutions correspond to a low temperature stellar source radiating at @xmath41k that dominates the optical spectrum . this suggests a significant contribution from a warm shell . at the maximum ( the model 1 in fig . 2 ) , the sed requires a larger nebular contribution ( @xmath42 = 2.3 in @xmath43 , emission measure @xmath44@xmath45 at @xmath46k ) , than prior to and after it ( @xmath47 in @xmath43 , @xmath48@xmath45 ) . the range of temperatures and scaling factors in the models of the stellar source corresponds to its effective radius @xmath49@xmath20 . we note that the absence of ultraviolet data did not allow us to determine the model parameters precisely . uncertainties in @xmath50 and @xmath51 are as high as 30 and 50% , respectively . the effect of the transient increase of the nebular emission at the maximum could be interpreted as a result of a mass ejection from the active object into the particle bounded nebula , because new emitters will convert the excess of ionizing photons there into the nebular radiation . the effect of brightening of symbiotic binary due to an increase in the mass - loss rate was originally suggested by @xcite . in addition , we searched for a short - term photometric variability on the timescales of minutes to hours ( fig . 3 ) . just prior to and after the active stage we observed an irregular variation within @xmath52mag ( top panels of fig . 3 ; previously measured also by * ? ? ? * ; * ? ? ? * ) , whereas around / after the maximum its amplitude increased by a factor of @xmath23 , to @xmath53mag ( bottom panels of fig . 3 ; the first detection to date ) . their source can be identified with the aid of the corresponding sed . the false @xmath54k photosphere , which is associated with the disk , dominated the optical domain , while contributions from the giant and the nebula were negligible ( fig . 2 ) . therefore , the @xmath20.06mag variations had to be produced by the disk . also the large size of the disk ( @xmath55@xmath20 ) supports this interpretation the timescale of the rapid photometric fluctuations is consistent with the dynamical time at the outer edge of the hot stellar source . figures 1 , 4 and 5 show evolution of the h@xmath5 , h@xmath6 , heii4686 , hei4713 , and the raman - scattered ovi 1032 profiles throughout the whole 2006 - 07 outburst . at the beginning of the 2006 active phase , prior to the initial rise ( spectrum from 15/12/2005 ) as well as prior to the maximum ( 03/06/2006 ) , the hydrogen and helium line profiles showed a simple emission core . at these days , emission wings of h@xmath5 extended to @xmath56 and @xmath57kms@xmath8 , respectively , while those of h@xmath6 persisted within @xmath58kms@xmath8 on both spectra . the emission core of the heii4686 line was placed symmetrically with respect to its reference wavelength ( fig . 1 ) . its strong flux in the 15/12/2005 spectrum , f(heii4686 ) = 2.3@xmath59@xmath60 , corresponded to the effective temperature of the ionizing source , @xmath61k ( f(h@xmath6 ) = 5.5 and f(hei4471 ) = 0.16@xmath59@xmath60 ; * ? ? ? * ) . in spite of a high temperature from nebular emission lines the raman - scattered ovi 1032 line disappeared entirely ( fig . our first spectrum from the maximum ( 25/07/2006 ) revealed a complex of strong absorptions in the h@xmath5 profile . one was extended to @xmath62kms@xmath8 and others cut the emission core at @xmath63kms@xmath8 and @xmath64kms@xmath8 ( fig . 4 ) . in addition , a pronounced s@xmath65 satellite component was present and located at + 1385kms@xmath8 . one week later ( 01/08/2006 ) strong absorptions at @xmath66kms@xmath8 and @xmath67kms@xmath8 appeared in the h@xmath6 and hei4713 lines ( figs . 1 and 4 ) . two satellite components to the h@xmath6 emission core were clearly recognizable and placed asymmetrically at @xmath68kms@xmath8 and @xmath69kms@xmath8 . from august 30th we observed typical broad emission wings with the extension of about @xmath70 and @xmath71kms@xmath8 in the h@xmath5 and h@xmath6 profile , respectively . they were accompanied by pronounced bipolarly located satellite emissions . from october to december their fluxes were becoming fainter and at the end of 2007 january only some remnants in a form of faint spikes could be recognized ( fig . we attribute these satellite components to bipolar jets . during the presence of jets , f(heii4686 ) fluxes decreased significantly ( f(heii4686)/f(h@xmath6)@xmath720.05 ) and their profiles were shifted blueward ( fig . 1 ) as during the 2000 - 01 maximum @xcite . the raman line was not detectable ( see also * ? ? ? * ) . it became to be visible on our 27/12/2006 spectrum as a faint redward - shifted emission around 6838 ( fig . our spectra taken after the maximum , from 11/04/2007 to 27/08/2007 , showed only a simple emission core of h@xmath5 and h@xmath6 lines with decreasing extension of their emission wings ( fig . heii4686 line became stronger , being located again around its reference wavelength as prior to the outburst ( fig . 1 ) . the raman emission band strengthened significantly and was comparable in the profile with that from quiescent phase ( fig . 5 ) . observed seds from the optical maximum require a very low temperature of the hot object , @xmath73k ( sect . in contrast , a significantly higher temperature of the ionizing source , @xmath61 and 90000k , was derived from the hydrogen and helium emission lines prior to the outbursts ( sect 3.2 ) and during its maximum @xcite , respectively . thus the hot active object was characterized by the two - temperature type of spectrum with @xmath74 as during the 2000 maximum . this situation can be explained by a disk - like structure of the hot object viewed under a high inclination angle . then the disk occults the hot ionizing source in the line of sight , while the nebula above / below the disk can easily be ionized ( see sect . * in detail ) . another effect resulting from the disk - structured hot object is the blueward shift of the heii4686 profile that develops during the outbursts , while out of active phases it is placed symmetrically with respect to the reference wavelength ( fig . 1 here and fig . 2 in the size of the he@xmath75 zone was probably very small at the maximum ( note the very low flux of the heii4686 line during the maximum , sect . 3.2 ) and thus the disk blocked a fraction of its redward shifted emission in the direction of the observer for the orbital inclination of about 75@xmath76 . the disappearance of the raman line during active phases at high @xmath77 ( sect . 3.2 and fig . 5 here ; * ; * ? ? ? * ) can also be a result of the disk - like structure of the hot object . the scattering process acts on the neutral atoms of hydrogen in the wind from the giant . so , if the ionizing source is indeed capable of producing the o@xmath78 ions ( note that the ionization potential @xmath79(o@xmath80ev requires @xmath81k @xcite ) , but no raman line is observed , implies that the o@xmath78 zone is probably too small ( cf . appendix b in * ? ? ? * ) , so that the disk blocks its radiation in directions to the densest parts of the neutral wind at the orbital plane . however , the arise of a faint _ redward_-shifted raman line on our spectra from 27/12/2006 and 30/01/2007 ( fig . 5 ) can be interpreted as due to the raman scattering in the outer parts of the neutral giant s wind that move _ from _ the original ovi photons . this signals a gradual dilution of the disk . following development of a strong raman emission with a typical _ blue_-shifted shoulder in the profile ( see fig . 5 ) reflects total dilution of the optically thick material at the orbital plane , because the original ovi photons then can be scattered in the densest part of the neutral wind from the giant along the binary axis that moves _ against _ them ( see also * ? ? ? * for a general interpretation of the raman line profile in symbiotic stars ) . the simultaneous disappearance of the disk and jets ( cf . figs . 4 and 5 ) is consistent with the necessity of the disk for the presence of jets ( e.g. * ? ? ? the disk - jet connection is discussed below in sect . finally , the disk - like structure of the hot _ active _ object in zand was indicated for the first time by the model seds from the 1985 outburst , which revealed directly the presence of the two - temperature uv spectrum ( sect . 5.3.4 . of * * ) . to isolate the jet components from the whole line profile , we formally fitted the emission line core and its extended foot component with two gaussian functions . then the residual jet emissions were compared with additional gaussians . from their fitted parameters ( the central wavelength , maximum , @xmath82 , and the width @xmath83 ) we derived the jet s radial velocity @xmath84 , flux @xmath85 and the width @xmath86 ( table 2 ) . corresponding fits are shown in fig . this approach and the resolution of our spectra allowed us to estimate uncertainties in the @xmath84 values to 10@xmath8723kms@xmath8 in h@xmath5 , @xmath88 widths were possible to adjust to observations within [email protected] and fluxes within 10@xmath8720% of the observed values . the smaller uncertainty corresponds to the stronger jet feature and vice versa . our observations and those published by @xcite and @xcite show an interesting behavior in @xmath84 . from the jets launching in 2006 july to the beginning of 2006 august , the satellite components were located asymmetrically , being shifted by about + 150kms@xmath8 from the symmetric position ( the top panel of fig . their average velocity was 1280kms@xmath8 . then , in the time of a few days , the jets became symmetric in their velocities . they had persisted at @xmath89kms@xmath8 to about mid of september when reduced suddenly their @xmath90 to @xmath21100kms@xmath8 till the end of their detection . the middle panel of fig . 6 plots the velocity ratio of the faster jet to its slower counterpart that reflects directly the evolution in the jets position . at the initial stage of the jets presence , this ratio was 1.21.3 , whereas during the following period the velocity ratio was @xmath91 . the bottom panel of fig . 6 shows the flux ratio of the red jet component to the blue one , @xmath92 , for the h@xmath5 line as a function of the orbital phase . we interpret the decrease of this ratio , indicated around the inferior conjunction of the giant , as an occultation effect ( sect . . relatively small width of the satellite components ( @xmath93kms@xmath8 , table 2 ) with respect to , for example , the broad wings of the h@xmath5 line ( @xmath94(h@xmath95 4000kms@xmath8 , e.g. fig . 4 ) implies that ( i ) the jets are highly collimated , and ( ii ) the emitting particles can be characterized with one constant velocity within the visible part of the jet . in the following two subsections we discuss some consequences of these jet properties . the winds of luminous hot stars are accelerated by the radiation field , produced by the underlying photosphere ( e.g. * ? ? ? they are called line driven winds , because they are driven by absorption in spectral lines . in this process the photons transfer their energy into the kinetic energy of the wind , and , due to the doppler shifts in the radial direction , the process acts throughout the _ entire _ wind , accelerating gradually its particles from the zero to the terminal velocity , @xmath96 . as a result , we observe broad emission lines in the spectra of these stars . for symbiotic binaries , @xcite suggested that the broad wings of the h@xmath5 profiles ( @xmath97kms@xmath8 ) are formed in the stellar wind from their hot components . the relatively narrow width of the jet features suggests that they are driven by a different mechanisms . if the jets were driven like the wind , their observed profile would be a few thousands kms@xmath8 broad , because of the large velocity dispersion in the radial direction ( @xmath98 ) , and thus would hardly be detectable due to their low emission . qualitatively , the narrow jet profiles imply that the jet engine works like a gun accelerating material through a _ short _ path ( relatively to the jet length ) to a high velocity , at which the shot - out - material continues to further distances . also the connection between the disk and the jets , we discuss in sect . 4 , suggests that the jet launching mechanism is very different from that accelerating the wind . generally , the jet engine transforms the kinetic energy of the inner parts of the disk to the kinetic energy of the flow in the vertical direction . in this respect , @xcite and @xcite suggested a scenario how jets could be blown from accretion disks . the narrow @xmath88 allow us to assume a constant velocity of the emitting particles within the visible part of the jet . then , if the broadening of the satellite components is caused by thermal motions , the corresponding doppler width ( @xmath99kms@xmath8 , i.e. @xmath100 in h@xmath5 ) would require unrealistically high temperature of a few millions of kelvins . in addition , the profiles of satellite lines have sharp edges , do not have typical wings , and their peaks are often flat ( fig . 1 , e.g. 10/09/2006 ) . if the jet consisted of a parallel beam of particles , i.e. having the geometry of a narrow column , we could not observe any real broadening . therefore , we ascribe the observed line width to the dispersion in the line - of - sight velocity components of the jet particles . this can be satisfied if the jet emitting medium has the geometry of a narrow conus with the peak at the central object and characterized with a small opening angle @xmath101 . ccccccccccccccc date & line & & & & @xmath102 & & @xmath103 & & @xmath104 + dd / mm / yy & & @xmath105 & @xmath106 & @xmath105 & @xmath106 & @xmath105 & @xmath106 & & @xmath105 & @xmath106 & & @xmath105 & @xmath106 & + 15/12/05&h@xmath5&&1.0& & & & & & 1.3 + 03/06/06&h@xmath5&&1.6& & & & & & 0.57 + 25/07/06&h@xmath5 & -1120 & + 1385 & & 8.2 & & 355&2.3& & 7.3&1.2 & & 1.7 & + 01/08/06&h@xmath6 & -1100 & + 1408 & 1.1 & 1.9 & 253 & 289&3.6&6.6&5.9&1.2 & 0.76&1.1 & + 30/08/06&h@xmath5 & -1250 & + 1214 & 5.3 & 5.5 & 227 & 255&1.9&5.2&6.0&1.6 & 0.91&1.0 & 2.5 + & h@xmath6 & -1267 & + 1212 & 2.2 & 2.3 & 204 & 225&2.8&4.6&5.3&1.8 & 0.86&0.97 & + 10/09/06&h@xmath5 & -1152 & + 1208 & 5.1 & 4.8 & 301 & 280&1.7&7.5&6.6&1.5 & 1.2 & 1.1 & 1.5 + & h@xmath6 & -1164 & + 1216 & 3.1 & 2.7 & 275 & 275&2.6&6.7&6.5&2.3 & 1.4 & 1.3 & + 18/10/06&h@xmath5 & -1156 & + 1165 & 4.0 & 2.9 & 237 & 226&1.6&5.9&5.5&1.0 & 0.83&0.66&2.7 + 09/11/06&h@xmath5 & -1100 & + 1100 & 2.1 & 1.7 & 237 & 215&1.5&6.1&5.6&0.56&0.59&0.49&2.4 + 23/11/06&h@xmath5 & -1120 & + 1100 & 2.0 & 1.9 & 237 & 215&1.3&6.0&5.6&0.57&0.58&0.52&2.7 + 27/12/06&h@xmath5 & -1110 & + 1110 & 0.87 & 0.60 & 247 & 237&1.2&6.3&6.1&0.22&0.38&0.33&2.5 + 30/01/07&h@xmath5&&1.1 & & & & & & 1.0 + 11/04/07&h@xmath5&&0.83& & & & & & 2.5 + 22/06/07&h@xmath5&&0.80& & & & & & 1.5 + 31/07/07&h@xmath5&&0.84& & & & & & 1.3 + 27/08/07&h@xmath5&&0.82& & & & & & 0.78 + the conus geometry of the jet and the constant velocity of its particles ( sect . 3.4.2 . ) allow us to express the observed width of the satellite component as a function of its opening angle @xmath101 and the orbital inclination @xmath107 . assuming that the jets were launched with the velocity , @xmath108 , perpendicularly to the disk plane that coincides with the orbital one , the maximum dispersion of the line - of - sight jet velocity component corresponds to @xmath109 , where the half width at the zero intensity of the jet ( @xmath110 ) is in kms@xmath8 . then for the measured central jet s velocity , @xmath111 , the opening angle can be approximated as @xmath112.\ ] ] a similar relation was also used by @xcite . corresponding parameters from table 2 , orbital inclination @xmath113 @xcite and adopting @xmath114 , yield its average value as @xmath115 where the uncertainty represents a total differential of function ( 1 ) for uncertainties in @xmath88 , @xmath84 ( sect . 3.4 ) and that in the orbital inclination , @xmath116 . it is generally considered that the jet velocity in all the jet - producing objects is of the order of the escape velocity from their central stars , which indicates that the jet - type outflows originate from the vicinity of the accretors ( e.g. * ? ? ? * and references therein ) . recently , @xcite and @xcite performed analytical estimates on how the plasma close to the central object , specifically in the boundary layer of the disk , could be accelerated to velocities larger than the local escape velocity . our observations seem to be consistent with this scenario . the high velocities of jets , @xmath117 kms@xmath8 , observed at large distances from the central star ( sec . 3.8.1 ) , imply @xmath118 for all possible distances of the jet ejection . however , from observations it is difficult to determine accurately the distance from the accreting body , at which jets are launched . @xcite also could not rule out the possibility that jets can be launched somewhat away from the accretor , up to @xmath119 the boundary layer radius . they noted that this case would require a strong perturbation in the disk . therefore , for a rough estimate only , we will assume that the jets were launched from the wd surface at @xmath120 . then the average value of @xmath121kms@xmath8 ( @xmath122kms@xmath8 , table 2 , @xmath113 ) and a typical wd radius of @xmath123 = 0.01@xmath20 correspond to the mass of the accretor in zand , @xmath124 . we note that this value satisfies perfectly the relation between mass and radius of wds ( e.g. * ? ? ? @xcite derived @xmath125@xmath12 from the spectroscopic orbit for @xmath126 . our value of @xmath127 derived for @xmath113 requires just a larger total system mass of @xmath128@xmath12 to satisfy the mass function of 0.024@xmath12 @xcite . however , in the real case , the same value of @xmath127 can be obtained for different combinations of the @xmath129 ratio and the distance of the jet ejection . finally , we note that the large effective radius of the active star during the optical maximum derived from the sed ( @xmath130@xmath20@xmath131 , sect . 3.1 ) does not contradict to launching the jets from the vicinity of the wd surface , because of the disk - like structure of the hot active object seen under a high inclination angle ( sect . 3.3 ) . in this section we derive some constraints supporting that the jet emission is due to the photoionization of hydrogen and that the medium is optically thin in the direction of the observer . \(i ) according to fig . 3 of @xcite and the @xmath132 fluxes from table 2 , the average balmer decrement @xmath132(h@xmath133(h@xmath134 , which is close to the theoretical value of 2.75 , given by the recombination process at the electron temperature of 20000k ( e.g. * ? ? ? equivalent widths presented by burmeister & leedjrv were converted to fluxes with the aid of our continuum calibration in table 2 . the somewhat lower values of the observed decrement than the theoretical one could be caused by a partial opacity of the nebular jet medium in the balmer lines . generally , the opacity in the h@xmath5 line is larger than in the h@xmath6 . the effect , however , should not be significant . also the high orbital inclination and the small value of @xmath101 correspond to a relatively small intersection , @xmath135 , of the jet conus with the line of sight , and thus to a low value of the optical depth ( @xmath136 ) , which supports rather optically thin regime in the direction of the observer . for a comparison , the jets produced by the symbiotic star mwc560 are optically thick , because they are seen nearly pole - on , i.e. their @xmath135 and thus @xmath137 are very large , which gives rise the blueshifted jet _ absorption _ components in the spectrum @xcite . \(ii ) if the jet emission is due to the recombination of free electrons with protons , then it can be created only within the ionized fraction of the ejected material . in appendix a we calculate the extension of the jet nebula as the distance from the central hot star , at which ionizing photons are completely consumed by jet particles along paths outward from the ionizing star . according to eq . ( a2 ) the jet radius is limited by its strmgren sphere at the distance @xmath138 from the ionizing source , i.e. @xmath139 where the parameters , @xmath140 are explained in appendix a. in section 3.8 and figure 7 ) we demonstrate that the jets radius , determined independently from their observed luminosity and the opening angle ( eq . ( 7 ) ) , agrees well with the radius of the strmgren sphere , given by the luminosity and the temperature of the ionizing source . this result confirms that the satellite components are really created by recombinations within the jets , ionized by the central star . evolution in the fluxes of jets is consistent with such an ionization structure . between the beginning of 2006 august and the end of september , their fluxes settled ( with some fluctuations ) around a maximum ( * ? ? ? * ; * ? ? ? * this paper ) . no increasing trend , as could be expected from an increase of the emission measure of jets ( see eq . ( 5 ) ) due to their expansion , was observed during this period . this implies that the emitting mass in jets was approximately constant , and thus suggests that the satellite components represent only the illuminated part of jets ionized by the hot central object , which is bounded by the strmgren sphere . figure 9 ( see appendix b ) shows that the _ number _ of hydrogen ionizing photons is approximately constant for temperatures between about 50000 and 130000k at a fixed luminosity of a black - body source . therefore , the hot star produced roughly constant flux of ionizing photons , and thus also relevant fluxes of jets , in spite that its temperature increased from 75000k to 115000k during the above mentioned period @xcite . observed fluctuations were probably caused by those in the luminosity of the ionizing object . the recombination process of the radiation by the jet plasma , the resulting fluxes , and the geometrical and kinematics parameters of the jets put some constraints to determine the mass loss rate through jets . assuming that the jets were expelled into the solid angle @xmath141 $ ] of the jet nozzle ( sect . 3.4.2 . ) , then the corresponding mass - loss rate , @xmath142 , and the mean particle concentration , @xmath143 , in the jets volume , are connected via the mass continuity equation as @xmath144 where @xmath24 is the mean molecular weight and @xmath145 is the mass of the hydrogen atom . total luminosity produced by jets through the recombination transition of the h@xmath5 line , @xmath146 ) , is related to the line emissivity , @xmath147 ( erg@xmath45s@xmath8 ) , by @xmath148 where @xmath149 is the volume emission coefficient in h@xmath5 , @xmath150 and @xmath151 are concentrations of electrons and protons and @xmath152 and @xmath103 are the volume and emission measure of the jets , respectively . for the optically thin medium of jets , the luminosity can be determined from the observed fluxes as @xmath153 . further , eq . ( 5 ) assumes a completely ionized medium ( i.e. @xmath154 ) , radiating at a constant electron temperature , i.e. @xmath149 is constant throughout the jet emitting volume . according to sect . we approximate its geometry by the conus , i.e. @xmath155 substituting eq . ( 6 ) into eq . ( 5 ) , we can express the jet radius by means of the parameters , obtained directly from observations ( @xmath156 and @xmath157 ) , as @xmath158 this expression allows us to rewrite eq . ( 4 ) as @xmath159}\right)^{2/3 } \frac{v_{\rm jet}}{[{\rm km\,s^{-1}}]}~ m_{\sun}\,{\rm yr^{-1}},\ ] ] and/or as a function of the radius @xmath160 as @xmath161 } \frac{\delta\omega}{[{\rm sr } ] } \frac{l_{\rm jet}}{[l_{\sun}]}\right)^{1/2 } \frac{v_{\rm jet}}{[{\rm km\,s^{-1}}]}~ m_{\sun}\,{\rm yr^{-1}},\ ] ] where the factor @xmath162 = @xmath163 or @xmath164 and @xmath165 = @xmath166 or @xmath167 for luminosities in h@xmath5 or h@xmath6 , respectively . volume emission coefficients for @xmath168k , @xmath149 = 1.83 and @xmath169 @xcite . according to eqs . ( 8) and ( 9 ) , determination of @xmath142 requires to estimate reasonable values of @xmath143 or @xmath160 . figure 7 plots dependencies between these parameters . below we discuss some constraints that allow us to determine more accurate ranges of the jet parameters . \(i ) @xcite introduced conditions for thermally launching jets from accretion disks around wds . in their model the accreted material is strongly shocked due to large gradients of physical quantities in the boundary layer and cools on the timescale longer than its ejection time from the disk . the model requires large accretion rates of @xmath170@xmath14 . some observational evidences of the disk - jet connection , as described in section 4 , could support this scenario . in agreement with this suggestion and the fact that there is a large disk encompassing the wd ( sects . 3.1 . and 3.3 . ) , we can exclude rates of @xmath171@xmath14 ( i.e. @xmath172@xmath14 , sect . 4.1 ) that correspond to @xmath173@xmath45 and @xmath174au ( see fig . 7 ) , because of too small jet radii that would be significantly occulted by the disk for the orbital inclination of @xmath175 . observations do not indicate this case ( e.g. table 2 , fig . 6 ) . from the other side , we exclude rates @xmath176@xmath14 ( i.e. @xmath177@xmath45 , @xmath178au , fig . 7 ) , because this would require too high accretion rates of @xmath179@xmath14 ( see also sect . 4.1 ) . as the mass of the disk can be estimated to @xmath180@xmath12 ( e.g. @xmath181@xmath20 , @xmath182 @xcite and average density @xmath183 = -7 to -8 ) , the accretion rates of about @xmath184@xmath14 would exhaust the disk during a short time . however , observations indicate the presence of a large disk with jets for about six months , from 2006 july to 2007 january ( sect . 3.3 , fig . 5 ) . these conditions thus limit @xmath142 to @xmath185@xmath14 and consequently , @xmath186@xmath45 and @xmath187au . \(ii ) during active phases of zand the emission measure produced by the ionized hot star wind was determined to @xmath188@xmath45 ( table 1 in * ? ? ? * ) . it represents the so - called low - temperature nebula , which is subject to eclipses in active symbiotic systems @xcite , and thus it is located within the radius of the giant , @xmath189 , around the hot component . for zand , the average particle concentration in the wind around the hot star within the radius @xmath190@xmath20 is @xmath191@xmath45 . for a comparison , the mean concentration of the jet particles within this distance , @xmath192@xmath45 , which corresponds to @xmath193@xmath14 ( fig . this suggests that @xmath194 and also the normalized mass loss rate ( e.g. into 1sr ) via the jets is larger than that through the wind . \(iii ) the transient decrease in the flux ratio , @xmath195 , measured during the inferior conjunction of the giant ( fig . 6 ) , could be caused by the occultation of the red jet component by the stellar disk of the giant ( also noted by * ? ? ? as @xmath196au , this observation suggests that the radii of jets are @xmath197au . however , uncertainties in the orbital elements do not allow a more accurate estimate ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? finally , we note that some theoretical works , devoted to modeling the jets in symbiotic stars , also adopt the jet radius of 1au ( e.g. * ? ? ? * ; * ? ? ? therefore , we determined the mass loss rate @xmath142 in table 2 for a representative value of @xmath198au . the mean concentration and the volume of jets determine their emitting mass as @xmath199 which can be rewritten with the aid of eqs . ( 6 ) and ( 7 ) as @xmath200 } \frac{l_{\rm jet}}{[l_{\sun}]}\right)^{1/2 } \left(\frac{r_{\rm jet}}{[au]}\right)^{3/2 } ~m_{\sun},\ ] ] where @xmath201 or 9.3@xmath202 for luminosities in h@xmath5 or h@xmath6 , respectively . the average value of @xmath203 ( sect . 3.5 ) , @xmath204 in h@xmath5 ( from the 2006 august september maximum ) and @xmath160 = 1au yield the emitting mass of @xmath205 in both jet components . @xcite derived significantly larger value of @xmath206@xmath12 for the similar luminosity of the jet pair they observed on 12/08/2006 . this difference results from their adopted mean density of only @xmath207@xmath45 , which requires very large size of one jet , @xmath208au . however , such an extension of the jet is beyond the limits we estimated in sect . 3.8 . with respect to our approximation , in which the jet medium has a constant particle density and emits within its strmgren radius , we approximate the total mass released by the jets during the time of their detection , @xmath209 , as @xmath210 where @xmath211 is the average mass loss rate via jets . values from table 2 ( @xmath212@xmath14 for the jet pair and @xmath213 days ) yield @xmath16@xmath12 . it is generally well known that during active phases the hot components in symbiotic binaries lose their mass in the form of wind . to estimate the rate of the mass loss produced by the active star through the stellar wind , we used the h@xmath5 method as suggested by @xcite . the method assumes that the broad wings are due to kinematics of the photoionized and optically thin stellar wind . profiles from the maximum were modeled with the same input parameters as those from the 2000-maximum . the disk radius , @xmath214@xmath20 , and the disk thickness at its edge , @xmath215 = 1.92@xmath20 , were adjusted to the effective radius of the hot object derived from the sed assuming a flared disk with @xmath182 . prior to and after the outburst we used @xmath216@xmath20 and @xmath215 = 1.05@xmath20 . the results do not depend critically on these parameters . examples of comparison between the modeled and observed profiles are shown in figs . 4 and 8 . synthetic profiles fit well the observed wings for @xmath217kms@xmath8 . corresponding mass loss rates through the wind , @xmath104 are around 2@xmath218 ( table 2 ) . other model parameters are the terminal velocity , which runs between 1500kms@xmath8 and 3000kms@xmath8 , and the acceleration parameter @xmath219 ( cf . ( 2 ) in * ? ? ? it is of interest to note that the observation made just prior to the outbursts ( the het spectrum ) revealed a more complex structure in the wind outflow . the @xmath220 plot shows a faint and very broad ( fwhm @xmath221 1200kms@xmath8 ) emission component superposed on the extended blue h@xmath5 wing ( fig . the red giant in zand loses mass via the wind at @xmath222@xmath14 ( * ? ? ? * table 3 ) . according to the recent study on the wind accretion in binary stars @xcite the wd companion can accrete with efficiency up to 10% , i.e. at @xmath223@xmath14 during _ quiescent phases _ of zand ( the total system mass of 3.2@xmath12 and the wind velocity of 30kms@xmath8 at the accretor were adopted ) . this quantity corresponds to the accretion luminosity of @xmath224@xmath19 ( parameters from sect . 3.6 ) , which is at least a factor of 30 smaller than the observed luminosity of the hot stellar object ( e.g. * ? ? ? this discrepancy led to suggestion that an additional source of the energy in a form of steadily burning material on the white dwarf surface has to be present ( e.g. * ? ? ? * ) . during the recent active phase , the @xmath142 rate put limits for the accretion rate through the disk , @xmath225 , because both the rates are proportional . typical values of @xmath226 are in the range [email protected] ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , although @xcite found that @xmath142 can almost be equal to @xmath225 , if the accretion energy is converted efficiently into magnetic energy and is emitted in the form of a magnetically dominated outflow or jet . here we will consider the ratio @xmath227 . accordingly , the observed @xmath228@xmath14 thus requires @xmath229@xmath14 . two possible mechanisms how to increase the accretion rate are either a disk instability or an increase in the mass loss rate from the giant . if the latter possibility had been the case , the normal red giant in zand would have increased its @xmath230 by 3 orders of magnitude from a few times @xmath231@xmath14 during quiescence to @xmath232@xmath14 for a short time during the activity , to power the observed jets . this seems to be unlikely . thus , assuming that @xmath230 is more or less constant , a disk instability had to be responsible for the transient , but significant , increase in @xmath225 through the disk . we summarize critical observations supporting this suggestion as follows : 1 . the rapid photometric variability originates in the disk ( sects . 3.1 and 3.3 ) . figure 3 shows the change from an irregular low - amplitude variation prior to the outburst to a slower , but higher - magnitude variation , observed during the maximum when the jets were launched . according to @xcite , who examined the disk - jet connection in chcyg , the disappearance of the fastest variations indicates that the innermost disk was disrupted around the time when the jets were produced . this interpretation is based on the relationship between the disk radius and variability timescale ( slower variations come from larger disk radii and vice versa , as summarized by * ? ? ? * ) . also the gas - dynamical modeling of the flow structures during the zand outburst suggested that the sudden increase of the accretion rate could result from the disruption of the disk @xcite . the disk disruption , i.e. the removal of material from its inner parts , implies that the jets were accretion - powered . dramatic spectroscopic evolution around the optical maximum ( 2006 july / august ) , as indicated by complex absorption / emission profiles accompanied by the asymmetric ejection of jets ( fig . 4 , sect . 3.2 here and fig . 2 in * ? ? ? * ) , could also be a direct result of the disk disruption . it is of interest to note that similar behavior was also observed during chcyg outbursts in 1982 - 84 , 1992 - 94 and 1998 - 00 , but on somewhat longer timescales ( see figs . 1 and 4 of * ? ? ? all the outbursts were followed by jet - extended features on the radio images @xcite . however , for zand these features have been recorded for the first time . for example , during its 2000 - 03 major outburst no jets were present ( see * ? ? ? * ; * ? ? ? * ) . simultaneous presence of both the disk and jets during the period 2006 july december and their disappearance from 2007 january ( sect . 3.3 , figs . 4 and 5 ) suggest that bulk of the accretion energy was released in the form of jets . as a result the hot stellar source in zand had reduced its radiation significantly and therefore we observed a rapid decline in the star s brightness from 2007 january ( see figs . 1 and 5 ) . from this point of view , the absence of jets during the previous ( 2000 - 03 ) major outburst is consistent with a slower decline in the brightness from the 2000-maximum ( see fig . 1 ) and the disk presence for a longer time ( @xmath21.5 years , * ? ? ? * ) . in both the outbursts the presence of a large disk was terminated observationally by a small optical rebrightening accompanied by the emergence of the raman line ( fig . 5 here and sect . 3.6 in * ? ? ? we note that the disk - jet connection in zand , as summarized in the points above , has some similarities with a hard state of the microquasar grs1915 + 105 as described by @xcite and later discussed by @xcite . for the symbiotic star chcyg this similarity was pointed out by @xcite . the asymmetric jet episode in zand lasted for less than one month from the jets launching in 2006 july to the beginning of 2006 august ( fig . 6 , sect . the velocity asymmetry in jets was so far revealed only for other objects and at significantly larger timescales of years to decades ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? thus , the very short duration of the asymmetric jets in zand was probably connected with a disk instability induced during the optical maximum . qualitatively , the instability could cause different conditions on the accretor s poles for the ejection of jets . particularly , the anti - correlation of the opening angles to the jet velocities at the initial stage of the jets launching ( fig . 6 ) could be a possible cause of the observed jet s asymmetry . the larger the jet nozzle is , the lower ejection velocity can be expected to drive the jet flow at the same rate ( eq . this possibility was already suggested by @xcite for magnetically driven jets if the opening angles are different due to different pressure gradients . the short - term , @xmath4mag , photometric variations represent another type of variability , which developed during the jet ejection ( fig . 3 , sect . their source was associated with the disk ( sect . 3.1 ) . in the previous section we ascribed the evolution in the short - term variability to disruption of the inner parts of the disk . the origin of both these effects could become better understood by investigating a disk instability due to an increase of the hot component luminosity at the outburst maximum . according to @xcite irradiation of the disk by the central star can lead an originally planar disk to become twisted and tilted out of the orbital plane . in our case , additional energy liberated during the outburst by thermonuclear burning on the wd surface increases formally the efficiency @xmath233 of the accretion process ( see eq . ( 3.10 ) in * ? ? ? * ) , which then could lead to radiation - driven warping close to the disk s center even for wd accretors . for zand , the radiation - induced warping occurs at all radii @xmath234@xmath20 for the luminosity of @xmath235@xmath19 around the optical maximum ( e.g. * ? ? ? * ) , the mass accretion rate from the wind of @xmath236@xmath14 ( sect . 4.1 ) and the wd mass of 0.64@xmath12 ( sect . 3.6 ) . with respect to the disk radius of @xmath23710@xmath20 ( sect . 3.1 ) , the warping will act from inner parts of the disk . however , connections between jets and disk variability in symbiotics have not been studied theoretically in detail yet . current theories on the warping and wobbling disks were elaborated for different types of objects and are characterized by significantly larger timescales than those suggested by observations of zand ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the main results of this paper can be summarized as follows : \(i ) between 2006 july 19 and 28 our photometric monitoring of the prototypical symbiotic star zand revealed the highest maximum of its brightness that has ever been recorded by the multicolor photometry ( @xmath2388.0 , fig . 1 ) . around the mid of august the brightness declined to @xmath239 and persisted around this level to 2007 january . during this period , rapid photometric variation ( @xmath53mag ) on the timescale of hours developed ( fig . \(ii ) the sed models and spectral characteristics from the optical maximum can be explained by a disk - like structure of the hot active object ( sect . the short - term photometric variation was produced by the disk ( sect . \(iii ) during the optical maximum a mass ejection from the active object was indicated photometrically through an increase of the nebular emission ( fig . 2 ) and spectroscopically by the development of complex absorptions on the blue side of the hydrogen and helium lines and a strong s@xmath65 jet emission ( fig . 4 ) . \(iv ) high velocity satellite components to h@xmath5 and h@xmath6 emission lines developed in the spectrum from the optical maximum . their presence in the spectrum was transient . our first and last detection was on 25/07/2006 and 27/12/2006 , respectively . their spectral properties ( table 2 ) indicated ejection of highly collimated bipolar jets . we summarize their main characteristics as follows : 1 . the jets were asymmetrical in their velocities for less than one month from their launching in 2006 july ( @xmath240 = 1.2 1.3 ) . after this episode they became suddenly symmetrical until the end of their presence ( fig . the jets were collimated within an average opening angle of [email protected] for the orbital inclination @xmath113 ( sect . 3 . the average velocity of jets was 4960kms@xmath8 . the ratio @xmath242 implies mass of the accreting wd of @xmath243 ( sect . the average outflow rate via jets was @xmath244@xmath14 , during their august september maximum , which corresponds to the emitting mass in jets , @xmath245@xmath12 ( sect . 3.8.2 ) . 5 . the total mass released by the jets during their detection was approximated to @xmath246 ( sect . 3.8.3 ) . \(v ) the short duration of the asymmetry in the jet velocities and evolution in the rapid photometric variability resulted from a disruption of the innermost disk . in particular , we pointed out a possibility of the radiation - induced warping , which can occur close to the disk s center due to the additional source of energy from the thermonuclear burning on the wd surface . for zand parameters , this type of disk instability could occur at all radii @xmath234@xmath20 from the accretor ( sect . the authors thank mario livio for constructive comments that helped improve the clarity of the paper . dmitri bisikalo and drahomr chochol are thanked for discussions . martin vako and pavel schalling are thanked for taking some photometric observations and miroslav lechta for initial reduction of the spectra taken at the ondejov observatory . this research was supported in part by a grant of the slovak academy of sciences no . 2/7010/27 and by the grant agency of the czech republic , grant no . 205/06/0217 and 205/08/h005 . jb acknowledges the support from the pennsylvania state university and the nsf - nato grant dge-0312144 and the marie curie international reintegration grant mirg - ct-2007 - 200297 . the hobby - eberly telescope ( het ) is a joint project of the university of texas at austin , the pennsylvania state university , stanford university , ludwig - maximillians - universitt mnchen , and georg - august - universitt gttingen . the het is named in honor of its principal benefactors , william p. hobby and robert e. eberly . here we calculate the boundary between the ionized and neutral part of the jet for the case that the sources of both the ionizing photons and particles are located at the central star . the boundary is determined by the locus of points , at which the flux of ionizing photons is balanced by the rate of ionization / recombination acts inside the nebula . thus the equilibrium condition between the number of ionizing photons @xmath247 , emitted into the fraction of the sphere @xmath248 around the direction of the jet , and the number of recombinations , can be expressed as @xmath249 it is assumed that the ionizing photons are emitted by the central source spherically symmetrically , @xmath250 stands for the total hydrogenic recombination coefficient in the case @xmath43 , @xmath150 and @xmath151 are concentrations of electrons and protons and @xmath251 measures the distance from the central ionizing source along the jet . for the fully ionized hydrogen plasma ( @xmath252 ) with a mean particle concentration , @xmath143 , eq . ( a1 ) can be integrated to give @xmath253 in the sense of this relation , the radius of the visible part of the jet , @xmath160 , represents that of the strmgren sphere . to determine the total number of photons , @xmath247 , capable to ionize neutral hydrogen and helium , emitted by the star of temperature @xmath254 , we integrated the plank curve from the corresponding ionization limits ( i.e. 13.598ev , @xmath255s@xmath8 and 54.416ev , @xmath256s@xmath8 for hydrogen and helium , respectively ) to @xmath257 . figure 9 shows the @xmath247 quantity as a function of the temperature for a luminosity of the ionizing source of 10000@xmath19 . it is worth to note that the function @xmath258 is slightly decreasing from @xmath273000k to higher temperatures , in spite that the total energy beyond the ionization limit increases . therefore , irrespective to a large temperature changes between 50000 and 125000k , the number of the hydrogen ionizing photons varies only within 15% of the maximum value at 73000k at a given luminosity ( see fig . 9 ) .

spectra of extranuclear clouds within the ionization cone of circinus galaxy are presented and modelled assuming photoionization by the seyfert 2 nucleus but no preconceived assumption concerning the spectral shape of the ionizing agn radiation or the gas density distribution . the most important result is that , regardless of the assumed agn spectral shape or density distribution , the metal abundances are remarkably well (.2 dex ) constrained by the data . oxygen and neon are found to be a factor of 5 below solar , while nitrogen is solar ( cf . table [ tab_z ] ) . the large n / o overabundance is in agreement with the chemical enrichment expected from the old ( circum)nuclear starburst . similar studies of extranuclear clouds in other objects could therefore provide a powerful tool to determine metallicities and trace past starburst activity in agns . other implications such as the role of shock excitation , which is effectively excluded , and the intrinsic shape of the agn ionizing continuum , which is poorly constrained by the data , are also discussed .

astro-ph9811177

the circinus galaxy ( a1409 - 65 ) is a nearby ( @xmath14 mpc ) gas rich spiral lying close to the galactic plane in a region of relatively low ( @xmath11.5 mag ) interstellar extinction ( freeman et al . several observed characteristics indicate that this galaxy hosts the nearest seyfert 2 nucleus known . these include optical images showing a spectacular [ oiii ] cone ( marconi et al . 1994 hereafter @xcite ) ; optical / ir spectra rich in prominent and narrow coronal lines ( oliva et al . 1994 , hereafter @xcite , moorwood et al . 1996 , hereafter @xcite ) ; x - ray spectra displaying a very prominent fe - k fluorescent line ( matt et al . @xcite ) and optical spectropolarimetric data which reveal relatively broad h@xmath2 emission in polarized light ( oliva et al . complementary to these is observational evidence that this galaxy has recently experienced a powerful nuclear starburst which is now traced by the near ir emission of red supergiants ( oliva et al . @xcite , maiolino et al . @xcite ) and which may have propagated outwards igniting the bright ring of o stars and hii regions visible in the image ( @xcite ) . such a starburst could have been triggered by gas moving toward the nuclear region and eventually falling onto the accretion disk around the black hole powering the agn . a debated issue is whether nuclear starbursts are common features of agns and if they are more common in type 2 than in type 1 seyferts , as suggested by e.g. 10@xmath3 m observations ( maiolino et al . @xcite ) and studies of the stellar mass to light ratios ( @xcite ) . since starbursts are predicted and observed to deeply modify the chemical abundances of the host galaxy ( e.g. matteucci & padovani @xcite ) , such an effect should also be evident in this and other seyferts . however , to the best of our knowledge , no reliable measurement of metallicity for the narrow line region clouds of seyfert 2 s exists in the literature . in particular , although has been since long known that the large [ nii]/h@xmath2 ratio typical of seyferts can not be easily explained using simple models with normal nitrogen abundances ( e.g. osterbrock @xcite , komossa & schulz @xcite ) , the question of whether its absolute ( n / h ) or relative ( e.g. n / o ) abundance is truly different than solar is still open . finding a reliable method to derive metallicities and , therefore , to trace and put constraints on past starburst activity is the main aim of this paper . -9pt # 1 lrrcrrcc & & + & & + & & + & f & i & & f & i & & @xmath4 + @xmath5 } \,\lambda\lambda$]3727 & 77 & 270 & & 78 & 133 & & 1.49 + @xmath6 } \,\lambda$]3869 & 45 & 136 & & 41 & 66 & & 1.45 + @xmath7 } \,\lambda\lambda$]4073 & & & & 4 & 7 & & 1.40 + @xmath84102 & & & & 17 & 24 & & 1.40 + @xmath94340 & & & & 35 & 46 & & 1.34 + @xmath10 } \,\lambda$]4363 & & & & 16 & 21 & & 1.34 + @xmath114686 & 32 & 41 & & 54 & 60 & & 1.24 + @xmath12 } \,\lambda$]4711 & & & & 9 & 10 & & 1.23 + @xmath12 } \,\lambda$]4740 & & & & 9 & 10 & & 1.22 + @xmath134861 & 100 & 100 & & 100 & 100 & & 1.18 + @xmath10 } \,\lambda$]4959 & 365 & 320 & & 317 & 300 & & 1.15 + @xmath10 } \,\lambda$]5007 & 1245 & 1025 & & 1048 & 965 & & 1.14 + @xmath14 } \,\lambda$]5146 & & & & @xmath158 & @xmath157 & & 1.09 + @xmath16 } \,\lambda\lambda$]5199 & 28 & 18 & & 9 & 7 & & 1.08 + @xmath17 } \,\lambda$]5303 & @xmath1517 & @xmath1510 & & @xmath157 & @xmath155 & & 1.05 + @xmath115411 & & & & 5.9 & 4.4 & & 1.02 + @xmath18 } \,\lambda$]5721 & 23 & 9 & & 8.9 & 6.0 & & 0.95 + @xmath19 } \,\lambda$]5755 & & & & 7.2 & 4.7 & & 0.95 + @xmath205876 & 35 & 12 & & 15 & 9.7 & & 0.92 + @xmath18 } \,\lambda$]6087 & 36 & 11 & & 16 & 9.7 & & 0.89 + @xmath21 } \,\lambda$]6300 & 165 & 42 & & 46 & 26 & & 0.85 + @xmath22 } \,\lambda$]6312 & 20 & 5 & & 11 & 6 & & 0.85 + @xmath21 } \,\lambda$]6364 & 56 & 14 & & 16 & 8.6 & & 0.84 + @xmath23 } \,\lambda$]6374 & 55 & 14 & & @xmath154 & @xmath152 & & 0.84 + @xmath24 } \,\lambda$]6435 & & & & 5.4 & 2.9 & & 0.83 + @xmath19 } \,\lambda$]6548 & 535 & 116 & & 154 & 81 & & 0.81 + @xmath256563 & 1390 & 298 & & 565 & 295 & & 0.81 + @xmath19 } \,\lambda$]6583 & 1620 & 343 & & 457 & 237 & & 0.81 + @xmath206678 & & & & 5.2 & 2.6 & & 0.80 + @xmath7 } \,\lambda$]6716 & 490 & 96 & & 128 & 64 & & 0.79 + @xmath7 } \,\lambda$]6731 & 496 & 96 & & 113 & 57 & & 0.79 + @xmath24 } \,\lambda$]7006 & 10 & 2 & & 14 & 6.4 & & 0.75 + @xmath26 } \,\lambda$]7136 & 148 & 22 & & 61 & 27 & & 0.73 + @xmath27 } \,\lambda$]7291 & @xmath1520 & @xmath153 & & @xmath154 & @xmath152 & & 0.70 + @xmath5 } \,\lambda\lambda$]7319 & 45 & 6 & & 12 & 5.2 & & 0.70 + @xmath5 } \,\lambda\lambda$]7330 & 40 & 5 & & 10 & 4.5 & & 0.70 + @xmath26 } \,\lambda$]7751 & 38 & 4 & & 12 & 4.8 & & 0.64 + @xmath28 } \,\lambda$]7892 & 112 & 11 & & @xmath159 & @xmath153 & & 0.62 + @xmath29 } \,\lambda$]8617 & @xmath1540 & @xmath153 & & @xmath1513 & @xmath154 & & 0.52 + @xmath22 } \,\lambda$]9069 & 980 & 53 & & 220 & 65 & & 0.48 + @xmath22 } \,\lambda$]9531 & 2900 & 134 & & 570 & 155 & & 0.44 + @xmath30 } \,\lambda$]9850 & 70 & 2.9 & & @xmath1533 & @xmath159 & & 0.42 + @xmath31 } \,\lambda$]9913 & 100 & 4.1 & & & & & 0.41 + @xmath3210049 & 90 & 4 & & & & & 0.40 + @xmath29 } \,\lambda$]16435 & 470 & 7.1 & & 30 & 5 & & 0.17 + @xmath33 } \,\lambda$]19629 & 640 & 8 & & @xmath1590 & @xmath1515 & & 0.12 + @xmath3421213 & 400 & 4.7 & & 50 & 8 & & 0.11 + @xmath3521655 & 230 & 2.7 & & 18 & 3 & & 0.11 + flux@xmath36 & 10 & 1350 & & 5 & 36 & & + # 1#2 we chose the circinus galaxy as a benchmark because its emission line spectrum is characterized by remarkably narrow ( @xmath37150 km / s , @xcite ) emission lines which are particularly easy to measure and which indicate relatively low dynamical activity . this last aspect may be used to put tight constrains on the possible contribution of shock excitation which may complicate the modelling of the observed spectrum and the determination of metallicities . this paper presents new optical and infrared spectroscopic data and is structured as follows . observations and data reduction are described in sect . 2 and the results are analyzed in sect . 3 . in sect . 4 we constrain the excitation conditions of the gas and model the observed spectra in terms of photoionization from the agn . the derived chemical abundances are discussed in sect . 5 and in sect . 6 we draw our conclusions . long slit optical spectra were collected at the eso ntt telescope in march 1995 using a dichroic beam splitter feeding both blue ( 1024@xmath38 tek - ccd with 0.37/pix ) and red ( 2048@xmath38 tek - ccd with 0.27/pix ) arms of emmi . simultaneous blue ( 37005000 with @xmath11.7 / pix ) and red ( 500010000 with @xmath11.2 / pix ) spectra were obtained through a 2slit centered on the optical nucleus at two position angles ( cf . [ showslit ] ) . the first was at pa=318@xmath39 , roughly aligned with the [ oiii ] cone axis and including the brightest [ oiii ] emitting knots , while the second was at pa=243@xmath39 and along the low excitation rim visible in the [ sii ] and `` true colour '' images shown in fig . 5 and fig . 10 of @xcite,.oliva ] several exposures were averaged and the total integration times of the spectra at pa=318@xmath39 were 100 , 75 , 25 minutes over the 37005000 , 50007300 and 770010080 wavelength ranges , respectively . the spectra at pa=243@xmath39 covered 3700 - 5000 and 50007300 with a total integration time of 50 minutes per wavelength interval . data were flux calibrated using observations of ltt3218 and reduced with midas using standard procedures . infrared spectra were collected in march 1995 , also at the eso ntt using the spectrometer irspec equipped with a 62x58 sbrc insb array whose pixel size was 2.2along the slit and @xmath15 along the dispersion direction . spectra of [ feii]1.644 , h@xmath40 2.121 , and [ sivi]1.962 were obtained using a 4.4slit at pa=318@xmath39 . each long - slit spectrum consisted of 4 abba cycles ( a = source , b=sky , i.e. a region 300e ) with 2x60 sec integrations per position . the data were reduced using the irspec context of midas which was developed by one of us ( eo ) and which also allows the accurate subtraction of time variable oh sky lines . the same airglow emission was used as wavelength reference ( oliva & origlia @xcite ) and the spectra were corrected for instrumental and atmospheric transmission using spectra of featureless early o stars and flux calibrated using measurements of photometric standard stars . # 1 [ cols= " < , > , > , > , > , > , > , > , > , > , > , > " , ] # 1#2 -16pt the quasi - complete spectra of the nucleus and of knot c are shown in figs [ spec_nuc ] , [ spec_knc ] respectively , and the derived line fluxes are summarized in table [ tab_obs1 ] . dilution by a stellar continuum is particularly strong in the nuclear spectrum where the equivalent width of [ oiii]5007 is only 50 ( cf . table [ tab_obs2 ] ) and a factor of @xmath4110 lower than found in typical seyfert 2 s . the stellar contribution is normally estimated and subtracted using either off nuclear spectra extracted from the same 2d long slit frames , or a suitable combination of spectra of non active galaxies used as templates ( e.g. ho @xcite , ho et al . however , neither of the methods proved particularly useful because line emission contaminates the stellar emission all along the slit , and we could not find any template which accurately reproduces the prominent stellar absorption features typical of quite young stellar populations . the fluxes of weak lines ( @xmath155% of the continuum ) in the nucleus are therefore uncertain and , in a few cases , quite different than those reported in @xcite , the largest discrepancy being for [ ni ] which is a factor of 2 fainter here . the spectrum of knot c has a much more favourable line / continuum ratio and shows many faint lines which are particularly useful for the modelling described in sect . [ photion_model ] . the spatial variation of the most important lines is visualized in figs . [ velcont ] , [ spec_all ] which show contour plots of the continuum subtracted long slit spectra and selected spectral sections of the various knots respectively . the fluxes are summarized in table [ tab_obs2 ] together with the extinctions which were derived from hydrogen recombination lines assuming standard case - b ratios ( hummer & storey @xcite ) . a remarkable result is the large variations of the typical line diagnostic ratios [ oiii]/ , [ oi]/ , [ nii]/ and [ sii]/ which are plotted in fig . [ knotdiag ] and range from values typical of high excitation seyferts ( nucleus , knots a , b , c , d ) , to low excitation liners ( knots h , i ) and normal hii regions ( knots e , l ) . another interesting result is the steep extinction gradient between the regions outside ( knots c , d , h , i ) and those close to the galactic disk ( nucleus and knots a , e , l ) . however , a comparison between the map ( moorwood & oliva @xcite ) , the images ( @xcite ) and the observed br@xmath2 flux from the whole galaxy ( @xcite ) , do not show evidence of more obscured ( @xmath42 ) ionized regions such as those observed in ngc4945 and other starburst galaxies ( e.g. moorwood & oliva @xcite ) . nevertheless , these data can not exclude the presence of deeply embedded ionized gas which is obscured even at 4@xmath3 m ( i.e. @xmath4350 mag ) . particularly interesting is the variation of the line ratios between the adjacent knots c and d. the ratio [ fevii]/ is a factor of 2 larger in knot d than in c , but this most probably reflects variations of the iron gas phase abundance ( see also sect . [ iron ] ) . much more puzzling is the spatial variation of the low excitation lines [ oi ] , [ sii ] , [ nii ] which drop by a factor 1.8 , while the high excitation lines heii , [ oiii ] , [ neiii ] , [ ariii ] together with the [ sii ] density sensitive ratio and [ oiii]/[oii ] vary by much smaller amounts ( cf . table [ tab_obs2 ] ) . this can not therefore be explained by variations of the ionization parameters , which should first of all affect the [ oiii]/[oii ] ratio . a possible explanation for this is discussed in sect . [ model_others ] . temperature and density sensitive line ratios are summarized in table [ tab_diagnostic ] . note that only a few values are available on the nucleus because of the strong stellar continuum which prevents the measurement of faint lines . the relatively large ( oiii ) , slightly lower ( siii ) and much cooler [ nii ] , [ sii ] temperatures are typical of gas photoionized by a `` typical agn '' , i.e. a spectrum characterized by a power law continuum with a super imposed uv bump peaked at @xmath4450100 ev ( e.g. mathews & ferland @xcite ) . as ( oiii ) mainly depends on the average energy of @xmath45 ev ionizing photons , `` bumpy '' spectra , which are quite flat between 13 and 54 ev , yield hot [ oiii ] . the lower ionization species , such as [ nii ] and [ sii ] , mostly form in the partially ionized region , heated by soft x rays , whose temperature can not exceed 10@xmath46 k due to the powerful cooling by collisionally excited ly@xmath2 and 2photon emission . the contrast between the oiii and nii temperatures can be further increased if sub solar metallicities are adopted , because [ oiii ] is a major coolant while [ nii ] only plays a secondary role in the cooling of the partially ionized region . an alternative explanation for the oiii , nii etc . temperature differences is to assume that part of the line emission arises from density bounded clouds . in this case there is no need to adopt a `` bumpy '' agn spectrum and detailed models , assuming a pure power law ionizing continuum , were developed by binette et al . ( 1996 , hereafter @xcite ) . however , it should be kept in mind that ( oiii)@xmath43(nii ) does not necessarily indicate the presence of density bounded clouds . there is a clear trend between and excitation of the species used to determine the density . the [ sii ] red doublet yields densities lower than [ oii ] which is compatible with a single - density cloud for the following reason . if the flux of soft x - rays ( 200 - 500 ev ) is strong enough , then [ sii ] lines are mostly produced in the x - ray heated region where the average hydrogen ionization fraction is quite low ( @xmath370.1 ) . the lines of [ oii ] on the contrary can only be produced in the transition region , where the ionization degree is close to unity , because of the very rapid o h charge exchange reactions . hence ( sii)@xmath15(oii ) most probably indicates the presence of a strong soft - x flux , as one indeed expects to be the case for an agn spectrum . this is also confirmed by the detailed modelling described below . the higher densities in the fully ionized region are new results because the blue [ ariv ] doublet is usually too weak in agn spectra and the fir [ nev ] lines are only accessible with the iso - sws spectrometer ( @xcite ) . the [ ariv ] density of knot c is equal , within the errors , to that derived by [ oii ] thus indicating that no large variations of densities are present in this cloud . # 1 lcc ratio & value & diagnostic + + [ oii ] 3727/7325 & 25@xmath06 & 300@xmath15@xmath156000 + [ sii ] 6731/6716 & [email protected] & 400@xmath15@xmath15800 + [ nev]@xmath36 14.3@xmath3m/24.3@xmath3 m & [email protected] & 2000@xmath15@xmath1512000 + [ siii ] 9531/6312 & @xmath4725 & @xmath3714000 + + [ oii ] 3727/7325 & 14@xmath04 & 1400@xmath15@xmath1511000 + [ sii ] 6731/6716 & [email protected] & 200@xmath15@xmath15500 + [ ariv ] 4711/4740 & [email protected] & 1500@xmath15@xmath1510000 + [ nii ] 6583/5755 & 50@xmath013 & 9000@xmath15@xmath1512500 + [ oiii ] 5007/4363 & 46@xmath09 & 14000@xmath15@xmath1518000 + [ sii ] 6731/4073 & @xmath378 & @xmath3712000 + [ siii ] 9531/6312 & 26@xmath07 & 11500@xmath15@xmath1516000 + @xmath36 [ nev ] fir lines from m96 several photoionization models for the circinus galaxy have been discussed in the literature . these used line intensities and ratios measured from the central regions of the galaxy and were mostly directed to constraining the intrinsic shape of agn photoionizing continuum . quite different conclusions were drawn by @xcite , who found evidence for a prominent uv bump centered at @xmath170 ev , and binette et al . ( 1997 , hereafter @xcite ) who showed that equally good ( or bad ) results could be obtained using a combination of radiation and density bounded clouds photoionized by a pure power law spectrum . in all cases the metal abundances were assumed to be solar and no attempt was made to constrain metallicities . here we mainly concentrate on knot c , an extranuclear cloud whose rich emission line spectrum and simple geometry ( a plane parallel region ) are better suited for deriving physical parameters from photoionization modelling . we first analyze the observational evidence against shock excitation and then describe in some detail the new modelling procedure which is primarily aimed at determining the gas metal abundances , ( cf . sects . [ details_photion ] and [ knc_abund ] ) . we also analyze the problems to reproduce the observed [ fevii]/[feii ] and [ nii]/[ni ] ratios ( sects . [ iron ] , [ nii_ni ] ) , discuss the role of dust and reanalyze several crucial aspects of the line emission from the brightest regions closest to the nucleus . strong evidence against shock excitation comes from the very low strength of [ feii ] which indicates that iron is strongly underabundant , i.e. hidden in grains , within the partially ionized region ( cf . figs . [ figz ] , [ figznoch ] ) . shocks are very effective in destroying dust grains , and velocities as low as 100 km / s are enough to return most of the iron to the gas phase ( e.g. draine & mckee @xcite ) . this is confirmed by observations of nearby herbig haro objects ( e.g. beck - winchatz et al . @xcite ) and supernova remnants whose near ir spectra are dominated by prominent [ feii ] ir lines ( e.g. oliva et al . although the low fe@xmath48 gas phase abundance could , in principle , be compatible with a very slow shock ( @xmath3750 km / s ) , this falls short by orders of magnitude in producing high excitation species . another argument comes from the heii/ and heii / hei ratios which are a factor @xmath432 larger than those predicted by shocks models with velocities @xmath49 km / s ( dopita & sutherland @xcite ) . it should be noted that , although stronger heii could probably be obtained by increasing the shock speed to @xmath411000 km / s , these velocities are incompatible with the observed line profiles . more generally , the observed line widths ( @xmath15150 km / s , @xcite ) are difficult to reconcile with the fact that only shocks faster than 300 km / s can produce prominent high excitation lines such as [ oiii ] ( dopita & sutherland @xcite ) . this argument becomes even stronger when interpreting the nuclear spectra where the highest excitation lines , [ fex , xi ] , do not show any evidence of large scale motions at velocities @xmath43200 km / s . ( cf . @xcite and work in preparation ) . finally , it is worth mentioning that detailed shock + photoionization composite models recently developed by contini et al . ( @xcite ) suggest that shock excited gas does not contribute significantly to any of the optical / ir lines from circinus . we constructed a grid of models covering a wide range of ionization parameters ( @xmath50 ) , densities ( @xmath51 ) , metallicities ( @xmath52 , @xmath53 ) and shape of the ionizing continuum which we parameterized as a combination of a power law extending from 1 ev to 10 kev with index @xmath54 and a black body with @xmath55 , the relative fraction of the two components being another free parameter . all the parameters were varied randomly and about 27,000 models were produced using cloudy ( ferland @xcite ) which we slightly modified to include the possibility of varying the n h charge exchange rate . the line intensities were also computed using the temperature and ionization structure from cloudy and an in house data base of atomic parameters . the agreement with the cloudy output was good for all the `` standard '' species while large discrepancies were only found for coronal species ( e.g. [ fex ] ) whose collision strengths are still very uncertain and much debated . out of this large grid we selected about 400 models whose [ oiii]/ , [ oiii]/[oii]/[oi ] , [ arv]/[ariv]/[ariii ] , [ siii]/[sii ] , [ oiii]5007/4363 , [ oii]3727/7325 , [ sii]6731/6716 , [ ariv]4711/4740 line ratios were reasonably close to those observed in knot c. these were used as the starting point for computing the `` good models '' , with adjusted values of relative metal abundances , which minimized the differences between predicted and observed line ratios . note that to reproduce both [ fevii ] and [ feii ] we were in all cases forced to vary the iron gas phase abundance between the fully and partially ionized regions ( cf . also sect . [ iron ] ) . the results of the best model are summarized in table [ tab_modelknc ] and discussed in sect . [ detail_knc_model ] . note the large discrepancy for [ ni ] which is overpredicted by a factor of @xmath415 , this problem is discussed in sect . [ nii_ni ] . the most important results are the abundance histograms shown in figs . [ figz ] , [ figznoch ] which were constructed by including all the `` good '' models with @xmath56 . although the choice of the cutoff is arbitrary , it should be noticed that variations of this parameter do not alter the mean values , but only influence the shape of the distributions whose widths roughly double if the @xmath57 cutoff is increased to values as large as 30 . the large grid described above was also organized to have at least 4 models with the same photon flux , continuum shape and abundances but different densities , and these were used as starting points to construct multi - density models . we simulated 2density clouds by coupling all available models with different weights and selected about 300 of them , which were used as starting points to compute the `` good models '' following the same procedure adopted for the single - density case . we also simulated clouds with @xmath583 density components , but these results are not included here because this complication had virtually no effect on the quality of the fit . the most important result is that single and multi - density models are equally good ( or bad ) in reproducing the observed line ratios . obviously , this does not necessarily imply that knot c is a single - density cloud , but rather indicates that the stratifications which probably exist have little effect on the available density sensitive line ratios . we constructed a grid of about 10,000 models photoionized by a power law agn continuum with index @xmath59 , which turns into @xmath60 beyond 500 ev , and covering the same range of physical parameters as the radiation bounded clouds described above . the column density of the radiation bounded component was always large enough to reach a temperature @xmath153000 k at its outer edge . for each model we also computed line intensities from a density bounded cloud which we defined as a layer with thickness @xmath61 equal to 5% of the nominal radius of the hii strmgren sphere , numerically ( @xmath62= filling factor ) @xmath63 the relative contribution of density and radiation bounded regions was set by imposing heii4686/=0.6 this approach is somewhat similar to that adopted by @xcite apart from the following details . the radiation bounded component here is formed by high density gas , a condition required to match the observed [ ariv]4711/4740 , and is concentrated within the projected size of knot c , i.e. @xmath1580 pc . in practice , the two components have the same densities and see un filtered radiation from the agn . the `` good models '' were optimized , and the best abundances derived using the the same procedure depicted above . worth mentioning is that most of the `` good models '' have spectral slopes in the range 1.31.5 which are in good agreement with those used by @xcite . the most important result is that these mixed models give similar , though slightly poorer ( cf . [ detail_knc_model ] ) results than radiation bounded clouds . dust can modify the ionization and temperature structure because it competes with gas in absorbing the uv photons , and because it hides refractory elements such as iron . given the low ionization parameters , however , the first effect is negligible , i.e. the ionization structure of the fully ionized region of knot c is not affected by dust although this plays an important role in modifying the heating cooling balance of the partially ionized region heated by x rays . cooling : the refractory species hidden in the grains can not contribute to the line cooling and this effect is correctly computed by cloudy . for example , depleting fe on grains produces higher gas temperature and stronger [ oi ] , [ sii ] etc . lines , because it suppresses the near ir lines of [ feii ] which are among the major coolants for quasi solar fe / o gas phase abundances . heating : the metals hidden in the dust still contribute to the heating of the gas because the x rays have energies much larger than the binding energy of the grains , and can not therefore recognize metals in dust from those in the gas phase . the major problem is that cloudy ( and probably other photoionization models ) does not include the x ray heating from the grain metals and therefore underestimates the temperature of the partially ionized region and hence may predict weaker fluxes of [ oi ] , [ sii ] and other low excitation lines . therefore , the models for knot c , having most of iron depleted on grains , are not fully self consistent and most probably require a too high flux of x rays to reproduce e.g. [ oi]/[oiii ] . this may imply that the `` true '' models should have somewhat softer spectra than those of fig . [ agn_cont ] . table [ tab_modelknc ] lists the results of the best photoionization models , selected from the several hundred which provided a reasonable fit to the spectrum of knot c. note that the results of multi - density clouds ( sect . [ details_multidens ] ) are not included because they are virtually indistinguishable from those of single - density models . [ agn_cont ] shows the agn continuum adopted for the `` best models '' of table [ tab_modelknc ] . radiation bounded models require a prominent uv bump centered ( in @xmath64 ) at about 100 ev . this is slightly bluer than that found by @xcite and somewhat harder than model spectra of more luminous accretion disks ( cf . e.g. laor @xcite ) , but in qualitative agreement with the predicted dependence of spectral shape with agn luminosity , i.e. that lower luminosity accretion disks should have harder spectra ( netzer et al . @xcite ) . # 1 # 1#2to 100pt#1 # 2 lccc adopted parameters & model 1 & model 2 & model 3 + element & + helium & 11.1 & 11.1 & 11.1 + nitrogen & 8.05 & 8.20 & 8.05 + oxygen & 8.17 & 8.27 & 8.17 + neon & 7.27 & 7.37 & 7.27 + sulphur & 6.98 & 6.90 & 6.75 + argon & 6.38 & 6.45 & 6.20 + silicon & 6.85 & 6.95 & 6.85 + fe low - ion@xmath36 & 5.80 & 5.80 & 5.80 + fe high - ion@xmath65 & 6.90 & 7.00 & 7.00 + gas density ( @xmath66 ) & 2000 & 3000 & 2000 + ionizing continuum@xmath67 & uv - bump & @xmath68 & @xmath69 + ionization parameter@xmath70 ( @xmath71 ) & 3.5 @xmath72 & 4.7 @xmath72 & 2.0 @xmath72 + fraction radiation bounded@xmath73 & 100% & 4% & 3% + @xmath74()/@xmath75()@xmath76 & 100% & 6% & 6% + & model 1 & model 2 & model 3 + & 0.60 & 0.60 & 0.60 + & 6.8 & 5.8 & 6.3 + & 0.83 & 0.79 & 0.77 + & 8.0 & 6.5 & 4.7 + & 41 & 57 & 39 + & 10 & 9.8 & 9.3 + & 31 & 43 & 42 + & 0.12 & .057 & 0.12 + & .027 & .024 & .028 + & 14 & 12 & 13 + & 0.68 & 0.70 & 0.73 + & 2.2 & 1.5 & 0.12 + & 0.20 & 0.20 & 0.19 + & 0.95 & 0.88 & 0.91 + & 3.0 & 2.4 & 2.6 + & 20 & 22 & 21 + & .089 & .098 & .092 + & 0.56 & 0.57 & 0.25 + & 0.17 & 0.17 & .032 + & 0.88 & 0.96 & 0.88 + & .031 & .030 & .005 + & 0.70 & 1.2 & 4.1 + & .045 & .030 & .005 + & 1.6 & 1.8 & 1.8 + # 1#2 the amplitude of the uv bump could be significantly decreased by relaxing the assumption on the gas density distribution and , with a properly tuned combination of density and radiation bounded clouds , one may probably get a similarly good fit with a power law . it is nevertheless instructive to analyze why our mixed models provide a worse fit to the data and , in particular , are unable to simultaneously reproduce the oii / oiii balance and the [ arv]/[ariii ] and [ fevii]/[fevi ] ratios ( cf . last two columns of table [ tab_modelknc ] ) . model # 2 has a high ionization parameter and correctly predicts high excitation lines but underestimates all [ oii ] lines while model # 3 , with a lower value of u , correctly reproduces the [ oii]/[oiii ] ratio but predicts very faint high excitation lines . the main reason for the above difference is that , for a given gas density , the [ oii]/[oiii ] ratio is a measure of the flux of soft ionizing photons ( @xmath77=1354 ev ) , while [ arv]/[ariii ] and [ fevii]/[fevi ] depend on the flux of harder radiation ( @xmath78 80 ev ) . the spectrum of knot c is characterized by a quite large [ oii]/[oiii ] ratio , which points towards low fluxes of soft photons , and strong high excitation lines , which require large fluxes of hard photons . these somewhat contradictory requirements can be easily satisfied by a spectrum which steeply rises beyond the lyman edge , and peaks at @xmath44100 ev , i.e. a spectrum similar to that of model # 1 . + independently on the detailed results of the models , the following arguments indicate that density bounded clouds may indeed play an important role . + ionizing spectra with pronounced uv bumps tend to produce too high [ oiii ] temperatures and model # 1 is indeed quite hot , though still compatible ( within 1.5@xmath79 ) with the somewhat noisy measurement of [ oiii]4363 ( cf . [ spec_knc ] and tab . [ tab_modelknc ] ) . curiously , though , the result of model # 1 is the opposite of that obtained by most previous models in the literature which failed to produce hot enough [ oiii ] . therefore , the classical `` [ oiii ] temperature problem '' may just reflect the fact that past models were mostly biased toward high metallicities and rather flat ( i.e. without strong bumps ) continua . + the flux from knot c is only @xmath15% of that expected if the gas absorbed all the ionizing photon impinging on the 2 x 2(40 x 40 pc@xmath38 ) geometrical cross section of the cloud . such a small `` effective covering factor '' could , in principle , be obtained by assuming a suitable distribution of radiation bounded , geometrically thin clouds or filaments . however , density bounded clouds seem to provide a more self consistent interpretation because the measured value is remarkably close to the 6% predicted by models # 2 and # 3 ( cf . note @xmath62 of tab . [ tab_modelknc ] ) . + the variation of line ratios between the different knots can not be explained by radiation bounded clouds illuminated by the same ionizing continuum , but require e.g. intrinsic beaming of the agn continuum and/or filtering by matter bounded clouds closer to the nucleus . mixed models could give a more natural explanation to the spatial variation of line ratios ( see also @xcite ) . # 1 # 1#2to 155pt#1 ( # 2 ) lr & @xmath80 + effective covering factor of knot c@xmath65 & @xmath415% + & @xmath81 + & @xmath82 + fraction of @xmath83(h)@xmath84 intercepted by gas@xmath73 & @xmath411% + & @xmath85 + & @xmath86 + # 1#2 the ionizing photon flux seen by knot c is constrained by the @xmath71sensitive and density sensitive line ratios which yield @xmath87 this can be translated into @xmath83(h)@xmath84 , the total number rate of ionizing photons from the agn , once the angular distribution of the uv ionizing radiation is known . assuming that it arises from a geometrically thin , optically thick accretion disk , one finds @xmath88 ( e.g. laor & netzer @xcite ) and @xmath89 where @xmath90 is the projected distance of knot c i.e. 15.5or @xmath91=300/@xmath92 pc from the nucleus , @xmath93 being the inclination angle of the cone relative to the plane of the sky . detailed kinematical studies indicate @xmath94 ( elmouttie et al . @xcite ) . the agn luminosity in the ionizing continuum is therefore @xmath95 where @xmath96 is the average photon energy which is here assumed to be @xmath150 ev . the ionizing luminosity is therefore very large but compatible , within the errors , with the observed fir luminosity @xmath97 ( siebenmorgen et al . this implies that the agn emits most of its energy in the ionizing continuum or , equivalently , that the agn intrinsic spectrum has a prominent peak or bump in the ionizing uv , and much weaker emission at lower energies . this is in good agreement with computed models of accretion disks which also predict that low luminosity agns , such as circinus , should be characterized by a quite hard ionizing continuum ( cf . laor @xcite , netzer et al . @xcite ) . the global properties of the agn are summarized in table [ tab_globalagn ] which also includes a comparison between @xmath98 and the observed recombination rate , based on iso observations of the br@xmath2 h recombination line at 4.05 @xmath3 m ( cf . the difference is remarkably large , with only @xmath11% of the agn ionizing photons being accounted for by emission from `` normal '' ionized gas . this indicates that the bulk of the lyman continuum radiation either goes into ionizing regions which are obscured even at 4 @xmath3 m ( i.e. @xmath4350 mag ) , or is directly absorbed by grains in dusty clouds lying very close to the agn . the observed [ fevii]6087/[feii]16435@xmath471 ratio can not be explained using the same iron gas phase abundance in the heiii strmgren sphere , where fevii is formed , and in the partially ionized region , where iron is predominantly feii due to the rapid charge exchange recombination reactions with neutral hydrogen . it should be noted that this is a fundamental problem unrelated to the details of the photoionization models and primarily reflects the fact that the [ fevii ] line has a very small collision strength ( @xmath99 ) which is factor of about 10 lower than the [ feii ] transition . a fe@xmath100/fe@xmath101 integrated relative abundance is thus required to reproduce the observed line ratio . this number is incompatible with the relative sizes of the heiii and partially ionized regions which are constrained by e.g. the heii and [ oi ] lines . it is also interesting to note that this problem is even exacerbated if the shock models of dopita & sutherland ( @xcite ) are adopted because these predict fe@xmath100/fe@xmath48 ratios much smaller than the photoionization models described above . a possible solution could be to advocate that the [ fevii ] collision strengths are underestimated by a factor of @xmath4110 which would reconcile the observed [ fevii ] and [ feii ] intensities with a low ( 2% of solar ) iron abundance ( cf . figs . [ figz ] , [ figznoch ] ) . however , although the collision strengths of coronal fe lines are known to be very uncertain , and vary by factors @xmath4310 depending on whether resonances are included in the computation ( cf . sect . 5 of oliva @xcite ) , the available collision strengths for [ fevii ] increase by only 20% between the old dw computations of nussbaumer & storey ( @xcite ) and the newer r mat ( i.e. including resonances ) values of keenan & norrington ( @xcite ) . more detailed modelling of the [ fevii ] lines and spectroscopic studies of nearby astrophysical laboratories ( e.g. high excitation planetary nebulae ) are required to clarify this issue . the alternative possibility is to assume that the iron depletion is much larger in the partially than in the fully ionized region , as already suggested by netzer & laor ( @xcite ) and ferguson et al . ( @xcite ) . however , we could not find any simple explanation for such a stratification in knot c which lies far from the agn and receives a relatively modest flux of hard uv photons . therefore , fe bearing dust can not be photo evaporated and the only mechanism to destroy grains is sputtering . the slow shock produced by the ionization front which propagated outward when the agn turned on was too slow ( @xmath10240 km / s ) to effectively return fe to the gas phase ( e.g. draine & mckee @xcite ) . faster shocks ( @xmath58100 km / s ) are a natural and efficient source of sputtering but can not explain the observations because they are expected , and observed , to emit prominent [ feii ] lines from the dust free recombining gas . a possibility to overcome this problem is a combination of shocks and photoionization where e.g. the dust free gas processed by the shock is kept fully ionized by the agn radiation . however , this situation is short lived because , after a few thousand years , the gas piled up behind the shock will eventually reach a column density high enough to become radiation bounded , and shield the recombining gas which will therefore show up in [ feii ] . the photoionization models of knot c systematically overestimate by large factors ( @xmath436 ) the strength of [ ni ] relative to [ nii ] ( cf . table [ tab_modelknc ] and fig . [ figz ] ) . although [ ni]5200 has a quite low critical density ( @xmath11500 @xmath103 ) , this error can not be attributed to the presence of higher density clouds because these would also effect the [ sii ] density sensitive ratio . in other words , multi - density models which correctly reproduce the high [ nii]/[ni ] ratio inevitably predict too large [ sii]6731/6716 ratios . also , we can exclude observational errors because , in knot c , the [ ni ] doublet has an equivalent width of about 2.5 and is only marginally affected by blending with neighbouruing stellar absorption lines . a possible solution is to argue that the rate coefficients for n h charge exchange are overestimated , as already suggested by e.g. ferland & netzer ( @xcite ) . in the partially ionized region , nii is mostly neutralized via charge exchanges with h@xmath104 and adopting lower charge exchange efficiencies yield larger [ nii]/[ni ] ratios . this is evident in fig . [ nitrogen ] where this ratio is plotted as a function of the assumed value of @xmath105(n ) , the charge exchange rate coefficient . assuming a n h charge exchange a factor of @xmath4130 lower than presently adopted yields the correct [ nii]/[ni ] ratio . noticeably , the problem we find here is exactly the opposite of what reported by stasinska ( @xcite ) whose models systematically underpredicted the [ ni]/[nii ] ratio in objects with low [ oi]/[oii ] . a puzzling aspect of the line emission from regions very close to the nucleus is the simultaneous presence of high ( e.g. [ fexi ] ) and low ( e.g. [ sii ] , [ oi ] ) ionization species . more specifically , the images of @xcite clearly show that [ sii ] peaks at a distance of only @xmath10.5 or 10 pc from the nucleus ( cf . fig . 9 of @xcite ) . this result is incompatible with the standard idea according to which low excitation lines are produced in clouds with low ionization parameters . numerically , combining the ionizing continuum inferred from the spectrum of knot c with @xmath106=1.2@xmath107 @xmath103 , the highest density compatible with the fir [ nev ] doublet ( table [ diagnostic ] ) , yields @xmath108 at @xmath109=10 pc from the agn , an ionization parameter far too large for the production of low excitation species . not surprisingly therefore , all the models so far developed for the high excitation lines ( @xcite , @xcite , @xcite ) predict that [ sii ] should peak much farther out and have a much lower surface brightness than that observed . a simple and indeed natural solution is to assume that the low excitation lines are formed in dusty clouds . at these large ionization parameters dust dominates the absorption of uv ionizing photons and , therefore , quenches the heiii and hii strmgren spheres . consequently , the x ray dominated partially ionized region starts at much smaller radii , and is also slightly hotter than in the dust free model ( cf . fig [ modelnuc ] ) . therefore , dusty clouds have a [ sii ] peak much closer to the nucleus and a @xmath4310 times higher surface brightness than dust free clouds . it should be noted , however , that the total luminosity of low excitation optical lines is similar in the two cases , fundamentally because the available luminosity of soft x rays is the same . although the combined emission of dusty and dust free clouds can account for the observed emission of low ionization and coronal species , it falls short by large factors in producing [ oiv ] and other relatively low ionization species which form within the heiii sphere this is an intrinsic limitation of `` compact models '' such as those of fig . [ modelnuc ] , and can be understood as follows . compact models are characterized by large `` ionization parameters '' ( cf . sect . 3 of oliva @xcite for a critical discussion on this parameter ) and therefore have large fluxes of oiv ionizing photons which keep oxygen in higher ionization stages ( mostly ovi and ovii ) at all radii inside the heiii strmgren sphere . outside the heiii region , on the contrary , oxygen can not be ionized above oiii because most of the oiii ionizing photons have already been absorbed by heii . therefore , oiv can only exist in a very narrow range of radii , just at the edge of the heiii sphere , and its relative abundance is therefore very low . in practice , we found it impossible to construct a single model which simultaneously produces a compact coronal line region , such as that observed in [ fexi ] ( @xcite ) , and which comes anywhere close to the [ oiv]25.9/[oiii][email protected] observed ratio . we did indeed construct many thousands of randomly generated dusty and dust free models , and attempted an approach similar to that used for knot c ( sect . [ details_photion ] ) but , in no case , could we find a model which satisfies these contradictory constraints . it should also be noted that @xcite independently come to a similar conclusion . the main conclusion therefore is that , regardless of the details of the models , the nuclear spectrum and line spatial distribution can only be modeled by adding a third `` diffuse '' component ( e.g. with a lower filling factor ) to the dusty and dust free clouds discussed above and depicted in fig . [ modelnuc ] . given the large number of free parameters , we abandoned the idea of using photoionization models to constrain abundances and other physical properties of the gas , we made some attempt to verify that a mixture of clouds exposed to the same continuum , and with the same abundances as model # 1 of knot c ( table [ tab_modelknc ] and fig . [ agn_cont ] ) could reasonably well reproduce the observed properties of the nuclear spectrum . however , the results are not too encouraging and , apart from the much improved [ sii ] surface brightness and [ oiv]/[oiii ] ratio , the solutions are not significantly better than those already discussed by @xcite and @xcite , and are not therefore discussed here . an analysis similar to that used for knot c was also applied to the other extra nuclear knots using the more limited number of lines available in their spectra . the results can be summarized as follows . the abundances derived in the seyfert - type knots a , b , d , g , f ( cf . fig . [ knotdiag ] ) are similar those found in knot c but affected by much larger errors because of the more limited numbers of lines available for the analysis . in particular , the density sensitive [ ariv ] doublet and the u sensitive [ arv ] line is not detected in any of these knots , and the reddening correction for [ oii]@xmath110 could be very uncertain in the high extinction regions ( cf . note @xmath111 of table [ tab_obs2 ] ) . we also attempted to verify if the observed line ratios in these knots could be explained as photoionization by the same agn continuum seen by knot c but could not find any satisfactory solution using radiation bounded clouds exposed to the same continuum . adding matter bounded clouds alleviates the problem ( as already stressed by @xcite ) but requires an _ ad hoc _ choice of their photoelectric opacities , i.e. the radius at which the ionization structure is cut . in particular , explaining the drop of low excitation lines between the adjacent knots c and d ( cf . sect . [ spatial_line_distribution ] ) requires matter bounded clouds cut at about 1.2@xmath112 the hii strmgren radius , the exact position of the cut depending on the assumed shape of the agn continuum . another parameter affecting the ratio of low to high excitation lines ( e.g. [ oi]/[oiii ] ) is the iron gas phase abundance which influences the cooling of the partially ionized region ( cf . [ role_of_dust ] ) . if iron is more abundant in knot d , as indicated by its stronger [ fevii ] line emission ( cf . table [ tab_obs2 ] and sect . [ spatial_line_distribution ] ) , than [ oi ] , [ sii ] , [ nii ] could be depressed by the increased [ feii ] cooling . the abundances derived for the liner like knots ( h , i ) are very uncertain ( @xmath01.3 dex at least ) . their spectra are not compatible with illumination from the same continuum seen by knot c but require a harder ( i.e. more x rays relative to 1380 ev photons ) spectrum which could be in principle obtained by filtering the agn continuum through an absorber with a carefully tuned photoelectric opacity . alternatively , the weak ( in surface brightness ) spectrum of these low excitation knots could be explained by shock excitation , in which case one expects [ feii]16435/@xmath11 and a factor @xmath4310 larger than in the case of pure photoionization . finally , the oxygen lines in the highly reddened hii like knots ( e , l ) are too weak to allow any reliable abundance analysis . deriving abundances of agn clouds from photoionization models has generally been considered to be unreliable because the shape of the agn ionizing continuum and the density distribution of the emitting regions ( clouds ) are both basically unknown . the more typical approach therefore has been to assume a metallicity and use photoionization models to constrain the agn spectrum and/or the gas density distribution , or just to demonstrate that the gas is photoionized . although explicit statements concerning the nitrogen abundance are often found in the literature ( e.g. storchi - bergmann & pastoriza @xcite , simpson & ward @xcite ) these are based on a very limited choice of photoionization model parameters and in most cases find oxygen abundances close to solar , in disagreement with what is derived here . the only other piece of work which covers a model parameter range comparable to that presented here is that by komossa & schulz ( @xcite ) who analyzes a much more limited numbers of lines , e.g. do not include [ ariv , v ] , in a large sample of seyferts . they find that , on average , oxygen is underabundant by a factor of @xmath412 and that the n / o ratio is only a factor of 1.52.0 above the solar value . we believe that the analysis presented here leads to more reliable metallicity estimates . the results presented here indicate that , in spite of the above uncertainties , reliable metallicities can indeed be derived from spectra including a large enough number of lines , such as that of knot c. our method has been described in sect . [ details_photion ] and , in short , is based on the computation of a large number of photoionization models with a minimal personal bias and preconceived ideas on the agn ionizing spectrum and the gas density distribution . in particular , we also considered mixed models with combinations of density and radiation bounded clouds ( @xcite ) as well as models with multiple density components . we spanned a very wide range of model parameters which were varied randomly , and selected the relatively few ( about 200 ) good models which came closest to reproducing the observed line ratios . the main results are summarized in figs . [ figz ] , [ figznoch ] which show the distribution of element abundances required to reproduce the observed line ratios . the most remarkable feature is that the metal abundances are quite well constrained in spite of the very different assumptions made for the gas density distribution and shape of the agn continua . in other words , models with very different abundances fail to match the observed lines ratios in knot c regardless of the agn spectral shape and/or gas density distribution assumed . another encouraging result is that lines from different ionization stages yield similar abundances which simply reflects the fact that the models reproduce the observed line ratios reasonably well . there are however remarkable exceptions , such as the [ nii]/[ni ] and [ feii]/[fevii ] ratios which are both predicted too high . possible explanations for these differences have been discussed above ( sects . [ iron ] , [ nii_ni ] ) . we stress here , however , that the uncertainties on [ ni ] have little effects on the derived nitrogen abundance because n@xmath48 is the most abundant ion within the partially ionized region . therefore , the best n abundance decreases by only a factor @xmath11.3 once the n@xmath48/n@xmath104 is increased to match the observed [ nii]/[ni ] ratio ( cf . fig.[figznoch ] ) . note also that the he / h abundance is only poorly constrained by the models , and although models with he / [email protected] are somewhat favoured , no firm conclusion about he overabundance can be drawn from the data . + # 1 lccccc element & 12+log(x / h ) & [ x / h ] & & + & & & & obs & model + nitrogen@xmath113 & 8.0 & @xmath114 & & @xmath115 & @xmath116 + oxygen & 8.2 & @xmath117 & & & + neon & 7.3 & @xmath118 & & @xmath119 & @xmath120 + sulphur & 6.9 & @xmath121 & & @xmath122 & n.i . + argon & 6.4 & @xmath120 & & @xmath123 & n.i . + stellar metallicity & & @xmath117 & & & + # 1#2 the derived abundances are summarized in table [ tab_z ] where the most striking result is the large overabundance of nitrogen relative to oxygen , + 0.7 dex above the solar value , whose implications are discussed below . an independent estimate of metallicity can be derived from the measured equivalent widths of co stellar absorption features , using the new metallicity scale proposed and successfully applied to young lmc / smc clusters by oliva & origlia ( @xcite ) . in short , the method is based on the strength of the co(6,3 ) band head at 1.62 @xmath3 m whose behaviour with metallicity is modelled using synthetic spectra of red supergiants . the equivalent width of the stellar co lines from the central 100 x 100 pc@xmath38 of circinus are reported in table 2 of oliva et al . ( @xcite ) and yield an average metallicity of @xmath124 , a value remarkably close to the oxygen abundance derived above ( cf . table [ tab_z ] ) . the nitrogen overabundance is of particular interest in view of its possible relationship with the ( circum)nuclear starburst and n enrichment from material processed through the cno cycle . according to chemical evolutionary models of starburst events , the n / o relative abundance reaches a maximum value of [ n / o]@xmath125 ( i.e. 4 times the solar value ) at about @xmath126 yr and remains roughly constant for several @xmath127 years ( cf . fig . 4 of matteucci & padovani @xcite ) . the nitrogen overabundance mostly reflects the effect of the winds from he burning red supergiants whose surface composition is strongly n enriched by gas dredged up from the shell where hydrogen was burned through the cno cycle . the amount and temporal evolution of the n / o abundance depends on model details , e.g. the shape of the imf and the duration of the starburst , as well as on poorly known parameters such as the efficiency of the dredge up and the contribution of primary n production by massive stars ( e.g. matteucci @xcite ) . it is however encouraging to find that the observed n / o abundance ( table [ tab_z ] ) is very close to that predicted at a time which is compatible with the age of the starburst in circinus ( cf . fig . 9 of maiolino et al . it should also be noticed that the observed absolute abundances are about an order of magnitude lower than the model predicted values , but this can be readily explained if the starburst transformed only @xmath128 of the available gas into stars , in which case the chemical enrichment was diluted by a similar factor . this hypothesis is in good agreement with the fact that circinus is a very gas rich galaxy ( e.g. freeman et al . @xcite ) . in short , the observed n / o overabundance is fully compatible with what expected for a quite old ( several @xmath127 yr ) starburst . the [ nii]/ and other line ratios measured in the cone of circinus are similar to those observed in many others sy2 s and several observational results indicate that relatively old starburst events are common in type 2 seyferts ( e.g. maiolino et al . this may therefore indicate that nitrogen is typically overabundant in sy2 s due to enrichment by the starburst associated with the agns . this tantalizing conclusion should be however verified by detailed spectroscopic studies and analysis of a sufficiently large number of objects . by modelling the rich spectrum of an extranuclear cloud in the ionization cone of the circinus galaxy we have found that metal abundances are remarkably well constrained , regardless of the assumptions made on the shape of the ionizing continuum and gas distribution . this new result may open a new and interesting field of research using photoionization models to derive metallicities in agns which could in turn be related to the star formation activity in the recent past , i.e. old nuclear starbursts . in the case of circinus , the large n / o overabundance found here is fully compatible with what expected from chemical evolution models of starbursts ( sect . [ n_and_starburst ] ) . much less encouraging are the results on the agn ionizing continuum whose shape can not be constrained by the observed line ratios but depends on the assumed gas density distribution . within the limits of the model parameters spanned here we somewhat favour an agn spectrum with a `` uv bump '' but can not exclude that , with a different and better tuned combination of density and radiation bounded clouds , one could achieve similarly good fits with a power law agn continuum ( sect . [ detail_knc_model ] ) . we also found that photoionization models can not reproduce the observed [ fevii]/[feii ] and [ nii]/[ni ] ratios and argued that these may reflect errors in the collision strengths for [ fevii ] and rate coefficient of n h charge exchange reactions . it should be noted , however , that the problem [ nii]/[ni ] does not significantly influence the derived nitrogen abundance . we thank roberto maiolino and francesca matteucci for useful discussions , luis ho and thaisa storchi bergman for providing information on template stellar spectra , gary ferland for making cloudy available to the community , and an anonymous referee for many comments and criticisms which have been fundamental for improving the quality of the paper .

the invariant approach is a powerful method for studyingviolation for specific lagrangians . the method is particularly useful for dealing with discrete family symmetries . we focus on theproperties of unbroken invariant lagrangians with yukawa - like terms , which proves to be a rich framework , with distinct aspects of , making it an ideal group to investigate with the invariant approach . we classify lagrangians depending on the number of fields transforming as irreducible triplet representations of . for each case , we construct-odd weak basis invariants and use them to discuss the respectiveproperties . we find thatviolation is sensitive to the number and type of representations . * invariant approach to in unbroken * + gustavo c. branco , ivo de medeiros varzielas , stephen f. king + _ centro de fsica terica de partculas - cftp , _ _ instituto superior tcnico - ist , universidade de lisboa - ul , _ _ avenida rovisco pais , 1049 - 001 lisboa , portugal _ _ school of physics and astronomy , university of southampton , _ _ so17 1bj southampton , united kingdom _

1505.06165

the origin and nature of @xmath0and its violation remains a mystery both within and beyond the standard model ( sm ) . in addressing the question of @xmath0it was observed some time ago that phases which appear in the yukawa matrices for example are not robust indicators of @xmath0violation since their appearance is dependent on the choice of basis . on the other hand , physical @xmath0violating observables only depend on particular combinations of yukawa matrices which are invariant under different choices of basis . such weak basis invariants , which have the property that they are zero if @xmath0is conserved and non - zero if @xmath0is violated therefore provide unambiguous signals of @xmath0violation which are closely related to experimentally measurable quantities . the use of such @xmath0-odd weak - basis invariants ( cpis ) , rather than particular phases in a given basis , is generally referred to as the invariant approach ( ia ) to @xmath0violation . in the ia to @xmath0violation @xcite , one starts by separating the full lagrangian of the theory in two parts , one denoted @xmath4 that is known to conserve @xmath0 , typically the kinetic terms and pure gauge interactions @xcite ,- like transformations that include a family symmetry transformation was introduced in @xcite , in an attempt to obtain a connection between quark masses and mixing angles . ] and the remaining lagrangian , denoted @xmath5 . the crucial point is that @xmath4 allows for many different @xmath0transformations and as a result , @xmath0is violated if and only if none of these @xmath0transformations leaves @xmath5 invariant . in the case of the sm , @xmath4 includes the gauge interactions and the kinetic energy terms , while the relevant components of @xmath5 are the yukawa interactions . using the ia , one can readily derive @xcite some specific conditions that the yukawa couplings have to satisfy in order to have @xmath0invariance . it is well known that the yukawa couplings in the sm have a large redundancy which results from the freedom that one has to make redefinitions of the fermion fields which leave the gauge interactions invariant but change e.g. the quark yukawa couplings @xmath6 , @xmath7 without changing the physics . the great advantage of the ia is that it allows one to derive cpis which , if non - vanishing , imply @xmath0violation . in the sm , it has been shown @xcite that the relevant cpi is @xmath8^ 3 $ ] , where we define the hermitian combinations @xmath9 and @xmath10 . for the 3 fermion generation case this cpi leads to the jarlskog invariant @xcite . the ia can be applied to any extension of the sm , in particular to extensions of the sm with majorana neutrinos @xcite . it should be emphasised that the ia not only enables one to verify whether a given lagrangian violates @xmath0 , but also provides an idea of how suppressed @xmath0violation might be . a notable example is the possibility of showing why @xmath0 in the sm is too small to generate the baryon asymmetry of the universe ( bau ) . one simply observes that the dimensionless number @xmath11^ 3 / v^{12}$ ] is of order @xmath12 , where we used the hermitian quark mass matrices and @xmath13 gev denotes the scale of electroweak symmetry breaking . this dimensionless number should be compared to the size of observed bau , @xmath14 @xcite . the ia , leading to basis invariant quantities , also identifies what combination of parameters are physical such that , e.g. there is no need to count how many phases can be eliminated through rephasing , which can be laborious in complicated lagrangian , and specially in the presence of family symmetries . recently @xcite the use of cpis , valid for any choice of @xmath0transformation , was advocated as a powerful approach to studying specific models of @xmath0violation in the presence of discrete family symmetries . examples based on @xmath15 and @xmath1 family symmetries were discussed and it was shown how to obtain several known results in the literature . in addition , the ia was used to identify how explicit ( rather than spontaneous ) @xmath0violation arises , which is geometrical in nature , i.e. persisting for arbitrary couplings in the lagrangian . here we intend both to further highlight the usefulness of the ia in dealing with discrete family symmetries and also to systematically explore the @xmath0properties of @xmath1 . by using the ia , we are able to construct cpis independently of the specific group and need to consider the group details only to compute coupling matrices by using the respective clebsch - gordan coefficients in any particular basis . by combining the coupling matrices with the cpis , basis - independent quantities are obtained which indicate if there is @xmath0violation . in this paper we explore in depth the @xmath0properties of unbroken @xmath1 invariant lagrangians using the ia as outlined in @xcite ( see also the proceedings @xcite ) . the method is based on @xcite . we focus on @xmath1 since it involves many features which may be encountered in more general discrete groups , such as complex representations and multiple non - trivial singlet representations . it therefore constitutes a rich playground for exploring the ia in the case of discrete family symmetries . although the cases discussed do not represent realistic models , since the @xmath1 is unbroken , the work here lays the foundation for future models based on spontaneously broken @xmath1 . following the ia we consider several cases that highlight how the @xmath0properties depend both on the field content and on the type of contractions considered ( which may be controlled e.g. by additional symmetries , even though here we do nt always consider those explicitly ) . we focus on tri - linears terms , which we refer to as yukawa - like couplings , keeping in mind that most ( but not all ) cases considered are meant as fermion - fermion - scalar terms . we start by considering lagrangians with just @xmath1 singlets where the ia identifies the relative phases that are physical , and then concentrate on yukawa - like terms involving a triplet , an anti - triplet , and a singlet . when the yukawa - like couplings are between triplet , anti - triplet and singlet , we study cases with a single independent triplet , with independent triplet and anti - triplet , and with three or more independent 3-dimensional irreducible representations . some of the cases considered are similar to adding @xmath1 to a type ii 2 higgs doublet model ( 2hdm ) or @xmath16 higgs doublet model ( nhdm ) . for each of these frameworks the ia allows to identify if the lagrangian has the possibility to explicitly violate @xmath0 , how this depends on how many different @xmath1 singlets are coupled , and what cpis are relevant and non - vanishing when there is @xmath0violation . this serves to further illustrate the convenience and power of the ia for the study of @xmath0properties of specific lagrangians . the layout of the remainder of this paper is as follows . in section [ ia ] we briefly review the ia to @xmath0 in family symmetry models . we continue with the group theory of @xmath1 in section [ d27 ] . section [ singlets ] considers just singlets . in section [ one ] the field content includes one triplet . two triplets are considered in detail in section [ 2triplets ] , where we differentiate also based on the number and type of singlets present . we generalise to three triplets in section [ three ] and to four and more triplets in appendix [ sec:4more ] . for comparison with the ia , we present some examples of specific @xmath0matrices in appendix [ spe ] . finally we conclude in section [ conclusions ] . as mentioned in the introduction , the ia as outlined in @xcite is based on @xcite , where to study the @xmath0properties of a given lagrangian one starts by splitting it @xmath17 where @xmath4 denotes the part that is known to conserve @xmath0(kinetic terms and gauge interactions , as pure gauge interactions conserve @xmath0@xcite ) . @xmath5 includes non - gauge interactions such as the yukawa couplings . a review of how the ia is applied to the standard model ( sm ) lepton sector can be found in @xcite , which also includes its application to a model of spontaneously broken @xmath15 and to a model of @xmath1 which features explicit geometrical @xmath0violation . in this paper we study many different lagrangians invariant under unbroken @xmath1 , relying on the ia . as pointed out in @xcite , the presence of a family symmetry does not change the most general @xmath0transformation which leaves invariant @xmath4 - these are the kinetic terms and the gauge terms which are flavour blind . before continuing with the ia , a relevant question is what role is played by the consistency relations @xcite in this type of analysis . the consistency relations can be obtained by considering that a lagrangian invariant under both a family symmetry and a @xmath0symmetry should be the same whether one considers doing a consistent @xmath0transformation before or after a family symmetry transformation . the concept is illustrated in fig . [ fig : cons ] for a field @xmath18 . transformation @xmath19 is consistent with the group @xmath20 , as following it with the transformation @xmath21 associated with element @xmath22 of @xmath20 , and then with @xmath23 , is equivalent to the transformation @xmath24 associated with some other element @xmath25 of @xmath20 . [ fig : cons],width=6 ] when this is considered rigorously one obtains a relationship between @xmath0transformations @xmath19 and family symmetry transformations @xmath21 @xmath26 if we find an explicit @xmath0transformation that leaves the lagrangian which respects some family symmetry @xmath20 invariant ( even one is enough ) then we can be quite sure that the theory conserves @xmath0as well as respecting the @xmath20 . in this case the consistency relation in eq.[eq : consistency ] is automatically satisfied . this is clear since under a @xmath0transformation , followed by a family symmetry transformation , followed by another @xmath0transformation , etc . , leaves the lagrangian invariant , @xmath27 the consistency of @xmath0and @xmath20 is clear since the lagrangian is left invariant at each stage . however we can be even more explicit than this in order to demonstrate the equivalence of the two approaches . consider a mass term @xmath28 in the lagrangian , then define @xmath29 suppose that the lagrangian is invariant under some family symmetry transformation , @xmath21 , then this implies that the mass term in the lagrangian remains unchanged under a family symmetry transformation and hence @xmath30 the condition for the invariance of the lagrangian under a @xmath0transformation , @xmath19 , requires that the mass term swaps with the @xmath31 mass term hence , @xmath32 taking the complex conjugate of eq.[cond00 ] we find , @xmath33 using eq.[cond0 ] for the last equality . from eq.[cond01 ] and eq.[cond0 ] we obtain , @xmath34 hence @xmath35 where we have used eq.[cond00 ] but for a different group element @xmath25 in the last equality . by comparing both sides of eq.[cond03 ] we identify , @xmath36 which is just the consistency condition in eq.[eq : consistency ] . as we further illustrate in the following sections , using the ia one need not specify a @xmath0transformation . for a given lagrangian it is sufficient to input the invariance conditions imposed by the symmetries . this makes the ia very useful to study @xmath0violation in the presence of family symmetries . the group @xmath1 @xcite , a member of the @xmath37 subgroups of @xmath38 @xcite , has featured prominently in studies related with @xmath0 . it leads to geometrical @xmath0violation as shown in @xcite and @xcite , a feature that was analysed using different methods by @xcite.violation has also been found in other groups @xcite . ] in addition , the @xmath0transformations consistent with @xmath1 triplets were presented in @xcite . using the ia , @xcite found that a specific @xmath1 invariant lagrangian features a different type of geometrical @xmath0violation , where @xmath0is explicitly violated ( rather than spontaneously ) . in this paper we go beyond this result , considering many @xmath1 invariant lagrangians to study in depth how the interplay between @xmath1 and @xmath0changes depending on the representations and how they couple with one another . to do so , some understanding of the group properties is required . @xmath1 has three @xmath39 generators @xcite but we only need to use two , which we refer to as @xmath40 ( for cyclic , with @xmath41 ) and @xmath42 ( for diagonal , with @xmath43 ) . this notation refers to their respective 3-dimensional representation matrices in the basis we use . we define @xmath44 . starting with the 9 distinct singlets which we conveniently label as @xmath45 , the generators are represented by @xmath46 and @xmath47 for that particular singlet . a field transforming as a @xmath48 ( trivial singlet ) is explicitly invariant under @xmath1 transformations , and the other 8 singlets simply get multiplied by the respective powers of @xmath49 when acted upon by @xmath40 or @xmath42 . the other irreducible representations of @xmath1 are triplets , two distinct ones which we take as @xmath50 and @xmath51 . the generator @xmath40 is represented equally for both @xmath52 @xmath42 is represented as a diagonal matrix with entries that are powers of @xmath49 , with the exponents denoted by the indices of the triplet representation @xmath53 the determinant of the matrices is 1 ( @xmath1 is a subgroup of @xmath38 ) and the two indices identify @xmath54 , @xmath55 . the representations @xmath50 and @xmath51 behave as a triplet and anti - triplet , so in analogy with @xmath38 we refer to them mostly as the @xmath56 and @xmath57 representations . the subscript notation is useful to remember the powers of @xmath49 that each component transforms with under @xmath58 so we refer to it occasionally throughout the paper , e.g. if we take @xmath59 transforming as triplet @xmath60 and @xmath61 transforming as ( anti-)triplet @xmath62 , the explicit construction of the trivial singlet is @xmath63 . this can be verified by acting on @xmath64 and @xmath65 with generators @xmath40 and @xmath42 and checking that the prescribed @xmath66 remains invariant . indeed , @xmath67 and rules for constructing the non - trivial singlets from triplet and anti - triplet follow @xmath68 @xmath69 @xmath70 @xmath71 @xmath72 @xmath73 @xmath74 @xmath75 all these can be verified by acting on the triplets with the generators and tracking how each product transforms . to illustrate how the ia would proceed , we start by considering yukawa - like terms without @xmath1 triplets . throughout we refer to fields transforming as singlets under @xmath1 as @xmath76 where the subscript refers to the field being assigned as a @xmath45 under @xmath1 . a simple example where the field content is @xmath77 , @xmath78 , @xmath79 would have yukawa - like terms @xmath80 it is clear there are further @xmath1 invariant terms such as @xmath81 , but for the sake of illustrating the ia we consider the @xmath0properties of the yukawa - like terms in @xmath82 by itself as the only part of the full lagrangian that can violate @xmath0(i.e . as the @xmath5 for this case ) . the next step in the ia is to consider the most general @xmath0transformation for each field consistent with the kinetic terms etc . , in this case this means each singlet transforms with its own phase which we denote as @xmath83 @xmath84 when we apply these transformations on @xmath82 and demand it remains invariant , we obtain a set of necessary and sufficient conditions for @xmath0conservation restricting the parameters in @xmath82 @xmath85 we note that the @xmath0conservation conditions on couplings @xmath86 and @xmath87 are independent of @xmath88 , @xmath89 . to build cpis we combine conditions that cancel dependence on the @xmath0transformations , which involving @xmath86 and @xmath87 requires only the cancellation of all dependence on @xmath90 . a simple and useful cpi is then @xmath91 $ ] , as @xmath92 = 0 \ , , \label{0110_rel}\ ] ] where the @xmath93 are complex numbers so @xmath94 . the cpi vanishing is a necessary ( but not necessarily sufficient ) condition for @xmath0conservation and it constrains the relative phase between the two couplings . there are also cpis of this type constraining the relative phases between @xmath95 and the other two @xmath93 couplings . cpis involving @xmath96 can also be built noting that the cpi needs to cube the other couplings e.g. @xmath97 $ ] , as @xmath98=0 \,.\ ] ] if any of these cpi are non - zero , then @xmath0is violated . other cpis like @xmath99 $ ] ( or @xmath100 $ ] ) do not provide useful constraints as they automatically vanish . as a first generalisation of @xmath82 we add a field @xmath101 . this allows a mixed invariant whose coupling constant will be sensitive to the @xmath0transformation of the non - trivial singlets . we continue to consider only yukawa - like terms in @xmath5 so we write @xmath102 the new @xmath0conditions are @xmath103 with the added couplings , we have from the direct generalisations of eq . [ 0110_rel ] cpis involving @xmath104 like @xmath105 \ , , \\ \mathrm{im } & [ y_{02 } y_{01}^ * ] \ , , \\ \mathrm{im } & [ y_{01 } y_{02}^ * ] \,,\end{aligned}\ ] ] which constrain the relative phases of the @xmath93 couplings . it is however not possible to obtain such simple , useful constraints on the relative phase of the mixed coupling @xmath106 , because it involves @xmath88 and @xmath107 . the simplest cpi , @xmath108 $ ] automatically vanishes and is therefore not useful . the conclusion from this cpi alone would be that the phase of @xmath106 is unconstrained by requiring @xmath109 to conserve @xmath0 . however , we note that by using a more complicated invariant involving @xmath110 , @xmath111 and either @xmath96 or 3 insertions of any @xmath93 , we can build non - trivial cpis involving @xmath106 , such as @xmath112 \ , , \label{z_rel } \\ \mathrm{im } [ y_{1}^3 y_{ij}^ { * 3 } z_{01}^ * z_{02}^ * ] \,.\end{aligned}\ ] ] the situation involving the coupling of mixed terms of the type @xmath106 qualitatively changes if there are sufficient mixed terms . consider now a field content with all 9 @xmath1 singlets @xmath76 . to reduce the number of allowed terms we may impose a @xmath39 symmetry where each @xmath76 transforms equally . such a @xmath113 symmetry forces @xmath114 ( in addition it forbids a multitude of other yukawa - like terms like @xmath115 , where @xmath116 would play the role of @xmath117 ) . there are 9 yukawa - like terms like @xmath118 ( one for each singlet ) . cpis involving the @xmath119 will be like eq . [ z_rel ] . focusing solely on the mixed terms like @xmath120 , there are 12 combinations of singlets whose indices add up to a mixed invariant @xmath121 the @xmath0conservation condition for each coupling depends on the 3 phases of the respective singlets , e.g. @xmath122 with these conditions it is now possible to combine several of the mixed couplings to form a cpi . an example is @xmath123 \,,\ ] ] meaning this particular combination of couplings is constrained by @xmath0conservation to be real . other combinations of this type can be built from the couplings in @xmath124 . we will now consider lagrangians involving yukawa - like couplings with just one @xmath1 triplet @xcite . in order to make invariants , the terms will necessarily involve the conjugate of that triplet . in this case it is not possible to construct yukawa couplings involving a @xmath1 triplet weyl fermion . if @xmath125 , while we do have @xmath126 , the @xmath1 invariant of type @xmath127 is not itself invariant under lorentz symmetry as e.g. @xmath128 implies @xmath129 . we construct the lagrangian with one scalar triplet @xmath130 and 2 scalar singlets @xmath78 , @xmath79 , such that the yukawa - like terms are @xmath131 the most general @xmath0transformations are @xmath132 where @xmath133 is a general unitary matrix . assuming @xmath0invariance of @xmath134 and using matrices @xmath135 corresponding to the couplings @xmath93 we have @xmath136 we can build a useful cpi for this lagrangian @xcite @xmath137 the cpi applies for any matrices @xmath138 and @xmath139 . imposing @xmath1 invariance we have from eq . [ ab01],[ab10 ] @xmath140 @xmath141 which we input into @xmath142 and obtain @xmath143 finding a non - vanishing cpi means that @xmath0is violated , which clearly happens for any non - zero values of @xmath86 and @xmath87 . given that @xmath0is explicitly violated by a phase only originating from the group structure and not from arbitrary lagrangian parameters , this is a minimal case with explicit geometrical @xmath0violation @xcite . in @xcite it was pointed out that @xmath1 provides an example of a group where not all clebsch - gordan coefficients can be made real by a change of basis , when several of the singlets are used . indeed , this fact was already referred in the earlier @xmath1 works @xcite where cp is violated spontaneously and therefore only a few singlets were used . the change of basis analysis presented explicitly in @xcite further clarifies the connection between the inevitability of complex clebsch - gordan coefficients ( which are basis - dependent ) and the presence of multiple singlets . the physical consequences are of course basis - independent as illustrated elegantly in the invariant approach , and depend crucially on the field content , not just of singlets but also of triplets as shown in the following sections . we continue our exploration of unbroken @xmath1 invariant lagrangians with yukawa - like terms by considering in detail the class with two distinct 3-dimensional representations . we take these to be explicitly a triplet @xmath144 and an anti - triplet @xmath145 . , whose @xmath31 @xmath146 would transform as an anti - triplet . ] the notation we are following is suggestive of identifying the @xmath56 and @xmath57 as fermions , as considered in @xcite . nevertheless , it is also possible to consider just scalars @xcite , or that one of the triplets is a scalar and yukawa couplings are formed by having @xmath1 singlet fermions , as considered in @xcite . the conclusions we derive with the ia apply to all cases . as a further point of notation , if e.g. @xmath147 does refer to the sm quark fields , the most general @xmath0transformation consists in @xmath148 here we need not specify if @xmath147 is a fermion or scalar , and in particular we are more concerned with identifying which matrix corresponds to each field ( in eq . [ fermioncp ] , @xmath133 corresponds to @xmath147 ) . for these reasons we use a simplified notation along the lines of @xmath149 which is exact for scalars , and is more convenient to identify which general @xmath0transformation corresponds to each field ( in eq . [ scalarcp ] , @xmath150 corresponds to @xmath147 ) . we assume the @xmath5 part of the lagrangian consists of yukawa - like terms between triplet @xmath144 , anti - triplet @xmath145 , and singlets @xmath151 . this class of lagrangians is a good framework to illustrate several interesting points , so we go into some detail of what happens when varying the number of coupled singlets . the first model we consider of this type is with @xmath147 , @xmath152 and 2 singlets @xmath79 and @xmath78 @xcite @xmath153 before applying the ia to lagrangians of this type , lets consider what happens when applying a specific @xmath0transformation . arguably the simplest @xmath0transformation is the trivial @xmath0transformation , which we refer to as @xmath154 . this corresponds to @xmath155 in eq . [ scalarcp ] , i.e. @xmath156 where we used the subscript to denote that @xmath157 transforms as a @xmath158 , given that under action by generator @xmath42 , @xmath159 and therefore we must have @xmath160 under @xmath42 ( as expected from complex conjugation ) . similarly , @xmath161 for the two @xmath1 singlets @xmath79 and @xmath78 in @xmath162 @xmath163 where the conjugated versions transform under @xmath1 as @xmath164 and @xmath165 respectively . this means that under the trivial @xmath0transformation all four fields go into their respective conjugate @xmath1 representations . the yukawa - like terms in @xmath162 are explicitly invariant under @xmath1 and @xmath87 , @xmath86 are arbitrary complex numbers . we now impose additionally that @xmath162 is invariant under @xmath154 . for the @xmath87 coupling , expanding from eq . [ ab20 ] and using eq . [ cp1a ] , [ cp1b ] , [ cp1c ] @xmath166 in identifying how @xmath167 has transformed under @xmath154 , note the @xmath154-transformed product @xmath167 still transforms as a @xmath164 under @xmath1 , as it picks up a phase of @xmath168 when acted by @xmath40 . for the @xmath86 coupling , expanding from eq . [ ab02 ] and using eq . [ cp1d ] @xmath169 in contrast , because of the action in eq . [ cp1b ] , @xmath170 picks up a phase @xmath49 when acted by @xmath42 , we identify that the @xmath154-transformed product @xmath171 transforms as a @xmath172 under @xmath1 . what are then the physical consequences of imposing @xmath154 on the @xmath1-invariant @xmath162 ? we need to compare eq . [ cp1yc ] and eq . [ cp1 yd ] to the @xmath31 part of @xmath162 . in the case of eq . [ cp1 yd ] this reveals exactly the same expression , except that @xmath173 appears , therefore the conclusion is clear - imposing @xmath154 on @xmath162 forces @xmath174 . however , when we compare what we obtained in eq . [ cp1yc ] to @xmath175 the only way to make the expressions match ( to have @xmath162 be invariant under@xmath154 ) is to require @xmath176 . that there is some incompatibility with @xmath87 was already hinted at by the fact that eq . [ cp1yc ] is explicitly not invariant under @xmath1 , which we denoted through the subscripts - @xmath177 transforms as a @xmath164 , as does @xmath178 . our interpretation of these results is not that @xmath154 becomes incompatible with @xmath1 when the theory includes field @xmath79 together with @xmath147 and @xmath152 . rather , that it is always possible to add @xmath154 to a @xmath1 invariant lagrangian regardless of the field content , but we interpret this result as indicating that there will be physical consequences on the couplings that make the theory consistent by making the lagrangian invariant under the full symmetry imposed . in other words , it is more correct to state the incompatibility is not with the field content but rather with the couplings . that @xmath154 restricts @xmath162 does not mean that @xmath0is violated though , and in fact a @xmath0transformation can be defined that leaves the lagrangian invariant . a good way to check @xmath162 is @xmath0conserving is to use the ia . before doing so , lets examine the effects of another possible @xmath0transformation @xmath179 , keeping the same transformations as @xmath154 for the singlets but where the triplet transforms as @xmath180 where some components swapped positions . similarly , @xmath181 note that unlike what happens with @xmath154 , given that under action from generator @xmath42 we have @xmath182 and therefore @xmath183 , this @xmath184 transforms as a @xmath185 , and similarly this @xmath186 transform as a @xmath62 . checking how the @xmath187 terms transform under @xmath154 we have @xmath188 in identifying how @xmath167 has transformed under @xmath179 , it now picks up a phase of @xmath49 when acted by @xmath40 . the other combination @xmath189 and one can verify that the @xmath167 has transformed under @xmath179 into an expression that picks up a phase of @xmath168 when acted by @xmath42 . by comparing eq . [ cp2yc ] , [ cp2 yd ] we conclude that by imposing @xmath179 invariance on eq . [ l3 ] we are forcing @xmath87 to be real and @xmath190 ( contrast with @xmath154 ) . while we used @xmath1 and @xmath154 ( and @xmath179 ) in this explicit example , our interpretation is general - we have the freedom to impose any family symmetry ( discrete or not ) together with any @xmath0symmetry . eventually what may happen in extreme cases , is that it will not be possible to form non - trivial combinations that are invariant under both symmetries . we feel it is important to stress that in this interpretation , the @xmath0symmetry is consistently treated equally to other symmetries - the transformation is defined and it has consequences for the lagrangian . it is important to stress again that even if imposing a specific @xmath0symmetry on a theory restricts the couplings of the lagrangian , this does not mean that the lagrangian violates @xmath0 . @xmath162 is an example of that as we now show using the ia . it is convenient to first rewrite @xmath162 in terms of coupling matrices @xmath138 and @xmath139 @xmath191 and to specify the general @xmath0transformation properties @xmath192 one can verify that @xmath154 and @xmath179 are particular cases of this general @xmath0transformation . imposing invariance under the general @xmath0transformation requires @xmath193 we again wish to build combinations of the yukawa - like couplings that eliminate the dependence on the general transformation . unlike what we considered with @xmath154 and @xmath179 , @xmath147 and @xmath152 transform in general with distinct unrelated matrices . this forces cpis to alternate between @xmath135 and @xmath194 . it is therefore convenient to define the hermitian combinations @xmath195 and @xmath196 ( similar hermitian combinations appear in the sm cpi , @xmath197 ^ 3 $ ] ) . they are useful because general @xmath0invariance requires @xmath198 involving for each only the matrix associated with @xmath152 and @xmath147 , respectively . the cpis are of the type ( with integers @xmath199 ) @xmath200 \ , , \label{ia2sgp } \\ \mathrm{im } { \mathrm{tr}\ , } [ h_{01}^{n_1 } h_{10}^{n_2 } h_{01}^{n_3 } ( ... ) ] \ , . \label{ia2shp}\end{aligned}\ ] ] when referring to cpis of this type we are including also cpis like @xmath201 \ , , \label{splitg } \\ \mathrm{im } { \mathrm{tr}\ , } [ g_{01}^{n_1 } y_{10}^\dagger y_{01 } g_{10}^{n_2 } y_{01}^\dagger y_{10 } g_{01}^{n_3 } ( ... ) ] \label{splitg2 } \,.\end{aligned}\ ] ] the first one alternates between @xmath202 and @xmath203 by having a single additional @xmath204 in the middle , and eventually goes back to @xmath202 due to the lone @xmath138 that is required to cancel the dependence on @xmath88 . the second one is similar by having @xmath205 between @xmath202 , then requiring @xmath206 somewhere else in order to cancel the dependence on @xmath88 , @xmath89 - note the ordering of the inserted @xmath135 and @xmath194 is not arbitrary . it is also possible to mix and match odd and even insertions of @xmath135 and @xmath194 @xmath207 \label{splitg3 } \,.\ ] ] we refer to the more complicated cpis as being of the type eq . [ ia2sgp ] , [ ia2shp ] because they are obtained by iteratively inserting some @xmath208 or @xmath209 inside an existing @xmath210 or @xmath211 , thus separating the constituent @xmath212 and @xmath194 . and split one of the @xmath202 in the middle with @xmath213 and redefine the integers to obtain eq . [ splitg ] . or in reverse , remove @xmath214 from eq . [ splitg3 ] , which allows an @xmath215 to be created and included into @xmath216 by redefining the integers and arrive back at eq . [ splitg ] - which can then be related with eq . [ ia2sgp ] . ] when considering the yukawa - like matrices that comply with @xmath1 invariance @xmath140 @xmath141 that we have seen before , we have @xmath217 and it is possible to check that all these cpis vanish automatically , for any complex @xmath87 , @xmath86 . therefore @xmath162 conserves @xmath0for any @xmath87 , @xmath86 , even though it is not in general invariant under some of the particular @xmath0transformations like @xmath154 and @xmath179 . the conclusion is a bit more general and applies beyond the pair of singlets @xmath78 and @xmath79 . using the ia it is relatively easy to verify that a lagrangian of type @xmath162 automatically conserves @xmath0for any two singlets @xmath1 ( even if a specific @xmath0like @xmath154 imposes restrictions on these lagrangians ) . the field content is then @xmath147 , @xmath152 and any two singlets @xmath76 , @xmath218 . there are only two yukawa - like terms and associated matrices @xmath219 the cpis are trivial generalisations of the previous cases , e.g. @xmath220 \ , , \label{ia2sg } \\ \mathrm{im } & { \mathrm{tr}\ , } [ h_{ij}^{n_1 } h_{kl}^{n_2 } h_{ij}^{n_3 } ( ... ) ] \ , , \label{ia2sh}\end{aligned}\ ] ] and the other types discussed above . due to invariance under @xmath1 we have @xmath221 regardless of the singlets used . using these relations allows one to conclude that all of these cpis automatically vanish for any couplings @xmath93 , @xmath222 , meaning @xmath0is automatically conserved in lagrangians of @xmath162 type - yukawa - like couplings of any 2 @xmath1 singlets to 2 triplets . however , the possibility for explicit @xmath0violation exists for 3 singlets and beyond - see also @xcite , but note that the conclusion depends on the number of triplets as well ( e.g. 2 singlets are sufficient to violate @xmath0 in the presence of only one triplet @xcite ) . applying the ia to lagrangians of @xmath162 type with 3 singlets allows us to identify explicit geometrical @xmath0violation . in this case we change notation from triplet @xmath147 and anti - triplet @xmath152 to the notation used in the model of @xcite , to allow an easier comparison . we introduce the sm fermions @xmath223 and also @xmath224 and singlet scalars @xmath77 , @xmath78 , @xmath79 . this model is a model of leptons , with a charged lepton lagrangian that gives in this basis a diagonal mass matrix , with a @xmath1 triplet scalar @xmath225 @xcite @xmath226 we focus here on the @xmath0properties of the neutrino lagrangian @xmath227 @xmath228 this is similar to the @xmath162 lagrangian , with yukawa - like terms between the triplets and the singlets . the general @xmath0transformations are @xmath229 assuming @xmath0invariance of @xmath227 and expressing the couplings in terms of yukawa matrices @xmath135 , the @xmath0conservation requirements are @xmath230 @xmath1 imposes @xmath231 @xmath140 @xmath232 because we have 3 yukawa matrices , we can build a cpi that does not involve @xmath208 or @xmath209 combinations @xmath233 this cpi is qualitatively different from the ones that could be built with only 2 matrices . indeed if we calculate it for this particular choice of 3 singlets @xmath234 this means that in general , for arbitrary ( non - zero ) couplings , this lagrangian violates @xmath0as the condition @xmath235 , necessary for @xmath0conservation , is not fulfilled . this is the case originally identified to have explicit geometrical @xmath0violation in @xcite , with the phase appearing in @xmath236 independent of the arbitrary phases of couplings . we can again check what are the consequences of adding specific @xmath0transformations , this time to a lagrangian where explicit @xmath0violation is possible . from the analysis of the rather similar @xmath162 lagrangian , we can conclude that imposing @xmath154 to @xmath227 leads to @xmath237 , and that imposing @xmath179 leads to @xmath190 . as may have been expected , both make @xmath236 vanish . this is a clear demonstration of the meaning of adding different @xmath0transformations - both lead to @xmath0conservation but not necessarily the same consequences . , @xmath152 and singlets @xmath78 , @xmath79 and @xmath238 in the basis where these singlets correspond to @xmath172 , @xmath239 and @xmath164 of @xmath1 respectively , then impose @xmath154 in that basis . the two couplings @xmath87 and @xmath240 are forced to vanish . conversely , @xmath179 forces only one coupling to vanish , @xmath86 . if one changes the basis it may be that e.g. @xmath154 is no longer associated with the @xmath241 matrix in the new basis , but it will nevertheless force two couplings to vanish whereas @xmath179 forces only one coupling to vanish . see @xcite for more specific considerations regarding basis changes . ] the type of cpi exemplified in @xmath236 is very useful in the study of yukawa - like terms between triplet , anti - triplet and singlets . for @xmath0to be conserved it is necessary that it vanishes . in cases with with 3 @xmath1 singlets , it is also sufficient that the respective cpi vanishes for this type of lagrangian to conserve @xmath0 . the combinations of 3 singlets that automatically conserve @xmath0can be found @xcite to be 12 out of the total 84 combinations of 3 singlets . curiously , these combinations are the 12 combinations of 3 singlets appearing in each of the 12 terms of @xmath124 in eq . [ ls9 ] of section [ singlets ] . to be clearer , consider explicitly the @xmath242 term in eq . we construct a lagrangian similar to @xmath227 but with singlets @xmath77 , @xmath78 , @xmath101 @xmath243 then use yukawa - like matrices @xmath135 and build the cpi similar to @xmath236 @xmath244 then we repeat these steps for the @xmath245 term in eq . [ ls9 ] and so on . when inserting the conditions imposed by @xmath1 on the yukawa - like matrices @xmath135 , these 12 cpis vanish automatically for any values of the yukawa - like couplings @xmath246 one can explicitly build @xmath0transformations to prove that indeed the 12 respective lagrangians @xmath247 , @xmath248 , @xmath249 , @xmath250 automatically conserve @xmath0(for any values of the 3 yukawa - like couplings of each respective lagrangian ) . as illustrated by @xmath236 itself , the cpis for the other 3 singlet combinations do not vanish automatically and appear with the respective @xmath251 , multiplied by a factor of @xmath252 or @xmath253 . they are further examples of explicit geometrical @xmath0violation . any choice of 4 or more singlets will necessary include combinations of 3 that would allow @xmath0violation . for example , by adding any other singlet to the set @xmath77 , @xmath78 , @xmath101 in @xmath247 , we have a singlet with @xmath76 with @xmath254 and any of the @xmath236-type cpis involving @xmath135 with @xmath255 , @xmath138 , @xmath256 @xmath257 each of which falls in one of the 72 cases that are in general non - zero ( unless the respective couplings are vanishing ) . we continue investigating lagrangians with yukawa - like terms between triplets and singlets , in the presence of 3 triplets of @xmath1 . in contrast to the situation studied in @xcite where there is a scalar @xmath1 triplet coupling to fermions , we consider a situation where we generalise the @xmath162 lagrangian with additional anti - triplet @xmath258 . the triplet @xmath144 now contains the sm quark @xmath259 doublets , and two anti - triplets @xmath260 contain the up and down quark @xmath259 singlets . the scalars are higgs doublets @xmath261 and @xmath262 ( we deviate here slightly from the notation used for scalars in other sections ) . with a @xmath263 symmetry it is possible to have the @xmath258 couple only to @xmath264 and @xmath152 couple only to @xmath265 , leading to a type ii 2hdm ( see e.g. @xcite ) where the actual yukawa terms are constrained by @xmath1 . the lagrangian is @xmath266 we express the lagrangian in terms of yukawa matrices @xmath6 , @xmath7 and apply the ia to this lagrangian and @xmath152 are explicitly fermions , we continue using the abridged notation of eq . [ scalarcp ] rather than the more rigorous one of eq . [ fermioncp ] . ] @xmath267 the conditions on @xmath6 , @xmath7 from imposing general @xmath0invariance on @xmath268 are @xmath269 we see that we can without loss of generality redefine @xmath270 to absorb @xmath271 and @xmath272 to absorb @xmath273 . we choose then @xmath274 and redefine @xmath270 , @xmath272 accordingly . and @xmath78 in @xmath162 as there was only one anti - triplet @xmath152 . ] in effect what this means is that we have , for @xmath6 and @xmath7 , the same type of @xmath0conservation requirements that they would have in the sm . so we extrapolate from @xcite , rely on the hermitian combinations @xmath275 to eliminate the dependence in @xmath276 @xmath277 and conclude that @xmath278 ^ 3 = 0 $ ] @xcite is a necessary and sufficient condition for @xmath0conservation for @xmath268 . we can now use the @xmath6 and @xmath7 that @xmath1 would impose @xmath279 which happens to correspond to the usual basis where @xmath280 is diagonal and @xmath281 in the limit of unbroken @xmath1 in fact @xmath282 so it is clear that @xmath0is automatically conserved for any @xmath283 , @xmath284 . indeed , by using the ia on the lagrangian we conclude that the only way to build cpis is to keep yukawa structures that couple to @xmath152 together , and also to keep yukawa structures that couple to @xmath258 together , as that is the only way to cancel the respective @xmath270 and @xmath272 dependence . in addition to combining @xmath280 and @xmath285 , one can use just @xmath286 or just @xmath287 , but not mix both . continuing from the type ii 2hdm yukawa lagrangian , using the ia it is relatively straightforward to generalise the conclusions for an increasing number of scalars . it is useful to classify each scalar and their respective yukawa matrix according to their sector , i.e. an @xmath288 couples to @xmath152 or an @xmath289 couples to @xmath258 . it is in general no longer possible to absorb the respective phases into @xmath276 . the respective @xmath0conservation requirements include @xmath290 which cancel the dependence on phases @xmath291 , @xmath292 , but as we have seen already cpis with the hermitian combinations are automatically verified in unbroken @xmath1 . however , it is now possible to build cpis of the type of @xmath236 by using either 3 different @xmath293 or 3 different @xmath294 - without mixing the two sectors . using the conclusions derived for the lagrangians with 2 triplets ( 1 triplet and 1 anti - triplet ) in section [ 2triplets ] , we can extrapolate to this 3 triplet case ( 1 triplet and 2 anti - triplets ) . doing so , we conclude that each sector remains automatically @xmath0conserving when coupled to up to any 2 @xmath1 singlets , and can remain automatically @xmath0conserving when coupled to 3 @xmath1 singlets ( if the 3 singlets are one of the 12 special combinations , as discussed in section [ 2triplets ] ) . if a specific representation is repeated in the up and down sector this still represents 2 fields , as they are distinguished by the type ii @xmath263 symmetry that distinguishes the sectors . on the other hand , the 3 singlet representations chosen for the up sector and for the down sector need not be the same , so one can couple up to 6 different singlet representations without enabling @xmath0violation . conversely , coupling the triplets to 7 or more distinct singlet representations will not allow automatic @xmath0conservation , and the minimal singlet content that enables @xmath0violation is 3 singlet representations all in the same sector . recall also that even without additional symmetries , for @xmath1 triplet weyl fermions one can not construct @xmath295 , @xmath296 or @xmath297 due to lorentz invariance ( even though the combinations would be @xmath1 invariant ) . if one continues to generalise this class of lagrangians to 4 or more triplets the conclusion is indeed that we can separately treat each yukawa - like sector ( a distinct triplet to anti - triplet pairing ) . this is because the general @xmath0matrices appearing for each sector are unrelated , and that constrains the types of cpis that can be built in the ia . in @xmath1 , each sector can couple to as many as 3 different singlets before @xmath0violation arises as a possibility . this extends the conclusion derived in section [ 2triplets ] for lagrangians with a single yukawa - like sector , @xmath187 , @xmath227 and the 12 special combinations @xmath247 , @xmath249 . it is interesting to note that with 3 sectors , which arises as a possibility with 4 or more triplets , the full representation content of @xmath1 with all 9 singlets can be present while @xmath0is still automatically conserved . given that cases with 4 or more triplets no longer have a counterpart with the sm quarks , we relegate a more detailed analysis to appendix [ sec:4more ] . in addition we present some examples of specific @xmath0transformations in appendix [ spe ] , which includes an existence proof of @xmath0transformations for a lagrangian with all the irreducible representations of @xmath1 . the group @xmath1 is very interesting from the point of view of @xmath0properties . in this work we considered several @xmath1 invariant lagrangians with yukawa - like terms ( tri - linears ) and studied them with the invariant approach . our dual purpose was to demonstrate the usefulness of the invariant approach in lagrangians invariant under discrete family symmetries , and simultaneously to explore the @xmath0properties of @xmath1 . the method is independent of the group when the @xmath0-odd invariants are constructed , and the group details are needed only to obtain coupling matrices in some convenient basis , which can then be used in the @xmath0-odd invariants to obtain basis - independent quantities signalling @xmath0violation . starting with simple cases where the field content includes only 1-dimensional representations of @xmath1 ( singlets ) , the invariant approach reveals what are the relevant physical phases , which turn out to be specific relative phases of the complex couplings . we then turned to consider yukawa - like terms involving @xmath1 triplet and anti - triplet , starting with a single 3-dimensional representation ( triplet ) and progressing to two and more triplets , where it becomes helpful to refer to sectors of distinct pairs of triplet and anti - triplet . the conclusions derived for the two triplet case with one sector are that @xmath0is automatically conserved for yukawa - like terms involving up to any 2 @xmath1 singlets and for 12 special combinations out the total 84 combinations of 3 singlets . the other cases are examples of explicit geometrical @xmath0violation . based on these results , the invariant approach allows us to extrapolate for cases with three or more triplets . the same type of conclusion holds independently for each sector , and therefore with 3 sectors it is even possible to have all 9 @xmath1 singlets present while automatically conserving @xmath0 . we have therefore completed a fairly exhaustive analysis of unbroken @xmath1 lagrangians . the analysis here should provide a useful guide for formulating future realistic models in which @xmath1 is spontaneously broken . however the main motivation was to highlight the utility and power of the invariant approach for lagrangians with discrete family symmetries . this project has received funding from the european union s seventh framework programme for research , technological development and demonstration under grant agreement no pief - ga-2012 - 327195 sift . the authors also acknowledge partial support from the european union fp7 itn - invisibles ( marie curie actions , pitn- ga-2011- 289442 ) , and by fundao para a cincia e a tecnologia ( fct , portugal ) through the projects cern / fp/123580/2011 , ptdc / fis - nuc/0548/2012 and cftp - fct unit 777 ( uid / fis/00777/2013 ) which are partially funded through pocti ( feder ) , compete , qren and eu . we thank cern for hospitality . the final generalisation that we consider is to add more fields transforming as triplet representations . following from the three triplet case with triplet @xmath147 and anti - triplets @xmath152 , @xmath258 , we add another anti - triplet @xmath298 . we continue to assume that each anti - triplet has its own sector of @xmath1 singlets denoted as @xmath288 , @xmath289 , @xmath299 due to e.g. an abelian symmetry . using the ia and considering how cpis can be constructed we extend the previous results to conclude that the relevant cpis are of @xmath236 type for each sector , due to the different @xmath272 , @xmath270 , @xmath300 matrices @xmath301 @xmath302 it is interesting that at 4 triplets ( in this case 1 triplet and 3 anti - triplets ) we have reached a situation where @xmath0can be automatically conserved even with fields transforming as each of the 9 @xmath1 singlets ( one example is @xmath303 , @xmath304 , @xmath305 , @xmath306 , @xmath307 , @xmath308 , @xmath309 , @xmath310 , @xmath311 ) . at this stage it should be relatively clear that adding further anti - triplets @xmath312 , @xmath313 , @xmath249 , with their own singlet sector , the ia still shows the relevant cpis to be of @xmath236 type for each sector because of the different @xmath314 , @xmath315 , @xmath249 . conversely , the ia also allows to generalise to cases where there are multiple triplets and multiple anti - triplets . considering distinct triplets @xmath147 , @xmath316 , and several anti - triplets , in general the conclusion depends on how many separate sectors are present , but there are some subtleties . for example , take the lagrangian @xmath317 with the usual general transformations for the singlets each with their own phase and @xmath318 in this case there are 3 sectors like in @xmath319 , because some additional symmetry distinguishes triplets @xmath147 and @xmath316 such that @xmath316 pairs only with @xmath320 , @xmath321 . consider instead a situation where @xmath147 and @xmath316 couple to the same anti - triplets and singlets , like in the lagrangian @xmath322 which has 6 sectors . note though that while e.g. @xmath323 and @xmath324 appear to be distinct sectors , if @xmath323 has automatic @xmath0conservation having up to 3 @xmath288 singlets , this applies also to @xmath324 , because the singlets @xmath288 are the same . it is interesting to compare the ia to the construction of specific @xmath0transformations for lagrangians of @xmath162 type with any of the singlets @xmath325 @xmath326 for each yukawa - like term @xmath333 we have a @xmath0conservation requirement on the yukawa - like matrix @xmath135 @xmath334 therefore in general the @xmath335 and @xmath272 matrices that respect this requirement are different for each @xmath135 . within this point of view , the possibility of @xmath0violation arises when it is impossible to have even a single set of @xmath335 , @xmath272 and @xmath83 transformations that simultaneously fulfil the distinct requirements of all @xmath135 that are present in the lagrangian . checking this can be quite laborious as it should be done with complete generality and indeed one of the advantages of the ia is that usually one need not check for the existence of such transformations . for illustration purposes we fix for simplicity the respective @xmath336 } $ ] , @xmath155 , and @xmath272 to be diagonal . for these choices we present for each @xmath135 the respective @xmath272 @xmath337 corresponds to the requirement of the @xmath77 , @xmath78 and @xmath101 yukawa - like coupling ; @xmath338 corresponds to the requirement of the @xmath79 , @xmath339 and @xmath340 yukawa - like coupling ; @xmath341 corresponds to the requirement of the @xmath238 , @xmath342 and @xmath117 yukawa - like coupling . these @xmath0transformations are with loss of generality ( e.g. @xmath272 need not be diagonal ) , but they are still an existence proof of a valid @xmath0transformation for each singlet by itself , and for the specific groups of 3 singlets shown . it naturally agrees with what was obtained through the ia for the more general case . for example , we knew already that with a single sector @xmath323 , these 3 combinations of three singlets belong to the 12 that automatically conserve @xmath0 . furthermore , with 3 sectors @xmath323 , @xmath343 and @xmath344 we have in addition to @xmath272 also @xmath270 and @xmath300 , enabling the possibility to have all 9 @xmath1 singlets in yukawa - like terms with triplets while automatically conserving @xmath0for any arbitrary ( non - zero ) complex value of the nine @xmath93 . in this 3 sector case with @xmath319 in eq . [ l_4q ] , an existence proof for @xmath303 , @xmath304 , @xmath305 , @xmath306 , @xmath307 , @xmath308 , @xmath309 , @xmath310 , @xmath311 follows by keeping @xmath336 } $ ] , @xmath155 , @xmath345 and taking @xmath270 and @xmath300 respectively as the diagonal matrices appearing in eq . [ ud10],[ud20 ] . j. bernabeu , g. c. branco and m. gronau , phys . b * 169 * ( 1986 ) 243 . w. grimus and m. n. rebelo , phys . * 281 * ( 1997 ) 239 [ hep - ph/9506272 ] . g. ecker , w. grimus and w. konetschny , nucl . b * 191 * ( 1981 ) 465 ; g. ecker , w. grimus and h. neufeld , nucl . b * 247 * ( 1984 ) 70 . c. jarlskog , phys . * 55 * ( 1985 ) 1039 . g. c. branco , l. lavoura and m. n. rebelo , phys . b * 180 * , 264 ( 1986 ) ; g. c. branco , m. n. rebelo and j. i. silva - marcos , phys . lett . * 82 * , 683 ( 1999 ) [ hep - ph/9810328 ] ; h. k. dreiner , j. s. kim , o. lebedev and m. thormeier , phys . rev . d * 76 * , 015006 ( 2007 ) [ hep - ph/0703074 [ hep - ph ] ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1502.01589 [ astro-ph.co ] , r. h. cyburt , b. d. fields , k. a. olive and t. h. yeh , arxiv:1505.01076 [ astro-ph.co ] . g. c. branco , i. de medeiros varzielas and s. f. king , arxiv:1502.03105 [ hep - ph ] . i. de medeiros varzielas , arxiv:1503.02633 [ hep - ph ] . g. c. branco , p. m. ferreira , l. lavoura , m. n. rebelo , m. sher and j. p. silva , phys . * 516 * ( 2012 ) 1 [ arxiv:1106.0034 [ hep - ph ] ] . f. feruglio , c. hagedorn and r. ziegler , jhep * 1307 * ( 2013 ) 027 [ arxiv:1211.5560 [ hep - ph ] ] . m. holthausen , m. lindner and m. a. schmidt , jhep * 1304 * ( 2013 ) 122 [ arxiv:1211.6953 [ hep - ph ] ] . g. c. branco , j. m. gerard and w. grimus , phys . b * 136 * ( 1984 ) 383 . i. de medeiros varzielas , s. f. king and g. g. ross , phys . b * 648 * ( 2007 ) 201 [ hep - ph/0607045 ] . e. ma , mod . lett . a * 21 * ( 2006 ) 1917 [ hep - ph/0607056 ] . c. luhn , s. nasri and p. ramond , j. math . * 48 * ( 2007 ) 073501 [ hep - th/0701188 ] . h. ishimori , t. kobayashi , h. ohki , y. shimizu , h. okada and m. tanimoto , prog . theor . phys . suppl . * 183 * ( 2010 ) 1 [ arxiv:1003.3552 [ hep - th ] ] . i. de medeiros varzielas and d. emmanuel - costa , phys . d * 84 * ( 2011 ) 117901 [ arxiv:1106.5477 [ hep - ph ] ] . i. de medeiros varzielas , d. emmanuel - costa and p. leser , phys . lett . b * 716 * ( 2012 ) 193 [ arxiv:1204.3633 [ hep - ph ] ] . g. bhattacharyya , i. de medeiros varzielas and p. leser , phys . * 109 * ( 2012 ) 241603 [ arxiv:1210.0545 [ hep - ph ] ] . i. de medeiros varzielas and d. pidt , j. phys . g * 41 * ( 2014 ) 025004 [ arxiv:1307.0711 [ hep - ph ] ] . i. medeiros varzielas and d. pidt , jhep * 1311 * ( 2013 ) 206 [ arxiv:1307.6545 [ hep - ph ] , arxiv:1307.6545 ] . m. c. chen , m. fallbacher , k. t. mahanthappa , m. ratz and a. trautner , nucl . b * 883 * ( 2014 ) 267 [ arxiv:1402.0507 [ hep - ph ] ] . i. p. ivanov and c. c. nishi , jhep * 1501 * ( 2015 ) 021 [ arxiv:1410.6139 [ hep - ph ] ] . m. fallbacher and a. trautner , nucl . b * 894 * ( 2015 ) 136 [ arxiv:1502.01829 [ hep - ph ] ] . i. de medeiros varzielas , jhep * 1208 * ( 2012 ) 055 [ arxiv:1205.3780 [ hep - ph ] ] . i. p. ivanov and l. lavoura , eur . j. c * 73 * ( 2013 ) 4 , 2416 [ arxiv:1302.3656 [ hep - ph ] ] . i. de medeiros varzielas , j. phys . * 447 * ( 2013 ) 012038 [ arxiv:1302.3991 [ hep - ph ] ] . c. c. nishi , phys . d * 88 * ( 2013 ) 3 , 033010 [ arxiv:1306.0877 [ hep - ph ] ] .

we analyze conditions for violation of the bell inequality in the clauser - horne - shimony - holt form , focusing on the josephson phase qubits . we start the analysis with maximum violation in the ideal case , and then take into account the effects of the local measurement errors and decoherence . a special attention is paid to configurations of the qubit measurement directions in the pseudospin space lying within either horizontal or vertical planes ; these configurations are optimal in certain cases . besides local measurement errors and decoherence , we also discuss the effect of measurement crosstalk , which affects both the classical inequality and the quantum result . in particular , we propose a version of the bell inequality which is insensitive to the crosstalk .

0707.0036

in 1935 einstein , podolsky , and rosen ( epr ) have shown in the classical paper @xcite that quantum mechanics contradicts the natural assumption ( the `` local realism '' ) that a measurement of one of two spatially separated objects does not affect the other one . this `` spooky action at a distance '' known as an entanglement is now recognized as a major resource in the field of quantum information and quantum computing . @xcite the paradox has led epr to conclude that quantum mechanics is an incomplete description of physical reality , thus implying that some local hidden variables are needed . the epr paradox remained at the level of semi - philosophical discussions until 1964 , when john bell contributed an inequality for results of a spin - correlation experiment , @xcite which should hold for any theory involving local hidden variables , but is violated by quantum mechanics . inspired by bell s idea , in 1969 , clauser , horne , shimony , and holt @xcite ( chsh ) proposed a version of the bell inequality ( bi , the generic name for a family of inequalities ) , which made the experimental testing of local hidden - variable theories possible . the main advantage of the chsh inequality in comparison with the original bi is that it does not rely on an experimentally unrealistic assumption of a perfect anticorrelation between the measurement results when two spin-1/2 particles ( in the spin-0 state ) are measured along the same direction . many interesting experiments @xcite have been done since then . the results of these experiments clearly show a violation of the bell inequalities , in accordance with quantum mechanical predictions . the bi violation has been mostly demonstrated in the experiments with photons;@xcite it has been also shown in the experiments with ions in traps @xcite and with an atom - photon system ; @xcite a bell - like inequality violation has been also demonstrated in an experiment with single neutrons.@xcite experimental violation of the bi in solid - state qubits would be an important step towards practical quantum information processing by solid - state devices.@xcite experiments on observation of the bi violation in josephson phase qubits are currently underway . @xcite theoretical study related to the bi violation in solid - state systems has also attracted significant attention in recent years . @xcite in this paper we discuss the bell inequality ( in the chsh form ) for solid - state systems , focusing on experiments with superconducting phase qubits . we study effects of various factors detrimental for observation of the bi violation , including local measurement errors , decoherence , and interaction between qubits ( crosstalk ) , and analyze optimal conditions in presence of these nonidealities . the paper is organized as follows . in sec . [ ii ] we review the chsh type of the bi , and also discuss tomography - type measurements using qubit rotations . in sec . [ iii ] we consider the ideal case and describe all situations for which the bi is violated maximally . sections [ iv]-[vii ] are devoted to the effects of various nonidealities on the observation of the bi violation . in sec . [ iv ] we discuss the effect of local measurement errors , using a more general error model than in the previous approaches . @xcite analytical results for maximally entangled states and numerical results for general two - qubit states are presented . in sec . [ vi ] we consider the effect of local decoherence of the qubits . in sec . [ vii ] we discuss measurement crosstalk , which affects both the bi ( since crosstalk is a classical mechanism of communication between qubits ) and the quantum result . we also propose a version of the bi , which is not affected by the crosstalk . section [ viii ] provides concluding remarks . we begin with a brief review of the chsh inequality , @xcite a type of the bell inequality usually used in experiments . let us consider a pair of two - level systems ( qubits ) @xmath0 and @xmath1 . assuming that a realistic ( classical ) theory based on local observables @xcite holds and there is no communication between the qubits ( i.e. , no crosstalk ) , the two - qubit measurement results should satisfy the chsh inequality @xcite -2s2 , where s = e(,)-e(,)+e(,)+ e(, ) . here @xmath2 and @xmath3 ( @xmath4 and @xmath5 ) are the unit radius - vectors on the bloch sphere along the measurement axes for qubit @xmath0 ( @xmath1 ) and @xmath6 is the correlator of the measurement results : e(,)=p_++(,)+p_(,)- p_+-(,)-p_-+ ( , ) , where @xmath7 is the joint probability of measurement results @xmath8 and @xmath9 for qubits @xmath0 and @xmath1 , respectively . the sum of the probabilities in eq . equals one . this can be used to recast eq . in the form s=4t+2 , where t = p(,)-p(,)+p(,)+ p(,)-p_a()-p_b ( ) . here @xmath10 , whereas @xmath11 [ or @xmath12 is the probability of the measurement result `` + '' for qubit @xmath0 ( or @xmath13 irrespective of the measurement result for the other qubit [ in the classical theory in absence of communication between qubits , @xmath14 is obviously independent of the direction @xmath4 and even independent of the very fact of the qubit @xmath1 measurement ; similarly for @xmath15 . thus , instead of the probabilities @xmath16 , one can equivalently use the probabilities @xmath17 , @xmath18 , and @xmath19 , and the inequality can be recast in an equivalent form @xcite -1t0 . notice that both inequalities and can have somewhat different meanings in different physical situations . in particular , the results `` @xmath20 '' and `` @xmath21 '' may correspond to the presence or absence of the detector `` click '' ( so called `` one - channel measurement '' @xcite ) ; in this case a low - efficiency detector significantly increases the chance of the result `` @xmath21 '' . another possibility is the so - called `` two - channel measurement '' , in which the qubit states``@xmath20 '' and `` @xmath21 '' are supposed to produce clicks in different detectors ; @xcite in this case inefficient detection leads to three possible results : `` @xmath20 '' , `` @xmath21 '' , and `` no result '' . significant inefficiency of the optical detectors leads to the so - called `` detector loophole , '' @xcite which arises because effectively not the whole ensemble of the qubit pairs is being measured . this problem is often discussed in terms of contrasting the clauser - horne @xcite ( ch ) and chsh interpretations of the inequalities , which differ by considering either the whole ensemble or a subensemble of the qubit pairs . it is important to mention that in the case of the josephson phase qubits ( which formally belongs to the class of one - channel measurements ) the whole ensemble of qubit pairs is being measured , and therefore there is no detector loophole ( if one avoids @xcite corrections for measurement errors ) , as well as there is no difference between chsh and ch interpretations . in some cases , as for josephson phase qubits , the measurement ( detector ) axis can not be physically rotated . however , instead of the detector rotation , one can rotate the qubit state . @xcite let us show the equivalence of the two methods explicitly , using the example of the phase qubit and assuming ideal ( orthodox ) measurement . the hamiltonian of the phase qubit in a microwave field in the subspace of the two lowest states in the qubit potential well @xmath22 and @xmath23 is & h_q=&(_q/2)(|_1_1| -|_0_0| ) + & & + ( t)(t+ ) ( |_0_1|+|_1_0| ) , where @xmath24 is the planck constant , @xmath25 is the qubit resonance frequency , @xmath26 is the microwave frequency , @xmath27 is the time - dependent rabi frequency , @xmath28 is the dipole - moment matrix element , and @xmath29 is the electric - strength amplitude of the microwave field . transforming to the qubit basis ( the rotating frame ) @xmath30 , @xmath31 , neglecting fast oscillating terms in the hamiltonian ( the rotating - wave approximation ) and assuming the resonance condition @xmath32 , we arrive at the hamiltonian h(t)= , where the unit vector @xmath33 lies in the @xmath34 plane making the angle @xmath35 with the @xmath36 axis , whereas @xmath37 is the vector of the pauli matrices . @xcite here and below we associate state @xmath38 ( @xmath39 ) with the measurement result `` @xmath20 '' ( `` @xmath21 '' ) and with the eigenvalue @xmath40 ( @xmath41 ) of @xmath42 , so that @xmath43 . as follows from eq . , the microwave pulse rotates the qubit state such that the initial density matrix @xmath44 becomes @xmath45 , where & u_r&=e^-i/2 = ( /2)-i(/2 ) + & & = ( cc ( /2)&e^-i(/2 ) + -e^i(/2)&(/2 ) ) . here @xmath46 , where @xmath47 and @xmath48 are the pulse starting and ending time moments . the probability of the qubit to be found in state @xmath49 is p_i = tr(|ii|_q)=tr(p_i_q ) ( i=0,1 ) . here @xmath50 is the projection operator @xmath51 , i.e. , & & p_1()= ( cc 1+&e^-i + e^i&1- ) = ( i+ ) , + & & p_0()=i - p_1()= ( i- ) , where @xmath52 is the identity matrix and @xmath2 is the unit vector = ( , , ) defining the measurement axis . equations and show explicitly the equivalence between the qubit and detector rotations . namely , one can interpret @xmath53 ( @xmath54 ) as the probability of the qubit to be found in the state with the pseudospin parallel ( antiparallel ) to the measurement axis @xmath2 . notice that the microwave phase @xmath55 is naturally defined modulo @xmath56 , while the rabi rotation angle @xmath57 can be always reduced to a @xmath56 range ( we will assume @xmath58 ) . nevertheless , with this restriction there are still two sets of angles @xmath59 corresponding to the same measurement direction @xmath2 , because @xmath2 is invariant under the change ( , ) ( -,+ ) . it is easy to make a one - to - one correspondence between the measurement direction @xmath2 and angles @xmath59 by limiting either @xmath57 or @xmath55 to a @xmath60-range ( instead of @xmath56 ) . however , we prefer not to do that because it is convenient and natural physically to have a @xmath56 range for one angle when the other angle is fixed . so , as follows from eqs . and , the polar ( zenith ) and azimuth spherical coordinates of @xmath2 are equal to , respectively , @xmath57 and @xmath55 when @xmath61 , and @xmath62 and @xmath63 when @xmath64 . the joint probability of the two - qubit measurement can be written as p_ij ( , ) = tr[p_i^a ( ) p_j^b ( ) ] , where @xmath65 is the two - qubit density matrix , @xmath66 , and @xmath67 . the purpose of this paper is to analyze conditions needed to observe the bi violation in experiment . since it is usually easier to observe an effect when it is maximal , we start the analysis with the situations where violation of the bi is maximal . equations and yield @xmath68 , where a = p_1^a()-p_0^a()= _ a and similarly @xmath69 . here @xmath70 and @xmath71 , the eigenvalues of @xmath72 and @xmath73 being @xmath74 . correspondingly , as follows from eq . , s = tr(b ) , where the bell operator @xcite @xmath75 is = ab - ab+ab+ab ( here @xmath76 and @xmath77 ) . the maximum and minimum values of @xmath78 ( so - called cirelson s bounds @xcite ) s_=2 can be obtained , for example , in the following way . @xcite since the bell operator @xmath75 is hermitian , @xmath79 are equal to the maximum and minimum eigenvalues of @xmath75 . these eigenvalues can be found analyzing the eigenvalues of @xmath80 : ^2=4+[a , a][b , b ] = 4 - 4(_a ) ( _b ) ( here the vector product is taken before the scalar product ) . since the eigenvalues of the pauli matrices are equal to @xmath81 , the largest eigenvalue of @xmath80 is 8 , achieved when @xmath82 and @xmath83 . in this case the maximum and minimum eigenvalues of @xmath75 are @xmath84 , thus leading to eq . ( [ 19 ] ) . ( both values @xmath84 are realized because in this case the other eigenvalue of @xmath80 is 0 , and the sum of all four eigenvalues of @xmath75 should be equal to 0 since @xmath85 . ) consider some useful properties of @xmath78 . as follows from eqs . and , the value of @xmath78 is invariant under arbitrary local unitary transformations @xmath86 and @xmath87 , ( u_au_b)(u_a^u_b^ ) , if simultaneously @xmath88 or , equivalently , r_a , r_a , r_b , r_b , where @xmath89 ( @xmath90 ) is the rotation matrix corresponding to @xmath86 ( @xmath87 ) , so that , e.g. , @xmath91 . this invariance is an obvious consequence of the equivalence between the qubit and detector rotations , discussed in the previous section . as a result of the invariance , if some state is known to violate the bi for a given configuration of the detectors , one can obtain many other states and the corresponding detector configurations providing the same bi violation , by using eqs . with all possible local rotations . note also that @xmath92 inverts the sign if the pair of vectors @xmath93 ( or @xmath94 ) inverts the sign . correspondingly , for a given state s - s -,-(-,- ) . as follows from eq . , there is a one - to - one correspondence between the classes of detector configurations maximizing and minimizing @xmath78 for a given state . for any given detector configuration satisfying the condition , each of the cirelson s bounds is achieved for a unique maximally entangled state . @xcite in contrast , for a given maximally entangled state there can be many optimal detector configurations giving the maximal bi violation . to the best of our knowledge , only the configurations with the detector axes lying in one plane have been usually considered in the literature , though generally the detector axes for different qubits may lie in two different planes . moreover , the bi violation has been studied mainly for one of the bell states , @xcite & & |_=(|10|01)/ , [ 28 ] + & & |_=(|00|11)/ , while an arbitrary maximally entangled state can be written as |_me=(|_1^a_1^b+ |_2^a_2^b)/ , where @xmath95 is an orthonormal basis for qubit @xmath96 . our purpose here is to determine all optimal detector configurations for any maximally entangled state . let us start , assuming that the qubits are in the singlet state @xmath97 . for this state latexmath:[$e(\vec{a},\vec{b})=\langle\psi_-| ( \vec{a}\cdot\vec{\sigma})\otimes(\vec{b}\cdot\vec{\sigma } ) maximizing this formula over @xmath2 and @xmath99 , we should choose @xmath2 to be parallel to @xmath100 , while @xmath3 should be antiparallel to @xmath101 . therefore@xcite @xmath82 , since @xmath100 and @xmath102 are mutually orthogonal . similarly , rewriting @xmath78 as @xmath103 , we can show that maximization of @xmath78 requires @xmath83 . in this way we easily show that the _ necessary and sufficient condition _ for reaching the upper bound @xmath104 for the singlet state is , = -(+)/ , =(-)/. because of the symmetry ( [ 3.25 ] ) , the necessary and sufficient condition for reaching the lower bound @xmath105 can be obtained by the inversion of the detector directions for one of the qubits : , = ( + )/ , =(-)/. equations and show that for the singlet state the maximum bi violation requires that the detector axes for both qubits lie in the same plane . however , the orientation of this plane is arbitrary , since eqs . and determine only the angles between the detector axes . all configurations maximizing ( or minimizing ) @xmath78 for the singlet state can be obtained from _ one _ maximizing ( minimizing ) configuration by all possible rotations of the plane containing the detector axes . as the initial maximizing case we can choose the most standard configuration@xcite when all detector axes are within @xmath106 plane , @xmath107 and the polar ( zenith ) angles of the detector directions @xmath2 , @xmath3 , @xmath4 , @xmath5 are @xmath108 then all detector configurations with @xmath109 can be parametrized by three euler angles @xcite @xmath110 , @xmath111 , and @xmath112 ( @xmath113 , @xmath114 ) , which describe an arbitrary rotation of the configuration ( [ conf - st-1])([conf - st-2 ] ) . similarly , all minimizing configurations ( @xmath115 ) can be obtained from the standard @xmath106 case [ eq . ( [ conf - st-1 ] ) ] with @xmath116 by arbitrary rotations of this configuration , characterized by three euler angles @xmath117 . any maximally entangled two - qubit state can be obtained from the singlet state by a unitary transformation of the basis of one of the qubits @xcite ( i.e. a one - qubit rotation ) . therefore , an arbitrary case corresponding to the bounds @xmath118 can be reduced to the singlet state considered above by a unitary transformation @xmath87 of the qubit @xmath1 basis and simultaneous corresponding rotation of the detector axes @xmath4 and @xmath5 for the second qubit . since the transformation @xmath87 can also be characterized by three euler angles @xmath119 , @xmath120 , and @xmath121 , an arbitrary situation with @xmath109 can be characterized by six independent parameters @xmath122 , using the standard configuration ( [ conf - st-1])([conf - st-2 ] ) as a starting point . similarly , any situation with @xmath115 is characterized by the same six parameters , starting with the @xmath106 configuration ( [ conf - st-3 ] ) . since these six parameters can describe arbitrary directions of four measurement axes @xmath123 still satisfying the conditions @xmath82 and @xmath83 , it is obvious that any such four - axes configuration produces @xmath109 for exactly one entangled state and also produces @xmath115 for another entangled state . notice that the sign of @xmath78 can obviously be flipped by a @xmath60-rotation of qubit @xmath0 ( or @xmath1 ) around the axis @xmath124 ( @xmath125 ) instead of the @xmath60-rotation ( [ 3.25 ] ) of its detector axes . also notice that six independent parameters for an optimal configuration can be alternatively chosen as any parameters characterizing the four measurement axes , which are pair - wise orthogonal : @xmath82 and @xmath83 . an important special case is the class of `` odd '' maximally entangled states |=(|10 + e^i|01)/ ( 02 ^ -1/20.707 . notice that when the measurement fidelities for the both qubits are the same , @xmath126 , eq . reduces to the threshold fidelity f>0.5 + 2 ^ -5/40.920 , while if the measurement for one of the qubits is ideal ( for example , @xmath127 ) , then the bi violation requires f_b>0.5 + 2 ^ -3/20.854 . let us first analyze the extremum values of @xmath78 for the class of maximally entangled states ( [ 3.18 ] ) . since in this case @xmath128 , we obtain from eq . that for maximally entangled states s=2(f_1^a - f_0^a)(f_1^b - f_0^b ) + ( f_1^a+f_0^a-1)(f_1^b+f_0^b-1)s_0 is directly related to the corresponding quantity @xmath129 in the absence of errors . therefore the extremum values of @xmath78 for maximally entangles states are s_=2(f_1^a - f_0^a)(f_1^b - f_0^b ) 2(f_1^a+f_0^a-1)(f_1^b+f_0^b-1 ) , and they are achieved under the same conditions as discussed in sec . [ iii ] . when the asymmetry of measurement fidelities is similar for both qubits ( @xmath130 and @xmath131 or both inequalities with `` @xmath132 '' sign ) , the first term in eq . is positive , and therefore the bi @xmath133 can be stronger violated for positive @xmath78 than for negative @xmath78 . similarly , if the asymmetries are opposite ( for example , @xmath130 and @xmath134 ) , then it is easier to violate the bi for negative @xmath78 . if fidelities are symmetric at least for one qubit [ eq . ] , then the first term in eq . vanishes , and therefore @xmath135 , as discussed in the previous subsection . if the fidelities are the same for both qubits , @xmath136 and @xmath137 , then the positive @xmath78 is preferable and s_+=2(f_1-f_0)^2 + 2(f_1+f_0 - 1)^2 ( see the dashed lines in fig . [ f1 ] ) . this value of @xmath138 reaches the maximum @xmath139 when both errors vanish ( @xmath140 ) and decreases with the decrease of each fidelity in the interesting region @xmath141 ( more accurately , as long as @xmath142 ) . the condition for the bi violation in this case is ( f_1-f_0)^2+(f_1+f_0 - 1)^2>1 . this threshold of the bi violation on the @xmath143-@xmath144 plane is shown by the lowest dashed line in fig . it is an arc of the ellipse ( corresponding to @xmath145 ) , which is symmetric with respect to the line @xmath146 and is centered at @xmath147 . however , as seen from fig . [ f1 ] , this threshold looks quite close to a straight line on the @xmath143-@xmath144 plane . notice that in the case @xmath148 the threshold reduces to the most well - known condition @xcite f_1>2 - 20.828 , while in the case of symmetric error , @xmath149 , we recover eq .. ( the maximum quantum value of @xmath78 ) versus the measurement fidelities @xmath143 and @xmath144 ( assumed equal for both qubits ) , optimized over all two - qubit states . the dashed lines show the result of @xmath138 maximization over the maximally - entangled states only [ eq . ] . the bell ( chsh ) inequality can be violated when @xmath141 . , width=264 ] to optimize the chsh inequality violation in presence of the measurement errors over all two - qubit states , including non - maximally entangled states , we have used numerical calculations . the analysis has been performed in two different ways ( with coinciding results ) . first , we have searched for the maximum violation by finding extrema of the eigenvalues of the modified bell operator @xmath150 defined by eq . ( [ 4.3 ] ) . since @xmath150 is a hermitian operator and @xmath151 , for fixed measurement directions @xmath123 the maximum and minimum eigenvalues of @xmath150 are equal to the maximum and minimum values of @xmath78 , optimized over the two - qubit states . therefore , optimization of the eigenvalues over the measurement directions gives the extrema of @xmath78 . similar method has been previously used @xcite for the case of identical local errors with @xmath148 , while we apply this method to our more general error model . the numerical maximization ( minimization ) of the largest ( smallest ) eigenvalue of the bell operator @xmath150 has been performed using the software package mathematica . the full optimization should be over all four measurement directions ( @xmath152 , described by 8 angles total . however , because of the invariance of @xmath78 under local transformations , it is sufficient to optimize @xmath150 over only two angles : the angle between @xmath2 and @xmath3 and the angle between @xmath4 and @xmath5 , while the other angles are kept fixed . our numerical results show that in general the optimal values of these two angles are different from each other ; however , for equal fidelity matrices [ eq . ] these angles are equal , so that @xmath153 . this result has been obtained previously @xcite for the special case @xmath148 . our numerical results also show that @xmath154 for positive values of the product @xmath155 and @xmath156 when this product is negative , similar to the result for the maximally entangled states [ see discussion after eq . ] . we have checked that the numerical results for @xmath138 and @xmath157 obtained via optimization of the eigenvalues of the bell operator @xmath150 coincide with the results ( see fig . [ f1 ] ) obtained by our second numerical method based on the direct optimization of @xmath78 . the second method happened to be more efficient numerically ; as another advantage , it provides the optimal measurement directions together with optimal values @xmath138 and @xmath158 , while the bell - operator method gives only @xmath138 and @xmath158 . in principle , direct optimization of @xmath78 ( for fixed measurement fidelities ) implies optimization over the two - qubit density matrix and over 8 measurement directions . however , there is a simplification . it is obviously sufficient to consider only pure states , since probabilistic mixtures of pure states can not extend the range of @xmath78 . moreover , it is sufficient to consider only the states of the form |=(/2)|10+(/2)|01 , since any pure two - qubit states can be reduced to this form by local rotations of the qubits ( which are equivalent to rotations of the measurement directions ) ; this fact is a direct consequence of the schmidt decomposition theorem.@xcite the angle @xmath159 can be limited within the range @xmath160 because the coefficients of the schmidt decomposition are non - negative . this range can be further reduced to @xmath161 since @xmath60-rotation of both qubits about @xmath36-axis ( or any horizontal axis ) exchanges states @xmath162 and therefore corresponds to the transformation @xmath163 . our numerical optimization of @xmath78 within the class of two - qubit states ( [ 4.16 ] ) has shown that for non - zero measurement errors ( we considered @xmath164 ) the optimal measurement directions ( @xmath152 always lie in the same vertical plane [ such configuration is described by eq . ] . this vertical plane can be rotated by an arbitrary angle about @xmath165-axis [ such rotation is equivalent to an overall phase factor in eq . ] ; therefore we can assume @xmath166 in eq . . notice that for the state the vectors @xmath167 and @xmath168 in eq . are along the @xmath165-axis , @xmath169 . these vectors are zero for the maximally entangled state ( @xmath170 ) , then the vertical configuration is no longer preferential ; however , the maximally entangled state is optimal only when there are no measurement errors . for the state ( [ 4.16 ] ) and vertical configuration of the detector axes the expression for @xmath78 has the form & & s=2_-^a_-^b-_+^a_+^b(g - h ) + & & + 2(_+^a_-^b_a -_-^a_+^b_b ) , where & g=&_a_b-_a_b + _ a_b + & & + _ a_b , + & h=&_a_b-_a_b + _ + & & + _ a_b. the numerical maximization and minimization of @xmath78 in this case involves optimization over 5 parameters : @xmath159 , @xmath171 , @xmath172 , @xmath173 , and @xmath174 . nevertheless , in our calculations this procedure happened to be faster than optimization over only two parameters in the method based on the bell operator eigenvalues ( we used mathematica in both methods ) . the solid lines in fig . [ f1 ] show the contour plot of maximum value @xmath138 on the plane @xmath143-@xmath144 for the case when the measurement fidelities for two qubits are equal [ eq . ; in this case @xmath175 . notice that the line for @xmath145 ends at the points@xcite @xmath176 and @xmath177 ( strictly speaking , this line corresponds to @xmath178 since @xmath179 can be easily realized without entanglement ) . the dashed lines , which correspond to the optimization over the maximally entangled states only [ eq . ] , coincide with the solid lines at the points @xmath146 , because in this case the optimum is achieved at the maximally entangled states , as follows from the discussion after eq . ( [ 4.23 ] ) . when @xmath180 , the use of non - maximally entangled states gives a wider range of measurement fidelities allowing the bi violation . however , as seen from fig . [ f1 ] , the difference between the solid and dashed lines significantly shrinks with the increase of @xmath138 , so that there is practically no benefit of using non - maximally entangled states for the bi violation stronger than @xmath181 . notice that the solid and dashed lines in fig . [ f1 ] are symmetric about the line @xmath146 since the interchange @xmath182 does not change @xmath138 , as was discussed after eq . ( [ trsf ] ) . numerical calculations show that in the case of equal fidelity matrices , eq . , the optimal detector configurations for a nonmaximally entangled state have a `` tilted - x '' shape : @xmath183 and @xmath184 . in this case the number of parameters to be optimized in reduced from 5 to 3 , significantly speeding up the numerical procedure . notice that each optimal configuration within the class of states corresponds to a 6-dimensional manifold of optimal configurations , obtained by simultaneous local rotations of the measurement axes and the two - qubit state ( see discussion in sec.[iiib2 ] ) . the detailed analysis of the effects of decoherence will be presented elsewhere . @xcite in this sections we discuss only some results of this analysis , and also discuss the combined effect of local measurement errors and decoherence . to study effects of decoherence we assume for simplicity that the qubit rotations are infinitely fast . thus , we assume that after a fast preparation of a two - qubit state @xmath65 there is a decoherence during time @xmath185 resulting in the state @xmath186 , which is followed by fast measurement of @xmath186 ( including tomographic rotations ) . now @xmath78 is given by eq . ( in the absence of errors ) or and ( in the presence of errors ) where @xmath65 should be substituted by @xmath186 . to obtain @xmath186 we assume independent ( local ) decoherence of each qubit due to zero - temperature environment , described by the parameters @xmath187 and @xmath188 ( here @xmath189 ) where @xmath190 and @xmath191 are the usual relaxation times for the qubit @xmath96 ( @xmath192 ) . as the initial state we still assume the state of the form ( even though in presence of decoherence this state actually does not always provide @xcite the extrema of @xmath78 ) . it can be shown analytically @xcite that in the absence of measurement errors the maximum violation of the bi for the state can be achieved when the detector axes lie in either a horizontal [ eq . ] or vertical [ eq . ] plane ( any other detector configuration can not give stronger violation ) . in the case of only population relaxation ( @xmath193 ) the horizontal configuration is better , while in the case of only @xmath194-effect ( @xmath195 ) the vertical configuration is better . when local measurement errors are considered together with decoherence , the optimal detector configurations may be neither vertical nor horizontal . to elucidate this fact , note that in eq . with @xmath65 replaced by @xmath186 the vectors @xmath167 and @xmath168 remain vertical in the presence of decoherence . as a result , when in the absence of errors the optimal configuration is horizontal , measurement errors may make the optimal detector axes to go out of the horizontal plane . note , however , that for some parameter ranges the vertical and horizontal configurations are still optimal . in numerical calculations we should optimize @xmath78 over 8 parameters : @xmath159 and 7 detector angles ( one of the angles @xmath55 can be fixed because of the invariance of @xmath78 under identical rotations of the qubits around the @xmath165 axis ) . we have performed such optimization for a few hundred parameter points , choosing the measurement fidelities @xmath196 and decoherence parameters @xmath197 and @xmath198 randomly from the range ( 0.7 , 1 ) . for many ( more than half of ) parameter points the optimal configuration was still found to be either vertical or ( in much smaller number of cases ) horizontal . even when the optimal configuration was neither vertical nor horizontal , we found that restricting optimization to only the vertical and horizontal configurations gives a very good approximation of the extrema @xmath79 ( within 0.01 for all calculated parameter points ) . such restriction significantly speeds up the calculations , since we need to optimize over only 5 parameters instead of 8 parameters . assuming initial state and replacing @xmath65 by @xmath186 in eq . we obtain & s=&2_-^a_-^b+_+^a_+^b\{[1-_a-_b- ( _ a-_b)]g + & & + _ a_bh } + 2_+^a_-^b(_a+_a-1)_a + & & + 2_-^a_+^b(_b-_b-1)_b , when the detector axes are in a vertical plane [ @xmath199 and @xmath200 are defined in eq . ] , and & & s=2_-^a_-^b+_+^a_+^b_a_b for a horizontal detector configuration . to find the extrema of @xmath78 within the class of vertical configurations , eq . ( [ 6.4 ] ) should be numerically optimized over the parameter @xmath159 and four angles @xmath57 . the optimization of @xmath78 within the class of horizontal configurations is much simpler , because the term in the square brackets in eq . can be optimized independently of @xmath159 . this optimization is exactly the same as in the ideal case [ see eqs . and with @xmath201 , therefore the term in the square brackets has extrema @xmath202 , and therefore the extrema of eq . are reached at @xmath203 ( maximally entangled state ) , thus yielding a rather simple formula s_=2 _ -^a_-^b2 _ + ^a_+^b_a_b , which depends only on the @xmath194-relaxation and measurement fidelities . versus the measurement fidelities @xmath143 and @xmath144 for the initial state in presence of decoherence with @xmath204 and @xmath205 , in absence ( solid lines ) or presence ( dashed lines ) of the symmetric crosstalk with @xmath206 . solid and dashed lines coincide for @xmath145 . , width=264 ] for the numerical example shown by solid lines in fig . [ f5 ] we assume that decoherence is identical for the two qubits and choose @xmath204 and @xmath205 , that corresponds to realistically good values for phase qubits : @xmath207 ns , @xmath208 ns , and @xmath209 ns . we also assume identical errors for both qubits , eq . , which implies @xmath210 ( as in the absence of decoherence ) , so in fig . [ f5 ] we show the contour plot only for @xmath138 . we have found that for these decoherence parameters the vertical detector configuration is better than any other configuration [ assuming initial state ] for any measurement fidelities in the analyzed range ( @xmath211 ) . notice that for the assumed decoherence parameters @xmath212 in the absence of measurement errors , and the bi violation requires @xmath213 ( for @xmath149 ) , that should be compared to the threshold @xmath214 in absence of decoherence . let us mention that the error model previously discussed in relation to the bi violation in the two - photon interference @xcite can be shown to be formally equivalent to the special case of our model with pure dephasing ( @xmath195 ) and identical errors with @xmath148 . then our quantities @xmath144 and @xmath215 correspond , respectively , to the detector efficiency and visibility in ref . . the case of pure dephasing in the absence of errors has been also considered in connection with the bi violation in mesoscopic conductors . the nonidealities discussed above are common for many types of qubits . now let us discuss a more specific type of error : the measurement crosstalk for josephson phase qubits . @xcite the crosstalk error originates from the fixed ( capacitive ) coupling between the qubits , which is still on in the process of measurement . the mechanism of the crosstalk is the following.@xcite if the measurement outcome for the phase qubit @xmath0 is `` 1 '' , then this qubit is physically switched to a highly excited state ( outside of the qubit hilbert space ) , and its dissipative oscillating evolution after the switching affects the qubit @xmath1 . as a result , the extra excitation of the qubit @xmath1 may lead to its erroneous switching in the process of measurement , so that instead of the measurement outcome `` 1,0 '' we may get `` 1,1 '' with some probability @xmath216 . similarly , because of the crosstalk from the qubit @xmath1 to the qubit @xmath0 , we may obtain the measurement result `` 1,1 '' instead of `` 0,1 '' with some probability @xmath217 . the values of @xmath216 and @xmath217 significantly depend on the timing of the measurement pulses applied to the qubits.@xcite if the qubit @xmath0 is measured few nanoseconds earlier than the qubit @xmath1 , then @xmath218 ; if the qubit @xmath1 is measured first , then @xmath219 . in the case when the measurement pulses are practically simultaneous , the crosstalk probability becomes significantly lower and @xmath220 . let us model the crosstalk in the following simple way . even though physically the crosstalk develops at the same time as the measurement process and its description is quite nontrivial , @xcite we will assume ( for simplicity ) that the crosstalk effect happens after the `` actual '' measurement ( characterized by measurement fidelities as in sections [ iv ] and [ vi ] ) , so that the only effect of the crosstalk is the change of the outcome `` 1,0 '' into `` 1,1 '' with probability @xmath216 and the change of the outcome `` 0,1 '' into `` 1,1 '' with probability @xmath217 . moreover , we assume that the probabilities @xmath216 and @xmath217 do not depend on the measurement axes ( @xmath2 , @xmath4 , etc . ) . notice that the measurement crosstalk obviously violates the fundamental assumption of locality , on which the bi is based ( so , strictly speaking the bi approach is not applicable in such situation ) . in this section we discuss the modification of the classical bound for @xmath78 , taking the crosstalk into account ( this bound is now model - dependent , in contrast to the usual bi ) , and we also analyze the effect of the crosstalk on the quantum result for @xmath79 . besides that , we discuss a simple modification of the experimental procedure , which eliminates the effect of crosstalk by using only the `` negative result '' outcomes . first , let us briefly review the derivation of the chsh inequality , presented in ref . . in a local realistic theory s = s ( ) f ( ) d , where @xmath221 is a distribution of the hidden variable @xmath222 and & s ( ) = & a(,)b(,)- a(,)b(, ) + & & + a(,)b(,)+ a(,)b(, ) . here the measurement outcomes @xmath223 and @xmath224 can take only the values @xmath74 , depending on the hidden variable @xmath222 and the detector orientations for the qubits @xmath0 and @xmath1 . ( notice that in sections [ iii ] and [ iv ] the notation @xmath72 and @xmath73 has been used for operators ; now they are classical quantities . also notice that the outcome value @xmath41 is associated with the result `` 0 '' . ) it is easy to check that @xmath225 under the _ locality assumption _ : the result @xmath223 does not depend on the orientations @xmath4 of the qubit @xmath1 and vice versa ; similarly , @xmath221 does not depend on @xmath2 and @xmath4 . after integration ( [ 7.1 ] ) , this leads to the chsh inequality . in our model the measurement crosstalk can not change the positive product of outcomes @xmath226 ; however , it changes @xmath227 into @xmath226 with the probability @xmath228 if @xmath229 or with probability @xmath230 if @xmath231 . ( the locality assumption is obviously violated , since the value of @xmath72 now depends not only on @xmath222 and @xmath2 but also on @xmath73 and thus implicitly on @xmath4 ; similarly for @xmath73 . ) notice that we assume that the crosstalk probabilities @xmath216 and @xmath217 may in principle depend on @xmath222 ( this is a slight generalization of the more natural @xmath222-independent model for @xmath216 and @xmath217 ) . the random change from @xmath227 to @xmath226 due to the crosstalk leads to the modification of the chsh inequality . here we mention only the main points of the derivation of the modified chsh inequality ; the details are in the appendix . we start with fixing @xmath222 and considering all possible changes of the quantity @xmath232 in eq . due to the crosstalk , thus obtaining the new values of @xmath233 together with their probabilities for each of 16 realizations of the vector @xmath234 . it is easy to see that due to the crosstalk @xmath232 can get the values @xmath235 , which are outside of the limits @xmath236 . averaging the value @xmath237 over the crosstalk scenarios and then maximizing and minimizing the result over 16 realizations of the vector @xmath238 , we get -2 + 4\{p_c^a ( ) , p_c^b ( ) } s ( ) 2 + 2|p_c^a()-p_c^b()| . finally averaging this result over @xmath222 , we obtain the modified chsh inequality : -2 + 4\{p_c^a , p_c^b}s2 + 2|p_c^a - p_c^b| , which is the main result of this subsection . let us consider two special cases . for a symmetric crosstalk , @xmath239 , the inequality becomes -2 + 4p_cs2 , while for a fully asymmetric crosstalk , @xmath240 , it becomes -2s2 + 2p_c^a . ( similarly for @xmath241 ) . it is interesting to notice that the inequality in the symmetric case is more restrictive than the bi ( `` easier '' negative bound and no change of the positive bound ) . this is actually quite expected because in the limiting case @xmath242 the crosstalk makes all @xmath243 products equal 1 , so that @xmath179 always , as also follows from eq . . in contrast , for the fully asymmetric crosstalk the inequality is less restrictive than ( `` harder to violate '' positive bound and no change of the negative bound ) . in the case of a finite crosstalk asymmetry , both the positive and negative bounds change [ eq . ( [ 6.6 ] ) ] . let us emphasize that in contrast to the derivation of the original chsh inequality , @xmath244 may be significantly outside of the range @xmath245 ; it can get the values @xmath246 for any ( symmetric or asymmetric ) crosstalk . so the fact that the lower bound for @xmath78 never decreases and the upper bound increases only slightly for small crosstalk , is due to the statistical averaging of the random increase and decrease of @xmath244 due to the crosstalk . in the quantum case the crosstalk changes the measurement probabilities @xmath247 as & & p_00^c = p_00 , p_10^c=(1-p_c^a)p_10 , p_01^c=(1-p_c^b)p_01 , + & & p_11^c = p_11+p_c^ap_10+p_c^bp_01 . then using the definitions and we obtain that the measured value of @xmath78 becomes s^c=2_c+(1-_c)s+2(p_c^a - p_c^b)[p_b()-p_a( ) ] , where @xmath248 , while @xmath78 , @xmath14 , and @xmath249 are the quantities obtained in the absence of crosstalk [ @xmath14 and @xmath249 are defined after eq . ] . in a general case the maxima and minima @xmath250 of eq . can be found numerically . to estimate the effect of the crosstalk , let us calculate eq . for a maximally entangled state in the absence of errors and decoherence . then @xmath251 , and using @xmath118 we obtain s^c_+=2-(2 - 2)_c , s^c_-=-2+(2 + 2)_c . as we see , both extrema are affected by the crosstalk , making the range narrower from both sides ; however , the lower boundary is affected much stronger than the upper boundary . comparing eq.([6.10 ] ) with the modified chsh inequality ( [ 6.6 ] ) , we see that the lower bound shifts up for the quantum result always faster than for the classical bound ; therefore the gap between the quantum and classical bounds always shrinks due to the crosstalk . the classical - quantum gap at positive @xmath78 shrinks from both sides due to the crosstalk . in the case of symmetric crosstalk , @xmath239 , we can easily consider non - maximally entangled states , measurement errors and decoherence , since eq . in this case reduces to @xmath252 . therefore , @xmath250 are simply related to the values @xmath79 without crosstalk ( but with measurement errors and decoherence ) : s^c_=2p_c+(1-p_c)s _ [ similar dependence was used in eq . ( [ 6.10 ] ) ] . a violation of the upper bound of the modified chsh inequality can be observed when @xmath253 , which yields @xmath141 , while a violation of the lower limit in eq . requires @xmath254 , which yields @xmath255 . quite surprisingly , the symmetric crosstalk does not change the conditions for the bi violation . ( of course , the violation of the increased lower bound is not as convincing psychologically as the violation on the increased upper bound . ) let us discuss the combined effect of local errors , decoherence , and symmetric crosstalk for the numerical example considered in sec.[vi ] , assuming symmetric crosstalk with @xmath206 . now for the state in the absence of local measurement errors ( @xmath140 ) we get @xmath256 , which is slightly less than the value @xmath257 obtained in the absence of the crosstalk . the dependence of @xmath258 on the measurement fidelities @xmath143 and @xmath144 in this case is illustrated by the dashed lines in fig.[f5 ] . in accordance with the above discussion , the solid line for @xmath145 and the dashed line for @xmath259 coincide ( the bi violation boundary is not affected ) , while the comparison of the solid and dashed lines for @xmath260 and @xmath261 shows that the crosstalk makes an observation of a given bi - violating value of @xmath138 more difficult . figure [ f6 ] shows a similar contour plot on the @xmath143-@xmath144 plane for the lowest quantum value @xmath262 assuming the same parameters as in fig . comparing figs . [ f5 ] and [ f6 ] we see that it is more difficult to violate the lower classical bound ( even though it is now only @xmath263 ) than the upper classical bound of 2 . this is because we assumed the same measurement fidelities for both qubits , that generally shifts the quantum result up [ as in eq . ( [ 25 ] ) ] . notice that in fig.[f6 ] the reference bound of @xmath264 can still be violated , though only for almost perfect fidelities . versus the measurement fidelities @xmath143 and @xmath144 ( same for both qubits ) optimized over the states . we assume symmetric crosstalk with @xmath206 and decoherence parameters @xmath204 and @xmath205 ( same as for fig . [ f5 ] ) . the classical bound is shifted from @xmath264 to @xmath263 by the crosstalk . , width=264 ] slight modification of the chsh inequality ( [ 3 ] ) can make it insensitive to the crosstalk . the main idea is to use only experimental outcomes with the result `` 0 '' , when a qubit does not switch , and therefore the crosstalk does not occur . such `` negative - result '' ( `` null - result '' ) experiments with the josephson phase qubits are very interesting from the quantum point of view;@xcite however , here we are interested only in the classical consequence ( or rather absence of it ) for a measurement with a null result . instead of the inequality ( [ 3 ] ) let us use the equivalent inequality ( [ 6 ] ) , and let us change the definition of @xmath265 in eq . ( [ 7 ] ) by interchanging the measurement outcomes `` 1 '' and `` 0 '' . then by symmetry the inequality ( [ 6 ] ) is still valid , so we get the classical bounds @xmath266 for @xmath267 where the probability @xmath268 is for measuring the qubit @xmath0 only ( without measuring the qubit @xmath1 ) , while @xmath269 is for measuring only the qubit @xmath1 . with this simple modification , the chsh inequality becomes insensitive to the mechanism of the measurement crosstalk@xcite considered in this section . notice , however , that in performing experiment in this way it is still important to check the absence of a direct crosstalk ( due to the measurement pulse itself ) . this can be done by applying a measurement pulse to the well - detuned qubit @xmath1 ( so that it can not switch ) and checking that this does not affect switching probabilities for the qubit @xmath0 ( and similarly for @xmath0 interchanged with @xmath1 ) . in this paper we have considered the conditions for the violation of the bell inequality in the chsh form@xcite for the entangled pairs of solid - state qubits , when instead of the rotation of optical polarizers ( detectors ) we have to rotate the states of two qubits before the measurement , which itself is always performed in the logical @xmath165-basis ( @xmath39,@xmath38 ) for each qubit . while most of our results are applicable to many types of qubits , we have focused on the experiments with the josephson phase qubits . @xcite we have analyzed the bi violation for the ideal case as well as in presence of various nonidealities , including local measurement errors , local decoherence , and measurement crosstalk . in the ideal case the maximum violation of the bi ( @xmath270 , while the classical bound is @xmath133 ) can be realized for any maximally entangled state . the optimal configuration of the measurement directions in this case can be realized with three degrees of freedom for each maximally - entangled state ( the measurement direction in our terminology actually refers to the qubit rotation before the measurement ) . however , in presence of nonidealities there is typically less freedom in choosing the optimal configuration . for the `` odd '' two - qubit states involving superpositions of the states @xmath271 and @xmath272 , we have focused on the `` vertical '' measurement configurations , for which all measurement directions ( @xmath152 are within the same vertical plane of the bloch sphere , and the `` horizontal '' configuration , for which the four measurement axes are within the @xmath36-@xmath273 plane . the qubit measurement with finite local errors ( characterized by the fidelities @xmath143 and @xmath144 for each qubit ) shrinks the quantum range for @xmath78 . we have found that for a maximally entangled state the bi violation is still possible when the classical bounds @xmath236 are exceeded by the extrema of the quantum result given by eq . ( [ 25 ] ) . in particular , when two qubits have the same fidelities , the violation condition is given by eq . ( [ 27 ] ) and shown by the lowest dashed line in fig . [ f1 ] ; it can be crudely approximated by the condition @xmath274 . a significantly softer violation condition can be obtained when allowing the two - qubit state to be non - maximally entangled ; @xcite this condition is shown by the lowest solid line in fig . however , the trick of using a non - maximally - entangled state does not help much when we need a bi violation with a significant margin ( not just barely ) ; this can be seen by comparing the solid and dashed lines in fig . [ f1 ] . for non - maximally entangled odd two - qubit states in presence of local measurement errors , the vertical measurement configuration is found to be preferable in comparison with other configurations . analyzing the effect of local decoherence of the qubits for the odd two - qubit states , we have found that either vertical or horizontal configuration of the measurement directions is optimal , depending on the parameters . in particular , in the case of population ( energy ) relaxation in @xmath165-basis , the horizontal configuration is optimal , while for pure dephasing the vertical configuration is optimal . in presence of both the decoherence and local measurement errors , the optimal configuration can be neither horizontal nor vertical ; however , restricting optimization to only these two classes of configurations gives a very good approximation of the extrema @xmath79 . obviously , both the decoherence and the measurement errors make the observation of the bi violation more difficult . we have also analyzed the effect of the measurement crosstalk @xcite which plays an important role in measurement of the capacitively - coupled phase qubits . since the crosstalk is a mechanism of classical communication between the qubits , strictly speaking the bi is inapplicable . however , for a particular model of the crosstalk it is possible to derive a modified chsh inequality [ see eq . ( [ 6.6 ] ) ] . in particular , we have found that the symmetric crosstalk does not change the upper classical bound but increases the lower classical bound . the crosstalk also affects the quantum bounds , which are given by eq . ( [ 6.10 ] ) for the maximally entangled state in the otherwise ideal case with arbitrary crosstalk and by eq . ( [ 6.9 ] ) for an arbitrary case but assuming a symmetric crosstalk . quite unexpectedly , the symmetric crosstalk does not change the threshold condition for the observation of the bi violation . however , the crosstalk always reduces the gap between the classical and quantum bounds and makes an observation of the bi violation with a finite margin more difficult . it is important to mention that the detrimental effect of the crosstalk can be eliminated by a slight change of the chsh inequality ( by using only negative - result outcomes ) , which makes it insensitive to the crosstalk [ see eqs . ( [ crt-1 ] ) and ( [ crt-2 ] ) ] . we have performed the numerical simulations with the parameters similar to the experimental values for the best present - day experiments with the josephson phase qubits . @xcite our results ( see fig . [ f5 ] ) show the possibility of the chsh inequality violation with a significant margin even without further improvement of the phase qubit technology . we thank qin zhang , john martinis , nadav katz , and markus ansmann for useful discussions . the work was supported by nsa and dto under aro grant w911nf-04 - 1 - 0204 . to derive the inequality , we fix @xmath222 and use the abbreviated notation @xmath275 , @xmath276 , @xmath277 , @xmath278 and similarly for @xmath73 and @xmath279 . let us introduce the vectors @xmath234 and @xmath280 , so that @xmath281 . the vector @xmath238 can assume 16 values , whereas without the crosstalk @xmath282 can take 8 values , since the number of pluses or minuses in @xmath282 is even . each pair of @xmath238 and @xmath283 yields one value of @xmath284 . generally , the crosstalk effect differs for @xmath238 and @xmath283 , except for the symmetric case , @xmath239 , when crosstalk depends only on @xmath282 . we start the analysis with the symmetric case and then consider the general asymmetric case . it is easy to check that without crosstalk @xmath285 or @xmath264 . to obtain the upper bound of the modified bi , we consider four values of @xmath282 corresponding to @xmath285 . the crosstalk can not change @xmath286 , hence we discuss three other values : @xmath287 , and @xmath288 . for any of these vectors , the crosstalk makes @xmath233 to take the values 0 , 2 , and 4 with the probabilities @xmath289 , and @xmath290 , respectively , where @xmath291 . indeed , the change of any @xmath41 to 1 in @xmath282 occurs with the probability @xmath290 , yielding @xmath292 for the change of the first @xmath41 in @xmath293 and @xmath294 for the change of any other @xmath41 , whereas @xmath295 is the probability of no change or change to @xmath296 , both cases yielding @xmath285 . thus , though now @xmath233 can achieve the maximal mathematically possible value 4 , it is easy to see that the maximal value for the average of @xmath233 over crosstalk is still @xmath297 . to obtain the lower limit of the inequality , we consider four values of @xmath282 corresponding to @xmath298 in the absence of the crosstalk . consider first @xmath299 . this vector can be changed by the crosstalk to any of 16 possible combinations of four pluses and minuses , yielding the values @xmath300 with the probabilities @xmath301 , respectively ( note that the sum of the above probabilities equals 1 ) . the above probabilities can be easily obtained if one takes into account that @xmath302 results from the change of only the first @xmath41 in @xmath282 , @xmath298 occurs when either @xmath282 have not changed or the first and one of the last 3 components have changed , @xmath292 occurs when the crosstalk results in @xmath282 with the first and two other components equal to 1 or @xmath41 , @xmath285 occurs when two of the last three components or all the components of @xmath282 change the sign , and @xmath294 results from the changes of all the last three components in @xmath282 . as a result , in the case when only @xmath299 is realized , we obtain s=-2 + 4p_c . finally , let us consider the values @xmath303 , @xmath304 , and @xmath305 . for any of these vectors the crosstalk makes @xmath233 to take the values @xmath264 , 0 , 2 with the probabilities @xmath306 , and @xmath307 , respectively . these probabilities follow from the fact that @xmath298 when @xmath293 is not changed , @xmath292 results from the change of only one component @xmath41 , and @xmath285 results from the change of the both negative components . it is easy to check that for any of the three above vectors , we again obtain eq . . as mentioned above , when @xmath309 , the crosstalk yields different results for @xmath238 and @xmath283 . however , recalling that the change @xmath310 occurs with the probability @xmath216 if @xmath311 and @xmath312 or with the probability @xmath217 if @xmath313 and @xmath314 , we can obtain a symmetry relation for the probability of a change of @xmath315 with a given @xmath238 to some @xmath316 as a function of @xmath216 and @xmath217 : p_a(p_c^a , p_c^b , c ) = p_a(p_c^b , p_c^a ,- c ) . to extend the results of sec . [ a1 ] to the general case of asymmetric crosstalk , we should reconsider all possible changes of @xmath317 discussed in sec . [ a1 ] with the account of the two vectors @xmath238 and @xmath283 corresponding to each @xmath317 . as a result , the probability of each value of @xmath233 obtained above is replaced by two probabilities , which differ from each other by the change @xmath318 [ cf . eq . ] . it happens that for most combinations @xmath282 the probability obtained in sec . [ a1 ] is replaced by two probabilities just by replacing @xmath319 with @xmath216 or with @xmath217 . one of two exceptions is the case @xmath320 . then for @xmath321 the quantity @xmath285 changes to @xmath322 with the probabilities @xmath323 , and @xmath324 , respectively , where @xmath325 . this yields @xmath326 . the other value of @xmath327 [ corresponding to @xmath328 is obtained from this formula by substitution @xmath318 , so that @xmath329 . the other exception is @xmath330 . in this case we obtain @xmath331 ( for both possible values of @xmath238 ) . finally , by choosing the worst cases for the lower and upper bounds for @xmath332 and averaging the result over @xmath222 , we obtain eq . . z. y. ou and l. mandel , phys . lett . * 61 * , 50 ( 1988 ) ; y. h. shih and c. o. alley , ibid . * 61 * , 2921 ( 1988 ) ; p. r. tapster , j. g. rarity , and p. c. m. owens , ibid . * 73 * , 1923 ( 1994 ) ; p. g. kwiat , k. mattle , h. weinfurter , a. zeilinger , a. v. sergienko , and y. shih , ibid . * 75 * , 4337 ( 1995 ) ; w. tittel , j. brendel , h. zbinden , and n. gisin , ibid . * 81 * , 3563 ( 1998 ) ; g. weihs , t. jennewein , c. simon , h. weinfurter , and a. zeilinger , ibid . * 81 * , 5039 ( 1998 ) . this parametrization is performed as follows . in view of eq . , any maximizing detector configuration is determined by @xmath2 and @xmath3 , and hence can be equivalently described by a right - handed system of axes @xmath333 such that @xmath334 . this configuration can be obtained from the standard configuration ( [ conf - st-1])([conf - st-2 ] ) ( with the corresponding frame @xmath335 and @xmath336 ) by the axes rotation described by the euler angles @xmath110 , @xmath111 , and @xmath112 . this rotation is performed in three stages : ( i ) the rotation around @xmath337 by @xmath110 , ( ii ) the rotation around the new position of @xmath338 by @xmath111 , ( iii ) the rotation around @xmath339 by @xmath112 . our measurement error model for the qubit @xmath96 can be interpreted as the exchange of the measurement results @xmath344 with probability @xmath345 and the predetermined result with probability @xmath346 ( result 0 if @xmath347 or 1 if @xmath347 ) , while the `` correct '' operation is realized with the remaining probability @xmath348 . since the bound @xmath349 obviously remains valid in all the cases when one or both qubits operate in the mode of result exchange or in the mode of predetermined result , and since the erroneous operation modes are independent of the correct results , the averaging over these modes ( for each qubit ) can not invalidate the bound @xmath349 .

from 2mass infra - red photometry we find two red clump ( rc ) populations co - existing in fields toward the galactic bulge at latitudes.5 , ranging over in longitude and 20 in latitude . these rc peaks indicate two stellar populations separated by.3 kpc ; at ()=( ) the two rcs are located at 6.5 and 8.8.2 kpc . the double - peaked rc is inconsistent with a tilted bar morphology . most of our fields show the two rcs at roughly constant distance with longitude , also inconsistent with a tilted bar ; however , an underlying bar may be present . stellar densities in the two rcs changes dramatically with longitude : on the positive longitude side the foreground rc is dominant , while the background rc dominates negative longitudes . a line connecting the maxima of the foreground and background populations is tilted to the line of sight by , similar to claims for the tilt of a galactic bar . the distance between the two rcs decreases towards the galactic plane ; seen edge - on the bulge is x - shaped , resembling some extra - galactic bulges and the results of n - body simulations . the center of this x is consistent with the distance to the galactic center , although better agreement would occur if the bulge is 23 gyr younger than 47 tuc . our observations may be understood if the two rc populations emanate , nearly tangentially , from the galactic bar ends , in a funnel shape . alternatively , the x , or double funnel , may continue to the galactic center . from the sun this would appear peanut / box shaped , but x - shaped when viewed tangentially . 0.5 in

1008.0519

following the detection of an apparent h i bar within 2kpc of the galactic center , by liszt & burton ( 1980 ) , blitz & spergel ( 1991 ) found that the 2.4@xmath7 imaging data of matsumoto et al . ( 1982 ) showed the presence of a tilted bar in the galactic bulge . that part of the bar closest to the sun was at positive galactic longitudes . these conclusions were primarily based on the observed fluxes for latitudes @xmath8=39@xmath1 and longitudes at @xmath9=0,@xmath510@xmath1 . stanek et al . ( 1994 , 1997 ) studied v and i photometry of red clump ( rc ) stars toward the bulge at longitudes of @xmath9@xmath21,@xmath55@xmath1 ( latitudes near @xmath10=@xmath114@xmath1 ) and found a systematically fainter rcs from positive to negative longitude . this was interpreted , and modelled , as a distance effect which was fit with a tri - axial structure , or tilted bar , with the near - side at positive galactic longitudes . later work on the cobe / dirbe near and far infrared sky brightnesses data by dwek et al . ( 1995 ) and binney , gerhard , & spergel ( 1997 ) confirmed the existence of the tilted galactic bar . distances from rc stars , studied for fields at various positions , from babusiaux and gilmore ( 2005 ) , nishiyama et al . ( 2005 , 2006 ) , and rattenbury et al . ( 2007 ) confirm the general picture of a tilted galactic bar . in particular , nishiyama et al . ( 2005 , 2006 ) claimed to find a secondary inner bar . the recent radial velocity study of bulge m giants stars , by howard et al . ( 2008 , 2009 ) , found rotation that is best fit with bar models . in particular , they find that the bulge at @xmath118@xmath12 and @xmath114@xmath12 rotates cylindrically , as do boxy bulges of other galaxies . recently , mcwilliam et al . ( 2010 ) and zoccali ( 2010 ) independently found evidence for two red clumps toward the galactic bulge region . in the current paper we outline and expand on the evidence of this joint discovery and map the extent of the double red clump over the bulge region . the data discussed here come from the following four photometric catalogs available in the literature : _ i ) _ the 2mass point source catalog ( skrutskie et al . 2006 ) . _ ii ) _ optical @xmath13 wfi photometry at the 2.2 m eso / mpg telescope at la silla , for a @xmath14 arcmin field at @xmath15 ( zoccali et al . 2003 ) . _ iii ) _ near ir @xmath16 photometry from sofi at the ntt in la silla , for a @xmath17 arcmin field centered at @xmath18 ( zoccali et al . 2003 ) . _ iv ) _ the ogle @xmath19 maps of the galactic bulge , from udalski et al . ( 2002 ) . in figure [ cmd_b6 ] we show the color magnitude diagram ( cmd ) of the bulge region centered approximately at @xmath20 , along the minor axis , from four independent photometric data sets , all of them showing a double peaked horizontal branch red clump . the upper panels show the wfi ( left ) and ogle ( right ) optical cmds together with the luminosity functions ( lfs ) of rgb stars falling inside the diagonal strip marked in the cmd . the reddening free magnitude @xmath21 magnitude is defined as @xmath22 . ] was used , following rattenbury et al . ( 2007 ) , and adopting the extinction values given in sumi et al . ( 2004 ) for the ogle fields . the ogle field shown here is the number 7 , centered at @xmath23 . it is worth noticing that rattenbury et al . ( 2007 ) data also show a double peaked red clump in this field ( c.f . , their fig . 5 ) . the red thin histograms in both panels show the lf of the ogle field centered in baade s window ( sc46 ) , where the red clump is significantly narrower and uni - modal . the lower panels of figure [ cmd_b6 ] show the near ir cmds and lfs from the sofi photometry ( left ) and the 2mass point source catalog ( right ) . again , the red clump is double peaked in this field . the lf for baade s window is also shown as a red thin histogram . the latter was obtained by selecting a 0.8 square degree field , centered at at @xmath24 , from the 2mass point source catalog . unfortunately at @xmath25 the completeness of the 2mass catalog drops abruptly below the red clump , therefore the comparison is not as conclusive as the one for the ogle catalogs . the 2mass color - magnitude diagram in figure [ fig - cmd-10 - 80 ] shows the two rcs at ( @xmath3)=(@xmath111,@xmath118 ) , which appear tilted , consistent with the predicted metallicity - dependence of the teramo isochrones ( pietrinferni et al . 2004 ) . the figure also clearly shows that the two rcs possess similar range and mean ( j@xmath11k ) colors ; as in figure 1 this suggests that the reddening is foreground and small , and that the two rc populations possess a similar range and mean metallicities . the mean ( j@xmath11k)@xmath26 color differences between bright and faint rcs for the 16 fields in this paper that contain both rcs is [email protected] magnitudes . thus , the colors are practically identical , but in all cases the brighter rc is slightly redder than the faint rc . this systematic difference is in the right sense and size to be due to the change in color of the background red giant branch population between the bright and faint rcs . a small metallicity difference ( about 0.1 dex ) , between bright and faint rcs could produce the color shift , but it would still be necessary to subtract the red giant branch background effect ; thus , any metallicity difference must be smaller than @xmath20.1 dex . figure [ cmd_b6 ] and [ fig - cmd-10 - 80 ] suggests that the double clump might be due to the presence of two populations at two different distances . a magnitude difference of @xmath27 at @xmath28 kpc would correspond to a distance difference of @xmath29 kpc . we note here for the first time that in the outer bulge , along the minor axis , bright and faint red clumps coexist , as if the near and far side of the bar both extend towards the minor axis . figure [ cmd_b6 ] demonstrates that the horizontal branch red clump in the field at @xmath20 is significantly broader than the one of baade s window at @xmath25 . if the double - peaked rcs are due to the distances of two populations , the observations immediately appear inconsistent with a single tilted bar . it is , therefore , important to ask whether the double - peaked rcs could have resulted from stellar evolution , or to effects other than distance . a few points can be addressed by looking at these figures . first of all , the double clump is real . it is not an artifact of bad photometry , such as a bad match of mosaic data , because it is present in several independent catalogs . we have checked that it is present in each of the 8 chips of the wfi mosaic . second , the two peaks can not be due to the rgb bump ( nor the agb bump ) falling close to the rc because in that case the two would also occur in baade s window . also , as we will see in figures [ fig - lon ] & [ fig - lon7 ] , the relative strength of the two rc peaks changes dramatically with longitude , while the population ratios of rgb bump , agb bump and hb red clump depend on the evolutionary time of these phases , and therefore should be independent of the line of sight . the double clump can not be due to two extinction patches in this particular direction , because in that case the separation between the two peaks would be smaller in the @xmath30 band , compared to the @xmath31 band . instead , the separation is roughly comparable in all the filters . in addition , as mentioned above , the two peaks look very similar in the 8 chips of the wfi mosaic . this would not be the case if they were due to extinction patches . finally , were extinction responsible for the two peaks , the clump that is @xmath32 magnitudes fainter would also be @xmath33 magnitudes redder than the other one in @xmath34 , which is clearly not the case . the predicted rc@xmath11rgb bump i - band magnitude differences from the teramo stellar evolution code ( e.g. , pietrinferni et al . 2004 ) increase with metallicity , and a match to the observed difference between our putative rcs occurs for 12gyr isochrone at solar metallicity . however , the predicted rc numbers exceed those of the rgb bump by roughly a factor of 10 . this presents a particularly severe problem for the negative longitude fields ( see figure [ fig - lon ] ) , where the faint component dominates , so the proposed rgb bump population exceeds the rc , completely at odds with current stellar evolution ideas . the rgb bump / rc ratio for the negative longitude bulge fields of figure [ fig - lon ] exceed the stellar evolution predictions by more than a factor of 30 . this might be qualitatively understood if metal - rich bulge stars largely terminate their evolution prematurely , after the rgb bump but before the rc , say if their metallicity is higher than currently realized and they experienced significantly enhanced mass - loss . however , as noted above , it would then be necessary to understand why stellar evolution on the giant branch is so different on the positive and negative longitude sides of the bulge ; as a result there would still be two separate bulge populations . the similarity of the j@xmath11k colors of the two populations suggests that there is very little metallicity difference . given these difficulties , the idea that our two rcs are due to rc plus rgb bump is not a tenable hypothesis . it is likely that the rgb bump is detected as a very small peak in some of the panels of figure [ fig - lon ] ; the main effect of the rgb bump is to reduce the clarity of the fainter rc peaks . theoretical isochrone rc k - band magnitudes and ( j@xmath11k ) colors display metallicity - dependence : more metal - rich rc stars are redder in ( j@xmath11k ) and brighter in k. this effect is clearly evident in the teramo group stellar evolution models ( e.g. , pietrinferni et al . 2004 ) . the theoretical predictions show that rc m@xmath35 is brighter for younger ages ; however , no single age can account for the observed 0.65 mag . difference in rc k@xmath36 seen of figure [ fig - cmd-10 - 80 ] and [ fig - lon ] for solar - metallicity isochrones , unlike the case for the i - band . the closest that the theoretical rc predictions come to to the observations , at solar composition , is for ages at 2 and 14 gyrs , but this only accounts for @xmath37m@xmath35 of 0.28 mag . a combination of age and metallicity effects could explain the brightness of the observed rc peaks , but this requires unrealistic ages and a mean [ fe / h ] near @xmath110.5 dex , which is in conflict with measured bulge abundances . for latitude @xmath118@xmath12 the age explanation would require the majority of stars at positive longitudes to be 2gyr old and the majority of stars at negative longitudes to be 14 gyr , with a roughly equal mixture of both at longitudes near @xmath9@xmath20 . however , this is excluded by the age data of zoccali et al . ( 2003 ) , at ( @xmath9,@xmath10)=(0.3,@xmath116.2 ) , and clarkson et al . ( 2008 ) at ( @xmath9,@xmath10)=(1.3,@xmath112.7 ) . both studies find an old bulge @xmath3810 gyr with almost no trace of a younger population . again , this would not explain how two vastly different age stellar populations could be maintained on opposite sides of the galactic center . we conclude that it is not possible to account for the observed k - band rc magnitudes with two stellar populations of different ages at a single distance . thus , we attribute the observed differences in rc magnitude as primarily a distance effect . we note the similarity of our bulge - wide double rc to the double rc in the bulge globular cluster terzan 5 , found by ferraro et al . it is tantalizing that the galactic longitude of terzan 5 puts it very close to the peak of our bright rc ( but closer to the plane than our fields ) , and that ferraro et al.s preferred terzan 5 distance , at [email protected] kpc , is very similar to our estimated distance for the bright rc , at 6.5 @xmath50.2 kpc . ferraro et al . ( 2009 ) fit the rc region using two isochrones , one at 12 gyr with [ fe / h][email protected] dex , and a brighter one with an age of 6 gyr and [ fe / h][email protected] dex . the young population is critical for explaining the terzan 5 double rc . if this explanation were true for the bulge - wide double rc found here it might indicate that the bright rc is due to a huge accreted stellar system . while a second , young , metal - rich , population is acceptable for an individual globular cluster , it is in conflict with the distribution of _ cmd _ ages found by zoccali et al . ( 2003 ) and clarkson et al . ( 2008 ) , who find almost no trace of bulge stars with ages less than 10 gyr . the zoccali et al . ( 2003 ) field is closest to our fields near @xmath40=@xmath41 , which shows a large fraction of bright and faint rc star populations . thus , for double rcs due to age and metallicity , rather than distance , there should be large numbers of both young and old populations in the zoccali et al . ( 2003 ) field ; yet no significant 6 gyr population is evident in the zoccali et al . ( 2003 ) data . the two rcs in ter 5 @xmath42 are shifted in mean ( j@xmath11k ) color by @xmath20.15 magnitudes , consistent with the claimed [ fe / h ] difference . however , for our bulge - wide fields the bright and faint rcs have nearly identical mean colors , no matter what the reddening , suggesting no significant metallicity differences . without a metallicity difference the required ages of the two bulge - wide rc populations , for [ fe / h]@[email protected] , needs to be 2 and 14 gyrs , as stated above ; such age differences are easily ruled - out . thus , there seems to be no way to appeal to an acceptable combination of age and metallicity to explain the brightness and color of the two bulge - wide double rcs discussed in this paper ; thus , we assume that the double rcs reflect a distance effect . a caveat is that this conclusion relies on the assumption that the age distribution at @xmath40=@xmath41 is similar to the old ages in the @xmath43 field of zoccali et al . ( 2003 ) and the @xmath44 field of clarkson et al . ( 2008 ) . k ) color - magnitude diagram for the field at ( @xmath9,@xmath10)=(@xmath111,@xmath118 ) . the two red clumps are visible showing a slight upward trend . red lines indicate the observed k - band rc peaks in the plaut field at ( @xmath9,@xmath10)=(@xmath391,@xmath118 ) . the upward pointing vector shows the change in rc k,(j@xmath11k ) from [ fe / h][email protected] to 0.00 , predicted from the teramo stellar evolution models ( e.g. , pietrinferni et al . 2004 ) and is consistent with the observed rc slopes . also shown is a downward reddening vector for @xmath37a@xmath45=1.0 mag . , width=340 ] in order to map out the extent of the double rc region toward the galactic bulge we have produced cmds and lfs similar to figure [ cmd_b6 ] for a large number of regions . in figure [ fig - lon ] we show some of these regions for latitude @xmath46 , with longitudes ranging from @xmath47@xmath12 to @xmath48@xmath12 . the mean reddening of each field was estimated from the mean 2mass ( j@xmath11k ) color of the red clump . here we adopt an unreddened 2mass ( j@xmath11k)@xmath36=0.61 for the @xmath118@xmath1 bulge red clump , based on the alves ( 2000 ) mean cit v@xmath11k=2.35 for solar neighborhood rc stars , the alonso et al . ( 1999 ) color - t@xmath49 relations and the cit2mass color transformations of carpenter ( 2005 unpublished ) . the mean metallicity of the alves ( 2000 ) solar neighborhood rc stars is similar to the mean metallicity of the @xmath118@xmath1 bulge field , at [ fe / h][email protected] dex , based on a linear metallicity gradient computed with the metallicities of nearby fields from zoccali et al . thus , the alves ( 2000 ) rc calibration has the appropriate metallicity for this bulge field , and roughly the right t@xmath49 and ( j@xmath11k ) color , although because the bulge is significantly older the luminosity of the solar neighborhood rc stars is different . note that because the rc ( j@xmath11k ) colors show a small metallicity sensitivity our reddening values are close , but not strictly correct , for bulge fields with different mean metallicities . in our analysis we employ the winkler ( 1997 ) reddening law , from which we find a@xmath35=0.64 e(j@xmath11k ) . we note that nishiyama et al ( 2009 ) find a slightly different reddening law for the bulge , for which they obtain a@xmath35=0.528 e(j@xmath11k ) . our k - band extinction correction for ( @xmath9,@xmath10)=(@xmath111,@xmath118 ) at 0.10 mag . would be reduced to 0.08 mag . if we employed the nishiyama et al . ( 2009 ) reddening law . because these values are very small , being less than the bin width in the luminosity functions presented here , the choice of reddening law does not affect our conclusions . we prefer to use the ( j@xmath11k ) color difference to estimate the average reddening and k - band extinction for each field rather than the ( h@xmath11k ) color because the bulge rc ( h@xmath11k ) color , in our fields , suffers significant overlap with foreground disk main sequence and rc stars . we do not attempt to correct for extinction of individual rc stars using the reddening - free k@xmath50 magnitudes employed by nishiyama et al . ( 2005 ) , because this introduces unacceptable errors for our fields . the formula used by nishiyama et al . ( 2005 ) effectively adopts reddening values for each star based on the h@xmath11k color distance from the mean rc value . however , the rc is not a discrete point , but has an intrinsic color width even for single populations ; in 47 tuc the rc has a 2mass color width of 0.10 magnitudes in ( j@xmath11k ) and 0.13 magnitudes in ( h@xmath11k ) . because the bulge contains stars with a range of metallicity the intrinsic width of the rc is even larger than in 47 tuc . for this reason , distance from the mean rc color is a poor reddening indicator in the low - extinction bulge fields considered here . if we employ k@xmath50 from nishiyama et al . ( 2005 ) our bulge luminosity functions would effectively be convolved with an error distribution and the peaks more difficult to identify . because the k - band extinctions of our fields are small , typically less than 0.10 magnitudes , it is more reliable to simply correct for the average extinction for each field , implied from the mean rc ( j@xmath11k ) color , than to employ the putative reddening - free k@xmath50 magnitudes for individual stars . because of the metallicity - dependence of the rc ( j@xmath11k ) and m@xmath35 values our adopted mean extinctions and implied distances are sensitive to systematic metallicity differences between fields . in figure [ fig - lon ] we show the dereddened 2mass k - band luminosity function , for 8 fields covering longitudes from @xmath9=@xmath116 to @xmath9=@xmath397 degrees , at a latitude of @xmath10=@xmath118 degrees . each field was constructed using 2mass data from a 30 arc - minute radius circular aperture on the sky . the luminosity function was derived from a vertical strip , 0.10 magnitudes wide in ( j@xmath11k)@xmath36 , centered on the mean rc ( j@xmath11k)@xmath36 color for each field . the color - magnitude diagrams of all 8 fields show that the bright and faint rc possess similar ranges and mean ( j@xmath11k)@xmath36 values , as also seen in figures 1 and 2 . this is most easily understood if the two rcs have similar metallicities and reddenings ( i.e. , due to foreground reddening ) . for fields beyond the ranges of figure [ fig - lon ] we can see the same rcs individually , but not both together , over a large area of sky approximately 20@xmath5120 degrees in extent . globular clusters have a negligible effect on our luminosity functions . in figure [ fig - lon ] only the field at ( @xmath9,@xmath10)=@xmath393,@xmath118 contains a globular cluster , ngc 6624 , which covers an extremely small fraction of the field . the distance modulus of ngc 6624 ( harris et al . 1996 ) indicates k@xmath36=14.1 , which coincides with a minuscule peak in the luminosity function of figure [ fig - lon ] , and is comparable to the noise . we determine distances of the two rc populations using the observed 47 tuc rc plus theoretical corrections for metallicity . for the bulge field in figure [ fig - lon ] at ( @xmath9,@xmath10)=(@xmath391,@xmath118 ) the bright and faint rc peaks are located at k@xmath36=12.64 and 13.29 respectively . zoccali et al . ( 2008 ) suggests that [ fe / h][email protected] for a bulge latitude of @xmath52 ; koch & mcwilliam ( 2008 ) find [ fe / h ] for 47 tuc of @xmath110.76 dex . the teramo stellar evolution models indicate that the bulge rc is 0.20 magnitudes more luminous than the 47 tuc in the k - band , due to this metallicity difference , assuming that both systems are 12 gyr old . if the bulge is younger or more metal - rich the size of the correction increases . the 47 tuc rc 2mass k@xmath36 magnitude is 11.98 and its distance modulus is 13.22 ( koch & mcwilliam 2008 ) . combined with the metallicity k - band correction we find distances for the bright and faint bulge rcs of 6.5 and 8.8 @xmath50.2 kpc respectively . as noted earlier , the ratio of bright to faint rc in figure [ fig - lon ] varies strongly with longitude , with the faint rc dominant at negative longitudes , and the bright peak strongest on the positive longitude side of the bulge . for the latitude @xmath52 fields of figure [ fig - lon ] the faint rc peaks around @xmath53 and the bright rc peaks near @xmath54 . we note that the foreground peak , at @xmath54 contains many more rc stars than the peak of the background population , at @xmath53 . because the background population is at a higher distance above the plane for @xmath52 , a lower star count is expected for the background rc . this provides a qualitative explanation why the negative longitude side of the bulge is fainter at nir wavelengths than at positive longitudes away from the galactic plane . the panels in figure [ fig - lon ] show that the two rcs remain at similar k@xmath36 magnitudes , at latitude @xmath52 , although the relative strength of the faint and bright rcs changes strongly with longitude , with the bright one dominating at positive longitudes , and the faint one at negative longitudes . closer to the plane , at latitudes @xmath55 , similar behavior is observed : a ( single ) bright clump at positive @xmath9 , becoming progressively fainter as one crosses the minor axis and moves towards negative @xmath9 . while the rc populations in the panels of figure [ fig - lon ] appear at nearly constant k@xmath36 magnitude , the ( @xmath3)=(@xmath56 ) and ( @xmath57 ) panels provide the strongest evidence of locations where the foreground rc is more distant and the background rc closer than the norm . it would be interesting to know whether these distance changes are consistent with spiral locii . the same two - component rc behavior is seen on the positive latitude side , above the galactic plane , as evident in figure [ fig - lon7 ] . compared to the fields at @xmath52 the double rc peaks are less pronounced at @xmath58 . on the positive latitude side the peak of the rc number counts is near @xmath59 and @xmath60 for the foreground and background populations respectively , which is very similar to the @xmath61 side . also , like the @xmath61 side , there is marginal evidence in figure [ fig - lon7 ] that the foreground and background peaks are closer to the galactic center at @xmath62 and @xmath63 respectively . = @xmath391 to @xmath3910@xmath1 at latitude @xmath10=@xmath398 , normalized to the maximum of the bright rc and shifted for clarity . the position of the bright rc appears unchanged over 9@xmath1 in longitude , suggesting roughly constant distance . the faint rc can also be seen in the profiles at @xmath9=@xmath391@xmath1 and @xmath392@xmath1 , and marginally at @xmath393@xmath1 longitude.,width=302 ] in figure [ fig - over8 ] we overplot the luminosity function at @xmath10=@xmath398@xmath1 for longitudes from @xmath391 to @xmath3910@xmath1 , showing that the foreground rc lies at almost constant magnitude , although it does seem slightly fainter ( more distant ) at @xmath9=@xmath3910@xmath1 longitude . this structure is a major component of the bulge at positive longitudes ; its near - constant distance is completely at odds with a bar tilted at @xmath220@xmath1 to the line of sight , which should have changed distance by more than 2 kpc , over 9@xmath1 in longitude , or about 0.7 magnitudes in brightness . plane , for longitudes @xmath64 to @xmath65 , if we interpret the clumps as tracers of two populations ( m@[email protected] assumed for the @xmath61 field ) . point sizes are proportional to the number of rc stars in the peak , 0.1 mag . , bin above the background . the bright ( faint ) population dominates at positive ( negative ) longitudes . approximate uncertainty on these distances is @xmath20.3 kpc . the `` gc '' point marks the distance of the central super - massive black hole ( i.e. for @xmath66 ) from ghez et al . ( 2008 ) , at [email protected] kpc , but projected onto the @xmath61 plane , for a distance of 8.1 kpc . , width=340 ] in figure [ distdot-8 ] we plot the rc distances for 12 fields in the latitude @xmath52 plane , using the rc peak magnitudes measured from figure [ fig - lon ] , supplemented with four extra fields . the distances assumed [ fe / h][email protected] for @xmath61 with a metallicity correction of @xmath110.20 mag . applied to the observed 47 tuc rc m@xmath35 ( see koch & mcwilliam 2008 ) , based on the theoretical rc calibration of pietrinferni et al . ( 2004 , 2006 ) . the fields range in longitude from @xmath64 to @xmath67 . the sizes of the dots in figure [ distdot-8 ] are proportional to the number of stars at the peak of the rc . from figures [ fig - lon ] and [ distdot-8 ] , our best estimate of the maximum counts of the foreground and background rc populations are @xmath68 and @xmath69 respectively . a line connecting these background and foreground peaks is tilted to the @xmath70 line by @xmath220@xmath6 , which is identical to the tilt of the galactic bar claimed by numerous studies . the line joining the foreground and background rc peaks intersects the @xmath70 line at a distance of @xmath27.7 kpc from the sun , within 1@xmath71 of the distance to the galactic center ( ghez et al . 2008 ) , projected onto the @xmath61 plane . the @xmath70 intersection with this putative bar line is close to the half - way distance between the foreground and background , indicating symmetry . the above facts suggest that the rc populations we have found reside at the ends of the galactic bar ; presumably , at @xmath72 we are seeing vertical projections from the bar ends . the almost constant distances found for the foreground rc population in figure [ fig - over8 ] most likely reflect stars from the nearby end of the bar , spread out into an orbital arc , or partial arc . our observation that far from the bar ends the rcs appear closer to the galactic center distance suggests arcs extending from the bar ends ; they seem closer to the center than expected from a circular orbit . because the bar is nearly pointing toward us , the arcs extend roughly perpendicular to the line of sight direction ; this can explain why the distance to the foreground rc does not change much over 9@xmath1 in longitude . although several groups have modeled the galactic bar using the red clump magnitude as a measure distance , almost all used low - latitude data , with @xmath73 . the exception is rattenbury et al . ( 2007 ) , who included data from 45 ogle - ii fields ; only two of these fields exceeded @xmath74 , and were located near @xmath75 . rattenbury et al . ( 2007 ) found that beyond longitude @xmath76@xmath26@xmath1 the v@xmath11i luminosity functions are inconsistent with their initial simplistic tilted bar model used to fit the inner regions . at large negative longitudes the i@xmath77 luminosity functions on the right panel of figure 7 in rattenbury et al . ( 2007 ) appear to have the same peak value , consistent with no tilt to the bar at all , although they did not comment on this . for large positive longitudes , _ l_@xmath786@xmath1 the ogle v@xmath11i luminosity functions in figure 7 of rattenbury et al . ( 2007 ) were much broader than the negative longitude data with a shorter distance modulus . for these ogle fields ( numbers 813 ) the bar model predicted much shorter distances than the data . to fix these inconsistencies rattenbury et al . ( 2007 ) introduced a tri - axial model for the bar , but this added complexity still did not provide satisfactory fits to the luminosity functions at large longitudes positive or negative . the cmd the ( @xmath9,@xmath10)=(0,@xmath391 ) field of babusiaux & gilmore ( 2005 ) , published in figure 4 and 5 of their paper , indicate a double , tilted , rc separated by 0.7 mag . in the k - band . the two rcs show the same ( j@xmath11k@xmath79 ) range and upward tilts indicating that the ( j@xmath11k@xmath79 ) range is dominated by metallicity dispersion , rather than reddening dispersion . while this apparently confirms our double rc finding to @xmath10=@xmath391 , the separation is larger than expected from figure [ fig - lat ] in this work . they also found a second rc at l=@xmath395 , indicating a more distant structure . nishiyama et al . ( 2005 ) noted the presence of a weaker rc in _ b_=@xmath391 fields with @xmath80 , at k@xmath81@xmath82 13.5 . they also found a significantly shallower slope of dereddened k magnitude versus longitude in within @xmath83@xmath114 , compared to larger longitudes , and suggested that this is evidence of an inner bar . a further confirmation of the existence of a double clump at @xmath84 comes from the analysis of vieira et al . ( 2007 ) , who measured proper motions for @xmath85 stars in plaut s window . the double clump is barely visible in the cmd of their figure 4 , but it is evident in their figure 10 , although they did not comment on it . the separation between the two clumps is @xmath86 mag , consistent with what we find in figure [ fig - lon ] , and slightly larger than the separation between the clumps at @xmath20 , as if the two populations get further away from each other when going far from the galactic plane . in figure [ fig - lat ] we show a vertical cut in the 2mass data at constant longitude , @xmath87 , with latitudes ranging from @xmath88 to @xmath89 degrees . the main point to notice in figure [ fig - lat ] is that the two rc populations are present over a 20 degree range in latitude , although we do not have information for the inner @xmath55 degrees . we note that data for the panels at @[email protected] degrees were taken from rectangular boxes , 1.5 degrees in latitude by 4 degrees longitude , for an increased area of sky , necessary to detect the two rc components above the background noise . thus , these panels in figure [ fig - lat ] should be divided by a factor of 7.64 to normalize to the counts of the lower latitude fields ( 30 arc - minute radius circles ) . figure [ fig - lat ] ( and the @xmath391,@xmath118 panel in figure [ fig - lon ] ) show that the separation between the foreground and background rcs decreases toward the galactic plane . at @xmath8@xmath2@xmath510@xmath1 the separation in k@xmath36 is 0.71 mag . , at @xmath10=@xmath118@xmath1 it is 0.64 mag . , while at @[email protected]@xmath1 the separation is 0.44 mag . this indicates that the two rcs are closer together at lower galactic latitudes . this is consistent with the de - reddened ogle v@xmath11i photometry of baade s window ( at @xmath10=@xmath114 ) in figure 1 , showing a single , wide rc . in figure [ fig - lat ] the bright rc appears approximately unchanged with latitude , while the faint rc becomes brighter toward the galactic plane . this suggests that the bright and faint rcs are closer together in distance at lower latitudes . because the k@xmath36 value of the bright rc does not change significantly it might be assumed that the bright rc is at a fixed distance , and that the faint peak is closer at lower latitudes , forming a k - shape morphology . however , for figure [ fig - lat ] the computation of the de - reddened k - band magnitudes employed a single rc m@xmath35 value , which we believe is correct for the metallicity of the field at @xmath10=@xmath118@xmath1 , but is not correct for fields with different mean metallicities . in order to determine more reliable distances from the two rcs we adopt metallicities as a function of galactic latitude from zoccali et al . ( 2008 ) , and the metallicity sensitivity of rc m@xmath35 from the teramo stellar evolution models ( e.g. , pietrinferni et al . we note that the rc m@xmath35 is also sensitive to age ; in particular , the teramo rc models show steeper metallicity - dependence for older ages . studies of the galactic bulge age ( e.g. , zoccali et al . 2003 ; clarkson et al . 2008 ) show that it is in the range 1014 gyr , with a very small population @xmath905% possibly as young as 5 gyr . we have adopted an age of 12 gyr and assume a 1@xmath71 uncertainty of@xmath21 gyr . we simply add the theoretical rc corrections to the observed m@xmath35 of the 47 tuc rc . in figure [ fig - x ] we show the distances of each rc for each latitude field in figure [ fig - lat ] . figure [ fig - x ] shows a remarkable x - shape structure ; again , we note that if the metallicity gradient were not taken into account figure [ fig - x ] would appear more k - shaped . corrected for zoccali et al . ( 2008 ) mean [ fe / h ] values , based on teramo group theoretical rc m@xmath35 predictions ( pietrinferni et al . 2004 ) . a dominant , nearly symmetric , x - shape morphology is obvious . the `` gc '' point marks the geometric galactic center distance determined by ghez et al . ( 2008 ) , at 8.0 @xmath50.6 kpc . the x - shape appears centered at 7.3 @xmath50.3 kpc ; thus , to within the measurement uncertainties , the x - shape center agrees to within the uncertainties with the galactic center distance . however , better agreement would occur if the bulge is 23 gyr younger than 47 tuc , width=340 ] an alternative interpretation is that the metallicity of the background field increases toward the galactic plane , while the foreground field remains at constant mean metallicity . this explanation requires a metallicity increase of @xmath20.7 dex from @xmath8=10.25 to 5.5@xmath1 , which is much greater than the observed metallicity gradient of zoccali et al . we therefore abandon this possibility . recently , zoccali et al . ( 2010 ) , hill et al . ( 2010 , unpublished ) and babusiaux et al . ( 2010 ) claimed that the bulge metallicity gradient of zoccali et al . ( 2008 ) is an artifact , due to a changing ratio of two populations of bulge stars with different metallicities and kinematics . the larger velocity dispersion bulge stars have lower mean metallicities ( e.g. , minniti 1996 ) , and reside further from the galactic plane than the higher metallicity population . we note that the range of [ fe / h ] from zoccali et al . ( 2008 ) is such that almost all of the helium - burning stars in both populations go through the rc phase . our use of the mean metallicities is appropriate to determine average distances as a function of longitude and latitude , but it is probable that the two populations do not share the same morphology . we note that there must also be a third population , the inner - halo , as evidenced by the bulge rr lyrae stars ; this population possesses a spherical , rather than bar - like , morphology ( majaess 2010 ) . however , the inner halo does not affect our results because most of its stars would not appear on the rc and because the peak of the halo metallicity function , near [ fe / h][email protected] dex , was not detected in the study of zoccali et al . ( 2008 ) . thus , our picture of the 2mass bulge now suggests a latitudinal narrowing of the two rc populations toward the galactic center , with roughly constant distances in the longitudinal direction at high latitudes , together with an asymmetry in longitude . this suggests a three dimensional x - structure . we estimate the minimum fraction of rc stars in the peaks in figure [ fig - lon ] by subtracting the rgb background below the rc peaks , through interpolation of the rgb above and below the rc region . the minimum counts between the two rgb background - subtracted rc peaks provides a means to estimate the minimum contribution of the peaks to the total rc population . either the minimum is due to the overlapping wings of the peak profiles , or a smooth rc population , due to the bar , underlying the rc peaks . between 20 and 100% of the bulge rc at ( @xmath3)=(@xmath4 ) is in either the foreground or background peaks . at ( @xmath3)=(@xmath91 ) the minimum extends practically to the background , so the entire population appears to be either background or foreground at this latitude , with no detectable bar at intermediate distances . for latitudes with @xmath92 ( c.f . figure [ fig - lat ] ) it is difficult to determine whether the foreground / background components are a few percent or 100% of the population . detailed modelling will be required for more precise estimates . the narrowing of the distance between the foreground and background rc populations suggests that the bar is significantly shorter than the distance between the two rcs at @xmath93 . it is also possible that the x - shape dominates over the bar even close to the galactic plane ; however , the brava velocity dispersion is consistent with a bar component , even at @xmath61 . here we investigate published evidence relevant to differences between the bright and faint rc populations . if our bright and faint rc bulge populations are on opposite sides of the galactic center , then their different distances can be revealed by proper motion studies . mao & paczyski ( 2002 ) predicted a proper motion difference between bright and faint rc bulge sub - populations near 1.6 mas yr@xmath94 for baade s window , assuming that they correspond to the near and far halves of the galactic bar . sumi et al . ( 2003 ) measured proper motions of 47,000 bulge stars from ogle survey data ( at @xmath95 ) , with a baseline of 4 years . they found a proper motion difference of [email protected] mas yr@xmath94 between bright and faint rc sub - populations , in good agreement with the expectations of mao & paczyski ( 2002 ) . our analysis of the sumi et al . ( 2003 ) data shows that the proper motion difference is reduced to [email protected] mas yr@xmath94 if the bright end of the bright rc population box is reduced to resemble our bright rc sub - population limits . this result indicates that the bright and faint rc populations are , indeed , separated in distance as we assume here for our fields at higher latitudes . an additional proper motion test can be obtained from the photographic data of vieira et al . ( 2007 , henceforth v07 ) for the plaut field at @xmath96 , with a baseline of 21 years . from 328 bright rc and 365 faint rc stars in the v07 data , with 2mass photometry , we find only a 1@xmath71 difference between the mean proper motions of bright and faint rcs , in the longitude direction , at 0.19@xmath97 mas yr@xmath94 . in the galactic latitude direction the proper motion difference is 0.51@xmath98 mas yr@xmath94 , or about 3@xmath71 , with the faint rc mean proper motion smaller than the bright rc s . our analysis of the v07 data also shows a 3.4@xmath71 difference in the proper motion dispersions of the bright and faint rc populations in the latitude direction . the faint rc dispersion is smaller than for the bright rc , consistent with the faint rc being more distant , although not as large as would be expected from our adopted distances . however , we expect that the signal will be diluted by contamination from foreground / background rgb stars in the faint / bright rc samples . thus , while the v07 mean longitudinal proper motion differences between the two rc populations are not consistent with distance separation , both the mean proper motion and the proper motion dispersion in the latitude direction are consistent with the faint rc at a significantly greater distance . two radial velocity studies of the galactic bulge have been undertaken recently : the extensive brava fiber survey ( howard et al . 2008 , 2009 ) of bright bulge giants , and a fabry - perot investigation of bulge giants and rc stars by rangwala , williams & stanek ( 2009 , henceforth rws09 ) . the two studies show consistent mean velocities and velocity dispersions in the overlap regions . the brava study is quite extensive , covering a range of 20@xmath1 in longitude , at latitudes of @xmath99 and @xmath100 . the resultant brava velocities are consistent with cylindrical rotation with a peak - to - peak amplitude of @xmath2150 km / s for fields at @xmath99 and @xmath100 , or a maximum velocity of @xmath275 km / s . howard et al . ( 2008 , 2009 ) conclude that this is consistent with a pseudo - bulge , but not a classical bulge . shen et al . ( 2010 ) model the brava velocities and find a pure bar morphology with no trace of a classical bulge . because the brava survey did not reach the faint rc , it contains no information on the bright versus faint rc kinematics . a very important point is that both the brava and the rws09 velocities are consistent with orbital motion , showing the expected change in sign where stars at positive longitudes , near @xmath54 , are foreground , but background near @xmath101 . this strongly confirms our assumption that the fields near these longitudes are separated in distance , rather than at the same distance with differing rc magnitudes , due to unspecified population effects . one might ask how the brava velocity - longitude plots can appear continuous and symmetric when it samples stars on opposite sides of the galactic center that are asymmetrically distributed in longitude . the answer is that for a line of sight through a circular orbit the foreground and background have the same radial velocity , but opposite proper motions . the symmetry of the brava velocities with longitude indicates that the foreground and background populations are equally distant from the center of rotation . in addition to the overall velocity field rsw09 also dissected the rc into bright and faint components . they found bright minus faint rc velocity differences in their @xmath102 fields of @xmath1135@xmath511 km / s , which they attributed to non - circular streaming motions . closer to the minor axis , at @xmath103 , rsw09 found no velocity differences between the two rc populations . the velocity difference at @xmath55@xmath1 might be understood as the signature of off - center bar rotation , with the axis of rotation closer to the background rc population ; however , this possibility appears to be ruled - out by the symmetry of the brava velocities . rws09 noted that this velocity difference is opposite to the predictions of mao & paczyski ( 2002 ) : faint rc stars have more positive velocities , not the expected more negative values . thus , the bright / faint rc velocity differences of rws09 must be investigated further . although many metallicity studies of the galactic bulge have been undertaken , only rangwala & williams ( 2009 , henceforth rw09 ) have compared [ fe / h ] values from bright and faint rc stars . rw09 measured metallicities from the 8542 ca - triplet line in their earlier fabry - perot data . based on a calibration with high resolution [ fe / h ] values they inferred a mean [ fe / h ] for the field at @xmath104 of @[email protected] dex , and @[email protected] dex for the @xmath101 field . this metallicity difference suggests that the bright and faint rc are distinctly different populations . however , such a longitudinal , lop - sided , metallicity difference is very difficult to understand for most morphologies ; if correct , accretion of a dwarf galaxy seems plausible . because a lop - sided metallicity trend would have significant implications for our understanding of the bulge , independent verification of this result is critical . the presence of stars in the rw09 sample with reported [ fe / h ] up to @xmath393 dex gives cause for concern that large systematic uncertainties may be present . andreas koch ( 2010 , private communication ) has made provisional metallicity measurements from the brava spectra ; he found no significant asymmetry with longitude , contradicting the unusual metallicities of rw09 . we , therefore , significantly downgrade the weight given to the metallicity asymmetry claimed by rw09 . from 2mass k@xmath36 , ( j@xmath11k)@xmath36 color - magnitude diagrams we found two distinct rc populations towards the galactic bulge , separated by @xmath37k@[email protected] magnitudes at latitude @xmath52 , and co - existing in fields over 13@xmath1 in longitude and 20@xmath1 in latitude . the presence of the two rc populations is particularly obvious at a galactic latitude of @xmath118@xmath1 . we detect the individual rc populations over a @xmath220@xmath10520@xmath1 area of sky , roughly symmetric about the galactic center . thus , the two populations cover essentially the entire galactic bulge / bar region ; however , the 2mass data do not probe the rc within @xmath25@xmath1 of the galactic plane , due to confusion limitations . we also find that the faint rc is dominant on the negative longitude side of the galactic center , while the bright rc is the principal population at positive longitudes . based on the age and metallicity sensitivity of the rc , predicted by theoretical stellar isochrones of the teramo group ( pietrinferni et al . 2004 ) , we find that the two rcs can not be due to any allowed combination of age or [ fe / h ] in the galactic bulge / bar at a single distance . number statistics firmly rule - out the possibility that the two luminosity peaks are due to rc plus red giant branch bump ( rgb bump ) or asymptotic giant branch bump ( agb bump ) . heretofore unknown stellar evolution effects are also militated against , due to the change in number ratio between positive and negative longitude sides of the galactic bulge . the obvious , and natural , explanation is that the faint and bright rc peaks reflect different distances of two populations . at ( @xmath3)=(@xmath4 ) we estimate distances of 6.5 and 8.8@xmath106 kpc for the two rcs , based on m@[email protected] , computed with theoretical offsets from the observed 47 tuc k - band absolute magnitude , and assuming an age of 12 gyr and a mean [ fe / h][email protected] dex . figure [ distdot-8 ] and [ fig - x ] should be consulted for distances at other locations . we note that these distances are based on the assumption of a bulge age equal to that of 47 tuc , near 12 gyr ; our estimated distances increase for younger bulge ages . radial velocity data , particularly from the brava survey ( howard et al . 2008 , 2009 ) , are consistent with the faint rc population being more distant than the bright rc population . proper motion results from sumi et al . ( 2003 ) and v07 are also generally consistent with a distance interpretation . an immediate conclusion is that these two co - existing populations are not consistent with the body of a bar , since the line of sight through a bar should occur at one distance , not two . other evidence against these populations being in the body of a bar is the fact that both populations mostly exist at fixed distances , independent of longitude ; in particular , the foreground rc peak k@xmath36 magnitude is unchanged over 9@xmath1 in longitude . this is in stark contrast to a bar , which other studies have claimed is pointed almost directly towards us , tilted by only @xmath220@xmath1 to the line of sight . for a bar tilted at 20@xmath1 the distance change over 9@xmath1 in longitude is expected to exceed 2kpc , or roughly 0.7 magnitudes . when we plot distance of the peaks versus longitude for latitude @xmath52 , and add approximate number counts , we find that the faint and bright rc counts peak near longitudes @xmath107 and @xmath54 , respectively . a line joining these two maxima is tilted , approximately 20@xmath1 to the line of sight and crosses the @xmath70 line within 1@xmath71 of the distance to the galactic center , projected onto the @xmath52 plane ; this intersection point is roughly midway between the foreground and background populations . as a function of latitude our luminosity plots show that the two rc peaks are closer in distance at lower galactic latitudes ( i.e. , toward the plane ) , mostly because of an increase in brightness of the background rc . the two rcs appear to merge around @xmath108 . when we make a crude correction to the rc m@xmath35 for the metallicity gradient with latitude along the minor axis , by zoccali et al . ( 2008 ) , we found that a vertical cut through the bulge / bar , at @xmath103 ( from @xmath109 to @xmath110 , i.e. , as viewed from the side ) shows an x - shaped morphology . the center of the x - shape occurs near [email protected] kpc , which is equal to , within the uncertainties , the galactic center distance of ghez et al ( 2008 ) , at [email protected] kpc . better agreement between these two centers would be obtained if the bulge is @xmath223 gyr younger than 47 tuc , or if the metallicity difference between bulge and 47 tuc rc stars is larger than we assume , or if there is an unspecified systematic error in the theoretical isochrones . the rc peak counts above the troughs in the double peaked rc luminosity functions gave a minimum estimate for the fraction of stars in the x versus stars in a broader rc distribution , perhaps from a bar or spheroidal component . this is a minimum because it is possible that the troughs are partly , or entirely , due to the overlapping wings of the distributions of the foreground / background rc components . we found that at ( @xmath3)=(@xmath4 ) the foreground / background rc x components are at least 20% of the total rc population ; at ( @xmath3)=(@xmath111 ) the foreground / background x - components must be close to 100% of the total rc ; on the other hand , near @xmath112 the estimated fraction due to foreground / background components ranges from almost negligible to 100% of the population . our analysis of the 2mass data contains no compelling direct evidence of a galactic bar , although one may be present in a smooth rc population between the rc peaks . as noted above , at ( @xmath3)=(@xmath111 ) , the 2mass data show no room for a bar . detailed modelling will be required to constrain a bar population in this data . it seems reasonable , however , that the x - shape may merge into a short bar at latitudes below @xmath113 . we note that the radial velocity data of the brava survey shows an increased dispersion near @xmath114 , as expected from a bar . detailed models of these radial velocities by shen et al . ( 2010 ) provide excellent fits with a pure bar structure ; they found that a spheroidal component can be no more than 8% of the total . our finding that at least 20% of rc stars are in the x - shape component at @xmath52 , rather than a bar , may be in conflict with the 8% limit of shen et al . ( 2010 ) ; however , their limit refers to a spherical distribution , rather than arcs at the ends of a bar , so it is possible that no conflict exists . a reasonable , qualitative , model to explain the observations presented here is that the high - latitude bulge ( above @xmath1156@xmath1 ) is dominated by the vertical extent of stars near the foreground and background ends of a galactic bar . in this scenario stars extend in all directions from the ends of the bar , i.e. , in both longitude and latitude . in longitude the stars occupy arcs , or partial orbits , emanating from the bar ends . for this reason , a line connecting the peaks in the foreground/ background rc populations trace the direction of the bar , tilted at @xmath220@xmath1 to the line of sight , assuming that there is no significant rotational lag . the axis of rotation of the bar is marked by the point where this line intersects with @xmath70 . in figure [ fig - x ] this point is located at 7.7 kpc from the sun ; it happens to lie roughly midway between the two rcs and within 1@xmath71 of the galactic center distance . this arrangement should produce symmetry in the mean radial velocity curve , similar to the brava observations . because the bar is almost end - on , the stars in the arcs , near the bar ends , spread almost tangentially to the line of sight , and thus appear at a nearly constant distance , as observed . however , at large angular distance , in longitude , from the bar ends , the stars along the arcs / orbits are closer to the galactic center distance , also observed . since stars emanating from the bar ends do so in all directions , they appear as two roughly circular areas , and take - on a peanut appearance when viewed nearly end - on to the bar ; a slight flattening of the peanut may be expected due to the gravitational potential in the latitude direction . when viewed from the side , the structure takes on an x - shape morphology , as the stars emanating from the bar ends merge into the bar . in this case the main component of the bar resides at low galactic latitudes , and perhaps for this reason the end components are relatively easy to detect at high latitude . an alternative description is that there is only an x - shaped component which extends to high latitudes , but without connecting to a bar . as before , since we are looking almost end - on to the axis of the x , the foreground and background appear at nearly constant distances . another , alternative , explanation of the observed two rc components is that we have detected concentrations in the thick disk near the ends of the bar . if correct , this scenario might naturally explain the similarity of the composition of thick disk and bulge red giant stars found by alves - brito et al . ( 2010 ) . if our two populations represent extensions from the ends of a bar , then their velocities should show cylindrical rotation , similar to bar models , but with a distribution not considered by shen et al . one major difficulty is that our results suggest a end - to - end bar length of @xmath22.5 kpc , whereas the best fit model of shen et al . ( 2010 ) indicates a bar half - length of 4kpc . x - shape , boxy , and peanut morphologies of extra - galactic bulges are well known phenomena , e.g. , ic 4767 , ngc 128 , ngc 4469 , ic 2531 ( e.g. , whitmore & bell 1988 ; bureau et al . 2006 . ) the x - shapes are relatively minor components of extra - galactic bulges , and significant image processing is usually required to highlight them , in stark contrast to the strong signal we find here for the galaxy . n - body simulations of isolated disk galaxies and the growth of bars ( e.g. , patsis et al . 2002 ; athanassoula 2005 ) find x - shape morphologies resulting from bar growth . in particular , the * x1v1 * resonance mode of patsis et al . ( 2002 ) appears very similar to our results for the galaxy . if these predictions are correct , the implication is that the bulge x - shape resulted from bar instabilities , and that the galaxy contains a pseudo - bulge , and a bar , rather than a classical bulge . x - morphologies also result from mergers , although bars are still involved : mihos et al . ( 1995 ) performed numerical simulations of satellite accretion by s0 galaxies , and found that prograde accretion induced the formation of bars , which subsequently buckled and produced x structures . since the formation of a bar from an isolated disk galaxy relies on interactions between its disk and halo , the overlap between these two mechanisms is greater than at first apparent . typical timescales for pseudo - bulge growth is @xmath21 gyr ( e.g. , athanassoula 2008 ) , which may be problematic for a rapid formation timescale for the galactic bulge ( e.g. , ballero et al . 2007 ) , as implied by the enhanced [ mg / fe ] ratios ( e.g. , fulbright et al . 2007 ; lecureur et al . 2007 ; mcwilliam & rich 1994 ) ; although , both the timescale and the mg enhancements have associated uncertainties which may bring them into rough agreement . another difficulty is that pseudo - bulges / bars are thought to result in well mixed orbits , such that metallicity gradients may not be expected ( zoccali 2010 ; howard et al . 2009 . ) however , the bulge metallicity gradient is well - known ( zoccali et al . 2008 ) , and the metallicities correlate with kinematic properties . it is possible that the gradient simply results from the super - position of two populations ( zoccali et al . 2010 ; babusiaux et al . 2010 ; hill et al . howard et al . ( 2009 ) speculated that dissipative processes during bar formation might explain the metallicity gradient . while an x - shape bulge qualitatively explains the 2mass photometry and much of the published kinematic data , some literature results are problematic . the @xmath1135 km / s radial velocity difference between bright and faint rc found by rws09 is difficult to understand ; asymmetric bar rotation could explain it , but such an asymmetry in velocities is not obvious in the brava data . also , the 0.4 dex difference in [ fe / h ] between @xmath116 and @xmath117 bulge components , found by rw09 , suggests an accreted system , rather than a bar or symmetric x - component ; however , provisional metallicities from the brava spectra , by koch ( 2010 , private communication ) do not support an [ fe / h ] asymmetry . perhaps the most important difficulty is that the x - shape is evident in our 2mass data when the dependence of mean [ fe / h ] on galactic latitude is taken into account , without the metallicity corrections the morphology is more k - shaped . this is a problem because the expectation is that the bar should not possess a metallicity gradient , and while x - morphologies are well known in external galaxies , we are unaware of k - morphologies . the final piece of confusing evidence is the lack of difference between the mean proper motions of bright and faint rc stars in the data of vieira et al . ( 2007 ) . these apparent inconsistencies demand that future work on the kinematics and composition of rc bulge stars be pursued for a more complete picture of the galactic bulge region . it is particularly important to verify whether there is a metallicity gradient with latitude in the rc x - populations ; if so , this would support the idea of howard et al ( 2009 ) , that the bar formed with dissipation ; if not , then the bulge is k - shaped . if the gradient simply reflects a changing ratio between two populations with different mean [ fe / h ] , then composition studies of these two populations would be interesting and useful . a definitive investigation of the possible metallicity asymmetry with longitude proposed by rangwala & williams ( 2009 ) is also necessary . extant age studies by zoccali et al . ( 2003 ) and clarkson et al . ( 2008 ) should be supplemented with a photometric survey for the ages of bulge stars at @xmath40=@xmath118 , in order to completely rule - out the possibility that age is responsible for the luminosity difference between the two rcs at this latitude . other important future work includes detailed modelling of the 2mass data , in order to determine the relative proportions of a bar and the foreground / background rc components . a high spatial resolution infrared photometric survey , such as the vista variable in the va lctea survey ( vvv , minniti et al . 2010 ) , covering roughly the same longitudinal extent as the current work , but going to much smaller latitudes will help to determine whether the x - shape continues to the galactic center , or joins the bar . analysis of the premilinary vvv catalogues is ongoing ( saito et al . 2010 , in preparation ) . independent verification of the foreground / background distances , at high latitude , with standard candles , such as miras , strong - lined rr lyrae stars , and eclipsing binaries , will also be very helpful . after this paper was submitted for publication nataf et al . ( 2010 ) reported a split rc toward the galactic bulge . their findings are broadly consistent with the results of this paper ; in particular , they find very nearly equal ( v@xmath11i ) colors , consistent with similar metallicity for the two rc components . part of this work was performed by am while visiting the kitp in santa barbara , which is supported under nsf grant pho5 - 51164 . am thanks andreas koch for a useful conversation . mz acknowledges the fondap center for astrophysics 15010003 , the basal center for astrophysics and associated technologies pfb-06 , fondecyt regular 1085278 , and the mideplan milky way millennium nucleus p07 - 021-f . alonso , a. , arribas , s. , & martinez - roger , c. 1999 , a&as , 140 , 261 alves , d.r . 2000 , , 539 , 732 alves - brito , a. , melndez , j. , asplund , m. , ramrez , i. , yong , d. 2010 , a&a , 513 , 35 athanassoula , e. 2005 , , 358 , 1477 athanassoula , e. 2008 , in `` formation and evolution of galaxy bulges '' , iau symposium , volume 245 , p. 93 - 102 , eds . martin bureau , e. athanassoula & beatriz barbuy ( cambridge : cup ) mcwilliam , a. , fulbright , j. , & rich , r.m . 2010 , in `` chemical abundances in the universe : connecting first stars to planets '' , iau symposium , volume 265 , p. 279 - 284 , katia cunha , monique spite & beatriz barbuy ( cambridge : cup ) nishiyama , s. , nagata , t. et al . 2006 , journal of physics : conference series , volume 54 , proceedings of `` the universe under the microscope - astrophysics at high angular resolution '' , eds . rainer schoedel , andreas eckart , susanne pfalzner & eduardo ros , pp . 62 - 66 ( 2006 ) . zoccali , m. 2010 , in `` chemical abundances in the universe : connecting first stars to planets '' , iau symposium , volume 265 , p. 271 - 278 , eds . katia cunha , monique spite & beatriz barbuy ( cambridge : cup )

we present optical ( to ) off - nuclear spectra of 26 powerful active galaxies in the redshift range , obtained with the mayall and william herschel 4-meter class telescopes . the sample consists of radio - quiet quasars , radio - loud quasars ( all with ) and radio galaxies of fanaroff & riley type ii ( with extended radio luminosities and spectral indices comparable to those of the radio - loud quasars ) . the spectra were all taken approximately 5 arcseconds off - nucleus , with offsets carefully selected so as to maximise the amount of galaxy light falling into the slit , whilst simultaneously minimising the amount of scattered nuclear light . the majority of the resulting spectra appear to be dominated by the integrated stellar continuum of the underlying galaxies rather than by light from the non - stellar processes occurring in the active nuclei , and in many cases a 4000 break feature can be identified . the individual spectra are described in detail , and the importance of the various spectral components is discussed . stellar population synthesis modelling of the spectra will follow in a subsequent paper ( nolan et al . 2000 ) . galaxies : stellar content galaxies : active quasars : general

astro-ph0002021

understanding the host galaxies of active galactic nuclei ( agn ) is now recognised as an important step on the path towards reaching an understanding of agn themselves - how they form , how they are fuelled and how the differences between the various classes of object arise . in two areas in particular the nature of the host galaxy gives us a direct insight into the workings of the agn : galaxy properties seem to play a role in determining the radio loudness of the central engine ( powerful radio sources are almost never found in spiral or disc - dominated systems though see mchardy et al . 1994 ) ; and amongst radio - loud objects the host galaxies offer a powerful , orientation - independent means of testing models which attempt to unify different types of agn via beaming and viewing angle effects ( _ eg _ urry & padovani 1995 ) . the observations presented here attempt to address both of these issues by comparing the galaxies associated with the three main types of powerful agn : radio - quiet quasars ( rqqs ) , radio - loud quasars ( rlqs ) and radio galaxies ( rgs ) of fanaroff & riley type ii . only once the problem of separating the diffuse galaxy emission from the wings of the quasar point spread function ( psf ) has been overcome can one can begin to describe and classify the morphologies , brightness profiles and interaction histories of the quasar hosts . over the last decade improvements in ground - based techniques and the advent of the hubble space telescope ( hst ) have revolutionised our understanding of quasar host galaxies . evidence for mergers or interactions in the form of morphological disturbances and close companions is a common feature of these images , but a significant number of quasars are also found in what appear to be undisturbed hosts . in addition , the idea that radio - loudness is a straightforward consequence of the host galaxy type has had to be abandoned . although some radio - quiet quasars are found in spirals ( _ eg _ hutchings et al . 1994 ; rndahl , rnnback & van groningen 1997 ) in general the hosts of both radio - loud and radio - quiet quasars tend to have properties consistent with early - type galaxies ( _ eg _ vron - cetty & woltjer 1990 ; disney et al . 1995 , hutchings & morris 1995 , bahcall et al . 1997 , mcleod , rieke & storrie - lombardi 1999 ) and typically have luminosities @xmath4 ( _ eg _ dunlop et al . 1993 , bahcall , kirhakos & schneider 1994 , 1995ab , 1996 ; hutchings et al . 1994 , boyce et al . 1998 , hooper , impey & foltz 1997 ) and in many cases are comparable in mass to brightest cluster galaxies ( bcgs ) . meanwhile , mcleod & rieke ( 1995a ) find evidence for a _ lower _ limit on the @xmath5-band luminosities ( and hence the mass of the red , established stellar populations ) of galaxies hosting radio - quiet agn , which appears to increase as the nuclear luminosity increases , implying that the nuclear activity is closely linked to the mass of the bulge component of the host . this impression has been reinforced by mclure et al . ( 1999 ) , who find that _ all _ the rqqs with @xmath6 in their hst sample occur in massive elliptical galaxies , with only the least luminous radio - quiet objects lying in disc - dominated hosts . many long - held views about the triggering of nuclear activity and the origins of radio loudness are currently being reassessed in the light of these imaging studies , but images can not tell the whole story . a completely independent way of characterising the host galaxies of agn is via analysis and classification of their stellar populations and starformation histories . the aim of the observations described in this paper was to obtain high signal - to - noise spectra of quasar hosts and radio galaxies for use in spectrophotometric modelling to determine the nature and history of their stelllar constituents . previous off - nuclear spectroscopy of the host galaxies of quasars has produced mixed results . boroson & oke ( 1982 ) were the first to detect an unequivocably stellar continuum from the nebulosity surrounding the radio - loud quasar 3c48 and subsequent studies revealed stellar continua and emission / absorption features around several other quasars ( boroson , oke & green 1982 ; boroson & oke 1984 ; boroson , persson & oke 1985 ; hickson & hutchings 1987 ; hutchings & crampton 1990 ) . however , except for the general result that there appeared to be systematic differences between the spectra of radio - loud and radio - quiet quasar hosts , these studies produced little advance in our understanding of the relationship between rqqs , rlqs and rgs for the reasons outlined below . in general the quasar targets were chosen virtually at random , often for the sole reason that they were ` interesting ' and/or unusual . until now a programme of off - nuclear spectroscopy for statistically useful and properly matched samples of rqqs and rlqs has never been carried out , nor has any attempt been made to compare the off - nuclear spectra of rlqs with equivalent _ off - nuclear _ spectra of radio galaxies . the early work also focussed mostly on emission - line activity , with discussion of the stellar continuum being confined to classification as red or blue , and the identification of a few stellar features . only limited attempts were made to use the form of the spectrum to investigate the composition and evolution of the stellar population . finally , few of the host galaxy spectra were taken sufficiently off - nucleus - _ eg _ the spectra taken by boroson and collaborators were taken only 3@xmath7 from the quasar . this was done in the belief that any further off - nucleus the host galaxy would be too faint for a reasonable spectrum to be obtained , but inevitably resulted in significant contamination of the host galaxy spectrum by scattered light from the quasar nucleus . a scaled version of the quasar spectrum had therefore to be subtracted from the off - nuclear spectrum to reveal the spectrum of the underlying host , but the extra noise and systematic errors introduced by this process severely limited the quality of the final spectra . thus , the main obstacles in these previous attempts to classify the hosts of powerful agn were the difficulty in separating the underlying starlight from the glare of the quasar and - less immediately , but still of some importance - the lack of well - defined samples of a sufficient size to carry out straightforward statistical comparisons . as far as possible , the design of the current study was intended to circumvent both these problems with the goal of obtaining clean spectra of the stellar component of radio - loud and radio - quiet quasar hosts and radio galaxies , which could then be used to search for systematic differences and similarities between the stellar populations and of the galaxies hosting each type of activity . despite that fact that the radio galaxies lack bright nuclear point sources we were careful to adopt exactly the same observing strategy as used for the quasars to ensure that the data would be directly comparable . in this paper we describe the observations and present the spectra . a second paper ( nolan et al . 2000 ) will describe the results of spectrophotometric modelling to estimate the ages and starformation histories of the galaxies ( preliminary results have already been reported by kukula et al . the current paper is organised as follows . section 2 describes the samples used and section 3 details the observing and data reduction strategies chosen to optimise the amount of starlight collected by the various instruments employed throughout the study . section 4 gives an overview of the data obtained and sections 5 and 6 contain more detailed information on the individual spectra . dunlop et al . ( 1993 ) and taylor et al . ( 1996 ) observed a sample of intermediate - redshift ( @xmath8 ) radio - loud and -quiet quasarsw hz@xmath9sr@xmath9 . ] and fanaroff - riley type ii radio galaxies ( rgs ) in the near infrared ( @xmath10-band:2.2@xmath11 m ) in order to compare the luminosities and morphologies of the galaxies associated with these three main types of powerful active nucleus . the choice of waveband was informed by the low quasar : host ratio in the near infrared , which allowed accurate determination of the galaxy properties from the ground with the minimum amount of confusion from the point spread function of the central quasar . their sample was carefully constructed to ensure that the different types of object could be compared directly with one another : the radio - loud and -quiet quasars both have the same distribution of optical luminosities ( @xmath12 ) and both the radio - loud quasars and the radio galaxies have similar extended radio luminosities and morphologies and steep radio spectra . the sample is therefore ideal for investigating the influence of galaxy properties on the ` radio loudness ' of otherwise very similar quasars , and also for testing unified models of rlqs and rgs which predict that the properties of their hosts should be identical . a substantial subset of the sample has also formed the basis for an @xmath13-band imaging study using hst ( kukula et al . 1999 , mclure et al . 1999 , dunlop et al . 2000 ) , in which the enhanced angular resolution of hst has allowed both unambiguous identification of the host morphology and identification of detailed substructure in the quasar hosts and radio galaxies . this sample provides an ideal starting point for a spectroscopic study of host galaxies because , in addition to the careful selection criteria , the existence of deep near - infrared images of every object provides us with a unique opportunity to minimise the contamination of the galaxy spectra by quasar light . armed with knowledge of the extent and orientation of the host galaxy on the sky one is able to choose a slit position which is far enough from the nucleus to avoid the worst excesses of scattered quasar light , but which simultaneously maximises the amount of galaxy light falling onto the slit . out of the the 40 objects in the original taylor et al . sample a total of 26 objects were observed in the current study ( 9 rqqs , 10 rlqs and 7 rgs ) . details are listed in table 1 . note that the radio source 3c59 has been shown by meurs & unger ( 1991 ) to consist of three separate sources , of which only the weakest appears to be associated with the quasar 0204 + 292 . as a result , the radio luminosity of 0204 + 292 places it below our dividing line of @xmath14w hz@xmath9sr@xmath9 and we classify it as an rqq . cllcccl iau & & v & @xmath15 & telescope & alternative + name & ra ( @xmath16 ) & dec ( @xmath17 ) & & & & names + + 0007@xmath18106 & 00 10 30.98 & @xmath1810 58 28.4 & 15.40 & 0.089 & m4 m & iii zw 2 + 0054@xmath18144 & 00 57 09.92 & @xmath1814 46 11.0 & 15.71 & 0.171 & m4m / wht & phl909 + 0157@xmath18001 & 01 59 49.72 & @xmath1800 23 41.0 & 15.69 & 0.164 & m4m / wht & mrk 1014 + 0204@xmath18292 & 02 07 02.16 & @xmath1829 30 45.4 & 16.00 & 0.109 & wht & 3c59 + 0244@xmath18194 & 02 47 41.22 & @xmath1819 40 53.6 & 16.66 & 0.176 & wht & ms 02448 + 19 + 1549@xmath18203 & 15 52 02.28 & @xmath1820 14 01.6 & 16.50 & 0.250 & wht & 1e15498 + 203 , lb906 , ms 15498 + 20 + 1635@xmath18119 & 16 37 46.54 & @xmath1811 49 49.8 & 16.50 & 0.146 & wht & mc1635 + 119 , mc 2 + 2215@xmath19037 & 22 17 47.35 & @xmath1903 32 48.4 & 17.20 & 0.241 & wht & ms 22152 - 03 , ex2215 - 037 + 2344@xmath18184 & 23 47 25.71 & @xmath1818 44 58.4 & 15.90 & 0.138 & m4m / wht & e2344 + 184 + + 0137@xmath18012 & 01 39 57.24 & @xmath1801 31 46.8 & 17.07 & 0.258 & m4 m & phl1093 + 0736@xmath18017 & 07 39 18.01 & @xmath1801 37 04.6 & 16.47 & 0.191 & m4m / wht & s0736 + 01 , oi061 + 1004@xmath18130 & 10 07 26.12 & @xmath1812 48 56.4 & 15.15 & 0.240 & wht & s1004 + 13 , ol107.7 , 4c13.41 + 1020@xmath19103 & 10 22 32.79 & @xmath1910 37 44.2 & 16.11 & 0.197 & m4 m & s1020 - 103 , ol133 + 1217@xmath18023 & 12 20 11.90 & @xmath1802 03 42.3 & 16.53 & 0.240 & wht & s1217 + 02 , um492 + 2135@xmath19147 & 21 37 45.24 & @xmath1914 32 55.6 & 15.53 & 0.200 & wht & s2135 - 14 , phl1657 + 2141@xmath18175 & 21 43 35.56 & @xmath1817 43 49.1 & 15.73 & 0.213 & wht & ox169 + 2201@xmath18315 & 22 03 15.00 & @xmath1831 45 38.3 & 15.58 & 0.298 & m4 m & b2 2201 + 31a , 4c31.63 + 2247@xmath18140 & 22 50 25.40 & @xmath1814 19 49.9 & 15.33 & 0.237 & m4m / wht & pks2247 + 14 , 4c14.82 + 2349@xmath19014 & 23 51 56.14 & @xmath1901 09 12.8 & 15.33 & 0.173 & wht & pks2349 - 01 , pb5564 + + 0230@xmath19027 & 02 32 43.14 & @xmath1902 33 34.1 & 19.20 & 0.239 & wht & pks0230 - 027 + 0345@xmath18337 & 03 48 46.9 & @xmath1833 53 16 & 19.00 & 0.244 & wht & 3c93.1 + 0917@xmath18459 & 09 21 08.95 & @xmath1845 38 55.8 & 17.22 & 0.174 & wht & 3c219.0 + 1215@xmath19033 & 12 17 55.3 & @xmath1903 37 23 & 18.90 & 0.184 & wht & + 1330@xmath18022 & 13 32 53.27 & @xmath1802 00 45.0 & 18.27 & 0.215 & m4 m & 3c287.1 + 1334@xmath18008 & 13 37 31.38 & @xmath1800 35 29.0 & 19.00 & 0.299 & wht & + 2141@xmath18279 & 21 44 11.66 & @xmath1828 10 18.9 & 18.15 & 0.215 & m4m / wht & 3c436 + lcccccrccl source & type & z & date&airmass&seeing&exposure&slit & @xmath20-range & + & & & & & ( @xmath7 ) & ( sec ) & width&( ) & + + 0007 + 106 & g & 0.089 & 25/09/92 & 1.10 & 1.7 & 5@xmath211800 & 3@xmath7&3665 - 7121 & 5@xmath7n , 90@xmath22 + & n & 0.089 & 25/09/92 & 1.13 & 1.6 & 1@xmath211800 & 3@xmath7&3665 - 7121 & zero offset + 0054 + 144 & g & 0.171 & 26/09/92 & 1.19 & 1.8 & 6@xmath211800 & 3@xmath7&3659 - 7121 & 3@xmath7s , 4@xmath7e , 15@xmath22 + & n & 0.171 & 26/09/92 & 1.60 & 1.6 & 1@xmath21900 & 3@xmath7&3659 - 7121 & zero offset + 0157 + 001 & g & 0.164 & 25/09/92 & 1.26 & 1.8 & 5@xmath211800 & 3@xmath7&3662 - 7121 & 5@xmath7n , 90@xmath22 + & n & 0.164 & 25/09/92 & 1.55 & 1.9 & @xmath23 & 3@xmath7&3662 - 7121 & zero offset + 2344 + 184 & g & 0.138 & 24/09/92 & 1.05 & 1.9 & 5@xmath211800 & 3@xmath7&3665 - 7124 & 5@xmath7e , 177@xmath22 + & n & 0.138 & 24/09/92 & 1.15 & 2.5 & 1@xmath211800 & 3@xmath7&3662 - 7121 & zero offset + + 0137 + 012 & g & 0.258 & 24/09/92 & 1.34 & 2.0 & 5@xmath211800 & 3@xmath7&3662 - 7121 & 5@xmath7e , 155@xmath22 + & n & 0.258 & 26/09/92 & 1.70 & 1.8 & 1@xmath21900 & 3@xmath7&3659 - 7121 & zero offset + 0736 + 017 & g & 0.191 & 19/03/93 & 1.20 & 2.9 & 5@xmath211800 & 3@xmath7&3530 - 7004 & 5@xmath7s , 90@xmath22 + & g & 0.191 & 20/03/93 & 1.19 & 1.4 & 4@xmath211800 & 3@xmath7&3530 - 7004 & 5@xmath7s , 90@xmath22 + & n & 0.191 & 19/03/93 & 1.40 & 3.1 & 1@xmath21900 & 3@xmath7&3530 - 6989 & zero offset + & n & 0.191 & 20/03/93 & 1.17 & 1.4 & 1@xmath211200 & 3@xmath7&3530 - 6989 & zero offset + 1020@xmath19103 & g & 0.197 & 19/03/93 & 1.41 & 2.9 & 3@xmath211800 & 3@xmath7&3530 - 7004 & 3.5@xmath7w , 3.5@xmath7n , 45@xmath22 + & g & 0.197 & 20/03/93 & 1.38 & 1.4 & 4@xmath211800 & 3@xmath7&3530 - 7004 & 3.5@xmath7w , 3.5@xmath7n , 45@xmath22 + & n & 0.197 & 19/03/93 & 1.36 & 2.9 & 1@xmath21900 & 3@xmath7&3530 - 6989 & zero offset + 2201 + 315 & g & 0.298 & 25/09/92 & 1.09 & 1.8 & 5@xmath211800 & 3@xmath7&3665 - 7121 & 5@xmath7n , 84@xmath22 + & n & 0.298 & 25/09/92 & 1.00 & 1.8 & 1@xmath211800 & 3@xmath7&3665 - 7121 & zero offset + 2247 + 140 & g & 0.237 & 24/09/92 & 1.26 & 1.2 & 5@xmath211800 & 3@xmath7&3665 - 7124 & 4@xmath7e , 3@xmath7s , 38.7@xmath22 + & n & 0.237 & 24/09/92 & 1.06 & 1.2 & 1@xmath211800 & 3@xmath7&3665 - 7124 & zero offset + + 1330 + 022 & g & 0.215 & 20/03/93 & 1.18 & 1.4 & 5@xmath211800 & 3@xmath7&3548 - 7004 & 2.5@xmath7n , 2.5@xmath7e , 135@xmath22 + & n & 0.215 & 19/03/93 & 1.18 & 3.1 & 2@xmath211800 & 3@xmath7&3548 - 7004 & zero offset + 2141 + 279 & g & 0.215 & 26/09/92 & 1.03 & 1.8 & 4@xmath211800 & 3@xmath7&3542 - 7124 & 4@xmath7e , 0@xmath22 + & n & 0.215 & 26/09/92 & 1.13 & 1.6 & 4@xmath211800 & 3@xmath7&3542 - 7124 & zero offset + lcccccrcccl source & type&z&date&airmass&seeing&exposure&slit & @xmath20-range&@xmath20 join & + & & & & & ( @xmath7)&(sec ) & width&( ) & ( ) & + + 0054 + 144 & g & 0.171 & 01/09/94 & 1.30 & 0.9 & 4@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 3.5@xmath7n , 4@xmath7w , 228@xmath22 + & n(m4 m ) & 0.171 & 26/09/92 & 1.60 & 1.6 & 1@xmath21900 & 3@xmath7 & 3659 - 7121 & & zero offset + 0157 + 001 & g & 0.164 & 16/11/93 & 1.15 & 1.6 & 4@xmath211800 & 2@xmath7 & 3456 - 9060 & 6100 & 5@xmath7n , 107.5@xmath22 + & n(m4 m ) & 0.164 & 25/09/92 & 1.55 & 1.9 & 1@xmath211200 & 3@xmath7 & 3662 - 7121 & & zero offset + 0204 + 292 & g & 0.109 & 17/11/93 & 1.03 & 1.0 & 5@xmath211800 & 2@xmath7 & 3456 - 9060 & 6100 & 4.5@xmath7w , 0@xmath22 + 0244 + 194 & g & 0.176 & 02/09/94 & 1.06 & 1.7 & 4@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 5@xmath7w , 180@xmath22 + 1549 + 203 & g & 0.250 & 30/03/95 & 1.05 & 1.2 & 4@xmath211800 & 2@xmath7 & 3450 - 8991 & 6100 & 4@xmath7w , 3@xmath7n , 228@xmath22 + 1635 + 119 & g & 0.146 & 13/05/94 & 1.12 & 1.3 & 5@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 4.75@xmath7w , 1@xmath7s , 172@xmath22 + 2215@xmath19037 & g & 0.241 & 01/09/94 & 1.60 & 0.9 & 3@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 4.25@xmath7w , 3@xmath7n , 210@xmath22 + 2344 + 184 & g & 0.138 & 02/09/94 & 1.20 & 1.6 & 4@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 3@xmath7n , 4@xmath7w , 215@xmath22 + & n(m4 m ) & 0.138 & 24/09/92 & 1.15 & 2.5 & 1@xmath211800 & 3@xmath7 & 3662 - 7121 & & zero offset + + 0736 + 017 & g & 0.191 & 16/11/93 & 1.14 & 1.0 & 5@xmath211800 & 2@xmath7 & 3456 - 9060 & 6100 & 3.5@xmath7s , 3.5@xmath7w , 135@xmath22 + & n(m4 m ) & 0.191 & 19/03/93 & 1.40 & 3.1 & 1@xmath21900 & 3@xmath7 & 3530 - 6989 & & zero offset + 1004 + 130 & g & 0.240 & 30/03/95 & 1.20 & 1.5 & 4@xmath211800 & 2@xmath7 & 3450 - 8991 & 6100 & 3.5@xmath7n , 3.5@xmath7w , 43@xmath22 + 1217 + 023 & g & 0.240 & 11/05/94 & 1.20 & 0.8 & 4@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 4@xmath7e , 3@xmath7n , 334@xmath22 + 2135@xmath19147 & g & 0.200 & 02/09/94 & 1.60 & 1.2 & 4@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 3.5@xmath7w , 3.5@xmath7s , 315@xmath22 + 2141 + 175 & g & 0.213 & 13/05/94 & 1.60 & 1.0 & 3@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 5@xmath7n , 290@xmath22 + & n(wht ) & 0.213 & 13/05/94 & 1.30 & 1.1 & 1@xmath21900 & 2@xmath7 & 3240 - 8700 & 6050 & zero offset + 2247 + 140 & g & 0.237 & 16/11/93 & 1.04 & 1.6 & 2@xmath211800 & 2@xmath7 & 3456 - 9060 & 6100 & 4@xmath7e , 3@xmath7s , 38.7@xmath22 + & n(m4 m ) & 0.237 & 24/09/92 & 1.06 & 1.2 & 1@xmath211800 & 3@xmath7 & 3665 - 7124 & & zero offset + 2349@xmath19014 & g & 0.173 & 04/09/94 & 1.35 & 0.7 & 4@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 1@xmath7s , 5@xmath7w , 324@xmath22 + + 0230@xmath19027 & g & 0.239 & 02/09/94 & 1.30 & 2.0 & 2@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 3.5@xmath7s , 3.5@xmath7w , 320@xmath22 + & n(wht ) & 0.239 & 01/09/94 & 1.20 & 1.5 & 1@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & zero offset + 0345 + 337 & g & 0.244 & 02/09/94 & 1.30 & 3.0 & 4@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 5@xmath7n , 0.5@xmath7w , 265@xmath22 + 0917 + 459 & g & 0.174 & 19/11/93 & 1.10 & 1.7 & 5@xmath211800 & 2@xmath7 & 3456 - 9060 & 6100 & 5@xmath7n , 90@xmath22 + 1215@xmath19033 & g & 0.184 & 13/05/94 & 1.19 & 0.9 & 3@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 4@xmath7s , 3@xmath7e , 50@xmath22 + 1334 + 008 & g & 0.299 & 30/03/95 & 1.60 & 2.0 & 2@xmath211800 & 2@xmath7 & 3450 - 8991 & 6100 & 3.5@xmath7w , 3.5@xmath7s , 135@xmath22 + 2141 + 279 & g & 0.215 & 02/09/94 & 1.20 & 1.0 & 4@xmath211800 & 2@xmath7 & 3240 - 8700 & 6050 & 4.5@xmath7n , 2@xmath7e , 296@xmath22 + & n(m4 m ) & 0.215 & 26/09/92 & 1.13 & 1.6 & 4@xmath211800 & 3@xmath7 & 3542 - 7124 & & zero offset + in order to asssess the feasibility of our observing strategy initial observations of 11 of the nearest and brightest objects in the sample were carried out using the mayall 4-m telescope at kitt peak national observatory . this was followed by a larger programme of observations using the 4.2-m william herschel telescope ( wht ) , part of the isaac newton group of telescopes on la palma . tables 2 and 3 list the observations made on the mayall 4-m telescope and the wht respectively . six objects ( 0054@xmath18144 , 0157@xmath18001 , 0736@xmath18017 , 2141@xmath18279 , 2247@xmath18140 and 2344@xmath18184 ) were observed with both telescopes in order to provide a check for consistency between the two sets of observations . the spectra for these objects are discussed in more detail in section 6 . observations of 11 objects were carried out with the r. c. spectrograph on the mayall 4-m telescope at kitt peak in 1992 september and 1993 march ( table 1 ) . the long - slit spectrograph uses a tektronix @xmath24 chip with 24-@xmath11 m pixels , designated t2 kb . the slit width was set to 3@xmath7 and the instrument was configured to give a spectral resolution of 1.9 / pixel and a spatial resolution of 0.69@xmath7/pixel . the slit was first centred on the quasar ( or , if the object was a radio galaxy , on the galaxy centroid ) , before being rotated and offset to the desired position , usually 5 arcseconds off - nucleus . to allow the removal of cosmic rays five 1800-second off - nuclear exposures were obtained for each object , along with a shorter exposure of the quasar itself . the spectra typically spanned wavelengths from 3500 to 6000 , although the precise range varied from object to object according to redshift ; details for individual objects are given in table 2 . data reduction was carried out using standard iraf routines . after the two runs at kitt peak had demonstrated that galaxy light could indeed be separated from that of the quasar and that useful spectra could be obtained , further observations were made of 22 objects ( 6 of which had already been observed at kitt peak ) with the intermediate dispersion spectroscopic and imaging system ( isis ) at the 4.2 m william herschel telescope ( wht ) on la palma . isis uses a dichroic mirror to split the incoming light into red and blue beams which are then treated separately in spectrographs which have been optimised for the appropriate wavelength ranges ( @xmath25 for the blue arm and @xmath26 for the red ) . this arrangement allows a larger wavelength coverage than is possible with the single spectrograph on the mayall 4-m , enabling us to extend our spectra further into the red and thus giving greater scope for constraining models of spectrophotometric evolution . the data were obtained in four separate observing runs in 1993 november , 1994 may , 1994 september and 1995 march . the instrumental set - up differed slightly between the runs , resulting in variations in the wavength ranges obtained . for the two sessions in 1994 the detectors in both the blue and red arms were tektronix ( ` tek ' ) ccds , with 24-@xmath11 m pixels . in november 1993 and march 1996 the tektronix chip in the red arm was replaced by an eev p88300 chip , with 22.5-@xmath11 m pixels , which allowed a slightly larger wavelength coverage . r158 gratings were used in each arm , giving a spectral resolution of 2.88 pixel@xmath9 for the blue tek chip and 2.90 ( tek ) or 2.72 pixel@xmath9 ( eev ) for the red chips . as with the observations at kitt peak the slit was first centred on the quasar nucleus , or the optical peak of the radio galaxy , and then offset to the desired position , 5@xmath7 from the quasar , and rotated to be at right angles to the direction of the offset . the slit width was set to 2@xmath7 , placing the inner edge of the slit at least 4@xmath7 from the quasar position . once again exposures were limited to 1800 seconds duration . whenever possible we aimed to obtain five such frames per object , giving a total on - source exposure time of 2.5 hours . on - nuclear quasar spectra were also taken when time permitted . data reduction was carried out using the figaro package , part of the starlink suite of astronomical software . the galaxies observed with the wht tend to be fainter and/or at greater redshifts than those observed on the mayall 4-m at kitt peak , and so additional care needed to be taken during the reduction of the wht spectra . in particular , the procedure employed to optimize the extraction of the off - nuclear spectrum from the ccd frame means that the final flux calibration is only relative and not absolute - comparisons with the absolute fluxes obtained on the mayall 4-m are therefore not meaningful . the use of the red and blue arms of isis , whilst extending the wavelength coverage significantly , also introduces its own special problems . the large wavelength range made available by isis means that the spectra are affected by several prominent atmospheric emission features ( figure 1 ) which must be removed of these the strongest are the two oxygen lines at 5577 and 6300 , the sodium d line at 5890 , and the series of oh bands at wavelengths @xmath27 . sky lines were removed by fitting a third order polynomial to two empty strips of sky on either side of the target spectrum but the removal process sometimes left a residual imprint of these features and this constitutes a major source of noise in some of the fainter spectra in our sample . where this is the case the affected regions are mentioned in the description of the individual spectrum . a second problem involves the splicing together of the two spectra from the red and blue arms of the instrument to give continuous wavelength coverage from 3200 to 9000 . the reflection and transmission responses of the isis dichroic cross at @xmath28 and the instrument settings were designed to ensure a large overlap between the wavelength coverages of the two arms so that the full reflection / transmission curves could be followed in the blue and the red spectra respectively . however , the faintness of many of the spectra effectively causes the flux calibration to fail at the red end of the blue spectrum and the blue end of the red , producing an artificial ` lump ' in the spliced spectrum at the crossover point ( 6100 or 6050 , depending on the exact instrumental settings at the time of the observations - see column 8 in table 3 ) . accordingly , in plotting the wht spectra in figure 2 we have blanked out the data points for 100 on either side of the join region and replaced them with an averaged bridging section . we have masked out this section of the data in all subsequent modelling due to its inherrent unreliability ( see nolan et al . 2000 ) . due to the luminosity of the quasars themselves the on - nuclear spectra obtained with the wht do not suffer from this problem and the red and blue sections have been spliced together directly with no noticeable discontinuity in flux at the join . the galaxy spectra themselves are displayed in figure 2 along with nuclear spectra for the same object , where such are available . all the data have been smoothed to 10- bins , and the wht data are displayed with the region linking the individual red and blue spectra blanked out as described above . the expected position of various stellar absorption features , including the 4000 break are indicated by dotted lines in the off - nuclear spectra and the wavelengths of redshifted [ oiii]@xmath29 and h@xmath30 are also marked . a cursory examination of figure 2 is enough to show that the quality of the off - nuclear spectra varies considerably from object to object . in the two initial observing runs at kitt peak priority was given to nearby objects with bright , prominent host galaxies , and the eleven spectra obtained with the mayall 4-m enjoy a high signal - to - noise ratio . by contrast , the objects observed in later runs on the wht tend to have larger redshifts and the signal - to - noise in many of these spectra is correspondingly lower due to the rapid reduction in surface brightness and increase in galactocentric radius with @xmath15 . the primary goal of the observing program - to obtain spectra from the stellar component of the host whilst avoiding scattered emission from the active nucleus - appears to have been satisfied to a large extent . this can be demonstrated most easily by comparing the off - nuclear ( galaxy ) and nuclear spectra for the same object in figure 2 . the quasar spectra show prominent broad lines , notably those of h@xmath31 and h@xmath32 , with an increase in flux towards shorter wavelengths ( particularly when the ` blue bump ' continuum feature begins to emerge at @xmath33 ) . whilst emission lines do occur in the off - nuclear spectra , they tend to be relatively narrow forbidden lines such as [ oiii]@xmath34 . where permitted lines occur , they lack the extremely broad profiles seen in the quasar nuclei , and are thus unlikely to result from scattering - induced contamination by nuclear light . in many cases the line ratios in the off - nuclear spectra also differ from those measured in the quasars , indicating that the emission arises under different conditions than those prevailing in the active nucleus . however , in spectra obtained under conditions of poor seeing , there is likely to be some degree of nuclear contamination and this will be exacerbated if the object was observed far from the zenith , where differential atmospheric refraction may also lead to signifcantly more contamination at the blue end of our wavelength range than in the red . although the extent of such contamination is difficult to measure directly , tables 2 and 3 list the atmospheric seeing and airmass at the time each spectrum was obtained , to allow a rough assessment of the problem to be made . ccc iau & telescope & break + name & & strength + + 0007@xmath18106 & m4 m & @xmath35 + 0054@xmath18144 & m4 m & @xmath36 + & wht & @xmath36 + 0157@xmath18001 & m4 m & @xmath36 + & wht & @xmath36 + 0204@xmath18292 & wht & @xmath37 + 0244@xmath18194 & wht & @xmath38 + 1549@xmath18203 & wht & @xmath39 + 1635@xmath18119 & wht & @xmath40 + 2215@xmath19037 & wht & @xmath35 + 2344@xmath18184 & m4 m & @xmath40 + & wht & @xmath35 + + 0137@xmath18012 & m4 m & @xmath40 + 0736@xmath18017 & m4 m & @xmath35 + & wht & @xmath41 + 1004@xmath18130 & wht & @xmath42 + 1020@xmath19103 & m4 m & @xmath40 + 1217@xmath18023 & wht & @xmath42 + 2135@xmath19147 & wht & @xmath43 + 2141@xmath18175 & wht & @xmath44 + 2201@xmath18315 & m4 m & @xmath35 + 2247@xmath18140 & m4 m & @xmath45 + & wht & @xmath39 + 2349@xmath19014 & wht & @xmath46 + + 0230@xmath19027 & wht & @xmath47 + 0345@xmath18337 & wht & @xmath48 + 0917@xmath18459 & wht & @xmath39 + 1215@xmath19033 & wht & @xmath49 + 1330@xmath18022 & m4 m & @xmath37 + 1334@xmath18008 & wht & @xmath50 + 2141@xmath18279 & m4 m & @xmath51 + & wht & @xmath52 + the characteristic shape of the stellar continuum , including features such as the 4000 break and various stellar absorption lines , is easily recognisable in most cases . the 4000 break is particularly important for comparison between data and models of the stellar population . since it covers a large wavelength interval it is relatively insensitive to the effects of instrument resolution and noise ( hamilton 1985 ) . the break amplitude ( defined as the ratio of the average flux density between rest - frame 4050 and 4250 to that between 3750 and 3950 ) is therefore widely used as a tracer of spectral evolution ( _ eg _ bruzual 1983 ) . the discontinuity results from the combined effect , shortwards of 4000 , of lines of several elements heavier than helium , in a variety of ionization states , along with the crowding of higher order balmer lines . if a significant population of massive young stars is present the enhanced degree of ionization causes the feature to weaken ; it is most prominent in the spectrum of a well - established stellar population in which the most massive stars have had time to evolve away from the main sequence . hence the 4000 break is sensitive to both spectral type and metallicity , although if we assume a constant metallicity ( a reasonable assumption for a particular galaxy at a fixed radius ) it becomes a good indicator of the mean age of the local stellar population . the strength of this feature , as measured in each of the current spectra , is listed in table 4 . other stellar absorption features , such as g band ( 4300 to 4320 ) and the mg ib ( 5173 ) and fe@xmath53 lines , are also clearly present in many of the spectra . several of the off - nuclear spectra ( _ eg _ 0054 + 144 , 2201 + 315 ) , despite showing a clear break at 4000 and a continuum longwards of this wavelength which can be fitted extremely well by a passively ageing stellar population ( nolan et al . 2000 ) , also display a contribution from a component which rises steeply towards the blue . the slope of this feature closely resembles that of a quasar sed , and its presence is often ( but not always ) accompanied by emission lines characteristic of quasar nuclei , suggesting that it is in fact scattered light from the quasar itself , the result either of atmospheric scattering or ( since the feature is not always correlated with poor observing conditions ) of scattering within the ism of the host galaxy . another possibility is that the blue excess indicates the presence of a substantial population of young stars within the host galaxy . if this latter case were true then it would pose a serious problem for the unification of rlqs and rgs since none of the radio galaxies display such a component . the issue of excess blue continuum is raised on a case - by - case basis in the following section , but a full discussion in the light of detailed stellar population synthesis modelling is deferred to the companion paper by nolan et al . ( 2000 ) . the average values of the 4000 break strength for the three types of object in the sample are 1.4 ( rqq hosts ) , 1.5 ( rlq hosts ) and 1.7 ( rgs ) . these are all somewhat lower than the value of @xmath54 measured for local inactive elliptical galaxies by hamilton ( 1985 ) , but we note that there is a wide scatter in our sample and that the lowest values are all associated either with spectra in which the signal to noise is particularly poor ( _ eg _ 0244 + 292 , 1004 + 130 , 1334 + 008 ) or those which clearly show an extra source of continuum emission at short wavelengths ( _ eg _ the quasars 0054 + 144 , 1217 + 023 , 2141 + 175 ) . the cleanest spectra generally have break strengths which are consistent with those seen in ` normal ' well - established galaxies . as a final caveat we note that these spectra tell us only about the stellar composition of the region of the galaxy covered by the slit - we can not , for example , rule out the presence of a significantly different stellar population closer to the nucleus , or concentrated in clumps which the slit happens to avoid . however , the slit has a width of at least 2 arcsec and , as can be seen from the contour plots in figure 2 , its length cuts across a significant fraction of the galaxy in the transverse direction . the area covered often amounts to several square arcseconds ( equivalent to several tens of square kiloparsecs at typical redshifts ) and therefore represents a good general sample of the outer regions of the host . objects are listed under their iau names ( alternative names are listed in table 1 ) , in order of increasing right ascension . the classification of each object as either an rqq , an rlq or a radio galaxy is indicated in parentheses , along with the name(s ) of the telescope(s ) on which off - nuclear spectra were obtained . the form of the spectrum is described , noting any peculiar features as well as the presence or otherwise of a 4000 break at the expected observed wavelength ( @xmath55 ) . we also note the morphology of the galaxy ( disc or elliptical ) based on its surface brightness profile in the @xmath10 or @xmath13-band continuum images by taylor et al . ( 1996 ) ( @xmath10-band ; ukirt ) or mclure et al . ( 1999 ) and dunlop et al . ( 2000 ) ( @xmath13-band ; hst ) . for a more detailed description of previous imaging studies of the host galaxies see dunlop et al . ( 1993 ) ( rlqs and rqqs ) or taylor et al . ( 1996 ) ( rgs ) . * 0007@xmath18106 ( rqq ; m4 m ) : * the nuclear spectrum of this radio - quiet quasar shows prominent broad h@xmath56 and h@xmath57 emission as well as narrower forbidden line emission from species including [ 0iii ] and [ fevii ] . the off - nuclear spectrum of the host galaxy has a high signal - to - noise and displays little sign of contamination from the quasar : the contribution from emission lines is very small , and the 4000 break is clearly visible ( redshifted to 4356 ) despite the rapid increase in quasar flux towards the blue end of the spectrum , which would tend to mask the break if scattering were significant . g band and mg ib absorption are also present . we note that the slit crosses the optical arc - like structure to the north of the quasar which hutchings et al . ( 1984 ) suggest may be a spiral arm . previous spectroscopy of this region by green , williams & morton ( 1978 ) showed narrow emission lines ( with different line ratios from those in the nucleus ) and a red continuum which they attribute to starlight . however , this region only constitutes a small fraction of the galaxy light intercepted by the slit in the current observations . taylor et al . ( 1996 ) find that an exponential disc profile provides a good fit to the nir surface brightness distribution of the galaxy . ( morphology : disc ; taylor et al . 1996 . ) * 0054@xmath18144 ( rqq ; m4 m & wht ) : * quasar continuum emission dominates the nuclear spectrum of this rqq , although h@xmath58 and [ oiii]@xmath59 lines are visible . these lines are not prominent in either the mayall 4-m or wht off - nuclear spectra of the host ( particularly the latter spectrum ) but bluewards of the ( relatively weak ) 4000 break ( at @xmath60 ) the galaxy spectrum displays a marked increase in flux , very similar in form to that displayed by the quasar itself . this component is seen in both off - nuclear spectra , which were taken at different times , under different seeing conditions , and used different slit positions , so it is not clear whether we are seeing quasar light which is being scattered into our line of sight either by the atmosphere or by the interstellar medium of the host galaxy , or whether the blue excess is due to a population of young stars . g band , mg ib and fe5270 absorption features appear in the mayall 4-m spectrum . mclure et al . ( 1999 ) classify the galaxy as an elliptical : its light profile is very well described by an @xmath61 law and the galaxy itself is quite red ( @xmath62 ) , but a tidal interaction with a nearby companion is suggested by the extension to the nw of the nucleus ( dunlop et al . ( morphology : elliptical ; mclure et al . 1999 . ) * 0137@xmath18012 ( rlq ; m4 m ) : * the 4000 break in this object is quite clear ( at @xmath63 ) and there is little evidence of nuclear emission lines . ( morphology : elliptical ; mclure et al . 1999 . ) * 0157@xmath18001 ( rqq ; m4 m & wht ) : * the off - nuclear spectrum obtained with the mayall 4-m does not appear to be strongly contaminated by emission from the nucleus and the 4000 break is visible at @xmath64 . the g band absorption feature is also present . however , in the spectrum taken with the wht using a similar slit position , there appears to be a significant contribution from the quasar continuum bluewards of the break . the signal : noise ratio in the wht spectrum is much reduced longwards of 7500 due to residuals from the subtracted oh bands . in both cases the slit intercepts the prominent tidal arm which extends north and nw of the nucleus ( mackenty & stockton 1984 ) and is known to contain several emission line regions ( stockton & mackenty 1987 ) , although emission lines are not strongly evident in the integrated spectra presented here . previous spectroscopy of this structure by heckman et al . ( 1984 ) showed a velocity difference of @xmath65 km s@xmath9 between the arm and the quasar itself . mclure et al . ( 1999 ) find that the underlying smooth @xmath13-band continuum light is well described by an @xmath61-law , though conceivably this might be another result of the tidal interaction . ( morphology : elliptical ; mclure et al . 1999 . ) * 0204@xmath18292 ( rqq ; wht ) : * the 4000 break ( @xmath66 ) is quite strong in this wht spectrum , but residuals from oxygen and oh features from the sky spectrum are also present . ( morphology : disc ; taylor et al . 1996 ) * 0230@xmath19027 ( rg ; wht ) : * the off - nuclear spectrum of this small , faint radio galaxy suffers from low signal to noise and residual sky features are visible . however a break can be seen at @xmath67 . the nuclear spectrum of this object is also dominated by starlight , although narrow emission lines are present . ( morphology : elliptical ; dunlop et al . 2000 . ) * 0244@xmath18194 ( rqq ; wht ) : * the signal to noise in this off - nuclear spectrum is very low . apart from residual sky features there is some evidence for a drop in the continuum level around @xmath68 ( the expected position of the 4000 break ) , but also for a blue component shortwards of this , perhaps indicative of nuclear contamination . ( morphology : elliptical ; mclure et al . 1999 . ) * 0345@xmath18337 ( rg ; wht ) : * the spectrum is noisy , but a break feature appears to be present at @xmath69 . the wht slit intercepts a bright knot nw of the quasar which taylor et al . ( 1996 ) suggest may be an embedded companion galaxy . ( morphology : elliptical ; mclure et al . 1999 . ) * 0736@xmath18017 ( rlq ; m4 m & wht ) : * the wht spectrum , though noisier and suffering from oh - band residuals at long wavelengths , agrees well with the spectrum obtained previously at kitt peak . the 4000 break is quite clear at @xmath70 and weak [ oiii ] , h@xmath30 and h@xmath71 features can also be seen , though they appear to be narrower than those in the nuclear spectrum . the galaxy itself is highly disturbed , but mclure et al . ( 1999 ) fit an @xmath61-law profile to the smooth component of the @xmath13-band continuum . ( morphology : elliptical ; mclure et al . 1999 . ) * 0917@xmath18459 ( rg ; wht ) : * the off - nuclear spectrum is pure stellar continuum , with a strong break at @xmath72 . the galaxy isophotes are complex but mclure et al . ( 1999 ) find the underlying distribution to be well described by an @xmath61-law . ( morphology : elliptical ; mclure et al . 1999 . ) * 1004@xmath18130 ( rlq ; wht ) : * oh - band residuals are the only prominent feature of this off - nuclear spectrum , with little evidence for either emission lines or a 4000 break ( at @xmath73 ) . the absence of a strong break may reflect the fact that the elliptical host galaxy of this quasar is known to possess unusual ` spiral ' features close to the nucleus ( mclure et al . 1999 ) , perhaps indicating the presence of a significant population of young stars . stockton & mackenty ( 1987 ) note that there is no significant extended [ oiii ] emission in this object . ( morphology : elliptical ; mclure et al . 1999 . ) * 1020@xmath19103 ( rlq ; m4 m ) : * despite the presence of stellar absorption features such as g band and a 4000 break at @xmath74 , the off - nuclear spectrum also contains many emission lines as well as a blue excess , which may indicate a significant contribution from scattered quasar light ( the seeing was quite poor for much of the observations ) . dunlop et al . ( 1993 ) note that the quasar is off - centre and that the galaxy isophotes appear to be swept back towards the sw , providing evidence for disturbance in this object . ( morphology : elliptical ; dunlop et al . 1999 . ) * 1215@xmath19033 ( rg ; wht ) : * this spectrum displays only a weak break at @xmath75 . oh - band residuals dominate longwards of 7500 . ( morphology : elliptical ; dunlop et al . 2000 . ) * 1217@xmath18023 ( rlq ; wht ) : * a weak break feature is present at @xmath73 but shortwards of this the spectrum is dominated by a component which rises towards the blue possibly indicating scattered nuclear continuum ( there is however little evidence for accompaning nuclear line emission , and the seeing during the observations was excellent ) . at the red end , poor subtraction of oh bands has reduced the signal to noise ratio of the spectrum . ( morphology : elliptical ; dunlop et al . 2000 . ) * 1330@xmath18022 ( rg ; m4 m ) : * weak [ oiii ] lines occur in the off - nuclear spectrum of this radio galaxy and the stellar continuum shows g band absorption and a strong 4000 break feature at @xmath76 . by contrast , the nuclear spectrum shows evidence for broad h@xmath77 and the less prominent break may indicate a contribution from a quasar - type continuum or perhaps a nuclear starburst region . ( morphology : elliptical ; dunlop et al . 2000 . ) * 1334@xmath18008 ( rg ; wht ) : * a very noisy spectrum obtained under poor conditions , the underlying continuum is confused by many residual sky features . the apparent increase in flux shortwards of 4500 almost certainly reflects a failure of the flux calibration at very low light levels . however , there is some evidence for a break at the expected wavelength of @xmath78 . the slit intercepts one of the secondary nuclei reported by taylor et al . ( morphology : elliptical ; taylor et al . 1996 . ) * 1549@xmath18203 ( rqq ; wht ) : * the extended nebulosity around this rqq , though confused by a foreground galaxy cluster , shows a strong break feature at the expected wavelength of @xmath79 and little evidence for emission lines . oh - band residuals add to the noise levels at long wavelengths . ( morphology : elliptical ; dunlop et al . 2000 . ) * 1635@xmath18119 ( rqq ; wht ) : * the 4000 break is clearly visible at @xmath80 , but residual sky features degrade the quality of the spectrum towards the red end . ( morphology : elliptical ; mclure et al . 1999 . ) * 2135@xmath19147 ( rlq ; wht ) : * generally low signal to noise with prominent residuals due to sky features . however , there is weak evidence for a break at @xmath81 . the galaxy appears to be disturbed and possesses a secondary nucleus to the se of the quasar ( stockton 1982 ) which may also be active ( hickson & hutchings 1987 ) . the wht slit intercepts a region to the sw of the quasar at which stockton & mackenty ( 1987 ) report extended [ oiii ] emission . narrow h@xmath30 is present in the current spectrum but the [ oiii ] lines fall within the join region where the signal is unreliable . ( morphology : elliptical ; dunlop et al . 2000 . ) * 2141@xmath18175 ( rlq ; wht ) : * low signal to noise and the presence of a strong blue component may serve to mask any evidence of a 4000 break in this object ( expected at @xmath82 ) . the idea that this blue component originates as scattered quasar light is bolstered by the possible presence of the h@xmath30 line , but the issue is confused by strong sky residuals ( although the seeing was good , the airmass during the observations was relatively high , so differential refraction might explain the presence of quasar contamination at shorter wavelengths without requiring the presence of a strong h@xmath30 line ) . the elongated appearance of the galaxy is apparently the result of an edge - on tidal arm consisting of old stars ( stockton & farnham 1991 ) . however , the wht slit crosses the galaxy to the ne of the quasar , where the starlight appears to follow a bulge - dominated @xmath61-law ( mclure et al . ( morphology : elliptical ; mclure et al . 1999 . ) * 2141@xmath18279 ( rg ; m4 m & wht ) : * the off - nuclear spectrum taken at kitt peak shows a weak [ oiii]@xmath83 line , a break feature at @xmath76and mg ib absorption . the wht spectrum is much noisier , but generally consistent with the features in the earlier data . the nuclear spectrum of this radio galaxy is also dominated by starlight , although prominent narrow lines are present . ( morphology : elliptical ; mclure et al . 1999 . ) * 2201@xmath18315 ( rlq ; m4 m ) : * this object shows prominent g band absorption but only a weak 4000 break at @xmath84 , along with low - equivalent - width [ oiii ] lines . a blue component shortwards of the break is consistent with scattering of the nuclear quasar continuum . ( morphology : ambiguous ; taylor et al . 1996 . ) * 2215@xmath19037 ( rqq ; wht ) : * the spectrum is relatively noisy with prominent residuals from all the bright sky features . the apparent ` hump ' at @xmath85 is an artifact that results from the sodium d sky line coinciding with the beginning of the masking region that links the red and blue halves of the the spectra from isis . there is little evidence for a prominent break at the expected wavelength of @xmath86 . both hutchings et al . ( 1989 ) and mclure et al . ( 1999b ) classify the galaxy as a non - interacting elliptical system . ( morphology : elliptical ; dunlop et al . 2000 . ) * 2247@xmath18140 ( rlq ; m4 m & wht ) : * the galaxy s 4000 break is visible at @xmath87 in the mayall 4-m spectrum , along with g band absorption . the data obtained on the wht are consistent with this , although residual sky features from oxygen and sodium d lines and the oh bands are also present . ( morphology : elliptical ; mclure et al . 1999 . ) * 2344@xmath18184 ( rqq ; m4 m & wht ) : * both the mayall 4-m and wht spectra show a break at @xmath88 as well as g band absorption and are generally consistent with one another despite their differing slit positions . the nuclear spectrum of this object is also dominated by starlight ; 2344@xmath18184 is one of the least powerful quasars in the current study and technically qualifies as a type 1 seyfert galaxy . the surface brightness profile is that of a disc galaxy , although a bulge dominates in the central regions ( mclure et al . 1999 ) and there is also evidence for a bar ( hutchings , janson & neff 1989 ) . ( morphology : disc ; mclure et al . 1999 . ) * 2349@xmath19014 ( rlq ; wht ) : * there is a weak break feature at @xmath89 and also evidence for weak h@xmath71 , [ oiii]@xmath83 and h@xmath30 lines . the red end of the spectrum is dominated by residual atmospheric oh band emission . ( morphology : elliptical ; mclure et al . 1999 . ) six objects in the current sample ( the rqqs 0054@xmath18144 , 0157@xmath18001 and 2344@xmath18184 , the rlqs 0736@xmath18017 and 2247@xmath18140 , and the radio galaxy 2141@xmath18279 ) were observed with both the mayall 4-m telescope and the wht . this duplication allows us to check for systematic differences between the spectra obtained with each instrument and to this end the are replotted one above the other in figure 3 . since the observations were often made under different atmospheric conditions they also provide a means of assessing the degree to which any contamination by nuclear light can be attributed to either the airmass and seeing conditions at the time of the observations or scattering within the host galaxy itself . where a different slit position was used , the two spectra allow us to examine the degree of homogeneity in the stellar composition of the galaxy . it should be noted that due to the optimization method used to extract the wht spectra , their flux calibration can only be considered as relative , not absolute . different slit widths were also used on the two instruments . comparisons of the flux densities obtained on the two telescopes are therefore not meaningful . * 0054@xmath18144 ( rqq ) : * different slit positions were used for the mayall 4-m and wht observations . the spectrum obtained with the mayall 4-m shows a contribution from h@xmath71/[oiii ] at @xmath90 , which is lacking in the wht data . however , the underlying continuum is very similar in both spectra and the measured depth of the ( weak ) 4000 break is very similar in each . this implies that the emission lines detected in the mayall 4-m spectrum are a local feature , produced _ in situ _ rather than being due to scattered light from emission - line regions in the nucleus , since they do not appear to be accompanied by a corresponding amount of scattered nuclear continuum . the consistency in the strength of the 4000 break at the two observing epochs , along with its relative weakness , suggests the presence of a significant population of young stars in the host galaxy . mclure et al . ( 1999 ) report a tidal feature visible in their @xmath13-band hst image of the elliptical host galaxy which may be linked to the origin of this young stellar population . * 0157@xmath18001 ( rqq ) : * the slit positions used in the two sets of observations were essentially the same . however , the wht spectrum shows excess blue continuum shortwards of this feature . the two spectra were taken under what were ostensibly very similar atmospheric conditions , but we note that a very good approximation of the wht spectrum can be obtained simply by adding a scaled version of the nuclear spectrum to the off - nuclear mayall 4-m data . this suggests that atmospheric scattering is to blame for the blue excess in the wht spectra . * 0736@xmath18017 ( rlq ) : * the two spectra were both taken under good atmospheric conditions and using the same slit position . the agreement between the two is excellent . * 2141@xmath18279 ( rg ) : * the wht spectrum has a much lower signal : noise ratio than the mayall 4-m spectrum , and uses a different slit position . however , the overall continuum shape is in good agreement with the earlier data . both slits intercept an extension to the ne of the nucleus which may be a tidal feature caused by interaction with a northern companion . * 2247@xmath18140 ( rlq ) : * the same slit position was used for both spectra , although the seeing during the wht run was quite poor ( @xmath91 ) . the two datasets are consistent with one another , with a relatively strong 4000 break and little sign of an additional blue continuum component from either young stars or scattered quasar light . * 2344@xmath18184 ( rqq ) : * different slit positions were used on the two different instruments , and the wht spectrum shows an excess of blue emission and a correspondingly weaker 4000 break . the seeing conditions were actually worse during the m4 m observations , making atmospheric scattering of quasar light an unlikely culprit . moreover , we note that the nuclear spectrum of this low - luminosity quasar is not particularly blue and itself shows prominent stellar continuum features ( see figure 2 ) . more likely is that the wht slit crossed a region of the galaxy containing a large number of young stars . the host of 2344@xmath18184 is a disc galaxy , with a central bulge and prominent spiral arms ( mclure et al . ( hutchings et al . ( 1989 ) also suggest the presence of a bar . ) we describe and present optical spectra of 26 galaxies hosting powerful nuclear activity . the sample contains carefully matched subsamples of all three types of powerful agn - radio - quiet and radio - loud quasars , and frii radio galaxies - enabling us to investigate the relationship between the host galaxy and the radio properties of the resident agn and also to test unified models for radio - loud objects . the spectra were taken 5 arcsec _ off - nucleus _ and , via a careful choice of slit position , aim to maximise the amount of galaxy light entering the instrument whilst avoiding contamination from the active nucleus . in the majority of cases this approach appears to have been successful ; the continuum is clearly overwhelmingly stellar in origin and even when the presence of scattered nuclear emission is suspected , features such as the 4000 break are still clearly discernable . in almost all objects we detect at least some stellar signatures . a second paper ( nolan et al . 2000 ) will discuss spectrophotometric modelling of the galaxy spectra in order to determine the ages and starformation histories of their constituent stellar populations . the authors would like to thank the staff at kpno and the ing for their assistance during the observations , and the anonymous referee whose comments and suggestions resulted in several improvements in the final paper . dhh and mjk acknowledge pparc support . 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 . bahcall j. n. , kirhakos s. & schneider d. p. , 1994 , apj , 435 , l11

we define relative versions of the classical invariants of legendrian and transverse knots in contact 3-manifolds for knots that are homologous to a fixed reference knot . we show these invariants are well - defined and give some basic properties .

0909.4316

a smooth oriented knot ( an embedding of @xmath0 ) in a contact 3-manifold @xmath1 is called _ legendrian _ if it is everywhere tangent to the contact planes @xmath2 and _ transverse _ if it is everywhere transverse to @xmath2 . null - homologous legendrian knots in a closed contact 3-manifold have two classical invariants , the thurston - bennequin invariant and the rotation number . null - homologous transverse knots in a contact 3-manifold have a classical invariant , the self - linking number . our goal is to generalize these ideas to the case when @xmath3 is not null - homologous in @xmath4 ( that is , @xmath5\neq 0\in h_1(m;\z)$ ] ) , in particular , to the case when @xmath3 is homologous to another fixed knot @xmath6 . we prove the following theorem . [ main ] consider homologous knots @xmath3 and @xmath6 in a contact 3-manifold @xmath1 with @xmath7 for an embedded oriented surface @xmath8 and orient @xmath3 and @xmath6 as boundary components of @xmath8 so that @xmath9=[k]-[j]$ ] . a. if @xmath3 and @xmath6 are legendrian , then the relative thurston - bennequin number of @xmath3 relative to @xmath6 is well - defined . b. if @xmath3 and @xmath6 are legendrian , then the relative rotation number of @xmath3 relative to @xmath6 is well - defined up to an integer @xmath10 such that @xmath11)$ ] , where @xmath12\in h_2(m , k\cup j)$ ] . for a tight contact 3-manifold @xmath1 , this integer ambiguity vanishes so the relative rotation number is well - defined . c. if @xmath3 and @xmath6 are transverse , then the relative self - linking number of @xmath3 relative to @xmath6 is well - defined up to an integer @xmath10 such that @xmath11)$ ] , where @xmath12\in h_2(m , k\cup j)$ ] . for a tight contact 3-manifold @xmath1 , this integer ambiguity vanishes so the relative self - linking number is well - defined . in order to prove theorem [ main ] , we first discuss the dependence of each relative invariant on the seifert surface for @xmath3 and @xmath6 . then we show that although legendrian ( resp . transverse ) isotopies do not preserve the homology type of @xmath3 in @xmath13 , they do not change the values of the respective relative invariants . since the contact structure @xmath2 is trivial over the compact seifert surfaces with boundary , we do not require that @xmath2 be globally trivializable over @xmath4 . i would like to express deep gratitude to john etnyre for his guidance and help with many fundamental and technical aspects of this work . i am also grateful to my advisor danny ruberman for his patience and support throughout my graduate years when the ideas of this paper were developed . i would like to also thank vladimir chernov and jeremey van horn - morris for helpful discussions and correspondence . if @xmath14 is a null - homologous legendrian knot in a contact 3-manifold , then from the exact sequence for the relative homology of the pair @xmath15 , we see that there exists a relative class @xmath16 such that @xmath17 maps to @xmath5\in h_1(k;\z)$ ] under the boundary map @xmath18 , moreover , there exists a smooth embedded oriented seifert surface @xmath19 for @xmath3 such that @xmath12=\alpha$ ] . if @xmath14 is a null - homologous legendrian knot in a closed oriented contact 3-manifold , then the _ thurston - bennequin invariant _ of @xmath3 is defined as @xmath20 , which measures the number of @xmath21-twists ( with sign ) of the contact framing with respect to the seifert framing along @xmath3 . equivalently , one can define @xmath22 as @xmath23 , where @xmath24 is a push - off of @xmath3 in the direction normal to the contact planes along @xmath3 ( since @xmath2 is trivial over @xmath8 and , in particular , along @xmath25 , see @xcite , it is coorientable along @xmath3 ) . the linking number @xmath26 is well - defined and equals to an integer @xmath27 , where @xmath28 is generated by a meridional disc with homology class @xmath29 $ ] and @xmath30\in h_1(m\setminus k)$ ] is in this kernel with @xmath30=n[\mu]$ ] . the value of @xmath22 depends only on the relative homology class @xmath12\in h_2(m , k;\z)$ ] since the seifert framing for null - homologous @xmath3 is well - defined , so @xmath31 for any @xmath32 with @xmath33=[\sigma]$ ] , and for a choice of another @xmath34\in h_2(m , k;\z)$ ] , @xmath35 , which could be nontrivial if @xmath36 . additionally , for a null - homologous legendrian knot @xmath3 in @xmath1 , the _ rotation number _ @xmath37 is defined for a trivialization of @xmath38 , given by a nonzero section @xmath39 , whose restriction to @xmath3 yields a non - zero vector field @xmath40 tangent to @xmath3 . the rotation number @xmath37 of @xmath3 with respect to this trivialization is given by @xmath41)$ ] , where @xmath42 is the euler class of @xmath2 ( for a choice of coorientation of @xmath2 ) and @xmath43 is the relative euler class of @xmath44 . equivalently , for a non - zero vector field tangent to @xmath3 , the rotation number is the obstruction of extending @xmath40 to a nonzero vector field in @xmath38 . more directly , one can compute the rotation number as the winding number @xmath45 of the image of @xmath40 in @xmath46 under the map @xmath47 which is the restriction of the trivialization @xmath38 to @xmath3 . the rotation number does not depend on the trivialization @xmath48 ( so we will omit it from the notation ) , but it does depend on the relative homology class @xmath12\in h_2(m , k;\z)$ ] and on the orientation of @xmath8 . for a choice of another relative homology class @xmath33\in h_2(m , k;\z)$ ] , we have @xmath49-[\sigma])$ ] , where @xmath33-[\sigma]\in h_2(m;\z)$ ] . for a null - homologous transverse knot @xmath14 in an oriented contact 3-manifold @xmath4 , let @xmath19 be a seifert surface for @xmath3 , then we can define the _ self - linking number _ of @xmath3 as @xmath50 , where @xmath24 is a push - off of @xmath3 in the direction of ( the restriction along @xmath3 of ) a nonzero vector field @xmath51 on @xmath8 ( such a vector field exists since @xmath38 is trivial ) . the self - linking number can be interpreted as the obstruction of extending a non - zero vector field @xmath40 ( which points out of @xmath8 ) along @xmath3 in @xmath52 to a non - zero vector field in @xmath38 , in which case a push - off @xmath24 of @xmath3 along @xmath40 would have linking @xmath53 with @xmath3 relative to @xmath8 . similarly to the rotation number , upon fixing a coorientation of the contact plane field , a choice of a different relative homology class @xmath33\in h_2(m , k;\z)$ ] gives us @xmath54-[\sigma])$ ] , where the sign depends on whether the orientation of @xmath3 coincides with the transverse orientation ( or coorientation ) of @xmath2 . for more details , read @xcite . we are considering homologous oriented knots @xmath3 and @xmath6 in a contact 3-manifold @xmath4 . equivalently , the oriented link @xmath55 is null - homologous in @xmath4 , so the usual argument generalizes directly to show there exists a smooth embedded surface @xmath19 , where @xmath12\in h_2(m , k\cup j;\z)$ ] is mapped to @xmath56\in h_1(k\cup j;\z)$ ] by the boundary homomorphism , that is , @xmath8 is a seifert surface for @xmath55 . always orient @xmath3 and @xmath6 as boundary components of the oriented @xmath8 so that @xmath9=[k]-[j]$ ] . we generalize the classical invariants to the case when @xmath3 is homologous to another knot @xmath6 in @xmath1 as follows . let @xmath3 and @xmath6 be homologous legendrian knots in a contact 3-manifold @xmath1 oriented accordingly with @xmath7 for an oriented embedded seifert surface @xmath8 so that @xmath9=[k]-[j]$ ] . define the _ thurston - bennequin invariant of @xmath3 relative to @xmath6 _ by @xmath57 where @xmath58 denotes the seifert framing that @xmath3 ( resp . @xmath6 ) inherits from @xmath8 , and @xmath59 denotes the number of @xmath21-twists ( with sign ) of the contact framing relative to @xmath58 along @xmath3 or @xmath6 . for push - offs @xmath24 and @xmath60 of @xmath3 and @xmath6 in the direction normal to the contact planes , @xmath61 . let @xmath3 and @xmath6 be homologous legendrian knots in a contact 3-manifold @xmath1 oriented accordingly with @xmath7 for an oriented embedded seifert surface @xmath8 so that @xmath9=[k]-[j]$ ] . the restriction to @xmath3 of the trivialized contact 2-plane field @xmath38 gives a map @xmath62 , under which a non - zero tangent vector field @xmath40 to @xmath3 traces out a path of vectors in @xmath46 . we can then compute the winding number @xmath45 and similarly for @xmath6 . then define the _ relative rotation number of @xmath3 _ by @xmath63 equivalently , @xmath64)$ ] . let @xmath3 and @xmath6 be homologous transverse knots in a contact 3-manifold @xmath1 oriented accordingly with @xmath7 for an oriented embedded seifert surface @xmath8 so that @xmath9=[k]-[j]$ ] . the contact 2-plane field @xmath2 is trivial over @xmath8 , so there exists a nonzero vector field @xmath51 in @xmath38 . take @xmath24 and @xmath60 to be the push offs of @xmath3 and @xmath6 in the direction of @xmath51 . then define the _ relative self - linking number _ of @xmath3 with respect to @xmath6 by @xmath65 next we give some basic properties of the relative invariants and discuss their dependence on the seifert surface . let @xmath3 , @xmath24 , and @xmath6 be oriented legendrian knots in a contact 3-manifold accordingly with @xmath7 so that @xmath9=[k]-[j]$ ] for an oriented embedded surface @xmath8 , and let @xmath24 be homologous to @xmath3 and @xmath6 via a surface @xmath32 oriented accordingly with @xmath66 so that @xmath67=[k]-[k']$ ] and assume @xmath68 is and embedded surface oriented with @xmath69 so that @xmath70=[-k']-[-j]$ ] . assume that @xmath8 , @xmath32 and @xmath68 are disjoint away from their common boundaries . then @xmath71 note also that @xmath72 , and @xmath73 . to see this , check that @xmath74 and @xmath75 are independent of the orientations of @xmath3 and @xmath6 . under ( de)stabilization ( see @xcite ) , we have @xmath76 , therefore @xmath77 and @xmath78 , where @xmath32 denotes the surface after ( de)stabilization . [ framing ] consider a contact 3-manifold @xmath1 and homologous legendrian knots @xmath3 and @xmath6 in @xmath4 oriented accordingly with @xmath7 so that @xmath9=[k]-[j]$ ] , where @xmath19 is an oriented embedded surface . then the seifert framing @xmath58 on @xmath3 ( resp . , on @xmath6 ) is uniquely defined by the relative homology class @xmath12\in h_2(m , k\cup j;\z)$ ] . take neighborhoods @xmath79 and @xmath80 , we have the maps ( with @xmath81-coefficients ) @xmath82 @xmath83 the map @xmath84 is an isomorphism because @xmath85 deformation retracts onto @xmath86 . the map @xmath87 is an isomorphism from excising @xmath88 , and @xmath89 is the usual boundary map on relative homology . let @xmath90 , then @xmath91)$ ] gives the homology classes @xmath30\in h_1(\partial n_k)$ ] and @xmath92\in h_1(\partial n_j)$ ] of corresponding push - offs of @xmath3 and @xmath6 into @xmath8 . so @xmath30 $ ] and @xmath92 $ ] are uniquely defined by @xmath12\in h_2(m , k\cup j)$ ] . the following states that framings on @xmath3 and @xmath6 induced by different relative homology classes in @xmath93 differ by the same number of twists along each boundary component . [ independent ] consider knots @xmath3 and @xmath6 oriented accordingly with @xmath94 for an oriented seifert surface @xmath95 in a 3-manifold @xmath4 so that @xmath96=[k]-[j]$ ] . then for any other oriented seifert surface @xmath97 with @xmath98 so that @xmath99=[k]-[j]$ ] , @xmath100 consider a homology class @xmath101\in h_2(m , j\cup k)$ ] and a surface @xmath97 with @xmath101=[\sigma_2]\in h_2(m , j\cup k)$ ] , then @xmath101-[\sigma_2]$ ] is a class in @xmath102 . so any other surface for @xmath55 is obtained from @xmath95 by adding on a closed surface @xmath103 of some homology class @xmath104\in h_2(m)$ ] . we construct a surface @xmath105 with boundary @xmath106 representing the class @xmath101+[a]$ ] and show that the above equality is true for @xmath105 . first , observe that @xmath107 for any @xmath108 $ ] . we can ensure this is true by modifying the interior of @xmath109 as follows . replace small closed 2-discs in @xmath109 around consecutive intersection points ( with opposite signs ) of @xmath109 along @xmath6 with a cylinder whose boundary components are the boundaries of the two 2-discs and which runs in a neighborhood of @xmath6 . this eliminates such intersection points in pairs . the intersections between @xmath8 and @xmath109 consist of @xmath110 ribbon arcs and possibly some circles on the interior of @xmath8 . we first eliminate the circle intersections by cutting - open the oriented surfaces along each circle and gluing up the pieces to obtain a new oriented surface ( there is an unique obvious way to do this , which depends on the orientations of the surfaces , see figure [ f10 ] below , in particular , reversing the orientation of one of the surfaces results in the alternate gluing up of the cut open surfaces ) . and @xmath109 . ] next , consider the ribbon intersection arcs and label them @xmath111 . note that each arc @xmath112 runs on @xmath8 from a point @xmath113 to a point @xmath114 with @xmath115 , so that on @xmath109 , @xmath116 and @xmath117 are interior points and @xmath112 is an interior arc ( see figure [ f1 ] ) . to resolve the @xmath112 , cut along the interior of each and glue the surfaces ( only one way to do this so that we get an oriented surface , see figure [ f11 ] below ) and note that the twists at each endpoint of the arc are opposite . lemma [ independent ] implies that the relative thurston - bennequin invariant is independent not only of the seifert surface but also of the relative homology class . this will be useful when proving invariance under legendrian isotopies . for the relative rotation number , consider legendrian knots @xmath3 and @xmath6 oriented as boundary components of an oriented embedded surface @xmath8 , we have @xmath118 , @xmath119 . under ( de)stabilization of @xmath3 , only the winding number along @xmath3 would change by @xmath120 , therefore , @xmath121 , where @xmath32 denotes the new surface after ( de)stabilization . the relative rotation number does not depend on the trivialization of @xmath38 but it does depend on the seifert surface through its relative homology class @xmath12\in h_2(m , k\cup j;\z)$ ] and its orientation . and @xmath109 adds one twist along each @xmath6 and @xmath3 , with opposite signs . ] [ trivializationdependence ] for homologous legendrian knots @xmath3 and @xmath6 in a contact 3-manifold @xmath1 oriented accordingly with @xmath7 for an embedded oriented surface @xmath19 so that @xmath9=[k]-[j]$ ] , the relative rotation number @xmath118 does not depend on the choice of trivialization of @xmath38 . consider two trivializations @xmath122 given by two nontrivial sections @xmath123 and @xmath124 so that there exists a function @xmath125 such that @xmath126 for all @xmath127 . therefore , the winding number @xmath128 , where the degree of @xmath87 over @xmath3 can be computed by the induced homomorphism @xmath129 in homology . similarly , @xmath130 . in particular , since @xmath131 , we have that @xmath132 , so @xmath133 @xmath134 [ independent2 ] let @xmath3 and @xmath6 be homologous legendrian knots oriented as @xmath135 for oriented seifert surfaces in a contact 3-manifold @xmath4 with relative homology classes @xmath101,[\sigma_2]\in h_2(m , k\cup j;\z)$ ] so that @xmath136=[k]-[j]$ ] . then @xmath137-[\sigma_2].\ ] ] notice that if @xmath101=[\sigma_2]$ ] , the lemma shows that the relative rotation number is independent of the particular relative homology representative . this is a direct generalization of the classical argument ( see @xcite ) . consider the closed oriented surface @xmath138 , the union taken along @xmath55 , which is oriented as the boundary of @xmath8 . the two trivializations @xmath139 and @xmath140 coincide along a segment @xmath141 on @xmath3 and a segment @xmath142 on @xmath6 , which means that we have a trivialization of @xmath2 over @xmath143 with a 2-disc @xmath144 removed and a 2-disc @xmath145 removed , where @xmath146 and @xmath147 . since @xmath3 and @xmath6 are oriented as boundary components of @xmath95 , there are compatible orientations for @xmath144 and @xmath145 which coincide with their natural orientation from @xmath143 . over @xmath144 and @xmath145 , there exist unique trivializations of @xmath2 up to homotopy . assume that the trivialization @xmath148 coincides along @xmath149 with the trivialization defined by the positive tangent vector @xmath40 to @xmath3 , and similarly , that the trivialization @xmath150 and the trivialization of @xmath2 over @xmath145 given by the positive tangent vector @xmath151 along @xmath6 . then by definition we have @xmath152 and @xmath153 , where @xmath154 denotes the trivialization @xmath155 . therefore , we have @xmath156 since @xmath157 , the above expression becomes @xmath158 first , consider @xmath159 . the first winding number is equal to the number of rotations of @xmath40 relative to @xmath160 along @xmath161 , which is oriented homotopic rel endpoints to the segment @xmath162 . the second winding number represents the number of rotations of @xmath163 along @xmath164 , which is homotopic rel boundary to the segment @xmath165 . we have analogous interpretation of the term @xmath166 . next take a segment @xmath167 from a point on @xmath168 to a point on @xmath169 and consider a neighborhood @xmath170 of @xmath17 , whose boundary consists of two arcs parallel to @xmath17 and of an arc @xmath171 along @xmath168 such that @xmath172 rotates once and , similarly , an arc @xmath173 along @xmath169 such that @xmath174 also rotates once . since @xmath175 is trivial , @xmath2 rotates in the same way along @xmath171 and @xmath173 , however , the orientations of @xmath171 and @xmath173 are reversed by an orientation - preserving isotopy between them through the product structure @xmath176\cong n(\alpha)$ ] . therefore , @xmath170 does not modify the value of the quantity @xmath158 then the euler number @xmath177-[\sigma_2]$ ] measures the number of rotations of the trivialization of @xmath2 over @xmath178 along the boundary of the 2-disc @xmath179 relative to a constant vector field which extends @xmath40 and @xmath180 over @xmath170 . this is equal to the number of rotations of the trivialization of @xmath2 over @xmath181 over @xmath182 with respect to extensions of @xmath40 and @xmath180 , respectively . therefore , @xmath183-[\sigma_2])$ ] . a direct way of obtaining this result is to choose @xmath154 so that they are both zero along @xmath6 so the argument directly reduces to the case when we compare the rotations of the trivializations only along @xmath3 , which is just the classical case . for the relative self - linking number , consider the transverse knots @xmath3 and @xmath6 oriented as boundary components of the oriented surface @xmath8 , over which @xmath2 is trivial and , in particular , it is cooriented along @xmath184 . then @xmath185 depends on how the orientations of the components of @xmath184 compare with orientation of ( the normal direction of ) the contact planes , so reversing the orientation of either @xmath8 or @xmath38 causes @xmath185 to reverse sign . [ independent3 ] consider a contact 3-manifold @xmath1 and homologous transverse knots @xmath3 and @xmath6 in @xmath4 oriented as @xmath7 for an embedded oriented surface @xmath19 so that @xmath9=[k]-[j]$ ] . the relative self - linking number @xmath185 does not depend on the choice of trivialization of @xmath38 . follows by the same argument as in the proof of lemma [ trivializationdependence ] . let @xmath95 and @xmath97 be seifert surfaces for the homologous oriented transverse knots @xmath3 and @xmath6 in a contact 3-manifold @xmath4 representing the relative homology classes @xmath101,[\sigma_2]\in h_2(m , k\cup j;\z)$ ] so that @xmath136=[k]-[j]$ ] . then @xmath186-[\sigma_2].\ ] ] notice that if @xmath101=[\sigma_2]$ ] , the lemma shows that the relative self - linking number is independent of the particular relative homology representative . the argument is similar to the proof of lemma [ independent ] . assume without loss of generality that the coorientation of @xmath2 makes @xmath3 positively transverse . again take the union @xmath187 and arrange that the trivializations of @xmath139 and @xmath140 coincide along an interval @xmath188 and an interval @xmath189 . thus we obtain a trivialization of @xmath2 over @xmath190 given by a nonzero vector field @xmath191 . note that @xmath192 is also trivial , the trivialization of @xmath148 given by a constant vector field @xmath40 tangent to @xmath193 and the trivialization of @xmath150 given by a constant vector field @xmath151 tangent to @xmath194 . then , as in lemma [ independent ] , one can use a neighborhood @xmath170 of an arc between @xmath168 and @xmath145 to reduce the computation of the trivialization of @xmath2 to the boundary of the 2-disc @xmath179 and see that it computes an euler number . this euler number @xmath177-[\sigma_2])$ ] measures the difference of the winding number of @xmath195 with respect to @xmath191 along @xmath168 minus the winding number of @xmath196 with respect to @xmath191 along @xmath169 ( note that @xmath168 and @xmath169 are oriented as boundaries of @xmath144 and @xmath145 on the oriented @xmath197 , so the negative sign comes form the fact that @xmath151 is pointing in the direction of @xmath198 ) . now let @xmath24 and @xmath60 be push - offs of @xmath3 and @xmath6 , respectively , in the direction of @xmath191 . then by definition , @xmath199 and @xmath200 from the construction , there is no contribution to @xmath201 along @xmath188 and there is no contribution to @xmath202 along @xmath189 for @xmath203 . then along @xmath161 , @xmath201 is equal to the winding number of @xmath191 relative to @xmath195 measured along this segment , because @xmath141 is homotopic , rel bondary , to @xmath162 , by an orientation - reversing homotopy , and @xmath141 is homotopic , rel bondary , to @xmath165 by an orientation - preserving homotopy . therefore , this computes @xmath204 similarly , to measure the winding number of @xmath191 relative to @xmath196 , we look at @xmath205 , keeping in mind that @xmath206 is homotopic to @xmath207 by an orientation - preserving homotopy and @xmath206 is homotopic to @xmath208 by an orientation - reversing homotopy . we are working with @xmath209 , so switching the signs , we obtain @xmath210 which computes precisely the last two remaining terms of the difference of the relative self - linking numbers coming from @xmath95 and @xmath97 , completing the proof . in this section , we study the problem of finding a seifert surface for oriented @xmath55 under smooth isotopies . given a surface @xmath211 with @xmath212 , consider a smooth isotopy @xmath213,$ ] of @xmath3 which fixes @xmath6 . if the isotopy lives in the complement of @xmath6 , then it extends to a global isotopy by the isotopy extension theorem , and , in particular , to an isotopy of @xmath8 through embedded surfaces . more formally , @xmath214 is a seifert surface for @xmath215 for all @xmath216 $ ] . note that there are versions of the isotopy extension theorem for transverse and legendrian isotopies . if @xmath217 intersects @xmath218 for some @xmath219 , we show how to construct a surface for @xmath215 after the intersection occurs . [ construction ] consider homologous knots @xmath3 and @xmath6 in a contact 3-manifold @xmath1 oriented as boundary components of an embedded oriented surface @xmath19 so that @xmath9=[k]-[j]$ ] , and consider a smooth isotopy @xmath220 of @xmath3 which fixes @xmath6 . additionally , assume that @xmath217 intersects @xmath6 transversely in a point @xmath221 , that is , @xmath222 and @xmath223 for all @xmath224 . for small @xmath225 , let @xmath226 and let @xmath227 , then there exists an oriented embedded surface @xmath228 with @xmath229 . since @xmath230 and @xmath231 are isotopic , they are homologous , so there exists an embedded oriented surface @xmath103 with @xmath232 , and we can arrange that @xmath233 . we construct @xmath234 out of the surfaces @xmath103 and @xmath235 by eliminating their intersections . intersection arcs that run boundary - to - boundary on one surface and intersect only the interior of the other surface are called _ ribbon arcs _ ( figure [ f1 ] ) . ribbon arcs can be eliminated by locally pushing the interior of the one surface ( e.g. , @xmath97 in figure [ f4 ] ) which contains the arc in its interior across the part of the other surface ( e.g. , @xmath95 in figure [ f4 ] ) that the arc bounds with the boundary ( this requires an innermost - arc - first procedure ) . and @xmath236 . ] if this part of surface @xmath97 contains any other intersections with @xmath95 ( circles ) , eliminate them first by locally isotoping the interior of @xmath97 across ( see figure [ f2 ] for an example involving @xmath235 and @xmath103 ) . . ] note that for this to be always possible , we need @xmath4 to be irreducible . parallel to @xmath230 . ] however , @xmath103 lives in a solid torus ( which is irreducible ) neighborhood and all the isotopies of the interior of @xmath235 can be restricted to live in this solid torus , so no such general assumptions for @xmath4 are necessary . so we assume there are no ribbon intersections and next eliminate non - ribon intersection arcs ( parallel to @xmath230 ) by isotoping the interior of @xmath235 locally across the half - disc that such arc bounds with @xmath230 ( see figure [ f3 ] ) . next we eliminate circle intersections between @xmath235 and @xmath103 , which we claim can not be multiples of @xmath237\in h_1(a;\z)$ ] . assume that @xmath238 is a multiple of @xmath239 $ ] . recall that @xmath240 , which produces the intersection arc @xmath17 that runs from @xmath6 to @xmath230 on @xmath235 and @xmath241 to the boundary @xmath230 on @xmath103 ( figure [ f4 ] ) . . ] take a framed neighborhood @xmath242 of @xmath103 so that @xmath230 is the inverse image of the core under a diffeomorphism from @xmath242 t the solid torus ( see @xcite or theorem 4.1.12 in @xcite ) with the framing given by @xmath243 . for a small @xmath225 , @xmath244 so @xmath245 is a parallel copy of @xmath230 . if there are multiple non - trivial circles , let @xmath246 denote the one `` closest '' to @xmath230 ( as in figure [ f4 ] ) . let @xmath247 denote the part of @xmath235 bounded by @xmath230 and @xmath246 . it has no other boundary components so let @xmath248 . this is a closed oriented surface . since @xmath6 is closed of complementary dimension to @xmath249 , their geometric intersection is 0 . so @xmath6 intersects @xmath249 at another point besides @xmath241 , and since @xmath240 , @xmath6 intersects @xmath247 . but @xmath235 is embedded and can not self - intersect . an intersection arc is a _ clasp _ if it runs from boundary - to - interior on each surface and a _ singular clasp _ if it runs boundary - to - boundary on one surface ( figure [ f5 ] ) . and a singular clasp @xmath250.,title="fig : " ] and a singular clasp @xmath250.,title="fig : " ] there are exactly @xmath251 clasps @xmath252 and exactly one singular clasp @xmath17 in @xmath253 ( @xmath17 is unique since another such arc @xmath254 would make @xmath255 a self - intersection of the embedded @xmath235 ) . clasps ( and singular clasps ) can be resolved standardly to give an embedded oriented surface ( see figures [ f6_0 ] and [ f6 ] ) . it is an _ orientation preserving cut _ ( see @xcite ) , a _ smoothing of a singular arc in an orientable way _ , or an _ umschaltung _ ( m. dehn ) . it consists of cutting open the interior of the clasp and joining the edges appropriately , whereby an endpoint of the clasp becomes a singular point . cut open at this point ( see figure [ f6 ] for an example ) to obtain an embedded oriented surface . consider the clasps @xmath252 . again , we take a framed neighborhood @xmath256 of @xmath230 with the framing @xmath243 . cut open the interiors of @xmath17 and each @xmath252 ( e.g. , figure [ f7 ] ) . and the clasps @xmath252 with a cut along @xmath257 . ] next join the edges accordingly ( e.g. , see figure [ f8 ] ) and cut along singular points to obtain an embedded oriented surface . . ] more precisely , two edges of a cut - open clasp are _ adjacent _ if , given any two neighborhoods @xmath258 and @xmath259 of these edges in the respective surface , @xmath260 is a nonempty segment along @xmath230 , properly containing the common boundary point of the two edges . near two adjacent edges , the surfaces @xmath235 and @xmath103 are also `` adjacent '' . we resolve the clasp arcs @xmath252 and the singular clasp @xmath17 by joining adjacent edges ( look at figures [ f7 ] and [ f8 ] ) . this produces an embedded oriented surface @xmath234 with @xmath229 . in the construction of lemma [ construction ] above , resolving all other possible intersections besides the clasps required only isotoping the interior of @xmath235 locally , so we may assume that such intersections do not occur . [ twistingnearbeta ] consider a standard framed neighborhood @xmath242 with the product framing on @xmath230 induced by @xmath243 as in the construction of lemma [ construction ] prior to resolving the clasp intersections between @xmath235 and @xmath103 . denote by @xmath261 the product framing induced on the core @xmath230 and its parallel translation @xmath231 . then @xmath262 consider a push - off @xmath263 of @xmath264 into @xmath235 giving the seifert framing @xmath265 along @xmath230 so that @xmath266 . the intersection @xmath267 consists of a point on each clasp arc @xmath268 and a point on the singular clasp arc @xmath269 . now , isotop @xmath263 in @xmath235 through each arc @xmath252 across @xmath231 such that the intersection points along @xmath252 are eliminated and @xmath263 intersects @xmath103 once along @xmath17 . this implies that away from a neighborhood of @xmath17 , @xmath263 links with @xmath230 relative to the product framing @xmath261 in the same way that it links with @xmath231 relative to the product framing @xmath261 because @xmath263 is disjoint from the annulus @xmath103 away from @xmath17 . resolving the @xmath252 modifies the interior of @xmath235 away from @xmath263 . resolving @xmath17 modifies @xmath263 into @xmath270 , which coincides with @xmath263 away from a neighborhood of @xmath17 and does not link with @xmath231 rel product framing , we obtain @xmath270 from @xmath263 by removing a small segment and adding another one , each isotopic to the other rel boundary in the complement of @xmath231 . observe that since @xmath234 is constructed from @xmath271 by resolving their intersections , @xmath270 is a push - off of @xmath231 into @xmath234 . relative to the product framing , it links with @xmath231 in the same way as @xmath263 linked with @xmath230 except for the @xmath272 which is an additional twist and contributes @xmath120 . so with respect to the product framing , @xmath273 . [ twistingnearalpha ] in the construction of lemma [ construction ] , @xmath274 resolving the clasps @xmath252 changes the interior of @xmath235 away from @xmath6 so it does not affect the framing at @xmath6 . we need to see what changes occur to the framing @xmath265 along @xmath6 when we resolve the singular clasp @xmath17 . prior to resolving it , take ( the closure of ) a standard framed tubular neighborhood @xmath275 with framing @xmath265 via a push - off @xmath276 into @xmath235 . consider an arc @xmath277 that ends at @xmath241 such that @xmath278 $ ] . after @xmath17 is resolved , @xmath279 , take a push - off @xmath60 of @xmath6 into @xmath234 . since @xmath235 and @xmath234 are isotopic away from a neighborhood of the resolved arc @xmath17 , we can choose that @xmath60 and @xmath280 coincide away from @xmath241 . observe that @xmath60 intersects @xmath281 , moreover , @xmath282 by the construction of @xmath234 . the winding number of the normal to @xmath6 defined by @xmath280 under the diffeomorphism to @xmath283 equals 0 , and with the winding number of the normal to @xmath6 defined by @xmath60 , we get @xmath282 . this gives @xmath284 . [ alphatwist ] let @xmath234 be the surface constructed in lemma [ construction ] . then @xmath234 is smoothly isotopic to a surface obtained from @xmath235 by the following local operation in a neighborhood @xmath285 of the singular clasp @xmath17 ( figure [ f9 ] ) . cut @xmath285 along its boundary arcs that run across @xmath230 to @xmath6 in the interior of @xmath235 . then create an oriented twist in @xmath285 depending on the sign of @xmath286 , as shown figure [ f9 ] , and re - glue back by the identity diffeomorphism . this gives the diffeomorphism type of @xmath234 as constructed in lemma [ construction ] . by a local twist in @xmath235 around the singular clasp @xmath17 . since only @xmath287 is fixed , the diffeomorphism type of @xmath234 depends on the arc @xmath17 . the twist is determined by the sign of the intersection of @xmath6 and @xmath103 at @xmath241 . ] to see this , note that resolving the clasp intersections @xmath252 creates a surface which is smoothly isotopic to @xmath288 since away from the singular clasp @xmath17 , @xmath288 deformation retracts to @xmath235 . so the diffeomorphism type of @xmath234 is determined only by the resolution of @xmath17 . now , since @xmath230 and @xmath231 are oriented as boundary components of the oriented annulus @xmath103 , make sure that the orientations on @xmath230 as a boundary component of @xmath235 and a boundary component of @xmath103 coincide , consider @xmath289 . if @xmath290 , then in the construction of lemma [ construction ] , resolving @xmath17 produces an oriented surface which locally is obtained from @xmath235 by a left twist in the disc @xmath285 . similarly , if @xmath291 , the twist on @xmath285 is right . thus the diffeomorphism type of @xmath234 depends on the arc @xmath17 and the sign of @xmath272 . in particular , if @xmath17 and @xmath254 are two isotopic arcs ( with common boundary point @xmath221 ) in @xmath235 running from @xmath3 to @xmath221 , then the resulting surfaces @xmath234 and @xmath292 obtained by resolving @xmath17 and @xmath254 are isotopic . consider homologous legendrian ( resp . transverse ) knots @xmath3 and @xmath6 oriented as @xmath7 for an oriented embedded surface @xmath8 in a contact 3-manifold @xmath1 so that @xmath9=[k]-[j]$ ] and a legendrian ( resp . transverse ) isotopy @xmath293\rightarrow(m,\xi)$ ] of @xmath3 in the complement of @xmath6 , which fixes @xmath6 . we can extend it to all of @xmath4 by the isotopy extension theorem ( in the legendrian or transverse category , resp . , see @xcite ) with @xmath294 on a neighborhood of @xmath6 . because @xmath220 extends to an isotopy of the seifert surface @xmath8 , the relative invariants are preserved . that is , for a legendrian isotopy @xmath220 , @xmath295 and @xmath296 , and for a transverse isotopy @xmath220 , @xmath297 for all @xmath219 . now consider a legendrian ( resp . transverse ) isotopy which , as in lemma [ construction ] , isotops @xmath3 across @xmath6 , that is , @xmath298 crosses @xmath6 for some @xmath299 . then @xmath220 no longer extends to an isotopy of @xmath8 . for @xmath225 , let @xmath300 and @xmath235 be as in lemma [ construction ] and assume the trace of @xmath220 and transverse to @xmath6 at @xmath301 , we will say that @xmath220 is _ locally transverse _ to @xmath6 ( see @xcite for a discussion of the case when @xmath220 is not locally transverse to @xmath6 ) . we claim that in this case the relative invariants are well - defined for all @xmath224 and are indeed invariant under @xmath220 . [ legendriantransverseinvariance ] consider homologous legendrian ( resp . transverse ) knots @xmath3 and @xmath6 oriented as boundary components of an oriented embedded surface @xmath8 in a contact 3-manifold @xmath1 so that @xmath9=[k]-[j]$ ] . consider a legendrian ( resp . transverse ) isotopy @xmath293\longrightarrow(m,\xi)$ ] of @xmath3 , and suppose that @xmath302 and @xmath223 for @xmath224 . a. if @xmath220 is legendrian , then @xmath303 for all @xmath224 , where @xmath33\in h_2(m,\varphi_t(k)\cup j;\z)$ ] . b. if @xmath220 is legendrian , then @xmath304 for all @xmath224 , where @xmath33\in h_2(m,\varphi_t(k)\cup j;\z)$ ] . c. if @xmath220 is transverse , then @xmath305 for all @xmath224 , where @xmath33\in h_2(m,\varphi_t(k)\cup j;\z)$ ] . we will use the construction of @xmath234 in the case of smooth isotopies ( lemma [ construction ] ) . resolving some intersections of @xmath288 involved isotoping the interior of @xmath235 locally , which does not affect the relative invariants . resolving the clasp intersections , however , may . we assume that we have resolved all intersections besides the ( possibly multiple ) clasps @xmath252 and the singular clasp @xmath17 . assume first that @xmath220 is legendrian and let @xmath234 be constructed as in the proof of lemma [ construction ] . first , we prove the invariance of the relative thurston - bennequin number . prior to resolving any of the clasp intersections between @xmath235 and @xmath103 , take a framed legendrian neighborhood @xmath256 of @xmath230 such that the product framing @xmath261 is given by @xmath243 as in lemma [ twistingnearbeta ] , and so @xmath231 is a parallel copy of the legendrian core @xmath230 . since @xmath230 and @xmath231 are legendrian isotopic , @xmath306 , equivalently , @xmath307 . the reason for this is the fact that there exists an immersed annulus @xmath103 ( the trace of the isotopy ) that satisfies this , so by lemma [ independent ] , the embedded surface @xmath103 satisfies this condition as well . equivalently , since we are in a framed legendrian neighborhood @xmath256 , @xmath308 or @xmath309 . by lemma [ independent ] , we could use @xmath234 to prove the invariance of the relative thurston - bennequin number . by the discussion at the beginning of this section , it is sufficient to prove that @xmath310 which , by definition , is equivalent to @xmath311 or @xmath312 since @xmath313 , the left - hand side of equation ( [ eq1 ] ) equals @xmath314 since @xmath231 is legendrian isotopic to @xmath230 , @xmath308 , so the above expression simplifies to @xmath315 since @xmath316 by lemma [ twistingnearbeta ] , the expression above ( and the left - hand side of equation ( [ eq1 ] ) ) equals @xmath317 . now , the right - hand side of equation ( [ eq1 ] ) equals @xmath318 which simplifies to @xmath319 . by lemma [ twistingnearalpha ] , @xmath320 , which implies that @xmath321 , and equation ( [ eq1 ] ) holds . we set some general notation . for a contact 3-manifold an a legendrian knot bounding an oriented surface and oriented as the boundary of the surface in the 3-manifold , let @xmath322 denote the winding number of the positive tangent vector field to the knot measured as a winding number of a loop of vectors under the identification with @xmath46 at each point of the knot given by the trivialization @xmath323 of the contact structure over the surface restricted to the boundary . in the standard sense , @xmath322 is the rotation number of the legendrian knot . we start with the following simple observation . [ rotnumber ] consider a legendrian unknot @xmath324 in standard contact 3-ball around the origin in @xmath325 . the rotation number of @xmath141 ( with respect to the global trivialization on @xmath326 given by the non - zero section @xmath327 of @xmath328 ) is given by the degree of the tangent vector to the lagrangian projection @xmath329 , where @xmath330 ( see @xcite ) . note that since @xmath141 bounds a 2-disc in @xmath331 , we can consider its oriented projection under @xmath332 and compute the winding number of its oriented boundary @xmath333 , in the lagrangian projection , the positively oriented parts of the 2-disc point in the positive @xmath334-direction of the @xmath335-plane , so the rotation number computation amounts to counting the number of positively oriented ( upward ) circles in the projection of the 2-disc . now take another oriented legendrian knot @xmath336 which co - bounds an annulus @xmath337 with @xmath141 , so its lagrangian projection links with @xmath141 ( positively if @xmath338 and negatively if @xmath339 ) . orient @xmath337 so that the orientation that @xmath141 inherits as its boundary is the opposite to the one it inherits as boundary of @xmath340 . then @xmath341 is trivial over @xmath337 , and @xmath342 . let @xmath343 denote the trivialization from @xmath327 on @xmath341 restricted to @xmath141 , then @xmath344 this explicitly tells us how the rotation number of @xmath141 is computed for another trivialization coming from another seifert surface , purely in terms of a linking number with a knot which defines the new seifert framing for @xmath141 . this observation is easy to see by working with legendrian knots with corners " in @xmath345 , note that it holds in this case as well . another approach is to use the characteristic foliations on the 2-dsic @xmath340 and the annulus @xmath337 . in our situation , we have the trivializations @xmath346 induced by each seifert surface on @xmath347 and their restrictions to the knots @xmath300 and @xmath6 . we need to show @xmath348 which , by definition , is equal to @xmath349 or , rearranging terms , @xmath350 we will use a local model near the intersections @xmath17 and @xmath252 . this is justified by the fact that away from intersection points along @xmath230 , @xmath231 , and @xmath6 , we can assume that the trivializations @xmath351 and @xmath352 coincide along @xmath353 , with the special point that along @xmath230 , we would need to change the sign since @xmath230 acquires opposite orientation from the surface @xmath235 and the surface @xmath103 . therefore , we only need to see what the local contributions to the winding numbers are near the arcs of intersection . first we deal with the arcs @xmath252 , refer to figures [ f7 ] and [ f_beta ] . take an arc @xmath252 and consider a disc with corners @xmath354 around @xmath252 and boundary components @xmath355 with @xmath356 . isotop the @xmath357 to make them legendrian and chose the corners of @xmath358 so that the winding numbers @xmath359 and @xmath360 are well - defined and @xmath361 and @xmath362 . now take a small 2-disc with corners @xmath363 around @xmath252 with legendrian boundary @xmath364 , such that @xmath365 and @xmath366 . note that resolving the arc @xmath252 in constructing the surface @xmath234 is a local operation and we can arrange the two discs @xmath358 and @xmath367 to be contained in an arbitrarily small 3-ball neighborhood of the arc @xmath252 . let @xmath368 with @xmath369 be the new surface in our 3-ball neighborhood after the arc @xmath252 has been resolved . note that @xmath368 coincides with @xmath370 in a neighborhood of @xmath371 . take a contactomorphism @xmath87 between the 3-ball neighborhood containing @xmath368 and @xmath326 . now , take 2-disc strips to complete @xmath358 to an annulus @xmath372 with legendrian @xmath373 and complete @xmath367 to an annulus @xmath374 with legendrian @xmath375 . arrange that @xmath376 and @xmath377 . note also that @xmath378 bounds a 2-disc @xmath379 and that @xmath380 is an annulus which coincides with @xmath381 near @xmath371 . now take an oriented push - off @xmath382 of @xmath383 into @xmath374 and note that @xmath384 , because @xmath382 is isotopic to @xmath385 . also take an oriented push - off @xmath386 of @xmath378 into @xmath337 and note that @xmath387 . . ] by remark [ rotnumber ] , @xmath388 and since @xmath389 , @xmath390 subtracting these and noting that @xmath391 , we have @xmath392 and subtracting the equal contributions of rotations along @xmath393 and @xmath394 , we have @xmath395 which implies that as a result of the local resolution of the arc @xmath252 , there is no contribution to the winding number of @xmath231 as compared to the winding number of @xmath230 with respect to @xmath235 . thus the winding numbers of @xmath396 and @xmath397 , are fixed , regardless of which trivialization ( @xmath351 or @xmath352 ) is used . since resolving the @xmath252 arcs is done in the complement of @xmath6 , we have no contribution to @xmath398 and @xmath399 from this , so there is no contribution to the value of the relative rotation number of @xmath231 relative to @xmath6 as a result of resolving the @xmath252 . now we resolve the arc @xmath17 , we will clear the notation we used in resolving @xmath252 , but stick to the notation rules we have set . we expect that it would change the individual rotation numbers of @xmath231 and @xmath6 , but these changes would subtract out in the value of the relative rotation number . for the following , refer to figure [ f7 ] and figure [ f_alpha1 ] to understand the local picture near the arc @xmath17 . we know that @xmath287 lies on @xmath6 , let the other point in @xmath400 be @xmath401 . take a half - disc @xmath354 with legendrian boundary around @xmath17 with corners at points @xmath402 and label @xmath403 , where @xmath404 contains the point @xmath405 with @xmath406 and @xmath407 is contained in the interior of @xmath358 . on @xmath235 , choose an arc @xmath408 with @xmath409 , where @xmath410 and let the arc in @xmath6 bounded by @xmath241 and @xmath411 be @xmath412 so that @xmath413 bound a 2-disc @xmath414 . note that @xmath379 has four corners @xmath415 , @xmath411 , @xmath241 , and @xmath405 . similarly , choose an arc @xmath416 with @xmath417 , where @xmath418 and let the arc in @xmath6 bounded by @xmath241 and @xmath419 be @xmath236 so that @xmath420 bound a 2-disc @xmath421 with four corners @xmath422 , @xmath419 , @xmath241 , and @xmath405 , such that @xmath423 . let @xmath424 and @xmath425 . . ] in addition , we choose a standard 3-ball neighborhood around @xmath358 , @xmath379 , and @xmath426 and a contactomorphism @xmath87 to @xmath326 that will be useful later . since @xmath17 is the only intersection between @xmath235 and @xmath234 , assume that @xmath351 and @xmath352 coincide along @xmath230 and @xmath6 away from @xmath17 , in particular , assume that the two trivializations coincide along @xmath427 and @xmath428 . by construction , @xmath429 now let @xmath430 , let @xmath431 , where @xmath337 be an annulus with legendrian boundary @xmath432 . then @xmath433 and @xmath434 , so remark [ rotnumber ] implies @xmath435 where @xmath436 . since @xmath437 we obtain @xmath438 observe that we can assume that the trivializations @xmath439 and @xmath440 coincide along @xmath441 ( with opposite sign ) , and that we can construct @xmath337 so that @xmath441 and @xmath442 have equal ( and opposite ) contribution to the winding number of their respective boundary components . therefore , @xmath443 by construction , @xmath444 and ( with appropriate orientations ) @xmath445 thus , @xmath446 so in @xmath447 , this gives @xmath448 @xmath449 now to compute the term @xmath450 , we invert " the picture in figure [ f_alpha1 ] , see figure [ f_alpha2 ] . consider the legendrian knots @xmath451 and @xmath452 , then @xmath453 and let @xmath454 for a 2-disc @xmath455 . observe that @xmath456 can be isotoped so that they coincide along @xmath457 and so that @xmath367 and @xmath358 coincide along @xmath441 . also note that @xmath458 . take a 2-disc @xmath459 with corners in the complement of @xmath455 so that @xmath460 is legendrian and @xmath461 , so that it extends @xmath455 to an oriented embedded annulus @xmath462 with legendrian boundary components @xmath463 and @xmath464 . choose @xmath459 so that @xmath465 . consider points @xmath466 and @xmath467 and take two parallel legendrian copies @xmath468 and @xmath469 with @xmath470 and such that @xmath471 . now take a push - off @xmath472 of @xmath281 into @xmath367 and a push - off @xmath473 of @xmath281 into @xmath455 such that @xmath474 and @xmath475 . isotop @xmath476 rel boundary to make them legendrian and let @xmath477 and @xmath478 . moreover , choose @xmath476 so that @xmath479 . note that we have freedom to adjust the points @xmath480 and @xmath481 in order to ensure that the arcs satisfy these properties . so now we have two legendrian unknots @xmath141 and @xmath336 , and notice that @xmath141 bounds a 2-disc @xmath379 which can be isotoped in the interior to coincide with @xmath455 near @xmath281 and to coincide with @xmath459 near @xmath482 , while @xmath336 bounds a 2-disc @xmath426 which can be isotoped in the interior to coincide with @xmath459 near @xmath483 and to coincide with @xmath367 near @xmath484 . observe that we can isotop the interior of @xmath379 so that it intersects with @xmath367 precisely along the arc @xmath17 since @xmath336 necessarily intersects @xmath379 along the segment @xmath484 . . ] in this setup , remark [ rotnumber ] applies to @xmath141 and @xmath336 to give @xmath485 equivalently , @xmath486 which , after subtracting the equal contributions along the @xmath487 , equals @xmath488 equivalently , @xmath489 note that @xmath490 so @xmath491 by construction , @xmath492 so we have that @xmath493 now consider the term @xmath494 . look back at the 2-disc @xmath358 with @xmath403 . since the 2-disc @xmath379 coincides with @xmath455 near @xmath281 , and @xmath495 , we can isotop the interior of @xmath455 away from @xmath17 and @xmath281 so that @xmath496 , a clasp arc such that @xmath497 . in order to keep consistent orientations , we will need to use the discs @xmath358 , @xmath367 , @xmath455 , and @xmath379 , in addition to the annulus @xmath337 that @xmath141 and @xmath336 co - bound , note that @xmath498 . first , @xmath358 is given an orientation coming from @xmath103 , and @xmath455 is given the same orientation , since it coincides with @xmath358 near @xmath499 . on the other hand , by construction , we orient @xmath367 so that its common boundary segment @xmath500 with @xmath358 receives opposite orientations as a boundary to each disc . also , @xmath379 is oriented consistently with @xmath455 so that the strip between @xmath501 and @xmath281 along which the two discs coincide is oriented the same way from each 2-disc . then the annulus @xmath337 is oriented so that @xmath141 receives opposite orientations as an element of @xmath502 and @xmath503 . this gives an orientation on @xmath336 and a sign to @xmath504 . reversing the orientation on @xmath358 and carrying out this check reveals that the sign @xmath505 does not change and the sign of @xmath494 also does not change . moreover , for each choice of orientations , an easy check reveals that @xmath506 , so these two numbers are equal . therefore , and we obtain @xmath507 then substituting in @xmath447 and @xmath508 , we have @xmath509 @xmath510 @xmath511 @xmath512 therefore , the construction of the new surface after a legendrian isotopy intersects the reference knot " dies not change the relative rotation numbers . we start with the equivalent of remark [ rotnumber ] in the transverse category . [ slnumber ] consider a transverse unknot @xmath324 in standard contact 3-ball around the origin in @xmath325 . we can compute the self - linking number of @xmath141 with respect to the global trivialization of @xmath328 or with respect to a trivialization of @xmath341 over a seifert surface ( in this case , a 2-disc ) for @xmath141 , in which case we take a non - zero section of the trivialization and measure the linking number with @xmath141 of a push - off of @xmath141 under this section . we set some notation , let @xmath513 denote the push - off of a non - zero section of @xmath514 , so , in particular , for @xmath515 , we have @xmath516 . now take another oriented transverse knot @xmath336 which co - bounds an annulus @xmath337 with @xmath141 . orient @xmath337 so that the orientation that @xmath141 inherits as its boundary is the opposite to the one it inherits as boundary of @xmath340 . then @xmath517 in the case of transverse knots and a transverse isotopy between @xmath230 and @xmath231 , with the setup in lemma [ construction ] , we need to prove @xmath518 , which by definition is equivalent to @xmath519 or , @xmath520 again we consider the cases when resolving the arcs @xmath252 and the arc @xmath17 separately . by remark [ slnumber ] above , the arguments follow exactly the argument in the proof of part ( b ) . we will use the figures and labels from part ( b ) . we consider a local picture near @xmath252 and reduce the discussion to computing the local contributions to the self - linking numbers . this is justified by working with the characteristic foliation and arranging that away from a local 2-dsic neighborhood of the intersection arcs on each surface , the contribution to the self - linking difference is zero . use the fact that self - linking is additive under relative connected sums ( see @xcite ) . we provide some more detail of the setup . again , consider the setup in figure [ f_beta ] , choose the non - zero section of @xmath521 and the arcs @xmath357 so that along @xmath357 , we have no contribution to the quantity @xmath522 . similarly , choose the endpoints of the arcs @xmath523 so that they have equal ( and opposite in sign ) contributions to @xmath522 . note that this is possible since we know that @xmath524 . do the same for the arcs @xmath112 and choose the arc @xmath525 so that @xmath526 . resolving @xmath252 is a local modification away from @xmath6 , so @xmath527 . we claim that @xmath528 is also equal to zero in an appropriate sense computed locally near @xmath252 . consider @xmath529 and recall that @xmath530 , so the above expression equals @xmath531 observe that @xmath532 , where @xmath533 is a push - off of @xmath231 into @xmath103 , and , similarly , @xmath534 , where @xmath263 is a push - off of @xmath230 into @xmath103 . therefore , @xmath447 becomes @xmath535 rearranging terms , we obtain @xmath536 by the local construction in part ( b ) for the case for @xmath252 , remark [ slnumber ] yields that @xmath537 and @xmath538 . so the quantity @xmath447 is equal to zero . now for resolving @xmath17 , we again look at the expression @xmath520 use the setup and terminology from figure [ f_alpha2 ] to conclude that @xmath539 . together with lemma [ construction ] , lemmas [ independent ] , [ independent2 ] , and [ independent3 ] imply that the relative thurston - bennequin number and relative rotation number are legendrian isotopy invariants , and the relative self - linking number is a transverse isotopy invariant , even when the isotopy of @xmath3 intersects the reference knot " @xmath6 . in the case when our contact manifold @xmath1 is tight , we can use convex surface theory and the characteristic foliation of the surfaces ( see @xcite ) to compute the relative invariants and apply them to relative connected sums and classifications ( see @xcite ) . in @xcite , chantraine looked at legendrian knots that are cobordant via a lagrangian cylinder in and conjectures an explicit formula which essentially gives the relative thurston - bennequin invariant in this construction . in @xcite , we study the more general setup of cobordant legendrian knots with no restrictions on the embedded surface in the cobordism , and prove a relative slice genus bound . in @xcite , chernov defines and proves well - definedness of relative self - linking numbers using homotopy methods and transverse homotopy instead of isotopy . this requires certain assumptions on the contact structure of the 3-manifold ( tightness , or , equivalently , coorientability ) . the relative thurston - bennequin invariant and the relative self - linking number we have defined here behave like the affine self - linking invariant in theorem 2.0.2 . parts 1 and 2a in @xcite .

the most intense monitoring observations yet made in the optical ( ) and near - infrared ( ) wave bands were carried out for several well - known nearby seyfert 1 galaxies of ngc 5548 , ngc 4051 , ngc 3227 , and ngc 7469 . over three years of observations , after the operation of the magnum telescope started in early 2001 , clear time - delayed response of the-band flux variations to the-band flux variations was detected for all of these galaxies . their color temperature was estimated to be k from the observed flux variation gradients . this estimated color , along with the smoothness of the flux variation in the near - infrared in contrast with rapid fluctuations in the optical , supports a view that the bulk of the flux should originate in the thermal radiation of hot dust that surrounds the central engine . cross - correlation analysis was performed in order to quantify the lag time corresponding to the light - travel distance of the hot dust region from the central engine . the measured lag time is days for ngc 5548 , days for ngc 4051 , about days for ngc 3227 , and days for ngc 7469 . we found that the lag time is tightly correlated with the optical luminosity as expected from dust reverberation ( ) , while only weakly with the central virial mass , which suggests that an inner radius of the dust torus around the active nucleus has a one - to - one correspondence to central luminosity . in the lag time versus central luminosity diagram , the-band lag times place an upper boundary on the similar lag times of broad - emission lines in the literature . this not only supports the unified scheme of agns , but also implies a physical transition from the blr out to the dust torus that encircles the blr . furthermore , our-band flux variations of ngc 5548 on timescales of up to 10 days are found to correlate with x - ray variations and delay behind them by one or two days , indicating the thermal reprocessing of x - ray emission by the central accretion flow .

astro-ph0511697

the unified scheme of active galactic nuclei ( agns ) and radio - quiet quasars explains that seyfert 1 nuclei with broad emission - line regions ( blr ) would be classified as seyfert 2 if high velocity clouds emitting broad emission lines were hidden by the dust tori that encircle such clouds ( antonucci 1993 ) . outside the dust torus there would be a narrow emission - line region ( nlr ) that extends to more than kilo - parsecs and is often observed as resolved images . however , the angular scale of the inner dust torus even in the nearest agn is too small to be spatially resolved with any current technology in optical and near - infrared imaging observations . it is generally accepted that the near - infrared emission from seyfert galaxies and quasars is thermally produced by hot dust heated by uv radiation from the central engine . barvainis ( 1987 , 1992 ) developed a dust - emission model that naturally explains the bump or excess in the near - infrared continuum of agns by thermal emission from grains at their sublimation temperature in a range of @xmath11 k. this model , which is supported by the fact that a variety of agns have blackbody components of nearly uniform temperature of @xmath12 k ( kobayashi et al . 1993 ) , necessarily claims a dust distribution with central hole . a typical uv luminosity of @xmath13 ergs s@xmath14 for seyfert galaxies therefore gives an inner radius of @xmath15 pc or @xmath16 light days at which hot dust sublimates . consequently , if we monitor their continuum fluxes in the uv / optical and near - infrared , a lag time would be expected that corresponds to the light - travel distance between the central engine and the hot dust region from which the bulk of near - infrared emission radiates ( e.g. , clavel , wamsteker & glass 1989 ; nelson 1996 ; glass 2004 ) . the most successful similar technique is the reverberation mapping of the blr carried out for many seyfert 1 galaxies and quasars during the last decade ( e.g. , peterson 1993 ; wandel , peterson & malkan 1999 ; kaspi et al . if we verify that the infrared lag is longer than the lag of broad emission lines of lower ionization energy , the inner radius of the dust torus should be larger than the outer extension of the blr , which provides strong support to the unified scheme of agns . however , there have only been a few cases where lag times for both blr and dust torus were measured for the same agn ( clavel , wamstekar & glass 1989 ; minezaki et al . 2004 ; suganuma et al . 2004 ) , mostly because long - term simultaneous monitoring observations in the uv / optical and near - infrared for dust reverberation have rarely been attempted . the magnum ( ulticolor ctive alactic clei onitoring ) project was proposed to carry out such monitoring observations for many agns ( kobayashi et al . 1998a ; yoshii 2002 ) . this project built a new 2-m optical - infrared telescope at the university of hawaii s haleakala observatory on the hawaiian island of maui , and started preliminary observations in early 2001 . light curve data of high photometric accuracy and frequent sampling have been accumulating since then . this paper reports early results for four nearby seyfert 1 galaxies included in the magnum program , including ngc 5548 , ngc 4051 , ngc 3227 , and ngc 7469 for which the extent of the blr and the central virial mass were previously derived from reverberation mapping of the blr . the intensive monitoring observations using the magnum telescope , with an improved technique for calculation of cross - correlation functions , enable the most accurate measurement of lag times between the @xmath3 and @xmath2 flux variations for the above four galaxies , and comparison with lag times for the broad emission lines . in [ obs ] we describe the target agns , the observational conditions , the image reduction , and photometry procedures . in [ var ] we discuss the characteristics of multicolor light curves of the target agns and their optical to near - infrared colors . we interpret the clear lag between optical and near - infrared variations in terms of dust reverberation . in [ ccfana ] we perform a cross - correlation analysis to quantify the lag time between the @xmath3 and @xmath2 flux variations . here , we describe how to calculate the cross - correlation functions based on two methods optimized for our time series data . in [ discus ] we discuss how well the measured lag times for the target agns relate to their optical nuclear luminosities and central masses , in combination with our earlier result of ngc 4151 ( minezaki et al . 2004 ) and previous lag time measurements for other agns in the literature . we compare these lag times with similar lag - time measurements for the broad emission lines , in order to see how the inner edge of the dust torus relates to the blr . the characteristics of the target agns ( ngc 5548 , ngc 4051 , ngc 3227 , and ngc 7469 ) are listed in table [ objlst ] . all of them are classified as seyfert 1 galaxies . the right ascension and declination for an epoch of 2000 are given in columns 2 and 3 , respectively . the apparent magnitude of average nuclear flux is given in column 4 . the recession velocity and the galactic @xmath3-band extinction are given in columns 5 and 6 , respectively , from the ned database . assuming a hubble constant of @xmath17 km s@xmath14mpc@xmath14 , the absolute @xmath3 magnitude estimated from column 4 is given in column 7 , after being corrected for galactic extinction in column 6 , based on the average fluxes given in the next section ( table [ varstat ] ) . the light curve data have been accumulated over the last several years , since the magnum observatory started its operation in early 2001 . a part of the early result has already been published as letters ( ngc 4151 , minezaki et al . 2004 ; ngc 5548 , suganuma et al . 2004 ) . monitoring observations were made using the multi - color imaging photometer ( mip ) mounted on the magnum telescope ( kobayashi et al . 1998a , b ) . the field of view of the mip is 1.5 @xmath18 1.5 arcmin@xmath19 , and it enables us to obtain the images simultaneously in the optical and the near - infrared bands by splitting the incident beam into two different detectors such as an site ccd ( 1024 @xmath18 1024 pixels , 0.277 arcsec / pixel ) and an sbrc insb array ( 256 @xmath18 256 pixels , 0.346 arcsec / pixel ) . monitoring observations in the @xmath3 and @xmath2 bands started in 2001 , ngc 4051 in january , ngc 5548 in march , ngc 7469 in june , and ngc 3227 in november , while the @xmath20 observations started later . the nuclear fluxes with the @xmath3 and @xmath2 filters were observed as the first priority , and the sampling interval for each target agn was set shorter than one tenth of the expected lag time between the @xmath3 and @xmath2 flux variations . the @xmath21 and @xmath22 filters were used as the second priority , and the @xmath23 and @xmath24 filters as the third priority . the telescope was pointed , typically six times , at each of the target agn and two reference stars a and b , alternately . the pointings were repeated first in order of a @xmath25 agn @xmath25 b , then in the reverse order with small shift of several arcsecs , and so on , which gave us dithered field images of the agn and the stars . in each pointing , the ccd detector was exposed at once during which the insb array was read out several times or more to avoid saturation . high signal - to - noise ratios were achieved with total on - source integration time of several minutes for each object . digitized sky survey ( dss ) images for four target fields are shown in figures [ field_n5548 ] to [ field_n7469 ] . the fwhms of the psf during the observations were typically @xmath26 in the optical and @xmath27 in the near - infrared . one standard star was observed each night to calibrate the flux of the reference stars . twilight flat was obtained each evening for the @xmath23-band images , and dome flat for other bands was obtained at dawn after the observation . our monitoring observations were performed with the highly automated observing system we developed ( kobayashi et al . 2003 ) . the images were reduced using iraf . we followed the standard procedures for image reduction , with small corrections for the non - linear response of the detectors . for the infrared image reduction , a sky image assembled from dithered images pointed at the reference stars was subtracted prior to flat - fielding . then , flat - field corrections were applied using dome flats for the optical images and illumination - corrected dome flats for the near - infrared images . twilight flats were used for the @xmath23-band images . the nuclear fluxes of the galaxies were measured relative to the reference stars , and then the fluxes of these stars were calibrated . aperture photometry within @xmath28 with the sky reference of @xmath29 ring was applied to all images , then the fluxes of the nuclei and the reference stars were compared . we adopted a large aperture rather than that generally used for a psf - like object because the variation of the psf during the observation increases the scatter of the flux , which often dominates the uncertainties in the aperture photometry applied for the nucleus on the extended source of the host galaxy . for each object and band , we used two comparison methods of relative photometry considering their conditions . in one method , the average fluxes of the nuclei and the reference stars over all dithering sets were calculated before they were compared . in the other method , the fluxes of the nuclei and the reference stars were compared for each dithering set before the relative fluxes were assembled . the latter method can cancel relatively slow changes in transparency of the atmosphere during the observation , but has a statistical disadvantage because the number of the position is only six or less . we chose it mainly for the optical bands and chose the former mainly for the near - infrared bands . rapid fluctuation of the flux between the dithering positions in the same night was attributed to thin cloud , and such observations were omitted from the light curve data . the uncertainty in the measurements is determined as the rms variations , divided by square root of the number of frames in the first method or the number of positions in the second method , about these averages in the relative photometry . the errors for each night were usually better than @xmath30 0.01 mag . the flux ratios of two reference stars @xmath31 in various bands were monitored to confirm that they were not variable . the fluxes of the reference stars were calibrated with respect to photometric standard stars taken from landolt ( 1992 ) and hunt et al . the magnitudes of the reference stars were determined using fixed airmass coefficients for many data sets for which the nucleus and the standard stars were observed in the same night . the airmass coefficients were stable on clear nights , and we adopted pre - measured values of 0.212 ( @xmath21 ) , 0.116 ( @xmath3 ) , 0.016 ( @xmath24 ) , and 0.011 ( @xmath22 ) for unit airmass . we adopted the fitted value for each calibration , ranging from 0.04 to 0.08 for the @xmath2 band . the calibration errors were less than 0.01 mag in most bands . the coordinates and magnitudes of the reference stars in the optical and near - infrared bands are shown in tables [ refmagopt ] and [ refmagir ] , respectively . the aperture magnitude of the nucleus was determined each night , relative to those of the reference stars obtained each night . small corrections for the color term in the optical filters were also applied . the fluxes in the aperture are significantly contaminated by starlight of the host galaxy . the offset galaxy fluxes for the target agns were estimated as follows : the surface profile of the galaxy was fitted using galfit ( peng et al . 2002 ) where the images of the reference stars in the same night were used as the psf . this profile was subtracted from the observed image , and the flux of the residual psf - like image was measured . then , the offset galaxy flux in the aperture was determined as the difference between the total flux in the aperture and the flux of the residual psf - like image . the offset galaxy flux estimated from the data of best psf was subtracted from the observed light curve . in the case of ngc 7469 , which has a ring - shaped star - forming region with a radius of about @xmath32 around the nucleus , the flux of the psf - like image was estimated without obtaining the galaxy profile with galfit . instead , we estimated the nuclear flux by substracting the central psf image from the observed image and regarding the residual at the center as the galaxy offset . a lower nuclear flux was then obtained from a larger offset by equating such residual with the brightness at the radius of the star - forming ring , while a larger nuclear flux from a lower offset by equating such residual with the brightness at the radius outside which the galaxy profile dominates over the profile of star - forming ring . the offset galaxy fluxes thus estimated in the @xmath28 are given in table [ galfx ] . their uncertainty estimated from the scatter over different nights or the residual flux at the origin of the flux to flux diagram is @xmath33 % , sometimes giving a significant error in the nuclear flux which may result in false variations of the nuclear color . note that the galaxy subtraction provides an offset flux only and does not affect the cross - correlation analysis in [ ccfana ] . the host galaxy flux in the aperture depends on the seeing fwhm . this effect makes a systematic change in the measured nuclear flux up to 0.03 mag depending on the central profile of the host galaxy , and causes false fluctuations in the light curve . in order to estimate a relation between the seeing fwhm and the corresponding nuclear flux , the images in the best seeing condition were convolved with several fwhm values , as shown in figure [ seef_gal ] . thus , the nuclear flux measured each night was corrected using the fitting formulae : @xmath34 , \label{seecor_eq}\ ] ] where @xmath35 and @xmath36 are the constants and their values for each nucleus and band are given in table [ seecor ] . however , such correction was not attempted in the cases of @xmath37 for ngc 4051 and @xmath38 for ngc 3227 , because the photometric errors of each night were found to dominate over the seeing effect . figure [ sed ] shows the spectral energy distribution ( sed ) observed in the optical ( @xmath0 ) and near - infrared ( @xmath1 ) for each nucleus . the fluxes are those averaged over the nights when the seds were observed . the error bars include their rms variations and the uncertainties in the offset subtraction . we see that the sed is steep in the near - infrared ( @xmath1 ) while it is flat in the optical ( @xmath0 ) except for ngc 3227 . this is consistent with a view that the hot dust emission dominates towards longer wavelengths in the near - infrared , while the flat power - law spectra from nuclei of seyfert galaxies dominates in the uv and optical . however , ngc 3227 shows a steep sed in the optical which is different from the other nuclei . this may originate from offset errors in the @xmath3 band ( see [ var_opt ] ) , rather than originating from the central engine . the light curves based on aperture fluxes of four target agns are shown in figures [ lc_ngc5548 ] to [ lc_ngc7469 ] . the correction for the seeing effect has been applied and the offset flux of the host galaxy has been subtracted here . the sampling characteristics of the @xmath3 and @xmath2 light - curve data are given in table [ smpstat ] . the whole monitoring period for each galaxy and the maximum intervals between adjacent sampling epochs are given in columns 2 and 3 , respectively . the number of data points , the average and median intervals for the @xmath3 band are given in columns 4 , 5 and 6 , respectively , and those for the @xmath2 band are in columns 7 , 8 and 9 . the maximum interval corresponds to the sampling gap due to solar conjunction . the average interval is longer than the median , because of sampling gaps of three months due to solar conjunction and occasional blanks of several weeks caused by bad weather or facility maintenance . the median interval of @xmath39 days over @xmath40 years is much shorter than any other monitoring observations in the near - infrared . because of changing weather conditions and/or sudden problems with instruments , there are cases where the observations in the @xmath3 band were successful while not in the @xmath2 band , and vice versa . therefore , the number of @xmath3-band data points does not necessarily coincide with that for the @xmath2 band . the statistics that characterize the observed variability in the @xmath3 and @xmath2 bands are given in table [ varstat ] . the values in this table were calculated after the offset flux of the host galaxy in the aperture was subtracted . for each target agn , the average flux @xmath41 , the maximum to minimum flux ratio @xmath42 , and the normalized variability amplitude @xmath43 are given . here , we define @xmath43 as the variance of the flux @xmath44 , corrected for the measurement uncertainty @xmath45 , divided by the average flux : @xmath46 where @xmath47 and @xmath48 it is evident from the values of @xmath43 in this table that the four agns underwent significant @xmath3 and @xmath2 variations during our monitoring observations . comparison of flux variations between the @xmath21 and @xmath3 bands for each agn indicates that their long timescale variations as well as small , short timescale fluctuations are almost synchronized ( figures [ lc_ngc5548 ] to [ lc_ngc7469 ] ) . this tendency is also true of the @xmath23-band variations for ngc 7469 . figure [ fvg_bv ] shows the @xmath3 flux plotted against the @xmath21 flux for each agn . each point represents a pair of nuclear fluxes in a certain night from which the flux of the host galaxy has been subtracted , and the error bars represent the photometric errors . there is a tight linear relation between the flux variations in the @xmath21 and @xmath3 bands . winkler et al . ( 1992 ) and winkler ( 1997 ) reported this linear relation between the optical flux variations for many seyfert galaxies in the southern hemisphere . they considered that the slope of the linear relation in the flux - to - flux diagram , called a flux variation gradient ( fvg ) , would represent the actual color of the variable component , which appears to be constant for individual agns . if this is the case , the offset of the linear regression line from the origin is due to non - variable components , such as errors in subtracting the galaxy flux , unresolved narrow - line fluxes , etc . a tight linear relation and no sign of significant non - linearity in figure [ fvg_bv ] support the above view , and the color of the central engine derived from the fvg , which is unaffected by non - variable components , would be more reliable than those inferred from the sed in figure [ sed ] . our estimate of nuclear @xmath49 color from the fvg , after being corrected for galactic extinction , gives @xmath50 ( ngc 5548 ) , @xmath51 ( ngc 4051 ) , @xmath52 ( ngc 3227 ) , and @xmath53 ( ngc 7469 ) . these values are similar to a typical value for seyfert 1 galaxies reported by winkler et al . ( 1992 ) , and correspond to the power - law nuclear flux @xmath54 with @xmath55 ( ngc 5548 ) , @xmath56 ( ngc 4051 ) , @xmath57 ( ngc 3227 ) , and @xmath58 ( ngc 7469 ) , provided the effects of emission - line fluxes such as h@xmath59 are not taken into account . in contrast with the optical light curves showing the short - term fluctuations , the near - infrared light curves in the @xmath1 bands are smooth ( figures [ lc_ngc5548 ] to [ lc_ngc7469 ] ) . this is consistent with a view that the bulk of near - infrared emission comes from the thermal reprocessing region whose size exceeds the distance over which light travels in the timescale of short - term fluctuations in the optical . figure [ fvg_hk ] shows the @xmath2 flux plotted against the @xmath22 flux for each agn . there is a linear relation between the flux variations in the @xmath2 and @xmath22 bands . our estimate of nuclear @xmath4 color from the fvg gives @xmath60 ( ngc 5548 ) , @xmath61 ( ngc 4051 ) , @xmath62 ( ngc 3227 ) , and @xmath63 ( ngc 7469 ) , which is consistent with the fvg colors for dozens of seyfert galaxies measured by glass ( 2004 ) . these colors correspond to the black - body temperature of @xmath5 k which agrees with the sublimation temperature of graphite grains , and support the thermal origin of near - infrared emission , especially in the @xmath2 band . there is a scatter in the fvg in figure [ fvg_hk ] , and the relation between the @xmath22 and @xmath2 flux variations is not as tight as that for the @xmath21 and @xmath3 fluxes . this implies that the thermal reprocessing region that emits the @xmath22 flux may not completely coincide with that for the @xmath2 flux , and/or the central source could contaminate differently in the @xmath22 and @xmath2 fluxes . the most important feature of the near - infrared variation , besides its smoothness , is its lag time behind the optical variation . figures [ lc_ngc5548 ] to [ lc_ngc7469 ] show that the near - infrared light curve resembles the optical light curve , if shifted backwards by about ten days ( ngc 4051 and ngc 3227 ) or several tens of days ( ngc 5548 and ngc 7469 ) . figure [ fvg_vk ] shows the @xmath2 flux plotted against the @xmath3 flux for each agn . because of the short - term fluctuations present only in the @xmath3 band , and because of the delay between the @xmath3 and @xmath2 variations , there is a significant scatter in figure [ fvg_vk ] . the lag time is approximately estimated by shifting either one of the @xmath3 or @xmath2 light curves horizontally . figures [ lc_ngc5548_vk ] to [ lc_ngc7469_vk ] show the observed @xmath3 and @xmath2 light curves for each agn . the dotted curve in each figure is the @xmath2 light curve shifted backwards by the lag time determined approximately by sight . a more reliable estimate of lag time will be given in the next section . quantitative estimates of lag time between the optical and near - infrared light curves are possible based on the cross - correlation analysis commonly used in studies of multiple time series . for example , the application of this type of analysis to spectroscopic monitoring observations of broad emission lines dates back more than a decade ( for a review , see peterson 2001 ) . let @xmath64 and @xmath65 be the irradiating ( @xmath3 band ) and responding ( @xmath2 band ) light curves at time @xmath66 , respectively . then they are related to each other via @xmath67 where @xmath68 is the transfer function with respect to the lag time @xmath69 . this equation indicates that the transfer function represents the time - smeared response to a @xmath45-function outburst of incident radiation . as a function of lag time @xmath69 , the cross - correlation function ( ccf ) and the auto - correlation function ( acf ) are defined , respectively , as @xmath70 and @xmath71 the acf is therefore the cross - correlation of the incident light curve with itself . use of equation [ tf ] to replace @xmath65 in equation [ ccf ] gives @xmath72 the ccf is therefore the convolution of the transfer function with the acf . the cross - correlation does not necessarily give an unambiguous measure of @xmath68 , but simply quantifies the time delay @xmath69 between the two light curves , which approximately represents the location of the peak or the centroid of @xmath68 . for intuitive interpretation of the ccf , we rewrite equation [ ccf ] as follows : @xmath73 ^ 2 dt\;\;\;. \label{ccf4}\end{aligned}\ ] ] the first and second terms are nearly constant , if the interval of time integration is much larger than @xmath69 . therefore , the third term , which is an integration of square difference between @xmath65 and @xmath74 , determines the shape of ccf(@xmath69 ) . we see that the ccf becomes maximum when the third term becomes minimum . equation [ ccf4 ] supports our sight estimation of lag time by shifting either one of the irradiating and responding light curves with respect to time ( figures [ lc_ngc5548_vk ] to [ lc_ngc7469_vk ] ) , if two light curves are linearly related to each other . the ccf is equivalent to the correlation coefficient for the pairs of time series @xmath75 . thereby , practical calculation is given by @xmath76[\,i(t_i)-\left < i\right>\ , ] } { \sigma_c\,\sigma_i } \;\ ; , \label{ccf5}\ ] ] where @xmath77 is the number of pairs , and @xmath78 and @xmath79 are the averages of @xmath74 and @xmath65 , respectively , and @xmath80 and @xmath81 are their standard deviations . note that this equation is valid only for the period during which @xmath74 and @xmath65 overlap in time . the data points in equation [ ccf5 ] should be regularly spaced in time , because each point in one light curve must be paired with a corresponding point in the other . for generating the pairs @xmath75 , or equivalently @xmath82 in our case , linear interpolation of either one of the @xmath3 or @xmath2 light curves has generally been used with small modifications if necessary ( e.g. , gaskell & peterson 1987 ; white & peterson 1994 ) . this interpolation method , however , sometimes does not provide a sufficient number of pairs for calculation , so that the shape of ccf is noisy and its peak difficult to detect . furthermore , this method does not guarantee a symmetrical acf with respect to a lag time of zero , when the sampling of the data has a highly unbalanced weight for time . here , we calculate ccfs based on different methods of interpolation , `` bidirectionally interpolated '' ( bi ) ccfs and `` equally sampled '' ( es ) ccfs . let @xmath83 and @xmath84 be the fluxes of two observed light curves irregularly sampled at times @xmath85 and @xmath86 , respectively . then , for the data pairs for which we calculate the standard linear correlation coefficient for lag time @xmath69 , bi - ccfs combine two types of pairs , ( @xmath87 ) and ( @xmath88 ) , where @xmath89 and @xmath90 are linearly interpolated fluxes from @xmath91 and @xmath92 , respectively . es - ccfs collect the pairs of ( @xmath93 ) , where @xmath66 is sampled by equal interval @xmath94 independent of @xmath95 and @xmath96 . these two methods provide symmetrical acfs with respect to a lag time of zero . the bi - ccfs put weight on the observed data , yet are affected by the unbalanced sampling of observations concurrently . the es - ccfs have contrary characteristics . the problem in the interpolation , which occurs inevitably in both bi and es methods , can be overcome by simulation of flux variations , as described below . figures [ ccf_ngc5548 ] to [ ccf_ngc7469 ] show the ccfs and acfs for four target agns calculated by the two methods of bi and es . table [ tp_vk ] shows the cross - correlation results for several subsets of the observed light curve data used in the calculation . the monitoring period for each subset , expressed in the modified julian date ( mjd ) , is given in column 3 . the ccf peak position @xmath97 and the ccf peak value @xmath98 for the bi method are in columns 5 and 6 , respectively , and those for the es method are in columns 7 and 8 . monte carlo simulations are reliable for quantitative estimates of lag time . maoz & netzer ( 1989 ) introduced the basic scheme for this type of simulation . a large number of artificial light curves are created for both irradiating and reprocessing sources , and their ccfs are calculated to determine the peak position @xmath99 for each pair of light curves . the empirical distribution of @xmath99 is called ` cross - correlation - peak - distribution ( ccpd ) ' . the uncertainty of lag time is determined from the ccpd . in this technique , it is important to simulate the light curve such that the causes of the uncertainty should be adequately reflected in the uncertainty of lag time . there are two major sources of uncertainty in lag time measurements . one is flux errors in individual data points , and the other is the sampling gaps that often significantly underestimate the real variations . in our photometric monitoring data , the former is not effective , because the photometric data have been obtained from homogeneous observation with very high s / n ratios , especially for nearby seyfert galaxies . on the other hand , the latter seriously affects the analysis , because there are inevitable gaps of three months due to solar conjunction and blanks of a few weeks due to bad weather conditions or facility maintenance . artificial light curves should be simulated in a less model - dependent way , based on the observed light curves , because we have a limited knowledge of the variability mechanism in the @xmath3 band and the transfer function from the @xmath3-band variations to the @xmath2-band variations . peterson et al . ( 1998 ) introduced a model - independent monte carlo method , called `` flux randomization / random subset selection ( fr / rss ) '' , where the simulated light curve is based on the observed one . the fr method modifies the observed fluxes in each simulation by random gaussian deviations based on the quoted error of each observed point , and accounts for the effects of flux - measurement uncertainties . the rss method randomly extracts the same number of data points from the observed light curve , allowing for multiple extraction , to construct an artificial light curve . this method is then able to estimate the uncertainty due to the different time sampling of the data points . however , multiple extraction results in the removal of the observed data points at certain times in each simulation , which underestimates the real variation of the light curve . here we introduce a new technique of realizing the light curves , which simulate the flux variations between the observed data points using stochastic processes characterized by the structure function ( sf ) from the observed light curves . the sf is often used for the analysis of single time series ( simonetti et al . it measures the distribution of power over timescales for which the variations are correlated . a first - order structure function , for a series of flux measurements @xmath100 , @xmath101 is defined as @xmath102 ^ 2 \;\;,\ ] ] where the sum is over all pairs for which @xmath103 , and @xmath104 is the number of pairs . the sf of most agns is given by @xmath105 which increases up to some characteristic lag time of months or years . at first , the mean square variation @xmath106 in regard to interval @xmath107 is determined based on the sf of the observed light curve . in simulation of each light curve , an epoch during a sampling gap is randomly determined , and the flux is generated by random gaussian deviations , which is equivalent to the fr method except for using the variability parameter @xmath106 from the linearly interpolated flux from the observed data points . once the point is achieved , it is regarded as an observed data point in the light curve , so that the realized data should have self - correlation seen in the observed light curves . similar efforts are repeated until the mean sampling interval of the updated light curve decreases to one day . the light curve , generated this way , is used in each ccf method , instead of using the interpolated fluxes in either of @xmath89 or @xmath90 . table [ sf ] shows the sfs fitted by equation [ sf_pl ] using all the observed data in the @xmath3 and @xmath2 bands for each target agn . the mean square variation @xmath108 is determined directly from the observed sf . however , in a particular case of the @xmath2 band for ngc 5548 , it was determined as @xmath109 @xmath110 , by iteration until the simulated sf agrees with the observed sf . in this way we generated 1000 pairs of artificial @xmath3 and @xmath2 light curves , some of which are shown in figures [ simlc_ngc5548 ] to [ simlc_ngc7469 ] , together with the observed data . the lag time was calculated for each subset of the monitoring period , because we are interested in whether phase - to - phase changes in lag time exist , and also because we know that the variation in the @xmath2 band does not always respond linearly to that in the @xmath3 band . the lag time for the entire monitoring period was also calculated , but this is only for reference . in principle , the lag time was calculated from the data taken over a period between two adjacent solar conjunctions . only for ngc 5548 , we combined the second season of year 2002 and a part of the third season of year 2003 , so that the lag time was calculated over a full period including the minimum luminosity state . the ccpds from monte carlo simulations are shown in figures [ ccpd_ngc5548 ] to [ ccpd_ngc7469 ] , where the frequency distribution of @xmath99 over simulation runs is plotted as a histogram . here , such runs of low ccf peaks with confidence levels of less than 95 % are excluded , where the confidence level is derived from the peak value of ccf and the number of observed data points in the overlapping period of the two time series shifted after by @xmath69 . table [ lag_vk ] shows the cross - correlation results from the simulated light curves . the lag time was determined as a median of @xmath99 , or 50 % fraction of the cumulative ccpd . the uncertainty is measured by @xmath11134.1 % fraction from the median , thus corresponding to 1@xmath44 error for a normal statistical distribution . the lag time for ngc 4051 needs careful interpretation . there are two comparable peaks in the ccf , located at @xmath112 days and @xmath113 days . for the latter case , however , the major ascending and descending parts of the @xmath2 light curve are shifted into the blank periods of the @xmath3 light curve , so that the high correlation may not be real . the rapid @xmath3-band variations on short timescales of up to a month are not at all correlated with the @xmath2-band light curve if shifted backwards by @xmath113 days . according to equation [ ccf4 ] , the @xmath3 and @xmath2 variations of larger amplitude and longer timescale in the overlapping period tend to maximize the ccfs more effectively by chance , even if those of shorter timescale are in fact highly correlated . it is very likely that the global flux variation for ngc 4051 contains information on some other effect of time shift in addition to the light - travel effect . considering the above , the lag time shown in the table is calculated from the histogram below the dip at @xmath114 days . we detected the lag time for ngc 3227 for the first time , though the uncertainty is not very small . as seen from figure [ lc_ngc3227_vk ] , the optical variation of ngc 3227 is very rapid like that of ngc 4051 . the light curve data are undersampled because of some unfortunate gaps , especially in 2003 . more observations are needed to accurately measure the lag time for such an agn with low - luminosity and rapid optical variation . there is a possibility that the lag time may be underestimated , not representing the light - travel time from the central engine to the hot dust region . the contribution of the @xmath2-band flux from the central engine , if any , would make the phase of near - infrared variations closer to the corresponding phase in the optical . in this case , the lag time should be systematically smaller than the light - travel time from the central engine to the hot dust region . if the power - law optical component from the central engine is extended to the near - infrared wavelength region , its maximum contribution in the @xmath2-band flux is about 10 % . subtraction of this component from the @xmath2-band flux observed on the same night would increase the lag time by @xmath115 % . in most cases , the ccpds are smooth and have clear peaks , and a lag between the @xmath3 and @xmath2 variations is detected ( figures [ ccpd_ngc5548 ] to [ ccpd_ngc7469 ] ) . generally , bi - ccfs tend to have a sharp jump , in contrast to es - ccfs for which the pairs are sampled densely and constantly . such a jump appears close to the peak and affects the values of @xmath97 . we therefore use the es - ccf result in table [ lag_vk ] as representing the lag time in following discussions . note that the lag time for all data is only for reference and is not shown in the plots . the dust reverberation models , applied to previous observations of agns , have estimated an equilibrium temperature at the inner radius of the dust torus ( clavel , wamsteker & glass 1989 ; nelson 1996 ) , which is consistent with the thermal emission in the near - infrared originating from the hot dust heated at its sublimation temperature in the range of @xmath11 k ( barvainis 1992 ) . as a result , there would be an inner hole of hot dust torus whose radius is proportional to a square root of central luminosity . the lag time between uv / optical continuum variations and near - infrared ( @xmath1162 @xmath117 m ) continuum variations should be the light - travel time from the center to the dust region . our lag measurements of the four agns ( ngc 5548 , ngc 4051 , ngc 3227 , ngc 7469 ) are given in table [ lag_vk ] , and are plotted against the absolute @xmath3 magnitude in figure [ dtmv_all ] , together with previous measurements . the filled symbols are our results , including our early result for ngc 4151 ( minezaki et al . the others are taken from the literature ( see minezaki et al . 2004 and references therein ) . it is evident from this figure that a good correlation of @xmath118 , indicated earlier from @xmath119 to @xmath120 by minezaki et al . ( 2004 ) , now extends down to @xmath121 , spanning a range of 1000 times in luminosity . in particular , we have expanded the range of @xmath122 faintwards by about three magnitudes by measuring , for the first time , the lag time of ngc 4051 which sits in the faint end of `` classical '' seyfert nuclei . this indicates that agns of low luminosity might also have the dust torus expected from the dust reverberation models . the good correlation between lag time and absolute @xmath3 magnitude strongly supports our idea of using @xmath118 as a distance indicator for agns ( kobayashi et al . 1998b ; yoshii 2002 ) . given a narrow range of flux variations throughout our monitoring period , we could not find clear differences among successive lag measurements for individual agns . observations covering larger flux variations are desirable for investigating how the hot dust would be redistributed against changes in the power of the central engine . one of other fundamental quantities that characterize agns is the central mass , which could be regarded as the mass of a super - massive black hole . for the high - energy region , comparison of the characteristic frequency of x - ray variability power spectra of agns with that of black hole x - ray binaries has led to an interpretation that the size of the x - ray - emitting region would linearly scale with gravitational radius or gravitational mass , so that the x - ray variability has been proposed as a method of measuring the mass of agns ( hayashida et al . 1998 ; uttley et al . 2002 ) . for the uv or optical region , the optically thick and geometrically thin accretion disk is generally believed to be a source of uv and optical continuum from agns , and the standard model expects a radial temperature distribution in the disk that depends on both accretion rate and central mass ( frank , king , & raine 1992 ) . therefore , it is worthwhile to examine how the inner radius of the dust torus scales with central mass . recent measurements of lag time and velocity dispersion for several broad emission lines in ngc 5548 have given evidence for kepler motion of the blr gas ( peterson & wandel 1999 ) . assuming that the virial theorem applies similarly to about three dozen seyfert nuclei and quasars that have been targets of reverberation mapping observations for the blr ( e.g. , wandel , peterson & malkan 1999 ; kaspi et al . 2000 ; peterson et al . 2004 ) , their central masses have been estimated with a typical uncertainty of a factor of 2 or 3 over a range of @xmath123 . figure [ dtmbh ] shows the infrared lag measurements plotted against the central masses that have been measured from reverberation radii and widths of broad emission lines in individual agns . the meanings of filled and open symbols are the same as in figure [ dtmv_all ] for infrared lag measurements . the central masses are taken from peterson et al . the inclined lines correspond to @xmath124 , @xmath125 , @xmath126 , and @xmath127 , where @xmath128 is the light - travel distance or inner radius of the dust torus , and @xmath129 is the gravitational radius which scales linearly with central mass @xmath130 . correlation between the central mass and the lag time ( figure [ dtmbh ] ) is weaker than between the central optical luminosity and the lag time ( figure [ dtmv_all ] ) . furthermore , the inner radius of dust torus @xmath128 does not scale with gravitational radius @xmath131 that encloses the central mass . it is therefore evident that the inner radius of dust torus is not determined by the dynamics under the influence of a supermassive black hole at the center , but rather by the strength of uv / optical radiation from the central engine . as a consequence , the scatter in figure [ dtmbh ] might be interpreted as the difference in mass - to - luminosity ratio that should be determined by the accretion rate and the radiative efficiency in accretion flow in individual agns . confirmation of a dust torus outside the blr not only verifies the unification scheme of agns , but also yields knowledge of physical conditions in the central regions of agns for which we can not yet obtain spatially resolved images . previously , clavel , wamsteker & glass ( 1989 ) measured both blr and dust reverberation radii during the same monitoring campaign for a bright seyfert 1 galaxy of fairall 9 , and concluded that the hot dust is outside the blr . we now examine whether this view is supported by our measurements of infrared lags . figure [ blrdust ] shows our measurements of infrared lags ( ngc 5548 , ngc 4051 , ngc 3227 , ngc 7469 ; ngc 4151 , minezaki et al . 2004 ) , together with previous measurements of broad - line lags in the literature ( fairall 9 , clavel , wamstekar & glass 1989 , rodriguez - pascual et al . 1997 , peterson et al . 2004 ; ngc 3783 , reichert et al . 1994 ; ngc 7469 , wanders et al . 1997 , kriss et al . 2000 , collier et al . 1998 ; ngc 5548 , krolik et al . 1991 , peterson & wandel 1999 , dietrich et al . 1993 , korista et al . 1995 ; ngc 4151 , clavel et al . 1990 , maoz et al . 1991 , kaspi et al . 1996 ; ngc 3227 , winge et al . 1995 , onken et al . 2003 ; ngc 4051 , shemmer et al . 2003 ) . the optical luminosities corresponding to the broad - line lags are averaged over the monitoring period in each item in the literature . each agn has a range of lag time for the blr depending on the broad emission line used ( korista et al . 1995 ; peterson & wandel 1999 ; see also figure [ blrdust ] ) . in general , there is a tendency for broad emission lines of higher ionization to have smaller lag times , and for those of lower ionization to have larger lag times , which suggests that the blr has a radially stratified ionization structure . therefore , the blr extends out to the light - travel distance corresponding to the largest lag time in each agn . we note that a correlation of @xmath132 has been suggested for broad h@xmath59 emission lines for a wide range of central luminosity of agns including quasars brighter than @xmath133 ( kaspi et al . 2000 ) , although their lags for seyfert galaxies show a fairly large scatter . figure [ blrdust ] shows that the lag times for broad lines are located below the boundary set by the infrared @xmath94 versus @xmath122 relation . some lags of broad emission lines are beyond the infrared lag in the same agn . however , this occurs when such lags for the blr were measured in a bright state , while the infrared lag for hot dust was measured in a faint state . such lags for the blr , after being corrected for their luminosity dependence , become smaller than the infrared lag , as actually reported for ngc 5548 ( suganuma et al . 2004 ) . thus , the outer radius of blr nearly corresponds to the inner radius of the dust torus . in other words , it is suggested that the bulk of the emitting region of high - velocity gas is restricted inside the `` wall '' of the dust torus which is located at a distance determined by the continuum luminosity . netzer & laor ( 1993 ) proposed that the dust in the narrow line - emitting gas might separate the blr from the nlr by suppressing the emission lines in the dust region , and that this might explain the large difference in observed properties of broad and narrow emission lines , such as line width and covering factor . the result presented in figure [ blrdust ] supports their view that the blr size is determined by the dust sublimation radius . it has been proposed that the uv / optical flux variation in agns originates in the thermal reprocessing of x - ray variability by the optically thick and geometrically thin accretion disk in its outer part , while the bulk of the steady component is generated by viscous processes in the accretion disk itself ( e.g. , collin - souffrin 1991 ) . in this scheme , the variation of reprocessed component in the uv / optical flux should show a lag time of days or less behind the variation of x - rays released from the geometrically thick hot corona in the inner part of the accretion disk . for ngc 5548 , clavel et al . ( 1992 ) reported a uv lag of @xmath134 days behind the x - ray , by comparing the light curves in the hard x - ray ( @xmath135 kev ) and the uv ( 1350 ) . it is not easy to detect a clear correlation between x - ray and uv / optical variations on short timescales , nor the lag time between them , because generally the amplitude of uv / optical variation on a short timescale is small and difficult to observe . from observations of optical ( 5100 ) and hard x - ray ( @xmath135 kev ) fluxes of light curves spanning six years for ngc 5548 , uttley et al . ( 2003 ) reported a strong correlation in their long - timescale variation , while only weakly restricting @xmath136 days . since the amplitude of such optical variation is found to exceed that of hard x - ray , they argued that x - ray reprocessing is not a main driver of the long - timescale optical variation , at least for ngc 5548 . thus , the problem yet to be answered is whether there is a clear correlation between x - ray and optical variations on short timescales . as shown below , because of accurate photometry and frequent sampling of the @xmath3-band light curve data , we detected , for the first time , the day - basis optical variation having a small amplitude and a clear lag time behind the x - ray variation , which is well within a framework of reprocessing the x - ray flux into the optical flux . the top panel of figure [ ngc5548xv ] shows our @xmath3 light curve for ngc 5548 ( figure [ lc_ngc5548 ] ) and the x - ray ( @xmath135 kev ) light curve obtained for the same period from the rossi x - ray timing explorer ( rxte , uttley et al . these two light curves strongly correlate with each other , but the amplitude of the optical variation is much smaller than that of the x - ray . the bottom panel shows a portion of the light curves for a period of @xmath137 with an expanded vertical scale for the optical flux . although the optical data were not sampled as often as the rxte , a day - basis correlated variation as well as an optical lag of one day or two behind the x - ray is clearly seen . cross - correlation analysis gives a lag time of @xmath138 days between the two light curves . a preliminary result was reported in suganuma et al . ( 2004 ) . the size of the x - ray emitting region in agns is considered to be as small as a light - travel distance of hours , since a significant fraction of x - ray flux varies on this timescale ( figure [ ngc5548xv ] ) . the observed correlated variability indicates that the optical emitting region in ngc 5548 is located at a light - travel distance of one day or two from the x - ray emitting region , and favors the thermal reprocessing of x - ray variability by the optically thick accretion disk , rather than a non - thermal process which emits both x - ray and uv / optical fluxes from almost the same region . on the other hand , the long - timescale x - ray and optical variations might be generated by some other mechanisms , such as instabilities of the accretion disk itself . our result also shows a geometrical relation between accretion disk and blr . as seen in figure [ blrdust ] , the broad emission line of highest ionization in ngc 5548 gives the shortest lag of a few days ( korista et al . 1995 ; peterson & wandel 1999 ) . the detected lag time of one or two days indicates that the uv / optical emitting region of the accretion disk is located in the innermost part of the blr in ngc 5548 . we carried out the most intense monitoring observations yet made in optical and near - infrared wavebands for four nearby seyfert 1 galaxies , including ngc 5548 , ngc 4051 , ngc 3227 , and ngc 7469 . clear time - delayed response of the @xmath2-band flux variations to the @xmath3-band variations was detected for all the galaxies , which can be interpreted as light - travel time between the central engine and the inner edge of the dust torus . we modified a cross - correlation analysis to measure the lag times with high reliability , and our infrared lag times are found to range from 10 to 80 days , depending on the optical luminosity of the agn . comparing our infrared lag measurements with other quantities ( optical luminosity , central virial mass , and broad - line lag ) based on our sample and others in the literature , we summarize our findings as follows : 1 . the infrared lag time is tightly correlated with the optical luminosity according to @xmath139 in agreement with the expected relation of dust reverberation . 2 . a large scatter between lag time and central virial mass suggests that an inner radius of the dust torus is not directly related to the dynamics of a supermassive black hole at the center . 3 . infrared lags place an upper boundary on similar lag measurements of broad - emission lines in the literature . this not only supports the unified scheme of agns , but also implies a physical transition from the blr out to the dust torus that encircles the blr . our @xmath3 light curve for ngc 5548 shows a clear lag of one day or two behind the x - ray light curve obtained from the rxte satellite , which indicates that the optical variation on short timescale originates in from the thermal reprocessing of x - rays by the optically thick accretion disk . ongoing monitoring observations for agns in our sample , aided by the results from other samples and different wavelengths , would enable us to understand the size and structure of dust torus of agns , not only from their statistical properties but also from their time evolution . simultaneous reverberation studies of near - infrared continuum and broad emission lines of low ionization for several years would provide a clear view of the relation between dust torus and broad emission - line regions . we thank mamoru doi and kentaro motohara for discussions and advice during the observations . we also thank colleagues at the haleakala observatories for their help in facility maintenance . this research has been supported partly by the grant - in - aid of scientific research ( 10041110 , 10304014 , 12640233 , 14047206 , 14253001 , 14540223 ) and coe research ( 07ce2002 ) of the ministry of education , science , culture and sports of japan , and 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 . ccccrcc ngc5548 ...... & @xmath140 @xmath141 @xmath142 & @xmath143 @xmath144 @xmath145 & 15.3 & 5359 & 0.068 & @xmath146 ngc4051 ...... & @xmath147 @xmath148 @xmath149 & @xmath150 @xmath151 @xmath152 & 15.0 & 924 & 0.043 & @xmath153 ngc3227 ...... & @xmath154 @xmath155 @xmath156 & @xmath157 @xmath158 @xmath159 & 14.4 & 1480 & 0.075 & @xmath160 ngc7469 ...... & @xmath161 @xmath148 @xmath162 & @xmath163 @xmath164 @xmath165 & 14.5 & 4521 & 0.228 & @xmath166 clccc r@c@l r@c@l ngc 5548 & refa & @xmath140 @xmath141 @xmath167 & @xmath168 @xmath169 @xmath170 & - & 14.422&@xmath111&0.006 ( 19 ) & 13.782&@xmath111&0.003 ( 37 ) + & refb1 & @xmath171 @xmath172 @xmath173 & @xmath174 @xmath175 @xmath176 & - & 13.742&@xmath111&0.004 ( 19 ) & 13.129&@xmath111&0.002 ( 37 ) + ngc 4051 & refa & @xmath147 @xmath177 @xmath178 & @xmath179 @xmath180 @xmath181 & - & 13.568&@xmath111&0.007 ( 13 ) & 12.636&@xmath111&0.003 ( 27 ) + & refb2 & @xmath147 @xmath148 @xmath182 & @xmath179 @xmath183 @xmath184 & - & 14.472&@xmath111&0.005 ( 13 ) & 13.874&@xmath111&0.006 ( 27 ) + ngc 3227 & refa & @xmath154 @xmath155 @xmath185 & @xmath186 @xmath187 @xmath188 & - & 13.394&@xmath111&0.003 ( 7 ) & 12.771&@xmath111&0.003 ( 12 ) + & refb & @xmath154 @xmath189 @xmath190 & @xmath186 @xmath191 @xmath192 & - & 12.506&@xmath111&0.003 ( 7 ) & 11.959&@xmath111&0.003 ( 12 ) + ngc 7469 & refa & @xmath161 @xmath148 @xmath193 & @xmath194 @xmath195 @xmath196 & [email protected] ( 6 ) & 13.918&@xmath111&0.003 ( 4 ) & 13.180&@xmath111&0.003 ( 18 ) + & refb & @xmath161 @xmath197 @xmath198 & @xmath199 @xmath200 @xmath201 & [email protected] ( 6 ) & 13.891&@xmath111&0.003 ( 4 ) & 12.998&@xmath111&0.003 ( 18 ) + clcc l@c@l@l l@c@l@l l@c@l ngc 5548 & refa & @xmath140 @xmath141 @xmath167 & @xmath168 @xmath169 @xmath170 & & - & & & 12.227&@xmath111&0.004&(3 ) & 12.196&@xmath111&0.006 ( 9 ) + & refb1 & @xmath171 @xmath172 @xmath173 & @xmath174 @xmath175 @xmath176 & & - & & & & - & & & 11.531&@xmath111&0.004 ( 9 ) + & refb2 & @xmath171 @xmath172 @xmath202 & @xmath174 @xmath175 @xmath203 & & - & & & 13.072&@xmath111&0.009&(3 ) & & - & + ngc 4051 & refa & @xmath147 @xmath177 @xmath178 & @xmath179 @xmath180 @xmath181 & 10.606&@xmath111&0.004&(4 ) & 10.06&@xmath111&0.01 & ( 1 ) & 9.948&@xmath111&0.004 ( 13 ) + & refb1 & @xmath147 @xmath148 @xmath204 & @xmath179 @xmath183 @xmath205 & 10.545&@xmath111&0.004&(4 ) & & - & & & 10.243&@xmath111&0.003 ( 13 ) + & refb2 & @xmath147 @xmath148 @xmath182 & @xmath179 @xmath183 @xmath184 & & - & & & 12.34&@xmath111&0.01 & ( 1 ) & & - & + ngc 3227 & refa & @xmath154 @xmath155 @xmath185 & @xmath186 @xmath187 @xmath188 & 11.483&@xmath111&0.005&(5 ) & 11.143&@xmath111&0.004&(9 ) & 11.092&@xmath111&0.004 ( 13 ) + & refb & @xmath154 @xmath189 @xmath190 & @xmath186 @xmath191 @xmath192 & 10.806&@xmath111&0.005&(5 ) & 10.508&@xmath111&0.004&(9 ) & 10.459&@xmath111&0.004 ( 13 ) + ngc 7469 & refa & @xmath161 @xmath148 @xmath193 & @xmath194 @xmath195 @xmath196 & 11.74&@xmath111&0.01 & ( 2 ) & 11.366&@xmath111&0.010&(4 ) & 11.308&@xmath111&0.006 ( 9 ) + & refb & @xmath161 @xmath197 @xmath198 & @xmath199 @xmath200 @xmath201 & 11.22&@xmath111&0.01 & ( 2 ) & 10.778&@xmath111&0.010&(4 ) & 10.696&@xmath111&0.004 ( 9 ) + c c c c c c c ngc 5548 ...... & - & 1.4 & 3.7 & - & 20 & 16 + ngc 4051 ...... & - & 4.7 & 8.6 & 41 & 47 & 39 + ngc 3227 ...... & - & 3.0 & 6.4 & 50 & 66 & 60 + ngc 7469 ...... & 3.3 & 5.4 & 8.6 & 41 & 54 & 56 + cc r@c@l r@c@lc ngc 5548 ...... & @xmath21 & @xmath2060&.&025 & 0&.&058 + & @xmath3 & 0&.&001 & 0&.&095 + & @xmath22 & 0&.&23 & 0&.&28 + & @xmath2 & 0&.&41 & & - & + ngc 4051 ...... & @xmath3 & @xmath2060&.&02 & 0&.&066 + & @xmath2 & 0&.&006 & 0&.&39 + ngc 7469 ...... & @xmath23 & 0&.&058 & & - & + & @xmath21 & @xmath2060&.&007 & 0&.&066 + & @xmath3 & 0&.&066 & 0&.&075 + & @xmath24 & 0&.&77 & & - & + & @xmath22 & 0&.&54 & 0&.&81 + & @xmath2 & 0&.&67 & 0&.&55 + ccc rrc c rrc ngc 5548 ...... & 893 & 110 & 101 & 8.9 & 3 & & 95 & 9.5 & 4 + ngc 4051 ...... & 915 & 112 & 84 & 11.0 & 5 & & 89 & 10.4 & 5 + ngc 3227 ...... & 589 & 151 & 40 & 15.1 & 4 & & 40 & 15.1 & 4 + ngc 7469 ...... & 748 & 154 & 88 & 8.6 & 3 & & 86 & 8.8 & 3 + cccccccccc ngc 5548 ...... & 2.6 & 0.9 & 0.32 & 3.2 & & 16 & 4 & 0.26 & 2.9 + ngc 4051 ...... & 3.6 & 0.6 & 0.16 & 2.0 & & 31 & 7 & 0.22 & 2.5 + ngc 3227 ...... & 6.6 & 0.4 & 0.06 & 1.3 & & 18 & 3 & 0.18 & 2.0 + ngc 7469 ...... & 5.8 & 0.6 & 0.10 & 1.8 & & 41 & 3 & 0.06 & 1.2 + cccr r@lc c r@lc ngc 5548 ...... & all data & @xmath207 & 101 & 45&.1 & 0.909 & & 49&.2 & 0.920 + & year 2001 & @xmath208 & 36 & 47&.8 & 0.962 & & 48&.0 & 0.961 + & year 2002 - 2003 & @xmath209 & 52 & 49&.5 & 0.968 & & 48&.3 & 0.947 + & year 2003 & @xmath210 & 42 & 57&.0 & 0.952 & & 53&.1 & 0.946 + ngc 4051 ...... & all data & @xmath211 & 84 & @xmath2125&0 & - & & @xmath2125&0 & - + & year 2001 & @xmath213 & 28 & 9&.0 & 0.347 & & 10&.3 & 0.479 + & year 2002 & @xmath214 & 30 & 11&.3 & 0.894 & & 11&.4 & 0.886 + & year 2003 & @xmath215 & 26 & 25&.9 & 0.804 & & 17&.7 & 0.779 + ngc 3227 ...... & all data & @xmath216 & 40 & 10&.9 & 0.640 & & @xmath2178 & & - + & year 2002 & @xmath218 & 12 & 16&.7 & 0.701 & & 18&.9 & 0.772 + & year 2003 & @xmath219 & 28 & 6&.7 & 0.714 & & @xmath2175 & & - + ngc 7469 ...... & all data & @xmath220 & 88 & 87&.7 & 0.716 & & 86&.0 & 0.606 + & year 2001 & @xmath221 & 53 & 88&.8 & 0.798 & & 87&.9 & 0.822 + & year 2002 & @xmath222 & 30 & 52&.9 & 0.846 & & 55&.4 & 0.826 + lcccclcc ngc 5548 ...... & @xmath223 & 1.46 & 60 & & @xmath224 & @xmath225 & 60 ngc 4051 ...... & @xmath226 & 0.71 & 30 & & @xmath227 & @xmath228 & 30 ngc 3227 ...... & @xmath229 & 0.91 & 20 & & @xmath230 & @xmath231 & 50 ngc 7469 ...... & @xmath232 & 0.82 & 30 & & @xmath233 & @xmath234 & 50 cc ccc cr@l cr@l ngc 5548 ...... & all data & @xmath235 & @xmath236 $ ] & & & @xmath237&@xmath238 & & @xmath239&@xmath240 + & year 2001 & @xmath241 & @xmath242 $ ] & & & @xmath243&@xmath244 & & @xmath245&@xmath246 + & year 2002 - 2003 & @xmath247 & @xmath248 $ ] & & & @xmath249&@xmath250 & & @xmath251&@xmath252 + & year 2003 & @xmath253 & @xmath248 $ ] & & & @xmath254&@xmath255 & & @xmath256&@xmath257 + ngc 4051 ...... & all data & @xmath258 & @xmath259 $ ] & & & * @xmath260*&*@xmath261 * & & * @xmath262*&*@xmath263 * + & year 2001 & @xmath264 & @xmath265 $ ] & & & * @xmath266*&*@xmath267 * & & * @xmath268*&*@xmath269 * + & year 2002 & @xmath270 & @xmath271 $ ] & & & @xmath272&@xmath273 & & @xmath274&@xmath275 + & year 2003 & @xmath276 & @xmath277 $ ] & & & @xmath278&@xmath279 & & * @xmath280*&*@xmath281 * + ngc 3227 ...... & all data & @xmath282 & @xmath283 $ ] & & & * @xmath284*&*@xmath285 * & & - & + & year 2002 & @xmath282 & @xmath286 $ ] & & & * @xmath287*&*@xmath288 * & & * @xmath289*&*@xmath290 * + & year 2003 & @xmath282 & @xmath283 $ ] & & & @xmath291&@xmath292 & & - & + ngc 7469 ...... & all data & @xmath293 & @xmath294 $ ] & & & * @xmath295*&*@xmath296 * & & * @xmath297*&*@xmath298 * + & year 2001 & @xmath299 & @xmath300 $ ] & & & @xmath301&@xmath302 & & @xmath303&@xmath304 + & year 2002 & @xmath305 & @xmath306 $ ] & & & @xmath307&@xmath308 & & * @xmath309*&*@xmath310 * +

we present a general model to account for the multimode nature of the quantum electromagnetic field in projective photon - counting measurements . we focus on photon - subtraction experiments , where non - gaussian states are produced conditionally . these are useful states for continuous - variable quantum information processing . we present a general method called mode reduction that reduces the multimode model to an effective two - mode problem . we apply this method to a multimode model describing broadband parametric downconversion , thereby improving the analysis of existing experimental results . the main improvement is that spatial and frequency filters before the photon detector are taken into account explicitly . we find excellent agreement with previously published experimental results , using fewer free parameters than before , and discuss the implications of our analysis for the optimized production of states with negative wigner functions .

0902.2867

the ability to prepare and measure specific quantum states of the light is the keystone of many quantum information processing ( qip ) protocols . these states can be described either with discrete variables in terms of photons , or with continuous variables in terms of waves . in the latter case , the physical quantities of interest are the amplitude and the phase of the light wave , or their cartesian counterparts called quadratures @xmath0 and @xmath1 . a very convenient representation of the quantum state is then provided by the wigner function @xmath2 , which corresponds to a quasi - probability distribution of the quadratures , ` quasi- ' because @xmath3 may assume negative values . an important task for qip is the ability to undo effects of decoherence by ` distillation ' : to obtain a single quantum state that is more pure from two or more copies that have undergone decoherence . since states of light with gaussian wigner functions can not be distilled with gaussian operations @xcite , one is left with two strategies : either to distill gaussian states with non - gaussian operations , or to distill non - gaussian states with gaussian operations @xcite . this paper is a contribution to the latter strategy , and focuses on the preparation of the non - gaussian states rather than on their distillation . the negativity of the wigner function is a standard figure of merit , quantifying at the same time how non - gaussian and how non - classical a quantum state is @xcite . one way of obtaining non - gausian states is by conditional photon - counting measurements , as first proposed by dakna _ _ @xcite . it was soon realized that such conditional measurements can improve quantum teleportation of continuous variables @xcite . in recent years , several experiments @xcite combining continuous- and discrete - variable tools allowed for preparing and observing quantum states of free - propagating light with negative wigner functions @xcite . many of these experiments are based on the use of a squeezed vacuum produced by parametric fluorescence , which involves many optical modes @xcite . this multimode nature is exhibited in both continuous - wave ( cw ) operation , using optical parametric oscillators ( opo ) below threshold @xcite , and in pulsed experiments with a single - pass high amplification . in order to make accurate predictions , it is therefore crucial to develop multimode theoretical models . this was done in @xcite for setups using an opo , and in @xcite for pulsed @xmath4-photon fock state tomography in a case of very low squeezing , when fock states expansion are limited to @xmath4 photon only . however , these models do not fully account for all phenomena linked to the non - constant space and time profiles of the modes under study , the spatial pulse profile in the transverse direction for example , and they especially do not account for gain - induced distortions in the parametric amplification process @xcite . as we will see later on , these phenomena are one main signature of this multimode nature , and are critical in the case of single - pass pulsed experiments . in this paper we propose an alternative general framework to describe the generation of squeezed light , with a twofold goal : first , to introduce a method that reduces a multimode model to an effective two - mode description . second , to show that a specific spatio - temporal multimode model for photon - subtraction experiments and our mode - reduction analysis thereof give an improved understanding of state - of - the - art photon - subtraction experiments . in sec . [ secmodered ] we show how to reduce a multi - mode model to an effective two - mode model . this mode - reduction procedure is then applied in sec . [ secapp ] to give an improved analysis of the photon - subtraction experiments of ref . we discuss the method and its application and conclude in sec . [ secdiscconc ] . some technicalities are deferred to two appendices . as a starting point we have a complete set of spatial , temporal , or spatio - temporal optical modes , in terms of which the light propagation can be described . the modes have field operators @xmath5 and @xmath6 with bosonic commutation relations @xmath7=\delta_{jk}$ ] . the @xmath8 may also stand for continuous operators like @xmath9 , where @xmath10 is time , with commutation relation @xmath11=\delta(t - t')$ ] , in which case sums over modes are to be replaced with integrals . to simplify the notation even further , we introduce @xmath12 , the column vector of all the @xmath5 . similarly @xmath13 is the row vector which has operators @xmath14 as components . we shall also need the row vector @xmath15 and the column vector @xmath16 . similar notation will be used for other vectors . in this section we consider a general unitary transformation @xmath17 in which output quadratures linearly depend on input quadratures , thus preserving the gaussian nature of the quantum fields . in fact , the only non - gaussian operation will be the projective measurement , corresponding to a detection event in a subset of the output modes by an avalanche photodiode ( apd ) , see figure [ gensetup ] . . after the transformation a few modes are filtered out and result in a detection event in the avalanche photodiode ( apd ) . such a detection event in the apd prepares a state in the single mode which is mode analyzed by homodyne detection . ] we will assume the evolution operator @xmath17 in fig . [ gensetup ] to be a bogoliubov transformation , where the output field operators depend linearly on the input ones : @xmath18 where @xmath19 and @xmath20 are two matrices which satisfy @xmath21 whereby the commutation relations of the field operators are preserved . the output state of the light can be characterized by doing homodyne measurements on a normalized mode described by @xmath22 , which has the mode operator @xmath23 this mode can be defined as the mode that perfectly matches the local oscillator of the homodyne detector . after the bogoliubov transform ( [ bogoliubov ] ) it is described by @xmath24 without conditioning upon photon detection , and since the gaussian nature of the initial vacuum state is preserved by the bogoliubov transform , the homodyne measurements will show gaussian wigner functions corresponding to squeezed vacuum states of light . the point is now how to describe the output state conditional upon detection of a photon by the apd , which is a projective measurement . analogous to the homodyne detection , we assume the photon detection mode @xmath25 to be described by a normalized state @xmath26 , with corresponding field operator @xmath27 after the time evolution described by eq . ( [ bogoliubov ] ) , the output at the photon detector becomes : @xmath28 if we only know that a photon has been detected but not in which detection mode , then we should average the conditional output state over all these detection modes , as is shown in more detail below . at this point we have a multimode output state that in principle we can update given the detection of a photon in the detector @xmath25 . in practice it is convenient to first simplify the multimode expressions ( [ ahout ] ) and ( [ adout ] ) . the first and central step in the mode - reduction procedure is to rewrite the homodyne mode eq . ( [ ahout ] ) in the form @xmath29 in terms of new mode operators @xmath30 with standard commutation relations . the coefficients @xmath31 , and @xmath32 in eq . ( [ modereductionahout ] ) are found as follows . besides having standard commutations , @xmath33 and @xmath34 in eq . ( [ modereductionahout ] ) should annihilate the vacuum state , so that @xmath33 must contain all annihilation operators in eq . ( [ ahout ] ) : is followed by a perfect non - degenerate optical parametric amplifier ( ndopa ) with squeezing parameter @xmath35 . ] @xmath36 with @xmath37 fixed up to a phase factor by @xmath38=1 $ ] . we choose @xmath37 to be real - valued and positive so that @xmath39 furthermore , from @xmath40=0 $ ] it follows that @xmath41=\frac{1}{\eta}\vec{\psi}_h^\dag \,u\,v^t\ , \vec{\psi}_h^*.\end{aligned}\ ] ] a complex value for @xmath42 can be removed with a redefinition of the phase of the homodyne mode , which just means that we can assume that the state is squeezed in the @xmath43 or @xmath44 direction . finally , @xmath32 can be found from the fact that @xmath45 and @xmath46 have commutator one , as have @xmath47 and @xmath48 : @xmath49 here again we used the freedom to choose @xmath32 real - valued and nonnegative . this completes the mathematics of the mode reduction of the multimode homodyning signal . the physical argument that two and only two modes should remain goes as follows . the squeezed vacuum after the bogoliubov transform can only be a centered gaussian state , hence it is fully described by only the variances @xmath50 and @xmath51 . the squeezed vacuum output can therefore be modeled @xcite by a perfect single - mode degenerate optical parametric amplifier ( dopa ) with squeezing parameter @xmath52 , followed by a perfect non - degenerate optical parametric amplifier ( ndopa ) with squeezing parameter @xmath53 , as presented in figure [ modelu ] . this is a two - mode model , with a hilbert space @xmath54 . the values of the parameters @xmath52 and @xmath55 can be deduced from the two independent coefficients in eqs . ( [ eta]-[beta ] ) , using : [ equ_gr ] @xmath56 a related squeezing parameter that we will also use in the following is @xmath57 . we now condition upon the measurement of a click in the photon detector ( apd ) . we assume to be in the limit that the average number of photons per pulse entering the photon detector is much less than one . then a single click in the detector corresponds to the detection of a single photon . one can make a reduced - mode description of the photon detection operator @xmath58 of eq . ( [ ad ] ) , analogous to eq . ( [ modereductionahout ] ) . the operator @xmath59 can be expanded into a part acting on @xmath54 , plus a component @xmath60 acting on the complementary space @xmath61 orthogonal to @xmath54 : @xmath62 note that @xmath60 on the right - hand side contains all the terms acting on @xmath61 , including creation operators . the coefficients in ( [ adperp_mode_red ] ) can again be found by taking commutators , for example : [ gammadelta ] @xmath63=\frac{1}{\eta}\vec{\psi}_h^\dag \,u\,v^t\ , \vec{\phi}_d^*(j ) \label{gamma}\\ \delta_j & = & [ \hat{a}_1,\hat{a}_{d,{\rm out}}(j)]=\frac{1}{\beta^*}[\vec{\phi}_d^\dag(j ) \,v\,v^\dag\ , \vec{\psi}_h-\alpha^*\gamma_j ] \label{delta}.\end{aligned}\ ] ] in general , a click recorded in the apd corresponds to the measurement of at least one photon . in the limit of low detection probability , the action of the detection is the subtraction of a _ single _ photon . note that this assumption is practically always obeyed if the parameter @xmath25 is in a continuum ( like in the case of spectral filtering ) , since the probability to have two photons exactly in the same mode is then negligible . henceforth we assume to be in this limit . in the heisenberg picture , a photon detection then corresponds to the application of the operator @xmath58 to the initial state ( _ i.e. _ to the vacuum state ) , followed by the normalization of the result : @xmath64 is the detection probability for the mode @xmath25 : @xmath65 below we will use that the vacuum expectation value of @xmath66 can be expressed as @xmath67 . before continuing , it can be instructive to recall the concision allowed by the heisenberg picture . in a single - mode problem , a photon - subtracted squeezed state is equivalent to a squeezed single - photon state : this case corresponds to the simple bogoliubov transform @xmath68 , which directly gives a pure @xmath4-photon state ( after normalization ) when applied to the vacuum . as the states do not evolve in the heisenberg picture , they all can be considered as ` input ' states ; but when measured using output quadratures , this @xmath4-photon state will appear to be squeezed . we are going to use the same approach in the multimode case . one can first note that the conditioned state @xmath69 in eq . ( [ psicond ] ) is already a @xmath4-photon state . this state , however , does not belong to @xmath54 only , so that measurements output are not so obvious to compute . in fact , we are solely interested in expectation values @xmath70 of operators describing the output that is measured in the homodyne detector . such expectation values can be written as @xmath71 in eq . ( [ rewriteexpvalb ] ) , the trace can be separated into a trace over @xmath54 and a trace over @xmath72 , and the latter does not act on the function @xmath73 , whose expectation value is then : @xmath74 all quantities of interest can therefore be deduced from the input state reduced density matrix @xmath75 , acting in @xmath54 , and the crucial advantage of mode reduction is to allow a simple expression for this matrix : writing @xmath76 , where @xmath77 and @xmath78 are the ground states of @xmath54 and @xmath61 , respectively , and using eqs . ( [ adperp_mode_red],[psicond ] ) , we directly obtain : @xmath79 in terms of the _ modal purity _ @xmath80 and where @xmath81 is an operator that creates a single photon in a superposition of mode 0 and mode 1 . the state ( [ rho2 ] ) produced from a detection event in mode @xmath25 is a mixed state , mixing vacuum and a single - photon state with weight @xmath82 . without conditioning or in the limit @xmath83 , we have @xmath84 . _ squeezed vacuum. _ before determining the output wigner function corresponding to the conditional state ( [ rho2 ] ) , it is instructive to first determine the wigner function of the output state in the simplest experimental situation , where we ignore the photon detector . the input state is then @xmath85 . we will make use of the standard wigner functions of the vacuum @xmath86 and of single - photon states @xmath87 , both with @xmath88 . clearly , @xmath89(x_{0},p_{0};x_{1},p_{1})$ ] equals @xmath90 . in order to obtain the output wigner function for the homodyne mode , we wish to express @xmath89 $ ] as a function of @xmath91 , @xmath92 ( defined from the output homodyne mode @xmath93 given by eq . ( [ modereductionahout ] ) ) . this requires the introduction of another mode @xmath94 orthogonal to @xmath95 , so that the transformation @xmath96 is symplectic ( _ i.e. _ commutation relations are preserved ) . using the model of fig . [ modelu ] , one can choose @xmath97 of the form : @xmath98 this form is by no means unique , but this does not pose a problem since mode @xmath99 will eventually be integrated out . one can now invert the relations ( [ modereductionahout],[a1out ] ) , thereby expressing @xmath100 as a function of @xmath101 , and @xmath102 . after tracing over mode @xmath99 , which amounts to integrating over @xmath103 and @xmath102 , we find the output signal entering the homodyne detector to be a squeezed vacuum state with a gaussian wigner function @xmath104 where @xmath43 and @xmath44 stand for @xmath105 and @xmath106 and with variances [ defvar ] @xmath107 we can now invert eq . ( [ defvar ] ) and rewrite the three mode - reduction parameters @xmath108 , and @xmath32 in eq . ( [ modereductionahout ] ) in terms of the variances , giving : [ etaalphabetavxvp ] @xmath109 in the following , we will keep writing @xmath108 , and @xmath32 to shorten notation . it should be kept in mind , however , that eq . ( [ etaalphabetavxvp ] ) directly expresses these parameters in terms of the measurable variances @xmath110 of the squeezed vacuum . in particular , @xmath32 vanishes for minimal - uncertainty states . notice also that the parametrization for the mode - reduction parameters ( [ etaalphabetavxvp ] ) is equivalent to the one in eq . ( [ equ_gr ] ) in terms of squeezing parameters @xmath52 and @xmath111 . thus @xmath52 and @xmath111 can be expressed in terms of the variances @xmath110 , and _ vice versa_. _ photon - subtracted squeezed vacuum. _ as for the squeezed vacuum , we now calculate the wigner function for the photon - subtracted squeezed vacuum , starting with the initial state ( [ rho2 ] ) . the mode ( [ athetaj ] ) has a one - photon excitation in state ( [ rho2 ] ) . the orthogonal mode with creation operator @xmath112 is not excited . hence the wigner function corresponding to the state ( [ rho2 ] ) is @xmath113 note that in this expression , quadratures @xmath114 , @xmath115 can be easily expressed as functions of quadratures @xmath100 using ( [ athetaj ] ) . as before , the wigner function for the output signal is found by using the symplectic transformation defined by eqs . ( [ modereductionahout ] ) and ( [ a1out ] ) . by tracing again over the mode @xmath99 , we obtain ( see appendix [ appendixwigner ] ) @xmath116 the constants in this wigner function are given by [ constantsinw ] @xmath117 note that the wigner function ( [ wj ] ) of the photon - subtracted squeezed state differs from the wigner function of the squeezed vacuum @xmath118 of eq . ( [ 0_sqz ] ) only because of the polynomial in @xmath43 and @xmath44 between the large brackets . the same quantities @xmath110 as in eq . ( [ defvar ] ) show up , with or without conditioning . since in general @xmath119 @xcite , we find the condition @xmath120 . _ averaged wigner functions. _ practical detectors do not resolve with infinite precision when and where photons are detected . we should therefore average over all possible microscopic states that agree with the detection record . we assumed in sec . [ secconditioning ] that the average number of photons detected per pulse in the apd is much smaller than one . averaging over unresolved detection events is then equivalent to averaging over single - photon subtraction events . the wigner transformation of the density matrix is a linear transformation . therefore , the averaged wigner function @xmath121 is simply obtained by replacing @xmath122 ... @xmath123 in ( [ wj ] ) by @xmath124 ... @xmath125 , with the notation @xmath126 here @xmath127 is the sum of the probabilities @xmath128 of microscopic states that agree with the detection record . from eq . ( [ constantsinw ] ) it follows that averaged quantities @xmath124 ... @xmath125 involve sums like @xmath129 ; in the following we will write the respective averages as @xmath130 , @xmath131 or @xmath132 . the wigner function ( [ wj ] ) of the photon - subtracted state and its detection - averaged version have a very general significance . before going to sec . [ secapp ] , devoted to a practical implementation of these results , let us finish with some reflections on the detection modes . all the previous results were derived through the use of a set of detection modes @xmath133 . the coefficients entering in the averaged wigner function @xmath134 involve quantities like @xmath127 , @xmath130 , @xmath131 or @xmath132 , in which the detection operators only appear through their projection operator : @xmath135 hence one would find the same predicted averages if one would employ a different set of detection modes that has the same associated projector . this projector describes how the setup filters the signal before it enters the photon detector . for example , a time - domain filtering system will be described by @xmath136 , where @xmath137 can be written in terms of a set of modes @xmath9 that are labeled by the time @xmath10 : @xmath138 where @xmath139 when the apd is switched on , and @xmath140 otherwise . to give another important example , a spectral slit can be described with a set of modes @xmath141 with an associated projector @xmath142 where @xmath143 if the frequency @xmath144 is filtered out , and @xmath145 otherwise . above we have assumed that the filtering of the signal after its production in the dopa was included in the bogoliubov transform @xmath17 . below we will give an alternative description , in which @xmath17 is separated into the transformation due to the production of the squeezed light in the dopa , and the subsequent filtering before detection . this alternative description will enable a more straightforward comparison with the empirical model in sec . [ secapp ] . so , instead of the input - output transform eq . ( [ adout ] ) for the photon detection operator , we now write @xmath146 where the bogoliubov matrix @xmath147 accounts for filters ( as the filters are passive , we have @xmath148 ) ; this matrix @xmath147 is of course unitary , even if the filters can present losses . in fact , losses will be modeled using beamsplitters , where the lost energy is reflected into auxiliary non - relevant modes . these modes do not interact with the rest of the experiment ( _ i.e. _ they are unaffected by bogoliubov transform @xmath149 ) and will not reach the apd . but all the other modes , referred to as relevant modes , should be considered as detection modes , and we then have : @xmath150 where @xmath151 is the projector onto the subspace of relevant modes . let @xmath152 be the projector on the non - relevant modes . as the latter are unaffected by the transform @xmath149 , we have @xmath153 . inserting the relation @xmath154 into ( [ adoutp ] ) then leads to @xmath155 here the last term on the right annihilates vacuum , and commutes with annihilation operators like @xmath33 or @xmath34 : this term will add no contribution to the results of the previous subsection . the only change therefore consists in the substitution @xmath156 , so that the operator @xmath157 in eq . ( [ equprojp ] ) should be replaced by @xmath158 where @xmath159 and where we have used standard properties of projection operators ( @xmath160 , @xmath161 ) . the operator @xmath162 represents the action of the filters restricted to the subspace of relevant modes . if @xmath162 is a projector , such as @xmath137 of eq . ( [ equprojpt ] ) or @xmath163 of eq . ( [ equprojpomega ] ) , then the effect of @xmath164 is the same as of @xmath157 in eq . ( [ equprojp ] ) . this can be easily understood : it is equivalent to say that the filtered modes are blocked , or that they are first redirected into auxiliary modes , and then blocked . the operator @xmath164 of eq . ( [ equprojp2 ] ) is a more general quantity than @xmath157 in eq . ( [ equprojp ] ) , however , since @xmath164 need not be a projection operator . it can for instance account for partial absorption of the modes . in that case , the spectral transmission @xmath165 in eq . ( [ equprojpomega ] ) can assume any value between @xmath166 and @xmath4 , to account for filtering systems more complex than a simple spectral slit . furthermore , the above expressions can simply be generalized to situations where several filters are used . for example , if a spectral slit @xmath163 is followed by a time - domain filter @xmath137 , then the above expressions for @xmath162 and @xmath164 become @xmath167 , leading to @xmath168 . at this stage , we have a complete description of the final state starting from the bogoliubov transform ( [ bogoliubov ] ) . the results of the previous section are generally valid , since we started with a multimode model that was left unspecified . our purpose now is to establish a concrete link with the photon - subtraction experiment as described in ref . @xcite , and to improve its analysis . in the experiment by ourjoumtsev _ et al . _ @xcite , pulses of squeezed light are produced . the setup is sketched in figure [ figdopa ] . are transformed into pairs of photons of frequency @xmath144 ( in the experiment , the central frequency @xmath169 corresponds to a central wavelength of about @xmath170 ) . the output signal of the dopa is sampled by a beam splitter with low reflectivity @xmath171 . if a photon is detected by the apd , then ideally it has been subtracted from the squeezed vacuum . ] a squeezed vacuum , produced in a single - pass dopa ( a @xmath172 crystal ) by down - conversion of frequency - doubled femtosecond laser pulses , is sampled by a beam splitter with low reflectivity @xmath173 . two mode filters are placed in front of the apd : a spatial mode filter , that consists in a single - mode fiber , and a spectral slit of width @xmath174 . if a photon is detected by the apd , then ideally it has been subtracted from the squeezed vacuum . this subtraction leads to a @xmath4-photon squeezed state , which is very close to a ` schr " odinger kitten ' state . quantum state tomography with a balanced homodyne detector @xcite allows the complete reconstruction of this highly non - gaussian quantum state of light . the wigner function ( [ wj ] ) was derived assuming that the mode reduction was performed on the mode entering the homodyne detector . to relate our results to the empirical model discussed in the next subsection , we here choose to perform the mode reduction to the signal directly after the dopa . we model the dopa using the scheme presented in fig . [ modelu ] , where the parameters @xmath52 and @xmath111 are linked to the bogoliubov transform through eqs . ( [ eta]-[equ_gr ] ) . for the calculation of the modes @xmath175 detected by the apd , the sampling beamsplitter and the mode filters can be separately added to this transform , as explained in sec . [ subsecdetmodes ] . as stated above , here we chose not to include the sampling beam splitter into the mode reduction . the wigner function ( [ wj ] ) then describes the signal just after the dopa . we therefore still need to account for this sampling beam splitter between the dopa and the homodyne detection , as well as for other losses . for example , one usually accounts for imperfections of the homodyne detection by adding a fictitious beam splitter of transmission @xmath176 just before the homodyne detection , where @xmath176 is the homodyne detection efficiency . both those beam splitters can be easily implemented by replacing the variances according to : @xmath177.\ ] ] and by multiplying @xmath178 , @xmath179 and @xmath180 by @xmath181 . before going into detailed calculations , let us first recall the empirical model that was proposed in ref . @xcite to account for experimental results . it is useful to recall the empirical model proposed in ref . @xcite to explain the experiments , and to see by what assumptions our multimode model reduces to it . the dopa is again modeled as in fig . [ modelu ] , producing the same squeezed vacuum . however , in the empirical model it is assumed that the detected photon is either in the homodyne mode with probability @xmath182 , or in an orthogonal mode with probability @xmath183 . in the latter case , the detection event is not correlated with the homodyne measurement , and one simply performs a homodyne measurement on squeezed vacuum . the output density matrix obtained with the empirical model is similar to our @xmath184 in eq . ( [ rho2 ] ) . in fact , the two would be identical if the detected photon was only due to photons in the mode @xmath185 or from @xmath72 . this is in general not the case , however , as there will be an admixture from @xmath186 in the photon detection operator . in fact , in order to completely account for the multimode nature of this experiment , the empirical model should be modified in the way depicted on fig . [ figmodelubeamsplit ] , with the insertion of a beamsplitter of amplitude reflection and transmission coefficients @xmath187 and @xmath188 that allows interference between @xmath189 and @xmath190 . a photon detection event in such a setup can indeed be equivalent to the application of @xmath191 to the initial vacuum [ see eqs . ( [ rho2],[athetaj ] ) ] , provided @xmath192 with @xmath193 . these angles @xmath194 and @xmath195 are mixing angles that fix the probability amplitudes of detection of a photon of mode 0 and of mode 1 . our angle @xmath195 in general depends both on the squeezing properties of the light source , and on the filtering of the signal before the photon detector , whereas the empirical @xmath196 only depends on the source . within a narrow - filter approximation that will be detailed in next subsection , such a setup can also account for averaged quantities ( [ xbar ] ) with the use of an average angle @xmath197 instead of @xmath195 ( see eq . ( [ theta bar ] ) in the following ) . this possibility of interference between the homodyne signal and the @xmath190 signal is the crucial difference between the multimode and the empirical models : more interference makes the empirical model worse . the essential assumption of the empirical model is thus that the photon detection operator does not have a contribution from @xmath190 . then @xmath198 and @xmath199 could be replaced by @xmath42 and @xmath32 , respectively , according to eqs . ( [ modereductionahout ] ) and ( [ adperp_mode_red ] ) , and the angle @xmath195 in eq . ( [ rho2 ] ) by @xmath196 ( in which case , according to eq . ( [ equivsetup ] ) , the beamsplitter @xmath200 in the equivalent model in fig . [ figmodelubeamsplit ] can be removed ) . , the resulting wigner function is equivalent to the two - mode model depicted in the figure . the dopa may in general be represented by an ideal single mode dopa and a two - mode ndopa as in fig . [ modelu ] , and the photon counting is a combination of dark counts as well as a coherent mixture of the two output ports of the ndopa . compared to the empirical model developed in ref . @xcite , the only difference is the presence of the beam splitter @xmath200 which was not present in the empirical model . ] the empirical wigner function can be easily deduced from eq . ( [ wj ] ) with the above replacements , and has the same form after the replacement of @xmath201 by @xmath202 . coefficients @xmath203 and @xmath204 are obtained by multiplying @xmath205 , @xmath206 by @xmath207 , and replacing @xmath208 and @xmath209 by @xmath42 and @xmath32 , respectively . this gives @xmath210 the coefficient @xmath211 is given by @xmath212 , and @xmath213 vanishes . the empirical model produces intuitive results . however , it requires justification . if large spectral slits would be used , then the homodyne mode and many other orthogonal modes would hardly be affected by the slit . if the detected photon could have come from many modes orthogonal to @xmath185 , then the modal purity @xmath182 would be unacceptably low , and also a large admixture of @xmath190 would enter the detection signal . indeed , some of us found experimentally that the spectral slit should be as narrow as possible , while still allowing the detection of a signal , in order to find the highest modal purities ( see also sec . [ sec : concretemultimode ] ) . consequently , narrow slits have been used in the photon - subtraction experiment @xcite . although it is obvious that filtering is necessary , the use of a narrow spectral slit before the photon detector does not make the empirical model automatically valid . a quantitative comparison of both models is therefore needed to test the validity of the empirical model , as given below . let us now develop a simple spatio - temporal multimode model for which the bogoliubov transformation can be written explicitly . we assume that light propagation inside the dopa is described by modes of the form @xmath214 where the plane wave is exactly phase - matched , and where the amplitude @xmath178 satisfies the slowly - varying envelope approximation ( svea ) . this approximation does not hold for all the light that exits the nonlinear crystal , but the homodyne mode is supposed to be phase - matched , and we will assume that the filters before the apd block the modes that are not phase - matched . we will furthermore neglect diffraction effects within the dopa . in the basis @xmath215 , where @xmath43 and @xmath216 are spatial variables running on the dopa s transverse plane , the @xmath19 and @xmath20 of the bogoliubov transform ( [ bogoliubov ] ) then become diagonal ( see appendix [ appendixbogoliubov ] ) , and we have @xmath217 in terms of operators that we assume to have commutation relations @xmath218= \delta(x - x')\delta(y - y')\delta(t - t').\ ] ] coefficients in the transformation ( [ bogoliubov2 ] ) have the form [ uvxyt ] @xmath219 , \label{uxyt}\\ v(x , y , t ) & = & \sinh[\,q l \ , e_p(x , y , t)\ , ] . \label{vxyt}\end{aligned}\ ] ] here @xmath220 is the pump - beam amplitude , which we assume to be real - valued . the parameter @xmath221 takes into account the nonlinearity of the crystal and @xmath222 is its length . one can allow for group velocity mismatch ( gvm ) in the crystal by convoluting @xmath223 by a rectangular unit gate of duration @xmath224 , the time separation induced by the gvm after passing the crystal ( see appendix [ appendixbogoliubov ] ) . more precisely , this convolution should be made twice , as we also have gvm for the second harmonic generation ( shg ) of the pump beam . the @xmath19 and @xmath20 are real - valued functions if the pump beam @xmath223 is so , which is a valid assumption if there is no frequency chirp . the homodyne mode @xmath225 will also be taken real - valued in the following . the homodyne signal is then given by @xmath226 , where integration over @xmath227 is implied . mode reduction now starts with the identification @xmath228 which is eq . ( [ a0 ] ) specified for our spatiotemporal model . the mode - reduction parameters are now given by spatio - temporal integrals , for example [ alphaeta2 ] @xmath229 further parameters can be found analogously . _ averaging over photon detection events. _ we have seen in sec . [ subsecdetmodes ] that all averaged quantities can be obtained through the determination of the operator @xmath157 defined in ( [ equprojp2 ] ) when the filters are separately added to the bogoliubov transform . in the considered experiment two filters are used : a rectangular spectral slit , that can be described using ( [ equprojpomega ] ) ; and a monomode fiber that selects a single spatial mode @xmath230 , which we suppose to be real - valued , and therefore corresponds to the projector @xmath231 . note that in this experiment detection times are unknown at the scale of pulses duration , so that there is no time - domain filtering . ( in the analysis one should average over all possible photon detection times . ) we therefore have to use @xmath232 where @xmath233 if @xmath144 enters into the spectral slit , @xmath166 otherwise . here @xmath171 is the sampling beamsplitter reflectivity , and @xmath234 accounts for all other losses in the conditioning arm ( apd efficiency , optics losses ... ) . as can be seen from eq . ( [ svea0 ] ) , the field amplitudes are defined around a central frequency @xmath169 ( or @xmath235 for the dopa pump beam , see appendix [ appendixbogoliubov ] ) , so that the frequency @xmath236 for the amplitude fourier transform corresponds in fact to this central frequency ; in that way , a rectangular spectral slit well centered around this central frequency can be defined as @xmath237 $ ] . the operator @xmath157 defined in eq . ( [ equpmodel ] ) should be applied to fourier - transformed mode functions : @xmath238 using eqs . ( [ pj ] ) , ( [ equpmodel ] ) and ( [ fourier ] ) , the average total photon detection probability per pulse then becomes @xmath239 where the average is taken over the detection modes @xmath240 . the expression ( [ ptotcalc ] ) increases linearly with the filter width @xmath174 , but is valid only if @xmath174 is small enough to warrant the svea . furthermore , when conditioning upon a click in the detector in sec . [ secconditioning ] , we assumed that @xmath241 , an assumption that can now be tested with the explicit formula ( [ ptotcalc ] ) . in general , the parameters @xmath242 and @xmath243 appear in @xmath134 as product sums like @xmath244 , or @xmath245 . in our concrete model , the evaluation of detection - averaged coefficients in the wigner function involves integrals of the type : @xmath246 for example , the average @xmath247 is found by substituting both @xmath248 and @xmath249 in ( [ integral ] ) by the time - domain fourier transform of the function @xmath250 . [ here and in the following , we abbreviate products like @xmath251 by @xmath252 . ] since @xmath253 and @xmath254 are fourier transforms of real - valued functions , the corresponding integrals ( [ integral ] ) are real - valued as well . hence , all mode - reduction parameters and coefficients in the wigner functions are also real - valued . in particular , the coefficient @xmath255 in @xmath256 vanishes , see eq . ( [ dj ] ) . there is no difficulty to numerically evaluate integrals ( [ integral ] ) and we will do that below , but let us first focus on an additional approximation that can considerably simplify these results , without becoming inaccurate . _ narrow - filter approximation. _ we previously discussed the experimental observation that the spectral slit should be as narrow as possible . another simplification is possible in that case , that simply consists in neglecting in the integrals ( [ integral ] ) the frequency dependence of the mode profiles within the narrow width @xmath174 , _ i.e. _ @xmath257 for @xmath258 . let us recall that this value @xmath236 corresponds to the central frequency @xmath169 of the pulses , where the real - valued amplitudes present a maximum . this has two consequences : first , the presence of this maximum justifies a zero - order taylor approximation , provided the spectral width @xmath174 is much smaller than the spectral width of the pulses ( typically @xmath259 , where @xmath188 is the pulse duration ) ; second , if all functions involved in ( [ integral ] ) present a maximum at @xmath236 , then the narrow - filter approximation generates an upper - bound for these integrals , and therefore for quantities like @xmath260 , @xmath261 or the modal purity @xmath262 ( see [ xij],[xbar ] ) . this approximation will be applied and tested in sec . [ secnumerical ] , dedicated to the numerical results . this approximation brings the following simplification in the integrals ( [ integral ] ) : @xmath263 evidently , we end up with separate integrals over @xmath264 and @xmath265 , and using the definitions ( [ gammadelta ] ) we obtain the averages [ gammabdeltab ] @xmath266 in the narrow - filter approximation , averages of products are simply given by products of averages , @xmath267 , and @xmath268 , etc . essentially in the @xmath269 limit the filter removes any temporal information about the time the photon was emitted from the dopa . the photodetection then corresponds to a single mode with @xmath270 , regardless of the average over detection times . we therefore find for the photon - subtracted squeezed state an average wigner function of the form ( [ constantsinw ] ) , with coefficients [ abarbbar ] @xmath271 and @xmath272 and @xmath273 . using the same substitution in ( [ xij],[athetaj ] ) we can also introduce the averaged modal purity @xmath274 and the average angle @xmath275 defined by : @xmath276 _ constant profiles. _ before dealing with a more realistic case , it is interesting to focus on the case of constant profiles . let us assume a constant value for @xmath277 , @xmath223 , and @xmath278 within a space - time support of volume @xmath279 . the normalization of the homodyne mode implies @xmath280 , and equation ( [ uvxyt],[alphaeta2 ] ) leads to @xmath281 , @xmath282 , and @xmath283 . as @xmath32 vanishes , the mode @xmath34 is no more defined , and eqs . ( [ delta],[deltab ] ) can not be used anymore . in fact the mode - reduction procedure now leads to an effective _ single_-mode model rather than a two - mode model . the homodyne mode is now in the single - mode space @xmath284 spanned by @xmath33,@xmath285 . one can simply put @xmath286 in ( [ adperp_mode_red ] ) , and hence @xmath287 . this leads to the average angle @xmath288 . the modal purity becomes @xmath289 , as it should for a single - mode model . most importantly , we find @xmath290 , which according to eq . ( [ constantsinw ] ) corresponds to the most negative value for the wigner function at the origin , @xmath291 . so , with constant profiles and a narrow filter slit , we recover from our multimode model the single - mode description for photon subtraction experiments . since this limit leads to the most negative wigner function , it represents the ideal limit for producing states for qip applications , at least according to our simple multimode model . this shows that the multimode nature essentially appears through the mode distortions due to the non - constant space and time profiles of the pulses . the central role of gain - induced distortions is particularly clear with regard to the multimode nature of the squeezed vacuum produced by the dopa : assuming a constant pump field , that is assuming no gain - induced distortions , is in fact enough to obtain @xmath283 . let us now return to a more realistic model , taking into account these profiles . _ gaussian profiles. _ as a more realistic simplification , we assume gaussian profiles for the various fields . for instance , we write the homodyne field @xmath277 as @xmath292 where @xmath293 is the beam waist , @xmath188 the duration of the gaussian pulse , and @xmath294 a normalization constant . the pump beam is usually obtained by shg , in a crystal pumped by a beam identical to the homodyne beam . in the lowest order of the shg process , the profiles of @xmath223 and @xmath295 have the same shapes , so we can assume another gaussian profile : @xmath296 where one expects the pump pulse duration to be @xmath297 . however , if gvm is taken into account , this gaussian profile ( [ epxyt ] ) must be convoluted by rectangular gates . in practice , such convolutions lead to beam profiles that are still very close to gaussians . the final gaussian profile to be introduced here is the spatial mode @xmath230 of the filter in front of the apd . it is the @xmath298 mode of a monomode fiber that is well approximated by a normalized gaussian of waist @xmath299 . here our goal is twofold : first , to compare our multimode analysis with the empirical model that was used before to analyze photon subtraction experiments . second , by exploring our multi - parameter multimode model , we look for parameter regimes that are best suited for producing states with the most negative wigner functions . _ expansions in pump field. _ for our numerical work it is convenient to write all fields as taylor expansions in the pump field @xmath223 . from eq . ( [ uvxyt ] ) it follows directly that [ uuuvvv ] @xmath300 = \sum_m b_m e_p^m , \\ u\,v(x , y , t ) & = & \frac{1}{2}\sinh[2 q l e_{p } ] = \sum_m c_m e_p^m , \\ v^2(x , y , t ) & = & - \frac{1}{2 } + \frac{1}{2}\cosh[2 q l e_{p } ] = \sum_m d_m e_p^m,\end{aligned}\ ] ] with @xmath301 as in eq . ( [ epxyt ] ) . this defines the constant coefficients @xmath302 , and @xmath303 . for @xmath304 these expansions converge quite quickly . we then only have to insert relations ( [ uuuvvv ] ) into the various integrals for an efficient numerical evaluation . for instance , we can rewrite ( [ eta2 ] ) as : @xmath305 with @xmath306 in the same way we have [ numeric ] @xmath307 with [ numeric2 ] @xmath308 after fixing parameters , these expansions in the pump field can be readily used for numerical evaluations . _ fixing basic parameters. _ first we fix some parameters of our multimode model in order to present numerical results and to see how much our analysis differs from the one in ref . @xcite , where filtering before the photon detection was not modeled explicitly . we take @xmath309 and a transmission @xmath310 of the sampling beam splitter . moreover , we fix @xmath311 , which is compatible with the coupling efficiency into the filtering monomode fiber ( approximately @xmath312 , see ref . @xcite ) . regarding efficiency of homodyne detection , the mode @xmath313 considered in sec . [ secmodered ] was defined as the mode that perfectly matches the local oscillator of the homodyne detection , in other words the matching efficiency equals unity by definition in our model . the transmission of the optics and the photodetection efficiency together lead to an overall efficiency of homodyne detection @xmath176 . we put @xmath314 , in agreement with ref . @xcite . as stated above , if gvm is taken into account , the almost gaussian profile of the pump pulse is convoluted twice by a rectangular gate with time window @xmath315 . a @xmath316 crystal of length @xmath317 has @xmath318 . for an initial duration of the homodyne pulse @xmath319 , the convolutions indeed lead to a nearly gaussian beam profile with @xmath320 . we assume the identity @xmath321 in the following . _ negative wigner functions. _ as stated in the introduction , the global minimum of a wigner function is the standard figure of merit for the nonclassicality and ` non - gaussianity ' of the corresponding state . after subtraction of a single photon , the wigner function @xmath322 is always most negative in the origin ( since @xmath323 ) . figure [ w0fs ] shows how @xmath324 depends on the squeezing factor @xmath325 . the most negative values are obtained in the low - squeezing limit @xmath326 . this can be understood as there is less gain - induced distortions in that case . , as a function of squeezing parameter @xmath325 , which is varied by changing the quantity @xmath327 . the narrow - filter approximation was made for the spectral slit . fixed parameters : pulse parameters @xmath328 , @xmath329 , transmission of sampling beam splitter @xmath330 , efficiency of homodyne detection @xmath331 . ] , as a function of the spectral slit transmission for the homodyne mode . calculated for @xmath332 , using a complete evaluation of integrals ( [ integral ] ) , i.e. without the narrow - filter approximation . the narrow - filter approximation is well satisfied for low transmissions . ] in ref . @xcite , the best experimental results ( highest modal purities ) were obtained for @xmath332 . for this value of @xmath333 , which can be selected by choosing the right value for the quantity @xmath327 , we obtain @xmath334 and @xmath335 ; the latter value is close to what was observed in @xcite , without correction for the detection efficiency . at this stage , it can be interesting to compare this result , obtained using the narrow - filter approximation , with a more accurate calculation based on a complete evaluation of integrals ( [ integral ] ) . figure [ w0modelcompl ] presents the numerical results obtained for @xmath324 at @xmath332 as a function of the spectral slit transmission for the homodyne mode . ( this transmission can be increased by making the spectral slit width @xmath174 larger . ) a minimal value is clearly reached for low transmissions , justifying _ a posteriori _ the use of narrow spectral slits in the experiment of ref . @xcite . since for low transmissions , @xmath336 does not differ much from its minimal value , the narrow - filter approximation that we made in sec . [ sec : concretemultimode ] gives accurate results . , as a function of @xmath337 , where @xmath338 is the waist of the pump field and @xmath293 is the waist of the homodyne field . @xmath339 is fixed such that squeezing parameter @xmath332 . other parameter values as in fig . [ w0fs ] . ] figure [ w0fw ] predicts the behavior of @xmath324 when varying the size of the pump beam . experimental values for the widths were related by @xmath340 @xcite . [ w0fw ] clearly shows that one can await a high increase of the negativity from a larger pump beam . this result was intuitive , as there is less gain - induced distortions in that case , but is here quantified . this can motivate the use of amplified pulses @xcite , in order to have a spatially broader pump beam ( _ i.e. _ with larger @xmath341 ) , but with the same intensity . _ comparison with empirical model. _ measured negative wigner functions were interpreted in @xcite using the empirical model as introduced in sec . [ secempirical ] , where the photon is subtracted in the ` good ' mode @xmath45 with probability @xmath182 , and where the state is left in the squeezed vacuum with probability @xmath183 . as explained before , the main difference between the empirical and our models is that the conditioned state in the former corresponds to the single - photon initial state @xmath342 with @xmath343 , while it corresponds to @xmath344 in the latter , with @xmath345 . these angles @xmath194 and @xmath346 are mixing angles that fix the probability amplitudes of detection of a photon of mode 0 and of mode 1 . figure [ theta ] shows @xmath196 and @xmath197 as a function of the squeezing parameter @xmath333 . clearly , @xmath196 and @xmath197 do not differ too much , and by less than @xmath347 for @xmath332 . ( solid line ) and @xmath196 ( dashed line ) of modes 0 and 1 , as a function of squeezing parameter @xmath333 , which is varied by changing @xmath327 . other parameter values as in fig . [ w0fs ] . ] another important difference between our model and the empirical model is that modal purities @xmath348 in our model are fixed by the relation ( [ xij ] ) , whereas the parameter @xmath182 in the empirical model is a free parameter . this freedom can be used to fit the data , _ i.e. _ to have @xmath349 or @xmath350 . it is however not _ a priori _ possible to fit both parameters @xmath205 and @xmath206 ( [ abar],[bbar ] ) from ( [ abempirical ] ) using only the fitting parameter @xmath182 . in neither model should the variances @xmath50 and @xmath51 be considered as free fitting parameters of the photon - subtraction experiment , at least their values should agree with the values for @xmath110 obtained by homodyne measurements of the squeezed vacuum . in our model the mixing angles @xmath351 and their average @xmath346 take into account the filtering of the signal that is used for conditioning . in the empirical model , the corresponding angle @xmath194 is independent of the filtering . thus it is to be expected that this inaccuracy of the empirical model will lead to optimally fitted modal purities @xmath352 in the empirical model that are systematically lower than the average modal purity @xmath353 in our model . this is indeed what we find for the curves in figure [ figwigner ] : the best fit in the present example is obtained for @xmath354 , a value that is indeed smaller but still close to @xmath355 . the high quality of this fit ( with an error less than @xmath356 ) is directly linked to the fact that in the present case @xmath357 . there is a possibility to improve this result if @xmath111 is considered as a fitting parameter as well . we obtained an error of less than @xmath358 between the wigner functions for @xmath359 and @xmath360 , _ i.e. _ for a value of @xmath111 that differs by @xmath361 from the value given by the multimode model . in other words , if @xmath111 has a great influence on @xmath203 and @xmath204 in ( [ abempirical ] ) , it has a very low impact on @xmath50 , @xmath51 ; the change from @xmath334 to @xmath360 modifies the values of @xmath50 , @xmath51 by only a few @xmath362 , and for this reason it is very difficult to accurately measure @xmath111 from squeezed vacuum @xcite . these considerations explain why the empirical model can fit experimental data so successfully ; even when @xmath196 is not equal to @xmath275 , the parameter @xmath111 gives a supplementary freedom for fitting . and @xmath363 of the wigner function @xmath322 ( solid line ) , and corresponding best fits using the empirical model ( dashed line ) with @xmath364 . parameter values as in fig . results of our model and the empirical fits almost overlap . ] we have introduced a straightforward and physically intuitive procedure that we call ` mode reduction ' to simplify the multimode description of squeezed light to the bare essentials . for photon - subtraction experiments , this means that the homodyne signal is reduced to an effective two - mode description and the detector signal requires one extra orthogonal effective mode . we derived the wigner function of the homodyne signal conditional upon the detection of a single photon , and we also showed how to average over possible measurement outcomes . the general mode - reduction formalism was then applied to a detailed model describing photon subtraction of gaussian spatiotemporal pulses of squeezed light . this model features many experimental parameters such as beam waists and duration of the pulses that can be independently measured . indeed , our model does not have free fitting parameters . this allows one to study in detail what are the crucial experimental parameters to produce optimally negative wigner functions with pulses of squeezed light . we compared our new model to the empirical model that was used before to analyze photon - subtraction experiments in @xcite . in fact , the formulae for the output wigner functions look similar . one crucial difference is that the empirical model does have a free parameter , namely the quantity called the modal purity . in our model modal purities also occur , be it with a slightly different meaning , but they are fixed quantities . a good agreement between our model and experiments therefore gives more understanding than an accurate fit with the empirical model . we found that in the range of parameters of the measurements in @xcite , both our model and the empirical model are accurate . we reasoned that modal purities in our model would be systematically higher , and in our numerical example we found this to be the case . the accuracy of the empirical model strongly depends on the availability of the free parameter . it was nevertheless a surprise in the theoretical analysis that the mixing angles @xmath194 and @xmath346 , describing the relative probability of measuring a photon in either one of two effective modes , differ at most @xmath365 in a whole range of squeezing parameters . our mode - reduction procedure is closely related to the analysis of photon - subtraction experiments of refs . one could express our mode - reduction parameters in terms of elements of the covariance matrix of refs . . our output wigner function in eq . ( [ wj ] ) then reduces to the one in ref . @xcite , but only in the special case that all our mode reduction parameters are real - valued so that @xmath366 in eq . ( [ dj ] ) vanishes . this we assumed for simplicity in sec . [ sec : concretemultimode ] . our mode - reduction procedure is carried out in the heisenberg picture . we think that our approach has some advantages . in our approach it becomes quite intuitive in what sense it goes beyond the empirical model of ref . @xcite . in our concrete multimode analysis , we include effects not considered in ref . @xcite , such as the transverse beam profile , for which we found that wider pump beams lead to more negative wigner functions . in conclusion , we presented a very concise model that can account for the multimode nature of projective photon - counting measurements . it gives an intuitive picture of photon - subtraction experiments , close to the empirical model previously published . this multimode model therefore gives consistent results , in agreement with previously published experiments where pulses of light with negative wigner functions were produced conditionally . our model can be used to predict the changes in the output upon variation of experimentally relevant parameters , and to optimize the setup design . we thank k. mlmer for useful discussions . this work has been supported by the danish research council through quantop , by compas , and by the niels bohr international academy . in order to derive eq . ( [ wj ] ) from eq . ( [ winitial ] ) , one has first to invert eqs . ( [ modereductionahout],[a1out ] ) , leading to @xmath367 one should then replace in eq.([winitial ] ) @xmath100 by @xmath368 , and @xmath102 , and integrate over @xmath369 , @xmath102 . it is however convenient to make a change of variables so that the integral is over @xmath370 , @xmath371 instead of @xmath369 , @xmath102 . in this case the only transforms needed for this calculation is [ rel_x0p0 ] @xmath372 as well as the transformation of the integral @xmath373 one should then note that the wigner function ( [ winitial ] ) is the product of a polynomial in @xmath43 , @xmath44 , and of a gaussian term @xmath374 , with @xmath375 with eq . ( [ rel_x0p0 ] ) , the exponent can be rewritten as @xmath376 the integral ( [ x1p1integral ] ) with eq . ( [ winitial ] ) as its integrand can then be found by replacing in the integrand the squares @xmath377 and @xmath378 by @xmath379 and by replacing the first - order terms according to @xmath380 the goal of this appendix is the derivation of the local bogoliubov transformation ( [ bogoliubov2 ] ) . we assume that inside the dopa the pump pulse with an angular frequency @xmath235 travels at a speed @xmath381 , with negligible absorption . this field can therefore be written as @xmath382 where @xmath383 is an arbitrary time delay and where ` @xmath384 ' is a purely conventional phase factor . let us write the probe beam as @xmath385 where the phase - matching condition @xmath386 is assumed to be satisfied . by using the svea in maxwell s equations , neglecting diffraction terms and considering the first - order dispersion , we obtain @xmath387 where @xmath388 . the substitution of @xmath389 by @xmath10 then leads to @xmath390 where @xmath391 is the gvm . with the assumption that @xmath223 is real - valued , the solution to eq . ( [ svaaeq ] ) becomes @xmath392\,a_{\rm in}+\sinh[\,q l f_p \,]\,a^*_{\rm in},\ ] ] where the @xmath215-dependence was suppressed . the effective pump field @xmath393 is given by @xmath394 with @xmath395 the time separation induced by the gvm after crossing the crystal . ( [ fpintegral ] ) shows that the effective pump field @xmath393 is a convolution of the pump field @xmath223 with a rectangular unit gate of duration @xmath224 , which for @xmath396 is centered around the origin . as the quantized version of eq . ( [ aoutsolved ] ) , we then find eq . ( [ bogoliubov2 ] ) of the main text . the pump field @xmath397 of the main text is to be understood as the effective pump field @xmath393 derived here . evidently , @xmath398 in the limit @xmath399 ( no gvm ) . j. eisert , s. scheel , and m.b . plenio , phys . 89 * , 137903 ( 2002 ) . g. giedke and j.i . cirac , phys . a * 66 * , 032316 ( 2002 ) . browne , j. eisert , s. scheel , and m.b . plenio , phys . a * 67 * , 062320 ( 2003 ) . kim , e. park , p.l . knight , and h. jeong , phys . a * 71 * , 043805 ( 2005 ) . a. biswas and g.s . agarwal , phys . a * 75 * , 032104 ( 2007 ) . m. dakna , t. anhut , t. opatrn ' y , l. kn " oll , and + d .- g . welsch , phys . a * 55 * , 3184 ( 1997 ) . t. opatrn ' y , g. kurizki , and d .- welsch , phys . a * 61 * , 032302 ( 2000 ) . lvovsky , h. hansen , t. aichele , o. benson , j. mlynek , and s. schiller , phys . lett . * 87 * , 050402 ( 2001 ) . j. wenger , r. tualle - brouri and p. grangier , phys . lett . * 92 * , 153601 ( 2004 ) . a. ourjoumtsev , r. tualle - brouri , j. laurat , and ph . grangier , science * 312 * , 83 ( 2006 ) , and supporting online material . a. ourjoumtsev , r. tualle - brouri , and ph . grangier , phys . lett . * 96 * , 213601 ( 2006 ) . k. wakui , h. takahashi , a. furusawa and m. sasaki , e - print quant - ph/0609153v1 . neergaard - nielsen , b. nielsen , c. hettich , k. mlmer , and e.s . polzik , phys . lett . * 97 * , 083604 ( 2006 ) . a. ourjoumtsev , h. jeong , r. tualle - brouri , p. grangier , nature * 448 * , 784 ( 2007 ) . v. parigi , a. zavatta , m. kim , and m. bellini , science * 317 * , 1890 ( 2007 ) . kim , j. phys . b : at . mol . . phys . * 41 * , 133001 ( 2008 ) . m. avenhaus , h. b. coldenstrodt - ronge , k. laiho , w. mauerer , i. a. walmsley and c. silberhorn , phys . * 101 * , 053601 ( 2008 ) . m. sasaki and s. suzuki , phys . a * 73 * , 043807 ( 2006 ) . k. mlmer , e - print quant - ph/0602202v1 . nielsen and k. mlmer , phys . a * 75 * , 023806 ( 2007 ) . nielsen and k. mlmer , phys . a * 76 * , 033832 ( 2007 ) . f. grosshans and ph . grangier , phys . lett . * 88 * , 057902 ( 2002 ) . t. aichele , a.i . lvovsky and s. schiller , eur . j. d * 18 * , 237 ( 2002 ) . a. la porta and r. e. slusher , phys . a * 44 * , 2013 ( 1991 ) . gardiner and p. zoller , _ quantum noise _ ( springer , berlin , 2000 ) . j. wenger , j. fiurek , r. tualle - brouri , n.j . cerf , and p. grangier , phys . a * 70 * , 053812 ( 2004 ) . u. leonhardt , _ measuring the quantum state of light _ ( cambridge university press , 1997 ) .

we study classes of dynamic programming ( dp ) algorithms which , due to their algebraic definitions , are closely related to coefficient extraction methods . dp algorithms can easily be modified to exploit sparseness in the dp table through memorization . coefficient extraction techniques on the other hand are both space - efficient and parallelisable , but no tools have been available to exploit sparseness . we investigate the systematic use of homomorphic hash functions to combine the best of these methods and obtain improved space - efficient algorithms for problems including linear sat , set partition , and subset sum . our algorithms run in time proportional to the number of nonzero entries of the last segment of the dp table , which presents a strict improvement over sparse dp . the last property also gives an improved algorithm for cnf sat with sparse projections .

1203.4063

_ coefficient extraction _ can be seen as a general method for designing algorithms , recently in particular in the area of exact algorithms for various np - hard problems @xcite ( cf . @xcite for an introduction to exact algorithms ) . the approach of the method is the following ( see also @xcite ) : 1 . define a variable ( the so - called coefficient ) whose value ( almost ) immediately gives the solution of the problem to be solved , 2 . show that the variable can be expressed by a relatively small formula or circuit over a ( cleverly chosen ) large algebraic object like a ring or field , 3 . show how to perform operations in the algebraic object relatively efficiently . in a typical application of the method , the first two steps are derived from an existing dynamic programming ( dp ) algorithm , and the third step deploys a carefully selected algebraic isomorphism , such as the discrete fourier transform to extract the desired solution / coefficient . algorithms based on coefficient extraction have two key advantages over dp algorithms ; namely , they are space - efficient and they parallelise well ( see , for example , @xcite ) . yet , dp has an advantage if the problem instance is _ sparse_. by this we mean that the number of candidate / partial solutions that need to be considered during dp is small , that is , most entries in the dp table are not used at all . in such a case we can readily adjust the dp algorithm to take this into account through _ memorization _ so that both the running time and space usage become proportional to the number of partial solutions considered . unfortunately , it is difficult to parallelise or lower the space usage of memorization . coefficient extraction algorithms relying on interpolation of sparse polynomials @xcite improve over memorization by scaling proportionally only to the number of _ candidate _ solutions , but their space usage is still not satisfactory ( see also @xcite ) . this paper aims at obtaining what is essentially the best of both worlds , by investigating the systematic use of homomorphisms to `` hash down '' circuit - based coefficient extraction algorithms so that the domain of coefficient extraction and hence the running time matches or improves that of memorization - based dp algorithms , while providing space - efficiency and efficient parallelisation . the key idea is to take an existing algebraic circuit for coefficient extraction ( over a sparsely populated algebraic domain such as a ring or field ) , and transform the circuit into a circuit over a smaller domain by a homomorphic hash function , and only then perform the actual coefficient extraction . because the function is homomorphic , by hashing the values at the input gates and evaluating the circuit , the output evaluates to the hash of the original output value . because the function is a hash function , the coefficient to be extracted collides with other coefficients only with negligible probability in the smaller domain , and coefficient extraction can be successfully used on the new ( hashed - down ) circuit . we call this approach _ homomorphic hashing_. we study sparse dp / coefficient extraction in three domains : ( a ) the univariate polynomial ring @xmath0 $ ] in section [ sec : sss ] , ( b ) the group algebra @xmath1 $ ] in section [ sec : setpart ] and ( c ) the mbius algebra of the subset lattice in section [ sec : unionhashing ] . the subject of sparse dp or coefficient extraction is highly motivated and well - studied @xcite . in @xcite , a sparse polynomial interpolation algorithm using exponential space was already given for ( a ) and ( b ) ; our algorithms improve these domains to polynomial space . in @xcite a polynomial - space algorithm for finding a small multilinear monomial in @xmath1 $ ] was given . in @xcite a general study of settings ( a ) and ( b ) was initiated , but sparsity was not addressed . our main technical contribution occurs with ( c ) and hashing down to the `` solomon algebra '' of a poset . our methods work for general arithmetic circuits similarly as in @xcite , and most of our algorithms work for counting variants as well . but , for concreteness , we will work here with specific decision problems . although we mainly give improvements for sparse variants of these problems , we feel the results will be useful to deal with the general case as well ( as we will see in section [ sec : setpart ] ) . [ [ subset - sum ] ] subset sum + + + + + + + + + + the problem is the following : given a vector @xmath2 and integer @xmath3 , determine whether there exists a subset @xmath4 $ ] such that @xmath5 . it is known to be solvable @xmath6 time and @xmath7 space @xcite , and solving it faster , or even in @xmath8 time and polynomial space are interesting open questions @xcite . recently , a polynomial space algorithm using @xmath9 time was given in @xcite . [ thm : mainsss ] an instance @xmath10 of the problem can be solved 1 . in @xmath11 expected time and polynomial space , and 2 . in @xmath12 time and polynomial space , where @xmath13 is the number of distinct sums , i.e. @xmath14\}|$ ] . here it should be noted that standard sparse dp gives an @xmath11 time and space algorithm . informally stated , our algorithms hash the instances by working modulo randomly chosen prime numbers and applying the algoritm of @xcite . while interesting on their own , these results may be useful in resolving the above open questions when combined with other techniques . [ [ linear - sat ] ] linear sat + + + + + + + + + + the linear sat problem is defined as follows : given a matrix @xmath15 , vectors @xmath16 and @xmath17 , and an integer @xmath18 , determine whether there is there a vector @xmath19 such that @xmath20 and @xmath21 . variants of linear sat have been studied , perhaps most notably in @xcite , where approximability was studied . in @xcite the fixed parameter tractability was studied for parameterizations of various above guarantees . there , it was also quoted from @xcite that ( a variant of ) linear sat is `` as basic as satisfiability '' . it can be observed that using the approach from @xcite , linear sat can be solved in @xmath22 time and @xmath22 space . also , using standard `` sparse dynamic programming '' , it can be solved in @xmath23 time and @xmath23 space , where @xmath24 is the rank of @xmath25 . we obtain the following polynomial - space variants : [ thm : mainls ] an instance @xmath26 of can be solved by algorithms with constant one - sided error probability in 1 . @xmath23 time and polynomial space , and 2 . @xmath27 time and polynomial space . the first algorithm hashes the input down using a random linear map and afterwards determines the answer using the walsh - hadamard transform . the second algorithm uses a win / win approach , combining the first algorithm with the fact that an @xmath25 with high rank can be solved with a complementary algorithm . [ [ satisfiability ] ] satisfiability + + + + + + + + + + + + + + the cnf - sat problem is defined as follows : given a cnf - formula @xmath28 over @xmath29 variables , determine whether @xmath30 is satisfiable . there are many interesting open questions related to this problem , a major one being whether it can be solved in time @xmath31 ( the ` strong exponential time hypothesis ' @xcite states this is not possible ) , and another being whether the number of satisfying assignments can be counted in time @xmath31 for some @xmath32 ( e.g. @xcite ) . a _ prefix assignment _ is an assignment of 0/1 values to the variables @xmath33 for some @xmath34 . a _ projection _ ( _ prefix projection _ ) of a cnf - formula is a subset @xmath35 $ ] such that there exists an assignment ( prefix assignment ) of the variables such that for every @xmath36 it satisfies @xmath37 if and only if @xmath38 . an algorithm for cnf - sat running in time linear in the number of _ prefix _ projections can be obtained by standard sparse dp . however , it is sensible to ask about complexity of cnf - sat if the number of projections is small . we give a positive answer : [ thm : maincnf ] satisfiability of a formula @xmath39 can be determined in @xmath40 time and @xmath41 space , where @xmath42 : \pi \text { is a projection of } \phi \}|$ ] . we are not aware of previous work that studies instances with few projections . although most instances will have many projections , we think our result opens up a fresh technical perspective that may contribute towards solving the above mentioned and related questions . underlying theorem [ thm : maincnf ] is our main technical contribution ( theorem [ thm : unionhashsmallsupp ] ) that enables us to circumvent partial projections and access projections directly , namely homomorphic hashing from the mbius algebra of the lattice of subsets of @xmath43 $ ] to the solomon algebra of a poset . a full proof of theorem [ thm : unionhashsmallsupp ] is given in the appendix ; we give a specialized , more direct proof of theorem [ thm : maincnf ] in section 5 . the proofs of claims marked with a `` @xmath44 '' are relegated to the appendix in order to not break the flow of the paper . lower - case boldface characters refer to vectors , while capital boldface letters refer to matrices , @xmath45 being the identity matrix . the rank of a matrix @xmath25 is denoted by @xmath24 . if @xmath46 and @xmath13 are sets , and @xmath13 is finite , denote by @xmath47 the set of all @xmath48-dimensional vectors with values in @xmath46 , indexed by elements of @xmath13 , that is , if @xmath49 , then for every @xmath50 we have @xmath51 . we denote by @xmath52 and @xmath53 the set of integers and non - negative integers , respectively , and by @xmath54 the field of integers modulo a prime @xmath55 . an arbitrary field is denoted by @xmath56 . for a logical proposition @xmath57 , we use iverson s bracket notation @xmath58 $ ] to denote a @xmath59 if @xmath57 is true and a @xmath60 if @xmath57 is false . for a function @xmath61 and @xmath62 , the preimage @xmath63 is defined as the set @xmath64 . for an integer @xmath29 and @xmath65 , denote by @xmath66 the characteristic vector of @xmath67 . sometimes we will state running times of algorithms with the @xmath68 notation , which suppresses any factor polynomial in the input size . for a ring @xmath46 and a finite set @xmath13 , we write @xmath47 for the ring consisting of the set @xmath47 ( the set of all vectors over @xmath46 with coordinates indexed by elements of @xmath13 ) equipped with coordinate - wise addition @xmath69 and multiplication @xmath70 ( the _ hadamard product _ ) , that is , for @xmath71 and @xmath72 , @xmath73 we set @xmath74 and @xmath75 for each @xmath76 , where @xmath69 and the juxtaposition denote addition and multiplication in @xmath46 , respectively . the inner - product @xmath71 is denoted by @xmath77 . for @xmath78 denote by @xmath79 the _ support _ of @xmath80 , that is , @xmath81 , where @xmath60 is the additive identity element of @xmath46 . a vector @xmath80 is called a _ singleton _ if @xmath82 . we denote by @xmath83 the singleton with value @xmath84 on index @xmath85 , that is , @xmath86 $ ] for all @xmath87 . if @xmath46 is a ring and @xmath88 is a finite semigroup , denote by @xmath89 $ ] the ring consisting of the set @xmath47 equipped with coordinate - wise addition and multiplication defined by the convolution operator @xmath90 , where for @xmath71 , @xmath91 we set @xmath92 for every @xmath76 . if @xmath93 are rings with operations @xmath94 and @xmath95 respectively , a _ homomorphism from @xmath46 to @xmath13 _ is a function @xmath96 such that @xmath97 and @xmath98 for every @xmath99 . [ obs : hom ] let @xmath46 be a ring , and let @xmath88 and @xmath100 be finite semigroups . suppose @xmath101 such that for every @xmath102 we have @xmath103 . then the function @xmath104 \rightarrow r[t]$ ] defined by @xmath105 where @xmath106 for all @xmath107 is a homomorphism . a _ circuit _ @xmath108 over a ring @xmath46 is a labeled directed acyclic graph @xmath109 where the elements of @xmath110 are called _ gates _ and @xmath111 has a unique sink called the _ output gate of @xmath108_. all sources of @xmath108 are called _ input gates _ and are labeled with elements from @xmath46 . all gates with non - zero in - degree are labeled as either an _ addition _ or a _ multiplication gate_. ( if multiplication in @xmath46 is not commutative , the in - arcs of each multiplication gate are also ordered . ) every gate @xmath112 of @xmath108 can be associated with a ring element in the following natural way : if @xmath112 is an input gate , we associate the label of @xmath112 with @xmath112 . if @xmath112 is an addition gate we associate the ring element @xmath113 with @xmath112 , and if @xmath112 is a multiplication gate we associate the ring element @xmath114 with @xmath112 where @xmath115 are the ring elements associated with the @xmath116 in - neighbors of @xmath112 , and @xmath69 and @xmath90 are the operations of the ring @xmath46 . suppose the ground set of @xmath46 is of the type @xmath117 where @xmath118 are sets . then @xmath108 is said to have _ singleton inputs _ if the label of every input - gate of @xmath108 is a singleton vector of @xmath46 . let @xmath46 and @xmath13 be rings , let @xmath119 be a homomorphism , and suppose that @xmath108 is a circuit over @xmath46 . then , the circuit @xmath120 over @xmath13 _ obtained by applying @xmath121 to @xmath108 _ is defined as the circuit obtained from @xmath108 by replacing for every input gate the label @xmath122 by @xmath123 . note that the following is immediate from the definition of a homomorphism : [ obs : homcir ] suppose @xmath108 is a circuit over a ring @xmath46 with output @xmath124 . then the circuit over @xmath13 obtained by applying a homomorphism @xmath119 to @xmath108 outputs @xmath125 . in this section we will study the problem and prove theorem [ thm : mainsss ] . as mentioned in the introduction , it should be noted that this merely serves as an illustration of how similar problems can be tackled as well since the same method applies to the more general sparse polynomial interpolation problem . however , to avoid a repeat of the analysis of @xcite , we have chosen to restrict ourselves to the problem . our central contribution over @xcite is that we take advantage of sparsity . given @xmath126 and an integer @xmath127 , let @xmath128 be defined by @xmath129 : \sum_{e \in x } a_e \equiv j\ ( { \operatorname{mod}}{p } ) \bigg\}\bigg|\text { \ for every } j\in{\mathbb{z}}_p.\ ] ] we also use the shorthand @xmath130 . we use a corollary from @xcite : [ cor : ln ] given an instance @xmath10 of and an integer @xmath131 , @xmath132 can be computed in @xmath133 time and @xmath134 space . we will also need the following two results on primes : [ thm : pnt ] if @xmath135 , then the number of prime numbers at most @xmath136 is at least @xmath137 . [ thm : sam ] there exists an algorithm @xmath138 running in @xmath139 time that , given integer @xmath140 as input , outputs either a prime chosen uniformly at random from the set of primes at most @xmath136 or @xmath141 . moreover , the probability that the output is @xmath141 is at most @xmath142 . we will run a data reduction procedure similar to the one of claim 2.7 in @xcite , before applying the algorithm of corollary [ cor : ln ] . the idea of the data reduction procedure is to work modulo a prime of size roughly @xmath143 or larger : [ lem : hashsss ] let @xmath144 and let @xmath145 be an upper bound on the number of bits needed to represent the integers , i.e. @xmath146 . then , @xmath147\geq \frac{1}{2}$ ] , where the probability is taken uniformly over all primes @xmath148 . suppose @xmath149 . then there exists an integer @xmath150 such that @xmath151 and @xmath152 . this implies that @xmath131 is a divisor of @xmath153 , so let us bound the probability of this event . since @xmath154 , it has at most @xmath155 distinct prime divisors . let @xmath156 . it is easy to check that for sufficiently large @xmath145 and @xmath29 : @xmath157 \leq \frac { \beta + \log n}{\frac{\gamma}{\log \gamma + 2}}\leq \frac { \beta + \log n}{\frac{\gamma}{3(n + \log \beta ) } } \leq \frac{1}{2{s}},\ ] ] where we use theorem [ thm : pnt ] in the first inequality and @xmath158 in the second inequality . applying the union bound over the at most @xmath159 elements of @xmath160 , the event that there exists a @xmath150 with @xmath151 and @xmath161 has probability at most @xmath162 . now we give two algorithms utilizing homomorphic hashing , one for the case where @xmath159 is known , and one for the case where @xmath159 is not known . there exists an algorithm that , given an instance @xmath10 of the problem and an integer @xmath144 as input , outputs a nonnegative integer @xmath163 in @xmath164 time and polynomial space such that ( i ) @xmath165 implies @xmath166 and ( ii ) @xmath167 \geq \frac{1}{4}$ ] . the algorithm is : first , obtain prime @xmath168 using lemma [ thm : sam ] . second , compute and output @xmath132 using corollary [ cor : ln ] . condition ( i ) holds since @xmath169 implies @xmath166 for any @xmath170 . moreover , condition ( ii ) follows from lemma [ lem : hashsss ] and lemma [ thm : sam ] since @xmath171 . the time and space bounds are met by corollary @xmath172 because @xmath173 . the algorithm is the following : maintain a guess @xmath159 of @xmath160 , initially set to @xmath29 . obtain a prime @xmath168 using lemma [ thm : sam ] , and compute @xmath132 using corollary [ cor : ln ] . if @xmath169 output * no * since @xmath166 ; otherwise , attempt to construct a subset @xmath4 $ ] such that @xmath5 using self - reduction . if this succeeds , return * yes*. otherwise , double @xmath159 and repeat . the expected running time ( taken over all primes @xmath55 ) is @xmath174 because when @xmath175 the probability of succesfully constructing a solution or concluding that none exist is at least @xmath162 by the arguments in the proof of lemma [ lem : hashsss ] . the derandomization for theorem [ thm : mainsss](b ) is given in the appendix . in this section we assume that @xmath56 is a field of non - even characteristic and that addition and multiplication refer to operations in @xmath56 . for @xmath176 , we denote by @xmath177 the _ walsh - hadamard matrix _ , defined for all @xmath178 by @xmath179 . [ lem : wht ] the walsh - hadamard matrix satisfies @xmath180 and , for every @xmath181 $ ] , it holds that @xmath182 . we first prove the following general theorem , of which theorem [ thm : mainls](a ) is a special case . [ thm : z2 ] there exists a randomized algorithm that , given as input 1 . a circuit @xmath108 with singleton inputs over @xmath1 $ ] , 2 . an integer @xmath183 , and 3 . an element @xmath184 , outputs the coefficient @xmath185 with probability at least @xmath162 , where @xmath186 $ ] is the output of @xmath108 . the algorithm runs in time @xmath164 and uses @xmath164 arithmetic operations in @xmath56 , and requires storage for @xmath134 bits and elements of @xmath56 . @xmath187 let @xmath188 + 1 . choose a matrix @xmath189 uniformly at random from the set of all @xmath190 matrices with binary entries . let @xmath191 \rightarrow { { \mathbb{f}}}[{\mathbb{z}}_2^s]$ ] be the homomorphism defined by @xmath192 where @xmath193 for all @xmath194 . apply @xmath121 to @xmath108 to obtain the circuit @xmath195 . @xmath197 let @xmath198\rightarrow{{\mathbb{f}}}$ ] be the homomorphism defined by @xmath199 for all @xmath200 $ ] . apply @xmath201 to @xmath195 to obtain the circuit @xmath202 . evaluate @xmath202 and return the output . the algorithm is given in algorithm [ alg : find1 ] . let us first analyse the complexity of this algorithm : steps 1 and 2 can be performed in time polynomial in the input . step 3 also be done in polynomial time since it amounts to relabeling all input gates with @xmath203 where @xmath204 was the old label . indeed , we know that @xmath205 $ ] is a singleton @xmath206 , so @xmath203 is the singleton @xmath207 and this can be computed in polynomial time . step 4 takes @xmath164 operations and calls to @xmath208 , so for the complexity bound it remains to show that each call to @xmath208 runs in polynomial time . step 5 can be implemented in polynomial time similar to step 3 since the singleton @xmath209 is mapped to @xmath210 . finally , the direct evaluation of @xmath202 uses @xmath211 operations in @xmath56 . hence the algorithm meets the time bound , and also the space bound is immediate . the fact that @xmath187 returns @xmath212 with probability at least @xmath162 is a direct consequence of the following two claims . let @xmath213 be the output of @xmath195 . [ clm : one ] @xmath214 \geq \frac{1}{2}$ ] . [ clm : two ] algorithm @xmath187 returns @xmath215 . for @xmath216 and @xmath217 denote by @xmath218 the @xmath219th row of @xmath25 and define @xmath220 \in \mathbb{q}[{\mathbb{z}}_2^m]$ ] by @xmath221}= \begin{cases } { \left\langle 1,{\boldsymbol 0 } \right\rangle } & \text{if } i = w=0,\\ 0 & \text{if } i=0 \text { and } w\neq 0,\\ { \boldsymbol f}[i-1,w ] + { \boldsymbol f}[i-1,j-\omega_i ] * { \left\langle 1,{\boldsymbol a^{(i ) } } \right\rangle } & \text{otherwise . } \end{cases}\ ] ] it is easy to see that for every @xmath222 , @xmath223 , and @xmath224 , the value @xmath225_{{\boldsymbol y}}$ ] is the number of @xmath226 such that @xmath227 and @xmath228 where @xmath229 and @xmath230 are obtained by truncating @xmath231 and @xmath25 to the first @xmath219 rows . hence , we let @xmath108 be the circuit implementing [ eq : setpart ] and let its output be @xmath232}$ ] . thus , @xmath233 is the number of @xmath19 with @xmath20 and @xmath234 . also , @xmath235 since any element of the support of @xmath80 is a sum of rows of @xmath25 and hence in the row - space of @xmath25 , which has size at most @xmath236 . to apply theorem [ thm : z2 ] , let @xmath237 and observe that the computations are in fact carried out over integers bounded in absolute value poly - exponentially in @xmath29 and hence the operations in the base field can also be executed polynomial in @xmath29 . the theorem follows from theorem [ thm : z2 ] . to establish theorem [ thm : mainls](b ) , let us first see how to exploit a high linear rank of the matrix @xmath25 in an instance of linear sat . by permuting the rows of @xmath25 as necessary , we can assume that the first @xmath24 rows of @xmath25 are linearly independent . we can now partition @xmath238 into @xmath239 , where @xmath240 has length @xmath24 and @xmath241 has length @xmath242 . there are @xmath243 choices for @xmath241 , each of which by linear independence has at most one corresponding @xmath240 such that @xmath20 . given @xmath241 , we can determine the corresponding @xmath240 ( if any ) in polynomial time by gaussian elimination . thus , we have : [ obs : highrank ] linear sat can be solved in @xmath244 time and polynomial space . this enables a `` win / win approach '' where we distinguish between low and high ranks , and use an appropriate algorithm in each case . compute @xmath24 . if @xmath245 , run the algorithm of observation [ obs : highrank ] otherwise , run the algorithm implied by theorem [ thm : mainls](a ) . we now give a very similar application to the set partition problem : given an integer @xmath3 and a set family @xmath246 where @xmath247 , @xmath248 , determine whether there is a subfamily @xmath249 with @xmath250 such that @xmath251 and @xmath252 . the _ incidence matrix _ of a set system @xmath253 is the @xmath254 matrix @xmath25 whose entries @xmath255 $ ] are indexed by @xmath256 and @xmath257 . [ thm : setpartexpspacerank][thm : setpart5 ] there exist algorithms that given an instance @xmath258 of set partition output the number of set partitions of size at most @xmath3 with probability at least @xmath162 , and use ( a ) @xmath23 time and polynomial space , and ( b ) @xmath259 time and space , where @xmath25 is the incidence matrix of @xmath253 . in this section our objective is to mimic the approach of the previous section for @xmath260 $ ] , where @xmath261 is the semigroup defined by the set union @xmath262 operation on @xmath263 , the power set of an @xmath29-element set @xmath264 . the direct attempt to apply a homomorphic hashing function , unfortunately , fails . indeed , let @xmath121 be an arbitrary homomorphism from @xmath261 to @xmath265 with @xmath266 . let @xmath267 and consider the minimum value @xmath268 with @xmath269 ; in particular , for @xmath270 we have @xmath271 , which signals failure since we can not isolate @xmath272 from @xmath264 . instead , we use hashing to an algebraic structure based on a poset ( the `` solomon algebra '' of a poset due to @xcite ) that is obtained by the technique `` iterative compression '' . this gives the following main result . for reasons of space we relegate a detailed proof to the appendix ; here we will give a simplified version of the proof in the special case of theorem [ thm : maincnf ] in this section . [ thm : unionhashsmallsupp ] let and @xmath273 . there are algorithms that , given a circuit @xmath108 with singleton inputs in @xmath260 $ ] outputting @xmath80 , compute 1 . a list with @xmath274 for every @xmath275 in @xmath276 time , 2 . @xmath277 in time @xmath278 if @xmath279 such that @xmath280 . the above result is stated for simplicity in the unit - cost model , that is , we assume that arithmetic operations on integers take constant time . for the more realistic log - cost model , where such operations are assumed to take time polynomial in the number of bits of the binary representation , we only mention here that our results also hold under some mild technical conditions . let us first show that theorem [ thm : maincnf](a ) indeed is a special case of theorem [ thm : unionhashsmallsupp ] : use a circuit over @xmath281},\cup)]$ ] that implements the expression @xmath282 where @xmath283 $ ] ( respectively , @xmath284 $ ] ) is the set of all indices of clauses that contain a positive ( respectively , negative ) literal of the variable @xmath285 . then use theorem [ thm : unionhashsmallsupp ] to determine @xmath286}$ ] , the number of satisfying assignments of @xmath30 . now we proceed with a self - contained proof theorem [ thm : maincnf ] . given poset @xmath287 , the _ mbius function _ @xmath288 of @xmath57 is defined for all @xmath289 by @xmath290 the _ zeta transform _ @xmath291 and _ mbius transform _ @xmath292 are the @xmath293 matrices defined by @xmath294 $ ] and @xmath295 for all @xmath289 . for a cnf - formula @xmath30 denote @xmath296 for the set of all projections of @xmath30 . recall in theorem [ thm : maincnf ] we are given a cnf - formula @xmath297 over @xmath29 variables . for @xmath298 define @xmath299 . then we have the following easy observations 1 . @xmath300 , 2 . @xmath301 for every @xmath298 , 3 . @xmath302 for every @xmath298 . given the above lemma and observations , we will give an algoritm using a technique called _ iterative compression _ @xcite . as we will see , by this technique it is sufficient to solve the following `` compression problem '' : [ lem : compresssimple ] given a cnf - formula @xmath303 and a set family @xmath304}$ ] with @xmath305 , the set @xmath296 can be constructed in @xmath306 time . in what follows @xmath307 refers to an assignment of values to the @xmath29 variables in @xmath30 . define @xmath308}}$ ] for all @xmath309 $ ] by @xmath310 \text { it holds that } { \boldsymbol a } \text { satisfies } c_i \text { iff } i \in x \}|.\ ] ] it is easy to see that @xmath311 , so if we know @xmath312 for every @xmath313 we can construct @xmath296 in @xmath314 time . towards this end , first note that for every @xmath315 $ ] it holds that @xmath316 \text { it holds that } { \boldsymbol a } \text { satisfies } c_i \text { only if } i \in y \}|.\end{aligned}\ ] ] second , note that the last quantity can be computed in polynomial time : since every clause outside @xmath317 must not be satisfied , each such clause forces the variables that occur in it to unique values ; any other variables may be assigned to arbitrary values . that is , the count is 0 if the clauses outside @xmath317 force at least one variable to conflicting values , otherwise the count is @xmath318 where @xmath319 is the number of variables that occur in none of the clauses outside @xmath317 . now the algorithm is the following : for every @xmath313 compute @xmath320 in polynomial time as discussed above . then we can use algorithm @xmath321 as described below to obtain @xmath312 for every @xmath313 since it follows that @xmath322 from the definition of @xmath292 and the fact that @xmath323 . algorithm @xmath321 clearly runs in @xmath324 time , so this procedure meets the claimed time bound . let @xmath326 such that @xmath327 implies @xmath328 . @xmath330 @xmath241 . recall that we already know that @xmath331 . now , for @xmath332 we set @xmath333 and use @xmath334 to obtain @xmath335 using lemma [ lem : compresssimple ] . in the end we are given @xmath336 and since @xmath337 is exactly the original formula , the input is a yes - instance if and only if @xmath43 \in { \mathrm{supp}}(\phi_m)$ ] . the claimed running time follows from observations 1 and 3 above and the running time of algorithm @xmath321 . we will now give an application of theorem [ thm : unionhashsmallsupp](b ) to set cover : given a set family @xmath338 where @xmath273 and an integer @xmath339 , find a subfamily @xmath340 such that @xmath341 and @xmath342 . given an instance of set cover , let @xmath279 be the largest real such that @xmath343 . then the instance can be solved in @xmath344 time ( and exponential space ) . let @xmath345 , and for every @xmath346 and @xmath347 define @xmath348}$ ] as follows : @xmath349}= \begin{cases } { \left\langle 1,\emptyset \right\rangle } & \text{if } i = j=0,\\ { \boldsymbol f}[i-1,j ] + { \boldsymbol f}[i-1,j-1 ] * { \left\langle 1,s_i \right\rangle } & \text{otherwise . } \end{cases}\ ] ] it is easy to see that for every @xmath350 we have that @xmath351_x$ ] is the number of @xmath352 such that @xmath353 and @xmath354 . hence the theorem follows directly by applying theorem [ thm : unionhashsmallsupp](b ) by interpreting the above recurrence as a circuit in order to determine the value @xmath355_u$ ] . * appendix * by reducing modulo @xmath55 we can assume that @xmath357 . we have @xmath358 where the first equality holds by definition and the second equality follows from the fact that @xmath359 when @xmath360 or @xmath361 . the latter expression can be evaluated in the claimed resource bounds using theorem [ thm : ln ] . take an integer @xmath362 uniformly at random , and check whether it is a prime using the polynomial time algorithm of @xcite . if @xmath219 is prime , then output @xmath219 and halt ; otherwise repeat . if no prime is found after @xmath363 repetitions , output @xmath141 and halt . the upper bound on the probability of failure follows from theorem [ thm : pnt ] since the probability that @xmath141 is returned is at most @xmath364 . let @xmath365 . call a prime @xmath55 _ bad _ if @xmath149 . analoguously to the proof of lemma [ lem : hashsss ] , there are at most @xmath366 bad primes . let @xmath367 be the first @xmath122 prime numbers in increasing order where @xmath368 . since there are only @xmath369 bad primes , the majority of the set of integers @xmath370 will be equal to @xmath356 . then use the folklore majority voting algorithm @xcite as in algorithm [ alg : prfapp ] : the correctness follows from the above discussion and the correctness of the majority voting algorithm @xcite that is folklore and implemented in the algorithm . for the running time , note that @xmath379 is @xmath164 by theorem [ thm : pnt ] and hence the running time is @xmath380 since step 6 takes time @xmath164 . it is clear that the algorithm can be implemented using polynomial space . for every @xmath381 $ ] we have @xmath382 and hence @xmath121 is easily seen to be a homomorphism by observation [ obs : hom ] . thus , by observation [ obs : homcir ] we know that @xmath383 , that is , for every @xmath384 @xmath385 hence , if @xmath386 , there must exist @xmath387 such that @xmath388 and @xmath389 . equivalently , @xmath390 . for any @xmath391 with @xmath392 we have @xmath393=\prod_{i=1}^s{\operatorname{prob}}_{{\boldsymbol h}}[({\boldsymbol x}{\boldsymbol h})_i = 0]=2^{-s},\ ] ] where the probability is taken uniform over all binary @xmath394 matrices , and the two equalities follow from the fact that the random variables @xmath395 for @xmath396 are independent and uniformly distributed . now the claim follows by taking the union bound over all elements in the support : @xmath397 \leq { \operatorname{prob}}[\exists { \boldsymbol x } \in { \mathrm{supp}}({\boldsymbol v}):\ { \boldsymbol x}\neq { \boldsymbol t } \wedge { \boldsymbol x}{\boldsymbol h}={\boldsymbol t}{\boldsymbol h } ] \leq |s|2^{-s } \leq \frac{1}{2}.\ ] ] let @xmath398 $ ] such that @xmath399 for every @xmath194 . it suffices to show that @xmath400 since @xmath187 returns @xmath401 as can bee seen from line 4 , and this is equal to @xmath402 by lemma [ lem : wht ] . for proving that @xmath400 , we first claim that @xmath201 is a homomorphism from @xmath403 $ ] to @xmath53 since , for @xmath404 $ ] , we have that @xmath405 equals @xmath406 and @xmath407 equals @xmath408 then , by observation [ obs : homcir ] , @xmath197 returns @xmath409 . for @xmath410 $ ] we have @xmath411 so @xmath412 and hence @xmath400 . use theorem [ thm : mainls ] . assume @xmath413 and create the instance @xmath414 of linear sat where @xmath25 is the incidence matrix of the set system @xmath415 , @xmath416 , @xmath417 and @xmath418 . it is easy to see that the algorithm of theorem [ thm : mainls ] returns exactly the number of set partitions of size at most @xmath3 consider the following circuit @xmath108 over @xmath422 $ ] : @xmath423&= \begin{cases } \displaystyle { \left\langle 1,{\boldsymbol 0 } \right\rangle } & \text{if } i = j=0\\ \displaystyle 0 & \text{if } i=0\text { and } j\neq0\\ \displaystyle \sum_{h=0}^j f[i-1,h]g[j - h ] & \text{otherwise , where } \end{cases}\\ \label{eq : expspace2 } g[j ] & = \sum_{i=1}^n \big[|s_i|=j\big ] { \left\langle 1,s_i \right\rangle}\end{aligned}\ ] ] for every @xmath424 and non - negative integers @xmath219 and @xmath425 , the coefficient @xmath351_{{\boldsymbol x}}$ ] counts the number of ways to choose an @xmath219-tuple of sets in @xmath426 such that their sizes sum up to @xmath425 and their characteristic vectors sum to @xmath238 in @xmath427 . thus @xmath428 ) \leq 2^{{\operatorname{rk}}({\boldsymbol a})}$ ] . furthermore , @xmath429_{{\boldsymbol 1}}$ ] is the number of set partitions of size @xmath3 times @xmath430 . indeed , if a @xmath3-tuple of sets from @xmath426 contributes to @xmath431_{{\boldsymbol 1}}$ ] , each element of @xmath264 must occur in a unique set in the @xmath3-tuple . it remains to compute @xmath432)_{{\boldsymbol 1}}$ ] . for this we will use algorithm @xmath187 with @xmath433 , except that we replace line 4 with the following to compute @xmath434 , where @xmath213 is the output of @xmath195 : the correctness follows from claim [ clm : one ] and the observation that the inversion formula from theorem [ lem : wht ] is returned on line 7 . indeed , @xmath121 is a homomorphism and @xmath438 is a bijective homomorphism , so observation [ obs : homcir ] enables us to compute @xmath437 using . to establish the time and space complexity , we observe that steps 5 and 6 take @xmath439 time by elementary analysis and theorem [ thm : fastwht ] , and that step 7 takes @xmath440 time since we can compute @xmath441})$ ] via in @xmath442 operations in @xmath443 by relying on the stored values @xmath444){\boldsymbol \phi}$ ] , where each operation requires @xmath445 time by theorem [ thm : fastwht ] . this section is dedicated to the proof of theorem [ thm : unionhashsmallsupp ] . instead of a combinatorial proof similar to the one of section [ sec : unionhashing ] , we use the algebraic perspective in this proof since we feel it gives a more fundamental insight into the hashing function used . to obtain theorem [ thm : unionhashsmallsupp ] , we prove the following generalization of lemma [ lem : compresssimple ] . [ thm : unionhashing ] there is an algorithm that , given circuit @xmath108 over @xmath260 $ ] with singleton inputs outputting @xmath80 , and a set family @xmath338 such that @xmath446 computes a list with @xmath274 for every @xmath275 in @xmath447 time . we now combine lemma [ thm : unionhashing ] with the iterative compression technique to obtain theorem [ thm : unionhashsmallsupp ] in a way very similar to section [ sec : unionhashing ] . instead of the subformula @xmath448 from section [ sec : unionhashing ] , we need the following notion . [ defi : restri ] given a circuit @xmath108 over @xmath260 $ ] and @xmath350 , the _ restriction of @xmath108 to @xmath272 _ is the circuit obtained by applying the function @xmath121 to it , where , for @xmath449 @xmath450 assume @xmath267 , and for every @xmath451 , let @xmath452 be the restriction of @xmath108 to @xmath453 and @xmath454 be the output of @xmath455 . then , since taking restrictions is homomorphic , we have by observation [ obs : homcir ] that @xmath456 this implies that if @xmath457 , then @xmath458 and hence @xmath459 we now describe how to implement the theorem . first note that @xmath460 . for @xmath461 do the following : set @xmath462 . then by , @xmath463 . hence , by invoking the algorithm of lemma [ thm : unionhashing ] using @xmath455 as the circuit and @xmath464 as the set family , we can obtain @xmath465 for every @xmath457 in @xmath466 time which is @xmath467 time . from these values we can easily obtain the support of @xmath454 . after @xmath29 steps , we have computed @xmath468 for every @xmath275 . the claimed running time follows from the fact that @xmath469 for every @xmath461 . the remainder of this section is devoted to the proof of lemma [ thm : unionhashing ] . the idea is to use the set family @xmath334 to create a poset and homomorphically hash @xmath260 $ ] to the so - called solomon algebra of the poset . we will first recall all necessary notions and properties , second introduce the hash function , and third give the algorithm implementing lemma [ thm : unionhashing ] . let @xmath287 be a poset . an element @xmath470 is said to be an _ upper bound _ of a set @xmath471 if @xmath472 holds for all @xmath473 . the set @xmath272 is said to have a _ join _ in @xmath57 if there exists an upper bound @xmath474 ( called _ the join _ ) of @xmath272 such that , for all upper bounds @xmath116 of @xmath272 , it holds that @xmath475 . for @xmath471 , let us write @xmath476 for the join of @xmath272 ; for the join of @xmath477 we write simply @xmath478 . the _ open interval _ @xmath479 is the poset induced by the set @xmath480 . we write @xmath481 , @xmath482 $ ] and @xmath483 $ ] for the analogous ( half-)closed interval . let @xmath57 be a poset . the _ mbius function _ @xmath288 of @xmath57 is defined for all @xmath289 by @xmath484 the _ zeta transform _ @xmath291 and _ mbius transform _ @xmath292 are the @xmath293 matrices defined by @xmath294 $ ] and @xmath295 for all @xmath289 . use induction on the number of elements in the interval @xmath479 . if @xmath486 , @xmath479 contains only the empty chain , which is even . if @xmath487 , group all chains on their smallest element @xmath488 . the contribution of all these chains is exactly @xmath489 since odd chains are extended to even chains and vice - versa by adding @xmath488 . it is known that @xmath292 and @xmath291 are mutual inverses . to see this , note that @xmath490 can be interpreted as the number of even chains minus the number of odd chains in the interval @xmath482 $ ] which can be seen to be @xmath60 if @xmath491 by a pairing argument , and @xmath59 otherwise . this principle is called _ mbius inversion_. let @xmath287 be a poset . solomon algebra _ $ ] is the set @xmath493 equipped with coordinate - wise addition @xmath494 and the _ solomon product _ @xmath495 defined for all @xmath496 $ ] and @xmath497 by @xmath498 use induction of @xmath29 . for @xmath502 , the statement clearly holds . otherwise , we have by definition of @xmath495 that @xmath503 \mu(r , s ) \prod_{i=1}^{k } v^i_{a_i}\\ & = \sum_{a_1,\ldots , a_k \in p } \left ( \sum_{a_1,\ldots , a_{k } \leq q } \mu(q , s ) \right ) \prod_{i=1}^{k } v^i_{a_i } \end{aligned}\ ] ] for every @xmath506 we have @xmath507 \mu(q , z ) f_x g_y \\ & = \sum_{x , y , q \in p } [ x , y \leq q \leq w ] \left(\sum_{q \leq z \leq w } \mu(q , z)\right ) f_x g_y \\ & = \sum_{x , y , q \in p } [ x , y \leq q \leq w ] [ q = w ] f_x g_y \\ & = \sum_{x , y \in p } [ x , y \leq w ] f_x g_y = \left(\sum_{x \leq w}f_x\right ) \left ( \sum_{y \leq w } g_y \right ) = ( { \boldsymbol f}{\boldsymbol{\zeta}})_{w } ( { \boldsymbol g}{\boldsymbol{\zeta}})_{w}.\end{aligned}\ ] ] the singleton @xmath509 is the multiplicative identity because @xmath510 where the last equality follows from mbius inversion . to see that @xmath492 $ ] is a commutative ring , note that @xmath291 is an isomorphism from @xmath492 $ ] to the ring @xmath493 . indeed , theorem [ thm : zethom ] shows that @xmath291 is a homomorphism , and @xmath511 shows that @xmath291 is bijective . let @xmath264 be an @xmath29-element set and let @xmath287 be a poset that satisfies @xmath516 and has a minimum element @xmath508 . define the function @xmath517\rightarrow { \mathbb{n}}[p]$ ] by setting , for all @xmath518 $ ] , @xmath519 [ lem : homhashunion ] if @xmath287 is a poset with minimum element , @xmath516 and @xmath520 $ ] such that every @xmath275 has a join in @xmath57 and for every @xmath521 , @xmath522 then for every @xmath523 , @xmath524 . denote @xmath525 . then , @xmath526 \prod_{i=1}^{|x| } [ a_i = e_i]\\ & \ { x \text { has a join by assumption since it is in the support } \ } \\ & = \sum_{x \in { \mathrm{supp}}({\boldsymbol v } ) } \left ( \sum_{e_1 \vee \ldots \vee e_{|x| } \leq q \leq y } \mu(q , y ) \right ) v_x\\ & \ { \text{m\"obius inversion } \ } \\ & = \sum_{x \in { \mathrm{supp}}({\boldsymbol v } ) } [ e_1\vee\ldots\vee e_{|x| } = y ] v_x\\ & \ { \text{by the assumption stated in~\eqref{eqn : reqprop } } \ } \\ & = v_x\end{aligned}\ ] ] for every @xmath527 $ ] we have @xmath528 @xmath529 \}\\ & = { \bigoplus}_{x \subseteq u } \biggl\langle\sum_{v \cup w = x}v_vw_w,\hat 0\biggr\rangle{\otimes}{\bigotimes}_{e \in x } { \left\langle 1,e \right\rangle}\\ & \ { \text{by definition of $ { \oplus}$ } \}\\ & = { \bigoplus}_{x \subseteq u } { \bigoplus}_{v \cup w = x}{\left\langle v_vw_w,\hat 0 \right\rangle}{\otimes}{\bigotimes}_{e \in x } { \left\langle 1,e \right\rangle}\\ & \ { \text{by lemma \ref{lem : idempot } and commutativity of $ { \otimes}$ from lemma \ref{lem : solring } } \ } \\ & = { \bigoplus}_{x \subseteq u } { \bigoplus}_{v \cup w = x}{\left\langle v_vw_w,\hat 0 \right\rangle}{\otimes}{\bigotimes}_{e \in v } { \left\langle 1,e \right\rangle } { \otimes}{\bigotimes}_{e \in w } { \left\langle 1,e \right\rangle}\\ & \{\text{using that } v \text { and } w \text { determine } x \}\\ & = { \bigoplus}_{v , w \subseteq u } { \left\langle v_vw_w,\hat 0 \right\rangle}{\otimes}{\bigotimes}_{e \in v } { \left\langle 1,e \right\rangle } { \otimes}{\bigotimes}_{e \in w } { \left\langle 1,e \right\rangle}\\ & \ { \text { by lemma \ref{lem : idempot } } \}\\ & = { \bigoplus}_{v , w \subseteq u } ( { \left\langle v_v,\hat 0 \right\rangle}{\otimes}{\left\langle w_w,\hat 0 \right\rangle}){\otimes}{\bigotimes}_{e \in v } { \left\langle 1,e \right\rangle } { \otimes}{\bigotimes}_{e \in w } { \left\langle 1,e \right\rangle}\\ & \ { \text { by distributivity from lemma \ref{lem : solring } } \}\\ & = \left({\bigoplus}_{x \subseteq u } { \left\langle v_x,\hat 0 \right\rangle } { \otimes}{\bigotimes}_{e \in x } { \left\langle 1,e \right\rangle}\right ) { \otimes}\left({\bigoplus}_{x \subseteq u } { \left\langle v_x,\hat 0 \right\rangle}{\otimes}{\bigotimes}_{e \in x } { \left\langle 1,e \right\rangle}\right)\\ & = h({\boldsymbol v}){\otimes}h({\boldsymbol w}).\end{aligned}\ ] ] we are now ready to give the proof of lemma [ thm : unionhashing ] . as mentioned before , the proof idea is to construct a poset @xmath57 from the given set family and hash the given circuit to a circuit @xmath530 over the solomon algebra @xmath492 $ ] with the homomorphic hash function @xmath121 . then by theorem [ thm : zethom ] and lemma [ lem : solhom ] , the zeta - transform of the output of @xmath530 can be computed fast using point - wise multiplication , and then using mbius inversion the original output can be computed . because of the hash property from lemma [ lem : homhashunion ] of @xmath121 , the required output of @xmath108 can then be read from the output of @xmath530 . @xmath532 construct the poset @xmath533 . construct the circuit @xmath195 over @xmath492 $ ] obtained from @xmath108 by applying @xmath121 . @xmath535 @xmath536 construct the circuit @xmath202 over @xmath53 obtained from @xmath195 by applying @xmath537 . evaluate @xmath202 and return the output . @xmath325 let @xmath326 such that @xmath327 implies @xmath328 . @xmath330 @xmath241 . the algorithm is algorithm @xmath538 . let us first analyse the complexity . recall that to establish the claimed bounds we assume that each operation in @xmath53 takes one time unit . step 1 can easily be implemented in @xmath539 time . step 2 expands every singleton ( of @xmath260 $ ] ) in @xmath108 according to into a product of at most @xmath540 singletons of @xmath492 $ ] , and thus can be implemented in time and storage @xmath541 . step 6 can be implemented in time and space @xmath542 because the singleton @xmath543 is mapped to @xmath544v$ ] . in step 7 , the evaluation of @xmath202 over @xmath53 uses @xmath211 operations in @xmath53 . the complexity of algorithm @xmath321 is easily seen to be @xmath447 , and since the loop at step 3 is the only other time - consuming part the resource bounds are clearly met . we proceed with the proof of correctness of algorithm @xmath538 . first note that since @xmath446 , we have by construction that for every @xmath545 , @xmath546 , since every other upper bound in @xmath57 on @xmath547 must be above the element @xmath272 of @xmath57 . this directly implies that the requirement imposed by is met by @xmath57 . hence by lemma [ lem : homhashunion ] we have that @xmath524 for every @xmath275 . thus to prove correctness it suffices to show that the vector @xmath548 is equal to @xmath549 . since @xmath121 is a homomorphism by lemma [ lem : solhom ] , observation [ obs : homcir ] implies that @xmath550 outputs @xmath549 . then , we claim that for every @xmath551 it holds that @xmath552 . to see this note that on line 7 first the zeta transform is applied to @xmath195 which is a homomorphism to @xmath553 by theorem [ thm : zethom ] , and then we restrict to the coordinate @xmath163 . the claim than follows by observation [ obs : homcir ] since both transformations are homomorphisms . the proof of item ( b ) is fairly similar to the proof of item ( a ) . the main difference is the use of the poset @xmath57 . in this section , we will use a poset larger than the support of @xmath80 , but the advantages are that 1 . we not need to construct the poset using iterative compression , and hence we do not need to determine all components of @xmath80 , 2 . the poset will be decomposable , implying that the bottleneck in ( a ) , the mbius inversion step , can be improved . let us now start with the more formal treatment . given two posets @xmath557 and @xmath558 , the _ direct product _ @xmath559 is the poset on the set @xmath560 such that @xmath561 if @xmath562 and @xmath563 . [ lem : dirprodmob ] let @xmath57 and @xmath564 be posets and let @xmath565 and @xmath566 denote their respective mbius functions . denote @xmath567 and @xmath568 for the mbius function and order of @xmath559 , then for @xmath561 it holds that @xmath569 [ lem : yatmob ] there is an algorithm that , given a poset @xmath573 ordered by set inclusion and containing @xmath574 and @xmath264 , computes a table with @xmath575 and @xmath576 for every @xmath577 in @xmath578 time and space in this proof @xmath579 $ ] denotes the set @xmath580 for an integer @xmath219 . let @xmath581 $ ] . we first define @xmath582 \rightarrow \mathbb{n}$ ] by letting @xmath583 it is easy to see that @xmath584 . also , define for every @xmath461 : @xmath585g(s ) & \text{if } i=0 \text { and } s\neq\emptyset\\ f_{i-1}(s ) + [ i \in s]f_{i-1}(s \setminus i ) & \text{otherwise}. \end{cases}\ ] ] then it holds that @xmath586 for every @xmath587 . note that by symmetry ( that is , reversing the order of the poset ) , this procedure can be used as well to compute @xmath576 for every @xmath577 in the same time bounds . by theorem [ thm : zethom ] , @xmath291 is a homomorphism from @xmath492 $ ] to @xmath493 . hence , similarly as in the proof of theorem [ thm : unionhashsmallsupp](a ) , we can compute @xmath596 in time polynomial in the input size . then the claim follows directly by applying lemma [ lem : yatalg ] . @xmath597 obtain @xmath594 using claim [ clm : incexc ] let @xmath598 and @xmath599 and construct the poset @xmath600 . construct the circuit @xmath195 over @xmath492 $ ] obtained from @xmath108 by applying the homomorphism @xmath121 as defined in . @xmath534 . compute @xmath601 for every @xmath602 using lemma [ lem : yatmob ] . @xmath603 the arguments for the correctness of this algorithm are all similar to the arguments of for the correctness of algorithm @xmath604 : first note that @xmath605 . this in turn implies the condition of lemma [ lem : homhashunion ] , and hence if @xmath606 is the output of @xmath550 we indeed have that @xmath607 . as shown before in the proof of lemma [ thm : unionhashing ] , @xmath608 . hence it remains the show that on line 7 the expression is indeed the mbius inversion formula . this follows from the definition of @xmath292 , the direct product property of the mbius function from lemma [ lem : dirprodmob ] and the fact that in the subset lattice @xmath609 for @xmath610 . for the running time of the above algorithm , note that line 1 takes @xmath595 time . for line 2 , we have @xmath611 and @xmath612 , and hence @xmath613 . thus , lines 2 - 5 take @xmath614 time . line 6 takes @xmath595 time , and using these values , line 7 can be performed in time at most @xmath614 . simultaneously try all integers @xmath616 , and for every @xmath588 run the algorithm of lemma [ lem : trd ] . terminate whenever any of these algorithms terminates . this procedure runs in time @xmath617 where the inequality is achieved by taking @xmath618 .

we study theoretically and experimentally the frequency and temperature dependence of resistivity noise in semiconductor heterostructures-doped by mn . the resistivity noise is observed to be non - monotonous as a function of frequency . as a function of temperature , the noise increases by two orders of magnitude for a resistivity increase of about 50% . we study two possible sources of resistivity noise dynamic spin fluctuations and charge fluctuations , and find that dynamic spin fluctuations are more relevant for the observed noise data . the frequency and temperature dependence of resistivity noise provide important information on the nature of the magnetic interactions . in particular , we show how noise measurements can help resolve a long standing debate on whether the mn - doped gaas is a- zener / ruderman - kittel - kasuya - yosida ( rkky ) or double exchange ferromagnet . our analysis includes the effect of different kinds of disorder such as spin - glass type of interactions and a site - dilution type of disorder . we find that the resistivity noise in these structures is well described by a disordered rkky ferromagnet model dynamics with a conserved order parameter .

1202.3207

the mechanism of magnetic ordering in dilute magnetic semiconductors ( dmss ) has traditionally attracted much attention of investigators . one of the issues being most intensely studied @xcite concerns the microscopic nature of magnetism in bulk dmss ; in particular , whether the microscopic interaction governing the ferromagnetism ( fm ) is of zener / rkky type or double - exchange type . let us recall that the zener / rkky model is applicable when the effective coupling between local magnetic moments and the spins of charge carriers is smaller than the carrier bandwidth , and , this model is widely used in the context of magnetism in the kondo lattice systems @xcite . on the other hand , the double exchange model is believed to describe the magnetism in manganites @xcite and double perovskites @xcite where the hund s rule coupling is known to be large compared to the carrier bandwidth . while the ordering of local moments of magnetic metal in both cases is mediated by the charge carriers of the semiconductor ( sc ) host , the two models differ in the relative importance of the magnitudes of the effective coupling of the local moments with the spins of charge carriers and the intersite hopping integrals ( in other words the bandwidth for charge carriers ) . the current opinion is that both mechanisms of fm can take place in bulk dmss . the conditions of the realization of either mechanism strongly depend not only on the concentration of magnetic metal ions , but also on the character of their distribution in the sc host . the last factor plays an important role in fm ordering in bulk dmss and is crucially driven by the details of their growth technology . besides , technological or fabrication aspects drastically affect the character of fluctuations of crystal - field and exchange potentials in the sc host as well as the degree of ionization of magnetic metal atoms and effective concentration of carriers supplied by them when doping the host . these aspects also determine the relative importance of phase segregation in the system , as well as the contribution of wide and narrow bands in the spectrum of carriers to the coupling between local magnetic moments of the metal . an emerging trend in the studies of dmss concerns the investigation of hybrid layered heterostructures ( in the following we shall also use the term `` 2d dms structures '' ) containing magnetic ultrathin metal layers ( so - called @xmath0-layers ) embedded into a nonmagnetic sc heterostructure . this interest in 2d dms structures has a twofold motivation . the first one is practical : present - day microelectronic devices have a planar geometry and it becomes necessary to understand the mutual interaction of different parts of the structure such as the @xmath0-layer and quantum wells . a number of studies of 2d dms structures have been reported in the literature @xcite . in refs . , the fm state in mn @xmath0-doped gaas / algaas heterostructures was observed at rather high temperatures . however , these devices have low mobility values because the authors of refs . aimed at achieving the highest possible hole density just in the vicinity of mn ions to maximize the curie temperature @xmath3 . note here that mn ions are responsible not only for magnetism but also act as acceptors and a source of the random potential that limits mobility . second , properties of 2d magnets are often qualitatively different from those of their bulk counterparts , and studies in heterostructures can sometimes clarify issues of basic research on two - dimensional magnetism in solids . issues peculiar to the heterostructures , such as magnetic anisotropy , limit their use in a general 2d ferromagnet problem . however for properties such as the frequency dependence of the magnetization noise , the universality of the behavior is not sensitive to issues such as magnetic anisotropy since the universality is dictated by other properties such as the existence of a conserved order parameter , and heterostructures provide a good platform . note also that as we have studied resistivity noise of the hole gas in the 2d quantum well , our analysis will be useful for studies of magnetic impurity effects in low - dimensional conductors . it is also perhaps worth observing here that the idea of using resistivity noise for probing magnetization dynamics that we develop here is not specifically for heterostructures and can be adapted for the study of other 2d ferromagnets . as a rule , in the bulk dmss there are no insurmountable problems in obtaining detailed information on fm ordering from direct magnetic measurements although it is often difficult to correctly interpret the obtained data in terms of some specific theoretical model . in 2d dms structures , the situation turns out to be more complicated and direct magnetic measurements that are able to detect fm ordering in the metal layers are often tedious , troublesome , and sometimes even spurious as opposed to the bulk dmss . moreover , direct measurements of the magnetization are impractical given the small size of the magnetically active region in 2d dms structures . in such a situation , an indirect retrieval of magnetic characteristics ( for example , from resistivity and hall effect measurements ) becomes decisive , whereas for the bulk dmss , it is in general a subsidiary tool . however , even in the bulk dmss , the resistivity anomaly , which is a peak or shoulder at the temperature dependence of resistivity , is widely used as an evidence for the onset of significant fm correlations and the measure of @xmath3 . @xcite in an earlier paper ( ref . ) , it was demonstrated that in contrast to the resistivity anomaly in bulk dmss , which typically appears near ( and above ) the curie temperature , the resistivity anomaly observed in 2d dms structures typically appear far below the mean - field curie temperature and may even exist in the absence of an actual magnetic phase transition . nevertheless , the resistivity anomaly ( a peak or shoulder ) in 2d dmss can be also regarded as an indication of the onset of significant fm correlations ) . @xcite the use of indirect experimental methods to reveal magnetic properties in 2d dms structures introduces additional considerations into the interpretation of data compared to bulk samples . for example , magnetic proximity effects and long - range coulomb interaction between different components of such structures have a crucial influence on the magnetic and transport characteristics of the system . recently , structures with a single fm @xmath0-layer deposited near or inside the sc quantum well ( qw ) forming the 2d channel of conductivity have emerged as a test system to study magnetic proximity effects . for example , in ref . , photoluminescence polarization was observed in al@xmath4ga@xmath5as / gaas / al@xmath4ga@xmath5as structures with a 0.5 monolayer of mnas placed at 5 nm from a gaas qw . crucially , these structures had a gate , which allowed shifting holes from the qw to the fm layer . it was shown that the holes in the qw are spin - polarized and that their degree of polarization strongly increases ( from 0.4% to 6.3% ) with the `` pressing '' gate voltage . in the control structures , lacking the fm @xmath0-layer , the photoluminescence polarization was not observed . induced photoluminescence polarization effect was recently observed in other structures such as gaas / in@xmath6ga@xmath7as / gaas ( @xmath8 ) containing mn @xmath0-layer separated from the ingaas qw by a 3 - 5 nm thick gaas spacer . @xcite it is clear that the aforementioned results can hardly be explained by a simple tunneling of holes from the qw to fm layer through the spacer , since even under a pressing voltage the characteristic depth of such tunneling does not exceed 1 nm . in 2d dms structures containing spatially separated 2d hole gas in the qw and fm @xmath0-layers , the mutual influence of these two subsystems has three effects that have a bearing on the magnetic properties . first , interpenetration of the wave function tails between the qw and the fm layers polarizes the spins of itinerant charge carriers in the qw and modifies the effective coupling between local spins in the fm layer . @xcite second , quantum fluctuations in the qw stabilize magnetic order in the fm layer , suppressing at the same time the amplitude of the magnetic moment and the transition temperature with respect to those found by mean - field estimates . @xcite third , electrostatic charge redistribution occurs between the qw and the fm layers due to their different density of states and depths ; and as a result , modification of the magnetic characteristics of the fm layer occurs even on purely classical grounds . @xcite recently , different ways have been proposed to describe self - consistently ferromagnetic ordering and spin polarization of charge carriers in the 2d dms structures under discussion . assuming that mn atoms form isolated paramagnetic centers inside the @xmath0-layer , the authors of refs . proposed a model of indirect rkky - type exchange coupling between local moments of mn atoms in the @xmath0-layer via the `` tails '' of the wave function of itinerant hole states in qw . in the framework of the same assumption about of a localized character of the hole states in the @xmath0-layer , the authors of ref . formulated the problem in terms of the anderson - fano model of configuration interaction between the localized hole states at mn centers and itinerant hole states in the qw . in the framework of the method of ref . , the spin splitting of itinerant hole states in an external magnetic field is strongly enhanced due to their hybridization with localized hole states . this leads to a resonant enhancement of the interband radiation recombination of itinerant holes and causes circular polarization of the luminescence in qw . on the other hand , an alternative approach based on the assumption of an itinerant character of the hole states in both the qw and fm @xmath0-layer was successfully used @xcite . it has been shown that a thin layer of fm metal located in a bulk sc matrix induces quasi-2d spin - polarized collective states and a half - metallic type of an electron spectrum of the system is formed @xcite . these states ( called also confinement states ) have rather extended character along the structure axis of growth because of their small binding energies , so they penetrate deep into the spacer and can reach the qw . it may be shown ( see the next section ) that hybridization of confinement states in the fm @xmath0-layer , with one - electron states in qw leads to the spin polarization of carriers in the qw . in the model @xcite , the holes in the qw play mostly an `` observer '' ( passive ) role in the exchange coupling inside the @xmath0-layer , although this does not mean that they have no effect on the fm ordering in the system . @xcite if one assumes an easy - plane character of magnetic anisotropy of the `` free '' @xmath0-layer ( i.e. in the absence of the qw ) , then introducing a qw near the @xmath0-layer can drastically change the type of magnetic anisotropy . as a result , the easy axis of magnetization appears to be directed along the normal to the @xmath0-layer plane . @xcite experiments @xcite as well as recent direct measurements of magnetic anisotropy @xcite give clear indication of the existence of such an orientational transition . in addition to the effects of induced spin polarization of holes in the qw and the reorientation of magnetization in the fm @xmath0-layer , there exist purely electrostatic ( coulomb ) effects of charge redistribution between these two subsystems . as it was shown in ref . , the long - range fluctuations of electrostatic potential , which are inevitable due to an inhomogeneous distribution of magnetic metal ions over the fm @xmath0-layer , are projected onto the qw and give rise to similar fluctuations of electrostatic potential in the qw . due to this fact , the transport measurements involving charge carriers in the qw can reveal some tiny details of disorder in the distribution of magnetic metal ions in the fm @xmath0-layer , the latter being inaccessible by the direct magnetic measurements . it is evident from the above discussion that in comparison with bulk dms systems , 2d dms structures have a number of additional experimental and theoretical complications arising from the use of indirect probes for magnetism as well as the interplay between the magnetic @xmath0-layer and the qw . the choice of the indirect probe is dictated not only by the convenience of measurement , but also by the level of detail one is seeking as regards the properties . resistivity measurement , although an indirect probe of magnetism , is one of the most widely used probes in both bulk and 2d dms systems given its simplicity . in both systems , the observation of an anomaly in the temperature dependence of the resistivity is associated with the onset of significant ferromagnetic correlations . unfortunately , such a measurement is unsuitable for shedding light on the microscopic interactions governing fm : both the rkky and double - exchange models can describe the resistivity anomaly observed in dmss . this is not surprising since as far as such static properties are concerned , both can be described by effective heisenberg models . the differences manifest themselves , however , in the dynamic properties such as the autocorrelation function , since the damping of magnetic excitations is sensitive to the microscopic details governing the fluctuations . in the following , we shall show how the magnetization dynamics can be probed through resistivity noise measurements . while our analysis is applied to 2d dms structures , we note that this method of deducing the mechanism of fm from the analysis of dynamic susceptibility is quite general and can be used for other systems . in this paper , we present a theoretical and experimental study of resistivity noise as an indirect probe of magnetic dynamics in 2d dms structures @xmath0-doped by magnetic atoms . we show how resistivity noise can distinguish between rkky / zener and double - exchange mechanisms . we find that in our structures the noise measurements are consistent with a disordered rkky picture . the experiments reported in this paper were performed using the gaas / in@xmath9ga@xmath7as / gaas qw structure @xmath0-doped by mn . such structures produced by selective doping exhibit a high enough hole mobility ( more than 2000 @xmath10 at 5 k ) and show clear evidence of a 2d excitation spectrum as well as fm at moderately high temperatures of about @xmath11 @xcite . in fig . [ fig : structure ] , we illustrate a schematic layout of the fabricated structures . similar 2d heterostructures were also reported elsewhere ; however , the ferromagnetic ordering was obtained at a much lower ( millikelvin ) range of temperatures . @xcite . depending on mn content , the samples of the type presented in fig . [ fig : structure ] exhibit metallic or activation conductivity on both sides of the metal - insulator transition . the insulating samples are most interesting as they can give us valuable insights into the mechanism of ferromagnetism in these dms heterostructures and we will focus on the sample very close to the percolation transition since it demonstrates some features of both metallic and insulating behavior . ( color online ) layout of the mn @xmath0-doped gaas / in@xmath9ga@xmath7as / gaas qw heterostructures used in this work and in ref . . the carbon ( c ) layer is a nonmagnetic source of holes . noise measurements reported in this paper correspond to sample 4 in ref . , an insulating sample on the border of a percolative metal - insulator transition . ] the rest of the paper is organized as follows . in sec . [ sec : proximity ] , we clarify the role of magnetic proximity effects in the interplay of ferromagnetism and carrier transport in 2d dms structures with a spatially separated fm @xmath0-layer and the qw . in sec . [ sec : models ] , we present our model for the resistivity noise in qw and obtain its relation to charge and magnetization dynamics in the fm @xmath0-layer . it is shown how magnetization dynamics is sensitive to the microscopic nature of fm correlations and can be used to distinguish between rkky and double - exchange types of coupling . disorder effects are also analyzed . in sec . [ sec : comparison - with - experiment ] , we compare these findings with our experimental data on frequency and temperature dependence of resistivity noise . it is argued that the data supports the picture of a disordered rkky type of fm over a double - exchange type of fm . section [ sec : discussion ] contains a discussion of our results . before proceeding to the point at issue of our work , in this section we shall try to shed additional light on a key physical problem of the systems under consideration , namely , how magnetic correlations in the fm @xmath0-layer affect charge transport in the qw . let us recall the microscopic anderson - type model for a single fm @xmath0-layer embedded into the sc host proposed in refs . . it has been shown that fm order in such system can be attributed to the intrinsic physical properties of the @xmath0-layer . the relevant effects described by this model are the hybridization of the electron states of the metal and semiconductor atoms , the strong charge and spin redistributions around the @xmath0-layer , and the electron - electron correlation on the metal atom . following refs . , confinement states in the form of quasi-2d spin - polarized subbands located within the sc band gap arise in the host near the @xmath0-layer . this leads to the half - metallic character of the electron spectrum of the system . under certain approximations , a picture of the low energy branches of these confinement states can be qualitatively reproduced by a simple phenomenological hamiltonian @xmath12 @xmath13 here the 3d hamiltonian @xmath14 describes itinerant electron states in a bulk sc , @xmath15 is the effective one - dimensional spin polarized potential of the charge carriers in a single fm @xmath0-layer @xmath16\delta(z+l ) < 0 , \ , \ , ( v_1 , v_2 ) > 0 . \label{eq : vdelta}\end{aligned}\ ] ] the @xmath17 axis is oriented along the direction of the structure growth , the fm @xmath0-layer lies in the @xmath18 plane , @xmath19 is the delta - function , @xmath20 and @xmath21 are parameters derived from the model described in ref . , @xmath22 is the vector composed of the pauli matrices in the spin space , @xmath23 is the unit vector along the direction of magnetization of the fm @xmath0-layer . let us introduce the energy parameters @xmath24 specifying the positions of the spin - polarized subbands of confinement states described by hamiltonian for a single fm @xmath0-layer in the absence of the qw , with the wave functions @xmath25 and characteristic lengths @xmath26 ( @xmath27 is the crystal momentum in the ( @xmath28 ) plane perpendicular to the @xmath17 axis , @xmath29 is the spin - projection index in the diagonal representation for the @xmath30 matrix . below , we discuss a half - metallic regime , when @xmath31 and the @xmath32 states are empty at all @xmath27 , while the @xmath33 states are partially occupied depending on the fermi level position . the low - energy branches in the spectrum of charge carriers in a single nonmagnetic qw are successfully modeled by a phenomenological hamiltonian @xmath34 @xmath35 here @xmath36 is an effective one - dimensional potential of the charge carriers in a single nonmagnetic qw : @xmath37 without the effects of charge redistribution between the fm @xmath0-layer and qw one can simply assume that @xmath38 at @xmath39 and @xmath40 at @xmath41 and @xmath42 , where @xmath43 is the nominal qw thickness ; the charge redistribution leads to a more intricate form of @xmath44 due to appearance of a qw / spacer interface . let us introduce the energy parameters @xmath45 specifying the position of the @xmath46th spin - degenerate subband of 2d states in the spectrum of qw described by hamiltonian of a single qw in the absence of the fm @xmath0-layer ( @xmath47 ) . the corresponding wave functions @xmath48 exponentially decay with @xmath49 away from qw , @xmath50 , where @xmath51 are characteristic lengths , and @xmath52 is the effective mass of charge carriers . in the simplest case , only one spin - degenerate subband of 2d states , @xmath53 with the parameters @xmath54 and @xmath55 exists in the qw . the states @xmath53 are partially occupied depending on the fermi level position . let us now treat the 2d dms structure containing both the fm @xmath0-layer and the qw as a triple layer quantum system with an effective one - dimensional potential @xmath56 composed of two wells with different widths and depths separated by the barrier ( spacer ) of width @xmath57 . we can write the model hamiltonian @xmath58 describing the low - energy states of the charge carriers in the following form : @xmath59 evidently , the effective potential @xmath56 can not be correctly modeled as a simple sum of the potentials @xmath36 and @xmath15 , due to a strong redistribution of the carriers between the qw and fm @xmath0-layer . in principle , using an appropriate form of @xmath56 , one can calculate ( analytically or numerically ) the eigenenergies and eigenstates of hamiltonian at all values of the spacer thickness @xmath57 , but this method seems to be tedious . however , to reveal the main physics of the magnetic proximity effect between the fm @xmath0-layer and qw in the quantum structure described by hamiltonian , we consider only the situation of a relatively thick spacer , @xmath60 . we assume that the minima of @xmath61 and @xmath62 lie at @xmath63 and the occupation number of @xmath33 states in fm @xmath0-layer significantly exceeds that of @xmath53 states in the qw . thus , in the half - metallic regime we have @xmath64 and @xmath65 and the fermi level @xmath66 lies between two minima , @xmath67 and @xmath68 , i.e. @xmath69 . this corresponds to the fact that the charge carrier density in the fm @xmath0-layer is much higher than that in the qw . we now write the wave function @xmath70 of the system with hamiltonian in the variational form @xmath71 where @xmath72 and @xmath73 are coefficients obeying an evident normalization condition : @xmath74 @xmath75 note that the functions @xmath53 and @xmath25 are not orthogonal , so @xmath76 . obviously , this simplest approach treats only qualitatively the problem of interference of the charge carrier states of two subsystems as a simple superposition of the wave functions @xmath53 and @xmath25 . moreover , it is unable to capture in full measure possible resonance effects ( for example , arising additional bound states ) . however , even with such an approximate method , we can project the states @xmath25 of the fm @xmath0-layer onto the states @xmath53 of qw using a redefinition @xmath77 , where @xmath78 and treating @xmath79 as a small perturbation to @xmath36 at @xmath60 . omitting the cumbersome but straightforward calculations , we write the final expression for the energy spectrum of the charge carriers in qw in the second order of the perturbation series expansion in terms of @xmath79 : @xmath80 where @xmath81 ^ 2 /[\varepsilon_0(\bm{q } ) - \omega_{\alpha}(\bm { q } ) ] , \label{eq : delta - alpha}\end{aligned}\ ] ] @xmath82 @xmath83 the approach discussed above , although very rough , nevertheless captures the main physics of the magnetic proximity mechanism acting in our system due to the tunneling of confined spin - polarized charge carriers from the fm @xmath0-layer to the qw . as we see from eq . , the energy bands @xmath84 originating from the nonmagnetic qw states become spin - polarized due to their quantum interference with the spin - polarized states of the fm @xmath0-layer . it can be easily shown that to exponential accuracy , the spin splitting of the states in the qw @xmath85 contains different contributions decreasing at @xmath60 , but the principal contribution at @xmath86 is @xmath87 $ ] . note that the lengths @xmath88 decrease at large values of the crystal momentum , @xmath89 , i.e. corresponding components of @xmath90 are exponentially vanishing . that is why , using eq . , we shall limit ourselves only to the range of small crystal momenta @xmath91 . qualitative estimates show that for our system , @xmath92 nm , @xmath93 nm , and @xmath94 nm , so for the hole densities in qw @xmath95 m@xmath96 , our approximation seems to be reasonable ( here @xmath97 is the fermi crystal momentum of holes in the qw ) . thus , we regard the energies given by eq . ( at small wavevectors @xmath91 ) as eigenvalues of an effective 2d hamiltonian @xmath98 of charge carriers in the carriers in qw @xmath99 @xmath100 , \label{eq : j - u}\end{aligned}\ ] ] where @xmath101 is a 2d lateral vector in the @xmath102 plane , @xmath23 is the unit vector along the arbitrary quantization axis . hamiltonian describes 2d carriers in the qw moving in an `` external '' homogeneous field induced by their quantum interference with the spin - polarized carriers in the fm @xmath0-layer . this field has the `` coulomb '' ( @xmath103 ) and `` exchange '' ( @xmath104 ) components . let us now suppose that due to the lateral long - range fluctuations of the mn density within the fm @xmath0-layer on length scales exceeding @xmath92 nm , the components @xmath103 and @xmath104 become slowly varying functions of @xmath101 due to the corresponding variation of the potentials @xmath105 in eq . . in other words , we may treat @xmath106 and @xmath107 as slowly varying components of an effective random field projecting the long - range disorder from the fm @xmath0-layer to the qw . we can also include in @xmath106 the `` true '' long - range fluctuating coulomb potential @xmath108 of charge carriers provided by their inhomogeneous distribution in the @xmath0-layer . the effective random field ( @xmath109 ) may induce a percolative metal - insulator transition in the qw and the formation of an inhomogeneous structure with the activation - type conductivity . in this situation , the qw can be considered as a set of metallic fm droplets separated by insulating spacers . it is evident from eq . that for a hopping between @xmath110th and @xmath111th fm droplets with different orientations ( @xmath112 ) of their magnetic moments , a charge carrier in the qw has to overcome an additional energy barrier @xmath113 $ ] resulting from the effect of the proximity - induced spin polarization in this section , we first of all recall some important methodological aspects of our preceding study @xcite of the gaas / in@xmath9ga@xmath7as / gaas qw structure @xmath0-doped by mn and discuss existing principal difficulties of an adequate description of magnetism in these multi - component system . strictly speaking , there is no universal model of magnetism in 2d dms structures and the choice of a suitable description of fm ordering in such structures is in some aspects more difficult than that in bulk dmss . for example , let us qualitatively discuss the zener / rkky and double exchange pictures of magnetism in the context of the gaas / in@xmath9ga@xmath7as / gaas qw structure @xmath0-doped by mn earlier studied in ref . . these structures demonstrate an activation character of conductivity and the key problem is how their fm behavior may be mediated by the holes both in the qw and in the @xmath0-layer . let us suppose that the effect of `` external '' holes in the qw on the fm ordering in the @xmath0-layer is less significant than that of `` intrinsic '' holes in the @xmath0-layer itself . the surface density of manganese ions estimated for the sample studied in the paper ( see sec . iv , sample 4 in refs . is @xmath114 m@xmath96 ) , corresponding to the inter - manganese distance of about 1.6 nm . we consider @xmath115 as a nominal concentration of local magnetic moments in the @xmath0-layer . unfortunately , we are unable to directly extract from the conductivity measurements the density of charge carriers ( holes ) in the @xmath0-layer , @xmath116 , due to their low mobility . we can roughly treat @xmath115 as an upper limit for @xmath116 , while the compensation between acceptors and donors in the @xmath0-layer can diminish this estimate by an order of magnitude . assuming further , that only holes in the qw participate in the conductivity , it is possible to determine a lower limit for @xmath116 as being of the order of the hole density in the qw , @xmath117 . taking @xmath118 m@xmath119 , from the conductivity data , we obtain the inter - hole distance in the @xmath0-layer @xmath120 falling within the 1.67 nm range . a naive estimate of the hole effective radius in the @xmath0-layer ( @xmath121 ) in the model of a single shallow acceptor gives @xmath122 nm , where @xmath123 is the bohr radius of a light hole in gaas . this value can be treated as an upper limit for @xmath121 , while it is obviously overestimated and even exceeds the thickness of the @xmath0-layer . in the more realistic model of a single deep impurity acceptor , we obtain @xmath124 nm for the heavy hole in gaas using experimentally known binding energy @xmath125 mev for mn@xmath126 ion in gaas . this value can be treated as a lower limit for @xmath121 . so , we get an estimate 0.9 nm @xmath127 5.3 nm . at @xmath128 , i.e. at a sufficiently high hole density , the zener / rkky mechanism is relevant for our system and describes well the fm ordering in the @xmath0-layer within the framework of an effective heisenberg spin hamiltonian with an effective coupling via itinerant holes . on the other hand , the magnetism in the polaron model arises by double exchange mechanism which is relevant at a sufficiently low concentration of the holes in the @xmath0-layer , such that @xmath129 . in the polaron percolation picture of kaminski and das sarma @xcite , the holes are localized at the length scale of @xmath121 with strong hund s rule coupling to mn atoms . the above estimates are _ a priori _ unable to show preference to one or another model of ferromagnetism in our system , due to evident uncertainties in the parameters @xmath120 and @xmath121 . thus , we have to present additional arguments supporting or rejecting the model under consideration . in the following , we analyze the peculiarities of dynamic spin fluctuations in different models and calculate the frequency and time dependence of the resistivity noise power . we show that magnetization dynamics is sensitive to the microscopic nature of fm correlations and thus can be used to distinguish between zener / rkky and double - exchange models of fm . now , let us come back to the main issue of this paper . in an earlier work , @xcite , we had studied the role of charge disorder and magnetization in the mn @xmath0-layer on the resistivity of the quantum well . at low charge carrier densities , the random distribution of ionized mn atoms in the mn layer creates a fluctuating potential for the holes and leads to accumulation of holes in droplets . conduction takes place through hopping of holes between the droplets . the building of magnetic correlations between the charge droplets gives rise to a dip / shoulder - like feature in the temperature dependence of resistance . the temperature dependence of the resistivity of the droplet phase in these dms heterostructures was modeled as @xmath130 where @xmath131 is the angle between the orientations of the magnetizations @xmath132 and @xmath133 associated with the neighboring @xmath134 and @xmath135 droplets , respectively , and @xmath136 the temperature dependence of the static magnetic correlation for a 2d ferromagnet above the transition temperature will in general take the form @xmath137 where @xmath138 is the magnetic correlation length . the static magnetic susceptibility @xmath139 is related to @xmath138 through @xmath140 the two mechanisms that are generally considered for ferromagnetism in dms systems are double exchange and @xmath1-@xmath2 zener / rkky . the low - energy excitations of both these models are known to have same the dispersion relation @xmath141 and as far as static properties are concerned , both can be represented by an effective heisenberg model of spins at low energies . therefore , the two models predict the same behavior of the resistivity and can not distinguish between the two mechanisms of ferromagnetism in the mn layer . for a heisenberg model of spin interactions , @xmath142 the magnetic correlation length is given by @xmath143 the real system is likely to deviate from an isotropic heisenberg ferromagnet due to spin - orbit interactions as well as the 2d distribution of the mn doping . for instance , for a uniaxial ferromagnet , we have @xmath144 for this uniaxial magnet @xmath145 is a conserved order parameter . for small but finite @xmath146 the model shows ising - type behavior at sufficiently low temperatures , undergoing a transition to a ferromagnetically ordered state @xcite at a temperature @xmath147 @xmath148 for temperatures @xmath149 the correlation length is approximately given by eq . . as the temperature approaches @xmath150 eq . ( [ eq : xi - heisenberg ] ) for the magnetic correlation length should perhaps be replaced with a power law , @xmath151 where @xmath152 for the limiting case of the ising model . for a general anisotropic heisenberg model , the order parameter is not conserved . in ref . , we considered an isotropic heisenberg model and obtained quantitative agreement with the resistivity of the insulating samples over a broad temperature range . as is clear from the above discussion , resistivity measurements in these dms structures are unable to distinguish between rkky and double - exchange mechanisms since the static correlations in both cases can be described by an equivalent effective heisenberg model . it is then natural to examine dynamic properties such as the ( frequency - dependent ) resistivity noise . resistivity noise can of course also arise from charge fluctuations in the puddles in the quantum well . however , if noise from magnetic fluctuations dominates that from charge fluctuations , resistivity noise can be a useful tool for probing the microscopic origin of ferromagnetic interactions . this happens to be true in our case . we develop now a theory for the effect of magnetic and charge fluctuations on the resistivity noise . we will also examine the effect of various kinds of disorder on the noise . we consider the magnetic fluctuations first . we will study the implication of increasing magnetic anisotropy on the frequency dependence of the resistivity noise . for homogeneously disordered magnets , it is known that the resistivity noise @xmath153 and the magnetization noise @xmath154 are related through @xcite @xmath155 a similar relation can be obtained for our phase - segregated model . we assume that the magnetic moments in a droplet are aligned with the polarization in the mn layer directly above the droplet and disregard fluctuations of the magnitude of the droplet magnetic moment . our model for resistivity is a simple nearest - neighbor hopping type and the hopping is a function of the relative orientation of the magnetizations at the sites of the two puddles in question : @xmath156 fluctuations of the orientation of the magnetizations of the droplets cause resistivity fluctuations . for our model of resistivity , @xmath157 we thus obtain @xmath158 here @xmath159 and @xmath160 refer to the static values , and the time dependence is expressed in the fluctuations @xmath161 the greek labels refer to the orientation of the magnetization and the latin indices label the charge puddles in the quantum well . @xmath110 and @xmath111 in the above expressions refer to nearest - neighbor puddles in the quantum well and are not summed over . the resistivity noise takes the form @xmath162 where @xmath163 is the correlation function of the magnetization , @xmath164 refer to magnetization directions and @xmath165 refer to neighboring droplets . from eqs . ( ) and ( ) it thus follows that @xmath166\nonumber \\ \approx(j_{0}/t)^{2}\sum_{\alpha}\left[c_{\alpha\alpha } ( 0,\omega)+e^{-r_{ij}/\xi_{m}(t)}c_{\alpha\alpha}(r_{ij},\omega ) \right].\label{eq : r - noise}\end{aligned}\ ] ] the spin correlation function is related to the dynamic susceptibility through the fluctuation - dissipation relationship , @xmath167 the magnetization dynamics is crucially dependent on whether the order parameter is a conserved quantity . if the order parameter is not conserved , then the spin relaxation has a finite lifetime even as @xmath168 this is the case for an ising or anisotropic heisenberg ferromagnet . if the order parameter is conserved , then the spin relaxation lifetime diverges as @xmath169 this would be the case , for example , for a uniaxial or isotropic heisenberg ferromagnet . we follow the hydrodynamic approach of hohenberg and halperin @xcite for low energy dynamics of all these cases . consider first the case where the order parameter is not a conserved quantity . this could be the case , for example , for anisotropic magnets . such a case corresponds to model a of ref . , defined by the markoffian equations of motion @xmath170 @xmath171 is a gaussian white noise source,@xmath172 and @xmath173 are arbitrary external fields . the dynamic susceptibility for this model has the form@xmath174 the constant @xmath175 will in general depend on @xmath138 and @xmath176 but the relaxation rate @xmath177 as @xmath178 @xcite . at finite temperatures , the autocorrelation function will exhibit an exponential relaxation with time,@xmath179 the resistivity noise from magnetic fluctuations is of the random telegraph type : @xmath180 consider now the case where the order parameter is conserved ( model b in the parlance of ref . ) . this would be the case , for example , for a uniaxial or isotropic ferromagnet . the dynamical susceptibility has a form similar to eq . , except that now spin relaxation lifetime diverges as @xmath181 so that @xmath182 the rkky and double exchange models as well as the pure heisenberg model model fall in this category . for a pure heisenberg model , the damping mechanism is magnon - magnon scattering and one has @xmath183 . thus , in this case @xmath184 the autocorrelation function @xmath185 is given by @xmath186 for separations @xmath187 such that @xmath188 eq . has the following limiting behavior : @xmath189 equation ( ) should be contrasted with the lorentzian behavior of the autocorrelation function for random telegraph noise . it is also useful to study the time dependence of the autocorrelation function , @xmath190 at small separations @xmath187 such that @xmath191 we may ignore the bessel function in eq . and obtain @xmath192 next , we consider the rkky mechanism in a clean metal . in this case , the magnons can decay into particle - hole excitations for which one can show @xcite @xmath193 however , in the presence of impurities , one must take into account diffusion corrections @xcite which results in @xmath194 where @xmath195 and @xmath196 are the uniform spin susceptibility and spin diffusion constant respectively for the disordered system . note that this gives us the same results for the autocorrelation function as the pure heisenberg model with conserved spin dynamics . the mechanism of damping is however different - here it is magnon decay into incoherent particle - hole excitations . finally let us consider the double exchange model for which the spin wave life time is given by @xmath197 in two dimensions @xcite . for this model , we have @xmath198 the autocorrelation function @xmath185 is given by @xmath199 for small enough @xmath200 the autocorrelation function for the double exchange ferromagnet has the following limiting behavior @xmath201 thus , at long times , the autocorrelation function ( and consequently the resistivity noise ) for the 2d double exchange ferromagnet decays as @xmath202 this is to be contrasted with the @xmath203 decay for the heisenberg and disordered rkky ferromagnets . as we shall see in the following , the experimental data on resistivity noise is consistent with the rkky model and not the double exchange model . in sec . [ sub : resistivity - noise - for ] , we considered the effect of disorder on the dynamic susceptibility of the holes in the mn layer . we found that unlike the case of a clean metal where @xmath204 weak localization effects give us @xmath205 instead . thus , the magnetization dynamics of disordered rkky magnets incorporating weak localization effects and a purely heisenberg magnet are the same . in this section , we discuss the effects of disorder arising from randomness in the positions of the mn spins . we model such disorder in the form of site dilution and randomly varying exchange interaction , respectively , starting from a uniform heisenberg model . first , we consider a nearest - neighbor heisenberg model with a fraction @xmath206 of missing sites , which mimics random concentration of mn atoms . @xmath206 does not refer to the monolayer doping density of mn atoms ; rather , it refers to missing sites on an effective sublattice made of the mn atoms with a lattice constant of the order of the typical inter - mn separation . the exchange energy scale in the model is set by the value at typical mn - mn separations . a missing site can be related to regions with no mn atoms on the sublattice . for @xmath207 ( the critical threshold for percolation ) , the ground state is ferromagnetic and the low - energy spin dynamics is the same as that for the pure heisenberg model , albeit with a reduced zero - temperature spin - wave stiffness.@xcite . the temperature dependence of the magnetic correlation length for @xmath208 is thus @xmath209.$]@xcite in particular , for the 2d isotropic model on a square lattice , @xmath210 for @xmath211 @xcite and @xmath212 for @xmath213@xcite thus disorder due to site dilution will not affect the frequency and temperature dependences of noise for @xmath208 as long as one remains above the site percolation threshold on this sublattice . in actual systems , the long range of exchange interactions means that the percolation transition may not occur and one remains in the heisenberg universality class even for random concentration changes . next , we consider a nearest - neighbor heisenberg model with random exchange interactions chosen with zero mean . more realistically , the exchange interactions between pairs of spins will be a rapidly oscillating function of their separation . however , for the spin - glass state we are describing , the long - range nature of the interaction is not believed to play an important role and we therefore discuss the nearest - neighbor heisenberg model with couplings of random sign as opposed to the actual system with random site positions . the low - energy excitations of this model are gapless linearly - dispersing magnons@xcite ( as in a heisenberg antiferromagnet ) , and the lower critical dimension is two . the transverse susceptibility at low temperatures is expected to behave like @xmath214 as is the case for the heisenberg ferromagnet / antiferromagnet models . the dynamics of the spin - glass depends on whether one has dissipative equations of motion or the total spin is conserved.@xcite for dissipative dynamics , model a discussed above is evidently appropriate while for conserved dynamics , the spin dynamics is diffusive ( like in model b ) . @xcite to summarize , disorder in the form of site dilution ( while remaining above the percolation threshold ) as well as in the form of a rapidly oscillating function of the inter - mn separation does not qualitatively change the results for model a and model b dynamics discussed above for pure systems . finally , let us consider the case where the disordered magnetic system is in a griffiths phase . at a finite temperature , the system is organized into clusters that are weakly coupled to their neighbors but within the clusters , the spins are ferromagnetically ordered ( if the exchange coupling @xmath215 to a neighboring spin is less than the temperature @xmath216 we may regard the coupling to be weak ) . the long - time behavior of the autocorrelation function is dominated by the slow relaxation of atypically large , compact clusters . the relaxation of the cluster spin is through spin diffusion . the probability that a site @xmath110 belongs to a cluster of size @xmath57 is of the order of @xmath217 where @xmath218 is the magnetic correlation length , and @xmath219 is a constant . the autocorrelation function , @xmath220 is dominated by contributions from clusters of size @xmath221 whence @xcite @xmath222 the resistivity noise at low frequencies has the following behavior @xmath223 resistivity noise can arise from charge fluctuations due to inter - droplet hopping . the effect of inter - droplet hopping on the resistivity noise has been studied in the literature in the context of electronic phase separation in manganites @xcite , and we make a similar analysis for our droplet phase . we begin with the ground state consisting of a large number @xmath224 of neutral ( uncharged ) droplets . at finite temperatures , let @xmath225 be the number of neutral droplets and @xmath226 be the number of droplets with one extra ( less ) charge . evidently @xmath227 and @xmath228 let @xmath229 denote the charging energy of the droplets , and @xmath230 be the rate of escape of a hole from a charge - neutral droplet . in equilibrium , @xmath231 there are four elementary charge transfer processes - ( 1 ) from one neutral droplet to another , ( 2 ) from a droplet with one excess charge to another with one deficient charge , ( 3 ) from a droplet with one excess charge to a neutral droplet , and ( 4 ) from a neutral droplet to another with one deficient charge . the latter two processes do not involve a free energy cost . at low temperatures , we may ignore processes involving transfers of larger charges . the rates for processes ( 1 ) and ( 2 ) are , respectively , @xmath232 where @xmath233 is the distance between the two droplets . for processes ( 3 ) and ( 4 ) one has @xmath234 in the presence of magnetic correlations , we should replace @xmath230 by @xmath235 similar considerations also apply for the fluctuations of the number of charged and neutral droplets ; for example , for process ( 2 ) , @xmath236 where the summation is over all droplets with one deficient charge . the average fluctuation of @xmath237 is @xmath238 resistivity fluctuations are related to fluctuations of @xmath237 through the following relationship @xcite : @xmath239 this leads us to @xmath240 where the sum @xmath241 is over all pairs of droplets corresponding to process ( 2 ) . the dominant contribution is from relaxation to nearest neighbor droplets . this is especially true for the insulating samples where the inter - droplet distance @xmath242 retaining only the nearest neighbor contributions , the normalized resistivity noise due to droplet charge fluctuations takes the form @xmath243 where @xmath17 is the droplet coordination number and @xmath244}.\label{eq : tau2m}\end{aligned}\ ] ] the temperature dependence of @xmath245 is different from that of the resistivity [ see eq . ] but nevertheless shows an anomalous behavior around a temperature where @xmath246 in the low temperature limit where @xmath247 @xmath248 . in the high temperature regime , we have @xmath249 the measurements of the resistivity noise were performed with the sample , the structure of which is presented in fig . [ fig : structure ] . parameters of this sample are as follows : mn content is 0.35 ml , and it corresponds to @xmath250 cm@xmath96 surface concentration ; the thickness of the spacer separating mn @xmath0-layer and the qw is 3 nm ; indium concentration in the ga@xmath7in@xmath6as qw @xmath252=0.17 and so its depth is about 70 mev ; the hole density @xmath1 and their mobility @xmath253 in qw are @xmath254 cm@xmath96 ( 77 k ) or @xmath255 cm@xmath96 ( 5 k ) and 2370 @xmath256/v@xmath257s ( 77 k ) or 3400 @xmath256/v@xmath257s ( 5 k ) , respectively . as it was shown ref . , this sample is close to the metal - insulator transition of the percolation type having high enough value of resistance @xmath258 k@xmath259 and @xmath260 . the temperature dependence of its resistivity can be fitted to the arrhenius law with an activation energy @xmath261 k. ferromagnetic ordering is established for this sample through observations of the anomalous hall effect and resistivity anomaly ( peak at the temperature dependence of the resistance ) . the direct magnetic measurements for samples from the same set show magnetic hysteresis even for samples with smaller mn content . @xcite the temperature corresponding to the onset of strong ferromagnetic correlations for this sample was found to be about 30 k as measured by the resistivity anomaly . as it was shown previously @xcite , the mn @xmath0-layer consists of fm droplets affecting the magnetic state in the qw giving rise to a set of metallic fm droplets , which is responsible for metal - insulator transition . so , there are two characteristic temperatures : first that corresponds to the fm ordering inside droplets and the second related to the formation of the long - range fm state in the sample , which can be treated as a kind of the `` fm percolation transition '' . the characteristic temperature found from the resistive anomaly corresponds to the second case . at this temperature magnetic correlations became significant . the data ( see fig . [ fig : freq - dep ] ) on the frequency dependence of noise show @xmath262 ( @xmath263 ) as a function of frequency for a number of temperatures ( mostly below the resistivity peak ) . the noise is non - monotonous at all these temperatures , and especially so at lower temperatures . the curves at higher temperatures suggest more than one relaxation time while the low temperature data indicates a single relaxation time . ( color online ) plots of the measured frequency dependence of @xmath262 for various values of the temperature . ] the data in fig . [ fig : freq - dep ] correspond to temperatures well below the resistivity anomaly . this is the regime where ferromagnetic correlations are significant . note that the resistivity noise amplitude decreases with increase of temperature for the high frequency as well as the low - frequency limits . thus , we try power - law fits in the low- and high - frequency range away from the peak in the resistivity noise ( see fig . [ fig : freq - dep-4k ] ) . this behavior is consistent with both the magnetic and charge fluctuation models . we will discuss the temperature dependence of resistivity noise later in sec . [ sub : temperature - dependence ] . we consider power - law fits at the low- and high - frequency ends . at high frequencies , we find ( see fig . [ fig : freq - dep-4k ] ) for example for the @xmath264 data , @xmath265 which should be contrasted with a @xmath266 decrease expected for a lorentzian . at low frequencies , one can find a fit to @xmath267 or @xmath268 the former would be consistent with a lorentzian ( no conserved order parameter ) while the logarithmic behavior in the latter can arise from model b dynamics ( conserved order parameter ) . ( color online ) frequency dependence of noise at @xmath264 ( solid curve ) together with fits to the low and high frequency regimes . at the low - frequency end , the dashed curve and the dotted curve are fits to @xmath269 and @xmath270 respectively . at the high - frequency end , the fit is to @xmath271 by analyzing the long - time behavior of @xmath272 we find that the agreement is better with model b dynamics . [ fig : griffiths4.4k ] shows the time dependence of the resistivity autocorrelation function numerically obtained by a discrete fourier transform of the frequency data . the long - time behavior shows a good fit to @xmath273 for disordered rkky ferromagnets , one expects @xmath274 while for the double exchange case , @xmath275 we found that fits to exponential relaxation , @xmath276 or a griffiths relaxation , @xmath277 are not as good as the power - law fits . the @xmath203 power - law relaxation supports the case of the rkky ferromagnet . ( color online ) plot showing the long time dependence of the resistivity autocorrelation function @xmath278 extracted from the resistivity noise data at @xmath279 ( solid curve ) together with fits to a power - law time dependence with logarithmic corrections . the dashed curve is a fit to @xmath280 while the dot - dashed curve is a fit to @xmath281 in two dimensions , @xmath282 behavior is expected for a disordered rkky ferromagnet ( see eq . ) and @xmath283 for double exchange ferromagnets ( see eq . ) . the logarithmic time dependence indicates @xmath284 noise contributions . the fit to the rkky model is better than to the double exchange . ] time traces of resistivity fluctuations showing distinct telegraph - like switching at low temperatures . at temperatures higher than around 20k , the telegraph - like switching is not so prominent . ] in fig . [ fig : telegraph - noise ] , the time traces of resistivity fluctuations are shown , while the main stable contribution to the resistivity is subtracted . note that the time traces of the resistivity show distinct telegraph - like switching at low temperature . this switching is not so prominent once the temperature exceeds about 200 k : this is near to but below the temperature , at which one observes the anomalous peak in the resistivity and noise . the advent of telegraph - like switching can be understood if the magnetic model has uniaxial symmetry instead of complete rotational symmetry . this is possible given the planar geometry of the structure . nevertheless , at the moment we are unable to say whether at still longer times , the autocorrelation relaxation will remain a power law or become exponential . the logarithmic term indicates @xmath284 noise , possibly from other mechanisms . figure [ fig : temp - dep ] shows the temperature dependence of the resistivity noise at two difference frequencies . the temperature range shown in in the figure corresponds to the low frequency regime for both @xmath285 and @xmath286 hz . this can be deduced from fig . [ fig : freq - dep ] where the crossover frequency ( between the low- and high - frequency regimes ) for @xmath287 k is around 200 hz , and the data in fig . [ fig : temp - dep ] are taken at @xmath288 k. we analyze the temperature dependence in the context of the magnetic and charge fluctuation models we have developed above . for models a and b as well as for the griffiths picture , the temperature dependence of the normalized resistivity noise is directly related to the temperature dependence of the magnetic correlation length @xmath289 as well as the coefficients @xmath175 ( model a ) and @xmath290 ( model b ) . in all cases , @xmath291 we do not yet know the contribution to @xmath175 and @xmath290 from the change of hole resistivity with temperature . suppose the holes in the quantum well do contribute to @xmath175 and @xmath292 in this case , as the temperature is reduced , @xmath175 and @xmath290 can show a decrease mirroring the resistivity , and that it has an anomaly when the correlation length becomes comparable with the inter - droplet separation . ( color online ) temperature dependence of the resistivity noise @xmath293 measured at @xmath285 hz ( solid curve ) and @xmath286 hz ( dashed curve ) . ] for models a and b in the low - frequency regime , in accordance with eqs . and , we fit the temperature dependence data in fig . [ fig : temp - dep ] to @xmath294 where @xmath295 is a constant . the exponential temperature dependence of @xmath159 and @xmath138 together with the scatter in the data makes it difficult to extract the sub - leading power - laws and logarithms ; thus , we do not attempt to distinguish models a and b here . for the charge fluctuation model , following eq . , we tried fitting the data to @xmath296 where @xmath297 is a constant . we consider @xmath3 and @xmath298 as fitting parameters . we could not obtain a satisfactory fit for the charge fluctuation result for any positive value of @xmath299 ( color online ) temperature dependence of the resistivity noise @xmath293 for @xmath286 hz ( dots ) and a fit ( solid curve ) to our model of magnetic fluctuations , @xmath300 the fit corresponds to @xmath301 k. the model does not take into account changes in the damping parameters @xmath175 or @xmath290 across the resistivity anomaly . ] [ fig : tempdep - fit ] shows the resistivity noise data for @xmath286 hz and a fit to our model of magnetic fluctuations . possible variation of @xmath175 or @xmath290 across the resistivity anomaly is not taken into account in the fit . from this fitting we have got @xmath301 k that by two times exceeds the temperature of the resistance anomaly and maximum of @xmath293 versus temperature , while it is in a good agreement with @xmath302 k , which was found from fitting of the resistivity temperature dependence . @xcite that additionally proves the earlier result that there are two characteristic temperatures in such systems : the local curie temperature at which ferromagnetic ordering occurs inside the magnetic droplets and the global curie temperature of the long - range ferromagnetic ordering establishment at the scale of the whole sample `` magnetic percolation transition '' . we experimentally studied and analyzed the resistivity noise in 2d dms structures . we have shown how noise measurements can be used to probe the microscopic dynamics governing ferromagnetism in dms heterostructures . for insulating dms systems , we obtained a relation between the resistivity noise and the spin autocorrelation function which enables us to probe magnetization dynamics through transport measurements . we studied a number of microscopic models describing ferromagnetism in the mn @xmath0-layer motivated by the hydrodynamic approach of hohenberg and halperin.@xcite the models studied fall into two broad classes : ( a ) those where the order parameter is not a conserved quantity ( model a ) such as ising magnets and ( b ) those where the order parameter is conserved ( model b ) such as uniaxial heisenberg ferromagnets , rkky ferromagnets , and double - exchange ferromagnets . model b dynamics , where observed , can be very useful in understanding the microscopic origin of ferromagnetism . in particular , we showed that resistivity noise arising from magnetization fluctuations of rkky and double - exchange ferromagnets are qualitatively different and can thus be used to address the long standing question of whether ferromagnetism in dms systems arises from an rkky or a double exchange mechanism . we also analyzed the effect of disorder on the magnetization dynamics . our results are summarized as follows . \(a ) the resistivity noise for uniaxial heisenberg ferromagnets as well as disordered rkky ferromagnets decreases as @xmath303 for times long compared with the time scale @xmath304 of spin diffusion over the magnetic correlation length . in contrast , @xmath305 for a double exchange ferromagnet . these cases all involve a conserved order parameter . when the order parameter is not conserved , such as in an ising ferromagnet ( or an antiferromagnet ) , the resistivity noise has a random - telegraph behavior and follows @xmath306 at long times . the differences arise from the momentum dependence of the damping rate @xmath307 of magnetic fluctuations . @xmath308 for the ising case ; @xmath309 for uniaxial heisenberg ferromagnets and disordered rkky ferromagnets ; and @xmath310 for the 2d double exchange case . \(b ) magnetic disorder in the form of site dilution and random sign of nearest neighbor exchange interaction was considered . as long as one remains above the classical percolation threshold , site dilution was seen to degrade the curie temperature and spin stiffness but otherwise retain the same dynamics as the undiluted case . this is in effect a confirmation of the validity of the harris criterion in our systems . when the sign of inter - impurity exchange interaction varies randomly , as may be the case at low mn density together with an rkky interaction of the mn atoms , the ground state is a spin glass instead of a ferromagnet . it was shown that the damping rate of magnetic fluctuations for this case also goes as @xmath311 \(c ) fluctuations of charge in the puddles in the quantum well result in a random telegraph noise of the form @xmath312 at long times . in terms of frequency , this is a lorentzian behavior that has been obtained in sec . [ sec : charge - fl ] and discussed in relation to the experimental data in sec . [ sub : freq - time - dep ] . \(d ) our experimental data indicate that @xmath303 , which is evidence in favor of a disordered rkky ferromagnet for these samples . however , as we have discussed in sec . [ sec : proximity ] and sec . [ sec : models ] , the rkky behavior is not universal , and depends on sample parameters such as the ratio of the inter - hole distance and the bohr radius in the mn layer . the temperature dependence of resistivity noise qualitatively agrees with the theory although the agreement is not as good as the frequency or time dependence . \(e ) our method of probing the microscopic interactions governing ferromagnetism is not specific to 2d dms structures and can prove useful in the study of other ferromagnetic systems . we are grateful to d. dhar , a.m. finkelstein , d.e . khmelnitskii , e.z . meilikhov , p. mahadevan and t.v . ramakrishnan for helpful discussions . the work was supported by the russian foundation for basic research , projects nos . 09 - 02 - 92675-ind , 11 - 02 - 00363 , and 11 - 02 - 00708 , and by the indo - russian collaboration program , grant no . int / rfbr / p-49 . we also acknowledge the partial support of euromagnet under eu contract no . rii3-ct-2004 - 506239 . v.t . and k.d . acknowledge the support of tifr . k. sato , l. bergqvist , j. kudrnovsk , p.h . dederichs , o. eriksson , i. turek , b. sanyal , g. bouzerar , h. katayama - yoshida , v.a . dinh , t. fukushima , h. kizaki , and r. zeller , rev . phys . * 82 * , 1633 ( 2010 ) . aronzon , m.a . pankov , v.v . rylkov , e.z . meilikhov , a.s . lagutin , e.m . pashaev , m.a . chuev , v.v . kvardakov , i.a . likhachev , o.v . vihrova , a.v . lashkul , e. lhderanta , a.s . vedeneev , and p. kervalishvili , j. appl . phys . * 107 * , 023905 ( 2010 ) . aronzon , v.a . kulbachinskii , p.v . gurin , a.b . davydov , v.v . rylkov , a.b . granovskii , o.v . vikhrova , yu.a . danilov , b.n . zvonkov , y. horikoshi , and k. onomitsu , pisma zh . . fiz . * 85 * , 32 ( 2007 ) [ jetp lett . * 85 * , 27 ( 2007 ) ] . vasileva , yu.a . danilov , ant.a . ershov , b.n . zvonkov , e.a . uskova , a.b . davydov , b.a . aronzon , s.v . gudenko , v.v . rylkov , a.b . granovsky , e.a . ganshina , n.s . perov , and a.n . vinogradov , fiz . . poluprovod . * 39 * , 87 ( 2005 ) [ semiconductors * 39 * , 77 ( 2005 ) ] .

the power of jets from black holes are expected to depend on both the spin of the black hole and the structure of the accretion disk in the region of the last stable orbit . we investigate these dependencies using two different physical models for the jet power : the classical blandford - znajek ( bz ) model and a hybrid model developed by meier . in the bz case , the jets are powered by magnetic fields directly threading the spinning black hole while in the hybrid model , the jet energy is extracted from both the accretion disk as well as the black hole via magnetic fields anchored to the accretion flow inside and outside the hole s ergosphere . the hybrid model takes advantage of the strengths of both the blandford - payne and bz mechanisms , while avoiding the more controversial features of the latter . we develop these models more fully to account for general relativistic effects and to focus on advection - dominated accretion flows ( adaf ) for which the jet power is expected to be a significant fraction of the accreted rest mass energy . we apply the models to elliptical galaxies , in order to see if these models can explain the observed correlation between the bondi accretion rates and the total jet powers . for typical values of the disk viscosity parameter and mass accretion rates consistent with adaf model expectations , we find that the observed correlation requires ; i.e. , it implies that the black holes are rapidly spinning . our results suggest that the central black holes in the cores of clusters of galaxies must be rapidly rotating in order to drive jets powerful enough to heat the intracluster medium and quench cooling flows . [ firstpage ] accretion , accretion disks black hole physics galaxies : active galaxies : jets x - rays : galaxies mhd

astro-ph0612354

recent high resolution _ chandra _ observations of the cores of elliptical galaxies provide dramatic illustration of the impact of jets launched from active galactic nuclei ( agn ) on the interstellar medium of their host galaxies and/or on the intracluster medium if the galaxies reside in clusters . these observations have revealed prominent x - ray surface brightness depressions corresponding to cavities or bubbles inflated by the jets as they interact with the surrounding hot gas ( e.g. , @xcite , hereafter a06 ; @xcite ) . this interaction is presumed to deposit large amounts of energy in their environments ( e.g. , @xcite ) , modifying the hierarchy of galaxy formation ( e.g. , @xcite ) and altering the evolution of the intracluster medium by counteracting radiative losses ( e.g. , @xcite ) . based on the energetics of creating the observed cavities , the minimum energy associated with the jets range from @xmath2 ergs in galaxies , groups , and poor clusters to @xmath3 ergs in rich clusters , ranking the outbursts as among the most powerful phenomena in the universe while a detailed understanding of the extragalactic agn jet phenomena remains elusive , the combination of physically insightful analytic studies ( e.g. , @xcite ) and sophisticated general relativistic , magnetohydrodynamic ( mhd ) numerical simulations ( e.g. , @xcite ) are beginning to yield important insights . there is now a general consensus that jets are fundamentally mhd events . the currently favoured models presuppose an accretion flow threaded by large - scale magnetic fields flowing onto a supermassive black hole . in the neighborhood of the black hole , the flow settles into a disk - like structure , in which the rate of inward flow of matter depends on the efficiency with which its angular momentum can be transferred outward . this transfer of angular momentum is mediated by mhd turbulence . both mhd turbulence and differential rotation of the plasma in the body of the disk generate and intensify toroidal magnetic fields @xcite . when the pressure associated with the toroidal fields grows strong enough , the field lines escape from the disk forming a rotating helical tower of field lines above and below the disk . centrifugal forces associated with this rotating magnetic field helix drives any plasma trapped onto the field lines upward and out of the disk , generating outflows . this is the crux of the blandford - payne @xcite model for jets , wherein the magnetic fields extract energy from the rotation of accretion disk itself to power the outflows . depending on the detailed structure of the magnetic fields and the accretion disk , this mechanism is expected to generate outflows ranging from broad , uncollimated winds to highly collimated jets . in the event that the black hole itself is spinning , the magnetic field can also extract the rotational energy of the central black hole to power the outflows . in the blandford - znajek ( bz ) model @xcite , the magnetic fields are assumed to be connected directly to the horizon of a spinning black hole while in the @xcite model , the magnetic fields associated with jet production are anchored to the inflowing plasma inside the black hole s ergosphere . in both these models , the dragging of inertial frames , relative to an observer at infinity , within the ergosphere of the rotating black hole results in a rotating , tightly wound vertical tower of field lines , and hence powerful outflows . moreover , @xcite has shown that in the punsly - coroniti - like models , the differential dragging of the frames will also act as a dynamo to amplify the magnetic field at the expense of the black hole s rotational energy and that this will have the effect of further enhancing the jet power . the results of numerical simulation studies @xcite are , in a broad brush sense , consistent with the bz and punsly - coroniti - meier type of models , indicating that the power of jets depends on both the mass accretion rate as well as the spin of the black hole . in fact , @xcite argues that understanding the radio - loud and radio - quiet dichotomy in the qso population requires the jet power to exhibit such dependencies . and while this assertion appears to be indicated by observations of black hole candidate systems in our galaxy @xcite , very little is known about the prevailing conditions underlying the jet phenomena in extragalactic agns . the recent _ chandra _ observations of bubbles , however , offer a unique opportunity to remedy this . the x - ray observations not only provide an estimate of the power of the jets emitted by the agns but also the rate at which matter is accreting onto the black hole - accretion flow system . recently @xcite ( hereafter a06 ) analyzed the data for nine nearby , x - ray luminous elliptical galaxies and found a remarkably tight correlation between the bondi accretion rates and the jet powers . within the context of viable models for the jet - black hole - accretion flow system , such a correlation not only provides insight into the efficiency with which the rest energy of the material accreting onto the black hole is converted into jet energy but also the spin distribution of the black holes powering agn . in this paper , we seek to shed light on these issues . we begin by considering two physical models for the jet power : the bz model as described above , and a hybrid model proposed by @xcite ( see also , @xcite ) , in which the magnetic field threads the plasma throughout in the body of the disk where frame - dragging is negligible as well as within the ergosphere . this latter model combines the blandford - payne - like disk acceleration mechanism with the blandford - znajek - like scheme that draws upon the rotational energy of the black hole . we improve upon these models by incorporating important general relativistic effects that previously were introduced via highly restricted approximation . following @xcite , we further couple our models for the jet production to the advection - dominated accretion flow model ( hereafter adaf , @xcite , hereafter n98 ; @xcite ) . there is a considerable body of work indicating that the launching of jets is most efficient when the accretion flow is advection - dominated ( e.g. , @xcite ) and that jet production is suppressed in the standard thin accretion disk normally associated with radiatively efficient agns @xcite . this paper is organized as follows : in [ sec : model ] , we describe our models for the accretion flow and the jet power . we summarize the findings of a06 in [ sec : correl ] . in [ sec : constr ] , we compare the correlations between the jet power and accretion rate implied by our models with the results of a06 and consider the resulting implications for jet efficiencies and the distribution of black hole spins . we also compare results for accretion models with and without winds and explore the implications of uncertainties in a06 estimates of jet power and mass accretion rates on our findings . lastly , we highlight our main results in [ sec : conclusion ] and offer some concluding remarks . there are several lines of evidence that suggest that the accretion flow onto agns with strong jets is best described as an adaf : first , the observed bolometric ( radiative ) luminosities of radio - loud agns are typically many orders of magnitude smaller than the luminosities expected if the mass were flowing onto the central black holes along thin accretion disks ( e.g. , @xcite ) ; in the latter case , the bolometric luminosities correspond to roughly 10% of the rest - mass energy of the accreting matter @xcite . adafs can not only account for the observed radiative quiescence of the jet - emitting agns but also reproduce their nuclear x - ray luminosities with accretion disk mass flow rates comparable to their bondi rates @xcite . the exceptional systems ( such as the radio - loud quasars ) that are both highly luminous and radio - loud are currently an intriguing puzzle : possibly they are high accretion rate systems in which the disk is puffed up by the trapped energy despite the system s low opacity ( e.g. , @xcite ) . second , the magnitude and the structure of the magnetic fields associated with adafs are much more conductive to the extraction of spin energy from the hole than those associated with standard thin disks ( e.g. , @xcite ) . third , x - ray binaries with black hole candidates ( xrbs ) in the `` low / hard '' state display a strong correlation between the x - ray and radio emissions , which can best be understood in the context of an adaf - jet system @xcite . low - luminosity agns with jets show similar correlations and hence , are thought to be scaled - up versions of the galactic xrbs @xcite . one characteristic of the adaf - jet system is low mass flow rates . the observations suggest that at high flow rates , the black hole systems switch to the `` high / soft '' state and the jet activity is suppressed @xcite . this observed `` quenching '' of the jet ( as well as the transition from the `` low / hard '' to the `` high / soft '' ) is believed to be due to a change from an adaf ( powerful jets ) at low mass flow rates to a thin accretion disk structure ( weak jets ) at high mass flow rates ( e.g. , @xcite ) . we describe the structure of the adaf using the self - similar equations of @xcite . we are especially interested in the analytical equations ( see appendix i ) that describe the vertical half - thickness of the disk @xmath4 , the angular velocity of the disk @xmath5 and the magnetic field strength @xmath6 near the black hole in terms of radius @xmath7 , black hole mass @xmath8 , accretion rate onto the black hole @xmath9 and advection parameter @xmath10 ( assumed @xmath11 ) . we require these as inputs to our models for the jet power described in [ sec : bz ] and [ sec : disk ] below . we note that although the @xcite equations are based on an early model for the adaf that did not allow for mass loss in the form of winds from the accretion flow , we have verified that these self - similar solutions provide a good approximation to the structure of the inner regions of adaf models , like that of @xcite ( hereafter bb99 ) , that do allow for such losses . the equations for @xmath4 , @xmath6 and @xmath5 also depend on the properties of the accreting fluid , such as its adiabatic index @xmath12 , its viscosity parameter @xmath13 and the ratio of gas to magnetic pressure @xmath14 in the fluid . these quantities are not independent . the value of @xmath12 depends on @xmath14 via the relationship @xmath15 @xcite and based on the mhd numerical simulations of the evolution of the magnetorotational instability in accretion disks @xcite , @xmath16 . we assume that the magnetic pressure is related to the field strength as @xmath17 . we constrain the value of @xmath13 from recent mhd numerical simulations of radiatively inefficient accretion flows , which take into account self - consistently the role of the maxwell stresses in establishing its value . the general relativistic simulations ( e.g. , @xcite ) find that in the inner portions of the disk @xmath18 . although @xmath13 is not explicitly quoted , we can estimate @xmath13 from these values of @xmath14 using the relationship between the two , obtaining @xmath19 . for computational purposes , we will derive results for @xmath20 and @xmath21 . to put these @xmath13-values in context , we note that the simulations around schwarzschild holes of @xcite suggest that near the innermost stable circular orbit , @xmath13 reaches high values : @xmath22 . moreover , recent adaf models of xrbs require values of @xmath23 in order to account for the observations @xcite . to the above model for the adaf , we introduce three important modifications to take into account general relativistic effects induced by the kerr metric : first , we take as input for our jet model the values of @xmath4 , @xmath6 and @xmath5 evaluated at @xmath24 , the radius of the marginally stable orbit of the accretion disk . this radius depends sensitively on the dimensionless black hole spin parameter @xmath25 ( `` @xmath26 '' in geometrized units , @xmath27 is the specific angular momentum ) , where @xmath28 is the angular momentum of the hole and @xmath29 is the maximal angular momentum @xcite . second , an observer at infinity will see the disk and the magnetic fields near the black hole rotate , in the boyer - lindquist coordinate system , not with an angular velocity @xmath5 but @xmath30 , where @xmath31 is the angular velocity , in the same coordinate system , corresponding to the local spacetime rotation enforced by the spinning black hole @xcite . and third , we take into account the field - enhancing shear caused by frame - dragging when calculating the magnetic field strength in the inner region of the disk , as first suggested by @xcite and tentatively observed in recent mhd numerical simulations @xcite . following @xcite , we relate the amplified , azimuthal component of the magnetic field to the unamplified magnetic field strength derived from the self - similar adaf solution as @xmath32 , where @xmath33 is the field - enhancing factor . the amplitude of this factor depends on the black hole spin through the angular velocity of spacetime rotation , @xmath34 . in the case of a non - rotating black hole , @xmath35 and @xmath36 ( i.e. , no field enhancement ) . the potential importance of these effects has been recognized previously and @xcite even incorporated them into his jet power model , albeit in the form of simplifications that blunted their impact . for example , he took @xmath37 to be a free parameter rather than explicitly relate it to spacetime rotation induced by the rotating black hole , he adopted a simple approximation for @xmath34 , and he evaluated his jet model only for two limiting radii : @xmath38 ( corresponding to @xmath39 ) and @xmath40 ( @xmath41 ) . in our approach , we are able to capture more fully the dependence of the important model parameters on @xmath42 and by doing so , are able to explore the sensitivity of the accretion flow and the jet power solutions to variations in the value of the black hole spin parameter . finally , we follow @xcite in assuming that the poloidal and azimuthal components of the magnetic field are related to each other as @xmath43 where @xmath7 is the radius in cylindrical coordinates , a relationship based on the assertion that the strength of the poloidal field is limited by the vertical extent of turbulent eddies in the disk . in the case of an adaf , this yields @xmath44 because @xmath45 . according to the bz model , the total power of the resulting jet is given by ( e.g. , @xcite ) : @xmath46 where @xmath47 g m_\bullet / c^2 $ ] is the horizon radius , @xmath48 is the strength of the magnetic field normal to the horizon , and the factor @xmath49 depends on the angular velocity of the field lines @xmath50 relative to that of the hole , @xmath51 . following the usual practice , we assume that @xmath52 , which maximizes the power output ( e.g. , @xcite ) . as for @xmath48 , @xcite has argued that the field threading the horizon ought to be comparable in strength to the field threading the inner regions of the accretion flow . hence , we take the field strength at the horizon to be the same as that at the radius of the marginally stable orbit of the accretion disk ; that is , @xmath53 . these field configurations are consistent with those seen in the in the jet launch regions in numerical simulations of accretion onto kerr black holes ( e.g. , @xcite ) . finally since the accretion rate enters the jet power only through the field strength , the jet power only depends on the accretion rate measured at the marginally stable orbit of the accretion flow @xmath54 . if the mass is conserved in the disk as in the early version of the adaf model ( @xcite , n98 ) then @xmath9 is constant with radius , but contemporary adaf models often allow for mass loss from the accretion disk in the form of winds and outflows ( see [ sec : power ] ) . the primary advantage of the bz model is its simplicity . the picture describes a black hole whose event horizon is equivalent to a rotating conducting surface with surface charges and currents . in effect , the horizon is a unipolar inductor onto which the magnetic fields are attached . the magnetic fields are torqued by the rotation of the black hole and in turn , drive an outflow . there are , however , concerns about the model . @xcite have argued that the very notion of flux being emitted from the event horizon renders the model unphysical because the event horizon is casually disconnected from events upstream . disturbances generated at the event horizon can only propagate inward . moreover , the bz model is also formulated on the assumption of zero accretion that underlies the bz model ; the bz model is derived in the limit of vanishing plasma density and pressure @xcite . on the other hand , @xcite have reported good agreement between several aspects of the bz model and their numerical simulation results , and for this reason , we consider this model in the present work . we do note , however , that @xcite also reported a growing difference between the model and their results with increasing black hole spins , a result that can partly be attributed to the fact that the bz model is derived in the limit of slowly rotating black holes . contrary to the premise of very low plasma densities that underlies the bz model , accretion of matter is presumed to be a key element of real agn systems . in fact , numerical simulations show that the coupling between the accretion flow and the magnetic fields is an essential element of jet production . for this reason , we consider a second jet model , the hybrid model of @xcite . as noted previously , this model is constructed so that large - scale magnetic fields thread the accretion disk outside the ergosphere as well as rotating plasma within the ergosphere that is flowing onto the black hole . hence , the model is able to draw upon both the rotational energy of the accretion disk as well as the spinning black hole in order to drive outflows , though the extraction of the black hole rotational energy occurs indirectly through field lines anchored to the plasma subject to spacetime rotation imposed by the black hole . this model also takes into account the effects of field amplification by both the differential rotation of the plasma in the body of the disk and the differential frame - dragging . as discussed by @xcite , the jet power in this model is a strong function of the thickness of the accretion disk and the black hole spin ; strong magnetic fields and rapid rotation , the necessary ingredients for the launching of powerful jets , only arise when the disk is thick and the hole is spinning rapidly . following @xcite , the total jet power for the hybrid model is given by @xmath55 where @xmath32 and @xmath30 . all quantities are evaluated at @xmath56 , which is also assumed to be the approximate characteristic size of the jet - formation region . we note that we denote the jet power for this model with the superscript `` disk '' in order to highlight that in addition to drawing upon the rotational energy of the spinning black hole , this model also draws energy from the accretion disk . as a matter of interest , we note that there is considerable support for the punsly - coroniti - meier component of the hybrid model from recent numerical simulations studies @xcite . on the other hand , the relevance of the blandford - payne mechanism , which describes the extraction of jet power from the rotation of the accretion disk and which is incorporated in our hybrid model , is very much a matter of debate . for example , @xcite have recently argued that this mechanism is not at all important for understanding the collimated jet outflows seen in their numerical simulations while @xcite has suggested that such conclusions are premature and that the simulation results are dependent on the initial conditions adopted . there are several interesting properties of our jet models that ought to be highlighted . first , the jet power for neither the bz nor the hybrid models ( equations [ eq : powerbz ] and [ eq : powerdisk ] , respectively ) depends on the black hole mass . moreover , the dependence on the accretion rate enters the jet power only through the field strength ( see appendix i ) and for the combined accretion flow - jet power model under consideration here , the jet powers for both models can be expressed as @xmath57 ; that is , the jet power depends linearly on the mass accretion rate . however , both have a complicated dependence on @xmath42 and @xmath13 ( see appendix i ) . figure [ fig : newpower ] illustrates the spin dependence of both the bz and the hybrid @xmath58 models , for viscosity parameters @xmath20 and @xmath21 , and a fiducial value of @xmath59 . this corresponds to one tenth of the mean bondi accretion rate of the elliptical galaxies studied by a06 . ) and the hydrid model ( equation [ eq : powerdisk ] ) as a function of black hole spin parameter @xmath42 . the curves are computed assuming a fiducial value of @xmath59 , and for two different values of the viscosity parameter : @xmath20 and @xmath21 . the right axis shows the jet efficiency @xmath60 . since the jet powers scale linearly with @xmath61 , @xmath62 depends only on @xmath42 and @xmath13 . ] as figure [ fig : newpower ] shows , both the bz and the hybrid @xmath58 models behave similarly and give comparable results for intermediate and high values of @xmath42 . the slight difference in the jet power between the two models is not significant given the uncertainty in our understanding of the detailed mechanisms underlying jet formation . at low spin values , however , the jet power in the hybrid model is considerably higher than that for the bz case . the drop - off in the bz case reflects the decline in the black hole rotational energy available to power the jet while in the hybrid case , the decline is limited by contributions from the accretion disk . the most important point made by figure [ fig : newpower ] is that the jet power for both the bz and the hybrid models is a strong function of the black hole spin , spanning a range of three orders of magnitude as the black hole spin varies from 0 to 1 . this strong dependence is largely due to our improvements over previous models , which did not incorporate carefully the physics of the kerr metric @xcite . this type of @xmath42-dependence for the jet power has previously only been seen in complex numerical mhd simulations of jet formation @xcite . the right axis of figure [ fig : newpower ] shows the jet efficiency factor , defined as @xmath63 . since our model jet powers scale linearly with @xmath64 , the corresponding jet efficiency factor depends only on the black hole spin parameter @xmath42 and the viscosity parameter @xmath13 . for high values of @xmath42 , the efficiencies predicted by our two jet models are comparable to the novikov - thorne thin disk radiative efficiency , which is related to the binding energy of the innermost stable circular orbit @xcite , and the jet carries away a significant fraction of the rest mass energy flowing through the disc . we note that for the maximal black hole spin , @xmath65 , as implied by the analysis of @xcite , the jet efficiencies are @xmath66 ( bz model ) and @xmath67 ( hybrid model ) for @xmath20 . on the other hand , if @xmath68 , the efficiencies drop to @xmath69 ( bz model ) and @xmath70 ( hybrid model ) . given a distribution of gas about a central black hole , the most simple configuration describing the accretion of the gas onto the the black hole is the bondi flow model @xcite , which assumes a non - luminous central source and a spherically symmetric flow with negligible angular momentum . the resulting bondi accretion rate can be written as @xmath71 , where @xmath72 is the accretion radius , @xmath73 is the gravitational constant , @xmath8 is the black hole mass , @xmath74 is the sound speed of the gas at @xmath75 , @xmath76 is the density of gas at @xmath75 and @xmath77 is a numerical coefficient that depends on the adiabatic index of the gas . this estimate is frequently used in studies of the central x - ray emitting gas in elliptical galaxies ( e.g. , @xcite ) . recently , a06 derived the bondi accretion rates and jet powers from _ chandra _ x - ray observations of nine nearby , x - ray luminous giant elliptical galaxies that show evidence of jet - inflated cavities in their central regions , and found a tight correlation between the two . the jet powers were estimated from the energies and timescales required to inflate cavities observed in the surrounding x - ray emitting gas such that @xmath78 , where @xmath79 is the energy required to create the observed bubbles and @xmath80 is the age of the bubble . a06 estimates are based on the assumption that the x - ray bubbles are inflated slowly . the values of @xmath81 for the systems were calculated from the observed gas temperature and density profiles . in most cases , the bondi radius is not observed directly , and the appropriate density and temperature are determined by extrapolating the observed data , typically by a factor of 3 or greater , in radius ( cf . , @xcite ) . the black hole masses were deduced from the optical velocity dispersion measurements using the correlation between central black hole mass and velocity dispersion of @xcite . since a06 presents a state of the art analysis of current data , we accept the estimates of the bondi accretion rate and jet power at face value in what follows . clearly it will be important to see how these observational constraints improve with future x - ray missions . in section [ sec : rapidexpansion ] , we consider the impact of uncertainties in the calculation of the jet power from the physical properties of cavities observed in the chandra data . the a06 results for the nine systems are shown in figure [ fig : comp ] , where @xmath82 is the total accretion power released for an efficiency of 10% , @xmath83 . the error bars plotted include a systematic uncertainty of 0.46 dex in @xmath84 , implied by the intrinsic dispersion of 0.23 dex in the @xmath85 relation ( see a06 ) . the correlation between @xmath86 ( or equivalently , @xmath82 ) and jet power @xmath58 was expressed by a06 as a power - law model of the form @xmath87 with @xmath88 and @xmath89 . this power - law model is shown in figure [ fig : comp ] as a solid line . ) and jet power ( @xmath58 ) for nine nearby , x - ray luminous giant elliptical galaxies derived by a06 . the error bars plotted include a systematic uncertainty of 0.46 dex in @xmath84 . the fitted power - law model predicted by our jet models ( @xmath90 and @xmath91 ) is represented by the dashed ( best - fit ) and dotted ( error bars ) lines ; the best - fit power - law model determined by a06 is shown as the solid line . ] the a06 results are interesting in two regards : first , they show that the central agns in these systems are extremely sub - eddington ; that is , they have extremely low accretion rates compared to the eddington rate , @xmath92 , where @xmath93 . our adaf model for the accretion flow is only valid for such highly sub - eddington flows . second , as noted by a06 , the correlation implies that a non - negligible fraction ( @xmath94 for @xmath95 ) of the energy associated with the rest mass of the gas entering @xmath75 is channeled into jet power . presumably less matter than the amount predicted by the bondi rate gets down to the black hole for several reasons . as the gas in an adaf has angular momentum , accretion is driven not just by gravity as in the bondi flow but also by the rate of angular momentum transport characterized by the parameter @xmath13 in our adaf model ; given a certain ambient density in the external medium , the accretion rate predicted by the adaf model is lower than the bondi accretion rate : @xmath96 ( e.g. , n98 ; @xcite ) , with @xmath97 . furthermore , some part of the gas may be prevented from being accreted due to winds and/or convection ( e.g. , bb99 ; @xcite ) occuring in the adaf , reducing even more @xmath61 compared to @xmath86 . we allow @xmath61 to be smaller than @xmath86 by introducing the parameter @xmath98 such that @xmath99 . the parameter @xmath98 represents the fraction of material supplied by the external medium at the bondi radius that ultimately reaches the innermost stable circular orbit of the accretion flow and gets accreted afterwards . therefore this parameter encompasses our ignorance about the possible physical processes that may modify the density profile of the accretion flow with respect to the adaf solution . we note that here we are only considering the simplest situation where the density profile is affected . winds too will reduce the rate of energy and angular momentum accretion onto the central black hole . we defer the discussion of such complications to [ sec : winds ] . including this modification into our models gives jet power relations of the form @xmath100 this relationship has the same functional form as the correlation found by a06 ( equation [ eq : powerlaw ] ) and we can now use the above equation to constrain the main model parameters ( @xmath13 , @xmath98 and @xmath42 ) using a06 results . comparing equations [ eq : powerlaw ] and [ eq : power ] , it follows that our models predicts the slope @xmath91 , which is somewhat higher than the value obtained by a06 , @xmath89 , although the difference is not statistically significant ( see fig . [ fig : comp ] ) . as @xmath101 and @xmath6 in equation [ eq : powerlaw ] presumably are strongly correlated , we fit a06 s data to a power - law model with the fixed value @xmath91 to find the corresponding value of @xmath101 . using the @xmath102 fit statistics , which accounts only for errors in the values of @xmath82 , we find @xmath90 with @xmath103 for 8 degrees of freedom , which indicates that our @xmath91 fit also provides a good description of the data , and hints that the systematic errors in the data points are not fully independent . the power - law model with @xmath104 and @xmath91 is plotted in figure [ fig : comp ] as the dashed line ; the dotted lines delimitate the corresponding statistical uncertainty in this power - law . in the @xmath91 model , the value of the parameter @xmath101 is a measure of the jet efficiency @xmath62 . setting @xmath91 in eq . [ eq : powerlaw ] we have @xmath105 which can be recast as @xmath106 figure [ fig : eta ] shows the jet efficiency factor corresponding to the observed value of @xmath90 ( @xmath91 ) , as a function of the mass accretion parameter @xmath98 . the value of @xmath107 derived by a06 is , in fact , the minimum value and corresponds to @xmath108 or @xmath109 . since @xmath98 is expected to be significantly less than unity , the required jet efficiency will be higher . and @xmath98 required to reproduce the observed correlation between @xmath86 and @xmath58 . each line correspond to a different value of @xmath101 ( shown above the lines ) , the thicker line represents the best - fit value of @xmath101 . we limit the value of efficiency to the theoretical maximum for the hybrid model with @xmath110 ( @xmath111 ) . the dotted line corresponds to the maximum efficiency for the blandford - znajek model with @xmath110 . these correspond to maximally spinning black holes with @xmath65 . the inferred efficiencies decline with increasing @xmath13 since the plausible values of @xmath98 increase . for @xmath68 , the maximal efficiencies are @xmath69 ( bz model ) and @xmath70 ( hybrid model ) . ] since in our models for the jet power , the jet efficiency factor is a function of the viscosity parameter @xmath13 and the black hole spin @xmath42 , the above equations imply that the value of the parameter @xmath101 is a function of the parameters @xmath13 , @xmath98 and @xmath42 . and therefore , for a given value of @xmath13 , we can use the observed value of @xmath90 to explore the range of possible values of @xmath98 and @xmath42 . as described previously , various different analysis and arguments suggest that @xmath13 ranges between 0.04 and 0.3 . in the discussion below , we will derive results for the two values that bound the range : @xmath20 and @xmath21 . figure [ fig : contour ] shows the parameter space ( @xmath42 , @xmath98 ) corresponding to the observed values of @xmath101 for the bz ( dashed line , [ sec : bz ] ) and the hybrid ( solid line , [ sec : disk ] ) jet models . the two panels show the results for the two different values of @xmath13 near the innermost stable circular orbit , @xmath68 ( @xmath112 , left panel ) and @xmath20 ( @xmath113 , right panel ) . we remind the reader that @xmath14 is the ratio of the gas pressure to the magnetic pressure near the radius of the marginally stable orbit . the label beside each contour is the value of @xmath101 for that line . we show the effect of considering an uncertainty of @xmath114 in @xmath101 . note that the range we have adopted assumes that the uncertainty in each data point is independent . @xmath101 would be smaller if , for example , the jet power was systematically under - estimated in all systems ( see [ sec : rapidexpansion ] ) . based on the arrangement of the contours in this figure , we find that the observed tight correlation between accretion rates and jet powers implies a narrow range of black hole spins for the elliptical galaxies of the sample of a06 , irrespective of the value of @xmath13 and the specific model adopted , with the main result being that the central black holes powering the jets in these systems must be rapidly spinning . specifically , we find that if @xmath115 ( figure [ fig : contour]a ) , as required to ensure an agreement between adaf model predictions for xrbs and the observations @xcite , then the hybrid model requires @xmath116 , while the bz model implies @xmath117 . if the value of the viscosity parameter is on the low side , @xmath118 ( figure [ fig : contour]b ) , both jet models require @xmath116 . in all cases , this corresponds to @xmath119 . these lower limits for both @xmath42 and @xmath62 correspond to the situation where rate of mass accretion onto the black hole equals the bondi rate ( @xmath120 ) . as discussed above , only a fraction of the bondi rate is likely to make its way to the black hole and realistically , @xmath121 . this restricts the allowed region of parameter space in figure [ fig : contour ] to that below the horizontal dotted lines ( @xmath122 ) corresponding to @xmath123 . subject to these constraints on the mass accretion rate , the observed correlations between mass accretion rate and jet power imply that in the case of the hybrid model , the central black holes must have spins @xmath124 if @xmath115 . the corresponding range for the mass accretion rate onto the black hole is @xmath125 . if @xmath13 is on the low side , then the black holes in the hybrid model must be spinning rapidly ( @xmath126 ) and the corresponding range for the mass accretion rate is @xmath127 . for a given value of @xmath42 , the corresponding jet efficiency factor can be read off from figure [ fig : eta ] . for example , if @xmath128 and @xmath129 , @xmath130 , giving an overall efficiency of @xmath131 . in the case of the bz model , the spin distribution corresponds to @xmath126 ( @xmath68 ) and @xmath132 ( @xmath20 ) . the latter is unphysical and indicates that values of viscosity parameters as low as @xmath20 are disfavored . for this line of reasoning , we find that the bz model requires @xmath133 in order to guarantee @xmath134 and @xmath135 . it is also possible these `` problems '' reflect the limitations of the bz model . as indicated previously , the bz model is , strictly speaking , applicable only in the limit of slow rotation where the unipolar approximation is reasonable . for this reason , and for the clarity of the discussion , we shall hereafter only discuss the results in the context of the hybrid model though we shall continue to show the results for both models . since our results indicate high spin values for jet - emitting black holes , it useful to bear in mind that recent numerical relativistic mhd simulations of thick - disks of @xcite suggest that the accreting plasma will bring the black hole in spin equilibrium not at the maximal value of @xmath65 @xcite but at a lower value of @xmath136 . this is not a trivial difference . at high spin rates , a @xmath137 reduction in @xmath42 translates into a factor of @xmath138 reduction in the jet power . in the specific case of the a06 elliptical galaxies , we find that black holes in these galaxies must be fed at rates @xmath139 , depending on the value of @xmath13 and the mechanism of jet powering at work . in the particular case of a hybrid model with @xmath115 , the black holes can not be fed at rates much lower than @xmath140 ( i.e , @xmath141 , where the lower limit is due to the reduced efficiency associated with @xmath142 and corresponds to @xmath123 ) , limiting the potential role of mass loss through winds . a06 uses the x - ray observations of the intracluster medium cavities to estimate the jet power . to derive the energy @xmath79 required to create the observed bubbles in the x - ray emitting gas , a06 assume that the cavities are inflated slowly and obtain the relationship @xmath143 for @xmath144 ( relativistic plasma ) , where @xmath145 is the thermal pressure of the surrounding x - ray emitting gas , @xmath146 is the volume of the cavity and @xmath12 is the mean adiabatic index of the fluid within the cavity . this is the minimum energy required to inflate the cavities and does not take into account any additional energy that may have gone into heating the intracluster medium . a more realistic scenario is likely to involve overpressurized bubbles , which upon injection expand rapidly to reach pressure equilibrium with their surroundings , and in the process generate weak shock waves that heat the x - ray emitting gas . @xcite calculate the total energy @xmath147 deposited by the jets in such cases and find that @xmath148 where @xmath149 is the thermal pressure of the surrounding gas at the location where the bubble is injected . equation [ eq : nusser ] suggests that if a bubble is injected with overpressure @xmath150 then @xmath151 . we consider the possibility that as an extreme case the bubble energies are twice the value assumed by a06 ( @xmath152 ) , and calculate the impact of this on the observed jet powers and on the results derived from our models . following a06 , we calculate the modified jet powers as @xmath153 and refit the power - law model with @xmath91 to the modified data , taking into account the proper error propagation on the values of @xmath154 . we obtain @xmath155 with @xmath103 for 8 degrees of freedom . as the assumed value of @xmath147 implies higher jet powers , our models need higher values of @xmath42 to reproduce the increased values of @xmath154 . in particular , if @xmath115 these values of @xmath101 imply @xmath156 for the hybrid model and @xmath157 for the bz model . therefore , a narrower range of still large spins is required to explain @xmath154 and the implied numerical values of @xmath98 and @xmath158 also increase slightly . one important uncertainty in the above calculation ( and also in that of a06 ) is the estimate of the age of the bubbles . a06 calculate the ages using the formula @xmath159 , where @xmath160 is the distance of the bubble centre from the black hole and @xmath74 is the adiabatic sound speed , but as discussed by @xcite ( see also @xcite ) there are two other ways of estimating the age of the cavities ( equations 3 and 4 of @xcite ) which result in longer timescales when compared to @xmath161 . an underestimate of the timescales involved impacts the estimate of jet power , as determined above , by artificially enhancing the jet power above the `` true '' value . to take this into account , we consider the possibility that the a06 ages are too low by a factor of @xmath162 @xcite , such that the modified age is @xmath163 , implying a smaller jet power @xmath164 . with reference to the original jet power estimates of a06 , we find that revising the timescales as suggested here results , not surprisingly , in a systematic decrease in lower limits of @xmath42-values . for example , in the case of the @xmath68 hybrid model , we find that @xmath165 . in spite of the above - mentioned concerns with the a06 estimates of the total jet energy and the associated timescales , it seems likely that their end - product , the estimates of the jet power in the nine elliptical galaxies , is reasonable . the impact of underestimating the total jet energy very nearly cancels out the impact of underestimating the timescales involved and the analysis that we have presented thus far will only be minimally affected . finally , we have alluded previously to the possibility that the accretion flow may be modified by mass loss in the form of winds . winds from the accretion flow are likely to be the norm rather than the exception ( e.g. , @xcite ) and these winds will remove mass , angular momentum and energy from the flow . as a consequence , adaf - like accretion flows with winds ( advection - dominated inflow - outflow solution , adios model , bb99 ) will have a different dynamical structure than adafs without winds . for instance given the same value of @xmath9 near the black hole , the angular velocity , scale height , total pressure and magnetic field strength predicted by the adafs with and without winds will be different . and since our models for the jet power are linked to the accretion flow model through the quantities @xmath4 , @xmath5 and @xmath6 , it is not inconceivable that jet powers linked to no - wind adafs could be very different from jet powers linked to adios models . the adios solution has three parameters in addition to those characterizing the no mass loss adaf model : @xmath166 , @xmath167 and @xmath168 . these parameters describe how much mass , energy and angular momentum , respectively , the wind removes from the accretion flow and specific choice of values for these three parameters leads to different types of winds ( for more information see bb99 ) . to assess the implications for jet power if the true underlying model is more correctly described by the adios solution , we begin by fixing the parameters common to the two models ( the adios and the no - wind model ) as follows : @xmath169 , @xmath170 , and @xmath171 ( we have confirmed that varying these particular variables does not affect our conclusions ) . we then consider various adios solutions arising from varying the wind parameters across the following range : @xmath172 , @xmath173 and @xmath174 . this range of values encompasses all the interesting types of winds . we found that while some features of the two models are very different , the solutions for @xmath4 , @xmath5 and @xmath6 at the radius of marginally stable orbit in the adios model are , to first order , comparable to the corresponding solutions derived from the no - wind model . in detail , the differences are such that the adios jet powers are always smaller than the no - wind values by factor of order unity , regardless of the combination of values of the wind parameters . lower jet powers in the adios model , in turn , imply both a narrower range of spins for jet - emitting black holes and that these black holes must be spinning faster than those embedded in the corresponding no mass loss adaf in order to reproduce the correlations derived by a06 . in other words , if winds are indeed the norm , then the central black holes must be spinning even more rapidly than suggested by our analysis above . a06 reported an efficiency of conversion of @xmath81 into jet power of @xmath175 for @xmath95 . in our modelling , this corresponds to the case where @xmath108 ( i.e. @xmath176 ) . since it is more likely that @xmath177 , our models suggest that the efficiency of conversion of the accreting matter into jet power is considerably higher . for instance , @xmath62 may reach values as high as @xmath178 for high spin rates . the extraction of spin energy from the holes is responsible for this noticeable increase in the jet efficiency . we note that the increase in the jet power with black hole spin has been verified in numerical simulations of jets ( e.g. , @xcite ) . the uppper limit we have obtained for @xmath62 is more than an order of magnitude larger than that reported by @xcite . this result is mainly due to the fact that we have included a kerr metric shear - driven dynamo @xcite , which enhances the field strength and was not included by these authors ( see [ sec : riaf ] and appendix ) . based on the calculations of @xcite , @xcite have recently asserted that accretion flow - jet models of the kind we have considered can not account for the jet powers of a sample of low accretion rate 3cr fr i radio galaxies observed using the hubble space telescope ( jet powers in the range @xmath179 ) . as a result of much higher efficiencies , adopting the critical accretion rate @xmath180 @xcite above which the adaf solution ceases to be valid , and taking @xmath65 and @xmath68 , our jet models yield @xmath181 , which is more than enough to account for the observed jet power in the sample used by @xcite . the adopted model for the jet - formation mechanism predicts a linear relation between accretion rate and jet power , while the measured best - fit slope of the relation measured by a06 is marginably better fit by a non - linear relationship . more than likely , this marginal discrepancy is due to the small number of systems in the a06 sample . however , we note that if @xmath98 has a slight dependence on @xmath42 , as suggested by recent numerical simulations of @xcite , then @xmath98 will also exhibit a weak dependence on @xmath58 and the agreement of the model with the observed correlation will be improved . we verified that if @xmath182 the slope predicted by the model agrees with the observed best - fit slope . in the numerical simulations , the dependence of @xmath61 on @xmath42 arises due amplified magnetic fields that act to transfer angular momentum from the hole to the accretion disk . in addition , it is not inconceivable that the dependence @xmath183 might also emerge as a consequence of feedback effects of the jet on the material fueling the accretion flow . we have employed two physical models for the black hole - accretion flow - jets with features that are broadly consistent with the results of numerical simulation to understand the empirical correlation between accretion rates and jet powers of x - ray luminous elliptical galaxies derived by a06 : the classical blandford - znajek model and a hybrid model proposed by @xcite . in the bz model the energy is extracted electromagnetically from the spinning black hole ; in the hybrid model the magnetic field threads both the inner and the outer regions of the accretion flow and the energy that powers the outflow is extracted from both the rotation of the disk as well as from the rotation of the black hole , via the frame - dragged accretion flow inside the hole s ergosphere . we assume that the accretion flow is advection - dominated ( adaf ) and take into account general relativistic effects not fully appreciated before in these models . in the absense of disk winds and feedback , the model suggests that the jet power should be linearly dependent on the accretion mass flow rate . the normalisation of this relation is dependent on the disk viscosity and the black hole spin , with the dispersion around the best - fit correlation of a06 being caused by different values of the black hole spin , @xmath42 , and the ratio of the disk mass flow rate to the accretion rate onto the black hole , @xmath98 . we find that the jet efficiency ( @xmath184 ) can exceed @xmath185 so that the jet may carry away an appreciable fraction of the rest mass energy of the accreted matterial . we compared our jet power models to the jet power estimates made by a06 . adopting typical values of the viscosity parameter @xmath0 , the @xmath86 vs. @xmath186 correlation implies a narrow range of spins @xmath187 and accretion rates @xmath188 . if we further demand that the mass accretion rate be restricted to @xmath189 , as is both likely to be the case and consistent with adaf model expectations , we find that the observed correlations require @xmath190 ; i.e. , the correlations imply rapidly spinning black holes in all of the target galaxies . if additionally , the adaf accretion flows also experience mass , energy and angular momentum loss via winds as in the adios model proposed by bb99 , the correlations indicate nearly - maximally spinning black holes ( see [ sec : winds ] ) . it is reassuring that semi - analytic cosmological simulations of the spin evolution of black holes through mergers and gas accretion @xcite and estimates of the radiative efficiencies of global populations of quasars based on soltan - type arguments ( e.g. , @xcite ) suggest that most nearby massive holes are rapidly rotating . the relatively small scatter in the correlation between @xmath191 and @xmath58 of a06 , combined with the strong dependence of jet power on black hole spin implies large black hole spins is probably a general result valid for all elliptical galaxies . of course the sample considered by a06 is relatively small , but if this result holds - up in larger samples , and can be shown to also apply to the larger central radio galaxies of galaxy clusters @xcite , our results suggest that they have the most powerful jets because they have sufficient black hole mass to host relatively high mass flow rates while the disk remains in the adaf state ( and hence is able to generate strong poloidal magnetic fields near the black hole ) . in this picture , all central ellipticals have the capacity to produce powerful jets , and the resulting jet power is determined by the structure of the accretion disk . our results reveal a potentially fundamental connection between black holes and the formation of the most massive galaxies : the central holes in the cores of galaxy clusters must be rapidly rotating , in order to make the radio jets powerful enough to provide an effective feedback mechanism and quench the cooling flows , therefore preventing star formation and explaining the observed galaxy luminosity function ( e.g. , @xcite ) as well as accounting for the observed x - ray luminosity temperature and x - ray luminosity sunyaev - zeldovich effect correlations for the intracluster medium ( e.g. , @xcite ) . we are very grateful to david l. meier and chris done for their valuable guidance and discussions . rsn thanks teddy cheung , feng yuan and xinwu cao for helpful discussions and acknowledges the hospitality of the institute for computational cosmology ( icc , durham university ) where part of this work was carried out . rsn and tsb acknowledges support from the brazilian institutions cnpq , capes and fapergs . rgb thanks pparc for the support of a senior research fellowship . ab thanks the icc for hospitality shown to him during his tenure there as leverhulme visiting professor , and is deeply appreciative of support from the leverhulme trust as well as nserc ( canada ) . this work was supported by the european commission s alfa - ii programme through its funding of the latin - american european network for astrophysics and cosmology ( lenac ) . allen , s. w. , r. j. h dunn , fabian , a. c. , taylor , g. b. , reynolds , c. s. 2006 , mnras , 372 , 21 ( a06 ) armitage , p. j. , natarajan , p. 1999 , apj , 523 , l7 babul , a. 2002 , mnras , 330 , 329 balbus , s. a. , hawley , j. f. 1998 , reviews of modern physics , 70 , 1 bardeen , j. m. , press , w. h. , teukolsky , s. a. 1972 , apj , 178 , 347 birzan , l. , rafferty , d. a. , mcnamara , b. r. , wise , m. w. , nulsen , p. e. j. 2004 , apj , 607 , 800 begelman , m. c. , blandford , r. d. , rees , m. 1984 , rev . phys . , 56 , 255 blandford , r. d. , znajek , r. 1977 , mnras , 179 , 433 blandford , r. 2005 , in merloni a. , nayakshin s. , sunyaev r. a. , eds , eso astrophysics symposia , growing black holes : accretion in a cosmological context . springer - verlag , p. 477 blandford , r. d. , payne , d. g. 1982 , mnras , 199 , 883 blandford , r. d. , begelman , m. 1999 , mnras , 303 , l1 ( bb99 ) bondi , h. 1952 , mnras , 112 , 195 bower , r. g. , benson , a. j. , malbon , r. , helly , j. c. , frenk , c. s. , baugh , c. m. , cole , s. , lacey , c. g. 2006 , mnras , 370 , 645 cao , x. , rawlings , s. 2004 , mnras , 349 , 1419 churazov , e. , sunyaev , r. , forman , w. , bhringer , h. 2002 , mnras , 332 , 729 churazov , e. , sazonov , s. , sunyaev , r. , forman , w. , jones , c. , bhringer , h. 2005 , mnras , 363 , 91 croton , d. j. , et al . 2006 , mnras , 365 , 11 cui , w. , zhang , s. n. , chen , w. 1998 , apj , 492 , l53 dalla vecchia , c. , bower , r. g. , theuns , t. , balogh , m. l. , mazzotta , p. , frenk , c. s. 2004 , mnras , 355 , 995 de villiers , j. p. , hawley , j. f. , krolik , j. h. , hirose , s. 2005 , apj , 620 , 878 di matteo , t. , johnstone , r. m. , allen , s. w. , fabian , a. c. 2001 , apj , 550 , l19 di matteo , t. , allen , s. w. , fabian , a. c. , wilson , a. s. , young , a. j. 2003 , apj , 582 , 133 esin , a. a. , mcclintock , j. e. , narayan , r. 1997 , apj , 489 , 865 fabian , a. c. , sanders , j. s. , taylor , g. b. , allen , s. w. , crawford , c. s. , johnstone , r. m. , iwasawa , k. 2006 , mnras , 366 , 417 falcke , h. , krding , e. , markoff , s. 2004 , a&a , 414 , 895 ferrari , a. 1998 , ara&a , 36 , 539 gallo , e. , fender , r. p. , pooley , g. g. 2003 , mnras , 344 , 60 gammie , c. f. , shapiro , s. l. , mckinney , j. c. 2004 , apj , 602 , 312 ghosh , p. , abramowicz , m. a. 1997 , mnras , 292 , 887 greene , j. e. , ho , l. c. , ulvestad , j. s. 2006 , apj , 636 , 56 hawley , j. f. , gammie , c. f. , balbus , s. a. 1995 , apj , 440 , 742 hawley , j. f. , krolik , j. h. 2006 , apj , 641 , 103 hirose , s. , krolik , j. h. , de villiers , j. p. , hawley , j. f. 2004 , apj , 606 , 1083 igumenshchev , i. v. , narayan , r. , abramowicz , m. a. 2003 , apj , 592 , 1042 koide , s. , meier , d. l. , shibata , k. , kudoh , t. 2000 , apj , 536 , 668 koide , s. 2003 , phys . d , 67 , 104010 komissarov , s. s. 2005 , mnras , 359 , 801 krding , e. g. , jester , s. , fender , r. 2006 , mnras , 372 , 1366 livio , m. , ogilvie , g. i. , pringle , j. e. 1999 , apj , 512 , 100 loewenstein , m. , mushotzky , r. f. , angelini , l. , arnaud , k. a. , quataert , e. 2001 , apj , 555 , l21 maccarone , t. j. , gallo , e. , fender , r. 2003 , mnras , 345 , l19 macdonald , d. , thorne , k. s. 1982 , mnras , 198 , 345 mccarthy , i. g. , balogh , m. l. , babul , a. , poole , g. b. , horner , d. j. 2004 , apj , 613 , 811 mckinney , j. c. 2005 , apj , 630 , l5 mckinney , j. c. , gammie , c. f. 2004 , apj , 611 , 977 meier , d. l. 1999 , apj , 522 , 753 meier , d. l. 2001 , apj , 548 , l9 merloni , a. , heinz , s. , di matteo , t. 2003 , mnras , 345 , 1057 narayan , r. , mahadevan , r. , quataert , e. 1998 , in abramowicz m. a. , bjrnsson g. , pringle j. e. , eds , the theory of black hole accretion disks . cambridge univ . press , p. 148 ( n98 ) narayan , r. 2005 , ap&ss , 300 , 177 narayan , r. , yi , i. 1995 , apj , 452 , 710 nemmen , r. s. , storchi - bergmann , t. , yuan , f. , eracleous , m. , terashima , y. , wilson , a. s. 2006 , apj , 643 , 652 novikov , i. d. , thorne , k. s. 1973 , in black holes , de witt c. , de witt b. s. , eds , black holes . gordon & breach , new york , p. 344 nusser , a. , silk , j. , babul , a. 2006 , mnras , 373 , 739 pellegrini , s. 2005 , apj , 624 , 155 proga , d. , begelman , m. c. 2003 , apj , 592 , 767 punsly , b. , coroniti , f. v. 1990 , apj , 354 , 583 quataert , e. , narayan , r. 1999 , apj , 520 , 298 quataert , e. , gruzinov , a .. 2000 , apj , 539 , 809 rafferty , d. a. , mcnamara , b. r. , nulsen , p. e. j. , wise , m. w. 2006 , apj , 652 , 216 rees , m. j. , begelman , m. c. , blandford , r. d. , phinney , e. s. 1982 , nature , 295 , 17 sikora , m. , stawarz , l. , lasota , j .- 2007 , apj , 658 , 815 sijacki , d. , springel , v. 2006 , mnras , 366 , 397 shakura , n. i. , sunyaev , r. a. 1973 , a&a , 24 , 337 soltan , a. 1982 , mnras , 200 , 115 soria , r. , fabbiano , g. , graham , a. w. , baldi , a. , elvis , m. , jerjen , h. , pellegrini , s. , siemiginowska , a. 2006 , apj , 640 , 126 taylor , g. b. , sanders , j. s. , fabian , a. c. , allen , s. w. 2006 , mnras , 365 , 705 thorne , k. s. 1974 , apj , 191 , 507 thorne , k. s. , price , r. h. , macdonald , d. a. , eds . 1986 , black holes : the membrane paradigm . yale univ . press , new haven , p. 132 tremaine , s. , et al . 2002 , apj , 574 , 740 volonteri , m. , madau , p. , quataert , e. , rees , m. 2005 , apj , 620 , 69 wang , j .- m . , chen , y .- m . , ho , l. c. , mclure , r. j. 2006 , apj , 642 , l111 yu , q. , tremaine , s. 2002 , mnras , 335 , 965 we list the equations we used to compute the dependence of the jet power on @xmath13 , @xmath42 and @xmath9 using the blandford - znajek model ( [ sec : bz ] ) and the hybrid model ( [ sec : disk ] ) . the jet power is given by equations [ eq : powerbz ] ( bz model ) and [ eq : powerdisk ] ( hybrid model ) . the code that implements the equations described in this work and returns the jet power is available at the url http://www.if.ufrgs.br/@xmath192rns/jetpower.htm . the following equations describe the self - similar adaf structure @xcite , where we use the black hole mass in solar units ( @xmath193 ) , accretion rates in eddington units ( @xmath194 , @xmath195 is the eddington accretion rate defined in [ sec : correl ] ) and radii in schwarzschild units ( @xmath196 ) : @xmath197 the constants @xmath198 , @xmath199 and @xmath200 are defined as @xmath201 ^{1/2 } , \\ c_3 = c_2 ^ 2/\epsilon ' , \\ \epsilon ' \equiv \frac{1}{f } \left ( \frac{5/3-\gamma}{\gamma-1 } \right ) , \\ g'(\alpha,\epsilon ' ) \equiv \left [ 1 + \frac{18 \alpha^2}{(5 + 2\epsilon')^2 } \right]^{1/2}. \end{gathered}\ ] ] the relations among @xmath13 , @xmath12 and @xmath14 are given in [ sec : riaf ] . the angular velocity of the field seen by an outside observer at infinity in the boyer - lindquist frame is @xmath30 , where the angular velocity of the local metric is given by @xcite @xmath202 using geometrized units ( @xmath203 ) . we estimate the field - enhancing shear caused by the kerr metric following @xcite as @xmath33 , such that the azimuthal component of the field is given by @xmath32 ( see [ sec : riaf ] ) . the poloidal component is related to the azimuthal component following @xcite as @xmath204 . lastly , we insert all the quantities defined above into equations [ eq : powerbz ] and [ eq : powerdisk ] , and then evaluate the resulting equations at the marginally stable orbit of the accretion disk @xmath24 , given by @xcite @xmath205^{1/2 } \right\rbrace , \\ z_1 \equiv 1 + ( 1-j^2)^{1/3 } \left [ ( 1+j)^{1/3 } + ( 1-j)^{1/3 } \right ] , \nonumber \\ z_2 \equiv ( 3 j^2 + z_1 ^ 2)^{1/2}. \nonumber\end{gathered}\ ] ] taking @xmath56 in the bz model corresponds to assume that the strength of the magnetic field at the horizon of the hole is very similar to the corresponding strength at the marginally stable orbit . this is a reasonable assumption according to recent numerical simulations of jet formation ( e.g. , @xcite ) .

super - resolution imaging aims at improving the resolution of an image by enhancing it with other images or data that might have been acquired using different imaging techniques or modalities . in this paper we consider the task of doubling the resolution of tomographic grayscale images of binary objects by fusion with double - resolution tomographic data that has been acquired from two viewing angles . we show that this task is polynomial - time solvable if the gray levels have been reliably determined . the task becomes-hard if the gray levels of some pixels come with an error of or larger . the-hardness persists for any larger resolution enhancement factor . this means that noise does not only affect the quality of a reconstructed image but , less expectedly , also the algorithmic tractability of the inverse problem itself . = 1

1701.04399

the results presented in this paper are motivated by the task of enhancing the resolution of reconstructed tomographic images obtained from binary objects representing , for instance , crystalline structures , nanoparticles or two - phase samples @xcite . the reconstructed tomographic image might contain several gray levels , which , depending on the accuracy of the reconstruction , result from the fact that low - resolution pixels may cover different numbers of black high - resolution pixels . for turning the grayscale image into a higher - resolution binary image we utilize the gray levels and two additional higher - resolution projections , which may have been acquired by different imaging techniques or modalities ( e.g. , via scanning transmission electron microscopy @xcite ) . more precisely , we study the task of reconstructing binary @xmath2-images from row and column sums and additional constraints , so - called _ block constraints _ , on the number of black pixels to be contained in the @xmath3-blocks resulting from a subdivision of each pixel in the @xmath4 low - resolution image . we remark that we do not require that the x - ray data is taken from orthogonal directions . in our context it suffices that the x - ray data has been taken with double - resolution according to the discretization of the lower - resolution image . figure [ fig : process ] illustrates the process . the example given in this figure also shows that the block - constraints can help to narrow down the solution space . in fact , the solution shown in this figure is uniquely determined by the input ( the row and column sums and block constraints ) . the row and column sums alone do not determine the solution uniquely ( and , of course , neither do the block constraints ) . apart from super - resolution imaging @xcite , and its particular applications @xcite in discrete tomography @xcite , the tasks discussed in this paper are also relevant in other contexts . from a combinatorial point of view , they can be viewed as reconstructing binary matrices from given row and column sums and some additional constraints . unexpected complexity jumps based on the results of the present paper are discussed in @xcite . background information on such problems involving different kinds of additional constraints can be found in ( * ? ? ? * sect . 4 ) . other applications belong to the realm of _ dynamic discrete tomography _ @xcite . for instance , in plasma particle tracking , some particles are reconstructed at time @xmath5 between @xmath6 and the next time step @xmath7 the particles may have moved to other positions . the task is to reconstruct their new positions , again from few projections . one way of incorporating additional prior knowledge about the movement leads to block constraints of the kind discussed in this paper ; see @xcite . as the task of reconstructing a binary matrix from its row and column sums can be formulated as the task of finding a @xmath8-matching in a bipartite graph , we remark that our results can also be seen as results on finding @xmath8-matchings that are subject to specific additional constraints . for related ( but intrinsically quite different ) results on matchings subject to so - called budget constraints ; see , e.g. , @xcite . the two main contributions of this paper are as follows . on the one hand , we show that reliable bimodal information can be utilized efficiently , i.e. , the resolution can be doubled in polynomial time if the gray levels of the low resolution image have been determined precisely ( theorem [ thm : main1 ] ) . on the other hand , the task becomes already intractable ( unless @xmath9 ) if the gray levels of some pixels come with some small error of @xmath1 ( theorem [ thm : main2 ] and corollary [ cor : largebox ] ) . this proves that noise does not only affect the quality of the reconstructed image but also the algorithmic tractability of the inverse problem itself . hence the possibility of compensating noisy imaging by including bimodal information may in practice be jeopardized by its algorithmic complexity . let @xmath10 @xmath11 and @xmath12 denote the set of integers , natural numbers , and non - negative integers , respectively . for @xmath13 set @xmath14 @xmath15:=\{1,\dots , k\},$ ] and @xmath15_0:=\{0,\dots , k\}.$ ] with @xmath16 we denote the all - ones vector of the corresponding dimension . the cardinality of a finite set @xmath17 is denoted by @xmath18 in this paper we use cartesian coordinates ( rather than matrix notation ) to represent pixels in an image . in particular , the two numbers @xmath19 and @xmath20 in a pair @xmath21 denote the @xmath22- and @xmath23-coordinate of a point , respectively . in the following we consider grids @xmath24\times[n].$ ] the set @xmath25 $ ] and @xmath24\times\{j\}$ ] is called _ column @xmath19 _ and _ row @xmath26 _ respectively . let , in the following @xmath27 any set @xmath28 $ ] is a _ vertical strip ( of width @xmath29)_. a _ horizontal strip ( of width @xmath29 ) _ is a set of the form @xmath24\times\{(j-1)k+1,\dots , jk\}.$ ] sets of the form @xmath30\times[c , d])\cap\mathbb{z}^2,$ ] with @xmath31 and @xmath32 @xmath33 are called _ boxes . _ for @xmath34 let @xmath35_0 ^ 2.$ ] defining for any @xmath36 and @xmath37 the set of _ ( lower - left ) corner points _ @xmath38\times [ n])\cap(k\mathbb{n}_0 + 1)^2,$ ] we call any box @xmath39 with @xmath40 a _ block_. the blocks form a partition of @xmath24\times[n],$ ] i.e. , @xmath41\times[n].$ ] for @xmath42 with @xmath43 we define the task of ( noisy ) super - resolution nsr@xmath44 + @xmath45 & & & & & & & & & \textnormal{\hspace*{1ex}gray value measurem.)}\\ & & & \omit \rlap{$\displaystyle v(i , j)\in[k^2]_0,$ } & & & & ( i , j)\in c(m , n , k ) , & & \textnormal{(gray value measurem.)}\\[1.2ex ] & \textnormal{task:}\quad & & \omit \rlap{find $ \xi_{p , q}\in\{0,1\},$ \:$(p , q)\in[m]\times[n],$ with } \\ \displaystyle & & & \sum_{\mathclap{p\in[m ] } } \xi_{p , q}&&=r_q , & & q\in[n ] , & & \textnormal{(row sums ) } \\ \displaystyle & & & \sum_{q\in[n ] } \xi_{p , q}&&=s_p , & & p \in [ m ] , & & \textnormal{(column sums)}\\ \displaystyle & & & \sum_{\mathclap{(p , q)\in b_k(i , j)}}\xi_{p , q}&&=v(i , j ) , & & ( i , j)\in r , & & \textnormal{(block constraints)}\\ \displaystyle & & & \sum_{\mathclap{(p , q)\in b_k(i , j)}}\xi_{p , q}&&\in v(i , j)+[-\varepsilon,\varepsilon ] , & \hspace*{3ex } & ( i , j)\in c(m , n , k)\setminus r,\:\:&&\textnormal{(noisy block constraints)}\\ & & & \omit \rlap{or decide that no such solution exists.}\\\end{aligned}\ ] ] the numbers @xmath46 and @xmath47 are the row and column sum measurements of the higher - resolution binary @xmath2 image , @xmath48_0 $ ] corresponds to the gray value of the low - resolution @xmath3-pixel at @xmath21 of the low - resolution @xmath4 grayscale image , and @xmath49 is the set of low - resolution pixel locations for which we assume that the gray values have been determined reliably , i.e. , without error . the number @xmath50 is an error bound for the remaining blocks . the task is to find a binary high - resolution image satisfying the row and column sums such that the number of black pixels in each block adds up to a gray value for the corresponding @xmath3-pixel in the lower - resolution image that lies in the specified interval . clearly , a necessary ( and easily verified ) condition for feasibility is that @xmath51}r_q=\sum_{p\in[m]}c_p.\ ] ] in the following we will assume without loss of generality that this is always the case . our special focus is on double - resolution imaging , i.e. , on the case @xmath52 for @xmath53 we define ndr@xmath54nsr@xmath55 in the reliable situation , i.e. , for @xmath56 we simply speak of _ double - resolution _ and set dr@xmath57nsr@xmath58 ( then , of course , the set @xmath49 can be omitted from the input . ) in the present paper , we show that the resolution of any reliable grayscale image can in fact be doubled in each dimension in polynomial - time if x - ray data is provided from two viewing angles at double resolution . in other words , we show [ thm : main1 ] @xmath59 figures [ fig : drexample2 ] and [ fig : drexample1 ] illustrate the performance of the algorithm for dr applied to two phantoms . the results have been obtained within a fraction of a second on a standard pc . in figure [ fig : drexample2 ] , the original phantom is a binary @xmath60 image of a crystalline sample taken from @xcite . it is assumed that the half - resolution grayscale image shown in figure [ fig : drexample2](a ) has been obtained by some imaging method and that double - resolution x - ray information in the two standard directions is also available . application of the algorithm yields the binary image shown in figure [ fig : drexample2](c ) . in figure [ fig : drexample1 ] , the original phantom is a binary @xmath61 image , for which the half - resolution grayscale image shown in figure [ fig : drexample1](a ) is available . from this image and the row and column sums ( counting the black pixels ) of the original phantom the algorithm returns the binary image shown in figure [ fig : drexample1](b ) . the results shown in figures [ fig : drexample2 ] and [ fig : drexample1 ] satisfy , of course , the row and column sums and block constraints . however , in both cases the solutions are still not unique . this can also be detected in polynomial time . [ main : unique ] for every instance of dr it can be decided in polynomial time whether the instance admits a unique solution . while the additional x - ray information ( at double resolution ) does in general reduce the ambiguity , typically it will still not lead to uniqueness . of course , a standard way of dealing with non - uniqueness in practice is regularization @xcite . to illustrate the possibility of adapting a regularization scheme to our context , we show in figure [ fig : drexample1](c ) a binary solution minimizing the _ total variation ( tv ) . _ in fact , the solution is obtained from figure [ fig : drexample1](a ) by applying @xmath62 so - called local switches , each of which are strictly decreasing the value of the total variation functional @xmath63}\sum_{q\in[n]}||(\nabla x_{p , q}^{(1)},\nabla x_{p , q}^{(2)})^t||_2,\ ] ] where @xmath64 and @xmath65 denotes the euclidean norm . see , e.g. , @xcite for some background information . of course , we obtain the same results for other @xmath66-norms ( including the @xmath67-norm ) , because our images are binary . we now turn to the task @xmath68 where small `` occasional '' uncertainties in the gray levels are allowed . first observe that a constant number of uncertainties does not increase the complexity . however , if we allow a ( as we will see , comparably small but ) not constant number of uncertainties the problem becomes hard . [ thm : main2 ] ndr@xmath69 is @xmath0-hard for any @xmath70 as it turns out , the problem remains @xmath0-hard for larger block sizes . [ cor : largebox ] nsr@xmath44 is @xmath0-hard for any @xmath43 and @xmath70 in other words , noise does not only affect the reconstruction quality . it also affects the algorithmic tractability of the inverse problem . ( for background material on complexity theory , see , e.g. , @xcite . ) the present paper is organized as follows . we deal with the case of reliable data in section [ sect:3 ] . results on ndr@xmath69 and nsr@xmath44 involving noisy data are contained in section [ sect:4 ] . section [ sect:5 ] concludes with some additional remarks on certain extensions and an open problem . in this section we discuss the task dr , i.e. , @xmath71 and @xmath72 we remark that , in the following , our emphasis is on providing brief and concise arguments for polynomial - time solvability rather than to focus on computationally or practically most efficient algorithms . recall , however , our comments on the computational performance of our algorithms and the effect of the additional information for the example images depicted in figures [ fig : drexample2 ] and [ fig : drexample1 ] . the main result of this section is theorem [ thm : main1 ] . in its proof we show that dr decomposes into five problems that can be solved independently . the five problems are single - graylevel versions of dr where each block is required to contain the same number @xmath73_0 $ ] of ones . moreover , we allow for restricted problem instances ( and in particular row and column sums ) that are defined only on subsets @xmath74\times[n]\ ] ] of the grid @xmath24\times[n]$ ] defined by means of some @xmath75 let @xmath76:\exists j\in[n]:(i , j)\in i\}\ ] ] and @xmath77:\exists i \in [ n]:(i , j)\in i\}.\ ] ] then we define for @xmath73_0:$ ] dr@xmath78 + @xmath79 & \textnormal{task:}\quad & & \omit \rlap{find $ \xi_{p , q}\in\{0,1\ } , \:(p , q)\in g(i)$ with } \\ \displaystyle & & & \sum_{\mathclap{p:(p , j)\in g(i)}}\xi_{p , j+l}&&=r_{j+l } , & & j\in \pi_y(i ) , \:\ : l\in \{0,1\}&&\textnormal{(row sums ) } \\ \displaystyle & & & \sum_{\mathclap{q:(i , q)\in g(i)}}\xi_{i+l , q}&&=c_{i+l } , & & i\in \pi_x(i ) , \:\ : l\in \{0,1\}&&\textnormal{(column sums)}\\ \displaystyle & & & \sum_{\mathclap{(p , q)\in b(i , j)}}\xi_{p , q}&&=\nu,&\:\ : & ( i , j)\in i , & \:\:&\textnormal{(block constraints)}\\ & & & \omit \rlap{or decide that no such solution exists.}\end{aligned}\ ] ] of course , the tasks dr@xmath80 @xmath81 are trivial . in fact , the only potential solution satisfying the block constraints is given by @xmath82 @xmath83 and it is checked easily and in polynomial time whether this satisfies the row and column sums . the next lemma and corollary deal with @xmath84 , @xmath85 its statement and proof uses the notation @xmath86 and @xmath87 for the number of blocks in the same vertical strip as but below @xmath88 or in the same horizontal strip as but left of @xmath89 respectively . more precisely , let @xmath90)\cap i|\qquad \textnormal{and}\qquad \rho_j(i):=|([i]\times\{j\})\cap i|\ ] ] for @xmath91 [ lemma : mono ] a. [ mono2 ] an instance @xmath92 of dr@xmath93 is feasible if , and only if , for every @xmath94 we have @xmath95 b. [ mono3 ] the solution of a feasible instance of dr@xmath93 is unique if , and only if , for every @xmath94 we have @xmath96 c. [ mono1 ] dr@xmath97 clearly , for the feasibility of a given instance of dr@xmath93 the conditions @xmath98 are necessary . now , suppose that the conditions are satisfied . for every @xmath94 and @xmath99 we set @xmath100 figure [ fig : example1 ] gives an illustration . with @xmath101 in gray color . ( right ) solution defined by .,scaledwidth=45.0% ] by this definition we have satisfied all block constraints . a simple counting argument shows that the row and columns sums are also as required . in fact , for @xmath102 and @xmath103 we have @xmath104 this concludes the proof of ( [ mono2 ] ) . to show ( [ mono3 ] ) , we consider a solution @xmath105 to @xmath106 suppose that @xmath107 and @xmath108 for some @xmath109 then , there exist @xmath110 $ ] with @xmath111 by the block constraints we have @xmath112 applying a switch to @xmath113 @xmath114 @xmath115 and @xmath116 i.e. , setting @xmath117 @xmath118 and @xmath119 otherwise , yields thus another solution . as the same argument holds also for the column sums we have shown that the condition in ( [ mono3 ] ) is necessary . the condition is , on the other hand , also sufficient since the 0-values of the row and column sums leave only a single position @xmath120 in each block @xmath89 @xmath121 for which @xmath122 can take on a non - zero value ( in fact , the value needs to be 1 to satisfy the block constraints ) . we now turn to ( [ mono1 ] ) . the algorithm checks the condition ( [ mono2 ] ) . if it is violated , it reports infeasibility ; otherwise , a solution is constructed through clearly , all steps can be performed in polynomial time . [ cor : mono3 ] a. [ cormono2 ] an instance @xmath92 of dr@xmath123 is feasible if , and only if , for every @xmath94 we have @xmath124 b. [ cormono3 ] the solution of a feasible instance of dr@xmath123 is unique if , and only if , for every @xmath94 we have @xmath125 c. [ cormono1 ] dr@xmath126 the results follow directly from the results for @xmath127 in lemma [ lemma : mono ] as the former are obtained from the latter by inversion and vice versa , i.e. , by replacing @xmath128 by @xmath129 and @xmath130 by @xmath131 @xmath121 @xmath132 and , in the solutions , all @xmath67 s by @xmath133 s and @xmath133 s by @xmath67 s . next we prove polynomial - time solvability of @xmath134 to this end we need two lemmas . the first can be viewed as dealing with a certain `` two - color version '' of discrete tomography . the second deals with specific switches or interchanges ( in the sense of @xcite ) . [ lem : unimod2 ] the problem , given @xmath135 @xmath136 and @xmath137 @xmath121 decide whether there exist non - negative integers @xmath138 @xmath139 @xmath140 such that @xmath141 and , if so , determine a solution , can be solved in polynomial time . we assemble the variables @xmath138 @xmath142 for @xmath143 into an @xmath144-dimensional vector and rephrase the constraints in matrix form @xmath145 @xmath146 and @xmath147 where @xmath148 contains the @xmath149 s , @xmath150 contains the @xmath151 s , and @xmath152 denotes the @xmath153-identity matrix . the problem is then to determine an integer solution of @xmath154 where @xmath155 the submatrix @xmath156 is totally unimodular as it is the node - edge incidence matrix of a bipartite graph ( one of the two parts of the partition is the set of the last @xmath144 rows of @xmath157 ) . but then @xmath158 is also totally unimodular as it results from appending a subset of rows of @xmath159 to the totally unimodular matrix @xmath160 in other words , @xmath161 is an integral polyhedron , i.e. , it coincides with the convex hull of its integral vectors . hence , a vertex of this polyhedron can be found in polynomial time by linear programming . see ( * ? ? ? 16&19 ) for background material . in a solution of dr we can have 16 different block types . they are shown in figure [ fig : types:1 ] . several of them have the same label as they can not be distinguished by the row sums . we introduce now the concept of local switches . for this we remark that if we consider a horizontal strip with a block labeled @xmath162 and a block labeled @xmath163 ( labeling as in figure [ fig : types:1 ] ) , we can perform a switch that turns the blocks into blocks labeled @xmath164 and @xmath165 , respectively . we call such a switch a _ horizontal local switch of class @xmath166 _ in a similar way we define the _ horizontal local switches of class @xmath167-@xmath168 _ see table [ table : switchesexample ] , column 1 - 3 , and column 4 for an illustration . note that the illustration does not show all horizontal local switches but rather gives a typical example . ccclc@cclass & turns & into & & illustration&illustration + & & & & ( horizontal local switch ) & ( vertical local switch ) + ( 1 ) & @xmath169&@xmath170 & & & + & & & & & + ( 2 ) & @xmath171&@xmath172 & & & + & & & & & + ( 3 ) & @xmath173&@xmath174 & & & + & & & & & + ( 4 ) & @xmath175&@xmath176 & & & + & & & & & + ( 5 ) & @xmath177&@xmath178 & & & + & & & & & + ( 6 ) & @xmath165&@xmath165 & & & + & & & & & + ( 7 ) & @xmath179&@xmath180 & & & + + let @xmath181 denote the coordinate transformation @xmath182 a switch in a solution @xmath105 is called a _ vertical local switch of class @xmath183 @xmath184,$ ] _ if it is a horizontal local switch of class @xmath185 in @xmath186 see table [ table : switchesexample ] , column 5 , for an illustration . a switch is called _ local _ if it is a horizontal or vertical local switch of some class @xmath183 @xmath184.$ ] a solution is called _ reduced _ if no local switch of any class @xmath183 @xmath184,$ ] can be applied . [ lem : reduction ] a. [ red1 ] the set of solutions of a given instance of dr is invariant under local switches . + b. [ red2 ] if an instance of @xmath187 has a solution , then there is also a reduced solution . to prove ( i ) just observe that the local switches do neither change the row and column sums nor the number of ones contained in each block . let us now turn to ( ii ) . suppose the instance has a solution . the local switches of class @xmath183 @xmath188,$ ] increase the number of blocks @xmath88 with @xmath189 the process of applying local switches of class @xmath183 @xmath188,$ ] thus needs to terminate since the number of blocks in each problem instance is finite . further , note that local switches of class @xmath190 applied in horizontal strips increase the number of blocks labeled @xmath191 the vertically applied switches of that class do not decrease the number of such blocks , but increase the number of blocks containing at least a one in the rightmost column of the block . [ lem : dr2 ] @xmath192 our general strategy is as follows . first we show that a given instance is feasible if , and only if , there exists a specific reduced solution that contains only three block types . then we give a polynomial - time algorithm that finds such a solution or determines infeasibility . in the following we will again use the notational convention that specific settings of variables and parameters are signified by a superscript @xmath193 by possibly rearranging the rows and columns , we assume that @xmath194 now , suppose that the given instance has a solution . by lemma [ lem : reduction ] there exists a reduced solution . in the following , we consider such a reduced solution . the potential block types , and corresponding variables @xmath195 @xmath196 and @xmath165 counting the number of occurrences of the respective block types in the strip @xmath24\times\{j , j+1\}$ ] of a reduced solution , are shown in figure [ fig : types:1](b ) . block types counted by the same variable have equal row sums . consider now for fixed @xmath101 the blocks in @xmath197\times\{j , j+1\}).$ ] counting the block types , any solution needs to satisfy @xmath198 which implies @xmath199 as the solution is @xmath123-reduced we need to have @xmath200 or @xmath201 this , together with @xmath202 and , implies @xmath203 the same argument applies to the other horizontal strips , and a similar argument holds for the vertical strips . hence , we conclude that @xmath204 amounts to the task of assigning one of the three block types shown in figure [ fig : types:2 ] to each of the blocks @xmath89 @xmath121 in such a way that the number of type 1 and type 2 blocks in each horizontal and vertical strip , respectively , equals some prescribed numbers . instances considered in the proof of lemma [ lem : dr2 ] . value @xmath67 and @xmath133 are indicated in black and white , respectively.,scaledwidth=25.0% ] now , we introduce @xmath205-variables @xmath138 @xmath139 @xmath121 that should be interpreted as follows : @xmath206 and @xmath207 if , and only if , @xmath88 is of type 1 ; @xmath208 and @xmath209 if , and only if , @xmath88 is of type 2 , and @xmath210 if , and only if , @xmath88 is of type 3 . then we ask for @xmath205-solutions that satisfy @xmath211 such solutions can be found in polynomial time by lemma [ lem : unimod2 ] . now with these preliminary results we can turn to the proof of theorem [ thm : main1 ] . we show that the task of finding a solution to a given problem instance can be decomposed into five independent subtasks of solving instances of @xmath212 @xmath213 the subtasks are then polynomial time solvable by lemma [ lemma : mono ] , corollary [ cor : mono3 ] and [ lem : dr2 ] ( and trivially for @xmath214 and @xmath215 ) . the key step in our proof is to deduce from the data the row and column sums for the particular subtasks . this deduction is possible for reduced solutions . their existence for feasible problem instances is guaranteed by lemma [ lem : reduction ] . further , we will , if needed apply some preprocessing . first , if the problem instance contains block constraints of the form @xmath216 @xmath81 we need to have @xmath217 in this case , we fix the values of these variables , reduce the row and column - sums accordingly , and consider only the remaining blocks . for notational convenience , we assume from now on that our instance does not contain such block constraints . second , by possibly rearranging the rows and columns , we assume that @xmath218 suppose now that the problem instance has a solution . considering a horizontal strip @xmath24\times \{j , j+1\}$ ] with @xmath219\cap(2\mathbb{n}_0 + 1),$ ] we compute @xmath220\cap(2\mathbb{n}_0 + 1):v(i , j)=1\}|,\\ b_j:=|\{i\in [ m]\cap(2\mathbb{n}_0 + 1):v(i , j)=2\}|,\\ c_j:=|\{i\in [ m]\cap(2\mathbb{n}_0 + 1):v(i , j)=3\}|.\end{aligned}\ ] ] a count of the block types of the solution in that strip ( the variables for the block types are indicated in figure [ fig : types:1 ] ) , yields @xmath221 from the system we obtain @xmath222 by lemma [ lem : reduction ] there is a reduced solution ; the corresponding variable setting will be signified by the superscript @xmath193 suppose now in such a reduced solution we have @xmath223 as we can not apply any further local switch of class @xmath185 , @xmath224 we need to have @xmath225 hence by and @xmath226 and @xmath227 using and we obtain @xmath228 which imply @xmath229 a contradiction to . hence , we have @xmath230 in the following we distinguish three cases . first , suppose that in every reduced solution we have @xmath231 as we can not apply any further local switch of class @xmath183 @xmath232 we must have @xmath233 hence @xmath234 and @xmath235 eq . implies @xmath236 as @xmath237 we thus have @xmath238 further , all values of the variables are determined since yields @xmath239 for the second case , suppose that there is a reduced solution with @xmath240 and in any such solution we have @xmath241 then , @xmath242 and as we can not apply any further local switch of class @xmath243 we need to have @xmath244 hence @xmath235 eq . implies @xmath245 as @xmath246 we thus have @xmath247 further , all values of the variables are determined since yields @xmath248 finally , for the third case , suppose that there is a solution with @xmath249 then @xmath250 and @xmath251 . . implies @xmath252 as @xmath253 we thus have @xmath254 further , all values of the variables are determined since yields @xmath255 since , by assumption , the given instance has a solution and , hence a reduced one , exactly one of the three cases above must occur . which one it is , can be determined from the data , simply by testing whether we have @xmath256 in all three cases we know the particular values of the @xmath257 @xmath258 @xmath259 @xmath260 @xmath261 @xmath262 and @xmath263 therefore we know for each @xmath264 $ ] the row sums @xmath265 and @xmath266 where @xmath267 denotes the set @xmath268 in particular , we have @xmath269 a similar argument applies to the vertical strips . we can therefore decompose the reconstruction problem into the three independent subproblems @xmath212 @xmath264,$ ] which , by lemma [ lemma : mono ] , corollary [ cor : mono3 ] , and lemma [ lem : dr2 ] , are solvable in polynomial time . note that the above proof is constructive , fully detailing the steps of a polynomial - time algorithm for solving dr . we also remark that the solutions returned by the proposed algorithm are always reduced solutions . this is easily seen by ( i ) verifying that the values @xmath257 @xmath258 @xmath259 @xmath260 @xmath261 @xmath262 and @xmath270 @xmath219,$ ] in each of the three cases in ( and correspondingly for the variables in the vertical strips ) do not allow an application of any of the local switches of class @xmath183 @xmath184\setminus\{6\},$ ] because for each of the local switches there is always a corresponding variable of value zero ; and ( ii ) by noting that the proposed algorithm for @xmath204 returns , by definition , only solutions where no local switch of class @xmath271 can be applied . now we turn to uniqueness . by theorem [ thm : main1 ] , infeasibility is detected or a solution of an instance is determined in polynomial time . we can thus assume that the problem instance has a solution . clearly , a solution is unique if , and only if ( i ) there is only one reduced solution , and ( ii ) every solution is a reduced solution . the solution returned by the algorithm is a reduced solution . any two reduced solutions define the same instances of the subproblems @xmath212 @xmath264,$ ] because the problem instances are defined by means of , i.e. , the definition depends only on the particular values of the @xmath272 and @xmath273 hence , there is only one reduced solution if , and only if , the solution to each of the subproblems @xmath212 @xmath264,$ ] is unique . the conditions in lemma [ lemma : mono]([mono3 ] ) and corollary [ cor : mono3]([cormono3 ] ) , and hence uniqueness for the instances of @xmath127 and @xmath274 can be checked in polynomial time . for @xmath204 we have by lemma [ lem : unimod2 ] , a vertex solution @xmath275 of the linear program at our disposal . now we minimize the linear objective function @xmath276 over the same feasible region , which again can be done in polynomial time . the solution @xmath275 is unique if , and only if , @xmath277 hence uniqueness for @xmath204 , and therefore ( i ) , can be checked in polynomial time . suppose now that the reduced solution @xmath105 returned by the algorithm is unique among all reduced solutions , i.e. , ( i ) is satisfied . we have shown in lemma [ lem : reduction ] that every solution can be reduced by applying a sequence of local switches . reversing the sequence and the local switches ( the _ reversed local switch _ of the local switch that changes @xmath278 into @xmath279 is the switch that reverts these changes ) , we thus see that ( ii ) holds if , and only if , there is no reversed local switch that can be applied to @xmath280 there are @xmath281 possible tuples @xmath278 that need to be checked whether they form a reversed local switch , hence ( ii ) can also be checked in polynomial time . in the proof of theorem [ thm : main2 ] we use a transformation from the following @xmath0-hard problem ( see @xcite ) 1-in-3-sat + [ cols= " < , < " , ] for a given instance of 1-in-3-sat we will construct a circuit board that contains an initializer , several connectors , and clause chips . a truth assignment is transmitted through the circuit board , the clause chips ensure that the clauses are satisfied . the structure of our proof is in some ways similar to the proof from @xcite that establishes the @xmath0-hardness of the task of reconstructing lattice sets from x - rays taken in three or more directions . however , we need to deviate from @xcite in several key aspects , because the block constraints do not give a direct way of controlling points over large distances , and the elimination of `` unwanted solutions '' inside individual blocks seems also problematic . a fundamental difference to @xcite is , for instance , that we encode the boolean values for the variables by specific types of blocks , and the satisfiability of the clauses is verified via row and column sums . at first glance it seems that part of the variable assignment information is lost after verification , but it turns out that we can recover it from `` redundant '' information in our encoding . _ key ideas of the proof of theorem [ thm : main2 ] . _ before we start with the detailed proof , we illustrate the general ideas by means of an example . suppose , the instance @xmath92 of 1-in-3-sat is given by @xmath282 we will define an instance @xmath283 of ndr@xmath69 such that there is a solution for @xmath283 if , and only if , there is one for @xmath284 to this end , we construct a circuit board contained in the box @xmath285\times[34].$ ] with each @xmath286\times[34]$ ] we associate a variable @xmath287 the circuit board consists of three major types of components : an _ initializer _ , _ connectors _ , and _ clause chips_. the initializer and the connectors are rather similar . the initializer contains for every variable @xmath288 @xmath289,$ ] a so - called _ @xmath290-chip _ , the connectors contain for every variable a _ @xmath291-chip . _ the clause chips are more complex as they consist of two _ collectors _ , two _ verifiers _ , and a _ transmitter_. figure [ fig : p2example](a ) shows the circuit board for our instance . the gray areas indicate blocks that are set to zero via block constraints . by setting to zero suitable row and column sums we make sure that non - zero components @xmath292 are only possible in the white bold - framed boxes . the non - zero row and column sums are shown on the top and on the right of the figure . the bold - framed @xmath293 boxes are _ boolean chips _ , which are , according to their position on the circuit board , further classified into @xmath290 and @xmath291-chips , @xmath294.$ ] in these chips we will encode the respective boolean values for the variables @xmath290 and @xmath291 , @xmath294.$ ] all blocks , except for those set to zero , are allowed to contain at most two ones . the blocks containing the @xmath291-chips are required to contain precisely two ones . in each clause chip there are two strips playing a special role . one is a horizontal strip containing a so - called _ horizontal verifier _ , the other is a vertical strip containing a so - called _ vertical verifier _ ( see rows 25 and 26 , and columns 9 and 10 in figure [ fig : p2example](a ) ) . a key property of our construction is that the truth assignments will be in 1-to-1 correspondence with the solutions of the ndr@xmath69 instance if the row and column sum constraints related to the verifiers are left unspecified . the row and column sums constraints related to the verifiers will ensure that the truth assignments are in fact satisfying truth assignments . we can already begin to see how a truth assignment for our particular instance @xmath92 will provide a solution for @xmath295 utilizing the type 1 and type 2 blocks shown in figure [ fig : p2types ] , we set each @xmath290-chip , @xmath289,$ ] of the initializer to type 1 if @xmath296 otherwise we set it to type 2 . - and @xmath297-chips , @xmath289.$ ] in @xmath290-chips , the types 1 and 2 represent the boolean values true and false , respectively . , scaledwidth=35.0% ] let us first consider @xmath298 and the sequence of boolean chips marked a , b , c , d , and e. see also figure [ fig : transmission ] . the chips a and b lie in the same vertical strip , and since a is of type 1 , b needs to be of type 2 to satisfy the column sums in the vertical strip . chip c needs to be of type 1 to satisfy the row sums in the horizontal strip . then , the remaining points in the two vertical strips intersecting chip c are uniquely determined by the column sums . as chip d needs to contain two ones ( since the vertical strip containing chip d contains four ones , two of which need to be contained in chip e via block constraints ) we conclude from the row sums that d is of type 1 . considering now the vertical strip intersecting d we deduce that chip e is of type 2 , and therefore , together with chip a , satisfies the row sums in the horizontal strip . thus , the @xmath299-chips are consistently of type 1 , the @xmath300-chips are of type 2 . note that we have satisfied all row and column sums for the strips intersecting one of the chips a , b , c , d , and e. moreover , the block constraints in these chips are also satisfied . . , scaledwidth=45.0% ] turning to the boolean chips for @xmath301 we conclude with a similar reasoning that the @xmath302- and @xmath303-chips are consistently of type 1 and type 2 , respectively . we have now satisfied the block constraints for the @xmath302- and @xmath303-chips , additionally , we satisfied the row and column sums for the strips intersecting one of these chips . clearly , we have not violated any of the previously satisfied constraints as there is no row or column that intersects both a boolean chip for @xmath298 and @xmath304 we say that we have deduced the types of the boolean chips for @xmath305 by _ vertical transmission _ , because in our previous chain of arguments we started by considering the vertical strip containing the respective boolean chip of the initializer ( see also figure [ fig : transmission ] ) . note that such a vertical transmission does not directly work for the assignment false of the variables @xmath306 @xmath307 since the column sums do not allow us to deduce a unique solution for the respective @xmath291-chip . hence for the boolean chips for @xmath308 we resort to _ horizontal transmission _ to deduce the types of the respective @xmath290- and @xmath291-chips . in this way we conclude that the @xmath290- and @xmath291-chips are consistently type 2 and type 1 , respectively . figure [ fig : p2example](b ) shows the solution that we obtain by the previous arguments . all constraints , in particular also the row and column sums in the verifiers , are satisfied . we will later see in general that there is a single one in each of the two verifiers of the @xmath309-clause chip , @xmath310 if , and only if , exactly one of the three literals appearing in the @xmath309-th clause is true . now , consider the reverse direction for the proof , i.e. , suppose a solution @xmath105 to our instance @xmath283 of figure [ fig : p2example](a ) is given . note that if we can ensure that the types of the boolean chips in the initializer are either 1 or 2 then by the previous arguments we have a satisfying truth assignment for @xmath92 by setting @xmath311 @xmath294,$ ] if , and only if , the corresponding @xmath290-chip of the initializer is of type 1 . the respective block , row and column sum constraints allow only the four possible types of blocks for the boolean chips of the initializer shown in figure [ fig : p2types ] . the types 3 and 4 , however , are ruled out by the following `` global '' argument : suppose , for some @xmath289,$ ] the @xmath290-chip of the initializer is of type @xmath312 by horizontal transmission , we then conclude that the @xmath291-chips need to be of type 1 ( by transmitting , as before in a unique way , through each clause chip ) . by vertical transmission , however , we conclude that the @xmath291-chips are of type 2 , which is a contradiction . we remark that effectively we utilize the noisy block constraints at three different places in our construction . they are needed , because we can not prescribe the exact number of ones in the @xmath290-chips of the horizontal and vertical collectors , and in the @xmath313-configurations that are introduced later ( see also figures [ fig : p2collectors ] , [ fig : p2collectorfilled1 ] , and [ fig : p2collectorfilledx ] ) . having illustrated the key ideas behind the construction , we now turn to the formal proof . let in the following @xmath314 denote an instance of 1-in-3-sat . we will define an instance @xmath283 of ndr@xmath69 such that there is a solution for @xmath283 if , and only if , there is one for @xmath284 the circuit board will be contained in the box @xmath315\times[n]:=[s(6t+2)+2t]^2.\ ] ] with each @xmath316\times[n]$ ] we associate a variable @xmath287 figure [ fig : p2layout ] illustrates the general layout of the construction . ) . non - zero components @xmath292 of a solution will be possible only for those @xmath120 contained in the white area.,scaledwidth=43.0% ] we define the instance @xmath317 of ndr@xmath69 by specifying the row sums @xmath272 the column sums @xmath318 and the block constraints . for notational convenience we define the block constraints via a function @xmath319 in this notation , @xmath320 @xmath321 denotes the block constraint @xmath322 while @xmath323 signifies the block constraint @xmath324\cap \mathbb{z}).\ ] ] the set @xmath49 of reliable block constraints for @xmath283 is then @xmath325 the different components of the circuit board are placed at specific positions . for a more compact definition of these positions we set @xmath326_0.\ ] ] turning to the definition of the initializer , we remark that the initializer is contained inside the box @xmath327_0 ^ 2.$ ] we will specify the row and column sums when the corresponding connectors are introduced . in terms of the block constraints we define for every @xmath328_0 ^ 2:$ ] @xmath329 the block @xmath330 @xmath289,$ ] will be called _ @xmath331-chip ( of the initializer)_. for an illustration of an initializer , see the bottom right @xmath332\times[8]$]-box in figure [ fig : p2example](a ) . we define @xmath333 connectors . each connector is contained inside a box @xmath334_0 ^ 2 $ ] with @xmath335_0.$ ] in terms of block constrains we define for every @xmath336_0 $ ] and @xmath337_0 ^ 2:$ ] @xmath338 for the row and column sums we define for every @xmath336_0 $ ] and @xmath339_0:$ ] @xmath340 the block @xmath341 @xmath342_0,$ ] @xmath289,$ ] will be called _ @xmath343-chip ( of the @xmath309-th connector)_. for an illustration of a connector , see the bottom left @xmath332\times[8]$]-box in figure [ fig : p2example](a ) . next we define the @xmath344 clause chips . each clause chip is contained in a box @xmath345_0 ^ 2 $ ] with @xmath346_0.$ ] the box @xmath347_0\times[4t-1]_0 $ ] is a _ ( vertical ) collector _ , the box @xmath348_0\times[4t-1]_0 $ ] is a _ ( vertical ) verifier _ , the box @xmath349_0 ^ 2 $ ] is a _ transmitter _ , the box @xmath350_0\times[1]_0 $ ] is a _ ( horizontal ) verifier _ , and the box @xmath351_0\times[2t-1]_0 $ ] is a _ ( horizontal ) collector_. for an illustration of a clause chip , see the top left @xmath352\times[26]$]-box in figure [ fig : p2example](a ) . for @xmath353_0 $ ] we define the row and column sums for the verifiers ( we will call these row and column sums _ verifier sums _ ) by @xmath354 the remaining row and columns sums for the transmitters , i.e. , for @xmath353_0 $ ] and @xmath355_0 $ ] are defined by @xmath356 note that sums for the rows and columns meeting the collectors are already defined as there is a corresponding connector in the same row or , respectively , column ( see also figure [ fig : p2layout ] ) . we have thus completed our definition of the row and column sums of our instance and continue now with the remaining block constraints . for the vertical collector , we define for every @xmath353_0 $ ] and @xmath328_0 ^ 2:$ ] @xmath357 the box @xmath358_0\times[1]_0,$ ] @xmath359_0,$ ] will be called _ @xmath290-chip ( of the @xmath360-th vertical collector)_. for the horizontal collector we define similarly for every @xmath353_0 $ ] and @xmath361_0 ^ 2:$ ] @xmath362 the box @xmath363_0\times[1]_0,$ ] @xmath359_0,$ ] will be called _ @xmath290-chip ( of the @xmath360-th horizontal collector)_. note that the @xmath290-chips , @xmath289,$ ] of the collectors are not contained in blocks . in fact , each such @xmath290-chip intersects two blocks ( which , in turn , are allowed to contain at most two ones ) . for any @xmath289 $ ] we will also refer to the @xmath290- and @xmath291-chips on the circuit board as _ boolean chips . _ for the empty space above the vertical connector , i.e. , for @xmath353_0 $ ] and @xmath364_0 ^ 2,$ ] we define @xmath365 now we turn to the block constraints for the clause - dependent part of chip , i.e. , the verifiers and the transmitters . with @xmath366 and @xmath367 @xmath368,$ ] we denote the indices of the variables that appear unnegated and , respectively , negated in the @xmath309-th clause . for instance , for @xmath369 ( regarded as the first clause ) we have @xmath370 and @xmath371 for the vertical verifiers we define for every @xmath335,$ ] @xmath294 , $ ] and @xmath372:$ ] @xmath373 similarly , for the horizontal verifiers we define for every @xmath335,$ ] @xmath294,$ ] and @xmath374:$ ] @xmath375 for the transmitters we define for every @xmath335 $ ] and @xmath376:$ ] @xmath377 the remaining block constraints in the transmitter are set to zero , i.e. , for every @xmath335,$ ] @xmath378_0 ^ 2 $ ] with @xmath379 we set @xmath380 for @xmath368 $ ] and @xmath381 $ ] we refer to the sets of points @xmath382 as _ @xmath313-configuration_. figure [ fig : p2collectors ] gives an illustration . outside the initializer , connectors , and clause chips we set everything to zero , i.e. , for every @xmath383\times[n])\cap c(m , n,2)$ ] with @xmath384_0 ^ 2)\cup\bigcup_{s\in[s-1]_0}\left(((a_s , a_s)+[2(t-1)]_0 ^ 2)\cup ( ( a_s , a_s+2t)+[6t+1]_0 ^ 2))\right)\ ] ] we set @xmath385 this concludes the formal definition of the instance @xmath295 we shall now show that @xmath283 admits a solution if , and only if , the 1-in-3-sat instance @xmath92 admits a solution . let @xmath386\times[n]$ ] denote the set of points of the boolean chips and @xmath313-configurations , @xmath387\times[t].$ ] * claim 1 : * in any solution @xmath105 of @xmath283 we have @xmath388 for every @xmath389\times[n])\setminus \mathcal{g}.$ ] to see this , we first remark that the claim holds for all points outside each clause chip since the only blocks there that are not set to zero are the boolean chips of the initializer and connectors . for the points inside a clause chip , we need to distinguish three cases since there are three differently structured @xmath313-configurations , in which different blocks are set to zero . they are shown in figure [ fig : p2collectors ] . the figure shows , however , that in all three cases it holds that the points outside the corresponding boolean chips and the @xmath313-configuration are either set to zero by block constraints or by zeros of suitable row or column sums . this shows the claim . * claim 2 : * in any solution @xmath105 of @xmath283 every @xmath290- and @xmath291-chip , @xmath289,$ ] is of one of the types @xmath390 shown in figure [ fig : p2types ] . for this , consider a fixed index @xmath381.$ ] each @xmath290-chip is contained in a horizontal or vertical strip that contains , except for a @xmath291-chip , no other points of @xmath391 since we required via block constraints that each @xmath291-chip contains two ones and since the respective row or column sums in the strip sum up to four there need to be exactly two ones also in each @xmath290-chip . there are four possibilities to place two ones into a @xmath293 box if it is required ( as is the case via our row and column sums ) that no two ones are in the rightmost column and upper row of the box . the four possibilities are shown in figure [ fig : p2types ] . * claim 3 : * in any solution @xmath105 of @xmath283 it holds that for every @xmath294 $ ] the @xmath290-chips are either all of type 1 or type 2 . to show this , consider again a fixed index @xmath381.$ ] let @xmath392 denote the sequence with ( i ) @xmath393 is the @xmath290-chip of the initializer , ( ii ) @xmath394 is the @xmath291-chip of the @xmath395-th connector , @xmath396,$ ] ( iii ) @xmath397 is the @xmath290-chip of the @xmath309-th vertical collector , @xmath368,$ ] ( iv ) @xmath398 is the @xmath313-configuration of the @xmath309-th clause , @xmath368,$ ] and ( v ) @xmath399 is the @xmath290-chip of the @xmath309-th horizontal collector , @xmath368.$ ] by vertical transmission ( see figure [ fig : p2collectorfilled1 ] ) and claim 2 , we have the following implication : if a @xmath290-chip of @xmath400 @xmath401,$ ] is of type 1 , 3 or @xmath402 then the @xmath290-chip of @xmath403 is of type 1 , and the points of @xmath404 are also uniquely determined ( here , @xmath405 ) . block . ( a , b , c ) transmission through the three differently structured @xmath313-configurations . ( d , e ) transmission through a connector.,scaledwidth=90.0% ] by horizontal transmission ( see figure [ fig : p2collectorfilledx ] ) and claim 2 , we have the following implication : if a @xmath290-chip of @xmath400 @xmath401,$ ] is of type 2 , 3 or @xmath402 then the @xmath290-chip of @xmath406 is of type 2 , and the points of @xmath407 are also uniquely determined ( here , @xmath408 ) . block . ( a , b , c ) transmission through the three differently structured @xmath313-configurations . ( d , e ) transmission through a connector.,scaledwidth=90.0% ] since @xmath409 we have proved claim 3 , and , in particular , we have ruled out that any @xmath290-chip is of type 3 and 4 . the types 1 and 2 can be associated with boolean values . in @xmath290-chips , and in particular those of the initializer , we associate with type 1 the value true and with type 2 the value false . * claim 4 : * there is a 1-to-1 correspondence between the ( not - necessarily satisfying ) truth - assignments for @xmath92 and the solutions @xmath105 to the @xmath410-variant of @xmath283 that allows violation of the verifier sums . by construction , there is neither a horizontal nor a vertical strip that intersects the boolean chips for two variables @xmath411 @xmath412 with @xmath413 the arguments from claim 3 are thus independently valid for each @xmath289.$ ] we can hence consider an arbitrary ( not - necessarily satisfying ) truth - assignment for @xmath414 fill in the corresponding types in the initializer and extend them to yield a solution to the @xmath410-variant of @xmath283 with unspecified verifier sums . vice versa , by reading off the chip types in the initializer of a solution to @xmath410 yields a truth - assignment for @xmath284 we have already seen in the proof of claim 3 that the type of each @xmath290-chip , @xmath289,$ ] in the initializer determines all ones in the corresponding @xmath291-chips and @xmath313-configurations , @xmath335.$ ] hence , the correspondence is indeed 1-to-1 . * claim 5 : * the satisfying truth - assignments of @xmath92 are in 1-to-1 correspondence with the solutions of @xmath295 this can now be seen by taking into account that the @xmath290-chip , @xmath289,$ ] of the @xmath309-th horizontal and vertical verifier , @xmath368,$ ] contributes a one to the horizontal and , respectively , vertical verifier in the @xmath309-th clause chip if , and only if , the @xmath290-chip is true ( is of type 1 ) with @xmath415 or the @xmath290-chip is false ( is of type 2 ) with @xmath416 ; see figures [ fig : p2collectors ] , [ fig : p2collectorfilled1](a , b , c ) , and [ fig : p2collectorfilledx](a , b , c ) . the verifier sums thus ensure that exactly one literal in the @xmath309-th clause is true . this , together with claim 4 , proves the claim . by claim 5 it holds that @xmath283 admits a solution if , and only if , @xmath92 admits a solution . finally , we note that the transformation runs in polynomial time . in section [ sect : notation ] we remarked that @xmath68 can be viewed as version of @xmath187 where small `` occasional '' uncertainties in the gray levels are allowed . reviewing our proof of theorem [ thm : main2 ] , the term `` occasional '' can be quantified as meaning `` of the order of the square root of the number of blocks . '' let us further remark that the arguments in the above proof do not rely on the particular value of @xmath53 in the noisy block constraints . these constraints can be replaced by any other types of block constraints as long as they allow for 0 , 1 , and 2 ones to be contained in each of the corresponding blocks . by this observation it is possible to adapt our proof to several other contexts where the reconstruction task involves other types or combinations of block constraints ; see @xcite . finally , note that the transformation in the proof of theorem [ thm : main2 ] is parsimonious ( see claim 5 in the above proof ) . hence the problem of deciding whether a given solution of an instance of ndr@xmath69 with @xmath53 has a non - unique solution is @xmath0-complete ; for an @xmath0-hardness proof of @xmath417 see ( * ? ? ? * lemma 4.2 ) . we use a transformation from ndr@xmath418 let @xmath419 denote an instance of ndr@xmath418 we set @xmath420 and @xmath421 for @xmath422\times[n])\cap c(m , n,2)$ ] and @xmath423 $ ] we define @xmath424,\\ 0&:&\textnormal{otherwise,}\end{array}\right.\\ c'_{\frac{k}{2}(i-1)+l}&:=\left\{\begin{array}{lll}c_{i+l}&:&l\in[2],\\ 0&:&\textnormal{otherwise,}\end{array}\right.\\ v'(k(i-1)/2 + 1,k(j-1)/2 + 1))&:=v(i , j),\end{aligned}\ ] ] and @xmath425 this defines an instance @xmath426 of nsr@xmath427 clearly , the instance @xmath92 of ndr@xmath69 admits a solution if , and only if , @xmath283 of nsr@xmath44 admits a solution ( by filling / extracting the @xmath428-blocks of @xmath92 into / from the @xmath3-blocks of @xmath283 ) . this is a polynomial - time transformation , proving the theorem . recall that in the proof of theorem [ thm : main1 ] we have made extensive use of the concept of local switches . each such switch involves four points . figure [ fig : largeswitchingcomp ] shows , on the other hand , two solutions of a dr instance that differ by @xmath429 pixels , and for which it is easily checked that they are the only two solutions for that instance ( such a structure has been utilized in the proof of theorem [ thm : main2 ] ) . in this case there is thus no sequence of small switches transforming one solution into the other . the existence of such large irreducible switches is well - known for the @xmath0-hard reconstruction problem from more than two directions @xcite ( and would follow independently from @xmath430 ) . the problem @xmath431 however , admits large switches albeit is polynomial - time solvable . in addition to the block constraints our computational tasks rely only on row and column sums , i.e. , projection data from two viewing angles . for more than two viewing directions we obtain the @xmath0-hardness of nsr@xmath44 for any @xmath43 and @xmath53 directly from the construction in @xcite . the @xmath0-hardness of the corresponding variants of nsr@xmath432 for @xmath433 can also be derived from the construction in @xcite . first the construction is thinned out so as to make sure that each block contains at most one element of the grid of candidate points . then , a second copy with inverted colors is interwoven by applying a translation by @xmath434 or @xmath435 this ensures that each non - empty block contains exactly one one . g. van gompel , k. j. batenburg , e. van de casteele , w. van aarle , and j. sijbers . a discrete tomography approach for superresolution micro - ct images : application to bone . in _ 2010 ieee intern . biomedical imaging : from nano to macro _ , pages 816819 , 2010 . p. gritzmann . on the reconstruction of finite lattice sets from their x - rays . in _ discrete geometry for computer imagery ( eds . : e. ahronovitz and c. fiorio ) , dcgi97 , lecture notes on computer science 1347 , springer _ , pages 1932 , 1997 . c. kisielowski , p. schwander , f. h. baumann , m. seibt , y. kim , and a. ourmazd . an approach to quantitative high - resolution transmission electron microscopy of crystalline materials . , 58(2):131155 , 1995 . p. schwander , c. kisielowski , f. h. baumann , y. kim , and a. ourmazd . mapping projected potential , interfacial roughness , and composition in general crystalline solids by quantitative transmission electron microscopy . , 71(25):41504153 , 1993 . j. zhu , j. gao , a. ehn , m. aldn , z. li , d. moseev , y. kusano , m. salewski , a. alpers , p. gritzmann , and m. schwenk . measurements of 3d slip velocities and plasma column lengths of a gliding arc discharge . , 106(4):04410114 , 2015 .

we propose and investigate an exactly solvable model of non - equilibrium luttinger liquid on a star graph , modeling a multi - terminal quantum wire junction . the boundary condition at the junction is fixed by an orthogonal matrix , which describes the splitting of the electric current among the leads . the system is driven away from equilibrium by connecting the leads to heat baths at different temperatures and chemical potentials . the associated non - equilibrium steady state depends on and is explicitly constructed . in this context we develop a non - equilibrium bosonization procedure and compute some basic correlation functions . luttinger liquids with general anyon statistics are considered . the relative momentum distribution away from equilibrium turns out to be the convolution of equilibrium anyon distributions at different temperatures . both the charge and heat transport are studied . the exact current - current correlation function is derived and the zero - frequency noise power is determined . l s 1_1 o * v * * luttinger liquid in non - equilibrium + steady state * + mihail mintchev and paul sorba + 1.5 truecm _ istituto nazionale di fisica nucleare and dipartimento di fisica delluniversit di pisa , largo pontecorvo 3 , 56127 pisa , italy + _ laboratoire de physique thorique dannecy - le - vieux , cnrs , + 9 , chemin de bellevue , bp 110 , f-74941 annecy - le - vieux cedex , france

1210.5409

the universal features of a large class of one - dimensional quantum models , exhibiting gapless excitations with linear spectrum , are successfully described @xcite by the tomonaga - luttinger ( tl ) liquid theory @xcite-@xcite . this theory applies to various systems , including nanowire junctions and carbon nanotubes , which are available nowadays in experiment @xcite-@xcite . for this reason the study of non - equilibrium phenomena in the tl liquid phase attracts recently much attention @xcite-@xcite . a typical non - equilibrium setup , considered in the literature , is the junction of two or more semi - infinite leads with electrons at different temperatures and/or chemical potentials . the junction is an interval of finite length @xmath5 , where the electrons injected from the leads interact among themselves . this interaction drives the system away from equilibrium . differently from the equilibrium tl liquid on the line , the non - equilibrium model defined in this way , is not exactly solvable . nevertheless , it is extensively studied @xcite-@xcite by various methods , including linear response theory , bosonization combined with the non - equilibrium keldish formalism and perturbation theory . one of the main goals of the present paper is to explore the possibility to construct and analyze an alternative _ exactly solvable _ model for a non - equilibrium tl junction . since the universal features of such a system are expected to manifest themselves in the critical ( scale invariant ) limit , it is natural to shrink the domain of the non - equilibrium interaction to a point , taking @xmath6 . for a complete description of the critical regime it is essential to take into account all point - like interactions , which ensure a unitary time evolution of the system . these interactions can be parametrized by a scattering matrix @xmath0 localized in the junction point , as shown in the multi - terminal setup displayed in fig.[fig1 ] . ( 600,120)(40,250 ) and @xmath7 semi - infinite leads , connected at infinity to thermal reservoirs with temperatures @xmath8 and chemical potentials @xmath9.,title="fig : " ] each lead contains a tl liquid , which at infinity is in contact with a heat reservoir with ( inverse ) temperature @xmath8 and chemical potential @xmath9 . our first step below is to show that there exists a non - equilibrium steady state ( ness ) , which describes the tl configuration in fig.[fig1 ] . this state is characterized by non - trivial time independent electric and heat currents , flowing in the leads . the scattering matrix @xmath0 is implemented by imposing specific boundary conditions at the junction . it turns out that the boundary conditions , which describe the splitting of the electric steady current in the junction , lead to an exactly solvable problem . in fact , we establish the operator solution in this case and investigate the relative non - equilibrium correlation functions in the ness representation . the tl theory has been introduced originally @xcite-@xcite for describing fermion systems . it has been understood later on @xcite-@xcite that the fermion tl liquid is actually an element of a more general family of _ anyon _ tl liquids , which obey abelian braid statistics . in this paper we explore the general anyon tl liquid , obtaining the conventional fermionic and bosonic ones as a special cases . from the two - point anyon correlation functions we extract the ness distribution of the tl anyon excitations . in momentum space this non - equilibrium distribution is a _ nested convolution _ of equilibrium distributions at different temperatures and chemical potentials . as expected , the convolution depends on the scattering matrix @xmath0 , which drives the system away from equilibrium . we investigate also the ness correlators of the electric and energy currents , describing in detail the charge and heat transport in the junction . the zero - frequency noise power is deduced from the two - point current - current correlation function , whose exact expression in terms of hypergeometric functions is established . we prove the breakdown of time reversal invariance as well . the paper has the following structure . in the next section we construct non - equilibrium chiral fields in a ness on a star graph modeling the junction . we derive here the non - equilibrium casimir energy and the heat current and compare the latter with the conformal field theory result . in section 3 we develop a non - equilibrium finite temperature operator bosonization procedure . we also establish the operator solution , subject to the current splitting boundary condition at the junction . we show that this condition covers two different physical situations , corresponding to a junction with and without charge dissipation . the non - equilibrium correlation functions are investigated in section 4 , where the anyon ness distributions are derived . the charge and heat transport as well as the noise are also studied there . section 5 provides a concise outlook of the paper and contains some general observations . the appendix collects some results about the asymptotic properties of the anyon ness correlators . the fundamental building blocks of bosonization away from equilibrium are the free massless scalar field @xmath10 and its dual @xmath11 . the fields @xmath10 and @xmath11 propagate on a star graph @xmath12 , which is shown in fig . [ fig2 ] and models the quantum wire junction . ( 500,70)(-150,20 ) with @xmath7 edges modelling the junction of @xmath7 quantum wires.,title="fig : " ] the edges @xmath13 are half - lines and each point @xmath14 in the bulk @xmath15 of @xmath12 is uniquely determined by its coordinates @xmath16 , where @xmath17 is the distance to the vertex @xmath18 and @xmath19 labels the edge . besides the massless klein - gordon equation , the fields @xmath10 and @xmath11 satisfy the duality relations @xmath20 the initial conditions are fixed by the equal - time canonical commutation relations @xmath21_{{}_- } = [ { \widetilde{\varphi}}(t , x , i)\ , , \ , { \widetilde{\varphi}}(t , y , j)]_{{}_- } = 0 \ , , \label{ecc1}\ ] ] @xmath22_{{}_- } = [ ( \der_t{\widetilde{\varphi}})(t , x , i)\ , , \ , { \widetilde{\varphi}}(t , y , j)]_{{}_- } = -\ri \delta_{ij}\delta ( x - y ) \ , . \label{ecc2}\ ] ] in order to determine the dynamics completely , one must impose some boundary conditions at the vertex @xmath23 . these conditions are conveniently formulated in terms of the combinations @xmath24 which depend on @xmath25 and @xmath26 respectively and define right and left chiral fields @xmath27 on @xmath12 . the most general _ scale invariant _ boundary conditions , generating a _ unitary time evolution _ of @xmath10 and @xmath11 , are parametrized by the orthogonal group @xmath28 and read @xcite-@xcite @xmath29 these simple conditions capture the universal features of the system and @xmath0 has a straightforward physical interpretation : the vertex @xmath18 of @xmath12 represents a _ scale invariant point - like defect _ , @xmath0 being the associated scattering matrix . our next step is to construct a steady state @xmath30 , which captures the evolution of the chiral fields @xmath27 on @xmath12 , whose edges are attached at infinity to thermal reservoirs at inverse temperatures @xmath8 as shown in fig . [ fig1 ] . in the boson case we take all chemical potentials to be equal . ] , setting @xmath31 in all reservoirs . the system is away from equilibrium if @xmath0 contains at least one non - trivial transmission coefficient among reservoirs with different temperature . the construction of @xmath30 , described below , follows the scheme developed in @xcite and is based on scattering theory . it adapts to the case under consideration some modern ideas @xcite-@xcite about ness . the framework is purely algebraic and generalizes the definition @xcite of equilibrium gibbs state over the algebra of canonical commutation relations ( ccr ) . we start by observing that the massless klein - gordon equation and the relations ( [ dual ] ) lead to the following representation @xmath32 \ , , \label{fil1 } \\ & & { \varphi_{i , l}}(\xi ) = \int_0^\infty \frac{{{\rm d}}k}{\pi \sqrt{2}}\ , \sqrt { \delta_\lambda ( k ) } \left [ a_i^\ast ( -k ) \e^{\ri k \xi } + a_i(-k ) \e^{-\ri k \xi}\right ] \ , , \label{fil2}\end{aligned}\ ] ] @xmath33 being some distribution to be fixed below . using that @xmath0 is a real matrix , the boundary condition ( [ bc1 ] ) implies the constraints @xmath34 where @xmath35 @xmath36 is the heaviside step function and @xmath37 indicates the transpose of @xmath0 . from the equal - time commutation relations ( [ ecc1 ] , [ ecc2 ] ) one infers that the elements @xmath38 generate the following deformation @xmath39 of the algebra of ccr : @xmath40_{{}_- } = [ a^*_i ( k)\ , , \ , a^*_j ( p)]_{{}_- } = 0 \ , , \label{rta1}\ ] ] @xmath41_{{}_- } = 2\pi [ \delta ( k - p)\delta_{ij } + s_{ij}(k)\delta(k+p ) ] \ , . \label{rta2}\ ] ] moreover , ( [ ecc1 ] , [ ecc2 ] ) imply that @xmath42 there exist a one - parameter family of tempered distributions , which solve this equation in @xmath43 . a convenient representation of this family is given by @xcite @xmath44\ , , \label{distr2}\ ] ] where @xmath45 is a free parameter with dimension of mass , having well - known infrared origin . the above structure is very general and equations ( [ fil1],[fil2 ] ) apply to any representation of the algebra @xmath39 , which is a simplified version of the so called reflection - transmission ( rt ) algebra @xcite-@xcite , describing factorized scattering in integrable models with point - like defects in one dimension . the fock and the gibbs state over @xmath39 describe equilibrium physics and have been largely explored . we will investigate here the ness @xmath30 , which describes the physical situation shown in fig . [ fig1 ] . for this purpose we first observe that the sub - algebras @xmath46 and @xmath47 , generated by the elements @xmath48 and @xmath49 respectively , parametrize the _ asymptotic incoming _ and _ outgoing _ fields . accordingly , both @xmath46 and @xmath47 are conventional ccr algebras ; in fact the @xmath50 term in ( [ rta2 ] ) vanishes if both momenta are negative or positive . it is worth stressing that ( [ constr1 ] ) relate @xmath46 with @xmath47 and that the whole rt algebra @xmath39 can be generated via ( [ constr1 ] ) either by @xmath46 , or by @xmath47 . the main idea for constructing @xmath51 is based on this kind of asymptotic completeness property . starting with an equilibrium state on @xmath46 , we will extend it by means of ( [ constr1 ] ) to a non - equilibrium state on the whole algebra @xmath39 . for this purpose we introduce the edge hamiltonian and number operators @xmath52 which describe the _ asymptotic _ dynamics at @xmath53 ( i.e. before the interaction ) in terms of @xmath46 . defining @xmath54 we introduce the equilibrium gibbs state over @xmath46 in the standard way @xcite . for any polynomial @xmath55 over @xmath46 we set @xmath56\ , , \qquad \qquad \qquad \label{def1}\end{aligned}\ ] ] where @xmath57 and @xmath58 . all the expectation values ( [ def1 ] ) can be computed @xcite by purely algebraic manipulations and can be expressed in terms of the two - point functions , which are written in terms of the familiar bose distribution @xmath59}}{1- \e^{-\beta_i [ |k| - \mu_b ] } } \label{fbd1}\ ] ] in the following way @xmath60 @xmath61 \delta_{ij } 2\pi \delta ( k - p)\ , . \label{2b}\ ] ] we stress that ( [ 2a],[2b ] ) hold on @xmath46 , i.e. only for _ negative _ momenta . the common for all reservoirs chemical potential @xmath62 allows to avoid in ( [ 2a],[2b ] ) the infrared singularity at @xmath63 . we anticipate that @xmath64 has nothing to do with the _ fermion _ chemical potentials , appearing in the non - equilibrium bosonization procedure described in the next section , where the limit @xmath65 exist and will be performed . the next step is to extend ( [ def1]-[2b ] ) to the whole rt algebra @xmath39 , namely to _ positive _ momenta . employing ( [ constr1 ] ) one finds @xmath66 \delta ( k - p ) \nonumber \\ + \bigl [ \theta(-k)b_i(k)\ , \s^t_{ij } + \theta(k)\s_{ij}\ , b_j(k ) \bigr ] \delta ( k+p ) \bigr\}\ , . \qquad \ ; \ ; \ , \label{cor1}\end{aligned}\ ] ] the expression for @xmath67 is obtained from ( [ cor1 ] ) by the substitution @xmath68 } } \ , . \label{fbd2}\ ] ] the final step is to compute a generic correlation function . by means of the commutation relations ( [ rta1],[rta2 ] ) , this problem is reduced to the evaluation of correlators of the form @xmath69 , which can be computed by iteration via @xmath70 we would like to mention in conclusion that the use of the rt algebra @xmath39 in the construction of @xmath71 represents only a convenient choice of coordinates , which has a simple physical interpretation in terms of scattering data and applies to a variety of systems @xcite with point - like defects . in order to illustrate the physical properties of @xmath72 , it is instructive to investigate the non - equilibrium energy density and transport associated with the scalar field @xmath10 . the equations of motion imply the conservation @xmath73 of the energy - momentum tensor @xmath74:(t , x , i)\ , , \label{emtph1}\\ \theta_{xt}(t , x , i ) & = & \frac{1}{2 } : \left [ ( { \partial}_x { \varphi } ) ( { \partial}_t { \varphi } ) - { \varphi}({\partial}_x { \partial}_t { \varphi})\right ] : ( t , x , i)\ , , \label{emtph2}\end{aligned}\ ] ] where @xmath75 denotes the normal product in the algebra @xmath39 . the boundary condition ( [ bc1 ] ) implies the kirchhoff rule @xmath76 which , combined with ( [ econs ] ) , ensures energy conservation . the derivation of @xmath77 and @xmath78 is based on the expectation value @xmath79 + \s_{i_1i_2 } b_{i_2}(k ) \cos[k(t_{12}-{\widetilde{x}}_{12 } ] + \nonumber \\ b_{i_1}(k)\ , \s^t_{i_1i_2 } \cos[k(t_{12}+{\widetilde{x}}_{12 } ] + \sum_{j=1}^n \s_{i_1j}\ , b_j(k)\ , \s^t_{ji_2 } \cos[k(t_{12}-x_{12}]\bigr \}\ , , \label{ncf}\end{aligned}\ ] ] where @xmath80 . plugging ( [ ncf ] ) in the definitions ( [ emtph1],[emtph2 ] ) , one gets @xmath81 where @xmath82 is the dilogarithm function . ( [ tr1 ] ) determines the casimir energy , whereas ( [ tr2 ] ) describes the energy ( heat ) transport . both are time - independent , thus confirming that we are dealing with a steady state . notice also that the energy density is @xmath83-dependent , which reflects the breaking of translation invariance by the junction and is consistent with the conservation law ( [ econs ] ) . since @xmath0 is an orthogonal matrix , @xmath84 obviously satisfies the kirchhoff s rule ( [ k1 ] ) . the energy density @xmath85 can be written in the equivalent form @xmath86 where the last term vanishes at equilibrium and describes therefore the _ non - equilibrium contribution _ to the casimir energy . for a junction with @xmath87 wires there are two one - parameter families @xmath88 of scattering matrices with @xmath89 . for both families one finds @xmath90\ , , \label{ctr1}\ ] ] which deserves a comment . the heath transport has been investigated recently in the framework of conformal field theory ( cft ) for generic central charge @xmath91 in @xcite . ( [ ctr1 ] ) confirms the result of @xcite for @xmath92 and extends this result in two directions : imperfect junction with transmission probability @xmath93 and @xmath94 . we observe in this respect that derivation of the dilogarithm terms in ( [ ctr1 ] ) is problematic in a cft context , because both @xmath95 and @xmath96 are non - trivial dimensionless parameters . in the limit @xmath65 the @xmath83-dependent integral in ( [ tr1 ] ) can be evaluated explicitly and one finds @xmath97 ^ 2}\ , , \label{tr3 } \\ \t_i(\beta , 0 ) & = & \frac{\pi}{12}\sum_{j=1}^n \left ( \delta_{ij } - \s_{ij}^2 \right ) \frac{1}{\beta_j^2}\ , . \qquad \label{tr4}\end{aligned}\ ] ] in the non - equilibrium bosonization procedure , developed below , we will need the correlation functions of the chiral fields ( [ fil1],[fil2 ] ) ) in the ness @xmath71 . it is easily seen that all of them can be expressed in terms of the distribution @xmath98}}{1- \e^{-\beta [ k - \mu_b ] } } \e^{\ri k \xi } + \frac{1}{1- \e^{-\beta [ k - \mu_b ] } } \e^{-\ri k \xi}\right ] \ , . \label{w}\ ] ] the full @xmath99-dependence and the singularity at @xmath100 of ( [ w ] ) are captured by @xcite @xmath101\right \ } + o(\mu_b)\ , , \label{wa}\ ] ] where @xmath102 stands for @xmath99-independent terms , which vanish in the limit @xmath65 . it is convenient at this point to relate the ( up to now free ) infrared regularization parameter @xmath103 to @xmath64 by means of @xmath104 the limit @xmath65 in ( [ w ] ) now exists and gives the distribution prescription , adopted throughout the paper , indicates as usual the weak limit @xmath105 . ] @xmath106 \ , , \label{w1}\ ] ] which is the fundamental block of the ness chiral correlation functions . in fact , for the two - point correlators one gets @xmath107 where @xmath108 . as expected , the point - like interaction in the vertex of @xmath12 induce a non - trivial left - right mixing described by ( [ lr],[rl ] ) . the possibility to express fermions in terms of bosons in 1 + 1 dimensional space - time has been discovered long ago by jordan and wigner @xcite . the bosonization technique in the fock representation of the fields @xmath10 and @xmath11 has been applied for solving the tomonaga - luttinger ( tl ) model in @xcite-@xcite . the framework has been extended later @xcite to the finite temperature gibbs representation of @xmath10 and @xmath11 . both the fock and gibbs representations describe equilibrium physics . our goal in what follows we will to apply the ness representation , constructed in the previous section , for investigating the non - equilibrium tl liquid in the multi - terminal configuration shown in fig . [ fig1 ] . the bulk dynamics is governed by the tl lagrangian density @xmath109 where @xmath110 are complex fermion fields , @xmath111 is the fermi velocity and @xmath112 are the coupling constants . the bulk theory has an obvious @xmath113 symmetry . in fact , the lagrangian density ( [ lagr ] ) is left invariant by the two independent phase transformations @xmath114 implying the current conservation laws @xmath115 where the charge and current densities are given by @xmath116(t , x , i)\ , , & z = l\ , , \\ [ \psi^*_2\psi_2](t , x , i)\ , , & z = r\ , , \end{cases } \qquad j_{z } ( t , x , i ) = \begin{cases } [ \psi^*_1\psi_1](t , x , i)\ , , & z = l\ , , \\ -[\psi^*_2\psi_2](t , x , i)\ , , & z = r\ , . \end{cases } \label{lrcurrents}\ ] ] the currents @xmath117 have simple physical meaning : @xmath118 and @xmath119 represent the particle excitations moving along the edge @xmath13 towards and away of the vertex @xmath18 respectively . interpreting the vertex as a defect , which can be characterized by some scattering matrix , the currents @xmath120 and @xmath121 describe therefore the _ incoming _ and _ outgoing _ flows . it is well known @xcite-@xcite that the tl model ( [ lagr ] ) is exactly solvable on the line @xmath43 . on the graph @xmath12 the situation is more involved , because one should take into account the boundary conditions in the vertex @xmath18 . the conditions @xmath122 which work in the free case @xmath123 , do not lead @xcite to exactly solvable problem after switching on the tl interactions . the fact that on @xmath43 the quartic bulk interactions in ( [ lagr ] ) are solved exactly via bosonization suggest to try boundary conditions which , differently from ( [ bc2 ] ) , are formulated in terms of real boson fields . in this spirit and according our previous comments on the chiral currents ( [ lrcurrents ] ) , it is quite natural to consider @xmath124 which has been proposed and explored first in the two - terminal case in @xcite . an advantage of ( [ bc3 ] ) is the direct interpretation in terms of _ gauge invariant physical observables _ , which represent the basic building blocks of algebraic quantum field theory ( see e.g. @xcite ) . in fact , ( [ bc3 ] ) describes the splitting in the vertex @xmath18 of the outgoing current @xmath125 along the edge @xmath13 in incoming currents @xmath126 along the edges @xmath127 . for this reason we refer to @xmath128 the as the _ current splitting matrix _ and show in the next subsection that @xmath128 actually coincides with the boson scattering matrix @xmath0 . referring for the details to @xcite , we recall here the anyon operator solution of the tl model on a star graph @xmath12 . the solution provides a _ unified description _ of all anyon luttinger liquids and is expressed in terms of the chiral fields ( [ fil1],[fil2 ] ) and the parameters @xmath129 and the sound velocity @xmath130 as follows : @xmath131}:\ , , \label{psi1}\\ \psi_2(t , x , i ) & = & \eta_i:\e^{\ri \sqrt { \pi } \left [ \tau { \varphi_{i , r}}(vt - x ) + \sigma { \varphi_{i , l}}(vt+x)\right ] } : \ , . \label{psi2}\end{aligned}\ ] ] here @xmath132 denotes the normal product in the rt algebra @xmath39 and @xmath133 are some klein factors , controlling the statistics of @xmath134 . in this respect we impose the general anyon exchange relation @xmath135 where @xmath136 is the sign function and @xmath137 is the so called _ statistical parameter _ which interpolates between bosons ( @xmath138 - even integer ) and fermions ( @xmath138 - odd integer ) . a simple realization of the klein factors is @xmath139 where @xmath140 generate the auxiliary algebra @xmath141 = [ \gamma^*_i\ , , \ , \gamma^*_j ] = 0\ , , \qquad [ \gamma_i\ , , \ , \gamma^*_j ] = \ri \frac{\k}{2 } \epsilon_{ij } \ , , \label{kl2}\ ] ] with @xmath142 for @xmath143 , @xmath144 and @xmath145 for @xmath146 . in order to fix the solution ( [ psi1],[psi2 ] ) completely , one should determine the parameters @xmath147 , @xmath148 and @xmath149 in terms of coupling constants @xmath150 and the statistical parameter @xmath138 . using a standard short distance expansion and ( [ psi1],[psi2 ] ) one gets the charge and current densities and @xmath151 . ] @xmath152 , \label{rhopm}\ ] ] @xmath153\ , , \label{jpm}\ ] ] where for convenience the variables @xmath154 have been introduced . the normalization of ( [ rhopm ] ) is fixed @xcite by the ward identities associated with the electric charge @xmath155 and the helicity @xmath156 defined by @xmath157 the normalization of ( [ jpm ] ) in turn is determined by the conservation law @xmath158 plugging ( [ psi1],[psi2],[rhopm ] ) in the quantum equations of motion @xmath159 \psi_\alpha ( t , x , i)= \nonumber \qquad \qquad \qquad \qquad \\ 2 [ g_+:\rho_+(t , x , i)\psi_\alpha : ( t , x , i ) -(-1)^\alpha g_- : \rho_-(t , x , i)\psi_\alpha : ( t , x , i)]\ , , \label{tleqm}\end{aligned}\ ] ] one finds @xmath160 moreover , the exchange relation ( [ an1 ] ) implies @xmath161 eqs . ( [ sys1],[sys2],[sys3 ] ) provide a system for determining @xmath149 and @xmath162 ( or equivalently @xmath147 and @xmath148 ) in terms of @xmath138 and @xmath150 . the solution is @xmath163 where the positive roots are taken in the right hand side . the relations ( [ z ] ) and ( [ v ] ) represent the anyon generalization @xcite of the well known result for canonical fermions @xmath164 , where an alternative and frequently used notation @xcite is @xmath165 considering the general anyon solution ( [ z ] , [ v ] ) , we assume in what follows that the parameters @xmath166 belong to the domain @xmath167 which ensures that @xmath147 , @xmath148 and @xmath149 are real and finite . let us discuss finally the current splitting boundary condition ( [ bc3 ] ) and establish the relation between @xmath128 and @xmath0 . expressing the chiral currents @xmath168 in terms of the chiral fields @xmath27 , one finds @xmath169 which , according to ( [ bc1 ] ) satisfy the current splitting boundary condition ( [ bc3 ] ) , provided that @xmath170 the symmetry content of the tl junction is strongly influenced by ( [ bc4 ] ) . the point is that in the presence of a defect the continuity equation ( [ vcons ] ) alone is not enough to ensure the electric charge conservation . a direct computation shows indeed that @xmath171 the independence of @xmath172 implies that the electric charge @xmath155 is conserved if and only if @xmath173 which , as expected , is equivalent to the kirchhoff rule @xmath174 since @xmath175 , one infers from ( [ k3 ] ) that @xmath37 satisfies ( [ k3 ] ) as well . therefore , the electric charge @xmath155 is conserved for those @xmath0 , whose entries along each column ( line ) sum up to 1 . in geometric terms , these scattering matrices belong to the stability subgroup @xmath176 of the @xmath7-vector @xmath177 . an explicit parametrization of @xmath178 in terms of angular variables is given in @xcite . summarizing , the condition @xmath179 guaranties the energy conservation in the tl - junction . concerning the electric charge @xmath155 , one must distinguish two different regimes . @xmath155 is conserved for @xmath180 . if instead @xmath0 belongs to the complement @xmath181 , there is an external incoming or outgoing charge flow in the junction and @xmath155 is not conserved . the possibility to describe such imperfect junctions is a remarkable feature of the current splitting boundary condition ( [ bc3 ] ) . the physical details about the charge transport in the junction are discussed in section 4.2 below . the crucial property of the operator solution ( [ psi1],[psi2],[z],[v ] ) is that it is universal , meaning that it applies for any representation of the chiral field algebra generated by @xmath27 . the fock and gibbs representations have been largely studied and describe the equilibrium properties of the tl model on @xmath12 . in order to explore the behavior of the luttinger liquid away from equilibrium , we investigate below the operator solution in the ness representation of the rt algebra @xmath39 , constructed in section 2.2 . the first step in this direction is the introduction of the _ fermion _ chemical potentials @xmath182 where @xmath183 defines the fermi energy for @xmath164 and @xmath184 is the external voltage applied to the thermal reservoir in the edge @xmath13 of fig . [ fig2 ] . in what follows we keep @xmath183 fixed and vary eventually the gate voltages @xmath184 . as already mentioned , the _ boson _ chemical potential @xmath62 has been introduced for avoiding some infrared singularities at the boson level and has nothing to do with @xmath9 . in fact , in the chiral correlators ( [ ll]-[rr ] ) we already performed the limit @xmath65 . in order to recover @xmath9 , following @xcite we introduce the shift @xmath185 , defined by @xmath186 and , consistently with the boundary condition ( [ bc1 ] ) , @xmath187 the transformations ( [ fcp1],[fcp2 ] ) extend to an automorphism @xmath188 on the whole algebra generated by the chiral fields @xmath27 , which is directly implemented in the operator solution ( [ psi1],[psi2],[rhopm],[jpm ] ) . at this stage the tl correlation functions in the ness are defined by @xmath189 \cdots \o_k[{\varphi}_{i_k , z } ] \rangle_{\beta,\mu } = \langle \o_1[\alpha_\mu { \varphi}_{i_1,z } ] \cdots \o_k[\alpha_\mu { \varphi}_{i_k , z}]\rangle_{\beta}\ , . \label{gcf}\ ] ] in the rest of the paper we focus on the correlation functions ( [ gcf ] ) , which capture the physical properties of the luttinger liquid with the current splitting boundary condition ( [ bc3 ] ) away from equilibrium . we will show in particular that ( [ gcf ] ) satisfy the kubo - martin - schwinger ( kms ) condition @xcite at equilibrium , which justifies the introduction of the chemical potentials @xmath9 by means of ( [ fcp2],[fcp3 ] ) . we derive here the two - point correlators of @xmath190 defined by ( [ psi1 ] , [ psi2 ] ) in the ness and discuss their properties . for this purpose we extend away from equilibrium the finite temperature results of @xcite . using ( [ gcf2 ] ) , for @xmath191 one finds @xmath192}\right \}^{\sigma \tau \s_{ij}^t } \left \{\frac{1}{\frac{\beta_j}{\pi } \sinh \left [ \frac{\pi}{\beta_j}(vt_{12 } - { \widetilde{x}}_{12 } ) - \ri \varepsilon \right ] } \right \}^{\sigma \tau \s_{ij } } \quad \ ; \ ; \nonumber \\ \left \{\frac{1}{\frac{\beta_i}{\pi } \sinh \left [ \frac{\pi}{\beta_i}(vt_{12}+ x_{12 } ) - \ri \varepsilon \right ] } \right \}^{\tau^2 \delta_{ij } } \prod_{k=1}^n \left \{\frac{1}{\frac{\beta_k}{\pi } \sinh \left [ \frac{\pi}{\beta_k}(vt_{12 } - x_{12 } ) - \ri \varepsilon \right ] } \right \}^{\sigma^2 \s_{ik}\s_{kj}^t } \ , , \label{anc1}\end{aligned}\ ] ] where @xmath193 @xmath194 + \sigma[(vt_1-x_1)\sum_{k=1}^n\s_{ik}\mu_k -(vt_2-x_2)\sum_{k=1}^n\s_{jk}\mu_k]\}/(\sigma+\tau ) } \ , . \label{anc4}\ ] ] the @xmath195-correlator has the analogous form , @xmath196 the tl junction involves two types of @xmath191-@xmath195 interactions . first , the lagrangian ( [ lagr ] ) contains a @xmath191-@xmath195 _ bulk coupling _ proportional to @xmath197 . second , the current splitting boundary condition ( [ bc3 ] ) provides an additional _ boundary interaction _ described by the mixed left - right correlators ( [ lr ] , [ rl ] ) ) . consequently , the mixed @xmath191-@xmath195 correlators are non - trivial and have the form @xmath198}\right \}^{\tau^2 \s_{ij}^t } \left \{\frac{1}{\frac{\beta_j}{\pi } \sinh \left [ \frac{\pi}{\beta_j}(vt_{12 } - { \widetilde{x}}_{12 } ) - \ri \varepsilon \right ] } \right \}^{\sigma^2 \s_{ij } } \quad \ ; \ ; \nonumber \\ \left \{\frac{1}{\frac{\beta_i}{\pi } \sinh \left [ \frac{\pi}{\beta_i}(vt_{12}+ x_{12 } ) - \ri \varepsilon \right ] } \right \}^{\sigma \tau \delta_{ij } } \prod_{k=1}^n \left \{\frac{1}{\frac{\beta_k}{\pi } \sinh \left [ \frac{\pi}{\beta_k}(vt_{12 } - x_{12 } ) - \ri \varepsilon \right ] } \right \}^{\sigma \tau \s_{ik}\s_{kj}^t } \ , , \label{manc1}\end{aligned}\ ] ] where @xmath199 @xmath200 + \sigma[(vt_1-x_1)\sum_{k=1}^n\s_{ik}\mu_k-(vt_2+x_2)\mu_j ] \}/(\sigma+\tau ) } \ , . \label{manc4}\ ] ] finally , @xmath201 as expected , in the equilibrium limit @xmath202 and @xmath203 for all @xmath204 , the correlators ( [ anc1])([manc22 ] ) simplify and satisfy the kms condition , which represents a non - trivial check both on the computation and on the shift ( [ fcp2 ] , [ fcp3 ] ) introducing the chemical potentials . let us consider for instance ( [ anc1 ] ) , which in this limit takes the form @xmath205\mu /(\sigma+\tau)}\times \qquad \qquad \nonumber \\ \left \{\frac{1}{\frac{\beta } { \pi } \sinh \left [ \frac{\pi}{\beta } ( vt_{12}+ { \widetilde{x}}_{12 } ) - \ri \varepsilon \right ] } \right \}^{\sigma \tau \s_{ij}^t } \left \{\frac{1}{\frac{\beta } { \pi } \sinh \left [ \frac{\pi}{\beta } ( vt_{12 } - { \widetilde{x}}_{12 } ) - \ri \varepsilon \right ] } \right \}^{\sigma \tau \s_{ij } } \quad \ ; \ ; \nonumber \\ \left \{\frac{1}{\frac{\beta } { \pi } \sinh \left [ \frac{\pi}{\beta } ( vt_{12}+ x_{12 } ) - \ri \varepsilon \right ] } \right \}^{\tau^2 \delta_{ij } } \left \{\frac{1}{\frac{\beta } { \pi } \sinh \left [ \frac{\pi}{\beta } ( vt_{12 } - x_{12 } ) - \ri \varepsilon \right ] } \right \}^{\sigma^2 \delta_{ij } } \ , . \quad \label{anc5}\end{aligned}\ ] ] recalling that the kms automorphism @xmath206 acts on @xmath207 as follows , @xmath208 ( t , x , i ) = \e^{\ri s \mu}\ , \psi_\alpha ( t + s / v , x , i)\ , , \label{anc6}\ ] ] one can check that the equilibrium correlator ( [ anc5 ] ) satisfies the kms condition @xmath209(t_2,x_2,j)\rangle_{\beta , \mu } = \langle \left [ \varrho_{s}\psi_1\right ] ( t_2,x_2,j)\psi_1^*(t_1,x_1,i)\rangle_{\beta , \mu } \label{anc7}\ ] ] for all values of the statistical parameter @xmath138 . the critical scaling dimensions @xmath210 can be extracted from ( [ anc1])([manc22 ] ) in the limit @xmath211 and @xmath212 . because of the operator mixing , this is a subtle issue , which has been discussed in full detail in @xcite . one gets , @xmath213 where @xmath214 are the eigenvalues of @xmath0 . as already observed in @xcite , the impact of the vertex interaction is captured by the term @xmath215 , which preserves unitarity in the sense of conformal field theory because @xmath216 . a remarkable special case is obtained by setting @xmath217 . in this case the bulk @xmath191-@xmath195 coupling vanishes and one is left only with the boundary interaction induced by the current splitting boundary condition ( [ bc3 ] ) . from ( [ v ] ) one gets @xmath218 and ( using ( [ sys1 ] ) with @xmath137 and @xmath219 ) @xmath220 inserting ( [ nst ] ) in ( [ anc1])([anc22 ] ) and localizing the fields in the same edge ( i.e. setting @xmath221 ) , one finds that the correlation functions simplify to @xmath222}\right \}^{\k}\ , , \label{panc1}\end{aligned}\ ] ] @xmath223}\right \}^{\k \s_{ik}^2}. \label{panc2}\end{aligned}\ ] ] the condition @xmath224 and the left - right asymmetry of the ness construction in section 2 imply that only left moving ( incoming ) excitations contribute to @xmath225 , which therefore coincides with the equilibrium correlator @xcite . all the non - equilibrium features are captured by @xmath226 , which involves only right moving ( outgoing ) excitations . in fact , in spite of being localized in the edge @xmath13 of the graph , ( [ panc2 ] ) depends on the temperatures and chemical potentials of the all @xmath7 edges . it is instructive for this reason to derive and compare the fourier transforms of ( [ panc1 ] , [ panc2 ] ) . we will show first that they can be expressed in terms of the finite temperature tl _ anyon distribution _ discovered in @xcite . consider in fact @xmath227 plugging ( [ panc1 ] ) in ( [ ad1 ] ) one gets @xmath228 where the @xmath229-function fixes the dispersion relation and @xmath230 is the equilibrium anyon momentum distribution @xcite @xmath231 @xmath232 and @xmath12 being the beta and gamma functions ( euler s integrals of first and second kind respectively ) . notice that for @xmath233 the distribution ( [ ad3 ] ) depends on both @xmath234 and @xmath235 and not only on the dimensionless combination @xmath236 . ( [ ad3 ] ) defines a smooth function of @xmath237 , which satisfies @xmath238 and has the following asymptotic behavior : @xmath239 for positive integer @xmath138 ( i.e. for fermions and bosons ) the distribution ( [ ad3 ] ) simplifies to @xmath240 ! } \prod_{j=1}^{n-1 } \bigl |j - \frac{1}{2 } - \frac{\ri}{2\pi } p\beta \bigr |^2\ , , & \k = 2n-1 \ , , \\ \\ \frac{1}{(\e^{\beta p}-1 ) } \frac{2\pi \beta^{-2n } } { ( 2n-1 ) ! p } \prod_{j=0}^{n-1 } \bigl |j- \frac{\ri}{2\pi } p\beta \bigr |^2\ , , & \k = 2n\ , , \end{cases } \label{intad}\ ] ] where @xmath241 and the familiar fermi and bose distributions appear as prefactors . the first two fermion and boson distributions are @xmath242 as expected , in the fermion point @xmath164 of the tl liquid one gets the familiar fermi distribution . in spite of the fact that the remaining boson and fermion points ( @xmath243 ) have been established in @xcite more then a decade ago , to our knowledge their physical meaning and potential applications of ( [ intad ] ) have not been fully explored . ( 500,110)(-120,20 ) at fixed temperature @xmath244 for @xmath245 ( dotted blue line ) and @xmath246 ( dashed black line ) compared to the fermi distribution @xmath164 ( continuous red curve).,title="fig : " ] ( 500,110)(-120,20 ) at fixed @xmath247 for different temperatures @xmath248 ( continuos red line ) , @xmath249 ( dashed black line ) and @xmath250 ( dotted blue line).,title="fig : " ] in order to give an idea about the anyon distributions in the interval @xmath251 , we show some of them in fig.[fig3 ] , where the standard fermi distribution ( continuous red curve ) is given for comparison . fig.[fig4 ] displays the behavior of the anyon distribution ( [ ad3 ] ) for fixed @xmath247 and different temperatures . for @xmath252 and with decreasing of the temperature @xmath253 one observes the formation of a sharp peak at @xmath254 ( in agreement with eq . ( [ as0 ] ) ) , which signals a condensation - like phenomenon @xcite . concerning the fourier transform of ( [ panc2 ] ) , it is useful to consider first the case when all the temperatures are equal ( @xmath202 ) , the system being driven away from equilibrium only by the voltages @xmath184 . in this case @xmath255}\right \}^{\k}. \label{panc22}\end{aligned}\ ] ] and therefore @xmath256 one has still the equilibrium distribution , with the energy shifted by a linear combination of the chemical potentials @xmath257 , whose coefficients are the @xmath0-matrix elements . finally , in the coordinate space the general expression ( [ panc2 ] ) is a product of @xmath225-factors with different temperatures and chemical potentials . one gets therefore in momentum space the nested convolution formula @xmath258 being a convolution of distributions , ( [ ad223 ] ) is also a well defined distribution . the ness @xmath259 has therefore a remarkable property : the associated _ non - equilibrium distribution _ is simply a _ convolution of equilibrium distributions _ with different temperatures and chemical potentials . since the general form of ( [ ad223 ] ) is quite complicated , it is instructive to consider below the case @xmath87 and @xmath260 , focusing on @xmath261 using the @xmath83-space representation , at equal temperatures one finds the relation @xmath262 for @xmath263 the convolution @xmath264 defines a new distribution . since we were not able to determine its explicit analytic form , we give some plots which are obtained numerically . the plots in figs . [ fig5 ] and [ fig6 ] illustrate the behavior of @xmath264 for different values of @xmath265 , @xmath138 and @xmath36 . we see that even for @xmath263 the distribution @xmath264 is similar to @xmath230 with a kind of effective " temperature and statistical parameter depending on @xmath265 , @xmath138 and @xmath36 . ( 500,110)(-120,20 ) at fixed @xmath245 and @xmath266 for different temperatures @xmath267 ( continuos red line ) , @xmath268 ( dashed black line ) and @xmath269 ( dotted blue line).,title="fig : " ] ( 500,110)(-120,20 ) at fixed @xmath270 , @xmath271 and @xmath272 for statistical parameters @xmath273 ( continuos red line ) , @xmath274 ( dashed black line ) and @xmath275 ( dotted blue line).,title="fig : " ] summarizing , we derived above the two - point tl anyon correlation functions away from equilibrium . the results ( [ anc1],[manc1 ] ) are expressed as products of @xmath0-dependent powers of equilibrium correlators at different temperatures . in agreement with this fact the momentum space anyon ness distribution is the convolution of equilibrium anyon distributions ( [ ad223 ] ) . the equilibrium limit satisfies the kms conditions . at criticality one is dealing with a @xmath92 conformal field theory , whose anomalous dimensions ( [ dimensions0 ] ) depend not only on the coupling constants @xmath150 , but also on the scattering matrix @xmath0 . the above technique allows to compute higher anyon correlation functions as well , but in order to investigate the transport properties of the system , we concentrate below on the electric and energy current correlators away from equilibrium . the charge transport in the ness is described by @xmath276 which follow by substituting ( [ jpm],[sym1 ] ) in ( [ gcf ] ) . ( [ ct00 ] ) describes the external charge flow in the junction : it is constant in time and is incoming for @xmath277 and outgoing for @xmath278 . ( [ ct01 ] ) determines instead the value of the currents along the leads . the charge balance @xmath279 is satisfied and represents an useful check . if @xmath180 , the electric charge is conserved @xmath280 and the @xmath183-dependence in ( [ ct01 ] ) drops out , leading to @xmath281 the current ( [ ct1 ] ) satisfies the kirchhoff rule ( [ k2 ] ) and vanishes at equilibrium ( @xmath282 for all @xmath204 ) as it should be . the dependence on the statistical parameter @xmath138 is explicit and deserves a comment . in the physical domain @xmath283 , defined by ( [ physcond ] ) , the overall coefficient in front of the sum in ( [ ct1 ] ) is positive , @xmath284 for @xmath285 the coefficient @xmath286 decreases monotonically with @xmath287 . for @xmath288 one has that @xmath289 in the physical domain @xmath283 . in this case the coefficient @xmath286 increases in the interval @xmath290 , reaching the maximal value @xmath291 and decreases for @xmath292 . this behavior is illustrated in fig . [ fig7 ] . ( 500,110)(-120,20 ) at @xmath293 ( dashed black line ) and @xmath294 ( continuous red line).,title="fig : " ] the current ( [ ct1 ] ) is proportional to the applied external voltages . non - linear effects are absent in the critical regime under consideration , which implies the conductance tensor @xmath295 we see that the ness approach , adopted in this paper , confirms the result for @xmath296 , obtained previously for @xmath164 by different methods , including renormalization group techniques @xcite , linear response theory @xcite and conformal field theory @xcite . the novelty in ( [ ct2 ] ) is the explicit dependence on the statistical parameter @xmath138 , shown in fig . [ fig7 ] . a similar computation gives the energy ( heat ) flow @xmath297 + \frac{\pi v^2}{12}\sum_{j=1}^n \left ( \delta_{ij } - \s_{ij}^2 \right ) \frac{1}{\beta_j^2}\ , , \label{ct04}\ ] ] which satisfies the kirchhoff rule ( [ k1 ] ) for all @xmath175 . for @xmath180 the expression ( [ ct04 ] ) takes the form @xmath298 + \frac{\pi v^2}{12}\sum_{j=1}^n \left ( \delta_{ij } - \s_{ij}^2 \right ) \frac{1}{\beta_j^2}\ , . \label{ct4}\end{aligned}\ ] ] the heat flow depends therefore not only on @xmath183 and the voltages @xmath184 , but also on the temperatures @xmath8 . as before , we consider for illustration the case of @xmath87 wires . inserting the scattering matrices ( [ ex1 ] ) in ( [ ct01 ] ) one has @xmath299\ , , & \ ; \ ; { \rm det}\ , \s = 1\ , , \\ gv_f[(\mu_1 + \mu_2)(\sin \theta -1)+(\mu_1-\mu_2)\cos \theta ] \ , , & \ ; \ ; { \rm det}\ , \s = -1\ , . \\ \end{cases } \label{ex2}\ ] ] therefore , @xmath300 corresponding respectively to full reflection ( disconnected edges ) and complete transmission in the junction . finally , @xmath301 + \frac{\pi v^2\sin^2 \theta}{12}\left ( \frac{1}{\beta_1 ^ 2}-\frac{1}{\beta_2 ^ 2}\right ) \ , , \label{ex4}\end{aligned}\ ] ] for both families in ( [ ex1 ] ) . in this section we derive the _ noise power _ in the tl junction in fig . [ fig1 ] . for this purpose we need @xcite the two - point _ connected _ current - current correlator @xmath302 after some algebra one finds @xmath303 } \delta_{ij } + \sum_{l=1}^n \s_{il } \frac{1}{\beta_l^2 \sinh^2\left [ \frac{\pi}{\beta_l}(vt_{12}- x_{12 } ) - \ri \varepsilon \right ] } \s_{lj}^t \quad \nonumber \\ -\s_{ij } \frac{1}{\beta_j^2 \sinh^2\left [ \frac{\pi}{\beta_j}(vt_{12}-{\widetilde{x}}_{12 } ) -\ri \varepsilon \right ] } - \frac{1}{\beta_i^2 \sinh^2\left [ \frac{\pi}{\beta_i } ( vt_{12}+{\widetilde{x}}_{12 } ) -\ri \varepsilon \right ] } \s^t_{ij } \biggr \ } \ , . \quad \label{n2}\end{aligned}\ ] ] one easily verifies that the equilibrium limit ( @xmath304 for all @xmath204 ) of ( [ n2 ] ) satisfies the kms condition @xmath305(t_2,x_2,j ) \rangle_{\beta , \mu}^{\rm conn } = \langle [ \varrho_{s } j_+](t_2,x_2,j ) j_+(t_1,x_1,i ) \rangle_{\beta , \mu}^{\rm conn}\ , , \label{currkms1}\ ] ] where the kms automorphism @xmath306 acts on @xmath307 as follows , @xmath308(t , x , i ) = j_+(t+s / v , x , i ) \ , . \label{currkms2}\ ] ] the explicit expression ( [ n2 ] ) contains fundamental physical information about the ness . first of all , since @xmath309 the ness breaks down _ time reversal _ invariance , even if the junction interaction preserves it , i.e. if @xmath310 @xcite . nevertheless , _ time translation _ invariance is preserved , which allows one to use the conventional definition @xcite of _ noise power _ @xmath311 eq . ( [ n3 ] ) defines a complex matrix whose entries can be expressed @xcite in terms of the hypergeometric function @xmath312 , namely @xmath313 \beta_i^{-1 } \delta_{ij } + \qquad \qquad \nonumber \\ \sum_{l=1}^n\s_{il}\beta_l^{-1}\bigl [ f_-(\omega , \beta_l , -x_{12 } ) - f_+(\omega , \beta_l , -x_{12})\bigr ] \s^t_{lj } - \qquad \qquad \qquad \qquad \nonumber \\ \s_{ij}\beta_j^{-1}\bigl [ f_-(\omega , \beta_j,-{\widetilde{x}}_{12 } ) - f_+(\omega , \beta_j,-{\widetilde{x}}_{12})\bigr ] - \bigl [ f_-(\omega , \beta_i , { \widetilde{x}}_{12 } ) - f_+(\omega , \beta_i,{\widetilde{x}}_{12})\bigr ] \beta_i^{-1 } \s^t_{ij } \bigl \}\ , , \nonumber \\ \label{a1}\end{aligned}\ ] ] with @xmath314 from ( [ a1 ] , [ a2 ] ) one can deduce the zero - frequency limit ( _ zero - frequency noise power _ ) @xmath315 using @xmath316 = \frac{1}{2\pi v}\ , , \label{n5}\ ] ] one gets @xmath317 where @xmath286 , defined by ( [ ct3 ] ) , captures the dependence ( see fig.[fig7 ] ) of the noise on the statistical parameter @xmath138 . as expected , @xmath318 turns out to be a @xmath319-independent real symmetric matrix . if the electric charge is conserved ( @xmath320 ) , the noise power ( [ n6 ] ) satisfies in addition the kirchhoff rule @xmath321 the expression ( [ n6 ] ) admits the typical johnson - nyquist @xmath322 behavior and shows the non - trivial interplay between the different temperatures and the scattering matrix . for example , in the two - terminal case with @xmath323 one finds @xmath324 where @xmath325 is the absolute temperature and @xmath326 is the boltzmann constant . the eigenvalues of @xmath327 @xmath328 are nonnegative in agreement with the positivity of the two - point function ( [ n1 ] ) . analogous result holds for @xmath329 corresponding to @xmath330 . in this paper we constructed and investigated an exactly solvable model of a non - equilibrium luttinger junction . the basic points of our approach are : ( i ) : : a scale invariant point - like interaction , which is described by a scattering matrix @xmath0 and drives the system away from equilibrium ; ( ii ) : : a representation generated by a ness @xmath259 , which encodes the point - like interaction in the chiral fields @xmath27 ; ( iii ) : : an exact operator solution of the tl model ( in terms of @xmath27 ) on a star graph with the current splitting boundary condition in the vertex ; ( iv ) : : an extension of the conventional fermion luttinger liquid to anyon statistics . combining these ingredients , we derived the basic correlation functions in the state @xmath259 . the essential characteristic features of these functions are : ( a ) : : the non - equilibrium two - point anyon correlations are products of @xmath0-dependent powers of equilibrium correlations at the temperatures and chemical potentials of the heat baths , connected to the leads ; ( b ) : : accordingly , the corresponding momentum space distribution is the convolution of equilibrium anyon distributions at different temperatures and chemical potentials ; ( c ) : : the fourier transform of the leading terms in the large distance expansion of the anyon correlations gives cauchy - lorentz distributions , which after convolution reproduce themselves with appropriate width and median ; ( d ) : : in the critical limit one has a @xmath92 conformal field theory with @xmath0-dependent anomalous dimensions , which are explicitly derived ; ( e ) : : the expected breakdown of time reversal invariance is manifest in the current - current correlator . we investigated in detail the energy and charge transport in the junction for all values of the statistical parameter . the energy is conserved for @xmath175 , which covers both possibilities of a junction without and with electric charge dissipation . in the latter case we determined the exact expression for the charge flow leaving or entering the junction . the connected current - current correlation is a linear combination of hypergeometric functions . the associated zero - frequency noise power depends linearly on the temperatures . our investigation above has been focused essentially on the critical properties of anyon luttinger liquids away from equilibrium . it will be interesting to study the noncritical aspects as well . the generalization of the results of this paper beyond the luttinger liquid paradigm , when the nonlinearity of the dispersion relation becomes essential , is also a challenging open problem . we thank b. douot and i. safi for an inspiring discussion , which stimulated our interest in non - equilibrium luttinger liquids . would like also to thank the laboratoire de physique thorique dannecy - le - vieux for the kind hospitality during the preparation of the manuscript . the behavior of the correlators ( [ panc1],[panc2 ] ) at large large space separation @xmath331 is encoded in @xmath332 where @xmath333 , the dots stand for sub - leading contributions and @xmath334 the fourier transform @xmath335 is the familiar cauchy - lorentz distribution , where @xmath336 is the half width at half maximum and @xmath337 is the statistical median . using that the class of cauchy - lorentz distributions is closed under convolution , one finds @xmath338 summarizing , the fourier transform of the leading term in the long distance expansion of both ( [ panc1 ] ) and ( [ panc2 ] ) is a cauchy - lorentz distribution . notice that the width and the median of ( [ ld6 ] ) depend on the temperatures and chemical potentials of all heat baths , as well as on @xmath0 . a. liguori , m. mintchev and l. pilo , nucl . b * 569 * ( 2000 ) 577 [ hep - th/9906205 ] . n. ilieva and w. e. thirring , phys . b * 504 * ( 2001 ) 201 , [ hep - th/0010030 ] . p. calabrese and m. mintchev , phys . b * 75 * ( 2007 ) 233104 [ cond - mat/0703117 [ cond - mat ] ] . o. i. patu , v. e. korepin and d. v. averin , europhys . * 86 * ( 2009 ) 40001 . b. bellazzini , m. mintchev and p. sorba , j. phys . a * 40 * ( 2007 ) 2485 , [ arxiv : hep - th/0611090 ] . b. bellazzini , m. burrello , m. mintchev and p. sorba , proc . pure math . * 77 * ( 2008 ) 639 , arxiv:0801.2852 [ hep - th ] . m. mintchev , j. phys . a a * 44 * ( 2011 ) 415201 [ arxiv:1106.5871 [ math - ph ] ] . a. liguori , m. mintchev and l. zhao , commun . phys . * 194 * , 569 ( 1998 ) [ arviv : hep - th/9607085 ] . m. mintchev , e. ragoucy and p. sorba , phys . b * 547 * , 313 ( 2002 ) [ arxiv : hep - th/0209052 ] . m. mintchev , e. ragoucy and p. sorba , j. phys . a * 36 * , 10407 ( 2003 ) [ arxiv : hep - th/0303187 ] . b. bellazzini , p. calabrese and m. mintchev , phys . b * 79 * 085122 ( 2009 ) , [ arxiv:0808.2719 ] . b. bellazzini , m. mintchev and p. sorba , phys . rev . b * 80 * , 25441 ( 2009 ) , arxiv:0907.4221[hep - th ] . s. lal , s. rao , and d. sen , phys . b * 66 * ( 2002 ) 165327 .

brown dwarfs form mineral clouds in their atmospheres , where charged particles can produce large - scale discharges in form of lightning resulting in substantial sudden increase of local ionisation . brown dwarfs are observed to emit cyclotron radio emission . we show that signatures of strong transient atmospheric ionisation events ( flash ionisation ) can be imprinted on a pre - existing radiation . detection of such flash ionisation events will open investigations into the ionisation state and atmospheric dynamics . such events can also result from explosion shock waves , material outbursts or ( volcanic ) eruptions . we present an analytical model that describes the modulation of a pre - existing electromagnetic radiation by a time - dependent ( flash ) conductivity that is characteristic for flash ionisation events like lightning . our conductivity model reproduces the conductivity function derived from observations of terrestrial gamma ray flashes , and is applicable to astrophysical objects with strong temporal variations in the local ionization , as in planetary atmospheres and protoplanetary disks . we show that the field responds with a characteristic flash - shaped pulse to a conductivity flash of intermediate intensity . more powerful ionisation events result in smaller variations of the initial radiation , or in its damping . we show that the characteristic damping of the response field for high - power initial radiation carries information about the ionisation flash magnitude and duration . the duration of the pulse amplification or the damping is consistently shorter for larger conductivity variations and can be used to evaluate the intensity of the flash ionisation . our work suggests that cyclotron emission could be probe signals for electrification processes inside bd atmosphere . [ firstpage ] physical data and processes : instabilities radiation mechanisms : non - thermal atmospheric effects methods : analytical stars : atmospheres stars : brown dwarfs

1601.07474

the search for lightning in the solar system has benefited from an increasing variety of space missions ( e.g. venus express , cassini ) . lightning is the most observed flash transient process , that is a powerful but brief , short term processes leading to a local burst of ionisation . characteristic optical and radio emission provide growing evidence for lightning in solar - system planets @xcite . powerful flash transient processes can act destructively , or they can trigger the formation of new ( e.g. pre - biotic ) molecules @xcite . on earth , lightning is one of the most significant sources of natural ozone production e.g. @xcite . associated with terrestrial lightning is powerful gamma ray emission ( terrestrial gamma ray flashes ( tgfs ) ; e.g. , @xcite ) . we are particularly interested in lighting as an example for flash transient processes in atmospheres of brown dwarfs and giant gas planets with respect to observations of cyclotron maser emission that is observed in increasing number of brown dwarfs @xcite . brown dwarfs have clouds and weather - like variation that determines their dynamic atmospheres @xcite . recent spitzer observations suggest the detection of global winds @xcite . @xcite suggest that in such clouds small - scale spark discharge can be expected at cloud bottom while large - scale discharge should be more common near the top of the cloud or above it . hence , flash ionisation processes can be expected to occur rather frequently in the cloud layers of these ultra - cool objects . @xcite show that such flash ionisation events may affect a larger volume of the atmosphere in brown dwarfs than in the atmosphere of earth . the observation of cyclotron emission from brown dwarfs confirms that strong magnetic fields are present in these ultra - cool objects @xcite . the observation of a strong and coherent cyclotron emission further confirms an existence of a source of ionisation to produce the electron beams responsible for the emission . for hotter stars than m - dwarfs and brown dwarfs , coronal mass ejections are the sources of accelerated electrons like for example on the radio emitting flare stars like uv ceti or the magnetic chemically peculiar , extremely fast rotating radio pulsar cu virginis @xcite . the solar wind provides the majority of electrons for planetary cyclotron radio emission @xcite for most of the solar system planets . most brown dwarfs , however , occur as single stars or in brown dwarf - brown dwarf / ultra - low mass binary systems . non of the above sources of free electrons is , hence , available to brown dwarfs , unless they are part of a binary system where an active m - dwarf producing coronal mass ejections or a white dwarf provides high - energy irradiation ( @xcite ) . cyclotron emission from brown dwarfs is powerful and coherent , and the observations of this emission are consistent and reproducible . possible sources of free electrons leading to strong radio emissions in ultra - cool objects like brown dwarfs and free - floating planets are under investigation and include dust - dust collisions @xcite , alfven ionisation @xcite and cosmic rays impact @xcite . electric discharges ( lightning , transient luminescent events , small - scale but frequent coronal discharges ) are additional sources for electrons . given that brown dwarfs are powerful radio emitters and that clouds and winds form in their atmospheres , we suggest using cyclotron emission as a probe signal ( carrier signal ) to search for its transformation by atmospheric processes . recently , @xcite observed that the lighting radio emission originating from relativistic electrons accelerated in the earth magnetic field imprint their signal onto the radio signal of cosmic ray induced atmospheric air showers . in @xcite , the carrier signal for the effect of lighting is the radio emission from a cosmic ray airshower , while a more powerful source is required for the carrier source in the astrophysical context . existing data from brown dwarfs may already carry the fingerprints of lightning in a form of specific time - variations . by studying how a coherent ( cyclotron ) emission is modulated by flash transient processes , we suggest a new detection method for transient events and for the atmospheres where they occur . we demonstrate how the electromagnetic field of the cyclotron radiation is transformed and how the signatures of the flash ionisation is imprinted onto the pre - existing cyclotron emission . _ flash ionisation _ describes the fast transient events leading to rapid ionization of surrounding medium lightning being one example @xcite . other discharges include those induced by cosmic rays @xcite , explosions , plasma jets or lightning in volcano plumes . section 2 contains our approach with details of electron cylotron emission and a reflection of emission symmetries . the concept of the time - dependent conductivity as a model for lightning is outlined in sect . 3 . section [ sec:2 ] presents our mathematical modelling for the electromagnetic field transformed by a flash ionisation process represented by a time - dependent conductivity . an exact analytical solution to the equations for a general conductivity function is given in sect . [ ssec:2.2 ] . the temporal variation of the particular flash - like conductivity is modelled in sect . [ ssec:2.3 ] . section [ ssec:3.4 ] presents a test case for our model by comparing the conductivity time - variations with tgf observations . the formal body derived in sect [ sec:2 ] is not restricted to radio waves only . section [ sec:3 ] presents a parameter study that shows how the electromagnetic field responds to different conductivity time - flashes . we demonstrate how the amplitude and the damping character changed depending on the strength of the intersected flash ionisation event . section [ sec:4 ] discusses the observational effects of flash ionisation events on radio emission from brown dwarfs . we summarise this section in a recipe for observations. this section outlines the approach that we take to derive the signatures of flash ionisation events like lighting imprinted onto a pre - existing cyclotron emission . we first summarise necessary facts about electron cyclotron maser emission ( sect . [ ss : ce ] ) and outline our approach to cyclotron emission probing lightning - active regions in sect [ ss : how ] . electron cyclotron emission is a localised - source emission , which is known to emerge out of a star / planet s atmosphere shaped as a hollow cone centred around a magnetic pole ( e.g. , @xcite ) . each cone is associated with a distinct source region as indicated in fig [ f:11 ] . such radiation cones form due to a high directivity property of the local cyclotron emission ( e.g. , @xcite ) , with its annular symmetry reflecting the magnetic field symmetry around the pole . rotation of the object causes the periodicity of the radiation peaks detected when the cone s walls cross the receiver s line - of - sight . the width of the observed radiation peaks ( e.g. fig . 1 in @xcite ) is determined by the extension of the source and by the path length along which the radiation propagated and along which it experiences scattering and dispersion effects . it is therefore proportional to the cone - wall thickness , @xmath0 , as depicted in figs . [ f:11 ] and [ f:1 ] . cyclotron maser emission occurs when the velocity distribution function of the electron beam , that travels into increasing magnetic field , is continually transformed by magnetic compression to form a horseshoe - shape distribution in velocity space@xcite . when eventually the horseshoe is confined enough for the major electron population to satisfy the cyclotron resonance condition , @xmath1 ( is a local cyclotron frequency , @xmath2 is a relativistic factor , @xmath3 , @xmath4 is the speed of light , and @xmath5 and @xmath6 are wave number and particle velocity parallel to the beam direction , respectively ] ) , the beam emits the cyclotron maser radiation at the location where this resonance happened . various factors influence the atmospheric altitude at which this resonance occurs , including the initial electrons distribution in velocity space , the location where the electrons coupled to the magnetic field , and the magnetic field values , gradient and topology . it is known from earth and saturn direct in - source observations ( e.g. , @xcite to happen inside plasma cavities with rarefied plasma . this does not imply the average local density should be low , while higher density can influence the radiation escape . being emitted from a localised region with small spread along the altitude , the radiation at the source has a sharp frequency spectrum . in addition to the in - situ observations , this was confirmed by scaled laboratory experiment @xcite . the emission frequency therefore corresponds to the local magnetic field and depends on the altitude at which the emission happened . the cyclotron radiation is emitted nearly perpendicular to the magnetic field lines ( e.g. , @xcite ) , but it can be refracted by the atmospheric gas and clouds before it emerges from the atmosphere in the shape of a hollow cone ( fig.[f:11 ] ) . this can also be caused by the non - uniform magnetic field along the path and , consequently , non - uniform plasma frequency and non - uniform refractive index . if the source is far enough from the ionosphere and the radiation is not influenced by propagation effects ( case 1 in fig.[f:11 ] ) , the radiation will propagate in a very wide straight cone ( almost plane - like , hence @xmath7 very close to @xmath8 ) with very thin cone walls ( opening angles @xmath9 , hence cone wall thickness @xmath10 in fig . [ f:11 ] ) . radio pulsars are examples of the sources of such straight ( not refracted ) propagation . typical width of the radio emission cones for pulsars , @xmath0 , is around @xmath11@xmath12 ( e.g. , @xmath13 , @xmath14 and @xmath15 found in @xcite ) . a summary of the underlying physics and geometry consideration addressing the question why the cyclotron emission appears in hollow cones is given in @xcite . radio observations from brown dwarfs , however , often suggest smaller cone opening angles , @xmath2 , and thicker cone walls ( @xmath16 ) , @xmath0 , due to broadened ( by propagation ) radiation pulses in the observed phase curves . the pulse broadening would be caused by interaction of the emitted radiation with the atmospheric gas and clouds , hence , larger broadening advocating a larger path lengths ( case 3 or 4 in fig . [ f:11 ] ) in contrast to very narrow pulses of non - refracted radiation ( case 1 in fig . [ f:11 ] ) . higher frequency cyclotron radiation from the same astrophysical source is emitted deeper into the magnetic field / atmosphere ( closer to the magnetic pole where the magnetic field lines converge ; see cases 3 and 4 in fig . [ f:11 ] ) , and hence it would refract steeper than lower frequency one . in this case , it is likely to pass through more of the stratified atmospheric volume becoming more affected by the propagational effects , which results in thicker walls of the radiation cones . note that the effect of different refraction of different frequencies cones could only be noticeable in observations for a substantial separation of the frequencies to allow for a substantial separation of the local medium properties . e.g. , with the cyclotron frequency proportional to the local magnetic field ( @xmath17 ) , which decreases from the star / planet s surface as @xmath18 ( @xmath19 being a radial distance from the object s magnetic dipole centre ) , the square of the refractive index determined as @xmath20 ( where @xmath21 is the local plasma frequency and @xmath22 is the frequency of the radiation ) will also have roughly @xmath18 dependence . therefore , the difference between @xmath23 at different locations along the magnetic field lines would be proportional to @xmath24 , which means that the relative difference in refraction index , @xmath25 , is @xmath26 times smaller than the correspondent relative frequencies difference , @xmath27 . the effect of two well - separated ( in frequencies ) cones refracting differently on their way out of the star s magnetosphere is significant in observations from cu virginis . the effect of refraction was suggested as an explanation for multi - frequency observations to explain why cu virginis radio observations ( e.g. , ( kellett et al . ( 2007 ) ; trigilio et al . ( 2008 ) ) ) found a higher frequency cone ( i.e. , radiated deeper into the magnetic field and the atmosphere , at lower altitude ) inside a lower frequency one , radiated at higher latitude ( lo et al . the clearly seen refraction there resulted in the cones walls , @xmath0 , broadening up to @xmath28 and @xmath29 ( kellett et al . ( 2007 ) ) . the idea of cyclotron emission radiated in the shape of hollow cones was applied by @xcite to derive an opening angle of @xmath30 and a cone wall thickness @xmath31 for neptune s radio emission , and @xcite applies the concept to jovian radio emission . observations by @xcite for the brown dwarf tvlm 513 suggest an estimation of the opening angle from their fig.3 being @xmath32 , and the observed fwhh of the radiation peak in the same plot can provide an estimation of the radiation cone wall thickness as @xmath33 . observations by @xcite present the phase curve of the emission ( their fig . 4 ) from which the cone wall thickness can be estimated as @xmath34 ( beam angle in @xcite ) . this small @xmath0 may indicate the cyclotron source located rather high in the atmosphere , comparable to case 2 in fig [ f:11 ] . @xcite s observation for nltt 33370ab ( their figs . 3 , 4 ; rotation period p=3.7h ) could be interpreted as a more complicated scenario with a rather small radiation cone with an opening angle estimated from the phase curve as @xmath35 and a wall thickness of @xmath36 . such wide beams ( large @xmath0 ) ( comparable to the refracted cu - virginis beams while being broader than non - refracted pulsar ones ) with a relatively small opening angle ( @xmath2 ) may be represented by case 4 in fig [ f:11 ] . we note that the number of cones depends on the magnetic field geometry as demonstrated in @xcite . brown dwarf atmospheres and similarly cold atmospheres , like those of extrasolar giant gas planets , are made of chemically diverse oxygen - rich , h@xmath37-dominated gases . clouds form and have a distinct influence on the atmosphere through their opacity and through element depletion . brown dwarf atmospheres can also be expected to be dynamic as brown dwarfs are fast rotators . giant gas planets may be close to their host star such that irradiation drives strong winds inside their atmospheres . it is plausible to expect that such dynamic , cloud forming atmospheres generate lighting activity ( @xcite ) more intensive than in the solar system and affecting larger atmospheric volumes ( @xcite ) . ultra - cool atmospheres may show different kind of discharges in form of thunderstorms including inter - cloud and intra - cloud discharges . transient luminous events ( sprites , elves , jets , etc . ) are frequently observed on earth in conjunction with lighting . sprites and jets are substantially more extended than lighting , and therefore increase the atmospheric volume affected by flash ionisation events . these flash ionisation events perturb the upper atmosphere and magnetosphere ( e.g. @xcite ) . for a brown dwarf with a radius of @xmath38 km and a cyclotron peak s half - width of about @xmath39 @xcite , the spread of the beam ( @xmath0 in fig [ f:11 ] ) emerging from the atmosphere is @xmath40km@xmath41 km . if this radiation beam underwent a ten - fold spatial dispersion ( for example case 3 in fig . [ f:11 ] ) , the source region near the magnetic pole would have an extension ( i.e. , the cone s wall thickness ) about 244 km across . this horizontal ( with respect to the star / planet s surface ) extend of a storm between @xmath42 km and @xmath43 km determines the volume of the column of the atmosphere which could be probed by the cyclotron beam . recent observation from saturn s storm area indicate that the storm that emerges at the surface alone has a head diameter of @xmath44 km . @xcite demonstrate how the atmospheric dynamic for brown dwarfs produces similar pattern . these rough ideas of the radiation / storm longitudinal extensions suggest them to be reasonably comparable for the storms being able to affect a significant portion of the beam s radiation . we suggest that if the ecm radiation travels through the extended atmosphere of the brown dwarf / planet , it may be modulated by conductivity changes in the atmosphere produced by flash ionisation and can proposed to probe the extended atmosphere it travels through . in the following , we consider a local fragment of the radiation cone at a direction that will rotate into the observer s view . cyclotron emission is a frequency - resonant phenomenon and can therefore be represented by a harmonic wave . we study how this coherent wave changes after experiencing a flash ionisation event affecting a volume it paths through . we consider a pre - existing coherent radiation ( cyclotron radiation or similar ) travelling through an atmosphere environment with flash - ionisation events ( e.g. lighting , sprites , flares ) . its coherent character allows to model this incident electromagnetic field as a harmonic wave . we are interested in how the electromagnetic field of a pre - existing emission is transformed by the flash ionisation . the interaction between an electromagnetic wave and a medium rapidly changing in time is highly non - linear . current approaches that model radiation , emitted by flash ionization processes , are focused on microscopic modelling of charged particle production ( electrons and ions ; e.g. , @xcite ) , in some cases including electric ( @xcite ) or magnetic ( @xcite ) fields . as a result , microscopic approaches predict linear growth rates of the number density of charged particles , and hence , the linear growth stage of electromagnetic emission only . in order to determine the maximum of the flash ionisation event , its decline and duration , the non - linear saturation and the consequent drop in the particles velocities and production rate needs to be taken into account . this will allow us to qualify and quantify the electromagnetic field response . in the case of an external field probing the flash - ionisation area , non - linear interactions between field and the transient , ionised medium can be expected to be stronger compared to the discharge s own radiation field because of the higher amplitude of the initial ( cyclotron ) field . we therefore apply the concept of a parameterised time - dependent conductivity in a macroscopic model to allow us to account for non - linear interactions between the ionised medium and the field . we use maxwell s equations formalism to study the field transformation by the transient processes . we model the flash ionisation by a time - dependent conductivity with a flash - like shape . a flash - shape like conductivity function first rises describing the fast linear growth of the ionised particles at a breakdown ionisation stage , then reaching saturation and eventually declines decribing the ending of the ionisation process ( see fig.[f:4 ] ) . this time - dependent functional form is inspired by experimental and observational results ( e.g. @xcite ) . flash ionisation events like lightning , coronal discharges , energetic explosions or eruptions , can be suitably described by a time - dependent conductivity without expensive small - scale modelling , because suitable variation in amplitude , duration , shape and the flash front s gradient can be adopted . conductivity shapes derived from tgfs is a particular well studied case which we use as test case in sect . [ ssec:3.4 ] . in sect . [ ssec:2.1 ] , the changing electric field component from maxwell s equations for a time - dependent conductivity flash is derived , and the exact solution of the derived equations ( eqs . ) is presented in sect . [ ssec:2.2 ] . the time - dependent conductivity function is discussed in sect . [ ssec:2.3 ] and its relvance of natural discharges in sect . [ ssec:3.4 ] . the electromagnetic fields , @xmath45 and @xmath46 , that interact with a fast - changing medium influence the medium parameters ( i.e. dielectric permittivity or conductivity ) , and these induced changes feed back into the fields . in the model presented in this paper , the electromagnetic fields in a transient medium are described by the linear maxwell s equations . however , the material coefficients , dielectric permittivity , magnetic permeability or conductivity , as part of maxwell s equations in turn depend on the fields , and therefore the field equations are intrinsically non - linear . this section presents the derivation of the equations for the electric component of the electromagnetic field for a time - dependent conductivity . the interaction of the pre - existing coherent radiation field with the transient event starts at the time @xmath47 . at this moment , the medium s conductivity begins changing in time in a flash - like manner due to the ionisation process . we consider the problem in one spacial dimension , @xmath48 , and assume that the electromagnetic fields only have components which are perpendicular to that dimension , @xmath49 the fields should satisfy maxwell s equations ( in cgs units ) , @xmath50 where @xmath51 is a dielectric permittivity of the medium , which is dimensionless and is considered to be constant in time , @xmath4 is the velocity of light [ cm / s ] , @xmath52 is the time - dependent conductivity [ 1/s ] , and @xmath53 [ statv / cm ] and @xmath54 [ oe ] are the electric and magnetic components of the electromagnetic field , correspondingly . considering 1d electric and magnetic field components only , as in eq . , the following equation is derived from the system of two maxwell s equations for @xmath53 and @xmath54 components of the electromagnetic field ( eq . ) , by excluding @xmath54 and collecting the time - dependent conductivity term in the right - hand side : @xmath55 the left - hand side of eq . is a wave equation for homogeneous media . we assume that the transient processes start at @xmath47 . the left - hand side of eq . has a known green s function @xcite , which is a solution of a wave equation with the dirac s delta function @xmath56 right - hand side ( rhs ) denoting a point source of radiation . applying convolution of the green s function to its right - hand side @xcite the following integral equation for the field results @xcite : @xmath57 where @xmath58 , and @xmath59 is the initial electric component of the field , describing the pre - existing cyclotron radiation ( i.e. the field which propagated through the medium before the transient processes starts at @xmath60 ) . we consider the initial field @xmath59 as a plane wave , @xmath61 , and assume the transformed field s spatial dependence in the similar form , as @xmath62 equation can then be reduced to the following equation for the field s time - dependence @xmath63 , after integrating eq . over @xmath48 , @xmath64 equation can be transformed after double integration into a second - order differential equation for the field s time - dependence , @xmath63 , @xmath65 this is a linear differential equation , which accounts for nonlinear effects of the flash ionisation in maxwell s equations , eq . , by introducing the time - dependent conductivity , @xmath66 . substituting @xmath67 with a dimensionless @xmath68 by @xmath69 , eq . takes a dimensionless form with respect to a normalised electric field , @xmath70 @xmath71 where the function @xmath72 represents the conductivity time - dependence , @xmath73 the initial conditions for @xmath68=0 are @xmath74 note , that the differential equation eq . could be derived without using the green s function and integral equation for the field , but using the integral equation allows finding the right initial condition for the field s first derivative , eq . . we seek to solve eqs . to study how an electromagnetic field in an initial form of a harmonic wave can be transformed by transient events . we aim to derive a signature of transients ( like lightning ) in a pre - radiated field such as coherent cyclotron radiation . for that purpose , we firstly reduce the eq . to a first - order riccati differential equation , by substituting the field s time - dependence , @xmath75 @xmath76 the resulting riccati equation for the function @xmath77 is : @xmath78 assuming that @xmath79 is known according to eq . . there are no known solutions for such an equation in general . we tested various representations of the functions @xmath77 and @xmath79 , i.e. different representation of the electromagnetic field and the time - dependent conductivities , @xmath75 and @xmath80 . we found that if the function @xmath79 is expressed through a function @xmath81 , such that @xmath82 ( @xmath83 is an arbitrary constant ) , an exact particular solution to the riccati eq . can be written as @xmath84 the representation of the conductivity function @xmath79 in terms of another function @xmath81 in eq . is just a substitution of one function with another , in order to get mathematical advantages . it is valid for any function and do not involve any assumptions on the functions . the resulting riccati eq . and eq . are therefore exact equivalents of the wave equation eq . for the electric component of electromagnetic field . a particular solution for the electromagnetic field determined by eq is then @xmath85 to find a general solution for the electromagnetic field , the substitution @xmath86 in terms of a new introduced function @xmath87@xmath88 @xmath89 transforms the initial differential eq . for the field , with the conductivity - related function @xmath79 expressed as in eq . , into a first - order ordinary linear differential equation for the derivative of @xmath90 , @xmath91 which can be solved easily . the solution of eq . , @xmath92 gives the following general solution for the time - dependent electromagnetic field : @xmath93 the constants @xmath94 and @xmath95 are found from the initial conditions in eq . to be : @xmath96 the solution for the field ( eq . ) is an _ exact general solution _ describing how the radiation field is transformed by time - variations of conductivity as described in a general form by eq . . advantages of exact analytical solution over numerical solutions are much more profound for powerful fast processes ( i.e. , mathematically , for large - amplitude rapidly varying coefficients of differential equations at a time - scale comparable to the wave period ) , as conventional numerical techniques fail to produce correct reliable results in this case . we are interested in the case when the conductivity s time - dependence has a flash - like shape , describing localised ionisation or discharge events . we demonstrate in sect . [ ssec:2.3 ] how we model the time - dependent conductivity by the choice of a function @xmath81 that describes a flash - like ionisation event . the function @xmath81 determines the time - dependence of the conductivity by eq . , as well as the solution for the electromagnetic field by eq . . we wish to describe the conductivity flash variation most accurately by allowing it to be as flexible as possible . we require our model to be able to describe two different types of lightning discharges : a capacitor - like one and one induced by runaway electrons . a time - dependence corresponding to a runaways induced discharge , like that known for cosmic ray induced lightning discharges @xcite , starts as exponential growth . unlike a capacitor - like discharge , which would start with almost linear growth and requires a high degree of charge separation , runaway discharges require less initial potential difference ( see e.g. @xcite ) . the runaway - initiated discharges are therefore more easy to start , subject to existence of a small population of runaway electrons . the time - dependent conductivity of such initiated discharges starts as an exponential growth of the conductivity due to the breakdown of the avalanche . we represent the time - dependent conductivity by the function @xmath81 which allows us to describe both the capacity and the runaway breakdown discharges as part of the same formulae : @xmath97 the conductivity changing with time from eq . is now @xmath98 the parameters @xmath99 , @xmath100 , @xmath83 , @xmath101 , @xmath102 and @xmath103 are used to describe the desired variety of the particular time - dependence s shape . figure [ f:4 ] demonstrates various cases of the time evolution of the flash - like conductivity that we are interested in . the blue line in fig . [ f:4](a ) shows a capacitor - like discharge , while two other curves demonstrate a runaway - initiated discharge ( like one triggered by cosmic rays ) . figure [ f:4](a ) shows _ intermediate_-value conductivity flashes , while fig . [ f:4](b ) shows small variations of conductivity within 10% of the background value , i.e. the value existing there before the flash ionisation event ( e.g. , background thermal ionisation in brown dwarfs ) . equation allows for the analytical solution of the problem of electromagnetic field transformation , while describing the desired features of the conductivity variations . the general solution for the electric field component is then described by eq . with the function @xmath81 as given in eq . . [ f:4a ] + [ f:4b ] this general solution for the electric field component describes a pre - existing radiation s response to a flash ionisation event . such events can take place in cloudy substellar atmospheres but also in , for example , protoplanetary discs . time - depending conductivity correspondent to a flash ionisation process could be expected in a form of a flash - like shape , firstly rising due to fast growth of the ionised particles at a breakdown ionisation stage , then reaching saturation and eventually declining when the process is over . direct and indirect ( through measuring a current ) evidence of a flash - shape time - dependent conductivity can be found in observations ( e.g. , on earth @xcite or on jupiter @xcite ) and experiments ( e.g. , nuclear lightning @xcite or laboratory experiment , @xcite ) . to check the relevance of our conductivity - flash model to lightning discharges and similar phenomena and to conclude on quantitative parameters which can be expected for the conductivity flash , we look into the observational results of a discharge associated with terrestrial gamma ray flash ( tgf ) analysed in @xcite . detection of radio signals in ulf and vlf was reported , with gamma rays detected within 0.2 ms of the fast discharge . similar distinctive variations of the waveforms were reported for another tgf - associated intra - cloud discharge @xcite . tgf production was reported to be associated with the upward propagating leader during the initial development of compact ( 1.52 km channel ) intra - cloud flash within 30 km of the sub - satellite point . we use the integral ohm s law @xmath104 to retrieve the conductivity variation in time , correspondent to currents and fields which resemble the observed waveforms @xcite . @xmath105 and @xmath63 in eq . are the electrons current and electric field , correspondingly . the current and the field have a flash - like pulse waveforms in observations for the events like lightning discharges or tgfs . often , this form is modelled with double - exponential representation @xcite . we use the representation @xmath106 for them , which describes the pulse shape reasonably accurate and allows analytical laplace transforms , while not leading to divergence in inverse laplace transform , as a double - exponential representations would do . the parameters @xmath107 , @xmath108 , @xmath109 and @xmath110 as well as the integration constants @xmath4 and @xmath83 are determined from reproducing tgf associated lighting discharge from @xcite in fig . [ f:9 ] . for known waveforms of the current and the field , eq . becomes an integral equation for the conductivity @xmath66 . the integral ohm s law reflects the fact that the current and the field are not only affecting each other at any particular moment , but the full pre - history of their time - evolution has its influence . eq . is a volterra type integral equation of a convolution type . this allows transforming it into an algebraic equation by laplace transform , @xmath111 , and deriving then the conductivity as follows : @xmath112 the time - dependent conductivity can be found from after applying an inverse laplace transform , @xmath113 the current and the field in eq . are determined by eq . . we use the current and the field resembling those in @xcite , their fig.3 , to retrieve the conductivity waveform ( its time - dependence ) . [ f:9 ] shows the current ( red , solid ) and the field ( blue , dashed ) and the resulting retrieved conductivities for two cases of different time delays between the field waveform emission and the discharge current . while the shape of the conductivity time variation does not depend on the time delay , its peak value depends on it dramatically . estimation of the time delay however relay on accuracy of estimations for the distance from the detectors to the discharge and for the correspondent velocities . the duration of the retrieved conductivity flash does not depend on that delay , and is almost ten times longer than that of the current or field pulses ( flashes ) . similar results were obtained for the current and field waveform reported in @xcite for saturn observations , but the conductivity flash duration is comparatively shorter , being about five times longer than that of the current and the field . our model described by eq . , with conductivity plotted in fig . [ f:4 ] is well representative of the flash ionisation events like lightning discharges , as comparison with the retrieved conductivity ( fig . [ f:9 ] ) suggests . [ ka ] ( red ) and of the magnetic component of electromagnetic field , @xmath100[nt ] ( blue ) , with the retrieved conductivity time dependencies [ 1/s ] for the correspondent @xmath114 and @xmath100 combinations plotted below them . the electric current and the field have the waveform parameters similar to the observational results for a tgf - associated lightning discharge reported in @xcite . two different plots are for different delays of the electromagnetic production with respect to the current . ] flash ionisation events like lightning , coronal discharges , energetic explosions or eruptions , can be suitably described by the time - dependent conductivity ( eq . ) , because this formula allows to adapt a suitable variation in amplitude , duration , shape and the flash front s gradient . we have demonstrated in sec . [ ssec:3.4 ] that our model does reproduce the well - studied case of tgfs current ( fig . [ f:4 ] , [ f:4 ] ) . we consider the conductivity s temporal variations through flash ionisation in a medium with a small value of natural background conductivity , e.g. through thermal ionisation ( @xcite ) . this background ionisation is physically required to start the flash ionisation . from a mathematical point of view , the background ionisation is not necessary , and all the derived formulae remain valid if the initial ( background ) conductivity is zero . figure [ f:4](a ) shows larger amplitude changes in conductivity , while fig . [ f:4](b ) demonstrates the changes within a few percent of the background value . the purpose of this paper is to study if and how the electric field changes for different flash ionisation events ( like coronal discharges , lightning or sprites ) , which we represent by various conductivity profiles ( eq . ; fig . [ f:4 ] ) . as we aim to find signatures of the flash ionisation events in pre - existing cyclotron radiation , we consider radio frequencies as the incident radiation field in our analysis , i.e. as a carrier signal for the modulation imposed by a flash - ionisation . the analytical results from the previous sections are valid for any frequency , i.e. for any relation between the initial wave frequency and the time scale of the ionisation process , as long as the assumption of a simultaneous conductivity change in different points of the local medium is reasonable . using the model formulae presented in sect . [ sec:2 ] , fig . [ f:5 ] demonstrates how the electric field ( eq . ) responds to a flash - like pulse due to a fast change of conductivity : * the response of the electric field begins when the conductivity flash has a high positive gradient . * the maximum of the field response corresponds to the time of the maximum conductivity gradient . * the field responds on a significantly shorter time - scale than the time - scale of the underlying process responsible for the rapid conductivity change . * the pulse duration of the response field is typically about ten times shorter than the conductivity flash duration . [ f:5a ] + [ f:5b ] for a fixed duration of the conductivity time - flash , the duration of the response field pulse s vs the conductivity peak value is shown in fig . [ f:7 ] . remarkably , * the field response becomes shorter in time the more powerful the ionisation process is ( i.e. for a larger peak value of the flash s conductivity ) . figure [ f:8 ] demonstrates that the field response also starts slightly later for flashes with larger peak value of conductivity , when making the comparison at the same fixed position of the conductivity maximum . this can be attributed to the fact that the maximum gradient of the flash is reached slightly later for larger flashes . the duration of the responding electric field pulse ( fig . [ f:5 ] ) is at least one order of amplitude shorter than the duration of the ionisation flash . , at its maximum value , vs the maximum ( peak ) value of relative conductivity change with respect to the background conductivity , @xmath115 . dashed and solid curves correspond to cases of conductivity rising linearly up at the beginning of the flash process @xmath86corresponding to the blue curve in fig . [ f:5](a)@xmath88 and conductivity with exponential growth at the beginning of the flash @xmath86corresponding to the red curve in fig . [ f:5](a)@xmath88 . one can see that bigger field response does not always correspond to bigger flash of conductivity ( i.e. larger peak value of conductivity ) . ] investigating the dependence of the amplitude of the electric field response pulse on the amplitude of the conductivity flash ( fig . [ f:6 ] ) shows that for small variations of the conductivity in time @xmath86within a few percent , as shown in fig . [ f:4](b)@xmath88 the field s change is not noticeable . * for _ intermediate _ changes in time , however , the response amplitude will be larger for more powerful ionisation ( larger flash of conductivity , see fig . [ f:8 ] ) . note , that the conductivity can be described as sums of the products of densities for all available charges and the charges mobilities , @xmath116 where @xmath117 is the electrons drift velocity and @xmath118 is the local electrostatic electric field . during the avalanche ( breakdown ) process , the mobility of electrons as well as their number density is much higher than those of ions or charged dust grains , so we only take electrons motion into account . the conductivity is then determined by @xmath119 where @xmath120 is a charge of the electron , @xmath121 is electrons number density and @xmath122 is the mobility of the electrons . a higher peak value of conductivity requires therefore a combination of strong ionisation and charge acceleration . . the duration is measured at @xmath123 of the field s maximum value at its peak . colour code is the same as in fig . the response field pulse is consistently shorter for larger flash of conductivity ( larger peak conductivity value ) . ] , on maximum value of relative conductivity change with respect to the background conductivity , @xmath115 . the start time is determined on the field - rising slope at @xmath123 of the field s maximum value at its peak . colour code is the same as in fig . the field response starts consistently later for larger flash of conductivity ( larger peak conductivity value ) . ] for larger peak conductivities , the response field peak value becomes less sensitive to the further increase of the conductivity , with the dependence shown in fig . [ f:6 ] reaching saturation . points shown with a triangle , a circle and a diamond in fig . [ f:6 ] correspond to the curves marked with the same symbols in fig . [ f:5](a ) and lay within _ intermediate _ flash ionisation intensity . increasing the peak conductivity even further ( considering more powerful flash ionisation processes ) leads to reversal of the effect : the field response becomes smaller for larger conductivity flash , i.e. the normalised field amplitude , @xmath124 decreases with increasing the relative conductivity change ( @xmath115 ) . it is known that lightning discharges radiate an electromagnetic field ( e.g. , @xcite ) . this radiation occurs when runaway electrons , produced by ionising discharges , form conducting currents which act as radiating antennae . the radiation is produced during the energetic phase of the discharge , when the runaway electrons are still fast enough to radiate , and when their radiation is not yet fully damped by collisional losses . similarly , the short - term amplification of the pre - existing radiation in a flash - ionised medium , as seen in fig . [ f:5](a ) , is caused by contributions of the energetic electrons radiative currents . both , mobility and charges number density , influence conductivity , as seen in eq . . as noted in sect . 1 , the conductivity in our model is a combination of all effects from ionization and charges movement , including the reactive and the dissipative responses . the peak of the field amplification corresponds roughly to the maximum gradient of the rising conductivity . this reflects on the dynamics between two mechanisms of rising conductivity : the one associated with increasing charges mobility and the one associated with increasing number density of charges . runaway breakdown at the first stage of the discharge dramatically increases the charge mobility aspect of the conductivity , while consequent continuing increase in conductivity is caused mostly by the increased number of charges from collisional ionisation . this second stage of the conductivity rise will effectively damp the produced radiation / amplification . * * while one might intuitively expect that a more powerful ionisation process would lead to a higher response in the electromagnetic field s amplitude , our results show that there is some threshold in the peak value of conductivity ( which is different for different initial conditions and shapes of the conductivity time - flashes ) , corresponding to the peaks , after which ( i.e. for the conductivity change bigger than that value ) the field response decreases for larger flashes . the reason for this is that , above a certain threshold , the conductive medium would be efficiently damping the field , allowing the damping to prevail over the effect of pumping the field by the transient event s energy . an ionised medium ( like the ionosphere ) is generally non - transparent to most of the radio frequencies . this is one of the biggest problems with observing lightning discharges from outside of the solar system or terrestrial planets . with additional energy coming into the radiation field from powerful ionisation processes , such a field has a chance to penetrate through even an initially ionised medium . however , processes of higher extremity ( involving larger peak value of conductivity flash ) would result in a metal - like ionised area which would damp the electromagnetic field dramatically . the classification is based on the effect which is produced by the transient event on the electromagnetic field , e.g. , ` strong ' event ( ` big ' conductivity flash ) attributes to damping the field , as in fig . [ f:5]b , while ` intermediate ' event denotes the flash resulting in the field amplification , as in fig . small events ( i.e. , the ionisation events which change the conductivity within a few percent of its initial value ) do not produce noticeable signatures . the field s response amplitude can not therefore be used as a single criterion for the power of the underlying ionisation process , and most powerful flash ionisation processes ( including , for example , discharges / lightning and explosions / eruptions ) can be missed by observations . the duration of the electric field response pulse , however , provides the information about the magnitude of the conductivity flash , as it universally decreases with the increasing conductivity flash amplitude for the same fixed duration of the conductivity flash . we apply our model to the case of radio wavelength being the pre - existing signal that can be modulated by a flash - ionisation event . our choice of parameters is guided by observations of cyclotron radio emission from brown dwarfs and late m dwarfs . 4.8 ghz frequency observations were reported from a late m dwarf @xcite , and observations of radio emission from brown dwarfs in the frequency range 4.4 4.9 ghz were presented in @xcite . @xcite also observed cyclotron emission from brown dwarfs at 4.88 ghz and 8.44 ghz , and radio emission at 5.8 ghz was reported by @xcite . the radiation frequency , being related to local cyclotron frequency of electrons , is proportional to local magnetic field . brown dwarfs and other ucds magnetic field is in order of a few kg @xcite , providing the cyclotron radiation frequency in order of a few ghz . a few gauss planetary magnetic field results in a cyclotron radio emission in tens khz few mhz frequency range ( e.g. , @xcite for jupiter ) . [ f:2a ] + [ f:2b ] all our typical results are shown for @xmath125=4.8@xmath126 ( cyclic frequency ) in figs . [ f:5]-[f:8 ] . they are representative for other frequencies as well , as we demonstrate by fig . [ f:2 ] for the alternative incident radiation frequency @xmath127 ( ) . the transformed electric field given by eqs . do not scale with the frequency exactly . however , within the radio frequency range the temporal characteristics of the field ( start time and duration of the response , figs . [ f:7 ] , [ f:8 ] ) still roughly scale with the frequency as @xmath128 , which makes our analysis more general . another key input value for the model of the flash is a peak value of conductivity . we showed in sect.[ssec:3.4 ] that the peak conductivity corresponding to terrestrial tgfs can be expected within a wide range of values , depending on an actual delay time between the measured current and field . for a cloud - to - ground discharge on earth , the peak value of temporal variation of the local temperature @xmath129 is typically about @xmath130k @xcite . a value for the conductivity derived by these authors was calculated to be @xmath131 . ] ] this is the conductivity that can be expected within an active lightning channel on earth . for the surrounding area , where the lightning would produce flash ionisation of some ( sometimes substantial ) volume , the conductivity is smaller than in the channel , which could correspond to both ` big ' and ` intermediate ' flash regimes ( as defined in sect . [ sec:4 ] ) . however there are no values for the peak conductivity for those areas in literature . only the ` intermediate ' flash effects can be seen by short - time amplification as in fig . [ f:5](a ) . however , the damping of the field as shown in fig . [ f:5](b ) can also serve as an indication of an atmospheric flash , though high intensity initial ( cyclotron ) radiation or high signal resolution , sufficient for detecting 30%60% variations in amplitude @xmath86such degree of damping of the initial field is suggested by fig . [ f:5](b)@xmath88 is required . the conductivity of a medium can be calculated using eq . . runaway electrons from the channel ionise the surrounding medium . the conductivity is then proportional to the electron drift velocity , @xmath117 . the velocity , and therefore the mobility , is exponentially decaying with time @xcite after the short initial period of electrons acceleration by the ionisation flash , @xmath132 the number density of electrons is growing exponentially at runaway breakdown before it reaches a saturation at about or after the moment when the electrons start slowing down , @xmath133 . a product of these two exponents , as in eq . , one of which is time - shifted ( with the decay starting with some initial time delay ) and another is growing till the moment of saturation , results in the flash time - dependence of conductivity . according to @xcite , the ` burning point ' at lightning discharge moves at @xmath134 km / s , while the tip velocity of sprites from microdischarge model is @xmath135 km / s @xcite . exponential decay in the runaway electrons velocity , @xcite would result in an average peak conductivity across the transient region of about @xmath136 s@xmath137 . the @xmath138 s duration of the conductivity flashes was considered above as this would be within average values for the duration of a cloud - to - ground discharge on earth . the peak conductivity of @xmath135 s@xmath137 corresponding to this duration is small enough to be classified as an ` intermediate ' flash which amplifies the field , as fig . [ f:5]a suggests . with high - pressure atmospheres like those of bds the number of branches in lightning trees grow significantly with increasing pressure @xcite . this makes our model of a homogeneous time - dependent conductivity more easily applicable to atmospheres of brown dwarfs . lightning on other planets and on brown dwarfs can be more powerful and longer in time than on earth , see e.g. @xcite . the duration of the flash ionisation of the area surrounding the lightning channel can also be expected to be longer than the lightning flash itself . for longer conductivity flash , the peak conductivity as @xmath139 s@xmath137 would still be within ` intermediate ' flash limits , being therefore detectable by the field amplification . longer duration of the flash ionisation process would also mean better chance of detecting the response signal under the same time - resolution of detection . thus , high time - resolution of detection would be sufficient for detecting highly amplified pulse response of the field from ` intermediate ' conductivity time - flash , while both high time resolution as well as intensity resolution ( and so , high signal - to - noise ratio ) are needed to detect stronger flash events . powerful flash ionisation events like tgfs , are expected to produce effects of ` big ' flashes , according to the high peak conductivity values as in fig . they are therefore detectable only by the field s damping . note , that when mentioning ` big ' and ` intermediate ' flashes of conductivity , we compare the conductivity peak values for the same duration of the conductivity flashes . for the duration being in order of @xmath140 s the boundary between the ` strong ' and the ` intermediate ' flash events is around the conductivity peak values at few thousand @xmath141 . for longer duration , higher conductivity peak values are required to classify the flash event as strong. + + + + lllrr cases in fig [ f:1 ] : & a & b & c & d + + a & @xmath142 & @xmath143 & @xmath144 & @xmath145 + b & @xmath146 & @xmath146 & @xmath147 & @xmath147 + c & @xmath148 & @xmath149 & @xmath150 & @xmath151 + d & -1.44 & -1.44 & -23.04 & -23.04 + @xmath102 & @xmath152 & @xmath152 & @xmath153 & @xmath153 + k & 2.0 & 2.0 & 2.0 & 2.0 + + duration @xmath154 [ s ] & 30 & 30 & 120 & 120 + peak value @xmath155 [ s@xmath137 ] & @xmath157 & @xmath158 & @xmath159 & @xmath160 + & intermediat & large & large & large + & flash & flash & flash & flash + + & amplified & damped & damped & damped + the discussed above signatures of the cyclotron emission transformation by a flash ionization event can be expected in a form shown in fig . 3 in the case when the emission is observed continuously . due to high directivity of the cyclotron emission and to the emitting body s rotation , this can only happen if the geometry of rotation allows the radiated beam / cone staying at the same place with respect to the observer , or if the observing station is following the body s ( planet s ) rotation . in most of the cases the emission is seen as periodic bursts at the phases when the beam turns towards earth ( e.g. , @xcite ) , as in fig . [ f:1 ] , top . if a flash ionization event happens on the path of one of the periodic beams ( fig . [ f:11 ] ) out of the atmosphere , it would not change the periodicity but influence the signal s amplitude . figure [ f:1 ] visualises four example cases how flash ionisation can alter a cyclotron signal . the beam shapes were calculated from @xmath161+\mbox{sech}[10\cdot(2t-3)]+\mbox{sech}[10\cdot(2t-5)]$ ] ( @xmath162 , t time in units of phase ) but any suitable function ( like gaussian ) could be used . please note that the plots in fig . 9 do not include rigorous modelling of the propagation of the transformed ( by flash ionization ) signal towards the observer . instead , a model modulation function ( imitating effects of the emission source s rotation ) shown in fig . 9 ( a ) is used . the plots in fig . 9 ( b - d ) are obtained as products of the field transformed by the transient event and the modulation function . this way , the plots represent predictions for the observations , while plots in fig . 4 present the transformed field at the source.the top panel in fig . 9 summarizes the ideal , unperturbed situation where two emission beams reoccur after one rotational period , @xmath163 . the half - height width of the peaks is determined by the wall thickness , @xmath0 , of the emission cone as shown in fig . [ f:11 ] . the time interval between the two beams , i.e. between the emission maxima , is related to the emission cone opening angle , @xmath2 ( fig . [ f:11 ] ) . the changing peak height and width of the cyclotron emission due to an encounter with a lighting flash , for example , is show in the second and third row in fig . these figures result from solving eqs . [ e:16 ] , [ e:17 ] for the electric field component applying a parametrised conductivity function ( eq . [ e:19 ] , see also sect . [ ssec:2.3 ] ) . the necessary parameters for the presented results are summarized in table [ t : abcn ] . depending on the peak value of the conductivity flash induced by the ionisation and on its duration , the field modulation by the flash is either a short powerful burst of radiation ( sharp increase of the amplitude , as in fig . [ f:1 ] ( _ case ( a ) _ ) , or a damping of the whole radiation peak , fig . [ f:6 ] ( _ cases ( c ) , ( d ) _ ) , or its fragment , fig . [ f:1 ] ( _ case ( b ) _ ) . the transformation of cyclotron emission which is seen in observations of jo746 + 20 in @xcite , their fig . 5.01 , looks similar to shown in fig . [ f:1 ] and can potentially be a sign of flash ionisation events affecting the cyclotron emission . detecting a signal pattern as shown in fig . [ f:1 ] can be indicative of flash ionization events in the object s atmosphere and magnetosphere ( like lightning or tgf ) . the cyclotron emission is produced by an electron beam moving along the star / dwarf / planet s magnetic field lines . its power depends on the beam s properties , including its geometrical properties , and number density and velocity distributions . significant amplification of a regularly seen emission , like shown in fig . [ f:1 ] ( b ) and [ f:1 ] ( c ) , is unlikely to happen due to sudden changes in the beam , because of unrealistically high amplitude for cyclotron emission . damping of the whole signal or its fragment , however , could also be caused by the electron beam variations . simultaneous optical observations can help to exclude or confirm the beam s variations , as visible aurora is induced by the same beam as the radio cyclotron emission . variation seen in the visual aurora s geometrical pattern and intensity are expected to correlate with variations in radio emission from the same beam . the radio waves refracts differently compared to optical waves , hence , radio ways will travel through different part of the atmosphere than optical waves . therefore , if the radio waves pass through a localised flash ionisation event , the optical light may miss it . in a situation when the visible light is still affected by the same event , this effect should be smaller than that for radio emission , because the visible light is emitted and propagates in a thicker cone . therefore the observed visible signal is an integral effect of the light emerging from a significantly larger external surface of the atmosphere than that of a directed radio cyclotron beam . this way , only a small portion of this visible light can be affected by a local event , making the effect of flash ionisation on visible light smaller compared to its effect on radio waves . another reason for cyclotron emission variability is spin - modulation of the emission due to a character of the star / planet s rotation . in particular , when a planet s magnetic dipole is offset from its rotation axis , the solar / star wind electrons at different rotation phase have different paths towards the magnetic poles . this results in the emission produced at different altitudes , where different values of magnetic field account for the emission variability @xcite . the duration of the response field pulse ( including both cases of amplification and damping in the pre - existing radiation waveform ) is determined by duration,@xmath154 , of the conductivity flash . as discussed in sect . [ ssec:3.4 ] , for the single discharges in earth atmosphere the response field pulse is about ten times shorter than the duration of the discharge . for saturn it was found that it is only five times shorter @xcite . though single branches of lightning in earth atmosphere are known to last for a few @xmath164s to a few @xmath165 , a lightning event consists of a tree of strokes , coming as groups of branches . @xcite , studying multiple - discharge flashes in the atmosphere of earth , found that flashes in each group are bunched together for 15-20 min . there are also reasons to expect longer discharges in extraterrestrial atmospheres with different compositions and higher local pressure . the duration of discharges in air is limited by electron attachment because free electrons are easily lost in air through attachment to oxygen . duration of discharges in other gases and under higher than earth atmospheric pressure can be significantly longer than in air , as proved by experiments in @xcite . the field s response stretches in time proportionally to the longer integral flash duration , reaching about 5 - 10 folds of the flash s duration , which makes it observable with current instrumentation for large / average expected flashes of lightning . a time resolution on a scale shorter than the response pulse duration is needed to follow the shape of the response filed pulse ( waveform ) . however , for its detection , even one measurement within its duration will be indicative ( fig . [ f:1 ] , lower panels ) which can be obtained even when the resolution is lower than the response duration . this makes the detectability subject to the statistical occurrence of transient ionisation events which is well studied on earth , and to a lesser extend for the solar system planets . from fig . [ f:1 ] one also concludes that for an example period of @xmath166 hours a large flash would cause a substantial damping of the emission peak by @xmath167 with a time - dip of @xmath168 ( _ case ( d ) _ in fig . [ f:1 ] ) . intermediate flashes do produce a substantial amplification over a very short time window ( _ case ( a ) _ in fig . [ f:1 ] ) . we have shown that transient ionisation events can leave an imprint on radiation that intercepts a time - dependently ionised gas , like an atmosphere affected by lightning discharges . in the case of brown dwarfs , cyclotron maser emission is a good candidate for the probe signal. observing the cyclotron periodic bursts for several rotational periods ( not necessarily consequent ) is needed to spot a variability . to interpret it as a signature of a flash ionisation event which happened at a location along the cyclotron signal s path , we suggest the following analysis of the detected varied electromagnetic field : 1 . several orders of magnitude s flash amplification of the initial signal , on a timescale comparable to the cyclotron radio emission wave - period , is a strong indication of a flash ionisation event . the detection of such amplitude amplification of the cyclotron emission ( or any other signal carrier ) requires a high time - resolution . the amplification seen in observations ( fig . [ f:5](a ) for continuous emission , [ f:1](b)-(c ) for periodically seen emission ) suggests an ` intermediate ' peak value of the conductivity flash produced by the ionisation being the cause . 2 . flash ( e.g. , brief , short term , on a time - scale comparable to the cyclotron emission wave - period ) damping of the observed radio signal ( fig . [ f:5](b ) for continuous emission , [ f:1](c)-(e ) for periodically seen emission ) can be a signature of a strong flash ionisation . it can though be confused with effects of beam variations and with spin - modulation of the cyclotron emission due to the source s rotation . 3 . simultaneous observations in optical wavelength can help to distinguish the damping effect of the flash ionisation from effects of the beam variations ( i.e. , variations in the beam of electrons producing both optical aurora and radio emission ) . if there is no correlation between the radio variability and the variability in the optical aurora , the influence of beam variations can be ruled out as both radio and optical emissions are produced by the same beam of electrons . spin - modulation of the cyclotron radio emission due to the source s rotation @xcite can be ruled out or subtracted from the observations if observing the object for several consequent periods of rotation . brown dwarfs are thought to have turbulent atmospheres , as concluded from variability measurements and cloud detections ( @xcite ) , e.g. , variability seen in luhman 16b @xcite . the turbulent atmospheres have a higher rate of transient flash ionisation events like discharges or explosions , as seen for saturn @xcite . some of brown dwarfs known for radio cyclotron emission are also fast rotators , e.g. , tvlm513 has a period of just 1.9673 h @xcite . this makes them easier objects for multi - period observations . the motivation behind this research is to find a way to probe atmospheric discharge processes which are otherwise inaccessible . for this purposes , we investigated if and how a coherent emission ( e.g. a cyclotron maser radio emission ) could be modulated by passing through a region with flash ionisation processes in , for example , an atmospheric environment or a magnetosphere . our results are applicable to a wider variety of problems in astrophysics when a pre - existing radiation travels through a rapidly ionising medium . these can include discharges in protoplanetary disks @xcite and possibly fast radio bursts ( see e.g. @xcite which could be attributed to flash transient events if their variations are not periodically detected . our model ( eqs . ) can also be applied to microwave background changed during the reionisation epoch @xcite if the limitations of 1d space - uniform consideration are reasonable for a particular scenario . we develop a model to describe the influence of flash ionisation events on pre - existing radiation that passes through the region where the transient events happen . we solved a ( direct ) problem of electromagnetic field transformation as a result of the flash transient events and provide a recipe for observers in sect . [ ssec:5.1 ] . our main conclusions are 1 . the transformation of coherent emission by fast transient processes in the medium of its propagation can be observable ; 2 . attributing larger signal perturbations to more powerful processes is not always correct , as more powerful processes can lead to smaller response signals ; 3 . flash processes of intermediate intensity can be detected by short - term amplification of the signal . the response field s duration can be related to the flash ionisation s time scale and intensity ( fig . [ f:7 ] ) ; 4 . if the initial emission signal is powerful enough to detect the character of the damping , the small pulse s amplitude and duration in the response field can provide information about the ionisation flash magnitude ( peak value of conductivity ) and duration ; 5 . if the detection of the field s variation is possible for several related flashes ( attributed to the same undergoing event , like multiple strokes of the same cloud - to - ground lightning discharge ) and allows deriving the conductivity peak values from additional information , it would be possible to determine whether the discharges started as a runaway electrons breakdown or rather like capacity discharges , by deriving the dependence of the field responses maximum values on maximum value of the conductivity change . absence of a noticeable plateau in this dependence ( see fig . [ f:6 ] ) suggests that the runaway breakdown discharge is likely to be a source of the ionisation , which requires about ten times lower voltage to enable the discharge in comparison to a conventional capacity - like discharge . if the plateau is noticeable , then a capacitor - like discharge is more probable . the authors gratefully acknowledge the support from the european community under the fp7 by the erc starting grant 257431 and are thankful for computer support from the university of st andrews , supa school of physics & astronomy . majority of reference search was done with a help of ads which is appreciated . we also thank paul rimmer for reading and commenting on our manuscript .

we present a realistic theoretical treatment of a three - level system in a hot atomic vapor interacting with a coupling and a probe field of arbitrary strengths , leading to electromagnetically - induced transparency and slow light under the two - photon resonance condition . we take into account all the relevant decoherence processes including collisions . velocity - changing collisions ( vccs ) are modeled in the strong collision limit effectively , which helps in achieving optical pumping by the coupling beam across the entire doppler profile . the steady - state expressions for the atomic density - matrix elements are numerically evaluated to yield the experimentally measured response characteristics . the predictions , taking into account a dynamic rate of influx of atoms in the two lower levels of the , are in excellent agreement with the reported experimental results forhe*. the role played by the vcc parameter is seen to be distinct from that by the transit time or raman coherence decay rate .

0901.3790

electromagnetically - induced transparency ( eit ) in three - level @xmath0-systems is a phenomenon in which an initially absorbing medium is rendered transparent to a resonant weak probe laser when a strong coupling laser is applied to a second transition @xcite . it is based on quantum interference effects involving coherence between the two lower states of the @xmath0 . the quest has been on for simple room - temperature systems capable of demonstrating eit for applications ranging from narrow transparencies to switchable and controlled broadband delays , and slowing of light @xcite for use in quantum - information processing . room - temperature atomic vapors have been found to be attractive candidates for such applications , necessitating a treatment of the phenomenon taking into account effects of atomic motion and collisions . there are some early studies @xcite of doppler - broadening effects in eit for a system of moving atoms . in the limit of vanishing probe field and under the assumption that all atoms are trapped in the dark state ( which is a coherent superposition of the two lower levels of the @xmath0 ) , it was found that power - broadening of the eit linewidth takes place : @xmath2 ( where @xmath3 is the rabi frequency of the coupling field and @xmath4 is the doppler half - width at half - maximum ) , which is similar to the well - known result for a homogeneously - broadened system : @xmath5 ( where @xmath6 is the homogeneous linewidth ) . this dependence was experimentally verified in ref.@xcite . in the limit of relatively low probe field intensity , @xmath7 , and under the same assumption of full coherent trapping ( i.e. , neglecting the two - photon coherence decay ) , other workers @xcite have derived the following result for the eit linewidth : @xmath8 , where @xmath9 is the rabi frequency of the probe field . for eit in a room - temperature gas , the decay of two - photon ( raman ) coherence is caused by several mechanisms , such as transit - time broadening , population exchange , and atom - atom and atom - wall collisions . insight into the most significant decoherence mechanism can be gained by measuring the width of the eit resonance as a function of the coupling field intensity . an existing theoretical treatment of eit in doppler - broadened gases @xcite , assuming the population exchange between the lower levels to be the main source of decoherence , predicts a _ nonlinear _ dependence of the eit width @xmath10 on the coupling beam intensity for weak coupling powers . in the limit of very large coupling intensity , it is shown to reduce to the power - broadening case . et al_. @xcite consider a closed atomic model scheme and argue that such a model gives a description almost equivalent to the one for an open system in which atoms decay ( out of the interaction region ) with the rate @xmath11 , and atoms come into the interaction region with equally populated lower levels . though it is a theory for eit in a doppler - broadened medium , the role of collisions is completely neglected in ref.@xcite . most existing experiments in atomic vapors @xcite have shown the dependence of the width of the eit resonance on the coupling field intensity ( @xmath12 ) to be _ linear _ , even for weak coupling powers ( with the exception of the work by ye and zibrov @xcite which was performed without a buffer gas and with a very small beam diameter ) . as an example , we focus on the data of our recent demonstration @xcite that metastable @xmath1he ( he * ) at room temperature is a simple system capable of yielding eit and slow light . for the coupling field strengths used , @xmath10 is expected to evolve linearly with @xmath3 according to ref.@xcite , with a slope depending on @xmath11 . the lower - level relaxation in a gas is mainly determined by the transit time of the atoms through the laser beam different beam sizes would lead to different transit times and hence different values of @xmath11 . our experimental results @xcite have clearly shown that ( i ) @xmath10 evolves quadratically with @xmath3 , and ( ii ) the slope of this evolution is the same for different beam sizes , i.e. , for different values of @xmath11 . the suggestive role played by collisions between metastable and ground - state atoms in yielding this experimental result is not described by the theory of ref.@xcite . if all the atoms across the entire doppler profile are assumed to be initially optically pumped by the coupling beam to the probe ground level , a calculation @xcite of the response of the medium up to first order in the probe field leads to a linear dependence of the eit linewidth on the coupling beam intensity @xcite . to obtain this result , one also supposes that the decoherence in the lower states is caused by pure dephasing , contrary to the assumptions of population exchange in ref.@xcite . in a realistic situation , the special initial condition of the entire atomic population being in the probe ground level does not hold good the population is equally likely to be in the probe and the coupling ground levels initially . thus , population exchange between the two lower levels can not be ignored . also , a treatment in first ( linear ) order in the probe field can not possibly give results when the coupling field is small , viz . of the order of the probe field . in this paper , we address this deficiency in the existing theory of eit and slow light in a hot atomic vapor , and attempt a complete , realistic analysis taking into account all the relevant decoherence processes , for arbitrary strengths of the probe and the coupling fields , and without assumptions of any special initial condition . we consider the influx of fresh atoms in the lower levels and the outflux from all the levels at a diffusive transit rate in the gas . we allow for unequal rates of influx in the lower levels to take into account optical pumping by the control field and return of coherently - prepared atoms into the interaction region . the phase - interrupting and velocity - changing collisions ( vccs ) of active atoms are also modeled effectively . apart from the transparency width discussed so far , there are other features of interest associated with eit . for non - zero detunings of the coupling field from the center of the doppler - broadened transition frequency , the transmitted intensity profiles become asymmetric about the two - photon resonance ( raman detuning = 0 ) @xcite . this fano - like feature is a signature of the two - photon process of eit , and emerges naturally from our model . the narrow spectral hole in the absorption profile of the eit medium is accompanied by a strong dispersion of the index of refraction according to the kramers - kronig relations , inducing a low group velocity . the evolutions of the peak transmission and the group delay with the coupling beam intensity predicted from our analysis faithfully reproduce the experimentally observed behaviors . the paper is organized as follows . in sec.ii , all the different relaxation processes for eit in a three - level @xmath0-system are discussed . the fraction of atoms that come back to the beam from outside being still coherently - prepared are suitably modeled in this section . in sec.iii , vccs are dealt with separately , and the density matrix equations are written with various relaxations including that due to vccs . the steady - state solutions for the density matrix elements are presented in the strong - collision approximation with a model for the vccs under rapid vcc coverage . this is followed by our theoretical results in sec.iv on the doppler - averaged fano - like eit profiles , the variation of the eit width , the peak transmission and the group delay with the coupling intensity , all of which agree very well with the experimental data for the he * system . the general dependence of these features on the vcc parameter , the unequal atomic influx parameter , the raman decoherence rate and the initial transmission are also probed . it is shown that the unequal feeding back of atoms into the lower levels , more being in the coherently - prepared dark state @xmath13 , has an important effect on various characteristics of an eit medium . it is also shown that the impact of vccs is distinct and can not be incorporated by simply modifying the transit time decay ( and hence the raman coherence relaxation ) rate . the conclusions are presented in sec.v . consider a @xmath0 atomic system as in fig.[lambda ] . levels @xmath14 and @xmath13 are coupled by a weak probe field , the interaction energy being given by its rabi frequency @xmath9 . another strong coupling field of rabi frequency @xmath3 couples the same excited level @xmath14 with level @xmath15 along the other arm of the @xmath0 . both fields are treated classically . the probe detuning is @xmath16 with @xmath17 as the probe frequency , and likewise , the coupling field detuning is @xmath18 with @xmath19 as the coupling frequency . the raman detuning is @xmath20 the probe and the coupling fields propagate in the same direction , and the frequency difference @xmath21 is small enough so that the residual doppler shift @xmath22 can be ignored @xcite . for eit , the system should be prepared by optical pumping so that the initial population is entirely concentrated in the dark state @xmath23 . the coupling field creates a quantum interference between the probability amplitudes of transition @xmath13 @xmath24 @xmath14 via two different channels , ( i ) a direct absorption process from @xmath13 to @xmath14 , and ( ii ) an indirect stimulated raman process from @xmath13 to @xmath14 to @xmath15 to @xmath14 . under the appropriate condition of two - photon resonance @xmath25 = 0 , the medium becomes effectively transparent ( zero absorption ) for the probe field , leading to eit . scheme for eit , scaledwidth=40.0% ] in the example of @xmath1he , the first excited state 2@xmath26s@xmath27 is a metastable state with a lifetime of approximately 8000 s. the transition between 2@xmath26s@xmath27 and the second excited state 2@xmath26p at a gap of 277 thz or 1083 nm is conveniently used . the levels 2@xmath26s@xmath27 and 2@xmath26p@xmath27 are each split into three sublevels with @xmath28 = @xmath291 , 0 and 1 . out of these six levels , a @xmath0 system is carved out by optical pumping ( with the help of proper choice of polarization of light , selection rule and allowed stimulated / spontaneous transitions ) @xcite , with @xmath28 = @xmath291 and 1 of 2@xmath26s@xmath27 forming the two lower levels , and @xmath28 = 0 of 2@xmath26p@xmath27 the upper level . at zero magnetic field the lower levels are degenerate and @xmath30 . the dispersion and absorption of the medium with respect to the probe field of rabi frequency @xmath9 are determined by the off - diagonal element @xmath31 of the atomic density matrix . this describes the atomic coherence or the atomic polarization . hence , it is important to investigate the different relaxation processes affecting the optical coherence as well as the raman coherence of the two lower levels that lead to the dark state and eit . it is known that collisions of active atoms , which only perturb the phase or amplitude of an oscillating atom without changing its velocity , lead to homogeneous line - broadening and a shift of its line center . but collisions can also result in changes in the velocity of active atoms , in addition to being phase - interrupting , and affect atomic coherences of the system under consideration @xcite . the present theoretical analysis examines the density matrix equations for a three - level @xmath0-system interacting with two fields , in the presence of the following different sources of relaxation : 1 . the spontaneous decay from the excited state @xmath14 transfers atoms to the ground states with equal decay rates @xmath32 , 2 . the transit of the atoms through the laser beam at a rate @xmath33 replaces atoms ( in all states ) in the laser interaction region by fresh atoms arriving ( only in the lower states ) from the volume outside that region . the transit rate depends on the pressure ( diffusion coefficient ) of the gas . all the populations and coherences are affected because of this motion . but it is the lower - state relaxation that is mainly determined by the transit rate @xmath33 , 3 . the collisions ( phase - interrupting ) damp the atomic polarizations and coherences , 4 . the collisions also change velocities of the active atoms without changing their internal state . in our example of the he * system , the atom - wall collisions are not very significant for the dynamics . de - excitation of he * on collision with cell walls leaves inert he atoms in the ground state ( 1@xmath34s@xmath35 ) , which can not be detected : there are no background atoms to contribute to noise , unlike experiments which use true ground - state atoms such as the alkali metals . thus , with our system , there is an advantage of collisions of the active atoms with the walls of the cell which result in quenching . the decay for the optical coherences is : @xmath36 where @xmath37 is the collisional dephasing due to pressure , and the raman coherence decay for @xmath38 is : @xmath39 where @xmath40 is the collisional term proportional to gas pressure , and @xmath41 is the dephasing due to inhomogeneity in residual magnetic field and other possible dephasing mechanisms . the mean free - path of the he * atoms is , in a hard sphere model , of the order of 0.1 mm . if we consider that the atoms cross the beam in a one - dimensional random walk , we can see that , at 300 k , they experience about 10@xmath1 collisions during their trip across a 1-cm - diameter beam , leading to a diffusive transit time @xmath42 of the order of 0.5 ms . according to a rigorous calculation by fitzsimmons @xcite , the diffusion constant at 1 torr and 300 k is @xmath43 = 500 @xmath44/s . thus , in a 1-d diffusion model , the variance in the displacement is @xmath45 . for @xmath46 = 1 cm , we again obtain the transit time , @xmath47 = 1 ms . for metastable helium at room temperature and pressure of 1 torr inside a cylinder shielded with @xmath48-metal to avoid stray magnetic fields , the various decay rates are typically : @xmath49 the coherences between the lower levels may benefit from the fact that the atoms can diffuse out of the interaction region and return before decohering . the rates at which the atoms return to the lower states @xmath13 and @xmath15 are not likely to be equal , since inside the beam , the populations in these two states are made unequal by the control - field optical pumping from @xmath50 to @xmath51 , and hence the population diffusing outside is also likely to be unbalanced . the rate of return from outside can not be a constant but would depend dynamically on the population difference , @xmath52 . this feature is incorporated by using unequal influx rates : @xmath53 , \\ \left.\frac{\mathrm{d}\rho_{cc}}{\mathrm{dt}}\right|_{\mathrm{in } } & = & \frac{\gamma_{\mathrm{t}}}{2 } \left [ 1 - \beta \left ( \rho_{bb } - \rho_{cc } \right)\right],\end{aligned}\ ] ] while maintaining a single departure rate of @xmath33 from the beam in all the states . a value of @xmath54 would indicate that all atoms going back to the beam are prepared for eit in the dark state @xmath13 . on the other hand , @xmath55 = 0 would indicate that equal number of atoms enter the beam in states @xmath13 and @xmath15 . the overall atomic population is , of course , conserved . a physical picture of @xmath55 can be thought as arising from the treatment of the atomic motion outside the laser beam in a diffusion equation , by assuming a random distribution of the durations spent by the atoms outside the interaction region @xcite . velocity - changing collisions ( vccs ) which shuffle atoms between different velocity classes represent an important source of relaxation for the lower states . it can modify the atomic velocity without affecting the atomic coherence in the lower states : in this case the atoms prepared by the laser radiation in a dark state are transferred to other velocity classes @xcite . if this transfer applies with high efficiency , all the atomic velocity classes are pumped into the dark state , either by direct pumping or by vcc @xcite . our active atoms are in a three - level @xmath0 configuration . it is taken that collisions do not possess sufficient energy to induce transitions between the upper and the lower levels . this assumption effectively allows one to treat the scattering of each active atom separately , such that one can apply standard quantum mechanical scattering theory using a different total energy for each active atom level . we use the _ impact approximation _ in which the active atom - perturber atom collisions are viewed to occur instantaneously , i.e. , the duration @xmath56 of the typical collision is assumed to be much less than the various time scales in the problem ( with the exception of the optical period @xmath57 , where @xmath58 is a transition frequency ) . in particular , the assumption that @xmath56 is much lesser than the time @xmath59 between collisions is called the _ binary collision approximation_. in our example of metastable @xmath1he , the buffer or perturber atoms are the @xmath1he atoms in the ground ( 1@xmath34s@xmath35 ) state . typically , @xmath60 s at 1.0 torr of buffer gas pressure and @xmath61 s. the binary collision approximation is valid easily up to a gas pressure of 500 torr . the implications of the impact approximation are easily understood . each collision produces a change in the density matrix @xmath62 associated with active atoms . between collisions , the external field produces a time rate of change for the density matrix . since changes produced by the external fields during collision are assumed to be negligible , the impact approximation allows one to represent the contributions to @xmath63 as arising independently from the collisions and external radiation fields . in this way , one obtains a ` master ' or transport equation for @xmath62 @xcite . the influence of vcc on eit is formally treated by writing the density matrix equations for each velocity class and introducing in those equations the probability for vcc between different classes @xcite . the contribution of vccs to the density - matrix equations for the populations and coherences , in general , is @xmath64_{\mathrm{vcc } } = - \gamma_{ij\mathrm{vcc } } \rho_{ij}(v ) + \int k_{ij}(v ' \rightarrow v ) \rho_{ij}(v ' ) dv ' . \label{vccgen}\ ] ] here the vcc process is described in terms of the collisional relaxation rates @xmath65 , and the associated collisional kernels are @xmath66 . the collision kernel @xmath66 [ of the dimension of inverse length ] gives the probability density per unit time that a collision changes the velocity of an active atom in state @xmath67 from @xmath68 to @xmath69 . changes in @xmath69 occur at some average rate @xmath65 , which is related to the kernel in the following way @xcite : @xmath70 the first term on the right - hand side in eq.([vccgen ] ) can be viewed as the out term " resulting from collisions that remove active atoms in state @xmath67 from the velocity subclass @xmath69 , and the last term is the in term " bringing atoms from other velocity subclasses into the subclass @xmath69 . @xmath71 is related to the differential scattering cross - section , and @xmath72 is related to the corresponding total scattering cross - section . since the collision interaction depends on the internal atomic states of the atoms , a complete description requires a separate collision kernel for each atomic state population and each coherence . in general , an excited atom has a slightly larger collisional cross - section than a ground - state atom because excited atoms are bigger than the ground - state atoms , and therefore the excited atoms suffer a stronger collisional damping of speeds . the opposite holds true for the case of active alkali atoms colliding with neon where an excited atom has a slightly lower collisional cross - section than a ground - state atom . however , in a low - pressure regime where the decay rate @xmath73 ( @xmath74 10@xmath75 s@xmath76 ) at the optical transition ( @xmath77 or @xmath78 ) is larger than the collision rate @xmath79 ( @xmath74 10@xmath80 ) , the transport of coherence in the optical transition from one velocity group to another is of not much importance . for the lower - state coherence , the situation is different , as the effective lower state relaxation ( @xmath74 10@xmath1 s@xmath76 ) is smaller than the collision rate . we can thus assume that the collision kernel is zero for the optical coherences , is approximately the same for all populations , and is different for the lower - state coherence : @xmath81 thus , for the three - level @xmath0 system , the added vcc contributions are : @xmath82_{\mathrm{vcc } } & = & - \gamma_{0\mathrm{vcc } } \rho_{ii}(v ) + \int k_0(v ' \rightarrow v ) \rho_{ii}(v')~ dv ' , ~~(i = a , b , c ) , \label{eq8 } \\ \frac{d}{dt } [ \tilde{\rho}_{cb}(v)]_{\mathrm{vcc } } & = & - \gamma_{1\mathrm{vcc } } \tilde{\rho}_{cb}(v ) + \int k_1(v ' \rightarrow v ) \tilde{\rho}_{cb}(v ' ) dv ' , \label{eq9}\end{aligned}\ ] ] where @xmath83 , @xmath84 , and @xmath85 . here we have further neglected the velocity - dependence of the collision rate @xmath79 because it is usually a slowly - varying function of velocity . we first convert the usual density matrix elements @xmath86 to slowly - varying variables @xmath87 in order to remove the fast optical oscillations by using the following transformations : @xmath88 taking into account the various relaxations mentioned above , including the dynamic atomic influx into the beam , and incorporating the effect of vccs expressed in eqs.([eq8])-([eq9 ] ) , the evolution of the slowly - varying density matrix elements @xmath89 for atoms with velocity @xmath69 may be written under the rotating - wave approximation as : @xmath90 \nonumber \\ & & - i\frac{\omega_{\mathrm{c}}}{2 } [ \tilde{\rho}_{ac}(v ) - \tilde{\rho}_{ca}(v ) ] + \int k_0(v ' \rightarrow v ) \rho_{aa}(v')~ dv ' , \label{eq14 } \\ \frac{d \rho_{bb}(v)}{dt } & = & \frac{\gamma_0}{2 } \rho_{aa}(v ) - \left ( \gamma_{\mathrm{t } } + \gamma_{0\mathrm{vcc } } \right ) \rho_{bb}(v ) + \frac{\gamma_{\mathrm{t}}}{2 } \left [ w(v ) + \beta \left ( \rho_{bb}(v ) - \rho_{cc}(v ) \right ) \right ] \\ & & + i\frac{\omega_{\mathrm{p}}}{2 } [ \tilde{\rho}_{ab}(v ) - \tilde{\rho}_{ba}(v ) ] \nonumber + \int k_0(v ' \rightarrow v ) \rho_{bb}(v')~ dv ' , \label{eq15 } \\ \frac{d \rho_{cc}(v)}{dt } & = & \frac{\gamma_0}{2 } \rho_{aa}(v ) - \left ( \gamma_{\mathrm{t } } + \gamma_{0\mathrm{vcc } } \right ) \rho_{cc}(v ) + \frac{\gamma_{\mathrm{t}}}{2 } \left [ w(v ) - \beta \left ( \rho_{bb}(v ) - \rho_{cc}(v ) \right ) \right ] \\ & & + i\frac{\omega_{\mathrm{c}}}{2 } [ \tilde{\rho}_{ac}(v ) - \tilde{\rho}_{ca}(v ) ] \nonumber + \int k_0(v ' \rightarrow v ) \rho_{cc}(v')~ dv ' , \label{eq16 } \\ \frac{d \tilde{\rho}_{ab}(v)}{dt } & = & - \left [ \frac{\gamma}{2 } - i ( \delta_{\mathrm{p } } - k v ) \right ] \tilde{\rho}_{ab}(v ) + i\frac{\omega_{\mathrm{c}}}{2 } \tilde{\rho}_{cb}(v ) - i\frac{\omega_{\mathrm{p}}}{2 } [ \rho_{aa}(v ) - \rho_{bb}(v ) ] , \label{eq17 } \\ \frac{d \tilde{\rho}_{ca}(v)}{dt } & = & - \left [ \frac{\gamma}{2 } + i ( \delta_{\mathrm{c } } - k v ) \right ] \tilde{\rho}_{ca}(v ) - i\frac{\omega_{\mathrm{p}}}{2 } \tilde{\rho}_{cb}(v ) + i\frac{\omega_{\mathrm{c}}}{2 } [ \rho_{aa}(v ) - \rho_{cc}(v ) ] , \label{eq18 } \\ \frac{d \tilde{\rho}_{cb}(v)}{dt } & = & - \left [ \gamma_{\mathrm{r } } + \gamma_{1\mathrm{vcc } } - i \delta_{\mathrm{r } } \right ] \tilde{\rho}_{cb}(v ) - i\frac{\omega_{\mathrm{p}}}{2 } \tilde{\rho}_{ca}(v ) + i\frac{\omega_{\mathrm{c}}}{2 } \tilde{\rho}_{ab}(v ) \nonumber \\ & & + \int k_1(v ' \rightarrow v ) \tilde{\rho}_{cb}(v')~ dv ' . \label{eq19}\end{aligned}\ ] ] note that the _ total _ atomic population in state @xmath67 at a time @xmath47 is given by @xmath91 and @xmath92 for a closed system . @xmath89s have the dimensions of inverse speed . equations ( [ eq14])-([eq19 ] ) are to be solved for @xmath93 to get the susceptibility at @xmath17 . these integro - differential equations can be solved using iterative techniques which may be taken up in future . the solution can be worked out for various limiting forms of the collision kernel , and here we follow the strong collision model , in which @xmath94 is a jump process . the effect of collisions is ` strong ' , i.e. , it washes out the memory of the pre - collision value of the velocity . a single collision , on average , thermalizes the velocity distribution . the rate of collisions is taken as an average rate given by the inverse of the mean free - time between collisions . the collision kernel is then greatly simplified and can be approximated by @xmath95 where @xmath96 is the maxwell - boltzmann distribution for the velocity vector in one direction , given by @xmath97 where @xmath98 is the most probable speed : @xmath99 @xmath100 being the mass of an atom . for a temperature of 300 k , the most probable speed of he atoms is about 1100 ms@xmath76 . in the presence of light of wave - number @xmath101 , the 1/@xmath102 doppler half - width is @xmath103 . for our system , with a laser at a frequency @xmath104 = 1.74 @xmath105 10@xmath106 rad / s ( or wavelength @xmath107 = 1.083 @xmath48 m ) , the doppler half - width at half maximum ( hwhm ) , @xmath108 = 0.9 ghz . note that under the assumption that the lhs of eq.([eq11 ] ) is independent of the initial velocity @xmath68 , the rhs is the only allowed form , consistent with the detailed balance of transitions @xmath109 and the conservation of probability ( [ eq7 ] ) . the excitation by a single - mode laser is velocity - selective . in the absence of vccs , optical pumping with a single - mode laser polarizes only a small portion of the thermal velocity distribution . in strong thermalizing vccs , the root - mean - square velocity change @xmath110 is much larger than the width of the resonant velocity bin " ( @xmath111 , with @xmath73 the homogeneous linewidth ) . to ensure rapid thermalization to access the entire velocity profile , the number of vccs occurring during the lower - state orientation relaxation time @xmath112 must be large compared to the total number of velocity bins ( @xmath113 ) @xcite . also , when the photon absorption rate @xmath114 is large compared to the rate of diffusion of atoms across the laser beam ( @xmath115 ) but small compared to the rate of vccs ( @xmath116 ) , the redistribution of atoms in the different velocity classes rebuild the maxwell - boltzmann velocity distribution ( i.e. , vccs thermalize the lower - state velocity distributions rapidly compared to an absorption - emission cycle ) . this leads to a velocity - independent type of optical pumping @xcite . since atoms jump from one velocity class to another , when the number of vccs occurring per cycle is much greater than the number of pump photons , the pumping spreads over the entire doppler distribution in each optical pumping cycle . the laser is depleting a single velocity bin at a rate @xmath114 , but since the rate @xmath79 at which the vccs are replenishing it with atoms from the entire doppler distribution is so much faster , the velocity distribution stays thermalized even during the pumping process . under the above conditions and considering the broadening due to the thermal velocity distribution of the atoms , we re - write the density matrix equations @xcite with @xmath117 here @xmath118s are dimensionless , and @xmath119 , for a closed system without dissipation . the simplified assumption ( [ thermal ] ) would imply that the effect of the vccs is to bring about a complete redistribution of the population over all the velocity classes such that the inhomogeneous media is similar to a homogeneous one but with a width given by the inhomogeneous doppler - broadened width . hence , the problem effectively reduces to that of a homogeneous system with a doppler - broadened pumping rate . however , vccs affect populations and coherences in a slightly different manner . though there is a redistribution of the atomic population over the entire inhomogeneous width , vccs can still lead to a decoherence of the raman coherence . in order to take account of any depolarization because of collisions that leads to a decoherence of the prepared eit medium , we additionally modify the collision kernel in eq.([eq11 ] ) for the lower - state coherence @xcite with strong thermalizing vccs by inserting a depolarization ratio @xmath120 as @xmath121 with @xmath122 from eq.([eq7 ] ) . we define the parameter @xmath123 as a deviation from complete coherence preservation . @xmath124 would imply a loss of coherence by vccs , and hence a loss of coherently - prepared , dark - state atoms from the maxwell - boltzmann distributed system . @xmath125 corresponds to a complete redistribution of population over the doppler width by vcc , without any loss of polarization . the vcc contribution gets added to the transit time decay ( and hence the raman coherence relaxation ) rate as @xmath126 ( with @xmath127 0 ) . the populations in the lower levels are aided by the influx of fresh atoms at the dynamic transit rate ( a part of which has no contribution from the vcc ) , yielding the inhomogeneity in the density matrix equations in the steady state , necessary for non - trivial solutions . the steady - state solutions of eqs.([eq14])-([eq19 ] ) with particular combinations such as @xmath128 , @xmath129 and @xmath130 are considered . using eqs . ( from the above set of equations using these doppler - broadened rates , we finally obtain the steady - state value of @xmath136 which yields @xmath31 . @xmath137 & = & -\frac{\gamma_{\mathrm{x3 } } ( \delta + \delta_{\mathrm{r}})}{\omega_{\mathrm{c } } } \mbox{im } [ r_{cb } ] - \frac{\gamma_{\mathrm{y3 } } ( \delta + \delta_{\mathrm{r}})}{\omega_{\mathrm{c } } } \mbox{re } [ r_{cb } ] + \frac{\gamma_{\mathrm{y4 } } ( \delta + \delta_{\mathrm{r}})}{\omega_{\mathrm{p } } } ( r_{aa } - r_{bb } ) , \\ \mbox{im } [ r_{ab } ] & = & \frac{\gamma_{\mathrm{x3 } } ( \delta + \delta_{\mathrm{r}})}{\omega_{\mathrm{c } } } \mbox{re } [ r_{cb } ] - \frac{\gamma_{\mathrm{y3 } } ( \delta + \delta_{\mathrm{r}})}{\omega_{\mathrm{c } } } \mbox{im } [ r_{cb } ] - \frac{\gamma_{\mathrm{x4 } } ( \delta + \delta_{\mathrm{r}})}{\omega_{\mathrm{p } } } ( r_{aa } - r_{bb}),\end{aligned}\ ] ] where @xmath138 = \frac{r_{cb } + r_{bc}}{2}$ ] and @xmath139 = \frac{r_{cb } - r_{bc}}{2}$ ] . the susceptibility for the doppler - broadened medium is given as : @xmath140 where @xmath141 is the normalization constant , which is obtainable from the initial probe transmission at resonance in the absence of the coupling field . we now test the validity of our theory by comparing our predictions against known experimental results in hot vapors , viz . in a he * cell @xcite . for he * , collisions are expected to play a favorable role through five different effects : ( i ) vccs enable one to span the entire doppler profile quickly and efficiently during optically pumping ; ( ii ) collisions aid in the feeding distortion in pumping back more atoms in @xmath13 compared to that in @xmath15 when they are entering the beam ; ( iii ) collisions increase the transit time of the atoms through the beam and hence the raman coherence lifetime @xcite ; and ( iv ) this is possible because collisions involving he atoms in the zero spin and angular momentum ground state do not depolarize the colliding he * @xcite ; and ( v ) penning ionization ( pi ) among identically polarized he * atoms is almost forbidden @xcite . we probe the general dependence of various features of interest doppler - averaged fano - like transmission profiles , the variation of the eit width , the peak transmission and the group delay with the coupling intensity on the following system characteristics : the vcc parameter @xmath142 , the unequal atomic influx parameter @xmath55 , the raman decoherence rate @xmath11 , and the initial transmission @xmath143 which carries the information about the number density of the participating atoms . in our model , the effect of collisions is incorporated through collisional dephasing of both the optical and raman coherences given in eqs.([optcoh ] ) and ( [ ramancoh ] ) , the collision - induced diffusive ( as opposed to ballistic ) transit rate @xmath33 of the atoms , a complete redistribution of the atomic population over all velocity classes leading to a velocity - independent optical pumping under rapid vcc coverage , and the vcc decoherence rate @xmath144 entering when the vcc parameter @xmath145 . for the he * system , we find that the decoherence due to vcc is very small . as @xmath142 is a deviation from the complete polarization - conservation by vcc , it should not depend on the beam size . we choose a value of @xmath146 , which helps us to obtain good fits to the measured evolutions with the coupling intensity for each of eit width , peak transmission and group delay , for all the measured beam diameters , with @xmath147 , which is simply the number of collisions per second as mentioned earlier . the excess incoming rate , @xmath148 to state @xmath13 over @xmath15 also has a distinct impact on the various features . in particular , the parameter @xmath55 = 0.1 gives satisfactory results in comparing the theoretical results with the experimentally measured evolutions . with the precisions of our measurements in he * , a constant value of @xmath55 is found to fit well for all beam diameters . the choice of the best - fit values of the parameters @xmath142 and @xmath55 for the he * system has been elaborated using the group delay plots in subsection d below . there was a slight over - estimation of the initial transmission @xmath143 for the eit experiments in he * , reported in ref.@xcite directly from the measured values , which included an effect of partial saturation by the probe power . this has been adjusted in the present paper . when the coupling beam frequency is no longer at the center of the doppler profile , transmitted intensity profiles become asymmetric . this is similar to the fano profiles obtained in the case of eit in a homogeneously - broadened medium and which have been shown to be due to interferences between a direct process and stimulated raman scattering in the overall transition probability @xcite . however , here , these profiles are modified by the fact that they have to be convoluted with the inhomogeneous doppler profile . figure [ fig2 ] shows the evolution of the transmission versus raman detuning @xmath25 for different values of the detuning @xmath149 of the coupling beam with respect to the center of the doppler profile . on the left - hand side , the measured values through a he * cell with initial transmission at resonance @xmath143 = 0.56 for a beam diameter of 1.5 cm , a coupling power of 11 mw and a probe power of 140 @xmath48w are reproduced from ref.@xcite . the base of the symmetric ( black ) curve at resonance corresponds to a transmission higher than 0.56 , because of partial saturation of the medium by the probe intensity . theoretically , the transmission profiles are generated from the imaginary part of the susceptibility at different detunings @xmath149 of the coupling beam with respect to the center of the doppler profile mentioned above . the corresponding plots on the right - hand side of fig.[fig2 ] are obtained for a beam diameter of 1.5 cm corresponding to different values of @xmath149 , with the same initial transmission , coupling and probe powers as in the experiment , using @xmath150 = 4.3 khz , @xmath151 = 0.41 khz , @xmath146 , and @xmath55 = 0.1 . the transmission profiles at all values of @xmath149 increases slightly with an increase in @xmath55 . this is also seen in the behavior of peak transmission shown in subsection c. this has been checked for a range of @xmath55 from 0 to 0.1 . there is a distinct and sensitive dependence on @xmath142 , as the tip of each profile decreases slightly with an increase of @xmath142 , and the effect is the largest on the resonant profile . as a result , there is also a change in the relative placement of the profiles . this has been checked for a range of values of @xmath142 from 10@xmath152 to 0 . the reason for this is simple . since a deviation of the value of @xmath142 from 0 indicates a loss of coherence , the transmission at resonance will be greater for @xmath142 closer to 0 , because the system is more coherent as @xmath142 approaches 0 . additionally , the effect will be more pronounced when the atomic system is in resonance with the two fields , which is the condition for a perfect eit . we thus find that the predictions from our model indeed agree very well with the experimental results . the mechanism of decoherence in eit in a room - temperature gas can be probed by measuring the width of the eit resonance as a function of the coupling beam intensity . as mentioned in the introduction , the theoretical treatment of eit in doppler - broadened gases @xcite , assuming the population exchange between the lower levels to be the main source of decoherence , predicts a _ nonlinear _ dependence for weak coupling powers : the eit width is expected to evolve with the coupling beam rabi frequency @xmath3 @xcite according to : @xmath153 where @xmath154 gives the effective width over which the atoms are pumped into the probe ground state for a fixed value of @xmath3 . in the case when @xmath155 , @xmath154 is dependent on @xmath3 and @xmath11 : @xmath156 , and the coherent population trapping is shown to be velocity selective , i.e. , it occurs only for those atoms whose frequencies are close to resonance with the coupling field . in the opposite regime , when @xmath157 , ref.@xcite predicts that @xmath158 , i.e. , in the limit of very large intensity , it is reduced to the power - broadening case . in contrast , in ref.@xcite it is assumed that all the atoms across the doppler profile are initially optically pumped by the coupling beam to the probe ground level , and the decoherence in the lower states is caused by pure dephasing , and not population exchange . then the response of the medium up to first order in probe field yields the following _ linear _ dependence of the eit linewidth on the coupling beam intensity : @xmath159 an example of the evolution of the eit width versus the intensity of the coupling beam is displayed in fig.[fig3](a ) . in the experiment with he * @xcite , the logarithm of the transmitted intensity is calculated and its full width at half maximum ( fwhm ) is then measured , in order to determine precisely the width of the susceptibility @xmath160 of the medium . the measured sub - natural ( @xmath161 ) widths for a beam diameter of 1.5 cm , a probe power of 100 @xmath48w and an initial transmission @xmath143 = 0.46 are shown as dots . theoretically , the imaginary part of the susceptibility is fitted with a lorentzian to obtain the fwhm corresponding to a particular coupling intensity . the continuous line in fig.[fig3 ] is the best fit from our model with the same parameters as in the experiment , using @xmath162 = 4.3 khz , @xmath163 = 0.41 khz , @xmath146 , and @xmath55 = 0.1 . the eit width @xmath10 is seen to evolve linearly with the coupling beam intensity , i.e. , quadratically with the coupling beam rabi frequency @xmath3 . with the experimental parameters ( @xmath164 at 1 torr , @xmath165 , @xmath166 ghz ) @xcite , one obtains @xmath167 . since the maximum rabi frequencies @xmath3 used in the experiment are smaller than @xmath168 , the measurements are in the first regime of @xcite , where @xmath169 . thus @xmath10 is predicted to evolve linearly with @xmath3 @xcite , with a slope depending on @xmath11 , but these predictions are violated in the experiment . if we use eq.([javan ] ) to fit the linear evolution of @xmath10 versus the coupling intensity , we obtain @xmath170 , which is of the same order of magnitude as @xmath171 , showing that a major part of the doppler profile takes part in the eit process . in our model we achieve the straight line feature of @xmath10 versus coupling intensity as in ref.@xcite , but without any assumptions of special initial conditions . moreover , if the data are fitted to the straight line given by eq.([lvovsky ] ) , the resulting estimate of @xmath150 does not reproduce accurately the measured evolutions of the peak - transmission and delay @xcite . our present model fits the straight line data for atomic @xmath172rb vapor at temperatures 60 - 100@xmath173c given in ref.@xcite , with appropriate values of the parameters for the optically thick system . keeping all other parameters constant , a change in the initial transmission @xmath143 , and hence in the number density of the participating atoms in the cell , does not affect the evolution of the eit width with coupling intensity in our model . likewise , a change in the unequal feeding parameter @xmath55 does not produce any visible change in the eit width . this has been checked for a range of values of @xmath174 0 to 0.5 . this is understandable as these two parameters do not affect the raman coherence lifetime but affect the pumping efficiency ( e.g. the fraction of atoms that participate in the eit phenomenon ) . figure [ fig3](b ) shows the variation of the eit width with the coupling intensity from our model for different values of the raman decoherence rate @xmath11 around the best - fit value , keeping @xmath142 , @xmath55 and @xmath143 constant . it clearly shows that the slope of the evolution is the same for different values of @xmath11 . note that different laser beam diameters would lead to different transit times of the atoms through the beam , and hence different @xmath33 . the motion of the atoms through the beam is assumed to be diffusive , as stated previously , because of the large number of collisions they undergo . the raman decoherence rate @xmath11 given by eq.(15 ) thus contains a variable @xmath33 plus other terms . for the he * measurements , we estimate that @xmath175 4 khz for a low pressure of 1 torr and an ambient magnetic field inhomogeneity . this is obtained by estimating first @xmath33 for a particular beam diameter , and then @xmath11 from the best fit of ( primarily ) the eit - width data . in the experiment , the beam diameters are generally not precisely known , because the beam emerging from the acousto - optic modulators is not perfectly gaussian . this affects the precise determination of the transit times as well as beam intensities ( the probe and coupling powers measured before the he * cell were , of course , their average values ) and rabi frequencies . we have , however , checked that the experimental data for different beam sizes yield the @xmath33 values following the diffusive transit scenario . figure [ fig3](c ) shows the variation for different values of the vcc parameter @xmath142 . the slope of the evolution is the same for different values of @xmath120 but the intercept increases with an decrease of @xmath142 . from figs.[fig4](b ) and ( c ) , the width - intercept at @xmath3 = 0 are seen to depend on the parameters @xmath11 and @xmath142 , yielding a narrower resonance for a lower @xmath11 or a lower @xmath142 . the inset shows that the measured data fit better with @xmath145 and @xmath176 . however , the lines do not fit eq.([lvovsky ] ) , with @xmath11 of ref.@xcite simply replaced here by @xmath177 . as pointed out earlier , in the density - matrix equations , the vcc contribution gets added to the transit time decay ( and hence the raman coherence relaxation ) rate as @xmath178 , except in the inhomogeneous term signifying atomic influx in the populations in the lower levels at the transit rate @xmath33 which has no contribution from the vcc . it is therefore not surprising that the net effect of @xmath142 can not be taken care of by @xmath33 or @xmath11 . figure [ fig3](d ) shows the variation of the eit width with the coupling intensity from our model for different values of the optical coherence decay rate @xmath73 . the slope of the line changes inversely with changes in the values of @xmath4 and @xmath73 . we next consider the maximum probe transmission at resonance ( @xmath149 = 0 ) corresponding to different coupling intensities . figure [ fig4](a ) shows the evolution of the cell transmission versus coupling beam intensity . the measured values ( dots ) through a he * cell for a beam diameter of 1.5 cm are reproduced from ref.@xcite . the corresponding continuous curve is obtained from our theory by calculating the transmission at two - photon resonance @xmath179 , with @xmath146 , @xmath55 = 0.1 at the same initial transmission and probe power as in the experiment : probe power of 70 @xmath48w , @xmath143 = 0.46 , @xmath150 = 4.3 khz , and @xmath163 = 0.41 khz . using the same hypotheses as for the derivation of eq.([lvovsky ] ) @xcite , the peak transmission is predicted to evolve as @xcite : @xmath180 it is seen that there is a better agreement of the experimental result with our present theory than what was obtained using eq.([lvovsky_trans ] ) in ref.@xcite with the value of @xmath11 deduced from the eit - width data , in spite of the fact that we do not assume @xmath181 = 1 , i.e. , a system perfectly prepared by optical pumping at @xmath47 = 0 as in ref.@xcite , is redundant here in the presence of rapid vcc coverage . figure [ fig4](a ) also shows the variation of the peak transmission with the coupling intensity from our model for different values of the vcc parameter @xmath142 . the dotted and the dashed plots correspond to @xmath142 = 0 and @xmath182 , respectively . it can be thus inferred that the peak transmission increases with a decrease in @xmath142 . this also has support from the effect of @xmath142 on the doppler - broadened profiles in fig.[fig2 ] for the same reason . the best - looking fit for @xmath146 emphasizes that the depolarization due to vccs is a small effect , as expected from the work of shlyapnikov _ et al . _ @xcite . figure [ fig4](b ) shows the variation of the peak transmission with the coupling intensity from our model for different values of the raman decoherence rate @xmath11 around the best - fit value , when @xmath142 , @xmath55 and @xmath143 are kept constant @xcite . it is clear that a lower decoherence rate @xmath11 leads to a more coherent system , and thus @xmath150 = 3.2 khz ( dashed curve ) yields the highest peak transmission compared to the rest shown in this plot . thus the evolutions from our model are consistent with our physical understanding . also note that figs.[fig4](a ) and ( b ) show that an increase in @xmath11 has a similar effect as a decrease in @xmath142 . figure [ fig4](c ) shows the same for different values of the unequal feeding parameter @xmath55 from the theory . the effect of @xmath55 here is clearly very distinct . as @xmath55 increases , there are more atoms entering the laser beam in the dark state @xmath13 . thus , the peak transmission at two - photon resonance will also increase as the transparency is more when more atoms can participate in the eit phenomenon . figure [ fig4](d ) shows the effect of different values of the initial transmission @xmath143 . it is obvious that a higher initial transmission @xmath143 would lead to a higher peak transmission for non - zero coupling intensities . figure [ fig5](a ) shows the measured evolution of the group delay ( through a he * cell of length 2.5 cm ) versus coupling beam intensity for a beam diameter of 1.5 cm ( dots ) , reproduced from ref.@xcite . the results have been obtained with gaussian probe pulses of duration of 70 @xmath48s with a peak power of 35 @xmath48w and with the coupling and probe beam frequencies at the center of the doppler profile ( @xmath183 ) . in the slow - light experiments performed with he * , reported in ref.@xcite , the measured delays for non - zero coupling intensities were slightly overestimated as the reference used was the probe pulse in the absence of a coupling beam . theoretically , the group delay is calculated from the derivative of the real part of the susceptibility with respect to frequency at two - photon resonance , for the experimentally used parameters . the continuous curve in fig.[fig5](a ) is the best fit obtained from our model with @xmath150 = 4.3 khz , @xmath163 = 0.41 khz , @xmath146 and @xmath55 = 0.1 , for the same probe power of 35 @xmath48w and the same initial transmission @xmath143 = 0.46 as in the experiment . the group velocity derived @xcite from the susceptibility of ref.@xcite leads to a group delay at the line center ( @xmath183 ) given by : @xmath184 ^ 2}\;. \label{lvovsky_delay}\ ] ] the maximum value of the group delay is reached for @xmath185 and its value is @xmath186 . again , there is a better agreement of the experimental results with our present theory than what was obtained in ref.@xcite using eq.([lvovsky_delay ] ) , with the value of @xmath11 deduced from the eit - width data . figure [ fig5](a ) also shows the comparison of the group delay profiles from our theory for different values of @xmath187 ( dashed curve ) and 0 ( dotted curve ) around the best - fit value corresponding to @xmath146 ( continuous curve ) , for the 1.5 cm - diameter beam . the magnified part shown in the inset justifies the choice of the vcc parameter value of @xmath146 for the measured data for he * , which is distinct from that of @xmath125 . as mentioned earlier , the measured data were over - estimated and a good fit running slightly below the data points would require a sensitive adjustment of @xmath188 ( as shown ) . figure [ fig5](b ) shows the comparison of the group delay profiles from the theory for different values of @xmath11 around the best - fit value , for the 1.5 cm - diameter beam . @xmath11 indeed has a great impact on the delay which is clear because lower the value of @xmath11 ( for example , the dashed curve for @xmath150 = 3.2 khz ) , more is the coherence , hence higher are the delays achieved in the system . figure [ fig5](c ) shows the same as above for different values of @xmath55 around the best - fit value , for a range of @xmath55 from 0 to 0.5 . we see that with an increase of @xmath55 , the delays at different coupling intensities decrease . on comparison with the plots of peak transmission for different values of @xmath55 in fig.[fig4](c ) , one deduces that the effect of @xmath55 on these two characteristics are opposite . we can understand this as follows . if an increase of @xmath55 increases the peak transmission ( as seen and explained above ) , then from the relationship of resonant peak transmission and delay , it becomes clear that an increased transmission would lead to a decrease in the group delay . this is also manifested in eqs.([lvovsky_trans ] ) and ( [ lvovsky_delay ] ) . the inset at a magnified scale shows the comparison of the plots for @xmath189 and @xmath190 with reference to the measured data points it is clear that @xmath190 provides a distinctly better fit than @xmath189 . figure [ fig5](d ) shows the comparison of the group delay profiles from the theory for different values of @xmath143 around the best - fit value , for the 1.5 cm - diameter beam . the group delay increases with a decrease in @xmath143 and the effect is also supported by eq.([lvovsky_delay ] ) . a decrease in the initial transmission in the absence of the coupling field signifies a proportionate increase in the number of participating atoms . @xmath143 , however , does not affect the eit width . thus the delay - bandwidth product , which is a figure of merit for a delay / storage medium , turns out to be proportional to the number density of the medium . in comparison to the width and the peak transmission plots shown in figs.[fig3 ] and [ fig4 ] respectively , we note that @xmath11 , @xmath142 , @xmath55 and @xmath143 have a greater impact on the delay as seen from fig.[fig5 ] , especially in determining the peak of the delay curve . we have presented a realistic analysis of eit and slow light in a hot atomic vapor , taking into account all the relevant decoherence processes , for arbitrary strengths of the probe and the coupling fields , and without assuming any special initial condition . we have considered the influx of fresh atoms in the lower levels and the outflux from all the levels at a transit rate in the gas . unlike the theory for eit in a doppler - broadened medium in ref.@xcite in which the role of collisions is completely neglected , our analysis includes phase - interrupting collisions of active atoms as well as velocity - changing collisions , modeled effectively in the strong collision limit . we have demonstrated the role of vccs in redistributing the atomic population over all velocity classes and hence a velocity - independent optical pumping . the initial condition of @xmath191 = 1 , i.e. , a system perfectly prepared by optical pumping at @xmath47 = 0 in ref.@xcite , is redundant here in the presence of rapid vcc coverage . the steady - state solutions for the atomic density matrix elements are presented in the strong - collision approximation with a model for the vccs under rapid vcc coverage . a value of the vcc parameter @xmath192 indicates a loss of coherence by vccs . as observed from the results , all the eit characteristics have a sensitive dependence on @xmath142 . in our example of the he * system @xcite , it was found that a small value of @xmath146 gives the best fits for all the measured characteristics for different beam sizes . for he * , the motion of the atoms through the beam is assumed to be diffusive , because of the large number of collisions they undergo . further , because of diffusion and favourable collisions outside the interaction region , we have allowed a slightly greater fraction of atoms to enter the beam prepared in state @xmath13 than those in state @xmath15 using a parameter @xmath55 . the best fit value of @xmath193 supports this fact . for all systems in which other decoherence effects are small so that atoms can diffuse out of the interaction region and return before decohering , the unequal feedback fraction would model such a positive contribution to the coherence , similar in effect to a decrease in the number density or an increase in the initial transmission @xmath143 . a constant value of @xmath55 is found to fit different beam sizes , given the precision of the reported measurements . for non - zero detunings @xmath149 of the coupling field from the center of the doppler - broadened transition frequency , the transmitted intensity profiles become asymmetric about the two - photon resonance ( raman detuning @xmath25 = 0 ) . this fano - like feature is a signature of the two - photon process of eit , and emerges naturally from our model . the eit width @xmath10 , simulated from our model , shows a linear dependence on the coupling beam intensity , i.e. , a quadratic dependence on the coupling beam rabi frequency @xmath3 , as observed in experiments . the evolutions of the peak transmission and the group delay with the coupling beam intensity predicted from our analysis faithfully reproduce the experimentally observed behaviors . in the evolution of all these features of interest , an increase in the raman decoherence rate @xmath11 seems in general to have a similar effect as an increase in the vcc parameter @xmath142 . but whatever the beam diameter , @xmath142 should remain constant , while @xmath33 and hence @xmath11 should decrease when the beam diameter increases ( keeping @xmath194 constant ) . eit has recently found applications in white - light cavities @xcite for use in gravitational wave detection . for applications of slowing of light for use in quantum - information processing , and in particular , in quantum memory , the medium needs to be optically dense . our model is applicable to all such systems of hot atomic vapor , which are attractive candidates for practical applications requiring large - bandwidth controllable delays , for example , for signal processing at the appropriate wavelength . the system can be generalized to model tripod - like systems @xcite in hot vapor . this work is supported by an indo - french networking project funded by the department of science and technology , government of india , and the french ministry of foreign affairs . the work was initiated during a stay of rg in france , supported also by `` cnano ile - de - france '' and `` triangle de la physique '' . the authors wish to thank e. arimondo , h. gilles , d. kumar , m. leduc , p. j. nacher and g. tastevin for useful discussions . the work of jg is supported by the council of scientific and industrial research , india . the school of physical sciences , jawaharlal nehru university , is supported by the university grants commission , india , under a departmental research support scheme . s. knappe , r. wynands , j. kitching , h. g. robinson and l. hollberg , j. opt . b * 18 * , 1545 ( 2001 ) ; j. vanier , a. godone , and f. levi , phys . a * 58 * , 2345 ( 1998 ) ; m. erhard and h. helm , phys . a * 63 * , 043813 ( 2001 ) ; i. novikova , d. f. phillips , a. s. zibrov , r. l. walsworth , a. v. taichenachev and v. i. yudin , opt . lett . * 31 * , 2353 ( 2006 ) . in both refs.@xcite and @xcite , in order to avoid the complexity in the integration with a maxwell - boltzmann velocity distribution , the usual gaussian distribution has been approximated with a lorentzian function , leading to small deviations . the interesting effect of diffusion - induced ramsey narrowing " has the same origin ; see , for example , y. xiao , i. novikova , d. f. phillips and r. l. walsworth , phys . * 96 * , 043601 ( 2006 ) . it occurs due to the random distribution of the durations spent by the atom outside the interaction region . instead of treating the atomic motion in a separate diffusion equation , the effect here has been easily incorporated in the parameter @xmath55 . e. arimondo , phys . a * 54 * , 2216 ( 1996 ) . note that in this paper the collisional relaxation rate and kernel for populations in eq.(15b ) should have the superscript @xmath195 , and not @xmath196 . in deriving eqs.(16 ) , it has been assumed that @xmath197 = 0 , and @xmath198 in the cpt system . also , the maxwell distribution @xmath96 in eq.(18 ) has a mistake : the @xmath101 in the denominator should be deleted . the distribution @xmath96 is then the usual normalized maxwell distribution with the dimensions of inverse speed . the dimensionality of eq.(17 ) is now balanced . @xmath199s in eq.(11 ) are dimensionless . then @xmath200 introduced in eq.(21 ) has the dimensions of speed . but the @xmath200s given in the solution ( 25 ) are dimensionless . in deriving the dynamical eqs.(16 ) and ( 22 ) , the steady - state values of the optical coherences have been used . a. v. phelps , phys . rev . * 99 * , 1307 ( 1955 ) ; m. g. payne _ et al_. , phys lett . * 35 * , 1154 ( 1975 ) ; m. pinard and f. lalo , j. phys . france * 41 * , 799 ( 1980 ) ; b. zygelman , phys . a * 43 * , 575 ( 1991 ) . other than the anomaly related to the actual beam size inside the cell as mentioned in the text , note that the three sets of data for the eit width and the two sets of data for peak transmission in ref.@xcite were generated under different experimental conditions , notably the initial transmission @xmath143 , signifying differences in the number density of the participating atoms in the he * discharge cell . likewise , the transit time of the atoms in the beam , influenced by collisions , and the collisional dephasing @xmath201 , would be different for these different measurements . the best fits from our theory are consistent with the logical adjustments of both @xmath143 and @xmath11 for these different sets of data . note that in the calculation of the control field rabi frequency @xmath3 in ref.@xcite , the oscillator strength for the transition has been erroneously taken as the total @xmath202 instead of @xmath203 ( in case of degenerate zeeman sublevels , when a circularly - polarized light probes the transition along any one arm of the @xmath0-system , the oscillator strength is half of the total @xmath204 . ) . it has resulted in a scaling of @xmath3 , which has been corrected here . this has no bearing on the conclusions drawn in ref.@xcite .

the possible occurrence of equilibrium off - equatorial tori in the gravitational and electromagnetic fields of astrophysical compact objects has been recently proved based on non - ideal mhd theory . these stationary structures can represent plausible candidates for the modelling of coronal plasmas expected to arise in association with accretion discs . however , accretion disc coronae are formed by a highly diluted environment , and so the fluid description may be inappropriate . the question is posed of whether similar off - equatorial solutions can be determined also in the case of collisionless plasmas for which treatment based on kinetic theory , rather than fluid one , is demanded . in this paper the issue is addressed in the framework of the vlasov - maxwell description for non - relativistic multi - species axisymmetric plasmas subject to an external dominant spherical gravitational and dipolar magnetic field . equilibrium configurations are investigated and explicit solutions for the species kinetic distribution function are constructed , which are expressed in terms of generalized maxwellian functions characterized by isotropic temperature and non - uniform fluid fields . the conditions for the existence of off - equatorial tori are investigated . it is proved that these levitating systems are admitted under general conditions when both gravitational and magnetic fields contribute to shaping the spatial profiles of equilibrium plasma fluid fields . then , specifically - kinetic effects carried by the equilibrium solution are explicitly provided and identified here with diamagnetic , energy - correction and electrostatic contributions . it is shown that these kinetic terms characterize the plasma equation of state by introducing non - vanishing deviations from the assumption of thermal pressure .

1309.3979

this paper concerns the vlasov - maxwell description of collisionless magnetized plasmas @xcite related to axisymmetric discs arising in the combined gravitational and electromagnetic ( em ) fields of astrophysical compact objects . in particular , in this work an application of the kinetic theory developed by @xcite is considered , to the treatment of equilibrium and possibly non - neutral equatorial as well as off - equatorial plasma tori . remarkably , the existence of equilibrium levitating ( i.e. , off - equatorial ) structures has been recently pointed out by @xcite , based on a fluid non - ideal magnetohydrodynamics ( mhd ) description . indication of existence of structures of this type occurring in collisionless plasmas is also supported by other recent works based on test - particle approach @xcite . in addition , in the context of charge - separated pulsar magnetospheres , @xcite found an indication of the possible development of stable off - equatorial tori that can be revealed in his early numerical simulations of magnetized collisionless plasmas . indeed , from the physical point of view , off - equatorial plasma configurations are intrinsically different from the case of equatorial disc systems . the astrophysical relevance of these structures lies in the possibility of modelling coronal plasmas characterized by low - density and high - temperature conditions @xcite . in this regard , two different issues arise , which deserve a detailed investigation . first , under these conditions fluid descriptions may become inappropriate , requiring in principle the adoption of a kinetic treatment . however , it still remains to be ascertained whether the levitating structures can be recovered as equilibrium solutions in the framework of a kinetic description . second , a priori it is not obvious whether the kinetic theory developed previously for equatorial plasmas can be extended to the description of off - equatorial tori , how this can be achieved in practice and what are the physical implications as far as their occurrence in real systems is concerned . the statistical description of plasma dynamics can be carried out in terms of either fluid or kinetic treatments , with the choice of the appropriate framework depending on the plasma phenomenology to be addressed and the relevant features of the phenomena to be studied @xcite . the majority of fluid approaches are based on hydrodynamic or mhd treatments @xcite . in the case of collisionless plasmas , when these are formulated independently of an underlying kinetic theory , some limitations can arise . first , it is well known that the set of fluid equations may not be closed , requiring in principle the prescription of arbitrary higher - order fluid fields and closure conditions , including in particular the equation of state ( eos ) or the pressure tensor @xcite . second , in these approaches typically no account is given of microscopic phase - space particle dynamics together with phase - space plasma collective phenomena . on the other hand , only in the context of kinetic theory these difficulties can be consistently overcome , as this treatment permits to obtain well - defined constitutive equations for the relevant fluid fields describing the plasma state and to solve at the same time the closure problem @xcite . these issues become relevant in the case of collisionless or weakly - collisional multi - species plasmas subject to em and gravitational fields where phase - space particle dynamics is expected to play a dominant role . in particular , kinetic theory is essential for studying both stationary configurations and dynamical evolution of plasmas when kinetic effects are relevant , such as ones associated with conservation of particle adiabatic invariants @xcite , temperature and pressure anisotropies , diamagnetic and finite larmor - radius ( flr ) effects as well as energy - correction contributions @xcite . the problem of formulating a kinetic theory appropriate for the description of collisionless plasmas in quasi - stationary ( i.e. , equilibrium)configurations of astrophysical accreting systems and laboratory scenarios has been presented in series of works , based on the non - relativistic vlasov - maxwell description . in our context , several issues have been treated , ranging from laboratory plasmas occurring in tokamak devices @xcite , axisymmetric accretion disc plasmas characterized by locally - nested magnetic surfaces @xcite and current - carrying magnetic loops @xcite around compact objects as well as spatially non - symmetric systems in astrophysical and laboratory contexts @xcite . it was shown that consistent solutions of the vlasov equation can be determined for the species kinetic distribution function ( kdf ) describing collisionless plasmas , based on the identification of the relevant single - particle invariants . the equilibrium kdfs were expressed in terms of generalized bi - maxwellian distributions , characterized by temperature anisotropy , non - uniform fluid fields and local plasma flows . chapman - enskog representations of these equilibria were obtained by developing a suitable perturbative kinetic theory , which in turn made possible the analytical calculation of the fluid fields and the identification of the relevant kinetic effects included in the corresponding mhd description . as a basic consequence , it was shown that these solutions can exhibit non - vanishing current densities which can also support a kinetic dynamo mechanism for the self - generation of em fields in which the plasma is immersed @xcite . finally , more recently the kinetic theory has been extended to describe axisymmetric plasmas characterized by strong shear - flow and/or supersonic velocities @xcite , while @xcite reports a kinetic analysis of the stability properties of particular equilibrium solutions with respect to axisymmetric em perturbations . a notable feature of the kinetic equilibria mentioned here is the unique prescription of the functional dependences of an appropriate set of fluid fields , carried by the species kdfs , which are related to physical observables of the system . these fields are referred to as _ structure functions _ and are denoted as @xmath0 ( see in particular @xcite and the definition below ) , with the subscript s being the species index . depending on the kinetic regime being considered , according to the classification scheme presented by @xcite , these dependences are expressed in terms of the poloidal flux function @xmath1 of the magnetic field and/or the effective potential @xmath2 defined as@xmath3where @xmath4 and @xmath5 are the gravitational and electrostatic ( es ) potentials respectively , with @xmath6 and @xmath7 denoting the species particle mass and charge . thus , in the general case kinetic theory requires that @xmath8 . it must be stressed that , behind the apparent simplicity of the result , for practical applications one ultimately needs obtaining a representation of @xmath9 in terms of spatial coordinates , e.g. cylindrical ones @xmath10 . assuming an axisymmetric configuration where @xmath1 and @xmath2 are generic functions of both @xmath11 , the representation @xmath12 applies . hence , the complete solution of the problem actually requires determining the explicit representation of the potentials @xmath13 in terms of the spatial cylindrical coordinates . this can be a very difficult task , since in the general case the plasma itself contributes to the generation of the gravitational filed , through its non - vanishing mass - density , and , more important , of the em fields by means of non - vanishing charge and current densities . in the present discussion we ignore , however , the self - generation of the gravitational field , focusing only on the generation of em fields . it follows that , in order to obtain explicitly the relationship between the em potentials and the coordinate system , one has necessarily to solve ( numerically ) the coupled set of vlasov - maxwell equations to determine @xmath14 and @xmath15 , where the source terms of the fields are prescribed functions of the potentials . on the other hand , such a solution is also demanded for the following additional reasons : a ) in order to calculate explicitly the characteristic kinetic effects which enter the equilibrium kdf and are associated with diamagnetic - flr and energy - correction effects ( see @xcite and the discussion in section 7 ) ; b ) in order to establish the diffeomorphism @xmath16 which relates cylindrical ( and similarly , spherical coordinates ) to local magnetic coordinates , where @xmath17 is an angle - like coordinate defined on equipotential magnetic surfaces @xmath18 the complexity of the theory , as far as this issue is concerned , might represent a possible limit for the practical realization of kinetic equilibria of this type for configurations of astrophysical interest . this may be relevant especially when demanding a comparison between kinetic and fluid treatments based on analytical solutions . therefore , the question arises of whether such a difficulty can be actually encompassed in some scenarios , possibly by invoking suitable asymptotic kinetic orderings to be imposed on the system . in particular , this concerns the existence of configurations which allow for the construction of the diffeomorphism @xmath19 based on analytical solutions of the potentials @xmath20 in such a way to afford an explicit treatment of the spatial dependences of the equilibrium fluid fields carried by the kdf according to the constraints posed by the vlasov equation . a second point of crucial importance concerns the determination of the eos that characterizes disc plasmas in the collisionless state . as mentioned above , this is usually realized by prescribing the form of the scalar pressure , or more generally the components of the pressure tensor , which represent the closure condition for the euler momentum equation in mhd treatments . since collisionless plasmas are intrinsically characterized by the occurrence of phase - space anisotropies which cause the equilibrium kdf to generally deviate from a simple isotropic maxwellian distribution , the knowledge of the correct form of the eos is not a trivial task . in fact , one can only prescribe the pressure tensor in consistent way on the base of the kinetic theory ( kinetic closure conditions ) . in this regard , the problem consists in the identification of the kinetic effects which must be included in the pressure tensor , the understanding of their physical origin and the way they influence the macroscopic configuration of the plasma system . among the relevant ones , contributions associated with es corrections can play a central role , since they arise from microscopic charge interactions and carry information about local es fields generated by the validity of quasi - neutrality condition in rotating plasmas or deviations away from it in non - neutral systems . finally , from the astrophysical point of view a further motivation is represented by @xcite , where the discovery of the possible occurrence of stationary configurations of disc plasmas in off - equatorial tori is reported . in that work , a newtonian axisymmetric model of non - conductive , charged and perfect fluid tori orbiting in the combined gravitational and dipolar magnetic fields generated by a central compact object is presented . the result is obtained in the framework of a non - ideal mhd description and shows that the interplay between gravitational and magnetic fields can effectively enhance vertically extended structures in a plasma torus , which may correspond to localized concentrations of matter above and under the equatorial plane . the importance of this conclusion lies in the possibility of interpreting these off - equatorial tori as forming the surrounding material usually invoked to explain the spectral emission / absorption features in accretion - disc systems . possible examples of this type are provided by coronal halos consisting of non - neutral ion - electron plasmas or by obscuring dusty - plasma tori that are believed to be produced in galactic nuclei @xcite . the complete understanding of the physical properties of these structures is far from being satisfactory and deserves further investigations , both theoretical and observational . in particular , here we consider the possibility of a low - density and high - temperature coronal plasma for which the collisionless state applies . following the arguments discussed above , under these conditions the proper framework for the description of these plasmas is represented by the kinetic theory . in particular , the question is posed of whether kinetic equilibria can be proved to exist for collisionless magnetized plasmas which exhibit off - equatorial maxima in the matter distribution and how these solutions can be possibly related to fluid mhd ones . this amounts at identifying the appropriate kinetic regimes which meet these conditions and determining the spatial dependences of the corresponding fluid fields , including the pressure tensor and the intrinsically - kinetic effects contained in the eos . to be successful , a program of this type must make possible an analytical approach in accordance with the considerations presented above , to extend the kinetic theory developed in @xcite and to permit its practical application to gain insights into an astrophysical issue of notable importance connected with accretion - disc phenomenology . to conclude , we should comment on the fact that the reference works cited above on kinetic theory and off - equatorial tori as well as the present investigation are carried out in the framework of a non - relativistic description ( both with respect to the treatment of the gravitational field and the plasma velocities ) . however , the problem of general - relativistic solutions could in principle be posed , since there is a number of situations in which the formation of accretion discs is due to strong gravitational fields , e.g. around black holes , and general - relativistic corrections must be considered . although a complete theory of this type is still missing , non - relativistic treatments can nevertheless provide the reference framework for the inclusion of some relevant features characteristic of general - relativistic theories . this can be achieved for example by the adoption of pseudo - newtonian potentials for the description of spherically - symmetric gravitational fields @xcite . it has been shown , in fact , that the precision of the pseudo - newtonian description of stationary general - relativistic phenomena can be very high @xcite . an alternative consists in the inclusion of post - newtonian corrections . in this reference , a kinetic theory of self - gravitating collisionless gases adopting such a technique can be found in @xcite for spherical solutions , and in @xcite for the case of axially - symmetric solutions . in view of the considerations presented above , the purpose of this paper is the formulation of a kinetic theory appropriate for the analytical treatment of collisionless disc plasmas in axisymmetric off - equatorial tori ( see fig . [ fig-1 ] for a schematic view of the configuration geometry ) . the results of the investigation are the following ones : .[fig-1 ] \1 ) the identification of a physically - realizable astrophysical configuration in which the spatial profiles of the potentials @xmath21 can be analytically prescribed , when suitable plasma orderings apply . in the framework of an asymptotic theory , this permits to decouple the vlasov equation from the maxwell equations , at least to leading - order , with the possibility of an explicit treatment of the spatial dependences contained in the equilibrium plasma fluid fields . 2 ) the construction of species equilibrium kdfs which are consistent with the kinetic constraints imposed by microscopic phase - space conservation laws for the single particle dynamics . these are proved to be expressed in terms of generalized maxwellian kdfs characterized by non - uniform fluid fields and isotropic temperature , both to be associated with a finite set of structure functions . \3 ) the identification of the relevant kinetic regimes which can in principle arise in the configuration determined at point 1 ) above and the prescription of the corresponding functional dependences on the structure functions . 4 ) the development of a perturbative theory which makes possible a representation of the equilibrium kdfs in terms of a chapman - enskog series . as a result , this permits first the analytical evaluation of the plasma fluid fields , and second to distinguish the characteristic kinetic effects which enter the solution and their main physical properties . \5 ) the proof that the gravitationally - bound and magnetized plasma regime provides the most general functional dependences in the structure functions carried by the equilibrium kdf , which are consistent both with the analytical treatment of the fluid fields and the occurrence of off - equatorial tori . as an illustration of the technique , explicit calculation of the leading - order species number density and velocity profiles is provided for two configurations of physical interest . \6 ) the calculation of the eos for the equilibrium plasma , to be expressed in terms of the species pressure tensor associated with the kdf . by making use of the perturbative treatment , this includes also the identification of both the kinetic effects and the es corrections that can effectively contribute to the eos . in particular , the latter are shown to determine non - trivial deviations from the assumption of having thermal pressure for the collisionless plasma . \7 ) to determine the constraints to be imposed on the kinetic solution for the validity of the theory and the physical configuration realized , which arise from the maxwell equations for the self - generated equilibrium em fields . these represent necessary conditions which must be verified a posteriori for the complete solution of the vlasov - maxwell problem and the explicit treatment of es corrections in comparison with fluid - based approaches . the scheme of the paper is as follows . in section 3 the model assumptions and the fundamental em orderings are presented . section 4 deals with the definition of plasma orderings and the introduction of corresponding kinetic regimes which characterize collisionless plasmas treated here . in section 5 the equilibrium species kdfs are explicitly determined and their realizations are given for three different plasma regimes . in section 6 a suitable perturbative theory is developed , which allows for the analytical treatment of the equilibrium kdfs and the corresponding fluid fields . the conditions for the occurrence of off - equatorial tori are then investigated in the case of collisionless plasmas belonging to the gravitationally - bound and magnetized plasma regime . in section 7 the expression of the kinetic corrections which characterize the kinetic equilibria is provided , while the corresponding contributions in the eos are computed in section 8 . section 9 contains analysis of the poisson and the ampere equations and the constraints which they pose on the kinetic solution . finally , concluding remarks are summarized in section 10 . for the construction of kinetic equilibria , we ignore the possible existence of weakly - dissipative effects ( coulomb collisions and turbulence ) and em radiation effects @xcite . it is assumed that the kdf and the em fields associated with the plasma obey the system of vlasov - maxwell equations , with maxwell s equations being considered in the quasi - static approximation . for definiteness , we shall consider here a plasma consisting of @xmath22-species of charged particles which are characterized by proper mass @xmath6 and total charge @xmath7 . in particular , given a generic species kdf @xmath23 defined in the phase - space @xmath24 , with @xmath25 and @xmath26 being the configuration and velocity space respectively , the vlasov equation determines the dynamical evolution of @xmath27 and is given by@xmath28 the plasma is taken to be : a ) non - relativistic , in the sense that the flow velocities of all species are small compared to the speed of light @xmath29 , that the gravitational field can be treated within the classical newtonian theory , and that the non - relativistic vlasov kinetic equation is used as the dynamical equation for the kdf ; b ) collisionless , so that the mean free path of the plasma particles is much longer than the largest characteristic scale length of the plasma ; c ) axisymmetric , so that the relevant dynamical variables characterizing the plasma ( e.g. , the fluid fields ) are independent of the azimuthal angle @xmath30 when referred to a set of either cylindrical coordinates @xmath31 or spherical coordinates @xmath32 . thanks to the axisymmetry assumption , as a shortway in the following we shall denote with @xmath33 the configuration state vector , where @xmath33 denotes either @xmath34 or @xmath35 . we are concerned here with quasi - stationary configurations , namely solutions which are slowly - varying in time . this condition is also referred to as equilibrium configuration . for a generic physical quantity @xmath36 that depends on spatial coordinates @xmath33 and time @xmath37 , the quasi - stationarity is expressed by letting in the following @xmath38 , with @xmath39 being a small dimensionless parameter to be suitably defined ( see below ) and @xmath40 an integer . similar considerations apply for the equilibrium kdf @xmath27 , which is denoted in the following as @xmath41 . from the symmetry properties introduced here one can immediately derive the fundamental quantities which are conserved for the single - particle dynamics . in particular , under the assumptions of axisymmetry the canonical momentum conjugate to the azimuthal angle @xmath42 is an integral of motion . this is given by:@xmath43furthermore , from the condition of quasi - stationarity , the total particle energy@xmath44represents an adiabatic invariant of prescribed order , with @xmath2 being defined in eq . ( [ effective ] ) . following the discussion in @xcite , here we recall that a generic quantity @xmath45 defined in the phase - space is an adiabatic invariant of order @xmath46 with respect to @xmath47 when it satisfies the condition @xmath48 , where @xmath49 is a suitable integer and @xmath50 is the cyclotron frequency . this means that adiabatic invariants are conserved in asymptotic sense , namely up to a prescribed order of accuracy determined by the parameters @xmath47 and @xmath46 . we consider solutions of the equilibrium magnetic field @xmath51 which admit , at least locally , a family of nested axisymmetric toroidal magnetic surfaces @xmath52 , where @xmath1 denotes the poloidal magnetic flux of @xmath51 . the magnetic surfaces can be either locally closed @xcite or locally open @xcite in the configuration domain occupied by the plasma . in both cases a set of magnetic coordinates ( @xmath53 ) can be defined locally , where @xmath17 is a curvilinear angle - like coordinate on the magnetic surfaces @xmath54 by construction , magnetic coordinates are related to cylindrical or spherical coordinates by a diffeomorphism @xmath55 which must be consistently determined , as discussed above . each relevant physical quantity @xmath56 can then be conveniently expressed either in terms of the set @xmath33 or as a function of the magnetic coordinates , i.e. @xmath57 . consistent with these assumptions , we require the em field to be slowly - varying in time , i.e. , of the form @xmath58 in particular , we assume the magnetic field to be represented as@xmath59where @xmath60 and @xmath61 denote the self - generated magnetic field produced by the plasma and a finite external axisymmetric magnetic field ( vacuum field ) . for definiteness , in this treatment both contributions are assumed to exhibit only non - vanishing poloidal components , to be denoted in the following as @xmath62 . notice that , for what concerns the self - field , this assumption must be verified a posteriori to be consistent with the constraints placed on the kinetic solution by the maxwell equations . hence , the two fields are represented as@xmath63so that the total poloidal magnetic field takes the form@xmath64with @xmath65 . in particular , for the purpose of the present work , the external magnetic field is taken to coincide with a dipolar field . in such a case , the flux function @xmath66 is written in terms of the spherical coordinates @xmath67 as@xmath68with @xmath69 being the magnitude of the dipole magnetic moment . charged particles are also assumed to be subject to the effective potential @xmath70 defined by eq . ( [ effective ] ) . in principle , both the es potential @xmath71 and the gravitational potential @xmath72 can be produced by the plasma itself and by external sources . however , in the following it is assumed that @xmath71 is uniquely generated by the plasma charge density , while we shall neglect the self contribution of the plasma to @xmath4 . hence , for an axisymmetric disc , the gravitational potential is taken as being stationary and to coincide identically with the potential generated by the central compact object . the latter is expressed here in terms of the spherically - symmetric newtonian potential as@xmath73where @xmath74 is the newton gravitational constant and @xmath75 is the mass of the compact object . given validity of these assumptions , in order to address the first issue posed in the section 2 we proceed introducing the following fundamental orderings for the self em fields:1 ) the self component of the equilibrium magnetic field @xmath60 is ordered with respect to the external field @xmath61 as@xmath76with @xmath77 , and in general @xmath78.2 ) the equilibrium es potential @xmath5 satisfies the ordering assumption@xmath79with @xmath77 . this means that the es potential energy is small with respect to the gravitational potential energy , where @xmath47 will be properly defined below . in the following the case @xmath80 in eqs . ( [ basic ] ) and ( [ eees ] ) will be considered . it is important to remark that the two conditions ( [ basic ] ) and ( eees ) pose strong constraints from the physical point of view on the realizability of the equilibrium configuration for the collisionless disc plasma . in particular , they must be verified a posteriori in order to warrant the consistency of the kinetic equilibrium solution for the species kdf with the validity of the maxwell equations . ( [ basic ] ) and ( [ eees ] ) are the fundamental em orderings which permit the analytical treatment of the spatial dependences of the equilibrium fluid fields as prescribed by the kinetic solution to be determined below . in fact , when these orderings hold , to leading - order the magnetic flux function @xmath1 and the effective potential @xmath2 coincide respectively with the vacuum fields , namely the external dipolar flux function @xmath81 in eq . ( [ dipolo ] ) and with the gravitational potential @xmath4 given by eq . ( newton ) . hence , at this order it is possible to construct explicitly the diffeomorphism which relates the spherical coordinates @xmath67 with the potentials @xmath82 . this is expressed by letting@xmath83the corresponding relationship holding for cylindrical coordinates can then be obtained from the spherical ones , giving in particular@xmath84 in this section we introduce the relevant orderings for collisionless plasmas , which permit to identify corresponding kinetic regimes characterized by different equilibrium solutions and to verify their consistency with the existence of off - equatorial tori . to this aim we determine a classification scheme based on the method outlined in @xcite and suitable for the treatment of the physical setting indicated in the previous section . the first step is the definition of the dimensionless species - dependent parameters @xmath85 , @xmath86 and @xmath87 . these are prescribed in such a way to be all independent of single - particle velocity and at the same time to be related to the characteristic species thermal velocities . both perpendicular and parallel thermal velocities ( defined with respect to the magnetic field local direction ) must be considered . these are defined respectively by @xmath88 and @xmath89 , with @xmath90 and @xmath91 denoting here the species perpendicular and parallel temperatures . in detail , the first parameter is defined as @xmath92 , where @xmath93 is the species larmor radius , with @xmath94 being the characteristic scale - length of the spatial variations of all of the fluid fields associated with the kdf and of the em fields . the second parameter @xmath86 is related to the particle canonical momentum @xmath95 . by denoting @xmath96 , @xmath86 is identified with @xmath97 . hence , @xmath86 effectively measures the ratio between the toroidal angular momentum @xmath98 and the magnetic contribution to the toroidal canonical momentum , for all particles in which @xmath99 is of the order @xmath100 while @xmath1 is assumed as being non - vanishing . in particular , here the magnetic flux can be estimated as @xmath101 , with @xmath102 denoting the characteristic length - scale of flux variations and @xmath103 the magnitude of the poloidal magnetic field . note that , by definition , @xmath104 , but , in principle , we can also have @xmath105 locally . we also notice that the present definition of @xmath86 is not the only possibility , as one could also define the parameter @xmath86 such that it is related to the azimuthal flow velocity , namely letting @xmath106 , with @xmath107 being the corresponding angular frequency . finally , @xmath87 is related to the particle total energy @xmath108 and is prescribed in this work as @xmath109 . it follows that @xmath87 measures the ratio between particle kinetic and es potential energy with respect to the gravitational potential energy , for all particles having velocity @xmath110 of the order @xmath111 , with @xmath4 being assumed as non - vanishing . we notice that the definition of @xmath87 differs from that used in previous works @xcite while it is consistent with the ordering assumption ( [ eees ] ) . in the following we shall denote as thermal subset of velocity space the subset of the euclidean velocity space in which the asymptotic conditions @xmath112 holds . a comment is in order regarding the role of the magnetic field in the two parameters @xmath86 and @xmath85 . in the first case , the magnetic field is represented by means of the poloidal flux @xmath1 which contributes to the toroidal canonical momentum @xmath95 , while @xmath85 depends on the magnitude of the total magnetic field . invoking the definitions for @xmath86 and @xmath85 given above , it follows that @xmath113 , where @xmath94 and @xmath102 are respectively the characteristic scale - lengths of equilibrium fluid and em fields and of the poloidal flux . in general , the two quantities should be considered as independent , with @xmath104 and @xmath114 ( in the present context one has identically @xmath115from eq . ( [ b field ] ) ) . indeed , the parameter @xmath86 determines the particle spatial excursions from a magnetic flux surface , while @xmath85 measures the amplitude of the larmor radius with respect to the inhomogeneities of the background fluid fields . these two effects correspond to different physical magnetic - related processes , due respectively to the larmor - radius and magnetic - flux surface confinement mechanisms . in this work we assume that the ordering condition@xmath116holds for the collisionless plasma considered here . this amounts at requiring that the larmor radius remains small with respect to the scale - length @xmath94 , which , as shown in @xcite , represents a condition that is expected to be easily verified in accretion - disc systems . hence , one can consistently identify the small parameter @xmath47 introduced above with @xmath117 . the classification that is introduced in this work is based on the magnitude of the two parameters @xmath86 and @xmath87 . in detail , plasma species will be distinguished as belonging to the following regimes:1 ) _ gravitationally - bound _ if @xmath118 and @xmath119.2 ) _ magnetized _ if @xmath120 and @xmath121.3 ) _ gravitationally - bound and magnetized _ if both @xmath118 and @xmath120 . in the case of regimes 1 and 3 the following asymptotic expansion holds for the total particle energy @xmath122:@xmath123 . \label{order - sigma}\]]similarly , for regimes 2 and 3 the particle canonical momentum @xmath124 admits the expansion@xmath125 . \label{ordering - epsilon}\]]it is instructive to analyze the main features of these regimes and the physical conditions for their occurrence . the action of some energy non - conserving mechanisms is required for the establishment of the case of gravitationally - bound plasmas . in particular , plausible physical mechanisms that can be responsible for the decrease of the single - particle kinetic energy , in both collisionless and collisional ad plasmas , are em interactions ( e.g. , binary coulomb collisions among particles and particle - wave interactions , such as landau damping ) and/or radiation emission ( radiation - reaction ) . these can in principle be ascribed also to the occurrence of em instabilities and em turbulence . for single particles these processes can be dissipative . as a consequence , these particles tend to move towards regions with higher gravitational potential ( in absolute value ) . after multiple interactions of this type , the process can ultimately reach an equilibrium state which corresponds to the gravitationally - bound regimes . as far as the magnetic - field based classification , we notice that the requirement @xmath120 ( regimes 2 and 3 ) means that a particle trajectory remains close to the same magnetic surface @xmath126 , while satisfying the ordering ( [ eps - emme ] ) . finally , for greater generality , in the rest of the treatment we shall assume that , in the regimes in which @xmath118 and/or @xmath120 , the orderings @xmath127 and @xmath128 apply . in this section we proceed with the construction of the species equilibrium kdf and its characterization to the plasma regimes identified above . we consider both exact as well as asymptotic representations for the solution , the latter being expressed in terms of a chapman - enskog series . to reach the goal , here we adopt the solution technique developed in @xcite , which consists in the construction of solutions of the vlasov equation of the form @xmath129 , where @xmath130 is a suitable adiabatic invariant . this amounts at requiring that @xmath130 is necessarily a function of particle adiabatic invariants . in view of the model assumptions introduced above , it follows that the general form of the equilibrium kdf in the present context is given by@xmath131with @xmath40 and where slow - time dependences are assumed to be uniquely associated with the particle energy . here @xmath132 denotes the so - called structure functions , i.e. , functions which depend implicitly on the particle state @xmath133 . in order for @xmath134 to be an adiabatic invariant , @xmath132 must also be functions of the adiabatic invariants . this restriction is referred to here as a kinetic constraint . the precise form of the functional dependences of @xmath132 is characteristic of each plasma regime , as discussed below . in order to determine an explicit representation of @xmath130 according to eq . ( [ fss ] ) , we impose the following requirements : \1 ) the kdf must be characterized by species - dependent non - uniform fluid fields , azimuthal flow velocity and isotropic temperature , to be suitably prescribed in terms of the structure functions . \2 ) open , locally nested magnetic flux surfaces : the magnetic field is taken to allow quasi - stationary solutions with magnetic flux lines belonging to locally nested and generally open magnetic surfaces . \3 ) kinetic constraints : suitable functional dependences are imposed on the structure functions @xmath132 which depend on the regime being considered and such to warrant @xmath130 to be an adiabatic invariant . \4 ) in all regimes , @xmath130 is required to be asymptotically close ( in a suitable sense to be defined below ) to a local maxwellian kdf . this requires the possibility of determining a posteriori a perturbative representation of the kdf equivalent to the chapman - enskog expansion for the analytical treatment of implicit phase - space dependences contained in the structure functions , with the consistent inclusion of es corrections , flr - diamagnetic and/or energy - corrections contributions . 5 ) the kdf @xmath130 must be a strictly - positive real function and it must be summable , in the sense that the velocity moments of the form@xmath135must exist for a suitable ensemble of weight functions @xmath136 , to be prescribed in terms of polynomials of arbitrary degree defined with respect to components of the velocity vector field @xmath137 . then , following @xcite , we express the equilibrium kdf @xmath130 as@xmath138which is referred to as the _ generalized maxwellian kdf . _ here the structure functions are identified with the set @xmath139 , where @xmath140 , @xmath141 and @xmath142 are related to the species number density , isotropic temperature and azimuthal angular velocity respectively . invoking the definitions ( [ p_fi ] ) and ( [ total_energy ] ) , eq . ( [ gm ] ) can also be written as@xmath143 } { \left ( 2\pi /m_{\mathrm{s}}\right ) ^{3/2}t_{\ast \mathrm{s}}^{3/2}}\exp \left\ { -\frac{m_{\mathrm{s}}\left ( \mathbf{v}-\mathbf{v}_{\ast \mathrm{s}}\right ) ^{2}}{2t_{\ast \mathrm{s}}}\right\ } , \]]where @xmath144 and@xmath145 it is worth pointing out that the form of the solution ( [ gm ] ) holds for all the plasma kinetic regimes identified in the previous section . the difference in the three cases concerns the prescription of the kinetic constraints to be imposed on @xmath132 . in particular , consistent with the requirements listed above , these are assigned as follows : 1 ) for gravitationally - bound plasmas it is required that the functional dependence of @xmath132 is of the type@xmath146for which eq . ( [ order - sigma ] ) applies , while implicit dependences with respect to @xmath147 remain excluded in such a case . \2 ) for magnetized plasmas , the kinetic constraint is realized by imposing@xmath148for which eq . ( [ ordering - epsilon ] ) applies , while implicit dependences with respect to @xmath149 are excluded . \3 ) for gravitationally - bound and magnetized plasmas both eqs . ( order - sigma ) and ( [ ordering - epsilon ] ) hold , so that the general form of the kinetic constraint is given by@xmath150 the connection between the realization of these regimes and the occurrence of off - equatorial tori will be investigated in the next section . here it must be noticed that , because of the constraints ( [ gb - k])@xmath151([gb - m - k ] ) , at this stage the structure functions can not be regarded as fluid fields , since they are defined in the phase - space , namely they depend on the single particle velocity via the particle energy @xmath108 and the canonical momentum @xmath95 . instead , the fluid fields associated with @xmath130 must be properly computed as velocity moments according to eq . ( [ velomo ] ) and they are unique once the precise form of the structure functions is explicitly prescribed in @xmath130 . in this section we first proceed determining a chapman - enskog representation for @xmath130 that makes possible the treatment of the implicit phase - space functional dependences carried by the structure functions as well as the analytical evaluation of the equilibrium fluid fields and the associated kinetic contributions . we then apply the result to prove the validity of the kinetic theory developed here as far as the description of off - equatorial toroidal structures is concerned . this task can be achieved by implementing an appropriate perturbative theory for @xmath130 which was first developed in @xcite and which is based on a taylor expansion of @xmath132 with respect to the dimensionless parameters @xmath87 and @xmath86 . it is understood that the basic feature of such a kinetic perturbative technique is that it is strictly applicable only in localized subsets of velocity space ( thermal subsets ) , namely to particles whose velocity satisfies the asymptotic ordering ( [ order - sigma ] ) and/or ( ordering - epsilon ) . a notable consequence of such an approach is that , for each kinetic regime , quasi - stationary , self - consistent , asymptotic solutions of the vlasov - maxwell equations ( kinetic equilibria ) can be explicitly determined by means of suitable taylor expansions of @xmath130 . in particular , it is found that maxwellian - like kdfs can be obtained locally in phase - space , where the appropriate convergence conditions hold . this procedure provides also the correct constitutive equations of the leading - order fluid fields as well as the precise form of the es , flr - diamagnetic and/or energy - correction contributions to the kdf . in detail , invoking eqs . ( [ order - sigma ] ) and ( [ ordering - epsilon ] ) , a linear asymptotic expansion for the structure functions can be obtained . in the general case , neglecting corrections of @xmath152 as well as of @xmath153 and @xmath154 , with @xmath155 , this is given by@xmath156 _ { \psi _ { \ast \mathrm{s}}=\psi , \phi _ { \ast \mathrm{s}}=\frac{m_{\mathrm{s}}}{z_{\mathrm{s}}e}\phi _ { \mathrm{g } } } \nonumber \\ & & + \left ( \phi _ { \ast \mathrm{s}}-\frac{m_{\mathrm{s}}}{z_{\mathrm{s}}e}\phi _ { \mathrm{g}}\right ) \left [ \frac{\partial \lambda _ { \ast \mathrm{s}}}{\partial \phi _ { \ast \mathrm{s}}}\right ] _ { \psi _ { \ast \mathrm{s}}=\psi , \phi _ { \ast \mathrm{s}}=\frac{m_{\mathrm{s}}}{z_{\mathrm{s}}e}\phi _ { \mathrm{g } } } , \label{qq}\end{aligned}\]]where @xmath9 is the leading - order term which uniquely follows for each regime ( see below ) . when eq . ( [ qq ] ) is applied to the equilibrium kdf @xmath130 and the ordering ( [ eees ] ) is also invoked , the following chapman - enskog representation is found:@xmath157 , \label{ce}\]]where the leading - order contribution @xmath158 coincides with a drifted isotropic maxwellian kdf carrying non - uniform number density , azimuthal differential flow velocity and isotropic temperature . in detail:@xmath159where @xmath160 is the leading - order drift velocity carried by @xmath158 . here @xmath161 represents the leading - order species number density and is given by@xmath162 , \label{nde}\]]with @xmath163 being referred to as the pseudo - density . the leading - order structure functions @xmath9 coincide now with the set of functions @xmath164 which are defined in the configuration space , with @xmath165 and @xmath107 being respectively the leading - order species temperature and azimuthal rotation angular frequency . in addition , the quantities @xmath166 , @xmath167 and @xmath168 represent first - order kinetic corrections . in particular , @xmath169 is referred to as flr - diamagnetic contribution , @xmath167 carries energy - correction contributions ( with respect to both kinetic and es potential energies ) , while @xmath170 represents a purely es term . the precise expression of these functions will be given below in a separate section , where we discuss the relevance of these kinetic effects and their physical meaning in the framework of the present perturbative theory . for the moment , it is sufficient to say that all the first - order corrections are part of the kinetic equilibrium , and can not be neglected for the consistent formulation of the solution . it must be also stressed here that eq . ( [ ce ] ) is very general : while @xmath170 is non - vanishing for all the kinetic regimes considered above , the existence of @xmath169 and @xmath167 depends instead on the type of kinetic constraints . in particular , we distinguish the following features : 1 ) for gravitationally - bound plasmas ( regime 1 ) @xmath171 and @xmath9 is subject to the constraint@xmath172 \2 ) for magnetized plasmas ( regime 2 ) @xmath173 and the functional dependence of @xmath9 becomes@xmath174 \3 ) for gravitationally - bound and magnetized plasmas ( regime 3 ) in general both @xmath175 and @xmath176 , while for @xmath9 one has in this case@xmath177 we notice that , to the leading - order , the equilibrium solution determined here does not depend on the es potential , but only on the gravitational potential @xmath4 and the magnetic flux @xmath178 , which are assigned and known functions of the spatial coordinates . the perturbative theory developed here represents the starting point for the application of the kinetic theory to the modelling of off - equatorial plasma tori in axisymmetric disc systems . this is based on the analysis of the spatial dependences which characterize the leading - order kinetic solution and that can be dealt with analytically thanks to the fundamental em ordering assumptions introduced in section 3 . in particular , for each of the three regimes considered here , the proof that the kinetic equilibria admit off - equatorial solutions follows by analyzing the spatial profile of the leading - order number density defined by eq . ( [ nde ] ) under the requirement of having maxima out of the equatorial plane , namely for @xmath179 . this requires preliminary to assign the functional form of the structure functions @xmath9 characterizing eq . ( [ nde ] ) in terms of the potentials @xmath1 and/or @xmath4 . a detailed discussion of this type for all of the three plasma regimes is beyond the scope of this work and will be addressed separately in future studies . for the purpose of the present investigation , it is sufficient to consider here the case of regime 3 . in fact , this provides the most general conditions for the occurrence of levitating structures , while regimes 1 and 2 can be viewed as special realizations of regime 3 . in particular , we notice that the latter is expected to represent also the most plausible realization in real systems , in which both the gravitational and magnetic fields contribute to determine the profiles of the fluid fields . let us then discuss the case of plasmas belonging to regime 3 . the number density profile is prescribed according to eqs . ( [ nde ] ) and ( [ r3 ] ) . we notice that , thanks to the analytical relationships ( [ d1 ] ) and ( d2 ) and the orderings ( [ basic])@xmath151([eees ] ) , any function of @xmath180 can be equivalently expressed in terms of @xmath181 . because of this , the rhs of eq . ( nde ) becomes now a generic function of @xmath181 , namely of the form @xmath182 . this represents the most general kind of spatial dependence which is admitted by the kinetic equilibrium . hence , in the general case and in the absence of other particular restrictions , suitable prescriptions of @xmath183 can be determined for regime 3 plasmas , according to the real system to be studied , which admit maxima out of the equatorial plane . this conclusion has a general character of validity and assures the consistency of the kinetic theory presented here for collisionless axisymmetric plasmas with the possible occurrence of levitating tori in the external gravitational and magnetic fields of the type prescribed above . we can now explore in more detail the present conclusion by considering explicitly two possible physical realizations of this type of solution : _ case a : _ in this first example we assume that both @xmath163 and @xmath165 in eq . ( [ nde ] ) are constant . from the physical point of view , the requirement @xmath184 means that the spatial variations of the number density are uniquely determined by the exponential term ( maxwellian factor ) , which in turn depends on the leading - order plasma temperature as well as on the rotational frequency , gravitational potential and magnetic flux @xmath1 . this choice is consistent with previous literature ( see for example @xcite ) . concerning the condition @xmath185 , this corresponds to a leading - order plasma isothermal profile that is consistent with the kinetic constraints that characterize the solution ( see section 7 ) . we remark that in the present framework , the isothermal condition can only be satisfied to the leading - order , while for non - uniform plasmas , the full temperature profile is generally non - constant because of higher - order kinetic effects . these issues will be discussed in detail in sections 7 and 8 . in validity of the prescription of constant @xmath163 and @xmath165 , the only freedom left concerns the functional dependence of @xmath186 , which is considered of the form ( [ r3 ] ) . under this assumption , the number density still remains of the type @xmath183 . in this situation , it is convenient to prescribe @xmath183 consistent with the requirement of exhibiting maxima out of the equatorial plane , independently of its actual representation given by eq . ( [ nde ] ) . the prescription of a physically - acceptable profile of @xmath183 must be done in such a way to reproduce observational or experimental data . once the profile of @xmath183 is set , since also @xmath163 and @xmath165 are constant in this example , then eq . ( [ nde ] ) can be inverted and used to uniquely derive the expression of the corresponding species angular frequency @xmath187 which determines the levitating structure . in particular , the latter is obtained by solving the quadratic algebraic equation@xmath188 hence , under these conditions , it is possible to introduce a density profile which is in agreement with physical configurations and the existence of off - equatorial tori . for leading - order isothermal systems this also prescribes the form of the corresponding plasma rotation frequency according to eq . ( [ om ] ) . the extension of this solution method to the case of a non - isothermal plasma species recquires the additional prescription of the temperature profile , namely @xmath189 , while the frequency @xmath107 can still be obtained from eq . ( [ om ] ) . in both cases we notice that the kinetic equilibrium thus determined generally allows for the existence of two separate roots for @xmath107 . if both are real , they should correspond to two different admissible equilibria with opposite direction of plasma rotation with respect to the external dipolar magnetic field orientation . _ case b : _ as a second example , we assume the validity of the kinetic constraint eq . ( [ r3 ] ) for all the three structure functions . in particular , here we treat the situation in which the condition@xmath190is identically satisfied in the configuration domain occupied by the collisionless plasma species . from the physical point of view , eq . ( req1 ) means that the number density profile @xmath161 is not modified by the maxwellian exponential factor and coincides with the pseudo - density @xmath191 . this requirement is satisfied when the exponential factor in eq . ( [ nde ] ) is one . hence , this condition is met for the species angular frequency satisfying the algebraic quadratic equation@xmath192 we notice that again eq . ( [ omg ] ) generates two roots for the frequency @xmath107 , as for the case a discussed above . in addition , eq . ( [ omg ] ) holds for both isothermal and non - isothermal plasmas . finally , in validity of the @xmath193ordering and the ordering ( [ eees ] ) , when @xmath194 , one can infer that the solutions of @xmath107 from eq . ( omg ) are asymptotically close to those from eq . ( [ om ] ) , although the number density and the temperature are not necessarily so . to conclude this section , it is useful to make a qualitative comparison of the results obtained here with those presented in ref . @xcite , where the existence of off - equatorial structures has been proved on the basis of a fluid non - ideal mhd description . this involves in particular the inspection of eq . ( 41 ) for the pressure profile given in ref . @xcite , which can give rise to off - equatorial maxima for suitable choices of the coefficients entering the same equation ( see discussions in sections 3 and 4 in the same reference ) . indeed , pressure and density profiles are proportional ( at least to the leading - order ) when the condition @xmath185 applies ( see also eq . ( [ ptot ] ) below ) . in such a case , it is immediate to verify that the rhs of eq . ( 41 ) in ref . @xcite can be effectively expressed as a function of @xmath1 and @xmath4 only , in agreement with the prescription holding for regime 3 plasmas . although the two treatments ( i.e. , the present one and ref . @xcite ) consider different physical conditions for the levitating plasma , the consistency pointed out here establishes a notable result . in fact , first it shows that , as anticipated in the introduction , the kinetic theory developed in this paper and its analytical formulation allow for direct comparisons with previous literature works based on fluid approaches . second , it proves that fluid results can in principle be reproduced consistently on the basis of a kinetic treatment , thus extending their validity to a wider class of plasma regimes . third , in turn it reinforces the statement given above concerning the general character of the present kinetic theory for regime 3 plasmas in providing a suitable mathematical and physical framework for the investigation of off - equatorial structures . in this section we provide the explicit representation of the kinetic corrections @xmath169 , @xmath195 and @xmath170 introduced in the chapman - enskog representation of the equilibrium kdf given by eq . ( [ ce ] ) . the inclusion of these contributions is necessary for the complete solution of the equilibrium problem in the framework of vlasov - maxwell description of collisionless plasmas . in particular , these terms represent the deviations of the kdf from a maxwellian distribution and arise because of the constraints imposed by single - particle phase - space conservation laws on the solution itself . the precise definition of the first - order terms @xmath169 , @xmath167 and @xmath170 is also required to distinguish the solution among the three kinetic regimes pointed out above . in detail , the first - order corrections @xmath169 and @xmath167 originate from the perturbative treatment of the implicit phase - space dependences carried by the structure - functions @xmath132 entering the equilibrium kdf @xmath130 . they are found to be given by:@xmath196 v_{\varphi } \nonumber \\ & & + \frac{cm_{\mathrm{s}}}{z_{\mathrm{s}}e}r\left [ \frac{e_{\mathrm{s}}-\omega _ { \mathrm{s}}p_{\varphi \mathrm{s}}}{t_{\mathrm{s}}}-\frac{3}{2}\right ] a_{3\mathrm{s}}v_{\varphi } , \label{h1}\end{aligned}\]]@xmath197 c_{3\mathrm{s}}\left ( \frac{1}{2}v^{2}+\frac{{z_{\mathrm{s}}e}}{m_{\mathrm{s}}}\phi \right ) \nonumber \\ & & + \left [ c_{1\mathrm{s}}+\frac{p_{\varphi \mathrm{s}}\omega _ { \mathrm{s}}}{t_{\mathrm{s}}}c_{2\mathrm{s}}\right ] \left ( \frac{1}{2}v^{2}+\frac{{z_{\mathrm{s}}e}}{m_{\mathrm{s}}}\phi \right ) , \label{h2}\end{aligned}\]]where the following definitions have been introduced:@xmath198hence , @xmath169 and @xmath195 are polynomial functions of the particle velocity which contain diamagnetic and energy - correction contributions and depend on the so - called thermodynamic forces @xmath199 and @xmath200 . the latter represent the gradients of the structure functions across equipotential magnetic and gravitational surfaces and arise in collisionless plasmas characterized by non - uniform fluid fields . consistency of these expressions with the @xmath86 and @xmath87 ordering assumptions requires that@xmath201 v_{\varphi } \lesssim o\left ( \varepsilon _ { \mathrm{s}}\right ) , \]]@xmath202 c_{3\mathrm{s}}\left ( \frac{1}{2}v^{2}+\frac{{z_{\mathrm{s}}e}}{m_{\mathrm{s}}}\phi \right ) \lesssim o\left ( \sigma _ { \mathrm{s}}\right ) , \]]which implies that @xmath165 must actually be of the form @xmath203 , with @xmath40 , i.e. at most slowly - dependent on @xmath1 and @xmath4 . this conclusion motivates the choice done in section 6 to treat isothermal plasmas ( to leading - order ) . as pointed out above , the contribution @xmath169 is null for gravitationally - bound plasmas , while @xmath167 vanishes for magnetized plasmas . instead , both terms are present in the equilibrium solution for plasmas belonging to regime 3 . we also notice that the @xmath193expansion generates contributions in @xmath167 which are proportional to both particle kinetic energy and es potential . in particular , terms which depend on @xmath5 contribute to the occurrence of es corrections to the kinetic solution and the corresponding fluid fields . finally , the last contribution @xmath170 originates from the validity of the @xmath204ordering ( [ eees ] ) when this is taken into account in the expression for the leading - order number density , and results in the following term@xmath205it must be stressed that the es contributions arising in @xmath167 and @xmath170 originate from different perturbative treatments . in fact , @xmath206 follows from the @xmath204ordering and is common to all the regimes considered here when eq . ( [ eees ] ) applies . instead the terms in @xmath167 can only be included when the @xmath207ordering applies ( regimes 1 and 3 ) . to conclude the section , it is worth pointing out that the treatment of the first - order kinetic corrections displayed here requires the following preliminary steps:1 ) the precise identification of the appropriate plasma collisionless kinetic regime.2 ) the prescription of the leading - order spatial profiles of the structure functions , consistent with the kinetic constraints for each regime.3 ) the evaluation of the thermodynamic forces and the explicit calculation of the es potential.in particular , the existence of the equilibria determined here is subject to the validity of the maxwell equations , i.e. the poisson equation for the es potential @xmath5 and ampere s equation ( see section 9 ) . in this section we proceed with the calculation of the eos corresponding to the kinetic equilibrium determined here . this requires in particular to compute the species pressure tensor @xmath208 carried by the kdf @xmath130 and defined as@xmath209where @xmath26 denotes the velocity domain of integration . since @xmath130 is isotropic with respect to quadratic particle velocity dependences , one can immediately infer that each species pressure tensor is isotropic and of the form @xmath210 , where @xmath211 denotes the thermal scalar pressure , with @xmath212 and @xmath213 being respectively the species total number density and temperature associated with @xmath214 . the calculation of @xmath215 can be carried out analytically for thermal particles when the chapman - enskog representation ( [ ce ] ) applies . in the following we consider this case . furthermore , consistent with the @xmath86 and @xmath87 orderings , in the first - order terms @xmath166 and @xmath167 we approximate the canonical momentum @xmath216 and the energy @xmath108 respectively with @xmath217 and @xmath218 . hence , under these assumptions , one can prove that the scalar pressure can be represented as@xmath219 , \label{ptot}\]]where @xmath220 is the leading - order term , with @xmath161 being defined by eq . ( [ nde ] ) . in addition we notice that the term @xmath169 associated with the @xmath221expansion does not contribute to the eos because it is odd in the azimuthal component of particle velocity . instead , @xmath170 does not depend explicitly on particle velocity and therefore it is not affected when the integral ( [ pten ] ) is computed on @xmath26 . hence , its contribution in eq . ( [ ptot ] ) is simply proportional to @xmath222 and represents part of the es corrections which enter the definition of the total pressure @xmath215 . finally , invoking eq . ( [ h2 ] ) , explicit calculation gives for @xmath223 the following result:@xmath224where@xmath225 from this result it is interesting to point out that , although to the leading - order the species pressure coincides with the thermal pressure , the first - order corrections introduce deviations in the eos that are distinctive for collisionless plasmas . in particular , the energy - correction contributions which enter through @xmath226 are associated with the gradients of structure functions across gravitational equipotential surfaces and include also es corrections proportional to @xmath5 . these terms however vanish in uniform collisionless plasmas . on the other hand , the es correction to the eos carried by @xmath170 is independent and follows uniquely from the @xmath204ordering introduced above between es and gravitational potential energy . clearly , all the first - order contributions in the eos arise as specifically - kinetic effects , which characterize the kinetic treatment of collisionless plasmas . in this section we analyze the constraints which are posed by the maxwell equations on the kinetic equilibria . these concern in particular the validity of the orderings ( [ basic ] ) and ( [ eees ] ) and for this reason they apply to all the plasma regimes identified above . we consider first the poisson equation for the es potential , which is written as@xmath227 in the general case the solution is non - trivial , because the total number density @xmath228 depends both implicitly and explicitly on the es potential itself ( see for example @xcite ) . however , the solution simplifies in validity of the sub - ordering expansion ( [ eees ] ) introduced above . in fact in this case the es potential enters the kinetic solution only through the first - order corrections to the equilibrium kdf . therefore , consistent with the orderings introduced in the present work and the perturbative theory developed here , one can obtain an asymptotic solution for @xmath5 by considering only the leading - order contribution to the species number density . thus , when the said sub - ordering applies , neglecting corrections of @xmath229 and @xmath230 and invoking eq . ( [ nde ] ) , the poisson equation becomes to this accuracy:@xmath231where the source term @xmath232 is defined as@xmath233 . \label{tesader}\]]here @xmath234 does not depend on @xmath5 , and therefore the es potential can be readily obtained by integrating eq . ( [ po2 ] ) yielding@xmath235with @xmath236 being the corresponding green function . for the consistency of the theory , the solution for @xmath5 given by the previous equation must be checked a posteriori to satisfy the initial ordering ( [ eees ] ) . in particular , this can represent a constraint condition for the magnitude of the species number densities which contribute to the es potential through the system charge density ( [ tesader ] ) . manifestly , the validity of the ordering ( [ eees ] ) is necessary for the present theory to apply , and for this reason the calculation of @xmath5 represents the ultimate step to be done in order to warrant the consistency of the treatment . similar considerations apply to the ampere equation , which determines the self - generation of magnetic field by the equilibrium collisionless plasma . the ampere equation is written as@xmath237where @xmath60 is defined in eq . ( [ bself ] ) and @xmath238 is the total current density , which is given by@xmath239with @xmath240 being the species flow velocity . it is immediate to prove that , in the present formulation , @xmath240 is purely azimuthal at equilibrium ; in fact additional components of the velocity can only arise in the presence of temperature anisotropy , see for example @xcite . again , consistent with the perturbative treatment presented here , one can retain only the leading - order contributions to @xmath238 in eq . ( [ ampere-2 ] ) . under this assumption eq . ( [ ampere-2 ] ) becomes@xmath241where @xmath161 is given by eq . ( [ nde ] ) and @xmath160 ( see eq . ( [ fm ] ) ) . ( [ gss ] ) represents a generalized grad - shafranov equation for the poloidal magnetic flux @xmath242 in which the source term on the rhs depends only on explicitly known quantities . the solution for @xmath60 which results from eq . ( [ gss ] ) must then be checked a posteriori to verify the ordering condition ( [ basic ] ) introduced above , which is necessary in order to warrant the validity of the theory and its analytical development . in this case eq . ( [ basic ] ) can represent a constraint for the magnitude of the species rotation angular frequencies which contribute to the system charge current . in this paper , a theoretical investigation of equilibrium configurations for collisionless non - relativistic and axisymmetric plasmas has been presented , taking into account the role of a central spherically - symmetric gravitational field . the formulation is based on a multi - species kinetic approach developed in the framework of the vlasov - maxwell description . the case of astrophysical plasmas arising in the gravitational field of compact objects and in the presence of both an external dipolar magnetic field and self electromagnetic fields has been treated . three different plasma regimes have been identified which are characetrized by distinctive kinetic orderings . it has been proved that in all cases consistent kinetic equilibria can be determined , with the kinetic distribution function being expressed in terms of generalized maxwellian functions . it has been shown that the three regimes differ by the form of the kinetic constraints which are imposed on the equilibrium solutions and which uniquely follow from phase - space single - particle conservation laws . by imposing appropriate orderings on the self electromagnetic fields and by developing a suitable perturbative theory , an analytical treatment of the equilibria has been proposed . in terms of this , several issues have been addressed . first , the conditions of existence of equilibrium structures corresponding to off - equatorial tori have been investigated . it has been shown that these systems can generally arise for the regime which has been referred to here as magnetized and gravitationally - bound plasmas . this analysis can be important from the astrophysical point of view , since off - equatorial tori may represent a physically - realizable model of magnetized coronal plasmas which are believed to characterize accretion discs . in addition , the treatment based on kinetic theory can pose the basis for comparison with analogous fluid results carried out in terms of mhd theory . as a second application , the plasma equation of state has been determined analytically and expressed in terms of the pressure tensor . it has been shown that the latter exhibits deviations from the thermal pressure characteristic of collisional plasmas because of the existence of specifically - kinetic effects . these have been identified with diamagnetic , energy - correction and electrostatic contributions which apply in combination with the occurrence of non - uniform fluid fields . finally , the validity of the poisson and ampere equations have been addressed , showing that they can introduce non - trivial constraints on the magnitude of the plasma number density and flow velocity for the consistency with the orderings introduced in the theory developed here . the conclusions established in this work can be relevant for future investigations of astrophysical plasmas in equilibrium configurations , with particular focus on collisionless plasmas in accretion discs and off - equatorial tori associated with compact objects . financial support by the italian foundation angelo della riccia ( firenze , italy ) is acknowledged by c.c . , j.k . , p.s . and z.s . would like to express their acknowledgment for the institutional support of faculty of philosophy and science , silesian university in opava ( czech republic ) and thank the project synergy cz.1.07/2.3.00/20.0071 sustaining the international collaboration . thanks czech science foundation gar 13 - 00070j .

the collective behavior of spins in a dilute magnetic semiconductor is determined by their mutual interactions and influenced by the underlying crystal structure . hence , we begin with the atomic quantum - mechanical description of this system using the proposed variational - perturbation calculus , and then turn to the emerging macroscopic picture employing phenomenological constants . within this framework we study spin waves and exchange stiffness in the zener model of ( ga , mn)as , its thin layers and bulk crystals described by the tight - binding approximation . analyzing the anisotropic part of exchange , we find that the dzyaloshinskii - moriya interaction may lead to a cycloidal spin arrangement and uniaxial in - plane anisotropy of diagonal directions in thin layers , resulting in a surface - like anisotropy in thicker films . we also derive and discuss the spin - wave contribution to magnetization and curie temperature . our theory reconstructs the values of stiffness determined from the temperature dependence of magnetization , but reproduces only partly those obtained from analyzing precession modes in ( ga , mn)as thin films .

1006.1271

dilute magnetic semiconductors , such as ( ga , mn)as , form a class of materials with many outstanding properties@xcite and functionalities.@xcite a number of these result from the complex structure of the valence band , which hosts holes mediating magnetic order between spins localized on transition metal impurities . the strong force which lies behind this mechanism is the carrier - mediated exchange interaction . it gives the low - lying energy states of such a system the character of spin waves , which contribute to both its equilibrium ( e.g. spontaneous magnetization and curie temperature ) and nonequilibrium magnetic properties ( ferromagnetic resonance and relaxation).@xcite at the same time , it allows us to replace the atomic quantum - mechanical description of the system with the classical , continuous micromagnetic theory , where it appears in form of exchange stiffness.@xcite quite generally , micromagnetic properties of any ferromagnet are determined by the magnitudes of the exchange stiffness and magnetocrystalline anisotropy.@xcite the first describes the exchange energy associated with nonuniform distributions of local directions of magnetization . the other is the energy needed to change the total magnetization direction with respect to the crystal axes , which involves the competing crystal - field and spin - orbit interactions . while the main part of the exchange energy is isotropic , a consequence of its electrostatic origin , the relativistic spin - orbit coupling can create its small anisotropy , namely the dependence on the crystalline orientation of magnetization . in zinc - blende bulk crystals and thin layers with broken space inversion symmetries , it leads to many interesting effects , related to the anisotropic@xcite and dzyaloshinskii - moriya@xcite exchange , which have not been hitherto studied in ( ga , mn)as . since the total spin is no longer conserved , they are indispensable when considering any usage of spins for information storage and processing . the spin - wave spectrum and the isotropic exchange stiffness in bulk ( ga , mn)as were computed by knig , jungwirth and macdonald,@xcite and by brey and gmez - santos@xcite within the @xmath0 zener model employing the six - band @xmath2 hamiltonian,@xcite which neglects the inversion asymmetry specific to the zinc - blende lattice . it was found that the actual magnitude of the exchange stiffness is much greater when one takes into account the complex structure of the valence band , as compared to the case of a simple parabolic band.@xcite this , as well as the highly anisotropic fermi surface , were shown to explain@xcite why the mean - field approximation@xcite is so accurate in ( ga , mn)as . it was also found that the values of exchange and anisotropy energies obtained within the same formalism describe quantitatively@xcite the width of stripe domains in films with perpendicular magnetic anisotropy . more recently , bouzerar@xcite employed a self - consistent local random - phase approximation in order to evaluate the spectral density of spin - wave excitations in ga@xmath3mn@xmath4as . the magnitudes of spin - spin exchange integrals @xmath5 were obtained from first principles computations within the local spin - density approximation ( lsda ) and tight - binding linear muffin - tin orbital approach neglecting the spin - orbit interaction . the theory allows to treat disorder and thermal fluctuations , and shows that the calculated spectral density has well - defined maxima up to about one half of the relevant debye wave vector @xmath6 , where @xmath7 is the lattice constant . this made it possible to determine the spin - wave dispersion @xmath8 in the range @xmath9 , from which the magnitude of spin - wave stiffness was obtained.@xcite experimentally , potashnik _ et al._@xcite analyzed the temperature dependence of magnetization in a series of ga@xmath3mn@xmath10as samples , which provided the values of spin - wave stiffness from the @xmath11 bloch law . in later experiments , the stiffness was determined by examining spin precession modes excited by optical pulses@xcite and under ferromagnetic resonance conditions.@xcite the values obtained for some films with thickness greater than 120 nm , either as - grown@xcite or annealed,@xcite are in good agreement with those predicted by bouzerar.@xcite however , the values for thinner films@xcite or another series of annealed samples@xcite were about three times smaller . similarly small magnitudes of spin - wave stiffness were found by analyzing the domain structure of annealed ga@xmath12mn@xmath13as.@xcite the experimental works@xcite demonstrate that spectral positions of spin - wave resonances are strongly affected by the character of spin pinning at the sample borders . the same effect is caused by magnetic anisotropy changes along the growth direction , particularly strong at the film interface and surface.@xcite therefore , theoretical predictions concerning spin - wave excitations in both thick and thin layers may provide a useful guide to better understanding of magnetization dynamics in real samples . in this paper we investigate spin waves and related micromagnetic constants in ferromagnetic ( ga , mn)as , as described by the @xmath0 zener model.@xcite the validity of this model is supported by photoemission experiments@xcite and _ ab initio _ computations , in which the inaccuracies of the lsda are partly reduced by self - interaction.@xcite furthermore , the magnitude of low - temperature quantum corrections to conductivity indicates that in the concentration range relevant to ferromagnetism , the density of states assumes values expected for valence band holes.@xcite the carrier band is formed from @xmath14-type - like states of the gaas semiconductor structure , which are mostly built from the anion @xmath15 orbitals , but has a rather large @xmath16 component as the result of a strong @xmath17 interaction . we describe it by the @xmath1 tight - binding approximation for thin layers and , by applying periodic boundary conditions , for bulk crystals.@xcite our analysis of the anomalous hall effect in ( ga , mn)as in refs . and provides a thorough comparison of this method with other band structure models.@xcite one of its advantages is that it captures the inversion asymmetry of the zinc - blende lattice , which produces the dresselhaus spin splitting of the conduction band,@xcite and additionally the structure inversion asymmetry in thin layers , which creates the bychkov - rashba spin splitting.@xcite the growth - induced biaxial strain is included by changing the atoms arrangement , according to the strain tensor values : @xmath18 and @xmath19 , where @xmath20 is the strain - induced change of the lattice constant , and @xmath21 is the ratio of elastic moduli . also , the on - site energies of the @xmath16 orbitals depend linearly on the strain tensor values.@xcite the @xmath17 exchange coupling is modelled using the virtual crystal and mean - field approximations , while taking into account the appropriate weights of ga and as orbitals in the wavefunctions close to the center of the brillouin zone.@xcite in this way , spin polarization of individual mn moments is replaced by a molecular field , which creates a @xmath22-dependent zeeman - like splitting of the host bands . for heavy holes in the @xmath23 point it is given by @xmath24 , where @xmath25 is the fraction of cation site density @xmath26 substituted by mn forming a spin @xmath27 , and @xmath28 mevnm@xmath29 is the @xmath0 exchange integral.@xcite the validity of this approach can be questioned if the magnetic ions produce bound states,@xcite but the metallic character of the carrier states of interest here means that this does not occur due to many - body screening . at the same time we neglect the effect of this screening on potentials produced by magnetization fluctuations , which is justified as long as @xmath30 is smaller than the fermi energy , so that spin and charge density fluctuations are decoupled . a part of mn atoms form unintentional defects such as interstitials ( which can passivate single substitutional spins ) and antisites . both being double donors , they significantly lower the hole density , which can be partly remedied by removing the interstitials by post - growth annealing . the paper is arranged as follows . in sec . [ sec : theoreticalmodel ] we use the proposed variational - perturbation calculus to describe the system of lattice spins interacting via hole carriers , and its spin - wave excitations . in the micromagnetic theory , the system is described by the set of constants related to magnetocrystalline anisotropy and exchange energy . section [ sec : exchange ] provides quantitative results on the spin - wave stiffness expressed as the dimensionless parameter @xmath31 , comparing it between the @xmath2 and @xmath1 tight - binding models . in sec . [ subsec : magntemp ] we derive the spin - wave contribution to magnetization and curie temperature . section [ sec : experimental ] compares our theory to related experimental findings.@xcite the last section of the paper contains the summary of our work . in this section , we demonstrate a physically transparent perturbation - variational method of treating the systems in question . we use it to find the effective hamiltonian of lattice ions interacting through hole carriers , written in the _ interaction representation _ of creation and annihilation operators . next , we calculate the dispersion dependence of its low lying energy states by transforming the hamiltonian to the _ spin - wave representation_. these two pictures are associated with different sets of phenomenological parameters describing the macroscopic system , whose properties we shall investigate later on . we consider the ferromagnetic phase of a system consisting of @xmath32 carriers and @xmath33 magnetic lattice ions , described by the hamiltonian @xmath34 and coupled by the @xmath0 exchange interaction @xmath35 , @xmath36 where @xmath37 and @xmath38 are the @xmath39-th carrier s and @xmath40-th ion s spin operators , while @xmath41 and @xmath42 are their respective positions . the strength of the @xmath0 exchange interaction between these two spins is described by a smooth function @xmath43 , localized around the @xmath40-th magnetic ion . in the absence of external fields , @xmath34 depends on the carriers degrees of freedom only.@xcite the dynamics of magnetic ions coupled to the system of hole carriers requires a self - consistent description , which takes into account how the holes react to the ions magnetization changes . therefore , we use the lwdin perturbation method specifically adapted for multiparticle hamiltonians,@xcite to derive an effective hamiltonian @xmath44 for ions only . we choose the multiparticle basis states of @xmath45 as @xmath46 . the ion part @xmath47 is an eigenstate of the unperturbed ions - holes system , and the hole part @xmath23 is a slater determinant of eigenstates of the one - particle hole hamiltonian @xmath48 where @xmath49 describes the host band structure and @xmath30 is the spin splitting induced by polarized lattice ions . the lwdin calculus consists in dividing the multiparticle basis states into two subsets , @xmath50 and @xmath51 , @xmath52 set @xmath50 contains all states @xmath53 , where @xmath54 is the @xmath32-particle ground state of @xmath55 . set @xmath51 contains all the remaining states , in which at least one hole is excited above the fermi level . we construct the effective hamiltonian for the states from set @xmath50 only , adding their coupling with set @xmath51 as the second order perturbation , @xmath56 where @xmath57 . the term @xmath58 is independent of the ion configurations , so we set it to zero for simplicity . thus , the effective hamiltonian @xmath44 depends only on the ion degrees of freedom but , thanks to the lwdin method , takes hole excitations into account and can be used to calculate the spin - wave dispersion in a self - consistent manner . the variational part of our method consists in searching for the energy @xmath59 in the range where we expect to find the lowest eigenenergies of @xmath45 , which are the ones that we are interested in . for a known average spin splitting @xmath30 , we can set @xmath59 to the total energy of the hole multiparticle state @xmath54 , @xmath60 , where @xmath61 is the energy of the band with wave vector @xmath62 of hamiltonian @xmath55 , and the sum goes over all occupied eigenstates @xmath63 in @xmath54 . the states @xmath64 are of the form @xmath65 , @xmath66 . to simplify the sum over @xmath64 , we approximate the diagonal matrix element @xmath67 , which depends on both @xmath47 and @xmath68 , by the total energy of the multiparticle hole state @xmath68 , @xmath69 . it describes the interaction of @xmath68 with the average configuration of the ions spins corresponding to the spin splitting @xmath30 . we can thus write the hamiltonian in the following form : @xmath70 the factor @xmath71 under the sum can be written as @xmath72 , where @xmath73 and @xmath74 . since the denominator in eq . is independent of @xmath75 , summing over @xmath75 is equivalent to inserting an identity operator , which allows us to write the last term as @xmath76 we can thus treat @xmath44 as a hamiltonian acting on ion states only , @xmath77 since the @xmath0 exchange term in @xmath78 , which produces the extra - diagonal matrix element @xmath79 , is the interaction of a single hole with an ion , the only @xmath68 states which have a non - zero contribution to the sum over @xmath68 in eq . are those which are created from @xmath54 by just one excitation , @xmath80 . hence , we have @xmath81 and hamiltonian can be written as @xmath82 where @xmath83 is the fermi - dirac distribution coefficient . the fraction in the above sum looks dangerous , as it may diverge in the presence of the energy bands crossings , which would make our perturbation calculus invalid . however , the effective hamiltonian for ions depends on the average of these factors , and will be shown immune to this problem . we write hamiltonian using ion spin operators and integrate out the hole degrees of freedom : @xmath84 the coefficients @xmath85 and @xmath86 are given by @xmath87 where due to the condition @xmath66 in eq . , for @xmath88 the summation goes over @xmath89 , and @xmath90 where @xmath91 for @xmath92 and @xmath93 = \hat s^z$ ] by convention . to obtain the above expressions , we substituted the hole - only part of hamiltonian @xmath94 by the sum of @xmath32 one - particle hamiltonians @xmath49 from eq . . we also used the formula @xmath95 . it arises from the fact that the bloch states @xmath96 building the multiparticle hole states @xmath23 are one - particle eigenstates of hamiltonian @xmath55 , which includes spin splitting @xmath30 . assuming that @xmath97 is a slowly - varying function of @xmath98 and @xmath99 is constant within the unit cell around the @xmath40-th ion and vanishes outside of it , we obtained @xmath100 , where @xmath101 is the crystal volume . the second term of @xmath44 describes the exchange interaction between the lattice ions . in the presence of the spin - orbit coupling it has an antisymmetric part in form of the dzyaloshinskii - moriya interaction , @xmath102 where @xmath103 is the antisymmetric levi - civita symbol with @xmath104 , while @xmath105 is a pseudovector and exists only in the systems with broken inversion symmetry , @xmath106 using the lwdin perturbation - variational calculus , we have thus described the problem as a lattice spin system coupled by exchange interaction the low - lying energy states of such systems are wavelike and can be modeled in the small oscillations approximation . for this purpose , we replace the spin operators with certain nonlinear functions of bosonic creation and annihilation operators @xmath107 and @xmath108 , carrying out the holstein - primakoff bosonization.@xcite next , we approximate these functions with their power expansions around the state of saturation magnetization : @xmath109 leaving in the hamiltonian the terms which are quadratic in creation and annihilation operators , as only these terms influence the dispersion relation . this approximation works very well in the long - wave limit , @xmath110 , as the neglected magnon - magnon interactions are proportional to @xmath111.@xcite furthermore , it justifies neglecting any short - range interactions between localized spins . at the same time , the continuous medium approximation becomes valid , which allows us to carry out the fourier transform @xmath112 . after simple algebraic transformations , we arrive at the final form of the harmonic hamiltonian , @xmath113 \ . \end{split}\ ] ] we call it the _ interaction _ representation as it describes the perturbation of the ground state by the isotropic coulomb interaction ( first term ) and by the spin - orbit interaction , coupling modes of different @xmath114 ( remaining terms ) . the spin susceptibility of the holes is given by @xmath115 where @xmath116 is the density of localized spins @xmath117 in the sample volume @xmath101 and @xmath118 . the presence of the energy denominator shows that @xmath119 corresponds to the second - order part of the hamiltonian . as promised , vanishing of the denominator is not harmful , due to the de lhospital rule . the formula implies that @xmath120 is symmetric in @xmath114 , while @xmath121 , @xmath122 inherit the symmetry of the @xmath123 eigenstates . hence , we expect the latter to be symmetric with respect to @xmath114 for systems which preserve space inversion symmetry,@xcite and otherwise for systems which do not . the @xmath114-independent term describes the interaction of a single magnetic ion with a molecular field arising from the intraband spin polarization of the carriers , @xmath124 the corresponding term reflecting the interband polarization , @xmath125 arises from both @xmath86 and the second part of @xmath85 coefficient , and cancels exactly in the full hamiltonian @xmath126 . hamiltonian in the interaction representation describes the spin system in terms of circularly polarized plane waves ( first term ) , which interact with each other and deform in time ( remaining terms ) . we want to obtain the dispersion relation of independent , stable magnons . for this purpose , we diagonalize @xmath126 by the bogoliubov transformation from @xmath127 to @xmath128 operators , keeping in mind that we deal with the system which breaks the space inversion symmetry . the final form of the effective hamiltonian in the _ spin - wave _ representation reads @xmath129 where excitation modes are spin waves with dispersion @xmath130 in the case of @xmath131 , fulfilled for the systems invariant under space inversion , the above formula simplifies to knig s _ _ et al.__@xcite solution , @xmath132 furthermore , neglecting the spin - orbit coupling , when @xmath133 , bogoliubov transformation is unnecessary , and the effective hamiltonian is already diagonalized by @xmath127 operators . the negative sign of the hamiltonian is a consequence of using the electronic convention to describe the hole - ion system . apart from spin waves , which are the eigenstates of the stationary hamiltonian @xmath126 , dynamic excitations of different physical origin may occur ( e.g. stoner spin - flips ) . they transfer a single carrier across the fermi level to the state excited by the energy @xmath134 ( see the denominator of eq . , and may lead to the spin waves damping . in the presence of the spin - orbit coupling , we find them likely to appear at very low energetic cost throughout the whole @xmath114-vector range . on the other hand , within the mean - field approximation and well below the curie temperature , substitutional and thermal disorder are characterized by a short correlation length @xmath135 , which ensures a well - defined spin - wave excitation spectrum for @xmath136 . figure [ fig : figure1 ] presents a typical spin - wave dispersion spectrum @xmath137 in bulk ( ga , mn)as calculated by the tight - binding computational scheme described in sec . i. the hole concentration @xmath14 equals @xmath138nm@xmath139 and the spin splitting @xmath140ev is applied along the easy axis @xmath141 fixed to the [ 001 ] direction by the biaxial strain @xmath142.@xcite the latter causes a small anisotropy between @xmath137 for spin waves propagating in the @xmath143 $ ] direction and in the @xmath144 plane . furthermore , due to the lack of inversion symmetry in the unit - cell geometry , @xmath145 in general . this means that , contrary to the results of the six - band @xmath2 model,@xcite @xmath146 can be asymmetric with respect to the sign of @xmath114 or @xmath30 ( inset ) . ( this effect was studied experimentally and theoretically in the context of dielectric susceptibility in another zinc - blende crystal , insb.@xcite ) related to the very small dresselhaus spin splitting of the conduction band , the also small @xmath146 asymmetry is observed for all but [ 100 ] and [ 111 ] @xmath114 directions , and decreases with the growth of spin splitting . this result remains in line with our previous studies showing that static properties of ( ga , mn)as like curie temperature or anisotropies , to which @xmath137 belongs , depend mainly on the properties of the carrier bands.@xcite $ ] ( equivalent to @xmath147 $ ] ) and @xmath143 $ ] directions in bulk ( ga , mn)as . inset : the @xmath137 asymmetry due to the bulk inversion asymmetry of gaas lattice modeled by the @xmath1 tight - binding method , as compared to the six - band @xmath2 model preserving the symmetry . lower and upper solid lines denote the [ 412 ] and the opposite @xmath148 $ ] propagation direction . ] figure [ fig : figure2 ] presents @xmath137 in ( ga , mn)as thin layer for @xmath149nm@xmath139 and @xmath150ev along the easy axis [ 110 ] . the host crystal consists of two unstrained infinite monolayers ( ga , as , ga , as ) grown in the [ 001 ] direction . the new effect , related to the structure asymmetry , is a shift of the dispersion minimum to a non - zero @xmath151 value . $ ] direction to @xmath152nm@xmath153 can be observed . ] according to the above plots , @xmath146 for small @xmath114-vectors can be described by the following general formula : @xmath154 where indices @xmath155 denote spatial directions used in the einstein sum convention . the @xmath156 and @xmath157 constants are the spin - wave stiffness tensor and the dzyaloshinskii - moriya coefficient , respectively , while @xmath158 is the spin - wave gap created by magnetocrystalline anisotropies . the higher order terms arising from the bulk inversion asymmetry can be skipped as negligibly small for considered @xmath114-vectors . alternatively , we can expand the @xmath114-dependent terms in @xmath146 , considering their properties implied by eq . , @xmath159 where @xmath160 is the saturation magnetization of the spin system . we obtain that , up to quadratic terms , the spin - wave stiffness in eq . depends only on the @xmath121 term , @xmath161 the above constants are the subject of micromagnetics , which we shall investigate in the next section . it ignores the quantum nature of the atomic matter and uses classical physics in the limit of a continuous medium . the atomic - scale effects investigated in the previous section lead to the wavelike behavior of the spin system . in micromagnetics , these spins are replaced by classical vectors with their slow - varying direction @xmath162 described by the free energy functional : @xmath163 = \int \bigl [ \sum_{j=1}^\infty \mathcal k_j^{\mu } n_{\mu}^{2j } & + \mathcal{a}_{\alpha\beta}^{\mu\nu } \partial_\mu n^\alpha \partial_\nu n^\beta \\ & + u^\mu \epsilon_{\alpha\beta } n^\alpha \partial_\mu n^\beta \bigr ] { \text{d}}^3{\mathbf{r}}\ , \end{split}\ ] ] where indices @xmath155 and @xmath164 denote spatial and magnetization directions , respectively , and @xmath165 is the antisymmetric levi - civita symbol . the first term describes the anisotropy energy , which depends on the orientation of the magnetization with respect to the easy axis @xmath141 . consecutive orders of the magnetocrystalline anisotropy tensor @xmath166 in principal - axis representation are numbered by @xmath40 . the next term is the symmetric exchange energy , where @xmath167 is the exchange stiffness tensor , with @xmath168 . the antisymmetric part of exchange is expressed by the last , dzyaloshinskii - moriya term.@xcite since we describe the magnetization fluctuations around the easy axis , we do not include the derivatives of @xmath169 in the sum , as they are of higher order . the exchange stiffness @xmath167 can be split into two parts . the first one is isotropic in the magnetization direction , but can bear the anisotropy of the exchange interaction in space ( e.g. caused by biaxial strain ) , @xmath170 the remaining part is anisotropic with respect to the magnetization direction , @xmath171 we define indices @xmath172 , referring to corresponding spin components , so that @xmath173 and @xmath174 . it follows from the tensors definitions that @xmath175 and @xmath176 . we shall use the new notation to rewrite the exchange energy , including the dzyaloshinskii - moriya term , in the following form : @xmath177 { \text{d}}^3{\mathbf{r } } \ . \end{split}\ ] ] in analogy to microscopic approach , we transform the exchange energy functional to the reciprocal space , @xmath178 \ . \end{split}\ ] ] we compare the above result to the effective hamiltonian in the interaction representation . it is clear that @xmath179 ( together with @xmath157 ) and @xmath180 correspond to coefficients of the @xmath121 and @xmath181 expansions in eq . , respectively . furthermore , we can identify the components of the above form with different physical mechanisms governing the spin behavior . the term involving @xmath179 describes the energy of a circularly polarized spin wave , as it arises from the isotropic part of exchange interaction . hence , we shall call it the _ isotropic _ exchange stiffness tensor . the two terms involving @xmath180 account for the anisotropic exchange , as we have chosen in eq . , induced by the spin - orbit coupling . hence , we shall call it the _ relativistic _ exchange stiffness tensor . its non - zero elements imply that the tilting of an individual spin from the easy axis @xmath141 to different directions has different energetic cost . as a consequence , the polarization of the spin wave deforms and acquires an elliptical shape . the linear term characterized by the constant @xmath182 represents the minimum shift of the spin - wave dispersion dependence observed in thin ( ga , mn)as layers ( fig . [ fig : figure2 ] ) , associated with the asymmetric exchange of dzyaloshinskii - moriya . the energy @xmath158 in eq . is related to the anisotropy constant @xmath183 in the full free energy functional . we calculate the above tensors for the bulk ( ga , mn)as from fig . [ fig : figure1 ] and two monolayers from fig . [ fig : figure2 ] . for this purpose , we fit the coefficients of the @xmath121 and @xmath181 expansions in eq . on ca 1nm@xmath153 edge cube ( or square ) in @xmath114-space , centered around zero . the obtained tensors describe the energy of spin waves polarized in the plane perpendicular to the easy axis @xmath141 . if they propagate in this plane , we call them longitudinal waves . transverse spin waves propagate along @xmath141 . the dispersion spectrum of spin waves propagating in bulk ( ga , mn)as along two main crystal axes , @xmath184 \parallel x$ ] ( equivalent to @xmath147 $ ] ) and @xmath143 \parallel z$ ] , is presented in fig . [ fig : figure1 ] . the simulated system is biaxially strained , @xmath185 , with the hole concentration @xmath186nm@xmath139 and spin splitting @xmath187ev along the easy axis @xmath188 $ ] . the energy cost of exciting a circularly polarized wave is given by the exchange stiffness tensor @xmath189 it is expressed by a diagonal form with eigenvectors pointing along crystal axes , as the magnetization in the spins ground state is uniform . the difference between its elements reflects the anisotropy of the exchange interaction in space ( between the @xmath190 plane and the growth direction @xmath191 ) caused by the biaxial strain . in the presence of the spin - orbit coupling , the circular polarization can deform into an ellipse . this polarization anisotropy is described by the relativistic @xmath192 tensor , which depends on the mutual orientation of magnetization and spin - wave propagation directions . in our system with the easy axis @xmath141 along the spatial @xmath191 direction , it takes the following form:@xmath193 its zero diagonal component means that the polarization of transverse spin waves is circular , while the non - zero elements indicate that longitudinal spin waves have elliptical polarization , with the shorter axis of the ellipse rotated to the @xmath114 direction . the resulting polarizations are illustrated with fig . [ fig : figure4]a . , in ( a ) the bulk ( ga , mn)as from fig . [ fig : figure1 ] with @xmath188 $ ] and ( b ) after changing @xmath141 to [ 010 ] . ] for an arbitrary propagation direction , the shape of the spin - wave polarization can be calculated from the following polarization matrix : @xmath194 where @xmath195 , or in more detail @xmath196 the ellipse is the solution of the equation @xmath197 where @xmath198 and @xmath199 are the in - plane components of the magnetization unit vector @xmath200 . the ratio of its longer and shorter main axes , @xmath7 and @xmath201 , can be derived from the eigenvalues of the polarization matrix as @xmath202 kinetically , a longitudinal elliptical spin wave can be viewed as a circular one , which experiences the lorentz contraction in the direction of motion , @xmath203 , traveling with the velocity @xmath204 , where @xmath205 is the speed of light . figure [ fig : figure3 ] presents this _ spin - wave relativistic velocity _ in the relevant range of @xmath114 vectors along [ 100 ] and [ 110 ] directions . ( dotted line ) and [ 100 ] ( solid line ) , as a function of the @xmath114-vector magnitude . ] as already mentioned , @xmath180 is induced by the spin - orbit coupling , which connects the symmetries of the lattice with spins . thus , similarly to magnetocrystalline anisotropies , it depends on the magnetization direction with respect to the crystal axes . the @xmath180 tensor calculated in the analyzed ( ga , mn)as system with the easy axis @xmath141 changed to the @xmath147 $ ] direction would acquire the following form : @xmath206 which reveals its dependence on the biaxial strain . ( an appropriate rearrangement of the matrix elements gives the @xmath192 tensor for @xmath141 along [ 100 ] . ) now the longitudinal spin waves propagate in the @xmath207 plane and experience the ` lorentz contraction ' as described above , while the transverse waves propagate along the new easy axis @xmath141 . their polarizations are additionally deformed by the potential of the strained crystal . the latter is stretched equally in the [ 100 ] and [ 010 ] directions and compressed in the [ 001 ] direction , and so are the polarizations . the magnitude of their deformations is given by the @xmath114-independent terms of the polarization matrix @xmath208 . figure [ fig : figure4]b is illustrative of this effect . technically , the relativistic exchange might lead to macroscopic anisotropies in the system , if they did not average out for spin waves propagating in different directions . however , one can imagine a weak anisotropy arising from the described phenomena in asymmetrically shaped samples , where the largest number of similarly polarized modes can exist along the longest dimension . it would then be a candidate for an explanation of the weak uniaxial anisotropy of the [ 100 ] and [ 010 ] crystal axes observed in some ( ga , mn)as samples.@xcite in thin layers of ( ga , mn)as ( fig . [ fig : figure2 ] ) , we observe the minimum shift of the spin - wave dispersion to a non - zero @xmath151 value , which was not present in bulk . it is a hallmark of the dzyaloshinskii - moriya asymmetric exchange , arising from structure inversion symmetry breaking . the mechanism of this interaction is demonstrated in our numerical simulations for the two monolayers of ( ga , mn)as from fig . [ fig : figure2 ] . : 1 s are situated at @xmath209 , 2 at @xmath210 , 3 at @xmath211 and 4 at @xmath212 . the molecular field of magnetic mn ions is introduced according to sec . [ sec : section1 ] . the strongest interactions between the nearest neighbors are marked by solid bonds . the created solid lines define mirror planes in the zinc - blende crystal , ( 110 ) and @xmath213 . since the analyzed structure has two - dimensional periodicity , the carrier momenta in the growth direction are quantized and spin waves propagate in - plane only . ] the structure of the modeled lattice is shown in fig . [ fig : figure5 ] . thanks to the lack of inversion symmetry at the midpoint of each ga - as bond , the exchange interaction of dzyaloshinskii - moriya is allowed , @xmath214 . according to moriya s rules,@xcite the @xmath105 vector of each atom pair , indicated by an arrow at each bond , is perpendicular to their mirror plane . its sense is always the same when we go from ga to as atom . since in our theoretical approach outlined in sec . [ subsec : effhamiltonian ] we have used the holstein - primakoff transformation to describe spin waves as small fluctuations around the magnetization direction @xmath141 , only the @xmath215 terms contribute to the spin - wave dispersion relation . for the above reasons , we expect the maximum effect of the dzyaloshinskii - moriya interaction in systems magnetized along the @xmath216 vectors : perpendicular to @xmath217 $ ] or @xmath218 $ ] . first , we consider the magnetization @xmath141 set along the @xmath217 $ ] direction and a spin wave propagating perpendicular to it , @xmath219 $ ] . the dotted line is the wavefront , along which all spins must be in phase . along this wavefront , the constant @xmath216 vector ( big arrows ) tilts the spins perpendicular to itself and to each other to minimize the energy of the dzyaloshinskii - moriya interaction . the @xmath220 vector is a function of the distance between the spins , @xmath221 . hence , when we move from one spin pair to another in a regular lattice structure , their chirality @xmath222 minimizing the energy is constant in magnitude and antiparallel to @xmath220 . therefore , each spin will be rotated with respect to its neighbors by a constant angle around a constant axis . the rotation direction is parallel or antiparallel to the spin - wave propagation direction @xmath114 , depending on whether @xmath220 is parallel or antiparallel to @xmath141 . in this way , a cycloidal structure with a period @xmath223 forms in the spin system . since the modulation occurs along the @xmath218 $ ] direction , this is where we observe the dispersion minimum shift by @xmath224 ( fig . [ fig : figure2 ] ) . for the magnetization @xmath141 pointing along the @xmath218 $ ] direction , the @xmath216 vectors ( small arrows ) cancel out when looking along this direction , and no frustrated structure of lower energy can arise . since the dzyaloshinskii - moriya interaction operates in the sample plane , neither will we observe its hallmarks when @xmath141 is perpendicular to this plane . for the quantitative analysis of the described effects , we fit the spin - wave dispersion presented in fig . [ fig : figure2 ] with the form . in fig . [ fig : figure6 ] ( full circles ) we sweep the magnetization @xmath141 in the sample plane ( 001 ) and report the obtained angle between @xmath141 and @xmath151 together with the energy gain @xmath225 due to the dzyaloshinskii - moriya interaction , and the magnitude of @xmath151 . from simple algebra we have @xmath226 and @xmath227 . as expected , the energy gain is the largest when @xmath141 and @xmath151 are perpendicular to each other , for @xmath228 $ ] , and drops to zero for @xmath229 $ ] ( fig . [ fig : figure6]a , b ) . the first arrangement is accompanied by the strongest frustration of the spin system , @xmath230nm@xmath153 ( fig . [ fig : figure6]c ) , resulting in the spin cycloid with period @xmath231 nm . in the other arrangement , no spin frustration arises and the collinear spin order is preserved . this behavior accounts for an in - plane anisotropy with the easy and hard axes along the @xmath217 $ ] and @xmath218 $ ] directions , respectively . it is easy to notice that if we mirror - reflect the sample ( or equivalently , move the bottom ga1 layer to the top ga5 position ) , the two axes switch[110 ] will become the hard axis and @xmath218 $ ] the easy axis ( fig . [ fig : figure6 ] , open circles ) . consequently , the sign of @xmath151 and the orientation of the spin cycloid will change . the described anisotropy arises from the net dzyaloshinskii - moriya interaction . although it is a surface effect , it should not be confused with the surface anisotropy , which does not occur in the zinc - blende ( 001 ) surface . in the above reasoning , it is important to remember that we deal with the hole - mediated ferromagnetism . while the magnetic lattice ions substitute only ga sites and would seem oblivious to the inversion asymmetry of the ga - as pairs , the carriers interact with all surrounding atoms . thus , the system of magnetic ions interacting through the hole carriers , as described by eq . , is sensitive to all symmetry properties of the crystal . in two ( ga , mn)as monolayers from fig . [ fig : figure2 ] as a function of the magnetization direction @xmath141 for the analyzed system ( full circles ) and for its mirror - reflection ( open circles ) . a ) the direction of the @xmath151 vector with respect to @xmath141 , b ) energy gain @xmath225 , c ) magnitude of the minimum shift @xmath151 . ] as we have shown above , the dzyaloshinskii - moriya interaction in thin ( ga , mn)as layers leads to the frustration of spins in the ground state . while in bulk ( ga , mn)as the diagonal form of tensor @xmath179 depicts the uniform ground state magnetization , the exchange stiffness tensors in thin layers reveal a spin cycloid oriented along one of the mirror planes : @xmath232 their eigenvectors point along the highest symmetry axes , [ 110 ] and @xmath218 $ ] . this effect is equivalent to applying strain along one of these directions , which would account for a magnetoelastic anisotropy , like the one in biaxially strained samples . however , contrary to the case of a uniformly strained sample , all effects related to the dzyaloshinskii - moriya interaction vanish in the bulk limit . the @xmath179 , @xmath180 and @xmath233 constants let us think of the magnetic system in terms of the interactions by which it is governed , as described by the hamiltonian in the interaction representation . alternatively , if we want to deal with independent stationary magnons ( with already deformed polarization ) , represented by the hamiltonian , we calculate the spin - wave stiffness @xmath156 , related to @xmath179 by the formula . the difference between these two pictures is especially apparent at finite temperatures , where the length of the magnetization vector , @xmath234 , decreases and the formula becomes @xmath235 the higher the temperature , the stronger we have to tilt the magnetization vector in order to excite the quantum of spin waves ( a magnon ) . hence , @xmath236 decreases with temperature faster than @xmath237 . this section discusses the relation between the spin - wave stiffness and curie temperature , based on which we define the normalized spin - wave stiffness parameter , @xmath31 . depending only weakly on the hole density @xmath14 and spin splitting @xmath30 , it makes a convenient tool for experimentalists to estimate the spin - wave stiffness values @xmath238 given the curie temperature and the magnetization of the sample . we provide quantitative numerical results on @xmath31 for bulk ( ga , mn)as obtained in the @xmath1 tight - binding computational scheme ( see sec.i ) , and compare them to the outcome of the previously employed six - band @xmath2 model.@xcite we also clarify that the surprisingly large magnitude of @xmath238 in these systems results from the @xmath14-like character of the periodic part of bloch function . the standard relation between spin - wave stiffness @xmath238 and curie temperature @xmath239 in a cubic crystal , which is often used in the literature on the topic , reads @xmath240 where @xmath241 is the nearest neighbor distance.@xcite derived for short - range interactions , it is interesting to find out how it is modified when considering the actual nature of the spin - spin exchange . this question has a number of implications . for instance , both carrier relaxation time , which is limited by magnon scattering , and the quantitative accuracy of the mean - field approximation grow with @xmath242 , as the density of spin waves at given temperature @xmath243 diminishes when @xmath238 increases . it is worth noting that a simple adaption of this formula to ( ga , mn)as by replacing @xmath241 with @xmath244 resulted in an overestimation of the @xmath0 exchange integral by an order of magnitude.@xcite allowing for spatially modulated structures , the magnetic ordering temperature @xmath239 for a carrier - controlled ferromagnet is given by a solution of the mean - field equation,@xcite @xmath245 where @xmath246 and @xmath247 are the carrier and lattice ion spin susceptibilities . in the simple case of a parabolic band they are proportional to the pauli and curie - weiss magnetic susceptibilities , respectively . first , we consider the two - band model of carriers residing in a simple parabolic band , where all that is left of the @xmath248 tensor of eq . is the scalar isotropic exchange stiffness @xmath50 . according to the previous section , it is related to the quadratic coefficient of the expansion @xmath249 , by @xmath250 . additionally , we assume that the ground state of the system corresponds to the uniform ferromagnetic ordering , @xmath251 , and take @xmath252 in the curie form , @xmath253 using eqs . and , we obtain @xmath254 if the values of both spin splitting @xmath30 at @xmath255 and @xmath256 are much smaller than the magnitude of the fermi energy @xmath257 , the carrier susceptibility is given by @xmath258 the total density of states at @xmath259 is @xmath260 , where @xmath261 is the carrier effective mass , and @xmath262 is the lindhard function . we obtain from these equations @xmath263 we see that , in agreement with the notion that magnetic stiffness increases with the range of the spin - spin interaction , @xmath241 in eq . is replaced by @xmath264 in eq . , which scales with the range of the carrier - mediated interactions . indeed , according to the ruderman - kittel - kasuya - yosida theory,@xcite the magnitude of the ferromagnetic exchange integral decays at small spin - spin distances @xmath265 as @xmath266 and reaches the first zero at @xmath267 . since in semiconductors @xmath268 , the above formulae imply that @xmath242 is rather large in systems with carrier - controlled ferromagnetism . a question arises as to how the ratio @xmath269 would be modified , if the complexities of the valence band were taken into account . as already noticed by knig _ et al._,@xcite the values of exchange stiffness for ( ga , mn)as are greater in the six - band model with a spin - orbit coupling than in the case of a simple parabolic band . as argued by these authors , due to the multiplicity of the valence bands , the fermi level for a given carrier concentration is much lower than in the two - band model , and hence both carrier polarization and exchange stiffness are greater.@xcite on the other hand , brey and gmez - santos@xcite assign large values of @xmath238 to higher magnitudes of @xmath239 in the multi - band model . to check these suggestions we plot in fig . [ fig : figure7 ] the values of dimensionless parameter @xmath270 for various hole concentrations @xmath14 and @xmath271 . the ratio @xmath242 for bulk ( ga , mn)as is computed using the @xmath1 tight - binding ( sec . [ sec : section1 ] ) and the previously employed@xcite six - band @xmath2 model of the semiconductor band structure . guided by the results of the two band model , one could expect @xmath272 in the limit of small spin polarizations , @xmath273 . in contrast to these expectations , @xmath31 is of the order of eleven and , moreover , weakly dependent on the polarization , altered here by changing the hole concentration and the spin - orbit splitting . furthermore , the experimentally observed biaxial strain magnitudes have only slight effect on the stiffness tensor ( see fig . [ fig : figure1 ] and eq . .@xcite this indicates that the single parameter @xmath31 can serve to estimate the magnitudes of @xmath238 and @xmath50 if only the mn magnetization @xmath47 and hole concentration @xmath14 are known . in bulk ( ga , mn)as as a function of the hole concentration @xmath14 and the spin - orbit coupling ( inset ; the value of the spin - orbit coupling in ( ga , mn)as is indicated by dotted line ) for different spin splittings @xmath30 . filled and empty markers indicate the results of the @xmath1 tight - binding and the six - band @xmath2 model , respectively . ] knowing that neither the spin - orbit coupling nor the multiplicity of carrier bands can explain the large spin - wave stiffness , we turn our attention to the matrix elements @xmath274 in the spin susceptibility of holes . neglecting the spin - orbit coupling , @xmath275 is a product of the spin part @xmath276 and the real spatial part @xmath277 . thus , we can write @xmath278 for @xmath279 . in the parabolic two - band model with its carrier wave functions described by plane waves @xmath280 , the periodic part @xmath281 . hence @xmath282 and @xmath283 , and @xmath284 simplifies to the lindhard function . more realistic models take into account the periodic lattice potential , which mixes different atomic orbitals and leads to the @xmath62-dependent modulation of @xmath275 . ( the eventual composition of hole states in the @xmath1 tight - binding model is presented in fig . [ fig : figure8]a - c . ) from this it follows that the scalar products @xmath285 are @xmath114-dependent , and can be only smaller than in the simple model . as a consequence , @xmath286 has a steeper slope , as presented in fig . [ fig : figure8]d . this explains why the magnitude of the stiffness tensor is so large in ( ga , mn)as and related ferromagnets . actually , the fact that the @xmath14-type character of the carrier wave functions affects in this way the @xmath287-dependence of dielectric susceptibility has been already noted.@xcite -vector . it is given by the squares of norms of projections of the carrier state on atomic @xmath288 , @xmath14 , @xmath16 or @xmath289 orbitals of as or ga atoms ( solid and dotted lines ) . panel d presents the lindhard function @xmath290 calculated in the tight - binding model ( @xmath186nm@xmath139 and @xmath140ev , with the spin - orbit coupling set to zero ) and in the parabolic band model with the same fermi energy @xmath291ev . dashed lines are their square form fits . ] in this section , we address the problem of applicability of the mean - field model to the description of temperature dependence of magnetization in the analyzed systems . while it is a known fact that the critical fluctuations do not change much the mean - field curie temperature @xmath292 in the presence of long - range exchange interactions,@xcite several papers@xcite discuss how it is lowered by spin waves . also , the influence of dilution and disorder on spin waves and the curie temperature , which can be crucial in samples with low mn content , was studied by many authors with varying conclusions.@xcite in response , we propose a simple model which takes into account the correct number of spin - wave excitations and thermal depolarization of spins . however , we argue that spin waves also do not lower the mean - field @xmath292 due to thermal decoherence of quantum spin system . the mean - field approximation employed in our model allows for a reasonable overall description of ferromagnetism in the analyzed systems . it reduces the problem of lattice spins coupled by the exchange interaction to that of noninteracting spins in the molecular field @xmath293 . their magnetization is described by the self - consistent equation , @xmath294 the brillouin function @xmath295 ignores the actual nature of thermal fluctuations and their correlations , and assumes that every spin fluctuates independently , which in our theory corresponds to the high-@xmath287 limit of spin - wave excitations ( see fig . [ fig : figure1 ] ) . at low temperatures , however , the long - wavelength magnons of much lower energies can exist . since each of them carries a moment of @xmath296 , they yield a strong contribution to the temperature dependence of magnetization,@xcite @xmath297 where @xmath298 is the thermal average of spin - wave excitations in each mode . the latter can be modeled by the bose - einstein distribution , @xmath299 , as spin waves are bosons . then , replacing the summation in eq . by an integral and putting @xmath300 , one obtains the well known @xmath11 bloch law@xcite @xmath301 where @xmath302 is the riemann zeta function . remembering about the spin - wave gap created by magnetic anisotropy , which allows for ferromagnetism in low dimensional systems despite the mermin - wagner theorem,@xcite we have @xmath303 . ( we neglect the dzyaloshinskii - moriya coefficient , which vanishes in bulk limit . ) in that case , the above law is modified to@xcite @xmath304 where @xmath305 is de jonquire s function . furthermore , the bose - einstein statistics allows for the unlimited number of spin waves in each mode ( on the other hand , knig _ et al._@xcite assume that there can be only @xmath306 spin waves of each @xmath114 , which is too strict ) . in fact , their total number can not be larger than @xmath307 , corresponding to complete magnetization reversal . we handle this by introducing a fictious mode with zero energy , which is ` occupied ' by the spin waves which have not been excited in reality . in this way we can treat the problem with classic bose - einstein condensate methods:@xcite the total number of bosons occupying all modes is always @xmath307 , their zero - energy mode constitutes the ` condensate ' phase , while the excited spin waves constitute the ` thermal cloud ' . the total number of spin waves in the limit of infinite crystal volume is therefore given by @xmath308 additionally , while at zero temperature the system is described by a pure state @xmath309 , where exciting a spin wave costs the energy @xmath310 at non - zero temperatures it is described by a mixture of pure states @xmath311 with a certain number of spins flipped by one - particle thermal excitations , @xmath312 . thus , its magnetization drops to @xmath234 according to the mean - field brillouin function . the energy cost of exciting a spin wave is now given by @xmath313\ .\ ] ] since the spin - wave dispersion @xmath146 depends on temperature @xmath243 almost exclusively via the spin splitting @xmath30 and is approximately proportional to it , we can estimate the above expression by @xmath314 , where @xmath315 is the magnetization in the state @xmath311 . spin waves are thus perturbations of the thermal state of lattice spins and not of the ground state , which we can model just by replacing @xmath316 with @xmath234 ( or @xmath30 with @xmath317 ) in the dispersion relation @xmath137 . they additionally lower the magnetization to @xmath318 . remaining in the limit of small oscillation approximation , we obtain the following set of equations : @xmath319 where @xmath320 . the spin splitting @xmath317 is induced by the mean field of the lattice spins described by the first equation , while @xmath321 additionally takes into account their depolarization due to spin waves . , calculated by the mean - field model , @xmath11 bloch law and its modified version , and the ` spin - wave condensate ' . the experimental dependence for the sample from ref . is analyzed in sec . [ sec : experimental ] . ] figure [ fig : figure9 ] presents the temperature dependence of magnetization as described by the above methods for the bulk ( ga , mn)as from fig . [ fig : figure1 ] . the @xmath11 bloch law and its modified version apply to low temperatures only . we see that the values of @xmath238 determined neglecting the anisotropy gap would be overestimated . the ` spin - wave condensate ' includes the magnon contribution to magnetization in the whole temperature range , and leads to significantly lower curie temperature than in the mean - field model . this method uses the correct bound for the number of spin - wave excitations . it also naively attempts to solve the problem of the well - known shortcomings of the spin - wave theory@xcite introduced by the holstein - primakoff transformation , as it includes thermal disorder by depolarizing the lattice spins with temperature . however , quantum - mechanical intuition suggests that the spin waves should vanish completely at higher temperatures . non - zero temperature leads to the loss of information about the system : pure states are replaced by mixed states , namely the thermal states @xmath322 , and the quantum correlation of spins vanishes . this can be reflected in the @xmath0 hamiltonian by replacing @xmath323 with @xmath324 , which is simply the mean - field model . similarly , in the vicinity of the curie transition , where the problem of critical fluctuations arises , the mean - field models of systems with long - range exchange interactions are known to work very well.@xcite the next section provides further insight into the applicability of the presented models by comparing their results to experiment . as already mentioned in sec . i , the theory of exchange stiffness developed within the six - band @xmath0 zener model@xcite describes quantitatively the width of stripe domains in ( ga , mn)as.@xcite recently , gourdon _ et al._@xcite carried out a detailed analysis of the magnetic domain structure and magnetic properties of an annealed 50nm - thick ga@xmath12mn@xmath13as layer with a perpendicular magnetic easy axis and the curie temperature of 130k . two employed experimental methods yielded an upper and lower limit of the isotropic exchange stiffness @xmath50 as a function of temperature @xmath243 . as found by these authors from examining the domain - wall velocity , the higher values of @xmath325 , determined from the lamellar domain width , are reliable . we model the sample in the tight - binding computational scheme for bulk ( ga , mn)as ( see sec . [ sec : section1 ] ) . in order to determine the material parameters for numerical simulations , we start by estimating the effective mn content @xmath326 from the measured low - temperature spontaneous magnetization @xmath327ka / m . taking into account the hole contribution , @xmath328ka / m,@xcite implies the magnetization of the mn spins @xmath329ka / m . this value corresponds to the effective mn content @xmath330 and the spin splitting @xmath140ev . no direct measurements of the hole concentration are available for this sample , so we estimate its magnitude from the effective and total mn content , @xmath326 and @xmath331 . assuming that interstitial mn donors had survived the annealing process and formed antiferromagnetic pairs with the substitutional mn acceptors,@xcite we obtain @xmath332nm@xmath139.@xcite from fig . [ fig : figure7 ] , for the given value of @xmath14 and @xmath30 we find @xmath333 , which gives the spin - wave stiffness @xmath334mevnm@xmath335 ( @xmath336pj / m ) at @xmath337k . knowing this , we can calculate the temperature dependence of magnetization according to sec . [ subsec : magntemp ] ( fig . [ fig : figure9 ] ) . in the mean - field picture ( solid lines ) , the magnetization of lattice spins @xmath338 is described by the brillouin function . the magnitude of hole magnetization @xmath339 decreases with temperature proportionally to @xmath338 , according to eq . . the resulting magnetization @xmath234 is compared to the experimental curve ( circles ) . we obtain a good agreement with the measured data , especially near the curie transition , and @xmath340k . at low temperatures , we plot the outcomes of the @xmath11 bloch law and its modified version employing the calculated spin - wave stiffness value @xmath238 ( dotted lines ) . the modified bloch law , adjusted to include the spin - wave gap , gives very good agreement with the experimental dependence , which indicates that in this regime the spin - wave excitations are solely responsible for demagnetization . near the curie transition , we reconstruct the measured curie temperature and magnetization values with the brillouin function , which suggests that the temperature destroys the spin - wave coherence and recalls the mean - field picture . as a function of temperature @xmath243 obtained in ref . from the analysis of the lamellar domain width in ga@xmath12mn@xmath13as . ] to reconstruct the @xmath325 trend obtained from the magnitudes of lamellar domain width,@xcite we again make use of the fact that the susceptibility @xmath341 depends on temperature almost exclusively via spin splitting . thus , and according to the formula , the exchange stiffness scales with temperature as @xmath342.@xcite we use the experimentally determined @xmath234 and the calculated @xmath343pj / m to estimate the exchange stiffness values for the remaining temperatures as @xmath344 . as shown in fig . [ fig : figure10 ] , this procedure correctly reproduces the experimental @xmath325 trend . however , the same value of @xmath345 , which has been shown to successfully describe the experimental @xmath234 dependence with the modified bloch law , is twice as large as the exchange stiffness constant determined from the measured lamellar domain width . potashnik _ et al._@xcite evaluated the isotropic exchange constant @xmath346 for the set of optimally annealed ( ga , mn)as layers with varying mn content . the values derived from the temperature dependence of magnetization , using the standard bloch law , and from the curie temperature within the three - dimensional heisenberg model,@xcite were similar . for samples with varying effective mn content @xmath326 , as determined from the temperature dependence of magnetization using the @xmath11 bloch law ( large crosses ) and from the curie temperature ( small crosses ) in ref . . empty circles indicate the region where the experimental @xmath292 exceeds the theoretical predictions . inset : experimental temperature dependence of magnetization for the sample with @xmath347 fit by the standard bloch law with @xmath348 mev ( ref . ) , the modified bloch law with the theoretically obtained @xmath349 mev and by the mean - field brillouin function . ] our theoretical reconstruction of the experiment by the @xmath1 tight - binding model of bulk ( ga , mn)as is demonstrated in fig . [ fig : figure11 ] . to obtain the presented results , we have estimated the effective mn content @xmath326 from the measured low - temperature magnetization @xmath350 , and then increased it by about @xmath351 to include the hole contribution.@xcite assuming that annealing removed all interstitials and each remaining substitutional mn produces one hole carrier , the hole density @xmath352 . for the obtained values of @xmath14 and @xmath326 we have calculated the mean - field curie temperatures @xmath239 and the exchange constant @xmath353 assuming that mn ions form a cubic lattice . we note that this form is equivalent up to a few percent to that employed in ref . : @xmath354 , where @xmath25 is the total mn content ( n. samarth , private communication ) . since for the samples with low mn content our values of @xmath239 are much lower than the experimental ones , we conclude that the corresponding estimates of @xmath346 ( indicated by empty circles ) are not reliable . on the other hand , we reconstruct curie temperatures for samples with @xmath355 ( filled circles ) , but this time the theoretical values of @xmath346 are much smaller than the experimental ones . this discrepancy points to the importance of the anisotropy - induced energy gap in the spin - wave spectrum . as illustrated in fig . [ fig : figure11 ] ( inset ) for the sample with @xmath347 , fitting the experimental @xmath234 curves with the standard bloch law neglecting the gap leads to higher values of @xmath346 than those expected theoretically . at the same time , the modified bloch law employing the theoretical exchange constant @xmath349 mevnm@xmath335 reconstructs the analyzed @xmath234 trend . we notice that it perfectly describes the mild slope of the low - temperature @xmath234 , contrary to the standard bloch law.@xcite similarly to the case of the gourdon _ et al . _ experiment,@xcite the mean - field model works very well at higher temperatures . in a series of experiments , the spin - wave stiffness was determined by examining spin precession modes excited by optical pulses@xcite and under ferromagnetic resonance conditions.@xcite according to these works , the experimental findings are strongly affected by gradients of magnetic anisotropy , presumably associated with carrier depletion at the surface and interface , which also affect the character of spin pinning . we also note that no influence of the magnetic field on the hole spins , visible as a deviation of the land factor from the value @xmath356,@xcite was taken into account in the employed landau - lifshitz equations . with these reservations we show in fig . [ fig : figure12 ] the experimentally evaluated values of @xmath238 plotted as a function of the nominal mn concentration @xmath25 for various as - grown and annealed samples of ga@xmath3mn@xmath10as . these findings are compared to the results of _ ab initio _ computations@xcite ( dashed line ) and our theory for the hole concentration @xmath357 and @xmath358 ( solid lines ) . when comparing theoretical and experimental results , one should take into account that the actual mn concentration @xmath326 is smaller than @xmath25 , particularly in as - grown samples . as seen , our theory describes properly the order of magnitude of the spin - wave stiffness @xmath238 but can not account for a rather large dispersion in the experimental data . as a function of the nominal mn concentration @xmath25 in ga@xmath3mn@xmath4as . solid symbols denote annealed as well as etched and hydrogenated samples ( ref . ) . empty symbols denote as - grown samples . the value @xmath356 was used to convert @xmath238 in magnetic units ( tnm@xmath335 ) to energy units ( mevnm@xmath335 ) . ] in thin ( ga , mn)as layers , described in sec . [ sec : layers ] , the dzyaloshinskii - moriya interaction can be observed in form of a spin - wave dispersion minimum shift . it leads to the formation of a cycloidal spin structure and uniaxial in - plane anisotropy of the [ 110 ] and @xmath218 $ ] directions , with the easy axis determined by the sample geometry . both these phenomena have been a subject of considerable interest in recent years.@xcite while the long - period spin structures have not been hitherto observed , the uniaxial anisotropy is commonly present in ( ga , mn)as.@xcite however , it was shown by gradual etching@xcite or by investigating different samples@xcite to be insensitive to the layer thickness . this is not the case for the anisotropy caused by the dzyaloshinskii - moriya interaction , which is a surface effect and vanishes with increasing layer thickness . to test our theory , one should pick very thin high - quality samples grown in the [ 001 ] direction , which additionally excludes the anisotropy of the surface , with the easy axis along one of the in - plane diagonal directions . the cycloidal spin structure could be then observed e.g. under a magnetic force microscope or by neutron scattering . the sample should be probed along the [ 110 ] and @xmath218 $ ] directions to find the long - period modulation of magnetization only along the one perpendicular to the easy axis . we have investigated spin waves and exchange stiffness in thin layers and bulk crystals of ferromagnetic ( ga , mn)as described by the @xmath1 tight - binding computational scheme . using the proposed variational - perturbation calculus , we have described the analyzed systems and their spin - wave excitations . their properties have been expressed by the phenomenological parameters of micromagnetic theory . we have noticed that the strength of ferromagnetic order described by the isotropic exchange stiffness is significantly amplified by the @xmath14-like character of carrier wave functions , as compared to the simple parabolic band model . furthermore , we have found various effects reflecting the tendency of the spin - orbit interaction to pervade every aspect of carrier dynamics . they produce the relativistic corrections to spin - waves given by the anisotropic exchange stiffness tensor and the asymmetric dzyaloshinskii - moriya coefficient . the latter accounts for the cycloidal spin arrangement and the accompanying uniaxial in - plane anisotropy of diagonal ( [ 110]/@xmath218 $ ] ) directions in thin layers , which can result in a surface - like anisotropy in thicker films . quantitative results on the stiffness constant have been provided in form of a normalized parameter , which assumes the value @xmath359 over a wide range of mn and hole concentrations in ( ga , mn)as . they agree with the previous @xmath2 calculations@xcite but predict significantly smaller values of spin - wave stiffness than those resulting from _ ab initio _ computations.@xcite the above theories have been applied to analyze the related experimental data on the stiffness parameter and the temperature dependence of magnetization . our basic theoretical model has not managed to reconstruct all stiffness values obtained by various experimental methods . in all cases we have reconstructed the entire range of magnetization dependence on temperature . at low temperatures , it can be understood within the the modified bloch law@xcite employing the values of spin - wave stiffness calculated by our model . at higher temperatures , the mean - field theory becomes justifiable owing to thermal decoherence and the long range character of spin - spin interactions . at the same time , we have reproduced only partly the stiffness values obtained from analyzing precession modes in ( ga , mn)as thin films . our results may allow to separate bulk and surface effects , as well as bring to light the pining phenomena and the role of inhomogeneities in experiments examining precession modes in slabs of carrier - controlled ferromagnetic semiconductors . a.w . acknowledges support by the president of polish academy of sciences and ec network semispinnet ( pitn - ga-2008 - 215368 ) , and t.d . acknowledges support from the european research council within the `` ideas '' 7th framework programme of the ec ( fundms advanced grant ) .

understanding of the kinematic evolution of coronal mass ejections ( cmes ) in the heliosphere is important to estimate their arrival time at the earth . it is found that kinematics of cmes can change when they interact or collide with each other as they propagate in the heliosphere . in this paper , we analyze the collision and post - interaction characteristics of two earth - directed cmes , launched successively on 2012 november 9 and 10 , using white light imaging observations from stereo / secchi and in situ observations taken from wind spacecraft . we tracked two density enhanced features associated with leading and trailing edge of november 9 cme and one density enhanced feature associated with leading edge of november 10 cme by constructing j - maps . we found that the leading edge of november 10 cme interacted with the trailing edge of november 9 cme . we also estimated the kinematics of these features of the cmes and found a significant change in their dynamics after interaction . in in situ observations , we identified distinct structures associated with interacted cmes and also noticed their heating and compression as signatures of cme - cme interaction . our analysis shows an improvement in arrival time prediction of cmes using their post - collision dynamics than using pre - collision dynamics . estimating the true masses and speeds of these colliding cmes , we investigated the nature of observed collision which is found to be close to perfectly inelastic . the investigation also places in perspective the geomagnetic consequences of the two cmes and their interaction in terms of occurrence of geomagnetic storm and triggering of magnetospheric substorms .

1408.0352

coronal mass ejections ( cmes ) are huge magnetized plasma expulsions from the sun into the heliosphere and are considered as primary drivers of geomagnetic storms and various other space weather effects @xcite . it is therefore required to improve the physics and our understanding of evolution of cmes in the solar wind , which is found to be dependent on heliospheric environment @xcite . various techniques based on stereoscopic observations which estimate the 3d kinematics of cmes in the heliosphere have been developed @xcite . using these techniques , association of remote observations of cmes with in situ observations has been carried out by @xcite . also , several models have been developed for predicting the arrival time of cmes at 1 au @xcite . sometimes , cmes are observed to erupt in quick succession which under certain favorable initial conditions , can interact or merge with each other during their propagation in the heliosphere . in such a scenario , space weather prediction schemes may not be successful without taking into account the post - interaction cmes characteristics . the compound streams ( interaction of cme - cir or cme - cme ) were first inferred by @xcite using observations from helios and iseee 3 spacecraft . their study suggested that a large geomagnetic storm may be associated with compound streams . the observational evidence for merging of cmes ( or cannibalism ) was first given by @xcite based on the long - wavelength radio and soho / lasco observations . thus , these interactions are of great importance from space weather point of view . before the era of wide angle imaging far from the sun , the understanding of involved physical mechanisms in cme - cme or cme - shock interaction was achieved mostly from magnetohydrodynamic ( mhd ) numerical simulations of the interaction of a shock wave with an magnetic cloud ( mc ) @xcite , the interaction of two ejecta @xcite , and the interaction of two mcs @xcite . @xcite have shown that a forward shock can cause an intense southward magnetic field of long duration in the preceding magnetic cloud . such modifications in the preceding cloud are important for space weather prediction . therefore , events involving cme - cme and cme - shock interactions are important candidates to investigate their kinematics and arrival time at near earth . after the launch of _ solar terrestrial relation observatory ( stereo ) _ in 2006 @xcite , it is possible to continuously image the cmes and to determine the 3d locations of their features in the heliosphere and hence have direct evidence of cme interaction using images from heliospheric imager ( hi ) on sun earth connection coronal and heliospheric investigation ( secchi ) package @xcite . however , immediately after the launch of stereo , during deep extended solar minimum , not many interacting cmes were observed . as the current solar cycle progressed , interaction of cmes now appears a fairly common phenomenon , in particular around solar maximum . the interacting cmes of 2010 august 1 have been studied extensively by using primarily the stereo white light imaging , near earth in situ and , s - waves radio observations @xcite . also , @xcite have reported a clear deflection of 2010 may 23 - 24 cmes after their interaction . as mentioned above , therefore , only a few studies on cme interaction have been reported . several key questions regarding cme interaction need to be addressed , which are not well understood quantitatively , viz . ( 1 ) how the dynamics of cmes change after interaction ? what is the regime of interaction , i.e. elastic , inelastic or super - elastic ? @xcite ( 2 ) what are the consequences of the interaction of cme - shock structure ? how the overtaking shock changes the plasma and the magnetic field properties into preceding magnetic cloud ? @xcite ( 3 ) what are the favorable conditions for the cme cannibalism and the role of magnetic reconnection in it ? @xcite ( 4 ) what is the possibility for production of reverse shock at cme - cme interaction site ? @xcite ( 5 ) whether these interacted structures produce different geomagnetic consequences than individual cmes , on their arrival to magnetosphere ? @xcite . ( 6 ) what are the favorable conditions for the cme s deflection and enhanced radio emission during cme - cme interaction in light of aforementioned questions and only a few studies reported , it is clear that the prediction of arrival time of interacting cmes and association of remote observations of such cmes with in situ is challenging . in the present study , we focus on the identification , evolution and propagation of two cmes launched on 2010 november 9 and 10 , as they travel from the corona to the inner heliosphere . this study , in which the launch time of the selected cmes are separated by about 14 hr , gives an opportunity to understand the cme - cme interaction unambiguously for simpler scenario , contrary to complex events around 2010 august 1 , wherein four cmes were launched in quick succession . we find that cmes of november 9 and 10 are directed towards the earth and their initial characteristics indicate a probability of their interaction . in section [ intcmes ] , observations and estimated kinematics of these cmes are presented to investigate if they interacted during their propagation in the heliosphere . in section [ momentum ] , nature of collision , momentum and energy transfer during collision phase of cmes are described using their estimated true mass and kinematics . in section [ insitu ] , in situ observations of remotely tracked cme features are described . using the derived kinematics combined with drag based model ( dbm ) of @xcite , the arrival time and transit speed of tracked features of cmes at l1 is predicted in section [ arrtime ] . the geomagnetic response of interacted cmes structures is described in section [ geomagnetic ] . in section [ results ] , the results and discussions of the present study are emphasized and conclusions are given in section [ conclu ] . secchi suite consists of an extreme ultraviolet imager ( euvi ) , two coronagraphs ( cor1 and cor2 ) and , two heliospheric imagers ( hi1 and hi2 ) . these instruments can image the heliosphere from near the sun to the earth and beyond . the euvi ( field - of - view ( fov ) : 1 - 1.7 @xmath0 ) , cor1 ( fov : 1.5 - 4 @xmath0 ) and cor2 ( fov : 2.5 - 15 @xmath0 ) fov are centered on the sun . the hi1 and hi2 have their optical axis aligned in the ecliptic plane and are off - centered from the sun at solar elongation of @xmath1 and @xmath2 , respectively . hi1 and hi2 have wide fov of @xmath3 and @xmath4 , respectively . with this identical imaging package ( secchi ) on twin spacecraft , a cme can be tracked from 0.4 to @xmath5 elongations . at the time of observations around november 9 , the stereo a & b were @xmath6 westward and @xmath7 eastward from the sun - earth line at a distance of 0.96 au and 1.08 au from the sun , respectively . the in situ properties were investigated through data collected by instruments on - board wind spacecraft @xcite . in order to determine the true direction and speed of cmes launched on november 9 and 10 in quick succession , 3d reconstruction technique is applied to cme features close to the sun . we find that the cme of november 9 ( hereinafter referred as , cme1 ) has a slow speed than fast speed november 10 cme ( hereinafter referred as , cme2 ) and both move in a similar direction towards the earth . these observations indicate the probability of their interaction in the heliosphere . using secchi / hi observations , we attempt to investigate if these cmes indeed interacted with each other in the heliosphere . we also attempt to identify various cme structures in the in situ observations taken at 1 au . a partial halo cme1 ( angular width of @xmath8 ) was observed in soho / lasco - c2 fov at 15:12 ut on 2012 november 9 ( http://cdaw.gsfc.nasa.gov/cme_list/ ) . in the secchi / cor1-b and a , this cme was observed at 14:25 ut in the sw and se quadrants , respectively . another partial halo cme2 ( angular width @xmath9 ) was observed in lasco c2 fov at 05:12 ut on november 10 . this cme was detected by secchi / cor1-b and a in the sw and se quadrants , respectively , at 05:05 ut on november 10 , 2012 . figure [ scc ] shows true kinematics of cme1 and cme2 from 3d reconstruction of selected features along the leading edges of corresponding cmes using tie - pointing procedure ( scc_measure : @xcite ) on secchi / cor2 data . this tie - pointing procedure uses the concept of epipolar geometry @xcite and due to epipolar constraints , the identification of same feature in two images ( from stereo - a and b ) is always possible along the same epipolar line in both images . before implementing this technique , the processing of secchi / cor2 images and creation of minimum intensity images and then its subtraction from the sequence of processed cor2 images was carried out following the approach described in @xcite . as the separation angle of the stereo spacecraft is large , it is better to use two independent methods to confirm the results of 3d reconstruction . therefore , we have also used the visual fitting of both cmes in the secchi / cor2 fov using the graduated cylindrical shell ( gcs ) model @xcite for the 3d reconstruction of these cmes . this model represents the flux rope structure of cmes with two shapes : the conical legs and the curved ( tubular ) fronts , also known as the `` hollow croissant '' . we used contemporaneous image triplets of cme1 around 17:39 ut on november 9 from stereo - a / cor2 , stereo - b / cor2 and soho / lasco - c3 . the best visual fit for cme1 is found in the direction of w02s14 , with a half angle of 19@xmath10 , tilt angle of 9@xmath10 around the axis of symmetry of the model ( i.e. rotated 9@xmath10 anticlockwise out of the ecliptic plane ) , and an aspect ratio of 0.52 . at this time , cme1 was at a distance of about 9.6 @xmath0 from the sun . using the obtained fitted values of half angle and aspect ratio of cme1 , its 3d face - on and edge - on angular width is estimated as 100@xmath11 and 62@xmath11 , respectively . we also carried out the visual fitting using gcs model for cme2 around 06:39 ut on november 10 by exploiting the concurrent image triplets of stereo - a / cor2 , stereo - b / cor2 and soho / lasco - c3 . the best fit for cme2 is obtained in the direction of w06s25 , at a tilt angle of @xmath12 around the axis of symmetry of the model , with a half angle of @xmath13 , and an aspect ratio of 0.19 . at this time cme2 was at a distance of 8.2 r@xmath14 from the sun . using the gcs model fitted values , the 3d face - on and edge - on angular width of cme2 is estimated as 46@xmath11 and 22@xmath11 , respectively . the gcs model fit for cme1 and cme2 is shown in the top and bottom panel of figure [ fm ] , respectively . we find that the estimated latitude and longitude of both cmes from gcs model and tie - pointing procedure are in agreement within few degrees . this shows that the results of both tie - pointing and gcs model are reliable and can be used for estimating kinematics of cmes in the coronagraphic fov . from a study of 3d kinematics of the two cmes using tie - pointing approach ( scc_measure : @xcite ) , we found that cme1 is slow with a true speed of 620 km s@xmath15 around 15 r@xmath14 while cme2 is faster with a speed of 910 km s@xmath15 at a distance of approximately 15 r@xmath14 ( figure [ scc ] ) . latitude and longitude of reconstructed features for both cmes suggest that these were earth - directed and could possibly interact in the interplanetary ( ip ) medium . if the speed of cmes is assumed to be constant beyond the outer edge of cor2 fov , then these cmes should have collided at approximately 130 r@xmath14 on november 11 around 02:30 ut . however , previous studies have shown that the speed of a cme can significantly change after the coronagraphic fov @xcite . therefore , further tracking of cme features in the heliosphere is required to determine the exact location and time of interaction of these cmes . evolution of cme1 and cme2 in the running difference images of cor2 , hi1 and hi2 fov is shown in figure [ evolution ] . to track and estimate the arrival times of the cmes in the heliosphere using hi1 & 2 images , we constructed j - maps based on the method developed by @xcite and derived the variation of elongation of selected features with time . the details about extraction of strip of constant position angle , selection of bin size and adopted procedure for construction of j - maps and derivation of elongation angle from it has been described in section 3.1.1 of @xcite . the positively inclined bright features in the j - maps ( figure [ j - maps ] ) correspond to enhanced density structure of the cmes . we tracked the leading and trailing edge ( marked in green and red ) of bright features corresponding to slow cme1 and leading edge ( blue ) of bright feature corresponding to the fast cme2 . prior to interpreting the tracked features of cme1 in j - map as its leading and trailing edge , respectively , the derived elongations corresponding to all the three tracked features were overplotted on the base difference hi1 images . on careful inspection of the sequence of these images , we noticed that the two tracked features of cme1 correspond to first density enhancement at the front and second density enhancement at the rear edge of cme1 , respectively . the base difference hi1-a images with overplotted contours of elongation corresponding to tracked features of cme1 and cme2 , are shown in figure [ basediff ] . it is to be noted that the term ` trailing edge ' of cme1 does not correspond to the rear most portion of cme1 but corresponds to the part behind the front and is not an image artifact in j - map constructed from running difference images . since , the tracked features of cme2 were not observed well in stereo b ecliptic j - map , we could not implement the stereoscopic reconstruction technique to estimate the cme kinematics . instead , we used harmonic mean ( hm ) approximation which requires single spacecraft observation to derive the elongation variations with time for the tracked features which could then be converted into heliocentric distances from the sun @xcite . in the hm approximation , geometry of a cme is considered as a self - similar expanding sphere attached to the sun and its centre moves along a fixed longitude . the choice of the hm method is based on our earlier study on comparison of various reconstruction techniques involving both stereoscopic and single spacecraft observations @xcite . it was found that amongst the single spacecraft techniques , the hm method is the most suitable for arrival time prediction . implementing the 3d reconstruction technique in section [ 3d ] ( figure [ scc ] ) , we have shown that the estimated longitude of cme1 and cme2 are approximately 10@xmath10 and 2@xmath10 east of the sun - earth line . we assume that beyond the cor2 fov , cmes continue to propagate along the same direction i.e. we ignore any heliospheric longitudinal deflection of cmes . therefore , the aforesaid estimated values of longitude were used in the hm approximation to convert the derived elongation from j - maps to radial distance from the sun . figure [ kinematics ] shows the distance and speed estimated for different tracked features of the two cmes . the estimated speed is derived from the adjacent distances using a numerical differentiation with three point lagrange interpolation and therefore has systematic variations due to slight errors in the estimated distance . it is clear that the leading edge ( le ) of cme1 has higher speed ( @xmath16 500 km s@xmath15 ) than its trailing edge ( te ) speed ( @xmath16 350 km s@xmath15 ) , averaged over few data points at the entrance of hi1 fov . also , the le and te of cme1 have lower speeds than le of cme2 . le of cme2 shows a large radial speed of approximately 1100 km s@xmath15 ( ecliptic speed = 950 km s@xmath15 ) in the cor2 fov i.e.(2.5 - 15 r@xmath14 ) ( figure [ scc ] ) . beyond 10 r@xmath14 distance , the le of cme2 continuously decelerates for @xmath16 10 hours up to 46 @xmath0 where its speed reduces to 430 km s@xmath15 . the extreme fast deceleration of le of the cme2 , starting from cor2 fov , seems to be due to possible interaction with the preceding cme1 . it is likely that the cme1 had large spatial scale due to which the trailing plasma and magnetic fields from cme1 created sufficiently dense ambient medium acting as a huge drag force for cme2 which resulted in its observed deceleration . the extreme fast deceleration of cme2 le can also be due to closed magnetic structure of cme1 which may act like a magnetic obstacle for the cme2 @xcite . from estimated kinematics of tracked features in heliosphere ( top and bottom panel of figure [ kinematics ] ) , it is clear that around november 10 at 11:30 ut speed of cme1 te started to increase from 365 km s@xmath15 with simultaneous decrease of cme2 le speed from 625 km s@xmath15 . such an observation of acceleration of cme1 and deceleration of cme2 provides an evidence for the commencement of collision @xcite . we define collision phase , as the interval during which the two cmes come in close contact with each other and show opposite trend of acceleration relative to one another until they attain an approximately equal speed or their trend of acceleration is reversed . carefully observing the variations of speed of tracked cme1 te and cme2 le , the start and end boundary of collision phase is drawn as vertical lines in the bottom panel of figure [ kinematics ] . we noticed that at the end of the collision phase around november 10 at 17:15 ut , speed of cme1 te is @xmath16 450 km s@xmath15 and speed of cme2 le is 430 km s@xmath15 . in the beginning of the marked collision phase , cme1 te is at a distance of 37 r@xmath17 and cme2 is at 30 r@xmath17 . at the end of the marked collision phase , the cme1 te and cme2 le are at a distance of 50 r@xmath17 and 46 r@xmath17 , respectively . here , we must highlight that throughout the collision phase , the estimated heliocentric distance of cme1 te and cme2 le is never found to be equal . this is because of the large scale structure of cmes and we suggest that the tracked te of cme1 and le of cme2 using j - maps , are not strictly the rearmost trail of cme1 and outermost front of cme2 , respectively . therefore , we had defined ` collision ' when tracked features of cmes ( i.e. time - elongation tracks ) come in close contact ( separation of @xmath16 5 r@xmath17 ) with one another . such small separation between observed collided features is possible if forward shock driven by cme2 interacts with the rear most portion of the cme1 as also suggested by @xcite . to understand the momentum exchange during collision of cmes , we require their true masses . since , appearance of a cme is due to thomson scattered photospheric light from the electrons in the cme @xcite , therefore , the recorded scattered intensity can be converted into the number of electrons and hence mass of a cme can be estimated , if composition of cme is known . but an observer from different vantage points receives different amount of scattered light by the electrons , therefore , the true location of electrons of cme , i.e. propagation direction of cme must be known to estimate the true mass of cme . for a long time , mass of a cme was calculated assuming the cme location in the observer s plane of sky @xcite . although , the propagation direction of cme is calculated in our study using tie - pointing @xcite and forward modeling @xcite methods and described in section [ 3d ] , but to avoid any bias of the reconstruction methods , we use the method of @xcite which is based on thomson scattering theory , to estimate the true propagation direction and true mass of both the cmes . we used the simultaneous image pair of secchi / cor2 to estimate the true mass of cmes . first , we estimated the mass of cme i.e. m@xmath18 and m@xmath19 , from stereo - a and stereo - b locations . then we took the ratio of equations ( 7 ) and ( 8) of @xcite and obtained a new expression as @xmath20 , where @xmath21@xmath18 is the angle of propagation direction of cme measured from the plane - of - sky ( pos ) of _ stereo - a _ , b@xmath22(@xmath21@xmath18 ) is the brightness of a single electron at an angular distance of @xmath21@xmath18 from the pos and @xmath23 is the separation angle between both _ stereo - a _ and _ stereo - b _ from the sun - earth line . using the measured m@xmath18 and m@xmath19 in aforementioned expression , the value of @xmath24 can be estimated . however , we noticed that multiple values of @xmath24 correspond to same ratio of m@xmath18 and m@xmath19 but the correct value of @xmath24 among them can be selected by visual inspection of cme in cor2 images . we obtained the true mass of cme using the derived value of @xmath24 in equation ( 4 ) of @xcite . the estimated mass of cme1 at 18:39 ut on november 9 at @xmath16 15 r@xmath14 , from stereo - a and stereo - b vantage point is 4.60 @xmath25 10@xmath26 kg and 2.81 @xmath25 10@xmath26 kg , respectively . for cme2 , the estimated mass at 07:24 ut on november 10 at @xmath16 15 r@xmath14 , from stereo - a and stereo - b locations is 2.25 @xmath25 10@xmath26 kg and 1.31 @xmath25 10@xmath26 kg , respectively . the estimated propagation direction of cme1 and cme2 is 19@xmath11 and 21@xmath11 west from the sun - earth line , respectively . we estimated the true mass of cme1 and cme2 as m@xmath27 = 4.66 @xmath25 10@xmath26 kg and m@xmath28 = 2.27 @xmath25 10@xmath26 kg , respectively . we assume that after crossing the cor2 fov and during collision of cme1 and cme2 , their estimated true masses ( m@xmath27 and m@xmath28 ) remain constant as estimated earlier . from the figure [ kinematics ] , the observed velocity of cme1 and cme2 before the collision is estimated as ( u@xmath27,u@xmath28 ) = ( 365,625 ) km s@xmath15 and observed velocity of cme1 and cme2 after the collision is ( v@xmath27,v@xmath28 ) = ( 450,430 ) km s@xmath15 . to understand the nature of collision of large scale magnetized plasmoids , we attempt to estimate the coefficient of restitution ( @xmath29 ) of colliding cme1 and cme2 . the coefficient of restitution measures the bounciness ( efficiency to rebound ) of a pair of objects in collision and is defined as ratio of their relative velocity of separation to relative velocity of approach . hence , for @xmath29 @xmath30 1 , @xmath29 = 1 , and @xmath29 @xmath31 1 the collision is termed as inelastic , elastic and super - elastic and consequently the kinetic energy of the system after the collision is found to decrease , equal and increase than before the collision , respectively . we restrict ourselves not to estimate the value of @xmath29 directly by using the pre and post collision velocity of cmes . this is because the velocities estimated from reconstruction method have some errors and do not guarantee conservation of momentum , a necessary condition of collision , therefore , can lead to erroneous estimation of @xmath29 value . in our approach , considering the condition of conservation of total momentum for the collision , the velocity of cme1 and cme2 after the collision can be estimated theoretically ( v@xmath32,v@xmath33 ) , provided that velocity of cme1 and cme2 before the collision and the coefficient of restitution ( @xmath29 ) value are known . using the velocities ( u@xmath27,u@xmath28 ) and masses ( m@xmath27,m@xmath28 ) , we iterate for a range of @xmath29 values and estimate the ( v@xmath32,v@xmath33 ) each time . using this approach , the most suitable value of @xmath29 , using which the theoretically estimated ( v@xmath32,v@xmath33 ) is found to be closest to the observed ( v@xmath27,v@xmath28 ) i.e. for which the variance , + @xmath34 , is minimum , can be obtained . the details of the above approach and the equations used , are explained in @xcite . using the aforementioned approach , the value of @xmath29 = 0.1 is found with ( v@xmath32,v@xmath33 ) = ( 458,432 ) km s@xmath15 and @xmath35 = 9 . with this theoretically estimated @xmath29 value , total kinetic energy of cmes is found to decrease by 6.7% of its value before the collision . using the estimated velocities of the cmes before and after the collision ( figure [ kinematics ] ) , the value for @xmath29 is calculated equal to 0.08 , which is approximately same as obtained from iterations described above . therefore , our analysis suggests that the observed collision between the cmes are close to perfectly inelastic in nature . the kinetic energy of cme1 and cme2 before the collision was 3.1 @xmath25 10@xmath36 joules and 4.4 @xmath25 10@xmath36 joules , respectively . after the collision , based on the observed speeds , we found that the kinetic energy of cme1 increased by 51% to its value before the collision while the kinetic energy of cme2 decreased by 54.5% to its value before the collision . we also noticed that after the collision , momentum of cme1 increased by 23% and momentum of cme2 decreased by 31% to their values before the collision . such calculations support the claim that significant exchange of kinetic energy and momentum take place during cme - cme collision @xcite . considering the uncertainties of @xmath37 100 km s@xmath15 in the speed , we repeated the above analysis . taking the ( v@xmath27,v@xmath28 ) = ( 550,530 ) , we found @xmath29 = 0 , i.e. perfectly inelastic . but in this case , ( v@xmath32,v@xmath33 ) is estimated as ( 450,450 ) km s@xmath15 which highlight that theoretically estimated velocity is not close to the observed velocity ( v@xmath27,v@xmath28 ) . if we consider the ( v@xmath27,v@xmath28 ) = ( 350,330 ) , then it implies that the velocities of both cmes after the collision are less than their values before the collision . therefore , such velocities also can not be considered for our case ( as defined in section [ hi ] ) . as , there are many sources of errors in the estimation of true mass of cmes , therefore , the effect of uncertainty in mass must also be examined before confirming the nature of observed collision . in our case , the mass ratio ( m@xmath27/m@xmath28 ) is equal to 2.05 , varying this ratio between 0.5 to 3.0 in a step of 0.25 , and repeating the iterative procedure for estimating the most suitable value of @xmath29 corresponding to each mass ratio , we obtain @xmath29 close to 0 . figure [ mass ] shows the variation of @xmath29 and @xmath35 with varying mass ratio . it is clear that despite taking large uncertainties in the mass of cme1 and cme2 , the nature of observed collision remains close to perfectly inelastic for the case of interacting cmes of 2012 november 9 - 10 . we analyzed the in situ data taken from wind spacecraft located at l1 to identify the tracked density enhanced features of cmes . figure [ insitu ] shows magnetic field and plasma measurements during november 12 , 12:00 ut to november 15 , 12:00 ut . the arrival of a forward shock ( labeled as s ) marked by a sudden enhancement in speed , temperature and density is noticed at 22:20 ut on november 12 . the region between the first and second vertical line represents the turbulent sheath region . based on the cme identification criteria of @xcite , the region bounded between second vertical line at 08:52 ut on november 13 and third vertical line at 02:25 ut on november 14 , is identified as cme structure . in this region , we observed expansion of cme characterized by a monotonic decrease in proton speed and temperature . based on the predicted arrival times which is derived using the estimated kinematics of the remotely observed tracked features of cme1 and cme2 as inputs in the dbm ( explained in section [ arrtime ] ) , this region is associated with earth - directed cme1 launched on november 9 . during the passage of cme1 , the magnetic field is observed to be high ( @xmath16 20 nt ) , plasma beta is less than unity ( @xmath38 @xmath30 1 ) with smooth rotation in magnetic field vector . also , the latitude ( @xmath21 ) value of magnetic field vector decreased from 43@xmath11 to -43@xmath11 and its longitude ( @xmath39 ) decreased from 203@xmath11 to 74@xmath11 . therefore , this region can be classified as a magnetic cloud ( mc ) and based on the observed arrival time it is associated with cme1 . due to the interaction of the cmes , the region associated with cme1 is found to be at higher temperature than found generally for a normal isolated cme . after the third vertical line , magnetic field strength decreased reaching a minimum value of 6 nt around 4:00 ut on november 14 . this interval of sudden drop in magnetic field is associated with a sudden rise in density , temperature and plasma @xmath38 , suggestive of a possible magnetic hole ( mh ) @xcite which is considered as a signature of magnetic reconnection @xcite . another region of magnetic field depression from 08:05 ut to 10:15 ut on november 14 , reaching a minimum value of 3 nt is also noticed . corresponding to this minimum value of magnetic field , the plasma @xmath38 and temperature is found to increase . the region during 03:45 ut - 08:05 ut on november 14 , bounded between two distinct mh like structures , has enhanced magnetic field ( @xmath16 15 nt ) , and plasma @xmath38 less than unity . this region , between two mh , seems to be a magnetic field remnant of reconnecting cmes structures . based on the extremely elevated interval of plasma beta , temperature , observation of magnetic hole ( mh ) and probably sudden fast rotation in magnetic field vector during the region bounded between third and fourth vertical line , in figure [ insitu ] , this region is identified as interaction region ( labeled as ir ) of cme1 te with cme2 le . another structure is identified based on the elevated fluctuating magnetic field and temperature during 12:00 ut - 21:21 ut on november 14 , bounded between fourth and fifth vertical line in figure [ insitu ] . during this interval , we noticed high magnetic field ( 9 nt ) with no monotonic decrease in temperature and speed profile as well as plasma beta ( @xmath38 ) is not less than unity . from these observations , i.e. lack of mc signatures and the short duration ( 9.5 hr ) of the structure associated with cme2 , we infer that the wind spacecraft perhaps intersected the flank of cme2 @xcite . this is also confirmed by the estimated latitude ( -25@xmath11 ) of cme2 using gcs model ( described in section [ 3d ] ) . on examining the evolution of cme2 in hi1-a movie , we noticed that cme2 is directed towards the southern hemisphere which is a favorable condition for its flank encounter with in situ wind spacecraft . we observed that the magnitude of magnetic field in mc region is constant around @xmath16 20 nt which may be due to the passage of shock from cme1 . we also noticed that average temperature in first half of cme1 is high ( @xmath16 10@xmath40 k ) than found in general ( @xmath16 10@xmath41 k ) . the high temperature of the cme1 may occur due to its collision with cme2 , thereby resulting in its compression . another possibility of high temperature of cme1 is due to the passage of forward shock driven by cme2 as reported in earlier studies by @xcite . however , in the present case , the following cme2 is also observed with a high temperature ( @xmath16 5 @xmath25 10@xmath40 k ) , which has not been reported in earlier studies of interacting cmes . we noticed significant high density at the front of cme1 , possibly due to overall compression by sweeping of the plasma of cme1 at its leading edge by cme2 driven shock . we used the estimated speed at the last point of measurement ( up to where cmes could be tracked unambiguously in hi ) and used it as input in the dbm developed by @xcite to estimate the arrival time of tracked features at l1 . @xcite have shown that drag parameter lies in the range of ( 0.2 - 2.0 ) @xmath25 10@xmath42 km@xmath15 . using these values of the drag parameter in the dbm , the arrival time of cme at l1 can be predicted with a reasonably good accuracy @xcite . the estimated speed , time and distance ( v@xmath43 , t@xmath43 and r@xmath17 ) of le of cme1 ( green track in j - map ) are used as inputs in the dbm , corresponding to extreme range of drag parameter , to predict its arrival time and transit speed at l1 . in situ observations show a peak in density , @xmath16 0.5 hr after the shock arrival around 23:00 ut on november 12 with a transit speed of 375 km s@xmath15 , which is expected to be the actual arrival of tracked le feature corresponding to cme1 . the predicted values of arrival time and transit speed of features and errors from the actual values are shown in table [ tabarr ] . for te feature of cme1 and le of cme2 , we assume that they encounter the dense ambient solar wind medium created by the preceding cme1 le . therefore , their kinematics with the maximum value of the statistical range of the drag parameter ( 2.0 @xmath25 @xmath44 km@xmath15 ) , are used as inputs in the dbm . we consider that the te of cme1 corresponds to the density enhancement at trailing front of cme1 . at the rear edge of cme1 , a density enhancement ( 12 particles @xmath45 ) is observed around 23:30 ut in wind observations on november 13 , which is considered as the actual arrival of te of cme1 . the predicted arrival time and transit speed of this feature and errors therein , are given in table [ tabarr ] . we further noticed that in situ data , an enhancement in density corresponding to le of cme2 around 12:00 ut on november 14 ( marked as arrival of cme2 with fourth vertical line in figure [ insitu ] ) is observed which can be considered as the actual arrival time of le of cme2 . the predicted values of cme2 le , using its kinematics with dbm , and errors therein are also listed in table [ tabarr ] . the error for cme2 le is large , but , several factors can lead to such large errors which are discussed in section [ results ] . it must be highlighted that if the 3d speed of cmes estimated at final height in cor2 fov is assumed as constant up to l1 , then the predicted arrival time of cme1 and cme2 will be @xmath16 10 - 16 hr and 44 hr earlier , respectively , than their predicted arrival times using post - collision speeds combined with dbm . this emphasizes the use of hi observations and post - collision speeds of cmes as inputs in the dbm for improved arrival time prediction of interacting cmes . similarly , using hi observations , @xcite have shown that linear fit of deprojected height - time above 50 r@xmath17 gives a half a day improvement over cme arrival time estimated using lasco data . as mentioned in the introduction section , very few studies have been dedicated to the study of the geomagnetic consequences of interacting cmes in the past . the cmes of 2012 november 9 - 10 resulted in a single strong geomagnetic storm with dst index @xmath16 -108 nt at 8:00 ut on november 14 , therefore it is important to investigate the impact of the interaction of the cme1 of november 9 with the cme2 of november 10 on the terrestrial magnetosphere - ionosphere system ( mi ) in details . figure [ geomag](a - e ) reveal the variations in the solar wind parameters in the geocentric solar ecliptic ( gse ) coordinate system during 2012 november 12 - 15 . these parameters include solar wind proton density ( in @xmath46 ) , velocity ( in km @xmath47 , negative x - direction ) , ram pressure ( in npa ) , z - component of the interplanetary magnetic field ( imf @xmath48 , in nt ) and y - component of the interplanetary electric field ( @xmath49 , in mv / m ) respectively . these data with cadence of 1 minute are taken from the nasa gsfc cdaweb ( link : www.cdaweb.gsfc.nasa.gov / istp _ public/[www.cdaweb.gsfc.nasa.gov / istp _ public/ ] ) . it is also important to note that the solar wind parameters presented in figure [ geomag](a - e ) are corrected for propagation lag till the nose of the terrestrial bow shock . in order to compare the variations of these parameters with the magnetospheric and ionospheric parameters , additional time lags that account for the magnetosheath transit time and the alfven transit time are calculated @xcite . therefore , the solar wind parameters presented in figure [ geomag](a - e ) are corrected for the propagation lag , point by point , till ionosphere . figure [ geomag](f ) represents the variation in the polar cap ( pc ) index . the pc index is shown @xcite to capture the variations in the ionospheric electric field over polar region efficiently . figure [ geomag](g ) shows the variation in the sym - h ( in nt ) index which primarily represents the variation in the magnetospheric ring current @xcite . the sym - h index is essentially the high temporal resolution ( 1 min ) version of the dst index . figure [ geomag](h ) shows variations in westward ( midnight sector ) auroral electrojet ( al ) current ( in nt ) which captures auroral substorm processes reasonably well . further , in figure [ geomag ] , the arrival of the shock is denoted by s , different vertical lines mark the arrival of different features of cme1 and cme2 and are labeled in the same manner as in figure [ insitu ] . figure [ geomag](a ) reveals the arrivals of two distinctly enhanced density structures , first enhancement occurred during arrival of the shock - sheath region before cme1 le ( on november 12 20:00 ut to november 13 04:00 ut ) and second enhancement occurred during arrival of the cme1 te - ir region ( november 13 12:00 ut to november 14 12:00 ut ) . during the arrival of shock - sheath region before cme1 le , the peak density reached @xmath1668 @xmath46 which is more than two times of the corresponding peak density ( @xmath1630 @xmath46 ) observed during the arrival of cme1 te - ir . sharp enhancement in the solar wind velocity was also observed vis - a - vis the sharp density enhancement during arrival of shock - sheath region ( figure [ geomag]b ) when the velocity reached from @xmath16300 km s@xmath15 to @xmath16470 km s@xmath15 . however , solar wind velocity did not change sharply and significantly during arrival of cme1 te - ir . the changes in the density and the velocity resulted in changes in the solar wind ram pressure shown in figure [ geomag](c ) . the peak ram pressure during shock - sheath region before cme le ( @xmath1618 npa ) was almost twice than that observed during cme1 te - ir ( @xmath169 npa ) . figure [ geomag](d ) reveals that imf @xmath48 was predominantly southward in both density enhanced intervals although fast fluctuations were observed corresponding to the arrival of the shock - sheath region . the first peak in imf @xmath48 is @xmath16 - 8 nt before the arrival of the shock . thereafter , imf @xmath48 fluctuated sharply between @xmath16 - 20 nt to @xmath16 + 20 nt during arrival of shock - sheath region before arrival of cme1 le . during cme1 te - ir , imf @xmath48 reached a peak value of @xmath16 - 18 nt . no significant change in imf @xmath48 is observed during the passage of the cme2 on november 14 - 15 , when its magnitude hovered around the zero line . figure [ geomag](e ) elicits the variation in @xmath49 during 2012 november 12 - 15 . it is observed that the peak value of @xmath49 before the arrival of the shock was @xmath162 mv / m . however , during the shock event , @xmath49 fluctuated between @xmath37 @xmath508 mv / m . during arrival of cme1 te - ir , the peak value of @xmath49 reached @xmath167.5 mv / m . in fact , similar to imf @xmath48 , a sharp polarity change in @xmath49 was noticed at @xmath5009:40 ut on november 14 . no significant change in @xmath49 is observed during the passage of the cme2 on november 14 - 15 , when its magnitude hovered around the zero line . figure [ geomag](f ) shows that the pc index increased during both density enhanced intervals and the peak value is @xmath165 for both the intervals . figure [ geomag](g ) shows variation in the sym - h index which revealed the development of a geomagnetic storm during cme1 te region . the sym - h index reached a value of @xmath16 - 115 nt during the passage of ir on november 14 . the ring current activity was not significant during arrival of the shock - sheath region before cme1 . lastly , figure [ geomag](h ) elicits the variation in the al index . it is seen that al reached @xmath16 - 600 nt during the arrival of shock - sheath before cme1 and @xmath16 - 1400 nt during the arrival of cme1 te - ir . therefore , significant intensification of westward auroral electrojet occurred during the arrival of cme1 te - ir . the above observations ( from figure [ geomag ] ) reveal several interesting points . first , the magnitude of al seems to remain unaffected by large amplitudes of fluctuations in imf @xmath48 and @xmath49 during the shock - sheath region . second and the most important point is that the duration of occurrence of the al intensification during shock - sheath before cme1 le and in cme1 te - ir region were nearly identical with the duration of the southward imf @xmath48 and positive @xmath49 phases . however , the al amplitudes seem to be more closely correlated with the @xmath49 amplitudes during cme1 te - ir region compared to shock - sheath region . third , although the peak amplitudes of pc index were nearly same during shock - sheath region before cme1 le and cme1 te - ir region , the peak amplitudes of the al index were significantly different during these two intervals . fourth , the substorm activity seems to be over during the passage of the cme2 , therefore , cme2 did not have any bearing on the triggering of substorms . figure [ geomag ] demonstrates the direct role played by the positive @xmath49 ( or southward imf @xmath48 ) in the storm - time al intensification particularly when the terrestrial magnetosphere encounters the cme1 te and ir region . further , the arrival of cme2 did not affect the terrestrial magnetosphere - ionosphere system . as aforementioned , major geomagnetic response is noticed during the arrival of trailing edge of the preceding cme ( cme1 ) and the interaction region ( ir ) of the two cmes near the earth . we also conclude that the following cme ( cme2 ) failed to cause a significant geomagnetic activity possibly due to the fact that the in situ spacecraft encountered the flank of this cme as mentioned in section [ insitu ] . we understand that due to interaction and collision between the trailing edge of the preceding cme1 and the leading edge of the following cme2 as revealed in imaging observations ( described in section [ hi ] ) , the parameters responsible for geomagnetic activity were significantly intensified at cme1 rear edge and in interaction region found between cme1 and cme2 . these results bring out the importance of cme - cme interaction in the formation of interaction region and its role in the significant development of geomagnetic disturbances . in this paper , we have studied the interactions between two successively launched earth directed cmes from the sun on 2012 november 9 and 10 , respectively . the analysis involved remote sensing observations of cme propagation in inner heliosphere from secchi / cor2 and hi images and the associated in situ measurements from wind spacecraft . the first cme ( cme1 ) associated with a filament eruption is observed to be slower ( 600 km s@xmath15 ) than the successive cme ( cme2 ) launched on november 10 with a higher velocity ( 900 km s@xmath15 ) . both the cmes were launched in approximately the same direction towards the earth , indicating a possibility of interaction because of their relative speeds and the same direction . different features were tracked as brightness enhancements in j - maps obtained from cor2 and hi images and subsequently hm technique was applied to estimate the kinematics of the tracked features of the cmes . from the estimated kinematics , the site of collision of cmes could be located at a distance of approximately 35 r@xmath14 at 12:00 ut on november 10 , which is at least 85 r@xmath14 before and 15 hours earlier than as predicted by using the initial kinematics of cmes in cor2 fov . the estimated kinematics were then used as inputs in the dbm to determine the arrival times of the tracked features . after the interaction , cme2 transferred some of its velocity to cme1 and both were observed to propagate together , showing a distinct deceleration of cme2 . as described earlier , cme1 le is found to be propagating with higher speed than cme1 te . therefore , cme1 le ( green track ) did not collide with cme2 le ( blue track ) . with the start of collision phase , the le of cme1 is also found to accelerate and after the collision phase , significant acceleration of cme1 le is noticed . this may occur either because of sudden impact ( push ) from cme1 te to cme1 le during collision or due to the passage of a shock driven by the cme2 or a combined effect of both . beyond the collision phase , up to an estimated distance of nearly 100 r@xmath14 , le of cme2 was found moving behind the te of cme1 and both these features propagated together decelerating slowly . for few hours after the observed collision phase , it is noticed that speed of the le and te of cme1 is slightly higher than le of cme2 , which can increase the observed separation of these structures . the j - maps show that both these features can be tracked for further elongations after interaction . however , due to the limitation of the hm method for higher elongations @xcite , we restricted our measurements on the tracked features up to 100 r@xmath14 . our analyses show an evidence that after the interaction , the two features did not merge . it appears that le of cme2 interacted with te of cme1 and continued to propagate with a reduced speed of 500 km s@xmath15 . in the secchi / hi - b images these features can not be well distinguished and therefore it is difficult to use any stereoscopic method to infer about a possible longitudinal deflection ( based on stereoscopic methods ) of these features after interaction . during the collision between cme1 and cme2 , we noticed a large deceleration of fast cme2 while relatively less acceleration of slow cme1 , till both approached an equal speed . this is expected to occur if the mass of cme1 is larger than that of cme2 , which is indeed the case . in section [ momentum ] we have shown that the mass of cme1 is @xmath16 2.0 times larger than the mass of cme2 . this result is an important finding and is in agreement with the second scenario of interaction described in @xcite . it also must be noted that we have estimated the mass of cmes in cor2 fov while interaction takes place in hi fov . further , we can not ignore the possibility of increase in mass of cme due to mass accretion at its front via snowplough effect in the solar wind beyond cor fov @xcite . we must also mention that in our calculation of momentum and energy transfer , although we use the total mass of cme1 but only a part of cme1 ( i.e. te of cme1 ) takes part in collision with cme2 . considering the uncertainties in the derived speeds and mass of the cmes , our analysis reveals that the nature of collision remains close to perfectly inelastic . however , we also acknowledge that defining collision phase is crucial and seems to play an important role in this calculation . in this context , it must be emphasized that marking of the start of collision is often difficult . this is because the following fast speed cme2 starts to decelerate , well before the actual merging or collision is revealed in imaging observations . this is also shown in earlier studies for other interacting cmes @xcite . the interaction of cme2 with streamers or overlying coronal magnetic field lines @xcite may be responsible for deceleration of following cme2 before its collision with cme1 takes place . however , the contribution of other factors , e.g. , the solar wind drag acting on cme2 , interaction of cme2 with trailing edge of preceding cme1 , and magnetic interaction of cme2 with preceding cme1 can not be completely ignored . also , possibly a shock launched by the following cme2 can contribute in acceleration of the preceding cme1 before the actual merging of both cmes . therefore , different timing and large time - interval of acceleration of one cme and deceleration of other , prevent us to pinpoint the exact start and end of the collision phase . we concluded that the collision of interacting cmes of 2012 november 9 - 10 is close to perfectly inelastic in nature . here , it is noted that we are dealing with large scale magnetized three - dimensional structures , however , in the present study we have reduced the calculations to one - dimensional collision which seems to be accurate as tracked features of both cme1 and cme2 are considered to move along approximately the same trajectory . this implies that there will be a small difference between velocity of tracked feature of one cme and its component along the direction of propagation ( i.e. longitude ) of tracked feature of other cme . we must admit that possibility of change in longitude of tracked features during their collision can partially contribute to some errors in the estimated speed . in light of earlier reported observations @xcite , it seems that collision of cmes can occur in all i.e. elastic , inelastic or super - elastic regimes . an in - depth study is required further for understanding the nature of cme - cme collision which may depend on the characteristics of cmes , locations and duration of collision phase in the heliosphere . at the last point of measurement in hi fov around 120 r@xmath14 , the estimated speed of te of cme1 and le of cme2 are approximately equal ( @xmath16 470 km s@xmath15 ) . therefore , both features are expected to arrive at l1 at approximately the same time . comparing to in situ measured actual arrival time , the delayed arrival of le of cme2 is possible due to higher drag force acting on it resulting in its deceleration . however , the propagation direction of cme2 in southward hemisphere , as noticed in hi1-a images and also estimated in cor2 fov using 3d reconstruction , can also account for flank encounter of cme2 and thus its delayed sampling by wind spacecraft at l1 . this is consistent with our interpretation in section [ insitu ] . since , in the hm method , the front of the cme is assumed spherical which may not be the real case for cme2 , as it was moving with fast speed and has large probability to distort and flatten its front while interacting with structured solar wind @xcite . in this case , the estimated speed using hm method will lead to some error and can result in the delayed arrival of le of cme2 . keeping all these issues in mind , we believe that it is quite probable that the tracked feature corresponding to le of cme2 ( blue track in j - map ) is not sampled by in situ spacecraft . therefore , erroneous predicted arrival time for le of cme2 is related to incorrect identification of tracked cme2 le in in situ data which is taken as a reference ( actual ) for remotely observed tracked cme2 le . the present study not only estimates the kinematics and arrival times based on hi observations and their application to the dbm model of cme propagation , it also attempts to associate tracked brightness enhancements as observed in remote sensing cor and hi ( j - maps ) with the in situ observations @xcite . the association of hi observations with in - situ measurements from wind leads to many interesting results . in spite of two cmes launched from the sun in succession in the earthward direction , we observe only one shock in in situ data which may suggest the merging of shocks driven by cme1 and cme2 , if both cmes would have driven shocks . however , such a claim can not be made unless we are well familiar with the in situ signatures of merged shock and plasma structure following it . the sweeping of plasma to high density at the front of cme1 and its compressed heating is most likely due to passage of cme2 driven shock through the mc associated with cme1 . based on the predicted arrival time of tracked features , it seems that the cme2 driven shock and cme1-sheath region is tracked as cme1 le in j - map . therefore , we infer that cme1 le is propagating probably into an unperturbed solar wind . our study also provides a possibility of formation of interaction region ( ir ) at the junction of trailing edge of preceding cme and leading edge of following cme . we show that during collision of the cmes , kinetic energy exchange up to 50% and momentum exchange between 23 to 30% . our study also demonstrate that the arrival time prediction significantly improved using his on - board stereo compared to cor2 observations , and also emphasizes the importance of understanding of post - collision kinematics in further improving the arrival time prediction for a reliable space weather prediction scheme . our study reveals clear signatures of interaction of these cmes in remote and in situ observations and also helps in identification of separate structures corresponding to these cmes . in spite of the interaction of the two cmes in the interplanetary medium which generally results in complex structures as suggested by @xcite , in our case , we could identify interacted cmes as distinguished structures in wind spacecraft data . even after collision of these cmes , they did not merge which may be possibly because of strong magnetic field and higher density of cme1 than cme2 . this needs further confirmation , and therefore , it is worth to investigate what decides the formation of merged cme structure or complex ejecta during cme - cme interaction . due to single point in situ observations of cme , we acknowledge the possibility of ambiguity in marking the boundaries of cmes . in the present case , the boundaries for cme1 and ir are distinctly clear . also , slight ambiguity in the boundary of cme2 ( flank encounter ) will not change our interpretation because the main geomagnetic response is caused due to enhanced negative bz in trailing portion of cme1 and its extension in ir . here , we also point out that temperature in ir is lesser than the temperature for cme2 region but temperatures in both regions are elevated as compared to a normal non - interacting cme . the observations of unexpected larger temperature in cme2 region than ir region may be due to the possibility that sheath region of cme2 is intersected by in situ spacecraft , as we have interpreted . the association of geomagnetic storms with isolated single cmes has been carried out extensively for a long time @xcite . however , only few studies have been dedicated to understand the role of interacting cmes in the generation of geomagnetic storms @xcite . our study is important as it focuses on the role of interacting cmes in the generation of geomagnetic storms as well as substorms . in the context of substorms , our study highlights that persistence of imf bz in the southward direction is more important than the amplitude in driving the substorm activity as manifested by the al intensification . using windmi model , @xcite have shown that the ip shock and sheath features for cmes contribute significantly to the development of storms and substorms . but , in our study , sharp and large southward excursions in the midst of fluctuating imf bz associated with shock ( shock - sheath region before cme1 ) were found less effective in producing strong substorm activity . therefore , further investigations are required regarding characteristics ( geometry , intensity ) of shock and preceding cme in the context of triggering of substorms , as has been shown in earlier studies @xcite . regarding geomagnetic storms , our study suggests that the trailing edge of preceding cme ( cme1 ) and ir formed between two interacting cmes are efficient candidates for intense geomagnetic storms . is that the nearly equal amplitude responses of the pc index corresponding to the shock - sheath region ( imf bz sharply fluctuating between southward and northward directions ) preceding the cme1 interval vis - - vis the cme1 interval ( imf bz steadily turning southward ) . this is interesting as the responses of sym - h and al during these two intervals are quite different in terms of amplitudes of variations . @xcite have inferred that the pc index for the sunlit polar cap ( summer hemisphere , high polar ionospheric conductance ) responds mainly to the geo - effective interplanetary electric field whereas the pc index for the dark winter cap ( winter hemisphere , low polar ionospheric conductance ) responds better to the particle precipitation in the auroral zone like the ae and al indices . it is to be noted here that , in the present study , no distinction is made between the variations of pc index in the northern ( pcn ) and southern ( pcs ) hemispheres . on the other hand , it is known that al index is constructed based on magnetometer observations around the northern auroral oval only . the discrepancy between the variations in the pc index and the al index may , therefore , suggest towards the hemispherical asymmetry in the response of polar ionosphere corresponding to this event . this is not unexpected as asymmetric auroral intensities in the northern and southern hemispheres are reported in literature @xcite . further investigations are needed to understand this aspect . the cmes of november 9 and 10 , provide us a rare opportunity to investigate the consequences of cme - cme interaction . a combination of heliospheric imaging and in situ observations are used for improving our understanding of cme kinematics , post - collision characteristics and nature of collision . main results of our study can be concluded as follows : 1 . the analysis of propagation kinematics as obtained from j - maps provide evidence that the cme1 and cme2 collide at 35 r@xmath14 much earlier than by using the estimated kinematics in the cor2 fov . this emphasizes the importance of heliospheric imaging particularly , for interacting cmes and ascertaining their impact and arrival at the earth . our analysis shows that post interaction and collision kinematics is required for a better prediction of cme arrival time at 1 au . in the present case of november 9 and 10 cmes , the speeds and momentum of cmes changed from 23% to 30% compared to their values before the collision . our results also highlight that estimated kinematics , in particular after collisions are important to combine with dbm for improving the estimation of arrival times of different features of cme which experience different drag forces during their propagation through the heliosphere . based on this study , we conclude that cmes can not be treated as completely isolated magnetized plasma blobs , especially when they are launched in quick succession . 3 . using estimated mass and kinematics before and after the collision , we estimated the total kinetic energy of the system before and after the collision . we found that total kinetic energy of the system decreased by 6.7% to its value before the collision and the nature of collision is close to perfectly inelastic . the study also supports the idea that cme interaction or collision can lead to the heating and compression of both preceding and following cmes . it is clear that if the trailing edge of november 9 cme would not have been tracked using imaging observation , one could not have witnessed the collision . further , trailing edge of this cme had strong negative component of magnetic field for long duration ( @xmath16 13 hr ) which may be due to its collision with the leading edge of november 10 cme . therefore , our study reveals that tracking of different features of cme seems to be necessary for better understanding of cme - cme interaction . it is observed in the present case that the persistence of imf @xmath48 in the southward direction is more important ( rather than the amplitude ) in driving the substorm activity as manifested in the al intensification . sharp and equally large southward excursions in the midst of fluctuating imf @xmath48 associated with shock were found less effective in producing equally strong substorm activity . it is identified that the interaction region ( ir ) , formed due to collision between cme1 te and cme2 le , has intensified plasma and magnetic field parameters which are responsible for major geomagnetic activity . the present case study of interacting cmes of 2012 november 9 - 10 , highlight the importance of heliospheric imaging for estimating the kinematics of cme features before and after their collision and interactions . our study support that interacting cmes can result in strong geomagnetic storms and also in substorms . further statistical study of such interacting cmes is required to understand their nature of collision and to investigate the characteristics of the cmes ( mass , strength and orientation of magnetic field , speed and direction of propagation , duration of collision phase ) which are responsible for their interaction . we acknowledge the uk solar system data centre for providing the processed level-2 stereo / hi data . the in situ measurements of solar wind data from ace and wind spacecraft were obtained from nasa cdaweb ( http://cdaweb.gsfc.nasa.gov/ ) . we also acknowledge the use of dbm , developed by bojan vr@xmath51nak and available at http://oh.geof.unizg.hr/cadbm/cadbm.php , in this study . we also thank the reviewer for useful comments , which helped improve the manuscript . the work by n.s . partially contributes to the research for european union seventh framework programme ( fp7/2007 - 2013 ) for the coronal mass ejections and solar energetic particles ( comesep ) project under grant agreement no .

in a former paper ( garcia - rissmann 2005 ; hereafter paper i ) , we have presented spectra of 64 active , 9 normal and 5 starburst galaxies in the region around the near - ir calcium triplet absorption lines ( cat ) and the [ siii] line . in the present paper we analyze the cat strength ( ) , and kinematical products derived in that study , namely stellar ( ) and ionized gas ( ) velocity dispersions . our main results may be summarized as follows : ( 1 ) seyfert 2s show no sign of dilution in with respect to the values spanned by normal galaxies , even when optical absorption lines such as the caii k band at 3933 are much weaker than in old , bulge - like stellar populations . ( 2 ) the location of seyfert 2s in the- plane is consistent with evolutionary synthesis models . the implication is that the source responsible for the dilution of optical lines in these agn is a young stellar population , rather than an agn featureless continuum , confirming the conclusion of the pioneer study of terlevich , daz & terlevich . ( 3 ) in seyfert 1s , both}$ ] and tend to be diluted due to the presence of a non - stellar component , in agreement with the unification paradigm . ( 4 ) a comparison of with ( obtained from the _ core _ of the [ siii ] emitting line ) confirms the existence of a correlation between the typical velocities of stars and clouds of the narrow line region . the strength and scatter around this correlation are similar to those previously obtained from the [ oiii] line width .

0809.3178

the stellar absorption lines from the near - ir calcium triplet ( cat ) made their debut in the field of active galactic nuclei ( agn ) with the work of terlevich , daz & terlevich ( 1990 , hereinafter tdt ) . at that time , it was thought that the optical spectrum of seyfert 2s contained a featureless continuum ( fc ) from the agn , which , in light of unification scenarios ( antonucci 1993 ) , should be associated with scattered light from the hidden seyfert 1 nucleus ( cid fernandes & terlevich 1995 ) . tdt realized that this idea could not be fully correct , since their data indicated that , unlike for absorption lines in the optical , the strength of the cat in these sources shows no signs of dilution with respect to the values found in normal galaxies . their interpretation of this fact was that what was called a `` non - stellar fc '' was in fact a stellar component associated to young stellar populations . this conclusion was thoroughly confirmed by numerous studies in the past decade ( see cid fernandes 2004 for a review ) . since then , interest in the cat shifted towards its use as a tracer of stellar kinematics , particularly in agn . the location of the cat in a relatively clean spectral region makes it an ideal feature to measure stellar velocity dispersions ( @xmath6 ) , which trace the gravitational potential of the host galaxy s bulge . in an influential work , nelson & whittle ( 1996 ; nw ) used the cat in a comparison between stellar and gaseous kinematics of agn , finding that the typical velocity narrow line region ( nlr ) clouds ( @xmath3 ) correlates with @xmath6 , albeit with significant scatter . more recently , a strong correlation between @xmath6 and black - hole mass ( tremaine 2002 and references therein ) indirectly boosted interest in observations of the cat ( nelson 2000 ; botte 2004 ) . for completely different reasons , recent work on normal galaxies has also raised the interest in the cat . empirical investigations have revealed rather puzzling behavior of the cat in bulges and elliptical galaxies . firstly , detailed population synthesis models tend to over - predict the cat strength ( saglia 2002 ; cenarro 2003 , 2004 ) , particularly for giant ellipticals , though for dwarf ellipticals the match between data and models is satisfactory ( michielsen 2007 ) . secondly , while classic metallicity tracers like the mg@xmath7 index are known to correlate with @xmath6 ( eg , terlevich 1981 ) , tracing the well - known mass - metallicity relation , the strength of the cat appears to be _ anti - correlated _ with @xmath6 ( cenarro 2003 , 2004 ; falcn - barroso 2003 ; michielsen 2003 ) . the interpretation of these results is still not clear , and the situation is likely to be even more complex in systems with varied star formation histories like agn ( cid fernandes 2004 ; wild 2007 ) . one thus sees that , more than 15 years after tdt introduced the cat in agn research , there are still plenty of reasons to keep studying it . with this general motivation , in paper i we have carried out a spectroscopic survey of seyfert galaxies in the region including the cat and the [ siii]@xmath09069 emission line . paper i concentrated in the presentation of the data and the derivation of four main data products : stellar velocity dispersions ( @xmath6 ) , [ siii ] emission line widths ( @xmath8}$ ] , which is representative of the highly ionized component of the gas of the narrow line region in agns ) , [ siii ] equivalent widths ( @xmath9}$ ] ) , and the cat strength ( @xmath1 ) . here we use these data to address the following issues : ( 1 ) examine the contribution of a non - stellar component in the near - ir spectra of seyfert galaxies and test the consistency of @xmath9}$ ] and @xmath1 data with the unified model , ( 2 ) study the sensitivity of @xmath1 to stellar population properties , ( 3 ) investigate the connection between nlr and stellar kinematics and cat strength , and ( 4 ) report results of spatially resolved spectroscopy , not reported in paper i. to study these questions , we complement the data in paper i with values of the equivalent width of the caii k line ( @xmath4 ) obtained from previous studies . as shown by cid fernandes ( 2001 ) , the caii k line is a powerful tracer of stellar populations even in type 2 seyferts . by comparing @xmath1 with @xmath4 we can assess whether the cat is a useful stellar population tracer , and determine whether the combination of cak and cat strengths is explained as due to stars alone or if an fc component is necessary . we also add in literature information on the width ( @xmath10}$ ] ) of the [ oiii]@xmath05007 line , which is useful to test whether our results are somehow affected by the choice of [ siii]@xmath09069 as a tracer of nlr motions . this paper is organized as follows . section [ sec : cat_nonstellarlight ] presents a series of studies related to the cat strength . after showing cat observational properties in [ subsec : cat in sy ] , we compare @xmath1 with @xmath4 in [ subsec : dilution ] and investigate whether the location of the points in the @xmath1 versus @xmath4 diagram is consistent with the existence of an fc at near - ir wavelengths . in [ stellar models ] we use evolutionary synthesis models to track the behavior of both @xmath1 and @xmath4 as a function of age and metallicity , and overlay the models onto the data in the @xmath1 versus @xmath4 diagram . in [ subsec : spatial cat profiles ] we study the spatial behavior of @xmath2 and @xmath1 for about half of our sample . in section [ sec : kinematics ] we analyze stellar and ionized gas kinematics : stellar velocity dispersions and its relation to cat are studied in [ subsec : stellar velocity dispersions ] , while in [ subsec : stellar vs. nlr kinematics ] we discuss the link between stellar and gas kinematics . finally , section [ sec : summary ] summarizes our main results . one of our goals is to evaluate the usefulness of the cat as a stellar population diagnostic . since our sample is dominated by agn , before discussing stellar populations we must first examine to which extent non - stellar light affects our cat measurements . as explained in paper i , we adopt the @xmath1 and @xmath11 definitions of cenarro et al ( 2001a ) . they offer two definitions of the cat equivalent width : ` cat ' ( which we call @xmath1 ) , which consists of a sum of the equivalent widths of all three cat lines , and ` cat * ' ( called @xmath11 here ) , which corrects @xmath1 for contamination by paschen line absorption . these equivalent widths are measured with respect to a continuum defined by fitting the spectrum in 5 windows in the 8474 - 8792 range . because of the presence of unwanted features in the observed spectra , we opted to measure @xmath12 and @xmath11 on the synthetic spectra . on the other hand , our spectral base used to fit the observed spectra does not have stars with the spectral types with paschen lines , so we use @xmath12 as equivalent widths measurements for our spectra . all nuclear cat data refers to appertures of @xmath13 2 @xmath14 2 ( paper i ) . to take into account the different velocity dispersions of our galaxies , we broadened each spectrum to match the largest value of @xmath15 of our sample , corresponding to ngc 3115 ( @xmath16 and @xmath17 km / s ) . this correction changes the value of @xmath1 by typically @xmath18 with respect to the uncorrected values given in paper i. @xmath4 measurements from cid fernandes et al ( 2004 ) are also used in our analysis . it is well known that a non stellar light is present in seyfert 1s . the contribution of such a component in seyfert 2s has been a topic of much controversy in the past . one way to address this issue is to compare seyfert 1s with seyfert 2s through cat . [ fig_hist_ews ] ( top panel ) shows the distributions of @xmath12 for seyfert galaxies in our sample , with seyfert 1s ( including subtypes 11.9 ) marked by the filled areas . almost all seyfert 2 lie between 4 and 8 , with a median of 6.5 . @xmath1 tends to be smaller in type 1 nuclei , with a median of 4.8 . emission lines from the nlr follow the same pattern , with smaller equivalent widths in seyfert 1s than in seyfert 2s , as shown for [ siii]@xmath19 in the bottom panel of fig . [ fig_hist_ews ] . @xmath19 emission line ( bottom ) for seyfert 1 ( filled areas ) and seyfert 2 ( empty areas).,width=340 ] another way of illustrating this result is shown in fig . [ fig_cat_x_pl ] , where we plot the cat strength as a function the fractional contribution of power - laws to the continuum in the cat region . this fraction , which we denote @xmath20 , is deduced from the spectral fits presented in paper i , where , besides observed stars , power - laws of several slopes were included to emulate the effects of a template mismatch in the @xmath6 measurements . table [ newcat_and_pl ] list @xmath20 , along with the new ( corrected ) @xmath1 values mentioned at the beginning of this section and their errors . lllrlllr + name & type & @xmath1 ( ) & @xmath20 ( @xmath21 ) & name & type & @xmath1 ( ) & @xmath20 ( @xmath21 ) + ngc 205 & normal & [email protected] & 32 & ngc 7410 & sy 2 & [email protected] & 22 + ngc 526a & sy 1.9 & [email protected] & 49 & ngc 7469 & sy 1.5 & [email protected] & 65 + ngc 526b & normal & [email protected] & 19 & ngc 7496 & sy 2 & [email protected] & 37 + ngc 1019 & sy 1.5 & [email protected] & 35 & ngc 7582 & sy 2 & [email protected] & 14 + ngc 1068 & sy 2 & [email protected] & 24 & ngc 7590 & sy 2 & [email protected] & 17 + ngc 1125 & sy 2 & [email protected] & 21 & ngc 7714 & sy 3 & [email protected] & 53 + ngc 1140 & starburst & [email protected] & 37 & mark 0001 & sy 2 & [email protected] & 25 + ngc 1142 & sy 2 & [email protected] & 0 & mark 0003 & sy 2 & [email protected] & 47 + ngc 1241 & sy 2 & [email protected] & 18 & mark 0040 & sy 1 & [email protected] & 38 + ngc 1365 & sy 1.8 & [email protected] & 85 & mark 0078 & sy 2 & [email protected] & 15 + ngc 1380 & normal & [email protected] & 5 & mark 0079 & sy 1.2 & [email protected] & 55 + ngc 1386 & sy 2 & [email protected] & 5 & mark 0273 & sy 2 & [email protected] & 10 + ngc 1433 & normal & [email protected] & 4 & mark 0348 & sy 2 & [email protected] & 21 + ngc 1672 & starburst & [email protected] & 14 & mark 0372 & sy 1.5 & [email protected] & 6 + ngc 1808 & starburst & [email protected] & 21 & mark 0461 & sy 2 & [email protected] & 25 + ngc 2110 & sy 2 & [email protected] & 19 & mark 0516 & sy 1.8 & [email protected] & 28 + ngc 2639 & sy 1.9 & [email protected] & 27 & mark 0573 & sy 2 & [email protected] & 9 + ngc 2997 & normal & [email protected] & 16 & mark 0705 & sy 1.2 & [email protected] & 41 + ngc 3081 & sy 2 & [email protected] & 24 & mark 0915 & sy 1.8 & [email protected] & 28 + ngc 3115 & normal & [email protected] & 15 & mark 1066 & sy 2 & [email protected] & 31 + ngc 3256 & starburst & [email protected] & 55 & mark 1073 & sy 2 & [email protected] & 25 + ngc 3281 & sy 2 & [email protected] & 19 & mark 1210 & sy 2 & [email protected] & 38 + ngc 3783 & sy 1.5 & [email protected] & 66 & mark 1239 & sy 1.5 & [email protected] & 81 + ngc 4339 & normal & [email protected] & 6 & eso 362-g08 & sy 2 & [email protected] & 22 + ngc 4507 & sy 2 & [email protected] & 24 & eso 362-g18 & sy 1.5 & [email protected] & 28 + ngc 4593 & sy 1 & [email protected] & 60 & ic 2560 & sy 2 & [email protected] & 14 + ngc 4748 & sy 1 & [email protected] & 63 & ic 3639 & sy 2 & [email protected] & 31 + ngc 4968 & sy 2 & [email protected] & 20 & ic 5169 & sy 2 & [email protected] & 19 + ngc 5135 & sy 2 & [email protected] & 37 & iras 01475 - 0740 & sy 2 & [email protected] & 36 + ngc 5929 & sy 2 & [email protected] & 17 & iras 04502 - 0317 & sy 2 & [email protected] & 19 + ngc 6300 & sy 2 & [email protected] & 17 & mcg -01 - 24 - 012 & sy 2 & [email protected] & 27 + ngc 6814 & sy 1.5 & [email protected] & 47 & mcg -02 - 08 - 039 & sy 2 & [email protected] & 5 + ngc 6860 & sy 1.5 & [email protected] & 48 & mcg -06 - 30 - 015 & sy 1.5 & [email protected] & 47 + ngc 6907 & normal & [email protected] & 4 & mcg + 8 - 11 - 11 & sy 1.5 & [email protected] & 84 + ngc 6951 & sy 2 & [email protected] & 10 & ugc 1395 & sy 1.9 & [email protected] & 33 + ngc 7130 & sy 1.9 & [email protected] & 25 & ugc 12138 & sy 1.8 & [email protected] & 14 + ngc 7172 & sy 2 & [email protected] & 25 & ugc 12348 & sy 2 & [email protected] & 12 + ngc 7184 & normal & [email protected] & 19 & ugc 3478 & sy 1.2 & [email protected] & 66 + ngc 7212 & sy 2 & [email protected] & 33 & akn 564 & sy 1.8 & [email protected] & 32 + versus percentage of power - law contribution to the synthesis . hereafter , symbols are like in this figure.,width=340 ] the plot shows that @xmath20 is smaller for seyfert 2s than for seyfert 1s . furthermore , the values of both @xmath20 and @xmath1 obtained for seyfert 2s span approximately the same range as those found for the non - active galaxies in the sample . the fact that @xmath23 and similarly @xmath24}({\rm sey 1 } ) < w_{[siii]}({\rm sey 2})$ ] is consistent with the unified model , which predicts a dilution of the equivalent widths of the absorption and emission lines in sources where the nuclear non - stellar continuum is seen directly . the fact that the range of cat strengths spanned by seyfert 2s is similar to that spanned by non - active galaxies indicates that their scattered fc , if present , is too weak to produce a noticeable effect on the near - ir range . while the results reported above suggest one does not need to worry about non - stellar light in the cat range in seyfert 2s , it is desirable to test this idea more conclusively before proceeding to an analysis of @xmath1 in terms of stellar populations alone . this can be done combining our cat data with information from a different wavelength range . the equivalent width of the caii k absorption at 3933 ( @xmath4 ) provides the necessary information . the cak line , which originates in old , late - type stars from the host s bulge , is very sensitive to dilution by an underlying blue continuum . as usual , this continuum can be either due to an agn or to a young stellar population , and the value of @xmath4 by itself can not distinguish between these two possibilities ( e.g. , storchi - bergmann 2000 ) . the combination of @xmath4 and @xmath1 data helps lifting this ambiguity . the idea , first developed by tdt , is simple : if the cak is diluted by a truly featureless continuum , and if this fc extends from the optical to the near ir , then the cat would be diluted too , and by a predictable amount . from observations of type 1 seyferts and quasars , we know that the agn optical continuum is well described by a power - law @xmath25 , with @xmath26 in the 12 range or , equivalently , @xmath27 , with @xmath28 . this fc is superposed to the bulge stellar light . suppose that the bulge has intrinsic cak and cat equivalent widths @xmath29 and @xmath30 , and that the ratio of the stellar continuum fluxes over these two lines is @xmath31 . trivial algebra then leads to the following relation between the observed cak and cat equivalent widths : @xmath32 @xmath33 where @xmath34 is the fraction of the total continuum at 3933 which comes from the fc . we modeled a mixing of a fc and bulge continuum in a form of `` fc + ssp '' mixing lines , where ssp refers to a simple stellar population representative of 10 gyr old , typical of elliptical galaxies and bulges , as obtained from bruzual & charlot ( 2003 , bc03 ) models for solar metallicity . practical use of equation [ eq : dilution ] requires stipulating fiducial values for @xmath29 , @xmath35 and @xmath36 . we adopt @xmath37 and @xmath38 . equation [ eq : dilution ] is over plotted to the data points in the @xmath1 versus @xmath4 diagram ( fig . [ fig_cat_x_cak_a ] ) . the lines indicate the bulge + fc mixing lines for different values of the slope @xmath26 and for @xmath39 decreasing from 100% at @xmath40 to 0% at @xmath41 . the @xmath4 measurements come from the optical studies of cid fernandes ( 1998 , 2001 , 2004 ) . of the 78 galaxies in our sample , 42 overlap with these studies , most of which ( 31 ) correspond to seyfert 2s . we see that some seyfert nuclei line up along the mixing lines . however , in order to explain all of the objects , it is necessary a contribution of @xmath39 of about 50% or greater . there are some seyfert with high @xmath1 and low @xmath4 that are not explained in terms of these mixing lines for @xmath26 between 1 ( @xmath42=1 ) and 2 ( @xmath42=0 ) , which are representative of type 1 agn . for these objects , only if @xmath26 were as low as @xmath43 or less the dilution lines would go approximately through the observed values , but not even quasars have such an extremely blue continuum . such a blue fc could in principle be produced by scattering of the agn light by dust particles . nevertheless , the required values of the @xmath39 fraction ( of the order of @xmath44 ) would imply that scattered broad lines in the optical should become visible , in which case the galaxy would not be classified as a seyfert 2 in the first place ( an argument first put forward by cid fernandes & terlevich 1995 ) . besides that , about half of the seyfert 2s with @xmath39 greater than 50% show high order balmer lines ( hobl ) in absorption in their spectra ( gonzlez delgado 2001 ) . so , not diluted hobl argue against a dilution by a power - law in these objects . ) for bulge plus fc power - laws , with slopes @xmath26 = 1 and 2 ( typical range of type 1 agn ) , as labeled . upper scale shows the percentage of @xmath39 contribution . symbols as in fig . [ fig_cat_x_pl].,width=340 ] we have to point out that `` fc + ssp '' in fig . [ fig_cat_x_cak_a ] are computed with a solar ssp values . for another metallicities , fidutial numbers are a litlle different ( mainly lower @xmath1 ) , and the lines would go down for about 1 . in that way , mixing lines in fig . [ fig_cat_x_cak_a ] should be considered as an upper limit for `` fc + ssp '' models . we also computed `` fc + e '' models ( not shown in fig . [ fig_cat_x_cak_a ] ) , where e refers to ngc 2110 , the earliest ( t=-3 ) seyfert 2 galaxy in our sample for which we have determined @xmath4 and @xmath1 . for this model , @xmath45 , and the dilution lines would stay even more distant from the data for the same range of @xmath26 considered before . not even for @xmath26 as low as @xmath13 0 ( not realistic ) the models would along up the data . so , our main conclusions about fc + bulge mixing models for seyfert 2 nuclei remains valid . tdt have performed the same analysis , but using the mgib line at 5175 instead of cak ( see also jimnez - benito 2000 ) . they conclude that if the observed dilution of mgib is due to a power - law fc of reasonable slope , then the cat would also be diluted , contrary to their empirical result . here , we considered `` fc + ssp '' mixing models with some reasonable combination of @xmath29 , @xmath30 and @xmath36 values representative of old stellar populations . these models shows that dilution by a power - law fc seems to be not completely adequate to account for the observed strengths of optical and near - infrared ranges simultaneously . hence , a simple mixture of a bulge - like component plus an agn - like fc , envisaged in the early days of agn research ( e.g. , koski 1978 ) , is just not a viable description of the data . we will analyse another explanation for @xmath4 and @xmath1 behavior in the next section . now that we considered dilution by non - stellar light in the cat range in type 2 seyferts , we will proceed to a stellar population analysis . the behavior of the cat equivalent width as a function of stellar parameters ( @xmath46 , @xmath47 , [ fe / h ] ) has been addressed by several studies ( e.g. , daz , terlevich & terlevich 1989 ) , and this knowledge has been incorporated into population synthesis models which map the cat dependence on the age and metallicity of coeval stellar populations ( e.g. , idiart , thevenin & de freitas pacheco 1997 ) . in a recent and comprehensive study , cenarro and coworkers revisited this topic , analyzing the cat in a large library of stars and producing state - of - the - art predictions for the behavior of the cat in integrated stellar populations ( cenarro 2001a , b,2002 ; vazdekis 2003 ) . despite early hopes that the cat could serve as a metallicity indicator for galaxies ( as it does for globular clusters , at least below @xmath48 ; armandroff & zinn 1988 ; idiart 1997 ; vazdekis 2003 ) , or as tracers of recent star - formation through its sensitivity to the presence of red super - giants ( hereafter rsg ; see tdt ; garca - vargas , moll & bressan 1998 ; mayya 1997 ) , observations have revealed that the cat shows remarkably little variation in strength for galaxies spanning wide ranges in morphological and stellar population properties . this small sensitivity is qualitatively consistent with the new generation of synthesis models ( vazdekis 2003 ) , which show that , for metallicities higher than @xmath49 dex , this feature varies little as a function of age and metallicity . overall , these results raise serious doubts as to the usefulness of the cat as a stellar population diagnostic . in this section we take another look at this issue , tackling it from an empirical perspective . as in the previous section , information on the cak line will be incorporated in the analysis . let us assume that strongly diluted cak lines in seyfert 2s are the result of the presence of a starburst component , in line with previous investigations . in particular , cid fernandes ( 2001 ) propose that seyfert 2s with @xmath50 can be safely identified as composite starburst + agn systems , where the starburst is unambiguously identified by independent features such as the wr bump or pcygni line profiles in the uv . investigating the behavior of @xmath1 as a function of @xmath4 thus provides a fully empirical test of whether the cat is or is not sensitive to stellar populations . a cat versus cak diagram was already shown in fig . [ fig_cat_x_cak_a ] , where most points correspond to seyfert 2s . the plot shows that @xmath1 bears little , if any , relation to @xmath4 . it is therefore hard to evade the conclusion that for the mean ages and metallicities of metal rich stellar populations , @xmath1 is a poor diagnostic : objects as diverse as mrk 1210 , which contains a powerful young starburst , including wolf - rayet stars ( storchi - bergmann , cid fernandes & schmitt 1998 ) , and ngc 7172 , which is dominated by @xmath51 gyr stars ( cid fernandes et al 2004 ) have @xmath1 values indistinguishable within the errors ( @xmath52 and @xmath53 , respectively ) . ( left ) and @xmath4 ( right ) in seyfert 2s . note the contrast of the behavior of @xmath4 , which is very sensititive to young stars , with the approximate constancy of @xmath1.,width=340 ] in fig . [ fig_perc_x_ews ] we combine our cat data with the results of the detailed spectral synthesis analysis of type 2 seyferts carried out by cid fernandes ( 2004 ) for 14 sources in common with our sample . we compare their young stars ( age @xmath54 myr ) plus `` fc '' light fractions at 4020 ( @xmath55 ) with both @xmath1 ( left panel ) , and @xmath4 ( right ) . as discussed in that paper , which dealt exclusively with spectra in the 35005200 range , what the synthesis returns as a fc is in most cases a reddened starburst component instead of a true non - stellar component , such that @xmath55 is in practice a measure of the starburst strength . the near constancy of @xmath1 , already evident in fig . [ fig_cat_x_cak_a ] , is even more striking in fig . [ fig_perc_x_ews ] . the inexistent correlation between cat strength and @xmath55 contrasts with the strong decrease in @xmath56 as the starburst component increases in strength . again , one is lead to the conclusion that , by itself , @xmath1 is not a stellar population tracer for our galaxies . the most natural interpretation of figs . [ fig_cat_x_cak_a ] and [ fig_perc_x_ews ] is that the dilution of optical lines in some seyfert 2 nuclei is due to a young stellar population which causes little or no dilution of the cat . in this section we use models to check whether this is a viable scenario . to investigate the behavior of the cat and cak lines as a function of stellar population properties , we have computed the evolution of @xmath1 and @xmath4 for ssp for 4 metallicities . bc03 evolutionary synthesis models were used for this purpose . ssp models are computed with the stelib library ( le borgne 2003 ) , chabrier ( 2003 ) mass function and padova 1994 tracks ( see bc03 for details ) . because @xmath1 , @xmath11 and @xmath4 are not provided in the standard bc03 distribution , they were computed directly from the theoretical spectral energy distributions , as explained at the beginning of section [ sec : cat_nonstellarlight ] . for compatibility with our @xmath12 measurements , we broadened all model spectra as explained at the beginning of section [ sec : cat_nonstellarlight ] , adopting for the instrumental resolution that of the stelib library ( @xmath57 km / s at @xmath58 ) . , @xmath1 and @xmath59 absorption line indices for ssp , from bc03 models . long dashed , dashed , solid and dotted lines correspond to metallicities of 0.2 , 0.4 , 1 and 2.5 @xmath60 , respectively.,width=340 ] fig . [ fig_bc03_x_t_a ] shows the results for four different metallicities : 0.2 , 0.4 , 1 and 2.5 @xmath60 . @xmath4 is practically null during the first @xmath61 yr . after these initial phases , the cak strength increases monotonically with time , which makes it good to discriminate very young ssp ( younger than @xmath62 yr ) from ssp older than @xmath63 gyr . on the other hand , @xmath1 starts from small values up to a few myr , when the first rsg appear . the exact timing and duration of this phase depends on the metallicity . from @xmath51 myr to about 1 gyr , @xmath1 oscillates as rsg are replaced by red - giants . throughout these different phases , however , @xmath1 spans a small range , from @xmath64 to 8 . taking into account the uncertainties in both models and data , this is in practice a constant value . only in the first myrs of evolution does @xmath1 have a useful diagnostic power , as it assumes much smaller values . notice that in these early phases @xmath1 is contaminated by paschen lines , which explains why it does not start from 0 at @xmath65 . in fig . [ fig_cat_x_cak_b ] these models are overlayed to the data in the @xmath1 versus @xmath4 diagram . seyfert 1s are not plotted because of their non - stellar continuum , which affects both cak and cat . numbers along the model tracks mark ages of @xmath66 , @xmath67 , @xmath68 , @xmath69 and @xmath70 yr . the first thing to notice is that , unlike the bulge + fc mixing lines in fig . [ fig_cat_x_cak_a ] , these models do span the region occupied by objects in our sample , showing that stellar populations alone are capable of explaining the behavior of seyfert 2s in the cak versus cat diagram . secondly , fig . [ fig_cat_x_cak_b ] seems to indicate that sub - solar abundances may be required to match the data on seyfert 2s , particularly those with weak cat . this is somewhat surprising , given that we are dealing with the nuclei of early type spirals , where @xmath71 is expected to be relatively high . one must nevertheless recall that these are ssp models , whereas actual galaxies contain a mixture of stellar populations . this is particularly true for type 2 seyferts , where previous studies have revealed a wide variety of star formation histories ( cid fernandes 2001 ; 2004 ; wild 2007 ) . more realistic star formation histories can be modeled as a mixture of ssp , whose effect on this diagram can be conceived drawing imaginary mixing lines connecting any series of points along the instantaneous burst tracks . to have a rough idea of the effects of such a mixture , in fig . [ fig_cat_x_cak_c ] we repeat the spectral mixing experiment of fig . [ fig_cat_x_cak_a ] but this time using a 1 myr population ( to represent ongoing star - formation ) instead of a non stellar power - law continuum . solar and 2.5 solar bc03 models are used . we further allow for the possibility that the young population suffers more extinction than the old one , so we consider three cases , @xmath72 = 0 , 1.5 and 3 mag . again , we do not plot seyfert 1 nuclei , as they present dilution due to the non - stellar continuum and thus are out of the scope of this experiment . this exercise is essentially the same as that performed with a power - law fc in fig . [ fig_cat_x_cak_a ] , except that the continuum shape is not the same , and , due to contamination by paschen lines , @xmath1 is not @xmath73 for a pure starburst ( see fig . [ fig_bc03_x_t_a ] , middle panel ) . as is evident in the figure , these mixing lines span the observed points for most of the sy 2 sample . it is important to emphasize that only for very young populations one expects a relatively low cat . as shown in the plot , galaxies with @xmath74 and @xmath75 require such a population to be adequately modeled . not even the @xmath76 @xmath60 models are capable of modeling these objects without a very young population . it is also important to note that although our 1 myr + 10 gyr mixing models span most of the observed points in the cak versus cat plane , galaxies with @xmath1 above @xmath64 can be equally well modeled with an older young population . for metallicites @xmath77 @xmath60 , @xmath1 rises above 6 as early on as @xmath78 myr , with the appearence of the first rsgs . this also happens for continuous star formation models , as shown by the solid line in the cat - cak diagram . mixing models for @xmath79 myr + 10 gyr populations are not shown for clarity , but , as can be deduced from fig . [ fig_cat_x_cak_b ] , they would span most of the upper part of cak versus cat plane , where many sy 2s are located . we note in passing that tdt originally suggested that the cat in seyfert 2s signal the presence of rsg in a starburst . while this may well be true in some cases , our modeling shows that there is no way of telling it just from the strength of the cat , since we are able to fit the @xmath1 and @xmath4 data both with and without rsgs . given the simplicity of these models , the overlap between data and models in the cat - cak diagram is highly encouraging . nevertheless , at the middle of the diagram there are two sy 2 that are far away from our models . the two `` outliers '' are mrk 3 and ngc 7212 . it may be that a bona - fide agn continuum is present in these galaxies , which would help explaining their low cak and specially cat lines . the spectral synthesis analysis of ngc 7212 by cid fernandes ( 2004 ) indicates that @xmath80 of the light at 4020 comes from an fc component . they also identify a weak broad h@xmath42 , strengthening the case for the presence of a non - stellar continuum . furthermore , for both ngc 7212 and mrk 3 , imaging ( pogge & de robertis 1993 ; kotilainen & ward 1997 ) , spectropolarimetry ( tran 1995 ) , and spectral modelling ( gonzlez delgado 2001 ) all point to the existence of an fc component ( see cid fernandes 2001 for a detalied discussion ) . none of these studies find compelling evidence for significant on - going star - formation in these galaxies . it is therefore likely that their low @xmath1 values ( the smallest amongst seyfert 2s ) are indeed due to dilution by non - stellar light , explaining why they deviate from the region spanned by our purely stellar models in fig . [ fig_cat_x_cak_c ] . but without seyfert 1 nuclei and over plotting simple stellar populations , as computed from the bc03 models . lines as in fig . [ fig_bc03_x_t_a ] . numbers in stellar lines denotes log(age[yr ] ) . symbols as in fig . [ fig_cat_x_pl].,width=340 ] now overploting old plus redenned young stellar population models for solar and 2.5 solar metallicity . upper scale shows the percentage of the starburst contribution in the cak region , @xmath81 . mixture lines corresponds to different extinction magnitudes and the asterisks denote null ( @xmath82 , upper right ) and full ( @xmath83 , lower left ) young contribution . the csf line refers to continuous star formation from ages between 1 myr and 10 gyr . dots on this line indicates bursts of 5 and 10 myr old.,width=340 ] a consistent scenario thus emerges , where the blue continuum which dilutes the optical spectrum of some seyfert 2s is due to young stellar populations , while their undiluted cat strengths can be explained as due to rsgs or older populations . as explained above , in the few cases where the cat is low , a combination of an old plus a very young ( pre - rsg ) population is needed . in these cases , one may expect to see paschen lines in absorption , which is not observed in our galaxies , as reported in paper i. to investigate this potential problem we have examined the strength of paschen lines in the theoretical spectra of the mixing models shown in fig . [ fig_cat_x_cak_c ] . we pay particular attention to the pa14 line at @xmath84 , which is strategically placed between ca2 and ca3 . we find that , for @xmath85 , pa14 only becomes visible for @xmath86 or @xmath87 for z = 1 and 2.5 @xmath88 , respectively . these fractions falls to about 75 @xmath21 is we consider 3 magnitudes of differential extinction . these very large fractions result from the fact that the young population is much bluer than the old one , so any mixture where young stars contributes significantly to the cat range , automaticaly implies that this same component completely dominates in the optical . spectral synthesis studies in the optical range never find such huge starburst contribution in seyfert 2s . thus , the non detection of paschen lines may be simply due to the fact that old stars dominate the flux in the near ir , even when younger populations dominate the optical light . a further factor which may be related to the non detection of paschen absorption is filling by an emission component , powered either by the agn or a circumnuclear starburst . a few of our galaxies do show a hint of pa14 emission , like the starburst galaxy ngc 7714 , where pa14 has about 0.5 emission equivalent width ( see also gonzlez delgado 1995 ) . in other cases , emission and absorption may approximately compensate each other . this seems to be the case in the super star clusters of ngc 1569 , where wr features , high - order balmer absorption lines and the cat are all clearly detected ( gonzlez delgado 1997 ) , but paschen lines are not seen either in absorption or emission . hence , even though further studies concerning paschen lines are desirable , their no detection do not invalidate our conclusions . all cat data reported so far pertains to nuclear regions . for 34 of the 78 galaxies studied in paper i our long - slit data are of sufficient quality to warrant the extraction of spatially resolved spectra . detailed results for all such sources are presented in asari ( 2006 ) . in this section we summarize the main results concerning the spatial behavior of the cat strength . [ fig : spatial_wcat ] shows @xmath1 spatial profiles for 6 galaxies , chosen to represent seyfert 1s ( panels a and b ) , seyfert 2s ( c and d ) and composite starburst + seyfert 2 ( e and f ) . there is a clear drop in @xmath1 as one approaches the nucleus in seyfert 1s , whereas seyfert 2s present remarkably flat @xmath89 profiles . at a few hundred pc from the nucleus , however , the cat strength becomes similar for both seyfert types . this illustrates once again how sensitive @xmath1 is to an underlying fc , in agreement with the results reported in section [ sec : cat_nonstellarlight ] . in a way , these results are arguably more convincing proof of the fc effects . since the analysis in section [ sec : cat_nonstellarlight ] was based on a comparison of cat strengths for seyfert 1s and 2s , or with fiducial values , one could worry that for some reason the stellar populations of the two kinds of agn are intrinsically different , invalidating a comparative analysis . here , on the other hand , we show that at least immediately outside their nuclear regions , seyfert 1s and 2s are very similar , and that dilution is really a nuclear effect present only in type 1s . to further illustrate this , fig . [ fig_hist_ews_spatial ] repeats the comparison performed in fig . [ fig_hist_ews ] , but this time using off - nuclear @xmath1 measurements . depending of the galaxy redshift , the furthest off - nuclear measurements correspond to less than 1 kpc to @xmath13 3 kpc . in fig . [ fig_hist_ews_spatial]a we proceed as follows : in the first cases ( nearest galaxies ) we take the furthest points from the nucleus ( for instance , ngc 1433 in fig . [ fig : spatial_wcat ] ) , and in the second cases ( furthest galaxies ) , we firstly obtained the interpolated cat values at exactly 1 kpc taking into account the data at each side of the nucleus , and then we calculate the average of the interpolated values ( for instance , ngc 3081 in fig [ fig : spatial_wcat ] ) . we see that the differences between seyfert 1s and 2s disappear when considering these off - nuclear values , which we call `` _ @xmath901kpc _ '' data . [ fig_hist_ews_spatial]b does the same thing , but using the values corresponding to the furthest points from the nucleus for which we have signal , and we call them `` _ furthest _ '' data . numerical results of @xmath1 spatial measurements are summarized in table [ tablespatialcat ] . a ks test applied to the distributions yields probablitities of @xmath91 ( `` _ @xmath901kpc _ '' ) and @xmath92 ( `` _ furthest _ '' ) . we see that even for short distances ( @xmath90 1 kpc ) there is no significant dilution of cat in seyfert 1s , as we expected from the regions outside the nuclei . but for off - nuclear measurements : ( a ) 1 kpc data , ( b ) furthest point for which we have signal ( see text for details ) . scale and areas as in fig [ fig_hist_ews].,width=340 ] .statistics for spatial @xmath1 measurements : minimum , median and maximum values are shown for seyfert 1 and seyfert 2 and for nuclear and off - nuclear data . for off - nuclear data , we considered _ @xmath901 kpc _ data , which refers to the values at 1 kpc or less from the nucleus , and _ furthest _ data , which are the corresponding values of furthest points from the nucleus of each galaxy for which we have signal . [ cols="<,^,^,^",options="header " , ] it is also interesting to point out that these results are essentially identical to those obtained from the spatial analysis of the cak strength by cid fernandes ( 1998 ) and storchi - bergmann ( 1998 , their fig . 1 ) . hence , the fc which dilutes the optical spectrum in seyfert 1s is the same which decreases the cat strength in the near ir . this component is not present in type 2s . the behavior of @xmath89 for starburst @xmath93 seyfert 2 composites ( figs . [ fig : spatial_wcat]e and f ) , on the other hand , contrasts with that derived from @xmath4 radial profiles . contrary to the strong spatial dilution by young stars detected in the cak and other optical lines , the cat profiles show very little , if any , variation as a function of distance to the nucleus in composite systems . in ngc 7582 there is a marginal hint of dilution . from optical work we know that this galaxy hosts a very dusty central starburst , so it could represent a case where the starburst light does contribute to the flux in the cat range with a non negligible fraction . however , even in this most favorable case the nuclear cat is diluted by less than 15% with respect to its value at @xmath94 pc . in contrast , the optical absorption lines are much more diluted : at the wavelengths of the cak , g - band and mgib lines , the dilution factors are 70 , 60 and 57% , respectively . to better appreciate this strong difference , we invite the reader to compare fig . [ fig : spatial_wcat]e with fig . 29 of cid fernandes ( 1998 ) , where the radial profiles of several optical absorption lines in ngc 7582 are shown . the difference in @xmath1 and @xmath4 profiles is striking , but not surprising . in fact , our finding that the cat does not mimic the cak spatial behavior is fully consistent with the conclusion laid out in the section above that only under extreme conditions ( very young age and large extinction ) would a starburst strongly dilute both cak and cat lines . our spectroscopic data ( paper i ) provides estimates of two characteristic velocities : the typical velocity of stars through the host s bulge , @xmath6 , and the typical nlr cloud velocity dispersions , inferred from the width of the [ siii]@xmath09069 line : we fitted the profile of [ siii]@xmath09069 with one or two gaussians , one representing the core of the emission line and the other , if necessary , representing the wings . the resulting fwhm of the core component divided by @xmath95 is what we call @xmath8}$ ] . we have determined @xmath8}$ ] by this method for 31 seyfert and 4 starburst nuclei . the distributions obtained for @xmath6 and @xmath96}$ ] are shown in fig . [ fig_histograms_vd ] . they show that both types of nuclei have essentially the same ( statistical ) values : for seyfert 1s @xmath6 = 128 @xmath22 37 km / s and @xmath96}$ ] = 111 @xmath22 29 km / s , while for seyfert 2s @xmath6 = 134 @xmath22 45 km / s and @xmath8}$ ] = 137 @xmath22 76 km / s . firstly , we will discuss the stellar velocity dispersions ( @xmath6 ) and then we will analyze the nlr kinematics through @xmath8}$ ] and its possible link to @xmath6 . , top ) and [ siii]-based gas velocity dispersions ( @xmath8}$ ] , bottom ) . areas as explained in fig . [ fig_hist_ews].,width=340 ] as reviewed in section [ sec : introduction ] , recent work on normal galaxies has dedicated much attention to the fact that @xmath1 is slightly anti - correlated with velocity dispersion all the way from dwarf to giant elliptical galaxies and bulges ( saglia 2002 ; cenarro 2003 ; falcn - barroso 2003 ; michielsen 2003 ) . even though our cat data are not of the same quality as that employed in these studies , it is interesting to check whether starburst and active galaxies follow the same @xmath12-@xmath6 as normal ones . .,width=340 ] fig . [ fig_cat_x_vd_all ] shows the results of this test . clearly , as a whole , our galaxies do not follow a well defined @xmath12-@xmath2 relation at all . the few normal galaxies in our sample ( crosses ) , as well as the high-@xmath2 seyfert 2s , do tend to follow a similar anti - correlation found by saglia et al ( 2002 ) for ellipticals . however , there is a large population of galaxies with low-@xmath2 and relatively low @xmath1 which has no correspondance in previous studies of this relation . excluding mrk 3 and ngc 7212 ( which are diluted by fc , as discussed in [ mixed models ] ) , the most deviant points are the starbursts ngc 1140 , ngc 3256 and ngc 7714 , located in the bottom left part of the plot . next , at @xmath1 between 5 and 6.5 and @xmath97 150 km / s , are the composite starburst - seyfert 2s nuclei , like ngc 5135 , ngc 7582 , ic 3639 , mark 1 . it thus seems that the complex star formation histories of these objects are somehow related to their displacement with respect to the @xmath12-@xmath2 relation defined for more well behaved systems . interestingly , the normal galaxy ngc 205 is located away the anticorrelation stated for normal nuclei . this dwarf elliptical galaxy has a low metallicity ( mateo , 1998 ) , which indicates that it could also influence the cat-@xmath2 relation . the question concerning the acceleration of gas in the nlr has been analyzed in several papers ( nw , tdt ) . as pointed out by green & ho ( 2005 ; gh ) , the nlr is small enough to be illuminated by the active nucleus and large enough to feel the gravitational forces of the bulge of the host galaxy . comparison between gas and stellar kinematics was previously carried out by nw using the [ oiii]@xmath05007 line , who find that @xmath6 and the fwhm of [ oiii]@xmath05007 are correlated , but with substantial scatter . recent work on seyfert 1s ( jimnez - benito 2000 ) and sdss galaxies ( heckman 2004 , gh ) confirms this finding . the implication is that the nlr clouds are at least partly dominated by virial motions in the host galaxy s bulge . given the existence of this relation , forbidden line widths have sometimes been used as a surrogate for @xmath6 in the absence of information on the stellar dynamics ( nelson 2000 ; grupe & mathur 2004 ) . the case of narrow - line seyfert 1 galaxies ( nls1 ) is particularly interesting . in a recent paper , komossa and xu ( 2007 ) have shown that , for these seyfert galaxies , @xmath98}$ ] can be used as a proxy for @xmath2 only after removing the ( usually blue ) wings of the [ oiii ] profile . in other words , @xmath98}$ ] should correspond only to the core of the line , similarly to what we have done in the present paper in the fit of the [ siii ] emission - line profiles to obtain @xmath8}$ ] . based on the width of the [ siii]@xmath09069 line . symbols as in fig . [ fig_cat_x_pl ] . ordinary least squares regressions are applied over all data but mrk 1 and ngc 1068 : solid , @xmath99 bisector fit ; dashed , @xmath100 and @xmath101 ( shallowest and steepest lines , respectively).,width=340 ] in fig . [ fig_vdsiii_x_vdstar ] we compare our values of @xmath8}$ ] and @xmath6 for all seyfert nuclei . we note that two of them , namely mrk 1 and ngc 1068 , have @xmath96}$ ] much larger than @xmath2 . evidence of jets were found in these galaxies , which may explain the acelleration of the gas via shocks produced in the interaction within jets and nlr . excluding these two nuclei , we have fitted all the other galaxies by linear relations given by ordinary least square ( @xmath99 ) regressions ( isobe et al , 1990 ) . solid line represents the @xmath99 bisector fit , while dashed lines are @xmath100 and @xmath101 ( shallowest and steepest lines , respectively ) . the ols bisector fit gives a slope of 0.93 ( n = 29 , @xmath102 = 0.37 , p = 0.046 ) . taking into account just seyfert 2s , we find a slope of 1.03 ( n = 21 , @xmath102 @xmath13 0.40 , p = 0.07 ) and 0.73 for seyfert 1s , for which there is not a clear correlation ( @xmath102 @xmath13 0.31 , p = 0.45 ) . the slopes and correlation coefficients do not vary substantially when we exclude worst quality data . using the @xmath103 very good , @xmath104 good , and @xmath105 not - too - bad quality flags adopted in paper i , we find for @xmath106 a slope of 1.00 ( n = 23 , @xmath102 = 0.39 , p = 0.06 ) , and 1.00 for @xmath107 sources alone ( n = 20 , @xmath102 = 0.3 , p = 0.2 ) . all these is summarized in table [ ols_for_[siii ] ] . interestingly , there is also a tendency for 4 starburst nuclei , with a slope of about 0.56 and @xmath102 @xmath13 0.81 ( not shown here ) . lccccccl + & all & sy 1 & sy 2 & all ( @xmath106 ) & all ( @xmath107 ) + slope & 0.93 & 0.73 & 1.03 & 1.00 & 1.00 + n & 29 & 8 & 21 & 23 & 20 + @xmath102 & 0.37 & 0.31 & 0.40 & 0.39 & 0.30 + p & 0.046 & 0.45 & 0.07 & 0.06 & 0.20 + we repeated this analysis using @xmath108 } } = { \rm fwhm}_{\rm [ oiii ] } / ( 8 \ln 2)^{1/2}$ ] taken from the literature ( whittle , 1992 ; bassani 1999 ; schmitt 2003 ) . ngc 1068 , mrk 1 and mrk 78 present very different values of @xmath108}}$ ] and @xmath2 . further excluding the worst quality data ( `` flag @xmath109 '' in whittle 1992 and our flag @xmath110 ) the ols bisector fit gives a slope of 1.48 ( @xmath111 ) for the whole sample ( 28 objects ) , while splitting into seyfert 1s and 2s the slopes are 1.13 ( @xmath112 ) and 1.60 ( @xmath113 ) respectively . this is described in table [ ols_for_[oiii ] ] . we point out that for this analysis we used fwhm data of [ oiii ] line , which are ( in median ) larger than our estimates of @xmath3 through the fwhm of the @xmath114 of [ siii ] line , so that the slope of a fit applied to [ oiii ] data yields larger values than those obtained for [ siii ] data . we confirm this fitting @xmath115}}$ ] vs @xmath108}}$ ] , which gives a slope of @xmath13 0.85 ( @xmath102 @xmath13 0.44 , 20 objects ) . lccccccl + & all & sy 1 & sy 2 + slope & 1.48 & 1.13 & 1.60 + n & 28 & 12 & 16 + @xmath102 & 0.42 & 0.81 & 0.33 + . [ ols_for_[oiii ] ] there are , therefore , correlations between nlr and stellar motions , but with considerable scatter . one way to study this scatter is to look at the residuals @xmath115}}$]/@xmath2 about unity . fig . [ fig_histogram_vdratio ] shows this distribution , which is a moderately peaked ( kurtosis @xmath13 -1 ) gaussian for the majority of seyfert 2s . without taking into account the two galaxies with abnormally high gaseous velocity dispersion ( for which it was shown that there are several kinematical components ) , @xmath115}}$]/@xmath2 = 0.97 @xmath22 0.32 ( average and rms dispersion ) , while for seyfert 1s and 2s @xmath115}}$]/@xmath2 are 1.02 @xmath22 0.40 and 0.95 @xmath22 0.29 , respectively , i.e. , @xmath115}}$]/@xmath2 is the same for both types of objects . [ fig_histogram_vdratio ] is similar to fig . 7 of nw , which uses [ oiii]-based gas kinematics . although nw have twice as many objects , we note that our distribution is more concentrated around unity , mainly because we have fwhm determinations of gaussians fitted to the _ core _ of the [ siii ] line . all this study shows that the [ siii ] core emitting zone of the nlr is dynamically linked to the gravitational bulge potential , a result similar to that obtained by komossa and xu ( 2007 ) for nls1 galaxies . however , we see that there are possible secondary factors besides gravitational potential that could explain the scatter about unity in nlr kinematics . shocks and other acceleration mechanisms could account for the high velocity clouds observed in a few galaxies , but in general a complete study of gas - stellar kinematics remains to be done . } } $ ] /@xmath2 . areas as explained in fig . [ fig_hist_ews].,width=340 ] a detailed analysis of stellar and gaseous kinematics of nlr was made by gh taking precisely @xmath115}}$]/@xmath2 as the central quantity of their study . they found that none but one of the nuclear activity indicators is correlated with @xmath115}}$]/@xmath2 ; only with @xmath116 , which is indirectly a measure of bulge mass , there is a hint of relationship . although we have few points , we note this trend in our data , which is shown in fig . [ fig_ratiovd_x_vdstar ] together with gh s best fit . again , applying ols bisector fit , we obtain @xmath117}}$]/@xmath118 , with @xmath119 . a least squares fit gives @xmath117}}/\sigma_{\star } = -0.56 \log \sigma_{\star } + 1.13 $ ] ( @xmath102 = -0.55 , p = 0.022 ) , nearly the same relation obtained by them . only for seyfert 2s there is a weaker ( but real ) trend , @xmath117}}/\sigma_{\star}=-0.41 \log \sigma_{\star}+0.81 $ ] ( n = 21 , @xmath120 , p = 0.09 ) , while for seyfert 1s the correlation is moderately strong , @xmath121}}/\sigma_{\star}=-0.74 \log\sigma_{\star}+1.50 $ ] ( n = 8 , @xmath122 , p = 0.036 ) . we checked whether @xmath115}}/\sigma_\star$ ] correlates with other properties , like morphological and activity types , redshift , inclination , etc . , but ( like gh ) found no strong trend . the absence of a trend with the inclination of the host galaxy suggests that the gas in the nlr has random velocities and so the velocity field is similar to that of the stars , as previously found by nw ( their fig . our data thus support the existence of an anticorrelation between @xmath8}$ ] and @xmath2 , albeit with a large scatter . using @xmath123}}$ ] as a proxy of @xmath2 tends to underestimate velocity dispersion in bulge of massive galaxies . in any case , given the existence of a correlation between both ( gas and stellar ) velocity dispersions , and taking into account that [ siii ] is a high ionization line produced very near the active nucleus and often with strong wings like [ oiii ] , our study supports the idea of using @xmath115}}$ ] as a kinematic tracer of the nlr . } } $ ] /@xmath2 with @xmath2 . solid line is the least square fit for all seyferts . dotted line , green&ho s fit . symbols as in fig . [ fig_cat_x_pl].,width=340 ] as an extension of our spatial analysis , we report results for the variations in kinematical properties of our sub - sample . fig . [ fig : spatial_sig ] shows examples of spatial variations in @xmath6 . some recent works ( barbosa ( 2006 ) , garca - lorenzo 1999 ; emsellen 2001 ; mrquez 2003 ) have discussed the possibility of a _ sigma - drop _ detection , i.e. , that @xmath2 is appreciably lower towards nuclear regions due to the presence of a star - forming cold nuclear disc . we see a nuclear drop in @xmath2 in only two galaxies ( fig . [ fig : spatial_sig]a and b ) . for some objects there is even a hint of a rise in @xmath2 in their central parts ( e.g. , fig . [ fig : spatial_sig]c and d ) . we have to keep in mind , however , that our spectra do not have the spatial resolution needed to detect this behavior in @xmath2 , which would require a resolution of a few tens parsecs . another interesting feature that can be seen for some of our objects is a symmetric off - nuclear drop in @xmath6 , as in ngc 4593 ( fig . [ fig : spatial_sig]e ) . this is in agreement with the detailed integral field spectroscopy of this object made by barbosa ( 2006 ) , who found a partial ring of low-@xmath2 nearly at the same distance from the nucleus . a similar behavior is seen in ngc 6951 ( fig . [ fig : spatial_sig]f ) . u - band images of these galaxies ( muoz martn et al , 2007 ) shows the presence of young stars at the locations where the off nuclear @xmath2 drops are observed . spatial variations . dotted lines mark @xmath124 to + 200 pc region ; dashed lines , the @xmath125 to + 1 kpc region.,scaledwidth=50.0% ] in paper i we presented a spectroscopic atlas of 78 galaxies in the region of cat lines , and measured the equivalent widths of cat and [ siii]@xmath09069 lines , and stellar and gaseous velocity dispersions . in this paper we used these data , as other measurements , to investigate the cat strength and the kinematical properties of seyfert nuclei . our conclusions can be summarized as follows : 1 . the equivalent widths of cat and [ siii ] are diluted in seyfert 1s , due to the presence of a non - stellar component , while in seyfert 2s there is no sign of dilution . we show that the nuclear dilution of cat lines in seyfert 1s and the non observed dilution in seyfert 2s holds spatially , i.e. , within approximately the central kpc . at this distance the effects disappears , thus implying that the stellar lines are diluted due the nuclear non - stellar continuum directly seen in seyfert 1s . 2 . the cat strength turns out not to be a good tracer for the mean ages and metallicities of metal rich stellar populations . nonetheless , its combination with cak line yields a useful constraint on the nature of the continuum emission from optical to near - ir wavelengths . we show that the location of seyfert 2s in the cat - cak plane could be explained satisfactorily by stellar population synthesis models , by considering mixtures of an old plus very young stellar populations with a extinction of just two or three magnitudes . the hypothesis of seyfert 2 nuclei composition as a mixing of an old stellar population plus power - law central source can not account for the majority of the data . 4 . there is a correlation between nuclear @xmath6 and @xmath8}$ ] , as well as an anticorrelation between @xmath8}$]/@xmath2 and @xmath2 , both with substantial scatter . this means that care must be exercised when using @xmath8}$ ] as a proxy for @xmath6 . these results are compatible with previous results of nw and gh . lrv , nva , rcf , agr and tsb acknowledge the support of capes and cnpq . lrv acknowledge the support from secyt , and the hospitality of ufsc . rgd acknowledges support by spanish ministry of science and technology ( mcyt ) through grant aya-2001 - 3939-c03 - 01 . we thank laboratrio nacional de astrofsica for the allocation of time on eso 1.52 m and financial support during the runs . we also thank the referee , javier cenarro , for the very useful comments which greatly improved the content of the paper . part of the data described here were taken at kitt peak national observatory , national optical astronomy observatories , which are operated by aura , inc . , under a cooperative agreement with the national science foundation . basic research at the nrl is supported by 6.1 base funding . this research made use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , caltech , under contract with nasa . tremaine , s. ; gebhardt , k. ; bender , r. ; bower , g. ; dressler , a. ; faber , s. m. ; filippenko , a. v. ; green , r. ; grillmair , c. ; ho , l. c. ; kormendy , j. ; lauer , t. r. ; magorrian , j. ; pinkney , j. ; richstone , d. , 2002 , apj , 574 , 740

we present a comprehensive analysis for the determination of the confusion levels for the current and the next generation of far - infrared surveys assuming three different cosmological evolutionary scenarios . we include an extensive model for diffuse emission from infrared cirrus in order to derive absolute sensitivity levels taking into account the source confusion noise due to point sources , the sky confusion noise due to the diffuse emission , and instrumental noise . we use our derived sensitivities to suggest best survey strategies for the current and the future far - infrared space missions _ spitzer _ , _ akari _ ( _ astro - f _ ) , _ herschel _ , and _ spica_. we discuss whether the theoretical estimates are realistic and the competing necessities of reliability and completeness . we find the best estimator for the representation of the source confusion and produce predictions for the source confusion using far - infrared source count models . from these confusion limits considering both source and sky confusions , we obtain the optimal , confusion limited redshift distribution for each mission . finally , we predict the cosmic far - infrared background ( cfirb ) which includes information about the number and distribution of the contributing sources . [ firstpage ] cosmology : observations infrared : galaxies galaxies : evolution ism : structure methods : observational

astro-ph0603163

the detection limits of space infrared ( ir ) telescope systems are governed by instrumental , photon , and confusion noise sources . the instrumental noise is intrinsic to the detector and electronics system , e.g. , readout noise and dark current fluctuations . the photon noise is due to poissonian fluctuations in the photon counts from the sky background emission and the telescope thermal emission which can be reduced by decreasing the telescope temperature . on the other hand , the confusion noise is due to both the superposition of point sources in crowded fields and the brightness fluctuation of extended structures at scales of order of the telescope beam size . in the far - ir bands , due to the relatively large beam sizes , astronomical observations are mostly affected by the sky confusion due to the galactic cirrus structure @xcite and the source confusion due to the unresolved extragalactic sources . though the confusion level is the unavoidable limit for the detection of faint point sources , the source structure below the confusion limit also creates background fluctuation in the observed image . this means that the source distribution can be obtained by measuring the background fluctuation which results from a convolution of the background sources and the telescope optics @xcite . after the successful infrared astronomy satellite mission ( _ iras _ ) @xcite , subsequent space ir telescopes such as the infrared space observatory ( _ iso _ ) @xcite and _ spitzer _ @xcite , have increased the levels of sensitivity and spatial resolution at far - infrared wavelengths . in addition , the japanese _ akari _ ( formerly known as _ astro - f _ ) satellite was launched successfully on februrary 21st 2006 @xcite , and will perform an all sky survey in 4 far - ir bands with much improved sensitivities , spatial resolutions and wider wavelength coverage than the previous _ iras _ survey over two decades ago . moreover , within the next decade , the herschel space observatory ( _ herschel _ ) @xcite and the space infrared telescope for cosmology and astrophysics ( _ spica _ ) @xcite will observe the much deeper universe with large aperture sizes of 3.5 m. in our previous paper ( jeong et al . 2005 , hereafter paper i ) , we numerically generated a high spatial resolution map of the galactic cirrus and investigated the effect of sky confusion for current and future space far - ir missions . in addition to the simulated sky confusion noise , this work includes realistic source confusion effects assuming various cosmological evolutionary scenarios . we also obtain the expected cosmic far - ir background ( cfirb ) . the fluctuations in the surface brightness of extended structure on scales smaller than the resolution of the telescope / instrument beam can produce spurious events that can be easily mistaken for genuine point sources , since the existence of a source is usually simply derived from the difference in signal between the on source and some background ( off source ) position . such extended structure is observed in wispy neutral interstellar dust in the milky way that is heated by the interstellar radiation field and is known as the infrared cirrus @xcite , and is one of the main noise sources in the far - ir range . cirrus emission peaks at far - ir wavelengths but was detected in all 4 _ iras _ bands . there have been realistic estimations of the sky confusion from observational data by _ iras _ and _ iso _ @xcite . the intensity of galactic cirrus is a function of galactic latitude and is serious for wavelengths longer than 60 @xmath0 m . using high resolution maps extended from currently available galactic emission maps , we have estimated the sky confusion noise for various missions , based on fluctuation analysis and detailed photometry over realistically simulated images . in paper i , we concluded that the sky confusion is expected to be two orders of magnitude lower for the next generation of space missions with larger aperture sizes such as _ herschel _ and _ spica _ , but on the other hand , current 6090 cm aperture missions such as _ spitzer _ and _ akari _ , will have to endure a sky dominated by confusion noise at long wavelengths ( @xmath1 m ) . in this paper , we also discuss the composite effect of sky confusion and source confusion from the implementation of a realistic model for the source distribution . note that additional components affecting the confusion noise are the presence of faint , unresolved , asteroids and the zodiacal light . though bright asteroids can be easily rejected via source confirmation between 2 pointings / scans over the same sky position , faint asteroids may contribute to fluctuations assumed to be from faint galaxies . although @xmath2brah@xmath3 m et al . found no such fluctuations on small scales from the isophot ( photometer on board the _ iso _ ) data , more sensitive instruments may detect them . thus , these two components may show a non - negligible contribution to the far - ir confusion near the ecliptic plane . this paper is structured as follows . in section [ sec : confused ] , we briefly describe the confusion noise due to sky brightness fluctuations and the source confusion due to extragalactic point sources . we present our source count models including galaxy evolution and the simulated images in section [ sec : models ] . based upon the specifications of each ir mission , we estimate the source confusion noise through fluctuation and photometric analyses in section [ sec : simulation ] . we present the expected results from our simulations in section [ sec : expected ] . our conclusions are summarised in section [ sec : conclusions ] . throughout this work we assume a concordance cosmology of @xmath4 , @xmath5 = 0.3 , and @xmath6 = 0.7 , unless otherwise explicitly stated . the galaxy confusion limit is defined as the threshold of the fluctuations in the background sky brightness below which sources can not be discretely detected in the telescope beam @xmath7 @xmath8 , where @xmath9 is the observation wavelength and @xmath10 is the telescope diameter . thus , the fluctuation noise arises from the same origin as the galaxies that one is aiming to detect . if we assume galaxies are distributed as a power law in flux , @xmath11 , down to some detection limit @xmath12 , @xmath13 where @xmath14 is the number of sources per unit solid angle with flux greater than @xmath11 , @xmath15 is the slope of the integral source counts ( where @xmath16 for a euclidean universe ) and @xmath17 is the number of sources brighter than @xmath12 per unit solid angle . assuming that the counts flatten at some faint flux , @xmath18 , i.e. @xmath19 , then the intensity of the background ( in jy / sr ) up to some maximum flux , @xmath20 , corresponding to these sources is given by , @xmath21 the fluctuations contributed by sources below the detection limit @xmath12 is given by the second moment of the differential source counts @xmath22 , @xmath23 , @xmath24 assuming the power law distribution of sources given in equation ( [ counts ] ) , equation ( [ fluctuations ] ) can be evaluated to give @xmath25.\ ] ] for the euclidean case , the dominant sources contributing to the background intensity are those just below the detection limit @xmath26 @xcite . however , the strong evolution detected in the galaxy population steepens the source counts and produces super euclidean slopes ( @xmath27 1.5 ) and the sources around the detection limit also contribute significantly to the fluctuations in the background intensity . rigorous theoretical definitions of confusion have been presented by scheuer and condon . hogg has highlighted more practical aspects of galaxy confusion noise . an analytical derivation broadly following franceschini et al . is given below . note that the clustering of sources will complicate the confusion noise [ e.g. , as in the case of radio sources , condon ] although here , for clarity , we do not treat this effect . some authors @xcite have investigated the effect of the clustering of sources on the confusion limit . they found that clustering possibly increases the level of the source confusion by 10% for _ akari _ mission @xcite and by 10 @xmath28 15% for _ spitzer _ and _ herschel _ missions @xcite in the far - ir range . assuming that the sources are distributed randomly over the sky described by a power law form @xmath29 and a corresponding differential distribution given by @xmath30 where @xmath31 , the detector response to a source of flux @xmath11 , at a position @xmath32 from the axis of a detector beam of profile ( point spread function ) @xmath33 is given by @xmath34 . hence the mean number of responses with amplitudes between @xmath35 , @xmath36 in a solid angle @xmath37 is given by ; @xmath38 where @xmath39 is the solid angle of the beam in steradians . note that for the power law distribution of sources discussed above , equation ( [ responses1 ] ) can be rewritten as , @xmath40 where @xmath41 is the effective beam size @xcite . taking the second moment of the @xmath42 distribution ( the variance ) gives the fluctuation of the response , @xmath43 ; @xmath44 where @xmath45 is a cut off response introduced to stop the variance from diverging at bright source fluxes . more practically the confusion limit @xmath45 ( corresponding to a cut off flux @xmath46 ) is set to some factor of the confusion noise such that @xmath47 , where the factor @xmath48 ( usually chosen values between 3 and 5 ) limits the largest response included in the evaluation of the confusion noise @xmath23 [ values of @xmath28 5 are assumed in the calculations of franceschini et al . ] . the difference in assuming a cut off in the response as opposed to a cut off in flux is that weak contributions from strong sources are not neglected , as even a strong source far from the axis of the beam may contribute significantly to the point spread function of the beam . assuming for clarity in the calculations , a circular gaussian beam profile , @xmath49 such that @xmath50 where @xmath51 is the full width at half maximum ( fwhm ) of the beam , integrating equation ( [ noise ] ) over the solid angle of the beam gives ; @xmath52 or @xmath53 where @xmath54 , effectively the power in the fluctuations , is given by , @xmath55 thus , the confusion limit can be directly related to the fwhm of the instrumental beam . for the simple power law representation of the distribution of extragalactic sources given previously and the definitions of equations ( [ responses1 ] ) and ( [ responses2 ] ) , the confusion limit is given by , @xmath56 therefore , the confusion noise limit will be a complex function of the beam size @xmath51 , the source counts @xmath57 , the cut off in flux @xmath46 or response @xmath45 and the factor @xmath48 . for the assumed symmetric gaussian beam profile , @xmath58 . the confusion due to faint galaxies will be worse at longer wavelengths and smaller telescope diameters . since the confusion noise is related to the mean number of responses ( the source density ) and the cut off response @xmath59 , a useful , practical benchmark for the confusion limit can be set by limiting the number sources per beam before the beam becomes confused . ideally , the confusion limit would be determined by the flux at which the source density becomes 1 source per beam although more realistically a limit of between 1/20 - 1/50 sources per beam ( 20 - 50 beams per source ) is assumed [ e.g. , hogg ] . note that this difference is due to the contribution from faint sources undetected below a certain limit . many authors @xcite have described the source confusion due to extragalactic point sources and the sky confusion due to galactic cirrus as separate issues since most extragalactic surveys are limited to low background regions . however , in order to cover much larger survey areas , we also naturally have to include the medium - to - high background regions . in addition , since the cirrus structure covers both large and small scales , both of these contributions should be considered together . by using various source distributions and a realistically simulated cirrus background , we address the effects of sky confusion and source confusion simultaneously with a more realistic framework . the input catalogues are prepared using the models of pearson ( hereafter cpp ) . cpp is a far - ir model based on the _ iras _ colours and luminosity function of galaxies [ see also pearson & rowan - robinson ] . the model incorporates a 4 component parameterization of galaxy spectral energy distributions ( sed ) segregated by _ iras _ colours @xcite . a normal galaxy population modelled on the cool 100 @xmath0m/60 @xmath0 m colours identified with infrared cirrus @xcite . a starburst population based upon the warm 100 @xmath0m/60 @xmath0 m colours of _ iras _ galaxies with the archetypical starburst galaxy m82 as a template sed @xcite . an ultraluminous galaxy population @xcite representing the high luminosity tail of the _ iras _ starburst galaxy population and representative of the archetypical ulig arp220 @xcite . an agn ( seyfert 1 @xmath60 seyfert 2 ) population modelled on a 3 - 30 @xmath0 m dust torus component @xcite defined by hot 25 @xmath0m/60 @xmath0 m colours . though recent _ spitzer _ observations showed silicate dust emissions in luminous and low - luminosity agn @xcite , we do not include it . the input seds are shown in figure [ sed ] . these seds have been shown to provide good fits to the seds and colours of sources in the _ iso_-elais survey at 6.7 , 15 , 90 and 170 @xmath0 m @xcite and also recently for the first results from the _ spitzer _ swire legacy survey @xcite . = 8.5 cm to produce the input source distributions , we calculate the total number of sources per unit solid angle at observation wavelength , @xmath61 , down to some flux limit @xmath62 ; @xmath63 where @xmath64 and @xmath65 are evolutionary functions in the luminosity and density of the source population respectively . the integration is made over the luminosity function ( number density of objects as a function of luminosity ) , @xmath66 and the cosmological volume @xmath67 , enclosed inside a limiting redshift @xmath68 defined as the redshift at which a source of luminosity , @xmath69 , falls below the sensitivity , @xmath70 . luminosity functions , @xmath66 , are determined from the all sky _ iras _ pscz catalogue at 60 @xmath0 m by saunders et al . for starburst and normal galaxies following colour criteria akin to those of rowan - robinson @xmath60 crawford . similarly , the _ hot _ agn population is well represented by the 12 @xmath0 m sample of rush et al . using the luminosity function of lawrence et al . . in addition to the above , following cpp , we utilize a log exponential luminosity function , defined at 60 @xmath0 m , to represent the ulig population , [ referred to as the _ burst _ model ] which was originally implemented to address the paradigms of the strong evolution in the galaxy source counts observed with the _ iso _ in the mid to far - ir , at sub - mm wavelengths with the submillimetre common user bolometer array ( scuba ) and the detection of the cirb at @xmath71 m @xcite . this model was found to provide a good fit to both the number counts and redshift distributions of galaxies from sub - mm to near infrared wavelengths as well as the cosmic infrared background . to shift the luminosity function from the wavelength at which the luminosity function is defined , @xmath72 , to the observation wavelength , @xmath73 , the ratio @xmath74 is obtained via the model template spectra . we prepare 3 types of input catalogue and produce many versions of each catalogue to reduce statistical errors . 1 . _ no - evolution model _ - no evolution is assumed for all galaxy components _ luminosity evolution model _ - luminosity evolution is included following pearson @xmath60 rowan - robinson ( 1996 ) . only evolution in the luminosity of the source population is assumed in this scenario . this luminosity evolution follows a parametric form as a function of redshift of @xmath75 , where @xmath76 is independently defined for each galaxy type to produce the best fits to the far - ir counts . burst evolution model _ - luminosity and density evolution is included following the cpp model . the assumed evolution in both luminosity and density incorporates a burst in a specific redshift range and follows a parametric form as a function of redshift of @xmath77 $ ] , where the parameters @xmath76 , @xmath78 , @xmath79 are independently defined for the luminosity and density evolution and galaxy type to produce the best fits to the far - ir counts . details of the evolution included in the models can be found in cpp . figure [ fig_comp_scount ] shows the evolutionary source count models for the _ spitzer _ far - ir galaxy counts at 70 @xmath0 m and 160 @xmath0 m respectively @xcite . the two evolutionary models provide reasonable fits and also cover the spread in uncertainty shown in the observations . therefore , in this work we consider the _ burst evolution _ and _ luminosity evolution _ models as upper and lower limits to the numbers of far - ir sources respectively . we do not attempt to fit the mid - near - infrared _ spitzer _ counts . for specific updated evolutionary models , please see pearson and pearson . = 16 cm based upon the 3 types of input catalogue , we generated simulated images in each band for various space missions . the image size for the distributed source simulation is @xmath28 5.3 square degrees . in order to check the effect of sky confusion noise , we also include high resolution cirrus maps whose mean cirrus brightness ranges from 0.3 mjy sr@xmath80 to 25 mjy sr@xmath80 at 100 @xmath0 m as the background by using the method described in paper i. as we described in section [ sec : confused ] , there are many definitions for the source confusion . we compare these definitions and propose an appropriate definition for source confusion . in table [ tab_inst_para ] , we list the basic instrumental parameters of present and future ir space missions ; the aperture of the telescope , full width at half maximum ( fwhm ) of the beam profile and the pixel size for each detector . @ccrlrlrl + & aperture & & & + & ( meter ) & & & + space mission & & sw & lw & sw & lw & sw & lw + + _ spitzer _ @xmath81 & 0.85 & 70 & 160 & 17 & 35 & 10 & 16 + _ akari _ @xmath82 & 0.69 & 75 & 140 & 23 & 44 & 27 & 44 + _ herschel _ @xmath83 & 3.5 & 70 & 175 @xmath84 & 4.3 & 9.7 & 3.2 & 6.4 + _ spica _ & 3.5 & 70 & 160 & 4.3 & 9.7 & 1.8 & 3.6 + _ @xmath85 _ fwhm of diffraction pattern . + _ @xmath81 _ two mips bands for the _ spitzer _ mission have 3 bands with central wavelengths of 24 @xmath0 m , 70 @xmath0 m and 160 @xmath0 m . we use 70 @xmath0 m band as sw band and 160 @xmath0 m band as lw band in this paper . + _ @xmath82 _ _ akari / fis _ ( far infrared surveyor ) has a wide - s band in sw and a wide - l band in lw . + _ @xmath83 _ pacs has a ` blue ' array in two short wavelength bands ( centered at 70 @xmath0 m and 110 @xmath0 m ) and a ` red ' array at longer wavelengths ( centered at 175@xmath0 m ) . in this paper , we use only the 70 @xmath0 m band of the ` blue ' array . + _ @xmath86 _ since one of our motivations in this paper is to compare the confusion limits , we use only common bands among all far - ir bands in the considered space missions . + _ @xmath84 _ note that though the central wavelength of the ` red ' array in pacs is 175 @xmath0 m , our estimated results are for 160 @xmath0 m . due to this wider beam , the confusion noise is expected to increase by @xmath28 15% at 175 @xmath0 m compared to that at 160 @xmath0 m . first , we estimate the source confusion from the classical definition of source confusion , beams per source ( sources per beam ) . though often cited in the literature as sources per beam , we use the term beams per source for our definition of source confusion in order to compare the effect of source confusion according to the different beam sizes for each mission . we check the source confusion by changing the number of beams per source criterion from 10 to 50 . [ fig_beam_sconf ] shows the source confusion assuming this definition for each mission and each model . hogg showed that 30 beams per source is the minimum photometric criterion where the source counts are steep , and suggested 50 beams per source for the definition of source confusion . rowan - robinson adopted 40 beams per source . in recent papers , dole et al . ( 2003 , 2004b ) suggested that source confusion could be defined by a source density criteria , corresponding to @xmath28 12 beams per source for the _ spitzer _ mission . they have estimated a source confusion limit for the _ spitzer _ mission of 3.2 mjy for sw band and 40 mjy for lw band , respectively . these results are similar to our estimations with the definition of 12 beams per source for the luminosity evolution model in the sw band and for the burst evolution model in the lw band , respectively . these consistences suggest that the source distribution model used for the estimation of the source confusion limit for the case of _ spitzer _ @xcite also predicts that a starburst component , and ulig component dominate the sw and lw bands respectively . in table [ tab_sc_beams ] , we list the source confusion limits estimated by the definition of 12 and 40 beams per source . as seen from fig . [ fig_beam_sconf ] , we found that the source confusion did not increase at a constant rate and that the slope of the source confusion slightly varies according to the source distribution model and the resolution of the mission , especially for burst evolution model . therefore , we conclude that we can not apply the same definition of beams per source in a generic way for different missions . @ccccccccccccc + & & & + & & & + & & & & & & + space mission & 12b @xmath85 & 40b @xmath81 & 12b & 40b & 12b & 40b & 12b & 40b & 12b & 40b & 12b & 40b + + _ spitzer _ & 0.53 & 1.9 & 8.8 & 23 & 2.1 & 5.1 & 17 & 34 & 6.9 & 13 & 42 & 69 + _ akari _ @xmath82 & 1.9 & 6.0 & 13 & 34 & 5.1 & 12 & 23 & 47 & 14 & 26 & 52 & 87 + _ herschel _ & _ spica _ & 0.014 & 0.066 & 0.65 & 2.1 & 0.10 & 0.47 & 2.6 & 6.8 & 0.11 & 0.65 & 4.3 & 20 + _ @xmath85 _ source confusion defined by dole et al . ( 2004b ) : flux corresponding to 12 beams per source . + _ @xmath81 _ source confusion defined by rowan - robinson ( 2001 ) : flux corresponding to 40 beams per source . + _ @xmath82 _ wide - s band for sw and wide - l band for lw . = 8.5 cm = 8.5 cm another criterion for the quantification of source confusion can be defined by the fluctuation from beam to beam due to the distribution of the point sources @xcite . since the beam size is large and the source counts are steep , the usual definition of ` beams per source ' may not be valid in the case of far - ir photometry . for the psf , we assume an ideal circular aperture airy pattern corresponding to the aperture size of the telescopes for each mission except for _ akari _ for which we use the theoretical psfs estimated from the telescope design @xcite . to visualize the iteration procedure , we plot the @xmath87 ratio as a function of @xmath88 for the _ spitzer _ and _ herschel _ & _ spica _ missions for the case of the burst evolution model ( see figure [ fig_sn_sconf ] ) . for the sw band of the _ herschel _ & _ spcia _ missions , the @xmath89 ratio is always greater than 5 , which means that we can not obtain a solution for source confusion , even for @xmath48 = 5 . dole et al . estimated the source confusion with their best estimator for the _ spitzer _ mission . with optimized @xmath48 parameters of 3.8 in sw band and 6.8 in lw band , they obtained source confusion limits of 3.2 mjy and 36 mjy at 70 @xmath0 m and 160 @xmath0 m , respectively . using our source count models and the same @xmath48 parameters , we obtained the source confusion limits of 3.7 mjy and 21 mjy for the luminosity evolution model and 12 mjy and 60 mjy for the burst evolution model at the same wavelengths . as we showed in section [ sec : sc_def_beam ] , though the source density of the source count model used in dole et al . is similar to the luminosity evolution model in the sw band and the burst evolution model in the lw band , we find that the fluctuation due to distributed sources in the burst evolution model is much stronger than that of the model of dole et al . . this amount of fluctuation may degrade the final detection limits . = 8.5 cm in table [ tab_sc_fluc ] , we list the source confusion limits estimated from the definition by fluctuation for the cases of @xmath48 = 3 and @xmath48 = 5 . for @xmath48 = 5 , there are no solutions for the sw band of _ herschel _ & _ spica _ missions . even in the lw band , we could not find a reasonable solution since the estimated source confusion for the two evolutionary models gave identical results . however , when we attempted photometry on the simulated images including the point sources for each source count model , we encountered some limits to the source detection . therefore , we conclude that we can not apply a constant @xmath48 to the estimation of source confusion generically for all cases . @ccccccccccccc + & & & + & & & + & & & & & & + space mission & @xmath48 = 3 & @xmath48 = 5 & @xmath48 = 3 & @xmath48 = 5 & @xmath48 = 3 & @xmath48 = 5 & @xmath48 = 3 & @xmath48 = 5 & @xmath48 = 3 & @xmath48 = 5 & @xmath48 = 3 & @xmath48 = 5 + + _ spitzer _ & 0.065 & 0.28 & 3.8 & 10 & 0.51 & 1.9 & 14 & 33 & 0.51 & 1.9 & 41 & 87 + _ akari _ & 0.36 & 1.4 & 5.6 & 15 & 2.4 & 6.8 & 20 & 43 & 2.4 & 23 & 59 & 115 + _ herschel _ & _ spica _ & @xmath90 @xmath85 & 0.002 & 0.042 & 0.34 & @xmath90 & 0.006 & 0.18 & 1.5 & @xmath90 & @xmath90 & 0.18 & 1.5 + _ @xmath85 _ no solution for this case . in an attempt to define the source confusion for all cases , we implement a composite definition by fluctuation and photometry . as a first step , we attempt photometry on the simulated images created from each source count model . the definition of completeness and reliability for quantifying the source detection efficiency is widely used in photometry . the ` completeness ' is defined as the fraction of detected sources to the original input catalogue sources and the ` reliability ' as the fraction of real sources to all detected sources.for our photometry , we removed the false or spurious detections by comparing the positions of detected sources with those in the input catalogue . the real sources as defined in the reliability , refers to sources of which the measured flux agrees with the input flux to within a 20% error . note that the reliability at a given flux range is not always higher than the completeness . an excess of sources near the detection limit , or more likely an overestimation of the flux of sources at or near the detection limit can be caused by a step effect where the underlying , unresolved sources are entering the psf , affecting the sky - subtraction . we carried out aperture photometry on the simulated images using the sextractor software _ _ v__2.2.2 @xcite . the most influential parameters are the size of background mesh and the threshold for the source detection for the aperture photometry . since the size of background mesh is related to the range of scales of the background fluctuation , as we set a smaller mesh size , we can detect smaller fluctuation . therefore , if we use smaller size of background mesh and thresholds , we obtained many more spurious sources . we set the size of the background mesh to @xmath28 2 times the measuring aperture , and the detection threshold as 4 , which is optimized for better reliability of the detected sources thus reducing the false detection rate . as discussed in section [ sec : sc_def_fluc ] , we could not use a constant value of @xmath48 for the estimation of source confusion . current and future space missions , will detect much fainter sources to higher sensitivity , which means that we will observe high source densities to extremely faint detectable flux levels . therefore , a significant factor contributing to the source detection comes from both the faint sources below the detection limit and the high source density above or around the detection limit . with this assumption , we include the contributed fluctuations from the sources above the detection limit as well . in order to find the limiting flux affecting the source detection , we use a photometric method on the simulated images . we set the limiting flux @xmath88 in equation [ fluctuations ] to be the flux that the completeness reaches @xmath28 80% where the completeness level abruptly falls off . we do not use the reliability criterion in this definition since the reliability can be improved by using optimized photometric methods . we also assume that the sources above this flux level do not contribute significantly to the source confusion . we obtain the final source confusion from a 4@xmath91 fluctuation in order to compare with the threshold used in the photometry . in table [ tab_sc_phot ] , we list the source confusion limits from our best estimators . @crlrlrl + & & & + & & & + space mission & sw & lw & sw & lw & sw & lw + + _ spitzer _ & 0.48 & 9.4 & 2.0 & 23 & 6.5 & 66 + _ akari _ & 2.0 & 14 & 6.3 & 33 & 17 & 87 + _ herschel _ & _ spica _ & 0.0095 & 0.63 & 0.077 & 2.2 & 0.10 & 4.4 + though we could not obtain the results for the sw band of _ herschel _ & _ spica _ missions with the original definition of the fluctuation with constant @xmath48 , we could obtain the results by our new best estimator . we also find that the source confusion limits by our definition are mostly consistent with a completeness of @xmath28 65% for all missions . therefore , we conclude that our definition can explain the behaviour of source confusion well in the far - ir range , regardless of individual space mission characteristics . in our study of sky confusion in paper i , we predicted the confusion limits considering both sky confusion due to cirrus structure and source confusion for each mission from a rather simple ( power law ) source distribution models and plotted the sky confusion levels for an assumed range of average cirrus brightness @xmath92 in table [ tab_sc_cirrus ] ( see details in paper i ) . kiss et al . have also estimated the all - sky cirrus confusion for various ir missions . [ fig_comp_kiss ] shows a comparison of sky confusion between kiss et al . and jeong et al . for the _ akari _ mission at 170 @xmath0 m . since the kiss et al . ( 2005 ) model assumes a linear relationship between the cirrus brightness and sky confusion , it often gives different results in low to medium cirrus brightness . = 7.5 cm @crlrlrl + & + & & & + space mission & sw & lw & sw & lw & sw & lw + + _ spitzer _ & 0.10 & 7.1 & 1.2 & 10 & 11 & 19 + _ akari _ & 0.73 & 9.4 & 5.4 & 18 & 27 & 42 + _ herschel _ & _ spica _ & 0.003 & 0.16 & 0.027 & 0.21 & 0.31 & 0.37 + _ @xmath85 _ mean brightness of cirrus emission since , in general , the cirrus fluctuations are not represented by gaussian noise , we can not directly sum the two noise contributions ( sky confusion and source confusion ) for the estimation of the final confusion noise . however , in a recent paper , h@xmath93raudeau et al . found that the distribution of cirrus fluctuation is near - gaussian , at least at the isophot 90 and 170 @xmath0 m spatial frequencies . thus , we use equation [ eq_tot_no ] to estimate final confusion limits @xmath94 in this paper by summation of the two confusion noise components : @xmath95 where @xmath96 is the sky confusion limit due to cirrus structure and @xmath97 is the source confusion limit . fig . [ fig_conf_all ] shows the final confusion limits due to both sky and source confusion for each mission and each evolutionary model . the sky confusion will be negligible for the _ herschel _ and _ spica _ missions even in regions of high cirrus brightness and the dominant confusion contribution is predicted to be the source confusion in high galactic latitude regions for the purpose of cosmological studies . however , care must be taken when considering the _ spitzer _ and _ akari _ missions in the mean cirrus brightness range @xmath98 @xmath99 20 mjy sr@xmath80 for the lw band . in order to check the effect caused by the combination of both sky and source confusion , we attempt photometry on the simulated images including various cirrus background levels and source count models and compare them with the photometric results considering only source confusion . these results are presented in appendix [ app : phot_results ] . = 8.5 cm = 8.5 cm we summarize the confusion limits including both sky and source confusion for each mission in table [ tab_final_sc ] for the low cirrus background regions dominated by source confusion . the mean brightness used in the estimation of sky confusion is 0.5 mjy sr@xmath80 and 1.0 mjy sr@xmath80 for the sw and lw bands , respectively . the sensitivity for the _ akari _ mission is estimated based upon the recent laboratory experiments , and various detector characteristics and observational environments which may affect the final sensitivity ( jeong et al . 2003 , 2004 ; matsuura et al . 2002 ; shirahata et al . 2004 ) . though the single scan sensitivity in survey mode of the _ akari _ mission is much lower than the confusion limit , the sensitivity at higher ecliptic latitudes where many scans overlap should approach the confusion limit . for the _ herschel _ mission , since the sensitivity is 3 mjy , we expect that most observations in the sw band will be dominated by instrumental noise and those in the lw band will be near the confusion limit . however , the _ spica _ mission will achieve the confusion limit in both bands since it will have a large aperture telescope cooled to very low temperatures @xcite . @cccrlrlrl + & & & & + & & & & + space mission & sw & lw & sw & lw & sw & lw & sw & lw + + _ spitzer _ & 6 & 15 & 0.49 & 12 & 2.0 & 25 & 6.6 & 67 + _ akari _ @xmath85 & 12 ( 200 ) & 12 ( 400 ) & 2.2 & 18 & 6.4 & 34 & 17 & 88 + _ spica _ @xmath81 & 3 & 3 & 0.010 & 0.66 & 0.077 & 2.3 & 0.10 & 4.4 + _ herschel _ @xmath82 & 3 & 3 & 0.010 & 0.73 & 0.077 & 2.6 & 0.10 & 5.1 + _ @xmath85 _ 5@xmath91 sensitivity for slow scan mode ( scan speed : 8 arcsec / sec , reset : 1 sec ) and survey mode ( parenthesized values , 1 scan ) . the sensitivity of the survey mode at high ecliptic latitude regions can be improved due to the high redundancy . + _ @xmath81 _ 5@xmath91 sensitivity without source confusion and sky confusion . in the case of the _ herschel _ & _ spica _ mission , we commonly use the 5@xmath91 sensitivity of _ herschel _ mission @xcite . + _ @xmath82 _ our actual estimations in the lw band are for those in 160 @xmath0 m . we considered the effect of wider beam at 175 @xmath0 m of the lw band . the detection limits as a function of integration time can be an important indicator of the confusion level . based upon recent hardware measurements for the _ akari _ mission and the _ spitzer _ results from dole et al . ( 2004b ) , we compare the detection limits as a function of an integration time in figure [ fig_noise_intg ] . in case of the _ akari _ mission , we also plot the results of different sampling modes for the high background regions . the decreasing component of the noise is the usual behavior of noise with increasing integration time while the flattening component is mainly due to the confusion . since a low cirrus region was assumed in this estimation , the flattening of the noise results from the source confusion . in the sw band , the detection limits are mostly determined by the instrumental noise and the source confusion noise irrespective of the amount of cirrus background except for the very highest cirrus regions , i.e. near the galactic plane . however , in the lw band , we find that the sky confusion affects the noise level severely in the regions brighter than @xmath28 20 mjy sr@xmath80 for the _ akari _ mission and @xmath28 30 mjy sr@xmath80 for the _ spitzer _ mission , respectively ( see also the plot of confusion limits in fig . [ fig_conf_all ] ) . if the different sampling mode for high background regions available for the _ akari _ mission is used , the detection limits are mostly limited by instrumental and photon noises in the sw band unless the exposure time is very long . however , in the lw band , we found that the sky confusion limits compete with the instrumental and photon noises even for short exposure observations . = 6.5 cm = 6.5 cm = 6.5 cm = 6.5 cm based upon two evolutionary ( and a no evolution ) scenarios of source distributions , we have estimated the expected confusion limits for various ir missions in section [ sec : conf_limits ] . the two evolutionary models , the luminosity evolution model and burst evolution model , provide upper and lower limits for the final detection limits . here , we discuss the results expected from the two evolutionary models from our estimated sensitivities . once the realistic detection limits have been determined , we can obtain the expected optimal confusion limited redshift distribution from the models . the number - redshift distribution to our limiting flux @xmath88 , can be extracted directly from our simulated catalogue . in figure [ fig_rsdist ] , we show the expected redshift distributions and expected number of sources per square degree for each mission at the predicted confusion limit . we consider this as the optimal redshift distribution for each mission . since a fairly large number of sources below @xmath100 1.0 are detected in the lw bands for the no evolution model , the redshift distribution for each mission has a significant difference depending on whether there is evolution or not . though we can find the peaks in the sw band of the burst evolution model and the lw band of luminosity evolution model for the _ spitzer _ and _ akari _ missions , we still have difficulty in finding a significant difference between the two evolutionary scenarios . however , we expect that the large 3.5 m aperture missions ( e.g. _ herschel _ and _ spica _ ) will have advantages over the 6090 cm aperture missions ( _ spitzer _ and _ akari _ ) in distinguishing between the evolutionary scenarios due to the larger numbers of sources detected in the redshift range of 13 . since the source confusion should be more severe for stronger evolutionary models , the total number of detected sources in the no evolution model is largest in the lw band except for the _ herschel _ and _ spica _ missions . from figure [ fig_rsdist ] , we show that the confusion limit for the _ spitzer _ and _ akari _ missions , restricts the majority of the sources to redshifts of around or less than unity . for these missions , we could expect of the order of a few tens of sources per redshift bin per square degree out to redshifts of 2 - 3 ( note that the mips 70 @xmath0 m band is more sensitive in theory but suffers from detector array problems in practice ) . however , for the _ herschel _ and _ spica _ missions , we could expect 500 - 1000s of sources per square degree per redshift bin out to redshifts of 2 - 3 . this would allow direct probing of the dusty starburst populations detected at 24 @xmath0 m with _ spitzer _ @xcite and at 850@xmath0 m with _ scuba _ [ e.g. chapman et al . ] resulting in the construction of much more accurate far - ir luminosity functions out to these redshifts with large number statistics . = 8 cm = 8 cm = 8 cm = 8 cm = 8 cm = 8 cm the measurements of the infrared background have been carried out in several wavelength bands . also , there has been rapid progress in resolving a significant fraction of the background with deep galaxy counts at infrared wavelengths . the flux from extragalactic sources below the detection limits create fluctuations in the background . the cosmic far - infrared background ( cfirb ) intensity @xmath101 produced by all sources with the flux below the limiting flux @xmath12 , is obtained from : @xmath102 in addition , the cfirb fluctuations @xmath103 from sources with a random distribution below a given detection limit @xmath12 can be estimated as follows , @xmath104 the detection limits @xmath88 can be found in table [ tab_final_sc ] . we list the expected cfirb intensity and resolved fraction for each mission in table [ tab_int_cfirb ] . dole et al . predicted that the _ spitzer _ mission can resolve 18% of the cfirb at 160 @xmath0 m from their source count model . according to our estimation with our source count model , the resolved cfirb is expected to be 9 17% of the total cfirb . as shown in table [ tab_int_cfirb ] , the _ herschel _ & _ spica _ missions will resolve a much larger fraction of the cfirb , i.e. 90 94% in the sw band and 60 72% in the lw band , compared to other missions . @ccccccccc + & & + & & & & + & intensity & resolved & intensity & resolved & intensity & resolved & intensity & resolved + space mission & ( mjy / sr ) & ( % ) & ( mjy / sr ) & ( % ) & ( mjy / sr ) & ( % ) & ( mjy / sr ) & ( % ) + + _ spitzer _ & 0.063 & 48 & 0.33 & 17 & 0.12 & 52 & 0.69 & 9 + _ akari _ & 0.094 & 32 & 0.31 & 14 & 0.21 & 27 & 0.66 & 7 + _ herschel _ & _ spica _ & 0.013 & 90 & 0.16 & 60 & 0.015 & 94 & 0.21 & 72 + = 8 cm table [ tab_fluc_cfirb ] lists the expected cfirb fluctuations for each mission . we can not find a significant difference in the fluctuation levels of the _ spitzer _ and the _ akari _ missions in the lw band since the detection limits already approach a flux range with a monotonic increase of fluctuation ( see fig . [ fig_fluc_ulig ] for the burst evolution model ) . in the far - ir range , lagache & puget and matsuhara et al . have studied the detection of the cfirb fluctuation with _ iso _ data in marano 1 region and the lockman hole , respectively . for comparison , we list the fluctuation estimated from our model for the _ iso _ mission in table [ tab_fluc_comp ] . we find that the observed fluctuations are mostly located between the results from our two evolution models except for the observations at 90 @xmath0 m . comparing with the model of lagache , dole & puget , our estimated fluctuations from the burst evolution model are in good agreement with those from their model . @crlrl + & & + & & + space mission & sw & lw & sw & lw + + _ spitzer _ & 35 & 1600 & 290 & 12000 + _ akari _ & 130 & 1800 & 1200 & 13000 + _ herschel _ & _ spica _ & 0.4 & 120 & 0.6 & 290 + @cccccc + @xmath9 & @xmath7 & @xmath20 & observations & predicted @xmath85 & predicted ( this work ) @xmath81 + ( @xmath0 m ) & ( arcmin ) & ( mjy ) & ( jy@xmath105/sr ) & ( jy@xmath105/sr ) & ( jy@xmath105/sr ) + + 90 & 0.4 @xmath90 20 & 150 & 13000 @xmath106 3000 @xmath82 & 5300 & 2100 @xmath90 7200 + 170 & 0.6 @xmath90 4 & 100 & 7400 @xmath83 & 12000 & 3800 @xmath90 14000 + 170 & 0.6 @xmath90 20 & 250 & 12000 @xmath106 2000 @xmath82 & 16000 & 5500 @xmath90 18000 + _ @xmath85 _ model from lagache , dole & puget . + _ @xmath81 _ lower limit is estimated from luminosity evolution model and upper limit from burst evolution model . + _ @xmath82 _ observational analysis from matsuhara et al . . + _ @xmath83 _ observational analysis from lagache & puget . the power spectrum of cirrus emission at high galactic latitudes ( @xmath99 80 degree ) also has a fluctuation of about 10@xmath107 jy@xmath105/sr at 0.01 arcmin@xmath80 at 160 @xmath0 m with a power index of -2.9 @xmath106 0.5 . in order to distinguish the cfirb fluctuations from the observed data that include both cfirb and cirrus emission effectively , we require an area larger than 115 , 125 , and 19 arcmin@xmath105 for the _ spitzer _ , _ akari _ , and the _ herschel _ & _ spica _ missions , respectively for the case of luminosity evolution model . if we adopt the burst evolution model , the minimum required area for the resolution of the estimated power spectrum should increase to 463 , 490 , 36 arcmin@xmath105 for the _ spitzer _ , _ akari _ , and the _ herschel _ & _ spica _ missions , respectively . the shaded area in figure [ fig_cfirb_fluc ] covers the fluctuations for both evolutionary models . in this estimation , we do not consider the clustering of extragalactic sources . as sky confusion noise increases in the low galactic latitude regions , detection limits should also degrade , and the cfirb fluctuation becomes larger ( see figure [ fig_cfirb_cirrus ] ) . as can be seen in figure [ fig_cfirb_cirrus ] , the fluctuations show a monotonic increase to medium cirrus regions . since the fluctuation power in low galactic latitude regions should be more than 10@xmath108 jy@xmath105/sr at 0.01 arcmin@xmath80 at 160 @xmath0 m , it is difficult to extract the cfirb fluctuations from the analysis of power spectrum [ note that this cfirb fluctuation is expected to have spatial structure below 1 arcmin@xmath80 ( see figure [ fig_cfirb_fluc ] ) ] . however , we expect to detect cfirb fluctuations in most of the low - to - medium cirrus regions . = 8 cm = 6.0 cm = 8 cm = 6.0 cm in order to probe the confusion limit for infrared observations , we generated source catalogues assuming a concordance ( i.e. flat , dark energy dominated ) cosmological world model ( @xmath109 , @xmath110 , @xmath111 ) for 2 evolutionary scenarios defined as the luminosity evolution model and burst model following pearson and pearson et al . , respectively . we also considered the sky confusion due to the infrared cirrus . though the sky confusion is not a dominant noise source at high galactic latitudes , we should take into account the effects of fluctuations of the sky brightness for large area surveys . based upon the fluctuation analysis and photometry on simulated images , we find a composite estimator that represents the source confusion well . from our analysis of source confusion and source distribution models including galaxy evolution , we have estimated final confusion limits of 2.0 6.6 mjy and 25 67 mjy at 70 @xmath0 m and 160 @xmath0 m for the _ spitzer _ mission , 6.4 17 mjy and 34 88 mjy at 75 @xmath0 m and 140 @xmath0 m for the _ akari _ mission , 0.077 0.10 mjy and 2.3 4.4 mjy at 70 @xmath0 m and 160 @xmath0 m for the _ spica _ mission , and 0.077 0.10 mjy and 2.6 5.1 mjy at 70 @xmath0 m and 175 @xmath0 m for the _ herschel _ mission in low cirrus regions . if the source distribution follows the evolutionary models , the current and planned infrared missions will be mostly limited by source confusion . other components affecting confusion are the fluctuation from the zodiacal light , asteroids , and clustering of sources . as yet the fluctuation of the zodiacal light on small scales has not been observed although we may expect this information from the current or next generation of space missions . based upon the number distribution @xcite and sed @xcite of asteroids , we roughly estimated the effect of source confusion from asteroids . since the slope of the flux - number distribution is shallower than that of euclidean space and the number density at any given flux is much smaller than that of our evolutionary source distributions , we expect that the effect from asteroids near the ecliptic plane is not severe , i.e. , less than 2% and 0.5% in the sw and lw bands , respectively . the fluctuation from the clustering of sources results in a 10 @xmath28 15% increase in the confusion @xcite , and can be comparable to the sky confusion in regions of medium cirrus brightness in the lw band . this effect of clustering on the confusion should be detectable in the sw band or in regions of low cirrus brightness . using our study for sky confusion and source confusion , we have generated all sky maps for final confusion limits . these maps are shown in appendix [ app : conf_ass ] . we have also obtained the optimal ( confusion limited ) redshift distribution from each source count model . the redshift distributions for each mission show significant differences between the no evolution scenario and evolutionary models . however , in order to distinguish between the two evolution scenarios , higher sensitivity / resolution missions ( e.g. , _ herschel _ and _ spica _ ) are required compared to the present , relatively low resolution missions . we have estimated that the cfirb will be resolved to 48 52% and 9 17% at 70 @xmath0 m and 160 @xmath0 m for the _ spitzer _ mission , 27 32% and 7 14% at 75 @xmath0 m and 140 @xmath0 m for the _ akari _ mission , and 90 94% and 60 72% at 70 @xmath0 m and 160 @xmath0 m for the _ herschel _ & _ spica _ missions . we also found that we can detect the cfirb fluctuations in most low - to - medium cirrus regions . we thank an anonymous referee for valuable comments . this work was financially supported in part by the kosef grant r14 - 2002 - 058 - 01000 - 0 . cpp acknowledges support from the japan society for the promotion of science . 7 @xmath2brah@xmath3 m p. , leinert c. , lemke d. , 1997 , a&a , 328 , 702 altieri b. et al . , 1999 , a&a , 343 , l65 aussel h. , cesarsky c.j . , elbaz d. , starck j.l . , 1999 , a&a , 342 , 313 biviano et al . , 2000 , in mezure a. , le fevre o. , le brun v. , eds , asp conf . ser . 200 , clustering at high redshifts , astron . , san fransisco , p. 101 bertin e. , arnouts s. , 1996 , a&as , 117 , 393 chapman s.c . , blain a. , ivison r. , smail i. , 2003 , nat , 422 , 695 condon j.j . , 1974 , apj , 188 , 279 dole h. et al . , 2001 , a&a , 372 , 364 dole h. , lagache g. , puget j .- l . , 2003 , apj , 585 , 617 dole h. et al . , 2004a , apjs , 154 , 87 dole h. et al . , 2004b , apjs , 154 , 93 efstathiou a. , rowan - robinson m. , siebenmorgen r. , 2000a , mnras , 313 , 734 efstathiou a. et al . , 2000b , mnras , 319 , 1169 efstathiou a. , rowan - robinson m. , 2003 , mnras , 343 , 322 elbaz d , 2000 , in lemke d , stickel m.k . , eds , iso surveys of a dusty universe , springer - verlag , berlin , p. 121 fixsen d.j . , dwek e. , 2002 , apj , 578 , 1009 flores h. et al . , 1999 , a&a , 343 , 389 franceschini a. , toffolatti l. , danese l. , de zotti d. , 1989 , apj , 344 , 35 franceschini a. , toffolatti l. , mazzei p. , danese l. , de zotti d. , 1991 , a&as , 89 , 285 gallagher d.b . , irace w.r . , werner m.w . , 2002 , in mather j.c . , 4850 , ir space telescopes and instruments . spie , bellingham , wa , p. 17 gautier t.n . iii , boulanger f. , p@xmath93rault m. , puget j.l . , 1992 , aj , 103 , 1313 gruppioni c. et al . , 1999 , mnras , 305 , 297 harwit m. , 2004 , adv . space res . , 34 , 568 hao l. et al . , 2005 , apj , 625 , l75 helou g. , beichman c.a . , 1990 , proc . of the 29th liege international astrophysical coll . esa publications division , noordwijk , p. 117 h@xmath93raudeau ph . et al . , 2004 , mnras , 354 , 924 herbstmeier u. et al . , 1998 , a&a , 332 , 739 hogg d. , 2001 , aj , 121 , 1207 hughes d. et al . , 1998 , nat , 394 , 241 jeong w .- s . 2003 , pasj , 55 , 717 jeong w .- s . 2004 , adv . space res . , 34 , 573 jeong w .- s . 2005 , mnras , 357 , 535 ( paper i ) kawara k. et al . , 1998 , a&a , 336 , l9 kessler m.f . et al . , 1996 , a&a , 315 , 27 kim s. , lee h.m . , nakagawa t. , hasegawa s. , 2003 , jkas , 36 , 21 kiss cs . , @xmath2brah@xmath3 m p. , klaas u. , juvela m. , lemke d. , 2001 , a&a , 379 , 1161 kiss cs . et al . , 2003 , a&a , 399 , 177 kiss cs . , klaas u. , lemke d. , 2005 , a&a , 430 , 343 lagache g. , haffner l.m . , reynolds r.j . , tufte s.l . , 2000a , a&a , 354 , 247 lagache g. , puget j.l . , 2000b , a&a , 355 , 17 lagache g. , dole h. , puget j.l . , 2003 , mnras , 338 , 555 lagache g. et al . 2004 , apjs , 154 , 112 lawrence a. , walker d. , rowan - robinson m. , leech k.j . , penston m.v . , 1986 , mnras , 219 , 687 low f.j . et al . , 1984 , apj , 278 , l19 matsuhara h. et al . , 2000 , a&a , 361 , 407 matsuura s. et al . , 2002 , in mather j.c . , ed . 4850 , ir space telescopes and instruments . spie , bellingham , wa , p. 902 m@xmath112ller t.g . , lagerros j.s.v . , 1998 , a&a , 338 , 340 murakami h. , 1998 , in bely p.y . , breckinridge j.b . , eds , proc . 3356 , ir space telescopes and instruments . spie , bellingham , wa , p. 471 nakagawa t. , 2001 , in pilbratt g.l . , cernicharo j. , heras a.m. , prusti t. , harris r. , eds , esa sp-460 , the promise of the herschel space observatory , esa publications division , p. 67 nakagawa t. , spica working group , 2004 , adv . space res . , 34 , 645 negrello m. , magliocchetti m. , moscardini l. , de zotti g. , granato g.l . , silva l. , 2004 , mnras , 352 , 493 oliver s.j . et al . , 1997 , mnras , 289 , 471 papovich c. et al . , 2004 , apjs , 154 , 70 pearson c.p . , rowan - robinson m. , 1996 , mnras , 283 , 174 pearson c.p . 2001 , mnras 325 , 1511 ( cpp ) pearson c.p . , matsuhara h. , onaka t. , watarai h. , matsumoto t. , 2001b , mnras , 999 , 1014 pearson c.p . et al . , 2004 , mnras , 347 , 1113 pearson c.p . , 2005 , mnras , 358 , 1417 pearson c.p . , 2006a , mnras , in preparation pearson c.p . et al . , 2006b , mnras , in preparation pilbratt g. l. , 2002 , in mather j.c . , ed . , proc . 4850 , ir space telescopes and instruments . spie , bellingham , wa , p. 586 poglitsch a. , waelkens c. , geis n. , 2002 , in mather j.c . , ed . spie vol . 4850 , ir space telescopes and instruments . spie , bellingham , wa , p. 662 puget j.l . et al . , 1999 , a&a , 345 , 29 rieke g.h . , 2004 , apjs , 154 , 25 rodighiero g. et al . , 2005 , mnras , astro - ph/0511563 rowan - robinson m. , crawford p. 1989 , mnras , 238 , 523 rowan - robinson m. , 1995 , mnras , 272 , 737 rowan - robinson m. , 2001 , apj , 549 , 745 rowan - robinson m. et al . , 2004 , mnras , 351 , 1290 rowan - robinson m. et al . , 2005 , aj , 129 , 1183 rush b. , malkan m. , spinoglio l. , 1993 , apjs , 89 , 1 sanders d.b . , mirabel i.f . , 1996 , ara&a , 34 , 725 saunders w. , rowan - robinson m. , lawrence a. , efstathiou g. , kaiser n. , ellis r.s . , frenk c.s . , 1990 , mnras , 242 , 318 saunders w. et al . , 2000 , mnras , 317 , 55 scheuer p.a . , 1957 , proc . cambridge phil . , 764 scott s.e . et al . , 2000 , mnras , 331 , 817 serjeant s.b.g . et al . , 2000 , mnras , 316 , 768 shibai h. , 2004 , adv . space res . , 34 , 589 shirahata m. et al . , 2004 , in mather j.c . , spie vol . 5487 , optical , infrared , and millimeter space telescopes . spie , bellingham , wa , p. 369 siebenmorgen r. , haas m. , kr@xmath113gel e. , shultz b. , 2005 , a&a , 436 , l5 smail i. , ivison r.j . , blain a.w . , 1997 , apj , 490 , l5 soifer b.t , houck j.r . , neugebauer g. , 1987 , ara&a , 25 , 187 sturm e. et al . , 2005 , apj , 629 , l21 takeuchi t.t . , ishii t.t . , 2004 , apj , 604 , 40 tedesco e.f . , cellino a. , zappal@xmath3 v. , 2005 , aj , 129 , 2869 vaisanen p. , tollestrup e.v . , fazio g.g . , 2001 , mnras , 325 , 1241 werner m.w . et al . , 2004 , apjs , 154 , 1 xu c. et al . , 2001 , apj , 562 , 179 we have carried out photometry on our simulated images under various conditions . the results of the photometry is shown below . [ fig_phot_all_nobgr ] shows the photometric results for each mission considering only source confusion . the results for the no evolution model are clearly segregated from those of the evolution models due to the low source confusion . we also found that the completeness and the reliability fall rapidly since the source confusion is more severe for relatively low resolution observations such as is the case for the lw bands of _ akari _ and _ spitzer_. = 13 cm fig . [ fig_phot_fis_noevol ] shows the photometric results for the _ akari _ mission with the no evolution model and various levels of cirrus background . since the contribution from the sky confusion in the lw band is comparable to the source confusion ( see the top - left panel of fig . [ fig_conf_all ] ) , the completeness and reliability are well segregated as the cirrus background increases . = 8.7 cm fig . [ fig_phot_fis ] shows the photometric results for the _ akari _ mission for the two evolutionary models and various levels of cirrus background . since the contribution from the source confusion in the lw band is stronger than that in no evolution model ( fig . [ fig_phot_fis_noevol ] , we find that the plots of completeness and reliability gather closer compared to those in the no evolution model . = 13 cm we also show the photometric results for the _ spitzer _ mission in the case of the two evolutionary models in fig . [ fig_phot_sirtf_evols ] . although the cirrus background becomes higher , the completeness and reliability show similar values in the case of both the sw and lw bands with a mean brightness @xmath98 @xmath114 10 mjy sr@xmath80 due to the relatively more severe source confusion compared to the case of the no evolution model . in addition , we found that though the _ akari _ and _ spitzer _ missions have similar apertures ( 0.69 m and 0.85 m , respectively ) , the larger pixel size of _ akari _ makes both the sky and source confusion more severe . = 13 cm fig . [ fig_phot_spica ] shows the photometric results for the _ herschel _ and _ spica _ missions from the two evolutionary models . since the source confusion is much more significant in most of the cirrus background regions ( see the lower panels of fig . [ fig_conf_all ] ) , the completeness does not change significantly , irrespective of the amount of cirrus background . however , from the reliability plot we see that the sky confusion still affects the measurement of the source flux . using our results for sky confusion and source confusion , we have generated all sky maps for the final confusion limits . since the final confusion limits in the sw band are dominated by source confusion , we show our results only for the case of the lw band ( see figs [ fig_conf_ass_sirtf ] , [ fig_conf_ass_fis ] , and [ fig_conf_ass_spica ] ) except for all sky survey of the _ akari _ mission . since the instrumental noise of the all sky survey of the _ akari _ mission decreases in proportion to the square root of the number of scans , we show here the final detection limit map including instrumental noise by considering the visibility for the all sky survey ( see figure [ fig_detlim_ass_fis ] ) .

note : this is a book chapter that will appear in _ handbook of optoelectronic device modeling and simulation _ , taylor & francis books , 2017 . editor : joachim piprek . for table of contents , see http://www.nusod.org/piprek/handbook.html

1609.09130

quantum cascade lasers ( qcls ) are high - power , coherent light sources emitting in the mid - infrared ( mid - ir ) and terahertz ( thz ) frequency ranges @xcite . qcls are electronically driven , unipolar devices whose active core consists of tens to hundreds of repetitions of a carefully designed stage . the qcl active core can be considered a superlattice ( sl ) , in which each stage is a multiple - quantum - well ( mqw ) heterostructure , where confined electronic states with specific energy levels are formed because of quantum confinement . the concept of achieving lasing in semiconductor sls was first introduced by kazarinov and suris @xcite in 1971 . the first working qcl was demonstrated by faist _ . _ @xcite two decades later . qcls are typically iii - v material systems grown on gaas or inp substrates . molecular beam epitaxy ( mbe ) @xcite and metal - organic chemical vapor deposition ( mocvd ) @xcite are the techniques that enable precise growth of thin layers of various iii - v alloys . it is also possible to incorporate strain into the structure , as long as the total strain in a stage is balanced . both the precision and the possibility of introducing strain bring great flexibility to the design of the qcl active core , so lasing over a wide range of wave lengths ( from 3 to 190 @xmath0 m ) has been achieved . the growth techniques produce high - quality interfaces , with atomic - level roughness . mid - ir qcls ( wave - length range @xmath1 m ) have widespread military and commercial applications . a practical portable detector requires mid - ir qcls to operate at room - temperature ( rt ) , in continuous - wave ( cw ) mode , and with high ( watt - level ) output power . furthermore , these qcls must also have high wall - plug efficiency ( wpe , the ratio of emitted optical power to the electrical power pumped in ) and long - term reliability under these high - stress operating conditions . as the stress likely stems from excessive nonuniform heating while lasing @xcite , improving device reliability and lifetime goes hand - in - hand with improving the wpe . in qcls , multiple conduction subbands are formed in the active core by means of quantum confinement . qcls are unipolar devices , meaning that lasing is achieved through radiative intersubband transitions ( transitions between two conduction subbands ) instead of radiative interband transitions ( transitions between the conduction and valence bands ) in traditional quantum well ( qw ) semiconductor lasers . as a result , electrons do not combine with holes after the radiative transitions and can be used to emit another photon . in order to reuse electrons , the same mqw heterostructure is repeated many times ( 2570 ) in the qcl active core ( the so - called cascading scheme ) . figure [ fig : bandstructure ] depicts a typical conduction - band diagram of two adjacent stages in a qcl under an electric field . each stage consists of an injector region and an active region . the injector region has several thin wells separated by thin barriers ( @xmath2 ) , so a miniband is formed , with multiple subbands that are close in energy and whose associated wavefunctions have high spatial overlap . typically , the lowest few energy levels in the miniband are referred to as the injector levels . the injector levels collect the electrons that come from the previous stage and inject them into the active region . the active region usually consists of 23 wider wells ( @xmath3 ) separated by thin barriers . consequently , a minigap forms in the active region between the upper lasing level ( 3 ) and the lower lasing level ( 2 ) . another important energy level in the active region is the ground state ( 1 ) . there is a thin barrier ( usually the thinnest among all layers ) between the injecting region and the active region , called the injection barrier . by design , the injector levels are close in energy and strongly coupled to the upper lasing level because of the thin injection barrier . the upper and lower lasing levels have large spatial overlap , which allows a radiative transition between the two levels ; the wave length of the emitted light is determined by the energy spacing between these two levels . the lower lasing level overlaps with the ground state for efficient electron extraction . electron emission of longitudinal optical ( lo ) phonons is the dominant mechanism for electron extraction , so the energy spacing between the lower lasing level and the ground state is designed to be close to the lo phonon energy to facilitate extraction . with careful design , the electron lifetime in the upper lasing level is longer than in the lower lasing level , so population inversion can be achieved . after reaching the ground state , electrons tunnel through the injector into the the upper lasing level of the next stage , and the process is repeated . of course , the lasing mechanism description above is idealized . in reality , the efficiency of the radiative transition between the upper and lower lasing levels is very low @xcite . in recent years , considerable focus has been placed on improving the wpe and output power of qcls for rt cw operation . @xcite showed 8.4 % wpe and 1.3 w output power around 4.6 @xmath0 m in 2008 . shortly thereafter , lyakh _ et al . _ @xcite reported 12.7% wpe and 3 w power at 4.6 @xmath0 m . watt - level power with 6% wpe at 3.76 @xmath0 m and then lower power at 3.39 @xmath0 m and 3.56 @xmath0 m are reported by bandyopadhyay _ _ @xcite demonstrated 21% wpe and 5.1 w output power around 4.9 @xmath0 m in 2011 . much higher wpe and/or output power has been achieved at lower temperatures or at pulsed mode @xcite near 4.8 @xmath0 m . a summary of recent developments can be found in review papers @xcite . while good output powers and wpes have been achieved , long - term reliability of these devices under rt cw operation remains a critical problem @xcite . these devices are prone to catastrophic breakdown owing to reasons that are not entirely understood , but are likely related to thermal stress that stems from prolonged high - power operation @xcite . this kind of thermal stress is worst in short - wave length devices that have high strain and high thermal impedance mismatch between layers @xcite . in addition to improved device lifetime , we seek better cw temperature performance ( higher characteristic temperatures @xmath4 and @xmath5 , defined below ) @xcite . the first aspect is a weaker temperature dependence of the threshold current density . empirically , the threshold current density ( the current density at which the device starts lasing ) has an exponential dependence on the operating temperature @xmath6 : @xmath7 . higher characteristic temperature @xmath4 is preferred in qcl design , as it means less variation in @xmath8 as the temperature changes . another key temperature - dependent parameter is the differential quantum efficiency ( also called the slope efficiency of external quantum efficiency ) , defined as the amount of output optical power @xmath9 per unit increase in the pumping current @xmath10 : @xmath11 . the differential quantum efficiency is directly proportional to the wpe ( @xmath12 , where @xmath13 is the feeding efficiency ) . therefore , the higher the @xmath5 , the closer @xmath14 is to unity , and the higher the wpe . recently , deep - well structures with tapered active regions have demonstrated significant improvements in @xmath4 and @xmath5 with respect to the conventional @xmath15 device @xcite , underscoring that the suppression of leakage plays a key role in temperature performance @xcite . still , the microscopic mechanisms and leakage pathways that contribute to these empirical performance parameters remain unclear . under high - power , rt cw operation , both electron and phonon systems in qcls are far away from equilibrium . in such nonequilibrium conditions , both electronic and thermal transport modeling are important for understanding and improving qcl performance . electron transport in both mid - ir and thz qcls has been successfully simulated via semiclassical ( rate equations @xcite and monte carlo @xcite ) and quantum techniques ( density matrix @xcite , nonequilibrium green s functions ( negf ) @xcite , and lately wigner functions @xcite ) . inp - based mid - ir qcls have been addressed via semiclassical @xcite and quantum transport approaches ( 8.5-@xmath16 @xcite and 4.6-@xmath16 @xcite devices ) . there has been a debate whether electron transport in qcls can be described using semiclassical models , in other words , how much of the current in qcls is coherent . theoretical work by iotti and rossi @xcite show that the steady - state transport in mid - ir qcls is largely incoherent . monte carlo simulation @xcite has also been used to correctly predict transport near threshold . however , short - wavelength structures @xcite have pronounced coherent features , which can not be addressed semiclassically @xcite . negf simulations accurately and comprehensively capture quantum transport in these devices , but are computationally demanding . density - matrix approaches have considerably lower computational overhead than negf , but are still capable of capturing coherent - transport features . a comprehensive review of electron - transport modeling was recently written by jirauschek and kubis @xcite . electronic simulations that ignore radiative transitions are applicable for modeling qcls below or near threshold , where the interaction between electrons and the laser electromagnetic field can be ignored . such simulations are useful for predicting quantities such as threshold current density and @xmath4 . however , in order to accurately model qcls under lasing operations , the effect of the laser field on electronic transport would have to be included . in some cases , the effects of the laser field can be very strong @xcite , especially for high wpe devices , where the field - induced current can be dominant @xcite . when included in simulations , the laser field is typically either modeled as an additional scattering mechanism @xcite or as a time - dependent sinusoidal electric field @xcite . in this work , we ignore the effect of the laser field on electron dynamics . thermal transport in qcls is often described through the heat diffusion equation , which requires accurate thermal conductivity in each region , a challenging task for the active core that contains many interfaces @xcite . it is also very important to include nonequilibrium effects , such as the nonuniform heat - generation rate stemming from the nonuniform temperature distribution @xcite and the feedback that the nonequilibrium phonon population has on electron transport @xcite . in this chapter , we present a multiphysics ( coupled electronic and thermal transport ) and multiscale ( bridging between a single stage and device level ) simulation framework that enables the description of qcl performance under far - from - equilibrium conditions @xcite . we present the electronic ( sec . [ sec : electron ] ) and thermal ( sec . [ sec : thermal ] ) transport models , then bring them together for electrothermal simulation of a real device structure ( sec . [ sec : example ] ) . we strive to cover the basic ideas while pointing readers to the relevant references for derivation and implementation details . depending on the desired accuracy and computational burden , one can model electronic transport in qcls with varying degrees of complexity . the goal is to determine the modal gain ( proportional to the population inversion between the upper and lower lasing levels ) under various pumping conditions ( current or voltage ) and lasing conditions ( pulsed or continuous wave ) . a typical electron transport simulator relies on accurately calculated quasibound electronic states and associated energies in the direction of confinement . electronic wavefunctions and energies are determined by solving the schrdinger equation or the schrdinger equation combined with poisson s equation in highly doped systems . section [ sec : kp ] introduces a @xmath17 schrdinger solver coupled with a poisson solver . more information about other solvers for electronic states can be found in the review paper @xcite and references therein . the simulations of electronic transport fall into two camps depending on how the electron single - particle density matrix is treated . the diagonal elements of the density matrix represent the occupation of the corresponding levels and off - diagonal elements represents the `` coherence '' between two levels . transport is semiclassical or incoherent when the off - diagonal coherences are much smaller than the diagonal terms , and can be approximated as proportional to the diagonal terms times the transition rates between states @xcite . in that case , the explicit calculation of the off - diagonal terms is avoided and only the diagonal elements are tracked , which simplifies the simulation considerably . when the off - diagonal terms are appreciable , transport is partially coherent and has to be addressed using quantum - transport techniques , discussed below . semiclassical approaches assume that electronic transport between stages is largely incoherent `` hopping '' transport . the key quantities are populations of electronic states that are confined in the qcl growth direction , and electrons transfer between them due to scattering events . the scattering rates can be obtained empirically or more rigorously , via fermi s golden rule . common semiclassical approaches are the rate equations and ensemble monte carlo ( emc ) , the latter solving a boltzmann - like transport equation stochastically . in the rate - equation approach @xcite , scattering between relevant states , i.e. , the injector level , the upper and lower lasing levels , and the ground state , is captured through transition rates . the rates include all relevant ( radiative and nonradiative ) scattering mechanisms , and can be either empirical parameters or calculated @xcite . the computational requirements of rate - equation models are low , so they are suitable for fast numerical design and optimization of different structures @xcite . the heterostructure in the qcl active core is a quasi - two - dimensional ( quasi-2d ) system , where electrons are free to move in the @xmath18 plane , while confined cross - plane , in the @xmath19direction ; the confinement results in the formation of quasibound states and discrete energy levels corresponding to the bottoms of 2d energy subbands . the electron wavefunctions in 3d are plane waves in the @xmath18 plane and confined wavefunctions in @xmath20 . electronic transport is captured by a boltzmann - like semiclassical transport equation @xcite , which can be solved via the stochastic emc technique assuming instantaneous hops between states in 3d due to scattering @xcite . the simulation explicitly tracks the energy level and in - plane momentum of each particle in the simulation ensemble ( typically @xmath21 particles ) . tracking in - plane dynamics makes it more detailed than the rate - equation model . the transition rates are generally computed directly from the appropriate interaction hamiltonians , and therefore depend on the energy levels as well as the wavefunction overlaps between different electronic states @xcite . emc allows us to include nonequilibrium effect into transport , which is covered in more detail in sec . [ sec : emc ] . density matrix and negf are the two most widely used techniques to describe quantum transport in qcls . recently , a wigner - function approach was also successfully used to model a superlattice @xcite . in semiclassical approaches , the central quantity of interest is the distribution function @xmath22 , the probability of an electron occupying an eigenstate @xmath23 and having an in - plane kinetic energy @xmath24 . the quantum - mechanical analogue is the single - electron density matrix , @xmath25 , where the diagonal elements @xmath26 are occupations and the off - diagonal elements @xmath25 are the spatial coherences between states @xmath23 and @xmath27 at the in - plane energy @xmath24 . when employing semiclassical methods , off - diagonal matrix elements are assumed to be much smaller than diagonal elements . this approximation may fail in some cases , for example , when two eigenstates with a large spatial overlap have similar energies . this scenario often arises when modeling terahertz qcls @xcite , but can also come up in mid - ir qcls @xcite . in these cases , semiclassical models fail . the density - matrix models that have been employed for qcl modeling can be categorized into two groups . the first includes hybrid methods , where transport is treated semiclassically within a region of the device ( typically a single stage ) while the effects of tunneling between different regions , separated by barriers , is treated quantum mechanically using a density - matrix formalism with phenomenological dephasing times @xcite . the second group involves completely quantum - mechanical methods that rely on microscopically derived markovian master equations that guarantee positivity of the density matrix @xcite . both methods are more computationally expensive than their semiclassical counterparts , because the density matrix contains many more elements than its diagonal semiclassical analogue . the nonequilibrium green s function technique ( see a good overview in @xcite ) relies on the relationships between single - particle time - ordered green s functions and correlation functions @xcite . the correlation function @xmath28 , often referred to as the lesser green s function @xcite , is one of the central quantities and can be understood as a two - time generalization of the density matrix , where @xmath29 refers to the magnitude of in - plane wave vector . the correlation function contains both spatial correlations ( terms with @xmath30 ) as well as temporal correlations between times @xmath31 and @xmath32 ( not included in semiclassical or density - matrix models ) . typically , the potential profile is assumed to be time independent , in which case the correlation function only depends on the time difference @xmath33 . fourier transforming over the time difference into the energy domain gives the energy - resolved correlation function @xmath34 , which is the quantity which is usually solved for numerically @xcite . the main advantages of the negf formalism are that it provides spectral ( energy - resolved ) information and it includes the effects of collisional broadening ( the broadening of energy levels due to scattering ) , which is particularly important when the states are close in energy . these advantages carry a considerable computational cost , so negf calculations are much more time consuming than density - matrix approaches @xcite . here , we focus on presenting semiclassical modeling of electron transport in qcl structures via emc @xcite . the solver consists of two parts , a coupled schrdinger poisson solver and a transport kernel . we solve for the electronic states using the coupled schrdinger poisson solver and feed the energy levels and the wavefunctions of the relevant electronic states to the transport kernel . the transport kernel keeps track of the electron momentum , energy , and distribution among subbands . if the electron density inside the device is high , transport kernel will periodically feed the electron distribution back to the schrdinger poisson solver and update the electronic states . this loop is repeated until the electron distribution converges . by doing so , we solve for both electron transport and the electronic band structure self - consistently . since the active qcl core consists of repeated stages , the wavefunctions in any stage can be obtained from the wavefunctions in any other stage by translation in space and energy . this translational symmetry makes it possible to simulate electron transport in only one generic central stage instead of in the whole qcl core @xcite . typically , electronic states in nonadjacent stages have negligible overlap , which also means that the transition rates between them are negligible . as a result , it is sufficient to limit interstage scattering events to only those between adjacent stages . figure [ fig : emcsche ] shows a schematic of three adjacent stages under an applied field . we simulate electron transport in the central stage @xmath35 , while nearest - neighbor interstage ( @xmath36 ) and intrastage ( @xmath37 ) scattering is allowed . periodic boundary conditions ( pbcs ) are applied in the simulation , i.e. , whenever one electron scatters from the central stage out to the next stage ( process @xmath38 ) , an electron scatters from the previous stage into the central stage ( process @xmath39 ) and vice versa ( process @xmath40 and process @xmath41 ) . pbcs are justified by the cascading scheme . we employ the @xmath17 method to solve the schrdinger equation and couple it to a poisson solver @xcite . the @xmath17 method is an efficient way to solve for the electronic band structure near the band edges , where the transport happens in qcls . the @xmath17 method considers the contribution from the conduction band ( c ) , light - hole band ( lh ) , and the spin - orbit split - off band ( so ) ( the heavy - hole band ( hh ) decouples from the other three at the band edge ) @xcite . the contributions from the lh and so are especially important for narrow - gap materials , such as inp . moreover , in modern qcls , strain - balanced structures have been employed to obtain enhanced performance . in these structures , alternate layers are compressively or tensilely strained while the whole structure is strain free , with carefully designed thickness of each layer . the @xmath17 method allows for convenient inclusion of the effects of strain on the band structure . the implementation details of the @xmath17 solver can be found in @xcite . the @xmath17 solver can only solve for a finite structure rather than an infinite periodic one . as a result , we need to simulate a finite number of stages and add artificially high barriers to the two ends to confine all the states . if a stage is far enough from the boundaries , the calculated band structure in it should be the same as if we were to solve for the whole periodic structure . tests have confirmed that three stages , which we also use in emc , are enough when solving for the electronic states to ensure that the central - stage states are unaffected by the simulation - domain potential boundaries . the states from the central stage are then translated in energy and position to the neighboring stages according to the stage length and the applied electric field . when we need to solve for electron transport and electronic states self - consistently , it is necessary for the solver to be able to automatically pick out the electronic states belonging to the central stage . one intuitive criterion is to calculate the `` center of mass '' for each state ( the expectation value of the cross - plane coordinate , @xmath42 ) and assign those falling in the central stage to that stage . however , in our three - stage scheme , this method may pick up the states that are too close to the boundary . one can either extend the number of stages in the @xmath17 solver to five , so the three stages in the middle are all far from the boundary , or use additional criteria , such as that there be more than 50% possibility of finding an electron in the central stage , based on the probability density distribution , or requiring that the location of the probability - density peak be in the central stage . additional criteria requiring strong confinement of states have been explored in @xcite . the emc kernel tracks the hopping transitions of electrons between subbands and stages until convergence , and outputs the transport information for us to calculate the experimentally relevant quantities such as current and modal gain @xcite . in the transport kernel , both electron electron interactions and electron lo - phonon interactions are considered . other scattering processes such as intervalley scattering , impurity scattering , interface roughness scattering can be considered under different circumstances @xcite . photon emission is not considered , either . because emc tracks individual particles , nonequilibrium electron transport can be automatically captured . ( emc tracks individual simulation particles , each of which might represent thousands of real electrons . ) the most important scattering mechanism in qcls is electron lo - phonon scattering , which facilitates the depopulation of the lower lasing level . as shown in @xcite , phonon confinement has little effect on the electronic transport , therefore , for simplicity , lo phonons are treated as bulklike dispersionless phonons with energy @xmath43 . the transition rate between an initial state @xmath44 with energy @xmath45 and a final state @xmath46 with @xmath47 can be derived from fermi s golden rule as @xmath48 where @xmath49 is the electronic charge while @xmath50 and @xmath51 are static and high - frequency electronic permittivities of the material , respectively . the integrals are over the in - plane kinetic energy @xmath24 of the final state and the cross - plane momentum transfer @xmath52 . @xmath53 is the in - plane momentum transfer . latexmath:[\[\label{eq : oi } is defined as the overlap integral ( oi ) between the initial and final states , where @xmath52 is the cross - plane momentum transfer . the integration is over the angle between initial and final in - plane momenta @xmath55 and @xmath56 ( @xmath57 ) , cross - plane momentum component of the final state ( @xmath58 ) , and the kinetic energy of the final state ( @xmath59 ) . @xmath60 represents the number of lo phonons with momentum @xmath61 . the expression can be further simplified in the equilibrium case , where @xmath60 follows the bose - einstein distribution @xcite . in order to model nonequilibrium phonon effects , we numerically integrate the expression using a phonon number histogram according to both @xmath62 and @xmath52 @xcite . according to the uncertainty principle , position and momentum can not both be determined simultaneously . since our electrons are all confined in the central stage ( @xmath63 is finite ) , the cross - plane momentum is not exactly conserved during the scattering process ( @xmath64 ) @xcite . this analysis does not affect the momentum conservation in the @xmath18 plane , because we assume infinite uncertainty in position there . previously , the cross - plane momentum conservation has been considered through the momentum - conservation approximation ( mca ) @xcite and a broadening of @xmath52 according to the well width @xcite . the mca forbids a phonon emitted between subbands @xmath65 and @xmath66 to be re - absorbed by another transition between @xmath67 and @xmath68 if @xmath69 or @xmath70 , and thus might underestimate the electron - lo interaction strength @xcite . the concept of well width is hard to apply in a mqw structure such as the qcl active core @xcite . we observe that the probability of a phonon with cross - plane momentum @xmath52 being involved in an interaction is proportional to the overlap integral in equation ( [ eq : oi ] ) . figure [ fig : overlap ] depicts the typical overlap integrals for both intersubband ( @xmath71 and @xmath72 ) and intrasubband ( @xmath73 ) transitions . as a result , in each electron lo - phonon scattering event , we randomly select a @xmath52 following the distribution from the overlap integral ( fig . [ fig : overlap ] ) . depending on the mechanism ( absorption or emission ) , a phonon with ( @xmath74 ) is removed / added to the histogram according to the 2d density of states ( dos ) and the effective simulation area @xcite . once the phonons with a certain momentum are depleted , transitions involving such phonons become forbidden . from eq . ( [ eq : oi ] ) versus cross - plane phonon wave vector @xmath52 for several transitions ( intersubband @xmath71 and @xmath72 ; intrasubband @xmath73 ) . figure reproduced from @xcite , y. b. shi and i. knezevic , j. appl . 116 , 123105 ( 2014 ) , with the permission of aip publishing.,title="fig : " ] in order to couple the emc solver to the thermal transport solver , we need to keep a detailed log of heat generation during electron transport . in all the relevant scattering events , electron lo - phonon scattering is the only inelastic mechanism and therefore is the only mechanism that contributes to heat generation . as a result , the total energy emitted and absorbed in the form of lo phonons is recorded during each step of the emc simulation . the nonequilibrium phonons decay into acoustic longitudinal acoustic ( la ) phonons via a three - phonon anharmonic decay process . the formulation and the parameters here follow @xcite . the simulation results of emc including nonequilibrium phonons are shown in section [ sec : example ] . the dominant path of heat transfer in a qcl structure is depicted in fig . [ fig : heatflow ] . the operating electric field of a typical qcl is high , which means that considerable energy is pumped into the electronic system . these energetic , `` hot '' electrons relax their energy largely by emitting lo phonons . lo phonons have high energies but flat dispersions , so their group velocities are low and they are poor carriers of heat . an lo phonon decays into two la phonons via a three - phonon process referred to as anharmonic decay . la phonons have low energy but high group velocity and are the main carriers of heat in semiconductors @xcite . if we neglect the diffusion of optical phonons , the flow of energy in a qcl can be described by the equations @xmath75 where @xmath76 , @xmath77 , and @xmath78 are the lo phonon , acoustic phonon , and electron energy densities , respectively . @xmath79 is the thermal conductivity in the system and @xmath80 is the acoustic - phonon ( lattice ) temperature . the term @xmath81 describes heat diffusion , governed by acoustic phonons . we have also used the fact that the rate of increase in the lo - phonon energy density equals the difference between the rate of its generation by electron lo - phonon scattering and the rate of anharmonic decay into la phonons . in a nonequilibrium steady state , both the lo and la energy densities are constant , so @xmath82 as described in the previous section , the right - hand side of equation ( [ eq : heatdiff ] ) is the heat - generation rate @xmath83 and can be obtained by recording electron lo - phonon scattering events in electronic emc @xcite @xmath84 where @xmath85 is the electron density ( @xmath86 is the sheet density and @xmath87 is the length of a single stage ) while @xmath88 and @xmath89 are the number of simulation particles and the simulation time , respectively . @xmath90 and @xmath91 are the energies of the emitted and absorbed lo phonons , respectively . to solve equation ( [ eq : heatdiff ] ) , we need information on both the thermal conductivity @xmath79 and the heat - generation rate @xmath83 ; they are discussed in subsections [ sec : iiivsl ] and [ sec : device ] , respectively . the qcl active core is a sl : it contains many identical stages , each with several thin layers made from different materials and separated by heterointerfaces . the thermal - conductivity tensor of a sl system reduces to two values : the in - plane thermal conductivity @xmath92 ( in - plane heat flow is assumed isotropic ) and the cross - plane thermal conductivity @xmath93 . experimental results have shown that , in sls , the thermal conductivity is very anisotropic @xcite ( @xmath94 ) while both @xmath95 are smaller than the weighted average of the constituent bulk materials @xcite . both effects can be attributed to the interfaces between adjacent layers @xcite . here , we discuss a semiclassical model for describing the thermal - conductivity tensor of iii - v sl structures . note that the model described here is in principle applicable to sls in other material systems , as long as they have high - quality interface and thermal transport is mostly incoherent @xcite . in particular , we focus on thermal transport in iii - arsenide - based sls , as they are most commonly used in mid - ir - qcl active cores @xcite . under qcl operation conditions of interest ( @xmath96 k , and typically near rt ) , thermal transport is dominated by acoustic phonons and is governed by the boltzmann transport equation ( bte ) . to obtain the thermal conductivity , we solve the phonon bte with full phonon dispersion in the relaxation - time approximation @xcite . to capture both the anisotropic thermal transport and the reduced thermal conductivity in sl systems , we need to observe the twofold influence of the interface . first , it reduces @xmath92 by affecting the acoustic - phonon population close to the interfaces @xcite . second , it introduces an interface thermal boundary resistance ( itbr ) , which is still very difficult to model @xcite . common models are the acoustic mismatch model ( amm ) and the diffuse mismatch model ( dmm ) @xcite ; the former assumes a perfectly smooth interface and only considers the acoustic mismatch between the two materials , while the latter assumes complete randomization of momentum after phonons hit the interface . as most iii - v based qcls are grown by mbe or mocvd , both well - controlled techniques allowing consistent atomic - level precision , neither amm nor dmm captures the essence of a iii - v sl interface . figure [ fig : slsche ] shows a schematic of interface roughness in a lattice - matched sl . the jagged dashed boundaries depict transition layers of characteristic thickness @xmath97 between the two materials . . this effective interface roughness leads to phonon - momentum randomization and to interface resistance in cross - plane transport . figure reproduced from @xcite , s. mei and i. knezevic , j. appl . 118 , 175101 ( 2015 ) , with the permission of aip publishing.,title="fig : " ] we introduce a simple model that calculates a more realistic itbr ( a key part in calculating @xmath93 ) by interpolating between the amm and dmm transmission rates using a specularity parameter @xmath98 . the model has a single fitting parameter : the effective interface rms roughness @xmath97 . since the growth environment is well controlled , using one @xmath97 to describe all the interfaces is justified . we use @xmath97 to calculate a momentum - dependent specularity parameter @xmath99 where @xmath100 is the magnitude of the phonon wave vector and @xmath57 is the angle between @xmath101 and the normal direction to the interface . consistent with the twofold impact of interface roughness , @xmath97 affects the thermal conductivity through two channels . apart from calculating the itbr , an effective interface scattering rate @xmath102 dependent on the same specularity parameter @xmath103 is added to the internal scattering rate to calculate modified @xmath92 ( see detailed derivations in @xcite ) . by adjusting only @xmath97 , typically between 1 - 2 , the calculated thermal conductivity using this model fits a number of different experiments @xcite . thermal transport inside the active core of a qcl is usually treated phenomenologically : @xmath92 is typically assumed to be 75% of the weighted average of the bulk thermal conductivities of the constituent materials , while @xmath93 is treated as a fitting parameter ( constant for all temperatures ) to best fit the experimentally measured temperature profile @xcite . we calculated the thermal - conductivity tensor of a qcl active core @xcite and showed that the typical assumption is not accurate and that the degree of anisotropy is temperature dependent ( fig . [ fig : sltc ] ) . ga@xmath106as and in@xmath107al@xmath108as . the solid curve , dashed curve , and dashed - dotted curve show the calculated in - plane , cross - plane , and averaged bulk thermal conductivity , respectively . @xmath109 in the calculations . the inset shows the ratio between the calculated in - plane and the averaged bulk thermal conductivities . figure reproduced from @xcite , s. mei and i. knezevic , j. appl . 118 , 175101 ( 2015 ) , with the permission of aip publishing.,title="fig : " ] the ratio between @xmath92 and the averaged bulk value ( inset to fig . [ fig : sltc ] ) varies between 45% and 70% over the temperature range of interest . @xmath93 has a weak dependence on temperature , in keeping with the common assumption in simplified models ; the weak temperature sensitivity means that itbr dominates cross - plane thermal transport . these results show that it is important to carefully calculate the thermal - conductivity tensor in qcl thermal simulation and we will use this thermal - conductivity model in the device - level simulation . the active core is not the only region we need to model in a device - level thermal simulation . figure [ fig : qclsche ] shows a typical schematic ( not to scale ) of a qcl device in thermal simulation with a substrate - side mounting configuration @xcite . the active core ( in this case , consisting of 36 stages and 1.6 @xmath0 m thick ) with width @xmath110 is embedded between two cladding layers ( 4.5-@xmath0m - thick gaas ) . the waveguide is supported by a substrate ( gaas ) with thickness @xmath111 . an insulation layer ( si@xmath112n@xmath113 ) with thickness @xmath114 is deposited around the waveguide and then etched away from the top to make the contact . finally , a contact layer ( au ) with thickness @xmath115 and a thin layer of solder ( @xmath116 ) are deposited on top . there is no heat generation in the regions other than the active core . further , these layers are typically thick enough to be treated as bulk materials . bulk - substrate ( gaas or inp ) thermal conductivities are readily obtained for iii - v materials from experiment , as well as from relatively simple theoretical models @xcite ( table [ table : bulktc ] ) . .thermal conductivity as a function of temperature for materials in a qcl structure . [ cols="<,<",options="header " , ] the length of a qcl device is much greater than its width , therefore we can assume the length is infinite and carry out a 2d thermal simulation . the schematic of the simulation domain ( not to scale ) is shown in fig . [ fig : qclsche ] . the boundary of the simulation region is highlighted in green , and certain boundary conditions ( heat sink at fixed temperature , convective boundary condition , or adiabatic boundary condition ) can be applied ( independently ) to each boundary . typically , the bottom boundary of the device is connected to a heat sink while other boundaries have the convective boundary condition at the environment temperature ( single - device case ) or the adiabatic boundary condition ( qcl - array case ) . typical values for the layers thickness are @xmath117 m , @xmath118 m , @xmath119 m , @xmath120 m , @xmath121 m . we use the finite - element method to solve for the temperature distribution . the whole device is divided into different regions according to their materials properties . each stage of the active region is treated as a single unit with the heat - generation rate tabulated in the device table in order to capture the nonuniform behavior among stages . the active core is very small , but is also the only region with heat generation , small thermal conductivity , and spatial nonuniformity . to capture the behavior of the active region while saving computational time , we use a nonuniform mesh in the finite - element solver to emphasize the active core region . figure [ fig : qclmesh ] shows a mesh generated in the simulation . it is known that among all the stages in the active core , the temperature @xmath122 and the electric field @xmath123 ( @xmath65 represents the stage index ) are not constant @xcite , but we have no _ a priori _ knowledge of how they depend on the stage index . however , we know that the charge current continuity equation must hold , and in the steady state @xmath124 ; this implies that the current density @xmath125 must be uniform , as the current flow is essentially in one dimension , along @xmath20 . this insight is key to bridging the single - stage and device - level simulations . from sec . [ sec : emc ] , we can obtain the heat - generation rate @xmath83 inside the active core by running the single - stage emc simulation . each single - stage emc is carried out at a specific electric field @xmath126 and temperature @xmath6 and outputs both the current density @xmath127 and the heat - generation rate @xmath128 . by sweeping @xmath126 and @xmath6 in range of interest , we obtain a table connecting different field and temperature @xmath129 to proper current density and heat - generation rate @xmath130 [ @xmath131 . however , from the discussion above , the input in the thermal simulation needs to be the constant parameter @xmath125 . therefore , we `` flip '' the recorded @xmath132 table to a so - called device table @xmath133 , suitable for coupled simulation @xcite . figure [ fig : thermalflchart ] depicts the flowchart of the device - level electrothermal simulation @xcite . before the simulation , we obtain the device table [ @xmath133 ] , as discussed above . we also have to calculate the thermal conductivities ( @xmath92 and @xmath93 ) of the active region as a function of temperature and tabulate them , based on the model described in sec . [ sec : iiivsl ] . we also need the bulk thermal conductivity of other materials in the device ( cladding layer , substrate , insulation , contact , and solder ) as a function of temperature . these material properties are standard and already well characterized . and temperature profile @xmath122 across the whole device . based on the tabulated information from the single - stage simulation and assumed @xmath135 , we get stage - by - stage profile for the electric field @xmath123 and the heat - generation rate @xmath136 profiles . an accurate temperature - dependent thermal conductivity model , which includes the boundary resistances of layers , and the temperature profile guess are used as input to the heat diffusion equation , which is then iteratively solved ( with updated temperature profile in each step ) until the thermal boundary conditions are satisfied.,title="fig:",width=5 ] each device - level thermal simulation is carried out in a certain environment ( i.e. , for a given set of boundary conditions ) and with a certain current density @xmath125 . at the beginning of the simulation , an initial temperature profile is assigned . with the input from the device table and the thermal conductivity data in each region , we use a finite - element method to iteratively solve the heat diffusion equation until convergence . at the end of the simulation , we obtain a thermal map of the whole device . further , from the temperature @xmath122 in each stage and the injected current density @xmath125 , we obtain the nonuniform electric field distribution @xmath123 . with the electric field in each stage and given the stage thickness , we can accurately calculate the voltage drop across the device and obtain the current voltage characteristic . by changing the mounting configuration ( @xmath110 , @xmath111 , @xmath114 , @xmath115 , @xmath116 ) or the boundary conditions , the temperature profile can be changed . in this section , we present detailed simulation results of a 9-@xmath0 m gaas / al@xmath137ga@xmath138as mid - ir qcl @xcite based on a conventional three - well active region design . the chosen structure has 36 repetitions of the single stage ; each stage has 16 layers . starting from the injection barrier , the layer thicknesses in one stage ( in @xmath139 ) are * 46*/19/*11*/54/*11*/48/*28*/34/*17*/30/**//**//*26*/30 . here , the barriers ( al@xmath137ga@xmath138as ) are in bold while the wells ( gaas ) are in normal font ; the underlined layers are doped to a sheet density of @xmath140 . the results at 77 k are shown here . figure [ fig : pageband ] shows the electronic states of the chosen structure under the design operating field of 48 kv / cm , calculated from the coupled @xmath17poisson solver ( see sec . [ sec : kp ] ) . the active - region states of the central stage are represented in bold red curves ; 1 , 2 , and 3 are the ground state and the lower and upper lasing levels , respectively . injector states are labeled @xmath141 and @xmath142 . other blue states together form the miniband . ( when the electron density in the qcl is high , the electronic bands have to be calculated self - consistently with emc . ) -valley subbands in two adjacent stages of the simulated gaas / algaas - based structure . the bold red curves denote the active region states ( 1 , 2 , and 3 represent the ground state and the lower and upper lasing levels , respectively ) . the blue curves represent injector states , with @xmath141 and @xmath142 denoting the lowest two . figure reproduced from @xcite , y. b. shi and i. knezevic , j. appl . 116 , 123105 ( 2014 ) , with the permission of aip publishing.,title="fig : " ] the current density @xmath125 vs field @xmath126 curve , one of the key qcl characteristics at a given temperature , is intuitive to obtain in emc . after calculating the electronic band structure at a certain field @xmath126 , the wavefunctions , energy levels , and effective masses of each subband and each stage are fed into the emc solver . in the emc simulation , we include all the scattering mechanisms described in sec . [ sec : emc ] . since we employ periodic boundary conditions , the current density @xmath125 can be extracted from how many electrons cross the stage boundaries in a certain amount of time in the steady state . the net flow @xmath144 of electrons is calculated by subtracting the flow between the central stage and the previous stage ( @xmath145 ) from the flow between the central stage and the next stage ( @xmath146 ) in each time step . the current density is then calculated as @xmath147 where @xmath148 is the time interval during which the flow is recorded . @xmath149 is the effective in - plane area of the simulated device . since doping is the main source of electrons , the area is calculated as @xmath150 where @xmath151 is the number of simulated electrons and @xmath86 is the sheet doping density ( in @xmath152 ) in the fabricated device . in the current simulation , @xmath153 and @xmath154 . due to the stochastic nature of emc , we need to average the current density over multiple time steps . in practice , one can record the net cumulative number of electrons per unit area that leave a stage over time and obtain a linear fit to this quantity in the steady state ; the slope yields the steady - state current density . from each individual simulation , we extract the current density at a given electric field and temperature . to obtain the @xmath143 curve at that temperature , we sweep the electric field . to demonstrate the importance of including nonequilibrium phonons effects , we carry out the simulation with thermal phonons alone and with both thermal and excess nonequilibrium phonons . figure [ fig : jf ] is the @xmath143 curve for the simulated structure with ( filled squares ) and without ( empty squares ) nonequilibrium phonons at 77 k. it can be seen that the current density at a given field considerably increases when nonequilibrium phonons are included and the trend holds up to 60 kv / cm . this difference is prominent at low temperatures ( < 200 k ) and goes away at rt @xcite . ) curve of the simulated device with ( filled squares ) and without ( empty squares ) the nonequilibrium phonon effect at 77 k. the inclusion of nonequilibrium phonons considerably increases the current density at a given field up to 60 kv / cm . , title="fig : " ] we calculate the modal gain as @xcite @xmath156 where @xmath157 is the permittivity of free space . some constants are obtained from experiment : waveguide confinement factor @xmath158 , stage length @xmath159 nm , optical - mode refractive index @xmath160 , and full width at half maximum @xmath161 @xcite . the dipole matrix element between the upper and lower lasing levels ( @xmath162 nm ) and the emission wave length ( @xmath163 m ) are also estimated in experiment @xcite , but we calculate these two terms directly . the dipole matrix element is calculated as @xmath164 the value is slightly different at different fields , as the band structure changes . at 48 kv / cm , the calculated matrix element is @xmath165 nm . similarly , the wave length of emitted photon also changes at different fields . one can calculate the value from the energy difference between the upper and lower lasing levels . the calculated wave length at 48 kv / cm is 8.964 @xmath0 m . @xmath166 is the population inversion obtained from emc . again , due to the randomness of emc , the population inversion needs to be averaged over a period of time after the steady - state has been reached . figure [ fig : gm ] shows the modal gain of the device with nonequilibrium ( filled squares ) and thermal ( empty squares ) phonons as a function of ( a ) electric field and ( b ) current density at 77 k. horizontal dotted line indicates the total estimated loss in the device , which is used to help find the threshold current density , @xmath8 . lasing threshold is achieved when the modal gain @xmath155 equals the total loss @xmath167 . we consider two sources of loss , mirror ( @xmath168 ) and waveguide ( @xmath169 ) , so the total loss is @xmath170 . the intercepts between the total loss line and the @xmath155 vs @xmath126 [ fig . [ fig : gm](a ) ] and @xmath155 vs @xmath125 [ fig . [ fig : gm](b ) ] curves give the threshold field @xmath171 and threshold current density @xmath8 , respectively . like the current density , the modal gain of the device is also considerably higher when nonequilibrium phonons are considered , which leads to a lower @xmath171 and a lower @xmath8 . the reason for the increased current density and modal gain with nonequilibrium phonons can be attributed to the enhanced injection selectivity and efficiency @xcite . ) of the simulated device with nonequilibrium ( filled squares ) and thermal ( empty squares ) phonons as a function of ( a ) applied electric field and ( b ) current density @xmath125 at 77 k. horizontal dotted line shows the total estimated loss of the device.,title="fig : " ] the way to obtain the heat - generation rate @xmath83 is similar to how we get the current density @xmath125 . we record the cumulative net energy emission as a function of time and fit a straight line to the region where the simulation has reached a steady state . the slope of the line is used in place of @xmath172 . figure [ fig : hgen ] shows the heat - generation rate as a function of electric field at 77 k. the filled squares and the empty squares depict the situation with and without nonequilibrium phonons , respectively . at 77 k with ( filled squares ) and without ( empty squares ) nonequilibrium phonons.,title="fig : " ] this section serves to illustrate how the described simulation is implemented in practice , and what type of information it provides at the single - stage and device levels . first , the single - stage coupled simulation has to be performed at different temperatures , as in fig . [ fig : singlestagejf](a ) . we note the the calculated @xmath143 curves show a negative - differential - conductance region , which is typical for calculations , but generally not observed in experiment . instead , a flat @xmath143 dependence is typically recorded @xcite . at every temperature and field , we also record the heat - generation rate , as depicted in fig . [ fig : singlestagejf](b ) . second , the thermal model for the whole structure is developed . considering that growth techniques improve over time , structures grown around the same time should have similar properties . since the device studied here was built in 2001 @xcite , we assume the active core should have similar effective rms roughness @xmath97 to other lattice - matched gaas / alas sls built around the same time @xcite . from our previous simulation work on fitting the sl thermal conductivities @xcite , we choose an effective rms roughness @xmath173 in this calculation . figure [ fig : pagetc ] shows the calculated thermal conductivities @xmath92 ( solid line ) and @xmath93 ( dashed line ) , along with the calculated bulk thermal conductivity ( dash - dotted line ) for the substrate gaas . ; solid line ) and cross - plane ( @xmath93 ; dashed line ) thermal conductivities of the active core , along with the bulk thermal conductivity of the gaas substrate ( dash - dotted line ) . the effective rms roughness @xmath97 is taken to be @xmath174,title="fig : " ] the structure we considered operated in pulsed mode at 77 k. depending on the duty cycle , the temperature distribution in the device can differ considerably . figure [ fig : tempprofileschematic ] depicts a typical temperature profile across the device , while fig . [ fig : tempprofiledutycycle ] depicts the profile across the active core alone at duty cycles of 100% ( essentially continuous wave lasing , if the device achieved it ) and 0.01% ( as in experiment @xcite ) . clearly , cw operation would results in dramatic heating of the active region . finally , fig . [ fig : jv ] shows the @xmath175 curve of the entire simulated device at 77 k with a duty cycles of 0.01% , 100% , and as observed in experiment @xcite . and the duty cycle is 100%.,title="fig:",width=302 ] ) .,title="fig : " ] we overviewed electronic and thermal transport simulation of qcls , as well as recent efforts in device - level electrothermal modeling of these structures , which is appropriate for transport below threshold , where the effects of the optical field are negligible . we specifically focused on mid - ir qcls in which electronic transport is largely incoherent and can be captured by the ensemble monte carlo technique . the future of qcl modeling , especially for near - rt cw operation , will likely include improvements on several fronts : 1 ) further development of computationally efficient yet rigorous quantum - transport techniques for electronic transport , to fully account for coherent transport features that are important in short - wavelength mid - ir devices ; 2 ) a better understanding and better numerical models for describing the role of electron electron interaction , impurities , and interface roughness on device characteristics ; 3 ) holistic modeling approaches in which electrons , phonons , and photons are simultaneously and self - consistently captured within a single simulation . the goal of qcl simulation should be nothing less than excellent predictive value of device operation across a range of temperatures and biasing conditions , along with unprecedented insight into the fine details of exciting nonequilibrium physics that underscores the operatiuon of these devices . the authors gratefully acknowledge support by the u.s . department of energy , basic energy sciences , division of materials sciences and engineering , physical behavior of materials program , award no . de - sc0008712 . the work was performed using the resources of the uw - madison center for high throughput computing ( chtc ) . 80ifxundefined [ 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 * * , ( ) link:\doibase 10.1109/5.649646 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.332655 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3385159 [ * * , ( ) ] link:\doibase 10.1109/jstqe.2012.2237387 [ * * , ( ) ] \doibase http://dx.doi:10.1038/nphoton.2009.262 [ * * , ( ) ] \doibase http://dx.doi.org/10.1038/nphoton.2012.143 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.2894569 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3238263 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3496489 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4719110 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3586773 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3484279 [ * * , ( ) ] link:\doibase 10.1364/oe.23.008462 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.2907862 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/j.sse.2010.03.015 [ * * , ( ) ] link:\doibase 10.1049/el.2012.0017 [ * * , ( ) ] @noop * * , ( ) \doibase http://dx.doi.org/10.1063/1.1474613 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.1494457 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.1882768 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.87.146603 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.1644337 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.2711153 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4896400 [ * * , ( ) ] link:\doibase 10.1103/physrevb.67.085315 [ * * , ( ) ] link:\doibase 10.1103/physrevb.80.245316 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.79.165322 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.205311 [ * * , ( ) ] http://stacks.iop.org/1367-2630/12/i=3/a=033045 [ * * , ( ) ] \doibase http://10.1007/s10825 - 016 - 0869 - 3 [ ( ) , http://10.1007/s10825 - 016 - 0869 - 3 ] , link:\doibase 10.3390/photonics3020038 [ * * , ( ) ] link:\doibase 10.1103/physrevb.66.245314 [ * * , ( ) ] link:\doibase 10.1002/pssb.201350322 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) \doibase http://dx.doi.org/10.1063/1.4895123 [ * * , ( ) ] http://stacks.iop.org/0034-4885/68/i=11/a=r02 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3147943 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4863665 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3284523 [ * * , ( ) , http://dx.doi.org/10.1063/1.3284523 ] link:\doibase 10.1109/jstqe.2013.2239613 [ * * , ( ) ] link:\doibase 10.1109/jqe.2008.922327 [ * * , ( ) ] link:\doibase 10.1007/s00340 - 011 - 4744 - 4 [ * * , ( ) ] link:\doibase 10.1007/s10825 - 012 - 0397 - 8 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4935142 [ * * , ( ) ] @noop ( ) , link:\doibase 10.1103/physrevb.73.245320 [ * * , ( ) ] link:\doibase 10.1109/jqe.2002.1005404 [ * * , ( ) ] link:\doibase 10.1109/jqe.2007.907563 [ * * , ( ) ] _ _ , @noop ph.d . thesis , ( ) _ _ , @noop ph.d . thesis , ( ) \doibase http://dx.doi.org/10.1063/1.2136420 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0370-1573(01)00029-1 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.2387485 [ * * , ( ) ] @noop _ _ ( , ) link:\doibase 10.1007/s10825 - 008 - 0175 - 9 [ * * , ( ) ] link:\doibase 10.1103/physrevb.39.7852 [ * * , ( ) ] http://stacks.iop.org/0022-3719/15/i=28/a=021 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.2172225 [ * * , ( ) ] link:\doibase 10.1103/physrevb.50.14179 [ * * , ( ) ] link:\doibase 10.1109/jproc.2006.879794 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4832615 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.98526 [ * * , ( ) ] link:\doibase 10.1115/1.2911404 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.114919 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0921-4526(95)00858-6 [ * * , ( ) ] link:\doibase 10.1103/physrevb.59.8105 [ * * , ( ) ] link:\doibase 10.1115/1.2824212 [ * * , ( ) ] link:\doibase 10.1103/physrevb.57.14958 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.1455693 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.165406 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevb.88.155318 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.61.605 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.2222074 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.1374520 [ * * ( ) , http://dx.doi.org/10.1063/1.1374520 ] \doibase http://dx.doi.org/10.1063/1.2840136 [ * * , ( ) ]

we propose a many - body method for band - structure calculations in strongly correlated electron systems and apply it to nio . the method may be viewed as a translationally invariant version of the cluster method of fujimori and minami . thereby the coulomb interaction within the-shells is treated by exact diagonalization and the-shells then are coupled to a solid by an extension of the cluster perturbation theory ( cpt ) due to senechal et al . the method is computationally no more demanding than a conventional band structure calculation and for nio we find good agreement between the calculated single particle spectral function and the experimentally measured band structure .

cond-mat0409071

band structure calculations based on the single particle picture have enjoyed considerable success in solid state theory . single particle picture here is meant to imply that the ground state is obtained by filling up according to the pauli principle the energy levels calculated for a single electron in an ` effective potential ' . the effective potential thereby is usually constructed within the framework of the local density ( lda ) or local spin density approximation ( lsda ) to density functional theory ( dft)@xcite and despite the well - known fact that the eigenvalues of the kohn - sham equations should not be identified with the single - particle excitation energies of a system , the resulting band structures often give an almost quantitative description of angle - resolved photoemission spectroscopy ( arpes ) . + however , there are also some classes of solids which defy such a description , most notably transition metal compounds with partially filled @xmath0 and @xmath1-shells and strong coulomb interaction between the electrons in these . a frequently cited example is nio , where lsda band structure calculations correctly predict an antiferromagnetic and insulating ground state , but only a small ` slater gap ' of a fraction of an ev@xcite , whereas experimentally nio is an insulator with a bandgap of 4.3 ev@xcite and stays so even above the magnetic ordering temperature . while dft thus gives reasonable answers within its domain of validity - namely ground state properties - the non - correspondence between the kohn - sham eigenvalues and the single - particle excitation energies of the solid obviously has to be taken literal for this compound ( if the band gap is not read off from the lsda band structure but expressed as the difference of ground state energies it is in fact possible to calculate it within the framework of dft , as shown by norman and freeman@xcite ) . it is generally believed that the reason for the discrepancy is the strong coulomb interaction between the electrons in the ni @xmath2-shell , which leads to a substantial energy splitting between @xmath3 configurations with different @xmath4 . this leads to a very pronounced ` pinning ' of the @xmath0-shell occupation number @xmath4 , in the case of nio to the value @xmath5 . final states for photoemission or inverse photoemission then correspond to a single @xmath0-shell being in either a @xmath6 , a @xmath7 or a @xmath8 configuration , in each case with very small admixture of configurations with other @xmath4 . the corresponding ` defect ' then may be thought of propagating through the crystal with definite @xmath9 . this pinning of the electron number in both initial and final states can not be reproduced by a wave function which takes the form of a simple slater - determinant - such as the ground state deduced from the kohn - sham equations . the situation is improved somewhat in the self - interaction corrected version of dft@xcite , which renders a certain fraction of the @xmath0-orbitals completely localized , so that their occupation number in fact does become pinned - for the remaining delocalized @xmath0-orbitals , however , the problem remains . another way to achive the pinning of the @xmath0-shell occupancy is the use of an orbital - dependent potential in the framework of the so called lda+u method@xcite . speaking about gap values the calculations based on the gw - approximation@xcite also need to be mentioned - these give _ ab initio _ gap values which are in good agreement with experiment but do not seem to reproduce the high energy ` satellite ' in the photoemisison spectrum . + in addition to the pinning of the @xmath0-shell occupancy , the multiplet structure of the metal ion poses a problem for single - particle theories as well . it is quite well established that the multiplet structure ( appropriately modified by the crystal field splitting ) of the isolated metal ion persists in the solids . clear evidence for this point of view comes from the fact , that angle integrated valence band photoemission spectra@xcite as well as x - ray absorption spectra@xcite of many transition metal compounds can be reproduced in remarkable detail by configuration interaction calculations solving exactly the problem of a single @xmath0-shell hybridizing with a ` cage ' of ligands . in these calculations it is crucial , however , that the intra - shell coulomb repulsion is treated in full detail . while the cluster method is spectacularly successful for angle - integrated quantities its ` impurity ' character unfortunately makes it impossible to extract the dispersion relations of @xmath10-resolved single particle excitations . + actual dispersion relations in the presence of strong coulomb interaction were first studied by hubbard@xcite , thereby taking an entirely different point of view as compared to the single particle picture on which conventional band structure calculations are based . thereby the @xmath0-shells first are considered as isolated , and their affinity and ionization spectra obtained , thereby treating the coulomb repulsion exactly . in his famous papers hubbard used a much simplified model , where the orbital degeneracy of the @xmath0-level was neglected whence ionization and affinity spectrum of the ` half - filled ' @xmath0-shell collapse to single peak each , with the two peaks separated by the coulomb energy @xmath11 . upon coupling the individual atoms to the solid , the ionization and affinity states of the individual atoms then are systematically broadened to form the two ` hubbard bands ' . the coupling to the solid was achieved originally by the famous hubbard i approximation , but in fact this may be interpreted as a particularly simple form of the cluster perturbation theory ( cpt ) , proposed by senechal _ et al._@xcite , where the individual ` clusters ' consist of just a single @xmath0-shell . this suggests immediately to relax hubbard s simplifications and take into account the full complexity of a transition metal oxide including the orbital degeneracy of the @xmath0-shell , the full coulomb interaction between @xmath0-electrons in these and the sublattice of ligands . this is essentially the purpose of the present manuscript . + an important complication is due to the sublattice of ligands . it has been shown by fujimori and minami@xcite that in discarding altogether the sublattice of ligands ( in the case of nio : the oxygen atoms ) , hubbard actually went one step to far in his simplification of the model . namely fujimori and minami showed that the top of the valence band in nio is composed of states , where a hole is predominantly in an oxygen atom , but somehow ` associated ' with an @xmath4-conserving excitation of a neighboring @xmath0-shell - i.e. a magnon or a @xmath12 exciton . this type of state might be viewed as generalization of a zhang - rice singlet@xcite in the cuo@xmath13 planes of cuprate superconductors . it was then found by zaanen , sawatzky and allen@xcite that there is a crossover between this so - called charge - transfer insulator and a more conventional mott - hubbard insulator as a function of two key parameters , the coulomb repulsion @xmath11 between electrons in the @xmath0-shell and the charge transfer energy @xmath14 , which are defined as @xmath15 strong experimental support for the picture proposed by fujimori and minami is provided by the resonance behavior of the photoemission intensity as seen in photoemission with photon energies near the @xmath16 absoption threshold@xcite . + adopting this point of view and using a simplified ` kondo - heisenberg ' model , in which the charge degrees of freedom on ni where projected out , bala _ et al._@xcite then obtained dispersion relations of quasiparticles in nio which in fact do contain the key feature seen in arpes@xcite : the coexistence of strongly dispersive oxygen bands on one hand and a complex of practically dispersionless ( i.e. : massively renormalized ) bands which form the top of the valence band on the other hand . + in the present theory no reduction of the hamiltonian to a t - j - type model is performed . rather we use the same basic idea as in the treatment of the kondo lattice in ref . @xcite : the system is divided into sub - units which are treated exactly and the hybridization between the sub - units is treated approximately . to do so , we define the ground state for vanishing hybridization as the ` vacuum state ' and treat the charge fluctuations created by the hybridization as ` effective fermions ' , for which an approximate hamiltonian can be derived and solved . it has been shown in ref . @xcite that the hubbard - i approximation for the single - band hubbard model can be re - derived in this fashion if the sub - units are taken to be only a single site - including more complex ` composite particles ' which extend over several unit cells then improves the agreement with numerical results . the generalized zhang - rice singlets discussed above may be viewed as such composite particles . it is shown in appendix i of ref . @xcite that this treatment is in fact equivalent to the original cluster perturbation theory of senechal _ et al._@xcite provided the sub - units into which the system is divided are non - overlapping . this last requirement poses a substantial problem for transition metal oxides , because the rocksalt lattice of nio can not be easily divided into non - overlapping sub - units which are still amenable to exact diagonalization without artificially breaking a symmetry of the lattice ( which would lead to artificial symmetry breaking in the band structure ) . we therefore need to adjust the concept of cluster perturbation theory to this situation - which also is an objective of the present work . the remainder of the paper is organized as follows : in section ii we discuss a simplified 1d model , in section iii we present the general theory , in section iv we apply the theory to the 1d model and compare the obtained single particle spectra with result from exact diagonalization , in section iv we discuss the arpes spectra of nio and section v gives the conclusions . for a start we consider the following minimal version of a 1d charge transfer model ( see figure [ fig3 ] ) : @xmath17 this model describes a 1d chain consisting of strongly correlated metal(@xmath0 ) orbitals and uncorrelated ligand ( @xmath18 ) orbitals . henceforth we choose @xmath19 as the unit of energy and unless otherwise stated set @xmath20 . the relevant filling ( which we will consider henceforth ) of the model is @xmath21 electrons ( or @xmath22 hole ) per unit cell . while our goal ultimately is to study realistic models for compounds such as nio , our motivation for studying this highly oversimplified model is as follows : it is simple enough so that reasonably large clusters ( up to @xmath23 unit cells ) can be treated by exact diagonalization ( ed ) and the obtained exact results for the single - particle spectral function then can serve as a benchmark for the analytical theory . the very simple nature of the model thereby is highly desirable , because it results in a small number of ` bands ' so that the comparison with theory is more significant than e.g. in the case of nio . + schematic representation of the hamiltonian ( [ model ] ) and its parameters . ] the quantity of main interest is the photoemission and inverse photoemission spectrum , defined as @xmath24 where @xmath25 denotes the type of d - orbital , and @xmath26 ( @xmath27 ) denote the @xmath28 eigen state ( eigen energy ) with @xmath4 electrons - thereby @xmath29 corresponds to the ground state . the spectral function for @xmath18-electrons is defined in an analogous way . + to get an idea how to construct an adequate theory it is useful to compare the paramagnetic mean - field solution of the model ( i.e. with @xmath30 ) and the results of an ed calculation , see figure [ fig4 ] . the mean - field solution gives two bands @xmath31 single particle spectral functions @xmath32 and @xmath33 obtained by mean - field solution of the model and by exact diagonalization of a system with @xmath23 unit cells . the wave vector @xmath34 increases from the lowermost to the uppermost panel in steps of @xmath35 . @xmath36-functions have been replaced by lorentzians of width @xmath37 . the part to the left ( right ) of the vertical dashed line shows @xmath32 ( @xmath33 ) . ] the lower ( fully occupied ) one of these has predominant @xmath18-character , the upper ( half - occupied ) one has predominant @xmath0-character . in the spectra obtained by exact diagonalization , the @xmath18-like bands persists with an almost unchanged dispersion . this is not really surprising , because an electron in the respective state moves predominantly on the @xmath18-sublattice and thus will not feel the strong coulomb interaction on the @xmath0-sites very much . the band with predominant @xmath0-character , on the other hand disappears completely in the exact spectra . there is a diffuse band somewhat below @xmath38 and a second one at the energy @xmath39 . clearly , these two resemble the ` hubbard bands ' expected for a strongly correlated system and the respective final states have the character of an empty or doubly occupied @xmath0-orbital . in addition to these hubbard bands , however , there is a third group of peaks at energies @xmath40 which energetically is close to the @xmath18-like band , has a mixed @xmath18-@xmath0 character and which does in fact form the first ionization states of the system . its closeness to the @xmath18- band would seem to suggest that the final states have a hole predominantly on the @xmath18-sites , but it also has a significant admixture of @xmath0-weight and moreover is closer to the fermi energy than the @xmath18-like band . in the charge transfer system under consideration , the @xmath0-like band thus actually splits up into _ three bands _ , rather than the two hubbard bands which one might expect . + despite the highly oversimplified nature of the 1d model , there is actually already a clear analogy to nio : lda bandstructure calculations@xcite produce two well - separated band complexes , the lower one ( i.e. : the one more distant from the fermi energy ) with predominant oxygen character , the upper one with ni character . this is quite similar to the mean - field solution in figure [ fig4 ] . the actual photoemission spectra , however , show first of all a broad structure at binding energies @xmath41 below the top of the valence band . resonant photoemission experiments@xcite show , that the final states observed in this energy range have predominantly @xmath6 character - clearly they should be identified with the @xmath0-like band at @xmath42 in our model . next , at binding energies @xmath43 and @xmath44 arpes experiments@xcite find a group of strongly dispersive bands which closely resemble the oxygen - like bands obtained from an lda band structure calculation - they obviously correspond to the dispersive @xmath18-like ` remnant ' of the free - electron band in our model . finally , the top of the valence band in nio is formed by a group of almost dispersionless bands , whereby the mixed ni @xmath6 and @xmath45 character of these states is established by resonant photoemission@xcite - these states then would correspond to the low intensity band which forms the top of the electron annihilation spectrum . the cluster calculation of fujimori and minami@xcite suggests , that these states should be viewed as hole - like ` compound objects ' where a hole on oxygen is bound to an excited state of @xmath7 - a type of state that might be viewed as a generalization of the zhang - rice singlet in the cuo@xmath13 planes of cuprate superconductors . as already conjectured by fujimori and minami their ` compound nature ' would make these quasiholes very heavy , which immediately would explain the lack of dispersion seen in the arpes spectra@xcite . finally , inverse photoemission shows the presence of an upper hubbard band in nio which is also present in the spectra of the 1d model . + the above comparison shows , that in addition to @xmath18-like holes we will need three types of ` effective particles ' to reproduce the correlated band structure of the model . the two standard hubbard bands , which correspond to @xmath46-like final states are not sufficient here . to get an idea what these states should be , let us start from the ionic limit , @xmath47 . the ground state then corresponds to a constant number of electrons , @xmath4 , in each metal @xmath0-shell and completely filled ligand @xmath18-shells , see figure [ fig10]a . switching on the hybridization integral @xmath19 then will produce charge fluctuations : in a first step , a hole is transferred into a @xmath18 orbital , thus producing a @xmath48 state in @xmath0-orbital number @xmath49 , see figure [ fig10]b . the @xmath48 state has an energy of @xmath50 relative to the original @xmath3 state , and will become our first ` effective particle ' - these ` particles ' form the unoccupied hubbard band . in a second step , the @xmath18-like hole can be transferred into a @xmath0-orbital @xmath51 , thus producing a @xmath52 state , see figure [ fig10]c . the latter has an energy of @xmath53 relative to the @xmath3 state and provides the second type of ` effective particle ' - actually the one that forms the ` satellite ' in the spectral function . finally , the hole in @xmath54 can be transferred back into a neighboring @xmath18-level , thereby leaving the orbital @xmath54 in a state @xmath55 , i.e. an eigenstate of @xmath3 other than the orginal one , see figure [ fig10]d . the ` compound object ' consisting of @xmath55 and a hole in a neighboring @xmath18-orbital will be the third type of effective particle , its energy relative to the original @xmath3 state is @xmath56 , i.e. appropriate to give top of the valence band . these states maight be viewed as generalizations of a zhang - rice singlet , or an extreme case of an ( either spin- or orbital - like ) ` kondo object ' . level repulsion due to hybridization between the @xmath52 and the @xmath57 states will push the latter up to higher energies relative to the @xmath18-like bands - in this way , these states become the first ionization states . in the following , we will try to give the above considerations a more solid theoretical foundation and apply them to the calculation of ` correlated band structures ' . charge fluctuation processes relative to the purely ionic configuration . ] we consider a typical transition metal oxide and restrict our basis to the oxygen @xmath58-orbitals and transition metal @xmath0-orbitals . taking the energy of the @xmath18-level as the zero of energy the single - particle terms in the hamiltonian then take the form @xmath59 where @xmath60 creates a spin-@xmath61 electron in the @xmath0-orbital @xmath62 on metal site @xmath49 and @xmath63 creates an electron in @xmath18-orbital @xmath64 on oxygen site @xmath65 . the @xmath66 combine charge - transfer energy and crystalline electric field . the coulomb interaction between the @xmath0-electrons is @xmath67 where we have suppressed the site label @xmath49 , @xmath68 and @xmath69 denotes the @xmath70-component of the orbital angular momentum . the matrix - elements @xmath71 can be expressed in terms of the @xmath21 racah - parameters , @xmath72 @xmath73 and @xmath74 . + for a start we take all the hybridization matrix elements @xmath75 to be zero . in this limit each oxygen is in a @xmath76 configuration , and each transition metal ion in one of the ground states of @xmath77 with @xmath4 electrons . in general , this ground state is degenerate , and we deal with this by choosing one of these ground states , which we call @xmath78 , for each metal ion - for example in the case of @xmath79 we would choose the direction of the spin @xmath80 of the @xmath0-shell to oscillate between the two sublattices , so as to describe the antiferromagnetic order in the system . we call @xmath78 the corresponding ` reference state ' on transition metal site @xmath49 , it obeys @xmath81 . in the following , we consider the product state of the @xmath78 and the completely filled oxygen-@xmath18 sublattice as the ` vacuum ' of our theory . for the 1d model ( [ model ] ) we introduce two @xmath0-sublattices @xmath72 and @xmath73 and choose latexmath:[\[\begin{aligned } so as to model the antiferromagnetic spin correlations in the system . + next , we assume that the hybridization between the sub - systems is switched on . this will create charge fluctuations in the vacuum : in a first step an electron from a @xmath18-shell will be transferred into one of the @xmath0-orbitals of the neighboring metal ion @xmath49 , a process frequently denoted as @xmath83 . due to the many - body character of the hamiltonian , the resulting state , @xmath84 , in general is not an eigenstate of @xmath85 - rather , we can express it as a superposition of eigenstates : @xmath86 are the eigenstates of the @xmath0-shell with @xmath87 electrons . since we are dealing with a single @xmath0-shell these can be obtained by exact diagonalization , which gives us the eigenenergies @xmath88 of the states @xmath89 as well as the coefficients @xmath90 . if the @xmath89 are chosen to be eigenstates of @xmath91 ( as we will assume in all that follows ) only a small fraction of the @xmath92 is different from zero . we now represent the state where the metal ion @xmath49 is in the state @xmath89 by the presence of a ` book keeping fermion ' , created by @xmath93 , at the site @xmath49 . the spin index @xmath61 thereby gives the difference in @xmath70-spin between the state @xmath89 and the reference state @xmath78 - in principle this is redundant , but we add it so as to make the analogy with a free - particle hamiltonian more obvious . an important technical point is , that in case the @xmath89 are not eigenstates of @xmath91 this labelling is not possible - there may exist states @xmath89 which can be reached by transferring an electron of either spin direction into the @xmath0-shell @xmath49 . + all in all , the charge fluctuation process then can be described by the hamiltonian @xmath94 it is understood that states @xmath89 which have @xmath95 for both directions of @xmath61 should be omitted from this hamiltonian . one can see , that the ` bare ' hopping integrals @xmath96 are multiplied by the coefficients @xmath90 , which have a modulus @xmath97 . the effective fermions @xmath98 thus in general have a weaker hybridization with the @xmath18-orbitals than the original electrons , which will naturally lead to some kind of correlation narrowing of all bands of appreciable @xmath0-character . + in the 1d model ( [ model ] ) there is only one state with @xmath87 electrons , namely the state @xmath99 this has an energy of @xmath100 , whence @xmath101 . for a site @xmath49 on the @xmath72 sublattice we thus identify @xmath102 and the part @xmath103 becomes @xmath104 plus an analogous term which describes the charge fluctuations on the sites of the @xmath73-sublattice . + we proceed to the next type of state which is admixed by the hybridization . first , an electron from the metal - ion with @xmath87 electrons may be transferred back to the oxygen atom with the hole , i.e. @xmath105 . if the metal ion @xmath49 thereby returns to the reference state @xmath78 this process is described by the term ` @xmath106 ' in ( [ h1 ] ) . if the metal ion returns to an @xmath4-electron state @xmath107 other than @xmath78 we should model this by a bosonic excitation @xmath108 . here we will neglect these latter processes - this is presumably the strongest and least justified approximation in the present theory . it implies for example that we are neglecting ( in the case of nio ) the coupling to @xmath12 excitons and the influence of the quantum spin fluctuations ( spin waves ) in the 3d s=1 heisenberg antiferromagnet formed by the ni - moments . + a second type of state can be generated by filling the hole in the oxygen shell @xmath65 with an electron from a @xmath0-shell @xmath51 , that means @xmath109 . this leaves the @xmath0-shell on @xmath54 in an eigenstate @xmath110 of @xmath111 electrons , the net effect is the transfer of an electron between the @xmath0-shells @xmath112 , i.e. precisely the process considered originally by hubbard . we write @xmath113 ionization state by the presence of a hole - like book - keeping fermion , created by @xmath114 . in an analogous way as above we arrive at the following effective hamiltonian to describe the second type of charge fluctuation : @xmath115 since @xmath116 creates a hole - like particle , the presence of terms like @xmath117 is not unusual - these terms describe particle - hole correlations , not particle - particle correlations as in bcs theory . + in the @xmath118 model , the only state with @xmath111 electrons is the empty site @xmath119 , which has the energy @xmath120 , whence @xmath121 . we identify , for a site @xmath49 on the @xmath73 sublattice : @xmath122 and the term @xmath123 reads @xmath124 plus an analogous term describing the @xmath72 sublattice . + as already stated , the second type of charge transfer excitation describes the transfer of an electron between two @xmath0-shells via an intermediate state with a hole in an oxygen @xmath18-shell . if there are direct @xmath0-@xmath0 transfer integrals , this process can also occur in one step . the respective part of the effective hamiltonian can be constructed in an entirely analogous fashion as the parts above - since it is lengthy , we do not write it down in detail . + we proceed to the last type of ` effective particle ' that we will consider . if the @xmath0-shell on atom @xmath49 is in an eigen state of @xmath111 electrons ( which would be described by the @xmath116-particle ) it is possible that an electron from a neighboring @xmath18-shell is transferred to the @xmath0-shell , thereby leaving the @xmath0-shell @xmath49 in an eigenstate @xmath107 of @xmath4 electrons other than the reference state @xmath78 . we will consider the ` compound object ' consisting of the ` excited ' @xmath3 state @xmath125 on site @xmath49 and a hole in a linear combination @xmath126 of @xmath18-orbitals on the nearest neighbors of @xmath49 as a further effective particle , created by @xmath127 . here @xmath25 denotes the symmetry of the linear combination of @xmath18-orbitals , which is such as to hybridize with exactly one of the @xmath60 on site @xmath49 . for an ideal tetrahedral cage of oxygen atoms around each metal ion there is exactly one combination @xmath126 for each @xmath128 . the creation and annihilation of the @xmath70-particles can be described by the term @xmath129 here @xmath130|0\rangle$ ] denotes the kinetic energy of the combination @xmath126 - it can be expressed in terms of the integrals @xmath131 and @xmath132 . also , @xmath133 | 0 \rangle$ ] is the hybridization matrix element between @xmath126 and @xmath60 and can be written in terms of @xmath134 and @xmath135 . + in the 1d model , the only excited state on the @xmath72 sublattice is the state @xmath136 the only possible combination of @xmath18 orbitals which hybridizes with a @xmath0-orbital is @xmath137 , which has @xmath138 . there is therefore just one @xmath139-like particle on the @xmath72-sublattice , which corresponds to site @xmath49 being in the state @xmath136 and having an extra hole in the combination @xmath140 . the total energy is @xmath141 whence @xmath142 and the corresponding hamiltonian reads @xmath143 and , again , a corresponding term for the @xmath73-sublattice . this concludes the types of state which we take into account . we are thus assuming that the hole always is on a nearest neighbor of the @xmath55 state - this means that we truncate the ` kondo cloud ' which is not exactly true . in principle this approximation could be relaxed by including more complex composite particles but here we do not include these . + in order for the mapping between the actual system and the ` book - keeping fermions ' to be a faithful one , we must require that the occupation of any @xmath0-site is either @xmath144 or @xmath22 - otherwise , the state of the respective @xmath0-shell is not unique . this implies that the book - keeping fermions @xmath145 , @xmath146 and @xmath139 have to obey a hard - core constraint , in exactly the same way as e.g. the magnons in spin - wave theory for the heisenberg antiferromagnet ( haf ) . it is well known , however , that linear spin - wave theory for the haf , which neglects this hard - core constraint alltogether and treats the magnons as free bosons , gives an excellent description of the antiferromagnetic phase , even in the case of @xmath147 and @xmath148 , where quantum fluctuations are strong . the reason is , that the density @xmath4 of magnons / site obtained self - consistently from linear spin wave theory is still relatively small , whence the probability that two magnons occupy the same site and thus violate the constraint is @xmath149 . for the same reason we expect that relaxing the constraint in the present case and treating the book - keeping fermions as free fermions will be a very good approximation - its physical content is the assumption , that the probability of charge fluctuations is small , which is certainly justified in a mott- or charge - transfer - insulator . for completeness we note that there is also a certain interference between the @xmath139 particle and the holes on oxygen in the sense that the respective creation and annihilation operators do not exactly anticommute . again , we neglect this , with the justification again being the very low density of the @xmath139 and @xmath150 particles . + adding the various terms in the hamiltonian and the direct @xmath151 hopping @xmath152 we obtain a hamiltonian which describes the lowest order charge fluctuation processes while still being readily solvable by fourier and bogoliubov transform with the result : @xmath153 where @xmath154 is a band index . quantities of physical interest now can be readily calculated . let us first discuss the electron count . we assume that the reference states for the @xmath0-shell have @xmath4 electrons each . then , the total number of electrons / unit cell is @xmath155 where @xmath156(@xmath157 ) denote the number of metal ( oxygen ) atoms in the unit cell . on the other hand we have @xmath158 here the sums over @xmath65 and @xmath49 run over the @xmath18 and @xmath0-shells in one unit cell and the equation follows readily from the electron / hole - like character of the various effective fermions . this can be rewritten as @xmath159 where @xmath160 denotes the total number of ionization states in the unit cell which can be reached by extracting a spin @xmath61 electron from one of the @xmath78 , and and @xmath161 the total number of @xmath139-particles in the unit cell which couple to one of these ionization states . if we assume that the band structure is spin independent ( which is the case for nio ) these numbers must be independent of @xmath61 and we obtain the following requirement for the electron number : @xmath162 the number of occupied bands in the system thus is @xmath163 . since the total number of bands produced by our formalism is @xmath164 , we find that the chemical potential falls exactly into the gap between the @xmath165 bands which correspond ( in the limit of vanishing hybridization ) to the oxygen @xmath58 states , the ionization states of the @xmath0-shell and the @xmath45-type states on one hand and the @xmath166 bands , which correspond to the affinity states of the @xmath0-shells on the other hand . obviously , this is the physically correct position of the chemical potential . + the quantity of main interest to us , the single - particle spectral function @xmath167 can be obtained from the eigenvectors of the hamilton matrix once the resolution of the @xmath0-electron creation / annihilation operator is known . were it not for the presence of the @xmath139-like particles , we could write @xmath168 which is easily verified by taking matrix - elements between the right an left - hand side . however , the presence of the @xmath139-like ` particle ' complicates this . due to their ` compound nature ' the processes by which the electron annihilation operator couples to a @xmath139-particle are rather complicated . for example , one might envisage a process in which an electron in a @xmath48 configuration on site @xmath49 is annihilated , leaving the @xmath0-shell in a @xmath55 state ( i.e. an eigenstate of @xmath3 other than the reference state @xmath78 on site @xmath49 ) . then , if simultaneously a hole happens to be present in a @xmath18-orbital next to site @xmath49 , this process would create a @xmath139-like particle on site @xmath49 , leading to an operator product of the type @xmath169 to describe this process . similarly , if a @xmath55 state is somehow created on a site @xmath49 ( this is not possible in the framework of the hamiltonian which we wrote down above - it would necessitate terms including the bosonic excitations @xmath108 discussed above ) and a hole is created in a @xmath18-orbital next to this site , this would result in the creation of a @xmath139-like particle on site @xmath49 . this process could be described by a product of the type @xmath170 . due to the fact that these processes all involve products of three operators one might expect that they lead predominantly to an incoherent continuum in the spectral function . all in all , we may thus expect that this type of process will not contribute substantially to the photoemission intensity for the dominant peaks . clearly , the problem in calculating the spectral weight is a drawback of the theory - it should be noted , however , that there is a very clear physical reason for this problem , namely the ` compound nature ' of the @xmath171 particles and this should be reflected in any theoretical description . ground state energy and @xmath0-occupation as a function of @xmath11 for @xmath172 , @xmath173 as obtained by exact diagonalization of a system with @xmath23 unit cells and from the present theory . the trivial contribution of @xmath174 has been subtracted off from the ground state energy . ] our theory involves a number of strong approximations which need to be checked in some way . here we present a comparison of results obtained for the 1d model ( [ model ] ) and exact diagonalization of finite clusters . for the 1d model systems with @xmath23 unit cells easily can be solved exactly on a computer and we use these results as a benchmark to check our theory . the simplicity of the model actually makes the comparison more significant than e.g. in the case of a realistic model for nio , because we expect only a small number of ` bands ' whence any disagreement will be more obvious . + before we discuss results for physical quantities let us address one of the key approximations of our theory , namely the neglect of the hard - core constraint which in principle should be obeyed by the book - keeping fermions . solving the 1d model with @xmath172 , @xmath175 gives the gs expectation values @xmath176 , @xmath177 and @xmath178 . this implies that the probability for a violation of the constraint on any given @xmath0-site is @xmath179 , i.e. entirely negligible . simply relaxing the constraint thus is probably an excellent approximation . + next , figure [ fig9 ] compares the total energy / site and the @xmath0-occupancy , which is a measure for the charge - transfer form @xmath180 , as obtained by exact diagonalization and from the theory . obviously , there is good agreement . single particle spectral functions @xmath32 and @xmath33 obtained by the present theory and by exact diagonalization of a system with @xmath23 unit cells . the wave vector @xmath34 increases from the lowermost to the uppermost panel in steps of @xmath35 , to that end the figure combines spectra obtained with periodic and antiperiodic boundary conditions . the part to the left ( right ) of the vertical dashed line shows @xmath32 ( @xmath33 ) . ] figure [ fig5 ] shows a comparison between the single particle spectral function obtained from the theory and by exact diagonalization of a system with @xmath23 unit cells . to obtain a denser mesh of @xmath34-points , spectra for a system with periodic and antiperiodic boundary conditions have been used for the exact diagonalization part , that means @xmath181 and @xmath182 have been calculated with pbc , the ones for @xmath183 and @xmath184 have been obtained with abc . although there is no rigorous proof for this , experience shows that combining spectra with pbc and abc gives quite ` smooth ' dispersion relations - as can also be seen in the present case . in order to suppress the luttinger - liquid behaviour expected for @xmath118 systems , a staggered magnetic field of @xmath185 was applied . the agreement between theory and exact diagonalization then is obviously quite good . the dispersion and spectral character of the main ` bands ' in the numerical spectra is reproduced quite well . the main difference concerns the very strong damping of the lower hubbard band at @xmath186 , which actually forms broad continuum rather than a well - defined band in the numerical spectra . moreover , the upper hubbard band at @xmath187 has some ` fine structure ' in the numerical spectra , which is not reproduced by the theory . on the other hand , our theory does not include any damping mechanism such as the coupling to spin excitations , so one can not expect it to reproduce such details . another slight discrepancy concerns the bandwidth of the oxygen band at @xmath188 , which is somewhat underestimated by theory . apart from that and a few low - intensity peaks in the numerical spectra , however , there is a rather obvious one - to - one correspondence between the bands in the theoretical spectra and the exact ones . same as figure ( [ fig5 ] ) but with different parameter values . ] next , we consider the spectra for a nonvanishing @xmath18-@xmath18 hopping , @xmath189 , see figure [ fig6 ] . again , there is good agreement between theory and numerics , with the main discrepancy being again the damping of the satellite and the fine structure of the upper hubbard band . still , there is a clear one - to - one correspondence between theory and exact spectra . an interesting check is provided by inverting the sign of @xmath190 . one might expect at first sight that the only effect is to invert the dispersion of the @xmath18-like band . inspection of the hamiltonian ( [ h3_simp ] ) shows , however , that inverting the sign of @xmath190 also affects the energy of the @xmath139-particle , and hence should lead to a shift of the corresponding band . the actual spectra in figure [ fig7 ] then show , that this is indeed the case in the numerical spectra . the @xmath70-like band is shifted to higher energies by very nearly the amount of @xmath191 expected from theory , so that the lowest hole - addition states now belong to the @xmath18-like band . the fact that the inversion of the sign of @xmath190 has precisely the effect predicted by theory is a strong indication , that this is indeed the correct interpretation of the low energy peaks in the spectra . same as figure ( [ fig5 ] ) but with different parameter values . ] summarizing the results of the preceeding section we may say that the theory reproduces the numerical spectra and the trends under a change of parameters remarkably well , an indication that despite its simplicity the theory really captures the essential physics of the two - band model . this is encouraging to apply it to a real material , nio . in applying the above procedure to nio we first performed a standard lda band structure calculation in the framework of the lmto - method@xcite for nio ( thereby assuming a paramagnetic ground state ) and obtained the lcao parameters by a fit . for simplicity no overlap integrals were taken into account . lda band structure for paramagnetic nio and lcao fit . the fermi energy is taken to be zero . ] a comparison between the lda band structure and the lcao fit is shown in figure [ fig1 ] ( the lda result is essentially identical to that of ref.@xcite ) , the hybridization integrals and site energies obtained by the fit are given in table [ tab1 ] . we have also obtained lcao parameters for an antiferromagnetic lsda band structure , and those parameters which can be compared ( such as the hybridization integrals ) do not differ significantly . all in all this procedure gives quite reliable estimates for the values of the various hopping integrals . the lda bandstructure broadly can be divided into two complexes of bands : the lower one at energies between @xmath192 and @xmath193 has almost pure oxygen @xmath18-character . in other words , a hole in this bands would move almost exclusively on the oxygen sublattice and have only a very small probability to be on a ni ion . one may thus expect , that these states persist essentially unchanged in the the correlated ground state . next , the complex between @xmath193 and @xmath194 has almost exclusively ni @xmath2 character . the lda band structure thus would seem to suggest , that there are states , where a hole is moving essentially from one ni site to another , which have a less negative binding energy - i.e. which are closer to the fermi energy - than the states where the hole is moving in the oxygen sublattice . clearly , in view of the value of the charge transfer energy @xmath195 , which is consistently suggested by a variety of methods@xcite , this is a quite wrong picture of the electronic structure . .hybridization integrals and site - energies ( in @xmath196 ) obtained by a lcao fit to the paramagnetic lda band structure of nio . [ cols="^,^,^,^,<",options="header " , ] next , we consider the racah parameters @xmath73 and @xmath74 . these differ only slightly from their values for free ions and we took the values from fujimori and minami@xcite @xmath197 , @xmath198 for @xmath7 and @xmath199 , @xmath200 for @xmath8 . in general , these parameters are screened by covalency between @xmath0-orbitals and ligands@xcite but for simplicity we keep the ` bare ' values . + this leaves us with two parameters , which require a special treatment , namely the racah parameter @xmath72 , which is subject to substantial solid - state screening , and the difference of site energies between the ni @xmath2-level and the oxygen @xmath58-level . the racah parameter @xmath72 is related to the coulomb energy @xmath11 , which can be obtained from ` pure @xmath0-quantities ' according to @xmath201 . here we used the values @xmath202 and @xmath203 . similar values for @xmath11 have been obatined by fujimori and minami@xcite from a cluster fit of the valence band photoemission spectrum , by van elp _ et al._@xcite from a cluster fit to the x - ray absorption spectrum and by norman and freeman@xcite from density functional calculations . the value of @xmath14 is somewhat small compared to others , which are around @xmath204 . + single particle spectral densities ( see ( [ specdef ] ) for a definition ) for antiferromagentic nio obtained by the present theory . the momenta are along the @xmath205 direction , the top of the valence band is the zero of energy . @xmath36-functions are replaced by lorentzians of width @xmath206 , the @xmath0-like spectral density is multiplied by a factor of @xmath207 . ] then , the problem of a single @xmath0-shell was solved by exact diagonalization in the @xmath208 , @xmath209 and @xmath210-electron subspaces . the maximum dimension of the hilbert space was 120 for @xmath211 . a nonvanishing cef parameter @xmath212 was applied in order to stabilize the correct @xmath213 @xmath214 ground state for @xmath7 in @xmath215-symmetry . to account for the antiferromagnetic nature of the gs of nio , we chose the reference state @xmath78 to be the @xmath216 member of the @xmath214 multiplet on the ni - sites of one sublattice , and the @xmath217 member on the other one . since we neglect spin - orbit coupling the direction of the spin quantization axis is arbitrary and has no influence on the spectral function . the kinetic energies of the @xmath218 and @xmath219-like combinations of @xmath18-orbitals , which enter the energy of the @xmath70-like particles are @xmath220 and @xmath221 , the respective hybridization integrals are @xmath222 and @xmath223 . all in all , the rank of the effective hamilton matrix to be diagonalized was @xmath224 i.e. quite moderate . to improve the agreement with experiment , the following minor adjustments of parameters were made : the @xmath151 hybridization integrals were reduced by a factor of @xmath225 , and the @xmath226 hybridization integrals were increased by a factor of @xmath227 . + the full single particle spectral functions obtained along the @xmath205 direction for antiferromagnetic nio then is shown in figure [ fig11 ] . comparison between the experimental peak dispersions determined by arpes in non - normal emission ( taken from figure 12 of ref . @xcite ) and the position of ` significant peaks ' in the theroretical spectra . the labels on the ` bands ' indicate a possible correspondence between experiment and theory . ] it differs quite significantly from the what one would expect on the basis of the lda band structure ( see figure([fig1 ] ) ) but instead shows the same overall structure as in the 1d model , compare figure [ fig4 ] . as was the case for the 1d model one can broadly speaking distinguish four complexes of bands . at binding energies @xmath228 , there is a broad continuum of bands with strong @xmath0-weight . analysis of the wave functions shows , that the respective states have ( mainly ) @xmath146 ( i.e. : @xmath6 ) character , with some admixture of @xmath139 ( i.e. : @xmath45 ) and ( less ) admixture of @xmath229 character . clearly these bands should be identified with the ` satellite ' in the experimental nio spectra . by analogy with the 1d model we may expect that these high energy states undergo substantial broadening as is indeed seen in experiment . in figure [ fig11 ] the satellite by and large disperses upwards as one goes away from @xmath230 - shen _ at al._@xcite interpreted their data as showing a downward dispersion of the satellite . on the other hand this feature is rather broad and composed of many ` subpeaks ' so that it may be difficult to make really conclusive statements about the dispersion of the spectral weight without a full calculation of the spectral weight , including the ` radiation characteristics ' of the individual @xmath0-orbitals , final states effects etc . + next , there is a group of strongly dispersive bands of predominant @xmath229 character , which closely resembles the lower complex of bands in the lda calculation , see figure [ fig1 ] . in view of their almost pure oxygen character it is no surprise that these bands are hardly influenced by whatever happens on the ni sites . next comes a group of practically dispersionless bands which form the top of the valence band . they have mainly @xmath139 ( i.e. : @xmath45 ) character with some admixture of @xmath146 ( i.e. : @xmath6 ) . due to their strong @xmath139-character these bands probably are influenced most strongly by our approximation to omit any terms involving @xmath139-operators in the spectral weight operator ( [ resolution ] ) . we may expect that taking the prcocesses discussed there will probably enhance the weight of these states and also add some more @xmath18-like weight to these peaks . + the topmost peak is rather intense and actually composed of several ` subpeaks ' - it is in fact the only feature in this energy range which shows significant dispersion . below this broad peak , there are several bands with lower intensity and practically no dispersion - all of this exactly as seen in the arpes experiment by shen _ et al._@xcite . figure [ fig12 ] shows a more detailed comparison of the dispersion of ` significant peaks ' in the photoemission part of the theoretical spectra with the experimental peak dispersions as reported by shen _ et al._. it can be seen , that the agreement is quite good . along both @xmath205 and @xmath231 the main discrepancy is the position of bands @xmath74 and @xmath232 ( or @xmath233 along @xmath231 ) which are somewhat higher in energy in the theory - still , the discrepancy is only a fraction of an @xmath196 . in view of the fact that we have used the simplemost set of parameters this is quite good agreement . the band portion @xmath234 which is unusual due to its downward curvature has actually been observed by shen _ at al . _ in normal emission ( see figure 6 of ref . the part @xmath235 seems to correspond to the experimental band @xmath236 itself - it has rather low spectral weight for momenta close to @xmath230 . + fine structure of the broad peak which forms the top of the valence band structure in antiferromagnetic nio . the symbols give the positions of peaks with appreciable weight . ] finally , figure [ fig8 ] gives the dispersion of the ` sub - bands ' of the broad structure @xmath72 at the valence band edge . this fine structure has not yet been resolved experimentally as yet - however , shen _ et al . _ found evidence for at least three ` subpeaks ' and also for a quite substantial dispersion , although this made itself felt only as a dispersion of the line shape of the broad peak . looking at figures 9 and 10 of ref . @xcite it would appear that along @xmath205 there is an overall ` upward ' dispersion of the topmost peak @xmath72 as one moves from @xmath237 with two local maxima of the upper edge of @xmath72 just after @xmath230 und just before @xmath238 with the whole band complex being most narrow approximately halfway between @xmath230 and @xmath238 . it can be seen already from figure [ fig12 ] that this dispersion of the peak - shape of @xmath72 is reproduced quite well by theory . similarly , along @xmath231 the brod band complex seems to have its minimum width halfway between @xmath230 and @xmath238 . at least these qualitative results are quite consistent with the dispersion in figure [ fig8 ] . clearly a more detailed study of the fine structure of feature @xmath72 would provide an interesting check of the present and other theories for the band structure of nio . another stringent check for theory would be to unravel the orbital character of the individual flat bands such as @xmath74 and @xmath232 by studying their intensity as a function of photon polarization and energy . + finally , we mention the upper hubbard band , with the corresponding final states having predominantly @xmath145-character . the insulating gap has a magnitude of @xmath239 , which is consistent with experiment@xcite . figure [ fig2 ] shows the angle integrated ( i.e. : @xmath9-integrated ) photoemission and inverse photoemission spectrum . by and large there is reasonable agreement with experiment . the fact that theory puts the dispersionless bands @xmath74 and @xmath232 too close to the top of the valence bands leads to a too weak shoulder on the negative binding energy side of the ` main peak ' at the top of the valence bands . momentum integrated spectral weight for antiferromagnetic nio . to simulate a photoemisison spectrum the lorentzian broadening has been taken energy dependent according to @xmath240 . ] in summary , we have presented a theory for the single particle - excitations of charge transfer insulators . the basic idea is to interpret the charge fluctuations out of the purely ionic configuration as ` effective fermions ' and derive and solve an effective hamiltonian for these . this is the same physical idea which is underlying both the hubbard i approximation and the cluster perturbation theory and , as demonstrated above , when applied to a realistic model of a charge - transfer insulator these methods , which so far have been restricted to more ` model - type ' systems , give indeed quite satisfactory agreement with experiment . the key approximation , namely to treat the hubbard - like operators describing the charge fluctuations as free fermions thereby is well justified , because of the low density of these effective fermions , which renders their ( strong ) interaction largely irrelevant . a systematical way to relax this approximation would be the @xmath241-matrix approach , as demonstrated by kotov _ et al_@xcite . it should be noted that the calculation is computationally no more demanding than a conventional band structure calculation and can be ` automated ' almost completely . the weakest link in the chain thereby is the necessity to perform an lcao - fit to an lda band structure . + one important conceptual problem is the necessity to break the symmetry which originates from the degeneracy of the ground state multiplet of a single transition metal ion and choose the ` reference states ' @xmath78 ` by hand ' . however , one might as well consider choosing an _ ansatz _ for these reference states which takes the form of a linear combination @xmath242 , where the sum extends over the gs multiplet , and determine the coefficients @xmath243 from the requirement of minimum total energy . in this way spin and orbital ordering could be studied in much the same way as lattice parameters are optimized in conventional lda calculations and since all ` ingredients ' for the goodenough - kanamori rules@xcite are taken into account , this may be a quite promising method . since spin - orbit coupling also can be trivially included in the exact diagonalization of the isolated @xmath0-shells one might even hope to address magnetic anisotropies and/or anisotropic exchange interactions . a procedure for the improvement of cpt calculations on model hamiltonians which is similar in spirit has been proposed by potthoff _ et al._@xcite . + one major drawback of the theory clearly is the approximate nature of the calculation of the spectral weight . it should be noted , however , that there is a very clear physical reason for this problem , namely the ` compound nature ' of the zrs - like states which form the top of the valence band . if the present interpretation of these states is the correct one , basically any theory will face similar problems . one possible way out would be to derive a version of the original cpt which can work with site - sharing clusters . + finally , we would like to discuss the relationship between our theory and previous workers in the field . et al._@xcite and takahashi and igarashi@xcite have calculated the quasiparticle band structure of nio more along the lines of conventional many - body theory . starting from a paramagnetic lda band structure ( ref . @xcite ) or an antiferromagnetic hartree - fock band structure ( ref . @xcite ) these authors added a self - energy constructed within the local approximation to three - body scattering theory . the obtained band structures show the same ` large scale features ' as the one obtained here , but there are also significant differences , particularly so near the top of the valence band . more detailed comparison with experiment seems necessary to discuss the merits of the various theories . + next , there is a clear analogy between the present theory and the cluster method of fujimori and minami@xcite and va elp _ at al._@xcite . with the exception of @xmath244 states in the photoemissions spectrum the present theory employs the same type of basis states as the cluster calculations . the only difference is , that we designate one of the degenerate ground states of @xmath3 as a ` vacuum state ' and interpret the other states as ` deviations ' from this vacuum state . those deviations which carry the quantum number of an electron then are considered as effective free fermions . as discussed above , the low density of these effective fermions probably make this a very good approximation . + there is also an obvious relationship between the present theory and the work of unger and fulde@xcite . using the projection technique developed by becker and fulde@xcite these authors constructed an equation of motion for single - particle spectral functions of the cuo@xmath13 plane , which is very similar to the ones which would be obtained from our effective hamiltonians . + finally we address the work of bala _ et al._@xcite , which is very similar in spirit to the present theory . these authors derived a ` kondo - heisenberg'-like model operating in the subspace of @xmath45 type states by eliminating - via canconical transformation - the charge fluctuations between states of the type @xmath245 and states of the type @xmath6 ( their theory was concerned with the motion of a single hole in an @xmath229 orbital ) . accordingly , their theory produced ( in addition to the free - electron - like @xmath229 bands ) two weakly dispersive bands - one for each of the ` flavours ' @xmath219 and @xmath218 whereby the flavour stands for the symmetry of the linear combination of @xmath229 orbitals around a given @xmath246 site . thereby bala _ at al . _ actually went one step beyond the present theory by taking into account the coupling of @xmath229-like holes to the antiferromagnetic magnons - which is omitted in the present theory . in the present theory , no canonical transformation is performed , so that also the high energy features ( satellite and upper hubbard band ) are reproduced . moreover , we also take the excited multiplets of @xmath7 and their covalent mixing with the @xmath6 multiplets into account , whence we obtain a larger number of zrs - like bands - consistent with experiment . experimentally the impact of the coupling to magnons which is ignored in the present theory but treated accurately in the work of bala _ at a. _ could be studied only by considering the ` fine structure ' of the broad peak at the valence band top . these states seem to have an appreciable dispersion which might or might not be influenced by the coupling to magnons . w. kohn and l. j. sham , phys . rev . * 140 * , a1133 ( 1965 ) . k. terakura , t. oguchi , a. r. williams and j. kuebler , phys . b * 30 * , 4734 ( 1984 ) . g. a. sawatzky and j. w. allen , phys . 53 * , 2239 ( 1984 ) . m. r. norman and a. j. freeman , phys . b * 33 * , 8896 ( 1986 ) . a. svane and o. gunnarsson , phys . * 65 * , 1148 ( 1990 ) . z. szotek , w. m. temmerman , and h. winter , phys . b * 47 * , 4029 ( 1993 ) . v. i. anisimov , j. zaanen , and o. k. andersen , phys . rev . b * 44 * , 943 ( 1991 ) . v. i. anisimov , i. v. solovyev , m. a. korotin , m. t. czyzyk , and g. a. sawatzky , phys . rev . b*48 * , 16929 ( 1993 ) . f. aryasetiawan and o. gunnarsson , phys . rev . lett . * 74 * , 3221 ( 1995 ) . s. massidda , a. continenza , m. posternak , and a. baldereschi phys . b * 55 * , 13494 ( 1997 ) . a. fujimori and f. minami , phys . rev . b * 30 * , 957 ( 1984 ) . j. van elp , h. eskes , p. kuiper , and g. a. sawatzky , phys . rev . b * 45 * , 1612 ( 1992 ) . j. hubbard , proc . london , ser . a 277 , * 237 * ( 1964 ) , and * 281 * , 401 ( 1964 ) . d. senechal , d. perez , and m. pioro - ladriere , phys . lett . * 84 * , 522 ( 2000 ) . f. c. zhang and t. m. rice , phys . b * 37 * , 3759 ( 1987 ) . j. zaanen and g. a. sawatzky j. w. allen , phys . lett . * 55 * , 418 ( 1985 ) . oh , j. w. allen , i. lindau , and j. c. mikkelsen , jr . , b * 26 * , 4845 ( 1982 ) . o. tjernberg , s. sderholm , u. o. karlsson , g. chiaia , m. qvarford , h. nyln , and i. lindau , phys . b * 53 * , 10372 ( 1996 ) . l. h. tjeng , c. t. chen , j. ghijsen , p. rudolf , and f. sette , phys . lett . * 67 * , 501 ( 1991 ) . j. bala , a. m. oles , and j. zaanen , phys . * 72 * , 2600 ( 1994 ) ; j. bala , a. m. oles , and j. zaanen , phys . rev . b * 61 * , 13573 ( 2000 ) . h. kuhlenbeck , g. odrfer , r. jaeger , g. illing , m. menges , th . mull , h .- j . freund , m. phlchen , v. staemmler , s. witzel , c. scharfschwerdt , k. wennemann , t. liedtke , and m. neumann , phys . b * 43 * , 1969 ( 1991 ) . z. x. shen , r. s. list , d. s. dessau , b. o. wells , o. jepsen , a. j. arko , r. barttlet , c. k. shih , f. parmigiani , j. c. huang , and p. lindberg , phys . b * 44 * , 3604 ( 1991 ) . r. eder , o. rogojanu , and g. a. sawatzky , phys . b * 58 * , 7599 ( 1998 ) . a. dorneich , m. g. zacher , c. grber , and r. eder , phys . b * 61 * , 12816 ( 2000 ) . m. g. zacher , r. eder , e. arrigoni , and w. hanke , phys . rev . b * 65 * , 045109 ( 2002 ) . o. k. andersen , phys . b * 12 * , 3060 ( 1975 ) . r. g. shulman and s. sugano , phys . rev . * 130 * , 506 ( 1963 ) . v. n. kotov , o. sushkov , z. weihong , and j. oitmaa , phys . lett . * 80 * , 5790 ( 1998 ) . j. b. goodenough , phys . 100 , * 564 * ( 1955 ) , and j. phys . solids * 6 * , 287 ( 1958 ) . j. kanamori , j. phys . solids 10 , * 87 * ( 1959 ) . m. potthoff , m. aichhorn , and c. dahnken , phys . lett . * 91 * , 206402 ( 2003 ) . f. manghi , c. calandra , and s. ossicini , phys . lett . * 73 * , 3129 ( 1994 ) . m. takahashi and j. igarashi , phys . b * 54 * , 13566 ( 1996 ) ; m. takahashi and j. igarashi , ann . phys . * 5 * , 247 ( 1996 ) . p. unger and p. fulde phys . b * 47 * , 8947 ( 1993 ) ; p. unger and p. fulde , phys . b * 48 * , 16607 ( 1993 ) ; p. unger and p. fulde phys . rev . b * 51 * , 9245 ( 1995 ) . k. w. becker and p. fulde , z. phys . b * 72 * , 423 ( 1988 ) ; k. w. becker and w. brenig , z. phys . b * 79 * , 195 ( 1990 ) .

we study the evolution of tensor metric fluctuations in a class of non - singular models based on the string effective action , by including in the perturbation equation the higher - derivative and loop corrections needed to regularise the background solutions . we discuss the effects of such higher - order corrections on the final graviton spectrum , and we compare the results of analytical and numerical computations .

gr-qc0101019

recent studies , including strong coupling corrections to the string effective action , have provided the pre - big bang scenario @xcite with a number of promising models for the transition from the growing to the decreasing curvature regime @xcite . however , a crucial question remains : is it possible to test this cosmological scenario and , in particular , to distinguish it from other , more conventional descriptions of the very early universe ? the answer to this essential question is in principle positive , since the pre - big bang scenario incorporates a dilaton field which modifies the standard inflationary kinematics and which , through its non - minimal coupling , also directly modifies the evolution of perturbations ( of the metric , and of the other background fields ) @xcite . it is well known , on the other hand , that the transition from an inflationary period to the usual friedman - robertson - walker ( frw ) phase amplifies tensor metric perturbations , and is associated with the production of a stochastic background of relic gravitational waves @xcite . such a primordial background decouples from matter immediately below the planck scale , unlike the electromagnetic radiation which underwent a complicated history until recombination , and has been transmitted almost unperturbed down to our present epoch . as a consequence , its present spectrum should be a faithful portrait of the very early universe @xcite , thus opening a window for the observation of processes occurring near to the planck scale , and for discriminating among various models of primordial evolution . in the context of string cosmology , several mechanisms may contribute to the generation of gravitational waves from the initial vacuum state . one contribution comes from the process of dynamical dimensional reduction @xcite , when the internal dimensions shrink down to a final compactification scale . another contribution is due to the time - dependence of the dilaton field and thus of the effective gravitational coupling constant during the pre - big bang phase @xcite . finally , there is the usual contribution arising from the accelerated expansion of the external three - dimensional space . both the dilaton and the full , higher - dimensional metric thus contribute to the external pump " field which is responsible for the parametric amplification of tensor metric fluctuations , normalised to an initial vacuum fluctuation spectrum . with respect to the standard inflationary scenario , the amplification of tensor perturbations in the pre - big bang scenario is strongly enhanced for large comoving wavenumber @xmath0 @xcite , in such a way that the produced background of gravitational waves could in principle be detected in the near future by various planned experiments @xcite . arguments in favour of this conclusion arise from the particular kinematics of string cosmology . first , since the hubble parameter is increasing in the pre - big bang phase , higher frequency modes will cross the horizon at higher values of @xmath1 , which implies that the amplitude of the spectrum increases with frequency . in this context , the comoving amplitude of perturbations may grow even outside the horizon @xcite , instead of being frozen as happens in standard inflation . second , the peak amplitude of the spectrum ( naturally located in the high frequency regime , near the end point of the spectrum ) is normalised so as to match the string scale @xcite , i.e. nearly eight orders of magnitude above the high frequency normalisation of the standard scenario @xcite . hence , the pre - big bang phase could lead to a very efficient production of gravitons in the sensitivity range of present gravitational antenna . in fact such a relic background could be in principle be detected by the second ( planned ) generation of interferometric detectors @xcite . to date , extended studies of tensor perturbations have been mostly performed in the context of the tree - level effective action , with and without the possible contribution of a high curvature stringy phase . in all cases the background cosmologies that have been investigated have been singular , i.e. the curvature and dilaton have become singular at some fixed proper time . there have been a number of suggested ways to tackle the singularity problem , considering also anisotropic @xcite and inhomogeneous @xcite backgrounds , or the presence of a non - local dilaton potential @xcite . however , if we confine our attention to homogeneous and isotropic metrics , and to a local potential , one of the most promising approaches to the graceful exit problem @xcite suggests that the curvature singularities may be cured by adding higher - order corrections to the string effective action ( see for instance @xcite ) . these include both higher - derivative @xmath2 corrections , due to finite - size effects , and quantum corrections , due to the loop expansion in the string coupling parameter . in the context of cosmological perturbation theory , it is known @xcite that such corrections may induce an additional amplification when the curvature becomes large in string units . the aim of this paper , therefore , is to discuss the evolution and the spectrum of tensor perturbations in the context of a class of non - singular cosmological backgrounds , by taking into account ( in the perturbation equation ) the full contribution of those corrections that are responsible for the regularisation of the background evolution . in this way it will be possible ( for the first time , to the best of our knowledge ) to follow the complete evolution of perturbations , from the initial vacuum down to the present time , through a continuous numerical integration of the linearised perturbation equations . the paper is organised as follows . in section 2 we recall the general form of the first order @xmath2 corrections arising in the context of the massless bosonic sector of the low - energy heterotic string action . we extend them by including possible one- and two - loop quantum corrections , and we write down the corresponding background equations . in section 3 we compute the linearised tensor perturbation equation , including such higher - order corrections . in section 4 we present a general discussion of the evolution of perturbations in the presence of higher - order curvature corrections . we then check the main points of our discussion through a numerical integration of the perturbation equations for regular backgrounds , by using an appropriate dilaton potential , or particle production effects , to stabilise the dilaton in the post - big bang era . we find that the general effect of the higher - order corrections is to flatten the slope of the high - frequency part of the spectrum , but the effect is probably too small in the class of models discussed in this paper to make the graviton background visible " to the planned advanced detectors . the main results of this paper are finally summarised in section 5 . in the context of the pre - big bang scenario , our present frw universe is expected to emerge from the string perturbative vacuum as a consequence of some process for instance , plane wave collisions @xcite triggering the gravitational collapse of a sufficiently large portion of spacetime @xcite . the initial evolution is driven by the kinetic energy of the dilaton , and can be described , in the string frame , as inflationary expansion with growing coupling and curvature . after reaching a maximal scale controlled by the fundamental string length @xmath3 , @xmath4 , it is then expected that the universe is smoothly connected to the frw regime , with a constant dilaton field . in the early stage of the dilaton - driven inflationary ( ddi ) period , strings are propagating in a background of small curvature , and the fields are weakly coupled , so that the cosmological evolution is consistently described by the lowest order string effective action which , by assuming static internal dimensions , can be written as : @xmath5 + { \cal l}_{c } \right\}.\ ] ] here @xmath6 , we have adopted the convention @xmath7 , @xmath8 , @xmath9 and set our units such that @xmath10 . following @xcite , we have introduced an additional lagrangian @xmath11 allowing for the inclusion of higher - order corrections to the tree - level effective action , a non - perturbative potential yet to be determined and/or the backreaction of the produced radiation . considering this additional lagrangian as an independent source we shall assume that the associated energy - momentum tensor is diagonal , homogeneous and isotropic , and can be written in the perfect fluid form as @xmath12 . the resulting equations of motion for the background are obtained from the variation of eq . ( [ e : low ] ) with respect to the metric and the dilaton field : @xmath13 a dot represents the differentiation with respect to proper time @xmath14 and @xmath15 is the hubble parameter . @xmath16 arises from the variation of the additional lagrangian with respect to @xmath17 . finally , eq . ( [ e : e_constraint ] ) is the conservation equation , a direct consequence of eq . ( [ e : eq_rho])eq . ( [ e : eq_phi ] ) . to lowest order , the homogeneous and isotropic version of the pre - big bang scenario , without non - local dilaton potential , inevitably faces a curvature and dilaton singularity at the end of the ddi phase . however , a singular behaviour is often a late manifestation of the breakdown of the description , extended beyond its domain of validity . hence , the low - energy dynamics should be supplemented by corrections , in order to give a reliable description of the high curvature regime , expected to be crossed before reaching the frw universe . such corrections are generally of two types : tree - level @xmath2 corrections , referring to the finite size of the string , and quantum corrections , resulting from a more conventional loop expansion in the string coupling parameter . it has been shown in @xcite that the higher curvature @xmath2 terms , arising from the expansion around the point - particle limit , may eventually stabilise the evolution of the universe into a de - sitter like regime of constant curvature @xmath18 and linearly growing ( in cosmic time ) dilaton . in this paper we will consider the most general tree - level @xmath2 corrections to the gravi - dilaton string action up to fourth - order in derivatives ( see @xcite for a detailed analysis ) , which can be written in the form @xcite : @xmath19 @xmath20 the constraint eq . ( [ e : constraint ] ) on the coefficients is required so that the action reproduces the usual string scattering amplitudes @xcite . the parameter @xmath21 allows us to move between different string models , and we will later use @xmath22 to agree with previous studies of the heterotic string @xcite . finally , @xmath23 is the gauss - bonnet combination , guaranteeing the absence of derivatives higher than two in the equations of motion . another conventional expansion of the string effective action is controlled by the string coupling parameter , @xmath24 . initially , @xmath25 . however , the growth of the dilaton during the ddi era implies that the late evolution of the background could be dominated by quantum loop corrections , which will eventually force the universe to escape from the fixed point determined by the @xmath2 corrections @xcite , for a smooth connection to the post - big bang regime . unfortunately , there is as yet no definitive calculation of the full loop expansion in string theory , so we are left to speculate on plausible terms that will eventually make up the loop contribution . multiplying each term of the tree - level @xmath2 correction by a suitable power of the string coupling is the approach which has already met with some success @xcite , and which we shall adopt also in this paper . since the quantum corrections are not formally derived from an explicit computation , we shall allow for different coefficients @xmath26 ( @xmath27 ) at one - loop ( two - loop ) in the string coupling , replacing the coefficients @xmath28 of the tree - level action , where @xmath26 and @xmath27 are not necessarily subject to the constraint eq . ( [ e : constraint ] ) . with such assumptions , the effective lagrangian density @xmath11 of eq . ( [ e : low ] ) , which is in general the sum of the tree - level @xmath2 and loop corrections , @xmath29 , in the present case will take the form @xmath30 where @xmath31 is given by eq . ( [ e : class ] ) , with the constant parameters @xmath32 and @xmath33 actually controlling the onset of the loop corrections . the very similarity between the first order @xmath2 and loop corrections enables a simple form for their contributions to the background equations of motion . from now on , we introduce an additional parameter @xmath34 in the exponential of the correction @xmath11 , such that @xmath2 corrections correspond to @xmath35 , whereas one- or two - loop corrections correspond to @xmath36 and @xmath37 respectively . for instance , using the notation @xmath38 to represent the one - loop contribution ( @xmath36 inside the brackets ) , we easily get the source terms for the system eq . ( [ e : eq_rho])eq . ( [ e : eq_phi ] ) : @xmath39 and the terms in brackets are given respectively by : @xmath40 \dot{\phi}^3 \bigr\ } , \\ & & \nonumber \\ \{p_{c}\}_{n } & = & \alpha ' \lambda_{0}e^{(n-1)\phi } \bigl\{\left[8 c_1 ( n-1)^2 - 3 c_2 \right ] \dot{\phi}^2 h^2 - 2 c_2(n-1 ) \dot{\phi}^3 h - 2 c_{3 } \dot{\phi}^2 \ddot{\phi}\\ & & \hspace{2,1 cm } + 8 c_1 ( n-1 ) \ddot{\phi } h^2 + 16 c_1(n-1 ) \dot{\phi } h \left[\dot{h } + h^2 \right ] \nonumber \\ & & \hspace{2,1 cm } - 4 c_2 \dot{\phi } \ddot{\phi } h -2 c_2 \dot{\phi}^2 \dot{h } \nonumber + \left [ c_4 - c_3(n-1 ) \right]\dot{\phi}^4\bigr\},\\ & & \nonumber \\ \{\delta_{\phi}\}_{n } & = & \alpha ' \lambda_{0 } e^{(n-1)\phi } \bigl\{24c_1(n-1 ) h^2 \left[\dot{h } + h^2 \right ] - 3 c_4\dot{\phi}^2 \left [ 4 \ddot{\phi } + 4 \dot{\phi}h + ( n-1 ) \dot{\phi}^2 \right ] \nonumber \\ & & \hspace{2,1cm}+ c_3 \dot{\phi } \left [ ( n-1)^2 \dot{\phi}^3 + 4 ( n-1 ) \dot{\phi } \ddot{\phi } - 6 \dot{\phi } \left(\dot{h } + 3 h^2 \right ) - 12 \ddot{\phi } h \right]\bigr\ } \nonumber \\ & & \hspace{2,1cm}- 3 c_2 \left[2 \ddot{\phi } h^2 + 6 \dot{\phi } h^3 + 4 \dot{\phi } h \dot{h } + ( n-1 ) \dot{\phi}^2 h^2 \right ] \bigr\}. \nonumber\end{aligned}\ ] ] here , of course , for @xmath36 the coefficients @xmath28 are replaced by @xmath26 , while for @xmath37 the @xmath28 are replaced by @xmath27 . note that for @xmath35 , @xmath41 , @xmath42 , one easily recovers the @xmath2 corrections used in @xcite . in general , the combination of tree - level @xmath2 and quantum loop corrections does not lead to a fixed value for the dilaton in a finite amount of time . although a non - perturbative ( supersymmetry breaking ) potential @xcite is expected to stabilise the dilaton , here we shall follow @xcite by assuming the dilaton is frozen out by radiation , after the transition to the post - big bang regime . to this aim we will introduce by hand " the presence of radiation , coupled to the dilaton and satisfying the conservation equation @xmath43 where @xmath44 represents the decay width of the dilaton . the solutions of the system of equations eq . ( [ e : eq_rho ] ) eq . ( [ e : eq_phi ] ) , including the radiation and the quantum corrections , provide the cosmological gravi - dilaton background in which we will study the propagation of tensor perturbations . we will consider a spatially flat frw manifold , and we will perturb the above equations keeping the dilaton and all sources fixed , @xmath45 , @xmath46 . to lowest order , the tensor fluctuations will obey the usual string frame perturbation equation @xcite , including the non - minimal coupling to a time - dependent dilaton . on the other hand , the inclusion of higher - order corrections , needed to regularise the background , will inevitably lead to a modification of such a perturbation equation ( see @xcite for initial studies taking into account higher curvature contributions to the evolution of perturbations ) . it is well known that gravitational waves , arising from linearised tensor perturbations , do not couple to pressure and energy density , and hence do not contribute to classical gravitational instabilities . nevertheless , they are of interest as a specific signature of pre - big bang cosmologies since , as stressed in a number of papers @xcite , the production of high - frequency gravitons is strongly enhanced , in string cosmology , with respect to the standard inflationary scenario . in this study we will restrict our attention to tensor metric perturbations around a @xmath47 spatially flat background , parameterised by the transverse , trace - free variable @xmath48 , @xmath49 where @xmath50 denotes covariant differentiation with respect to the background metric . note also that the indices of @xmath48 are raised or lowered with the unperturbed metric , @xmath51 . the linear evolution equation for the metric fluctuations can be obtained by perturbing the action eq . ( [ e : low ] ) up to terms quadratic in the tensor variable @xmath48 : @xmath52 \nonumber \\ & & + \lambda_{0 } \int d^{4}x \left [ \{\delta^{(2)}{\cal l}\}_{0 } + a\{\delta^{(2)}{\cal l}\}_{1 } + b\{\delta^{(2)}{\cal l}\}_{2 } \right],\end{aligned}\ ] ] where the integrand of the @xmath34-correction is given by : @xmath53 \partial_{\mu } \phi \partial_{\nu } \phi\right ) \nonumber \\ & & \hspace{2cm}+ \delta^{(2)}\left(c_3\sqrt{-g}\box \phi ( \partial_{\mu } \phi)^{2}\right ) + \delta^{(2)}\left(c_4\sqrt{-g}(\partial_{\mu } \phi)^{4 } \right)\biggr\}.\end{aligned}\ ] ] following @xcite , it is convenient to use the synchronous gauge where @xmath54 this enables us to write the perturbed action as a quadratic form depending on the first and second derivatives of the symmetric , trace - free matrix @xmath55 , with time - dependent coefficients fixed by the background fields @xmath56 , @xmath57 . introducing the matrix notation @xmath58 , and using @xmath59 for the laplace operator in a flat space , we have @xmath60 h^2 \nonumber \\ & & \hspace{3,5cm}- h\ddot{h } - 4 h h \dot{h } - \frac{3}{4 } \dot{h}^2 + \frac{1}{4 } h \frac{\nabla^2}{a^2}h \biggr\ } \nonumber \\ & & + \lambda_{0 } \int d^{4}x \;\textrm{tr}\ ; \left [ \{\otimes\}_{0 } + a\{\otimes\}_{1 } + b\{\otimes\}_{2 } \right],\end{aligned}\ ] ] where the integrand of the @xmath34-correction is given by : @xmath61 h^{2 } \\ & & \hspace{1,8 cm } - \left [ c_{1 } \dot{h } + 7 c_1 h^2 + \frac{1}{8}c_{2}\dot{\phi}^2\right ] \dot{h}^2 - 4 c_{1 } h^2 h \ddot{h } - 2 c_{1 } h \dot{h } \ddot{h } \nonumber \\&&\hspace{1,8 cm } - \left [ 8 c_{1 } h ( \dot{h } + 2 h^2 ) + c_{2 } \dot{\phi}^2 h + \frac{1}{2 } c_{3 } \dot{\phi}^3 \right ] h \dot{h } + 2 c_{1 } \dot{h } \frac{\nabla^2}{a^2}\dot{h } \nonumber \\&&\hspace{1,8 cm } + \left[c_{1 } h^{2 } + c_{1}\dot{h } + \frac{1}{8 } c_{2 } \dot{\phi}^2 \right ] h\frac{\nabla^2}{a^2}h + 4 c_1 h \dot{h}\frac{\nabla^2}{a^2}h + 2 c_1 \ddot{h}\frac{\nabla^2}{a^2}h\biggr\}. \nonumber\end{aligned}\ ] ] here , and in what follows , the coefficients @xmath28 are to be replaced by @xmath26 and @xmath27 for @xmath36 and @xmath37 , respectively . again , for @xmath35 , @xmath41 , @xmath42 , one recovers the first order @xmath2 corrections discussed in @xcite . we can now integrate by parts all the terms with more than two partial derivatives acting on @xmath62 , as well as the terms in @xmath63 and @xmath64 . this will drop all terms with more than two derivatives thanks to the gauss - bonnet combination , leading to @xmath65 , \label{e : eq_2_var}\end{aligned}\ ] ] where we have @xmath66 \dot{\phi}^2 h^2 - 2 c_2 ( n-1 ) \dot{\phi}^3 h \\ & & \hspace{1,9 cm } + \left[c_4 - c_3 ( n-1)\right ] \dot{\phi}^4 -2 c_3 \dot{\phi}^2 \ddot{\phi } + 8 c_1 ( n-1 ) \ddot{\phi } h^2 \nonumber\\ & & \hspace{1,9 cm } + 16 c_1 ( n-1 ) \dot{\phi } h \left[\dot{h } + h^2\right ] -4 c_2 \dot{\phi } \ddot{\phi } h -2 c_2 \dot{\phi}^2 \dot{h } \bigr ) \nonumber \\ & & \hspace{1,9cm}+ h\frac{\nabla^2}{a^2}h \left ( \left [ c_1 ( n-1)^2 + \frac{1}{8 } c_2\right ] \dot{\phi}^2 + c_1(n-1 ) \ddot{\phi}\right ) \nonumber \\ & & \hspace{1,9cm}+ \dot{h}^2 \left ( c_1 ( n-1 ) \dot{\phi } h - \frac{1}{8 } c_2 \dot{\phi}^2 \right)\biggr\}.\end{aligned}\ ] ] we stress that all tree - level @xmath2 and quantum loop corrections disappear in the limit of a constant dilaton field , since in that case the contribution of the gauss - bonnet term corresponds a topological invariant . this ensures that the resulting perturbation equation reduces to the klein - gordon form , typical of a minimally coupled scalar , as we enter the frozen dilaton , post - big bang regime . the rather complicated form of eq . ( [ e : eq_2_var ] ) can be simplified as the coefficient of the @xmath67 term vanishes identically , thanks to the background equation eq . ( [ e : eq_p ] ) . by decomposing the matrix @xmath68 into the two physical polarisation modes of tensor perturbations , @xmath69 and @xmath70 , @xmath71 we can finally write the action , for each polarisation mode @xmath72 , as @xmath73 with @xmath74 , \nonumber \\ \{\otimes\}_{n } & = & \alpha ' \lambda_0 e^{n\phi } \left[4 c_1 ( n-1)^2 \dot{\phi}^2 + \frac{1}{2 } c_2 \dot{\phi}^2 + 4 c_1 ( n-1 ) \ddot{\phi } \right],\end{aligned}\ ] ] where @xmath62 is now a scalar variable standing for either one of the two polarisation amplitudes @xmath69 , @xmath70 . the variation of the action with respect to @xmath62 gives then the modified perturbation equation : @xmath75 \ddot{h } \nonumber\\ & & + \left [ 3h - \dot{\phi } + \{\otimes\}_{0 } + a\{\otimes\}_{1 } + b \{\otimes\}_{2}\right ] \dot{h } \nonumber \\ & & - \left [ 1 + \{\ominus\}_{0 } + a\{\ominus\}_{1 } + b \{\ominus \}_{2}\right ] \frac{\nabla^2}{a^2 } h , \label{e : eq_pol}\end{aligned}\ ] ] with @xmath76 eq . ( [ e : eq_pol ] ) controls the time evolution of the fourier components @xmath77 of the two polarisation modes , and is valid even during the high curvature regime connecting the pre - big bang branch to our present frw universe , since it incorporates the contributions of the higher - order corrections needed to regularise the background evolution . we shall now focus on the generation of gravitational waves in the context of the pre - big bang scenario . to normalise the graviton spectrum to the quantum fluctuations of the vacuum we need the canonical variable that diagonalises the perturbed action , and that represents in this case the normal modes of tensor oscillations of our gravi - dilaton background . to this purpose we note that introducing the conformal time coordinate @xmath78 , defined by @xmath79 , the action eq . ( [ e : action_pol ] ) can be re - written @xmath80 where a prime denotes differentiation with respect to the conformal time , and @xmath81 , \label{e : y_gen } \\ z^{2}(\eta ) & = & e^{-\phi } \left[a^2 + \{\otimes\}_{0 } + a\{\otimes\}_{1 } + b \{\otimes\}_{2 } \right ] , \label{e : z_gen}\end{aligned}\ ] ] with @xmath82 introducing the canonical variable @xmath83 enables us to diagonalise the kinetic part of the action : @xmath84 for each fourier mode , @xmath85 , we can now obtain from the perturbed action a linearised wave equation in terms of the eigenstates of the laplace operator , @xmath86 which takes the explicit form : @xmath87 this linear perturbation equation is the major result of the first part of this paper , since it encodes ( through eq . ( [ e : y_gen ] ) eq . ( [ e : z_gen ] ) ) the full contribution of those higher - order corrections which can be used to regularise the background evolution . we note that the above equation defines an effective time - dependent mode , @xmath88 , whereas the effective potential is determined by the evolution of the generalised pump field @xmath89 . these new features of our generalised perturbation equation will be discussed in detail in the next section . the graviton spectrum of pre - big bang models generated in the context of the lowest - order perturbation equation , with and without the contribution of a high curvature string phase , has been discussed extensively in the literature ( see for example @xcite ) . in the remaining part of this section we shall briefly recall some general results of these studies , so as to highlight the modifications induced by the higher - order corrections in the perturbation equation , and in the spectrum of the relic pre - big bang gravitons . in the extremely weak coupling and low curvature regime , the tree - level solutions of the pre - big bang scenario are fully adequate to describe , asymptotically , the initial evolution of the quantum fluctuations of the gravi - dilaton background . in normalising the fluctuations in the asymptotic past @xmath90 , we shall thus neglect any correction to the low - energy action , considering the standard perturbation equation for the fourier mode @xmath91 : @xmath92 \psi_{k } = 0.\label{e : psi_reduced}\ ] ] such an equation describes the evolution of a classical harmonic oscillator undergoing a parametric amplification driven by the effective potential @xmath93 , with @xmath94 . by recalling the ( conformal time ) expression of the tree - level solutions @xcite , @xmath95 we obtain the external pump field @xmath96 . hence , the power - law type of the effective potential leads to two distinct regimes for the evolution equation eq . ( [ e : psi_reduced ] ) . sub - horizon modes @xmath97 are freely oscillating , and no quantum particle production occurs . however , the non - adiabatic behaviour of modes having left the hubble radius @xmath98 in the dilaton - driven phase , are described asymptotically by the general solution @xcite @xmath99 where @xmath100 , and @xmath101 is the comoving amplitude of tensor perturbation for the given mode ( @xmath102 and @xmath103 indicate the coefficients of the growing and decaying solutions , respectively ) . the amplitude of super - horizon modes is thus frozen , modulo logarithmic corrections . the overlapping of these two extreme behaviours is given by the exact solution to the bessel - type equation eq . ( [ e : psi_reduced ] ) , which can be expressed in a linear combination of the first- and second - kind hankel functions : @xmath104 where @xmath105 encompasses the background dynamics . in the asymptotic past ( @xmath106 ) any given mode @xmath0 is well inside the hubble radius ( as @xmath107 blows up as we go back in time in pre - big bang models ) , and the two types of hankel functions @xmath108 are oscillating , corresponding to negative and positive frequency modes respectively . it is therefore possible to normalise tensor perturbations to a spectrum of quantum fluctuations of the initial background . indeed , the annihilation and creation operators resulting from the expansion of the @xmath48 field , with @xmath109 , obey canonical commutation relations @xmath110 = [ a_{k}^{\dag},a_{k'}^{\dag } ] = 0 $ ] and @xmath111 = \delta_{k k'}$ ] . for the normalisation to an initial vacuum state , we choose to restrict to positive energy states , @xmath112 , hence @xmath113 and @xmath114 . the effective potential of the perturbation equation is expected to grow monotonically during the initial dilaton - driven regime , up to a maximal value reached during the string phase , while it is expected to decrease rapidly to zero at the onset of the standard , radiation dominated phase . emerging from the asymptotic past , a typical mode @xmath0 will thus first oscillate inside the horizon and then progressively feel the potential barrier . the process of quantum particle production will take place and carry on until the background enters the standard frw - regime . after that , the effective potential identically vanishes , and eq . ( [ e : psi_reduced ] ) yields the free oscillating solution : @xmath115 with incoming and outgoing waves . the mean number of produced gravitons is obtained from the square modulus of the bogoliubov coefficient , @xmath116 , using the matching of the tensor perturbation variable @xmath62 and @xmath117 at the onset of the frw phase . the low - energy perturbation equation may allow for estimating a lower bound on the spectrum of the produced gravitons @xcite . for the tree - level solution with @xmath118 , by defining the spectral energy density per logarithmic interval of frequency , @xmath119 , one obtains in particular @xcite : @xmath120 where @xmath121 refers to the maximal height of the effective potential barrier , @xmath122 . we recall that the cubic slope of the spectrum is a direct consequence of the evolution of the background during the ddi branch . in @xcite , the authors argued that the contribution of a high curvature string phase would lead in general to a reduction of the slope of the spectrum of relic gravitational waves . this was explicitly confirmed for a model based only on the tree - level @xmath2 corrections @xcite . however , we stress that the investigation of the evolution of high frequency modes , hitting the potential barrier during the high curvature regime , needs to be performed using the full perturbation equation eq . ( [ e : psi_full ] ) , including also the loop corrections . in this section we highlight the modifications to the tensor perturbation equations induced by the corrections adopted to regularise the background evolution . to this purpose , we shall first restrict the discussion to the tree - level @xmath2 corrections ; the reason for this choice is twofold . first , @xmath2 corrections are derived unambiguously since we require them to reproduce the string scattering amplitude , as opposed to our lack of precise knowledge for quantum loop corrections , which forces us to speculate on plausible terms . second , if any clear imprint of a string phase " like the one introduced in @xcite is to be found in the spectrum of relic gravitons , it should be a direct consequence of a long enough string phase , where @xmath123 is the number of e - folds . ] characterised by a constant hubble parameter . as suggested in @xcite , this can be achieved by assuming that the evolution of the universe is driven by @xmath2 corrections into a fixed point in the @xmath124 space . in the final part of this section we will comment on the impact of the quantum loop corrections , and confront our predictions with the results obtained through exact numerical integrations of the perturbation equation . the lowest - order effective action of string theory provides an adequate description of the dilaton - driven branch of the pre - big bang scenario . in this weakly coupled and low curvature regime , no correction is required either to the effective potential nor to the wavelength of a given perturbation , and the evolution equation of each fourier mode reduces to its usual form eq . ( [ e : psi_reduced ] ) . however , it is well known that the kinetic energy of the dilaton necessarily drives an isotropic and homogeneous background towards a curvature singularity , imposing the need to supplement the tree - level action with corrections in order to establish a reliable description of the high curvature regime . the very presence of the tree - level @xmath2 corrections eq . ( [ e : class ] ) , subject to the constraint eq . ( [ e : constraint ] ) , may lead in general to a regime of constant hubble parameter and a linearly growing dilaton , @xmath125 const , @xmath126 , which seems to represent an exact conformal solution of string theory also to all orders in @xmath2 @xcite . however , at any given finite order , we recall that not all sets of coefficients of the truncated action lead to such a good fixed point in the @xmath124 plane ( see @xcite for a detailed analysis ) . we shall thus restrict the analysis to the region of parameters which in principle may allow for a graceful exit , when loop corrections are included . in the @xmath127 parameter space , such a region is delimited on the one hand by @xmath128 , referring to a change of sign of the time derivative of the shifted dilaton , @xmath129 @xcite . on the other hand , tree - level @xmath2 corrections , truncated to first order , are not expected to violate the null energy condition , and thus can not lead in principle to fixed points located after the einstein bounce @xcite ( i.e , the bounce of the scale factor in the einstein frame ) . this implies @xmath130 for @xmath131 , to a first approximation . the location of such fixed points , satisfying the above constraints , can be seen in fig . [ f : shift ] . the figure on the left , from ref . @xcite , shows the location in the @xmath132 space of the fixed points for different choices of the parameters @xmath133 and @xmath134 and the corresponding ratio @xmath135 . the blue line shows the @xmath133-dependence of the fixed points for a constant @xmath136 , while the green line shows the @xmath137-dependence for a constant @xmath138 . the point labeled by ( 1 ) corresponds to @xmath139 , with @xmath140 . the point labeled by ( 2 ) corresponds to @xmath141 , with @xmath142 . the point labeled by ( 3 ) corresponds to @xmath143 , with @xmath144 . on the right , we show the tree - level @xmath2 asymptotic shift @xmath145 given by eq . ( [ e : classical_shift ] ) as a function of the parameter @xmath133 and of the fixed point distribution . the blue line shows the evolution of the shift for a constant @xmath136.,title="fig:",width=226,height=188 ] the figure on the left , from ref . @xcite , shows the location in the @xmath132 space of the fixed points for different choices of the parameters @xmath133 and @xmath134 and the corresponding ratio @xmath135 . the blue line shows the @xmath133-dependence of the fixed points for a constant @xmath136 , while the green line shows the @xmath137-dependence for a constant @xmath138 . the point labeled by ( 1 ) corresponds to @xmath139 , with @xmath140 . the point labeled by ( 2 ) corresponds to @xmath141 , with @xmath142 . the point labeled by ( 3 ) corresponds to @xmath143 , with @xmath144 . on the right , we show the tree - level @xmath2 asymptotic shift @xmath145 given by eq . ( [ e : classical_shift ] ) as a function of the parameter @xmath133 and of the fixed point distribution . the blue line shows the evolution of the shift for a constant @xmath136.,title="fig:",width=226,height=188 ] to describe the evolution of the high frequency modes , which are expected to leave the hubble radius during the high curvature regime , we must use the perturbation equation eq . ( [ e : psi_full ] ) , including those corrections required to regularise the background evolution . as mentioned in section @xmath146 , the incorporation of the higher - order corrections in the perturbation equation implies a time - dependence of the effective wavenumber @xmath0 . indeed , eq . ( [ e : psi_full ] ) can be conveniently rewritten @xmath147 - v(\eta)\right\ } \psi_{k } = 0 , \label{41}\ ] ] where we have introduced the effective shift in the comoving frequency @xmath148 which , at next - to - leading order in the string tension expansion , reads @xmath149 in the weak coupling regime , and for small curvature @xmath150 , the shift in the frequency is zero , hence no modification to the ddi cubic branch of the relic gravitational spectrum is expected . however , this frequency shift will be no longer negligible when @xmath151 . in general , we find that @xmath152 tends to grow , although a short decrease ( which still satisfies @xmath153 ) may happen at the onset of the string phase . we also observe that the modification arising in the asymptotic regime , where the curvature is saturated by a linearly growing dilaton , corresponds to a constant shift of the comoving frequency , as first suggested in @xcite . in that case , by setting @xmath154 and @xmath155 , we can rewrite the asymptotic version of eq . ( [ 1 ] ) as @xmath156 during such an intermediate string phase the external potential reduces to its tree - level version , namely @xmath157 , and the enhancement in frequency associated to eq . ( [ e : classical_shift ] ) is the same for all modes , since the shift @xmath145 does not depend on the wavenumber @xmath0 . in fig . [ f : shift ] we have represented the effective shift as a function of the parameter @xmath133 , highlighting the @xmath133-dependence by drawing the curve for @xmath136 . although the asymptotic shift can not be expressed as a function of an unique variable , it is however possible to extract its minimal value for the type of @xmath2 correction we are considering . the numerical simulations suggest that the lowest shifts correspond to fixed points located near the branch change line , @xmath158 . the bound @xmath159 attached to the sign change of @xmath160 implies that , in the limit of large values of @xmath161 , the effective asymptotic shift is always positive , and is given by @xmath162 . does the frequency shift have any implication on the spectral distribution ? in principle the answer is yes , since higher frequency modes tend to leave the hubble radius at larger values of @xmath163 in the pre - big bang scenario . as a consequence , a given comoving mode will spend less time under the potential barrier , resulting in a smaller amplification of the frequency attached to this mode . although it is difficult to quantify such an effect , we believe the higher - curvature corrections amount to an overall rescaling , by a numerical factor of order unity , of the total energy density of the background , as already discussed in @xcite . finally , the form of the effective shift eq . ( [ e : classical_shift ] ) suggests that a background undergoing a non - singular evolution in the vicinity of @xmath164 will automatically imply a rapid growth ( if not a divergence ) of the amplitude of the tensor fluctuations . hence , we can understood the denominator of eq . ( [ e : classical_shift ] ) as a source of quantum gravitational instability , of the type already discussed in @xcite . in recent studies of the graceful exit problem , in the context of the pre - big bang scenario , the regularisation of the background curvature and the transition to the post - big bang regime , are obtained by including quantum loop corrections , according to eq . ( [ e : efflag ] ) . in this context , we may note that the tree - level @xmath2 shift eq . ( [ e : classical_shift ] ) provides an interesting constraint on the choice of parameters we can use in order to implement such a graceful exit , since in general the divergence curve for the high curvature shift is very close to the condition for the fixed point in the @xmath132 plane . in @xcite , it has been shown that for a constant @xmath137 the fixed point value of the hubble parameter tends to be reduced when @xmath133 is decreasing . simultaneously the location of the singularity curve also decreases in the @xmath132 plane , so that the range of coefficients leading to a non - singular evolution of the background , and avoiding the instability of quantum fluctuations , is strongly restricted to values very close to the simplest case of @xmath2 correction , @xmath165 ( as also discussed in @xcite ) . this is an unexpected , although fortunate constraint on the loop corrections , which allows restricting the range of values for the parameters @xmath26 and @xmath27 . in our attempt to characterise the power spectrum of metric perturbations generated in such a class of non - singular models , we will thus consider the perturbation equation eq . ( [ 41 ] ) with a full shift determined by @xmath166 - 4 \left(\phi''- 2\frac{a'\phi'}{a}\right ) \left[1 + 4 b e_1 e^{2\phi}\right]\right\ } } { 8 a^2-\alpha'\phi'\left\{8\frac{a'}{a}\left[1 + b e_1 e^{2\phi}\right ] -\phi ' \left[c_2 + a d_2 e^{\phi } + b e_2 e^{2\phi}\right]\right\}}.\ ] ] here the coefficients of the high curvature , one- and two loop corrections are respectively @xmath167 and @xmath27 , with @xmath168 from the constraint eq . ( [ e : constraint ] ) . as already stressed , this frequency shift reduces to a constant in the asymptotic stage of the string phase . the loop corrections re - establish the time - dependence during the graceful exit , but its contribution soon becomes negligible at the end of the phase of accelerated evolution . indeed , eq . ( [ e : y_gen ] ) eq . ( [ e : z_gen ] ) rapidly reduce to their tree - level expressions @xmath169 , hence we expect @xmath170 before we enter the radiation - dominated epoch , where the frequency shift vanishes identically . we are interested in the generic features of the primordial gravitational waves , generated by quantum fluctuations of the background metric during the pre - big bang phase , which could be detected by both ground - based detectors such as ligo @xcite and virgo @xcite , and space - based experiments such as lisa @xcite ( see also @xcite for a recent review and references therein ) . for such a purpose , we use the dimensionless spectral energy density of perturbations , @xmath171 , to describe the background of gravitational radiation . in critical units , the energy density is defined by @xmath172 where @xmath173 is the present energy density in the stochastic gravitational waves per logarithmic interval of frequency , @xmath174 is the physical ( angular ) frequency of the wave , and @xmath175 is the critical energy density required to close the universe . following the standard procedure for the computation of the energy spectrum @xcite , we must determine the expectation number of gravitons per cell of the phase space , @xmath176 , which only depends on the frequency for an isotropic and stochastic background . neglecting @xmath2 and loop corrections , both in the background and perturbation equations , the lowest order result eq . ( [ e : rho_spectrum ] ) leads to @xcite @xmath177 where @xmath178 is the maximal amplified proper frequency , associated with the value @xmath179 of the hubble expansion parameter at the end of the string phase . including higher - order corrections , the relation between the ultraviolet cutoff of the spectrum and the time of horizon crossing is in general more complicated , and a precise definition of the maximal amplified frequency is out of reach without a complete analytic solution for the background dynamics . however , a reliable estimate of the slope of the intermediate string branch of the spectrum can be obtained by using the leading features of the string phase . first , we shall neglect ( during the string phase ) the effective frequency shift in the perturbation equation , whose role is simply expected to induce an overall rescaling of the amplitude @xcite , as already pointed out in section 4.2 . second , we will assume that the duration of the exit phase " is small compared to the intermediate string phase , and that no dramatic physics take place during the transition to the standard radiation - dominated epoch ( indeed , in some cases where the exit is catalysed by the loop corrections , the process is almost instantaneous ) . this is a big assumption , of course , which could possibly underestimate pre - heating " and re - heating " effects . however , the accuracy of these assumptions can be confirmed with numerical simulations . in summary , we suppose that the evolution of the background consists mainly of three phases : an initial , low energy dilaton - driven branch , the intermediate string phase with a constant hubble parameter ( @xmath180 ) and a linearly growing dilaton ( @xmath181 const ) , and the final radiation phase with decreasing hubble parameter and frozen dilaton . it follows from our assumptions that the spectral graviton distribution can be correctly estimated through the low - energy perturbation equation , even for the high frequency branch of the spectrum . in that case , it is already well known @xcite that the slope of the high - frequency modes , crossing the horizon in the high curvature , stringy regime , and re - entering in the radiation era , is fully determined by the fixed point values @xmath182 . indeed , during such a string phase , the scale factor undergoes the usual de sitter exponential expansion , while the logarithmic evolution of the dilaton , in conformal time , is weighted by the ratio @xmath183 , i.e. @xmath184 const @xcite . by introducing the convenient shifted variable @xmath185 , and referring the spectrum to a fixed point allowing a subsequent ( loop catalysed ) exit , i.e. @xmath186 , one easily finds @xcite @xmath187 where @xmath188 is the limit frequency marking the transition to the high curvature regime . in our model of the background evolution , on the other hand , the allowed fixed points @xmath182 are located in the ( @xmath189 ) plane between the `` branch change line '' , @xmath190 , and the `` einstein bounce '' , @xmath191 , see fig . 1 and @xcite . it follows that , in the context of our approximations , the spectrum of tensor fluctuations , at high frequency , has a slope constrained by @xmath192 in order to check this important analytical result , we analysed numerical solutions for different choices of coefficients of the tree - level @xmath2 corrections . the spectral distribution of relic gravitational waves can be obtained by numerical integration of the perturbation equation eq . ( [ e : psi_full ] ) , in the background eq . ( [ e : eq_rho][e : e_constraint ] ) . the initial conditions are chosen in the low curvature and weakly coupled regime and are thus based on the tree - level solutions , which provide an adequate description of this phase . we then evolve the system until a given mode re - enters the hubble radius , in the late time frw radiation - dominated epoch , where we extract the bogoliubov coefficients by comparing the result of the simulation and the free - oscillating solution eq . ( [ e : psi_radiation ] ) . to highlight the impact of the time - dependence of the frequency shift , we first present the spectral distribution in a case where we do not consider the @xmath2 and loop corrections in the perturbation equation eq . ( [ e : psi_full ] ) . fig . [ f : spectrum ] illustrates the results of such a simulation , for the particular model of background corresponding to @xmath193 , @xmath194 ( all the other @xmath195 ) , @xmath196 , @xmath197 and @xmath198 . we show , in particular , the non - singular evolution of @xmath163 and @xmath199 , and the evolution of the quantity @xmath200 , which enables us to determine if a given mode leaves the hubble radius during the ddi phase or during the intermediate string phase . finally , we present the graviton distribution for such a background , computed from eq . ( [ e : psi_full ] ) , in units @xmath201 , where @xmath202 is a maximal amplified frequency . the choice of coefficients for the tree - level @xmath2 corrections is @xmath203 and @xmath204 ( and the same for @xmath26 and @xmath27 ) , corresponding to the case @xmath205 in fig . [ f : shift ] . the left figure shows a non - singular evolution for the hubble parameter @xmath15 and for @xmath206 , as a function of the number of e - folds , @xmath207 . the middle figure shows the evolution of @xmath208 as function of @xmath209 . the low - energy , dilaton - driven phase takes place approximatively for @xmath210 . after a short transition , this initial period is followed by a string phase with nearly constant hubble parameter and linearly growing dilaton , for @xmath211 . after a successful exit triggered by loop corrections , the background evolution enters the frw radiation - dominated phase at @xmath212 . comoving modes leaving the hubble radius during the string phase ( @xmath213 ) are characterised by a slope smaller than @xmath214 in the spectral distribution , as illustrated in the right figure which shows @xmath215 in units of @xmath201 . the upper dashed line corresponds to the cubic slope typical of the low frequency part of the spectrum , emerging from the dilaton - driven epoch.,title="fig:",width=170,height=136 ] the choice of coefficients for the tree - level @xmath2 corrections is @xmath203 and @xmath204 ( and the same for @xmath26 and @xmath27 ) , corresponding to the case @xmath205 in fig . [ f : shift ] . the left figure shows a non - singular evolution for the hubble parameter @xmath15 and for @xmath206 , as a function of the number of e - folds , @xmath207 . the middle figure shows the evolution of @xmath208 as function of @xmath209 . the low - energy , dilaton - driven phase takes place approximatively for @xmath210 . after a short transition , this initial period is followed by a string phase with nearly constant hubble parameter and linearly growing dilaton , for @xmath211 . after a successful exit triggered by loop corrections , the background evolution enters the frw radiation - dominated phase at @xmath212 . comoving modes leaving the hubble radius during the string phase ( @xmath213 ) are characterised by a slope smaller than @xmath214 in the spectral distribution , as illustrated in the right figure which shows @xmath215 in units of @xmath201 . the upper dashed line corresponds to the cubic slope typical of the low frequency part of the spectrum , emerging from the dilaton - driven epoch.,title="fig:",width=170,height=136 ] the choice of coefficients for the tree - level @xmath2 corrections is @xmath203 and @xmath204 ( and the same for @xmath26 and @xmath27 ) , corresponding to the case @xmath205 in fig . [ f : shift ] . the left figure shows a non - singular evolution for the hubble parameter @xmath15 and for @xmath206 , as a function of the number of e - folds , @xmath207 . the middle figure shows the evolution of @xmath208 as function of @xmath209 . the low - energy , dilaton - driven phase takes place approximatively for @xmath210 . after a short transition , this initial period is followed by a string phase with nearly constant hubble parameter and linearly growing dilaton , for @xmath211 . after a successful exit triggered by loop corrections , the background evolution enters the frw radiation - dominated phase at @xmath212 . comoving modes leaving the hubble radius during the string phase ( @xmath213 ) are characterised by a slope smaller than @xmath214 in the spectral distribution , as illustrated in the right figure which shows @xmath215 in units of @xmath201 . the upper dashed line corresponds to the cubic slope typical of the low frequency part of the spectrum , emerging from the dilaton - driven epoch.,title="fig:",width=170,height=136 ] the effects of the higher - order corrections are expected to arise when the curvature scale becomes large in string units . as a consequence , the low frequency branch of the spectrum is unaffected by such corrections , and remains characterised by a cubic slope . however , modes leaving the hubble radius during the intermediate string phase are strongly affected by the background dynamics . the resulting slope of the spectrum is found to be reduced , but remains confined between @xmath216 and @xmath214 , as expected from eq . ( [ e : slope ] ) . does the inclusion of the time - dependence of the comoving wavenumber lead to any modification ? we stress that when such a shift in the frequency is considered , we have @xmath217 at horizon crossing . as a consequence , the shortest scale to leave the hubble radius is given by @xmath218 , instead of @xmath219 . this is illustrated in fig . [ f : spectrum2 ] , where we compare the results of the numerical integration with and without corrections in the perturbation equation , for identical non - singular background evolutions ( the same model of fig . no modification appears during the low energy epoch , where the frequency shift is negligible . however , higher - order corrections in the perturbation equation , which result in the time - dependent shift , lead to a reduction of the maximal amplified frequency and of the time spent on super - horizon scales . using the same choice of coefficients as in fig . [ f : spectrum ] , we compare the evolution of @xmath208 ( solid curve ) and @xmath220 ( dashed curve ) as a function of @xmath209 . when higher - order corrections are included in the perturbation equation , the high frequency cutoff and the duration of the super - horizon epoch are found to be reduced . the right hand figure is a magnified image of the end point of the string branch of the spectral distribution , which highlights the impact due to a non - trivial frequency shift : the significant new feature is the reduction of the high frequency peak and a further reduction of the slope , which may also become smaller than two , for comoving modes leaving the hubble radius during the `` exit phase''.,title="fig:",width=226,height=188 ] using the same choice of coefficients as in fig . [ f : spectrum ] , we compare the evolution of @xmath208 ( solid curve ) and @xmath220 ( dashed curve ) as a function of @xmath209 . when higher - order corrections are included in the perturbation equation , the high frequency cutoff and the duration of the super - horizon epoch are found to be reduced . the right hand figure is a magnified image of the end point of the string branch of the spectral distribution , which highlights the impact due to a non - trivial frequency shift : the significant new feature is the reduction of the high frequency peak and a further reduction of the slope , which may also become smaller than two , for comoving modes leaving the hubble radius during the `` exit phase''.,title="fig:",width=226,height=188 ] the impact of the time - dependence of the comoving wavenumber @xmath0 is non - trivial for very high - frequency modes , leaving the hubble radius during the graceful exit to the frw - radiation dominated epoch . in that case , the result eq.([e : omega ] ) no longer applies , and the slope can be smaller than two , as shown in fig . however , the impact remains negligible for modes well inside the string branch of the spectrum i.e. for modes crossing the horizon during the asymptotic , fixed point regime . in that case , our analytical approximations for the slope of the spectrum are confirmed by the numerical simulations . in table [ t : slope ] , we present the predicted and measured slopes well inside the string phase , for the three different choices of coefficients illustrated in fig . [ f : shift ] . the agreement is striking . .[t : slope]choices of coefficients for the tree - level @xmath2 corrections with @xmath221 and the remaining coefficient satisfying @xmath222 , according to eq . ( [ e : constraint ] ) . we have compared the predicted slope determined by the fixed point , according to eq . ( [ e : omega ] ) , with the measured slope well inside the string branch of the spectrum , obtained by numerical integration of eq . ( [ e : psi_full ] ) , including all corrections . [ cols="<,^,^,^,^ " , ] it is well known @xcite that the spectrum of relic gravitons , in the context of the pre - big bang scenario , easily evades constraints arising from both the cmb anisotropy at cobe scale @xcite and pulsar timing data @xcite , because of its rapid growth at low frequency . hence , the peak amplitude of the spectrum is only constrained by the primordial nucleosynthesis bound @xcite , as well as by primordial black hole production @xcite . the relic graviton background is in general compatible with these constraints if it is normalised so as to match the string scale at the high - frequency end point of the spectrum . in that case one finds @xcite that the high frequency peak is typically of order @xmath223 , for a maximal amplified frequency @xmath224 $ ] . saturating this high - frequency end point , and using the prediction eq . ( [ e : slope ] ) for the slope of the string branch of the spectrum , the energy density of relic gravitons from a pre - big bang phase could be at most of @xmath225 at @xmath226 $ ] , regardless of the duration of the string phase with constant hubble parameter , i.e. far below the sensitivity of the second ( planned ) generation of interferometric gravity wave detectors @xcite . it seems thus difficult to detect the relic gravitons from a high curvature string phase , associated to a fixed point of the truncated effective action at least within the model of background considered in this paper unless the exit phase is able to affect a sufficiently wide band of the high frequency spectrum . finally , the numerical integration of eq . ( [ e : psi_full ] ) enables us to follow carefully the time evolution of modes , from the ddi epoch up to the frw - radiation dominated era . we have explicitly checked that the amplitude of tensor perturbations is rapidly frozen in on super - horizon scale , regardless of the particular higher - order dynamics of the background during the string phase and the exit epoch . in this paper we have discussed the evolution of tensor perturbations in a class of non - singular cosmological models based on the gravi - dilaton string effective action , expanded up to first order in @xmath2 , and including one - loop and two - loop corrections in the string coupling parameter . the coefficients of the loop expansion have been chosen in such a way as to avoid the curvature singularities in the background , and the quantum instability of metric fluctuations . in addition , we have included an appropriate radiation backreaction to stabilise the dilaton in the final , post - big bang regime . we have been able to obtain , in this way , a class of regular cosmological solutions in which the initial universe , emerging from the string perturbative vacuum , is first trapped in a fixed point of high , constant curvature , and then spontaneously undergoes a smooth transition to the standard frw regime . the fixed point is determined by the @xmath2 corrections , while the unstability and the decay of the high curvature regime is triggered by the quantum loop corrections . we have determined , in this class of backgrounds , the linearised equation governing the evolution of tensor metric fluctuations , taking into account all the @xmath2 and loop corrections contributing to the background regularisation . we have discussed , on this ground , the amplification of tensor perturbations normalised to the quantum fluctuations of the vacuum , and we have estimated their final spectral energy distribution through analytical and numerical methods . our results confirm previous expectations , that the low frequency modes , crossing the horizon in the low - curvature regime , are unaffected by higher - order corrections . also , we have explicitly checked ( through numerical simulations ) that the so - called string branch " of the spectrum , associated with the asymptotic , fixed point regime , can be reliably estimated by means of the low - energy perturbation equation , that its slope is determined by the asymptotic constant values of @xmath163 and @xmath199 , and that it is flatter than at low frequency . however , when the background regularisation crosses a fixed point , and when the fixed point is determined by the @xmath2 expansion truncated to first order ( like in the class of models of this paper ) , it turns out that the slope of the string branch is at least quadratic , and thus probably too steep to be compatible with future observations by planned advanced detectors . in the transition regime dominated by the quantum loops , on the other hand , the influence of the higher - order corrections on tensor perturbations is more dramatic , and the corresponding slope may be much flatter than the one of the preceding string phase . in summary , the results of this paper agree with the general expectation that the shape of the spectrum of the relic graviton background , obtained in the context of the pre - big bang scenario , is strongly model - dependent . however , it turns out that the formulation of a complete and consistent regular scenario , associated to a visible " graviton spectrum , may be more difficult than previously expected at least within a truncated perturbative approach , like the one adopted in this paper . cc is supported by the swiss nsf , grant no . 83eu-054774 and ors/1999041014 . ejc was partially supported by pparc . we thank r. durrer , m. giovannini , s. leach , l. mendes , c. ungarelli , g. veneziano , f. vernizzi and d. wands for helpful and stimulating discussions . m. gasperini and g. veneziano . pre - big bang in string cosmology . , 1:317339 , 1993 , hep - th/9211021 . an updated collection of papers and references on the pre - big bang scenario is available at the url , http://www.to.infn.it/@xmath227gasperin/. r.r . metsaev and a.a . order @xmath228 ( two - loop ) equivalence of the string equations of motion and the sigma model weyl invariance conditions : dependence on the dilaton and the antisymmetric tensor . , b293:385 , 1987 . m. gasperini . elementary introduction to pre - big bang cosmology and to the relic graviton background . 1999 , hep - th/9907067 . proc . of the second sigrav school on gravitational waves ( centre a. volta , como , april 1999 ) , eds . i. ciufolini et al . ( iop publishing , bristol , 2001 ) , p. 280 .

a constructive version of newton puiseux theorem for computing the puiseux expansions of algebraic curves is presented . the proof is based on a classical proof by abhyankar . algebraic numbers are evaluated dynamically ; hence the base field need not be algebraically closed and a factorization algorithm of polynomials over the base field is not needed . the extensions obtained are a type of regular algebras over the base field and the expansions are given as formal power series over these algebras .

1304.6770

newton puiseux theorem states that , for an algebraically closed field @xmath0 of zero characteristic , given a polynomial @xmath1[y]$ ] there exist a positive integer @xmath2 and a factorization @xmath3 where each @xmath4[y]$ ] . these roots @xmath5 are called the _ puiseux expansions _ of @xmath6 . the theorem was first proved by newton @xcite with the use of newton polygon . later , puiseux @xcite gave an analytic proof . it is usually stated as : _ the field of fractional power series , i.e. the field @xmath7 , is algebraically closed _ abhyankar @xcite presents another proof of this result , the `` shreedharacharya s proof of newton s theorem '' . this proof is not constructive as it stands . indeed it assumes decidable equality on the ring @xmath8 $ ] of power series over a field , but given two arbitrary power series we can not decide whether they are equal in _ finite _ number of steps . we explain in this paper how to modify his argument by adding a separability assumption to provide a constructive proof of the result : the field of fractional power series is _ separably _ algebraically closed . in particular , the termination of newton puiseux algorithm is justified constructively in this case . this termination is justified by a non constructive reasoning in most references @xcite , with the exception of @xcite ( for an introduction to constructive algebra , see @xcite ) . following that , we show that the field of fractional power series algebraic over @xmath9 is algebraically closed . another contribution of this paper is to analyze in a constructive framework what happens if the field @xmath0 is not supposed to be algebraically closed . the difference with @xcite , which provides also such an analysis , is that we do not assume the irreducibility of polynomials to be decidable . this is achieved through the method of _ dynamic evaluation _ @xcite , which replaces factorization by gcd computations . the reference @xcite provides a proof theoretic analysis of this method . + with dynamic evaluation we obtain algebras , _ triangular separable algebras _ , as separable extensions of the base field and the puiseux expansions are given over these algebras . theorem [ dynamicnewton - extended ] shows that the extensions produced by the algorithm are minimal in the sense that if @xmath10 is one such extension and @xmath11 is any other algebra over the base field such that @xmath12 factors linearly over @xmath13 $ ] for some positive integer @xmath14 , then @xmath11 splits @xmath10 which in case @xmath11 and @xmath10 were fields would be equivalent to saying that @xmath11 contains the normal closure of @xmath10 . but this then shows that @xmath10 splits itself , which in case @xmath10 is a field is equivalent to saying that @xmath10 is a normal extension ( corollary [ spliteachother ] ) . theorem [ dynamicnormal2 ] will then show that any two triangular separable algebras @xmath11 and @xmath15 that split each other are in fact powers of a some triangular separable algebra , i.e. @xmath16 and @xmath17 for some triangular separable algebra @xmath10 and positive integers @xmath18 . this algorithm gives less information than duval s rational puiseux expansion algorithm @xcite since we can easily obtain the classical puiseux expansions of a polynomial from the rational ones ( rational puiseux expansions describe the roots of the polynomial by pairs of power series , i.e. a parametrization , with rational coefficients ) . in @xcite the rational expansions of a polynomial @xmath19 $ ] are given as long as @xmath12 is absolutely irreducible , i.e. irreducible in @xmath20 $ ] , where @xmath21 is the algebraic closure of @xmath0 . it would be interesting to also justify duval s algorithm in a constructive framework . we recall that a ( discrete ) field is defined to be a non trivial ring in which any element is @xmath22 or invertible . for a ring @xmath10 , the formal power series ring @xmath23 $ ] is the set of sequences @xmath24 , with @xmath25 @xcite . an _ apartness _ relation @xmath26 on a set is a symmetric relation satisfying @xmath27 and @xmath28 . an apartness is tight if it satisfies @xmath29 . in addition to the ring identities , a ring with apartness satisfies @xmath30 , @xmath31 and @xmath32 , see @xcite . next we define the apartness relation on power series as in . [ apartd ] let @xmath10 be a ring with apartness . for @xmath33 $ ] we define @xmath34 if @xmath35 . the relation @xmath26 as defined above is an apartness relation and makes @xmath23 $ ] into a ring with apartness @xcite . this definition of @xmath26 applies to the ring of polynomials @xmath36 \subset r[[x]]$ ] . we note that , if @xmath0 is a discrete field then for @xmath37 $ ] we have @xmath38 iff @xmath39 is invertible for some @xmath40 . for @xmath41[y]$ ] , we have @xmath42 iff @xmath43 is invertible for some @xmath40 and @xmath44 . let @xmath10 be a commutative ring with apartness . then @xmath10 is an _ integral domain _ if it satisfies @xmath45 for all @xmath46 . a _ heyting _ field is an integral domain satisfying @xmath47 . the heyting field of fractions of @xmath10 is the heyting field obtained by inverting the elements @xmath48 in @xmath10 and taking the quotient by the appropriate equivalence relation , see . for @xmath49 and @xmath50 in r we have @xmath51 iff @xmath52 . for a discrete field @xmath0 , an element @xmath38 in @xmath8 $ ] can be written as @xmath53 with @xmath54 and @xmath55 . it follows that the ring @xmath8 $ ] is an integral domain . if @xmath55 we have that @xmath56 is invertible in @xmath8 $ ] . we denote by @xmath57 , the heyting field of fractions of @xmath8 $ ] , we also call it the heyting field of laurent series over @xmath0 . thus an element apart from @xmath22 in @xmath57 can be written as @xmath58 with @xmath55 and @xmath59 , i.e. as a series where finitely many terms have negative exponents . unless otherwise qualified , in what follows , a field will always denote a discrete field . let @xmath10 be a ring . a polynomial @xmath60 $ ] is separable if there exist @xmath61 $ ] such that @xmath62 , where @xmath63 $ ] is the derivative of @xmath64 . [ sepdivisorsep ] let @xmath10 be a ring and @xmath60 $ ] separable . if @xmath65 then both @xmath66 and @xmath67 are separable . let @xmath62 for @xmath68 $ ] . then @xmath69 , thus @xmath67 is separable . similarly for @xmath66 . [ separablevarchangepersist ] let @xmath10 be a ring . if @xmath70 $ ] is separable and @xmath71 a unit then @xmath72 $ ] is separable . the following result is usually proved with the assumption of existence of a decomposition into irreducible factors . we give a proof without this assumption . it works over a field of any characteristic . [ squarefree ] let @xmath66 be a monic polynomial in @xmath73 $ ] where @xmath0 is a field . if @xmath74 is the derivative of @xmath66 and @xmath67 monic is the gcd of @xmath66 and @xmath74 then writing @xmath75 we have that @xmath76 is separable . we call @xmath76 the separable associate of @xmath66 . let @xmath49 be the gcd of @xmath76 and @xmath77 . we have @xmath78 . let @xmath79 be the gcd of @xmath49 and @xmath80 . we have @xmath81 and @xmath82 , with @xmath83 and @xmath84 coprime . the polynomial @xmath49 divides @xmath85 and hence that @xmath81 divides @xmath86 . it follows that @xmath83 divides @xmath87 and since @xmath83 and @xmath84 are coprime , that @xmath83 divides @xmath88 . also , if @xmath89 divides @xmath64 then @xmath90 and @xmath91 . hence @xmath92 divides @xmath93 . since @xmath83 divides @xmath88 , this implies that @xmath94 divides @xmath95 . so @xmath96 divides @xmath97 . since @xmath49 divides @xmath66 and @xmath74 , @xmath49 divides @xmath67 . we show that @xmath89 divides @xmath67 for all @xmath98 by induction on @xmath98 . if @xmath89 divides @xmath67 we have just seen that @xmath96 divides @xmath99 . also @xmath96 divides @xmath100 since @xmath49 divides @xmath77 . so @xmath96 divides @xmath101 . on the other hand , @xmath96 divides @xmath102 . so @xmath96 divides @xmath67 which is the gcd of @xmath66 and @xmath74 . this implies that @xmath49 is a unit . if @xmath6 is in @xmath23[y]$ ] we let @xmath103 be the derivative of @xmath6 with respect to @xmath104 . [ apartness ] let @xmath0 be a field and let @xmath105[y]$ ] be separable over @xmath57 , then @xmath106 since @xmath6 is separable over @xmath57 we have @xmath107 for @xmath108[y]$ ] and @xmath109 $ ] . from this we get that @xmath110 is equal to the constant term on the left hand side , i.e. @xmath111 . thus @xmath106 . one key of abhyankar s proof is hensel s lemma . we formulate a little more general version than the one in @xcite by dropping the assumption that the base ring is a field . [ hensel ] let @xmath10 be a ring and @xmath112 be a monic polynomial in @xmath23[y]$ ] of degree @xmath113 . given monic @xmath114 $ ] of degrees @xmath115 respectively and @xmath116 $ ] such that @xmath117 , @xmath118 and @xmath119 ; we can find @xmath120[y]$ ] of degrees @xmath121 respectively , such that @xmath122 and @xmath123 , @xmath124 . the proof is almost the same as abhyankar s @xcite , we present it here for completeness . + since @xmath23[y ] \subsetneq r[y][[x]]$ ] , we can rewrite @xmath12 as a power series in @xmath125 with coefficients in @xmath126 $ ] . let @xmath127 , with @xmath128 $ ] . now we want to find @xmath129[[x]]$ ] such that @xmath130 . if we let @xmath131 and @xmath132 , then for each @xmath133 we need to find @xmath134 for @xmath135 such that @xmath136 . we also need @xmath137 and @xmath138 for @xmath139 . + we find such @xmath140 by induction on @xmath133 . we have that @xmath141 . assume that for some @xmath142 we have found all @xmath140 with @xmath143 and @xmath144 for @xmath145 and @xmath146 . now we need to find @xmath147 such that + @xmath148 . we let @xmath149 , and we can see that @xmath150 . we are given that @xmath151 . multiplying by @xmath152 we get @xmath153 . by euclidean division we can write @xmath154 for some @xmath155 with @xmath156 . thus we write @xmath157 . we can see that @xmath158 since @xmath159 . since @xmath160 is monic of degree @xmath161 , @xmath162 . we take @xmath163 . + now , we can write @xmath164 as monic polynomials in @xmath104 withe coefficients in @xmath23 $ ] , with degrees @xmath121 respectively . it should be noted that the uniqueness of the factors @xmath165 and @xmath166 proven in @xcite may not necessarily hold when @xmath10 is not an integral domain . if @xmath167 is an element of @xmath23 $ ] we write @xmath168 to mean that @xmath169 for @xmath170 and @xmath171 to mean furthermore that @xmath172 is invertible . [ keylemma ] let @xmath0 be an algebraically closed field of characteristic zero . + let @xmath173[y]$ ] be a monic non - constant polynomial of degree @xmath174 separable over @xmath57 . then there exist @xmath175 and a proper factorization @xmath176 with @xmath165 and @xmath166 in @xmath177[y].$ ] we can assume w.l.o.g . that @xmath178 . this is shreedharacharya s trick @xcite ( a simple change of variable @xmath179 ) . the simple case is if we have @xmath180 for some @xmath181 . in this case @xmath182 $ ] and @xmath183 . thus @xmath184 . for any root @xmath185 of @xmath186 we have then a proper decomposition @xmath187 with @xmath188 and @xmath166 coprime , and we can use hensel s lemma [ hensel ] to conclude ( in this case we can take @xmath189 ) . in general , we know by lemma [ apartness ] that for @xmath190 or @xmath191 we have @xmath192 is apart from @xmath22 . we then have @xmath193 invertible for some @xmath194 . we can then find @xmath64 and @xmath2 , @xmath195 , such that @xmath196 is invertible and @xmath197 whenever @xmath198 ( see explanation below ) . we can then write @xmath199 with @xmath200 . as in the simple case , we have a proper decomposition @xmath201 with @xmath202 monic of degree @xmath203 in @xmath204 and @xmath205 monic of degree @xmath133 in @xmath204 , with @xmath206 . we then take @xmath207 and @xmath208 we note that since the polynomial is of finite @xmath104 degree the search for @xmath2 and @xmath64 is finite . for example if the polynomial is of @xmath104 degree @xmath209 ( see figure [ searchfinite ] ) and if @xmath210 and @xmath211 we need only search the finite number of pairs to the left of the dotted line . [ searchfinite ] and @xmath64 , lemma [ keylemma].,title="fig:",scaledwidth=100.0% ] [ abhyankarproof ] let @xmath0 be an algebraically closed field of characteristic zero . + let @xmath173[y]$ ] be a monic non - constant polynomial separable over @xmath57 . then there exist a positive integer @xmath2 and factorization @xmath212\ ] ] if @xmath12 is separable over @xmath57 then @xmath213 for some positive integer @xmath2 is separable over @xmath214 . the proof follows from lemma [ sepdivisorsep ] and lemma [ keylemma ] by induction . [ sepalgclose ] let @xmath0 be an algebraically closed field of characteristic zero . the heyting field of fractional power series over @xmath0 is separably algebraically closed . let @xmath215 $ ] be a monic separable polynomial of degree @xmath113 . let @xmath216 be the product of the denominators of the coefficients of @xmath6 . then we can write @xmath217 for @xmath218[z]$ ] . by lemma [ separablevarchangepersist ] we get that @xmath6 , hence @xmath165 , is separable in @xmath204 over @xmath57 . by theorem [ abhyankarproof ] , @xmath219 factors linearly over @xmath177 $ ] for some positive integer @xmath2 . consequently we get that @xmath220 factors linearly over @xmath214 . in the following we show that the elements in @xmath221 algebraic over @xmath9 form a discrete algebraically closed field . [ rootorderbound ] let @xmath0 be a field and @xmath222 $ ] be a non - constant monic polynomial such that @xmath223 . if @xmath224 is a root of @xmath225 , then @xmath226 for some positive integer @xmath79 . we can find @xmath227 $ ] such that @xmath228[y]$ ] with @xmath229 . let @xmath230 . if @xmath231 then so is @xmath232 for @xmath233 . but we know that in @xmath234 there is a non - zero term with @xmath235degree @xmath79 . thus @xmath236 ; consequently @xmath237 note that if @xmath238 are algebraic over @xmath9 then @xmath239 and @xmath240 are algebraic over @xmath9 . [ discreteness ] let @xmath0 be a field . the set of elements in @xmath221 algebraic over @xmath9 is a discrete set ; more precisely @xmath26 is decidable on this set . it suffices to show that for an element @xmath110 in this set @xmath241 is decidable . let @xmath242 $ ] be a monic non - constant polynomial . let @xmath224 be a root of @xmath243 . if @xmath244 then @xmath245 . otherwise , @xmath6 can be written as @xmath246 with @xmath247 and @xmath248 . by lemma [ rootorderbound ] we can find @xmath79 such that any element in @xmath214 that is a root of @xmath249 has an order less than or equal to @xmath79 . thus @xmath241 if an only if @xmath226 . if @xmath250 is algebraic over @xmath9 then @xmath251 is algebraic over @xmath9 . thus the set of elements in @xmath221 algebraic over @xmath9 form a field @xmath252 . this field is in fact algebraically closed _ in _ @xmath221 . since for an algebraically closed field @xmath0 we have shown @xmath221 to be only _ separably _ algebraically closed , we need a stronger argument to show that @xmath253 is algebraically closed . [ algebraicallyclosed ] for an algebraically closed field @xmath0 of characteristic zero , the field @xmath253 is algebraically closed . let @xmath254 $ ] be a monic non - constant polynomial of degree @xmath98 . by lemma [ discreteness ] @xmath253 is a discrete field . by lemma [ squarefree ] we can decompose @xmath6 as @xmath255 with @xmath256 $ ] a non - constant monic separable polynomial . by corollary [ sepalgclose ] , @xmath166 has a root @xmath257 in @xmath221 . since @xmath253 is algebraically closed in @xmath221 we have that @xmath258 . we can draw similar conclusions in the case of real closed fields . [ seprealclose ] let @xmath10 be a real closed field . then 1 . for any @xmath259 we can find @xmath260 such that @xmath261 or @xmath262 . a separable monic polynomial of odd degree in @xmath263 $ ] has a root in @xmath264 . since @xmath10 is real closed , the first statement follows from the fact an element @xmath265 $ ] with @xmath266 has a square root in @xmath23 $ ] . let @xmath267[y]$ ] be a monic polynomial of odd degree @xmath113 separable over @xmath268 . we can assume w.l.o.g . that @xmath269 . since @xmath6 is separable , i.e. @xmath270 for some @xmath271 $ ] , then by a similar construction to that in lemma [ keylemma ] we can write @xmath272 for @xmath273[z]$ ] such that @xmath274 for all @xmath275 . since @xmath10 is real closed and @xmath276 has odd degree , @xmath276 has a root @xmath14 in @xmath10 . we can find proper decomposition into coprime factors @xmath277 . by hensel s lemma[hensel ] , we lift those factors to factors of @xmath278 in @xmath279[z]$ ] thus we can write @xmath130 for monic non - constant @xmath280[y]$ ] . by lemma [ sepdivisorsep ] both @xmath165 and @xmath166 are separable . either @xmath165 or @xmath166 has odd degree . assuming @xmath165 has odd degree greater than @xmath281 , we can further factor @xmath165 into non - constant factors . the statement follows by induction . let @xmath10 be a real closed field . by lemma [ discreteness ] we see that @xmath282 is discrete . a non - zero element in @xmath283 can be written @xmath284 for @xmath285 with @xmath55 . then @xmath286 is positive iff its initial coefficient @xmath287 is positive @xcite . we can then see that this makes @xmath282 an ordered field . [ realclosed ] for a real closed field @xmath10 , the field @xmath282 is real closed . let @xmath283 . since @xmath282 is discrete , by lemma [ seprealclose ] we can find @xmath288 such that @xmath261 or @xmath262 . + let @xmath289 $ ] be a monic polynomial of odd degree @xmath98 . applying lemma [ squarefree ] several times , by induction we have @xmath290 with @xmath291 $ ] separable non - constant monic polynomial . for some @xmath292 we have @xmath293 of odd degree . by lemma [ seprealclose ] , @xmath293 has a root in @xmath282 . thus @xmath6 has a root in @xmath282 . the goal of this section is to give a version of theorem [ abhyankarproof ] over a field @xmath0 of characteristic @xmath22 , not necessarily algebraically closed . a commutative ring @xmath10 is _ ( von neumann ) regular _ if for every element @xmath294 there exist @xmath295 such that @xmath296 and @xmath297 . this element @xmath185 is called the quasi - inverse of @xmath49 . a ring is regular iff it is zero - dimensional and reduced . it is also equivalent to the fact that any principal ideal ( and hence any finitely generated ideal ) is generated by an idempotent . if @xmath49 is an element in @xmath10 and @xmath298 then the element @xmath299 is an idempotent such that @xmath300 and @xmath10 is isomorphic to @xmath301 with @xmath302 and @xmath303 . furthermore @xmath49 is @xmath22 on the component @xmath304 and invertible on the component @xmath305 . we define strict bzout rings as in . a ring @xmath10 is a ( strict ) bzout ring if for all @xmath306 we can find @xmath307 such that @xmath308 , @xmath309 and @xmath310 . if @xmath10 is a regular ring then @xmath36 $ ] is a strict bzout ring ( and the converse is true @xcite ) . intuitively we can compute the gcd as if @xmath10 was a field , but we may need to split @xmath10 when deciding if an element is invertible or @xmath22 . using this , we see that given @xmath311 in @xmath36 $ ] we can find a decomposition @xmath312 of @xmath10 and for each @xmath292 we have @xmath313 in @xmath314 $ ] such that @xmath308 , @xmath309 and @xmath310 with @xmath67 monic . the degree of @xmath67 may depend on @xmath292 . [ extension ] if @xmath10 is regular and @xmath64 in @xmath36 $ ] is a separable polynomial then @xmath315=r[x]/{\langle p\rangle}$ ] is regular . if @xmath316 is an element of @xmath315 $ ] with @xmath133 in @xmath36 $ ] we compute the gcd @xmath67 of @xmath64 and @xmath133 . if @xmath317 , we can find @xmath318 and @xmath319 in @xmath36 $ ] such that @xmath320 since @xmath64 is separable . we then have @xmath321 and @xmath322 . it follows that @xmath323 is idempotent and we have @xmath324 . a _ triangular separable @xmath0algebra _ @xmath325,~p_1(a_1 ) = 0,~p_2(a_1,a_2 ) = 0,\dots\ ] ] is a sequence of separable extension starting from a field @xmath0 , with @xmath326 in @xmath73 $ ] , @xmath327 in @xmath328[x]$ ] , @xmath329 all monic and separable polynomials . a triangular separable algebra is thought of as an approximation of the algebraic closure of @xmath0 , and is determined by a list of polynomials @xmath330 ( this is related to the way @xcite avoids the algebraic closure , by adding only constants as needed , with the difference that we do nt assume an irreducibility test . ) it follows from lemma [ extension ] that each triangular separable algebra defines a regular algebra @xmath331 $ ] . in this case however , the idempotent elements have a simpler direct description . if we have a decomposition @xmath332 with @xmath333 in @xmath334 $ ] then since @xmath335 is separable , we have a relation @xmath336 and @xmath337 are then idempotent element . we then have a decomposition of @xmath10 in two triangular separable algebras @xmath338 and @xmath339 . if we iterate this process we obtain the notion of _ decomposition _ of a triangular separable algebra @xmath10 in finitely many triangular algebra @xmath340 . this decomposition stops when all polynomials @xmath341 are irreducible , i.e. when @xmath10 is a field . for a triangular separable algebra @xmath10 and an ideal @xmath342 of @xmath10 , if @xmath343 is a triangular separable algebra then we describe @xmath343 as being a _ refinement _ of @xmath10 . thus a refinement of @xmath344,p_1, ... ,p_n$ ] is of the form @xmath345,q_1, ... ,q_n$ ] with @xmath346 . the following is a corollary of lemma [ squarefree ] . [ squarefreemonic ] let @xmath66 be a monic polynomial in @xmath36 $ ] where @xmath10 is a triangular separable @xmath0algebra . if @xmath74 is the derivative of @xmath66 then there exist a decomposition @xmath347 and on each @xmath348 we can find polynomials @xmath349 in @xmath314 $ ] such that @xmath350 , @xmath351 and @xmath352 with @xmath76 monic and separable . [ sum1decomp ] let @xmath10 be a regular ring and let @xmath353 such that @xmath354 . then we can find a decomposition @xmath355 such that for each @xmath348 we have @xmath356 a unit in @xmath348 for some @xmath357 . we have a decomposition @xmath358 with @xmath234 unit in @xmath11 and zero in @xmath15 . we have @xmath359 in @xmath15 . the statement follows by induction . [ apartcontinvert ] let @xmath10 be a triangular separable algebra over a field @xmath0 of characteristic @xmath22 . let @xmath360[y]$ ] be a monic polynomial such that @xmath361 for some @xmath362[y]$ ] and @xmath241 in @xmath8 $ ] . then we can find a decomposition @xmath363 of @xmath10 such that in each @xmath348 we have @xmath364 a unit for some @xmath2 and @xmath365 or @xmath366 . since @xmath367 $ ] we have @xmath368 a unit for some @xmath194 . since @xmath361 , we have @xmath369 with @xmath370 and @xmath371 . then we have @xmath372 . by lemma [ sum1decomp ] we have a decomposition @xmath363 of @xmath10 such that in @xmath348 we have @xmath364 is a unit for some @xmath2 and @xmath373 . lemma [ keylemma ] becomes in this way . [ keylemmadyn ] let @xmath10 be a triangular separable algebra over a field @xmath0 of characteristic @xmath22 . let @xmath374[y]$ ] be a monic non - constant polynomial of degree @xmath174 such that @xmath361 for some @xmath362[y]$ ] and @xmath241 in @xmath8 $ ] . there exists then a decomposition @xmath375 of @xmath10 and for each @xmath292 there exist @xmath175 and a proper factorization @xmath176 with @xmath165 and @xmath166 in @xmath376[y]$ ] where @xmath377 $ ] is a separable extension of @xmath348 . by lemma [ apartcontinvert ] we have a decomposition @xmath378 of @xmath10 such that in each @xmath379 we have @xmath364 a unit for some @xmath2 and @xmath365 or @xmath366 . the rest of the proof proceeds as the proof of lemma [ keylemma ] , assuming w.l.o.g . we then find a decomposition of each @xmath379 ; thus a decomposition @xmath375 of @xmath10 and for each @xmath203 we can then find @xmath2 and @xmath64 such that @xmath196 is invertible and @xmath197 whenever @xmath198 in @xmath380 . we can then write @xmath199 with @xmath381 a unit . we then find a further decomposition @xmath382 of @xmath380 and for each @xmath133 a number @xmath161 and a separable extension @xmath383 $ ] of @xmath384 such that @xmath385 with @xmath386 invertible . using hensel s lemma [ hensel ] , we can lift this to a proper decomposition @xmath201 with @xmath202 monic of degree @xmath387 and @xmath205 monic of degree @xmath318 . we take @xmath388 and @xmath389 we can then state the following version of newton puiseux algorithm . [ abhyankarproofdyn ] let @xmath0 be a field of characteristic @xmath22 . let @xmath390 in @xmath8[y]$ ] be a monic non - constant polynomial separable over @xmath57 . there exists then a triangular separable algebra @xmath10 over @xmath0 and @xmath175 and a factorization @xmath391\ ] ] the algorithm for computing this factorization proceeds by induction on @xmath98 , using lemma [ keylemmadyn ] . more precisely the algorithm proceeds as follows . at a given point , we have computed 1 . a triangular separable extension @xmath10 of @xmath0 2 . a number @xmath2 and a partial decomposition @xmath392 with all @xmath393[y]$ ] monic in @xmath104 . the algorithm stops if all @xmath293 are of degree @xmath281 in @xmath104 . otherwise , we apply lemma [ keylemmadyn ] to the first polynomial @xmath394 of degree @xmath395 in @xmath104 to compute a decomposition of @xmath10 and for each algebra @xmath396 in this decomposition a separable extension @xmath397 $ ] , a positive integer @xmath64 and a proper decomposition @xmath398 . we select then one algebra , and we proceed with the decomposition @xmath399 the previous algorithm is not deterministic when selecting an algebra in a decomposition . the goal of this section is to compare two possible triangular separable algebras that can be obtained by this algorithm . we are going to show that they are both powers of a common triangular algebra . in the following we refer to the elementary symmetric polynomials in @xmath98 variables by @xmath400 taking @xmath401 . [ symmrootszeros ] let @xmath10 be a reduced ring . given @xmath353 , if @xmath402 for @xmath403 then @xmath404 . we have @xmath405 . hence , @xmath406 for @xmath407 and since @xmath10 is reduced , @xmath408 . [ orderlemma ] let @xmath10 be a reduced ring . given @xmath409 $ ] such that for some positive rational number @xmath79 we have @xmath410 for @xmath407 . then @xmath411 for @xmath407 . let @xmath412 . we show that @xmath197 if @xmath413 . assume that we have @xmath197 for @xmath414 . we show then @xmath415 for @xmath416 . the coefficient of @xmath417 in @xmath418 is @xmath419 . since @xmath420 we get that @xmath421 and hence by lemma [ symmrootszeros ] we get that @xmath415 for @xmath416 . [ splitlemma ] for a ring @xmath10 and a reduced extension @xmath422 , let @xmath423 be an element of @xmath23[y]$ ] such that @xmath424 with @xmath425 in @xmath279[z]$ ] for some @xmath426 . if @xmath427 factors linearly over @xmath428 $ ] for some @xmath175 then @xmath429 factors linearly over @xmath11 . we have @xmath430 , @xmath431 $ ] + and hence we have @xmath432 , @xmath433 $ ] and @xmath434 since @xmath435 is in @xmath279[z]$ ] we have @xmath436 . + since @xmath437 , lemma [ orderlemma ] shows that @xmath438 for @xmath439 . hence @xmath440 is in @xmath441 $ ] and since @xmath442 , we have that @xmath429 factors linearly over @xmath11 , of roots @xmath443 . [ splitdef ] let @xmath444,p_1, ... ,p_n$ ] be a triangular separable algebra with @xmath445 of degree @xmath446 and @xmath11 an algebra over @xmath0 . then @xmath11 splits @xmath10 if there exist a family of elements @xmath447 such that @xmath448 for @xmath449 we can view the previous definition as that of a tree of homomorphisms from the subalgebras of @xmath10 to @xmath11 . at the root we have the identity homomorphism from @xmath0 to @xmath11 under which @xmath326 factors linearly , i.e. @xmath450 . from this we obtain @xmath451 homomorphisms @xmath452 from @xmath328 $ ] to @xmath11 each taking @xmath453 to a different @xmath454 . if @xmath327 factors linearly under say @xmath455 , i.e. @xmath456 then we obtain @xmath84 different ( since @xmath327 is separable ) homomorphisms @xmath457 from @xmath458 $ ] to @xmath11 . similarly we obtain @xmath84 different homomorphisms from @xmath458 $ ] to @xmath11 by extending @xmath459 , thus having @xmath460 homomorphism in total . continuing in this fashion we obtain the @xmath2 different homomorphisms of the family @xmath461 . we note that if an algebra @xmath11 over @xmath0 splits a triangular separable algebra @xmath10 over @xmath0 then @xmath462}$ ] . if @xmath11 is a field then the converse is also true as the following lemma shows . [ splitfield ] let @xmath463 be a field and @xmath444,p_1, ... ,p_n$ ] a triangular separable algebra . then @xmath464}$ ] only if @xmath465 splits @xmath10 . let @xmath466 , @xmath467 = m = \prod_{i=1}^n m_i$ ] and let @xmath468}$ ] . then there exist a system of orthogonal idempotents if @xmath469 and @xmath470 . ] @xmath471 such that @xmath472 . let @xmath473 be the image of @xmath474 in @xmath475 . then we have @xmath476 since otherwise we will have the ideals @xmath477 for some @xmath469 . since @xmath326 is separable there are up to @xmath451 different images @xmath478 of @xmath453 . thus the size of the set @xmath479 is equal to @xmath451 only if @xmath326 factors linearly over @xmath465 . similarly , for each different image @xmath480 of @xmath453 there are up to @xmath84 possible images of @xmath481 in @xmath465 since the polynomial @xmath482 is separable . thus the size of the set @xmath483 is equal @xmath460 only if @xmath326 factors linearly over @xmath465 and for each root @xmath480 of @xmath326 the polynomial @xmath482 factors linearly over @xmath465 . continuing in this fashion we find that the size of the set @xmath484 is equal to @xmath485 only if @xmath465 splits @xmath10 . [ decompsplitdiv ] let @xmath11 be a triangular separable algebra over a field @xmath0 and let @xmath64 be a monic non - constant polynomial of degree @xmath2 in @xmath486 $ ] such that @xmath487 with @xmath488 . if @xmath67 is a monic non - constant polynomial of degree @xmath98 such that @xmath489 then we have a decomposition @xmath490 such that for any @xmath491 in the product @xmath492 with @xmath493 the image in @xmath491 of some @xmath494 . let @xmath495 for @xmath496 . let @xmath497 . then @xmath498 . we can find a decomposition of @xmath11 into triangular separable algebras @xmath499 such that @xmath500 in @xmath501 and @xmath502 is a unit in @xmath503 in which case @xmath504 in @xmath505 . by induction we can find a decomposition of @xmath11 into a product of triangular separable algebras @xmath506 such that @xmath67 factors linearly over @xmath348 . from definition [ splitdef ] it is obvious that if an algebra @xmath11 splits a triangular separable algebra @xmath10 then @xmath507 splits @xmath10 for any ideal @xmath342 of @xmath11 . [ splitpersist ] let @xmath11 and @xmath10 be triangular separable algebras over @xmath0 such that @xmath11 splits @xmath10 . let @xmath15 be a refinement of @xmath10 . then we can find a decomposition @xmath508 into a product of triangular separable algebras such that @xmath379 splits @xmath15 for @xmath509 . let @xmath510,p_1, ... . then @xmath511 , g_1, ... ,g_n$ ] where @xmath512 for @xmath513 . let @xmath514 and @xmath515 for @xmath516 . since @xmath11 splits @xmath10 we have a family of elements @xmath447 satisfying the condition of definition [ splitdef ] . we have @xmath517 . by lemma [ decompsplitdiv ] we decompose @xmath11 into the product @xmath518 such that for any given @xmath519 in the product we have @xmath520 and @xmath521 with @xmath522 for @xmath523 . since each @xmath524 is an image of some @xmath525 and @xmath526 factors linearly over @xmath11 we have that @xmath527 factors linearly over @xmath519 but then @xmath528 divides @xmath527 and thus by lemma [ decompsplitdiv ] we can decompose @xmath519 into the product @xmath529 such that for a given @xmath530 in the product we have @xmath531 and @xmath532 . by induction on the @xmath451 values of @xmath524 we can find a decomposition @xmath533 such that in each @xmath534 we have @xmath535 and @xmath536 for @xmath537 . continuing in this fashion we can find a decomposition of @xmath11 such that each algebra in the decomposition splits @xmath15 . [ decompsplitsplit ] let @xmath11 and @xmath15 be triangular separable algebras such that @xmath538 and each @xmath379 splits @xmath15 . then @xmath11 splits @xmath15 . let @xmath539,g_1, ... ,g_n$ ] with @xmath540 . then we have a family of elements @xmath541 in @xmath379 satisfying the conditions of definition [ splitdef ] . we claim that the family @xmath542 of @xmath11 elements satisfy the conditions of definition [ splitdef ] . since we have a factorization @xmath543 over @xmath379 , we have a factorization @xmath544 over @xmath11 . since for @xmath545 we have a factorization @xmath546 of in @xmath379 , we have a factorization @xmath547 . continuing in this fashion we verify that the family @xmath461 satisfy the requirements of definition [ splitdef ] . [ splitalgsplitrefinement ] let @xmath11 and @xmath15 be triangular separable algebras such that @xmath11 splits @xmath15 . then @xmath11 splits any refinement of @xmath15 . lemmas [ splitlemma ] , [ decompsplitsplit ] and corollary [ splitalgsplitrefinement ] allow us to extend lemma [ keylemmadyn ] as follows . [ keylemmadynext ] let @xmath444,p_1, ... ,p_n$ ] be a triangular separable algebra with @xmath466 . let @xmath548[y]$ ] be a monic non - constant polynomial of degree @xmath174 such that @xmath361 for some @xmath362[y]$ ] , @xmath549 $ ] with @xmath241 . there exists then a decomposition @xmath375 of @xmath10 and for each @xmath292 there exist @xmath175 and a proper factorization @xmath176 with @xmath165 and @xmath166 in @xmath376[y]$ ] where @xmath550,q$ ] is a separable extension of @xmath348 . + moreover , let @xmath11 be a triangular separable algebra such that @xmath11 splits @xmath10 and let @xmath551 be the family of elements in @xmath11 satisfying the conditions in definition @xmath552 . if @xmath553 factors linearly over @xmath428 $ ] for @xmath554 where @xmath555 for some positive integer @xmath319 then @xmath11 splits @xmath556 . the proof proceeds as the proof of lemma [ keylemma ] , assuming w.l.o.g . we first find a decomposition @xmath375 of @xmath10 and for each @xmath203 we can then find @xmath2 and @xmath64 such that @xmath196 is invertible and @xmath197 whenever @xmath198 in @xmath380 . we can then write @xmath199 with @xmath200 . since @xmath11 splits @xmath10 then by lemma[splitpersist ] we can find a decomposition @xmath557 of @xmath11 such that each @xmath379 splits @xmath380 for each @xmath203 . we then find a further decomposition @xmath382 of @xmath380 and for each @xmath387 a number @xmath161 and a separable extension @xmath558 $ ] of @xmath559 such that @xmath560 with @xmath386 invertible . similarly , we can decompose each @xmath379 further into @xmath561 such that each @xmath505 splits each @xmath559 for all @xmath562 . let the family @xmath563 be the image of the family @xmath551 in @xmath505 . then @xmath505 splits @xmath10 with @xmath564 as the family of elements of @xmath505 satisfying definition [ splitdef ] . but then @xmath565 factors linearly over @xmath505 . for some subfamily @xmath566 of elements in @xmath505 we have that @xmath505 splits @xmath559 . thus @xmath567 factors linearly over @xmath505 for all @xmath568 in the family . by lemma [ splitlemma ] we have that @xmath569 factors linearly over @xmath505 for all @xmath568 . thus @xmath505 splits the extension @xmath558 $ ] . but then by lemma [ decompsplitsplit ] we have that @xmath11 splits @xmath558 $ ] . using hensel s lemma [ hensel ] , we can lift this to a proper decomposition @xmath201 with @xmath202 monic of degree @xmath387 and @xmath205 monic of degree @xmath318 . we take @xmath388 and @xmath389 we can then extend theorem [ abhyankarproofdyn ] as follows . [ dynamicnewton - extended ] let @xmath173[y]$ ] be a monic non - constant polynomial separable over @xmath57 . there exists then a triangular separable algebra @xmath10 over @xmath0 and @xmath175 and a factorization @xmath391\ ] ] moreover , if @xmath11 is a triangular separable algebra over @xmath0 such that @xmath12 factors linearly over @xmath570 $ ] for some positive integer @xmath161 then @xmath11 splits @xmath10 . as we shall see in the examples below , the result of the computation is usually several triangular separable algebras @xmath363 over the base field @xmath0 with linear factorizations of @xmath6 over @xmath571, ... $ ] for some @xmath572 . the previous theorem allows us to state the following about these algebras . [ spliteachother ] let @xmath11 and @xmath15 be two triangular separable algebras obtained by the algorithm of theorem [ abhyankarproofdyn ] . then @xmath11 splits @xmath15 and @xmath15 splits @xmath11 . consequently , a triangular separable algebra obtained by this algorithm splits itself . thus given any two algebras @xmath305and @xmath573 obtained by the algorithm and two prime ideals @xmath574 and @xmath575 we have a field isomorphism @xmath576 . therefore all the algebras obtained are approximations of the same field @xmath465 . since @xmath465 splits all the algebras and itself is a refinement , @xmath465 splits itself , i.e. @xmath577}$ ] and @xmath465 is a normal , in fact a galois extension of @xmath0 . classically , this field @xmath465 is the field of constants generated over @xmath0 by the set of coefficients of the puiseux expansions of @xmath6 . the set of puiseux expansions of @xmath6 is closed under the action of @xmath578 , where @xmath21 is the algebraic closure of @xmath0 . thus the field of constants generated by the coefficients of the expansions of @xmath6 is a galois extension . the algebras generated by our algorithm are powers of this field of constants , hence are in some sense minimal extensions . even without the notion of prime ideals we can still show interesting relationship between the algebras produced by the algorithm of theorem [ abhyankarproofdyn ] . the plan is to show that any two such algebras @xmath11 and @xmath15 are essentially isomorphic in the sense that each of them is equal to the power of some common triangular separable algebra @xmath10 , i.e. @xmath16 and @xmath17 for some positive integers @xmath18 . to show that @xmath16 we have to be able to decompose @xmath11 . to do this we need to constructively obtain a system of orthogonal nontrivial ( unless @xmath579 already ) idempotents @xmath471 . since @xmath11 and @xmath15 split each other , the composition of these maps gives a homomorphism from @xmath11 to itself . we know that a homomorphism between a field and itself is an automorphism thus as we would expect if there is a homomorphism from a triangular separable algebra @xmath11 to itself that is not an automorphism we can decompose this algebra non trivially . we use the composition of the split maps from @xmath11 to @xmath15 and vice versa as our homomorphism this will enable us to repeat the process after the initial decomposition , that is if @xmath580 are algebras in the decompositions of @xmath11 and @xmath15 , respectively , we know that they split each other . this process of decomposition stops once we reach the common algebra @xmath10 . [ dynamicnormal1 ] let @xmath11 be a triangular separable algebra over a field @xmath0 and let @xmath581 be @xmath0homomorphism . then @xmath582 is either an automorphism of @xmath11 or we can find a non - trivial decomposition @xmath583 . let @xmath584,p_1, ... ,p_l$ ] with @xmath585 . let @xmath582 map @xmath474 to @xmath586 , for @xmath587 . then @xmath586 is a root of @xmath588 . the set of vectors @xmath589 is a basis for the vector space @xmath11 over @xmath0 . if the image @xmath590 is a basis for @xmath11 , i.e. @xmath591 is a linearly independent set then @xmath582 is surjective and thus an automorphism . + assuming @xmath582 is not an automorphism , then the kernel of @xmath582 is non - trivial , i.e. we have a non - zero non - unit element in @xmath592 , thus we have a non - trivial decomposition of @xmath11 . [ dynamicnormal2 ] let @xmath593 be triangular separable algebras over a field @xmath0 such that @xmath11 splits @xmath15 and @xmath15 splits @xmath11 . then there exist a triangular separable algebra @xmath10 over @xmath0 and two positive integers @xmath18 such that @xmath594 and @xmath595 . first we note that by corollary [ splitalgsplitrefinement ] if @xmath11 splits @xmath15 then @xmath11 splits any refinement of @xmath15 . trivially if @xmath11 splits @xmath15 then any refinement of @xmath11 splits @xmath15 . since @xmath11 and @xmath15 split each other then there is @xmath0homomorphisms @xmath596 and @xmath597 . the maps @xmath598 and @xmath599 are @xmath0homomorphisms from @xmath11 to @xmath11 and @xmath15 to @xmath15 respectively . if both @xmath582 and @xmath600 are automorphisms then we are done . otherwise , by lemma [ dynamicnormal1 ] we can find a decomposition of either @xmath11 or @xmath15 . by induction on @xmath601 the statement follows . theorems [ dynamicnormal2 ] and [ dynamicnewton - extended ] show that the algebras obtained by the algorithm of theorem [ abhyankarproofdyn ] are equal to the power of some common algebra . this common triangular separable algebra is an approximation , for lack of irreducibility test for polynomials , of the normal field extension of @xmath0 generated by the coefficients of the puiseux expansions @xmath602 $ ] of @xmath6 , where @xmath21 is the algebraic closure of @xmath0 . the following are examples from a haskell implementation of the algorithm . we truncate the different factors unevenly for readability . applying the algorithm to @xmath603[y]$ ] @xmath604,a=0,\ ; b^2 - 13/36=0,\ ; c^2 - 3=0 \\ & f(x , y ) = \\ & \;\;\ ; ( y+(-b-1/6)x+(-31b/351 - 7/162)x^3+(-415b/41067 - 29/1458)x^5 + ... ) \\ & \;\;\ ; ( y+(b-1/6)x+(31b/351 - 7/162)x^3+(1415b/41067 - 29/1458)x^5 + ... ) \\ & \;\;\ ; ( y - c+x/6 + 5cx^2/72 + 7x^3/162 + 185cx^4/10368 + 29x^5/1458+ ... )\\ & \;\;\ ; ( y+c+x/6 - 5cx^2/72 + 7x^3/162 - 185cx^4/10368 + 29x^5/1458+ ... )\end{aligned}\ ] ] @xmath604 , a^2 - 3=0,b - a/3=0,c^2 - 13/36=0\\ & f(x , y)=\\ & \;\;\ ; ( y - a+x/6 + 5ax^2/72 + 7x^3/162 + 185ax^4/10368 + 29x^5/1458 + ... ) \\ & \;\;\ ; ( y+(-c-1/6)x+(-31c/351 - 7/162)x^3+(-415c/41067 - 29/1458)x^5+ ... )\\ & \;\;\ ; ( y+(c-1/6)x+(31c/351 - 7/162)x^3+(1415c/41067 - 29/1458)x^5+ ... )\\ & \;\;\ ; ( y+a+x/6 - 5ax^2/72 + 7x^3/162 - 185ax^4/10368 + 29x^5/1458 + ... ) \end{aligned}\ ] ] @xmath604 , a^2 - 3=0,b+2a/3=0,c^2 - 13/36=0\\ & f(x , y ) = \\ & \;\;\;(y - a+x/6 + 5ax^2/72 + 7x^3/162 + 185ax^4/10368 + 29x^5/1458+ ... )\\ & \;\;\ ; ( y+a+x/6 - 5ax^2/72 + 7x^3/162 - 185ax^4/10368 + 29x^5/1458+ ... )\\ & \;\;\ ; ( y+(-c-1/6)x+(-31c/351 - 7/162)x^3+(-415c/41067 - 29/1458)x^5 + ... ) \\ & \;\;\ ; ( y+(c-1/6)x+(31c/351 - 7/162)x^3+(1415c/41067 - 29/1458)x^5+ ... )\end{aligned}\ ] ] the algebras in the above example can be readily seen to be isomorphic . however , as we will show next , this is not always the case . to illustrate theorem [ dynamicnormal2 ] we show how it works in the context of an example computation . the polynomial is @xmath605 . the following are two of the several triangular separable algebras obtained by our algorithm along with their respective factorization of @xmath12 . @xmath606 , p_1,p_2,p_3,p_5\\ & p_1=y^4 - 4 , \ ; p_2 = y - a/5 , \ ; p_3 = y^2 - 1/4,\\ & p_4 = y^3 + 2 a^2y/3 + 20 a^3/27,\ ; p_5 = y^2 + 3 d^2/4 + 2 a^2/3 \\ & f(x , y ) = \big ( y - ax^{\tfrac{1}{2}}+3a^3x^{\tfrac{3}{2}}/16 + ... \big ) \big(y - cx^2 + ... \big ) \big(y+cx^2 + ... \big ) \\ & \;\;\;\ ; \big ( y+(-d+a/3)x^{\tfrac{1}{2}}+(-3ad^2/16-a^2d/16 - 7a^3/48)x^{\tfrac{3}{2}}+ ... \big ) \\ & \;\;\;\ ; \big ( y+(-e+d/2+a/3)x^{\tfrac{1}{2}}+\\ & \;\;\;\;\;\;\;\;\;\;\;\;(3ade/16-a^2e/16 + 3ad^2/32+a^2d/32-a^3/48)x^{\tfrac{3}{2}}+ ... \big ) \\ & \;\;\;\ ; \big ( y+(e+d/2+a/3)x^{\tfrac{1}{2}}+\\ & \;\;\;\;\;\;\;\;\;\;\;\;(-3ade/16+a^2e/16 + 3ad^2/32+a^2d/32-a^3/48)x^{\tfrac{3}{2}}+ ... \big ) \\ \\ & b = q[r , t , u , v , w ] , q_1,q_2,q_3,q_5\\ & q_1= y^4 - 4,\ ; q_2= y+4r/5,\ ; q_3= y,\;q_4=y^2 - 1/4,\ ; q_5=y^2+r^2 \\ & f(x , y ) = ( y - rx^{\tfrac{1}{2}}+3r^3x^{\tfrac{3}{2}}/16 + ... ) ( y+rx^{\tfrac{1}{2}}-3r^3x^{\tfrac{3}{2}}/16 + ... ) \\ & \;\;\;\ ; ( y - vx^2 + ... ) ( y+vx^2 + ... ) \\ & \;\;\;\ ; ( y - wx^{\tfrac{1}{2}}-3r^2wx^{\tfrac{3}{2}}/16 + ... ) ( y+wx^{\tfrac{1}{2}}+3r^2wx^{\tfrac{3}{2}}/16 + ... ) \end{aligned}\ ] ] we now show that the two algebras indeed split each other . over @xmath15 the polynomial @xmath326 factors as @xmath607 . each of these factors partly specify a homomorphism taking @xmath49 to a zero of @xmath326 in @xmath15 . for each we get a factorization of @xmath608 over @xmath15 . * @xmath609 + @xmath610 * @xmath611 + @xmath612 * @xmath613 + @xmath614 * @xmath615 + @xmath616 for each of the @xmath617 mappings of @xmath49 we get @xmath618 mappings of @xmath79 . now we see we have @xmath619 different mappings arising from the different mappings of @xmath49 and @xmath79 . each of these @xmath619 mappings will give rise to @xmath620 different mappings of @xmath621 ( factorization of @xmath622) ... etc . thus we have a number of homomorphisms equal to the dimension of the algebra , that is @xmath623 homomorphisms . we avoid listing all these homomorphisms here . in conclusion , we see that @xmath15 splits @xmath11 . similarly , we have that @xmath11 splits @xmath15 . we show only one of the @xmath624 homomorphisms below . the polynomial @xmath625 factors linearly over @xmath11 as @xmath626 . under the map @xmath627 we get a factorization of @xmath628 over @xmath11 as @xmath629 now to the application of theorem [ dynamicnormal2 ] . under the map above we have an endomorphism @xmath630 and @xmath631 . thus in the kernel we have the non - zero element @xmath632 and as expected @xmath633 divides @xmath608 . using this we obtain a decomposition of @xmath634 . we have @xmath635,p_1 , p_2,p_3,g_4,p_5 $ ] with @xmath636 and @xmath637,p_1 , p_2,p_3,h_4,p_5 $ ] with @xmath638 . with @xmath639 in @xmath640 , @xmath641 and we can see immediately that @xmath642 . similarly , we can decompose the algebra @xmath643 , where @xmath644,p_1,p_2,p_3,h_4 , g_5 $ ] with @xmath645 and @xmath646,p_1,p_2,p_3,h_4 , h_5 $ ] with @xmath647 . the polynomial @xmath628 factors linearly over both @xmath648 and @xmath649 as @xmath650 . we can readily see that both @xmath648 and @xmath649 are isomorphic to @xmath15 , through the @xmath648 automorphism @xmath651 . thus proving @xmath652 . we are grateful to henri lombardi for the useful comments and discussions during the work leading to this paper . we thank the referees for comments and suggestions that helped to improve the article . * s basu * , * r pollack * , * m - f roy * , http://dx.doi.org/10.1007/3-540-33099-2[_algorithms in real algebraic geometry ( algorithms and computation in mathematics ) _ ] , springer - verlag new york , inc . , secaucus , nj , usa ( 2006 ) * j della dora * , * c dicrescenzo * , * d duval * , http://dx.doi.org/10.1007/3-540-15984-3_279[_about a new method for computing in algebraic number fields _ ] , from : `` eurocal 85 '' , ( b caviness , editor ) , lecture notes in computer science 204 , springer berlin / heidelberg ( 1985 ) 289290

+ since its launch in 2004 , the swift satellite has monitored the x - ray afterglows of several hundred gamma - ray bursts , and revealed that their x - ray light - curves are more complex than previously thought , exhibiting up to three power - law segments . energy injection into the relativistic blast - wave energizing the burst ambient medium has been proposed most often to be the reason for the x - ray afterglow complexity . we examine 117 light - curve breaks of 98 swift x - ray afterglows , selected for their high - quality monitoring and well - constrained flux decay rates . thirty percent of afterglows have a break that can be an adiabatic jet - break , in the sense that there is one variant of the forward - shock emission from a collimated outflow model that can account for both the pre- and post - break flux power - law decay indices , given the measured x - ray spectral slope . if allowance is made for a steady energy injection into the forward - shock , then another 56 percent of x - ray afterglows have a light - curve break that can be explained with a jet - break . the remaining 12 percent that are not jet - breaks , as well as the existence of two breaks in 19 afterglows ( out of which only one can be a jet - break ) , suggest that some x - ray breaks arise from a sudden change in the rate at which energy is added to the blast - wave , and it may well be that a larger fraction of x - ray light - curve breaks are generated by that mechanism . the fractional increase in the shock energy , inferred from the energy injection required to account for the observed x - ray flux decays , may be anticorrelated with the grb prompt output , whether the x - ray break is a jet - break or an energy - injection break . that anticorrelation can also be seen as bursts with a higher energy output being followed by faster fading x - ray afterglows . to test the above two mechanisms for afterglow light - curve breaks , we derive comprehensive analytical results for the dynamics of outflows undergoing energy injection and for their light - curves , including closure relations for inverse - compton afterglows and for the emission from spreading jets interacting with an wind - like ambient medium . + + 2 3 4 8 c 51 52 53 54 ii o radiation mechanisms : non - thermal , relativistic processes , shock waves , gamma - ray bursts , ism : jets and outflows

1212.1486

the power - law temporal decays displayed by the radio , optical , and x - ray fluxes of grb afterglows ( @xmath8 ) have provided strong support for the prediction ( meszaros & rees 1997 ) that afterglows arise from the relativistic blast - wave generated as the grb outflow ejecta interacts with the ambient medium . if the grb outflow is a collimated jet then a break in the smooth power - law decay of the afterglow is expected for both either a conical jet ( panaitescu , meszaros & rees 1998 ) or a sideways - spreading jet ( by rhoads 1999 ) . the breaks found in the flux decay light curves of most of the well - sampled afterglows that were localized by the bepposax satellite seemed to provide solid observational support for the existence of these predicted jet - breaks ( e.g. frail et al 2001 ) . it was also realized early on that there should be a substantial diversity in the steepness of the afterglow flux decay , even when the spectral slope is constant , if the fundamental forward - shock parameters vary as the shock traverses the ambient medium . rees & meszaros ( 1998 ) derived the expected power - law flux decay for the forward - shock synchrotron flux in the case of an outflow whose kinetic energy - per - solid angle @xmath9 is not constant in time , either because the shock s energy increases due to fresh ejecta arriving at the blast - wave or because @xmath9 is anisotropic . two important microphysical parameters ( @xmath10 and @xmath11 ) , which quantify the fraction of the post - shock energy acquired by electrons and magnetic field , also determine the intensity the afterglow synchrotron flux , and their variability could also yield a variety of decay indices @xmath12 . swift measurements of grb x - ray afterglows provide support for that shock parameters vary . the standard forward - shock model with constant energy and shock micro - parameters predicts a closure relation between the temporal decay index @xmath12 and the slope @xmath13 of the x - ray continuum ( @xmath14 ) of the form @xmath15 . the early x - ray flux of some swift afterglows decays slower than expected for the measured spectral slope @xmath13 , which was attributed to a variable kinetic energy of the forward shock ( nousek et al 2006 , panaitescu et al 2006 , zhang et al 2006 ) . moreover , the flux decay indices @xmath12 displayed by swift x - ray afterglows after the slow decay phase are not correlated with the x - ray spectral slope @xmath13 ( panaitescu 2007 ) , as would be expected from the @xmath16 closure relation for the forward - shock emission . when optical light - curves are taken into account , the standard jet model fares even worse , being inconsistent with practically all swift afterglows ( liang et al 2008 ) . while more than half of the well - sampled optical afterglows display coupled optical and x - ray light - curves ( panaitescu & vestrand 2011 ) , exhibiting similar flux decay indices and/or simultaneous light - curve breaks , many other afterglows show a decoupling of their optical and x - ray light - curves , with a light - curve break occurring only in the x - ray . at least four possibilities for this light - curve decoupling , which is not predicted by the standard jet model , have been proposed : @xmath17 ) the x - ray afterglow emission arises occasionally from the same not - yet - understood mechanism which produces the prompt burst emission ( ghisellini et al 2007 ) , @xmath18 ) there is a dominant contribution of the reverse - shock to the afterglow emission ( uhm & beloborodov 2007 ) , @xmath19 ) the forward - shock emission is reprocessed by bulk - scattering and/or inverse - compton scattering outside the forward - shock ( panaitescu 2008 ) , or @xmath20 ) the optical and x - ray afterglow flux arise from different parts of the forward - shock , from different jets ( e.g. racusin et al 2008 ) . in this work , _ we assume that swift x - ray afterglows arise from the forward - shock _ , and attribute the failure of the standard jet model in accounting for the x - ray flux decay indices to the assumption that the forward shock energy is constant . evolving micro - physical parameters ( e.g. ioka et al 2006 ) could , in principle , explain the measured x - ray flux decay indices , but the existence of light - curve breaks requires an unnatural , sudden change in their evolution . for that reason , we maintain in this work the basic assumption of the standard jet model that micro - physical parameters do not evolve . we also assume that the ambient medium with which the post - grb outflow interacts is uniform , without discontinuities that could yield breaks in the afterglow light - curve at a frequency . a notable model thus excluded is that where the afterglow - producing outflow interacts with a prior outflow ( ioka et al 2006 ) whose density increases radially outward up to its termination shock , yielding a slower decay of the afterglow flux than expected for a homogeneous ambient medium , followed by a break to a steeper decay . one reason for relaxing the constant shock energy constraint is the observational evidence for x - ray flares in swift afterglows ( e.g. chincarini et al 2007 ) . the short timescale of those flares indicates that the grb progenitor produces relativistic ejecta after the burst phase , whose arrival at the decelerating forward - shock could increase substantially the blast - wave s energy . our hypothesis is that such an injection of energy in the forward - shock is , in general , a more persistent and continuous process . however , we note that the fluence of x - ray flares is about 10 percent that of the burst ( falcone et al 2007 ) , hence the late outflow powering the flares could carry a dynamically - important energy only if that outflow is much less efficient than the grb outflow in producing x - ray emission . we derive in [ anal ] the closure relations @xmath16 for a continuous energy injection in the forward - shock . for the ease of extracting from observations the required injected power history , we assume that the forward - shock energy increases as a power with observer - time , @xmath21 , which is motivated by that the resulting afterglow flux should be power - law in observer time . this assumption circumvents a calculation of the kinematics for the ejecta forward - shock interaction . in [ swift ] , the closure relations for the forward - shock emission with energy injection are applied to a set of swift afterglows with well sampled light - curves . there , we examine two plausible scenarios for the origin of afterglow x - ray light - curve breaks ( break in energy injection , jet - break ) , three dynamical regimes of the forward - shock ( spherical expansion , conical jet , spreading jet ) , and two radiation processes that may yield the afterglow x - ray flux ( synchrotron , inverse - compton ) . there are three mechanisms that could produce a change in the flux power - law decay index @xmath12 , i.e. an afterglow light - curve break . \(1 ) passage of a spectral break . x - ray light - curve breaks are , generally , not accompanied by a spectral evolution ( nousek et al 2006 ; liang , zhang & zhang 2007 ; willingale et al 2007 ) , hence their origin in the passage of the spectral feature is largely ruled out , but there are other reasons why such an x - ray light - curve break origin is unlikely . the slower evolution of the cooling frequency @xmath22 ( the synchrotron characteristic frequency at which radiate the electrons that cool radiatively on the hydrodynamical timescale ) leads to extended x - ray light - curve breaks when @xmath22 crosses the 0.3 - 10 kev band . the peak frequency @xmath23 of the @xmath24 synchrotron spectrum decreases faster , but it is likely to be lower than the x - rays even at early times , hence it can not cross the x - ray band during afterglow observations . \(2 ) jet - break . a light - curve break occurs at the ` jet - break time ' , when the outflow lorentz factor @xmath25 drops below the inverse of the jet half - angle @xmath26 . for a ` conical jet ' that does not spread laterally ( as in the case of an outflow core embedded in an envelope that prevents the spreading of the core but yields a negligible emission ) , the steepening of the x - ray flux decay is solely due to that , after the jet - break time , the average kinetic energy per solid angle @xmath9 over the visible cone of angle @xmath27 ( where relativistic doppler boost is effective ) is decreasing with observer - frame time . for a ` spreading jet ' ( as expected for a top - hat distribution of @xmath9 ) , in addition to the above - described geometrical - relativistic effect , there is a second contribution to the x - ray flux decay steepening from the enhanced jet deceleration produced by the jet lateral spreading . \(3 ) energy injection break . if energy injection occurs routinely in grb afterglows , then a light - curve break could be produced by a change in the injected power . the dynamical behaviours of the forward - shock model to be considered are : @xmath17 ) a ` spherical ' outflow , i.e. a wide jet whose edge becomes visible to the observer after the last x - ray measurement ( at @xmath28 ) , which occurs if jet half - angle satisfies @xmath29 , and @xmath18 ) a narrow jet whose edge is already visible at the first x - ray observation ( at @xmath30 ) , occurring when @xmath31 . for the latter model , we will consider both a conical jet and a spreading jet . these dynamical models are applicable also to x - ray afterglows without light - curve breaks , in which case the energy injection law must be unevolving ( a single index @xmath32 ) . we will investigate the ability of the energy - injection break and jet - break models above to accommodate the observed x - ray flux decays before and after the break , with allowance for all three dynamical models ( spherical outflow , conical jet , spreading jet ) , for two types of ambient medium ( homogeneous and wind - like ) , two radiation mechanisms dominating the x - ray flux ( synchrotron and inverse - compton ) , and two possible locations of the x - ray relative to the cooling frequency @xmath22 but , first , a word about their rationale . wind - like media , having a @xmath33 particle density , are expected at 0.1 - 1 pc from the source of long grbs , i.e. where the @xmath34 afterglow emission is produced , because the progenitors of such bursts are massive wolf - rayet stars that drive powerful winds . the termination - shock of such winds is of several pcs ( garcia - segura et al 1996 ) , thus a homogeneous medium at the afterglow location seems unjustified . however , if the grb progenitor resides in a highly - pressurized bubble , the wind termination - shock radius could be as low as 0.5 pc ( chevalier et al 2004 ) , with a further reduction of that radius if the wr wind is weak , as expected along the rotation axis ( ramirez - ruiz et al 2005 ) , or if the wr star moves sufficiently fast relative to the ism ( van marle et al 2006 ) , thus the afterglow could occur in shocked wind , which is quasi - homogeneous . synchrotron emission is generally believed to produce all the afterglow emission , but the likelihood of inverse - compton emission being dominant is higher at higher photon frequency , like x - rays . admittedly , that requires a high environment density , perhaps ten times higher ( panaitescu 2011 ) than implied by the mass - loss rates and terminal wind velocities measured for galactic wolf - rayet winds ( nugis & lamers 2000 ) . the @xmath16 closure relations expected for a given variant of the forward - shock model are derived from the evolution of the spectral characteristics @xmath35 ( peak flux ) , @xmath23 ( frequency of the @xmath36 spectrum peak ) , and @xmath22 ( cooling frequency ) , and from the shape of the @xmath37 spectrum at frequencies higher than its @xmath38 peak : @xmath39 for @xmath40 , @xmath41 for @xmath42 and @xmath43 for @xmath44 , where @xmath45 is the power - law index of the electron distribution with energy produced by the shock ( @xmath46 ) . therefore , the flux received at observing frequency @xmath47 satisfies @xmath48 for @xmath42 and @xmath49 for @xmath44 ( @xmath40 implies an x - ray spectrum with @xmath50 that is harder than measured for nearly all swift x - ray afterglows ) . for a spherical outflow ( or a jet before the jet - break time @xmath51 ) and for synchrotron emission , @xmath52 , @xmath53 , @xmath54 . here @xmath55 is the mass of the ambient medium swept - up by the forward - shock ( i.e. the number of radiating electrons ) ; @xmath56 is the post - shock magnetic field , whose energy density @xmath57 is assumed to be a fraction @xmath11 of the post - shock energy density @xmath58 ( from the shock jump conditions ) ; @xmath59 is the external density at the shock s location ; @xmath60 is the typical electron random lorentz factor , parametrized as a fraction @xmath10 of the proton random lorentz factor @xmath61 ; and @xmath62^{-1}$ ] is the random lorentz factor of the electrons whose radiative cooling timescale equals the comoving - frame shock s age @xmath63 , @xmath64 being the outflow radius and @xmath65 the compton parameter . we assume the more likely situations of either electrons cooling mostly through synchrotron emission ( @xmath66 ) or , if inverse - compton is dominant ( @xmath67 ) , that @xmath68 , in which case @xmath69 is constant and does not affect the evolution of @xmath22 . for a jet ( conical or spreading ) after the jet - break time and for _ synchrotron _ emission , we have @xmath70 , with @xmath23 and @xmath22 as above . thus @xmath71 @xmath72 @xmath73 the characteristics of the _ inverse - compton _ spectrum are immediately related with those of the synchrotron spectrum : @xmath74 ( where @xmath75 is the optical thickness to electron scattering of the shocked ambient medium ) , @xmath76 , and @xmath77 , thus @xmath78 @xmath79 @xmath80 the above equations for the spectral characteristics show that their evolution and , thus , the resulting flux decay index @xmath12 , is set by the forward - shock dynamics : @xmath81 , @xmath82 , and @xmath83 . the energy - per - particle behind that shock is @xmath84 , hence the lab - frame energy of the forward - shock is @xmath85 and its dynamics is given by @xmath86 , where @xmath87 is the kinetic energy of the ejecta ( all transferred to the forward - shock ) at the time @xmath88 when the assumed power - law energy injection ( @xmath21 ) begins . the other equations for the forward - shock dynamics are @xmath89 , with @xmath90 the density of the ambient medium at radius @xmath64 ( @xmath91 for a homogeneous medium , @xmath92 for a wind ) and @xmath93 the increase of the jet opening at the comoving - frame sound speed ( @xmath94 ) , and @xmath95 for the observer - frame arrival time @xmath96 of photons emitted along the direction toward the observer . defining @xmath97 , @xmath98 with @xmath99 ( the radius where the lateral spreading almost doubles the initial opening of an adiabatic jet interacting with a homogeneous medium ) , and @xmath100 , the above equations for the outflow dynamics become : @xmath101 @xmath102 @xmath103 @xmath104 equation ( [ dyn2 ] ) allows the conversion of the jet dynamics [ @xmath105 , @xmath106 , @xmath107 from lab frame to observer frame [ @xmath108 , @xmath82 , @xmath109 . for a spherical outflow , a conical jet , or a spreading jet before the jet - break time , @xmath110 is constant and the above equations can be solved easily , leading to @xmath111 @xmath112 @xmath113 @xmath114 which can be replaced in the equations for the synchrotron and inverse - compton spectral characteristics ( equations [ spsy1][spsy3 ] and [ spic1][spic3 ] ) to obtain their evolutions and , then , the power - law index of the afterglow flux decays given in lines 1 , 2 , 4 , and 5 of table [ spek ] and table [ alfa ] , respectively . lcccccccc & & + & @xmath115 & @xmath116 & @xmath117 & @xmath115 & @xmath116 & @xmath117 + syn spherical & @xmath32 & -2@xmath118 & -2@xmath119 & -2@xmath120 & -2@xmath118 & 2@xmath119 + + syn conical jet & 4@xmath121 & -2@xmath118 & -2@xmath119 & @xmath122 & -2@xmath118 & 2@xmath119 + + syn spreading jet & @xmath123 & -@xmath124 & @xmath125 & @xmath122&@xmath126&@xmath32 + ic spherical & 4@xmath127 & @xmath128 & -4@xmath127 & -1 & @xmath126&@xmath129 + + ic conical jet & 2@xmath130 & @xmath128 & -4@xmath127 & -2@xmath118&@xmath126&@xmath129 + + ic spreading jet & @xmath131 & @xmath132 & @xmath133&@xmath134&@xmath135&@xmath136 + [ spek ] for a jet spreading laterally , the dynamics equations can be solved analytically in the case of no energy injection and for a homogeneous medium . that has been done by rhoads ( 1999 ) , who obtained an exponential jet deceleration for a lateral spreading given by @xmath137 : @xmath138 , @xmath139 , @xmath140 at @xmath141 , where @xmath142 . because the differential lateral spreading is on top " of a conical jet and not over " a cylindrical jet ( as in rhoads formalism ) , the correct lateral expansion is @xmath143 . for @xmath144 and @xmath91 , equations ( [ dyn1])([dyn2 ] ) ( which use the latter prescription for jet spreading ) have the solution @xmath145 , @xmath146 , @xmath147 for @xmath148 , which can be shown to lead to the same flux power - law decay indices as for rhoads prescription because the increase of the jet radius is only logarithmic in observer time ( note that , for a nearly constant jet radius @xmath64 , @xmath143 implies @xmath137 ) . the above exponential solutions for the dynamics of an adiabatic jet interacting with a homogeneous medium can be expressed in observer time by integrating @xmath149 , which leads to @xmath150 for @xmath151 . thus @xmath152 from where it can be shown that @xmath153 @xmath154 @xmath155 where @xmath156 is the jet energy and @xmath157 . lcccccccc & & + & @xmath158 & @xmath159 & @xmath158 & @xmath159 + syn spherical & 4@xmath160 4@xmath161 & 4@xmath162 4@xmath163 & 4@xmath164 4@xmath165 & 4@xmath162 4@xmath163 + + syn conical jet & 4@xmath166 4@xmath167 & 4@xmath168 4@xmath161 & 4@xmath168 4@xmath161 & 4@xmath166 4@xmath167 + + syn spreading jet & @xmath169 3@xmath161&@xmath169 3@xmath163 & @xmath169 2@xmath165 & @xmath169 2@xmath163 + ic spherical & 8@xmath170 8@xmath171 & 8@xmath172 8@xmath173 & @xmath169 2@xmath174 & @xmath175 2@xmath176 + + ic conical jet & 8@xmath177 8@xmath178 & 8@xmath179 8@xmath180 & @xmath181 2 @xmath1822@xmath176 & @xmath169 2 @xmath183 2@xmath165 + + ic spreading jet & 2@xmath160 @xmath184 & 2@xmath185 2@xmath176 & 2@xmath186 & 2@xmath187 + [ alfa ] in the more general case of energy injection and/or for a wind - like medium , the solution to the dynamics equations ( [ dyn1])([dyn2 ] ) can not be found analytically . they can be cast as logarithmic derivatives @xmath188 @xmath189 @xmath190 and integrated numerically , which showed that the left - hand sides are ( asymptotically ) constant , i.e. the dynamics [ @xmath82 , @xmath191 , @xmath83 ] of a spreading jet undergoing energy injection according to @xmath21 is a power - law in observer time . that dynamics is between the extreme cases of an adiabatic jet interacting with a homogeneous medium ( which yields the fastest deceleration ) and a spherical outflow with energy injection ( which yields the slowest deceleration ) . given that , in both of those two extreme cases , the dynamical quantities have a power - law evolution with time , it is not surprising that an intermediate case manifests that feature too . replacing power - law solutions for @xmath193 , @xmath110 , and @xmath194 in equations ( [ f])([x ] ) , allows the calculation of the temporal exponents by requiring that the right - hand sides are also constants , i.e. without a @xmath195 dependence . the result is @xmath196 @xmath197/2 } \theta \propto t^{\frac{1}{2}(1-\frac{e}{3-s})}\ ] ] @xmath198 it is worth mentioning that the above dynamics of spreading jets undergoing energy injection can be recovered from the dynamics of a spreading adiabatic jet interacting with a homogeneous medium ( equations [ m ] , [ teta ] ) if one replaces @xmath21 ( the prescription for energy injection ) and @xmath199 . the latter relation is trivial for an outflow decelerating as a power - law with radius ( as for a spherical outflow or a conical jet ) and is approximately satisfied for an exponential deceleration ( as for an adiabatic spreading jet interacting with a homogeneous medium ) , which suggests that it should also stand in any other case of an intermediate - strength deceleration ( that is verified with the numerical integration of the jet dynamics ) . either from equations ( [ dyn3 ] ) and ( [ dyn4])or by replacing @xmath156 and @xmath200 in equations ( [ gm ] ) and ( [ tprim ] ) with the above prescriptions , it can be shown that @xmath201/2 } \gamma \propto t^{-\frac{1}{2}(1-\frac{e}{3-s } ) } \label{dyn5}\ ] ] @xmath202/2 } ( \propto r/\gamma ) t ' \propto t^{\frac{1}{2}(1+\frac{e}{3-s } ) } ( \propto r/\gamma ) \label{dyn6}\ ] ] for a spreading jet and with energy injection . by substituting equations ( [ dyn3])([dyn6 ] ) in the relations for the synchrotron and inverse - compton spectral characteristics ( equations [ spsy1][spsy3 ] and [ spic1][spic3 ] ) , one can derive the evolutions from lines 3 and 6 of table [ spek ] and the power - law indices of the afterglow flux decay from lines 3 and 6 of table [ alfa ] . we acknowledge that the dynamics of spreading jets undergoing energy injection was derived using a simplistic toy model for the jet lateral spreading , where the jet was assumed to have a uniform kinetic energy per solid angle @xmath9 , although both lateral spreading and energy injection could / should introduce a substantial angular gradient of @xmath9 , decreasing with angle measured from the jet s spine . [ cols="^,^,^,^,^,^,^,^,^,^ " , ] ( 1 ) : x - ray spectral slope ( 90 percent cl ) + ( 2 ) : epoch of first swift x - ray measurement + ( 3 ) : index of x - ray flux power - law decay at @xmath203 ( 90 percent cl ) + ( 4 ) : epoch of x - ray light - curve break + ( 5 ) : index of x - ray flux power - law decay at @xmath204 ( 90 percent cl ) + ( 6 ) : epoch of last measurement + ( 7 ) : 10 kev10 mev isotropic energy release during prompt phase ( uncertain by up to 50 percent ) + ( 8) : redshift + out of the 634 afterglow light - curves listed at the swift - xrt light - curve repository ( evans et al 2007 ) posted until the end of year 2011 , we have retained only those for which @xmath205 a light - curve break is obvious or seems possible , @xmath206 a power - law light - curve segment is monitored for longer than a factor 5 in time before and after the break , and @xmath207 the spectral slope @xmath208 listed at the swift - xrt spectra repository ( evans et al 2007 ) has a 90 percent confidence level ( cl ) less than 0.16 . the light - curves have been piece - wise fit with power - laws , using measurements sufficiently far from the break time , to capture the asymptotic power - law flux decay . shallow breaks for which the data could be fit satisfactorily ( i.e. reduced @xmath209 ) with a single power - law were excluded . the 90 percent confidence level ( @xmath210 ) was determined from the variation of @xmath211 around the best power - law fit . as shown in table [ alfa ] , for an adiabatic outflow ( @xmath144 ) , the power - law index is @xmath212 + const = @xmath213/2 + const , with the coefficient of @xmath208 being as large as 9/8 for other models . thus , an equal importance of measurement uncertainties @xmath210 and @xmath214 requires that we retain only afterglows whose decay indices can be determined with a 90 percent cl smaller than @xmath215 . the above two basic selection criteria led to the sample of 98 afterglows and 117 x - ray light - curve breaks listed in table [ ag1 ] . forty - four afterglows have temporal coverage starting from the burst end ( for them , we determine the fractional increase of the outflow energy during the afterglow phase ) and a measured redshift , which allows the calculation of the grb output and of the afterglow jet energy . the light - curves of the afterglows listed in table [ ag1 ] are morphologically diverse . a minority display a continuous , slowly - decaying power - law flux from the burst phase to the afterglow , the afterglow beginning being chosen when the flux fluctuations disappear . the majority of grbs display a sharp decay after the prompt phase ( the grb tail ) followed by a much slower decay and spectral hardening ; the afterglow beginning is chosen at the epoch when the slower flux decay starts . the closure relations @xmath16 listed in table [ alfa ] are used to calculate the required energy injection indices from the measured pre- and post - break x - ray flux decay indices @xmath12 and the x - ray slope @xmath13 ( which gives the electron index @xmath45 ) . for a model to be validated , the resulting energy injection index @xmath32 @xmath205 should be positive ( i.e. @xmath216 , with the uncertainty @xmath217 calculated from the propagation of @xmath218 and @xmath219 ) , and @xmath206 for jet - break models , should lead to a decelerating outflow , so that the jet boundary can become visible to the observer at some time . from equation [ gm ] , this condition is @xmath220 . for adiabatic jets ( @xmath144 ) , the observables @xmath221 and @xmath222 yield two constraints . as there are no free model parameters , aside the various possibilities related to the jet dynamics , location of the cooling frequency , ambient medium type , and dominant x - ray emission process , the model is over - constrained . thus , it should not be surprising that only a small fraction of swift x - ray breaks satisfy @xmath223 both before and after the break , i.e. can be explained by any single variant of the adiabatic jet break model ( upper part of table [ mod ] ) . lcccccccc & & + & @xmath224 & @xmath225 & @xmath224 & @xmath225 + + j syn cjet & 3 & 11 & 0 & 9 + j syn sjet & 0 & 7 & 0 & 7 + j ic cjet & 2 & 15 & 0 & 9 + j ic sjet & 2 & 7 & 0 & 7 + + j syn cjet & 48 & 47 & 12 & 29 + j syn sjet & 19 & 39 & 15 & 42 + j ic cjet & 64 & 52 & 0 & 31 + j ic sjet & 67 & 18 & 0 & 16 + + ei syn cjet & 111 & 98 & 113 & 77 + ei syn sjet & 115 & 106 & 115 & 105 + ei syn sph & 77 & 31 & 106 & 31 + ei ic cjet & 114 & 4 & 116 & 80 + ei ic sjet & 114 & 47 & 114 & 54 + ei ic sph & 105 & 37 & 115 & 36 + [ mod ] evidently , swift x - ray afterglows may not be all of the same forward - shock model type . even when we take into account all these possible variants , we find that adiabatic jet - break model can explain only 30 afterglows ( indicated in boldface in table [ ag1 ] ) , i.e. about 30 percent of the sample considered here . six of those 30 breaks occur in afterglows with two light - curve breaks , the break compatible with an adiabatic jet - break being always the second . for all six , there is always a spherical energy - injection break model with @xmath144 after the break that accounts for the first break , and compatible with the jet - break model , i.e. the two break models have the same location of cooling frequency relative to the x - rays , the same type of external medium , and the same x - ray emission process . for the jet - break model , we require that the x - ray light - curve break arises solely from the outflow collimation , without any contribution from a break in the injected power , i.e. we require a constant energy injection index @xmath32 across the x - ray light - curve break : @xmath226 . thus , the model is still over - constrained , with two observables ( @xmath221 and @xmath222 ) that must be accommodated by a single model parameter ( @xmath32 ) . considering that the jet - break model with steady energy injection injection is less over - constrained than the adiabatic jet model , it may not be surprising that the former model can account for a larger number of swift breaks ( mid part of table [ mod ] ) . with allowance for all jet - break model variants listed in table [ mod ] ( i.e. for either location of the cooling frequency relative to the x - ray , for either radiation process , for either type of medium , and either post jet - break dynamics , conical or spreading ) , the jet - break model with steady energy injection ( including adiabatic jet - breaks , for which the energy injection parameter @xmath32 is consistent with zero ) can account for 103 of the 117 x - ray light - curve breaks considered here ( the unexplained breaks are shown in italics in tables [ ag1 ] ) . 37 of those potential jet - breaks occur in afterglows with two light - curve breaks ; with the exception of the first break of afterglow 081222 , all other 36 jet - breaks can be paired consistently with an energy - injection break for the remaining break , i.e. the two break models have the same medium stratification , the same x - ray radiative process , the same location of x - ray relative to the cooling frequency , and compatible ejecta dynamics . because only one break can arise from the outflow collimation and 19 afterglows in our sample have two light - curve breaks , a more correct way of assessing the success of the jet - break model with steady energy injection is to say that it can explain at least one of the breaks observed for 85 afterglows out of 98 or , conversely , that it can not account for any of the x - ray light - curve breaks of 12 percent of afterglows . excluding the 30 breaks ( of 30 afterglows ) that can be accounted for with an adiabatic jet - break , the jet - break model with non - zero energy injection can explain one x - ray light - curve break for 55 of the 98 afterglows considered here , which argues strongly that energy injection may be often present in swift afterglows . given that 12 percent of afterglows have an x - ray light - curve break that can not be explained with a jet - break even when a steady energy injection is allowed , and that some afterglows display two breaks , out of which only one can be a jet - break , it is natural to speculate that some x - ray light - curve breaks must arise from a change in the power at which energy is added to the forward - shock . in the extreme , one could speculate that all x - ray breaks could be such energy injection breaks , but the large fraction of breaks that can be explained as jet - breaks argues against that extreme speculation . the broken power - law energy injection model is not over - constrained , as there are two observables ( @xmath221 and @xmath222 ) for two model parameters ( @xmath227 and @xmath228 ) , the only restriction being that @xmath229 . thus , it should not be surprising that there are variants of the energy - injection break model that can account for most of the swift x - ray breaks ( lower part of table [ mod ] ) . there is only one x - ray break that can not be explained with an energy injection break : grb afterglow 060607 . from table [ alfa ] , it can be seen that , for @xmath230 , the steepest possible decay for an adiabatic outflow is that for the inverse - compton emission from a conical jet interacting with a wind , and for @xmath231 : @xmath232 . for grb afterglow 060607 , @xmath233 , hence the steepest possible decay for the forward - shock emission has @xmath234 , which is slower than the measured post - break @xmath235 . after having identified the power - law energy injection that accommodates each of the x - ray light - curve breaks , one can calculate the increase in the forward - shock energy from first to last x - ray measurement . the interesting result is that the fractional increase of the forward - shock energy over a source - frame time - interval common to all afterglows ( from 100 s to 100 ks ) , @xmath236 , is anticorrelated with the isotropic - equivalent grb output @xmath237 , for any model variant that can account for a significant fraction of the afterglow breaks listed in table [ ag1 ] . the @xmath238 anticorrelation is illustrated in figure [ f1 ] for two jet - break models and in figure [ f2 ] for two energy - injection break models . for the synchrotron jet - break model ( black points in figure [ f1 ] ) , the linear correlation coefficient in log - log space is @xmath239 , yielding a probability for a stronger anticorrelation @xmath240 in the null hypothesis . for the inverse - compton jet - break model ( red points in figure [ f1 ] ) , @xmath241 and @xmath242 . for the synchrotron energy - injection break model ( black points in figure [ f2 ] ) , @xmath243 , and @xmath244 , while for the inverse - compton energy - injection break model ( red points in figure [ f2 ] ) , @xmath245 , and @xmath246 . similarly high chance probabilities are obtained for other models ; thus the @xmath238 anticorrelation is only tentative " . the above linear correlation coefficients ( pearson s ) @xmath64 have been calculated assuming a 50 percent uncertainty ( in natural scale ) for both quantities plotted , the chance probabilities of those anticorrelations have been estimated conservatively for a linear correlation coefficient @xmath247 . the best - fits shown in figures [ f1 ] and [ f2 ] were obtained by minimizing @xmath211 using the uncertainties of both quantities . that @xmath17 ) the fractional increase of the forward - shock energy over a fixed time interval is determined by the exponent @xmath32 of the power - law energy injection which accommodates the measured x - ray flux decay index @xmath248 , and that @xmath18 ) @xmath249 ( a stronger energy injection yields a slower flux decay ) suggest that the @xmath238 correlation may originate from a @xmath250 correlation . indeed , as illustrated in figure [ f3 ] , more energetic bursts are followed by faster decaying x - ray afterglows . the correlation between these two observables is slightly stronger than the @xmath238 correlation for afterglows with jet - breaks at the x - ray light - curve epoch , but significantly weaker if the x - ray light - curve breaks are energy - injection breaks . the @xmath238 correlation implies that either the efficiency of the grb mechanism or the energy of the ejecta producing the grb emission are anticorrelated with the kinetic energy of the ejecta interacting with the forward - shock after the burst phase . if the dissipation mechanism leading to the production of @xmath251-rays is internal shocks ( rees & meszaros 1994 ) in a fluctuating relativistic wind , then the burst dissipation efficiency is determined by the ratio between the lorentz factors of interacting shells . for an _ impulsive _ ejecta release ( short - lived engine ) , all the ejecta are produced quasi - instantaneously and the energy injection occurs as the inner parts of the outflow catch - up with the ever - decelerating forward - shock ( rees & meszaros 1998 ) . in this model , the @xmath238 correlation implies an anticorrelation between the average shells lorentz factor ratio with the energy of the shells behind the leading edge of the outflow , smoother outflows ( i.e. with a smaller lorentz factor contrast ) having more energy localized toward the trailing edge . for an _ extended _ ejecta release ( long - lived engine ) , where the ejecta are released over a source - frame duration comparable to the observer - frame time during which energy injection takes place , the @xmath238 correlation would imply an anticorrelation among bursts between the lorentz factor contrast in the leading , grb - producing part of the outflow and the kinetic energy of the outflow expelled by the central engine at later times . this is a puzzling implication , which makes the long - lived engine scenario less likely . irrespective of the dissipation mechanism , if the grb output is correlated with the energy of the grb - producing ejecta , then the @xmath238 correlation implies a relationship between the grb ejecta energy and the radial distribution of the afterglow outflow energy . for a short - lived central engine , where the grb and afterglow outflows are the same , the @xmath238 correlation would require that more energetic outflows have less energy behind the leading front . for a long - lived central engine , where the grb outflow is the early afterglow blast - wave , the @xmath238 correlation implies that engines that are initially more energetic release less energy after the burst . perhaps this scenario appears more plausible because it opens the possibility of a constant total ejecta energy . unfortunately , we do not find that @xmath252 is universal ; instead , its distribution extends over at least two decades , for the @xmath253 calculated in any light - curve break model . we note that the detection of x - ray flares in some swift afterglows indicates an extended ejecta release , but the energetic output of those flares is only about 10 percent of the prompt grb output ( falcone et al 2007 ) . in contrast , the energy injection required during the afterglow phase corresponds to an increase of the outflow energy after the burst by a factor up to @xmath254 ( figures [ f1 ] and [ f2 ] ) . to reconcile the flare and afterglow energetics , the internal shocks that produce flares must be very inefficient x - ray radiators compared to the forward - shock ( see also maxham & zhang 2009 ) or , else , the x - ray flares could be produced by the central engine , bearing little correlation with the energy of the ejecta released after the burst . the large increase of the forward - shock energy , by a factor @xmath255 ( fig [ f1 ] ) or @xmath256 ( fig [ f2 ] ) , required to account for the light - curve breaks of swift x - ray afterglows raises the question if the total outflow energy is not too large and incompatible with the expected energy reservoir for a few solar - masses black - hole and its sub solar - mass torus formed in the collapse of a wr massive star . the jet energy can be estimated by assuming that , at the first x - ray measurement ( @xmath30 ) , the isotropic - equivalent forward - shock energy is the grb output @xmath237 . this assumption introduces an underestimation of the jet energy by a factor up to a few because energy could be injected into the forward - shock even before @xmath30 ( 100 1000 s ) , maybe starting as early as the burst end ( 10 100 s ) , at a rate @xmath257 ( this is the range of energy injection laws that accommodate the early x - ray flux decays of table [ ag1 ] ) , and with the estimated jet energy satisfying @xmath258^{3/4}$ ] for a homogeneous medium and @xmath258^{1/2}$ ] for a wind . for a short - lived engine , where all the outflow produces the burst emission and only its leading edge yields the early afterglow , we have @xmath259 ; in this case , the total jet energies calculated here may be substantially overestimated . for a long - lived engine , where the leading part of the outflow produces both the burst and early afterglow emissions , @xmath260 is equivalent to assuming a ( reasonable ) 50 percent grb efficiency . the jet initial aperture @xmath26 , needed to calculate the jet energy from isotropic equivalents , can be constrained as following . for the model of a narrow jet whose edge becomes visible before the first x - ray measurement and the x - ray light - curve break is an energy injection , @xmath31 . for the model where the x - ray light - curve break ( at @xmath261 ) is a jet - break undergoing a steady energy injection , @xmath262 . for the model of a wide jet whose edge becomes visible after the last x - ray measurement ( @xmath28 ) , @xmath29 . prior to the jet - break , the jet lorentz factor is given by the dynamics of a spherical blast - wave : @xmath263^{1/8 } \left ( \frac{t}{z+1}\right)^{-3/8}\ ] ] for a homogeneous medium of particle density @xmath264 per @xmath265 and @xmath266^{1/4 } \left ( \frac{t}{z+1}\right)^{-1/4}\ ] ] for a wind - like medium of density @xmath267 normalized to that corresponding to progenitor mass - loss rate - to - terminal speed ratio of @xmath268 , respectively , with @xmath269 being the blast - wave isotropic - equivalent energy at observer time @xmath96 . the latter is calculated assuming that @xmath260 and for the energy injection law @xmath21 required for each model to account for the measured x - ray flux decay . the observer time in the above equations corresponds to the arrival - time of photons emitted by the fluid moving at an angle @xmath27 relative to the direction toward the observer , from where arises most of the high - energy emission received by the observer at time @xmath96 . putting together all the elements described above , we have calculated the jet final energy ( at @xmath28 ) for the x - ray afterglows whose light - curve breaks can be explained by the a jet - break model with steady energy injection or an energy - injection break model ( the models listed in the mid and lower parts of table [ mod ] ) . the jet energies for @xmath270 and @xmath271 , averaged over all the afterglows that can be accounted for by each model ( fewer than indicated in table [ mod ] , because the jet energy calculation requires the grb output / burst redshift ) , are listed in table [ eng ] . for jet - breaks models , the jet opening is determined by the light - curve break epoch . for energy injection - break models , the epoch of the first ( last ) x - ray measurement provide an upper ( lower ) limit on the jet - break epoch , which yields an upper ( lower ) limit on the jet opening and , consequently , on the jet energy . numbers in round brackets give the dispersion of individual jet energies ( or their lower / upper limits ) . lccccccccc & & + & @xmath224 & @xmath225 & @xmath224 & @xmath225 + + j - syn - cjet & 50.4(0.5 ) & 49.9(0.6 ) & 51.1(0.6 ) & 49.9(0.3 ) + j - syn - sjet & 50.3(0.8 ) & 50.1(0.7 ) & 51.5(0.6 ) & 50.2(0.4 ) + j - ic - cjet & 50.7(0.6 ) & 49.8(0.7 ) & & 50.0(0.3 ) + j - ic - sjet & 50.6(0.7 ) & 49.6(0.6 ) & & 49.7(0.3 ) + + ei - syn - cjet @xmath272 & 50.0(0.6 ) & 49.7(0.6 ) & 51.7(0.7 ) & 49.9(0.5 ) + ei - syn - sjet @xmath272 & 50.8(0.7 ) & 50.2(0.7 ) & 51.7(0.6 ) & 50.3(0.4 ) + ei - syn - sph @xmath273 & 51.0(0.5 ) & 50.6(0.4 ) & 51.8(0.6 ) & 50.6(0.3 ) + ei - ic - cjet @xmath272 & 50.2(0.7 ) & 49.7(0.7 ) & 53.6(0.8 ) & 50.0(0.5 ) + ei - ic - sjet @xmath272 & 50.4(0.7 ) & 48.9(0.8 ) & 55.0(1.3 ) & 49.5(0.6 ) + ei - ic - sph @xmath273 & 51.4(0.6 ) & 50.7(0.5 ) & 53.3(0.8 ) & 50.6(0.3 ) + [ eng ] a potentially important result shown in table [ eng ] is that models where x - ray is inverse - compton , the ambient medium is a wind , and x - ray is lower than the cooling frequency , can be ruled out on energetic grounds because , for them , the required jet total energy exceeds @xmath274 erg . those models lead to a high final jet energy because a substantial increase of the jet energy is required to account for the measured x - ray flux decay rate , which stems from that the corresponding adiabatic jet model yields a steep x - ray flux decay ( table [ alfa ] : @xmath275 , for a conical jet ) , from that energy injection has a weak effect in mitigating the flux decay ( @xmath276 , for the spreading jet model ) , or a combination of those two factors ( for the spherical outflow model ) . assuming the same jet energy at the beginning of the afterglow , one can also calculate the ratios of jet energies among the three types of x - ray break models and outflow dynamics : @xmath205 a narrow - jet model with an energy - injection break ( models e - cjet and e - sjet in table [ eng ] ) , @xmath206 a jet - break model with steady energy injection ( models j - cjet and j - sjet ) , and @xmath207 a wide - jet model with energy - injection break ( model e - sph ) . the interesting result is that , going from type @xmath205 to type @xmath206 models above increases on average ( i.e over afterglows ) the total jet energy by a factor 2 - 3 , with another increase by a factor 2 - 3 required when going from type @xmath206 to @xmath207 models . these results hold for either type of ambient medium , either location of the cooling frequency relative to the x - ray , and for either x - ray radiative process , if the jet does not spread sideways , and only for inverse - compton emission , if the jet is spreading laterally . thus , the most economical model for a light - curve break is that of a narrow jet with an early jet - break and a broken power - law energy injection origin for the x - ray light - curve break . however , these conclusions do not hold for spreading jets if the x - ray radiation process is synchrotron . in this case , we find that the required jet energy is about the same for all model types above . note that these relative energy requirements can not be deduced from the average jet energies listed in table [ eng ] . here , we are referring to averages over afterglows of jet energy ratios between two models that explain the same afterglow break , while table [ eng ] gives jet energies for a given model averaged over all the afterglows that can be explained by that model . the first caveat / weakness of the model testing presented above was already discussed : the energy - injection break origin for x - ray light - curve breaks is only weakly falsifiable by a comparison between the model @xmath16 closure relation and the observed flux power - law decay index @xmath12 and spectral slope @xmath13 . the only way for this model to fail is an x - ray flux that decays steeper than expected for an adiabatic shock , as energy injection can only mitigate the afterglow flux decay . furthermore , a comparison of the model light - curve breaks against observations is needed to validate possible x - ray light - curve break models , as the sharpness of x - ray light - curve breaks varies among afterglows . the majority of the well - sampled light - curve breaks used here are sharp , lasting less than a factor two in time . quite likely , the sharpest breaks can not be accounted by a wind - like medium , whatever is the break mechanism , owing to the slower jet deceleration produced by a decreasing external density ( kumar & panaitescu 2000 ) . without taking into account such limitations , testing only the ability of some models to account for the observed flux power - law decay indices will overestimate the fraction of afterglows that can be explained by a given model . we emphasize the importance of using afterglows whose light - curves have been sampled over a sufficiently long time before and after the break , so that the asymptotic flux decay indices can be measured accurately . otherwise , shortly - monitored light - curves with breaks can lead to an underestimation of the break magnitude @xmath277 , which will favour a steady energy injection interpretation , because its effect is to reduce the break magnitude @xmath278 . we note that eighty percent of the power - law segments used in this work last a decade or more in time . inclusion of the optical light - curves could provide a further test to the origin of x - ray light - curve breaks ( e.g. liang et al 2007 ; curran et al 2009 ; liang et al 2008 ) in the two models discussed above ( jet - break and energy - injection break ) . both models yield _ achromatic _ light - curve breaks , appearing at the same time in all light - curves . for those afterglows where it can be established that the optical and x - ray emissions arise from the same source , either because the light - curve breaks are achromatic or because the optical and x - ray spectra are on the same power - law continuum , testing the light - curve break origin and determining the model details ( location of cooling frequency , ambient medium stratification , radiation process , outflow dynamics ) can be done by comparing the optical flux decay indices @xmath279 with the model expectations , for the energy injection parameter @xmath32 required by the x - ray flux decay indices @xmath248 . in grblog database of optical light - curves ( quimby , mcmahon & jeremy 2004 ) , we find five afterglows ( 050401 , 061121 , 070420 , 090424 , 090618 ) listed in table [ ag1 ] with chromatic x - ray breaks , hence their x - ray and optical emissions arise from different parts of the outflow , and eight with achromatic breaks , thus a comprehensive test of the x - ray break models can not be performed at this time . we note that , of those 8 afterglows with achromatic breaks , two ( 060729 , 081029 ) have @xmath280 , for five ( 050922c , 051109 , 060614 , 080710 , 090424 ) @xmath281 , and for one ( 090510 ) @xmath282 . in general , @xmath283 ( as for 060729 ) is expected if the cooling frequency is not between the optical and x - ray and @xmath284 if @xmath22 is in between . according to table [ spek ] , a decreasing @xmath22 ( which leads to @xmath285 ) , is obtained only for a homogeneous medium , thus all six afterglows with @xmath286 require such an ambient medium . energy injection accelerates the evolution of @xmath22 and increases the index difference @xmath287 above the expectation for an adiabatic outflow : @xmath288 for a spherical / conical outflow and @xmath289 for a spreading jet . thus , the energy injection indices @xmath32 in the range @xmath290 required by the x - ray decays of those six afterglows imply that @xmath291 , consistent with what is measured for five of them , while the large @xmath287 measured for 090510 indicates that the optical and x - ray are from different radiative processes ( synchrotron and inverse - compton , respectively ) . in this work , we have considered only the forward - shock emission and ignored the emission from the reverse - shock , which energizes the grb / post - grb ejecta , the agent of energy injection into the blast - wave . the two shocks dissipate equally the kinetic energy of the ejecta , which indicates that the reverse - shock could make a comparable contribution to the afterglow emission . for a short - lived ejecta source , it can be shown from the kinematics of the catch - up between the grb ejecta and the decelerating forward - shock that the ratio of the lorentz factor @xmath292 of the ejecta arriving at the forward - shock at time @xmath96 to that of the swept - up ambient medium @xmath108 is @xmath293^{1/2 } \gta 1 $ ] ( evidently , the ratio of those two lorentz factors must be above unity , but can not be much above unity , else the catch - up would have occurred earlier ) . in this case , the reverse shock is only mildly relativistic and the ejecta electrons accelerated by the reverse - shock could be much less energetic than the electrons behind the forward - shock , making the emission of the latter dominant at x - ray photon energies . for a long - lived central source , the contrast between the lorentz factor of the unshocked outflow and that of the forward - shock can be arbitrarily large , the reverse - shock could be highly relativistic , provided that the ejecta are not too dense ( e.g. sari & piran 1995 ) . the decay of the reverse - shock flux can be calculated in a way similar to that shown here for the forward - shock ; however , the parametrization of the reverse - shock emission is not that simple because @xmath17 ) its dynamics depends on the lorentz factor of the ejecta and their density , and @xmath18 ) its peak flux depends on the total ejecta mass having entered the reverse - shock . thus , the calculation of the reverse - shock emission requires at least two parameters that quantify the energy and mass mass injection into the blast - wave , leading to a model even less constrainable ( observationally ) than the forward - shock ( which requires only one parameter , @xmath32 , for the energy injection law ) . by comparing the flux decay index ( @xmath12 ) with the spectral slope ( @xmath13 ) measured for a set of 98 well - sampled swift x - ray afterglows with the expectations for the forward - shock model , we find that about a third of those afterglows display a break that could be the traditional ` jet - break ' claimed to have been observed in a dozen pre - swift optical afterglows ( [ adb ] ) . for this test of _ adiabatic _ jet - breaks , we have considered various details of the forward - shock model : homogeneous or wind - like ambient media , spectral cooling frequency higher or lower than the x - rays , synchrotron or inverse - compton as the dominant x - ray process , a collimated conical or spreading outflow . if a steady energy injection into the forward - shock occurs , then the jet - break model may account for 88 percent of the 117 light - curve breaks or , else put , for at least one break in 87 percent of the 98 afterglows selected here ( [ jbei ] ) . this vast improvement over the adiabatic jet model suggests that energy injection could be a prevalent process in grb afterglows . the remaining 12 percent of breaks that can not be explained by a jet - break undergoing a steady energy injection , and the 19 afterglows with two breaks ( only one being , at most , a jet - break ) , indicate that some x - ray breaks may originate from a change in the rate at which energy is added to the forward - shock ( an energy - injection break ) , and it is possible that some ( many ? ) of the identified jet - breaks are also energy - injection breaks ( [ eibr ] ) . it should be recognized that the complete success of the energy - injection break model over the traditional jet - break model could be due mostly to that the latter model is over - constrained ( observations provide two constraints : flux pre- and post - break power - law decay indices for one model parameter : the energy injection power - law exponent ) , while the former model is not . we find a tentative correlation between the grb output and the afterglow x - ray flux decay rate , more energetic bursts being followed by faster decaying afterglows ( [ fag ] ) . if energy injection is a common process in grb afterglows , then the slower decaying afterglows should be identified with a stronger total energy injection . this means that the grb energy afterglow ( x - ray flux ) decay rate correlation should lead to an anticorrelation of the grb output with the increase of the outflow kinetic energy during the afterglow phase , i.e. more energetic bursts should be followed by afterglows with less energy being injected . this anticorrelation is , indeed , observed for both the jet - break and the energy injection break models for x - ray light - curve breaks , and for various model details ( dynamics , x - ray afterglow emission process , radial stratification of the ambient medium ) . if the grb progenitor is a long - lived source of relativistic ejecta , the above anticorrelation implies an anticorrelation of the kinetic energy of the leading outflow ( producing the burst ) with that of the following ejecta ( which inject energy in the blast - wave during the afterglow phase ) , but the total ejecta energy ( burst plus afterglow ) is far from being universal . the energy budget required by most of the possible variants of the forward - shock model ranges from @xmath294 to @xmath295 ergs , and only variants involving inverse - compton emission in the x - rays and a wind - like medium can be excluded on energetic grounds because , in those models , the required jet - energy exceeds @xmath296 erg ( [ jeten ] ) . considering both models for light - curve breaks ( jet - break with steady energy - injection and energy - injection break ) , we find that the most energetically economical model is that of a narrow jet whose jet - break occurs early on ( before first x - ray measurement ) with the light - curve break being an energy - injection break , followed by a model where the x - ray light - curve break is a jet - break , with the most `` wasteful '' model being that of a wide jet whose break occurs late ( after last x - ray measurement ) , the x - ray break being an energy - injection break . however , the average ( over afterglows ) energy ratio among pairs of these three break models is only 23 . this work was supported by an award from the laboratory directed research and development program at the los alamos national laboratory and made use of data supplied by the uk science data center at the university of leicester , by the grb log " site for optical light - curves ( http://grblog.org/grblog.php ) , and by the grb afterglow repository of gcn circulars ( http://www.mpe.mpg.de/@xmath297jcg/grbgen.html ) . 99 chevalier r. , li z. , fransson c. , 2004 , apj , 606 , 369 chincarini g. et al , 2007 , apj , 671 , 1903 curran p. et al , 2009 , mnras , 395 , 580 evans e. et al , 2007 , a&a , 469 , 379 evans e. et al , 2009 , mnras , 397 , 1177 falcone a. et al , 2007 , apj , 671 , 1921 ghisellini g. , ghirlanda g. , nava l. , 2007 , apj , 658 , l75 hawley j. & krolik j. , 2006 , apj , 641 , 103 ioka k. et al , 2006 , a&a 458 , 7 kumar p. , panaitescu a. , 2000 , apj , 541 , l9 liang e. , zhang b .- b . , zhang b. , 2007 , apj , 670 , 565 liang e. et al , 2008 , apj , 675 , 528 van marle a. et al , 2006 , a&a , 460 , 105 maxham a. & zhang b. , 2009 , apj , 707 , 1623 mckinney j. , 2005 , apj , 630 , l5 mszros p. & rees m. , 1997 , apj , 476 , 232 nousek j. et al , 2006 , apj , 642 , 389 nugis t. & lamers h. , 2000 , a&a , 360 , 227 panaitescu a. , mszros p. , rees m. , 1998 , apj , 503 , 314 panaitescu a. et al , 2006 , mnras , 366 , 1357 panaitescu a. , 2007 , mnras , 379 , 331 panaitescu a. , 2008 , mnras , 383 , 1143 panaitescu a. , 2011 , mnras , 414 , 1379 panaitescu a. & vestrand t. , 2011 , mnras , 414 , 3537 quimby r. , mcmahon e. , jeremy m. , 2004 , grbs : 30 years of discovery " , aip conf proc , 727 , 529 racusin j. et al , 2008 , nature , 455 , 183 ramirez - ruiz e. et al , 2005 , apj , 631 , 435 rees m. & mszros p. , 1994 , apj , 430 , l93 rees m. & mszros p. , 1998 , apj , 496 , l1 rhoads j. , 1999 , apj , 525 , 737 sari r. & piran t. , 1995 , apj , 455 , l143 uhm z. & beloborodov a. , 2007 , apj , 665 , l93 willingale r. et al , 2007 , apj , 662 , 1093 zhang b. et al , 2006 , apj , 642 , 354

states in quantum field theory ( qft ) are represented by many - particle wave functions , such that a state describing particles depends on spacetime positions . since a general state is a superposition of states with different numbers of particles , the wave function lives in the configuration space identified with a product of an infinite number of 4-dimensional minkowski spacetimes . the squared absolute value of the wave function is interpreted as the probability density in the configuration space , from which the standard probabilistic predictions of qft can be recovered . such a formulation and probabilistic interpretation of qft allows to interpret the wave function as a pilot wave that describes deterministic particle trajectories , which automatically includes a deterministic and continuous description of particle creation and destruction . in particular , when the conditional wave function associated with a quantum measurement ceases to depend on one of the spacetime coordinates , then the 4-velocity of the corresponding particle vanishes , describing a trajectory that stops at a particular point in spacetime . in a more general situation a dependence on this spacetime coordinate is negligibly small but not strictly zero , in which case the trajectory does not stop but the measuring apparatus still behaves as if this particle has been destroyed . pacs : 11.10.-z , 03.70.+k , 03.65.ta

0904.2287

the bohmian interpretation of nonrelativistic quantum mechanics ( qm ) @xcite is the best known and most successfull attempt to explain quantum phenomena in terms of `` hidden variables '' , that is , objective properties of the system that are well defined even in the absence of measurements . according to this interpretation , particles allways have continuous and deterministic trajectories in spacetime , while all quantum uncertainties are an artefact of the ignorance of the initial particle positions . the wave function plays an auxiliary role , by acting as a pilot wave that determines motions of particles for given initial positions . yet , in its current form , the bohmian interpretation is not without difficulties . an important nontrivial issue is to make the bohmian interpretation of many - particle systems compatible with special relativity . manifestly relativistic - covariant bohmian equations of motion of many - particle systems have been proposed in @xcite and further studied in @xcite , but for a long time it has not been known how to associate probabilistic predictions with such relativistic - covariant equations of motion . recently , a progress has been achieved by realizing that the _ a priori _ probability density @xmath1 of a single particle at the space - position @xmath2 at time @xmath3 should _ not _ be interpreted as a probability density in space satisfying @xmath4 , but as a probability density in _ spacetime _ satisfying @xmath5 @xcite . the usual probabilistic interpretation in space is then recovered as a conditional probability , corresponding to the case in which the time of detection has been observed . for an @xmath0-particle wave function it generalizes to @xmath6 , where @xmath7 and @xmath8 . as shown in @xcite , such a manifestly relativistic - invariant probabilistic interpretation is compatible with the manifestly relativistic - covariant bohmian equations of motion . in this paper we confirm this compatibility , through a more careful discussion in appendix [ appb ] . the main remaining problem is how to make the bohmian interpretation compatible with quantum field theory ( qft ) , which predicts that particles can be created and destructed . how to make a theory that describes deterministic continuous trajectories compatible with the idea that a trajectory may have a singular point at which the trajectory begins or ends ? one possibility is to explicitly break the rule of continuous deterministic evolution , by adding an additional equation that specifies stochastic breaking of the trajectories @xcite . another possibility is to introduce an additional continuously and deterministically evolving hidden variable that specifies effectivity of each particle trajectory @xcite . however , both possibilities seem rather artificial and contrived . a more elegant possibility is to replace pointlike particles by extended strings , in which case the bohmian equation of motion automatically contains a continuous deterministic description of particle creation and destruction as string splitting @xcite . however , string theory is not yet an experimentally confirmed theory , so it would be much more appealing if bohmian mechanics could be made consistent without strings . the purpose of this paper is to generalize the relativistic - covariant bohmian interpretation of relativistic qm @xcite describing a fixed number of particles , to relativistic qft that describes systems in which the number of particles may change . it turns out that the natural bohmian equations of motion for particles described by qft automatically describe their creation and destruction , without need to add any additional structure to the theory . in particular , the new artificial structures that has been added in @xcite or @xcite turn out to be completely unnecessary . the main new ideas of this paper are introduced in sec . [ sec2 ] in a non - technical and intuitive way . this section serves as a motivation for styding the technical details developed in the subsequent sections , but a reader not interested in technical details may be satisfied to read this section only . [ sec3 ] presents in detail several new and many not widely known conceptual and technical results in standard qft that do not depend on the interpretation of quantum theory . as such , this section may be of interest even for readers not interested in interpretations of quantum theory . the main purpose of this section , however , is to prepare the theoretical framework needed for the physical interpretation studied in the next section . sec . [ sec4 ] finally deals with the physical interpretation . the general probabilistic interpretation and its relation to the usual probabilistic rules in practical applications of qft is discussed first , while a detailed discussion of the interpretation in terms of deterministic particle trajectories ( already indicated in sec . [ sec2 ] ) is delegated to the final part of this section . finally , the conclusions are drawn in sec . [ sec5 ] . in the paper we use units @xmath9 and the metric signature @xmath10 . in this section we formulate our main ideas in a casual and mathematically non - rigorous way , with the intention to develop an intuitive understanding of our results , and to motivate the formal developments that will be presented in the subsequent sections . as a simple example , consider a qft state of the form @xmath11 and a 2-particle state @xmath12 . for example , it may represent an unstable particle for which we do not know if it has already decayed into 2 new particles ( in which case it is described by @xmath12 ) or has not decayed yet ( in which case it is described by @xmath13 ) . however , it is known that one allways observes either one unstable particle ( the state @xmath13 ) or two decay products ( the state @xmath12 ) . one never observes the superposition ( [ e2.1 ] ) . to answer this question , let us try with a bohmian approach . one can associate a 1-particle wave function @xmath14 with the state @xmath13 and a 2-particle wave function @xmath15 with the state @xmath12 , where @xmath16 is the spacetime position @xmath17 , @xmath18 , of the particle labeled by @xmath19 . then the state ( [ e2.1 ] ) is represented by a superposition @xmath20 however , the bohmian interpretation of such a superposition will describe _ three _ particle trajectories . on the other hand , we should observe either one or two particles , not three particles . how to explain that ? to understand it intuitively , we find it instructive to first understand an analogous but much simpler problem in bohmian mechanics . consider a single non - relativistic particle moving in 3 dimensions . its wave function is @xmath21 , where @xmath22 and the time - dependence is suppressed . now let us assume that the particle can be observed to move either along the @xmath23 direction ( in which case it is described by @xmath24 ) or on the @xmath25-@xmath26 plane ( in which case it is described by @xmath27 ) . in other words , the particle is observed to move either in one dimension or two dimensions , but never in all three dimensions . but if we do not know which of these two possibilities will be realized , we describe the system by a superposition @xmath28 however , the bohmian interpretation of the superposition ( [ e2.3 ] ) will lead to a particle that moves in all three dimensions . on the other hand , we should observe that the particle moves either in one dimension or two dimensions . the formal analogy with the many - particle problem above is obvious . fortunately , it is very well known how to solve this analogous problem involving one particle that should move either in one or two dimensions . the key is to take into account the properties of the _ measuring apparatus_. if it is true that one allways observes that the particle moves either in one or two dimensions , then the total wave function describing the entanglement between the measured particle and the measuring apparatus is not ( [ e2.3 ] ) but @xmath29 where @xmath30 is a position - variable that describes the configuration of the measuring apparatus . the wave functions @xmath31 and @xmath32 do not overlap . hence , if @xmath30 takes a value @xmath33 in the support of @xmath34 , then this value is not in the support of @xmath35 , i.e. , @xmath36 . consequently , the motion of the measured particle is described by the conditional wave function @xcite @xmath37 . the effect is the same as if ( [ e2.3 ] ) collapsed to @xmath27 . now essentially the same reasoning can be applied to the superposition ( [ e2.2 ] ) . if the number of particles is measured , then instead of ( [ e2.2 ] ) we actually have a wave function of the form @xmath38 the detector wave functions @xmath31 and @xmath32 do not overlap . hence , if the particle counter is found in the state @xmath34 , then the measured system originally described by ( [ e2.2 ] ) is effectively described by @xmath15 . now , what happens with the particle having the spacetime position @xmath39 ? in general , its motion in spacetime may be expected to be described by the relativistic bohmian equation of motion @xcite @xmath40 where @xmath41 is an auxiliary scalar parameter along the trajectory . however , in our case the effective wave function does not depend on @xmath39 , i.e. , the derivatives in ( [ e2.6 ] ) vanish . consequently , all 4 components of the 4-velocity ( [ e2.6 ] ) are zero . the particle does not change its spacetime position @xmath42 . it is an object without an extension not only in space , but also in time . it can be thought of as a pointlike particle that exists only at one instant of time @xmath43 . it lives too short to be detected . effectively , this particle behaves as if it did not exist at all . now consider a more realistic variation of the measuring procedure , taking into account the fact that the measured particles become entangled with the measuring apparatus at some finite time @xmath44 . before that , the wave function of the measured particles is really well described by ( [ e2.2 ] ) . thus , before the interaction with the measuring apparatus , all 3 particles described by ( [ e2.2 ] ) have continuous trajectories in spacetime . all 3 particles exist . but at time @xmath44 , the total wave function significantly changes . either ( i ) @xmath30 takes a value from the support of @xmath34 in which case @xmath45 becomes zero , or ( ii ) @xmath30 takes a value from the support of @xmath35 in which case @xmath46 and @xmath47 become zero . after time @xmath44 , either the particle 1 does not longer change its spacetime position , or the particles 2 and 3 do not longer change their spacetime positions . the effect is the same as if the particle 1 or the particles 2 and 3 do not exist for times @xmath48 . in essence , this is how relativistic bohmian interpretation describes the particle destruction . in order for this mechanism to work , we see that it is essential that each particle possesses not only its own space coordinate @xmath49 , but also its own time coordinate @xmath50 . the corresponding particle trajectories are illustrated by fig . the picture on the left shows the trajectories of particles 1 and 2 in spacetime , for the case in which the particle 1 is destructed at time @xmath44 . the trajectory of the destructed particle looks discontinuous . however , the trajectories of all particles are described by continuous functions @xmath51 with a common parameter @xmath41 , so the set of all 3 trajectories ( because @xmath19 ) in the 4-dimensional spacetime can be viewed as a single continuous trajectory in the @xmath52 dimensional configuration space . the picture on the right of fig . [ fig1 ] demonstrates the continuity on the @xmath53-@xmath54 plane of the configuration space . one may object that the mechanism above works only in a very special case in which the absence of the overlap between @xmath31 and @xmath32 is _ exact_. in a more realistic situation this overlap is negligibly small , but not exactly zero . in such a situation neither of the particles will have exactly zero 4-velocity . consequently , neither of the particles will be really destroyed . nevertheless , the measuring apparatus will still behave as if some particles have been destroyed . for example , if @xmath30 takes value @xmath33 for which @xmath55 , then for all practical purposes the measuring apparatus behaves as if the wave function collapsed to the second term in ( [ e2.5 ] ) . the particles with positions @xmath56 and @xmath57 also behave in that way . therefore , even though the particle with the position @xmath58 is not really destroyed , an effective wave - function collapse still takes place . the influence of the particle with the position @xmath58 on the measuring apparatus described by @xmath33 is negligible , which is effectively the same as if this particle has been destroyed . a `` real '' particle ( having non - zero 4-velocity ) surrounded by a sea of `` vacuum '' particles ( having zero 4-velocities).,width=226 ] of course , the interaction with the measuring apparatus is not the only mechanism that may induce destruction of particles . any interaction with the environment may do that . ( that is why we use the letter @xmath59 to denote the states of the measuring apparatus . ) or more generally , any interactions among particles may induce not only particle destruction , but also particle creation . whenever the wave function @xmath60 does not really vary ( or when this variation is negligible ) with some of @xmath16 for some range of values of @xmath16 , then at the edge of this range a trajectory of the particle @xmath61 may exhibit true ( or apparent ) creation or destruction . in general , a qft state may be a superposition of @xmath0-particle states with @xmath0 ranging from @xmath62 to @xmath63 . thus , @xmath60 should be viewed as a function that lives in the space of infinitely many coordinates @xmath16 , @xmath64 . in particular , the 1-particle wave function @xmath14 should be viewed as a function @xmath65 with the property @xmath66 for @xmath67 . it means that any wave function in qft describes an infinite number of particles , even if most of them have zero 4-velocity . as we have already explained , particles with zero 4-velocity are dots in spacetime . the initial spacetime position of any particle may take any value , with the probability proportional to @xmath68 . thus , the bohmian particle trajectories associated with the 1-particle wave function @xmath65 take a form as in fig . [ fig2 ] . in addition to one continuous particle trajectory , there is also an infinite number of `` vacuum '' particles which live for an infinitesimally short time . it is intuitively clear that a particle that lives for an infinitesimally short time is not observable . however , we have an infinite number of such particles , so could their overall effect be comparable , or even overwhelming , with respect to a finite number of `` real '' particles that live for a finite time ? there is a simple intuitive argument that such an effect should not be expected . the number of `` vacuum '' particles is equal to the cardinal number of the set on natural numbers , denoted by @xmath69 . this set has a measure zero with respect to a continuous trajectory , because a continuous trajectory corresponds to a set of real numbers , the cardinal number of which is @xmath70 @xcite . intuitively , the number of points on a single continuous trajectory is infinitely times larger than the number of points describing the `` vacuum '' particles . consequently , the contribution of the `` vacuum '' particles to any measurable effect is expected to be negligible . the purpose of the rest of the paper is to further elaborate the ideas presented in this section and to put them into a more precise framework . let @xmath71 be some orthonormal basis of 1-particle states . a general normalized 1-particle state is @xmath72 where the normalization condition implies @xmath73 . from the basis @xmath71 one can construct the @xmath0-particle basis @xmath74 , where @xmath75 denotes the symmetrization over all @xmath76 for bosons , or antisymmetrization for fermions . the most general state in qft describing these particles can be written as @xmath77 . now let as assume that the number of particles is measured . it implies that the particles become entangled with the environment , such that the total state describing both the measured particles and the environment takes the form @xmath78 the environment states @xmath79 , @xmath80 are macroscopically distinct . they describe what the observers really observe . when an observer observes that the environment is in the state @xmath79 or @xmath80 , then one says that the original measured qft state is in the state @xmath81 or @xmath82 , respectively . in particular , this is how the number of particles is measured in a state ( [ e3.3 ] ) with an uncertain number of particles . the probability that the environment will be found in the state @xmath79 or @xmath80 is equal to @xmath83 or @xmath84 , respectively . of course , ( [ e3.3 ] ) is not the only way the state @xmath85 can be expanded . in general , it can be expanded as @xmath86 where @xmath87 are some normalized ( not necessarily orthogonal ) states that do not need to have a definite number of particles . a particularly important example are coherent states ( see , e.g. , @xcite ) , which minimize the products of uncertainties of fields and their canonical momenta . each coherent state is a superposition of states with all possible numbers of particles , including zero . the coherent states are overcomplete and not orthogonal . yet , the expansion ( [ e3.5 ] ) may be an expansion in terms of coherent states @xmath87 as well . furthermore , the entanglement with the environment does not necessarily need to take the form ( [ e3.4 ] ) . instead , it may take a more general form @xmath88 where @xmath89 are macroscopically distinct . in principle , the interaction with the environment may create the entanglement ( [ e3.6 ] ) with respect to any set of states @xmath90 . in practice , however , some types of expansions are preferred . this fact can be explained by the theory of decoherence @xcite , which explains why states of the form of ( [ e3.6 ] ) are stable only for some particular sets @xmath90 . in fact , depending on details of the interactions with the environment , in most real situations the entanglement takes either the form ( [ e3.4 ] ) or the form ( [ e3.6 ] ) with coherent states @xmath87 . since coherent states minimize the uncertainties of fields and their canonical momenta , they behave very much like classical fields . this explains why experiments in quantum optics can often be better described in terms of fields rather than particles ( see , e.g. , @xcite ) . in fact , the theory of decoherence can explain under what conditions the coherent - state basis becomes preferred over basis with definite numbers of particles @xcite . there is one additional physically interesting class of sets @xmath90 . they may be eigenstates of the particle number operator defined with respect to _ bogoliubov transformed _ ( see , e.g. , @xcite ) creation and destruction operators . thus , even the vacuum may have a nontrivial expansion of the form latexmath:[\[\label{u3.3 } |0\rangle = c_0 |0'\rangle + \sum_{n=1}^{\infty } \sum_{i'_1,\ldots , i'_n } c_{n;i'_1,\ldots , i'_n } @xmath0-particle states with respect to the bogoliubov transformed number operator . in fact , whenever the two definitions of particles are related by a bogoliubov transformation , the vacuum for one definition of particles is a squeezed state when expressed in terms of particles of the other definition of particles @xcite . thus , if the entanglement with the environment takes the form @xmath92 then the vacuum ( [ u3.3 ] ) may appear as a state with many particles . indeed , this is expected to occur when the particle detector is accelerated or when a gravitational field is present @xcite . the theory of decoherence can explain why the interaction with the environment leads to an entanglement of the form of ( [ u3.4 ] ) @xcite . thus , decoherence induced by interaction with the environment can explain why do we observe either a definite number of particles or coherent states that behave very much like classical fields . however , decoherence alone can not explain why do we observe some particular state of definite number of particles and not some other , or why do we observe some particular coherent state and not some other . consider a free scalar hermitian field operator @xmath93 satisfying the klein - gordon equation @xmath94 the field can be decomposed as @xmath95 where @xmath96 and @xmath97 can be expanded as @xmath98 } , & \nonumber \\ & \hat{\psi}^{\dagger}(x)=\displaystyle\int d^3k \ , f({\bf k } ) \ , \hat{a}^{\dagger}({\bf k } ) e^{i[\omega({\bf k})x^0-{\bf k}{\bf x } ] } . & \end{aligned}\ ] ] here @xmath99 is the @xmath100 component of the 4-vector @xmath101 , and @xmath102 and @xmath103 are the usual creation and destruction operators , respectively . the function @xmath104 is a real positive function which we do not specify explicitly because several different choices appear in the literature , corresponding to several different choices of normalization . all subsequent equations will be written in forms that do not depend on this choice . we define the operator @xmath105 the symbol @xmath106 denotes the symmetrization , reminding us that the expression is symmetric under the exchange of coordinates @xmath107 . ( note , however , that the product of operators on the right hand side of ( [ e3.10 ] ) is in fact automatically symmetric because the operators @xmath108 commute , i.e. , @xmath109=0 $ ] . ) the parameter @xmath110 is a normalization constant determined by the normalization condition that will be specified below . the operator ( [ e3.10 ] ) allows us to define @xmath0-particle states in the basis of particle spacetime positions , as latexmath:[\[\label{e3.11 } |x_{n,1 } , \ldots , x_{n , n}\rangle = \hat{\psi}^{\dagger}_n(x_{n,1 } , \ldots , x_{n , n } ) vacuum @xmath81 , form a complete and orthogonal basis in the hilbert space of physical states . if @xmath112 is an arbitrary ( but normalized ) @xmath0-particle state , then this state can be represented by the @xmath0-particle wave function @xmath113 we also have @xmath114 we choose the normalization constant @xmath110 in ( [ e3.10 ] ) such that the following normalization condition is satisfied @xmath115 however , this implies that the wave functions @xmath116 and @xmath117 , with different values of @xmath0 and @xmath118 , are normalized in different spaces . on the other hand , we want these wave functions to live in the same space , such that we can form superpositions of wave functions describing different numbers of particles . to accomplish this , we define @xmath119 where @xmath120 @xmath121 are volumes of the corresponding configuration spaces . in particular , the wave function of the vacuum is @xmath122 this provides that all wave functions are normalized in the same configuration space as @xmath123 where we use the notation @xmath124 @xmath125 note that the physical hilbert space does not contain non - symmetrized states , such as a 3-particle state @xmath126 . it also does not contain states that do not satisfy ( [ e3.9 ] ) . nevertheless , the notation can be further simplified by introducing an extended kinematic hilbert space that contains such unphysical states as well . every physical state can be viewed as a state in such an extended hilbert space , although most of the states in the extended hilbert space are not physical . in this extended space it is convenient to denote the pair of labels @xmath127 by a single label @xmath61 . hence , ( [ e3.19 ] ) and ( [ e3.20 ] ) are now written as @xmath128 @xmath129 similarly , ( [ e3.16 ] ) with ( [ e3.15 ] ) is now written as @xmath130 the particle - position basis of this extended space is denoted by @xmath131 ( which should be distinguished from @xmath132 which would denote a symmetrized state of an infinite number of physical particles ) . such a basis allows us to write the physical wave function ( [ e3.14 ] ) as a wave function on the extended space @xmath133 now ( [ e3.18 ] ) takes a simpler form @xmath134 the normalization ( [ e3.18 ] ) corresponds to the normalization in which the unit operator on the extended space is @xmath135 while the scalar product is @xmath136 with @xmath137 . a general physical state can be written as @xmath138 it is also convenient to write this as @xmath139 where the tilde denotes a wave function that is not necessarily normalized . the total wave function is normalized , in the sense that @xmath140 implying @xmath141 next , we introduce the operator @xmath142 from the equations above ( see , in particular , ( [ e3.7])-([e3.12 ] ) ) , it is easy to show that @xmath143 satisfies @xmath144 introducing a hermitian number - operator @xmath145 with the property @xmath146 one finds that a general physical state ( [ e3.26 ] ) satisfies the generalized klein - gordon equation @xmath147 we also introduce the generalized klein - gordon current @xmath148 from ( [ e3.31 ] ) one finds that , in general , this current is not conserved @xmath149 where @xmath150 and @xmath151 . from ( [ e3.34 ] ) we see that the current is conserved in two special cases : ( i ) when @xmath152 ( a state with a definite number of physical particles ) , or ( ii ) when @xmath153 ( any physical state of massless particles ) . in the extended hilbert space it is also useful to introduce the momentum picture through the fourier transforms . we define @xmath154 where @xmath155 and @xmath156 corresponds to the number of different values of the label @xmath61 . in the basis of momentum eigenstates @xmath157 we have @xmath158 @xmath159 it is easy to check that the normalizations as above make the fourier transform @xmath160 and its inverse @xmath161 consistent . we can also introduce the momentum operator @xmath162 the wave function ( [ e3.35 ] ) is the momentum eigenstate @xmath163 in particular , the wave function of the physical vacuum is given by ( [ e3.17 ] ) , so @xmath164 we see that ( [ e3.17 ] ) can also be written as @xmath165 showing that the physical vacuum can also be represented as @xmath166 intuitively , it means that the vacuum can be thought of as a state with an infinite number of particles , all of which have vanishing 4-momentum . similarly , an @xmath0-particle state can be thought of as a state with an infinite number of particles , where only @xmath0 of them have a non - vanishing 4-momentum . finally , let us rewrite some of the main results of this ( somewhat lengthy ) subsection in a form that will be suitable for a generalization in the next subsection . a general physical state can be written in the form @xmath167 the corresponding unnormalized @xmath0-particle wave functions are @xmath168 there is a well - defined transformation @xmath169 from the physical hilbert space to the extended hilbert space , so that the general state ( [ s1 ] ) can be represented by a single wave function @xmath170 in this subsection we discuss the generalization of the results of the preceding subsection to the case in which the field operator @xmath171 does not satisfy the free klein - gordon equation ( [ e3.7 ] ) . for example , if the classical action is @xmath172 , \ ] ] then ( [ e3.7 ] ) generalizes to @xmath173 where @xmath174 is the field operator in the heisenberg picture . ( from this point of view , the operator @xmath93 defined by ( [ e3.8 ] ) and ( [ e3.9 ] ) and satisfying the free klein - gordon equation ( [ e3.7 ] ) is the field operator in the interaction ( dirac ) picture . ) thus , instead of ( [ s2 ] ) now we have @xmath175 where @xmath85 and @xmath81 are states in the heisenberg picture . assuming that ( [ s2 ] ) has been calculated ( we shall see below how in practice it can be done ) , the rest of the job is straightforward . one needs to make the transformation ( [ s3 ] ) in the same way as in the free case , which leads to an interacting variant of ( [ s4 ] ) @xmath176 the wave function ( [ s4 ] ) encodes the complete information about the properties of the interacting system . now let us see how ( [ s2 ] ) can be calculated in practice . any operator @xmath177 in the heisenberg picture depending on a single time - variable @xmath3 can be written in terms of operators in the interaction picture as @xmath178 where @xmath179 @xmath180 is some appropriately chosen `` initial '' time , @xmath44 denotes the time ordering , and @xmath181 is the interaction part of the hamiltonian expressed as a functional of field operators in the interaction picture ( see , e.g. , @xcite ) . for example , for the action ( [ e3.44 ] ) we have @xmath182 where @xmath183 denotes the normal ordering . the relation ( [ e3.46 ] ) can also be inverted , leading to @xmath184 thus , the relation ( [ e3.10 ] ) , which is now valid in the interaction picture , allows us to write an analogous relation in the heisenberg picture @xmath185 where @xmath186 by expanding ( [ e3.47 ] ) in powers of @xmath187 , this allows us to calculate ( [ e3.10 ] ) and ( [ s2 ] ) perturbatively . in ( [ s2 ] ) , the states in the heisenberg picture @xmath85 and @xmath81 are identified with the states in the interaction picture at the initial time @xmath188 and @xmath189 , respectively . to demonstrate that such a procedure leads to a physically sensible result , let us see how it works in the special ( and more familiar ) case of the equal - time wave function . it is given by @xmath190 calculated at @xmath191 . thus , ( [ s2 ] ) reduces to @xmath192 using @xmath193 and @xmath194 the expression further simplifies @xmath195 in practical applications of qft in particle physics , one usually calculates the @xmath196-matrix , corresponding to the limit @xmath197 , @xmath198 . for hamiltonians that conserve energy ( such as ( [ e3.48 ] ) ) this limit provides the stability of the vacuum , i.e. , obeys @xmath199 where @xmath200 is some physically irrelevant phase @xcite . essentially , this is because the integrals of the type @xmath201 produce @xmath202-functions that correspond to energy conservation , so the vacuum remains stable because particle creation from the vacuum would violate energy conservation . thus we have @xmath203 the state @xmath204 is not trivial , but whatever it is , it has some expansion of the form @xmath205 where @xmath206 are some coefficients . plugging ( [ e3.55 ] ) and ( [ e3.56 ] ) into ( [ e3.53 ] ) and recalling ( [ e3.10])-([e3.12.1 ] ) , we finally obtain @xmath207 this demonstrates the consistency of ( [ s2 ] ) , because ( [ e3.55.1 ] ) should be recognized as the standard description of evolution from @xmath197 to @xmath208 ( see , e.g. , @xcite ) , showing that the coefficients @xmath209 are the same as those described by standard @xmath196-matrix theory in qft . in other words , ( [ s2 ] ) is a natural many - time generalization of the concept of single - time evolution in interacting qft . in secs . [ sec3.2 ] and [ sec3.3 ] we have discussed in detail scalar hermitian fields , corresponding to spinless uncharged particles . in this subsection we briefly discuss how these results can be generalized to any type of fields and the corresponding particles . in general , fields @xmath210 carry some additional labels which we collectively denote by @xmath211 , so we deal with fields @xmath212 . for example , spin-1 field carries a vector index , fermionic spin-@xmath213 field carries a spinor index , non - abelian gauge fields carry internal indices of the gauge group , etc . thus eq . ( [ e3.10 ] ) generalizes to @xmath214 where @xmath215 is a collective label @xmath216 . the symbol @xmath217 denotes symmetrization ( antisymmetrization ) over bosonic ( fermionic ) fields describing the same type of particles . hence , it is straightforward to make the appropriate generalizations of all results of secs . [ sec3.2 ] and [ sec3.3 ] . for example , ( [ e3.27 ] ) generalizes to @xmath218 with self - explaining notation . to further simplify the notation , we introduce the column @xmath219 and the row @xmath220 . with this notation , the appropriate generalization of ( [ e3.26.1 ] ) can be written as @xmath221 for the case of states that contain fermionic particles , eq . ( [ e3.26.1gen ] ) requires further discussion . as a simple example , consider a 1-particle state describing one electron . in this case , ( [ e3.26.1gen ] ) can be reduced to @xmath222 where @xmath223 is a dirac spinor . in this expression , the quantity @xmath224 must transform as a lorentz scalar . at first sight , it may seem to be in contradiction with the well - known fact that @xmath225 transforms as a time - component of a lorentz vector . however , there is no true contradiction . let us explain . the standard derivation that @xmath226 transforms as a vector @xcite starts from the assumption that the matrices @xmath227 do not transform under lorentz transformations , despite of carrying the index @xmath228 . however , such an assumption is not necessary . moreover , in curved spacetime such an assumption is inconsistent @xcite . in fact , one is allowed to define the transformations of @xmath223 and @xmath227 in an arbitrary way , as long as such transformations do not affect the transformations of measurable quantities , or quantities like @xmath226 that are closely related to measurable ones . thus , it is much more natural to deal with a differently defined transformations of @xmath227 and @xmath223 , such that @xmath227 transforms as a vector and @xmath223 transforms as a scalar under lorentz transformations of spacetime coordinates @xcite . the spinor indices of @xmath227 and @xmath223 are then reinterpreted as indices in an internal space . with such redefined transformations , ( [ e3.58 ] ) is fully consistent . the details of our transformation conventions are presented in appendix [ appa ] . in this subsection we adopt and further develop the probabilistic interpretation introduced in @xcite ( and partially inspired by earlier results @xcite ) . the quantity @xmath229 is naturally interpreted as the probability of finding the system in the ( infinitesimal ) configuration - space volume @xmath230 around a point @xmath231 in the configuration space . indeed , such an interpretation is consistent with our normalization conditions such as ( [ e3.26.1 ] ) and ( [ e3.26.1gen ] ) . in more physical terms , it gives the joint probability that the particle @xmath232 is found at the spacetime position @xmath39 , particle @xmath233 at the spacetime position @xmath234 , etc . similarly , the fourier - transformed wave function @xmath235 defines the probability @xmath236 , which is the joint probability that the particle @xmath232 has the 4-momentum @xmath237 , particle @xmath233 the 4-momentum @xmath238 , etc . as a special case , consider an @xmath0-particle state @xmath239 . it really depends only on @xmath0 spacetime positions @xmath240 . with respect to all other positions @xmath241 , @xmath242 is a constant . thus , the probability of various positions @xmath241 does not depend on @xmath241 ; such a particle can be found anywhere and anytime with equal probabilities . there is an infinite number of such particles . nevertheless , the fourier transform of such a wave function reveals that the 4-momentum @xmath243 of these particles is necessarily zero ; they have neither 3-momentum nor energy . for that reason , such particles can be thought of as `` vacuum '' particles . in this picture , an @xmath0-particle state @xmath244 is thought of as a state describing @xmath0 `` real '' particles and an infinite number of `` vacuum '' particles . to avoid a possible confusion with the usual notions of vacuum and real particles in qft , in the rest of the paper we refer to `` vacuum '' particles as _ dead _ particles and `` real '' particles as _ live _ particles . or let us be slightly more precise : we say that the particle @xmath61 is dead if the wave function in the momentum space @xmath235 vanishes for all values of @xmath245 except @xmath246 . similarly , we say that the particle @xmath61 is live if it is not dead . the properties of live particles associated with the state @xmath143 can also be represented by the wave function @xmath247 . by averaging over physically uninteresting dead particles , ( [ e4.1 ] ) reduces to @xmath248 which involves only live particles . this describes the probability when neither the space positions of detected particles nor times of their detections are known . to relate it with a more familiar probabilistic interpretation of qm , let us consider the special case ; let us assume that the first particle is detected at time @xmath249 , second particle at time @xmath250 , etc . in this case , the detection times are known , so ( [ e4.2 ] ) is no longer the best description of our knowledge about the system . instead , the relevant probability derived from ( [ e4.2 ] ) is the _ conditional _ probability @xmath251 where @xmath252 is the appropriate normalization factor . ( for more details regarding the meaning and limitations of ( [ e4.3 ] ) in the 1-particle case see appendix [ appb ] . ) the probability ( [ e4.3 ] ) is sometimes also postulated as a fundamental axiom of many - time formulation of qm @xcite , but here ( [ e4.3 ] ) is derived from a more fundamental and more general expression ( [ e4.2 ] ) ( which , in turn , is derived from an even more general axiom ( [ e4.1 ] ) ) . an even more familiar expression is obtained by studding a special case of ( [ e4.3 ] ) in which @xmath253 , so that ( [ e4.3 ] ) reduces to @xmath254 where @xmath255 is given by ( [ e4.4 ] ) evaluated at @xmath253 . now let us see how the wave functions representing the states in interacting qft are interpreted probabilistically . consider the wave function @xmath190 given by ( [ s2 ] ) . for example , it may vanish for small values of @xmath256 , but it may not vanish for their large values . physically , it means that these particles can not be detected in the far past ( the probability is zero ) , but that they can be detected in the far future . this is nothing but a probabilistic description of the creation of @xmath0 particles that have not existed in the far past . indeed , the results obtained in sec . [ sec3.3 ] ( see , in particular , ( [ e3.57 ] ) ) show that such probabilities are consistent with the probabilities of particle creation obtained by the standard @xmath196-matrix methods in qft . having developed the probabilistic interpretation , we can also calculate the average values of various quantities . we are particularly interested in average values of the 4-momentum @xmath257 . in general , its average value is @xmath258 where @xmath259 is given by ( [ e3.40 ] ) . if @xmath239 , then ( [ e4.6 ] ) can be reduced to @xmath260 similarly , if the times of detections are known and are all equal to @xmath3 , then the average space - components of momenta are given by a more familiar expression @xmath261 finally , note that ( [ e4.6 ] ) can also be written in an alternative form @xmath262 where @xmath263 is the probability density and @xmath264 here @xmath265 is given by an obvious generalization of ( [ e3.32 ] ) @xmath266 the expression ( [ e4.9 ] ) will play an important role in the next subsection . the idea of the particle - trajectory interpretation is that each particle has some trajectory @xmath51 , where @xmath41 is an auxiliary scalar parameter that parameterizes the trajectories . such trajectories must be consistent with the probabilistic interpretation ( [ e4.1 ] ) . thus , we need a velocity function @xmath267 , so that the trajectories satisfy @xmath268 where the velocity function must be such that the following conservation equation is obeyed @xmath269 = 0.\ ] ] namely , if a statistical ensemble of particle positions in spacetime has the distribution ( [ e4.10 ] ) for some initial @xmath41 , then ( [ e4.13 ] ) and ( [ e4.14 ] ) will provide that this statistical ensemble will also have the distribution ( [ e4.10 ] ) for _ any _ @xmath41 , making the trajectories consistent with ( [ e4.1 ] ) . the first term in ( [ e4.14 ] ) trivially vanishes : @xmath270 . thus , the condition ( [ e4.14 ] ) reduces to the requirement @xmath271 = 0.\ ] ] in addition , we require that the average velocity should be proportional to the average momentum ( [ e4.9 ] ) , i.e. , @xmath272 in fact , the constant in ( [ e4.16 ] ) is physically irrelevant , because it can allways be absorbed into a rescaling of the parameter @xmath41 in ( [ e4.13 ] ) . the physical 3-velocity @xmath273 , @xmath274 , is not affected by such a rescaling . thus , in the rest of the analysis we fix @xmath275 as a first guess , eq . ( [ e4.16 ] ) with ( [ e4.17 ] ) suggests that one could take @xmath276 . however , it does not work in general . namely , from ( [ e4.10 ] ) and ( [ e4.11 ] ) we see that @xmath277 , and we have seen in ( [ e3.33 ] ) that @xmath278 does not need to be conserved . instead , we have @xmath279 = j(\vec{x } ) , \ ] ] where @xmath280 is some function that can be calculated explicitly whenever @xmath281 is known . so , how to find the appropriate function @xmath267 ? the problem of finding @xmath282 is solved in @xcite for a very general case ( see also @xcite ) . since the detailed derivation is presented in @xcite , here we only present the final results . applying the general method developed in @xcite , one obtains @xmath283 , \ ] ] where @xmath284 @xmath285 @xmath286 eqs . ( [ e4.21])-([e4.22 ] ) provide that ( [ e4.19 ] ) obeys ( [ e4.15 ] ) , while ( [ e4.20 ] ) provides that ( [ e4.19 ] ) obeys ( [ e4.16])-([e4.17 ] ) . we note two important properties of ( [ e4.19 ] ) . first , if @xmath287 in ( [ e4.18 ] ) , then @xmath276 . in particular , since @xmath287 for free fields in states with a definite number of particles ( it can be derived for any type of particles analogously to the derivation of ( [ e3.34 ] ) for spinless uncharged particles ) , it follows that @xmath276 for such states . second , if @xmath281 does not depend on some coordinate @xmath288 , then both @xmath289 and @xmath290 . [ to show that @xmath290 , note first that @xmath280 defined by ( [ e4.18 ] ) does not depend on @xmath288 when @xmath281 does not depend on @xmath288 . then the integration over @xmath291 in ( [ e4.21 ] ) produces @xmath292 , which kills the dependence on @xmath288 carried by ( [ e4.22 ] ) ] . this implies that dead particles have zero 4-velocity . the results above show that the relativistic bohmian trajectories are compatible with the spacetime probabilistic interpretation ( [ e4.1 ] ) . but what about the more conventional space probabilistic interpretations ( [ e4.3 ] ) and ( [ e4.5 ] ) ? _ a priori _ , these bohmian trajectories are not compatible with ( [ e4.3 ] ) and ( [ e4.5 ] ) . nevertheless , as discussed in more detail in appendix [ appb ] for the 1-particle case , the compatibility between measurable predictions of the bohmian interpretation and that of the `` standard '' purely probabilistic interpretation restores when the appropriate theory of quantum measurements is also taken into account . having established the general theory of particle trajectories by the results above , now we can discuss particular consequences . the trajectories are determined uniquely if the initial spacetime positions @xmath293 in ( [ e4.13 ] ) , for all @xmath18 , @xmath294 , are specified . in particular , since dead particles have zero 4-velocity , such particles do not really have trajectories in spacetime . instead , they are represented by dots in spacetime , as in fig . [ fig2 ] ( sec . [ sec2 ] ) . the spacetime positions of these dots are specified by their initial spacetime positions . since @xmath295 describes probabilities for particle creation and destruction , and since ( [ e4.14 ] ) provides that particle trajectories are such that spacetime positions of particles are distributed according to @xmath295 , it implies that particle trajectories are also consistent with particle creation and destruction . in particular , the trajectories in spacetime may have beginning and ending points , which correspond to points at which their 4-velocities vanish ( for an example , see fig . [ fig1 ] ) . for example , the 4-velocity of the particle a vanishes if the conditional wave function @xmath296 does not depend on @xmath16 ( where @xmath297 denotes the actual spacetime positions of all particles except the particle @xmath61 ) . one very efficient mechanism of destroying particles is through the interaction with the environment , such that the total quantum state takes the form ( [ e3.4 ] ) . the environment wave functions @xmath298 , @xmath299 do not overlap , so the particles describing the environment can be in the support of only one of these environment wave functions . consequently , the conditional wave function is described by only one of the terms in the sum ( [ e3.4 ] ) , which effectively collapses the wave function to only one of the terms in ( [ e3.3 ] ) . for example , if the latter wave function is @xmath300 , then it depends on only @xmath0 coordinates among all @xmath16 . all other live particles from sectors with @xmath301 become dead , i.e. , their 4-velocities become zero which appears as their destruction in spacetime . more generally , if the overlap between the environment wave functions is negligible but not exactly zero , then particles from sectors with @xmath301 will not become dead , but their influence on the environment will still be negligible , which still provides an effective collapse to @xmath300 . since decoherence is practically irreversible ( due to many degrees of freedom involved ) , such an effective collapse is irreversible as well . another physically interesting situation is when the entanglement with the environment takes the form ( [ e3.6 ] ) , where @xmath87 are coherent states . in this case , the behavior of the environment can very well be described in terms of an environment that responds to a presence of classical fields . this explains how classical fields may appear at the macroscopic level , even when the microscopic ontology is described in terms of particles . since @xmath87 is a superposition of states with all possible numbers of particles , trajectories of particles from sectors with different numbers of particles coexist ; there is an infinite number of live particle trajectories in that case . similarly , an entanglement of the form of ( [ u3.4 ] ) explains how accelerated detectors and detectors in a gravitational field may detect particles in the vacuum . for example , let us consider the case of a uniformly accelerated detector . in this case , @xmath302 corresponds to the rindler vacuum , while @xmath81 is referred to as the minkowski vacuum @xcite . the particle trajectories described by ( [ e4.13 ] ) are those of the minkowski particles . the interaction between the minkowski vacuum and the accelerated detector creates new minkowski particles . for instance , if the detector is found in the state @xmath79 , then the minkowski particles are in the state @xmath302 , which is a squeezed state describing an infinite number of live particle trajectories . such a view seems particularly appealing from the point of view of recently discovered renormalizable horava - lifshitz gravity @xcite that contains an absolute time and thus a preferred definition of particles in a classical gravitational background @xcite . let us also give a few remarks on measurements of 4-momenta and 4-velocities . if @xmath303 in ( [ e3.1 ] ) denote the 4-momentum eigenstates , then ( [ e3.4 ] ) describes a measurement of the particle 4-momenta . since the 4-momentum eigenstates are also the 4-velocity eigenstates , ( [ e3.4 ] ) also describes a measurement of the particle 4-velocities . thus , as discussed also in more detail in @xcite , even though the bohmian particle velocities may exceed the velocity of light , they can not exceed the velocity of light when their velocities are measured . instead , the effective wave function associated with such a measurement is a momentum eigenstate of the form of ( [ e3.35 ] ) , where @xmath304 for live particles and @xmath246 for dead particles . this also explains why dead particles are not seen in experiments : their 3-momenta and energies are equal to zero . finally , we want to end this subsection with some conceptual remarks concerning the physical meaning of the parameter @xmath41 . this parameter can be thought of as an evolution parameter , playing a role similar to that of the absolute newton time @xmath3 in the usual formulation of nonrelativistic bohmian interpretation @xcite . to make the similarity with such a usual formulation more explicit , it may be useful to think of @xmath41 as a coordinate parameterizing a `` fifth dimension '' that exists independently of other 4 dimensions with coordinates @xmath305 . however , such a 5-dimensional view should not be taken too literally . in particular , while the time @xmath3 is measurable , the parameter @xmath41 is not measurable . given the fact that @xmath41 is not measurable , what is the physical meaning of the claim that particles have the distribution @xmath306 at some @xmath41 ? we can think of it in the following way : we allways measure spacetime positions of particles at some values of @xmath41 , but we do not know what these values are . consequently , the probability density that describes our knowledge is described by @xmath306 averaged over all possible values of @xmath41 . however , since @xmath306 does not depend on @xmath41 , the result of such an averaging procedure is trivial , giving @xmath306 itself . in this paper we have extended the bohmian interpretation of qm , such that it also incorporates a description of particle creation and destruction described by qft . unlike the previous attempts @xcite and @xcite to describe the creation and destruction of pointlike particles within the bohmian interpretation , the approach of the present paper incorporates the creation and destruction of pointlike particles _ automatically _ , without adding any additional structure not already present in the equations that describe the continuous particle trajectories . one reason why it works is the fact that we work with a many - time wave function , so that the 4-velocity of each particle may vanish separately . even though the many - time wave function plays a central role , we emphasize that the many - time wave function , first introduced in @xcite , is a natural concept when one wants to treat time on an equal footing with space , even if one does not have an ambition to describe particle creation and destruction @xcite . another , even more important reason why it works is the entanglement with the environment , which explains an effective wave function collapse into particle - number eigenstates ( or some other eigenstates ) even when particles are not really created or destroyed . as a byproduct , in this paper we have also obtained many technical results that allow to represent qft states with uncertain number of particles in terms of many - time wave functions . these results may be useful by they own , even without the bohmian interpretation ( see , e.g. , @xcite ) . this work was supported by the ministry of science of the republic of croatia under contract no . 098 - 0982930 - 2864 . at each point of spacetime , one can introduce the tetrad @xmath307 , which is a collection of four spacetime vectors , @xmath308 . the bar on @xmath309 denotes that @xmath309 is not a spacetime - vector index , but only a label . by contrast , the index @xmath310 is a spacetime - vector index . the tetrad is chosen so that @xmath311 where @xmath312 is the spacetime metric and @xmath313 are components of a matrix equal to the minkowski metric . the spacetime - vector indices are raised and lowered by @xmath312 and @xmath314 , respectively , while @xmath309-labels are raised and lowered by @xmath315 and @xmath313 , respectively . thus , ( [ a1 ] ) can also be inverted as @xmath316 now let @xmath317 be the standard dirac matrices @xcite . from them we define @xmath318 the spinor indices carried by matrices @xmath317 and @xmath319 are interpreted as indices of the spinor representation of the _ internal _ group so(1,3 ) . thus , @xmath317 transform as scalars under spacetime coordinate transformations . similarly , the spinors @xmath320 are also scalars with respect to spacetime coordinate transformations , while their spinor indices are indices in the internal group so(1,3 ) . likewise , @xmath321 is also a scalar with respect to spacetime coordinate transformations . it is also convenient to define the quantity @xmath322 which is also a scalar with respect to spacetime coordinate transformations . thus we see that the quantities @xmath323 are both scalars with respect to spacetime coordinate transformations , and that the quantities @xmath324 are both vectors with respect to spacetime coordinate transformations . finally , we note that the relation with the more familiar ( but less general ) formalism in minkowski spacetime @xcite can be established by using the fact that in flat spacetime there is a global choice of coordinates in which @xmath325 however , ( [ a7 ] ) is not a covariant expression , but is only valid in one special system of coordinates . in this appendix we present a more carefull discussion of the meaning of our probabilistic interpretation that treats time on an equal footing with space . the first subsection deals with the pure probabilistic interpretation that does not assume the existence of particle trajectories , while the second subsection deals with the bohmian interpretation in which the existence of particle trajectories is also taken into account . for simplicity , we study the case of 1 particle only , while the generalization to many particles is straightforward . in a pure probabilistic interpretation of qm , one does not assume the existence of particle trajectories . instead , the main axiom is that @xmath320 determines the spacetime probability density @xmath326 this probability describes a statistical ensemble consisting of a large number of events , where each event is an appearance of the particle at the spacetime point @xmath327 . now , if we pick up a subensemble consisting of all events @xmath327 that have the same time coordinate @xmath328 , then the distribution in this subensemble is given by the conditional probability @xmath329 where the normalization factor @xmath330 is equal to the marginal probability that the particle from the initial ensemble will have @xmath328 . ( in fact , when @xmath320 is a superposition of plane waves with positive frequencies only , then @xmath331 does not depend on @xmath3 @xcite . ) a more formal way to understand ( [ appb1 ] ) is as follows . we start from the identity @xmath332 formally , it is convenient to use a discrete notation @xmath333 where the function @xmath334 can be thought of as an eigenstate of the operator @xmath327 with the eigenvalue @xmath335 . thus , ( [ appb3 ] ) can be written as @xmath336 the probability that the particle will be found in the eigenstate @xmath334 is equal to @xmath337 . recalling ( [ appb4 ] ) , this again leads to ( [ appb1 ] ) . yet another way to understand ( [ appb1 ] ) and ( [ appb2 ] ) is through the theory of _ ideal _ quantum measurements . the measured particle with the position @xmath327 entangles with the measuring apparatus described by a pointer position @xmath30 . thus , instead of ( [ appb5 ] ) we have the total wave function @xmath338 where @xmath339 are normalized apparatus states that do not overlap in the @xmath30-space . the main axiom ( [ appb1 ] ) now includes @xmath30 as well , so that ( [ appb1 ] ) generalizes to @xmath340 . therefore , the probability that @xmath30 will have a value from the support of @xmath339 is equal to @xmath337 , which again leads to ( [ appb1 ] ) . ( [ appb2 ] ) emerges from ( [ appb1 ] ) as a conditional probability within the statistical ensemble of measurement outcomes . of course , a real experiment is not ideal . if the departure from ideality is small , then the distributions of measurement outcomes ( [ appb1 ] ) and ( [ appb2 ] ) are good approximations . however , a real experiment may in fact be far from being ideal . in particular , in a real experiment the functions @xmath334 ( which are localized in both space and time ) in ( [ appb6 ] ) may be replaced by functions @xmath341 which are well localized in space , but very extended in time . in such an experiment , ( [ appb6 ] ) is replaced by an entanglement of the form @xmath342 for definiteness , @xmath341 may be taken to be time independent and proportional to @xmath343 for @xmath344 $ ] , but vanishing for other values of @xmath3 . now the measuring apparatus measures the space position @xmath2 of the particle very well , but time @xmath3 at which the particle attains this position remains uncertain with uncertainty equal to @xmath345 . instead of ( [ appb2 ] ) , the measured distribution of particle positions is now @xmath346 where @xmath347 and @xmath348 is the normalization factor chosen such that @xmath349 . namely , the probability that @xmath30 will have a value from the support of @xmath350 is given by ( [ appb2 ] ) . we also stress that the difference between ( [ appb2 ] ) and ( [ appb2 ] ) is irrelevant in most practical situations . namely , in actual experiments one usually deals with wave functions that are well approximated by energy eigenstates with a trivial time dependence proportional to @xmath351 . consequently , @xmath352 is almost independent on @xmath3 , which makes ( [ appb2 ] ) and ( [ appb2 ] ) practically indistinguishable . now let us assume that particles have trajectories @xmath353 where @xmath354 is such that @xmath355 are such trajectories compatible with the probabilistic predictions studied in sec . [ secappb1 ] ? clearly , they are compatible with ( [ appb1 ] ) because @xmath356 but are they compatible with ( [ appb2 ] ) ? to answer this question , it is useful to eliminate @xmath41 from ( [ appb7 ] ) by introducing velocities @xmath357 in general , ( [ appb7 ] ) implies that @xmath358 inequality ( [ appb10 ] ) shows that , _ a priori _ , ( [ appb7 ] ) is not compatible with ( [ appb2 ] ) ( see also @xcite ) . nevertheless , the compatibility restores when the theory of quantum measurements is also taken into account . namely , we extend ( [ appb7 ] ) and ( [ appb7 ] ) such that @xmath359 also satisfies a bohmian equation of motion compatible with @xmath340 . for ideal quantum measurements , ( [ appb6 ] ) implies that the probability that @xmath33 will take a value from the support of @xmath339 is equal to @xmath337 . this means that the statistics of measurement outcomes is given by ( [ appb1 ] ) . thus , ( [ appb2 ] ) emerges from ( [ appb1 ] ) as a conditional probability within the statistical ensemble of measurement outcomes . similarly , in a more realistic measurement based on ( [ appb6 ] ) , the probability that @xmath33 will take a value from the support of @xmath350 is given by ( [ appb2 ] ) . thus we see that the space probability density ( [ appb2 ] ) does not necessarily need to be correct . instead , the space probability density depends on how exactly it is measured . however , the important points are ( i ) that the space probability density can in principle be predicted from the fundamental axiom ( [ appb1 ] ) when the measuring procedure is well defined , and ( ii ) that the probabilistic predictions of the bohmian interpretation agree with those of the `` standard '' ( pure probabilistic ) interpretation for ideal measurements based on ( [ appb6 ] ) , as well as for more realistic measurements based on ( [ appb6 ] ) . d. bohm , phys . rev . * 85 * , 166 ( 1952 ) . d. bohm , phys . rev . * 85 * , 180 ( 1952 ) . d. bohm and b. j. hiley , phys . rep . * 144 * , 323 ( 1987 ) . p. r. holland , _ the quantum theory of motion _ ( cambridge university press , cambridge , 1993 ) . k. berndl , d. drr , s. goldstein , and n. zangh , phys . a * 53 * , 2062 ( 1996 ) . h. nikoli , found . * 18 * , 549 ( 2005 ) . h. nikoli , aip conf . proc . * 844 * , 272 ( 2006 ) [ quant - ph/0512065 ] . h. nikoli , hep - th/0702060 , to appear in found . h. nikoli , int . j. quantum inf . * 7 * , 595 ( 2009 ) [ arxiv:0811.1905 ] . d. drr , s. goldstein , r. tumulka , and n. zangh , j. phys . a * 36 * , 4143 ( 2003 ) . d. drr , s. goldstein , r. tumulka , and n. zangh , phys . lett . * 93 * , 090402 ( 2004 ) . h. nikoli , found . * 17 * , 363 ( 2004 ) . h. nikoli , found . 18 * , 123 ( 2005 ) . d. drr , s. goldstein , and n. zangh , j. stat . phys . * 67 * , 843 ( 1992 ) . r. penrose , _ the road to reality _ ( jonathan cape , london , 2004 ) . l. e. ballentine , _ quantum mechanics : a modern development _ ( world scientific publishing , singapore , 2000 ) . m. schlosshauer , _ decoherence and the quantum - to - classical transition _ ( springer , berlin , 2007 ) . o. kbler and h. d. zeh , ann . phys . * 76 * , 405 ( 1973 ) . j. r. anglin and w. h. zurek , phys . d * 53 * , 7327 ( 1996 ) . n. d. birrell and p. c. w. davies , _ quantum fields in curved space _ ( cambridge press , new york , 1982 ) . l. p. grishchuk and y. v. sidorov , phys . d * 42 * , 3413 ( 1990 ) . b. l. hu and a. matacz , phys . d * 49 * , 6612 ( 1994 ) . j. audretsch , m. mensky , and r. mller , phys . d * 51 * , 1716 ( 1995 ) . p. kok and u. yurtsever , phys . d * 68 * , 085006 ( 2003 ) . cheng and l .- f . li , _ gauge theory of elementary particle physics _ ( clarendon press , oxford , 1984 ) . j. d. bjorken and s. d. drell , _ relativistic quantum fields _ ( mcgraw - hill book company , new york , 1965 ) . j. d. bjorken and s. d. drell , _ relativistic quantum mechanics _ ( mcgraw - hill book company , new york , 1964 ) . e. c. g. stckelberg , helv . acta * 14 * , 322 ( 1941 ) . e. c. g. stckelberg , helv . acta * 14 * , 588 ( 1941 ) . l. p. horwitz and c. piron , helv . acta * 46 * , 316 ( 1973 ) . a. kyprianidis , phys . rep . * 155 * , 1 ( 1987 ) . j. r. fanchi , found . * 23 * , 487 ( 1993 ) . s. tomonaga , prog . . phys . * 1 * , 27 ( 1946 ) . w. struyve and a. valentini , j. phys . a * 42 * , 035301 ( 2009 ) . p. horava , phys . d * 79 * , 084008 ( 2009 ) . h. nikoli , arxiv:0904.3412 . h. nikoli , phys . b * 678 * , 218 ( 2009 ) . h. nikoli , found . phys . * 38 * , 869 ( 2008 ) .

we give a construction of quandle cocycles from group cocycles , especially , for any integer , quandle cocycles of the dihedral quandle from group cocycles of the cyclic group . we will show that a group 3-cocycle of gives rise to a non - trivial quandle 3-cocycle of . when is an odd prime , since , our 3-cocycle is a constant multiple of the mochizuki 3-cocycle up to coboundary . dually , we construct a group cycle represented by a cyclic branched covering branched along a knot from the quandle cycle associated with a colored diagram of .

1012.3729

a quandle , which was introduced by joyce @xcite , is an algebraic object whose axioms are motivated by knot theory and conjugation in a group . in @xcite , carter , jelsovsky , kamada , langford and saito introduced a quandle homology theory , and they defined the quandle cocycle invariants for classical knots and surface knots . the quandle homology is defined as the homology of a chain complex generated by cubes whose edges are labeled by elements of a quandle . on the other hand the group homology is defined as the homology of a chain complex generated by tetrahedra whose edges are labeled by elements of a group . so it is natural to ask a relation between quandle homology and group homology . this question also arises from the fact that the quandle cocycle invariants were defined as an analogue of the dijkgraaf - witten invariants , which are defined using group cocycles . in @xcite , the authors defined a simplicial version of quandle homology and constructed a homomorphism from the usual quandle homology to the simplicial quandle homology . the important point is that this homomorphism gives a triangulation of a knot complement in algebraic fashion . this enables us to relate the quandle homology to the topology of knot complements . in this paper , we apply the results of @xcite to construct quandle cocycles from group cocycles . first , we demonstrate how to give a quandle cocycle of the dihedral quandle @xmath1 from a group cocycle of the cyclic group @xmath2 for any integer @xmath0 in section [ sec : dihedral ] . we will show that a generator of @xmath6 gives rise to a non - trivial quandle 3-cocycle of @xmath7 . when @xmath3 is an odd prime , since @xmath4 , our quandle 3-cocycle is equal to a constant multiple of the mochizuki 3-cocycle @xcite up to coboundary . then we generalize the construction to wider classes of quandles . let @xmath8 be a group and @xmath9 be an element of @xmath8 , then the set @xmath10 forms a quandle by conjugation . ( it is known by joyce @xcite that any faithful homogeneous quandle has such a presentation . ) when some obstruction in second cohomology vanishes , we will construct a quandle cocycle of @xmath11 from a group cocycle of @xmath8 . dually , we relate the quandle cycle associated with an arc and region coloring ( shadow coloring ) of a knot @xmath5 to a group cycle represented by a cyclic branched covering branched along @xmath5 . let @xmath12 be a diagram of @xmath5 . we can define the notion of arc and region colorings of @xmath12 by a quandle @xmath11 . a pair of an arc and a region coloring is called a _ shadow coloring_. we can associate a cycle of a quandle homology group to a shadow coloring of @xmath12 . using the homomorphism constructed in @xcite , we will construct a group cycle of @xmath8 represented by a cyclic branched covering branched along @xmath5 . this reveals a close relationship between the shadow cocycle invariant of a knot and the dijkgraaf - witten invariant of the cyclic branched cover . we remark that hatakenaka and nosaka defined an invariant of 3-manifolds called the _ 4-fold symmetric quandle homotopy invariant _ @xcite , based on the fact that any 3-manifold can be represented as a 4-fold simple branched covering of @xmath13 along a link . as an application , they showed that the shadow cocycle invariant of a link for the mochizuki @xmath14-cocycle is equal to a scalar multiple of the dijkgraaf - witten invariant of the double branched cover along the link . this paper is organized as follows . in section [ sec : group_homology ] , we recall the definition of group homology and show how to represent a group cycle by a triangulation with a labeling of its @xmath15-simplices . we give a presentation of the fundamental group of a cyclic branched covering branched along a knot in section [ sec : pres_of_cyc_bra_cov ] , which is independent from the other sections . in section [ sec : cyc_rep_by_cyc_bra_cov ] , we construct a group cycle represented by a cyclic branched cover . we recall the definition of quandles and their homology theory in section [ sec : quandle_and_homology ] . we review some results from @xcite in section [ sec : main_map ] and apply them to construct quandle cocycles of the dihedral quandle @xmath1 in section [ sec : dihedral ] . the reader who is interested in the form of the @xmath14-cocycle , consult ( [ eq : the_cocycle ] ) ( and ( [ eq : formula_of_d ] ) ) . we will generalize the construction to wider classes of quandles in section [ sec : gen_const ] . in section [ sec : quandle_cyc_as_cyc_bra_cov ] , we construct a group cycle represented by a cyclic branched covering from the quandle cycle associated with a shadow coloring . the reader who is only interested in the construction of quandle cocycles from group cocycles may skip sections [ sec : group_homology ] [ sec : cyc_rep_by_cyc_bra_cov ] except [ subsec : group_homology ] . * acknowledgement * the author thanks takefumi nosaka for useful discussions . he also thanks the referees for helpful comments to improve the exposition . the author is supported by jsps research fellowships for young scientists . in this section , we collect basic facts on group homology . in [ subsec : group_homology ] , we review the definition of group homology . we refer to @xcite for details . the material discussed in [ subsec : cycles_represented_by_triangulations ] and [ subsec : group_cycles_and_representations ] was developed in @xcite . let @xmath8 be a group . let @xmath16 be the free @xmath17$]-module generated by @xmath18-tuples @xmath19 $ ] of elements of @xmath8 . define the boundary map @xmath20 by @xmath21 ) = & g_1[g_2|\dots|g_n ] \\ & + \sum_{i=1}^{n-1}(-1)^i [ g_1|\dots|g_ig_{i+1}|\dots|g_n ] + ( -1)^n[g_1|\dots|g_{n-1 } ] . \end{split}\ ] ] we remark that the chain complex @xmath22 is acyclic , where @xmath23 \to \mathbb{z}$ ] is the augmentation map . so the chain complex @xmath24 gives a free resolution of @xmath25 . let @xmath26 be a right @xmath17$]-module . the homology of @xmath27 } c_n(g)$ ] is called the _ group homology _ of @xmath26 and denoted by @xmath28 . in other words , @xmath29}(m , \mathbb{z})$ ] . let @xmath30 be the free @xmath25-module generated by @xmath31 . then @xmath30 is a left @xmath17$]-module by the action @xmath32 . define the boundary operator of @xmath30 by @xmath33 @xmath24 and @xmath34 are isomorphic as chain complexes . in fact , the following correspondence gives an isomorphism : @xmath35 & \leftrightarrow ( 1,g_1,g_1g_2 , \dots , g_1 \cdots g_n ) \\ ( \quad g_0[g_0^{-1}g_1 | g_1^{-1}g_2| \dots | g_{n-1}^{-1 } g_n ] & \leftrightarrow ( g_0,\dots , g_n ) \quad ) \end{split}\ ] ] the notation using @xmath36 is called _ homogeneous _ and the one using @xmath37 $ ] is called _ inhomogeneous_. factoring out @xmath16 by the degenerate subcomplex , that is generated by @xmath37 $ ] such that @xmath38 for some @xmath39 , we obtain the _ normalized _ chain complex and its homology group . it is known that the group homology using the normalized chain complex coincides with the homology using the unnormalized one . in the homogeneous notation , we factor out @xmath30 by the subcomplex generated by @xmath40 such that @xmath41 for some @xmath39 to define the normalized chain complex . for a left @xmath17$]-module @xmath42 , the group cohomology @xmath43 is defined as the cohomology of the cochain complex @xmath44}(c_n(g),n)$ ] . let @xmath45 be an abelian group . a cocycle of @xmath46 in the homogeneous notation is a function @xmath47 satisfying the following conditions : 1 . @xmath48 , 2 . @xmath49 for any @xmath50 . ( left invariance ) if @xmath51 also satisfies 1 . @xmath52 if @xmath53 for some @xmath39 , then @xmath51 is a normalized @xmath18-cocycle . we can show that any @xmath18-cocycle is cohomologous to a normalized @xmath18-cocycle . let @xmath54 be an @xmath18-dimensional simplex . we label the vertices of @xmath54 by @xmath55 . a face of @xmath54 is presented by a subset of @xmath56 . let @xmath57 be the face spanned by @xmath58 with a vertex ordering given by @xmath59 . any face inherits a vertex ordering from the vertex ordering of @xmath54 i.e. @xmath57 with @xmath60 . let @xmath61 be a cw - complex obtained by gluing a finite number of @xmath18-dimensional simplices along their @xmath62-dimensional faces in pairs by simplicial homeomorphisms . we denote the @xmath63-skeleton of @xmath61 by @xmath64 . we assume that the gluing maps preserve the vertex orderings of the faces . then @xmath65 is homeomorphic to a topological @xmath18-manifold ( not orientable in general ) . when @xmath65 is oriented , we call @xmath61 an _ ordered @xmath18-cycle_. consider an @xmath18-cycle @xmath66 of @xmath67 . then @xmath66 is represented by a sum @xmath68\ ] ] where @xmath69 and @xmath70 . for each @xmath71 $ ] , take an @xmath18-simplex @xmath72 . then label the edge @xmath73 of @xmath72 by @xmath74 for @xmath75 . in particular , we have : @xmath76 we denote the label of @xmath77 by @xmath78 . for @xmath79 , label the oriented edge @xmath77 by @xmath80 . for any 2-dimensional face @xmath81 , we have @xmath82 . rewriting them in the homogeneous notation , we assign labels to the vertices of @xmath72 as @xmath83 up to the left action of @xmath8 ( figure [ fig : labeling ] ) . since @xmath84 , @xmath62-dimensional faces cancel in pairs . gluing @xmath72 s along their faces according to such pairings , we obtain an @xmath18-cycle @xmath61 . at any @xmath62-simplex of @xmath61 , there exist exactly two adjacent @xmath18-simplices . the labelings of the @xmath62-simplex derived from these two @xmath18-simplices coincide . thus we have a well - defined labeling of @xmath15-simplices @xmath85 satisfying , 1 . [ item : circle ] @xmath82 for any 2-dimensional face @xmath81 , 2 . [ item : inverse ] @xmath86 . we call a labeling of @xmath15-simplices satisfying the conditions ( [ item : circle ] ) and ( [ item : inverse ] ) _ @xmath8-valued @xmath15-cocycle_. orient @xmath72 positive if @xmath87 and negative if @xmath88 . since these orientations agree on face pairings , thus we have an orientation on @xmath61 . therefore @xmath61 is an ordered @xmath18-cycle with a @xmath8-valued @xmath15-cocycle @xmath89 . in general , @xmath61 may not be connected , but we assume that @xmath61 is connected because we can treat each connected component separately in our arguments . conversely any ordered @xmath18-cycle @xmath61 with a @xmath8-valued @xmath15-cocycle @xmath89 represents an @xmath18-cycle of @xmath67 . suppose a cycle @xmath90 is represented by an ordered @xmath18-cycle @xmath61 with a @xmath8-valued @xmath15-cocycle @xmath89 . then @xmath66 induces a homomorphism from @xmath91 to @xmath8 as follows . let @xmath92 be the universal covering of @xmath61 and @xmath93 be the covering map . then the simplices of @xmath61 lift to simplices of @xmath92 and each lift has an induced vertex ordering compatible with adjacent @xmath18-simplices . the @xmath8-valued @xmath15-cocycle @xmath89 of @xmath61 induces a @xmath8-valued @xmath15-cocycle of @xmath92 . consider a `` fundamental domain '' of @xmath61 , that is a contractible subcomplex @xmath12 of @xmath92 such that * @xmath94 , * @xmath95 , for any @xmath96 , where we regard @xmath91 as the group of deck transformations . by definition , the number of @xmath18-simplices in @xmath12 coincides with the number of @xmath18-simplices in @xmath61 , in particular finite . we fix a base point @xmath97 in the interior of @xmath12 . each @xmath62-simplex on @xmath98 is glued to another @xmath62-simplex on @xmath98 . we denote such pair of faces by @xmath99 . let @xmath100 be a path in @xmath61 which starts at @xmath101 , traverses @xmath102 in the direction from @xmath103 to @xmath104 and ends at @xmath105 . these paths form a system of generators of the fundamental group @xmath106 . the relations are given at any @xmath107-simplices , around which there are a finite number of @xmath102 s . fix an @xmath18-simplex @xmath54 in @xmath12 and a labeling of vertices of @xmath54 derived from @xmath89 . then @xmath18-simplices adjacent to @xmath54 inherit labelings of vertices from @xmath89 . in this way , all vertices of @xmath12 are labeled by elements of @xmath8 . now consider the labeling of vertices of @xmath108 and @xmath109 . since these reduce to the same labeling of edges , they coincide up to left multiplication . therefore there exists an element of @xmath8 which sends the labeling of vertices of @xmath109 to the one of @xmath108 . denote the element by @xmath110 . this @xmath111 induces a homomorphism @xmath112 . conversely if we have an ordered @xmath18-cycle @xmath61 and a homomorphism @xmath112 , we can construct a @xmath8-valued @xmath15-cocycle @xmath89 and then a cycle of @xmath67 up to boundary as follows . since @xmath111 induces a map @xmath113 , we obtain a labeling of @xmath15-simplices @xmath89 of @xmath61 . this gives rise to a @xmath8-valued @xmath15-cocycle @xmath89 and a homology class in @xmath114 . the @xmath8-valued @xmath15-cocycle @xmath89 is well - defined up to the coboundary action . a map @xmath115 acts on a @xmath8-valued @xmath15-cocycle @xmath89 as a coboundary action by @xmath116 we can show that the homology class obtained from @xmath89 does not change under the coboundary action . for any @xmath50 , the cycle corresponding to the representation @xmath117 is obtained from @xmath89 by the coboundary action by @xmath118 . as a result , the homology class obtained from @xmath111 depends only on the conjugacy class of @xmath111 . for a closed oriented @xmath18-manifold @xmath26 and a representation @xmath119 , we have a homology class defined by the image of the fundamental class @xmath120 $ ] under the map @xmath121 . when @xmath26 is homeomorphic to an ordered @xmath18-cycle @xmath61 , the homology class is represented by a @xmath8-valued @xmath15-cocycle @xmath89 of @xmath61 associated with @xmath111 . in this situation , we say that the homology class defined by @xmath26 and @xmath111 is _ represented _ by @xmath61 and @xmath89 . in this section , we give a presentation of the fundamental group of a cyclic branched covering from the wirtinger presentation . let @xmath5 be a knot in @xmath13 and @xmath12 be a diagram of @xmath5 . then @xmath122 is presented by generators and relations , called the wirtinger presentation . let @xmath123 be the generators of the wirtinger presentation , that correspond to the arcs of @xmath12 . each crossing ( figure [ fig : wirtinger_relation ] ) gives rise to a relation @xmath124 . for any integer @xmath125 , let @xmath126 be the @xmath127-fold cyclic covering of @xmath5 , i.e. the manifold corresponding to the kernel of @xmath128 putting back the knot @xmath5 to @xmath126 , we obtain the @xmath127-fold cyclic branched covering @xmath129 of @xmath5 . [ prop : presentation_of_pi_one ] @xmath130 has the following presentation : @xmath131 the inclusion map @xmath132 is given by @xmath133 if we take appropriate base points . by adding a relation @xmath134 , we obtain a presentation of @xmath135 . a method for obtaining a presentation of the fundamental group of a branched covering is given in @xcite . but we give a proof here because some techniques will be used to construct a group cycle represented by @xmath129 later . first , we construct a handle decomposition of the knot complement associated with the wirtinger presentation ( figure [ fig : trefoil ] ) . then we lift the handle decomposition to a handle decomposition of @xmath126 ( figure [ fig : lifted_splitting ] ) . after that , we read the relations given by attaching 2-handles . let @xmath136 be a regular neighborhood of @xmath5 . we shall give a handle decomposition of @xmath137 . we represent @xmath13 by the one point compactification of @xmath138 . let @xmath139 and @xmath140 . we denote the equatorial sphere of @xmath13 by @xmath141 . put @xmath5 in a position such that the projection to @xmath142 has only double points . let @xmath18 be the number of the crossings of this projection . we deform @xmath5 in the @xmath143-direction so that @xmath5 intersects @xmath142 at @xmath144 points and each of @xmath145 consists of @xmath18 arcs . we call @xmath146 over - crossing arcs and @xmath147 under - crossing arcs . index the arcs of @xmath148 by @xmath100 @xmath149 . now @xmath150 is homeomorphic to a handlebody of genus @xmath18 ( see figure [ fig : trefoil ] ) . projecting the over - crossing arcs to @xmath142 , we obtain a meridian disk system of the handlebody . we denote the meridian disk corresponding to @xmath100 by @xmath151 . for each under - crossing arc , attach a 2-handle @xmath152 along @xmath153 to @xmath154 ( see figure [ fig : trefoil ] ) . then the resulting manifold is homeomorphic to @xmath155 where @xmath156 is a 3-ball . attaching @xmath156 along the boundary to @xmath157 , the resulting manifold is homeomorphic to @xmath137 . so we have a handle decomposition of @xmath137 into one 0-handle , @xmath18 1-handles , @xmath18 2-handles and one 3-handle . in this handle decomposition , 1-handles correspond to the wirtinger generators and 2-handles to the wirtinger relations . we denote the set of @xmath39-handles by @xmath158 and @xmath159 . next we consider the preimage of @xmath160 in @xmath126 . cut @xmath160 along the meridian disks @xmath151 , and denote the resulting manifold by @xmath161 , which is homeomorphic to a 3-ball . let @xmath105 be a point in @xmath162 . we take a loop in @xmath160 which starts at @xmath105 , then intersects the meridian disk @xmath151 and ends at @xmath105 . orient the loop so that it corresponds to the generator of @xmath163 . by abuse of notation , we also denote this loop by @xmath100 . take a lift @xmath164 of @xmath161 and denote the preimage of @xmath165 in @xmath166 by @xmath167 . then there exists a unique lift @xmath168 of the loop @xmath100 starting at @xmath167 . since @xmath100 corresponds to the generator of @xmath169 , @xmath168 ends at another lift of @xmath161 , we denote it by @xmath170 . similarly the lift of @xmath100 starting at @xmath170 ends at another lift , we denote it by @xmath171 . continuing this , we see that @xmath172 and all lifts of @xmath161 will appear . therefore the preimage of @xmath160 is decomposed into @xmath127 3-balls @xmath173 . the intersection of @xmath174 and @xmath175 consists of @xmath18 disks each of which is a lift of a meridian disk @xmath151 . we denote this lifted disk by @xmath176 ( figure [ fig : lifted_splitting ] ) . it is easy to check that @xmath177 forms a meridian disk system of the preimage of @xmath160 . denote the lift of @xmath100 starting at @xmath174 ( ending at @xmath175 ) by @xmath178 . then the generator of @xmath179 corresponding to the meridian disk @xmath176 is given by @xmath180 we denote this element by @xmath181 . to give a simple presentation of @xmath179 , we add extra generators @xmath182 corresponding to @xmath183 respectively and relations @xmath184 . finally we consider the relations given by the lifts of 2-handles . we see that the relation @xmath124 lifts to @xmath185 see figure [ fig : lifted_splitting ] . since the generator @xmath181 is represented by ( [ eq : generator_in_cyclic_cover ] ) , the inclusion map @xmath186 is given by @xmath187 . this proves the second statement . by adding a 2-handle to @xmath126 along @xmath188 and capping off the resulting sphere , we obtain a manifold homeomorphic to the cyclic branched covering @xmath129 . therefore a presentation of the cyclic branched covering is obtained by adding a relation @xmath134 . for a representation @xmath189 , we have the restriction map @xmath190 given by @xmath191 if @xmath192 , it reduces to a representation @xmath193 and there is a group cycle given by @xmath194 and @xmath195 . in this section , we construct an explicit ordered @xmath14-cycle and its @xmath8-valued @xmath15-cocycle representing the homology class given by @xmath194 and @xmath195 . first we give a triangulation of @xmath137 associated with the wirtinger presentation . ll ( a ) & ( b ) + & + & + ( c ) & ( d ) + & + ( e ) & + let @xmath5 be a knot and fix a diagram of @xmath5 . as in the proof of proposition [ prop : presentation_of_pi_one ] , we define @xmath196 and give a heegaard splitting of @xmath154 with the meridian disks @xmath151 . cutting the handlebody @xmath154 along the meridian disks @xmath151 , the result is a ball with @xmath144 2-cells on the boundary . we denote the resulting 3-ball by @xmath161 and the pair of 2-cells corresponding to @xmath151 by @xmath108 and @xmath104 so that the wirtinger generator corresponding to @xmath100 runs from @xmath108 to @xmath109 ( figure [ fig : cut_heegaard](a ) ) . now consider the attaching regions of the two handles corresponding to under - crossing arcs , which consist of annuli @xmath197 on @xmath198 . each annulus is divided by @xmath151 s into 4 rectangles on @xmath199 . we define a graph on @xmath199 with vertices consisting of @xmath99 and edges consisting of these rectangles ( figure [ fig : cut_heegaard](b ) ) . each vertex of this graph has valency at least four . we can make all vertices of the graph into trivalent vertices ( figure [ fig : cut_heegaard](c ) ) by adding extra @xmath15-handles and @xmath200-handles ( stabilizations of the heegaard splitting , see figure [ fig : stabilization ] ) . we denote the new vertices by @xmath201 , @xmath202 , ... which originally belonged to @xmath99 . the dual of the graph gives a triangulation of @xmath199 ( figures [ fig : cut_heegaard](d ) and [ fig : cut_heegaard](e ) ) . by abuse of notation , we denote the triangles dual to the vertices @xmath201 , @xmath202 , ... by the same symbols . taking a cone from an interior point of @xmath161 , we obtain a triangulation @xmath61 of @xmath161 into @xmath203 tetrahedra . regluing the triangles @xmath204 , @xmath205 , ... to @xmath206 , @xmath207 , ... , we obtain a triangulation of @xmath137 into @xmath203 tetrahedra , which was explained in @xcite . we remark that this is not a triangulation in the usual sense : it is not a simplicial complex , moreover the link of some @xmath208-simplex is not homeomorphic to the @xmath200-sphere . actually there exists only three @xmath208-simplices , one is the cone point in @xmath161 ( north pole in @xcite ) , second is the @xmath208-simplex corresponding to the complementary regions of the diagram ( south pole ) , the last one is a @xmath208-simplex whose small neighborhood is homeomorphic to the cone over the torus @xmath209 . using this triangulation , we construct a triangulation of the cyclic branched covering @xmath129 . let @xmath210 be @xmath127 copies of @xmath161 and @xmath211 be the triangulation of @xmath174 we have constructed . we denote the triangles @xmath201 , @xmath202 , ... on @xmath212 by @xmath213 , @xmath214 , ... respectively . by abuse of notation , we regard @xmath213 , @xmath214 , ... simply as @xmath215 . glue @xmath211 s along their boundary triangles by the following identification maps @xmath216 denote this triangulation by @xmath217 . we define an ordering of each tetrahedron by assigning @xmath208 to the interior vertex of @xmath174 , @xmath15 to the vertex corresponding to complementary region of the diagram , @xmath200 to the vertex corresponding to under - crossing arc , @xmath14 to the vertex corresponding to over - crossing arc , respectively ( figure [ fig : ordering ] ) . the orderings are compatible under the gluing maps . so @xmath217 is an ordered @xmath14-cycle . here @xmath217 is a triangulation of @xmath129 except near the @xmath208-simplex corresponding to @xmath5 , whose small neighborhood is homeomorphic to a cone over a torus . we can resolve this singularity by inserting a suspension of a @xmath218-gon around @xmath5 ( figure [ fig : prism ] ) . as a result , we obtain an ordered @xmath14-cycle homeomorphic to @xmath129 . ( this procedure is called `` blowing up '' at a @xmath208-simplex in @xcite . ) we construct a group cycle given by @xmath129 and a representation @xmath219 using the triangulation @xmath217 . here @xmath195 is given by the set @xmath220 satisfying @xmath221 give a labeling of vertices on @xmath211 for each @xmath222 . let @xmath223 be the labeling of vertices on @xmath224 and @xmath225 be the labeling of vertices on @xmath226 . if these are related by @xmath227 then we obtain a @xmath8-valued @xmath15-cocycle @xmath89 on @xmath217 by gluing @xmath211 using @xmath195 . to obtain a group cycle given by @xmath129 and @xmath195 , we insert a suspension over a @xmath218-gon to @xmath217 at the @xmath208-simplex corresponding to @xmath5 . we give an ordering on the vertices of each tetrahedron of the suspension compatible with the ordering on the @xmath218-gon , e.g. order the central @xmath208-simplex maximal . these tetrahedra inherit a vertex labeling by @xmath8 on the boundary faces of the suspension . we assign any labeling at the central @xmath208-simplex of the suspension . then upper and lower tetrahedra have the same labelings with different orientations . so these cancel out in pairs and give no contribution to the group cycle represented by @xmath129 and @xmath195 . therefore the homology class given by @xmath217 and @xmath89 represents the homology class given by @xmath129 and @xmath195 . in this section , we review the definitions of quandles , rack ( co)homology and quandle ( co)homology . our treatment of quandle ( or rack ) homology follows that of @xcite . in [ subsec : shadow ] and [ subsec : cocycle_invariant ] , we recall the notions of colorings and quandle cocycle invariants defined in @xcite . a _ quandle _ @xmath228 is a set with a binary operation @xmath105 satisfying the following axioms : * @xmath229 for any @xmath230 , * the map @xmath231 defined by @xmath232 is a bijection for any @xmath233 , * @xmath234 for any @xmath235 . we denote the inverse of @xmath236 by @xmath237 . for a quandle @xmath228 , we define the _ associated group _ @xmath238 by @xmath239 a quandle @xmath228 has a right @xmath238-action in the following way . let @xmath240 be an element of @xmath238 where @xmath241 and @xmath242 . define @xmath243 . one can easily check that this is a right action of @xmath238 on @xmath228 . so the free abelian group @xmath244 $ ] generated by @xmath228 is a right @xmath245$]-module . let @xmath246 be the free left @xmath245$]-module generated by @xmath247 . we define the boundary map @xmath248 by @xmath249 figure [ fig : boundary_map ] shows a graphical picture of the boundary map . let @xmath250 be the @xmath245$]-submodule of @xmath246 generated by @xmath251 with @xmath53 for some @xmath39 . now @xmath250 is a subcomplex of @xmath246 . let @xmath252 . for a right @xmath245$]-module @xmath26 , we define the _ rack homology _ of @xmath26 by the homology of @xmath253 } c^r_n(x)$ ] and denote it by @xmath254 . we also define the _ quandle homology _ of @xmath26 by the homology of @xmath255 } c^q_n(x)$ ] and denote it by @xmath256 . the homology @xmath257 , here @xmath25 is the trivial @xmath245$]-module , is equal to the usual quandle homology @xmath258 . let @xmath259 be a set with a right @xmath238-action . for any abelian group @xmath45 , the abelian group @xmath260 $ ] freely generated by @xmath259 over @xmath45 is a right @xmath245$]-module . the homology group @xmath261)$ ] is usually denoted by @xmath262 ( @xcite ) . let @xmath42 be a left @xmath245$]-module . we define the _ rack cohomology _ @xmath263 by the cohomology of @xmath264 } ( c^r_n(x ) , n)$ ] . the _ quandle cohomology _ @xmath265 is defined in a similar way . for a set @xmath259 with a right @xmath238-action and an abelian group @xmath45 , we let @xmath266 be the left @xmath245$]-module generated by functions @xmath267 , where the action is defined by @xmath268 for @xmath269 and @xmath270 . the cohomology group @xmath271 is usually denoted by @xmath272 ( @xcite ) . let @xmath228 be a quandle . let @xmath273 be an oriented link in @xmath13 and @xmath12 be a diagram of @xmath273 . an _ arc coloring _ of @xmath12 is an assignment of elements of @xmath228 to arcs of @xmath12 satisfying the following relation at each crossing , where @xmath274 . by the wirtinger presentation of the knot complement , an arc coloring determines a representation of @xmath275 into the associated group @xmath238 . this is obtained by sending each meridian to its color . let @xmath259 be a set with a right @xmath238 action . region coloring _ of @xmath12 is an assignment of elements of @xmath259 to regions of @xmath12 satisfying the relation for any pair of adjacent regions , where @xmath276 and @xmath230 . a pair @xmath277 is called a _ shadow coloring_. if we fix a color of a region of @xmath12 , then colors of other regions are uniquely determined . therefore there always exists a region coloring compatible with a given arc coloring . we define a cycle @xmath278 of @xmath279)$ ] for a shadow coloring @xmath280 . assign @xmath281 for a positive crossing and @xmath282 for a negative crossing colored by respectively . then define @xmath283),\ ] ] where @xmath284 . we can easily check that [ prop : not_depend_on_region_coloring ] @xmath278 is a cycle and the homology class @xmath285 $ ] is invariant under reidemeister moves . moreover it does not depend on the choice of the region coloring if the action of @xmath238 on @xmath259 is transitive . so the homology class @xmath285 $ ] is an invariant of the arc coloring @xmath286 in many cases . there are two important sets with right @xmath238-action : one is when @xmath259 consists of one point @xmath287 and the other is when @xmath288 . eisermann showed that the cycle @xmath285 $ ] for @xmath289 is essentially described by the monodromy of some representation of the knot group along the longitude ( @xcite , @xcite ) . so we concentrate on the invariant @xmath285 $ ] in the case of @xmath288 from now on . let @xmath228 be a quandle with @xmath290 . let @xmath45 be an abelian group and @xmath51 be a cocycle of @xmath291 . we define the _ ( shadow ) quandle cocycle invariant _ by @xmath292.\ ] ] here the sum is finite because there are only a finite number of shadow colorings of @xmath12 . this is an invariant of oriented knots by proposition [ prop : not_depend_on_region_coloring ] . if @xmath238 acts on @xmath228 transitively ( i.e. @xmath228 is connected ) , @xmath293 does not depend on the choice of a region coloring @xmath294 by proposition [ prop : not_depend_on_region_coloring ] , thus @xmath295 coincides with the quandle cocycle invariant , where @xmath294 is a region coloring compatible with @xmath286 . we can regard @xmath296 as an element of @xmath297 by @xmath298 this gives a homomorphism @xmath299 . therefore a quandle @xmath14-cocycle @xmath300 gives rise to a quandle cocycle invariant . explicitly , the cocycle invariant has the following form : @xmath301\ ] ] where @xmath302 are given by @xmath286 and @xmath303 are given by @xmath294 . let @xmath228 be a quandle . let @xmath306 . we define the boundary operator of @xmath307 by @xmath308 since @xmath228 has a right action of @xmath238 , the chain complex @xmath307 has a right action of @xmath238 by @xmath309 . let @xmath26 be a left @xmath245$]-module . we denote the homology of @xmath310 } m$ ] by @xmath311 . for any abelian group @xmath45 , we can also define the cohomology group @xmath312 in a similar way . let @xmath313 be the set consisting of maps @xmath314 . we let @xmath315 denote the cardinality of the set @xmath316 . for each generator @xmath317 of @xmath318)$ ] , where @xmath319 , we define @xmath320 for any @xmath321 . fix an element @xmath322 . for each @xmath323 , we define a homomorphism @xmath324 ) \longrightarrow c^{\delta}_{n+1}(x ) \otimes_{\mathbb{z}[g_x ] } \mathbb{z}\ ] ] by @xmath325 for example , in the case @xmath326 ( see figure [ fig : main_map ] ) , @xmath327 and in the case @xmath328 , @xmath329 the map @xmath330 ) \longrightarrow c^{\delta}_{n+1}(x ) \otimes_{\mathbb{z}[g_x ] } \mathbb{z}$ ] is a chain map . therefore @xmath331 induces a homomorphism @xmath332 ) \to h^{\delta}_{n+1}(x;\mathbb{z } ) $ ] . we remark that the induced map @xmath332 ) \to h^{\delta}_{n+1}(x;\mathbb{z } ) $ ] does not depend on the choice of @xmath322 ( @xcite ) . in general , it is easier to construct cocycles of @xmath333 from group cocycles of some group related to @xmath228 than those of @xmath334)$ ] . if we have a function @xmath51 from @xmath335 to some abelian group @xmath45 satisfying 1 . @xmath336 , 2 . @xmath337 for any @xmath233 , 3 . @xmath338 if @xmath53 for some @xmath39 , then @xmath51 is a cocycle of @xmath339 and @xmath340 is a cocycle of @xmath341 . moreover , @xmath340 can be regarded as a cocycle in @xmath342 by @xmath343 we will construct functions satisfying these three conditions from group cocycles . for any integer @xmath344 , let @xmath1 denote the cyclic group @xmath2 with quandle operation defined by @xmath345 . actually this operation satisfies the quandle axioms . the quandle @xmath1 is called the _ dihedral quandle_. in this section , we construct quandle cocycles of @xmath1 from group cocycles of @xmath346 . in the next section , we will propose a general construction of quandle cocycles from group cocycles . let @xmath8 be the cyclic group @xmath2 ( @xmath3 is an integer greater than @xmath200 ) . the first cohomology @xmath347 is generated by the 1-cocycle @xmath348 defined by @xmath349 the connecting homomorphism @xmath350 of the long exact sequence corresponding to @xmath351 maps @xmath348 to a generator of @xmath352 , and the reduction @xmath353 maps it to a generator @xmath354 of @xmath355 . explicitly we have @xmath356 where @xmath357 is an integer @xmath358 such that @xmath359 . cup products of @xmath348 s and @xmath354 s are also cocycles . moreover , when @xmath3 is an odd prime , it is known that any element of @xmath360 can be presented by a cup product of @xmath348 s and @xmath354 s , see e.g. ( * ? ? ? * proposition 3.5.5 ) . we remark that @xmath348 and @xmath354 and their products are normalized cocycles . for an integer @xmath361 , let @xmath51 be a normalized @xmath63-cocycle of @xmath362 . regarding @xmath1 as @xmath363 , we obtain a map @xmath364 satisfying 1 . @xmath336 , 2 . @xmath338 if @xmath53 for some @xmath39 , by using the homogeneous notation ( [ subsec : group_homology ] ) . if @xmath51 also satisfies the condition 1 . @xmath337 for any @xmath365 , then @xmath51 gives rise to a quandle @xmath63-cocycle of @xmath366 by the construction of section [ sec : main_map ] . define @xmath367 by @xmath368 then @xmath369 satisfies the condition ( 2 ) by the left invariance of the homogeneous group cocycle . it is easy to check that @xmath369 also satisfies the condition ( 1 ) and ( 3 ) . so we obtain a quandle @xmath63-cocycle . we give an explicit presentation of the @xmath14-cocycle arising from @xmath370 . let @xmath371 then @xmath372 is a 2-cocycle . ( we can check that @xmath372 is cohomologous to @xmath373 . ) then by the definition ( equation ( [ eq : average_of_cocycle ] ) ) , @xmath374 is given by @xmath375 \mapsto x \cdot d(y , z).\ ] ] by definition we have @xmath376 and @xmath377 we remark that the cocycle @xmath372 can be easily understood geometrically . identify @xmath378 with the complex number @xmath379 where @xmath380 . then @xmath381 if @xmath382 is counterclockwise , @xmath383 if @xmath382 is clockwise and @xmath384 if @xmath382 is degenerate ( figure [ fig:2-cocycle ] ) . this interpretation and the equation ( [ eq : symmetric ] ) make various calculations easy . the quandle @xmath14-cocycle arising from @xmath370 has the following presentation : @xmath385 this is a non - trivial quandle 3-cocycle of @xmath1 with @xmath2 coefficient . in ( [ eq : main_map ] ) , since the map @xmath386 does not depend on the choice of @xmath387 , we let @xmath388 . then we have @xmath389 for @xmath390 . rewriting in inhomogeneous notation , this is equal to @xmath391 - [ 2y - x|x - y|z - y ] \\ & - [ 2z - x| x - y|y - z ] + [ 2z-2y+x| y - x|y - z ] . \\ \end{split}\ ] ] the evaluation of @xmath374 on this cycle is @xmath392 we will see that this cocycle is non - trivial because the evaluation on the cycle given by a shadow coloring @xmath280 of the @xmath393-torus link ( figure [ fig:(2,p)-torus ] ) is non - zero . color two arcs by @xmath394 as in the figure [ fig:(2,p)-torus ] . then other arcs must be colored by @xmath395 by the relations at the crossings . let @xmath396 be the color of the central region . then we have @xmath397 we assume that @xmath398 and @xmath399 . evaluation of the cocycle on @xmath278 is equal to @xmath400 by the next lemma , this is equal to @xmath401 . let @xmath361 be an integer . for @xmath402 , we have @xmath403 when @xmath404 , the left hand side is @xmath208 . when @xmath404 , this is straightforward since @xmath405 for any @xmath406 . when @xmath407 is a unit in @xmath2 , by ( [ eq : formula_of_d ] ) we have @xmath408 this is equal to @xmath409 when @xmath3 is odd . when @xmath3 is even , @xmath410 since @xmath411 is a unit in @xmath2 . when @xmath407 is not a unit in @xmath2 , let @xmath412 be the greatest common divisor of @xmath407 and @xmath3 then @xmath413 since @xmath414 is a unit in @xmath415 , this reduces to the previous case . since @xmath200 is divisible in @xmath2 when @xmath3 is odd , we have when @xmath361 is an odd number , @xmath416 is a non - trivial quandle 3-cocycle of @xmath1 with @xmath2 coefficient . when @xmath3 is prime , it is known that @xmath4 . therefore our cocycle is a constant multiple of the mochizuki 3-cocycle @xcite . we remark that when @xmath3 is prime , @xmath417 was calculated by nosaka @xcite for any @xmath18 , moreover he gave a system of generators of @xmath418 . when @xmath3 is an odd integer , nosaka showed in @xcite that @xmath419 . since @xmath420 is zero , we have @xmath421 . this means that there exists a non - trivial quandle 3-cocycle of @xmath1 with @xmath2 coefficient . in this section , we generalize the construction in the previous section to wider classes of quandles . we will construct a quandle cocycle of a faithful homogeneous quandle @xmath228 from a group cocycle of @xmath422 when an obstruction living in the second cohomology of @xmath422 vanishes . [ [ section ] ] let @xmath8 be a group . fix an element @xmath423 . let @xmath424 . now @xmath11 has a quandle operation by @xmath425 . in this section , we shall construct a quandle cocycle of @xmath11 from group cocycle of @xmath8 . first we shall show that this class of quandles is not so special . let @xmath228 be a quandle . we denote the group of the quandle automorphisms of @xmath228 by @xmath422 . we regard an automorphism acts on @xmath228 from the right . for @xmath230 , let @xmath426 be the map which sends @xmath407 to @xmath427 . by the axioms ( q2 ) and ( q3 ) , @xmath426 is a quandle automorphism . @xmath228 is called _ faithful _ if @xmath428 is injective . a quandle @xmath228 is _ homogeneous _ if @xmath422 acts on @xmath228 transitively . the following lemma was essentially shown in theorem 7.1 of @xcite , but we include a proof for completeness . every faithful homogeneous quandle @xmath228 is represented by @xmath11 with some group @xmath8 and @xmath423 . for @xmath230 and @xmath429 , we have @xmath430 . in fact , @xmath431 for any @xmath233 . let @xmath432 and fix an element @xmath433 . put @xmath434 . because @xmath228 is homogeneous , for any @xmath230 , there exists @xmath50 such that @xmath435 . so we have @xmath436 , namely @xmath437 . therefore we obtained a homomorphism @xmath438 . this is surjective since @xmath439 , and injective since @xmath228 is faithful . let @xmath440 be the centralizer of @xmath9 in @xmath8 . a map @xmath441 is well - defined and bijective . let @xmath442 . then @xmath443 , so @xmath444 and @xmath445 . this means that @xmath446 and @xmath447 belong to the same right coset . therefore the map is well - defined . by a similar argument , we can show the injectivity . surjectivity is trivial by definition . from now on we study the quandle structure on @xmath448 and construct a section of the projection @xmath449 . the quandle operation on @xmath11 induces a quandle operation on @xmath448 : @xmath450 this quandle operation on @xmath448 lifts to a quandle operation @xmath97 on @xmath8 by @xmath451 we can easily check that @xmath97 satisfies the quandle axioms ( the inverse operation is given by @xmath452 ) and the projection map @xmath453 is a quandle homomorphism . we remark that the quandle operation given by ( [ eq : lift_of_the_operation ] ) has been already studied by joyce @xcite and eisermann @xcite . let @xmath454 be a section of @xmath455 , i.e. a map ( not a homomorphism ) satisfying @xmath456 . since @xmath457 and @xmath458 are in the same coset in @xmath448 , there exists an element @xmath459 satisfying @xmath460 [ lem : section ] if @xmath461 is an abelian group , @xmath462 is a quandle 2-cocycle . if the cocycle @xmath412 is cohomologous to zero , we can change the section @xmath222 to satisfy @xmath463 . for @xmath464 and @xmath465 , we have @xmath466 therefore @xmath467 and @xmath468 for any @xmath469 . comparing ( [ eq : ditribution_1 ] ) and ( [ eq : ditribution_2 ] ) , we have @xmath470 by @xmath471 , we also have @xmath472 . if @xmath412 is cohomologous to zero , then there exists a map @xmath473 satisfying @xmath474 . put @xmath475 , then @xmath476 satisfies @xmath477 . the @xmath200-cocycle @xmath412 has already appeared in @xcite in a similar context . let @xmath8 be the dihedral group @xmath478 where @xmath3 is an odd number greater than @xmath200 . then we have @xmath479 and @xmath480 . we can identify @xmath481 with @xmath482 . define a section @xmath454 by @xmath483 then we have @xmath484 therefore @xmath485 for any @xmath486 . [ [ section-1 ] ] let @xmath8 be a group . fix @xmath423 with @xmath487 ( @xmath125 ) . in the following we assume : [ asm : vanishing ] @xmath461 is abelian and the 2-cocycle corresponding to @xmath488 is cohomologous to zero . under the assumption , we can take a section @xmath454 satisfying @xmath463 by lemma [ lem : section ] . let @xmath489 be a normalized group @xmath63-cocycle of @xmath8 in the homogeneous notation , where @xmath45 is an abelian group . then @xmath51 satisfies 1 . @xmath336 , 2 . @xmath490 for any @xmath50 ( left invariance ) , 3 . @xmath338 if @xmath53 for some @xmath39 . since it is convenient to use a right invariant function in the following construction , we modify the condition ( 2 ) by 1 . @xmath491 for any @xmath50 ( right invariance ) , by replacing @xmath492 by @xmath493 . define @xmath494 by @xmath495 for @xmath496 . the function @xmath369 satisfies the conditions ( 1 ) , ( 2 ) and ( 3 ) of section [ sec : main_map ] . therefore @xmath369 gives rise to a @xmath63-cocycle of @xmath497 . it is clear that ( 1 ) and ( 3 ) are satisfied from the conditions on a normalized group cocycle in the homogeneous notation . we only have to check the second property . @xmath498 combining with the arguments of section [ sec : main_map ] , we have if @xmath461 is abelian and the 2-cocycle corresponding to @xmath488 is cohomologous to zero , then there is a homomorphism @xmath499 for any abelian group @xmath45 . [ [ section-2 ] ] we return to the case of @xmath1 discussed in the previous section . we assume that @xmath3 is an odd integer greater than @xmath200 . let @xmath8 be the dihedral group @xmath478 . consider the short exact sequence @xmath500 we regard @xmath501 as @xmath502 by taking coset representatives in @xmath503 . then @xmath501 acts on @xmath2 by @xmath504 . this induces the restriction map @xmath505 we can show that this homomorphism is an isomorphism ( proposition iii.10.4 of @xcite ) . to obtain a group cocycle of @xmath503 from a group cocycle of @xmath2 , we need the inverse map , which is called the _ transfer_. the transfer map is described as follows ( see also @xcite ) . let @xmath396 be the map @xmath506 defined by @xmath507 and @xmath508 . for a cocycle @xmath51 of @xmath509 , the image @xmath510 of the transfer is given by @xmath511 in the homogeneous notation . when restricted to the image of @xmath454 , this is equal to the map defined in ( [ eq : average_of_cocycle ] ) . applying our construction for this group cocycle , we obtain a quandle 3-cocycle of @xmath1 , which is twice the cocycle constructed in the previous section . [ [ subsec : quandle_to_group ] ] we end this section by giving another homomorphism from @xmath512 to @xmath513 arising from more general context . let @xmath228 be a quandle and @xmath26 be a right @xmath245$]-module . we can construct a map from the rack homology @xmath514 to the group homology @xmath515 . the following lemma is well - known , e.g. see lemma 7.4 of chapter i of @xcite . let @xmath516 be a chain complex where @xmath517 are projective ( e.g. free ) . let @xmath518 be an acyclic complex . any homomorphism @xmath519 can be extended to a chain map from @xmath520 to @xmath521 . moreover such a chain map is unique up to chain homotopy . so there exists a unique chain map from @xmath522 to @xmath523 up to homotopy . this map induces @xmath524 } c^r_*(x ) \to m \otimes_{\mathbb{z}[g_x ] } c_*(g_x)$ ] and then @xmath525 . using normalized chains in group homology , we can also construct a map @xmath526 . we give an explicit chain map . let @xmath527 be a generator of @xmath246 . we define a map @xmath528 by @xmath529\ ] ] where @xmath530 is defined for a permutation @xmath66 and @xmath531 as follows . let @xmath532 be the maximal set of numbers satisfying @xmath533 . then define @xmath534 a graphical picture of this map is given in figure [ fig : quandle_cycle_to_group_cycle ] . for example , when @xmath328 , @xmath535 - [ x|z|y*z]+[y|z|(x*y)*z]-[y|x*y|z ] \\ & + [ z|x*z|y*z]-[z|y*z|(x*y)*z ] , \end{split}\ ] ] for @xmath536 . dually , we also have a map @xmath537 . we apply this map for @xmath11 . since there exists a natural homomorphism from the associated group @xmath538 to @xmath8 , we have a homomorphism @xmath539 fenn , rourke and sanderson defined the _ rack space _ @xmath540 in @xcite . since @xmath541 is isomorphic to @xmath238 , there exists a unique map , up to homotopy , from @xmath540 to the eilenberg - maclane space @xmath542 which induces the isomorphism between their fundamental groups . this map induces a homomorphism @xmath537 , which is equal to the map we have constructed in this subsection . clauwens showed in proposition 25 of @xcite that this map vanishes under some conditions on @xmath228 and @xmath26 . in particular , when @xmath3 is odd prime , @xmath543 vanishes for @xmath544 . in this section , we study the dual of the previous construction . we will show that the cycle @xmath278 associated with a shadow coloring @xmath277 of a knot @xmath5 gives rise to a group cycle represented by a cyclic branched covering along @xmath5 and the representation induced from the arc coloring @xmath286 . let @xmath228 be a quandle . let @xmath12 be a diagram of a knot @xmath5 . for a shadow coloring @xmath545 of @xmath12 whose arcs and regions are colored by @xmath228 , define @xmath546 and @xmath547 for @xmath548 by @xmath549 by the axiom ( q3 ) , @xmath550 is also a shadow coloring . in the following , we assume that @xmath551 for some group @xmath8 and @xmath423 and satisfies assumption [ asm : vanishing ] . as in the previous section , let @xmath552 be a section satisfying @xmath553 for @xmath554 . let @xmath555 composing with @xmath556 ) \longrightarrow c^{\delta}_{n+1}(x)$ ] , we have a map @xmath557 ) \xrightarrow{\varphi } { } c^{\delta}_{3}(x ) \xrightarrow{\iota } { } c_{3}(g ; \mathbb{z}).\ ] ] for a shadow coloring @xmath280 , @xmath558 is not a cycle of @xmath559 in general . but we can show [ thm : main_branched ] let @xmath560 be a shadow coloring of a diagram @xmath12 of a knot @xmath5 by @xmath11 . let @xmath561 be the color of an arc of @xmath12 . if @xmath562 , @xmath563 is a group cycle represented by the @xmath127-fold cyclic branched covering @xmath129 along the knot @xmath5 and the representation @xmath564 induced from the arc coloring @xmath286 . let @xmath565 be the arcs of the diagram @xmath12 such that @xmath566 . we denote the color @xmath567 by @xmath568 ( @xmath569 ) . the arc coloring @xmath286 induces a representation @xmath570 . using @xmath571 , @xmath111 is given by @xmath572 we have @xmath573 for any @xmath574 . so we have @xmath575 since @xmath576 by ( [ eq : covering_representation ] ) , we have @xmath577 let @xmath578 be copies of the triangulation of a @xmath14-ball with a vertex ordering on each tetrahedra constructed in section [ sec : cyc_rep_by_cyc_bra_cov ] ( figure [ fig : cut_heegaard ] and [ fig : ordering ] ) . then we define a labeling of vertices of @xmath211 by @xmath579 ( see figure [ fig : labeling_of_crossing ] ) . the vertices of the face @xmath580 is labeled by @xmath581 and the vertices of the face @xmath582 is labeled by @xmath583 where @xmath396 is the color of the region as indicated in figure [ fig : labeling_of_crossing ] . by ( [ eq : gluing_relation ] ) the labelings of the face @xmath584 and @xmath585 are related by @xmath586 therefore this gives a @xmath8-valued @xmath15-cocycle on @xmath217 as constructed in section [ sec : cyc_rep_by_cyc_bra_cov ] . thus the chain given by ( [ eq : cycle_represented_by_cyclic_branched_cover ] ) is a cycle represented by the cyclic branched cover @xmath129 and the representation induced from the arc coloring @xmath286 . we end this section by comparing the shadow cocycle invariant of the @xmath393-torus knot ( @xmath3 : odd ) for the @xmath14-cocycle obtained in section [ sec : dihedral ] and the dijkgraaf - witten invariant of the lens space @xmath587 . let @xmath8 be a finite group and @xmath588 be a cocycle of @xmath589 where @xmath45 is an abelian group . dijkgraaf and witten defined an invariant of closed oriented 3-manifolds for each @xmath588 . for an oriented closed manifold @xmath26 , it is defined by @xmath590 \rangle \in \mathbb{z}[a]\ ] ] where @xmath591 is the pull - back of @xmath588 by the classifying map @xmath592 corresponding to @xmath111 . since @xmath587 is a double branched cover along the @xmath393-torus knot , the cycle obtained from a shadow coloring by @xmath1 gives rise to a group @xmath14-cycle of @xmath503 represented by @xmath587 by theorem [ thm : main_branched ] . since every representation of @xmath593 into @xmath503 reduces to a representation into @xmath2 , it is natural to ask whether the shadow cocycle invariant for our quandle @xmath14-cocycle coincides with the dijkgraaf - witten invariant for @xmath594 . as we remarked in the introduction , these invariants coincide up to some constant by the result of hatakenaka and nosaka @xcite when @xmath3 is prime . let @xmath280 be the shadow coloring indicated in figure [ fig:(2,p)-torus ] . since the homology class @xmath285 $ ] does not change under the action of @xmath1 on the shadow coloring @xmath280 ( see lemma 4.5 of @xcite ) , we assume that @xmath399 . as computed in [ subsec:3_cocycle_of_dihedral ] , the evaluation of the @xmath278 at the @xmath14-cocycle derived from @xmath595 is equal to @xmath401 . therefore the shadow cocycle invariant is @xmath596/ ( t^p-1 ) \cong \mathbb{z}[\mathbb{z}/p].\ ] ] we compute the dijkgraaf - witten invariant of the lens space @xmath597 for the group @xmath14-cocycle @xmath594 . although this was computed in @xcite , we give a proof based on a triangulation . we represent @xmath597 by an ordered @xmath14-cycle and give a labeling of @xmath15-simplices as indicated in figure [ fig : lens ] . here the triple @xmath598 , @xmath599 and @xmath412 must satisfy @xmath600 and @xmath601 . using multiplicative notation , the evaluating of @xmath595 on this cycle is @xmath602 ) = \sum_{i=0}^{p-1 } a \cdot b_2(ia+b , -q a ) = -q a^2.\ ] ] the second equality follows form the fact that @xmath603 runs over @xmath2 and the equation ( [ eq:2_cocycle_of_cyclic_group ] ) when @xmath598 is a unit , and also when @xmath598 is not a unit by a similar argument . therefore the dijkgraaf - witten invariant of @xmath597 is equal to @xmath604/ ( t^p-1 ) \cong \mathbb{z}[\mathbb{z}/p].\ ] ] therefore this coincides with the shadow cocycle invariant up to a constant . 99 d. j. benson , _ representations and cohomology i : basic representation theory of finite groups and associative algebras _ , cambridge studies in advanced mathematics , 30 . cambridge university press , cambridge , 1991 . xii+224 pp . s. kamada , _ quandles with good involutions , their homologies and knot invariants _ , intelligence of low dimensional topology 2006 , 101108 , ser . knots everything , 40 , world sci . hackensack , nj , 2007 .

time series photometry of 20 cataclysmic variables detected by the catalina real - time transient survey is presented . 14 of these systems have not been observed previously and only two have been examined in - depth . from the observations we determined 12 new orbital periods and independently found a further two . eight of the cvs are eclipsing systems , five of which have eclipse depths of more than 0.9 mag . included in the sample are six su uma systems ( three of which show superhumps in our photometry ) , a polar ( sss1944 - 42 ) and one system ( css1417 - 18 ) that displays an abnormally fast decline from outburst . [ firstpage ] techniques : photometric - binaries : close , eclipsing - stars : novae , cataclysmic variables , dwarf novae

1310.3642

we present the latest results of a photometric follow - up study of faint cataclysmic variable stars ( cvs ; see @xcite ) mostly in the southern hemisphere . specifically , in this paper we discuss the observations of 20 cvs identified by the catalina real - time transient survey ( crts ; see @xcite ) . this work forms part of a survey with the aim of characterising newly identified cvs , determining their orbital periods , searching for sub - orbital periodicities and selecting targets for in - depth studies on large telescopes and multi - wavelength campaigns . in the previous papers in this series ( see @xcite and references therein ) , the focus was on faint nova remnants and cvs identified by the sloan digital sky survey ( sdss , e.g. @xcite ) . recently , it has shifted to cvs discovered by the crts @xcite . the crts is a large scale transient survey that observes 30@xmath0000 deg@xmath1 of the sky in search of transients ( see @xcite ) . they make use of data from the catalina sky survey @xcite , which searches for near - earth objects . the observing strategy is to observe a field four times at 10-minute intervals , then return to the field up to four times per lunation @xcite . transients are detected by looking for variations of more than 2 mag in the @xmath2 filter . as most of the fields have 7- to 8-year baselines and the crts can reach a depth of @xmath323 by co - adding images ( although individual pointings reach @xmath319 to 21 ) , a variety of transients are discovered . amongst these are supernovae , blazars , flare stars and cvs . as evidenced by the number of cvs the crts has detected - more than 1000 to date ( september 2013 ) - the crts is particularly efficient at finding them . this is for a number of reasons . for example , the magnitude variation cut - off limit of 2 mag , which was designed to restrict the number of artifacts and pulsating variables , is approximately the minimum amplitude of a dwarf nova outburst , so large numbers of dwarf novae ( dne ) are detected . this range also ensures that the high inclination systems with deep eclipses are detected , as well as a number of magnetic systems that show high and low states . as the crts commonly reaches down to @xmath321 , or fainter if the object has a high state brighter than this limit , it will find faint cvs such as the double degenerate am cvn ( am canum venaticorum ) systems . also advantageous to our photometric follow - up study , is the excellent coverage in the southern hemisphere . follow - up observations on the objects identified by large scale surveys , such as the crts , are becoming increasingly necessary as existing projects continue to detect new transients and surveys such as the large synoptic survey telescope ( lsst ; @xcite ) are planned . on average , from 2008 to 2012 the crts detected 45 new cvs per quarter . for cvs , finding the orbital periods for a greater sample of systems is necessary for evolutionary studies ( e.g. @xcite ) . additionally , amongst the newly discovered cvs will be systems that need to be observed with larger telescopes ( e.g. , salt : woudt et al . 2010 ) and multi - wavelength campaigns ( e.g. cc sculptoris : woudt et al . 2012b ) . particularly important are the eclipsing systems , from which we can obtain system characteristics such as the mass ratio through eclipse deconvolution ( littlefair et al . 2008 ; savoury et al . photometric surveys , such as the kind presented in this paper , help to identify these targets . in this paper , in section [ sec : observations ] we describe the observing procedure , the instrumentation and the data reduction . the results of each of the individual cvs follow . section [ sec : conclusions ] contains a summary and discussion of the results . differential photometry was performed using the university of cape town ( uct ) ccd @xcite and the new sutherland high - speed optical cameras ( shoc ; @xcite , @xcite ) mounted on the 74-in and 40-in reflector telescopes of the south african astronomical observatory ( saao ) . for the first time in this series of papers , we have used the shoc systems . two instruments utilizing andor ixon 888 cameras are available for use at the saao . the field of view is 2.85x2.85 arcmin@xmath1 on the 40-in telescope and 1.29x1.29 arcmin@xmath1 on the 74-in , as compared to 1.23x1.82 arcmin@xmath1 and 0.57x0.83 arcmin@xmath1 on the uct ccd . for these observations both the uct ccd and shoc were used in frame - transfer mode and shoc was operated in 1mhz conventional mode with 2.4 preamplifier gain . no filters were used , but the data were calibrated to the sloan @xmath4 photometric system . papers i to vi in this series used hot white dwarfs as calibration standards @xcite , which gave @xmath2 accurate to @xmath50.1 mag @xcite . using the survey archival data we compared the calibration offset using hot white dwarfs and those using sdss photometry of comparison stars on the target field on nights when both were available . over the range @xmath6=0.2 to 1.0 , it was found that there was a stable zero - point offset of [email protected] mag between @xmath2 and sdss @xmath4 - which is consistent with the photometric transformation in @xcite . subsequently , we calibrated our white light photometry to the @xmath4 photometric system using sdss stars on the target field that have @xmath6 colours in the range 0.2 - 1.0 . the calibration is accurate to @xmath50.1 mag . in the cases where a suitable calibration star was not observed over the course of the evening , we applied the calibration determined for another night during the observing run . the runs calibrated in this way are labelled with a colon in the observing log ( table [ tbl : log ] ) . the observing procedure varies according to the behaviour of a given cv . a target is initially observed for a few hours ( @xmath54 h ) . if the light curve shows modulation , then it is observed further over consecutive nights . if there are no modulations present or the cv appears to have a long orbital period ( greater than 5 hours ) , then no further observations are taken as the telescope time can be spent more profitably . longer - term monitoring ( weeks to years ) is carried out in order to reduce the error margin on orbital periods or to observe cvs in an alternate outburst state in order to obtain a superhump period , an orbital period or to examine quasi - periodic oscillations ( qpos ) and dwarf nova oscillations ( dnos ) . data reductions were performed with the program duphot for the uct ccd @xcite and in iraf @xcite using standard reduction routines for the shoc observations . two different techniques were used to find periodicities ( such as the orbital period ) in the photometry . the cvs which had sinusoidal light curves were analyzed by fourier analysis using the starlink period package . those with non - sinusoidal light curves ( e.g. the eclipsing systems ) were analyzed by phase dispersion minimization ( pdm , @xcite ) . the former technique has been explained in detail in previous papers in this series . briefly , pdm determines the true period by folding the light curve on a series of test periods and determining which produces the least scatter . in order to quantify the scatter , each folded light curve is binned and the overall variance of all the bins is divided by the variance of the unbinned data . the result - the pdm statistic @xmath8 - will be close to 1 for false periods and be small for true periods . a pdm periodogram will thus have local minima at the true periods ( see for example figure [ fig : css0116_pdm ] ) . full details are given in @xcite . the distinction in treatment between the non - sinusoidal and sinusoidal systems is due to the fact that in attempting to fit sharp features by a series of sinusoids , fourier analysis spreads the power to harmonics of the period , thereby reducing the power of the fundamental . pdm does not suffer from this problem . unless otherwise stated , before the pdm periodogram or fourier transform ( ft ) is calculated , the individual runs are linearly detrended and mean - subtracted . this reduces the spurious noise at low frequencies introduced by observing the cv over the same airmass range over consecutive nights . in order to determine an uncertainty on the periods obtained from the ft , a sine - curve is fitted to the photometry using the ft period as an initial guess . by varying the period , a markov chain monte carlo method ( mcmc ) - specifically the metropolis hastings algorithm ( @xcite and @xcite ) - then samples the probability distribution of solutions . the median value and standard deviation of this distribution are then quoted as the period and uncertainty respectively . in cases where there are additional strong peaks in the ft , more than one sine - wave is fitted simultaneously . for those systems which have highly non - sinusoidal light curves , the uncertainty can not be determined in this way . the uncertainty derived from pdm for these cvs is obtained by bootstrapping using the monte carlo algorithm for case resampling @xcite . the sampling distribution for the period is estimated by creating new light curves by selecting n of the n data points ( repetition is allowed ) and performing pdm on each new sample . as is the case for the ft , the period and uncertainty given in this paper are the median and standard deviation of the bootstrap distribution . a variety of other techniques are used to select the correct period from the ft / pdm periodogram and determine the reality of a peak . these include phase - folding , using the window function - which shows the aliasing structure introduced by sampling - and the fisher randomisation test @xcite . in the latter , the amplitude values on the light curves are shuffled to create a large number of new light curves and fts are calculated for each . the probability of a period being real and equal to the quoted value on the original ft is then determined by looking at the proportion of shuffled light curves that gave a higher peak at that frequency . the observing log is given in tab . 1 . each crts transient has an i d of the form css yymmdd : hhmmss@xmath7ddmmss . the first three letters indicate in which of the three component surveys it was discovered ( css : catalina sky survey , mls : mount lemmon survey and sss : siding springs survey , see @xcite ) . the following six digits indicate the date on which it was classified . the right ascension and declination follow the colon . in the observing log and the discussion the i d is abbreviated to the form css hhmm@xmath7dd . [ tbl : log ] lllllllll object & type & run no . & date of obs . & hjd of first obs . & length & t@xmath9 & tel . & @xmath4 + & & & ( start of night ) & ( + 2450000.0 ) & ( h ) & ( s ) & & ( mag ) + * css0116 + 09 * & dn & s8018@xmath10 & 27 november 2010 & 5528.2811 & 2.3485 & 30 & 40-in & 16.8@xmath11 + & & s8021@xmath10 & 28 november 2010 & 5529.2925 & 2.3928 & 30 & 40-in & 17.1@xmath11 + & & s8024@xmath10 & 29 november 2010 & 5530.2755 & 3.5503 & 15 & 40-in & 17.3@xmath11 + & & s8033 & 5 december 2010 & 5536.2887 & 1.1579 & 20,25 & 74-in & 18.9@xmath11 + & & s8040 & 8 december 2010 & 5539.3026 & 2.0642 & 25 & 74-in & 19.1@xmath11 + & & s8098@xmath12 & 5 october 2011 & 5840.4774 & 2.8041 & 20 & 74-in & 18.8@xmath11 + & & s8100@xmath12 & 6 october 2011 & 5841.4096 & 1.4116 & 20 & 74-in & 18.9@xmath11 + * css0411 - 09 * & su & s7891@xmath13 & 21 december 2009 & 5187.2848 & 3.6556 & 8 & 74-in & 15.3 + & & s7893@xmath13 & 22 december 2009 & 5188.2872 & 1.4372 & 8 & 74-in & 15.3 : + & & s7898@xmath13 & 24 december 2009 & 5190.2908 & 1.4061 & 8 & 74-in & 15.7 : + * css0438 + 00 * & dn & s8022 & 28 november 2010 & 5529.4480 & 2.0579 & 30 & 40-in & 19.2@xmath11 + & & s8025 & 29 november 2010 & 5530.4410 & 3.7803 & 20 & 40-in & 19.3@xmath11 + & & s8050 & 26 december 2010 & 5557.3107 & 1.3357 & 70 & 74-in & 19.5:@xmath11 + * css0449 - 18 * & dn & s8059 & 30 january 2011 & 5592.3146 & 3.6642 & 45 & 40-in & 17.9:@xmath11 + & & s8061 & 31 january 2011 & 5593.2846 & 3.1261 & 45 & 40-in & 17.8:@xmath11 + & & s8062 & 1 february 2011 & 5594.2851 & 4.3161 & 45 & 40-in & 17.9:@xmath11 + & & s8063 & 6 february 2011 & 5599.3015 & 3.1400 & 45 & 40-in & 17.9:@xmath11 + * sss0501 - 48 * & dn & s8144@xmath12 & 21 january 2012 & 5948.2779 & 0.4890 & 20 & 74-in & 17.3 + & & s8145@xmath12 & 22 january 2012 & 5949.2795 & 5.5672 & 20 & 74-in & 17.2 + & & s8147@xmath12 & 23 january 2012 & 5950.2802 & 5.3910 & 20 & 74-in & 17.3 + * css0558 + 00 * & dn & s8031 & 4 december 2010 & 5535.4867 & 1.0567 & 40 & 74-in & 19.2 + & & s8041 & 8 december 2010 & 5539.3984 & 4.5805 & 40 & 74-in & 19.4 + & & s8042 & 9 december 2010 & 5540.3889 & 4.6192 & 40,50,60 & 74-in & 19.6 + & & s8044 & 12 december 2010 & 5543.4696 & 2.8842 & 60 & 74-in & 19.3 + & & s8048 & 24 december 2010 & 5555.4143 & 2.4899 & 40 & 74-in & 18.9 + * css0902 - 11 * & dn & s7927@xmath10 & 22 march 2010 & 5278.2856 & 1.1053 & 6 & 74-in & 16.5 : + & & s7932 & 31 march 2010 & 5287.2415 & 5.0953 & 30 & 40-in & 17.8 : + & & s7934 & 1 april 2010 & 5288.2411 & 4.4381 & 30 & 40-in & 17.8 + & & s8032 & 4 december 2010 & 5535.5401 & 0.9011 & 20 & 74-in & 17.6 + & & s8039 & 7 december 2010 & 5538.5174 & 1.5656 & 10 & 74-in & 17.7 : + * css0942 - 19 * & su ? & s8082 & 11 may 2011 & 5693.2133 & 4.1458 & 60 & 74-in & 19.6 + & & s8085 & 12 may 2011 & 5694.2114 & 4.1342 & 60 & 74-in & 19.5 + & & s8088 & 13 may 2011 & 5695.2133 & 3.9792 & 60 & 74-in & 19.3 + * css1052 - 06 * & su & s7949@xmath14 & 8 april 2010 & 5295.2194 & 3.4922 & 10 & 40-in & 16.0 + & & s7952@xmath14 & 9 april 2010 & 5296.2272 & 1.9869 & 10 & 40-in & 16.3 : + & & s7955@xmath14 & 10 april 2010 & 5297.4247 & 0.6506 & 10 & 40-in & 16.4 + & & s7964 & 13 april 2010 & 5300.3142 & 3.2947 & 20 & 40-in & 18.7 + * sss1128 - 34 * & dn & s8091 & 14 may 2011 & 5696.2570 & 4.9031 & 30 & 74-in & 18.7 : + & & s8094@xmath10 & 15 may 2011 & 5697.2270 & 3.1458 & 40 & 74-in & 16.0 : + & & s8095@xmath10 & 16 may 2011 & 5698.2932 & 2.5563 & 40 & 74-in & 16.0 : + * css1221 - 10 * & dn & s8056 & 28 january 2011 & 5590.5679 & 1.4753 & 90 & 40-in & 19.3 : + & & s8058 & 29 january 2011 & 5591.4973 & 3.0451 & 120 & 40-in & 19.3 : + & & s8060 & 30 january 2011 & 5592.5197 & 1.8894 & 100 & 40-in & 19.3 : + & & s8064 & 6 february 2011 & 5599.4637 & 3.9494 & 120 & 40-in & 19.4 : + & & s8066 & 7 february 2011 & 5600.4846 & 1.2688 & 120 & 40-in & 19.7 : + & & s8068 & 5 march 2011 & 5626.4949 & 3.6354 & 120 & 40-in & 19.6 : + & & s8069 & 6 march 2011 & 5627.3897 & 4.5690 & 120 & 40-in & 19.4 : + & & s8071 & 8 march 2011 & 5629.3847 & 6.4775 & 120 & 40-in & 19.4 + * sss1224 - 41 * & dn & s8162@xmath12 & 28 february 2012 & 5986.5230 & 2.4560 & 120 & 40-in & 19.4@xmath11 + & & s8177@xmath12 & 17 march 2012 & 6004.4454 & 5.1158 & 120 & 74-in & 19.3@xmath11 + & & s8180@xmath12 & 18 march 2012 & 6005.4696 & 4.1647 & 120 & 74-in & 19.3@xmath11 + & & s8182@xmath12 & 19 march 2012 & 6006.3926 & 6.0487 & 120 & 74-in & 19.3@xmath11 + & & s8185@xmath12 & 20 march 2012 & 6007.4541 & 4.5465 & 120 & 74-in & 19.3@xmath11 + * sss1340 - 35 * & dn & s8188@xmath12 & 20 april 2012 & 6038.3205 & 2.9894 & 60 & 40-in & 18.4@xmath11 + * css1417 - 18 * & su & s7992@xmath10 & 6 july 2010 & 5384.2382 & 2.3681 & 20 & 74-in & 16.8 + & & s7993@xmath10 & 7 july 2010 & 5385.2267 & 2.5268 & 20 & 74-in & 18.8 + & & s8080 & 10 may 2011 & 5692.3198 & 4.5998 & 120 & 74-in & 19.9 + & & s8083 & 11 may 2011 & 5693.4119 & 3.0331 & 120 & 74-in & 20 + & & s8089 & 13 may 2011 & 5695.4009 & 3.0666 & 120 & 74-in & 19.7 + + lllllllll object & type & run no . & date of obs . & hjd of first obs . & length & t@xmath9 & tel . & @xmath4 + & & & ( start of night ) & ( + 2450000.0 ) & ( h ) & ( s ) & & ( mag ) + * css1556 - 08 * & dn & s8078@xmath10 & 9 may 2011 & 5691.4659 & 4.3199 & 10 & 74-in & 16.9 : + & & s8081 & 10 may 2011 & 5692.5194 & 2.9558 & 10 & 74-in & 18.0 + & & s8086 & 12 may 2011 & 5694.3974 & 3.7759 & 10 & 74-in & 18.4 : + * css1727 + 13 * & dn & s7967 & 19 may 2010 & 5336.5160 & 2.3258 & 45 & 74-in & 19.0 : + & & s7970 & 20 may 2010 & 5337.4814 & 3.4521 & 45 & 74-in & 19.7 : + & & s7972 & 21 may 2010 & 5338.4636 & 3.0020 & 45 & 74-in & 19.7 : + & & s7976 & 22 may 2010 & 5339.4875 & 2.1676 & 55 & 74-in & 19.7 : + * sss1944 - 42 * & p & s8084 & 11 may 2011 & 5693.5527 & 2.9464 & 10 & 74-in & 17.4 + & & s8087 & 12 may 2011 & 5694.5753 & 2.4897 & 10 & 74-in & 17.6 : + & & s8090 & 13 may 2011 & 5695.5555 & 3.1700 & 10 & 74-in & 17.2 + * sss2003 - 28 * & su & s8093 & 14 may 2011 & 5696.5973 & 2.0719 & 35 & 74-in & 19.1:@xmath11 + & & s8096@xmath10 & 16 may 2011 & 5698.5522 & 2.9172 & 90,100 & 74-in & 16.4:@xmath11 + * css2054 - 19 * & su & s8005@xmath14 & 31 october 2010 & 5501.2447 & 4.0508 & 45 & 74-in & 16.5 : + & & s8008@xmath14 & 1 november 2010 & 5502.2457 & 3.4636 & 45 & 74-in & 16.7 + & & s8010@xmath14 & 2 november 2010 & 5503.2729 & 2.5758 & 30 & 74-in & 17.3 : + * css2108 - 03 * & dn & s8107@xmath15 & 11 october 2011 & 5846.2343 & 4.5482 & 30 & 74-in & 17.7@xmath11 + & & s8109@xmath15 & 15 october 2011 & 5850.2412 & 2.4834 & 60 & 40-in & 17.6@xmath11 + & & s8111@xmath15 & 16 october 2011 & 5851.2308 & 3.7767 & 60 & 40-in & 17.6@xmath11 + & & s8113@xmath15 & 19 october 2011 & 5854.2789 & 0.6000 & 60 & 40-in & 17.8@xmath11 + + css0116 + 09 was discovered by the crts on december 20 , 2008 when it was observed in outburst at @xmath163 mag . it has a counterpart in data release 8 of the sloan digital sky survey ( sdss , @xcite ) with @xmath17=19.0 , @xmath18=19.1 and @xmath4=19.0 . @xcite further observed it at @xmath19=0.24@xmath20 , @xmath2=18.89@xmath21 , @xmath22=0.64@xmath23 . based on its colour , they confirmed its crts classification as a cv . we caught css0116 + 09 on the decline from outburst in november 2010 and in quiescence on two further observing runs ( see tab . the outburst light curves are given in fig . [ fig : css0116_outburst ] and show css0116 + 09 to be a deeply eclipsing system . 0.13@xmath24hz is indicated by a vertical bar . ] the pdm periodogram ( see fig . [ fig : css0116_pdm ] ) of these observations together with those taken in quiescence a week later ( s8033 and s8040 ) , gives an orbital period of [email protected]@xmath0d . bootstrapping the results gives p@[email protected]@xmath0d , where the orbital period and uncertainty are the median and standard deviation of the distribution respectively . the ephemeris for minimum light is5 ) denotes an error of 0.00005 . ] @xmath26 plot a of fig . [ fig : css0116_quiescent ] shows the average light curve of s8033 and s8040 folded on the orbital period . the average of runs s8098 and s8100 is shown in plot b of the figure . unfortunately these runs were too far removed in time to reduce the uncertainty in eq . [ eq : css0116 + 09ephem ] . from s8040 to s8098 , the uncertainty on the eclipse timing amounted to more than a period and so combining these runs would have introduced a cycle ambiguity . the eclipse depth of the light curves ranges from 0.8@xmath0mag to 1.4@xmath0mag . it is at its minimum in outburst , when the accretion disc contributes a larger fraction of the light . this depth indicates that css0116 + 09 has an inclination of greater than approximately 70@xmath27 . the crts light curve of css0411 - 09 has a number of high amplitude outbursts ( up to 3.8@xmath0mag ) over the 8-year baseline . no follow - up work had been undertaken on this object prior to this paper . based on our observations and the quoted crts quiescent magnitude of @xmath2=19.4 , we observed the system during an outburst of more than 4@xmath0mag . the light curves ( fig . [ fig : css0411 ] ) show superhumps with amplitudes decreasing from approximately 0.5 to 0.3@xmath0mag over four days - indicating that this was a superoutburst and that css0411 - 09 is of the su uma class . mag respectively . the time given is the fractional part of the heliocentric julian date . ] 0.02@xmath24hz , and its first and second harmonics are marked . middle panel : the corresponding window function shows the alias structure in the ft . lower panel : pdm periodogram of the combined runs , with the superhump period marked . ] hz ( 25.58@xmath0s ) . ] the combined ft of runs s7891 , s7893 and s7898 is shown in the upper panel of fig . [ fig : css0411 ft ] . only the mean was subtracted from each run , not the linear trend . the middle panel shows the window function , which reveals some structure in the ft and accounts for the broadness of the alias structure . power at the first ( 349.18@xmath28hz ) and second harmonic ( [email protected]@xmath29hz ) of the main peak at [email protected]@xmath29hz allows the unambiguous identification of the superhump period of [email protected]@xmath0d . the pdm periodogram ( see bottom panel of fig . [ fig : css0411 ft ] ) gives a consistent solution . bootstrapping the pdm gives p@[email protected]@xmath0d . using p@[email protected]@xmath0d and equation 1 of @xcite gives an approximate orbital period of 0.0645(@xmath75)@xmath0d for css0411 - 09 . the ft of run s7898 ( shown in fig . [ fig : css0411ftdno ] ) , when css0411 - 09 started its descent towards quiescence ( 0.4@xmath0mag fainter than the previous observation two days earlier ) , shows a clear peak at [email protected]@xmath0s with an amplitude of 4@xmath0mmag . the amplitude and frequency of this rapid optical modulation is typical of dwarf nova oscillations ( see @xcite ) . over the course of the 6.5-year crts observations of css0438 + 00 , three outbursts have been recorded . the first was observed at @xmath162.5@xmath0mag above the quiescent magnitude ( @xmath2=19.3 ) and the second , at @xmath162@xmath0mag . it is within the sdss survey area and was observed in quiescence at @xmath17=19.4 , @xmath18=19.5 and @xmath4=19.4 mag . [ fig : css0438 ] gives light curves for the three observing runs we obtained for css0438 + 00 . it shows an eclipse depth of @xmath31mag - indicative of a system with an inclination greater than approximately 70@xmath27 . the magnitude of the system , when out of eclipse , varies between orbits ( for example , compare s8025 before and after phase 0 ) . this type of variation is usually attributed to changes in the mass transfer rate from the secondary . in the combined pdm periodogram of s8022 and s8025 it produced a very high amplitude peak at 0.131@xmath0d . prewhitening at this frequency gave an orbital period of 0.0657 d. taking the median and standard deviation of the bootstrapped sample as the orbital period and uncertainty , yields an ephemeris for minimum light of @xmath32 unfortunately s8050 could not be used to reduce the uncertainty on p@xmath25 . over the 27 days between s8025 and s8050 , the uncertainty on the orbital period accumulated to the point where there was a cycle ambiguity . css0449 - 18 shows frequent outbursts in its crts light curve , despite the fact that it is not as well sampled as the other css systems . the quiescent magnitude is quoted at @xmath2=17.7 and the outbursts were captured up to 2.8@xmath0mag above quiescence . our photometry was taken at close to minimum light . the light curves show orbital humps and rounded , shallow eclipses ( see fig . [ fig : css0449 ] ) that vary in profile from night to night . this indicates that it is likely that only the bright spot and part of the accretion disc were eclipsed . . the average is plotted twice for display purposes . ] from bootstrapping the pdm , we obtain an orbital period of @[email protected]@xmath0d . the ephemeris for minimum light is @xmath34 the average light curve is shown in fig . [ fig : css0449avg ] . the eclipses have a depth of @xmath35mag and a duration of approximately 26@xmath0min . they resemble those of u gem ( see for example @xcite and @xcite ) , consequently the inclination angle should be similar to the @xmath572@xmath27 estimated for u gem by @xcite . sss0501 - 48 shows a variation in its quiescent magnitude in the crts light curve , ranging between [email protected] and 19@xmath0mag . two outbursts were also recorded up to [email protected] . a substantial amount of flickering can be seen in our observations of sss0501 - 48 ( see fig . [ fig : sss0501 ] ) , which could partly explain the scatter seen in the crts quiescent magnitudes . despite the long length of the observing runs , the fts did not show any periodic behaviour . h and 5.39@xmath0h long respectively and are displaced vertically by 0.5 and 1.4@xmath0mag for display purposes . ] css0558 + 00 was discovered on january 14 , 2010 during an outburst of approximately @xmath36=2.4@xmath0mag . two previous outbursts were also recorded in the crts archival data . it has a counterpart in the sdss dr8 with @xmath17=20.6 , @xmath18=20.3 and @xmath4=19.0 . spectroscopic follow - up observations were carried out by @xcite , who observed it during an outburst that was not captured by the crts . the single spectrum they obtained was typical of a dwarf nova in outburst and showed weak , broad h@xmath37 line in emission . h ( s8031 ) and 4.6@xmath0h ( s8041 ) . each run is offset vertically by the magnitude indicated in parentheses . ] 0.03@xmath29hz ) and its first harmonic are marked . the power at lower frequencies is caused by the finite length of the runs . ] we observed this system during quiescence . each of the runs showed flickering ( see fig . [ fig : css0558 ] ) , which caused broad , low - amplitude peaks in the individual fts . figure [ fig : css0558 ft ] gives the ft and pdm periodogram for the four longest runs . both contained a peak at 0.06809@xmath0d and at the first harmonic . a fisher randomization test on the fundamental gives a false alarm probability ( fap ) lying between 0 and 0.01 with 95% confidence , that this period is not present in the data . it gives the same fap that the period is not equal to the quoted value . these two results confirm that the 0.06809@xmath0d period is real . bootstrapping the pdm gives an orbital period of 0.06808(@xmath71)@xmath0d . phasefolding the observations gives the ephemeris for maximum light as @xmath38 the quiescent magnitude for css0902 - 11 , as given by the crts , is @xmath2=17.5@xmath0mag . it was discovered on february 1 , 2009 when it went into a 2.8@xmath0mag or higher outburst . @xcite took time series spectroscopy of this cv , from which they estimated it to be at a distance of 1100(+350,-260 ) pc and have a secondary star of spectral type k7@xmath39 and magnitude [email protected]@xmath0 . additionally , based on the small velocity amplitude of the secondary ( 100@xmath40km@xmath0s@xmath41 ) , they predicted that the system would not be eclipsing . this is confirmed by our photometry . h modulation detected by @xcite . the data - length of the two runs , s7932 and s7934 , were 5.1@xmath0h and 4.4@xmath0h respectively . ] their absorption spectra showed a clear 6.62@xmath42h modulation . only two of our runs approached this length - at 5.1 and 4.4@xmath0h respectively . the combined ft gives a period of 3.3@xmath43h , half that of @xcite s value . phase - folding our two longest runs ( s7932 and s7934 ) on the 6.62@xmath0h period ( fig . [ fig : css0902 ] ) shows that css0902 - 11 is a double hump system . only one orbital hump was covered entirely in each of our two runs , which produced significant power in the ft at half the orbital period . in conclusion , our photometry is consistent with the 6.62@xmath0h orbital period detected by @xcite . the crts light curve for css0942 - 19 is sparsely sampled , but shows two outbursts with amplitudes of approximately 4@xmath0mag . judging from the large amplitudes , these may have been superoutbursts . this system was also caught in outburst by the palomar quest sky survey @xcite . we observed css0942 - 19 in quiescence in may 2011 at @xmath44 ( see tab . 1 ) four months after the last recorded outburst by crts . the pdm periodogram of the combined runs on this object ( see fig . [ fig : css0942pdm ] ) , shows strong peaks at the orbital frequency ( [email protected]@xmath29hz ) and at its first harmonic . bootstrapping the pdm , we obtain an orbital period and ephemeris for maximum light of @xmath45 the average light curve of css0942 - 19 , folded on this ephemeris , is shown in fig . [ fig : css0942 ] . 0.001@xmath0d photometric period . ] since css1052 - 06 was detected by the crts on april 8 , 2010 , numerous outbursts of approximately @xmath36=3@xmath0mag have been captured . no follow - up observations have been performed on it prior to this paper . during our first three observing runs ( s7949 , s7952 and s7955 ) , css1052 - 06 was in superoutburst and superhumps were present ( see fig . [ fig : css1052 ] ) . three days later , during s7964 , the system had returned to quiescence . mag respectively . ] the ft and the pdm periodogram of the two long runs in superoutburst are given in fig . [ fig : css1052 ft ] . the highest ( or lowest in the case of the pdm ) peaks coincide . bootstrapping the pdm gives a superhump period of 0.07938(@xmath73)@xmath0d . a fisher randomisation test gives a false alarm probability , that the quoted period is different to the true period , of between 0 and 0.01 with a 95% confidence level . using the relation between the superhump and orbital periods in equation 1 of @xcite , the approximate orbital period for this system is 0.0765@xmath46d . the ft of the quiescent run s7964 did not show a peak at this period . this could be attributed to the large flickering in the light curve and the run length of less than two orbits . d ) and its first harmonic are marked . ] the crts have detected multiple outbursts of sss1128 - 34 since it was first observed in 2006 . their light curve gives an average quiescent magnitude of @xmath319 , although the individual observations range from @xmath318 to @xmath319.8 . this range is produced by the orbital variation , as shown in our photometry in fig . [ fig : sss1128 ] . s8091 was taken during quiescence , whereas s8094 and s8095 were taken in outburst . the latter two runs were displaced vertically for display . ] each of the light curves show a clear double hump structure . the orbital hump of amplitude @xmath50.4@xmath0mag at phase 1 is produced by the bright spot orbiting into and out of our line of sight . the smaller hump at phase 0.5 is probably produced by the bright spot shining through a semi - transparent disc . during our initial run ( s8091 ) , sss1128 - 34 was in quiescence . the following day the system was 1.7 mag brighter and s8094 and s8095 were taken in outburst . in the ft of the combined runs , the orbital period ( 0.096(@xmath71)@xmath0d ) was chosen from two possible aliases of approximately equal height , based on the power of their respective first harmonics ( see fig . [ fig : sss1128 ft ] ) . the double hump structure produces large amounts of power at the first harmonic , so the ambiguity in the fundamental is removed because only one of the two likely orbital periods has a strong first harmonic . only the mean , not the linear trend , was subtracted from these runs before calculating the ft , as removing the linear trend would greatly affect the shape of s8094 and s8095 . using the metropolis - hastings algorithm to sample the period probability distribution ( based on sine - curve fitting of the light curve ) , we get p@[email protected]@xmath0d . the period from the ft was used as the initial guess and the median and standard deviation are quoted as the period and uncertainty . the ephemeris for maximum light for these observations is @xmath47 which places sss1128 - 34 in the period gap . the crts detected css1221 - 10 when it went into an outburst of magnitude @xmath163.5 on march 24 , 2008 . it does not have a counterpart in the sdss . as seen in fig . [ fig : css1221 ] , css1221 - 10 has shallow eclipses with varying profiles reminiscent of those seen in css0449 - 18 ( fig . [ fig : css0449 ] ) . this indicates that the system has a low inclination and is showing grazing eclipses . h ( s8066 ) to 6.5@xmath0h ( s8071 ) . each run is shifted vertically by the magnitude indicated in parentheses . ] the first harmonic of the orbital period was clear in the pdm periodogram of the combined runs s8056 , s8058 , s8064 , s8068 , s8069 and s8071 . the ephemeris for minimum light is @xmath48 where the uncertainty on the orbital period was obtained by bootstrapping the results . [ fig : css1221_average ] displays the folded average light curve of the observations . the eclipse depth is approximately 0.2@xmath0mag and the system shows an orbital hump . d orbital period . two orbits are shown for display purposes . ] the crts quiescent value for sss1224 - 41 is @xmath2=19.1 , but the light curve shows scatter between 17 and 20.6 . this is due to outbursts and , as we see in our photometry , eclipses . we obtained five runs on sss1224 - 41 over february and march 2012 . the corresponding light curves , with their @xmath50.6@xmath0mag eclipses , are shown in fig . [ fig : sss1224 ] . as it turns out , run s8182 was 142s short of a complete orbital cycle . 0.6@xmath0mag the gaps in the light curves were caused by passing cloud . the indicated vertical offsets are for display purposes . ] the pdm periodogram showed peaks at the orbital period ( 0.2537@xmath0d ) and its first harmonic . bootstrapping the results to refine the orbital period and determine the uncertainty , as well as phase - folding the observations , gave an ephemeris for minimum light of @xmath49 sss1224 - 41 has a quiescent magnitude of @xmath318.4@xmath0mag , but has a number of points that were taken at v@xmath5020 . as shown by our photometry , these points were taken during eclipse . 0.9 , indicating a system inclination of greater than approximately 70@xmath27 . ] we obtained one run on sss1340 - 35 , which contained two eclipses of depth @xmath50.9@xmath0mag ( see fig . [ fig : sss1340 ] ) . the corresponding pdm periodogram gives an orbital period of 0.059(@xmath71)@xmath0d , where the uncertainty was obtained by bootstrapping . the ephemeris for minimum light is @xmath51 the crts detected one superoutburst of css1417 - 18 at @xmath5215 - approximately 5 magnitudes higher than the sdss dr8 quiescent magnitude of @xmath4=20.3 . we obtained two sets of observations of this object , one in outburst at @xmath4=16.8 and one in quiescence at @xmath4=20 . this outburst was not captured by the crts . mag respectively . ] all the observations are given in fig . [ fig : css1417 ] . consider first the quiescent light curves shown in the bottom panel . they show a clear orbital hump profile with a @xmath50.2@xmath0mag amplitude . using the highest amplitude peak in the ft as the initial guess for sine - curve fitting , the mcmc method produced a distribution of periods with median value 0.0845@xmath0d and standard deviation 0.0001@xmath0d , which we quote as the orbital period and uncertainty . the ephemeris for maximum light is @xmath53 now consider the two runs taken in outburst . s7992 did not show a modulation near the orbital period because , at @xmath4=16.8 , the high disc luminosity overwhelmed all other variations . the gap in s7993 , caused by passing cloud , was too large to pick up an orbital variation in this run . in the one day separating s7992 and s7993 , css1417 - 18 declined in magnitude by @xmath4=2@xmath0mag . in dwarf novae ( dne ) the decline rate from outburst is related to the orbital period via equation 3.5 in @xcite : @xmath54 using the 0.0845@xmath0d orbital period obtained from the quiescent runs , we would expect a decline rate of @xmath50.96@xmath0mag@xmath0d@xmath41 , instead of the observed 2@xmath0mag@xmath0d@xmath41 . this value falls below the scatter on the plot of log(@xmath55 ) versus log(p@xmath25 ) for dne ( see figure 3.11 of @xcite ) . further observations of this system in decline are needed . css1556 - 08 is listed as @xmath2=18.4 by the crts . on january 31 , 2011 , @xcite took two spectra of it showing broad h@xmath37 and h@xmath56 lines in emission . in march 2012 it went into superoutburst at [email protected]@xmath0mag , which was observed by @xcite , @xcite and @xcite ( the system is refered to as ot j155631.0 - 080440 ) . @xcite gave an initial value for the superhump period as 0.1 d , which was then refined to [email protected]@xmath0d by @xcite . @xcite determined the superhump period to be [email protected]@xmath0d . the average profile phased on this period is shown in figure 70 of @xcite . our photometry of css1556 - 8 includes one run in outburst ( s8078 ) and two in quiescence ( s8086 and s8081 ) - it is shown in fig . [ fig : css1556 ] . using equation 1 from @xcite and the [email protected]@xmath0d superhump period from @xcite , gives an estimate for the orbital period as [email protected]@xmath0d . although this period is present in the fts of s8078 and of s8081 and s8086 combined , it is not the highest power alias and does not have any high power harmonics . the crts quiescent magnitude for css1727 + 13 is listed as @xmath2=19.5 , but is probably lower because many of the observations only yielded upper limits - most of them at @xmath320.3 . @xcite observed this object spectroscopically . based on the fwhm of the h@xmath37 line ( 1200@xmath0km@xmath0s@xmath41 ) , they concluded that css1727 + 13 has an intermediate inclination . our light curves ( fig . [ fig : css1727 ] ) show flickering , but the ft and pdm of the combined runs only show low amplitude , broad peaks . the @xmath57-year long crts light curve for sss1944 - 42 shows two low states and a high state lasting from august 2009 to june 2011 ; no outbursts have been captured . this behaviour is typical of a magnetic system and the absence of outbursts indicates that it may be a polar . our observations were taken during the high state . only the mean of each of the individual runs was subtracted , not the linear trend . the pdm periodogram of the combined runs has a peak at the orbital period ( 0.063855@xmath0d ) . bootstrapping yields p@[email protected]@xmath0d . the ephemeris for maximum light for this system is @xmath58 fig . [ fig : sss1944 ] shows the average light curve folded on this ephemeris . . two orbits are shown for display purposes . ] over the course of the crts observations ( from april 2005 to september 2012 ) , only one outburst has been recorded for this object . it occurred on the 15th of june 2010 , when the system rose to @xmath2=15.4 from the listed quiescent value of @xmath2=18.8 . on the 16th of may 2011 , we caught this system in an outburst that was not captured by the crts . this run ( s8096 ) , as well as the quiescent run taken two days previously , are both displayed in fig . [ fig : sss2003 ] . the photometry shows eclipses of depth @xmath591.9 , indicating an inclination of greater than approximately 70@xmath27 . the ephemeris for the time of mid - eclipse for these observations is @xmath60 where the orbital period and uncertainty were obtained by bootstrapping the pdm . since the detection of css2054 - 19 on the 29th of august 2009 , numerous outbursts have been captured by the crts , the highest amplitude being @xmath36=3.5 above the quoted quiescent value . the large amplitudes indicated that this may be an su uma system . our observations confirm this . the photometry ( fig . [ fig : css2054 ] ) shows superhumps of amplitude @xmath50.2@xmath0mag , indicating that css2054 - 19 was in superoutburst . 0.2@xmath0mag . s8005 and s8010 are displayed at the correct brightness ; s8008 has been shifted by 0.2@xmath0mag . ] 0.001@xmath0d and [email protected]@xmath0d . ] the ft of the combined runs on this object gave two possible aliases for the superhump period , namely [email protected]@xmath0d and [email protected]@xmath0d , where the former was marginally higher . the corresponding pdm periodogram gave the same results ( see fig . [ fig : css2054ftandpdm ] ) . bootstrapping the pdm produced a peak at both aliases . the first peak , at [email protected]@xmath0d , accounted for 88.5% of the data points , while the second peak ( at [email protected]@xmath0d ) accounted for the remainder . based on these results , we have a preference for the [email protected]@xmath0d period , but ca nt exclude the one day alias . using p@[email protected]@xmath0d , equation 1 of @xcite estimates an orbital period of [email protected]@xmath0d . this estimate would place css2054 - 19 within the period gap . the crts light curve for css2108 - 03 gives the quiescent magnitude as @xmath2=18 and its highest amplitude outburst at @xmath2=14.9 . its counterpart in the sdss dr8 was observed in quiescence at @xmath17=18.7 , @xmath18=18.6 and @xmath4=18 @xcite . the crts light curve for this object gives strong indications that it is an eclipsing system , as it has a number of points that are found more than a magnitude below quiescence . the faintest point is an upper - limit for detection at @xmath2=20.7 . photometry on css2108 - 03 confirms this , showing deep eclipses and large amplitude flickering ( see fig . [ fig : css2108 ] ) . . the shortest run ( s8113 ) is not shown . s8109 and s8111 have been shifted by 1.6 and 3@xmath0mag for display purposes . ] @xcite observed this system photometrically and determined an orbital period of 0.156926798(@xmath79)@xmath0d . independently we determined an orbital period of 0.15699(@xmath75)@xmath0d by bootstrapping the pdm of runs s8107 , s8109 and s8111 combined . the ephemeris of minimum light for our observations is @xmath61 the average of the runs folded on this ephemeris is presented in fig . [ fig : css2108_average ] . it has an eclipse depth of @xmath51.5@xmath0mag . . the eclipse depth is @xmath51.5@xmath0mag . ] [ tbl : summary ] llllll object & type & p@xmath25 & p@xmath30 & @xmath4 & remarks + & & ( d ) & ( d ) & & + css0116 + 09 & dn & 0.06582(@xmath75 ) & - & 18.9@xmath62 , 17.1@xmath63 & eclipsing , [email protected] mag + css0411 - 09 & su & - & 0.06633(@xmath71 ) & 15.4@xmath10 & superhump + css0438 + 00 & dn & 0.06546(@xmath79 ) & - & 19.3@xmath62 & eclipsing , depth@xmath51 mag + css0449 - 18 & dn & 0.15554(@xmath74 ) & - & 17.9@xmath62 & eclipsing , [email protected] mag + sss0501 - 48 & dn & - & - & 17.3@xmath64 & + css0558 + 00 & dn & 0.06808(@xmath71 ) & - & 19.3@xmath64 & + css0902 - 11 & dn & 0.2758(@xmath74)@xmath65 & - & 17.7@xmath64 , 16.5@xmath10 & + css0942 - 19 & su ? & 0.147(@xmath71 ) & - & 19.5@xmath64 & + css1052 - 06 & su & - & 0.07938(@xmath73 ) & 16.2@xmath10 , 18.7@xmath64 & superhump + sss1128 - 34 & dn & 0.0985(@xmath71 ) & - & 18.7@xmath64 , 16@xmath10 & + css1221 - 10 & dn & 0.14615(@xmath71 ) & - & 19.4@xmath66 & eclipsing , [email protected] mag + sss1224 - 41 & dn & 0.25367(@xmath73 ) & - & 19.3@xmath66 & eclipsing , [email protected] mag + sss1340 - 35 & dn & 0.059(@xmath71 ) & - & 18.4@xmath66 & eclipsing , [email protected] mag + css1417 - 18 & su & 0.0845(@xmath71 ) & - & 17.8@xmath10 , 19.9@xmath64 & fast decline from outburst + css1556 - 08 & su & - & 0.089309(@xmath753)@xmath67 & 16.9@xmath10 , 18.2@xmath64 & + css1727 + 13 & dn & - & - & 19.5@xmath64 & + sss1944 - 42 & p & 0.06385(@xmath72 ) & - & 17.4@xmath64 & + sss2003 - 28 & su & 0.05871(@xmath74 ) & - & 19.1@xmath66 , 16.4@xmath68 & eclipsing , [email protected] mag + css2054 - 19 & su & - & 0.09598(@xmath78)@xmath69 & 16.8@xmath10 & superhump + css2108 - 03 & dn & 0.15699(@xmath75)@xmath70 & - & 17.7@xmath66 & eclipsing , [email protected] mag + + we observed 20 cvs identified by the crts with the aim of classifying them , determining orbital periods and selecting targets for further observations with large telescopes . of these 20 systems , only 6 have been observed prior to this work . four systems were confirmed to be cvs by means of spectra and standardized photometry and a superhump period was determined for one of these ( css1556 - 08 , @xcite ) . css0902 - 11 has been followed - up in detail with time - series spectroscopy @xcite and css2108 - 03 was observed photometrically by @xcite . the results are summarised in tab . we determined 12 new orbital periods and independently discovered periods for a further two cvs , namely css0902 - 11 and css2108 - 03 . for three of the systems ( css0411 - 09 , css1052 - 06 and css2054 - 19 ) , we determined superhump periods . css1556 - 08 had a pre - determined superhump period ( @xcite and @xcite ) , but we could not determine an orbital period from our photometry . the remaining 2 cvs did not show any periodic modulations . most of the cvs were dne systems . this is as we should expect , as the crts identifies transients based on their variability . there was also a polar in the sample ( sss1944 - 42 ) , that was picked up by the crts because it showed high and low states that differed by more than a magnitude . the orbital periods of these cvs fall predominantly below the period gap ( see @xcite and @xcite ) , but there were two within the gap ( sss1128 - 34 and possibly css2054 - 19 ) and six above it . the predominance of cvs with periods below the period gap in the crts dataset is not fully explained . @xcite discussed this bias . they compared the cumulative distribution functions of the outburst amplitudes of the crts cvs and those in the survey region that were listed in rkcat @xcite that were not detected by the crts . they found that the crts shows a bias against low amplitude outbursts up to 6@xmath0mag . cvs with larger outburst amplitudes spend more time above the 2@xmath0mag cut - off limit that the crts employs and are thus more likely to be detected . furthermore , they plotted the outburst amplitude versus the orbital period for cvs within the crts footprint ( values from rkcat ) and found a trend for short period dne to have larger outbursts . combined , these two findings indicate that there is a bias in the crts towards shorter period systems @xcite . eight of the cvs in this sample were eclipsing systems . as mentioned in section [ sec : observations ] , the long - term crts light curves of the deeply eclipsing systems sometimes give indications of the eclipses . as each field is observed three times at 10-min intervals , it is possible that one or two of the observations are taken in eclipse . if it is sufficiently deep , these points can show up on the long - term light curve at a magnitude or more below the quiescent level . of the five systems with eclipse depths of more than 0.9@xmath0mag in our sample , four showed these characteristics . a number of the cvs presented in this paper provide promising targets for more in - depth studies with larger telescopes - such as the eclipsing systems ( css0116 + 09 , css0438 + 00 , css0449 - 18 , css1221 - 10 , sss1224 - 41 , sss1340 - 35 , sss2003 - 28 and css2108 - 03 ) , which through eclipse deconvolution can yield parameters such as the inclination and mass ratio . css1417 - 18 shows an unusually fast decline from outburst . the new orbital periods will also contribute towards population and evolutionary studies . we thank the anonymous referee for the suggestions and comments which helped to improve this paper . the authors also gratefully acknowledge funding from the south african square kilometre array project , the erasmus mundus programme sapient , the national research foundation of south africa ( nrf ) , the nederlandse organisatie voor wetenschappelijk onderzoek ( the dutch organisation for science research ) , the university of cape town , the national astrophysics and space science programme , the south african astronomical observatory ( saao ) and the claude leon foundation postdoctoral fellowship program . funding for the shoc camera was provided by the nrf , specifically the research infrastructure support programme s national equipment programme ( grant uid # 74428 ) . this research uses observations made at the saao and has also made use of nasa s astrophysics data system bibliographic services and of sdss - iii data . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions ( see http://www.sdss3.org ) , the national science foundation , and the u.s . department of energy office of science . the css survey is funded by the national aeronautics and space administration under grant no . nng05gf22 g issued through the science mission directorate near - earth objects observations program . the crts survey is supported by the u.s . national science foundation under grants ast-0909182 and ast-1313422 . h. et al . , 2011 , apjs , 193 , 29 e. et al . , 2012 , aas / division for planetary sciences meeting abstracts , 44 , 210.13 r. et al . , 2013 , pasp , 152 , 976 a.j . et al . , 2009 , aj , 696 , 870 s.g . et al . , 2011 , arxiv:1102.5004 s.g . et al . , 2008 , astronomische nachrichten , 329 , 263 b.t . et al . , 2009 , mnras , 397 , 2170 b. , 1979 , ann . , 7 , 1 a.a.s . , odonoghue d. , fourie p. , rust m. , sass c. , stoffels j. et al . , 2011a , epsc - dps joint meeting 2011 , 1173 , w.k . , 1970 , biometrika , 57 , 97 s. et al . , 2005 , aj , 130 , 873 t. et al . , 2013 , pasj , 65 , 23 t. , 2013 , http://ooruri.kusastro.kyoto-u.ac.jp/mailarchive/vsnet-alert/15618 c. , 2006 , mnras , 373 , 484 c. , baraffe i. , patterson j. , 2011 , apjs , 194 , 28 a. , 1992 , aj , 104 , 340 s.p . , dhillon v.s . , marsh t.r . , gnsicke b.t . , southworth j. , baraffe i. , watson c.a . , copperwheat c. , 2008 , mnras , 388 , 1582 n. , rosenbluth a.w . , rosenbluth m.n . , et al . , 1953 , j. chem . , 21 , 1087 a.f . , nemec j.m . , 1985 , aj , 90 , 2317 d. , 1995 , baltic astron . , 4 , 519 , 2012 , http://ooruri.kusastro.kyoto-u.ac.jp/mailarchive/vsnet-alert/14406 , 2012b , http://ooruri.kusastro.kyoto-u.ac.jp/mailarchive/vsnet-alert/14416 h. , kolb u. , 2003 , aa , 404 , 301 c.d . et al . , 2011 , mnras , 415 , 2025 r.f . , 1978 , apj , 224 , 953 d. et al . , 2009 , bulletin of the american astron . soc .. 41 , 460.01 j.r . , skinner j.n . , 2012 , aj , 144 , 81 d. , 1986 , proc . spie , 627 , 733 b. , 1995 , cataclysmic variable stars . cambridge univ . press , cambridge b. , 2004 , pasp , 116 , 115 b. , nather e.r . , 1971 , 152 , 219 b. , woudt p.a . , pretorius m.l . , 2003 , mnras , 344 , 1193 p.a . , warner b. , odonoghue d. , buckley d.a.h . , still m. , romero - colemero e. , visnen p. , 2010 , mnras , 401 , 500 p.a . , warner b. , de bude d. , macfarlane s. , schurch m.p.e . , zietsman e. , 2012 , mnras , 421 , 2414 p.a . et al . , 2012b , mnras , 427 , 1004 e. , marsh t.r . , morales - rueda l. , 2006 , mnras , 369 , 805 e.h . , robinson e.l . , 1987 , apj , 321 , 813

we model the extinction profiles observed in the small and large magellanic clouds with a synthetic population of dust grains consisting by core - mantle particles and a collection of freeflying polycyclic aromatic hydrocarbons . all different flavors of the extinction curves observed in the magellanic clouds can be described by the present model , that has been previously ( successfully ) applied to a large sample of diffuse and translucent lines of sight in the milky way . we find that in the magellanic clouds the extinction produced by classic grains is generally larger than absorption by polycyclic aromatic hydrocarbons . within this model , the nonlinear faruv rise is accounted for by polycyclic aromatic hydrocarbons , whose presence in turn is always associated to a gap in the size distribution of classical particles . this hints either a physical connection between ( e.g. , a common cause for ) polycyclic aromatic hydrocarbons and the absence of middlesized dust particles , or the need for an additional component in the model , that can account for the nonlinear faruv rise without contributing to the uv bump at nm , e.g. , nanodiamonds .

1507.08550

interstellar extinction curves ( isecs ) are now available for many lines of sight in the milky way galaxy ( mwg ) , in the small and large magellanic clouds ( smc , lmc ) , in m31 , in nearby galaxies , and increasingly in objects at high redshift . the isecs have a great variety of shapes . this evidence may seem to suggest that grain models must be complex , allowing for different carriers for each of the different components of the isecs . however , all the measured extinction profiles are clearly members of the same family , and in fact , isecs can be phenomenologically described by means of a very limited number of empirical parameters @xcite . such a smooth topological transition among extinction profiles may be naturally interpreted as the response of dust to the local physics . dust grains are composed mainly of silicates and carbons , as made clear by spectroscopic evidence and depletion data ( e.g. , @xcite ) . they are formed in the envelopes of cool stars and in novae and supernovae , and are destroyed in shocks , but with different efficiencies , silicates being more resistant than carbons @xcite . even if the silicates and carbons can be produced as separate populations , the destruction of solid carbons to atoms and small molecules will lead to carbon deposition on silicate cores . according to the scenario originally envisaged by @xcite , gaseous carbon in diffuse clouds is gradually deposited on silicate cores in hydrogen rich @xmath1 form , and is eventually annealed and darkened to graphitic ( h - poor ) @xmath2 kind by the interstellar radiation field . in other words , dust is not immutable and is expected to evolve in time ( e.g. , @xcite ) . such dust formation and evolution scenario has been implemented in a model the [ cm]@xmath3 model @xcite successfully applied to a sample of 329 galactic lines of sights @xcite by @xcite . subsequently , the carbon evolution prescriptions of this scenario were condensed in a set of timedependent equations , describing carbon deposition onto the silicate cores and its following photo - darkening , and matched to the observationally based inferences of dust physical parameters @xcite , closing the circle . most of the isecs in the @xcite sample correspond to dust mantles that restarted growing after being almost completely removed , with the leftover fully rehydrogenated , by , e.g. , occasional shocks . a natural extension of this work is its application to ( and test with ) environments markedly different from the interstellar medium of the solar neighborhood , in which the life cycle of the interstellar medium is known to be very different either globally , e.g. , much more frequent shocks as in starburst galaxies , or locally , because of independent observational constraints on individual lines of sight . in this paper we shall apply the [ cm]@xmath3 model to a number of lines of sight in the magellanic clouds ( mcs ) , whose data have been provided by @xcite and @xcite . these two irregular dwarf galaxies are nearby ( @xmath4 kpc ) , gasrich , actively star - forming , and have sub - solar metallicities ( e.g. , @xcite ) . the smc average isec presents a featureless profile with no 217.5 nm bump and a steep far ultraviolet rise , while the lmc one shows a hint of the bump and a firm far - ultraviolet rise , stronger than in the mwg but less sharp than in the smc . we describe the specific lines of sight , their known physical properties , and how they are modelled in section [ mod ] , the resulting characterisation of dust grains along such extinction paths in section [ res ] , section [ discussection ] contains our discussion and the last section summarises our conclusions . @xcite and @xcite exploited a collection of newly taken and archive data to obtain individual extinction curves for a total of 24 lines of sight : 19 for lmc and 5 for smc . the list of sight lines reported in table [ one ] contains also their absolute visual magnitude , @xmath5 , the ratio of total to selective extinction , @xmath6 , and the gas to dust ratio expressed as @xmath7 @xmath8 mag@xmath9 . this quantity has been constructed for specific lines of sight using data taken from @xcite and references therein : the atomic hydrogen column density is derived from absorption lyman-@xmath10 profiles in all cases but sk-69 279 , for which we used data from the hydrogen 21 cm emission ; the molecular hydrogen column density is obtained from _ fuse _ h@xmath11 absorption measurements when available , or otherwise exploiting the relation linking @xmath12 to @xmath13 , as found by @xcite for the mwg . in addition to individual lines of sight , we also considered the `` average '' extinction curves given by @xcite for the smc bar , the lmc 2 supershell ( ss ) , and the overall lmc average . these average extinction curves were treated exactly as the individual ones , to see if this resulted in a sensible description of representative dust properties for the areas they refer to . lccc + los & @xmath5 & @xmath6 & @xmath7 + & ( mag ) & & ( @xmath14 @xmath8 mag@xmath9 ) + + azv18 & 0.49@xmath15 & 2.90@xmath15 & 67.19@xmath16 + azv23 & 0.48@xmath3 & 2.65@xmath3 & 49.29@xmath17 + azv214 & 0.35@xmath3 & 2.40@xmath3 & 41.03@xmath17 + azv398 & 0.68@xmath3 & 3.14@xmath3 & 55.63@xmath17 + _ smc bar _ & & 2.74@xmath3 & 24.55@xmath18 + + azv456 & 0.57@xmath15 & 2.19@xmath15 & 10.39@xmath16 + + sk-66 19 & 0.86@xmath3 & 3.44@xmath3 & 29.59@xmath16 + sk-66 88 & 1.03@xmath3 & 3.67@xmath3 & 20.71@xmath17 + sk-67 2 & 0.56@xmath15 & 3.75@xmath15 & 18.55@xmath16 + sk-68 23 & 1.04@xmath3 & 3.35@xmath3 & 6.03@xmath17 + sk-68 26 & 0.64@xmath15 & 3.45@xmath15 & 20.77@xmath16 + sk-68 129 & 0.57@xmath15 & 3.37@xmath15 & 32.73@xmath16 + sk-69 108 & 0.98@xmath3 & 3.15@xmath3 & 10.88@xmath17 + sk-69 206 & 0.96@xmath3 & 3.68@xmath3 & 35.33@xmath17 + sk-69 210 & 1.36@xmath3 & 3.32@xmath3 & 25.75@xmath17 + sk-69 213 & 0.63@xmath3 & 3.96@xmath3 & 22.55@xmath17 + _ lmc _ & & 3.41@xmath3 & 12.88@xmath18 + + sk-68 140 & 0.67@xmath15 & 3.34@xmath15 & 28.77@xmath16 + sk-68 155 & 0.56@xmath15 & 2.81@xmath15 & 21.82@xmath16 + sk-69 228 & 0.53@xmath15 & 3.54@xmath15 & 28.50@xmath16 + sk-69 256 & 0.11@xmath3 & 0.64@xmath3 & 15.67@xmath17 + sk-69 265 & 0.32@xmath3 & 1.68@xmath3 & 24.43@xmath17 + sk-69 270 & 0.44@xmath3 & 2.34@xmath3 & 18.99@xmath16 + sk-69 279 & 0.75@xmath15 & 3.54@xmath15 & 21.80@xmath19 + sk-69 280 & 0.56@xmath3 & 3.12@xmath3 & 23.03@xmath17 + sk-70 116 & 0.65@xmath3 & 3.41@xmath3 & 16.82@xmath17 + _ lmc2 ss _ & & 2.76@xmath3 & 21.88@xmath18 + @xmath15 @xcite ; @xmath3 @xcite ; @xmath20 atomic hydrogen from lyman @xmath10 ; @xmath21 molecular hydrogen from _ fuse _ ; @xmath22 molecular hydrogen estimate ( see text ) ; @xmath23 atomic hydrogen from 21 cm ; @xmath24 @xcite [ one ] the dust model consists in a distribution of core - mantle grains and a mixture of polycyclic aromatic hydrocarbons ( pahs ) . classical grains are constructed assembling four concentric components : a cavity , whose volume is in fixed proportion @xmath25 for all core sizes ; the silicate shell of radius @xmath26 ; the @xmath2 layer of thickness @xmath27 , @xmath28 being the mantle thickness ; and the @xmath1 layer of thickness @xmath29 . we consider a size distribution for the silicate cores , given by a power law @xmath30 . such distribution allows for a gap in particle sizes , so that two populations of dust grains , big and small ones , may be present , each characterised by lower ( @xmath31 , @xmath32 ) and upper size ( @xmath33 , @xmath34 ) limits in the distribution . the molecular component is represented by a mixture of 54 pahs in the size range @xmath35 c atoms , in the charge states 0 , @xmath36 , and + 2 @xcite . the equations describing the model have been reported in several papers ( e.g. , @xcite ) , and are not repeated here . the model produces a theoretical extinction curve as a function of its parameters . a leastsquares procedure @xcite yields the bestfitting model parameter values for each line of sight . the variancecovariance matrix for the bestfitting parameters , measuring how tightly they are constrained by the fit , is obtained by synthetic statistics , by repeating the fit multiple times on data perturbed according to their errors . the details of this procedure , and the modifications in the mpfit levenbergmarquardt implementation to cope with underdetermination of some parameters , are described in @xcite . the evolutionary model of carbon mantles is detailed in @xcite . the carbon cycle in and out of dust is described by a set of ordinary differential equations that follow the evolution in time of the available gasphase carbon , the deposited polymeric @xmath1 carbon , and its radiation - annealed @xmath2 counterpart . the main free parameters are the photodarkening time the time scale for graphitization ( e.g. , @xcite ) and the initial conditions . the differential equations thus constitute a oneparameter system : solutions starting from the same initial conditions , but with different photodarkening times , have no other points in common but the starting point and the asymptotic limit . of course , the evolutionary model also has a parametric dependence on the assumed environmental conditions , i.e. the elemental abundances and so on . however , this dependence is rather mild , and the solutions of the model retain their overall behaviour with minor differences for any sensible environmental conditions @xcite . fitted isecs are shown in figs . [ smcfitcurves ] ( smc ) and [ lmcfitcurves ] ( lmc ) , while the inferred dust parameters are reported in tables [ classictable ] ( classical dust ) and [ pahtable ] ( pahs ) . in the same tables we also show the results for the average extinction profiles . [ cols="^,^ " , ] cccccccccccc @xmath37 & @xmath38 & @xmath39 & @xmath28 & @xmath40 & @xmath41 & @xmath42 & @xmath43 & @xmath44 & @xmath45 ( @xmath46 ) & @xmath47 & @xmath48 & & & & & & & & & & & @xmath49(@xmath8 ) + + + azv18 & 1.80 ( 0.14 ) & 0.00 ( @xmath50 ) & 2.0 ( 0.9 ) & 0.00 ( @xmath50 ) & 7 ( 3 ) & 42 ( 3 ) & 130 ( 18 ) & 282 ( 110 ) & 87 ( 19 ) 3.51 ( 0.04 ) & 137 ( 18 ) + azv23@xmath51 & 3.336 ( 0.003 ) & 0.67 ( @xmath50 ) & 4.70 ( 0.02 ) & 0.77 ( @xmath50 ) & 5.00 ( @xmath50 ) & 136.0 ( 0.7 ) & & & 0.0 ( @xmath50 ) & 3.3199 ( 0.0007 ) & 0.0 ( 0.0 ) + azv23@xmath52 & 3.87 ( 0.13 ) & 0.00 ( @xmath50 ) & 0.3 ( 0.5 ) & 0.6 ( 0.4 ) & 5.07 ( 0.22 ) & 45.7 ( 2.6 ) & 118 ( 17 ) & 339 ( 200 ) & 72 ( 18 ) & 3.760 ( 0.016 ) & 178 10 + azv214 & 1.79 ( 0.12 ) & 0.55 ( 0.06 ) & 0.64 ( 0.14 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 607 ( 240 ) & & & 0.0 ( @xmath50 ) & 3.527 ( 0.015 ) & 68 ( 13 ) + azv398 & 1.83 ( 0.12 ) & 0.01 ( 0.02 ) & 1.7 ( 0.7 ) & 0.00 ( @xmath50 ) & 5.1 ( 0.4 ) & 44.0 ( 2.3 ) & 153 ( 50 ) & 345 ( 160 ) & 109 ( 50 ) & 3.52 0.03 & 179 21 + _ smc bar _ & 2.10 ( 0.06 ) & 0.00 ( @xmath50 ) & 0.6 ( 0.3 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 44 ( 4 ) & 130 ( 8) & 269 ( 16 ) & 86 ( 10 ) & 3.583 ( 0.016 ) & + + + + + azv456 & 1.21 ( 0.11 ) & 0.39 ( 0.05 ) & 1.5 ( 0.7 ) & 0.00 ( @xmath50 ) & 6.2 ( 1.6 ) & 66 ( 4 ) & 195 ( 26 ) & 410 ( 15 ) & 129 ( 28 ) & 3.458 ( 0.021 ) & 251 ( 11 ) + + + + + sk-66 19 & 1.00 ( 0.04 ) & 0.24 ( 0.03 ) & 1.5 ( 0.6 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 26 ( 4 ) & 60 ( 8) & 530 ( 30 ) & 34 ( 5 ) & 3.489 ( 0.007 ) & ( 97 ) ( 14 ) + sk-66 88 & 1.026 ( 0.022 ) & 0.29 ( 0.02 ) & 2.7 ( 0.4 ) & 0.00 ( @xmath50 ) & 5.4 ( 1.5 ) & 34 ( 5 ) & 62 ( 15 ) & 542 ( 29 ) & 28 ( 10 ) & 3.487 ( 0.007 ) & 109 ( 19 ) + sk-67 2 & 1.0706 ( 0.0027 ) & 0.00 ( @xmath50 ) & 1.09 ( 0.05 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 37.54 ( 0.01 ) & 152.6 ( 0.4 ) & 337 ( 12 ) & 115.1 ( 0.4 ) & 3.4891 ( @xmath53 ) & 192.3 ( 2.5 ) + sk-68 23 & 0.966 ( 0.008 ) & 0.26 ( 0.01 ) & 1.1 ( 0.4 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 19.2 ( 1.9 ) & 62.0 ( 2.6 ) & 538 ( 9 ) & 42.8 ( 1.8 ) & 3.4896 ( 0.0021 ) & 145 ( 8) + sk-68 26 & 0.966 ( 0.029 ) & 0.40 ( 0.03 ) & 1.7 ( 0.5 ) & 0.27 ( 0.17 ) & 5.00 ( @xmath50 ) & 24 ( 4 ) & 47.9 ( 1.8 ) & 610 ( 16 ) & 24 ( 6 ) & 3.479 ( 0.004 ) & 88 ( 7 ) + sk-68 129 & 1.00 ( 0.08 ) & 0.29 ( 0.09 ) & 1.5 ( 0.9 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 43 ( 10 ) & 83 ( 26 ) & 544 ( 80 ) & 40 ( 17 ) & 3.481 ( 0.015 ) & 89 ( 19 ) + sk-69 108 & 0.97 ( 0.04 ) & 0.34 ( 0.04 ) & 1.3 ( 0.4 ) & 0.00 ( 0.01 ) & 5.05 ( 0.21 ) & 26 ( 4 ) & 59 ( 6 ) & 539 ( 50 ) & 34 ( 3 ) & 3.480 ( 0.006 ) & 168 ( 22 ) + sk-69 206 & 1.093 ( 0.021 ) & 0.48 ( 0.02 ) & 1.64 ( 0.27 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 27.3 ( 2.4 ) & 64.4 ( 2.0 ) & 589 ( 40 ) & 37.2 ( 2.4 ) & 3.452 ( 0.005 ) & 72 ( 9 ) + sk-69 210 & 1.038 ( 0.009 ) & 0.43 ( 0.01 ) & 0.56 ( 0.09 ) & 0.58 ( 0.09 ) & 5.00 ( @xmath50 ) & 33.3 ( 1.1 ) & 74.4 ( 1.6 ) & 545 ( 40 ) & 41.1 ( 1.6 ) & 3.474 ( 0.004 ) & 185 ( 12 ) + sk-69 213 & 1.01 ( 0.05 ) & 0.25 ( 0.09 ) & 1.15 ( 0.25 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 53 ( 4 ) & 137 ( 16 ) & 485 ( 60 ) & 84 ( 16 ) & 3.492 ( 0.009 ) & 153 ( 16 ) + _ lmc _ & 1.14 ( 0.20 ) & 0.25 ( 0.13 ) & 3.5 ( 1.5 ) & 0.01 ( 0.02 ) & 10 ( 6 ) & 47 ( 6 ) & 158 ( 70 ) & 435 ( 60 ) & 112 ( 70 ) & 3.46 ( 0.04 ) & + + + + + sk-68 140 & 1.08 ( 0.11 ) & 0.16 ( 0.09 ) & 3.2 ( 1.1 ) & 0.00 ( @xmath50 ) & 5.5 ( 1.2 ) & 45 ( 4 ) & 122 ( 13 ) & 413 ( 90 ) & 77 ( 13 ) & 3.473 ( 0.020 ) & 146 ( 12 ) + sk-68 155 & 1.12 ( 0.05 ) & 0.39 ( 0.08 ) & 0.88 ( 0.15 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 40 ( 5 ) & 51 ( 3 ) & 684 ( 210 ) & 11 ( 5 ) & 3.493 ( 0.014 ) & 80 ( 9 ) + sk-69 228 & 0.910 ( 0.025 ) & 0.25 ( 0.08 ) & 4.7 ( 0.3 ) & 0.00 ( @xmath50 ) & 5.2 ( 0.6 ) & 35.1 ( 1.5 ) & 36.9 ( 1.9 ) & 574 ( 90 ) & 1.3 ( 1.1 ) & 3.461 ( 0.008 ) & 28 ( 3 ) + sk-69 256@xmath54 & 15.62 ( 0.12 ) & 0.67 ( @xmath50 ) & 1.7 ( 1.3 ) & 0.27 ( 0.22 ) & 7 ( 5 ) & 13 ( 8) & 96 ( 22 ) & 109 ( 26 ) & 82 ( 21 ) & 3.28 ( 0.03 ) & 104 ( 20 ) + sk-69 256@xmath55 & 16.8 ( 1.2 ) & 0.611 ( 0.014 ) & 1.10 ( 0.04 ) & 0.45 ( 0.19 ) & 5.00 ( @xmath50 ) & 11.3 ( 1.1 ) & 120 ( 5 ) & 162 ( 7 ) & 109 ( 6 ) & 3.35 ( 0.04 ) & 0.0 ( 0.0 ) + sk-69 265 & 1.20 ( 0.04 ) & 0.42 ( 0.05 ) & 0.68 ( 0.10 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 56 ( 5 ) & 62 ( 8) & 497 ( 120 ) & 4 ( 5 ) & 3.454 ( 0.009 ) & 56 ( 11 ) + sk-69 270 & 1.19 ( 0.04 ) & 0.01 ( 0.02 ) & 1.03 ( 0.28 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 47.2 ( 2.3 ) & 131 ( 15 ) & 296 ( 100 ) & 83 ( 15 ) & 3.476 ( 0.010 ) & 179 ( 11 ) + sk-69 279 & 0.92 ( 0.06 ) & 0.17 ( 0.09 ) & 2.3 ( 0.5 ) & 0.00 ( @xmath50 ) & 5.5 ( 2.4 ) & 39 ( 11 ) & 52 ( 22 ) & 630 ( 160 ) & 13 ( 14 ) & 3.488 ( 0.012 ) & 48 ( 13 ) + sk-69 280 & 0.91 ( 0.04 ) & 0.05 ( 0.04 ) & 1.0 ( 0.5 ) & 0.00 ( @xmath50 ) & 13.0 ( 2.3 ) & 40.7 ( 2.9 ) & 81 ( 8) & 425 ( 50 ) & 41 ( 9 ) & 3.493 ( 0.012 ) & 125 ( 13 ) + sk-70 116 & 1.124 ( 0.017 ) & 0.51 ( 0.03 ) & 0.95 ( 0.17 ) & 0.00 ( @xmath50 ) & 5.00 ( @xmath50 ) & 25.2 ( 1.8 ) & 71.3 ( 2.8 ) & 692 ( 160 ) & 46.1 ( 2.8 ) & 3.464 ( 0.007 ) & 61.1 ( 6.7 ) + _ lmc2 ss _ & 1.09 ( 0.19 ) & 0.21 ( 0.14 ) & 2.2 ( 1.6 ) & 0.00 ( @xmath50 ) & 8 ( 4 ) & 52 ( 8) & 149 ( 60 ) & 415 ( 80 ) & 98 ( 60 ) & 3.45 ( 0.04 ) & + + we adopt @xmath56 , @xmath57 , @xmath58 + @xmath51 fit solution for azv23 without pahs + @xmath52 fit solution for azv23 including pahs + @xmath54 fit solution for sk69 256 with similar contributions by classical dust and pahs + @xmath55 fit solution for sk69 256 without pahs [ classictable ] cccc + + & & & + & & & + + + azv18 & 2.8 ( 0.4 ) & -0.053 ( 0.012 ) & 0.051 ( 0.013 ) + azv214 & 1.9 ( 0.4 ) & -0.007 ( 0.021 ) & 0.073 ( 0.010 ) + azv23@xmath51 & 3.72 ( 0.21 ) & 0.007 ( 0.012 ) & 0.075 ( 0.004 ) + azv398 & 2.6 ( 0.3 ) & -0.010 ( 0.017 ) & 0.068 ( 0.004 ) + _ smc bar _ & 3.4 ( 0.3 ) & 0.006 ( 0.022 ) & 0.072 ( 0.010 ) + + azv456 & 4.40 ( 0.19 ) & 0.060 ( 0.008 ) & 0.045 ( 0.008 ) + + sk-66 19 & 1.13 ( 0.17 ) & 0.015 ( 0.020 ) & 0.063 ( 0.011 ) + sk-66 88 & 1.06 ( 0.19 ) & -0.003 ( 0.013 ) & 0.064 ( 0.006 ) + sk-67 2 & 3.43 ( 0.04 ) & 0.025 ( 0.001 ) & 0.051 ( 0.001 ) + sk-68 129 & 1.6 ( 0.3 ) & -0.012 ( 0.011 ) & 0.071 ( 0.005 ) + sk-68 23 & 1.40 ( 0.08 ) & -0.001 ( 0.009 ) & 0.063 ( 0.005 ) + sk-68 26 & 1.38 ( 0.10 ) & 0.000 ( 0.010 ) & 0.056 ( 0.006 ) + sk-69 108 & 1.72 ( 0.22 ) & 0.004 ( 0.016 ) & 0.055 ( 0.009 ) + sk-69 206 & 0.75 ( 0.09 ) & 0.015 ( 0.008 ) & 0.056 ( 0.007 ) + sk-69 210 & 1.36 ( 0.09 ) & 0.026 ( 0.006 ) & 0.050 ( 0.006 ) + sk-69 213 & 2.43 ( 0.26 ) & 0.034 ( 0.010 ) & 0.065 ( 0.006 ) + _ lmc _ & 2.6 ( 0.6 ) & 0.019 ( 0.022 ) & 0.068 ( 0.010 ) + + sk-68 140 & 2.18 ( 0.17 ) & -0.030 ( 0.010 ) & 0.066 ( 0.007 ) + sk-68 155 & 1.43 ( 0.16 ) & -0.009 ( 0.013 ) & 0.076 ( 0.005 ) + sk-69 228 & 0.53 ( 0.06 ) & -0.013 ( 0.021 ) & 0.069 ( 0.012 ) + sk-69 256@xmath52 & 9.5 ( 1.8 ) & 0.05 ( 0.06 ) & 0.07 ( 0.03 ) + sk-69 265 & 1.8 ( 0.4 ) & -0.005 ( 0.007 ) & 0.061 ( 0.007 ) + sk-69 270 & 4.06 ( 0.26 ) & 0.007 ( 0.009 ) & 0.073 ( 0.007 ) + sk-69 279 & 0.64 ( 0.17 ) & 0.010 ( 0.018 ) & 0.065 ( 0.006 ) + sk-69 280 & 2.24 ( 0.23 ) & -0.003 ( 0.010 ) & 0.055 ( 0.007 ) + sk-70 116 & 0.94 ( 0.10 ) & -0.006 ( 0.016 ) & 0.074 ( 0.007 ) + _ lmc2 ss _ & 2.7 ( 0.6 ) & 0.008 ( 0.020 ) & 0.068 ( 0.010 ) + we adopt @xmath59 , @xmath60 + @xmath51 fit solution for azv23 including pahs + @xmath52 fit solution for sk69 256 with similar contributions by classical dust and pahs + lccccccccc + + & si / h ( ppm ) & & & si / c + & & & mantle & pah & total & + + + azv18 & 4.8 ( 1.0 ) & & 2.7 ( 1.3 ) & 12.1 ( 1.6 ) & 14.8 ( 1.6 ) & 0.3 ( 0.1 ) + azv214 & 7.6 ( 1.2 ) & & 2.0 ( 0.4 ) & 11.3 ( 2.2 ) & 13.3 ( 2.5 ) & 0.6 ( 0.2 ) + azv23 @xmath51 & 6.8 ( @xmath61 ) & & 65.0 ( 0.3 ) & & 65.0 ( 0.3 ) & 0.3 ( 0.1 ) + azv23 @xmath52 & 6.9 ( 1.4 ) & & 0.7 ( 1.2 ) & 20.0 ( 1.1 ) & 20.7 ( 1.2 ) & 0.1 ( @xmath61 ) + azv398 & 7.1 ( 1.7 ) & & 3.5 ( 1.5 ) & 14.9 ( 1.8 ) & 18.4 ( 2.4 ) & 0.4 ( 0.1 ) + + azv456 & 17.8 ( 1.1 ) & & 9 ( 5 ) & 93 ( 4 ) & 102 ( 7 ) & 0.2 ( @xmath61 ) + + sk-66 19 & 13.8 ( 0.6 ) & & 4.0 ( 1.5 ) & 13.1 ( 1.9 ) & 17.1 ( 2.9 ) & 0.8 ( 0.1 ) + sk-66 88 & 20.5 ( 0.8 ) & & 10.7 ( 1.2 ) & 19 ( 3 ) & 29 ( 4 ) & 0.7 ( 0.1 ) + sk-67 2 & 18.1 ( 0.7 ) & & 5.4 ( 0.2 ) & 69.4 ( 0.9 ) & 74.8 ( 1.1 ) & 0.2 ( @xmath61 ) + sk-68 129 & 12.0 ( 1.4 ) & & 4.0 ( 2.3 ) & 16 ( 3 ) & 20 ( 5 ) & 0.6 ( 0.2 ) + sk-68 23 & 61.5 ( 0.8 ) & & 14 ( 4 ) & 78 ( 4 ) & 91 ( 8) & 0.7 ( 0.1 ) + sk-68 26 & 18.1 ( 0.6 ) & & 10 ( 4 ) & 22.8 ( 1.7 ) & 33 ( 5 ) & 0.6 ( 0.1 ) + sk-69 108 & 31.0 ( 1.1 ) & & 9.2 ( 2.1 ) & 50 ( 6 ) & 59 ( 8) & 0.5 ( 0.1 ) + sk-69 206 & 12.0 ( 0.3 ) & & 5.0 ( 0.8 ) & 7.9 ( 1.0 ) & 12.9 ( 1.3 ) & 0.9 ( 0.1 ) + sk-69 210 & 13.5 ( 0.5 ) & & 4.7 ( 0.7 ) & 17.6 ( 1.1 ) & 22.3 ( 1.0 ) & 0.6 ( @xmath61 ) + sk-69 213 & 17.4 ( 1.9 ) & & 5.3 ( 1.3 ) & 43 ( 5 ) & 48 ( 4 ) & 0.4 ( 0.1 ) + + sk-68 140 & 11.5 ( 1.3 ) & & 7.8 ( 2.7 ) & 25.3 ( 2.0 ) & 33 ( 4 ) & 0.4 ( 0.1 ) + sk-68 155 & 17.2 ( 2.4 ) & & 3.9 ( 0.7 ) & 18.4 ( 2.0 ) & 22.2 ( 1.9 ) & 0.8 ( 0.1 ) + sk-69 228 & 16.1 ( 2.2 ) & & 12.7(1.1 ) & 6.5 ( 0.7 ) & 19.3 ( 1.5 ) & 0.9 ( 0.2 ) + sk-69 256@xmath54 & 5.0 ( 1.0 ) & & 18 ( 10 ) & 39 ( 7 ) & 57 ( 13 ) & 0.1 ( @xmath61 ) + sk-69 256@xmath55 & 10.4 ( 0.4 ) & & 21(5 ) & & 21 ( 5 ) & 0.5 ( 0.2 ) + sk-69 265 & 10.0 ( 1.4 ) & & 2.0 ( 0.3 ) & 12.1 ( 2.4 ) & 14.2 ( 2.4 ) & 0.7 ( 0.2 ) + sk-69 270 & 13.2 ( 2.7 ) & & 3.9 ( 1.1 ) & 50 ( 3 ) & 53.9 ( 2.9 ) & 0.3 ( 0.1 ) + sk-69 279 & 23 ( 4 ) & & 8.7 ( 1.9 ) & 10.4 ( 2.8 ) & 19 ( 3 ) & 1.2 ( 0.3 ) + sk-69 280 & 14.7 ( 1.1 ) & & 2.0 ( 1.2 ) & 30 ( 3 ) & 32 ( 3 ) & 0.5 ( 0.1 ) + sk-70 116 & 23.3 ( 1.7 ) & & 5.9 ( 1.0 ) & 19.0 ( 2.1 ) & 25.0 ( 2.2 ) & 0.9 ( 0.1 ) + we adopt @xmath62 + @xmath51 fit solution for azv23 obtained with classical dust only ( no pahs ) + @xmath52 fit solution for azv23 including pahs + @xmath54 fit solution for sk69 256 with similar contributions by classical dust and pahs + @xmath55 fit solution for sk69 256 obtained with classical dust only ( no pahs ) as described in @xcite , the fit is severely underconstrained , mainly due to degeneracy in how many different pah mixtures can add up to very nearly the same cumulative cross section . as a result , only some of the parameters are well determined by the fit , namely the ones defining the classical dust , and some collective properties of the pah mixture . generally , the respective contributions of classic dust and pahs to the modeled isecs ( as shown in figs . [ smcfitcurves ] and [ lmcfitcurves ] ) are very well constrained . this was the case for all isecs we fitted in the mwg and for all but two exceptions in the mcs , i.e. azv23 and sk69 256 . in the latter two exceptions , to add insult to injury , there is more than one acceptable solution , with qualitatively different shapes of the classic dust and pahs contributions that produce almost the same profile . these degenerate fits are shown in fig . [ degeneratefits ] , and parameters for both solutions are given in tables [ classictable ] , [ masstable ] , and [ classicmasstab ] . this reflects an actual physical degeneracy in the possible ways to account for the observed shapes of these specific extinction curves , which could be broken by using additional information ( e.g. , abundance constraints or observations of aromatic features in emission ) . azv23 has an extremely weak bump , if any , and a correspondingly weak nonlinear faruv rise . this can be achieved by a plain monomodal distribution of classic dust with negligible pahs ( red curve in fig . [ smcfitcurves ] ) ; alternatively the same isec results from opening a gap in the grain distribution , thereby creating a broad break in the classical dust extinction which almost perfectly cancels with the bump due to pah absorption . @xmath3 model . top panel : observed ( shaded area ) and fitted ( solid black curve ) isec towards sk67 2 ; the dashed line shows the contribution of classical grains to the extinction . in blue , the total extinction ( solid ) and the contribution of classical dust grains ( dashed line ) resulting from exactly the same [ cm]@xmath3 model with the only difference of artificially removing the gap in the size distribution . bottom panel : extinction produced by classical dust grains in the bestfitting [ cm]@xmath3 model ( black ) , in the same model with the size gap artificially removed ( blue ) , and the contribution of dust grains with sizes in the gap ( red ) . ] to make evident the effect on the extinction curve of a gap in the size distribution of classical dust grains , in the top panel of fig . [ figuranogap ] we show the isec towards sk67 2 , the best fitting extinction curve ( which has a gap ) and its components in black , with superimposed , in blue , the extinction curve that would result from exactly the same extinction model but with no gap . in the case of sk69 256 there is a hint of a bump , but due also to the low signaltonoise ratio a similar degeneracy of acceptable solutions occurs , i.e. pahs may or may not be present . the remaining isecs in the mc sample do not appear to suffer from such degeneracy , as their fits unambiguously and consistently converge to the same solutions regardless of the starting point of the iterative fitting procedure . the welldefined pah collective properties we identify ( see @xcite for details ) are the total column density of carbon atoms locked in pahs per unit visual magnitude , @xmath63 ( or given the gas to dust ratio @xmath64 ) , and the fractions in each charge state summarized as an average charge per carbon atom , @xmath65 = \sum_i n_i q_i/ \sum_i n_i { \cal n}^i_{\rm c}$ ] , where @xmath66 , @xmath67 , and @xmath68 are column density , charge , and number of carbon atoms for the @xmath69th pah molecule , respectively . in table [ pahtable ] we also report the charge dispersion , @xmath70 $ ] , a measure of the homogeneity of the charging within the pah distribution . the total absolute column density of carbon locked in pahs is not an outcome of the fit , since we are dealing with normalized extinction , but it can derived through the absolute extinction ( see table [ one ] ) . the statistics of the mc sample are relatively small ( 24 lines of sight in total ) . nevertheless , some trends are already clear . the fitting results presented in table [ classictable ] occupy a slightly larger volume of the parameter space with respect to the galactic corresponding values . in particular , while for the mwg lines of sight the size distribution of classical grains is always bimodal with no exceptions ( see table 1 in @xcite ) , with a gap in size @xmath71 between @xmath72 and 200 nm , there are a few mc lines of sight for which the size distribution shows a negligible gap ( i.e. azv214 , sk-69 228 and one of the degenerate solutions for azv23 ) . in general , while there is considerable scatter in both , median values of @xmath73 are smaller in the mcs with respect to those we previously found for the mwg . the same can be seen if one compares the fits for the average mwg isec to the average isecs for the mcs . another important difference between the two samples resides in the predominant presence of @xmath1 carbon mantles in the mcs . the @xmath74 fraction is different from zero only in four cases , namely sk-68 26 , sk-69 256 , sk-69 210 , and azv23 ( see table [ classictable ] ) , while in the mwg carbonaceous mantles tend to be overwhelmingly aromatic . this is particularly significant as mcs have generally stronger radiation fields than the mwg ( up to ten times higher , @xcite ) . from figs . [ smcfitcurves ] and [ lmcfitcurves ] , it is evident that classical dust dominates the extinction profile , with some exceptions in the bump spectral region . in only one isec ( sk-69 256 ) pahs may be dominant in the whole range , depending on which of the formally acceptable solutions is the real one . the pah charge dispersion decreases with increasing pah charge ( fig . [ charge ] ) suggesting that when the mixture charge is not negligible all molecular species tend to assume the same charge value , as characteristic during e.g. , photoionization ( modulo the differences in the photo - absorption cross - sections ) . on the other side , large charge dispersions are expected when a superposition of different environments , with different physical conditions , are crossed by a line of sight . in this case the average charges of individual regions merge to a median , close to zero , value , but with a large spread around it . figure [ abundances ] shows in graphical form the total abundances of c and si locked in dust and pahs that we obtained in this work for the sightlines to the mcs examined here , also listed in table [ masstable ] , compared with the corresponding values we previously obtained for the mwg @xcite and with some reference values for the total elemental abundances . the level of compatibility of the amounts of si and c locked in dust in the mcs with abundance constraints is at least as good as that for the mwg , with a similar fraction of data points exceeding mildly the nominal ( average ) budget of available si , c constraints almost always respected , one spectacular failure for lmc , one for the smc and one for the mwg ( not visible , because it falls out of the plot ) . this can be interpreted as a validation of the [ cm]@xmath3 model for the mcs . lccccccccc + + los & @xmath75 & @xmath76 & @xmath77 & @xmath78 + + + azv18 & 0.075 ( 0.023 ) & 0 ( 0.0 ) & 0.0034 ( 0.0020 ) & 0.014 ( 0.004 ) + azv214 & 0.063 ( 0.013 ) & 0 ( 0.0 ) & 0.0013 ( 0.0004 ) & 0.0068 ( 0.0016 ) + azv23@xmath79 & 0.083 ( 0.005 ) & 0.049 ( 0.003 ) & 0.0071 ( 0.0005 ) & 0 ( 0.0 ) + azv23@xmath80 & 0.084 ( 0.018 ) & 0 ( 0.0 ) & 0.0007 ( 0.0012 ) & 0.0180 ( 0.0015 ) + azv398 & 0.12 ( 0.03 ) & 0.000003 ( 0.000016 ) & 0.0047 ( 0.0021 ) & 0.018 ( 0.004 ) + + azv456 & 0.066 ( 0.010 ) & 0 ( 0.0 ) & 0.0028 ( 0.0015 ) & 0.025 ( 0.004 ) + + sk-66 19 & 0.140 ( 0.019 ) & 0.00001 ( 0.00003 ) & 0.0033 ( 0.0013 ) & 0.0097 ( 0.0019 ) + sk-66 88 & 0.163 ( 0.020 ) & 0.000001 ( 0.000004 ) & 0.0070 ( 0.0011 ) & 0.0109 ( 0.0023 ) + sk-67 2 & 0.069 ( 0.028 ) & 0 ( 0.0 ) & 0.0017 ( 0.0007 ) & 0.019 ( 0.008 ) + sk-68 129 & 0.09 ( 0.04 ) & 0.00000 ( 0.0000011 ) & 0.0025 ( 0.0017 ) & 0.009 ( 0.004 ) + sk-68 23 & 0.158 ( 0.012 ) & 0.000002 ( 0.000005 ) & 0.0028 ( 0.0008 ) & 0.0146 ( 0.0014 ) + sk-68 26 & 0.095 ( 0.024 ) & 0.0019 ( 0.0016 ) & 0.0022 ( 0.0009 ) & 0.0088 ( 0.0023 ) + sk-69 108 & 0.144 ( 0.015 ) & 0.00002 ( 0.00010 ) & 0.0034 ( 0.0008 ) & 0.0169 ( 0.0028 ) + sk-69 206 & 0.151 ( 0.019 ) & 0.000002 ( 0.000008 ) & 0.0052 ( 0.0010 ) & 0.0073 ( 0.0013 ) + sk-69 210 & 0.196 ( 0.017 ) & 0.0038 ( 0.0008 ) & 0.0012 ( 0.0003 ) & 0.0187 ( 0.0019 ) + sk-69 213 & 0.086 ( 0.019 ) & 0.000005 ( 0.000022 ) & 0.0021 ( 0.0006 ) & 0.015 ( 0.003 ) + + sk-68 140 & 0.09 ( 0.03 ) & 0.00000 ( 0.0000015 ) & 0.0050 ( 0.0024 ) & 0.015 ( 0.005 ) + sk-68 155 & 0.10 ( 0.04 ) & 0.0000 ( 0.0000008 ) & 0.0019 ( 0.0007 ) & 0.0081 ( 0.0027 ) + sk-69 228 & 0.09 ( 0.04 ) & 0 ( 0.0 ) & 0.0061 ( 0.0027 ) & 0.0028 ( 0.0013 ) + sk-69 256@xmath81 & 0.018 ( 0.008 ) & 0.0024 ( 0.0028 ) & 0.0026 ( 0.0016 ) & 0.010 ( 0.004 ) + sk-69 256@xmath82 & 0.038 ( 0.014 ) & 0.0037 ( 0.0022 ) & 0.0019 ( 0.0009 ) & 0 ( 0.0 ) + sk-69 265 & 0.064 ( 0.015 ) & 0 ( 0.0 ) & 0.00107 ( 0.00024 ) & 0.0057 ( 0.0015 ) + sk-69 270 & 0.065 ( 0.018 ) & 0.00001 ( 0.00004 ) & 0.0015 ( 0.0005 ) & 0.018 ( 0.004 ) + sk-69 279 & 0.14 ( 0.05 ) & 0 ( 0.0 ) & 0.0045 ( 0.0017 ) & 0.0048 ( 0.0020 ) + sk-69 280 & 0.084 ( 0.016 ) & 0 ( 0.0 ) & 0.0009 ( 0.0006 ) & 0.0126 ( 0.0026 ) + sk-70 116 & 0.102 ( 0.019 ) & 0.000001 ( 0.000003 ) & 0.0021 ( 0.0005 ) & 0.0062 ( 0.0012 ) + @xmath51 fit solution for azv23 obtained with classical dust only ( no pahs ) + @xmath52 fit solution for azv23 including pahs + @xmath54 fit solution for sk69 256 with similar contributions by classical dust and pahs + @xmath55 fit solution for sk69 256 obtained with classical dust only ( no pahs ) table [ classicmasstab ] reports the column mass densities derived from our fits of the individual extinction curves , while figure [ classicmassfig ] shows the correlation of the total dust mass column densities versus total hydrogen mass column density . they can be compared with the results of @xcite , obtained fitting a dust emission model to herschel data . while of course our data points are very sparse , compared to all the pixels in the maps of the mcs , as used by @xcite , they do fall in the same area of the @xmath83@xmath84 plane covered by their data , and are therefore fully consistent with them . this is rather interesting , since the two datasets were obtained using completely different observations ( emission versus absorption ) and dust models . and normalized @xmath2 mantle carbon fraction , with relative errors for the smc ( red asterisks ) , the lmc ( blue asterisks ) , and the mwg ( pale grey symbols ) . the black symbol marked `` mw '' represents the values corresponding to the average galactic isec . values corresponding to average isecs in the mcs are also indicated . mwg data are taken from @xcite . ] in @xcite we outlined a selfconsistent evolutionary scenario for dust in the diffuse ism ( dism ) in the mwg . in such model , silicate grain particles , being relatively longlived , progressively accrete c atoms in an hrich environment ( the dism ) , thereby forming aliphatic carbonaceous mantles . the older layers of these mantles are progressively photoprocessed to aromatic , while new aliphatic material simultaneously continue to deposit on top of them . the deposition of carbon is assumed to be proportional to the rate of collisions of c atoms onto dust grains , with a sticking factor calibrated to match observational constraints on co abundances . since dust evolution is modelled in diffuse and translucent clouds , which are rich in atomic h , all freshly deposited carbonaceous material is assumed to be highly hydrogenated , i. e. aliphatic . upon uv irradiation , carbonaceous material is assumed to be dehydrogenated , becoming more and more aromatic . to simplify the model , carbonaceous material , instead of having a continuous , smooth transition between deeper , older , and thus more aromatic material , and freshly deposited , aliphatic material , is represented as just two layers , the deeper one completely aromatic , the other completely aliphatic . occasionally , shocks peel away most of this material , restarting the cycle and ejecting the fragments of carbonaceous mantles into the gas , where they can possibly be a source of pahs . shocks are not explicitely included in the set of differential equations giving the time evolution of the grain mantles , but are taken into account in terms of `` instantaneous '' episodes , removing most of the carbonaceous mantles , rehydrogenating the small leftover , and putting back into the gas phase the resulting carbon ; the time evolution , according to the differential equations of the model , is then resumed from these new initial conditions . upon trying to use this kind of evolutionary scenario to interpret the isecs observed in the mcs , as shown in fig.[mcevofig ] , it is apparent that the annealing time scale must be almost always much longer than the average time between mantleshattering events . the annealing scale is defined by the competition between energetic processing by uv radiation ( and cosmic rays ) and rehydrogenation by hot h atomic gas , turning aromatic carbonaceous material back into aliphatic . in the mwg , rehydrogenation was deemed to be by and large negligible , but this may be not the case in the mcs . carbonaceous mantles in the mcs result from the fit to be almost always completely aliphatic , with only few exceptions . this points to significantly higher starformation activities , and associated supernova rates , and/or much longer photoprocessing timescales , possibly due to more effective rehydrogenation . this is in agreement with the current star formation rates in the mcs ( @xmath85 in the smc , @xcite and @xmath86 @xmath87 yr@xmath9 in the lmc , @xcite ) that per unit area are an order of magnitude higher than in the mwg ( @xmath88 yr@xmath9 , @xcite ) . in general , observations of stellar populations in dwarf galaxies suggest that the star formation bursts are sporadic , separated by millions to billions of years even in isolated systems ( e.g. , @xcite ) . according to their starformation histories @xcite , the mcs appear to be typical members of this class of galaxies . mctype isecs may be then prototypical of dwarf galaxies of low metallicity , with high gas content ( see e.g. , @xcite ) . in the mwg dism the photo - darkening time is @xmath89 yr @xcite , @xmath90 being local fluctuations around the average intensity of the interstellar uv radiation field . such time should be compared with lifetime of carbon mantles against sputtering ( whose estimate is not a trivial task ) . assuming a plausible shock frequency of @xmath91 yr@xmath9 @xcite , we derive that dust mantles in the mwg dism have enough time to be converted in aromatic structure . @xcite show that the evolutionary age of isecs peaks at times of few myr ( assuming typical diffuse interstellar conditions ) comparable to the probable average age of diffuse clouds . translating these considerations to the dism in the mcs , we may conclude that the carbon recycling time must be much shorter than the local photodarkening time , @xmath92 yr ( @xmath93 , @xcite ) . however , such result should be taken with care , as many effects are concurring in dust destruction and formation processes . for instance , small grains are more susceptible to destruction by sputtering in shock waves than large grains @xcite , because larger grains decouple from the gas and are therefore less exposed to ions . in strong shocks , graingrain interaction may lead to shattering and thus to the creation of smaller grains @xcite , which are then more likely to be sputtered away . hence , the timescale of dust destruction may be inversely proportional to the abundance of dust , since the rate of collisions is proportional to the grain number density ( e.g. , @xcite ) . although pahs are clearly less abundant in the mcs than in the mwg ( see table [ pahtable ] ) , a potential problem in our description of the carbon cycle in galaxies is posed by the the sharp decline of the source of aromatic material in the dism , since most of dust mantles are destroyed before they experience a substantial annealing . pah formation rates in the cold , carbonrich winds of evolved stars are too slow , the injection time being approximately 2 gy . surveys in smc also revised down the contribution of agb stars to dust production ( e.g. , @xcite ) . without a net positive contribution from supernovae to the dust budget , this suggests that dust must grow in the dism or be formed by another yet unknown mechanism ( see @xcite and @xcite ) . recent _ herschel _ photometric and spectroscopic observations of the supernova 1987a by @xcite may suggest that supernovae can be an important source of dust in the interstellar medium , from local to high - redshift galaxies . however , the point is controversial as it depends on the efficiency of supernova remnants in destroyng interstellar grains . @xcite determined dust lifetimes in the mcs exploiting the available data of supernova remnants for these galaxies . these authors inferred dust lifetimes in the mcs significantly shorter then in the mwg , as the cumulative dust production rates by agb stars and supernovae are one order of magnitude lower than dust destruction in interstellar shocks produced by the expansion of supernova remnants conversely , we are left with the puzzle of what becomes of the aliphatic fragments released into the dism of the mcs by the shattering events . @xcite proposed that they should be rapidly evaporated into polyyines , and thereafter quickly photodissociated under dism conditions . they would therefore have a negligible effect on the extinction . if instead a significant population of aliphatic carbonaceous nanoparticles were able to survive for a significant time , thereby achieving nonnegligible abundances , they should be visible both in extinction , exhibiting @xmath94 features contributing mainly to the nonlinear faruv rise , and given their small sizes possibly producing aliphatic ch emission . indeed , some emission in the aliphatic ch stretch at 3.4 @xmath95 m is detected toward the mcs @xcite . classical dust extinction is generally larger than pah extinction with a few exceptions . in these latter cases , the pah contribution dominates only in the bump region . in all these relatively `` pahrich '' cases , the classical size distribution is strongly bimodal , with a gap ranging from 40 to 130 nm , with a tendency to increase the non linearity of the faruv rise with increasing size breaks . all the remaining isecs are reproduced with classic grain size distributions that tend to have markedly lower gaps ( see table [ classictable ] ) . in @xcite we concluded that , in the framework of [ cm]@xmath3 model , bumpless extinction curves are not necessarily lines of sight devoid of pahs , but rather that the @xmath96 pah absorption can be very effectively masked by a large gap between the maximum size limit of small dust grains and the minimum size limit of large ones . mwg data were not conclusive since mc - type isecs are rare in our galaxy . the present results suggest a different interpretation based on the fact that some isecs show a more or less weak non - linear far - uv rise without showing an appreciable bump . this is particularly evident in the smc sub sample . in the lmc the most evident trends are small bumps with linear far - uv rises . within the current model dust populations , for the smc - type isec the only possibility is the introduction of pahs that , through @xmath97 absorptions , may produce such rise . however , as a by product , pah molecules exhibit @xmath96 transitions who generate the bump . thus , the only viable solution , for bumpless isecs , is the removal of mid - size grains , whose extinction power falls in the bump region , from the grain size distribution . this may be interpreted in two alternative ways . the first one is that we must be missing an @xmath98dominated carbon component , whose sizes are intermediate between small solids and macromolecules . this is also hinted by the numerical tendency during the fitting procedure to go below the lowest limit of 5 nm in particle size , that is systematic for all the lines of sight in the mwg and in the mcs . such small particles might be carbon mantle fragments , possibly produced during some destructive events ( as discussed in the previous paragraph ) , such as e. g. nanodiamonds @xcite . the second interpretation is instead that in our model we are missing no important component contributing to extinction , and hence the fact that pahs are abundant only if middlesized dust grains are absent is real , and must have a sound physical explanation . one might argue that if an interstellar cloud , and the dust therein , underwent some shocks in its history , they may have shattered middlesized grains , producing fragments ( small grains ) and releasing pahs from the disrupted carbonaceous mantles . this would be a consistent explanation for the mwg , where carbonaceous mantles are almost entirely aromatic , so that their disruption would indeed produce pahlike fragments . but in the mcs carbonaceous mantles result to be almost entirely aliphatic , so their destruction can hardly be a viable source of pahs . in the case of lmctype isecs the results of the fitting procedure are more straightforward than for the smctype , bumpless ones : in this galaxy pahs are present but are a minor component , their presence being only evident in the bump spectral region . the possible extra aliphatic `` mesoscopic '' component is marginal in this case . conversely , the problem of the lack of a suitable source of aromatic material is therefore more pressing for lmc than smc . we here concisely summarise the main conclusions of this work . * the [ cm]@xmath3 dust model can very precisely match all the isecs observed in the mcs , yielding estimates for the total dust mass column density versus total hydrogen column density that are fully consistent with those obtained by @xcite in a completely different , independent way . * abundance of elements locked up in [ cm]@xmath3 models fitting the isecs in the mcs are at least as compatible with abundance constraints as those for the mwg @xcite . * when isecs have vanishingly weak bump and faruv nonlinear rise , they can be equally well fitted either by classical dust alone or by classical dust _ and _ pahs . in such cases models without pahs use much more si and c atoms , which may put them at odds with abundance constraints . * carbonaceous mantles in the mcs resulting from the [ cm]@xmath3 model fits are overwhelmingly aliphatic , whereas those in the mwg are predominantly aromatic . together with the observational constraint that radiation fields in mcs are typically 10 times stronger than the average in the mwg @xcite , in the [ cm]@xmath3 model this implies that mantleshattering events must be @xmath99 times more frequent in the mcs than in the mwg , and/or much more efficient . desorption from such mantles can not be a significant source of pahs in the mcs . * bumpless isecs with large nonlinear faruv rises can be fitted within the [ cm]@xmath3 model , but this requires a finely tuned match of the gap in the size distribution of classical dust grains , resulting in a dip in the extinction they produce that precisely cancels with the bump . either there is some hitherto not understood physical relation connecting the population of pahs with the size distribution of silicate grains in the diffuse interstellar medium , or this hints that the [ cm]@xmath3 model must be missing some component different from pahs that can can produce a nonlinear faruv rise without simultaneously contributing to the bump . we thank the referee for her / his constructive criticism , that helped us to improve this work . we acknowledge the support of the autonomous region of sardinia , project crp 26666 ( regional law 7/2007 , call 2010 ) 99 asano , r. , takeuchi , t.t . , hirashita , h. , nozawa , t. , 2013 , , 432 , 637 asplund m. , grevesse n. , sauval a. j. & scott p. , 2009 , ara&a , 47 , 481 bocchio , m. , jones , a.p . & slavin j.d . 2014 , , 570 , a32 bolatto , a.d . et al . , 2011 , , 741 , 12 boulanger , f. , onaka , t. , pilleri , p. , joblin , c. 2011 , eas publications series 46 , 399 boyer , m.l . et al . , 2012 , , 748 , 40 cardelli , j.a . , clayton , g.c . , & mathis , j.s . 1989 , , 345 , 245 cartledge , s.i.b . , clayton , g.c . , gordon , k.d . , 2005 , , 630 , 355 cecchi - pestellini , c. , malloci , g. , mulas , g. , joblin , c. , & williams , d.a . 2008 , , 486 , l25 cecchi - pestellini , c. , cacciola , a. , iat , m.a . , 2010 , mnras , 408 , 535 cecchi - pestellini , c. , casu , s. , mulas , g. , & zonca , a. 2014 , , 785 , 41 cecchi - pestellini , c. , viti , s. & williams , d.a . , 2014 , , 788 , 100 draine , b.t . 1987 , in _ physical processes in interstellar clouds _ , eds morfill & m. scholer , p. 423 draine , b.t . 2003 , , 41 , 241 duley , w.w . & williams , d.a . , 1984 , , 211 , 97 dwek , e. , staghun , j. , arendt , r.g . , kovacs , a. , su , t. , benford , d.j . , 2014 , , 788 , l30 fitzpatrick , e.l . , & massa , d. 2007 , , 663 , 320 gordon , k.d . , clayton , g.c . , misselt , k.a . , landolt , a.u . , & wolff , m.j . 2003 , , 594 , 279 harris , j. & zaritsky , d. , 2004 , , 127 , 1531 harris , j. & zaritsky , d. , 2009 , , 138 , 1243 howarth , i.d . , 2011 , , 413 , 1515 iida s. , ohtaki t. , seki t. , 1984 , am . 120 , 258 israel , f.p . 1993 , , 276 , 25 jones , a.p . , duley , w.w . , williams d.a . 1990 , q. j. r. astron . , 31 , 567 levenberg , k. 1944 , quarterly of applied mathematics 2 , 164 malloci , g. , joblin , c. , & mulas , g. 2007 , cp , 332 , 353 markwardt c.b . 2009 , astronomical data analysis software and systems xviii , 411 , 251 mathis j.s . , 2000 , j. geophys . , 105 , 10269 matsuura , m. et al . , 2011 , science , 333 , 1258 matsuura , m. et al . , 2015 , , 800 , 50 mattsson , l. , de cia , a. , andersen , a.c . , zafar , t. , 2014 , , 440 , 1562 mckee , c.f . , 2011 , science , 333 , 1227 mennella v. , muoz caro g.m . , ruiterkamp r. , schutte w.a . , greenberg j.m . , brucato j.r . , colangeli l. , 2001 , , 367 , 355 mori , t.i . , sakon , i. , onaka , t. , kaneda , h. , umehata , h. , ohsawa , r. 2012 , 744 , 68 mulas , g. , zonca , a. , casu , s. , & cecchi - pestellini , c. 2013 , apjs , 207 , 7 rai , r. , rastogi , s. , 2010 , , 410 , 2722 rai , r. , rastogi , s. , 2012 , , 423 , 2941 rai , r.k . , botet , r.s . , 2014 , , 444 , 303 rezaei , s. , javadi , a. , khosroshahi , h. & van loon , j .- th , 2014 , arxiv : 1409.0790v1 robitaille , t.p . & whitney , b.a . , 2010 , , 710 , l11 romanduval , j. et al . 2014 , , 797 , 86 slavin , j.d . , jones , a.p . , tielens a.g.g.m . , 2004 , , 614 , 796 temim , t. , dwek , e. , tchernyshyov , k. , boyer , m.l . , meixner , m. , gall , c. , & romanduval , j. 2015 , , 799 , 158 valencic , l.a . , clayton , g . c. , & gordon , k.d . 2004 , , 616 , 912 weisz , d. r. , skillman , e. d. , cannon , j. m. , et al . 2008 , apj , 689 , 160 welty , d.e@xmath3 . , xue , r. & wong , t. 2012 , , 745 , 173 zhukovska , s. & henning , t. , 2013 , in _ the life cycle of dust in the universe : observations , theory , and laboratory experiments _ , eds . a. andersen , m. baes , h. gomez , c. kemper & d. watson , p. 16 zonca , a. , cecchi - pestellini , c. , mulas , g. , & malloci , g. 2011 , , 410 , 1932 zubko v. , dwek e. & arendt g. , 2004 , , 152 , 211

high - rate space - time block codes ( stbc with code rate ) in multi - input multi - output ( mimo ) systems are able to provide both spatial multiplexing gain and diversity gain , but have high maximum likelihood ( ml ) decoding complexity . since group - decodable ( quasi - orthogonal ) code structure can reduce the decoding complexity , we present in this paper systematic methods to construct group - decodable high - rate stbc with full symbol - wise diversity gain for arbitrary transmit antenna number and code length . we show that the proposed group - decodable stbc can achieve high code rate that increases almost linearly with the transmit antenna number , and the slope of this near - linear dependence increases with the code length . comparisons with existing low - rate and high - rate codes ( such as orthogonal stbc and algebraic stbc ) are conducted to show the decoding complexity reduction and good code performance achieved by the proposed codes . space - time block codes ( stbc ) , group - decodable code structure , code construction .

1011.3272

space - time codes ( stc ) in multi - input multi - output ( mimo ) systems have been extensively studied for their ability to provide transmit diversity gain and spatial multiplexing gain @xcite . space - time trellis codes ( sttc ) @xcite and space - time block codes ( stbc ) @xcite are able to provide diversity gain and have code rate limited by 1 . on the other hand , bell labs layered space - time ( blast ) system @xcite , high - rate linear dispersion ( ld ) codes @xcite , golden code @xcite , perfect codes @xcite , ps - sr code@xcite , etc . , have code rate @xmath0 and are able to provide multiplexing gain ( the latter four have diversity gain too ) . to achieve higher code rates with low joint - decoding complexity , many stbc with code rate @xmath1 have been designed to be group - decodable ( quasi - orthogonal ) @xcite . in contrast , there were much fewer designs of group - decodable stbc with code rate @xmath0 ( high - rate stbc ) . in @xcite , square 2-group - decodable stbc of code rate 1.25 for 4 transmit antennas were obtained by computer search ; in @xcite , 2-group - decodable stbc of code rate @xmath2 for @xmath3 ( @xmath4 ) transmit antennas were constructed . in @xcite , it was also shown that the group - decodable code structure is beneficial to diversity - embedded ( de ) space - time codes as it avoids interference between the different diversity layers in the de codes and helps to guarantee the designed diversity levels . in this paper , group - decodable high - rate stbc with arbitrary number of transmit antennas and code lengths is considered , then systematic methods to construct them with full symbol - wise diversity are presented . their maximum achievable code rate and decoding complexity are analyzed . specific code examples are constructed and simulated . the rest of this paper is organized as follows . high - rate stbc with code rate @xmath0 will be abbreviated as stbc . in section ii , the system model is described and group - decodable stbc is defined . unbalanced 2-group - decodable stbc and balanced 2-group - decodable stbc are constructed systematically in section iii and section iv , respectively . comparisons of the decoding complexity and ber performance are shown in section v. finally , this paper is concluded in section vi . in this paper , bold lower case and upper case letters denote vectors and matrices ( sets ) , respectively ; @xmath5 and @xmath6 denote the real and the complex number field , respectively ; @xmath7 and @xmath8 stand for the real and the imaginary part of a complex vector or matrix , respectively ; @xmath9^*$ ] , @xmath9^t$ ] , @xmath9^h$ ] and @xmath10 denote the complex conjugate , the transpose , the complex conjugate transpose and the frobenius norm of a matrix , respectively ; @xmath11 and @xmath12 represent dimension of a vector / matrix space and rank of a matrix , respectively ; @xmath13 denotes an identity matrix . we consider a space - time block coding system employing @xmath14 transmit antennas and @xmath15 receive antennas . the transmitted signal sequences are partitioned into independent time blocks for transmission over @xmath16 symbol durations using stbc matrix @xmath17 of size @xmath18 . following the signal model in @xcite , @xmath17 can be denoted as : @xmath19 where @xmath20 are real valued symbols representing the real and imaginary components of complex constellation symbols , @xmath21 are called dispersion matrices . thus , the code rate is @xmath22 considering complex symbol transmission . the average energy of the code matrix is constrained to @xmath23 . the received signals @xmath24 of the @xmath25th receive antenna at time @xmath26 can be arranged in a @xmath27 matrix @xmath28= [ \tilde{r}_{tm}]$ ] . thus , the transmit - receive signal relationship can be presented as : @xmath29 = \sqrt{\rho}\textbf{x}\tilde{\textbf{h}}+\tilde{\textbf{z}}\ ] ] where @xmath30 $ ] is the channel matrix with independent entries @xmath31 ; @xmath32 @xmath33=[\tilde{z}_{tm}]$ ] is the additive noise matrix with independently , identically distributed ( i.i.d . ) @xmath34 entries @xmath35 ; @xmath36 is the average signal - to - noise ratio ( snr ) at each receive antenna . the received signal can also be rewritten as @xcite : @xmath37 where @xmath38,~ \bar{\textbf{h}}=\left [ \begin{array}{cccccccc } \tilde{\textbf{h}}^r_{1}\\ \tilde{\textbf{h}}^i_{1}\\ \vdots \\ \tilde{\textbf{h}}^r_{m}\\ \tilde{\textbf{h}}^i_{m } \end{array } \right],~\textbf{s}=\left [ \begin{array}{cccccccc } s_1\\ s_2\\ \vdots\\ s_l \end{array } \right],~\textbf{z}=\left [ \begin{array}{cccccccc } \tilde{\textbf{z } } ^r_1\\ \tilde{\textbf{z } } ^i_1\\ \vdots\\ \tilde{\textbf{z } } ^r_{m}\\ \tilde{\textbf{z } } ^i_{m } \end{array } \right ] , \end{split}\ ] ] @xmath39=\left [ \begin{array}{cccccccc } \mathscr{c}_1\bar{\textbf{h } } & \mathscr{c}_2\bar{\textbf{h } } & \cdots & \mathscr{c}_l\bar{\textbf{h } } \end{array } \right ] , \\ & \mathscr{c}_l=\left [ \begin{array}{cccccccc } \mathcal { c}_l & \textbf{0 } & \cdots & \textbf{0}\\ \textbf{0 } & \mathcal { c}_l & \cdots & \textbf{0}\\ \vdots & \vdots & \ddots & \vdots \\ \textbf{0 } & \textbf{0 } & \cdots & \mathcal { c}_l \end{array } \right]_{2tm\times 2nm},~\mathcal { c}_l=\left [ \begin{array}{cccccccc } \textbf{c}^r_l & -\textbf{c}^i_l\\ \textbf{c}^i_l & \textbf{c}^r_l \end{array } \right]_{2t\times 2n } \end{split}\ ] ] and @xmath40 . the maximum likelihood ( ml ) decoding of stbc is to find the solution @xmath41 so that @xmath42 to avoid rank deficiency at the decoder , @xmath43 is required , which means that @xmath44 should be tall , i.e. , @xmath45@xcite@xcite , which implies that the receiver antenna number satisfies @xmath46 . firstly , _ linear independence of matrices _ is defined as follows : [ rank_matrix ] _ the matrices @xmath48 are said to be linearly independent if no nontrivial linear combination of them is equal to zero . in other words , with @xmath49 @xmath50 only when @xmath51 . _ it is easy to show that the linear independence among @xmath48 is equivalent to the linear independence among vectors @xmath52 , where @xmath53 with @xmath54 , and @xmath55 is the matrix - to - vector mapping function @xmath56}{0pt}{}{a~c}{b~d}}\xrightarrow{g}[a~b~c~d]^t$ ] . the main idea of group - decodable stbc is to divide the @xmath57 real transmitted symbols embedded in a code matrix into several orthogonal groups such that after linear channel matched filtering , the ml detection metric of the transmitted symbols can be decoupled into independent submetrics , each containing a smaller group of symbols . assume that the transmitted symbols can be separated into @xmath47 groups and each group has @xmath58 symbols , then @xmath59 . let the set of indexes of symbols in the @xmath60th group be denoted as @xmath61 . for an stbc to be @xmath47-group - decodable , two conditions should be satisfied : \(i ) @xmath62 where @xmath63 ; \(ii ) @xmath64 where @xmath65 $ ] , @xmath66 . condition ( i ) means that the stbc is group - decodable and condition ( ii ) guarantees that no decoder of any group is rank deficient . to satisfy the condition ( i ) , _ yuen _ et al . @xcite have established a necessary and sufficient condition as follows : [ qoc_th ] _ the necessary and sufficient condition to make @xmath67 and @xmath68 in the stbc matrix ( [ lstbc ] ) to be orthogonal ( i.e. , to achieve @xmath62 ) is @xmath69 _ regarding the condition ( ii ) , @xmath64 implies that @xmath70 should be linearly independent . [ th_independence ] _ the necessary and sufficient condition for @xmath70 to be linearly independent is that @xmath56}{0pt}{}{\textbf{c}^r_{i_1}}{\textbf{c}^i_{i_1}}}$ ] , @xmath71}{0pt}{}{\textbf{c}^r_{i_2}}{\textbf{c}^i_{i_2}}},\cdots,$ ] @xmath71}{0pt}{}{\textbf{c}^r_{i_{l_i}}}{\textbf{c}^i_{i_{l_i}}}}$ ] must be linearly independent . _ the proof of theorem [ th_independence ] is given in appendix [ proofindependence ] . from the above , a formal definition of @xmath47-group - decodable stbc can be presented as : [ group_decodable_stbc ] _ an stbc is said to be @xmath47-group - decodable if _ _ ( i ) @xmath72 ; _ _ ( ii ) @xmath56}{0pt}{}{\textbf{c}^r_{i_1}}{\textbf{c}^i_{i_1}}},\cdots , { \genfrac{[}{]}{0pt}{}{\textbf{c}^r_{i_k}}{\textbf{c}^i_{i_k}}},\cdots , { \genfrac{[}{]}{0pt}{}{\textbf{c}^r_{i_{l_i}}}{\textbf{c}^i_{i_{l_i}}}}$ ] are linearly independent where @xmath73 . _ in this paper , we focus on 2-group - decodable stbc , i.e. , @xmath74 . for 2-group - decodable stbc , the total transmitted symbols @xmath75 where @xmath76 and @xmath77 are the number of symbols in the first group and second group , respectively . we will consider two cases . the first case is @xmath78 and @xmath79 ( a special case of @xmath80 ) , called _ unbalanced _ 2-group - decodable stbc ; the other case is @xmath81 , called _ balanced _ 2-group - decodable stbc . the former will be used to construct the latter . considering the unbalanced 2-group - decodable stbc with @xmath78 and @xmath79 , we have @xmath82 for brevity , @xmath83 is simplified as @xmath84 $ ] , and @xmath85 with @xmath86 is represented by @xmath87 $ ] , @xmath88 . _ symbol - wise diversity is denoted as the minimum rank of the dispersion matrices in an stbc@xcite@xcite . _ to achieve full symbol - wise diversity gain , @xmath89 is required to be full rank , i.e. , @xmath90 . then , ( [ draft_eq_qoc ] ) can be written as @xmath91 it is easy to show that ( [ eq_draft ] ) can be converted into scalar equations as : @xmath92 where @xmath93 , @xmath94 . in turn , ( [ eq_dis ] ) can be rewritten in matrix form as : @xmath95 where @xmath96 of size @xmath97 and @xmath98}{0pt}{}{\textbf{y}^r}{\textbf{y}^i}})$ ] of size @xmath99 with mapping functions @xmath100 and @xmath55 given in ( [ candy ] ) , @xmath101 $ ] and @xmath102 @xmath103 $ ] . @xmath104,~~\bar{\textbf{y}}=g({\genfrac{[}{]}{0pt}{}{\textbf{y}^r}{\textbf{y}^i}})=\left [ \begin{array}{cccccc } y^r_{11 } \\y^i_{11}\\\vdots\\ y^r_{t1}\\y^i_{t1}\\y^r_{12 } \\y^i_{12}\\\vdots \\y^r_{t2}\\y^i_{t2}\\\vdots \\y^r_{1n } \\y^i_{1n}\\\vdots \\y^r_{tn}\\y^i_{tn } \end{array } \right]\ ] ] as @xmath105 is of size @xmath99 , the solution space of ( [ eq_matrix ] ) , @xmath106 , is of dimension @xmath107 . let @xmath108 @xmath109 be the basis of @xmath106 , which are linearly independent . denoting @xmath110 as the inverse function of @xmath55 in ( [ candy ] ) , linearly independent matrices @xmath56}{0pt}{}{\textbf{y}^r_1}{\textbf{y}^i_1}},$ ] @xmath56}{0pt}{}{\textbf{y}^r_2}{\textbf{y}^i_2 } } , \cdots , { \genfrac{[}{]}{0pt}{}{\textbf{y}^r_{2tn - rank(\mathcal { c})}}{\textbf{y}^i_{2tn - rank(\mathcal { c})}}}$ ] can be obtained as @xmath56}{0pt}{}{\textbf{y}^r_i}{\textbf{y}^i_i}}=g^{-1}(\bar{\textbf{y}}_i)$ ] with @xmath111 . from definition [ group_decodable_stbc ] , if @xmath112 and @xmath89 in ( [ eq_draft ] ) are used as the dispersion matrices , the resultant stbc will be an unbalanced 2-group - decodable stbc of code rate @xmath113 with 1 real symbol in the first group and @xmath107 real symbols in the second group . from the discussions above , we can summarize the systematic construction of unbalanced 2-group - decodable stbc as follows : step 1 : pick a @xmath18 matrix @xmath89 with full rank as the dispersion matrix @xmath83 in the first group ; step 2 : based on the matrix @xmath89 , obtain the matrix @xmath96 following equation ( [ candy ] ) ; step 3 : based on the matrix @xmath114 , solve equation ( [ eq_matrix ] ) and obtain its solution space represented as @xmath115 @xmath116 subject to the condition that all @xmath117 ( @xmath118 ) lead to full - rank dispersion matrices in step 4 ; step 4 : using the vector - to - matrix mapping function @xmath110 ( inverse function of @xmath55 in ( [ candy ] ) ) , obtain matrices @xmath56}{0pt}{}{\textbf{y}^r_i}{\textbf{y}^i_i}}=g^{-1}(\bar{\textbf{y}}_i)$ ] with @xmath118 . using @xmath119 as the dispersion matrices in the second group , obtain the resultant 2-group - decodable stbc as @xmath120 where @xmath121 is in the first group , while @xmath122 to @xmath123 are in the second group ; step 5 : use the constellation rotation technique @xcite to optimize the proposed code . since the code symbols are divided into mutually orthogonal groups , this constellation rotation can be done group by group . since the code rate of unbalanced 2-group - decodable stbc is @xmath113 , its upper bound depends on the lower bound of @xmath124 . regarding @xmath125 , we have the following theorem : [ rank_c_prop ] _ + _ _ i ) when @xmath126 , i.e. , @xmath127 , then @xmath114 in ( [ eq_matrix ] ) is of full rank and @xmath128 ; _ _ ii ) when @xmath129 , i.e. , @xmath130 , then the lower bound of @xmath124 is @xmath131 and it is reached when @xmath89 ( after suitable permutations ) takes the form of @xmath132 $ ] . _ the proof of theorem [ rank_c_prop ] is given in appendix [ proof_rank_c ] . from theorem [ rank_c_prop ] , it can be deduced that there are @xmath133 ( when @xmath126 ) or @xmath134 ( when @xmath135 ) dispersion matrices in the second group . then the following proposition on the maximum code rate of unbalanced 2-group - decodable stbc can be obtained : _ for an unbalanced 2-group - decodable stbc for @xmath14 transmit antennas over @xmath16 symbol durations , its maximum achievable code rate is @xmath136 for @xmath126 , or @xmath137 for @xmath135 . for the former , when @xmath138 , the code rate @xmath139 approaches @xmath14 asymptotically , i.e. , the code approaches full rate . _ the code rate variation of the proposed unbalanced 2-group - decodable stbc as a function of @xmath14 and @xmath16 is shown in fig . [ rate_un2groupstbc ] . transmit antennas over @xmath16 symbol durations . for illustration purpose , @xmath140 are shown.,width=326 ] clearly , the ml decoding complexity order of the proposed 2-group - decodable stbc is mainly decided by that of the larger group . following @xcite , the ml decoding complexity order @xmath141 can be shown as : @xmath142 where @xmath143 denotes the number of real symbols ( need not be orthogonal ) in the largest group , @xmath144 denotes the number of orthogonal real symbols in that group ( @xmath145 if the largest group is fully non - orthogonal ) , @xmath146 denotes the size of the signal constellation applied with transmission bit rate @xmath147 and stbc code rate @xmath148 . for the proposed unbalanced 2-group - decodable stbc , @xmath149 ( when @xmath126 ) or @xmath134 ( when @xmath135 ) , while @xmath150 ( when @xmath126 ) or @xmath137 ( when @xmath135 ) . in this subsection , we present a step - by - step example of the construction of an unbalanced 2-group - decodable stbc for 4 transmit antennas over 2 symbol durations . this code can be used in 2 ways : ( i ) to form an unbalanced 2-group - decodable code for 2 transmit antennas ; ( ii ) to construct a balanced 2-group - decodable code for 4 transmit antennas in section [ example_b_nt4 ] . step 1 : pick a 2@xmath1514 matrix @xmath83 with full rank ( rank 2 ) as the dispersion matrix in the first group : @xmath152\ ] ] note that @xmath83 satisfies theorem [ rank_c_prop](ii ) , hence it achieves the code rate bound . step 2 : based on @xmath83 , obtain matrix @xmath114 with @xmath153 from ( [ candy ] ) : @xmath154}\ ] ] step 3 : solve equation ( [ eq_matrix ] ) with matrix @xmath114 , and obtain the solution space of dimension @xmath155 as : @xmath156}+ k_2{\small\left [ \begin{array}{ccccccccccccccccccc } 0 \\ 1 \\ 0 \\ 1 \\ 0 \\ -1 \\ 0 \\ 1 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ \end{array } \right]}+ k_3{\small\left [ \begin{array}{ccccccccccccccccccc } 0 \\ 1 \\ 0 \\ -1 \\ 0 \\ 1 \\ 0 \\ 1 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ \end{array } \right]}+ k_4{\small\left [ \begin{array}{ccccccccccccccccccc } 0 \\ 1 \\ 0 \\ 1 \\ 0 \\ 1 \\ 0 \\ -1 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ \end{array } \right]}\ ] ] step 4 : under the vector - to - matrix mapping function @xmath110 , obtain @xmath56}{0pt}{}{\textbf{y}^r_i}{\textbf{y}^i_i}}=g^{-1}(\bar{\textbf{y}}_i)$ ] with @xmath157 as:@xmath158}{0pt}{}{\textbf{y}^r_1}{\textbf{y}^i_1}}=\left [ \begin{array}{cccccccccc } -1 & 1 & 0 & 0\\ 1 & 1 & 0 & 0\\ 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0\\ \end{array } \right],~ { \genfrac{[}{]}{0pt}{}{\textbf{y}^r_2}{\textbf{y}^i_2}}=\left [ \begin{array}{cccccccccc } 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0\\ 1 & -1 & 0 & 0\\ 1 & 1 & 0 & 0 \\ \end{array } \right],~ { \genfrac{[}{]}{0pt}{}{\textbf{y}^r_3}{\textbf{y}^i_3}}=\left [ \begin{array}{cccccccccc } 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0 \\ 1 & 1 & 0 & 0\\ -1 & 1 & 0 & 0\\ \end{array } \right],~ { \genfrac{[}{]}{0pt}{}{\textbf{y}^r_4}{\textbf{y}^i_4}}=\left [ \begin{array}{cccccccccc } 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0\\ 1 & 1 & 0 & 0\\ 1 & -1 & 0 & 0\\ \end{array } \right].\ ] ] so , we have @xmath119 as : @xmath159},~ \textbf{y}_2={\small\left [ \begin{array}{cccccccccc } j & -j & 0 & 0\\ j & j & 0 & 0 \\ \end{array } \right]},~ \textbf{y}_3={\small\left [ \begin{array}{cccccccccc } j & j & 0 & 0\\ -j & j & 0 & 0 \\ \end{array } \right]},~ \textbf{y}_4={\small\left [ \begin{array}{cccccccccc } j & j & 0 & 0\\ j & -j & 0 & 0\\ \end{array } \right]}.\ ] ] we emphasize that since ( [ eq_matrix ] ) is under - determined , there will be many possible solutions of @xmath160 . typically , we choose the set of @xmath160 leading to @xmath161 full - rank dispersion matrices , in order to achieve full symbol - wise diversity gain @xcite@xcite ; as many orthogonal dispersion matrices as possible , in order to achieve a large k in ( [ dc_order ] ) . since the dispersion matrices @xmath83 in ( [ un_2x4_c ] ) and @xmath162 to @xmath163 in ( [ un_2x4_y ] ) transmit no information on the third and fourth antennas , so they can be reduced to the followings , @xmath162 to @xmath163 will be used in section [ example_b_nt4 ] to construct a balanced 2-group - decodable code example . ] without loss in code rate or diversity : @xmath164;~ \textbf{y}_1=\left [ \begin{array}{cccccccccc } -1 & 1 \\ 1 & 1 \\ \end{array } \right],~ \textbf{y}_2=\left [ \begin{array}{cccccccccc } j & -j \\ j & j \\ \end{array } \right],~ \textbf{y}_3=\left [ \begin{array}{cccccccccc } j & j \\ -j & j \\ \end{array } \right],~ \textbf{y}_4=\left [ \begin{array}{cccccccccc } j & j \\ j & -j \\ \end{array } \right].\ ] ] hence an unbalanced 2-group - decodable stbc with the dispersion matrices in ( [ un_2x2_cy ] ) for 2 transmit antennas can be obtained as : @xmath165\ ] ] where @xmath121 is in the first group , while @xmath122 to @xmath166 are in the second group . furthermore , @xmath122 to @xmath167 are orthogonal , which leads to @xmath168 and @xmath169 in the decoding complexity order formula ( [ dc_order ] ) for this code . @xmath170 has code rate @xmath171 , hence , its decoding complexity order calculated following ( [ dc_order ] ) is : @xmath172 in this subsection , we present the code example of a 2-group - decodable stbc for 4 transmit antennas over 4 symbol durations . step 1 : pick a 4@xmath1734 matrix @xmath83 with rank 4 as the dispersion matrix in the first group : @xmath174.\ ] ] step 2 - 4 : since @xmath175 , @xmath176 dispersion matrices can be obtained in the second group as : @xmath177,~~\textbf{y}_2=\left [ \begin{array}{cccccccccc } 0 & 0 & j & 0 \\ 0 & 0 & 0 & -j \\ j & 0 & 0 & 0 \\ 0 & -j & 0 & 0 \\ \end{array } \right],~\textbf{y}_3=\left [ \begin{array}{cccccccccc } 0 & 0 & 0 & 1 \\ 0 & 0 & -1 & 0 \\ 0 & 1 & 0 & 0 \\ -1 & 0 & 0 & 0 \\ \end{array } \right],~\textbf{y}_4=\left [ \begin{array}{cccccccccc } 0 & 0 & 0 & j \\ 0 & 0 & j & 0 \\ 0 & j & 0 & 0 \\ j & 0 & 0 & 0 \\ \end{array } \right],~\\ \end{split}\ ] ] @xmath178,~~\textbf{y}_6=\left [ \begin{array}{cccccccccc } 0 & 1 & 0 & 0 \\ -1 & 0 & 0 & 0 \\ 0 & 0 & 0 & 1 \\ 0 & 0 & -1 & 0 \\ \end{array } \right],~\textbf{y}_7=\left [ \begin{array}{cccccccccc } 0 & j & 0 & 0 \\ j & 0 & 0 & 0 \\ 0 & 0 & 0 & -j \\ 0 & 0 & -j & 0 \\ \end{array } \right],~ \textbf{y}_8=\left [ \begin{array}{cccccccccc } j & 0 & 0 & 0 \\ 0 & j & 0 & 0 \\ 0 & 0 & j & 0 \\ 0 & 0 & 0 & j \\ \end{array } \right],\\ \textbf{y}_{9}&=\left [ \begin{array}{cccccccccc } 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & -1 \\ -1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ \end{array } \right ] , ~\textbf{y}_{10}=\left [ \begin{array}{cccccccccc } 0 & 0 & j & 0 \\ 0 & 0 & 0 & j \\ j & 0 & 0 & 0 \\ 0 & j & 0 & 0 \\ \end{array } \right],\textbf{y}_{11}=\left [ \begin{array}{cccccccccc } j & 0 & 0 & 0 \\ 0 & -j & 0 & 0 \\ 0 & 0 & -j & 0 \\ 0 & 0 & 0 & j \\ \end{array } \right],\textbf{y}_{12}=\left [ \begin{array}{cccccccccc } 0 & 0 & 0 & 1 \\ 0 & 0 & 1 & 0 \\ 0 & -1 & 0 & 0 \\ -1 & 0 & 0 & 0 \\ \end{array } \right],\\ \textbf{y}_{13}&=\left [ \begin{array}{cccccccccc } 0 & 1 & 0 & 0 \\ -1 & 0 & 0 & 0 \\ 0 & 0 & 0 & -1 \\ 0 & 0 & 1 & 0 \\ \end{array } \right],~\textbf{y}_{14}=\left [ \begin{array}{cccccccccc } 0 & j & 0 & 0 \\ j & 0 & 0 & 0 \\ 0 & 0 & 0 & j \\ 0 & 0 & j & 0 \\ \end{array } \right],\textbf{y}_{15}=\left [ \begin{array}{cccccccccc } 0 & 0 & 0 & j \\ 0 & 0 & -j & 0 \\ 0 & -j & 0 & 0 \\ j & 0 & 0 & 0 \\ \end{array } \right],\textbf{y}_{16}=\left [ \begin{array}{cccccccccc } j & 0 & 0 & 0 \\ 0 & -j & 0 & 0 \\ 0 & 0 & j & 0 \\ 0 & 0 & 0 & -j \\ \end{array } \right ] . \end{split}\ ] ] the resultant unbalanced 2-group - decodable stbc is : @xmath179 } } \end{split}\ ] ] where @xmath121 is in the first group , while @xmath122 to @xmath180 are in the second group . furthermore , @xmath122 to @xmath181 are orthogonal , which leads to @xmath182 and @xmath183 in the decoding complexity order formula ( [ dc_order ] ) for this code . @xmath184 has code rate @xmath185 , hence , its decoding complexity order calculated following ( [ dc_order ] ) is : @xmath186 in the 3gpp standardization effort , a 2-antenna stbc that fits into 3 time slots ( instead of the typical 2 time slots ) are desired due to peculiarity in the existing protocol @xcite@xcite . our code construction framework is able to easily obtain a 2-group decodable stbc @xmath187 to meet such atypical specifications , while achieving the maximum rate 3/2 and full symbol - wise diversity . @xmath188}}\ ] ] where @xmath121 is in the first group , @xmath122 to @xmath189 are in the second group , and @xmath190;\\ \textbf{c}_2&=\left [ \begin{array}{cccccccccc } j & 0 \\ 0 & -j \\ 0 & 0 \\ \end{array } \right],~\textbf{c}_3=\left [ \begin{array}{cccccccccc } 0 & 1 \\ -1 & 0 \\ 0 & 0 \\ \end{array } \right],~\textbf{c}_4=\left [ \begin{array}{cccccccccc } 0 & j \\ j & 0 \\ 0 & 0 \\ \end{array } \right],~\textbf{c}_5=\left [ \begin{array}{cccccccccc } j & 0 \\ 0 & j \\ 0 & 0 \\ \end{array } \right],~\\ \textbf{c}_6&=\left [ \begin{array}{cccccccccc } j & 0 \\ 0 & -j \\ 0 & 1 \\ \end{array } \right],~\textbf{c}_7=\left [ \begin{array}{cccccccccc } 0 & 1 \\ -1 & 0 \\ 1 & 0 \\ \end{array } \right],~\textbf{c}_8=\left [ \begin{array}{cccccccccc } 0 & j \\ j & 0 \\ j & 0 \\ \end{array } \right],~\textbf{c}_9=\left [ \begin{array}{cccccccccc } j & 0 \\ 0 & j \\ 0 & j \\ \end{array } \right].\\ \end{split}\ ] ] furthermore , @xmath122 to @xmath167 are orthogonal , which leads to @xmath191 and @xmath169 in the decoding complexity order formula ( [ dc_order ] ) for this code . @xmath187 has code rate @xmath192 , hence , its decoding complexity order calculated following ( [ dc_order ] ) is : @xmath193 we now present a method of constructing balanced 2-group - decodable stbc for @xmath14 transmit antennas over @xmath16 ( @xmath16 even ) symbol durations from two unbalanced 2-group - decodable stbc . [ balancedprop ] _ suppose that @xmath194 and @xmath195@xmath196 are the dispersion matrices of two unbalanced 2-group - decodable stbc for @xmath14 transmit antennas over @xmath197 symbol durations where @xmath198 satisfies the qoc with @xmath199 , @xmath200 satisfies the qoc with @xmath201 , @xmath56}{0pt}{}{\textbf{a}^r_2}{\textbf{a}^i_2}},\cdots , { \genfrac{[}{]}{0pt}{}{\textbf{a}^r_{l}}{\textbf{a}^i_{l}}}$ ] are linearly independent , and @xmath56}{0pt}{}{\textbf{b}^r_2}{\textbf{b}^i_2}},\cdots , { \genfrac{[}{]}{0pt}{}{\textbf{b}^r_{l}}{\textbf{b}^i_{l}}}$ ] are linearly independent too . let @xmath202,\cdots , \left [ \begin{array}{cccc } \textbf{a}_{l } \\ \textbf{b}_1 \\ \end{array } \right ] , \left [ \begin{array}{cccc } \textbf{a}_i \\ -\textbf{b}_1 \\ \end{array } \right]\right\ } , \end{split}\ ] ] @xmath203,\cdots , \left [ \begin{array}{cccc } \textbf{a}_1 \\ \textbf{b}_{l } \\ \end{array } \right ] , \left [ \begin{array}{cccc } -\textbf{a}_1 \\ \textbf{b}_{k } \\ \end{array } \right]\right\ } , \end{split}\ ] ] where @xmath204 . then , the matrices @xmath205 @xmath206 satisfy the qoc with @xmath207 ; @xmath56}{0pt}{}{\textbf{u}^r_1}{\textbf{u}^i_1}},$ ] @xmath208}{0pt}{}{\textbf{u}^r_{l}}{\textbf{u}^i_{l}}}$ ] are linearly independent , and @xmath56}{0pt}{}{\textbf{v}^r_1}{\textbf{v}^i_1}},\cdots , { \genfrac{[}{]}{0pt}{}{\textbf{v}^r_{l}}{\textbf{v}^i_{l}}}$ ] are linearly independent too . note that the @xmath194 and @xmath195@xmath196 can be the same or different . _ based on definition [ group_decodable_stbc ] , @xmath209 and @xmath210 @xmath211 in proposition [ balancedprop ] can be applied as the dispersion matrices of a balanced 2-group decodable stbc . transmit antennas over @xmath16 symbol durations constructed following proposition [ balancedprop ] . for illustration purpose , @xmath140 are shown.,width=326 ] _ for the balanced 2-group - decodable stbc for @xmath14 transmit antennas over @xmath16 ( even ) symbol durations constructed following proposition [ balancedprop ] , its code rate can approach @xmath212 for @xmath213 , or @xmath214 for @xmath215 . for the former , when @xmath138 , the code rate @xmath212 approaches @xmath14 asymptotically , i.e. , the code approaches full rate . _ for the dispersion matrices @xmath216 and @xmath217 in proposition [ balancedprop ] , we have shown in section [ sec_un_coderate ] that the maximum achievable @xmath57 is @xmath218 ( when @xmath219 ) or @xmath220 ( when @xmath221 ) . therefore , the balanced 2-group - decodable stbc constructed from proposition [ balancedprop ] is of code rate @xmath222 ( when @xmath219 ) or @xmath214 ( when @xmath221 , including @xmath175 ) . for the former , when @xmath138 , the code rate @xmath223 approaches @xmath14 asymptotically , i.e. , the code approaches full rate . the code rate variation of the proposed balanced 2-group - decodable stbc as a function of @xmath14 and @xmath16 is shown in fig . [ rate_b2groupstbc ] . note that the 2-group - decodable stbc proposed in @xcite supports @xmath175 , @xmath224 transmit antennas , and code rate @xmath2 . they are indicated as big square markers in fig . [ rate_b2groupstbc ] . clearly , our proposed construction is more scalable in code length , transmit antennas number and code rate . following the code construction in section [ section_un_construction ] , another set of dispersion matrices @xmath225 for an unbalanced 2-group - decodable stbc for 4 transmit antennas over 2 symbol durations can be obtained as : @xmath226;\ ] ] @xmath227},~ \textbf{y}'_2={\small\left [ \begin{array}{cccccccccc } 0 & 0 & j & -j \\ 0 & 0 & j & j \\ \end{array } \right]},~ \textbf{y}'_3={\small\left [ \begin{array}{cccccccccc } 0 & 0 & j & j \\ 0 & 0 & -j & j \\ \end{array } \right]},~ \textbf{y}'_4={\small\left [ \begin{array}{cccccccccc } 0 & 0 & j & j \\ 0 & 0 & j & -j \\ \end{array } \right]}.\ ] ] the resultant balanced 2-group - decodable stbc for 4 transmit antennas over 4 symbol durations is : @xmath233 } \end{split}\ ] ] where @xmath121 to @xmath234 are in the first group , while @xmath181 to @xmath235 are in the second group . for this code , @xmath236 , @xmath145 and @xmath171 , hence , its decoding complexity order calculated following ( [ dc_order ] ) is : @xmath237 in this section , we investigate the ber performance and ml decoding complexity order of the 2-group - decodable stbc examples shown earlier . in all simulations , the mimo channel is assumed to be quasi - static rayleigh fading in the sense that the channel coefficients do not change during one codeword transmission , and the channel state information is perfectly known at the receiver . in this subsection , we compare the unbalanced 2-group - decodable stbc @xmath238 in ( [ example_un_matrixnt2 ] ) with alamouti code @xcite , blast @xcite and golden code @xcite in a 2@xmath1512 mimo system with 4 bits per channel use . due to the different code rates , alamouti code , blast and golden code are simulated with 16-qam , 4-qam and 4-qam , respectively . on the other hand , @xmath238 is of code rate @xmath239 . we let one real symbol be drawn from 4-pam , and other 4 real symbols ( viewed as 2 complex symbols ) be drawn from 8-qam , then the bit rate of @xmath238 is 4 bits per channel use . the parameters of these codes are compared in table [ table_comp_2gp2nt ] , including the decoding complexity order following ( [ dc_order ] ) . table [ table_comp_2gp2nt ] shows that the proposed code has much lower decoding complexity order than golden code due to group - decodable code structure , and higher decoding complexity order than alamouti code due to higher code rate . for example , with @xmath240 bits per channel use , the decoding complexities of golden code , the proposed code and alamouti code are in decreasing order of @xmath241 . we plot the ber curves of these codes in fig . [ ber_nt2_4bps ] . to achieve full diversity , constellation rotations for @xmath238 are obtained by computer search ( drawn from 4-pam ) , @xmath242 for @xmath121 and @xmath234 ( drawn from 8-qam ) , @xmath243 for @xmath122 and @xmath244 ( drawn from 8-qam ) . ] . [ ber_nt2_4bps ] shows that the proposed @xmath238 can achieve full diversity ( same ber slope as alamouti code and golden code ) . the ber curve of @xmath238 lies between those of golden code and alamouti code , which is in accordance with their code rates . 2 mimo system with 4 bits per channel use.,width=412 ] we compare the proposed unbalanced 2-group - decodable stbc with orthogonal stbc ( ostbc ) @xcite , quasi - orthogonal stbc ( qostbc ) @xcite , perfect code @xcite and ps - sr code @xcite in a 4@xmath1512 mimo system with about 4 bits per channel use . the ostbc , qostbc , perfect code and ps - sr code are simulated with 32-qam ( 3.75 bits per channel use ) , 16-qam , 16-qam ( 3.64 bits per channel use ) , 4-qam and 4-qam , respectively . to achieve code rate 2 , we simulate the unbalanced 2-group - decodable stbc @xmath245 in ( [ example_un_matrixnt4 ] ) with @xmath180 removed . then the bit rate of @xmath245 with 4-qam is 4 bits per channel use . the parameters of these codes are listed in table [ table_comp_2gp4nt ] . it shows that the proposed code has lower ml decoding complexity order than the perfect code due to group - decodable code structure , and higher decoding complexity order than the ostbc and qostbc due to higher code rate . due to fast decoding code structure , the ps - sr code has a lower decoding complexity that that of the proposed . for example , 3.4 mm with @xmath240 bits per channel use , the decoding complexities of perfect code , the proposed code , ps - sr code , qostbc , and ostbc are in decreasing order of @xmath246 ( approximate ) . we plot the ber curves in fig . [ ber_nt4_4bps ] , where the optimum constellation rotation proposed in @xcite is applied for qostbc and the constellation rotations for @xmath245 are obtained by computer search for @xmath247 and @xmath166 , @xmath248 for @xmath249 and @xmath181 , @xmath248 for @xmath250 and @xmath167 , @xmath251 for @xmath252 and @xmath253 , @xmath254 for @xmath189 and @xmath235 , @xmath255 for @xmath256 and @xmath257 , @xmath258 for @xmath259 and @xmath260 , @xmath261 for @xmath262 and @xmath263 . ] . from fig . [ ber_nt4_4bps ] , we can see that the proposed @xmath245 has the same full diversity gain as the perfect code and the ps - sr code ( the ps - sr code has the best ber performance ) , and performs much better than ostbc and qostbc due to higher code rate . 2 mimo system with about 4 bits per channel use.,width=412 ] in this subsection , we compare the proposed balanced 2-group - decodable stbc @xmath264 in ( [ example_b_matrixnt2 ] ) with the 2-group - decodable stbc presented in @xcite in a 4@xmath1512 mimo system with 2.5 bits per channel use . since their code rates are 5/4 , they will be simulated with 4-qam . both codes have the same decoding complexity order . we plot their ber curves in fig . [ ber_nt4_25bps ] , where the constellation rotations for @xmath264 are obtained by computer search for @xmath121 and @xmath244 ( similarly @xmath265 and @xmath266 ) , @xmath267 for @xmath122 and @xmath234 ( similarly @xmath268 and @xmath235 ) , 0 for @xmath269 and @xmath270 . ] . such constellation rotation optimization are feasible because the information symbols are group - decodable and hence can be optimized separately . from fig . [ ber_nt4_25bps ] , we can see that both codes achieve full diversity gain , and the proposed code has a small 0.3 db coding gain over the code in @xcite probably because our constellation rotation angles are slightly more optimal . 2 mimo system with 2.5 bits per channel use.,width=412 ] in this subsection , we compare the proposed 3-time - slot stbc @xmath187 in ( [ example_un_matrixt3 ] ) with the other 3-time - slot stbc @xmath271 presented in @xcite in a 2@xmath1512 mimo system with 3 bits per channel use . as the code rates of @xmath187 and @xmath271 are 1.5 and 1 , they are applied with 4-qam and 8-psk , respectively . we plot their ber curves in fig . [ ber_t3_3bps ] , where the constellation rotations for @xmath187 are obtained by computer search for @xmath121 , @xmath272 for @xmath122 and @xmath265 , @xmath272 for @xmath244 and @xmath268 , @xmath273 for @xmath269 and @xmath266 and @xmath274 for @xmath234 and @xmath270 . ] . from fig . [ ber_t3_3bps ] , we can see that the proposed @xmath187 achieves a much better performance than the @xmath271 @xcite due to higher diversity gain . 2 mimo system with 3 bits per channel use.,width=412 ] in this paper , we first derive unbalanced 2-group - decodable high - rate stbc for @xmath14 transmit antennas over @xmath16 symbol durations with code rates upper - bounded by @xmath136 for @xmath126 , or @xmath137 for @xmath129 , then use them to systematically construct balanced 2-group - decodable high - rate stbc with code rates @xmath212 for @xmath213 , or @xmath214 for @xmath215 . the proposed high - rate stbc are able to achieve full symbol - wise diversity , and their code rates increase almost linearly with the transmit antenna number @xmath14 and approach @xmath14 asymptotically when @xmath138 . performance studies show that with constellation rotation optimization , the proposed 2-group - decodable stbc can achieve the same full diversity as the algebraic stbc , and much better ber performance than the ( quasi-)orthogonal stbc . the proposed code is very scalable in code length , transmit antenna number and code rate . its constellation rotation optimization is also easier to perform because its symbols are group - orthogonal and hence can be optimized separately . ( necessary condition ) assume that @xmath56}{0pt}{}{\textbf{c}^r_{i_1}}{\textbf{c}^i_{i_1}}},~ { \genfrac{[}{]}{0pt}{}{\textbf{c}^r_{i_2}}{\textbf{c}^i_{i_2}}},\cdots,$ ] @xmath71}{0pt}{}{\textbf{c}^r_{i_{l_i}}}{\textbf{c}^i_{i_{l_i}}}}$ ] are not linearly independent , i.e. , @xmath275 @xmath276 in ( [ r_hs ] ) are not linearly independent . then there exists @xmath277 where not all the scalars @xmath278 are zero . since @xmath279 , we have @xmath280 in other words , the assumed premise on @xmath281 @xmath282 is violated . therefore , the necessary condition is proved . ( sufficient condition ) assume that @xmath70 are not linearly independent , i.e. , @xmath283 where not all the scalars @xmath278 are zero . we can obtain that : @xmath284 where @xmath285 is of size @xmath286 . since @xmath285 is the channel coefficient vector with independent entries , we have @xmath287 . then , @xmath288 must be 0 . in other words , @xmath289 . therefore , @xmath290 are linearly dependent , i.e. , @xmath56}{0pt}{}{\textbf{c}^r_{i_1}}{\textbf{c}^i_{i_1}}},~ { \genfrac{[}{]}{0pt}{}{\textbf{c}^r_{i_2}}{\textbf{c}^i_{i_2}}},\cdots , { \genfrac{[}{]}{0pt}{}{\textbf{c}^r_{i_{l_i}}}{\textbf{c}^i_{i_{l_i}}}}$ ] are not linearly independent . hence , the sufficient condition is proved . \i ) when @xmath126 , after some row / column permutations , @xmath114 can be rewritten as ( b.29 ) where @xmath291 @xmath292^t(i=1,2,\cdots , n)$ ] are highlighted in dashed boxes and @xmath293 stand for the other elements in @xmath114 . in fact , @xmath294 because @xmath295 are linearly - independent row vectors when @xmath127 . this will be proved below using proof by contradiction . recall that @xmath296 $ ] and @xmath297 @xmath298 $ ] . let @xmath299 $ ] , then @xmath300 where @xmath301 is a complex - vector - to - real - vector mapping function and @xmath302 . suppose that @xmath295 are linearly dependent , then , since @xmath295 are real , there is @xmath303 where not all the real scalars @xmath304 are zero . therefore , under the inverse function for @xmath301 , ( [ real_dependent ] ) can be presented as @xmath305 since not all the values @xmath306 are zero , @xmath307 are linearly dependent and @xmath308 , which is contrary to the original premise . therefore , @xmath295 are linearly independent , @xmath294 and @xmath309 . \ii ) when @xmath129 , @xmath310 can be written as @xmath311 $ ] where @xmath312 . after some row / column permutation , @xmath114 can be rewritten as ( b.32 ) where @xmath313^t$ ] are highlighted in _ dashed _ boxes with @xmath314 and @xmath315 @xmath316^t$ ] are highlighted in _ dotted _ boxes with @xmath317 . as stated above , as @xmath318 , it can be proved by contradiction that @xmath319 are linearly independent , and @xmath320 are linearly independent too . then , @xmath321 for @xmath322 and @xmath323 for @xmath317 . therefore , @xmath324 . v. tarokh , n. seshadri , and a. r. calderbank , space - time codes for high data rate wireless communication : performance criterion and code construction , _ ieee trans . inf . theory _ , 744 - 765 , mar . 1998 . k. lu , s. fu , and x .- xia , closed - form designs of complex orthogonal space - time block codes of rates @xmath326 for @xmath327 or @xmath328 transmit antennas , _ ieee trans . inf . theory _ , vol . 4340 - 4347 , dec . 2005 . k. pavan srinath and b. sundar rajan , low ml - decoding complexity , large coding gain , full - rate , full - diversity stbcs for 2x2 and 4x2 mimo systems , _ ieee j. of sel . topics signal process . _ , 3 , no . 6 , pp . 916 - 927 , dec . 2009 . 3rd generation partnership project ; technical specification group radio access network ; evolved universal terrestrial radio access ( e - utra ) ; physical channels and modulation ( release 8) , 3gpp ts 36.211 v8.3.0 ( 2008 - 05 ) .

we study hamiltonian walks ( hws ) on the family of three dimensional modified sierpinski gasket fractals , as a model for compact polymers in nonhomogeneous media in three dimensions . each member of this fractal family is labeled with an integer . we apply an exact recursive method which allows for explicit enumeration of extremely long hamiltonian walks of different types : closed and open , with end - points anywhere in the lattice , or with one or both ends fixed at the corner sites , as well as some hamiltonian conformations consisting of two or three strands . analyzing large sets of data obtained for and 4 , we find that numbers of hamiltonian walks , on fractal lattice with sites , for behave as . the leading term is characterized by the value of the connectivity constant , which depends on , but not on the type of hw . in contrast to that , the stretched exponential term depends on the type of hw through constant , whereas exponent is determined by alone . for larger values , using some general features of the applied recursive relations , without explicit enumeration of hws , we argue that asymptotical behavior of should be the same , with $ ] , valid for all . this differs from the formulae obtained recently for hamiltonian walks on other fractal lattices , as well as from the formula expected for homogeneous lattices . we discuss the possible origin and implications of such result . _ keywords _ : solvable lattice models ; polymers , copolymers , polyelectrolytes and biomolecular solutions ; structures and conformations ( theory )

0910.2348

the fact that compact conformations of polymers are principal configurations of the native states of globular proteins makes them an important subset of all the physically accessible conformations . since compact conformations occupy space as densely as possible , one of the simplest ways to model them is to use hamiltonian walks ( hws ) on a lattice , which are , by definition , self - avoiding walks ( saws ) that visit all the lattice sites exactly once @xcite . in order to make this model more capable for capturing different features of various physical phenomena ( such as protein melting @xcite or protein folding @xcite ) local interactions can be introduced , but even in its simplest form , with no interactions taken into account , the problem of enumeration and classification of hamiltonian walks proved to be extremely difficult . studies of hamiltonian walks are primarily focused on finding the overall numbers @xmath2 of open and closed hws on lattice with @xmath3 sites . it is expected that limiting value of @xmath14 exists when @xmath15 , and its particular value @xmath16 corresponds to the configurational entropy per monomer ( site ) . this means that to the lowest approximation , @xmath2 behaves as @xmath6 and , therefore , the so - called connectivity constant @xmath7 can be interpreted as average number of steps available to the walker having already completed a large number of steps . the leading corrections to the exponential term are expected to have the power - law form @xmath17 ( as in the case of ordinary self - avoiding walks ) , and the stretched exponential form @xmath9 , with @xmath10 , so that @xmath2 should scale as @xmath18 expectations of such forms are based on the exact studies of hws on the manhattan @xcite and some fractal lattices @xcite , as well as on results obtained for closely related models of collapsed interacting self - avoiding walks on square @xcite and cubic lattice @xcite . in the case of collapsed saws appearance of the stretched exponential term , which is not present in the scaling form for non - interacting saws , was explained as a consequence of surface effects . namely , a collapsed saw forms a compact globule , with a sharp boundary separating it from the surrounding solvent , so that monomers on the boundary have smaller number of contacts with other monomers than those in the bulk of the globule , and therefore surface tension should arise . for homogeneous lattices one can assume that the boundary itself is homogeneous surface , and then straightforward arguments @xcite lead to the conclusion that term @xmath9 , with @xmath19 ( @xmath20 being the dimensionality of the lattice ) , should appear . exact studies @xcite , series analysis of data obtained via exact enumeration @xcite , as well as monte - carlo simulations @xcite of collapsed saws on homogeneous lattices , certainly confirm existence of the stretched exponential term with the proposed formula for @xmath11 . although it is believed that hw model corresponds to the interacting saw model at temperature @xmath21 , in spite of the continuous improvements of exact @xcite and monte carlo @xcite enumeration techniques , direct confirmation of scaling relation for hws on non - oriented homogeneous lattices has not been achieved yet . apart from manhattan lattice , scaling forms for hamiltonian walks were obtained only for some fractal lattices , where stretched exponential corrections for both open and closed hws were found only for @xmath22-simplex fractals with even @xmath22 @xcite . even in that cases , its presence can not be explained using a simple generalization of the argument used for collapsed self - avoiding walks on homogeneous lattices . however , results of the studies on fractals suggest that stretched exponential term can be expected on lattices where larger number of entangled conformations is possible . in order to get a deeper insight into this issue , here we apply an exact recursive method for enumeration and classification of hws on modified three - dimensional sierpinski gasket family of fractals . in contrast to most of the previously in this context studied fractals , which were embedded either in two - dimensional spaces or spaces with dimensionality higher than 3 , each member of the family studied in the present paper is embedded in three - dimensional space . therefore , hws on these fractals can be understood as a toy model for compact polymers critical behaviour in realistic nonhomogeneous 3d media , which , to the best of our knowledge , has not been studied so far . the paper is organized as follows . modified 3d sierpinski gasket fractals , as well as the method itself , are described in the next section . explicit forms of the recursion relations , obtained for the types of hamiltonian walks which are needed for generation and enumeration of all closed conformations are presented for the first three members of the fractal family . numerically analyzing these relations we find scaling form @xmath23 . using some general features of these relations ( obtained in [ ap : kb ] and [ app : xasimptotika ] ) , we argue that such form should be correct for the whole fractal family , and derive the closed - form formula for the exponent @xmath11 . in section 3 we extend the method to open hws , explicitly apply it again on the first three fractals , and then generalize it . it turns out that the number of open hws scales as @xmath24 , with the same values of the connectivity constant @xmath25 and the exponent @xmath11 as for the closed hws , but with @xmath26 . general features of the recursion relations for open hws are derived in [ app : ac ] and [ app : dh ] . all the obtained results are summarized and discussed in the section 4 , and some auxiliary considerations , needed for the discussion , are given in appendix e. three - dimensional modified sierpinski gasket ( 3d msg ) fractal is constructed recursively , starting with a unit tetrahedron . the first step of the construction , so called generator @xmath27 of order @xmath28 , is obtained by joining @xmath29 unit tetrahedrons into a @xmath8 times larger tetrahedral structure ( see ) in such a way that vertices of neighboring unit tetrahedrons are connected via infinitesimal junctions . . vertices of neighboring unit tetrahedrons are connected via infinitesimal junctions . ] enlarging the generator @xmath8 times , and substituting the smallest tetrahedrons with @xmath27 , and then repeating this procedure @xmath30 times , one obtains @xmath31 - generator of order @xmath32 , which contains @xmath33 sites . the complete 3d msg fractal with parameter @xmath8 is obtained when @xmath34 , and its fractal dimension @xmath35 is equal to @xmath36/{\ln b}\ , .\ ] ] one should note here that ordinary 3d sierpinski gasket ( sg ) fractal with parameter @xmath8 is constructed in a similar way , with the only difference being that neighboring unit tetrahedrons in 3d sg lattice are not moved apart , as is the case for the 3d msg fractal . this small change does not alter the basic properties of the lattice , such as its fractal dimension , but it simplifies the complicated scheme needed for enumeration of closed hamiltonian walks on 3d sg fractals @xcite and allows for enumeration of extremely long open hws on 3d msg fractals with @xmath1 , and 4 . in addition , general conclusions about scaling forms of the numbers of hws on 3d msg fractal with arbitrary @xmath13 can be derived , as will be explained in what follows . by definition , each hamiltonian walk on @xmath37 structure visits all its @xmath38 sites exactly once . in msg fractal . the generator consists of 40 vertices , which are numbered consecutively from the first ( arbitrarily chosen ) vertex visited by the hw , to the last one ( which is connected with the first one ) . one can notice that hw conformation within the unit tetrahedron can be either one - stranded ( @xmath39-type ) , like the one connecting the sites 26 - 27 - 28 - 29 , or two - stranded ( @xmath40-type ) , like the conformation traversing the tetrahedron with vertices 4 - 5 - 12 - 13 . ] an example of closed hamiltonian walk on the generator @xmath41 is shown . one can notice that this hw can be decomposed into @xmath42 parts ( within the ten unit tetrahedrons ) , which are either one- or two - stranded . this observation can easily be generalized to generators of higher order and any @xmath8 : any closed hw on @xmath31 can be decomposed into @xmath43 parts within the same number of generators @xmath37 , from which the generator @xmath31 is made of . these parts traverse each @xmath44 either once or two times , and we shall call the corresponding configurations @xmath39- or @xmath40-type hw steps , respectively , whereas numbers of hws of these types within a @xmath44 will be denoted by @xmath45 and @xmath46 , respectively . using these numbers , the overall number of closed hws on @xmath31 can be expressed as @xmath47 where @xmath48 are the numbers of closed hws configurations within @xmath31 , with @xmath49 steps of @xmath40type , and @xmath50 @xmath39steps . for instance , the path presented in is one of @xmath51 possible closed hws within the @xmath41 , contributing to the term @xmath52 in the corresponding equation . the upper limit @xmath53 in the sum in is equal to 0 for @xmath54 , whereas for @xmath13 it can be shown ( see [ ap : kb ] ) that @xmath55 due to the self - similarity of the lattices under study , the numbers @xmath48 do not depend on @xmath30 , and , also , numbers @xmath45 and @xmath46 fulfil recursion relations of the following form : @xmath56 where coefficients @xmath57 and @xmath58 depend only on @xmath49 and @xmath8 , and are positive integers for all @xmath13 , including zero for @xmath54 . for instance , for @xmath54 these relations are recursion relations are the same as for the 4-simplex fractal lattice @xcite . ] @xmath59 and we were able to find the explicit form of relations for @xmath60 and 4 also , by direct computer enumeration of possible hw conformations within the msg generator . the corresponding coefficients @xmath57 and @xmath58 are presented in , @ccccccccccccc & & & & & & + @xmath49&@xmath57&@xmath58&&&@xmath49&@xmath57&@xmath58&&&@xmath49&@xmath57&@xmath58 + 0&-&22&&&2&-&4308&&&7&-&26465392 + 1&-&0&&&3&-&1936&&&8&-&99652120 + 2&6&0&&&4&-&5808&&&9&-&151443088 + 3&4&4&&&5&3192&1888&&&10&23848720&199987864 + 4&2&1&&&6&848&2534&&&11&58605536&204194352 + -&-&-&&&7&1728&1056&&&12&78351952&172479256 + -&-&-&&&8&664&596&&&13&81469824&126633376 + -&-&-&&&9&332&160&&&14&66418856&78454776 + -&-&-&&&10&64&32&&&15&43526336&41200784 + -&-&-&&&-&-&-&&&16&22989024&18548660 + -&-&-&&&-&-&-&&&17&9642816&6662824 + -&-&-&&&-&-&-&&&18&3032724&1901008 + -&-&-&&&-&-&-&&&19&626056&397392 + -&-&-&&&-&-&-&&&20&62434&42514 + while for larger @xmath8 values they could not be reached within the reasonable time with the computer facilities available to us . however , one can show ( see [ ap : kb ] ) that for general @xmath13 lower limits of the sums in relations are equal to @xmath61 which is a result essential for establishing some general conclusions , as will be explained in the following paragraphes . once the explicit form of recursion relations and is established , starting with the initial values @xmath62 , and @xmath63 , corresponding to the unit tetrahedron , one can calculate the numbers @xmath45 , @xmath46 , and subsequently the overall number @xmath64 of closed hws , in principle for any @xmath30 . however , since these numbers quickly become extremely large , it is useful to introduce new variable @xmath65 . then , from it follows that recursion relation for the variable @xmath66 , for @xmath13 case is the same as for 4-simplex fractal lattice , which is given in @xcite . ] , has the form @xmath67 numerical analysis of this recursion relation for @xmath60 and @xmath68 cases , shows that starting with the initial value @xmath69 , @xmath66 quickly tends to 0 , as @xmath30 grows . assuming that this is correct for general @xmath13 , for large values of @xmath30 the above equation can be approximated as @xmath70 from which follows @xmath71 where @xmath72 is some constant , whose value depends on @xmath8 , but it is always less than 1 . for @xmath54 the corresponding relation is @xmath73 @xcite , and particular values of @xmath72 for @xmath1 and 4 cases are given in . @crcllllll & @xmath35&@xmath53&@xmath72&@xmath25 & @xmath74 & @xmath75&@xmath11&@xmath76 + 2 & @xmath77&0&0.8366&1.0876&0.8366&0.9147&0.5&0.5 + 3 & @xmath78&2&0.8835&1.4404&0.8963&0.9554&0.4471&0.4471 + 4 & @xmath79&7&0.8639&1.4686&0.8696&0.9496&0.3667&0.4628 + recursion relation for numbers @xmath46 of two - stranded hws , with the variable @xmath65 obtains the form @xmath80 from which one gets @xmath81 numerically iterating this recursion relation , together with , one finds that @xmath82 where @xmath25 ( see ) is constant larger than 1 . on the other hand , the overall number @xmath64 of closed hws on the generator of order @xmath32 can be expressed as @xmath83 so that @xmath84 from the asymptotical behavior of the number @xmath66 , and from the fact that @xmath85 tends to constant value @xmath86 when @xmath87 , it then follows that @xmath88 where @xmath89 is overall number of vertices within the generator of order @xmath30 . to find the leading - order correction to the asymptotic behavior of @xmath90 we first introduce variable @xmath91 which , as follows from , satisfies the relation @xmath92\ , .\ ] ] then , using , one obtains @xmath93\ , , \ ] ] from which , taking into account the large @xmath49 behavior of @xmath94 , it follows that @xmath95 substituting this relation into one derives @xmath96 with @xmath97}\ , , \label{eq : sigma}\ ] ] and @xmath98 at the end of this section we want to emphasize that the crucial step in preceding derivation , which led to the scaling form , with the value of exponent @xmath11 given by formula , was the assumption that @xmath66 tends to 0 for general @xmath8 ( explicitly confirmed only up to @xmath68 ) . its direct consequence is approximate difference equation , which is obtained from the exact relation . on the other hand , the key ingredient of that relation is the fact that @xmath99 , due to which @xmath66 behaves as @xmath100 , for @xmath101 . in that sense , exact expressions for @xmath102 and @xmath103 , obtained in appendix a for general @xmath13 , are essential for establishing the scaling form , whereas the particular values of the coefficients @xmath57 and @xmath58 of the recursion relations do not affect neither its general form , nor the value of @xmath11 . however , whether @xmath66 tends to 0 or not certainly depends on the values of @xmath57 and @xmath58 . analyzing data given in for @xmath60 and @xmath68 , one can observe that @xmath104 , for @xmath105 . it can be shown ( see [ app : xasimptotika ] ) that such inequality for general @xmath13 is sufficient to prove that numbers @xmath66 , which satisfy difference equation with the initial condition @xmath69 , tend to 0 when @xmath34 . unfortunately , the inequality itself we were not able to prove , but , since it seems plausible , we think that scaling relation can be accepted as valid for general @xmath8 . any open hw on generator @xmath31 of order @xmath32 can be decomposed into @xmath43 parts within its @xmath43 constitutive generators @xmath37 of order @xmath30 . the parts which contain ending points ( see ) can be of four different types : * @xmath106-type , which consists of one hw , with one end at one vertex of @xmath44 and the other end at any other site of @xmath37 , including its vertices ; * @xmath107-type , composed of two non - intersecting saws , one with both ends at the vertices of the @xmath37 , and the other with one end at the third vertex of the @xmath37 and the second anywhere within it - these two saws together visit all the sites of the @xmath37 ; * @xmath108-type , consisting of two non - intersecting saws , each of them starting at different vertex of @xmath37 and ending anywhere within it , in such a way that all sites of @xmath37 are visited ; * @xmath109-type , comprised of three non - intersecting saws , each of them starting at different vertex of @xmath37 , one ending at the fourth vertex , and the remaining two anywhere within @xmath37 - again , all sites of @xmath37 should be visited by these three saws . msg fractal . points at which hw begins or terminates are marked as full red circles , for the sake of better recognition . if both ends of the walk are in the same tetrahedron the two - leg hw conformation within that tetrahedron is either of @xmath108- or @xmath109-type . otherwise , if end - points of the walk belong to different tetrahedrons , the corresponding one - leg conformations are of @xmath106- or @xmath107-type . ] @xmath106 and @xmath107type configurations have one dangling end , therefore we shall call them one - leg configurations ( steps ) , whereas @xmath108 and @xmath109configurations , with two dangling ends , will be called two - leg steps . if both end - points of the complete hw lie in the same @xmath37 , that @xmath37 contains a two - leg step ( of @xmath108 or @xmath109type ) , whereas the remaining @xmath110 generators @xmath37 contain either @xmath39 or @xmath40step . otherwise , when end - points are in two different @xmath37 generators , the complete open hw has two one - leg steps ( of @xmath106 or @xmath107type ) , and the remaining @xmath111 parts of the walk are of either @xmath39 or @xmath40type . therefore , one concludes that overall number @xmath112 of open hws within @xmath31 is equal to @xmath113 where @xmath114 , @xmath115 , @xmath116 , @xmath117 and @xmath118 are polynomials in @xmath45 and @xmath46 , of power @xmath111 or @xmath110 , and @xmath119 , @xmath120 , @xmath121 and @xmath122 are overall numbers of corresponding type hws within @xmath37 generator . numbers of one - leg conformations fulfill recursion relations of the following form : @xmath123 with @xmath124 while numbers of two - leg conformations obey relations of the form @xmath125 with @xmath20 and @xmath126 being some polynomials in @xmath45 and @xmath46 . for instance , for @xmath54 relations and are : @xmath127 they coincide with the corresponding relations for 4-simplex fractal lattice , as well as does the complete further analysis , which can be seen in @xcite . in the remaining part of this section we present the general analysis for @xmath13 3d msg fractals . for general @xmath13 it can be shown ( see [ app : ac ] ) that polynomials @xmath128 and @xmath129 , appearing in , have the form @xmath130 starting with the initial values for the numbers @xmath45 , @xmath46 , @xmath119 , @xmath120 , @xmath121 , and @xmath122 , and using the recursive relations , , and , one can calculate these numbers in principle for any @xmath30 , and substituting them into , eventually find the overall number @xmath112 of open hws . since all these numbers increase rapidly with @xmath30 , it is useful to introduce variables @xmath131 in addition to already defined @xmath65 . with these variables @xmath112 can be rewritten as @xmath132\ , , \label{eq : otvoreneuv}\end{aligned}\ ] ] and explicitly calculated using corresponding recursion relations for @xmath133 , @xmath134 , @xmath135 , and @xmath136 , simultaneously with and . in [ app : ac ] it is shown that the upper limits of the sums in are equal to @xmath137 so that by dividing with recursion relation for @xmath46 , given in , for @xmath133 and @xmath134 one obtains recursion relations @xmath138 where @xmath139 are functions of @xmath66 of the following form : @xmath140 and @xmath141 is defined in . since @xmath142 and @xmath143 when @xmath34 , it follows that @xmath144 , @xmath145 , implying that @xmath133 tends to 0 , and @xmath134 to some constant value . this is indeed correct for @xmath60 and @xmath68 cases , for which we managed to find the complete set of coefficient @xmath146 ( see ) . @cccccccccc & & + & & + @xmath49&@xmath147&@xmath148&@xmath149&@xmath150 & @xmath147&@xmath148&@xmath149&@xmath150 + 0&378&1050&202&620&726000&2078460&454340&1343670 + 1&1722&5256&798&2854&6785652&20162052&4124238&12825924 + 2&2748&8964&1962&8696&30159948&92803776&18522682&61201186 + 3&3588&15144&2912&15172&85107696&270289488&56914854&200289608 + 4&2976&9576&3126&24064&171919920&559983864&135608360&503818972 + 5&-&-&4632&23684&265804824&875876784&260879696&1011455276 + 6&-&-&2420&27088&319608912&1041132024&412200376&1653387800 + 7&-&-&1304&4308&290292720&895898352&548029844&2243585880 + 8&-&-&-&-&185693208&493595664&612939852&2479455304 + 9&-&-&-&-&51662736&71546160&552106184&1451395108 + 10&-&-&-&-&-&-&407928692&1451395108 + 11&-&-&-&-&-&-&226652724&546796688 + 12&-&-&-&-&-&-&68954256&26465392 + starting with the initial values @xmath151 , @xmath152 , _ i.e. _ @xmath153 and @xmath154 , and numerically iterating relations , for both @xmath60 and @xmath68 , already after five iterations one obtains limiting values of @xmath134 : @xmath155 for @xmath60 , and @xmath156 for @xmath68 . in [ app : dh ] it is shown that recursion relations for the variables @xmath135 and @xmath136 can be put into the following matrix form @xmath157 where @xmath158 and @xmath159 are functions of @xmath66 of the form @xmath160 with @xmath161 and @xmath162 being some positive constant integers . using these relations , one can calculate @xmath135 and @xmath136 , and by putting them , together with @xmath66 , @xmath133 , @xmath134 and @xmath46 , into , finally can evaluate @xmath163 , for any @xmath30 . in order to find asymptotic behavior of @xmath163 , we first notice that relations can be combined with the relation for @xmath66 , so that one can establish recursion relations for @xmath164 and @xmath165 , terms through which variables @xmath135 and @xmath136 appear in . using the facts that @xmath142 , @xmath166 , @xmath167 , and @xmath168 , when @xmath34 , it can be shown that @xmath169\ , , \label{eq : wqasim}\ ] ] where @xmath170 and @xmath171 are some constants ( see [ app : dh ] ) . this relation implies that @xmath164 and @xmath165 tend to zero as @xmath34 , and consequently , comparing the terms containing @xmath164 and @xmath165 with those with @xmath133 and @xmath134 in , one can conclude that @xmath172 this means that for @xmath101 number @xmath112 behaves as @xmath173 so that , using , one obtains @xmath174 consequently , taking into account , it follows that overall number @xmath175 of open hws scales as @xmath176 where , as in , @xmath177 , and particular values of @xmath25 , @xmath75 and @xmath72 are given in . in this paper we have analyzed asymptotic behavior of the numbers of open and closed hamiltonian walks on three - dimensional modified sierpinski gasket family of fractals . numbers of extremely long hws on these lattices can be generated by applying an exact recursive enumeration scheme , based on the fact that any hw on the @xmath178th step of the construction of the 3d msg fractal , @xmath31 , consists of @xmath179 hws within its @xmath43 constitutive @xmath37 structures , which can be of one of six possible types : @xmath180 , and @xmath109 ( and ) . numbers of these hws , @xmath119 , @xmath45 , @xmath120 , @xmath121 , @xmath46 , and @xmath122 , fulfil closed set of recursive relations , , and , which , due to the self - similarity of the lattices under study , do not depend on @xmath30 , and therefore can be obtained by explicit enumeration and classification of hw conformations on the first step of the fractal construction , @xmath27 . as @xmath8 grows , number of hw conformations rapidly increases already on @xmath27 , so that we have managed to find explicit form of the recursive relations only up to @xmath68 . however , we have shown that for any @xmath13 these recursive relations have some features that enable general analysis , leading to conclusion that overall numbers @xmath181 and @xmath175 of closed and open hws , respectively , on any 3d msg fractal , scale with the number @xmath89 of the lattice sites as @xmath182^{n_l^\sigma}\ , , \quad z_l^o\sim \omega^{n_l}\left[\lambda^{\frac{k_c+3}{n_g-3}\,4^{-\sigma}}\right]^{n_l^\sigma}\ , , \ ] ] with @xmath11 given by formula : @xmath183 constants @xmath25 and @xmath72 can be obtained numerically , using relations @xmath184 and their particular values for @xmath1 , and 4 are given in . number @xmath185 represents the maximal number of @xmath37 generators within the @xmath31 they belong to , which are traversed by two - stranded @xmath40-type hw conformation ( @xmath40-step ) within any closed @xmath186 on @xmath31 ( see equation and [ ap : kb ] ) . as one can see , hw conformations that traverse @xmath37 , and recursion relations for the corresponding numbers @xmath45 and @xmath46 , are sufficient for obtaining connectivity constant @xmath25 that governs the leading exponential term in the scaling forms for @xmath181 and @xmath175 , as well as constant @xmath72 which appears in their correction terms . whereas the leading term @xmath187 is the same for both closed and open hws , correction term in both cases has the same stretched exponential form @xmath188 , but with different values for @xmath189 and @xmath190 , indicating that number of open hws is larger than the number of closed hws . one should notice here that the fact that scaling form obtained for open hws is determined by the behavior of the numbers @xmath45 and @xmath46 only , means that contribution of one and two leg hw conformations ( @xmath106 , @xmath107 , @xmath108 and @xmath109type walks ) , _ i.e. _ hws with their ends in the interior of @xmath37 fractal structures , is not significant . furthermore , in sections 2 and 3 it was elaborated that the main mathematical reason for obtaining scaling forms with the stretched exponential correction term is the fact that @xmath191 , meaning that entangled conformations ( with large number of two - stranded parts ) dominate over those in which hws rarely return to already visited fractal generators . results of the study of hws on 3d msg fractals , presented in this paper , should be compared with the results recently obtained for other fractal lattices @xcite . for @xmath22-simplex fractal lattices with even values of @xmath22 the same asymptotic behavior of the hw numbers was found , @xmath192 , but with @xmath193 , which differs from the formula derived here for @xmath11 . the existence of the term @xmath9 was also related with the facts that : ( 1 ) entangled hws prevail , and ( 2 ) hws with `` interior '' terminating points do not contribute to the asymptotic behavior of hw numbers , as in the 3d msg fractals case . for odd @xmath22 , stretched exponential terms in the scaling forms for hws on @xmath22-simplex fractals do not exist , hws with `` interior '' terminating points do affect the scaling forms for the numbers of open hws , and entangled conformations do not dominate in this case . for given - mandelbrot and 2d modified sierpinski gasket fractals ( which are two - dimensional generalizations of the sierpinski gasket and 3-simplex fractal , respectively ) stretched exponential terms are also absent , whereas again , one and two leg conformations are necessary for obtaining the scaling forms for open hws , and , due to the specific topology of these lattices , only one - stranded hw conformations are possible . therefore , results of the present study confirm the assumption that existence of the stretched exponential term is related with the issue of hw entanglement . however , the discrepancy between the formulas obtained for the exponent @xmath11 for different fractal families , and , more generally , the question which properties of the underlying lattice and in what way determine @xmath11 , deserves further consideration . to this end we recall that stretched exponential terms were obtained for hws on manhattan @xcite , as well as for low - temperature saws on square @xcite and cubic @xcite lattice . in this context , one should also mention the spiral saws on 2d regular lattices @xcite and lattice animals on some hierarchical lattices . whereas , to the best of our knowledge , physical interpretation of the stretched exponential term for the 2d spiral saw models has never been proposed , its presence in the scaling forms obtained for lattice animals on hierarchical lattices was directly connected with the existence of the sets of sites with different coordination numbers @xcite . utilizing similar idea , in the case of low - temperature ( collapsed ) saws it was explicitly explained in @xcite why term of the form @xmath9 , with @xmath19 for regular lattices , should arise . however , as was elaborated in @xcite , direct application of such approach on hws on fractal lattices does not give satisfactory result . yet , the scaling forms obtained in this paper can be expound in the spirit of the physical reasoning given in @xcite , as will be explained in the following paragraph . we shall first focus on closed hws on @xmath37 with the maximal number of two - stranded @xmath40 conformations within the unit tetrahedrons . such hws represent maximally entangled closed ( mec ) compact conformations and they accomplish the maximal possible connectedness between the generators @xmath37 of all orders @xmath30 . now , from all the @xmath38 sites visited by such maximally entangled hw , observe those which belong to one - stranded conformations ( @xmath39-steps ) within the unit tetrahedrons , and which are directly connected only with the sites belonging to the same unit tetrahedron . for instance , vertices 9 and 10 , or 15 and 16 in represent examples of such sites . since these sites are not directly connected with the other tetrahedrons of the lattice ( via hw ) , they are the maximally isolated sites within the maximally entangled closed hw conformation . it is shown in [ app : diskusija ] that number @xmath194 of such sites is equal to @xmath195 and one can easily see that , using this expression , scaling relation for the numbers od closed hws can be transformed into the following form @xmath196 in a similar way ( see [ app : diskusija ] ) , one can show that scaling relation for open hws can be expressed as @xmath197 where @xmath198 is the number of maximally isolated sites on @xmath37 visited by maximally entangled open hw ( which is an open hw with the maximal number of @xmath40-steps on unit tetrahedrons and with both loose ends being of type @xmath107 , on all levels @xmath30 ) . comparing forms and with the scaling relations expected for hws on homogeneous lattices , one can say that in this case terms @xmath199 and @xmath200 play the role of the stretched exponential correction term @xmath9 in the case of homogeneous lattices . indeed , for homogeneous lattices @xmath201 , with @xmath19 , is proportional to the number of sites on the lattice boundary , which have smaller number of neighbors than the bulk sites , and in that sense they are similar to the maximally isolated sites within the maximally entangled hws on 3d msg lattices . therefore , one might generally expect that scaling relation for the number of hws on nonhomogeneous lattices instead of stretched exponential correction term has a term @xmath202 , where @xmath203 is the number of conveniently defined `` maximally isolated sites '' . relation between @xmath203 and the overall number of lattice sites @xmath3 depends on the particular topology of the lattice under study , which is the reason why different scaling forms , as functions of @xmath3 , are obtained . conclusion of the previous paragraph can be supported by performing a similar analysis in the case of hws on previously studied fractal families @xcite . for instance , for @xmath22-simplex fractals maximally isolated sites can be defined in a similar way as for 3d msg fractals , _ i.e. _ these are the sites which are directly connected by maximally entangled hw only with the sites within the same fractal generator they belong to . here , maximally entangled hws are those that contain maximal number of maximally - stranded hw parts . then , it can be shown @xcite that @xmath204 and @xmath205 , where @xmath206 . since overall numbers of closed and open hws on @xmath22-simplex fractal with even @xmath22 asymptotically behave as @xmath207 ( see ( 5.8 ) in @xcite ) and @xmath208 ( formula ( 5.14 ) in @xcite ) , respectively , they immediately fit into the scaling relation @xmath209 , where @xmath210 and @xmath211 for both types of hws . for odd values of @xmath22 , however , it can be shown @xcite that number @xmath203 of maximally isolated sites , for both open and closed hws , is proportional to @xmath38 , so that any term of the form @xmath202 can only contribute to the leading term @xmath6 . this certainly is in accord with the scaling forms found in @xcite for odd @xmath22 , which do not have stretched exponential corrections . to conclude , we may say that presented exact study of hamiltonian walks on modified three - dimensional sierpinski gasket fractals supports the idea that entangled conformations are of the most physical importance for the behavior of compact polymers in inhomogeneous media . the introduced concept of `` maximally isolated sites '' proved to be useful in physical interpretation of different scaling forms obtained for hamiltonian walks on all so far studied fractals . whether such concept can be applied on homogeneous and other nonhomogeneous lattices will be the matter of our further investigations . this paper has been done within the project no oi 141020b , funded by the serbian ministry of science and environmental protection . in this appendix we prove that numbers @xmath53 , @xmath102 and @xmath103 are given by formulae and . the number @xmath53 is the maximum number of @xmath40-steps within the closed hw configuration ( see ) , whereas the numbers @xmath102 and @xmath103 are related to the maximum number of @xmath40-steps within the hw configurations of @xmath39 and @xmath40-type ( see ) , _ i.e. _ the conformations that traverse the generator of order @xmath32 once or two times , respectively . by , , @xmath40-step here we imply any hw conformation consisting of two mutually avoiding strands , both traversing the generator of order @xmath30 ( see ) . in a similar way , , , @xmath39-step is any hw conformation that consists of one strand traversing @xmath37 . according to our definition of the modified sierpinski gasket fractals , generators @xmath37 within the generator @xmath31 are connected via infinitesimal junctions . each of these junctions connects two , three or four neighboring @xmath37 generators . , , twofold junction connects vertices of neighboring @xmath37 , both lying in the same edge of @xmath31 ( for instance , vertices 3 and 4 in are joined by such junction ) . vertices of neighboring @xmath37 that lie inside the faces of @xmath31 ( such as sites 5,6 and 19 in ) are connected by three - folded junction , whereas four - folded junction occur in the interior of @xmath31 , and can exist only for @xmath212 . in we explicitly indicate some of the junctions connecting the vertices of the generators @xmath213 , within the generator @xmath214 . of order @xmath32 , for @xmath68 3d msg , is presented , as it is seen from above . gray - shaded small tetrahedrons represent generators @xmath213 of order @xmath30 , lower layers being darker . vertices lying in the same horizontal plane are indicated with the same symbol , being also the same for two adjacent planes containing vertices that belong to different generators . on the right - hand side of the picture horizontal layers of tetrahedrons are split and slightly magnified in order that all vertices , as well as some junctions can be seen . in particular , twofold junctions in the lowest layer of vertices are indicated by yellow lines , whereas junctions joining three vertices are given in green color . there is only one four - fold junction the corresponding four connected vertices are encircled , and in the magnified circle these vertices are presented as seen in three dimensions , with the junction indicated in orange color . one can check that overall number of junctions is @xmath215 ( three twofold junctions for each of six @xmath214 edges , three three - fold junctions for each of four @xmath214 faces , and one four - fold interior junction ) , which certainly is in accord with formula . ] it is not difficult to show that the number of junctions which connect @xmath43 generators @xmath37 into the generator @xmath31 is equal to @xmath216 from the restraint that each site of the lattice has to be visited exactly once , follows that each junction ( _ i.e. _ its middle point ) can be traversed at most once . now , suppose that closed hw conformation within @xmath31 consists of @xmath217 @xmath39-steps and @xmath218 @xmath40-steps . each @xmath37 is traversed by hw , which implies that @xmath219 each @xmath39step uses two junctions and each @xmath40step uses four junctions . however , since each junction connects two @xmath37 generators , the overall number of visited junctions is equal to @xmath220 . in addition , steps through corner @xmath44 generators for the closed hw have to be of the @xmath39-type ( see ) , which means that one of the three junctions for each of four corner generators is certainly not used . this implies that inequality @xmath221 is satisfied . then , from and directly follows that @xmath222 _ i.e. _ the maximum number of @xmath40-steps within the closed hw conformation is indeed equal to @xmath223 , viewed from above , with the upper three layers of constitutive generators @xmath224 ( gray - shaded tetrahedrons - triangles ) moved to the right - hand side of the picture , for the sake of better visuality . red oriented line represents a @xmath39-type hw configuration , which contains 16 @xmath40-steps , _ i.e. _ the maximum possible number of them , given by formula . remaining 19 steps are of the @xmath39-type , and they are depicted as straight lines connecting only two vertices of the traversed @xmath224 generator , but it is implied that all sites within it are visited . the numbers in the figure denote points where hw leaves or enters the lower part of @xmath225 . ] next , we consider a @xmath39type hw configuration on @xmath31 generator , with @xmath217 @xmath39steps , and @xmath218 @xmath40steps . steps through two corner @xmath37 generators , at which hw starts or terminates , can be of either @xmath39 or @xmath40type . however , both remaining two corner @xmath37 generators must be traversed by @xmath39step ( see ) , meaning that two of @xmath226 junctions are never used . on the other hand , each @xmath39step uses two junctions , unless it is the first or the last step of the hw configuration , in which case it utilizes only one junction ( since the corner vertex does not belong to any junction ) . similarly , @xmath40step which traverses interior @xmath37 generators uses four junctions , whereas possible @xmath40step through either the first or the last traversed @xmath37 uses only three junctions . suppose that both the first and the last visited corner generators contain @xmath40-step . then , these two steps together use six junctions , and the remaining @xmath227 , , interior @xmath40steps use @xmath228 , whereas @xmath217 @xmath39steps use @xmath229 junctions . since every junction can be traversed only once , and two consecutive steps share junctions , it follows that altogether @xmath230/2 $ ] junctions are visited by such hw conformation . this implies the following inequality @xmath231\leq n_j-2\ , , \ ] ] which , together with the relation @xmath232 , gives @xmath233 in the remaining two possible situations , when the first and the last @xmath37 are traversed by either both @xmath39steps , or one of them by @xmath39step and the other by @xmath40step , in a similar way it can be shown that inequality holds as well . type hw conformation on @xmath225 with 19 @xmath40steps , which is the maximum possible number @xmath234 of them , given by formula . the two strands of the conformation are depicted as oriented red and blue lines . the red strand starts at vertex labeled with number 1 , and terminates at vertex 7 , whereas the numbers 2 to 6 mark the sites at which the strand leaves or enters the second layer of @xmath235 generators . ] as one can see in an @xmath40type hw conformation within @xmath31 can use all three junctions of any corner @xmath44 , which happens when such @xmath44 is traversed by @xmath40-step . this also means that all @xmath226 junctions within @xmath31 can be utilized . suppose that all four corner @xmath44 generators are traversed by @xmath40steps altogether 12 junctions are used by these four @xmath40-steps . if the whole hw conformation contains @xmath218 @xmath40steps , the remaining @xmath236 interior @xmath40-steps take @xmath237 junctions , whereas to all @xmath217 @xmath39steps @xmath229 junctions correspond . every used junction connects two steps , therefore in this case inequality @xmath238\leq n_j\ , , \ ] ] holds , which , again with @xmath232 , leads to @xmath239 in a similar manner one can show that in the remaining four possibilities : one , two , three or four @xmath39steps through the corner @xmath37 generators , inequality is valid again . note that formulae , and are not correct in the @xmath54 case . in this case all four constitutive @xmath240 generators of the @xmath241 generator are at its corners , and each of them is connected with each of the other three by two - folded junction . since two corner generators have a junction in common , the number of usable junctions in relations ( a.3 ) and ( a.6 ) is @xmath242 and @xmath243 instead of @xmath244 and @xmath242 , respectively . if we represent each @xmath245 by a point ( vertex ) then @xmath246 represents a complete graph of four points . since there can be neither subgraph with three points of degree three and one point of degree one , nor subgraph with two points of degree three and two of degree one , terms in relation with @xmath247 and @xmath248 are not possible . as it was stressed in section [ sec : zatvorene ] asymptotical behavior of the parameter @xmath65 is of the greatest importance for obtaining the scaling form of hws . here we want to prove that assumption @xmath104 , for @xmath105 , leads to the conclusion that @xmath142 , when @xmath34 . numbers @xmath66 satisfy the exact relation , from which follows that @xmath249 where we used the relation @xmath250 , obtained in the previous appendix , the fact that all coefficients @xmath58 and @xmath57 are positive , as well as the proposed inequality . this means that sequence of positive numbers @xmath66 is monotonically decreasing , and since @xmath69 , it follows that this sequence has non - negative limiting value @xmath251 . if @xmath252 , again from , when @xmath34 , follows equation @xmath253 which , after dividing by @xmath254 , can be transformed to @xmath255 however , the latter equation can not be satisfied , since its right - hand side has the form of polynomial in @xmath254 , with all its coefficients being positive numbers . consequently , one concludes that @xmath256 . in this appendix we first show that polynomials @xmath257 , appearing in recursion relations for numbers @xmath119 and @xmath120 of one - leg hw conformations , have the form , with @xmath258 given by . it is obvious that dangling end - step of one - leg configuration within @xmath31 can be either of @xmath106 or @xmath107type . the remaining part of the hw traverses the remaining @xmath110 generators @xmath37 , and consists of @xmath217 @xmath39steps and @xmath218 @xmath40steps , so that equation @xmath259 is satisfied . consider an @xmath106type hw configuration on generator @xmath31 . one of its end - steps is fixed at one of the four corner generators @xmath37 and it can be either @xmath39 , @xmath40 , or @xmath107step , respectively using one , three , or two junctions . two of the remaining three corner generators are traversed by @xmath39steps , each of them using two junctions of the corresponding @xmath37 , whereas its third junction can not be traversed . the last corner generator is either traversed by a @xmath39step , or it contains dangling @xmath106 or @xmath107step , and such steps use two , one , or three junctions , respectively . if a corner generator contains @xmath106step , only one of its corresponding three junctions can be traversed . now , suppose that , , corner steps are @xmath260 , whereas the dangling @xmath106-step , using one junction , is in the interior generator . such hw traverses @xmath261 junctions , and , since corner @xmath39-steps can not use altogether three of all @xmath226 junctions , it follows that @xmath262 which with gives @xmath263 in a similar manner one can analyze all the other possibilities and obtain corresponding maximum value of @xmath218 . from the , in which all of these possibilities , together with the corresponding numbers of used and usable junctions , as well as @xmath264 are given , one can see that @xmath265 . @ccccccc dangling end - step & , , corner & , , corner & number of & number of + and its position & passing - steps & fixed - step & used junctions & usable junctions&@xmath266 + @xmath106 - interior @xmath37&@xmath267&@xmath40&@xmath220&@xmath268 & + @xmath106 - interior@xmath37&@xmath267&@xmath39&@xmath220&@xmath269&@xmath270 + @xmath106 - corner @xmath37&@xmath271&@xmath40&@xmath220&@xmath244&@xmath272 + @xmath106 - corner @xmath37&@xmath271&@xmath39&@xmath220&@xmath273&@xmath274 + @xmath107 - interior @xmath37&@xmath267&@xmath40&@xmath275&@xmath268&@xmath272 + @xmath107 - interior @xmath37&@xmath267&@xmath39&@xmath275&@xmath269&@xmath274 + @xmath107 - corner @xmath37&@xmath271&@xmath40&@xmath275&@xmath242 & + @xmath107 - corner @xmath37&@xmath271&@xmath39&@xmath275&@xmath244&@xmath270 + @xmath107 - corner @xmath37&@xmath267&@xmath107&@xmath275&@xmath244&@xmath270 + @xmath107type hw configuration on @xmath31 consists of two strands : one of @xmath39type , _ i.e. _ with two ends fixed at two corners of the @xmath31 , and the other of @xmath106type , with one end fixed at the third corner , whereas its second end is free , _ i.e. _ it can be anywhere within the @xmath31 , in some of interior generators @xmath37 , as well as in any of the four corner - generators @xmath37 . the three steps containing fixed ends can be of either @xmath39 or @xmath40type , whereas the corner - generator , which does not contain the fixed - end , is either traversed by a @xmath39step or it contains the dangling @xmath106 or @xmath107step . the other relevant details of possible situations , regarding the type and the position of the step containing the dangling end , are listed in . one can see that hw configurations with @xmath106 dangling end - step and maximal number of @xmath40steps are those with the dangling end - step in the interior @xmath37 , whereas the three steps with ends fixed at corner - generators are of @xmath40type , and the fourth corner - generator is traversed by a @xmath39step , and thus conclude that @xmath276 . configurations with @xmath107 dangling end and maximal number of @xmath40steps have also all three fixed - steps of @xmath40type , while the dangling end is in the fourth corner - generator , therefore @xmath277 . @ccccccc dangling end - step & , , corner & , , corner & number of & number of + and its position & fixed - steps & passing - step & used junctions & usable junctions&@xmath266 + @xmath106 - interior @xmath37&@xmath267&@xmath39&@xmath278&@xmath279&@xmath274 + @xmath106 - interior @xmath37&@xmath280&@xmath39&@xmath278&@xmath269&@xmath272 + @xmath106 - interior @xmath37&@xmath281&@xmath39&@xmath278&@xmath268&@xmath282 + @xmath106 - interior @xmath37&@xmath283&@xmath39&@xmath278&@xmath243 & + @xmath106 - corner @xmath37&@xmath267&@xmath106&@xmath278&@xmath284&@xmath285 + @xmath106 - corner @xmath37&@xmath280&@xmath106&@xmath278&@xmath273&@xmath270 + @xmath106 - corner @xmath37&@xmath281&@xmath106&@xmath278&@xmath244&@xmath286 + @xmath106 - corner @xmath37&@xmath283&@xmath106&@xmath278&@xmath242&@xmath287 + @xmath107 - interior @xmath37&@xmath267&@xmath39&@xmath220&@xmath279&@xmath285 + @xmath107 - interior @xmath37&@xmath280&@xmath39&@xmath220&@xmath269&@xmath270 + @xmath107 - interior @xmath37&@xmath281&@xmath39&@xmath220&@xmath268&@xmath286 + @xmath107 - interior @xmath37&@xmath283&@xmath39&@xmath220&@xmath243&@xmath287 + @xmath107 - corner @xmath37&@xmath267&@xmath107&@xmath220&@xmath273&@xmath274 + @xmath107 - corner @xmath37&@xmath280&@xmath107&@xmath220&@xmath244&@xmath272 + @xmath107 - corner @xmath37&@xmath281&@xmath107&@xmath220&@xmath242&@xmath282 + @xmath107 - corner @xmath37&@xmath283&@xmath107&@xmath220&@xmath226 & + @xmath107 - corner @xmath37&@xmath288&@xmath39&@xmath220&@xmath273&@xmath274 + @xmath107 - corner @xmath37&@xmath289&@xmath39&@xmath220&@xmath244&@xmath272 + @xmath107 - corner @xmath37&@xmath290&@xmath39&@xmath220&@xmath242&@xmath282 + in the remaining part of this appendix we show that polynomials @xmath128 and @xmath129 , appearing in the recursion relation for the overall number @xmath112 of open hws on @xmath31 , have the form given by . polynomials @xmath114 , @xmath115 and @xmath116 correspond to hws whose two step - ends are two @xmath106steps , one @xmath106 end one @xmath107step , and two @xmath107steps , respectively . since the whole hw configuration has @xmath43 steps , its , , interior part has @xmath111 steps which are either of @xmath39 or @xmath40type , implying that @xmath291 are homogeneous polynomials in @xmath45 and @xmath46 of power @xmath111 . in we list all possible arrangements of steps within open hw configuration with two one - leg end - steps , together with the corresponding numbers of used and usable junctions . using these data one can recognize arrangements with maximum possible number of @xmath40steps , and conclude that this number is equal to @xmath292 . @ccccc steps through & steps through & number of used & number of usable & @xmath234 + corner - generators & interior generators & junctions & junctions & + @xmath293&@xmath294 @xmath39 , @xmath218 @xmath40&@xmath275&@xmath273&@xmath274 + @xmath295&@xmath296 @xmath39 , @xmath218 @xmath40 , 1 @xmath106&@xmath275&@xmath269&@xmath270 + @xmath297&@xmath298 @xmath39 , @xmath218 @xmath40 , 2 @xmath106&@xmath275&@xmath244 & + @xmath299&@xmath294 @xmath39 , @xmath218 @xmath40&@xmath300&@xmath244&@xmath270 + @xmath295&@xmath296 @xmath39 , @xmath218 @xmath40 , 1 @xmath107&@xmath300&@xmath269&@xmath274 + @xmath301&@xmath296 @xmath39 , @xmath218 @xmath40 , 1 @xmath106&@xmath300&@xmath268 & + @xmath297&@xmath298 @xmath39 , @xmath218 @xmath40 , 1 @xmath106 , 1 @xmath107&@xmath300&@xmath244&@xmath270 + @xmath302&@xmath294 @xmath39 , @xmath218 @xmath40&@xmath303&@xmath242 & + @xmath301&@xmath296 @xmath39 , @xmath218 @xmath40 , 1 @xmath107&@xmath303&@xmath268&@xmath270 + @xmath297&@xmath298 @xmath39 , @xmath218 @xmath40 , 2 @xmath107&@xmath303&@xmath244&@xmath274 + if an open hw has both its ends in the same @xmath37 , then this corresponds to @xmath108 or @xmath109step within it , whereas remaining part of that hw consists of @xmath217 @xmath39-steps and @xmath218 @xmath40steps such that @xmath304 . this means that @xmath117 and @xmath118 are homogeneous polynomials in @xmath45 and @xmath46 of power @xmath110 . an @xmath109step can exist only in interior @xmath37 , in which case all steps through corner - generators are of @xmath39-type , and four junctions can never be utilized . since such hw takes @xmath300 junctions , it follows that @xmath305 . finally , @xmath108step can exist in any @xmath37 , either corner or interior one , but in both cases the number of used junctions is @xmath275 , and four junctions corresponding to corner - generators can not be taken by such hw . thus , one obtains that maximum number of @xmath40steps is @xmath306 . here we give some details of derivation of the conclusion that numbers of two - leg hw configurations are not needed for obtaining the scaling form of the overall number of open hws . first we notice that recursion relation for the two - leg hw configuration of the type @xmath108 , given in , follows from the fact that dangling ends of the two strands can reside either ( 1 ) in the same generator @xmath37 , thus forming @xmath108 or @xmath109step within it , or ( 2 ) in two different generators @xmath37 , thus forming two one - leg steps , either of @xmath106 or @xmath107type . it is not difficult to see that case ( 1 ) can be obtained by cutting a step of some @xmath39type configuration . therefore , each @xmath39type configuration with @xmath49 @xmath39steps and @xmath50 @xmath40steps gives rise to @xmath49 @xmath108type configurations with @xmath307 @xmath39steps , @xmath50 @xmath40steps and one @xmath108step , which are obtained by cutting a @xmath39step , whereas by cutting its @xmath40steps ( which can be done in two ways for each @xmath40step ) one can obtain @xmath308 different @xmath108type configurations with @xmath49 @xmath39steps , @xmath309 @xmath40steps and one @xmath109step . in a similar way , by cutting an @xmath40type hw configuration , one can obtain an @xmath109type configuration . from these observations straightforwardly follows that @xmath310 applying a reasoning similar to one used in the previous appendix , one can obtain that polynomials @xmath311 and @xmath312 , corresponding to two - leg configurations with dangling ends in different generators , have the following forms @xmath313 with @xmath314 and @xmath315 being constant positive integers , @xmath316 , and @xmath317 next , recursion relations are obtained by dividing with the recursion relation for @xmath46 , given in . therefore , coefficients @xmath158 are defined as @xmath318 and @xmath319 as @xmath320 where @xmath321 corresponds to @xmath322 , respectively . then , from and relations directly follow . for @xmath101 , @xmath66 tends to 0 , so that from one obtains @xmath323 finally , using these approximate forms while multiplying with , relation is obtained , with coefficients @xmath170 equal to @xmath324 and @xmath325 in this appendix we want to derive the formulae and for the numbers @xmath194 and @xmath326 of maximally isolated sites within the maximally entangled closed ( mec ) and maximally entangled open ( meo ) hamiltonian walks , respectively . since mec hws on the generator @xmath37 of order @xmath30 are closed hws with the maximal number of @xmath40-steps , and maximally isolated sites within mec hw are `` interior '' vertices of @xmath39-steps made through unit tetrahedrons ( _ i.e. _ not the vertices at which hw enters or exits the tetrahedron ) , number @xmath194 is equal to @xmath327 where @xmath328 is the number of unit tetrahedrons traversed by @xmath39-steps . in order to calculate @xmath328 , and consequently @xmath194 , we observe maximally entangled @xmath39-type ( meb ) hws , _ i.e. _ @xmath39-type hws with the maximal number of @xmath40-steps , and , similarly , maximally entangled @xmath40-type ( mee ) hws . for meb hw passing through @xmath31 generator , we introduce label @xmath329 for the numbers of unit tetrahedrons traversed by @xmath39-step . in a similar way , let @xmath330 be the number of unit tetrahedrons of @xmath31 , traversed by @xmath39-step , for mee hw . then , from the recursion relation for the numbers @xmath45 and @xmath46 , follows the matrix recursion relation for the numbers @xmath331 and @xmath332 : @xmath333 since @xmath102 and @xmath103 are given by , and for the unit tetrahedron one has @xmath334 and @xmath335 , solving the obtained recursion relation one obtains @xmath336 taking into account that mec hw on @xmath31 consists of @xmath53 mee hws on @xmath37 generators and @xmath337 meb hws on @xmath37 , as implied by , it follows @xmath338 and consequently @xmath339 since @xmath340 , @xmath89 , and @xmath341 , as given by , formula straightforwardly follows . in a similar manner , from , , , , , and for meo one obtains . 12 mayer j - m , guez c and dayantis j , 1990 _ b _ * 42 * 660 ; shakhnovich e and gutin a , 1990 _ j.chem.phys . _ * 93 * 5967 ; pande v s , joerg c , grosberg a yu and tanaka t , 1994 6231 ; kloczkowski a and jernigan r l , 1997 _ comput . sci . _ * 7 * 163 ; kloczkowski a and jernigan r l , 1998 _ j.chem.phys . _ * 109 * 5134 ; kloczkowski a and jernigan r l , 1998 _ j.chem.phys . _ * 109 * 5147 ; jensen i , 2001 _ comput . commun . _ * 142 * 109 ; bousquet - mlou m , guttman a j and jensen i , 2005 9159 ; peto m , sen t z , jernigan r l and kloczkowski a , 2007 _ j.chem.phys . _ * 127 * 044101 ; jacobsen j l , 2007 14667 ramakrishnan r , pekny j f and caruthers j m , 1995 _ j.chem.phys . _ * 103 * 7592 ; jaeckel a , sturm j and dayantis j , 1997 2345 ; lua r , borovinskiy a l and grosberg a yu , 2004 _ polymer _ * 45 * 717 ; mansfield m l , 2006 _ j.chem.phys . _ * 125 * 154103 ; oberdorf r , ferguson a , jacobsen j l and kondev j , 2006 _ e _ * 74 * 051801 ; jacobsen j l , 2008 118102

the detection of a gamma - ray burst ( grb ) in the solar neighborhood would have very important implications for grb phenomenology . the leading theories for cosmological grbs would not be able to explain such events . the final bursts of evaporating primordial black holes ( pbhs ) , however , would be a natural explanation for local grbs . we present a novel technique that can constrain the distance to gamma - ray bursts using detections from widely separated , non - imaging spacecraft . this method can determine the actual distance to the burst if it is local . we applied this method to constrain distances to a sample of 36 short duration grbs detected by the interplanetary network ( ipn ) that show observational properties that are expected from pbh evaporations . these bursts have minimum possible distances in the 10 cm ( 710 au ) range , consistent with the expected pbh energetics and with a possible origin in the solar neighborhood , although none of the bursts can be unambiguously demonstrated to be local . assuming these bursts are real pbh events , we estimate lower limits on the pbh burst evaporation rate in the solar neighborhood .

1512.01264

the composition of the short - duration , hard - spectrum gamma - ray burst ( grb ) population is not yet fully understood . it is believed that most of the bursts are generated in compact binary mergers @xcite and while the handful of optical counterparts and host galaxies discovered to date does not contradict this view , it is also thought that the population probably contains up to 8% extragalactic giant magnetar flares as well @xcite . for the majority of the short - duration grb population , however , there is simply not enough evidence to determine their origin unambiguously . hawking radiation from primordial black holes ( hereafter pbh ) was one of the very first explanations proposed for cosmic gamma - ray bursts @xcite , and it continues to be proposed today @xcite . the pbh lifetime and burst duration depend on its mass , so pbhs bursting today have similar masses and durations , and release similar energies , making them in essence ` standard candles ' . the typical pbh gamma - ray burst is not expected to be accompanied by detectable intrinsically - generated extended emission or have an afterglow , although accompanying bursts at other wavelengths or afterglows may arise if , for example , the pbh is embedded in a high density magnetic field or plasma @xcite . in the standard emission scenario , so - called because it uses the standard model of particle physics @xcite , the pbh gamma - ray burst is strongest in the final second of the burst lifetime , has a hard energy spectrum , and should be detectable in the vicinity of the earth . for a typical interplanetary network detector sensitive to bursts of fluence 10@xmath3 erg @xmath4 and above , pbh events could in principle be detected out to a distance of a few parsecs , depending on the emission model . pbhs evaporating today do not have enough luminosity to be detected at cosmological distances even by the most sensitive current instruments , so searching for them locally is a logical step . when observed by a single detector , the properties of a pbh burst might not appear to be significantly different from those of other short bursts ; instruments with localization or imaging capabilities would obtain their arrival directions as they would for an infinitely distant source . indeed many attempts to find evidence for the existence of pbh bursts have to date been based mainly on the spatial distribution and time histories of a subset of short bursts @xcite . other search methods have employed atmospheric cherenkov detectors @xcite , air shower detectors @xcite , radio pulse detection @xcite , spark chamber detection @xcite , and grb femtolensing @xcite . table [ comparison ] gives a comparison of these various methods . to widely spaced interplanetary network ( ipn ) detectors , however , a local pbh burst could look significantly different when compared with bursts from distant sources , due to the curvature of the received wavefront . in this paper , we use this fact to explore the possibility that some short bursts may originate in the solar neighborhood , and estimate lower limits to the pbh burst evaporation rate assuming these bursts are real pbh bursts . this paper is organized as follows . in section [ blackhole ] we derive the fluence expected in the detector from a pbh burst using the standard emission model and , as a maximal alternative , the hagedorn emission model . in section [ methodology ] we explain how we localized the detected bursts in 3d relaxing the assumption that they are at infinite distances . the detailed discussion of the methodology is given in appendix a. our data selection criteria are described in section [ data_selection ] . our results and pbh burst rate limit calculation are given in section [ results ] . in section [ discussion ] , we discuss implications and limitations of our results . as a pbh hawking - radiates , its mass is decreased by the total energy carried off by the emission , and the black hole temperature , which is inversely proportional to its mass , increases . in the standard emission model ( sem ) @xcite , the black hole directly hawking - radiates those particles which appear fundamental on the scale of the black hole . once the black hole temperature reaches the qcd transition scale ( @xmath5 mev ) , quarks and gluons are directly hawking - radiated . the pbh gamma - ray burst spectrum is the combination of the directly hawking - radiated photons and those produced by the decay of other directly hawking - radiated particles . an sem pbh with a remaining emission lifetime of @xmath6 sec has a mass of @xmath7 @xcite and a remaining rest mass energy of @xmath8 the expected fluence arriving at the detector from a pbh at a distance @xmath9 from earth is then @xmath10 where @xmath11 and @xmath12 is the fraction of the pbh energy that arrives in the energy band of the detector , and the maximum distance from which the sem pbh is detectable is @xmath13 where @xmath14 is the sensitivity of the detector . the sem analysis is consistent with high energy accelerator experiments @xcite . however , an alternative class of pbh evaporation models was proposed before the existence of quarks and gluons was confirmed in accelerator experiments and these models continue to be discussed in the pbh burst literature . in such models ( which we label hm scenarios ) , a hagedorn - type exponentially increasing number of degrees of freedom become available as radiation modes once the black hole temperature reaches a specific threshold such as the qcd transition scale . in the hm scenarios , we assume that the remaining pbh mass is emitted quasi - instantaneously as a burst of energy @xmath15 once the black hole mass reaches some threshold @xmath16 ; for the qcd transition scale , @xmath17 g and @xmath18 erg . proceeding as above , the maximum distance from which the hm pbh burst is detectable is @xmath19 where @xmath20 is the fraction of the hm pbh energy that arrives in the energy band of the detector . when a pair of ipn spacecraft detects a burst , if the distance to the source is taken to be a free parameter , the event is localized to one sheet of a hyperboloid of revolution about the axis defined by the line between the spacecraft . if the burst is assumed to be at a distance which is much greater than the interspacecraft distance , the hyperboloid intersects the celestial sphere to form the usual localization circle ( or annulus , when uncertainties are taken into account ) . another widely spaced spacecraft would produces a second hyperboloid which intersects the first one to define a locus of points which is a simple hyperbola . note that both hyperboloids have a common focus . again , if the burst is assumed to be at a large distance from the spacecraft , the hyperbola intersects the celestial sphere at two points to define two possible error boxes . a fourth , non - coplanar spacecraft even at a moderate distance from earth , such as _ konus - wind , can often be used to eliminate one branch of the hyperbola and part of the second branch . a terrestrial analogue to this method is time difference of arrival ( tdoa ) , with the important exception that grb sources can be at distances which are effectively infinite . further details are given in appendix a. while a single instrument with imaging or localization capability would obtain the correct sky position for a pbh burst regardless of its distance , the same is not true of an ipn localization , for which the derived arrival direction depends on the source distance . _ in this paper , we relax the assumption that bursts are at infinite distances . if a burst is detected by three widely spaced spacecraft , then according to the previous discussion , the possible location of the burst traces a simple hyperbola in space as illustrated in figure [ fig : figure3 ] . in an earth - centered coordinate system , this hyperbola has a closest distance to the earth , that is , a distance lower limit . as we explain in section [ ul_estimate ] , this fact can be used to calculate a lower limit to the pbh burst density rate in the solar neighborhood assuming that the bursts that we consider are actual pbh bursts . in principle , detections by three spacecraft can rule out a local origin for a burst , but it is impossible to prove a local origin with only three non - imaging spacecraft . in the case where a burst is observed by four widely spaced non - imaging spacecraft , the burst can be localized to a single point in space ( or a region in space if uncertainties are taken into consideration ) . this scenario is illustrated in figure [ fig : figure4 ] . thus in order to prove the local origin of a burst using non - imaging spacecraft , one needs detections from at least four satellites that are at interplanetary distances from each other . as mentioned in appendix a , in the special case of two widely separated spacecraft , where one spacecraft has precise imaging capability , it is in principle possible to demonstrate a local origin . we have explored this case in detail and defer treatment of it to a future paper . none of the events in this paper are in that category . the ipn database contains information on over 25,000 cosmic , solar , and magnetar events which occurred between 1990 and the present ( ` http://ssl.berkeley.edu/ipn3/index.html ` ) . during this period , a total of 18 spacecraft participated in the network . some were dedicated grb detectors , while others were primarily gamma - ray detectors with grb detection capability . indeed , the composition of the ipn changed regularly during this time , as old missions were retired and new ones were launched . however , all the instruments were sensitive to bursts with fluences around @xmath21 or peak fluxes @xmath22 and above , resulting in a roughly constant detection rate . all known bursts , regardless of their intensity or duration , or the instruments which detected them , are included in this list . we have searched it for gamma - ray bursts with the following properties . * confirmed cosmic bursts which occurred between 1990 october and 2014 december ( 24.25 y ; 10795 grbs survived this cut ) . * bursts observed by three or more spacecraft , of which two were at interplanetary distances ; 839 grbs survived this cut . this small number is due firstly to the relatively high sensitivity thresholds of the distant ipn detectors ( roughly @xmath23 ) , and their somewhat coarser time resolutions , and secondly to a 2.5 year period between 1993 august and 1996 february when there was only one interplanetary spacecraft in the network . * bursts with no x - ray or optical afterglow , either because there were no follow - up observations , or because searches were negative . in addition , as discussed before , the arrival direction derived from ipn localization depends on its assumed distance , and the bursts were initially triangulated assuming an infinite distance . thus even if a simultaneous search had taken place , it might not have identified an event within the error box if the burst was local . other selection effects come into play starting with the launch of the hete spacecraft in 2000 october , and later with the launch of swift in 2004 november , namely that x - ray and optical observations were often done rapidly , leading to more x - ray and optical detections and the elimination of the bursts from further consideration here . on the other hand , the launches of suzaku in 2005 july and fermi in 2008 june resulted in an increase in the short burst detection rate which more than compensated for the previous effect . * bursts with durations @xmath24 1 s and no extended emission ( ee ) . no cut was made based on the light curve shape ; we discuss this in section [ discussion ] . the detection rate of short bursts was roughly , but not exactly , constant over the period of this study . these cuts , which are commutative , were carried out in the order described above , so as to minimize the required analysis of the full sample of 10795 events . 36 bursts satisfied these criteria . none were detected by an imaging instrument with good spatial resolution . while we believe that the overall ipn detection rate of short bursts was roughly constant , the data selection resulted in a rate which varied significantly from year to year . in order to obtain distance lower limits as described in appendix a , these bursts were triangulated assuming that their distances from earth were free parameters . in all cases , however , an infinite distance was also compatible with the data . the results appear in table [ distances ] . of the 10795 grbs which survived the first cut , we would expect roughly 20% , or 2150 , to be short - duration events ( @xmath24 1 s ) . of those , perhaps 10% , or 215 , would display extended emission @xcite , bringing the sample to 1935 non - ee bursts . we would expect about 80% of them to have no optical counterpart , either because none was detected or none was searched for ( this number applies only to short bursts ) . this reduces the sample to about 1550 . since only 36 events survived all the data cuts , we can estimate the average ipn efficiency for this selection procedure to be about 2.3% . thus if we exclude the 2.5 year period when the ipn had a single interplanetary spacecraft , the observed event rate is 1.7/year , and the true rate is about 72/year . according the methodology described in section [ methodology ] and appendix a , we have calculated the minimum possible distances to the sample of 36 bursts selected in section [ data_selection ] . this burst sample is shown in table [ distances ] and the 12 columns give : 1 . the date of the burst , in ` yymmdd ` format , with suffix a or b where appropriate , 2 . the universal time of the burst at earth , in seconds of day , 3 . the spacecraft which were used for the triangulation ; a complete list of the spacecraft which detected the burst may be found on the ipn website ( ` http://ssl.berkeley.edu/ipn3/index.html ` ) , 4 . the burst duration , in seconds , 5 . the fluence of the burst in erg @xmath4 , 6 . the energy range over which the duration and fluence were measured , in kev , 7 . the lower limit to the burst distance , obtained by triangulation , in cm , 8 . the distance to which this burst could have been detected if it were a pbh burst of energy 10@xmath25 erg , assuming that all the energy went into the measured fluence ( this is essentially the maximum possible detectable distance ) , 9 . the maximum detectable distance assuming the sem model ( equation [ eq : semdmax ] ) in terms of the undetermined parameter @xmath26 , 10 . the maximum detectable distance assuming the hm model ( equation [ eq : hdmax ] ) in terms of the undetermined parameter @xmath27 , 11 . whether or not counterpart searches took place and if so , their references , 12 . references to the duration , peak flux , and/or fluence measurements , and/or to the localization . the shortest burst in table [ distances ] has a duration of 60 ms . due to the relatively coarse time resolutions of interplanetary detectors , bursts with shorter durations must have greater intensities to be detected , effectively setting a higher detection threshold for very short events . the weakest event has a fluence of @xmath28 . the bursts in table [ distances ] could not have come from distances less than the distance lower limits in column 7 ; however , all of them have time delays which are also consistent with infinite distances . figure [ fig : pbh_min_dist_histo ] shows a histogram of these minimum distances . the detector - dependent distance upper limits in column 8 are calculated assuming the extreme case that these events are caused by @xmath29 erg hm - type bursts from primordial black holes of mass @xmath30 g and that all of the emitted energy spectrum is contained within the detector measurement limits . table [ localizations ] gives the coordinates of the centers and corners of the error boxes for the events in table [ distances ] , assuming that the sources are at infinity . if in fact the sources are local , the arrival directions are distance - dependent , and different from the ones in table [ localizations ] . these coordinates represent the intersections of annuli , and in some cases the curvature of the annuli would make it inaccurate to construct an error box by connecting the coordinates with straight - line segments . all previous direct pbh burst searches resulted in null detections @xcite . in this case , one can derive an upper limit on the local pbh burst rate density , that is , an upper limit on the number of pbh bursts per unit volume per unit time in the local solar neighborhood . however , in our case , we have pbh burst candidates with short duration , no known afterglow detection and minimum distances that are sub - light - years . since we have pbh candidates , we should be able to derive an actual measurement of the pbh burst rate density under the assumption that the candidates are actual pbh bursts . thus , the actual pbh burst rate density is @xmath31 where @xmath32 is the number of pbh bursts , @xmath33 is the effective pbh detectable volume and @xmath34 is the observed duration . the selection efficiency of the ipn is @xmath35 . if all the candidates identified in section [ data_selection ] are real pbh bursts , then we have 36 pbh bursts , i.e. , @xmath3636 . hence , our pbh burst rate density estimate is , @xmath37 next we need to estimate the values of @xmath34 , @xmath33 , and @xmath35 . because we have studied ipn bursts collected over 21.75 years ( the 2.5 year ipn non sensitivity period is excluded ) , our observed duration is @xmath34=21.75 years . the effective pbh detectable volume , @xmath33 , calculation for this 21.75 year period , however , is not obvious . each pbh candidate has a distance consistent with some minimum distance up to infinity . we also know that pbh bursts are not bright enough to be detected from large distances . the maximum possible detectable distance of a pbh burst depends on the high - energy physics model used to calculate the final pbh burst spectrum @xcite . currently there are no accurate calculations for final pbh burst photon spectra in the kev - mev energy range . thus as a conservative maximum possible detectable pbh burst distance , we can take the maximum value of the minimum distances in our candidate pbh burst sample . this corresponds to a distance of 0.47 parsecs ( @xmath38 cm ) . because all the pbh burst candidates in the sample are actual ipn detections , this distance value is model - independent . on the other hand , it is important to note that the ipn is not capable of detecting all the bursts within this distance over the entire observation duration due to various factors such as the orientation of the satellites , and/or instrument duty cycles . thus the effective volume calculated from the above maximum possible detectable pbh burst distance is an overestimate . hence the pbh burst rate calculated in equation [ pbh_rate_36 ] is in reality a lower limit on the pbh burst rate density , @xmath39 in section [ data_selection ] , we made a rough estimate of the selection efficiency of ipn , @xmath35 , for pbh bursts . however , we note that it is very challenging to calculate @xmath35 accurately due to a number of unknown factors such as the fraction of bursts without ee , the fraction of bursts without afterglows , the fraction of bursts to which the ipn is not sensitive ( for example due to orientation or deadtime ) , etc . using the estimated effective pbh detectable volume @xmath33 , observed duration @xmath34 , and selection efficiency @xmath35 , we can now estimate the lower limit of the pbh burst rate in the best case scenario where all the pbh burst candidates are actual pbh bursts . in this case , our pbh burst lower limit is @xmath40158.5 bursts @xmath41 . if we assumed 100% efficiency then the pbh burst lower limit is @xmath40 3.6 bursts @xmath41 . if only one of the candidates is an actual pbh burst then the value of the rate density lower limit depends on the minimum distance to that particular burst . if the burst with the largest minimum distance ( grb 140807 ) is the pbh burst , then the pbh burst rate density lower limit is @xmath40 0.1 bursts @xmath41 . if the burst with the smallest minimum distance ( grb 970902 ) is real then the pbh burst rate density lower limit is @xmath42 bursts @xmath41 ( this value is excluded by other high - energy experiments , however ) . all these estimates assume that pbhs are distributed uniformly in the solar neighborhood . the ipn pbh burst rate density lower limit values are shown in figure [ fig : pbh - limit - final ] . pbh burst upper limits from various other searches are also shown in the figure . in the worst case scenario where none of our candidates is a real pbh burst , we can not estimate a lower limit to the pbh burst rate density and instead consider to estimate an upper limit to the pbh burst rate density . however , the assumption that none of the bursts in the sample is real but still we have candidates implies that our criteria to identify pbh bursts defined in section [ data_selection ] is not sufficient . this means our method is not capable of setting an upper limit on the pbh burst rate density . the detection of gamma - ray transients points to very high energy explosive phenomena in the universe . their detection in the solar neighborhood would indicate a previously unrecognized and potentially exotic phenomenon in our cosmic backyard . the sample of bursts identified in this paper are candidates for such explosions . they have short durations , no known afterglow detections , and have distance limits consistent with the solar neighborhood . in principle our methodology is capable of proving the local origin of bursts . however in order to do that we need either four widely separated non - imaging spacecraft or two spacecraft that include one with imaging capability ( see section [ methodology ] and appendix a ) . this is not the case for any of the bursts we considered in our study . while some events were indeed detected by four or more spacecraft , the spacecraft were not widely separated , i.e. at interplanetary distances . with four widely separated non - imaging spacecraft detections or detections by one imaging spacecraft and one non - imaging spacecraft , it would be possible to prove that some of these bursts are in the solar neighborhood and this would definitely point to an exotic origin for these bursts . lacking that however , we can look at other properties of these bursts and discuss how likely it is that they may have a pbh origin . firstly , it is of interest to investigate the sky distribution of our pbh burst candidates . for example , @xcite have argued that , due to the fact that very short duration grbs ( i.e. , grbs with duration @xmath43 100 msec ) have a non - isotropic sky distribution , they may be drawn from a different grb population , possibly from pbh bursts . in order to investigate this we have calculated burst density maps using the gaussian kernel density methodology described in @xcite . since the sky locations of the pbh candidates depends on their distance from earth , we started by assuming all the bursts are at their minimum distances and calculated their sky density map . this map is shown in figure [ sky_map_pbh_candidate_min ] . the map is presented in galactic coordinates with a 25 degree smoothing radius . it shows some relatively high density areas , but the probability of generating this density contrast by chance , in the case when the true sky distribution is uniform , is @xmath400.2 , estimated using a monte carlo simulation @xcite . thus the density structure seen in figure [ sky_map_pbh_candidate_min ] is consistent with a uniform source distribution . if these pbh candidates are real , they can not be further away than @xmath4010 parsecs ( which is the maximum possible distance they can be detected assuming the optimistic hagedorn - type model ) . we then also calculated the sky locations of the pbh candidates assuming they are at 10 parsecs and derived the sky density map as shown in figure [ sky_map_pbh_candidate_10pc ] . this map is also consistent with a uniform source distribution . this is the behaviour one would expect if these pbh burst are local with maximum detectable distance @xmath4010 parsecs . according to the standard model for hawking radiation @xcite , pbh bursts are standard candles , that is , all pbh bursts are intrinsically identical at the source . however , the way the burst appears at large distances may vary depending on its host environment . the final burst properties of the pbh burst depend on its mass and the number of fundamental particle degrees of freedom available at various energies @xcite . in principle , by measuring the photon flux arriving from a pbh burst candidate in a given energy range , we can calculate the distance to that burst . however , during the last second of the pbh lifetime , its temperature is well above @xmath40 1 tev and the physics governing these high energies is not fully understood . thus , the full spectrum of a pbh burst is difficult to calculate . @xcite have calculated pbh burst light curves in the 50 gev 100 tev energy range using the standard model of particle physics . this calculation considered both the direct hawking radiation of photons from the black hole and the photons created due to the fragmentation and hadronization of the directly hawking - radiated quarks and gluons . the expected pbh light curve is a power - law with an index of @xmath44 and at various sub - energy bands within the 50 gev 100 tev energy range has an interesting inflection point due to the brief dominance of the directly hawking - radiated photons around @xmath45 second before the pbh expiration . this is potentially a unique signature of a pbh burst in the gev / tev energy range . the detectors in the ipn are sensitive to photons in the energy range 10 kev 100 mev . there are no published calculations of pbh burst light curves in the kev / mev range using the standard model of particle physics to which we can fit our light curves and extract fit parameters . nonetheless it is of interest to look at the light curves of our pbh burst candidates . firstly , we note that , although a pbh will be emitting in the kev / mev range before it becomes hot enough to emit in the gev / tev range , a kev / mev burst signal can only be generated by the low energy component of a pbh that is also emitting in the gev / tev range : this is because a 10 tev black hole has a remaining burst lifetime of @xmath401 s whereas a @xmath241 gev black hole has a lifetime @xmath46 yr and observationally would be a stable source not a burst . this low energy photon component , in turn , is predominantly generated by other higher energy hawking - radiated species via decays or the inner bremsstrahlung effect @xcite and is not the directly hawking - radiated photon flux which decreases in the kev / mev band as the burst progresses . acknowledging the uncertainty in the pbh light curves in the kev / mev range , it is also possible that the pbh burst signal may have a longer or shorter duration in the kev / mev range than in the gev / tev range due to differences both in production at the source and in detector sensitivity , and that the duration difference varies with the distance to the pbh . figure [ lc_grid ] shows the light curves of the 36 ipn bursts in our sample . some are clearly single - peaked , others are clearly multi - peaked , and some were not recorded with sufficient statistics to determine the true number of peaks . it is interesting to note that bursts such as grb 970921 , grb 080222 , grb000607 , grb 101009 , grb 121127 , grb 131126a , and grb 141011a display a kev / mev time profile that resembles the pbh light curve profile calculated by @xcite for the gev / tev energy range . in this paper , we introduced a novel method to constrain the distances to grbs using detections from multiple spacecraft . utilizing detections from three non - imaging spacecraft we could only constrain the minimum distances to our current sample of bursts . the maximum distance is constrained by the energy available during the final second of the pbh burst . however , the amount of energy released in the kev / mev energy band is not known and may be highly model - dependent . on the other hand , with detections by four widely separated ( @xmath40 au distances ) non - imaging spacecraft or one non - imaging spacecraft and one imaging spacecraft , we can constrain burst distances independent of any high energy physics model , and potentially show that some bursts are local . such a detection will not only prove the existence of pbh bursts , by fitting light curves and spectra derived using various beyond the standard model physics theories , we can also identify which theory describes nature . support for the ipn was provided by nasa grants nnx09au03 g , nnx10au34 g , nnx11ap96 g , and nnx13ap09 g ( _ fermi ) ; nng04gm50 g , nng06ge69 g , nnx07aq22 g , nnx08ac90 g , nnx08ax95 g , and nnx09ar28 g ( integral ) ; nnx08an23 g , nnx09ao97 g , and nnx12ad68g(_swift ) ; nnx06ai36 g , nnx08ab84 g , nnx08az85 g , nnx09av61 g , and nnx10ar12 g ( _ suzaku ) ; nnx07ar71 g ( messenger ) ; nag5 - 3500 , and jpl contracts 1282043 and y503559 ( _ odyssey ) ; nnx12ae41 g , nnx13ai54 g , and nnx15ae60 g ( ada ) ; nnx07ah52 g ( konus ) ; nag5 - 13080 ( rhessi ) ; nag5 - 7766 , nag5 - 9126 , and nag5 - 10710 , ( _ bepposax ) ; and nng06gi89 g . tnu acknowledges support from the laboratory directed research and development program at the los alamos national laboratory ( lanl ) . the konus - wind experiment is partially supported by a russian space agency contract and rfbr grants 15 - 02 - 00532 and 13 - 02 - 12017-ofi - m . we would also like to thank jim linnemann ( msu ) , dan stump ( msu ) , brenda dingus ( lanl ) , and pat harding ( lanl ) for useful conversations on the analysis . assume two spacecraft , @xmath47 and @xmath48 , separated by a distance @xmath9 , observe a grb . for any assumed distance between the grb and the spacecraft , the difference in arrival times @xmath49t@xmath50 must be constant . let @xmath51 , @xmath52 , @xmath53 and @xmath54 , @xmath55 , @xmath56 be the coordinates of the two spacecraft . then the locus of points @xmath57 which describes the possible source locations is given by consider first the two - dimensional problem for simplicity . let @xmath47 and @xmath48 define the z - axis of a coordinate system whose origin is halfway between the spacecraft . the positions of @xmath47 and @xmath48 are the foci of the hyperbola : this is shown in figure [ fig : hyperbola ] . here @xmath61 is the difference between the distances of any point on the hyperbola from the foci , so @xmath62 , and @xmath63 . for every point on this hyperbola , the difference in the arrival times is @xmath64 . if we assume an infinite distance for the source , the asymptotes of the hyperbola define the two possible arrival directions of the grb . in practice , we will have two or more hyperboloids generated by three or more spacecraft , and we will want to work in earth - centered cartesian coordinates with one axis oriented towards right ascension zero , declination zero , and another axis oriented towards declination 90@xmath66 . consider the three - spacecraft case . a spacecraft pair will define two foci of a hyperboloid ; the line joining the two spacecraft , which defines the axis of rotation of the hyperboloid , will be oriented with respect to earth - centered coordinates such that it represents a rotation and a translation . we want to express the formula for the hyperboloid in the earth - centered system . here the primed coordinate system is the one defined by the foci of the hyperboloid ; its origin is the same as that of the unprimed , earth - centered system , and it is rotated , but not translated , with respect to it . now perform a translation of the primed system so that its origin is at the midpoint of the two foci . if the coordinates of the two spacecraft , expressed in the unprimed system , are @xmath51 , @xmath52 , @xmath53 and @xmath54 , @xmath55 , @xmath56 , the origin of the translated system will be at ( @xmath51 + @xmath54)/2 , ( @xmath52 + @xmath55)/2 , ( @xmath53 + @xmath56)/2 . the formula for the hyperboloid , expressed in the earth - centered system , becomes where @xmath69 and @xmath70 refer to the hyperboloid for spacecraft 1 and 2 . a similar equation describes the hyperboloid for spacecraft 1 and 3 . although a third equation can be derived for spacecraft 2 and 3 , it is not independent of the other two , because it is constrained by the condition @xmath64 + @xmath71 + @xmath72 = 0 . the locus of points describing the intersection of two hyperboloids is a simple hyperbola , contained in a plane . this is shown in figure [ fig : figure3 ] . this hyperbola contains all the points satisfying the time delays for the two spacecraft pairs , @xmath64 and @xmath71 , when the grb distance is allowed to vary . the two branches of the hyperbola intersect the celestial sphere at two points ; if the distance is taken to be infinite , the two points are the possible source locations . it follows that a grb observed by three , and only three , widely separated non - imaging spacecraft , can not be unambiguously proven to originate at a local distance ; on the other hand , in the case where the hyperbola degenerates to a single point , that point must be at an infinite distance , and a local origin can be ruled out . none of the bursts in this sample were in this category . the lower limit to the source distance is the point on the hyperbola ( @xmath57 ) which is closest to earth . this can be found by solving for the minimum of the expression @xmath74 ( the earth distance ) subject to the constraints imposed by equations a1 and a2 . in practice there are uncertainties associated with @xmath64 and @xmath71 , and we have used the most probable values to derive the lower limits . since @xmath75 , @xmath76 , and @xmath77 vary along the hyperbola , the apparent arrival direction for an observer at earth depends on the assumed distance ; if the source distance is assumed to be infinite , the derived right ascension and declination will not be correct if the source is actually local . for example , if grb 140807 were at its minimum allowable distance ( @xmath78 cm , or @xmath79 au ) , the angle between its true coordinates and the coordinates for an infinitely distant source would be @xmath80 . but for grb 101129 , whose minimum allowable distance is only @xmath81 cm , or 2.6 au , the angle would be 54.2 . eliminates one branch of the hyperbola and part of the second branch , leading to a larger distance lower limit . thus , in the case of four widely separated spacecraft , it is in principle possible to rule out an infinite distance and prove that the origin is local as illustrated in figure [ fig : figure4 ] . however , this was not the case for any of the bursts in this study ; they are all consistent with both local and infinite distances . note that this method does not depend on the properties of the grb itself , such as duration or intensity ; the lower limit is determined by the ipn configuration ( through the spacecraft coordinates ) and the direction of the burst ( through the time delays ) . thus , for example , it can be applied equally to long- and short - duration bursts , and a sample of long - duration events would yield a distribution of lower limits which was comparable to a sample of short - duration events . one special case should be noted here . with just two widely separated spacecraft , if one has precise imaging capability , the problem reduces to finding the intersection of a hyperboloid and the vector defined by the precise localization from the imager . in principle , a local origin can be demonstrated , or a distance lower limit can be obtained . we have studied approximately two hundred grbs which are in this category , and analysis of the results is ongoing . abdo , a. a. , abeysekara , a. u. , alfaro , r. , et al . 2015 , astroparticle physics , 64 , 4 d. e. alexandreas , g. e. allen , d. berley , et al . 1993 , physical review letters 71 , 2524 m. amenomori , z. cao , b. z. dai , et al . 1995 , proc . international cosmic ray conference 2 , 112 barnacka , a. , glicenstein , j .- f . , and moderski , r. 2012 86 , 043001 bhat , p. , gopalakrishnan , n. , gupta , s. , ramana murthy , p. , sreekantan , b. , and tonwar , s. 1980 284 , 433 cline , d. & hong , w. 1996 , astroparticle physics 5 , 175 czerny , b. , et al . 2011 , new astronomy 16 , 33 cline , d. , et al . 1997 , 486 , 169 cline , d. , et al . 1999 , 527 , 827 cline , d. , et al . 2003 , astroparticle physics 18 , 531 cline , d. , et al . 2005 , 633 , l73 czerny , b. , et al . 2011 , new astronomy 16 , 33 cline , d. & hong , w. 1992 , 401 , l57 eichler , d. , livio , m. , piran , t. , and schramm , d. 1989 , 340 , 126 fegan , d. , mcbreen , b. , obrien , d. , and osullivan , c. 1978 , 271 , 731 fichtel , c. , bertsch , d. , dingus , b. , et al . 1994 , 434 , 557 frontera , f. , guidorzi , c. , montanari , e. , et al . 2009 , 180 , 192 j. glicenstein , a. barnacka , m. vivier , et al . 2013 , arxiv:1307.4898 goldstein , a. , burgess , j. m. , preece , r. , et al . 2012 199 , 19 goldstein , a. , preece , r. , mallozzi , r. , et al . 2013 208 , 21 golenetskii , s. , aptekar , r. , mazets , e. , et al . 2011a , gcn circ . 12110 golenetskii , , s. , aptekar , r. , frederiks , d. , et al . 2011b , gcn circ . 12111 golenetskii , s. , aptekar , r. , frederiks , d. , et al . 2011c , gcn circ . 12249 golenetskii , s. , aptekar , r. , frederiks , d. , et al . 2011d , gcn circ . 12271 golenetskii , s. , aptekar , r. , mazets , e. , et al . , 2012a gcn circ . 13313 golenetskii , s. , aptekar , r. , frederiks , d. , et al . 2012b , gcn circ . 13315 golenetskii , s. , aptekar , r. , mazets , e. , et al . 2012c , gcn circ . 13620 golenetskii , s. , aptekar , r. , frederiks , d. , et al . 2012d , gcn circ . 13621 golenetskii , s. , aptekar , r. , mazets , e. , et al . 2012e , gcn circ . 13627 golenetskii , s. , aptekar , r. , mazets , e. , et al . 2012f , gcn circ . 14021 golenetskii , s. , aptekar , r. , frederiks , d. et al . 2012 g , gcn circ . 14022 golenetskii , s. , aptekar , r. , mazets , e. , et al . 2013a , gcn circ . 14561 golenetskii , s. , aptekar , r. , frederiks , d. , et al . 2013b , gcn circ . 14565 golenetskii , s. , aptekar , r. , palshin , v. , et al . 2013c , gcn circ . 15550 golenetskii , s. , aptekar , r. , frederiks , d. , et al . 2013d , gcn circ . 15551 golenetskii , s. , aptekar , r. , mazets , e. , et al . 2014a , gcn circ . 16801 golenetskii , s. , aptekar , r. , frederiks , d. , et al . 2014b , gcn circ . 16807 golenetskii , s. , aptekar , r. , palshin , v. , et al . 2014c , gcn circ . 16906 golenetskii , s. , aptekar , r. , frederiks , d. , et al . 2014d , gcn circ . 16907 guiriec , s. , briggs , m. , connaughton , v. , et al . 2010 , 725 , 225 hawking , s. 1974 , 248 , 30 hurley , k. , berger , e. , castro - tirado , a. , et al . 2002 , 567,447 hurley , k. , rowlinson , a. , bellm , e. , et al . 2010 , 403 , 342 hurley , k. , goldsten , j. , mitrofanov , i. , et al . 2011a , gcn circ . 12247 hurley , k. , goldsten , j. , mitrofanov , i. , et al . 2011b , gcn circ . 12269 ishida , y. , tashiro , m. , terada , y. , et al . 2012 , gcn circ . 14044 jelley , j. , baird , g. , and omongain , e. 1977 , 267 , 499 keane , e. , stappers , b. , kramer , m. , and lyne , a. 2012 , 425 , l71 e. t. linton , r. w. atkins , h. m. badran , et al . 2006 , jcap 1 , 13 macgibbon , j. h. and webber , b. r. 1990 41 , 3052 macgibbon , j. h. , carr , b. j. , & page , d. n. 2008 , 78 , 064043 masetti , n. , palazzi , e. , pian , e. , et al . 2000 , gcn circ . 720 mazets , e. , aptekar , r. , cline , t. , et al . 2008 , 680 , 545 paciesas , w. , meegan , c. , pendleton , g. , et al . 1999 122 , 465 paciesas , w. , meegan , c. , von kienlin , a. , et al . 2012 199 , 18 pagani , c. , and evans , p. a. 2014 , gcn circ . 17134 page , d. & hawking , s. 1976 , 206 , 1 page , d. 1976 , 13 , 198 page , d. n. , carr , b. j. , & macgibbon , j. h. 2008 , , 78 , 064044 palshin , v. , hurley , k. , svinkin , d. , et al . 2013 , 207 , 38 park , h. , williams , g. , perez , d. , et al . 2000 , gcn circ . 873 pelassa , v. , and meegan , c. 2013 , gcn circ . 15573 phinney , s. , and taylor , j. 1979 277 , 117 porter , n. , and weekes , t. 1977 212 , 224 porter , n. , and weekes , t. 1978 183 , 205 porter , n. , and weekes , t. 1979 277 , 199 price , p. , axelrod , t. , and schmidt , b. 2000 , gcn circ . 898 rees , m. 1977 , 266 , 333 sakamoto , a. , tashiro , m. , terada , y. , et al . 2012 , gcn circ . 13350 singer , l. , kasliwal , m. , and cenko , s. 2013 , gcn circ . 15572 singer , l. , private communication 2014 svinkin , d. , hurley , k. , aptekar , r. , et al . 2015a , 447 , 1028 svinkin , d. , et al . 2015b , in preparation terekhov , o. , denisenko , d. , lobachev , v. , et al . 1994 , astron . 20(3 ) , 265 g. tesic and veritas collaboration 2012 , journal of physics conference series 375 052024 ukwatta , t. n. , & woniak , p. r. 2016 , , 455 , 703 ukwatta , t. n. , stump , d. , macgibbon , j. h. , et al . 2015 , arxiv:1510.04372 von kienlin , a. , meegan , c. , paciesas , w. , et al . 2014 211 , 13 von kienlin , a. 2013 , gcn circ . 14560 von kienlin , a. 2014 , gcn circ . 16905 vreeswijk , p. , and rol , e. 2000 , gcn circ . 908 xiong , s. meegan c. 2012 , gcn circ . 13644 yasuda , t. , terada , y. , tashiro , m. , et al . 2011 , gcn circ . 12114 yasuda , t. , tashiro , y. , terada , y. , et al . 2013 , gcn circ . 14600 rotated to obtain one sheet of a hyperboloid of rotation of two sheets . the two spacecraft are aligned along the z - axis , and are the foci of the hyperboloids . each hyperboloid defines the loci of possible source distances in the three - dimensional problem . if the distance is assumed to be infinite , the circle defined by the intersection of the hyperboloid with the celestial sphere gives the possible grb arrival directions . [ fig : hyperboloid],scaledwidth=80.0% ] . this map is normalized to represent a probability density function ( pdf ) that integrates to 1 over the entire sphere . the smoothing parameter is taken to be 25 degrees . the circles indicate the locations of the individual bursts . the maximum and the minimum density values in this map are 0.166 and 0.041 , respectively . the probability of generating this density contrast by chance in the case when the true sky distribution is uniform , is @xmath400.2 estimated using a monte carlo simulation @xcite . thus the density structures seen in the map are consistent with a uniform distribution . [ sky_map_pbh_candidate_min],scaledwidth=90.0% ] atmospheric cherenkov & @xmath83 & 200 mev-10 tev & @xmath84 & 1 + air shower & @xmath85 & @xmath86 ev & @xmath87 & 2 + radio pulse & @xmath88 & 430 - 1374 mhz & @xmath89 & 3 + spark chamber & @xmath90 & 20 mev-1 gev & @xmath91 & 4 + spatial distribution & @xmath92 & 15 kev-10 mev & & 5 + 910206 & 31529 & uly , pvo , phe & 0.7 & @xmath93 & 100100000 & @xmath94 & @xmath95 & @xmath96 & @xmath97 & no & 1,6 + 920414 & 84162 & uly , bat , pvo & 0.96 & @xmath98 & 20 2000 & @xmath99 & @xmath100 & @xmath101 & @xmath102 & no & 2,3,6 + 970902 & 27561 & uly , kon , near & 0.44 & @xmath103 & 10 10000 & @xmath104 & @xmath105 & @xmath101 & @xmath106 & no & 4,5 + 970921 & 83828 & uly , kon , near & 0.06 & @xmath107 & 10 10000 & @xmath108 & @xmath109 & @xmath110 & @xmath111 & no & 4,5 + 981102 & 28554 & uly , kon , near & 0.57 & @xmath93 & 10 10000 & @xmath112 & @xmath95 & @xmath113 & @xmath114 & no & 5,13 + 990712 & 27919 & uly , bat , kon , near & 0.62 & @xmath115 & 10 10000 & @xmath116 & @xmath117 & @xmath118 & @xmath119 & no & 5,8 + 000607 & 08689 & uly , kon , near & 0.09 & @xmath120 & 10 10000 & @xmath121 & @xmath105 & @xmath122 & @xmath123 & 14,15 & 5,13 + 001025 & 71369 & uly , kon , near & 0.48 & @xmath124 & 10 10000 & @xmath125 & @xmath105 & @xmath126 & @xmath127 & 15,16 & 5,13 + 001204 & 28869 & uly , kon , sax , near & 0.27 & @xmath128 & 10 10000 & @xmath129 & @xmath130 & @xmath131 & @xmath132 & 15,17,18 & 5,7,13 + 010104 & 62490 & uly , kon , sax , near & 0.76 & @xmath133 & 10 10000 & @xmath134 & @xmath135 & @xmath136 & @xmath137 & no & 5,7,13 + 010119 & 37177 & uly , kon , near & 0.18 & @xmath138 & 10 10000 & @xmath139 & @xmath119 & @xmath140 & @xmath141 & no & 5,7,13 + 080203 & 08456 & ody , kon , int , mes & 0.4 & @xmath142 & 10 10000 & @xmath140 & @xmath143 & @xmath144 & @xmath145 & no & 5,13 + 080222 & 37262 & ody , kon , mes & 0.11 & @xmath146 & 10 10000 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & no & 5,13 + 080530 & 58296 & ody , int , mes , suz & 0.41 & @xmath151 & 1001000 & @xmath152 & @xmath153 & @xmath154 & @xmath155 & no & 6 + 090228 & 17600 & ody , kon , rhe , mes & 0.08 & @xmath156 & 10 10000 & @xmath157 & @xmath158 & @xmath159 & @xmath160 & no & 5,6,9,10,19 + 090523 & 34075 & ody , kon , mes , suz & 0.5 & @xmath161 & 10 10000 & @xmath118 & @xmath162 & @xmath163 & @xmath164 & no & 5,6,13 + 100223 & 09491 & ody , kon , mes , suz & 0.19 & @xmath165 & 10 10000 & @xmath166 & @xmath162 & @xmath167 & @xmath168 & no & 5,6,9,10 + 100629 & 69243 & ody , kon , int , mes & 0.46 & @xmath128 & 10 10000 & @xmath169 & @xmath97 & @xmath154 & @xmath170 & no & 5,6,9,10 + 100811 & 09349 & ody , kon , int , mes & 0.33 & @xmath171 & 10 10000 & @xmath172 & @xmath109 & @xmath126 & @xmath173 & no & 5,6,9,11 + 101009 & 24858 & ody , kon , int , mes & 0.18 & @xmath161 & 10 10000 & @xmath174 & @xmath109 & @xmath167 & @xmath164 & no & 5,6,13 + 101129 & 56371 & ody , kon , int , mes & 0.32 & @xmath175 & 10 10000 & @xmath176 & @xmath130 & @xmath101 & @xmath177 & no & 5,9,11 + 110510 & 80844 & ody , kon , swi , mes & 0.84 & @xmath178 & 10 10000 & @xmath179 & @xmath180 & @xmath181 & @xmath182 & no & 6,13 + 110705 & 13031 & ody , kon , int , mes & 0.22 & @xmath183 & 10 10000 & @xmath184 & @xmath185 & @xmath186 & @xmath102 & no & 9,11,20,21,22 + 110802 & 55157 & ody , kon , int , mes & 0.54 & @xmath187 & 10 10000 & @xmath188 & @xmath189 & @xmath190 & @xmath191 & no & 13,23,24 + 120222 & 01776 & ody , kon , mes , fer & 0.87 & @xmath192 & 10 10000 & @xmath193 & @xmath148 & @xmath194 & @xmath195 & no & 9,11 + 120519 & 62294 & ody , kon , mes , fer & 0.91 & @xmath196 & 10 10000 & @xmath197 & @xmath198 & @xmath149 & @xmath177 & no & 9,11,25,26,27 + 120811 & 01230 & ody , kon , mes , fer & 0.31 & @xmath199 & 10 10000 & @xmath200 & @xmath119 & @xmath126 & @xmath173 & no & 9,11,28,29,30,31 + 121023 & 27857 & ody , mes , fer & 0.51 & @xmath201 & 10 1000 & @xmath202 & @xmath203 & @xmath204 & @xmath205 & no & 9,11 + 121127 & 78960 & ody , kon , int , mes & 0.53 & @xmath206 & 10 10000 & @xmath207 & @xmath123 & @xmath208 & @xmath209 & no & 11,12,32,33,34 + 130501 & 00831 & ody , kon , mes , suz & 0.41 & @xmath210 & 10 10000 & @xmath211 & @xmath212 & @xmath213 & @xmath214 & no & 13 + 130504b & 27123 & ody , kon , mes & 0.36 & @xmath215 & 10 10000 & @xmath216 & @xmath217 & @xmath96 & @xmath130 & no & 35,36,37,38 + 131126a & 14050 & ody , kon , mes , fer & 0.12 & @xmath138 & 10 10000 & @xmath218 & @xmath212 & @xmath219 & @xmath141 & 39 & 40,41,42 + 140414b & 80735 & ody , kon , int , mes & 0.97 & @xmath138 & 10 10000 & @xmath220 & @xmath212 & @xmath221 & @xmath222 & no & 13 + 140807 & 43173 & ody , kon , mes , fer & 0.57 & @xmath223 & 10 10000 & @xmath224 & @xmath198 & @xmath167 & @xmath225 & 43 & 13 + 140906c & 85869 & ody , kon , int , mes & 0.13 & @xmath138 & 10 10000 & @xmath101 & @xmath119 & @xmath140 & @xmath141 & no & 44,45 + 141011a & 24380 & ody , kon , mes , fer & 0.06 & @xmath226 & 10 10000 & @xmath227 & @xmath228 & @xmath149 & @xmath229 & no & 46,47,48 + 910206 & & + center & 87.5650 & 17.3372 + corners & 86.7457 & 17.8617 + & 88.4371 & 16.5674 + & 86.6713 & 18.2592 + & 88.3683 & 16.9025 + or & & + center & 86.9621 & 36.1678 + corners & 86.0419 & 35.5941 + & 87.9516 & 36.9783 + & 85.9902 & 35.1926 + & 87.8875 & 36.6407 + 920414 & & + center & 90.5234 & -76.3115 + corners & 90.6025 & -76.3287 + & 90.3189 & -76.2150 + & 90.7309 & -76.4079 + & 90.4446 & -76.2943 + 970902 & & + center & 351.0259 & + 7.3294 + corners & 351.0744 & + 7.3080 + & 351.1009 & + 7.3100 + & 350.9774 & + 7.3507 + & 350.9509 & + 7.3486 + 970921 & & + center & 235.8826 & -24.8310 + corners & 236.0899 & -24.3906 + & 236.2034 & -24.1627 + & 235.6864 & -25.2436 + & 235.5867 & -25.4367 + 990712 & & + center & 123.4627 & + 6.6755 + corners & 123.8544 & + 7.2952 + & 123.2666 & + 6.3351 + & 123.2036 & + 6.2494 + & 123.7301 & + 7.1201 + or & & + center & 125.3550 & + 9.4408 + corners & 124.9180 & + 8.8500 + & 125.6024 & + 9.7478 + & 125.6597 & + 9.8378 + & 125.0364 & + 9.0294 + 000607 & & + center & 38.4971 & + 17.1419 + corners & 38.4656 & + 17.1016 + & 38.5472 & + 17.2317 + & 38.5287 & + 17.1823 + & 38.4470 & + 17.0523 + 001025 & & + center & 275.3488 & -5.1067 + corners & 275.4272 & -5.1411 + & 275.1851 & -4.9912 + & 275.2707 & -5.0723 + & 275.5140 & -5.2229 + 001204 & & + center & 40.2997 & + 12.8817 + corners & 40.2796 & + 12.8928 + & 40.3478 & + 12.8979 + & 40.3199 & + 12.8707 + & 40.2516 & + 12.8656 + 010104 & & + center & 317.3689 & + 63.5116 + corners & 317.6201 & + 63.5227 + & 317.5851 & + 63.4776 + & 317.1183 & + 63.5004 + & 317.1523 & + 63.5455 + 010119 & & + center & 283.4446 & + 11.9964 + corners & 283.4892 & + 12.0007 + & 283.4159 & + 11.9832 + & 283.4000 & + 11.9921 + & 283.4734 & + 12.0096 + 080203 & & + center & 48.2086 & + 24.3289 + corners & 48.1704 & + 24.3216 + & 48.2401 & + 24.0161 + & 49.1541 & + 20.8021 + & 48.9850 & + 21.5848 + & 48.2469 & + 24.3362 + & 47.7229 & + 27.1306 + & 47.5349 & + 27.9115 + & 48.1002 & + 24.6272 + 080222 & & + center & 68.7192 & + 56.9390 + corners & 69.2723 & + 57.0738 + & 68.1039 & + 56.6937 + & 68.1686 & + 56.7979 + & 69.3407 & + 57.1754 + 080530 & & + center & 5.3443 & 28.2246 + corners & 5.2701 & 28.3257 + & 5.4170 & 28.1570 + & 5.2717 & 28.2920 + & 5.4186 & 28.1232 + 090228 & & + center & 98.7014 & -28.7863 + corners & 98.7827 & -28.8531 + & 98.5307 & -28.6112 + & 98.5698 & -28.6709 + & 98.8219 & -28.9123 + 090523 & & + center & 22.8383 & -62.2047 + corners & 23.0738 & -62.1859 + & 23.2792 & -62.2169 + & 22.6023 & -62.2230 + & 22.3971 & -62.1907 + 100223 & & + center & 101.6711 & + 12.8302 + corners & 101.9024 & + 12.6261 + & 101.4091 & + 13.4941 + & 101.5176 & + 12.9034 + & 102.0085 & + 12.0774 + 100629 & & + center & 227.6190 & + 29.5479 + corners & 227.6295 & + 29.2934 + & 227.5705 & + 29.7415 + & 227.6081 & + 29.8012 + & 227.6671 & + 29.3536 + 100811 & & + center & 344.9605 & + 20.6223 + corners & 344.9395 & + 20.5411 + & 345.0289 & + 20.6525 + & 344.9816 & + 20.7032 + & 344.8922 & + 20.5919 + 101009 & & + center & 75.6068 & -33.9325 + corners & 75.8367 & -33.8146 + & 76.0209 & -33.6624 + & 75.1904 & -34.1986 + & 75.3765 & -34.0492 + 101129 & & + center & 268.6211 & -7.7232 + corners & 272.2014 & -9.1377 + & 265.1168 & -6.9176 + & 272.2546 & -9.3668 + & 265.1792 & -7.0998 + 110510 & & + center & 336.1407 & -44.0726 + corners & 344.4202 & -52.3621 + & 344.5801 & -52.3041 + & 327.7177 & -35.8320 + & 327.8448 & -35.7920 + 110705 & & + center & 156.0322 & 40.1169 + corners & 156.1434 & 40.1805 + & 156.0390 & 39.9001 + & 156.0246 & 40.3325 + & 155.9211 & 40.0532 + 110802 & & + center & 44.4655 & 33.0037 + corners & 44.4647 & 33.6264 + & 44.4285 & 32.4871 + & 44.5087 & 33.5009 + & 44.4754 & 32.3503 + 120222 & & + center & 296.9268 & 14.7102 + corners & 290.7915 & -5.6861 + & 291.1995 & -2.6858 + & 304.9807 & 31.3395 + & 307.1167 & 33.7741 + 120519 & & + center & 178.0457 & 21.9477 + corners & 178.2141 & 22.2660 + & 178.2885 & 22.1813 + & 177.8043 & 21.7128 + & 177.8781 & 21.6273 + 120811 & & + center & 43.7599 & -31.9123 + corners & 43.8831 & -31.9728 + & 43.7269 & -32.0579 + & 43.7928 & -31.7664 + & 43.6370 & -31.8516 + 121023 & & + center & 313.3540 & -4.6025 + corners & 315.3258 & -8.9120 + & 311.5238 & -0.2314 + & 315.1761 & -8.9571 + & 311.3903 & -0.3093 + 121127 & & + center & 176.4314 & -52.4316 + corners & 176.2965 & -52.4618 + & 176.4549 & -52.2522 + & 176.4080 & -52.6106 + & 176.5661 & -52.4011 + 130501 & & + center & 350.5102 & 16.7266 + corners & 350.4295 & 16.8223 + & 350.4890 & 15.7404 + & 350.5512 & 17.6829 + & 350.5910 & 16.6309 + 130504b & & + center & 347.1594 & -3.8325 + corners & 347.0950 & -3.8347 + & 349.6913 & -9.6653 + & 345.9151 & -0.2257 + & 347.2238 & -3.8303 + 131126a & & + center & 202.8995 & 51.5581 + corners & 202.9690 & 51.6343 + & 203.1321 & 51.5592 + & 202.6668 & 51.5564 + & 202.8303 & 51.4819 + 140414b & & + center & 121.8042 & -38.1211 + corners & 121.2343 & -35.5319 + & 121.0519 & -33.7849 + & 122.7335 & -42.2323 + & 122.4453 & -40.6251 + 140807 & & + center & 188.8547 & 36.2259 + corners & 188.9062 & 36.3820 + & 188.7683 & 36.3184 + & 188.9406 & 36.1333 + & 188.8030 & 36.0695 + 140906c & & + center & 314.9144 & 2.0606 + corners & 314.8105 & 2.5705 + & 314.9946 & 2.0906 + & 314.8341 & 2.0305 + & 315.0226 & 1.5373 + 141011a & & + center & 257.9473 & -9.6520 + corners & 258.0072 & -9.3553 + & 257.7749 & -9.3571 + & 258.1189 & -9.9487 + & 257.8862 & -9.9489 +

we consider singular solutions of the biharmonic nls . in the-critical case , the blowup rate is bounded by a quartic - root power law , the solution approaches a self - similar profile , and a finite amount of-norm , which is no less than the critical power , concentrates into the singularity ( `` strong collapse '' ) . in the-critical and supercritical cases , we use asymptotic analysis and numerical simulations to characterize singular solutions with a peak - type self - similar collapsing core . in the critical case , the blowup rate is slightly faster than a quartic - root , and the self - similar profile is given by the standing - wave ground - state . in the supercritical case , the blowup rate is exactly a quartic - root , and the self - similar profile is a zero - hamiltonian solution of a nonlinear eigenvalue problem . these findings are verified numerically ( up to focusing levels of ) using an adaptive grid method . we also calculate the ground states of the standing - wave equations and the critical power for collapse in two and three dimensions .

0912.1233

the focusing nonlinear schrdinger equation ( nls ) @xmath2 where @xmath3 , and @xmath4 is the laplacian , has been the subject of intense study , due to its role in various areas of physics , such as nonlinear optics and bose - einstein condensates ( bec ) . it is well - known that the nls possesses solutions that become singular in a finite time @xcite . of special interest is the critical ( @xmath5 ) nls @xmath6 which models the collapse of intense laser beams that propagate in a bulk kerr medium . in this study , we consider the focusing biharmonic nonlinear schrdinger equation ( bnls ) @xmath7 where @xmath8 is the biharmonic operator . equation admits waveguide ( standing - wave ) solutions of the form @xmath9 , where @xmath10 satisfies the `` standing - wave '' equation @xmath11 the bnls is called `` _ @xmath0-critical _ '' , or simply `` critical '' if @xmath12 . in this case , the @xmath0 norm ( `` _ power _ '' ) is conserved under the bnls dilation symmetry @xmath13 . the critical bnls can be rewritten as @xmath14 correspondingly , the bnls with @xmath15 is called subcritical , and the bnls with @xmath16 is called supercritical . this is analogous to the nls , where the critical case is @xmath17 . in @xcite , ben - artzi , koch and saut proved that the bnls is locally well - posed in @xmath18 , when @xmath19 is in the _ @xmath18-subcritical _ regime @xmath20 global existence and scattering of bnls solutions in the @xmath18-critical case @xmath21 were studied by miao , xu and zhao @xcite and by pausader @xcite . the latter work also showed well - posedness for small data . the @xmath18-critical defocusing bnls was studied by miao , xu and zhao @xcite and by pausader @xcite . the above studies focused on non - singular solutions . in this work , we study singular solutions of the bnls in @xmath18 , i.e. , solutions that exist in @xmath22 over some finite time interval @xmath23 , but for which @xmath24 the first study of singular bnls solutions was done by fibich , ilan and papanicolau @xcite , who proved the following results : [ thrm : ge_subcritical ] solutions of the subcritical ( @xmath15 ) focusing bnls exist globally . [ thrm : ge_critical ] let @xmath25 , where @xmath26 , and @xmath10 is the ground state of with @xmath27 . then , the solution of the critical focusing bnls exists globally . the simulations in @xcite suggested that there exist singular solutions for @xmath12 and @xmath16 , and that these singularities are of the blowup type , namely , the solution becomes infinitely localized . however , in contradistinction with nls theory , there is currently no rigorous proof that solutions of the bnls can become singular in either the critical or the supercritical case . to the best of our knowledge , the only work , apart from @xcite , which considered singular solutions of the bnls is by chae , hong and lee @xcite , who proved that if singular solutions of the critical bnls exist , then they have a power - concentration property . see section [ ssec : power_conc ] for more details . in this work , we consider singular solutions of the focusing bnls in the @xmath18-subcritical regime . our purpose is to characterize these singular solutions : their profile , blowup rate , power concentration , et cetera . in some cases , we assume radial symmetry , i.e. , that @xmath28 where @xmath29 in these cases , equation reduces to @xmath30 where @xmath31 is the radial biharmonic operator . specifically , the critical bnls reduces to @xmath32 the paper is organized as follows : in section [ sec : invariance ] , we use noether theorem to derive conservation laws for the bnls . we recall that in the critical nls , the conservation law which follows from invariance of the action integral under dilation leads to the nls `` variance identity '' , which can be used to prove the existence of singular solutions . in section [ ssec : varid ] we use a similar procedure to derive the `` variance identity '' for the critical bnls , and then generalize it to the supercritical bnls . however , since it is not clear that the `` bnls variance '' is positive definite , this identity does not lead to a proof of the existence of singular solutions . the ground states of the bnls standing - wave equation were previously computed only in the one - dimensional case @xcite , since they were computed using a shooting method , which can not be easily generalized to multi - dimensions . in section [ sec : stationary ] we use the spectral renormalization method to compute the ground - states of the critical bnls for one , two and three dimensions . the calculated ground - states provide the first numerical estimate of the critical power for collapse @xmath33 in the two - dimensional and three - dimensional cases , see section [ sec : critical_power ] . direct simulations of the critical bnls suggest that the constant @xmath34 in theorem [ thrm : ge_critical ] is optimal . in section [ sec : rigorous ] we use rigorous analysis to study the critical bnls . the blowup rate is shown to be lower - bounded by a quartic root , i.e. , @xmath35 the corresponding bound for the critical nls is a square root , i.e. , @xmath36 we then prove that singular solutions converge to a self - similar profile strongly in @xmath37 , for any @xmath19 in the @xmath18-subcritical regime . this implies that the singular solutions have the power - concentration property , whereby the amount of power that enters the singularity point is at least @xmath34 . these rigorous results mirror those of the critical nls . let us denote the location of the maximal amplitude of a radially - symmetric solution by @xmath38 singular solutions are called `` _ peak - type _ '' when @xmath39 for @xmath40 , and `` _ ring - type _ '' when @xmath41 for @xmath42 . in this work , we use asymptotic analysis and numerics to find and characterize peak - type singular solutions of the bnls equation . ring - type singular solutions of the bnls will be studied elsewhere @xcite . in section [ sec : crit_peak ] we use asymptotics and numerics to show that peak - type singular solutions of the critical bnls collapse with the quasi self - similar profile @xmath43 where the self - similar profile is the ground state @xmath44 . the blowup rate is shown to be slightly faster than the quartic - root bound . this is analogous to the critical nls , where the blowup rate of peak - type solutions is slightly faster than the square - root bound , due to the loglog correction ( the `` loglog law '' ) . it is an open question whether the correction to the bnls blowup rate is also a loglog term . in section [ sec : supercrit_peak ] we consider peak - type singular solutions of the supercritical bnls . in this case , asymptotics and numerics show that singular solutions are of the quasi self - similar form @xmath45 the blowup rate of @xmath46 is exactly a quartic - root , and the self - similar profile @xmath47 is different from the ground - state . rather , as in the supercritical nls , the self - similar profile @xmath48 is the zero - hamiltonian solution of a nonlinear eigenvalue problem . although @xmath47 is not in @xmath0 , it can be the self - similar profile of a collapsing @xmath18 solution , since the collapsing solution is `` only '' quasi self - similar . section [ sec : num_meth ] presents the numerical methods . the computations of singular bnls solutions that focus by factors of @xmath1 and more , necessitated the usage of adaptive grids . we develop a modified version of the static grid redistribution method @xcite , which is more convenient for the biharmonic problem . calculating the bnls standing waves in multi - dimensions is done using the spectral renormalization method . in this study , we use rigorous theory , asymptotic theory and numerics to analyze singular solutions of the bnls . all the results presented in this work mirror those of the nls , `` up to a change by a factor of @xmath49 '' in the blowup rate ( @xmath50 ) , in the critical value of @xmath19 ( @xmath51 ) , et cetera . however , several key features of nls theory are still missing from bnls theory . first , the `` variance identity '' for the bnls can not be used to prove that singular solutions exist . second , the critical nls is invariant under the pseudo - conformal ( `` lens - transformation '' ) symmetry which can also be used to construct explicit singular solutions . at this time , it is unknown whether an analogous identity for the critical bnls exists . third , in critical nls theory , the self - similar profile is known to possess a quadratic radial phase term , i.e. , @xmath52 this term represents the focusing of the solution towards @xmath53 , and plays a key role in the rigorous and asymptotic theory of the critical nls . at this time , we do not know the analogous radial phase term for the critical bnls . finally , we note that a similar `` up to a factor of @xmath49 '' connection exists between singular solutions of the nonlinear heat equation , see @xcite , and the biharmonic nonlinear heat equation , see @xcite . for example , the @xmath54 norm of singular solutions blows up as @xmath55 for the nonlinear heat equation , and as @xmath56 for the biharmonic heat equation . the `` similarity up to a factor of @xmath49 '' , however , is not perfect . for example , the self - similar spatial variable is @xmath57 for the nonlinear heat equation , and @xmath58 { { { t_{\rm c}}}-t } $ ] for the biharmonic nonlinear heat equation . another difference between the equations is that singular solutions are _ asymptotically _ self - similar for the nonlinear heat equation , and truly self - similar for the biharmonic heat equation . in contrast , the nls possesses self - similar singular solutions , whereas for the bnls it is unknown whether singular solutions are truly , or only asymptotically , self - similar . the bnls is the euler - lagrange equation of the action integral @xmath59 where @xmath60 is the lagrangian density @xmath61 therefore , the conserved quantities of the bnls can be found using noether theorem , see appendix [ app : noether ] . as in the standard nls , invariance of the action integral under phase - multiplications @xmath62 implies conservation of the `` power '' ( @xmath0 norm ) , i.e , . @xmath63 similarly , invariance under temporal translations @xmath64 implies conservation of the hamiltonian @xmath65 = { \left\vert{\delta\psi}\right\vert}_2 ^ 2 -\frac{1}{\sigma+1 } { \left\vert{\psi}\right\vert}_{2(\sigma+1)}^{2(\sigma+1)},\ ] ] and invariance under spatial translations @xmath66 implies conservation of the linear momentum , i.e. , @xmath67 in the critical case @xmath68 , the action integral is also invariant under the dilation transformation @xmath69 the corresponding conserved quantity is @xmath70 we recall that the action integral of the critical nls is invariant under the dilation transformation @xmath71 the corresponding conserved quantity is @xmath72 in addition , the integral term @xmath73 is the time - derivative of the variance , i.e. , @xmath74 therefore , it follows that @xmath75 in the supercritical nls , the second derivative of the variance is not related to this conservation law . nevertheless , direct differentiation shows that @xmath76 therefore @xmath77 since @xmath78 , the variance identity shows that solutions of the critical and supercritical nls , whose hamiltonian is negative , become singular in a finite time @xcite . we next extend the analogy between and to the non - critical case . in the case of the bnls , direct differentiation shows that @xmath79 therefore , if we define @xmath80 where @xmath81 is a positive constant , we get the bnls variance identity @xmath82 in order to use this identity to prove singularity formation , however , one must show that @xmath83 , as defined by , has to remain positive . direct integration by parts gives that @xmath84 while the second term on the rhs is a temporal derivative of a positive - definite quantity , the first term on the rhs is not . therefore , it remains an open question whether @xmath83 has to be positive . the bnls equation admits the standing - wave solutions @xmath85 the equation for the standing - wave profile is @xmath86 where @xmath87 . for example , in one dimension , equation is given by @xmath88 and in two dimensions by @xmath89 [ eqs : radial_stationary_state ] if we impose radial symmetry , eq . reduces to @xmath90 where @xmath91 is given in . at @xmath53 , all the odd derivatives vanish , and so the solution of is subject to the boundary conditions @xmath92 the solution of eq . was computed numerically in the one - dimensional case in @xcite as follows . in the @xmath93 case , can be integrated once , yielding an explicit relation between @xmath94 and @xmath95 . this parameter - reduction enables the usage of a one - parameter shooting approach . unfortunately , such a parameter reduction is not possible in higher dimensions . therefore , in multi - dimensions we compute the ground - states using the spectral renormalization method ( srm ) , which was introduced by petviashvili in @xcite , and more recently by albowitz and musslimani in @xcite . see section [ ssec : srm ] for further details . in figure [ fig : srm ] we display the results in the critical case , i.e. , the ground states of @xmath96 figure [ fig : srm_1d ] displays the ground - state of eq . in the critical @xmath93 case , as calculated by the srm . the solution is in excellent agreement with the solution computed in @xcite using the shooting method . figure [ fig : srm_2d ] and figure [ fig : srm_3d ] display the ground state in the critical @xmath97 and @xmath98 cases . we note that , while the srm method that we use does not enforce radial symmetry , the calculated ground states for @xmath99 and @xmath100 are radially symmetric ( data not shown ) . as noted in @xcite , the ground - states of the bnls are non - monotonic in @xmath101 and change their sign , in contradistinction with the ground - states of the nls which are monotonically - decreasing and strictly positive . theorem [ thrm : ge_critical ] shows that the critical power for collapse in the critical bnls is @xmath102 , when @xmath10 is the ground - state of . the computation of @xmath10 , see section [ sec : stationary ] , allows for the numerical calculation of the critical power @xmath34 . the case @xmath103 was found in @xcite to be @xmath104 using the calculated ground state in the two - dimensional case , see figure [ fig : srm_2d ] , we now calculate the critical power in the two - dimensional case , giving @xmath105 similarly , using the calculated ground state in the @xmath98 case , see figure [ fig : srm_3d ] , gives @xmath106 we now ask whether theorem [ thrm : ge_critical ] is sharp , in the sense that for any @xmath107 , there exists an initial condition @xmath108 such that @xmath109 and the corresponding solution of the critical bnls becomes singular . as noted , at present there is no proof that solutions of the bnls can become singular . therefore , in particular , it is unknown whether theorem [ thrm : ge_critical ] is indeed sharp . hence , we will explore this issue numerically . we recall that in the critical nls , the necessary condition for collapse @xmath110 is sharp , since for any @xmath107 the initial condition @xmath111 becomes singular in a finite time @xcite . therefore , we now check numerically whether for @xmath112 , the solution of the critical bnls with the initial condition @xmath113 becomes singular . to do this , we solve the one - dimensional and two - dimensional critical bnls equations with the initial condition @xmath114 , see fig [ fig : peak_ric ] . in both cases , the solution appears to blow up , suggesting that theorem [ thrm : ge_critical ] is sharp . see section [ ssec : sgr ] for the numerical methodology used for solving the bnls . we also note that in the critical nls , if the initial condition is different from the ground state , the power input required for collapse is strictly larger than @xmath115 , see @xcite . this the case also in the bnls . for example , in the one - dimensional critical bnls with a gaussian shaped initial condition @xmath116 , the input power required for collapse is strictly larger than @xmath117 , see @xcite . in the two - dimensional critical bnls with a gaussian shaped initial condition , the input power required is strictly larger than @xmath118 , see @xcite . in @xcite , cazanave and weissler proved that the blowup rate for singular solutions of the critical nls is not slower than a square - root , i.e. , that @xmath119 . the analogous result for the critical bnls is as follows : [ thrm : low - bound ] let @xmath120 be a solution of the critical bnls ( [ eq : cbnls ] ) that becomes singular at @xmath121 , and let @xmath122 . then , @xmath123 such that @xmath124 we follow the proof given by merle @xcite for the critical nls . for a fixed @xmath125 , @xmath126 , let us define @xmath127 then , @xmath128 is defined for @xmath129 , and satisfies the bnls equation @xmath130 since @xmath131 this implies that @xmath132 , i.e. , that @xmath133 becomes singular as @xmath134 . in addition , [ eqs : psi1_0_bounds ] @xmath135 from the definition of @xmath128 and power conservation it follows that @xmath136 using equations ( [ eq : delta_psi1_bound ] ) and ( [ eq : psi1_bound ] ) and the cauchy - schwartz inequality gives @xmath137 together , the three formulae ( [ eqs : psi1_0_bounds ] ) imply that for any fixed @xmath138 @xmath139 in other words , for each @xmath125 , the initial @xmath18 norm of @xmath128 is bounded by a function of @xmath140 . specifically , this bound is independent of @xmath125 . from the local existence theory @xcite , @xmath128 exists in @xmath141 $ ] , where @xmath142 therefore , it follows from that @xmath143 depends on @xmath144 , but is independent of @xmath125 . since @xmath128 blows up at @xmath145 we have that @xmath146 from which the result follows . in @xcite , weinstein showed that the collapsing core of all singular solutions of the critical nls approaches a self - similar profile . we now prove the analogous result for the critical bnls : [ thrm : self - similarity ] let @xmath147 and let @xmath148 be a solution of the radially - symmetric critical bnls ( [ eq : cbnls ] ) with initial conditions @xmath149 , that becomes singular at @xmath121 . let @xmath150 and let @xmath151 then , for any sequence @xmath152 there is a subsequence @xmath153 such that @xmath154 strongly in @xmath155 , for all @xmath156 such that , where @xmath19 is in the @xmath18-subcritical regime .]@xmath157 in addition , @xmath158 , where @xmath10 is the ground state of equation . let @xmath159 and define @xmath160 from the definition of @xmath161 it follows that @xmath162 = { { \mathit l } } ^4h[\psi(t_k)].\ ] ] therefore , using cauchy - schwartz , @xmath163 since @xmath164 is bounded , it follows that there exists a subsequence of @xmath161 which converges weakly in @xmath18 to a function @xmath165 . from the compactness lemma [ lem : compactness ] , see appendix [ app : compactness ] , it follows that @xmath166 strongly in @xmath155 , for all @xmath156 given by . next , we prove that @xmath167\leq0 $ ] . since @xmath168 , it follows that @xmath169 , and so @xmath170 . additionally , since @xmath171 for some @xmath172 , we have that @xmath173 and so @xmath174 \leq \lim_{k\to\infty } h[\phi_k ] = \lim_{k\to\infty } { { \mathit l } } ^4h[\psi_0 ] = 0.\ ] ] in addition , since @xmath175 = \lim_{k\to\infty } \left ( 1-\frac{1}{\sigma+1}{\left\vert{\phi_k}\right\vert}_{2(\sigma+1)}^{2(\sigma+1 ) } \right),\ ] ] it follows that @xmath176 so @xmath177 . therefore , corollary [ cor : p2h_bound ] , see appendix [ app : gagli - niren ] , implies that @xmath178 . solutions of critical nls have the power concentration property , whereby the amount of power that collapses into the singularity is at least @xmath179 , see @xcite . in what follows , we prove the analogous results for the critical bnls . [ cor : power - concentration ] let @xmath147 , and let @xmath148 be a solution of the radially - symmetric critical bnls ( [ eq : radial_cbnls ] ) that becomes singular at @xmath121 . then , @xmath180 , @xmath181 where @xmath182 the result shall follow directly from corollary [ cor : power - conc - lower - bound ] . the following corollary shows that the rate of power - concentration is not slower than the blowup rate @xmath183 . the nls analogue is due to tsutsumi @xcite and weinstein @xcite . [ cor : power - conc - lower - bound ] let @xmath147 , let @xmath148 be a solution of the radially - symmetric critical bnls ( [ eq : radial_cbnls ] ) that becomes singular at @xmath121 , and let @xmath150 . then , 1 . for any monotonically - decreasing function @xmath184 such that @xmath185 we have that @xmath186 2 . for any @xmath187 , @xmath188 such that @xmath189 see appendix [ app : proof - power - conc - lower - bound ] . since the second part of corollary [ cor : power - conc - lower - bound ] is true for all @xmath187 and since @xmath190 , corollary [ cor : power - concentration ] follows . the next corollary shows that the power - concentration rate has a quartic - root upper bound . the analogue in nls theory , which is a square - root upper bound , was proved in @xcite . [ cor : power - conc - upper - bound ] let @xmath147 and let @xmath148 be a solution of the radially - symmetric critical bnls ( [ eq : radial_cbnls ] ) that becomes singular at @xmath121 . then , 1 . for any monotonically - decreasing @xmath191 such that @xmath192 we have that @xmath186 2 . for any @xmath187 , @xmath188 such that @xmath193 theorem [ thrm : low - bound ] implies that @xmath194 . therefore , the result follows immediately from corollary [ cor : power - conc - lower - bound ] . recently , chae , hong and lee @xcite used the harmonic analysis method of bourgain @xcite to prove that singular solutions of the critical bnls for @xmath195 have the power - concentration property @xmath196 where @xmath197 is a positive constant . this result is more general than corollaries [ cor : power - concentration],[cor : power - conc - lower - bound],[cor : power - conc - upper - bound ] in that it does not assume radial symmetry . the proof given here , however , is considerably simpler . more importantly , it shows that @xmath198 . in this , we consider radially - symmetric singular solutions that are `` _ peak - type _ '' , i.e. , for which @xmath39 for @xmath40 , where @xmath199 is the location of the maximal amplitude . the critical nls admits singular solutions that collapse with the universal @xmath200 profile , i.e. , @xmath201 , where @xmath202 the self - similar profile @xmath203 is the ground - state solution of @xmath204 the blowup rate of @xmath46 is given by the _ loglog law _ @xcite @xmath205 since the blowup rate is slightly faster than a square root @xmath206 therefore , the phase term @xmath207 in vanishes as @xmath208 . hence , the blowup profile reduces to @xmath209 we now look for the `` corresponding '' peak - type singular solutions of the critical bnls . theorem [ thrm : self - similarity ] suggests that the collapsing core of the singular solution approaches a self - similar form , i.e. , @xmath210 where @xmath211 and @xmath212 . substituting into and requiring that @xmath213 \sim \left [ \delta\psi \right ] \sim \left [ |\psi|^{d/2}\psi \right ] $ ] suggests that @xmath214 let us consider the self - similar profile @xmath47 . in the singular region @xmath215 we have that @xmath216 and @xmath217 hence , @xmath47 satisfies @xmath218 theorem [ thrm : low - bound ] shows that that blowup rate of @xmath46 is lower - bounded by a quartic - root . in the critical nls the blowup rate of peak - type solutions is slightly faster than the analogous square - root rate , due to the loglog correction . hence , we expect that the blowup rate of peak - type critical bnls solutions is slightly faster than a quartic root , i.e. , @xmath219{{{t_{\rm c}}}-t } } \to 0.\ ] ] in that case , @xmath220 and reduces to the standing - wave equation . since the ground - states of attain their maximal amplitudes at @xmath221 , see section [ sec : stationary ] , these are peak - type solutions . the above informal analysis thus leads to the following conjecture : [ conj : crit_peak_rate_profile ] the critical bnls admits peak - type singular solutions such that : 1 . the collapsing core approaches the self - similar profile + [ eqs : crit_peak_qss ] @xmath222 where @xmath223 + and @xmath10 is the ground - state of equation . the blowup rate of @xmath46 is slightly faster than a quartic - root , i.e. , @xmath224 in section [ ssec : crit_peak_simulations ] we provide numerical evidence in support of conjecture [ conj : crit_peak_rate_profile ] . the one - dimensional critical bnls @xmath225 was solved with the gaussian initial condition @xmath226 with @xmath227 , whose power is @xmath228 . the maximal amplitude of the solution @xmath229 as a function of time is plotted in fig [ fig : peak_amp_ls_1d ] . the amplitude increases abruptly by a factor of @xmath230 around @xmath231 , suggesting that the solution becomes singular in a finite time . the simulation was repeated for the radially - symmetric two - dimensional critical bnls @xmath232 with the gaussian initial condition @xmath233 with @xmath234 , whose power is @xmath235 the amplitude increases abruptly by a factor of @xmath1 . around @xmath236 , see fig [ fig : peak_amp_ls_2d ] , again suggesting that the solution becomes singular in a finite time . we next consider the self - similar profile of the collapsing solutions from figure [ fig : peak_amp_ls ] . in order to verify that it is given by , we rescale the solutions as @xmath237 with @xmath238 . the rescaled solutions at focusing levels of @xmath239 and @xmath240 are indistinguishable , see fig [ fig : crit_peak_rescaled ] , indicating that the collapsing core is indeed self - similar according to . a predicted , the self - similar profile is very close to the ground - state @xmath10 in the core region @xmath241 . we next compute the blowup rate @xmath242 , defined by the relation@xmath243 to do that , we perform a least - squares fit of @xmath244 with @xmath245 , see figure [ fig : crit_peak_powerlaw ] , obtaining a value of @xmath246 for both @xmath103 and @xmath99 . this value of @xmath242 is slightly above @xmath247 , implying that the quartic - root lower - bound given by theorem [ thrm : low - bound ] is close to the actual blowup - rate of peak - type singular solutions . next , we provide two indications that the blowup rate is faster than @xmath247 . first , if the blowup rate is exactly @xmath247 , then @xmath248 should be finite and strictly negative . however , up to focusing level of @xmath240 , @xmath249 does not appear to converge to a negative constant , but rather to increase slowly towards @xmath250 , see figure [ fig : crit_peak_l3lt ] . second , according to the informal analysis in section [ ssec : crit_peak_analysis ] , the blowup rate is faster than a quartic root if and only if the self - similar profile @xmath47 satisfies the standing - wave equation , which is indeed what we observed numerically in figure [ fig : crit_peak_rescaled ] . * remark : * in the critical nls the blowup rate of peak - type solutions is slightly faster than the analogous square - root lower - bound , due to the well - known loglog - correction . figure [ fig : crit_peak_powerlaw ] shows that the blowup rate is slightly faster than a quartic root , and figure [ fig : crit_peak_l3lt ] shows that @xmath251 very slowly . together , this suggests that the blowup rate in the critical bnls is only slightly faster than the analogous quartic root . at present , we do not know if the blowup rate of peak - type solutions of the critical bnls is a quartic root with a @xmath252 correction . we note , however , that the loglog correction in the critical nls can not be determined numerically @xcite , and can only be derived analytically . therefore , we expect that the determination of the analogous correction to the @xmath247 blowup rate of the critical bnls will also have to be done analytically , and not numerically . in contrast to the extensive theory on singularity formation in the critical nls , much less is known about the supercritical nls . numerical simulations and formal calculations ( see , e.g. , ( * ? ? ? * chapter 7 ) and the references therein ) suggest that peak - type singular solutions of the supercritical nls collapse with a universal @xmath253 profile , i.e. , [ eq : intro_psiq ] @xmath254 , \\ \psi_\text{non - singular}(t , r)\quad & r\ge r_c , \end{cases}\ ] ] where @xmath255 note that the singular region @xmath256 $ ] is constant in the coordinate @xmath101 . therefore , in the rescaled variable @xmath257 , the singular region @xmath258 $ ] becomes infinite as @xmath259 . this is in contradistinction with the critical case , wherein the singular region @xmath260 $ ] is constant in the rescaled variable @xmath261 , but shrinks to a point in the original coordinate @xmath101 . the self - similar profile @xmath262 is the solution of @xmath263 solutions of ( [ eq : ode4q ] ) are complex - valued , and depend on the parameter @xmath264 and on the initial condition @xmath265 . solutions of ( [ eq : ode4q ] ) whose amplitude @xmath266 is monotonically - decreasing in @xmath261 , and which have a zero hamiltonian , are called _ admissible solutions _ @xcite . for each choice of @xmath267 , equation has a unique admissible solution ( up to a multiplication by a constant phase @xmath268 ) . this solution is attained for specific real values of @xmath264 and @xmath269 , which we denote as @xmath270 the blowup rate of @xmath46 is a square - root , i.e. , @xmath271 numerical simulations and formal calculations suggest that : 1 . the self - similar profile of singular peak - type solutions of the nls is an admissible solution of . since @xmath272 attains its maximal amplitude at @xmath221 , the solution is peak - type . the constant @xmath264 of the blowup rate is equal to @xmath273 . hence , in particular , @xmath264 is universal ( i.e. , is independent of the initial condition @xmath274 ) . the admissible solution @xmath272 satisfies @xmath275 thus , @xmath276 . nevertheless , @xmath272 can be the self - similar profile of @xmath277 solutions , since @xmath278 only for @xmath279 $ ] , see @xcite . as in the supercritical nls , we expect that singular peak - type solutions of the supercritical bnls collapse as @xmath280 where @xmath281 is a self - similar profile , to be determined . as in the supercritical nls , the singular region @xmath256 $ ] , is constant in the coordinate @xmath101 . therefore , in the rescaled variable @xmath257 , the singular region @xmath258 $ ] becomes infinite as @xmath259 . this is again in contradistinction with the critical - bnls case , where the singular region @xmath260 $ ] is constant in the rescaled variable @xmath261 , but shrinks to a point in the original coordinate @xmath101 . the bnls is invariant under the dilation symmetry @xmath282 , where @xmath283 is a constant . in the supercritical case @xmath16 , this suggests that @xmath284 similar arguments as in section [ ssec : crit_peak_analysis ] show that @xmath285 therefore , as in the supercritical nls , we expect the collapsing part of the solution to approach the self - similar profile , hence the self - similar profile reduces to . ] @xmath286 theorem [ thrm : low - bound ] showed that in the critical case , if @xmath287 , then @xmath288 . the following lemma extends this result to peak - type solutions of the supercritical bnls . [ lem : supercrit_peak_rate_profile ] let @xmath16 , and let @xmath120 be a peak - type singular solution of the bnls that collapses with the @xmath281 profile . if @xmath289 , then @xmath290 . furthermore , @xmath291 if and only if the self - similar profile @xmath47 satisfies the equation @xmath292 if @xmath293 , then @xmath294 and @xmath295 hence , the equation for @xmath48 is @xmath296 implying that @xmath297 should be bounded as @xmath298 . since @xmath299 , it follows that @xmath300 . if @xmath291 , then @xmath301 , and equation reduces to . let us consider the fourth - order nonlinear ode for the self - similar profile @xmath47 . its solution requires four boundary conditions , and the determination of the parameter @xmath264 . radial symmetry implies that @xmath302 . since the solution is invariant up to rescaling @xmath303 , one can set , with no loss of generality , @xmath304 . therefore , we require two additional constraints in order to determine @xmath264 and @xmath305 . we recall that in the supercritical nls the `` admissible value '' of @xmath264 is determined from the requirements that @xmath262 has a zero hamiltonian and a monotonically - decreasing amplitude , see section [ ssec : supercrit_peak_nls_review ] . for the bnls , the following informal argument suggests that the zero - hamiltonian condition should also holds . indeed , from hamiltonian conservation it follows that @xmath306 $ ] is bounded , because otherwise the non - singular region would also have an infinite hamiltonian . calculating the hamiltonian of @xmath281 , we have @xmath307 \sim l^{-4/\sigma+d-4 } \left [ \int_{\rho=0}^{r_c / l(t ) } \left ( { \left\vert { \delta_\rho b(\rho ) } \right\vert}^2 -\frac{1}{1+\sigma}{\left\vert{b}\right\vert}^{2 + 2\sigma } \right ) \rho^{d-1 } d\rho \right].\ ] ] from @xmath18-subcriticality , see , it follows that @xmath308 as @xmath309 . therefore , if @xmath310 $ ] remains bounded as @xmath208 then @xmath311 = \int_{\rho=0}^{\infty } \left ( { \left\vert { \delta_\rho b } \right\vert}^2 -\frac1{1+\sigma } { \left\vert { b } \right\vert}^{2 + 2\sigma } \right ) \rho^{d-1}d\rho = 0.\ ] ] wkb analysis of shows that , see appendix [ app : wkb ] , @xmath312 where @xmath313 and @xmath314 . equation therefore has two algebraically - decaying solutions , @xmath315 and @xmath316 , an exponentially - increasing solution @xmath317 , and an exponentially - decreasing solution @xmath318 . since @xmath16 , the exponent @xmath319 of @xmath316 is larger that the exponent @xmath320 of @xmath315 , hence @xmath321 as @xmath322 . [ lem : admissible_b23 ] let @xmath47 be a zero - hamiltonian solution of . then , @xmath323 and @xmath324 furthermore , @xmath325 . the exponentially increasing solution @xmath317 must vanish identically if the integrals are to converge , hence @xmath326 . next , the @xmath327 phase term in @xmath316 implies that @xmath328 hence , the integral @xmath329 diverges in the @xmath18-subcritical regime @xmath330 , i.e. , @xmath331 . since , in addition , @xmath332 the hamiltonian can be finite only if @xmath333 , in which case @xmath325 . lemma [ lem : admissible_b23 ] shows that the condition @xmath334=0 $ ] imposes the two constraints @xmath333 and @xmath326 . hence , zero - hamiltonian solutions of equation satisfy the five boundary conditions @xmath335 therefore they form a discrete set of solutions and of values of @xmath264 . we conjecture that the additional condition of monotonicity of @xmath336 will lead to a unique admissible solution @xmath48 and a unique value of @xmath264 . [ cor : admissible_b23_power ] let @xmath47 be a zero - hamiltonian solution of . then , @xmath337 . nevertheless , @xmath338 . since @xmath339 , @xmath340 following the arguments of @xcite , the profile @xmath281 satisfies @xmath341 in summary , we conjecture the following : [ conj : supercrit_peak_rate_profile ] + let @xmath120 be peak - type singular solution of the supercritical bnls . then , [ eqs : supercrit_peak_qss ] 1 . the collapsing core approaches the self - similar profile @xmath281 , i.e. , @xmath342 where @xmath343 2 . the self - similar profile @xmath47 is the solution of @xmath344=0 , \end{gathered}\ ] ] where @xmath345 and @xmath334 $ ] is the hamiltonian of @xmath48 , see . 3 . in particular , @xmath346 . 4 . equation has a unique `` admissible solution '' with a unique `` admissible value '' of @xmath347 , such that @xmath348 is monotonically decreasing . additionally , @xmath349 as @xmath322 . the admissible solution is the self - similar profile @xmath48 of @xmath281 , see . the blowup rate of singular peak - type solutions is exactly a quartic root , i.e.,@xmath350{{{t_{\rm c}}}-t } , \qquad \kappa>0.\ ] ] 7 . the coefficient @xmath264 of the blowup rate of @xmath46 is equal to the value of @xmath264 of the admissible solution @xmath48 , i.e. , @xmath351{{{t_{\rm c}}}-t } } = \kappa(\sigma , d).\ ] ] in particular , @xmath264 is universal ( i.e. , it does not depend on the initial condition ) . in section [ ssec : supercrit_peak_simulations ] we provide numerical evidence in support of conjecture [ conj : supercrit_peak_rate_profile ] . [ cols="^,^,^,^,^,^",options="header " , ] the radially - symmetric bnls was solved in the supercritical case @xmath352 with the initial condition @xmath353 , and in the supercritical case @xmath354 with the initial condition @xmath355 . in both cases , the solutions blowup at a finite time , see figure [ fig : supercrit_peak_amp_ls ] . to check whether the solutions collapse with the self - similar profile , the solution was rescaled according to . the rescaled solutions at the focusing levels @xmath239 and @xmath240 are indistinguishable in both the one - dimensional case ( figure [ fig : supercrit_peak_rescaled_1d ] ) and the two - dimensional case ( figure [ fig : supercrit_peak_rescaled_2d ] ) , providing numerical support that the solution collapses with the @xmath281 profile . as predicted , the self - similar profile is different than the ground - state @xmath10 . indeed , figure [ fig : supercrit_peak_rescaled_far ] shows that as @xmath356 , the self - similar profile of @xmath120 decays as @xmath357 , which is in agreement with the decay rate of @xmath358 . we next verify that the solution converges to the asymptotic profile for @xmath256 $ ] , i.e. , for @xmath258 $ ] . to do this , we plot in figure [ fig : supercrit_peak_convergence ] the rescaled solution at focusing levels of @xmath359 , as a function of @xmath360 . the curves are indistinguishable at @xmath361 , but bifurcate at increasing values of @xmath362 . these `` bifurcations positions '' are marked by circles in figure [ fig : supercrit_peak_convergence ] , and their @xmath362 values are listed in table [ tab : supercrit_peak_convergence ] . the `` bifurcation positions '' are linear in @xmath363 , indicating that the region where @xmath293 is indeed @xmath258 $ ] , which corresponds to @xmath256 $ ] . to compute the blowup rate @xmath242 , we performed a least - squares fit of @xmath244 with @xmath245 , see figure [ fig : supercrit_peak_powerlaw ] . the resulting values are @xmath364 in the @xmath352 case and @xmath365 in the @xmath354 case . next , we provide two indications that the blowup rate is exactly @xmath247 , i.e. , that @xmath366{{{t_{\rm c}}}-t},\qquad \kappa>0.\ ] ] first , if the blowup rate is exactly a quartic root , then @xmath367 indeed , figure [ fig : supercrit_peak_l3lt ] shows that in the case @xmath103 , @xmath368 , @xmath369 , implying that @xmath370{4\cdot0.289 } \approx 1.037\,.\ ] ] in the case @xmath99 , @xmath371 , @xmath372 , implying that @xmath373{4\cdot0.171 } \approx 0.909\,.\ ] ] since @xmath249 converges to a finite , negative constant , this shows that the blowup rate is exactly @xmath247 . second , according to lemma [ lem : supercrit_peak_rate_profile ] , if @xmath374 the self - similar profile @xmath47 should not satisfy the standing - wave equation . in figure [ fig : supercrit_peak_rescaled ] we saw that the rescaled self - similar bnls solutions and the corresponding ground - states differ considerably ( compare with figure [ fig : crit_peak_rescaled ] ) , indicating that the blowup rate is exactly @xmath247 . therefore , the numerical results again support conjecture [ conj : supercrit_peak_rate_profile ] . finally , we verified that the value of @xmath264 in the blowup rate is universal . we solve the bnls in the case @xmath352 with the initial condition @xmath375 . in this case , the calculated value of @xmath376 is @xmath377{-4l_tl^3}\approx1.037 $ ] , which is equal , to first @xmath378 significant digits , to the previously obtained value , see , for the initial condition @xmath353 . similarly , in the case @xmath354 , we solve the equation with the initial condition @xmath379 . the calculated value of @xmath380 is @xmath377{-4l_tl^3}\approx0.913 $ ] , which is equal , to first @xmath49 significant digits , to the previously obtained value , see , for the initial condition @xmath381 . in this study , we computed singular solutions of the bnls equation . these solutions become highly - localized , so that the spatial scale - difference between the singular region @xmath382 and the exterior regions can be as large as @xmath383 . in order to resolve the solution at both the singular and non - singular regions , we use an adaptive grid . we generate the adaptive grids using the _ static grid redistribution _ ( sgr ) method , which was first introduced by ren and wang @xcite , and later simplified and improved by gavish and ditkowsky @xcite . using this approach , the solution is allowed to propagate ( self - focus ) until it becomes under - resolved . at this stage , a new grid , with the same number of grid - points , is generated using deboors ` equidistribution principle ' , wherein the grid points @xmath384 are spaced such that a certain weight function @xmath385 $ ] is equidistributed , i.e. , that @xmath386dr = \text{const},\ ] ] see @xcite for details . the algorithms of @xcite keeps a recursive set of grids , the coordinates of each is given in the reference frame of the previous grid . this enables the simulation to reach high focusing levels , where the grid - points position can not be stored at the physical reference frame due to loss of significant digits . in this study , however , we implement a simplified version of the method of @xcite , in which we dispense with the above hierarchy of grids and store the grid - points position in terms of the original reference frame . on this highly non - uniform grid , we use a standard non - uniform finite - difference approximation of the radial biharmonic operator . the approximation is third - order accurate , with a seven - point stencil . thus , we are limited by the standard machine accuracy - bit floating points , with relative machine error of @xmath387 . using @xmath388-bit floating points will , for all practical purposes , eliminate the above restriction . ] to focusing levels of no more than @xmath389 . however , the resulting gain in software ( and numerical ) simplicity justifies this limitation . the method in @xcite allows control of the fraction of grid points that migrate into the singular region , preventing under - resolution at the exterior regions . this is done by using a weight - function @xmath390 , which penalizes large inter - grid distances . however , we found that this numerical mechanism , while necessary , is insufficient for our purposes . in order to understand the reason , let us consider the grid - point spacings @xmath391 . using the method of @xcite with both @xmath392 and @xmath390 causes a very sharp bi - partition of the grid points to those inside the singular region , whose spacing is determined by @xmath392 and is @xmath393 , and to those outside the singular region , whose spacing is determined by @xmath390 and is @xmath394 , see figure [ fig : w3_effect_old ] . inside each of these regions , the finite difference approximation we use is well conditioned . however , at the transition between these two regions , the finite - difference stencil , seven - points in width , spans grid - spacings with @xmath395 scale - difference leading to under - resolution which completely violates the validity of the finite - difference approximation . in order to overcome this limitation , we improve the algorithm of @xcite by adding a third weight function @xmath396 which penalizes the second - difference @xmath397 operator of the grid locations , allowing for a smooth transition between the singular region and the non - singular region , see fig [ fig : w3_effect_new ] . on the sequence of grids , the equations are solved using a predictor - corrector crank - nicholson scheme , which is second - order in time . here , we describe the adaptation of the spectral renormalization method ( srm ) to the standing - wave bnls equation . denoting the fourier transform of @xmath398 by @xmath399({{\mathbf { k}}})$ ] , equation transforms to @xmath400({{\mathbf { k}}})=\frac{1}{k^4 + 1 } \mathcal f\left[|r|^{2\sigma}r\right],\ ] ] where @xmath401 , leading to the fixed - point iterative scheme @xmath402= \frac{1}{k^4 + 1 } \mathcal f\left[|r_m|^{2\sigma}r_m\right ] , \qquad m=0,1,\dots~.\ ] ] typically this iterative scheme diverges either to @xmath403 or to @xmath404 . in order to avoid this problem , we renormalize the solution as follows . multiplying equation by @xmath399^*$ ] and integrating over @xmath405 gives the integral relation : [ eqs : srm ] @xmath406 \equiv \int{\left\vert{\mathcal f[r]}\right\vert}^2d{{\mathbf { k } } } , \quad sr[r ] \equiv \int \frac{1}{k^4 + 1 } \mathcal f\left[|r|^{2\sigma}r\right]\mathcal f[r]^*d{{\mathbf { k}}}.\ ] ] we now define @xmath407 such that the integral relation is satisfied by @xmath408 , i.e , that @xmath409 = c_m^2sl[r_m ] = c_m^{2\sigma+2}sr[r_m]=sr\left[r_{m+1/2}\right],\ ] ] leading to @xmath410}{sr[r_m]}\right)^{\frac{1}{2\sigma } } , $ ] and hence to @xmath411}{sr[r_m]}\right)^{1+\frac{1}{2\sigma } } { \left\vert{r_m}\right\vert}^{2\sigma}r_m.\ ] ] the spectral renormalization method is therefore given by the iterations @xmath412}{sr[r_m]}\right)^{1+\frac{1}{2\sigma } } \frac{1}{k^4 + 1 } \mathcal f\left(|r_m|^{2\sigma}r_m\right ) , \qquad m=1,2\dots~.\ ] ] in this work , we use the srm to solve for the cases @xmath413 without imposing radial symmetry . alternatively , one might have solved the radial equation using a modified hankel - like transform instead of the fourier transform . our main reason for not doing so is the convenience and cost - effectiveness of using the fast fourier transform . we also note that our non - radially - symmetric method produced a radially - symmetric solution , which suggests that the ground state is radially symmetric . we thank nir gavish for useful discussions . this research was partially supported by grant # 123/2008 from the israel science foundation ( isf ) . here we provide an extension of the compactness lemma for @xmath414 functions @xcite , to the case of @xmath415 : ( compactness lemma)[lem : compactness ] let @xmath147 and let @xmath416 be in the @xmath18-subcritical regime . then , the embedding @xmath417 is compact , i.e. , every bounded sequence @xmath418 has a subsequence @xmath419 which converges strongly in @xmath420 . if @xmath421 , then the sequence @xmath422 has a subsequence @xmath419 which converges weakly to @xmath423 in @xmath18 . since the limit of radial functions is a radial function , @xmath424 . in addition , since for any bounded domain @xmath425 , the embedding @xmath426 is compact , there is a subsequence which converges strongly to @xmath423 in @xmath427 , i.e. , @xmath428 . from the gagliardo - nirenberg inequality on the bounded domain @xmath425 , see @xcite , @xmath429 and since @xmath430 in the @xmath18-subcritical case , it follows that @xmath431 strongly in @xmath432 , so that @xmath433 next , strauss radial lemma @xcite for @xmath434 functions gives that @xmath435 and @xmath436 , @xmath437 so that @xmath438 s.t . @xmath439 @xmath440 finally , since @xmath441 the convergence in @xmath442 is obtained . from the proof of theorem [ thrm : self - similarity ] , @xmath443 and since @xmath444 , we have that @xmath445 @xmath446 since @xmath447 , it follows that @xmath448 , and so @xmath449 this is true @xmath450 , and so @xmath451 for the second result , since @xmath452 , it follows that for all @xmath187 there exist @xmath453 such that @xmath454 therefore , since @xmath455 a similar argument as in the previous section gives @xmath456 in @xmath0-critical case @xmath12 , which is always in the @xmath18-subcritical regime , the appropriate gagliardo - nirenberg inequality in @xmath18 is @xcite : ( gagliardo - nirenberg inequality ) let @xmath12 , and let @xmath457 , then @xmath458 we note that the ground - state @xmath10 of equation is the minimizer of the gagliardo - nirenberg inequality @xcite , and that its @xmath0 norm , the critical power , satisfies @xmath459 hence , the gagliardo - nirenberg inequality implies the following corollary : [ cor : p2h_bound ] let @xmath460 and @xmath12 , then @xmath461 \geq \left [ 1-\left ( \frac{{\left\vert{f}\right\vert}_2 ^ 2}{{{p_{\rm cr } } } } \right)^{\sigma } \right ] \cdot { \left\vert{\delta f}\right\vert}_2 ^ 2,\ ] ] so that @xmath462\geq0.\ ] ] as @xmath356 , the nonlinear term in becomes negligible , and reduces to @xmath463 where @xmath464 in order to apply the wkb method , we substitute @xmath465 , and expand @xmath466 substituting @xmath467 and balancing terms shows that @xmath468 , and that the equation for the leading - order , the @xmath469 terms , is @xmath470 therefore , @xmath471 the equation for the next order , the @xmath472 terms , is @xmath473 implying that @xmath474 the next - order terms are @xmath475 and can be neglected . [ eqs : wkb_solutions ] we therefore obtain the three solutions @xmath476 since is a fourth order ode , another solution is required . to obtain the fourth solution , we substitute @xmath477 in and obtain that the equation for the leading - order , the @xmath478 terms , is @xmath479 and that the next order is @xmath480 and can be neglected . the fourth solution is therefore @xmath481 the lagrangian density of the bnls is @xmath482 for example , the bnls action integral is invariant under the phase - multiplication @xmath486 in this case , @xmath487 and so theorem [ thrm : noether ] implies that the integral @xmath488 d{{\mathbf { x } } } \\ & = \int \left [ \frac i2\psi^ * \left ( \psi_t \cdot 0 + \nabla \psi \cdot 0 - i \psi \right ) - \frac i2\psi \left ( \psi_t^ * \cdot 0 + \nabla \psi ^*\cdot 0 + i \psi^ * \right ) -\mathcal{l}\cdot 0 \right ] d{{\mathbf { x } } } \\ & = { \left\vert{\psi}\right\vert}_2 ^ 2,\end{aligned}\ ] ] i.e. , the power , is a conserved quantity . other conservation laws can be found in a similar manner . height 2pt depth -1.6pt width 23pt , _ global well - posedness and scattering for the focusing energy - critical nonlinear schrdinger equations of fourth order in the radial case _ , j. of differ . equations , 246 ( 2009 ) , pp . 3715 3749 .

we have obtained high dynamic range , good natural seeing images of bl lacertae objects ( bl lacs ) to search for the agn host and thus constrain the source redshift . these objects are drawn from a sample of bright flat - spectrum radio sources that are either known ( via recent _ fermi _ lat observations ) gamma - ray emitters or similar sources that might be detected in continuing gamma - ray observations . all had spectroscopic confirmation as bl lac sources , but no redshift solution . we detected hosts for 25/49 objects . as these galaxies have been argued to be standard candles , our measured host magnitudes provide redshift estimates ( ranging from 0.21.0 ) . lower bounds are established on the redshifts of non - detections . the mean of the fit redshifts ( and lower limits ) is higher than those of spectroscopic solutions in the radio- and gamma - ray- loud parent samples , suggesting corrections may be needed for the luminosity function and evolution of these sources .

1002.1343

bl lac objects , being highly continuum - dominated , are a perennial problem for those wishing to study the evolution of agn populations . the lack of visible broad lines and , even more , the low equivalent width of host absorption features makes redshift determinations extraordinarily difficult . yet , since the continuum domination is a manifestation of the good alignment of the relativistic jet outflow to the earth line - of - sight , @xcite knowledge of the distance , and hence luminosity scale , of these sources is very interesting . this problem has become particularly important with gamma - ray detection of large numbers of blazars with the _ fermi _ lat @xcite . it has been known since _ egret _ that radio loud flat spectrum blazars dominate the bright extragalactic sources in the gev sky @xcite ; these objects are flat - spectrum radio quasars ( fsrq ) and radio loud bl lacs . we have good techniques in place to identify likely radio counterparts for the gamma - ray sources . extensive spectroscopy campaigns , especially on the egret - like ` cgrabs ' sample @xcite have provided nearly complete redshifts and characterizations of the fsrq . however , despite extensive observation , including 8+m telescope integrations , less than half of the bl lac counterparts have spectroscopic ids . in some cases , high quality spectroscopy can give useful lower limits on the source redshift @xcite , but many remain unconstrained . in the first lat blazar catalog @xcite , the problem is even more pronounced since the excellent high energy response of the lat favors detection of hard spectrum sources . bl lacs are substantially harder ( @xmath1 ) than fsrq ( @xmath2 ) and so provide a larger fraction , @xmath3% of the _ fermi _ blazar sample , than was seen by _ we attempt here to constrain the redshifts of radio loud bl lacs which have shown no convincing spectroscopic redshift solution , despite sensitive observations on large telescopes , by high dynamic range imaging searches for the agn host . our targets are drawn from the ` crates ' catalog of bright flat - spectrum @xmath4 radio sources @xcite . specifically , in this program we targeted the subset of crates consisting of @xmath5 ) sources selected as likely gamma - ray blazars before the fermi mission ( i.e. cgrabs sources , healey et al . 2008 ) and @xmath6 ) sources that are likely counterparts to lat sources detected early in the mission @xcite . all sources are at declination @xmath7 and all have been shown to display featureless optical spectra @xcite , with sensitive ( 4m+ class ) observations . we were able to image @xmath8 of the sources satisfying these critera at the time of the observing campaigns . it has been claimed @xcite that bl lac host galaxies detected with hst imaging are remarkably uniform giant ellipticals with @xmath9 ; accordingly host detections can give redshift estimates and upper limits on the host flux can give lower limits on the distance . here we do not test this assumption or the possible biases that would select a modest magnitude range for the detected host sample . we simply apply the method to extract redshift constraints , noting that while individual redshifts are doubtless imprecise , the estimates can still be useful for statistical studies of bl lac evolution and as a guide to and comparison with other methods of redshift estimation . conversely , when precision spectroscopic redshifts become available , our measurements can be used to help constrain the host evolution . since we will be searching for de vaucouleurs profile excesses in the wings of the stellar psf of the bright bl lac core , we require good natural seeing and a moderate field of view for adequate comparison stars . for example , while near - ir adaptive optics can deliver superior psf cores , the wings of the source at @xmath1090% encircled energy are extensive and often quite variable over the small corrected fov . this prevents the accurate psf modeling and subtraction required to obtain host measurements whose integrated magnitude can be 10% or less of the core flux . all observations were carried out using the wiyn 3.6 m telescope at kitt peak national observatory , which features good natural seeing and a @xmath11 fov with the standard mosaic imagers . we employed optic on the nights of 2008 oct . 31 nov . 2 for our initial imaging run . this 2@xmath122k@xmath124k camera covers 9.6@xmath13 at 0.14@xmath14/pixel and features lincoln labs orthogonal transfer ccds , which allow rapid electronic ( ot ) tracking using an appropriate in - field guide star . with a relatively short ( 25s ) read - out time , we were able to combine short exposures that avoided core saturation with longer exposures that probe for the host galaxies in the psf wings . the targets were imaged with the sdss @xmath0 filter kpno 1586 . we observed 31 targets , with total on - source time ranging from 210s to 2100s ( table [ optic_results ] ) . sky conditions were photometric through the run , and we were able to employ ot guiding for the bulk of the sources , with correction rates up to 50 hz . the final @xmath0 image quality had a median fwhm of 0.65@xmath14 ; the best 25% of the sources had image stacks with fwhm @xmath15 . a period during which mirror cooling was lost resulted in some observations with poor psf ( up to 1@xmath14 ) . software problems in addition cost several hours . while ot guiding significantly improves the source psf , which is crucial for our science goals , it does create some complications during data reduction . in particular , flat - field frames must be created for each individual exposure , with weighted exposure per pixel following the history of ot offsets during each individual exposure . this creates some challenges in assembling the final mosaic images , as defects can affect adjacent pixels . the ` smoothing ' of the flat - field response also affects the noise statistics of the final image . the data were processed using routines in the iraf package ` mscred ` . with the excellent psf ( as small as 0.4@xmath14 ) achieved for some frames , we required superior camera distortion corrections at each position angle ; for some orientations a lack of field stars limited the accuracy of the astrometric solution . after registration and exposure - weighting , median combined image stacks were prepared for further analysis . observations were made using the mini - mosaic camera ( minimo ) on the nights of 2009 march 24 - 25 , under highly variable conditions . the 2@xmath122k@xmath124k mosaic similarly has a plate scale of 0.14@xmath14/pixel and covers a field of 9.6 arcmin . we again employed the sdss @xmath0 filter . because of the long ( @xmath163 minute ) readout time of the mini - mosaic camera , most objects had 3 - 5 dithered exposures of 300s each . these relatively long exposures meant that many of the bl lac targets had saturated cores , despite the typically poorer final image quality during this run ( median final fwhm 0.84@xmath14 ) . the exposure sequences and final diq for each object are listed in table [ minimo_results ] . in all , 18 bl lacs were observed . since early lat detections had been announced by this run @xcite we were able to specially target known gamma - ray emitters lacking redshift solutions . the sky transparency was variable and half the run was lost to high winds and poor seeing . minimo employs conventional ccds , and so standard reductions were conducted with the iraf package ` mscred ` . camera distortions were mapped using usno a2.0 catalog stars ; these were also used to assign a wcs to each frame . after processing and bad pixel correction the individual dither frames were aligned using unsaturated stars near the bl lac , exposure weighted and median combined using standard sigma - clipping algorithms to produce our final images . minimo suffers from ` ghost ' images due to amplifier cross - talk . care was taken during target acquisitions to ensure that these did not fall in the vicinity of the bl lac . we made no attempt to correct these ghosts , but avoid using any affected comparison stars . a number of the target fields were covered by the sdss , so we were able to establish our zero point directly from standard aperture photometry of unsaturated field stars ( 16 @xmath17 19 ) . for the optic run only six fields had suitable stars . the median rms of these zero point measurements was 0.052 mag . while this was significantly larger than the intrinsic sdss photometric errors , the scatter is smaller than our typical host photometric errors and much smaller than the scatter in the claimed absolute magnitude of the bl lac host ( @xmath18mag ) . for the exposures lacking sdss comparison stars , we applied the mean zero point after correction for atmospheric extinction @xcite . given that conditions were photometric throughout , we are conservative in assigning the 0.068 scatter between the zero point measurements as the final zero - point error . while conditions were variable during the minimo observations , 12/18 fields had stars suitable for direct sdss cross - calibrations . in particular for four observations logged as cloudy we were able to establish in - field calibration . taking the rms zero point from the remainder of the fields , we established , after extinction correction , a zero point with scatter of 0.036 mag . this was applied to the remaining target fields . accurate psf modeling is essential to extract the bl lac host from the wings of the nuclear point source . we built an effective psf in the stacked image of each bl lac using the iraf / daophot ` psf ` routine . psf stars were selected from those that appeared in all sub - frames making up the image stack , had stellar fwhm and appeared isolated . psf stars were selected by hand and were individually checked for consistency with a stellar fwhm . in practice the process was iterative ; some otherwise desirable psf template stars had faint unresolved sources in the wings , so we modeled and subtracted these sources before recombining for an improved psf . point sources as faint as @xmath19 were removed in this manner . each template psf has a radius of 5.6@xmath14 and covers over 11 fwhm even for our worst diq images . the background level for each psf star was determined from the mode of a 2@xmath204@xmath14 wide annulus starting at 7@xmath14 from the star . in a few of the images with poor seeing the background was determined from larger radius , to ensure insignificant contamination from the psf wings . to best model the psf wings , we also used some stars with nonlinear cores . above the non - linearity level ( 30,000 adu for optic , 40,000 adu for minimo ) , the cores were masked in the psf stack . we found that a psf model with linear gradients across the image @xcite was required to take full advantage of the small fwhm of the best optic data . to guide our fitting we examined the spatial dependence of the fwhm . three terms might be expected to contribute to psf degradation . first , we expect the efficacy of the low order atmospheric corrections to fall off with distance from the guide star . we attempted to mitigate this effect by guiding in the target quadrant , whenever possible . second , distortions in the camera system are expected to increase away from the optical axis ( assumed to be near the field center ) . finally , since our principal frame registration is on the bl lac core , registration or residual rotation errors could degrade the psf as a function of distance from the target itself . to test for these effects , we collected fwhm measurements of unresolved sources across a number of fields . in no case did we find the psf was primarily correlated with distance to the target either in the individual exposures or in the final combined images , implying that stacking error was not important . in some cases the psf degradation appeared correlated with distance from the optical axis , implicating camera distortions , but for other images the distance from the guide star appeared to dominate ( see figure [ fig : comparison ] ) . with the poorer seeing experienced during our minimo run , we found that the spatial variation of the psf was much less important , with the fwhm varying by no more than a few @xmath21pixels ( @xmath22 ) across the field of view . again stacking errors were negligible . however for consistency we fit for and applied linear gradients in the psf model . thus , in forming the final model psf for the analysis , we were able to draw stars from across the fields avoiding only the outermost corners of the frames , where there was notable degradation . in general between @xmath16 1/6 and @xmath16 1/5 of the psf stars used were saturated in their cores ( only @xmath16 1/9 with minimo , given the poorer seeing ) . tables [ optic_results ] and [ minimo_results ] list the number of psf stars used in each image . the median number of psf stars for each field was 30 for both optic and minimo . previous studies @xcite indicate that bl lac hosts are better fit by de vaucouleurs profiles than exponential disks . we therefore adopt simple de vaucouleurs profiles with fixed sersic index 4 , convolving the model with the image psf , in our galaxy fitting . we did find examples of disturbed galaxies and hosts with faint companions , but no host appeared disk - dominated . our bl lac image model includes up to five parameters , introduced hierarchically . we start by fitting for nuclear ( psf ) flux plus possible host . we assume that the host galaxy isophotes are centered on the nucleus @xcite , and fit or bound the host flux , initially assuming a a fixed effective radius of @xmath23 ( 10 kpc at @xmath24 ; we assume @xmath25 , @xmath26 and @xmath27 km s@xmath28 mpc@xmath28 throughout this paper ) . we next allow the de vaucouleurs profile @xmath29 to vary , when required by a high significance for a measurement of @xmath29 in the fit . finally for a few bl lacs , we find a significant detection of host ellipticity ; in these cases we include the ellipticity @xmath30 and position angle @xmath31 as additional free parameters in the fit . after fitting the host normalization , we report galaxy magnitudes by extrapolating the integrated light to infinity . to prepare for host fitting , we determine the centroid position of the core psf , using the daophot task ` peak ` , which matches to the model psf , excluding pixels above our ( conservative ) non - linearity threshold ( 30,000 adu for optic , 40,000 adu for minimo ) , and delivering centroid coordinates accurate to @xmath32 pixels . since a number of our images were saturated , this clipping was important for obtaining accurate centroids . in almost all cases , the galaxy surface brightness dropped below the background surface brightness fluctuations within our standard @xmath33 psf radius . accordingly the bl lac image model was fit to @xmath34 cutouts from the image stack centered on the bl lac core pixel . daophot provides a 2@xmath12-oversampled psf model . this was re - binned to the original pixel size , using the precisely determined centroid and then used in the subsequent fitting . this psf centering and re - sampling proved to be quite accurate dipole structure was never evident in residuals to our final bl lac model , whereas with even 0.05 pixel shifts for the bl lac core position such residuals were obvious . we apply our bl lac fit to a subset of the pixels in the cutouts around each bl lac . first all pixels beyond @xmath33 ( outside the psf model ) are masked . pixels associated with resolved objects within the fitting radius but more than three fwhm from the bl lac core are also masked ( in practice we exclude rectangular regions , see figure [ fig : j0814_blanking ] ) . unresolved objects more than three fwhm away are fit to the psf model and then treated as fixed background . occasionally , very bright nearby stellar objects show significant residuals at the psf core after subtraction ; in these cases a few pixel rectangular region is masked at the core before the final bl lac fit . the occasional column or pixel defect or cosmic ray are similarly masked . finally we mask the pixels within 1 fwhm in diameter of the bl lac core . in roughly 25% of the optic images this was required in any case due to non - linear adu levels . however , we found that sampling noise at 1 fwhm could dominate the fit statistics , so a consistent exclusion of pixels at the core was used for all objects . for a few of the brightest objects , we extended the core exclusion region to 1.5 fwhm , while for a few of the faintest , we used 0.5 fwhm . the final results were not sensitive to the precise value used . for each bl lac model the background was determined from an annulus starting at 7@xmath14 containing 2 - 3@xmath12 the pixel count of the psf model region ; for a few poor seeing images and bl lacs with very extended hosts we used a larger annulus . [ h ] preparing the fitting region for j0814 + 6431 ( image fwhm @xmath35 ) . one resolved companion and pixels near the centroid of the nuclear component are masked . the corrected fov is @xmath36 in radius ; n up , e left . ] for all multi - parameter fits , the parameters are determined by minimization of the quantity : @xmath38 where the sum is over the @xmath39 unmasked pixels in the fitting region , @xmath40 is the observed intensity in adu , @xmath41 is the model intensity in adu and @xmath42 is the expected variance . the model is the sum of the constant background , psf and host galaxy terms , where the analytic host model is integrated over a given pixel to determine the galaxy contribution . the variance is estimated according to simple poisson statistics to be : @xmath43 where for optic the gain was @xmath44 and for minimo 1.4@xmath45 . the read noise @xmath46 is @xmath47 ( optic ) or 5.5@xmath48 ( minimo ) and our final image stack is the from the median of @xmath49 normalized single exposures . this is a simplification since for optic the orthogonal transfer during integration correlated the sky noise ( likely an artifact of imperfect flat - fielding ) . also the residual pixel noise in the unmasked region adds to the variance for sharp images . nevertheless we find that this variance is a reasonable first approximation to the local noise , and we can determine the best - fit model by @xmath37 minimization . we use the simulated annealing algorithm from press et al . ( 1986 ) . the final size of the simplex in @xmath50 dimensions gives the relative errors between the @xmath51 estimates of the fitting parameters . we would like to relate this to the 1@xmath52 statistical error , to provide a standard statistical error estimate for each parameter . we do this by scaling to an externally estimated 1@xmath52 error determined for each point source flux @xmath53 . we determine this error by analogy with standard aperture photometry . in our case the central fwhm is masked in the fit so we compute the fractional error in the psf amplitude for an annulus outside the masked region of width 1 fwhm as @xmath54 where @xmath55 is the source flux ( adu ) within the photometry annulus , @xmath56 is the number of pixels within the annulus , @xmath57 is the standard deviation of the background ( adu ) , and @xmath58 is the number of pixels used to determine the background value . the first term under the radical is poisson noise . the second is a statistical uncertainty due to the random fluctuations and unresolved sources in the background . the final term accounts for the possibility of a poisson error in background level . the first term generally dominates , and thus @xmath59 is @xmath60 as the annulus contains roughly 1/2 the flux of an aperture without the central mask . we found a typical @xmath59 for our psf stars of @xmath61 . with the large number of counts for the point sources measured here , even small centroiding errors can in principle contribute to the error , since a varying fraction of the point source falls in the central masked pixels . to test for this in each psf fit , we artificially shifted the model psf position by @xmath32pixels , the maximum centroiding error , in each coordinate . the rms variation in the included counts was computed to estimate @xmath62 , the fractional error in flux estimation due to imprecise centroiding . although this error only became comparable to @xmath59 for a few of the brightest point sources , we sum it in quadrature to compute our statistical photometry error as @xmath63 , where @xmath64 is the total number of counts for the point source in adu . as noted above , the size and shape of the simplex determined from our @xmath37 minimization provides _ relative _ fitting errors . we adjusted our convergence criterion such that the rms spread about the mean fit value for the psf amplitude was always close to @xmath53 which , since the nuclear point source always dominates the model , is a very good estimate of the true model - fit statistical error . thus , by rescaling the simplex amplitude such that the rms spread in the nucleus flux estimates was equal to this photometric error @xmath53 , we have normalized the simplex - determined error to be 1@xmath52 ( statistical ) in each quantity . we call this error @xmath65 . since our models are dominated by the bright core psf , which is different for each exposure , and since we have a limited number of stars in each frame to generate a psf model , our final uncertainty must be dominated by systematic errors due to imperfections in this model . this is in contrast to the hst - based fitting of @xcite , where the psf is , of course , stable and well - modeled and unresolved background structure dominated the final uncertainty . we therefore wish to estimate our systematic psf uncertainty . we do so by fitting for a de vaucouleurs host around unresolved stellar sources . in each image stack we choose such test stars relatively close to the bl lac with comparable brightness . on average we were able to measure seven such stars per image . ideally none would have been used in forming the psf , but since these are the stars most suitable for psf modeling , they were largely included . before fitting , each star was treated exactly as its associated bl lac , including the same masking of central pixels . we then fit for the psf amplitude and ` host ' amplitude , with the de vaucouleurs @xmath23 fixed at our default radius . the fit host amplitude could either be negative ( indicating excess wings on the model psf ) or positive ( psf model too narrow ) . note that decreasing the host angular size would weaken constraints on its amplitude , since it would be more highly covariant with the psf . our adopted value ( appropriate for a typical @xmath66kpc at @xmath24 ) is reasonably conservative ; even at the minimum angular size at @xmath67 , this would only decrease by 28% . the dispersion in the amplitudes of the fitted ` host ' fluxes provides an estimate of our systematic uncertainty . however , since the stellar fluxes covered a substantial range , we elected to scale the ` host ' flux significance to the statistical error on this flux for each test star @xmath65 the rms spread in the ` host ' fluxes fit around the test stars , in units of @xmath65 , then provides an estimate of the systematic error , for each individual bl lac field . figure [ fig : systematic ] shows the distribution of ` host ' significance ( @xmath68 ) for the test stars in the optic and minimo fields , along with the distribution of the rms significance for the individual fields . the rms significance of these ` host ' fits for optic and minimo were 8.3@xmath69 and 8.0@xmath69 , respectively . this suggests that our statistical ( poisson ) errors on the host flux substantially underestimate the true errors by @xmath70 . however , it should be noted that the distributions are approximately normal and that the mean fitted host flux is not significant . for the optic data , we obtain @xmath71 while for minimo the average is 0.1@xmath69 . in view of the systematic errors we expect random offsets of @xmath72 and @xmath73 for the two sets of test stars . we conclude that the frame - to - frame spread in upper limits on surrounding host galaxies are appreciably larger than expected from pure poisson statistics this can be attributed to errors in forming the individual psf models and to unresolved structure in the psf stars and test stars backgrounds . however there is no _ overall _ introduced host flux ; implying that our psf model is a fair , albeit uncertain , representation of the data . to be conservative , we adopt for our final error the _ sum _ of the statistical and systematic errors , namely @xmath74 , where @xmath75 is the rms value of the individual test stars in the stacked image of an individual bl lac . this is conservative since while we expect the statistical errors to have a normal distribution , it would be very surprising if the much larger systematic errors had wings as large as a gaussian out to many times the fwhm of the distribution . we tested this error estimate by inserting poisson realizations of model hosts around isolated stars in several fields . for artificial galaxies assigned the total counts at the 3@xmath76 detection limit , we recovered host fluxes with rms errors of @xmath77 . in no case was the difference between the injected and recovered counts larger than @xmath78 . we constrain the properties of the bl lac host by the hierarchical fit . we first fit for a circularly symmetric host with fixed angular radius @xmath79 , just as for the test star measurements . the final uncertainty on the host counts is taken to be @xmath76 where the statistical error is taken from the simplex rescaled to the aperture photometry error for the bl lac core , while the systematic error is the multiple of this statistical error determined by ` host ' fits to the test stars in this bl lac s field , as described above . if the host flux is larger than @xmath80 ( which is dominated by the systematic uncertainty ) , we deem it significant . if the significance does not reach @xmath80 , we infer an upper limit of @xmath81 , or @xmath82 , whichever is larger . if there is a significant detection of a bl lac host , we next re - run the fits allowing the host angular radius @xmath29 to vary . the only fitting constraint is @xmath83 . if the best fit value of @xmath29 is less than the fwhm for the image , we do not deem it significant . a value for @xmath29 is only quoted if @xmath84 , where the final statistical plus systematic error is estimated from the re - scaled simplex , as above . significant estimates of @xmath29 were found for 25% of the optic - measured bl lacs and 33% of those measured with minimo . note that the overall flux and its fit error were free to vary , as well . as a consistency check , we note that in no case where a significant @xmath29 measurement was found , did the host flux measurement decrease in significance to below @xmath85 . finally , we attempted to fit for a significant host ellipticity , with two additional parameters @xmath30 and the position angle @xmath31 added to the model . these parameters both exceeded @xmath82 significance for only two of the bl lacs observed with optic . again the host flux and size @xmath29 were free in this final fit . tables [ optic_results ] and [ minimo_results ] contain the final fit nucleus and host amplitudes ( converted to magnitudes , see below ) and the host physical size ( when available ) . details of the observations ( exposure lengths , fwhm of the stellar sources and the number of stars used in the psf model ) are also listed . since @xcite have argued that bl lac host galaxies are standard candles with @xmath86 , we can use our measurements and upper limits on the host flux to extract redshift estimates and lower bounds . to do this we compute an @xmath0 hubble diagram for our kpno 1586 filter . we improve on the similar r - band hubble diagram of @xcite , by including host evolution , assuming a host ( elliptical ) formation redshift of @xmath87 @xcite . with this assumption , we adopt an elliptical galaxy spectrum computed at the appropriate age from the pegase model of @xcite , where the age is computed for each @xmath88 for our standard cosmology . to lock the overall normalization to the @xmath89 of @xcite , we fixed the flux of the @xmath90 , evolved elliptical spectrum to @xmath91 . this required folding through an @xmath46 filter . unfortunately the precise kpno kron - cousins @xmath46 filter used is not recorded , except in a private 1995 communication between j. holtzman and landolt @xcite . accordingly we use the transmission curve for kron - cousins r filter kpno w004 , first obtained in 1994 ; this may be the precise filter used in the original study . in any case , after normalizing with this filter at @xmath90 , we take the model flux from each observation redshift , convert to the observed frame using the luminosity distance @xmath92 from our standard cosmology and convolve with the @xmath0 kpno 1586 transmission ( figure [ fig : filters ] ) to obtain the apparent magnitude . the resulting hubble diagram is shown in figure [ fig : hubble_diagram_all ] . for comparison we have also computed the @xmath46-band hubble diagram for this evolving model . the curve is quite close to @xcite s analytic approximation , although departures caused by the evolution are evident . hubble diagram computations with other evolving galaxy models yielded similar curves , with differences small compared to those introduced by uncertainty in the host absolute magnitude . to compare with our measurements , we converted our model fit host counts ( in adu ) to @xmath0 magnitudes , including atmospheric @xcite and galactic @xcite extinction corrections . the establishment of the flux scale is discussed above ( 3 ) , while the assumed galactic extinctions from the dust maps at the bl lac locations are listed in tables [ optic_results ] and [ minimo_results ] . finally , these host magnitudes ( and lower limits ) are converted to redshifts using our hubble diagram . of course the host absolute magnitude uncertainty translates into a range of possible redshifts . there are two contributions to the full redshift uncertainty . the first is host measurement uncertainty , including both statistical and systematic photometry errors . we use the @xmath93 range for the maximum and minimum host magnitude to infer asymmetric error bars on the host redshifts . this is augmented by adding , in quadrature , the photometry errors estimated for our zero point this increase is however very small . these values are listed as the first error flags in the final columns of tables [ optic_results ] and [ minimo_results ] . in addition , the claimed @xmath94mag dispersion in the host luminosity leads to a redshift range . we propagate the @xmath90 @xmath46 absolute magnitude uncertainty through observed @xmath0 and hence to @xmath95 as a function of redshift ( see figure [ fig : error_comparison ] ) . this nearly always completely dominates the redshift uncertainty and is listed as the second error flag in our tables . for objects with a fit angular @xmath29 , the central redshift estimate is used to convert to proper kpc ; the values are comfortably close to the otherwise assumed @xmath66kpc . finally , for undetected hosts , we take the minimum @xmath96 allowed by the @xmath82 host magnitude limit , the photometric scale errors , and the standard value for the host luminosity to obtain a lower limit on the host redshift . again this is listed in the last column of tables [ optic_results ] and [ minimo_results ] . we have treated our measurement errors conservatively , having summed the systematic and statistical uncertainties in our error flags and requiring @xmath97 detections before claiming a measurement is significant . on the other hand , we consider only the assumed dispersion in the host luminosity . clearly an improved calibration of the host magnitudes ( if uniform ) can make these measurements much more useful . hubble diagrams calculated in kpno w004 @xmath46 and kpno 1586 @xmath0 . the analytic estimate for the @xmath46 hubble diagram of @xcite is shown for comparison.,width=359 ] hubble diagrams calculated in kpno w004 @xmath46 and kpno 1586 @xmath0 . the analytic estimate for the @xmath46 hubble diagram of @xcite is shown for comparison.,width=359 ] [ h ] we were able to detect hosts for 14/31 optic - observed bl lacs and 11 of 18 observed with minimo . the redshift estimates varied from @xmath98 and the lower limits from @xmath99 . for optic the median value was @xmath100 for minimo @xmath101 . it is interesting to compare this with the _ hst _ host detections of @xcite ; in this study nearly all bl lacs at @xmath102 were resolved but only 6/23 sources with known @xmath103 yielded host detections . while at first sight it might seem surprising that a ground - based program could be competitive , it should be remembered that the _ hst _ snapshot bl lac survey observed with the f606w and f702w filters . the @xmath0 filter blue cut - on is @xmath104 nm redder than for the bulk of the _ hst _ observations , and so we do not expect to suffer appreciable surface brightness attenuation from the 4000 break until @xmath105 . further the typical hst snapshot exposures were @xmath106s with a 2.4 m aperture @xmath107 900@xmath202100s with a 3.6 m aperture . the combination of redder band and deeper exposure allow some additional sensitivity to the low surface brightness host wings . of course , our sources were selected from those _ lacking _ spectroscopic redshifts , making it likely that these are a higher redshift sub - sample . however it is also worth noting that the typical @xmath108 core host ratio plotted in figure [ fig : nuchost ] represents an even stronger core dominance , compared to the _ hst _ sample , than one might think since the continua of bl lacs is quite blue compared to the hosts , and our @xmath109 bandpass is redder than that of previous @xmath46 or @xmath110 studies . [ h ] for 14 of the 25 detected hosts we obtained estimates of the host angular size . the inferred median effective radii were 9.2 kpc ( optic ) and 8.0 kpc ( minimo ) in good agreement with previous bl lac host estimates @xcite . the other detected hosts had @xmath29 measurements with @xmath111 significance or best fit radii smaller than the fwhm and thus were judged to be insignificantly resolved . in two cases we have a formal detection of ellipticity . for j0211 + 1051 the small value @xmath112 accords well with the visual impression ( figure [ fig : coresub ] ) . in the other case , j0348@xmath201610 , the ellipticity is also small @xmath113 , but here the result is suspect , as there is a bright companion within @xmath114 . indeed it may be no coincidence that the two objects with significant @xmath30 both have a large number of companions . while the upper limits on @xmath30 are in some cases constraining , we conclude that with our modest ground - based resolution we can not in most cases probe the host structure . one arena where the comparison with hst - based measurements should be fairly robust is the host - nucleus flux ratio . here we find a marked difference with the hst sample . of the 69 bl lac hosts resolved by @xcite , 54% of these had @xmath115 . even counting the 42 unresolved bl lacs 34% had a nuclear - host ratio @xmath116 . in contrast , we find only 1 object of our 49 bl lacs has @xmath117 . indeed 63% of our hosts are @xmath118 fainter than the nucleus ( figure [ fig : nuchost ] ) . two selection effects may account for this difference . first , these objects have been selected as bright flat spectrum radio core sources . further a significant fraction of the minimo targets were in addition known to be active gamma - ray emitters . this implies that the earth line - of - sight is even more nuclear component dominated ( on - axis ) than for the typical bl lac . second , these are the subset of these systems _ lacking _ previous spectroscopic redshifts . again , this suggests that they should be even more core continuum dominated than the typical bl lac . we present here a sample of azimuthally averaged @xmath0 instrumental surface brightness plots , compared with our best fit model components ( figures [ fig : radopdet ] @xmath20 [ fig : radmi ] ) . the surface brightness data are measured from excess counts above the above the fitted background level measured in annuli about the core position . the model curves are from integrations over these radial bins . the error bars shown are 1@xmath52 , calculated in a fashion consistent with the variance in 5 . we show two bl lacs with host detections using the optic camera , as well as two fits providing upper limits . for minimo we show one detected host and one unresolved source . it has been noted @xcite that bl lac hosts are surrounded by a significant excess of nearby galactic objects . following @xcite , we count companions within a projected proper distance of 50kpc of the bl lac centroid . when we do not have a @xmath96 estimate , we count only companions within @xmath119 of the nucleus ( 50kpc at the minimum angular diameter distance @xmath120 ) . we count only resolved objects . we find at least one companion for 51% of our bl lacs ( 25/49 ) , in reasonable agreement with the detection rate of @xcite ( 47% ) and @xcite ( 62.5% ) . of course , with our ground - based imaging , some of the unresolved nearby sources may represent additional compact galactic companions . the number of companions @xmath121 within a projected distance of @xmath122 kpc , the magnitudes of the two brightest companions ( and the comparison of the brightest to the host flux @xmath123 ) and a flag indicating evidence for interaction are given in table [ companionship ] . as noted above , our two hosts with nominal detections of ellipticity have both large numbers of nearby companions and morphological evidence for interaction . these may in fact represent recent mergers in compact groups . j1440 + 0610 has an extended companion with centroid at @xmath124kpc which extends within 50kpc . for two objects ( j0050@xmath200929 and j0743 + 1714 ) the companion at small radius may be physically contiguous with the undetected host . clear evidence for interaction was seen only for the better - seeing optic data , so additional companion interaction is likely . [ h ] c c c c c c c c name & @xmath125 & @xmath126 & @xmath127 & @xmath123 & @xmath128 ( @xmath129 ) & @xmath128 ( kpc ) & int . ? + + [ -2ex ] j0050@xmath200929 & 1 & @xmath10 19.5 & - & - & @xmath130 0.6 & - & m + j0202 + 4205 & 1 & 22.7 & - & 1.3 & 4.1 & 33 & n + j0203 + 7232 & 1 & 22.3 & - & - & 5.6 & - & n + j0211 + 1051 & 7 & 21.9 & 23.6 & 5.0 & 7.3@xmath131 & 24@xmath131 & y + j0219@xmath201842 & 1 & 21.5 & - & 1.7 & 3.6 & 25 & y + j0348@xmath201610 & 5 & 21.5 & 21.6 & 2.9 & 2.7 & 14 & y + j0607 + 4739 & 1 & 23.0 & - & - & 4.6 & - & y + j0610@xmath201847 & 1 & 23.3 & - & 3.2 & 4.9 & 35 & n + j0625 + 4440 & 2 & 22.5 & 24.2 & - & 3.8@xmath131 & - & m + j0650 + 2502 & 2 & 21.6 & 22.5 & 2.6 & 5.3@xmath131 & 31@xmath131 & n + j0743 + 1714 & 2 & @xmath10 21.8 & 23.7 & - & @xmath130 0.6 & - & m + j0814 + 6431 & 1 & 22.3 & - & 3.9 & 5.2 & 25 & m + j0817@xmath200933 & 2 & 22.4 & 23.8 & 2.0 & 4.9@xmath131 & 35@xmath131 & n + j0835 + 0937 & 1 & 22.2 & - & - & 3.8 & - & m + j0907@xmath202026 & 1 & 21.9 & - & 3.6 & 8.0 & 39 & n + j0915 + 2933 & 1 & 22.1 & - & 3.7 & 9.4 & 47 & n + j1008 + 0621 & 3 & 22.1 & 23.7 & 2.3 & 4.1 & 27 & m + j1243 + 3627 & 1 & 23.9 & - & 4.6 & 4.1 & 25 & m + j1253 + 5301 & 1 & 24.1 & - & 4.3 & 7.5 & 50 & m + j1427 + 2347 & 1 & 22.7 & - & 5.4 & 8.1 & 30 & m + j1440 + 0610 & 1 & 20.2 & - & 0.6 & 8.1 & 64 & n + j1542 + 6129 & 1 & 22.8 & - & 4.1 & 7.7 & 40 & m + j1624 + 5652 & 1 & 23.0 & - & 2.1 & 5.0 & 35 & n + j2241 + 4120 & 2 & 22.6 & 23.0 & 3.5 & 2.8@xmath131 & 18@xmath131 & y + j2305 + 8242 & 1 & 23.1 & - & 3.1 & 5.6 & 42 & n + as a part of our on - going spectroscopic campaign to obtain identifications and redshifts for the _ fermi_-detected blazars , some of these imaging targets have had additional spectroscopic exposure . in a few cases we obtain true spectroscopic redshifts @xmath132 . in several other cases we have lower limits on the redshift @xmath133 from detection of intervening intergalactic absorption - line systems . finally , in several cases we were able to use spectroscopic limits on absorption line - strengths ( typically ca h & k and g - band limits ) , together with the assumption of a uniform host magnitude ( as in this paper ) to place lower redshift limits @xmath134 on the bl lac @xcite . we have obtained a direct redshift measurement for only one of these targets , using keck lris spectroscopy @xcite . the value is within @xmath135 ( statistical+systematic ) of the imaging estimate : j2022 + 7611 @xmath136 ( @xmath137 ) . for three objects our imaging @xmath138 estimates are in agreement with lower limits from our own spectroscopy @xcite . the low redshift of j1427 + 2347 is particularly interesting in view of the recent veritas tev detection of this source ( ong et al . 2009 ) . j0712 + 5033 @xmath139 ( @xmath140 ) . j1427 + 2347 @xmath141(@xmath142 ) . j1440 + 0610 @xmath143(@xmath144 ) . for two objects our imaging redshift constraints are stronger than those obtained from the spectroscopic constraint on the hk line strengths : j0909 + 0200 @xmath145 ( @xmath146 , @xcite . j0049 + 0237 @xmath147 ( @xmath148 @xcite . for two objects our lower limit on the redshift from imaging is not as strong as that from our spectroscopy @xcite : j0050@xmath200929 @xmath149 ( @xmath150 ) and j2050 + 0407 @xmath151 ( @xmath152 ) . finally for two objects the imaging redshifts are in disagreement with the lower limits from limits on the hk absorption line strength : j1253 + 5301 @xmath153 ( @xmath154 ; @xmath155 higher ) and j1542 + 6129 @xmath156 ( @xmath157 ; @xmath158 higher ) . these last two disagreements may partly be attributed to differences in the host size and slit losses assumed in the spectroscopic method . however , statistically we should expect some @xmath138 estimates to be low , as malmquist bias will assure that fluctuations causing detections make the sources appear artificially close . of course it is also likely that outliers in the bl lac host magnitude distribution exist : these will cause larger disagreements . as we continue to collect spectroscopic redshifts and limits for these sources , we should be able to probe the fraction of sub - luminous hosts and test the utility of the standard candle hypothesis of @xcite . we have detected hosts for half of the bl lac objects imaged in this wiyn campaign . these detections , assuming a standard host luminosity , give redshift estimates with a median value of @xmath159 . the upper limits on host flux for the remaining objects also give reasonable redshift constraints with a median value for the bound of @xmath160 . this may be compared with the spectroscopic redshifts already obtained for the two parent bl lac populations : for the bl lacs in the early _ fermi _ blazar list one finds @xmath161 , while for the radio selected cgrabs bl lacs , we have @xmath162 ( figure [ fig : zhist ] ) . we can further quantify the difference by making kolmogorov - smirnoff comparisons of the distributions . while the lbas and cgrabs sets are quite consistent , with a ks probability of 0.58 , the imaging set differs from both with a ks probability of similarity of only @xmath163 . if we include the lower @xmath96 bounds in the distribution the probability drops to @xmath164 . the evident lack of low redshift imaging detections is doubtless a selection effect : such objects will in general show the strongest absorption features from the host and thus are most likely to provide spectroscopic redshifts . with the ground - based imaging results , we obtain a number of additional redshift estimates , with good efficiency for host detection out to @xmath165 . the distribution of our upper limits suggests that space - based imaging searches , using near - ir filters , can be productive to even higher redshift . in any event , it appears that it would be a mistake to assume that the objects in the radio and gamma - ray samples lacking spectroscopic redshifts are similar to those with spectroscopic distances . the imaging results suggest , in contrast , that these other sources are appreciably more distant and , on average have higher luminosity . it will be important to include these imaging @xmath96 estimates and bounds in attempts to measure the bl lac luminosity function . one other major conclusion of this analysis is that host searches should be at the @xmath5 band or redder . here the contrast with the relatively blue nuclear continuum flux is improved : we have seen that this allows host detections in sources with extreme core dominance . the increased sensitivity to the low surface brightness halos of the redshifted hosts should also materially improve the detection fraction at @xmath166 . again this will be important for study of the population and evolution of bl lac hosts galaxies . finally , it will be important to pursue spectroscopic confirmations of these redshift estimates , as only with such data will it be possible to test and extend the claimed uniformity of the bl lac host population and to probe its possible evolution . this work was supported in part by nasa grant nnx08aw30 g through the fermi guest investigator program . we thank the staff of the wiyn observatory for excellent support and in particular steve howell for help with optic and di harmer for the minimo set - up . we in addition thank j. windschitl for sharing notes on optic reduction techniques and mike shaw and steve healey for help with comparison to the spectroscopic bl lac sample . * j0050@xmath200929 * : this high galactic latitude field is particularly sparse , with a deficit of suitable of psf stars ( 13 ) . the bl lac host shows asymmetric structure to e , which we flag in table [ companionship ] as due to a close companion . even with a relatively good 0.59@xmath14 fwhm its nature is not clear . in fact , the fixed @xmath29 fit suggests a resolved host with significance @xmath167 , but gives an exceptionally high @xmath168 . this implies that the source is not well modeled by a de vaucouleurs host . thus we interpret the @xmath85 non - detection as a redshift lower limit , inferring a distance consistent with the @xcite spectroscopically derived bound @xmath169 . * j0110 + 6805 * : this target is at @xmath170 and so is not a member of the crates catalog . it is otherwise known as 4c+67.04 . an unresolved object of comparable brightness appears @xmath171 fwhm from the agn nucleus . we did a psf subtraction of this star and masked the peak pixels . since our psf fits are sequential , not simultaneous , the substantial overlap with the bl lac core psf introduced additional uncertainty in the bl lac host fit . we approximate this by adding the aperture photometry psf uncertainties of the neighbor star and bl lac nucleus in quadrature , using this as the nucleus statistical uncertainty and propagating through the rest of the analysis . * j0203 + 7232 * : a bright @xmath172 star appears @xmath173 from the bl lac centroid . as for j0110 + 6805 we increase the nucleus photometric error by adding in the uncertainty in the amplitude fit for this star . * j0211 + 1051 * : the detectable bl lac host extends beyond the standard fitting radius @xmath33 ; accordingly we extend the radius of the fitting region to @xmath174 , and increase the size of the sky annulus . * j0348@xmath201610 * : a very bright @xmath175 field star @xmath176 from the bl lac produced noticeable scattering wings across the fitting region . we estimated this scattered flux by radially averaging the stellar wings over regions free of background objects and then subtracted this flux from the fitting region to remove the small induced gradient . after subtraction , we applied the standard bl lac fitting . * j0743 + 1714 * : two bright ( @xmath177 ) stars are @xmath178 and @xmath179 from the bl lac . subtraction of a radially averaged template of the scattering wings , as for j0348@xmath201610 , produced a flat background acceptable for the bl lac fitting . * j0817@xmath200933 * : two bright ( @xmath180 ) stars are @xmath181 and @xmath182 from the bl lac . subtraction of a radially averaged template of the scattering wings , as for j0348@xmath201610 , produced a flat background acceptable for the bl lac fitting . * j1427 + 2347 * : the bright bl lac nucleus is highly saturated , resulting in bleeding . in this case the central exclusion was increased to diameter of 3 fwhm and the bleed trail was carefully masked . the test stars were fit with a matching central exclusion region . while the sky background annulus radius was increased to @xmath188 , the bl lac model fit was performed in the standard @xmath33 radius region . abdo , a.a . et al . , 2009a , apj , 700 , 597 abdo , a.a . et al . , 2009b , apjs , 183 , 46 abdo , a.a . et al . , 2010 , apj , subm.davis , l. 1994 , `` a reference guide to the iraf / daophot package '' falomo , r. et al . 2000 , apj , 542 , 731 fioc , m. & rocca - volmerange , b. 1997 , aa , 326 . 950 hartmann , r.c . et al . 1999 , apjs , 123 , 79 healey , s.e . 2007 , apjs , 171 , 61 healey , s.e . et al . 2008 , apjs , 175 , 97 holtzman , j. a. et al . 1995 , , 107 , 1065 massey , p. et al . 2002 , `` direct imaging manual for kitt peak '' mattox , j.r . 2001 , apjs , 135 , 155 odowd , m. & urry , c. m. 2005 , apj , 627 , 972 ong , r.a . 2009 , atel # 2084 press , w.h . 1986 , `` numerical recipes '' , ( new york : cambridge university press , 1986 ) . pursimo , t. et al . 2002 , aa , 381 , 810 sbarufatti , b. et al . 2005 , apj , 635 , 173 sbarufatti , b. et al . 2006 , aj , 132 , 1 scarpa , r. et al . 2000 , apj , 532 , 740 schlegel , d. j. , finkbeiner , d. p. & davis , m. 1998 , apj , 500 , 525 shaw , m.s . , et al . 2009 , apj , 704 , 477 shaw , m.s . , 2010 , apj , in prep . soward - emmerd et al . 2005 , apj , 626 , 95 urry , c.m . & padovani , p. , pasp , 107 , 83 urry , c.m . 2000 , apj , 532 , 816 c c c c c c c c c c c name & exposures ( s ) & seeing ( @xmath129 ) & @xmath189 & a@xmath190 & n@xmath191 & @xmath192 & @xmath193 & @xmath194 & @xmath29 ( kpc ) & @xmath96 + [ .2ex ] + [ -2ex ] j0004@xmath201148 & [ 180,180,300,300,300 ] & 0.74 & @xmath195 & 0.06 & 25 & 19.66 & @xmath10 21.02 & @xmath130 0.28 & - & @xmath10 0.86 + [ .3ex ] j0049 + 0237 & [ 300,300,300,300,300 ] & 1.04 & @xmath196 & 0.05 & 29 & 17.92 & @xmath10 20.97 & @xmath130 0.06 & - & @xmath10 0.84 + [ .3ex ] j0050@xmath200929 & [ 60,60,60,60,60,60,60,60,60 ] & 0.59 & @xmath197 & 0.06 & 13 & 15.31 & @xmath10 17.65 & @xmath130 0.12 & - & @xmath10 0.27 + [ .3ex ] j0110 + 6805@xmath198 & [ 180,180 ] & 0.76 & 5.29 & 2.38 & 61 & 14.48 & 17.90@xmath199 & 0.04 & - & 0.29@xmath200 + [ .3ex ] j0202 + 4205 & [ 180,300,300,300,300 ] & 0.46 & @xmath201 & 0.18 & 24 & 18.05 & 21.39@xmath202 & 0.05 & @xmath203 & 0.94@xmath204 + [ .3ex ] j0203 + 7232 & [ 180,180,180 ] & 0.49 & 10.4 & 1.37 & 52 & 16.21 & @xmath10 18.64 & @xmath130 0.11 & - & @xmath10 0.39 + [ .3ex ] j0211 + 1051 & [ 60,60,60,180,180,180 ] & 0.50 & @xmath205 & 0.27 & 27 & 15.32 & 16.91@xmath206 & 0.23 & @xmath207 & 0.20@xmath208 + [ .3ex ] j0219@xmath201842 & [ 180,180,180,180,180 ] & 0.76 & @xmath209 & 0.07 & 20 & 17.65 & 19.85@xmath210 & 0.14 & - & 0.60@xmath211 + j0348@xmath201610 & [ 180,180,180,180,180 ] & 0.70 & @xmath212 & 0.09 & 24 & 16.67 & 18.65@xmath213 & 0.16 & @xmath214 & 0.39@xmath215 + [ .3ex ] j0433 + 2905 & [ 180,180,300,300,300 ] & 0.46 & @xmath216 & 1.49 & 30 & 17.53 & @xmath10 19.20 & @xmath130 0.21 & - & @xmath10 0.48 + [ .3ex ] j0502 + 1338 & [ 300,300,300,300,300 ] & 0.49 & @xmath217 & 0.99 & 28 & 18.18 & 19.05@xmath218 & 0.44 & @xmath219 & 0.45@xmath220 + [ .3ex ] j0509 + 0541 & [ 60,60,60,60,60 ] & 0.53 & @xmath221 & 0.21 & 28 & 14.75 & @xmath10 18.60 & @xmath130 0.03 & - & @xmath10 0.38 + [ .3ex ] j0527 + 0331 & [ 300,300,300,300,300,300,300 ] & 0.67 & @xmath222 & 0.31 & 27 & 18.64 & 21.24@xmath223 & 0.10 & @xmath224 & 0.90@xmath225 + [ .3ex ] j0607 + 4739 & [ 180,180,180,180,180,180 ] & 0.78 & 12.9 & 0.36 & 30 & 15.53 & @xmath10 19.56 & @xmath130 0.02 & - & @xmath10 0.54 + [ .3ex ] j0610@xmath201847 & [ 180,180,180 ] & 0.83 & @xmath226 & 0.17 & 24 & 17.45 & 20.07@xmath202 & 0.09 & - & 0.64@xmath227 + [ .3ex ] j0612 + 4122 & [ 130,90,120,120,120 ] & 0.50 & @xmath228 & 0.39 & 28 & 16.67 & @xmath10 20.28 & @xmath130 0.04 & - & @xmath10 0.69 + [ .3ex ] j0625 + 4440 & [ 180,180,180,180,180 ] & 0.60 & 14.4 & 0.29 & 23 & 17.12 & @xmath10 20.60 & @xmath130 0.04 & - & @xmath10 0.77 + [ .3ex ] j0712 + 5033 & [ 240,120,120,120 ] & 0.60 & 23.9 & 0.13 & 28 & 16.42 & 19.17@xmath229 & 0.08 & @xmath230 & 0.47@xmath231 + [ .3ex ] j0814 + 6431 & [ 90,90,90,90,90 ] & 0.45 & 33.2 & 0.12 & 26 & 15.51 & 18.38@xmath232 & 0.07 & @xmath233 & 0.35@xmath234 + [ .3ex ] j0909 + 0200 & [ 180,180,180,180,180 ] & 0.52 & 31.4 & 0.06 & 28 & 17.90 & @xmath10 20.91 & @xmath235 & - & @xmath10 0.83 + [ .3ex ] j1813 + 0615 & [ 180,180,180 ] & 0.83 & 11.3 & 0.41 & 106 & 17.40 & @xmath10 20.31 & @xmath130 0.07 & - & @xmath10 0.70 + [ .3ex ] j190311 + 554044 & [ 120,120,120,120 ] & 0.53 & 20.5 & 0.12 & 36 & 16.27 & @xmath10 19.75 & @xmath130 0.04 & - & @xmath10 0.58 + [ .3ex ] j1927 + 6117 & [ 120,120,120,120,120 ] & 0.69 & 19.5 & 0.13 & 70 & 16.95 & 19.57@xmath236 & 0.09 & - & 0.54@xmath237 + [ .3ex ] j2009 + 7229 & [ 180,180,180,180,180 ] & 0.66 & 20.2 & 0.53 & 30 & 17.23 & @xmath10 20.50 & @xmath130 0.05 & - & @xmath10 0.74 + [ .3ex ] j2022 + 7611 & [ 180,180 ] & 0.64 & 21.1 & 0.47 & 37 & 16.68 & 19.26@xmath238 & 0.09 & - & 0.49@xmath239 + [ .3ex ] j2050 + 0407 & [ 180,180,180,180,180 ] & 0.83 & @xmath240 & 0.18 & 29 & 17.93 & @xmath10 20.48 & @xmath130 0.10 & - & @xmath10 0.74 + [ .3ex ] j2200 + 2137 & [ 300,300,300,300,300 ] & 0.67 & @xmath241 & 0.20 & 32 & 18.50 & @xmath10 20.64 & @xmath130 0.14 & - & @xmath10 0.77 + [ .3ex ] j2241 + 4120 & [ 300,300,300,300,300 ] & 0.70 & @xmath242 & 0.47 & 31 & 17.68 & 19.46@xmath243 & 0.19 & @xmath244 & 0.52@xmath237 + [ .3ex ] j224356 + 202101 & [ 90,60,60 ] & 0.56 & @xmath245 & 0.09 & 39 & 15.10 & @xmath10 18.64 & @xmath130 0.04 & - & @xmath10 0.39 + [ .3ex ] j2305 + 8242 & [ 300,300,300 ] & 0.94 & @xmath246 & 0.46 & 31 & 19.99 & 19.97@xmath247 & 1.03 & - & 0.62@xmath248 + [ .3ex ] j2346 + 8007 & [ 300,300,300,300 ] & 1.01 & @xmath249 & 0.45 & 68 & 17.90 & @xmath10 21.07 & @xmath130 0.14 & - & @xmath10 0.87 + [ .7ex ] c c c c c c c c c c c name & exposures ( s ) & seeing ( @xmath129 ) & @xmath189 & a@xmath190 & n@xmath191 & @xmath250 & @xmath193 & @xmath194 & @xmath29 ( kpc ) & @xmath96 + [ .2ex ] + [ -2ex ] j024330 + 712012 & [ 300,300 ] & 1.09 & 10.4 & 1.50 & 51 & 16.95 & @xmath10 18.70 & @xmath130 0.20 & - & @xmath10 0.40 + [ .3ex ] j065047 + 250304 & [ 60,300,300,300,300 ] & 0.68 & 11.0 & 0.19 & 31 & 16.14 & 19.03@xmath236 & 0.07 & - & @xmath251 + [ .3ex ] j0743 + 1714 & [ 300,300,300 ] & 0.65 & 19.0 & 0.07 & 41 & 18.43 & @xmath10 20.82 & @xmath130 0.11 & - & @xmath10 0.81 + [ .3ex ] j0817@xmath200933 & [ 300,300,300,300 ] & 0.78 & 14.3 & 0.20 & 42 & 16.61 & 20.37@xmath238 & 0.03 & @xmath252 & 0.71@xmath253 + [ .3ex ] j083543 + 093659 & [ 300,300,300 ] & 0.83 & 27.5 & 0.10 & 30 & 17.93 & @xmath10 18.89 & @xmath130 0.41 & - & @xmath10 0.42 + [ .3ex ] j0907@xmath202026 & [ 300,300,300 ] & 0.97 & 18.0 & 0.34 & 35 & 14.46 & 18.32@xmath254 & 0.03 & @xmath255 & @xmath256 + [ .3ex ] j091553 + 293326 & [ 300,300,300 ] & 0.68 & 42.9 & 0.05 & 23 & 15.71 & 18.45@xmath257 & 0.08 & @xmath258 & @xmath259 + [ .3ex ] j095301@xmath20084034 & [ 300,300,300 ] & 0.78 & 33.9 & 0.92 & 34 & 15.29 & @xmath10 18.71 & @xmath130 0.04 & - & @xmath10 0.40 + [ .3ex ] j1008 + 0621 & [ 300,300,300 ] & 0.73 & 46.0 & 0.05 & 26 & 17.68 & 19.80@xmath260 & 0.14 & @xmath261 & @xmath262 + [ .3ex ] j103742 + 571158 & [ 300,300,300,300,300 ] & 0.76 & 51.8 & 0.01 & 20 & 15.97 & @xmath10 19.95 & @xmath130 0.03 & - & @xmath10 0.62 + [ .3ex ] j105912@xmath20113424 & [ 300,300,300 ] & 0.85 & 42.7 & 0.05 & 21 & 15.86 & @xmath10 19.08 & @xmath130 0.05 & - & @xmath10 0.46 + [ .3ex ] j124313 + 362755 & [ 300,300,300,300 ] & 0.86 & 80.5 & 0.02 & 26 & 15.72 & 19.32@xmath263 & 0.04 & - & @xmath264 + [ .3ex ] j125311 + 530113 & [ 300,300,300,300,300 ] & 0.93 & 64.1 & 0.02 & 22 & 16.83 & 19.83@xmath218 & 0.06 & @xmath265 & @xmath266 + [ .3ex ] j142700 + 234802 & [ 300,300,300 ] & 0.71 & 68.2 & 0.11 & 28 & 13.99 & 17.30@xmath267 & 0.05 & - & @xmath268 + [ .3ex ] j144052 + 061033 & [ 300,300 ] & 0.98 & 56.6 & 0.07 & 26 & 16.89 & 19.59@xmath236 & 0.08 & @xmath269 & @xmath270 + [ .3ex ] j154225 + 612950 & [ 300,300,300 ] & 1.02 & 45.4 & 0.03 & 33 & 14.98 & 18.68@xmath271 & 0.03 & - & @xmath272 + [ .3ex ] j1624 + 5652 & [ 300,300,300 ] & 0.89 & 42.3 & 0.02 & 33 & 17.93 & 20.07@xmath271 & 0.14 & - & @xmath273 + [ .3ex ] j1749 + 4321 & [ 300,300,300 ] & 1.09 & 29.2 & 0.07 & 30 & 17.41 & @xmath10 20.73 & @xmath130 0.05 & - & @xmath10 0.79 + [ .7ex ]

crum s theorem in one - dimensional quantum mechanics asserts the existence of an associated hamiltonian system for any given hamiltonian with the complete set of eigenvalues and eigenfunctions . the associated system is iso - spectral to the original one except for the lowest energy state , which is deleted . a modification due to krein - adler provides algebraic construction of a new complete hamiltonian system by deleting a finite number of energy levels . here we present a discrete version of the modification based on the crum s theorem for the ` discrete ' quantum mechanics developed by two of the present authors . yukawa institute kyoto + dpsu-10 - 1 + yitp-10 - 15 + * modification of crum s theorem for ` discrete ' quantum mechanics + + * * leonor garca - gutirrez , satoru odake and ryu sasaki * yukawa institute for theoretical physics , + kyoto university , kyoto 606 - 8502 , japan + department of physics , shinshu university , + matsumoto 390 - 8621 , japan

1004.0289

crum s seminal paper of 1955 @xcite has played an essential role in elucidating the structure of one - dimensional quantum mechanical systems in general and exactly solvable ones , in particular . throughout this paper , we mean ` exact solvability ' in the schrdinger picture , namely a quantum system is exactly solved when the complete set of eigenvalues and eigenfunctions are known . many exactly solvable quantum mechanical hamiltonians were constructed and investigated by combining shape invariance @xcite and crum s theorem @xcite , or the factorisation method @xcite or the method of the so - called supersymmetric quantum mechanics @xcite . it is interesting to note that most of these shape invariant systems are also solvable in the heisenberg picture @xcite . exactly solvable quantum mechanical systems of one and many degrees of freedom are not only important in their own right but also have fundamental applications in various disciplines of physics / mathematics , _ e.g. _ the fokker - planck equations @xcite and their discretised version , birth and death processes @xcite , to name a few . shape invariance is a sufficient condition for exactly solvable quantum mechanical systems . the number of shape invariant systems , however , was quite limited ; only about a dozen until the recent discovery @xcite of the several types of infinitely many shape invariant hamiltonians @xcite which led to the infinitely many exceptional laguerre , jacobi , wilson and askey - wilson polynomials . many methods were proposed to derive exactly solvable but non - shape invariant quantum mechanical systems from known shape invariant ones @xcite . ( we apologise to those whose work we have missed . ) among them krein - adler s modification @xcite of crum s theorem is the most comprehensive way to generate infinitely many variants of exactly solvable hamiltonians and their eigenfunctions , starting from an exactly solvable one . the derived system is _ iso - spectral _ with the original one except that a finite number of energy levels are deleted . if the original system has polynomial eigenfunctions , as is usually the case , the derived systems have also polynomial eigenfunctions . by construction , these polynomials constitute a complete set of orthogonal functions . but they do not qualify to be called _ exceptional orthogonal polynomials _ @xcite since some members of certain degrees are _ missing _ due to the _ deletion_. the discrete quantum mechanics is a deformation of the ordinary quantum mechanics in the sense that the schrdinger equation is a second order _ difference _ equation instead of differential . in the formulation of odake and sasaki @xcite , the algebraic and analytical structure of quantum mechanics as well as shape invariance and exact solvability are retained in the discrete version . the eigenfunctions of the exactly solvable one - dimensional discrete quantum mechanics are the askey - scheme of hypergeometric orthogonal polynomials and their @xmath2-versions @xcite , _ e.g. _ the continuous hahn , the wilson and the askey - wilson polynomials . these examples are all shape invariant and they are also solvable in the heisenberg picture @xcite . the dynamical symmetry algebra of these algebras are the askey - wilson algebras @xcite and degenerations , which contain the @xmath2-oscillator algebra @xcite . the discrete version of crum s theorem was also established recently @xcite . in this paper we present the discrete quantum mechanics version of adler s modification @xcite of crum s theorem . it allows to generate an infinite variety of exactly solvable discrete quantum hamiltonian systems . the insight obtained from crum s theorems and their modification , in the ordinary and the discrete quantum mechanics , is essential for the recent derivation of the infinite numbers of shape invariant systems and the new exceptional orthogonal polynomials @xcite . we will discuss the main results , the specialisation to the cases of polynomial eigenfunctions and simplest example for various exactly solvable cases ; first for the ordinary quantum mechanics and then for the discrete versions . the reason is two - fold ; firstly to introduce appropriate notion and notation in the familiar cases of the ordinary quantum mechanics . secondly we choose to reveal the underlying logical processes which are not easy to fathom in adler s paper @xcite or in crum s original article @xcite . as seen in the subsequent sections , the logical structures of the associated hamiltonian systems and their modification by _ deletion _ of energy levels are shared by the ordinary and the discrete quantum mechanics . this paper is organised as follows . in section two , adler s modification of crum s theorem is recapitulated in appropriate notation for our purposes . the specialisation to the cases of polynomial eigenfunctions is discussed in some detail . section three provides the discrete quantum mechanics version of the modification of crum s theorem . again the specialisation to the cases of polynomial eigenfunctions is mentioned . appendix gives the simplest examples of the modified hamiltonian systems obtained by deleting the lowest lying @xmath3 excited states for various exactly solvable hamiltonians . appendix a provides three examples from the ordinary quantum mechanics , the harmonic oscillator , the radial oscillator , the darboux - pschl - teller potential . appendix b is for the four examples from the discrete quantum mechanics , the hamiltonians of the meixner - pollaczek , the continuous hahn , the wilson and the askey - wilson polynomials @xcite , which are known to reduce to the hermite , the laguerre and the jacobi polynomials in certain limits , respectively . let us start with a generic one - dimensional quantum mechanical ( qm ) system having discrete semi - infinite energy levels only : @xmath4 here we have chosen the constant part of the hamiltonian so that the groundstate energy is zero . then the hamiltonian is _ positive semi - definite _ and can be factorised , @xmath5 here a real and smooth function @xmath6 is called a _ prepotential _ and it parametrises the groundstate wavefunction @xmath7 , which has _ no node _ and can be chosen real and positive : @xmath8 it is trivial to verify @xmath9 in one dimension all the energy levels are non - degenerate . by construction all the eigenfunctions are square - integrable and orthogonal with each other and form a complete basis of the hilbert space : @xmath10 where @xmath11 is the set of non - negative integers @xmath12 . it is well - known that the @xmath13-th excited wavefunction @xmath14 has @xmath13 nodes in the interior . for simplicity we choose all the eigenfunctions to be real . a few exactly solvable examples are given in appendix . let us choose a set of @xmath3 distinct non - negative integers @xmath15 , satisfying the condition @xmath16 this condition means that the set @xmath17 consists of several clusters , each containing an _ even number _ of _ contiguous _ integers @xmath18 where @xmath19 . if @xmath20 for the lowest lying cluster , it could contain an even or odd number of contiguous integers . the set @xmath17 specifies the energy levels to be _ deleted_. this simply reflects the fact that no singularity arises when two neighbouring levels are deleted . in the ordinary qm , the zeros of the two neighbouring eigenfunctions @xmath21 and @xmath22 interlace with each other . this fact is essential for the non - singularity of the potential after deletion . see adler s paper @xcite for a proof . the situation is essentially the same in the discrete qm . however , in the dqm to be discussed in the subsequent section , due to the lack of general theorem , the interlacing of the zeros of the two neighbouring eigenfunctions @xmath21 and @xmath22 must be verified for each specific hamiltonian . deleting an arbitrary number of contiguous energy levels starting from the groundstate ( @xmath20 ) is achieved by the original crum s theorem @xcite . next we will construct hamiltonian systems corresponding to the successive deletions @xmath23 ( and @xmath24 , @xmath25 , etc . ) step by step , algebraically . it should be noted that some quantities in the intermediate steps could be singular . for given @xmath26 the first hamiltonian @xmath27 can be expressed in two different ways : @xmath28 unless @xmath29 , @xmath30 is singular due to the zeros of @xmath31 . it is very important to note that @xmath32 in is a ` formal adjoint ' of @xmath33 . we stick to this notation , since the algebraic structure of various expressions appearing in the deletion processes , from to , are best described by using the ` formal adjoint ' . these define a new hamiltonian system @xmath34 with the ` eigenfunctions ' @xmath35 note that the energy level @xmath26 is now deleted , @xmath36 , from the set of ` eigenfunctions ' @xmath37 of the new hamiltonian @xmath38 . suppose we have determined @xmath39 and @xmath40 with @xmath41 deletions . they have the following properties @xmath42 we have also @xmath43 next we will define a new hamiltonian system with one more deletion of the level @xmath44 . we can show @xmath45 these determine a new hamiltonian system with @xmath46 deletions : @xmath47 after deleting all the @xmath48 energy levels , the resulting hamiltonian system @xmath49 , @xmath50 , etc has the following form : @xmath51 now that @xmath52 has the lowest energy level @xmath53 : @xmath54 with the groundstate wavefunction @xmath55 @xmath56 as usual the hamiltonian system can be expressed simply in terms of the groundstate wavefunction @xmath55 , which we will denote by new symbols @xmath57 , @xmath58 , etc : @xmath59 as shown by krein - adler @xcite , the results can be expressed succinctly : @xmath60(x ) } { \text{w}[\phi_{d_1},\phi_{d_2},\ldots,\phi_{d_\ell}](x ) } \quad(n\in\mathbb{z}_+\backslash\mathcal{d } ) , \label{qmphib1bsn}\\ & \bar{u}(x)\equiv u_{d_1,\ldots , d_\ell}(x)=u(x ) -2\frac{d^2}{dx^2}\bigl ( \log\text{w}\,[\phi_{d_1},\phi_{d_2},\ldots,\phi_{d_\ell}](x)\bigr ) \quad(\ell\geq 0 ) , \label{ub1bs}\end{aligned}\ ] ] in which the wronskian determinant is defined by @xmath61(x ) { \stackrel{\text{def}}{=}}\det\bigl(\frac{d^{j-1}f_k(x)}{dx^{j-1}}\bigr)_{1\leq j , k\leq n}.\ ] ] for @xmath62 , we set @xmath63(x)=1 $ ] . in deriving the determinant formulas and use is made of the properties of the wronskian @xmath64(x ) = g(x)^n\text{w}[f_1,f_2,\ldots , f_n](x ) , \label{wformula1}\\ & \text{w}\bigl[\text{w}[f_1,f_2,\ldots , f_n , g ] , \text{w}[f_1,f_2,\ldots , f_n , h]\,\bigr](x){\nonumber\\}&=\text{w}[f_1,f_2,\ldots , f_n](x)\ , \text{w}[f_1,f_2,\ldots , f_n , g , h](x ) \qquad(n\geq 0 ) . \label{wformula2}\end{aligned}\ ] ] let us note that @xmath65 and @xmath66 are symmetric with respect to @xmath67 , and thus @xmath68 is independent of the order of @xmath69 . let us state adler s theorem again ; _ if the set of deleted energy levels @xmath67 satisfy the condition , the hamiltonian @xmath70 with is well - defined and hermitian , and its complete set of eigenfunctions are given by . _ crum s theorem corresponds to the choice @xmath71 , and the resulting lowest energy level is @xmath72 and there is no deleted energy levels above the new groundstate . in this subsection we consider the typical case of shape invariant systems in which the eigenfunctions consist of the orthogonal polynomials @xmath73 : @xmath74 in which @xmath75 is called the _ sinusoidal coordinate_. as shown in detail in the examples in appendix a , @xmath76 for the harmonic oscillator ( the hermite polynomials ) , @xmath77 for the radial oscillator ( the laguerre polynomials ) and @xmath78 for the darboux - pschl - teller potential ( the jacobi polynomials ) . the groundstate wavefunction @xmath7 provides the orthogonality weight function @xmath79 in this case , the modification of crum s theorem produces the eigenfunctions @xmath80 which again consist of polynomials in @xmath75 . by using and @xmath81(x ) = \bigl(\tfrac{d\eta(x)}{dx}\bigr)^{\frac12n(n-1 ) } \text{w}[f_1,f_2,\ldots , f_n](\eta(x ) ) , \label{wformula3}\ ] ] we obtain a simple expression of the eigenfunctions @xmath82(\eta(x ) ) } { \text{w}[p_{d_1},p_{d_2},\ldots , p_{d_\ell}](\eta(x))}.\ ] ] this simply means that the resulting eigenfunctions are again polynomials in @xmath75 : @xmath83(\eta(x))},\quad \mathcal{p}_n(\eta){\stackrel{\text{def}}{=}}\text{w}[p_{d_1},p_{d_2},\ldots , p_{d_\ell},p_n](\eta ) , \label{polyphi0}\end{aligned}\ ] ] satisfying the orthogonality relation @xmath84 let us emphasise that @xmath13 is not the _ degree _ in @xmath85 and by construction , @xmath3 members are missing : @xmath86 . therefore these polynomials can not be called _ exceptional orthogonal polynomials _ @xcite . let us begin with a few general remarks on the one - dimensional discrete qm with pure imaginary shifts . see @xcite for the general introduction to the discrete quantum mechanics with pure imaginary shifts and @xcite for the crum s theorem in the discrete qm . in the discrete qm , the dynamical variables are , as in the ordinary qm , the coordinate @xmath87 , which takes value in an infinite or a semi - infinite or a finite range of the real axis and the canonical momentum @xmath88 , which is realised as a differential operator @xmath89 . since the momentum operator appears in exponentiated forms @xmath90 , @xmath91 , in a hamiltonian , it causes finite pure imaginary shifts in the wavefunction @xmath92 . this requires the wavefunction as well as other functions appearing in the hamiltonian to be _ analytic _ in @xmath87 within a certain complex domain including the physical region of the coordinate . let us introduce the @xmath93-operation on an analytic function , @xmath94 . if @xmath95 , @xmath96 , then @xmath97 , in which @xmath98 is the complex conjugation of @xmath99 . obviously @xmath100 and @xmath101 . if a function satisfies @xmath102 , we call it a ` real ' function , for it takes real values on the real line . the starting point is again a generic one dimensional discrete quantum mechanical hamiltonian with discrete semi - infinite energy levels only . again we assume that the groundstate energy is chosen to be zero @xmath103 , so that the hamiltonian is positive semi - definite . the generic factorised hamiltonian reads @xcite @xmath104 since the @xmath93-operation for @xmath105 , @xmath106 and @xmath107 satisfies @xmath108 they map a ` real ' function to a ` real ' function @xmath109 by specifying the function @xmath110 , various explicit examples are obtained @xcite . a few exactly solvable examples are given in appendix . the corresponding schrdinger equation @xmath111 is a _ difference equation _ @xmath112 instead of differential in the ordinary qm . although this equation looks rather complicated , the equation for the polynomial eigenfunctions has a familiar form of difference equations . again the groundstate wavefunction @xmath7 is determined as a zero mode of @xmath113 , @xmath114 ( @xmath115 @xmath116 ) , namely , @xmath117 this dictates how the ` phase ' of the potential function @xmath118 is related to that of the groundstate wavefunction @xmath119 . here we also assume that the groundstate wavefunction @xmath7 has no node and chosen to be real and positive for real @xmath87 . due to the lack of generic theorems in the theory of difference equations , let us assume that all the energy levels are non - degenerate and that all the eigenfunctions are square - integrable and orthogonal with each other and form a complete basis of the hilbert space : @xmath120 in most explicit examples these statements can be verified straightforwardly . for simplicity we choose all the eigenfunctions to be real on the real axis @xmath121 , which is made possible by . the formulation of the modified crum s theorem in the discrete quantum mechanics goes almost parallel to that in the ordinary quantum mechanics . again the presentation is purely algebraic . let us note that various quantities in intermediate steps might have singularities and hamiltonians might not be hermitian . we choose a set of distinct non - negative integers @xmath122 , satisfying the condition as before . first let us note that the hamiltonian @xmath27 can be rewritten by incorporating the level @xmath26 as : @xmath123 these define a new hamiltonian system @xmath124 note that the energy level @xmath26 is now deleted , @xmath36 , from the set of ` eigenfunctions ' @xmath37 of the new hamiltonian @xmath38 . suppose we have determined @xmath39 and @xmath40 with @xmath41 deletions . they have the following properties @xmath125 where @xmath126 . we have also @xmath43 next we will define a new hamiltonian system with one more deletion of the level @xmath44 . we can show the following : @xmath127 these determine a new hamiltonian system with @xmath46 deletions : @xmath128 where @xmath129 . after deleting all the @xmath48 energy levels , the resulting hamiltonian system @xmath49 , @xmath50 , etc has the following form : @xmath130 now that @xmath52 has the lowest energy level @xmath53 : @xmath131 with the groundstate wavefunction @xmath55 @xmath132 then the hamiltonian system can be expressed simply in terms of the groundstate wavefunction @xmath55 , which we will denote by new symbols @xmath133 , @xmath134 , etc : @xmath135 the discrete counterpart of the determinant formulas requires a deformation of the wronskian , the casorati determinant , which has a good limiting property : @xmath136(x ) { \stackrel{\text{def}}{=}}i^{\frac12n(n-1 ) } \det\bigl(f_k(x+i\tfrac{n+1 - 2j}{2}\gamma)\bigr)_{1\leq j , k\leq n},\\ & \lim_{\gamma\to 0}\gamma^{-\frac12n(n-1 ) } \text{w}_{\gamma}[f_1,f_2,\ldots , f_n](x ) = \text{w}\,[f_1,f_2,\ldots , f_n](x),\end{aligned}\ ] ] ( for @xmath62 , we set @xmath137(x)=1 $ ] . ) . it satisfies @xmath136^*(x ) = \text{w}_{\gamma}[f_1^*,\ldots , f_n^*](x),\\ & \text{w}_{\gamma}[gf_1,gf_2,\ldots , gf_n ] = \prod_{j=1}^ng(x+i\tfrac{n+1 - 2j}{2}\gamma)\cdot \text{w}_{\gamma}[f_1,f_2,\ldots , f_n](x ) , \label{dwformula1}\\ & \text{w}_{\gamma}\bigl[\text{w}_{\gamma}[f_1,f_2,\ldots , f_n , g ] , \text{w}_{\gamma}[f_1,f_2,\ldots , f_n , h]\,\bigr](x){\nonumber\\}&=\text{w}_{\gamma}[f_1,f_2,\ldots , f_n](x)\ , \text{w}_{\gamma}[f_1,f_2,\ldots , f_n , g , h](x ) \quad(n\geq 0 ) . \label{dwformula2}\end{aligned}\ ] ] by using the casorati determinant we obtain ( @xmath138 ) @xmath139(x ) } { \text{w}_{\gamma}\,[\phi_{d_1},\ldots,\phi_{d_\ell } ] ( x - i\tfrac{\gamma}{2 } ) } , \label{phib1bsn}\\ & \bar{v}(x)\equiv v_{d_1\,\ldots\,d_\ell\,\mu}(x)= \sqrt{v(x - i\tfrac{\ell}{2}\gamma)v^*(x - i\tfrac{\ell+2}{2}\gamma)}{\nonumber\\}&\phantom{v_{d_1\,\ldots\,d_\ell\,\mu}(x)=}\times \frac{\text{w}_{\gamma}\,[\phi_{d_1},\ldots,\phi_{d_\ell } ] ( x+i\tfrac{\gamma}{2 } ) } { \text{w}_{\gamma}\,[\phi_{d_1},\ldots,\phi_{d_\ell } ] ( x - i\tfrac{\gamma}{2})}\ , \frac{\text{w}_{\gamma}\,[\phi_{d_1},\ldots,\phi_{d_{\ell}},\phi_{\mu } ] ( x - i\gamma ) } { \text{w}_{\gamma}\,[\phi_{d_1},\ldots,\phi_{d_{\ell}},\phi_{\mu}](x)}. \label{vb1bs+1}\end{aligned}\ ] ] we also have ( @xmath138 ) @xmath140 ( x - i\tfrac{\gamma}{2 } ) } { \text{w}_{\gamma}\,[\phi_{d_1},\ldots,\phi_{d_\ell } ] ( x+i\tfrac{\gamma}{2})}. \label{prodv}\ ] ] therefore @xmath141 and @xmath66 are symmetric with respect to @xmath67 , and @xmath142 is independent of the order of @xmath69 . let us state the discrete qm analogue of adler s theorem ; _ if the set of deleted energy levels @xmath143 satisfy the condition , the modified hamiltonian is given by @xmath144 with the potential function given by and its eigenfunctions are given by _ . the discrete qm version of crum s theorem @xcite corresponds to the choice @xmath71 and the new groundstate is at the level @xmath72 and there is no vacant energy level above that . due to the lack of generic theorems in the theory of difference equations , the hermiticity of the resulting hamiltonian @xmath133 and the non - singularity of the eigenfunctions @xmath145 can not be proved categorically for the discrete qm , even when the condition is satisfied by the deleted levels . see appendix a of @xcite for a detailed discussion of the self - adjointness of the hamiltonians in discrete qm . it should be stressed that in most practical cases , in particular , in the cases of polynomial eigenfunctions , the hermiticity of the hamiltonian @xmath133 and non - singularity of the eigenfunctions @xmath80 are satisfied . in this subsection we consider the typical case of shape invariant systems in which the eigenfunctions consist of the orthogonal polynomials @xmath73 : @xmath146 in which @xmath75 is called the _ sinusoidal coordinate _ , see . the groundstate wavefunction @xmath7 provides the orthogonality weight function @xmath147 the difference equation for @xmath73 looks much simpler than the schrdinger equation : @xmath148 for the explicit forms of @xmath110 , see for example , these are the equations that determine the hypergeometric orthogonal polynomials , _ e.g. _ the meixner - pollaczek ( mp ) , the continuous hahn ( ch ) , the wilson ( w ) and the askey - wilson ( aw ) polynomials . in fact , the above form of the _ difference equation _ is independent of the fact that @xmath149 is a polynomial or not . it is obtained simply by the similarity transformation of the hamiltonian in terms of the groundstate wavefunction @xmath7 : @xmath150 in the case of polynomial eigenfunctions , the modification of crum s theorem produces the eigenfunctions @xmath80 which again consist of polynomials in @xmath75 . by using the property we have @xmath151(x ) = \prod_{j=1}^{\ell } \phi_0\bigl(x+i\tfrac{\ell+1 - 2j}{2}\gamma\bigr)\cdot \text{w}_{\gamma}[\check{p}_1,\ldots,\check{p}_{\ell}](x),\\ & \text{w}_{\gamma}[\phi_1,\ldots,\phi_{\ell},\phi_n](x ) = \prod_{j=1}^{\ell+1 } \phi_0\bigl(x+i\tfrac{\ell+2 - 2j}{2}\gamma\bigr)\cdot \text{w}_{\gamma}[\check{p}_1,\ldots,\check{p}_{\ell},\check{p}_n](x),\end{aligned}\ ] ] where @xmath152 . corresponding to the formula in the ordinary qm , we have @xmath153(x ) = \varphi_n(x)\times\bigl(\text{polynomial in $ \eta(x)$}\bigr),\end{aligned}\ ] ] in which @xmath154 is defined in ( and @xmath155 is defined in ) . these simply mean that the resulting eigenfunctions @xmath80 are again polynomials in @xmath75 : @xmath156 in which @xmath157 and @xmath158 are certain polynomials in @xmath75 defined by @xmath159 = \varphi_{\ell+1}(x)\times\mathcal{p}_n(\eta(x)),\quad \text{w}_{\gamma}[\check{p}_{d_1},\ldots,\check{p}_{d_\ell } ] = \varphi_{\ell}(x)\times\mathcal{q}(\eta(x ) ) . \label{calpqdef}\ ] ] the polynomials @xmath160 form a complete basis of the hilbert space and satisfy the orthogonality relations @xmath84 let us emphasise that @xmath13 is the level of the original eigenfunction and not the _ degree _ in @xmath85 . the degree of @xmath161 depends on the set @xmath17 , and it can be calculated explicitly from . by construction @xmath3 members are missing : @xmath86 . therefore these polynomials can not be called _ exceptional orthogonal polynomials _ @xcite . theory of exactly solvable discrete qm is less developed than that of the ordinary qm . up to date , the known exactly solvable discrete quantum systems are all shape invariant @xcite and in one to one correspondence with the known ( @xmath2)-hypergeometric orthogonal polynomials @xcite@xcite . small progress was made in this direction @xcite by introducing several new sinusoidal coordinates for the construction of new types of exactly solvable hamiltonians . roughly speaking , this approach attempts to create the discrete analogues of various morse type potentials and the soliton potentials . in this paper we pursue another direction ; to construct infinitely many exactly solvable quantum systems by _ deforming _ the known exactly solvable one . in the ordinary qm , the modification of crum s theorem @xcite due to krein - adler @xcite allows to produce an essentially iso - spectral hamiltonian by deleting a finite number of energy levels from the original system . the set of deleted level must satisfy certain condition , but there are infinitely many possible deletions leading to infinitely many exactly solvable systems starting from a known one . the discrete analogue of adler s modification is presented in this paper in parallel with the original version , since the algebraic structure is common . we also comment on the practical cases when the eigenfunctions consist of orthogonal polynomials . the eigenfunctions of the resulting system also consist of orthogonal polynomials . but certain members of these polynomials are missing due to the deletion . very special and simple examples , in which all the excited states from the first to the @xmath3-th are deleted ( see fig.2 ) , are presented explicitly in appendix . as will be commented shortly , these examples were instrumental for the discovery of the infinitely many shape invariant systems and the corresponding infinitely many exceptional orthogonal polynomials @xcite . in the ordinary qm , the corresponding prepotential has a very simple form : @xmath162 to be compared with the prepotential for the hamiltonian of the @xmath3-th exceptional orthogonal polynomial ( 14 ) and ( 28 ) of @xcite : @xmath163 in the discrete qm , the corresponding formula is @xmath164 to be compared with the corresponding formula for the hamiltonian of the @xmath3-th exceptional orthogonal polynomial ( 30 ) of @xcite : @xmath165 the addition ( multiplication ) of the deforming polynomial with the shifted parameters @xmath166 would achieve the shape invariance . since the harmonic oscillator has no shiftable parameter , we have @xmath167 . this also ` explains ' non - existence of exceptional hermite polynomials . in contrast to the hermite polynomial , the continuous hahn polynomial has four real parameters . we can construct the corresponding exceptional continuous hahn polynomials with three real parameters , which will be reported elsewhere . the actual function forms of the deforming polynomial @xmath168 in appendix are not the same as those for the exceptional orthogonal polynomials . for the ordinary qm examples , see vs. ( 13 ) and ( 27 ) in @xcite and for the discrete qm examples , see vs. ( 64 ) and ( 78 ) in @xcite . but they share some interesting features . before closing this section , let us remark that the present modification of crum s theorem is applicable to the hamiltonian systems of various species of the infinite family of exceptional orthogonal polynomials @xcite , as well as to those of the classical orthogonal polynomials including the wilson and the askey - wilson polynomials . we thank reido kobayashi for useful comments and discussion . r.s . is supported in part by grant - in - aid for scientific research from the ministry of education , culture , sports , science and technology ( mext ) , no.19540179 . thanks mext for the research studentship . * appendix * in appendix we present very special and simple examples of an application of adler s theorem , in which the eigenstates @xmath169 are deleted . in other words , @xmath170@xmath171 , that is , the modified groundstate level is the same as that of the original theory @xmath172 . the situation is illustrated in fig.2 , which should be compared with fig.1 , depicting the generic case discussed in sections [ genordqm ] and [ gendqm ] . figure1 : generic case figure2 : special case the black circles denote the energy levels , whereas the white circles denote _ deleted _ energy levels . we write @xmath173 , @xmath174 , @xmath175 , @xmath176 etc . as @xmath177 , @xmath178 , @xmath179 , @xmath180 etc . this hamiltonian @xmath181 is non - singular for even @xmath3 but may be singular for odd @xmath3 . since algebraic formulas such as the wronskians and casoratians are valid for even and odd @xmath3 , we present various formulas without restricting to the even @xmath3 . the original systems are shape invariant but the @xmath182-deleted systems @xmath177 are not . the rightmost vertical line in fig.2 corresponds to the hamiltonian system @xmath183 , which is shape invariant and it is obtained from @xmath177 by one more step of crum s method . this study helped us to find the new shape invariant systems and exceptional orthogonal polynomials @xcite . here we apply adler s theorem to the harmonic oscillator , the radial oscillator and the darboux - pschl - teller potential , whose eigenfunctions are described by the classical orthogonal polynomials . that is , the hermite , laguerre and jacobi polynomials , to be abbreviated as h , l and j , respectively . these original systems are shape invariant , meaning a very special form of parameter dependence , , . here we display the parameter dependence explicitly by @xmath184 , which represents the set of the parameters . here we summarise various properties of the original hamiltonian systems to be compared with the specially modified systems to be presented in [ sec : a2 ] . let us start with the hamiltonians , schrdinger equations and eigenfunctions ( @xmath185 ) : @xmath186 here @xmath75 is the sinusoidal coordinate , @xmath187 is the prepotential and @xmath188 is the @xmath13-th energy eigenvalue : @xmath189 -\frac12x^2+g\log x&:\text{l}\\[1pt ] g\log\sin x+h\log\cos x&:\text{j } \end{array}\right.\!\ ! , \quad \mathcal{e}_n({\boldsymbol}{\lambda}){\stackrel{\text{def}}{=}}\left\ { \begin{array}{ll } 2n&:\text{h}\\ 4n&:\text{l}\\ 4n(n+g+h)&:\text{j } \end{array}\right.\!\!.\end{aligned}\ ] ] the eigenfunction consists of an orthogonal polynomial @xmath190 , a polynomial of degree @xmath13 in @xmath85 , ( @xmath191 for @xmath192 ) : @xmath193 l_n^{(g-\frac12)}(\eta)&:\text{l}\\[1pt ] p_n^{(g-\frac12,h-\frac12)}(\eta)&:\text{j } \end{array}\right.\!\ ! , \quad c_n({\boldsymbol}{\lambda}){\stackrel{\text{def}}{=}}\left\ { \begin{array}{ll } 2^n&:\text{h}\\ \frac{(-1)^n}{n!}&:\text{l}\\[1pt ] \frac{(n+g+h)_n}{2^nn!}&:\text{j } \end{array}\right.\!\!,\end{aligned}\ ] ] in which @xmath194 is the pochhammer symbol . shape invariance means @xmath195 or equivalently , @xmath196 the action of @xmath197 and @xmath198 on the eigenfunction is : @xmath199 here the coefficients @xmath200 and @xmath201 are the factors of @xmath188 : @xmath202 the forward and backward shift operators , @xmath203 and @xmath204 , are defined by : @xmath205 and their action on the polynomial is : @xmath206 note that @xmath203 and @xmath204 can also be expressed in terms of @xmath85 only @xcite . the orthogonality reads @xmath207 \frac{1}{2\,n!}\gamma(n+g+\frac12)&:\text{l}\\[4pt ] \frac{\gamma(n+g+\frac12)\gamma(n+h+\frac12 ) } { 2\,n!(2n+g+h)\gamma(n+g+h)}&:\text{j } \end{array}\right.\!\!.\end{aligned}\ ] ] the prepotential of the modified system is obtained from up to an additive constant : @xmath208 it is a polynomial ( @xmath209 ) deformation of the shape invariant one @xmath210 . note that the normalization of @xmath168 does not affect the hamiltonian . the explicit forms of the deforming polynomial @xmath211 will be given in . for even @xmath3 , the polynomial @xmath209 has no zero in the range of @xmath87 and the modified hamiltonian system is hermitian , which reads : @xmath212 this system is not shape invariant . as mentioned in section [ summary ] , the above form of the deformed prepotential is closely related to that of the exceptional laguerre and jacobi polynomials . a degree @xmath3 polynomial in @xmath85 , @xmath211 is defined by @xmath213(\eta;{\boldsymbol}{\lambda } ) { \stackrel{\text{def}}{=}}\prod_{k=1}^{\ell}k!\,c_k({\boldsymbol}{\lambda})\cdot \xi_{\ell}(\eta;{\boldsymbol}{\lambda}),\ ] ] and the explicit forms are : @xmath214 the eigenfunctions are @xmath215(\eta;{\boldsymbol}{\lambda } ) { \stackrel{\text{def}}{=}}\prod_{k=1}^{\ell}k!\,c_k({\boldsymbol}{\lambda})\cdot(-1)^{\ell } p_{\ell , n}(\eta;{\boldsymbol}{\lambda})\quad \bigl(\rightarrow p_{\ell,0}(\eta;{\boldsymbol}{\lambda})=1\bigr).\end{aligned}\ ] ] note that @xmath216 is a polynomial of degree @xmath13 in @xmath85 and @xmath217 and @xmath216@xmath218 for @xmath219 . we set @xmath220 for @xmath192 . for even @xmath3 , the eigenpolynomial @xmath221 ( @xmath222 ) has @xmath223 zeros in the range of @xmath87 . the operators @xmath224 and @xmath225 connect the modified system @xmath226 to the shape invariant system @xmath227 with the parameters @xmath228 , which is denoted by the rightmost vertical line in fig.2 . the @xmath13-th level ( @xmath222 ) of the modified system @xmath229 is _ iso - spectral _ with the @xmath230-th level of the new shape invariant system @xmath231 : @xmath232 here @xmath233 and @xmath234 are defined by @xmath235 b_{\ell , n-1}({\boldsymbol}{\lambda}){\stackrel{\text{def}}{=}}b_{n-1}({\boldsymbol}{\lambda})\times a^{-1 } , \end{array}\quad a=\left\ { \begin{array}{ll } ( -2)^{\ell}(n-\ell)_{\ell}&:\text{h}\\ 1&:\text{l}\\ ( -2)^{-\ell}(n+g+h+1)_{\ell}&:\text{j } \end{array}\right.\!\ ! , \label{qmflnbln}\ ] ] and they factorise @xmath188 , @xmath236 . the forward and backward shift operators @xmath237 and @xmath238 , which act on the polynomial eigenfunctions , are defined by : @xmath239 their action on the polynomials is ( @xmath222 ) : @xmath240 note that @xmath237 and @xmath238 can be expressed in terms of @xmath85 @xcite . for @xmath222 , the above relation provides a simple formula of the modified eigenpolynomial @xmath216 in terms of @xmath211 and the original eigenpolynomial @xmath241 : @xmath242 in which the coefficient @xmath243 is given by @xmath244 the orthogonality relation for even @xmath3 is : @xmath245 a few historical remarks are in order . dubov et al @xcite derived in 1992 an exactly solvable hamiltonian system of a deformed harmonic oscillator , which corresponds to the @xmath246 case of this appendix . their paper , written about two years before adler s , relied on rather heuristic arguments . recently quesne @xcite derived exactly solvable and non - shape invariant systems of deformed radial oscillator and deformed dpt potential , both are called type iii . these results again correspond to the @xmath246 cases of the radial oscillator and the dpt potential of this appendix . here we apply adler s theorem to the shape invariant , therefore solvable , systems whose eigenfunctions are described by the orthogonal polynomials ; the meixner - pollaczek ( we set the parameter @xmath247 ) , continuous hahn , wilson and askey - wilson polynomials @xcite , to be abbreviated as mp , ch , w and aw , respectively . see @xcite and @xcite for the discrete qm treatment of these polynomials . here we summarise various properties of the original hamiltonian systems to be compared with the specially modified systems to be presented in [ sec : b2 ] . let us start with the hamiltonians , schrdinger equations and eigenfunctions ( @xmath185 ) : @xmath248 the set of parameters @xmath184 are @xmath249 where @xmath250 . the the sinusoidal coordinate @xmath75 is , @xmath251 the potential function @xmath252 and energy eigenvalue @xmath188 are @xmath253 \bigl(2ix(2ix+1)\bigr)^{-1}\prod_{j=1}^4(a_j+ix)&:\text{w}\\[2pt ] \bigl((1-e^{2ix})(1-qe^{2ix})\bigr)^{-1}\prod_{j=1}^4(1-a_je^{ix } ) & : \text{aw } \end{array}\right.\!\ ! , \label{vform}\\ & \mathcal{e}_n({\boldsymbol}{\lambda}){\stackrel{\text{def}}{=}}\left\ { \begin{array}{lll } 2n&&:\text{mp}\\ n(n+b_1 - 1),&b_1{\stackrel{\text{def}}{=}}a_1+a_2+a_1^*+a_2^*&:\text{ch}\\[1pt ] n(n+b_1 - 1),&b_1{\stackrel{\text{def}}{=}}a_1+a_2+a_3+a_4&:\text{w}\\[1pt ] ( q^{-n}-1)(1-b_4q^{n-1}),&b_4{\stackrel{\text{def}}{=}}a_1a_2a_3a_4&:\text{aw } \end{array}\right.\!\!.\end{aligned}\ ] ] the eigenfunction is described by the orthogonal polynomial @xmath190 , a polynomial of degree @xmath13 in @xmath85 : @xmath254 w_n(\eta;a_1,a_2,a_3,a_4)\!\!&:\text{w}\\[1pt ] p_n(\eta;a_1,a_2,a_3,a_4|q)\!\!&:\text{aw } \end{array}\right.\!\ ! , \ \ c_n({\boldsymbol}{\lambda}){\stackrel{\text{def}}{=}}\left\ { \begin{array}{ll } \frac{1}{n!}\,2^n\!\!&:\text{mp}\\[1pt ] \frac{1}{n!}\,(n+b_1 - 1)_n\!\!&:\text{ch}\\[1pt ] ( -1)^n(n+b_1 - 1)_n\!\!&:\text{w}\\[1pt ] 2^n(b_4q^{n-1}\,;q)_n\!\!&:\text{aw } \end{array}\right.\!\!,\end{aligned}\ ] ] in which @xmath255 is the @xmath2-pochhammer symbol . we set @xmath191 for @xmath192 . the shape invariance relations involve one more parameter @xmath256 : @xmath257 ( \frac12,\frac12)&:\text{ch}\\[1pt ] ( \frac12,\frac12,\frac12,\frac12)&:\text{w,\,aw } \end{array}\right.\!\ ! , \quad \kappa{\stackrel{\text{def}}{=}}\left\ { \begin{array}{ll } 1&:\text{mp,\,ch,\,w}\\ q^{-1}&:\text{aw } \end{array}\right.\!\!,\end{aligned}\ ] ] or equivalently , @xmath258 the groundstate wavefunction @xmath259 is determined by @xmath260 and its explicit forms are : @xmath261 \sqrt{\gamma(a_1+ix)\gamma(a_2+ix)\gamma(a_1^*-ix)\gamma(a_2^*-ix ) } & : \text{ch}\\[2pt ] \sqrt{(\gamma(2ix)\gamma(-2ix))^{-1 } \prod_{j=1}^4\gamma(a_j+ix)\gamma(a_j - ix)}&:\text{w}\\[2pt ] \sqrt{(e^{2ix}\,;q)_{\infty}(e^{-2ix}\,;q)_{\infty } \prod_{j=1}^4(a_je^{ix}\,;q)_{\infty}^{-1}(a_je^{-ix}\,;q)_{\infty}^{-1 } } & : \text{aw } \end{array}\right.\!\!.\ ] ] we introduce an auxiliary function @xmath155 with the properties : @xmath262 the sinusoidal coordinate @xmath75 has the following properties : @xmath263 2\eta(x)\cosh\tfrac{k\gamma}{2}&:\text{aw } \end{array}\right.\!\!,\\ & \eta(x - i\tfrac{k\gamma}{2})\eta(x+i\tfrac{k\gamma}{2 } ) = \left\ { \begin{array}{ll } \eta(x)^2+\tfrac14k^2&:\text{mp,\,ch}\\ \bigl(\eta(x)+\tfrac14k^2\bigr)^2&:\text{w}\\[1pt ] \eta(x)^2+\sinh^2\tfrac{k\gamma}{2}&:\text{aw } \end{array}\right.\!\!. \label{etaprod}\end{aligned}\ ] ] these mean that for a polynomial @xmath264 in @xmath85 , @xmath265 is another polynomial in @xmath75 . the action of @xmath197 and @xmath198 on the eigenfunctions is @xmath199 the factors of the energy eigenvalue , @xmath200 and @xmath201 , @xmath266 , are given by @xmath267 the forward and backward shift operators @xmath203 and @xmath204 are defined by @xmath268 and their action on the polynomials is @xmath206 the orthogonality relation is @xmath269 2\pi n!\,(n+b_1 - 1)_n\prod_{1\leq i < j\leq 4}\gamma(n+a_i+a_j)\cdot \gamma(2n+b_1)^{-1}&:\text{w}\\[2pt ] 2\pi(b_4q^{n-1};q)_n(b_4q^{2n};q)_{\infty}(q^{n+1};q)_{\infty}^{-1 } \prod_{1\leq i < j\leq 4}(a_ia_jq^n;q)_{\infty}^{-1}&:\text{aw } \end{array}\right.\!\!.\end{aligned}\ ] ] the potential function , the hamiltonian and the schrdinger equation of the modified system are : @xmath270 the explicit forms of the deforming polynomial @xmath211 will be given in . for even @xmath3 , @xmath209 has no zero in the rectangular domain in the complex @xmath87 plane , @xmath271 , @xmath272 . note that the normalization of @xmath168 does not affect @xmath177 . this system is not shape invariant . the second line of the expression for @xmath273 , , is obtained from the first line by repeated use of the formula of the auxiliary function @xmath274 . as mentioned in section [ summary ] , this form of the deformed potential function is closely related to that of the exceptional wilson and askey - wilson polynomials . it is convenient to introduce an auxiliary function @xmath275 : @xmath276}\prod_{k=1}^{\ell-2 } \bigl(\varphi(x - i\tfrac{k}{2}\gamma)\varphi(x+i\tfrac{k}{2}\gamma ) \bigr)^{[\frac{\ell - k}{2 } ] } , \label{varphildef}\ ] ] where @xmath277 $ ] denotes the greatest integer not exceeding @xmath87 . note that @xmath278 + 2\sum_{k=1}^{\ell-2}[\frac{\ell - k}{2 } ] = \frac12\ell(\ell-1)$ ] . the deforming polynomial @xmath211 is defined by @xmath279(x;{\boldsymbol}{\lambda } ) \qquad\quad\bigl ( \check{p}_n(x;{\boldsymbol}{\lambda}){\stackrel{\text{def}}{=}}p_n(\eta(x);{\boldsymbol}{\lambda})\bigr){\nonumber\\}&{\stackrel{\text{def}}{=}}\prod_{k=1}^{\ell}c_k({\boldsymbol}{\lambda})\cdot \varphi_{\ell}(x ) \xi_{\ell}(\eta(x);{\boldsymbol}{\lambda } ) \times\left\ { \begin{array}{ll } \prod_{k=1}^{\ell}k!\!\!&:\text{mp,\,ch,\,w}\\ \prod_{k=1}^{\ell}\prod_{j=1}^k\sinh\frac{-j\gamma}{2}\!\!&:\text{aw } \end{array}\right.\!\!.\end{aligned}\ ] ] as in the ordinary qm cases , it is expressed in terms of the polynomial @xmath280 of the original system with shifted parameters : @xmath281 note that @xmath282 and @xmath283 for the mp , w and aw cases . the eigenfunctions are @xmath284(x;{\boldsymbol}{\lambda } ) { \stackrel{\text{def}}{=}}\prod_{k=1}^{\ell}c_k({\boldsymbol}{\lambda})\cdot \varphi_{\ell+1}(x)(-1)^{\ell } p_{\ell , n}(\eta(x);{\boldsymbol}{\lambda})\\ & \qquad\qquad\qquad\qquad\qquad\qquad \times\left\ { \begin{array}{ll } \prod_{k=1}^{\ell}k!&\!\!:\text{mp,\,ch,\,w}\\ \prod_{k=1}^{\ell}\prod_{j=1}^k\sinh\frac{-j\gamma}{2}&\!\!:\text{aw } \end{array}\right . \bigl(\rightarrow p_{\ell,0}(\eta;{\boldsymbol}{\lambda})=1\bigr ) . \nonumber\end{aligned}\ ] ] for even @xmath3 , @xmath221 ( @xmath222 ) has @xmath223 zeros in the range of @xmath87 . note that @xmath216 is a polynomial of degree @xmath13 in @xmath85 and @xmath217 and @xmath220 for @xmath219 . we set @xmath220 for @xmath192 . the operators @xmath224 and @xmath225 connect the modified system @xmath285 to the shape invariant system @xmath286 with the parameters @xmath228 , which is denoted by the rightmost vertical line in fig.2 . the @xmath13-th level ( @xmath222 ) of the modified system @xmath229 is _ iso - spectral _ with the @xmath230-th level of the new shape invariant system @xmath231 : @xmath232 here , @xmath233 and @xmath234 are the factors of the energy eigenvalue , @xmath236 , and are defined by @xmath235 b_{\ell , n-1}({\boldsymbol}{\lambda}){\stackrel{\text{def}}{=}}b_{n-1}({\boldsymbol}{\lambda})\times a^{-1 } , \end{array}\quad a=\left\ { \begin{array}{ll } 2^{\ell}&:\text{mp}\\ ( b_1+n)_{\ell}&:\text{ch}\\ ( -1)^{\ell}(n-\ell)_{\ell}(b_1+n)_{\ell}&:\text{w}\\ q^{-\frac12\ell n}(q^{n-\ell};q)_{\ell}(b_4q^n;q)_{\ell}&:\text{aw } \end{array}\right.\!\!. \label{dqmflnbln}\ ] ] the forward and backward shift operators @xmath237 and @xmath238 which act on the polynomial eigenfunctions , are defined by : @xmath287 and their action on the polynomials is @xmath288 for @xmath222 , the above formula provides a simple formula of the modified eigenpolynomial @xmath216 in terms of @xmath211 and the original eigenpolynomial @xmath241 : @xmath289 the orthogonality relation for even @xmath3 is : @xmath290 s.odake and r.sasaki , `` unified theory of annihilation - creation operators for solvable ( ` discrete ' ) quantum mechanics , '' j. math . phys . * 47 * ( 2006 ) 102102 ( 33pages ) , exact solution in the heisenberg picture and annihilation - creation operators , " phys . lett . * b641 * ( 2006 ) 112 - 117 . h.risken , _ the fokker - planck equation _ ( 2nd . ed . ) , springer - verlag , berlin , ( 1996 ) . d.gmez-ullate , n.kamran and r.milson , `` an extension of bochner s problem : exceptional invariant subspaces , '' j. approx theory * 162 * ( 2010 ) 987 - 1006 ; `` an extended class of orthogonal polynomials defined by a sturm - liouville problem , '' j. math . * 359 * ( 2009 ) 352 - 367 ; exceptional orthogonal polynomials and the darboux transformation , " arxiv:1002.2666[math - ph ] . c.quesne , `` exceptional orthogonal polynomials , exactly solvable potentials and supersymmetry , '' j. phys . * a41 * ( 2008 ) 392001 ; b.bagchi , c.quesne and r.roychoudhury , isospectrality of conventional and new extended potentials , second - order supersymmetry and role of pt symmetry , " pramana j. phys . * 73 * ( 2009 ) 337 - 347 . s.odake and r.sasaki , `` infinitely many shape invariant potentials and new orthogonal polynomials , '' phys . lett . * b679 * ( 2009 ) 414 - 417 . s.odake and r.sasaki , infinitely many shape invariant discrete quantum mechanical systems and new exceptional orthogonal polynomials related to the wilson and askey - wilson polynomials , " phys . lett . * b682 * ( 2009 ) 130 - 136 . s.odake and r.sasaki , infinitely many shape invariant potentials and cubic identities of the laguerre and jacobi polynomials , " j. math * 51 * ( 2010 ) april issue , in press , arxiv:0911.1585[math - ph ] . s.odake and r.sasaki , `` another set of infinitely many exceptional ( @xmath291 ) laguerre polynomials , '' phys . * b684 * ( 2010 ) 173 - 176 . l.ho , s.odake and r.sasaki , `` properties of the exceptional ( @xmath291 ) laguerre and jacobi polynomials , '' dpsu-09 - 7 , yitp-09 - 70 , arxiv:0912.5447[math - ph ] . s.yu.dubov , v.m.eleonski and n.e.kulagin , `` equidistant spectra of anharmonic oscillators , '' soviet phys . jetp * 75 * ( 1992 ) 446 - 451 ; chaos * 4 * ( 1994 ) 47 - 53 . m.g.krein , doklady acad . cccp , * 113 * ( 1957 ) 970 - 973 ; v..adler , `` a modification of crum s method , '' theor . math . phys . * 101 * ( 1994 ) 1381 - 1386 . p.b.abraham and h.e.moses , changes in potentials due to changes in the point spectrum : anharmonic oscillators with exact solutions , " phys . rev . * a22 * ( 1980 ) 1333 - 1340 ; b.mielnik , factorization method and new potentials with the oscillator spectrum , " j. math . phys . * 25 * ( 1984 ) 3387 - 3389 ; m.m.nieto , relationship between supersymmetry and the inverse method in quantum mechanics , " phys . lett . * b145 * ( 1984 ) 208 - 210 . g.junker and p.roy , conditionally exactly solvable problems and nonlinear algebras , " phys . * a232 * ( 1997 ) 155 - 161 ; conditionally exactly solvable potentials : a supersymmetric construction method , " annals of phys . * 270 * ( 1998 ) 155 - 177 . s.odake and r.sasaki , `` shape invariant potentials in ` discrete ' quantum mechanics , '' j. nonlinear math * 12 * suppl . 1 ( 2005 ) 507 - 521 ; `` equilibrium positions , shape invariance and askey - wilson polynomials , '' j. math . phys . * 46 * ( 2005 ) 063513 ( 10 pages ) . s.odake and r.sasaki , `` calogero - sutherland - moser systems , ruijsenaars - schneider - van diejen systems and orthogonal polynomials , '' prog . theor . phys . * 114 * ( 2005 ) 1245 - 1260 . s.odake and r.sasaki , orthogonal polynomials from hermitian matrices , " j. math . phys . * 49 * ( 2008 ) 053503 ( 43 pages ) . s.odake and r.sasaki , exactly solvable ` discrete ' quantum mechanics ; shape invariance , heisenberg solutions , annihilation - creation operators and coherent states , " prog . * 119 * ( 2008 ) 663 - 700 . a.f.nikiforov , s.k.suslov , and v.b.uvarov , _ classical orthogonal polynomials of a discrete variable _ , springer , berlin , ( 1991 ) . s.odake and r.sasaki , @xmath2-oscillator from the @xmath2-hermite polynomial , " phys . lett . * b663 * ( 2008 ) 141 - 145 . p.gaillard and v.b.matveev , wronskian and casorati determinant representation for darboux - pschel - teller potentials and their difference extensions , " j. phys . * a42 * ( 2009 ) 404009 .

we examine theoretically coherent electron transport through the single - molecule magnet mn , bridged between au(111 ) electrodes , using the non - equilibrium green s function method and the density - functional theory . we analyze the effects of bonding type , molecular orientation , and geometry relaxation on the electronic properties and charge and spin transport across the single - molecule junction . we consider nine interface geometries leading to five bonding mechanisms and two molecular orientations : ( i ) au - c bonding , ( ii ) au - au bonding , ( iii ) au - s bonding , ( iv ) au - h bonding , and ( v ) physisorption via van der waals forces . the two molecular orientations of mn correspond to the magnetic easy axis of the molecule aligned perpendicular [ hereafter denoted as orientation ( 1 ) ] or parallel [ orientation ( 2 ) ] to the direction of electron transport . we find that the electron transport is carried by the lowest unoccupied molecular orbital ( lumo ) level in all the cases that we have simulated . relaxation of the junction geometries mainly shifts the relevant occupied molecular levels toward the fermi energy as well as slightly reduces the broadening of the lumo level . as a result , the current slightly decreases at low bias voltage . our calculations also show that placing the molecule in the orientation ( 1 ) broadens the lumo level much more than in the orientation ( 2 ) , due to the internal structure of the mn . consequently , junctions with the former orientation yield a higher current than those with the latter . among all of the bonding types considered , the au - c bonding gives rise to the highest current ( about one order of magnitude higher than the au - s bonding ) , for a given distance between the electrodes . the current through the junction with other bonding types decreases in the order of au - au , au - s , and au - h . importantly , the spin - filtering effect in all the nine geometries stays robust and their ratios of the majority - spin to the minority - spin transmission coefficients are in the range of 10 to 10 . the general trend in transport among the different bonding types and molecular orientations obtained from this study may be applied to other single - molecular magnets .

1003.2750

advances in experimental techniques have led to a great number of experimental studies on electron transport through molecular junctions formed by single molecules bridged between electrodes or molecular monolayers adsorbed onto surfaces , using three - terminal set - ups or scanning tunneling microscopy ( stm ) measurements . recently , molecular junctions based on single - molecule magnets ( smms ) connected to electrodes or monolayers of smms at surfaces , have been fabricated and their electron transport characteristics @xcite have been measured , as well as their mechanical , electronic , and magnetic properties @xcite . electron transport through an smm drew a lot of attention because of the intriguing interplay between its transport properties and the internal magnetic degrees of freedom , which is absent in transport through small organic molecules . an smm consists of several transition metal ions interacting through organic or inorganic ligands via super - exchange interactions . the spin configuration of the ground state of an smm is determined by a delicate balance among the super - exchange interactions of different strengths between the transition metal ions . thus , the magnetic structure of an smm must be taken into account in understanding its electron transport and other quantum properties . recently , first - principles calculations of transport through an smm were performed on a prototype smm mn@xmath0 terminated with a thiol ( -s ) group within au electrodes.@xcite the calculations carried out in refs . [ ] suggest that the mn@xmath0 molecule can function as a spin filter with low bias voltage even with non - magnetic electrodes . even though the mn@xmath0 is chemically bonded to the au electrodes , the broadening of the relevant molecular orbitals is so small compared to its charging energy that the kondo temperature is expected to be extremely low . in addition , these @xcite and other calculations @xcite demonstrate the importance of the internal magnetic degrees of freedom of the mn@xmath0 in electron transport , in contrast to typical quantum dots , regardless of the specific details of the coupling of the molecule to the electrodes . systematic studies on molecular junctions based on small organic molecules reveal that the current through the molecular junctions is highly sensitive to properties of interfaces between the molecules and electrodes because the interfaces determine the degree of the overlap between the molecular levels and conduction channels of the electrodes.@xcite a thiol group is most commonly used to build a strong chemical link between the single molecules and the au electrodes in molecular junctions . the conductance of such single - molecule junctions is typically several orders of magnitude smaller than @xmath3 ( the conductance quantum ) . binding of the s atoms to the hollow sites of the au surface gives rise to different conductance ( by a factor of about 2 to 3 ) from binding of the same s atoms to on - top sites of the au surface.@xcite molecular junctions terminated with an amine ( -n ) group within au electrodes reveal even lower conductance than those with a thiol group.@xcite in order to increase conductance of molecular junctions , linker molecules including au atoms @xcite or isocyanide derivatives @xcite were used between the single molecules and electrodes . in addition , recent experiments show that conductance through molecular junctions based on small single molecules can be enhanced to an order of @xmath4 by their direct bonding to the electrodes without a thiol group or any other linker molecules.@xcite this enhancement is attributed to a strong coupling between the molecules and the electrodes , which places the transport in the transparent regime rather than in the tunneling regime.@xcite however , corresponding systematic studies have not yet been carried out for electron transport through an smm . in the case of quantum dots , the properties of interfaces are negligible in transport . however , for molecular junctions based on smms , the molecules are chemically bonded to the electrodes , and thus their transport properties can change with different interfaces even if the molecules are weakly coupled to the electrodes in the sense that the charging energy is much greater than the broadening of the relevant molecular levels . in this paper , we investigate how bonding types , interface geometries , and geometry relaxation influence transport characteristics of an smm mn@xmath0 when it is bridged between au electrodes . an smm mn@xmath0 @xcite consists of four inner mn@xmath5 ions ( @xmath6 ) surrounded by eight outer mn@xmath7 ions ( @xmath8 ) which are all antiferromagnetically coupled through o anions , as well as c and h atoms , as shown in fig . [ fig : isolated - mn12geo ] . the dominant exchange interactions are between the inner mn@xmath5 ions and the outer mn@xmath7 ions.@xcite in the ground state the magnetic moments of the four mn@xmath5 ions are antiparallel to those of the eight mn@xmath7 ions , such that the total spin becomes @xmath9 . @xcite an smm mn@xmath0 initially synthesized by lis @xcite consists of ligands which do not form direct chemical bonding to an au surface . thus , most transport experiments through an mn@xmath0 within au electrodes were performed on systems where the mn@xmath0 molecules are chemically bonded to surfaces or electrodes via linker molecules such as a thiol group,@xcite or physically adsorbed onto surfaces or electrodes without any linker molecules.@xcite to form direct au - s bonding , some of the original ligands of the mn@xmath0 molecules @xcite were substituted by s - containing ligands , or the mn@xmath0 molecules were deposited onto an au surface that was initially functionalized with s - containing alkane chains.@xcite so far , the measured electric current through an mn@xmath0 molecule or its derivative in various experiments is in the range of 1 pa to 100 pa at bias voltage of a few tens of mv.@xcite compared to binding a small organic molecule to au electrodes through a thiol or amine group , binding an mn@xmath0 to au electrodes via linker molecules bears the following differences : ( a ) an mn@xmath0 has much more binding sites to the linker molecules , and ( b ) the orientation of an mn@xmath0 relative to the electrodes is , to a great extent , determined by binding sites of the linker molecules to the mn@xmath0 , rather than their binding sites to the electrodes . molecule : mn ( purple ) , o ( red ) , c ( gray ) , h ( white ) . the ground - state spin is @xmath10 or the magnetic moment is 20 @xmath11.,height=241 ] full control of the properties of interfaces is extremely difficult to achieve in molecular junctions and tunneling measurements using stm or scanning tunneling spectroscopy ( sts).@xcite thus , in transport experiments , for a given bonding type , several binding sites are plausible , and hundreds of fabricated samples of the bonding type give rise to a histogram of conductance for given gate voltage or a series of current - voltage curves . in the present study , we take into account five different bonding types between an mn@xmath0 and au electrodes , and for each bonding type , some representative interface geometries are examined . even though we do not simulate all possible configurations of the interface that may be realized in fabricated samples for a given bonding type , the general trend in our calculated transport properties will elucidate the effects of linker group , molecular orientation , and geometry relaxation , as well as the effectiveness of different bonding types and interface geometries in transport through an mn@xmath0 and other smms . in this work , we consider an mn@xmath0 molecule bridged between au(111 ) electrodes ( two - terminal set - up ) via nine different ways ( fig . [ fig : geo ] ) , and investigate their current - voltage characteristics using the density - functional theory ( dft ) and the nonequilibrium green s function method ( negf ) . the nine different ways can be categorized according to their bonding types : ( i ) au - s bonding , ( ii ) au - c bonding , ( iii ) au - au bonding , ( iv ) au - h bonding , and ( v ) physisorption through van der waals forces . dft does not fully capture the nature of van der waals forces where the orbitals considered do not overlap . we note , however , that if molecular orbitals do not overlap with the orbitals of electrodes at a molecular junction , the transport properties of the junction belong to the tunneling regime . a recent study reveals that first - principles calculations based on dft and negf provide qualitative features of transport in such a regime.@xcite for example , when a pt break junction becomes broken with vacuum between the electrodes , a first - principles calculation showed exponentially decaying conductance as a function of distance between the electrodes . @xcite this result suggests that our methodology can be used to understand the transport in the case of physisorption , although the physical mechanism in our case differs from that in the broken junction . in the present calculations , we consider two molecular orientations relative to the electrodes : when the magnetic easy axis of the mn@xmath0 is normal [ orientation ( 1 ) ] or parallel [ orientation ( 2 ) ] to the transport direction , the @xmath12 axis , as shown in fig . [ fig : geo ] . the structure of the mn@xmath0 dictates that in the orientation ( 1 ) one linker molecule can be attached to each electrode without significantly deforming the mn@xmath0 , while in the orientation ( 2 ) , two linker molecules are possibly accommodated for each electrode without much distortion . as illustrated in fig . [ fig : geo ] , the nine different interface geometries are named based on the following rule . adapted from the notations used in refs . [ ] , we refer to the orientation ( 1 ) as geo 1 , and to the orientation ( 2 ) as geo 2 . for each interface geometry , the orientation of an mn@xmath0 is first specified , then the type of adsorption ( either chemical or physical ) is given . for chemical bonding , the type of bonding and the name of the linker group are noted , while for physisorption , whether linker molecules are placed or not is stated . for example , geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow [ fig . [ fig : geo](a ) ] denotes an interface geometry with the mn@xmath0 in the orientation ( 1 ) and the linker molecule ( sc@xmath13h@xmath14)@xmath15 bonded to the hollow sites of the electrodes via au - s bonding . we will first outline our model and computational method . then we will discuss the interface geometries in detail , and the effects of geometry relaxation , molecular orientation , bonding type , and linker group on transport properties of the mn@xmath0 . a brief conclusion will follow . we use the quantum transport code smeagol @xcite interfaced with the dft code siesta@xcite . the au electrodes are treated semi - infinitely using siesta . the scattering region consists of the mn@xmath0 , linker molecules , and six flat au atomic layers on each side of the mn@xmath0 ( @xmath16 surface atoms per layer ) , as shown in fig . [ fig : geo ] . even if the number of the au atomic layers increases from six to nine on each side of the mn@xmath0 , the transmission probability does not change at all . this was tested and confirmed for the interface geometry with the au - c bonding [ geo 1:au-(c@xmath13h@xmath14)@xmath15 , fig . [ fig : geo](c ) ] , where the broadening of the relevant molecular level turns out to be the largest among the interface geometries considered . the electronic structures of the electrodes and scattering region are calculated within the spin - polarized perdew - burke - ernzerhopf ( pbe ) generalized - gradient approximation ( gga ) @xcite , using siesta . we generate troullier - martins pseudopotentials @xcite for au , mn , s , o , c , and h with scalar relativistic terms and core corrections except for h. we also build corresponding basis sets of the elements using the scheme presented in ref . [ ] . for the mn basis set , 3@xmath17 orbitals as a semicore are required in order for the molecular orbital levels of an isolated mn@xmath0 to be comparable to those obtained from all - electron calculations @xcite . with these pseudopotentials and basis sets , self - consistent dft calculations including spin - orbit coupling @xcite are performed for an isolated mn@xmath0 . using a modified version @xcite of siesta , we obtain the total magnetic moment of 20 @xmath11 and the magnetic anisotropy barrier of 66.4 k,@xcite . this is in good agreement with experiment @xcite and with the barrier @xcite computed using the dft code vasp@xcite . to reduce the computational cost , a small au basis set of a single @xmath18 orbital is used for transport calculations , while a large au basis set of both @xmath19 and @xmath18 orbitals is used for geometry relaxation . the transmission at the fermi level in au is dominated by the @xmath18 states because the @xmath19 orbital levels are well below the fermi level . it is checked that in the scattering region , the density of states projected onto mn @xmath19 orbitals with the large au basis set is fairly similar to that with the small au basis set , which justifies the utilization of the small au basis set in the transport study . for a given interface geometry , the distance @xmath19 between the electrodes was determined such that all of the terminating atoms from the linker molecules ( such as s , c , or au atoms ) are bonded to the lowest - energy sites of the the au slab with reasonable bond lengths . for example , for the au - s bonding , the distance @xmath19 was obtained from an optimum bond length between a s - terminated alkane chain and the hollow site of a small au cluster . in constructing the initial geometry of the scattering region , each component of the region , such as the electrodes , the mn@xmath0 , and the linker molecules , are separately optimized in advance . the largest force is found at the interface between the linker molecules and the electrodes.@xcite thus , the geometry relaxation with a fixed @xmath19 allows the linker molecules and the mn@xmath0 to relax further and lowers the force at the interface . prior to the transport calculations , all of the geometries of the scattering region or interface geometries considered are relaxed with a fixed distance @xmath19 between the electrodes , using siesta , until the magnitude of the maximum force exerted on the atoms becomes less than 0.1 ev / , unless stated otherwise . to avoid any quantum confinement effects,@xcite for the scattering region , periodic boundary conditions are applied in the transverse directions and @xmath20 @xmath21-points are sampled . the retarded green function for the scattering region , @xmath22 , has the form of @xmath23^{-1}$ ] , where @xmath24 and @xmath25 are the overlap and hamiltonian matrices for the scattering region , respectively.@xcite here @xmath26 and @xmath27 are self - energies arising from the interactions of the mn@xmath0 with the left and right electrodes , respectively . these self - energies depend on energy @xmath28 and @xmath21 vector , and they are obtained from spin - polarized calculations of retarded surface green functions for the electrodes which are computed using the scheme developed in ref . the initial magnetic moment of the scattering region is set to 20 @xmath11 . interactions with phonons or additional electron correlations such as on - site coulomb repulsion ( hubbard - like @xmath29 term ) are not taken into account . the density matrix of the scattering region is self - consistently solved within the negf formalism until it converges . after the convergence of the density matrix , the total magnetic moment slightly increases , such as 20.3 @xmath11 for geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow , due to a small amount of spin polarization in the au atomic layers caused by the mn@xmath0 . then the transmission @xmath30 at low bias voltage @xmath31 is calculated as follows : @xmath32(e , v_b),\end{aligned}\ ] ] where @xmath33 and @xmath34 denote the broadening of molecular levels induced by coupling to the left and right electrodes . accurate calculations of @xmath30 require high resolution in energy @xmath28 ( such as @xmath35ev for very sharp transmission peaks ) due to the weak coupling between the mn@xmath0 and the electrodes . since the electrodes are treated semi - infinitely and periodic boundary conditions are employed , the transmission coefficients are obtained after their integration over the @xmath21-points . the current @xmath36 as a function of @xmath31 is obtained from @xmath37 , \label{eq : iv}\end{aligned}\ ] ] where @xmath38 and @xmath39 are the fermi - dirac distribution functions of the left and right electrodes . the electronic temperature used is 10 k. we examine the nine interface geometries ( fig . [ fig : geo ] ) leading to the five bonding types and two molecular orientations that were discussed in sec . i. their molecular orientations , linker groups , bonding types , binding sites , and distances @xmath19 between the electrodes are listed in table 1 . in the au - s bonding , four interface geometries are considered : fig . [ fig : geo](a ) , ( b ) , ( e ) , and ( f ) . in fig . [ fig : geo](a ) , geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow , one sc@xmath13h@xmath14 linker molecule is bonded to each au electrode , where the s atom is adsorbed at the hollow sites of the flat au surface . for a fixed distance @xmath19 , after the geometry relaxation , this geometry bears the shortest distances between the s and au surface atoms in the range of 2.52 to 2.82 . if the s atom is bonded to the on - top site of the au surface [ geo 1:au-(sc@xmath13h@xmath14)@xmath15-ontop , fig . [ fig : geo](b ) ] , the total energy increases by 0.094 ev compared to the hollow - site case , and the shortest distance between the s and au atoms becomes 1.92 . in fig . [ fig : geo](e ) and ( f ) , two s linker atoms [ geo 2:au - s@xmath40 or two sc@xmath15h@xmath41 linker molecules [ geo 2:au-(sc@xmath15h@xmath41)@xmath40 are bonded to the hollow sites of the surface of each electrode . for these two geometries the shortest distances between the s and au atoms are in the range of 2.52 to 2.68 . our study of the au - c bonding was motivated by recent measurements on the conductance of a molecular junction based on benzene directly bonded to pt electrodes.@xcite it is prevalent that chemical bonding between a metallic surface and c atoms is difficult to achieve under ordinary conditions , especially between an au surface and c atoms . however , the direct au - c bonding does occur in a few exceptional cases , such as isocyanide derivatives adsorbed on au surfaces.@xcite in the geometry with the au - c bonding [ geo 1:au-(c@xmath13h@xmath14)@xmath15 , fig . [ fig : geo](c ) ] , one c@xmath13h@xmath14 linker molecule is adsorbed at the on - top site of each electrode , and the shortest distance between the au and outermost c atoms is 2.28 ( 2.37 ) for the left ( right ) linker molecule . the study of the au - au bonding was inspired by an experimental effort to increase conductance through small molecules using au - au bonding.@xcite in the geometry with the au - au bonding [ geo 1:au-(auc@xmath13h@xmath14)@xmath15 , fig . [ fig : geo](d ) ] , one auc@xmath13h@xmath14 linker molecule is bonded to the hollow sites of each electrode , and the shortest distances between the au surface and the au atoms from the linker molecules are in the range of 2.73 to 2.79 . the study of the geometries with physisorption and without linker molecules , was motivated by the transport measurement through the mn@xmath0 without any linker molecules @xcite . we consider such geometries in the orientation ( 2 ) . in the geometry with the au - h bonding [ geo 2:au - h - no - linker , fig . [ fig : geo](h ) ] , no linker molecules are placed between the mn@xmath0 and the electrodes . however , the short distance @xmath19 in that geometry allows au - h bonding to be initially formed at four different locations . we choose one of the four au - h bonding distances in its initial geometry to be slightly shorter than the three distances , and then upon geometry relaxation , only one au - h bonding remains effective as shown in fig . [ fig : geo](h ) . in the case of physisorption , two geometries are considered : figs . [ fig : geo](g ) and ( i ) . in fig . [ fig : geo](g ) , geo 2:phys-(ch@xmath13)@xmath41 , two ch@xmath13 linker molecules are attached to each side of the mn@xmath0 . upon geometry relaxation ( with a fixed @xmath19 until the magnitude of the maximum force is less than 0.1 ev / ) , in this geometry , the shortest distance between the au and the c atoms from the linker molecules is 2.41 , which is long enough such that chemical bond is not formed between the linker molecules and the electrodes . in fig . [ fig : geo](i ) , geo 2:phys - no - linker , no linker molecules are placed between the mn@xmath0 and the electrodes , and the au surface atoms are well separated from the h atoms from the mn@xmath0 in the range of 2.97 to 3.10 . thus , no chemical bond is formed between the electrodes and the mn@xmath0 in this geometry , either . the geometry of geo 2:phys - no - linker was built from the optimized geometry of geo 2:au - h - no - linker with an increased distance @xmath19 . hereafter , when a particular bonding type is discussed , its corresponding binding site is also implicitly assumed . for an isolated mn@xmath0 molecule in the gas phase , without the on - site coulomb repulsion @xmath29 , our dft calculations using siesta show that the energy gap between the highest occupied molecular orbital ( homo ) and the lowest unoccupied molecular orbital ( lumo ) levels is 0.38 ev . since the mn@xmath0 has a high magnetic moment of 20 @xmath11 in the ground state , both the homo and lumo arise from the majority - spin mn @xmath19 orbitals . this feature of the homo and lumo does not change with inclusion of a proper value of the @xmath29 term , although the homo - lumo energy gap increases with the @xmath29 term . this was shown in previous dft+u calculations @xcite , in contrast to the result discussed in ref . [ ] . such a characteristic of the homo and lumo was also earlier revealed by all - electron dft calculations@xcite , and it was indirectly demonstrated in experiments on locally charged mn@xmath0 molecules@xcite . orbitals of the ( a ) initial and ( b ) optimized geometries , geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow . refer to fig . [ fig : geo ] for numbering of the mn sites . the densities of states for mn(3 ) and mn(4 ) sites are the same as those for mn(1 ) and mn(2 ) . , title="fig:",width=294,height=226 ] orbitals of the ( a ) initial and ( b ) optimized geometries , geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow . refer to fig . [ fig : geo ] for numbering of the mn sites . the densities of states for mn(3 ) and mn(4 ) sites are the same as those for mn(1 ) and mn(2 ) . , title="fig:",width=294,height=226 ] when an mn@xmath0 is bridged between the electrodes in the orientation ( 1 ) , the s@xmath41 symmetry of an isolated mn@xmath0 is broken and the molecular levels in the interface show approximately twofold symmetry . in addition , the magnetic easy axis of the mn@xmath0 is slightly tilted from the axis perpendicular [ @xmath42 axis , fig . [ fig : geo](a ) ] to the transport direction , in order to allow the linker molecules to be bonded to the electrodes . furthermore , geometry relaxation at a fixed @xmath19 renders some noticeable changes in the molecular levels . to compare the molecular levels before and after the geometry relaxation , we examine the geometry geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow . for its initial geometry , the homo and lumo levels of the mn @xmath19 orbitals , are separated by 0.3 ev , which is similar to the homo - lumo energy gap of 0.38 ev for the isolated mn@xmath0 . with the geometry optimization , the mn @xmath19 orbital levels right below the fermi level @xmath43 approach toward @xmath43 , but the mn @xmath19 levels right above @xmath43 become split ( fig.[fig : pdos - geo1 ] ) . as a result , the separation between the homo and lumo levels reduces to 0.15 ev . the homo level is located near @xmath44 ev , while the lumo level is near @xmath43 [ fig . [ fig : pdos - geo1](b ) ] ( this is due to some charge transfer from the au electrodes to the mn@xmath0 @xcite ) . the same kind of shift is expected for the rest of the geometries considered in the orientation ( 1 ) . now when an mn@xmath0 is bridged between the electrodes in the orientation ( 2 ) , the s@xmath41 symmetry is by large preserved in the molecular levels in the interface . if the distance @xmath19 is not too short , the geometry relaxation will shift the molecular levels for geometries in the orientation ( 2 ) as well . one exception is geo 2:au - s@xmath41 , where a combination of the strong au - s bonding with the short distance @xmath19 prevents the mn@xmath0 and the four s atoms from being significantly rotated or stretched / compressed with geometry relaxation . at zero bias as a function of energy @xmath28 relative to the fermi level @xmath43 for the five interface geometries with the orientation ( 1 ) , and ( b ) for the five geometries with the orientation ( 2 ) . the arrows indicate the transmission peaks associated with the homo and lumo levels of the mn@xmath0.,title="fig:",width=294,height=226 ] at zero bias as a function of energy @xmath28 relative to the fermi level @xmath43 for the five interface geometries with the orientation ( 1 ) , and ( b ) for the five geometries with the orientation ( 2 ) . the arrows indicate the transmission peaks associated with the homo and lumo levels of the mn@xmath0.,title="fig:",width=294,height=226 ] we discuss the effect of geometry relaxation on @xmath45 . similarly to the effect on the molecular levels , the geometry relaxation gives rise to a upward shift in the transmission peaks right below @xmath43 and more spread peaks right above @xmath43 , as shown in the first and the third panels of fig . [ fig : trc](a ) from the top . according to our calculations , the current is carried via the lumo level of the mn@xmath0 , and thus the broadening of the lumo level is relevant to the current - voltage characteristics . without contributions from interactions with phonons , in resonance , the individual transmission peak centered at @xmath46 for given @xmath31 ( for a single conductance channel ) is solely described by the broadening of the corresponding molecular level as follows:@xcite @xmath47 where @xmath48 and @xmath49 are the broadening of the @xmath50th molecular level caused by the left and right electrodes , respectively , and they are determined by full widths at half maximum of the transmission peak centered at @xmath46 . here @xmath51 is the average broadening given by @xmath52 . ( notice that for symmetric coupling , @xmath53 , eq . ( [ eq : trc ] ) implies that @xmath54 becomes unity at @xmath55 . ) the orbital broadening @xmath56 is related to the hopping integral @xmath57 ( between the electrodes and the mn@xmath0 ) as @xmath58 , where @xmath59 is the density of states at the @xmath50th level.@xcite the values of @xmath57 and @xmath59 depend on bonding types and binding sites . we compute the values of @xmath60 and @xmath61 from fitting the transmission peaks to eq . ( [ eq : trc ] ) . for geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow , the geometry relaxation reduces the value of @xmath62 from 0.0028 to 0.0020 ev , while it enhances the value of @xmath63 from 2.6@xmath64 to 5.0@xmath65 ev , as listed in table 1 . since the value of @xmath62 decreases to a small degree with the geometry relaxation , a slightly lower current flows through the relaxed junction than via the initial one ( the topmost left panel of fig . [ fig : iv ] ) . similar behavior is expected for the rest of the geometries . henceforth , we discuss optimized interface geometries only , unless specified otherwise . ) characteristics for six interface geometries : majority - spin contribution ( red box ) , minority - spin contribution ( blue star ) . the topmost left panel shows @xmath66 curves for both the initial ( brown circle ) and the optimized geometries geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow ( red box ) . note that geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow and geo 2:phys-(ch@xmath13)@xmath41 have different vertical scales from the rest of the geometries . ] for the interface geometries with the orientation ( 1 ) , fig . [ fig : geo](a ) , ( b ) , ( c ) , and ( d ) , the linker molecules are closer to the mn(5 ) , mn(6 ) , mn(9 ) , and mn(10 ) sites than the rest of the eight mn sites , as illustrated in fig . [ fig : geo ] . among the former four mn sites , the mn(5 ) and mn(9 ) sites predominantly contribute to the lumo of the mn@xmath0 [ fig . [ fig : ldos](a ) ] , as well as to some degree to the homo [ figs . [ fig : pdos - geo1](b ) and [ fig : ldos](b ) ] . notice that the homo and lumo arise from the majority - spin mn @xmath19 orbitals . a much greater contribution to the homo arises from the mn(8 ) and mn(12 ) sites [ fig . [ fig : ldos](b ) ] , but these sites are not as close to the linker molecules as the former four sites . thus , for the geometries with the orientation ( 1 ) , the broadening of the lumo level is expected to be somewhat larger than that of the homo level , as shown in table 1 and figs . [ fig : pdos - geo1](b ) , [ fig : trc](a ) , and [ fig : pdos - geo1a ] . for the geometries with the orientation ( 2 ) , in the case of chemisorption , fig . [ fig : geo](e ) , ( f ) , and ( h ) , the linker molecules are in a closer proximity to the mn(1 ) , mn(2 ) , mn(3 ) , mn(4 ) , mn(6 ) , mn(8 ) , mn(10 ) , and mn(12 ) sites , than to the other four mn sites , as shown in fig . [ fig : geo ] . the mn(6 ) , mn(8 ) , mn(10 ) , and mn(12 ) sites predominantly contribute to the homo [ fig . [ fig : ldos](d ) ] rather than the lumo [ fig . [ fig : ldos](c ) ] , and so the broadening of the homo level is much larger than that of the lumo level [ table 1 and fig . [ fig : trc](b ) ] . in fig . [ fig : trc](b ) , the transmission peak associated with the homo level for geo 2:au - s@xmath41 ( geo 2:au - h - no - linker ) , is very broad because it corresponds to four ( two ) molecular levels broadened right below @xmath43 including the homo level . in the case of physisorption , [ fig : geo](g ) and ( i ) , the mn(6 ) , mn(8 ) , mn(10 ) , and mn(12 ) sites are much closer to the electrodes than the mn(5 ) , mn(7 ) , mn(9 ) , and mn(11 ) sites . the homo originates mainly from the former four mn @xmath19 orbitals , while the lumo comes from the latter four mn @xmath19 orbitals . thus , for the geometries in the orientation ( 2 ) , either in the case of chemisorption or physisorption , the broadening of the homo level is expected to be much larger than that of the lumo level [ table 1 and fig . [ fig : trc](b ) ] . h@xmath14)@xmath15 , and onto ( c ) the lumo and ( d ) homo for geo 2:au - h - no - linker . the lumo and homo originate from the majority - spin mn @xmath19 orbitals . the criterion of the isosurface shown is @xmath67 electrons/@xmath68 , where @xmath69 is the bohr radius . the numbering of the mn ions is based on fig . [ fig : geo ] . the blobs far away from the core of the mn@xmath0 ( marked by l ) in ( a ) and ( b ) arise from the linker molecules.,title="fig:",width=294,height=181 ] h@xmath14)@xmath15 , and onto ( c ) the lumo and ( d ) homo for geo 2:au - h - no - linker . the lumo and homo originate from the majority - spin mn @xmath19 orbitals . the criterion of the isosurface shown is @xmath67 electrons/@xmath68 , where @xmath69 is the bohr radius . the numbering of the mn ions is based on fig . [ fig : geo ] . the blobs far away from the core of the mn@xmath0 ( marked by l ) in ( a ) and ( b ) arise from the linker molecules.,title="fig:",width=294,height=181 ] h@xmath14)@xmath15 , and onto ( c ) the lumo and ( d ) homo for geo 2:au - h - no - linker . the lumo and homo originate from the majority - spin mn @xmath19 orbitals . the criterion of the isosurface shown is @xmath67 electrons/@xmath68 , where @xmath69 is the bohr radius . the numbering of the mn ions is based on fig . [ fig : geo ] . the blobs far away from the core of the mn@xmath0 ( marked by l ) in ( a ) and ( b ) arise from the linker molecules.,title="fig:",width=200,height=181 ] h@xmath14)@xmath15 , and onto ( c ) the lumo and ( d ) homo for geo 2:au - h - no - linker . the lumo and homo originate from the majority - spin mn @xmath19 orbitals . the criterion of the isosurface shown is @xmath67 electrons/@xmath68 , where @xmath69 is the bohr radius . the numbering of the mn ions is based on fig . [ fig : geo ] . the blobs far away from the core of the mn@xmath0 ( marked by l ) in ( a ) and ( b ) arise from the linker molecules.,title="fig:",width=177,height=160 ] orbitals for ( a ) geo 1:au-(c@xmath13h@xmath14)@xmath15 and ( b ) geo 1:au-(auc@xmath13h@xmath14)@xmath15 . refer to fig . [ fig : geo ] for numbering of the mn ions . the densities of states for mn(3 ) and mn(4 ) sites are the same as those for mn(1 ) and mn(2 ) sites.,title="fig:",width=294,height=226 ] orbitals for ( a ) geo 1:au-(c@xmath13h@xmath14)@xmath15 and ( b ) geo 1:au-(auc@xmath13h@xmath14)@xmath15 . refer to fig . [ fig : geo ] for numbering of the mn ions . the densities of states for mn(3 ) and mn(4 ) sites are the same as those for mn(1 ) and mn(2 ) sites.,title="fig:",width=294,height=226 ] now we compare the level broadening for the geometries in the orientation ( 1 ) to that in the orientation ( 2 ) , for a given bonding type and fixed distance @xmath19 . for this comparison , we examine the following three geometries with the au - s bonding : one geometry in the orientation ( 1 ) , geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow , and two geometries in the orientation ( 2 ) , geo 2:au - s@xmath41 and geo 2:au-(sc@xmath15h@xmath41)@xmath41 . the distance @xmath19 in the former geometry , 25.73 , is much longer than those for the latter two geometries . thus , we extrapolate to find the broadening of a corresponding geometry in the orientation ( 2 ) with @xmath19=25.73 . the mn@xmath0 and the linker molecules play the role of a potential barrier in the electron transport , and so the broadening of the relevant molecular levels decays exponentially as a function of @xmath19 . thus , we apply @xmath70=@xmath71 , where @xmath72 and @xmath73 are positive constants , to the two geometries in the orientation ( 2 ) . extracting the values of @xmath72 and @xmath73 , we find that the extrapolated value of @xmath70 for @xmath19=25.73 is 0.0015 ev , which is less than the actual value of @xmath70 for the first geometry in the orientation ( 1 ) , 0.0020 ev . when the same logic is applied to the homo level , the extrapolated value of @xmath74 for such a geometry becomes 0.0069 ev,@xcite which is much greater than the actual value of @xmath74 for the first geometry , 0.00050 ev . consequently , for a given bonding type and fixed distance @xmath19 , the broadening of the lumo ( homo ) level for geometries with the orientation ( 1 ) is more ( less ) pronounced than that with the orientation ( 2 ) . combining this result with the finding that the lumo level broadens more than the homo level for the geometries with the orientation ( 1 ) and the opposite holds for those with the orientation ( 2 ) , we reach the following conclusion . although more statistics is desirable and other bonding types can be tested , our calculations indicate that as long as the current flows through the lumo level , ( for a given @xmath19 and bonding type ) geometries with the orientation ( 1 ) provide somewhat a higher current than those with the orientation ( 2 ) . however , if one can arrange the transport to occur through the homo level ( by application of gate voltage to the mn@xmath0 ) , geometries in the orientation ( 2 ) will give a much higher current than those in the orientation ( 1 ) . one caveat is that for a short distance @xmath19 the size of the mn@xmath0 prohibits interface geometries with the orientation ( 1 ) from being formed . orbitals for ( a ) geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow and ( b ) geo 1:au-(c@xmath13h@xmath14)@xmath15 : majority - spin ( black ) and minority - spin ( red ) . refer to fig . [ fig : geo ] for numbering of the mn ions . the dos projected onto the mn(1 ) and mn(2 ) sites are similar to those onto the mn(3 ) and mn(4 ) sites.,title="fig:",width=294,height=226 ] orbitals for ( a ) geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow and ( b ) geo 1:au-(c@xmath13h@xmath14)@xmath15 : majority - spin ( black ) and minority - spin ( red ) . refer to fig . [ fig : geo ] for numbering of the mn ions . the dos projected onto the mn(1 ) and mn(2 ) sites are similar to those onto the mn(3 ) and mn(4 ) sites.,title="fig:",width=294,height=226 ] orbitals for ( a ) geo 2:au-@xmath75 and ( b ) geo 2:au - h - no - linker : majority - spin ( black ) and minority - spin ( red ) . refer to fig . [ fig : geo ] for numbering of the mn ions . the dos projected onto the mn(1 ) and mn(2 ) sites are similar to those onto the mn(3 ) and mn(4 ) sites.,title="fig:",width=294,height=226 ] orbitals for ( a ) geo 2:au-@xmath75 and ( b ) geo 2:au - h - no - linker : majority - spin ( black ) and minority - spin ( red ) . refer to fig . [ fig : geo ] for numbering of the mn ions . the dos projected onto the mn(1 ) and mn(2 ) sites are similar to those onto the mn(3 ) and mn(4 ) sites.,title="fig:",width=294,height=226 ] orbitals for geo 2:au-(sc@xmath15h@xmath41)@xmath41 : majority - spin ( black ) and minority - spin ( red ) . refer to fig . [ fig : geo ] for numbering of the mn ions . the dos projected onto the mn(1 ) and mn(2 ) sites are similar to those onto the mn(3 ) and mn(4 ) sites.,width=294,height=226 ] for all of the geometries considered , the effect of spin filtering remains robust . the spin - filtering effect occurs in the transport through the mn@xmath0 because its majority - spin homo and lumo levels are well separated from the minority - spin homo and lumo levels and because the minority - spin homo level is located sufficiently below the majority - spin homo level ( fig . 2 in ref . [ ] , figs . [ fig : pdos - sp1 ] , [ fig : pdos - sp2 ] , and [ fig : pdos - sp3 ] ) . this feature of the homo and lumo does not change with interface geometry . however , the degree of the spin filtering depends on molecular orientations , bonding types , and the distance @xmath19 . we first compare the spin - filtering effect for the geometries in the orientation ( 1 ) to that in the orientation ( 2 ) . figure [ fig : trc - log ] shows the spin - polarized @xmath45 for six geometries [ fig . [ fig : geo](a ) , ( c ) , ( d ) , ( e ) , ( h ) , and ( i ) ] . in the last column of table 1 , the ratio of the majority - spin to the minority - spin transmission coefficients at the energy level associated with the lumo level , @xmath76 , is provided . for the geometries in the orientation ( 1 ) , the ratio varies from 6.11@xmath77@xmath78 to 5.26@xmath77@xmath79 , while for those in the orientation ( 2 ) , the ratio is in the range of 7.39@xmath77@xmath80 to 2.86@xmath77@xmath81 . a larger spin - filtering effect is obtained for the geometries in the orientation ( 1 ) than for those in the orientation ( 2 ) . this is partly because the linker molecules in the orientation ( 2 ) are in a closer proximity to the mn(1 ) , mn(2 ) , mn(3 ) , and mn(4 ) sites than those in the orientation ( 1 ) . these four mn sites mainly contribute to the minority - spin density of states as shown in figs . [ fig : pdos - sp1 ] and [ fig : pdos - sp2 ] . due to this proximity of the four mn sites , the minority - spin dos for the geometries in the orientation ( 2 ) is more spread than that for the geometries in the orientaion ( 1 ) . an additional reason is as follows . as discussed earlier and shown in figs . [ fig : pdos - sp1 ] , [ fig : pdos - sp2 ] , and [ fig : pdos - sp3 ] , the lumo arises from the majority - spin mn @xmath19 orbitals . at the energy level where the ratio is obtained , the majority - spin transmission is of the order of unity , independent of the distance @xmath19 , because the transport is carried by the lumo level . however , at that energy level , the transport of minority - spin electrons is in the tunneling regime , since there are no corresponding molecular levels of the minority - spin . thus , the minority - spin transmission decreases exponentially with @xmath19 . since the geometries in the orientation ( 1 ) have much longer distances @xmath19 than those in the orientation ( 2 ) , the coefficients of the minority - spin transmission for the former geometries are much smaller than those for the latter geometries . this renders the greatly enhanced ratio for the geometries in the orientation ( 1 ) . now let us briefly discuss the influence of bonding type on the spin - filtering effect . for the orientation ( 1 ) the minority - spin dos for the au - c bonding is slightly broader than that for the au - s bonding [ compare fig . [ fig : pdos - sp1](a ) to ( b ) ] . thus , the ratio for the au - c bonding is lower than that for the au - s bonding ( table 1 ) for a given distance @xmath19 . similarly , for the orientation ( 2 ) , the minority - spin dos for the au - s bonding is more delocalized than that for the au - h bonding [ compare fig . [ fig : pdos - sp2](a ) to ( b ) ] . so the ratio for the au - s bonding is lower than that for the au - h bonding for a given distance @xmath19 . for a specific bonding type , the ratio increases with increasing the distance @xmath19 because of the reason explained above . for example , for the au - s bonding , the minority - spin pdos for geo 2:au-(sc@xmath15h@xmath41)@xmath41 is much more localized than that for geo 2:au - s@xmath41 [ compare fig . [ fig : pdos - sp3 ] to [ fig : pdos - sp2](a ) ] . thus , the ratio for the former is two orders of magnitude greater than that for the latter . additionally , we note from figs . [ fig : trc ] and [ fig : trc - log ] that this ratio , @xmath76 , remains in the range of 100 to 10@xmath82 even at the energy level where the transport of both the majority spin and the minority spin is in the tunneling regime . the spin - filtering effect may be achieved for other smms with high magnetic moments and stable ground - state spin multiplets . we investigate the effect of binding site on transport properties for a given bonding type , for example , the au - s bonding in the orientation ( 1 ) , geo 1:au-(sc@xmath13h@xmath14)@xmath15 . we consider hollow and on - top binding sites of the s atoms to the au surface in that geometry , without geometry relaxation . the broadening of the lumo level somewhat decreases from the hollow - site to the on - top geometries , while the broadening of the homo level slightly increases ( table 1 ) . this difference in the broadening is much smaller than that induced by different bonding types . despite this small difference , the overall features of @xmath45 for the two geometries are fairly similar to each other , as shown in fig . [ fig : trc](a ) . thus , the @xmath36-@xmath83 curve for the on - top site geometry is expected to be similar to that for the hollow - site geometry , geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow , shown in fig . [ fig : iv ] . we compare the transport properties of the geometries with three bonding types in the orientation ( 1 ) , where the distance @xmath19 is fairly similar to one another : au - c bonding [ geo 1:au-(c@xmath13h@xmath14)@xmath15 ] , au - s bonding [ geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow ] , and au - au bonding [ geo 1:au-(auc@xmath13h@xmath14)@xmath15 ] . the broadening of the lumo level for the geometry with the au - c bonding is one order of magnitude larger than that with the au - s bonding , and it is twice as large as that with the au - au bonding , as shown in figs . [ fig : trc](a ) and [ fig : pdos - geo1a ] . accordingly , the current for the au - c bonding at 50 mv is one order of magnitude higher than that for the au - s bonding , and is twice as high as that for the au - au bonding ( fig . [ fig : iv ] ) . among all of the interface geometries considered , for a fixed distance @xmath19 , the geometry with the au - c bonding reveals the largest broadening of the lumo level , leading to the largest current ( fig . [ fig : iv ] ) . to understand the nature of the bonding between the linker molecules and the electrodes , we compute the density of states projected onto the @xmath17 orbitals of the s or c atoms of the linker molecules closest to the electrodes , and onto the @xmath18 orbitals of the three au surface atoms closest or bonded to those s or c atoms . the dos projected onto the au @xmath18 orbitals change with interface geometries due to the interactions between the linker molecules and the electrodes . for geo 1:au-(c@xmath13h@xmath14)@xmath15 , only the @xmath85 orbitals of the c atoms of the linker group greatly overlap with the au @xmath18 orbitals of the electrodes near @xmath43 [ the bottom four panels of fig . [ fig : pdos - s](a ) ] . the density of the lumo coincides with the peaks in the region bound by the dashed lines in fig . [ fig : pdos - s](a ) . strong overlap ( in terms of the peak height and the width of the region ) among the au @xmath18 orbitals , the c @xmath17 orbitals , and the lumo results in a high current through the mn@xmath0 . for geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow , the @xmath86 , @xmath87 , and @xmath85 orbitals of the s atoms have much weaker overlap with the au @xmath18 orbitals in the region where the lumo appears ( confined by the dashed lines ) than the au - c bonding case [ the top four panels of fig . [ fig : pdos - s](a ) ] , which gives rise to a reduced transmission probability . for geo 2:au - s@xmath41 , only the @xmath86 and @xmath87 orbitals of the s atoms bear some overlap with the au @xmath18 orbitals near @xmath43 ( in the region bound by the dashed lines ) [ the top four panels of fig . [ fig : pdos - s](b ) ] . this overlap is weaker than that for the au - c bonding but stronger than that for geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow or for geo 2:au-(sc@xmath15h@xmath41)@xmath41 [ the bottom four panels of fig . [ fig : pdos - s](b ) ] . this tendency agrees well with the trend in the value of @xmath62 and the current . additionally , our calculated spatially resolved density of states of the lumo [ fig . [ fig : ldos](a ) ] uncovers that the geometry with the au - c bonding gives rise to larger density in the linker molecules than any other geometries considered . thus , the transmission probability is highest for the geometry with the au - c bonding . the geometries with the au - s bonding do not have as high transmission probabilities as that for the au - c bonding or the au - au bonding . consequently , the au - c bonding provides the highest current and the au - s bonding gives rise to a lower current than the au - c bonding or the au - au bonding . orbitals of the three au surface atoms and @xmath86 , @xmath87 , @xmath85 orbitals of the s or c atom closest or bonded to the au atoms for ( a ) geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow and geo 1:au-(c@xmath13h@xmath14)@xmath15 , and ( b ) geo 2:au - s@xmath41 and geo 2:au-(sc@xmath15h@xmath41)@xmath41 . in the region confined by the dashed lines , the density peaks overlap with the lumo . the vertical scale in this figure differs from those in figs . [ fig : pdos - geo1 ] and [ fig : pdos - geo1a].,title="fig:",width=294,height=226 ] orbitals of the three au surface atoms and @xmath86 , @xmath87 , @xmath85 orbitals of the s or c atom closest or bonded to the au atoms for ( a ) geo 1:au-(sc@xmath13h@xmath14)@xmath15-hollow and geo 1:au-(c@xmath13h@xmath14)@xmath15 , and ( b ) geo 2:au - s@xmath41 and geo 2:au-(sc@xmath15h@xmath41)@xmath41 . in the region confined by the dashed lines , the density peaks overlap with the lumo . the vertical scale in this figure differs from those in figs . [ fig : pdos - geo1 ] and [ fig : pdos - geo1a].,title="fig:",width=294,height=226 ] we discuss the transport properties of the geometries with physisorption and without linker molecules . one of the geometries without linker molecules , geo 2:au - h - no - linker , has the shortest distance @xmath19 among the geometries considered . the broadening of the lumo level for that geometry is 54% of that with the shortest au - s bonding ( geo 2:au - s@xmath41 ) , and it is 21% of that with the au - c bonding , as shown in table 1 . the current for the geometry with the au - h bonding at 50 mv is 44% of that for geo 2:au - s@xmath41 , and it is 36% of that with the au - c bonding ( fig . [ fig : iv ] ) . in the case of physisorption , with the fixed distance @xmath19 , the broadening of the lumo level for geo 2:phys - no - linker is the same as that for geo 2:phys-(ch@xmath13)@xmath41 . this implies that the alkane chains do not play a role in the broadening of the lumo level . for the given distance @xmath19 , the broadening of the lumo level for these two physisorbed geometries takes up only 15% of that for the au - s bonding ( geo 2:au - s@xmath41 ) , as listed in table 1 . yet , the current for geo 2:phys-(ch@xmath13)@xmath41 at 50 mv is two orders of magnitude lower than that for geo 2:au - s@xmath41 ( fig . [ fig : iv ] ) . thus , in this case , the ratio of the values of @xmath70 for the two geometries does not have the same order as the ratio of the corresponding currents . the discrepancy between the two ratios arises from the difference in the response of the junctions or in the changes of the transmission with bias voltage . notice that fig . [ fig : trc ] shows the transmission spectra for zero bias . for chemisorbed junctions the molecular orbitals strongly overlap with the orbitals of the electrodes . thus , when bias voltage is applied , this overlap is enhanced especially in the regions between adjacent transmission peaks . however , for physisorbed junctions , the transmission peaks remain sharp even with bias voltage because of no overlap of the molecular orbitals with the orbitals of the electrodes . a combination of this effect with the ratio of the values of @xmath70 explains the ratio of the current through geo 2:phys-(ch@xmath13)@xmath41 to that through geo 2:au - s@xmath41 . our overall comparison of the transport behavior among different bonding types summarizes that the current through the mn@xmath0 decreases in the following order , for a given distance @xmath19 : au - c , au - au , au - s , au - h , and physisorption . we simulated a single - molecule junction based on the smm mn@xmath0 connected to au(111 ) electrodes via five different bonding mechanisms and two molecular orientations , and calculated their transport properties , using the non - equilibrium green s function method and the spin - polarized dft . geometry relaxation of the junction renders a small decrease in the electric current at low bias voltage . if the electron transport is carried by the lumo ( homo ) level of the mn@xmath0 , geometries in the orientation ( 1 ) provide a higher ( lower ) current that those in the orientation ( 2 ) . this is due to the structure of the mn@xmath0 . our two - terminal transport calculations reveal that the lumo level is relevant to the transport through the mn@xmath0 . in all of the interface geometries considered , the spin - filtering effect discussed in ref . [ ] remains robust , and it may occur to some other smms where the ground - state spin is large and the ground - state spin multiplet is reasonably well separated in energy from the low - lying excited spin multiplets . some experimental studies @xcite show that upon deposition of mn@xmath0 molecules on an au surface , the valence states of all of the mn ions are preserved , while some other experimental works @xcite reveal that such a deposition induces changes in the valence states of some of the mn ions . in cases where such changes do not occur @xcite , the spin - filtering effect is expected . we also find that the current through the mn@xmath0 depends on a bonding type and that it decreases in the following order for a fixed @xmath19 : au - c , au - au , au - s , au - h , and physisorption . this is because the overlap among the lumo , the s or c @xmath17 orbitals from the linker molecules , and the @xmath18 orbitals of the au surface atoms , decreases in that order . the degeneracy in the magnetic levels of the mn@xmath0 for a given spin @xmath88 is lifted by spin - orbit coupling . then inelastic transport through low - energy spin excitations must be included for quantitative comparison to experimental data . however , the general trend in the transport as a function of bonding type and interface geometry found in this study may still hold . is supported by nsf dmr-0804665 and the jeffress memorial trust funds . thanks the spanish ministerio de ciencia e innovacin for a juan de la cierva fellowship and the marie curie european itns funmols and nanoctm for funding . is supported by mec fis2006 - 12117 . computational support was provided by the sgi altix linux supercluster ( cobalt ) and intel 64 cluster ( abe ) at the national center for supercomputing applications under dmr060011 and by virginia tech linux clusters and advanced research computing . z. salman , k. h. chow , r. i. miller , a. morello , t. j. parolin , m. d. hossain , t. a. keeler , c. d. p. levy , w. a. macfarlane , g. d. morris , h. saadaoui , d. wang , r. sessoli , g. g. condorelli , and r. f. kiefl , nano lett . * 7 * , 1551 ( 2007 ) . u. del pennino , v. d. renzi , r. biagi , v. corradini , l. zobbi , a. cornia , d. gatteschi , f. bondino , e. magnano , m. zangrando , m. zacchigna , a. lichteinstein , and d. w. boukhvalov , surf . * 600 * , 4185 ( 2006 ) . j. r. friedman , m. p. sarachik , j. tejada , and r. ziolo , phys . lett . * 76 * , 3830 ( 1996 ) ; l. thomas and f. lionti and r. ballou and d. gatteschi and r. sessoli and b. barbara , nature , * 383 * , 145 ( 1996 ) . j. m. soler , e. artacho , j. d. gale , a. garca , j. junquera , p. ordejn , and d. snchez - portal , j. phys . : condens . matter * 14 * , 2745 ( 2002 ) ; j. junquera , . paz , d. snchez - portal and e. artacho , phys . b * 64 * , 235111 ( 2001 ) ; p. ordejn , d. a. drabold , m. p. grumbach and r. m. martin , phys . rev . b * 51 * , 1456 ( 1995 ) . d. w. boukhvalov , m. al - saqer , e. z. kurmaev , a. moewes , v. r. galakhov , l. d. finkelstein , s. chuizbaian , m. neumann , v. v. dobrovitski , m. i. katsnelson , a. i. lichtenstein , b. n. harmon , k. endo , j. m. north , and n. s. dalal , phys . rev . b * 75 * , 014419 ( 2007 ) . h. j. eppley , h .- tsai , n. de vries , k. folting , g. christou , and d. n. hendrickson , j. am . * 117 * , 301 - 317 ( 1995 ) ; m. soler , w. wernsdorfer , k. a. abboud , j. c. huffman , e. r. davidson , d. n. henrickson , and g. christou , j. am . soc . * 125 * , 3576 ( 2003 ) . r. basler , a. sieber , g. chaboussant , h. gdel , n. e. chakov , m. soler , g. christou , a. desmedt , and r. lechner , inorg . chem . * 44 * , 649 ( 2005 ) . for geo 2:au - s@xmath41 , the broadening of the homo level of the mn@xmath0 is buried under the single broad transmission peak ( width of 0.11 ev ) associated with four molecular orbitals right below the fermi level including the homo level . thus , in order to compute @xmath72 and @xmath73 for the homo level , we assume that the value of @xmath74 is @xmath89 ev . .average broadening of the homo and lumo levels ( @xmath63 , @xmath62 ) of the mn@xmath0 for the nine interface geometries . all geometries are optimized , unless stated otherwise . the broadening is calculated from zero - bias transmission spectra . the value of @xmath63 for geo 2:au - s@xmath41 ( geo 2:au - h - no - linker ) is marked by @xmath90 because it represents the full width at half maximum of the single broad transmission peak associated with the four ( two ) molecular levels right below @xmath43 including the homo level , as shown in fig . [ fig : trc](b ) . the ratio of the majority - spin to the minority - spin transmission , @xmath76 , is computed at the energy level corresponding to the lumo level . [ cols="^,^,^,^,^,^,^ " , ]

several variations of the shiryaev roberts detection procedure in the context of the simple changepoint problem are considered : starting the procedure at ( the original shiryaev roberts procedure ) , at for fixed , and at that has the quasi - stationary distribution . comparisons of operating characteristics are made . the differences fade as the average run length to false alarm tends to infinity . it is shown that the shiryaev roberts procedures that start either from a specially designed point or from the random `` quasi - stationary '' point are third - order asymptotically optimal . sequential analysis , sequential changepoint detection , shiryaev roberts procedure 62l10 , 62l15 , 60g40

1005.1129

the simple changepoint problem posits that one obtains a series of observations @xmath5 such that @xmath6 are independent and , for some value @xmath7 , @xmath8 ( the changepoint ) , @xmath9 have known density @xmath10 and @xmath11 have known density @xmath12 ( @xmath13 means that all observations have density @xmath10 and @xmath14 means that all observations have density @xmath12 ) . the changepoint @xmath7 is unknown , and the sequence @xmath15 is being monitored for detecting a change . a sequential detection policy is defined by a stopping time @xmath16 ( with respect to the @xmath17 s ) , so that after observing @xmath18 it is declared that apparently a change is in effect . by @xmath19 we denote the probability measure generated by the observations @xmath17 when the changepoint is @xmath7 and @xmath20 stands for the corresponding expectation . the notation @xmath13 , @xmath21 and @xmath22 correspond to the no - change scenario . in other words , under @xmath21 the observations @xmath23 are i.i.d . with density @xmath10 and under @xmath24 the observations @xmath23 are i.i.d . with density @xmath12 ( both with respect to a dominating measure @xmath25 ) . common operating characteristics of a detection policy @xmath16 are @xmath26 , the average run length ( expected time ) to false alarm ( assuming there is no change ) , and @xmath27 , the maximal expected delay to detection . subject to a lower bound @xmath28 on @xmath26 , the goal is to minimize the maximum expected delay . note that a uniformly optimal procedure that minimizes the expected detection delay @xmath29 for all @xmath8 over stopping times with @xmath30 does not exist , and we have to resort to the minimax setting . in 1961 , for the problem of detecting a change in the drift of a brownian motion , shiryaev introduced a detection procedure that is a limit of the bayes procedure when the parameter of the exponential prior distribution tends to zero ; see shiryaev @xcite . in particular , certain optimality properties of this procedure were established by shiryaev @xcite . in discrete time , a similar procedure was first considered by roberts @xcite as a particular case of the girschick - rubin @xcite bayesian procedure by setting the parameter of the geometric prior distribution to zero . therefore , this procedure is usually referred to as the shiryaev roberts procedure . the shiryaev roberts ( sr ) procedure and its modifications are the centerpiece of this paper . specifically , let @xmath31 denote the likelihood ratio for the observation @xmath32 , and define @xmath33 where @xmath34 is a positive threshold that controls the false alarm rate . for a connection between @xmath34 and the expected time to false alarm @xmath35 see pollak @xcite . when defining stopping times we always assume that @xmath36 , i.e. , @xmath37 if @xmath38 never reaches level @xmath34 . note the recursion @xmath39 ( with the null initial condition ) . pollak @xcite tweaked the procedure by starting it off at a random @xmath40 whose distribution is the quasi - stationary distribution @xmath41 of the sr statistic @xmath38 , defined by @xmath42 and showed that the stopping time @xmath43 where @xmath44 minimizes the maximal expected delay @xmath45 asymptotically as @xmath46 to within @xmath47 over all stopping times that satisfy @xmath48 , where @xmath34 is such that @xmath49 . we will refer to this randomized sr procedure as the shiryaev roberts pollak ( srp ) procedure . usually , @xmath50 can not be expressed in a closed form ( except in some rare cases ) . to compute @xmath50 and make the srp procedure implementable , moustakides et al . @xcite proposed a numerical framework . until recently the question whether the srp procedure is exactly optimal ( in the class of procedures with @xmath48 ) was an open question . moustakides et al . @xcite present numerical evidence that there exist procedures that are uniformly better . they regard starting off the original sr procedure at a fixed ( but specially designed ) @xmath51 , @xmath52 and defining the stopping time with this new deterministic initialization @xmath53 where @xmath54 they show by numerical examples that , for certain values of @xmath4 , apparently @xmath55 for all @xmath56 , where @xmath57 and @xmath34 are such that @xmath58 ( although the maximal expected delay is only slightly smaller for @xmath59 ) . we will refer to the procedure defined in and as the sr@xmath4 procedure . in @xcite , it is conjectured that the sr@xmath4 procedure with a specially designed @xmath60 is third - order asymptotically optimal ( i.e. , to within @xmath47 ) in the class of procedures with @xmath48 as @xmath61 . examples where the sr@xmath4 procedure is strictly minimax are provided by polunchenko and tartakovsky @xcite and tartakovsky and polunchenko @xcite . shiryaev @xcite showed for brownian motion that if a change takes place after many successive applications ( re - runs ) of a stopping time @xmath16 ( to a sequence @xmath5 , starting anew after each false alarm ) , then the expected delay is minimized asymptotically as @xmath62 ( i.e. , in a stationary mode ) over all multi - cyclic procedures with @xmath48 for every @xmath63 by the original ( multi - cyclic ) sr procedure . pollak and tartakovsky @xcite showed the same for discrete time . the goal of the present paper is to answer questions regarding comparisons between the various sr - type procedures introduced above the sr , sr@xmath4 , and srp procedures . is the stationary expected delay of the repeated sr procedure described in the previous paragraph similar to @xmath64 ? ( yes , see theorem [ th2 ] , theorem [ th3 ] and corollary [ cor1 ] . ) what can be said about the maximal expected detection delays of these detection procedures ? ( the srp procedure and the sr@xmath4 procedure with a specially designed @xmath4 are third - order asymptotically minimax , i.e. , to within a negligible term @xmath65 . see theorem [ th4 ] . this answer justifies the conjecture of moustakides et al . what can be said about @xmath64 , @xmath66 , and @xmath67 when all have the same average run length to false alarm @xmath28 ? ( the average delay to detection at infinity is the smallest for the original sr procedure @xmath68 , but the difference between them is @xmath47 as @xmath61 . see theorems [ th5 ] and [ th4 ] . ) we conclude with a numerical example that illustrates these phenomena . recall that @xmath7 denotes the unknown changepoint which is identified with the last time instant under the nominal regime . thus , conditional on @xmath69 , the joint density of the vector @xmath70 can be written as @xmath71 for any @xmath72 and @xmath73 provided that @xmath74 whenever @xmath75 . given observations @xmath76 , introduce the hypotheses @xmath77 that the change occurs somewhere within this stretch of observations and @xmath78 that there is no change . clearly , the latter hypothesis is equivalent to the hypothesis @xmath79 , @xmath80 . according to , the likelihood ratio of these hypotheses is @xmath81 where @xmath31 . therefore , the sr statistic can be interpreted as the average likelihood ratio averaged over a uniform improper prior distribution of the changepoint . by @xmath16 we denote a generic stopping time ( or a detection procedure ) and by @xmath82 the class of detection procedures ( stopping times ) for which the average run length ( arl ) to false alarm does not fall below a given number @xmath83 . the following two objects will be of the main interest in this paper : _ supremum average delay to detection _ @xmath84 and the limiting value of the average detection delay which we will refer to as _ average delay to detection at infinity _ @xmath85 as we mentioned in the introduction , we are interested in a minimax setting of minimizing the maximal expected delay @xmath86 over stopping times with the lower bound on the arl to false alarm @xmath30 , i.e. , in finding a procedure that would minimize @xmath86 in the class @xmath87 : @xmath88 . however , in general we are unable to find an exact solution to this problem for every @xmath63 and , for this reason , we focus on asymptotic solutions for a large arl to false alarm @xmath28 ; see polunchenko and tartakovsky @xcite and tartakovsky and polunchenko @xcite for examples where an exact minimax solution is available . we call the procedure @xmath89 * _ first - order asymptotically optimal _ * if @xmath90 i.e. , @xmath91 where @xmath92 as @xmath61 . we call the procedure @xmath93 * _ second - order asymptotically optimal _ * if @xmath94 where @xmath95 is bounded as @xmath96 . we call the procedure @xmath93 * _ third - order asymptotically optimal _ * if @xmath97 where @xmath47 tends to zero as @xmath96 . it follows from pollak @xcite that the srp procedure is third - order asymptotically optimal whenever @xmath98 . in section [ ss : add ] we prove the third - order asymptotic optimality property under the stronger second moment condition @xmath99 using different techniques . the second moment condition allows us to obtain higher - order asymptotic approximations for @xmath100 and @xmath101 ( up to a vanishing term ) . since the srp procedure is an equalizer , i.e. , @xmath102 does not depend on @xmath7 , it is sufficient to evaluate the average run length to detection @xmath103 assuming that the change is in effect from the very beginning . more importantly , using the ideas of moustakides et al . @xcite , we are able to design the initialization point @xmath4 in the sr@xmath4 procedure , which may or may not depend on the false alarm constraint @xmath28 , so that this procedure is also third - order asymptotically optimal . in this respect , the average delay to detection at infinity @xmath104 plays a critical role . to understand why , let us look at figure [ fig : fig1 ] which shows the average detection delay @xmath105 versus @xmath7 for several initialization values @xmath4 . this figure was obtained using integral equations and numerical techniques of moustakides et al . @xcite . for @xmath106 , this is the classical sr procedure whose average detection delay is monotonically decreasing to its minimum that is attained at infinity ( a steady state value ) . it is seen that there exists a value @xmath107 that generally may depend on the threshold @xmath34 for which the worst point @xmath7 is at infinity , i.e. , @xmath108 . this is a very important observation , since it allows us to build a proof of asymptotic optimality based on an estimate of @xmath109 . particular choices of the head start " @xmath110 will be discussed in the following sections . for various initialization strategies.,scaledwidth=70.0% ] the monotonicity of the curve for the average detection delay of the sr procedure allows us also to conclude ( intuitively only since this is only a numerical observation and there is no theoretical justification of monotonicity ) that the asymptotic lower bound for @xmath101 can be evaluated based on the value of @xmath111 . asymptotically @xmath103 , @xmath112 , and @xmath111 are the same since the mean of the quasi - stationary distribution is of order @xmath113 and the values of the head start @xmath114 that lead to the almost optimal performance are either fixed ( i.e. , @xmath115 ) or go to infinity in such a way that @xmath116 as @xmath117 , as we will see from the following study . in this section , we discuss the asymptotic behavior of the sr@xmath4 and srp detection procedures for large values of the threshold @xmath34 and the arl to false alarm @xmath28 . let @xmath118 denote the log - likelihood ratio for the @xmath119-th observation and let @xmath120 . introduce a one - sided stopping time @xmath121 let @xmath122 be an overshoot ( excess over the level @xmath123 at stopping ) , and let @xmath124 , \quad \varkappa = \lim_{a\to\infty } { { \mathsf{e}}}_0 \kappa_a . \ ] ] the constants @xmath125 and @xmath126 depend on the model and can be computed numerically . in general , @xmath127 and @xmath128 . [ th1]assume that @xmath107 where @xmath110 is either fixed or , more generally , @xmath129 in such a way that @xmath130 as @xmath117 . then for the sr@xmath4 procedure , uniformly in @xmath131 , @xmath132 where the constant @xmath125 is defined in . for the srp procedure @xmath133 the proof of this theorem is given in the appendix . let @xmath134 denote a random variable that asymptotically as @xmath135 has the same @xmath21-distribution as @xmath136 , i.e. , @xmath137 where @xmath138 is called the stationary distribution of @xmath136 . recall also that we denote by @xmath139 the quasi - stationary distribution of @xmath136 . we always assume that both quasi - stationary @xmath50 and stationary @xmath138 distributions exist , which is always true when @xmath140 is continuous . note that the process @xmath141 is a zero mean @xmath21-martingale and , hence , applying the optional sampling theorem yields @xmath142 , which can be used to approximate @xmath143 . using the above fact along with theorem [ th1 ] , for practical purposes we suggest the following approximations @xmath144 and @xmath145 where @xmath146 is the mean of the quasi - stationary distribution . note that the mean of the quasi - stationary distribution is of order @xmath113 as @xmath117 . indeed , by kesten ( * theorem 5 ) , @xmath147 which along with the fact that @xmath148 ( cf . pollak and siegmund @xcite ) yields @xmath149\,dx { \leqslant}\int_0^a [ 1-{{\mathsf{q}}}_{\mathrm{st}}(x)]\,dx = o(\log a).\ ] ] since @xmath150 as @xmath117 , it follows that @xmath151 , where @xmath152 as @xmath117 , so that @xmath153 we continue with obtaining asymptotic approximations ( as @xmath117 ) for the average delay to detection @xmath154 , including the case of the large changepoint @xmath7 , i.e. , for @xmath104 , as well as with deriving an asymptotic lower bound for @xmath101 . this will allow us to ascertain whether the sr@xmath4 procedure with a certain initialization @xmath4 ( which is either fixed or may depend on @xmath28 ) is third - order asymptotically optimal as @xmath96 recall that @xmath155 is the log - likelihood ratio for the observation @xmath32 , @xmath156 and @xmath126 is the limiting average overshoot in the one - sided test @xmath157 defined in . let @xmath158 and let @xmath159 . let @xmath160 denote the kullback leibler information number . [ lem1 ] let @xmath161 and assume that @xmath162 is non - arithmetic . let @xmath163 be such that @xmath164 and @xmath165 as @xmath166 for some @xmath167 . let @xmath168 and let @xmath169 be defined as in . then , as @xmath117 , @xmath170 } \bigg\ } + o(1 ) , \end{aligned}\ ] ] where @xmath65 as @xmath171 uniformly on @xmath172 . the proof of lemma [ lem1 ] is given in the appendix . [ rem3.1 ] let @xmath173 . note that @xmath174 is independent of @xmath175 and has the same @xmath19-distribution of all @xmath176 , i.e. , it is distributed as @xmath177 under @xmath24 . recall that by @xmath134 we denote a random variable that has the @xmath21-limiting ( stationary ) distribution of @xmath38 as @xmath178 , i.e. , @xmath179 . let @xmath180 = \int_0^\infty \int_0^\infty \log(1+x+y)\,d{{\mathsf{q}}}_{\mathrm{st}}(x ) \,d\widetilde{{{\mathsf{q}}}}(y),\ ] ] and @xmath181 = \int_0^\infty \log(1+r+y)\,d\widetilde{{{\mathsf{q}}}}(y),\ ] ] where @xmath182 . the following theorem , whose proof is based on lemma [ lem1 ] and can be found in the appendix , provides asymptotic approximations ( for large @xmath34 ) for the average delay to detection of the sr@xmath4 procedure ( for large @xmath7 and @xmath14 ) , and for the supremum average delay to detection @xmath183 of the srp procedure ( within vanishing terms @xmath47 ) . [ th2 ] if @xmath161 and @xmath162 is non - arithmetic , then for any @xmath168 @xmath184 } + o(1 ) \quad \text{as $ a\to \infty$},\ ] ] and @xmath185 } + o(1 ) \quad \text{as $ a\to \infty$},\ ] ] where @xmath65 as @xmath117 . define @xmath186 the following lemma provides the lower bound for the supremum average delay to detection in the class @xmath87 . this bound will be used to obtain an asymptotic lower bound in theorem [ th3 ] and for the proof of third - order asymptotic optimality of detection procedures in theorem [ th4 ] . [ lem2 ] let @xmath68 be the stopping time of the sr procedure that starts from zero and let the threshold @xmath187 be chosen so that @xmath188 . the following lower bound holds : @xmath189 obviously , for any stopping time @xmath16 @xmath190 { { \mathsf{p}}}_\infty(t > k)}{{{\mathsf{e}}}_\infty t}\\ & { \geqslant}\frac{\sum_{k=0}^\infty { { \mathsf{e}}}_k(t - k|t > k ) { { \mathsf{p}}}_\infty(t > k)}{{{\mathsf{e}}}_\infty t } = { { \mathcal{j}}}(t ) . \end{aligned}\ ] ] as follows from pollak and tartakovsky @xcite , the right hand side is minimized by the sr stopping time @xmath191 , so that @xmath192 and the proof is complete . the following theorem provides the asymptotic approximation for the lower bound @xmath193 . its proof is given in the appendix . [ th3 ] let @xmath194 be defined as in and @xmath195 as in . if @xmath161 and @xmath162 is non - arithmetic , then @xmath196 where @xmath65 as @xmath117 . theorem [ th3 ] also allows for the following interpretation . consider the following multi - cyclic detection procedure . let a stopping time @xmath16 be applied repeatedly after each alarm , so that @xmath197 are independent copies of @xmath16 and @xmath198 is the time interval between the @xmath199th and @xmath200th alarms . clearly , the number @xmath201 of false alarms before the changepoint @xmath7 is @xmath202 and the real change occurring at the point @xmath203 is detected at the time @xmath204 . for any fixed @xmath56 , the average delay to detection of the multi - cyclic ( repeated ) detection procedure is @xmath205 . assuming that the change occurs at a far time horizon ( i.e. , @xmath206 ) , introduce the _ stationary average detection delay _ @xmath207 by applying renewal theory , it can be shown that @xmath208 . to this end , see pollak and tartakovsky @xcite . furthermore , the sr procedure is exactly optimal in the sense of minimizing the stadd . therefore , the following corollary is a direct consequence of theorem [ th2 ] and theorem [ th3 ] . [ cor1]if @xmath162 is non - arithmetic and @xmath161 , then , as @xmath117 , @xmath209 and @xmath210 the following theorem establishes asymptotic optimality of the srp and sr@xmath4 detection procedures under moderate conditions . its proof is immediate from the above results . [ th4 ] let @xmath211 and let @xmath162 be non - arithmetic . \(i ) then @xmath212 } + o(1 ) \quad \text{as $ \gamma\to \infty$}.\ ] ] \(ii ) if in the srp procedure @xmath213 , then @xmath214 and @xmath215 } + o(1 ) \quad \text{as $ \gamma\to \infty$}.\ ] ] therefore , the srp procedure is asymptotically third - order optimal in the class @xmath87 : @xmath216 \(iii ) if in the sr@xmath4 procedure @xmath217 , and the initialization point @xmath4 is either fixed or tends to infinity with the rate @xmath218 and is selected so that @xmath219 , then @xmath220 and @xmath221 } + o(1 ) \quad \text{as $ \gamma\to \infty$}.\ ] ] therefore , the sr@xmath4 procedure is asymptotically third - order optimal : @xmath222 the asymptotic lower bound follows from lemma [ lem2 ] and theorem [ th3 ] . the asymptotic approximations and ( and statements of ( ii ) and ( iii ) in whole ) follow from theorems [ th1 ] and [ th2 ] . [ rem3.2]third - order asymptotic optimality of the srp procedure follows from pollak @xcite ( under the sole first moment condition ) , so that this result is not new . however , higher - order asymptotic approximation for the sadd is new . [ rem3.3 ] feasibility of selecting @xmath4 so that @xmath223 follows from numerical experiments performed by moustakides et al . @xcite as well as from the example in section [ s : example ] . see figure [ fig : fig1 ] in section [ s : prelim ] and figures [ fig : oc_vs_k ] and [ fig : srp_sr_r_addk_vs_k_arl_100 ] in section [ s : example ] . [ rem3.4 ] since for the sr procedure @xmath224 , it follows from theorem [ th2 ] ( setting @xmath106 in ) that @xmath225 where @xmath226 $ ] . since @xmath227 implies @xmath228 , it follows that with this choice of threshold @xmath229 + o(1 ) \quad \text{as $ \gamma \to \infty$}.\ ] ] comparing with the lower bound shows that @xmath230 thus , the sr procedure is only second - order asymptotically optimal and the difference is approximately equal to @xmath231 . this difference can be quite large when detecting small changes ( i.e. , when @xmath232 is small ) . it is worth noting that theorem [ th2 ] suggests that if the initializing point @xmath107 is selected from the equation @xmath233 , then for the large arl to false alarm @xmath28 the values of the average delays to detection at zero and infinity are approximately equal , @xmath234 \approx { { \mathsf{add}}}_\infty(t_a^{r^*})$ ] ( to within small terms @xmath47 ) . this choice of the head - start is intuitively appealing since we intend to make the sr@xmath235 procedure look like an equalizer as much as possible . obviously , the value of @xmath110 does not depend on @xmath28 , i.e. , it is a fixed number that depends on the model . this observation will be further elaborated in section [ s : example ] . note also that the fact that the limiting value @xmath236 equating @xmath234 $ ] and @xmath237 to within @xmath47 is a fixed number has been first noticed by moustakides and tartakovsky @xcite for the problem of detecting a change in the drift of a brownian motion . also , although starting at @xmath110 causes for a faster initial response than starting at @xmath106 , the resemblance to lucas and crosier s fir scheme is secondary : their method is designed to give a really fast initial response , whereas our goal is to attain asymptotic third - order optimality . it is interesting to ask how the average detection delays at infinity @xmath111 , @xmath104 , and @xmath238 are related when all three procedures have the same average run length to false alarm @xmath28 . it turns out that the @xmath239 is the smallest for the original sr procedure @xmath68 . theorem [ th5 ] below proves this statement . note also that by theorems [ th2 ] and [ th4 ] the difference between add s of all three procedures is @xmath47 as @xmath96 . this result can be proven in two steps : 1 ) to show that the arl to false alarm @xmath240 of the srp procedure is increasing in @xmath34 , the threshold ( the fact that the arl to false alarm of the sr@xmath4 procedure is increasing in @xmath34 for a fixed @xmath4 is obvious ) ; and 2 ) to show that the average delay @xmath238 of the srp procedure is increasing in @xmath34 ( obviously , the add s at infinity are the same for all three procedures when the same threshold is being used ) . since the sr procedure requires the lowest threshold to attain the same false alarm rate , this implies that the sr procedure has the lowest @xmath239 . we believe that @xmath240 and @xmath238 are both increasing in @xmath34 in the general case . however , we are able to prove this fact only when the cumulative distribution function of @xmath241 is concave , both pre - change and post - change , something that guarantees monotonicity properties of the markov detection statistics . this is restrictive , but it does hold , for example , in detection of a shift of a normal mean and in detection of a change of the parameter of an exponential distribution . it also holds for the example considered in the next section . for @xmath242 , regard the sequence defined by the recursion @xmath243 to prove the required result we need the following lemma whose proof can be found in the appendix . [ lem3 ] let @xmath244 be a cumulative distribution function of @xmath241 that is log - concave ( i.e. , @xmath245 is a concave function ) . then the process @xmath246 that has transition probabilities @xmath247 is a stochastically monotone markov process , i.e. , @xmath248 is non - decreasing and right - continuous in @xmath249 for all @xmath250 . [ rem3.5 ] note that the normal cumulative distribution function @xmath251 is log - concave , so the log - likelihood ratio of two normals whose means differ has a log - concave cdf . the same applies to two differing exponential distributions as well as to two differing beta - distributions considered in section [ s : example ] . we are now prepared to state the desired result . the details of the proof are given in the appendix . [ th5 ] suppose that the cdf of @xmath241 is log - concave both pre - change and post - change . let @xmath252 be fixed , and let @xmath253 be such that the arl to false alarm of the sr@xmath4 procedure @xmath254 is @xmath28 . then @xmath255 is an increasing function of @xmath4 and @xmath256 where @xmath257 is such that @xmath258 . in order to implement the asymptotic approximations we have to be able to compute the constants @xmath259 and @xmath195 defined by @xmath260 where @xmath261 is the stationary distribution of the sr statistic @xmath262 under @xmath21 and @xmath263 is the limiting distribution of @xmath264 under @xmath24 . assume that the distribution of @xmath140 is continuous . then for computing the constants we need to evaluate the two densities @xmath265 and @xmath266 . let @xmath134 and @xmath177 be random variables that are the limit ( in distribution , as @xmath267 ) of @xmath38 and @xmath268 , respectively , which have densities @xmath269 and @xmath270 . to find the desired densities , observe that , by recursion , @xmath134 and @xmath271 have the same density @xmath269 under @xmath21 . similarly @xmath177 and @xmath272 have the same density @xmath270 under @xmath24 . to see this note that , by the i.i.d . property of the data , @xmath268 has the same @xmath24-distribution as the random variable @xmath273 , which follows the recursion @xmath274 . therefore , we have the following integral equations for these densities @xmath275dy;~~ \tilde{q}(x)=-\int_0^\infty \tilde{q}(y)\left[\dfrac{\partial}{\partial x}f_0\left(\frac{1+y}{x}\right)\right]dy,\ ] ] where @xmath276 and @xmath277 are the corresponding distribution functions of the likelihood ratio @xmath140 . thus , @xmath269 and @xmath270 are the eigenfunctions corresponding to the unit eigenvalues of the linear operators defined , respectively , with the kernels @xmath278 the constants @xmath259 and @xmath195 are then obtained , usually by numerical integration . the next section offers a comparative performance analysis for an example where @xmath259 and @xmath195 are computable analytically . to verify the accuracy of the asymptotic approximations , we carried out an extensive performance evaluation of the procedures discussed in the earlier sections for the following example . suppose @xmath279 is a series of independent observations such that @xmath280 are @xmath281 each , and @xmath282 are @xmath283 each . put another way , the series undergoes a sudden and abrupt shift in the expected value from @xmath284 pre - change to @xmath285 post - change , while retaining the variance . to be specific , our goal is to verify the conditions and the accuracy of the asymptotic approximations given in theorem [ th2 ] , theorem [ th4 ] , and remark [ rem3.4 ] , i.e. , @xmath286 } , \\ { { \mathsf{sadd}}}(t_a ) & \approx \frac{1}{i}{\left[\log a+\varkappa - c_0\right ] } , \end{aligned}\ ] ] and the approximations for the arl to false alarm given in , i.e. , @xmath287 where @xmath288 is the mean of the quasi - stationary distribution . to undertake this task , it is necessary to be able to calculate the constants @xmath259 , @xmath195 , @xmath125 , and @xmath126 and also to compute the initialization point @xmath4 and the mean @xmath288 of the quasi - stationary distribution @xmath41 . while usually the constants @xmath259 and @xmath195 can be evaluated only numerically or by monte carlo , it turns out that for the @xmath289-model considered these constants are computable analytically . the pre- and post - change probability densities for this scenario are @xmath290 respectively , and the likelihood ratio for the @xmath291th observation is @xmath292 . the quasi - stationary distribution satisfies the integral equation @xmath293 where @xmath294 and @xmath295 since for any @xmath296 @xmath297 which is continuously differentiable with respect to @xmath249 , the equation for the quasi - stationary distribution @xmath50 for this model is @xmath298 note that the quasi - stationary distribution converges to the stationary distribution @xmath150 as @xmath117 ( cf . pollak and siegmund @xcite ) . equation can not be solved analytically for an arbitrary _ finite _ @xmath34 , but its limiting value when @xmath117 ( i.e. , the stationary distribution of the statistic @xmath136 ) does permit a closed - form solution . by , the stationary distribution @xmath138 satisfies the following equation @xmath299 and the solution is @xmath300{{\one_{\{x{\geqslant}0\}}}}$ ] . to derive the equation for @xmath301 , observe first that for @xmath296 @xmath302 which is identical to @xmath303 . as a result , the distribution @xmath301 satisfies precisely the same equation as @xmath138 and , therefore , @xmath304 using and , one is able to calculate the constants @xmath259 and @xmath195 _ exactly _ as @xmath305 in particular , @xmath306 . note that the kullback - leibler information number @xmath307 , so that @xmath308 unfortunately , neither the limiting average overshoot @xmath126 nor the limiting average `` exponential '' overshoot @xmath125 are computable exactly . monte carlo simulations with @xmath309 trials have been used to estimate the two as @xmath310 and @xmath311 with the standard error less than @xmath312 . specifically , these estimates were obtained from the formulas @xmath313 , \quad \zeta=\frac{1}{i}\exp\left\{-2\sum_{k=1}^\infty\frac{1}{k}\biggl[{{\mathsf{p}}}_0(s_k{\leqslant}0)+{{\mathsf{p}}}_\infty(s_k>0)\biggr]\right\},\ ] ] where @xmath314 , @xmath315 , and @xmath316 $ ] ( see , e.g. , woodroofe @xcite ) . the first fraction in the first formula is computable analytically . the only issue is the infinite sum . to evaluate this sum it was first truncated at @xmath317 . an extreme - value - theoretic argument shows that the weight of the dropped tail is of order of the machine precision . this makes it safe to assume that the sum of the first @xmath317 terms is effectively equal to the original infinite sum . the second source of errors is the expectations under the sum . these expectations are not computable analytically , and therefore monte carlo simulations were used : we generated @xmath309 trajectories of @xmath318 , and performed averaging across the trajectories to find @xmath319 $ ] for each @xmath320 . the sum in the second formula was evaluated in a similar manner . despite the fact that in the example considered the distributions @xmath138 and @xmath301 are obtainable exactly , neither the quasi - stationary distribution , required for the srp procedure , nor the conditional average delay to detection @xmath29 for @xmath8 and the arl to false alarm seem feasible to get analytically . to overcome this difficulty , these quantities were computed numerically , using the approach undertaken by moustakides et al . @xcite with the number of breakpoints set at @xmath321 , high enough to ensure a relative error in the order of a fraction of a percent . specifically , let @xmath322 . for @xmath323 , @xmath324 , and @xmath56 , define @xmath325 , @xmath326 $ ] , @xmath327 , and @xmath328 . using the markov property of the statistic @xmath136 , the following integral equations and recursions for operating characteristics are obtained by moustakides et al . @xcite : @xmath329dx , \quad j=0,\infty , \\ \delta_\nu(r ) & = \int_0^a\delta_{\nu-1}(x)\left[\dfrac{\partial}{\partial x } f_\infty\left(\dfrac{x}{1+r}\right)\right]dx,\quad \nu{\geqslant}1 , \\ p_\nu(r ) & = \int_0^ap_{\nu-1}(x)\,\frac{\partial}{\partial x}f_\infty\left(\frac{x}{1+r}\right)dx , \quad \nu{\geqslant}1 \end{aligned}\ ] ] with the initial conditions @xmath330 and @xmath331 . these integral equations yield the arl to false alarm @xmath143 and the sequence of average detection delays @xmath332 , @xmath333 as functions of the starting point @xmath324 . the distribution function @xmath334 is defined in and @xmath335 ^ 2 $ ] , @xmath336 . in order to implement the srp procedure @xmath337 as well as to evaluate its performance , we need to compute the quasi - stationary distribution @xmath50 , which satisfies the integral equation . the arl to false alarm @xmath240 and the average detection delay @xmath103 of the srp procedure are then computed as @xmath338\,d{{\mathsf{q}}}_a(r)=\int_0^a \phi_j(r)\,d{{\mathsf{q}}}_a(r),\quad j=\infty,0.\ ] ] we recall that the srp procedure is the equalizer , so that @xmath339 for all @xmath340 . finally , by moustakides et al . @xcite , the lower bound @xmath193 given in lemma [ lem2 ] is computed as @xmath341 , where @xmath342 is the solution of the integral equation @xmath343 } \,dx.\ ] ] the above integral equations allow us to compute numerically operating characteristics of both sr@xmath4 and srp procedures as well as the mean of the quasi - stationary distribution @xmath288 . of the quasi - stationary distribution @xmath50 as a function of the detection threshold @xmath34 . the @xmath344 function is plotted to demonstrate that @xmath345.,scaledwidth=90.0% ] at this point the only unresolved question is that of how to choose @xmath4 . to this end , several options are proposed by moustakides et al . @xcite ; one of the options is @xmath346 . recall that theorem [ th4 ] requires ( a ) @xmath347 as @xmath166 and ( b ) @xmath348 . if @xmath346 , then condition ( a ) is satisfied , since by and , @xmath349 and @xmath345 . this is also illustrated in figure [ fig : mu_a_a ] , which shows that the inequality @xmath350 and the equality @xmath345 are indeed satisfied . the condition ( b ) is also satisfied even for small values of the arl to false alarm , as can be seen from figure [ fig : oc_vs_k ] which shows how @xmath29 evolves as @xmath7 runs from 0 to 10 for the srp procedure and for the sr@xmath4 procedure with @xmath346 . the arl to false alarm is about 100 for both procedures . observe that the sr@xmath4 procedure attains supremum at infinity , i.e. , as @xmath206 . also , the stationary regime kicks in as early as at @xmath351 , and this is for @xmath352\approx 100 $ ] . in addition , figure [ fig : oc_vs_k ] illustrates theorem [ th5 ] @xmath353 is indeed the smallest for the sr procedure , while the difference is small . we iterate that it is easily shown that the log - concavity conditions of theorem [ th5 ] hold in the example considered , i.e. , @xmath354 $ ] and @xmath355 $ ] are concave functions . table [ tab : oc_vs_arl ] provides values of the supremum average delay to detection @xmath356 and the lower bound @xmath193 versus the arl to false alarm @xmath352 $ ] . also presented in parentheses are the corresponding theoretical predictions made based on the asymptotic approximations and . its is seen that the approximations for the arl to false alarm are fairly accurate even for small values of the arl such as 50 , while the approximations for the @xmath356 and the lower bound become accurate for the moderate false alarm rate ( @xmath357 and higher ) . of the sr , srp and sr@xmath4 ( @xmath346 ) procedures . the arl to false alarm @xmath358\approx 100$].,scaledwidth=95.0% ] [ cols="<,^,^,^,^,^,^",options="header " , ] [ tab : oc_vs_arl ] another possible way of starting the sr@xmath235 procedure is from the value of @xmath4 for which the average detection delay at the point @xmath14 is equal ( at least approximately ) to the @xmath239 ( i.e. , in the steady - state mode ) , as has been proposed in section [ ss : add ] . in the asymptotic setting this is equivalent to finding a point @xmath107 for which @xmath359 is equal to @xmath259 . clearly , @xmath110 is a _ fixed _ number that does not depend on @xmath34 since @xmath359 does not depend on @xmath4 and @xmath34 . using , we obtain the transcendental equation @xmath360 and the solution is @xmath361 . for the srp procedure and for the sr@xmath4 procedure with @xmath362 . the arl to false alarm @xmath358\approx 100$].,scaledwidth=95.0% ] figure [ fig : srp_sr_r_addk_vs_k_arl_100 ] shows the average delay to detection @xmath363 $ ] versus the changepoint @xmath7 for the sr@xmath235 procedure with @xmath364 and for the srp procedure . observe that for the sr@xmath235 procedure the average delay at @xmath14 is equal to that at infinity , as was planned . more importantly , the point @xmath14 is the worst ( supremum ) point ( along with large @xmath7 ) . also , it can be seen that the sr@xmath235 procedure is _ uniformly _ ( i.e. , for all @xmath8 ) better than the srp procedure , although in this example the difference is practically negligible . we also note that this initialization is better than starting off at the mean of the quasi - stationary distribution , while the difference in performance is very small the @xmath356 is equal to @xmath365 for @xmath346 and @xmath366 for @xmath107 . this allows us to conclude that the sr@xmath235 is robust with respect to the initialization point in a certain range . we considered three different versions of the shiryaev roberts procedure , with the difference being the starting point the conventional sr procedure , where @xmath0 , pollak s modification of this procedure , where @xmath3 is sampled from the quasi - stationary distribution of the sr statistic , and a generalization where @xmath367 is a specially designed deterministic number , proposed by moustakides et al . @xcite . for each of the procedures we derived asymptotic formulas for operating characteristics when the threshold @xmath34 is high and showed that asymptotically when the arl to false alarm is large the sr@xmath4 and srp procedures are both asymptotically third - order minimax . we emphasize that third - order asymptotic optimality of the sr@xmath4 procedure has been established under the conjecture that the worst changepoint is at infinity , which is justified numerically for several examples , including the one considered in section [ s : example ] . unfortunately , we have not been able to prove this conjecture analytically . in addition , we performed a comparative efficiency analysis of the detection procedures to verify the accuracy of the asymptotic approximations and demonstrated the proximity of the latter to the real values for a specific example . the results of numerical analysis allow us to conclude that ( in the minimax sense ) performance of the sr@xmath4 procedure that starts with the mean of the quasi - stationary distribution as well as with a point that equates the average detection delay at zero and infinity is almost indistinguishable from that of the srp procedure . we are grateful to a referee for useful suggestions that have improved the presentation in the article . [ lemsmax ] let @xmath368 be i.i.d . with @xmath369 and @xmath370 . let @xmath371 . then for all @xmath372 @xmath373 { } 0.\ ] ] applying doob s maximal submartingale inequality to the submartingale @xmath374 , we obtain @xmath375 } \\ & = \frac{1}{\varepsilon^2 n } { { \mathsf{e}}}{\left[{\left(\frac{s_n^2}{n}\right ) } { { \one_{\{\max\limits_{n { \leqslant}n } s_{n } { \geqslant}\varepsilon n\}}}}\right ] } . \end{aligned}\ ] ] first , it follows that @xmath376 { } 0 .\ ] ] now , we show that @xmath377 } \xrightarrow[n\to\infty ] { } 0,\ ] ] which implies that @xmath378 as @xmath379 , i.e. , the desired result . finally , for any @xmath383 we have @xmath384}+{}\\ & \qquad\qquad\qquad{}+{{\mathsf{e}}}{\left[{\left(\frac{s_n^2}{n}-l\wedge\frac{s_n^2}{n}\right)}{{\one_{\{\max\limits_{n{\leqslant}n } s_n > \varepsilon n\}}}}\right ] } \\ & { \leqslant}l{{\mathsf{p}}}{\left(\max\limits_{n{\leqslant}n } s_n > \varepsilon n \right)}+ { { \mathsf{e}}}{\left(\frac{s_n^2}{n}-l\wedge\frac{s_n^2}{n}\right ) } \\ & { \leqslant}\frac{l\sigma^2}{\varepsilon^2 n}+\sigma^2-{{\mathsf{e}}}{\left(l\wedge\frac{s_n^2}{n}\right ) } \\ & \xrightarrow[n\to\infty]{}\sigma^2-{{\mathsf{e}}}{\left(l\wedge\xi^2\sigma^2\right ) } \xrightarrow[l\to\infty]{}\sigma^2(1 - 1)=0 . \end{split}\end{aligned}\ ] ] observe that @xmath390 and @xmath391 hence , @xmath392 on @xmath393 , which implies that @xmath394 therefore , we have the following chain of equalities and inequalities : @xmath395 where the last inequality stems from @xmath396 ( so that @xmath397 ) and from @xmath398 } | t_{a - m } > t_a^r { \geqslant}m\ } \\ & { \leqslant}{{\mathsf{e}}}_\infty r_{t_a - m } = { { \mathsf{e}}}_\infty t_{a - m}. \end{aligned}\ ] ] note that @xmath399 is a nonnegative @xmath21-martingale with mean 1 , so that @xmath400 and we obtain @xmath401 let @xmath402 and @xmath403 so that @xmath404 and @xmath405 ( which can always be arranged ) . then , uniformly in @xmath406 , @xmath407 which along with the reverse inequality proves asymptotic equality whenever @xmath408 as @xmath117 , and if @xmath110 does not depend on @xmath34 the result obviously holds . similar to , @xmath409 thus , for the srp procedure , by conditioning on @xmath40 , we obtain @xmath410 where @xmath411 is the mean of the quasi - stationary distribution . by , @xmath412 and to obtain it suffices to take @xmath413 ( say ) . therefore , nonlinear renewal theory can be applied to the sequence @xmath421 note also that @xmath422 so that the sequence @xmath423 satisfies the conditions of the nonlinear renewal theorem of woodroofe ( * ? ? ? * theorem 4.5 ) uniformly in @xmath175 . indeed , the sequence @xmath424 is slowly changing and converges @xmath425a.s . ( as @xmath267 ) to the finite random variable @xmath426 . the nonlinear renewal theorem yields the following asymptotic approximation : @xmath427 } \bigg\ } + o(1 ) , \end{aligned}\ ] ] where @xmath65 as @xmath171 uniformly on @xmath172 . note that all the necessary conditions of this theorem hold trivially . the only condition that requires checking is the following : for some @xmath372 , @xmath428 { } 0 \quad \text{on $ \{r_\nu^r < a / n_a\}$ } .\ ] ] let @xmath429 , @xmath430 , and @xmath431 . changing the measure @xmath432 , we obtain that for any @xmath433 and @xmath434 @xmath435 } \\ & { \geqslant}{{\mathsf{e}}}_\nu\big[\exp{\left\{-s_{t_a^r}^{\nu+1}\right\ } } { { \one_{\ { { { { \mathcal{b}}}}_{a,\varepsilon},s_{t_a^r}^{\nu+1 } { \leqslant}c\ } } } } | t_a^r > \nu , r_\nu^r\big ] \\ & { \geqslant}e^{-c } \big [ { { \mathsf{p}}}_\nu{\left(t_a^r { \leqslant}\nu + l | t_a^r > \nu , r_\nu^r\right ) } - \\ & \quad { { \mathsf{p}}}_\nu \big(\max_{1{\leqslant}n { \leqslant}l } s_{\nu+n}^{\nu+1 } > c | t_a^r > \nu , r_\nu^r \big)\big ] . \end{aligned}\ ] ] setting @xmath436 and noting that @xmath437 we obtain @xmath438 where @xmath439 it remains to show that @xmath443 as @xmath117 for some @xmath444 . note that @xmath445 by lemma [ lemsmax ] , @xmath446 { } 0,\ ] ] so that @xmath447 . thus , condition holds , the asymptotic approximation follows and the proof is complete . write @xmath450 . note that @xmath451 as @xmath117 . obviously , @xmath452 note that for large enough @xmath34 @xmath453 and there exists @xmath454 such that for all @xmath455 , by kesten ( * ? ? ? * theorem 5 ) , @xmath456 ( when @xmath457 first condition on @xmath458 , conditional on @xmath459 ) , so that the second term in the last equality in is @xmath47 . by lemma [ lem1 ] , the first term in the last equality in is equal to @xmath460 } \\ & \quad - { { \mathsf{e}}}_\nu{\left[\log{\left(1 + \frac{v_{\nu,\infty}}{1+r_\nu^r}\right ) } \bigg | t_a^r > \nu , r_\nu^r < l_a\right ] } \bigg \ } + o(1 ) . \end{aligned}\ ] ] now , @xmath461 } \\ & = \frac{{{\mathsf{e}}}_\infty { \left[\log(1+r_\nu^r ) | t_a^r > \nu\right ] } - { { \mathsf{e}}}_\infty { \left[\log(1+r_\nu^r ) | t_a^r > \nu , r_\nu^r { \geqslant}l_a\right ] } { { \mathsf{p}}}_\infty ( r_\nu^r { \geqslant}l_a| t_a^r > \nu)}{1-{{\mathsf{p}}}_\infty ( r_\nu^r { \geqslant}l_a| t_a^r > \nu ) } \end{aligned}\ ] ] where by @xmath462 . also , since @xmath463 on @xmath464 , @xmath465 } { \leqslant}\log ( 1+a),\ ] ] so that @xmath466 } & = { { \mathsf{e}}}_\infty { \left[\log(1+r_\nu^r ) | t_a^r > \nu\right ] } \\ & \times { \left[1 + \frac{{{\mathsf{p}}}_\infty(r_\nu^r { \geqslant}l_a|t_a^r > \nu)}{1-{{\mathsf{p}}}_\infty(r_\nu^r { \geqslant}l_a|t_a^r > \nu)}\right ] } + o(1 ) \\ & = { { \mathsf{e}}}_\infty [ \log(1+r_\nu^r ) | t_a^r > \nu ] + o(1 ) \end{aligned}\ ] ] ( since @xmath467<\log ( 1+a)$ ] ) . by remark [ rem3.1 ] , @xmath174 is independent of @xmath175 and distributed as @xmath177 under @xmath24 . since @xmath104 exists , it follows that @xmath468 } - { { \mathsf{e}}}{\left[\log { \left(1+\frac{v_\infty}{1+r_\infty}\right)}\right]}\right\ } } + o(1 ) \\ & = \frac{1}{i } { \left\ { \log a + \varkappa - { { \mathsf{e}}}{\left[\log { \left(1+r_\infty + v_\infty\right)}\right ] } \right\}}+o(1 ) \quad \text{as $ a \to \infty$ } \end{aligned}\ ] ] and the proof of is complete . it remains to prove the validity of asymptotic approximation . putting @xmath14 in , we obtain that the stopping time of the sr@xmath235 procedure can be written as @xmath469 where @xmath470 and @xmath471 is the random walk with the drift @xmath232 ( under @xmath24 ) . the sequence @xmath472 is slowly changing and converges @xmath473a.s . to the random variable @xmath474 whose expectation is equal to @xmath259 . the crucial condition @xmath475 { } 0 \quad \text{for some}~\varepsilon > 0\ ] ] holds in an analogous way that yields . therefore , nonlinear renewal theory can be applied to yield @xmath476 = \frac{1}{i } ( \log a + \varkappa - c_r ) + o(1 ) \quad \text{as}~ a\to\infty,\ ] ] and the proof is complete . recall that @xmath165 . we have @xmath477 since @xmath478 and since for a sufficiently large @xmath34 , @xmath479 for the first term , denoted as @xmath480 , we have @xmath481 { } 0,\ ] ] so that @xmath482 as @xmath166 . write @xmath483 by lemma [ lem1 ] , uniformly in @xmath484 @xmath485 therefore , for the second term ( denote it as @xmath486 ) we have @xmath487 { { \mathsf{p}}}_\infty(t_a>\nu ) } { { { \mathsf{e}}}_\infty t_a}-{}\\ & { } \qquad\qquad\qquad\qquad\qquad\qquad-\frac{\sum_{\nu=1}^{n_a } d_a { { \mathsf{p}}}_\infty(t_a>\nu ) } { { { \mathsf{e}}}_\infty t_a } \\ & = d_a + o(1)-\frac{d_a \sum_{\nu=1}^{n_a } { { \mathsf{p}}}_\infty(t_a>\nu)}{{{\mathsf{e}}}_\infty t_a } = d_a + o(1 ) , \end{aligned}\ ] ] where the last equality follows immediately from the fact that @xmath488 { } 0 , \ ] ] where the inequality holds for a sufficiently large @xmath34 . the required result follows . if @xmath245 is concave , then @xmath496 , so @xmath497 is a non - increasing function of @xmath250 . therefore , we have @xmath498 } { \leqslant}0 , \end{aligned}\ ] ] which implies that @xmath499 is a non - increasing function of @xmath249 . let @xmath244 denote the cdf of @xmath241 , and let @xmath500 denote probability when @xmath244 is the cdf of @xmath241 , when observations are iid . the following applies both for @xmath244 defined by @xmath501 as well as for @xmath244 defined by @xmath502 . under the assumption of the theorem , @xmath244 is log - concave . if @xmath503 , then @xmath504 for all @xmath505 , so that @xmath506 crosses @xmath34 no later ( and often earlier ) than @xmath507 . hence , @xmath143 is a decreasing function of @xmath4 and @xmath253 is an increasing function of @xmath4 . ( note that this is true in general , with no assumption of log - concavity . ) note that @xmath518 , and therefore , @xmath519 exits @xmath520 at the same time that @xmath521 exits @xmath522 regardless of the distribution of @xmath523 . it follows that @xmath524 , where @xmath525 and @xmath526 are the @xmath21-stationary distributions of @xmath519 and @xmath521 , which are also the corresponding @xmath21-quasi - stationary distributions of the markov processes @xmath527 and @xmath516 . let @xmath528 . in a manner similar to , @xmath529 where @xmath493 , @xmath530 and @xmath531 . writing @xmath532 we obtain that @xmath533 since ( by the same consideration as in the proof of lemma [ lem3 ] ) @xmath534 is non - increasing function of @xmath535 , it follows that @xmath536 construct a sample space where @xmath539 . by , @xmath540 is stochastically larger than @xmath541 . hence , one can construct the probability space so that also @xmath542 . now , due to stochastic monotonicity of the transition probabilities , @xmath543 is stochastically larger than @xmath544 , and one can construct the probability space so that also @xmath545 . continuing this inductively , one obtains a sample space where @xmath546 for all @xmath547 . under @xmath21 , @xmath548 and @xmath549 tend in distribution to the quasi - stationary distributions @xmath526 and @xmath550 , respectively , so it follows that @xmath551 for all @xmath250 . finally , consider the process @xmath552 governed by @xmath553 and started at @xmath554 and the process @xmath555 governed by @xmath556 and started at @xmath557 . in just the same way as above , we can construct a single probability space with @xmath558 for all @xmath547 . therefore , the process @xmath552 will exit above @xmath559 no later than the process @xmath555 , and therefore the expected exit time of @xmath552 will not exceed that of @xmath555 . but these expectations are arl s to false alarm if @xmath560 and @xmath239 if @xmath561 . furthermore , the arl to false alarm and the @xmath239 of @xmath562 ( where @xmath563 ) are equal to those of @xmath552 . clearly , the first exit time of @xmath562 from @xmath564 is nothing but the srp stopping time @xmath565 . hence , it follows that both the average delay to detection @xmath509 ( @xmath566 ) and the arl to false alarm @xmath510 of the srp procedure are increasing functions of @xmath34 , and the proof is complete .

we study a class of nonlinear nonparametric inverse problems . specifically , we propose a nonparametric estimator of the dynamics of a monotonically increasing trajectory defined on a finite time interval . under suitable regularity conditions , we prove consistency of the proposed estimator and show that in terms of-loss , the optimal rate of convergence for the proposed estimator is the same as that for the estimation of the derivative of a trajectory . this is a new contribution to the area of nonlinear nonparametric inverse problems . we conduct a simulation study to examine the finite sample behavior of the proposed estimator and apply it to the berkeley growth data . .1 in .1 in _ department of statistics , university of california , davis _ .1 in * keywords : * autonomous differential equation ; nonlinear inverse problem ; monotone trajectory ; nonparametric estimation ; perturbation theory ; spline

1408.5339

monotone trajectories describing the evolution of state(s ) over time appear widely in scientific studies . the most widely studied are probably growth of organisms such as humans or plants ( milani , 2000 ; erickson , 1976 ; silk and erickson , 1979 ) . there are many parametric models for describing the features of growth curves , particularly in human growth ( hauspie et al . , 1980 ; milani , 2000 ) . most of these works focus on modeling the trajectories themselves or modeling the rate of change , i.e. , the derivative of the trajectories . other examples of monotone trajectories appear in population dynamics under negligible resource constraints ( turchin , 2003 ) , in dose - response analysis in pharmacokinetics ( kelly and rice , 1990 ) , in auction price dynamics in e - commerce ( jank and shmueli , 2006 ; wang et al . , 2008 ; liu and mller , 2009 ) , and in analysis of trajectories of aircrafts after take - off ( nicol , 2013 ) . some of these works are looking at function estimation with monotonic constraints and some of them are taking a functional data analysis approach . in contrast , our goal here is to estimate the functional relationship between the rate of change and the state , i.e. , the dynamics of the trajectory , through a nonparametric model . many systems such as growth of organisms or economic activity of a country / region are intrinsically dynamic in nature ( cf . ljung and glad , 1994 ) . a dynamics model provides a mechanistic description of the system rather than a purely phenomenological one . moreover , due to insufficient scientific knowledge , quite often there is a need for nonparametric modeling of the dynamical system . in addition , nonparametric fits can be used to develop measures of goodness - of - fit for hypothesized parametric models . there is a large literature in modeling continuous time smooth dynamical systems through systems of parametric differential equations ( see , e.g. , perthame , 2007 , strogatz , 2001 ) . these methods have been used to model hiv dynamics ( wu , ding and degruttola , 1998 ; wu and ding , 1999 ; xia , 2003 ; chen and wu , 2008a , 2008b ) , the dynamic behavior of gene regulation networks ( gardner _ et al . _ , 2003 ; cao and zhao , 2008 ) , etc . approaches for fitting a parametric dynamics model include the maximum likelihood or nonlinear least squares . a recent approach proposed by ramsay _ ( 2007 ) and cao _ et al . _ ( 2008 ) for parametric ordinary differential equations is based on the idea of balancing the model fit and the goodness of fit of the trajectories simultaneously . another popular approach to fit dynamics models is a two - stage procedure ( chen and wu , 2008a , 2008b ; , varah , 1982 ) , where the trajectories and their derivatives are first estimated nonparametrically and then the dynamics is fitted by regressing the fitted derivatives to the fitted trajectories . the two - stage approach can be easily adapted to estimate a nonparametric dynamics model . however , their performance is unsatisfactory due the difficulty of resolving the bias - variance trade - off in a data dependent way . brunel ( 2008 ) gives a comprehensive theoretical analysis of such an approach . very recently , hall and ma ( 2014 ) proposed a one - step estimation procedure that mitigates some of the inefficiencies of two - stage estimators . however , this approach does not seem to extend naturally to estimate nonparametric dynamical systems . there is also an extensive literature on the nonparametric estimation of monotone functions , e.g. , brunk ( 1970 ) , wright and wegman ( 1980 ) , mammen ( 1991 ) , ramsay ( 1988 , 1998 ) . however , most methods in this field are not concerned with the estimation of the gradient function , except for ramsay ( 1998 ) where the unknown function is modeled in terms of a second order differential equation and a smoothed estimate of its gradient is obtained as a byproduct . a key observation of estimating the dynamics of monotone trajectories is that for any smooth monotone trajectory , its dynamics can be described by a first order autonomous differential equation . specifically , if @xmath1 is positive , strictly monotone and differentiable on a finite time interval , then we can express @xmath2\ ] ] where @xmath3 is the gradient function . in this paper , we estimate the unknown gradient function @xmath4 nonparametrically from discrete noisy observations of @xmath5 . specifically , we model the gradient function by a basis representation where the number of basis functions grow with the sample size . we adopt a nonlinear least squares framework for model fitting . we then carry out a detailed theoretical analysis and derive the rate of convergence of the proposed estimator . we now highlight the major contributions of this work . although there is a large literature on linear nonparametric inverse problems ( cavalier _ et al . _ , 2004 ; cavalier , 2008 ; donoho , 1995 ; johnstone _ et al . _ , 2004 ) , especially on the nonparametric estimation of the derivative of a curve ( gasser and mller , 1984 ; mller et al . , 1987 ; fan and gijbels , 1996 ) , there is little theoretical development on nonlinear nonparametric inverse problems . thus , our work makes a new contribution to this important area . in this paper , we first quantify the degree of ill - posedness of the estimation of the gradient function @xmath4 as the number of basis functions grow to infinity . we then use this result to show that if @xmath4 is @xmath6 times differentiable then the @xmath0-risk of the proposed estimator has the same optimal rate of convergence , viz . , @xmath7 , as that of the estimator of the derivative of a trajectory assuming that the latter is @xmath8 times differentiable . in section [ sec : discussion ] , we show that the optimal rate of the proposed estimator is indeed the minimax rate for estimation of @xmath4 under @xmath0 loss if the class of estimators is restricted to be uniformly lipschitz . in the rest of the paper , unless otherwise specified , the phrase `` optimal rate '' refers to the best rate of convergence of the proposed estimator . among the few instances of nonparametric modeling of the gradient function known to us , xue , miao and wu ( 2010 ) dealt with a related but different problem of estimating a parametric ode with time - varying parameters , where the latter are modeled as unknown smooth functions of time . in a work most closely related to ours , wu et al . ( 2014 ) proposed a sparse additive model for describing the dynamics of a multivariate state vector and developed a combination of two - stage smoothing and sparse penalization for fitting the model . their model can be seen as a multi - dimensional generalization of the autonomous ode model studied here . in their paper , while deriving the risk bounds , it is assumed that whenever the gradient function @xmath4 is @xmath6 times differentiable , the state @xmath5 is at least @xmath9 times differentiable . however , due to the representation @xmath3 , it follows that @xmath4 is @xmath6 times differentiable if and only if @xmath5 is @xmath8 times differentiable . therefore , at least for the one - dimensional state variable case , the assumptions made in wu et al . ( 2014 ) are not satisfied in reality if @xmath6 indeed denotes the maximal order of smoothness of @xmath4 . this indicates that the rate of convergence their estimator of @xmath4 is not optimal for the current problem . it is also instructive to note that , due to the assumption about the additional degree of smoothness of the state variable , wu et al . ( 2014 ) did not encounter the technical challenge posed by the ill - posedness of the problem . the rest of the paper is organized as follows . in section [ sec : model ] , we briefly describe the model and the estimation procedure . we present the main theoretical results in section [ sec : theory ] and outline the main steps of the proof in section [ sec : proofs ] . we present a simulation study in section [ sec : simulation ] and an application to the berkeley growth data in section [ sec : real ] . we discuss the optimality of the estimation of @xmath4 in section [ sec : discussion ] . some proof details are provided in the appendix ( section [ sec : appendix ] ) . some derivations and graphical summaries are provided in the supplementary material ( sm ) . the class of models studied in this paper is of the form : @xmath10,\ ] ] where @xmath4 is an unknown smooth function which is assumed to be positive on the range of @xmath11\}$ ] . therefore , the sample trajectory @xmath1 is a strictly increasing function of time @xmath12 . the observations are @xmath13 where @xmath14 are observation times . the noise terms @xmath15 s are assumed to be i.i.d . with mean 0 and variance @xmath16 . our goal is to estimate the gradient function @xmath4 based on the observed data @xmath17s . we propose to approximate @xmath4 through a basis representation : @xmath18 where @xmath19 is a set of linearly independent compactly supported smooth functions . henceforth , we use @xmath20 to denote @xmath21 . in practice , the initial value @xmath22 and the right boundary @xmath36 may not be observed or may be observed with noise . the choice of the endpoints of the combined support of the basis functions then becomes a delicate matter . this is because evaluation of the trajectory is an initial value problem , so error in @xmath37 propagates throughout the time domain . we discuss this in more details in the appendix ( particularly , see figure [ fig : trajectory_envelop ] ) . in the following , we propose a modified estimation procedure when @xmath37 and @xmath38 are unknown . the basic idea is to first estimate the trajectory at the endpoints of a slightly smaller time interval @xmath39 $ ] for a small positive constant @xmath40 , and then estimate the gradient function using data falling within this time interval . throughout the paper , @xmath40 is treated as a fixed quantity . in practice , we may select @xmath40 to be the time point such that about 5% of the data fall in the intervals @xmath41 $ ] and @xmath42 $ ] . too small a value of @xmath40 may cause distortions of the estimated @xmath4 at the boundaries . we first obtain nonparametric estimates of @xmath43 and @xmath44 , denoted by @xmath45 and @xmath46 , respectively . we then define @xmath47 and @xmath48 , where @xmath49 is a small positive number satisfying @xmath50 which implies that @xmath51 as @xmath52 goes to infinity . at the same time , @xmath49 should be large enough so that @xmath53 which ensures that @xmath54 and @xmath55 as @xmath52 goes to infinity . for some technical considerations , to be utilized later , we also want @xmath56 . in practice , we may select @xmath49 to be @xmath57 where @xmath58 is the length of the smallest support among the basis functions @xmath59 . for more details on how to obtain @xmath60 , @xmath61 , see lemma [ lem : consistency_x_j_hat ] in section [ sec : theory ] . in addition , we also assume that @xmath60 , @xmath61 are estimated from a sample independent from that used in estimating @xmath34 . this can be easily achieved in practice by sub - sampling of the measurements . this assumption enables us to prove the consistency result ( in section [ sec : theory ] ) conditionally on @xmath60 , @xmath61 and treating them as nonrandom sequences converging to @xmath62 , @xmath61 . in this section , we discuss the consistency of the estimator @xmath71 defined by the loss function ( [ eq : loss_modified ] ) . the asymptotic framework is that the number of basis functions @xmath35 goes to infinity together with the number of measurements @xmath52 . the consistency of the estimator @xmath68 over @xmath72 $ ] is formulated in terms of the @xmath0-loss as : @xmath73 in theorem [ thm : optimal_rate ] we derive a bound on the rate of convergence of @xmath68 in terms of the @xmath0-loss as @xmath74 that depends upon the degree of smoothness of @xmath4 . specifically , the optimal rate is @xmath75 for @xmath76 . the following assumptions are made on the model . * @xmath77 , and @xmath78 on @xmath79 for some integer @xmath80 , where @xmath79 is an open interval containing @xmath25 $ ] . * the collection of basis functions @xmath81 satisfies : * * @xmath21 s have unit @xmath0 norm ; * * the combined support of @xmath82 is @xmath83 $ ] and for every @xmath84 , the length of the support of @xmath21 is @xmath85 ; * * @xmath86 for all @xmath84 ; * * @xmath87 , for @xmath88 ; * * the gram matrix @xmath89 is such that there exist constants @xmath90 , not depending on @xmath35 such that @xmath91 for all @xmath35 ; * * for every @xmath35 , there is a @xmath92 such that @xmath93}|x_g(t ) - x(t;\bs\beta^*)| = o(m^{-(p+1)})$ ] and @xmath94 } |g^{(j)}(u ) - g_{\bs\beta^*}^{(j)}(u)| = o(m^{-p+j})$ ] for @xmath88 , where @xmath95 and @xmath96 with @xmath66 as in ( [ eq : initial_beta_delta ] ) . * time points @xmath97 are realizations of @xmath98 , where @xmath99 s are i.i.d . from a continuous distribution @xmath100 supported on @xmath101 $ ] with a density @xmath102 satisfying @xmath103 for some @xmath104 . * the noise @xmath15 s are i.i.d . sub - gaussian random variables ( cf . vershynin , 2010 ) with mean 0 and variance @xmath105 . we give brief explanations of these assumptions . * a1 * ensures sufficient smoothness of the solution paths of the differential equation ( [ eq : basic ] ) . also by * a1 * , @xmath106 is @xmath8 times continuously differentiable on @xmath79 . assumptions ( i ) to ( v ) of * a2 * are satisfied by b - spline basis , rescaled to have unit norm , of order @xmath107 , with equally spaced knots . define @xmath108 which is used in determining the rates of convergence of the estimator . thus , by making use of * a4 * , we get the following results with respect to the estimates of @xmath109 and @xmath110 ( cf . fan and gijbels , 1996 ) . [ lem : consistency_x_j_hat ] suppose that * a1 * and * a4 * hold . consider using a kernel of sufficient degree of smoothness to obtain estimates @xmath60 for @xmath62 , @xmath111 , through local polynomial method with bandwidth of order @xmath112 . define @xmath113 . then @xmath114 and given @xmath115 , there exists @xmath116 such that @xmath117 with probability at least @xmath118 , where @xmath119 is as in ( [ eq : xi_n ] ) . if @xmath120 ( as in theorem [ thm : consistency ] ) and @xmath121 for some @xmath122 , then we have @xmath123 as @xmath124 . this ensures that @xmath125 $ ] is within the interval @xmath25 $ ] a.s . for large enough @xmath52 and hence the properties of the function @xmath4 hold on @xmath126 . in addition , @xmath126 contains the interval @xmath72 $ ] . therefore condition ( ii ) in * a2 * ensures that the combined support of the basis functions covers the range of the data used in estimating @xmath4 . conditions ( i ) to ( v ) of * a2 * are satisfied by many classes of basis functions , including normalized b - spline basis of order @xmath107 with equally spaced knots in the interval @xmath127 $ ] . we show in appendix b that if the b - splines basis of order @xmath128 with equally spaced knots in @xmath127 $ ] , then ( vi ) of * a2 * is also satisfied . condition ( vi ) of * a2 * ensures that a solution @xmath30 of ( [ eq : initial_general ] ) on @xmath129 $ ] exists for all @xmath130 sufficiently close to @xmath131 . this allows us to apply the perturbation theory of differential equations to bound the fluctuations of the sample paths when we perturb the parameter @xmath34 . assumption * a3 * on the randomness of the sample points allows us to work with the random variables @xmath132 defined as @xmath99 conditional on @xmath133 $ ] with conditional density @xmath134 given by @xmath135 . the properties of @xmath102 ensure that @xmath134 satisfies the same property on @xmath136 $ ] with possibly modified values of the constants @xmath137 and @xmath138 . it should be noted that the key derivations leading to the consistency of @xmath68 are conditional on @xmath139 and therefore @xmath140 is only a convenient assumption for describing the regularity of the time points . the asymptotic results ( theorems [ thm : consistency ] and [ thm : optimal_rate ] ) hold if instead of being randomly distributed , the time points form a fixed regular grid , say , with equal spacing . as mentioned earlier , the estimation of @xmath141 is a nonlinear inverse problem since @xmath142 is not directly observable . in addition , this is also an ill - posed estimation problem . let @xmath143 be the partial derivative of @xmath144 with respect to @xmath130 , where @xmath145 is the solution of ( [ eq : initial_beta_delta ] ) with @xmath146 . let @xmath147 be as in * a2*. define @xmath148 where the expectation is with respect to the distribution of @xmath149 . clearly @xmath150 is a positive semi - definite matrix . it becomes clear from the analysis carried out later that the degree of ill - posedness of the estimation problem is determined by the size of the operator norm of the matrix @xmath151 as a function of @xmath35 . the following proposition gives a precise quantification of the degree of ill - posedness . the situation here is in contrast with standard nonparametric function estimation problems where the corresponding matrix is well - conditioned . [ prop : kappa ] assume that assumptions * a1 * to * a3 * hold with @xmath80 . assume further that ( a ) @xmath152 ( a.s . ) and ( b ) @xmath153 . then ( a.s . ) @xmath154 by lemma [ lem : consistency_x_j_hat ] and the discussion that follows , under the condition of theorem [ thm : consistency ] , ( a ) and ( b ) of proposition [ prop : kappa ] hold . we now state the main result on the consistency of the estimate @xmath68 . [ thm : consistency ] suppose that the observed data @xmath155 follow the model described by equations ( [ eq : basic ] ) and ( [ eq : data_model ] ) and that assumptions * a1**a4 * are satisfied with @xmath156 . suppose further that the sequence @xmath35 is such that @xmath157 for some @xmath158 , @xmath121 , and @xmath119 be as defined in lemma [ lem : consistency_x_j_hat ] . let @xmath159 for some @xmath160 ( sufficiently small ) and @xmath161 for some @xmath162 . then as @xmath163 , with probability tending to one , there exists a local minimum @xmath164 of the objective function @xmath165 ( defined through ( [ eq : loss_modified ] ) ) , which is also a global minimum within radius @xmath166 of @xmath167 ( note that , @xmath168 by ( [ eq : m_alpha_optimal ] ) ) such that , with @xmath169 , @xmath170 the proof of theorem [ thm : consistency ] is given in section [ sec : proofs ] . [ rem : consistency ] assuming @xmath171 to be a constant , if @xmath35 is chosen to be of the order @xmath172 , then @xmath173 in ( [ eq : consistency ] ) simplifies to @xmath174 , which is within a factor of @xmath175 of the optimal rate in terms of the @xmath0-loss for estimating @xmath142 based on the data @xmath155 given by ( [ eq : basic ] ) when @xmath176)$ ] . the fact that an estimator of @xmath4 can attain this rate can be anticipated from the representation of @xmath4 as @xmath3 . for @xmath76 , we can improve the rate of convergence of @xmath68 slightly further , by dropping the factor of @xmath175 , as stated in the following result . [ thm : optimal_rate ] suppose that the conditions of theorem [ thm : consistency ] are satisfied with @xmath76 and , further , the sequence @xmath35 satisfies the condition that @xmath177 for some @xmath178 . let @xmath68 be as in theorem [ thm : consistency ] . then , @xmath179 with the optimal rate given by @xmath180 , which is obtained when @xmath181 for some @xmath178 . proof of theorem [ thm : optimal_rate ] is given in section [ subsec : optimal_rate ] of sm . we can also derive an approximate expression for the asymptotic variance of @xmath182 . using a consistent root @xmath164 , we can use the equation @xmath183 . using the asymptotic representation of @xmath184 used in the proof of theorem [ thm : optimal_rate ] ( see section [ subsec : optimal_rate ] of sm ) , and ignoring higher order terms and the contribution of the model bias , and finally evaluating the expressions at @xmath164 instead of @xmath167 ( which is unknown ) , we have @xmath185^{-1}.\ ] ] here the estimated noise variance @xmath186 can be computed as the mean squared error @xmath187 . the expression ( [ eq : approx_asymptotic_variance ] ) allows us to obtain an approximate asymptotic variance for @xmath188 by @xmath189 , for any given @xmath190 , where @xmath191 . in theorem [ thm : consistency ] we prove the rate of convergence for a local minimizer , which is a global minimizer within a radius of @xmath192 of @xmath167 for a suitable range of values of @xmath35 . therefore , we need an initial estimate which resides within this domain . in the following , we describe one way of obtaining such an initial estimate , through a two - stage approach , which is similar in spirit to the approaches by chen and wu ( 2008a , 2008b ) . suppose that we first estimate @xmath1 and @xmath142 by local polynomial smoothing and denote these estimates by @xmath193 and @xmath194 . then , we fit the regression model @xmath195 by ordinary least squares , where @xmath196 . we refer to the resulting estimator @xmath197 as the two - stage estimator of @xmath34 : @xmath198}(t_j)]^{-1 } ( \sum_{j=1}^n \hat x'(t_j ) \bs\phi(\hat x(t_j ) ) \mathbf{1}_{[\delta,1-\delta]}(t_j)).\ ] ] since @xmath1 is @xmath8 times continuously differentiable and @xmath142 is @xmath6-times continuously differentiable ( by * a1 * ) , and @xmath199 is sub - gaussian , with the optimal choice of bandwidths , we have @xmath200}(t_j ) & = & o((\sigma_\varepsilon^2/n)^{(p+1)/(2p+3)}\sqrt{\log n } ) \label{eq : rate_local_polynomial}\\ \max_{1\leq j \leq n } |\hat x'(t_j ) - x'(t_j)| \mathbf{1}_{[\delta,1-\delta]}(t_j ) & = & o((\sigma_\varepsilon^2/n)^{p/(2p+3)}\sqrt{\log n } ) \label{eq : rate_local_polynomial_deriv}\end{aligned}\ ] ] with probability tending to 1 . we state the following result about the rate of convergence of the two - stage estimator . the proof is given in section [ subsec : proofs_two_stage ] of sm . [ prop : two_stage_regression ] suppose that @xmath201 and * a1**a4 * hold and that the two - stage estimate of @xmath4 is given by @xmath202 where @xmath197 is defined in ( [ eq : beta_tilde_two_stage ] ) . then , supposing that @xmath203 , with probability tending to 1 , @xmath204 where @xmath205 when @xmath206 , the optimal value of @xmath207 is of the order @xmath208 is obtained when @xmath209 . it can be checked that for all @xmath80 , this rate is slower than the optimal @xmath210 for the nonlinear regression - based estimator @xmath68 derived in theorem [ thm : consistency ] . however , the rate of convergence of this estimator is faster than @xmath211 if @xmath212 . so , for these range of @xmath35 , which includes @xmath213 , the two - stage estimator resides within the ball of radius @xmath214 around @xmath167 , over which @xmath68 , the optimizer of ( [ eq : loss_modified ] ) , is a global optimum . in this section , we outline the main steps of the proof . some technical details are deferred to the appendix . the main idea behind the proof of theorem [ thm : consistency ] is to obtain a lower bound on the difference @xmath215 which is proportional @xmath216 when @xmath34 lies in an annular region around @xmath167 . the outer radius of the annular region depends on the degree of ill - conditioning of the problem , as quantified by proposition [ prop : kappa ] , and the smoothness of the function @xmath4 and the approximating bases , as indicated in condition * a2*. this lower bound then naturally leads to the conclusion about the existence and rate of convergence of a local minimizer @xmath68 . for convenience of notations , we define @xmath217 to be the sample path @xmath218 . since @xmath219 is given by ( [ eq : tilde_x_beta_closed ] ) in the appendix , @xmath220 in order to prove proposition [ prop : kappa ] , it suffices to find a lower bound on @xmath221^{2}\tilde f_t(t ) dt\ ] ] where @xmath222 with @xmath223 . by * a3 * , without loss of generality , we can take the density @xmath224 to be uniform on @xmath136 $ ] . we make use of the following result known as halperin - pitt inequality ( mitrinovic _ et al . _ , 1991 ) . [ lemma : halperin_pitt ] if @xmath225 is locally absolutely continuous and @xmath226 is in @xmath227)$ ] , then for any @xmath228 the following inequality holds with @xmath229 , @xmath230 now defining @xmath231 we have , @xmath232 x_*'(t)\\ & = & \left[\frac{g_{\mathbf{b}}'(x_*(t))}{g_{\bs{\beta}^*}(x_*(t ) ) } - \frac{g_{\mathbf{b}}(x_*(t))g_{\bs{\beta}^*}'(x_*(t))}{g_{\bs{\beta}^*}^2(x_*(t))}\right ] g_{\bs{\beta}^*}(x_*(t)).\end{aligned}\ ] ] by ( vi ) of * a2 * , we have @xmath93 } |x_g(t ) - x_*(t)| = o(m^{-(p+1)})$ ] and hence @xmath233 hence , using the facts that the coordinates of @xmath234 are @xmath235 and the coordinates of @xmath236 are @xmath237 , and all these functions are supported on intervals of length @xmath85 , we deduce that , @xmath238 an application of lemma [ lemma : halperin_pitt ] with @xmath239 and @xmath240 yields @xmath241 take @xmath242 for some @xmath243 , then by ( [ eq : r_prime_int ] ) , @xmath244 for constants @xmath245 dependent on @xmath246 . next , we write @xmath247 where @xmath248 which is bounded below by a positive constant on the interval @xmath249 $ ] . observe that by ( [ eq : x_star_boundary ] ) , the combined support of @xmath63 , viz . , @xmath127 $ ] , contains ( for sufficiently large @xmath35 ) the interval @xmath250 $ ] . also , @xmath251 and @xmath252 } h(v)\right)~ \mathbf{b}^t [ \int_{x_{0,m}}^{x_{1,m } } \bs\phi(v ) ( \bs\phi(v))^t dv - o(1)]\mathbf{b } ~\geq~ k_3,\end{aligned}\ ] ] for some constant @xmath253 , for sufficiently large @xmath35 . thus , by appropriate choice of @xmath254 , we have @xmath255 for some constant @xmath256 , which yields ( [ eq : g_star_condition ] ) . from ( [ eq : diff_l_delta_beta ] ) and ( [ eq : diff_l_l_tilde_delta_beta ] ) , we deduce that @xmath271 where @xmath272}(t_j ) \\ & & \hskip-.2 in -\frac{2}{n } \sum_{j=1}^n ( x_g(t_j ) - x(t_j;\bs\beta^*))(x(t_j;\bs\beta^*;\widehat x_0 ) - x(t_j;\bs\beta^ * ) ) \mathbf{1}_{[\delta,1-\delta]}(t_j ) \nonumber\\ u_{4n}(\bs\beta,\bs\beta^ * ) & \hskip-.1in= & \hskip-.1in\frac{2}{n } \sum_{j=1}^n ( x(t_j;\bs\beta ) - x(t_j;\bs\beta^*))(x(t_j;\bs\beta;\widehat x_0)-x(t_j;\bs\beta ) ) \mathbf{1}_{[\delta,1-\delta]}(t_j).\end{aligned}\ ] ] using the fact that @xmath273 can be expressed as @xmath274,\ ] ] provided @xmath275 for @xmath276 $ ] , we have , for all @xmath260 , @xmath277 } \sup_{t \in [ \delta,1-\delta ] } |x(t;\bs\beta , a_0 ) - x(t;\bs\beta , x_{0,\delta})| \leq c_1\xi_n\ ] ] for some @xmath278 . here , we have used the fact that for @xmath129 $ ] , and @xmath279 $ ] , @xmath280 and that @xmath281}| g_{\bs\beta}(x ) - g_{\bs\beta^*}(x)| = o(\bar\alpha_n m^{1/2 } ) = o(m^{-3/2}),\ ] ] so that , by using ( [ eq : x_diff_bound ] ) , and the fact that @xmath282 , @xmath283 } x(1-\delta;\bs\beta , a_0 ) ] \subset [ x_{0,m},x_{1,m}]\ ] ] for large enough @xmath35 and @xmath52 . we now bound individual terms in the expansion ( [ eq : diff_l_tilde_delta_beta_expand ] ) . first , we have the following lower bound on @xmath284 , the proof of which is given in appendix c. [ lemma : gamma_beta_beta_star_bound ] let @xmath284 be as defined in ( [ eq : gamma_beta_beta_star_def ] ) . then given @xmath115 , there exist constants @xmath285 and @xmath286 independent of @xmath287 such that @xmath288 uniformly in @xmath260 with probability at least @xmath118 . since @xmath289 , and the constant @xmath290 can be chosen to be small enough so that we can conclude from ( [ eq : gamma_beta_beta_star_bound ] ) that given @xmath115 , there exists @xmath291 such that @xmath292 next , by cauchy - schwarz inequality , we have @xmath293 ) , we have @xmath294 and hence @xmath295 and @xmath296}(t_j ) } { \sigma_\varepsilon \sqrt{\sum_{j=1}^n ( x(t_j;\bs\beta ) - x(t_j;\bs\beta^*))^2\mathbf{1}_{[\delta,1-\delta]}(t_j)}}~,\ ] ] and setting @xmath297 being zero if the denominator is zero , we have @xmath298 be a @xmath299-net for @xmath300 . then @xmath301 . then , by using lemma [ lem : subgaussian_hoeffding ] in sm , and ( [ eq : gamma_beta_beta_star_bound_refined ] ) , we conclude that given @xmath115 , there exist constants @xmath302 , and a set @xmath303 with @xmath304 , such that for all @xmath305 , @xmath306 for some constant @xmath307 . thus , taking @xmath40 to be sufficiently small , say , @xmath308 for @xmath309 large enough , and using the smoothness of the process @xmath297 as a function of @xmath34 , we can show that given any @xmath115 , there exists @xmath310 , such that for all @xmath305 , @xmath311 very similarly , defining @xmath312}(t_j ) } { \sigma_\varepsilon \sqrt{\sum_{j=1}^n ( x(t_j;\bs\beta;\widehat x_0)-x(t_j;\bs\beta))^2\mathbf{1}_{[\delta,1-\delta]}(t_j)}},\ ] ] expressing @xmath313 , and using ( [ eq : v_1n_bound ] ) , we have , for any given @xmath115 , there exists @xmath314 and a set @xmath315 with @xmath316 , such that for all @xmath317 , @xmath318 finally , by * a2 * we have the bound @xmath319 } |x_g(t ) - x(t;\bs\beta^*)|^2 \leq c_2 m^{-2(p+1)}\ ] ] for some @xmath320 . combining ( [ eq : u_2n_bound])([eq : d_n_star_bound ] ) , we claim that , given @xmath115 , there exist constants @xmath321 , @xmath322 , and constants @xmath323 , @xmath324 , not depending on @xmath287 , such that uniformly on @xmath300 @xmath325 with probability at least @xmath326 . from ( [ eq : l_tilde_beta_diff_bound ] ) and ( [ eq : gamma_beta_beta_star_bound_refined ] ) , and a careful choice of the constant @xmath327 in the definition ( [ eq : alpha_n_def ] ) of @xmath210 , and with @xmath35 as in [ eq : m_alpha_optimal ] , we conclude that for any @xmath115 , there exists @xmath328 such that , uniformly in @xmath260 , @xmath329 with probability at least @xmath326 . from this , we can conclude that with probability at least @xmath326 there exists a local minimum @xmath164 of @xmath165 , which is also a global minimum within radius @xmath166 of @xmath167 and which satisfies @xmath330 . in this section , we conduct a simulation study to examine the finite sample performance of the proposed estimation procedure , as well as to compare it with the two - stage estimator described in section [ subsec : two_stage_regerssion ] . in the simulation , the true gradient function @xmath4 is represented by @xmath331 b - spline functions with knots at @xmath332 and respective coefficients @xmath333 ( shown by the blue curve in figure [ fig : simu ] ) . we set the initial value @xmath334 in equation ( [ eq : basic ] ) to generate the true trajectory @xmath335 . we then simulate @xmath336 independent data sets according to equation ( [ eq : data_model ] ) . specifically , for each data set , we first randomly choose an integer @xmath52 from @xmath337 . then @xmath52 observation times @xmath338 are uniformly sampled from @xmath101 $ ] . finally , the @xmath17 s are generated according to equation ( [ eq : data_model ] ) with added noise @xmath339 . the observed data from one such replicate is shown in figure [ fig : obs_trajectory ] in sm together with the true trajectory @xmath335 . we fit the proposed estimator @xmath340 with @xmath35 b - spline basis functions with equally spaced knots on @xmath341 $ ] . we consider @xmath342 and choose @xmath35 by an approximate leave - one - out cv score criterion similar to that used in paul et al . ( 2011 ) . out of the @xmath336 replicates , @xmath343 times the model with @xmath344 ( the true model ) is chosen and @xmath345 times the model with @xmath346 is chosen . we also consider the two - stage estimator , where in the first stage , the sample trajectory @xmath335 and its derivative @xmath347 are estimated by applying local linear and local quadratic smoothing with gaussian kernel , respectively , to the observed data @xmath348 . the bandwidths are chosen by cross - validation . in the second stage , a quadratic smoothing of @xmath349 versus @xmath350 is performed to get an estimate of @xmath141 . figure [ fig : simu ] shows the estimated gradient functions ( red curves ) of these @xmath336 independent replicates overlayed on the true gradient function ( blue curve ) . it can be seen from this figure that , the proposed estimator shows little bias . its sampling variability is somewhat larger on the left side of the observed @xmath190 domain than on the right side of the observed @xmath190 domain . it performs much better than the two - stage estimator which shows both high bias and high variance . indeed , the bias of the two - stage estimator would not go away even when in the second stage the true model is used to estimate @xmath4 ( through a least - squares regression of @xmath349 versus @xmath350 ) . figure [ fig : simu_traj ] shows the estimated trajectories ( red curves ) of these @xmath336 independent replicates overlayed on the true trajectory ( blue curve ) . in the left panel of the figure , the estimated trajectories are solved from equation ( [ eq : basic ] ) using the 4th - order runge - kutta method with @xmath4 being the proposed estimator @xmath340 . in the right panel of the figure , the trajectories are estimated by applying local linear smoothing of the observed data ( which are then used in the two - stage fitting for @xmath141 ) . the estimated trajectories from the proposed procedure follow the true trajectory very well with little bias , whereas the estimator from the first - stage smoothing of the two - stage procedure shows more bias and more variability . figure [ fig : simu_traj_deriv ] in sm shows the estimated derivative of the trajectory . again , the proposed procedure gives a much better estimate of @xmath351 than the presmoothing estimate ( by local quadratic smoothing ) used in the two - stage procedure . in this section , we apply the proposed model to the berkeley growth data ( tuddenham and snyder , 1954 ) . although in the literature , there are many studies of growth curves ( hauspie et al . , 1980 ; milani , 2000 ) , most of them try to model either the growth trajectories ( i.e. , @xmath335 ) or the rate of growth ( i.e. , @xmath347 ) . on the contrary , our goal is to estimate the gradient function , i.e. , the functional relationship between @xmath347 and @xmath335 which provides insights of the growth dynamics , such as at what height the growth rate tends to be the highest . specifically , we fit the proposed model to each of the @xmath352 female subjects in this data set . for each girl , her heights were measured at @xmath353 time points from @xmath354 year old to @xmath355 years old . we use @xmath35 b - spline basis functions with equally spaced knots . we consider @xmath356 and for each subject we choose the `` best '' @xmath35 using an approximate leave - one - out cv score . in @xmath357 out of @xmath352 subjects , the model with @xmath358 is chosen , and for the rest @xmath359 subjects , the model with @xmath360 is chosen . figure [ fig : fit_gradient ] shows the fitted gradient functions for these @xmath352 subjects . from this figure , we can see that , most girls experienced two growth spurs , one at the birth ( when their heights are shortest ) and another when they were around either @xmath361 cm tall or @xmath362 cm tall . moreover figure [ fig : fit_gradient_se ] in sm shows the fitted gradient functions with the two - standard - error bands ( by equation ( [ eq : approx_asymptotic_variance ] ) ) for @xmath363 girls . figure [ fig : fit_trajectory ] in sm shows the observed ( red dots ) and fitted ( black curve ) growth trajectories for these @xmath363 girls . it can be seen that , the fitted trajectories fit the observed data very well . in this paper we have proposed an estimation procedure for nonparametrically estimating the unknown gradient function of a first order autonomous differential equation over a finite domain , when the trajectories are strictly monotone . in this section , we discuss the asymptotic rate optimality of the proposed estimator . we show that , if the estimators of the gradient function @xmath4 are restricted to a class of uniformly lipschitz function , the optimal rate for estimation of @xmath4 , i.e. , of the order @xmath364 , is the same as the optimal rate for estimation of the derivative of @xmath365 based on model ( [ eq : data_model ] ) in terms of the @xmath0 loss . we conjecture that the lipschitz requirement on the estimator of @xmath4 is not necessary and the minimax rate for estimation of @xmath4 is indeed of the order @xmath364 . in order to make this statement precise , we first specify the function class for @xmath4 as @xmath366 where @xmath367 and @xmath368 are constants . define the class of uniformly lipschitz functions @xmath369 where @xmath370 depends on ( at least as large as ) @xmath138 in ( [ eq : g_class ] ) . if @xmath371 , then we have @xmath372)$ ] and @xmath373)$ ] . in addition , we assume the observation model ( [ eq : data_model ] ) with the noise @xmath374 . let @xmath40 be as in section [ sec : model ] . by the condition @xmath375 , we know that there exist @xmath376 such that @xmath377 for all @xmath129 $ ] , for all @xmath371 . define , @xmath378 . then there are constants @xmath379 such that for any given estimator @xmath380 of @xmath4 , @xmath381 observe that @xmath382 . on the other hand , since @xmath372)$ ] , there exists an estimator @xmath383 with the property that , given @xmath384 , there exists constant @xmath385 such that @xmath386 for all @xmath387 . we define the estimator @xmath388 for @xmath389 . then , by triangle inequality , @xmath390 where , in the last step we have used the fact that @xmath380 . since @xmath373)$ ] , the minimax rate of estimation of @xmath389 in terms of the @xmath0 loss @xmath391 is of the order @xmath364 . this can be derived directly for @xmath4 restricted to @xmath392 by only slightly modifying the arguments in stone ( 1982 ) . combining this fact with ( [ eq : g_hat_error_bound ] ) , ( [ eq : x_g_minimax ] ) and ( [ eq : x_deiv_estimate_bound ] ) , we obtain that there exists @xmath393 , such that @xmath394 in other words , as long as @xmath68 is uniformly lipschitz , the rate @xmath364 is a lower bound on the rate for estimating @xmath4 in terms of the @xmath0-loss . we note that , the requirement @xmath380 can be relaxed by only requiring that this holds with probability approaching one as @xmath124 . the latter is satisfied by the estimator we proposed . thus , combining with theorem [ thm : optimal_rate ] , we deduce that the optimal rate of estimation of @xmath4 is @xmath364 for @xmath76 . in this section , we provide technical details for the proofs of the main results . specifically , in appendix a , we present results on perturbation analysis of differential equations that are central to controlling the bias in the estimates . in appendix b , we verify that condition ( vi ) of * a2 * is satisfied by a b - spline basis of sufficiently high order . in appendix c , we prove lemma [ lemma : gamma_beta_beta_star_bound ] . further technical details are given in the supplementary material . since @xmath335 satisfies the ode @xmath395,\ ] ] differentiating with respect to @xmath34 we obtain the the linear differential equations : @xmath396 for @xmath397 , where @xmath398 . the hessian of @xmath335 with respect to @xmath130 is given by the matrix @xmath399 , where @xmath400 , which satisfies the system of odes , for @xmath401 : @xmath402 , ~~{x}^{\beta_r,\beta_{r'}}(0)=0 . \end{aligned}\ ] ] with @xmath403 and @xmath404 denoting @xmath405 , we also have @xmath406 note that ( [ eq : tilde_x_beta ] ) , ( [ eq : tilde_x_beta_hessian ] ) and ( [ eq : tilde_x_a ] ) are linear differential equations . if the function @xmath407 is positive on the domain then the gradients of the trajectories can be solved explicitly as follows . @xmath408 @xmath409ds \nonumber\\ & & ~+ g_{\bs{\beta}}(x(t ) ) \int_0^t \frac{1}{g_{\bs{\beta}}(x(s ) ) } x^{\beta_r}(s)x^{\beta_{r'}}(s ) g_{\bs{\beta}}''(x(s ) ) ds . \end{aligned}\ ] ] and @xmath410.\ ] ] now we summarize approximations of various relevant quantities . the following result on the perturbation of the solution path in an initial value problem due to a perturbation in the gradient function is derived from deuflhard and bornemann ( 2002 ) . [ prop : perturb ] consider the initial value problem : @xmath411 where @xmath412 . on the augmented phase space @xmath413 , say , let the mappings @xmath225 and @xmath414 be continuous and continuously differentiable with respect to the state variable . assume that for @xmath415 , the initial value problem ( [ eq : ode_general ] ) , and the perturbed problem @xmath416 have the solutions @xmath190 and @xmath417 , respectively . if @xmath225 is such that @xmath418 for a function @xmath419 bounded on @xmath420 $ ] , and @xmath421 for some nonnegative function @xmath422 on @xmath420 $ ] , then @xmath423.\ ] ] we use the above result to compute bounds for the trajectories and their derivatives corresponding to the different values of the parameter @xmath34 in a neighborhood of the point @xmath167 . in order to keep the exposition simple , we assume that @xmath424 for @xmath425 and @xmath426 for @xmath427 with a differentiability requirement at the points @xmath428 and @xmath429 . our aim is to show that the range of the trajectories @xmath430 is contained in the set @xmath125 $ ] , for all @xmath129 $ ] and for all @xmath431 . let @xmath432 . also , let @xmath433 . as in the proof of proposition [ prop : kappa ] , we can easily show that @xmath72 \subset [ x_{0,m},x_{1,m}]$ ] for sufficiently large @xmath35 . on the other hand , by using the perturbation bound given by proposition [ prop : perturb ] progressively over small subintervals of the interval @xmath136 $ ] , it can be shown that @xmath434}|x(t;\bs\beta , x_{0,\delta } ) - x_g(t;x_{0,\delta})| \leq c_1 \gamma_n + c_2 \xi_n,\ ] ] for appropriate positive constants @xmath435 that depend on the value of @xmath4 and @xmath436 on the interval @xmath25 $ ] . now , using lemma [ lem : consistency_x_j_hat ] , the condition on @xmath210 as given in theorem [ thm : consistency ] , and the definitions of @xmath60 , @xmath62 and @xmath437 , for @xmath61 , we conclude that for large enough @xmath35 , the range of @xmath430 is contained in @xmath126 for all @xmath438 $ ] and for all @xmath439 . the scenario is depicted in figure [ fig : trajectory_envelop ] , where the dashed curves indicate the envelop of the trajectories @xmath430 , while the solid curve indicates the trajectory @xmath263 . next , we provide bounds for trajectories and their derivatives . in the following , @xmath440 is used to denote the @xmath441-norm over @xmath442 $ ] . first , by * a2 * we have the following : @xmath443 where @xmath444 and @xmath445 denote the @xmath446-th derivative of @xmath4 and @xmath447 , respectively . next , again from * a2 * , for @xmath35 large enough , solutions @xmath448\}$ ] exist for all @xmath130 such that @xmath449 . this also implies that the solutions @xmath450 and @xmath451 exist on @xmath136 $ ] for all @xmath130 such that @xmath449 , since they follow linear differential equations where the coefficient functions depend on @xmath452 . moreover , by _ gronwall s lemma _ ( deuflhard and bornemann , 2002 ) , ( [ eq : g_beta_estimates ] ) and the fact that @xmath453 for @xmath88 ( again by * a2 * ) . hence , if @xmath454 , then using proposition [ prop : perturb ] , the fact that @xmath455 for @xmath88 , and the expressions for the odes for the partial derivatives , we obtain ( almost surely ) : @xmath456 the same technique can be used to prove the following : @xmath457 whenever @xmath458 . to illustrate the key arguments , we prove ( [ eq : x_beta_bound ] ) and ( [ eq : x_diff_beta_bound ] ) . first , ( [ eq : x_beta_bound ] ) follows by ( [ eq : tilde_x_beta_closed ] ) , and the fact that @xmath459 and is supported on an interval of length @xmath85 . in fact it holds for all @xmath34 such that @xmath458 . next , note that the function @xmath460 is lipschitz with lipschitz constant @xmath237 and is supported on an interval of length @xmath85 . since ( [ eq : tilde_x_beta ] ) is a linear differential equation , using proposition [ prop : perturb ] with @xmath461 given by @xmath462 + \phi_{r}(x(t;\bs{\beta } ) ) - \phi_{r}(x(t;\bs{\beta}^*))\end{aligned}\ ] ] we obtain ( [ eq : x_diff_beta_bound ] ) by using ( [ eq : x_diff_bound ] ) and the following facts : @xmath463 } |x^{\beta_r}(t;\bs\beta)| = o(m^{-1/2})$ ] for all @xmath464 ; @xmath465 ; @xmath466 ; and @xmath467 . in this subsection , we verify that the condition ( vi ) of * a2 * is satisfied if @xmath63 is a normalized b - spline basis with equally spaced knots on @xmath127 $ ] and of order @xmath468 . in particular , we show that the rate of approximation of @xmath1 by @xmath469 with a carefully chosen @xmath470 satisfies the requirement that @xmath93 } |x(t ) - x(t;\bs\beta^*)| = o(m^{-(p+1)})$ ] and the conditions @xmath471 } |g^{(j)}(x ) - g_{\bs\beta^*}^{(j)}(x)| = o(m^{-p+j})$ ] for @xmath88 . the result is proved through the following lemmas proved in sm . [ lemma : path_bound_refined ] suppose that @xmath63 has combined support @xmath472 = [ x(\delta),x(1-\delta)]$ ] and satisfies ( ii)(v ) of * a2 * and @xmath167 furthermore has the property that @xmath473 } \left|\int_{x_{0,\delta}}^{x } \frac{g(u ) - g_{\bs\beta^*}(u)}{g(u ) } du \right| = a_m\ ] ] such that @xmath474 , uniformly in @xmath35 , for some @xmath475 $ ] and some @xmath476 . then , if @xmath477 , there exists @xmath122 such that @xmath478 } |x(t)-x(t;\bs\beta^*)| \leq c a_m.\ ] ] [ lemma : integral_g_spline_approx ] suppose that * a1 * holds with @xmath479 . let @xmath63 denotes the normalized b - spline basis of order @xmath480 with equally spaced knots on the interval @xmath127 $ ] . then there exists a @xmath481 such that @xmath482 satisfies @xmath483 } \left|\int_{x_{0,\delta}}^{x } \frac{g(u ) - g_{\bs\beta^*}(u)}{g(u ) } du \right| = o(m^{-(p+1)}).\ ] ] by a taylor expansion we have , for @xmath484 , @xmath485 where @xmath486 for all @xmath446 . from this , it follows that , for all @xmath260 , @xmath487}(t_j)\right ] ( \bs\beta - \bs\beta^ * ) \nonumber\\ & & - 3 \parallel \bs\beta - \bs\beta^*\parallel^2 \frac{1}{n } \sum_{j=1}^n \parallel x^{\bs\beta}(t_j;\tilde{\bs\beta}(t_j))- x^{\bs\beta}(t_j;\bs\beta^*)\parallel^2 \mathbf{1}_{[\delta,1-\delta]}(t_j),\end{aligned}\ ] ] where we have used @xmath488 . using proposition [ prop : kappa ] and lemma [ lem : g_gamma_quad_bound ] ( stated below ) we conclude , given @xmath115 , there exists @xmath489 such that , @xmath490}(t_j ) \right ] ( \bs\beta - \bs\beta^ * ) \\ & \geq & c_{10}(\eta ) \frac{1}{m^2 } \parallel \bs\beta - \bs\beta^*\parallel^2\end{aligned}\ ] ] for all @xmath260 , with probability at least @xmath118 . now , another application of the mean value theorem yields that for @xmath491 $ ] , @xmath492 where @xmath493 denotes the frobenius norm , and @xmath494 for all @xmath495 and @xmath496 . now , using ( [ eq : x_beta_beta_bound ] ) and ( [ eq : x_diff_beta_beta_bound ] ) , and combining the last three displays , we get ( [ eq : gamma_beta_beta_star_bound ] ) . [ lem : g_gamma_quad_bound ] suppose that * a1**a4 * hold . let @xmath497}(t_j).\ ] ] then , given @xmath115 , there exists constants @xmath498 such that , with probability @xmath118 , uniformly in @xmath499 , @xmath500 let @xmath501 . define @xmath502 . notice that @xmath503}(t_j ) ] = \mathbb{e}_{\tilde t}[\mathbf{v}_j\mathbf{v}_j^t ] = g_*,\ ] ] where the first expectation is with respect to the distribution of @xmath504 and the second with respect to that of @xmath149 . hence , we can write @xmath505 where @xmath506}(t_j)}{f_t(1-\delta)-f_t(\delta ) } - \mathbb{e}_{\tilde t}[\mathbf{v}_j\mathbf{v}_j^t]\right)\bs\gamma.\ ] ] note that , the random variables @xmath507 have zero conditional mean , are uniformly bounded , and are independent . moreover , the functions @xmath507 are differentiable functions of @xmath508 . then , since by ( [ eq : x_beta_bound ] ) , @xmath507 s are uniformly bounded by some @xmath509 , @xmath510 & \leq & k_1 \sum_{j=1}^{n}\mathbb{e}|u_j(\bs\gamma)| ~\leq~ 2k_1 n \bs\gamma^t g _ * \bs\gamma .\end{aligned}\ ] ] thus , by bernstein s inequality , for every @xmath511 and @xmath512 , @xmath513 on the other hand , by ( [ eq : g_star_condition ] ) , @xmath514 for some @xmath178 . by this , and the condition that @xmath515 , it is easy to see that @xmath516 . thus , using an entropy argument as in the proof of ( [ eq : z_beta_sup_bound ] ) , we conclude that given @xmath115 there exists @xmath517 such that @xmath518 recalling the definition of @xmath519 , and again using the fact that @xmath514 and @xmath515 , ( [ eq : g_gamma_quad_bound ] ) follows from ( [ eq : sum_u_bound ] ) . cavalier , l. , golubev , g. k. , lepskii , o. and tsybakov , a. b. ( 2004 ) . block thresholding and sharp adaptive estimation in severely ill - posed inverse problems . _ theory probability application _ , * 48 * , 426-446 . chen , j. and wu , h. ( 2008b ) . efficient local estimation for time - varying coefficients in deterministic dynamic models with applications to hiv-1 dynamics . _ journal of american statistical association _ * 103 * , 369384 . hauspie , r. c. , wachholder , a. , baron , g. , cantraine , f. , susanne , c. and graffar , m. ( 1980 ) . a comparative study of the fit of four different functions to longitudinal data of growth in height of belgian girls . _ annals of human biology _ , * 7*(4 ) , 347358 . liu , b. and mller , h .- ( 2009 ) . estimating derivatives for samples of sparsely observed functions , with application to online auction dynamics . _ journal of the american statistical association _ , * 104 * , 704716 . miao , h. , dykes , c. , demeter , l. m. and wu , h. ( 2009 ) . differential equation modeling of hiv viral fitness experiments : model identification , model selection , and multimodel inference . _ biometrics _ , * 65 * , 292300 . nicol , f. ( 2013 ) . functional principal component analysis of aircraft trajectories . _ isiatm 2013 , 2nd international conference on interdisciplinary science for innovative air traffic management , toulouse , france . _ poyton , a. a. , varziri , m. s. , mcauley , k. b. , mclellan , p. j. and ramsay , j. o. ( 2006 ) . parameter estimation in continuous dynamic models using principal differential analysis . _ computers & chemical engineering _ * 30 * , 698708 . qi , x. and zhao , h. ( 2010 ) . asymptotic efficiency and finite - sample properties of the generalized profiling estimation of parameters in ordinary differential equations . _ annals of statistics _ , * 38 * , 435481 . ramsay , j. o. , hooker , g. , campbell , d. and cao , j. ( 2007 ) . parameter estimation for differential equations : a generalized smoothing approach . _ journal of the royal statistical society , series b _ * 69 * , 741796 . wu , h. , lu , t. , xue , h. and liang , h. ( 2014 ) . sparse additive ordinary differential equations for dynamic gener regulatory network modeling . _ journal of the american statistical association _ , * 109 * , 700716 . xue , h. , miao , h. and wu , h. ( 2010 ) . sieve estimation of constant and time - varying coefficients in nonlinear ordinary differential equation models by considering both numerical error and measurement error . _ annals of statistics _ , * 38 * , 23512387 .

_ we discuss the principles to be used in the construction of discrete time classical and quantum mechanics as applied to point particle systems . in the classical theory this includes the concept of virtual path and the construction of system functions from classical lagrangians , cadzow s variational principle applied to the action sum , maeda - noether and logan invariants of the motion , elliptic and hyperbolic harmonic oscillator behaviour , gauge invariant electrodynamics and charge conservation , and the grassmannian oscillator . first quantised discrete time mechanics is discussed via the concept of system amplitude , which permits the construction of all quantities of interest such as commutators and scattering amplitudes . we discuss stroboscopic quantum mechanics , or the construction of discrete time quantum theory from continuous time quantum theory and show how this works in detail for the free newtonian particle . we conclude with an application of the schwinger action principle to the important case of the quantised discrete time inhomogeneous oscillator . _

hep-th9703079

there are various circumstances in mechanics where it is convenient or necessary to replace the continuous time ( temporal evolution ) parameter with a discrete parameter . computer simulation of waves is an example where the configuration of a system at time @xmath0 is calculated from a knowledge of its configuration at times @xmath1 and @xmath2 . there have been various attempts to construct classical and quantum mechanical theories based on this notion , such as the work of cadzow @xcite , logan @xcite , maeda @xcite and lee @xcite . the work of yamamoto et al . @xcite and klimek @xcite indicates that the subject continues to receive attention . this paper considers the question : by which principles if any should continuous time mechanical theories be discretised , that is , turned into discrete time analogues ? by discretisation we do not mean the numerical approximation of continuous time mechanics such as the work of bender et al . neither do we discretise space or the dynamical degrees of freedom . our attention is fixed solely on replacing a continuous dynamical evolution parameter with a discrete parameter . in this and the following paper , _ paper ii _ on discrete time classical field theory , our interest is the construction of exact , self consistent discrete time mechanics with well specified principles , equations of motion and predictions . this is motivated by the notion that at some unimaginably small scale , time is really discrete . this has echoes in modern theories such as string theory and quantum gravity , where the planck time of @xmath3seconds sets a scale at which conventional notions of space and time break down . it could be argued that relativity requires a symmetrical treatment between time and space but this leads to the situation of a space - time lattice approach which has lost all relativistic symmetries and rotational invariance . we argue that relativity does distinguish between timelike and spacelike , and by discretising only time our approach reduces the break with relativity to a minimum . lorentz covariance is broken in our approach to field theories , discussed in _ paper ii _ , but the residual euclidean invariances permit the construct of particle like states . it may be felt objectionable that there is no natural concept of velocity in discrete time mechanics . it could be argued that this lack destroys our intuitive feeling for dynamics based on the notion of ( say ) a particle system evolving from an initial position and an initial velocity . the right way to see the situation is in terms of real numbers . in continuous time mechanics , we normally consider a particle as having an instantaneous position and an instantaneous velocity ( we exclude brownian particle dynamics from our definition of continuous time mechanics ) . this information requires two real numbers for every degree of freedom . in discrete time mechanics , there is no natural concept of simultaneity analogous to this . what we mean by a particle is something with a position at time @xmath1 _ and _ a position at time @xmath2 , which also requires two real numbers for every degree of freedom . a particle here is more properly associated with the _ link _ between two successive points in discrete time , rather than those times separately . so ultimately , the only major difference in principle between continuous time and discrete time mechanics is the lack of the limit process @xmath4 . one problem with discrete time mechanics is a lack of guiding principles at key places , which our series of papers attempts to address . for example , consider the discretisation of a system with lagrangian @xmath5 . the approach taken by most authors would be to replace the temporal derivatives by differences , symmetrise the potential in some way , derive the analogue of the euler - lagrange equation , and finally evolve the system according to the resulting difference equation . quantities such as the energy @xmath6 which are conserved in continuous time mechanics would be monitored by calculating the value @xmath7 of the discretised hamiltonian . it is more than likely however that a naive discretisation of the hamiltonian would result in an expression @xmath8 which is not exactly conserved . this has been discovered by many authors . it is a particular merit of lee s approach @xcite that an invariant analogous to the energy drops out of the formalism , but only at the expense of a dynamically evolving discrete time interval . it is somewhat surprising therefore that a computer simulation based on the above principles should be judged good or bad according to how constant @xmath9 remains . in the absence of any proper principle for the construction of invariants of the motion it should come as no surprise to find that occasionally a quantity such as @xmath8 will vary enormously and unpredictably during the course of a simulation this happens because there are really three systems being confused ; 1 ) the original continuous time theory , 2 ) a discrete time system evolving exactly according to some well defined discretised euler - lagrange equation , and 3 ) some unknown discrete time system for which the naively discretised energy @xmath9 would be an exact constant of the motion , but only for evolution under its own discretised equation of motion , which could be very different to the equation of motion for 2 ) . on top of this there may be numerical uncertainties induced by the computer algorithms used . seen in this light , it would seem a wise policy to discretise according to definite principles which would establish conserved quantities rigorously . the construction of invariants of the motion therefore becomes one of the principal objectives of discrete time mechanics . an important first step in the process of constructing rigorous discrete time mechanics was the introduction of a discrete time action principle . this was done by cadzow @xcite , giving a discrete time analogue of the euler - lagrange equation . we shall call such an equation a _ cadzow s _ _ equation _ for the system . the construction of constants of the motion was considered by maeda @xcite in the case of continuous symmetries , whilst the construction of constants of the motion analogous to the energy had been considered earlier by logan @xcite . various features found in continuous time mechanics have discrete time analogues , including noether s theorem , whilst certain other features either do not or can not have discrete time analogues . an particular problem arises for example with hamiltonian evolution and equations of motion derived using poisson brackets . not only is there no possibility here of an infinitesimal translation in time ( which thereby renders the notion of a hamiltonian problematical ) but there is no natural concept of velocity as a limit either . this makes the standard definition of conjugate momentum as the partial derivative of the lagrangian with respect to a velocity just as problematical . this has not prevented a number of authors from constructing discrete time analogues of poisson brackets , however , with various degrees of success and utility , usually with the observation that the generator of time translations is not conserved . a feature of our approach is that we have found a clear principle for the definition of conjugate momentum in discrete time mechanics . it turns out _ not _ to be the partial derivative of the lagrangian with respect to a difference in general , but does reduce to it in various important cases . in addition , we have avoided trying to construct hamiltonians and equations of motion derived via poisson brackets . in our formalism the hamiltonian is displaced by a logan invariant , if such a quantity can be found . fortunately such an object does exist for the important case of the harmonic oscillator , which has ramifications in discrete time field theory discussed in the next paper in this series . the overall plan for this and subsequent papers is as follows . in this paper , _ paper i _ , we restrict our attention to classical and quantum point particle dynamics , reserving classical field theory to _ paper ii _ , quantum field theory to _ paper iii _ , and quantum electrodynamics to _ paper iv_. topics covered in _ paper i _ are as follows . first we introduce the central concept of _ system function_. this replaces the lagrangian as the key to the dynamics . with the system function we can calculate equations of motion , constructs invariants of the motion , and quantise the system . we give a prescription for constructing the system function from a given lagrangian . we may use this prescription to embed into the system function symmetries such as gauge invariance and hence construct electrodynamics . then we discuss the construction of invariants based on the work of maeda , noether , and logan , and apply it to the harmonic oscillator , which we discuss in detail . this key system lies at the heart of particle field theory , discussed in the following papers , and it displays some important properties , such as a natural cutoff for particle energy for example . we also discuss particle electrodynamics and the grassmannian oscillator . a major part of our programme is to develop discrete time quantum mechanics and so we conclude our paper with a discussion of the principles for first quantisation . this includes the concept of system amplitude , the construction of unequal - time commutators , and compatible operators . we then discuss the construction of discrete time quantum mechanics from standard quantum mechanics via a stroboscopic approach and give an explicit example . finally we apply the schwinger action principle to the discrete time inhomogeneous harmonic oscillator to construct the feynman propagator for the oscillator , in anticipation of its use in field theory . in continuous time mechanics lagrangian dynamics is conventionally formulated via an action principle based on the action integral @xmath10=\int_{t_i}^{t_f}dt\,l(\mathbf{q},\mathbf{\dot{q}},t ) , \label{action integral}\ ] ] where @xmath11 and @xmath12 are the initial and final times respectively along some given path @xmath13 . in our version of discrete time mechanics we postulate that the dynamical variables @xmath14 are observed or sampled at a finite number of times @xmath15 , where @xmath16 and @xmath17 , such that the intervals @xmath18 are all equal to some fundamental interval @xmath19 . for convenience we will write @xmath20 . it is possible to develop a theory where the time intervals vary dynamically along the path . such a mechanics was considered by lee @xcite . the extension of our methods to that particular situation is left for a further article . in our formulation of discrete time mechanics we replace the action integral ( [ action integral ] ) by an action sum of the form @xmath21=\sum_{n=0}^{n-1}f^n , \label{sum}\ ] ] where @xmath22 will be referred to as the _ system function_. the system function has the same central role in discrete time mechanics as the lagrangian has in continuous time mechanics . with it we may construct the equations of motion , define conjugate momenta , construct constants of the motion and attempt to quantise the system . in principle we could consider higher order system functions which depended on ( say ) @xmath23 , @xmath24 ... , @xmath25 @xmath26 @xmath27 but the case @xmath28 represents the simplest possibility which could give rise to non - trivial dynamics and will be considered exclusively from now on . such system functions are the discrete time analogues of lagrangians of the canonical form @xmath29 another reason for considering only a second order formulation @xmath30 is its direct relationship to hamilton s principal function , discussed presently . cadzow @xcite applied a variational principle to an action sum such as ( [ sum ] ) and derived the equation of motion @xmath31 where the symbol @xmath32 denotes an equality holding over a true or dynamical trajectory . we shall refer to @xmath33 as a _ cadzow s equation of motion _ for the system . we now discuss the interpretation of this equation . suppose we have a continuous time action integral of the form ( [ action integral ] ) . first , partition the time interval @xmath34 $ ] into @xmath35 equal subintervals . then the action integral may be written as a sum of sub - integrals , i.e. , @xmath10=\sum_{n=0}^{n-1}\int_{t_n}^{t_{n+1}}dt\,l(\mathbf{q}(t),% \mathbf{\dot{q}}(t),t).\ ] ] now suppose that we fixed the co - ordinates @xmath23 at the various times @xmath36 and then chose the path connecting each pair of points ( @xmath37 to be the true or dynamical path , that is , a solution to the euler - lagrange equations of motion for those boundary conditions . if this partially extremised path is denoted by @xmath38 then we may write @xmath39=\sum_{n=0}^{n-1}s^n , \label{action sum}\ ] ] where @xmath40 is known as hamilton s principal function , being just the integral of the lagrangian along the true path from @xmath23 at time @xmath41 to @xmath42 at time @xmath43 . we recall now that the canonical momenta @xmath44 , @xmath45 at the end points @xmath46 may be obtained from hamilton s principal function via the rule @xmath47 where the superscript @xmath48 denote that the momentum at the initial time @xmath49 carries information forwards _ _ , _ _ whereas the superscript @xmath50 denotes that the momentum at the final time @xmath12 is influenced by earlier dynamics with respect to the temporal interval concerned . at this stage the action sum ( [ action sum ] ) has not been extremised fully , as the intermediate points @xmath51 have been held fixed . now suppose we went further and extremised ( [ action sum ] ) fully by variation of the previously fixed intermediate co - ordinates @xmath52 . then we would find that @xmath53 this equation may be understood as the condition that the canonical momentum along the true path from @xmath54 to @xmath55 is continuous , that is , @xmath56 . we notice immediately that ( [ formal ] ) has the same formal structure as cadzow s equation ( [ cadzow ] ) provided we make the identification @xmath57 . another interpretation of cadzow s equation is that it endows the action sum with the additivity property of action integrals , which satisfy the relations @xmath58 this property holds for all trajectories in continuous time mechanics , not just for the true or classical trajectory . in the case of system functions we may write @xmath59 for some function @xmath60 of @xmath61 and @xmath42 , because cadzow s equation ( [ cadzow ] ) is equivalent to the statement that @xmath62 is independent of @xmath23 along dynamical trajectories . however , unlike action integrals , this property will not hold off the true or classical trajectory in general . @xmath63two important ideas emerge from the similarity between ( [ cadzow ] ) and ( [ formal ] ) : @xmath64 although the concept of velocity as a limit does not occur in discrete time mechanics , we will define a unique discrete time momentum @xmath65 conjugate to @xmath23 by the rule @xmath66 this should be compared with the approach of yamamoto et al . @xcite and most other workers , where the momentum is defined as a derivative of a discretised lagrangian with respect to a difference . in our terms cadzow s equation reduces simply to the statement that we may also calculate this momentum via the rule @xmath67 @xmath68 we will construct a system function @xmath69 from the temporal integral from @xmath41 to @xmath43 of a continuous time lagrangian , the question being which path to take . we can not in general consider using the true continuous time path , as this is meaningless in the context of discrete time mechanics and normally not known to us . for the particularly important case of the harmonic oscillator , however , we can evaluate hamilton s principal function precisely and this provides us with an important check on our formalism . the path chosen in the construction of the system function will be referred to as a _ virtual path_. it is possible to choose from a number of possible virtual paths , such as those inspired by @xmath70deformed mechanics @xcite . this does not alter any of the principles we employ and simply changes the details of the system function used and hence the sort of invariants of the motion we can find . in this paper we are interested in treating time homogeneously , and so we choose a temporal lattice with a constant fundamental time interval @xmath19 . our proposed solution for the virtual path in point particle mechanics is to take the geodesic or shortest geometric path from @xmath23 to @xmath71 , the metric being normally the euclidean one in physical space ( not in co - ordinate space ) . this prescription will normally provides us with a unique system function from a given lagrangian . moreover , it should be applicable to configuration spaces with curvature and is a co - ordinate frame independent concept . it allows us to construct a gauge - invariant discrete time prescription for electrodynamics , with a suitable modification . in _ paper ii _ of this series we shall show that we can apply this prescription successfully to field theories also . there may be important cases where the chosen virtual path is not a linear interpolation . this occurs for charged fields in the next paper in the series . in such cases , additional requirements such as gauge invariance will influence the choice of virtual path . to illustrate the procedure , consider a non - relativistic particle with position vector @xmath72 and lagrangian @xmath73 . then the virtual path @xmath74 taken between @xmath75 and @xmath76 is given by @xmath77 where @xmath78 and @xmath79 . with this choice of virtual path the time derivative become a difference operator . specifically , we find @xmath80 where we define @xmath81 then we construct the system function via the rule @xmath82 the use of this integration does not imply that continuous time is regarded as meaningful in the context of discrete time mechanics . we are interested only in the results , not in the means of obtaining these results . a useful analogy is with the use of classical mechanics to set up quantum mechanical models . once we have found our quantum theory , we need no longer to regard the classical model which generated it as any more than some approximation useful in some circumstances . our prescription allows us to embed into our system function fundamental properties such as gauge invariance and other symmetries of importance to physics . if the lagrangian is a real analytic function of its arguments then we may make a taylor expansion about @xmath83 and integrate term by term . this will be valid for lagrangians which are polynomial functions of @xmath84 and @xmath85 . in such cases the system function @xmath69 would be given by the formal expression @xmath86 where @xmath87 is the operator @xmath88 and @xmath89 and @xmath83 are considered independent at this stage . some examples will illustrate the process . for a particle in one dimension with lagrangian @xmath90 where the @xmath91 are constants , the system function is given formally by @xmath92 for instance , the anharmonic oscillator lagrangian @xmath93 gives the system function @xmath94 this differs from the anharmonic oscillator system function discussed in @xcite , which illustrates the general problem with discrete time mechanics . there may be many possible discretisation of a given continuous time system , all of which lead back to the continuous time theory when we take appropriate limits . the principle specified above gives us a unique discretisation ( subject to choice of virtual path ) . for the coulombic potential problem in three spatial dimensions , the lagrangian @xmath95 with virtual path @xmath96 gives the system function @xmath97 this system function leads to cadzow s equations of motion which preserve the discrete time analogue of orbital angular momentum . this system function is markedly different in form to the original coulombic lagrangian @xmath98 but if we consider trajectories for which we may write @xmath99 @xmath100 @xmath101 for each @xmath102 , then @xmath103 which corresponds with @xmath104 . however , it should be kept in mind that there will be many discrete time trajectories for which this limit can not be taken . for example , there may be trajectories where the particle repeatedly flips between two fixed positions only . this may happen with the discrete time harmonic oscillator , for example , and no limit such as the one discussed above exists for such a trajectory . discrete time mechanics is inherently richer in its set of possible trajectories than continuous time mechanics . in general , cadzow s equations lead to an implicit equation for @xmath105 involving @xmath106 and @xmath107 , although for certain systems such as the harmonic oscillator discussed below we may solve cadzow s equation to find @xmath108 explicitly . the situation is analogous to what happens in computer simulations of partial differential equations where not all equations give @xmath108 explicitly . in such cases we must use numerical techniques to solve for the @xmath105 in the classical theory . it is a special feature of our approach that our equations of motion involve only @xmath107 , @xmath106 , and @xmath108 , which is not always the case with finite difference schemes used to approximate differential equations . it is possible to find a discrete time analogue of noether s theorem in the case of continuous symmetries along the lines considered by maeda @xcite . we shall refer to constants of the motion found by this theorem as _ maeda - noether invariants . _ consider a system function @xmath109 which is invariant to some point transformation @xmath110 then we may write @xmath111 using cadzow s equation of motion . from this we deduce that the quantity @xmath112 will be conserved along dynamical trajectories , that is , @xmath113 this construction does not allow us to construct an analogue of the hamiltonian in the case of conserved systems because in our formulation we are not allowed to make infinitesimal jumps in time . logan @xcite gave a method for constructing constants of the motion which are not necessarily related to symmetries of the system function . consider a point transformation @xmath114 where @xmath115 is infinitesimal and @xmath116 is a function of @xmath117 and @xmath42 . then @xmath118 on the true trajectories . now suppose that the transformation @xmath119 is such that @xmath120 can be written in the form@xmath121 where @xmath122 then we immediately deduce that the quantity @xmath123 is conserved over the classical trajectories . such a constant of the motion will be referred to as a _ logan invariant . the discrete time harmonic oscillator in its generic form is given by the quadratic system function @xmath124 which gives the cadzow s equation of motion @xmath125 a logan invariant of the motion is found to be @xmath126 in this subsection we show how to solve the equation of motion @xmath127 and determine the behaviour of the oscillator as the discrete time tends to infinity . first we define the variables @xmath128 which will become the analogues of annihilation and creation operators in quantum theory . the constants @xmath129 are chosen to satisfy the condition @xmath130 under the equation of motion @xmath131 which implies @xmath132 condition @xmath133 gives @xmath134 we note that @xmath135 the logan invariant @xmath136 is given by @xmath137 which is a constant of the motion by inspection and is a form of great value in discrete time field theory . the complete solution to the problem is now readily obtained and given by @xmath138 x_1+\left [ \left ( \mu ^{-}\right ) ^{n-1}-\left ( \mu ^{+}\right ) ^{n-1}\right ] x_0}{\left ( \mu ^{+}-\mu ^{-}\right ) } , \;\;\;\eta ^2\neq 1 . \label{sol}\ ] ] for the case when @xmath139 we write @xmath140 and then we find @xmath141 whereas for @xmath142 we write @xmath143 ( assuming @xmath144 and then @xmath145 and similarly for @xmath146 the crucial result is that bounded , elliptic behaviour occurs when @xmath139 whereas unbounded ( hyperbolic ) behaviour occurs when @xmath147 this result gives a natural cutoff to particle momentum in field theory , as shown in _ paper ii_. the readily solved case when @xmath148 corresponds to the free particle and will be referred to as the _ parabolic _ case . when @xmath142 the system will be said to be _ hyperbolic _ and _ elliptic _ when @xmath149 . for the case @xmath139 it is useful to define @xmath150 and @xmath151 , \;\;\;a_n^{*}\equiv \mu ^{-n}\left [ x_{n+1}-\mu ^{-1}x_n\right ] , \label{ladder}\ ] ] the advantage being that these are constants of the motion , viz @xmath152 these are useful in the construction of particle states in quantum field theory because they correspond to annihilation and creation operators in the schrdinger picture . using the methods outlined in @xmath153 the continuous time lagrangian for the newtonian harmonic oscillator @xmath154 gives the system function @xmath155 which is equivalent to @xmath156 the equation of motion is given by @xmath157 which is equivalent to @xmath158 with the identification @xmath159 , which means @xmath160 using the results of the previous subsection , we deduce that elliptic behaviour occurs only when the time @xmath19 satisfies the condition @xmath161 an equivalent result is found in particle field theory , giving a natural cutoff for particle momentum . we may understand the relationship between hamilton s principal function for the interval @xmath162 $ ] and the system function by explicitly evaluating the former for the continuous time harmonic oscillator lagrangian @xmath163 . we find @xmath164 , \label{ham}\ ] ] and comparing this with @xmath165 we find @xmath166 we expect a similar relation to exist in the general case , but different potentials will modify the precise details . there is an apparent problem with @xmath167 whenever the time interval @xmath19 satisfies the condition @xmath168 , @xmath169 because the denominator @xmath170 vanishes at such times . this problem is an artefact of our representation of @xmath171 , because the definition of the principal function as a line integral over a finite contour of a bounded integrand means that @xmath171 can not diverge . the resolution of this apparent paradox is that at the _ recurrence times _ @xmath172 the endpoints @xmath106 and @xmath105 are no longer independent but are related by @xmath173 the physical interpretation of recurrence is simple . the harmonic oscillator has a fundamental period @xmath174 , independent of the initial conditions . an important construction for the harmonic oscillator are the variables @xmath175 , @xmath176 defined by @xmath177 , \nonumber \\ a_n^ { * } & \equiv & \frac{-ie^{-in\theta } } { \sin \left ( \theta \right ) } \left [ x_{n+1}-e^{-i\theta } x_n\right ] , \;\;\;\theta \equiv \omega t. \label{laddera}\end{aligned}\ ] ] these are constants of the motion , i.e. @xmath178 and are independent of @xmath19 . this means that recurrence must occur so as to cancel the zero of the denominator in @xmath167 at the recurrence times . to see what happens explicitly , we may invert the equations @xmath179 to find @xmath180 , \;\;\;x_{n+1}=\frac{{}_1}{{}^2}\left [ e^{i(n+1)\theta } a_n^{*}+e^{-i(n+1)\theta } a_n\right ] , \ ] ] so that at the recurrence times @xmath172 we have @xmath181 , \ ] ] from which we deduce @xmath182 in terms of @xmath175 and @xmath176 the principal function can be written as @xmath183 , \ ] ] a form which shows clearly that the principal function is not singular . moreover , we see that at the recurrence times @xmath184 it is a significant feature of the discrete time harmonic oscillator that it does not involve recurrence phenomena in this particular way , as no apparent singularities occur in the system function @xmath165 . this emphasises that discrete time mechanics is not equivalent to continuous time mechanics . given the discrete time harmonic oscillator system function @xmath378 we note that is an example of a normal system . this leads us to define the system amplitude to be @xmath379 \right ) , \label{asys}\end{aligned}\ ] ] where @xmath320 is some constant . from the unitarity condition @xmath380 we find ( [ zz1 ] ) and from @xmath323 we find @xmath381 , so we see that the above system co - ordinates are indeed normal . the self adjoint diagonal operator with operator component @xmath382 \label{op2}\ ] ] is compatible with the system amplitude @xmath383 the interpretation of this is that @xmath384 is the operator corresponding to the logan constant for the classical discrete time harmonic oscillator @xmath385 to see this explicitly , consider the operators @xmath386and @xmath387 the logan constant for the harmonic oscillator is quantised according to the standard rule @xmath388 a suitable co - ordinate representation of these operators with respect to the basis @xmath277 is @xmath389 and then the operator @xmath390 is represented by @xmath391 we see from the potential term in the differential operator @xmath392 that a complete set of physical states can be found as eigenstates of the operator provided @xmath393 this corresponds precisely to the elliptic region discussed in the classical theory . if the constants satisfy the elliptic condition then we may construct annihilation and creation operators for the system . these are diagonal with respect to any of the bases @xmath277 and are given by @xmath394 = e^{in\theta } \int dx\,|x , n\rangle \left\ { \sqrt{1-\eta ^2}x+\frac \hbar \beta \partial _ x\right\ } \langle x , n| , \nonumber \\ \hat{a}_n^{+ } & \equiv & -ie^{-i\theta } [ \hat{x}_{n+1}-e^{-i\theta } \hat{x}% _ n]=e^{-i\theta } \int dx\,|x , n\rangle \left\ { \sqrt{1-\eta ^2}x-\frac \hbar \beta \partial _ x\right\ } \langle x , n| . \nonumber \\ & & \end{aligned}\ ] ] these operators satisfy the commutation relation @xmath395 = \frac{2\hbar \sqrt{1-\eta ^2}}\beta .\ ] ] using the evolution relation ( [ pos ] ) and the operator equation of motion @xmath396 we find @xmath397 but this does not mean this operator is conserved . a conserved operator according to our definition must be compatible with the timestep operator @xmath398 we find that the creation and annihilation operators satisfy the relations @xmath399 which is reminiscent of various deformed commutators encountered in q - mechanics . however , the operator ( @xmath400 corresponding to the logan constant @xmath401 is compatible with the timestep operator and is therefore an invariant of the motion . we find @xmath402 this logan invariant is a close analogue of the oscillator hamiltonian in continuous time mechanics and the eigenstates of the former follow the same pattern as the eigenstates of the latter . for example , there is a ground state @xmath403 satisfying the relation @xmath404 with normalisable wave function @xmath405 this wave function is also an eigenstate of the logan invariant operator , with @xmath406 these results hold only for @xmath407 . we note from @xmath408 that @xmath409 and that @xmath410 when we identify our system with the newtonian oscillator . @xmath185we now consider the case of electrically charged particles interacting with electromagnetic fields . a more complete discussion of discrete time maxwell s equations is given in paper @xmath186 of this series . here we discuss only the case of test particles which are affected by external electric and magnetic fields but do not affect them . we shall find the cadzow equation of motion for such particles and show that in our prescription electric charge is conserved . consider a non - relativistic charged test particle of mass @xmath187 in external electromagnetic potentials . the continuous time lagrangian for such a system is given by @xmath188 where @xmath189 is the charge of the particle . this lagrangian is not gauge invariant but the equations of motion are gauge invariant , because under the gauge transformation @xmath190 the action integral transforms according to the rule @xmath191 _ { t_i}^{t_f } , \label{linesum}\ ] ] that is , the change in the action integral occurs only at the end points . if this property is preserved in any discretised version of electrodynamics then the equations of motion should be gauge invariant . our prescription for calculating the system function from the lagrangian does indeed preserve this property and therefore our discrete time equations of motion are gauge invariant . the first step is to construct discrete time electromagnetic potentials . these are discussed in full detail in paper @xmath186 , but the basic properties are the following . the magnetic vector potential @xmath192 differs to the scalar potential @xmath193 in that the former is defined at temporal lattice sites whereas the latter is defined on the links between these sites . if @xmath194 is the value of the vector potential at time @xmath102 at position @xmath72 , and @xmath195 is the scalar potential on the link at position @xmath72 from time @xmath102 to time @xmath196 then under a discrete time gauge transformation we have @xmath197 where @xmath198 is the value of the gauge transformation function at time @xmath102 and position @xmath72 . the electric and magnetic fields are defined by @xmath199 these are discrete time gauge invariant . by inspection the electric field is associated with temporal links whereas the magnetic field is associated with temporal sites . in order to apply our discretisation prescription to ( [ lem ] ) we specify the virtual paths between times @xmath41 and @xmath43 to be given by @xmath200 and then the system function is given by @xmath201 under a gauge transformation we find @xmath202 and so the action sum @xmath203changes according to the rule @xmath204 in agreement with @xmath205 therefore we expect cadzow s equations of motion obtained from @xmath206 to be gauge invariant . in general , the integrals over the external electromagnetic potentials in ( [ sysfun ] ) give complicated equations of motion and we will normally have only an implicit equation for @xmath76 , which however will be gauge invariant . we find @xmath207 in the limit @xmath4 we recover the usual lorentz force law @xmath208 for those trajectories where the limit exits . the charge density @xmath209 and current density @xmath210 are defined by the following functional derivatives with respect to the electromagnetic potentials of the action sum @xmath211@xmath212 we find @xmath213 which satisfy the discrete time analogue of the equation of continuity @xmath214 we may apply our methods to the grassmannian oscillator system , which serves as a prototype model for the dirac equation studied in _ paper ii_. our model consists of one complex anticommuting degree of freedom @xmath215 with equation of motion @xmath216 the lagrangian giving these equations is @xmath217 the equations @xmath218imply the harmonic oscillator equations of motion @xmath219 now consider discretisation using the linear virtual paths @xmath220 using @xmath221we find the system function is @xmath222 \nonumber \\ & & -\omega t\frac{\left\ { 2\psi _ { n+1}^{*}\psi _ { n+1}+\psi _ n^{*}\psi _ { n+1}+\psi _ { n+1}^{*}\psi _ n+2\psi _ n^{*}\psi _ n\right\ } } 6,\end{aligned}\ ] ] which leads to the equation of motion @xmath223 and similarly for the complex conjugate . it seems not possible to use ( @xmath224 to obtain the discretisation @xmath225 of the harmonic oscillator unless we change the virtual paths or renormalise the frequency @xmath226 . however , we can readily show that @xmath227 implies harmonic oscillator behaviour by the following method . first , rewrite @xmath228 as @xmath229 where @xmath230 if we define @xmath231 and shift the degrees of freedom according to the rule @xmath232 then the variables @xmath233 satisfy the discrete time oscillator equation @xmath234 where @xmath235 we find @xmath236 from which we deduce the upper limit @xmath237 for elliptic behaviour in the system . this is exactly one half of the upper limit found for the bosonic discrete time oscillator . we now discuss the possibility of quantising our classical discrete time mechanics . if we denote the process of quantisation by the symbol @xmath238 and the process of discretisation by the symbol @xmath239then the question arises , do these processes commute , i.e. does @xmath240 . in other words , does it matter if we discretise the quantum theory of some system with classical lagrangian rather than quantise the discretised version of the same classical lagrangian ? there is _ a priori _ no reason to expect these processes to commute . for one thing , we have not yet decided what @xmath241 might mean . also , there are some aspects of @xmath242 such as the lack of a hamiltonian which makes a dynamical quantum theory of discrete time point particle mechanics problematical . fortunately , there are some concepts from the conventional @xmath243programme which are useful and appear to survive @xmath244 . we shall comment on some of these aspects now and then consider the harmonic oscillator quantisation process @xmath245 in some detail in the following sections , reserving a discussion of the @xmath246 process for paper @xmath186 . in the following we discuss a system consisting of a point particle in one dimension . generalisation to more degrees of freedom is straightforward and we shall use the dirac bra - ket notation for convenience . in our approach to quantisation , we shall follow all the standard principles of orthodox quantum mechanics in the main . this means we face the same issues of rigour and interpretation as orthodox quantum theory . we shall not comment on those in general . we discuss below those aspects where discretisation of time requires some additional emphasis or comment . * proposition 1 : * _ physical states of a quantum system correspond one - to - one to rays in a separable hilbert space _ @xmath247 . _ a physical state vector _ @xmath248 _ will be in general normalised to unity , viz _ @xmath249 * proposition 2 : * _ for each integer time _ @xmath102 _ ( or more accurately , at each co - ordinate time _ @xmath250 _ @xmath247 _ is spanned by an improper basis _ @xmath251 _ _ , the elements of which satisfy the relation _ _ @xmath252 the resolution of the identity operator in @xmath247 is @xmath253 which holds for each @xmath102 . given a physical state @xmath254 in @xmath247 we may write for each @xmath102 @xmath255 where @xmath256 is the wavefunction at time @xmath102 , with the property that @xmath257 assuming normalisation to unity . * remark 1 : * _ the heisenberg picture is being used in the above . the time dependence of the basis sets _ @xmath258 _ allows us to use the schrdinger picture , discussed below . _ * remark 2 : * _ all the usual principles of quantum mechanics concerning the interpretation of the states in the hilbert space apply here . for example , a ) the superposition principle and its interpretation according to standard quantum mechanics holds and b ) given two physical states _ @xmath259 , @xmath260 _ then the inner product _ @xmath261 _ gives the conditional transition amplitude for the system to be found in state _ @xmath262 _ _ , given it is in state _ _ @xmath260 . * definition 1 : * _ an operator _ @xmath263 _ diagonal with respect to _ @xmath258 _ is one which can be written in the form _ @xmath264 _ where the component operator _ @xmath265 _ is some differential operator of finite order_. the action of such an operator on a typical state @xmath254 is given by @xmath266 with matrix elements given by @xmath267 * remark 3 : * _ the wavefunctions of the theory are elements of _ @xmath268 _ _ , the space of square integrable functions on _ _ @xmath269 _ _ and the operators ( including the observables of the theory ) which act on them are usually built up of functions of _ _ @xmath270 _ and _ @xmath271 _ if at a given time _ @xmath102 _ the component operator of some observable diagonal with respect to _ @xmath272__happens to be represented by say a multiple of _ @xmath273 _ this carries no implication that the observable is related to a velocity ( which would normally be implied in conventional wave mechanics , where the momentum operator is represented by _ @xmath274 _ _ there is no concept of velocity in the normal sense in discrete time mechanics . _ _ keeping in mind the caveats discussed by concerning hermitian and adjoint operators , we define the hermitian conjugate or adjoint operator @xmath275 to have the property that @xmath276 for a dense set of physical states . if @xmath263 is diagonal with respect to @xmath277then assuming we may represent @xmath275 in the form @xmath278 for some operator component @xmath279 we find @xmath280 \psi _ n\left ( x\right ) .\ ] ] assuming that we are permitted to integrate by parts , which will be the case for normalisable wavefunctions falling off at spatial infinity , we can readily understand the relationship between @xmath281 and @xmath279 . given the former we can always work out the latter by integration by parts and vice - versa . if @xmath282 then @xmath283 and the operator is self - adjoint . physical observables of the theory which are diagonal with respect to @xmath258 will be assumed to have this property . the dynamical content of the theory is expressed in terms of unitary _ timestep operators _ @xmath284 , one for each @xmath102 . * proposition 3 : * _ for each _ @xmath102 _ there is a unitary operator _ @xmath284 _ such that _ @xmath285 from this we deduce the relations @xmath286 * remark 4 : * _ the operator _ @xmath284 _ provides an isometry between _ @xmath258 _ to _ @xmath287 _ _ , that is _ _ @xmath288 using the resolution of the identity ( [ res ] ) we may represent the timestep operators in the non - diagonal form @xmath289 we are now in a position to define the fundamental functions of the quantum theory , the _ system amplitudes _ @xmath290 , defined by @xmath291 from which we arrive at the non - diagonal expressions @xmath292 * remark 5 : * _ the system amplitudes will not be differential operators of finite order in general but well - behaved complex valued functions of two real variables . neither will they be singular distributions . they are similar in function to integrated feynman transition kernels encountered in the path integral formulation of standard quantum mechanics . _ the condition that the timestep operators are unitary , viz @xmath293 leads to the closure condition @xmath294 on the system amplitudes . * definition 2 : * _ a system for which the system amplitudes are independent of time , such that we may write _ _ @xmath296 _ and for all _ @xmath102 _ _ , will be said to be _ _ _ autonomous_. * definition 3 : * _ an autonomous system for which the system amplitude _ @xmath296 _ carries the symmetry _ @xmath297 _ will be said to be time - reversal invariant_. * remark 6 : * _ most of the system amplitudes of interest to us will be autonomous and time - reversal invariant . this will be so whenever we construct system amplitudes from system functions which have been obtained from conventional time - translation invariant and time - reversal invariant lagrangians using the virtual path approach discussed above_. the heisenberg picture description of the physical states used so far means that we may write @xmath298 from which we deduce @xmath299 from this we see that the system amplitudes play a role analogous to finite time scattering kernels in conventional quantum mechanics . equation @xmath300 is about the closest we come in this theory to something analogous to a time dependent schrdinger wave equation . we may set up a formal description in the schrdinger picture as follows . given a heisenberg picture state @xmath254 and a knowledge of the component functions @xmath256 , define the sequence of states @xmath301 for some chosen time @xmath187 . then if @xmath302 we find @xmath303 it is straightforward to extend this to jumps over more than one time interval . this establishes the schrdinger picture in this theory . given an improper basis @xmath304 we may construct a self - adjoint position operator @xmath305 which has the property @xmath306@xmath307 * remark 7 : * _ for convenience we shall follow the traditional abuse of notation and use the symbol _ @xmath270 _ both for the position operator and for a particular eigenvalue of that operator . it will be clear normally from the context what is meant whenever a clash of notation occurs . _ the position operators have the merit of being diagonal with respect to the appropriate basis , that is , @xmath308 is diagonal with respect to @xmath309 the position operators are not necessarily diagonal with respect to bases at other values of @xmath102 . from the condition @xmath310 we find @xmath311 which is self - adjoint but not necessarily diagonal with respect to @xmath312 * remark 8 : * _ it is possible for _ @xmath313 _ to reduce to diagonal form with respect to the _ @xmath258 _ basis but this depends on the details of the system amplitudes_. from the above we arrive at the fundamental expression for the commutators : @xmath314 = \int dxdydz\,\,|x , n\rangle u_n^{*}\left ( y , x\right ) \,y\left ( z - x\right ) u_n\left ( y , z\right ) \langle z , n| . \label{comm}\ ] ] in this subsection we discuss the quantisation of a large family of systems for which the following property holds : * definition 4 : * _ co - ordinates for a system for which the right hand side of the commutator _ ( [ comm ] ) _ is a multiple of the identity operator for each value of _ @xmath102 _ will be called normal_. * remark 9 : * _ we shall see below that the co - ordinates for the important system equivalent to the harmonic oscillator are normal . _ a class of system amplitudes for which the co - ordinates are normal may be constructed from autonomous , time - reversal invariant system functions of the form @xmath315 where @xmath316 is a non - zero constant . the cadzow s equation for this system is @xmath317 which has the merit of giving @xmath105 explicitly in terms of @xmath106 and @xmath318 . now define the system amplitude to be given by @xmath319 where @xmath320 is some constant . then from the unitarity condition @xmath321 we find @xmath322 from @xmath323 we find @xmath314 = \frac{-i\hbar } \beta , \label{zz2}\ ] ] so that the above co - ordinates for this system are normal . moreover , with the momentum @xmath324 conjugate to @xmath106 defined by @xmath325 we recover the conventional commutator @xmath326 = -i\hbar .\ ] ] this result is not expected to hold for systems which are not normal . the operator equations of motion are found to be @xmath327 where @xmath328 is the diagonal operator @xmath329 from this we obtain the discrete time version of ehrenfest s theorem ; i.e. @xmath330 for expectation values over a physical state . in our theory the quantum dynamics is completely determined by the system amplitude . suppose now that the system is autonomous and time - reversal invariant . this means that for each time @xmath102 we may write @xmath331 where @xmath296 is independent of @xmath102 . for such a system there may be constants of the motion comparable to the maeda - noether and logan invariants discussed in the classical theory . consider an operator @xmath332 diagonal with respect to @xmath258 , viz . @xmath333 where @xmath334 is some differential operator of finite order . matrix elements of the commutator of @xmath332 with @xmath335 are given by @xmath336 |\psi \rangle = \int dxdy\,\phi _ n^{*}\left ( x\right ) \left\ { c(x,\partial _ x)u\left ( x , y\right ) -\tilde{c}% ^{*}(y,\partial _ y)u\left ( x , y\right ) \right\ } \psi _ n\left ( y\right ) , \ ] ] where @xmath254 and @xmath248 are arbitrary physical states . from this we arrive at the result * theorem : * _ a diagonal operator commutes with the timestep operator of an autonomous system if _ * definition 5 : * _ a diagonal operator which commutes with the timestep operator of an autonomous system will be said to be compatible ( with the timestep operator ) . _ * remark 10 : * _ it is not necessary for a diagonal operator _ @xmath338 _ to be self - adjoint for it to be compatible with the timestep operator . _ * remark 11 : * _ from the above we deduce that compatible operators are invariants of the motion . to be explicit , consider a state _ @xmath260 _ which is an eigenstate of the diagonal operator _ @xmath332 _ with eigenvalue _ @xmath339 _ _ , i.e. _ _ @xmath340 _ then we can show _ @xmath341 _ and _ @xmath342 * remark 12 : * _ given the system amplitude _ @xmath296 _ it may be very hard or perhaps even impossible to find any compatible operators in closed form . it may be necessary to approximate such an operator via a perturbative expansion , for example . this is the quantum theory analogue of the problem of finding invariants of the motion for a classical discrete time theory given some system function . _ * remark 13 : * _ discrete time and continuous time quantum mechanics pose dual problems in the following sense . in continuous time quantum mechanics we are normally given a hamiltonian and the problem is to construct the time evolution operator . for a time independent hamiltonian a complete solution would require us to find all the eigenvalues _ @xmath343 _ and eigenstates _ @xmath344 _ of _ @xmath345 _ and then use them in the formal solution _ @xmath346 _ to write _ @xmath347 _ _ this is in general a formidable problem . in discrete time quantum mechanics the situation is the other way around . given a system amplitude , the problem is to find the compatible operators , if any . an important system where answers can be found to all of these questions in both approaches is the discrete time harmonic oscillator discussed in _ _ @xmath348 . in principle it should always be possible to construct examples of discrete time quantum systems by integrating the equation @xmath349 for the evolution operator @xmath350 in continuous time quantum mechanics , given the hamiltonian @xmath351 the boundary condition @xmath352 ensures a unique solution . from this point of view the discussion outlined above represents a stroboscopic approach , where the state vectors evolve continuously but are looked at periodically in a non - destructive ( mathematical ) sense . this examination of the state vector is * not * the same as an observation collapsing the wavefunction . for autonomous systems a formal solution to @xmath353 is @xmath354 in this approach the transition amplitude @xmath355 corresponds to our system amplitude when @xmath356 and may be evaluated in a number of ways . for example the feynman path integral method gives the formula @xmath357 \exp \left\ { \frac i\hbar \int_0^tdt^{\prime } l\left ( \dot{z},z , t^{\prime } \right ) \right\ } , \;\;\;t>0,\ ] ] where @xmath358 is the lagrangian , such amplitudes being functions of @xmath1 and the end - points @xmath270 and @xmath359 . the standard approach to the evaluation of such integrals is , rather interestingly , based on the discretisation of time . the time interval @xmath360 $ ] is partitioned into a finite number @xmath35 of equal steps , the integrand in the exponential is approximated suitably ( by what amounts to choosing a virtual path in our approach ) , the @xmath35 integrals evaluated , and then the limit @xmath361 taken . the relationship between this approach and our discrete time formalism should now be clear , the basic difference being that we do not take the limit @xmath361 . in a number of situations our system amplitude will actually take the form @xmath362 and then for @xmath35 timesteps the transition amplitude @xmath363 becomes @xmath364 which emphasises the relationship further . an important point which could be confusing is that the system function in ( [ point ] ) does * not * correspond to a logan constant for those systems such as the harmonic oscillator where such an invariant can be found . system functions in general are not expected to be invariants of the motion . an alternative method of constructing the transition amplitudes is to find the green s functions for the system . if the transition amplitude @xmath365 satisfies the homogeneous schrdinger equation @xmath366 with the boundary condition @xmath367 then the retarded and advanced green s functions @xmath368 and @xmath369 are related to the transition amplitude by @xmath370 and these satisfy the inhomogeneous equation @xmath371 if we can solve this equation then we can immediately construct the transition amplitude using the relation @xmath372 given the hamiltonian @xmath373 in continuous time mechanics we can readily find the greens functions in the quantum theory . we find @xmath374 from which we construct the transition amplitude @xmath375 this satisfies the closure condition @xmath376@xmath377 for a system amplitude and demonstrates the essential point that it is possible to construct examples of discrete time quantum mechanics from continuous time quantum mechanics . the converse need not be true . given a system amplitude , it may be impossible to find a compatible operator equivalent to some second order hamiltonian operator in continuous time mechanics . it is not difficult to find examples of normal systems where this occurs . in this section we discuss the inhomogeneous harmonic oscillator , which serves as a prototype for the application of our quantisation principles to field theories . we will use the source functional techniques of schwinger to obtain the ground state functional and various @xmath102-point functions of interest . because the schwinger method deals with time ordered products we should expect the discretisation of the time parameter to involve some changes in the details of the calculations . first , given a system function @xmath411 we are free to introduce an external source in any convenient way , as ultimately this will be set to zero . our choice is to define the system function @xmath412 $ ] in the presence of the external source as @xmath413 \equiv f^n+\frac{_1}{^2}tj_{n+1}x_{n+1}+\frac{_1}{^2}% tj_nx_n,\ ] ] a choice which allows the construction of time ordered product expectation values directly . the action sum from time @xmath414 to time @xmath415 ( @xmath416 now becomes @xmath417 = a^{nm}+\frac{_1}{^2}tj_mx_m+\frac{_1}{^2}% tj_nx_n+t\sum_{n = m+1}^{n-1}j_nx_n,\;\;\;m < n\ ] ] from which cadzow s equation of motion is found to be @xmath418 quantisation is introduced via the schwinger action principle modified for discrete time . we postulate that for an infinitesimal variation @xmath419 $ ] of the action operator then @xmath420 |\psi , m\rangle ^j,\;\;\;m < n\ ] ] for any states @xmath421 , @xmath422 at times @xmath415 , @xmath414 respectively , with evolution in the presence of the source . independent variation of the @xmath423 for @xmath424 then leads to the equations @xmath425 further application of the principle leads to expectation values of time ordered product of operators , such as @xmath426 where the symbol @xmath427denotes discrete time ordering . for example , @xmath428 where @xmath429 is the discrete step function , defined by @xmath430 and @xmath431 is the kronecker delta , defined by @xmath432 given the harmonic oscillator system function @xmath433 then the classical discrete time harmonic oscillator in the presence of the external source @xmath434satisfies the equation @xmath435 where @xmath436 @xmath437 with @xmath438 as discussed previously , elliptic ( oscillatory ) solutions occur for @xmath439 whereas hyperbolic solutions occur for @xmath147 we will now discuss these possibilities separately . the importance of the elliptic regime @xmath139 stems from the fact that in field theory this corresponds to physical particle configurations of the fields , i.e. , solutions which can be normalised . now define the action of the @xmath440 _ discrete time displacement operator _ @xmath441 by the rule @xmath442 for any function of the index @xmath443where @xmath102 is real . then @xmath444 may be written in the form @xmath445 to solve @xmath446 for the elliptic case we first define the following : since @xmath447we write @xmath448 and @xmath449 taking @xmath450 . if we define@xmath451 where @xmath452 is real then a useful identity is @xmath453 from this we deduce @xmath454 which is equivalent to @xmath455 we define the matrices @xmath456 , \;\;\;\;\;\lambda \equiv \lambda ^1,\ ] ] and use @xmath457 to prove @xmath458 if we write @xmath459 , \;\;\;\;\;\mathbf{j}_n\equiv \left [ \begin{array}{c } \frac t\beta j_n \\ 0 \end{array } \right]\ ] ] then @xmath446 may be written in the form @xmath460 this equation may be readily solved using the properties of the @xmath461 functions and by diagonalising the matrix @xmath462 . we choose feynman boundary conditions , specifying the particle to be at position @xmath463 in the past ( at time @xmath414 ) and at position @xmath464 in the future ( at time @xmath415 ) , giving @xmath465 @xmath466which is valid only for @xmath467 this can be tidied up into the form @xmath468 where @xmath469 then@xmath470 satisfies the inhomogeneous equation @xmath471 up to this stage we have taken @xmath472 with both finite , but normally we will be interested in the scattering limit @xmath473 also , we have appeared to have overlooked the possibility that @xmath474 vanishes in the denominator of the propagator @xmath475 for some values of @xmath35 and @xmath476 . we shall address both of these issues directly now . our method of avoiding possible singularities is to extend the feynman @xmath477 prescription to the @xmath478 parameter . by inspection of the equation @xmath479 we deduce that @xmath480 with this deformation of the @xmath478 parameter and taking the limit @xmath481 we find @xmath482 where @xmath483 satisfies the homogeneous equation @xmath484 and @xmath485 this is the discrete time analogue of the harmonic oscillator feynman propagator and reduces to it in the limit @xmath486 . a direct application of the discrete time schwinger action principle to the operator equation of motion @xmath487 then gives the ground state vacuum functional @xmath488 = z\left [ 0\right ] \exp \left\ { \frac{-it^2}{2\hbar } % \sum_{n , m=-\infty } ^\infty j_ng_f^{n - m}j_m\right\ } , \ ] ] essentially solving the quantum problem . using this result and @xmath489 we find that in the limit @xmath490@xmath491 from this we deduce @xmath492|0\rangle = \frac{-i\hbar } \beta , \ ] ] which agrees exactly with the discrete time oscillator commutation relation @xmath493=\frac{-i\hbar } \beta \label{com1}\ ] ] found previously . further ground state expectation values of commutators may be obtained by using the result @xmath494 for example , we find @xmath495 |0\rangle = \frac{-2i\hbar \eta } \beta , \ ] ] which agrees with the commutator @xmath496 = \frac{-2i\hbar \eta } \beta\ ] ] obtained from the operator equation of motion @xmath497 and the commutator @xmath498 another verification of the consistency of our methods is that we may use @xmath499 directly to find the ground state expectation value of the logan invariant ( [ op1 ] ) for the discrete time harmonic oscillator . we find @xmath500 which agrees exactly with previous results . because @xmath501 is real and positive the controlling parameter @xmath436 as given by ( [ eta ] ) takes vales only in the regions @xmath502 if we parametrise @xmath436 by the rule @xmath503 where @xmath504 is complex then if we take @xmath505 then the range of possibilities ( [ range ] ) corresponds to a contour @xmath506 in the complex @xmath507 plane which runs just below the real axis from the origin to @xmath508 and then runs from @xmath508 to @xmath509 the elliptic regime corresponds to values of @xmath504 on the first part of the contour , for which @xmath510 where @xmath115 is infinitesimal and positive , corresponding to the feynman @xmath511 prescription . the hyperbolic region corresponds to the part of the contour given by @xmath512 for this region analytic continuation of the @xmath513 functions leads to @xmath514 where @xmath515 from this the analytic continuation of the finite interval propagator ( [ dprop ] ) gives @xmath516 which satisfies the equation @xmath517 taking the limit @xmath518 gives the infinite interval propagator @xmath519 which satisfies the equation @xmath520 the elliptic and hyperbolic feynman boundary condition propagators can be summarised in the analytic form @xmath521 where @xmath522 and @xmath504 lies somewhere on the contour @xmath13 discussed above . then @xmath523 satisfies the equation @xmath524 however , physical states correspond only to points in the elliptic regime , as the wave functions will not remain normalisable in time otherwise . for example , we showed previously that the ground state wavefunction for the discrete time oscillator in the elliptic regime is given by @xmath525 demonstrating that analytic continuation to the hyperbolic regime will not give a normalisable ground state wave - function . in this paper we have shown that a consistent approach to the discretisation of time results in a consistent dynamical framework . once the initial psychological hurdle of accepting a dynamics without time derivatives has been jumped , then such a theory becomes as reasonable as continuous time dynamics . indeed , by taking the fundamental time interval @xmath19 small enough , it would appear possible to duplicate or approximate conventional theory as closely as required . the obvious question , _ why consider discrete time mechanics at all ? _ has two replies . first , it may be the case after all that there is some fundamental limit to time intervals , and so it becomes a matter of curiosity as to how far we can go along that road . second , there may be some novel properties in this approach which could prevent or alleviate the notorious problems with divergences which plague conventional field theories . it is worth considering any approach to the regularisation in field theory which is based on just one assumption , namely that of a discrete time . the behaviour of the discrete time oscillator holds the promise of potentially useful properties which may provide a cutoff for particle energy . in the next paper we apply our methods to classical field theories . however no divergence problems appear at that stage . in the third paper of this series , we shall consider second quantisation , and the issues to do with divergences of feynman diagrams will be discussed in subsequent papers in some detail . we are grateful to prof . j lukierski for providing us with important references and discussion on discrete time mechanics . * jaroszkiewicz g * , _ deformed mechanics and the discrete time quantised anharmonic oscillator _ , proceeding of @xmath526 karpacz winter school of theoretical physics , quantum groups , formalism and applications , university of wrocaw , poland , february 14 - 26 ( 1994 )

we have combined _ hst_/fos and ground - based spectra of the seyfert 1 galaxy ngc 5548 to study the narrow emission lines over the 1200 10,000 region . all of the spectra were obtained when the broad emission line and continuum fluxes were at an historic low level , allowing us to accurately determine the contribution of the narrow - line region ( nlr ) to the emission lines . we have generated multicomponent photoionization models to investigate the relative strength of the high ionization lines compared to those in seyfert 2 galaxies , and the weakness of the narrow mg ii line . we present evidence for a high ionization component of nlr gas that is very close to the nucleus ( pc ) . this component must be _ optically thin _ to ionizing radiation at the lyman edge ( i.e. , 2.5 ) to avoid producing [ o i ] and mg ii in a partially ionized zone . the very high ionization lines ( n v , [ ne v ] , [ fe vii ] , [ fe x ] ) are stronger than the predictions of our standard model , and we show that this may be due to supersolar abundances and/or a `` blue bump '' in the extreme ultraviolet ( although recent observations do not support the latter ) . an outer component of nlr gas ( at only pc from the continuum source ) is needed to produce the low ionization lines . we show that the outer component may contain dust , which further reduces the mg ii flux by depletion and by absorption of the resonance photons after multiple scatterings . we show that the majority of the emission in the nlr of ngc 5548 must arise within about pc from the nucleus . thus , the nlr in this seyfert 1 galaxy is very physically compact , compared to the typical nlr in seyfert 2 galaxies .

astro-ph9711324

ngc 5548 is a bright , low - redshift ( z @xmath4 0.0172 ) seyfert 1 galaxy that has received a considerable amount of attention over the past decade . in particular , ngc 5548 has been the subject of a number of intensive spectroscopic monitoring campaigns in the optical and uv ( korista et al . these efforts have yielded important results on the nature of the continuum source and the size , geometry , and kinematics of the broad - line region ( blr ) , for which the responsivity peaks at about 20 light days from the nucleus ( peterson et al . little attention has been paid to the narrow - line region ( nlr ) in ngc 5548 , however , although it could lead to a better understanding of the circumnuclear environment at much greater distances from the nucleus of this otherwise well - studied active galaxy . in general , studies of the nlr in active galaxies are important for understanding the nature of the nlr clouds and the interaction of the central continuum source with the surrounding galaxy on large scales . emission - line studies and detailed photoionization modeling are particularly useful for determining the range of physical conditions and reddening amongst the nlrs in active galaxies . a comparison of the nlr properties in seyfert 1 and seyfert 2 galaxies should be helpful in testing unified theories , which postulate that the two types are the same object viewed from different perspectives , such that the continuum source and blr are `` hidden '' in seyfert 2 galaxies ( miller & goodrich 1990 ; antonucci 1993 ) . if this basic hypothesis is correct , then the intrinsic properties of the nlrs in seyfert 1 and seyfert 2 galaxies should _ not _ show large systematic differences . the narrow lines in seyfert 1 galaxies are more difficult to measure than in seyfert 2 galaxies , due to blending with the broad lines . however , given spectra with sufficient signal - to - noise ratio and spectral resolution , these components can be isolated and measured reasonably well ( crenshaw & peterson 1986 ) . measurements of the narrow lines in the optical are given by cohen ( 1983 ) for a large number of seyfert 1 galaxies . in the uv , these measurements were difficult with the low spectral resolution of the _ international ultraviolet explorer _ ( _ iue _ ) , but are possible with instruments on the _ hubble space telescope _ ( _ hst _ ) , such as the faint object spectrograph ( fos ) . the fos uv spectra of ngc 5548 presented in a previous paper ( crenshaw , boggess , & wu 1993 ; hereafter paper i ) provide a good starting point for detailed studies of the nlr in a seyfert 1 galaxy . these observations happened to occur at a time when the broad emission lines ( and continuum fluxes ) were at an historic low in the uv , and the contrast between the broad and narrow components is thereby enhanced . paper i gives the uv spectrum and measurements of the broad and narrow lines in ngc 5548 . relative to the other narrow lines , c iv @xmath01549 is much stronger in ngc 5548 than in seyfert 2 galaxies , indicating a higher ionization parameter and/or harder continuum in the nlr of ngc 5548 . narrow mg ii @xmath02800 emission is very weak or absent in ngc 5548 , and paper i presents two possible explanations : 1 ) the nlr clouds lack the presence of a partially - ionized zone ( i.e. , they are optically thin to ionizing radiation ) , and/or 2 ) dust grains are present in the nlr clouds , and the mg ii flux is weak due to depletion and/or destruction from multiple scatterings and eventual absorption of the photons by dust ( kraemer & harrington 1986 ; ferland 1992 ) . we now have the opportunity to investigate the preliminary results from paper i in more detail , by including ground - based optical spectra and photoionization models . from the ground - based monitoring campaigns , we have selected spectra that cover the full optical range ( 3000 10,000 ) and were observed around the same time period as the uv data , when the broad - line fluxes were very low . the combination of optical and uv lines provides a wide range of emission - line diagnostics , as well as an opportunity to deredden the lines using the he ii recombination lines . we can then use multicomponent photoionization models to match the dereddened line ratios and probe the physical conditions in the nlr of ngc 5548 . we obtained the fos uv spectra of ngc 5548 through a 10 circular aperture on 1992 july 5 ut . paper i gives the details of the observations and measurements , along with the uv spectrum and emission - line fluxes ( with associated errors ) . the observations were made prior to the installation of costar on _ hst _ , so near - simultaneous _ iue _ spectra were used to adjust the absolute flux levels of the fos spectra . as noted in paper i , the scale factors needed to bring the fos continuum fluxes up to the _ iue _ levels are around 1.4 1.5 , which are somewhat higher than the values of 1.1 1.3 for our other seyfert observations . we concluded that the seyfert nucleus may not have been accurately centered in the aperture . koratkar et al . ( 1996 ) suggest that another possible explanation for the discrepancy in absolute fluxes is nonlinearity in the _ iue _ detectors . however , we have seen no evidence for this possibility in other observations at these flux levels , so we have no reason to distrust the _ iue _ fluxes . in addition , reprocessed versions of these spectra that we obtained from the _ hst _ and _ iue _ archives have not changed the original fluxes by more than 10% , so we continue to use the values from paper i. some of the emission lines in paper i have only a single number quoted for the flux ( as opposed to separate values for the broad and narrow components ) ; a single value represents the narrow - line contribution , since the broad component is either not present or too weak to be detected in these cases . we selected two optical spectra obtained during a four - year monitoring campaign on ngc 5548 ( peterson et al . 1994 ) , from a time interval of @xmath130 days when the h@xmath5 and continuum light curves were at their lowest levels to date . the spectra were chosen on the basis of their large wavelength coverage ( 3000 - 10000 ) , high signal - to - noise ratio ( @xmath650 per resolution element in the continuum at 5200 ) , and acceptable resolution ( @xmath18 ) . the spectra were obtained through a 40 x 100 aperture with the 3.0-m shane telescope @xmath7 kast spectrograph on 1992 april 21 and 1992 may 23 ut . additional details on the observations are given by peterson et al . the absolute flux levels were adjusted by scaling the optical spectra so that the [ o iii ] @xmath05007 flux is 5.58 x 10@xmath8 ergs s@xmath9 @xmath10 , a value determined from observations through large apertures on spectrophotometric nights ( peterson et al . the scale factors we used are 1.36 for the 1992 april 24 spectrum and 1.01 for the 1992 may 23 spectrum . plots of the optical spectra are shown in figure 1 ( the uv spectrum is shown in paper i ) . the contrast between the broad and narrow components of the permitted emission lines is most clearly seen in h@xmath5 . the 1992 april 21 spectrum was obtained at an historic low level , with a continuum flux of f@xmath11(5100 ) @xmath4 5.5 x 10@xmath12 ergs s@xmath9 @xmath10 @xmath9 and total h@xmath5 flux of f(h@xmath5 ) @xmath4 3.2 x 10@xmath8 ergs s@xmath9 @xmath10 . the flux levels are a little higher for the 1992 may 23 spectrum with f@xmath11(5100 ) @xmath4 6.0 x 10@xmath12 ergs s@xmath9 @xmath10 @xmath9 and f(h@xmath5 ) @xmath4 3.7 x 10@xmath8 ergs s@xmath9 @xmath10 . at the time of the fos observations on 1992 july 5 , the continuum and h@xmath5 fluxes were close to the same levels as those from the second optical spectrum , according to the light curves of peterson et al . ( 1994 ) . although the optical aperture is much larger than the one used for the uv observations , we have substantial evidence that it does not contain much additional nlr flux . unfortunately , there are no _ hst _ narrow - band images in [ o iii ] or other strong lines that could be used to directly determine the distribution of narrow - line emission close to the nucleus . however , there is significant evidence that the apparent size of the nlr is very small in ngc 5548 . peterson et al . ( 1995 ) find from a ground - based image that the [ o iii ] emission is pointlike , given a point - spread function that is characterized by a width of 20 ( fwhm ) . in addition , wilson & ulvestad ( 1982 ) show that in an aperture that is 42 in diameter , the [ o iii ] @xmath05007 fluxes at positions offset from the nucleus by 45 6@xmath13 are about 100 times weaker than the nuclear flux . more importantly , in paper i we found that the strongest uv lines in the _ iue _ 20@xmath13 x 10@xmath13 aperture have fluxes that are only @xmath1 20% higher than those in the fos 10 aperture . thus , the observed uv to optical line ratios that we quote are at most 20% too low , which has little effect on our comparisons with the model results . in order to measure the flux of each narrow optical line , we used a local baseline determined by linear interpolation between adjacent continuum regions or broad profile wings ( in the case of profiles consisting of broad and narrow components ) . for severely blended lines like h@xmath14 and [ n ii ] @xmath156548 , 6584 , we used the [ o iii ] @xmath05007 profile as a template to deblend the lines ( see crenshaw & peterson 1986 ) . the adopted flux for each narrow component is the average of the values from each of the two spectra . we determined the reddening of the narrow emission lines from the he ii @xmath01640/@xmath04686 ratio and the galactic reddening curve of savage & mathis ( 1979 ) . for the temperatures and densities typical of the nlr , the he ii lines are due to recombination , and this particular ratio only varies from 6.3 to 7.6 ( seaton 1978 ) ; we adopt an intrinsic value of 7.2 , consistent with our model values ( section 3 ) . the observed he ii @xmath01640/@xmath04686 ratio is 5.5 @xmath16 1.6 , which yields a reddening of e@xmath17 @xmath4 0.07 mag @xmath18 . the portion of the reddening that is due to our own galaxy is e@xmath17 @xmath4 0.03 mag , determined from a neutral hydrogen column density of n@xmath19 @xmath4 1.6 x 10@xmath20 @xmath10 ( murphy et al . 1996 ) and the relationship e@xmath17 @xmath4n@xmath19/5.2 x 10@xmath21 @xmath10 ( shull & van steenburg 1985 ) . we note that the intrinsic reddening of the narrow emission lines in this seyfert 1 galaxy , e@xmath17 @xmath3 0.04 mag , is much smaller than typical values of 0.2 0.4 mag obtained for seyfert 2 galaxies ( macalpine 1988 ; ferland & osterbrock 1986 ; kraemer et al . 1994 ) . we determined errors in the dereddened ratios from the sum in quadrature of the errors from three sources : photon noise , different reasonable continuum placements , and reddening . errors in the optical ratios are dominated by continuum placement , whereas errors in the uv to optical ratios are due to both continuum placement and uncertainties in the reddening correction . errors in the weak lines in both regions also have a significant contribution from photon noise . as we discussed earlier in this section , there are some possible sources of systematic error in the uv to optical line ratios , on which we placed upper limits of @xmath120% . table 1 gives the observed and dereddened narrow - line ratios relative to h@xmath5 , and errors in the dereddened ratios . cohen ( 1983 ) gives the next most comprehensive list of optical line ratios ; in general , cohen s observed ratios agree with ours to within the errors . a number of investigators have independently determined the narrow h@xmath5/[o iii ] @xmath05007 ratio in ngc 5548 ( cohen 1983 ; crenshaw & peterson 1986 ; peterson 1987 ; wamsteker et al . 1990 ; rosenblatt et al . 1992 ; wanders & peterson 1996 ) : these values range from 0.10 to 0.15 , compared to our value of 0.12 @xmath16 0.01 . estimates of the ionizing continuum are needed as input values for the photoionization models of the nlr . we choose the continuum data points given by krolik et al . ( 1991 ) , since they represent the historic mean levels for this object . as always , the greatest uncertainty is the shape of the extreme ultraviolet ( euv ) continuum . figure 2 gives the uv continuum point closest to the euv region , at 1340 , and the x - ray continuum points from krolik et al . turner & pounds 1989 ; clavel et al . 1991 ) in terms of luminosity ( ergs s@xmath9 hz@xmath9 ) , which we have adjusted for a hubble constant of h@xmath22 @xmath4 75 km s@xmath9 mpc@xmath9 . the dotted line in figure 1 gives krolik et al.s continuum fit in the euv , which is a power law determined from the uv data along with an exponential cutoff designed to meet the first x - ray point . we prefer a fit with two power laws ( in the form l@xmath23 @xmath4 k@xmath24 ) , given the evidence for an upturn in the spectrum at energies smaller than 1 2 kev ( i.e. , a soft x - ray excess ) . a fit to these data yields @xmath14 @xmath4 @[email protected] in the euv and soft x - rays , and @xmath14 @xmath4 @[email protected] in the hard x - rays ; the break point is at @xmath26 @xmath4 10@xmath27 hz@xmath9 ( 1.3 kev ) . ngc 5548 was monitored by the _ extreme ultraviolet explorer _ ( _ euve _ ) over a two month period during 1993 march may ( marshall et al . 1997 ) . during this time , the euv flux varied by a factor of four from peak to minimum , and the average flux ( corrected for galactic neutral hydrogen absorption ) was 135 @xmath28jy at @xmath176 . these observations provide an important constraint on the euv ionizing continuum , but were not used directly in our continuum fit for two reasons . first , the neutral hydrogen absorption due to our galaxy is well known ( see the previous section ) , but there could be additional absorption along the line of sight . second , the _ euve _ flux , averaged over two months , may not be representative of the average flux over many years . given these caveats , we plot the _ euve _ continuum point in figure 2 for comparison with our adopted continuum ; the error bar was determined from an estimate of @xmath1610% uncertainty in galactic n@xmath19 ( see murphy et al . 1996 ) . the _ euve _ point is slightly higher than the continuum fits in figure 2 , but appears to be consistent with our and krolik et al.s adoption of a relatively steep continuum . if we use krolik et al.s continuum or a continuum formed by joining the uv , _ euve _ , and x - ray points with line segments ( in log space ) , the total euv flux increases by factors of only 1.19 and 1.30 , respectively , and the flux at the frequency of the _ euve _ observation increases by factors of 1.40 and 1.84 , respectively . the effects of adopting these other continua are small , and will be discussed later in the paper . in modeling the narrow - line emission of ngc 5548 , we have adhered to our basic philosophy of keeping the number of free parameters to a minimum , by using the available observational constraints and the simplest assumptions possible . the parameters are varied until the best agreement is obtained with the observed line ratios , and additional input parameters are only included if they are needed to provide a reasonable match to the majority of the lines . discrepancies between the model predictions and specific lines are then investigated to provide further insight into the physical conditions . in some cases we generate variations on the standard model using additional parameters ( such as dust ) or nonstandard values of the initial parameters ( such as nonsolar abundances ) to illustrate our ideas for resolving the discrepancies . the basic modeling methodology that we employ is described in kraemer et al . ( 1994 ) and the details of the photoionization code are given in kraemer ( 1985 ) . to review the major points , we assume plane parallel geometry , which is reasonable if the gas is ionized by radiation from a central source at a distance that is large compared to the extent of the cloud . the gas is assumed to be atomic ( i.e. , there is no molecular component ) . for radiation bounded models , we stop the integration into the slab when the electron temperature falls below 5000 k and there is no longer any significant line emission . the emission line photon escape is through the ionized face of the slab . details of the treatment of dust in the models are described in kraemer ( 1985 ) . since the work of kraemer et al . ( 1994 ) , we have added iron to the elements modeled in the code . the atomic data that we used can be found in pradhan & peng ( 1994 ) , and references contained therein , as well as through ferland s `` cloudy and associates '' world wide web site ( http://www.pa.uky.edu/@xmath1gary/cloudy ) . the final output of these models is an emission line spectrum . the line strengths are tabulated relative to h@xmath5 . in addition , the model gives the volume emissivity of h@xmath5 , from which we can determine the mass of gas required to produce the observed line emission , the efficiency of production of h@xmath5 photons , and an estimate of the covering factor . in order to keep the input parameters to a minimum , we kept two of them fixed for our standard model : the shape of the ionizing continuum and the abundances . we used the simplest possible ionizing continuum consistent with the observations , as described in section 2.2 . in addition , we have assumed solar elemental abundances for the standard model as follows ( see lambert & luck 1978 ) : he = 0.1 , c = 3.4 x 10@xmath29 , o = 6.8 x 10@xmath29 , n = 1.2 x 10@xmath29 , ne = 1.1 x 10@xmath29 , s = 1.5 x 10@xmath29 , si = 3.1 x 10@xmath30 , mg = 3.3 x 10@xmath30 , fe = 4 x 10@xmath30 ( relative to hydrogen by number ) . our photoionization models are parameterized in terms of the density of atomic hydrogen ( n@xmath31 ) and the dimensionless ionization parameter at the illuminated face of the cloud : @xmath32 where _ l@xmath23 _ is the frequency - dependent luminosity of the ionizing continuum , _ d _ is the distance between the cloud and the ionizing source , and h@xmath33 = 13.6 ev . we show that we must add two enhancements to our standard model to obtain an acceptable match to the observations . first , we need two components of gas , characterized by different ionization parameters and densities . second , we show that the inner component must be optically thin ( i.e. , radiation bounded ) at the lyman edge ( 13.6 ev ) . we are then able to vary the ionization parameter and density of each component to match the observations . of course , the resulting standard model is an oversimplification , since it is likely that the nlr clouds are characterized by a number of different ionization parameters , densities , and optical depths . we have effectively averaged the initial conditions for each of these components to fit the largest selection of line ratios . given this simplification , the two - component model gives a surprisingly good fit to the observations . the narrow - line spectrum of ngc 5548 is dominated by high ionization lines , such as c iv @xmath01549 , n v @xmath01240 , and [ ne v ] @xmath15 3346 , 3426 , as well as the coronal lines of [ fe vii ] and [ fe x ] . this indicates that there is a high ionization component relatively near the central source , but presumably outside the blr , since these lines are much narrower ( @xmath34 500 km s@xmath9 fwhm ) than the broad lines ( @xmath1 5000 km s@xmath9 fwhm ) . also present in the spectrum are relatively strong lines of [ n ii ] @xmath156584 , 6548 and [ o ii ] @xmath03727 . these are typical of the narrow emission line regions of many seyfert 2 galaxies ( koski 1978 ; shuder & osterbrock 1981 ) , and presumably arise in a component of relatively low ionization and low density . the choice of density for these models is based on the relative strengths of certain forbidden emission lines in the spectrum of ngc 5548 . an emission line will not be an important coolant in gas with sufficiently large electron density that collisional de - excitation of the line will dominate over radiative transition ( see osterbrock 1989 ) . for example , the observed [ ne v ] @xmath03426/h@xmath5 ratio indicates that the density of the component in which that emission originates must be less than @xmath15 x 10@xmath35 @xmath36 ( derobertis & osterbrock 1984 ) . given the hardness of the ionizing continuum , one would expect high temperatures in this highly ionized gas . the lower limit on the density of this component can be estimated from the relative strength of the [ o iii ] @xmath05007 emission . the ratio of [ o iii ] @xmath05007/ [ ne v ] @xmath03426 is less than 5 , which indicates either low density , highly ionized gas , as is often seen in the extended nlr of seyfert 2 galaxies ( storchi- bergmann et al . 1996 ) , or density greater than 10@xmath37 @xmath36 and some collisional supression of the @xmath05007 line . the ratio of [ o iii ] @xmath04363/@xmath05007 is very high ( @xmath10.09 ) . in the low density limit , this ratio is @xmath17 x 10@xmath38 at a temperature of 10@xmath39 k ( osterbrock 1989 ) . at this low density , it is unlikely that temperatures consistent with photoionization equilibrium can increase this ratio to the observed value , and probable that the large @xmath04363/@xmath05007 ratio is due to the fact that some of this emission arises in gas of high density ( i.e. , @xmath40 10@xmath37 @xmath36 ) . at this density and level of ionization , the [ n ii ] @xmath156548 , 6584 emission from this component will be negligible , so there must be a lower ionization and lower density component present . in order for the [ n ii ] lines to be among the principal coolants from this component , its density must be less than 1 x 10@xmath41 @xmath36 . other studies ( cf . filippenko & halpern 1984 ; filippenko 1985 ; kraemer et al . 1994 ) have shown that a range of densities in the nlr of seyfert galaxies is likely , so it is not surprising that this condition exists in ngc 5548 . to summarize , the range in density , along with the presence of emission lines from a wide range of ionization states , indicates that more than one model component is needed to fit the nlr spectrum of ngc 5548 . the existence of strong high ionization lines such as c iv @xmath01549 , n v @xmath01240 , and [ ne v ] @xmath153346 , 3426 , and our evidence for high densities in the region that they are produced , requires a component of gas relatively close to the central source . the weakness of mg ii @xmath02800 and [ o i ] @xmath156300 , 6364 indicate that these gas clouds lack a significant partially ionized zone , and therefore must be _ optically thin _ to the ionizing radiation ( or `` matter bounded '' ) . we further investigate the conditions in these two components below . our approach in modeling ngc 5548 was to fit the high ionization component first and then add components as needed to fit the lower ionization lines ( in the end , only one additional component was needed ) . given the constraints and assumptions described in the previous section , we arrived at values of n@xmath31 @xmath4 1 x 10@xmath35 @xmath36 and u @xmath4 10@xmath42 for the high ionization component . substantially lower densities would result in [ o iii ] @xmath05007 being too strong , and higher densities would quench the [ ne v ] emission . a higher ionization parameter is possible but , given our euv continuum , would not increase the relative strengths of any of the high ionization lines other than [ fe x ] @xmath06374 , at the expense of putting this component at distances much closer than @xmath11 pc from the continuum source ( see section 4 ) . models were run to varying optical depth at the lyman limit @xmath2 , with the constraint that the mg ii and [ o i ] lines could not become too strong . after comparing the results of models run with @xmath2 @xmath4 1.5 to 10 , we found that @xmath2 @xmath4 2.5 gave the best fit . the emission line spectrum from this model , inner , is given in table 2 . a second component , outer , was needed to fit the lower ionization lines . we found that u @xmath4 10@xmath43 and n@xmath31 @xmath4 2 x 10@xmath39 @xmath36 gave a good simultaneous fit to the [ o iii ] @xmath05007/[o ii ] @xmath03727 and [ n ii ] @xmath06584/h@xmath5 ratios . unlike our models for mrk 3 and i zw 92 ( kraemer & harrington 1986 ; kraemer et al . 1994 ) , there was no need to add a third component for ngc 5548 , since there is no obvious contribution from a component of very low density ( @xmath44 10@xmath45 @xmath36 ) low ionization gas , such as very strong [ o ii ] @xmath03727 and [ n i ] @xmath05200 lines . the resulting emission line spectrum from outer is also included in table 2 . ( we will discuss two variations on inner and outer in the next section . ) in order to fit the observed ( and dereddened ) narrow - line spectrum of ngc 5548 , we combined the output spectrum of the two standard components inner and outer . in previous studies , we attempted to weigh the contributions from each component to fit specific emission line ratios . for the model of ngc 5548 , we simply took an equal contribution from inner and outer . the relative simplicity of the narrow - line spectrum and lack of a strong contribution from very low - ionization gas makes such a simple fit possible . the combined spectrum is given , along with the dereddened observed spectrum for comparison , in table 3 ( horizontal lines in the table indicate that the models do not predict the strengths of these emission lines ) . comparison of the model predictions to the dereddened observed spectrum in table 3 shows agreement , to within the errors , for most of the lines . in particular , these include c iv @xmath01549 , he ii @xmath01640 , [ o ii ] @xmath03727 , [ ne iii ] @xmath03869 , [ n ii ] @xmath06584 , and the balmer decrement . in radiation bounded gas , the ratio of the he ii lines to h@xmath5 is strongly dependent on the shape of the ionizing continuum , because neutral hydrogen is the dominant absorber of ionizing radiation between 13.6 ev and 54.4 ev , while above 54.4 ev , singly ionized helium dominates . if there is a component of matter bounded gas , he ii / h@xmath5 is less easily predicted . the accuracy of our fit to this ratio indicates that the relative contributions of the matter and radiation bounded components are approximately correct , given the observational constraints on the ionizing continuum . the fact that we have a reasonable fit for lines that span a wide range of ionization and critical densities supports our values for density and ionization parameter . most of the discrepancies between the observations and models are in the lowest and highest ionization lines , which we will address below . first , we address differences between the predicted and observed ratios for the low ionization lines . the mg ii @xmath02800 and [ o i ] @xmath156300 , 6364 lines are still predicted to be too strong by our standard model , by factors @xmath6 6 and 2.5 , respectively . nearly all this emission is coming from outer . two factors determine the strength of the [ o i ] lines : the hardness of the ionizing continuum and the physical depth of the emission line clouds . since it appears that we have a good fit for the ionizing continuum , the weakness in the observed [ o i ] lines gives a limit on the depth of the clouds . truncating the integration of outer at @xmath2 @xmath3 1000 would give a better fit to the [ o i ] without affecting the other important line ratios . this results in a cloud depth of @xmath12.5 x 10@xmath46 cm . the overprediction of the mg ii @xmath02800 line strength presents a somewhat different problem . in order to reduce the contribution of this line from inner , we assumed a matter bounded model for this component . to provide a better match to the observation of little or no mg ii emission , we can reduce the model contribution by modifying outer . the mg@xmath47 emissivity is greatest near the h@xmath47/h@xmath48 transition zone in outer , so a simple truncation at much lower optical depths is not feasible , as it would have a much greater effect on the other line ratios . an obvious explanation for the weak observed mg ii is depletion of the magnesium into dust grains , along with suppression of the resonance photons by multiple scatterings and eventual absorption by dust . this was suggested in paper i ( cf . kraemer & harrington 1986 ; ferland 1992 ) . for comparison with our standard model , we generated a version of outer that includes dust , assuming a dust to gas ratio that is 30% of that found in the galactic interstellar medium , with equal amounts of graphite and silicate grains and accompanying depletions . these assumptions were made to avoid biasing our results by simply having all of the mg depleted into dust grains . we assumed relative element depletions as calculated by seab & shull ( 1983 ) , and the grain size distributions determined by mathis et al . ( 1977 ) and draine & lee ( 1984 ) ; details of the treatment of dust in the code are given by kraemer ( 1985 ) . the results of the model are given in table 2 , and not only show a significant drop in the relative strength of mg ii @xmath02800 , but also a drop in the ly@xmath14 strength , as is expected due to the preferential dust absorption of multiply scattered uv resonance lines . the lower ly@xmath14/h@xmath5 ratio is a better fit to the observations . a substantially larger dust - to - gas ratio than we assumed would result in a ly@xmath14/h@xmath5 ratio that is lower than observed . therefore , it is likely that there is some dust mixed in with the low - ionization gas , although with a lower dust to gas ratio than found in the ism , and that depletion coupled with the resonance line suppression explain the weak mg ii . second , we address discrepancies in the high ionization lines . specifically , the lines of n v , [ ne v ] , [ fe vii ] , and [ fe x ] are too weak by factors of 2 to 4 compared to the observations . as we mentioned earlier , the density is well constrained , so increasing the ionization parameter brings the gas well within 1 pc , into the realm of the blr . however , these lines are relatively narrow ( fwhm @xmath34 500 km s@xmath9 , see moore et al . 1996 ) and they are not likely to arise very close to the blr . the strengths of the high ionization lines can also be enhanced relative to h@xmath5 by truncating the integration of inner at a lower optical depth . however , this has the problem of enhancing the he ii emission relative to h@xmath5 in the model . more likely solutions to the problem of underpredicting the high ionization lines include 1 ) shock ionization , 2 ) a large `` blue bump '' in the euv continuum , or 3 ) supersolar abundances . predicting the strengths of the coronal lines has always been a problem with simple photoionization models , as viegas - aldrovandi & contini ( 1989 ) discuss in some detail for the fe lines . they suggest that there may be shocked gas mixed in with the photoionized clouds and that these high ionization lines may arise there . although this is certainly a possible factor , there may be other plausible explanations which avoid adding another level of complexity . an obvious way in which the coronal lines might be enhanced is if there were a component of ionizing radiation that contributed significantly at energies between 100 and 500 ev . although there has been some speculation about the presence of a `` blue bump '' in the euv , recent work by zheng et al . ( 1997 ) on low - redshift quasars shows that the _ near _ euv continuum is likely to be much steeper than previously supposed . in ngc 5548 , the _ euve _ continuum point in figure 2 is further confirmation that a large blue bump is not present in the spectrum of ngc 5548 . another possibility may be diffuse radiation from the intercloud medium . tran ( 1995 ) has shown that there may be a contribution to the continuum radiation in some seyfert 2 galaxies from thermal emission from the intercloud medium responsible for the scattering of the hidden blr emission into the observer s line of sight . if a similar medium with temperatures @xmath3 5 x 10@xmath41 k exists in the nlr of ngc 5548 , it is possible that free - free radiation and line emission that arise within it may contribute to the ionization of the inner narrow - line gas . although this component would be weak compared to the continuum radiation emitted by the central source , it could have a significant local contribution to the ionization balance of clouds existing within this inner region . however , recent observations suggest that the extended uv continuum seen in some seyferts may be due to starbursts ( heckman et al . 1997 ) , which would not contribute to the high ionization lines . in studies of medium redshift qsos , ferland et al . ( 1996 ) found evidence of supersolar abundances . there is no direct evidence of elemental enhancements in seyfert galaxies , but it is certainly not implausible , particularly near the nucleus , where the most intense activity occurs . as oliva ( 1996 ) points out , the coronal line emission will be enhanced proportionally to the abundance of the atomic species . as a comparison , we ran a version of inner with a heavy element abundance that is twice solar , and the results are shown in table 3 . in particular , the relative strengths of the [ fe vii ] and [ ne v ] lines have increased , while many lines , such as c iv @xmath01549 and c iii ] @xmath01909 , show little change . it is possible , then , that the observed strength of some of these lines is in part due to enhanced abundances . note that we have not attempted to adjust the increase in abundances to fit assumptions about the type of star formation that might be expected . from our standard model , we can estimate several global properties of the nlr in ngc 5548 , including the covering factor and physical size , in addition to more local properties , including optical depth and presence of dust . we determine the covering factor of the nlr gas from the observed and model values of the `` conversion efficiency '' , @xmath49 , which is the ratio of h@xmath5 photons to ionizing photons . the covering factor is given by c @xmath4 @xmath49(observed)/@xmath49(model ) . a value of c @xmath40 1 would indicate that the ionizing radiation is anisotropic , which we would not expect to be the case for a seyfert 1 galaxy , since the central source is seen directly in such objects . assuming ho= 75 km s@xmath9 mpc@xmath9 , the observed h@xmath5 flux corresponds to a luminosity of 3.8 x 10@xmath50 ergs s@xmath9 , or 9.3 x 10@xmath51 h@xmath5 photons s@xmath9 . from the continuum observations described in section 2.2 , we calculate a total luminosity of ionizing photons of 1.09 x 10@xmath52 s@xmath9 . this yields an observed @xmath49 = 0.009 . our fit to the observed emission line spectrum assumed that each component in our model contributed 50% of the h@xmath5 emission . the resulting values of @xmath49 were 0.06 for inner and 0.11 for outer , and the covering factors are 0.07 and 0.04 , respectively , so c(nlr ) = 0.11 . the value is small compared to those found for seyfert 2 galaxies , which are often @xmath40 1 ( kinney et al . 1991 ; kraemer et al . 1994 ) , and there is no evidence for anisotropic radiation . note that this estimate of covering factor does not include the blr clouds , which contribute at least 50% of the flux in many of the strong lines , even when ngc 5548 is in its lowest state ( paper i ) . given the ionization parameters and densities of the two components from our standard model , as well as the ionizing luminosity , the characteristic sizes ( i.e. , radii ) for the two emitting regions are 1 pc for inner , and 70 pc for outer . ( using the higher continuum luminosity given by the _ euve _ point in figure 2 would increase these values by a factor of only @xmath53 , or 1.14 . ) thus , the nlr of ngc 5548 is _ physically _ compact , since the size of outer is much smaller than typical values of 200 1000 pc determined for seyfert 2 galaxies , using the same methods that we have described in this paper ( kraemer & harrington 1986 ; kraemer et al . although there have been reports of extended emission from ngc 5548 ( wilson et al . 1989 ) , the contribution to the integrated emission line spectrum from this region is small ( peterson et al . 1995 ) ; we estimated the contribution to the narrow uv lines outside of a 1@xmath13 aperture ( 330 pc for h@xmath22 @xmath4 75 km s@xmath9 mpc@xmath9 ) to be only @xmath120% . this is consistent with our finding that the majority of the narrow emission must arise in a region with a `` diameter '' of 140 pc . pogge ( 1989 ) and schmitt & kinney ( 1996 ) claim that the _ apparent _ size of the nlr in a seyfert 1 galaxy is typically much smaller than that of a seyfert 2 galaxy , although this may be due to a selection effect , since most of the seyfert 2s in the it hst archive were selected on the basis of their extended emission ( wilson 1997 ) . nevertheless , it is clear that the majority of the seyfert 1 galaxies in these studies are apparently compact . this can not be explained by viewing angle alone , since the opening angles of the presumed ionization cones are large , and in the simplest version of the unified model , the apparent extent of most seyfert galaxies should be much larger , even if viewed `` pole - on '' ( schmitt & kinney 1996 ) . our results show that a possible explanation for the small apparent size of the nlrs in many seyfert 1 galaxies is that they are truly ( i.e. , physically ) compact . schmitt & kinney explain this phenomenon as a result of the orientation of the obscuring torus with respect to the plane of the galaxy . however , their model does not explain the dominance of the high ionization lines in ngc 5548 and many other seyfert 1 galaxies , a feature not generally seen in seyfert 2 galaxies ( koski 1978 ; shuder & osterbrock 1981 ) . thus , if the unified model applies we might expect that the narrow emission line spectrum of this seyfert 1 galaxy would resemble that of seyfert 2 galaxies , and , further , we would expect to see a noticeable contribution to the spectrum from gas hundreds of parsecs from the central source . it is possible that the high ionization region in seyfert 2 galaxies is obscured by a torus , but this does not explain the absence of a low ionization component in ngc 5548 . not only do our models show that the narrow - line spectrum can be well fitted without such a component , but it is clear that the nlr of ngc 5548 is dominated by high ionization gas that must be located close to the central source . a few seyfert 1 galaxies do indeed show significant nlr emission at large distances from the central source . for example , _ hst _ observations of ngc 4151 ( evans et al . 1993 ; hutchings et al . 1997 ) reveal an array of knots and filaments out to a kiloparsec that are almost certainly ionized by the central continuum . it may be that some seyfert 1 galaxies , such as ngc 4151 , have extended emission line regions and narrow lines in their spectra that closely resemble seyfert 2 galaxies . cohen ( 1983 ) studied the optical narrow line spectra of a group of seyfert 1 galaxies and found some resemblance although , as a group , they appeared to be of somewhat higher ionization . certainly , some of the galaxies in his study have spectra that are indistinguishable from those of type 2 seyferts , but others , like ngc 5548 , appear to be dominated by high ionization lines , a condition that appears to be rare among seyfert 2 galaxies . it would be extremely interesting if one could determine if these high ionization objects are as compact as ngc 5548 appears to be . unfortunately , optical spectra alone are insufficient for detailed modeling of the nlr , and thus far , accurate measurements of the narrow - line strengths in the uv have only been obtained for ngc 5548 and ngc 4151 ( ferland & mushotzky 1982 ) . as we stated above , our finding that the inner component of gas is optically thin at the lyman limit is based both on the absence of strong mg ii emission and the relative strength of the high ionization lines . if this is indeed true , not only for ngc 5548 but for other seyfert 1 galaxies , it may be a clue to the origin of the inner narrow - line gas . thin filaments or knots of reasonably high density ( @xmath1 10@xmath35 @xmath36 ) could be the result of outflow from the blr , either as condensations in an expanding intercloud medium or as `` tails '' of blr clouds , driven out by radiation pressure . if so , the small physical depths of the clouds inferred from our modeling ( @xmath1 10@xmath54 cm ) may constrain the sizes of their blr progenitors . note that there has been some recent success in models of the blr using a component of optically thin clouds ( shields , ferland , & peterson 1995 ) . there have been other studies suggesting that the clouds in the nlr are matter bounded or that some mix of radiation bounded and matter bounded clouds exist ( wilson et al . 1997 ; viegas - aldrovandi 1988 ) . such a mix might give rise to the filamentary structure seen in some [ o iii ] images of seyfert galaxies rather than the typical molecular clouds found in spiral galaxies , and may give a clue about the origin of the nlr gas . one other point regarding the nlr gas is that it appears that there may be some dust present within the outer components of the emitting region . dust is certainly able to exist in clouds at this proximity to the central source , with dust temperatures reaching a few hundred k ( kraemer & harrington 1986 ) . nevertheless , the history of the dust in this gas is unknown . our finding that the outer clouds may not be purely radiation bounded would indicate that they are not simply interstellar molecular clouds that have an outer shell ionized by the central source . our finding that there is probably dust mixed in with this gas will be important in determining the origin of this component . finally , moore et al . ( 1996 ) find a correlation of ionization potential of the narrow emission lines with velocity width in ngc 5548 . combined with our finding that , to first order , the ionization level decreases with distance , this indicates that the radial velocities decrease with increasing distance . this trend is also seen in a more direct fashion in spatially resolved spectra of the inner nlr in ngc 4151 ( hutchings et al . we have analyzed uv and optical spectra of the seyfert 1 galaxy ngc 5548 that were obtained when it was at an historical minimum . we were able to isolate the narrow emission line components ( which do not vary in flux on short time scales ) , due to the relative weakness of the more rapidly variable broad emission lines that are usually blended with the narrow components . we have constructed photoionization models of the narrow - line region of this galaxy , and are able to successfully match the observed dereddened ratios of a large number of emission lines to within the errors , with the exceptions noted in section 3.3 . since we used the best direct observational evidence of the shape of the ionizing continuum , rather than making adjustments based on fitting emission line ratios , the quality of this fit is particularly satisfying . the fact that a good fit was obtained for the permitted emission lines , such as c iv @xmath01549 , and the forbidden lines , such as [ ne iii ] @xmath03869 , [ o iii ] @xmath05007 , [ o ii ] @xmath03727 , [ n ii ] @xmath06584 , etc . , indicates that the range of physical conditions assumed in these models is approximately correct . from our analysis and modeling of these spectra , we can make several statements regarding the physical conditions in the nlr of ngc 5548 . first , it is clear that the principal source of ionization in the nlr of ngc 5548 is the central continuum source . this conclusion is borne out by the quality of the fit to the emission line spectrum . the nlr covering factor is reasonably small ( c @xmath4 0.11 ) , so the nlr gas does not need to intercept much of the ionizing continuum to produce the observed emission lines fluxes . a second conclusion is that the highly ionized gas in the inner part of the nlr appears to be optically thin at the lyman limit ( @xmath2 @xmath3 2.5 ) , which yields constraints on the physical depths of these clouds and may provide a clue to their origin . we have also presented evidence for supersolar abundances in the inner portion of the nlr , and dust in the outer portion . the most important conclusion that we have reached in this study is that the nlr of ngc 5548 is physically compact , with the majority of emission coming from a distance @xmath34 70 pc from the nucleus . by contrast , the unified model of seyfert galaxies suggests that the _ physical _ dimensions of the nlr in seyfert 1 and 2 galaxies should be similar . additional studies of the type that we have presented in this paper , particularly of other seyfert 1 galaxies , are important for testing this aspect of the unified model . we thank the referee , andrew wilson , for helpful comments and suggestions . we are grateful to fred bruhweiler and pat harrington for helpful discussions on the physical properties of iron and the availability of atomic data . s.b.k . and d.m.c . acknowledge support from nasa grant nag 5 - 4103 . a.v.f acknowledges support from nasa grant nag 5 - 3556 , and b.m.p . acknowledges support from nsf grant ast-9420080 . oliva , e. 1997 , in emission lines in active galaxies : new methods and techniques , ed . peterson , f .- z . cheng , & a.s . wilson ( san francisco : astronomical society of the pacific ) , asp conference series , 113 , 288 wilson , a.s . 1997 , in emission lines in active galaxies : new methods and techniques , ed . peterson , f .- z . cheng , & a.s . wilson ( san francisco : astronomical society of the pacific ) , asp conference series , 113 , 264 lrrr ly@xmath14 @xmath01216 & 14.60 & 22.01 & ( @xmath167.53 ) + n v @xmath01240 & 1.82 & 2.70 & ( @xmath160.95 ) + o i @xmath01302 & 0.81 & 1.15 & ( @xmath160.60 ) + si iv / o iv ] @xmath01400 & 2.10 & 2.88 & ( @xmath160.99 ) + n iv ] @xmath01486 & 0.30 & 0.40 & ( @xmath160.15 ) + c iv @xmath01550 & 10.73 & 14.22 & ( @xmath164.10 ) + he ii @xmath01640 & 1.22 & 1.60 & ( @xmath160.50 ) + o iii ] @xmath01663 & 0.63 & 0.82 & ( @xmath160.26 ) + si iii ] @xmath01892 & 0.24 & 0.32 & ( @xmath160.13 ) + c iii ] @xmath01909 & 2.18 & 2.91 & ( @xmath160.83 ) + o iii]/c ii]@xmath02323 & 0.37 & 0.51 & ( @xmath160.23 ) + @xmath55ne iv ] @xmath02423 & 0.24 & 0.31 & ( @xmath160.15 ) + @xmath55o ii ] @xmath02470 & 0.21 & 0.27 & ( @xmath160.11 ) + mg ii @xmath02800 & @xmath440.15 & @xmath440.18 & + o iii @xmath03133 & 0.45 & 0.50 & ( @xmath160.18 ) + he ii @xmath03204 & 0.15 & 0.17 & ( @xmath160.07 ) + @xmath55ne v ] @xmath03346 & 0.57 & 0.62 & ( @xmath160.13 ) + @xmath55ne v ] @xmath03426 & 1.70 & 1.84 & ( @xmath160.30 ) + @xmath55fe vii ] @xmath03588 & 0.21 & 0.23 & ( @xmath160.07 ) + @xmath55o ii ] @xmath03727 & 0.79 & 0.84 & ( @xmath160.14 ) + @xmath55fe vii ] @xmath03760 & 0.45 & 0.48 & ( @xmath160.08 ) + @xmath55ne iii ] @xmath03869 & 1.25 & 1.32 & ( @xmath160.18 ) + h@xmath56 + he i @xmath03889 & 0.27 & 0.28 & ( @xmath160.18 ) + @xmath55ne iii ] + h@xmath57 @xmath03967 & 0.42 & 0.44 & ( @xmath160.07 ) + @xmath55s ii ] @xmath04072 & 0.12 & 0.13 & ( @xmath160.03 ) + h@xmath58 @xmath04102 & 0.21 & 0.22 & ( @xmath160.05 ) + h@xmath59 @xmath04340 & 0.45 & 0.46 & ( @xmath160.09 ) + @xmath55o iii ] @xmath04363 & 0.75 & 0.77 & ( @xmath160.11 ) + he ii @xmath04686 & 0.22 & 0.23 & ( @xmath160.05 ) + h@xmath5 @xmath04861 & 1.00 & 1.00 & + @xmath55o iii ] @xmath04959 & 2.61 & 2.60 & ( @xmath160.36 ) + @xmath55o iii ] @xmath05007 & 8.13 & 8.07 & ( @xmath160.88 ) + @xmath55fe vii ] @xmath05721 & 0.21 & 0.20 & ( @xmath160.05 ) + he i @xmath05876 & 0.22 & 0.21 & ( @xmath160.05 ) + @xmath55fe vii ] @xmath06087 & 0.48 & 0.45 & ( @xmath160.08 ) + @xmath55o i ] @xmath06300 & 0.33 & 0.31 & ( @xmath160.07 ) + @xmath55o i ] @xmath06364 & 0.11 & 0.10 & ( @xmath160.03 ) + @xmath55fe x ] @xmath06374 & 0.18 & 0.17 & ( @xmath160.06 ) + @xmath55n ii ] @xmath06548 & 0.27 & 0.25 & ( @xmath160.06 ) + h@xmath14 @xmath06563 & 3.30 & 3.06 & ( @xmath160.44 ) + @xmath55n ii ] @xmath06583 & 0.82 & 0.76 & ( @xmath160.15 ) + @xmath55s ii ] @xmath06716 & 0.36 & 0.33 & ( @xmath160.07 ) + @xmath55s ii ] @xmath06730 & 0.36 & 0.33 & ( @xmath160.07 ) + @xmath55o ii ] @xmath07325 & 0.21 & 0.19 & ( @xmath160.06 ) + @xmath55s iii ] @xmath09069 & 0.45 & 0.39 & ( @xmath160.09 ) + @xmath55s iii ] @xmath09532 & 0.79 & 0.68 & ( @xmath160.13 ) + lrrrr c iii @xmath0977 & 1.16 & 0.70 & 0.03 & 0.03 + n iii @xmath0990 & 0.16 & 0.11 & 0.00 & 0.00 + o vi @xmath01036 & 1.43 & 1.12 & 0.00 & 0.00 + si iii @xmath01206 & 0.05 & 0.05 & 0.02 & 0.02 + o v @xmath01216 & 1.94 & 1.64 & 0.00 & 0.00 + ly@xmath14 @xmath01216 & 36.24 & 35.59 & 38.02 & 20.26 + n v @xmath01240 & 2.02 & 1.86 & 0.00 & 0.00 + c ii @xmath01334 & 0.05 & 0.12 & 0.13 & 0.06 + si iv @xmath01398 & 0.52 & 0.54 & 0.08 & 0.06 + o iv ] @xmath01402 & 2.67 & 2.52 & 0.04 & 0.05 + s iv ] @xmath01417 & 0.14 & 0.13 & 0.01 & 0.01 + n iv ] @xmath01486 & 1.91 & 1.69 & 0.03 & 0.03 + c iv @xmath01550 & 30.87 & 30.08 & 0.66 & 0.45 + @xmath55ne v ] @xmath01575 & 0.04 & 0.05 & 0.00 & 0.00 + @xmath55ne iv ] @xmath01602 & 0.31 & 0.31 & 0.01 & 0.02 + he ii @xmath01640 & 2.64 & 2.39 & 1.57 & 1.71 + o iii ] @xmath01663 & 2.86 & 2.73 & 0.26 & 0.30 + n iii ] @xmath01750 & 0.92 & 0.99 & 0.16 & 0.19 + @xmath55ne iii ] @xmath01815 & 0.03 & 0.03 & 0.00 & 0.00 + si iii ] @xmath01883 & 0.00 & 0.00 & 0.05 & 0.05 + si iii ] @xmath01892 & 0.31 & 0.42 & 0.20 & 0.20 + c iii ] @xmath01909 & 8.08 & 8.78 & 2.19 & 1.98 + @xmath55o iii ] @xmath02321 & 0.51 & 0.59 & 0.04 & 0.05 + c ii ] @xmath02326 & 0.07 & 0.10 & 1.37 & 1.09 + @xmath55ne iv ] @xmath02423 & 0.03 & 0.03 & 0.26 & 0.29 + @xmath55o ii ] @xmath02470 & 0.01 & 0.02 & 0.29 & 0.31 + @xmath55 mg v ] @xmath02784 & 0.14 & 0.19 & 0.02 & 0.01 + mg ii @xmath02800 & 0.11 & 0.18 & 2.12 & 0.81 + @xmath55 mg v ] @xmath02929 & 0.04 & 0.05 & 0.00 & 0.00 + @xmath55ne v ] @xmath02974 & 0.02 & 0.02 & 0.00 & 0.00 + he ii @xmath03204 & 0.15 & 0.14 & 0.10 & 0.11 + @xmath55ne iii ] @xmath03342 & 0.02 & 0.02 & 0.00 & 0.00 + @xmath55ne v ] @xmath03346 & 0.29 & 0.41 & 0.04 & 0.04 + @xmath55ne v ] @xmath03426 & 0.79 & 1.11 & 0.10 & 0.10 + @xmath55n i ] @xmath03467 & 0.00 & 0.00 & 0.01 & 0.01 + @xmath55fe vii ] @xmath03588 & 0.11 & 0.16 & 0.01 & 0.00 + @xmath55s iii ] @xmath03722 & 0.01 & 0.01 & 0.04 & 0.04 + @xmath55o ii ] @xmath03727 & 0.00 & 0.00 & 1.61 & 1.70 + @xmath55fe vii ] @xmath03760 & 0.15 & 0.22 & 0.01 & 0.01 + @xmath55s iii ] @xmath03796 & 0.00 & 0.00 & 0.00 & 0.00 + @xmath55ne iii ] @xmath03869 & 1.11 & 1.67 & 1.28 & 1.42 + @xmath55ne iii ] @xmath03967 & 0.35 & 0.51 & 0.39 & 0.44 + @xmath55s ii ] @xmath04072 & 0.00 & 0.00 & 0.16 & 0.17 + h@xmath58 @xmath04100 & 0.26 & 0.26 & 0.26 & 0.26 + h@xmath59 @xmath04340 & 0.47 & 0.47 & 0.47 & 0.47 + @xmath55o iii ] @xmath04363 & 2.24 & 2.57 & 0.18 & 0.20 + he i @xmath04471 & 0.03 & 0.03 & 0.04 & 0.04 + mg i ] @xmath04571 & 0.00 & 0.00 & 0.03 & 0.03 + he ii @xmath04686 & 0.36 & 0.33 & 0.23 & 0.25 + @xmath55ne iv ] @xmath04720 & 0.07 & 0.08 & 0.00 & 0.00 + h@xmath5 & 1.00 & 1.00 & 1.00 & 1.00 + @xmath55o iii ] @xmath05007 & 3.02 & 4.43 & 17.63 & 18.05 + @xmath55n i ] @xmath05198 & 0.00 & 0.00 & 0.29 & 0.30 + @xmath55n i ] @xmath05200 & 0.00 & 0.00 & 0.23 & 0.23 + he ii @xmath05412 & 0.03 & 0.03 & 0.02 & 0.02 + @xmath55o i ] @xmath05577 & 0.00 & 0.00 & 0.02 & 0.02 + @xmath55fe vii ] @xmath05721 & 0.15 & 0.23 & 0.02 & 0.02 + @xmath55n ii ] @xmath05755 & 0.00 & 0.00 & 0.05 & 0.05 + he i @xmath05876 & 0.08 & 0.09 & 0.11 & 0.11 + @xmath55fe vii ] @xmath06087 & 0.22 & 0.34 & 0.03 & 0.03 + @xmath55o i ] @xmath06300 & 0.00 & 0.00 & 1.51 & 1.51 + @xmath55s iii ] @xmath06312 & 0.01 & 0.02 & 0.06 & 0.07 + @xmath55o i ] @xmath06364 & 0.00 & 0.00 & 0.50 & 0.50 + @xmath55fe x ] @xmath06374 & 0.13 & 0.13 & 0.00 & 0.00 + @xmath55n ii ] @xmath06548 & 0.00 & 0.00 & 0.62 & 0.68 + h@xmath14 @xmath06563 & 2.80 & 2.80 & 2.97 & 3.00 + @xmath55n ii ] @xmath06584 & 0.00 & 0.00 & 1.81 & 2.01 + @xmath55s ii ] @xmath156716 , 6731 & 0.00 & 0.00 & 1.19 & 1.18 + @xmath55o ii ] @xmath07325 & 0.01 & 0.02 & 0.38 & 0.43 + @xmath55s iii ] @xmath09069 & 0.00 & 0.01 & 1.21 & 1.33 + @xmath55s iii ] @xmath09532 & 0.01 & 0.02 & 2.95 & 3.24 + @xmath55s ii ] @xmath010,300 & 0.00 & 0.00 & 0.11 & 0.11 + @xmath55n i ] @xmath010,395 & 0.00 & 0.00 & 0.04 & 0.05 + @xmath55n i ] @xmath010,404 & 0.00 & 0.00 & 0.03 & 0.03 + lrr ly@xmath14 @xmath01216 & 37.10 & 22.01 ( @xmath167.53 ) + n v @xmath01240 & 1.00 & 2.70 ( @xmath160.95 ) + o i @xmath01302 & & 1.15 ( @xmath160.60 ) + si iv / o iv ] @xmath01400 & 1.65 & 2.88 ( @xmath160.99 ) + n iv ] @xmath01486 & 0.97 & 0.40 ( @xmath160.15 ) + c iv @xmath01550 & 15.80 & 14.22 ( @xmath164.10 ) + he ii @xmath01640 & 2.11 & 1.60 ( @xmath160.50 ) + o iii ] @xmath01663 & 1.56 & 0.82 ( @xmath160.26 ) + si iii ] @xmath01892 & 0.26 & 0.32 ( @xmath160.13 ) + c iii ] @xmath01909 & 5.14 & 2.91 ( @xmath160.83 ) + o iii]/c ii]@xmath02323 & 1.00 & 0.51 ( @xmath160.23 ) + @xmath55ne iv ] @xmath02423 & 0.13 & 0.31 ( @xmath160.15 ) + @xmath55o ii ] @xmath02470 & 0.15 & 0.27 ( @xmath160.11 ) + mg ii @xmath02800 & 1.12 & @xmath440.18 + o iii @xmath03133 & & 0.50 ( @xmath160.18 ) + he ii @xmath03204 & 0.12 & 0.17 ( @xmath160.07 ) + @xmath55ne v ] @xmath03346 & 0.16 & 0.62 ( @xmath160.13 ) + @xmath55ne v ] @xmath03426 & 0.44 & 1.84 ( @xmath160.30 ) + @xmath55fe vii ] @xmath03588 & 0.06 & 0.23 ( @xmath160.07 ) + @xmath55o ii ] @xmath03727 & 0.81 & 0.84 ( @xmath160.14 ) + @xmath55fe vii ] @xmath03760 & 0.08 & 0.48 ( @xmath160.08 ) + @xmath55ne iii ] @xmath03869 & 1.20 & 1.32 ( @xmath160.18 ) + h@xmath56 + he i @xmath03889 & & 0.28 ( @xmath160.18 ) + @xmath55ne iii ] + h@xmath57 @xmath03967 & 0.37 & 0.44 ( @xmath160.07 ) + @xmath55s ii ] @xmath04072 & 0.08 & 0.13 ( @xmath160.03 ) + h@xmath58 @xmath04102 & 0.26 & 0.22 ( @xmath160.05 ) + h@xmath59 @xmath04340 & 0.47 & 0.46 ( @xmath160.09 ) + @xmath55o iii ] @xmath04363 & 1.21 & 0.77 ( @xmath160.11 ) + he ii @xmath04686 & 0.29 & 0.23 ( @xmath160.05 ) + h@xmath5 @xmath04861 & 1.00 & 1.00 + @xmath55o iii ] @xmath04959 & 3.44 & 2.60 ( @xmath160.36 ) + @xmath55o iii ] @xmath05007 & 10.32 & 8.07 ( @xmath160.88 ) + @xmath55fe vii ] @xmath05721 & 0.08 & 0.20 ( @xmath160.05 ) + he i @xmath05876 & 0.09 & 0.21 ( @xmath160.05 ) + @xmath55fe vii ] @xmath06087 & 0.13 & 0.45 ( @xmath160.08 ) + @xmath55o i ] @xmath06300 & 0.76 & 0.31 ( @xmath160.07 ) + @xmath55o i ] @xmath06364 & 0.25 & 0.10 ( @xmath160.03 ) + @xmath55fe x ] @xmath06374 & 0.07 & 0.17 ( @xmath160.06 ) + @xmath55n ii ] @xmath06548 & 0.31 & 0.25 ( @xmath160.06 ) + h@xmath14 @xmath06563 & 2.89 & 3.06 ( @xmath160.44 ) + @xmath55n ii ] @xmath06583 & 0.91 & 0.76 ( @xmath160.15 ) + @xmath55s ii ] @xmath156716 , 6730 & 0.59 & 0.66 ( @xmath160.10 ) + @xmath55o ii ] @xmath07325 & 0.20 & 0.19 ( @xmath160.06 ) + @xmath55s iii ] @xmath09069 & 0.61 & 0.39 ( @xmath160.09 ) + @xmath55s iii ] @xmath09532 & 1.47 & 0.68 ( @xmath160.13 ) +

we propose a computational scheme for the _ ab initio _ calculation of wannier functions ( wfs ) for correlated electronic materials . the full - orbital hamiltonian is projected into the wf subspace defined by the physically most relevant partially filled bands . the hamiltonian obtained in this way , with interaction parameters calculated by constrained lda for the wannier orbitals , is used as an _ ab initio _ setup of the correlation problem , which can then be solved by many - body techniques , e.g. , dynamical mean - field theory ( dmft ) . in such calculations the self - energy operator is defined in wf basis which then can be converted back into the full - orbital hilbert space to compute the full - orbital interacting green function . using one can evaluate the charge density , modified by correlations , together with a new set of wfs , thus defining a fully self - consistent scheme . the green function can also be used for the calculation of spectral , magnetic and electronic properties of the system . here we report the results obtained with this method for srvo and vo . comparisons are made with previous results obtained by the lda+dmft approach where the lda dos was used as input , and with new bulk - sensitive experimental spectra .

cond-mat0407359

model hamiltonians used in the study of correlation effects in solids have a coulomb interaction term in a site - centered atomic - like orbital basis set which is not explicitly defined . when the correlated electrons are well localized , as , for example , 4@xmath6-states of rare - earth ions , atomic orbitals ( or atomic sphere solutions like muffin - tin orbitals ) are a good choice . however , the most interesting problems occur in the regime of metal - insulator transitions , where the states of interest become partially itinerant and rather extended . the error of using atomic orbitals is most severe in the case of materials with strong covalency effects , like late transition metal oxides , where partially filled bands are formed by the mixture of metallic @xmath7-orbitals and oxygen p - orbitals . for example , in high - t@xmath8 cuprates correlated states have the symmetry of cu-3@xmath7 @xmath9 orbitals , but are actually zhang - rice singlets formed by the combination of oxygen p - states centered around the cu ion and having @xmath9 symmetry . in model calculations the problem of defining the correlated orbitals is not very important , because it only affects model parameter values , which in any case are considered fitting parameters . however any attempt to construct an `` _ ab initio _ '' calculation scheme requires an explicit definition of the basis set for the coulomb interaction term . an important requirement for such a choice is that the orbitals must produce the partially filled bands where coulomb correlations occur while preserving the localized , site - centered atomic - like form . these requirements are fulfilled for wannier functions ( wfs ) @xmath10 defined as a fourier transformation of the bloch functions @xmath11 @xcite . here and below functions are labeled with band index n , lattice translation vector * t * and wave vector * k*. when there is more than one band crossing the fermi level , wfs are not uniquely defined . any * k*-dependent unitary transformation @xmath12 of the set of bloch functions @xmath11 for these bands produces a new set which can be used for the calculation of wfs via fourier transformation ( eq . [ psi_def ] , sec . [ w_def ] ) . if one imposes the requirement that the wfs should have the symmetry of atomic orbitals @xcite , this unitary transformation is well defined . the explicit form of the wfs allows one to compute coulomb interaction parameters in constrained local density approximation ( lda ) calculations . in this way the parameters for the _ ab initio _ many - body hamiltonian ( non - interacting hamiltonian @xmath1 and coulomb interaction ) in the wf basis can be computed by any first principle electronic structure calculation scheme ( below we use the lmto method ) . this hamiltonian can then be further investigated by one of the methods developed in the many - body community . in the present work we use the dynamical mean - field theory ( dmft ) @xcite . within dmft , the effective impurity problem corresponding to the many - body hamiltonian is solved by quantum monte - carlo simulations ( qmc ) @xcite . the dmft part of the proposed calculation scheme is essentially the same as the one used in the recently developed lda+dmft approach @xcite for the _ ab initio _ investigations of correlated electron materials @xcite . however , here we propose a more general procedure to compute the green function using the hamiltonian matrix and an integral over brillouin zone instead of the hilbert transform of the lda density of states ( dos ) . this particular method allows one to avoid the uncontrollable errors occuring in the computation of the green function using the hilbert transform of the lda dos . thus , to obtain an insulating solution we need to cut off the long ( metal - oxygen ) hybridization tails of the dos , renormalize it and shift the fermi energy to get an integer filling . in the present method we overcome the above - mentioned difficulties owing to the integer filling of wannier orbitals . the result of the dmft calculations is a local self - energy operator @xmath13 . in our scheme this operator is defined in the wf basis set @xmath14 and is acting in the subspace of partially filled bands which are used for the construction of the wfs . the paper is structured as follows . in sec . [ method ] the details of our scheme are presented . in sec . [ w_def ] we describe the construction of wfs , as well as the _ ab - initio _ hamiltonian matrix within this basis set in terms of bloch functions . in sec . [ wf_gf ] we propose a general method for the construction of wfs using the green function @xmath15 , which reduces to the results of sec . [ w_def ] in the non - interacting case . the reason for doing so is that the correlation effects can significantly renormalize the electronic states of the partially filled bands . hence the wfs computed from non - interacting bloch states are not an optimal choice for the basis set any more . in sec . [ wf_dmft ] we discuss how to calculate within dmft the local green function with the input of the hamiltonian matrix in wf basis set instead of the lda dos ( which is valid only in the case of degenerate bands ) . in sec . [ self - cons ] we show that the self - energy operator within the wf subspace , which is the solution of the correlation problem , can be transformed back into the full - orbital hilbert space , thus enabling the computation of the full interacting green function @xmath16 . it can then be used to calculate the spectral , magnetic and electronic properties of the system under investigation . in addition , to make the calculation scheme fully self - consistent , one can employ the @xmath15 to calculate the correlation affected charge density and thus the new lda potential . thereby the feedback from dmft to lda can be incorporated in a well - defined way . this is actually one of the great advantages of using the wf basis since in the lmto basis the feedback from dmft to lda is essentially uncontrolled . in sec . [ results ] the results for the electronic structure of the two vanadium oxides srvo@xmath4 and v@xmath5o@xmath4 obtained by the method developed in this work are presented and compared with the previous calculations by the simpler methods and new bulk - sensitive spectra . finally in sec . [ conclusion ] we close this work with a conclusion . let us consider the general case of the electronic structure problem . for the lda hamiltonian @xmath0 we have a hilbert space of eigenfunctions ( bloch states @xmath17 ) with the basis set @xmath18 defined by particular methods ( e.g. , lmto @xcite , or linearized augmented plane waves ( lapw ) @xcite , etc . ) . in this basis set the hamiltonian operator is defined as @xmath19 here and later greek indices are used for full - orbital matrices . if we consider a certain subset of the hamiltonian eigenfunctions , for example bloch states of partially filled bands @xmath11 , we can define a corresponding subspace in the total hilbert space . the hamiltonian matrix is diagonal in the bloch states basis . however , physically more appealing is a basis set which has the form of site - centered atomic orbitals . that is a set of wfs @xmath20 defined as the fourier transformation of a certain linear combination of bloch functions belonging to this subspace ( see below ( [ wf_psi ] ) ) . the hamiltonian operator @xmath1 defined in this basis set is @xmath21 the total hilbert space can be divided into a direct sum of the above introduced subspace ( of partially filled bloch states ) and the subspace formed by all other states orthogonal to it . those two subspaces are decoupled since they are the eigenfunctions corresponding to different eigenvalues . the hamiltonian matrix in the wf basis ( i.e. , a collection of the bases of the specific subspaces ) is block - diagonal so that the matrix elements between different subspaces are zero . the block matrix @xmath22 in ( [ ham_genwf ] ) corresponding to the partially filled bands can be considered as a projection of the full - orbital hamiltonian operator ( [ ham_gen ] ) onto the subspace defined by its wfs . all this concerns the non - interacting ( or lda ) hamiltonian . to treat coulomb correlations one also needs a definition of the localized orbitals where the electrons interact . wfs are a natural choice for such a definition . this choice leads to an important flexibility in the size of the basis set in the sense that the number of wfs can be changed by changing the set of bloch bands considered . the simplest case is a set of partially filled bands , for example the @xmath23-bands of vanadium oxides . this is a physically justified approximation because the coulomb interaction happens mainly between electrons ( or holes ) in the partially filled bands . if the problem to be solved concentrates on the excitation spectrum in a small energy window around the fermi level , this basis set is sufficient . however , if the excitations to higher lying states ( real or virtual ) are also important , the set of bloch bands used to construct the wfs need to be extended so that the coulomb interaction will be treated in a larger hilbert subspace . practically this means that the correlation problem is solved using a non - interacting few - orbital hamiltonian @xmath1 ( [ ham_genwf ] ) instead of the full hilbert space hamiltonian @xmath0 ( [ ham_gen ] ) . the interaction matrix elements of the model hamiltonian can be determined from constrained lda calculations for the specific wf basis set ( [ u_wf ] ) . projecting the full orbital hilbert space hamiltonian @xmath0 ( [ ham_gen ] ) onto the subspace of the partially filled bands gives us a few - orbital hamiltonian @xmath1 ( [ ham_genwf ] ) . this significantly decreases the complexity of the correlation problem , thus permitting its explicit solution . the many - body problem with a hubbard interaction then leads to a local self - energy operator @xmath24 which is naturaly defined in the basis of wfs centered on the same site : @xmath25-dependent projection matrix between the full orbital basis set and the orthonormalized wfs ( [ coef_wf_bloch ] ) . the definition ( [ sigma_gen ] ) allows one to convert the self - energy matrix @xmath26 back to the full hilbert space basis set ( subsection [ self - cons ] ) . with this the interacting green function can also be calculated in the full - orbital hilbert space ( subsection [ wf_gf ] ) . the concept of wfs has a very important place in the electron theory in solids since its first introduction in 1937 by wannier @xcite . wfs are the fourier transformation of bloch states @xmath17 @xmath27 where @xmath28 is the number of discrete @xmath29 points in the first brillouin zone ( or , the number of cells in the crystal ) . these extremely convenient orthogonal functions were widely investigated in the seventies @xcite . the strongly localized nature of the wfs together with all advantages of the atomic functions makes them a very useful tool where the atomic character of the electrons is highlighted . thus , using the wf method , significant progress was achieved in the fields of narrow - band superconductors , disordered systems , solid surfaces , etc . several methods for calculating wf for single and multiple bands in periodic crystals , and their generalization to non - periodic systems were proposed . the problem of non - unique definition of wfs in these methods was resolved by an iterative optimization of trial functions which have the same real and point group symmetry properties as wfs . among these methods , there are the variational koster - parzen principle @xcite which was generalized by kohn @xcite , the general pseudopotential formalism proposed by anderson @xcite , and the projection operator formalism by cloizeaux @xcite . however , all these computational schemes are restricted to simple band structures . wannier functions are not uniquely defined because for a certain set of bands any orthogonal linear combination of bloch functions @xmath17 can be used in ( [ wf_psi_def ] ) . in general it means that the freedom of choice of wannier functions corresponds to freedom of choice of a unitary transformation matrix @xmath30 for corresponding bloch functions @xcite : @xmath31 the resulting bloch function @xmath32 will generally not be an eigenfunction of the hamiltonian but has the meaning of a bloch sum of wannier functions ( see below @xmath33 in ( [ wf_psi ] ) ) . there is no rigorous way to define @xmath30 . this calls for an additional restriction on the properties of wfs . among others marzari and vanderbilt @xcite proposed the condition of maximum localization for wfs , resulting in a variational procedure to calculate @xmath30 . to get a good initial guess the authors of @xcite proposed choosing a set of localized trial orbitals @xmath34 and projecting them onto the bloch functions @xmath17 . it was found that this starting guess is usually quite good . this fact later led to the simplified calculating scheme proposed in @xcite where the variational procedure was abandoned and the result of the aforementioned projection was considered as the final step . the approach of @xcite has recently been used for the investigation of the row of 3@xmath7 transition metals ( fe , co , ni , and cu ) within the simplest many - body approximation , namely the unscreened hartree - fock approximation @xcite . another possibility to construct wfs was recenly developed by andersen @xmath35 @xcite . they proposed the nth - order muffin - tin orbital ( nmto ) scheme in which wannier - like low - energy mtos can be designed _ a priori_. using a new implementation of the lda+dmft approach they performed an investigation of the mott transition in orthorhombic 3@xmath36 perovskites @xcite . in this approach a realistic hamiltonian constructed with wannier orbitals ( on symmetrically orthonormalized nmtos ) was solved by dmft , including the non - diagonal part of the on - site self - energy . our projection procedure works as follows . first of all one needs to identify the physically relevant bands which will then be projected onto a wf basis . for example , in perovskites one usually takes the partially filled @xmath7-shell or some particular @xmath7-bands of transition metals , since they are mainly responsible for the physical properties of the system @xcite . these orbitals are well - separated and are , in our approach , easily extracted from the full orbital space as will be shown later . moreover , the projection method is applicable even in the case where the bands of interest differ and are strongly hybridized ( for example , cu-3@xmath7 and o-2@xmath37 states in high - t@xmath8 superconductors @xcite ) . to project bands of particular symmetry onto the wfs basis one can select either the band indices of the corresponding bloch functions ( @xmath38 ) , or choose the energy interval ( @xmath39 ) in which the bands are located . non - orthogonalized wfs in reciprocal space @xmath33 are then the projection of the set of site - centered atomic - like trial orbitals @xmath34 on the bloch functions @xmath17 of the chosen bands ( band indices @xmath40 to @xmath41 , energy interval ( @xmath39 ) ) : @xmath42 then the real space wfs @xmath43 are given by @xmath44 in the present work the trial orbitals @xmath34 are lmtos . note that in the multi - band case a wf in reciprocal space @xmath33 does not coincide with the bloch function @xmath11 due to the summation over band index @xmath45 in ( [ wf_psi ] ) . one can consider them as bloch sums of wfs analogous to the basis function bloch sums @xmath46 ( [ psik ] ) . the coefficients @xmath47 in ( [ wf_psi ] ) define ( after orthonormalization ) the unitary transformation matrix @xmath30 in ( [ psi_def ] ) . however , the projection procedure defined in ( [ wf_psi ] ) is more general than the unitary transformation ( [ psi_def ] ) . namely , the number of bands @xmath48 can be larger than the number of trial functions . in this case the projection ( [ wf_psi ] ) will produce @xmath28 new functions @xmath49 which define a certain subspace of the original @xmath48-dimensional space . this subspace will have the symmetry of the set of trial functions . in the next subsection we propose a way to determine wfs from the green function of the system ( [ wf1 ] ) rather than from a set of bloch states as in ( [ wf_psi ] ) . in this alternative projection procedure , trial functions are projected onto the subspace defined by the green function in a certain energy interval . the bloch functions in lmto basis ( or any other atomic orbital - like basis set ) are defined as @xmath50 where @xmath51 is the combined index representing @xmath52 ( @xmath53 is the atomic number in the unit cell , @xmath54 are orbital and magnetic quantum numbers ) , @xmath55 are the bloch sums of the basis orbitals @xmath56 @xmath57 and the coefficients have the property @xmath58 if @xmath59 in @xmath34 corresponds to the particular @xmath52 combination ( in other words @xmath34 is an _ orthogonal _ lmto basis set orbital ) , then @xmath60 , and hence @xmath61 for a non - orthogonal basis set see appendix [ ortho ] . in order to orthonormalize the wfs ( [ wf ] ) one needs to calculate the overlap matrix @xmath62 @xmath63 and its inverse square root @xmath64 is defined as @xmath65 in the derivation of ( [ o - s ] ) the orthogonality of bloch states @xmath66 was used . from ( [ wf ] ) and ( [ ss ] ) , the orthonormalized wfs in @xmath67-space @xmath68 can be obtained as @xmath69 with @xmath70 @xmath71 the real space site - centered wfs at the origin @xmath72 are given by the fourier transform of @xmath68 with @xmath73 . from ( [ wf_orth ] ) and ( [ psik ] ) one finds @xmath74 where @xmath75 and @xmath76 are the expansion coefficients of wf in terms of the corresponding lmto orbitals , in particular , @xmath77 here @xmath78 is an index counting the orbitals of the neighboring cluster for the atom where orbital @xmath59 is centered ( @xmath79 is the corresponding translation vector , @xmath80 is a combined @xmath52 index ) . the explicit form of the real space wf ( [ real_wf ] ) can be used to produce , e.g. , shapes of chemical bonds . for other applications only the matrix elements of the various operators in the basis of wf([wf_orth ] ) are needed . from ( [ wf_orth ] ) , ( [ coef_wf_bloch ] ) and ( [ real_wf ] ) the matrix elements of the hamiltonian @xmath81 in the basis of wf in real space where both orbitals are in the same unit cell are @xmath82 @xmath83 is the eigenvalue for a particular band . if , on the other hand , one of the orbitals corresponds to the wf for the atom @xmath84 shifted from its position in the primary unit cell by a translation vector @xmath85 , then the corresponding hamiltonian matrix element is @xmath86 matrix elements of the density matrix operator ( occupation matrix @xmath87 ) in the basis of wfs can be calculated as @xmath88 @xmath89 is step function , @xmath90 is the fermi energy . finally , the matrix elements of the hamiltonian @xmath81 in reciprocal space are @xmath91 the ( [ e_wf_k ] ) is valid only if the wfs computed by eqs . ( [ wf_orth])-([coef_wf_lmto ] ) . if the wfs were obtained in one calculation and then used to compute the hamiltonian matrix in another ( as is the case for the wfs ( [ wf1 ] ) in the green functions formalism ( see subsection [ wf_gf ] ) ) then eq . ( [ e_wf_k ] ) is not valid any more and the general expression must be used : @xmath92 thus , the transformation from lmto to wf basis set is defined by the explicit form of wfs ( [ wf_orth],[coef_wf_lmto ] ) , and by the expressions for matrix elements of the hamiltonian and density matrix operators in wf basis ( [ e_wf],[q_wf ] ) . the back transformation from wf to lmto basis can also be defined using ( [ wf_orth ] ) ( see subsection [ self - cons ] ) . finally , the coulomb matrix element @xmath93 needs to be calculated in the same wf basis . this requires a method similar to constrained lda @xcite , but now for wfs . to this end the wf - energy ( [ e_wf ] ) is computed as a function of its occupancy ( [ q_wf ] ) for a given wf @xmath59 . then the corresponding coulomb interaction parameter @xmath94 in the wf basis is given by @xmath95 as one can see @xmath94 depends on the wfs via ( [ e_wf],[q_wf ] ) . once the wfs have been recalculated ( for example in some self - consistent loop ) the interaction has to be recalculated as well . in many - body theory the system is usually not described by bloch functions @xmath17 ( [ psi ] ) and their energies @xmath83 but by the green function @xmath96 the green function matrix @xmath97 is defined via the non - interacting hamiltonian matrix @xmath98 and the self - energy matrix @xmath99 ( [ sigma_full ] ) as @xmath100 we define non - orthonormalized wf obtained by projecting the trial orbital @xmath101 on the hilbert subspace defined by the green function ( [ green1 ] ) in the energy interval ( @xmath39 ) , namely , @xmath102 and @xmath103 in the non - interacting case , the self - energy operator @xmath104 is absent , and hence we have @xmath105 where @xmath106 are the eigenvectors ( [ coef ] ) , and @xmath83 are the eigenvalues of @xmath107 . thus @xmath108 in ( [ coef_green_b ] ) becomes @xmath109 where @xmath110 are the band numbers which correspond to the energy interval ( @xmath39 ) . since this recovers the result of ( [ coef_b ] ) , we demonstrated that our general definition of wfs ( [ wf1 ] ) via green function reduces to that in terms of bloch functions ( [ wf ] ) in sec . [ w_def ] . to orthonormalize @xmath111 defined in ( [ wf1 ] ) , one can just follow the orthonormalizing procedure made in sec . [ w_def ] ( [ o - s]- [ coef_wf_lmto ] ) , which will not be repeated here . but it should be pointed out that in the green function formalism the overlap matrix @xmath62 is defined as @xmath112 the occupancy matrix in the orthogonalized wf basis ( [ wf1 ] ) is defined as @xmath113 by using ( [ green1 ] ) and orthogonalized ( [ wf1 ] ) , one finds @xmath114 with @xmath115 the energy matrix can be defined similarly ( except that the integral over energy is calculated in the @xmath116 interval where the corresponding wfs are defined ) as : @xmath117 with @xmath118 while ( [ e_wf4 ] ) looks similar to the non - interacting hamiltonian in wfs basis ( [ e_wf2 ] ) , it includes correlations via @xmath119 in ( [ green2 ] ) and hence is _ interacting_. in the previous subsection we showed ( eqs . [ green1]-[wf1 ] ) that the self - energy operator is needed to construct the wfs in terms of the full interacting green function . the dmft @xcite was recently found to be a powerful tool to numerically solve multi - band hubbard models . to define parameters of the correlated model hamiltonian ( hoppings , screened coulomb integrals ) , density functional theory within the lda was used @xcite . the combined lda+dmft computational scheme was successfully applied to a wide range of compounds with degenerate ( or almost degenerate ) orbitals ( for more details see @xcite ) . in these cases the non - interacting lda dos was used to obtain the green function of the system through a hilbert transformation . furthermore , the screened coulomb interaction parameters u and j were calculated by constrained lda @xcite . quite generaly , this scheme needs to be improved in two respects : ( i ) instead of the lda dos an lda - hamiltonian with a few , relevant orbitals should be used to calculate the green function , and ( ii ) a feedback from dmft to lda should be incorporated . both of these problems are solved by the new approach proposed in this work . in this method the hamiltonian matrix in the wf basis set @xmath120 is calculated from the lda hamiltonian via the projection procedure ( [ wf_psi],[e_wf_k ] ) . in the dmft self - consistency loop the local green function @xmath121 is then computed as an integral over the first brillouin zone ( bz ) : @xmath122^{-1}\right)_{nn'}.\end{aligned}\ ] ] the integration can actually be restricted to the irreducible part of the bz via the analytical tetrahedron method @xcite with a subsequent symmetrization of the green function matrix . the chemical potential @xmath123 is determined by the number of electrons on the @xmath28 interacting orbitals of interest @xcite . the dmft is based on the fact that in the @xmath124 limit the self - energy operator is local @xcite . its matrix @xmath26 ( @xmath125 - wf indices ) is defined in wf basis ( [ sigma_gen ] ) . if the trial functions in ( [ wf_psi],[wf1 ] ) are chosen as the basis functions of the irreducible representation of the point symmetry group of some particular real system @xcite , the green function matrix ( [ green2 ] ) and hence the self - energy matrix ( [ sigma_gen ] ) can be made diagonal @xcite in the @xmath59 index for on - site matrix elements . to set up the dmft equations one needs to define the bath green function @xmath126 in the usual way via the dyson equation@xcite : @xmath127 to obtain the green function @xmath128 of the effective single impurity anderson problem , various methods can be used : quantum monte - carlo ( qmc ) , numerical renormalization group ( nrg ) , exact diagonalization ( ed ) , non - crossing approximation ( nca ) , etc . ( for a brief overview of the methods see @xcite ) . with the condition @xmath129 one closes the self - consistent loop which can then be iterated until a converged solution for the self - energy @xmath130 is found . the self - energy operator @xmath24 obtained as a solution of dmft in sec [ wf_dmft ] is defined in the wf basis set ( [ sigma_gen ] ) . in order to compute the interacting green function in the full - orbital hilbert space ( [ green1]-[green2 ] ) one has to convert it back to the full - orbital ( lmto ) basis set . this can be easily done by using the linear expansion form of the wfs in terms of the full - orbital basis set ( [ wf_orth],[coef_wf_lmto ] ) , @xmath131 here we use the local form of the self - energy operator as obtained in dmft , but the formalism can be easily generalized . in the following we refer to this self - energy operator as the `` full - orbital '' self - energy . the matrix elements of the self - energy operator @xmath132 ( [ sigma_full ] ) together with the non - interacting hamiltonian matrix @xmath133 allow one to calculate the green function matrix @xmath134 ( [ green2 ] ) and thus the full - orbital interacting green function @xmath135 ( [ green1 ] ) . @xmath135 contains the full information about the system , and various electronic , magnetic and spectral properties can be obtained from it . in sec . [ full - orbital ] we use the full - orbital interacting green function computed within dmft(qmc ) to calculate the photoemission and x - ray absorption spectra for the strongly correlated vanadium oxides srvo@xmath4 and v@xmath5o@xmath4 , and to compare them with new bulk - sensitive experimental spectra . one can also calculate the charge density distribution modified by correlation effects via @xmath136 with this @xmath137 one can recalculate the lda - potential ( which is a functional of electron density ) . from the full - orbital green function ( [ green1 ] ) one can recalculate new wfs ( [ wf1],[coef_green_b ] ) which together with the new lda hamiltonian allows one to obtain new parameters for the non - interacting hamiltonian ( [ e_wf3 ] ) . with ( [ u_wf ] ) one can then compute a new coulomb interaction parameter @xmath93 . the set of new lda potential , wfs and coulomb interaction parameters calculated from the interacting green function ( [ green1 ] ) defines the input for the next iteration step and hence closes the self - consistency loop in the proposed computation scheme . for the feedback from dmft to lda in the particular case of the lmto method @xcite one needs a set of moments for the partial densities of states @xmath138 for every atomic sphere @xmath53 and the orbital moment @xmath139 @xcite in order to calculate the new charge density and hence the new lda potential : @xmath140 for clarity , in fig . [ scheme ] the essential steps of the wf scheme presented here are summarized . there are four interconnected parts in this scheme : ( i ) the basis of wf , ( ii ) the matrix elements of the hamiltonian and the self - energy in wf basis , ( iii ) the coulomb interaction between electrons on the wfs , and ( iv ) the projection into the few - orbital basis and back transformation to the full - orbital basis which retains the information about all orbitals . first the matrix elements of the non - interacting hamiltonian in reciprocal space @xmath141 ( [ e_wf_k ] ) and the interaction term @xmath142 ( [ sigma_gen ] ) are written in the basis of explicitly defined wfs @xmath68 ( [ wf_orth ] ) . the actual correlation problem , defined by the sum of these two terms ( [ gloc ] ) , is then solved within the lda+dmft(qmc ) approach @xcite . the local self - energy @xmath142 obtained thereby is then transformed back from the wannier basis to the full - orbital space ( see subsection [ self - cons ] ) . furthermore , with the full - orbital self - energy @xmath143 ( [ sigma_full ] ) the full - orbital green function @xmath144 ( [ green1],[green2 ] ) for the correlated electrons is calculated by a @xmath29-integration over the brillouin zone @xcite . the green function allows one to determine a new charge density distribution ( [ rho ] ) ( in the lmto case see ( [ m - g ] ) ) and a new set of wfs ( via ( [ wf1],[coef_green_b ] ) ) . this is used to construct a new lda - potential and new non - interacting hamiltonian . together with the new coulomb interaction parameters @xmath94 ( [ u_wf ] ) they serve as the input for the next iteration of calculations , thus completing the self - consistency loop . it should be stressed that in this scheme self - consistency involves not only the self - energy but also the basis of wfs in which it is defined , the charge density and lda potential used for constructing the non - interacting hamiltonian , and the interaction strength between electrons in the wfs . after convergency is reached the maximum entropy method ( mem ) @xcite can be used to obtain spectral functions . then , using the kramers - kronig transform , the green function on real axis @xmath145 is computed . with that one can construct the wfs basis self - energy @xmath146 ( see appendix [ acse ] ) . by the back transformation of @xmath146 with a subsequent @xmath29-integration the full - orbital green function @xmath147 is obtained , which now contains information not only about the states for which correlations are considered , but also for all other orbitals of the system . @xmath147 is used to obtain orbitally resolved spectral functions . this allows one , for example , to investigate the influence of the correlated orbitals ( e.g. , partially filled @xmath23 orbitals of v in srvo@xmath4 and v@xmath5o@xmath4 ) on other orbitals ( oxygen @xmath148 and vanadium @xmath149 orbitals ) . it also makes possible the computation of spectral functions in a wide energy region and not only in the vicinity of the fermi level . in our earlier lda+dmft calculation scheme the specific properties of a material entered only via the lda partial densities of states ( doss ) for the orbitals of interest . this procedure is valid for systems where the bands of interest are degenerate ( as in cubic crystals ; see appendix [ hilbert ] ) . for more complicated systems with lower symmetry one needs to employ the scheme proposed in this work , where the non - interacting hamiltonian @xmath150 ( [ e_wf_k ] ) ( projected on wfs ) describing the @xmath28 orbitals under consideration is used for the calculation of @xmath151 ( [ gloc ] ) within dmft ( see section [ wf_dmft ] ) . in this section we present results of lda+dmft calculations using the projected hamiltonian @xmath150 . the scheme was applied to srvo@xmath4 which has a cubic perovskite crystal structure , and to the trigonally distorted v@xmath5o@xmath4 ( both in the insulating and metallic phase ) . the results are compared with previous lda+dmft calculations where the hilbert transformation of the lda doss was used for srvo@xmath4 @xcite ( see section [ comp ] ) . one should note that the doss used for dmft calculations of v@xmath5o@xmath4 were obtained by the tb - lmto - asa code v. 47 @xcite in contrast to @xcite where the doss were calculated by the asw method @xcite . however , the doss obtained in both methods are very similar and do not produce much different lda+dmft results . the full - orbital calculation scheme proposed in this work allows one to answer an important question : how do coulomb correlations between some orbitals affect the other orbitals , and in particular , how does the interaction between the partially filled @xmath23 orbitals of the v3@xmath7-shell influence the filled o-2@xmath37 and the unoccupied v-3@xmath7(@xmath152 ) bands ? to answer these questions the matrix of the self - energy operator @xmath26 ( [ sigma_gen ] ) was converted back to the full - orbital basis set from the wf basis , and the full orbital interacting green function ( [ green2 ] ) was calculated . > from that , total and partial doss were computed to produce theoretical spectra for comparison with the experimental photoemission and x - ray absorption data . in this work , qmc simulations @xcite were used to solve the effective single impurity anderson problem in the dmft loop . the result of the dmft(qmc ) calculation is the self - energy on imaginary energy axis @xmath153 . to find the full - orbital self - energy @xmath154 on the real energy axis one has to perform an analytical continuation . the procedure is described in appendix [ continuation ] . a self - consistent computation of the charge density as described in section [ self - cons ] was not yet performed . investigations of correlation effects on the unoccupied cubic e@xmath155 states for both srvo@xmath4 and v@xmath5o@xmath4 are also a matter of future calculations . ( t@xmath157 ) spectral functions for srvo@xmath4 calculated via lda+dmft(qmc ) using : lda dos - light line , projected hamiltonian - black line . fermi level corresponds to zero.,scaledwidth=50.0% ] the cubic perovskite srvo@xmath4 can serve as a test case for our new method . for the cubic @xmath158 point symmetry group , the three 3@xmath7-orbitals @xmath159 transform according to the triply degenerate @xmath23 irreducible representation of this group . hence the corresponding green function and self - energy matrices have diagonal form with equal diagonal elements . as shown in appendix [ hilbert ] , the results of the hilbert transformation using lda dos must coincide with the results of the procedure of integration over the brillouin zone with the projected 3x3 t@xmath157 hamiltonian . we use the same lda dos and interaction parameters as in our previous papers @xcite . the v-3@xmath7(t@xmath157 ) states form a partially filled band in srvo@xmath4 . all t@xmath157 orbitals ( @xmath160 , @xmath161 , @xmath162 ) are equivalent due to the cubic symmetry of the lattice , so only the results for one of the t@xmath157 orbitals are presented in fig . [ dos_hmlt_srvo3 ] . in this figure v-3@xmath7(t@xmath157 ) spectral functions , calculated using the lda dos and the projected hamiltonian are shown . it is easy to see that both results are almost identical . the small differences between these two curves are due to the mem @xcite used for the calculation of the spectral function on the real energy axis from the dmft green function . we also applied our hamiltonain scheme to a more complicated system with lower symmetry where one can expect deviations from the results obtained using the lda dos . we performed lda+dmft(qmc ) calculations for the insulating and metallic phases of v@xmath5o@xmath4 with the projected 3x3 t@xmath157 hamiltonian and several @xmath93 values ( the values of @xmath93 are the same as in @xcite ) . in contrast to srvo@xmath4 , v@xmath5o@xmath4 has a non - cubic trigonal symmetry ( space group r@xmath163c ) . therefore we use the basis functions of the trigonal @xmath164 point group instead of the cubic @xmath158 group . in this basis set the v-3@xmath7-shall is split into two groups of bands . the three t@xmath157 bands are located around the fermi energy , the two degenerate e@xmath165 bands are at higher energies . interesting to us are the partially filled t@xmath157 bands which are formed by one @xmath166 and two degenerate @xmath167 orbitals . in fig . [ dos_asw_hmlt_v2o3 ] , we present v-3@xmath7(t@xmath157 ) spectral functions resulting from dos ( lmto and asw ) and hamiltonian calculations at u@xmath168ev , averaged over the three t@xmath157 bands . the spectrum computed with asw input is taken from ref . . ( t@xmath157 ) spectral functions for v@xmath5o@xmath4 in the metallic phase calculated via lda+dmft(qmc ) using : lda dos ( lmto ) - light line @xcite , lda dos ( asw ) - dashed line , projected hamiltonian - black line . fermi level corresponds to zero.,scaledwidth=70.0% ] as mentioned before , the differences between the curves calculated with asw and lmto dos as input are small ; there are only minor deviations in the peak height . in the comparison between the dos calculations and the hamiltonian calculation , it is interesting to note that for these averaged spectra , the differences are relatively small . we find the typical four peak structure ( with lower hubbard , quasiparticle peak and double - peaked upper hubbard band split by hund s rule coupling ) for all three calculations . only the position and the height of the quasiparticle peak are quite different for the hamiltonian calculation , indicating a more insulating solution than for the dos calculations at the same u - value . the similarity between the results for the dos and hamiltonian calculations is not surprising , because the trigonal distortion in v@xmath5o@xmath4 is relatively small . therefore the center of gravity and the bandwidth of the three t@xmath157 bands do not change much and the dos calculations can still produce accurate results . the differences between dos and hamilton calculations are more pronounced when one compares the band - resolved t@xmath157 spectra . in fig . [ dos_hmlt_v2o3i ] and fig . [ dos_hmlt_v2o3 m ] , the a@xmath169 and e@xmath170 spectra for the insulating and metallic crystal structure are presented for u@xmath171 and @xmath172 ev . ( t@xmath157 ) spectral functions for v@xmath5o@xmath4 in the insulating phase calculated via lda+dmft(qmc ) using : lda dos - light line , projected hamiltonian - black line . upper figures - @xmath166 , lower figures - @xmath167 orbitals . fermi level corresponds to zero.,scaledwidth=60.0% ] ( t@xmath157 ) spectral functions for v@xmath5o@xmath4 in the metal phase calculated via lda+dmft(qmc ) using : lda dos - light line , projected hamiltonian - black line . upper figures - @xmath166 , lower figures - @xmath167 orbitals . fermi level corresponds to zero.,scaledwidth=60.0% ] here the curves calculated by different schemes are distinctively different . this is especially clear from the upper part of the figures [ dos_hmlt_v2o3i ] , [ dos_hmlt_v2o3 m ] corresponding to the @xmath166-orbital of the v-3@xmath7(t@xmath157 ) subband . this is due to the fact that the hybridization effects are better accounted for by the projected hamiltonian , which includes not only intra - band hoppings but also inter - band on - site and inter - site hoppings . the latter effectively increases the bandwidth of the @xmath166-orbital . the corresponding spectral functions have a different peak structure and different intensities . another conclusion from this series of figures is that the results obtained with the hamiltonian procedure show a more insulating behavior for the same values of @xmath93 than the results with the scheme using the lda dos . . real part - full black line , imaginary part - full light line . fermi level corresponds to zero.,scaledwidth=50.0% ] the full - orbital calculations scheme proposed in this work produces an interacting green function @xmath135 ( [ green1 ] ) . the knowledge of the full green function allows us to calculate the spectral function not only for the states with correlations ( v-3@xmath7(@xmath23 ) orbitals in the case of srvo@xmath4 and v@xmath5o@xmath4 ) , but also their effect on the lower lying occupied oxygen 2@xmath37 states and the higher lying unoccupied v-3@xmath7(@xmath173 ) states , thus facilitating a comparison of the calculated and experimental spectra in a wide energy region . valence band photoemission spectroscopy ( pes ) using high photon energies and o @xmath174 x - ray absorption spectroscopy ( xas ) were performed on the beamline bl25su at the spring-8 synchrotron radiation facility . the pes spectra were taken using a gammadata - scienta ses-200 electron energy analyzer and xas spectra were measured by the total electron yield . the overall energy resolution was set to 0.2 ev . the pressure in the analyzer chamber was about 4 x 10@xmath175 pa . single crystals of srvo@xmath4 were cooled to 20 k and clean surfaces were obtained by fracturing _ in situ _ for the pes spectra , and by scraping _ in situ _ for the much more bulk sensitive xas spectra . well - oriented single - crystalline v@xmath5o@xmath4 samples were cleaved _ in situ _ at a temperature near the metal - insulator transition , yielding clean specular surfaces . the surface cleanliness was confirmed before and after the spectral run . we start with the results for the srvo@xmath4 system . in fig . [ srvo3_sigma ] , the self - energy on the real energy axis for srvo@xmath4 calculated via ( [ acse ] ) is shown . the @xmath176 ratio is calculated via @xmath177 . this self - energy was used for the calculation of total and partial doss in the full - orbital hilbert space ( see fig . [ srvo3_total],[srvo3_d ] ) . calculated via : lda - full light line , using the full - orbital self - energy from lda+dmft(qmc ) - dashed black line . fermi level corresponds to zero.,scaledwidth=50.0% ] spectral functions for srvo@xmath4 calculated using the full - orbital self - energy from lda+dmft(qmc ) . total @xmath7 - full light line , t@xmath157 - full black line , e@xmath178 - dashed black line . fermi level corresponds to zero.,scaledwidth=50.0% ] in fig . [ srvo3_total ] , the total spectral functions of srvo@xmath4 are presented . differences between lda and full - orbital lda+dmft(qmc ) spectral functions mainly occur near the fermi level . in particular , the lda spectral function has a more pronounced quasiparticle peak . the dos calculated using the full - orbital self - energy has a three - peak structure : lower hubbard band ( suppressed by oxygen states ) , quasiparticle peak and upper hubbard band located at about 3 ev . the origin of the upper hubbard band becomes clear from fig . [ srvo3_d ] . one can see that this broad peak is the sum of v-3@xmath7(e@xmath179 ) and v-3@xmath7(t@xmath157 ) states . it should be noted that correlations on the v-3@xmath7(e@xmath179 ) orbitals were not explicitly included here . we find , however , that the v-3@xmath7(e@xmath179 ) states are slightly modified by the full - orbital self - energy due to the of hybridization with the correlated v-3@xmath7(t@xmath157 ) orbitals . the question how correlations affect the position and width of v-3@xmath7(e@xmath179 ) states directly can only be answered by employing the full 3@xmath7-shell 5x5 projected hamiltonian . introducing correlations between t@xmath157 states changes significantly the total and partial lda doss of srvo@xmath4 . the main modification is a transfer of spectral weight from the energy region near fermi level to the lower and upper hubbard bands , and the reduction of the weight of the quasiparticle peak . our calculations find a strongly correlated but still metallic ground state for srvo@xmath4 . with spectral functions calculated via lda+dmft(qmc ) : taking into account only t@xmath157 - light line , using full - orbital self - energy - black line . theoretical spectra are multplied with the fermi function corresponding to 20k and broadened with a 0.2 ev gaussian to account for the instrumental resolution . intensities are normalized for quasiparticle peaks . fermi level corresponds to zero.,scaledwidth=50.0% ] to compare our results with the experimental pes we calculated a weighted sum of v-3@xmath7 and o-2@xmath37 spectral functions according to the photoemission cross section ratio 3:1 , corresponding to an experimental photon energy 900 ev . the theoretical spectra were multiplied with the fermi function corresponding to 20 k and broadened with a 0.2 ev gaussian to take into account the instrumental resolution . in fig . [ compar_pes_srvo3 ] one can see that the full - orbital spectra obtained in this way describe not only the quasiparticle peak , but also the peak at -6 ev and the shoulder at -3.5 ev of the pes spectra . since previous lda+dmft(qmc ) results were taking into account only the t@xmath157 states they were not able to describe pes spectra below -2 ev . it is interesting to note that previous experimental studies did not find any states at -3.5 ev ; only the new spectra reported here show this feature . real part of the self - energy from lda+dmft(qmc ) - full black line , imaginary part - full light line . fermi level corresponds to zero.,scaledwidth=50.0% ] the influence of correlation effects on the electronic structure will be more pronounced for systems close to the mott insulator transition . for this purpose we compare the total and partial doses for v@xmath5o@xmath4 in the insulating and metallic phases calculated via lda and full - orbital lda+dmft @xcite . since the v-3@xmath7(t@xmath157 ) orbitals in trigonal v@xmath5o@xmath4 split into non - equivalent a@xmath169 and e@xmath180 states , the self - energy will be different for these orbitals ( see fig [ v2o3_sigma ] ) . o@xmath4 calculated via : lda - full light line , using full - orbital self - energy from lda+dmft(qmc)- dashed black line . fermi level corresponds to zero.,scaledwidth=50.0% ] spectral functions for v@xmath5o@xmath4 calculated using full - orbital self - energy from lda+dmft(qmc ) . total d - full light line , t@xmath157 - full black line , e@xmath178 - dashed black line . fermi level corresponds to zero.,scaledwidth=50.0% ] in fig . [ v2o3_total ] one can see that the introduction of correlation effects changes the total dos drastically . whereas the lda dos is metallic for both metallic and insulating phases , the lda+dmft spectra clearly show the metal - insulator transition . another important feature are the modifications near the fermi energy and in the energy window from -3 ev to 7 ev . moreover , there are also changes in the e@xmath179 states which are caused by the hybridization with the correlated t@xmath157 bands , and are not due to a direct calculation of correlations in the e@xmath179 states . in fig . [ v2o3_d ] the total @xmath7-state dos above the fermi energy , consisting of t@xmath157 and e@xmath179 states is shown . o@xmath4 with spectral functions calculated via lda+dmft(qmc ) : taking into account only t@xmath157 - light line , using full - orbital self - energy - black line . theoretical spectra are multiplied with the fermi function and broadened with a 0.2 ev gaussian to simulate instrumental resolution . intensities are normalized for peaks situated around -1 ev . fermi level corresponds to zero.,scaledwidth=50.0% ] in fig . [ compar_pes_v2o3 ] a comparison of experimental pes and calculated lda+dmft(qmc ) ( t@xmath157 only and full - orbital ) spectra is presented . the full - orbital spectrum is a weighted sum of the calculated v-3@xmath7 and o-2@xmath37 spectral functions , according to the photoemission cross section ratio 2:1 , corresponding to the experimental photon energy of 500 ev . the theoretical spectra were multiplied with the fermi function and broadened with a gaussian of 0.2 ev.@xcite below the fermi energy the lda+dmft(qmc ) spectral functions ( for t@xmath157 only and the full - orbital scheme ) agree quite well with the pes spectra . however , the theory curves do not yet describe fine details . specifically , the pes spectrum shows two definite slope changes , at roughly -0.6 ev and -1.5 ev , producing a rather flat topped spectrum centered on -1 ev , whereas the theory curves show a single peak centered on -1 ev . x - ray absorption spectroscopy ( xas ) spectra of v@xmath5o@xmath4 with spectral functions calculated via lda+dmft(qmc ) using full - orbital self - energy . theoretical spectra are multiplied with the fermi function and broadened with a 0.2 ev gaussian to simulate instrumental resolution . intensities of the theory curves are normalized to the correct number of v-3@xmath7 electrons in the unit cell ( 8 electrons ) . the experimental curve is normalized to the same peak height as the full orbital curve . fermi level corresponds to zero.,scaledwidth=50.0% ] a comparison of the calculated lda+dmft(qmc ) full - orbital spectral functions and the o @xmath174 xas spectrum is shown on fig . [ compar_xas_v2o3 ] . the full - orbital spectra ( which are the partial o-2@xmath37 spectra for the unoccupied v-3@xmath7 states ) are found to be in rather good agreement with the structures of the experimental spectrum their relative intensities and their energies . this is due to the inclusion of o-2@xmath37 and v-3@xmath7(@xmath152 ) states in the calculations . the strong hybridization of the o-2@xmath37 with the d - states is described more correctly in the full - orbital calculations and the inclusion of @xmath152 states ( even without correlations ) significantly changes the v-3@xmath7-shell in the energy regions above the fermi edge . the inclusion of correlations in @xmath152 states within a 5x5 projected v-3@xmath7 hamiltonian is expected to add spectral weight in the energy interval between 3 ev and 5 ev ( see fig . [ compar_xas_v2o3 ] ) . we formulated a fully _ ab initio _ and self - consistent computational scheme based on wannier functions ( wf ) for the calculation of the electronic structure of strongly correlated compounds . the wf formalism provides an explicit strategy for the construction of the matrix elements of the required operators ( hamiltonian , self - energy operator , etc . ) , both in full - orbital and few - orbital bases , in real and reciprocal representations . the wf formalism allows one to project these operators from the full - orbital space to the few - orbital space and back , keeping the complete information about all orbitals . these projections are the essential ingredients of the computational scheme presented here . the self - consistency involves not only the self - energy but also the wf basis itself , the charge density with the lda - potential and the interaction strength parameters between the electrons on the wfs . the spectra obtained thereby found to be in good agreement with new bulk - sensitive experimental data . in the present work we did not yet employ the full scheme ( see fig . [ scheme ] ) to investigate spectral functions of strongly correlated systems but employed only few - orbital hamiltonians with t@xmath157 symmetry . clearly , full @xmath7-shell dmft(qmc ) results including e@xmath178 states will provide additional information about correlation effects in the system . such studies , as well as constrained lda calculations in the wf basis and investigations of the feedback from dmft to lda part are in progress now . this work was supported by russian basic research foundation grant rffi - gfen-03 - 02 - 39024_a , rffi-04 - 02 - 16096 and by the deutsche forschungsgemeinschaft through sonderforschungsbereich 484 , augsburg , the john von neumann institut fr computing , jlich , and in part by the joint uro - so project n22 , grant of president of russian federation for young scientists mk-95.2003.02 , dynasty foundation and international center for fundamental physics in moscow program for young scientists 2004 ) , russian science support foundation program for young phd of russian academy of science 2004 . work at the university of michigan was supported by the u.s . nsf under ( grant no . dmr-03 - 02825 ) and at postech by kosef through essc . all soft x - ray absorption and photoemission studies have been performed in spring-8 under the upproval and support of japan syncrotron radiation research institute . for cubic systems the matrix of the self - energy operator ( for example for @xmath23 orbitals ) is diagonal and all diagonal elements are equal . therefore the calculation of the green function within dmft by integration of the hamiltonian over the bz is equivalent to the hilbert transformation of the non - interacting ( lda ) dos @xmath181 : @xmath182 dmft produces green functions and self - energies on the imaginary energy axis . with the maximum entropy method @xcite , the spectral function on the real energy axis is calculated , which yields the imaginary part of the green function . in order to obtain the self - energy on the real energy axis , the full complex green function @xmath183 is calculated using its imaginary part obtained by mem : @xmath184 the self - energy for real energies is then calculated by solving the following two equations with the two variables @xmath185 and @xmath186 : @xmath187 where @xmath188 ( [ wf ] ) is valid only for orthogonal lmto orbitals . in the case of general non - orthogonal lmtos ( or any other atomic type orbital basis set ) , an orthogonalization procedure can be used by defining an orthogonal hamiltonian @xmath189 and the corresponding eigenvectors @xmath190 for the non - orthogonal hamiltonian @xmath191 and overlapping matrix @xmath192 : @xmath193 this orthogonalization is equivalent to the basis set transformation : @xmath194 then for the non - orthogonal lmto the trial function @xmath34 has to be replaced by @xmath195 and the eigenvectors with coefficients @xmath196 by @xmath197 . j. e. hirsch and r. m. fye , phys . lett . * 56 * , 2521 ( 1986 ) ; m. jarrell , phys . * 69 * , 168 ( 1992 ) ; m. rozenberg , x. y. zhang , and g. kotliar , phys . lett . * 69 * , 1236 ( 1992 ) ; a. georges and w. krauth , phys . rev . lett . * 69 * , 1240 ( 1992 ) ; m. jarrell in _ numerical methods for lattice quantum many - body problems _ , edited by d. scalapino , addison wesley , 1997 . k. held , i. a. nekrasov , n. blmer , v. i. anisimov , and d. vollhardt , int . j. mod b * 15 * , 2611 ( 2001 ) ; k. held , i. a. nekrasov , g. keller , v. eyert , n. blmer , a. k. mcmahan , r. t. scalettar , t. pruschke , v. i. anisimov , and d. vollhardt , avilable as cond - mat/0112079 ( published in _ quantum simulations of complex many - body systems : > from theory to algorithms _ , eds . j. grotendorst , d. marks , and a. muramatsu , nic series volume 10 ( nic directors , forschunszentrum jlich , 2002 ) p. 175 - 209 ( isbn 3 - 00 - 009057 - 6 ) ; k. held , i. a. nekrasov , g. keller , v. eyert , n. blumer , a. k. mcmahan , r. t. scalettar , th . pruschke , v. i. anisimov , and d. vollhardt , psi - k newsletter * 56 * , 65 ( 2003 ) [ psi-k.dl.ac.uk/newsletters/news_56/highlight_56.pdf ] ; a. i. lichtenstein , m. i. katsnelson , g. kotliar , in _ electron correlations and materials properties 2nd ed . _ , edited by a. gonis , nicholis kioussis and mikael ciftan , kluwer academic / plenum , p. 428 , new york ( 2002 ) , avilable as cond - mat/0112079 . pchelkina , d.e . kondakov , i.v . leonov , v.i . anisimov , v.a . gavrichkov , s.g . ovchinnikov , to be published in the physics of metals and metallography ( proceedings of easmag-2004 , http://eastmag-2004.kirensky.ru/registration/proc/ ) . in the calculations performed in the present paper we assumed the position of the fermi level to be unchanged when going from lda to dmft(qmc ) . it was found in previous investigations @xcite that this is a good approximation for vanadates . lambin and j. p. vigneron , phys . b * 29 * , 3430 ( 1984 ) ; o. jepsen and o. k. andersen , solid state commun . * 9 * , 1763 ( 1971 ) ; p. e. blchl , o. jepsen , and o. k. andersen , phys . b * 49 * , 16223 ( 1994 ) . i. a. nekrasov , g. keller , d. e. kondakov , a. v. kozhevnikov , th . pruschke , k. held , d. vollhardt , v. i. anisimov , cond - mat/0211508 ; a. sekiyama , h. fujiwara , s. imada , s. suga , h. eisaki , s. i. uchida , k. takegahara , h. harima , y. saitoh , i. a. nekrasov , g. keller , d. e. kondakov , a. v. kozhevnikov , th . pruschke , k. held , d. vollhardt , v. i. anisimov , cond - mat/0312429 , a. sekiyama , h. fujiwara , s. imada , h. eisaki , s. i. uchida , k. takegahara , h. harima , y. saitoh , s. sugacond - mat/0206471 . mo , j. d. denlinger , h .- kim , j .- h . park , j. w. allen , a. sekiyama , a. yamasaki , k. kadono , s. suga , y. saitoh , t. muro , p. metcalf , g. keller , k. held , v. eyert , v. i. anisimov , and d. vollhardt , phys . lett . * 90 * , 186403 ( 2004 ) .

we present measurements of the size of the vela pulsar in 3 gates across the pulse , from observations of the distribution of intensity . we calculate the effects on this distribution of noise in the observing system , and measure and remove it using observations of a strong continuum source . we also calculate and remove the expected effects of averaging in time and frequency . we find that effects of variations in pulsar flux density and instrumental gain , self - noise , and one - bit digitization are undetectably small . effects of normalization of the correlation are detectable , but do not affect the fitted size . the size of the pulsar declines from km ( fwhm of best - fitting gaussian distribution ) to less than 200 km across the pulse . we discuss implications of this size for theories of pulsar emission . # 10= -.025em0 - 0 .05em0 - 0 -.025em.0433em0 epsf physics department , university of california , santa barbara , california , 93106 , usa school of physics , university of melbourne , parkville 3052 , victoria , australia australia telescope national facility , epping , new south wales , 2121 , australia physics department , university of tasmania , hobart , 7001 , tasmania , australia institute of space and astronautical science , 3 - 1 - 1 yoshinodai , sagamihara , kanagawa 229 , japan hartebeesthoek radio astronomy observatory , krugersdorp , transvaal , south africa jet propulsion laboratory , california institute of technology , pasadena , california , 91109 , usa 1 truein

astro-ph9908254

pulsars emit meter- to centimeter - wavelength radiation with brightness temperatures of @xmath8 k , greater than any other astrophysical sources . the ultimate energy source for pulsar emission is rotation of the neutron star . combined with this rotation , the neutron star s magnetic field generates forces sufficient to accelerate electrons from the stellar surface . these energetic electrons or the gamma - rays they emit can pair - produce after a few centimeters . the consequent cascade of electron - positron pairs permeates the magnetic field of the pulsar out to the light cylinder . the electrons and positrons follow the lines of the strong magnetic field closely . particles on the `` open '' field lines , which pass through the light cylinder , gradually carry away part or all of the rotational kinetic energy of the pulsar as an electron - positron wind , leading to the observed spindown of the pulsar . the `` polar cap '' is the region defined by the open field lines at the surface of the neutron star , from which the wind is drawn . this basic picture has long been understood ( @xcite , @xcite , @xcite , @xcite ) . despite this understanding , the physical mechanism for conversion of a small fraction of the spindown power to a beam of radio emission , which sweeps past the observer each rotation period to produce the observed pulses , remains uncertain . observational tests of models for pulsar emission are difficult because the emission is so compact . it can not be resolved with instruments of earthlike or smaller dimensions . melrose ( 1996 ) and asseo ( 1996 ) recently reviewed theories of pulsar emission . in all models , collective behavior of the electron - positron plasma results in coherent emission from groups of particles . melrose divides theories into 4 categories , according to location of the emission region : ( i ) at a pair - production front at the polar cap ; ( ii ) in the electron - positron wind on open field lines above the polar cap ; ( iii ) in the electron - positron wind far above the cap , near the light cylinder ; and ( iv ) at , or even outside , the light cylinder . theories from all 4 categories appear viable . indeed , more than one emission mechanism may act , to produce the wide range of observed pulse morphologies ( @xcite ) . pulsars exhibit rich temporal structure , including complicated variations of structure of individual pulses about an average , and large pulse - to - pulse variations in flux density ; these offer clues to the emission process ( @xcite , @xcite ) . because the fundamental modes of lower - frequency electromagnetic waves are linearly polarized , either across or along the pulsar s magnetic field ( @xcite , @xcite , @xcite , @xcite ) , the strong linear polarizations of pulsars yield the direction of the pulsar s magnetic field at the locus of emission ( @xcite ) . polarization observations usually suggest that emission arises near the magnetic pole of the neutron star , which is offset from the rotation pole . pulsars show stable , sometimes complicated long - term average profiles , but with large pulse - to - pulse variations . the most successful classification of these profiles is that of rankin ( @xcite , @xcite ) , who divides emission into `` core '' and `` cone '' components , corresponding respectively to elliptical pencil beams and hollow cones directed from above the magnetic pole of the neutron star . a host of associated pulse properties support this classification ( @xcite , @xcite , @xcite ) . the existence of these different components may indicate the presence of more than one emission process ( @xcite , @xcite ) . a few core emitters show interpulses , suggesting that their magnetic poles lie near the rotation equator . comparison of pulse width with period for these pulsars , and extension of this relation to all core emitters , closely tracks the relation expected if the size of the polar cap sets the angular width of the beam for core emission ( @xcite ) . rankin concludes that core emission arises very near the surface of the neutron star . interstellar scattering offers the possibility of measuring the sizes of pulsars and the structure of the emission region , and its changes over the pulse , using effective instruments of au dimensions ( @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ) . in this paper , we describe measurement of the size of the vela pulsar , from the distribution of flux density in interstellar scintillation , observed on a short interferometer baseline . density fluctuations in the interstellar plasma scatter radio waves from astrophysical sources . these fluctuations act as a corrupt lens , with typical aperture of about 1 au . like such a lens , scattering forms a diffraction pattern in the plane of the observer . for a spatially - incoherent source , the intensity of the pattern is the convolution of the intensity of the pattern for a point source , with an image of the source ( goodman 1968 ) . the scattering system has resolution corresponding to the diffraction - limited resolution of the `` scattering disk '' , the region from which the observer receives radiation . the finite size of the scattering disk sets a minimum spatial scale for the diffraction pattern in the plane of the observer , at the diffraction limit . this scale is @xmath9 , where @xmath10 is the observing wavelength and @xmath11 is the angular standard deviation of the scattering disk , as seen by the observer . if the plasma fluctuations are assumed to remain `` frozen '' in the medium , while the motions of pulsar , observer , and medium carry the line of sight through it at speed @xmath12 , this spatial scale sets the timescale of scintillations , @xmath13 . differences in travel time from the different parts of the scattering disk set the minimum frequency scale of the pattern , at @xmath14^{-1}}$ ] , where @xmath15 is the characteristic distance from observer to scatterer , and @xmath16 is the characteristic distance from scatterer to source . observations with time resolution shorter than @xmath17 and with frequency resolution finer than @xmath18 are in the `` speckle limit '' of interstellar scattering . for an extended , spatially - incoherent source , overlap of different parts of the image of the source in the observer plane , after the convolution , reduces the depth of modulation of scintillation : `` stars twinkle , planets do not . '' the relation of cohen et al . ( 1967 ) and salpeter ( 1967 ) quantifies this fact as a relation between the size of the source and the depth of modulation of scintillations . the vela pulsar is particularly well - suited for studies of pulsar emission because it is strong , heavily scattered , and relatively nearby . in a typical `` scintle '' of observing time @xmath17 and bandwidth @xmath18 , a typical terrestrial radiotelescope can attain a signal - to - noise ratio of a few , for observations of the vela pulsar at decimeter wavelengths . a short observation can thus sample many scintles . moreover , the scattering disk is large enough at these wavelengths that the scale of the diffraction pattern is on the order of the size of the earth , so that different radiotelescopes on earth can sample different scintles . the nominal linear resolution of the scattering disk , acting as a lens , is about 1000 km at the pulsar . for the vela pulsar , the diameter of the light cylinder is 8500 km . in contrast , the size of the polar cap is tens of meters , for a dipole magnetic field and a @xmath19 km radius for the neutron star . thus , observations of the scintillation pattern can distinguish among different theories of pulsar emission , as discussed in [ psr_emission ] . among the properties that place the vela pulsar s pulse into the core class are its single component , varying little in width with frequency ; its short period and rapid spindown rate ; its circular polarization ; and its weak , irregular pulse - to - pulse variations ( @xcite , @xcite ) . however , vela also shows characteristics of cone emission , including strong , well - ordered linear polarization and a double profile in x- and gamma rays ( @xcite ) . indeed , radhakrishnan et al . ( 1969 ) first observed in the vela pulsar the s - shaped variation of the direction of linear polarization characteristic of cone pulsars . manchester ( @xcite ) has proposed that vela s pulse ( and those of some other young pulsars ) could represent one side of broad cone , which would help to explain the apparent inconsistency . on the other hand , romani & yadigaroglu ( 1995 ) propose that the radio pulse is indeed core emission , perhaps from very close to the polar cap , and that the double - peaked x- and gamma - ray emission arises far out in the magnetosphere , near the light cylinder . krishnamohan & downes ( 1983 ) performed an extensive study of 87040 pulses from the vela pulsar , and found that the pulse shape varied with peak intensity . notably , strong pulses tend to arrive earlier . they also found that the rate of change of angle of linear polarization changes with pulse strength , and interpreted this as activity from emission regions at different altitudes and magnetic longitudes . they inferred variations in the location of the emission region by about 400 km in altitude , and by about about @xmath20 in longitude . emission earlier in the pulse was found to arise further from the star , and from a larger region . interferometers present several advantages over single antennas for observations of scintillating sources . because noise does not correlate between antennas , no baseline response need be subtracted from observations . moreover , interferometers are nearly immune to interference and emission from extended or unrelated nearby sources . the statistics of interferometer response to signals and noise is well studied ( see , for example , @xcite ) . for a short interferometer baseline , interferometers measure the intensity of the source , commonly expressed as flux density . disadvantages of interferometers include the facts that they require coordinated observations at multiple antennas , and so often yield smaller effective apertures than the largest available single antenna ; and they often afford narrower bandwidths than some observing modes at single antennas . for the present work , careful treatment of statistics is essential and interferometers are preferred . we focus on observations of scintillating sources on short interferometric baselines in this paper . for a point source in strong scintillation , sampled in the speckle limit , the distribution of flux density follows an exponential distribution ( @xcite , @xcite ) : @xmath21 this equation holds if the source is small , compared with the resolution of the scattering disk treated as a lens , so that @xmath22 , where @xmath23 is the magnification of that lens , and @xmath24 where @xmath10 is the observing wavelength . the size of the source , parametrized as the standard deviation of a circular gaussian distribution of intensity , is @xmath25 . if @xmath26 is finite but small compared with 1 , then the distribution of flux density is the weighted sum of 3 exponentials ( @xcite ) . if , for example , the scattering disk is modeled as an elliptical gaussian , and the source is modeled as a circular gaussian , then the exponential scales are @xmath27 where @xmath28 and @xmath29 are the angular standard deviations of the scattering disk along the major and minor axes , and @xmath30 sets the scale of flux density . we assume that @xmath31 ; this is equivalent to the assertion that the scattering disk , treated as a lens , does not resolve the source . the probability distribution for flux density is then : @xmath32 at large flux denisty , the first term dominates and the distribution remains exponential , but with different scale @xmath30 and normalization than for a point source . at small flux density , the probability density falls to zero at zero flux density , in a way that depends on source size @xmath25 . the larger the source , the more sharply the distribution of flux density is peaked near the average flux density . figure [ 3exp_fig ] shows sample distributions . in this paper , we use this expression , with modifications for the effects of noise , to find the size of the vela pulsar in 3 gates across the pulse . source size reduces the intensity variation due to scintillation . the cohen - salpeter relation formalizes this fact as a relation between angular size and modulation index , @xmath33 . here the angular brackets @xmath34 denote an ensemble average , approximated in practice by a time average . for a circular gaussian source viewed through a elliptical scattering disk , in strong scattering ( @xcite ) , @xmath35 this expression is valid for arbitrarily large source size @xmath25 . the modulation index calculated from eq . [ psell ] is consistent with this expression in the limit @xmath36 . because the scintillation pattern varies with time and frequency , integration over either reduces the depth of modulation and narrows the distribution of amplitude . it thus affects the distribution in a way similar to finite source size . averaging leaves the mean flux density @xmath37 unchanged , but reduces the mean square flux density @xmath38 , and so reduces the modulation index . we can calculate the effects of averaging on the modulation index , and liken them to the effects of source size , to assess effects of averaging on estimates of source size . for a source with pointlike or gaussian intensity distribution , seen through a gaussian scattering disk , the correlation flunction of flux density with frequency follows a lorentzian distribution ( see , for example , @xcite ) : @xmath39 where @xmath40 and @xmath41 are the observing frequencies , and @xmath18 is the decorrelation bandwidth of the scintillations . for a small source with a gaussian distribution of intensity , the correlation function has the same form , with larger decorrelation bandwidth . for non - gaussian spectra of density fluctuations in the scattering material , the correlation function shows the same behavior , with a different functional form ( @xcite ) . after averaging in frequency by convolving with a `` boxcar '' of bandwidth @xmath42 , the mean intensity is unchanged , and the mean square intensity becomes @xmath43.\nonumber \\ & \equiv & \langle s(\nu)^2\rangle f_f\nonumber\end{aligned}\ ] ] here the bar on @xmath44 denotes averaging . this expression defines the convenient factor @xmath45 . a similar approach yields effects of averaging in time . the spatial scintillation pattern for a point source is the fourier transform of the screen phase ( @xcite , @xcite @xcite ) . commonly , it is assumed that scintillation in time represents the motion of the scintillation pattern across the instrument , while the screen and pattern remain unchanged . for an extended source , the pattern is the convolution of the point - source response with an image of the source . if both source and scattering disk follow gaussian distributions of intensity , then the correlation function in time is gaussian : @xmath46 where @xmath47 and @xmath48 are the times of observation , and @xmath17 is the timescale of interstellar scintillation . if the observer averages over time interval @xmath49 , the mean square flux density after that averaging is @xmath50 \equiv \langle s^2\rangle f_t \label{time_avg_s2}\ ] ] where @xmath51 . the effects of averaging can be expressed as a reduction in modulation index or an increase in inferred source size . the post - averaging modulation index @xmath52 is related to the pre - averaging value by @xmath53 using eq . [ modind_size ] we can infer a post - averaging size @xmath54 from @xmath52 , and so relate @xmath55 to the true size @xmath25 , using knowledge of the averaging bandwidth and time . this is the approach taken below . the effects of noise can be important for studies of scintillating pulsars because the scintles of lowest flux density are important in determining the size of the source . similarly , noise affects the distribution of intensity most strongly at the lowest intensities . thus , noise is important in determining source size . the interferometrist observes the signal from the source , along with noise , at 2 stations @xmath56 and @xmath57 and measures their correlation : @xmath58 . the correlation @xmath59 is often called the interferometric visibility , and is often expressed in units of flux density , using the measured or estimated gains of the antennas . for interferometric observations , @xmath59 is complex ( @xcite ) . in this paper , we use boldface characters to denote complex quantities . here @xmath60 are the electric fields in the observed polarization , at the two antennas ; and @xmath61 is the number of independent samples , the product of integration time and bandwidth . system noise may be in or out of phase with the signal , and so contributes to both real and imaginary components . signals from astrophysical sources are intrinsically noiselike , and their stochastic variations represent another source of noise , known as self - noise . self - noise contributes only in phase with the signal . without loss of generality , we take the phase of the signal to define the real axis . self - noise then affects only the real component . for a strongly - scintillating point source , the measured correlation @xmath59 is drawn from the distribution : @xmath62 where @xmath63 is the chi - square distribution with @xmath61 degrees of freedom ( see , for example , @xcite ) . again , @xmath30 is the mean flux density of the source , averaged over many scintles . we measure the correlation in units of flux density , using the gains of the antennas @xmath64 . the system noise in units of flux density is @xmath65 where @xmath66 are the system temperatures at stations 1 and 2 , and @xmath64 are the gains at stations 1 and 2 . this equation assumes that the noise from the source has gaussian statistics . in practice the radiation from all known astrophysical sources does appear to have gaussian statistics . in particular , radiation from pulsars follows gaussian statistics closely , with time - varying amplitude ( @xcite ) . for some pulsars , the amplitude varies quite dramatically over short times within the pulse ( @xcite , @xcite ) , although the vela pulsar seems to be relatively stable in this respect . when the number of independent samples @xmath61 is large , the chi - squared distribution is well approximated by a gaussian , and the distribution takes the form : @xmath67 the variances of the real and imaginary parts of the correlation are @xmath68 and @xmath69 the flux density of the source , @xmath70 , appears in the denominator of the exponent in eq . [ general_cdist ] , through @xmath71 . this reflects self - noise . for 2 identical antennas , system temperature and gain are equal , so that @xmath72 and @xmath73 , and the correlation @xmath74 is drawn from the distribution : @xmath75 unfortunately , this integral is not easily simplified . figure [ selfnoisedist_fig ] shows plots of the distribution of the amplitude of the correlation , @xmath76 , for a scintillating point source , for sample values of @xmath77 and @xmath61 . when @xmath61 is large , self - noise is negligible and the ratio @xmath78 is the average signal - to - noise ratio . when @xmath61 is small , self - noise becomes important . as the figure shows , self - noise increases the number of points at low amplitude , and flattens the distribution at high amplitude . if the number of independent samples @xmath61 is large , then self - noise can be ignored . the distribution of amplitude , @xmath79 , then takes the often - useful form : @xmath80 where @xmath81 is the error function ( @xcite ) . the noise , in flux density units , is @xmath82 . we have adopted the scaled parameters @xmath83 if the antennas are not identical , the distribution has the same form in this limit , but the definitions of the scaled parameters are slightly different . because self - noise can often be ignored , as figure [ selfnoisedist_fig ] suggests , this expression is often appropriate . for scintles much stronger than noise ( @xmath84 ) , and average flux density not much smaller than noise ( so that @xmath85 ) , this distribution approaches the purely - exponential form @xmath86 the constant of the exponential is the same as that for a noise - free scintillating point source , but the normalization is different . in effect , noise shifts the distribution toward greater amplitude . when the source has small but finite size , and self - noise can be ignored , the distribution of @xmath87 takes the form suggested by the combination of eqs . [ psell ] and [ erfc ] : @xmath88 here eq . [ sscales_def ] defines @xmath30 , @xmath89 , and @xmath90 , and eq . [ erfc ] defines @xmath91 ) . this relatively simple relationship holds even though the effects of noise on the amplitude distribution can not be described as a convolution . for scintles with amplitude much larger than nosie , and average flux density not much less than noise , ( @xmath84 and @xmath85 , the limits of eq . [ erfc_approx ] ) , only the first of the 3 terms is important , and this distribution also approaches the purely - exponential form : @xmath92 in this expression , effects of source size appear only through the normalization . this fact demonstrates that source size affects the form of the distribution only at small amplitude . if the gain of the observer s instrument varies with frequency or time , then the distribution of observed flux density will be the superposition of several distributions of the form given by eq . [ psexpdist ] ( or eq . [ psell ] , if the source is extended ) . for example , consider a scintillating point source observed with a distribution of gains @xmath93 , where @xmath94 is the gain . we take @xmath70 to be the true flux density of the source , and @xmath95 to be the antenna temperature , the observed quantity . the distribution of @xmath96 is then the convolution of the distribution of @xmath70 with that for @xmath97 : @xmath98 similarly , if the flux density of the source is not constant , then the distribution of measurements includes contributions from exponential distributions with different scales . the overall distribution of flux density will be the same ( eq . [ superposed_dists ] ) , although the effective gain variations will be stochastic rather than deterministic . in either case , the distribution function will be concave upward on a semilog scale , because the distributions with highest average flux density fall off most slowly . figure [ superposed_dist_fig ] shows examples , for flat distributions of gain about the mean . the effect is significant when the distribution gains includes very small or zero values . practical interferometers measure approximations to the visibility @xmath99 . in particular , interferometers almost always measure the normalized correlation ( @xcite , pp . 214 , 248 ) : @xmath100 here @xmath101 is a correlator - dependent constant , and @xmath70 is the flux density of the source . note that the denominator in eq . [ rho_0_ta_ts ] does not include the number of samples @xmath61 . the denominator is simply a normalization factor ; it does not arise from , or characterize , the noise in the measurement . most signals are digitized before correlation . for 1-bit sampling , as for the haystack correlator used for the observations described here , the measured correlation coefficient @xmath102 is related to @xmath103 by the van vleck relation : @xmath104 often correlators tabulate the correlation coefficient scaled by the linearized van vleck correction @xmath105 . because normalization and digitization become nonlinear at large amplitude , and noise and source structure are important at small amplitude , their effects are relatively easily separable . because normalization and digitization are important at large amplitude , where the distribution of amplitude is exponential even in the presence of source structure and noise , we investigate effects of normalization and digitization on such a distribution : @xmath106 . for 2 identical antennas observing an unresolved source , @xmath107 , and visbility is equal to the flux density , @xmath108 . then the distribution of @xmath103 will be @xmath109 note that the characteristic scale of the exponential is the signal - to - noise ratio in a single sample , @xmath77 . it is independent of integration bandwidth or time . at small @xmath103 , @xmath110 can be approximated by an exponential distribution : @xmath111 the additive constant `` 2 '' in the exponential arises from the factor of @xmath112 before the exponential in eq . [ rhodist_eq ] . we can also include the effects of 1-bit sampling , by including the van vleck relation in calculating this distribution . for the scaled correlation coefficient , @xmath113 we show sample distributions of @xmath110 and @xmath114 in figure [ rhodist_fig ] . as this figure demonstrates , the distribution of correlation departs from an exponential distribution at large amplitudes , when the signal - to - noise ratio is large in a single sample . the modulation index , @xmath115 , parametrizes source size @xmath116 , as eq . [ modind_size ] shows . one can estimate the modulation index directly using this relation , by replacing the ensemble averages with finite averages over a set of observations . such a procedure replaces @xmath117 with @xmath118 , and @xmath119 with @xmath120 , and yields an estimate @xmath121 for the modulation index . unfortunately , these finite sums are weighted by flux density , so that they are most sensitive to the scintles with the greatest flux density . for a nearly exponential distribution of flux density , as expected for a small source in strong scattering , these strong scintillations are rare . therefore , the sums are slow to converge . the measurement is heavily subject to the shot noise of scintillation ( @xcite ) . moreover , source size affects the weakest parts of the distribution , to which these sums are least sensitive ; whereas most of the systematic effects discussed above are greatest for the strongest parts of the distribution . to exclude the possibility that the estimate of size is biased by effects other than source structure , such as those discussed in [ variations_gain_flux_sec ] through [ average_modind ] , a careful study of the full distribution is the most accurate approach . the modulation index expresses the distribution as a single value . weakest scintles are also the most common , so the important parts of the distribution can be measured with greater accuracy than can the modulation index . although the shot noise of individual scintles can limit the accuracy of a measurement based on a fit to the distribution , the accuracy is much better than provided by estimates of modulation index . a pointlike scintillating source produces an exponential distribution of flux density , so that one expects an exponential distribution of visibility amplitude for an interferometer , if the baseline is short compared to the spatial scale of the diffraction pattern . averaging in frequency or in time can reduce the modulation due to scintillation . we show that variations in system gain or intrinsic flux density of the source affect this distribution , particularly if the variations are of order 100% . system noise and self - noise also change the distribution . system noise affects the form of the distribution function at small amplitudes , and by a change in the normalization at large amplitudes . self - noise affects the distribution at all amplitudes when the number of samples @xmath61 is small . normalization of the correlation function , and 1-bit sampling , affect the correlation at large amplitudes , particularly when the source is strong relative to system noise . for a source of small , but finite size , the distribution of flux density is the weighted sum of 3 exponentials . by `` small '' we mean here that the source is small compared with the resolution of the scattering disk seen as a lens . scales of two of these exponentials depend on the size of the source , and are thus useful for size determination , the subject of this paper . these two exponentials affect the distribution at small flux density . because effects of system noise and source size are both important at small correlation amplitude , both effects must be included in data reduction . considerations of finite size , noise , self - noise , normalization , and quantization can be important for single - dish observations of scintillating pulsars as well . the effects of these factors depend on the details of the detection and sampling schemes used at the antenna . jenet & anderson ( 1998 ) describe some of the important effects , for digitization and autocorrelation . we observed the vela pulsar and comparison quasars on 1992 oct 31 to nov 1 using radiotelescopes at tidbinbilla ( 70 m diameter ) , parkes ( 64 m ) , and hobart ( 25 m ) in australia ; hartebeesthoek ( 25 m ) in south africa ; and the 7 antennas of the very long baseline array of the us national radio astronomy observatory that could usefully observe the vela pulsar . each antenna observed right - circular polarized radiation in 14@xmath122 2-mhz bands between 2.273 and 2.801 ghz . these bands are sometimes referred to as `` if bands '' or `` video converters '' ; here we call them frequency bands or simply bands . the data were digitized to 1 bit ( that is , we measured only the sign of the electric field ) , and recorded with the mark iii recording system . we correlated the data with the mark iiib correlator at haystack observatory , with time resolution of 5 s and with 160 time lags . fourier transform of the correlation functions to the frequency domain yields cross - power spectra with 80 channels , with frequency resolution of 25 khz . because of correlator limitations , we correlated only 6 of the 14 recorded bands for most scans . correlations are tabulated as normalized fractional correlation with one - bit sampling , @xmath102 , as given by eqs . [ rho_0_ta_ts ] and [ van_vleck_relation ] . for each baseline , we correlated the data in 3 gates across the pulse . gate 1 covers approximately the 13 milliperiods up to the peak of the pulse , gate 2 covers the next 13 milliperiods , and gate 3 covers the next 25 milliperiods . the 3 gates have average intensity ratios of about 0.7:1:0.5 , as discussed further in [ obsintdist_sec ] below . the processor set the gates so that they were delayed according to the pulsar s measured dispersion ( @xcite , @xcite ) at the center frequency of each of the 14 bands . thus , dispersion smears the gates in time by no more than the dispersion across 1 mhz , or @xmath123 milliperiods . to the extent that structure of the pulsar s emission region changes with pulse phase , we can regard the gates as sampling diffraction patterns from sources with different structures , or from a single source viewed from different angles . the haystack correlator is a lag or `` xf '' correlator , so spectral data are affected by the fractional bitshift effect ( @xcite ) . after correlation , we removed the average phase slope introduced by this effect for each time sample . because the pulsar period is short compared to the fractional bit - shift rate , we did not encounter aliasing between the fractional bit - shift and the pulse period ( @xcite ) . for each baseline , we analyzed all 3 gates with identical phase models , with identical correlator parameters and the same phase model removed from each gate . this phase model consists of a phase offset and slopes in time and frequency , fit to the data in the 3 gates , for all correlated tracks over each 13-minute scan . this process is often called `` fringing '' the data . because we study the distribution of amplitude of correlation in this paper , the phase model influences the results only indirectly , through the effects of averaging . after fringing , we averaged the data in time to increase signal - to - noise ratio . we boxcar - averaged the data by 2 samples , to a resolution of 10 s in time . this resolution in frequency and time is finer than the diffraction pattern of the pulsar , but coarse enough to yield good signal - to - noise ratio . in this paper we focus on observations on the short tidbinbilla - parkes baseline , about @xmath124 km long . the tidbinbilla and parkes antennas are both large and have sensitive receivers . they can be regarded as nearly identical ; any difference is important only for details of the distribution of normalized correlation at high amplitude , discussed in [ normcorr_vanvleck_sec ] above and [ rhodist_obs_sec ] below . because the characteristic spatial scale of the diffraction pattern is @xmath125 , this short baseline effectively measures the flux density of the scintillation pattern at a single point in the plane of the observer . from observations of a strong continuum source , we found that gain varies within each recorded band . figure [ passband_fig ] shows the amplitude of a strong continuum source plotted with frequency , as an example . the source , 0826@xmath126373 , was observed on 1992 nov 1 from 0:22:30 ut to 0:35:30 ut . the figure shows the band between 2286.99 and 2288.99 mhz . the rolloff of gain at the high- and low - frequency ends of the band arises primarily from the filtering required to isolate the recorded band . these gain variations , and the differences in gain between recorded bands , can be measured from observations of a continuum source and , in principle , corrected . however , the noise does not follow the same profile , so this correction would distort the statistics of noise . because our measurements rely heavily on accurate knowledge of the noise level , we delete the outer part of each recorded band ( the lower 25 and the upper 20 of the 80 channels , as shown in figure [ passband_fig ] ) , and use the central part without gain calibration . we detect no significant variation of gain with time . we measured the system noise from the distribution of amplitude for 0826@xmath126373 , and compared the result with other estimates . for a strong continuum source , the distribution of correlation is expected to follow a 2-dimensional gaussian distribution in the complex plane , offset from the origin by the correlated flux density of the source , with standard deviation equal to the noise level @xmath116 ( @xcite ) . for a strong source , the distribution of the amplitude of the correlation follows a 1-dimensional gaussian distribution with the same standard deviation @xmath116 , centered on the mean amplitude . we found the distribution of amplitude about the mean for 0826@xmath126373 using the same data shown in figure [ passband_fig ] . each amplitude was measured with bandwidth 25 khz and time averaging of 10 s , in the central region of the band , as shown for a single band in figure [ passband_fig ] . we fit a gaussian distribution to the amplitude found in each of the 6 correlated frequency bands , with parameters of the standard deviation , normalization , and mean amplitude . table [ table1_cont_hist_6_1f2t_002231 ] summarizes results of these fits . the noise , given by standard deviation , should be identical in each band . the fits show a variation of up to 2.4% in the noise , or up to 2.2 times the standard errors . again , we do not regard these differences as significant . the mean amplitudes show variations of up to 3.0% , with high significance , reflecting variations in instrumental gain in the different bands . the differences in the noise do not correlate with the differences in gain . we could correct for the gain variations , but prefer to preserve identical conditions in all bands . figure [ cont_hist_off_fig ] shows the distribution of correlated flux density for all the recorded bands . so that the histogram reflects noise , rather than the variations in gain between bands , we removed the mean amplitude for each band , as given in table [ table1_cont_hist_6_1f2t_002231 ] , and added the overall mean of 1500 . the distribution of noise is nearly gaussian , as expected . a fit to this composite distribution yields noise level @xmath127 . this value is quite consistent with that found from other scans , and with the values for individual recorded bands , as table [ table1_cont_hist_6_1f2t_002231 ] shows . we can also compare this noise level with that found in observations that failed to detect any fringes , which should produce purely noise ; and in observations in a gate off the pulsar pulse . in either case the distribution function in the complex plane is a circular gaussian , centered at the origin ( @xcite ) ; the resulting distribution of amplitude follows a rayleigh distribution . in earlier work , we described the second approach ( @xcite ) , applied to observations made at the same epoch , but for a different scan on the pulsar , correlated on a different date and with a slightly different correlator configuration . expressed in the same units , this procedure yielded a noise level of @xmath128 this differs by less than the standard error from the value we measure here . we adopt @xmath129 . because the pulsar gate reduces the duty cycle , we must increase the noise level to reflect the shorter effective integration time , for comparisons with gated pulsar observations . gates 1 and 2 have duty cycles of 0.013 , and gate 3 has a duty cycle of 0.021 ; the noise increases by a consequent factor of 8.77 or 6.32 , respectively . quantization during analog - to - digital conversion affects the noise level . qualitatively , the presence of strong signals in the recorded band reduces the noise level in frequency channels where the signal is weak or absent ( @xcite ) . the `` dithering '' employed to enhance dynamic range in some commercial analog - to - digital conversions is a closely related phenomenon ( @xcite ) . quantitatively , the noise is not distributed evenly over the band , but is distributed according to the product of autocorrelation spectra at the 2 antennas ( which can be inferred from the cross - power spectrum ) plus `` white '' digitization noise . however , the summed , squared noise ( in the absence of self - noise ) remains the same as for zero correlation , or a source without spectral variation . we compute the expected noise level for weak signals for each time sample of the scintillation spectrum of the pulsar , and after excising outliers , use the mean value for the expected noise level . the resulting values , used in our analysis , are given in table [ table2_convfunc_fits_3best.out ] . relatively more conservative or more liberal cuts had little effect on the results of the fits for size , described below . we used the distributions of amplitude observed for the pulsar , in the 3 gates across the pulse , in combination with the measured system noise and the models discussed in [ theory_section ] , to find the size of the pulsar in the 3 gates . we used the ratio @xmath130 ( @xcite ) to set the ratio of the scales @xmath131 and @xmath132 for the fits . just as for the continuum source 0826@xmath126373 , the data were averaged for 10 s in time to yield high signal - to - noise ratio while remaining within the speckle limit , as discussed in [ iss_intro ] above . extrapolation from previous , published measurements yields a decorrelation bandwidth for the pulsar of @xmath133 khz at our observing frequency of 2.3 ghz , and a scintillation timescale of @xmath134 s ( @xcite ; @xcite ; @xcite ) . from our observations , as discussed in [ avg_time_freq_sec ] below , we infer a decorrelation bandwidth of @xmath135 khz and a timescale of @xmath136 s. figure [ 3best.gates123.1f2t.binvar_fig ] shows the observed distribution of correlation in the 3 gates . to each histogram , we fit a model for the size of the pulsar , of the form given by eq . [ 3cdist ] . we used the levenburg - marquardt algorithm for the nonlinear fit ( see , for example , @xcite ) . we fit for 3 parameters , corresponding to the overall normalization , the exponential constant @xmath30 , and the size of the source scaled by scattering parameters , @xmath137 . we searched broad ranges of initial values for the parameters , including the range from 0 to 0.3 for the scaled size of the source . table [ table2_convfunc_fits_3best.out ] summarizes the results of the fits . we found that the best - fitting model has nonzero size for the pulsar , with size that decreases across the pulse . the results for the normalization parameter were approximately consistent with the number of values in the histogram , as expected for a good fit . we can find the effects of averaging in time and frequency by theoretical calculations , and by empirically comparing results for data with different degrees of averaging . to find results of averaging , we boxcar - averaged the interferometric correlation over 1 , 2 , or 3 of the 25-khz channels output by the correlator in frequency , and over 1 , 2 , or 3 of the 5-sec samples in time . we then fit a model as described in [ obsintdist_sec ] , with a free parameter for source size , and with the strength of noise as found in [ ssec_gain_noise ] , scaled appropriately by the inverse square root of the averaging time and frequency . figure [ peakfit_fig ] shows the results for gate 2 , the gate in which the pulsar was strongest . results for the other gates are similar , although the extrapolated size of the pulsar with zero averaging are different . averaging in time or frequency increases the best - fitting size , as expected . the curves in figure [ peakfit_fig ] show the best - fitting forms for the expected variation of size @xmath138 as a function of averaging bandwidth and time interval , as given by eqs . [ modind_size ] , [ freq_avg_s2 ] , and [ time_avg_s2 ] , fitted to the data points . parameters of the fits are the decorrelation bandwidth @xmath18 and scintillation timescale @xmath17 , and the sizes of the source in the 3 gates in the absence of averaging . the best - fitting decorrelation bandwidth is @xmath139 khz , and the best - fitting scintillation timescale is @xmath140 sec . these are comparable to those extrapolated from previous measurements , reported in the literature ( see [ obsintdist_sec ] ) . we can remove effects of time and frequency averaging to find the true sizes for the source in the 3 gates , without averaging in time or frequency , inferred from these fits . table [ table2_convfunc_fits_3best.out ] gives the results . figure [ 3best.gates123.1f2t.binvar_fig ] shows both the best - fitting size , and the distribution predicted for a source of zero size , averaged in frequency and time by the same amount . in gates 1 and 2 a model with size greater than zero fits remarkably better . this indicates the significance of the fitted size after correction for time averaging : about 10 standard deviations in gate 1 , and about 6 standard deviations in gate 2 . in gate 3 , a model with size zero fits about as well as the best - fit , with @xmath141 . we measure variations of as much as 3.5% in gain across the bands , as figure [ passband_fig ] shows ; and we find variations of gain between bands of as much as 3.0% , as shown in table [ table1_cont_hist_6_1f2t_002231 ] . these are the maximum departures ; the distribution is actually much more concentrated than these figures indicate . the discussion in [ variations_gain_flux_sec ] shows that this level of gain variations affects the analysis by a completely negligible amount . individual pulses from vela show significant variability ( @xcite ) . these intensity variations are almost completely random in time . a measure of this variability is the the intrinsic , single - pulse modulation index , @xmath142 , where @xmath143 is the flux of the pulsar in a single pulse . the index @xmath144 is 0.4 for the average intensity of the pulse . it varies from 0.3 to 1.3 in individual bins of @xmath145 in pulse longitude , with the largest variations only at the beginning and end of the pulse , where the average intensity is low ( @xcite ) . our time averaging over 10 s averages together about 112 pulses . our gates cover @xmath146 or @xmath147 of pulse longitude . the modulation index for an average of @xmath148 pulses , assuming that the measurements are uncorrelated , is @xmath149 . we thus expect the modulation index of our 112-pulse average to be no more than 0.15 , at the very most ; and we expect the overall average of @xmath150 to be closer to @xmath151 . we adopt 15% as a conservative upper bound for the net effect of gain variations and intrinsic variability . figure [ superposed_dist_fig ] indicates that , even if the distribution of intensity follows a flat distribution with this width , effects of gain variations and intrinsic variability might be barely detectable , over 3 orders of magnitude in the probability . they will not significantly affect the measurements of pulsar size . for our observations , the product of bandwidth and integration time ( including the duty cycle of the pulsar gate ) is @xmath152 for gates 1 and 2 , and @xmath153 for gate 3 . the average amplitude of the source , @xmath30 , is between 12.1 and 4.9 times the noise level , @xmath154 . in these cases , as figure [ selfnoisedist_fig ] indicates , self - noise has a negligible effect on the distribution , to our accuracy . in other words , the distributions are indistinguishable from those for @xmath155 . self - noise has negligible impact on the distribution function . normalization of the correlation detectably affects the distribution of amplitude for the vela pulsar . figure [ rhofit_fig ] shows the distributions of amplitude in the 3 gates , with a logarithmic vertical axis . a purely exponential distribution would follow a straight line on this plot , but the observed distribution falls below this line at both small and large amplitudes . the figure shows the best - fitting models including effects of normalization of the correlation ( eq . [ rhodist_eq ] ) for each gate , which accurately describe the distributions at large amplitude . table [ table3_rhofit_table ] summarizes results of the fits of the distribution expected for a normalized correlation function , given by eq . [ rhodist_eq ] , to the data in fig . [ rhofit_fig ] . to improve the sensitivity of the fit to the high end of the distribution , we fit to the logarithm of the histogram . the fitted parameters agree well with expectations . the normalization is greater than the number of data in the histogram , probably because of the reduction in the distribution at small amplitudes by effects of noise and finite source size . models including effects of finite size and noise ( eq . [ 3cdist ] ) accurately describe the distributions at low amplitudes , where these effects become important , as figure [ 3best.gates123.1f2t.binvar_fig ] shows . if we compare the parameters of the 2 distributions at the intermediate amplitudes where both take a nearly exponential form , using the approximations in eqs . [ erfc_approx ] and [ rhodist_eq_approx ] , unsurprisingly we find that they agree very well , as figure [ rhofit_fig ] would suggest . our analysis indicates a decreasing size across the pulse . table [ table4_size_summary ] summarizes our results , expressed both in terms of the size parameter , and as the fwhm of the best - fitting gaussian distribution , in km . to calculate the size in km from the results of the fit requires the additional parameters @xmath156 , @xmath157 , and @xmath28 , @xmath29 . the angular broadening of the pulsar is @xmath158 , with the major axis at a position angle of @xmath159 ( @xcite ) . we use comparison of our measured decorrelation bandwidth from [ avg_time_freq_sec ] , @xmath139 hz , with the angular broadening and the pulsar distance of @xmath160 pc ( @xcite ) to obtain the characteristic distance of scattering material ( @xcite , @xcite ) . we find that the fractional distance of the scattering material from the earth to the pulsar is @xmath161 . here recall that @xmath15 is the distance from observer to scatterer , and @xmath16 is the distance from scatterer to pulsar . we thus obtain the magnification @xmath162 . we then use the fitted values for the size parameter @xmath163 to find the size in km in each gate , given in table [ table4_size_summary ] . note that the uncertainty in the magnification factor @xmath157 dominates the quoted uncertainties of the size . the uncertainty in the magnification factor stems , in turn , primarily from uncertainty in the distance to the pulsar , the fitted size is in reasonable agreement with our earlier results for gate 1 ( @xcite ) . the major differences in the analysis are the accounting for effects of averaging in time and frequency , for the effects of spectrally - varying signals on noise , and our use of the measured decorrelation bandwidth to determine the magnification factor , in this paper . among the 4 classes of processes for pulsar emission discussed by melrose ( 1996 ) , the measured size of about 440 km rules out only models in which the observed emission comes from close to the polar cap . this region has a size of less than a km , much smaller than the observed size . if the emission originates at this location , but is transported to a larger region , such models are still viable . perhaps because pulsar radiation is collimated into a beam , many pictures of pulsar emission treat the radiation as emerging in only a single direction from each point of the emitting surface , perhaps along the local magnetic field . however , for the emission region to have the size we measure , the observer must receive radiation from points at the emission surface separated by @xmath164 km . the observer measures the separations between these points as the size of the emission region . figure [ weather_retard_fig ] illustrates this fact . of course , different emitting regions could contribute at different times , so that the emitting region is a point at any instant , and the finite size is observed only by averaging over many points . our measured size is averaged over 10 s of observations , or 112 pulses , and over a range of pulse phase . aberration can significantly affect the observed size of the emission region . at a fraction of the radius of the light cylinder , the pulsar s magnetosphere travels at a fraction of lightspeed . this can affect the interpretation of the measured size . in particular , the measured size can include the height of the emission region . figure [ weather_retard_fig ] shows an example . aberration is particularly important in models where radiation is emitted at a range of heights . models for emission at a small fraction of the radius of the light cylinder , @xmath165 km , agree naturally with our observation . for a simple dipole field , the open field lines have a cross - section of about 440 km at an altitude of about 450 km . however , sources that emit at a range of altitudes can have measured sizes that include effects of the spread in altitude , so that emission could arise at altitudes of 0 to 440 km , for example . models for emission from near or at the light cylinder could also produce the observed size , although in that case some physical process must limit the emission to a region less than 10% of the diameter of the light cylinder in size . the size , and its decrease across the pulse , are in approximate agreement with the results of krishnamohan & downs ( 1983 ) . they concluded that the emission early in the pulse arose from a larger region than that late in the pulse , and that the spread in altitude of emission was about 400 km . although their methodology was quite different , their picture of the emission region has remarkably similar dimensions . the measured size is large for `` core '' emission , which is believed to arise near the stellar surface ( @xcite ) , because it statistically reflects the opening angle of the dipole field lines at the polar cap . if vela s pulse does show core emission , as some suggest , this leads to an apparent paradox . the third gate , with size consistent with zero , might represent the core emission , with the leading edge of the pulse one side of a cone . studies of size of the emission region as a function of frequency , and imaging of the source , may resolve this question . measurements of size for other pulsars would also be extremely valuable . such observations are now in progress . arons & barnard ( 1986 ) pointed out that radiation traveling nearly along pulsars magnetic field lines , with polarization in the plane defined by the pulsar s magnetic field and the radiation s wavevector ( the `` o - mode '' ) , interacts with the outflowing electron - positron wind . this radiation can be `` ducted '' along field lines ( @xcite ) . in principle , such ducting can lead to a large apparent emission region even though the original source of radiation , perhaps near the pulsar s polar cap , is quite compact . lyutikov ( 1998b ) points out that microphysics can amplify , as well as refract , these waves . magnetospheric refraction takes place for only one polarization . the vela pulsar is heavily linearly polarized , so this picture and the size measurements suggests that the dominant polarization represents the o - mode . however , the less - dominant polarization mode is present , and should show zero size if magnetospheric refraction sets the size . thus , polarimetric size measurements suggest a simple test of whether magnetospheric refraction is responsible for the observed size . such measurements should be little more difficult than those described here , and are presently being pursued . we present measurements of the size of the radio emission region of the vela pulsar , from the observed distribution of correlation amplitude on a short interferometer baseline . in strong scintillation , this distribution is exponential if the source is pointlike , and is the sum of 3 exponentials if the source is small but extended ( @xcite ) . the effects of finite size on this distribution are concentrated at small amplitudes . we find that averaging in frequency and time has effects similar to those of finite size . we calculate the distribution including effects of gaussian noise in the observing system , when an interferometer observes the source . this effect is likewise greatest at the lowest amplitudes , with a different functional form . we calculate the expected effects of variations in pulsar flux density and in instrumental gain : these effects tend to make the distribution more sharply peaked at zero amplitude and flatter at high amplitudes , and are small unless the variations in gain approach 100% . we calculate the effects of self - noise , which likewise make the distribution function higher at low amplitudes and flatter at high amplitudes , and are important when the number of independent samples , @xmath61 , is small . we finally consider the effect of normalization of the correlation function , and find that this makes the distribution function fall more rapidly than the exponential extrapolated from small amplitudes , when the correlation approaches 100% . we compare model distributions of amplitude with that observed for the vela pulsar on the parkes - tidbinbilla baseline at @xmath166 cm . we measure the noise level , and the variation in gain with frequency channel , from observations of a strong continuum source . models including a finite size for the pulsar in all 3 gates provide good fits . we correct the fitted sizes for the effects of averaging in frequency and time . the fitted size parameters are significant at about 10 times the standard errors , depending on the gate . table[table2_convfunc_fits_3best.out ] summarizes the results . the size decreases across the pulse . we find that effects of gain and pulsar flux - density variations , self - noise , the van vleck correction , and averaging in time and frequency are expected to be undetectably small for our observation . effects of normalization of the correlation function are detected , but do not significantly affect the size estimates . the linear size of the pulsar s emission region is about 440 km in the first gate , 340 km in the second , and less than 200 km in the third . we discuss size measurements for sources that , like pulsars , radiate into a collimated beam . the measured size is the size of the region on the source visible to the observer ; that size may not be identical from all lines of sight , so that the measured size may easily change as the source rotates , as we observe . aberration effects may also affect the measured size , so that the range in altitude of the emission region , as well as its width , may contribute to the size . our measured size is much larger than the sizes of postulated polar - cap emission regions , so that such models , without additional physics , are ruled out for this pulsar . however , magnetospheric refraction can duct radiation from a compact emission region to produce a much larger size ( @xcite , @xcite ) . the observed size is on the order of postulated emission regions in the lower open - field - line region , suggesting that this is a likely location for pulsar emission . it is much smaller than the outer open - field - line region or the light cylinder , but emission might arise in only parts of these regions , or might be beamed toward the observer from only certain parts . arons , j. 1983 , apj , 266 , 215 arons , j. , & barnard , j.j . 1986 , apj , 302 , 120 asseo , e. 1996 , in iau colloq . 160 , pulsars : problems and progress , ed . s. johnston , m.a . walker , and m. bailes , san francisco : astronomical society of the pacific , p. 147 backer , d.c . 1974 , apj , 190 , 667 backer , d.c . 1975 , a&a , 43 , 395 barnard , j.j . , & arons , j. 1986 , apj , 302 , 138 bartz , m. 1993 , microwaves & rf , 32 , 5:192 blandford , r.d . , & scharlemann , e.t . 1976 , mnras , 174 , 59 britton , m.c . 1997 , phd thesis , university of california , santa barbara codona , j.l . , creamer , d.b . , flatt , s.m . , frehlich , r.g . , & henyey , f.s . 1986 , radio science , 21 , 805 cordes , j.m . , weisberg , j.m . , & boriakoff , v. 1983 , apj , 268 , 370 cordes , j.m . , weisberg , j.m . , & boriakoff , v. 1985 , apj , 288 , 221 cordes , j.m . 1998 , personal communication cornwell , t.j . , anantharamaiah , k.r . , & narayan , r. 1989 , josa , a 6 , 977 desai , k.m . , gwinn , c.r . , reynolds , j.r . , king , e.a . , jauncey , d. , flanagan , c. , nicolson , g. , preston , r.a . , & jones , d.l . 1992 , apj , 393 , l75 goldreich , p. , & julian , w.h . 1969 , apj , 157 , 869 goodman , j.w . 1985 , statistical optics , new york : wiley gwinn , c.r . , bartel , n. , cordes j.m . 1993 , apj , 410 , 673 gwinn , c.r . , ojeda , m.j . , britton , m.c . , reynolds , j.e . , jauncey , d.l . , king , e.a . , mcculloch , p.m. , lovell , j.e.j . , flanagan , c.s . , smits , d.p . , preston , r.a . , & jones , d.l . 1997 , apj , 483 , l53 gwinn , c.r . , britton , m.c . , reynolds , j.e . , jauncey , d.l . , king , e.a . , mcculloch , p.m. , lovell , j.e.j . , & preston , r.a . 1998 , apj , 505 , 928 gwinn , c.r . , et al . 1999 , in preparation . hankins , t.h . 1972 , apj , 177 , l11 jenet , f.a . , & anderson , s.b . 1998 , pasp , 110 , 1467 krishnamohan , s. , & downs , g.s . 1983 , apj , 265 , 372 lyne , a.g . , & manchester , r.n . 1988 , mnras , 234 , 477 lyne , a.g . , & graham - smith , f. 1998 , pulsar astronomy , 2nd ed . , cambridge : cambridge university press lyutikov , m. 1998a , mnras , 293 , 447 lyutikov , m. 1998b , mnras , 298 , 1198 manchester , r.n . 1995 , japa , 16 , 107 melrose , d.b . 1996 , in iau colloq . 160 , pulsars : problems and progress , ed . s. johnston , m.a . walker , and m. bailes san francisco : astronomical society of the pacific , p. 139 melrose , d.b . and stoneham , r.j . , 1977 , proc . astr . soc . australia , 3 , 120 meyer , s.l . 1975 , data analysis for scientists and engineers , new york : wiley press , w.h . , flannery , b.p . , teukolsky , s.a . , & vetterling , w.t . 1989 , numerical recipes , cambridge uk : cambridge univ . press radhakrishnan , v. , & cooke , d.j . 1969 , aplett , 3 , 225 radhakrishnan , v. , & rankin , j.m . 1990 , apj , 352 , 258 rankin , j.m . 1983 , apj , 274 , 333 rankin , j.m . 1986 , apj , 301 , 901 rankin , j.m . 1990 , apj , 352 , 247 rickett , b.j . 1975 , apj , 197 , 185 roberts , j.a . , & ables , j.g . 1982 , mnras , 201 , 1119 romani , r.w . , & yadigaroglu , i.a . 1995 , apj , 438 , 314 ruderman , m.a . & sutherland , p.g . 1975 , apj , 196 , 51 scheuer , p.a.g . 1968 , nature , 218 , 920 smirnova , t. v. , shishov , v. i. , malofeev , v. m. 1996 , apj , 462 , 289 strickman , m.s . , grove , j.e . , johnson , w.n . , kinzer , r.i . , kroeger , r.a . , kurfess , j.d . , grabelsky , d.a . , matz , s.m . , purcell , w.r . , ulmer , m.p . , & jung , g.v . 1996 , apj , 460 , 735 sturrock , p.a . 1971 , apj , 164 , 529 taylor , j.h . , manchester , r.n . , & lyne , a.g . 1993 , apjss , 88 529 thompson , a.r . , moran , j.m . , & swenson , g.w . 1986 , interferometry and synthesis in radio astronomy , ( new york : wiley ) weatherall , j.c . & eilek , j.a . 1997 , apj , 474 , 407 wolszczan , a. & cordes , j.m . 1987 , apj , 320 , l35 lrrr 1 & @xmath167 & @xmath168 & @xmath169 + 2 & @xmath170 & @xmath171 & @xmath172 + 3 & @xmath173 & @xmath174 & @xmath175 + 4 & @xmath176 & @xmath177 & @xmath178 + 5 & @xmath179 & @xmath180 & @xmath181 + 6 & @xmath182 & @xmath183 & @xmath184 + all & @xmath185 & @xmath186 & @xmath187 + [ table1_cont_hist_6_1f2t_002231 ] lcccc & & & & + number of data & & @xmath188&@xmath189&@xmath190 + bin width & & @xmath191&@xmath192&@xmath193 + assumed noise level & & @xmath194&@xmath195&@xmath196 + * fitted parameters * & & & & + size parameter & @xmath197&@xmath198&@xmath199&@xmath200 + amplitude & @xmath30 & @xmath201&@xmath202&@xmath203 + normalization & & @xmath204&@xmath205&@xmath206 + standard deviation of residuals & & @xmath207&@xmath208&@xmath209 + & & & & + size parameter & & & & + after correction for averaging&@xmath210&@xmath211&@xmath212&@xmath213 + [ table2_convfunc_fits_3best.out ]

solving polynomial systems arising from applications is frequently made easier by the structure of the systems . weighted homogeneity ( or quasi - homogeneity ) is one example of such a structure : given a system of weights ,-homogeneous polynomials are polynomials which are homogeneous w.r.t the weighted degree . grbner bases for weighted homogeneous systems can be computed by adapting existing algorithms for homogeneous systems to the weighted homogeneous case . we show that in this case , the complexity estimate for algorithm 5 can be divided by a factor . for zero - dimensional systems , the complexity of algorithm ( where is the number of solutions of the system ) can be divided by the same factor . under genericity assumptions , for zero - dimensional weighted homogeneous systems of-degree , these complexity estimates are polynomial in the weighted bzout bound . furthermore , the maximum degree reached in a run of algorithm 5 is bounded by the weighted macaulay bound , and this bound is sharp if we can order the weights so that . for overdetermined semi - regular systems , estimates from the homogeneous case can be adapted to the weighted case . we provide some experimental results based on systems arising from a cryptography problem and from polynomial inversion problems . they show that taking advantage of the weighted homogeneous structure can yield substantial speed - ups , and allows us to solve systems which were otherwise out of reach . , ,

1412.7547

algorithms for solving polynomial systems have become increasingly important over the past years , because of the many situations where such algebraic systems appear , including both theoretical problems ( algorithmic geometry , polynomial inversion ) , and real - life applications ( cryptography , robotics ) . examples of such algorithms include eigenvalues methods for systems with a finite number of solutions , or resultant calculations for polynomial elimination ( see @xcite for a survey ) . the theory of grbner bases is another tool which has proved useful for this purpose , and many algorithms for computing grbner bases have been described since their introduction . they include direct algorithms , computing the grbner basis of any system : to name only a few , the historical buchberger algorithm @xcite , and later the faugre 4 @xcite and 5 @xcite algorithms ; as well as change of order algorithms , computing a grbner basis of an ideal from another grbner basis : the main examples are the algorithm @xcite for systems with a finite number of solutions , and the grbner walk @xcite for the general case . systems arising from applications usually have some structure , which makes the resolution easier than for generic systems . in this paper , we consider one such structure , namely _ weighted homogeneous _ polynomials : a polynomial @xmath11 is weighted homogeneous with respect to a system of weights @xmath0 ( or @xmath1-homogeneous ) if and only if @xmath12 is homogeneous in the usual sense . moreover , in order to obtain precise results , we will assume that the systems satisfy some _ generic _ properties , which are satisfied by almost any system drawn at random . this is a usual assumption for grbner basis complexity estimates . more generally , we will also consider affine systems with a _ weighted homogeneous structure _ , that is systems whose component of maximal weighted degree will satisfy these generic properties . the complexity estimates given in this paper can be applied to a wide range of grbner basis algorithms . however , we mainly focus on two algorithms : matrix-5 , which is a matrix variant of 5 described in @xcite , allowing for complexity analyses , and . [ [ sec : previous - work ] ] prior work + + + + + + + + + + the special case @xmath13 is the usual homogeneous case . in this case , all the results from this paper specialize to known results . furthermore , some hypotheses are always satisfied , making the properties and definitions simpler . in particular , the description of the hilbert series of a homogeneous complete intersection is adapted from @xcite , and the asymptotics of the degree of regularity of a semi - regular sequence were studied in @xcite . weighted homogeneous systems have been studied before , from the angle of singularity theory and commutative algebra . in particular , some results about the hilbert series and the hilbert function of weighted homogeneous ideals , including the weighted bzout bound , can be found in most commutative algebra textbooks . the computational strategy for systems with a weighted structure is not new either , for example it is already implemented ( partially : only for weighted homogeneous systems with a degree order ) in the computer algebra system magma @xcite . additionally , the authors of @xcite proposed another way of taking into account the weighted structure , by way of the hilbert series of the ideal . the authors of @xcite generalized this algorithm to systems homogeneous with respect to a multigraduation . their definition of a system of weights is more general than the one we use in the present paper . to the best of our knowledge , nobody presented a formal description of a computational strategy for systems with a weighted homogeneous structure ( not necessarily weighted homogeneous ) , together with complexity estimates . some of the results presented in this paper about regular sequences previously appeared in a shorter conference paper @xcite , of which this paper is an extended version : these results are the weak form of the weighted macaulay bound and the formal description of the algorithmic strategy for weighted homogeneous systems , with the complexity estimates and . this conference paper lacked a hypothesis ( reverse chain - divisible systems of weights ) , and as such lacked the precise description of hilbert series required to obtain results for semi - regular sequences . the sharp variant of the weighted macaulay bound , under the assumption of simultaneous noether position , was also added in the present paper . finally , the benchmarks section of the current paper contains additional systems , arising in polynomial inversion problems . the conference paper was using _ quasi - homogeneous _ to describe the studied structure , instead of _ weighted homogeneous_. while both names exist in the literature , _ weighted homogeneous _ seems to be more common , and to better convey the notion that this structure is a generalization of homogenity , instead of an approximation . the same notion is sometimes also named simply _ homogeneous _ ( in which case the weights are determined by the degree of the generators ; see for example @xcite ) , or homogeneous for a _ nonstandard graduation _ @xcite . [ [ sec : main - results ] ] main results + + + + + + + + + + + + by definition , weighted homogeneous polynomials can be made homogeneous by raising all variables to their weight . the resulting system can then be solved using algorithms for homogeneous systems . however , experimentally , it appears that solving such systems is much faster than generic homogeneous systems . in this paper , we show that the complexity estimates for homogeneous systems , in case the system was originally @xmath1-homogeneous , can be divided by @xmath14 , where @xmath15 is the complexity exponent of linear algebra operations ( @xmath16 for naive algorithms , such as the gauss algorithm ) . these complexity estimates depend on two parameters of the system : its _ degree of regularity _ @xmath17 and its _ degree _ @xmath18 . these parameters can be obtained from the _ hilbert series _ of the ideal , which can be precisely described under generic assumptions . to be more specific , we will consider systems defined by a _ regular sequence _ [ def : regular ] ) and systems which are in _ simultaneous noether position _ [ def : snp ] ) . [ thm : intro - comp ] let @xmath0 be a system of weights , and @xmath19 a zero - dimensional @xmath1-homogeneous system of polynomials in @xmath20 $ ] , with respective @xmath1-degree @xmath21 . the complexity ( in terms of arithmetic operations in @xmath22 ) of algorithm 5 to compute a @xmath1-grevlex grbner basis of @xmath23 is bounded by @xmath24 if @xmath25 is a regular sequence ( and in particular @xmath26 ) , then @xmath17 can be bounded by the _ weighted macaulay bound _ : @xmath27 if additionally @xmath25 is in simultaneous noether position w.r.t the order @xmath28 , then the weighted macaulay bound can be refined : @xmath29 the complexity of algorithm to perform a change of ordering is bounded by @xmath30 if @xmath25 forms a regular sequence , then @xmath18 is given by the _ weighted bzout bound _ @xmath31 in particular , the bound indicates that in order to compute a grbner basis faster for a generic enough system , one should order the variables by decreasing weights whenever possible . the hypotheses of the theorem are not too restrictive . in the homogeneous case , regularity and simultaneous noether position are generic properties . however , in the weighted homogeneous case , there are systems of weights and systems of weighted degrees for which they are not generic . in this paper , we identify large families of systems of weights and systems of weighted degrees for which they are ( prop . [ thm : gen-2 ] ) . all sequences in simultaneous noether position are regular . in the homogeneous case , conversely , all regular sequences are in simultaneous noether position up to a generic linear change of coordinates . in the weighted homogeneous case , it is no longer true . worse still , there are systems of weights for which there exists no non - trivial change of coordinates . in order to work around this limitation , we consider _ reverse chain - divisible _ systems of weights , that is systems of weights such that @xmath32 . this property ensures that there are non - trivial change of coordinates of the form @xmath33 for all @xmath34 , with @xmath35 a @xmath1-homogeneous polynomial with @xmath1-degree @xmath36 . under this assumption , many properties from the homogeneous case remain valid in a weighted setting , and in particular , any regular sequence is , up to a @xmath1-homogeneous change of coordinates , in simultaneous noether position ( th . [ thm : gen - linchg ] ) . for many systems from practical applications , the weights can be chosen to be reverse chain - divisible . we give a few examples in the last section of this paper . if @xmath37 , there is no regular sequence . instead , we will consider systems defined by a _ semi - regular _ sequence , that is systems for which no reduction to zero appear in a run of algorithm 5 . this property has several equivalent definitions in the homogeneous case . while these definitions can be easily extended to the weighted case , their equivalence is not necessarily true . however , we prove that these definitions are equivalent in the special case where the weights form a reverse chain - divisible sequence . in the homogeneous case , the property of being semi - regular is only conjectured to be generic , but this conjecture is proved in a handful of cases ( ( * ? ? ? 1.5 ) . in this paper , we adapt the proof of one of these cases , namely the case @xmath38 in a base field of characteristic @xmath39 . for semi - regular systems with @xmath38 , we obtain a bound on the degree of regularity of the system . more generally , in the homogeneous case , one can compute asymptotic estimates on the degree of regularity of a semi - regular sequence @xcite . these estimates can be adapted to the weighted homogeneous case . as an example , we give an asymptotic bound on the degree of regularity for semi - regular systems with @xmath40 for a given integer @xmath41 : [ thm : intro - dreg ] let @xmath42 and @xmath41 be two positive integers , and let @xmath40 . let @xmath43 and @xmath44 be two positive integers such that @xmath45 . consider the system of @xmath42 weights @xmath46 . let @xmath25 be a semi - regular sequence in @xmath20 $ ] , made of @xmath1-homogeneous polynomials with @xmath1-degree @xmath44 . then the highest degree reached in the computation of a @xmath1-grevlex grbner basis of @xmath47 is asymptotically bounded by @xmath48 where @xmath49 is the largest root of the @xmath41th hermite s polynomial . experimentally , if we lift the assumption that the system of weights is reverse chain - divisible , the degree of regularity does not appear to rise too far beyond the bound . future work on the topic could include characterizing the hilbert series of @xmath1-homogeneous semi - regular sequences in full generality , in order to obtain bounds on the @xmath1-degree of regularity . in practice , taking advantage of the weighted structure when applicable yields significant speed - ups . some instance of a weighted structure has already been successfully exploited for an application in cryptography @xcite . we also present timings obtained with several polynomial inversion problems , with speed - ups ranging from 12 to almost 100 . in particular , we use these techniques in order to compute the relations between fundamental invariants of several groups ( see @xcite ) . for some groups such as the cyclic-5 group or the dihedral group @xmath50 , computing these relations is intractable without considering the weighted structure of the system , while it takes only a few seconds or minutes when exploiting the weighted structure . all these systems are examples of applications where the weights giving the appropriate @xmath1-homogeneous structure are naturally reverse chain - divisible . these experimentations have been carried using 5 and with the grbner basis library fgb @xcite and 4 with the computer algebra system magma @xcite . there are other applications where grbner bases are computed for polynomial systems with a weighted - homogeneous structure , for example in coding theory , both for generating codes ( @xcite , @xcite ) and for decoding through guruswami - sudan s algorithm ( see @xcite for an overview ) . [ [ sec : organisation - paper ] ] organisation of the paper + + + + + + + + + + + + + + + + + + + + + + + + + in section [ sec : notat - prev - results ] , we define weighted graded algebras and some generic properties of weighted homogeneous systems . in section [ sec : regular - systems ] , we focus on regular systems and complete intersections . we describe the hilbert series of a weighted homogeneous complete intersection and give the sharp variant of the weighted macaulay bound . in section [ sec : semi - regular - systems ] , we consider semi - regular systems . we give some equivalent definitions of this property , and we show how asymptotic estimates of the degree of regularity can be adapted from the homogeneous case to the weighted case . additionally , we prove that frberg s conjecture in the case @xmath38 is true in the weighted case , as in the homogeneous case , provided that the base field is large enough . in section [ sec : cons - comp ] , we describe strategies for computing grbner bases for weighted homogeneous systems , and we give complexity estimates for these strategies . finally , in section [ sec : applications ] , we show how weighted structures can appear in applications , and we give some benchmarks for each example . let @xmath22 be a field . we consider the algebra @xmath20 = { \mathbb{k}}[\mathbf{x}]$ ] . this algebra can be graded with respect to a system of weights , as seen for example in @xcite . let @xmath51 be a vector of positive integers . let @xmath52 be a vector of nonnegative integers . let the integer @xmath53 be the , or _ weighted degree _ of the monomial @xmath54 . we say that the vector @xmath1 is a _ system of weights_. we denote by @xmath55 the system of weights defined by @xmath56 , associated with the usual grading ( in total degree ) on @xmath57 $ ] . any grading on @xmath57 $ ] comes from such a system of weights @xcite . when working with a @xmath1-graduation , to clear up any ambiguity , we use the adjective _ @xmath1-homogeneous _ for elements or ideals , or _ weighted homogeneous _ if @xmath1 is clear in the context . homogeneous _ will be reserved for @xmath55-homogeneous items . the following property is an easy consequence of the definition . let @xmath58,w)$ ] be a graded polynomial algebra . then the application @xmath59,w ) & { \operatorname*{\longrightarrow}\limits } & ( { \mathbb{k}}[t_{1},\dots , t_{n}],\mathbf{1 } ) \\ & f & { \longmapsto } & f(t_{1}^{w_{1}},\dots , t_{n}^{w_{n } } ) \end{matrix}\ ] ] is an injective graded morphism , and in particular the image of a weighted homogeneous polynomial is a homogeneous polynomial . the above morphism also provides a weighted variant of the grevlex ordering ( as found for example in @xcite ) , called the @xmath1-grevlex ordering : @xmath60 given a @xmath1-homogeneous system @xmath25 , one can build the homogeneous system @xmath61 , and then apply classical algorithms @xcite to that system to compute a grevlex ( resp . lex ) grbner basis of the ideal generated by @xmath61 . the _ @xmath1-degree of regularity _ of the system @xmath25 is the highest degree @xmath62 reached in a run of 5 to compute a grevlex grbner basis of @xmath61 . when the graduation is clear in the context , we may call it degree of regularity , and denote it @xmath17 . unlike what we could observe in the homogeneous case , this definition depends on the order of the variables ( we shall give an example in table [ tab : dreg ] in section [ sec : degr - regul - quasi ] , and another , with timings , in table [ tab : dreg - timings ] in section [ sec : generic - systems ] ) . let @xmath63 be a zero - dimensional ( not necessarily weighted homogeneous ) ideal in @xmath64 $ ] . in that case , we define the degree @xmath6 of the ideal @xmath63 as the ( finite ) dimension of @xmath65 , seen as a @xmath22-vector space : @xmath66 equivalently , if @xmath67 is the hilbert series ( with respect to the @xmath1-graduation ) of @xmath63 , this series is a polynomial in @xmath68 and @xmath69 this definition with the hilbert series can be extended to ideals with positive dimension . however , in a weighted setup , varieties can end up having rational ( not - necessarily integer ) degrees . this is the definition used by the software macaulay2 ( * ? ? ? * function ` degree(module ) ` ) . we will only consider the _ varieties associated with the ideals we consider . in particular , the dimension of @xmath70 is @xmath42 , and a zero - dimensional variety is defined by at least @xmath42 polynomials if the base field is algebraically closed . [ def : regular ] let @xmath0 be a system of weights , let be a system of and let @xmath19 be a sequence of @xmath1-homogeneous polynomials in @xmath20 $ ] , with @xmath1-degree @xmath6 . the system @xmath25 is called _ regular _ if it satisfies one of the following equivalent properties @xcite : 1 . @xmath71/\langle f_{1},\dots , f_{i-1 } \rangle$ ] ; 2 . the hilbert series of @xmath47 is given by @xmath72 [ def : snp ] let @xmath1 be a system of weights . let @xmath73 and @xmath74 be a sequence of @xmath1-homogeneous polynomials in @xmath20 $ ] the system @xmath25 is said to be _ in noether position w.r.t the variables @xmath75 _ if it satisfies the two following properties : * for @xmath76 , the canonical image of @xmath77 in @xmath57/i$ ] is an algebraic integer over @xmath78 $ ] ; * @xmath78 \cap i= 0 $ ] . the system @xmath25 is said to be _ in simultaneous noether position _ ( or in snp ) if for any @xmath79 , the system @xmath80 is in noether position w.r.t the variables @xmath81 . the following proposition enumerates useful characterizations of the noether position . they are mostly folklore , but we give a proof for completeness . [ prop : carac - np ] let @xmath73 , @xmath1 be a system of weights and @xmath6 be a system of @xmath1-degrees . let @xmath19 be a sequence of @xmath1-homogeneous polynomials , with @xmath1-degree @xmath6 . the following statements are equivalent : [ item : np1 ] the sequence @xmath25 is in noether position w.r.t . the variables @xmath75 ; [ item : np2 ] the sequence @xmath82 is regular ; [ item : np3 ] the sequence @xmath83 is in noether position w.r.t . the variables @xmath75 ; [ item : np4 ] the sequence @xmath84 is regular . ( [ item : np1 ] @xmath85 [ item : np2 ] ) . @xmath86 [ item : np2 ] can be found in @xcite , we give it again here for completeness . ] let @xmath63 be the ideal generated by @xmath25 . the geometric characterization of noether position ( see e.g. @xcite ) shows that the canonical projection onto the @xmath87 first coordinates @xmath88 is a surjective morphism with finite fibers . this implies that the variety is zero - dimensional , and so the sequence is regular . ( [ item : np2 ] @xmath85 [ item : np1 ] ) . let @xmath76 , we want to show that @xmath89 is integral over the ring @xmath78 $ ] . since @xmath90 defines a zero - dimensional ideal , there exists @xmath91 such that @xmath92 with @xmath93 for the grevlex ordering with @xmath94 . by definition of the grevlex ordering , we can assume that @xmath95 simply belongs to @xmath63 . this shows that every @xmath77 is integral over @xmath96/i$ ] . we get the requested result by induction on @xmath34 : first , this is clear if @xmath97 . now assume that we know that @xmath98/i$ ] is an integral extension of @xmath78 $ ] . from the above , we also know that @xmath99 is integral over @xmath98 $ ] , and so , since the composition of integral homomorphisms is integral , we get the requested result . finally , we want to check the second part of the definition of noether position . assume that there is a non - zero polynomial in @xmath78 \cap i$ ] . since the ideal is weighted homogeneous , we can assume this polynomial to be weighted homogeneous . either this polynomial has degree 0 , or it is a non - trivial syzygy between @xmath100 . so in any case , it contradicts the regularity hypothesis . ( [ item : np2 ] @xmath85 [ item : np4 ] ) . for any @xmath101 , write @xmath102 . since any permutation of a regular sequence is a regular sequence , @xmath103 is a regular sequence , that is , for any @xmath79 , @xmath104 is not a zero divisor in @xmath105/\langle x_{m+1},\dots , x_{n},f_{1},\dots , f_{i-1 } \rangle\ ] ] as a consequence , factoring in the quotient by @xmath106 , @xmath107 is no zero - divisor in @xmath108/\langle f'_{1},\dots , f'_{i-1 } \rangle.\ ] ] ( [ item : np4 ] @xmath85 [ item : np2 ] ) . for any @xmath34 , write @xmath109 with @xmath110 $ ] , and @xmath111 . let @xmath112 . assume that @xmath113 : @xmath114 as a consequence , considering only the monomials in @xmath115 $ ] @xmath116 since @xmath84 is regular , @xmath117 : @xmath118 and indeed , @xmath104 is no zero - divisor in @xmath20/\langle x_{m+1},\dots , x_{n},f_{1},\dots , f_{i-1}\rangle$ ] . it means that @xmath103 is a regular sequence . by permutation , we conclude that @xmath119 is a regular sequence . ( [ item : np4 ] @xmath86 [ item : np3 ] ) . the sequence @xmath120}^{m}$ ] is regular if and only if the sequence @xmath121 is regular . the equivalence between [ item : np3 ] and [ item : np4 ] is then a mirror of the equivalence between [ item : np1 ] and [ item : np2 ] . let @xmath1 be a system of weights . several properties from the homogeneous case turn out to be no longer true in the weighted case . for example , properties such as the noether normalization lemma are no longer available , since in general , we can not write any non trivial weighted homogeneous change of coordinates . however , if we add some constraints on the system of weights , some of these properties can be proved in a weighted setting . more precisely , we will consider _ reverse chain - divisible _ systems of weights , defined as follows . we say that @xmath1 is _ reverse chain - divisible _ if we have @xmath122 in this situation , the weights are coprime if and only if @xmath10 . the name `` chain - divisible '' can be found in @xcite , referring to a notion introduced in @xcite . in this setting , many results from the homogeneous case can now be adapted to the weighted homogeneous case . for example , the noether normalization lemma states that for homogeneous polynomials with an infinite base field , all regular sequences are in noether position up to a generic linear change of coordinates . in the weighted homogeneous case with reverse chain - divisible weights , all regular sequences are in noether position , up to a weighted homogeneous change of coordinates , with @xmath1-degree @xmath1 . more precisely , in the weighted homogeneous case , we can prove the following version of the noether normalization lemma ( see ( * ? ? ? 13.2.c ) for the homogeneous version of this lemma ) : let @xmath22 be an infinite field , @xmath1 be a reverse chain - divisible system of weights and @xmath123 $ ] be a non - constant polynomial , @xmath1-homogeneous with @xmath1-degree @xmath124 . then there are elements @xmath125 such that @xmath126 is a finitely generated module over @xmath127 $ ] . furthermore , if the field has characteristic @xmath39 or large enough , there exists a dense zariski - open subset @xmath128 such that for all @xmath129 , one can choose @xmath130 . we follow the proof of ( * ? ? ? for any @xmath131 , let @xmath132 , and let @xmath130 . we need to show that for generic @xmath133 , under this change of variables , @xmath95 is monic in @xmath134 : @xmath135 so the set of all @xmath133 s such that @xmath95 is monic in @xmath134 is exactly the set of all @xmath133 s such that @xmath136 , and since @xmath95 is @xmath1-homogeneous non - constant , this is a non - empty open subset of @xmath137 . then , as in the homogeneous case ( * ? ? 13.3 ) , a consequence of this lemma is noether s normalization theorem , which we restate in a weighted setting : [ thm : gen - linchg ] let @xmath1 be a reverse chain - divisible system of weights , and let @xmath25 be a zero - dimensional regular sequence in @xmath20 $ ] . then , for a generic choice of @xmath1-homogeneous polynomials @xmath35 with @xmath1-degree @xmath36 , the change of variable @xmath138 is such that @xmath139 is in simultaneous noether position with respect to the order @xmath140 . another property of reverse chain - divisible weights is the following proposition . in the homogeneous case , if @xmath141 are two non - negative integers , then any monomial with degree @xmath142 is divisible by a monomial with degree @xmath143 . when the system of weights is reverse chain - divisible , the following proposition states a similar result for the weighted case . assume that @xmath0 is a system of weights , such that . the following statements are equivalent : 1 . [ item : rcd1 ] the system of weights @xmath1 is reverse chain - divisible ; 2 . [ item : rcd2 ] let @xmath141 positive integers , @xmath144 , and @xmath145 a monomial of @xmath1-degree @xmath142 . assume that @xmath36 divides @xmath143 , and that @xmath145 is not divisible by any of the variables @xmath146 . then there exists a monomial @xmath147 with @xmath1-degree @xmath143 , such that . ( [ item : rcd1 ] @xmath85 [ item : rcd2 ] ) . fix @xmath143 . we shall prove by induction over @xmath142 that for any monomial @xmath145 with @xmath1-degree @xmath142 satisfying the hypotheses of ( [ item : rcd2 ] ) , there exists a monomial @xmath147 with @xmath1-degree @xmath143 dividing @xmath145 . the case @xmath148 is trivial . assume that @xmath149 , and let @xmath145 be a monomial of @xmath1-degree @xmath142 . let @xmath150 be the greatest index of a variable dividing @xmath145 , write @xmath151 , where @xmath152 is a monomial in @xmath153 $ ] , with @xmath1-degree @xmath154 . if @xmath155 , the result follows by induction . if @xmath156 , then @xmath157 , and @xmath158 has @xmath1-degree @xmath143 and divides @xmath145 so we can assume that @xmath159 and that @xmath160 . since @xmath1 is a reverse chain - divisible system of weights , @xmath161 divides @xmath162 for any @xmath41 in @xmath163 . hence , since @xmath164 $ ] and @xmath165 $ ] , @xmath166 and @xmath167 . by hypothesis , @xmath143 is divisible by @xmath36 , and in particular it is divisible by @xmath161 . all in all , this shows that @xmath168 is divisible by @xmath161 , and so it is divisible by @xmath169 . let @xmath170 the monomial @xmath147 has @xmath1-degree @xmath143 and divides @xmath145 . ( [ item : rcd2 ] @xmath85 [ item : rcd1 ] ) . assume that @xmath1 is a system of weights which is not reverse chain - divisible , we shall find integers @xmath141 and a monomial @xmath145 with @xmath1-degree @xmath142 which is not divisible by any monomial of @xmath1-degree @xmath143 . since @xmath1 is not reverse chain - divisible , there exists @xmath34 such that @xmath171 does not divide @xmath36 . in particular , @xmath172 and @xmath173 . without loss of generality , we may consider only the variables @xmath174 . let @xmath175 , . by paoli s lemma ( see for example @xcite or the discussion after ( * ? ? ? 5.1 ) ) , there exists exactly @xmath176 couple of non - negative integers @xmath177 such that @xmath178 . let @xmath145 be the monomial @xmath179 . the @xmath1-degree @xmath143 is divisible by @xmath36 , and @xmath145 is not divisible by @xmath146 . the maximal divisors of @xmath145 are @xmath180 as a consequence , @xmath145 is not divisible by any monomial of @xmath1-degree @xmath143 . this proposition essentially states that the staircase of a @xmath1-homogeneous ideal is reasonably shaped when @xmath1 is a reverse chain - divisible system of weights . for example , let @xmath1 be a reverse chain - divisible system of weights , and let @xmath63 be the ideal generated by all monomials of @xmath1-degree @xmath181 ( that is , the least common multiple of the weights ) . then the proposition proves that @xmath63 contains all monomials of @xmath1-degree greater than @xmath181 . if on the other hand the system of weights is not reverse chain - divisible , this property needs not hold . for example , consider the algebra @xmath182 $ ] graded w.r.t . the system of weights @xmath183 , the least common multiple of the weights being @xmath184 , and let @xmath63 be the ideal generated by all monomials of @xmath1-degree @xmath184 . consider the monomial @xmath185 : it has @xmath1-degree @xmath186 , yet it is not divisible by any monomial with @xmath1-degree @xmath184 , and so it does not belong to the ideal @xmath63 . we shall give some results about the genericity of regularity and noether position for weighted homogeneous sequences . the fact that they define zariski - open subsets of the sets of sequences of a given weighted degree is classical . for regular sequences , see for example @xcite . the proof for sequences in ( simultaneous ) noether position is a simple extension of the statement for regular sequences . however , we provide here a sketch of these proofs for completeness . [ thm : gen-1 ] let @xmath73 be two integers , @xmath0 a system of weights , and @xmath187 a system of @xmath1-degrees . then * the set of regular sequences , * the set of sequences in noether position with respect to the variables @xmath188 , and * the set of sequences in simultaneous noether position w.r.t . the order @xmath189 are zariski - open subsets of the affine space of @xmath1-homogeneous polynomials with @xmath1-degree @xmath6 . we shall prove that regular sequences form a zariski - open subset of the affine space of @xmath1-homogeneous polynomials of @xmath1-degree @xmath6 . the openness of noether position will then be a corollary , since by proposition [ prop : carac - np ] , @xmath190 is in noether position w.r.t the variables @xmath75 if and only if @xmath119 is regular . as for sequences in simultaneous noether position , they will be given by the intersection of @xmath87 open subsets , stating that the sequences @xmath80 , @xmath101 , are in noether position w.r.t . the variables @xmath81 . let @xmath19 be a family of @xmath87 generic quasi - homogeneous polynomials , that is polynomials in @xmath191[\mathbf{x}]$ ] , whose coefficients are algebraically independent parameters @xmath192 . we want to prove that regular sequences are characterized by some polynomial in these coefficients @xmath192 being non - zero , which implies that they belong to a zariski - open set . write @xmath193 . since the hilbert series characterizes regular sequences , @xmath25 is regular if and only if @xmath63 contains all monomials of @xmath1-degree between @xmath194 and @xmath195 , where @xmath196 is given by @xmath197 . this expresses that a given set of linear equations has solutions , and so it can be coded as some determinants being non - zero , as polynomials in the coefficients @xmath192 . this states that the set of regular sequences , sequences in noether position and sequences in simultaneous noether position are zariski - dense subsets if and only if they are not empty . unfortunately , depending on the weights and the weighted degrees , there may exist no regular sequence , and thus no sequences in ( simultaneous ) noether position either . for example , let @xmath198 and @xmath199 , the only @xmath1-homogeneous sequence with @xmath1-degree @xmath6 in @xmath200 $ ] is ( up to scalar multiplication ) @xmath201 , and it is not regular . however , this is only the case for very specific systems of @xmath1-degrees , for which there does not exist enough monomials to build non - trivial sequences . let @xmath73 be two integers , @xmath0 a system of weights , and @xmath187 a system of @xmath1-degrees . we say that @xmath6 is _ @xmath1-compatible _ if there exists a regular @xmath1-homogeneous sequence in @xmath20 $ ] with @xmath1-degree @xmath6 . we say that @xmath6 is _ strongly @xmath1-compatible _ if for any @xmath79 , @xmath202 is divisible by @xmath36 . using these definitions , we can identify the cases where the properties of being regular , in noether position or in simultaneous noether position are generic . [ thm : gen-2 ] let @xmath73 be two integers , @xmath0 a system of weights , and @xmath187 a system of @xmath1-degrees . for any @xmath79 , write @xmath203 and @xmath204 . write @xmath205 the affine space of @xmath1-homogeneous sequences of @xmath1-degree @xmath6 . then the following statements are true : 1 . [ item : gen-1 ] if @xmath6 is @xmath1-compatible , then regular sequences form a zariski - dense subset of @xmath205 ; 2 . [ item : gen-2 ] if @xmath6 is @xmath206-compatible , then sequences in noether position with respect to the variables @xmath75 form a zariski - dense subset of @xmath205 ; 3 . [ item : gen-4 ] if @xmath6 is strongly @xmath1-compatible , then @xmath6 is @xmath1-compatible , @xmath206-compatible , and for any @xmath34 , @xmath207 is @xmath208-compatible ; 4 . [ item : gen-5 ] if @xmath26 , @xmath6 is @xmath1-compatible and @xmath1 is reverse chain - divisible , then , up to some reordering of the degrees , @xmath6 is strongly @xmath1-compatible . the proofs of statements [ item : gen-1 ] and [ item : gen-2 ] follow the same technique : by theorem [ thm : gen-1 ] , we know that the sets we consider are zariski - open in @xmath205 . so in order to prove the density , we only need to prove that they are non empty . statement [ item : gen-1 ] is exactly the definition of the @xmath1-compatibility . for statement [ item : gen-2 ] , by @xmath206-compatibility , we know that there exists a @xmath1-homogeneous sequence @xmath19 with @xmath1-degree @xmath6 in @xmath115 $ ] , which is regular . as a consequence , the sequence @xmath119 is regular , and from the characterization [ item : np4 ] of noether position ( prop . [ prop : carac - np ] ) , this means that @xmath25 is in noether position with respect to the variables @xmath75 . in order to prove statement [ item : gen-4 ] , we need to exhibit regular sequences of length @xmath34 in @xmath209 $ ] for any @xmath79 . for @xmath79 , write @xmath210 , it is regular and each polynomial lies in @xmath209 $ ] . finally , statement [ item : gen-5 ] is a consequence of theorem [ thm : gen - linchg ] . let @xmath1 be a reverse chain - divisible system of weights , and @xmath6 a @xmath1-compatible system of @xmath1-degrees . up to reordering , we can assume that the polynomials are ordered so that @xmath211 ; this does not cancel the @xmath1-compatibility . let @xmath212 be a regular sequence , @xmath1-homogeneous with @xmath1-degree @xmath6 . by theorem [ thm : gen - linchg ] , there exist polynomials @xmath213 which are @xmath1-homogeneous with @xmath1-degree @xmath36 , and such that @xmath25 , under the change of variables @xmath214 , is in simultaneous noether position with respect to the order @xmath140 . from the characterization [ item : np4 ] of noether position , that means in particular that for any @xmath144 , @xmath215 belongs to a regular sequence , and thus is not zero . and by definition of reverse chain - divisible weights , its @xmath1-degree @xmath202 is a sum of multiples of @xmath36 , and so it is itself a multiple of @xmath36 . the statement [ item : gen-5 ] is a converse of [ item : gen-4 ] in the reverse chain - divisible case . in the non - reverse chain - divisible case , that converse is false : let @xmath216 , @xmath217 and consider @xmath218 in @xmath200 $ ] . the sequence @xmath25 is in simultaneous noether position w.r.t . the order @xmath219 , yet @xmath220 is neither divisible by @xmath221 nor by @xmath222 . the weaker converse that if @xmath6 is @xmath1-compatible , then @xmath6 is @xmath206-compatible is also false : with the same weights and algebra , let @xmath223 , the only polynomial with @xmath1-degree @xmath220 is ( up to scalar multiplication ) @xmath224 . it is non - zero , so @xmath225 is a regular sequence , but @xmath226 is not regular , so @xmath225 is not in noether position w.r.t @xmath227 . these examples lead to the following attempt at writing a general characterization of @xmath1-compatibility . let @xmath42 be a positive integer , @xmath0 a system of weights , and @xmath228 a system of @xmath1-degrees . further assume that * for all @xmath144 , @xmath229}_{d_{i } } \neq 0 $ ] * the formal series @xmath230 is a polynomial . is @xmath6 necessarily @xmath1-compatible ? the answer is _ no _ : take the system of weights @xmath231 , and the system of @xmath1-degrees @xmath232 . note that @xmath233 is the product of the weights , and @xmath234 the sum of the weights . the series @xmath235 is a polynomial . but at @xmath1-degree @xmath234 , there are only @xmath222 monomials , namely @xmath236 and @xmath237 , and they are not coprime , so we can not form a regular sequence of @xmath1-degrees @xmath238 . let @xmath0 be a reverse chain - divisible system of weights such that @xmath10 , and let @xmath228 be a system of @xmath1-degrees , such that for any @xmath144 , @xmath202 is divisible by all of the @xmath169 s . let @xmath239 $ ] be a polynomial algebra graded with respect to @xmath1 . the proof of statements and can be found in ( * ? ? ? * lemma 2.1 ) . this proof depends only on the value of @xmath251 , and since we assume it to be @xmath252 , it is also valid in our setting . it also proves as a side - result . * if @xmath254 : @xmath255 * if @xmath256 , then @xmath257 which implies @xmath258 and @xmath259 ( from statement ) . now consider the possible values of @xmath260 : * * if @xmath261 , then @xmath262 , and thus @xmath263 . it implies that @xmath264 is even , we shall prove that it is greater than or equal to @xmath39 . + from statement , @xmath265 , so if @xmath266 , we are done . if @xmath267 , by parity @xmath268 is odd , and thus @xmath269 . + it implies that : @xmath270 * * otherwise , @xmath271 , and in that case @xmath272 which implies that : @xmath273 and @xmath274 so we have : @xmath275 recall that @xmath276 , so by parity , @xmath277 , hence @xmath278 . the following theorem is a description of the shape of the hilbert series of a zero - dimensional complete intersection . it states that it is a self - reciprocal ( or palindromic ) polynomial , that is a polynomial with symmetrical coefficients , and that these coefficients increase at small degrees , then station , then decrease again . furthermore , between every strict increase , they reach a step , which has width @xmath279 . for an example , see figure [ fig : shape - hs - reg ] , where the width of the steps is 3 , and the width of the central plateau is 5 . this is a generalization of a known result in the homogeneous case , which has been proved for example in ( * ? ? ? 2.2 ) ( we will follow that proof for the weighted case ) . in the homogeneous case , there is no such step in the growth of the coefficients , and they are strictly increasing , then stationary , then strictly decreasing . [ thm : hs - regseq ] let @xmath0 be a reverse chain - divisible system of weights , and @xmath228 a system of degrees such that for any @xmath144 , @xmath202 is divisible by @xmath181 . consider the formal series @xmath280 the series @xmath281 is a self - reciprocal polynomial in @xmath68 ( i.e. for any @xmath282 , @xmath283 ) and its coefficients satisfy the inequalities : @xmath284 furthermore , if @xmath285 ( resp . @xmath286 ) , the coefficients increase ( resp . decrease ) with steps , and these steps have width @xmath279 : @xmath287 / [ evaluate = as using ( ) * ( ) , evaluate = as using min ( , ) , count = i from 0 ] in 0 / 1 , 1 / 1 , 2 / 1 , 3 / 2 , 4 / 2 , 5 / 2 , 6 / 3 , 7 / 3 , 8 / 3 , 9 / 3 , 10 / 3 , 11 / 3 , 12 / 2 , 13 / 2 , 14 / 2 , 15 / 1 , 16 / 1 , 17 / 1 ( pi ) at ( , ) ; ( , ) ( , ) ( , ) ; we adapt the proof from ( * ? ? ? 2.2 ) for the homogeneous case to the weighted case . up to permutation of the @xmath202 s , we can assume that for any @xmath34 , @xmath294 . we proceed by induction on @xmath42 . the result for the case @xmath295 is a consequence of the homogeneous case , since @xmath10 . let @xmath296 . let @xmath297 and @xmath298 , and consider the series @xmath299 the hilbert series @xmath300 can be computed from @xmath301 with @xmath302 and so for any @xmath124 , we have : @xmath303 where @xmath304 this proves that the polynomial is self - reciprocal : @xmath305 * if @xmath308 , then from statement in lemma [ lem : lemms ] , and the definition of @xmath253 and @xmath309 , @xmath254 and @xmath310 . let @xmath311 , then @xmath312 , and thus : @xmath313 let @xmath314 , that implies that @xmath315 , and thus : @xmath316 * if @xmath317 , then from statement again , @xmath256 and @xmath318 . let @xmath319 , we want to prove that @xmath320 is greater or equal to zero , depending on whether @xmath124 is divisible by @xmath279 . we shall consider two ranges of values for @xmath124 : * * if @xmath321 , then @xmath322 ( since @xmath323 ) . recall that @xmath324 . by hypothesis , @xmath307 is divisible by @xmath279 , and so , either both @xmath124 and @xmath325 are divisible by @xmath279 , or both are not . thus , @xmath326 * * if @xmath327 , then @xmath328 ; by definition , @xmath329 , so @xmath330 ; furthermore , @xmath331 , so in the end : @xmath332 since by construction , @xmath333 is divisible by @xmath279 , the same reasoning as before yields that @xmath334 and @xmath335 + still assuming that @xmath307 , let now @xmath336 , we want to prove that @xmath337 . if @xmath338 there is nothing to prove , so assume that @xmath339 and @xmath340 . but then @xmath341 , and so by symmetry , @xmath342 . the hypothesis that the weights are reverse chain - divisible is necessary . as a counter - example , let @xmath343 and @xmath344 . then the hilbert series of a complete intersection of @xmath1-degree @xmath6 is illustrated in figure [ fig : cex - hs - reg-1 ] . it is self - reciprocal , but the coefficients do not vary as predicted by theorem [ thm : hs - regseq ] . the hypothesis that each of the @xmath1-degrees should be divisible by @xmath181 is also necessary . as a counter - example , let @xmath345 and @xmath346 . then the hilbert series of a complete intersection of @xmath1-degree @xmath6 is illustrated in figure [ fig : cex - hs - reg-2 ] : the width of the steps is greater than @xmath279 . furthermore , following the proof , the parameters for this series should be defined by @xmath347 and @xmath318 , where @xmath348 , so that @xmath349 and @xmath350 . however , we can not reorder the degrees such that @xmath351 , and we can not deduce from statement in lemma [ lem : lemms ] that @xmath352 : indeed , we have @xmath353 but @xmath354 . however , the fact that the hilbert series is self - reciprocal for complete intersections is true even for general system of weights , and is a consequence of the gorenstein property of complete intersections ( see ( * ? ? ? 21 ) ; this property is also central to the proof of theorem [ prop : froberg - n+1 ] ) . the degree of regularity of a zero - dimensional homogeneous regular system is bounded by macaulay s bound @xmath362 and , in practice , that bound is reached for generic systems . the proof of this result uses the degree of the hilbert series of the system . however , in the weighted case , the best result we can obtain from the degree of the hilbert series is @xcite : @xmath363 and this bound is not sharp in general . in particular , it appears that this degree of regularity depends on the order we set on the variables . the following theorem is an improvement over the previous bound , under the additional assumption that the system is in simultaneous noether position . recall that this property is generic , and that for reverse chain - divisible systems of weights , it is always true for regular sequences , up to a weighted homogeneous change of coordinates . [ thm : dreg ] let @xmath0 be a ( not necessarily reverse chain - divisible ) system of weights and @xmath228 be a strongly @xmath1-compatible system of @xmath1-degrees . further assume that for any @xmath364 , @xmath365 . let @xmath212 be a system of @xmath1-homogeneous polynomials , with @xmath1-degree @xmath6 , and assume that @xmath25 is in simultaneous noether position for the variable ordering @xmath366 . then the @xmath1-degree of regularity of @xmath25 is bounded by @xmath367 so assume that @xmath369 . we consider the system @xmath370 defined by : @xmath371 this system is @xmath372-homogeneous , for @xmath373 . from the characterization [ item : np3 ] of noether position , the sequence @xmath370 is in simultaneous noether position . as a consequence , the induction hypothesis applies to @xmath370 , and the @xmath372-degree of regularity of @xmath370 is bounded by @xmath374 denote by @xmath375 the degree of the hilbert series of @xmath25 , that is @xmath376 . we want to prove that @xmath377 , i.e. that the grbner basis of @xmath25 need not contain any polynomial with @xmath1-degree greater than @xmath378 . equivalently , let @xmath309 be a monomial with @xmath1-degree @xmath379 , we will prove that @xmath309 is strictly divisible by a monomial in the initial ideal generated by @xmath25 . * if @xmath383 , then @xmath384 $ ] . by assumption , @xmath385 , @xmath386 so @xmath309 has @xmath1-degree greater than @xmath387 , and by induction hypothesis , @xmath309 is strictly divisible by a monomial in the initial ideal generated by @xmath370 ; * if @xmath388 , then consider @xmath389 , it is a strict divisor of @xmath309 . furthermore , since @xmath390 , @xmath391 . recall that @xmath375 is by definition the degree of the hilbert series of the ideal generated by @xmath25 , so @xmath309 lies in that ideal . the hypothesis stating that for any @xmath34 , @xmath392 is necessary . for example , let @xmath393 , @xmath394 and the system @xmath395 in @xmath200 $ ] , it is @xmath1-homogeneous with @xmath1-degree @xmath6 and in simultaneous noether position . this system has hilbert series @xmath252 ( the quotient vector span is generated by @xmath396 ) , which has degree @xmath397 . but the grbner basis of the system is given by @xmath25 itself , and contains @xmath227 , with @xmath1-degree @xmath222 . more generally , without that hypothesis , we obtain the following bound for @xmath62 : @xmath398 and the proof is the same as that of theorem [ thm : dreg ] , with the weaker induction hypothesis that @xmath399 , which does not need inequality . we give examples of the behavior of both bounds in table [ tab : dreg ] : we give the degree of regularity of a generic @xmath1-homogeneous system of @xmath1-degree @xmath6 , and show how this degree of regularity varies if we change the order of the weights @xmath1 . theorem [ thm : dreg ] gives an indication as to how to choose the order of the variables . generically , in order to compute a @xmath1-grevlex grbner basis of the system , the complexity estimates will be better if we set the variables in decreasing weight order . let @xmath0 be a system of weights , and @xmath228 a strongly @xmath1-compatible system of @xmath1-degrees . let @xmath400 $ ] be a generic system of @xmath1-homogeneous polynomials . let @xmath401 be the degree of the hilbert series of @xmath47 , and let @xmath44 be defined as @xmath402 where @xmath403 is the frobenius number of @xmath1 ( that is , the greatest @xmath1-degree at which the set of monomials is empty ) . in other words , @xmath44 is the degree of the first `` unexpected '' zero coefficient in the hilbert series ( by definition of the degree in the first case , and by self - reciprocality of the hilbert series in the second case ) . & speed - up + cyclic invariants , @xmath405 & 4.0 & 0.3 & 13.3 + cyclic invariants , @xmath406 , @xmath407 & 2705.8 & 73.4 & 36.9 + cyclic invariants , @xmath406 & > 90000^b^ & 370.0 & > 243 + cyclic invariants , @xmath408 , @xmath409 & 35922.4 & 2256.2 & 15.9 + cyclic invariants , @xmath408 , @xmath410 & > 300000^b^ & 7426.7 & > 40 + dihedral invariants , @xmath406 & > 40000^b^ & 18.5 & > 2162 + generic monomials , @xmath411 , @xmath412 , @xmath413 & 216.5 & 110.9 & 2.0 + generic monomials , @xmath411 , @xmath414 , @xmath415 & 31135.2 & 16352.2 & 1.9 + generic monomials , @xmath411 , @xmath416 , @xmath417 & 14919.2 & 8142.8 & 1.8 + generic monomials , @xmath411 , @xmath418 , @xmath419 & 5645.8 & 4619.0 & 1.2 + generic monomials , @xmath420 , @xmath421 , @xmath422 & 370.1 & 193.1 & 1.9 + generic monomials , @xmath420 , @xmath423 , @xmath424 & 4527.2 & 2904.6 & 1.6 + matrix minors , @xmath425 , @xmath426 , @xmath427 & 41220.0 & 26340.0 & 1.6 + matrix minors , @xmath425 , @xmath428 , @xmath427 & 48000.0 & 18060.0 & 2.7 + matrix minors , @xmath429 , @xmath428 , @xmath427 & 711690.0 & 390235.0 & 1.8 + matrix minors , @xmath405 , @xmath430 , @xmath431 & 613.9 & 325.4 & 1.9 + matrix minors , @xmath405 , @xmath426 , @xmath432 & 8059.4 & 3955.5 & 2.0 + matrix minors , @xmath405 , @xmath433 , @xmath434 & 71067.8 & 32721.5 & 2.2 + . & & & bardet , m. , dec . tude des systmes algbriques surdtermins . applications aux codes correcteurs et la cryptographie . phd thesis , universit pierre et marie curie - paris vi . http://tel.archives-ouvertes.fr/tel-00449609 bardet , m. , faugre , j .- c . , salvy , b. , yang , b .- y . , 2005 . asymptotic behaviour of the degree of regularity of semi - regular polynomial systems . in : mega05 , 2005 . eighth international symposium on effective methods in algebraic geometry . becker , t. , weispfenning , v. , 1993 . grbner bases . 141 of graduate texts in mathematics . springer - verlag , new york , a computational approach to commutative algebra , in cooperation with heinz kredel . http://dx.doi.org/10.1007/978-1-4612-0913-3 bosma , w. , cannon , j. , playoust , c. , 1997 . the magma algebra system . i. the user language . journal of symbolic computation 24 ( 3 - 4 ) , 235265 , computational algebra and number theory ( london , 1993 ) . http://dx.doi.org/10.1006/jsco.1996.0125 caboara , m. , de dominicis , g. , robbiano , l. , 1996 . multigraded hilbert functions and buchberger algorithm . in : engeler , e. , caviness , b. f. , lakshman , y. n. ( eds . ) , proceedings of the 1996 international symposium on symbolic and algebraic computation , issac 96 , zurich , switzerland , july 24 - 26 , 1996 . acm , pp . 7278 . http://doi.acm.org/10.1145/236869.236901 dalzotto , g. , sbarra , e. , 2006 . computations in weighted polynomial rings . analele stiintifice ale universitatii ovidius constanta 14(2 ) , 3144 . de boer , m. , pellikaan , r. , 1999 . grbner bases for codes . no . 4 in algorithms and computation in mathematics . springer , pp . http://www.win.tue.nl/~{}ruudp/paper/34.pdf[http://www.win.tue.nl/~\{}ruudp/paper/34.pdf ] faugre , j .- c . , 1999 . a new efficient algorithm for computing grbner bases @xmath435 . journal of pure and applied algebra 139 ( 1 - 3 ) , 6188 , effective methods in algebraic geometry ( saint - malo , 1998 ) . http://dx.doi.org/10.1016/s0022-4049(99)00005-5 faugre , j .- c . , 2002 . a new efficient algorithm for computing grbner bases without reduction to zero @xmath436 . in : proceedings of the 2002 international symposium on symbolic and algebraic computation . acm , new york , pp . 7583 ( electronic ) . http://dx.doi.org/10.1145/780506.780516 faugre , j .- c . , gianni , p. , lazard , d. , mora , t. , 1993 . efficient computation of zero - dimensional grbner bases by change of ordering . journal of symbolic computation 16 ( 4 ) , 329344 . faugre , j .- c . , safey el din , m. , verron , t. , 2013 . on the complexity of computing grbner bases for quasi - homogeneous systems . in : proceedings of the 2013 international symposium on symbolic and algebraic computation . issac 13 . acm , new york , ny , usa . faugre , j .- c . fgb : a library for computing grbner bases . in : fukuda , k. , hoeven , j. , joswig , m. , takayama , n. ( eds . ) , mathematical software - icms 2010 . 6327 of lecture notes in computer science . springer berlin / heidelberg , berlin , heidelberg , pp . http://www-polsys.lip6.fr/~jcf/papers/icms.pdf faugre , j .- c . , gaudry , p. , huot , l. , renault , g. , 2013 . using symmetries in the index calculus for elliptic curves discrete logarithm . journal of cryptology , 141 . gaudry , p. , 2009 . index calculus for abelian varieties of small dimension and the elliptic curve discrete logarithm problem . journal of symbolic computation 44 ( 12 ) , 1690 1702 , grbner bases in cryptography , coding theory , and algebraic combinatorics . http://www.sciencedirect.com/science/article/pii/s074771710800182x guerrini , e. , rimoldi , a. , 2009 . fglm - like decoding : from fitzpatrick s approach to recent developments . in : sala , m. , sakata , s. , mora , t. , traverso , c. , perret , l. ( eds . ) , grbner bases , coding , and cryptography . springer berlin heidelberg , pp . http://dx.doi.org/10.1007/978-3-540-93806-4_12 leonard , d. a. , 2009 . a weighted module view of integral closures of affine domains of type i. advances in mathematics of communications 3 ( 1 ) , 111 .

we investigate the role of radiative driving of shock ejected bipolar outflows from advective accretion discs in a self consistent manner . radiations from the inner disc affects the subsonic part of the jet while those from the pre - shock disc affects the supersonic part , and there by constitutes a multi stage acceleration process . we show that the radiation from the inner disc not only accelerate but also increase the mass outflow rate , while the radiation from the pre - shock disc only increases the kinetic energy of the flow . with proper proportions of these two radiations , very high terminal speed is possible . we also estimated the post - shock luminosity from the pre - shock radiations , and showed that with the increase of viscosity parameter the disc becomes more luminous , and the resulting jet simultaneously becomes faster . this mimics the production of steady mildly relativistic but stronger jets as micro - quasars moves from low hard to intermediate hard spectral states . [ firstpage ] hydrodynamics , radiation hydrodynamics , black hole physics , accretion , accretion discs , jets and outflows

1311.0264

jets are ubiquitous and are observed to accompany a variety of astrophysical objects such as , stellar - mass and super - massive black hole candidates , neutron stars , white dwarfs , young stellar objects ( ysos ) etc . these outflows exhibit different physical scales and power at one extreme , agns have jets with typical sizes @xmath0 few kpc to few @xmath1 kpc , jet velocities comparable to @xmath2 ( where @xmath2 is the light speed ) , luminosity range @xmath3erg / s and central mass in the range @xmath4 ( where @xmath5 is the mass of sun ) , while in the other extreme , yso jets have typical size @xmath6pc , outflow velocity @xmath7 , luminosity range @xmath8erg / s , and emerge from protostars with mass @xmath9 . in other words , only jets around compact objects are truly relativistic . jets around black hole candidates , be it from agns or micro - quasars , are shrouded in mystery . since black holes do not have hard surface and/or any atmosphere , therefore jets can only originate from the matter accreting onto it . moreover , the terminal speed of jets from around compact objects , though relativistic , but can vary widely too . for jets around grs 1915 + 105 or m87 exhibit terminal speed above ninety percent the speed of light @xcite , while the jet around ss433 is merely around @xmath10 @xcite . therefore , not only there is no consensus about the origin of jets , even the acceleration mechanism of jets are not well understood . there are few interesting properties of jets around compact objects . @xcite showed that the jet around m87 seems to originate within a region less than @xmath11 ( @xmath12 is the schwarzschild radius ) around the central object . in other words , entire accretion disc do not participate in generation of jets , but only the inner region of the accretion disc participate in jet generation . although the connection between jet states and spectral states of the accretion disc has not been conclusively established for massive black hole candidates like the agns , but for micro - quasars , this connection has firmly been established @xcite . persistent , quasi steady , mildly relativistic jets are observed in ` hard spectral state ' ( maximum power in the hard power law tail ) of the accretion disc . the jet seems to get stronger as the spectral state of accretion disc moves to the intermediate states . and truly relativistic jet blobs are generated during state transition to the steep power law state . no jet activity is observed in the canonical soft state ( maximum power in modified black body component ) @xcite . such close corelation of jet with the radiative states of the disc , points to the fact that the accretion disc physics is responsible for jet generation . matter accreting onto a compact object should possess some angular momentum and due to differential rotation some form of anomalous viscosity too . the first viscous disc model seriously considered by the community is the standard thin disc @xcite , although thin disc s inability to explain the origin of power - law photons , as well as , the theoretical inconsistencies like adhoc inner boundary condition , remained an overbearing concern . it was understood that a source of hot electron distribution a comptonizing corona , is required to explain the hard power law photons in the spectra of black hole candidates @xcite . since the boundary condition of black hole accretion is necessarily transonic , disc models with significant advective term gained popularity . the most popular model among the advective disc models was known as adaf or advection dominated accretion flow @xcite . this model is characterized by a single sonic point close to the horizon and subsonic elsewhere . however , it has been shown earlier that multiple sonic point may exist for inviscid rotating flow @xcite , and was confirmed that this is also true for dissipative advective accretion discs @xcite . and therefore it was shown by various authors that adaf is only a subset of general advective solution @xcite . general advective solutions which admits multiple sonic points may harbour standing or time - dependent shock solutions @xcite . shock in accretion has some interesting consequences , for the post shock hot electrons may inverse comptonize soft photons to produce the non - thermal hard radiation tail @xcite . a dominant shock associates with a low supply of soft photons from external keplerian disc , and the hot post - shock region inverse - comptonizes the intercepted photons and produces the low / hard state . and an increased supply of keplerian matter means supply of extra soft photons to cool down the post - shock region which results in weakening of , or , complete removal of the shock to produce the canonical high / soft state . in fact , the hardness intensity diagram ( hid ) of gro j 1655 - 40 was well explained by shocked accretion disc @xcite . these studies also showed that the steady state shock solutions are possible in a limited range of the parameter space , while oscillating or time dependent shocks are possible for a wide range of parameters or boundary conditions @xcite . and since these post - shock region also emits high energy radiations , therefore quasi - periodic oscillation of the shocked disc will give rise to quasi - periodic oscillation in hard radiations as well , and was furthered as a model for quasi - periodic oscillations or qpos @xcite . it was shown from observations that the spectral index increases ( hard to soft transition ) with the increase of qpo frequency , but remarkably the spectral index saturates ( indicating the attainment of high / soft state ) and then there are no qpo detected @xcite . the evolution of the qpo frequency with the spectral state for outburst sources like xtej1550 - 654 , gro 1655 - 40 etc , has been explained well with a model of inward drift of oscillating shock which translates into a spectral state transition from low / hard to intermediate hard . and the final disappearance of qpo with the disappearance of the inner disc or the post - shock disc @xcite . and it has been shown that indeed for the same outer boundary condition , the shock drifts inwards with the increase of the viscosity parameter @xcite . this implies that , with the increase of viscosity the pre - shock disc size increases , which increases the supply of soft photons . as the post - shock disc gets reduced , eventually the total output of the thermal comptonization ( dominant spectra in low / hard state ) decreases too . and since according to the advective accretion disc model , oscillation of post - shock disc gives rise to qpo , so the decrease of shock location increases the qpo frequency , until the shock itself disappears . hence the transition from hard to soft state , or in other words , the increase of spectral index should be correlated with the increase of qpo frequency , and finally the spectral index saturates @xcite as the contribution to the radiation from post - shock disc becomes negligible @xcite . in our previous paper @xcite , we have presented all possible advective viscous accretion solutions , including shocked and shock free solutions . a shocked accretion disc is more interesting , because the post - shock disc being hot can drive bipolar - outflows @xcite . since shock forms typically at @xmath13few @xmath14 , so the assertion that jet base is the inner part of the disc , favours the observational evidence from m87 that jet base is indeed @xmath15 @xcite . since it has also been shown that agns are just a scaled up version of the galactic black holes @xcite , therefore , following the above evidence one may argue that even for galactic black holes the jet base should be close to the central object . however , it is not just theoretical expectation that the inner region around agns and micro - quasars should be similar , therefore , the jet base for microquasars too would be close to the horizon . even in case of galactic black hole candidates ( bhc ) , the detection of strong radio flares ( read jets ) with simultaneous disappearance of qpos and the absence of the comptonized component , points out to the fact that the same region which gave rise to the comptonized component and generated the qpo has been ejected as relativistic jets @xcite . the shocked accretion disc model seems to satisfy all these criteria , starting with the direct evidence of m87 jet originating close to the black hole , the connection between growth of qpo frequency , comptonized component and the radio flare , to the disappearance of jet activity in the soft state ( no or very weak shock ) . @xcite showed ( fig . 5 of their paper ) that the correlation between thermally driven jets from shocked accretion discs and the spectral states from the same discs seems to follow the conclusions of observations of 10 microquasars @xcite . since post - shock disc produces the jets therefore when shock is absent or very weak , jets disappear mimicking the soft state . but thermally driven jets can achieve terminal speed up to @xmath16^{1/2}$ ] , where @xmath17 is the sound speed at the base , which for a shocked disc is the sound speed of the post - shock disc , and @xmath18 is the gravitational potential at the jet base . now the maximum sound speed physically possible is @xmath19 , and that the value of @xmath18 is quite significant because the jet base is rather close to the horizon @xcite . given these facts , the expression of terminal speed given above indicates that , truly relativistic jet terminal speed is not possible by only thermal driving . furthermore , jet states are correlated with the hard spectral states of the accretion disc , therefore , can the disc radiation accelerate the outflowing jet material ? in this respect , one may raise the issue that jet activities are not always observed during the state transition , which might be due to the lack of availability of simultaneous x - ray and radio / infra - red measurements . however , @xcite studied about 10 sources to draw the correlation between jet activity with the hard states , and the conclusion that jets are not seen during the canonical soft states . we would therefore like to study the interaction of radiations from the hard / hard intermediate states with the emanating jet , or in our parlance , the acceleration of jets with the radiation from a shocked disc , and how this radiative acceleration of jets is influenced with the change in viscosity parameter of the disc . it is to be remembered that , since we are investigating the disc in steady state , we are actually studying the generation and acceleration of steady , mildly relativistic jets generally associated with hard to hard intermediate states . the strong radio flares associated with relativistic blob ejections during hard intermediate to soft intermediate state transition @xcite is intrinsically a time dependent phenomenon and has not been addressed in this paper . since we have not incorporated all the physical aspects to mimic the transient phenomena , in this limited sense the results depend on the model assumptions . interaction of the disc radiation with the outflowing matter has been followed by many authors . @xcite studied radiative acceleration of jets above a keplerian disc , but in absence of radiation drag . @xcite showed that radiation drag is important for jets powered by radiation from a thick accretion disc @xcite . including radiation momentum deposition on an axial particle jet illuminated by an infinite thin disc in presence of drag term , @xcite showed that the upper limit of the terminal speed is around @xmath20 of @xmath2 . investigations on radiatively driven jets were extended by @xcite to generate relativistic terminal speeds for particle jets . moreover , the interaction of radiations from shocked accretion disc and jets were studied too @xcite . since shocked discs have two radiation sources , namely , the post - shock and pre - shock disc , therefore the redistribution of radiative power amongst these sources , could efficiently accelerate the jet around its axis . however , jets studied in the above mentioned papers were decoupled from the disc . in this paper we would like to study radiatively driven jet which is generated self consistently from the underlying accretion disc . in other words , we combine the accretion - ejection physics described by @xcite with the effect of radiative momentum deposition on the jets as shown by @xcite . in the next section , we present the simplifying assumptions and equations of motion . in section 3 , we present the methodology of solution . and in the last section we present the solutions and discussion . in this paper , we have assumed the axis - symmetric disc - jet system to be in steady state . the black hole is assumed to be non - rotating described by the pseudo - newtonian potential introduced by @xcite . the viscosity prescription in the accretion disc is described by the shakura - sunyaev @xmath21 viscosity . the jets are tenuous and should have less differential rotation than the accretion disc , as a result the viscosity in jets can be ignored . to negate any resulting torque , the angular momentum at the jet base is assumed same as that of the local value of angular momentum of the disc . we estimate the radiative moments , namely , the radiative energy density , flux , and radiative pressure from the disc . the jets being optically thin are subjected to radiation force including the drag terms . figure 1 , shows the schematic diagram of the disc - jet system , where the pre - shock , post - shock disc and the jet are clearly marked . the black hole is at bh . the accretion disc occupies the space on or about the equatorial plane , and the jet flow geometry is about the axis of symmetry . we have used the geometric unit system where , @xmath22 ( @xmath23 is the mass of the black hole , @xmath24 is the gravitational constant ) . therefore , in this representation the unit of length is , the schwarzschild radius or @xmath12 , the unit of mass is that of the central black hole @xmath23 , and that of time is @xmath25 , respectively , consequently the unit of speed is @xmath2 . since we are investigating self - consistent accretion - ejection system , we should present the governing equations together . however , since the accretion and jets follow different flow geometries , we present them separately . in section 2.1.1 , we present equations of motions , which when solved gives us the accretion solution . in section 2.1.2 , the equations governing the jet solution are presented . and finally in section 2.2 , we present the solution procedure , how a jet solution is simultaneously obtained from accretion disc through shocks . the equations of motion of the accreting matter around the equatorial plane , in cylindrical coordinates ( @xmath26 , @xmath27 , @xmath28 ) , have been extensively presented in the literature @xcite . here we give a brief account of them . these equations are the radial momentum equation , the accretion rate equation for a disc in vertical equilibrium , the angular momentum conservation equation and the entropy generation equation , and they are @xmath29 @xmath30 @xmath31 @xmath32 the local variables @xmath33 and @xmath34 in the above equations are the radial velocity , isotropic pressure , density and specific angular momentum of the flow , respectively . here , @xmath35 is vertically integrated density with @xmath36 being the local disc height from equatorial plane and @xmath37 is the viscous stress tensor . moreover , @xmath38 , @xmath39 are the entropy density and the local temperature respectively . the local heat gained ( due to viscosity ) and lost by the flow are given by @xmath40 and @xmath41 . the viscous stress , the local half height and the sound speed are given by @xmath42 where , @xmath43 is the dynamic viscosity coefficient , @xmath44 is the kinematic viscosity , @xmath21 is the shakura - sunyaev viscosity parameter , @xmath45 , @xmath46 and @xmath47 are the local angular velocity , local keplerian angular velocity and adiabatic index , respectively . here , all the variables have been made dimensionless by employing the unit system mentioned above section 2.1 , if @xmath48 , @xmath49 etc are dimensional radial coordinate and velocity , then @xmath50 , @xmath51 etc . integrating equation ( [ rme.eq ] ) with the help of eq . ( [ mf.eq]-[ege.eq ] ) , we find another constant of motion called the specific grand energy @xcite of the flow and is given by @xmath52 where , a , @xmath53 are local sound speed and the specific angular momentum of the flow on the horizon which is one of the constants of integration . using the adiabatic law , the definition of sound speed , and the mass - accretion equation ( eq . [ mf.eq ] ) , entropy - accretion rate is defined @xmath54 where @xmath55 . @xmath56 is the measure of local entropy and is constant for an inviscid and adiabatic flow . we rearrange and simplify eqs . ( [ rme.eq]-[ege.eq ] ) to the following forms ( see , kumar & chattopadhyay 2013 for details ) , @xmath57 since accreting matter at @xmath58 is subsonic , and close to the horizon it is supersonic , at some point in between , the flow would become transonic , where @xmath59 . this gives the so - called sonic point conditions or critical point conditions . and at @xmath60 or sonic point , the gradient is calculated by employing lhospital rule . the accretion solutions are obtained by solving eqs . ( [ grad.eq ] ) with the help of sonic point conditions . the jet streamline is approximated from the numerical simulations @xcite , where the jet flows through the funnel wall ( fw ) and the centrifugal barrier ( cb ) . a detail jet geometry has been described in @xcite . all the jet variables in the equations below have been made dimensionless by dividing the distances with @xmath61 , velocities with @xmath2 , mass with @xmath23 etc . the disc photons accelerate the jet by depositing the radiation momentum onto the jet . since the jet is optically thin so the radiation actually penetrates the jet . so every fluid parcel of the jet is accelerated by the radiation from behind , but the radiation field ahead also drags and decelerate the parcel . the momentum balance equation of the jet , correct up to first order in velocity , is given by @xcite @xmath62 here , @xmath63 , and @xmath64 , where , @xmath65 is the thompson scattering cross section , @xmath66 is the electron mass , and @xmath67 , @xmath68 , @xmath69 are radiative flux , radiative energy density and radiative pressure , respectively . the streamline is computed as @xmath70 . the first term in r. h. s of eq . ( [ radmjet.eq ] ) is the acceleration term , while negative terms are the radiative drag term . all the terms with the suffix ` j ' represents jet quantities . if @xmath71 , then eq . ( [ radmjet.eq ] ) can be integrated to give the bernoulli equation for jet , @xmath72 for purely thermally driven flow @xmath73 is a constant of motion , however , for a radiatively driven flow @xmath73 is a variable , as we will show later ( figs . 4e , 5e ) . the jet outflow rate is given by @xmath74 where @xmath75 is the cross - sectional area of the jet . since absorption and emission terms in the jet is zero , therefore , it is described by the polytropic equation of state @xmath76 . the entropy accretion rate ( from eq . [ jmf.eq ] ) for the jet is given by @xmath77 @xmath78 is a measure of entropy of the jet , and is constant along the jet since heating and cooling is ignored . equations ( [ radmjet.eq],[entroj.eq ] ) are simplified , to obtain , @xmath79 where , @xmath80 and , @xmath81 the sonic point ( @xmath82 ) condition for jet is obtained for the condition @xmath83 , @xmath84 , @xmath85 where , @xmath86,~~ a_2=-({\cal{e}}_{rj}+{\cal{p}}_{rj } ) , ~~ a_3=\left[{\cal{f}}_{rj}-\frac{1}{2(r_{jc}-1)^{2}}\left(\frac{dr_{j}}{dr}\right)_{r_{c } } + \frac{\lambda_{jc}^{2}}{x_{jc}^{3}}\left(\frac{dx_{j}}{dr}\right)_{r_{c } } \right].\ ] ] the gradient of jet velocity on the sonic point is obtained by lhospital rule @xmath87_c=\left[\frac{d{\cal n}/dr}{d{\cal d}/dr}\right]_c . \label{dvjc.eq}\ ] ] to obtain the jet solutions one has to integrate eq . [ gvjet.eq ] , with the help of eqs . ( [ jetc.eq]-[dvjc.eq ] ) . in this paper we have chosen the adiabatic index to be @xmath88 , and has been shown to be the typical value close to the black hole @xcite . the jet solution is dictated by the accretion solution , and now we discuss how we obtain the self - consistent accretion - ejection solution . the accretion solution depends on the following constants of motion @xmath89 , @xmath53 , and the viscosity parameter @xmath21 . therefore , our first step is to obtain the accretion solution . step 1 : initially we assume no jet @xmath90 and integrate eqs . ( [ rme.eq]-[ege.eq ] ) eqs . ( [ grad.eq ] ) outwards and find the inner sonic point ( @xmath91 ) iteratively by choosing appropriate @xmath92 and simultaneously checking for the sonic point conditions ( eqs . 16 - 19 , of kumar & chattopadhyay 2013 ) . the first hurdle to obtain the accretion solution is that although flow parameters are regular but there is a coordinate singularity on the horizon . however , the asymptotic values of @xmath93 , and @xmath94 very close to the horizon at @xmath95 , are obtained for appropriate values of @xmath56 @xcite , and are given by , @xmath96 , \ x\rightarrow 1 % r_{g } \label{asam.eq}\ ] ] and @xmath97^{1/2 } , \ x\rightarrow 1 % r_{g } \label{au.eq}\ ] ] where the function @xmath98 is @xmath99^{\frac{\gamma-1}{\gamma+1}}$ ] and the free fall velocity in the pseudo - newtonian potential is given by @xmath100 . without any loss of generality we choose @xmath101 , and with these asymptotic values at @xmath95 we start the integration of eqs . [ grad.eq ] and find the sonic points of the accretion solution . all kinds of solutions can be obtained , ranging from solutions passing through one sonic point ( adaf type and for low angular momentum flow bondi type , see kumar & chattopadhyay 2013 ) , to flow solutions harbouring multiple sonic points . only flows harbouring multiple sonic points can have shocks . once a flow passes through inner type sonic points @xmath91 , we integrate the flow equations while checking for the shock conditions ( eq . [ rmf.eq]-[ef1.eq ] ) and sonic point conditions to calculate shock location ( @xmath102 ) and outer sonic point ( @xmath103 ) simultaneously . the shock conditions are , @xmath104 @xmath105 and the mass flux @xmath106 where @xmath107 and @xmath108 represents the post - shock and pre - shock quantities , respectively and @xmath109 is vertically integrated pressure . here , @xmath110 is the compression ratio at the shock . if there is no shock then @xmath111 , and otherwise as has been mentioned above , the initial accretion solution is obtained by putting @xmath113 . step 2 : after we obtain the accretion temperature and number density distribution from the accretion solution , we compute the radiative moments @xmath114 , @xmath115 , and @xmath116 as is discussed in section 2.2.1 . these variables are used in jet equation . step 3 : the extra compression in the post - shock flow drives bipolar outflows , and the entire post - shock disc participates in jet generation @xcite . the jet base is the post - shock disc , so the jet is launched with the post - shock disc quantities . if @xmath117 is the jet launch site in the post - shock disc , then the specific energy of the disc at @xmath118 @xmath119 , and the post - shock specific angular momentum at @xmath118 @xmath120 and the density at the top of the disc at @xmath118 @xmath121 , are the flow variables of the jet at its base . here we have taken @xmath122 , and since the disc is assumed to be in hydrostatic equilibrium along the @xmath28 direction , @xmath123 , where , @xmath124 is the height of the disc at @xmath118 , and @xmath125 is disc height at shock location . the jet velocity @xmath126 and the sound speed @xmath127 at the jet base are not arbitrarily assigned , but are computed self consistently . we eliminate @xmath128 from eq . ( [ entroj.eq ] ) and eq . ( [ jetbern.eq ] ) , and express @xmath129 in terms of @xmath130 , @xmath126 , @xmath131 etc . since @xmath129 and @xmath131 are obtained from the post - shock disc and therefore known for the jet , we iterate with various values of entropy @xmath130 to obtain the correct value of @xmath126 and @xmath127 , with which the unique transonic solution is determined by integrating eq . ( [ gvjet.eq ] ) and checking for the sonic point conditions eqs . ( [ jetc.eq],[dvjc.eq ] ) . once the transonic jet solution is obtained , then @xmath132 is easily calculated from eq . ( [ ef1.eq ] ) . step 4 : now the information of @xmath132 is used in the shock condition ( eqs . [ rmf.eq ] - [ ef1.eq ] ) and then go back to step 1 to recalculate the new accretion solution , the new radiative moments ( step 2 ) , and the new jet solution ( step 3 ) . this process is repeated till the shock location converges to a value ( @xmath102 ) , and we find out the self - consistent , converged accretion - ejection solution . therefore , there are three iteration processes , namely , ( 1 ) to find out inner sonic point of accretion disc , ( 2 ) accretion shock and the outer sonic point , and ( 3 ) the jet sonic point when launched with the post - shock disc variables . [ lab : fig2 ] now as the jet is generated due to compression as well as shock heating from the post - shock disc ( see fig . [ lab : fig1 ] ) , the jet will be moving through the radiation field of the accretion disc . we now outline the method to compute radiative moments and the net radiative acceleration of the jet . the radiative acceleration term is proportional to the radiative flux @xmath67 , but the radiation drag term ( negative terms in r. h. s of eq . [ radmjet.eq ] ) depends on jet flow velocity @xmath133 , the radiation energy density @xmath68 and the radiation pressure @xmath69 . figure [ lab : fig1 ] shows the schematic diagram of the radiations coming from various parts of the disc , and how it can interact with the outflowing jet , generated due to shock heating of the inner disc . the radiative moments along the jet streamlines ( @xmath134 ) are calculated from the post - shock ( @xmath38 ) and pre - shock ( @xmath135 ) disc respectively . the radiative terms in eq . [ radmjet.eq ] are given by , @xmath136 @xmath137 @xmath138 since the jet streamline is close to the axis of symmetry , we calculate the radiative moments on the axis and approximate these to hold at the same radial distance on the jet streamline . in this paper , we are not dealing with the detailed features of the radiation spectrum from the accretion disc . rather , we are interested to see the effect of the total pre - shock and post - shock radiation on the acceleration of the jet . hence , we do not include radiative transfer dynamically into the hydrodynamic solution . to calculate the moments of radiation due to the pre - shock disc , as an example , we consider only synchrotron processes from the pre - shock disc . the synchrotron emissivity is due to the presence of stochastic magnetic field , where the magnetic pressure ( @xmath139 ) is in partial equipartition with the gas pressure ( @xmath140 ) , @xmath141 here , for @xmath142 implies no magnetic field and therefore the total pressure in eq . ( [ rme.eq ] ) @xmath143 , while @xmath144 implies strict equipartition between gas and magnetic pressure and therefore @xmath145 . the analytical expression for synchrotron emissivity is given by @xcite , and the resulting intensity is , @xmath146 where , @xmath147 and @xmath148 are the pre - shock local dimensionless temperature @xmath149 , electron number density , disc half height and angle from the axis of symmetry to the pre - shock disc surface , respectively . the dependence of @xmath150 on disc radius is through flow variable like @xmath151 . integrating @xmath150 over the pre - shock disc gives us the pre - shock luminosity @xmath152 . the radiative intensity from the post shock disc is @xmath153 , where @xmath154 is the post - shock luminosity in units of eddington luminosity and @xmath155 is the total surface area of the post - shock disc . the total luminosity is @xmath156 . the moments of the radiation field above the accretion disc was calculated before @xcite , and for the above mentioned approximations , they are given by , @xmath157^{3/2 } } ; ~~ { \cal e}_{r_jps}=2\pi{\cal j}_0\int^{x_{inj}}_{x_s}\frac{a^5zxdx } { u^2x^{3/2}(x-1)[(z - x~cot\theta_{ps})^2+x^2]^{3/2 } } , \ ] ] @xmath158 ^ 2 } ; ~~ { \cal f}_{r_jps}=2\pi{\cal j}_0\int^{x_{inj}}_{x_s}\frac{a^5z(z - x~cot\theta_{ps})xdx } { u^2x^{3/2}(x-1)[(z - x~cot\theta_{ps})^2+x^2]^2},\ ] ] @xmath159^{5/2 } } ; ~~ { \cal p}_{r_jps}=2\pi{\cal j}_0\int^{x_{inj}}_{x_s}\frac{a^5z(z - x~cot\theta_{ps})^2xdx } { u^2x^{3/2}(x-1)[(z - x~cot\theta_{ps})^2+x^2]^{5/2}},\ ] ] [ lab : fig3 ] where , @xmath160 , @xmath161 and @xmath162 for pre - shock disc , for @xmath163 . moreover , @xmath164 where , @xmath165 is the accretion rate in units of eddington accretion rate , @xmath65 is thomson scattering cross section , @xmath166 is mean molecular weight of the plasma , @xmath167 is the proton mass , @xmath66 is the electron mass and @xmath5 is the solar mass . it is to be noted that the pre - shock radiation would depend on the product @xmath168 . it is interesting to know that , the post - shock region by virtue of its geometry will block some of the pre - shock photons to the base of the jet , an effect coined as the shadow effect of the post - shock disc @xcite . that is to say , if the height of the jet is @xmath169 , then @xmath170 , where @xmath171 here , @xmath172 & @xmath173 are height and radius at outer edge of the disc . additionally due to the shadow effect , the inner edge of the pre - shock disc as seen by an observer at some height @xmath174 will be , @xmath175 in this paper , we use @xmath176 as a supplied parameter to calculate @xmath152 . also we use @xmath154 initially as a parameter ( except fig . 8 and fig . 9 ) in order to understand the effect of the relative proportions of radiation coming from different parts of disc on the ejected jet , but finally we will use the relation @xmath177 as a function of @xmath102 ( see appendix ) , to compute @xmath178 from the pre - shock radiation , and solve the accretion - ejection solution . in figs . 2a - d , all the plots are generated for a disc characterized by @xmath179 , which generates a shock at @xmath180 . in fig . 2a , @xmath150 is plotted as a function of @xmath26 , starting from @xmath102 outwards . in fig . 2b , we plot @xmath181 ( solid , online black ) , @xmath182 ( dotted , online blue ) and @xmath183 ( dashed , online red ) , the radiation moments due to the post - shock disc . in fig . 2c , we plot @xmath184 ( solid , online black ) , @xmath185 ( dotted , online blue ) and @xmath186 ( dashed , online red ) , the radiation moments due to the pre - shock disc . the shadow effect of the post - shock disc is clearly shown in the figure . in fig . 2d , we plot @xmath187 ( solid , online black ) , @xmath188 ( dotted , online blue ) and @xmath116 ( dashed , online red ) taking equal post - shock and pre - shock luminosities , @xmath189 , respectively . from the above figures it is clear that radiative moments from the post shock region peaks at a height closer to the black hole and the moments from the pre - shock disc peaks typically at a distance few times larger . however , the space dependent parts of the moments ( figs . 2b - c ) shows that the moments of radiation from the pre - shock disc is typically an order of magnitude weaker than those due to the post - shock disc . however , the second distinct bulge in fig . 2d shows , if the pre - shock luminosity is comparable to post - shock luminosity then the radiation moments peaks at two places , and hence presents a prospect for multi - stage acceleration scheme for the jets . in @xcite we have shown in details various cases of accretion solutions , and also have shown that post shock disc naturally produces bipolar outflows . in this paper , therefore we discuss only the shocked accretion solution . the accretion solutions are characteristic by flow parameters like the grand energy @xmath89 , the specific angular momentum at the horizon @xmath53 ( conversely , @xmath190 at the outer edge of the disc ) and the viscosity parameter @xmath21 . the jet is launched from the disc with the specific energy ( @xmath73 and angular momentum ( @xmath191 ) of the post - shock disc at the jet launch site ( @xmath192 ) . we also compute the moments of radiation fields ( @xmath187 , @xmath188 , @xmath116 ) above the disc , and are used to accelerate the jets . the pre - shock disc radiation depends on @xmath176 , where , @xmath193 is the ratio of the magnetic and the gas pressure and @xmath165 is the accretion rate in units of eddington rate . in order to find the effect of post - shock radiation and pre - shock radiation on jet acceleration , the post - shock luminosity @xmath154 and @xmath176 ( conversely , @xmath152 the pre - shock luminosity ) are supplied as independent parameters . however , since post - shock radiation is produced by inverse - comptonization of self generated and intercepted photons from the pre - shock disc , we compute the post - shock radiation self - consistently by employing the techniques of @xcite . and in figs . ( 8 , 9 , 10 ) the jets are accelerated by self consistent estimation of radiation field , both from pre and post - shock discs . the radiations produced from accretion is governed by the solution . however , we will now treat the post - shock luminosity as a parameter just to see how it affects the jet acceleration . the accretion mach number @xmath194 with @xmath195 is plotted for the parameters @xmath196 , in fig . 3 . we first consider @xmath197 and @xmath198 , the shock is at @xmath199 ( vertical dashed , online black ) , and drives bipolar outflow , where the jet mach number @xmath200 ( dashed , online black ) is plotted with @xmath201 , and the mass outflow rate is @xmath202 and the sonic point of the jet is at @xmath203 . radiation field from the accretion disc deposit its momentum and accelerate the shock generated jets . assuming @xmath204 for same set of accretion boundary condition , accretion - ejection solution is computed . the shock is found to be at @xmath205 ( solid , online blue ) and the jet @xmath200 distribution ( solid , online blue ) has a sonic point at @xmath206 and @xmath207 . the total luminosity for this case is @xmath208 . keeping @xmath176 same , we increase @xmath209 , so the total luminosity increases to @xmath210 , the accretion - ejection solution shows that the mass outflow increases to @xmath211 and the shock in accretion decreases to @xmath212 ( dotted , online red ) . in this case too , due to the increase of radiation , the sonic point decreases to @xmath213 , indicating stronger jet . it is interesting to note that , the pre - shock luminosity increases even though @xmath176 is kept constant , because , with the decrease in @xmath102 , the size of the pre - shock disc increases . [ lab : fig4 ] the jet solutions are explored in more details in the following figure . jet variables @xmath128 ( fig . 4a ) , @xmath214 ( fig . 4b ) , @xmath133 ( fig . 4c ) , @xmath215 ( fig . 4d ) , @xmath73 ( fig . 4e ) and @xmath216 ( fig . 4f ) are plotted with @xmath201 . each curve corresponds to @xmath197 ( dashed , online black ) , @xmath204 ( solid , online blue ) , and @xmath209 ( dotted , online red ) , and which are exactly the same cases of jet solutions as in fig . the increase in @xmath178 , accelerates the jets further , and therefore increases @xmath132 , this in turn decreases the post - shock pressure and the location of the shock front moves close to the horizon . as a result the jet base moves closer . at a given @xmath134 , we find @xmath133 is higher for higher @xmath154 , but difference in @xmath128 or @xmath216 are imperceptible . this shows , that enhanced jet acceleration is due to the radiative momentum deposition onto the jets and not due to conversion of thermal energy to the kinetic one . since higher @xmath154 accelerate the jet , the sonic point is formed closer to the horizon . higher @xmath154 not only means faster jet , but also a jet with higher entropy ( fig . [ lab : fig5 ] the jet terminal speed is defined as @xmath217 where @xmath218 . as shown in figs . 4 , @xmath219 increases appreciably with the increase of @xmath154 and @xmath152 , for a given value of disc viscosity parameter @xmath220 . we would like to see whether this behaviour of @xmath219 holds true for a range of @xmath21 . in fig . 5a , @xmath219 is plotted with @xmath154 , in figs . 5b & 5f , @xmath219 is plotted with @xmath102 and the jet base specific energy @xmath221 . and @xmath132 is plotted with @xmath154 ( fig . 5c ) and @xmath102 ( fig . each curve correspond for @xmath222 ( solid , online red ) , @xmath223 ( dotted , online blue ) , and @xmath224 ( dashed , online black ) . all these figures are generated with outer boundary condition @xmath225 and @xmath226 at @xmath227 . it is clear that @xmath219 increases with @xmath178 at a given value of @xmath21 , as well as , increases with @xmath21 at a given @xmath154 . since @xmath102 moves closer with the increase of both @xmath154 and @xmath21 , the jets are launched with higher @xmath221 , which in turn increases @xmath126 . this is also the reason that the relative mass outflow rate @xmath132 increases with the increase of both @xmath154 and @xmath21 . since @xmath126 increases with @xmath154 and @xmath21 , therefore the jets become supersonic at a distance nearer to the jet base , @xmath82 decreases with the increase of @xmath154 ( fig . 5e ) . [ fig6 ] the effect of radiation from pre - shock disc as it impinges on the jet is illustrated through figs . 6b - f . in fig . 6a , the mach number @xmath23 of the accretion solution is plotted with @xmath195 for disc parameters @xmath228 . the shock is at @xmath205 . we choose @xmath204 , but the pre - shock radiation is changed by varying @xmath176 . the solutions correspond to @xmath229 ( solid , online red ) , @xmath230 ( dotted , online blue ) , and @xmath231 ( dashed , online black ) . the post - shock disc actually hides the base of the jet from most of the radiation from the pre - shock disc , while shines its own light onto the jets ( see fig . [ lab : fig1 ] ) . as a result , if the post - shock radiation remains unaltered and the jet sonic point is formed in the portion of the funnel like region where pre - shock radiation is negligible , then the jet base velocity @xmath126 , the jet base @xmath232 or the jet base properties are likely to remain roughly same , keeping the massloss rate unaltered . consequently , the change in the accretion shock is imperceptible ( fig . all the jet variables closer to the base , @xmath233 ( fig . 6b ) , @xmath234 ( fig . 6c ) , @xmath235 ( fig . 6e ) , and @xmath236 are indistinguishable , while they differ from each other in the supersonic region , where the interaction of pre - shock radiation with the jet is significant too . it is to be remembered , that the temperature plotted here is the single temperature of the outflow . the corresponding electron temperature should be about 2 orders of magnitude less . however , the entropy accretion rates @xmath56 ( fig . 6d ) are distinguishable even at the base . once again it is clear from the temperature plot , that radiative driving is significant . it is also interesting to note from figs . 3 - 4 , that increasing @xmath154 , would result in faster jets , with higher @xmath132 and lower jet sonic point ( @xmath82 ) . while increasing the pre - shock radiation also results in faster jets , but with almost no change in @xmath132 and @xmath82 . since relative mass outflow rates affects the accretion solutions ( eq . [ ef1.eq ] ) , so the feed - back effect of the jet on the disc due to @xmath154 might be more significant than that due to @xmath237 . [ lab : fig7 ] in figs . 7a - d , we investigate how @xmath152 affects the jet solutions for a variety of @xmath21 , but for the same outer boundary condition as in fig . 6 with constant @xmath209 . in fig . 7a , we plot @xmath219 as a function of @xmath238 , where each curve represent disc solutions with @xmath222 ( solid , online red ) , @xmath223 ( dotted , online blue ) , and @xmath224 ( dashed , online black ) . in all these plots @xmath152 varies from @xmath239 . due to radiative driving , @xmath219 increases with @xmath238 , and at a given @xmath238 , it increases with @xmath21 . as has been explained in connection to the previous figure , increasing @xmath21 for a fixed outer boundary decreases @xmath102 , which means the jet base energy @xmath221 increases , resulting in faster jet . however , since the pre - shock disc primarily shines radiation on the supersonic part of the jet , therefore @xmath152 has marginal effect on @xmath82 , and @xmath126 . therefore , @xmath132 is almost constant with the change of @xmath152 ( fig . 7d ) , which in turn keeps @xmath102 almost unchanged ( fig . 7b ) , and the jet sonic point @xmath82 also remains unchanged ( fig . in other words , we may conclude , that the radiation from the inner torus of the accretion disc accelerate the jet , but also increases net mass - loss . on the other hand , radiation from pre - shock disc or the outer disc , accelerates the jet appreciably , although , has almost no effect on @xmath132 . therefore , in the second case we may obtain jets with higher kinetic luminosity . this conclusion is valid for any value of @xmath21 which admits accretion shock . [ lab : fig8 ] having investigated the role the pre - shock and post - shock radiation may have on jets , we actually calculate the @xmath154 from @xmath152 . in appendix a , we have discussed the different radiation processes in a general radiative transfer model ( chakrabarti & mandal 2006 ) . from the accretion disc spectrum , we calculate the ratio of post - shock to pre - shock luminosity ( @xmath240 ) using the viscous transonic solution . we then use a fitting formula of @xmath240 ( eq . [ poly.eq ] ) to calculate the ratio at any given shock location ( @xmath102 ) . in ( fig . [ lab : figa2]1b ) , a typical @xmath177 dependence on @xmath102 is obtained where the dots are the data points from model ( chakrabarti & mandal 2006 ) and solid line represents the fitting function . we assume that the behaviour of this ratio with shock location is generic . the procedure for calculating @xmath241 is as follows for a given set of values of @xmath242 , the accretion solution shows a shock at @xmath102 and a bipolar outflow with some @xmath132 . we then calculate @xmath152 ( eq . [ syn.eq ] ) by supplying @xmath168 and the density and temperature distribution between @xmath173 and @xmath102 . we use the fitting formula of @xmath240 ( eq . [ poly.eq ] ) to calculate @xmath154 . using these the jet solution is obtained . all the solutions presented in figs . ( 3 - 7 ) , we have solved the accretion ejection solution with the following fluid parameters @xmath89 , @xmath53 ( at the horizon , or equivalently , @xmath190 at the outer boundary ) , @xmath21 , and in addition the radiation parameters @xmath154 and @xmath168 ( equivalently @xmath152 ) . now , following the procedures described in appendix a , we reduce one parameter , namely , @xmath154 . in fig . 8a , we plot @xmath102 with @xmath53 , in fig . 8b , we plot the compression ratio @xmath243 with @xmath102 , and in fig . 8c , mass outflow rate @xmath132 with the compression ratio @xmath243 . all the plots are for accretion disc parameters @xmath244 , @xmath245 , and various results has been obtained by varying @xmath53 . the curves are for thermally driven jet ( @xmath246 ; dashed , online red ) , and thermal plus radiatively driven jets ( @xmath247 ; dotted online blue , and @xmath248 ; solid , online black ) . this shows that as the compression at the shock increases , it forces more matter into the jet channel . although @xmath243 increases as @xmath102 decreases , but smaller post - shock region means less matter can be driven as jets , so @xmath132 maximizes at some intermediate @xmath243 . it has also been shown earlier @xcite , that for @xmath111 no shock , @xmath249 for no shock there is no outflow . [ lab : fig9 ] in figs . 9a - l , we have plotted accretion and jet solutions for various @xmath21 and @xmath152 . all the accretion solutions ( @xmath23 with @xmath195 ) are for outer boundary parameters @xmath250 at the outer boundary @xmath227 . the viscosity parameters are @xmath251 ( 9a ) , @xmath252 ( 9d ) , @xmath253 ( 9 g ) , and @xmath254 ( 9j ) . the vertical jumps show the location of accretion shocks , and they are at @xmath255 ( 9a ) , @xmath256 ( 9d ) , @xmath257 ( 9 g ) , and @xmath258 9(j ) . the jet solutions corresponding to these accretion solutions , are presented by @xmath200 ( 9b , 9e , 9h , 9k ) and @xmath133 ( 9c , 9f , 9i , 9l ) . as @xmath21 increases @xmath102 decreases , therefore increasing the pre - shock disc . moreover , with decreasing @xmath102 , the post - shock disc becomes smaller and hotter . so as @xmath102 decreases , initially both @xmath154 & @xmath152 will increase , but at around @xmath259 , further decrease of @xmath102 will reduce @xmath240 and significantly increase @xmath152 ( fig . a1b ) . in figs . 9a , d , g , j , increase of @xmath21 , causes a shift of @xmath260 . consequently , @xmath152 increases from @xmath261 . the resulting jets are accelerated and the terminal velocity increases from @xmath262 as shock shifts from @xmath263 , with the corresponding change in luminosity . by considering the relative proportions of post - shock and pre - shock radiations , decrease of @xmath102 with increasing @xmath21 resembles the disc to move from hard state to hard intermediate state , and simultaneously the jet becomes stronger and faster ( both @xmath219 and @xmath132 increases ) . in fig . 10a - c , we show the comparison of shock parameter space ( @xmath264 ) of the accretion disc without massloss ( dotted , online red ) and with massloss but disc parameter @xmath265 ( long dashed , online black ) and @xmath266 ( dashed , online blue ) , and for various viscosity parameter @xmath267 ( fig . 10a ) , @xmath268 ( fig . 10b ) and @xmath269 ( fig . it is to be noted , that the bounded regions in @xmath264 parameter space , show the parameters for steady state shocks to occur , but non - steady shocks still exist outside the bounded region . the parameter space shrinks when massloss is considered , because with massloss , the post - shock pressure decreases , and the entire range for which steady shock may have existed in absence of jets , will not be able to satisfy the momentum balance across the shock front . moreover , shocks seem to exist for fairly high viscosity and in presence of massloss . [ lab : fig10 ] our main focus in this paper has been to study radiative and thermal driving of bipolar jets from dissipative accretion disc . the thermal driving is determined by the thermal energy of the jet base or the post - shock region . in other words , hotter the post - shock region , faster will be the jet . however , the thermal energy in the inner disc though hot , but can not produce jets with @xmath270 @xcite , for any realistic temperature ranges of the accretion disc . as has been explained in section 1 , only thermally driven outflows can achieve terminal speeds of the order of the sound speed at the base . radiative driving of jet is due to the momentum deposition of the accretion disc photons on to the shock driven jet . the issue of radiative acceleration of jets have been studied earlier @xcite , however , the jet base conditions in those works were assumed and were not self consistently obtained . in this paper , we do study the radiative acceleration of jets , but also by self - consistently solving the jet solution from the accretion disc solution itself . this is an exploratory study to investigate whether or not radiative driving is meaningful acceleration mechanism for jets . radiative moments were calculated a posteriori from viscous accretion solutions . we initially considered the post - shock luminosity ( @xmath154 ) as a parameter , and calculated the pre - shock radiation from the disc solution ( @xmath271 , @xmath272 ) and by supplying @xmath168 . similar to @xcite , in this paper too , we find that with the increase of @xmath21 for disc solutions starting with same outer boundary condition , the shock location moves towards the horizon . as @xmath102 decreases for fixed values of @xmath154 , the jets are stronger and faster since the post - shock intensity increases ( figs . 6 ) . for fixed values of @xmath21 , if @xmath154 is increased , the jet is accelerated , as well as , more matter flows into the jet channel . the depleted post - shock flow causes @xmath102 to decrease , in order to regain pressure balance across the shock . since enhanced @xmath154 increases the base velocity or @xmath126 , the jet sonic point decreases too . the effect of pre - shock luminosity is quite different . although pre - shock luminosity can accelerate as efficiently , as the post - shock radiation , however , since pre - shock radiation can ` see ' mostly the supersonic branch of the jet , so neither @xmath132 nor @xmath82 is affected appreciably . therefore , one may conclude that post - shock radiation both accelerates and controls the mass outflow rate , pre - shock radiation increases the kinetic energy of the jet . and since for both pre - shock and post - shock radiations @xmath273 at @xmath274 few@xmath275 , radiation drag effect is nullified at around that distance @xcite . which means faster jets can be obtained if enough radiation power can be supplied . and indeed we obtained terminal speeds of @xmath276 for total disc luminosity @xmath277 . although we have used @xmath278 ( which is acceptable for luminous sources ) , but assigning arbitrary values of @xmath154 may give wrong results , because @xmath154 crucially depends on pre - shock radiation . so in appendix 1 , we have estimated @xmath279 from correct radiative losses from a set of solutions of advective disc . we used this relation to estimate @xmath154 , for given values of @xmath152 and @xmath102 . we then increased @xmath21 for disc solutions starting with same @xmath280 , and calculated simultaneous disc - jet solution . we showed that as @xmath21 is increased , @xmath102 decreases and for reasons explained above , not only the disc moved from low intensity disc to brighter disc , but the jet became stronger and faster . although , we have not considered the keplerian disc component in our solution , but it resembles qualitatively , the transition of the disc from hard to intermediate hard states with associated strengthening of the jet from slow jets to faster and stronger jet , as has been reported in observations . we are still to include a few more physical processes in our disc - jet model , but our results qualitatively shows the correlation of disc spectral states and the jet states , as are observed in micro - quasars . moreover , our result also indicates that , with multi - stage acceleration mechanism , truly relativistic jets from accretion disc are a distinct possibility . 99 becker , p. a. , das , s. , le , t. , 2008 , , 677 , l93 biretta j. a. , 1993 , in burgerella d. , livio m. , oea c. , eds , space telesc . , vol . 6 , astrophysical jets . cambridge univ . press , cambridge , 263 chakrabarti , s. k. , 1989 , , 347 , 365 chakrabarti , s k. , titarchuk , l. , 1995 , , 455 , 623 . chakrabarti , s. k. , 1996 , , 464 , 664 chakrabarti , s. k. , 1999 , , 351 , 185 chakrabarti , s. k. , manickam , s. g. , 2000 , , 531 , l41 chakrabarti , s. k. ; das , s. , 2004 , , 349 , 649 chakrabarti , s. k. , mandal , s. , 2006 , , 642 , l49 . chakrabarti , s. k. , debnath , d. , nandi , a. , pal , p. s. , 2008 , , 489 , l41 chakrabarti , s. k. , dutta , b. g. , pal , p. s. , 2009 , , 394 , 1463 chattopadhyay , i. , chakrabarti , s. k. , 2000 , int . chattopadhyay , i. , chakrabarti , s. k. , 2002 , , 333 , 454 . chattopadhyay , i. , chakrabarti , s. k. , 2003 , in durouchoux ph . , fuchs y. , and rodriguez j. , eds , new views on microquasars , institut , detudes , scientifiques de cargese , corsica france , may 27-june 1 , 2002 , center for space physics , kolkata , india , p 126 . chattopadhyay , i. , das , s. , chakrabarti , s. k. , 2004 , , 348 , 846 . chattopadhyay , i. , 2005 , , 356 , 145 chattopadhyay , i. ; das , s. , 2007 , , 12 , 454 chattopadhyay , i. , 2008 , in chakrabarti s. k. , majumdar a. s. , eds , aip conf . ser . vol . 1053 , proc . 2nd kolkata conf . on observational evidence of back holes in the universe and the satellite meeting on black holes neutron stars and gamma - ray bursts . phys . , new york , p. 353 chattopadhyay , i. , chakrabarti , s. k. , 2011 , int . d , 20 , 1597 das , s. , chattopadhyay , i. , nandi , a. , chakrabarti , s. k. , 2001 , , 379 , 683 das , s. ; chattopadhyay , i. , 2008 , , 13 , 549d . das , s. ; becker , p. a. ; le , t. , 2009 , 702 , 649 doeleman , s. , s. , 2012 , science , 338 , 355 . feroci , m. , matt , g. , pooley , g. , costa , e. , tavani , m. , beloni , t. , 1999 , , 351 , 985 . fender r. , beloni , t. , 2004 , , 42 , 317 fender , r. p. , gallo , e. , russell , d. , 2010 , , 406 , 1425 . fukue , j. , 1987 , , 39 , 309 fukue , j. , 1996 , , 48 , 631 fukue , j. , tojyo , m. , hirai , y. , 2001 , , 53 , 555 fukumura , k. , tsuruta , s. , 2004 , , 611 , 964 gallo , e. , fender , r. p. , pooley , g. , g. , 2003 , 344 , 60 gu , wei - min ; lu , ju - fu , 2004 , chphl , 21 , 2551 icke , v. , 1980 , aj , 85(3 ) , 329 icke , v. , 1989 , 216 , 294 junor , w. , biretta , j. a. , livio , m. , 1999 , nature , 401 , 891 kumar r. , chattopadhyay i. , 2013 , mnras , 430 lanzafame , g. , cassaro , p. , schillir , f. , costa , v. , belvedere , g. , zappal , r. a. , 2008 , , 482 , 473 liang , e. p. t. , thompson , k. a. , 1980 , 240 , 271l lu , j. f. , gu , w. m. , & yuan , f. , 1999 , , 523 , 340 lynden - bell d. , 1978 , phys . , 17 , 185 mchardy , i. m. , koerding , e. , knigge , c. , fender , r. p. , 2006 , nature , 444 , 730 . mandal , s. , chakrabarti , s. , k. , 2010 , , 710 , l147 . margon , b. , 1984 , , 22 , 507 . mihalas d. , mihalas b. w. , 1984 , foundations of radiation hydrodynamics . oxford univ . press , oxford miller - jones , c. j. a. , , 2012 , mnras , 421 , 468 . mirabel i. f. , rodriguez l. f. , 1994 , nat , 371 , 46 molteni , d. , lanzafame , g. , chakrabarti , s. k. , 1994 , , 425 , 161 molteni , d. , ryu , d. , chakrabarti , s. k. , 1996a , apj , 470 , 460 molteni , d. , sponholtz , h. , chakrabarti , s. k. , 1996b , apj , 457 , 805 . narayan , r. , kato , s. , honma , f. , 1997 , 476 , 49 nagakura , h. , yamada , s. , 2009 , , 696 , 2026 novikov , i. d. ; thorne , k. s. , 1973 , in dewitt b. s. , dewitt c. , eds , black holes . gordon & breach , new york , p. 343 nishikawa , k. -i . , richardson , g. , koide , s. , shibata , k. , 2005 , , 625 , 60 paczyski , b. and wiita , p.j . , 1980 , , 88 , 23 . rushton , a. , spencer r. , fender , r. , pooley , g. , 2010 , , 524 , 29 shakura , n. i. ; sunyaev , r. a. , 1973 , , 24 , 337s . shaposhnikov , n. , titarchuk , l. , 2007 , , 663 , 445 . shaposhnikov , n. , titarchuk , l. , 2009 , , 699 , 453 shapiro , s. l. , teukolsky , s. a. , 1983 , black holes , white dwarfs and neutron stars , physics of compact objects . wiley - interscience , new york . 66 sikora m. , wilson d. b. , 1981 , , 197 , 529 . sunyaev , r. a. , titarchuk , l. , 1980 , , 86 , 121 . titarchuk , l. , g. , & fiorito , r. 2004 , apj , 612 , 988 . vadawale , s. , rao , a. r. , nandi , a. , chakrabarti , s. k. , 2001 , , 370 , l17 . we consider a general radiative transfer model ( chakrabarti & mandal 2006 ) which consists of two components , a keplerian disc on the equatorial plane and a sub - keplerian component on the top of the keplerian disc . the keplerian disc supplies the multi - colour black body photons and a fraction of that photons are inverse - comptonized by post - shock region . the pre - shock sub - keplerian disc emits radiation via bremsstrahlung and synchrotron process whereas the post - shock region produces the same as in pre - shock along with the comptonization of the local and intercepted soft photons . we calculate the radiation spectrum of accretion disc using the viscous transonic shocked solution for a given outer boundary condition @xmath281 at @xmath227 . for a given value of @xmath21 , a shock is formed at @xmath102 . the frequency integrated pre - shock and post - shock luminosity is then calculated . it has been shown in @xcite that for the same set of boundary conditions , @xmath102 decreases if @xmath21 is increased . in our case for a range of @xmath282 , for flows starting with the same injection values mentioned above , the shock location changes from @xmath283 ( fig . we calculate the ratio of post - shock to pre - shock luminosity @xmath240 for different values of @xmath102 in the range mentioned above and it has been plotted by dots in fig . ( a1b ) . in fig a1c we plot the associated variation in photon index , which shows the spectra softens as shock moves inward . the ratio ( @xmath240 ) increases initially as the shock move inward , reaching a maximum value and then decreases sharply . this behaviour is due to fact that initially as shock moves inward both post - shock temperature and density increases and hence post - shock luminosity increases but as shock reaches closer to central object the pre - shock luminosity @xmath152 increases and the increased supply of pre - shock photons cools down the post - shock flow , and therefore the ratio would decrease too . we have fitted a polynomial through the model data points and this general behaviour of @xmath240 with @xmath102 is used to calculate @xmath154 for given value of @xmath152 .

a _ chandra _ spectrum of the moderately active nearby f6 v star ori is used to study the coronal properties of mid - f dwarfs . we find that ori s coronal emission measure distribution is very similar to those of moderately active g and k dwarfs , with an emission measure peak near seeming to be ubiquitous for such stars . in contrast to coronal temperature , coronal abundances are known to depend on spectral type for main sequence stars . based on this previously known relation , we expected ori s corona to exhibit an extremely strong `` fip effect , '' a phenomenon first identified on the sun where elements with low `` first ionization potential '' ( fip ) are enhanced in the corona . we instead find that ori s corona exhibits a fip effect essentially identical to that of the sun and other early g dwarfs , perhaps indicating that the increase in fip bias towards earlier spectral types stops or at least slows for f stars . we find that ori s coronal characteristics are significantly different from two previously studied mid - f stars , procyon ( f5 iv - v ) and boo ( f7 v ) . we believe ori is more representative of the coronal characteristics of mid - f dwarfs , with procyon being different because of luminosity class , and boo being different because of the effects of one of two close companions , one stellar ( boo b : m2 v ) and one planetary .

1303.5630

although details about stellar magnetic field generation remain uncertain , it is accepted that coronal magnetic fields are generated by a magnetic dynamo operating within the convection zone . for main sequence stars , the convection zone narrows towards earlier spectral types , disappearing entirely around a spectral type of a5 v. at this point , coronae disappear as well @xcite , emphasizing the central role that the convection zone dynamo plays in generating stellar coronae . one general empirical goal of stellar coronal x - ray observations is to study how coronae change character as the convection zone narrows . any clear correlation between coronal properties and spectral type tells us something useful about how the dynamo and the fields it generates change with spectral type , thereby providing potentially crucial information for theoretical models of how stellar magnetic dynamos operate in general . coronal abundances represent one coronal property that appears to be tightly correlated with spectral type , at least for main sequence stars . on the sun , it has long been known that coronal abundances are different from photospheric abundances , elements with low first ionization potential ( fip ) having enhanced abundances in the solar corona and solar wind @xcite . analyses of x - ray and euv spectra have found a similar `` fip effect '' for some stars of low to moderate activity @xcite . however , for many other stars , particularly extremely active ones , the fip effect is either absent , or sometimes an inverse fip effect is observed , where low - fip elements have coronal abundances that are _ depleted _ relative to the high - fip elements @xcite . currently the only detailed theoretical model that is capable of explaining both solar - like fip effects and inverse fip effects is one that attributes the element fractionation to ponderomotive forces induced by alfvn and other mhd waves passing through or reflecting from the chromosphere @xcite . initially , it appeared that coronal abundances were primarily dependent on activity , as the most active stars clearly tend to have inverse fip effects instead of a solar - like fip effect . this activity dependence of abundances for active stars is illustrated and discussed in detail by ( * ? ? ? * see figure 37 ) . the abundance variations are not simply a uniform variation of low - fip elements relative to high - fip ones . the high - fip elements appear to fractionate with respect to each other as well , as illustrated by measurements of ne / o . the vast majority of spectroscopically observed stellar coronae exhibit high ratios of @xmath3 @xcite , but relatively inactive stars ( e.g. , the sun , @xmath4 cen ab , and procyon ) show @xmath5 @xcite . as more abundance measurements have been made , it has become apparent that abundances are also highly dependent on spectral type @xcite . in order to explore this further we have experimented with a high - fip / low - fip abundance ratio , @xmath6 , which quantifies a corona s fip bias as the average abundance of four high - fip elements ( c , n , o , and ne ) relative to fe ; normalized to photospheric abundance ratios . for normal main sequence stars with coronal x - ray luminosities of @xmath7 ( in ergs s@xmath8 ) , there is a surprisingly tight spectral type dependence for @xmath6 @xcite . in this correlation , early g stars all have a solar - like fip effect , which decreases towards late g and early k stars , reaching no fip effect at all at a spectral type of k5 v. later than k5 v , inverse fip effects are observed , with the magnitude of the effect increasing into the m dwarfs . we will refer to this as the `` fip - bias / spectral - type '' ( fbst ) relation . at least for stars with @xmath7 , any activity dependence for @xmath6 is lost in the scatter of the measurements . the fbst relation would predict that the solar - like fip effect should become even stronger as you move from solar - like early g stars to f stars . more precisely , instead of a factor of 4 enhancement of low fip elements in the corona as for the sun @xcite , the fbst relation would predict a factor of 6 enhancement for mid - f stars @xcite . however , observations of two mid - f dwarfs seem to contradict this expectation . the first of these is procyon ( f5 iv - v ) , the nearest and most frequently observed f star , which has been observed spectroscopically by euve , _ chandra _ , and xmm . all of these spectra demonstrate that procyon s corona exhibits no significant fip bias at all . its coronal abundances are consistent with its photospheric abundances @xcite . more recently , using xmm data @xcite have also found little difference in the abundances of low - fip and high - fip elements in the corona of @xmath2 boo ( f7 v ) , although they suggest that _ both _ are depleted in the corona by similar amounts . the procyon and @xmath2 boo measurements would seem to suggest that the fbst relation falls apart for the f dwarfs , possibly an effect of the thinning convection zones of such stars . this would be a very interesting result , but there are potential issues with both procyon and @xmath2 boo regarding whether either can be considered representative of mid - f dwarfs . for procyon , the problem is simply that it is not a pure main sequence star , as its `` iv - v '' luminosity class suggests . its radius of 2.05 r@xmath9 is certainly larger than that of a true mid - f main sequence star @xcite . for @xmath2 boo , the problem is the presence of two very different companions , one stellar and one planetary . the stellar companion , @xmath2 boo b , is an m2 v star that is unresolved from the primary in the xmm observation , which could in principle be contributing x - ray photons to the spectrum . given that m dwarfs are expected to have inverse fip effects @xcite , contamination from @xmath2 boo b would be expected to weaken any fip effect in the combined x - ray spectrum . as for the planetary companion , it is a very well - studied close - in giant planet , @xmath2 boo b , with an orbital period of only 3.31 days . there is already evidence that the presence of this massive planet is producing coronal activity on the star @xcite . is it possible that the planet could somehow be suppressing the expected strong coronal fip effect as well ? this is an exciting possibility , which makes it all the more important to know what the corona of a truly normal mid - f dwarf looks like . to that end , we here present an analysis of recent _ chandra _ observations of the nearby f6 v star , @xmath0 ori , which lies at a distance of 8.1 pc @xcite . the star has a rapid rotation rate of @xmath10 km s@xmath8 , corresponding to @xmath11 days @xcite . not surprisingly , given the known connection between stellar rotation and activity , @xmath0 ori also has a relatively high x - ray luminosity of @xmath12 @xcite , just barely within the @xmath7 activity regime mentioned above where the fbst relation remains tight . this is in fact one of the ten most luminous coronal x - ray sources within 10 pc , so it is curious that @xmath0 ori has never before been observed spectroscopically in x - rays . the @xmath0 ori spectrum promises to resolve the issue of what the coronal characteristics of moderately active mid - f main sequence stars really are . the _ chandra _ x - ray telescope observed @xmath0 ori twice in late 2010 , the first time on 2010 november 11 for 57.8 ksec , and a second time on 2010 november 23 for 20.1 ksec . the observations were performed with _ chandra _ s letgs configuration , combining the low energy transmission grating ( letg ) and the component of the high resolution camera detector used for spectroscopy ( hrc - s ) . the data were processed using version 4.3 of the _ chandra _ team s ciao software . the initial hrc - s image produced by the data processing consists of a zeroth - order image of the star , and an x - ray spectrum dispersed on both sides of the image , i.e. , a plus and minus order spectrum . further processing extracts one - dimensional spectra from the image , and then the plus and minus order spectra are coadded into a final spectrum covering the full @xmath13 wavelength region of letgs . photons in the time - tagged zeroth - order image are useful for assessing source variability . we find no significant variability at all for @xmath0 ori , and the x - ray fluxes appear to be identical at the two observing times , so we simply coadd the spectra taken at the two times into a final spectrum representing a combined exposure time of 77.9 ksec . this spectrum is shown in figure 1 . we use version 6.1 of the chianti atomic database to identify emission lines in our spectrum @xcite . the identified lines are listed in table 1 , and identifications are also provided in figure 1 . counts for the detected lines are measured by direct integration from the spectrum , and these raw counts are then converted to photon fluxes . both the counts and fluxes of the lines are provided in table 1 , together with 1@xmath14 uncertainties . the third column of the table lists line formation temperatures , based on ionization equilibrium computations from @xcite . there are many lines in table 1 with large quoted uncertainties that represent questionable detections . the emission measure ( em ) analysis described in section 2.3 is used to confirm _ a posteriori _ that these marginal detections are at least plausible . there are many line blends in table 1 . the em analysis and the line strengths in the chianti database are used to assess what lines are contributing significant flux to each emission feature . the line blends involving different species are ultimately ignored in the final em analysis . there are many density sensitive he - like triplets in the @xmath0 ori spectrum @xcite , but only the o vii triplet near 21.8 is sufficiently free of blends and with high enough signal - to - noise ( s / n ) for a decent density measurement . unfortunately , even for o vii we find we can only quote an upper limit for the electron density . based on the collisional equilibrium models of @xcite , the @xmath15 line flux ratio corresponds to an electron density ( in @xmath16 ) of @xmath17 . similarly low coronal densities have been found for many moderately active g and k dwarfs in the past @xcite , so there is no evidence that our moderately active f dwarf is any different from later type stars in terms of coronal density . inferring coronal abundances and temperature distributions from line flux measurements requires an em analysis , which is a nontrivial inversion problem . for this calculation , we use version 2.6 of the pintofale software package @xcite , which was developed to solve the em inversion problem using a markov chain monte carlo approach . the analysis also makes use of chianti line emissivities @xcite . our em analysis closely mirrors previous work , as described in particular detail by @xcite , so we refer the reader to that paper for details about this computation . the analysis considers the effects of interstellar absorption on the observed line fluxes . there is no measured h i column density towards @xmath0 ori itself , but there is a measurement of @xmath18 ( in @xmath19 ) towards @xmath20 ori @xcite , which is roughly at the same distance and in the same direction as @xmath0 ori , so this is the value we assume . we note , however , that this value is low enough that interstellar absorption does not have a large effect on line fluxes . the line fluxes by themselves allow the computation of the shape of the emission measure distribution , and the computation of relative coronal abundances . for the relative abundances , the natural reference element to use is fe , considering the large number of fe lines available in the letgs spectrum . in the initial em analysis we simply assume a solar photospheric abundance for our reference element , @xmath21}=-4.50 $ ] @xcite . table 2 lists the coronal abundances relative to fe computed in the em analysis , and figure 2 shows the derived emission measure distribution . the 90% confidence intervals shown for the emission measures and abundances in figure 2 and table 2 , respectively , are indicative of the uncertainties in solving the inversion problem , based on the magnitude of the uncertainties in the measured line fluxes . properly normalizing the em distribution requires the measurement of an absolute coronal [ fe / h ] ratio , which also allows the relative abundances in table 2 to be converted to absolute abundances . measurement of [ fe / h ] requires an assessment of the line - to - continuum ratio in the letgs spectrum , which can be done after the initial em analysis is completed . emission measures scale inversely with the assumed [ fe / h ] , since higher metal abundances mean less emission measure is required to account for the line fluxes . thus , increasing [ fe / h ] by a factor of two from @xmath21}=-4.50 $ ] to @xmath21}=-4.20 $ ] , for example , uniformly lowers the em distribution by a factor of two . synthetic line - plus - continuum spectra can be computed from the em distribution for various values of [ fe / h ] and compared with the data , in order to see which leads to the best match to the observed continuum level . this is illustrated in figure 3 , which shows the best fit for @xmath21}=-4.32 $ ] . this synthetic spectrum is also compared with the data in figure 1 . in letgs data , higher order spectra are superimposed onto the first order spectrum , so the synthetic spectrum accounts for the higher order spectra as well , specifically orders 25 . the higher order contributions are explicitly plotted in both figures 1 and 3 . the em distribution in figure 2 is the final normalized distribution assuming @xmath21}=-4.32 $ ] . the most striking feature of the @xmath0 ori em distribution in figure 2 is a sharp peak at @xmath1 . this peak actually seems to be very common for moderately active stars like @xmath0 ori . many examples are provided by @xcite , but the best of these are @xmath22 boo a ( g8 v ) and @xmath23 eri ( k2 v ) . the em distributions for these 2 stars are also shown in figure 2 . the @xmath22 boo a em distribution is a particularly good match for that of @xmath0 ori . this is perhaps not surprising considering that @xmath22 boo a and @xmath0 ori have similar x - ray luminosities of @xmath24 and @xmath12 , respectively @xcite , although the spectral types of the two stars are quite different . thus , there is no evidence that the thinner convection zone of @xmath0 ori has led to coronal temperatures any different from those of similarly active later type stars . interestingly enough , the @xmath1 peaks are also commonly observed for solar active regions @xcite . in figure 2 , the three stellar em distributions are compared with a solar one , which represents the average of 15 active region em distributions from @xcite , who emphasized the prevalence of @xmath1 em peaks on the sun . the shape of the solar active region em agrees beautifully with those of @xmath0 ori and @xmath22 boo a for @xmath25 , although the three stellar em distributions in figure 2 have much higher em at @xmath26 . @xcite have previously noted the similarity between the em distributions of solar active regions and those of @xmath22 boo a and @xmath23 eri , and have suggested that it implies that moderately active stars like these are covered almost completely with solar - like active regions . is it possible that these seemingly ubiquitous @xmath1 peaks are somehow a systematic artifact of the pintofale computations that produced them ? not only was pintofale s markov chain monte carlo approach used to compute the stellar em distributions in figure 2 , but it was also used by @xcite to compute the solar ones . note , however , that @xcite analyzed a completely different spectral region and a different set of emission lines than those used in the stellar analyses , suggesting that the @xmath1 peaks are not just a product of working with a particular set of lines . we explore this issue further using simulations . these simulations involve taking a known em distribution , computing line fluxes from that distribution for the set of lines in table 2 , and then plugging those line fluxes back into pintofale to see if the em inversion procedure can recover the original em distribution . uncertainties in the line fluxes are assumed to be the same as the relative uncertainties measured for the @xmath0 ori lines in table 1 . some results of these simulations are shown in figure 4 . the dotted line in figure 4 s `` simulation 1 '' is simply the @xmath0 ori distribution from figure 2 . the solid line shows the result when that is used as the input em distribution for the simulation . the computed em distribution agrees encouragingly well with the original distribution , being within the displayed error bars for 15 of the 17 temperature bins . most importantly , the simulation demonstrates pintofale s ability to recover the sharp @xmath1 em peak . in simulations 2 and 3 we simply shift the peak to @xmath27 and @xmath28 , respectively , and in simulation 4 we remove the peak entirely , assuming equal em values in the @xmath29 range . although the reconstructed em values are still generally within error bars , we do note that for simulations 24 the em is systematically overpredicted at @xmath1 and underpredicted at @xmath27 and @xmath28 . we conclude , therefore , that there may indeed be a modest tendency for pintofale to overemphasize the height and narrowness of the @xmath1 peaks , though the magnitude of this effect is not necessarily inconsistent with the inferred 90% confidence intervals . we can only speculate as to the source of this tendency . it does so happen that @xmath1 is the central formation temperature of the strongest lines in the letgs spectrum , the fe xvii lines between 1517 , so perhaps the need to fit those line fluxes particularly well leads to the modest overemphasis of em at @xmath1 , to the detriment of em at @xmath27 and @xmath28 . in any case , despite evidence for this small systematic error , the simulations provide support for the existence of a distinct maximum of em near @xmath1 for moderately active stars and solar active regions . the question then becomes , why is @xmath1 such a popular temperature for both a variety of solar active regions , and for moderately active stars of various spectral types ? is there something special about this particular temperature ? one notable thing about it is that there is a local minimum in the radiative cooling curve near @xmath1 ( e.g. , * ? ? ? * ; * ? ? ? the cooling curve relates how emissivity varies with temperature . plasma is stable against radiatively driven thermal instabilities where the cooling curve has a shallower slope than the temperature dependence of the heating function . the cooling curve has a steep negative slope for a variety of abundance sets in the range @xmath30 , which could imply thermal instability unless the heating curve has an even steeper negative slope at those temperatures . thus , @xmath1 is possibly the coolest stable temperature above @xmath31 . plasma initially between these two temperatures will most likely either heat up or cool down until a stable temperature is reached . perhaps the ubiquity of the @xmath1 peak implies something about the commonality of active region loops on the sun and solar - like main sequence stars , consistent with the idea that coronal emission from moderately active stars like @xmath0 ori and @xmath22 boo a is dominated by solar - like active regions that largely cover the stellar surface @xcite . on the sun , such loops are often found to have densities of @xmath32 @xmath16 ( e.g. , * ? ? ? such densities are consistent with those commonly measured in stellar coronae as well , at least at temperatures of @xmath33 @xcite . an explanation in terms of thermal stability presupposes that coronal plasma exists in structures ( loops ) that can support a quasi - steady state . no such explanation would be possible in a corona heated and supplied by more dynamic phenomena such as type ii spicules ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in simple static loop models there is a relation between a coronal loop s length and its apex temperature and density @xcite . thus , if coronal loops in various solar and stellar contexts tend to have similar densities and lengths , then presumably they would have similar temperatures as well . this presumes , of course , that these simple static models apply , which is questionable ( e.g. , * ? ? ? * ; * ? ? ? as discussed in section 2.3 , the em analysis provides coronal abundances measured relative to fe for many elements . all of these measurements , in the form @xmath34}$ ] , are listed in the second column of table 2 . we are interested in comparing coronal abundances with those of the photosphere . thus , in the third column of table 2 , we list stellar photospheric abundance ratios relative to fe , in the form @xmath34}_*$ ] . the photospheric abundances of @xcite are measured using line - by - line comparisons of solar and stellar spectral features , so the abundance measurements are intrinsically relative to solar abundances . thus , computing the stellar photospheric abundance ratios in table 2 requires the assumption of solar abundances . the assumed solar abundances of @xcite are listed in the last column of table 2 . for n and s , there are no stellar photospheric measurements available . we assume these elements behave like high - fip elements , such as o. thus , we assume @xmath35}\equiv \log{\rm [ s_*/s_{\odot}]}\equiv \log{\rm [ o_*/o_{\odot}]}=0.16 $ ] . neon is a special case . the fundamental problem with ne is that even for the sun there is no real photospheric abundance measurement , because there are no solar photospheric ne absorption lines . reference solar abundances for ne , such as the one listed in table 2 , are in reality based on measurements from the solar corona and transition region ( e.g. , * ? ? ? we know very well that abundances can be different from photospheric in the corona , but not necessarily in the transition region , where @xcite finds a low ne / o abundance , and no evidence for a fip effect in other elements . nevertheless , we do know that high - fip elements can be fractionated in the solar atmosphere , as he is depleted by a factor of 2 in the slow solar wind @xcite . this depletion of he ( and also possibly of ne ) is also a feature of models for the solar fip effect based on the ponderomotive force @xcite . complicating the issue further is the knowledge that moderately active and very active stars seem to univerally have coronal ne / o ratios much higher than that of the sun and other similarly inactive stars @xcite , raising the possibility that these stellar coronal ne abundances may be a better measure of the true cosmic ne abundance than the solar coronal ne abundance usually referenced as such . we here follow the same convention as in our past papers and assume that the average stellar coronal @xmath36 abundance ratio of @xcite is indicative of the true cosmic abundance of ne , and therefore applies to solar and stellar photospheres as well , including for @xmath0 ori . thus , the @xmath37}_*$ ] ratio in table 2 is derived assuming @xmath36 , instead of the much lower solar ratio of @xmath38 . the validity of this assumption is very much debatable , but it does lead to coronal ne abundances in better agreement with those of other high - fip elements ( see figure 5 ) . after subtracting the logarithmic photospheric abundance ratios , the logarithmic coronal abundance ratios in table 2 are in figure 5a plotted versus fip . the low - fip elements are all roughly consistent with the reference low - fip element , fe . the high - fip element abundances are lower by about @xmath39 dex , consistent with a fip effect nearly identical to that of the sun . the average value of the four high - fip elements in figure 5a is plotted explicitly as a horizontal dot - dashed line in the figure . this is the `` fip bias '' quantity , @xmath6 , that we have used in the past to reduce coronal abundances to a single number , which is then used to study the fbst relation described in section 1 @xcite . for @xmath0 ori , @xmath40 . the @xmath6 quantity has no clear variation with activity for main sequence stars with @xmath7 ( see figure 4 in * ? ? ? * ) , but it is worth noting that of the four high - fip elements that go into computing @xmath6 ( c , n , o , and ne ) , ne is once again an exception , as we know that the coronal ne / o ratio does vary within the @xmath7 sample , suggesting activity - dependent fractionation of ne even for less active stars @xcite . as alluded to above , the vast majority of spectroscopically observed @xmath7 stars , including @xmath0 ori , have @xmath3 . only a few truly inactive stars ( the sun and @xmath4 cen ab ) have @xmath5 @xcite . there are a couple characteristics of the low - fip elements that are worthy of note . @xcite found that the coronal ni abundance of the mid - f dwarf @xmath2 boo was anomalously high compared to other low - fip elements , being a factor of @xmath41 higher . we do not see this behavior for our mid - f dwarf , @xmath0 ori . the ni abundance may be a little high in figure 5 , but it is consistent with the other low - fip elements within the error bars . as for mg and si , in the past we had noted a tendency for mg abundances to be higher than si abundances in the coronae of many moderately active stars @xcite . this is not the case for @xmath0 ori , though , which if anything shows the opposite behavior , consistent with what @xcite found for @xmath2 boo . figure 5a shows the coronal abundance measurements relative to fe . but using the @xmath21}=-4.32 $ ] measurement from the line - to - continuum analysis , we can compute absolute abundances as well , and in figure 5b the absolute abundances are plotted , relative to stellar photospheric abundances as before . this figure casts doubt on whether the coronal abundance fractionation pattern of @xmath0 ori is truly solar - like or not . based on figure 5a , the @xmath0 ori coronal abundances were described above as very solar - like , because the @xmath40 measurement implies a factor of 4 underabundance of high - fip elements relative to low - fip elements in the corona , similar to the situation in the solar corona . but in the solar corona this is really due to an enhancement of low - fip elements , as opposed to a depletion of high - fip elements , meaning that for the sun the high - fip elements would be near 0.0 in figure 5b and the low - fip elements would be near 0.6 . in contrast , figure 5b would imply that for @xmath0 ori , the @xmath40 result is due to fractionation of _ both _ low - fip and high - fip elements , and is in fact mostly due to a depletion of high - fip elements . this is actually consistent with what we have found before for other moderately active stars @xcite . the question is whether this means that the fractionation behavior of these stars really is significantly different from the sun , or whether there is a systematic error in the line - to - continuum analysis that leads to systematic underestimates of @xmath21}$ ] . the history of solar coronal abundance measurements provides reason for caution with regards to the accuracy of the line - to - continuum analysis , as many such solar analyses have found the data more consistent with a high - fip depletion than a low - fip enhancement @xcite , contradicting other evidence , including direct solar wind measurements , that suggest low - fip enhancement is the primary effect @xcite . one concern is the possible presence of weak metal lines not listed in atomic line databases . these lines could collectively produce a significant pseudo - continuum in x - ray spectra , leading to overestimates of the strength of the real continuum , corresponding to an underestimate of @xmath21}$ ] . there is hope that modern databases such as chianti with more extensive line lists have greatly reduced this problem , but it is difficult to quantify what the systematic uncertainties really are . another potential issue is the effect of non - maxwellian distributions on the line - to - continuum analysis . non - maxwellian distributions will affect the analysis of both the line and continuum fluxes , but in different ways @xcite . all these issues provide justification for relying primarily on relative , rather than absolute abundances . the latter introduces the uncertain systematic errors involved in the line - to - continuum analysis , while the former does not . nevertheless , when using relative abundances , and quantities derived from them such as @xmath6 , it is important to keep in mind that the issue of absolute abundances has been left unresolved . a negative @xmath6 value , for example , could indicate either a low - fip enhancement or a high - fip depletion , and two stars with identical @xmath6 values could in principle be different in that respect . that being said , it is hard to imagine how the rather tight fbst relation shown in figure 6 ( and discussed in detail in the next section ) could exist without some consistency regarding how @xmath6 relates to absolute abundances . for example , the fbst relation suggests m4 v stars all have @xmath42 @xcite . it is hard to imagine why these stars would have different degrees of low - fip depletions and/or high - fip enhancements , but always calibrated so that @xmath43 . as described in section 1 , one of the main goals of the @xmath0 ori analysis is to see whether the fbst relation extends to f spectral types . the answer is provided in figure 6 . a full list of the stars plotted in the figure is provided by @xcite , but we now add the @xmath0 ori data point . based on the g , k , and m stars in the figure , @xmath0 ori would be expected to have @xmath44 , potentially the strongest solar - like fip effect ever observed . instead it has @xmath40 , comparable to the sun and other early g stars . there are three possible interpretations for the higher - than - expected @xmath6 measurement for @xmath0 ori : 1 . simply taken at face value , the @xmath0 ori measurement indicates that the fbst relation flattens towards spectral types earlier than g. observations of early f dwarfs would be very useful to verify this flattening . if real , the flattening of the fbst relation could be interpreted as an interesting manifestation of the narrowing convection zone . although somewhat higher than expected , it could be argued that the @xmath0 ori measurement is still close enough to the expected @xmath44 value to be considered consistent with a near - linear fbst relation , considering the amount of scatter that one sees in figure 6 . perhaps @xmath0 ori is actually too active to precisely follow the fbst relation . it is known that stars with @xmath45 lie significantly above the fbst relation @xcite . with @xmath12 @xcite , @xmath0 ori is close to this limit . if the @xmath46 limit is actually spectral type dependent , and lower for f dwarfs , @xmath0 ori would be expected to lie somewhat above the fbst curve . in section 1 , there were two f dwarfs discussed that are known from previously published work to be inconsistent with the fbst relation : procyon ( f5 iv - v ) and @xmath2 boo ( f7 v ) . based on measurements from @xcite and @xcite , we compute values of @xmath47 and @xmath48 for procyon and @xmath2 boo , respectively . these values are plotted in figure 6 , explicitly showing just how inconsistent these two stars are with both @xmath0 ori and the fbst relation . as for procyon , the problem is that it is not a pure main sequence star , as its `` iv - v '' luminosity class and large 2.05 r@xmath9 radius demonstrate @xcite . not only does procyon have coronal abundances very different from @xmath0 ori , but it has a much cooler corona as well , with an em peak near @xmath49 and no significant em above @xmath50 @xcite . cool coronae are associated with less active stars , but procyon s x - ray luminosity is @xmath51 , only a factor of 3 lower than @xmath0 ori , and comparable to many of the stars in the figure 6 sample with hotter coronae . for example , @xmath23 eri has a coronal x - ray luminosity of @xmath52 @xcite , lower than procyon , but it has an em distribution similar to @xmath0 ori ( see figure 2 ) . we believe that procyon s inconsistency with @xmath0 ori emphasizes just how sensitive coronal properties can be to luminosity class . procyon simply can not be considered to be a main sequence star , and it is dramatically inconsistent with the fbst relation as a consequence . as for @xmath2 boo , the inconsistency of @xmath2 boo with @xmath0 ori strengthens the argument made in section 1 that @xmath2 boo s high @xmath6 value is due to one of its two companions . either the m2 v companion , @xmath2 boo b , is contributing substantial flux to the xmm exposure , leading to erroneous abundance measurements , or the close - in giant planet , @xmath2 boo b , is somehow affecting the coronal abundances . the latter possibility is easily the more interesting of the two , as it would add a new twist to the current hot topic of star - planet interaction . 1 . spectral type . this is the fbst relation illustrated in figure 6 . there are numerous basic stellar , photospheric , and convection zone properties that are known to be tightly correlated with spectral type for main sequence stars ( stellar radius , temperature , mass , surface gravity , convection zone depth , etc . ) . the existence of the fbst relation demonstrates that somehow one or more of these characteristics also determines coronal abundances . it is worth noting that a spectral type dependence is also known to exist for t tauri stars and very active main sequence stars @xcite . 2 . activity and/or rapid rotation . the fbst relation seems to work very well for the vast majority of main sequence stars . however , there is clearly an activity dependence as well . this is most evident for extremely active stars with @xmath45 , which all lie significantly above the fbst relation in figure 6 @xcite . this activity dependence is also clearly seen in figure 37 of @xcite . an indication that activity dependence of coronal abundances extends to less active stars is the difference in ne / o seen between moderately active stars and truly inactive stars like the sun @xcite . for the activity extremes , it is possible that some aspect of high activity ( e.g. , high magnetic field strength ) is affecting coronal abundances . but it is also possible that the underlying cause of the extreme activity , rapid rotation , is the true determining factor . perhaps very rapid rotation affects photospheric and/or convection zone characteristics , thereby allowing coronal abundances to be altered in the same manner as spectral type affects abundances in the fbst relation . 3 . luminosity class . the best example of this is the @xmath0 ori / procyon dichotomy discussed in section 3.1 . there are evolved stars other than procyon that would lie above the fbst relation in figure 6 , but most are part of active binaries @xcite . for such stars , it is not clear whether the discrepancy from the fbst relation is truly due to luminosity class , or whether it is instead an effect of high activity or in some cases particularly close binarity . is there a common theme that can be found among these abundance determinants ? one avenue to explore concerns photospheric oscillation properties , which are known to depend on both spectral type and luminosity class ( # 1 and # 3 above ) , and may depend on rotation and activity as well ( # 2 ) . if coronal abundance fractionation is controlled by alfvn and other mhd waves traveling through the chromosphere , as suggested by @xcite , the photospheric oscillation characteristics might be expected to be crucial , since these oscillations are likely sources of many of the chromospheric waves . oscillation frequencies increase and amplitudes decrease towards later type stars on the main sequence @xcite , which have higher @xmath6 values . this would imply a potential connection between high @xmath6 and oscillations with high frequency and/or low amplitude . unfortunately , this hypothetical connection is not consistent with the @xmath0 ori / procyon dichotomy . oscillation frequencies decrease and amplitudes increase as stars evolve off the main sequence @xcite , so the aforementioned main sequence oscillation/@xmath6 connection would predict _ lower _ @xmath6 for procyon compared with @xmath0 ori . this is inconsistent with what we observe , so photospheric oscillation properties can not be the sole determinant of coronal abundances . [ [ a - possible - theoretical - framework - for - resolving - the - coronal - abundance - problem ] ] a possible theoretical framework for resolving the coronal abundance problem ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ turning to theoretical considerations , we outline a framework for trying to interpret the observed coronal abundance variations in solar - like stars . @xcite and @xcite argue that the alfvn waves giving rise to the ponderomotive force that generates the solar fip effect most plausibly have a coronal origin . this is because waves in resonance with a coronal loop having period @xmath53 , where @xmath54 is the coronal loop length and @xmath55 is the coronal alfvn speed , are required to produce the observed coronal depletion of he , and the only way to guarantee such a matching is if processes in the corona ( e.g. , nanoflares ) excite normal modes of the loop . @xcite present a `` toy model '' of the inverse fip effect , where fast mode waves propagating up through the chromosphere from below are reflected back down again in the region of the chromosphere where the alfvn speed is increasing with height . this wave population gives a ponderomotive force directed _ downwards _ , which competes with the _ upwards _ ponderomotive force induced by the coronal alfvn waves propagating downwards and reflecting back up . we argue that at the f and g dwarf end of the fbst relation , the positive fip effect from coronal alfvn waves dominates , while at the m dwarf end the inverse fip effect due to initially upward propagating fast mode waves dominates , with a transition between the two somewhere in the k dwarfs . why should this be so ? we discuss first the coronal alfvn waves . these are assumed to arise from coronal nanoflare reconnection events . several authors ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have argued that at least some , and maybe even most , of the magnetic energy released in reconnection should be converted to kinetic energy or waves in the surrounding plasma . it is plausible that such wave generation explains why surveys to find localized hot plasma as evidence of nanoflare reconnection have generally been unsuccessful ( e.g. , * ? ? ? instead , energy goes from magnetic field to waves , and is thus gradually dissipated as heat throughout the corona , and not quickly and locally as might have been expected . @xcite studied the efficiency of electron acceleration in reconnection with the ambient plasma @xmath56 ( the electron plasma @xmath57 , to be more precise ) . they found maximum energy input to electrons at @xmath58 , with reduced electron heating at higher @xmath59 , or lower magnetic field . we suggest that at the left hand side of the fbst relation , coronal reconnection primarily generates aflvn waves that end up causing positive fip fractionation when they reflect from the chromosphere . as one moves to the right , to later spectral type , coronal reconnection puts more energy into electrons , and less into waves . consequently , the positive fip effect diminishes . conversely , it is also possible that the amplitudes of fast mode waves penetrating the chromosphere from below and reflecting back down again may increase toward later spectral types , thus increasing the inverse fip effect . even though p - mode amplitudes decrease towards later type stars @xcite , it is possible that later type stars could be more efficient at converting the p - mode oscillations to fast mode mhd waves . different regions of the sun are different in this respect . it is known that sunspots are sinks of p - mode energy ( see * ? ? ? * and references therein ) , most likely through mode conversion or resonant absorption to alfvn or fast mode waves . there are many cases of sunspots , active regions , and flares observed to have reduced fip effects ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , although it is useful to note that an inverse fip effect has never been reported in any solar observation . an important factor that could act to restrict inverse fip effects to later spectral types is the expansion of the magnetic field within the chromosphere . in their `` toy model , '' @xcite found that the density and magnetic field scale heights , @xmath60 and @xmath61 , were required to satisfy @xmath62 to yield an inverse fip effect . although more realistic models are required to refine this relation , we note that @xmath60 should be smaller for later spectral types due to higher surface gravities , allowing the relation to be met more readily if @xmath61 does not also systematically decrease with spectral type . it is also worth noting that this relation will be met more readily for extremely active stars with magnetic fields so crowded that the fields can not expand in the chromosphere , which effectively means an increase in @xmath61 . this could potentially explain the tendency towards high @xmath6 for extremely active stars ( dependence # 2 in section 3.2 ) . the @xmath0 ori / procyon dichotomy ( see section 3.1 ) remains difficult to explain within this framework . procyon would have larger @xmath60 than @xmath0 ori , which would bias the star towards lower @xmath6 , not higher ( as observed ) , assuming magnetic field expansion is the central issue . if @xmath61 increases by much more than @xmath60 as one moves from the main sequence to more evolved stars , that would reverse this behavior , but we know of no physical reason to expect this to be the case . perhaps procyon has a fundamentally different coronal heating mechanism than the stars that follow the fbst relation . we have argued in section 2.4 that the commonly observed emission measure spike at @xmath63 would be inconsistent with a type ii spicule model of coronal heating and mass supply , if interpreted in terms of a thermal stability argument . but procyon does not possess this peak ( see section 3.1 ) , so a spicule - based coronal heating picture might work for procyon . such a model could also be consistent with procyon s lack of a fip effect , because the spicule flow speed up through the chromosphere would conceivably be too fast to allow fractionation to occur . in this picture , the lower surface gravity of procyon would allow upward propagating spicules to reach higher altitude and presumably higher temperature than on main sequence stars , potentially allowing the spicules to dominate the rather cool coronal emission observed from the star . drake et al . ( 1995 , their subsection 6.2 ) previously argued that the absence of fip effect in the corona of procyon is due to the effect of `` unresolved fine structures '' extending to higher temperatures there than they do on the sun . these speculations provide a framework for trying to understand how coronal abundances depend on stellar parameters . we hope this discussion can serve to focus attention on the critical observational and theoretical points required to subject it to further scrutiny . we would like to thank harry warren for providing us with the solar active region em distribution in figure 2 , and the referee manuel gdel for helpful comments . support for this work was provided by nasa through atp award nnh11aq23i and chandra award number go1 - 12012z issued by the chandra x - ray center ( cxc ) . allende prieto , c. , barklem , p. s. , lambert , d. l. , & cunha , k. 2004 , a&a , 420 , 183 ammler - von eiff , m. , & reiners , a. 2012 , a&a , 542 , a116 arnaud , m. , & rothenflug , r. 1985 , a&as , 60 , 425 asplund , m. , grevesse , n. , sauval , a. j. , & scott , p. 2009 , ara&a , 47 , 481 . audard , m. , behar , e. , gdel , m. , raassen , a. j. j. , porquet , d. , mewe , r. , foley , c. r. , & bromage , g. e. 2001 , a&a , 365 , l329 audard , m. , gdel , m. , sres , a. , raassen , a. j. j. , & mewe , r. 2003 , a&a , 398 , 1137 ball , b. , drake , j. j. , lin , l. , kashyap , v. , laming , j. m. , & garca - alvarez , d. 2005 , apj , 634 , 1336 braun , d. c. 1995 , apj , 451 , 859 brinkman , a. c. , et al . 2001 , a&a , 365 , l324 chaplin , w. j. , houdek , g. , appourchaux , t. , elsworth , y. , new , r. , & toutain , t. 2008 , a&a , 485 , 813 chaplin , w. j. , elsworth , y. , isaak , g. r. , miller , b. a. , & new , r. 2000 , mnras , 313 , 32 colgan , j. , abdullah , j. jr . , sherrill , m. e. , foster , m. , fontes , c. j. , & feldman , u. 2008 , apj , 689 , 585 de pontieu , b. , et al . 2011 , science , 33 , 55 dere , k. p. , landi , e. , mason , h. e. , monsignori fossi , b. c. , & young , p. r. 1997 , a&as , 125 , 149 dere , k. p. , landi , e. , young , p. r. , del zanna , g. , landini , m. , & mason , h. e. 2009 , a&a , 498 , 915 drake , j. f. , swisdak , m. , che , h. , & shay , m. a. 2006 , nature , 443 , 553 drake , j. j. , laming , j. m. , & widing , k. g. 1995 , apj , 443 , 393 drake , j. j. , laming , j. m. , & widing , k. g. 1997 , apj , 478 , 403 drake , j. j. , peres , g. , orlando , s. , laming , j. m. , & maggio , a. 2000 , apj , 545 , 1074 drake , j. j. , & testa , p. 2005 , nature , 436 , 525 dudk , j. , dzifkov , e. , karlick , m. , & kulinov , a. 2011 , a&a , 529 , a103 dudk , j. , kaparov , j. , dzifkov , e. , karlick , m. , & mackovjak , . 2012 , a&a , 539 , a107 feldman , u. , & laming , j. m. 2000 , phys . , 61 , 222 feldman , u. , widing , k. g. , & lund , p. a. 1990 , apj , 364 , l21 fludra , a. , & schmelz , j. t. 1995 , apj , 447 , 936 gdel , m. 2004 , a&ar , 12 , 71 gdel , m. 2007 , living rev . sol . phys . , 4 , 3 ( http://www.livingreviews.org/lrsp-2007-3 ) gdel , m. , et al . 2001 , a&a , 365 , l336 gdel , m. , skinner , s. l. , melnikov , s. y. , audard , m. , telleschi , a. , & briggs , k. r. 2007 , a&a , 468 , 529 huenemoerder , d. p. , canizares , c. r. , drake , j. j. , & sanz - forcada , j. 2003 , apj , 595 , 1131 huenemoerder , d. p. , canizares , c. r. , & schulz , n. s. 2001 , apj , 559 , 1135 jordan , c. 1980 , a&a , 86 , 355 judge , p. g. , de pontieu , b. , mcintosh , s. w. , & kosovare , o. 2012 , apj , 746 , 158 kashyap , v. & drake , j. j. 1998 , apj , 503 , 450 kashyap , v. , & drake , j. j. 2000 , bull . india , 28 , 475 kervella , p. , thvenin , f. , morel , p. , berthomieu , g. , bord , p. , & provost , j. 2004 , a&a , 413 , 251 kigure , h. , takahashi , k. , shibata , k. , yokoyama , t. , & nozawa , s. 2010 , pasj , 62 , 993 kjeldsen , h. , & bedding , t. r. 1995 , a&a , 293 , 87 kjeldsen , h. , & bedding , t. r. 2011 , a&a , 529 , l8 laming , j. m. 2004 , apj , 614 , 1063 laming , j. m. 2009 , apj , 695 , 954 laming , j. m. 2012 , apj , 744 , 115 laming , j. m. , & drake , j. j. 1999 , apj , 516 , 324 laming , j. m. , drake , j. j. , & widing , k. g. 1996 , apj , 462 , 948 laming , j. m. , drake , j. j. , & widing , k. g. 1995 , apj , 443 , 416 lenz , d. d. , deluca , e. e. , golub , l. , rosner , r. , & bookbinder , j. a. 1999 , apj , 517 , l155 liefke , c. , ness , j. -u . , schmitt , j. h. m. m. , & maggio , a. 2008 , a&a , 491 , 859 longcope , d. w. , & tarr , l. 2012 , apj , 756 , 192 maggio , a. , sanz - forcada , j. , & scelsi , l. 2011 , a&a , 527 , a144 martnez - sykora , j. , hansteen , v. , & moreno - insertis , f. 2011 , apj , 736 , 9 mcintosh , s. w. , de pontieu , b. , carlsson , m. , hansteen , v. , boerner , p. , & goossens , m. 2011 , nature , 475 , 477 ness , j. -u . , schmitt , j. h. m. m. , burwitz , v. , mewe , r. , raassen , a. j. j. , van der meer , r. l. j. , predehl , p. , & brinkman , a. c. 2002 , a&a , 394 , 911 perryman , m. a. c. , et al . 1997 , a&a , 323 , l49 phillips , k. j. h. , chifor , c. , & dennis , b. r. 2006 , apj , 647 , 1480 phillips , k. j. h. , pike , c. d. , lang , j. , watanabe , t. , & takahashi , m. 1994 , apj , 435 , 888 porquet , d. , mewe , r. , dubau , j. , raassen , a. j. j. , & kaastra , j. s. 2001 , a&a , 376 , 1113 raassen , a. j. j. , et al . 2002 , a&a , 389 , 228 rakowski , c. e. , & laming , j. m. 2012 , apj , 754 , 65 reale , f. 2010 , living rev . solar phys . , 7 , 5 , url : http://www.livingreviews.org/lrsp-2010-5 robrade , j. , schmitt , j. h. m. m. , & favata , f. 2008 , a&a , 486 , 995 rosner , r. , tucker , w. h. , & vaiana , g. s. 1978 , apj , 220 , 643 sanz - forcada , j. , affer , l. , & micela , g. 2009 , a&a , 505 , 299 sanz - forcada , j. , maggio , a. , & micela , g. 2003 , a&a , 408 , 1087 schmelz , j. t. , nasraoui , k. , rightmire , l. a. , kimble , j. a. , del zanna , g. , cirtain , j. w. , deluca , e. e. , & mason , h. e. 2009 , apj , 691 , 503 schmelz , j. t. , nasraoui , k. , roames , j. k. , lippner , l. a. , & garst , j. w. 2005 , apj , 634 , l197 schmitt , j. h. m. m. , golub , l. , harnden , f. r. , jr . , maxson , c. w. , rosner , r. , & vaiana , g. s. 1985 , apj , 290 , 307 schmitt , j. h. m. m. , & liefke , c. 2004 , a&a , 417 , 651 shkolnik , e. , bohlender , d. a. , walker , g. a. h. , & collier cameron , a. 2008 , apj , 676 , 628 simoniello , r. , finsterle , w. , garca , r. a. , salabert , d. , jimnez , a. , elsworth , y. , & schunker , h. 2010 , a&a , 516 , a30 sturrock , p. a. 1999 , apj , 521 , 451 telleschi , a. , gdel , m. , briggs , k. , audard , m. , ness , j. -u . , & skinner , s. l. 2005 , apj , 622 , 653 telleschi , a. , gdel , m. , briggs , k. r. , audard , m. , & scelsi , l. 2007 , a&a , 468 , 443 veck , n. j. , & parkinson , j. h. 1981 , mnras , 197 , 41 vesecky , j. f. , antiochos , s. k. , & underwood , j. h. 1979 , apj , 233 , 987 von steiger , r. , wimmer - schweingruber , r. f. , geiss , j. , & gloeckler , g. 1995 , adv . space res . , 15(7 ) , 3 walker , g. a. h. , et al . 2008 , a&a , 482 , 691 warren , h. p. , brooks , d. h. , & winebarger , a. r. 2011 , apj , 734 , 90 warren , h. p. , winebarger , a. r. , & brooks , d. h. 2012 , apj , 759 , 141 winebarger , a. r. , schmelz , j. t. , warren , h. p. , saar , s. h. , & kashyap , v. l. 2011 , apj , 740 , 2 wood , b. e. , laming , j. l. , & karovska , m. 2012 , apj , 753 , 76 wood , b. e. , & linsky , j. l. 2006 , apj , 643 , 444 wood , b. e. , & linsky , j. l. 2010 , apj , 717 , 1279 wood , b. e. , redfield , s. , linsky , j. l. , mller , h. -r . , & zank , g. p. 2005 , apjs , 159 , 118 young , p. r. 2005a , a&a , 439 , 361 young , p. r. 2005b , a&a , 444 , l45 lcccc si xiii & 6.648 & 6.99 & @xmath64 & @xmath65 + si xiii & 6.688 & 6.99 & & + si xiii & 6.740 & 6.99 & & + mg xii & 8.419 & 7.11 & @xmath66 & @xmath67 + mg xii & 8.425 & 7.11 & & + mg xi & 9.169 & 6.80 & @xmath68 & @xmath69 + mg xi & 9.231 & 6.80 & & + mg xi & 9.314 & 6.79 & & + ne x & 12.132 & 6.87 & @xmath70 & @xmath71 + ne x & 12.138 & 6.87 & & + fe xvii & 12.264 & 6.62 & @xmath72 & @xmath73 + fe xxi & 12.285 & 6.98 & & + ni xix & 12.435 & 6.69 & @xmath74 & @xmath75 + ne ix & 13.447 & 6.58 & @xmath76 & @xmath77 + ne ix & 13.553 & 6.58 & & + ne ix & 13.699 & 6.58 & @xmath78 & @xmath79 + fe xvii & 13.823 & 6.60 & @xmath80 & @xmath81 + ni xix & 14.043 & 6.68 & @xmath82 & @xmath83 + ni xix & 14.077 & 6.68 & & + fe xviii & 14.203 & 6.74 & @xmath84 & @xmath85 + fe xviii & 14.208 & 6.74 & & + fe xvii & 15.015 & 6.59 & @xmath86 & @xmath87 + fe xvii & 15.262 & 6.59 & @xmath88 & @xmath89 + o viii & 15.176 & 6.65 & & + fe xix & 15.298 & 6.83 & & + o viii & 16.006 & 6.63 & @xmath90 & @xmath91 + fe xviii & 16.005 & 6.73 & & + fe xviii & 16.072 & 6.73 & & + fe xvii & 16.778 & 6.58 & @xmath92 & @xmath93 + fe xvii & 17.053 & 6.58 & @xmath94 & @xmath95 + fe xvii & 17.098 & 6.58 & & + o vii & 18.627 & 6.34 & @xmath96 & @xmath97 + o viii & 18.967 & 6.59 & @xmath98 & @xmath99 + o viii & 18.973 & 6.59 & & + o vii & 21.602 & 6.32 & @xmath100 & @xmath101 + o vii & 21.807 & 6.32 & @xmath102 & @xmath103 + o vii & 22.101 & 6.31 & @xmath104 & @xmath105 + n vii & 24.779 & 6.43 & @xmath106 & @xmath107 + n vii & 24.785 & 6.43 & & + c vi & 33.734 & 6.24 & @xmath108 & @xmath109 + c vi & 33.740 & 6.24 & & + s xiii & 35.667 & 6.43 & @xmath110 & @xmath111 + si xi & 43.763 & 6.25 & @xmath112 & @xmath113 + si xii & 44.019 & 6.44 & @xmath114 & @xmath115 + si xii & 44.165 & 6.44 & @xmath116 & @xmath117 + si xii & 45.521 & 6.44 & @xmath118 & @xmath119 + si xii & 45.691 & 6.44 & @xmath120 & @xmath121 + fe xvi & 46.661 & 6.43 & @xmath122 & @xmath123 + fe xvi & 46.718 & 6.43 & & + si xi & 49.222 & 6.24 & @xmath124 & @xmath125 + fe xvi & 50.361 & 6.43 & @xmath126 & @xmath127 + fe xvi & 50.565 & 6.43 & @xmath128 & @xmath129 + si x & 50.524 & 6.15 & & + si x & 50.691 & 6.15 & @xmath130 & @xmath131 + si xi & 52.298 & 6.24 & @xmath132 & @xmath133 + fe xvi & 54.127 & 6.43 & @xmath134 & @xmath135 + fe xvi & 54.710 & 6.43 & @xmath136 & @xmath137 + mg x & 57.876 & 6.22 & @xmath138 & @xmath139 + mg x & 57.920 & 6.22 & & + fe xv & 59.405 & 6.32 & @xmath140 & @xmath141 + fe xvi & 62.872 & 6.42 & @xmath142 & @xmath143 + mg x & 63.152 & 6.21 & @xmath144 & @xmath145 + mg x & 63.295 & 6.21 & @xmath146 & @xmath147 + fe xvi & 63.711 & 6.42 & @xmath148 & @xmath149 + fe xvi & 66.249 & 6.42 & @xmath150 & @xmath151 + fe xvi & 66.357 & 6.42 & & + fe xv & 69.682 & 6.32 & @xmath152 & @xmath153 + fe xv & 69.941 & 6.32 & @xmath154 & @xmath155 + fe xv & 69.987 & 6.32 & & + fe xv & 70.054 & 6.32 & & + fe xv & 73.472 & 6.32 & @xmath156 & @xmath157 + fe xvi & 76.497 & 6.42 & @xmath158 & @xmath159 + ne viii & 88.082 & 5.96 & @xmath160 & @xmath161 + ne viii & 88.120 & 5.96 & & + fe xviii & 93.923 & 6.68 & @xmath162 & @xmath163 + fe xix & 101.550 & 6.80 & @xmath164 & @xmath165 + fe xviii & 103.937 & 6.68 & @xmath166 & @xmath167 + fe xix & 108.355 & 6.79 & @xmath168 & @xmath169 + fe xx & 121.845 & 6.88 & @xmath170 & @xmath171 + fe xxi & 128.752 & 6.95 & @xmath172 & @xmath173 + fe xx & 132.840 & 6.88 & @xmath174 & @xmath175 + fe xxiii & 132.906 & 7.12 & & + fe xxii & 135.755 & 7.03 & @xmath176 & @xmath177 + fe ix & 171.073 & 5.95 & @xmath178 & @xmath179 + fe x & 174.534 & 6.04 & @xmath180 & @xmath181 + cccc @xmath182 & 1.5 & 1.0 & 1.0 + @xmath21}$ ] & @xmath183 & @xmath184 & @xmath184 + @xmath185}$ ] & @xmath186 & 1.12 & 0.93 + @xmath187}$ ] & @xmath188 & ( 0.49 ) & 0.33 + @xmath189}$ ] & @xmath190 & 1.35 & 1.19 + @xmath37}$ ] & @xmath191 & ( 0.96 ) & 0.43 + @xmath192}$ ] & @xmath193 & 0.08 & 0.10 + @xmath194}$ ] & @xmath195 & 0.07 & 0.01 + @xmath196}$ ] & @xmath197 & ( @xmath198 ) & @xmath199 + @xmath200}$ ] & @xmath201 & @xmath202 & @xmath203 +

we present a level raising result for families of-adic automorphic forms for a definite quaternion algebra over . the main theorem is an analogue of a theorem for classical automorphic forms due to diamond and taylor . we show that certain families of forms old at a prime intersect with families of-new forms ( at a non - classical point ) . one of the ingredients in the proof of diamond and taylor s theorem ( which also played a role in earlier work of taylor ) is the definition of a suitable pairing on the space of automorphic forms . in our situation one can not define such a pairing on the infinite dimensional space of-adic automorphic forms , so instead we introduce a space defined with respect to a dual coefficient system and work with a pairing between the usual forms and the dual space . a key ingredient is an analogue of ihara s lemma which shows an interesting asymmetry between the usual and the dual spaces .

0903.3541

classical level raising results typically show that if the reduction mod @xmath0 of a level @xmath4 modular form @xmath5 has certain properties ( depending on a prime @xmath6 ) , then there exists a modular form @xmath7 of level @xmath8 , new at @xmath3 , with @xmath9 . an example of a level raising result for classical modular forms is the following , due to ribet @xcite : let @xmath10 be an eigenform , and let @xmath11 be a finite place of @xmath12 such that @xmath13 and @xmath5 is not congruent to an eisenstein series modulo @xmath14 . if @xmath15 is a prime number such that the following condition is satisfied , @xmath16 then there exists a @xmath3-new eigenform @xmath17 congruent to @xmath5 modulo @xmath14 . in this paper we prove an analogous level raising result for families of @xmath0-adic automorphic forms . in @xcite and part iii of @xcite , buzzard defines modules of overconvergent @xmath0-adic automorphic forms for definite quaternion algebras , and constructs from these a so - called ` eigencurve ' . the eigencurve is a rigid analytic variety whose points correspond to certain systems of eigenvalues for hecke algebras acting on these modules of automorphic forms . this space @xmath0-adically interpolates the systems of eigenvalues arising from classical automorphic forms . emerton has constructed eigenvarieties in a cohomological framework @xcite , but in the following we will work with buzzard s more concrete construction . since we first wrote the present text , we have also proven some cases of level raising for @xmath0-adic modular forms using the completed cohomology spaces investigated by emerton ( see @xcite ) . the first construction of an eigencurve was carried out for modular forms ( automorphic forms for @xmath18 ) in coleman and mazur s seminal paper @xcite . an important recent result is the construction of a @xmath0-adic jacquet - langlands map between an eigencurve for a definite quaternion algebra and the @xmath18 eigencurve ( interpolating the usual jacquet - langlands correspondence ) , as carried out in @xcite . we follow the general approach of the first part of diamond and taylor s paper @xcite , and our theorem [ supporttheorem ] is an analogue of theorem 1 in that paper , but several new features appear in our work . in particular , the level raising results in @xcite and @xcite for definite quaternion algebras are proved by utilising a pairing on finite dimensional vector spaces of automorphic forms . in our setting , the spaces of automorphic forms are banach modules over an affinoid algebra , so we introduce spaces of ` dual ' automorphic forms and work with the pairing between the usual space of automorphic forms and the dual space . we then prove suitable forms of ihara s lemma , our theorem [ lemma : trivialtorsion ] ( cf . lemma 2 of @xcite ) , for the usual and dual spaces of automorphic forms . an interesting asymmetry between the two situations can be observed . this investigation of level raising results was motivated by a conjecture made by a. paulin , prompted by results on local - global compatibility on the eigencurve in his thesis @xcite . paulin s conjecture was made for the @xmath18-eigencurve ; we may apply our theorem to the image of the @xmath0-adic jacquet - langlands map there to prove many cases of his conjecture . since we have applications to the eigencurve for @xmath19 in mind we work with definite quaternion algebras over @xmath2 in this paper , but some of the methods of section [ mainbody ] should apply to definite quaternion algebras over any totally real number field , although we do use the fact that weight space is one - dimensional in our arguments . we end this introduction by stating the conjecture made by paulin . we fix two distinct primes @xmath0 and @xmath3 , and an integer @xmath4 coprime to @xmath20 . let @xmath21 be the cuspidal eigencurve of tame level @xmath22 , parametrising overconvergent cuspidal @xmath0-adic modular eigenforms ( see @xcite for its construction ) . if @xmath23 is a point of @xmath21 , corresponding to an eigenform @xmath24 , paulin defines an associated representation of @xmath25 , denoted @xmath26 . we call an irreducible connected component @xmath27 of the eigencurve _ generically special _ if the @xmath25-representations associated to the points of @xmath27 away from a discrete set are special . we define _ generically unramified principal series _ similarly . denote by @xmath28 and @xmath29 the roots of the polynomial @xmath30 , where @xmath31 and @xmath32 are the @xmath33 and @xmath34 eigenvalues of @xmath24 . paulin makes the conjecture : suppose @xmath27 is generically unramified principal series . suppose further that there is a point @xmath23 on @xmath27 where the ratio of @xmath28 and @xmath29 becomes @xmath35 and @xmath26 is special . then there exists a generically special component @xmath36 intersecting @xmath27 at @xmath23 . chenevier raised the same question ( in a slightly different form ) in relation to the characterisation of the zariski closure of the @xmath3-new classical forms in the eigencurve . we address this issue in section [ neweigen ] . finally , in a recent preprint @xcite paulin has proved versions of his level raising ( and lowering ) conjectures ( even for ramified principal series ) . his techniques are completely different to ours , making use of deformation theory and requiring a recent important result of emerton showing that the space @xmath37 constructed by kisin in @xcite is equal to the @xmath18-eigencurve ( if one restricts to pieces of the two spaces where certain conditions are satisfied by the relevant mod @xmath0 galois representations ) . in this section we will prove the results we need about modules of @xmath0-adic overconvergent automorphic forms for quaternion algebras . let @xmath38 be a finite extension of @xmath39 . we call a normed @xmath38-algebra @xmath40 a _ banach algebra _ if it satisfies the following properties : * @xmath40 is noetherian , * the norm @xmath41 is non - archimedean , * @xmath40 is complete with respect to @xmath41 , * for any @xmath42 @xmath43 in @xmath40 we have @xmath44 . we will normally assume @xmath40 is a reduced affinoid algebra with its supremum norm . banach @xmath40-module _ is an @xmath40-module @xmath45 endowed with a norm @xmath41 such that * for any @xmath46 , @xmath47 we have @xmath48 , * @xmath45 is complete with respect to @xmath41 . given a set @xmath49 we define the banach @xmath40-module @xmath50 to be functions @xmath51 such that @xmath52 , with norm the supremum norm . by a finite banach @xmath40-module we mean a banach @xmath40-module which is finitely - generated as an abstract @xmath40-module . suppose @xmath45 is a banach module over a banach algebra @xmath40 . we say that @xmath45 is _ onable _ if it is isomorphic ( as a banach module ) to some @xmath50 . note that this terminology differs slightly from that of @xcite , where _ onable _ refers to modules _ isometric _ to some @xmath50 and _ potentially onable _ replaces our notion of onable . the banach @xmath40-module @xmath53 is said to satisfy the universal property @xmath54 if for every surjection @xmath55 of banach @xmath40-modules and continuous map @xmath56 , @xmath28 lifts to a continuous map @xmath57 such that the below diagram commutes : @xmath58^f\\ p\ar@{.>}[ur]^{\exists \beta}\ar[r]^\alpha & n } \ ] ] note that the universal property @xmath54 is not quite the same as the property of being projective in the category of banach @xmath40-modules , since an epimorphism of banach modules is not necessarily a ( set - theoretic ) surjection . a module @xmath53 having property @xmath54 is equivalent to @xmath53 being a direct summand of an onable module . ( see the end of section 2 in @xcite ) . let @xmath0 be a fixed prime . let @xmath1 be a definite quaternion algebra over @xmath2 with discriminant @xmath59 prime to @xmath0 . fix a maximal order @xmath60 of @xmath1 and isomorphisms @xmath61 for primes @xmath62 . note that these induce isomorphisms @xmath63 for @xmath62 . we define @xmath64 , where @xmath65 denotes the finite adeles over @xmath2 . write @xmath66 for the reduced norm map from @xmath67 to @xmath68 . note that if @xmath69 we can regard the @xmath0 component of @xmath7 , @xmath70 , as an element of @xmath71 . for an integer @xmath72 , we let @xmath73 denote the monoid of matrices @xmath74 such that @xmath75 , @xmath76 and @xmath77 . if @xmath78 is an open compact subgroup of @xmath79 and @xmath72 we say that @xmath78 has _ wild level _ @xmath80 if the projection of @xmath78 to @xmath81 is contained in @xmath73 . we will be interested in two key examples of open compact subgroups of @xmath79 . for @xmath45 any integer prime to @xmath59 , we define @xmath82 ( respectively @xmath83 ) to be the subgroup of @xmath79 given by the product @xmath84 , where @xmath85 for primes @xmath86 , and @xmath87 is the matrices in @xmath88 of the form @xmath89 ( respectively @xmath90 ) mod @xmath91 for all other @xmath92 . we can see that if @xmath93 divides @xmath45 , then @xmath83 has wild level @xmath80 . suppose we have @xmath72 , @xmath78 a compact open subgroup of @xmath79 of wild level @xmath80 and @xmath40 a module over a commutative ring @xmath94 , with an @xmath94-linear right action of @xmath73 . we define an @xmath94-module @xmath95 by @xmath96 where @xmath97 is embedded diagonally in @xmath79 . if we fix a set @xmath98 of double coset representatives for the finite double quotient @xmath99 , and write @xmath100 for the finite group @xmath101 , we have an isomorphism ( see section 4 of @xcite ) @xmath102 given by sending @xmath5 to @xmath103 . if @xmath104 for @xmath105 , then the groups @xmath100 are trivial ( this is proved in @xcite ) . for @xmath106 , @xmath107 with @xmath108 , we define @xmath109 by @xmath110 . note that we can now also write @xmath111 we can define double coset operators on the spaces @xmath95 . if @xmath78 , @xmath112 are two compact open subgroups of @xmath79 of wild level @xmath80 , and @xmath40 is as above , then for @xmath113 with @xmath114 we may define an @xmath94-module map @xmath115:\mathcal{l}(u , a ) \rightarrow \mathcal{l}(v , a)$ ] as follows : we decompose @xmath116 into a finite union of right cosets @xmath117 and define @xmath118 = \sum_i f|x_i.\ ] ] let @xmath119 be the rigid analytic space @xmath120 , defined over @xmath39 . the reader may consult lemma 2 of @xcite for details of this space s construction and properties . for example , @xmath119 is a union of finitely many open discs . the space @xmath119 is the _ weight space _ for our automorphic forms . the @xmath121-points @xmath122 of @xmath119 corresponding to characters @xmath123 with @xmath124 for some positive integer @xmath125 and finite order character @xmath126 are referred to as _ classical _ weights . let @xmath127 be a reduced connected @xmath38-affinoid subspace of @xmath119 , where @xmath128 is finite , and denote the ring of analytic functions on @xmath127 by @xmath129 . such a space @xmath127 corresponds to a character @xmath130 induced by the inclusion @xmath131 . if we have a real number @xmath132 for some @xmath133 , then we define @xmath134 to be the rigid analytic subspace of affine @xmath135-space over @xmath38 with @xmath121-points @xmath136 similarly ( for @xmath137 ) we define @xmath138 to be the rigid analytic subspace of affine @xmath135-space over @xmath38 with @xmath121-points @xmath139 a point @xmath140 corresponds to a continuous character @xmath141 . such maps are analytic when restricted to the set @xmath142 for small enough @xmath143 . if @xmath144 and @xmath143 have this property we call @xmath145 an @xmath143-analytic point . a point is @xmath143-analytic if and only if its correponding character extends to a morphism of rigid analytic varieties @xmath146 let @xmath127 be a @xmath38-affinoid subspace of @xmath119 as before , with associated character @xmath130 . we say that @xmath147 is @xmath143-analytic if every point in @xmath148 is @xmath143-analytic . fix a real number @xmath149 and let @xmath150 be the @xmath129-banach algebra @xmath151 , endowed with the supremum norm . if @xmath147 is @xmath152-analytic we can define a right action of @xmath73 on @xmath150 by , for @xmath153 , @xmath154 @xmath155 where @xmath140 ( with @xmath144 the associated character ) and @xmath156 . [ ocformdef]let @xmath127 be a @xmath38-affinoid subspace of @xmath119 as above , with @xmath157 the induced character . if we have a real number @xmath132 , some integer @xmath158 such that @xmath147 is @xmath152-analytic , and @xmath78 a compact open subgroup of @xmath79 of wild level @xmath159 , then define the space of @xmath143-overconvergent automorphic forms of weight @xmath127 and level @xmath78 to be the @xmath129-module @xmath160 if we endow @xmath161 with the norm @xmath162 , then the isomorphism @xmath163 induced by fixing double coset representatives @xmath164 is norm preserving . since the @xmath100 are finite groups , and @xmath150 is an onable banach @xmath129-module ( it is the base change to @xmath129 of @xmath165 , and all banach spaces over a discretely valued field are onable ) , we see that @xmath161 is a banach @xmath129-module , and satisfies property @xmath54 . note that if @xmath166 is a maximal ideal of @xmath129 , corresponding to a point @xmath167 for @xmath168 finite , then taking the fibre of the module @xmath161 at @xmath166 gives the space of overconvergent forms @xmath169 corresponding to the point @xmath145 of @xmath170 ( note that a point of @xmath170 is a reduced connected @xmath171-affinoid subspace ! ) . these spaces of overconvergent automorphic forms were first defined in @xcite , using ideas from the unpublished preprint @xcite . suppose @xmath40 is a banach algebra . given a banach @xmath40-module @xmath172 we define the _ dual _ @xmath173 to be the banach @xmath40-module of continuous @xmath40-module morphisms from @xmath172 to @xmath40 , with the usual operator norm . we denote the @xmath129-module @xmath174 by @xmath175 . if the map @xmath147 corresponding to @xmath127 is @xmath152-analytic , then @xmath73 acts continuously on @xmath150 , so @xmath175 has an @xmath129-linear right action of the monoid @xmath176 given by @xmath177 , for @xmath178 , @xmath179 and @xmath180 . if @xmath78 is as in definition [ ocformdef ] then its projection to @xmath181 is contained in @xmath182 , so it acts on @xmath175 . this allows us to make the following definition : for @xmath127 , @xmath147 , @xmath143 , @xmath28 and @xmath78 as above , we define the space of _ dual _ @xmath143-overconvergent automorphic forms of weight @xmath127 and level @xmath78 to be the @xmath129-module @xmath183 as in [ ocforms ] , we have a norm preserving isomorphism @xmath184 thus @xmath185 is a banach @xmath129-module . we note that it will not usually satisfy property @xmath54 , since ( unless @xmath127 is a point ) we expect that @xmath175 will not be onable . if @xmath78 , @xmath112 are two compact open subgroups of @xmath79 of wild level @xmath80 , then for @xmath113 with @xmath186 we get double coset operators @xmath115:\mathbf{v}^d_{x}(u;r ) \rightarrow \mathbf{v}^d_{x}(v;r)$ ] . for an integer @xmath187 , we define the _ hecke algebra away from _ @xmath187 , @xmath188 , to be the free commutative @xmath129-algebra generated by symbols @xmath189 for @xmath190 prime not dividing @xmath187 . if @xmath191 divides @xmath187 then we can define the usual action of @xmath188 by double coset operators on @xmath192 : for @xmath193 define @xmath194 to be the finite adele which is @xmath190 at @xmath190 and @xmath135 at the other places . abusing notation slightly , we also write @xmath195 for the element of @xmath79 which is @xmath196 at @xmath190 and the identity elsewhere . similarly set @xmath197 to be the element of @xmath79 which is @xmath198 at @xmath190 and the identity elsewhere . on @xmath192 we let @xmath199 act by @xmath200 $ ] and @xmath201 by @xmath202 $ ] . similarly on @xmath203 we define @xmath199 to act by @xmath204 $ ] and @xmath201 by @xmath205 $ ] . as usual we also have a compact operator acting on @xmath192 , namely @xmath206 $ ] . in this section @xmath127 , @xmath147 , @xmath143 , @xmath28 and @xmath78 will be as in definition [ ocformdef ] . we will denote by @xmath112 another compact open subgroup of wild level @xmath80 . we fix double coset representatives @xmath98 for the double quotient @xmath99 and let @xmath207 denote the order of the finite group @xmath101 . we can define an @xmath129-bilinear pairing between the spaces @xmath192 and @xmath203 by @xmath208 where @xmath209 , @xmath210 and on the right hand side of the above definition @xmath211 denotes the pairing between @xmath150 and @xmath175 given by evaluation . this pairing is independent of the choice of the double coset representatives @xmath164 , since for every @xmath212 , @xmath213 , @xmath214 , @xmath215 and @xmath210 we have @xmath216 combining this observation with the isomorphisms ( [ iso ] ) and ( [ dualiso ] ) we see that our pairing identifies @xmath203 with @xmath217 . the following proposition summarises a standard computation @xcite ( although these assume the level group is small enough that the finite groups @xmath100 are trivial ) , telling us how our pairing interacts with double coset operators . in particular , it implies that @xmath218 for @xmath219 when @xmath199 acts in the usual way . [ prop : pairinghecke ] let @xmath209 and let @xmath220 . let @xmath213 with @xmath221 . then @xmath222,\lambda\rangle=\langle f,\lambda|[vg^{-1}u]\rangle.\ ] ] for @xmath223 set @xmath224 . we have @xmath225=\sum_{v\in ( g^{-1 } u\eta)\cap v \backslash v}f|(g v),\ ] ] hence @xmath226,\lambda\rangle&=&\sum_{d\in d^\times\backslash d^\times_f / v } \gamma(d)^{-1}\langle f|[ug v](d),\lambda(d)\rangle\\ & = & \sum_{d\in d^\times\backslash d^\times_f / v}\sum_{v\in ( g^{-1 } ug)\cap v \backslash v } \gamma(d)^{-1}\langle f|(g v)(d),\lambda(d)\rangle\\ & = & \sum_{d\in d^\times\backslash d^\times_f / v}\sum_{v\in ( g^{-1 } ug)\cap v \backslash v } \gamma(d)^{-1}\langle f(dv^{-1}g^{-1})\cdot g_p v_p,\lambda(d)\rangle\\ & = & \sum_{x\in d^\times\backslash d^\times_f/(g^{-1 } ug)\cap v}\langle f(xg^{-1}),\lambda(x)\cdot g_p^{-1}\rangle\\ & = & \sum_{y\in d^\times\backslash d^\times_f / u\cap(g v g^{-1})}\langle f(y),\lambda(yg)\cdot g_p^{-1}\rangle\\ & = & \langle f,\lambda|[vg^{-1}u]\rangle\end{aligned}\]]where we pass from the third line to the fourth line by counting double cosets and the final line follows by similar calculations to the first @xmath227 lines . by the results of section 5 in @xcite and section 3 of @xcite we know that for a fixed @xmath228 , if @xmath127 is a sufficiently small affinoid whose norm is multiplicative ( with a precise bound given by thorme 5.3.1 of @xcite ) then since @xmath229 acts as a compact operator on @xmath192 , we have a @xmath229 stable decomposition @xmath230 where @xmath231 is the space of forms of slope @xmath232 . we need @xmath127 to be small enough that the newton polygon of the characteristic power series for @xmath229 acting on @xmath192 has the same slope @xmath233 part when specialised to any point of @xmath127 . from now on we fix @xmath234 and assume that @xmath127 is such that this slope decomposition exists . the key example of such an @xmath127 is an open ball of small radius . the space @xmath231 is a finite banach @xmath129-module with property @xmath54 , i.e. a projective finitely generated @xmath129-module . in fact this decomposition must be stable under the action of @xmath235 , since the @xmath199 and @xmath201 operators for @xmath236 commute with @xmath229 . we define @xmath237 to be the maps from @xmath192 to @xmath129 which are @xmath238 on @xmath239 . this space is also stable under the action of @xmath235 and is naturally isomorphic to the dual of @xmath231 . the following lemma implies that our pairing is perfect when restricted to @xmath240 . let @xmath172 be a finite banach @xmath129-module with property @xmath54 . then the usual natural map @xmath241 is an isomorphism . in other words , the @xmath129-module @xmath172 is _ reflexive_. since @xmath172 is finite we have a surjection of banach @xmath129-modules @xmath242 for some @xmath133 . applying the universal property @xmath54 to this surjection shows that we have a banach @xmath129-isomorphism @xmath243 for some module @xmath239 , so @xmath172 is a projective @xmath129-module . proposition 2.1 of @xcite states that the category of finite banach @xmath129-modules , with continuous @xmath129-linear maps as morphisms , is equivalent to the category of finite @xmath129-modules so we can just compute duals module - theoretically . we have exact sequences @xmath244&\mathbf{m}\ar[r]&{\ensuremath{\mathcal{o}}\xspace}(x)^{\oplus n}\ar[r]&\mathbf{n}\ar[r]&0}\ ] ] @xmath244&\mathbf{n}\ar[r]&{\ensuremath{\mathcal{o}}\xspace}(x)^{\oplus n}\ar[r]&\mathbf{m}\ar[r]&0}\ ] ] and since @xmath172 , @xmath239 , @xmath173 and @xmath245 are all projective as @xmath129-modules we can take the dual of these exact sequences twice to get commutative diagrams with exact rows @xmath246&\mathbf{m}\ar[r]\ar[d]&{\ensuremath{\mathcal{o}}\xspace}(x)^{\oplus n}\ar[r]\ar[d]&\mathbf{n}\ar[r]\ar[d]&0\\ 0\ar[r]&\mathbf{m}^{**}\ar[r]&{\ensuremath{\mathcal{o}}\xspace}(x)^{\oplus n}\ar[r]&\mathbf{n}^{**}\ar[r]&0\\ } \ ] ] @xmath246&\mathbf{n}\ar[r]\ar[d]&{\ensuremath{\mathcal{o}}\xspace}(x)^{\oplus n}\ar[r]\ar[d]&\mathbf{m}\ar[r]\ar[d]&0\\ 0\ar[r]&\mathbf{n}^{**}\ar[r]&{\ensuremath{\mathcal{o}}\xspace}(x)^{\oplus n}\ar[r]&\mathbf{m}^{**}\ar[r]&0\\ } \ ] ] where the vertical maps are the natural maps from a module to its double dual . since the central maps are isomorphisms , we conclude that the outer maps are too . we should remark here that our use of the dual banach modules @xmath203 is slightly unsatisfactory . for example , the modules do not satisfy property @xmath54 , and we must restrict to ` slope @xmath232 ' subspaces to get a perfect pairing . one could alternatively work with modules of _ all _ overconvergent automorphic forms , rather than imposing @xmath143-overconvergence for a particular @xmath143 . one defines @xmath247 where the ( compact ) transition maps in the direct system are induced by the inclusions @xmath248 for @xmath249 . if @xmath127 is a point ( so @xmath129 is a field ) then it is a standard result that the vector space @xmath250 is reflexive ( see proposition 16.10 of @xcite ) . using this fact it is fairly straightforward to show that for any @xmath127 , @xmath250 is a reflexive @xmath129-module , with dual the frchet space @xmath251 fix an integer @xmath252 ( the tame level ) coprime to @xmath0 and fix an auxiliary prime @xmath253 . let @xmath127 be an affinoid subspace of weight space with associated character @xmath147 which is @xmath152 analytic , for some integer @xmath158 . set @xmath254 , @xmath255 . to simplify notation we set @xmath256 we define a map @xmath257 by @xmath258+g|[u \eta_l v].\ ] ] since the map @xmath259 is defined by double coset operators with trivial component at @xmath0 it commutes with @xmath229 and thus gives a well defined map between these spaces of bounded slope forms . a simple calculation shows that these double coset operators act very simply . regarding @xmath5 and @xmath7 as functions on @xmath260 we have @xmath261=f$ ] , @xmath262=g|\eta_l$ ] . the image of @xmath259 inside @xmath45 will be referred to as the space of _ we also define a map @xmath263 by @xmath264,f|[v \eta_l^{-1 } u]).\ ] ] the kernel of @xmath265 is the space of _ newforms_. the maps @xmath259 and @xmath265 commute with hecke operators @xmath266 , where @xmath267 . the same double coset operators give maps @xmath268 @xmath269 using proposition [ prop : pairinghecke ] we have @xmath270 for @xmath271 , @xmath272 . similarly @xmath273 for @xmath274 , @xmath275 . an easy calculation shows that @xmath276 acts on the product @xmath277 by the matrix ( acting on the right ) @xmath278[u\eta_l u ] \\ [ u\eta_l u ] & l+1 \end{pmatrix}=\begin{pmatrix}l+1 & s_l^{-1}t_l \\ t_l & l+1 \end{pmatrix}.\ ] ] we have exactly the same double coset operator formula for the action of @xmath279 on the product @xmath280 . since the hecke operators @xmath281 act by @xmath282 , [ u\eta_l^{-1 } u]$ ] respectively on @xmath283 we deduce that , in terms of hecke operators , @xmath279 acts on @xmath284 by the matrix ( again acting on the right ) @xmath285 if the affinoid @xmath127 is sufficiently nice , then we can show that the map @xmath276 is injective . before we prove this , we note that in our setting a family of @xmath0-adic automorphic eigenforms over an affinoid @xmath131 is just a hecke eigenform @xmath5 in @xmath192 . [ propinj ] if @xmath127 is a one dimensional irreducible connected smooth affinoid , then the map @xmath276 is injective . let @xmath286 be the projective ( since @xmath129 is a dedekind domain ) finite banach @xmath129-module @xmath287 , and note that @xmath288 is stable under the action of all the hecke operators , since they all commute with @xmath276 . suppose @xmath286 is not zero . for @xmath289 in @xmath286 we have @xmath290 . eliminating @xmath7 we get @xmath291 , so projecting @xmath286 down to @xmath292 ( taking either the first or the second factor ) we see that the hecke operator @xmath293 acts as @xmath238 on a non - zero projective submodule of @xmath292 . this ( applying the local eigenvariety construction as described in section 6.2 of @xcite ) implies that there is a family of eigenforms over some one dimensional sub - affinoid of @xmath127 , all with the eigenvalue of @xmath293 equal to @xmath238 . now the hecke algebra element @xmath293 induces a rigid analytic function on the tame level @xmath4 eigencurve for @xmath1 ( by taking the appropriate eigenvalue associated to a point ) , so this function must vanish on the whole irreducible component containing the one dimensional family constructed above . however , every irreducible component contains a classical point , and these can not be contained in the kernel of @xmath293 since this would contradict the hecke eigenvalue bounds given by the ramanujan - petersson conjecture . note that the injectivity of @xmath276 implies the injectivity of @xmath259 . the above shows that if @xmath127 is as in the statement of proposition [ propinj ] , we have @xmath294 so our families in @xmath45 are not both old and new at @xmath3 . however , if @xmath127 is just a point , then @xmath276 may have a kernel - this corresponds to @xmath0-adic automorphic forms which are both old and new at @xmath3 . we denote the fraction field of @xmath129 by @xmath295 . if @xmath40 is an @xmath129-module we write @xmath296 for the @xmath295-vector space @xmath297 . we begin this section by noting that the injectivity of @xmath276 implies the injectivity of @xmath279 : suppose @xmath298 . then @xmath299 for all @xmath300 , so ( by proposition [ prop : pairinghecke ] ) @xmath301 for all @xmath300 . now since @xmath302 is an injective endomorphism of a finite dimensional vector space , it is an isomorphism , so we see that @xmath303 . hence @xmath279 ( thus a fortiori @xmath304 ) is injective . we now define two chains of modules which will prove useful : @xmath305 we note that @xmath306 , and that @xmath307 , with analogous statements for the starred modules . we fix the usual action of @xmath308 on all these modules . we can now describe some pairings between them which will be equivariant under the @xmath308 action . they will not all be equivariant with respect to the action of @xmath33 . we have a ( perfect ) pairing @xmath309 which , since @xmath304 is injective , induces a pairing @xmath310 which in turn induces a pairing @xmath311 the fact that this pairing is perfect follows from the following lemma : the pairing on @xmath312 induces isomorphisms @xmath313 and @xmath314 for the first isomorphism , the module @xmath315 correspond to @xmath316 such that @xmath317 for all @xmath47 . we have @xmath318 so @xmath317 for all @xmath47 if and only if @xmath319 for all @xmath47 , i.e. if and only if @xmath320 . the second isomorphism is obvious , since @xmath304 is injective . in exactly the same way , we have a perfect pairing @xmath321 the final pairing we will need is induced by the pairing between @xmath45 and @xmath322 . it is straightforward to check that this gives a perfect pairing : @xmath323 in classical level raising results ( such as @xcite ) analogues of ` ihara s lemma ' ( lemma 3.2 in @xcite ) are used to show that prime ideals of a hecke algebra containing the annihilators of certain modules of automorphic forms are in some sense ` uninteresting ' , or even to show that these modules are trivial . in this section we prove the appropriate analogue of ihara s lemma in our setting . from this section onwards we will assume that @xmath127 is a one dimensional irreducible connected smooth affinoid in weight space @xmath119 , so we can apply proposition [ propinj ] . we want to obtain information about the @xmath308 action on the quotients @xmath324 , @xmath325 , @xmath326 and @xmath327 . the pairings @xmath328 and @xmath329 allow us to use an analogue of ihara s lemma ( the following two propositions and theorem ) to obtain crucial information about all four quotients . recall that the radius of overconvergence @xmath143 equals @xmath330 for some positive integer @xmath133 . fix a positive integer @xmath331 such that @xmath332 contains all elements of @xmath333 congruent to @xmath135 modulo @xmath331 . we first need a lemma allowing us to control certain forms with weight a point in weight space . [ normtrivial ] let @xmath334 for some @xmath171 a finite extension of @xmath39 . 1 . let @xmath335 be non - zero . suppose @xmath43 factors through @xmath66 , that is @xmath336 for all @xmath337 with @xmath338 . then @xmath144 is a classical weight @xmath339 , and for all but finitely many primes @xmath340 mod @xmath331 , ( where @xmath331 is the fixed integer chosen above ) , @xmath341 . 2 . let @xmath342 . if @xmath43 factors through @xmath66 , that is @xmath336 for all @xmath337 with @xmath338 , then @xmath43 is zero . we first prove part ( i ) . suppose @xmath43 is as in the statement of that part . for @xmath343 we have @xmath344 for all @xmath213 . noting that @xmath345 for all @xmath346 , we see that @xmath347 for all @xmath346 , @xmath348 so @xmath349 is constant in @xmath350 , since non - constant rigid analytic functions have discrete zero sets . recall that @xmath254 , so @xmath351 is in @xmath352 , and for @xmath348 we have @xmath353 so @xmath144 must correspond to the classical weight given by @xmath354 for some character @xmath355 trivial on @xmath356 , where @xmath132 . this now implies that for each @xmath213 we have @xmath357 for all @xmath358 in the projection of @xmath359 to @xmath360 , since these matrices all have bottom right hand entry congruent to @xmath361 . we now follow @xcite to complete the proof of the first part of the lemma . there is a @xmath362 with @xmath363 , so @xmath364 is actually in @xmath333 and is congruent to @xmath135 mod @xmath331 . thus ( by the way we picked @xmath331 ) there is @xmath365 such that @xmath366 . now we have @xmath367 where to pass from the first line to the second we use the fact that @xmath43 factors through @xmath66 to commute @xmath43 s arguments , from the second to the third we use that @xmath43 is modular of level @xmath78 and in the final line we first substitute @xmath368 for @xmath369 ( since they have the same reduced norm ) , then commute @xmath43 s arguments and use the left invariance of @xmath43 under @xmath97 followed by the fact that @xmath370 implies that @xmath371 . we now give a proof of the second part of the lemma . first we perform a formal calculation . fix an isomorphism @xmath372 where @xmath373 is the ring of power series with coefficients in @xmath171 tending to zero ( @xmath374 a formal variable ) , and @xmath133 is a positive integer depending on @xmath143 . such an isomorphism exists since @xmath375 is just a disjoint union of finitely many affinoid discs . we then have an identification of @xmath376 with @xmath377\rangle$ ] , where @xmath378\rangle$ ] denotes the ring of power series with bounded coefficients in @xmath171 , and the pairing between @xmath379 and @xmath376 is given on each component by @xmath380 . now we can compute the action of @xmath381 on @xmath376 . let @xmath382 be an element of @xmath376 , with @xmath383 for each @xmath384 and @xmath385 fix @xmath386 to be the element of @xmath379 which is @xmath387 at the @xmath28 component , and zero elsewhere . we have @xmath388 we can now see that if we have @xmath389 we get @xmath390 for all @xmath259 and @xmath28 , which implies that @xmath391 . now we return to the statement in the lemma and suppose @xmath342 factors through @xmath66 . let @xmath392 be the element of @xmath393 with @xmath0 component equal to @xmath358 and all other components the identity . for all @xmath394 , @xmath395 since @xmath396 and @xmath397 since @xmath398 . hence @xmath399 and the above calculation shows that @xmath400 for all @xmath7 . the following two propositions apply the preceding lemma to give a form of ihara s lemma for modules of overconvergent automorphic forms and dual overconvergent forms . [ preihara ] for all but finitely many primes @xmath340 mod @xmath331 , @xmath401 annihilates the module @xmath402 for each maximal ideal @xmath166 of @xmath129 . fix @xmath403 mod @xmath331 with @xmath404 and set @xmath405 . let @xmath166 be a maximal ideal of @xmath129 and set @xmath406 . the maximal ideal @xmath166 corresponds to a point @xmath145 of @xmath407 , with corresponding weight @xmath408 the specialisation of @xmath147 at @xmath166 . we have a short exact sequence @xmath246&l^2\ar[r]^i&m\ar[r]&m / il^2\ar[r]&0 } .\ ] ] noting that @xmath409 and @xmath45 are @xmath129-torsion free , hence flat , and taking derived functors of @xmath410 gives an exact sequence @xmath246&\mathrm{tor}^{{\ensuremath{\mathcal{o}}\xspace}(x)}_1(m / il^2,k')\ar[r]^\delta&l^2\otimes_{{\ensuremath{\mathcal{o}}\xspace}(x)}k'\ar@/^2pc/[d]^i\\ 0&m / il^2\otimes_{{\ensuremath{\mathcal{o}}\xspace}(x)}k'\ar[l]&m\otimes_{{\ensuremath{\mathcal{o}}\xspace}(x)}k'\ar[l ] } .\ ] ] we have a commutative diagram @xmath246&l^2\ar[r]^i\ar[d]^{h_q}&m\ar[r]\ar[d]^{h_q}&m / il^2\ar[r]\ar[d]^{h_q}&0\\ 0\ar[r]&l^2\ar[r]^i&m\ar[r]&m / il^2\ar[r]&0\\ } , \ ] ] so by the naturality of the long exact sequence for @xmath411 the diagram @xmath412^\delta\ar[d]^{h_q}&l^2\otimes_{{\ensuremath{\mathcal{o}}\xspace}(x)}k'\ar[d]^{h_q}\\ \mathrm{tor}^{{\ensuremath{\mathcal{o}}\xspace}(x)}_1(m / il^2,k')\ar[r]^\delta&l^2\otimes_{{\ensuremath{\mathcal{o}}\xspace}(x)}k ' } \ ] ] commutes . to complete the proof it suffices to prove that @xmath413 annihilates the kernel of @xmath414 we proceed by viewing these modules as spaces of automorphic forms with weight @xmath145 ( a single point in weight space ) . we define two finite dimensional @xmath171-vector spaces : @xmath415 there are maps @xmath416 and @xmath417 as defined in section [ oldnew ] ( taking the weight @xmath127 in that section to be the point @xmath145 ) , but note that as now the weight is just a point in weight space , proposition [ propinj ] does not apply . in particular the map @xmath418 might not be injective . recall that @xmath292 and @xmath45 are finitely generated @xmath129 modules , whence @xmath419 and @xmath420 are finite dimensional @xmath171-vector spaces . since the newton polygon of the characteristic power series for @xmath229 acting on @xmath192 has the same slope @xmath233 part when specialised to any point of @xmath127 , and the specialisation of the banach @xmath129-modules @xmath192 and @xmath421 at @xmath166 gives @xmath422 and @xmath423 respectively , we have isomorphisms @xmath424 and @xmath425 which commute suitably with @xmath426 and @xmath416 . these isomorphisms also commute with double coset operators , so to prove the proposition it suffices show that @xmath413 annihilates the kernel of @xmath418 . suppose @xmath427 . then @xmath428 , so we have @xmath429 , @xmath430 . therefore @xmath431 and @xmath432 are both invariant under the action of the group @xmath78 , so @xmath431 is invariant under the action of the group generated by @xmath78 and @xmath433 in @xmath79 . ( note that every element of this group has projection to its @xmath0th component lying in @xmath73 . ) since by ii.1.4 , corollary 1 of @xcite @xmath434 and the @xmath3-factor of @xmath78 is @xmath435 , we have that @xmath431 is invariant under @xmath436 , where we embed @xmath436 into @xmath79 in the obvious way . denote by @xmath437 the algebraic subgroup of @xmath97 whose elements are of reduced norm @xmath135 . we have @xmath438 , since @xmath1 is split at @xmath3 . the strong approximation theorem applied to @xmath437 implies that @xmath439 is dense in @xmath440 , where @xmath441 is embedded diagonally in @xmath442 . for each @xmath443 we define @xmath444 since @xmath431 is continuous , @xmath445 is closed , and for @xmath446 , @xmath447 we have @xmath448 , since @xmath449 . therefore @xmath445 contains the dense set @xmath450 , so @xmath445 is the whole of @xmath442 . this shows that @xmath431 factors through @xmath66 . now the first part of lemma [ normtrivial ] applies . [ preiharadual ] the module @xmath451 is @xmath238 for all maximal ideals @xmath166 of @xmath129 we again set @xmath452 , and let @xmath167 be the point corresponding to @xmath166 . proceeding as at the beginning of the proof of proposition [ preihara ] we see that we must show that the map @xmath453 is injective . we define @xmath454 as in the previous proposition , it is sufficient to show that the map @xmath455 is injective . now we continue as in the proof of proposition [ preihara ] , and finally apply the second part of lemma [ normtrivial ] . the following consequence of the preceding two propositions will be the most convenient analogue of ihara s lemma for our applications . [ lemma : trivialtorsion ] 1 . there is a positive integer @xmath456 such that for all but finitely many primes @xmath457 , @xmath458 annihilates @xmath459 . therefore these hecke operators annihilate the modules @xmath460 and , by consideration of the pairing @xmath329 , @xmath327 . the module @xmath461 is equal to @xmath238 . therefore the modules @xmath462 and , by consideration of the pairing @xmath328 , @xmath326 are also equal to @xmath238 . we first prove the first part of the theorem . fix a @xmath463 with @xmath404 . the module @xmath459 is finitely generated ( and torsion ) over the dedekind domain @xmath129 , so it is isomorphic as an @xmath129-module to @xmath464 for some finite set of maximal ideals @xmath465 in @xmath129 . we set @xmath456 to be the maximum of the @xmath466 . set @xmath405 as before . we will show that @xmath467 annihilates @xmath459 . indeed , suppose that @xmath47 represents a nonzero torsion class in @xmath468 . thus , there exists a nonzero @xmath469 such that @xmath470 . we have @xmath471 . so we are required to prove that @xmath472 annihilates @xmath473 . since @xmath474 we can assume that @xmath475 , so it is enough to show that @xmath472 annihilates @xmath476 taking derived functors of @xmath477 of the short exact sequence @xmath246&{\ensuremath{\mathfrak{m}}\xspace}_i/{\ensuremath{\mathfrak{m}}\xspace}_i^{e_i}\ar[r]&{\ensuremath{\mathcal{o}}\xspace}(x)/{\ensuremath{\mathfrak{m}}\xspace}_i^{e_i}\ar[r]&{\ensuremath{\mathcal{o}}\xspace}(x)/{\ensuremath{\mathfrak{m}}\xspace}_i\ar[r]&0 } \ ] ] and applying proposition [ preihara ] and induction on @xmath466 ( note that @xmath478 is isomorphic to @xmath479 ) , we see that for each @xmath259 , @xmath480 annihilates @xmath481 . now @xmath482 for all @xmath259 , so @xmath472 annihilates all of @xmath473 . since @xmath28 was arbitrary , @xmath472 annihilates @xmath459 . the second part of the theorem follows easily from proposition [ preiharadual ] . we will call a maximal ideal @xmath483 in the hecke algebra @xmath484 _ eisenstein _ ( compare @xcite ) if there is some positive integer @xmath331 such that for all but finitely many primes @xmath457 , @xmath485 . the motivation for this definition is the following well - known lemma : suppose we have a galois representation @xmath486 and a positive integer @xmath331 such that for all but finitely many primes @xmath463 , the trace of frobenius at @xmath92 , @xmath487 . then @xmath488 is reducible . the formula for the traces implies that @xmath488 restricted to the cyclotomic field @xmath489 is isomorphic to @xmath490 , where @xmath491 is the trivial one dimensional representation , and @xmath492 is the @xmath3-adic cyclotomic character . denoting @xmath493 by @xmath494 and @xmath495 by @xmath496 and applying frobenius reciprocity we conclude that @xmath497 is non - zero . @xmath498 is just a direct sum of one dimensional representations , so @xmath488 is reducible . we recall that for an arbitrary ring @xmath94 the _ support _ of an @xmath94-module @xmath40 is the set of prime ideals @xmath499 such that the localisation @xmath500 is non - zero . if @xmath40 is finitely generated as an @xmath94-module then the support of @xmath40 is equal to the set of prime ideals in @xmath94 containing the annihilator of @xmath40 . we write @xmath501 for the image of @xmath484 in @xmath502 and similarly @xmath503 for the image of @xmath504 in @xmath505 . analogously we define @xmath506 and @xmath507 to be the image of the hecke algebra in the endomorphism rings of the relevant dual modules . note that there are natural embeddings @xmath508 and @xmath509 . if @xmath49 is an ideal of @xmath510 we write @xmath511 for the image of @xmath49 in @xmath501 and @xmath512 for the image of @xmath513 in @xmath503 we can now state and prove the main theorem of this section . [ supporttheorem]suppose @xmath483 is a non - eisenstein maximal ideal of @xmath510 containing @xmath293 . further suppose that @xmath514 is in the support of the @xmath501-module @xmath292 . then @xmath515 is in the support of the @xmath503-module @xmath516 . we write @xmath517 for the image of @xmath483 in @xmath506 and @xmath518 for the image of @xmath519 in @xmath507 . since @xmath514 is in the support of @xmath292 , and the perfect pairing between @xmath292 and @xmath283 is equivariant with respect to all of @xmath510 ( including @xmath33 ) , we know that @xmath517 is in the support of @xmath283 . consider the module @xmath520 since @xmath521 and @xmath522 we know that @xmath523 , i.e. @xmath517 is in the support of @xmath524 . we can view @xmath524 as a @xmath507-module , with @xmath525 in its support . theorem [ lemma : trivialtorsion ] implies that if @xmath525 is in the support of @xmath327 or @xmath462 then it is eisenstein , so it must be in the support of @xmath526 . this quotient is a homomorphic image of @xmath527 , so @xmath525 is in the support of @xmath527 . finally we can apply pairing @xmath528 ( which is equivariant with respect to @xmath529 ) to conclude that @xmath530 is in the support of @xmath531 . in this section we explain some applications of the preceding results , including the proof of some cases of the conjecture mentioned in the introduction . we firstly describe the application of theorem [ supporttheorem ] to the conjecture of our introduction . chenevier @xcite extended the classical jacquet - langlands correspondence to a rigid analytic embedding from the eigencurve for a definite quaternion algebra to some part of the @xmath18 eigencurve . we may use this to translate the results of the previous section to the @xmath18 eigencurve . we state the case of theorem @xmath532 in @xcite that we will use . we have primes @xmath533 and a coprime integer @xmath534 . fix another prime @xmath92 and a character @xmath535 of @xmath536 . let @xmath21 be the tame level @xmath537 and character @xmath535 reduced cuspidal eigencurve . let @xmath538 be a quaternion algebra ramified at the infinite place and at @xmath92 , and let @xmath539 be the corresponding reduced eigencurve of tame level @xmath540 and character @xmath535 . there is a closed rigid analytic immersion @xmath541 whose image is the zariski closure of the classical points of @xmath21 that are new at @xmath92 . this map is defined over weight space and is hecke equivariant . we get the same result if we change the level of @xmath21 to @xmath542 ( call this eigencurve @xmath543 ) and change the level of @xmath539 to @xmath544 ( call this eigencurve @xmath545 ) , where we construct these eigencurves using the hecke operators at @xmath3 in addition to the usual hecke operators away from the level . this allows us to relate @xmath543 and the two - covering @xmath546 of @xmath21 corresponding to taking roots of @xmath3th hecke polynomial . [ old ] there is a closed embedding @xmath547 , with image the zariski closure of the classical @xmath3-old points in @xmath543 . given @xmath127 an affinoid subdomain of @xmath119 , let @xmath548 and @xmath549 denote the banach @xmath129-modules of families ( weight varying over @xmath127 ) of overconvergent modular forms of tame levels @xmath537 and @xmath542 respectively , as defined in section 7 of @xcite . the two degeneracy maps from level @xmath542 to level @xmath537 give a natural embedding @xmath550 . denote the image of this map by @xmath551 - it is stable under all hecke operators ( including at @xmath3 ) . the lemma follows from the observation that applying the eigenvariety machine of @xcite to the banach modules @xmath551 ( with @xmath127 varying ) gives the space @xmath546 . let @xmath552 be the zariski closure of the classical points in @xmath543 corresponding to forms new at @xmath3 . proposition 4.7 of @xcite shows that @xmath27 can be identified with the points of @xmath543 lying in a one dimensional family of _ @xmath3-new _ points , where @xmath3-new means they come from overconvergent modular forms in the kernel of the map analogous to @xmath265 in the @xmath553 setting . the following theorem corresponds to the conjecture in the introduction for points of @xmath543 in the image of @xmath554 . [ raise ] suppose we have a point @xmath555 lying in the image of @xmath554 , with @xmath556 . let the roots of the @xmath3th hecke polynomial corresponding to @xmath23 be @xmath28 and @xmath557 where @xmath558 . then the point over @xmath23 of @xmath559 corresponding to @xmath28 lies in @xmath27 . we pick @xmath234 , @xmath143 and @xmath28 such that the automorphic form corresponding to the preimage of @xmath23 under @xmath554 is @xmath143-overconvergent of slope @xmath233 and level @xmath560 . now fix a closed ball in @xmath119 , containing the weight of @xmath23 , which is small enough ( note that ` small enough ' depends on @xmath234 , @xmath143 and @xmath28 ) to apply the local eigenvariety construction described in section 6.2 of @xcite . denote this ball by @xmath127 . as usual the @xmath152-analytic character @xmath561 induced by the embedding @xmath562 is denoted by @xmath147 . the system of hecke eigenvalues given by @xmath23 corresponds to a maximal ideal @xmath483 in the @xmath129 hecke algebra @xmath563 . we know that @xmath564 . if we set @xmath565 as before , and use the notation of the previous section , then @xmath514 is in the support of @xmath292 . at this stage theorem [ supporttheorem ] applies to the ideal @xmath483 , which is not eisenstein since the galois representation attached to @xmath23 is irreducible ( recall that we are working on the cuspidal part of the eigencurve ) . therefore we know that @xmath515 is in the support of @xmath516 . we can then take a height one prime ideal @xmath566 in the support of @xmath516 , and then this corresponds ( by proposition 6.2.4 of @xcite ) to a @xmath0-adic family of automorphic forms , new at @xmath3 , passing through a point @xmath567 with system of hecke eigenvalues the same as those for @xmath23 away from @xmath3 . now applying the map @xmath554 we see that one of the points over @xmath23 must lie in @xmath27 . a calculation using the fact that @xmath567 comes from an eigenform in the kernel of @xmath265 shows that the point corresponds to the root @xmath28 . to translate this theorem into the language of the introduction , note that @xmath559 corresponds to the generically unramified principal series components of @xmath543 , whilst @xmath27 corresponds to the generically special or supercuspidal components . we may identify the point over @xmath23 lying in @xmath27 as the one whose attached @xmath25-representation is special . we return to the situation of a definite quaternion algebra @xmath1 over @xmath2 with arbitrary discriminant @xmath59 prime to @xmath0 . denote the levels @xmath560 by @xmath78 and @xmath568 by @xmath112 , as before . we denote by @xmath539 the tame level @xmath544 reduced eigencurve for @xmath1 . suppose @xmath23 is a point of @xmath539 , with weight @xmath145 . we say that @xmath23 is _ @xmath3-new _ if it corresponds to a hecke eigenform in the kernel of the map @xmath569 , where this is defined as in section [ oldnew ] by @xmath570,f|[v \eta_l^{-1 } u])$ ] . denote by @xmath27 the zariski closure of the points in @xmath539 arising from _ classical _ @xmath3-new forms . we have the following proposition , due to chenevier . 1 . the set of @xmath145 in @xmath539 that are @xmath3-new is the set of points of a closed reduced analytic subspace @xmath571 . @xmath27 is a closed subspace of @xmath572 , and its complement is the union of irreducible components of dimension @xmath238 in @xmath572 . a point of @xmath572 lies in @xmath27 if and only if it lies in a one - dimensional family of points in @xmath572 . exactly as for proposition 4.7 of @xcite . we now apply the results of section [ mainbody ] to show that a point of @xmath573 lies in a one - dimensional family of points in @xmath572 , so by contradiction we can conclude that @xmath572 is equal to @xmath27 . [ thm : neweigen ] @xmath572 is equal to @xmath27 . in particular , @xmath572 is equidimensional of dimension @xmath135 . let @xmath23 be a point of @xmath573 with weight @xmath145 . the proof will proceed by showing that @xmath23 is also @xmath3-old , then raising the level at @xmath23 , as in the previous theorem , to show that it lies in a family of @xmath3-new points . we pick @xmath234 , @xmath143 and @xmath28 such that @xmath23 comes from a @xmath143-overconvergent automorphic form of slope @xmath234 and level @xmath574 . now fix a closed ball in @xmath119 , containing the weight of @xmath23 , which is small enough ( note that ` small enough ' depends on @xmath234 , @xmath143 and @xmath28 ) to apply the local eigenvariety construction described in section 6.2 of @xcite . denote this ball by @xmath127 . the point @xmath145 in @xmath127 corresponds to a maximal ideal @xmath166 of @xmath129 . if we set @xmath575 as before , then @xmath23 lies in a family corresponding to a hecke eigenvector in @xmath45 . as in the previous subsection , we have a closed embedding @xmath576 , where @xmath577 is a two - covering of the tame level @xmath78 reduced eigencurve for @xmath1 , and the image of this embedding is the zariski closure of the classical @xmath3-old points in @xmath539 , which also equals the space of @xmath3-old points in @xmath539 ( as is clear from applying the proof of lemma [ old ] to modules of overconvergent automorphic forms for @xmath1 ) . since the space @xmath539 is equidimensional of dimension @xmath135 , but @xmath23 does not lie in a family of points in @xmath572 , @xmath23 must lie in a family of points in @xmath577 . so @xmath23 is @xmath3-old and @xmath3-new . we now ` raise the level ' at @xmath23 . as in the proof of proposition [ preihara ] we specialise the @xmath129 modules @xmath578 and @xmath45 at the maximal ideal @xmath166 to give vector spaces @xmath579 and @xmath580 . since @xmath23 is @xmath3-new and @xmath3-old , it arises from an eigenform @xmath7 in @xmath581 . now a calculation using the explicit matrix for the map @xmath582 shows that @xmath7 is of the form @xmath583 , where @xmath5 is an eigenform in @xmath579 with @xmath584 , and the roots of the @xmath3th hecke polymonial for @xmath5 are @xmath28 and @xmath585 ( note that with our normalisations @xmath7 is the @xmath3-stabilisation of @xmath5 corresponding to the root @xmath28 ) . now applying the proof of theorem [ raise ] we see that @xmath23 lies in @xmath27 , so we have a contradiction and therefore must have @xmath586 . define the banach @xmath129-module @xmath587 for varying affinoid subdomains @xmath131 ( with corresponding character @xmath147 ) to be the kernel of the map @xmath588 one might wish to construct the rigid analytic space @xmath572 from these modules using buzzard s eigenvariety machine @xcite . the key issue is to show that the modules @xmath587 behave well under base change between affinoid subdomains . the author is not sure whethere this should be true or not - the results in this paper can be viewed as showing that these modules behave well under base change from a sufficiently small affinoid to a point and it is not clear that one can conclude something about base change between open affinoids from this . this paper is the result of research carried out whilst studying for a phd under the supervision of kevin buzzard , to whom i am grateful for suggesting the problem and providing such excellent guidance . i would also like to thank owen jones , david loeffler and kevin mcgerty for their helpful comments , and the referee for several useful remarks and corrections . r. coleman and b. mazur , _ the eigencurve _ , galois representations in arithmetic algebraic geometry ( durham , 1996 ) , london math . lecture note ser . 254 , cambridge univ . press , 1998 , pp . 1113 .

given a poset , the set , , of all scott closed sets ordered by inclusion forms a complete lattice . a subcategory of ( the category of posets and scott - continuous maps ) is said to be-faithful if for any posets and in , implies . it is known that the category of all continuous dcpos and the category of bounded complete dcpos are-faithful , while is not . ho & zhao ( 2009 ) asked whether the category of dcpos is-faithful . in this paper , we answer this question in the negative by exhibiting a counterexample . to achieve this , we introduce a new subcategory of dcpos which is-faithful . this subcategory subsumes all currently known-faithful subcategories . with this new concept in mind , we construct the desired counterexample which relies heavily on johnstone s famous dcpo which is not sober in its scott topology .

1607.03286

the collection , @xmath9 , of closed subsets of a topological space @xmath10 , ordered by inclusion , forms a distributive complete lattice often referred to as the _ closed set lattice _ of @xmath10 . if a lattice @xmath11 is isomorphic to the closed set lattice of some topological space @xmath10 , we say that @xmath10 is a _ topological representation _ of @xmath11 . it is natural to ask : [ quest : first question ] which lattices have topological representations ? seymour papert was the first to characterize such lattices as those which are complete , distributive , and have a base consisting of irreducible elements , @xcite . we can also ask how much of the topological structure of a space is encoded in its closed set lattice . following wolfgang thron , @xcite , we say that two topological spaces @xmath10 and @xmath12 are _ lattice - equivalent _ if their closed - set lattices are order - isomorphic . clearly , homeomorphic spaces are lattice - equivalent , but the converse fails . this then leads us to : [ quest : second question ] which classes @xmath2 of topological spaces are such that any two lattice - equivalent spaces @xmath10 and @xmath13 are homeomorphic , i.e. : @xmath14 sober topological spaces are exactly those that can be fully reconstructed from their closed set lattices , @xcite ; therefore the class @xmath15 of sober spaces is a natural choice in answer to question [ quest : second question ] . furthermore , any topological space @xmath10 is lattice - equivalent to its sobrification @xmath16 and so it follows that @xmath2 can not contain a non - sober space @xmath10 and its sobrification @xmath16 at the same time , in other words , @xmath15 is a _ maximal _ choice for @xmath2 . the two questions above can also be asked in the context of a particular class of topological spaces . the ones we have in mind were introduced by dana scott , @xcite , and are known collectively as _ domains_. the characteristic feature of domains is that they carry a partial order and that their topology is completely determined by the order . more precisely , let @xmath0 be a poset and @xmath17 a subset of @xmath0 . one says that @xmath17 _ scott open _ , if ( i ) @xmath17 is an upper set , and ( ii ) @xmath17 is inaccessible by directed joins . the set @xmath18 of all scott opens of @xmath0 forms the _ scott topology _ on @xmath0 , and @xmath19 is called the _ scott space _ of @xmath0 . in what follows , for a poset @xmath0 , we write @xmath1 to always mean the lattice of scott - closed subsets of @xmath0 . we may now relativize our definitions to the context of scott spaces . we say that a lattice @xmath11 has a _ scott - topological representation _ if @xmath11 is isomorphic to @xmath1 for some poset @xmath0 , and ask : [ prob : first problem ] which lattices have scott - topological representations ? although some work has been done on this problem , @xcite , as of now it remains open . in the special case of _ continuous domains _ a very pleasing answer was given independently by jimmie lawson , @xcite , and rudolf - eberhardt hoffmann , @xcite . they showed that a lattice @xmath11 has a scott - topological representation @xmath1 for some continuous domain @xmath0 if and only if @xmath11 is completely distributive . in the order - theoretic context the second question reads as follows : [ prob : second problem ] which classes of posets @xmath2 satisfy the condition @xmath20 a class @xmath2 of posets is said to be _ @xmath4-faithful _ if the above condition holds ( @xcite[p . 2170 , remark 2 ] ) . the following classes of posets are known to be @xmath4-faithful : 1 . @xmath21 of complete semilattices , @xcite , containing in particular all complete lattices ; 2 . @xmath22 of dcpos whose scott topologies are sober , containing in particular @xmath23 , the class of all continuous dcpos . a _ dcpo - completion _ of a poset @xmath0 is a dcpo @xmath24 together with a scott - continuous mapping @xmath25 such that for any scott - continuous mapping @xmath26 into a dcpo @xmath27 there exists a unique scott - continuous mapping @xmath28 satisfying @xmath29 . it was shown in @xcite that the dcpo - completion , @xmath30 , of a poset @xmath0 always exists ; furthermore @xmath31 . hence the class , @xmath32 , of all posets is _ not _ @xmath4-faithful . indeed any class of posets that is strictly larger than @xmath8 , the class of dcpos , is not @xmath4-faithful . this means that we can restrict attention entirely to dcpos , in particular , it leads one to ask : ( @xcite , ( * ? ? ? * remark 2 ) ) [ prob : ho - zhao problem ] is @xmath8 @xmath4-faithful ? this question was dubbed the _ ho - zhao problem _ in @xcite . the authors of this paper claimed that the two dcpos @xmath33,\leq ) \text { and } \psi = ( \{[0,a ] \mid 0 < a \leq 1\},\subseteq)\ ] ] show that @xmath8 is not @xmath4-faithful . however , it is easy to see that @xmath34,\leq)$ ] so that @xmath35,\leq)$ ] . on the other hand , @xmath36,\leq)_\bot$ ] which is not isomorphic to @xmath37 . this failure is only to be expected as both @xmath38 and @xmath39 are continuous dcpos and we already noted that @xmath23 is @xmath4-faithful . this paper comprises two parts . in section [ sec : positive ] we present a positive result by introducing the class @xmath40 of _ dominated dcpos _ and showing it to be @xmath4-faithful . importantly , @xmath40 subsumes all currently known @xmath4-faithful classes listed above . in the second part ( sections [ sec : johnstones - countereg][sec : hoxi - structure ] ) we show that the answer to the ho - zhao problem is negative . we construct a dcpo @xmath41 which is not dominated , and derive from it a dominated dcpo @xmath42 so that @xmath43 but @xmath44 . the construction makes use of johnstone s famous example of a dcpo @xmath45 whose scott topology is not sober ( @xcite ) . to familiarize the reader with johnstone s dcpo @xmath45 , we revisit it in section [ sec : johnstones - countereg ] , highlighting its peculiarities . this prepares us for the counterexample @xmath41 , presented in section [ sec : hoxi - informal ] . for quick accessibility we keep the description of @xmath41 in section [ sec : hoxi - informal ] informal . intrepid readers who are keen to pursue the detailed construction of @xmath41 and how it works in answering the ho - zhao problem may then continue their odyssey into section [ sec : hoxi - structure ] . for notions from domain theory we refer the reader to @xcite . a nonempty subset @xmath24 of a topological space @xmath46 is called _ irreducible _ if whenever @xmath47 for closed sets @xmath27 and @xmath48 , @xmath49 or @xmath50 follows . we say that @xmath24 is _ closed irreducible _ if it is closed and irreducible . the set of all closed irreducible subsets of @xmath10 is denoted by @xmath51 . two facts about irreducible sets will be important below : * if @xmath24 is irreducible then so is its topological closure . * the direct image of an irreducible set under a continuous function is again irreducible . in this paper we are exclusively interested in dcpos and the scott topology on them . when we say `` ( closed ) irreducible '' in this context we always mean ( closed ) irreducible with respect to the scott topology . it is a fact that every directed set in a dcpo is irreducible in this sense , while a set that is directed and closed is of course a principal ideal , that is , of the form @xmath52 . peter johnstone discovered in 1981 , @xcite , that a dcpo may have irreducible sets which are not directed , and closed irreducible sets which are not principal ideals ; we will discuss his famous example in section [ sec : johnstones - countereg ] and someone not familiar with it may want to have a look at it before reading on . given that we may view irreducible sets as a generalisation of directed sets , the following definition suggests itself : a dcpo @xmath53 is called _ strongly complete _ if every irreducible subset of @xmath54 has a supremum . in this case we also say that @xmath54 is an _ scpo_. a subset of an scpo is called _ strongly closed _ if it closed under the formation of suprema of irreducible subsets . note that despite this terminology , strongly complete partial orders are a long way from being complete in the sense of lattice theory . the basis of questions [ quest : first question ] and [ quest : second question ] in the introduction is the well - known ( dual ) adjunction between topological spaces and frames , which for our purposes is expressed more appropriately as a ( dual ) adjunction between topological spaces and coframes : @xmath55[r]{$\gamma:\mathbf{top}$}\ar@/^/[r]_\bot & \makebox[2em][l]{$\mathbf{cofrm}^{\mathrm{op}}:\mathrm{spec}$}\ar@/^/[l ] } \ ] ] here @xmath56 is the functor that assigns to a coframe its set of _ irreducible _ elements , that is , those elements @xmath57 for which @xmath58 implies @xmath59 or @xmath60 , topologized by closed sets @xmath61 . starting with a topological space @xmath46 we obtain the _ sobrification _ @xmath16 of @xmath10 by composing @xmath56 with @xmath4 . concretely , the points of @xmath16 are given by the closed irreducible subsets of @xmath10 and the topology is given by closed sets @xmath62 where @xmath63 . in order to present problems [ prob : first problem ] and [ prob : second problem ] in a similar fashion , it seems natural to replace @xmath64 with the category of _ dcpo spaces _ , i.e. , directed - complete partially ordered sets equipped with the scott topology . however , while we know that @xmath56 yields topological spaces which are dcpos in their specialisation order , it is _ not the case _ that the topology on @xmath65 equals the scott topology with respect to that order ; all we know is that every @xmath66 is scott - closed . rather than follow a topological route , therefore , we reduce the picture entirely to one concerning _ ordered sets_. to this end we restrict the adjunction above to the category @xmath67 of _ monotone convergence spaces _ ( * definition ii-3.12 ) ) on the topological side and compose it with the adjunction between monotone convergence spaces and dcpos @xmath68_\bot & \mathbf{mcs}:\mathrm{so}\ar@/^/[l ] } \ ] ] which assigns to a dcpo its scott space and to a space its set of points with the _ specialisation order_. we obtain a functor from @xmath8 to @xmath69 which assigns to a dcpo its coframe of scott - closed subsets and we re - use the symbol @xmath4 for it rather than writing @xmath70 . in the other direction , we assign to a coframe the _ ordered set _ of irreducible elements , where the order is inherited from the coframe . to emphasize the shift in perspective , we denote it with @xmath71 rather than @xmath56 or @xmath72 . it will also prove worthwhile to recall the action of @xmath71 on morphisms : if @xmath73 is a coframe homomorphism , then @xmath74 maps an irreducible element @xmath57 of @xmath75 to @xmath76 . altogether we obtain the following ( dual ) adjunction : @xmath55[r]{$\gamma:\mathbf{dcpo}$}\ar@/^/[r]_\bot & \makebox[2em][l]{$\mathbf{cofrm}^{\mathrm{op}}:{\mathrm{irr}}$}\ar@/^/[l ] } \ ] ] its unit @xmath77 maps an element @xmath78 of a dcpo @xmath54 to the principal downset @xmath52 , which is always closed and irreducible and hence an element of @xmath79 . the counit @xmath80 ( as a concrete map between coframes ) sends an element @xmath81 of a coframe @xmath11 to the set @xmath82 which is clearly closed under directed suprema , hence an element of @xmath83 . combining @xmath71 with @xmath4 yields a monad on @xmath8 which we denote with @xmath84 . concretely , it assigns to a dcpo @xmath54 the set @xmath85 of closed irreducible subsets ordered by inclusion . we call this structure the _ order sobrification of @xmath54_. if @xmath86 is a scott - continuous function between dcpos , and @xmath87 is a closed irreducible set , then @xmath88 is the scott closure of the direct image @xmath89 ( which is again irreducible as we noted at the beginning of this section ) . the monad unit is given by the unit of the adjunction and the multiplication @xmath90 maps a closed irreducible collection . ] of closed irreducible subsets to their union . we will see in section [ sec : hoxi - informal ] that this order - theoretic monad is _ not _ idempotent ; this is an important difference to its topological counterpart , the sobrification monad . the concrete description of @xmath91 as union allows us to conclude immediately : [ prop : scpo ] for any dcpo @xmath54 , @xmath85 is strongly complete . there is more to discover about our adjunction and the union operation : [ prop : union ] let @xmath54 be a dcpo . 1 . if @xmath27 a closed set of @xmath54 then @xmath92 . 2 . if @xmath93 is irreducible as a subset of @xmath85 then @xmath94 is irreducible as a subset of @xmath54 . if @xmath95 is scott - closed as a subset of @xmath85 then @xmath96 is scott - closed as a subset of @xmath54 . if @xmath27 is a closed set of @xmath54 then @xmath97 is strongly closed . \(1 ) this follows from the general description of the counit . \(2 ) assume we have @xmath98 with @xmath99 scott - closed subsets of @xmath54 . then @xmath100 by ( 1 ) and because every member of @xmath93 is irreducible . furthermore , @xmath97 and @xmath101 are scott - closed subsets of @xmath85 , so by assumption , one of them must cover the irreducible collection @xmath93 , say @xmath97 . it then follows that @xmath94 is contained in @xmath27 . \(3 ) this follows for categorical reasons because the union operation is the same as @xmath102 , but a direct proof is instructive for what follows : if @xmath103 is a directed subset of @xmath94 then each @xmath104 must belong to some @xmath105 . because each @xmath106 is a lower set of @xmath54 we have @xmath107 and because @xmath93 is a lower set of @xmath85 it follows that @xmath108 . the collection @xmath109 is directed and by assumption its supremum @xmath110{$\scriptstyle\uparrow$}}}}s$ ] belongs to @xmath93 as @xmath93 is scott - closed . if follows that @xmath111{$\scriptstyle\uparrow$}}}}s$ ] belongs to @xmath94 . \(4 ) let @xmath93 be an irreducible subset of @xmath97 . then by ( 2 ) we have that @xmath94 is irreducible . since @xmath94 is contained in the closed set @xmath27 , so is its closure @xmath24 . we know that @xmath24 is again irreducible and therefore this is the supremum of @xmath93 and it belongs to @xmath97 . strong closedness is crucial for the following result : [ lem : strong ] let @xmath54 be a dcpo and @xmath112 . 1 . @xmath113 . 2 . if @xmath95 is strongly closed then equality holds . the first statement is trivial by the definition of the counit @xmath114 . for the second , let @xmath24 be a closed irreducible subset of @xmath96 . we need to show that @xmath24 is an element of @xmath95 . every element @xmath78 of @xmath24 belongs to some @xmath115 . as we argued in part ( 3 ) of the preceding proof , it follows that for every @xmath116 , @xmath117 . we claim that the collection @xmath118 is irreducible as a subset of @xmath85 . this will finish our proof as we clearly have that @xmath24 is the supremum of @xmath93 and by assumption , @xmath95 is closed under forming suprema of irreducible subsets . so let @xmath93 be covered by two closed collections @xmath119 , in other words , every @xmath52 belongs to either @xmath120 or @xmath121 . it follows that each @xmath116 belongs to either @xmath122 or @xmath123 and these two sets are scott - closed by part ( 3 ) of the preceding proposition . because @xmath24 is irreducible , it is already covered by one of the two , and this implies that @xmath93 is covered by either @xmath120 or @xmath121 . we approach question [ prob : second problem ] via the monad @xmath84 . starting from the assumption @xmath124 we immediately infer @xmath125 and the question then becomes whether this isomorphism implies @xmath126 . our counterexample will demonstrate that in general the answer is `` no '' but in this section we will exhibit a new class @xmath40 of _ dominated dcpos _ for which the answer is positive , that is , we will show : @xmath127 before we do so , let us check that invoking the monad does not change the original question , in other words , the assumption @xmath128 is neither stronger nor weaker than @xmath124 : for arbitrary dcpos @xmath54 and @xmath129 , @xmath130 . the implication from right to left is trivial , so assume we are given an order isomorphism @xmath131 . the idea for an isomorphism @xmath132 from @xmath133 to @xmath134 is very simple : given @xmath135 we compute @xmath97 , the collection of all closed irreducible sets contained in @xmath27 . each of these can be replaced with its counterpart in @xmath129 via the given isomorphism @xmath136 . in @xmath129 , then , we simply take the union of the collection @xmath137 . using the maps that are provided to us by the adjunction , we can express @xmath132 as follows : @xmath138^{\varepsilon_{\gamma(d ) } } & \gamma({\widehat{d}})\ar[r]^{\gamma(i^{-1 } ) } & \gamma({\widehat{e}})\ar[r]^{\bigcup } & \gamma(e ) } \ ] ] for an inverse , we follow the same steps , starting at @xmath134 : @xmath139^{\varepsilon_{\gamma(e ) } } & \gamma({\widehat{e}})\ar[r]^{\gamma(i ) } & \gamma({\widehat{d}})\ar[r]^{\bigcup } & \gamma(d ) } \ ] ] in order to show that these are inverses of each other we use the fact that @xmath97 is _ strongly _ closed which we established in proposition [ prop : union](4 ) . since the concept of strong closure is purely order - theoretic we get that the direct image under @xmath136 is again strongly closed . this is crucial as it allows us to invoke lemma [ lem : strong](2 ) . the computation thus reads : @xmath140 the other composition , @xmath141 , simplifies in exactly the same way to the identity on @xmath134 , and since all maps involved are order - preserving , we have shown that the pair @xmath142 constitutes an order isomorphism between @xmath133 and @xmath134 . our new version of question [ prob : second problem ] requires us to infer @xmath126 from @xmath128 , and the most direct approach is to find a way to recognize _ purely order - theoretically _ inside @xmath85 those elements which correspond closed irreducible subsets of the form @xmath52 with @xmath143 . as our counterexample @xmath41 to the ho - zhao problem will show , this is not possible for general dcpos . the purpose of the present section is to exhibit a class of dcpos for which the direct approach works . given @xmath144 we write @xmath145 if there is @xmath116 such that @xmath146 . we write @xmath147 for the set @xmath148 . clearly , an element @xmath24 of @xmath85 is of the form @xmath52 if and only if @xmath149 holds , but this is not yet useful since the definition of @xmath150 makes explicit reference to the underlying dcpo @xmath54 . we can , however , record the following useful facts : let @xmath54 be a dcpo and @xmath87 closed and irreducible . 1 . @xmath151 2 . @xmath152 is irreducible as a subset of @xmath85 . the first statement is trivial because @xmath152 contains all principal ideals @xmath52 , @xmath116 . the proof of the second statement is essentially the same as that of lemma [ lem : strong ] . let @xmath54 be a strongly complete partial order and @xmath153 . we write @xmath154 if for all closed irreducible subsets @xmath24 of @xmath54 , @xmath155 implies @xmath156 . we say that @xmath143 is _ @xmath157-compact _ if @xmath158 , and denote the set of @xmath157-compact elements by @xmath159 . note that this is an _ intrinsic _ definition of a relation on @xmath54 without reference to any other structure . it is reminiscent of the way - below relation of domain theory but note that it is defined via _ closed _ irreducible sets . this choice has the following consequence , which is definitely not true for way - below : [ prop : prec - downset ] let @xmath54 be an scpo and @xmath143 . the set @xmath160 is scott - closed . let @xmath161 be a directed set of elements , each of which is @xmath157-below @xmath78 . we need to show that @xmath111{$\scriptstyle\uparrow$}}}}_{i\in i}a_i$ ] is also @xmath157-below @xmath78 . to this end let @xmath24 be a closed irreducible subset of @xmath54 with @xmath162 . since @xmath163 for all @xmath164 we have that every @xmath165 belongs to @xmath24 and because @xmath24 is closed , @xmath111{$\scriptstyle\uparrow$}}}}_{i\in i}a_i\in a$ ] follows . recall from proposition [ prop : scpo ] that @xmath85 is strongly complete for any dcpo @xmath54 , so on @xmath85 we can consider both @xmath150 and @xmath157 . we observe : @xmath145 implies @xmath166 for all @xmath167 . this holds because the supremum of a closed directed collection of closed irreducible subsets is given by union . our aim now is to give a condition for dcpos @xmath54 which guarantees the reverse implication . a dcpo @xmath54 is called _ dominated _ if for every closed irreducible subset @xmath24 of @xmath54 , the collection @xmath152 is scott - closed in @xmath85 . we are ready for the final technical step in our argument : a dcpo @xmath54 is dominated , if and only if @xmath166 implies @xmath145 for all @xmath167 . if : let @xmath168 . by assumption we have @xmath169 and in proposition [ prop : prec - downset ] we showed that the latter is always scott - closed . only if : let @xmath166 . we know that @xmath152 is irreducible and by assumption closed . we also know that @xmath151 so it must be the case that @xmath170 . we are ready to reap the benefits of our hard work . all of the following are now easy corollaries : for a dominated dcpo @xmath54 , the only @xmath157-compact elements of @xmath85 are the principal ideals @xmath52 , @xmath143 . also , the unit @xmath77 is an order isomorphism from @xmath54 to @xmath171 . let @xmath54 and @xmath129 be dominated dcpos . the following are equivalent : 1 . 2 . @xmath173 . 3 . @xmath174 . the class @xmath40 of dominated dcpos is @xmath4-faithful . let us now explore the reach of our result and exhibit some better known classes of dcpos which are subsumed by @xmath40 . [ prop : complete semilattice is dominated ] every scpo is dominated . if @xmath145 for closed irreducible subsets of an scpo , then by definition @xmath146 for some @xmath116 and hence @xmath175 . if @xmath176 is a directed family in @xmath152 , then @xmath111{$\scriptstyle\uparrow$}}}}_{i\in i}a_i\subseteq{\mathord{\downarrow}}{\mathop{\bigvee{}^{\,\makebox[0pt]{$\scriptstyle\uparrow$}}}}_{i\in i}(\bigvee a_i)$ ] , and the element @xmath111{$\scriptstyle\uparrow$}}}}_{i\in i}(\bigvee a_i)$ ] belongs to @xmath24 because @xmath24 is a scott - closed set . [ cor : all - dominated ] the following are all dominated : 1 . complete semilattices and complete lattices ; 2 . dcpos which are sober in their scott topologies ; 3 . @xmath85 for any dcpo @xmath54 . \(1 ) holds because complete semilattices have suprema for all bounded sets , which is all that is needed in the proof of the preceding proposition . \(2 ) holds because the only closed irreducible subsets of sober dcpos are principal ideals . \(3 ) holds because we showed in proposition [ prop : scpo ] that @xmath85 is always strongly complete . recall that a topological space is called _ coherent _ if the intersection of two compact saturated sets is again compact . it is called _ well - filtered _ if whenever @xmath177 is a filtered collection of compact saturated sets contained in an open @xmath178 , then some @xmath179 is contained in @xmath178 already . [ lem : ub(a ) is compact saturated ] let @xmath54 be a dcpo which is well - filtered and coherent in its scott topology . then for any @xmath180 , the set @xmath181 of upper bounds of @xmath24 is compact saturated . since @xmath182 , it is saturated . for any finite nonempty @xmath183 , @xmath184 is compact since @xmath54 is coherent . thus @xmath185 forms a filtered family of compact saturated sets whose intersection equals @xmath181 . if it is covered by a collection of open sets then by well - filteredness of @xmath54 some @xmath184 is covered already , but the latter is compact by coherence so a finite subcollection suffices to cover it . [ prop : filtered and coherent implies dominated ] every dcpo @xmath54 which is well - filtered and coherent in its scott topology is dominated . let @xmath176 be a directed family in @xmath152 . by the preceding proposition , the sets @xmath186 are compact saturated and form a filtered collection . suppose for the sake of contradiction that @xmath187 . since @xmath188 is scott open , by the well - filteredness of @xmath54 it follows that there exists @xmath189 such that @xmath190 . this contradicts the fact that every @xmath191 is bounded by an element of @xmath24 . at the juncture , the curious reader must be wondering why we have not yet given an example of a dcpo which is not dominated . now , as we showed , any dcpo whose scott topology is sober is dominated and so a non - dominated dcpo , if it exists , must be non - sober . peter johnstone , @xcite was the first to give an example of such a dcpo and it is thus natural to wonder whether it is dominated or not . in order to answer this question we explore his example @xmath45 in some detail in the next section . this will also help us to construct and study our counterexample to the ho - zhao conjecture . we denote with @xmath192 the _ ordered set _ of natural numbers augmented with a largest element @xmath193 . when we write @xmath194 we mean the natural numbers as a set , that is , _ discretely ordered_. johnstone s counterexample @xmath45 , depicted in figure [ fig : johnstones - non - sober - dcpo ] , is based on the ordered set @xmath195 , that is , a countable collection of infinite chains . we call the chain @xmath196 the _ @xmath197-th component _ of @xmath45 . similarly , for a fixed element @xmath198 we call the set @xmath199 the _ @xmath200-th level_. so @xmath201 is the unique element in the @xmath197-th component on level @xmath200 . we call the elements on the @xmath193-th level _ limit points _ , and refer to all the others as _ finite elements _ ( although they are not `` finite '' in the sense of domain theory ) . the order on @xmath45 is given by the product order plus the stipulation that the limit point @xmath202 of the @xmath197-th component be above every element of levels @xmath203 to @xmath197 . it will be convenient later to write this formally as follows : * @xmath204 if @xmath205 ( @xmath206 , @xmath207 ) * @xmath208 if @xmath209 ( @xmath210 ) johnstone s non - sober dcpo @xmath45 . ] as the hasse diagram makes clear , the only non - trivial directed sets of @xmath45 are the chains contained in a component , with supremum the component s limit point . since every finite element @xmath201 is also below the limit points @xmath211 , where @xmath212 , but not below any other finite element outside its own component , we see that none of them is _ compact _ in the sense of domain theory . it follows that @xmath45 is highly non - algebraic , and indeed it could nt be algebraic as algebraic dcpos are always sober spaces in their scott topology . let us take a closer look at the scott topology on @xmath45 . the two defining conditions of scott - closed sets manifest themselves in the following properties : * closed sets are lower sets : if the set contains a limit point @xmath202 , then it must contain the component @xmath213 and all levels @xmath214 . * closed sets are closed under the formation of limits : if the set contains infinitely many elements of any one component then it must contain the limit point of that component . taken together , we obtain the following principle ( which plays a crucial role in our construction as well ) : * if a scott - closed subset of @xmath45 contains infinitely many limit points , then it equals @xmath45 . this is because such a set contains infinitely many levels by the first property of scott - closed sets , which means that it contains infinitely many elements of every component , and therefore contains the limit points of all components by the second property . applying the first property again we see that it contains everything . it is now easy to see that @xmath45 itself is an irreducible closed set : if we cover @xmath45 with two closed subsets then at least one of them must contain infinitely many limit points . by @xmath215 that set then is already all of @xmath45 . of course , @xmath45 is not the closure of a singleton since there is no largest element so we may conclude , as peter johnstone did in @xcite , that @xmath45 is not sober in its scott topology . in the terminology of the last section , it also follows that @xmath45 is not strongly complete . the order sobrification @xmath216 of @xmath45 is depicted in figure [ fig : sobrification - of - johnstone ] . we will prove in subsection [ subsec : irred in johnstone dcpo ] that the picture is correct , that is , @xmath45 itself is indeed the only closed irreducible subset of @xmath45 which is not a principal downset . the order sobrification @xmath216 of johnstone s non - sober dcpo . ] for the moment , however , let us check whether @xmath45 is already a counterexample to the ho - zhao conjecture . unfortunately , this is not the case , the scott topologies of @xmath45 and @xmath216 are not isomorphic . we see this as follows : the extra point at the top of @xmath217 can not be reached by a directed set ; it is _ compact _ in the sense of domain theory . also , the image of @xmath45 under @xmath218 forms a scott - closed subset @xmath27 of @xmath216 . together this means that the largest element of @xmath219 is compact , while the largest element of @xmath220 is not : it is the directed limit of the closed sets @xmath221 , @xmath222 . as an aside , we may also observe that the order sobrification of @xmath216 would add yet another point , placed between the elements of @xmath27 and the top element @xmath216 which shows that order sobrification is not an idempotent process . in fact , our construction of @xmath41 addresses exactly this point , by making sure that the new elements that appear in the order sobrification are not compact and do not lead to new scott - closed subsets . the construction of @xmath41 may be viewed as an infinite process . we begin with @xmath45 and for every finite level @xmath223 of @xmath45 , we add another copy @xmath224 , whose infinite elements are identified with the elements of @xmath223 . no order relation between the _ finite _ elements of two different @xmath224 , @xmath225 is introduced . now the process is repeated with _ each _ finite level of _ each _ @xmath224 , adding a further @xmath226 many copies of @xmath45 . we keep going like this _ ad infinitum _ and in this way ensure all elements are limit elements . to make this a bit more precise , let @xmath227 be the set of lists of natural numbers . ( instead of `` list '' we also use the word `` string '' . ) we denote lists by @xmath228 and the empty list by @xmath229 . the result of adding the element @xmath200 to the front of list @xmath230 we write as @xmath231 , and for decomposing a list @xmath230 into its first element and the rest we use the notation @xmath232 and @xmath233 ( `` head '' and `` tail '' ) . note that @xmath234 and @xmath235 . the top of one component of @xmath41 . ] we use @xmath227 to index the many copies of johnstone s example that make up our dcpo @xmath41 . in a first step , we let @xmath236 be the disjoint union of all @xmath237 , @xmath238 . we label individual elements of @xmath236 with triples @xmath239 in the obvious way . on @xmath236 we consider the equivalence relation @xmath240 which identifies the finite element @xmath241 of @xmath237 with the infinite element @xmath242 of @xmath243 , for all @xmath244 and @xmath238 . we may now define @xmath41 as the quotient of @xmath236 by @xmath240 . we let the order on @xmath41 be the quotient order , that is , the smallest preorder such that the quotient map from @xmath236 to @xmath41 is monotone . we will show in the next section that this is in fact an order and that @xmath41 is a dcpo . figure [ fig : tree ] gives an impression of the resulting ordered set by showing the top of one component of @xmath41 . the constituent copies of @xmath45 are still very much present in @xmath41 : for a fixed @xmath238 we may collect all elements of the form @xmath241 , where @xmath206 , @xmath198 , into a substructure of @xmath41 that is order - isomorphic to @xmath45 . we call it the _ sheet _ @xmath237 . on @xmath245 we have again the _ finite levels _ @xmath246 consisting of the points @xmath247 for each @xmath222 . glueing together the copies of @xmath45 in this way does not create any new directed sets other than those ( essentially ) contained in a single sheet @xmath237 . from this it follows that the characterisation of scott - closed subsets given in section [ sec : johnstones - countereg ] above is still valid . the principle @xmath215 that we derived from this , however , now has greater reach : since every finite level in any sheet is simultaneously the set of limit points of a subsequent sheet , it holds that a scott - closed subset of @xmath41 can not contain infinitely many elements of _ any _ level without containing all of them . a close analysis of the situation ( which we carry out in the next section ) then shows that @xmath41 has one closed irreducible subset @xmath248 , which is not a point closure , for _ every _ @xmath238 , to wit , the sheet @xmath237 and everything below it . the top sheet of the order sobrification of @xmath41 . ] the ( inclusion ) order among the @xmath248 is exactly as indicated in figure [ fig : tree ] which implies in particular that none of them is compact in the sense of domain theory when viewed as elements of the order sobrification @xmath42 . more precisely , we have that @xmath248 is the limit ( i.e. , the closure of the union ) of the chain @xmath249 . if we look just at the top sheet @xmath250 in the order sobrification then we obtain the structure displayed in figure [ fig : sober - hoxi ] . the set @xmath27 of point closures ( indicated as open circles in figure [ fig : sober - hoxi ] ) no longer forms a scott - closed subset of @xmath42 . because a closed set must be downward closed , @xmath251 contains all @xmath252 , @xmath222 , and because it must be closed under taking limits , it contains @xmath253 as well , which means it equals all of @xmath42 . the same considerations hold on any sheet @xmath254 , @xmath238 , and we see that the scott topology of @xmath42 is the same as the sobrification topology , which in turn is always isomorphic to the topology of the original space , in our case the scott topology of @xmath41 . thus we have shown : [ thm : h = sobh ] the scott topologies of @xmath41 and @xmath42 are isomorphic . on the other hand , @xmath41 and @xmath42 are clearly not isomorphic as ordered sets : the latter has a largest element whereas the former does not . thus we have : [ cor : dcpo - not - faithful ] the category @xmath8 is not @xmath4-faithful . the structure of @xmath41 is sufficiently intricate to warrant a more formal examination of the claims made in the previous section . as explained informally above , we begin with the structure @xmath236 , the disjoint union of infinitely many copies @xmath237 of johnstone s dcpo , indexed by the set @xmath227 of strings of natural numbers . on @xmath236 we define the equivalence relation @xmath240 generated by @xmath255 for all @xmath206 , @xmath256 . explicit , then we can write the definition also as @xmath257 . ] we note that each equivalence class with respect to @xmath240 contains precisely the two elements @xmath258 and @xmath259 which appear in the definition , except for the elements of the form @xmath260 ( the maximal elements on sheet @xmath250 ) which are each in a class by themselves . for the arguments that follow below it is important to remember that @xmath240 connects elements from _ different _ sheets , to wit , @xmath237 and @xmath261 . the quotient preorder @xmath262 is defined as the transitive closure of @xmath263 . any preorder gives rise to an equivalence relation but in our case this is just @xmath240 again : the equivalence relation @xmath264 derived from the quotient preorder is equal to @xmath240 . by definition , two elements @xmath241 and @xmath265 are related by @xmath262 if and only if there is a sequence of elements @xmath266 , @xmath267 such that @xmath268 . we consider the first two steps of such a sequence ; since an element of the form @xmath269 is maximal in the sheet @xmath270 , the beginning of the sequence has to look as follows : @xmath271 with either @xmath272 and @xmath273 , or @xmath212 and @xmath274 . in the first case the sequence can continue with @xmath275 but then we are in sheet @xmath276 where the head element of the index @xmath277 is strictly larger than in @xmath231 . in the second case , we are already in a sheet where the index @xmath230 is strictly shorter than the starting index @xmath231 . this means that two elements @xmath241 and @xmath265 are related by @xmath264 only if they are related by @xmath240 already . we can now define @xmath41 as the quotient @xmath278 together with the quotient order @xmath279 . we denote the quotient map with @xmath280 . restricted to any sheet @xmath237 , the map @xmath280 is an order - isomorphism . this follows from the fact that the equivalence relation connects elements from different sheets only . this observation justifies speaking of the `` sheet @xmath237 '' irrespective of whether we are in @xmath41 or in @xmath236 . [ p : h - is - a - dcpo ] @xmath41 is a dcpo . elements of the form @xmath258 are maximal in the sheet @xmath237 , so in order to move from a finite element @xmath241 to a strictly larger one there is no other way than finding a larger element within @xmath237 , or to pass through an infinite element @xmath281 within @xmath237 and then to switch over to the element @xmath282 on sheet @xmath261 ( and to carry on from there ) . however , the strings in @xmath227 are of finite length , so the process of switching sheet can only be performed finitely often . it follows that any directed set of @xmath41 contains a cofinal subset entirely contained in a single sheet . because sheets are isomorphic to johnstone s dcpo @xmath45 by the preceding proposition , the supremum of the directed set can be found there as well . this proof also shows the following : the quotient map @xmath280 is scott - continuous . we are ready to show the main technical tool for establishing that @xmath41 is indeed a counterexample to the ho - zhao conjecture : the scott topology on @xmath41 equals the quotient topology derived from the scott topology on @xmath236 and the equivalence relation @xmath240 . since @xmath280 is scott - continuous , we have that @xmath283 is scott - open in @xmath236 for any scott - open set @xmath178 of @xmath41 . it follows that @xmath178 belongs to the quotient topology . conversely , if @xmath178 is a member of the quotient topology , then @xmath283 is scott - open in @xmath236 , which means that the set @xmath284 is scott - open for every @xmath238 . since we showed in the proof of proposition @xmath285 that directed sets of @xmath41 are essentially contained in a single sheet , it follows that @xmath178 is inaccessible by directed suprema ; in other words , scott - open since @xmath178 is upper . below we will have to make several ( and sometimes quite subtle ) constructions of open ( and closed ) sets on @xmath41 . by the proposition above we are allowed to do this by exhibiting a scott - open set @xmath286 for each sheet @xmath237 such that @xmath287 if and only if @xmath288 . the compatibility with @xmath240 together with the principle @xmath215 identified for johnstone s dcpo leads to the following principle for @xmath41 : * if a scott - closed subset of @xmath41 contains infinitely many elements from any level @xmath223 on any sheet @xmath237 , then it contains all of @xmath223 . this means that the closed ( resp . , open ) sets @xmath289 ( where @xmath290 is a closed set of @xmath41 ) are _ more specialized _ than just being scott - closed ( respectively , scott - open ) . closed sets of the form @xmath289 are called _ locally closed _ sets ( respectively , _ locally open _ ) . we will need to know exactly what such sets look like and the next proposition spells this out . the characterisation follows from the definition of scott - closedness , the order on @xmath45 , and @xmath291 . see figure [ fig : generic - closed ] for an illustration . the three parts of a generic locally closed set . ] [ p : closed ] a locally closed set @xmath292 is either all of @xmath45 or can be uniquely written as the union of the following three parts : @xmath293 \mbox{\emph{ii:}}&\bigcup_{1\leq p\leq h}l_p&\mbox{where } h\geq m_k&\mbox{(the lower rectangle)}\\[1ex ] \mbox{\emph{iii:}}&\bigcup_{j=1}^l{\mathord{\downarrow}}(n_j , q_j)&\mbox{where } n_1<n_2<\ldots < n_l\\ & & \mbox{and } n_j > h\mbox { for } j=1,\ldots , l&\mbox{(the finite columns ) } \end{array}\ ] ] if any of the numbers @xmath294 , @xmath295 and @xmath296 is equal to zero , then the corresponding part is empty , but because of the way the order is defined on @xmath45 , every component @xmath297 from part i forces the presence of a lower rectangle of at least height @xmath298 . the purpose of part ii is to account for the possibility that @xmath248 contains more complete rows than is required by the presence of limit points . we already know that @xmath45 is not sober in its scott topology since @xmath45 is closed and irreducible but not a point closure but we would like to establish that this is the only such subset . we do this in a fashion that prepares us for the more complicated case of @xmath41 . [ p : irrs ] the closed irreducible subsets of @xmath45 are @xmath103 itself and the closures of singleton sets . assume that @xmath24 is a closed proper subset of @xmath45 . we distinguish two cases : case 1 : @xmath24 contains no limit points . let @xmath299 be a maximal element of @xmath24 . we claim that @xmath300 is scott - closed . indeed , since @xmath24 contains no infinite chains at all we do nt need to worry about closure under limits . downward closure also holds because there is no order relationship between finite elements from different components of @xmath45 . we therefore have the decomposition @xmath301 which shows that @xmath24 can only be irreducible if @xmath302 and @xmath303 . case 2 : @xmath24 contains limit points . by the principle @xmath215 , there are only finitely many limit points in @xmath24 , so let @xmath202 be the one with the highest component number @xmath197 . for a decomposition we set @xmath304 and @xmath305 , and again we have @xmath306 . if we can show that @xmath27 is closed then the rest of the argument is as in the previous case . closure under directed sups is clear because the only non - trivial chains of @xmath45 are those infinite subsets that are contained in a single component , and we have left all components of @xmath24 unchanged except for @xmath213 but there we have removed a cofinal part . downward closure of @xmath27 follows because @xmath307 consists only of the limit points @xmath308 and none of these were members of @xmath24 to start with . [ prop : johnstone is dominated ] johnstone s dcpo @xmath45 is dominated . the only nontrivial directed sets in @xmath216 are the chains @xmath309 and their supremum is @xmath310 . if a closed irreducible subset @xmath24 of @xmath216 contains such a chain it must also contain its supremum , and because the supremum is a principal downset , it belongs to @xmath152 . by corollary [ cor : all - dominated ] , @xmath216 is dominated . from the preceding result , @xmath45 is dominated . hence @xmath311 ; otherwise , because both @xmath45 and @xmath216 are dominated , @xmath312 would imply @xmath313 , which we know is false . this argument then provides an alternate explanation as to why the pair of dcpos @xmath45 and @xmath216 is not a counterexample to the ho - zhao conjecture . let us now apply a similar technique in order to identify precisely all irreducible subsets of @xmath41 . our aim is to show the following : [ p : characterisation ] the closed irreducible subsets of @xmath41 are of the following two forms : 1 . downsets of individual elements of @xmath41 ; 2 . downsets of levels @xmath314 , where @xmath238 , @xmath222 , or @xmath315 . note that we could have expressed the second form as `` _ _ downsets of levels @xmath316 _ _ '' without the need to treat @xmath317 as a special case but for the arguments that follow we have found it advantageous to think of the level as consisting of finite elements . the proof of the proposition is in three parts . the sets mentioned in the previous proposition are closed and irreducible . sets of the first kind can also be seen as closures of singleton subsets and such sets are always irreducible . sets of the second kind are clearly irreducible by principle @xmath291 . the only concern is whether they are closed . for this we have to show that a set @xmath318 is locally closed on each sheet @xmath319 . we use the characterisation provided in proposition [ p : closed ] : if @xmath320 then @xmath321 consists of part ii only , namely , the levels @xmath322 . in case @xmath323 , @xmath324 is all of @xmath325 . if @xmath326 for some @xmath327 , then the limit points of @xmath328 are identified with the level @xmath329 of @xmath237 by the equivalence relation @xmath240 . therefore @xmath324 is either empty ( if @xmath330 ) or all of @xmath328 ( if @xmath331 ) . in either case it is locally closed . this argument repeats itself if @xmath332 is extended with further numbers . in all other cases ( i.e. , if @xmath230 is not a suffix of @xmath333 ) , @xmath324 is empty . [ p : clipping ] if a closed irreducible subset of @xmath41 contains a maximal element @xmath241 but not all of the level @xmath334 , then it equals the downset of @xmath241 . assume that @xmath335 contains @xmath241 as a maximal element but not every element on the corresponding level @xmath334 . we may assume that @xmath230 is of minimal length and that on @xmath237 , @xmath200 is maximal with these properties ( i.e. , if @xmath336 then the level @xmath337 contains no element of @xmath24 ) . as in the proof of proposition [ p : irrs ] , we need to distinguish two cases : case 1 : @xmath338 : first observe that since @xmath200 is maximal in @xmath248 , part i of @xmath248 is empty . we claim that there is a closed subset @xmath27 which does not contain @xmath241 such that @xmath339 . as in the proof of [ p : irrs ] we can then invoke irreducibility to conclude that @xmath340 and hence @xmath341 . we define @xmath27 via its locally closed sets @xmath342 , @xmath343 . here we go : if @xmath320 then we first observe that because @xmath241 is maximal in @xmath24 and @xmath338 , @xmath241 is one of the generators of part iii of @xmath248 . this means we can let @xmath344 and still have a subset that conforms with the structure exhibited in proposition [ p : closed ] . if @xmath326 for some @xmath327 , then the limit points of @xmath328 are identified with the level @xmath329 of @xmath237 by the equivalence relation . since we altered @xmath248 only on level n , we can restrict attention to the case where @xmath345 . the element @xmath241 appears in @xmath346 as @xmath347 and since we removed the former from @xmath248 we must remove the latter from @xmath348 , where it was one of the generators of part i of @xmath348 . ( note that we have assumed that @xmath24 does not contain the whole level @xmath334 , so @xmath348 does not contain all limit points of @xmath346 . ) however , in order to define @xmath349 it is not sufficient to remove the limit point @xmath347 as the result would not be closed under directed suprema . however , we can remove cofinally many elements of component @xmath350 , making sure we are not interfering with part ii of @xmath348 . the result is that the infinite column generated by @xmath347 has been replaced with a finite column generated by @xmath351 where @xmath352 with @xmath295 the height of the `` lower rectangle '' in the language of proposition [ p : closed ] . here . ] the construction continues in the same vein as we add more numbers to the front of the string @xmath353 . we always have to convert an infinite column into a finite one in case the generating limit point was removed in the parent sheet . if @xmath230 is not a suffix of @xmath333 , then we do nt need to manipulate @xmath324 at all in order to obtain @xmath354 . in summary , we have constructed the individual sets @xmath354 in such a way that they conform to the characterisation of locally closed set in proposition [ p : closed ] , that they are compatible with the equivalence relation @xmath240 , and that all elements in @xmath355 are below @xmath241 . crucially , @xmath356 does not contain the element @xmath241 we started out with . case 2 : @xmath357 : in this case we can also assume that @xmath358 because otherwise the maximal element @xmath258 is equivalent to @xmath259 on sheet @xmath261 , contradicting our assumption that the length of @xmath230 is minimal . since we know that @xmath24 does not contain the whole level @xmath359 , it contains only a finite subset of it by principle @xmath291 and @xmath260 is a generator of part i of @xmath360 . as before , we convert the infinite column generated by @xmath260 to a finite column generated by some @xmath361 and continue with this process on subsequent sheets . for the third and final step in our proof of [ p : characterisation ] we turn our attention to closed sets that do not contain any `` isolated '' maximal elements , that is , sets that contain the full level @xmath334 whenever they contain the element @xmath241 for some @xmath206 . levels are indexed by a string ( indicating the sheet ) and a number , but since the equivalence relation identifies @xmath316 with @xmath362 we only need strings of numbers to index them . we say that level @xmath363 is below @xmath364 if @xmath365 . this induces an order @xmath366 between strings of numbers generated by the following two rules : * @xmath367 for all @xmath222 , @xmath238 * @xmath368 for all @xmath369 , @xmath238 with respect to @xmath366 , @xmath227 is a tree with root @xmath229 , in other words , @xmath370 is linearly ordered for all @xmath238 . it is easy to write out the chain above @xmath230 and to check that it is upwards closed , so must be all of @xmath370 : @xmath371 a hasse diagram of @xmath372 would look exactly the same as the one shown in figure [ fig : tree ] ( though there we used a different labelling scheme ) . [ p : tree ] a closed irreducible subset @xmath24 of @xmath41 can only contain one complete level among its maximal elements . let @xmath364 be a maximal level contained in @xmath24 . we claim that @xmath373 is scott - closed . closure under suprema of directed sets is immediate since we are subtracting a downset . the issue at hand is downward closure , so assume there is @xmath374 , i.e. , there is @xmath375 with @xmath376 . the element @xmath78 belongs to some level @xmath377 and since @xmath378 is linearly ordered with respect to @xmath366 , we must have @xmath379 or @xmath380 . in the former case , @xmath381 because we assumed @xmath364 to be maximal in @xmath24 . in the latter case we have @xmath365 by the definition of @xmath366 . in either case we get a contradiction to the assumption that @xmath382 . we decompose @xmath24 as @xmath383 and irreducibility now implies that the second component must be empty . having identified exactly which subsets of @xmath41 are closed irreducible with respect to the scott topology , we can now deliver on our promise , made at the end of section [ sec : positive ] , of giving an example of a non - dominated dcpo . [ prop : hoxi is not dominated ] the dcpo @xmath41 is not dominated . by the preceding proposition , @xmath384 and for every @xmath385 , furthermore , for each @xmath385 , @xmath387 and hence @xmath388 is a directed family in @xmath389 . but @xmath390 equals @xmath41 and it is not the case that @xmath391 . thus , @xmath41 fails to be dominated . reassured by the above result , we now proceed to the next subsection to establish that the dcpos @xmath41 and @xmath42 form the desired counterexample to the ho - zhao conjecture . recall that the unit @xmath392 of the order sobrification monad maps @xmath78 to @xmath52 and is scott - continuous . we will employ this in our proof of theorem [ thm : h = sobh ] : we have the map @xmath393 described in the previous section ; it maps a closed set @xmath48 to @xmath394 . because @xmath395 is scott - continuous we have the map @xmath396 in the opposite direction . we show that they are inverses of each other . for the first calculation let @xmath48 be a scott - closed subset of @xmath41 . @xmath397 for the other composition , let @xmath398 be a scott - closed subset of @xmath42 . @xmath399 our proof will be complete if we can show the reverse of the first implication in the calculation above . for this we use our knowledge about the elements of @xmath42 , that is , the irreducible closed subsets of @xmath41 established in the previous subsection ( proposition [ p : characterisation ] ) . for irreducible subsets of the form @xmath52 the reverse of @xmath400 is trivially true , so assume that @xmath24 is the downset of some level @xmath334 for @xmath238 , @xmath222 , and @xmath401 . in particular we have @xmath402 for the elements @xmath241 of @xmath334 . because all elements of level @xmath197 on sheet @xmath346 are below @xmath403 , we have @xmath404 for all @xmath206 , and hence @xmath405 as the latter is downward closed . finally , @xmath406 so @xmath407 follows as desired . the case @xmath315 is similar . we have shown that the class of dominated dcpos is @xmath4-faithful and also seen that it is quite encompassing ( corollary [ cor : all - dominated ] ) . one may wonder whether there are other natural classes of dcpos on which @xmath4 is faithful or whether @xmath40 is in some sense `` maximal . '' strictly speaking , the answer to this question is no , since the singleton class @xmath408 is ( trivially ) @xmath4-faithful , yet as we showed not contained in @xmath40 . what is needed , then , is a proper definition of `` maximal '' before any attempt can be made to establish its veracity . this research was started when the first author visited the second in the autumn of 2015 , supported by grant sc7 - 1415 - 08 from the _ london mathematical society_. the second author would like to thank the participants of dagstuhl seminar 15441 on _ duality in computer science _ for their comments on an early presentation of the counterexample @xmath41 . he is also grateful to paul taylor and xiaodong jia for discussing and clarifying the categorical setting that gives rise to the order sobrification monad . g. gierz , k.h . hofmann , k. keimel , j.d . lawson , m.w . mislove , and d.s . _ continuous lattices and domains_. number 93 in encyclopedia of mathematics and its applications . cambridge university press , cambridge , 2003 .

in this paper we revisit the levy - perdew - sahni equation . we establish that the relation implicitly contains the conservation of energy density at every point of the system . the separate contributions to the total energy density are described in detail , and it is shown that the key difference to standard density functional methods is the existence of a general exchange - correlation potential , which does not explicitly depend on electron charge . we derive solutions for the hydrogen - like atoms and analyse local properties . it is found that these systems are stable due to the existence of a vector potential , related to electron motion , which leads to two general effects : ( i ) the root of the charge density acquires an additional complex phase ; and ( ii ) for single electrons , the vector potential cancels the effect of electrostatic repulsions . we determine the density of states of a free electron gas based on this model and find that the vectorpotential also accounts for the pauli exclusion principle . implications of these results for direct methods in density functional theory are discussed . it seems that the omission of vector potentials in formulating the kinetic energy density functionals may be the main reason that direct methods so far are not generally applicable . finally , we provide an orbital free self - consistent formulation for determining the groundstate charge density in a local density approximation .

cond-mat0608713

a key innovation in theoretical solid state physics in the last fifty years was the reformulation of quantum mechanics in a density formalism , based on the hohenberg - kohn theorem @xcite . despite initial resistance , in particular from quantum chemists , the method has replaced previous frameworks and provides , to date , the most advanced theoretical model for the calculation of atoms , solids , and liquids . however , its implementation relies on a rather cumbersome detour . while the hohenberg - kohn theorem is formulated exclusively in terms of electron densities and energy functionals , calculations today are based almost exclusively on the specifications given by kohn and sham one year after the initial theorem was made public @xcite . while this procedure is generally successful , and implemented today in numerical methods optimized for efficiency ( see e.g the ingenious way ionic and electron degrees of freedom are treated on much the same footing following a method developed by car and parinello @xcite ) , it is highly inefficient in one crucial conceptual point : if , according to the hohenberg - kohn theorem , the electron density is the only physically relevant variable of the system , then solving the schrdinger equation , setting up the eigenvectors , and computing the density of electrons is an operation , which creates a vast amount of redundant information . every information , pertaining to the solution of the single - particle schrdinger equation and the summation of single electron charges is discarded at the end of every step in the iteration cycle . one could therefore say that more than 90% of the information created in today s simulations is actually irrelevant . the question thus arises : _ do we have to create this information at all , or can we find a more direct way to arrive at the groundstate density of electrons without this cumbersome detour via the single - particle schrdinger equation ? _ in 1984 , levy , perdew , and sahni ( lps ) published an intriguing relation @xcite , in which the density of charge is described by a second order differential equation without any reference to kohn - sham orbitals @xcite . it introduced the possibility to formulate the many - electron problem with just one quantity , the electron density . in this respect it seemed to be more in line with the original hohenberg - kohn theorem @xcite than standard methods in density functional theory , based on the kohn - sham equations @xcite . in the last twenty years , developments in this field focussed on a search for suitable density functional approximations , especially for kinetic energy functionals . it is by now an expansive field of research in theoretical solid state physics and quantum chemistry . the main results of subsequent work on the lps relation @xcite are : the usage of systematically constructed harriman orbitals @xcite in a three - dimensional generalisation @xcite ; the introduction and analysis of the pauli potential @xcite ; the analysis of the uniqueness and asymptotic behaviour of the local kinetic energy @xcite ; the properties of the kinetic energy density @xcite ; and weighted or averaged density approximation @xcite . for a comprehensive review , see @xcite . the list is , of course , not complete . in particular it does not contain references to the development of the kohn - sham theory , e.g. the continued improvement of exchange - correlation functionals , as this is not the topic of the present paper . here , we want to reexamine the original lps relation and its properties . the outline of the paper is as follows : in section [ loc_prop ] we describe the local properties of the lps equation and find that it entails conservation of energy density throughout a quantum mechanical system . in section [ atom_ana ] we give the results of analytic applications of the lps relation for the hydrogen atom . we find that the effective potential is the sum of a non - zero hartree potential and an equally non - zero general exchange - correlation potential . the effective potential is zero due to cancellation of the two contributions . in section [ gen_xc_pot ] we analyse the physical origin of this cancellation and find that motion of single electrons creates a vector potential @xmath0 , called the _ inertial field _ due to its relation to electron motion . due to this potential the root of the charge density acquires an additional complex phase . the phase - shift is in line with an aharonov - bohm effect @xcite . based on these findings we formulate the general problem for an @xmath1-electron system in section [ gen_prob ] . finally , in section [ sum_dis ] we discuss the results and estimate their importance for the development of orbital free density functional methods . a closer analysis reveals that the genuine novelty of the lps relation seems to have been disregarded to this date . it is the conservation of energy at a local level . this is well in advance of the single - particle schrdinger equation or a many - body framework , where energy is conserved only globally . we start with the lps relation , @xmath2 \rho({\bf r})^{1/2 } = \mu \rho({\bf r})^{1/2},\ ] ] rearranging as @xmath3 and multiplying by @xmath4 results in @xmath5 we will show that the potential at a point @xmath6 multiplied by the charge density at this point , describes the potential energy density . thus the equation is nothing else but a description of energy conservation for every point of the system , @xmath7 here , we have symbolized the term @xmath8 by a total energy density @xmath9 . each term refers to a corresponding energy density . the resulting equation is equivalent to the lps relation , which therefore contains energy conservation also in its general form . since the relation between potentials and energy densities may not be directly accessible , we derive them in the following from fundamental considerations . the kinetic energy density @xmath10 is formulated for interacting bosonic @xcite particles . it can exactly be rewritten in a more useful manner , @xmath11 ^ 2}{\rho({\bf r})}= t(\rho({\bf r}),\nabla\rho({\bf r}),\nabla^2\rho({\bf r})).\ ] ] the usefulness of this formulation will be seen further down . it should be noted that the second term of the above expression is the von weizscker kinetic energy density @xcite . for electron charge contained in an infinitesimal volume @xmath12 around a point @xmath6 , the contribution to the interaction energy between electrons and nuclei @xmath13 will be @xmath14 where @xmath15 is the number of nuclei and @xmath16 the charge of the @xmath17th nucleus . the electron - nuclear energy density at a point @xmath6 is then @xmath18 the effective potential contains the explicit electron - electron interaction ( hartree potential , @xmath19 ) and a generalized exchange - correlation potential ( gxc - potential , @xmath20 ) which includes the exchange and correlation effects , as well as the potential contributions due to pauli exclusion , @xmath21 . the contribution to the hartree energy from the electron distribution between two infinitesimal volume @xmath12 and @xmath22 around points @xmath6 and @xmath23 respectively is @xmath24 integrating over @xmath22 and rearranging the results leads to the hartree energy density at point @xmath6 , @xmath25 the same behaviour is found for the general exchange - correlation energy density , @xmath26 although at this point the exact form of @xmath20 is not known . so far we have proved that the left hand sides of eqs . ( [ locenergy ] ) and ( [ locenergy2 ] ) are the same term by term . in order to prove that the total energy density is equal to the chemical potential times the electron density ( @xmath27 ) we employ the variational principle to minimize the total energy . this procedure also allows to determine the relationship between the functional derivatives of the energy terms , and the corresponding energy densities . integrating eq . ( [ locenergy2 ] ) over the whole space the total energy will be @xmath28&=&t[\rho]+e_{ext}[\rho]+e_{h}[\rho]+e_{gxc}[\rho]\nonumber\\ & = & \int d^3r\left[-\frac{1}{4}\nabla^2\rho({\bf r})+ \frac{1}{8}\frac{[\nabla\rho({\bf r})]^2}{\rho({\bf r})}\right ] \nonumber\\&&-\int d^3r\sum_{i=1}^m\frac{z_i\rho({\bf r})}{|{\bf r}-{\bf r}_i| } + \frac{1}{2}\int \int d^3rd^3r ' \frac{\rho({\bf r})\rho({\bf r}')}{|{\bf r}-{\bf r}'|}\nonumber\\ & & + \int d^3r v_{gxc}({\bf r})\rho({\bf r}).\end{aligned}\ ] ] the variational principle , including the condition for the total number of electrons @xmath1 with a lagrange multiplier @xmath29 , provides us with the groundstate energy , @xmath30- \mu\left(\int d^3r\rho({\bf r})-n\right)\right]=0.\ ] ] this leads to the following result : @xmath31 ^ 2}{\rho({\bf r})^2}- \sum_{i=1}^m\frac{z_i}{|{\bf r}-{\bf r}_i|}\nonumber\\ & & + \int d^3r'\frac{\rho({\bf r}')}{|{\bf r}-{\bf r}'|}+ v_{gxc}({\bf r})+\rho({\bf r})\frac{dv_{gxc}({\bf r})}{d\rho({\bf r } ) } = \mu.\nonumber\end{aligned}\ ] ] it must be noted that the energy term @xmath29 carries two separate meanings : mathematically , it is a lagrange multiplier ; physically , it is also the highest occupied level of the groundstate solution of the kohn - sham equations as well as the negative of the ionization energy in the exact dft as was proved earlier by perdew et al . @xcite . in the lps relation , it describes the eigenvalue of the problem , and the change of the total energy , if one electron is removed from the system if @xmath32 @xcite . here , we find the first unconventional implication of the lps equation . since it is not specified , in the derivation of the chemical potential from the lps relation @xcite , which electron is actually removed from the system , the chemical potential must be equal for the removal of any single electron . in this case , it is only possible , like in standard dft , to think of a many - electron system as composed of a discrete number of electrons with different energy levels , if we assume that the system will balance the removal of one electron by a change of all electronic eigenstates so that the highest electronic eigenstate remains empty . but from the previous derivation we may also conclude that @xmath33 describes the total energy of the system . in this case , @xmath29 must have a double meaning : it is , firstly , the energy eigenvalue of the @xmath1-electron system ; and @xmath34 is equal to the total energy . secondly , it is also the chemical potential of the system , or the negative ionization energy . this result differs from standard dft , where the total energy is a sum containing the discrete energy levels of every single eigenstate . to understand the difference we revert to the proof by lps that @xmath29 is the negative ionization energy of a single electron removed from the @xmath1-electron system , if the number of electrons is sufficiently large @xcite . suppose now , we remove one electron from the @xmath1-electron system . the related chemical potential is symbolized by @xmath35 , the negative ionization energy of one electron of the @xmath1-electron system . now , the system contains @xmath36 electrons . this means that the effective potential will be lower than in the first case : @xmath37 considering local energy conservation this implies also that @xmath38 . for calculating @xmath39 we used eq . ( [ eq : t ] ) and the following rule of functional derivatives : @xmath40=\int d^3rt(\rho({\bf r}),\nabla\rho({\bf r}),\nabla^2\rho({\bf r } ) ) , \nonumber\\ \text{then}&\ ; & \frac{\delta t}{\delta\rho}=\frac{\partial t}{\partial\rho } -\nabla\frac{\partial t}{\partial\nabla\rho } + \nabla^2\frac{\partial t}{\partial\nabla^2\rho}.\end{aligned}\ ] ] multiplying eq . ( [ fdere ] ) by the electron density @xmath4 gives , @xmath41 ^ 2}{\rho({\bf r})}- \sum_{i=1}^m\frac{z_i}{|{\bf r}-{\bf r}_i|}\rho({\bf r})\nonumber\\ & & + \int d^3r'\frac{\rho({\bf r}')\rho({\bf r})}{|{\bf r}-{\bf r}'|}+ v_{gxc}({\bf r})\rho({\bf r})\nonumber\\ & & + \rho({\bf r})^2\frac{dv_{gxc}({\bf r})}{d\rho({\bf r } ) } = \mu\rho({\bf r}).\end{aligned}\ ] ] comparing this with eqs . ( [ locenergy]),([locenergy2]),([eq : t]),([eq : eext]),([eq : eh ] ) and ( [ eq : exc ] ) it can be concluded that the energy densities for the different terms are , @xmath42 ^ 2}{\rho({\bf r})},\\ \varepsilon_{ext}({\bf r})&=&\frac{\delta e_{ext}}{\delta\rho } \rho({\bf r})=v_{ext}({\bf r})\rho({\bf r}),\\ \varepsilon_h({\bf r})&=&\frac{\delta e_h}{\delta\rho } \rho({\bf r})=v_h({\bf r})\rho({\bf r}),\\ \label{exc } \varepsilon_{gxc}({\bf r})&=&\frac{\delta e_{xc}}{\delta\rho } \rho({\bf r})=v_{gxc}({\bf r})\rho({\bf r})\;,\;\frac{dv_{gxc}}{d\rho}=0.\end{aligned}\ ] ] here , we arrive at an interesting consequence for the general exchange - correlation potential . the energy density for each separate term is described by the functional derivative times the electron density , even for the most complicated kinetic energy term . if this procedure holds also for the general exchange - correlation term , as written in eq . ( [ exc ] ) , then the general exchange - correlation potential does not explicitly depend on the electron density . this is in marked contrast to the standard formulations in dft , where exchange - correlations not only depend on the density of charge , but are generally parametrized in terms of the density ( e.g. @xcite ) . the energy conservation on a local level , i.e. , at every point @xmath43 of the system , can thus be directly deduced from the lps relation . comparing the energy densities with the functional derivatives of the corresponding energy terms has another consequence for the kinetic energy density @xmath10 and the kinetic energy functional @xmath44=\int d^3rt({\bf r})$ ] . starting from eq . ( [ eq : t ] ) , integrating over the whole space , performing the functional derivative , and multiplying by the electron density leads again to the kinetic energy density . from that cycle it can be clearly seen that the kinetic energy density _ can not _ contain arbitrary terms whose space integral vanishes . this finding is in contrast to previous assumptions @xcite . it is justified by the local energy conservation of the lps relation itself . let us analyse the asymptotic behaviour of eq . ( [ fdere ] ) as @xmath45 . after considering that @xmath46 and @xmath47 @xcite we arrive at @xmath48 which was found by yang et al . this means that there is no need of approximating the kinetic energy functional , it is exact and unique and has the correct asymptotic behaviour , although it describes non - interacting bosons . the detailed analysis reveals three important properties of the lps equation : * the energy density is conserved at a local level ; the total energy density is therefore constant throughout the system . * the kinetic energy density is unique , it does not contain any arbitrary terms . * the general exchange - correlation functional does not explicitly depend on the density of charge . from a practical point of view we note that self - consistency within the lps framework will be much faster to achieve , since the density distribution is much more restricted . in particular the requirement that the total energy density must be constant at every point of the system should make the construction of fast algorithms quite easy . in this respect it must also be noted that solutions of the equation scale with the volume of the system , thus the number of atoms : the method is therefore a true order - n method . in the following we analyse the results of the lps relation for simple atomic systems containing one electron like h , he@xmath49 , and li@xmath50 . previously , we have found that the direct consequence of the lps equation is the energy conservation on a local level . @xmath51 as a next step , we analyse the potentials and energy terms which contribute to the total energy , @xmath52 depending on an assumption about electron density . in general , the electron distribution in atoms is spherically symmetric , with the nucleus occupying the centre of the sphere . unless otherwise stated we use atomic units throughout the rest of the paper . we assume in the following that the electron density decays exponentially within the atom @xmath53 , with @xmath54 . this type of electron density is exact for hydrogen - like atoms , with the values ( hydrogen ) of @xmath55 . with this ansatz all the relevant variables of the atomic system can be explicitly written down . the number of electrons is @xmath56 the following quantities are necessary for evaluating the kinetic energy density : @xmath57 ^ 2}{\rho(r)}&=&\alpha\beta^2e^{-\beta r } , \nonumber \\ \nabla^2\rho(r)&=&\frac{d^2\rho(r)}{dr^2}+\frac{2}{r}\frac{d\rho(r)}{dr } = \alpha e^{-\beta r}\left[\beta^2-\frac{2\beta}{r}\right ] . \nonumber\end{aligned}\ ] ] thus the kinetic energy density will be @xmath58 ^ 2}{\rho(r ) } = \alpha e^{-\beta r}\left[-\frac{\beta^2}{8}+\frac{\beta}{2r}\right].\nonumber\\\end{aligned}\ ] ] the kinetic energy is @xmath59\nonumber\\&=&\pi\frac{\alpha}{\beta } = \frac{n\beta^2}{8}.\end{aligned}\ ] ] the chemical potential is given by @xmath60 = -\frac{\beta^2}{8}\ ] ] the external potential is @xmath61 , with @xmath62 being the nuclear charge ; the electron - nuclear energy is consequently @xmath63 the hartree potential can be obtained using gauss theorem via the radial component of the electric field : @xmath64,\nonumber\\ v_h(r)&=&-\int_\infty^r dr'e_{r'}(r')=\frac{4\pi\alpha}{\beta^3 } \left[\frac{2}{r}-e^{-\beta r}\left(\beta+\frac{2}{r}\right)\right]= \nonumber \\ & & + \frac{n}{r}-e^{-\beta r}\left[\frac{n\beta}{2}+\frac{n}{r}\right].\label{vh}\end{aligned}\ ] ] the hartree energy is thus @xmath65 \nonumber \\&=&\frac{5\pi\alpha n}{2\beta^2}=\frac{5}{16}n^2\beta\end{aligned}\ ] ] the gxc - potential can be obtained from eq . ( [ lps ] ) : @xmath66 \nonumber\\ & = & \frac{1}{r}\left[z - n-\frac{\beta}{2}\right]+ e^{-\beta r}\left[\frac{n\beta}{2}+\frac{n}{r}\right]\label{vgxc}\end{aligned}\ ] ] adding the hartree potential to the gxc - potential gives us all electron - electron related potentials , @xmath67 we get the gxc - energy as @xmath68 according to eq . ( [ etot ] ) the total energy will then be @xmath69 comparing the total energy to the kinetic energy in eq . ( [ tt ] ) , we find that @xmath70 . this demonstrates the validity of the virial theorem for atomic systems , i.e. @xmath71 . moreover , it becomes clear that @xmath29 is an _ average chemical potential _ , i.e. @xmath72 , where @xmath33 is the sum of the chemical potentials as we build up our system from 1 to n electrons : @xmath73 the single electron eigenvalues @xmath74 can directly be determined by dft . all terms in eq . ( [ lps ] ) are now parametrized . the atomic systems are described by three parameters @xmath75 , although , it has to be noted that the form of the electron density considerably changes for @xmath76 which means the above procedure is only valid for atomic systems with one electron . @xmath77 is directly related to the total energy ( @xmath33 ) , a fact known from experiments and dft calculations : @xmath78 now , we have all the tools to analyse the potentials for systems containing one electron ( @xmath79 ) . the electron distribution of hydrogen can of course be determined analytically , using the single particle schrdinger equation . however , to elucidate the physical features of the lps model , it is instructive to choose a different route . in case of taking the groundstate ( @xmath80 ) of the hydrogen the three parameters have the following values : @xmath81 . the calculated potentials and densities are shown in fig . [ fig : h ] . + is equal , but of opposite sign as @xmath82 . the effective potential @xmath83 is therefore zero . ] + the calculated values @xmath84 and @xmath85 are exactly the same as in the analytical treatment . however , we gain additional insights into the relation between hartree and exchange - correlation potentials : they are not zero , as they would be following the single particle schrdinger equation , but @xmath86 at every @xmath87 . this behavior also defines the limit for the effective potential @xmath88 . physically speaking , it indicates that a single electron does not interact with itself , not , as in the conventional picture , because electron - electron repulsion ( @xmath89 ) is zero , but because it is cancelled by exchange - correlation . for this reason we called @xmath90 the _ general exchange - correlation potential _ , because it is inconceivable , within a statistical many - electron treatment , that it will be non - zero for a single electron . another important feature of @xmath91 is that it can be negative . however , this feature was already analysed in detail by sim et al . @xcite . the mathematical condition for @xmath86 can be determined by comparing the terms in eqs . ( [ vh ] ) and ( [ vgxc ] ) . it is @xmath92 . since , according to eq . ( [ etot1 ] ) : @xmath93 which is always true for @xmath79 . if the total energy @xmath33 decreases to @xmath94 , then the decay @xmath77 , and , consequently , the effective potential , will be : @xmath95 the energy @xmath33 is therefore the minimum energy of the hydrogen atom ; the actual energy is bound from below . the atom is thus stabilized by repulsive electron - electron interactions , which _ can not _ be more than compensated by exchange - correlation interactions . a decrease of the energy , and thus an implosion of the atom , is prevented by the fact that exchange - correlation would have to be larger than the repulsive electron - electron interaction . the physical cause , invoked here to explain atomic stability , is different from the standard one , where it is thought that a decrease of the atomic volume would increase the energy uncertainty of the electron and thus increase the electron s energy . the standard account is based on the heisenberg uncertainty relations . the reason we arrive at a different account within the lps framework are the local properties . since within the standard framework only global properties are accessible , an argument can only be made on the basis of global properties , e.g. volume and energy . however , the lps framework provides direct access to all _ local _ properties . therefore the argument can be made on a local basis , via the local interactions of electron density and fields of interactions . it is tempting to extend the analysis to the general case of isolated electrons . in case of @xmath79 electron if @xmath62 becomes very small @xmath96 , the effective potential in ( [ veff:4 ] ) will remain zero , because @xmath97 . electron charge therefore will become close to homogeneous . but also in this case the effect of electron - electron repulsion is cancelled because of exchange - correlation . this seems to indicate , that regardless of the actual extension of electron charge , electrons will _ not _ experience any repulsion between the charge densities at different regions of the electron : electrons are therefore stable entities . in the general case @xmath98 the total energy of an arbitrary system of @xmath1 electrons must be greater than @xmath99 this value of @xmath33 gives the exact total energy for a system of non - interacting electrons . the total energy in this case can not decrease below @xmath33 because this would correspond to an attractive interaction between electrons . analysing @xmath100 and @xmath101 ions and the @xmath102 excited state of the electron in @xmath103 atom , all systems containing @xmath79 electron , @xmath104 was always found . to sum up the results of this analysis of hydrogen - like systems : ( i ) we found in all cases that the hartree potential is non - zero . this is in line with the general concept of dft , where extended distributions of electron charge always have an effect on electrostatic potentials . ( ii ) we also found that the atoms are stable due to the peculiar features of the general exchange correlation potential : it is a negative contribution , equal in magnitude to the hartree potential , which prevents the repulsion of electron charge to have any effect . the extent of the charge distribution is thus bound from above . but it is also bound from below , since the general exchange correlation potential does not exceed the magnitude of the hartree potential . hydrogen - like atoms as well as free electrons are therefore stable entities . however , at this point the physical nature of this potential remains mysterious : it does not explicitly depend on the charge distribution , as has been generally derived from the lps relation . it can also not be related to exchange correlation in standard dft , since the systems are composed of only one electron . at this point we have to look at the physical situation from a more fundamental point of view . in general , electrons are described by three physical properties : ( i ) mass , ( ii ) charge , and ( iii ) spin . from a fundamental point of view , one of these properties must be related to the existence and the properties of the general exchange correlation potential . it is straightforward to exclude charge itself as its origin , since both effects of charge , the electron - electron repulsion , and the electron - nuclear attraction , are part of the description . mass , even though it fits one of the characteristics of the general exchange correlation , i.e. the attractive nature , can not be responsible for it , because the coupling constant for gravitational interaction is tens of orders of magnitude smaller than the coupling constant for electrostatic interactions . this leaves only one viable option : the general exchange correlation potential must be related to the magnetic properties of electrons . in general , motion of charge is related to the existence of fields . conversely , fields will affect the motion of charge . from this perspective the motion of electron charge within a hydrogen atom is likely to create corresponding fields , which shall have an effect on the energy density at a particular point of the system . there is , however , one restriction : these fields can not be equal to electromagnetic @xmath105 and @xmath106 fields , as this would lead to a change of the energy of the atomic system due to radiation . the field in question can therefore only be a vector potential @xmath0 . the kinetic energy operator used in the previous sections does not contain any field components . for this reason it most likely gives only a limited account of the physical environment . the existence of a field @xmath0 and the ensuing kinetic energy operator @xmath107 will introduce imaginary components into the schrdinger equation . real functions @xmath108 are thus not sufficient to describe electrons in this situation . this requires to generalize the wavefunction for complex phases and to reinvestigate the lps relation according to this change . the motivation for this procedure is to determine the physical origin of the general exchange correlation potential . to this end we start with a free electron in a three - dimensional potential well . in this case the electron can be described as a plane wave , @xmath109 with @xmath110 the volume of the potential well and @xmath111 the momentum of the electron . the above wavefunction results in a constant electron density , @xmath112 from the above it seems to be clear that the lps equation , eq . ( [ lps_phi ] ) can not be complete , since for the free electron the square root of the electron density is constant which implies that @xmath29 has to be zero , which is obviously not the case . a similar remedy was suggested earlier by wang and carter , see section vi.1 of @xcite . here , we extend their results and assume the most general wavefunction to be complex . the wavefunction contains two important features of the electron : amplitude and phase , generally @xmath113 writing the schrdinger equation with this wavefunction in the most general way ( we omit space and time dependencies for brevity ) , @xmath114\psi= \nonumber\\ \hat h\psi&=&\left[-\frac{1}{2}\nabla^2 + v_{ext}+v_{h}+v_{gxc}\right]\psi\,\end{aligned}\ ] ] results in two equations for the real and imaginary parts , @xmath115 here , the most general @xmath90 is a complex potential . in a stationary state an eigenvalue equation applies , @xmath116\psi=\hat h\psi=\mu\psi,\end{aligned}\ ] ] which again results in two equations ( real and imaginary parts ) , @xmath117 eq . ( [ time_dep_ri ] ) restricts the wavefunction to the simpler form , @xmath118 the first of eqs . ( [ lps - modified ] ) is the modified lps equation where the phase function of the electron explicitly occurs modifying the kinetic energy of the electron system ( see next section ) . applying this equation to the free electron gives us the correct result for the total energy ( @xmath29 for @xmath79 electron ) : @xmath119 this modification accounts for the well known problem in orbital free methods , that the kinetic energy functionals of atomic systems , usually described by some modification of the von weizscker functional ( see above ) , or the kinetic energy functionals of free electron systems , like in many metals , are completely different . wang and carter proposed a lindhard function @xcite , which connects these two extreme cases in a common mathematical model . by contrast , we find here that the difference can be accounted for by the addition of a complex phase to the root of the charge density . we can sum up the results of this analysis by saying that it is essential , for the description of free electrons , to take into account both the amplitude and the phase information . as a consequence of this , the lps equation has to be modified . the phase information , we found , enters as an additional term in the kinetic energy density , @xmath120 . in previous sections we have shown that it is essential , for the stability of atoms , that the general exchange correlation is not zero . we claim that this is also the case for the free electron , see next section . furthermore , it was shown that a complex phase is essential for obtaining the correct energy eigenvalue . here , we want to determine the origin of this phase and show that the general exchange correlation is due to an inertial field of the electron . to this end let us consider the modified momentum of a charged particle travelling through a region of space with nonzero electromagnetic potentials . it is well known that the momentum of such an electron is @xmath121 where @xmath0 is the vectorpotential , while the curl of the vectorpotential is the external magnetic field , @xmath122 the aharonov - bohm effect @xcite established the importance of electromagnetic potentials in quantum physics . for example , in a system containing an external magnetic field , e.g. a solenoid , an electron is affected by the vectorpotential , even propagating through a region of space , where the external magnetic field is zero . it acquires the phase @xmath123 the effect is also observed in experiments @xcite . under the condition that an inertial field @xmath124 , i.e. a field , which is _ not _ due to external sources , but due to the _ propagation of the electron itself _ , exists , the properties of the general exchange - correlation potential can be readily derived . the hamiltonian now takes the form , @xmath125 ^ 2+v_{ext}({\bf r})+v_h({\bf r})\\&=&-\frac{1}{2}\nabla^2+\frac{1}{2}{\bf a}({\bf r})^2+\frac{i}{2}[{\bf a}({\bf r})\nabla+\nabla{\bf a}({\bf r})]\nonumber\\&&+ v_{ext}({\bf r})+v_h({\bf r}).\nonumber\end{aligned}\ ] ] comparing this to eq . ( [ lps_phi ] ) and using that @xmath126 it is straightforward to conclude that the generalized exchange - correlation operator is @xmath127\nonumber\\&= & \frac{1}{2}{\bf a}({\bf r})^2 + \frac{i}{2}[\nabla\cdot{\bf a}({\bf r})+2{\bf a}({\bf r})\nabla].\end{aligned}\ ] ] using this notation the hamiltonian can be written as @xmath128 the actions of each term of the hamiltonian on the time - space separable wavefunction , eq . ( [ psi- ] ) are @xmath129 ^ 2\right)\psi({\bf r},t)\nonumber\\&&-i\left ( \frac{1}{2}\nabla^2\varphi({\bf r})+ \frac{\nabla\rho({\bf r})^{1/2}}{\rho({\bf r})^{1/2}}\nabla\varphi({\bf r } ) \right)\psi({\bf r},t),\nonumber\\ \hat v_{ext}({\bf r})\psi({\bf r},t)&=&v_{ext}({\bf r})\psi({\bf r},t),\\ \hat v_h({\bf r})\psi({\bf r},t)&=&v_h({\bf r})\psi({\bf r},t),\nonumber\\ \hat v_{gxc}({\bf r})\psi({\bf r},t)&=&\left(\frac{1}{2}{\bf a}({\bf r})^2 -{\bf a}({\bf r})\nabla\varphi({\bf r})\right)\psi({\bf r},t)\nonumber\\&&+ i\left(\frac{\nabla\cdot{\bf a}({\bf r})}{2}+ \frac{\nabla\rho({\bf r})^{1/2}}{\rho({\bf r})^{1/2}}{\bf a}({\bf r } ) \right)\psi({\bf r},t).\nonumber\end{aligned}\ ] ] it is important to mention that the kinetic energy is modified by a term , @xmath130 ^ 2/2 $ ] which directly relates to the phase of the electron . this differs from the conception of e.g. vignale and kohn @xcite , where the field is related to current density . in previous work on orbital free dft this term was never taken into account in the way described here . adding all terms yields two equations for real and imaginary part , respectively , @xmath131 ^ 2+v_{ext}({\bf r})+v_h({\bf r}),\nonumber\\ 0&=&\left(\frac{1}{2}\nabla+\frac{\nabla\rho({\bf r})^{1/2}}{\rho({\bf r})^{1/2}}\right ) [ { \bf a}({\bf r})-\nabla\varphi({\bf r})].\end{aligned}\ ] ] the second equation leads to the simple solution @xmath132 where @xmath133 . the phase of the electron is therefore related to a quantity which can be interpreted as a vectorpotential . the result seems not surprising , considering the aharonov - bohm effect @xcite . however , an objection may be raised that this effect relates external @xmath0 fields to observed phase shifts , and that a phase - shift due to propagation of the electron itself will change the theoretical predictions for the experiments . the readers may convince themselves that this is not the case . if two electrons passing on either side of a solenoid possess the same inertial field @xmath0 , then the phase difference between any two paths with the same endpoint around the solenoid will be given by @xmath134 it is therefore determined exclusively by the external field . the main difference in our present model is that @xmath124 is the inertial field due to electron propagation , not a vectorpotential caused by external fields . the magnetic field due to this vectorpotential will vanish , since the curl of a gradient is always zero . substituting eq . ( [ a - gradphi ] ) into the first of eqs . ( [ lps - a ] ) gives @xmath135 the electron wavefunction takes the form , @xmath136 using this solution , we find that the real kinetic energy density @xmath137 is not the bosonic kinetic energy density , employed in the original lps relation , but that it possesses an additional term , depending on the inertial field @xmath0 . @xmath138 in this case we obtain a corresponding exchange correlation energy density described by : @xmath139 if the inertial field @xmath140 vanishes , the exchange correlation potential @xmath141 is zero . in addition , the sum of the real kinetic energy density @xmath137 and the exchange correlation energy density is always equal to the bosonic kinetic energy density : @xmath142 this entails that the fermi - ionic nature of electrons is due to their inertial fields . we shall show further down that a gauge transformation , which eliminates the inertial fields , reverts the problem back to a problem of interacting bosons . to demonstrate the consequences of the framework developed , let us analyse a free electron enclosed in a finite volume @xmath110 . the problem is quite interesting to analyse , as dft does not describe the system correctly . this is due to the fact that in dft both exchange correlation potential @xmath143 and hartree potential @xmath89 vanish , which is at odds with the physical situation comprising a finite distribution of electron charge . in our description we treat the hartree term in the correct manner . from general considerations we have found that @xmath144 for systems of one electron . it should also be noted that @xmath145 is zero within the box and infinity outside . as the phase of the wavefunction in eq.([psi - free - electron ] ) is @xmath146 , the inertial field of the free electron is @xmath147 this means that the inertial field is just the momentum of the free electron system which is uniform within the volume . this feature of free electrons was predicted by heuristic arguments some years ago @xcite . the curl of the inertial field is , of course , zero . the magnetic field therefore vanishes . taking this fact and the explicit form of @xmath90 in eq . ( [ v_gxc - a ] ) into account , we can write the modified lps equation as @xmath148\frac{1}{\sqrt{v}}e^{i{\bf k}{\bf r}}\\&= & \left[\frac{1}{2}{\bf k}^2+v_h({\bf r})-\frac{1}{2}{\bf k}^2 \right]\frac{1}{\sqrt{v}}e^{i{\bf k}{\bf r}}= \mu\frac{1}{\sqrt{v}}e^{i{\bf k}{\bf r}}.\nonumber\end{aligned}\ ] ] since @xmath149 , the chemical potential @xmath150 , and all potentials and also the kinetic energy density divided by the density have the same magnitude : @xmath151 in addition , they are also equal to the components of the total energy of the system , as @xmath79 : @xmath152 as @xmath153 , @xmath154 and @xmath155 . however , this is not the general formulation of the problem . in the general case , elaborated in detail for a homogeneous electron gas in the next section , the field @xmath156 will be complex , and we have to account for partial waves @xmath157 and @xmath158 traveling in opposing directions . the general treatment will be based on a modification of the lps equation to account for this situation . let us determine the inertial field of the electron in a hydrogen - like atom . the groundstate density and wavefunction can be written as @xmath159 the system has a total energy of @xmath160 . acting @xmath161 in eq . ( [ v_gxc - a ] ) on this wavefunction results in two equations for the real part and imaginary part , respectively , @xmath162 here , we assumed that @xmath163 has only radial components , @xmath164 . we have already calculated @xmath165 in eq.([vgxc ] ) , the parameters in our case are @xmath79 and @xmath92 . comparing this to the real part of @xmath90 , the inertial field can be obtained , @xmath166 , \nonumber\\ a_r(r)=\pm\sqrt{-\frac{2}{r}+2e^{-2zr}\left[z+\frac{1}{r}\right]}.\end{aligned}\ ] ] the inertial field @xmath167 in this case is imaginary . however , this results is only puzzling , as long as the direction of motion remains unconsidered . we have found , for the free electron , that @xmath167 is a vector field , which is parallel to the motion vector of the electron . if we assume , in case of the hydrogen electron , that the vector of motion is radial , then it can not be unique , because in this case the electron distribution can not be stable . this leads to the conclusion , that the standard time - independent solution must comprise two separate cases , with the vector of motion either parallel , or anti - parallel to the radial vector . the field @xmath167 should therefore be a superposition of two partial fields , @xmath168 , and @xmath169 . with the ansatz for the partial fields , @xmath170 we obtain for the total field @xmath164 @xmath171 the inertial field in hydrogen is thus described by a radial function @xmath172 , which complies with @xmath173}\right).\ ] ] for @xmath174 this leads to a well behaved solution , since @xmath175 the limit for @xmath176 is consequently : @xmath177 } = \sqrt{2z}.\ ] ] note that the real components of @xmath178 and @xmath179 possess opposite signs , in line with the previous findings for free electrons . the real and imaginary components of @xmath178 are shown in fig . [ fig : if ] . it is interesting to note that @xmath178 ( i ) does not show a singularity like the external potential , and ( ii ) does not decay exponentially . if we assume that the real part of the inertial field is related to electron motion , then the differential of the phase @xmath180 can be used to calculate the phase . since the phase is related to the wavelength of the electron via @xmath181 , we can also calculate the wavelength of the hydrogen electron as a function of radius : @xmath182 as seen in fig . [ fig : if ] the wavelength near the hydrogen core is about eight or @xmath183 atomic units , it saturates for high radii at about @xmath184 . within hydrogen . the components are equal at the limit of zero radius . the wavelength of the hydrogen electron ( dashed graph ) saturates at large radii . ] from a physical point of view , the real part of @xmath167 increases as the wavelength decreases . by contrast the imaginary part of @xmath167 decreases with decreasing density and increasing radius . tentatively , one could thus interpret the real part of @xmath167 as a physical quantity related to motion , and the imaginary part of @xmath167 as something akin to internal friction of the electron . at the boundary of the nucleus , or for @xmath176 , both components are equal , signifying that at this point the energy of motion is completely transformed into friction . here , we demonstrate that the inertial field also accounts for the pauli exclusion principle in a natural manner . separate pauli potentials , as proposed by march , or levy and ou - yang @xcite are therefore unnecessary . to this end we consider the simplest possible system , the system of free electrons . the inertial field of a single electron is given by @xmath185 , the corresponding energy of the electron is @xmath186 in a system of two electrons the two inertial fields will be superimposed . assuming , initially , that superimposed fields are parallel in space , the combined field of the two electrons will be @xmath187 since the density of charge will increase by a factor of two , the hartree potential and the general exchange correlation potential will increase by the same amount : a system of two electrons remains thus stable due to the superposition of the two inertial fields . a different way of expressing this situation would be that for every single electron the effective potential remains zero . for a system of @xmath1 electrons an equivalent description holds , so that the total energy of @xmath1 electrons is given by : @xmath188 the result is exactly the same as the one derived for a free electron gas in one dimension , which is usually accomplished using periodic boundary conditions . here , we have only used the property of any field , i.e. the superposition of amplitudes . however , in a general three dimensional system the direction of fields can vary . in this case the assumption of parallel vector potentials is not the state of minimum energy . then the state of minimum energy is described by : @xmath189 since the individual fields @xmath190 will have equal length @xmath191 , the minimum of the total field @xmath192 will be a sphere , equivalent to the fermi sphere in a three dimensional free electron model . note that in all these cases the addition of one electron increases the volume of occupied states in reciprocal space according to the pauli exclusion principle . the volume of one cell in this sphere is determined by the amplitude of the vector potential . in this section we assume a stationary electron state and investigate how it is transformed applying a gauge transformation . the effect on the schrdinger equation is also determined . this analysis shall aid us in developing methods for an effective solution of the electron problem . here , the aim is to find a suitable gauge for the inertial field @xmath0 . electromagnetic potentials can be transformed with a gauge transformation which results in the same electromagnetic fields , @xmath193 with @xmath194 the electromagnetic scalar potential and @xmath195 the gauge function . the wavefunction transforms like @xmath196 it is easy to show that the schrdinger equation transforms according to the following rule , @xmath197 \psi'({\bf r},t),\nonumber\end{aligned}\ ] ] where the initial state @xmath198 is supposed to be stationary . gauge transformations are potentially useful in reducing the complexity of the problem and finding general solutions . for example , the following transformation reverts a general wavefunction back to the root of the charge density , @xmath199 which describes a bosonic system as @xmath200\rho^{1/2}= i\frac{\partial\rho^{1/2}}{\partial t}=0.\ ] ] this is the original lps equation , see eq . ( [ lps_phi ] ) , without exchange correlation potentials . it describes interacting bosons ( @xmath201 ) , and does not comply with the pauli principle . the obvious conclusion from this result is that the pauli principle and the fermionic character of electrons is due to their phases . in previous sections it was established that the general problem of finding the groundstate of an @xmath1-electron system can be described as finding the six components - three real , three imaginary - of the complex vector field @xmath124 , the distribution of electron density @xmath4 , and the chemical potential @xmath29 . the complex vector field determines the phase of electrons , which is _ not _ described by charge density alone . from this perspective the seeming detour taken in kohn - sham dft is quite understandable . since charge density alone is insufficient to guarantee phase - continuity throughout a system , it has to be modelled by continuous phases of single electron states , taking into account electron - electron interactions via exchange - correlation functionals . within the present context , this detour is unnecessary , if the inertial field can be calculated . that in this case six additional values have to be determined for every point of the system is not a critical problem . after all , the whole computation still scales linearly , even though the prefactor might be somewhat larger . in addition , boundary conditions apply for the inertial field , and the chemical potential has to be constant throughout the system . given a complex vector field @xmath202 , the general problem can be formulated in two equations : @xmath203\rho^{1/2 } = 0,\nonumber \\ \left(\nabla{\bf a}_r - { \bf a}_r\nabla - 2{\bf a}_r{\bf a}_i\right)\rho^{1/2 } = 0.\end{aligned}\ ] ] it is straightforward to separate the different contributions and to assign them to potentials within standard dft . the square of the real part of the vector field accounts for the pauli principle , as discussed earlier . thus we may define a pauli potential by : @xmath204.\ ] ] the exchange and correlation potentials are defined in standard dft as the difference of the kinetic energy between interacting and non - interacting electrons . they correspond in this picture to an operator , @xmath205 , described by : @xmath206.\ ] ] the pauli potential and the exchange - correlation potential are not independent from each other , since they have to comply with the auxiliary condition : @xmath207\rho^{1/2}({\bf r } ) = 0.\ ] ] given this condition , the general problem is thus described in a similar fashion to the original lps equation by ( the dependency on coordinates is again omitted for brevity ) : @xmath208\rho^{1/2 } = 0.\ ] ] the full potential therefore has four distinct terms : ( i ) an external potential @xmath145 , which only depends on the distribution of positive charge , it is described in every dft framework ; ( ii ) a hartree potential @xmath89 , which depends only on the charge distribution , which also is present in standard dft methods ; ( iii ) a pauli potential @xmath209 , which depends on the amplitude of the inertial field , in standard kohn - sham theory it is accounted for by computing the solutions for single electron states ; and ( iv ) an exchange - correlation potential @xmath210 , which depends on the phase of the inertial field , described by local densities and their gradients in standard methods . note that within the present framework many - body effects are related to the inertial field @xmath0 ; in general they are thus mediated by field interactions . the effective potential , used in the original lps relation , is thus in fact an operator , given by : @xmath211 as an example , let us discuss the solution for a homogeneous electron gas . from the auxiliary equation eq . [ aux : eq ] , we may infer , as the simplest solution , that the real part of the vector potential is zero : @xmath212 from translational symmetry for the external potential and the hartree potential , and from the fact that the hartree potential scales with the density of charge , as well as from the translational invariance of the charge density @xmath213 , we get : @xmath214 thus we arrive at the nonlinear equation for the imaginary part of the potential , or : @xmath215 + \alpha \rho_0 - v_0 = \mu.\ ] ] choosing again , for translational invariance , a solution where @xmath216 , we finally arrive at : @xmath217 for a neutral system , where the external potential due to the positive background charge and the hartree potential must be equal , or @xmath218 , this amounts to : @xmath219 in this case the exchange - correlation potential of the homogeneous electron gas is described by @xmath220 . from a known chemical potential @xmath29 the inertial potential can therefore be calculated in a straightforward manner . under the same conditions the vector field for slowly varying densities , which we denote by @xmath221 , will be given by ( we denote by @xmath222 the chemical potential of the homogeneous electron gas of a specific density @xmath223 ) : @xmath224}\ , \frac{\nabla \mu_h[\rho({\bf r})]}{\left|\nabla \mu_h[\rho({\bf r})]\right|}.\ ] ] consistent with earlier findings , the exchange - correlation potential does not depend explicitly on the density of charge . the explicit pauli potential for the @xmath1 electrons of the homogeneous electron gas can be set to zero , because it is implicitly contained in the chemical potential @xmath222 . @xmath225 $ ] , in turn , can be determined by standard methods in dft . if the vector field is calculated from the homogeneous electron gas , then the missing terms in the general problem are accounted for . this amounts to a local density approximation for the general problem . assuming a distribution of @xmath15 ions , and a number of @xmath1 electrons within our system , and with @xmath226 from eq . [ ima : eq ] , the general problem in the local density approximation is described by : @xmath227 ^ 2}{\rho({\bf r})^2 } - \frac{1}{2}{\bf a}_h^2({\bf r } ) \nonumber \\ & & \pm \frac{1}{2 \rho^{1/2}({\bf r})}\left[{\bf a}_h({\bf r})\nabla - \nabla { \bf a}_h({\bf r})\right ] \rho^{1/2}({\bf r } ) \nonumber \\ & & - \sum_{i=1}^m\frac{z_i}{|{\bf r}-{\bf r}_i| } + \int d^3r'\frac{\rho({\bf r}')}{|{\bf r}-{\bf r}'| } , \\ { \bf a}_h({\bf r } ) & = & \sqrt{-2 \mu_h[\rho({\bf r})]}\ , \frac{\nabla \mu_h[\rho({\bf r})]}{\left|\nabla \mu_h[\rho({\bf r})]\right| } , \nonumber \\ n & = & \int d^3r ' \rho({\bf r } ' ) . \nonumber\end{aligned}\ ] ] it is interesting to note that the exchange - correlation potential contains not only the values of @xmath226 and @xmath4 , but also their derivatives . this is due to the fact that the inertial fields are related to the phases of the electrons . in this case all terms of the equation depend on the local charge and its derivatives . a self - consistency cycle then can proceed in the familiar manner by iterating the charge density distribution until the chemical potential is a minimum . as the only input parameter in the calculation is the density of charge itself , with the auxiliary quantities taken from the results obtained for a homogeneous electron gas , the method is a true order-@xmath1 method . however , it is not strictly necessary to remain at this level of approximation . assume , that a groundstate solution has been found . then the initial values of @xmath4 , @xmath228 , and @xmath29 are known . subsequently , the solution can be refined , by ( i ) using eq . [ aux : eq ] to determine the real part of the vector potential ; ( ii ) solving the general problem with the help of eq . [ gen : eq ] ; and ( iii ) iterating the vector field and the density distribution until the chemical potential is a minimum . in this case the solution is the true many - body solution of the problem . let us briefly summarize the findings in the previous sections . it was found that the lps relation describes local conservation of energy density . it can be generally derived by variational methods from the total energy functional , and emerges as the euler - lagrange equation of the energetic minimum . its application to hydrogen - like atoms showed that these systems are stable , because the effective potential , @xmath229 , is zero . however , the general exchange correlation potential , @xmath90 , which is equal in magnitude but of opposite sign as the hartree potential , can not be explained from standard exchange and correlation effects for systems containing only one electron . an analysis with a more general kinetic operator including the effect of @xmath0 fields showed that the gxc - potential is due to an inertial field @xmath0 , related to electron motion , which leads to a lowering of total energy . the phase of free electron is also due to this inertial field , @xmath230 which has the same origin as the aharonov - bohm effect @xcite . the inertial field @xmath0 was found to be equal to the momentum @xmath111 for the free electron . a system of @xmath1 free electrons will be determined in reciprocal space by a sphere of finite radius . the model in this case includes the pauli exclusion principle in a natural manner . the general problem can be formulated in terms of the local charge density distribution and the local complex vector field . from the results for a homogeneous electron gas we derived the solution for the vector field . this allowed to construct a local density approximation for the general problem , which is stated exclusively in terms of the density of charge , its derivatives , and auxiliary quantities inferred from a homogeneous electron gas . these results are quite unexpected . although it is probably too early to appreciate the full extent of these findings , some general points can be made . the existence of inertial fields @xmath0 , related to electron motion , seems to relate to the problem of finding transferable kinetic energy functionals in dft . if , as it seems , a crucial part of the physical picture has so far been omitted , then it is not surprising , that so far every formulation has been somewhat less than transferable . it is also important to note that the central result of the hohenberg - kohn theorem , i.e. the proof that the groundstate charge density minimizes the total energy , remains untouched by our findings . a complex phase added to the root of the charge density will not alter the charge density at any point . in this case the total energy of a given charge distribution is still a minimum , even though additional contributions due to the complex phase exist . from a practical point of view it could substantially aid in the development of local many - electron frameworks . the inertial fields are related to electron motion ; as they are fields , they should be superimposable and have to comply with boundary conditions . electron charge by itself does not have the same constraints . the well known and extensively developed framework of classical electrodynamics may in this case also become useful for the study of field properties in quantum mechanical systems . the field @xmath0 is related to magnetic properties of electrons , even if , as in case of free electrons or hydrogen - like electrons , the corresponding magnetic field @xmath106 is zero . it may well be that the ultimate answer to the question , what spin actually is , will come from a closer analysis of the field under varying external conditions . in addition , the finding could lay to rest a century - old problem , that of the self - energy of electrons due to electrostatic interactions . so far , all local electron models could not resolve the problem , why electrons would retain their size . here , we found that the effective potential @xmath231 for single electrons . this means , that no interactions exist between different regions of one extended electron . in principle , this allows for any size of electrons . from a purely mathematical perspective , and in view of direct methods in density functional theory , the proof of a successful application to general condensed matter problems is still missing . given that the restrictions this relation imposes on the charge density distribution are much more stringent than in current models , and that any direct method is potentially much more efficient than a method based on solutions of the kohn - sham equations , the results seem promising . progress , in this respect , will depend on the development of transferable and precise parameterizations of the general exchange - correlation potential , or the vector field @xmath0 . a task , which has yet to be accomplished . from a physical and more fundamental perspective , the emerging picture is quite intriguing . a quantum mechanical system , which finds its groundstate by equilibrating energy density throughout , is in principle not very different from a thermodynamic system of molecules , which accomplishes exactly the same on the level of molecular motion . there is thus no fundamental difference , if one goes from the analysis of isolated molecules to the analysis of electron charge . furthermore , it is quite difficult to explain to a layman , why an electron , as a point particle , does not immediately attach itself to the nucleus of a hydrogen atom . the present framework removes this problem , as well as the seeming contradiction , that extended electrons , with their inherent coulomb repulsion , are nevertheless stable . however , here we should add a note of caution . even though the results obtained so far look very promising , the success of the framework will ultimately depend on a very pragmatic fact : whether it can be used to make accurate predictions of solid state properties , and whether these predictions are in line with experimental data . wah is indebted to george darling , jacob gavartin , andres arnau and pedro echenique for stimulating discussions , criticism , and support , which to a large extent initiated or focussed research in this particular direction . he is supported by a royal society university research fellowship . kp is funded by the european commission under project nmp3-ct-2004 - 001561 . 99 s. hohenberg and w. kohn , phys . 136 , b864 ( 1964 ) w. kohn and l. j. sham , phys . 140 , a1133 ( 1965 ) r. car and m. parinello , phys . 55 , 2471 ( 1985 ) m. levy , j.p . perdew , and v. sahni , phys . a 30 , 2745 ( 1984 ) j. e. harriman , phys . a 24 , 680 ( 1981 ) s. k. ghosh and r. g. parr , j. chem . 82 , 3307 ( 1985 ) n. h. march , phys . 113a , 476 ( 1986 ) m. levy and h. ou - yang , phys . a 38 , 625 ( 1988 ) z. yang , s. liu , and y. a. wang , chem . 258 , 30 ( 1996 ) e. sim , j. larkin , and k. burke , j. chem . 118 , 8140 ( 2003 ) j. a. alonso and l. a. girifalco , phys . b 17 , 3735 ( 1978 ) e. chacon , j. e. alvarellos , and p. tarazona , phys . b 32 , 7868 ( 1985 ) p. garcia - gonzalez , j. e. alvarellos , and e. chacon , phys . a 54 , 1897 ( 1996 ) y. a. wang and e. a. carter , chapter 5 of `` theoretical methods in condensed phase chemistry '' in the series of `` progress in theoretical chemistry and physics '' , edited by s. d. schwartz , pp . 117 - 184 , kluwer , dordrecht ( 2000 ) y. aharonov and d. bohm , phys . 115 , 485 ( 1959 ) c. f. von weizscker , z. phys . 96 , 431 ( 1935 ) j. p. perdew , r. g. parr , m. levy , and j. l. balduz , phys . 49 , 1691 ( 1982 ) e. p. wigner , phys . 46 , 1002 ( 1934 ) l. wang and m. p. teter , phys . b 45 , 13196 ( 1992 ) r. webb , s. washburn , c. umbach , and r. laibowitz , phys . rev . 54 , 2696 ( 1985 ) b. schwarzschild , phys . today 39 , 17 ( 1986 ) g. vignale and w. kohn , phys 77 , 2037 ( 1996 ) w. a. hofer , physica a 256 , 178 ( 1998 )

we inquire how structure emerges during the process of protein folding . for this we scrutinise collective many - atom motions during all - atom molecular dynamics simulations . we introduce , develop and employ various topological techniques , in combination with analytic tools that we deduce from the concept of integrable models and structure of discrete nonlinear schrdinger equation . the example we consider is an-helical subunit of the hiv envelope glycoprotein gp41 . the helical structure is stable when the subunit is part of the biological oligomer . but in isolation the helix becomes unstable , and the monomer starts deforming . we follow the process computationally . we interpret the evolving structure both in terms of a backbone based heisenberg spin chain and in terms of a side chain based xy spin chain . we find that in both cases the formation of protein super - secondary structure is akin the formation of a topological bloch domain wall along a spin chain . during the process we identify three individual bloch walls and we show that each of them can be modelled with a very high precision in terms of a soliton solution to a discrete nonlinear schrdinger equation .

1511.07313

a domain wall is a prototype collective excitation in a physical system , and it is also the paradigm example of a topological soliton @xcite . a domain wall can appear whenever there is a global symmetry that becomes spontaneously broken . it constitutes the boundary that separates two neighbouring domains , in which the order parameter that detects the symmetry breaking has different values . in the case of a one dimensional heisenberg spin chain the order parameter is a three component unit length vector . when one of the three vector components vanishes identically , the heisenberg spin chain reduces to the xy spin chain @xcite . a domain wall along the spin chain is a localised excitation that interpolates between two different , ordered spin states in which the order parameter has different constant values . two major types of domain walls are commonly identified along the heisenberg chain @xcite . these are called the bloch wall and the nel wall , respectively . in the case of a bloch wall , the heisenberg spin variable rotates through the plane of the wall and in the case of a nel wall the rotation takes place within the plane of the wall itself . domain walls that are mixtures of these two , can also occur along the heisenberg spin chain , while along the xy spin chain , only domain walls of the bloch type can be present . in this article we demonstrate that the formation of super - secondary structures , during folding of a protein @xcite , can be understood in terms of a bloch domain wall that forms along a heisenberg spin chain , or along a closely related xy spin chain . we propose that the spin chain interpretation of a protein backbone provides a systematic framework for understanding and describing the process of protein folding . for this we employ all - atom force fields @xcite to scrutinise protein folding dynamics at the level of individual atoms and their oscillations . we analyse the folding pathway using a combination of topological techniques and global analytic tools . we isolate the collective oscillations which are pertinent for the folding process , from the noisy background of thermal and random individual atom fluctuations . in particular , we illustrate how the individual atom motions become organised and combined into a coherent structural excitation which we identify as the bloch wall . as a concrete example we consider an @xmath0-helical subunit of the hiv envelope glycoprotein gp41 @xcite , with protein data bank @xcite ( pdb ) code 1aik . there are six @xmath0-helical subunits in the biological assembly , shown in figure 1 . we consider in isolation the subunit , for which the first amino acid has number 628 in the pdb file . in isolation , the subunit is unstable and starts folding . the transmembrane glycoprotein 41 is itself a subunit of the retrovirus envelope protein complex . in the case of the hiv , its structure has been studied extensively . it is presumed to have substantial biological relevance to the initial viral infection . accordingly , the gp41 protein is a popular target for the development of an anti - viral immune response , to prevent and cure hiv infection . however , medical applications are beyond the direct scope of the present study . here we shall solely address and identify the physical mechanism , why and how an individual , initially @xmath0-helical subunit of 1aik becomes unstable in isolation , and starts folding . for our all - atom molecular dynamics simulations , we utilise the gromacs 4.6.3 package @xcite . we analyse the results using a variety of topological techniques and analytical tools . our approach derives from the mathematical structure of heisenberg and xy spin chains , in combination with properties of a discrete nonlinear schrdinger ( dnls ) hamiltonian @xcite . in particular , the dnls equation that describes the local extrema of the hamiltonian , enables us to analytically identify the profile of the domain wall , and to interpret it in terms of dnls soliton @xcite . here we present results from the detailed investigation of a particular example . however , we expect our observations and conclusions to be generic . indeed , the present results are fully in line with the previous findings @xcite obtained by using the coarse grained unres energy function @xcite in the case of protein a. the similitude of results that are obtained by analysing the protein folding process using different tools , built and based on phenomena with very different characteristic time and length scales , demonstrates that we have correctly identified the relevant collective motions that command the folding process . we have performed _ in silico _ experiments to fold one c - chain subunit of the core structure of gp41 @xcite . the structure comes from the hiv envelope glycoprotein with pdb code 1aik . the amino acid sequence is @xmath1 these amino acids are assigned the numbers 628 - 661 in the pdb entry of 1aik . we have used the molecular dynamics package gromacs 4.6.3 @xcite . we have analysed in detail a number of 80 ns long trajectories , with the crystallographic pdb conformation as the initial condition . we have chosen the length of the trajectories by inspecting , when major structural deformations take place . we have employed three different force fields , to eliminate force - field based artifacts . these are the united - atom force field gromos53a6 , and the all - atom force fields charmm27 and opls / aa . the 1aik subchain that we have investigated in detail , consists of 34 amino - acid residues , with pdb numbers 628 - 661 . there are 16200 atoms in the entire system that we have simulated , including the solvent . the simulation box has dimensions @xmath2 @xmath3 . this ensures that there is a 2 nm minimal distance between the protein atoms and the box walls , with periodic boundary conditions . we have described the solvent using the spc water model @xcite . we have neutralised the system at a salt concentration of 0.15 we have used steepest - descent for initial energy minimisation . the system was warmed up to 290 k by a simulated annealing in a 100 ps position - restraint simulation . we have chosen this relatively low temperature value for a better control of random thermal noise but without forgoing the underlying physical phenomena . for temperature control we have employed the berendsen - thermostat with a time constant 0.1 ps , and for pressure coupling the berendsen - barostat with a pressure set to 1 bar and a time constant 0.5 ps . constraints on all bonds were imposed with the lincs algorithm @xcite . we have used the particle mesh ewald ( pme ) method @xcite to compute the long - range electrostatic interactions , with van der waals and coulomb cutoff radii of 0.9 nm . for the 80 ns production run with a time step of 2 fs , that we analyse here in detail , we have changed the thermostat to v - rescale and the barostat to parrinello - rahman , keeping the initial time constants , to ensure the generation of a proper canonical ensemble @xcite . we have recorded the coordinates every 20 ps , which gives rise to 4000 frames that form the basis for our analysis . we have introduced , employed and developed a number of topological tools and analytic techniques to analyse and interpret the results of our gromacs simulations . we monitor the evolution of the protein geometry using frenet frames which are based on the backbone c@xmath0 atoms @xcite . the framing depends _ only _ on the c@xmath0 atom coordinates @xmath4 , where @xmath5 labels the residues and @xmath6 in the case of 1aik . at a given @xmath4 the frame consists of the unit backbone tangent ( @xmath7 ) , binormal ( @xmath8 ) and normal ( @xmath9 ) vectors , defined as follows , @xmath10 @xmath11 @xmath12 our aim is to identify and isolate the collective multi - atom motions that drive the protein folding process , from the background of the various random fluctuations . we expect that such coherent motions and oscillations have characteristic time scales , which are much longer than the period of an individual atom covalent bond oscillation . in average , over the relevant time scales , the distance between two consecutive c@xmath0 atoms can then be taken to be nearly constant , and equal to @xmath13 thus , at relevant time scales , the backbone dynamics can be described entirely in terms of the virtual backbone bond and torsion angles @xmath14 and @xmath15 , as the _ complete structural order parameters _ these angles are defined as follows , @xmath16 where @xmath17 { \label}{eq : s}\ ] ] conversely , the frame vectors ( [ eq : t])-([eq : n ] ) can be expressed in terms of these two order parameters iteratively , using the _ discrete frenet equation _ @xcite @xmath18 and the c@xmath0 backbone is calculated from @xmath19 unlike the tangent vector @xmath7 , the normal and binormal vectors @xmath9 and @xmath20 do not appear in equation ( [ eq : dffe ] ) . thus , if we rotate these two vectors simultaneously around the vector @xmath7 , the c@xmath0 geometry remains intact and only the way how it is framed changes . in particular , rotation by @xmath21 constitutes the discrete @xmath22 gauge transformation @xmath23 this transformation has been previously used extensively , to analyse protein loop structure @xcite . it will also be used in the sequel . according to ( [ eq : dffe ] ) the entire c@xmath0 backbone geometry is determined by the tangent vectors @xmath24 . thus , following @xcite we may visualise the backbone geometry in terms of these vectors : we take the base of @xmath7 to be at the location @xmath25 of the @xmath26 c@xmath0 atom . we identify the tip of @xmath7 as a point on the surface of a unit two - sphere @xmath27 that is centered at the point @xmath25 . we orient the coordinate system on the sphere so that the north - pole coincides with the tip of @xmath7 . thus , the north - pole is always in the direction of the next c@xmath0 , which is at the site @xmath28 . we proceed to characterise the direction of the next tangent vector @xmath29 _ i.e. _ the direction from @xmath28 towards the c@xmath0 atom at site @xmath30 , in terms of the longitude and latitude angles of the @xmath26 two - sphere @xmath31 . for this , we translate the center of @xmath27 from @xmath32 towards its north - pole , and all the way to the location @xmath28 of the @xmath33 c@xmath0 atom , without introducing any rotation of the sphere . we then record the direction of @xmath29 as a point on the surface of the translated @xmath27 . this defines the coordinate values ( @xmath34 ) , that determine how the backbone chain turns at site @xmath28 , to reach the @xmath35 central c@xmath0 atom at the point @xmath30 : the angle @xmath14 measures the latitude of @xmath29 on the translated two - sphere @xmath27 , from its north pole . the angle @xmath15 measures the longitude of @xmath29 , starting from the great circle that passes both through the north pole and through the tip of the binormal vector @xmath8 . when we repeat the above procedure for all c@xmath0 atoms , we obtain a ( @xmath36 ) distribution that characterises the overall geometry of a protein backbone . for a visualisation of this distribution we employ the geometry of a stereographically projected two - sphere : we project the @xmath37 coordinates from the south pole to the tangent plane of the north pole , of the two - sphere . if ( @xmath38 ) are the coordinates of this tangent plane the projection is defined by @xmath39 when we perform the projection for all c@xmath0 atoms in all crystallographic protein structures in pdb that have been measured with resolution better than 2.0 , we arrive at the statistical angular distribution that we show in figure [ fig-2 ] . it is the landscape for the shape of the protein backbones from the crystallographic data in pdb . by the way it is obtained , the crystallographic protein structure should be very close to a stationary minimum of the ensuing gibbs free energy . thus the figure [ fig-2 ] should be the collective landscape of stationary , minimum - energy protein structures . we observe that the pdb data is concentrated in an annulus which is roughly between the circles @xmath40 and @xmath41 . the exterior of the annulus @xmath42 is an excluded region , the ensuing conformations are subject to steric clashes . the interior @xmath43 is sterically allowed but in practice excluded in pdb structures . note that regular structures such as @xmath0-helices and @xmath44-strands are distinguished as highly localised regions in the figure [ fig-2 ] , with @xmath45 for @xmath0-helices and @xmath46 for @xmath44-strands . different regions in figure [ fig-2 ] can be connected by loops , which can be considered as trajectories along the variables ( @xmath47 ) . we have found that loops have the tendency to encircle the inner circle . in figure [ fig-3 ] a ) we show , as an example , a generic loop that connects the right - handed @xmath0-helical region with the @xmath44-stranded region . to describe a backbone segment analytically , we combine its c@xmath0 s bond and torsion angles into the three component unit vectors @xmath48 we interpret these vectors as the local order parameters along an imaginary _ linear _ one dimensional heisenberg spin chain , labeled by the index @xmath49 . this converts the c@xmath0 geometry into a configuration along a linear heisenberg chain in a one - to - one manner : in figure [ fig-3](b ) we have sketched how the ( generic ) trajectory shown in figure [ fig-3](a ) appears , figuratively , in terms of such a heisenberg spin chain configuration . since the spin variable ( [ si ] ) takes values only in the annulus @xmath50 of the two - sphere @xmath51 , it is apparent that a loop can be _ de facto _ identified as a domain wall akin the bloch wall along a heisenberg chain . the loop then interpolates between the two different regular secondary structures , denoted by state a and state b respectively , as shown in the figures 3 . the amino acid side chains can be similarly interpreted in terms of a one dimensional linear spin chain . in fact , there are several ways to identify the spin chain variable . here we utilise the directional vector that points from the c@xmath0 atom at @xmath32 towards the ensuing c@xmath44 atom , located at @xmath52 . this vector can be introduced for all amino acids except glycine ( g ) ; note that there is no glycine in ( [ lets ] ) . we start with the unit vector @xmath53 we recall the c@xmath0 based discrete frenet framing with the coordinates ( @xmath34 ) and represent ( [ sideo3 ] ) as three component unit vectors in this coordinate system , @xmath54 here ( @xmath55 ) are the spherical coordinates of the @xmath26 c@xmath44 atom , on the surface of the c@xmath0 centered two - sphere @xmath27 . in figure [ fig-4 ] we show the distribution of the vectors ( [ cbetai ] ) on the surface of the two - sphere , for all those crystallographic pdb structures that have been measured with resolution better than 1.0 . note that the sphere is the same as in figures [ fig-2 ] and [ fig-3](a ) but now there is no stereographic projection . we can interpret the distribution in figure [ fig-4 ] as the c@xmath44 landscape of stationary folded protein structures with minimum gibbs energy . the highly localized character of the distribution shows that there is a very strong correlation between the c@xmath0 ( backbone ) geometry and the c@xmath44 ( side chain ) geometry . accordingly , the ground state structures of the corresponding heisenberg spin chain hamiltonians must be very similar . we proceed to introduce a set of o(2 ) spin variables for the side - chain c@xmath44 . for this we define the projection of ( [ sideo3 ] ) onto the normal plane at the position of the @xmath26 c@xmath0 , @xmath56 for the next c@xmath44 along the chain , we introduce similarly the vector @xmath57 and compute its projection onto the _ same _ normal plane at the position of the @xmath26 c@xmath0 , @xmath58 we then define the relative angle @xmath59 , @xmath60 as shown in figure [ fig-5 ] , @xmath59 is the dihedral angle @xmath61 we note that the construction resembles that of newman projection in stereochemistry . in analogy with figure [ fig-3](b ) we identify the variable @xmath59 as an order parameter for a linear o(2 ) xy spin chain , @xmath62 like the heisenberg model , the xy model supports domain walls that interpolate between two configurations where the order parameter has different constant values . the domain wall of the xy model is akin the bloch domain wall of a heisenberg model . figure [ fig-4 ] shows that the ground state structure of the side chain xy model is closely related to that of the backbone heisenberg model , in the case of crystallographic pdb protein structures . the formation , evolution and structure of a loop along a folding protein can be monitored in terms of topologically determined folding indices . here we are interested in two particular examples of folding indices , one that relates to the backbone geometry and another one that relates to the side chain geometry . in the case of a heisenberg spin chain , there is a topological index akin a winding number that characterises and classifies its bloch domain walls . for the c@xmath0 bloch wall shown in figure [ fig-3 ] a ) , b ) this topological index counts the net number of times the corresponding trajectory encircles around the annulus in the figure _ i.e. _ around the north - pole of the two - sphere . we remind that due to steric constraints , the heisenberg variable ( [ si ] ) takes values in the annulus shown in figures [ fig-2 ] , [ fig-3 ] a ) . we also recall that for the first homotopy class of a circle @xmath63 which justifies the introduction of a topological concept . analytically , we may assign to each loop , and more generally to a backbone segment , between residues @xmath64 and @xmath65 the following folding index @xmath66 @xcite , @xmath67 { \label}{eq : gamma}\ ] ] where @xmath68 here @xmath69 $ ] denotes the integer part of @xmath70 . note that @xmath71 is the total rotation angle ( in radians ) that the projections of the c@xmath0 atoms of the consecutive loop ( segment ) residues make around the north pole . the @xmath72 label the first and last residue of the loop . commonly these are the last _ resp . _ first residues in the preceding and in the following regular secondary structures . the folding index is a positive integer when the rotation is counterclockwise , and a negative integer when the rotation is clockwise . the folding index can be used to classify individual loop structures and backbone segments , even entire protein backbones @xcite . note that the folding index is normalised so that it is equal to twice the number of times the vector in figure [ fig-3](b ) rotates around its axis , when the spin structure traverses a domain wall _ i.e. _ it assigns an even integer to the @xmath63 winding number , in the case of a closed trajectory . for example , for the trajectory shown in figures [ fig-3 ] the folding index has the value -1 . for a loop connecting an @xmath0-helix and a @xmath44-strand , the folding index is generically an odd integer . for a loop connecting two @xmath0-helices , or two @xmath44-strands , the folding index is generically an even integer . in the case of the side chains , we utilise the c@xmath44 angular xy spin variable ( [ etai ] ) to define a similar topological folding index @xcite . for this , we first choose a reference residue , _ e.g. _ the @xmath73 residue along the backbone . starting with this reference residue , we then evaluate the accumulated _ total _ angle @xmath74 over a segment with @xmath75 residues , @xmath76 and we define the ensuing index by @xmath77 { \label}{eq : etaprime2}\ ] ] again , the index acquires its topological justification from the fact that @xmath78 . the dihedral @xmath79 , the accumulated total angle ( [ eq : etaprime ] ) and the ensuing index ( [ eq : etaprime2 ] ) can all be used to study and classify loop structures , protein segments , and entire proteins . sect : landau a generic all - atom molecular dynamics simulation of a folding protein contains a wide range of intermediate conformations . some of these are essential for the correct folding pathway , while some are merely random transients with no inherent relevance to the folding process _ per se_. in order to identify the relevant conformational processes , we need systematic methods that smooth over and weave out the irrelevant random fluctuations . for this we recall the standard wilsonian universality arguments @xcite , to deduce the form of the landau free energy that emerges from the thermodynamical gibbs free energy . in the present context , the derivation is based on the following two assumptions @xcite : @xmath80 we assume that the characteristic length scales that are associated with spatial variations and deformations along the protein backbone around its thermal equilibrium configuration , are large in comparison to the covalent bond lengths . this presumes that there are no abrupt edges but only gradual slowly varying bends and twists along the backbone . from figures [ fig-2 ] and [ fig-4 ] we conclude that the steric constraints between the backbone and the side chain atoms act as powerful inhibitors of sharp , edgy motions . @xmath80 we also assume that the individual c@xmath0 virtual bond length oscillations have characteristic time scales , which are very short in comparison to the time scale which characterise a folding process . the characteristic time scale of a random covalent bond oscillation is around ten femtoseconds while in our simulations we record the individual atomic coordinates every 20 picosecond . we have tested that conformational changes which take place over shorter time scales , do not affect our conclusions . accordingly , we may adopt ( [ eq : aveva ] ) as the ( time averaged ) value for all the nearest neighbour c@xmath0-c@xmath0 distances . it has been shown that in the case of crystallographic pdb structures , the bond and torsion angles ( @xmath34 ) form a _ complete set of structural order parameters _ accordingly , in the vicinity of a gibbs free energy minimum we may expand the free energy in terms of these angles . for this we consider the response of the interatomic distances to variations in these angles , with @xmath81 where @xmath82 is the distance between _ any _ two c@xmath0 atoms @xmath83 and @xmath84 along the backbone . suppose that at a given local extremum of the free energy , the c@xmath0 bond and torsion angles have the equilibrium values @xmath85 we then consider a non - equilibrium conformation where the ( @xmath34 ) deviate from these extremum values . we denote the deviations by @xmath86 when the deviations are slowly varying in space , _ i.e. _ ( [ devi ] ) are small , we may taylor expand the gibbs free energy around the extremum , @xmath87 \ = \ ] ] @xmath88 @xmath89 @xmath90 the first term in the expansion evaluates the free energy at the extremum . since ( @xmath91 ) correspond to the extremum , the second term vanishes . thus we are left with the following expansion of the averaged free energy , @xmath92 @xmath93 @xmath94 when the characteristic length scale of spatial deformations around a minimum energy configuration is large in comparison to a covalent bond length , we may _ re - arrange _ the expansion ( [ fene ] ) in terms of the differences in the angles as follows : first come local terms , then come terms that connect the nearest neighbours , then come terms that connect the next - to - nearest neighbours , and so forth . when we re - order the expansion ( [ fene ] ) in this manner and demand that the free energy is invariant under local rotations in the ( @xmath95)-plane , we conclude @xcite that to the leading order the expansion of the gibbs free energy _ necessarily _ coincides with the energy of the following discrete nonlinear schrdinger equation @xcite @xmath96 @xmath97 @xmath98 + \sum\limits_{i=1}^{n-1 } ( \kappa_{i+1 } - \kappa_{i})^2 { \label}{e2old}\ ] ] this functional form of the free energy is simply the most general landau free energy that one can write down using the available variables ( @xmath99 ) , in a manner which is consistent with the symmetry principle that a local rotation of the ( @xmath100 ) frames has no effect on the backbone geometry . the corrections to ( [ e1old ] ) include next - to - nearest neighbours couplings and so forth , which are higher order terms from the point of view of our systematic expansion . we note that the expansion ( [ e1old ] ) has the property that in continuum limit it yields the coleman - weinberg derivative ( low momentum ) expansion @xcite @xmath101 where , following @xcite , we have identified the bond angle with the complex scalar field @xmath102 and the torsion angle with the u(1 ) gauge field @xmath103 , in the continuum limit . we _ emphasize _ that the approximation ( [ e1old ] ) is valid in the limit of slow spatial variations ( low momentum ) . that is , as long as there are no abrupt , sharp edges but only gradual bends and twists along the backbone . in particular , long range interactions are accounted for as long as they do not cause any localised sharp buckles along the backbone , and the angular variations respect the steric constraints . the wilsonian universality arguments are sufficient to conclude that in the limit of slowly varying backbone geometry _ any _ complete all - atom force field can be approximated by the energy function ( [ e1old ] ) . the parameters @xmath104 , @xmath105 , @xmath106 , @xmath107 , and @xmath108 depend on the atomic level physical properties and the chemical microstructure of the protein and its environment . in principle , these parameters can be computed from this knowledge . but as always in the case of a landau free energy , it remains a challenge to compute these parameters from the all - atom level . sect : solitons the free energy ( [ e1old ] ) relates to the dnls energy function @xcite . the non - linear , quartic bond angle contribution is the familiar double - well potential that gives rise to a spontaneous breakdown of the @xmath22 symmetry @xmath109 the spontaneous breakdown of this discrete symmetry is pivotal for the emergence of a loop structure , in the case of proteins . it gives rise to a bloch wave that interpolates between the two ground states @xmath110 . more generally , the quartic potential admits a non - symmetric profile of the form @xmath111 here @xmath83 and @xmath84 are the positions of the minima of the quartic potential , and @xmath112 is a force constant . by carefully taking the continuum limit of the c@xmath0 lattice , _ i.e. _ the limit where ( [ eq : aveva ] ) becomes small , and by introducing a mass - scaled variable @xmath113 with @xmath108 representing the effective mass of a residue , the pertinent dnls equation becomes @xmath114 where @xmath115 is the arc length parameter along the backbone . with @xmath116 and @xmath117 the solution of equation ( [ eq : schr ] ) is @xmath118 { \label}{kappaprof}\ ] ] where @xmath119 is the position of the inflection point , _ a.k.a . _ the center of a kink . in terms of the original variables and parameters @xmath120 } { \label}{eq : cosh}\ ] ] this is known as the dark soliton solution of the nonlinear schrdinger equation . it interpolates between the asymptotic values which correspond to the ( local ) minima of the potential , @xmath121 in the case of a protein , the soliton describes the bond angle profile of a super - secondary structure such as ( @xmath0-helix)-(loop)-(@xmath44-strand ) shown in figure [ fig-3 ] ; the parameters have the values @xmath122 and @xmath123 ( radians ) for the states @xmath124 and @xmath125 , shown in the figure . in the case of a protein chain , the arc length @xmath115 becomes replaced by a discrete variable which is equal to the position of the ensuing c@xmath0 in the sequence . the variables @xmath14 and @xmath15 are also mutually interacting , according to ( [ e1old ] ) . the soliton is constructed as the minimum of @xmath126 in equation ( [ e1old ] ) @xcite . it is the solution of a system of @xmath127 nonlinear equations in @xmath127 unknowns , where @xmath128 is the number of residues . in order to obtain the solution , we first solve for @xmath15 in terms of @xmath14 , @xmath129 = \frac{p}{r + q\,\kappa^2_{i } } \ \frac{u } { 1 + v \ , \kappa^2_{i } } { \label}{tauk}\ ] ] with @xmath130 and @xmath131 . by inserting equation ( [ tauk ] ) into equation ( [ e1old ] ) , the torsion angles @xmath132 are eliminated and we obtain a system of equations for the bond angles @xmath133 , @xmath134}{d\kappa_i^2 } \kappa_i \ \ \ \ \ ( i=1, ... ,n ) { \label}{nlse}\ ] ] where @xmath135 and @xmath136 = \frac{p}{r + q \ , \kappa^2 } + 2 ( 1- \lambda m^2 ) \kappa^2 + \lambda \ , \kappa^4 { \label}{u}\ ] ] here we recognize the discretised structure of equation ( [ eq : schr ] ) . the difference is in the first term on the right - hand side in equation ( [ u ] ) . however , it turns out that in the case of proteins , its effect is not that pronounced as the effect of the other terms ; it turns out that the first term is small in value when compared to the other two . we can construct the profile of the dark soliton solution to equation ( [ nlse ] ) numerically , by following the iterative procedure introduced in reference @xcite ; the explicit form of the solution is until now unknown , in terms of elementary functions . however , we obtain an _ excellent _ approximation @xcite by _ naively _ discretising the continuum dark nonlinear schrdinger equation soliton ( [ eq : cosh ] ) @xmath137 + \mu_{2 } \ , \exp\left [ - \sigma_{2 } ( i - s)\right ] } { \exp\left [ \sigma_{1 } ( i - s ) \right ] + \exp\left[- \sigma_{2 } ( i - s)\right ] } { \label}{eq : bond2}\ ] ] here @xmath138 \ { \rm mod}(2\pi)$ ] are parameters , which determine the amplitude of the variation of @xmath133 and the asymmetry of the inflection regions . the parameters @xmath139 and @xmath140 are related to the inverse of the range of the inflection region . we remark that in the case of proteins , the values of @xmath141 are determined entirely by the adjacent helices and strands . furthermore , far away from the soliton center we have in analogy with ( [ asymp ] ) @xmath142 the corresponding torsion angles are evaluated in terms of the bond angles using equation ( [ tauk ] ) . note that in the case of proteins , the profile of equation ( [ eq : bond2 ] ) becomes monotonically increasing when we add multiples of @xmath143 to the experimental values . since the values of @xmath144 s are defined @xmath145 , this does not affect the backbone geometry . the integer number of times the monotonically increasing variable @xmath14 covers its fundamental domain @xmath146 counts the number of solitons along the backbone . recall that negative values of @xmath14 are related to positive values of @xmath14 by @xmath22 symmetry ( [ eq : dsgau ] ) . finally , _ only _ the parameters @xmath139 and @xmath140 in ( [ eq : bond2 ] ) are intrinsically specific parameters for a given loop . but they specify only the length of the loop , not its shape which is determined _ entirely _ by the functional form of equation ( [ eq : bond2 ] ) and , as in the case of @xmath147 and @xmath148 , they are combinations of the parameters in equation ( [ u ] ) . in the expression ( [ tauk ] ) of the torsion angles @xmath149 , there are only two independent parameters @xmath150 and @xmath151 . consequently the profile of @xmath15 is determined entirely by @xmath14 , and by the structure of the adjacent regular secondary structures . it has been shown @xcite that most crystallographic protein structures in pdb can be described with very high precision in terms of such soliton profiles as their modular building blocks . moreover , it has been found that in the ensuing soliton profiles , the number of parameters is generically much smaller than the number of residues . thus , the energy function ( [ e1old ] ) has a very high predictive power , in describing folded proteins structures in pdb . its predictions can be subjected to stringent experimental scrutiny , both in the case of static and dynamic proteins . we now proceed to demonstrate , that all the concepts and structures we have identified are observed during an all - atom simulation of protein folding . we start with individual atom level scrutiny , even though our goal is to identify and model those _ collective _ conformational deformations that cause a protein to fold . we inquire how does self - organisation , in the case of a protein , relate to universal concepts such as formation of domain walls along spin chains . we study how accurately can the dynamics and structure of the important collective deformations be modelled by soliton profiles such as the one described by the dnls equation . we have subjected the single c - chain subunit of the core structure of gp41 @xcite with pdb code 1aik and residues 628 - 661 , to detailed all - atom simulations . we have used the gromacs 4.6.3 . package @xcite with three different force fields gromos53a6 , charmm27 and opls / aa having thus the protein described by 383 , 573 and 609 atoms , respectively . the final production simulation models 80 ns of the protein evolution . we have concluded this to be sufficient , to identify and analyse the important structural deformations that can take place . in figure [ fig-6](a ) we show the secondary structure of the final conformations that we have obtained using the three force fields , in our 80 ns simulations . we observe a clear deformation , in a segment that consists of the first 10 residues from the n - terminal ( upper part in the figure ) . in figure [ fig-7 ] we display the weighted root - mean - square - deviation ( rmsd ) of the protein backbone in the three force fields , @xmath152^{1/2 } { \label}{rmsd}\ ] ] here @xmath153 is the position of the atom @xmath49 at time @xmath154 , and @xmath155 where @xmath156 are the individual atom masses , and @xmath157 is the total mass of the backbone . the deformation is most intense in the gromos53a6 force field . with this force field , the initial @xmath0-helical structure begins collapsing within 45 ns . with the opls / aa force field , we find that the deformation starts after around 40 ns . in the case of the charmm27 force field , the helix tends to remain intact within the selected time range . the deformation begins only after a substantially longer simulation . apparently , this force field has a tendency to produce structures that have an overly @xmath0-helical content . after extended comparisons of the three force fields , including different time steps and simulation lengths , we have chosen a 80 ns gromos53a6 force field trajectory with 2 fs time step , for the final production simulation that we analyse here . qualitatively , the results that we present are independent of the force field and time step that we have chosen . in figure [ fig-8 ] we show the results from a do.dssp @xcite secondary structure analysis , in the case of the gromos53a6 simulation . the following _ qualitative _ observations can be made : @xmath80 after around 4 - 5 ns corresponding to frames 200 - 250 , there is an initial formation of a coil structure , according to do.dssp classification . the coil becomes initially stabilised around residue number 29 which corresponds to amino acid number 656 in the pdb file . the coil is connected to the c - terminal with a bend . at around 8 ns ( around frame 400 ) there are helical fluctuations in this structure towards n - terminal , and at around 22 - 24 ns ( frames 1100 - 1200 ) the coil structure moves back towards the c - terminal . the motion takes place in two steps , at around frame 1200 and then again at around frame 2800 after which the coil disappears , by merging into the apparently random fluctuations of the c - terminal residues . at the level of do.dssp secondary structure analysis the coil which emerges near the c - terminal and propagates along the backbone , is putatively akin the propagating loop structure that has been previously identified and studied in coarse grained unres simulations of the protein g related albumin - binding domain with pdb code 1gab @xcite . in particular the unres simulation @xcite identifies a displaced protein loop as a localised structure with a profile that can be described by the soliton solution of the discrete nonlinear schrdinger equation ( [ tauk ] ) , ( [ nlse ] ) . the simulation demonstrated that when the loop - soliton moves along the protein lattice , with cells matching the residues , there are waves that are emitted in its wake as vibrations in the lattice structure . these waves drain the kinetic energy of the soliton , and cause it to decelerate . eventually the kinetic energy of the soliton becomes depleted , and it can no longer cross over the energy barriers between lattice cells and becomes localised around a particular set of lattice cells . the energy barriers that prevent the soliton from translating along the backbone lattice were identified as peierls - nabarro barriers @xcite in @xcite . in the present case of the c - terminal coil , there is apparently a peierls - nabarro barrier that stops and prevents the coil that is supposedly modelled by a dnls , from propagating away from the c - terminal beyond the residues 28 - 29 . instead it becomes initially trapped , then moves towards the c - terminal and dissolves there . the soliton moves step - wise , it s crossing - over the ensuing peierls - nabarro barriers is boosted by thermal fluctuations . the soliton crosses a barrier whenever the amplitude of its thermal fluctuations exceeds the barrier - specific threshold value . @xmath80 at around the same time when the c - terminal coil forms , we observe a turn deformation that forms and proceeds away from the n - terminal , and then fluctuates thermally between residues 5 and 10 . after around 20 ns ( frame 1000 ) of simulation time , there is a rapid extended turn - like fluctuation that connects the n - terminal with a localised structure which is identified as a short coil by do.dssp . this is an apparent dnls soliton , emerging at the end of the extended turn - like structure and stabilising around the residues 6 - 8 ( residues 634 - 636 in the pdb file ) . we observe initially relatively strong fluctuations in the residues between the putative soliton and the n - terminal . but the amplitudes of these fluctuations become damped and after around 50 ns ( frame 2500 ) there are only minor fluctuations in the soliton . there is a bend between the soliton and the n - terminus which constitutes a peierls - nabarro barrier , high enough to prevent the soliton from moving towards the n - terminal , step - wise by thermal fluctuations . we proceed to analyse the dynamics _ quantitatively _ , and we start with the backbone folding index ( [ eq : gamma ] ) . for this we have divided the backbone into segments of varying length , and computed the folding index over the segments during the 80 ns time evolution . examples of results are shown in figure [ fig-9 ] , where we plot the numerical values of the folding index density ( [ eq : foldind ] ) . the first segment consists of the sites 4 - 11 corresponding to residues 632 - 639 in pdb . this segment covers the n - terminal soliton structure ( see figure [ fig-8 ] ) . the second segment consists of the sites 24 - 30 ( 652 - 658 in pdb ) . this covers the segment where the c - terminal coil initially appears in figure [ fig-8 ] . we observe the following : * figure [ fig-9](a ) : * initially , the folding index of the segment 4 - 11 vanishes . but in the vicinity of frame 1000 , coinciding with the formation of the n - terminal coil / soliton in figure [ fig-8 ] , the folding index starts fluctuating between the values @xmath158 and @xmath159 . we note how the pattern of the oscillations reflects the structural changes in the region between the coil and the n - terminal , shown in figure [ fig-8 ] : there are first fluctuations between turn and bend , during frames 1000 - 1500 . when the oscillations in the values of the folding index diminish and vanish near the frame 1500 , we observe a formation of helical structure between the coil and the n - terminal in figure [ fig-8 ] . the folding index then starts oscillating between the values @xmath160 , and this corresponds to a frame segment where the helix converts into a turn in the figure [ fig-8 ] . at around the frame 2500 , the folding index finally stabilises to the final value @xmath161 . this stabilisation concurs with the formation of a bend between the coil and the n - terminal , in figure [ fig-8 ] . * figure [ fig-9](b ) : * we observe the increase of folding index from @xmath162 to @xmath163 near frame 200 , and subsequent decrease back to @xmath162 near frame 1300 . according to figure [ fig-8 ] , these transitions coincide with the appearance of the c - terminal soliton , and its subsequent propagation towards the c - terminal , away from the segment 24 - 30 . accordingly , we have found that the variations in the values of the folding index , in particular in the case of the n - terminal soliton , coincide with the structural deformations that take place along the backbone segment which is located between the soliton and the n - terminal . in particular , the final stabilisation of the folding index concurs with the crossing over the peierls - nabarro barrier and subsequent stabilisation of the soliton , according to figure [ fig-8 ] . moreover , the c - terminal soliton emerges with @xmath163 and remains stable until the peierls - nabarro barrier crossing takes place . the evolution of the ensuing folding index is also fully in line with the results that we deduce from do.dssp . we conclude that the behaviour in the backbone segment , surrounding the soliton , directly correlates with the topological character of the soliton , in both cases . finally , in figure [ fig-10 ] we show the folding index density ( [ eq : foldind ] ) over the _ entire _ backbone and for all the 4000 frames . we observe that : @xmath80 initially , the folding index vanishes . this is consistent with the @xmath0-helical structure of the 1aik sub - chain . but there is a sudden initial transition to the value @xmath163 , presumably reflecting the initial stages of c - terminal soliton formation . @xmath80 up until the frame @xmath164 the folding index tends to vanish . but there are fluctuations , mainly between values @xmath165 which reflect the various processed that take place near the terminals . @xmath80 in the vicinity of frame 2700 there is a transition , and the value of the folding index starts stabilising toward the value @xmath163 . this stabilisation concurs with the stabilisation of the n - terminal soliton , and the final departure of the c - terminal soliton . the fluctuations also shift , oscillating between @xmath166 and @xmath167 and this shift is identified by the yellow arrows in the figure . in figure [ fig-11 ] we show a close - up to the last 50 frames in figure [ fig-10 ] . it confirms the stabilisation of the folding index towards the value @xmath163 , with occasional fluctuations where @xmath168 or @xmath169 . in figure [ fig-12 ] , following figure [ fig-3](a ) , we show the full trajectory for the entire final frame 4000 . the trajectory starts from the n - terminal which is located in the @xmath44-stranded region of figure [ fig-2 ] . it moves over to the @xmath0-helical region , then return to the @xmath44-stranded region to encircle the north pole . finally , the trajectory merges and ends with the @xmath0-helical region . the trajectory confirms that the final structure at frame 4000 indeed does support a twisting @xmath170 and that the twisting is furthermore located at the n - terminal soliton . the stabilisation of the folding index to the value @xmath163 confirms the global character of the remaining n - teminal soliton structure : there is a total twisting by @xmath170 along the final backbone , in comparison to the initial configuration and including the terminal residues , and this twisting is localised on the n - terminal soliton . moreover , we observe that the n - terminal residues are in a @xmath44-stranded position while the c - terminal residues are in the @xmath0-helical position . figure [ fig-4 ] shows the landscape of the ground state ( crystallographic structure ) c@xmath44 atom directions in the c@xmath0 centered discrete frenet frames . figure [ fig-13 ] shows how the directions of the c@xmath44 evolve during our entire gromos53a6 simulation . we find it remarkable how similar the dynamical landscape of figure [ fig-13 ] is with the static ground state landscape shown in figure [ fig-4 ] : the direction of c@xmath44 nutates _ tightly _ around its static ground state landscape . clearly , there must be strong correlations between the backbone c@xmath0 and side chain c@xmath44 , during the entire dynamics . accordingly , the information content in the angles @xmath59 in ( [ etai ] ) , ( [ etai2 ] ) should correlate strongly with the c@xmath0 geometry changes during the dynamical process . more generally , we expect that the various backbone and side chain spin models that we have introduced , are all in the same dynamical universality class . figure [ fig-14 ] shows the residue - wise accumulated distribution of all the individual angles @xmath59 in ( [ etai ] ) _ i.e. _ the ensuing landscape of the individual @xmath59 during the entire 80 ns gromos53a6 simulation . the conclusions that can be deduced from this figure are in line with those in figure [ fig-8 ] . in particular , we observe the presence of the n - terminal soliton , how it is centered around residues 5 - 6 . we also observe that the residues between sites 6 - 27 are in a helical position during the entire time evolution . we also observe the merging of the c - terminal soliton with the fluctuations of the c - terminal . figure [ fig-15 ] shows the time resolved landscape of all the @xmath59 angles . there is a remarkable similarity between this figure , and the figure obtained from the do.dssp backbone analysis shown in figure [ fig-8 ] . in particular , the formation and stabilisation of the n - terminal soliton around sites 5 - 6 is clearly visible . the appearance of the c - terminal soliton and its subsequent evolution is similarly visible : we observe how this soliton is formed at around frame 200 , and then propagates towards the c - terminal in a stepwise manner , crossing over the various peierls - nabarro barriers and eventually merging with the c - terminal thermal fluctuations . finally we note the apparent similarities between the structure of the landscape in figure [ fig-15 ] and the behaviour of the folding index density in figure [ fig-10 ] . we proceed to analyse the detailed properties of soliton structure and formation . our main focus will be on the n - terminal soliton structure . we are particularly interested in the phenomena that take place when the soliton is formed , _ i.e. _ the vicinity of the frame @xmath1711000 , and when the soliton moves over to a peierls - nabarro barrier and stabilises , _ i.e. _ the vicinity of the frame @xmath1712500 . we start with a closeup of the c - terminal part in figure [ fig-15 ] , shown in figure [ fig-16 ] . we observe how , in terms of the side chain @xmath59 angles , the soliton structure which forms with center at residue 28 subsequently propagates back - and forth , in a step - wise manner , towards the c - terminal . eventually it merges with the terminal , and dissolves into its fluctuations . the soliton motion is fully in line with figure [ fig-8 ] , and the initial motion is consistent with the folding index analysis in figure [ fig-9](b ) . in particular , the step - wise propagation of the soliton is fully consistent and in line with the presence of peierls - nabarro barriers . these barriers are high enough to trap the soliton momentarily , but low enough for the soliton to eventually cross over them when its thermal excitation energy fluctuates to high enough value . we remind that the present simulations have been performed at 290 k. we have chosen this relatively low temperature value , from the _ in vivo _ perspective , in order to restrain the soliton mobility and to dampen noisy thermal fluctuations . figure [ fig-17 ] shows a close - up of the n - terminal part in figure [ fig-15 ] , from two complementary perspectives . the soliton appears in the vicinity of frame 1000 . it subsequently translates one residue towards the n - terminal , in the vicinity of frame 2500 . this is an apparent crossing over a peierls - nabarro barrier by thermal fluctuation , and it is followed by a stabilisation of the soliton at the final position . note the correlation between the soliton motion and the extent of the n - terminal fluctuations . figure [ fig-18 ] shows the evolution of the individual @xmath59 angles in ( [ eq : etaprime ] ) , in the case of the n - terminal soliton structure . the panels display the deviation of @xmath59 from the initial average value for residues @xmath172 , which we have found to be those of primary interest . for @xmath173 and for @xmath174 and larger , the deviations from the initial average value fluctuate around zero . in figure [ fig-19 ] we scrutinise those segments of figure [ fig-18 ] , where the @xmath175 branch of the angle needs to be carefully resolved . in each of the panels in figure [ fig-19 ] we display the data in figure [ fig-18 ] , over a subset of frames and now evaluated around the value @xmath176 : in figure [ fig-19](a ) we show the segment 1500 - 2500 of figure [ fig-18](a ) , in figure [ fig-19](b ) we show the segment 2000 - 3000 of figure [ fig-18](b ) and in figure [ fig-19](c ) we show the segment 500 - 1500 of figure [ fig-18](c ) . by combining figures [ fig-18 ] and [ fig-19 ] we conclude that there are two major transitions , around frames 1000 and 2500 respectively . these transitions are concurrent with the major transitions in figures [ fig-8 ] , [ fig-15 ] , [ fig-17 ] and in particular figure [ fig-9 ] . the first transition corresponds to the creation of the n - terminal soliton , and the second one to its translation , by one site towards the n - terminal , and subsequent stabilisation . we also observe the presence of an extended transition process , visible in figure [ fig-18](b ) between frames 1100 - 1500 . in summary , we conclude from these figures that : @xmath80 in the vicinity of the frame 1000 , when the n - terminal soliton structure forms , there is an initial twisting of the @xmath177 dihedral which is close to @xmath178 and a twisting of the @xmath179 dihedral by an approximatively equal amount but in the opposite direction . thus , at this point the total twisting which is produced along the side chain segment vanishes . we note that the presence of two twists by an equal amount but opposite in direction , is the hallmark of a bloch domain wall pair production . but we recall that the backbone folding index detects only a single soliton , as shown in figure [ fig-9](a ) . @xmath80 after the initial bloch wall pair formation , the @xmath177 dihedral becomes slowly twisted back to the original value between frames 1200 - 1600 , so that after frame 1600 we are left with a total of @xmath180 twist . this process leaves us with a single soliton along the chain segment , with a total twisting around @xmath181 . @xmath80 finally , in the vicinity of frames 2100 - 2200 , we observe a rapid twisting of the @xmath182 dihedral by an amount close to @xmath183 . this is followed , in the vicinity of the frame 2400 - 2500 , by a twisting of the @xmath177 dihedral by an approximatively equal amount . there is an accompanying twist of the @xmath179 dihedral by an amount somewhat less than @xmath184 , over the same frames . these two twistings at @xmath185 accompany the peierls - nabarro barrier crossing , as can be deduced by comparison with figures [ fig-15 ] and [ fig-17 ] . @xmath80 when the soliton stabilises , after frame 3000 , we conclude that there is a total twisting in the side chain structure which is close to @xmath186 , and carried by the final soliton configuration . thus we assign to the final soliton the value @xmath187 of the index ( [ eq : etaprime2 ] ) . note that there is also certain ( small ) spillage of the @xmath188-values which is distributed among the nearby residues . in summary , the side chain analysis shows that at the level of the xy spin chain analysis , the n - terminal soliton structure forms by an initial rapid formation of a bloch domain wall _ i.e. _ soliton - antisoliton pair , followed by a slow twisting that apparently removes one of the two bloch wall solitons . this is then followed by a rapid transition , in combination with a peierls - nabarro barrier crossing , that forms the final stable soliton structure . accordingly , we may characterise the final soliton as a configuration with the c@xmath0 backbone folding index @xmath169 and the side chain xy folding index @xmath187 . figure [ fig-13 ] revels the presence of strong correlations between backbone and side chain dynamics . in particular , any formation of side chain soliton structure should correlate with the formation of corresponding backbone soliton . accordingly , we proceed to construct explicitly the backbone soliton that accompanies the n - terminal side chain soliton . we shall find that the backbone soliton can be modeled by a solution of the discrete nonlinear schrdinger equation ( [ tauk ] ) , ( [ nlse ] ) , with very high sub - atomic precision . as an example , we consider the profile of the n - terminal soliton structure at two different frames . other frames , and the c - terminal soliton structure , can be analysed similarly . we select the sites 4 - 11 ( pdb sites 632 - 639 ) for our analysis : the sites 0 - 3 are subject to fluctuations , and beyond the site 11 there is only a monotonic @xmath0-helix . we start with the frame 2000 , which is located in the regime where the side chain structure of the soliton has stabilised according to figure [ fig-18](b ) and the backbone folding index of the segment shown in ffigure [ fig-9](a ) has the value @xmath163 ; the side chain index over this segment is @xmath189 , according to figures [ fig-18 ] , [ fig-19 ] . in figure [ fig-20 ] we show the profile of the bond angle and the torsion angle over the segment 4 - 11 ( sites 632 - 639 in the pdb file ) in the frame 2000 , after we have implemented the @xmath22 gauge transformation ( [ eq : dsgau ] ) to identify the soliton profile . note that in order to compute a single bond angle , we need to know three residues while the evaluation of a torsion angle consumes four residues . thus , despite the smaller number of data points in the figure , the ensuing configuration engages 8 residues . we observe that the bond angle has the profile of a single domain wall soliton of the dnls equation , approximated by ( [ kappaprof ] ) . we use the software package propro that has been described at @xmath190 to numerically construct the ensuing soliton solution of the dnls equation . in figure [ fig-21 ] ( a ) we compare the profile of the bond and torsion angles in figure [ fig-20 ] with the profile of the soliton solution of ( [ tauk ] ) , ( [ nlse ] ) . in figure [ fig-22 ] ( top ) we compare the residue - wise distance between the all - atom configuration of frame 2000 , and the dnls solution . the average c@xmath0 rmsd between the two configurations is less than 0.1 ngstrm , and at no residue is the distance between the c@xmath0-atoms more than 0.2 ngstrm : the difference is truly negligible . the grey strip around the dnls soliton is a 0.2 ngstrm ( quantum mechanical ) fluctuation band @xcite . the figure [ fig-22 ] ( bottom ) shows the 3d overlay of the ensuing c@xmath0 backbones , for all practical purposes they are the same . in figure [ fig-23 ] we show the loop trajectories of the all - atom configuration of frame 2000 , and the corresponding dnls soliton , on the stereographically projected two - sphere of figure [ fig-2 ] and [ fig-3 ] . note that for both of these two loop trajectories the folding index , as defined in ( [ eq : gamma ] ) is vanishing , in that the trajectory does not encircle the north - pole ( center of the disk ) . a very short change either in the position of the residue labeled @xmath125 or in the position of the residue labeled c in figure [ fig-23 ] a ) , can shift the trajectory so that the line connecting them moves over to the other side of the north pole and the folding index becomes @xmath191 . this is consistent with the result shown in figure [ fig-9 ] that the folding index fluctuates between the values @xmath162 and @xmath191 , around the frame 2000 , with the posture shown in figure [ fig-23 ] being the more stable one . in figure [ fig-24 ] we show a close - up to the frame segment 2475 - 2525 around the soliton 2000 , in terms of the side chain angles @xmath188 . the close - up reveals the presence of fluctuations between the soliton and the n - terminal , while the helix between the soliton and the c - terminal displays very small fluctuations . thus , the fluctuations in the folding index around the frame 2000 are most likely due to shifts in the position of the residue labeled b in figure [ fig-23 ] ( a ) . figures [ fig-25]-[fig-28 ] show the same analyses for the conformation in frame 3500 . we find that the dnls soliton describes the domain wall soliton that we have constructed by all - atom simulations , with a very high sub - atomic precision . we note that the soliton in frame 3500 is a configuration that connects between the @xmath44-stranded region of figure [ fig-2 ] to the @xmath0-helical region , while in the case of the soliton at frame 2500 the initial residue is located in a sparsely populated region of the landscape in figure [ fig-2 ] . the major qualitative difference between frames 2000 and 3500 is between figures [ fig-23 ] and [ fig-28 ] . the soliton in frame 3500 is relatively stable . in particular , as shown in figure [ fig-9 ] , its folding index @xmath163 . we can understand the stability of the folding index by comparing figure [ fig-23 ] ( a ) with figure [ fig-28 ] ( a ) . in the later , the residues have assumed positions where the connecting arrows are stabilised against small perturbations , they are protected from crossing over the north pole in a manner that causes fluctuations in the value of the folding index . finally , we compare the structure of the solitons at frames 2500 and 3500 . from figures [ fig-20 ] , [ fig-25 ] we observe that in terms of the bond angles the soliton 3500 has indeed moved one site towards the n - terminal , from the position of soliton 2500 . in figure [ fig-29 ] we overlay their 3d structures . for this , we first translate the soliton in frame 3500 one residue away from the n - terminal , so that the two have the same location along the backbone . the figure shows the ensuing 3d interlaced c@xmath0 backbones , in a relative position where the rmsd is minimal . there is a visible difference , and the minimal rmsd is 2.0 ; the soliton has clearly become deformed . molecular dynamics enables the scrutiny of protein folding , at the level of individual atoms and over very short time intervals . however , it can leave us with the conceptual challenge to understand , how the individual atoms cooperate to produce the kind of large scale organisation that appears to be prevalent among crystallographic protein structures . we have performed detailed molecular dynamics simulations , with the aim to find out how organised structure emerges when a protein folds . we have first compared three different force fields using the gromacs 4.6.3 . package , to select the proper tools . we have chosen a c - chain subunit from hiv envelope glycoprotein with pdb code 1aik as a concrete example , partly due to its biomedical relevance even though this is an issue which has not been addressed by us . we have introduced and further developed various tools of modern theoretical physics , to systematise and analyse the data . these include topological tools , conceptual analogies drawn from the notion of spin chains , the notion of wilsonian universality , and methods based on the analytical structure of the discrete nonlinear schrdinger equation . in this manner we have arrived at the conclusion that the protein folding is a process that relates intimately to the emergence and interactions of solitons . in particular , a configuration such as the bloch domain wall along a spin chain appears to be most useful in comprehending how structure emerges and self - organises when a protein folds . we have inspected both the static and dynamic properties of domain wall solitons and observed that concepts which are familiar from the study of lattice systems , such as the peierls - nabarro barrier , also appear along protein backbones lattices , and in fact assume a central role in dictating how the folding proceeds . we hope that our observations help to pave a way for the powerful analytical and topological tools and techniques that have been introduced and developed in the context of integrable spin chains and related solvable models , to become part of the arsenal used describe emergence of structure and organisation in the case of proteins and other biological macromolecules . this research was supported in part by bulgarian science fund ( grant dnts - ch 01/9/2014 ) and china - bulgaria intergovernmental s&t cooperation project at ministry of science and technology of p.r . china ( 2014 - 3 ) . ajn acknowledges support from vetenskapsrdet , carl trygger s stiftelse fr vetenskaplig forskning and qian ren grant at bit , p.r . china . as was supported by national science center poland ( maestro umo-2012/06/a / st4/00376 )

strongly correlated fermi systems are among the most intriguing , best experimentally studied and fundamental systems in physics . there is , however , lack of theoretical understanding in this field of physics . the ideas based on the concepts like kondo lattice and involving quantum and thermal fluctuations at a quantum critical point have been used to explain the unusual physics . alas , being suggested to describe one property , these approaches fail to explain the others . this means a real crisis in theory suggesting that there is a hidden fundamental law of nature . it turns out that the hidden fundamental law is well forgotten old one directly related to the landau migdal quasiparticles , while the basic properties and the scaling behavior of the strongly correlated systems can be described within the framework of the fermion condensation quantum phase transition ( fcqpt ) . the phase transition comprises the extended quasiparticle paradigm that allows us to explain the non - fermi liquid ( nfl ) behavior observed in these systems . in contrast to the landau paradigm stating that the quasiparticle effective mass is a constant , the effective mass of new quasiparticles strongly depends on temperature , magnetic field , pressure , and other parameters . our observations are in good agreement with experimental facts and show that fcqpt is responsible for the observed nfl behavior and quasiparticles survive both high temperatures and high magnetic fields .

1111.4959

a. b. migdal s contribution to modern theoretical physics is very impressive . his endowment in and deep understanding of different domains of physics , including the nuclear and many - body physics , based on fermi - liquid approach , is outstanding . in migdal seminal papers a solid base for studying strongly interacting fermi systems and phase transitions occurring in them has been established @xcite . migdal s daring ideas of phase transitions related to @xmath0 condensation in nuclei and neutron stars @xcite inspired a theory of fermion condensation that has permitted to construct a new class of fermi liquids , new quasiparticles and new type of merging of single - particle levels of both finite and infinite fermi systems like nuclear , atomic and solid state systems @xcite . the new class of fermi liquids is represented by strongly correlated fermi systems where enormous number of experimental facts are collected . understanding the physics of these systems stimulates intensive studies of the possible manifestation of fermion condensation in other areas , as it has happened in the case of metal superconductivity , whose ideas were successfully used in describing atomic nuclei @xcite and in a possible explanation of the origin of the mass of elementary particles . therefore , we expect that the ideas associated with the new fermion condensation quantum phase transition @xcite in one area of research stimulates intensive studies of the possible manifestation of such a transition in other areas . strongly correlated fermi systems represented by heavy fermion ( hf ) metals and quasi - two - dimensional @xmath1he are among the most intriguing , best experimentally studied and fundamental systems in physics , which until very recently have lacked theoretical explanations @xcite . these are also a field never far from applications in synthesis of novel materials for cryogenics , rare earth magnets and applied superconductivity . the properties of these materials differ dramatically from those of ordinary fermi systems @xcite . their behavior is so unusual that the traditional landau quasiparticles paradigm does not apply to it . the paradigm states that the properties is determined by quasiparticles whose dispersion is characterized by the effective mass @xmath2 which is independent of temperature @xmath3 , the number density @xmath4 , magnetic field @xmath5 and other external parameters . the above systems are , however , in defiance of theoretical understanding . the ideas based on the concepts ( like kondo lattice involving quantum and thermal fluctuations at a quantum critical point ( qcp ) have been used to explain the unusual physics of these systems known as non - fermi liquid ( nfl ) behavior . alas , being suggested to describe one property , these approaches fail to explain the others . this means a real crisis in theory suggesting that there is a hidden fundamental law of nature , which remains to be recognized . it is widely believed that utterly new concepts are required to describe the underlying physics . there is a fundamental question : how many concepts do we need to describe the above physical mechanisms ? this can not be answered on purely experimental or theoretical grounds . rather , we have to use both of them . for instance , in the case of metals with heavy fermions , the strong correlation of electrons leads to a renormalization of the effective mass of quasiparticles , which may exceed the ordinary , `` bare '' , mass by several orders of magnitude or even become infinitely large at temperatures @xmath6 . moreover , the effective mass strongly depends on the temperature , pressure , or applied magnetic field . such metals exhibit nfl behavior and unusual power laws of the temperature dependence of the thermodynamic properties at low temperatures . the landau theory of the fermi liquid has remarkable results in describing a multitude of properties of the electron liquid in ordinary metals , fermi liquids of the @xmath1he type and nuclear liquid @xcite . the theory is based on the assumption that elementary excitations determine the physics at low temperatures . these excitations behave as quasiparticles , have a certain effective mass , and , judging by their basic properties , belong to the class of quasiparticles of a weakly interacting fermi gas . hence , the effective mass @xmath2 is independent of the temperature , pressure , and magnetic field strength and is a parameter of the theory . the landau fermi liquid ( lfl ) theory fails to explain the results of experimental observations related to the dependence of @xmath2 on the temperature @xmath3 , magnetic field @xmath5 , pressure , etc . ; this has led to the conclusion that quasiparticles do not survive in strongly correlated fermi systems and that the heavy electron does not retain its identity as a quasiparticle excitation , see e.g. @xcite . the unusual properties and nfl behavior observed in high-@xmath7 superconductors , hf metals and 2d fermi systems are assumed to be determined by various magnetic quantum phase transitions @xcite . indeed , when reasoning by analogy with respect to the second order phase transitions , one can assume that a phase transition responsible for the nfl behavior taking place up to lowest accessible temperatures is located at @xmath8 . since a quantum phase transition occurs at @xmath8 , the control parameters are the composition , electron ( hole ) number density @xmath4 , pressure , magnetic field strength @xmath5 , etc . a quantum phase transition occurs at a quantum critical point , which separates the ordered phase that emerges as a result of quantum phase transition from the disordered phase . it is usually assumed that magnetic ( e.g. , ferromagnetic and antiferromagnetic ) quantum phase transitions are responsible for the nfl behavior . the critical point of such a phase transition can be shifted to absolute zero by varying the above parameters . universal behavior can be expected only if the system under consideration is very close to a quantum critical point , e.g. , when the correlation length is much longer than the microscopic length scale , and critical quantum and thermal fluctuations determine the anomalous contribution to the thermodynamic functions of strongly correlated fermi system . quantum phase transitions of this type are so widespread @xcite that we call them ordinary quantum phase transitions @xcite . in this case , the physics of the phenomenon is determined by thermal and quantum fluctuations of the critical state , while quasiparticle excitations are destroyed by these fluctuations . conventional arguments that quasiparticles in strongly correlated fermi liquids `` get heavy and die '' at a quantum critical point commonly employ the well - known formula based on the assumptions that the @xmath9-factor ( the quasiparticle weight in the single - particle state ) vanishes at the points of second - order phase transitions @xcite . however , it has been shown that this scenario is problematic @xcite . the fluctuations in the order parameter developing an infinite correlation length and the absence of quasiparticle excitations are considered as the main reason for the nfl behavior of heavy - fermion metals , 2d fermion systems and high-@xmath7 superconductors @xcite . this approach faces certain difficulties , however . critical behavior in experiments with metals containing heavy fermions is observed at high temperatures comparable to the effective fermi temperature @xmath10 . for instance , the thermal expansion coefficient @xmath11 , which is a linear function of temperature for normal lfl , @xmath12 , demonstrates the @xmath13 temperature dependence in measurements involving ceni@xmath14ge@xmath14 as the temperature varies by two orders of magnitude ( as it decreases from 6 k to at least 50 mk ) @xcite . such behavior can hardly be explained within the framework of the critical point fluctuation theory . obviously , such a situation is possible only as @xmath6 , when the critical fluctuations make the leading contribution to the entropy and when the correlation length is much longer than the microscopic length scale . at a certain temperature @xmath10 , this macroscopically large correlation length must be destroyed by ordinary thermal fluctuations and the corresponding universal behavior must disappear . in the rest of this paper , we show that the fermion condensation quantum phase transition ( fcqpt ) @xcite is indeed responsible for the observed fascinating nfl behavior of strongly correlated fermi systems and quasiparticles survive both high temperatures and high magnetic fields . in section [ ehq ] , we give a detailed consideration of experimental evidences in favor of existence of quasiparticles , and formulate both a scaling behavior of strongly correlated fermi systems and the extended quasiparticle paradigm . then in section [ flfc1 ] , we consider the properties of landau fermi liquid . in section [ pom_m ] we demonstrate that the landau equation for the effective is not a phenomenological one and can be derived using the methods of density functional theory . thus , we establish the extended quasiparticle paradigm . fcqpt and a phase diagram of heavy fermion system located in the vicinity of fcqpt are investigated in section [ flfc ] . we propose that the phase diagram of systems located near fcqpt are strongly influenced by control parameters such as a chemical pressure , pressure or magnetic field . we find that under the application of the chemical pressure ( positive / negative ) qcp is destroyed or converted into a quantum critical line , correspondingly . in section [ hmf ] we establish that heavy fermion quasiparticles do exist in a very wide range of both temperatures @xmath3 and magnetic fields @xmath5 . finally , in section [ sum ] our results are summarized and discussed . there is a difficulty in explaining the restoration of the lfl behavior under the application of magnetic field @xmath5 , as observed in hf metals and in high-@xmath7 superconductors @xcite . for the lfl state as @xmath6 , the electric resistivity @xmath15 , the heat capacity @xmath16 , and the magnetic susceptibility @xmath17 . it turns out that the coefficient @xmath18 , the sommerfeld coefficient @xmath19 , and the magnetic susceptibility @xmath20 depend on the magnetic field strength b such that @xmath21 and @xmath22 , which implies that the kadowaki woods relation @xmath23 @xcite is @xmath5-independent and is preserved @xcite . such universal behavior , quite natural when quasiparticles with the effective mass @xmath2 playing the main role , can hardly be explained within the framework of approach that presupposes the absence of quasiparticles , which is characteristic of ordinary quantum phase transitions in the vicinity of qcp . indeed , there is no reason to expect that @xmath24 , @xmath25 and @xmath26 are affected by the fluctuations in a correlated fashion . for instance , the kadowaki woods relation does not agree with the spin density wave scenario @xcite and with the results of research in quantum criticality based on the renormalization - group approach @xcite . moreover , measurements of charge and heat transfer have shown that the wiedemann franz law holds in some high-@xmath7 superconductors @xcite and hf metals @xcite . all this suggests that quasiparticles do exist in such metals , and this conclusion is also corroborated by photoemission spectroscopy results @xcite . the inability to explain the behavior of heavy - fermion metals while staying within the framework of theories based on ordinary quantum phase transitions implies that another important concept introduced by landau , the order parameter , also ceases to operate . thus , we are left without the most fundamental principles of many - body quantum physics @xcite , and many interesting phenomena associated with the nfl behavior of strongly correlated fermi systems remain unexplained . [ ! ht ] as a function of the density @xmath4 versus @xmath3 @xcite . the density is depicted in the legend.,scaledwidth=47.0% ] the nfl behavior manifests itself in the power - law behavior of the physical quantities of strongly correlated fermi systems located close to their qcps , with exponents different from those of a fermi liquid @xcite . it is common belief that the main output of theory is the explanation of these exponents which are at least depended on the magnetic character of qcp and dimensionality of the system . on the other hand , the observed behavior of the thermodynamic properties can not be captured by these exponents as seen from figs . [ she ] [mhf ] . the behavior of the entropy @xmath27 of two - dimensional ( 2d ) @xmath28 @xcite shown in fig . [ she ] is positively different from that described by a simple function @xmath29 where @xmath30 is a constant and @xmath31 is the exponent . it is seen from fig . [ she ] that at the low densities @xmath32 @xmath33 the entropy demonstrates the lfl behavior characterized by a linear function of @xmath3 with @xmath34 . the behavior becomes quite different at the higher densities at which @xmath27 has an inflection point . obviously , at the inflection point @xmath27 can not be fit by the simple function @xmath29 . ht ] , @xmath35 , versus temperature @xmath3 as a function of magnetic field @xmath5 @xcite shown in the legend.,scaledwidth=47.0% ] as seen from fig . [ ybrhsi ] , the specific heat @xmath35 measured on @xmath36 @xcite exhibits a behavior that is to be described as a function of both temperature @xmath3 and magnetic @xmath5 field rather than by a single exponent . one can see that at low temperatures @xmath35 demonstrates the lfl behavior which is changed by the transition regime at which @xmath35 reaches its maximum and finally @xmath35 decays into nfl behavior as a function of @xmath3 at fixed @xmath5 . it is clearly seen from fig . [ ybrhsi ] that , in particularly in the transition regime , these exponents may have little physical significance . [ ! ht ] versus normalized temperature @xmath37 . @xmath38 is extracted from the measurements of the specific heat @xmath35 on @xmath36 @xcite shown in fig . the values of the field @xmath5 is listed in the legend.,scaledwidth=47.0% ] figure [ mhf ] displays measurements of @xmath35 on @xmath36 @xcite in wide range of both temperature @xmath3 and magnetic field @xmath5 variations . it is seen that both the nfl behavior and the range extends at least up to twenty kelvins and up to 18 t. figure [ mhf ] demonstrates that the lfl state at low temperatures and nfl one at higher temperatures are separated by the transition regime at which @xmath35 reaches its maximum value @xmath39 . thus , we again conclude that the observed nfl behavior exhibiting the maximum and extending to high temperatures can be hardly explained in the framework of the ordinary quantum phase transitions interpreting the exponents . in order to show that the behavior of @xmath40 and @xmath35 displayed in figs . [ she][mhf ] is of generic character , we remember that in the vicinity of qcp it is helpful to use `` internal '' scales to measure the effective mass @xmath41 and temperature @xmath3 @xcite . the internal scales of the thermodynamic functions such as @xmath40 or @xmath35 are related to `` peculiar points '' like the inflection or maximum . since the entropy has no maxima , its normalization is to be performed in the inflection point taking place at @xmath42 . note that @xmath43 is a function of @xmath4 , it is seen from fig . [ she ] that the inflection point moves towards lower temperatures at elevating @xmath4 . the normalized entropy @xmath44 as a function of the normalized temperature @xmath45 is reported in fig . [ shen ] . [ ! ht ] , extracted from the experimental facts displayed in fig . [ she ] , as a function of @xmath4 shown in the legend versus normalized @xmath46 . the arrow shows the inflection point.,scaledwidth=47.0% ] we normalize the entropy by its value at the inflection point @xmath47 . as seen from fig . [ shen ] , the normalization revels the scaling behavior of @xmath44 , that is the curves at different temperatures and densities @xmath4 merge into a single one in terms of the variable @xmath48 . we have excluded the experimental data taken at @xmath49 since the corresponding curves do not contain the inflection points . it is seen from fig . [ shen ] that @xmath50 extracted from the measurements is not a linear function of @xmath48 , as would be for a lfl , and shows the scaling behavior over three decades in normalized temperature @xmath48 . as seen from fig . [ ybrhsi ] , a maximum structure in @xmath51 at temperature @xmath52 appears under the application of magnetic field @xmath5 and @xmath52 shifts to higher @xmath3 as @xmath5 is increased . the value of the sommerfeld coefficient @xmath53 is saturated towards lower temperatures decreasing at elevated magnetic field . to obtain the normalized effective mass @xmath38 , we use @xmath54 and @xmath52 as `` internal '' scales : the maximum structure in @xmath35 was used to normalize @xmath35 , and @xmath3 was normalized by @xmath52 . in fig . [ ybrhsin ] the normalized effective mass @xmath55 as a function of normalized temperature @xmath37 is shown by geometrical figures . note that we have excluded the experimental data taken in magnetic field @xmath56 t. in that case , @xmath57 and the corresponding @xmath52 and @xmath54 are unavailable . ht ] versus normalized temperature @xmath46 . @xmath38 is extracted from the measurements of the specific heat @xmath35 on @xmath36 in magnetic fields @xmath5 @xcite listed in the legend . constant effective mass @xmath58 inherent in normal landau fermi liquids is depicted by the solid line.,scaledwidth=47.0% ] it is seen that the lfl state and nfl one are separated by the transition regime at which @xmath38 reaches its maximum value . figure [ ybrhsin ] reveals the scaling behavior of the normalized experimental curves : the curves at different magnetic fields @xmath5 merge into a single one in terms of the normalized variable @xmath59 . as seen from fig . [ ybrhsin ] , the normalized effective mass @xmath60 extracted from the measurements is not a constant , as would be for a lfl , and shows the scaling behavior over three decades in normalized temperature @xmath48 . it is seen from figs . [ ybrhsi ] and [ ybrhsin ] that the nfl behavior and the associated scaling extend at least to temperatures up to few kelvins . in order to get a deep insight in the behavior of @xmath35 displayed in fig . [ mhf ] , we again use the internal scales as it was done when constructing @xmath38 displayed in fig . [ ybrhsin ] . as a result , [ mnhf ] reveals the scaling behavior of the normalized experimental curves : the curves at different magnetic fields @xmath5 merge into a single one in terms of the normalized variable @xmath59 . as seen from fig . [ mnhf ] , the normalized effective mass @xmath60 is not a constant , as would be for a lfl , and shows the scaling behavior over two decades in normalized temperature @xmath48 . it is seen from figs . [ mhf ] and [ mnhf ] that the nfl behavior and the associated scaling extend at least both to high temperatures up to twenty kelvins and magnetic fields up to 18 t. [ ! ht ] versus normalized temperature @xmath37 . @xmath38 is extracted from the measurements of the specific heat @xmath35 on @xmath36 @xcite shown in fig . the values of the field @xmath5 is listed in the legend.,scaledwidth=47.0% ] taking into account the scaling behavior revealed in figs . [ shen][mnhf ] , we conclude that a challenging problem for theories considering the nfl behavior of strongly correlated fermi liquids is to explain both the scaling and the shape of the thermodynamic functions like @xmath44 and @xmath38 . while the theories calculating only the exponents that characterize @xmath44 and @xmath38 at @xmath61 deal with a part of the observed facts related to the problem and overlook , for example , consideration of the transition and lfl regimes . another part of the problem is the remarkably large ranges of temperature , magnetic field and the density over which the nfl behavior and the scaling are observed . concepts based on the kondo lattice and scenarios where fluctuations in the order parameter is sufficiently big and the correlation time is sufficiently large to develop the nfl behavior can hardly match up such high temperatures and explain the observed scaling behavior . as we will see below , the large temperature ranges are precursors of new quasiparticles , and it is the scaling behavior of the normalized effective mass that allows us to explain the thermodynamic of hf metals at the transition and nfl regimes . taking into account the simple behavior shown in figs . [ shen][mnhf ] , we ask the question : what theoretical concepts can replace the fermi - liquid paradigm with the notion of the effective mass in cases where fermi - liquid theory breaks down ? to date such a concept is not available @xcite . therefore , we focus on the concept of fcqpt preserving quasiparticles and intimately related to the unlimited growth of @xmath2 . it was shown that it is capable of revealing the scaling behavior of the effective mass and delivering an adequate theoretical explanation of a vast majority of experimental results in different hf metals @xcite . in contrast to the landau paradigm based on the assumption that @xmath2 is a constant as shown by the solid line in fig . [ ybrhsin ] , in fcqpt approach the effective mass @xmath2 of new quasiparticles strongly depends on @xmath3 , @xmath4 , @xmath5 etc . therefore , in accord with numerous experimental facts the extended quasiparticles paradigm is to be introduced . the main point here is that the well - defined quasiparticles determine as before the thermodynamic , relaxation and transport properties of strongly correlated fermi - systems in large temperature ranges , while @xmath2 becomes a function of @xmath3 , @xmath4 , @xmath5 etc . the fcqpt approach had been already successfully applied to describe the thermodynamic properties of such different strongly correlated systems as @xmath1he on the one hand and complicated hf compounds on the other @xcite . since we are concentrated on properties that are non - sensitive to the detailed structure of the system we avoid difficulties associated with the anisotropy generated by the crystal lattice of solids , its special features , defects , etc . , we study the universal behavior of strongly correlated fermi - systems located near their qcp at low temperatures using the model of a homogeneous hf liquid @xcite . the model is quite meaningful because we consider the scaling behavior exhibited by these materials at low temperatures @xcite . the scaling properties of the normalized effective mass that characterizes them , are determined by momentum transfers that are small compared to momenta of the order of the reciprocal lattice length . the high momentum contributions can therefore be ignored by substituting the lattice for the jelly model . while the values of the scales like the maximum @xmath54 of the effective mass and @xmath52 at which @xmath54 takes place are determined by a wide range of momenta and thus these scales are controlled by the specific properties of the system . on the other hand , the dependencies of these scales on magnetic fields , temperature , the density of system , pressure etc are again controlled by small momenta and can be analyzed within the model of hf liquid . one of the most complex problems of modern condensed matter physics is the problem of the structure and properties of fermi systems with large inter particle coupling constants . theory of fermi liquids , later called `` normal '' , was first proposed by landau as a means for solving such problems by introducing the concept of quasiparticles and amplitudes that characterize the effective quasiparticle interaction @xcite . the landau theory can be regarded as an effective low - energy theory with the high - energy degrees of freedom eliminated by introducing amplitudes that determine the quasiparticle interaction instead of the strong inter particle interaction . the stability of the ground state of the landau fermi liquid is determined by the pomeranchuk stability conditions : stability is violated when at least one landau amplitude becomes negative and reaches its critical value @xcite . we note that the new phase in which stability is restored can also be described , in principle , by the lfl theory . we begin by recalling the main ideas of the lfl theory @xcite . the theory is based on the quasiparticle paradigm , which states that quasiparticles are elementary weakly excited states of fermi liquids and are therefore specific excitations that determine the low - temperature thermodynamic and transport properties of fermi liquids . in the case of the electron liquid , the quasiparticles are characterized by the electron quantum numbers and the effective mass @xmath62 . the ground state energy of the system is a functional of the quasiparticle occupation numbers ( or the quasiparticle distribution function ) @xmath63 , and the same is true of the free energy @xmath64 , the entropy @xmath65 , and other thermodynamic functions . we can find the distribution function from the minimum condition for the free energy @xmath66 ( here and in what follows @xmath67 ) @xmath68 here @xmath69 is the chemical potential fixing the number density @xmath70 and @xmath71 is the quasiparticle energy . this energy is a functional of @xmath72 , in the same way as the energy @xmath73 is : @xmath74 . the entropy @xmath65 related to quasiparticles is given by the well - known expression @xcite @xmath75\frac{d{\bf p}}{(2\pi ) ^3},\label{fl3}\end{aligned}\ ] ] which follows from combinatorial reasoning . equation is usually written in the standard form of the fermi dirac distribution , @xmath76\right\}^{-1}.\label{fl4}\ ] ] at @xmath77 , ( [ fl1 ] ) and ( [ fl4 ] ) have the standard solution @xmath78 if the derivative @xmath79 is finite and positive . here @xmath80 is the fermi momentum and @xmath81 is the step function . the single particle energy can be approximated as @xmath82 , and @xmath83 inversely proportional to the derivative is the effective mass of the landau quasiparticle , @xmath84 in turn , the effective mass @xmath58 is related to the bare electron mass @xmath85 by the well - known landau equation @xcite @xmath86 where @xmath87 is the landau amplitude , which depends on the momenta @xmath88 and @xmath89 and the spins @xmath90 . for simplicity , we ignore the spin dependence of the effective mass , because @xmath58 is almost completely spin - independent in the case of a homogeneous liquid and weak magnetic fields . the landau amplitude @xmath91 is given by @xmath92 the stability of the ground state of lfl is determined by the pomeranchuk stability conditions : stability is violated when at least one landau amplitude becomes negative and reaches its critical value @xcite @xmath93 here @xmath94 and @xmath95 are the dimensionless spin - symmetric and spin - antisymmetric landau amplitudes , @xmath96 is the angular momentum related to the corresponding legendre polynomials @xmath97 , @xmath98.\ ] ] here @xmath99 is the angle between momenta @xmath89 and @xmath100 and the density of states @xmath101 . it follows from eq . that @xmath102 in accordance with the pomeranchuk stability conditions it is seen from eq . that @xmath103 , otherwise the effective mass becomes negative leading to unstable state when it is energetically favorable to excite quasiparticles near the fermi surface . it is common belief that the equations of the section [ flfc1 ] are phenomenological and inapplicable to describe fermi systems characterized by the effective mass @xmath2 strongly dependent on temperature , external magnetic fields @xmath5 , pressure @xmath104 etc . to derive the equation determining the effective mass , we consider the model of a homogeneous hf liquid and employ the density functional theory for superconductors ( scdft ) @xcite which allows us to consider @xmath73 as a functional of the occupations numbers @xmath105 @xcite . as a result , the ground state energy of the normal state @xmath73 becomes the functional of the occupation numbers and the function of the number density @xmath4 , @xmath106 , while eq . gives the single - particle spectrum . upon differentiating both sides of eq . with respect to @xmath89 and after some algebra and integration by parts , we obtain @xcite @xmath107 to calculate the derivative @xmath108 , we employ the functional representation @xmath109 it is seen directly from eq . that the effective mass is given by the well - known landau equation @xmath110 for simplicity , we ignore the spin dependencies . to calculate @xmath2 as a function of @xmath3 , we construct the free energy @xmath66 , where the entropy @xmath40 is given by eq . . minimizing @xmath91 with respect to @xmath111 , we arrive at the fermi dirac distribution , eq . . due to the above derivation , we conclude that eqs . and are exact ones and allow us to calculate the behavior of both @xmath112 and @xmath2 which now is a function of temperature @xmath3 , external magnetic field @xmath5 , number density @xmath4 and pressure @xmath104 rather than a constant . as we will see it is this feature of @xmath2 that forms both the scaling and the nfl behavior observed in measurements on hf metals . in lfl theory it is assumed that @xmath58 is positive , finite and constant since the integral on the right hand side of eq . represents a small correction in the case of normal metals . as a result , the temperature - dependent corrections to @xmath113 , the quasiparticle energy @xmath114 and other quantities begin with the term proportional to @xmath115 in 3d systems and with the term proportional to @xmath3 in 2d one @xcite . the effective mass is given by eq . , and the specific heat @xmath116 is @xcite @xmath117 and the spin susceptibility @xmath118 where @xmath119 is the bohr magneton and @xmath120 . in the case of lfl the electrical resistivity at low @xmath3 is given by @xcite @xmath121 where @xmath122 is the residual resistivity and @xmath26 is the coefficient determining the charge transport . the coefficient @xmath26 is proportional to the quasiparticle - quasiparticle scattering cross - section . equation symbolizes and defines the lfl behavior observed in normal metals . equation ( [ fll ] ) at @xmath8 , combined with the fact that @xmath123 becomes @xmath124 , yields the well - known result @xcite @xmath125 where @xmath126 , @xmath127 is the density of states of a free fermi gas and @xmath128 is the @xmath129-wave component of the landau interaction amplitude . because @xmath130 in the landau fermi - liquid theory , the landau interaction amplitude can be written as @xmath131 . provided that at a certain critical point @xmath132 , the denominator @xmath133 tends to zero , i.e. , @xmath134 , we find that @xcite @xmath135 where @xmath30 and @xmath31 are constants and @xmath136 is the `` distance '' from qcp @xmath132 at which @xmath137 . we note that the divergence of the effective mass given by eq . does preserve the pomeranchuk stability conditions for @xmath138 is positive , see eq . . equations and seem to be different but it is not the case since @xmath139 , while @xmath140 and eq . represents an implicit formula for the effective mass . the behavior of @xmath141 described by formula ( [ fl7 ] ) is in good agreement with the results of experiments @xcite and calculations @xcite . in the case of electron systems , eq . holds for @xmath142 , while for 2d @xmath1he we have @xmath143 so that always @xmath144 @xcite . such behavior of the effective mass is observed in hf metals , which have a fairly flat and narrow conductivity band corresponding to a large effective mass , with a strong correlation and the effective fermi temperature @xmath145 of the order of several dozen degrees kelvin or even lower ( e.g. , see ref . @xcite ) . the effective mass as a function of the electron density @xmath4 in a silicon mosfet ( metal oxide semiconductor field effect transistor ) , approximated by eq . ( [ fl7 ] ) , is shown in fig . the parameters @xmath30 , @xmath31 and @xmath132 are taken as fitting . we see that eq . provides a good description of the experimental results . [ ! ht ] in a silicon mosfet as a function of the electron number density @xmath4 . the squares mark the experimental data on the shubnikov - de haas oscillations , and the data obtained by applying a parallel magnetic field are marked by circles @xcite . the solid line represents the function ( [ fl7]).,scaledwidth=47.0% ] the divergence of the effective mass @xmath141 discovered in measurements involving 2d @xmath1he @xcite is illustrated in fig . figures [ fig5 ] and [ fig6 ] show that the description provided by eq . does not depend on elementary fermi particles constituting the system and is in good agreement with the experimental data . [ ! ht ] in 2d @xmath1he as a function of the density @xmath4 of the liquid , obtained from heat capacity and magnetization measurements . the experimental data are marked by squares @xcite , and the solid line represents the function given by eq . .,scaledwidth=47.0% ] it is instructive to briefly explore the scaling behavior of @xmath2 in order to illustrate the ability of the quasiparticle extended paradigm to capture the scaling behavior . let us write the quasiparticle distribution function as @xmath146 , with @xmath111 being the step function , and eq . then becomes @xmath147 at qcp @xmath148 , the effective mass @xmath141 diverges and eq . becomes homogeneous determining @xmath2 as a function of temperature while the system exhibits the nfl behavior . if the system is located before qcp , @xmath2 is finite , at low temperatures the integral on the right hand side of eq . represents a small correction to @xmath149 and the system demonstrates the lfl behavior seen in figs . [ ybrhsi ] and [ ybrhsin ] . the lfl behavior assumes that the effective mass is independent of temperature , @xmath150 , as shown by the horizontal line in fig . [ ybrhsin ] . obviously , the lfl behavior takes place only if the second term on the right hand side of eq . is small in comparison with the first one . then , as temperature rises the system enters the transition regime : @xmath2 grows , reaching its maximum @xmath54 at @xmath151 , with subsequent diminishing . as seen from fig . [ ybrhsin ] , near temperatures @xmath152 the last `` traces '' of lfl regime disappear , the second term starts to dominate , and again eq . becomes homogeneous , and the nfl behavior is restored , manifesting itself in decreasing @xmath2 as a function of @xmath3 . as shown in section [ pom_m ] , the pomeranchuk stability conditions do not encompass all possible types of instabilities and that at least one related to the divergence of the effective mass given by eq . was overlooked @xcite . this type of instability corresponds to a situation where the effective mass , the most important characteristic of quasiparticles , can become infinitely large . as a result , the quasiparticle kinetic energy is infinitely small near the fermi surface and the quasiparticle distribution function @xmath111 minimizing @xmath153 is determined by the potential energy . this leads to the formation of a new class of strongly correlated fermi liquids with fc @xcite , separated from the normal fermi liquid by fcqpt @xcite . it follows from ( [ fl7 ] ) that at @xmath8 and as @xmath154 the effective mass diverges , @xmath155 . beyond the critical point @xmath132 , the distance @xmath156 becomes negative and , correspondingly , so does the effective mass . to avoid an unstable and physically meaningless state with a negative effective mass , the system must undergo a quantum phase transition at the critical point @xmath157 , which , as we will see shortly , is fcqpt @xcite . because the kinetic energy of quasiparticles that are near the fermi surface is proportional to the inverse effective mass , the potential energy of the quasiparticles near the fermi surface determines the ground - state energy as @xmath148 . hence , at @xmath8 a phase transition reduces the energy of the system and transforms the quasiparticle distribution function . beyond qcp @xmath157 , the quasiparticle distribution is determined by the ordinary equation for a minimum of the energy functional @xcite : @xmath158 equation ( [ fl8 ] ) yields the quasiparticle distribution function @xmath159 that minimizes the ground - state energy @xmath73 . this function found from eq . ( [ fl8 ] ) differs from the step function in the interval from @xmath160 to @xmath80 , where @xmath161 , and coincides with the step function outside this interval . in fact , eq . coincides with eq . provided that the fermi surface at @xmath162 transforms into the fermi volume at @xmath163 suggesting that the single - particle spectrum is absolutely `` flat '' within this interval . a possible solution @xmath164 of eq . ( [ fl8 ] ) and the corresponding single - particle spectrum @xmath114 are depicted in fig . quasiparticles with momenta within the interval @xmath165 have the same single - particle energies equal to the chemical potential @xmath69 and form fc , while the distribution @xmath164 describes the new state of the fermi liquid with fc @xcite . in contrast to the landau , marginal , or luttinger fermi liquids @xcite , which exhibit the same topological structure of the green s function , in systems with fc , where the fermi surface spreads into a strip , the green s function belongs to a different topological class . the topological class of the fermi liquid is characterized by the invariant [ ! ht ] and the quasiparticle distribution function @xmath166 . because @xmath166 is a solution of eq . ( [ fl8 ] ) at @xmath8 , we have @xmath167 , @xmath168 , and @xmath169 , while @xmath170 and the band becomes completely flat . the fermi momentum @xmath80 satisfies the condition @xmath171.,scaledwidth=47.0% ] @xcite @xmath172 where `` tr '' denotes the trace over the spin indices of the green s function and the integral is taken along an arbitrary contour @xmath116 encircling the singularity of the green s function . the invariant @xmath173 in ( [ flvol ] ) takes integer values even when the singularity is not of the pole type , can not vary continuously , and is conserved in a transition from the landau fermi liquid to marginal liquids and under small perturbations of the green s function . as shown by volovik @xcite , the situation is quite different for systems with fc , where the invariant @xmath173 becomes a half - integer and the system with fc transforms into an entirely new class of fermi liquids with its own topological structure . we start with visualizing the main properties of fcqpt . to this end , again consider scdft . scdft states that the thermodynamic potential @xmath174 is a universal functional of the number density @xmath175 and the anomalous density ( or the order parameter ) @xmath176 , providing a variational principle to determine the densities . at the superconducting transition temperature @xmath7 a superconducting state undergoes the second order phase transition . our goal now is to construct a quantum phase transition which evolves from the superconducting one , since , as seen from fig . [ fig1 ] , the superconducting order parameter @xmath177 is finite over the region @xmath178 . let us assume that the coupling constant @xmath179 of the bcs - like pairing interaction @xcite vanishes , with @xmath180 making vanish the superconducting gap at any finite temperature . in that case , @xmath181 and the superconducting state takes place at @xmath8 while at finite temperatures there is a normal state . this means that at @xmath8 the anomalous density @xmath182 is finite , while the superconducting gap @xmath183 is infinitely small @xcite . in eq . , the field operator @xmath184 annihilates an electron of spin @xmath185 at the position @xmath186 . for the sake of simplicity , we consider the model of homogeneous hf liquid @xcite . then at @xmath8 , the thermodynamic potential @xmath174 reduces to the ground state energy @xmath73 which turns out to be a functional of the occupation number @xmath111 since in that case the order parameter @xmath187 . indeed , @xmath188 where @xmath189 and @xmath190 are normalized parameters such that @xmath191 and @xmath192 , see e.g. @xcite . upon minimizing @xmath73 with respect to @xmath111 , we obtain eq . . as soon as eq . has nontrivial solution @xmath164 then instead of the fermi step , we have @xmath161 in certain range of momenta @xmath193 with @xmath194 being finite in this range , while the single particle spectrum @xmath195 is flat . thus , the step - like fermi filling inevitably undergoes restructuring and forms fc when eq . possesses for the first time the nontrivial solution at @xmath196 which is qcp of fcqpt . in that case , the range vanishes , @xmath197 , and the effective mass @xmath2 diverges at qcp @xcite @xmath198 at any small but finite temperature the anomalous density @xmath199 ( or the order parameter ) decays and this state undergoes the first order phase transition and converts into a normal state characterized by the thermodynamic potential @xmath200 . indeed , at @xmath6 , the entropy @xmath201 of the normal state is given by eq . . it is seen from eq . that the normal state is characterized by the temperature - independent entropy @xmath202 @xcite . since the entropy of the superconducting ground state is zero , we conclude that the entropy is discontinuous at the phase transition point , with its discontinuity @xmath203 . thus , the system undergoes the first order phase transition . the heat @xmath204 of transition from the asymmetrical to the symmetrical phase is @xmath205 since @xmath206 . because of the stability condition at the point of the first order phase transition , we have @xmath207 . obviously the condition is satisfied since @xmath208 . at @xmath8 , a quantum phase transition is driven by a nonthermal control parameter , e.g. , the number density @xmath4 . as we have seen in section [ pom_m ] , at qcp , @xmath157 , the effective mass diverges . it follows from eq . that beyond qcp , the effective mass becomes negative . to avoid an unstable and physically meaningless state with a negative effective mass , the system undergoes fcqpt leading to the formation of fc . [ ! ht ] is taken as the control parameter and depicted as @xmath209 . the dashed line shows @xmath210 as the system approaches qcp , @xmath211 , of fcqpt which is denoted by the arrow . at @xmath212 and sufficiently low temperatures , the system is in the lfl state as shown by the shadow area . the new location of the system induced by the application of the positive pressure is shown by both the solid square and the arrow . the vertical arrow illustrates the system moving in the lfl - nfl direction along @xmath3 at fixed control parameter . at @xmath8 and beyond the critical point , @xmath213 , the system is at the quantum critical line depicted by the dashed line and shown by the vertical arrow . the critical line is characterized by the fc state with finite superconducting order parameter @xmath199 . at any finite low temperature @xmath214 , @xmath199 is destroyed , the system undergoes the first order phase transition , possesses finite entropy @xmath202 and exhibits the nfl behavior at any finite temperatures @xmath215.,scaledwidth=47.0% ] a schematic phase diagram of the system which is driven to the fc state by variation of @xmath4 is reported in fig . [ fig1 ] . upon approaching the critical density @xmath132 the system remains in the lfl region at sufficiently low temperatures as it is shown by the shadow area . the temperature range of the shadow area shrinks as the system approaches qcp , and @xmath216 diverges as shown by the dashed line and eq . . at qcp @xmath132 shown by the arrow in fig . [ fig1 ] , the system demonstrates the nfl behavior down to the lowest temperatures . beyond the critical point at finite temperatures the behavior remains the nfl and is determined by the temperature - independent entropy @xmath202 @xcite . in that case at @xmath217 , the system is again demonstrating the nfl behavior and approaching a quantum critical line ( qcl ) ( shown by the vertical arrow and the dashed line in fig . [ fig1 ] ) rather than a quantum critical point . upon reaching the quantum critical line from the above at @xmath6 the system undergoes the first order quantum phase transition , which is fcqpt taking place at @xmath206 . while at diminishing temperatures , the systems located before qcp do not undergo a phase transition and their behavior transits from nfl to lfl . it is seen from fig . [ fig1 ] that at finite temperatures there is no boundary ( or phase transition ) between the states of systems located before or behind qcp shown by the arrow . therefore , at elevated temperatures the properties of systems with @xmath213 or with @xmath212 become indistinguishable . on the other hand , at @xmath218 the nfl state above the critical line and in the vicinity of qcp is strongly degenerate , therefore the degeneracy stimulates the emergence of different phase transitions lifting it and the nfl state can be captured by the other states such as superconducting ( for example , by the superconducting state ( sc ) in @xmath219 @xcite ) or by antiferromagnetic ( af ) state ( e.g. af one in @xmath36 @xcite ) etc . the diversity of phase transitions occurring at low temperatures is one of the most spectacular features of the physics of many hf metals . within the scenario of ordinary quantum phase transitions , it is hard to understand why these transitions are so different from one another and their critical temperatures are so extremely small . however , such diversity is endemic to systems with a fc @xcite . as mentioned above , the system located above qcl exhibits the nfl behavior down to lowest temperatures unless it is captured by a phase transition . the behavior exhibiting by the system located above qcl is in accordance with the experimental observations that study the evolution of qcp in @xmath36 under negative chemical pressure induced with @xmath220 substitution @xcite . as reported , the negative pressure leaves an intermediate spin - liquid - type ground state over a finite magnetic field range . we assume a simple model that the application of negative pressure reduces @xmath4 and the system moves to a position over qcl . that is the @xmath220 substitution drives @xmath221 to a paramagnetic state over the finite magnetic field @xmath5 range , so that the electronic system of the hf metal is located above qcl shown in fig . [ fig1 ] and demonstrates the nfl behavior . in that case magnetic field @xmath5 represents an additional control parameter driving @xmath36 to the paramagnetic state . we predict that at lowering temperatures , the electronic system of @xmath36 is captured by a phase transition , since the nfl state above qcl is strongly degenerate . at diminishing temperatures , this degeneracy is to be removed by some phase transition which likely can be detected by the lfl state accompanying it . upon using nonthermal tuning parameters like the number density @xmath4 , the nfl behavior can be destroyed and the lfl one will be restored . in our simple model , the application of positive pressure @xmath104 makes @xmath4 grow removing the qcp of fcqpt shown in fig . [ fig1 ] , so that the electronic system of @xmath36 moves into the shadow area characterized by the lfl behavior at low temperatures . the new location of the system is shown by the arrow pointing at the solid square . as a result , the application of magnetic field @xmath222 does not drive the system to its qcp with the divergent effective mass because the qcp is already destroyed by the pressure . here @xmath223 is the critical magnetic field that eliminates the corresponding af order . at @xmath224 and raising temperatures , the system moving along the vertical arrow transits from the lfl regime to the nfl one . as seen from fig . [ fig1 ] , in the nfl regime the system properties coincide with that of the system at @xmath225 . the observed behavior is in accord with the experimental facts indicating the separation of the af order from qcp under the application of pressure in @xmath221 @xcite . as was mentioned in section [ ehq ] , one of the most interesting and puzzling issues in the research of hf metals is the nfl behavior taking place in wide range of both temperature @xmath3 and magnetic field @xmath5 variations @xcite . for example , measurements of the specific heat on @xmath36 under the application of magnetic field @xmath5 show that the range extends at least up to twenty kelvins as shown in fig . [ mhf ] . in fig . [ mnhf ] the normalized effective mass @xmath38 as a function of normalized temperature @xmath46 is shown by geometrical figures . figures [ mhf ] and [ mnhf ] reveal the remarkably large temperature ranges over which the nfl behavior and scaling one are observed . in this section , we analyze the thermodynamic properties of @xmath221 in both low and high magnetic fields . our calculations of the specific heat and magnetization allow us to conclude that under the application of magnetic field the heavy - electron system of @xmath36 evolves continuously without a metamagnetic transition . at low temperatures and high magnetic fields @xmath226 the system is completely polarized and demonstrates the lfl behavior , at elevated temperatures the hf behavior and related nfl one are restored . the obtained results are in good agreement with experimental facts as both temperature and magnetic field vary by more than two orders of magnitude . we use the landau equation to study the behavior of the effective mass @xmath227 as a function of the temperature and the magnetic field . for the model of homogeneous hf liquid at finite temperatures and magnetic fields , eq . acquires the form @xcite @xmath228 where @xmath229 is the landau amplitude dependent on momenta @xmath80 , @xmath129 and spin @xmath90 . for the sake of definiteness , we assume that the hf liquid is 3d liquid . the quasiparticle distribution function can be expressed as @xmath230\right\ } ^{-1},\label{hc2}\ ] ] where @xmath231 is the single - particle spectrum . in our case , the chemical potential may depend on the spin due to the zeeman splitting @xmath232 , where @xmath119 is the bohr magneton and @xmath69 is the chemical potential preserving the number of particles . we write the quasiparticle distribution function as @xmath233 . equation ( [ hc1 ] ) then becomes @xmath234 equation is solved with a quite general form of landau interaction amplitude @xcite . choice of the amplitude is dictated by the fact that the system has to be at the quantum critical point of fcqpt , which means that the first two @xmath129-derivatives of the single - particle spectrum @xmath195 should equal zero . since the first derivative is proportional to the reciprocal quasiparticle effective mass @xmath149 , its zero just signifies qcp of fcqpt . the second derivative must vanish ; otherwise @xmath235 has the same sign below and above the fermi surface , and the landau state becomes unstable @xcite . zeros of these two subsequent derivatives mean that the spectrum @xmath114 has an inflection point at @xmath80 so that the lowest term of its taylor expansion is proportional to @xmath236 . when the system is near fcqpt , it turns out that the normalized solution of eq . @xmath60 can be well approximated by a simple universal interpolating function . the interpolation occurs between the lfl ( @xmath237 ) and nfl ( @xmath238 ) regimes@xcite @xmath239 here @xmath240 and @xmath241 are constants , @xmath242 , @xmath243 and @xmath244 are fitting parameters , related to the landau amplitude . magnetic field @xmath5 enters eq . as @xmath245 making @xmath246 @xcite . we conclude that under the application of magnetic field the variable @xmath247 remains the same and the normalized effective mass is again governed by eq . . here @xmath223 is the critical magnetic field driving both hf metal to its magnetic field tuned qcp and corresponding nel temperature to @xmath8 . in some cases as in the hf metal @xmath248 , @xmath249 , see e.g. @xcite , while in @xmath36 , @xmath250 t @xcite and in @xmath251 the critical field is much larger @xmath252 t @xcite . therefore , in our simple model of the hf liquid @xmath223 is taken as a parameter . in what follows , we compute the effective mass using eq . and employ eq . for qualitative estimations of the considered values . based on eq . with magnetic field as the control parameter @xmath5 . the vertical and horizontal arrows show lfl - nfl and nfl - lfl transitions at fixed @xmath5 and fixed @xmath3 correspondingly . at @xmath253 the system is in antiferromagnetic ( afm ) state . the width of the transition regime @xmath254 is shown by the segment confined by the two vertical arrows . inset shows a schematic plot of the normalized effective mass versus the normalized temperature . transition regime , where @xmath38 reaches its maximum value at @xmath255 , is shown by the hatched area both in the main panel and in the inset.,scaledwidth=48.0% ] we now are in position to construct the schematic phase diagram of the hf metal @xmath36 at @xmath256 . the pase diagram is reported in fig . the magnetic field @xmath5 plays a role of the control parameter , driving the system towards its qcp . in our case this qcp is of fcqpt type . the fcqpt peculiarity occurs at @xmath257 , yielding new strongly degenerate state at @xmath258 . to lift this degeneracy , the system forms either superconducting ( sc ) or magnetically ordered ( ferromagnetic ( fm ) , antiferromagnetic ( afm ) etc ) states @xcite . in the case of @xmath36 , this state is afm one @xcite . as it follows from eqs . and and seen from fig . [ mt ] , at @xmath259 the system is in either nfl or lfl states . at fixed temperatures the increase of @xmath5 drives the system along the horizontal arrow from nfl state to lfl one . on the contrary , at fixed magnetic field and raising temperatures the system transits along the vertical arrow from lfl state to nfl one . the inset to fig . [ mt ] demonstrates the behavior of the normalized effective mass @xmath55 versus normalized temperature @xmath59 following from eq . . the @xmath260 regime is marked as nfl one since ( contrary to lfl case where the effective mass is constant ) the effective mass depends strongly on temperature . it is seen that temperature region @xmath261 signifies a transition regime between the lfl behavior with almost constant effective mass and nfl one , given by @xmath260 dependence . thus , temperatures @xmath262 , shown by arrows in the inset and main panel , can be regarded as a transition regime between lfl and nfl states . it is seen from eq . that the width of the transition regime @xmath254 is proportional to @xmath263 . it is shown by the segment between two vertical arrows in fig [ mt ] . these theoretical results are in good agreement with the experimental facts @xcite . in fig . [ mrm ] , our calculations of the normalized effective mass @xmath264 at fixed magnetic field @xmath265 making the system fully polarized are shown by the solid line . figure [ mrm ] reveals the scaling behavior of the normalized experimental curves - the scaled curves at different magnetic fields @xmath5 merge into a single one in terms of the normalized variable @xmath59 . it is seen from fig . [ mrm ] that our calculations are in accord with facts @xcite : at elevated temperatures ( @xmath266 ) the lfl state first changes into the transition regime and then disrupts into the nfl state . [ ! ht ] versus normalized temperature @xmath37 . @xmath38 is extracted from the data shown in fig . magnetic fields @xmath5 is listed in the legend . our calculations ( made at @xmath267 when the quasiparticle band is fully polarized ) are depicted by the solid curve tracing the scaling behavior of @xmath38.,scaledwidth=47.0% ] to get further insight into the behavior of the system at high magnetic fields , in fig . [ f2 ] we collect the curves @xmath264 both at low ( symbols in the upper box in fig . [ f2 ] ) and high ( symbols in the lower box ) magnetic fields @xmath5 . all curves have been extracted from the experimental facts @xcite . it is seen that while at low fields the low - temperature ends ( @xmath268 ) of the curves completely merge , at high fields this is not the case . moreover , the low - temperature asymptotic value of @xmath269 at low fields is around two times more then that at high fields . the physical reason for low - field curves merging is that the effective mass does not depend on spin variable so that the polarizations of subbands with @xmath270 and @xmath271 are almost equal to each other . this is reflected in our calculations , based on eq . for low magnetic fields @xmath272 . the result is shown by the dotted line in fig . [ f2 ] . [ ! ht ] at low ( upper box symbols ) and high ( lower box symbols ) magnetic fields extracted from the specific heat ( @xmath35 ) measurements of the @xmath36 @xcite . our low - field calculations are depicted by the dotted line tracing the scaling behavior of @xmath38 . our high - field calculations ( solid line ) are taken at @xmath273 when the quasiparticle band becomes fully polarized.,scaledwidth=49.0% ] it is also seen from fig . [ f2 ] that all low - temperature differences between high- and low field behavior of the normalized effective mass disappear at high temperatures . in other words , while at low temperatures the values of @xmath38 for low fields are two times more then those for high fields , at temperatures @xmath274 this difference disappear . it is seen that these high temperatures lie about the transition region , marked by hatched area in the inset to fig . this means that two states ( lfl and nfl ) separated by the transition region are clearly seen in fig . [ f2 ] displaying good agreement between our calculations ( dotted line for low fields and thick line at high fields ) and the experimental points ( symbols ) . contrary to the low fields case , at high fields @xmath273 ( symbols in the lower box of fig . [ f2 ] ) , in the low temperature lfl state the curves @xmath264 do not merge into single one and their values decrease as @xmath5 grows representing the full spin polarization of the hf band at the highest reached magnetic fields . as we have mentioned above , at temperature raising all effects of spin polarization smear down , yielding the restoration nfl behavior at @xmath275 . our high - field calculations ( solid line in fig . [ f2 ] ) reflect the latter fact and are also in good agreement with experimental facts . in order not to overload fig . [ f2 ] with unnecessary details , we show the calculations only for the case of the full spin polarization . [ ! ht ] of the functions @xmath35 versus magnetic field @xmath5 for @xmath36 . the maxima taken at the low fields @xcite are shown by the solid squares , the ones taken at the high fields @xcite are depicted by the solid circles . the solid curve is approximated by @xmath276 , @xmath277 is a fitting parameter.,scaledwidth=49.0% ] figure [ mb ] reports the maxima @xmath39 of the functions @xmath35 shown in fig . [ mhf ] versus @xmath5 depicted by the solid circles and the solid squares , corresponding measurements on @xmath221 taken at the low magnetic fields @xcite . the solid line represents the approximation describing the maxima value of @xmath278 calculated within the framework of fcqpt theory @xcite . it is seen that our calculations are in good agreement with the experimental facts over two decades in magnetic field @xmath5 . the agreement indicates that at @xmath279 the transition regime takes place and the nfl behavior restores at higher temperatures @xmath280 k. in fig . [ tb ] , the solid squares and circles denote temperatures @xmath281 at which the maxima of @xmath35 take place . to fit the experimental data @xcite the function @xmath282 defined by eq . it is seen from fig . [ tb ] that our calculations shown by the solid line are in accord with experimental facts , and we conclude that the transition regime of @xmath36 is restored at temperatures @xmath283 . [ ! ht ] at which the maxima of @xmath35 are located versus magnetic field @xmath5 for @xmath36 . the solid squares denote @xmath52 taken in measurements at the low magnetic fields @xcite . the solid circles denote @xmath52 at the high magnetic fields @xcite . the solid line represents the function @xmath284 , @xmath241 is a fitting parameter , see eq . .,scaledwidth=47.0% ] consider now the magnetization @xmath285 as a function of magnetic field @xmath5 at fixed temperature @xmath286 @xmath287 where the magnetic susceptibility @xmath25 is given by @xcite @xmath288 here , @xmath289 is a constant and @xmath290 is the landau amplitude related to the exchange interaction . at low magnetic fields , @xmath256 , our calculations show that the magnetization exhibits a kink at some magnetic field @xmath291 . the experimental magnetization demonstrates the same behavior @xcite . we use @xmath292 and @xmath293 to normalize @xmath5 and @xmath294 respectively . due to the normalization the coefficients @xmath289 and @xmath295 drop out from the result , and @xmath296 @xcite . so we can use eq . to calculate the magnetic susceptibility @xmath25 . the normalized magnetization @xmath297 both extracted from experiment ( symbols ) and calculated one ( solid line ) , are reported in the inset to fig . it shows that our calculations are in good agreement with the experimental facts , and all the data exhibit the kink ( shown by the arrow ) at @xmath298 taking place as soon as the system enters the transition region corresponding to the magnetic fields where the horizontal arrow in fig . [ mt ] crosses the hatched area . to illuminate the kink position , in the fig . [ kink ] we present the @xmath299 dependence in logarithmic - logarithmic scale . in that case the straight lines show clearly the change of the power dependence @xmath299 at the kink . [ ! ht ] ( symbols ) and straight lines which are guides for eye . the intersection of the straight lines visualize the kink at the crossover region in the fig . the inset : the field dependencies of the normalized magnetization @xmath294 of @xmath300 @xcite at @xmath301 k. the kink ( shown by the arrow ) is clearly seen at the normalized field @xmath302 . the solid curve represents our calculations.,scaledwidth=49.0% ] at magnetic field @xmath267 the quasiparticle band becomes fully polarized and a new kink appears @xcite , we call this kink as the second one . our calculations of the normalized magnetization ( line ) and the experimental points ( squares ) are shown in fig . [ kink2 ] . in that case both the magnetization and the field are normalized by the corresponding values at the second kink . ht ] as a function of the normalized magnetic field @xmath303 . the calculated magnetization is shown by the solid line , and the experimental facts @xcite is represented by the solid squares . the linear dependence of @xmath304 on @xmath303 is shown by the arrows . the inset demonstrates the experimental data with the subtracted linear dependence depicted by the stars . our calculations is shown by the solid line.,scaledwidth=47.0% ] in the fig . [ kink2 ] we plot our calculated magnetization ( in the variables @xmath5 and @xmath294 normalized to those in the second kink point ) versus experimental one . the marvelous coincidence is seen everywhere except the high - field part at @xmath305 where the normalized magnetization @xmath304 exhibits a linear dependence on @xmath303 depicted by the two arrows , while the calculated magnetization is approximately constant at these @xmath303 . such a behavior is the intrinsic shortcoming of the hf liquid model that takes into account only heavy electrons and omit the other conduction electrons that contribute to the magnetization @xcite . thus , we can consider the high - field ( at @xmath306 ) part of the magnetization as the contribution which does not included in our theory . to separate this contribution from the experimental magnetization curve , we ( numerically ) differentiate it , then subtract constant part at @xmath306 and integrate back the resulting curve . the coincidence between our calculations depicted by the solid curve and processed experimental data shown by the stars is reported in the inset to fig . [ kink2 ] . as we can see now , the coincidence between the theory and experiment is good in the entire magnetic field domain . taking into account the obtained results displayed in figs . [ f2][kink2 ] , we conclude that the hf system of @xmath300 evolves continuously under the application of magnetic fields . this observation is in agreement with experimental observations @xcite . we have shown that the physics of systems with heavy fermions is determined by the extended quasiparticle paradigm . in contrast to the paradigm that the quasiparticle effective mass is a constant , within the extended quasiparticle paradigm the effective mass of new quasiparticles strongly depends on the temperature , magnetic field , pressure , and other parameters . the quasiparticles are well defined and can be used to describe the scaling behavior of the thermodynamic hf metals in a wide range of both temperatures @xmath3 and magnetic fields @xmath5 . it was demonstrated that the phase diagram of systems located near the fermion condensation quantum phase transition are strongly influenced by control parameters such as chemical pressure , pressure , the number density or magnetic field . analyzing the phase diagram of @xmath36 , we have found that under the application of the chemical pressure ( positive / negative ) qcp is destroyed or converted into the quantum critical line , correspondingly . we have analyzed the thermodynamic properties of @xmath36 in both the low and the high magnetic fields . our calculations allow us to conclude that in magnetic field the hf system of @xmath36 evolves continuously without a metamagnetic transition and possible localization of heavy @xmath307 electrons . at low temperatures and under the application of magnetic field the hf system demonstrates the lfl behavior , at elevated temperatures the system enters the transition region followed by the nfl behavior . our observations are in good agreement with experimental facts and show that the fermion condensation quantum phase transition is indeed responsible for the observed nfl behavior and quasiparticles survive both high temperatures and high magnetic fields . we grateful to v.a . khodel , k.g . popov , and v.a . stephanovich for valuable discussions . this work was partly supported by the rfbr # 09 - 02 - 00056 .

we estimate the global minimum variance ( gmv ) portfolio in the high - dimensional case using results from random matrix theory . this approach leads to a shrinkage - type estimator which is distribution - free and it is optimal in the sense of minimizing the out - of - sample variance . its asymptotic properties are investigated assuming that the number of assets depends on the sample size such that as tends to infinity . the results are obtained under weak assumptions imposed on the distribution of the asset returns , namely it is only required the fourth moments existence . furthermore , we make no assumption on the upper bound of the spectrum of the covariance matrix . as a result , the theoretical findings are also valid if the dependencies between the asset returns are described by a factor model which appears to be very popular in financial literature nowadays . this is also well - documented in a numerical study where the small- and large - sample behavior of the derived estimator are compared with existing estimators of the gmv portfolio . the resulting estimator shows significant improvements and it turns out to be robust to the deviations from normality . * estimation of the global minimum variance portfolio in high dimensions * + taras bodnar , nestor parolya and wolfgang schmid + + _ department of empirical economics ( econometrics ) , leibniz university hannover , d-30167 hannover , germany _ + _ department of statistics , european university viadrina , po box 1786 , 15207 frankfurt ( oder ) , germany _ jel classification : g11 , c13 , c14 , c58 , c65 + _ keywords _ : global minimum variance portfolio , large - dimensional asymptotics , covariance matrix estimation , random matrix theory .

1406.0437

since markowitz ( 1952 ) presented his seminal work about portfolio selection , this topic has become a very fast growing branch of finance . one of markowitz s ideas was the minimization of the portfolio variance subject to the budget constraint . this approach leads to the well - known and frequently used portfolio , the global minimum variance portfolio ( gmv ) . there is a great amount of papers dealing with the gmv portfolio ( see , e.g. , jagannathan and ma ( 2003 ) , ledoit and wolf ( 2003 ) , okhrin and schmid ( 2006 ) , kempf and memmel ( 2006 ) , bodnar and schmid ( 2008 ) , frahm and memmel ( 2010 ) among others ) . we remind that the gmv portfolio is the unique solution of the following optimization problem @xmath8 where @xmath9 denotes the vector of portfolio weights , @xmath10 is a suitable vector of ones , and @xmath11 stands for the covariance matrix of the asset returns . note that in our paper @xmath0 is a function of the sample size @xmath1 and thus the covariance matrix depends on @xmath1 as well . this is shown by the index @xmath1 . the solution of ( [ vmin ] ) is given by @xmath12 the gmv portfolio ( [ gmv ] ) has the smallest variance over all portfolios . it is also used in multi - period portfolio choice problems ( see , e.g. , brandt ( 2010 ) ) . although this portfolio possesses several nice theoretical properties , some problems arise when the uncertainty about the parameters of the asset return distribution is taken into account . indeed , we do not know the population covariance matrix in practice and , thus , it has to be suitably estimated . consequently , the estimation of the gmv portfolio is strongly connected with the estimation of the covariance matrix of the asset returns . the traditional estimator is a commonly used possibility for the estimation of the gmv portfolio ( [ gmv ] ) . this traditional estimator is constructed by replacing in ( [ gmv ] ) the covariance matrix @xmath11 by its sample counterpart @xmath13 . okhrin and schmid ( 2006 ) derived the distribution of the traditional estimator and studied its properties under the assumption that the asset returns follow a multivariate normal distribution , whereas kempf and memmel ( 2006 ) analyzed its conditional distributional properties . furthermore , bodnar and schmid ( 2009 ) derived the distribution of the main characteristics of the sample gmv portfolio , namely its variance and its expected return . the traditional estimator is not a bad choice if the number of assets @xmath0 is fixed and it is significantly smaller than the number of observations @xmath1 in the sample . this case is often used in statistics and it is called standard asymptotics ( see , le cam and yang ( 2000 ) ) . in that case the traditional estimator is a consistent estimator for the gmv portfolio and it is asymptotically normally distributed ( okhrin and schmid ( 2006 ) ) . as a result , for a small fixed dimension @xmath14 we can use the sample estimator but it is not fully clear what to do if the number of assets in the portfolio is extremely large , say @xmath15 , comparable to @xmath1 . here we are in the situation when both the number of assets @xmath0 and the sample size @xmath1 tend to infinity . this double asymptotics has an interpretation when @xmath0 and @xmath1 are of comparable size . more precisely , when @xmath16 tends to a concentration ratio @xmath17 . this type of asymptotics is known as high - dimensional asymptotics or `` kolmogorov '' asymptotics ( see , bai and silverstein ( 2010 ) ) . under the high - dimensional asymptotics the traditional estimator behaves very unpredictable and it is far from the optimum one . it tends to underestimate the risk ( see , el karoui ( 2010 ) , bai and shi ( 2011 ) ) . in general , the traditional estimator is worse for larger values of the concentration ratio @xmath18 . imposing the assumption of a factor structure on the asset returns this problem was resolved in an efficient way by bai ( 2003 ) , fan et al . ( 2008 ) , fan et al . ( 2012 ) , fan et al . ( 2013 ) , etc . nevertheless , if the factor structure is not present the question of high - dimensionality remains open . further estimators for the weights of the gmv portfolio have been proposed in this situation . demiguel et al . ( 2009 ) suggested to involve some additional portfolio constraints in order to avoid the curse of dimensionality . on the other hand , shrinkage estimators can be used which are biased but can significantly reduce the risk of the portfolio by minimizing its mean - square error . the general shrinkage estimator is a convex combination of the traditional estimator and a known target ( for the gmv portfolio it can be the naive equally weighted portfolio ) . they were first considered by stein ( 1956 ) . recently , various authors showed that shrinkage estimators for the portfolio weights indeed lead to better results ( see , e.g. , golosnoy and okhrin ( 2007 ) , frahm and memmel ( 2010 ) ) . in particular , golosnoy and okhrin ( 2007 ) considered a multivariate shrinkage estimator by shrinking the portfolio weights themselves but not the whole sample covariance matrix . the same idea was used by frahm and memmel ( 2010 ) who constructed a feasible shrinkage estimator for the gmv portfolio which dominates the traditional one . there are several problems with these estimators : first , the normal distribution is usually imposed ; second , dominating does not mean optimal ; and third , the large dimensional behavior ( large @xmath0 and large @xmath1 ) seems not to be acceptable . the aim of the paper is to derive a feasible and simple estimator for the gmv portfolio which is optimal , in some sense , for small and large sample sizes and which is distribution - free as well . for that purpose we construct an optimal shrinkage estimator , study its asymptotic properties and estimate unknown quantities consistently . the estimator is obtained using random matrix theory , a fast growing branch of probability theory . the main result of this theory was proved by mar@xmath19enko and pastur ( 1967 ) and further extended under very general conditions by silverstein ( 1995 ) . nowadays it is called mar@xmath19enko - pastur equation . its importance arises in many areas of science because it shows how the real covariance matrix and its sample estimate are connected with each other . knowing this information we can build suitable estimators for high - dimensional quantities . the rest of the paper is organized as follows . in section 2 we present a shrinkage estimator for the gmv portfolio which is optimal in terms of minimizing the out - of - sample variance . the asymptotic behavior of the resulting shrinkage intensity is investigated for @xmath20 in section 2.1 and in case of @xmath21 in section 2.2 where it is shown that the shrinkage intensity tends almost surely to a deterministic quantity when both the sample size and the portfolio dimension increase . this result allows us to determine an oracle estimator of the gmv portfolio , while the corresponding bona fide estimator is presented in section 2.3 . in section 3 we provide a simulation study for different values of @xmath22 and under various distributional assumptions imposed on the data generating process . here , the performance and the convergence rate of the derived shrinkage estimator are compared with existing estimators of the gmv portfolio . the results of our empirical study are given in section 4 where we apply the suggested estimator as well as the existing estimators to real data consisting of returns on assets included in the s&p 500 ( standard & poor s 500 ) index . section 5 summarizes all of the obtained results . the lengthy proofs are moved to the appendix ( section 6 ) . let @xmath23 be the @xmath24 data matrix which consists of @xmath1 vectors of the returns on @xmath25 assets . let @xmath26 and @xmath27 for @xmath28 . we assume that @xmath29 as @xmath30 . this type of limiting behavior is also denoted as `` large dimensional asymptotics '' or `` the kolmogorov asymptotics '' . in this case the traditional estimators perform poor or even very poor and tend to over / underestimate the unknown parameters of the asset returns , i.e. , the mean vector and the covariance matrix . throughout the paper it is assumed that it exists a @xmath31 random matrix @xmath32 which consists of independent and identically distributed ( i.i.d . ) real random variables with zero mean and unit variance such that @xmath33 it is noted that the observation matrix @xmath34 consists of dependent rows although its columns are independent . the assumption of the independence of the columns can further be weakened by controlling the growth of the number of dependent entries , while no specific distributional assumptions are imposed on the elements of @xmath34 ( see , friesen et al.(2013 ) ) . the two main assumptions which are used throughout the paper are 1 . the covariance matrix of the asset returns @xmath11 is a nonrandom @xmath0-dimensional positive definite matrix . the elements of the matrix @xmath32 have uniformly bounded @xmath35 moments for some @xmath36 . these two regularity conditions are very general and they fit many practical situations . the assumption ( a1 ) is common for financial and statistical problems . it does not impose a strong restriction on the data - generating process , whereas the assumption ( a2 ) is purely technical moreover , it seems to influence only the convergence rate of the proposed estimator ( see , e.g. rubio et al . ( 2012 ) ) . + the sample covariance matrix is given by @xmath37 where the symbol @xmath38 stands for the identity matrix of an appropriate dimension . the traditional estimator for the gmv portfolio is obtained by replacing the unknown population covariance matrix @xmath11 in by the estimator . this leads to @xmath39 next , we derive the optimal shrinkage estimator for the gmv portfolio weights by optimizing with respect to the shrinkage parameter @xmath40 and fixing some target portfolio @xmath41 . its distributional properties are studied after that . the general shrinkage estimator ( gse ) for @xmath42 is defined by @xmath43 where @xmath44 is a given nonrandom ( or random but independent of the actual observation vector @xmath45 , i.e. , the last column of @xmath34 ) vector . no assumption is imposed on the shrinkage intensity @xmath40 which is the object of our interest . the aim is to find the optimal shrinkage intensity @xmath40 . for a given target portfolio @xmath41 which minimizes the out - of - sample risk + @xmath46 ( see , e.g. , frahm and memmel ( 2010 ) , rubio et al . ( 2012 ) ) . the loss function ( [ risk ] ) can be rewritten as @xmath47 where @xmath48 is the population variance of the gmv portfolio and @xmath49 is known as the out - of - sample variance of the portfolio with the weights @xmath50 . using ( [ gse ] ) we want to solve the following optimization problem @xmath51 where @xmath52 is the out - of - sample variance of the traditional estimator for the gmv portfolio weights . taking the derivative of @xmath53 with respect to @xmath40 and setting it equal to zero we get @xmath54 from the last equation it is easy to find the optimal shrinkage intensity @xmath55 given by @xmath56 in order to ensure that @xmath55 is the minimizer of ( [ optm ] ) we calculate the second derivative of @xmath53 which has to be positive . it holds that @xmath57 almost surely . the last inequality is always true because of the positive definiteness of the matrix @xmath11 and the fact that @xmath58 with probability zero . in theorem [ th1 ] we show that the optimal shrinkage intensity @xmath59 is almost surely asymptotically equivalent to a nonrandom quantity @xmath60 $ ] under the large - dimensional asymptotics @xmath61 . let @xmath62 be the variance of the target portfolio and let @xmath63 be the relative loss of the target portfolio @xmath41 . [ th1 ] assume ( a1)-(a2 ) . let @xmath64 for all @xmath1 . then it holds that @xmath65 where @xmath66 is the limit of @xmath67 . additionally , the out - of - sample variance @xmath68 of the traditional estimator for the gmv portfolio possesses the following asymptotic behavior @xmath69 the proof of theorem [ th1 ] is given in the appendix . theorem [ th1 ] provides us important information about the optimal shrinkage estimator of the gmv portfolio . especially , the application of theorem [ th1 ] immediately leads to consistent estimators for @xmath59 , @xmath70 , and @xmath71 which are presented in section 2.3 below . it is remarkable to note that the assumption @xmath72 is natural for financial markets . it ensures that the population variance of the gmv portfolio has a lower bound which is in - line with the capital asset pricing model since the portfolio variance can not be smaller than the market risk ( see , e.g. , elton et al . ( 2007 , chapter 7 ) ) . moreover , the assumption of the boundedness of the variance of the target portfolio @xmath73 is also well acceptable because it makes no sense to shrink to a portfolio with infinite variance . most importantly , this condition also holds even if the largest eigenvalue of the covariance matrix is unbounded . such a situation is present if the asset returns follow a factor model which is a very popular approach in financial literature nowadays ( see , e.g. , fan et al . ( 2008 ) , fan et al . ( 2012 ) ) . it is worth pointing out that the same result is true if we assume instead of @xmath74 the boundedness of the spectral norm of the population covariance matrix , i.e. , the uniformly bounded maximum eigenvalue of @xmath11 . the answer on the question about the performance of the traditional and the optimal shrinkage estimator for the gmv portfolio is given in corollary [ cor1 ] . [ cor1 ] a. under the assumptions of theorem [ th1 ] , we get for the relative loss of the traditional estimator for the gmv portfolio @xmath75 b. under the assumptions of theorem [ th1 ] , we get for the relative loss of the optimal shrinkage estimator for the gmv portfolio @xmath76 corollary [ cor1 ] is a straightforward consequence of theorem [ th1 ] . moreover , its first part generalizes the result of frahm and memmel ( 2010 , theorem 7 ) to an arbitrary distribution of the asset returns . using corollary [ cor1](a ) we can plot the behavior of the relative loss of the traditional estimator for the gmv portfolio as a function of the concentration ratio @xmath18 only , while the relative loss of the optimal shrinkage portfolio additionally depends on the relative loss of the target portfolio . furthermore , from both parts of corollary [ cor1 ] we get @xmath77 i.e. , the relative loss of the optimal shrinkage estimator for the gmv portfolio can asymptotically be presented as a linear combination of the relative loss of the traditional estimator and the relative loss of the target portfolio . because @xmath78 as @xmath79 and @xmath80 denote the left and right limits to the point @xmath81 , respectively . ] and @xmath82 we get that @xmath83 as @xmath79 , whereas the relative loss of the traditional estimator tends to infinity . figure [ fig1 ] presents the behavior of the traditional and the proposed oracle estimators of the gmv portfolio weights for different values of @xmath84 . the covariance matrix @xmath11 is taken as a @xmath85-dimensional matrix where we have taken @xmath86 of the eigenvalues equal to @xmath87 , @xmath88 equal to @xmath81 , and @xmath88 equal to @xmath89 . the matrix of eigenvectors @xmath90 is generated from the haar distribution has a haar measure over the orthogonal matrices , then for any unit vector @xmath91 , @xmath92 has a uniform distribution over the unit sphere @xmath93 . ] the target portfolio is chosen as the equally weighted portfolio , i.e. @xmath94 . in the figure we observe that the asymptotic relative loss of the traditional estimator for the gmv portfolio has a singularity point at one . the loss of the traditional estimator is relatively small up to @xmath95 but thereafter , as @xmath96 , it rises hyperbolically to infinity . in contrast to the traditional estimator of the gmv portfolio weights , the suggested optimal shrinkage estimator has a constant asymptotic relative loss which is always smaller than @xmath89 . this result is in - line with the theoretical findings discussed around corollary [ cor1 ] . figures [ fig1 ] and [ fig2 ] above here in figure [ fig2 ] we show the asymptotic behavior of the optimal shrinkage intensity @xmath97 as a function of the concentration ratio @xmath98 . the target portfolio @xmath41 and the covariance matrix @xmath11 are the same as in figure [ fig1 ] . in the interval @xmath98 the optimal shrinkage intensity @xmath97 is a nonlinearly decreasing ( in a convex manner ) function of the concentration ratio @xmath18 . we observe that the optimal @xmath97 tends to zero as @xmath18 approaches one and , thus , in the limiting case the only optimal choice would be the target portfolio @xmath41 . in case @xmath21 , the sample covariance matrix @xmath13 is singular and its inverse does not exist anymore . thus , we first have to find a reasonable replacement for @xmath99 . for the oracle estimator of the gmv portfolio weights we use the following generalized inverse of the sample covariance matrix @xmath13 @xmath100 where @xmath101 denotes the moore - penrose inverse . it can be shown that @xmath102 is the generalized inverse satisfying @xmath103 and @xmath104 . is not equal to the moore - penrose inverse because it does not satisfy the conditions @xmath105 and @xmath106 . nevertheless , in section 2.3 , where the bona fide estimator is constructed , we use the moore - penrose inverse of @xmath13 instead of @xmath102 in order to obtain a valuable approximation . ] obviously , in case @xmath20 the generalized inverse @xmath102 coincides with the usual inverse @xmath99 . moreover , if @xmath11 is proportional to the identity matrix then @xmath102 coincides with the moore - penrose inverse @xmath107 calculated for @xmath13 . it has also to be noted that @xmath102 can not be determined in practice since it depends on the unknown matrix @xmath11 . in this section , it is only used to determine an oracle estimator for the weights of the gmv portfolio , whereas the bona fide estimator is constructed in section 2.3 . based on @xmath102 in , the oracle traditional estimator for the gmv portfolio in case @xmath21 is first constructed and it is given by @xmath108 next , we determine the oracle optimal shrinkage estimator for the gmv portfolio weights expressed as @xmath109 similarly to section 2.1 , we deduce the optimal shrinkage intensity @xmath110 given by @xmath111 where @xmath112 is the oracle out - of - sample variance of the traditional estimator for the gmv portfolio . in theorem [ th12 ] we present the asymptotic properties of the optimal @xmath110 for @xmath21 . [ th12 ] assume ( a1)-(a2 ) . let @xmath64 for all @xmath1 . then it holds that @xmath113 where @xmath66 is the limit of @xmath67 . additionally , we get for the oracle out - of - sample variance @xmath114 of the traditional estimator for the gmv portfolio @xmath115 the proof of theorem [ th12 ] is given in the appendix . the asymptotic behavior of the relative loss calculated for the traditional oracle estimator of the gmv portfolio as well as for the oracle optimal shrinkage estimator is described in corollary [ cor2 ] . [ cor2 ] a. under the assumptions of theorem [ th12 ] , we get for the relative loss of the oracle traditional estimator for the gmv portfolio @xmath116 b. under the assumptions of theorem [ th12 ] , we get for the relative loss of the oracle optimal shrinkage estimator for the gmv portfolio @xmath117 similarly to the case @xmath20 , the relative loss of the optimal shrinkage estimator for the gmv portfolio is a linear combination of the relative loss of the traditional estimator and the relative loss of the target portfolio . furthermore , if @xmath118 , the relative loss of the traditional estimator tends to infinite is ill - behaved and not invertible at the point @xmath119 because in that case its smallest eigenvalue is very near to zero . ] , whereas for the relative loss of the shrinkage estimator we get @xmath120 which is bounded from above by @xmath121 , i.e. , it is finite . figure [ fig3 ] presents the asymptotic performance of the oracle traditional estimator and of the proposed oracle optimal shrinkage estimator for the gmv portfolio in case @xmath21 . a considerable improvement is present when the oracle optimal shrinkage estimator is applied where the average loss is always smaller than @xmath81 . in contrast , the average loss of the oracle traditional estimator possesses always larger values with a minimum of about @xmath122 reached around @xmath123 . figures [ fig3 ] and [ fig4 ] above here figure [ fig4 ] presents the asymptotic behavior of the optimal shrinkage intensity @xmath124 in case @xmath21 which is no longer a monotonic function of the concentration ratio @xmath18 as it is observed in figure [ fig2 ] . the optimal shrinkage intensity attains its maximum close to @xmath123 . moreover , @xmath124 remains positive even for large values of @xmath18 , i.e. the oracle optimal shrinkage estimator converges to @xmath41 for @xmath125 much slower as it was for @xmath126 . on the other hand , it converges to @xmath41 pretty fast for @xmath127 . as a result , we do nt have to expect the instability of the proposed shrinkage estimator neither in the neighborhood of @xmath119 nor for @xmath128 . in this subsection we show how the derived oracle estimators in case @xmath20 and @xmath21 , respectively , can be consistently estimated . this is achieved by estimating consistently the relative loss of the target portfolio @xmath67 . this result is presented in theorem [ th2 ] . [ th2 ] under the assumptions ( a1)-(a2 ) a consistent estimator of @xmath67 is given by @xmath129 the proof of theorem [ th2 ] is given in the appendix . applying theorems [ th1 ] and [ th2](a ) allows us to determine the _ bona fide _ estimator for the gmv portfolio weights in case @xmath98 . it is given by @xmath130 where @xmath131 is given above in ( [ estr ] ) . the expression ( [ bonafide ] ) presents the optimal shrinkage estimator for a given target portfolio @xmath41 because the shrinkage intensity @xmath132 tends almost surely to its optimal value @xmath97 for @xmath133 as @xmath134 the situation is more complicated in case @xmath21 . here , the quantity @xmath131 is not a bona fide estimator of the relative loss of the target portfolio , since the matrix @xmath102 depends on unknown quantities . for that reason we propose a reasonable approximation via the the application of the moore - penrose inverse @xmath107 . it is easy to verify that in case of @xmath135 equality holds . furthermore , both the extensive simulation study of section 3 and the empirical investigations of section 4 document that this approximation does a very good job even for dense population covariance matrix @xmath11 . the reason of this behavior could be the point that @xmath107 possesses a similar asymptotic behavior as @xmath102 . however , it is a very challenging mathematical problem to prove this result analytically and we leave this for the future research . in figure [ fig : moore ] we provide a short simulation with the same design as presented in figure [ fig2 ] in order to show that @xmath136 and @xmath137 are close asymptotically and justify the accuracy of our approximation . figures [ fig : moore ] above here taking into account the above discussion and the result of theorem [ th2 ] ( b ) , the bona fide estimator of the quantity @xmath66 in case @xmath21 is approximated by @xmath138 the application of leads to the bona fide optimal shrinkage estimator of the gmv portfolio in case @xmath21 expressed as @xmath139 where @xmath140 is the moore - penrose pseudo - inverse of the sample covariance matrix @xmath13 . it is noted that the estimator ( [ bonafide ] ) is the optimal estimator of the gmv portfolio for @xmath20 in terms of minimizing the out - of - sample variance , while the estimator ( [ bonafide+ ] ) is a suboptimal one in case @xmath21 . in order to summarize this section , we merge ( [ bonafide ] ) and ( [ bonafide+ ] ) into one _ bona fide _ optimal shrinkage estimator for the gmv portfolio weights in case @xmath17 given by is not theoretically handled but using the moore - penrose inverse and setting equal to zero the smallest eigenvalue we are still able to construct a feasible estimator in this situation . ] @xmath141 @xmath142 and @xmath143 where we use that @xmath144 if @xmath13 is nonsingular . in figure [ fig5 ] we investigate the difference between the oracle and the bona fide optimal shrinkage estimators for the gmv portfolio weights as well as between the oracle and the bona fide traditional estimators . the population covariance matrix is taken as a dense @xmath145-dimensional covariance matrix @xmath11 with @xmath146 of eigenvalues equal to @xmath147 , @xmath148 to @xmath149 , and last @xmath148 to @xmath150 . the eigenvectors are chosen in the same way as in the section about oracle estimator . the target portfolio is still the naive one , i.e. , @xmath94 . the observation matrix is generated from the normal distribution . a perfect fit of the bona fide optimal shrinkage estimator ( dotted red line ) to its oracle ( solid red line ) is observed for all of the considered values @xmath17 . the blue lines corresponds to the oracle traditional estimator ( solid blue line ) and the bona fide traditional estimator ( dash blue line ) . in contrast to the optimal shrinkage estimator , a difference between the bona fide traditional estimator and its oracle is present for @xmath21 which increases as @xmath18 becomes larger . for @xmath20 both the estimators coincide since in this case both the generalized inverse and the moore - penrose inverse are equal to the inverse of the sample covariance matrix . it is remarkable that the proposed bona fide optimal shrinkage estimator works well also at point @xmath119 although the corresponding oracle estimator is even not defined there . the reason is that we just set equal to zero the smallest eigenvalue of @xmath13 and use the moore - penrose inverse technique . the results of figure [ fig5 ] motivate the application of the moore - penrose inverse instead of the generalized inverse given at the beginning of section 2.2 in practice , whereas the traditional estimator should be used with care . we provide a further investigation of this point in the simulation study of section 3 . figure [ fig5 ] above here the last point , which has to be noted , is that the bona fide estimator ( [ bonafide ] ) is easy to use in practice since it can be fast computed . the target portfolio @xmath41 plays a crucial role in the determination of the optimal shrinkage estimator . the most obvious choice of @xmath41 would be the naive portfolio @xmath151 or a sparse portfolio . in the multi - period setting the weights of the previous period can be chosen as a target portfolio . theoretically , we can even take a random target portfolio but it should be independent of the actual observations . in particular , it can be a uniformly distributed random vector on the unit sphere ( suitably normalized ) or a uniformly distributed random vector on the simplex . choosing the optimal portfolio weights of the previous periods leads to more interesting example for a target portfolio which allows us to construct some sort of bayesian updating principle in the dynamic setting . in general , the answer on this question depends on the underlying data because the choice of the target weights is equivalent to the choice of the hyperparameter for the prior distribution of @xmath152 . this problem is well - known in bayesian statistics . the application of different priors leads to different results . so it is very important to choose the one which works well in most cases . the naive one is the equally weighted portfolio @xmath153 . obviously , the oracle shrinkage estimator with the prior weights as the true global minimum variance portfolio is a consistent estimator as shown in proposition [ prior_prop ] . moreover , including some new information about the true gmv portfolio into the prior can lead to a significant increase of performance ( see , bodnar et al . ( 2014 ) ) . for simplicity we take the naive portfolio in our simulation study in section 3 as well as in the empirical investigation of section 4 . consider the shinkage estimator as a vector function @xmath154 , where @xmath155 and @xmath156 are the @xmath0-dimensional vector spaces . in the following proposition we present some properties of the shrinkage estimator as a function of the target weights @xmath41 . [ prior_prop ] for the proposed shrinkage estimator @xmath157 it holds that 1 . @xmath158 is a consistent estimator for the gmv portfolio if the population covariance matrix @xmath159 for arbitrary @xmath160 and @xmath22 . 2 . @xmath161 is a consistent estimator for the gmv portfolio @xmath162 for all @xmath22 . the proof is a straightforward application of theorem [ th1 ] , theorem [ th12 ] and theorem [ th2 ] . in this section we demonstrate how the obtained results can be applied in practice . the first part of our simulations is dedicated to normally distributed data , while in the second part the asset returns are generated from the @xmath163-distribution with @xmath149 degrees of freedom . the target portfolio @xmath41 is taken as the naive portfolio @xmath164 . the results are presented in both cases @xmath20 and @xmath21 as well as for the covariance matrix with bounded ( section 3.1 ) and unbounded ( section 3.2 ) spectrum . the benchmark estimator is the dominating estimator of the gmv portfolio suggested by frahm and memmel ( 2010 ) . it is given by @xmath165 where @xmath166 is the estimated relative loss of the naive portfolio . the dominating estimator ( [ fm ] ) is derived under the assumption that the asset returns are normally distributed and it dominates over the traditional estimator in terms of the out - of - sample variance ( cf . frahm and memmel ( 2010 ) ) . nevertheless , it is not clear how far it is away from the optimal one for different values of the concentration ratio @xmath17 . its behavior for non - normally distributed data has not been studied yet as well . next , we compare the performance of the dominating estimator ( [ fm ] ) with the bona fide optimal shrinkage estimator ( [ bonafide ] ) . in order to find out the rates of convergence established in theorem [ th1 ] and [ th2 ] , we also consider the oracle optimal shrinkage estimator which can be easily constructed for @xmath20 and @xmath21 with the optimal shrinkage intensities given by ( [ alfa ] ) and ( [ alfa+ ] ) , respectively . as a performance measure we take the relative loss from section 2 . for an arbitrary estimator @xmath167 of the gmv portfolio it is defined by @xmath168 where @xmath169 and @xmath170 . in our simulation study we take @xmath0 as a function of @xmath1 . in particular , when @xmath171 and @xmath172 for @xmath173 $ ] the concentration ration @xmath18 is always equal to @xmath89 and @xmath0 increases together with @xmath1 exponentially . that is why the small dimensions are presented with more points and the large ones with less . similar choices of @xmath0 and @xmath1 are also performed for other values of @xmath174 . finally , it is noted that the simulation results show a good convergent rate in terms of the relative loss for the bona fide optimal shrinkage estimator to its oracle one already for @xmath175 . in this subsection , we assume that the covariance matrix possesses a bounded spectrum , i.e. with bounded maximum eigenvalue . here , we use the structure of the covariance matrix as in figure [ fig5 ] , i.e. , we take @xmath146 of its eigenvalues equal to @xmath147 , @xmath148 equal to @xmath149 and @xmath148 equal to @xmath150 . the high - dimensional covariance matrices constructed in this way possess uniformly the same spectral norm and their eigenvalues are not very dispersed . additionally , this choice of the covariance matrices ensures that when the dimension @xmath0 increases then the spectrum of the covariance matrices does not change its behavior . figures [ fig:6 ] and [ fig:7 ] above here in figures [ fig:6 ] and [ fig:7 ] we present the simulation results for normally distributed data and different values of the concentration ratio @xmath176 . figure [ fig:6 ] presents the _ global _ behavior of the considered estimators for different dimensions @xmath0 , while figure [ fig:7 ] shows the _ local _ distributional properties for a value of fixed @xmath177 . more precisely , under the _ global _ behavior we understand the evolution of the average relative loss with respect to the dimension @xmath0 and the _ local _ behavior presents the empirical cumulative distribution functions ( e.c.d.f . ) of the relative loss for one fixed value of @xmath0 , namely @xmath177 . the comparison in case of _ global _ setting is clear : the smaller the average loss the better is the estimator . the _ local _ study provides a more precise comparison in terms of the empirical distributions . in this case , the criterion of the best estimator is based on the observation that the e.c.d.f . with stochastically smaller values is dominating . this means that for two e.c.d . functions , the dominating one is placed on the left side from the other . this criterion is consistent with the stochastic dominance of order one . the only difference with respect to the stochastic dominance of order one is that the comparison is based on the empirical distribution functions instead of the population one . on global analysis we see that the bona fide optimal shrinkage estimator converges to the corresponding oracle one already for small values of @xmath0 in all of the considered cases @xmath178 . on the third place , the dominating estimator of frahm and memmel ( 2010 ) is ranked . it is always better than the traditional estimator which is the worst one , but it is always worse than the other two competitors . in terms of the values of the average relative loss , we observe that the difference between the estimators become more significant if @xmath18 increases and lies below @xmath179 . for instance , in case of @xmath180 , the average relative loss of the traditional estimator tends to @xmath146 , whereas it tends to @xmath81 for @xmath181 . these two results are in line with corollary [ cor1 ] , where it is proved that the average loss of the traditional estimator tends to @xmath182 under high - dimensional asymptotics . in case of @xmath183 , the difference between the average relative loss of the optimal shrinkage estimator and the dominating ( traditional ) estimators becomes very large . indeed , in this case the traditional estimator has an average relative loss which is asymptotically equal to @xmath184 . this means that the out - of - sample risk of the traditional estimator is @xmath150 times larger than the real risk . the dominating estimator clearly overperforms the traditional one but the relative loss is close to @xmath122 for small dimensions ( @xmath185 ) which means that its out - of - sample risk is @xmath149 times as large as the real risk . this is not acceptable anymore . in contrast , the _ bona fide _ optimal shrinkage estimator converges fast to its _ oracle _ one . the relative loss of the optimal shrinkage estimator is smaller than @xmath186 . figure [ fig:7 ] shows the same dominance in terms of the empirical distribution functions for a local analysis in the case @xmath177 . the best approaches are the _ oracle _ and the _ bona fide _ optimal shrinkage estimators . next , the dominating estimator is ranked followed by the traditional one . the plots also illustrate the fast convergence of the _ bona fide _ optimal shrinkage estimator to its oracle . the local analysis for @xmath177 confirms the almost sure convergence ( consistence ) of the _ bona fide _ optimal shrinkage estimator which is proved in theorem [ th1 ] . in figure [ fig:7 ] the relative risk of both the bona fide and the oracle optimal shrinkage estimators possesses a very small variance which vanishes when the dimension @xmath0 increases . at the same time , the dominating estimator possesses a significantly larger variance and it is unstable when @xmath18 is close to one . the traditional estimator shows a very crucial behavior and it is the worst one among the considered estimators . the most interesting situation is observed for @xmath187 in figures [ fig:6 ] and [ fig:7 ] which corresponds to the singular sample covariance matrix @xmath13 . here , we apply the results from section [ secc1 ] and [ bonaest ] and take the moore - penrose inverse @xmath107 instead of @xmath99 . note that we can not use the dominating estimator because it is not applicable for @xmath21 . the results are still impressing for both the global and the local regimes . again , the proposed bona fide optimal shrinkage estimator converges to its oracle . as a traditional estimator , we take the gmv portfolio constructed by using the moore - penrose inverse @xmath107 . the traditional estimator possesses a rapidly increasing average loss and the largest variance . it is not an acceptable estimator also for @xmath21 . in contrast , the bona fide optimal shrinkage estimator has a small variance and obeys a stable behavior even if @xmath21 . further we analyze the behavior of the considered estimators when the asset returns are no longer normally distributed . in particular it is interesting to study how strong is the impact of heavy tails on the estimators derived in the paper . for this reason , the @xmath163-distribution with @xmath149 degrees of freedom is used next in our simulation study . recently , authors have mentioned that @xmath149 degrees of freedom seems to be a suitable choice in practice ( see , venables and ripley ( 2002 ) ) . figures [ fig:8 ] and [ fig:9 ] above here in figures [ fig:8 ] and [ fig:9 ] we present the results for the @xmath163-distributed asset returns with @xmath149 degrees of freedom . the structure of the comparison study is the same as in case of normally distributed data . in general , the behavior observed in figures [ fig:8 ] and [ fig:9 ] does not differ significantly from those obtained for the normal distribution . the best estimator is , as usual , the proposed shrinkage estimator . the optimal shrinkage estimator dominates clearly other competitors over all @xmath176 . it is noted that the convergence rate of the _ bona fide _ optimal shrinkage estimator to its _ oracle _ is not effected by the presence of heavy tails . a similar asymptotic relative loss behavior for the optimal shrinkage estimator is established , i.e. , the average relative loss is asymptotically constant and it is smaller than @xmath89 . the traditional estimator possesses the worst behavior over all @xmath18 and @xmath0 . in this subsection we assume that the largest eigenvalue of the population covariance matrix @xmath11 increases as @xmath188 when @xmath189 . thus , the following structure of @xmath11 is considered here , namely @xmath146 of eigenvalues equal to @xmath147 , @xmath148 equal to @xmath149 , @xmath190 equal to @xmath150 and the last eigenvalue is equal to @xmath0 . + note that this structure corresponds to the case when a factor structure on the asset returns is imposed . the factor model can reduce significantly the number of dimensions so that the estimators do not suffer from the `` curse of dimensionality '' anymore ( see , e.g. , fan et al . ( 2013 ) ) . figures [ fig:10 ] to [ fig:13 ] above here in figures [ fig:10 ] to [ fig:13 ] we present the behavior of the estimators considered in the paper in case of a covariance matrix with unbounded spectrum . it is remarkable to note that the results are not very different from those obtained in case of a covariance matrix @xmath11 with bounded spectrum . the only difference is a somewhat greater variance of the estimators . on the other hand , the dominance behavior as well as the convergence rate of the bona fide optimal shrinkage estimator to its oracle is not effected by the largest eigenvalue of the population covariance matrix . this means that the proposed estimator is still applicable if the asset returns follow a factor model . even more , it does not lose its efficiency also in case @xmath21 . at the end , we note that , for the sake of interest , we have also simulated the @xmath163-distribution with @xmath87 degrees of freedom for both the bounded and the unbounded spectra . this change effects only the convergence rate but not the dominance behavior . in our theoretical framework we require the existence of the @xmath122th moment but the simulation study shows that this assumption can be relaxed or conjectured to be relaxed . as a result , the proposed optimal shrinkage procedure assures the efficiency in many important practical cases and , thus , can be applied in many real life situations . nevertheless , the empirical back - testing is still needed in order to check the behavior of the derived estimator for the gmv portfolio weights on a real data set . this is done is the next section . in this section , we apply the proposed optimal shrinkage estimator for the gmv portfolio ( [ bonafide ] ) to real data which consist of daily returns on the 417 assets listed in the s@xmath191p @xmath192 ( standard @xmath191 poor s @xmath192 ) index and traded during the period from 22.04.2013 to 19.03.2014 . it corresponds to the horizon of @xmath193 trading days . the s&p 500 index is based on the market capitalizations of 500 large companies having common stock listed on the nasdaq . in this empirical study we compare the performance of the derived optimal shrinkage estimator for the gmv portfolio weights given by ( [ bonafide ] ) with the traditional estimator and the dominating estimator suggested by frahm and memmel ( 2010 ) . the comparison is based a procedure which is similar to the rolling - window approach proposed by demiguel et al . in particular , we randomly pick up a portfolio of dimension @xmath194 from all @xmath195 portfolios and estimate the portfolio weights for the given estimation window of the length @xmath196 . we repeat this rolling - window procedure for the next step by including data of the next day and dropping out the data of the last day until the end of the data set is reached . the estimation window @xmath1 is chosen such that the concentration ratio @xmath197 lies in the set @xmath198 . in order to compare the performance of the estimators we consider the out - of - sample variance and the out - of - sample sharpe ratio . let @xmath199 be an estimator for the gmv portfolio which is based on the window with last observation at time @xmath163 and let @xmath200 be the vector of the asset returns for the next period @xmath201 . then the out - of - sample variance and the out - of - sample sharpe ratio are calculated by @xmath202 in order to measure the statistical significance we sample randomly @xmath203 different portfolios and calculate the e.c.d . functions of their out - of - sample variances and the corresponding out - of - sample sharpe ratios . the best strategy is chosen similarly to the stochastic dominance principle , i.e. we look for the strategy whose e.c.d.f . stochastically dominates the other ones . however , the dominance is defined differently in case of the out - of - sample variance and the out - of - sample sharpe ratio . for the out - of - sample variance , the e.c.d.f . of the best strategy should lie above the e.c.d . functions of the other competitors , i.e. larger values of the out - of - sample variance can take place with smaller probability . in contrast , criteria based on the out - of - sample sharpe ratio prefers the strategy whose e.c.d.f . lies below the other e.c.d . functions . in this case , the gmv portfolio constructed using the corresponding estimator would possess the highest out - of - sample sharpe ratio . figures [ fig : ecdf_var ] and [ fig : ecdf_sr ] above here in figures [ fig : ecdf_var ] and [ fig : ecdf_sr ] the e.c.d . functions of the out - of - sample variance and of the out - of - sample sharpe ratio are presented for three estimators of the gmv portfolio , namely for the optimal shrinkage estimator , the traditional estimator and the dominating estimator suggested by frahm and memmel ( 2010 ) . because , the dominating estimator can be constructed only in case @xmath20 , we drop it for @xmath204 and @xmath123 . in all of the considered cases , we observe a very good performance of the optimal shrinkage estimator . it overperforms the other estimation strategies for both considered criteria . the corresponding e.c.d.f . lies above the other e.c.d . functions in case of the out - of - control variance , whereas it is below the e.c.d . functions of other competitors in case of the out - of - sample sharpe ratio . on the second place , we rank the dominating estimator of frahm and memmel ( 2010 ) which is always better than the traditional estimator . the global minimum variance portfolio plays an important role in investment theory and practice . this portfolio is widely used as an investment opportunity in both static and dynamic optimal portfolio choice problems . although an explicit analytical expression for the structure of the gmv portfolio weights is available in literature , the estimation of the gmv portfolio appears to be a very challenging problem , especially for high - dimensional data . we deal with this problem in the present paper by deriving a feasible and robust estimator for the weights of the gmv portfolio when the distribution of the asset returns is not prespecified and no market structure is imposed . we construct an optimal shrinkage estimator for the gmv portfolio which is optimal in the sense of minimizing the out - of - sample variance . an analytical expression for the shrinkage intensity is obtained which appears to be a complicated function of the data and the parameters of the asset return distribution . we deal with the later problem by determining an asymptotically equivalent quantity of the shrinkage intensity under high - dimensional asymptotics . we estimate this asymptotically equivalent function consistently by applying recent results from random matrix theory . this is achieved under very weak assumptions imposed on the distribution of the asset returns . namely , we only require the existence of the fourth moment , whereas no explicit distributional assumption is imposed . moreover , our findings are still valid in both cases @xmath20 and @xmath21 as well as if the spectrum of the population covariance matrix is bounded or unbounded . as a result , the suggested method can be applied to heavy - tailed distributed asset returns as well as to asset returns whose dynamics can be modeled by a factor model which is a very popular approach in financial and econometric literature . finally , using simulated and real data , we compare the optimal shrinkage estimator for the gmv portfolio with existing ones . the theoretical findings as well as the results of the monte carlo simulations and an empirical study show that the suggested estimator for the gmv portfolio weights dominates the existing estimators in case @xmath17 . here the proofs of the theorems are given . first , we point out that for our purposes @xmath13 can be well approximated by @xmath205 since the matrix @xmath206 has rank one and , consequently , it does not influence the asymptotic behavior of the spectrum of the sample covariance matrix ( see , bai and silverstein ( 2010 ) , theorem a.44 ) . [ lem1 ] assume ( a1 ) and ( a2 ) . let a nonrandom @xmath207-dimensional matrix @xmath208 possess a uniformly bounded trace norm ( sum of singular values ) and let @xmath209 . then it holds that @xmath210 where @xmath211 * proof of lemma [ lem1 ] : * the application of theorem 1 in rubio and mestre ( 2011 ) leads to ( [ rm2011_id ] ) where @xmath212 is a unique solution in @xmath213 of the following equation @xmath214 the two solutions of ( [ eq1-lemma6_1 ] ) are given by @xmath215 in order to decide which of the two solutions is feasible , we note that @xmath216 is the stieltjes transform with a positive imaginary part . thus , without loss of generality , we can take @xmath217 and get @xmath218 which is positive only if the sign @xmath219 is chosen . hence , the solution is given by @xmath220 lemma [ lem1 ] is proved . + rubio and mestre ( 2011 ) studied the asymptotics of the functionals @xmath221 for a deterministic matrix @xmath222 with bounded trace norm at infinity . note that the results of theorem 1 of rubio and mestre ( 2011 ) also hold under the weaker assumption of the existence of the @xmath122th moments . this statement is obtained by using lemma b.26 of bai and silverstein ( 2010 ) on quadratic forms which we cite for presentation purposes as lemma [ lem2 ] below . [ lem2]*[lemma b.26 , bai and silverstein ( 2010 ) ] * let @xmath223 be a @xmath207 nonrandom matrix and let @xmath224 be a random vector with independent entries . assume that @xmath225 , @xmath226 , and @xmath227 . then , for any @xmath228 , @xmath229 where @xmath230 is some constant which depends only on @xmath231 . in order to obtain the statement of theorem 1 of rubio and mestre ( 2011 ) under the weaker assumption imposed on the moments , we replace lemma 2 of rubio and mestre ( 2011 ) by lemma [ lem2 ] in the case of @xmath228 . note that lemma 2 of rubio and mestre ( 2011 ) holds for @xmath232 while lemma [ lem2 ] is a more stronger result because it holds also in the case @xmath233 . this is the main trick which implies that lemma 3 of rubio and mestre ( 2011 ) holds also for @xmath228 ( instead of @xmath232 ) . lemma 4 of rubio and mestre ( 2011 ) has already been proved under the assumption that there exist @xmath35 moments . the last step is the application of lemma 1 , 2 and 3 of rubio and mestre ( 2011 ) with @xmath228 . finally , it can be easily checked that the further steps of the proof of theorem 1 of rubio and mestre ( 2011 ) hold under the existence of @xmath35 moments . in order to save space we leave the detailed technical proof of this assertion to the reader . * proof of theorem [ th1 ] : * let us recall the optimal shrinkage intensity expressed as @xmath234 it holds that @xmath235\nonumber\\ & = & \lim\limits_{z\rightarrow0^+}\text{tr}\left[\left(\dfrac{1}{n}{\mathbf{x}}_n{\mathbf{x}}^\prime_n - z{\mathbf{i}}\right)^{-1}{{\mbox{\boldmath$\sigma$}}}^{-\frac{1}{2}}_n{\mathbf{1}}{\mathbf{1}}^\prime{{\mbox{\boldmath$\sigma$}}}^{-\frac{1}{2}}_n\right ] \label{th31_eq1}\\ { \mathbf{1}}^\prime{\mathbf{s}}_n^{-1}{{\mbox{\boldmath$\sigma$}}}_n\mathbf{b}_n&=&\lim\limits_{z\rightarrow0^+}\text{tr}\left[({\mathbf{s}}_n - z{{\mbox{\boldmath$\sigma$}}}_n)^{-1}{{\mbox{\boldmath$\sigma$}}}_n\mathbf{b}_n{\mathbf{1}}^\prime\right]\nonumber\\ & = & \lim\limits_{z\rightarrow0^+}\text{tr}\left[\left(\dfrac{1}{n}{\mathbf{x}}_n{\mathbf{x}}^\prime_n - z{\mathbf{i}}\right)^{-1}{{\mbox{\boldmath$\sigma$}}}^{\frac{1}{2}}_n \mathbf{b}_n{\mathbf{1}}^\prime{{\mbox{\boldmath$\sigma$}}}^{-\frac{1}{2}}_n\right ] \label{th31_eq2}\\ { \mathbf{1}}^\prime{\mathbf{s}}_n^{-1}{{\mbox{\boldmath$\sigma$}}}_n{\mathbf{s}}_n^{-1}{\mathbf{1}}&= & \left.\dfrac{\partial}{\partial z}\text{tr}\left[\left(\frac{1}{n}{\mathbf{s}}_n - z{{\mbox{\boldmath$\sigma$}}}_n\right)^{-1}{\mathbf{1}}{\mathbf{1}}^\prime\right]\right|_{z=0}\nonumber\\ & = & \left.\dfrac{\partial}{\partial z}\text{tr}\left[\left(\frac{1}{n}{\mathbf{x}}_n{\mathbf{x}}^\prime_n - z{\mathbf{i}}\right)^{-1 } { { \mbox{\boldmath$\sigma$}}}^{-\frac{1}{2}}_n{\mathbf{1}}{\mathbf{1}}^\prime{{\mbox{\boldmath$\sigma$}}}^{-\frac{1}{2}}_n\right]\right|_{z=0 } \label{th31_eq3}\,.\end{aligned}\ ] ] let @xmath236 \quad \text{with } \quad \mathbf{\theta}_{\xi}={{\mbox{\boldmath$\sigma$}}}^{-\frac{1}{2}}_n{\mathbf{1}}{\mathbf{1}}^\prime{{\mbox{\boldmath$\sigma$}}}^{-\frac{1}{2}}_n\ ] ] and @xmath237 \quad \text{with } \quad \mathbf{\theta}_{\zeta}={{\mbox{\boldmath$\sigma$}}}^{\frac{1}{2}}_n\mathbf{b}_n{\mathbf{1}}^\prime{{\mbox{\boldmath$\sigma$}}}^{-\frac{1}{2}}_n\,.\ ] ] both the matrices @xmath238 and @xmath239 possess a bounded trace norm since @xmath240 and @xmath241 finally , using the equality @xmath246 we get @xmath247\right|=|\xi_n(z)-(1-c)^{-3}{\mathbf{1}}^\prime { { \mbox{\boldmath$\sigma$}}}^{-1}_n{\mathbf{1}}| \stackrel{a.s.}{\longrightarrow}0\ ] ] for @xmath248 as @xmath30 . consequently , @xmath249 ) and ( [ th31_eq9 ] ) leads to @xmath250 whereas additionally using ( [ th31_eq7 ] ) we get @xmath251 with @xmath252 for @xmath248 as @xmath30 . the quantity @xmath66 is the limit of @xmath67 which exists due to assumption @xmath253 and @xmath254 . these two equalities complete the proof of theorem [ th1 ] . * proof of theorem [ th12 ] : * in case of @xmath21 , the optimal shrinkage intensity is given by @xmath255 let @xmath256 and @xmath257 . using the definition of @xmath102 given in and the equality @xmath258 , we get @xmath259 = \text{tr}\left[{\mathbf{x}}_n\left({\mathbf{x}}_n^\prime{\mathbf{x}}_n\right)^{-2}{\mathbf{x}}_n^\prime\mathbf{\theta}_{\xi}\right ] \nonumber\\ & = & \left.\dfrac{\partial}{\partial z}\text{tr}\left[{\mathbf{x}}_n\left({\mathbf{x}}_n^\prime{\mathbf{x}}_n - z{\mathbf{i}}_n\right)^{-1}{\mathbf{x}}_n^\prime\mathbf{\theta}_{\xi}\right]\right|_{z=0}\\ { \mathbf{1}}^\prime{\mathbf{s}}_n^{*}{{\mbox{\boldmath$\sigma$}}}_n\mathbf{b}_n&=&\text{tr}\left[\left({\mathbf{x}}_n{\mathbf{x}}_n^\prime\right)^+\mathbf{\theta}_{\zeta}\right ] = \text{tr}\left[{\mathbf{x}}_n\left({\mathbf{x}}_n^\prime{\mathbf{x}}_n\right)^{-2}{\mathbf{x}}_n^\prime\mathbf{\theta}_{\zeta}\right]\nonumber\\ & = & \left.\dfrac{\partial}{\partial z}\text{tr}\left[{\mathbf{x}}_n\left({\mathbf{x}}_n^\prime{\mathbf{x}}_n - z{\mathbf{i}}_n\right)^{-1}{\mathbf{x}}_n^\prime\mathbf{\theta}_{\zeta}\right]\right|_{z=0}\\ { \mathbf{1}}^\prime{\mathbf{s}}_n^{*}{{\mbox{\boldmath$\sigma$}}}_n{\mathbf{s}}_n^{*}{\mathbf{1}}&= & \text{tr}\left[\left(\left({\mathbf{x}}_n{\mathbf{x}}_n^\prime\right)^+\right)^{-2}\mathbf{\theta}_{\zeta}\right ] = \text{tr}\left[{\mathbf{x}}_n\left({\mathbf{x}}_n^\prime{\mathbf{x}}_n\right)^{-3}{\mathbf{x}}_n^\prime\mathbf{\theta}_{\xi}\right]\nonumber\\ & = & \frac{1}{2}\left.\dfrac{\partial^2}{\partial z^2}\text{tr}\left[{\mathbf{x}}_n\left({\mathbf{x}}_n^\prime{\mathbf{x}}_n - z{\mathbf{i}}_n\right)^{-1}{\mathbf{x}}_n^\prime\mathbf{\theta}_{\xi}\right]\right|_{z=0}\,.\end{aligned}\ ] ] the application of the woodbury formula ( matrix inversion lemma , see , e.g. , horn and johnson ( 1985 ) ) @xmath260 leads to @xmath261\right|_{z=0}\\ { \mathbf{1}}^\prime{\mathbf{s}}_n^{*}{{\mbox{\boldmath$\sigma$}}}_n\mathbf{b}_n&=&\left.\dfrac{\partial}{\partial z } z \text{tr}\left[\left({\mathbf{x}}_n{\mathbf{x}}_n^\prime - z{\mathbf{i}}_p\right)^{-1}\mathbf{\theta}_{\zeta}\right]\right|_{z=0}\\ { \mathbf{1}}^\prime{\mathbf{s}}_n^{*}{{\mbox{\boldmath$\sigma$}}}_n{\mathbf{s}}_n^{*}{\mathbf{1}}&= & \frac{1}{2}\left.\dfrac{\partial^2}{\partial z^2}z\text{tr}\left[\left({\mathbf{x}}_n{\mathbf{x}}_n^\prime - z{\mathbf{i}}_p\right)^{-1}\mathbf{\theta}_{\xi}\right]\right|_{z=0}\,.\end{aligned}\ ] ] from the proof of theorem [ th1 ] we know that both the matrices @xmath238 and @xmath239 possess the bounded trace norm . then the application of lemma [ lem1 ] leads to @xmath262= \left.\dfrac{\partial}{\partial z}\frac{z}{x(z)-z}\right|_{z=0}{\mathbf{1}}^\prime { { \mbox{\boldmath$\sigma$}}}^{-1}_n{\mathbf{1}}\label{th12_eq1}\\ { \mathbf{1}}^\prime{\mathbf{s}}_n^{*}{{\mbox{\boldmath$\sigma$}}}_n\mathbf{b}_n&\stackrel{a.s.}{\longrightarrow}&\left.\dfrac{\partial}{\partial z}\frac{z}{x(z)-z}\right|_{z=0}\text{tr}\left[\mathbf{\theta}_{\zeta}\right]=\left.\dfrac{\partial}{\partial z}\frac{z}{x(z)-z}\right|_{z=0}\label{th12_eq2}\\ { \mathbf{1}}^\prime{\mathbf{s}}_n^{*}{{\mbox{\boldmath$\sigma$}}}_n{\mathbf{s}}_n^{*}{\mathbf{1}}&\stackrel{a.s.}{\longrightarrow } & \frac{1}{2}\left.\dfrac{\partial^2}{\partial z^2}\frac{z}{x(z)-z}\right|_{z=0}\text{tr}\left[\mathbf{\theta}_{\xi}\right]=\frac{1}{2}\left.\dfrac{\partial^2}{\partial z^2}\frac{z}{x(z)-z}\right|_{z=0}{\mathbf{1}}^\prime { { \mbox{\boldmath$\sigma$}}}^{-1}_n{\mathbf{1}}\label{th12_eq3}\end{aligned}\ ] ] for @xmath263 as @xmath30 , where @xmath212 is given in . finally , the application of the last two equalities together with , , and , leads to @xmath273 where @xmath274 and @xmath275 this completes the proof of theorem [ th12 ] . + * proof of theorem [ th2]*. first of all , we note that the asymptotic distribution of the quantity @xmath276 has already been derived in theorem [ th1 ] . from ( [ th31_eq6 ] ) we get that a consistent estimator of @xmath277 is given by @xmath278 in order to complete the proof of theorem [ th2 ] we need the following lemma of rubio and mestre ( 2011 ) + [ lem3 ] let @xmath279 be a sequence of i.i.d . real random vectors with zero mean and unit variance and with uniformly bounded @xmath35 moments for some @xmath36 and let @xmath280 be some nonrandom matrix with bounded trace norm at infinity . then it holds that @xmath281 next , we rewrite @xmath282 in the following way @xmath283 where @xmath284 is the @xmath285th column of the matrix @xmath32 . for the application of lemma [ lem3 ] we have to show that the matrix @xmath286 has a bounded trace norm at infinity . it holds that latexmath:[\[\label{rnorm } and , hence , the boundedness of the trace norm of the matrix @xmath286 follows directly from the assumption @xmath288 . the application of lemma [ lem3 ] leads to @xmath289 which together with ( [ con ] ) implies the statement of theorem [ th2 ] . bodnar , t. , gupta a.k . and n. parolya ( 2014 ) , on the strong convergence of the optimal linear shrinkage estimator for large dimensional covariance matrix , _ journal of multivariate analysis _ 132 , 215 - 228 . demiguel , v. , garlappi , l. , nogales , f. j. , and uppal , r. ( 2009 ) , a generalized approach to portfolio optimization : improving performance by constraining portfolio norms , _ management science _ 55 , 718 - 812 . rubio f. , mestre x. , and d.p . palomar ( 2012 ) , performance analysis and optimal selection of large minimum variance portfolios under estimation risk , _ ieee journal of selected topics in signal processing _ 6 , 337 - 350 . stein , c. ( 1956 ) , inadmissibility of the usual estimator for the mean of a multivariate normal distribution . in _ proceedings of the third berkeley symposium on mathematical statistics and probability _ , 1954 - 1955 , vol . i 197 - 206 . california press , berkeley .

dynamical linked cluster expansions are linked cluster expansions with hopping parameter terms endowed with their own dynamics . this amounts to a generalization from 2-point to point - link - point interactions . an associated graph theory with a generalized notion of connectivity is reviewed . we discuss physical applications to disordered systems , in particular to spin glasses , such as the bond - diluted ising model and the sherrington - kirkpatrick spin glass . we derive the rules and identify the full set of graphs that contribute to the series in the quenched case . this way it becomes possible to avoid the vague extrapolation from positive integer to , that usually goes along with an application of the replica trick . dynamical linked cluster expansions + for spin glasses hildegard meyer - ortmanns + and + thomas reisz institut fr theoretische physik bergische universitt wuppertal gaustrasse 20 d-42097 wuppertal , germany service de physique thorique de saclay ce - saclay f-91191 gif - sur yvette cedex , france institut fr theoretische physik universitt heidelberg philosophenweg 16 d-69120 heidelberg , germany

cond-mat0005412

linked cluster expansions ( lces ) have a long tradition in statistical physics . originally applied to classical fluids , later to magnetic systems ( @xcite,@xcite,@xcite and references therein ) , they were generalized to applications in particle physics in the eighties @xcite . there they have been used to study the continuum limit of a lattice @xmath10 field theory in 4 dimensions at zero temperature . in @xcite they were further generalized to field theories at finite temperature , simultaneously the highest order in the expansion parameter was increased to 18 . usually the analytic expansions are obtained as graphical expansions . because of the progress in computer facilities and the development of efficient algorithms for generating the graphs , it is nowadays possible to handle of the order of billions of graphs . the whole range from high temperatures down to the critical region becomes available , and thermodynamic quantities like critical indices and critical temperatures are determined with high precision ( the precision is comparable or even better than in corresponding high quality monte carlo results ) @xcite-@xcite . an extension of lces to a finite volume in combination with a high order in the expansion parameter turned out to be a particularly powerful tool for investigating the phase structure of systems with first and second order transitions by means of a finite size scaling analysis @xcite . linked cluster expansions are series expansions of the free energy and connected correlation functions about an ultralocal , decoupled theory in terms of a hopping parameter @xmath11 . the corresponding graphical representation is a sum in terms of connected graphs . the value of @xmath11 parametrizes the strength of interactions between fields at different lattice sites . usually they are chosen as nearest neighbours . in contrast to the ultralocal terms of a generic interaction we will sometimes refer to hopping terms as non - ultralocal . in this paper we develop dynamical linked cluster expansions ( dlces ) . these are linked cluster expansions with hopping parameter terms that are endowed with their own dynamics . such systems are realized in spin glasses with ( fast ) spins and ( slow ) interactions @xcite-@xcite . they also occur in variational estimates for su(n)-gauge - higgs systems , cf . @xcite . like lces they are expected to converge for a large class of interactions . formally dlces amount to a generalization of an expansion scheme from 2-point to point - link - point - interactions . these are interactions between fields associated with two points and with one pair of points called link . the points and links are not necessarily embedded on a lattice , and the links need not be restricted to nearest neighbours . we have developed a new theory in which a generalized notion of connectivity plays a central role . standard notions of equivalence classes of graphs like 1-line irreducible and 1-vertex irreducible graphs have been generalized , and new notions like 1-multiple - line irreducible graphs were defined in order to give a systematic classification . the paper is organized as follows . in sec . 2 we specify the models that admit a dlce . we introduce and explain the idea behind the abstract notions of theory . detailed definitions of , related notions and the computation of weights are given in sect . 4 treats the issue of renormalization in the sense of suitable resummations of graphs . applications to spin glasses are presented in sect . there it is of particular interest that dlces allow for the possibility of avoiding the replica trick . in the quenched limit we derive that dlces must be restricted to a subclass of the corresponding graphical expansion , so - called quenched dlces ( qdlces ) . we also list some examples for models whose phase structure is accessible to qdlces . to these belongs in particular the bond diluted ising model . 6 contains the summary and conclusions . in this section we first specify the class of models for which we develop dynamical linked cluster expansions . next we illustrate some basic notions of multiple - line graph theory , in particular the need for a new notion of connectivity . by @xmath12 we denote a finite or infinite set of points . one of its realizations is a hypercubic lattice in @xmath13 dimensions , infinite or finite in some directions with the topology of a torus . @xmath14 denotes the set of unordered pairs @xmath15 of sites @xmath16 , @xmath17 , also called unoriented links , and @xmath18 a subset of @xmath14 . we consider physical systems with a partition function of the generic form [ 2.zgen ] & & z(h , i , v ) + & & = u , with measures [ 2 m ] = _ x_0 d(x ) , u = _ l _ 1 du(l ) and action [ 2act ] s(,u , v ) = _ x _ s^((x ) ) + _ l _ 1s^1(u(l ) ) - _ x , y _ v(x , y)(x)u(x , y)(y ) , with non - ultralocal couplings & & v(x , y ) = v(y , x ) = 0 , + & & v(x , x ) = 0 . for later convenience the normalization via @xmath19 is chosen such that @xmath20=0 $ ] . the field @xmath21 is associated with the sites @xmath22 and the field @xmath23 lives on the links @xmath24 , and we write @xmath25 for @xmath26 . for definiteness and for simplicity of the notation here we assume @xmath27 and @xmath28 . in our actual applications to spin glasses the @xmath29s are the ( fast ) ising spins and the @xmath30s @xmath31 are the ( slow ) interactions . the action is split into two ultralocal parts , @xmath32 depending on fields on single sites , and @xmath33 depending on fields on single links @xmath34 . for simplicity we choose @xmath33 as the same function for all links @xmath24 . we may identify @xmath18 with the support of @xmath35 , _ 1 = \ { l=(x , y ) v(x , y ) = 0 } . the support of @xmath36 need not be restricted to nearest neighbours , also the precise form of @xmath32 and @xmath33 does not matter for the generic description of dlces , @xmath32 and @xmath33 can be any polynomials in @xmath29 and @xmath30 , respectively . the only restriction is the existence of the partition function . note that the interaction term @xmath37 contains a point - link - point - interaction and generalizes the 2-point - interactions @xmath38 of usual hopping parameter expansions . the effective coupling of the @xmath29 fields has its own dynamics governed by @xmath39 , the reason why we have called our new expansion scheme _ dynamical _ lce . dynamical linked cluster expansions are induced from a taylor expansion of @xmath40 about @xmath41 , the limit of a completely decoupled system . we want to express the series for @xmath42 in terms of connected graphs . let us consider the generating equation [ 2.genequation ] w / v(xy ) & = & 1/2 < ( x ) u(x , y ) ( y ) > + & = & 1/2 ( w_h(x ) i(x , y ) h(y ) + w_h(x)h(y)w_i(x , y ) + & + & w_h(x)i(x , y)w_h(y ) + w_i(x , y)h(y)w_h(x ) + & + & w_h(x)w_h(y)w_i(x , y ) ) . here @xmath43 denotes the normalized expectation value w.r.t . the partition function of eq . ( [ 2.zgen ] ) . subscripts @xmath44 and @xmath45 denote the derivatives of @xmath42 w.r.t . @xmath44 and @xmath46 , respectively . next we would like to represent the right hand side of eq . ( [ 2.genequation ] ) in terms of connected graphs . once we have such a representation for the first derivative of @xmath42 w.r.t . @xmath35 , grapical expansions for the higher derivatives can be traced back to the first one . for each @xmath42 in eq . ( [ 2.genequation ] ) we draw a shaded bubble , for each derivative w.r.t . @xmath47 a solid line , called a @xmath29-line , with endpoint vertex @xmath48 , and for each derivative w.r.t . @xmath49 a dashed line , called a @xmath30-line , with link label @xmath26 . the main graphical constituents are shown in fig . 1 . two @xmath29-lines with endpoints @xmath48 and @xmath50 are then joined by means of a dashed @xmath30-line with label @xmath51 , if the link @xmath51 has @xmath48 and @xmath50 as its endpoints , i.e. @xmath26 . according to these rules eq . ( [ 2.genequation ] ) , multiplied by @xmath36 and summed over @xmath48 and @xmath50 , is represented by fig . 2 . note that , because of the taylor operation , each solid line from @xmath48 to @xmath50 carries a factor @xmath36 . since the actual need for a new type of connectivity is not quite obvious from fig . 2 , because eq . ( [ 2.genequation ] ) does not contain higher than first order derivatives w.r.t . @xmath49 , let us consider a term ( 15.0,3.0 ) ( 15.0,7.0 ) w_h(x)w_h(y)w_h(r)w_h(s ) w_i(x , y)i(r , s ) occurring in the second derivative of @xmath42 w.r.t . @xmath36 , @xmath52 . according to the above rules this term would be represented as shown in fig . 3a . while the 2 vertices in the last term of fig . 2 are connected in the usual sense via a common ( solid ) line ( the dashed line with an attached bubble could be omitted in this case ) , the graph in fig . 3a would be disconnected in the old sense , since neither @xmath48 nor @xmath50 are line - connected with @xmath53 and @xmath54 , but -as a new feature of dlce graphs- the lines from @xmath48 to @xmath50 and from @xmath53 to @xmath54 are connected via the dashed lines emerging from a common bubble shown in the middle of the graph . as we see from fig . 3a , we need an additional notion of connectivity referring to the possibility of multiple - line connectivity . while the analytic expression is fixed , it is a matter of convenience to further simplify the graphical notation of fig . 3a at @xmath41 . two possibilities are shown in fig . 3b and fig . 3b we later refer in the formal definition of the new type of multiple - line connectivity . in the familiar standard notion of connectivity two vertices of a graph are connected via lines . the vertices are line - connected . already there , in a dual language , one could call two lines connected via vertices . the second formulation is just appropriate for our need to define when two lines are connected . the corresponding vertices mediating the connectivity of lines are visualized by tubes , in fig . 3b we have just one of them . the tubes should be distinguished from the former type of vertices represented as full dots which are connected via bare @xmath29-lines . in fig . 3c we show a simplified representation of fig . 3b that we actually use in graphical expansions . ( 15.0,7.0 ) the derivative terms have to be evaluated at @xmath41 . for @xmath41 we have a decomposition of @xmath42 according to w(h , i , v=0)= _ x _ 0 w^(h(x ) ) + _ l _ 1 w^1(i(l ) ) with [ 2.w0 ] z^(h ) = and [ 2.w1 ] z^1(i ) = . in eq.s ( [ 2.w0],[2.w1 ] ) we have omitted any single site or single link dependence , because we assume that @xmath32 and @xmath33 are the same for all @xmath55 and @xmath56 , respectively . therefore , at @xmath41 , the only non - vanishing derivatives of @xmath42 are [ 2.wd1 ] . w_h(x_1)h(x_2) ... h(x_n ) _ x_1,x_2, ... x_n and [ 2.wd2 ] . w_i(l_1)i(l_2) ... i(l_m ) _ v=0 = _ l_1,l_2, ... l_m , but mixed derivatives w.r.t . h and i vanish . as anticipated in fig.s 3b and 3c , for @xmath41 we replace the dashed bubbles and graphically distinguish between bubbles with @xmath29-lines and bubbles with @xmath30-lines . we define . ( 3.0,1.0 ) ( 0.0,0.0 ) ( 4.0,0.0 ) ( 0.0,0.0 ) ( 0.0,0.0)(1.0,0.8)(2.0,0.8 ) ( 0.0,0.0)(1.0,0.5)(2.0,0.5 ) ( 0.0,0.0)(1.0,-0.7)(2.0,-0.7 ) ( 1.2,0.2)(0.2,0)@xmath57 ( 1.2,-0.1)(0.2,0)@xmath57 ( 1.2,-0.4)(0.2,0)@xmath57 } n = v^c_n = ( ) _ h=0 for a connected n - point vertex with @xmath58 bare @xmath29-lines emerging from it and . ( 4.0,1.0 ) ( 0.0,-0.6 ) ( 4.0,0.0 ) ( 1.0,1.5)(2.0,2.1)(3.0,1.5 ) ( 1.0,1.0)(2.0,1.6)(3.0,1.0 ) ( 1.0,0.0)(2.0,0.6)(3.0,0.0 ) ( 1.6,1.1)(0.2,0)@xmath57 ( 1.6,0.8)(0.2,0)@xmath57 ( 1.6,0.5)(0.2,0)@xmath57 ( 2.0,1.85)(0,1)0.2 ( 2.0,1.6)(0,1)0.2 ( 2.0,1.35)(0,1)0.2 ( 2.0,1.1)(0,1)0.2 ( 2.0,0.85)(0,1)0.2 ( 2.0,0.6)(0,1)0.2 ( 2.0,0.35)(0,1)0.2 ( 2.0,0.1)(0,1)0.2 } = m^1 c _ = ( ) _ i=0 for a connected @xmath59-line consisting of @xmath59 bare lines . if @xmath60 , we often omit the dashed line . if the bare lines of a @xmath59-line are internal @xmath29-lines , they get vertices attached to their endpoints , if they are external @xmath30-lines , no vertices will be attached . let @xmath61 denote the lattice volume in @xmath13 dimensions . the taylor expansion of @xmath42 about @xmath41 to second order in @xmath35 then reads [ 2.long ] w(h , i , v ) & = & w(h , i , v=0 ) + & + & _ x , y _ 0 v(x , y ) w_h(x ) w_h(y)w_i(x , y ) + & + & _ x , y , r , s _ 0 v(x , y ) v(r , s ) + & ( & 4 w_h(y)w_h(s)w_h(r)h(x)w_i(x , y)w_i(r , s ) + & + & 2 w_h(x)h(r)w_h(y)h(s)w_i(x , y)w_i(r , s ) + & + & 4 w_h(y)w_h(s)w_h(r)h(x)w_i(x , y)i(r , s ) + & + & 2 w_h(r)h(x)w_h(y)h(s)w_i(r , y)i(x , s ) + & + & w_h(x)w_h(y)w_h(r)w_h(s)w_i(x , y)i(r , s ) ) _ v=0 + & + & o(v^3 ) , where we have used that @xmath62 . for each @xmath42 in the products of @xmath42s we now insert eq.s ( [ 2.wd1]),([2.wd2 ] ) . if we choose @xmath35 in a standard way as next - neighbour couplings v(x , y ) = 2k _ = 0^d-1 ( _ x+,y+ _ x-,y ) with @xmath63 denoting the unit vector in @xmath64-direction , eq . ( [ 2.long ] ) becomes in a graphical representation at @xmath65 [ 2.graph ] & = & ( 2k ) ( 2d ) ( 4.0,1.0 ) ( 0.0,0.0 ) ( 4.0,1.0 ) ( 0.0,0.2 ) ( 2.0,0.2 ) ( 0.0,0.2)(1,0)2.0 + & + & ( 2k)^2 \ { ( 2d)^2 ( 3.5,1.0 ) ( 0.0,0.0 ) ( 4.0,1.0 ) ( 0.0,-0.2 ) ( 0.0,0.6 ) ( 2.0,0.2 ) ( 0.0,-0.2)(1.0,-0.2)(2.0,0.2 ) ( 0.0,0.6)(1.0,0.6)(2.0,0.2 ) + ( 2d ) ( 4.0,1.0 ) ( 0.0,0.0 ) ( 4.0,1.0 ) ( 0.0,0.2 ) ( 2.0,0.2 ) ( 0.0,0.2)(1.0,-0.4)(2.0,0.2 ) ( 0.0,0.2)(1.0,0.8)(2.0,0.2 ) + & & + ( 2d)^2 ( 3.5,1.0 ) ( 0.0,0.0 ) ( 4.0,1.0 ) ( 0.0,-0.2 ) ( 0.0,0.6 ) ( 2.0,0.2 ) ( 0.0,-0.2)(1.0,-0.2)(2.0,0.2 ) ( 0.0,0.6)(1.0,0.6)(2.0,0.2 ) ( 0.9,0.2)(0,1)0.15 ( 0.9,0.4)(0,1)0.15 ( 0.9,0.6)(0,1)0.15 ( 0.9,0.0)(0,1)0.15 ( 0.9,-0.2)(0,1)0.15 ( 0.9,-0.4)(0,1)0.15 + ( 2d ) ( 4.0,1.0 ) ( 0.0,0.0 ) ( 4.0,1.0 ) ( 0.0,0.2 ) ( 2.0,0.2 ) ( 0.0,0.2)(1.0,-0.4)(2.0,0.2 ) ( 0.0,0.2)(1.0,0.8)(2.0,0.2 ) ( 1.0,0.2)(0,1)0.15 ( 1.0,0.4)(0,1)0.15 ( 1.0,0.6)(0,1)0.15 ( 1.0,0.0)(0,1)0.15 ( 1.0,-0.2)(0,1)0.15 ( 1.0,-0.4)(0,1)0.15 + & & + 2(2d ) ( 3.5,1.0 ) ( 0.0,0.0 ) ( 4.0,1.0 ) ( 0.0,0.0 ) ( 0.0,0.3 ) ( 2.0,0.0 ) ( 2.0,0.3 ) ( 0.0,0.0)(1.0,-0.5)(2.0,0.0 ) ( 0.0,0.3)(1.0,0.8)(2.0,0.3 ) ( 1.1,0.2)(0,1)0.15 ( 1.1,0.4)(0,1)0.15 ( 1.1,0.6)(0,1)0.15 ( 1.1,0.0)(0,1)0.15 ( 1.1,-0.2)(0,1)0.15 ( 1.1,-0.4)(0,1)0.15 } + & + & o(k^3 ) . for clarity , here we have written explicitly the topological symmetry factors and the lattice embedding numbers . ( usually graphs represent their full weights including these factors . ) note that the first two graphs of the second order contribution also occur in a usual lce with frozen @xmath30-dynamics , the next two differ by an additional dashed 2-line and the last one becomes even disconnected without the dashed line . as usual , graphical expansions for correlation functions , in particular susceptibilities , are generated from @xmath66 by taking derivatives w.r.t . the external fields @xmath47 and @xmath49 . graphically this amounts to attaching external @xmath29-lines and @xmath30-lines with in passing we remark that the conventional lce is included as a special case of the dlce , if the @xmath30-dynamics is `` frozen '' to some value @xmath67 , so that w^1(i ) & = & -s_1(u_0)+ i u_0 , + & = & u_0 , + & = & 0 , i.e. , no n - lines do occur with @xmath68 . in this case it becomes redundant to attach dashed lines to bare lines . as mentioned above , in an lce only the first three contributions would be left in eq . ( [ 2.graph ] ) . the definition of a as it will be given here is adapted to the computation of susceptibilities and the free energy , where the sum is taken over all possible locations of the fields . the definition easily generalizes to correlation functions . for details of the standard definiton of graphs in the framework of linked cluster expansions and related notions we refer e.g. to @xcite . here , for convenience , we briefly recall the very definition of a graph to point out the new properties of as defined below in this section . a ( standard lce ) graph or diagram is a structure = ( _ , _ , e _ , _ ) , where @xmath69 and @xmath70 are disjoint sets of internal lines and vertices of @xmath71 , respectively . @xmath72 is a map e _ : _ & & \{0,1,2 , } , + v & & e_(v ) that assigns to every vertex @xmath35 the number of external lines @xmath73 attached to it . finally , @xmath74 is the incidence relation that assigns internal lines to their two endpoints . a or is a structure = ( _ , _ , _ , e_^ ( ) , e_^(u ) , _ , _ ) . @xmath75 , @xmath76 and @xmath77 are three mutually disjoint sets , _ & = & , + _ & = & , + _ & = & . @xmath78 is a map e_^ ( ) : _ & & \{0,1,2 , } , + v & & e_^()(v ) that assigns to every vertex @xmath35 the number of bare external @xmath29-lines @xmath79 attached to @xmath35 . every such @xmath29-line represents a field @xmath29 . the number of external @xmath29-lines of @xmath80 is denoted by @xmath81 . similarly , @xmath82 is a map e_^(u ) : _ & & \{0,1,2 , } , + m & & e_^(u)(m ) that assigns to every multiple line @xmath83 the number of external @xmath30-lines @xmath84 attached to @xmath83 . every such @xmath30-line represents a field @xmath30 associated with a lattice link . the number of external @xmath30-lines of @xmath80 is given by @xmath85 . furthermore , @xmath86 and @xmath87 are incidence relations that assign bare internal lines to their endpoint vertices and to their multiple lines , respectively . we treat lines as unoriented . the generalization to oriented lines is easily done . more precisely , let @xmath88 be the set of unordered pairs of vertices @xmath89 with @xmath90 , @xmath91 . ( the bar implies unordered pairs , the prime the exclusions of @xmath92 , @xmath93 . ) as in the standard linked cluster expansion , self - lines are excluded . every bare internal line is then mapped onto its pair of endpoints via _ : _ ^. we say that @xmath35 and @xmath94 are the endpoint vertices of @xmath95 if @xmath96 . if there is such an @xmath95 , @xmath35 and @xmath94 are called neighbours . similarly , @xmath87 is a map _ : _ & & _ , + l & & _ ( l ) that maps every bare internal line to a multiple line . a multiple line @xmath97 is composed of bare internal lines @xmath95 which belong to @xmath83 in the sense that @xmath98 . @xmath99 is the total number of bare internal lines belonging to @xmath83 . with @xmath100 , @xmath83 is called a @xmath59-line . we always require that @xmath101 . on the other hand , every bare internal line belongs to one and only one multiple line . for simplicity we often identify a @xmath102-line with the only one bare line that belongs to it . next we introduce some further notions that will be used later . external vertices are vertices having external @xmath29-lines attached , _ , ext = \ { v _ e_^()(v)=0 } , whereas internal vertices do not , @xmath103 . similarly , external multiple lines have external @xmath30-lines attached , _ , ext = \ { m _ e_^(u)(m)=0 } , and the complement in @xmath76 are the internal multiple lines , @xmath104 . for every pair of vertices @xmath90 , @xmath91 , let @xmath105 be the set of lines with endpoint vertices @xmath35 and @xmath94 , and @xmath106 the number of these lines . thus @xmath105 is the set of lines @xmath35 and @xmath94 have in common . with @xmath79 denoting the number of external @xmath29-lines attached to @xmath93 , t__(v ) = _ w _ ^1(v , w ) + e_^()(v ) is the total number of bare lines attached to @xmath35 . some topological notions and global properties of graphs will be of major interest in the following . a central notion is the connectivity of a . recall that we want to consider the dlce expansion of the free energy and of truncated correlation functions as an expansion in connected graphs . as indicated in section 2 , the main generalization compared to the common notion of connectivity of a graph which is required here is that an additional type of connectivity is provided by multiple - lines . to define the connectivity of a @xmath80 , @xmath80 first is mapped to a ( standard ) lce graph @xmath107 to which the standard notion of connectivity applies . there are various equivalent ways to define such a map . we choose the following one . * for every multiple - line @xmath97 define a new vertex @xmath108 . let @xmath109 and define @xmath110 as the union of the vertices of @xmath80 and the new set of vertices originating from the multiple - lines . * for every bare internal line @xmath95 define two new internal lines @xmath111 and incidence relations ( l_1 ) & = & ( v_1 , w(_(l ) ) ) , + ( l_2 ) & = & ( v_2 , w(_(l ) ) ) , where @xmath112 and @xmath113 are the two endpoint vertices of @xmath51 . the set of all lines @xmath111 , for all @xmath95 , is denoted by @xmath114 . * define the external incidence relations : & & \{0,1,2 , } , + ( v ) & = & e_^()(v ) , , + ( v ) & = & e_^(u)(m ) , . now , @xmath115 is defined by = ( , , , ) . having defined the standard lce graph @xmath115 for any @xmath80 , we call @xmath80 multiple - line connected or just connected if @xmath115 is connected ( in the usual sense ) . in fig . 4 we have given two examples for a connected ( upper graph ) and a disconnected ( lower graph ) . ( 15.0,7.0 ) the next important notion is the topological equivalence of two . two [ top.equiv ] _ i = ( _ i , _ i , _ i , e^()_i , e^(u)_i , _ i , _ i ) , i=1,2 are called ( topologically ) equivalent if there are three invertible maps _ 1:_1 & & _ 2 , + _ 2:_1 & & _ 2 , + _ 3:_1 & & _ 2 , such that _ 2 _ 2 & = & _ 1 _ 1 , + _ 2 _ 2 & = & _ 3 _ 1 , and e_2^ ( ) _ 1 & = & e_1^ ( ) , + e_2^(u ) _ 3 & = & e_1^(u ) . here @xmath116 means decomposition of maps , and _ 1 : ^ & & ^ + _ 1(v , w ) & = & ( _ 1(v),_1(w ) ) . a symmetry of a @xmath117 is a triple of maps @xmath118 , @xmath119 and @xmath120 such that _ 2 & = & _ 1 , + _ 2 & = & _ 3 , and e^ ( ) _ 1 & = & e^ ( ) + e^(u ) _ 3 & = & e^(u ) . the number of these maps is called the symmetry number of @xmath80 . we denote by @xmath121 the set of equivalence classes of connected with @xmath122 bare internal lines , @xmath123 external @xmath29-lines and @xmath124 external @xmath30-lines . furthermore we set _ e_1,e_2 : = _ l0 _ e_1,e_2(l ) . a multiple line graph @xmath80 does not need to have a vertex . if @xmath125 , we have @xmath126 as well . if in addition @xmath80 is connected , @xmath76 consists of only one element , with all external @xmath30-lines attached to it . ( we anticipate that @xmath80 is 1-multiple - line irreducible ( 1mli ) by definition . for the definition of 1mli cf . section 4 below . ) the only graph of @xmath127 is given by = . ( 4.0,1.0 ) ( 0.0,0.0 ) ( 10.0,0.0 ) ( 0.9,0.6)(1.5,0.5)(2.1,0.6 ) ( 0.9,0.4)(1.5,0.3)(2.1,0.4 ) ( 0.9,-0.5)(1.5,-0.4)(2.1,-0.5 ) ( 0.9,-0.3)(1.5,-0.2)(2.1,-0.3 ) ( 1.5,-0.55)(0,1)0.15 ( 1.5,-0.35)(0,1)0.15 ( 1.5,-0.15)(0,1)0.15 ( 1.5,0.05)(0,1)0.15 ( 1.5,0.25)(0,1)0.15 ( 1.5,0.45)(0,1)0.15 ( 1.5,0.65)(0,1)0.15 ( 1.2,0.15)(0.2,0)@xmath57 ( 1.2,-0.05)(0.2,0)@xmath57 ( 1.2,0.05)(0.2,0)@xmath57 } e . it represents the leading term of the susceptibility [ susc.0e ] _ 0,e = _ l_1, ,l_e _ 1 < u(l_1 ) u(l_e ) > ^c and is given by @xmath128 . the index @xmath129 in ( [ susc.0e ] ) stands for truncated ( connected ) correlation . by removal of a @xmath59-line @xmath97 we mean that @xmath83 is dropped together with all bare internal lines and all external @xmath30-lines that belong to @xmath83 . this notion is explained in fig . ( it is used in section 4 for @xmath102-lines to define 1-particle irreducible ( 1pi ) and 1-line irreducible ( 1li ) . ) on the other hand , by decomposition of a @xmath59-line @xmath97 we mean that @xmath83 is dropped together with the external @xmath30-lines of @xmath83 , but all bare internal lines that belong to @xmath83 are kept in the graph , being identified now with @xmath102-lines . this notion will be used below to define 1mli and renormalized . it is illustrated in fig . 5b . similarly , decomposition of a vertex @xmath93 means to remove the vertex @xmath35 and to attach the free end of every line that entered @xmath35 before to a new vertex , a separate one for each line . this notion is used to define 1-vertex - irreducible ( 1vi ) and renormalized vertex moments for . for an example see fig . 5c . ( 15.0,7.0 ) in the last section we have defined and the notions of connectivity and equivalence of such graphs . the definition is chosen in such a way that the series expansions of the free energy and of truncated correlation functions are obtained as a sum over equivalence classes of connected . the number @xmath122 of bare internal lines of a @xmath80 counts the order in the expansion parameter @xmath36 to which @xmath80 contributes . if @xmath36 is of the form [ susc.vform ] v(x , y ) = 2k _ z(x ) _ y , z , with @xmath130 any finite @xmath48-dependent set of lattice sites , the contribution of @xmath80 is a multiple of @xmath131 . often used special cases are the nearest neighbour interactions [ susc.nn ] v(x , y ) = 2k _ = 0^d-1 ( _ x , y+ + _ x , y- ) and the uniform interaction [ susc.mf ] v(x , y ) = 2k ( 1 - _ x , y ) , which is used in models of spin glasses and partially annealed neural networks . susceptibilities of the @xmath29 and @xmath30 fields will be represented as [ susc.susc ] _ e_1,e_2 & = & _ x_1, ,x_e_1 _ 0 _ l_1, ,l_e_2 _ 1 < ( x_1 ) ( x_e_1 ) u(l_1 ) u(l_e_2 ) > ^c + & & _ x_1, ,x_e_1 _ 0 _ l_1, ,l_e_2 _ 1 . _ h = i=0 + & = & _ l0 ( 2k)^l _ _ e_1,e_2(l ) w ( ) with lattice volume @xmath61 and dimension @xmath13 . similar representations hold for higher moments @xmath64 . the weight @xmath132 of a @xmath133 is given as the product of the following factors * for every vertex @xmath93 a factor v^c_n = ( ) _ h=0 , where @xmath134 is the total number of bare lines attached to @xmath35 . * for every multiple line @xmath97 a factor [ susc.m1c ] m^1 c _ = ( ) _ i=0 , where @xmath100 , that is @xmath83 is a @xmath59-line , * a factor @xmath135 , where @xmath136 is the topological symmetry number of @xmath80 , * a factor counting the permutation symmetry of external @xmath29-lines , , * a factor counting the permutation symmetry of external @xmath30-lines , , * the lattice embedding number of @xmath80 , which is the number of ways @xmath80 can be embedded on a lattice of given geometry , e. g. on a hypercubic lattice . to this end , the vertices of @xmath80 ( if any ) are placed onto lattice sites . one arbitrary vertex is placed at a fixed lattice site , in order to account for the volume factor @xmath137 in ( [ susc.susc ] ) . a priori there is no exclusion principle . this means that any number of vertices can be placed at the same lattice site . ( this is sometimes called free embedding . ) two restrictions apply to the embeddings . the first constraint results from the fact that a bare internal line represents a hopping propagator @xmath36 , with lattice sites @xmath48 and @xmath50 at which the two endpoint vertices of the line are placed at . a reasonable computation of the embedding number takes into account the particular form of @xmath36 from the very beginning . the second constraint is that bare lines of the same multiple - line have to be mapped on the same pair of sites . + for example , if @xmath36 is the nearest neighbour interaction ( [ susc.nn ] ) , two vertices which have at least one line in common are to be placed at nearest neighbour lattice sites . on the other hand , a propagator @xmath36 of the form ( [ susc.mf ] ) implies a rather weak constraint in that @xmath48 and @xmath50 must be different , but otherwise can be freely placed over the lattice . we remark that in case of a non - trivial internal symmetry ( such as considered in section 7 ) the expressions of eq.s ( [ susc.susc])-([susc.m1c ] ) must be modified appropriately . in particular , the weight ( [ susc.m1c ] ) of a multiple - line does no longer take such a simple form . truncated correlation functions , susceptibilities and other moments are obtained as sums over that are connected . their number rapidly grows with increasing order , that is with increasing number of bare internal lines . the procedure of `` renormalization '' means that the connected moments are represented in terms of reduced ones . the reduced moments are obtained by summation over classes which are more restricted than just by their property of being connected . of course the number of graphs of such classes is smaller . only the most restricted classes must be constructed . the subsequent steps towards the moment computation are most conveniently done by operating analytically with the reduced moments . in particular , it is no longer necessary to generate all connected and the corresponding intermediate classes . a connected @xmath80 is called 1-particle irreducible ( 1pi ) if it satisfies the following condition . remove an arbitrary @xmath102-line of @xmath80 . there is at most one connected component left that has external lines attached . ( this notion is the same as the one used in the context of feynman graphs . ) on the other hand , if in addition the remaining graph is still connected , then @xmath80 is called 1-line irreducible ( 1li ) . in many cases it is sufficient to use only the second notion . it is for instance sufficient that all vertices are constrained to have only an even number of lines attached , or more generally , if graphs and subgraphs with one external line are forbidden . for notational simplicity we assume in the following that this is the case and henceforth refer only to the notion 1li . the generalization to the case in which 1li and 1pi graphs must be distinguished goes along the same lines as for lces , which was discussed in @xcite . by @xmath138 we denote the subset of @xmath133 that are 1li . 1li - susceptibilities are defined as series in the hopping parameter similarly as in ( [ susc.susc ] ) by restricting the summation to 1li graphs , _ e_1,e_2 ^ 1li = _ l0 ( 2k)^l _ _ e_1,e_2 ^ 1li(l ) w ( ) . susceptibilities are easily obtained in a closed form in terms of 1li - susceptibilities @xmath139 . it can be shown that the @xmath139s can be obtained by an appropriate legendre transform . for instance _ 2,0 & = & , + _ 2,1 & = & , where @xmath140 is the fourier transform of the hopping propagator @xmath36 , v(x , y ) = _ -^ e^-ik(x - y ) ( k ) . in lces the second important resummation comes from so called vertex renormalizations . this means partial resummation of graphs with specific properties such as having one external vertex only . these sums then are considered as `` renormalized vertices '' replacing the vertices of graphs with complementary properties . the procedure naturally leads to the notion of 1-vertex irreducibility ( 1vi ) and renormalized moments . in dlce we follow this procedure . the very definition of 1vi has to be modified slightly for because of the enhanced connectivity properties due to multiple - lines . in addition , as a natural generalization , we supplement vertex renormalization by multiple - line renormalization . a @xmath80 is called 1-vertex irreducible ( 1vi ) if it satisfies the following condition . decompose an arbitrary vertex @xmath93 . every connected component of the remaining graph has then at least one external line attached . it can be a @xmath29-line or a @xmath30-line . we write _ e_1,e_2 ^ 1vi(l ) = \ { _ e_1,e_2 ^ 1li(l ) } for the set of equivalence classes of graphs that are both 1li and 1vi , with @xmath123 external @xmath29-lines , @xmath124 external @xmath30-lines and @xmath122 bare internal lines . the renormalized vertex moment graphs are 1li graphs that have precisely one external vertex and no external multiple line , q_k(l ) = \ { _ k,0 ^ 1li(l ) } . a @xmath80 is called 1-multiple - line irreducible ( 1mli ) if it satisfies the following criterion . decompose an arbitrary multiple - line @xmath97 . every remaining connected component has then at least one external line attached . it can be a @xmath29-line or a @xmath30-line . we write _ e_1,e_2 ^ 1mli(l ) = \ { _ e_1,e_2 ^ 1li(l ) } . the renormalized multiple - line moment graphs are graphs that are 1li and have precisely one external multiple - line , but no external vertex , r_k(l ) = \ { _ 0,k^1li(l ) } . the equivalence classes of graphs that are both 1vi and 1mli are denoted by s_e_1,e_2(l ) = _ e_1,e_2 ^ 1vi(l ) _ e_1,e_2 ^ 1mli(l ) . with the renormalized moment graphs as defined above , the 1li - susceptibilities are now obtained in the form _ e_1,e_2 ^ 1li = _ l0 ( 2k)^l _ _ e_1,e_2(l ) ( ) . the weights @xmath141 are given as a product of factors as described in the last subsection , with the following two exceptions . * the vertex coupling constants @xmath142 are replaced by the renormalized vertex moments v^c_n v^c_n = _ l0 ( 2k)^l _ q_n(l ) w ( ) . * the multiple line coupling constants @xmath143 are replaced by the renormalized multiple line moments m^1 c _ m^c_= _ l0 ( 2k)^l _ r_(l ) w ( ) . in the series representations above , the @xmath132 are computed according to the rules of subsection 3.2 . in @xcite we have described an algorithmic construction of graphs that is the first step for an automatic computer aided generation . in exceptional cases dlces can be summed up in a closed form . otherwise , a computer aided generation of graphs is unavoidable , because the proliferation of dlce graphs is pronounced even compared to lce graphs . for lce graphs we know that billions of graphs must be included to obtain the critical exponents with the accuracy of some per mil . in this section we consider applications of dlces to disordered systems , in particular to spin glasses with `` slow '' interactions coupled to `` fast '' spins . the interactions @xmath144 are assumed to be in equilibrium with a thermal heat bath of inverse temperature @xmath145 , while the spins @xmath146 are equilibrated according to a second inverse temperature @xmath147 . both systems need not be mutually in equilibrium . let @xmath148 be the partition function that describes the equilibrium distribution of the spins for given @xmath144s , [ neur.zspin ] z_(j ) = _ \ { _ i = 1 } ) . the sum runs over pairs @xmath149 that need not be restricted to nearest neighbours only . we further assume that the dynamics of the time evolution of the slow interactions @xmath144 is governed by a langevin equation n j_(i , j ) = - ( j ) + _ ij(t ) with [ neur.h ] ( j ) = - + n _ i < j=1^n j_(i , j)^2 . here @xmath148 is given by ( [ neur.zspin ] ) , @xmath150 is the total number of spins , @xmath64 is a positive constant and @xmath151 is a stochastic gaussian white noise of zero mean and correlation < _ ij(t ) _ kl(t^ ) > = _ ( ij),(kl ) ( t - t^ ) . such a langevin equation for the @xmath144s can be derived from an ansatz which is motivated by neural networks @xcite-@xcite . moreover , since the time evolution of the @xmath144s is determined by a dissipative langevin equation , the equilibrium distribution of the slow variables is again a boltzmann distribution , governed now by the second temperature @xmath152 , [ neur.z ] z_^^ = _ -^_i < j = 1^n dj_(i , j ) , with @xmath19 some normalization that will be specified below . the effective hamiltonian @xmath153 of @xmath144 is given by ( [ neur.h ] ) . it is these equilibrium aspects of coupled systems of fast spins and slow interactions that we can treat analytically with dlces , as we will show below . apparently , because of integer @xmath0 , @xmath167 has the form of models to which dlce applies , with a hopping term [ neur.hop ] s_hop(j,^(a ) ) = - _ a=1^n _ i < j=1^n j_(i , j ) _ i^(a ) _ j^(a ) , a single link action [ neur.link ] s^1(j_(i , j ) ) = c j_(i , j ) + q n j_(i , j)^2 , and a single site action [ neur.site ] s^(_i^(a ) ) = - h _ depending on @xmath0 we distinguish the following cases . * @xmath168 . first we note that for @xmath168 we can directly apply dlce to @xmath169 and to derived quantities to obtain their series expansions in @xmath147 . but from a physical point of view , in a disordered system one is not interested in @xmath168 , because @xmath168 corresponds to the completely annealed situation , in which the fast spins and the slow interactions are in mutual equilibrium . ( in contrast , in particle physics one _ is _ interested in the @xmath168 case , cf . our applications of dlces in the framework of variational cumulant expansions of the @xmath170 higgs model @xcite . ) * @xmath68 , integer . again we apply dlce to @xmath171 , but have to account for the permutation symmetry between the replicas . formally , the replica symmetry plays a role similar to an internal symmetry , e.g. an @xmath172 symmetry in a scalar higgs model . dlces with nontrivial internal symmetries have been discussed in connection with the @xmath170 higgs model @xcite . thus we can study `` unquenched '' equilibrium aspects of systems with two temperatures and compare the results from dlces adapted to internal replica symmetry with monte carlo simulations for the same @xmath0 @xcite . * @xmath1 , the quenched limit . as we will show in the next section , in order to discuss the @xmath158 limit , we need not refer to @xmath0 times replicated systems @xmath164 characterized by ( [ neur.hop])-([neur.site ] ) , but just to @xmath148 given by ( [ neur.hop])-([neur.site ] ) with @xmath168 . by means of special dlces , so - called quenched dlces ( qdlces ) , we directly calculate the left hand side of ( [ neur.replica_trick ] ) . therefore , setting @xmath168 in ( [ neur.hop])-([neur.site ] ) in qdlces does _ not _ imply the completely annealed case , because we first take the logarithm of @xmath148 and then average over the @xmath144s . first we adapt the notation to section 2 to include more general cases . @xmath12 denotes the support of the spins , that is the set of lattice sites , with @xmath173 denoting their total number . @xmath18 @xmath174 are the pairs of sites whose spins interact . in accordance with ( [ neur.zre ] ) , we write for the normalized link - average of a function @xmath175 [ sg.average ] [ [ f(j ) ] ] = j f(j ) with j & = & _ l_1 d(j(l ) ) , + d(j ) & = & _ 1 dj , _ -^d(j ) = 1 . it is convenient to introduce the single link expectation values < g(j ) > _ 1 d(j ) g(j ) and the generating function @xmath176 by < > _ 1 . the way in which the replica trick can be avoided is examplified for the free energy density @xmath177 of the spin system averaged over the link couplings . the partition function of the spin system for a given distribution of the link interactions @xmath178 is given by = _ sp , where @xmath179 and s_sp(,j ) & = & - _ x , y_0 v(x , y ) ( x ) ( y ) j(x , y ) , + & = & _ x_0 d(x ) . without loss of generality we identify the support of the interaction @xmath180 with the set @xmath181 of lattice sites where @xmath182 is supported , _ 1 = \ { l=(x , y ) v(x , y)= 0}. for simplicity we assume @xmath36 to be of the form ( [ susc.vform ] ) , so that @xmath11 is a measure of the strength of the interactions @xmath36 . the free energy density of the spin system allows for a series expansion in the standard lce sense , with the link field @xmath183 playing the role of a `` background field '' , [ sg.wsp_series ] w_sp(j ) = _ l0 ( 2k)^l _ ^sp_0(l)w^sp(,j ) . here @xmath184 ( with @xmath185 ) denotes the set of equivalence classes of connected lce graphs with @xmath186 external lines and @xmath122 internal lines . the spin - weights @xmath187 are of the form [ sg.wsp_weight ] w^sp(,j ) = r^sp ( ) _ _ _ 1^ _ l_1 j(l)^m(l ) . the sum is taken over all non - vanishing lattice embeddings of the graph @xmath80 . it runs over all maps of internal lines of the graph @xmath80 to pairs of lattice sites of @xmath18 that are consistent with the graph topology in the sense discussed in section 3 . for every @xmath24 , @xmath188 denotes the number of lines of @xmath80 that are mapped onto the link @xmath51 by the embedding . all other factors that contribute to the weight are collected in the prefactor @xmath189 , including the inverse topological symmetry number of @xmath80 . next we want to express @xmath190 $ ] as a series in @xmath11 by means of dlce . toward this end we set @xmath191 and insert the series ( [ sg.wsp_series ] ) with ( [ sg.wsp_weight ] ) into ( [ sg.average ] ) . at this stage we are not concerned with question of ( uniform or dominated ) convergence and obtain [ sg.wspint ] [ [ w_sp(j ) ] ] & = & _ l0 ( 2k)^l _ _ 0^sp(l ) j w^sp(,j ) + & = & _ l0 ( 2k)^l _ _ 0^sp(l ) r^sp ( ) _ _ _ 1^ _ l_1 < j(l)^m(l)>_1 . the next step is to express the full single link expectation values in terms of the connected ones . they are related by [ sg.conn ] < j^m > _ 1 = _ ( ) _ p < j^|p| > _ 1^c , where @xmath192 denotes the set of all partitions of @xmath193 into non - empty , mutually disjoint subsets of @xmath194 . @xmath195 is the number of elements of the set @xmath196 . the relation ( [ sg.conn ] ) is equivalent to the partition of all lines of @xmath80 that are mapped to the same lattice link into multiple lines , with every multiple line contributing a factor [ sg.wdiff ] < j^|p| > _ 1^c = . _ i=0 = m^1 c_|p| . using ( [ sg.conn ] ) , ( [ sg.wdiff ] ) we rewrite ( [ sg.wspint ] ) as [ sg.wsptodlce ] [ [ w_sp(j ) ] ] = _ l0 ( 2k)^l _ _ 0^sp(l ) r^sp ( ) _ ( _ ) ( _ p m^1 c_|p| ) ( _ _ 1^ _ l_1 1 ) . the last summation in ( [ sg.wsptodlce ] ) is over all maps @xmath197 of the lines of @xmath80 to the lattice links of @xmath18 subject to the constraint that all lines that belong to the same multiple - line corresponding to some @xmath198 are mapped onto the same lattice link . finally we rewrite ( [ sg.wsptodlce ] ) as a sum over . to this end , we first observe that for every @xmath199 , every partition @xmath200 of the lines of @xmath80 into multiple - lines generates a @xmath201 in the obvious way . let us denote by @xmath202 the subset of of @xmath203 that stay connected after decomposition of all multiple lines . ( these are the which stay connected in the usual graph theoretical sense , when the dashed lines are omitted . ) for every @xmath204 there is a unique @xmath205 and at least one @xmath206 such that @xmath207 . let @xmath208 be the ( uniquely determined ) number of partitions @xmath206 with @xmath207 , and @xmath209 such an arbitrary partition . ( [ sg.wsptodlce ] ) then becomes [ sg.wdlce ] [ [ w_sp(j ) ] ] = _ l0 ( 2k)^l _ _ 0,0(l ) n_r^sp ( ( ) ) ( _ p ( ) m^1 c_|p| ) ( _ ( ) _ 1^ 1 ) . the last bracket of ( [ sg.wdlce ] ) is the lattice embedding factor of the @xmath210 . the second bracket from the right does not depend on the choice of @xmath209 and is the product of the multiple - line coupling constants as defined in section 3 . finally , @xmath211 is precisely the remaining part of the weight of @xmath210 that was described in detail in section 3 , endowed with the correct inverse topological symmetry number of the @xmath210 ( because of the factor @xmath208 ) . in summary , we obtain the series expansion of the link - averaged free energy density in terms of dlce graphs , [ sg.wseries ] [ [ w_sp(j ) ] ] = _ l0 ( 2k)^l _ _ 0,0(l ) w ( ) . the weight @xmath212 of a @xmath210 is defined and computed according to the rules given in section 3 . ( [ sg.wseries ] ) is the series representation of the link - averaged free energy density of the spin system , i.e. the free energy density of the @xmath1 replica system , in terms of dlce graphs . it looks much like the series representation of the @xmath102-replica system , which is given by [ sg.w1series ] w_1-repl = _ l0 ( 2k)^l _ _ 0,0(l ) w ( ) according to the discussion of section 2 . we recall that @xmath203 is the set of dlce vacuum graphs with @xmath122 bare lines that are connected in the generalized dlce sense . comparing ( [ sg.wseries ] ) and ( [ sg.w1series ] ) , the transition from @xmath168 to @xmath1 replicas is achieved by keeping only the subset @xmath213 of that are connected in the original ( lce ) sense . we emphasize that the restriction of dlces to qdlces is not an ad hoc ( or intuitively motivated ) assumption but a derived consequence of the fact that the logarithm is taken before the integration @xmath160 . this procedure accounts for _ all _ graphs that contribute to a given order in @xmath11 . thus we do have to truncate the series unless the series can be completely summed up , as it happens in exceptional cases . we expect that the series ( [ sg.wseries ] ) are convergent for a large class of interactions @xmath214 and @xmath36 if the coupling constant @xmath11 is sufficiently small . for special interactions most of the yield vanishing contributions so that we can further restrict the sum to a subset of @xmath215 . an example is given by the mean field type of interaction of the sherrington - kirkpatrick model , cf . sect . 5.2 . in the following we list some examples for systems of which we can study the phase structure by means of qdlces . their actions are special cases of ( [ neur.hop])-([neur.site ] ) with @xmath168 ( as explained above ) and the following choice of variables . * infinite range models . choose @xmath216 as before , @xmath217 , @xmath218 , v(x , y ) & = & k ( 1-_x , y ) , + _ 1 & & _ 1 + s^1(j ) & = & n j^2 . now the sum over the sites in ( [ neur.hop ] ) runs over arbitrary pairs @xmath149 , @xmath219 , and we obtain the infinite range sherrington - kirkpatrick model . for infinte range and in the thermodynamic limit ( @xmath220 ) , the phase structure can be solved by replica mean field theory , cf . e.g. @xcite . for qdlces the infinite range and @xmath220 limits imply that only tree graphs of 2-lines contribute to the series of the free energy density , such as & & + & & + + & & + ( 15.0,15.0 ) + + & & + & & the reason for that is that each 2-line gets a factor of @xmath221 from the @xmath222-part of the action , but each vertex gets a factor @xmath223 from the embedding onto a lattice @xmath12 . the contribution of every tree graph to the free energy is proportional to @xmath150 . if a chain of 2-lines connecting the vertices gets closed , forming a loop , there is one @xmath223 less in the total embedding factor . thus the contribution is suppressed by @xmath224 for every loop and vanishes in the thermodynamic limit . because of the simple tree structure there is a chance for summing up the series . this is currently under investigation . * finite range connectivity . the sum @xmath225 of the spins is now restricted to next - neighbours or , more generally , to a finite number of pairs . rather than specifying @xmath226 of ( [ neur.link ] ) , it is sufficient for dlces and qdlces to choose @xmath227 . let exp(-s_1(j ) ) = ( 1-p ) ( j ) + p ( j-1 ) with @xmath228 $ ] . the variables @xmath229 can then be interpreted as occupation numbers of the bonds . furthermore , if we choose @xmath230 we obtain a * * bond - diluted ising model . + choosing @xmath231 we obtain a * * bond - diluted @xmath232-state potts model . + if @xmath233 , we obtain a * * bond - diluted heisenberg model . qdlces provide a systematic analytic expansion for disordered systems with bond dilution in a quenched limit . coming from the high temperature ( small @xmath147 ) region one can study the phase structure as a function of the degree of dilution . work in this direction is in progress . in this paper we have introduced a new expansion scheme for 3-point interactions or , more precisely , for point - link - point interactions . this scheme generalizes linked cluster expansions for 2-point interactions by including hopping parameter terms endowed with their own dynamics . in chapters 3 - 4 we have developed a theory with an additional new type of multiple - line connectivity . we have introduced appropriate equivalence classes of graphs and discussed the issue of renormalization . these notions are required for an algorithmic generation of graphs . because of the fast proliferation of graphs already at low orders in the expansion , a computer aided implementation becomes unavoidable , if one is interested in higher orders of the expansion than we have computed so far . in sect . 5 we have shown how to avoid the replica trick for calculating the free energy of disordered systems in the quenched limit . dlces are a systematic expansion method to study the phase structure of disordered systems . it is systematic in the sense that we do not restrict the expansion to certain subclasses of graphs that can be summed up , but we identify and keep _ all _ graphs that contribute to a given order in the expansion parameter . dlces provide an analytic tool for studying systems in situations in which it has been impossible so far . we would like to thank reimar khn ( heidelberg ) for discussions . m. wortis , `` linked cluster expansion '' , in phase transition and critical phenomena , vol.3 , eds . c. domb and m.s . green ( academic press , london 1974 ) . c. itzykson , j .- drouffe , `` statistical field theory '' , vol.2 , cambridge university press , 1989 . guttmann , `` asymptotic analysis of power - series expansions '' , in phase transition and critical phenomena , vol.13 , eds . c. domb and j.l . lebowitz ( academic press ) . m. lscher and p. weisz , nucl . phys . * b300*[fs22 ] ( 1988 ) 325 . t. reisz , nucl . phys . * b450 * ( 1995 ) 569 . t. reisz , phys . lett . * 360b * ( 1995 ) 77 . m. campostrini , a. pelissetto , p. rossi and e. vicari , nucl . phys . * b459 * ( 1996 ) 207 . s. zinn , s .- n . lai and m. e. fisher , phys . rev . * e54 * ( 1996 ) 1176 . p. butera and n. comi , phys * e55 * ( 1997 ) 6391 . h. meyer - ortmanns and t. reisz , j. stat . phys . * 87 * ( 1997 ) 755 . d. sherrington and s. kirkpatrick , phys . lett * 35 * ( 1975 ) 1972 . v. dotsenko , s. franz and m. mzard , j. phys . * a27 * ( 1994 ) 2351 . r. w. penney , a. c. c. coolen , and d. sherrington , j. phys . a : math . * 26 * ( 1993 ) 3681 . a. c. c. coolen , r. w. penney , and d. sherrington , phys . * b48 * ( 1993 ) 16116 . h. meyer - ortmanns and t. reisz , int . j. mod . a14 ( 1999 ) 947 . t. reisz , nucl . b527 * ( 1998 ) 363 . k. h. fischer and j. a. hertz , `` spin glasses '' , cambridge university press , cambridge 1991 . j. zinn - justin , `` quantum field theory and critical phenomena '' , 3rd edition , oxford science publication , clarendon press , oxford 1996 . d. sherrington , j. phys . * a13 * ( 1980 ) 637 ; i. kondor , j. phys . * a16 * ( 1983 ) l127 . b. berg and w. janke , phys . lett * 80 * ( 1998 ) 4771

the _ ab initio _ quasirelativistic hartree - fock method developed specifically for the calculation of spectroscopic parameters of heavy atoms and highly charged ions was used to derive spectral data for the multicharged tungsten ion w . the configuration interaction method was applied to include the electron - correlation effects . the relativistic effects were taken into account in the breit - pauli approximation for quasirelativistic hartree - fock radial orbitals . the energy level spectra , radiative lifetimes , lande-factors are calculated for the , and configurations of the w ion .

1504.04567

[ intro ] in the present work we continue our theoretical studies @xcite of multicharged tungsten ions . the spectroscopic parameters of systems with the open 4d - shell are investigated . extensive interest in highly - charged tungsten ions @xcite is caused by its unique physical properties . metallic tungsten is widely exploited in high - temperature devices , including fusion reactors @xcite . although tungsten is difficult to melt and vaporize , its highly - charged ions emerge in fusion plasma and decrease its temperature . therefore one needs to control the concentration of these ions by monitoring their spectra . unfortunately , the experimental data for the multicharged tungsten ions are very sparse . this fact is very evident from the recent compilation @xcite of the experimental and semi - empirical data for the tungsten ions . it is absolutely clear from that review that the tungsten ions from the rubidium - like to the palladium - like systems , which have an open 4d - shell , are investigated very seldom . for the most ions , only a few levels from the ground @xmath6 configuration are determined . moreover , very few energy levels of the first excited @xmath7 and @xmath8 configurations are determined for some tungsten ions . this situation has encouraged us to perform the theoretical investigation of the spectroscopic parameters for these ions . the calculated data can substantially facilitate experimental studies of the corresponding spectra ; on the other hand , these data can be adopted directly for plasma spectra modelling . the w@xmath0 ion energy spectrum of the @xmath3 , @xmath4 and @xmath5 configurations consists of significantly larger number of energy levels , and the range of possible transition wavelengths is wider compared to the previously studied w@xmath9 @xcite and w@xmath10 @xcite ions . in order to keep the current paper concise , we present only the energy spectra and the characteristics of levels . the remaining spectroscopic parameters will be available from the database adamant ( applicable data of many - electron atom energies and transitions ) which is currently under development . likewise for the ions mentioned above ions , we employ a quasirelativistic ( qr ) approach @xcite with an extensive inclusion of correlation effects to investigate the w@xmath0 ion spectroscopic properties . since application of this approach has produced high - accuracy results for the w@xmath9 and w@xmath10 ions , there is no doubt about its applicability to the tungsten ion of slightly lower ionization degree . in section [ calc ] we provide a description of the adopted calculation method . since this approximation ultimately matches the one applied for the investigation of the w@xmath9 and w@xmath10 spectroscopic data in @xcite , here we provide only a brief summary of our method . produced results are discussed in section [ result ] . [ calc ] our applied quasirelativistic approximation significantly differs from widely used relativistic hartree - fock ( hfr ) method described in @xcite . the main differences arise from the set of quasirelativistic hartree - fock equations ( qrhf ) @xcite applied to determine radial orbitals ( ro ) . the details of these differences are summarized in @xcite , while the most complete and consistent description of our method to calculate spectroscopic parameters in the quasirelativistic approach is given in @xcite . furthermore , the methods to include effects of the finite nucleus size are described in @xcite . the radial orbitals for electrons representing the investigated configurations were determined by solving the qrhf equations . initially , the equations were solved for the ground configuration @xmath3 . afterwards the @xmath11-electron ro was calculated using the frozen - core potential for the @xmath4 configuration . the determined quasirelativistic radial orbitals were applied to describe electrons in all configurations . such an identical ro basis enables us to avoid the non - orthogonality problems of ro when the electron transition parameters are calculated . this basis is adopted to describe all the quasirelativistic admixed configurations , i.e. , the configurations obtained from the adjusted configurations by the one - electron and two - electron virtual excitations without changing the electron principal quantum number ( @xmath12 , in our case ) . the admixed configurations produced by the virtual excitations of one or two electrons from the @xmath12 shells to the states with @xmath13 were also included . the transformed radial orbitals ( tro ) with a variational parameter were applied to describe these electron states . initially , the tro were designed to include correlation effects in non - relativistic calculations @xcite and were employed successfully for that purpose ( see , e.g. , @xcite ) . the method to determine the tro in quasirelativistic approximation was presented in @xcite . in the current work , likewise in @xcite , we adopt the tro with two variational parameters , the integer even parameter @xmath14 , and the positive parameter @xmath15 : here the factor @xmath17 ensures the normalization of the determined tro , the first term in parenthesis performs the transformation of ro @xmath18 and the second term provides their orthogonality . the parameters @xmath14 and @xmath15 are chosen to gain the maximum of the averaged energy correlation correction calculated in the second order of perturbation theory ( pt ) . the level energies were calculated in two approximations , a and b. first of all , likewise in the previous calculations , tro were determined for the virtual electron excitations having the principal quantum number @xmath19 and all allowed values of the orbital quantum number @xmath20 . this is called the approximation a. afterward the tro basis was extended up to @xmath21 . this is approximation b. one can generate a huge amount of the admixed configurations for such bases of ro . for the selection of these configurations , we employed the mean weights of the admixed configurations @xmath22 determined in the second order of pt @xcite : here the numerator represents the sum of squared sub - matrix elements of the operators of kinetic energy and electrostatic interaction between the adjusted configurations @xmath24 and the admixed configurations @xmath25 . the nominator represents the statistical weight @xmath26 of adjusted configuration multiplied by the squared mean energy differences . different selection criteria @xmath27 values for the admixed configurations were adopted in case of different sizes of ro bases . we selected the admixed configurations with @xmath28 . here we must emphasize that the number of selected configurations changes only very slightly if the basis of tro is extended while the selection parameter @xmath27 remains the same . the full number of possible configuration state functions ( csf ) was further reduced by permuting the shells of virtually excited electrons in order to reduce the order of the hamiltonian being diagonalized , as it was described in @xcite . this procedure substantially decreased the number of the @xmath29-terms of admixed configurations without any changes to their correlation corrections . our computation abilities are restricted by the amount of memory ( ram ) of computer employed for calculations . in turn , the required amount is determined by the size of hamiltonian which depends on the number of csf having the same total @xmath29 momenta . so this parameter regulates the @xmath27 values in our calculations . the computation parameters for two approximations are given in in table [ tablea ] @xmath17 & & & & & + + + 1 & 0 & - & 0.694 & 1.5 & 49 & @xmath30&19 & @xmath31&16 & @xmath32&10 & @xmath33 + 2 & 122438 & 2.127e@xmath344 & 1.057 & 2.5 & 83 & @xmath30&10 & @xmath31 & & & & + 3 & 141702 & 2.586e@xmath344 & 1.209 & 1.5 & 39 & @xmath35&29 & @xmath30&29 & @xmath33 & & + 4 & 154216 & 2.010e@xmath346 & 1.063 & 3.5 & 45 & @xmath36&38 & @xmath30&15 & @xmath37 & & + 5 & 157527 & 6.824e@xmath344 & 1.868 & 0.5 & 59 & @xmath35&39 & @xmath33 & & & & + 6 & 157912 & 2.558e@xmath349 & 1.058 & 4.5 & 42 & @xmath38&42 & @xmath36&15 & @xmath30 & & + 7 & 221938 & 1.254e@xmath345 & 1.311 & 2.5 & 44 & @xmath35&30 & @xmath32&19 & @xmath37 & & + 8 & 274650 & 1.082e@xmath344 & 1.105 & 3.5 & 59 & @xmath30&37 & @xmath36 & & @xmath39 & & + 9 & 288865 & 1.886e@xmath344 & 1.142 & 4.5 & 51 & @xmath30&42 & @xmath38 & & & & + 10 & 301018 & 8.616e@xmath343 & 1.133 & 1.5 & 40 & @xmath35&29 & @xmath31&16 & @xmath30 & 9 & @xmath32 + 11 & 310060 & 1.364e@xmath344 & 1.466 & 0.5 & 59 & @xmath33&39 & @xmath35 & & & & + 12 & 317427 & 1.787e@xmath344 & 1.246 & 2.5 & 43 & @xmath31&29 & @xmath35&16 & @xmath37 & 9 & @xmath30 + 13 & 321357 & 4.357e@xmath344 & 1.091 & 5.5 & 98 & @xmath38 & & & & & & + 14 & 339668 & 2.705e@xmath344 & 1.108 & 2.5 & 40 & @xmath31&40 & @xmath37&13 & @xmath35 & & + 15 & 345943 & 3.332e@xmath344 & 1.102 & 3.5 & 80 & @xmath37&16 & @xmath36 & & & & + 16 & 400068 & 1.383e@xmath344 & 0.951 & 1.5 & 64 & @xmath32&24 & @xmath33 & 4 & @xmath31 & & + 17 & 438446 & 8.753e@xmath343 & 1.154 & 4.5 & 51 & @xmath36&32 & @xmath30&15 & @xmath38 & & + 18 & 480344 & 8.276e@xmath343 & 1.080 & 1.5 & 45 & @xmath31&32 & @xmath33&11 & @xmath35 & 8 & @xmath32 + 19 & 512314 & 6.113e@xmath343 & 1.164 & 2.5 & 61 & @xmath32&22 & @xmath37&11 & @xmath35 & & + 20 & 1233221 & 3.861e@xmath401 & 1.521 & 1.5 & 13 & @xmath41&11 & @xmath42 & 8 & @xmath43 & 8 & @xmath44 + 21 & 1329059 & 3.109e@xmath401 & 1.534 & 2.5 & 34 & @xmath42&14 & @xmath45&11 & @xmath44&10 & @xmath41 + 22 & 1356662 & 3.036e@xmath401 & 1.373 & 3.5 & 29 & @xmath42&12 & @xmath45 & 9 & @xmath46 & 7 & @xmath43 + 23 & 1357366 & 1.381e@xmath401 & 1.942 & 0.5 & 27 & @xmath42&13 & @xmath47 & 8 & @xmath41 & 8 & @xmath48 + 24 & 1364576 & 4.467e@xmath401 & 1.279 & 4.5 & 17 & @xmath42&13 & @xmath49&10 & @xmath45 & 9 & @xmath50 + 25 & 1368349 & 3.517e@xmath401 & 1.253 & 1.5 & 17 & @xmath44&14 & @xmath51&10 & @xmath52 & 9 & @xmath42 + 26 & 1388185 & 2.928e@xmath401 & 1.078 & 3.5 & 14 & @xmath53 & 8 & @xmath54 & 7 & @xmath55 & 6 & @xmath56 + 27 & 1391042 & 7.068e@xmath401 & 1.199 & 5.5 & 25 & @xmath57&16 & @xmath45&15 & @xmath58&11 & @xmath59 + 28 & 1405138 & 1.941e@xmath401 & 1.060 & 2.5 & 13 & @xmath51&13 & @xmath60 & 8 & @xmath61 & 7 & @xmath56 + 29 & 1407154 & 3.205e@xmath401 & 1.507 & 1.5 & 20 & @xmath44&10 & @xmath48&10 & @xmath62 & 8 & @xmath55 + 30 & 1425752 & 1.523e@xmath401 & 1.302 & 2.5 & 17 & @xmath44&10 & @xmath62 & 8 & @xmath63 & 6 & @xmath48 + 31 & 1442162 & 9.255e@xmath401 & 1.164 & 4.5 & 17 & @xmath45&10 & @xmath53 & 8 & @xmath54 & 7 & @xmath49 + 32 & 1447934 & 4.692e@xmath401 & 1.186 & 3.5 & 15 & @xmath45&12 & @xmath63&12 & @xmath51 & 7 & @xmath64 + 33 & 1465711 & 8.808e@xmath403 & 0.797 & 2.5 & 30 & @xmath61&10 & @xmath56 & 8 & @xmath65 & 8 & @xmath66 + 34 & 1475896 & 5.473e@xmath401 & 1.259 & 4.5 & 33 & @xmath42 & 9 & @xmath67 & 8 & @xmath53 & 7 & @xmath59 + 35 & 1477277 & 1.435e@xmath401 & 1.320 & 3.5 & 26 & @xmath42&12 & @xmath68 & 9 & @xmath69 & 9 & @xmath49 + 36 & 1486712 & 9.243e@xmath402 & 1.462 & 1.5 & 21 & @xmath42&11 & @xmath60&10 & @xmath44 & 9 & @xmath70 + 37 & 1489522 & 2.472e@xmath401 & 1.314 & 0.5 & 17 & @xmath55&13 & @xmath42&13 & @xmath48&12 & @xmath47 + 38 & 1499789 & 3.908e@xmath403 & 0.835 & 0.5 & 11 & @xmath71 & 8 & @xmath45 & 8 & @xmath72 & 8 & @xmath62 + 39 & 1500264 & 5.120e@xmath401 & 1.207 & 5.5 & 24 & @xmath45&18 & @xmath53&17 & @xmath73&14 & @xmath57 + 40 & 1508761 & 5.595e@xmath403 & 0.824 & 1.5 & 10 & @xmath74 & 9 & @xmath75 & 8 & @xmath60 & 7 & @xmath51 + 41 & 1508994 & 1.761e@xmath401 & 1.227 & 2.5 & 15 & @xmath60&12 & @xmath44 & 9 & @xmath46 & 6 & @xmath76 + 42 & 1510476 & 7.363e@xmath402 & 1.156 & 3.5 & 13 & @xmath53&12 & @xmath46&11 & @xmath77&10 & @xmath55 + 43 & 1519016 & 8.984e@xmath402 & 1.364 & 1.5 & 24 & @xmath48&15 & @xmath60&10 & @xmath78 & 8 & @xmath46 + 44 & 1520802 & 4.593e@xmath402 & 1.315 & 2.5 & 14 & @xmath55&10 & @xmath46&10 & @xmath68&10 & @xmath77 + 45 & 1521647 & 2.996e@xmath401 & 1.165 & 6.5 & 43 & @xmath73&31 & @xmath59&17 & @xmath53 & & + 46 & 1522174 & 1.371e@xmath401 & 1.145 & 3.5 & 15 & @xmath61&13 & @xmath60 & 9 & @xmath64 & 9 & @xmath51 + 47 & 1525207 & 1.036e@xmath401 & 1.163 & 4.5 & 18 & @xmath45&15 & @xmath53&10 & @xmath59 & 8 & @xmath51 + 48 & 1527002 & 4.811e@xmath401 & 1.187 & 5.5 & 20 & @xmath45&17 & @xmath79&13 & @xmath53&11 & @xmath64 + 49 & 1528587 & 1.826e@xmath401 & 1.187 & 2.5 & 9 & @xmath48 & 8 & @xmath47 & 6 & @xmath80 & 6 & @xmath68 + 50 & 1535449 & 1.066e@xmath401 & 1.075 & 4.5 & 18 & @xmath81&12 & @xmath82&11 & @xmath57&10 & @xmath83 + 51 & 1540027 & 1.732e@xmath344 & 1.141 & 7.5 & 55 & @xmath73&43 & @xmath84 & & & & + 52 & 1544245 & 6.118e@xmath402 & 1.168 & 3.5 & 22 & @xmath68&11 & @xmath49 & 8 & @xmath83 & 5 & @xmath85 + 53 & 1552067 & 1.110e@xmath401 & 0.987 & 2.5 & 20 & @xmath86&13 & @xmath87&12 & @xmath64 & 8 & @xmath61 + 54 & 1553351 & 8.317e@xmath402 & 1.171 & 4.5 & 18 & @xmath64&13 & @xmath57&11 & @xmath49&11 & @xmath67 + 55 & 1554926 & 3.973e@xmath402 & 1.087 & 5.5 & 31 & @xmath88&19 & @xmath59 & 9 & @xmath89 & 8 & @xmath81 + 56 & 1569249 & 4.704e@xmath402 & 0.892 & 1.5 & 15 & @xmath77&14 & @xmath51&11 & @xmath90&10 & @xmath62 + 57 & 1569592 & 9.282e@xmath401 & 1.114 & 6.5 & 28 & @xmath53&26 & @xmath79&25 & @xmath89&11 & @xmath84 + 58 & 1572802 & 2.678e@xmath402 & 1.147 & 3.5 & 13 & @xmath61 & 9 & @xmath68 & 7 & @xmath77 & 6 & @xmath43 + 59 & 1591339 & 1.387e@xmath401 & 1.178 & 5.5 & 25 & @xmath57&20 & @xmath50&12 & @xmath53&12 & @xmath79 + 60 & 1598739 & 3.767e@xmath402 & 1.093 & 1.5 & 16 & @xmath62&12 & @xmath63&10 & @xmath91 & 7 & @xmath92 + 61 & 1599077 & 2.100e@xmath401 & 0.684 & 0.5 & 35 & @xmath62&12 & @xmath55&11 & @xmath93&10 & @xmath92 + 62 & 1603882 & 4.091e@xmath402 & 1.157 & 4.5 & 14 & @xmath61&11 & @xmath82&11 & @xmath49 & 9 & @xmath42 + 63 & 1605525 & 2.308e@xmath402 & 1.157 & 2.5 & 10 & @xmath86 & 9 & @xmath41 & 8 & @xmath46 & 7 & @xmath94 + 64 & 1605815 & 1.949e@xmath402 & 1.075 & 3.5 & 13 & @xmath43&13 & @xmath95&10 & @xmath61 & 9 & @xmath96 + 65 & 1615746 & 1.691e@xmath402 & 1.029 & 4.5 & 15 & @xmath97&14 & @xmath81&12 & @xmath98 & 8 & @xmath57 + 66 & 1617094 & 5.836e@xmath402 & 1.186 & 2.5 & 12 & @xmath76&11 & @xmath99 & 7 & @xmath63 & 6 & @xmath52 + 67 & 1617342 & 1.571e@xmath402 & 1.307 & 1.5 & 12 & @xmath44 & 7 & @xmath100 & 7 & @xmath101 & 6 & @xmath71 + 68 & 1624193 & 2.842e@xmath402 & 1.112 & 4.5 & 12 & @xmath102&10 & @xmath57&10 & @xmath60 & 9 & @xmath77 + 69 & 1629216 & 5.448e@xmath402 & 1.086 & 3.5 & 14 & @xmath43&10 & @xmath96 & 9 & @xmath95 & 6 & @xmath103 + 70 & 1630865 & 5.680e@xmath403 & 0.992 & 2.5 & 13 & @xmath60&11 & @xmath75&10 & @xmath51&10 & @xmath61 + 71 & 1632610 & 5.893e@xmath403 & 0.969 & 1.5 & 15 & @xmath51&12 & @xmath55 & 9 & @xmath75 & 7 & @xmath62 + 72 & 1635037 & 3.657e@xmath402 & 1.110 & 5.5 & 20 & @xmath89&12 & @xmath81&12 & @xmath57&11 & @xmath59 + 73 & 1636150 & 7.169e@xmath403 & 0.329 & 0.5 & 29 & @xmath62&21 & @xmath55 & 8 & @xmath104 & 8 & @xmath100 + 74 & 1643750 & 3.717e@xmath402 & 1.056 & 1.5 & 21 & @xmath46&16 & @xmath60 & 7 & @xmath74 & 6 & @xmath72 + 75 & 1644103 & 1.999e@xmath402 & 1.233 & 2.5 & 12 & @xmath63&11 & @xmath101&10 & @xmath90 & 8 & @xmath76 + 76 & 1646826 & 3.138e@xmath403 & 1.440 & 0.5 & 14 & @xmath105&13 & @xmath71&12 & @xmath106&11 & @xmath48 + 77 & 1647104 & 8.568e@xmath402 & 1.172 & 4.5 & 16 & @xmath63&14 & @xmath77 & 8 & @xmath85 & 8 & @xmath61 + 78 & 1652080 & 1.758e@xmath402 & 1.087 & 3.5 & 14 & @xmath60&12 & @xmath107 & 8 & @xmath49 & 6 & @xmath108 + 79 & 1659647 & 1.416e@xmath401 & 1.200 & 5.5 & 23 & @xmath88&17 & @xmath45&15 & @xmath50&13 & @xmath64 + 80 & 1660774 & 1.107e@xmath402 & 0.977 & 3.5 & 22 & @xmath96&10 & @xmath50 & 9 & @xmath108 & 8 & @xmath109 + 81 & 1668491 & 3.112e@xmath403 & 1.197 & 1.5 & 14 & @xmath71&12 & @xmath55 & 9 & @xmath45 & 7 & @xmath104 + 82 & 1671927 & 2.435e@xmath402 & 0.870 & 0.5 & 18 & @xmath72&16 & @xmath100&13 & @xmath46&13 & @xmath93 + 83 & 1672404 & 8.518e@xmath403 & 1.193 & 2.5 & 10 & @xmath76 & 9 & @xmath110 & 6 & @xmath101 & 5 & @xmath41 + 84 & 1674931 & 1.072e@xmath402 & 0.920 & 4.5 & 44 & @xmath97 & 6 & @xmath81 & 6 & @xmath60 & 5 & @xmath96 + 85 & 1678742 & 3.615e@xmath402 & 1.163 & 2.5 & 10 & @xmath111&10 & @xmath49 & 8 & @xmath86 & 8 & @xmath41 + 86 & 1680925 & 2.265e@xmath402 & 1.179 & 3.5 & 19 & @xmath77 & 8 & @xmath85 & 6 & @xmath46 & 6 & @xmath44 + 87 & 1683416 & 6.847e@xmath401 & 1.183 & 6.5 & 61 & @xmath59&27 & @xmath73 & 8 & @xmath79 & & + 88 & 1684083 & 4.598e@xmath402 & 1.171 & 4.5 & 16 & @xmath49&10 & @xmath45&10 & @xmath57 & 8 & @xmath112 + 89 & 1687191 & 3.890e@xmath340 & 1.126 & 7.5 & 54 & @xmath84&43 & @xmath73 & & & & + 90 & 1690771 & 6.483e@xmath402 & 1.109 & 1.5 & 13 & @xmath74&13 & @xmath72&12 & @xmath70&11 & @xmath110 + 91 & 1690870 & 4.167e@xmath402 & 1.150 & 5.5 & 24 & @xmath64&16 & @xmath79&14 & @xmath82&11 & @xmath53 + 92 & 1692855 & 5.747e@xmath403 & 1.192 & 0.5 & 17 & @xmath113&16 & @xmath48&14 & @xmath114&11 & @xmath115 + 93 & 1695611 & 6.182e@xmath402 & 1.097 & 4.5 & 24 & @xmath67&12 & @xmath83&10 & @xmath81 & 9 & @xmath49 + 94 & 1700018 & 5.300e@xmath403 & 1.147 & 3.5 & 18 & @xmath68&11 & @xmath96 & 8 & @xmath108 & 5 & @xmath46 + 95 & 1700177 & 1.454e@xmath402 & 1.158 & 2.5 & 17 & @xmath62 & 7 & @xmath61 & 7 & @xmath43 & 6 & @xmath68 + 96 & 1704764 & 2.524e@xmath402 & 1.154 & 3.5 & 12 & @xmath43 & 7 & @xmath80 & 6 & @xmath50 & 6 & @xmath53 + 97 & 1710011 & 9.413e@xmath403 & 1.133 & 3.5 & 7 & @xmath116 & 6 & @xmath112 & 6 & @xmath115 & 5 & @xmath96 + 98 & 1710032 & 8.064e@xmath403 & 1.280 & 0.5 & 16 & @xmath115&15 & @xmath117&13 & @xmath76 & 8 & @xmath114 + 99 & 1711856 & 2.360e@xmath402 & 1.158 & 1.5 & 14 & @xmath92&11 & @xmath118&10 & @xmath68 & 8 & @xmath77 + 100 & 1713858 & 3.908e@xmath402 & 1.094 & 6.5 & 37 & @xmath89&26 & @xmath53&17 & @xmath84&11 & @xmath79 + 101 & 1717084 & 6.578e@xmath403 & 1.137 & 2.5 & 8 & @xmath77 & 7 & @xmath119 & 6 & @xmath101 & 5 & @xmath47 + 102 & 1727902 & 9.868e@xmath403 & 1.018 & 2.5 & 9 & @xmath111 & 7 & @xmath120 & 7 & @xmath49 & 6 & @xmath65 + 103 & 1729125 & 1.893e@xmath402 & 1.009 & 5.5 & 49 & @xmath97&15 & @xmath96&11 & @xmath98&10 & @xmath121 + 104 & 1729155 & 1.565e@xmath402 & 1.042 & 4.5 & 52 & @xmath96 & 8 & @xmath109 & 4 & @xmath50 & 4 & @xmath63 + 105 & 1729726 & 1.443e@xmath402 & 1.069 & 3.5 & 11 & @xmath94 & 8 & @xmath95 & 6 & @xmath85 & 6 & @xmath63 + 106 & 1730267 & 5.467e@xmath403 & 1.417 & 1.5 & 10 & @xmath122 & 7 & @xmath115 & 7 & @xmath123 & 6 & @xmath48 + 107 & 1745674 & 2.623e@xmath403 & 1.119 & 3.5 & 10 & @xmath104 & 8 & @xmath124 & 6 & @xmath95 & 5 & @xmath94 + 108 & 1748910 & 1.089e@xmath402 & 0.989 & 2.5 & 21 & @xmath125 & 9 & @xmath126 & 6 & @xmath127 & 6 & @xmath65 + 109 & 1752283 & 2.163e@xmath402 & 1.202 & 1.5 & 13 & @xmath70&10 & @xmath118 & 9 & @xmath41 & 7 & @xmath55 + 110 & 1754423 & 1.234e@xmath402 & 1.100 & 4.5 & 22 & @xmath50 & 8 & @xmath96 & 7 & @xmath63 & 7 & @xmath81 + 111 & 1758121 & 1.699e@xmath402 & 1.165 & 3.5 & 11 & @xmath69&10 & @xmath87 & 8 & @xmath43 & 8 & @xmath108 + 112 & 1765460 & 2.334e@xmath402 & 1.099 & 5.5 & 38 & @xmath58&12 & @xmath128 & 7 & @xmath88 & 6 & @xmath96 + 113 & 1766687 & 2.410e@xmath403 & 1.325 & 1.5 & 26 & @xmath106&22 & @xmath115&11 & @xmath114 & 6 & @xmath71 + 114 & 1768571 & 6.538e@xmath403 & 1.153 & 2.5 & 11 & @xmath115 & 9 & @xmath99 & 5 & @xmath125 & 5 & @xmath41 + 115 & 1776374 & 2.635e@xmath403 & 1.177 & 2.5 & 8 & @xmath129 & 8 & @xmath41 & 8 & @xmath115 & 7 & @xmath106 + 116 & 1778765 & 1.158e@xmath402 & 1.106 & 3.5 & 17 & @xmath94 & 8 & @xmath63 & 6 & @xmath67 & 6 & @xmath69 + 117 & 1780499 & 1.286e@xmath402 & 1.117 & 4.5 & 15 & @xmath50&11 & @xmath67&10 & @xmath103 & 7 & @xmath102 + 118 & 1781141 & 1.690e@xmath402 & 1.017 & 1.5 & 14 & @xmath62&11 & @xmath130 & 8 & @xmath99 & 7 & @xmath101 + 119 & 1785670 & 6.710e@xmath402 & 1.062 & 5.5 & 18 & @xmath58&13 & @xmath97&11 & @xmath131 & 9 & @xmath102 + 120 & 1787379 & 2.708e@xmath402 & 1.416 & 0.5 & 23 & @xmath132&18 & @xmath41&17 & @xmath93&11 & @xmath92 + 121 & 1799110 & 8.630e@xmath403 & 1.209 & 1.5 & 13 & @xmath122 & 9 & @xmath52 & 8 & @xmath51 & 6 & @xmath91 + 122 & 1800217 & 1.419e@xmath402 & 1.044 & 3.5 & 9 & @xmath54 & 9 & @xmath125 & 6 & @xmath119 & 6 & @xmath56 + 123 & 1801149 & 1.778e@xmath402 & 1.014 & 4.5 & 19 & @xmath97&15 & @xmath133&10 & @xmath95 & 9 & @xmath134 + 124 & 1802184 & 6.800e@xmath403 & 1.093 & 2.5 & 12 & @xmath124 & 8 & @xmath80 & 7 & @xmath126 & 6 & @xmath43 + 125 & 1806476 & 6.519e@xmath403 & 1.111 & 3.5 & 22 & @xmath125&19 & @xmath126&11 & @xmath65 & 6 & @xmath127 + 126 & 1809104 & 1.129e@xmath402 & 1.127 & 2.5 & 14 & @xmath43&10 & @xmath56&10 & @xmath52 & 9 & @xmath124 + 127 & 1821618 & 1.094e@xmath402 & 1.025 & 2.5 & 17 & @xmath135&16 & @xmath125 & 8 & @xmath65 & 7 & @xmath126 + 128 & 1822627 & 4.909e@xmath403 & 1.061 & 3.5 & 20 & @xmath133&13 & @xmath126 & 9 & @xmath109 & 6 & @xmath136 + 129 & 1822737 & 4.389e@xmath403 & 1.048 & 1.5 & 11 & @xmath126 & 9 & @xmath70 & 9 & @xmath113 & 9 & @xmath129 + 130 & 1822871 & 2.328e@xmath402 & 0.996 & 5.5 & 22 & @xmath121&21 & @xmath131&17 & @xmath97 & 8 & @xmath102 + 131 & 1828732 & 1.917e@xmath402 & 1.060 & 4.5 & 23 & @xmath54 & 9 & @xmath58 & 7 & @xmath102 & 7 & @xmath67 + 132 & 1830568 & 4.338e@xmath340 & 1.135 & 6.5 & 55 & @xmath97&26 & @xmath96 & 8 & @xmath131 & 7 & @xmath121 + 133 & 1831815 & 6.282e@xmath403 & 1.112 & 3.5 & 9 & @xmath69 & 9 & @xmath63 & 8 & @xmath96 & 8 & @xmath115 + 134 & 1835300 & 3.257e@xmath403 & 1.179 & 2.5 & 13 & @xmath86 & 8 & @xmath120 & 7 & @xmath47 & 7 & @xmath115 + 135 & 1840736 & 4.592e@xmath402 & 1.175 & 5.5 & 23 & @xmath50&17 & @xmath96 & 9 & @xmath88 & 8 & @xmath57 + 136 & 1844710 & 5.019e@xmath402 & 1.065 & 6.5 & 25 & @xmath137&25 & @xmath97&24 & @xmath131&12 & @xmath121 + 137 & 1846531 & 3.021e@xmath403 & 0.640 & 0.5 & 15 & @xmath114&13 & @xmath92&11 & @xmath55&10 & @xmath132 + 138 & 1847286 & 6.964e@xmath403 & 1.000 & 1.5 & 20 & @xmath126&13 & @xmath70 & 9 & @xmath135 & 8 & @xmath52 + 139 & 1851020 & 3.383e@xmath402 & 1.163 & 5.5 & 37 & @xmath96&17 & @xmath125 & 8 & @xmath50 & 8 & @xmath88 + 140 & 1854626 & 7.867e@xmath403 & 1.082 & 3.5 & 17 & @xmath69 & 8 & @xmath87 & 6 & @xmath109 & 5 & @xmath96 + 141 & 1859326 & 3.383e@xmath403 & 1.102 & 1.5 & 10 & @xmath138&10 & @xmath100 & 9 & @xmath101 & 8 & @xmath63 + 142 & 1862469 & 4.731e@xmath404 & 0.802 & 1.5 & 29 & @xmath75&10 & @xmath129&10 & @xmath139 & 7 & @xmath113 + 143 & 1865015 & 3.011e@xmath403 & 1.278 & 0.5 & 28 & @xmath105&15 & @xmath140 & 7 & @xmath113 & 7 & @xmath48 + 144 & 1865871 & 8.435e@xmath403 & 1.091 & 4.5 & 15 & @xmath96&14 & @xmath136&11 & @xmath95&10 & @xmath109 + 145 & 1867560 & 8.848e@xmath403 & 1.147 & 2.5 & 14 & @xmath101&10 & @xmath77 & 6 & @xmath90 & 6 & @xmath141 + 146 & 1873507 & 3.102e@xmath403 & 1.049 & 0.5 & 29 & @xmath115&13 & @xmath76&10 & @xmath113&10 & @xmath106 + 147 & 1878264 & 9.169e@xmath404 & 0.909 & 2.5 & 12 & @xmath75&11 & @xmath116 & 8 & @xmath65 & 5 & @xmath142 + 148 & 1879820 & 1.004e@xmath402 & 1.149 & 4.5 & 13 & @xmath63&10 & @xmath136 & 9 & @xmath126 & 8 & @xmath143 + 149 & 1880437 & 1.311e@xmath403 & 0.934 & 2.5 & 12 & @xmath56&11 & @xmath119 & 7 & @xmath65 & 6 & @xmath144 + 150 & 1882787 & 1.592e@xmath403 & 1.162 & 1.5 & 13 & @xmath106&11 & @xmath75&10 & @xmath115 & 9 & @xmath113 + 151 & 1883261 & 2.238e@xmath403 & 1.124 & 3.5 & 20 & @xmath108&10 & @xmath136 & 9 & @xmath49 & 8 & @xmath145 + 152 & 1888845 & 1.668e@xmath403 & 1.121 & 4.5 & 27 & @xmath95&16 & @xmath125&10 & @xmath75 & 4 & @xmath103 + 153 & 1890775 & 1.384e@xmath402 & 1.117 & 5.5 & 27 & @xmath125&17 & @xmath98&14 & @xmath96 & 9 & @xmath121 + 154 & 1891931 & 7.266e@xmath340 & 1.142 & 7.5 & 55 & @xmath97&42 & @xmath137 & & & & + 155 & 1892974 & 1.715e@xmath403 & 1.080 & 3.5 & 13 & @xmath75&12 & @xmath136&11 & @xmath120 & 6 & @xmath50 + 156 & 1893141 & 6.476e@xmath404 & 1.049 & 2.5 & 11 & @xmath124 & 9 & @xmath106 & 8 & @xmath104 & 7 & @xmath135 + 157 & 1895699 & 1.229e@xmath402 & 1.225 & 4.5 & 34 & @xmath126&15 & @xmath125&13 & @xmath65&11 & @xmath75 + 158 & 1897661 & 5.063e@xmath403 & 0.826 & 1.5 & 45 & @xmath126&10 & @xmath114 & 5 & @xmath115 & 4 & @xmath140 + 159 & 1899732 & 1.068e@xmath403 & 1.181 & 3.5 & 12 & @xmath115&12 & @xmath104 & 9 & @xmath95 & 7 & @xmath44 + 160 & 1907854 & 2.454e@xmath403 & 1.131 & 2.5 & 21 & @xmath126&13 & @xmath116&12 & @xmath115 & 7 & @xmath104 + 161 & 1910783 & 1.853e@xmath403 & 1.114 & 2.5 & 15 & @xmath126 & 9 & @xmath104 & 8 & @xmath115 & 8 & @xmath129 + 162 & 1921908 & 4.120e@xmath403 & 1.040 & 4.5 & 26 & @xmath134&16 & @xmath65&14 & @xmath98 & 8 & @xmath126 + 163 & 1925188 & 5.059e@xmath403 & 1.135 & 0.5 & 22 & @xmath106&22 & @xmath100&11 & @xmath46 & 9 & @xmath132 + 164 & 1931655 & 2.952e@xmath403 & 1.062 & 3.5 & 25 & @xmath126&17 & @xmath109&14 & @xmath125 & 5 & @xmath133 + 165 & 1931785 & 1.935e@xmath403 & 1.061 & 4.5 & 15 & @xmath128&12 & @xmath98 & 8 & @xmath125 & 7 & @xmath60 + 166 & 1934207 & 1.164e@xmath402 & 1.041 & 6.5 & 54 & @xmath137&24 & @xmath96&16 & @xmath97 & & + 167 & 1943495 & 1.936e@xmath403 & 1.116 & 5.5 & 44 & @xmath128&17 & @xmath65 & 8 & @xmath89 & 5 & @xmath134 + 168 & 1943675 & 2.381e@xmath403 & 1.401 & 0.5 & 23 & @xmath106&17 & @xmath146 & 9 & @xmath113 & 8 & @xmath140 + 169 & 1945844 & 9.744e@xmath404 & 1.121 & 2.5 & 17 & @xmath135&13 & @xmath116 & 7 & @xmath111 & 6 & @xmath75 + 170 & 1954377 & 1.372e@xmath403 & 1.158 & 1.5 & 14 & @xmath135&13 & @xmath114&10 & @xmath106 & 5 & @xmath52 + 171 & 1954660 & 1.396e@xmath403 & 1.085 & 3.5 & 13 & @xmath116&11 & @xmath136&10 & @xmath95 & 8 & @xmath126 + 172 & 1969549 & 2.916e@xmath403 & 1.110 & 3.5 & 18 & @xmath136&13 & @xmath124&10 & @xmath120 & 7 & @xmath115 + 173 & 1976415 & 1.556e@xmath403 & 1.004 & 1.5 & 22 & @xmath135&14 & @xmath139&11 & @xmath114 & 7 & @xmath118 + 174 & 1978953 & 9.192e@xmath404 & 1.125 & 4.5 & 19 & @xmath75&13 & @xmath128 & 8 & @xmath125 & 7 & @xmath136 + 175 & 1983558 & 6.177e@xmath404 & 1.064 & 1.5 & 17 & @xmath139&15 & @xmath113&10 & @xmath75 & 5 & @xmath48 + 176 & 1987517 & 4.045e@xmath402 & 1.109 & 6.5 & 35 & @xmath96&32 & @xmath131&16 & @xmath137&14 & @xmath121 + 177 & 1989260 & 1.099e@xmath403 & 1.082 & 2.5 & 13 & @xmath106 & 9 & @xmath75 & 8 & @xmath126 & 6 & @xmath86 + 178 & 1998655 & 5.870e@xmath404 & 1.070 & 2.5 & 14 & @xmath65&12 & @xmath139 & 8 & @xmath106 & 6 & @xmath75 + 179 & 2000506 & 3.300e@xmath404 & 1.231 & 3.5 & 23 & @xmath104&17 & @xmath115 & 9 & @xmath65 & 5 & @xmath124 + 180 & 2009090 & 4.252e@xmath403 & 1.164 & 1.5 & 24 & @xmath118&12 & @xmath52 & 7 & @xmath43 & 6 & @xmath138 + 181 & 2009423 & 2.085e@xmath403 & 1.011 & 2.5 & 13 & @xmath127&12 & @xmath116&12 & @xmath130 & 7 & @xmath144 + 182 & 2012363 & 4.245e@xmath340 & 1.125 & 7.5 & 55 & @xmath137&43 & @xmath97 & & & & + 183 & 2012371 & 3.978e@xmath404 & 1.077 & 3.5 & 24 & @xmath65&17 & @xmath120&10 & @xmath125 & 6 & @xmath116 + 184 & 2020332 & 4.658e@xmath404 & 1.083 & 1.5 & 18 & @xmath104&13 & @xmath123 & 9 & @xmath139 & 9 & @xmath129 + 185 & 2023581 & 9.205e@xmath404 & 1.045 & 2.5 & 18 & @xmath139&16 & @xmath120&15 & @xmath138 & 5 & @xmath127 + 186 & 2024253 & 4.292e@xmath404 & 0.450 & 0.5 & 23 & @xmath104&19 & @xmath114&17 & @xmath115 & 5 & @xmath113 + 187 & 2026321 & 3.993e@xmath402 & 1.162 & 5.5 & 34 & @xmath98&30 & @xmath125&11 & @xmath65 & 9 & @xmath128 + 188 & 2032070 & 3.112e@xmath403 & 1.166 & 4.5 & 34 & @xmath109&31 & @xmath126 & 9 & @xmath125 & 7 & @xmath98 + 189 & 2036286 & 4.545e@xmath404 & 1.065 & 5.5 & 32 & @xmath134&16 & @xmath131&16 & @xmath121&11 & @xmath65 + 190 & 2037223 & 5.150e@xmath404 & 1.070 & 4.5 & 19 & @xmath134&18 & @xmath133&15 & @xmath65 & 8 & @xmath75 + 191 & 2041950 & 4.078e@xmath404 & 1.190 & 2.5 & 17 & @xmath104&14 & @xmath115&12 & @xmath120 & 7 & @xmath106 + 192 & 2042038 & 1.366e@xmath403 & 1.171 & 1.5 & 17 & @xmath146&13 & @xmath139&12 & @xmath138&10 & @xmath123 + 193 & 2050015 & 5.729e@xmath404 & 1.414 & 1.5 & 22 & @xmath123&14 & @xmath138&11 & @xmath106 & 7 & @xmath146 + 194 & 2067603 & 5.970e@xmath404 & 1.161 & 3.5 & 15 & @xmath116 & 5 & @xmath80 & 5 & @xmath44 & 5 & @xmath87 + 195 & 2067788 & 8.705e@xmath404 & 0.880 & 0.5 & 19 & @xmath113&13 & @xmath146&13 & @xmath104&12 & @xmath105 + 196 & 2074398 & 4.469e@xmath404 & 1.121 & 3.5 & 23 & @xmath144 & 9 & @xmath133 & 9 & @xmath115 & 7 & @xmath136 + 197 & 2093800 & 4.972e@xmath404 & 1.043 & 4.5 & 24 & @xmath128&19 & @xmath65&10 & @xmath81 & 7 & @xmath95 + 198 & 2098868 & 7.315e@xmath404 & 1.120 & 5.5 & 29 & @xmath65&14 & @xmath128&12 & @xmath131 & 8 & @xmath125 + 199 & 2101721 & 8.179e@xmath404 & 1.220 & 1.5 & 20 & @xmath138 & 8 & @xmath146 & 7 & @xmath114 & 6 & @xmath76 + 200 & 2102761 & 6.455e@xmath404 & 1.177 & 2.5 & 20 & @xmath138 & 9 & @xmath106 & 9 & @xmath129 & 8 & @xmath75 + 201 & 2103251 & 6.053e@xmath404 & 1.126 & 3.5 & 15 & @xmath120&11 & @xmath75&10 & @xmath136 & 7 & @xmath144 + 202 & 2107061 & 7.605e@xmath404 & 1.163 & 4.5 & 24 & @xmath75&24 & @xmath136 & 8 & @xmath95 & 5 & @xmath63 + 203 & 2115092 & 8.220e@xmath404 & 0.433 & 0.5 & 23 & @xmath114&22 & @xmath104&14 & @xmath113&11 & @xmath146 + 204 & 2118427 & 5.520e@xmath404 & 0.905 & 1.5 & 32 & @xmath129&18 & @xmath104&10 & @xmath74 & 6 & @xmath49 + 205 & 2128377 & 5.333e@xmath404 & 1.029 & 2.5 & 19 & @xmath124&14 & @xmath104 & 9 & @xmath120 & 8 & @xmath129 + 206 & 2130494 & 4.535e@xmath404 & 1.139 & 3.5 & 47 & @xmath127&23 & @xmath144 & 7 & @xmath126 & 4 & @xmath124 + 207 & 2135286 & 4.458e@xmath404 & 1.092 & 3.5 & 25 & @xmath124 & 9 & @xmath109 & 7 & @xmath133 & 7 & @xmath116 + 208 & 2139995 & 6.770e@xmath404 & 1.068 & 6.5 & 57 & @xmath121&25 & @xmath131 & 8 & @xmath84 & 3 & @xmath53 + 209 & 2149065 & 8.702e@xmath404 & 1.087 & 1.5 & 16 & @xmath114&13 & @xmath135&13 & @xmath138 & 5 & @xmath104 + 210 & 2150540 & 5.402e@xmath404 & 1.033 & 2.5 & 19 & @xmath129&14 & @xmath135 & 9 & @xmath124 & 7 & @xmath127 + 211 & 2150766 & 8.186e@xmath404 & 1.044 & 4.5 & 31 & @xmath133&13 & @xmath134&12 & @xmath98&10 & @xmath109 + 212 & 2152006 & 1.141e@xmath403 & 1.050 & 5.5 & 35 & @xmath134&17 & @xmath121&13 & @xmath131&13 & @xmath98 + 213 & 2152478 & 1.376e@xmath403 & 1.020 & 0.5 & 42 & @xmath146&11 & @xmath100 & 8 & @xmath105 & 6 & @xmath106 + 214 & 2179055 & 7.409e@xmath404 & 1.347 & 1.5 & 32 & @xmath146&18 & @xmath113 & 9 & @xmath118 & 8 & @xmath123 + 215 & 2182142 & 7.517e@xmath404 & 1.146 & 2.5 & 23 & @xmath139&19 & @xmath138 & 6 & @xmath120 & 6 & @xmath129 + 216 & 2214969 & 2.568e@xmath404 & 1.063 & 3.5 & 19 & @xmath127&17 & @xmath144&14 & @xmath133&11 & @xmath116 + 217 & 2228247 & 1.756e@xmath404 & 0.867 & 2.5 & 40 & @xmath144&29 & @xmath127 & 6 & @xmath50 & 4 & @xmath116 + 218 & 2256608 & 2.542e@xmath404 & 0.335 & 0.5 & 19 & @xmath71&15 & @xmath43&13 & @xmath45&11 & @xmath140 + 219 & 2293388 & 3.578e@xmath404 & 1.971 & 0.5 & 28 & @xmath47&12 & @xmath117&11 & @xmath42&10 & @xmath45 + 220 & 2301564 & 2.253e@xmath404 & 1.130 & 1.5 & 27 & @xmath45&19 & @xmath71 & 8 & @xmath43 & 5 & @xmath51 + 221 & 2318231 & 2.896e@xmath404 & 1.175 & 2.5 & 20 & @xmath45&13 & @xmath51 & 6 & @xmath42 & 6 & @xmath71 + 222 & 2329881 & 4.164e@xmath404 & 1.145 & 3.5 & 16 & @xmath51&10 & @xmath45 & 6 & @xmath50 & 6 & @xmath59 + 223 & 2337488 & 3.743e@xmath404 & 1.058 & 4.5 & 12 & @xmath85&11 & @xmath102&10 & @xmath73&10 & @xmath58 + 224 & 2373640 & 2.978e@xmath404 & 1.620 & 1.5 & 29 & @xmath47&14 & @xmath42 & 9 & @xmath92 & 8 & @xmath78 + 225 & 2382670 & 1.114e@xmath404 & 1.101 & 2.5 & 9 & @xmath45 & 8 & @xmath42 & 6 & @xmath87 & 6 & @xmath47 + 226 & 2383244 & 3.921e@xmath404 & 1.051 & 3.5 & 17 & @xmath59&11 & @xmath51 & 9 & @xmath45 & 9 & @xmath64 + 227 & 2385634 & 6.034e@xmath404 & 0.237 & 0.5 & 21 & @xmath45&17 & @xmath68&13 & @xmath71 & 9 & @xmath70 + 228 & 2390409 & 6.719e@xmath404 & 1.082 & 4.5 & 16 & @xmath73&12 & @xmath51 & 9 & @xmath53 & 8 & @xmath64 + 229 & 2404064 & 8.347e@xmath404 & 1.057 & 5.5 & 24 & @xmath73&13 & @xmath79&13 & @xmath53&10 & @xmath82 + 230 & 2404688 & 2.996e@xmath404 & 0.938 & 1.5 & 16 & @xmath49&15 & @xmath68 & 8 & @xmath45 & 6 & @xmath110 + 231 & 2416352 & 2.982e@xmath404 & 1.041 & 2.5 & 14 & @xmath57&12 & @xmath49 & 6 & @xmath110 & 6 & @xmath111 + 232 & 2424028 & 8.706e@xmath404 & 1.060 & 6.5 & 26 & @xmath79&25 & @xmath84&17 & @xmath89&11 & @xmath73 + 233 & 2424548 & 4.176e@xmath404 & 0.999 & 3.5 & 12 & @xmath57&12 & @xmath59&10 & @xmath111 & 7 & @xmath107 + 234 & 2429397 & 8.010e@xmath404 & 0.986 & 4.5 & 27 & @xmath88&16 & @xmath59&10 & @xmath112 & 8 & @xmath82 + 235 & 2450180 & 2.385e@xmath404 & 1.237 & 1.5 & 10 & @xmath46 & 9 & @xmath71 & 9 & @xmath68 & 7 & @xmath72 + 236 & 2450798 & 1.483e@xmath404 & 1.228 & 3.5 & 17 & @xmath71 & 9 & @xmath44 & 9 & @xmath83 & 8 & @xmath119 + 237 & 2463349 & 1.537e@xmath404 & 1.245 & 2.5 & 17 & @xmath71&14 & @xmath47 & 8 & @xmath44 & 7 & @xmath87 + 238 & 2474294 & 9.119e@xmath404 & 1.179 & 3.5 & 10 & @xmath77 & 8 & @xmath57 & 8 & @xmath55 & 7 & @xmath42 + 239 & 2475624 & 1.786e@xmath404 & 1.005 & 2.5 & 20 & @xmath50&17 & @xmath119 & 6 & @xmath52 & 6 & @xmath108 + 240 & 2482952 & 4.099e@xmath403 & 1.161 & 4.5 & 20 & @xmath51&10 & @xmath53 & 9 & @xmath73 & 7 & @xmath61 + 241 & 2484114 & 2.607e@xmath404 & 1.214 & 2.5 & 18 & @xmath48&12 & @xmath91 & 8 & @xmath46 & 8 & @xmath69 + 242 & 2505601 & 2.979e@xmath404 & 0.870 & 1.5 & 24 & @xmath49&10 & @xmath43&10 & @xmath90 & 9 & @xmath91 + 243 & 2509043 & 1.344e@xmath404 & 1.150 & 3.5 & 13 & @xmath62&12 & @xmath83 & 8 & @xmath63 & 6 & @xmath71 + 244 & 2512282 & 1.205e@xmath403 & 1.228 & 0.5 & 21 & @xmath70&12 & @xmath41&11 & @xmath47 & 9 & @xmath68 + 245 & 2512970 & 1.773e@xmath402 & 1.108 & 5.5 & 26 & @xmath82&14 & @xmath59&13 & @xmath64&12 & @xmath73 + 246 & 2530193 & 1.830e@xmath402 & 1.048 & 6.5 & 36 & @xmath84&22 & @xmath79&16 & @xmath89&12 & @xmath73 + 247 & 2535719 & 4.458e@xmath404 & 1.344 & 1.5 & 12 & @xmath76&11 & @xmath41&10 & @xmath122 & 8 & @xmath49 + 248 & 2536655 & 2.087e@xmath404 & 0.985 & 3.5 & 16 & @xmath59&16 & @xmath67 & 8 & @xmath50 & 7 & @xmath80 + 249 & 2538392 & 5.998e@xmath404 & 1.067 & 2.5 & 11 & @xmath49 & 8 & @xmath74 & 7 & @xmath110 & 7 & @xmath111 + 250 & 2542661 & 6.466e@xmath404 & 1.354 & 0.5 & 15 & @xmath72&14 & @xmath47&13 & @xmath46&11 & @xmath41 + 251 & 2543216 & 4.617e@xmath404 & 1.123 & 4.5 & 18 & @xmath103&11 & @xmath88 & 9 & @xmath63 & 8 & @xmath50 + 252 & 2546808 & 5.833e@xmath404 & 0.988 & 2.5 & 16 & @xmath57&13 & @xmath52&13 & @xmath50 & 9 & @xmath43 + 253 & 2566921 & 2.050e@xmath404 & 1.187 & 3.5 & 16 & @xmath71 & 9 & @xmath87 & 9 & @xmath60 & 8 & @xmath107 + 254 & 2568158 & 2.777e@xmath404 & 1.209 & 1.5 & 15 & @xmath43&13 & @xmath92 & 8 & @xmath68 & 8 & @xmath130 + 255 & 2569207 & 5.521e@xmath405 & 1.069 & 2.5 & 17 & @xmath87&10 & @xmath52 & 9 & @xmath64 & 6 & @xmath110 + 256 & 2574020 & 2.857e@xmath404 & 1.091 & 4.5 & 15 & @xmath83&13 & @xmath107&10 & @xmath61 & 9 & @xmath102 + 257 & 2574923 & 2.673e@xmath404 & 1.104 & 5.5 & 28 & @xmath102&28 & @xmath81&19 & @xmath61&10 & @xmath89 + 258 & 2577404 & 4.730e@xmath405 & 0.809 & 0.5 & 19 & @xmath68&15 & @xmath43&13 & @xmath118 & 8 & @xmath41 + 259 & 2583652 & 2.723e@xmath404 & 1.066 & 3.5 & 27 & @xmath67&14 & @xmath50&11 & @xmath94 & 7 & @xmath63 + 260 & 2594129 & 1.200e@xmath404 & 1.056 & 4.5 & 36 & @xmath58&13 & @xmath54 & 7 & @xmath60 & 6 & @xmath103 + 261 & 2598239 & 3.975e@xmath405 & 1.174 & 2.5 & 19 & @xmath91&14 & @xmath48&11 & @xmath69&10 & @xmath90 + 262 & 2604784 & 2.248e@xmath404 & 1.059 & 3.5 & 26 & @xmath86&17 & @xmath112 & 7 & @xmath87 & 5 & @xmath119 + 263 & 2604983 & 3.030e@xmath404 & 1.075 & 1.5 & 15 & @xmath132&13 & @xmath99&12 & @xmath91 & 7 & @xmath90 + 264 & 2605314 & 1.674e@xmath404 & 1.149 & 2.5 & 17 & @xmath110&12 & @xmath69&10 & @xmath108 & 8 & @xmath68 + 265 & 2639152 & 3.367e@xmath404 & 1.435 & 0.5 & 18 & @xmath118&18 & @xmath76&16 & @xmath117&13 & @xmath70 + 266 & 2651589 & 2.598e@xmath404 & 1.081 & 1.5 & 23 & @xmath130&11 & @xmath63 & 8 & @xmath100 & 8 & @xmath101 + 267 & 2655168 & 2.948e@xmath404 & 1.272 & 3.5 & 16 & @xmath62&10 & @xmath63 & 8 & @xmath80 & 8 & @xmath55 + 268 & 2664175 & 3.863e@xmath404 & 1.176 & 4.5 & 15 & @xmath60&14 & @xmath51 & 9 & @xmath107 & 7 & @xmath58 + 269 & 2665465 & 3.099e@xmath404 & 1.117 & 5.5 & 23 & @xmath81&20 & @xmath61&13 & @xmath102&13 & @xmath89 + 270 & 2675532 & 3.546e@xmath404 & 1.117 & 3.5 & 13 & @xmath111 & 9 & @xmath86 & 8 & @xmath83 & 6 & @xmath71 + 271 & 2693487 & 7.979e@xmath405 & 1.272 & 1.5 & 17 & @xmath101&12 & @xmath100&12 & @xmath140 & 8 & @xmath122 + 272 & 2697685 & 3.749e@xmath404 & 0.999 & 0.5 & 29 & @xmath70&19 & @xmath72 & 9 & @xmath76 & 9 & @xmath46 + 273 & 2698564 & 1.819e@xmath403 & 1.079 & 4.5 & 16 & @xmath103&14 & @xmath88 & 9 & @xmath50 & 8 & @xmath63 + 274 & 2703296 & 1.042e@xmath404 & 1.062 & 2.5 & 15 & @xmath69&10 & @xmath52 & 8 & @xmath87 & 6 & @xmath43 + 275 & 2706142 & 2.115e@xmath404 & 1.089 & 2.5 & 19 & @xmath74&15 & @xmath94 & 9 & @xmath108 & 6 & @xmath110 + 276 & 2709104 & 1.566e@xmath404 & 0.892 & 1.5 & 16 & @xmath90&14 & @xmath130&11 & @xmath77&10 & @xmath132 + 277 & 2740035 & 1.759e@xmath404 & 1.109 & 3.5 & 26 & @xmath56&14 & @xmath54&12 & @xmath119 & 7 & @xmath62 + 278 & 2752237 & 2.348e@xmath404 & 1.035 & 2.5 & 21 & @xmath108&12 & @xmath130 & 9 & @xmath101 & 8 & @xmath119 + 279 & 2758501 & 1.742e@xmath404 & 1.229 & 1.5 & 16 & @xmath100&14 & @xmath140&13 & @xmath101 & 7 & @xmath132 + 280 & 2801938 & 2.651e@xmath404 & 0.709 & 0.5 & 53 & @xmath118&14 & @xmath43 & 6 & @xmath140 & 5 & @xmath70 + +

based on two - dimensional high resolution hydrodynamic numerical simulation , we study the mechanical and radiative feedback effects from the central agn on the cosmological evolution of an isolated elliptical galaxy . physical processes such as star formation and supernovae are considered . the inner boundary of the simulation domain is carefully chosen so that the fiducial bondi radius is resolved and the accretion rate of the black hole is determined self - consistently . in analogy to previous works , we assume that the specific angular momentum of the galaxy is low . it is well - known that when the accretion rates are high and low , the central agns will be in cold and hot accretion modes , which correspond to the radiative and kinetic feedback modes , respectively . the emitted spectrum from the hot accretion flows is harder than that from the cold accretion flows , which could result in a higher compton temperature accompanied by a more efficient radiative heating , according to previous theoretical works . such a difference of the compton temperature between the two feedback modes , the focus of this study , has been neglected in previous works . significant differences in the kinetic feedback mode are found as a result of the stronger compton heating and accretion becomes more chaotic . more importantly , if we constrain models to correctly predict black hole growth and agn duty cycle after cosmological evolution , we find that the favored model parameters are constrained : mechanical feedback efficiency diminishes with decreasing luminosity ( the maximum efficiency being ) and x - ray compton temperature increases with decreasing luminosity , although models with fixed mechanical efficiency and compton temperature can be found that are satisfactory as well . we conclude that radiative feedback in the kinetic mode is much more important than previously thought .

1403.0670

there is accumulating evidence showing that the evolution of host galaxies is tightly related to their central supermassive black holes ( smbhs ) @xcite . the most linkages are the strong correlations between the mass of smbh and properties of the galactic spheroid , including luminosity @xcite , stellar velocity dispersion @xcite and stellar mass @xcite . the general consensus is that powerful feedback from the central active galactic nuclei ( agns ) should play an important role in the formation and evolution processes of their host galaxies . simple energetic arguments indicate that agns should be capable of heating the inner regions of galaxies to offset radiative cooling and further regulate the black hole growth ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? many authors have proposed and performed successful theoretical models and numerical simulations to investigate how the feedback plays the role . however , because the length scales associated with agn activity are so tiny compared to their host galaxy , the feedback process must operate over a huge dynamical range . this implies that the detailed agn feedback mechanisms must be diverse and explains why this process is still incompletely understood ( for reviews , see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? to study agn feedback , ideally , we should cover the whole range of lengths and timescales , from black hole event horizon to the galactic scales . obviously , such a huge dynamical range is technically almost impossible to achieve by current 2- and 3-dimensional numerical simulations . in reality , different works focus on feedback at different scales . at the smallest black hole accretion scale , works have been done on outflow production driven by line force ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and on the radiative feedback due to the global compton scattering @xcite . the latter effect has been invoked to explain the intermittent behavior of the black hole activity of some compact young radio sources @xcite . on the other hand , many works ( e.g. * ? ? ? * ; * ? ? ? * ) have focused on the much larger galactic scale , from @xmath1 pc to tens of kpc and the timescale from a fraction of myr to several gyr . brighenti & mathews ( e.g. , @xcite ; @xcite , see also @xcite @xcite ) investigated the agn heating on the ism in details , and found that the thermal feedback could break the cool - core structures which are observed in many clusters , i.e. , the over - heating problem . gaspari et al . ( e.g. , @xcite ; @xcite ) performed detailed simulations on the mechanical feedback , and they found that the mechanical feedback is favoured to solve the cooling flow problem , to explain to cold rims , and meantime to avoid over - heating . many works on the agn feedback have been done in the context of isolated elliptical galaxies by ciotti , ostriker and their collaborators , both in one dimension @xcite and two dimensions @xcite . in these works , the inner boundary is small enough to resolve the black hole accretion flow , typically a few pc . this is important since it can compile the accretion rate in a physical way , which is crucial to evaluate the effect of agn feedback @xcite . when the bondi radius @xcite is not resolved , other more approximate methods of estimating the accretion rate ( e.g. , see * ? ? ? * ) must be utilized . the outer boundary is large enough to reach hundreds of kpc , i.e , the galactic outskirts . the timescale also covers a large range , from accretion timescale to galaxy evolution timescale . the interaction of the radiation and winds from the inner agns with the galactic gas and their effects on regulating the accretion are considered , together with physical processes such as supernovae heating and star formation . overall , the above mentioned works evidenced that the most satisfactory models are the combined models with both mechanical feedback and radiative feedback , in which mechanical feedback is very efficient to regulate the black hole growth , and radiative feedback can help to balance cooling and modulate the gas dynamics on the galactic scale . as is well - known , two `` modes '' of agn feedback have been identified @xcite . when the mass accretion rate is relatively large , @xmath2 ( where the eddington rate is defined as @xmath3 ) , we have the radiative or quasar mode . in this regime , a standard thin disk is operating in the central region of agns , the luminosity is high , and a strong wind is observed ( see @xcite @xcite for a review ) . when the accretion rate is lower , the agn is said to be in the kinetic mode , also known as the `` radio '' or `` maintanence '' mode . this mode corresponds to low - luminosity agns ( llagns ) . in this regime , the accretion flow is expected to be in the hot phase ( narayan & yi @xcite , @xcite ; see yuan & narayan @xcite for a recent review of the theoretical aspects of this model and its various astrophysical applications ) . the spectrum from a hot accretion flow is quite different from that of a standard thin disk , as we will describe in more detail in [ sec : radio - mode - feedback ] . moreover , recent theoretical studies indicate that , in addition to jet , winds also exist in hot accretion flows @xcite . observationally , such a theoretical prediction has been confirmed by the chandra observation to the accretion flow around the supermassive black hole in our galactic center @xcite . the importance of radiative feedback in the quasar mode is obvious and well recognized . however , in the radio mode , the role of radiation is usually assumed to be minor compared to mechanical feedback by winds or jet . the reasons are possibly twofold . first , it is usually thought that the radiative power is low compared to the kinetic power of jet . second , the efficiency of radiative heating of the emitted radiation on the interstellar medium is assumed to be not high . however , the radiative luminosity may actually be larger than the kinetic outflow power for luminosities @xmath4 @xcite . it should be noted that , regarding the radiative heating efficiency , it depends not only on the bolometric luminosity , but also the spectral energy distribution . taking the compton heating as an example , the compton heating is proportional to both the total radiative flux and the compton temperature @xmath5 ( defined as the effective temperature of the radiation field , * ? ? ? * ) , and the value of @xmath5 is determined by the spectral energy distribution ( see eq . [ eq : compton - temperature ] ) : in the quasar mode , typically @xmath6k @xcite , while in the radio mode , as we will describe in detail in [ sec : radio - mode - feedback ] , @xmath5 is likely to be as high as @xmath7k , which is about two orders of magnitude higher than that in the quasar mode @xcite . such a difference of @xmath5 in the two feedback modes has not been considered in the works mentioned above ; rather , it is often assumed to maintain the same @xmath5 at different feedback modes . therefore , the radiative feedback in the radio mode could potentially be much more important than previous thought . for example , @xcite studied two models with different mechanical feedback efficiency @xmath8 . in the first class of models ( `` models a '' ) , @xmath8 is a constant . in another family ( `` models b '' ) , the efficiency decreases with the decreasing luminosity . although there are some initial evidence that the assumption of model b leads to more realistic results , @xcite find that model b predicts too much growth of black hole mass , and incorrect agn duty cycle . will the results be changed when we correctly consider a stronger compton heating in the radio mode ? one of our main aims of the present work is to study the effect of a variable @xmath5 . in the present paper , following @xcite , we perform two - dimensional high resolution hydrodynamical numerical simulations to study the effect of agn feedback on the evolution of an isolated elliptical galaxy . both mechanical and radiative interaction are considered . special attention is paid to the effect of radiative feedback . especially , the compton temperature is self - consistently determined from the different accretion modes and a high ( consistent ) @xmath5 is adopted when the accretion is in the hot mode . the paper is organized as follows . in 2 , we introduce the physics of our model , including how do we treat mechanical and radiative feedback , the galaxy model we adopt , and the numerical treatments of stellar mass loss , supernovae , and star formation . the setup and boundary conditions of numerical simulation are described in 3 . our results are presented in 4 , while 5 is devoted to summary and discussion . in this section we introduce the physics of our feedback model . many aspects of the simulations such as the galaxy model , the treatment of stellar evolution , and the hydrodynamical equations are the same as in @xcite ; for completeness , we briefly recall them . in order to resolve the bondi radius , we carefully choose the inner boundary of our computational domain . yet , since the dynamics of accretion within the bondi radius ca nt be resolved , we still need to use a sub - grid model to describe the black - hole accretion onto the central agn . in order to guarantee accretion , in absence of explicit viscosity terms , following @xcite , the specific angular momentum of the gas is assumed to be low . in this case , it is generally assumed that a small disk will be formed at a small circularization radius inside the bondi radius . one important aspect to consider is the timescale for the gas at the inner boundary to fall onto the black hole . this is the sum of two timescales @xcite : ( 1 ) the free - fall timescale @xmath9 from the inner boundary to the circularization radius and ( 2 ) the accreting timescale @xmath10 at the circularization radius , which is determined by the viscous timescale of the small accretion disk . once the accretion timescale is known , it is possible to evaluate the mass accretion rate onto the black hole . first , we can calculate easily the mass inflow rate at the inner boundary of the simulation domain . as we described in 1 , only a part of the gas can finally reach the black hole because of the mass outflow from the unresolved inner disk . we then need to subtract the mass outflow rate to get the net mass accretion rate @xmath11 at the black hole horizon and subsequently trace the black hole growth by updating the black hole mass at each time step . the instantaneous radiative luminosity of the agn is determined by @xmath12 where @xmath13 is the radiative efficiency and @xmath14 is the speed of light . if the accretion flow is described by a standard thin disk ( i.e , the agn is in the quasar mode ) , the efficiency is only a function of black hole spin and is independent of accretion rate @xmath15 . we adopt @xmath16 in the present work . when the accretion is in the hot accretion mode ( i.e. , the agn is in the radio mode ) , the value of @xmath13 becomes smaller and is a function of @xmath15 . @xcite present the detailed study to this problem , and provide fitting formulae of @xmath13 as a function of @xmath15 . however , in the present paper , in order to isolate the effect of a time - dependent compton temperature @xmath5 , we simply follow previous works and adopt the following recipe @xmath17 with @xmath18 and @xmath19 ( @xcite @xcite , see also * ? ? ? we note that the efficiency described by the above equation is in general lower than the more exact results presented in @xcite . in the equation above , the dimensionless mass accretion rate is defined as @xmath20 where @xmath21 is the eddington luminosity . the interaction between the radiation and the galactic ism proceeds through two channels . one is via radiation pressure , and the other is the radiative heating . when evaluating radiative heating via compton scattering , an important concept is the compton temperature , which is the energy - weighted average photon energy of the radiation emitted by the agn . it is defined as ( e.g. , * ? ? ? * ) @xmath22 where @xmath23 is radiative flux , and @xmath24 and @xmath25 are the plank and boltzmann constants , respectively . obviously , the value of @xmath5 is determined by the spectrum of the radiation from the agn . sazonov et al . ( @xcite ) assessed a full range of observational data of quasars , computed the associated spectral energy distribution ( sed ) , and finally calculated the typical compton temperature of bright quasar , obtaining @xmath26 k. such a temperature is well above the central temperature of the cooling flow gas . so it is natural to expect that radiative feedback from quasars could have profound affects on their host galaxies . in fact , the radiative heating can be written as ( e.g. , * ? ? ? * ) @xmath27 where @xmath28 is the ionization parameter , @xmath29 is the flux of photoionizing photons and @xmath30 is the number density of the ism . now the question is what is the value of @xmath5 of the radiation spectrum produced by llagns , i.e. , in the radio mode . the previous works adopted the same value as in the quasar mode . however , observations show that the spectra produced by llagns and quasar are quite different , with the most characteristic one being the lack of the `` big blue bump '' in llagns ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * see also the review by yuan & narayan @xcite ) . in addition , the spectrum of llagns is more `` x - ray loud '' than that of quasar ( refer to fig . 7 in * ? ? ? * ) . this means that the spectra of llagns are much harder than that of quasars , i.e. , there is a larger fraction of high - energy photons . @xcite have made an initial calculation to the emitted spectrum from a hot accretion flow and found that the corresponding compton temperature @xmath5 could be as high as @xmath31 k. we would like to emphasize that there are two caveats in this calculation . one is that such value of @xmath5 is calculated from the theoretical spectrum emitted from hot accretion flows . in reality , since the accretion model of llagns also include a truncated thin disk @xcite , we should also include the contribution of such a thin disk . so that the best way is to directly calculate @xmath5 from the observed spectra of llagns at various luminosities . taking an llagn ngc3998 as an example , we have calculated the corresponding @xmath5 following eq . the multiwaveband spectrum of this source is combined in @xcite . the result is @xmath32k , although there are some uncertainties in the value of the high - energy cutoff in its x - ray spectrum . so this seems to indicate that the `` theoretical spectrum '' adopted in @xcite is not a bad approximation at least for some sources , although it is still necessary to check the other sources with various luminosities . the second caveat , which is perhaps more important , is that the relativistic effects are not included in @xcite and eq . ( 4 ) . obviously , combining the sed of various agns with various luminosities and calculating their @xmath5 after take into account all relativistic effects will be an important work ( yuan et al . , in preparation ) . including relativistic effects is likely reduce the value of @xmath5 , which will then weaken the radiative heating in the kinetic mode . however , this effect may be cancelled in some degree since the radiative efficiency we adopt in the present paper ( i.e. , eq . 2 ) is lower than the exact value calculated in @xcite . given these complications , in the present work , we simply adopt @xmath33 k. such a large value of @xmath5 means that even though the bolometric luminosity of llagns is low , its radiative feedback effects are potentially important . this is further strengthened by the possibility that galaxies may spend most of their time in the llagn phase . in this paper we intend to evaluate quantitatively effects of radiative feedback in the radio mode . to this aim we study two types of models , namely model `` f '' and model `` v '' . model `` f '' is for comparison purpose . in this model , the compton temperature is fixed to be @xmath34k , irrespective of the luminosity of the accretion flow , just as in previous works ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in model `` v '' , the compton temperature of the spectrum is variable , and is a function of the luminosity of the accretion flow . specifically , when the accretion rate is high , i.e. , the system is in the quasar mode , we adopt @xmath34k ; when the accretion rate is low , i.e. , the system is in the radio mode , we adopt @xmath35k . the key question is how to treat the transition between the radio and quasar modes . since the physics of black hole accretion is presumably independent of the black hole mass , we can gain some insight from the observations of black hole x - ray binaries . with the change of accretion rates , this kind of sources transit between two distinct states , namely soft and hard ones . the spectra of the two states are quite different , with the hard state being much harder than the soft state . the accretion flow in these two states are the standard thin disk and hot accretion flows respectively @xcite , and the physics of the transition between these two accretion modes is reviewed in @xcite . the dividing luminosity between the two states is @xmath36 @xcite , so that in model `` v '' we adopt the compton temperature as @xmath37 in addition to radiation , the nuclear wind from accretion flow is another important channel of agn feedback . we neglect the role of jet since ( at least for an isolated galaxy ) it is believed that jets simply drill through the surrounding gas and have little effect to the galaxy . winds provide mass , energy and momentum rushing into the ism . the existence of wind in the quasar mode has been directly confirmed by observations , e.g. , via the broad absorption lines in some quasars ( e.g. , * ? ? ? * ; * ? ? ? * ) . in the case of radio mode , outflows have also been detected recently in llagns . @xcite investigated the outflow from a sample of nearby agns . of the ten nearby seyfert 1 galaxies in their sample , six sources still have their bolometric luminosities below @xmath38 . their detailed study of the uv and x - ray absorbers clearly shows that a strong outflow exists in these sources . the bolometric luminosity of one source , ngc 4395 , is even as low as @xmath39 . most recently , the existence of outflow in sgr a * has been confirmed by chandra observations @xcite . theoretically , the existence of winds have been confirmed by mhd numerical simulation , although agreement has not been achieved on the mass flux of winds @xcite . following previous works , we adopt a phenomenological approach and notate the properties of wind , namely the fluxes of mass , energy , and momentum , as follows @xcite , @xmath40 where @xmath41 is the velocity of the wind , and @xmath8 is the mechanical efficiency , which describes the ratio of the wind power and the accretion power . as discussed in detail in @xcite , these expressions guarantee that the mass , energy , and momentum carried by the wind are self - consistent . the velocity of the wind in the case of quasar mode is relatively well observed , which is roughly @xmath42 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in the radio mode , however @xmath41 is poorly constrained by observations . from the theoretical point of view , yuan et al . ( @xcite ; see also * ? ? ? * ) discuss in a preliminary essay the terminal velocity of wind based on their mhd numerical simulation of hot accretion flow . more systematic study is ongoing ( yuan et al . 2014 in preparation ) . in the present work , we tentatively adopt the same @xmath41 as in the quasar mode . as for the mechanical efficiency @xmath8 , even in the case of relatively better observed quasar mode , its value is poorly constrained . given this situation , @xcite considered two prescriptions of mechanical feedback , and the corresponding families of models were indicated as `` models @xmath43 '' and `` models @xmath44 '' . for models @xmath43 , both the mechanical efficiency ( denoted as @xmath45 ) and the wind opening angles are independent of the accretion rate . in models @xmath44 , these two quantities vary with the accretion rate , and are arranged to be small at small accretion rates and reach a specified maximum at the eddington rate ( see also * ? ? ? that is , we have @xmath46 } \cr \displaystyle{{3\over 4}\cdot{\epsilon_w^m ~ l\over 1 + 0.25 l}},\quad\quad\quad\quad\ , { \rm [ b]}}\ ] ] where @xmath47 . it follows that the efficiency of models b is in general much lower than in models a. when @xmath48 , the mechanical efficiency @xmath8 of models b is identical to that of models a , but will drop rapidly with the decrease of @xmath49 . a natural question is then , which one is more realistic among model a and b. in the case of hot accretion flow , in which the outflow is likely mainly produced by mhd mechanism @xcite , it is still an open question how the efficiency changes with luminosity . but in the case of radiative line - driven winds , numerical simulations ( focused on the inner few hundred @xmath50 , where @xmath51 and @xmath52 is the gravitational constant ) indicate that @xmath8 does fall off when the luminosity decreases @xcite . in this sense , model b seems to be more realistic . following @xcite , we parameterize the angular distribution of the nuclear wind properties as @xmath53 where @xmath54 is a conserved quantity ( mass , energy , or radial momentum ) , @xmath55 is the total amount of the conserved quantity to be injected , and @xmath56 is the solid angle covered by winds . we set @xmath57 so that the half - opening angle enclosing half of the mechanical energy is @xmath58 . in terms of solid angle , this means that the wind is visible from @xmath59 of the available viewing angles @xcite . this fraction is also in agreement with observations of the fraction of obscured and unobscured agns @xcite under the assumption that the two populations are made up of a single population of objects that differ only in viewing angle . under the above assumptions , we can conveniently introduce in the numerical simulations the wind feedback by injecting the desired mass , energy , and momentum into the innermost cells of the simulation domain and self - consistently compute the radial transport of these quantities . the gravitational potential of the galaxy is contributed to by a dark matter halo and a stellar spheroid embedded in it , with a central black hole . as common , we ignore the self - gravity of the ism in the simulations . we note that the stellar population is of critical importance in our galactic scale simulations because : ( i ) the gravity due to the stellar spheroid is dominant at the scale of @xmath60 kpc from the galaxy center ; ( ii ) abundant gas will be released during the stellar evolution , which is the main gas resource of material fueling the black hole in our simulation . in fact , over a cosmological time span , the evolving stellar population will inject gas of mass summing up to @xmath61 of the total initial stellar mass into the galaxy , and this injected mass is therefore @xmath62 two orders of magnitude larger than the black hole mass observed in elliptical galaxies ( as @xmath63 , see * ? ? ? * ; * ? ? ? the stellar mass losses are thermalized by the relative motion of the stars and ism ( e.g. , * ? ? ? * ) ; in addition , supernova explosions provide additional source of mass , metals and energy . a full description of the build - up of the gaseous halo of the galaxy , up to onset of the first `` cooling catastrophe '' , has been given elsewhere ( e.g. , * ? ? ? here is sufficient to recall that when the ism density reaches a critical value , suddenly a large amount of cooling gas falls toward the center , a part of it accretes onto the central black hole , a part forms stars , and the remaining is expelled from the central region , until a new cycle repeats . in a word , the stellar population not only determines the source of the accreting gas but also is important to the gaseous dynamics and to the black hole growth . in the present models , we adopt the jaffe profile for the stellar component , @xmath64 where @xmath65 and @xmath66 are the total stellar mass and the scale - length of the galaxy , respectively . observations show that the total density distributions of ellipticals can usually be well described by a @xmath67 profile over a large radial range @xcite . for this reason , we set the density profile of dark matter halo so that the total mass profile decreases as @xmath67 . finally , the initial black hole mass is determined according to the magorrian relation @xcite . the velocity dispersion field of the stellar population is characterized by a central projected velocity dispersion @xmath68 , and all the other dynamical quantities relevant for the simulations are given elsewhere ( e.g. , * ? ? ? we calculate the stellar evolution during the simulations following @xcite & @xcite . in practice , the mass loss rate from the evolving stellar population declines as @xmath69 , and it is injected into the computational domain proportionally to @xmath70 . stellar evolution contributes further to the gas dynamics via type ia supernova ( sn ia ) explosions , which are the results of degenerate white dwarfs , e.g. , accreting white dwarfs which finally reach the chandrasekhar limit and mergers of two white dwarfs @xcite . we assume that each sn ia ejects @xmath71 erg of energy and @xmath72 of material into the ism . as anticipated , the sn ia rate evolves secularly as @xmath73 , so that the specific heating of the injected gas increases with time . the metal rich ism is an ideal place for the onset of radiative cooling instability and this , in combination with agn feedback , leads to inevitable recurrent starbursts @xcite . in the simulations , we compute the star formation rate at each radius @xmath74 using the standard schmidt - kennicut scheme . among the newly formed stars , there is a population of massive stars whose masses are @xmath75 . the massive stars have a relatively short lifetime and will finally evolve to type ii supernovae ( sne ii ) on a timescale of @xmath76 years . as for sn ia , we assume each sn ii ejects @xmath71 erg of energy into the ism and leaves behind a neutron star of @xmath72 . in our simulations , the evolution of the galaxy under the effect of the central agn is described by the following time dependent eulerian equations of hydrodynamics ( see @xcite @xcite for a full description ) : @xmath77 @xmath78 @xmath79 where @xmath80 , @xmath81 and @xmath82 are the gas mass , momentum and internal energy per unit volume , respectively , and @xmath83 is the velocity , @xmath84 is the gas pressure , and we adopt an adiabatic index @xmath85 . @xmath86 and @xmath87 are the net rates of radiative heating and cooling respectively , @xmath88 is the gravitational field of the galaxy ( i.e. , stars , dark matter , plus the time - dependent contribution of the growing central smbh ) . for simplicity , we do not take into account effects of the self - gravity of ism or the gravitational effect of the mass redistribution due to the stellar mass loss and star formation . the mass source term @xmath89 is the stellar mass loss rate , and @xmath90 corresponds to the thermalization of the stellar mass loss due to stellar velocity dispersion . we let @xmath91 and @xmath92 be the feedback rates of energy from type i and ii supernovae respectively . here @xmath93 is the mass return from short - lived type ii supernovae and @xmath94 , @xmath95 and @xmath96 are the sink terms of mass , momentum and energy due to star formation , respectively . we briefly recall the heating processes below , for more details we refer the readers to ciotti & ostriker ( e.g. , @xcite ) . in the energy equation ( eq . [ eq : engconsvr ] ) , we include 4 semi - analytic heating terms , i.e. , the radiative heating @xmath86 , stellar thermalisation @xmath90 , and supernovae heating @xmath91 and @xmath92 , which all are expected in real galaxies . however , it is hard to tell which process dominates over the others , because each heating process plays different roles in different regions , at different time . with stellar thermalisation @xmath90 , the ism could be heated up onto a level of the stellar velocity dispersion it is not enough to drive outflow . the heating due to supernovae ( i.e. , @xmath91 and @xmath92 ) is able to drive galactic - scale outflow but only on the galactic outskirt as the gravity is weak there , while the radiative heating @xmath86 could affect the ism in the whole galaxy , especially in the galaxy centre . as a qualitative comparison , the relative importance of the stellar thermalisation @xmath90 is always marginal compared to the supernovae heating , except in the innermost region where the stellar velocity dispersion increases because of the gravity of the supermassive black hole . regarding the supernovae heating items , sne ia heating @xmath97 is related to old stellar population , which is a secular process . while sne ii heating @xmath92 depends on the star formation , which is a transient process . generally speaking , @xmath91 dominates over @xmath92 , except where starburst occurs . in brief , the radiative heating @xmath86 and stellar heating processes ( @xmath90 , @xmath91 and @xmath92 ) play important roles in the inner region and the outskirt of the galaxy respectively , of course the relative importance also depends on the agn luminosity . radiative heating and cooling are computed by using the formulae presented in @xcite , which describe the net heating / cooling rate per unit volume of a cosmic plasma in photoionization equilibrium . in particular , compton heating and cooling , bremsstrahlung loss , photoionization , line and recombination cooling , are taken into account . the total radiation pressure gradient in eq.([eq : momconsvr ] ) can be divided into two parts @xmath98 radiation pressure due to _ electron scattering _ is computed as @xmath99 where @xmath100 is the electron scattering opacity . the radiation pressure due to all of the radiatively heating processes is calculated simply as ( see @xcite @xcite for details ) @xmath101 following @xcite , we choose the galaxy parameters so that the model obeys to the edge - on view of the fundamental plane @xcite and to the faber - jackson ( @xcite ) relation , in particular , the total stellar mass , @xmath102 , and the galaxy effective radius @xmath103 kpc , plus the contribution of the dark matter halo corresponding to a central projected stellar velocity dispersion @xmath104 , and to a stellar mass - to - light ratio @xmath105 in solar units . the initial mass of the black hole is @xmath106 . the gas density is initially set to be a very low value so that the gas in the simulations comes almost exclusively from explicit source terms arising from stellar evolution . this is often termed `` secular evolution '' to distinguish it from evolution induced by cosmological effects ( such as galaxy merging ) . we perform two dimensional hydrodynamic simulations with _ zeus - mp/2 _ @xcite in spherical coordinates ( @xmath74 , @xmath107 , @xmath108 ) , i.e. , all quantities are assumed to be axisymmetric . following @xcite , in the @xmath107 direction the mesh is divided homogeneously into 30 angular cells , while for the radial direction ( covering the radial range of 2.5 pc - 250 kpc ) , we use a logarithmic mesh with 120 bins to reach the needed resolution in the inner computational region . with this choice each cell is 10% larger than the previous inner cell . it is most important to resolve the fiducial bondi radius so that the accretion rate can be robustly estimated . in particular , the inner boundary radius ( @xmath109 pc ) is chosen to be within the compton and bondi radii when the compton temperature @xmath26 k. the compton radius @xmath110 is where the compton temperature is equal to the local virial temperature , mainly determined by @xmath111 @xcite . however , @xmath110 is not resolved for high @xmath5 , as in models `` v '' when the eddington ratio is low , i.e. , @xmath112 k and then @xmath113 pc with a black hole mass of @xmath114 . the typical values of the bondi radius ( for the central gas temperature and density of elliptical galaxies ) range between 10 and 100 pc . the bondi radius is estimated by using the sound speed @xmath115 at the inner boundary , @xmath116 since @xmath117 varies during the simulations , in the code the radial inflowing velocities at the inner boundary are limited to @xmath118 in case that the bondi radius is unresolved . this helps to avoid an unphysical large accretion rate . we stress that in the simulations bondi accretion _ is not imposed . _ the simulations in our paper are very time - consuming despite of the low numerical resolution . this is because of the large dynamical range . technically , it is too computationally expensive to implement both large dynamical range and high numerical resolution simultaneously . one of us ( gregory novak ) had tested the model in a previous ( but similar ) paper with different numerical resolutions and with different dynamical ranges @xcite . in his tests with higher resolution , similar results are found . however , when testing the model with a larger inner boundary ( i.e. , smaller dynamical range ) , the results changed significantly . therefore , in this paper we chose to sacrifice the numerical resolution to cover a large dynamical range . we assume reflecting boundary conditions on the @xmath107 boundaries occurring at each pole . on the inner / outer radial boundary , we use the standard `` outflow boundary condition '' in the _ zeus _ code ( see @xcite @xcite for more details ) . this allows both outflow and inflow depending on the state of gas just outside / inside the inner / outer boundary . we use a semi - implicit scheme for radiative cooling because the cooling timescales could be very short , e.g. , in galactic center where gas density could occasionally be very high . if the time step required by the courant - friedrich - levy ( cfl ) condition is shorter than the cooling time , the code updates gas temperature due to radiative cooling explicitly . otherwise we use an implicit scheme to update the energy integration . to avoid unphysical cooling and dramatic collapse which usually cause the code to crash , we also impose three limits dictating that the gas not drop below the temperature of the cosmic microwave background , the effective temperature associated with the agn radiation field ( note that this temperature is not equal to @xmath5 ) or the effective temperature of the stellar radiation field . in addition , we set a temperature floor of @xmath119 k in the cooling functions , since the gas can not reach these low temperatures by radiative cooling alone @xcite . as discussed in the introduction , in this paper we study the effects of agn radiative feedback in both the quasar and radio modes , paying special attention to the effects of a variable @xmath5 in the two modes . for this aim , we have run two groups of simulations according to the behavior of @xmath5 , namely models `` f '' and `` v '' . in models `` f '' @xmath5 is fixed , while in models `` v '' @xmath5 changes according to eq.([eq : model - v ] ) . table 1 shows various models we have run . capital letters `` a '' and `` b '' in the model names represent the two types of the nuclear wind model adopted ( see eq.[eq : eps - wind ] ) , while the numbers denote different mechanical efficiency @xmath45 of the wind , with increasing numbers corresponding to decreasing values of @xmath45 . in the following we first present an overview of the results , then we focus on some more specific property of the models . in general , the simulation results are similar to those presented in @xcite . for examples , figure [ fig : snapshot - b05v ] shows the snapshot of the density and temperature of the interstellar medium at different epochs for model b05v used here as a reference model , being the results of other models similar . at early times , the ism is very tenuous and the galaxy is in a quiescent phase , where only the passive evolution of stellar population is taking place . stellar mass losses enrich the gas content of the galaxy gradually . meanwhile , such gas is heated to a temperature near ( slithtly higher ) the local virial temperature as determined by the local stellar velocity dispersion and by type ia supernova explosions . in this evolutionary phase , the accretion onto the black hole is very low , and it is bondi - like . the black hole remains quiescent until the ism is globally cooled via radiative cooling . this catastrophic cooling leads to collapse of ism toward the center , i.e. , a `` cooling flow '' occurs , beginning with the formation of a cold shell of @xmath62 kpc radius @xcite . such a cold shell is unstable due to the rayleigh - taylor instability and to be disrupted quickly . on the path towards to the galactic center , the gas becomes even denser both due to the volume effect and to the accumulation of gas by the collapsing cold shell . the increase of gas density in turn enhances radiative cooling , which is proportional to density squared . such a cool and dense environment is ideal for star formation or even a starburst ( fig.[fig : snapshot - b05v ] left panel ) . actually , star formation is a very significant process of consuming the accreting gas ( about 1/3 of the total ) before it reaches the black hole . finally the black hole gets the fuel and agn activity is triggered ( fig . [ fig : snapshot - b05v ] , middle panel ) . once the central agn is accreting at a significant level , it produces radiation and a nuclear wind . they interact with the interstellar medium in the galaxy , which in turn regulates the central agn feeding . for radiative feedback , the ambient gas could be heated almost immediately because of the irradiation by agn via , e.g. , inverse compton scattering and photoionization . the mechanical feedback by outflow / wind is important mainly in the region very near to the black hole as mass flow can immediately mix into the ambient ism , affect the gas dynamics by ram pressure and heat the gas via shocks . as shown by the right panel of fig . [ fig : snapshot - b05v ] , when the ism is either heated or pushed outward as the consequence of radiative and mechanical feedback , gas is expelled from the galactic center , leaving behind a hot cavity . finally , a new loop of nucleus activity starts over again when the stellar mass losses replenish the galaxy , a critical density for catastrophic cooling is reached , and a new cold shell forms and collapses . a complete discussion of the life cycle of cold shells , their evolution under the action of radiative losses , agn feedback , star formation , can be found in @xcite and @xcite . @xcite also studied the synchrotron emission produced by the agn bursts , that behave like a giant supernova remnant . we note that agn feedback is not only _ negative _ , quenching the cooling flow , but could also be _ positive _ in some situations , inducing star formation in the central region of the galaxy @xcite the light curve of the agn luminosity over the whole cosmic epoch has been shown in @xcite for model a. figure [ fig : acc - his - b05v ] shows the light curve for a `` b '' type model ( in particular model b05v ) . the three panels correspond to different time resolutions . by comparison with fig . 6 in @xcite it can be seen that the overall luminosity evolution of models a and b is quite similar . the black hole accretion is very chaotic with a complicate temporal substructure , and many feedback loops are observed . a decline of the overall profile is clearly shown ( top panel ) . since the luminosity is shown in unit of @xmath21 , such a decline is mainly due to the mass growth of the central black hole . the luminosity evolution reflects the hydrodynamical phases briefly summarized in [ sec : overview ] . we can see that the luminosity valleys are followed by a relatively long - term climb ( middle panel ) . these are dividing points of feedback loops and signatures of strong agn feedbacks . agn feedback heats the ism , expel it out of the galactic center and stop the nuclear activities ( fig . [ fig : snapshot - b05v ] right panel ) , until a new cooling cycle begins followed by black hole accretion . in the bottom panel , we can see the luminosity fluctuations on short time scales , i.e. , the temporal substructure within a single accretion event . rayleigh - taylor instability and the interaction of direct and reflected shock waves ( see @xcite @xcite for a full discussion , see also * ? ? ? * ) are the main causes of such fluctuations . in fact , in fig . [ fig : snapshot - b05v ] ( left panel ) we can see that a cool dense shell forms around the galactic center : the density in the shell is higher than that of its inner gas , while the gravitational force points to the center . as a result , the cold shell fractures and falls onto the center piece by piece asynchronously , which finally induces the quick fluctuations in the light curves . figure [ fig : acc - his - a1 ] compares the different light curves of different models , namely a vs. b and f vs. v. the upper panel shows the light curves of model a1v ( red dashed line ) and model a1f ( blue solid line ) . both of the light curves are chaotic , and there is little qualitative difference between the two models . this is not surprising because in models `` a '' the mechanical agn feedback by nuclear winds is independent of @xmath120 , so that the change of radiative feedback ( f vs. v ) is not important . the behavior of the light curve changes for model b whose mechanical efficiency is smaller thus mechanical feedback is weaker . it can become not chaotic at all , as shown by the blue solid line in the lower panel of fig . [ fig : acc - his - a1 ] ( for model b05f ) . however , the light curve shown by the red dashed line ( model b05v ) is again chaotic . this is because in this model , although the mechanical feedback is weaker , the radiative feedback becomes stronger due to the consideration of a higher @xmath5 in the kinetic mode . in addition , during most epochs of the evolution , the accretion luminosity is significantly lower in model b05v than in model b05f . this result is because of the stronger radiative heating in model `` v '' due to the higher compton temperature which results in a overall smaller accretion rate during the evolution . in the third column of table [ tab : model - summary ] , we list the mass - weighted mean radiative efficiencies @xmath121 of the black hole accretion flow averaged over the evolution time as @xmath122 note that if the bulk of black hole accretion is in quasar mode , then according to eqs.([eq : eps - em])-([eq : mdot ] ) one expects @xmath123 . instead , if low luminosity accretion is important , @xmath124 . in all of the models except models a0f and a0v , @xmath121 lies in a narrow range @xmath125 . this efficiency is similar to that found by @xcite applying the soltan argument ( @xcite , see also * ? ? * ; * ? ? ? * ) but correcting for orbiting and ejected black holes . the values of @xmath121 in model a0f and a0v are the smallest . this is because of the strong mechanical feedback : the mechanical efficiency @xmath8 in these two models is the largest among all models . such a strong wind is very efficient in regulating mass accretion , since it pushes the gas away , thus the accretion rates are on average lower compared to other models . when the accretion rate is low , the efficiency will also become low . in general , the value of @xmath121 is slightly larger in models `` v '' than in `` f '' . the reason is as follows . refer to eq . [ eq : averaged_efficiency ] the numerator is obviously dominated by the contribution in the quasar mode , which is roughly same for the models `` v '' and `` f '' because the change of @xmath5 only applies in the low luminosity radio mode . the denominator is precisely equal to the growth of the black hole , i.e , the fourth column in table [ tab : model - summary ] . compared to model `` f '' , the compton heating in model `` v '' is stronger due to the higher @xmath5 . when the gas temperature is higher , the mass accretion rate will be lower . this results in an overall smaller black hole growth over the cosmological time , as discussed in the following section . the regulation of black hole growth at the centers of spheroids is an important effect of agn feedback . in fact , we recall that also in isolated galaxies ( i.e. , no merging ) , the gas produced by the evolving stars over the whole evolution is more than two orders of magnitude larger than the observed masses of black hole in the local universe . the fourth column in table [ tab : model - summary ] is the mass growth @xmath126 of the supermassive black hole during the whole evolution epoch . for both model `` v '' and `` f '' , we can see that the black hole growth is quite sensitive to the mechanical efficiency @xmath45 , namely @xmath126 drops rapidly when @xmath45 increases . physically , this is because with the increase of @xmath45 , the gas surrounding the black hole is pushed away more strongly , thus the mass accretion rate decreases . interestingly enough , we find that there exists a critical @xmath127 , below which the black hole growth @xmath126 is insensitive to @xmath8 . this is the case of models b3f and b4f ( or b3v and b4v ) . the explanation is that below such a small value of @xmath45 , the agn feedback becomes dominated by radiative feedback . in the regime of low mechanical efficiency ( @xmath128 ) , we found too much black hole growth . for example , in models b3f and b4f the final mass of the black hole is up to @xmath129 , which is beyond the typical mass range of a supermassive black hole . in the regime of high mechanical efficiency ( @xmath130 ) , the black hole growth is heavily suppressed , e.g. in models a0f and a0v , which results in a too low level of agn activity as argued in @xcite . thus we arrive at the important conclusion that the model - favored mechanical efficiency , for both type `` f '' and type `` v '' model , lies in a range of @xmath131 , which is consistent with the previous works and observational suggestions ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * see also @xcite@xcite ; @xcite @xcite ) . it is interesting to note that such a mechanical efficiency seems to be roughly consistent with the mhd numerical simulation to the outflow in hot accretion flows ( e.g. , * ? ? ? 34 therein ) . in the case of fixed compton temperature , we find that the mass of the black holes in models bf are all larger than in the corresponding models af . this is obviously because of the stronger mechanical efficiency @xmath8 in model a than in model b for given @xmath45 . for example , as shown in table [ tab : model - summary ] , the final black hole mass in a0f is smaller than that in b0f by nearly two orders of magnitude , and is nearly three orders of magnitude smaller than in model `` nowind.f '' , in which there is no nuclear wind at all . from table [ tab : model - summary ] , we can see that sometimes the final mass of the black hole can be as large as over @xmath129 . this obviously violates the observations , and indicates that when a fixed @xmath5 is adopted models b tend to overpredict the final black hole mass , although they may be more realistic than models a as we have mentioned in 1 . this conclusion is consistent with that in @xcite . it is therefore of great importance that the black hole growth @xmath132 in models `` v '' with variable compton temperature is systematically smaller . this is true for both models a and b. this is because a higher @xmath5 is adopted in model `` v '' in the radio mode . a higher @xmath5 implies that the gas surrounding the black hole is heated to a higher temperature , the mass accretion rate is smaller and less material is accreted by the black hole . this conclusion is also reinforced by the fact that when @xmath8 is smaller , the discrepancy of @xmath132 between models `` v '' and `` f '' is larger . this is because when @xmath8 is smaller , the role of mechanical feedback becomes less important compared to the radiative feedback therefore the effect of @xmath5 will be more significant . an important result in this context is that even for model b , when we consider a higher @xmath5 in the hot accretion phase ( i.e. , kinetic mode ) , i.e. , model `` v '' , the final mass of the black hole is not unrealistically large . the effect of a changing @xmath5 is so relevant that it is testified also by the limiting case model `` nowind.v '' , in which there is no wind feedback at all and the mass of the black hole is the largest , however also in this case the final mass of the black hole remains below @xmath133 . such a range of black hole mass is fully consistent with observations , thus models `` b '' are `` revived '' . one of the most important results that @xcite found is that for models b , in addition to the problem of black hole growth , also the agn duty cycle is not consistent with observations . when a high @xmath5 is adopted as in our model `` v '' , the dynamics of the gas fueling the black hole and subsequently the mass accretion rate is completely changed . as a consequence , also the agn duty cycle changes . in fig . [ fig : duty_cycle - b05 ] we plot agn duty cycle as a function of eddington ratio , taking model b05f ( left panels ) and b05v ( right panels ) as illustrative examples of the effect of a high @xmath5 . we remark that the results for other models are similar . in the top panels , the black and red lines show the cumulative time above and below the given eddington ratio , respectively . solid lines are the results compiled from the entire simulation time , while dashed lines only use the data in the final 2 gyr . the points are observational constraints taken from literature . the black / red points are the measurements of the fraction of objects above / below the given eddington ratio . the squares , cycles , upward- and downward- pointing triangles are from @xcite , @xcite , @xcite and @xcite , respectively , which are all compiled from low - redshift observations and should be compared with dashed lines . the star is a constraint compiled from high - redshift observations by @xcite , which should be compared with solid lines . the top / left panel shows that there are too many agn burst activities during the entire evolution time in model b05f ( solid line ) , but at the same time the agn activity in the last 2 gyr is perhaps too moderate compared to the observational constraints ( dashed line ) . these problems of models b with fixed @xmath5 were already pointed out in @xcite . for this reason it is particularly relevant that in the top / right panel , one can observe a remarkable improvement in model b05v , in which a higher @xmath5 is adopted in the phase of hot accretion flow . this result , combined with the more realistic growth of the black hole we have described in [ blackholegrowth ] , indicates that models b ( and their improvements ) have good potentialities to be used in realistic simulations . in fact , the importance of radiative feedback in the kinetic mode can be anticipated from the top / left panel of fig . [ fig : duty_cycle - b05 ] . we can see from the figure that the galaxy center spends @xmath134 of its evolution time with eddington ratios in a range of @xmath135 . such a range of eddington ratio belongs to the hot accretion phase ( i.e. , kinetic mode feedback ) , according to eq . [ eq : model - v ] . during this phase , the luminosity is at most two orders of magnitude lower than the typical quasar mode ( assumed to be 0.1 @xmath21 ) , while the compton heating efficiency is two orders of magnitude higher in the hot accretion phase than in the cold accretion phases , according to eqs . ( [ comptonheating])-([twotemperature ] ) . it is well - known that active galaxies spend most of the lifetimes in the radio / quiescent mode , as is shown in the top panels of fig . [ fig : duty_cycle - b05 ] . by contrast , the soltan argument indicates that most of the energy is emitted with high eddington ratio , e.g. , an acceptable model must have a significant fraction of the total energy emitted above a level corresponding to @xmath136 ( e.g. , * ? ? ? * ) . in the bottom panels of fig . [ fig : duty_cycle - b05 ] , we plot the `` energy - weighted '' duty cycle , the lines are of the same meanings as in the top panels , while they show the fractions of cumulative radiative energy during the evolution time . it is shown by the black solid lines that for model b05v there is 22% of the total energy emitted with a eddington ratio @xmath137 , while the value for model b05f is just 3% . in another word , models `` v '' , with a variable compton temperature , are much more favored than models `` f '' by the soltan argument . the fifth column in table [ tab : model - summary ] gives the total gas mass @xmath138 driven beyond 70 kpc ( i.e. , @xmath139 times the half - light radius of the stellar spheroid ) . in this paper , we account the gas driven in these region as the galactic wind . we can see that the values of @xmath138 are quite similar for all of the models despite of the large range of @xmath45 and different treatment of @xmath5 . the reason is that agn feedback by both winds and radiation mainly acts on the few inner kpc , and only in exceptionally strong accretion episodes the resulting shock waves can reach the galaxy outskirts . in the outer region of the galaxy , where most of the galactic winds originate from , agn feedback does not play an important role and it is mainly sn ia heating that controls the gas dynamics . the last column in table [ tab : model - summary ] presents the total ism mass @xmath140 remaining within 70 kpc at the end of the simulations . the values are similar for all models . this is because it is hard for the radiation and wind to influence such a large spacial region . however , we do find , although not shown here , the mass of the remaining gas in the inner region , say within @xmath141 kpc , is significantly smaller in model `` v '' than that in model `` f '' . this is again because of the higher @xmath5 thus stronger radiative heating and subsequently lower gas density . this is similar to the case of the growth of the black hole mass . to examine the effect of a high @xmath5 in a bit more detail , in figure [ fig : x_ray_properties ] we plot the x - ray surface brightness profiles ( 0.3 - 8 kev ; refer to pellegrini , ciotti & ostriker 2012 for the calculation of the surface brightness ) of the hot ism at quiescence for models b05f ( left panel ) and b05v ( right panel ) . the solid , dotted , dashed and dot - dashed lines correspond to increasing times , which are around 3 gyr , 6 gyr , 9 gyr , 12 gyr , respectively . we can see that within @xmath142 kpc , the emission profiles become systematically flatter in model `` v '' than in model `` f '' . because of the different compton heating in model `` v '' compared to model `` f '' , it is also expected that some difference will be present in the ism of the two families . in particular , the star formation may also be affected . to examine this effect , we have calculated the total mass of the newly formed stars during the whole evolution epoch . figure [ fig : enclosed - newstar_nowind ] shows the enclosed mass of the newly formed stars as a function of radius @xmath143 in models `` nowind.v '' and `` nowind.f '' . we can see that the star formation in the galaxy center is suppressed in model `` nowind.v '' due to the high compton temperature by nearly one order of magnitude compared to model `` nowind.f '' . however , the star formation in the outer region of the galaxy is just slightly disrupted . this implies that the range of radiative feedback is restricted to be @xmath144 1 kpc . the suppression of star formation is quite similar to the regulation of black hole growth . in the outer region of the galaxy , the feedback effect on star formation is dominated by supernovae . in the inner region , star formation relates directly to the gas accretion . the falling cold shell and filament contribute significantly to the mass accretion rate of the black hole , while we usually also find a very high star formation rate there . the strong radiative heating in the hot accretion phase trends to heat the gas into a state of high temperature and low density , which results in the suppression of star formation . we note that while mechanical feedback by winds could also suppress star formation , the mechanism is different . for mechanical feedback , the gas is expelled outward by ram pressure and probably heated via shocks . but ahead of the shock , gas will be compressed , and thermal instability will grow up rapidly . in this case , mechanical feedback could result in enhanced rather than suppressed star formation . such a phenomenon is already observed in the numerical simulations by @xcite . on the other hand , radiative feedback always suppresses star formation . for models `` v '' , we use a variable compton temperature @xmath5 as a piecewise function of the eddington ratio ( see eq . [ eq : model - v ] ) . in reality , @xmath5 could change somehow smoothly . in order to mimic such a situation , we design another family of models in which @xmath5 is of smoothed variation , i.e. @xmath145 where we choose @xmath146 k for comparison purpose , and make a test run ( denoted as `` b05s '' ) with the same model parameters as model b05v ( see table [ tab : model - summary ] for the details ) . we present the simulation results of model b05s in fig . [ fig : model - b05s ] . the right and middle panels are the agn duty cycles weighted by time and energy , respectively ( cf fig . [ fig : duty_cycle - b05 ] ) ; the left panel shows x - ray surface brightness profiles of the hot ism at quiescence ( cf fig . [ fig : x_ray_properties ] ) . we can see that the results are quite similar to those of model b05v , i.e. , a smoothed compton temperature does nt change our main results . we have performed two - dimensional high resolution hydrodynamical simulations to investigate the agn feedback in an isolated elliptical galaxy , focusing in particular on the so far unexposed regime of an accretion - dependent compton temperature @xmath5 . both radiative and mechanical feedback are taken into account and also physical processes on galactic scales such as star formation and supernovae are considered in the calculation . the inner boundary of our computational domain is carefully chosen so as to make sure that the fiducial bondi radius is resolved in our simulation , allowing for a robust estimate of the black hole accretion rate , which is crucial to study agn feedback . following previous works , two types of models have been considered . in `` a '' models , the mechanical efficiency of winds is a constant , while in `` b '' models , it decreases with the decreasing luminosity of agn as would be expected from radiatively driven winds @xcite . by construction models in the b family are perhaps more realistic than in the a family . however a few important systematic problems have been detected in the b family in previous works . in particular , previous calculations have shown that model b predicts a too high final black hole mass and incorrect agn duty cycle compared to observations . a major improvement of the present work compared with previous ones is that we allow for a different compton temperature of the radiation spectrum from the central agn . in the radiation mode , the accretion flow is described as a standard thin disk thus the corresponding compton temperature is relatively low , @xmath6k . but in the kinetic mode , the accretion flow is described by a hot accretion flow thus the corresponding compton temperature should be higher . following previous theoretical work , we adopt @xmath35k . this in general results in a stronger radiative feedback as adopted in @xcite . consequently , we find that the accretion processes are more chaotic and further suppressed . specifically , we find that model b now predicts a correct range of black hole mass and agn duty cycle . in more general sense , our study indicate that in the kinetic mode , radiative feedback is much more important than previous thought and should be seriously considered in future studies . some improvements can be made based on the present work in the future . first , as we have mentioned in the paper , we have simply chosen @xmath33 k as the compton temperature of the emitted spectrum from the hot accretion flows . this value comes from the calculation based on the theoretical spectrum from a hot accretion flow . in reality , the accretion flow is composed of an inner hot accretion flow plus an outer truncated thin disk ( see @xcite @xcite for a review ) . the best way is to combine the observed spectral energy distribution of llagns with various luminosities and calculate the compton temperature as a function of luminosity . moreover , relativistic effect may be important in calculating @xmath5 which was unfortunately neglected in the previous work . a complementary aspect of this study that needs an improvement is the description of the properties of the nuclear wind as a function of the black hole accretion state . in particular , issues line the mechanical efficiency @xmath8 , the wind speed and its angular distribution . in fact , as we have shown in this paper , @xmath8 is of crucial importance for the evaluation of mechanical feedback . but unfortunately this value is still poorly constrained by observations or by theories so we have to treat it as a free parameter , although some initial results have been obtained , e.g. , in the study of hot accretion flows @xcite . the angular distribution of wind , which is also expected to be important to mechanical feedback , is again poorly constrained . in this paper we assume a `` bipolar - like '' angular distribution function of the wind . in the future , it is expected that the theoretical studies will be able to better constrain these two properties . in fact , we are using the mhd numerical simulation data to study the wind from hot accretion flows and some initial results have been obtained . for example , we find that the winds do have a wide range of distribution angles , from @xmath147 . but different from the `` bipolar - like '' structure , we find that most of the mass flux of wind seem to be concentrate on the disk surface . it will be interesting to examine its effect in cooperation with the variations of @xmath5 . we thank scott tremaine for useful discussion of the soltan argument , and silvia pellegrini for the comments of the x - ray properties of the ism , and the referee for the helpful comments and suggestions . this work was supported in part by the natural science foundation of china ( grants 11103061 , 11133005 and 11121062 ) , the national basic research program of china ( 973 program , grant 2014cb845800 ) , and the strategic priority research program `` the emergence of cosmological structures '' of the chinese academy of sciences ( grant xdb09000000 ) . l.c . is supported by the grant prin miur 2010 - 2011 `` the chemical and dynamical evolution of the milky way and local group galaxie '' 2010ly5n27 . g.s.n . was supported by the erc european research council under the advanced grant program num 267399 momentum . the simulations were carried out at shanghai supercomputer center and shao super computing platform . notes : ( 1 ) parameter @xmath45 is the mechanical wind efficiency , @xmath121 is the mass - weighted mean radiative efficiency of the black hole accretion flow , @xmath148 is the change in the mass of the black hole in the evolution , @xmath138 is the total gas mass driven beyond 70 kpc ( which is far from the center and the ism there is recognized as galactic wind in this paper ) and @xmath149 is the total mass remaining within 70 kpc . all the masses are in unit of solar mass ; ( 2 ) there is no mechanical feedback of disk wind in models `` nowind.f '' and `` nowind.v '' .

the full counting statistics of the charge transport through an undoped graphene sheet in the presence of smooth disorder is studied . at the dirac point both in clean and diffusive limits , transport properties of a graphene sample are described by the universal dorokhov distribution of transmission probabilities . in the crossover regime , deviations from universality occur which can be studied analytically both on ballistic and diffusive sides . in the ballistic regime , we use a diagrammatic technique with matrix green functions . for a diffusive system , the sigma model is applied . our results are in good agreement with recent numerical simulations of electron transport in disordered graphene .

1006.1330

electron transport in graphene remains a field of intense experimental and theoretical activity @xcite . the hallmark of graphene is the massless dirac character of low - energy electron excitations . this gives rise to remarkable physical properties of this system distinguishing it from conventional two - dimensional metals . the most remarkable effects arise when the chemical potential is located in a close vicinity of the neutrality ( dirac ) point . in particular , a short - and - wide sample ( with width @xmath0 much exceeding the length @xmath1 ) of clean graphene exhibits at the dirac point pseudo - diffusive charge transport , @xcite with `` conductivity '' @xmath2 and with counting statistics ( characterizing fluctuations of current ) equivalent to that of a diffusive wire . @xcite in particular , the fano factor ( the shot noise power divided by the current ) takes the universal value @xmath3 that coincides with the well - known result for a diffusive metallic wire . @xcite this is at odds with usual clean metallic systems , where the conductance ( rather than conductivity ) is independent of @xmath1 and the shot noise is absent ( @xmath4 ) . the reason behind these remarkable peculiarities of transport in clean graphene at the dirac point is linearly vanishing density of states . this implies that the current is mediated by evanescent rather than propagating modes . the above theoretical predictions have been confirmed in measurements of conductance and noise in ballistic graphene flakes . @xcite recent advances in preparation and transport studies of suspended graphene samples also indicate that the system may be in the ballistic regime . @xcite effects of impurities on transport properties of graphene are highly unusual as well . in contrast to conventional metals , ballistic graphene near the dirac point conducts better when potential impurities are added.@xcite quantum interference in disordered graphene is also highly peculiar due to the dirac nature of the carriers . in particular , in the absence of intervalley scattering , the minimal conductivity@xcite @xmath5 is `` topologically protected '' from quantum localization . @xcite the exact value of the conductivity of such a system at the dirac point depends on the type of intravalley scattering ( random scalar or vector potential , or random mass , or their combination ) . for the case of random potential only ( which is experimentally realized by charged scatterers ) the conductivity in fact increases logarithmically with the length @xmath1 , in view of antilocalization . @xcite in our previous work@xcite , we have studied the evolution of conductance of a short - and - wide graphene sample from the ballistic to the diffusive regime . we have also shown that the leading disorder - induced correction to the noise and full counting statistics in the ballistic regime is completely governed by the renormalization of the conductance . this implies , in particular , that the fano factor @xmath6 remains unaffected to this order . indeed , the experiments@xcite give fano factor values in the vicinity of 1/3 at the dirac point for different system lengths @xmath1 . one could thus ask whether deviations from this value should be expected at all . in this work we present a detailed analysis of the shot noise and the full counting statistics in samples with long - range ( no valley mixing ) disorder . we show that to second order in the disorder strength a correction to the universal counting statistics of the ballistic graphene does arise . we calculate this correction and demonstrate that it suppresses the fano factor below the value 1/3 . for the case of random scalar potential , we also analyze the opposite limit of large @xmath1 when the system is deep in the diffusive regime . generalizing the analysis of weak - localization effects on the counting statistics by nazarov@xcite , we find that the fano factor returns to the value of 1/3 from below with increasing @xmath1 . the approach to 1/3 is however logarithmically slow , these results compare well with recent numerical works @xcite and particularly with the most detailed study by tworzydlo _ et al._@xcite the structure of the paper is as follows . in the sec . [ sec : mgfm ] , we describe the general matrix green function formalism and its application to the problem of full counting statistics . the model for graphene setup and disorder is introduced in sec . [ sec : model ] . we proceed with applying matrix green function method to the calculation of the distribution of transmission probabilities of the clean graphene sample in sec . [ sec : clean ] . in sec.[sec : ball ] we evaluate perturbative disorder corrections to the full counting statistics in ballistic regime . diffusive transport through disordered graphene is considered in sec . [ sec : dl ] within the sigma - model approach . the paper is concluded by sec . [ sec : sum ] summarizing the main results . technical details of the calculation are presented in three appendices . we begin with the general presentation of the matrix green function approach to the full counting statistics of a quasi - one - dimensional system . this formalism was developed by nazarov in ref . . consider a quasi - one - dimensional sample attached to two perfect metallic leads . transport characteristics of the system are encoded in the matrix of transmission amplitudes @xmath7 , where the indices enumerate conducting channels ( quantized transverse modes ) in the leads . eigenvalues of the matrix @xmath8 determine transmission probabilities of the system ( we use the `` hat '' notation for matrices in the space of channels ) . our main goal is to calculate the distribution of these transmission probabilities . the full counting statistics of the charge transport is given by the moments of this distribution or , equivalently , by the distribution itself . the first two moments of the transferred charge provide the conductance ( by landauer formula ) and the fano factor @xmath9 the starting point of our consideration is the relation between the matrix of transmission probabilities and the green function of the system , @xmath10 here @xmath11 are velocities in the @xmath12th and @xmath13th channel . the green function is taken in the mixed representation with real - space coordinates in the longitudinal direction and channel indices in transverse direction . the positions @xmath14 and @xmath15 should be taken in the left and right lead respectively in order to obtain the transmission matrix of the full system . the conjugate matrix @xmath16 is related to the retarded green function by a similar identity . the green functions are defined in the standard way as @xmath17 with energy @xmath18 and hamiltonian @xmath19 , the latter being an operator acting both on @xmath14 and in the channel space . with the help of eq . ( [ tg ] ) we can express all the moments of transmission distribution in terms of the green functions , @xmath20^n . \label{l - c}\ ] ] where @xmath21 is the velocity operator and @xmath14 and @xmath15 lie in the left and right lead respectively . for the first moment , @xmath22 , the above identity establishes an equivalence of the landauer and kubo representations for conductance . the complete statistics of the transmission eigenvalues can be represented by the generating function @xmath23^{-1}. \label{ffromtdt}\ ] ] once this function is known , all the moments of transmission distribution are easy to obtain by expanding the generating function in series at @xmath24 . an efficient method yielding the whole generating function was proposed in ref .. it amounts to calculating the matrix green function defined by the following equation : @xmath25 the parameter @xmath26 here corresponds to the source field mixing retarded and advanced components of the matrix green function . the positions @xmath27 and @xmath28 , where the source field is applied , lie within the left and right lead respectively . we will refer to this specific matrix structure as the ra ( retarded advanced ) space and denote such a matrices with the `` check '' notation . the main advantage of the matrix green function defined by eq.([matrixg ] ) is the following concise expression for the generating function of transmission probabilities : @xmath29\\ = \frac{1}{\sqrt{z } } \mathop{\mathrm{tr } } \left [ \begin{pmatrix } 0 & \hat v \\ 0 & 0 \end{pmatrix } \check g(x_l , x_l ) \right ] . \label{f}\end{gathered}\ ] ] the validity of this equation can be directly checked by expanding the green function in powers of @xmath26 with the help of perturbation theory and comparing this expansion termwise with eq . ( [ ffromtdt ] ) . the equivalence of these two expansions is provided by the identity ( [ l - c ] ) . another and , probably , most intuitive representation of the full counting statistics is given by the distribution function of transmission probabilities @xmath30 . this function takes its simplest form when expressed in terms of the parameter @xmath31 related to the transmission probability by @xmath32 . in terms of @xmath31 the probability density is defined by the identity @xmath33 . the definition of the generating function , eq . ( [ ffromtdt ] ) , implies a trace involving all transmission probabilities . with the distribution function of these probabilities we can express @xmath34 by the integral @xmath35 the function @xmath34 has a branch cut discontinuity in the complex @xmath26 plane running from @xmath36 to @xmath37 . the jump of the function across the branch cut determines the distribution function [ see ref . for derivation ] : @xmath38 . \label{pfromf}\ ] ] in other words , eq . ( [ ffromp ] ) solves the riemann - hilbert problem defined by eq . ( [ pfromf ] ) . the generating function @xmath34 can be related to the `` free energy '' of the system in the `` external '' source field . the free energy is defined in terms of the functional determinant @xmath39 the free energy can be calculated by standard diagrammatic methods and hence provides a very convenient representation of the full counting statistics . it is also convenient to parametrize the argument of the free energy by the angle @xmath40 according to @xmath41 . thus we have three equivalent representations of the full counting statistics by the functions @xmath34 , @xmath42 , and @xmath43 . in this paper we calculate the transport characteristics of a disordered graphene sample in terms of its free energy @xmath43 . the two other functions can be found with the help of identities @xmath44 the first of these relations directly follows from eq . ( [ omega ] ) while the second one is the result of the substitution of eq . ( [ ffromo ] ) into eq.([pfromf ] ) . the two most experimentally relevant quantities contained in the full counting statistics , namely , conductance and fano factor , eq . ( [ gf ] ) , can be expressed in terms of any of the functions introduced above . then the following expressions for the conductance and the fano factor hold : @xmath45 we will apply the matrix green function formalism outlined in this section to the problem of full counting statistics of a disordered graphene sample . our strategy is as follows . first , we calculate the matrix green function of a clean rectangular graphene sample and obtain the full counting statistics with the help of eq . ( [ f ] ) . then we introduce disorder in the model perturbatively . evaluation of the free energy by diagrammatic methods yields disorder corrections to the full counting statistics of a clean sample . we will adopt the single - valley model of graphene . more specifically , we will consider scattering of electrons only within a single valley and neglect intervalley scattering events . indeed , a number of experimental results show that in many graphene samples the dominant disorder scatters electrons within the same valley . first , this disorder model is supported by the odd - integer quantization @xcite of the hall conductivity , @xmath46 , representing a direct evidence@xcite in favor of smooth disorder which does not mix the valleys . the analysis of weak localization also corroborates the dominance of intra - valley scattering@xcite . furthermore , the observation of the linear density dependence @xcite of graphene conductivity away from the dirac point can be explained if one assumes that the relevant disorder is due to charged impurities and/or ripples . @xcite due to the long - range character of these types of disorder , the intervalley scattering amplitudes are strongly suppressed and will be neglected in our treatment . finally , apparent absence of localization at the dirac point down to very low temperatures @xcite points to some special symmetry of disorder . one realistic candidate model is the long - range randomness which does not scatter between valleys@xcite . the single - valley massless dirac hamiltonian of electrons in graphene has the form ( see , e.g. , ref . ) @xmath47 here @xmath48 ( with @xmath49 ) are pauli matrices acting on the electron pseudospin degree of freedom corresponding to the sublattice structure of the honeycomb lattice , @xmath50 , and the fermi velocity is @xmath51 cm / s . the random part @xmath52 is in general a @xmath53 matrix in the sublattice space . below we set @xmath54 and @xmath55 for convenience . we will calculate transport properties of a rectangular graphene sample with the dimensions @xmath56 . the contacts are attached to the two sides of the width @xmath0 separated by the distance @xmath1 . we fix the @xmath14 axis in the direction of current , fig . [ fig : sample ] , with the contacts placed at @xmath57 and @xmath58 . we assume @xmath59 , which allows us to neglect the boundary effects related to the edges of the sample that are parallel to the @xmath14 axis ( at @xmath60 ) . is placed between two parallel contacts . we assume @xmath61 throughout the paper . ] following ref . , metallic contacts are modelled as highly doped graphene regions described by the same hamiltonian ( [ ham ] ) . in other words , we assume that the chemical potential @xmath62 in the contacts is shifted far from the dirac point . in particular , @xmath63 , where @xmath18 is the chemical potential inside the graphene sample counted from the dirac point . ( all our results are independent of the sign of energy , thus we assume @xmath64 throughout the paper . ) we also assume zero temperature , that is justified provided @xmath65 . with the boundary conditions specified above , we are able to calculate explicitly the matrix green function ( [ matrixg ] ) for a clean graphene sample [ @xmath66 at zero energy . this calculation is outlined in appendix [ appg ] [ see eq . ( [ g0 ] ) ] . using this green function , we will study disorder effects in the framework of the diagrammatic technique for the free energy . in this section we apply the matrix green function formalism developed in sec.[sec : mgfm ] to the case of clean graphene . these results will play the role of the zeroth approximation for our perturbation theory . the matrix green function is derived in appendix [ appg ] . the generating function for the full counting statistics is given by eq . ( [ f ] ) . with the green function ( [ g0 ] ) , we obtain @xmath67 = \frac{w}{\pi l } \frac{\phi}{\sin \phi}.\ ] ] the corresponding dependence of the free energy on the source field @xmath40 follows from integration of eq . ( [ ffromo ] ) . this yields a simple quadratic function @xmath68 this remarkably simple result reveals the convenience of the source field parametrization @xmath41 . the clean sample responds linearly to the `` external '' field @xmath40 . the distribution of transmission probabilities given by eq . ( [ pfromo ] ) is just a constant , @xmath69 , in terms of @xmath31 . this means the distribution acquires the dorokhov form @xcite characteristic for disordered metallic wires @xmath70 hence electron transport in clean graphene at the dirac point is often called pseudodiffusive . let us now calculate an energy correction to the pseudodiffusive transport regime . in the vicinity of the dirac point ( we assume @xmath71 ) , we can account for finite energy @xmath18 by means of perturbation theory . the linear term is absent due to particle - hole symmetry of the dirac point . the lowest non - vanishing correction appears in the @xmath72 order and is given by the single diagram fig . [ fig : eloop ] , @xmath73 this integral of the product of two green functions is calculated in appendix [ appe ] . the result takes the form @xmath74 \!\right\ } , \label{oe_res}\ ] ] where @xmath75 is the digamma function . as explained above , the free energy @xmath76 contains information about the full counting statistics , i.e. all moments of the transfered charge . in particular , from eqs . ( [ gdef ] ) and ( [ fdef ] ) we obtain the following results for the conductance and fano factor : @xmath77 , \quad f = \frac{1}{3 } \left [ 1 + c_2 ( \epsilon l)^2 \right ] , \label{gfsmalle}\\ c_1 & = \frac{35 \zeta(3)}{3 \pi^2 } - \frac{124 \zeta(5)}{\pi^4 } \approx 0.101 , \label{c1}\\ c_2 & = -\frac{28 \zeta(3)}{15 \pi^2 } - \frac{434 \zeta(5)}{\pi^4 } + \frac{4572 \zeta(7)}{\pi^6 } \approx -0.052 . \label{c2}\end{aligned}\ ] ] these expressions coincide with the results of ref . obtained within an alternative ( transfer - matrix ) approach . let us now include the random part @xmath78 of the hamiltonian ( [ ham ] ) into consideration . there are in total four different types of disorder within the single valley dirac model : @xmath79 is the random potential ( charged impurities in the substrate ) , @xmath80 and @xmath81 correspond to the random vector potential ( e.g. long - range corrugations ) , and @xmath82 is the random mass . we will assume the standard gaussian type of disorder characterized by the correlation function @xmath83 the functions @xmath84 depend only on the relative distance @xmath85 and are strongly peaked near @xmath86 . thus we deal with isotropic and nearly white - noise disorder . however , in order to accurately treat ultraviolet divergencies arising in our calculation , we keep a small but finite disorder correlation length . the results will be expressed in terms of four integral constants characterizing the disorder strength @xmath87 within the specified gaussian disorder model , perturbative corrections to the free energy are given by the loop diagrams . the first and second order corrections are shown in fig . [ fig : diagrams ] . dashed lines in these diagrams denote disorder correlation functions ( [ dcorr ] ) . , ( a ) first - order , ( b ) and ( c ) second - order . ] the first - order correction to the free energy @xmath43 is given by the loop diagram containing two green functions and one impurity line , fig.[fig : diagrams]a . the green function at coincident points diverges . that is why we keep a finite correlation length calculating the first - order diagram . assuming the separation between two vertices of the diagram is given by the vector @xmath88 , we obtain the expression @xmath89.\end{gathered}\ ] ] now we substitute the green function from eq . ( [ g0 ] ) and expand the correction to the free energy in powers of @xmath88 . for the four possible disorder types , this yields @xmath90,\\ \omega_a^{(x),(y ) } & = \frac{\pi w}{l^2 } \int\limits_0^l dx \left [ \frac{-1}{4 \sin^2 \frac{\pi x}{l } } \pm \left ( \frac{1}{12 } + \frac{\delta_x^2 - \delta_y^2}{\pi^2 \bm{\delta}^4 } \right ) \right].\end{aligned}\ ] ] in these expressions we encounter two types of divergent terms : one with negative power of @xmath91 and one with an integral of @xmath92 , which diverges at @xmath93 and @xmath94 . these terms , however , are free of the source parameter @xmath40 and hence do not change any observables . the @xmath40-dependent terms are finite and , after integrating over @xmath88 in eq . ( [ omegaa ] ) , yield the simple result:@xcite @xmath95 it provides a linear ( in @xmath96 ) correction to the free energy of the clean sample , eq . ( [ omega0 ] ) , merely changing the overall prefactor ( conductance ) but preserving the quadratic dependence on @xmath40 and hence the form of the dorokhov distribution . thus the linear disorder correction does not destroy the pseudodiffusive character of transport in graphene at the dirac point . since the lowest disorder correction ( [ omegaa_res ] ) preserves the form of the dorokhov distribution , we proceed with higher order corrections . our aim is to find a deviation from the pseudodiffusive transport . the second - order correction to the free energy is due to the diagrams fig . [ fig : diagrams]b and [ fig : diagrams]c . the diagram with parallel impurity lines ( fig.[fig : diagrams]b ) yields @xmath97.\end{aligned}\ ] ] using the green function from eq . ( [ g0 ] ) , we expand the correction to the free energy in powers of @xmath88 and @xmath98 . then we drop all @xmath40-independent terms and average with respect to the directions of @xmath88 and @xmath98 . the following four contributions to the free energy are non zero : @xmath99.\end{gathered}\ ] ] after integrating with respect to @xmath14 and @xmath15 the above expression vanishes . thus we conclude that the diagram fig . [ fig : diagrams]b gives no contribution to the free energy , let us now consider the diagram fig . [ fig : diagrams]c with crossed impurity lines . this diagram contains no green functions at coincident points and hence does not require regularization . we can replace disorder correlation functions @xmath101 by equivalent delta - functions and obtain @xmath102 ^ 2.\end{gathered}\ ] ] with the green function ( [ g0 ] ) we find the following contribution to the sum in eq . ( [ omegacs ] ) : @xmath103 two - dimensional integrals with respect to @xmath14 and @xmath15 are straightforward due to periodicity of the integrand . as a result , the free energy is expressed as a single @xmath104 integral : @xmath105 . \label{omegac}\end{gathered}\ ] ] this integral diverges at @xmath106 . expanding near this point , we find that the integrand behaves as @xmath107 . the most singular part is @xmath40-independent and hence unobservable . integral of the second term diverges logarithmically and multiplies @xmath108 . this gives a logarithmic correction to the conductance of the system preserving the pseudodiffusive form of the transmission distribution . let us cut off the logarithmic integral at some ultraviolet scale @xmath109 that is the smallest scale where the massless dirac model with gaussian white - noise disorder applies , e.g. the scale of the disorder correlation length or lattice spacing in graphene . the upper cut - off is already embedded in the integrand of eq . ( [ omegac ] ) : the small @xmath104 expansion is valid for @xmath110 . thus we can isolate the divergent part of the integral ( [ omegac ] ) and the remaining @xmath111 correction , which has a non - trivial dependence on @xmath40 . @xmath112 \\ + ( \alpha_0 + 3 \alpha_z ) ( \alpha_0 - \alpha_z ) \omega_2(\phi ) \big\}. \label{omegac_res}\end{gathered}\ ] ] since the logarithmic term in the free energy contains an ultraviolet parameter @xmath113 defined up to a model - dependent constant , the functions @xmath114 are fixed up to an arbitrary constant . with this accuracy , we find @xmath115 the logarithmic correction in eq . ( [ omegac_res ] ) can be included into an effective @xmath1-dependence of the disorder strength parameters @xmath96 by renormalization group ( rg ) methods . the model of two - dimensional massless dirac fermions subject to gaussian disorder and its logarithmic renormalization appeared in various contexts . in particular , disorder renormalization in graphene was considered in refs . . one - loop rg equations for effective disorder couplings as functions of a running scale @xmath116 are @xmath117 [ rg ] parameters defined in eq . ( [ amu ] ) serve as initial conditions for the rg equations at an ultraviolet scale @xmath113 . integrating eqs . ( [ rg ] ) up to the largest scale that is the system size @xmath1 , we obtain effective disorder couplings @xmath118 and automatically take into account all leading logarithmic contributions like the one in eq . ( [ omegac_res ] ) . this allows us to replace the disorder parameters in the free energy by their renormalized values and drop the logarithm from eq . ( [ omegac_res ] ) . collecting in this way the contributions ( [ omega0 ] ) , ( [ omegaa_res ] ) , and ( [ omegac_res ] ) , we obtain the final expression for the free energy up to the second order in renormalized disorder parameters , @xmath119 . \label{omega_res}\end{gathered}\ ] ] thus we have established a deviation from pseudodiffusive transport regime ( @xmath120 ) in the second order in disorder strength . let us now derive a correction to the dorokhov distribution function of transmission probabilities . in the @xmath31 representation , the distribution function is given by eq . ( [ pfromo ] ) . using the result ( [ omega_res ] ) , we obtain @xmath121 . \label{p_res}\end{gathered}\ ] ] similarly to @xmath122 , the functions @xmath123 are defined up to a model - dependent constant . from eqs . ( [ omega1 ] ) and ( [ omega2 ] ) we obtain @xmath124 , \\ p_2(\lambda ) & = \mathrm{const } + \frac{\pi^2}{2 \sinh^2(2\lambda)}.\end{aligned}\ ] ] [ p12 ] and @xmath125 entering the disorder correction to the distribution of transmission probabilities ( [ p_res ] ) . in the case of random scalar potential ( @xmath126 ) , the distribution is determined by the sum @xmath127 only . ] the functions @xmath128 and @xmath125 are shown in fig . [ fig : p1p2p12 ] . ( when the only disorder is @xmath126 , correction to the distribution function is given by the sum @xmath127 also shown in the figure . ) the functions @xmath128 and @xmath125 can not be used for direct calculation of transmission moments due to their divergence at @xmath129 . this divergence signifies the break down of perturbative expansion in small values of disorder couplings close to @xmath130 ( that is @xmath131 ) . comparing disorder correction with the distribution in the clean sample , we conclude that the result ( [ p_res ] ) is valid provided @xmath132 . the deviation from pseudodiffusive transport regime can be experimentally demonstrated as a correction to the fano factor @xmath3 characteristic to the diffusive systems . divergence of the functions @xmath133 at @xmath134 prevents us from calculating transmission moments from the distribution function ( [ p_res ] ) . however , we can obtain transport characteristics from the free energy ( [ omega_res ] ) instead . with the help of eq . ( [ fdef ] ) , we find the fano factor up to quadratic terms in the renormalized disorder strength , @xmath135 remarkably , any weak disorder , irrespective of its matrix structure , suppresses the fano factor . [ note that the energy correction ( [ gfsmalle ] ) is also negative . ] the correction to the fano factor increases with increasing sample length @xmath1 due to renormalization ( [ rg ] ) . at some length @xmath136 , referred to as the mean free path , one of the renormalized disorder couplings reach a value of order unity and the perturbative rg treatment breaks down . this signifies the crossover from ballistic to diffusive transport regime . disorder correction to the fano factor becomes strong in this crossover region . to go beyond the mean free path scale we resort to other methods designed for diffusive systems when the system size exceeds the mean free path , the sample exhibits diffusive electron transport . on a semiclassical level , the system can be characterized by its conductivity per square in this limit . at the ballistics - diffusion crossover the conductivity of graphene is close to the quantum value @xmath137 . this signifies strong interference corrections to transport characteristics making semiclassical picture inadequate . these quantum effects lead to one of the four possible scenarios depending on the symmetry of disorder : 1 . if the only disorder is random potential ( @xmath126 ) , the system possesses time inversion symmetry @xmath138 and falls into symplectic symmetry class aii@xcite . quantum corrections to the conductivity are positive , leading to good metallic properties ( large dimensionless conductivity ) at large scales . 2 . in the case of random vector potential @xmath139 , the only symmetry of the problem is chirality , @xmath140 , signifying the chiral unitary symmetry class aiii . such disorder produces no corrections to the conductivity to all orders and can be effectively gauged out at zero energy @xcite . from the point of view of its transport properties , the system remains effectively clean and ballistic at all scales . if the only disorder is random mass ( @xmath141 ) , the hamiltonian has a bogolyubov de gennes symmetry @xmath142 characteristic for the symmetry class d. upon renormalization ( [ rg ] ) the disorder coupling gets smaller and the system becomes effectively clean . this means the absence of the mean free path scale and hence of the diffusive transport regime . 4 . in the generic case , when more than one disorder type is present and all symmetries are broken , the symmetry class is unitary ( a ) and transport properties are the same as at the critical point of the quantum hall transition . we will concentrate on the first case ( random potential ) when the system eventually acquires a large parameter dimensionless conductivity and can be quantitatively described by the proper effective field theory sigma model of the symplectic symmetry class . our consideration in this part of the paper is closely related to that of ref . . derivation of the sigma model with the source fields @xmath26 from eq.([matrixg ] ) is sketched in appendix [ appsigma ] . the symplectic sigma model operates with the matrix field @xmath143 of the size @xmath144 , where @xmath145 is the number of replicas . apart from replica space , matrix @xmath143 has retarded - advanced ( ra ) and particle - hole ( ph ) structures . the former is similar to the matrix green function while the latter is introduced in order to account for time - reversal symmetry of the problem . we will denote pauli matrices in ra space by @xmath146 . two constraints are imposed on @xmath143 , namely , @xmath147 and @xmath148 . this yields the target space @xmath149 characteristic for symplectic class systems . the sigma - model action is@xcite @xmath150 = \frac{\sigma}{16 } \int d\mathbf{r } \mathop{\mathrm{tr } } ( \nabla q)^2 . \label{sigma}\ ] ] here @xmath151 is the dimensionless ( in units @xmath137 ) conductivity of the two - dimensional disordered system . the source field is incorporated into boundary conditions , @xmath152 the free energy of the system in the source field @xmath40 is expressed through the @xmath153 limit of the sigma - model partition function as @xmath154 } \right ) . \label{omega_sigma}\ ] ] in a good metallic sample with @xmath155 , the @xmath143 integral in eq.([omega_sigma ] ) can be evaluated within the saddle point approximation . the action ( [ sigma ] ) is minimized by the following configuration of the field @xmath143 : @xmath156 replacing the integral in eq . ( [ omega_sigma ] ) with the value of the integrand at the saddle point , we obtain the semiclassical result for the full counting statistics , @xmath157}{n } = \frac{w \sigma \phi^2}{4 l}.\ ] ] this yields the dorokhov distribution of transmission probabilities in diffusive two - dimensional system.@xcite in order to find corrections to this result , we take into account fluctuations of the field @xmath143 near its saddle - point value @xmath158 . this is equivalent to the calculation of a cooperon loop , fig.[fig : cooperon ] , carried out in ref . . small fluctuations of @xmath143 near the saddle point @xmath158 are parametrized by the matrix @xmath159 as ( we write expressions involving @xmath159 up to the second order ) @xmath160 this parametrization of @xmath143 automatically fulfils the conditions @xmath161 and @xmath148 . the sigma - model action expanded up to the second order in @xmath159 takes the form @xmath150 = s[q_0 ] - \frac{\sigma}{16 } \int d\mathbf{r } \mathop{\mathrm{tr } } \left [ ( \nabla b)^2 - \frac{\phi^2}{4l^2 } \{\lambda_x , b\}^2 \right ] . \label{sigma_b}\ ] ] curly braces denote anticommutator . let us separate @xmath159 into the parts commuting and anticommuting with @xmath162 . these two parts do not couple to each other in the quadratic action ( [ sigma_b ] ) and only the former one couples to the source parameter @xmath40 . thus we can constraint the matrix @xmath159 by requiring its commutativity with @xmath162 . in terms of @xmath163 this yields @xmath164 and the action becomes @xmath150 = s[q_0 ] + \frac{\sigma}{8 } \int d\mathbf{r } \mathop{\mathrm{tr } } \left [ \nabla b \nabla b^t - \frac{\phi^2}{l^2}\ ; b b^t \right ] . \label{sigma_b}\ ] ] this quadratic form is diagonalized in momentum representation . component of momentum perpendicular to the leads takes quantized values @xmath165 with positive integer @xmath13 due to geometrical restrictions [ boundary conditions ( [ boundary ] ) fix @xmath166 at the interfaces with metallic leads ] . momentum parallel to the leads is continuous and unrestricted . for each value of the momentum there are @xmath167 independent matrix elements in @xmath163 . calculating the gaussian integral in eq . ( [ omega_sigma ] ) we obtain the free energy @xmath168 . \label{omega_dif}\end{gathered}\ ] ] in the result ( [ omega_dif ] ) , the sum diverges at large @xmath13 . the situation is similar to what we have encountered in the ballistic regime . expanding the sum in powers of @xmath40 , we see that the most divergent term is @xmath40-independent , while the next term multiplies @xmath108 and diverges logarithmically . this is nothing but the weak antilocalization correction . it renormalizes the conductivity but does not deform the full counting statistics . logarithmically divergent sum is cut at @xmath169 where @xmath136 is the mean free path . at larger values of @xmath13 the diffusive approximation ( gradient expansion in the sigma model ) breaks down . in terms of renormalized conductivity , the free energy reads @xmath170 , \label{omega_dif_renorm } \\ \tilde\sigma = \sigma + \frac{1}{\pi } \ln \frac{l}{l } \approx \frac{1}{\pi } \ln \frac{l}{l}. \label{tsigma}\end{gathered}\ ] ] the bare value of conductivity , @xmath151 , is of order one and hence negligible in comparison with the large renormalizing logarithm . the sum over @xmath13 in eq.([omega_dif_renorm ] ) is convergent and provides the deviation from semiclassical dorokhov statistics of transmission probabilities . in fact , a more rigorous procedure is to perform first a renormalization of the sigma model from the mean free path scale @xmath136 to the scale @xmath171 . then the free energy can be calculated perturbatively . it turns out , however , that this yields a result identical to the one obtained above within the perturbative analysis at the scale @xmath136 . indeed , the rg equation @xmath172 will lead exactly to the renormalization of conductivity @xmath173 , see eq . ( [ tsigma ] ) . the consequent evaluation of the perturbative contribution to @xmath174 yields eq . ( [ omega_dif ] ) with @xmath151 replaced by @xmath175 and the sum restricted to a finite ( independent of @xmath1 ) number of terms . in other words , the renormalization shifts the logarithmical contribution to @xmath151 from the second to the first term in square brackets in eq . ( [ omega_dif ] ) . let us derive the distribution function @xmath42 from the free energy ( [ omega_dif_renorm ] ) . applying eq . ( [ pfromo ] ) , we obtain the result in the form @xmath176 , \label{pdif}\\ p(\lambda ) = \frac{1}{\pi } \sum_{n = 1}^\infty \left [ \mathop{\mathrm{re } } \frac{\pi + 2 i \lambda}{\sqrt{\pi^2 n^2 - ( \pi + 2 i \lambda)^2 } } -\frac{1}{n } \right ] . \label{p}\end{gathered}\ ] ] at small values of @xmath31 , the sum in eq . ( [ p ] ) is determined by the term with @xmath22 . in the opposite limit , the sum can be estimated by the corresponding integral with the help of euler - maclaurin formula . thus we obtain the asymptotic expressions @xmath177 -\dfrac{1}{\pi}\ln\lambda , & \lambda \gg 1 . \end{cases}\ ] ] the function @xmath178 is shown in fig . [ fig : p ] . it is qualitatively similar to the numerical result of ref . . deviation from the semiclassical transport can be demonstrated by the correction to the fano factor . with the help of eq . ( [ fdef ] ) , we obtain @xmath179 a similar correction to the fano factor was found numerically in ref .. we compare the numerical results with eq.([fdif ] ) below . in the case of weak scalar disorder ( described by the coupling @xmath126 ) , the system undergoes a continuous crossover from ballistic to diffusive transport regime as the size @xmath1 grows . in both limiting cases , we encounter nearly dorokhov distribution of transmission probabilities with small corrections , eqs . ( [ fball ] ) and ( [ fdif ] ) , on both sides of the crossover . in the ballistic limit , we can formally introduce a dimensionless conductivity as @xmath180 . then the corrections to the fano factor are expressed in terms of the conductivity @xmath181 \dfrac{2\zeta(3)}{\pi^3 \sigma } , & \sigma \gg 1 . \end{cases } \label{fsigma}\ ] ] this fano factor as a function of conductivity is shown in fig . [ fig : f ] together with numerical results from ref . . in the numerical simulations , a single valley of graphene was modeled using a finite - difference approach . by construction , disorder in ref . has the symmetry of scalar potential which does not mix the valleys . it is this symmetry ( class aii ) which is considered in the present section . our results perfectly agree with the numerics in the diffusive limit ( see fig . [ fig : f ] ) in the range @xmath182 . on the ballistic side , the deviation is due to the non - universality of the ballistic transport . specifically , the function @xmath183 depends crucially on the microscopic details of disorder . in the numerical analysis of ref . , the model with strong scatterers was used , while in the present paper we adopt the model of weak gaussian white - noise disorder . for theoretical predictions on electron transport in the presence of strong scatterers see ref . . an earlier numerical study of ref . , based on the transfer - matrix description of the dirac problem , reported the value of the fano factor in the range @xmath184 ( for different samples ) with the conductivity , @xmath185 , of the largest systems varying from @xmath186 to @xmath187 . this is consistent with our predictions for the diffusive transport regime ( see fig.[fig : f ] ) . the behavior of @xmath188 in the ballistic regime is different due to the reasons described above ( strong vs. weak disorder ) . a non - monotonous dependence @xmath183 at the dirac point was also observed in ref .. the fano factor is @xmath6 both in the clean and strongly disordered limits . in the crossover from ballistics to diffusion , the fano factor strongly deviates from this universal value signifying the breakdown of the ( pseudo- ) diffusive description characterized by dorokhov distribution of transmission probabilities . ) . dashed line corresponds to the asymptotic value @xmath3 . solid symbols are numerical results from ref . , the size of rectangles corresponds to the error estimate . ] we have studied the full counting statistics of the charge transport through an undoped graphene sheet in the presence of weak and smooth ( not mixing valleys ) disorder . we have identified deviations from the dorokhov distribution of transmission probabilities both in ballistic [ eq . ( [ p_res ] ) , ( [ p12 ] ) ] and diffusive [ eqs . ( [ pdif ] ) , ( [ p ] ) ] regimes . in the former case , corrections are model - dependent while in the latter case only the symmetry of disorder matters . we have considered gaussian white - noise disorder in the ballistic regime and potential disorder ( symplectic symmetry class ) in diffusive limit . deviation from ( pseudo- ) diffusive transport always results in a negative correction to the fano factor , @xmath189 . our results are in good agreement with recent numerical simulations of electron transport in disordered graphene , see fig . [ fig : f ] . we are grateful to r. danneau , p. san - jose , and m. titov for stimulating discussions and to c. groth for providing us with the numerical data of ref .. the work was supported by rosnauka grant 02.740.11.5072 and by the eurohorcs / esf euryi award scheme ( i.v.g . ) . the full counting statistics of the electron transport is conveniently expressed in terms of the matrix green function @xcite in the external counting field @xmath41 , eq . ( [ matrixg ] ) . for the clean graphene sample attached to perfect metallic leads , fig.[fig : sample ] , this green function satisfies the following equation : @xmath190 since the operator in the left - hand side of the above equation commutes with the @xmath104 component of the momentum , we will first calculate the green function in the mixed coordinate - momentum representation , @xmath191 . inside the sample this function satisfies @xmath192 \check g_p ( x , x ' ) = \delta(x - x').\ ] ] we will look for a general solution of this equation in the form @xmath193 the chemical potential profile together with the infinitesimal terms @xmath194 in eq . ( [ keldysh ] ) defines the boundary conditions for the green function . the counting field @xmath26 can also be incorporated into the boundary conditions . in terms of @xmath195 we thus obtain @xmath196 delta function in the right - hand side of eq . ( [ eqgreen ] ) yields a jump of the green function at @xmath197 which provides the relation @xmath198 the matrices @xmath195 , and hence the green function , are completely determined by eqs . ( [ boundaryg ] ) and ( [ jump ] ) , @xmath199 fourier transform in @xmath200 yields the green function in the full coordinate representation . to facilitate further calculations , we decompose this green function into the following product of matrices : @xmath201 \sinh\dfrac{\phi y}{2l } & -i \cosh\dfrac{\phi y}{2l } \end{pmatrix}_{ra } \begin{pmatrix } \dfrac{1}{\sin \frac{\pi}{2l } ( x + x ' + iy ) } & \dfrac{1}{\sin \frac{\pi}{2l } ( x - x ' + iy ) } \\[8pt ] \dfrac{1}{\sin \frac{\pi}{2l } ( x - x ' - iy ) } & \dfrac{1}{\sin \frac{\pi}{2l } ( x + x ' - iy ) } \end{pmatrix}_\sigma \check \lambda \check u^{-1}(x ' ) , \label{g0 } \\ \check \lambda = \begin{pmatrix } \sigma_z & 0 \\ 0 & 1 \end{pmatrix}_{ra } , \qquad\qquad \check u(x ) = \begin{pmatrix } \sin\dfrac{\phi ( l - x)}{2l } & \cos\dfrac{\phi ( l - x)}{2l } \\[8pt ] i\cos\dfrac{\phi x}{2l } & i \sin\dfrac{\phi x}{2l } \end{pmatrix}_{ra}.\end{gathered}\ ] ] here we have used the source angle @xmath40 defined by @xmath41 . the matrices @xmath202 and @xmath203 operate in the retarded - advanced space only and hence commute with any disorder operators placed between the green functions . as a result , factors @xmath204 and @xmath205 drop from expressions for any closed diagrams . the matrices @xmath206 in the above equation allow us to decompose the green function into a direct product of the two operators acting in the ra space and in the sublattice space . in this appendix we evaluate the diagram fig . [ fig : eloop ] for the lowest energy correction to @xmath43 . substituting green function ( [ g0 ] ) into eq . ( [ oe ] ) and performing rescaling of integration variables we obtain @xmath207.\ ] ] the first ( second ) term in square brackets depends only on sum ( difference ) of @xmath14 and @xmath15 . this allows us to integrate over the difference ( sum ) of these variables . after some shifts of variables the remaining integral takes the form @xmath208 the last expression is the result of @xmath104 integration . it can be performed , e.g. , by closing the integration contour and summing up residues in the upper half - plane of imaginary @xmath104 . in order to make this sum convergent , one has to add a weak damping factor by an infinitesimal imaginary shift of @xmath40 . we proceed with the last integral in eq . ( [ uint ] ) by representing @xmath209 as a fourier series @xmath210.\ ] ] convergence of this fourier series should also be justified by a proper damping factor . this does not change the final result of the calculation hence we omit such extra factors for simplicity . performing the integration over @xmath211 we obtain @xmath212.\ ] ] the sum over @xmath13 diverges logarithmically . however , this divergence is independent of @xmath40 and hence does not influence any observable quantities , which are expressed as derivatives of the free energy . we can easily get rid of the divergent part by subtracting a similar sum over @xmath13 with @xmath213 . this yields the final result @xmath214 \\ = -\frac{w}{\pi l } \frac{(\epsilon l)^2}{\sin\frac{\phi}{2 } } \frac{\partial}{\partial \phi } \left\ { \cos\frac{\phi}{2 } \left [ \psi \left ( \frac{\pi + \phi}{2\pi } \right ) + \psi \left ( \frac{\pi - \phi}{2\pi } \right ) + 4 \ln 2 + 2 \gamma \right ] \right\}.\end{gathered}\ ] ] here @xmath75 is the digamma function and @xmath215 is the euler - mascheroni constant . the last expression yields eq . ( [ oe_res ] ) of the main text ( where we drop the unobservable constant ) . in order to carry out a parametrically controlled derivation of the sigma model , it is convenient to consider a modified problem with @xmath216 flavours of dirac fermions . to perform the disorder average of the free energy , we also introduce @xmath145 replicas . ( alternatively , one can use supersymmetry . as we will treat the sigma model perturbatively , the two approaches are fully equivalent . ) the derivation of the sigma model starts with the fermionic action generating the matrix green function ( [ matrixg ] ) . @xmath217 = \int d\mathbf{r } \sum_{a , b,\alpha } { \phi_a^\alpha}^\dagger \bigg ( \big\ { i0 \lambda_z - \bm{\sigma}\mathbf{p } - \sqrt{z } \sigma_x \big [ \lambda_+ \delta(x ) + \lambda_- \delta(x - l ) \big ] \big\}\delta_{ab } -v_{ab}(\mathbf{r } ) \bigg ) \phi_b^\alpha.\ ] ] here @xmath218 are matrices operating in ra space . this action is the functional of two independent grassmann two - component ( in @xmath151 space ) vector fields @xmath40 and @xmath219 . lower indices , @xmath113 and @xmath163 , refer to flavours while the upper index @xmath220 enumerates replicas . overall , there are @xmath221 independent grassmann variables in the lagrangian . the random matrix @xmath222 is symmetric , that insures the time - reversal symmetry of the model . we assume gaussian white - noise statistics for the matrix @xmath223 defined by the correlator @xmath224 \delta(\mathbf{r } - \mathbf{r } ' ) . \label{corr}\ ] ] using the time - reversal symmetry , we rewrite the action in terms of the single four - component field @xmath75 ( and its charge - conjugate version @xmath225 , that is linearly related to @xmath75 ) : @xmath226 this introduces an additional particle - hole ( ph ) structure of the fields . pauli matrices operating in ph space are denoted by @xmath227 . bar denotes the charge conjugation operation which has two important properties : @xmath228 and @xmath229 . the action takes the following form in terms of @xmath75 : @xmath230 = \int d\mathbf{r } \sum_{a , b,\alpha } \bar\psi_a^\alpha \bigg ( \big\ { i0 \lambda_z - \bm{\sigma}\mathbf{p } - \sqrt{z } \sigma_x \big [ \rho_+ \delta(x ) + \rho_- \delta(x - l ) \big ] \big\}\delta_{ab } - v_{ab}(\mathbf{r } ) \bigg ) \psi_b^\alpha.\ ] ] in this expression we have introduced the notation @xmath231 . now we are ready to average @xmath232 over the gaussian disorder distribution with the correlator ( [ corr ] ) . this yields an effective action with the quartic term . using the above - mentioned properties of charge conjugation , we recast the action in the form @xmath230 = \int d\mathbf{r } \left [ \sum_{a,\alpha } \bar\psi_a^\alpha \big\ { i0 \lambda_z - \bm{\sigma}\mathbf{p } - \sqrt{z } \sigma_x \big [ \rho_+ \delta(x ) + \rho_- \delta(x - l ) \big ] \big\ } \psi_a^\alpha + \frac{2\pi\alpha_0}{n } \sum_{a , b,\alpha,\beta } \mathop{\mathrm{tr } } \psi_a^\alpha \bar\psi_a^\beta \psi_b^\beta \bar\psi_b^\alpha \right].\ ] ] next , we decouple the quartic term introducing an auxiliary @xmath233 matrix @xmath234 by the hubbard - stratonovich transformation . this yields the action @xmath235 = \int d\mathbf{r } \left [ \frac{n \gamma^2}{8\pi\alpha_0 } \mathop{\mathrm{tr } } r^2 + \sum_{a,\alpha , \beta } \bar\psi_a^\alpha\bigg ( i \gamma r_{\alpha\beta } - \big\ { \bm{\sigma}\mathbf{p } + \sqrt{z } \sigma_x \big [ \rho_+ \delta(x ) + \rho_- \delta(x - l ) \big ] \big\}\delta_{\alpha\beta } \bigg ) \psi_a^\beta \right].\ ] ] parameter @xmath215 is an arbitrary number at this stage , its value will be fixed later . matrix @xmath236 couples to the product @xmath237 . this allows us to impose the corresponding symmetry constraint on the matrix @xmath234 : @xmath238 . finally , we integrate out the fermionic fields and obtain the action operating with the matrix @xmath234 only , @xmath239 = \frac{n}{2 } \mathop{\mathbf{tr } } \left ( \frac{\gamma^2 r^2}{4 \pi \alpha_0 } - \ln \bigg\ { i \gamma r - \bm{\sigma}\mathbf{p } - \sqrt{z } \sigma_x \big [ \rho_+ \delta(x ) + \rho_- \delta(x - l ) \big ] \bigg\ } \right ) . \label{trlog}\ ] ] the bold ` @xmath240 ' symbol implies the full operator trace including integration over space coordinates . derivation of the sigma model proceeds with the saddle - point analysis of the action ( [ trlog ] ) in the absence of the source field @xmath26 . we first look for a diagonal and spatially constant matrix @xmath234 minimizing the action . the saddle - point equation is identical to the self - consistent born approximation ( scba ) equation for the self energy @xmath241 , @xmath242 we fix @xmath215 to be the imaginary part of the scba self energy , @xmath243 with @xmath244 being ultraviolet energy cut - off ( band width ) . then the saddle - point configuration for the matrix @xmath234 is simply @xmath245 . this fixes the boundary conditions for the matrix @xmath234 at the contacts . since the leads are very good metals and fluctuations of @xmath234 are strongly suppressed there , @xmath246 for @xmath247 and @xmath248 . the matrix @xmath246 is not the only saddle point of the action ( [ trlog ] ) . other configurations minimizing the action can be obtained by rotations @xmath249 with any matrix @xmath250 which commutes with @xmath251 and preserves the constraint @xmath252 . matrix @xmath250 , and hence @xmath234 , is trivial in @xmath151 space . this allows us to reduce the dimension of @xmath234 to @xmath144 operating in @xmath116 , @xmath253 , and replicas only . the saddle manifold generated by matrices @xmath250 is @xmath254 . let us now restore the source term in the action and establish boundary conditions for @xmath234 . the matrix @xmath234 has a jump at the interfaces with the leads due to the delta functions in the action ( [ trlog ] ) . however , we can eliminate these jumps by a proper gauge transformation . let us perform a rotation @xmath255 with an @xmath14-dependent matrix @xmath256 . the action acquires the following form in terms of @xmath257 : @xmath258 = \frac{n}{2 } \mathop{\mathbf{tr } } \left [ \frac{\gamma^2 \tilde r^2}{2 \pi \alpha_0 } - \ln \left ( i\gamma \tilde r - \bm{\sigma}\mathbf{p } + i\sigma_x a^{-1 } \left\ { \frac{\partial a}{\partial x } + i \sqrt{z } \big [ \rho_+ \delta(x ) + \rho_- \delta(x - l ) \big ] a \right\ } \right ) \right].\ ] ] the source field drops from this action if we choose @xmath256 such that the expression in curly braces vanishes . this yields @xmath259 note that the matrix @xmath257 , defined with the help of the above matrix @xmath256 , fulfils the condition @xmath260 . since delta functions disappear from the action , we can infer that @xmath257 is continuous at the interfaces with the leads . in the left lead we have @xmath261 . this is the left boundary condition for the matrix @xmath257 . the right boundary condition is fixed by the identities @xmath262 and @xmath246 for @xmath248 . this yields @xmath263 we can further simplify this bulky expression by performing a constant rotation @xmath264 with the matrix @xmath265.\ ] ] after such a rotation the action and boundary conditions become @xmath266 = \frac{n}{2 } \mathop{\mathbf{tr } } \left [ \frac{\gamma^2 q^2}{2 \pi \alpha_0 } - \ln \left ( i\gamma q - \bm{\sigma}\mathbf{p } \right ) \right ] , \label{trlogq}\\ q(0 ) = \lambda_z , \qquad q(l ) = \lambda_z \cos\phi + \lambda_x \sin\phi.\end{gathered}\ ] ] thus we have reduced the boundary conditions to the form ( [ boundary ] ) . the matrix @xmath159 is chosen such that @xmath267 . hence the matrix @xmath143 obeys the symmetry constraint @xmath148 . the last step of the sigma - model derivation is the gradient expansion in eq.([trlogq ] ) . this expansion is straightforward for the real part of the action @xcite @xmath268 = -\frac{n}{4 } \mathop{\mathbf{tr } } \ln \big ( i\gamma q - \bm{\sigma}\mathbf{p } \big ) \big ( -i\gamma q - \bm{\sigma}\mathbf{p } \big ) = -\frac{n}{4 } \mathop{\mathbf{tr } } \ln \big ( p^2 + \gamma^2 + \gamma \bm{\sigma } \nabla q \big ) \simeq \frac{n}{16 \pi } \mathop{\mathrm{tr } } ( \nabla q)^2.\ ] ] the drude conductivity of the two - dimensional sample with @xmath13 flavours of massless dirac fermions at the dirac point is @xmath269 . with the dimensionless conductivity @xmath270 , we finally obtain eq.([sigma ] ) supplemented by the boundary conditions ( [ boundary ] ) . 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 ) ; j. low temp . phys . * 153 * , 374 ( 2008 ) ; solid state comm . * 149 * , 1050 ( 2009 ) . f. v. tikhonenko , d. w. horsell , r. v. gorbachev , and a. k. savchenko , phys . lett . * 100 * , 056802 ( 2008 ) ; d. w. horsell , a. k.savchenko , f. v. tikhonenko , k. kechedzhi , i. v. lerner , v. i. falko , solid state comm . * 149 * , 1041 ( 2009 ) ; f. v. tikhonenko , a. a.kozikov , a. k. savchenko , r. v. gorbachev , phys . lett . * 103 * , 226801 ( 2009 ) . an alternative possibility is that the dominant disorder is formed by resonant scatterers preserving the chiral symmetry@xcite . a detailed analysis of the evolution from ballistics to diffusion in this model will be published elsewhere . the result ( [ omegaa_res ] ) is twice smaller than the correction obtained in ref . within the transfer - matrix approach . the reason for this discrepancy is that in the latter case ultraviolet divergency was regulated by introducing a finite correlation length in the @xmath104 direction only . this resulted in the angle average @xmath271 being twice larger than for the isotropic model adopted in the current paper , @xmath272 . the sigma - model action contains also an imaginary @xmath273 topological term@xcite ommited in eq . ( [ sigma ] ) . this term is crucial for ensuring topological protection from anderson localization at small @xmath151 . here we are focusing , however , on the range of large @xmath151 , where the effect of topological term is exponentially small and can be discarded .

we present a nonparametric method for estimating the value and several derivatives of an unknown , sufficiently smooth real - valued function of real - valued arguments from a finite sample of points , where both the function arguments and the corresponding values are known only up to measurement errors having some assumed distribution and correlation structure . the method , _ moving taylor bayesian regression _ ( motabar ) , uses bayesian updating to find the posterior mean of the coefficients of a taylor polynomial of the function at a moving position of interest . when measurement errors are neglected , motabar becomes a multivariate interpolation method . it contains several well - known regression and interpolation methods as special or limit cases . we demonstrate the performance of motabar using the reconstruction of the lorenz attractor from noisy observations as an example .

1204.2841

the basic task of regression analysis the estimation of values of an unknown , sufficiently smooth function @xmath0 of one or several real arguments , given a finite amount of possibly noisy data occurs pervasively in many kinds of quantitative scientific research . most existing approaches to this task can roughly be classified into two groups : _ parametric _ or model - based approaches such as polynomial regression , spline smoothing @xcite , or kriging @xcite , and _ nonparametric _ approaches such as local regression , nearest or natural neighbour estimation @xcite , inverse distance weighting @xcite , or kernel smoothing . although parametric methods are usually based on more rigorous reasoning such as maximum likelihood estimation , bayesian updating , approximation theory , or some other kind of optimization , their results are only guaranteed to be reliable if the sought function is assumed to belong to some particular model or function class with a small number of parameters , e.g. , polynomials or splines of a fixed degree and fixed set or number of break points . if , as is often the case in empirical research , this strong assumption can not be justified , nonparametric methods are available which , however , are usually based on various forms of more heuristic reasoning , frequently involve the choice of several control parameters such as the number of neighbours or the choice of a weight or kernel function , and sometimes do not provide an easily interpreted assessment of reliability such as a standard error or confidence interval . also , many regression methods are not or only restrictedly applicable if one or several of the following conditions apply : ( i ) there might be measurement errors not only in the function values ( the `` dependent '' variable ) but also in the function arguments ( the `` independent '' variables ) , ( ii ) measurement errors might not be independent and identically distributed , ( iii ) the function is of more than one real argument , ( iv ) the measured sample is irregularly distributed in the function s argument space , and/or ( v ) also several derivatives of the function shall be estimated . in order to illustrate how plausible estimates can depend on the assumed amount and correlation of measurement errors , consider the minimal data @xmath1 , @xmath2 , and @xmath3 and assume that we want to estimate @xmath0 and @xmath4 on the interval @xmath5 $ ] on the basis of this data . in the left diagram in fig.[fig : hat3 ] , four different such estimates of @xmath0 are shown , which were produced with the method we will describe in this article , using different assumptions on the measurement error contained in the data . without measurement errors , i.e. , if both @xmath6 and @xmath7 data are precise , it is plausible to estimate @xmath8 and @xmath9 , as on the blue dotted line . without errors in @xmath6 but with large independent errors in @xmath7 , the apparent slope becomes quite uncertain , and one would rather estimate @xmath10 by the sample mean , @xmath11 , as on the yellow line . however , if the errors in @xmath7 are highly correlated , e.g. , because of a systematic but unknown bias in the measurement equipment , then one knows that the data is basically only shifted in the @xmath7 direction , which has no influence on slopes , so one would probably expect @xmath12 again , as on the green dashed line . the smaller the errors , the more the slope should resemble @xmath13 , as on the cyan dash - dotted line . similarly , if the @xmath7 data is precise but the @xmath6 data is highly uncertain , one would use the sample estimate if errors are uncorrelated , as on the yellow line in the right diagram in fig.[fig : hat3 ] . if errors in @xmath6 are correlated , one would retain some slope information which suggests that the values to the left and right of the three data points are rather below than above three . hence the estimate would be lower than with uncorrelated errors , as on the green dashed line . the occurrence of this lowering effect also shows that the effect of errors in the @xmath6 dimension is usually not the same as the effect of some `` equivalent '' amount of errors in the @xmath7 dimension . ( colour online ) illustration of the effect of error correlation on some motabar estimates . , title="fig:",scaledwidth=49.0% ] ( colour online ) illustration of the effect of error correlation on some motabar estimates . , title="fig:",scaledwidth=49.0% ] in this article , we develop a regression method which is nonparametric in the sense that it can estimate any sufficiently smooth function , but is still based on rigorous statistical reasoning , can deal with any of the above situations ( i)(v ) , and contains several existing methods as special or limiting cases . the method , _ moving taylor bayesian regression ( motabar ) _ , is based on the following ideas : 1 . approximate @xmath0 locally by a taylor polynomial at a moving position of interest @xmath14 . 2 . treat the unknown taylor coefficients as parameters in a statistical model . 3 . use the measured data to update prior beliefs about these parameters using bayesian updating . 4 . use the posterior mean ( or mode ) and variance of the parameters as estimates of the function s value and derivatives and the corresponding estimation uncertainty . in principle , these steps can be performed for any form of error and prior distribution , but the method becomes particularly transparent if the value measurement error is assumed to be multivariate gaussian and the involved prior distributions are also multivariate gaussian . in that case , the resulting posterior distributions can be written as mixtures of gaussians whose mean and variance can be derived analytically , and the resulting motabar estimator @xmath15 of @xmath16 turns out to be a mixture of smooth rational functions of @xmath14 that can be computed using simple linear algebra . the mixing is related to the distribution of argument measurement errors . without argument measurement errors , @xmath15 is a smooth rational function of @xmath14 whose algebraic form can be seen as a generalization of ordinary least squares regression estimators . for these reasons , we will focus on the case of gaussian priors and value errors in this article . the input data needed to apply motabar are then * the measured data , * variances and possibly covariances of both argument and value measurement errors , * a prior covariance matrix for the value of @xmath0 and its derivatives of order @xmath17 for some @xmath18 , * prior variances for @xmath0 s derivatives of order @xmath19 , * and , as the only free control parameter , an integer @xmath19 larger than the order of any derivative of @xmath0 one wants to estimate . for particular cases of error and prior distributions , it will turn out that the motabar estimate equal or approximate the results of some well - known other methods , including ordinary polynomial regression , inverse distance weighting , and linear interpolation on regular grids . this behaviour of our estimator is similar to that of a related approach by @xcite in which @xmath0 is also approximated by a taylor polynomial at a moving position of interest @xmath14 , but in which the taylor coefficients are then estimated not by bayesian updating but by minimizing a heuristic loss function motivated by approximation theory ( see sec.[sec : wang ] ) . other cases of error and prior distributions result in plausible generalizations or variants of well - known methods , e.g. , a new form of local polynomial interpolation and a bayesian variant of inverse distance weighted smoothing . in practise , the needed prior and error variances and covariances might themselves be estimated from other or even the same data , a question we do however not address in detail in this article . [ [ comparison - with - other - methods . ] ] comparison with other methods . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + motabar can be interpreted as a kind of _ local _ regression since although it takes into account also data points far away from @xmath14 , it gives them much less influence on the estimate than those close to @xmath14 . this fact is reflected in the occurrence of a _ weight matrix _ @xmath20 in the estimator equation . but unlike other local or piece - wise methods such as nearest or natural neighbour regression , locally weighted scatterplot smoothing ( loess ) @xcite , or splines , the motabar estimate @xmath15 is infinitely smooth ( i.e. , infinitely often differentiable ) , at least as long as the value measurement errors have nonzero variance , and there are either no argument measurement errors or their probability density is infinitely smooth ( e.g. , when argument measurement errors are gaussian as well ) . this is because in motabar the weight of each individual data point in the estimate depends smoothly on its distance from @xmath14 , whereas in other methods the weights can switch from zero to nonzero in a nonsmooth way as @xmath14 moves . the nonsmooth change in weights that other methods involve is also counter - intuitive when arguments can only be measured with some error . e.g. , suppose that measurements resulted in the argument - value pairs @xmath21 , @xmath22 , @xmath23 , and @xmath24 , where the argument measurements involved an error of magnitude @xmath25 , so that the last measurement might actually reflect @xmath26 instead of @xmath27 . a linear interpolant of the four measurements would have a slope of @xmath28 , but when the @xmath24 measurement is moved towards @xmath24 , the slope would discontinuously switch to @xmath29 . a similar effect occurs with splines . in other words , when the ranking of the argument measurements with respect to their distance from @xmath14 is uncertain due to measurement errors , it seems inappropriate to use a method that strongly relies on the correctness of this ranking , e.g. , by using only the @xmath30 nearest neighbours of @xmath14 ; although disregarding faraway measurements or extreme observations completely instead of just downweighting them might still be advisable to increase the robustness of the method if outliers might exist , e.g. , due to fat - tailed error distributions . one existing class of methods , inverse distance weighting ( idw ) @xcite , also uses smoothly decaying weights that only depend on the distance from @xmath14 . in a commonly used variant of idw , the weights are inversely proportional to the square or a larger power of the distance , and we will show that this method can be derived as a special case of motabar in which a noninformative prior distribution for @xmath31 is used . a common feature of many interpolation and smoothing methods , including idw and other weightings- or kernel - based methods , linear interpolation , and nearest or natural neighbour interpolation , is that the estimate can not exceed the largest measured values , even if the data strongly suggest a nonzero slope at the largest data point . such methods will therefore always underestimate the maxima of @xmath0 . other methods , like polynomial regression , can in some situations ` overshoot ' and result in estimates that lie far outside the measured range . spline methods are often considered a good compromise between these two behaviours regarding maxima . depending on the choice of @xmath19 , motabar will behave quite similar to spline methods in this respect . as a consequence , the weights of individual data values in the estimate might be @xmath32 or @xmath33 . similar to other nonparametric methods or methods with many parameters , the motabar estimate might fit the data too narrowly and show too much fluctuations due to this ` overfitting ' when @xmath19 and the prior variances are badly chosen . if the prior variances for higher derivatives grow too fast , the estimate is allowed to vary on smaller scales than the sampling density can resolve reliably in view of the assumed error distributions . the amount of overfitting can thus be controlled by varying @xmath19 and the priors . assume that we are interested in the value of a certain function @xmath34 and all its derivatives of order @xmath17 at a certain _ position of interest _ @xmath35 , where @xmath18 and @xmath0 is assumed to be @xmath19 times continuously differentiable . we take the convention to write elements of @xmath36 as column vectors and to enumerate their @xmath37 components with a subscript index , hence @xmath38 where the @xmath39 symbol denotes transposition . assume that the only information we have about @xmath0 is ( i ) @xmath40 _ measured data points _ @xmath41 with @xmath42 , ( ii ) some estimates of the magnitude of _ measurement errors _ , and ( iii ) some beliefs about the variability of @xmath0 and some of its derivatives . these assumptions will be made more precise later . we enumerate measurements with superscript indices in guillemots , hence @xmath43 . assume the @xmath44-th data point @xmath41 is the result of trying to measure @xmath0 at the argument @xmath45 , but the measurement might involve an error in both the argument and the value , so that the actual result @xmath46 is the value of @xmath0 at a slighly different argument @xmath47 ( where @xmath48 is the _ argument error _ ) plus some _ value error _ @xmath49 . in other words , @xmath50 [ [ multi - index - notation . ] ] multi - index notation . + + + + + + + + + + + + + + + + + + + + + for dealing with higher - order derivatives and multidimensional taylor polynomials of @xmath0 , it is convenient to use a _ multi - index _ @xmath51 consisting of nonnegative integers @xmath52 for @xmath53 . we then use the denotations @xmath54 @xmath55 where @xmath56 and @xmath57 . note that the order of differentiation in @xmath58 is unimportant for @xmath57 . our quantities of interest are then the derivatives @xmath59 for a given position of interest @xmath35 , a relationship between our quantities of interest @xmath60 and the data @xmath61 is given by taylor s theorem , which leads to @xmath62 where @xmath63\quad ( |\alpha|=p).\end{aligned}\ ] ] note that there are @xmath64 many choices for @xmath65 with @xmath66 and @xmath67 many choices for @xmath65 with @xmath68 . it will be convenient to stack all relevant quantities into the column vectors @xmath69 @xmath70 , @xmath71 , @xmath72 , @xmath73 , @xmath74 and the matrix @xmath75 note that the column vector @xmath76 has a row for each combination of @xmath65 and @xmath44 , and we write this row index as @xmath77 . this must not be confused with the row - and - column index pair @xmath78 of the matrix @xmath79 . in other words , @xmath80 is an @xmath81 vector , @xmath76 is @xmath82 , and @xmath79 is an @xmath83 matrix . eq.[eqn : taylor ] is now summarised in matrix notation as @xmath84 although this is formally a linear regression model , we are not interested in estimating its coefficient matrix @xmath79 ( which we know already up to some measurement errors ) , but in estimating the regressor @xmath80 , and this we need to do for each position of interest @xmath14 separately . this is the reason why we next apply bayes theorem to eq.[eqn : taylorshort ] . let us model our information about the quantities of interest @xmath85 and about the other unknown terms in eq.[eqn : taylorshort ] as bayesian beliefs , i.e. , in the form of ( subjective ) probability distributions , by treating @xmath80 and @xmath76 as random variables with lebesgue - integrable densities @xmath86 . because in general , each @xmath87 corresponds to an argument @xmath88 that is different from @xmath14 , we assume @xmath80 and @xmath76 are independent . also , we assume that the function - related quantities @xmath80 are independent from the measurement - related quantities @xmath89 and @xmath90 . the measured data @xmath91 then allow us to update our prior beliefs @xmath92 about @xmath80 by applying bayes theorem to eq.[eqn : taylorshort ] , which leads to posterior beliefs @xmath93 note that we follow the general convention to use the same symbol , here @xmath94 , to refer to all occurring probability densities since it will always be clear from the context which variable s density is meant . also , we suppress the domain of integration in the following , and we are not interested in multiplicative constants that do not depend on @xmath80 , and use the symbol @xmath95 to denote equality up to such a constant . to utilise eq.[eqn : bayes ] , we need to specify * a distribution @xmath96 for the argument errors that might depend on the arguments , * a distribution @xmath97 of the value errors that might depend on the arguments and argument errors , and * prior distributions @xmath92 and @xmath98 expressing our initial information about the values and derivatives of @xmath0 at @xmath14 before the measurements . the term @xmath99 can then be determined from @xmath98 using eq.[eqn : r ] , while the term @xmath100 can be determined from @xmath97 using eq.[eqn : taylorshort ] . as an estimate of @xmath80 one can then use any measure of central tendency of the posterior distribution as given by eq.[eqn : bayes ] , e.g. , the posterior mean , median , or mode , while an estimate of the estimation error would be given by a suitable measure of dispersion , e.g. , the posterior variance or the median distance from the median . although this strategy can in principle be applied to error and prior distributions of any form , our approach becomes especially simple in the gaussian case . for the rest of this article , we will assume that value errors and priors are ( multivariate ) gaussian , @xmath101 where we use _ precision matrices _ @xmath102 and omit to denote the dependency on @xmath103 to simplify the following equations . note that @xmath104 is a @xmath105 square matrix whose rows and columns we address using indices of the form @xmath106 with @xmath107 and @xmath108 . the case in which it is known that there are no value errors requires some special treatment since then @xmath109 so that @xmath104 does not exist . in that case , @xmath97 behaves like a dirac delta function , giving @xmath110 for all integrable functions @xmath111 of @xmath112 . the taylor remainders @xmath113 given @xmath89 and @xmath90 are then also gaussian with mean @xmath114 and covariance matrix @xmath115 given by @xmath116 note that for almost all choices of @xmath90 and @xmath117 , the matrix @xmath115 is nonsingular . we will therefore assume that @xmath90 has a continuous distribution and @xmath117 is nonsingular , so that almost surely @xmath115 is nonsingular . if it is known that @xmath118 , we assume instead that @xmath119 are in general position , in which case @xmath115 is also nonsingular . to get an understanding of the relative sizes of the entries in @xmath115 , consider the special case in which @xmath117 is block - diagonal with identical @xmath120 blocks @xmath121 , so that @xmath122 . then @xmath123 , showing how the covariance of the remainders @xmath124 and @xmath125 grows with the @xmath19-th power of the scalar product of @xmath126 and @xmath127 . if @xmath121 is diagonal , also @xmath115 is diagonal and the variance of the remainder @xmath124 grows with the @xmath128-th power of the distance between the actual argument of measurement @xmath126 and the position of interest @xmath14 . now the main step to solving eq.[eqn : bayes ] is to see that the posterior of @xmath80 given @xmath89 , @xmath129 , and @xmath90 is still gaussian , @xmath130 with precision matrix @xmath131 and mean @xmath132 given by @xmath133 where we call @xmath134 the _ squared weight matrix_. note that the basic behaviour of @xmath135 is to decrease with a growing distance of @xmath126 and @xmath127 from @xmath14 , similar to the weights used in idw . [ [ singularities . ] ] singularities . + + + + + + + + + + + + + + although in pathological cases , some of the involved matrices can be singular , they are nonsingular in general . more precisely , assume that @xmath90 has a continuous distribution and @xmath117 is nonsingular , so that @xmath115 is nonsingular , too ( see eq.[eqn : postr ] ) . then @xmath20 exist and is nonsingular for almost all choices of @xmath136 , including the case @xmath109 . in that case , and for generic @xmath137 and @xmath14 , @xmath79 has full rank @xmath138 ( like a vandermonde matrix ) , so that also @xmath139 is nonsingular when @xmath140 . finally , under these assumptions @xmath141 is nonsingular for almost all choices of @xmath142 , including the case @xmath143 . integrating over all possible values of the argument errors @xmath90 finally shows that the posterior of @xmath80 given @xmath89 and @xmath129 is a mixture of gaussians whose mean and covariance matrix are @xmath144 this posterior mean , which we call the _ motabar estimator _ , can now be used as a natural estimate of @xmath80 given @xmath89 , @xmath129 , and one can see that this estimate is * an affinely linear function of the measured values @xmath46 , with weights that decrease with growing distance between @xmath14 and @xmath45 , and * a mixture of values @xmath145 each of which is an infinitely smooth rational function of the position of interest @xmath14 that has no poles . because the entries of the inverse of an @xmath146 square matrix @xmath147 are rational functions of the entries of @xmath147 of degree at most @xmath148 , the degree of @xmath145 is at most @xmath149 . in practice , one can determine @xmath145 and @xmath150 like this , avoiding the large @xmath120 matrix inversions and solving instead two systems of linear equations : 1 . given @xmath14 , @xmath89 , and @xmath90 , compute @xmath79 , @xmath114 , and @xmath115 from eqns.[eqn : x ] and [ eqn : r ] . 2 . determine @xmath151 and @xmath152 by solving @xmath153 . 3 . compute @xmath154 . 4 . determine @xmath155 by solving @xmath156 . note that in pathological cases , either of the two systems might not be solvable uniquely , which possibility we do not discuss here . although the usually smaller @xmath146 matrix inversions needed to determine @xmath157 from in eq.[eqn : sigmapost ] are not as easily avoided , one can at least determine the posterior variance of @xmath158 by solving @xmath159 for @xmath160 and using @xmath161 in the integral in eq.[eqn : sigmapost ] to determine @xmath162 . if the argument errors @xmath90 are not known to be zero , the integrals over @xmath90 will usually have to be evaluated numerically even when @xmath90 is gaussian as well , since the integrand is a nonlinear function of @xmath90 . if @xmath90 has small variance , it can however be feasible to approximate the integrals by using a quadratic or 4th - order approximation of @xmath145 , which will be explored in a separate article . in addition to the obvious translation invariance in all dimensions , the motabar estimator is equivariant under various linear scaling transformations : ( i ) when @xmath136 , @xmath163 , and @xmath117 are all multiplied with the same constant @xmath164 , then @xmath165 remains unchanged and @xmath157 is multiplied by @xmath166 as well . ( ii ) when for all @xmath44 , all @xmath65 , and some @xmath30 , @xmath167 , @xmath168 , and @xmath169 are all multiplied with the same constant @xmath164 , the distribution of @xmath90 is stretched by a factor of @xmath166 along the @xmath30-th axis , and when @xmath170 , @xmath171 , the @xmath65-th row and column of @xmath163 , and the @xmath77-th row and column of @xmath117 are all divided by @xmath172 , then also @xmath173 and the @xmath65-th row and column of @xmath157 are divided by @xmath172 for all @xmath65 . to better understand the effects of the various inputs to motabar , let us consider a number of special and limit cases , some of which turn out to be equivalent to well - known existing methods , whereas others are presented to show limitations of our approach . fig.[fig : interpol12 ] illustrates the diversity of results one can get from the same data . ( colour online ) illustration of special and limit cases of motabar interpolation for an equidistant sample ( black dots ) from runge s function @xmath174 with one missing measurement at 0.2 . left : spline interpolation compared to motabar cases [ sec : nnbs ] ( piecewise polynomial based on four nearest neighbours ) and [ sec : lagrange ] ( lagrange polynomial ) . right : motabar cases [ sec : localpoly ] ( local cubic polynomial regression interpolation ) , [ sec : p2uncorr ] ( @xmath175 with uncorrelated errors and priors ) , [ sec : idw4 ] ( 4-th order idw ) , and a smoothing case ( @xmath176 , small error , penalised intermediate derivatives ) , all resulting in nonpolynomial rational functions . , title="fig:",scaledwidth=49.0% ] ( colour online ) illustration of special and limit cases of motabar interpolation for an equidistant sample ( black dots ) from runge s function @xmath174 with one missing measurement at 0.2 . left : spline interpolation compared to motabar cases [ sec : nnbs ] ( piecewise polynomial based on four nearest neighbours ) and [ sec : lagrange ] ( lagrange polynomial ) . right : motabar cases [ sec : localpoly ] ( local cubic polynomial regression interpolation ) , [ sec : p2uncorr ] ( @xmath175 with uncorrelated errors and priors ) , [ sec : idw4 ] ( 4-th order idw ) , and a smoothing case ( @xmath176 , small error , penalised intermediate derivatives ) , all resulting in nonpolynomial rational functions . , title="fig:",scaledwidth=49.0% ] for vanishing argument errors , @xmath96 becomes a dirac delta function , so that @xmath177 hence the estimate is a pole - free rational function of @xmath14 of degree at most @xmath149 . for most of the remainder of this section , we only derive @xmath178 and @xmath179 which can then be plugged into eqs.[eqn : estimator ] and [ eqn : sigmapost ] if argument errors exist . if there is no prior information about @xmath0 , an improper @xmath31-prior with @xmath180 and a centralised @xmath181-prior with @xmath182 can be used , in which case @xmath183 in this case , @xmath184 is that value of @xmath80 which minimises the quadratic function @xmath185 which provides an alternative way of numerical computation . although this is formally a weighted least - squares estimator , it does not estimate a global set of parameters @xmath186 as in a global linear regression model @xmath187 , but is a local model in which @xmath79 and @xmath20 depend on @xmath14 , so that the terms in the estimator have to be computed for each @xmath14 separately . [ [ consistent - estimation - of - polynomial - components . ] ] consistent estimation of polynomial components . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + if @xmath188 and @xmath189 , we have @xmath190 . hence if @xmath191 and @xmath129 is a monomial with @xmath192 for some @xmath65 with @xmath66 , then @xmath129 equals the @xmath65-th column of @xmath79 and hence @xmath193 equals the @xmath65-th column of @xmath194 . this means that the estimate @xmath195 is zero for @xmath196 and one for @xmath197 , which are the correct derivatives of the given monomial at @xmath191 . consequently , motabar with improper priors is also equivariant under the addition of polynomials , e.g. , a quadratic trend in a time series , a property shared with numerical differentiation via discrete legendre polynomials . however , while the latter has @xmath198 with an orthogonal coefficient matrix @xmath199 , the motabar coefficient matrix @xmath200 is not orthogonal in general . for proper priors with @xmath201 and @xmath202 , the absolute value of the @xmath65-th derivative is underestimated since the prior drags the estimate towards its mean at zero , and the absolute value of the other derivatives is slightly overestimated ( nonzero ) . this effect becomes smaller with increasing @xmath19 . for vanishing value errors , we have @xmath203 , hence @xmath204 and @xmath205 in the limit case of @xmath109 , this can also be derived directly from eq.[eqn : bayes ] by substituting @xmath206 . in that case , however , @xmath115 can become singular when @xmath207 for some @xmath44 , which has to be taken care of in the numerical solution , e.g. , by assuming very small but nonzero @xmath136 , or by using the following exact solution for the case @xmath208 : in that case , @xmath209 and , using the notation @xmath210 for a matrix @xmath211 without its @xmath212th row and @xmath30th column and a vector @xmath213 without its @xmath212th element , @xmath214 where @xmath215 are the prior precision and mean of @xmath216 conditional on @xmath209 , and @xmath217 is a vector of ones . if in addition to @xmath218 we have @xmath188 and @xmath189 , we get @xmath219 for general @xmath14 , while for @xmath220 , we get @xmath209 , @xmath221 , and @xmath222 with @xmath223 . this is interpolation based on local polynomial regression in which the weights @xmath224 decrease as a power of a quadratic form of @xmath225 and @xmath226 ( e.g. , the dash - dotted red line in fig.[fig : interpol12 ] right ) . hence , in contrast to other local polynomial methods such as loess , far away observations get a positive though small weight . at the other end of the error scale , for @xmath227 , we get @xmath228 and thus @xmath229 and @xmath230 . so if the @xmath31-prior is proper , the posterior approximates it , otherwise it diverges . the same behaviour obtains for a decreasingly informative @xmath181-prior , i.e. , for @xmath231 . this shows that while a noninformative @xmath31-prior might be chosen , the @xmath181-prior must be proper since it regulates the overall variability of the estimate . if the value error variance is finite ( @xmath232 ) , one might , on the other hand , assume that the derivatives of @xmath0 of degree @xmath233 are negligible compared to the scale of @xmath112 , e.g. , because @xmath0 is believed to be approximately a polynomial of degree at most @xmath234 . then , taking the limit @xmath182 and @xmath235 , we get @xmath236 which is a form of bayesian polynomial regression . although the squared weight matrix @xmath237 no longer depends on @xmath14 here , this is still not a global regression method since the @xmath31-prior might dependent on @xmath14 , e.g. , because one assumes some linear or nonlinear trend or some change in variability with @xmath14 . a truly global regression method can be obtained by using the noninformative @xmath31-prior with @xmath180 in addition to @xmath182 and @xmath235 . if @xmath238 , then @xmath239 in the limit , this is just ordinary global least - squares polynomial regression with polynomials of order @xmath234 and possibly correlated value errors of differing magnitude . for @xmath240 , @xmath175 , @xmath241 , and nonvanishing argument measurement errors , our model becomes the `` errors in the variables '' linear regression model @xmath242 since this is linear in @xmath243 , the integral over @xmath90 in eq.[eqn : bayes ] can be solved analytically . for @xmath191 , @xmath143 , @xmath244 , @xmath245 , @xmath246 , and @xmath247 , this results in the non - gaussian posterior @xmath248 the posterior mode has @xmath249 and @xmath250 where @xmath251 and @xmath252 , @xmath253 , @xmath254 , and @xmath255 . if @xmath256 , @xmath257 is the largest solution of the above equation , otherwise the smallest , and it has the same sign as @xmath258 . when the same model is estimated with the _ total least - squares _ method ( aka _ deming regression _ ) studied already by @xcite , the equation is @xmath259 instead , which is symmetric under an exchange of `` dependent '' and `` independent '' variables and leads to a larger absolute value of @xmath257 . for @xmath260 , this difference vanishes , and for @xmath261 , both solutions converge to the ordinary least - squares linear regression line with @xmath262 . for @xmath263 , total least squares gives @xmath264 independently of @xmath265 , while the cubic equation gives @xmath266 which still depends on @xmath265 and has @xmath267 for @xmath268 . this comparison shows that unlike in total least squares , it is essential in motabar whether we consider @xmath7 a function of @xmath6 or vice versa . to understand why the effects of value and argument errors are different in the linear model and why the cubic equation is not symmetric in @xmath269 , consider the simple case where the measurements are @xmath270 and only the middle one , @xmath271 , has errors in both dimensions which are independent with @xmath272 with equal probability then the real data @xmath273 are either @xmath274 or @xmath275 , both giving slope @xmath276 in linear regression , or @xmath277 or @xmath278 , both giving slope @xmath279 , so that the motabar posterior mean is @xmath280 , whereas total least squares results in @xmath276 because of the obvious symmetry . for @xmath240 , @xmath281 , and @xmath143 , we get lagrange interpolation with @xmath282 ( e.g. , the solid green line in fig.[fig : interpol12 ] left ) . for @xmath283 and @xmath284 , it depends on the placement of the @xmath126 whether the resulting polynomial is an interpolant or not ( see @xcite for an overview ) . a particularly simple case occurs for the minimal choice of @xmath19 , @xmath285 , and when both value errors and priors are uncorrelated , so that @xmath136 and @xmath117 are diagonal , and @xmath286 . then we have @xmath287 , @xmath288 , and @xmath289 , where @xmath290 the latter is the squared distance between @xmath126 and @xmath14 as measured by the quadratic norm that is weighted with the prior derivative variances @xmath291 . the posterior distribution of @xmath16 given @xmath90 is then gaussian with mean and variance given by @xmath292 this is a generalization of inverse distance weighting that takes into account value error via the occurrence of @xmath293 in eq.[eqn : bidww ] and prior information via the occurrence of @xmath294 in eq.[eqn : bidw ] . we propose to call this method _ bayesian idw smoothing . _ note that , as expected , @xmath295 is a rational function of @xmath14 of degree at most @xmath296 . setting @xmath109 , @xmath143 , and @xmath297 in the preceding , we get ordinary idw with squared distances , @xmath298 where @xmath295 is now a rational function of @xmath14 of degree exactly @xmath299 . note that while @xmath295 is independent from @xmath300 , the posterior variance is proportional to @xmath300 . in other words , our bayesian approach shows that the uncertainty of the idw estimate of order two is directly proportional to the scale of the derivatives of @xmath0 . if @xmath175 , @xmath240 , @xmath202 , @xmath244 , @xmath136 and @xmath117 are diagonal , and @xmath301 , then @xmath302 note that @xmath303 does not only depend on the distances of the @xmath126 from @xmath14 as encoded in @xmath304 , but also on the relative position of @xmath126 and @xmath127 , via the mixture terms in @xmath305 . see the dashed blue line in fig.[fig : interpol12 ] right for an example . this dependency vanishes if we let @xmath306 and @xmath307 , where we get an instance of the following case : assume that @xmath202 , @xmath244 , @xmath136 is diagonal , @xmath308 , and @xmath309 . if we let the prior for @xmath310 become noninformative by letting @xmath306 , but let the prior for @xmath60 with @xmath311 become sharp at @xmath312 by letting @xmath307 , then @xmath313 in other words , the motabar estimate converges to the ordinary idw smoothing ( or interpolation , if @xmath109 ) solution with exponent @xmath128 . e.g. , @xmath175 and @xmath109 gives 4-th idw interpolation ( solid green line in fig.[fig : interpol12 ] right ) . note that this , however , also shows that using idw with higher powers than two corresponds to implicitly assuming that some derivatives vanish which the result shows not to vanish after all . hence the only plausible form of idw is the one with squared distances . [ [ inconsistent - derivatives . ] ] inconsistent derivatives . + + + + + + + + + + + + + + + + + + + + + + + + + this example also illustrates an unintuitive feature of motabar : the estimates derivative @xmath314 need not coincide with the derivative of the estimated value @xmath315 with respect to @xmath14 . in idw smoothing , the estimated slope is zero as enforced by the sharp prior , although the estimated value is not constant . on the upper end of the order spectrum , for certain choices of priors , one can choose very large orders @xmath19 without running into numerical infeasibilities . a special limit case obtains with similar priors as above : assume @xmath109 , a noninformative prior for the function value @xmath158 ( so that @xmath316 ) , an uncorrelated and centralised @xmath181-prior with @xmath244 and @xmath309 , and that all derivatives of @xmath0 of order @xmath317 vanish at @xmath14 ( so that @xmath318 and @xmath319 for @xmath311 ) . in the limit for @xmath320 , then only the nearest neighbour @xmath126 of @xmath14 is used in the estimation . this is because then @xmath321 is diagonal and @xmath322 for @xmath323 , so that @xmath324 . in other words , motabar then estimates @xmath16 by @xmath325 , as in simple _ nearest neighbour estimation_. that is , for large @xmath19 and these priors , motabar estimate approximates a step function that is constant on the voronoi cells ( aka thiessen polyhedrons ) of the sample . as the 4-th order idw example in fig.[fig : interpol12 ] ( right ) shows , the step - like shape can already be seen for small values of @xmath19 . a step - like function with a similar smoothening at the edges also obtains for @xmath320 when argument errors @xmath90 are not vanishing , since then @xmath326 . on the other hand , nonvanishing value errors , even if very small , can lead to additional levels at the means of more than one nearest neighbour , showing that this limit behaviour is quite unstable ( see , e.g. , the thin yellow line in fig.[fig : interpol12 ] right ) . one can also get higher - order piecewise polynomial estimates , by assuming that only the derivatives of @xmath0 of order @xmath327 vanish at @xmath14 for some @xmath328 , and using an improper prior with @xmath329 for @xmath330 . then the motabar estimate approximates a piecewise polynomial interpolant of degree @xmath331 that uses the @xmath332 observations nearest to @xmath14 ( assuming those are in general position ) . for @xmath333 , this leads to a piecewise linear estimate that need not , however , coincide with ordinary linear interpolation since , e.g. , the two nearest neighbours of @xmath334 might both lie left of @xmath14 , leading to a discontinuity in the estimate to the right of @xmath14 . only when the positions @xmath126 build a regular polyhedral grid , the estimate approximates ordinary linear interpolation . for @xmath240 , this requires equidistant measurement positions , and for @xmath335 a regular triangular grid is required , whereas on a square grid , one gets a discontinuous linear interpolation with four square interpolation cells per grid cell , each corresponding to a different choice of three of the four corners of the grid cell . analogous results hold for higher - dimensional rectangular grids . for @xmath336 and @xmath240 , the limit result is a piecewise cubic polynomial which is , however , not a cubic spline since although its 2nd derivative is continuous , its value and slope are not ( e.g. , the dashed blue line in fig.[fig : interpol12 ] left ) . if the distance between @xmath14 and the data positions @xmath126 diverges , i.e. , if @xmath337 , we have @xmath338 . if the @xmath31-prior is informative ( @xmath201 ) , we then have @xmath339 that is , the data are too far away to drag the posterior significantly away from the prior . with a noninformative @xmath31-prior , on the other hand , the estimate for @xmath337 will either approximate the sample mean ( if @xmath285 ) or will diverge to @xmath340 ( almost surely for @xmath341 ) . if the @xmath126 are either regularly distributed or drawn from a sufficiently smooth distribution , the distance between @xmath14 and the closest @xmath126 will decay at a rate @xmath342 for @xmath343 . we conjecture that then the posterior precision grows at a rate @xmath344 for nonpathological error distributions and priors . the rationale for this conjecture is this : assume that @xmath109 , @xmath345 , and the @xmath40 arguments @xmath126 are sufficiently uniformly distributed over a @xmath37-dimensional ball of unit radius about @xmath14 , and let @xmath346 . then @xmath347 and this sum is asymptotically proportional to the integral @xmath348 } d^dx\,(x-\xi)^{\alpha+\beta}||x-\xi||_2^{-2p}\\ & \sim n\int_{z = n^{-1/d}}^1 dz\,z^{|\alpha|+|\beta|}z^{-2p } \sim n^{1+(2p-|\alpha|-|\beta|-1)/d}.\end{aligned}\ ] ] the conjecture implies @xmath349 , in particular @xmath350 . in particular , for @xmath240 , the conjectured rate of convergence of @xmath351 to @xmath16 is @xmath352 , the same as for spline interpolation . @xcite introduced a different approximation scheme that is also based on a local taylor expansion . adapting their terminology and notation to ours , that scheme can be described as defining an estimator @xmath353 for @xmath16 , where the weights @xmath354 are chosen to minimise the squared norm @xmath355 subject to a number of constraints @xmath356 for all @xmath65 with @xmath357 for some @xmath328 , where @xmath358 and @xmath359 are diagonal matrices whose entries are certain parameters @xmath360 and error variances , @xmath361 although the parameters @xmath362 correspond to the prior variances in our approach , wang et al.s rationale is not bayesian . their motivation is rather that @xmath363 is an estimator of the squared approximation error @xmath364 , and the constraints make sure that the estimator is correct if the true @xmath0 is a polynomial of degree @xmath365 . as we see from the usage of @xmath358 and @xmath359 instead of @xmath163 and @xmath366 , their scheme implicitly assumes uncorrelated `` priors '' and errors , but one can easily adapt it to deal with correlations by using @xmath367 instead of @xmath363 . also , they do not consider the case of argument errors , and for these it does not seem obvious how to deal with them consistently in their framework . to compare their scheme to ours in the case of no argument errors , note that the motabar estimate @xmath184 is a linear function of @xmath129 only when the prior for @xmath80 is improper and the prior for @xmath76 is centralised , while otherwise the estimate is only an _ affine _ function of @xmath129 . hence let us assume @xmath143 and @xmath244 in this section . in that case , the motabar estimator @xmath184 itself is that @xmath80 which minimises the quadratic function @xmath368 however , while in wang et al.s approach a constrained problem obtains , the above is an unconstrained problem . taking the corresponding limit of @xmath180 in wang et al.s scheme , their target function becomes @xmath369 so that their coefficients depend on the relative variances ( and correlation structure ) of @xmath80 but no longer on @xmath370 , while ours depends on the latter but not on the former . despite these differences , both approaches seem to have the same rate of convergence for @xmath343 and reduce to ordinary polynomial regression or idw for certain choices of control parameters . prior knowledge about @xmath80 , such as constraints of the form @xmath371 $ ] for some @xmath65 and @xmath372 , can easily be incorporated into motabar estimation by choosing a suitable prior distribution for @xmath80 , such as a uniform distribution of @xmath60 on @xmath373 $ ] or a gaussian restricted to this interval . if one then rewrites the chosen prior in the form @xmath374 which can be seen as a generalization of eq.[eqn : phiprior ] , it is easy to see that the resulting posterior simply involves the same factor @xmath375 , leading to a posterior distribution of @xmath376 with the same @xmath155 and @xmath141 as before . however , since the posterior mean is then no longer equal to @xmath155 , one either needs to integrate eq.[eqn : posteriorgeneralized ] explicitly or use the posterior mode instead . in some cases , the posterior mean can be determined analytically . e.g. , two - sided constraints of the form @xmath371 $ ] with finite bounds @xmath377 can most easily be modelled by putting @xmath329 and @xmath378 , where @xmath379 is the heaviside step function with @xmath380 for @xmath381 and @xmath382 otherwise . for one - sided constraints of the form @xmath383 , one can use @xmath384 . using the prior factor @xmath385 , the marginal posterior density of @xmath310 is proportional to @xmath386 with @xmath387 and @xmath388 . for a density of this form , the mean is given by @xmath389 for @xmath390 outside @xmath373 $ ] , this asymptotically equals @xmath391 with @xmath392 for @xmath393 ( good approximation for @xmath394 ) , or @xmath395 with @xmath396 for @xmath397 ( good approximation for @xmath398 ) . with this constraint , the motabar estimator becomes @xmath399 where @xmath400 depends on @xmath90 via @xmath390 and @xmath401 . letting @xmath307 in the preceding , we get @xmath402 to see that our methodology can be fruitful also in the case of non - gaussian @xmath181-priors and value errors , consider the case of @xmath240 , vanishing argument errors , an improper @xmath31-prior , independent laplace @xmath181-priors with @xmath403 for some @xmath404 , and independent value errors with asymmetric laplace distributions @xmath405 , where @xmath406 , @xmath407 , and @xmath408 $ ] . then @xmath409 where @xmath410 and the posterior mode of @xmath80 given @xmath89 , @xmath129 is the one that maximises @xmath411 . for @xmath412 , @xmath413 , and sharp @xmath181-priors with @xmath414 , this reduces to minimizing the quantile regression loss function , @xmath415 , so the above can be considered a novel form of local polynomial quantile regression . similar analytical forms of @xmath111 obtain also for other forms of @xmath181-priors , e.g. , gaussian or uniform ones . with a noninformative @xmath31-prior , the inconsistency between @xmath314 and the @xmath65-th derivative of @xmath416 with respect to @xmath14 will decrease with @xmath19 and vanish for @xmath320 . therefore , if one is interested in all derivatives of order @xmath365 , we suggest to use a noninformative prior for all @xmath60 with @xmath330 , and use a @xmath19 somewhat larger than @xmath331 , using either informative or noninformative priors for the derivatives of order @xmath417 . ( colour online ) typical effect of priors for data points placed uniformly at random ( black dots ) from @xmath418 ( thin dashed black line ) , with @xmath419 and @xmath176 . left : @xmath420 , no value error . right : @xmath421 , medium - sized iid value error with @xmath422 . in 100 such samples , the median @xmath423 distance between true and estimated function in the left[right ] case was 0.068[0.24 ] for proper correlated @xmath31-priors growing as @xmath424 with the correct frequency @xmath425 ( thick solid cyan line ) , 0.15[0.29 ] for proper uncorrelated priors of the same size ( thick dash - dotted cyan line ) , 0.22[0.48 ] for proper correlated priors with @xmath426 ( thin solid blue line ) , 0.25[1.7 ] for improper priors ( thin dash - dotted red line ) , and 0.56[0.55 ] for proper correlated priors with @xmath427 ( thin dotted green line ) . , title="fig:",scaledwidth=49.0% ] ( colour online ) typical effect of priors for data points placed uniformly at random ( black dots ) from @xmath418 ( thin dashed black line ) , with @xmath419 and @xmath176 . left : @xmath420 , no value error . right : @xmath421 , medium - sized iid value error with @xmath422 . in 100 such samples , the median @xmath423 distance between true and estimated function in the left[right ] case was 0.068[0.24 ] for proper correlated @xmath31-priors growing as @xmath424 with the correct frequency @xmath425 ( thick solid cyan line ) , 0.15[0.29 ] for proper uncorrelated priors of the same size ( thick dash - dotted cyan line ) , 0.22[0.48 ] for proper correlated priors with @xmath426 ( thin solid blue line ) , 0.25[1.7 ] for improper priors ( thin dash - dotted red line ) , and 0.56[0.55 ] for proper correlated priors with @xmath427 ( thin dotted green line ) . , title="fig:",scaledwidth=49.0% ] [ [ correlation - of - derivatives . ] ] correlation of derivatives . + + + + + + + + + + + + + + + + + + + + + + + + + + + assume that @xmath240 and @xmath428 is of the form @xmath429 with a spectrum @xmath430 and phases @xmath431 chosen uniformly at random . then the covariance of @xmath432 and @xmath433 is @xmath434 in other words , the @xmath30-th and @xmath435-th derivatives are fully correlated , and the @xmath30-th and @xmath436-th derivatives are fully anti - correlated ( see also @xcite ) . if the spectrum of @xmath0 is known approximately , eq.[eqn : corr ] the above is therefore a natural choice for the @xmath31-prior covariance structure . for unknown spectrum , however , using a prior with strong correlation between derivatives can result in worse results than using an uncorrelated or improper prior , as can be seen in the example in fig.[fig : sin_noerr_proper_improper ] . [ [ growth - of - derivative - variance . ] ] growth of derivative variance . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + if @xmath430 is concentrated at a single frequency @xmath437 , i.e. , @xmath438 , the variance of @xmath432 is growing exponentially : @xmath439 . for a broader but bounded spectrum , it is growing sub - exponentially . e.g. , a uniform spectrum of the form @xmath440 for @xmath441 and @xmath442 for @xmath443 gives @xmath444 . for typical unbounded spectra , in contrast , the variance growth is super - exponential . e.g. , for an exponentially decaying spectrum with @xmath445 for some @xmath446 , as is often assumed of chaotic dynamical systems , we get @xmath447 . similarly , for a gaussian decaying spectrum with @xmath448 for some @xmath449 , we get @xmath450 . in case of a power - law decaying spectrum , the variance even diverges for large @xmath30 . e.g. , for @xmath451 , we get @xmath452 for @xmath453 and @xmath454 for @xmath455 , justifying the use of an improper prior . without argument error , the _ posterior predictive distribution _ of @xmath46 from the taylor model about @xmath456 has @xmath457 if the model is correct , one therefore expects that approximately @xmath458 likewise , for any fixed @xmath14 , and if @xmath143 and @xmath459 , the posterior predictive distribution of @xmath460 from the taylor model about @xmath14 has @xmath461 if the model is correct , one therefore expects that approximately @xmath462 hence , assuming @xmath463 with a known matrix @xmath464 and unknown @xmath449 , one could either use the same @xmath465 for all @xmath14 and choose it so that the `` global '' eq.[eqn : postpredglobal ] is fulfilled , which is similar to the suggested parameter choice in @xcite , or use a different @xmath465 for each @xmath14 and choose it so that the `` local '' eq.[eqn : postpredlocal ] is fulfilled . if @xmath109 and @xmath466 , @xmath184 does not depend on @xmath465 , and eq.[eqn : postpredlocal ] becomes @xmath467 ( see eq.[eqn : postr ] ) . this value is also identical to the posterior mode of @xmath465 when @xmath465 is treated as a scale hyperparameter with a noninformative jeffreys prior of @xmath468 . hence the motabar interpolant with noninformative priors is @xmath469 to demonstrate the performance of motabar for complex data , we simulated a trajectory @xmath470 of the standard chaotic lorenz system given by the odes @xmath471 , @xmath472 , @xmath473 , where @xmath79 basically oscillates between @xmath474 with a frequency of @xmath475 ( see fig.[fig : lorenz],a ) . we then generated a sample of @xmath476 noisy observations @xmath477 , with iid errors @xmath478 ( which corresponds to @xmath479 of noise ) , for time - points @xmath480 that were regularly spaced at a distance @xmath481 ( which corresponds to @xmath482 data points per oscillation , see the black dots in fig.[fig : lorenz],f ) . from the sample , we reconstructed the original trajectory ( fig.[fig : lorenz],a ) up to a diffeomorphism , following the approach of @xcite by estimating the time evolution of the derivatives @xmath483 ( fig.[fig : lorenz],b ) , using different estimation methods . for motabar estimation , we used @xmath484 , an improper @xmath31-prior , and a @xmath181-prior whose variance @xmath485 was chosen to fit the approximate range and frequency of the oscillation . fig.[fig : lorenz](e ) shows that the motabar approach reproduces the shape of the trajectory much better than either spline smoothing ( c ) or numerical differentiation ( d ) , the latter being basically equivalent to an alternative phase space reconstruction method , the often used `` method of delays '' . note that there are other , more sophisticated state space reconstruction methods based on pca or discrete legendre polynomials , which deal better with noise @xcite and which will be compare to motabar in a separate paper . ( a ) ( colour online ) reconstruction of the lorenz attractor from noisy measurements of its @xmath79 component . ( a ) simulated true phase space trajectory . ( b ) derivatives of @xmath79 from numerical differentiation from simulated true @xmath79 data . ( c ) derivatives from natural cubic smoothing spline for noisy measurements of @xmath79 . ( d ) numerical differentiation from noisy measurements . ( e ) derivatives from motabar with @xmath484 and an improper @xmath31-prior . ( f ) detail of motabar estimated @xmath79 ( thick blue line ) compared to simulated true @xmath79 ( thin gray line ) , and noisy measurements ( black dots ) . , title="fig:",scaledwidth=45.0%](b ) ( colour online ) reconstruction of the lorenz attractor from noisy measurements of its @xmath79 component . ( a ) simulated true phase space trajectory . ( b ) derivatives of @xmath79 from numerical differentiation from simulated true @xmath79 data . ( c ) derivatives from natural cubic smoothing spline for noisy measurements of @xmath79 . ( d ) numerical differentiation from noisy measurements . ( e ) derivatives from motabar with @xmath484 and an improper @xmath31-prior . ( f ) detail of motabar estimated @xmath79 ( thick blue line ) compared to simulated true @xmath79 ( thin gray line ) , and noisy measurements ( black dots ) . , title="fig:",scaledwidth=45.0% ] + ( c ) ( colour online ) reconstruction of the lorenz attractor from noisy measurements of its @xmath79 component . ( a ) simulated true phase space trajectory . ( b ) derivatives of @xmath79 from numerical differentiation from simulated true @xmath79 data . ( c ) derivatives from natural cubic smoothing spline for noisy measurements of @xmath79 . ( d ) numerical differentiation from noisy measurements . ( e ) derivatives from motabar with @xmath484 and an improper @xmath31-prior . ( f ) detail of motabar estimated @xmath79 ( thick blue line ) compared to simulated true @xmath79 ( thin gray line ) , and noisy measurements ( black dots ) . , title="fig:",scaledwidth=45.0%](d ) ( colour online ) reconstruction of the lorenz attractor from noisy measurements of its @xmath79 component . ( a ) simulated true phase space trajectory . ( b ) derivatives of @xmath79 from numerical differentiation from simulated true @xmath79 data . ( c ) derivatives from natural cubic smoothing spline for noisy measurements of @xmath79 . ( d ) numerical differentiation from noisy measurements . ( e ) derivatives from motabar with @xmath484 and an improper @xmath31-prior . ( f ) detail of motabar estimated @xmath79 ( thick blue line ) compared to simulated true @xmath79 ( thin gray line ) , and noisy measurements ( black dots ) . , title="fig:",scaledwidth=45.0% ] + ( e ) ( colour online ) reconstruction of the lorenz attractor from noisy measurements of its @xmath79 component . ( a ) simulated true phase space trajectory . ( b ) derivatives of @xmath79 from numerical differentiation from simulated true @xmath79 data . ( c ) derivatives from natural cubic smoothing spline for noisy measurements of @xmath79 . ( d ) numerical differentiation from noisy measurements . ( e ) derivatives from motabar with @xmath484 and an improper @xmath31-prior . ( f ) detail of motabar estimated @xmath79 ( thick blue line ) compared to simulated true @xmath79 ( thin gray line ) , and noisy measurements ( black dots ) . , title="fig:",scaledwidth=45.0%](f ) ( colour online ) reconstruction of the lorenz attractor from noisy measurements of its @xmath79 component . ( a ) simulated true phase space trajectory . ( b ) derivatives of @xmath79 from numerical differentiation from simulated true @xmath79 data . ( c ) derivatives from natural cubic smoothing spline for noisy measurements of @xmath79 . ( d ) numerical differentiation from noisy measurements . ( e ) derivatives from motabar with @xmath484 and an improper @xmath31-prior . ( f ) detail of motabar estimated @xmath79 ( thick blue line ) compared to simulated true @xmath79 ( thin gray line ) , and noisy measurements ( black dots ) . , title="fig:",scaledwidth=45.0% ] [ [ outlook - alternative - local - approximations . ] ] outlook : alternative local approximations . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the simple form of the motabar estimator is due to our assumption of gaussian value measurement errors and the fact that taylor polynomials are linear in their coefficients . alternatively , one might approximate @xmath0 by other functions that can be parameterised by the low - order derivatives of @xmath0 at @xmath14 . for @xmath240 and if it is known that @xmath0 displays oscillatory behaviour , one such approximation could be @xmath486 with a remainder function @xmath487 with @xmath488 . because then @xmath489 , @xmath490 , @xmath491 , and @xmath492 , the four derivatives can be estimated from the model if a plausible prior for @xmath493 is used , although the estimate will not be a linear function in @xmath129 since the above approximation is not linear in the coefficients . if @xmath0 is known to be periodic with frequency @xmath425 , it might seem that one could also use partial sums of the corresponding fourier series instead , which are linear in their coefficients , but they are not parameterizable by a finite number of derivatives of @xmath0 at @xmath14 . [ [ software . ] ] software . + + + + + + + + + an open - source software package implementing motabar for use with the python programming language is under development and will be made available at http://www.pik-potsdam.de/members/heitzig/motabar . this work was supported by the german federal ministry for education and research ( bmbf ) via the potsdam research cluster for georisk analysis , environmental change and sustainability ( progress ) . the author thanks forest w. simmons , kira rehfeld , norbert marwan , bedartha goswami , and jrgen kurths for fruitful discussions .

the prominent broad emission blends in the spectra of active galactic nuclei have been shown to vary in response to continuum variations , but past attempts to measure the reverberation lag time of the optical lines have met with only limited success . here we report the detection of reverberation in two seyfert 1 galaxies , ngc 4593 and mrk 1511 , based on data from a program carried out at lick observatory in spring 2011 . light curves for emission lines including h and were measured by applying a fitting routine to decompose the spectra into several continuum and emission - line components , and we use cross - correlation techniques to determine the reverberation lags of the emission lines relative to-band light curves . in both cases the measured lag ( ) of is longer than that of h , although the inferred lags are somewhat sensitive to the choice of template used in the fit . for spectral decompositions done using the template of , we find()/(h ) in ngc 4593 and in mrk 1511 . the detection of highly correlated variations between and continuum emission demonstrates that the emission in these galaxies originates in photoionized gas , located predominantly in the outer portion of the broad - line region .

1304.4643

blends of emission lines are often among the most prominent broad emission features in the ultraviolet ( uv ) and optical spectra of broad - lined active galactic nuclei ( agns ) , and the integrated flux of emission in quasars can be greater than that of any other single emission line , including ly @xmath6 @xcite . despite more than three decades of observational and theoretical effort , the physical conditions that give rise to emission have remained very difficult to determine . this is due in part to the complex energy - level structure of the fe@xmath7 ion and the very large number of individual emission lines that appear in broad blends . recent spectral synthesis models for the fe@xmath7 ion include hundreds of energy levels and tens of thousands of individual transitions @xcite . a variety of processes in the broad - line region ( blr ) can contribute to line production , including collisional excitation as well as continuum and line fluorescence @xcite . it is not yet fully determined whether the production in agns occurs in gas heated solely by photoionization , or whether collisional ionization might play a significant or even dominant role @xcite . a particularly promising recent development has been the realization that anisotropy in the emission from high column density clouds can have a strong impact on the observed spectrum , and models for the emission from the shielded side of photoionized clouds have shown significant improvements in fitting observed spectra @xcite . the emission lines are closely linked to several fundamental issues in agn physics and phenomenology . from principal - component analysis of quasar spectra , the relative strength of emission is one of the major characteristics of `` eigenvector 1 '' @xcite , the component which accounts for much of the variance among the quasar population . empirically , eigenvector 1 represents the anticorrelations between and [ ] equivalent widths , and between strength and h@xmath0 width . the fundamental physical driver of the eigenvector 1 sequence is the eddington ratio @xmath8 @xcite , but the physical mechanism responsible for the increase in strength with eddington ratio is not fully understood . it may be linked to the presence of a soft x - ray excess at high eddington ratio , which would produce a larger partially ionized zone of warm , -emitting gas ( e.g. , * ? ? ? * ; * ? ? ? * ) , while orientation could play a secondary role @xcite . @xcite argue against the x - ray excess as the primary driver of strength , proposing instead that differences in gas - phase iron abundance , driven by selective depletion of iron onto grains , are primarily responsible for the wide range of observed line strengths in agns . in another interpretation proposed by @xcite and @xcite , the competing forces of gravity and radiation pressure set a critical column density for clouds to remain gravitationally bound within the blr , and at higher eddington ratio , the higher column density of surviving blr clouds would produce a larger /h@xmath0 flux ratio . for agns having extremely strong emission , it has also been suggested that the shock - heated gas associated with circumnuclear star formation may be responsible for the enhancement @xcite . furthermore , emission is potentially valuable as a tracer of chemical evolution . the strength of the uv emission relative to in quasar spectra has been used as a proxy for the iron to @xmath6-element abundance ratio in the blr ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , although the sensitivity of / line ratios to factors such as gas density and microturbulence means that observed line ratios are not straightforward indicators of the underlying abundance ratios @xcite . improved understanding of the physical conditions responsible for emission could have important ramifications for elucidating the history of metal enrichment in the densest regions of the universe at high redshift . a key quantity of interest is the spatial scale of the -emitting region . however , there are very few measurements available that directly constrain its size . @xcite carried out reverberation mapping of the uv lines in ngc 5548 and found a lag of about 10 days , similar to that of ly@xmath6 , indicating that the uv lines originate from within the blr . this has been the only measurement of a reverberation lag for the uv lines to date . aside from reverberation mapping , one additional direct constraint exists : a spectroscopic microlensing study of a lensed quasar by @xcite found evidence that the emission originates largely in the outer portion of the blr . previous attempts at reverberation mapping of the optical blends have generally not led to clear detections of reverberation lags . for the well - studied agn ngc 5548 , @xcite examined 13 years of monitoring data and found that the flux responded to continuum changes on timescales of less than several weeks , but the available data did not allow for a definite measurement of the lag . @xcite measured light curves for the blends in ark 120 and found a long - term variability trend that followed the continuum changes , but the data did not yield a significant cross - correlation lag . they concluded that the emission region in ark 120 was likely to be several times larger than the h@xmath0-emitting zone of the blr . an alternative possibility was that the emission might be powered by collisional excitation rather than by photoionization , although they were unable to fit the emission blends well with collisional excitation models . @xcite measured the light curve for pg 1700 + 518 using monitoring data from @xcite . while they were able to detect evidence of reverberation in the lines , the cross - correlation analysis did not yield a highly significant peak , and the measured lag of @xmath9 days was very uncertain . variability of the optical emission has been examined in other agns ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but most datasets have not been suitable for measurement of reverberation lags . the overall picture that emerges from these studies is that while optical emission does respond to continuum variations at least over long timescales , it tends to have a lower amplitude of variability than h@xmath0 , and it does not generally show a clear reverberation lag signature . in the absence of direct constraints on the relative sizes of the and h@xmath0 emission regions , indirect clues have come from line - profile measurements . @xcite and @xcite found that the velocity widths of the lines are generally similar to that of h@xmath0 , suggesting that and h@xmath0 originate from the same region within the blr . subsequent work has uncovered subtle systematic differences between h@xmath0 and optical widths , which hint at an origin for in the outer portion of the blr or perhaps within an `` intermediate - line region '' corresponding to the transition between the blr and the dusty torus @xcite . for a sample of 4000 sloan digital sky survey ( sdss ) quasar spectra , @xcite found that the full - width at half maximum intensity ( fwhm ) of the optical lines is typically about @xmath10 fwhm(h@xmath0 ) , albeit with substantial scatter , implying a size for the -emitting region that is typically about twice as large as the h@xmath0-emitting zone of the blr . here , we present new reverberation - mapping results for two nearby seyfert galaxies , ngc 4593 and mrk 1511 , from the lick agn monitoring project 2011 . section 2 gives a brief overview of our observing campaign . section 3 describes the fitting method used to isolate the continuum and emission - line components in the spectroscopic data and the measurement of light curves . in section 4 we describe the cross - correlation lag measurements and the dependence of the lags on the choice of template , and section 5 presents a discussion of the results . in an appendix , we present a brief discussion of results based on additional newly - available templates . while the reverberation lag measurements are modestly sensitive to the choice of template , both of these objects clearly show evidence of reverberation in response to continuum variations . initial results from the lick agn monitoring project 2011 were previously presented by @xcite and @xcite , and full details of the spectroscopic and photometric observing campaigns , analysis procedures , and light - curve data for all targets will be presented in a forthcoming series of papers . here , we briefly review the key aspects of the observing program . from early march through mid - june 2011 , we obtained queue - scheduled @xmath1-band images of ngc 4593 and mrk 1511 using the 0.76 m katzman automatic imaging telescope ( kait ) at lick observatory @xcite , the 0.9 m telescope at the brigham young west mountain observatory ( wmo ) , the faulkes telescope south at siding spring observatory , and the palomar 1.5 m telescope @xcite . our goal was to obtain nightly imaging for each target , but weather and scheduling issues led to some gaps in coverage , particularly during the initial portion of the campaign . the temporal sampling cadence of our observations is described in table [ lcstats ] . exposure times were typically 180300 s. all images were bias - subtracted and flattened , and cosmic - ray hits were removed using the la - cosmic routine @xcite . image - subtraction photometry was carried out using a version of the isis code @xcite modified by the high-@xmath11 supernova search team @xcite and with additional modifications by w. li , and also using the ` hotpants ` package by a. becker , which is based on the methods described by @xcite . for each telescope , a high - quality template image was chosen , and the template was then aligned with each night s image and convolved with a spatially varying kernel to match the point - spread function of that image . after subtracting the scaled template image , the variable agn flux is left as a point source in the subtracted image , allowing for aperture photometry using the iraf daophot package . the photometric aperture radius used for each telescope was set to match the average point - source fwhm for images from that telescope . lccc ngc 4593 photometry & 75 & 0.95 & 1.12 + ngc 4593 photometry ( averaged ) & 57 & 1.01 & 1.48 + ngc 4593 spectroscopy & 43 & 1.02 & 1.81 + mrk 1511 photometry & 71 & 1.00 & 1.51 + mrk 1511 photometry ( averaged ) & 59 & 1.05 & 1.82 + mrk 1511 spectroscopy & 40 & 1.01 & 1.94 + light curves were initially constructed separately with the data from each telescope , and then combined together to assemble a single light curve for each agn . normalization was done by modeling the variability using gaussian processes to obtain a finely - sampled model version of the light curve for each agn , following the same procedure described by @xcite . additive and multiplicative scaling factors were applied to match the modeled light curve to each telescope light curve , using the wmo light curve as the reference data for the model . the normalization code simultaneously constrained all scaling factors to obtain the best fit between all the telescope light curves using a markov chain monte carlo algorithm . then , scaling factors determined from the modeling were applied to the image - subtraction light curves , to normalize the data from all telescopes to a common flux scale . the final light curve was prepared by taking a weighted average of observations taken within 6 hours of one another . the @xmath1-band light curves are shown in figure [ figlightcurves ] , in arbitrary units of differential flux . our spring 2011 observing campaign consisted of 69 nights at the lick 3-m telescope between 2011 march 27 and june 13 ut . observations were done using the kast dual spectrograph @xcite . a d55 dichroic was used to separate the blue and red beams with a crossover wavelength of about 5500 . in this paper , we discuss only measurements from the blue arm of the spectrograph , where we used a 600 lines mm@xmath12 grism over @xmath135520 at a scale of 1.0 pixel@xmath12 . all observations of these two agns were done with a 4-wide slit oriented at a position angle of 45 . standard calibration frames including arc lamps and dome flats were obtained each afternoon , and flux standards were observed during twilight . exposure times were normally @xmath14 s and @xmath15 s for ngc 4593 and mrk 1511 , respectively . the weather at mt . hamilton was somewhat worse than average during spring 2011 ; we were able to observe ngc 4593 on 43 of the 69 nights , and mrk 1511 on 40 nights . spectroscopic reductions and calibrations followed standard methods implemented in iraf and idl . a large extraction width of 103 was used in order to accommodate the full extent of the agn spatial profiles observed on nights with poor seeing . error spectra were extracted and propagated through the full sequence of calibrations . in the reduced spectra , the median signal - to - noise ratio ( s / n ) per pixel in the range 46004700 is 122 for ngc 4593 and 71 for mrk 1511 . the reduced spectra were first normalized to a uniform flux scale by employing the procedure of @xcite . this method applies a flux scaling factor , a linear wavelength shift , and a gaussian convolution to each spectrum in order to minimize the residuals between the data and a reference spectrum constructed from several of the best - quality nights . the scaling is determined using a wavelength range containing the [ ] @xmath16 line , which is assumed to have constant flux . the upper panels of figure [ figmean - veron ] show the mean of all scaled spectra for each object . to assess the accuracy of the scaling procedure , we measured the light curve of the [ ] @xmath16 line in the scaled spectra and calculated the residual `` excess '' scatter @xmath17 in the [ ] light curve . we follow the usual definition for the normalized excess variance @xcite : @xmath18,\ ] ] where @xmath19 is the total number of observations , @xmath20 is the mean flux , and @xmath21 and @xmath22 are the individual flux values and their uncertainties . then @xmath17 gives a measure of the fractional scatter in the light curve due to random errors in flux scaling , over and above the scatter that would be expected based on the propagated photon - counting uncertainties . for ngc 4593 and mrk 1511 , we obtain an excess scatter of @xmath23 and 2.0% , respectively . these low values indicate that the scaling procedure worked well . as we describe below , these measures of residual scatter will be added to the error budget of the emission - line light curves to give a more realistic estimate of the total uncertainties . traditionally , broad emission - line fluxes in reverberation - mapping data have been measured by choosing continuum regions on either side of an emission line , fitting a line to the continuum , and integrating the flux above that line ( e.g. , * ? ? ? . however , in agn spectra there may be no pure continuum regions at all surrounding h@xmath0 , because , , and other emission features are present . choosing continuum regions that contain some emission - line flux could potentially bias a reverberation lag measurement , particularly for velocity - resolved reverberation signals in the faint high - velocity wings of h@xmath0 . an alternative approach is based on decomposition of spectra into continuum and emission - line components , so that the flux contributions of individual emission features can be isolated . our multi - component fitting method was previously applied to mrk 50 @xcite and is similar to the method used by @xcite . we carry out fits over the wavelength range extending from just blueward of the h@xmath24 line ( @xmath25 ) up to 5470 near the dichroic cutoff . over this wavelength range , each nightly spectrum was fit with a model consisting of several components : a power - law featureless continuum ( fc ) , a starlight template broadened in velocity by convolution with a gaussian , emission lines including h@xmath0 , [ ] @xmath26 , @xmath27 , and three lines ( 4471 , 4922 , and 5016 ) , and an template broadened in velocity by convolution with a gaussian . for the starlight template , we use an 11 gyr - old simple stellar population model at solar metallicity from @xcite . we experimented with adding younger stellar population components , but the 11 gyr model provided a sufficiently good fit that more complex stellar population models are not warranted , and the flux of a young population component is usually not very well constrained over this limited wavelength range . we used two different templates to model the lines , from boroson & green ( 1992 ; hereinafter bg92 ) , and vron - cetty et al . ( 2004 ; hereinafter v04 ) . the overall quality of the fits is very similar with the two templates , but there are some subtle differences that affect the reverberation lag measurements . we find that the v04 template produces better results than the bg92 template ( see section [ sec : templates ] ) , but we describe measurements and reverberation lags derived from both sets of spectral fits in order to illustrate the dependence of the results on the choice of template . the broad component of h@xmath0 , as well as the [ ] lines , are represented by fourth - order gauss - hermite functions , while gaussians were used to represent weaker features including narrow h@xmath0 , the broad and narrow components of , and the lines . we found that the complex shape of the h@xmath24 + [ ] @xmath28 blend was difficult to model , and instead of adding more model components we simply set the wavelength range subtended by this feature to have zero weight in the fit ( 42804395 rest wavelength ) . additionally , a uniform foreground extinction was applied to the model spectrum using the @xcite reddening law . this extinction correction represents the combined effects of galactic foreground extinction , extinction within the agn host galaxy , and wavelength - dependent slit losses resulting from the nonparallactic slit orientation @xcite . since these slit losses differ from night to night , we allowed the value of @xmath29 to vary freely in each fit . the full model includes 29 free parameters , and was fitted to the observed spectrum by @xmath30 minimization using a levenberg - marquardt technique @xcite . for each galaxy , the fit was first carried out on the high - s / n mean spectrum , and then the parameters from the best fit to the mean spectrum were used as starting parameter estimates for the fit to each nightly spectrum . for both objects , the fit optimization always drove the flux of @xmath31 to zero , using either template . however , inclusion of the other lines at 4922 and 5016 does significantly improve the fit quality compared to model fits that omit these lines , at least when using the v04 template . previous work has shown that these two lines can be important contributors to the `` red - shelf '' region redward of h@xmath0 @xcite . the red shelf also contains some emission @xcite and the i zw 1 templates contain some flux in this region . however , the relative amount differs between the bg92 and v04 templates , such that the v04 template has relatively weaker emission in the h@xmath0 red shelf , and this leads to noticeably different fitting results for the two templates . using the v04 template , we found that the template alone did not produce an adequate fit to the red shelf unless the lines were added as separate components . when fits are carried out using the bg92 template instead , the fluxes of the @xmath32 components in the h@xmath0 red shelf region go to zero or nearly zero , and essentially all of the flux in the h@xmath0 red shelf is taken up by the template . another difficulty in fitting this region is that the @xmath33 line is degenerate with the red wing of h@xmath0 ; in order to achieve a consistent deblending of this spectral region into h@xmath0 and flux , we constrained the 4922 and 5016 lines to have identical fluxes ( following * ? ? ? * ) and identical velocity widths . in the fits , these two components essentially track the same shape as two broad bumps in the broadened templates . due to the degeneracy between the several features contributing to the h@xmath0 red shelf , we are unable to determine a physically unique decomposition of this region into separate contributions from h@xmath0 , , , and [ ] . the fits are able to match the observed spectra well in this region using either template , but the slight differences in the red shelf decompositions may be partly responsible for the dependence of the reverberation lags on the choice of template . the strength and broadening of the component in the fit are primarily determined by the prominent blend around @xmath344700 containing several transitions from multiplets 37 and 38 . the velocity - broadening width of was allowed to vary independently to optimize the fit to each nightly spectrum , but the variation from night to night was relatively small . from the best fit to each night s data using the v04 template , the median velocity broadening ( i.e. , the dispersion of the gaussian broadening kernel ) applied to the template is @xmath35 km s@xmath12 for ngc 4593 and @xmath36 km s@xmath12 for mrk 1511 . the v04 template itself was constructed using lorentzian profiles for each line , with a width of fwhm = 1100 km s@xmath12 to match the line profiles in i zw 1 @xcite . we find a similarly small night - to - night scatter in the width of the broadening kernel when using the bg92 template ( @xmath37 km s@xmath12 ) , although the overall amount of broadening is substantially smaller due to the different template structure . the bg92 template is characterized by an intrinsic fwhm@xmath38900 km s@xmath12 ( e.g. , * ? ? ? * ) , but as an empirically - derived template it contains features with a range of widths , and shapes that do not correspond precisely to simple gaussian or lorentzian profiles . as a result , it is not straightforward to compare the inferred widths as measured using the two different templates , but for each template the fitting procedure consistently converges on a value for the broadening that varies only by a few percent from night to night . we comment further on the profile widths in section [ discussion ] . figures [ figmean - veron ] and [ figmean - boroson ] show the mean spectrum for each agn with its best - fitting model , and the individual fit components , for the fits done using the v04 and bg92 templates . the root - mean - square ( rms ) spectrum is constructed by taking the standard deviation of the scaled spectra ( e.g. , * ? ? ? in each of these figures we show the rms spectrum , as well as a modified version constructed by subtracting the agn fc and stellar - continuum components from each nightly spectrum before calculating the rms . removing the continuum components gives an improved rms spectrum by eliminating small residual stellar absorption - line features @xcite , and the continuum - subtracted rms spectrum more accurately depicts the rms variability profiles of the emission lines . fits done using the bg92 template assign relatively more flux to than fits with the v04 template . for the bg92 fits , the increased flux ( at the expense of the fc component ) results in a higher rms flux in the continuum - subtracted rms spectra . light curves were measured for individual emission lines from continuum - subtracted spectra . for h@xmath0 and the h@xmath24 + [ ] blend , we produced a residual spectrum by subtracting all other model components , and then measured the emission - line light curves by summation of the residual flux . the h@xmath24 and h@xmath0 integration regions were 43204450 and 48005000 for ngc 4593 , and 44204560 and 49205120 for mrk 1511 ( in the observed frame ) . the resulting light curves include the constant contributions of narrow h@xmath0 and h@xmath24 as well as [ ] @xmath28 blended with h@xmath24 . light curves for @xmath27 were measured by summation of the best - fitting broad and narrow model components . this produced a less noisy light curve than the alternate approach of calculating a residual spectrum for each night by subtracting all of the other model components . however , for ngc 4593 the emission was too weak to produce a useful light curve . the light curves were computed by summation of the best - fitting model over 44004900 , but any wavelength range would yield the same overall light - curve shape since the model fit assumes a uniform flux scaling factor for the template . the fc light curves were measured by integration over the same wavelength range used for . as described above , the residual scatter measured from the [ ] light curves represents a major contribution to the error budget in the light - curve measurements . therefore , we combine the residual scatter ( 1.2% for ngc 4593 and 2.0% for mrk 1511 ) in quadrature with the measurement uncertainties in the individual light - curve points to obtain a more realistic estimate of the true overall uncertainties . the emission - line light curves displayed in figure [ figlightcurves ] are shown with these expanded error bars . addition of the residual flux - scaling scatter to the error budget widens the measurement uncertainties in the cross - correlation lags described below , but only slightly . in figure [ figlightcurves ] we show only the spectroscopic light curves measured using the v04 template fits because the light curves measured from the bg92 fits are extremely similar in overall appearance . observations of other agns have shown that the amplitude of variability is typically somewhat smaller than that of h@xmath0 . for example , in ngc 5548 @xcite found that the variation amplitude was @xmath39%75% of that of h@xmath0 . to characterize the variability amplitude in ngc 4593 and mrk 1511 , we use @xmath17 as defined in equation 1 , which gives a measure of the normalized rms variability , corrected for measurement uncertainties . based on the v04 fits , the results for ngc 4593 are @xmath17(h@xmath0)@xmath40 and @xmath17()@xmath41 , while for mrk 1511 we measure @xmath17(h@xmath0)@xmath42 and @xmath17()@xmath43 . with the bg92 template fits , the @xmath17 values for h@xmath0 are essentially unchanged , while @xmath17 ( ) increases by @xmath44 relative to the values determined from the v04 fits . thus , both objects show a somewhat higher amplitude of variability in h@xmath0 than , similar to the behavior seen in ngc 5548 @xcite . for agns having a very low starlight fraction , it is often possible to measure the agn continuum flux from the spectra directly , without carrying out a decomposition to separate the agn flux from the host galaxy contribution ( e.g. , * ? ? ? . however , this approach is much less successful for agns having a substantial host galaxy component in the spectra , since the starlight dilutes the variability of the agn continuum and introduces additional random noise due to nightly variations in seeing and target centering within the slit ( e.g. , * ? ? ? given the substantial contributions of both stellar continuum and in these two objects , the spectral decomposition method is necessary in order to measure the agn continuum flux without contamination or dilution by these other components . in order to measure the cross - correlation function ( ccf ) for unevenly sampled time series , we employ the interpolation cross - correlation function ( iccf ) methodology and monte carlo error - analysis techniques described by @xcite , @xcite , and @xcite . emission - line lags for h@xmath0 , h@xmath24 , , and were measured relative to both the @xmath1-band and fc light curves . ccfs were computed over a temporal range of @xmath45 to @xmath46 days in increments of 0.25 days . for each ccf we compute two measures of the lag time : @xmath47 , which is the lag at the peak of the ccf , and @xmath2 , the centroid of the ccf for all points above 80% of the peak value @xcite . table [ xcortable ] lists the lag values . the quantity @xmath48 listed in table [ xcortable ] gives the peak amplitude of the ccf and is a measure of the significance of the correlation between the two light curves . we obtain higher @xmath48 values for ngc 4593 than for mrk 1511 , primarily owing to the higher s / n of the ngc 4593 light curves . lag values are given in the observed frame and can be converted to the agn rest frame by dividing by @xmath49 , where the redshifts are @xmath50 for ngc 4593 and @xmath51 for mrk 1511 . there are advantages and drawbacks to using either the @xmath1-band data or spectroscopic fc light curves in the cross - correlation analysis . the @xmath1-band light curves have more frequent sampling than the spectroscopic data , and better temporal sampling will always improve the determination of the ccf . additionally , the spectroscopic data suffer from systematic effects including variable slit losses due to miscentering and differential atmospheric refraction , while the photometric data are not susceptible to these problems . however , the @xmath1-band data include contributions of emission - line flux in addition to the agn continuum , while the spectroscopic decompositions can be used to produce a `` pure '' fc light curve without contamination by strong emission lines . a disadvantage of using the spectroscopic data in cross - correlation measurements is the effect of correlated flux - calibration errors between the emission - line and continuum light curves , which can introduce a spurious signal at zero lag . we find that using the @xmath1-band light curves produces consistently higher - quality results , but the reverberation lags are consistent when measured against either the @xmath1 band or fc light curves . for completeness , table [ xcortable ] presents all of the lag measurements based on the spectral decompositions using both the v04 and bg92 templates , and for cross - correlations computed with both the @xmath1-band and fc light curves . we tested the effect of detrending the data by subtracting a linear fit to the light curves before computing the ccfs . as described by @xcite , removing long - term secular variations from light curves can often improve the accuracy and significance of cross - correlation results . for mrk 1511 , detrending had almost no effect on the ccfs , because the light curves are not characterized by any long - term brightening or fading . ngc 4593 does show a net increase in luminosity over the duration of our monitoring program , but ccfs computed using detrended data were actually of poorer quality than the original ccfs , having values of @xmath48 that were lower by @xmath52 . therefore , we chose not to use detrended light curves for our final measurements . we first describe the results based on light curves measured from the v04 template fits . comparison with bg92 template fits is presented in section [ sec : templates ] . figure [ figccf ] illustrates the ccfs measured relative to the @xmath1-band data for emission - line light curves measured from the v04 template fits , as well as the autocorrelation function ( acf ) of the @xmath1-band light curve . for both agns the ccf has a shape roughly similar to that of the h@xmath0 ccf , but systematically shifted toward longer lags . lccc|ccc[t ! ] ngc 4593 : & & & & + h@xmath24 vs. @xmath1 & 0.95 & @xmath53 & @xmath54 & 0.94 & @xmath55 & @xmath54 + h@xmath24 vs. fc & 0.94 & @xmath56 & @xmath57 & 0.80 & @xmath58 & @xmath59 + h@xmath0 vs. @xmath1 & 0.92 & @xmath60 & @xmath61 & 0.94 & @xmath62 & @xmath63 + h@xmath0 vs. fc & 0.92 & @xmath64 & @xmath65 & 0.82 & @xmath66 & @xmath67 + vs. @xmath1 & 0.88 & @xmath68 & @xmath69 & 0.85 & @xmath70 & @xmath71 + vs. fc & 0.85 & @xmath72 & @xmath73 & 0.82 & @xmath74 & @xmath75 + mrk 1511 : & & & & + vs. @xmath1 & 0.84 & @xmath76 & @xmath77 & 0.83 & @xmath78 & @xmath79 + vs. fc & 0.81 & @xmath80 & @xmath81 & 0.79 & @xmath82 & @xmath83 + h@xmath24 vs. @xmath1 & 0.91 & @xmath84 & @xmath85 & 0.91 & @xmath86 & @xmath87 + h@xmath24 vs. fc & 0.83 & @xmath88 & @xmath89 & 0.80 & @xmath90 & @xmath91 + h@xmath0 vs. @xmath1 & 0.89 & @xmath92 & @xmath93 & 0.88 & @xmath94 & @xmath95 + h@xmath0 vs. fc & 0.82 & @xmath96 & @xmath97 & 0.81 & @xmath98 & @xmath99 + vs. @xmath1 & 0.80 & @xmath100 & @xmath101 & 0.81 & @xmath102 & @xmath103 + vs. fc & 0.75 & @xmath104 & @xmath105 & 0.73 & @xmath106 & @xmath107 + [ xcortable ] normally , the iccf method computes two versions of the ccf : first by interpolating the continuum light curve ( shifted by each trial value of the lag ) to the temporal steps of the emission - line light curve , and then by interpolating the emission - line light curve to the time sampling of the continuum light curve . with high - quality data , the two resulting ccfs are normally very nearly identical , and the final ccf is taken to be the mean of the two @xcite . however , when the two light curves have significantly different temporal sampling frequency , or when the light curves have poor s / n , the two interpolated ccfs can sometimes differ significantly in shape , and occasionally one of the two ccfs may contain features that are clearly spurious while the other has a more regular appearance . in such a situation it may be preferable to use just one of the two interpolated ccfs rather than taking their mean . for ngc 4593 , the two interpolated ccfs are nearly identical , and we base our measurements on the mean ccfs as illustrated in figure [ figccf ] . however , for mrk 1511 , we found that the second version of the ccf ( i.e. , from interpolating the light curve to the time steps of the @xmath1 band ) had a large and unphysical `` notch '' appearing just at the ccf peak . the other interpolation produced a smooth and regular ccf . overall the two interpolated ccfs had nearly identical shapes and widths , and they show the same overall lag response aside from the notch appearing at the peak of the first ccf . the presence of this notch in the mean ccf rendered the measurement of @xmath47 and @xmath2 suspect . since the first interpolated ccf results in a smoothly shaped peak , for mrk 1511 we use only the first interpolated version of the ccf to calculate @xmath47 and @xmath2 , rather than taking the mean of the two interpolations . this ccf is shown in figure [ figccf ] . days . this is consistent with the value listed in table [ xcortable ] but with a larger uncertainty due to the peculiar shape of the peak region of the mean ccf . the value of @xmath2(h@xmath0 ) is nearly unchanged if we use the mean ccf instead of the first interpolated version . ] the h@xmath0 lags measured relative to the @xmath1-band light curves are in the range of @xmath1084 days for ngc 4593 and @xmath1097 days for mrk 1511 . due to asymmetry in the ccfs , the values of @xmath2 and @xmath47 differ slightly , but not by a significant amount considering the uncertainties in the lag measurements . the observed trend of h@xmath0 having a longer lag than h@xmath24 is expected based on previous measurements ( e.g. , * ? ? ? similarly , the fact that the lag is unresolved in mrk 1511 is consistent with expectations based on previous measurements . in agns having h@xmath0 lags of @xmath110 days , the lag time is usually undetectable in datasets with nightly sampling ( e.g. , * ? ? ? * ; * ? ? ? the emission has a longer lag than h@xmath0 for both agns . in both cases the cross - correlation between the and @xmath1-band light curves produces a significant ccf peak , with @xmath111 for ngc 4593 and 0.80 for mrk 1511 . we also measured the emission - line lags relative to the agn fc component light curves . the resulting @xmath2 and @xmath47 values are generally consistent , within the uncertainties , with the lag values measured against the @xmath1-band data . however , the ccfs measured using the fc light curves have lower peak amplitude @xmath48 than the corresponding ccfs measured against the @xmath1 band , which we attribute to the poorer temporal sampling and greater calibration uncertainties of the spectroscopic data . thus , we consider measurements relative to the @xmath1 band to be the best determinations of the emission - line lags , while the lags measured relative to the fc serve as a useful consistency check . as another check , we measured the fc light curves over a different wavelength range ( 50505150 rest wavelength ) and measured the cross - correlations of the h@xmath0 light curves relative to this revised fc light curve . the measured lags only changed by a negligible amount ( much smaller than the 1@xmath112 uncertainties ) in comparison with the lags measured to the fc light curves as listed in table [ xcortable ] . figure [ figccf - fc ] shows the vs. fc ccfs for both sets of template fits . for mrk 1511 , the choice of template has very little impact on the ccf shape or the lag in this measurement . however , the difference between the two templates is significant for ngc 4593 , where the light curves measured from the bg92 template produce a very jagged and irregular ccf and yield very unreliable lag measurements with large uncertainties on @xmath2 and @xmath47 . the v04 template , on the other hand , gives a better - behaved result and smaller lag uncertainties . the cause of this difference is not obvious , but it does appear that the v04 template gives a clearly superior result at least for ngc 4593 . an additional possible concern is degeneracy between the stellar and agn continuum components in the fit . for both objects , the light curves of the starlight component do not show any significant time - dependent features other than random noise . we measured cross - correlations between the stellar continuum component and the @xmath1-band light curves in order to test whether the stellar continuum light curves might contain some residual agn flux . the resulting ccfs had very low peak amplitudes of @xmath113 and 0.43 for ngc 4593 and mrk 1511 , respectively , and had such irregular structure that no clear lag signal could be measured in either case . this provides further evidence that the spectral decompositions are properly separating the contributions of the continuum components . a small bias in reverberation lag measurements can occur when broad - band photometry is used for the continuum flux measurement , due to the contribution of emission - line flux to the photometric light curves . the passband of the @xmath1 filter includes flux from both h@xmath0 and , both of which lag the continuum variations . when cross - correlating spectroscopic emission - line light curves against the @xmath1-band light curves , this emission - line contamination in the @xmath1-band data will tend to bias the cross - correlation lag toward values that are lower than the true lag @xcite . to determine the level of this contamination for mrk 1511 as an example , we extrapolated our mean - spectrum fit over a broader wavelength range ( up to 6500 ) and carried out synthetic @xmath1-band photometry on the individual model components . in the mean spectrum , the ratios of emission - line to agn featureless continuum ( fc ) flux as measured through a @xmath1-band filter are @xmath114(h@xmath0)/@xmath114(fc ) @xmath115 , and @xmath114()/@xmath114(fc ) @xmath116 . we simulated the effect of the emission - line flux in the @xmath1-band light curves by creating mock reverberation datasets having properties similar to those of our observed data . following the method of @xcite , we began by generating simulated agn continuum light curves on a finely sampled temporal grid with bin size equal to 0.01 day , based on a power - density spectrum of the form @xmath117 ( similar to slopes measured from _ agn light curves by * ? ? ? each light curve was generated over a total duration of 1000 days . a shorter segment of length 100 days was then selected randomly and normalized to have an rms variability equal to 15% of the median flux . the h@xmath0 and lines were assumed to respond linearly to continuum variations , with a delta function as the transfer function for simplicity , and with lag times of 6 days for h@xmath0 and 9 days for . variability amplitudes for the emission lines were normalized so that h@xmath0 had the same 15% rms variability as the continuum , while was set to have a lower rms variability amplitude of 12% . then , a `` contaminated '' @xmath1-band light curve was created by adding scaled versions of the h@xmath0 and light curves to the fc light curve , in the relative proportions listed above for mrk 1511 . all of the light curves ( fc , @xmath1 band , h@xmath0 , and ) were then degraded to approximate the sampling and s / n of real data . from the finely sampled light curves , one flux point was chosen per night , randomly sampled from a window of @xmath118 hours relative to a uniform 24-hour cadence , and over a total spectroscopic monitoring duration of 80 days . weather losses were approximated by randomly removing 10% of the points from the fc and @xmath1-band light curves and 40% of the points from the emission - line light curves . random gaussian noise was added to attain s / n = 100 for the fc and @xmath1-band data , and s / n = 50 for the emission - line light curves . the light curve was cross - correlated against both the pure fc light curve and the contaminated @xmath1-band light curve , to determine the typical level of the bias , and this procedure was repeated @xmath119 times to build up a distribution of lag measurements for different initial realizations of the light - curve shape . we compiled the values of @xmath2 for the subset of simulations that produced `` successful '' measurements of lag , meaning that the ccf had @xmath120 and yielded a nonnegative value of @xmath2 ; by these criteria 83% of the simulations were successful . for this subset , the median lag of relative to the pure agn continuum is @xmath121 days , which is closely consistent with the input lag of 9 days . ( the uncertainty of @xmath122 days represents the 68% confidence interval on @xmath2 for the set of successful simulations . ) when the simulated light curves are cross - correlated against the simulated @xmath1-band data including emission - line contributions , we find a median lag of @xmath123 days . as expected , this is shorter than the lag measured with respect to the pure agn continuum , but the difference of 0.7 days in the median measurements is smaller than the uncertainty in the measured lag for either of our targets . thus , the impact of this bias on the emission - line lag measurements is fairly small , most likely below the level of the @xmath124 uncertainties in the lags . the effect of emission - line contamination in the @xmath1-band light curves would be similar for all of the cross - correlation measurements for a given agn , so the inferred relative sizes of the and h@xmath0 emission regions would be only modestly affected . among the reverberation measurements listed in table [ xcortable ] , the strongest differences between results based on the v04 and bg92 template fits are in the lag of relative to the @xmath1 band . the bg92 fits give @xmath2 values that are 23 days lower than the values measured from the v04 fits . this difference appears to stem from the different proportions of fc and fluxes for the two sets of spectral fits . as can be seen in figures [ figmean - veron ] and [ figmean - boroson ] , the broadened v04 template approaches zero flux density for the emission in the blue wing of h@xmath0 , as well as at the blue end of our fitting region . the broadened bg92 template , on the other hand , does not reach zero flux density anywhere in our fitting region , producing a higher `` pseudo - continuum '' level from blended lines and forcing the fc component to a lower overall flux than in the v04 fits . we suspect that the bg92 template fits are including a small amount of agn continuum flux into the fit component . this small contribution of agn continuum flux in the light curves would add a spurious signal at zero lag to the cross - correlations between and the @xmath1-band data , slightly biasing the lags toward low values . these differences stem from the methods used to construct the templates . the bg92 template is based on an observed spectrum of i zw 1 . emission lines from several transitions other than were removed , and the continuum was removed by fitting a polynomial to regions between strong emission lines , leaving an approximate emission spectrum . the v04 template is also based on an observed spectrum of i zw 1 , but was constructed by removing a much larger list of non- emission features , and also removing narrow permitted and forbidden lines which were found to be much stronger in i zw 1 than in most typical seyfert 1 spectra . each line in the remaining spectrum was then modeled with a lorentzian profile of fwhm@xmath125 km s@xmath12 , producing a noise - free template . the more rigorous removal of non- emission features and narrow lines by v04 is a major reason to prefer the v04 template over the bg92 template for these fits . furthermore , the empirical and somewhat subjective procedure used to remove the agn continuum by bg92 may be responsible for the fitting degeneracy between the and fc components described above . if the agn continuum were undersubtracted when constructing the bg92 template , there would be almost no noticeable impact on the quality of individual spectral fits that used the template , but the mixing of fc and emission would likely bias the reverberation lags measured for the component . for ngc 4593 , the ccf measured between the and fc light curves is certainly better when the v04 template is used , compared with the bg92 template . the reason for this difference is not clear , and it only appears to occur for ngc 4593 and not mrk 1511 , but it does add one additional reason to prefer the v04 template fits . for all of these reasons , we consider the measurements done using the v04 template fits to be our best - quality results . most importantly , our main results are independent of the template choice : in either case we find that emission does show a strong reverberation response , with a lag time longer than that of h@xmath0 . as described in the appendix , we find similar results when using the newly released multi - component templates of @xcite . with these new templates , we obtain slightly different values of @xmath2 for the emission lines , but the lags are still consistently longer than those of h@xmath0 , with @xmath2()/@xmath2(h@xmath0)@xmath126 . this further confirms that the larger size of the -emitting region ( compared with h@xmath0 ) is genuine and not an artifact of template choice . it is also worth noting that the different template fits yield differing amounts of and emission underlying the broad h@xmath0 profile . the uncertain amount of blending of h@xmath0 with other emission features ( each having a different lag relative to the continuum ) could represent a limiting factor for the accuracy of high - fidelity reverberation measurements , particularly for velocity - resolved lag measurements which attempt to determine the distribution of lag across the velocity width of the line . any ambiguity in decomposing the h@xmath0 spectral region into different emission - line components should be considered as a source of systematic uncertainty in determining the h@xmath0 lag distribution across the line profile . this is the first time that such clear reverberation signals have been seen for the optical blends in seyfert galaxies . our high - cadence monitoring data reveal that the emission in these galaxies does reverberate on short timescales in response to continuum variations , with a well - defined cross - correlation lag time . this gives direct evidence for an origin of the emission in photoionized gas in the blr . the ambiguity in previous reverberation results for ngc 5548 @xcite and ark 120 @xcite might be attributable to the lower cadence of the monitoring data used in these studies , although in both cases the monitoring duration spanned several years . it is also possible that the variability behavior in our two objects is not representative of the entire population of seyferts . measurement of variability over a broad range of agn properties should be a high priority for future high - cadence reverberation - mapping programs . by exploring a broad range of luminosities , it might be possible to test whether the optical emission follows a radius - luminosity relationship similar to that of the h@xmath0 line , with a luminosity dependence of approximately @xmath127 @xcite . it would be particularly interesting to test whether the ratio of to h@xmath0 radii varies systematically along the eigenvector 1 sequence as a function of @xmath8 . if the -emitting gas is infalling toward the central engine , as proposed by @xcite , then there might also be observable correlations between the reverberation lag and the redshift of the lines , although evidence for radial inflow remains controversial @xcite . the measured values of @xmath2 give an approximate mean radius for the zone from which each line is emitted within the blr . comparing the lags of and h@xmath0 for the preferred v04 template fits , we find @xmath3()@xmath128(h@xmath0 ) @xmath129 and @xmath5 in ngc 4593 and mrk 1511 , respectively . this gives a direct indication that the emission arises predominantly in the outer portion of the blr on larger scales than the h@xmath0 emission region . this conclusion is consistent with several other recent lines of evidence for an outer blr location for the emission @xcite . we use the term `` outer blr '' to denote emission from a region having a larger mean size than the h@xmath0 emission zone . aside from , the only broad emission lines seen to have lags longer than that of h@xmath0 are ] @xmath130 @xcite and h@xmath6 ( e.g. , * ? ? ? * ; * ? ? ? thus , the emission regions for these lines correspond to the outermost observable portion of the blr . the bg92 template fits give different ratios of to h@xmath0 lag , but we find that @xmath2 ( ) @xmath131 @xmath2(h@xmath0 ) for either template . the longer lag of relative to h@xmath0 does not mean that the h@xmath0 and emission regions are physically distinct ; in fact , there must be a very substantial radial overlap between them . recent progress in transfer - function modeling for reverberation - mapping data has provided direct illustrations of the radial extent of the blr as seen in balmer lines . for arp 151 , the transfer functions show that the h@xmath0 emission extends over a broad radial zone with an outer extent that is more than twice as large as @xmath132 , and the h@xmath6-emitting region is several times larger in extent than the mean h@xmath0 radius @xcite . if the ratios of to h@xmath0 size for ngc 4593 and mrk 1511 are typical , then the -emitting zone would encompass the outer portion of the h@xmath0-emitting zone of the blr and beyond , probably corresponding roughly to the region emitting h@xmath6 . this echoes the recent finding by @xcite that quasar h@xmath0 profiles generically contain an intermediate - width component which may be emitted from the same region that produces , while the very broad component of h@xmath0 would be emitted from smaller radii . further support for a link between the region and the outer portion of the balmer - line emitting zone of the blr comes from a new principal - component analysis of quasar spectra by @xcite . they demonstrate that one of the primary eigenspectra of their quasar sample is essentially the sum of two components : the emission spectrum and an intermediate - width core component of the balmer lines , which are readily interpreted as arising from a similar spatial region . @xcite found that line widths in sdss quasars are typically about 3/4 of the h@xmath0 widths . in order to test whether ngc 4593 and mrk 1511 are consistent with this trend , we examine the results of the spectral fits done using the bg92 template , since this closely follows the fitting method used by @xcite . we assume that the total velocity width of fwhm is given by the quadrature sum of the fwhm of the broadening kernel from the fit and the fwhm of the line profiles in the template itself which we take to be 900 km s@xmath12 for consistency with @xcite . we also correct the observed widths for an instrumental broadening of fwhm@xmath133 km s@xmath12 @xcite . then , we obtain widths of fwhm@xmath134 km s@xmath12 for ngc 4593 and @xmath135 km s@xmath12 for mrk 1511 , where the uncertainties are based on the night - to - night scatter in the width of the gaussian broadening kernel determined by the fitting procedure . the broad - h@xmath0 component models for the two agns have fwhm = @xmath136 km s@xmath12 for ngc 4593 and @xmath137 km s@xmath12 for mrk 1511 . in both cases , then , fwhm()/fwhm(h@xmath0 ) is almost precisely 0.75 , closely consistent with the average result for the @xcite sample . however , we also find that the inferred widths seem to be substantially dependent on choice of template . using our results from fitting with the v04 template , we obtain broader widths of fwhm = @xmath138 km s@xmath12 and @xmath139 km s@xmath12 for ngc 4593 and mrk 1511 , in both cases broader than the h@xmath0 fwhm values . this appears to be the result of a combination of factors including different relative strengths for individual lines between the two templates , and the fact that the v04 template is constructed using lorentzian models fitted to each line . one consequence of these lorentzian profiles is that the fwhm values of the template itself and of the gaussian broadening kernel do not simply add in quadrature to give the fwhm of the total line profile ; the fwhm of the gaussian - broadened template profile is broader than the quadrature sum of the template fwhm and broadening kernel fwhm . based on the fits done using the bg92 template , we can conclude that these two agns have fwhm()/fwhm(h@xmath0 ) ratios consistent with the mean value found by @xcite for a large sdss sample , so there is no evidence that they are outliers from the normal agn population . however , the absolute determination of width seems to be subject to substantial systematic uncertainty due to the details of how the fit is performed and choice of template . further investigation of the cause of this discrepancy in widths for different templates is beyond the scope of this paper , but this problem is likely to affect all inferences about profile widths measured by template fitting , particularly when the individual features are completely blended into a pseudocontinuum as is the case in these two objects . in addition to the systematic issues related to choice of template , there are some additional and important caveats to note about the measured lags . the ccf methodology only gives a single , simplistic measure of the lag time for an emission line , whereas in reality a given spectral line will be emitted over a broad range of radii . it might be possible to obtain more detailed information on the radial distribution of emissivity by applying the geometric modeling methods described by @xcite or by applying techniques to extract the shape of the transfer function from the data @xcite . however , the light curves are relatively noisy and higher - quality data might be required , perhaps over a longer monitoring duration , in order to go beyond the simple cross - correlation determination of the lag presented here . furthermore , our spectral fits assume a uniform flux scaling for the entire template , which results in an average lag measurement for the entire complex of blends , but this method is unable to explore the possibility of different response times for different lines . recent work has improved on traditional template - fitting methods by allowing for different behavior among different groups of lines @xcite , and @xcite demonstrated that in ark 564 , lines from different multiplets showed differing levels of correlation with continuum variations . application of such methods to reverberation - mapping data could potentially detect or constrain differences in reverberation timescale for different multiplets as well . the primary conclusions of this study are that measurement of the reverberation lag of the optical blends is indeed possible in favorable cases , and that in these two agns the emission responds directly to continuum variations with a lag time that corresponds to the outer portion of the blr , somewhat larger than the h@xmath0-emitting radius . with well - sampled data it is possible to detect this reverberation signature with a level of significance comparable to that of high - quality h@xmath0 reverberation measurements . detection of reverberation does require some special circumstances , in particular a high amplitude of continuum variability , as well as relatively strong emission overall . in all other objects from our 2011 sample , either the emission was too weak ( as in mrk 50 ; * ? ? ? * ) , or the overall flux variability was too low for measurements like these to be successful . finally , these results illustrate the value of carrying out spectral decompositions of reverberation - mapping data , in order to measure accurate light curves for , , and other weak or low - amplitude spectral features . application of these methods for measurement of emission - line light curves for our entire 2011 sample will be described in future papers in this series . we are extremely grateful to the lick observatory staff for their outstanding assistance during our 2011 observing run . the lick agn monitoring project 2011 is supported by nsf grants ast-110812 , 1107865 , 1108665 , and 1108835 . a.p . acknowledges support from the nsf through the graduate research fellowship program . a.v.f.s group at uc berkeley received additional funding through nsf grant ast-1211916 , gary & cynthia bengier , the richard & rhoda goldman fund , the tabasgo foundation , and the christopher r. redlich fund . kait and its ongoing operation were made possible by donations from sun microsystems , inc . , the hewlett - packard company , autoscope corporation , lick observatory , the nsf , the university of california , the sylvia & jim katzman foundation , and the tabasgo foundation . t.t . acknowledges a packard research fellowship . the work of d.s . and r.j.a . was carried out at jet propulsion laboratory , california institute of technology , under a contract with nasa . research by j.l.w . is supported by nsf grant ast-1102845 . j.h.w . acknowledges support by the national research foundation of korea ( nrf ) grant funded by the korea government ( mest ) ( no . 2012 - 006087 ) . the west mountain observatory receives support from nsf grant ast-0618209 . we thank the anonymous referee for helpful suggestions . we mourn the tragic passing of our friend and collaborator , weidong li , who devotedly oversaw the nightly operation of kait and taught us much about photometry . this work is dedicated to the memory of lick observatory staff member greg sulger . after this paper was submitted , the referee informed us that a public release of the multicomponent templates from @xcite had recently been announced @xcite . although a full study of variability using these new templates is beyond the scope of this paper , we carried out an initial examination by adapting our fitting method to use these templates . as described by @xcite and @xcite , the template set includes spectra describing lines from four separate multiplet groups ( denoted as the _ f _ , _ g _ , _ s _ , and _ p _ groups ) , as well as an additional template containing other lines found in the i zw 1 spectrum . carrying out fits using this set of five templates gives substantially more freedom to accurately fit observed spectra than the monolithic templates of bg92 or v04 , at the cost of adding four additional free parameters to allow for the individual flux scaling of each component . figure [ figmean - kov ] illustrates the fits to the mean spectra using these multicomponent templates . similar to the fits with the v04 and bg92 templates , the model reproduces the overall spectral shape well , but there are significant differences in the fit details compared with the monolithic templates . using the @xcite templates , the fit assigns relatively more flux to the starlight component and less to the fc . also , these fits force the fluxes of all three lines to zero , and the red shelf of h@xmath0 becomes dominated by , similar to the results from using the bg92 template . another notable difference is seen at the shortest wavelengths , below about @xmath140 . in this region , the primary contribution is from the _ p _ group , and the fit forces the normalization of this component to zero . this is most likely a spurious result due to degeneracies in the model fitting process , but it occurs consistently when fitting each individual spectrum for both agns . these multicomponent templates are optimally suited for use with agn - dominated spectra having very strong emission , such as the objects studied by @xcite and @xcite , but the high starlight fraction and relatively weaker emission in our targets presents a more difficult case study for constraining the weights of the five - component model . in order to check for differences with respect to the previous spectral fits , we measured h@xmath0 and light curves based on these decompositions following the same methods described previously , and carried out cross - correlations of these light curves against the @xmath1-band continuum . the light curve was integrated over the same spectral region used for the monolithic templates , in this case corresponding primarily to flux from the _ f _ group template . for ngc 4593 , we find @xmath2(h@xmath0)@xmath141 days , and @xmath2()@xmath142 days . for mrk 1511 , the results are @xmath2(h@xmath0)@xmath143 days and @xmath2()@xmath144 days . these results are generally consistent with the measurements done using the bg92 and v04 template fits , with the largest disagreements only being at slightly greater than the @xmath124 level . the modest disagreements further highlight the fact that the reverberation results are somewhat sensitive to the different structures of the templates . despite these differences , our primary result is essentially unchanged : the reverberation lags based on the new template fits are @xmath145 longer than those of h@xmath0 , pointing to an origin for the emission in the outer portion of the blr . in future work , these multicomponent templates may prove to be most advantageous when fitting very high s / n spectra of agns having smaller starlight contributions and stronger lines whose relative amplitudes can be more tightly constrained in the fits .

in quantum simulation , many - body phenomena are probed in controllable quantum systems . recently , simulation of bose - hubbard hamiltonians using cold atoms revealed previously hidden local correlations . however , fermionic many - body hubbard phenomena such as unconventional superconductivity and spin liquids are more difficult to simulate using cold atoms . to date the required single - site measurements and cooling remain problematic , while only ensemble measurements have been achieved . here we simulate a two - site hubbard hamiltonian at low effective temperatures with single - site resolution using subsurface dopants in silicon . we measure quasiparticle tunneling maps of spin - resolved states with atomic resolution , finding interference processes from which the entanglement entropy and hubbard interactions are quantified . entanglement , determined by spin and orbital degrees of freedom , increases with increasing covalent bond length . we find separation - tunable hubbard interaction strengths that are suitable for simulating strongly correlated phenomena in larger arrays of dopants , establishing dopants as a platform for quantum simulation of the hubbard model .

1507.06125

quantum simulation offers a means to probe many - body physics that can not be simulated efficiently by classical computers , using controllable quantum systems to physically realize a desired many - body hamiltonian@xcite . in the analog approach to quantum simulation exemplified by cold atoms in optical lattices@xcite , the simulator s hamiltonian maps to the desired hamiltonian . compared to digital quantum simulation , realized via complex sequences of gate operations@xcite , analog quantum simulation is usually carried out with simpler building blocks . for example , the heisenberg and hubbard hamiltonians of great interest in many - body physics are directly synthesized by cold atoms in optical lattices@xcite . although of immense interest and proposed long ago@xcite , analog simulation of fermionic hubbard systems has proven to be very challenging@xcite . the anticipated regime of the intensely debated spin liquid , unconventional superconductivity , and pseudogap@xcite has yet to be accessed even for cold atoms . here , the required low temperature @xmath0 is problematic due to the weak tunnel - coupling @xmath1 of cold atoms@xcite . moreover , experimentally resolving individual lattice sites , crucial elsewhere in bose - hubbard simulation@xcite , remains very challenging in quantum simulation of the hubbard model . here , we perform atomic resolution measurements resolving spin - spin interactions of individual dopants , realizing an analog quantum simulation of a two - site hubbard system . we demonstrate the much desired combination of low effective temperatures , single - site spatial resolution , and non - perturbative interaction strengths of great importance in condensed matter@xcite . the dopants physical hamiltonian @xmath2 , determined at the time of fabrication@xcite , maps to an effective hubbard hamiltonian @xmath3 , where @xmath4 is the on - site coulomb repulsion , @xmath5 ( @xmath6 ) creates ( destroys ) a fermion at lattice site @xmath7 with spin @xmath8 , @xmath9 is the number operator , and h.c . is the hermitian conjugate . here , it is desirable to achieve non - perturbative ( intermediate ) interaction strengths @xmath10 associated with quantum fluctuations and emergent phenomena@xcite , _ i.e. _ , beyond perturbative heisenberg interactions ( large @xmath10 ) realized in photon - based@xcite and ion - based@xcite simulations , and magnetic ions on metal surfaces@xcite . we focus on the system ground state , prepared by relaxation upon cooling@xcite , rather than system dynamics . = tunnel rate to tip , @xmath11 tunnel rate from reservoir ) the inter - acceptor coupling @xmath1 obeys @xmath12 . @xmath13 measures the interacting states quasi - particle wavefunction ( qpwf ) , contains interference processes from which we obtain two - body wavefunction amplitudes , and determine the entanglement entropy and effective hubbard interactions . b. acceptor pair ( double - protrusion ) in topography at @xmath14 v and @xmath15 pa ( top ) , and spectrally and spatially resolved @xmath13 taken at a bias @xmath16 v where topography is flat apart from atomic corrugation ( bottom ) . valence band ( vb ) , 2-hole ground state and 2-hole excited states are indicated . c. effective energy diagram of sequential hole tunneling through 2-hole ground and excited state of coupled acceptors . ] because the states of our artificial hubbard system are coupled and interacting , tunneling spectroscopy locally probes the spectral function . the spectral function is of key interest in many - body physics because it provides rich information on interactions@xcite , and is highly sought after in future `` cold - atom tunneling microscope '' experiments@xcite . for our few - body system , the local spectral function describes the quasi - particle wavefunction ( qpwf)@xcite and the discrete coupled - spin spectrum of the dopants . we find that interference of atomic orbitals directly contained in the qpwf allows us to quantify the electron - electron correlations and the entanglement entropy . the entanglement entropy is a fundamental concept for correlated many - body phases@xcite that has thus far evaded measurement for fermions . in the counterintuitive regime of our experiments , entanglement entropy increases as the valence bond is stretched , as coulomb interactions overcome quantum tunneling . in our system , the entanglement entropy is directly related to the hubbard interactions @xmath10 , and we find that @xmath10 is tunable with dopant separation , increasing from @xmath17 for @xmath18 , where @xmath19 nm is the effective bohr radius . this range , of interest to simulate unconventional superconductivity and spin liquids@xcite , is realized here due to the large bohr radii of the hydrogenic states . the semiconductor host allows for electrostatic control of the chemical potential@xcite , desirable to dynamically control filling - factor@xcite but not possible for ions on metal surfaces@xcite . * spectroscopy of coupled - spin system * subsurface boron acceptors in silicon were identified at 4.2 k as individual protrusions@xcite ( density @xmath20 @xmath21 ) in constant current images due to resonant tunneling at a sample bias @xmath22 v , and due to the acceptor ion s influence on the valence density of states at @xmath23 v. the sample was prepared by ultra high vacuum flash annealing at 1200 @xmath24c and hydrogen termination . the observed subsurface acceptors had typical depths@xcite @xmath25 nm , and correspondingly , a volume density @xmath26 times less than the bulk doping , @xmath27 @xmath28 . pairs of nearby acceptors with @xmath29 nm were also found , with a smaller density @xmath30 @xmath21 . the spectrum and spatial tunneling probability of the coupled acceptors were investigated at @xmath31 k via single - hole tunneling from a reservoir in the substrate to the dopant pair , to the tip@xcite ( fig . [ fig1]a ) . for the dopant pair in fig . [ fig1]b ( top ) , @xmath13 measured along the inter - dopant axis ( fig . [ fig1]b , bottom ) contains a peaks for each state entering the bias window , at @xmath32 v , 0.45 v , 0.55 v and 0.8 v. consistent with our single - acceptor@xcite and single - donor@xcite measurements near flat - band bias conditions , the bias for each peak in the spectrum ( fig . [ fig1]b , bottom ) is independent of tip position . this rules out distortion of our quantum state images by inhomogenous tip - induced potentials@xcite observed in other multi - dopant systems@xcite . these results can be attributed to weak electrostatic control by the tip ( fig . [ fig1]c ) and the states proximity to flat - band@xcite , though a large tip radius may also play a role . the spectral and spatially resolved measurements ( fig . [ fig1]b ) directly demonstrate that the holes are interacting , as follows . first , two peaks centred on dopant ions a or b are resolved in real space ( fig . [ fig1]b ) . second , energy differences between the peaks resolved in real space are smaller than the @xmath33 @xmath34ev thermal resolution . however , for orbitals at the same energy to not interact , their overlap must vanish . since the measured orbitals have a strong overlap , the sites are tunnel coupled , irrespective of the details of the tunneling current profile . the number of states observed , their energy differences , and their energies relative to the fermi energy confirm that they observed states are two - hole states ( fig . [ fdidufit ] and [ fsplitting ] ) . + * correlations , entanglement , and hubbard interactions * the ground state of a hubbard model with non - perturbative interactions is governed by @xmath35 in fig . [ fig2]a in the subspace of @xmath36 , @xmath37 , @xmath38 and @xmath39 , where @xmath5 creates a localized electron on site @xmath40 with spin @xmath41 , and @xmath42 is the vacuum state . the ground state is a superposition @xmath43 , where @xmath44 ( @xmath45 ) is the probability amplitude for a covalent ( ionic ) configuration ( fig . [ fig2]b ) . rewriting the state in a basis of even and odd orbitals , @xmath46 , where @xmath47 ( @xmath48 ) is the probability amplitude of the `` even / even '' ( `` odd / odd '' ) configuration . hybridizing singly and doubly occupied configurations , for sites @xmath49 ( red orbital ) and @xmath50 ( blue orbital ) . b. dependence of probability amplitudes on interactions @xmath10 : @xmath44 ( green dashed ) and @xmath45 ( green solid ) for configurations @xmath51 and @xmath52 ) , and @xmath47 and @xmath48 for @xmath53 and @xmath54 respectively . c. entanglement entropy @xmath55 increases with increasing hubbard interactions this occurs because of localization of red and blue orbitals associated with spins in the singlet , as illustrated in the insets . ] in limit of small tunnel couplings ( @xmath56 , fig . [ fig2]b ) the hubbard system may be described by perturbative heisenberg spin interactions . for vanishing @xmath10 , the ground state is a heitler - london singlet of localized spins , @xmath57 , with no contributions from @xmath58 and @xmath59 . due to vanishing wavefunction overlap the electrons can be associated with sites a and b ( they are distinguishable@xcite ) , and the spin at site a depends on the spin at site b as for a maximally entangled bell state . in the limit of vanishing interactions ( @xmath60 , fig . [ fig2]b ) corresponding to a tight - binding approximation , the spins delocalize and @xmath61 . in a molecular orbital ( mo ) basis , the ground state @xmath62 , which is a single slater determinant . although this state is a singlet ( one spin up , one spin down ) due to fundamental indistinguishability , the electrons can be ascribed independent properties because they occupy the same orbital , and the state is uncorrelated@xcite . in the regime of intermediate @xmath10 where tunneling and coulomb interactions compete non - perturbatively@xcite , tunnelling hybridizes the doubly - occupied configurations @xmath58 and @xmath59 into the ground state , such that the particles lose their individual identities . here , the von neumann entanglement entropy quantifies genuine entanglement ( inter - dependency of properties ) , distinguishing it from exchange - correlations due to indistinguishability@xcite . employing the convention@xcite @xmath63 ( 1 ) for zero ( maximal ) entanglement , @xmath64 increases as @xmath10 increases and coherent localization occurs ( fig . [ fig2]c ) , saturating at value of 1 . * * we now discuss the spatial tunneling maps of the two - hole ground states for different inter - acceptor distances . obtained by integrating the lowest voltage @xmath13 peak , the maps are shown in fig . [ fig3]a , [ fig3]b , and [ fig3]c for distances @xmath65 and @xmath66 ( @xmath19 nm ) having orientations @xmath67 from @xmath68 , @xmath69 from @xmath70 and @xmath71 from @xmath68 , respectively . the multi - nm spatial extent of the states reflects the extended wave - like nature of the acceptor - bound holes , owing to their shallow energy levels , which contrasts mn ions on gaas surfaces@xcite , magnetic ions on metals@xcite , and si(001):h dangling bonds@xcite . consequently , their envelopes are amenable to effective - mass analysis with lattice frequencies filtered out@xcite . consistent with measurements of single acceptors at similar depths on resonance at flatband@xcite , the states have predominantly s - like envelopes with slight extension along [ 110 ] directions , as expected when symmetry is not strongly perturbed by the surface . depths of the @xmath72 and @xmath73 pairs were estimated to be @xmath74 nm , and for @xmath75 , @xmath76 nm ( see fig . [ fdepth ] ) . we employed full - configuration interaction calculations of the singlet ground - state @xmath77 to confirm that coulomb correlations of coupled acceptors influence the ground state in a way that mimics the @xmath78 hubbard model . in particular , for @xmath79 , @xmath77 is predominantly composed of @xmath80 , a singlet of two even @xmath81``3/2 '' spin mos . with increasing @xmath82 , interactions enhance the probability amplitude of the @xmath83 singlet with two odd orbitals , analogous to the hubbard hamiltonian ( fig . [ fig2]b ) . the spins @xmath81``3/2 '' are predominantly composed of @xmath84 valence band bloch states . in particular , the low - lying @xmath81``1/2 '' spin excitations of each acceptor@xcite , which are predominantly composed of @xmath85 bloch states , do not qualitatively change the description . we also note that for @xmath86 , the mos are essentially linear combinations atomic orbitals having the effective bohr radii of single acceptors . single - hole tunneling transport through our coupled dopant system locally probes the spectral quasi - particle wavefunction@xcite . when @xmath87 ( fig . [ fig1]a ) , the single - hole tunneling rate is essentially governed by @xmath88 , the tunnel - out rate@xcite . in the present case , single - hole tunneling from the two - hole system to a single - hole final state @xmath89 ( fig . [ fig1 ] ) contributes @xmath90 , where @xmath91 is the qpwf , @xmath92 is the field operator , @xmath93 creates a single - hole mo eigenstate @xmath94 of the system@xcite , and the total tunnel rate is @xmath95 . from the qpwf description of coupled dopants , we obtain a spatial tunneling probability @xmath96 for the ground state . here , @xmath97 and ( @xmath98 ) contain constructive ( destructive ) interference corresponding to even ( odd ) linear combinations of atomic orbitals @xmath99 ( @xmath100 ) ( note : @xmath101 ) . to obtain @xmath98 , data were fit to @xmath102 , assuming linear combinations of parametrized s - like atomic orbitals for @xmath103 and @xmath104 appropriate for subsurface acceptors . the qpwf and atomic orbitals are described in figs . [ ftheoryqpwf ] , [ fsingleboron ] and [ fdoubleboron ] . the least - squares fits in figs . [ fig3]d , [ fig3]e , [ fig3]f ( colored lines ) of @xmath102 are in good agreement with data ( squares ) , for @xmath75 , @xmath105 and @xmath66 . for comparison with the data , grey curves are shown for both the uncorrelated ( maximally correlated ) state with @xmath106 ( @xmath107 ) in fig . [ fig3]d , [ fig3]e , and [ fig3]f . we note that all three separations exhibit interaction effects at the midpoint of the ions , where the quantum interference is strongest . we obtain @xmath108 , @xmath109 , and @xmath110 for @xmath65 and @xmath66 . data taken at higher tip heights gave identical results to within experimental errors ( see fig . [ fcorr ] and [ fivz ] ) , independently verifying that the tip does not influence our results . vs. @xmath82 . theory predictions are shown for coupled acceptors with @xmath68 orientations ( red line ) and @xmath70 ( blue line ) , alongside scaled h@xmath111 ( dashed black line ) . predicted localization is suppressed ( enhanced ) along @xmath68 ( @xmath70 ) relative to molecular hydrogen ( h@xmath111 ) , due to valence band anisotropy , which enhances ( suppresses ) @xmath1 . b. same as ( a ) , for the entanglement entropy @xmath55 . c. experimentally estimated hubbard interactions . error bars denote 95 % confidence intervals . ] the coulomb correlations , embodied both in @xmath112 ( fig . [ fig4]a ) and the entanglement entropy @xmath64 ( fig . [ fig4]b ) , could be evaluated directly from the fit , and both increase with increasing @xmath82 . the one - to - one mapping from @xmath55 to @xmath10 ( fig . [ fig2]c ) was used to determine the effective hubbard interactions from the entanglement entropy in fig . we obtain @xmath113 , 6.4 , and 14 , for @xmath75 , @xmath105 , and @xmath66 respectively ( fig . [ fig4]c ) , which increase as the tunnel coupling decreases . we conclude the analysis of the qpwfs with some critical remarks on correlations extracted from our fitting model , recalling that the large spatial overlap of the spectrally overlapping acceptor - bound holes directly shows their states are tunnel coupled . first , the coulomb correlations have a systematic effect on interference in the qpwf such that the least - squares error is significantly worse if @xmath98 is forced to zero in the fitting model ( table [ tsse ] ) . second , if applied to very far apart dopants where the ground state can still be resolved , our fitting model would not give a spurious result that the two dopants are highly correlated . this follows because the difference between @xmath114 and @xmath115 , which reflects the interference of atomic orbitals and is used to detect correlations , tends to zero as @xmath116 increases . data ( fig . [ fig3]a - c ) presented here are for coupled dopants that we found to be ( i ) well isolated from other dopants or dangling bonds , and ( ii ) at identical depths , as evidenced by the spatial extent and brightness of the atomic orbitals . when the latter is not satisfied , the atomic levels can be detuned , introducing more parameters to the fit . + * comparison with theory * these experimental results obey the trends predicted by our theory calculations for the spin - orbit coupled valence band . predictions in fig . [ fig4]a and [ fig4]b for displacements along @xmath70 ( blue solid line ) and @xmath68 ( red solid line ) both show increasing correlations and entanglement with increasing dopant separation . moreover , we find that the observed and predicted entanglement entropy qualitatively reproduce a single - band model ( fig . [ fig4]a , [ fig4]b , dashed lines ) . this result implies that inter - hole hubbard interactions follow an essentially hydrogenic trend with atomic separation , even for non - perturbative interactions @xmath117 . the hydrogenic nature of @xmath55 and @xmath10 persists in spite of the @xmath81``1/2 '' spin excited states of a single acceptors . such @xmath81``1/2 '' single - acceptor excited states states are found nominally @xmath118 mev above the @xmath81``3/2 '' spin ground state due to inversion symmetry breaking at the interface@xcite . although @xmath119 , @xmath55 and @xmath10 remain hydrogenic in our calculations because the `` 1/2 '' spin excited state has an s - like envelope whose spatial extent is similar to ( 1 ) the s - like @xmath81``3/2 '' ground state and ( 2 ) the scaled hydrogenic ground state . otherwise , single particle @xmath81``1/2 '' states would hybridize stronger than single particle @xmath81``3/2 '' states , form the 2-hole singlet at smaller separations , and localize more slowly relative to molecular hydrogen with increasing @xmath82 . furthermore , the polarization of the @xmath81``3/2 '' and @xmath81``1/2 '' states into @xmath120 and @xmath85 components , respectively , limits the mixing of @xmath81``1/2 '' states into the ground state . + * spin excited states and effective temperature * finally , we discuss the observed excited states , which confirm that the inter - acceptor tunnel - coupling dominates thermal and tunnel - coupling effects of the reservoir . the energies of the states were determined by fitting the single - hole transport lineshapes@xcite of the coupled acceptors ( fig . [ fdidufit ] and [ fsplitting ] ) . for the first excited state we found @xmath121 mev and @xmath122 mev for @xmath75 and @xmath66 respectively ( @xmath123 orientation ) , and @xmath124 mev for @xmath72 ( @xmath125 orientation ) . shown in fig . [ fig5]a , these energies are too small to add another hole , which would require @xmath126 mev for an acceptor in bulk silicon . however , the energies agree well with our predictions for two - hole excited states of coupled hole spins @xmath127 and @xmath128 , _ i.e. _ , @xmath129 mev and @xmath130 mev for @xmath131 and @xmath132 ( @xmath133 orientation ) , and @xmath134 mev ( @xmath135 orientation ) . here we note that some of the predicted coupled - spin excited states ( fig . [ fig5]b ) are unconventional : a singlet @xmath136 and triplet @xmath137 of two `` 3/2 '' holes ( orange lines ) and two `` 1/2 '' holes ( black lines ) , where @xmath138 is the ground state for all separations . more subtly , two manifolds @xmath139 , @xmath140 , @xmath141 , containing four states are predicted ( green lines ) , where one @xmath81``3/2 '' spin level and one @xmath81``1/2 '' spin level is occupied . for @xmath75 and @xmath105 ( @xmath73 ) , the measured energies are in better agreement with predictions for @xmath139 ( @xmath142 ) excitations . . singlets @xmath143 and triplets @xmath144 are present for interactions between two holes of @xmath145``3/2 '' spin ( orange ) and two holes of @xmath145``1/2 '' ( black ) spin . states @xmath146 and @xmath147 are sets of four closely spaced levels ( green ) with one 3/2 spin hole , and one `` 1/2 '' spin hole . error bars denote 95 % confidence intervals . ] the inter - acceptor tunnel couplings @xmath1 ( ratios @xmath148 ) were estimated to be 12 mev ( 30 ) , 7 mev ( 20 ) , and 3.5 mev ( 10 ) for @xmath65 and @xmath66 respectively , at @xmath31 k. such couplings @xmath1 exceed the reservoir coupling @xmath149 ( table [ tcouplingtable ] ) to the substrate by more than 50x . combined with bias @xmath150 v needed to bring the level into resonance , this rules out coherent interactions with substrate and tip reservoirs@xcite . note that the measured energy splittings imply small thermal excited - state populations of @xmath151 , @xmath152 , and @xmath153 for @xmath65 and @xmath66 respectively . we performed atomic resolution measurements resolving spin - spin interactions of interacting dopants , realizing quantum simulation of a two - site hubbard system . analyzing these local measurements of the spectral function@xcite , we find increasing coulomb correlations and entanglement entropy as the system is `` stretched''@xcite in the regime of non - perturbative interaction strengths @xmath10 . our experiment is the first to combine low effective temperatures @xmath154 at @xmath155 k and single - site measurement resolution , considered essential@xcite to simulate emergent hubbard phenomena@xcite . lower effective temperatures @xmath156 are possible at @xmath157 k. for example , @xmath158 hubbard lattices with @xmath159 and @xmath160 have recently been associated with both the pairing state and pseudogap in systems exhibiting unconventional superconductivity@xcite . the approach generalizes to donors , which can be placed in silicon with atomic - scale precision@xcite and spatially measured _ in - situ _ after epitaxial encapsulation@xcite . in contrast to disordered systems@xcite , atomically engineered dopant lattices will require weak coupling to a reservoir , displaced either vertically as demonstrated herein , or a laterally@xcite . strain could be used to further enhance the splitting between light and heavy holes , or suppress valley interference processes of electrons@xcite . interestingly , open hubbard systems which may exhibit unusual kondo behaviour@xcite could also be studied by this method . the demonstrated measurement of spectral functions could be used to directly determine excitation spectra , evaluate correlation functions@xcite , or obtain quasi - particle interference spectra@xcite , all of which contain rich information about many - body states , including charge - ordering effects . we envision _ in - situ _ control of filling factor@xcite , using a back - gate or patterned side - gate@xcite . these capabilities will allow for quantum simulation of chains , ladders , or lattices@xcite at low effective temperatures , having interactions that are engineered atom - by - atom . we thank h. wiseman , m. a. eriksson , m. s. fuhrer , o. sushkov , d. culcer , j .- s . caux , b. reulet , g. sawatzky , j. folk , f. remacle , m. klymenko and b. voisin for helpful discussions . this work was supported by the european commission future and emerging technologies proactive project multi ( 317707 ) , the arc centre of excellence for quantum computation and communication technology ( ce110001027 ) , and in part by the u.s . army research office ( w911nf-08 - 1 - 0527 ) and arc discovery project ( dp120101825 ) . s.r . acknowledges a future fellowship ( ft100100589 ) . m.y.s . acknowledges a laureate fellowship . the authors declare no competing financial interests . experiments were conceived by j.s , j.a.m , and s.r . j.s . carried out the experiments and analysis , with input from j.a.m . , r.r . , l.c.l.h , and s.r . theory modeling was carried out by j.s . , r.r . , and l.c.l.h and s.r . , with input from all authors . j.s . and s.r . wrote the manuscript with input from all authors . + + @xmath161 [email protected],@xmath162 [email protected] * sample preparation * samples were prepared by flash annealing a boron doped ( @xmath163 @xmath28 ) silicon wafer at @xmath164 @xmath24c in uhv followed by slow cooling at a rate @xmath165 @xmath24c@xmath166min@xmath167 to 340 @xmath24c . then , hydrogen passivation was carried out @xmath168 @xmath24c for ten minutes by thermally cracking h@xmath111 gas at a pressure @xmath169 mbar . + * measurements * atomic resolution single - hole tunneling spectroscopy was performed at 4.2 k using an ultra - high vacuum omicron low temperature scanning tunneling microscope ( lt - stm ) . current @xmath170 was measured as a function of sample bias @xmath171 and @xmath13 was obtained by numerical differentiation . details for the analysis of the data are provided in figs . [ fdidufit ] , [ fsplitting ] , [ fdepth ] , [ fsingleboron ] , [ fdoubleboron],[fcorr],and [ fivz ] , and appendices [ smeasuredspectrum],[sdepth],[sspatialmodel ] and [ stipheightspectra ] . + * theory * theory calculations of interacting states were carried out using the configuration interaction approach , in the luttinger - kohn representation including a realistic description of the heavy - hole ( @xmath172,@xmath173 ) , light - hole ( @xmath172,@xmath174 ) , and split - off hole ( @xmath175,@xmath174 ) degrees of freedom . details for the theory are provided in fig . [ ftheoryqpwf ] and appendices [ stheory],[shubbard ] and [ sentanglement ] . vs. @xmath171 for the @xmath75 , above ( black squares ) and away from ( orange squares ) the coupled acceptors . least square fits ( green line ) considering sequential hole transport through several peaks ( green dashed lines ) . b. same as a on logarithmic scale . c. tunneling of electrons from valence band ( v.b . ) to the tip . d. tunneling of electrons from the tip to the conduction band ( c.b . ) . schematic of hole tunneling at positive sample bias @xmath171 , from valence impurity band reservoir , to acceptor pair , to tip . ground and excited states for two - hole occupation are shown . electrons ( holes ) in the hole reservoir and stm tip are shaded blue ( white ) . ] in this section we discuss the extraction of the energies of the states observed in @xmath13 tunneling spectra . we find that the the number of observed states , the state s relative energies to eachother , and their energies relative to the sample s fermi energy , are only compatible with two interacting holes on pairs of acceptors . we further estimate the total tunnel coupling @xmath176 to the tip and sample reservoirs for sequential hole transport , based on the @xmath13 lineshape , and show that @xmath177 is much less than the inter - acceptor tunnel coupling @xmath1 for separations @xmath82 considered in the main text . for the acceptor pair with @xmath75 ( fig . [ fig3]a , main text ) , measured @xmath13 is plotted in linear scale ( fig . [ fdidufit]a ) and log scale ( fig . [ fdidufit]b ) . away from the acceptor pair ( fig . [ fdidufit]a / b , orange squares ) , direct tunneling of electrons from the valence band to the tip ( holes from the tip to the valence band ) is obtained for @xmath178 v as illustrated in fig . [ fdidufit]c , while direct tunneling of electrons from the tip to the conduction band occurs for @xmath179 v as illustrated in fig . [ fdidufit]d . above the pair ( fig . [ fdidufit]a / b , black squares ) , we observe a well - isolated peak at @xmath180 v , a collection of two strong ( closely spaced ) peaks at @xmath181 v and @xmath182 v , and fourth peak at @xmath183 v. plotted on a logarithmic scale , both the current ( not shown ) and numerically differentiated @xmath13 data ( fig . [ fdidufit]b ) , black squares ) grow exponentially with increasing @xmath171 for the first peak before reaching a local maximum . as suggested by this observation@xcite , we fit the first peak of the differential conductance to a state probed by sequential tunneling with broadening determined by the @xmath31 k temperature of the reservoir , @xmath184 , where @xmath185 c is the elementary charge , @xmath186 j / k is the boltzmann constant , @xmath187 is the peak voltage , and @xmath188 is the well - known lever arm from sequential transport@xcite describing the variation in the energy level @xmath189 due to a change in applied bias @xmath190 , as schematically illustrated in fig . [ fdidufit]e . the least - squares fit is in good agreement with the measured @xmath13 for both the full - width at half maximum @xmath191 and the exponential decay of the tail @xmath192 . a peak voltage @xmath193 mev and a lever arm @xmath194 were obtained for tunneling of holes from the reservoir , to the acceptor pair , to the tip . the value for @xmath188 is similar to those found for single - hole transport through individual acceptors in flashed p - type silicon@xcite and single arsenic donors in flashed n - type silicon@xcite . this excellent agreement shows that broadening of the single - hole transport peaks is due to the temperature associated with the fermi - dirac distribution in the hole reservoir . the small value of @xmath188 means that the `` broadened '' lineshape in fig . [ fdidufit]a and fig . [ fdidufit]b is due to weak electrostatic control of the sample potential by the stm tip , as expected due to strong screening by the carrier reservoir . this is further confirmed by the negligible `` spectral shift''@xcite , that is , the similarity of the apparent gap @xmath195 ev in @xmath13 measurements of fig . [ fdidufit]a / b to the actual @xmath196 ev band gap for heavily - doped p - type silicon at low temperature@xcite . of observed states relative to the lowest energy state ( energy @xmath197 ) , for @xmath75 , @xmath72 , and @xmath73 . for comparison , the energies of the predominantly @xmath81``3/2 '' and @xmath81``1/2 '' spin kramers doublets of a lone acceptor , @xmath198 nm from the si[001]:h interface , are shown on the right hand side . ] we extracted the lever arm @xmath188 and energies of states @xmath199 , by least - squares fitting of @xmath13 to a sum of single - hole tunneling peaks@xcite . fits to @xmath200 are found in fig . [ fdidufit]a and [ fdidufit]b ( solid green lines ) along with individual peaks ( dashed green lines ) , on linear and log scales , respectively . extracted energies @xmath201 of the excited states are plotted in fig . [ fsplitting ] , relative to the energy @xmath197 of the two - hole ground state , for @xmath75 , @xmath72 and @xmath73 . for all @xmath82 the extracted energy splittings between the states ( fig . [ fsplitting ] ) are too small to be associated with single - hole occupied states ( a@xmath202 states ) of coupled acceptors . for @xmath75 , two energetically nearby states were found at @xmath203 and @xmath204 mev , while a higher energy state was found at @xmath205 mev above those . here , the observed splittings are incompatible with the `` 1s '' hybridization energy @xmath206 mev between tunnel - hybridized single - hole states estimated by numerics ( section [ stheory ] ) . similar arguments hold for the @xmath72 ( @xmath73 ) , where the excited state splittings @xmath207 mev ( @xmath208 mev ) are incompatible with the single - hole coupling energy scale of @xmath209 mev ( @xmath210 mev ) . even more importantly , it follows from the electrostatic arguments given below that the coupled - acceptor peaks observed in experiments are too close to the hole reservoir s chemical potential to correspond to ionizations of the very `` deep '' single - hole ( a@xmath211 ) states into zero - hole ( a@xmath212 ) states . rather , they are only compatible with ionizations of the two - hole ( a@xmath213 ) ground and excited states of acceptor pairs into one - hole ( a@xmath211 ) states , which are energetically very similar to the ionization energy of a neutral ( a@xmath214 ) acceptor . for @xmath215 we found ionization transitions a@xmath216a@xmath212 for the last hole ( section [ stheory ] ) require more than @xmath217 mev energy . this exceeds the fermi energy @xmath218 mev of the reservoir ( @xmath219 is the valence band edge ) by more than @xmath220 mev . now , for @xmath221 v , the tip s contact potential bends states by an amount @xmath222 relative to the sample reservoir , where @xmath223 ev is the work function of degenerately doped p - type reservoir , and @xmath224 is tungsten tip workfunction . let us assume a work function @xmath225 ev typical of a tungsten tip@xcite . then , the states bend down by @xmath226 mev at zero bias , much smaller than the value @xmath227 mev required to depopulate a single hole a@xmath202 state ( a@xmath216a@xmath212 transition ) . therefore at zero bias , the observed acceptor pairs are in an a@xmath202 charge state , and a bias @xmath228 mv much less than @xmath229 v would be required to remove the last hole by tip - induced band bending . this is incompatible with @xmath230 to @xmath231 v of peaks observed in experiments . on the other hand , ionization energies of the two - hole states , a@xmath232a@xmath202 of coupled acceptors are calculated ( section [ stheory ] ) to be very energetically similar to a@xmath233 ionization energies . consequently , transitions populating ground and excited two - hole will occur for @xmath234 , explaining how they ( much like single - acceptor ionizing resonances@xcite ) are observed in our experiments . we have used the @xmath13 lineshape to obtain an upper bound on the tunnel coupling @xmath244 to the environmental reservoirs - the largest lifetime broadening energy scale@xcite @xmath177 that can be included in the lineshape analysis of the data@xcite . as necessary to observe molecular states , we find @xmath176 for data in fig . [ fig3]a , [ fig3]b , and [ fig3]c , to be considerably smaller than the inter - acceptor tunnel coupling @xmath1 estimated from the hybridization energies ( table [ tcouplingtable ] ) . tip height - dependent measurements ( section [ stipheightspectra ] ) confirmed the exponential dependence of the tunnel current on tip height , indicating the rate to the tip is the slower ( dominant ) rate . the dwell - time ( and coupling @xmath176 ) , determined by the fast rate , is therefore controlled by the coupling to the hole reservoir in the substrate , @xmath149 in fig . [ fdidufit]e . we have estimated the depth below the silicon surface of the acceptor dopant pairs , presented in fig . [ fig3]a , [ fig3]b and [ fig3]c of the main text , to be approximately @xmath245 nm , @xmath246 nm , and @xmath246 nm , respectively . the depth of isolated dopants in scanning tunneling spectroscopy is typically estimated fitting the spatial variation of a spectral feature such as a band edge , in a bias condition where the dopant is ionized@xcite , to a dielectric screened coulomb potential . however , as illustrated in fig . [ fdidufit]e , tunneling from the reservoir to the acceptor pair is a transition from a one - hole state to a two - hole state . consequently , for voltages @xmath171 below the lowest voltage peak at @xmath247 in fig . [ fdidufit]a and fig . [ fdidufit]b , the two - acceptor system is in coulomb blockade with single - hole occupation , rather than being doubly ionized , such that fitting to dielectric - screened single - ion potentials is inappropriate@xcite . for @xmath171 above @xmath187 the two - acceptor system has a high probability of occupation by two holes since @xmath248 . hence , the method of fitting the band profile to dielectric screened ion potentials used in the above references is problematic . to estimate the depth of our acceptors , we use the well - known result that the spatial extent of the bound state measured by sts increases with increasing dopant depth ( see _ e.g. _ , reference ) . we compare the full - width at half - maximum @xmath249 of the spatial tunneling distribution fit in section [ sspatialmodel ] to the same quantity calculated for subsurface acceptor - bound holes in the @xmath250 spin - orbit coupled valence band . we provide this estimate for completeness , since the fitting procedure described in section [ sspatialmodel ] does not rely on prior knowledge of the depth or bohr radius of the dopant . the full - width at half maximum @xmath249 for the 1s orbital density @xmath251 at @xmath252 can be directly solved giving @xmath253 . evaluating this quantity we obtain @xmath254 nm , @xmath255 nm , and @xmath256 nm for the measured molecules with @xmath75 , @xmath105 , and @xmath66 , respectively . comparing these results to the full - width at half maximum @xmath249 predicted for the orbitals using the numerical solution to kohn - luttinger @xmath250 hamiltonian presented in fig . [ fdepth ] , the @xmath249 extracted from experiments translate to depth estimates of @xmath257 nm , @xmath258 nm , and @xmath258 nm , respectively . when the acceptors are located deeper in the sample , the lever arms are anticipated to be smaller@xcite . the relative depths of the planes in which the acceptor pairs are estimated to be located agree with the extracted lever arms . we obtain a lever arm @xmath194 for the shallower @xmath75 pair . the other two @xmath72 and @xmath73 pairs which are slightly deeper have lever arms @xmath259 and @xmath260 , respectively . we note that fluctuations in the depth of the reservoir can also influence the lever arm . we anticipate that varying the dopant depths has a small effect on correlations in our experiments where the overall system is neutral . we have observed that the neutral level of near - surface acceptors ( @xmath261 ) remains bulk - like@xcite for depths 0.5 - 2.0 nm , as expected , due to a competition between the confinement and the dielectric mismatch . fixed ionization energy essentially fixes the bohr radius , fixing @xmath1 and @xmath262 . we note however that the apparent spatial extent of the wavefunction increases with increasing depth . this is not because of a change in the physical bohr radius . rather , it occurs because stm probes the wavefunction near the surface . we employed a microscopic configuration interaction framework to calculate eigenstates of interacting holes bound to proximate acceptors , in a kohn - luttinger @xmath250 spin - orbit coupled representation ( heavy holes , light holes , and split - off holes ) , and in a simplified single - band representation for reference . in both representations , the two - hole ground state at relatively small @xmath263 was found to be predominantly composed of the @xmath264 singlet of two even orbitals , where @xmath265 and @xmath266 denote + `` 3/2 '' and @xmath266``3/2 '' respectively for the valence band holes , and denote @xmath267 and @xmath268 respectively in the single - band approximation . the probability amplitude of the @xmath269 singlet of odd orbitals was found to increase with increasing @xmath82 , signaling interaction - driven generation of quantum correlations . the corresponding spatial tunneling probability distribution was predicted using the quasi - particle wavefunction ( qpwf ) density appropriate for few - particle states with quantum correlations@xcite , and are presented herein for the non - trivial @xmath250 spin - orbit coupled case . in the absence of spin - orbit coupling , wavefunctions of interacting spins can be classified into singlets and triplets@xcite . for holes in the valence band , the atomic orbitals have p - like symmetry , and the angular momentum @xmath270 couples to the spin ( intrinsic ) angular momentum @xmath78 , such that the resulting bloch states have definite @xmath271 and @xmath272 . while intrinsic angular momentum is not a good quantum number , spin can nevertheless be generalized to hole pseudospin , _ i.e. _ , kramers doublets of time reversal symmetric linear combinations of bloch states@xcite . coupling of hole spins is well understood in the kohn - luttinger framework@xcite . it has been theoretically shown that coupled hole pseudospin interact to produce singlets and triplets@xcite , as for coupled electron spins . in the kohn - luttinger approach generalized valence band holes spins as written as spinors @xmath273 , where @xmath274 are the envelope functions corresponding to the bloch states @xmath275 . due to spin - orbit coupling such spinors are @xmath276 % polarized ( rather than 100 % polarized ) into single bloch components . this produces a small band - mixing effect when the pseudospins are coupled together@xcite , that we take into account exactly in our numerics . we employ the kohn - luttinger fci approach developed for valence band holes in reference [ ] . two - hole eigenstates @xmath277 are expressed as superpositions of antisymmetrized two - hole slater determinants @xmath278 with probability amplitudes @xmath279 , @xmath280 in @xmath281 above , @xmath282 creates a single - hole state @xmath283 , where @xmath284 is the vacuum state . as discussed in detail below , the single - particle states @xmath285 were chosen as the eigenstates of the non - interacting hamiltonian @xmath286 of a pair of acceptor ion potentials and a hard - wall interface with dielectric mismatch . the hamiltonian for the two - hole system was taken as @xmath287 where @xmath288 is the coulomb repulsion term for the holes . substituting @xmath277 above into @xmath289 and taking the inner product with @xmath290 , we obtain for each @xmath293 , where @xmath294 is an off - diagonal ( interaction ) term evaluated in the spinor representation , @xmath295 is the direct coulomb interaction , and @xmath296 is a direct exchange interaction . finite @xmath297 terms can introduce correlations@xcite by creating eigenstates that are admixtures of two - particle slater determinants @xmath298 . the single particle hamiltonian employed was @xmath299 taking into account ion potentials @xmath300 of both acceptors , image charges @xmath301 due to dielectric mismatch@xcite , and an infinite ( hard - wall ) @xmath302 potential a distance @xmath303 nm away from the ion representing the semiconductor / vacuum interface . for @xmath304 we considered both the @xmath250 spin - orbit coupled kohn - luttinger valence band@xcite @xmath305 , as well as a single - band , isotropic , parabolic dispersion relation @xmath306 where @xmath307 is an effective mass chosen to reproduce boron s binding energy@xcite . the latter allows us to calculate states of a simple `` scaled hydrogen molecule '' without complex spin - orbit coupling and band structure effects . the ion potentials were taken as dielectric - screened coulomb potentials . for the @xmath250 scheme we included additional charge @xmath308 at the ion site as a crude core - correction to obtain the correct spatial extent @xmath309 and ionization energy of the states@xcite . the interaction term @xmath310 is a dielectric screened coulomb interaction including an image charge distribution produced by dielectric mismatch@xcite . for the single band theory , the four lowest single - electron eigenstates of @xmath286 were used to construct 6 possible two - particle configurations , all of which were included for a full configuration interaction ( fci ) approach . we considered two even , spin - degenerate orbitals @xmath311 , and two odd , spin - degenerate orbitals @xmath312 . for the @xmath250 kohn - luttinger calculations , we use a basis that reflects the four low - energy states of single acceptors with s - like envelopes - the predominantly @xmath173 ( `` 3/2 '' ) state , and the predominantly @xmath174 ( `` 1/2 '' ) state@xcite . near an interface , the `` 1/2 '' state experiences more confinement than the `` 3/2 '' state , making the `` 3/2 '' state the ground state@xcite . for lone subsurface acceptors @xmath198 nm from an si[001]:h interface , we have typically measured @xmath313 mev splittings ( see _ e.g. _ , lone acceptor data in fig . [ fsplitting ] ) , in good agreement with few mev values predicted by @xmath250 kohn - luttinger . see reference [ ] for further details . in the two - acceptor potential @xmath286 these hybridize into eight single - hole eigenstates , which were used to construct 28 possible two - hole configurations , all of which were included in the state vector for a fci approach . for the acceptor distances in the main text , four were predominantly composed of heavy holes ( `` 3/2 '' states ) and four were predominantly composed of light holes ( `` 1/2 '' states ) . in both groups of four states , two are found within a kramers doublet with even parity symmetry for the majority spin , and two are found within a kramers doublet having odd parity symmetry for the majority spin@xcite . a numerical representation of the fci matrix ( equation [ eq : ci ] ) was directly diagonalized to obtain state vectors @xmath314 and energies @xmath315 . the fci matrix was obtained by evaluating the coulomb matrix elements @xmath316 , exchange matrix elements @xmath317 , and the interaction terms @xmath297 by monte - carlo integration , using the vegas method for adaptive sampling and @xmath318 iterations per integral . the basis of single particle eigenstates required for the integrals was obtained numerically using the finite difference scheme , for both the single - band and @xmath250 representations , on a real - space grid of @xmath319 points with a discretization step size of @xmath320 nm , using luttinger parameters for silicon s valence band@xcite . to normalize the measured and theoretical ( @xmath250 and single - band ) inter - acceptor distances , we choose the value @xmath19 nm reproducing the @xmath321 mev binding energy of a boron acceptor in silicon@xcite in a single - band approximation . in this section , the ground state of the fci hamiltonian in section [ stheorymethodology ] is discussed . our fci predicts a singlet ground state for both the spin - orbit coupled @xmath250 representation and for the isotropic single - band case . in a single - band representation ( ignoring spin - orbit coupling ) , the ground state @xmath322 was obtained . with increasing @xmath82 , dynamical interactions ( @xmath323 s in equation [ eq : ci ] ) resulted in increasing @xmath48 and increasing quantum correlations @xmath324 , as predicted elsewhere@xcite . the ground state of interacting , spin - orbit coupled holes@xcite in the @xmath250 kohn - luttinger representation contains contributions for which majority pseudospin have even parity , @xmath325 and contributions for which for majority pseudospin have odd parity , @xmath326 for small @xmath263 , the dominant configuration ( approx . @xmath327 % ) in the ground state singlet was found to be @xmath328 , while for increasing @xmath82 , the probability amplitude @xmath329 was found to increase . from this , two conclusions can be drawn : ( 1 ) the increase in the probability amplitude @xmath329 with increasing @xmath82 signals the emergence of hole - hole quantum correlations@xcite , and ( 2 ) the predominance of `` 3/2 '' pseudospinspin orbitals signals that the band - mixing effect is weak for the values of @xmath82 in experiments . contributions @xmath330 and @xmath331 with @xmath1453/2 and @xmath3323/2 are much larger than the other `` band - mixing '' terms with @xmath1451/2 or @xmath332``1/2 '' . in this section , we discuss theoretical predictions of the correlation coefficient @xmath333 in fig . [ fig4]a , as well as theoretical calculation of the quantity measured for spatially resolved single - hole tunneling into multi - particle states with correlations , the qpwf density@xcite . the results of section [ stheorygroundstate ] are considered , for both the single - band and spin - orbit coupled @xmath250 kohn - luttinger representations . as described in section [ smeasuredspectrum ] , the tunneling probability density measured in experiments is determined by tunneling of a hole from the acceptor molecule to the tip , leaving behind a single - hole eigenstate @xmath334 on the acceptor molecule . the corresponding qpwf density @xmath335 is obtained@xcite , where @xmath336 , @xmath337 is the field operator , @xmath338 is position of the tip apex orbital , and @xmath339 is the ground state . for the single - hole transport processes probed , it is necessary to sum over the final ( single - hole ) states @xmath340 of the molecule after hole tunneling to the tip . the total tunneling probability density is then @xmath341 . for the @xmath250 spin - orbit coupled representation , the quasi - particle wavefunction densities are also readily obtained , such that @xmath344 , where @xmath345 $ ] . evaluating this expression results in a qpwf density that is a sum of even and odd qpwf contributions for all @xmath82 and both orientations @xmath70 and @xmath68 , as expected considering the generalized parity of holes@xcite . results for @xmath333 , again defined as twice the probability density of odd orbitals , are plotted in the main text ( fig . [ fig4]a , coloured lines ) for both @xmath70 and @xmath68 orientations . the predicted spatial tunneling probability density is shown @xmath165 nm above the ion in fig . [ ftheoryqpwf ] for an acceptor molecule with @xmath346 . the predicted correlation coefficient is @xmath347 , similar to @xmath348 extracted from measurements of the @xmath132 acceptor molecule , in fig . [ fig4]a of the main text . contributions to the spatial maps ( line profiles ) from even and odd qpwf densities are separately plotted in fig . [ ftheoryqpwf]a ( fig . [ ftheoryqpwf]d ) and fig . [ ftheoryqpwf]b ( fig . [ ftheoryqpwf]e ) , respectively . their sum , the total tunneling probability density ( line profile ) , is plotted in fig . [ ftheoryqpwf]c ( fig . [ ftheoryqpwf]f ) . upper and lower grey lines in fig . [ ftheoryqpwf]f denote normalized qpwf densities for the even qpwf and odd qpwf components , respectively , which in this case directly correspond to the profiles in fig . [ ftheoryqpwf]d and [ ftheoryqpwf]e , respectively . the experimentally obtained spatial tunneling probabilities for the two - hole ground state of acceptors , in fig . [ fig3]a , [ fig3]b , and [ fig3]c of the main text , were fit assuming even and odd linear combinations of atomic orbitals with s - like envelopes for @xmath349 and @xmath350 . this data was obtained using a tip height established by constant current imaging conditions , at a bias where the topography was nominally flat apart from dimer row corrugations . in this section we describe the fits and show that the probability density of an s - like wavefunction describes the measured probability density of the ground state and first excited state of an isolated acceptor , in agreement with what is expected for the s - like `` 3/2 '' pseudospin ground state and s - like `` 1/2 '' pseudospin excited state . for single acceptor ground state . b. integrated @xmath13 for first excited state , @xmath351 mev above the ground state . c. same as a with lattice frequencies filtered out . d. same as b with lattice frequencies filtered out . line profile of relatively tunneling probability parallel ( black squares ) and perpendicular ( green squares , offset for clarity ) to the dimer , and fit of profile parallel ( black line ) and perpendicular ( green line , offset for clarity ) to dimer direction . f. same as e , but for first excited state . scale bar : 1 nm ] spatially resolved tunneling spectra were measured for isolated acceptors , consistently revealing two @xmath13 peaks associated with the s - like @xmath172 ground state manifold . the @xmath352 mev splitting of the isolated acceptor in fig . [ fsplitting ] was found to be in good agreement with single - particle eigenstates calculated from our numerical kohn luttinger solver for depths determined as in previous work@xcite , that is , by a fitting the profile to the ionized acceptor s perturbation to the band edge@xcite . the spatial tunneling probability distributions for the lowest energy state and first excited state of an isolated acceptor were obtained by integrating the two low energy @xmath13 peaks . results are shown in fig . [ fsingleboron]a and fig . [ fsingleboron]b for the single acceptor in fig . [ fsplitting ] . the dimer lattice structure of the hydrogen - terminated @xmath353 surface is evident in the states fig . [ fsingleboron]a and fig . [ fsingleboron]b , running along the @xmath354 $ ] direction indicated . the dimer and lattice frequencies were fourier filtered out of the image@xcite , after which only the characteristic envelope of the probability density of the states remain . results for the ground state and first excited state are presented in fig . [ fsingleboron]c and [ fsingleboron]d , respectively . the tunneling probability density reflects the hole density , and the appearance of two s - like envelopes is in agreement with expectations of the s - like envelopes for the `` 3/2 '' and `` 1/2 '' states@xcite . we observe that the measured low - frequency envelope for the probability density is closer to isotropic for the ground state , and spatially elongated along the dimer direction for the excited state . the probability density was fit along a line in the plane ( @xmath252 ) of the surface of the form @xmath355 , where @xmath251 is a 1s orbital envelope function . in this expression , @xmath356 is the tip position , @xmath357 , @xmath358 and @xmath359 are centre coordinates of the orbital , @xmath360 is an effective depth , and @xmath361 is an effective bohr radius . the profile of the measured ground state ( solid squares ) and least - square fit ( solid line ) are shown for a profile along the direction parallel ( black ) and perpendicular ( green ) to the dimer row in fig . [ fsingleboron]e , demonstrating excellent agreement . the same quantities are plotted for the first excited state in fig . [ fsingleboron]f . a more noticeable anisotropy of the excited `` 1/2 '' state was found . motivated by the theoretical observation that the qpwf density partitions into contributions from even and odd states , we categorize the measured qpwf density into contributions from even and odd qpwfs , that is , linear combinations of atomic orbitals with even and odd parity , respectively . for a distance @xmath362 between acceptors , we use @xmath363 $ ] and @xmath364 $ ] , where @xmath365 and @xmath366 in atomic units . in the fits presented in the main text , @xmath361 , @xmath367 , @xmath368 , and @xmath369 are free parameters , while @xmath370 is simply a consequence of normalization . effective bohr radii @xmath371 nm and depths @xmath372 nm were obtained from the least squares fits . in our fits the values for @xmath361 and @xmath367 define a full - width at half maximum @xmath249 for the spatial tunneling probability of each acceptor . in section [ sdepth ] , the actual depth @xmath373 of the acceptors was estimated using @xmath249 derived from experimentally fit values @xmath361 , @xmath367 . the sum of square errors @xmath374 was used to determine the importance of the coulomb correlation parameter @xmath48 in the fits . the first column ( fig # 1 ) of table [ tsse ] gives the results for the @xmath236 when @xmath375 is a free parameter , while the second column ( fig # 2 ) of table [ tsse ] gives the results for @xmath236 when @xmath375 is artificially forced to zero . here we see the fit is considerably improved by including the coulomb correlation parameter in the fit . due to the very small lever arm @xmath376 , the data is acquired very near flat - band@xcite , and as such we do not observe the so - called `` ionization parabolas '' ( observed elsewhere and associated with tip - induced band bending@xcite ) for single b : si acceptors@xcite , or for b : si acceptor pairs ( fig . 1b , main text ) . . b. same as a , but for first excited state . c. spatial measurement of ground state for @xmath377 . d. same as c , but for first excited state . @xmath378 crystal directions directions are as indicated . scale bar : 1 nm ] to provide an independent check that the tip does not influence the results of the qpwf measurement , we present measurements on the @xmath75 and @xmath73 pairs from the main text at a tip height @xmath379 pm above the position where the data was taken in the main text . the normalized tunneling probability is shown in fig . [ fcorr]a for @xmath75 and in fig . [ fcorr]b for @xmath73 . line profiles of the normalized tunneling probability ( squares , fig . [ fcorr]c and fig . [ fcorr]d ) , are in excellent agreement with the same qpwf model ( solid red line , fig . [ fcorr]c and fig . [ fcorr]d ) used in the main text . reference curves are shown for the totally uncorrelated state ( @xmath380 ) and maximally correlated state ( @xmath381 ) . the extracted results for the correlations are @xmath382 for @xmath75 and @xmath383 for @xmath73 . these results for @xmath379 pm tip heights are within the experimental errors of the results for the data in the main text , which are @xmath384 for @xmath75 and @xmath385 for @xmath73 . for any given experimental signal - to - noise ratio , there always exists a large enough @xmath116 where it is impossible to distinguish the existence of correlations in the ground state qpwf . this follows from the fact that our model determines correlations based on the contrast between orbitals @xmath114 and @xmath115 . however , since @xmath386 , where @xmath387 are atomic orbitals at sites @xmath49 and @xmath50 , and the product @xmath388 is exponentially suppressed with increasing @xmath116 . the results of table [ tsse ] show that the coulomb correlations embodied by @xmath235 can be reliably detected , for the signal - to - noise ratio of our experiment . we present the ground and excited states of the acceptor pairs for @xmath75 and @xmath105 in fig . [ fdoubleboron ] . compared with the ground states for @xmath75 and @xmath72 in fig . [ fdoubleboron]a and fig . [ fdoubleboron]c respectively , the excited states for @xmath131 ( fig . [ fdoubleboron]b ) and @xmath377 ( fig . [ fdoubleboron]d ) are slightly spatially elongated along the @xmath389 direction . this is consistent with the expected qpwf density in the interpretation of fig . [ fig5 ] ( main text ) , that the excited state orbitals have one hole occupying a `` 1/2 '' pseudospin state and the other occupying a `` 3/2 '' pseudospin state . these states display slightly different spatial anisotropies ( fig . [ fsingleboron]c and fig . [ fsingleboron]d ) . while we have measured the spatial tunneling probabilities for the excited states , measuring excited - state quasi - particle wavefunctions requires slow tunnel - in from the tip into the dopant system ( not slow tunnel - out to the tip ) . slow tunnel - in from the tip is required to make the current limiting slow tunnel rates additive for excited states ( rather than competitive with the ground state as in the case herein ) , in the presence of coulomb blockade . some details are given in reference . vs sample bias @xmath171 for single acceptor . two distinct thermally broadened steps corresponding to @xmath390 states and @xmath391 states are observed@xcite . b. peak current @xmath392 reduces exponentially as a function of tip height . c. energy splitting to the @xmath391 excited state ] in this section we discuss our confirmation of our model for the thermally broadened sequential tunneling in fig . [ fdidufit]e , by detailed investigation of the tip - height dependence of the current , in agreement with our results on arsenic donors in si@xcite . this is shown in fig . [ fivz]a for tip heights that sequentially increase by @xmath393 pm per curve , above a single b : si acceptor presenting the unique topographical signature discussed elsewhere@xcite . the data shows broadened current steps centred at @xmath394 v and @xmath395 v that have an exponential dependence on the tip height , and which fit very well the lineshape of purely thermally broadened current resonance of a dopant@xcite . the peaks shown here correspond to the @xmath396 ground state and @xmath397 first excited state discussed in section [ sspatialmodel ] and reference . qualitatively , the exponential dependence on tip height verifies the assumption that @xmath398 , as expected for the large vacuum tunneling barrier , and as required to obtain the results in fig . [ fig3 ] of the main text . we performed least - squares fitting to extract the height @xmath392 of the thermally broadened current steps . these @xmath392 are plotted as a function of tip height @xmath399 in fig . [ fivz]b for the @xmath390 ground state ( red curve ) and the @xmath391 excited state ( blue curve ) . the data are in excellent agreement with a z - dependence @xmath400 where @xmath401 m@xmath167 . this value is in good agreement with the expected value for @xmath402 for the @xmath403 ev barrier expected for shallow valence band acceptor in silicon . moreover , we find that the lever arm ( not shown ) and energy splitting ( fig . [ fivz]c ) between the ground and excited state does is essentially independent of tip height . this provides further verification of the sequential tunneling model , and that the tip does not strongly influence the acceptor - bound states . the results for the two - site hubbard model and its relationship to the molecular orbital model in fig . [ fig2]b of the main text are straightforward to derive . the coefficients @xmath44 and @xmath45 of the singlet ground state @xmath404 ( fig . [ fig2]b , main text ) are readily found by direct diagonalization to be @xmath405 and @xmath406 where @xmath407 . the dependence of @xmath408 and @xmath48 on @xmath262 in fig . [ fig2]b can be found by a simple transformation . the orthonormal localized states of the hubbard dimer model created by operators @xmath409 and @xmath410 are @xmath411 respectively@xcite . here @xmath412 and @xmath413 create ( non - orthogonal ) atomic orbitals with spin @xmath8 , @xmath414 , @xmath415 , and @xmath416 is the overlap between the atomic orbitals . re - writing the even and odd eigenstates of the single - particle potential and equation the two singlet ground states we find @xmath417 and @xmath418 . these relationships are used to map the hubbard model solution to the molecular orbital model in fig . [ fig2]b of the main text . this mapping is independent of normalization constants @xmath419 , @xmath420 , and @xmath421 . for fermions , entanglement is defined in terms of slater decompositions@xcite rather than the schmidt decompositions of distinguishable particles , and can be expressed independent of basis by a degree of entanglement . here we follow derivation in reference [ ] for entanglement to describe correlations in a two - site fermi - hubbard system . for two particles that obey fermi statistics , @xmath422 is characterized by an anti - symmetric matrix @xmath423 , that is , a slater decomposition . transformed into a block diagonalized form @xmath424 $ ] through a unitary rotation of the single particle states , where @xmath425 the number of nonzero @xmath426 is called the slater rank , and if the slater rank is 1 , the quantum correlation of the state is zero@xcite . it has been shown@xcite that the @xmath427 are the eigenvalues of the basis independent quantity @xmath428 , and based on this observation , the degree of entanglement @xmath55 of the two particles was defined as @xmath429 for a superposition of `` even / even '' and `` odd / odd '' singlets @xmath430 , @xmath431 it directly follows that the non - zero eigenvalues @xmath427 of @xmath428 are @xmath97 and @xmath98 . then , the degree of entanglement for the state @xmath432 is @xmath433

transport through a single molecular conductor is considered , showing negative differential conductance behavior associated with phonon - mediated electron tunneling processes . this theoretical work is motivated by a recent experiment by leroy et al . using a carbon nanotube contacted by an stm tip [ nature * 432 * , 371 ( 2004 ) ] , where negative differential conductance of the breathing mode phonon side peaks could be observed . a peculiarity of this system is that the tunneling couplings which inject electrons and those which collect them on the substrate are highly asymmetrical . a quantum dot model is used , coupling a single electronic level to a local phonon , forming polaron levels . a `` half - shuttle '' mechanism is also introduced . a quantum kinetic formulation allows to derive rate equations . assuming asymmetric tunneling rates , and in the absence of the half - shuttle coupling , negative differential conductance is obtained for a wide range of parameters . a detailed explanation of this phenomenon is provided , showing that ndc is maximal for intermediate electron - phonon coupling . in addition , in absence of a gate , the `` floating '' level results in two distinct lengths for the current plateaus , related to the capacitive couplings at the two junctions . it is shown that the `` half - shuttle '' mechanism tends to reinforce the negative differential regions , but it can not trigger this behavior on its own .

cond-mat0510066

the prospect of using molecules as the fundamental building blocks of future nanoelectronics devices is rather innovating and exciting from the point of view of potential applications . on the fundamental side , the field of molecular electronics also opens new directions for research because of the prominent role of phonon ( vibron ) excitations in electronic transport . the nano - objects which are connected to metallic leads may consists of individual molecules , self - assembled monolayers , or conjugated systems such as polymers and carbon nanotubes . the interplay of electron transport and molecular vibrations has triggered much interest , and unambiguous signatures of phonons have been detected in several experiments @xcite . a recent work @xcite considered electron injection from a scanning tunneling microscope ( stm ) tip into a carbon nanotube . a single - wall carbon nanotube ( cnt ) is freely suspended over a trench . the stm - tip is located near the center of the suspended part of the cnt . a dc - bias voltage @xmath0 is applied between the substrate and the stm , and the current flowing through the stm - tip - cnt - substrate structure is measured at a given tunneling distance , controlled by the setpoint current . the motivation for this special geometry is to allow for free internal vibrations to occur in the suspended portion of the nanotube , in particular the so - called radial breathing modes ( rbm ) @xcite . contrary to acoustic modes in such system , this mode has a rather high oscillation frequency , and the authors observed that conductance peaks in the current - voltage characteristics of the nanotube were surrounded by phonon side peaks , due to emission or absorption of rbm phonons . indeed , cnt display a sizeable coupling of electrons to rbm modes @xcite . the contacts to the nanotube being rather resistive , transport in this system is dominated by the coulomb blockade regime , and phonon side peaks were observed around each coulomb blockade peak in differential conductance plots . interestingly , the authors mention frequent detection of negative differential conductance ( ndc ) regions . striking ndc features also appear in a very recent work by sapmaz _ @xcite , in a transport measurement of a suspended cnt , where current flows through the cnt , between two contacts at the substrate . in this work , phonon side peaks are attributed to longitudinal stretching modes , and the steps in the @xmath1 characteristics are followed by spikes , thus displaying ndc features . the purpose of the present theoretical work is to show that such ndc features can be described quite simply using a generic model which consists of a quantum dot with a single orbital level , coupled to on - site single phonon mode , and connected to leads by tunnel junctions . due to the weak tunnel couplings , the physics of ndc appear to be a consequence of the transport through small polaron states on the molecule . notice that polaron formation in cnt is suggested by several works @xcite . similar models have previously been considered in the literature @xcite . here we consider a molecular system or a nanotube setup whose tunneling matrix elements from the molecule to the leads are asymmetric , which is the case in ref . , and possibly also in ref . we propose that given this asymmetry , ndc can be obtained for a wide range of parameters and it can even be quantified by analytical means for strong asymmetry . furthermore , we argue that our approach for the description of ndc is by no means confined to the experimental geometry of ref . , which can be considered as an `` experimental paradigm '' for phonon modulated molecular transport . the present approach should apply to any molecular transport in the coulomb blockade regime where an optical - like phonon mode dominates . notice that ndc was found in ref . within a symmetric model , using an ansatz for the franck - condon factors coupling the leads to the polaron levels . catastrophic ndc was also proposed to occur in ref . , but due to an additional mechanism . in the framework of a single site model , ndc was also found in an adiabatic treatment of the polaron problem @xcite . it has been proposed to occur in the case of two competing molecular states @xcite or within a two - site model @xcite . besides electron - phonon coupling , other physical mechanisms can lead to ndc @xcite . in molecular electronics , one is tempted to use existing theoretical tools developed in the context of mesoscopic physics because the size of the devices allows a coherent description of the transport inside the molecule . when the molecule is connected to good contacts , and when the role of coulomb interaction and electron - phonon interactions is reduced , the landauer formulation of transport combined with a green s function calculation of the transmission coefficient serves as a good starting point for computing transport @xcite , but it then neglects the dynamical degrees of freedom of the molecule . phonons can be included in these approaches , analytically @xcite , or using numerical calculations @xcite based on the work of ref . . alternatively , the landauer approach can be generalized to a situation where interactions are restricted to a finite region @xcite using non - equilibrium green s functions . on the other hand , molecular contacts or the contacts to a nanotube are often of poor quality . it is then reasonable to think of the molecule as a quantum dot , which is subject to the coulomb blockade . this point of view was adopted by several authors @xcite , and is confirmed by experiments . one of the advantages of modecular electronics comes from the size of the individual nano - object which is connected to the electrodes . even at relatively large temperatures ( a few kelvins ) , the dynamics within the molecule / nanotube is fully phase coherent . on the other hand , the tunneling rates to the reservoirs are typically small compared to temperature . one can then derive rate equations for the electron population of the dot . here our goal is to use the simplest model which can account for the physics of the experiment of ref . . we therefore derive a kinetic equation approach which can account for the observed ndc effect , treating the electron - phonon coupling non - perturbatively while still allowing for an intuitive understanding of the physics at hand . furthermore , consequences can be derived on the electron transport as well as on the non - equilibrium phonon population . in the present work the local mode is not coupled to any environmental degrees of freedom , like phonon modes or electron - hole excitations in the substrate . although such coupling is probably relevant in some experiments , the very high phonon quality factors ( @xmath2 ) obtained in other works @xcite justify to neglect it as a first step . experimentally , the addition of voltage gates on a molecular transport setup is still challenging but it is possible @xcite to approach three metallic probes in a nanometer scale region . yet , there is also a motivation to study a setup where the molecular levels are `` floating '' instead of being fixed by a gate . besides the strong asymmetry of the tunneling rates , a specific feature of our work is that ndc can occur in such absence of gate voltage , and that the ratio of the right / left capacitance plays an important role , as it dictates the location of the molecular levels . the structure of the paper is as follows . the model is introduced in sec . ii , and the derivation of the rate equations is provided in sec . the general considerations about current transport and phonon occupation numbers are discussed in sec . numerical results are presented in sec . v , illustrating the role of the asymmetry of the capacitances , of the shift of the dot level , and of the so - called half - shuttle mechanism on the ndc . we conclude in sec . although the potential relevance of our approach applies to the carbon nanotube experiments of refs . , we refer to the central region between the leads as the quantum dot . this quantum dot is weakly coupled to two metallic electrodes by tunnel junctions . in the stm geometry of refs . , the electrodes represent the stm - tip and the substrate , to which we refer as the left ( @xmath3 ) and right ( @xmath4 ) electrodes , respectively . each tunnel junction ( @xmath5 ) is characterized by a resistance ( @xmath6 ) and a capacitance ( @xmath7 ) . while @xmath8 and @xmath9 are constant for a given contacted nanotube sample , @xmath10 and @xmath11 are functions of the tip - tube separation . the capacitances @xmath11 and @xmath9 are sample - dependent @xcite . for generality , one can add a gate voltage @xmath12 and a gate capacitance @xmath13 . we focus on the strong coulomb blockade regime , assuming that the number of electrons which can be added to the dot is restricted to one . the physical features revealed in the corresponding bias window can be easily extended to a full span of many coulomb blockade peaks , as well a several orbital levels as in refs . . the extra - charge electron state is locally coupled to a phonon mode : in refs . this mode is identified with the radial breathing mode ( rbm ) of the carbon nanotube @xcite and , in ref . , with the longitudinal stretching mode . in view of the phonon - assisted tunneling processes , the above - mentioned coulomb blockade regime implies that the charging energy of the dot is assumed to be infinitely large compared to the relevant energy scale determined ( at low temperature ) by the phonon energy @xmath14 and the bias voltage . in the absence of phonons , electron transport would occur only through a single electron state on the dot . in the coulomb blockade regime , far from the kondo regime , spin degrees of freedom are neglected . the hamiltonian of the system is written as follows : @xmath15 where @xmath16 @xmath17 @xmath18 here the operator @xmath19 ( @xmath20 ) annihilates ( creates ) an electron on a single dot level of energy @xmath21 ; similarly , @xmath22 ( @xmath23 ) annihilates ( creates ) an electron with momentum @xmath24 and energy @xmath25 in the @xmath26th lead . the rbm is linearly coupled ( with the coupling energy @xmath27 ) to the electric charge on the dot ; the rbm excitations are annihilated ( created ) by @xmath28 ( @xmath29 ) . in eq . ( [ v ] ) , @xmath30 is the energy associated with the tunneling coupling to the dot ; for simplicity , @xmath30 is assumed to be energy - independent ( constant density of states in the metallic leads ) , but we take into account the dependence of the tip - tube tunneling matrix amplitude on the boson coordinate @xmath31 : due to the `` breathing '' motion of the tube , the tip - tube tunneling distance deviates from its equilibrium value . explicitly , we assume an exponential @xmath31-dependence : @xmath32 where @xmath33 is determined by the ratio of the amplitude of the zero - point rbm oscillations to the electronic tunneling length ( @xmath34 ) which characterizes the tunnel barrier between the stm tip and the nanotube . such position dependent amplitudes have been introduced in the context of nano - mechanical electronic devices@xcite , where one refers to the `` shuttle '' mechanism as the central region oscillates between the two electrodes . on the other hand , in our situation the tube - substrate tunneling matrix amplitude is @xmath31-independent : @xmath35 . for @xmath36 , we refer here to this position dependent tunneling hamiltonian as the `` half - shuttle '' mechanism , because only one of the tunneling amplitude ( left ) is modified by the position . in the model hamiltonian ( [ h ] ) , formally describing the phonon - assisted tunneling of otherwise noninteracting electrons , the charging effects are taken into account via the bias - voltage dependence of the position of the dot level ( @xmath21 ) with respect to the chemical potentials of the leads , @xmath37 . a gate voltage @xmath12 and gate capacitance @xmath13 can be included . in our analysis , @xmath21 includes the change in the charging energy of the dot when one extra electron is added to the dot , and which is obtained from the electrostatic energy consideration : @xmath38 where @xmath39 is the `` bare '' energy level of the electron level in the dot , @xmath40 is the charging energy and @xmath41 is a background ( fractional ) charge of the dot . furthermore , the potential of the dot , @xmath42 , is given by @xmath43 @xmath44 in typical experiments @xmath45 . it is worth noticing that in the `` floating - level '' geometry which is considered in part of this work ( e. g. without the gate electrode , @xmath46 ) , the fractional charge is not fixed but it can instead be affected by changes in the capacitances of the junctions @xcite ( that is , @xmath47 may depend on the tip - tube separation ) . assuming weak coupling to the leads , it is convenient to eliminate the electron - phonon coupling in @xmath48 by the unitary transformation @xmath49 , with @xmath50 in the rotated basis , the electron state in the dot becomes `` dressed '' with phonons , forming a small polaron . this results in the `` polaron shift '' of the dot level energy , @xmath51 , and a renormalization of the dot - lead tunneling coupling . in the polaron representation , the hamiltonian reads : @xmath52 in eq . ( [ tildeht ] ) , we have used the fact that @xmath53 based on the assumption that the leads are in thermal equilibrium at given chemical potentials ( @xmath54 and @xmath55 ) , independently of the state of the dot , one can derive a kinetic equation for the reduced density matrix by tracing out the electrode degrees of freedom from the total density matrix . such equations have been used by several authors,@xcite but for sake of completeness we provide here a full derivation . in the polaron representation , the reduced density matrix of the phonon - coupled dot is given by @xmath56 where @xmath57 is the total density matrix in the interaction picture ( with respect to the tunneling coupling to the leads ) , which obeys the operator equation of motion : @xmath58 \ ; , \label{tilderhov_eq}\\ \tilde{{\cal v}}(t ) = e^{i t \left ( \tilde{{\cal h}}_0 + { \cal h}_{leads } \right ) } \ , \tilde{{\cal v } } \ , e^{-i t \left ( \tilde{{\cal h}}_0 + { \cal h}_{leads } \right)}~.\end{aligned}\ ] ] using the integral form of eq . ( [ tilderhov_eq ] ) , @xmath59 + \tilde{\rho}_v ( 0)\ ; , \label{tilderhov_inteq}\ ] ] substituting eq . ( [ tilderhov_inteq ] ) into eq . ( [ tilderhov_eq ] ) , and taking the trace over the electronic degrees of freedom of the leads with the above mentioned assumption of @xmath60 , where @xmath61 is the equilibrium density matrix of the leads , we arrive at the integro - differential equation of motion for the reduced density matrix in the polaron representation : @xmath62 + h.c . \ , , \label{kin_eq}\end{aligned}\ ] ] where @xmath63 is the density of states in the @xmath26th electrode , @xmath64 e^{- i \alpha p(\tau ) } ~,\ ] ] with @xmath65 and @xmath66 \label{ff}\ ] ] ( where `` p.v . '' stands for the principal value ) . from the above expressions for the kernels @xmath67 , it follows that the relevant retardation time is of the order of the inverse temperature , @xmath68 . assuming temperature to be high compared to the tunneling rates , @xmath69 , and considering the long - time behavior of the reduced density matrix , @xmath70 , we construct an asymptotic solution of eq . ( [ kin_eq ] ) by using an improved perturbation expansion @xcite : @xmath71 in eq . ( [ expansion ] ) , @xmath72 is a formal perturbation parameter that reflects a weakness of tunnel coupling and must be formally attributed to the right - hand side of eq . ( [ kin_eq ] ) : @xmath73 with @xmath74 . in the first - order equation in @xmath72 , retardation effects are neglected ( the markovian approximation ) , and in the above equation , we can replace @xmath75 . by constraction , @xmath76 is , on the time scale @xmath68 , a slowly evolving part of the reduced density matrix , while the rapidly oscillating ( with frequencies determined by a multiple integer of @xmath77 ) part of the right - hand side of eq . ( [ kin_eq2 ] ) is absorbed into @xmath78 . in what follows , we assume that the phonon frequency is high compared to the tunneling rates , @xmath79 . this `` anti - adiabatic '' condition allows us to neglect @xmath78 as being negligibly small compared to @xmath76 . under the above assumptions , eq . ( [ kin_eq2 ] ) reduces to a differential operator equation for @xmath80 . although @xmath76 is diagonal in the on - dot electron number ( no coherence between subsequent tunneling events at long times ) , @xmath81 yet it is non - diagonal in the phonon number due to the presence of the displacement operators ( @xmath82 ) in the right - hand side of eq . ( [ kin_eq2 ] ) . however , equations for the diagonal and off - diagonal parts of @xmath80 are decoupled from each other . as a result , in steady state ( @xmath83 ) , which we are interested in , @xmath76 becomes diagonal in both the electron and phonon numbers . introducing the electron - phonon joint probabilities @xmath84 for @xmath85 additional electrons and @xmath86 boson excitations on the dot , we obtain the system of rate equations for the joint probabilities , which in steady case can be written in the following form : @xmath87 with the normalization condition , @xmath88 . the charge / phonon transition rates are given by @xmath89 @xmath90 in the above equations , @xmath91^{-1}$ ] is the fermi distribution function of the @xmath26th electrode , @xmath92 , @xmath93 @xmath94 explicitly , we have for the oscillator matrix elements : @xmath95 from the solution of eq . ( [ rate_eqs ] ) we can calculate the dc current and expectation values of the phonon observables . it should be noticed that in the original ( non - polaron ) representation , the resulting phonon density matrix is given by @xmath96 in general , @xmath97 has contributions from both charge states @xmath98 and is non - diagonal in vibrational space ( two shifted oscillators ) . in order to express the current in terms of the joint probabilities @xmath99 , we have to average the current operator taken in the polaron representation , @xmath100 with the total density matrix @xmath101 where @xmath57 obeys eq . ( [ tilderhov_inteq ] ) . following the same steps and assumptions which we have made in the derivation of the rate equations , we obtain the average dc current flowing from the @xmath26th lead to the dot as ( we set @xmath102 ) @xmath103 where @xmath104 are the partial relaxation rates contributing to @xmath105 , eq . ( [ gamma_n ] ) , from the @xmath26th electrode : @xmath106 by virtue of the rate equations , we have conservation of the current , @xmath107 . in calculating @xmath1-characteristics from eqs . ( [ kin_eq ] ) and ( [ ij ] ) , we will focus on the case of highly asymmetric two - junction model with @xmath108 . this corresponds to the typical experimental situation with stm measurements where a contact to the stm tip plays the role of the high resistive tunneling junction . for instance , in stm measurements on suspended nanotubes , the typical ratio @xmath109 can be huge depending , in particular , on the tunneling distance between the stm - tip and the nanotube , @xmath110 typically . in the present section , no gate is present ( floating level ) thus @xmath46 . at the same time , according to the data in ref . obtained from spectroscopy measurements on suspended nanotubes , the ratio @xmath111 can be smaller as well as larger than unity depending on an effective length of the portion of the nanotube that is on the substrate , and which can be relatively short due to local defects induced by the edges of the trench . thus , there is no dominating asymmetry in the capacitive coupling of the dot to the both electrodes . as a result , in the voltage - biased system , the effective position of the polaron level with respect to the chemical potentials of the leads is strongly affected by the ratio @xmath112 which is not negligibly small like @xmath113 , or can even be larger than unity @xcite . it results that the polaron level ( assuming neither gate - induced nor intrinsic shift of the polaron level at zero voltage ) is not trivially stuck to the chemical potential of the far less resistive junction electrode , @xmath55 . considering the system of stationary rate equations , eq . ( [ rate_eqs ] ) , and writing explicitly expressions for the fermi factors entering eq . ( [ l_ls_gr ] ) , we obtain [ see also eqs . ( [ energy_level ] ) , ( [ phi_potential ] ) ] : @xmath114 where @xmath115 , and @xmath116 determines the position of the polaron level with respect to the lead chemical potentials at @xmath117 , considered as the reference level of zero energy . from eq . ( [ f_lr ] ) , the role of charging effects in the single - electron resonant level problem can be viewed as follows : when changing the bias voltage , the chemical potentials move in opposite directions ( depending on the sign of @xmath0 ) with different `` velocities '' determined by the capacitance ratios , @xmath118 and @xmath119 for the left and right electrode , respectively . the above picture is given in the `` reference frame '' of the polaron level , where its position is voltage - independent . for the floating - dot geometry , @xmath120 may be strongly affected by a background charge , @xmath47 . in the presence of a gate electrode @xcite , @xmath120 can be controlled by the gate potential , and @xmath121 in eq . ( [ phi_potential ] ) , must also include the gate capacitance , with @xmath122 . in this section , in order to make analytical progress on the prediction of the current transport , we formally assume temperature to be zero . because the rate equations have been derived in the high - temperature approximation ( @xmath123 ) , the zero - temperature assumption is justified for the voltage - biased cases where the chemical potentials , @xmath37 , are not very close to the phonon sidebands , so that thermally activated tunneling processes can be neglected . we first notice that in the case where the polaron level @xmath120 lies beyond the bias - voltage window , the system of rate equations , eq . ( [ rate_eqs ] ) , reduces to the following : @xmath124 which is satisfied for all phonon numbers @xmath125 . the ground state solution of eq . ( [ det_ball ] ) is either @xmath126 ( @xmath127 ) for @xmath128 , or @xmath129 for @xmath130 . as a result , in this case the current is zero : no contribution to the current occurs from the phonon sidebands alone ( cf . ref . ) . ( [ det_ball ] ) also works in the less trivial case when @xmath120 lies within the bias - voltage window , but @xmath131 . for instance , for @xmath132 , the solution of eq . ( [ det_ball ] ) ( valid for arbitrary @xmath133 ) reads @xmath134 , where @xmath135 and the current is given by @xmath136 for large @xmath137 , the exponential prefactor in eq . ( [ curr_through_polaron ] ) leads to a significant suppression of the current at low bias voltages ( the so - called franck - condon blockade @xcite ) . the role of @xmath33 ( `` half - shuttle '' ) will be discussed later . at higher voltages , such that phonon sidebands enter the bias - voltage window , eq . ( [ det_ball ] ) fails , and we have to go back to the full system of rate equations , eq . ( [ rate_eqs ] ) , taking into account the total balance of probabilities for transitions between different states of the dot . this renders the problem of calculating phonon distribution functions , @xmath99 , and eventually the @xmath1-characteristics , a rather complicated one to be solved analytically . however , for the case of high asymmetry in the tunneling couplings , it is possible to make some further analytical progress before passing to numerical calculations @xcite . we now exploit the condition @xmath108 for analyzing steady solutions of the rate equations . in the general case , when a bias voltage @xmath0 is applied and the current is not identically zero ( up to thermally activated processes ) , the polaron level is accompanied by @xmath138 phonon sidebands captured in the bias - voltage window , where we introduce ( `` int '' stands for the integer part ) @xmath139 independently of the sign of @xmath0 , the integer @xmath140 ( @xmath141 ) is defined as a number of phonon sidebands lying between the polaron level and the chemical potential of the electrode connected to the dot through the more ( less ) resistive junction . an example with @xmath142 and @xmath143 for @xmath144 is shown on fig . [ energy_dgrm ] . due to high asymmetry @xmath108 , only probabilities @xmath145 with @xmath146 $ ] do not vanish in this limit . for the example on fig . [ energy_dgrm ] , the probabilities to have an electron on the dot with any @xmath147 phonons , @xmath148 , are suppressed due to `` fast '' tunneling of the electron to the right electrode via @xmath149 open channels ; the probability to have the dot with an empty electron state but with the number of phonons @xmath150 ( and higher ) , @xmath151 , is also negligible due to the `` fast '' tunneling of an electron from the right electrode with absorbing two phonons . in other words , on a large time scale determined by @xmath152 , the polaron - hole states with 0 and 1 phonon excitation are ( quasi)steady states with respect to the tunneling coupling to the right electrode . notice that in the asymmetric situation , the phonon number distributions are nearly the same in the polaron basis ( the @xmath153 s ) and the original basis ( @xmath97 ) , as shown by eq . ( [ ph_distrib ] ) . the case @xmath154 can be treated in the same manner . we obtain that only probabilities @xmath155 with @xmath146 $ ] are not vanishing as @xmath113 . as a result , with a good accuracy , controlled by the smallness of @xmath113 , the current flowing from the left to the right , @xmath156 ( [ ij ] ) ] , can be written as @xmath157 where @xmath158 , and @xmath159 with @xmath160 . the current is given by a sum of partial currents , @xmath161 , representing @xmath141 conducting channels which are distributed with the corresponding phonon occupation probabilities . the magnitude of each partial current depends on the previously defined number of phonon sidebands @xmath140 captured between @xmath120 and @xmath54 . when varying @xmath0 , the number of captured phonon sidebands , @xmath140 and @xmath141 , can change . in the voltage ranges where @xmath140 and @xmath141 do not change , we obtain plateaus in the @xmath1 characteristics . assuming for concreteness @xmath144 , we see from eq . ( [ i_posv ] ) that if @xmath0 increases in such a way that only @xmath140 changes ( @xmath140 can only increase with increasing @xmath0 ) , the magnitude of each partial current , and hence the net current , will increase . thus , we obtain here the positive differential conductance ( pdc ) behavior which is commonly computed and observed . on the contrary , if when increasing @xmath0 , only @xmath141 changes ( i.e. increases ) , while @xmath140 remains constant , then a new phonon - assisted channel is added . this leads to redistribution of phonon occupation probabilities between all open channels . paradoxically , the net current may decrease , leading to a ndc behavior on @xmath1 @xcite . for @xmath144 , assuming that due to the voltage increase @xmath162 the number of phonon sidebands in the bias - voltage window has changed as @xmath163 , using eq . ( [ i_posv ] ) the variation of the current @xmath164 becomes : @xmath165 \ , . \label{deltai}\ ] ] thus , for this particular situation , assuming for simplicity @xmath166 , the condition for the ndc is @xmath167 : large values of the electron - phonon coupling do not favor ndc . the magnitude of the negative step on @xmath1 increases with increasing phonon occupancy @xmath168 . for @xmath154 , considering the same transition , @xmath169 , one finds : @xmath170 \ , , \label{deltai_negv}\ ] ] with the same ndc condition , @xmath167 , for @xmath171 . from the above expressions , we also notice that for positive ( negative ) bias voltage and non - zero but small half - shuttle @xmath172 , the ndc effect increases ( decreases ) for the first phonon - assisted step of the @xmath1 characteristic . as far as the bias voltage increases , the current eventually saturates and does not change in practice . for large bias voltages @xmath173 , the corresponding saturation currents @xmath174 are given by ( up to corrections @xmath175 ) : @xmath176 in the absence of the half - shuttle mechanism , we obtain that the saturation current does not depend on the phonon distribution , @xmath177 . for @xmath178 , considering equilibrated phonons , i.e. , forcing @xmath179 in the limit of strong phonon relaxation , we obtain the saturation currents : @xmath180 for non - equilibrated phonons , eq . ( [ isat_shuttle ] ) is expected to be accurate in the limit @xmath181 . nevertheless , for relatively small @xmath167 , the asymmetry in the current saturation values due to the half - shuttle can be noticeable at low bias voltages for @xmath182 because of the fast saturation of the partial currents @xmath161 ( fast convergency of the series in eqs . ( [ i_posv ] ) and ( [ i_negv ] ) ) . we now turn to the numerical solution of the rate equations , and explore the parameter space in order to find the signatures of ndc . after writting the system of the rate equations , ( [ rate_eqs ] ) , in the matrix form , @xmath183 , with @xmath184 , the problem reduces to finding the zero eigenvalue of the matrix @xmath185.@xcite the typical maximum number of phonon states has been taken around 40 , so that results of the calculation would not depend on this choice for the considered range of parameters . from numerically found solutions @xmath186 , the current has been calculated in the left and right electrodes using eq . ( [ ij ] ) . the achieved tolerance for @xmath107 was @xmath187 . contrary to the previous analytical arguments , calculations are performed at finite temperature . for calculation purposes we assume a large asymmetry in the tunneling rates , @xmath188 . yet , most of our results also hold for moderate asymmetries . the current @xmath189 is plotted in units of @xmath190 , the bias voltage @xmath0 , the polaron level @xmath120 and temperature @xmath191 are given in units of @xmath77 . in subsection a , one considers the floating level case . we consider the effect of a given shift of the dot level ( due to background or gate charges ) in subsection b. in these two subsections we will disregard the half - shuttle ( @xmath192 ) , which will be considered in the last subsection . we start by considering the case where the capacitances surrounding the dot fully specify the position of the polaron level . at zero bias voltage , the polaron level is aligned with the chemical potentials of the electrodes , @xmath193 . in this case , as it immediately follows from the system of rate equations , eq . ( [ rate_eqs ] ) , we have the symmetry relation @xmath194 and consequently @xmath195 . notice that for the `` floating - level '' geometry the case of @xmath193 is an exceptional one , and must be viewed as a reference point . [ ivpdc ] shows the @xmath1 characteristics for different @xmath137 s assuming small @xmath196 , so that the polaron level lies closer to @xmath55 . in the experiment of refs . , this means that the nanotube has a large overlap with the conducting substrate . as long as @xmath197 , that is if @xmath198 ( this condition is satisfied on the bias - voltage range which is plotted ) , @xmath199 , which results in the pdc behavior of @xmath1 . the pdc steps correspond to @xmath140 increasing by one each time @xmath0 passes through a multiple integer of @xmath200 . explicitly , the current - step amplitude at @xmath201 corresponding to the current increase is given by @xmath202 for smaller @xmath137 s the current - step amplitude decreases faster with increasing @xmath0 , which leads to the saturation of the current at lower voltages . note that the first step ( low bias voltage ) is rounded . this feature is specific to the fact that the capacitances are asymmetric . indeed , increasing the bias voltage from zero , the chemical potentials move away from the polaron level , but the chemical potential whose lead has the largest capacitance remains close to this level . in this event , thermally activated tunneling processes can be effective . when the bias voltage is further increased the phonon sidebands which contribute to new steps are those _ above _ the polaron level , and thus their corresponding steps are not thermally rounded . in the case where the asymmetry in the capacitances is opposite , which is shown in the next figure , the same reasoning applies . [ ivndc ] shows the @xmath1 characteristics for different @xmath137 s assuming a relatively large @xmath203 . in the experiment of refs . , this means that the nanotube has a small length , thus a small overlap with the conducting substrate , and thus a small capacitance compared to that of the tip - nanotube contact . alternatively , a large @xmath119 could be achieved if the nanotube contains impurities or bends . in the bias - voltage window , the polaron level is closer to @xmath54 than to @xmath55 . within the bias - voltage range which is plotted , the current changes by steps when @xmath0 passes through a multiple integer of @xmath204 , as before . at these points , @xmath141 increases by one ; correspondingly , one more phonon - assisted channel is added . for @xmath205 one still observes the pdc behavior , although the height of the current steps is strongly suppressed compared to the previous case on fig . [ ivpdc ] . for @xmath167 ndc occurs in the first step , and with decreasing @xmath137 , more ndc steps appear . for all curves , eq . ( [ deltai ] ) gives the current jump between the second and first plateaus on fig . [ ivndc ] . we have also performed calculations of the current voltage characteristics in this ndc regime while increasing gradually the temperature ( not shown ) . temperature tends to smooth the steps and to suppress ndc altogether when it becomes comparable to the spacing between the steps . at the same time that we monitor the electronic current , it is instructive to examine the phonon occupation numbers in order to quantify pdc or ndc behavior . the phonon distributions @xmath206 , which play the role of probabilities of open channels , at different bias voltages are shown on fig . [ pnndc ] . this figure demonstrates the increase of the number of phonon excitations with increasing @xmath141 ( see eq . ( [ mmm ] ) ) . it shows that the phonon number which are occupied is restricted by @xmath207 . thus , the phonon distribution allows to understand the height of the pdc or the ndc steps . in the two previous cases , two limits were considered : either @xmath181 or @xmath208 . on the way to the more general situation when @xmath119 and @xmath118 are comparable , let us now consider in more detail the capacitively symmetric case when @xmath209 . the @xmath1 characteristics for @xmath210 and 1.5 are shown on fig . [ iv_cl05e0 ] . for @xmath205 , the current shows a stepwise increased with essentially no ndc behavior . note that because the capacitances are equilibrated , the first step does not display substantial rounding due to thermal effects . for @xmath210 , one observes a stepwise increase of the current , but the onset of each step is associated with a small spike , thus exhibiting ndc behavior . this ndc behavior is thus qualitatively different from the one previously observed for the case of asymmetric capacitances . the spikes bringing the ndc features to the @xmath1 characteristics are robust with respect to temperature . in fig . [ smallpeak_tdependence ] one sees that these spikes are broadened with temperature , but ndc persists as long as the temperature is not comparable to the phonon frequency . on the @xmath1 for @xmath205 ( fig . [ iv_cl05e0 ] ) , the ndc - singularities appear only for higher voltages , however , the first phonon - assisted current step ( at @xmath211 ) is also deformed . 5.5 cm at low temperature , it is interesting to make a zoom of a given step in order to see how the spikes evolve when the electron phonon coupling is varied . fig . [ iv_peaks ] shows these `` singularities '' on the second step of the @xmath1 characteristic for a wide range of @xmath137 s ( cf . [ iv_cl05e0 ] ) . note that the ndc behavior is more pronounced with increasing @xmath212 . indeed , in the limit of vanishing electron phonon coupling , one gets only one step due to the polaron ( electron ) level , and all ndc features are absent . there is therefore an optimal value of @xmath137 which displays maximal ndc behavior . further increasing @xmath137 , ndc converts to the pdc case ( the curves for @xmath213 ) . in this case of equal capacitances ( @xmath214 ) , the polaron level lies in the middle of the bias - voltage window at all @xmath0 , so that both @xmath140 and @xmath141 increase by one when @xmath0 passes through a multiple integer of @xmath215 . in particular , at @xmath216 the transition @xmath217 occurs : below @xmath216 the polaron level is the only one which lies within the bias voltage window , while immediately above @xmath216 , two phonon sidebands are simultaneously captured by the bias window . as a result , in the vicinity of @xmath216 ( how close depends precisely on temperature ) , we have a competition between the pdc- and `` possible ndc''-type contributions to the current , corresponding to the processes of electron tunneling from the left electrode into the dot with emission and absorption of one phonon , respectively . in the transition region , @xmath218 , the current is given by ( here we take into account the fermi factors ) @xmath219 @xmath220 and @xmath221 . explicitly , we have @xmath222 f(\omega - v/2 ) \right ) \right ] \ , . % \label{}\ ] ] > from the numerics , it follows that up to @xmath223 , where @xmath224 is the half - width of the fermi distribution function ( see caption of fig . [ iv_peaks ] ) , we have @xmath225 , thus the current successively increases up to : @xmath226 this effect is due to the _ high asymmetry _ in the tunneling rates @xmath133 : the probability to have an empty dot with one phonon , @xmath227 , remains negligible , @xmath228 , until there is some probability to have thermally activated electrons of energy @xmath229 coming from the right electrode . only then , for @xmath230 , where @xmath231 has practically reached unity , @xmath232 starts deviating from zero . thus , we have @xmath233 @xmath234 \right ) \;. \label{g_vs_alpha}\ ] ] for @xmath230 , the @xmath0-dependence of the current is given by @xmath232 . with increasing @xmath0 , @xmath232 also increases and saturates to some quantity ( depending on @xmath137 ) which is comparable with @xmath235 ( the right - hand side plateau on fig . [ iv_peaks ] ) . from numerics we find that at @xmath236 , for instance , for @xmath237 , we correspondingly have @xmath238 . from eq . ( [ curr_diff ] ) and fig . [ g_plot ] we see that the ndc plateau disappears for small @xmath239 ; with increasing @xmath137 , the ndc contribution from an open channel increases - decreases having the maximum at @xmath240 , and for @xmath241 the ndc contribution converts to the pdc one ( cf . [ iv_peaks ] ) . up to now , we have considered only strong asymmetric capacitances and equal capacitances . a common feature of these choices is that the current voltage characteristics displays steps ( with or without ndc behavior ) whose spacing in voltage is essentially always the same at relatively low voltages . indeed , for the asymmetric capacitances plots each step is separated by ( approximately ) @xmath77 , while for @xmath214 , this spacing is @xmath215 . if one was to plot the differential conductance , as in the experiment of ref . , one would obtain a periodic series of peaks . what happens to this periodicity in an intermediate situation where the capacitances are comparable , but not equal ? [ iv_twoperiodic ] shows current - voltage characteristics with two steps of different length for the case @xmath242 . the number of phonon sidebands above the polaron level , @xmath140 , increases by one with the voltage period @xmath243 ; for these steps we _ always _ have a pdc behavior as it is demonstrated in the previous section . at the same time , the number of phonon sidebands below the polaron level @xmath244 with the voltage period @xmath245 : one more channel becomes open which yields the possibility of ndc . the larger @xmath137 , the more pdc - steps at low bias voltages are observed . in this subsection , we consider the more general situation where , due to gate or background charges , the polaron level @xmath120 is shifted by a constant value with respect to the chemical potentials of the leads at zero bias voltage . for @xmath246 , the symmetry @xmath247 is violated , moreover @xmath1 is not shifted in a trivial manner , meaning that it does not simply follow @xmath120 . [ iv_cl025eshifted ] shows two current - voltage characteristics for different values of @xmath120 . as it was mentioned in sec . iv , the current is zero when the polaron level lies beyond the bias voltage window . for negative @xmath0 , the polaron level becomes captured in this window at @xmath248 . at this point , the number of phonon subbands below the bare level @xmath140 in this case suddenly changes from @xmath249 to @xmath250 . in the curves of fig . [ iv_cl025eshifted ] corresponding to @xmath251 and @xmath252 , we observe a transition to @xmath253 and @xmath142 , respectively . this results in a fast saturation of the curves at negative voltages ( the saturation is faster for the curve with larger @xmath120 ) . previously , we found that small @xmath119 leads to pdc for the first phonon assisted peaks . here , although @xmath254 is relatively small in fig . [ iv_cl025eshifted ] , for positive @xmath0 , we observe ndc behavior already for the first phonon - assisted step on @xmath1 ( around @xmath255 ) for @xmath256 . this can be explained as follows . at small positive voltages , we first have a zero - current plateau until the polaron level becomes captured in the bias - voltage window ( @xmath257 ) at @xmath258 . this leads to the first pdc step on the @xmath1 curve at @xmath259 . then the second plateau persists until @xmath260 , which corresponds to the phonon sideband level @xmath229 becoming captured in the bias window ( @xmath249 , @xmath143 ) . because @xmath210 is less than the critical value @xmath261 ( see sec . iv ) , this transition gives the ndc step . next a plateau persists until , @xmath262 when the phonon sideband level @xmath263 is captured in the bias window ( @xmath264 , @xmath143 ) resulting in a pdc step on @xmath1 . the conditions to observe ndc for the first phonon - assisted step are therefore : @xmath265 the width of the ndc plateau is @xmath266 . ( [ conditionndc ] ) is fulfilled for the case @xmath267 on fig . [ iv_cl025eshifted ] , but it fails for @xmath268 , where we have the pdc behavior of the first phonon - assisted step on @xmath1 , for @xmath269 . [ iv_tdependence ] shows the temperature dependence of the ndc step for @xmath270 . with increasing temperature @xmath191 , the phonon - assisted channel opens for higher voltages when the one - phonon excited state of the empty dot becomes unreachable for thermally activated electrons from the right electrode at energy @xmath271 , ( as before , @xmath224 is the half - width of the fermi function ) . as a result , the ndc region with a decreased slope gradually shifts towards the next step on @xmath1 . fig . [ iv_tinducedndc ] shows the opposite situation : the ndc appears this time with _ increasing _ temperature . at low temperature , when the fermi half - width @xmath224 is negligibly small , we only observe pdc behavior for the first step on @xmath1 at @xmath258 , where we suddenly have a transition to @xmath272 . at this threshold , both the polaron level and the first phonon subband below it belong to the bias voltage window . then we obtain a plateau in the @xmath1 curve up to @xmath273 , where the next pdc step is associated with the increase of @xmath140 by one . however , if the temperature is increased so that @xmath224 becomes comparable with @xmath274 , now at @xmath275 the occupation probability @xmath168 will be negligibly small because of thermally activated electrons tunneling from the right electrode . @xmath168 starts developing only at @xmath276 leading to ndc behavior . here we have used that at @xmath277 , the half - width of the fermi function is about @xmath278 . at much higher temperature everything is of course washed out ( not shown ) . we conclude this subsection by some comments on the role of the asymmetry in the tunneling rates @xmath279 s on current transport . [ iv_cl05e025 ] shows @xmath1 characteristics for the different cases of junction asymmetry . as it is seen from the plots , the ndc steps in the cases of high and moderate asymmetry ( @xmath280 and 0.1 , respectively ) turn into pdc steps in the symmetric case ( @xmath281 ) . we also notice that the ndc steps are slightly shifted with respect to their pdc counterparts in the symmetric case . this shift is associated with the finite width of the fermi function , and it has been already been discussed in the two previous figures . the height of the phonon - assisted steps in the asymmetric case is noticeably large compared to the symmetric one . this is related to the phonon distribution , an example of which ( at @xmath282 ) is shown on fig . [ iv_cl05e025 ] for both cases ( symmetric and asymmetric ) . in the asymmetric case , the number of phonons is restricted by @xmath141 , while in the symmetric case the phonon distribution is more spread out . as a result , in the symmetric case , the phonon - assisted contribution to the current is weakened . in this subsection we discuss the role of the half - shuttle mechanism on @xmath1 characteristics . it is worth to be mentioned that in the absence of the half - shuttle ( @xmath192 ) , the rate equations ( [ rate_eqs ] ) are invariant under the transformation @xmath283 as a consequence , the current - voltage characteristics @xmath284 and @xmath285 corresponding to the cases of the polaron level shifted by @xmath120 and @xmath286 , respectively , are related as follows : @xmath287 the presence of half - shuttle mechanism ( @xmath178 ) violates this symmetry . [ shuttle_cl025 ] shows @xmath1 for the case when @xmath288 is relatively small and where the polaron level is not shifted , both of which are favorable for pdc . the current steps are suppressed at negative voltages ( differential conductance peaks in the inset of fig . [ shuttle_cl025 ] ) , while at positive bias voltage , the current steps have a tendency to increase . from eqs . ( [ i_posv ] ) , ( [ i_negv ] ) we obtain that the current - step amplitude at @xmath201 corresponding to the current increase is given by @xmath289 in other words , the half - shuttle mechanism works in favor of the formation of the polaron state : the probability of phonon - assisted tunneling onto the dot from the left electrode is increased , while the phonon - assisted tunneling from the dot to the left electrode is decreased . ( [ isat_shuttle ] ) for the saturation currents illustrates this fact . the case of relatively large @xmath290 is shown on fig . [ shuttle_cl075 ] . like in the previous case , here we also observe suppression of the current steps at negative voltages . at positive voltages , the ndc steps become more pronounced compared to the pdc ones . expressions for the one - phonon assisted steps are given by eqs . ( [ deltai ] ) , ( [ deltai_negv ] ) . we have provided concrete evidence that the half - shuttle mechanism alone , with reasonable values of the parameters , can not produce ndc when @xmath119 is small and when the bare polaron level is not shifted upwards . nevertheless , based on the results of figs . [ shuttle_eneg ] and [ shuttle_epos ] , we conclude that in situation when ndc is present ( shifted level or reversed capacitances ) , the addition of the half - shuttle tends to emphasize the ndc features . to summarize , we have provided an in - depth study of ndc behavior in molecular quantum dots or transistors . while this study covers a wide range of experimental parameters such as capacitances and electron - phonon coupling , here the focus was put on the case where the tunneling coupling from the dot to the source and drain leads is asymmetrical . this choice was motivated by the experiment of leroy and coworkers@xcite where a nanotube is suspended over a trench , allowing nearly free vibrations of the phonon breathing mode . electrons flow from an stm tip to the suspended region of the nanotube , and their tunneling amplitude is much smaller than that of the nanotube to the substrate . it turns out that this assumption allows us to quantify ndc behavior by analytical means . yet , moderate asymmetries , often encountered with two poor metallic contacts , qualitatively display the same physics . we have used a microscopic approach modeling the molecule / nanotube as a quantum dot , which is justified based on the early observation of coulomb blockade behavior in carbon nanotubes . a single phonon mode couples on - site to the quantum dot , in two different way . first , as the principal mechanism , it couples to the electron density on the dot , as in a polaron model . second , the half - shuttle coupling was introduced , with the motivation that it should be present in the experiment of refs . : when the nanotube vibrates , the tunneling distance between the nanotube wall and the stm tip oscillates accordingly , but the amplitude of this motion is believed to be rather small ( or the order of the zero point motion , which can be smaller than @xmath291 ) . with these ingredients , the polaron transformation eliminates the electron charge coupling to the phonon on the dot , and transfers this coupling to the tunneling hamiltonian . at this point , we have chosen to describe the situation corresponding once again to the experiment of ref . , where the time scale associated with tunneling events between the dot and the leads is large compared to the temperature , so that electrons evacuated in the leads effectively loose their phase coherence . on the other hand , the quantum mechanical nature of electron - phonon dynamics within the dot ( polaron formation ) is fully captured . this happens to be the regime were most molecular electronics experiments are performed nowadays , as these experiments typically do not require dilution refrigerator technology . from the density matrix of the total system , a kinetic equation for the reduced density matrix ( with the leads degrees of freedom integrated out ) was derived , leading to master equations for the electron population and their associated phonon numbers . at temperature high enough for this rate equation is valid , but low enough that thermally activated tunneling from the phonon sideband can be neglected , analytical predictions were made for the current - voltage characteristics , thanks to our assumption of a highly asymmetrical molecule lead configuration : the positive or negative differential conductance behavior depends both on the location of the bare polaron level and on the occupation of the associated phonon sidebands above and below this level which are included in the bias voltage window . turning to numerics , attention was first drawn to the role of the relative capacitances to the dot . if the molecule / nanotube is in better contact with the lead which evacuates the electrons ( the substrate ) , than with the one who injects electrons , it is plausible to believe that the capacitance of this junction will be larger than that of the injecting junction . in this case , no negative differential conductance is obtained . when increasing the bias voltage , the contribution of an increasing number of phonon sidebands leads to a standard staircase behavior in the current - voltage characteristics . thermal effects tend to round off the first step . in the experiment of ref . , the use of several nanotube samples with different lengths allowed to vary the relative capacitance of the injecting and of the tunneling contact . this provided a motivation to study the so - called `` reversed capacitance '' case , which according to our analytical prediction could justify the presence of an negative differential behavior . this was indeed observed numerically , but it was emphasized that intermediate phonon couplings are needed to observe ndc : `` small '' electron - phonon coupling leads to barely noticeable ndc , while `` large '' electron - phonon coupling suppresses ndc altogether , thus the need to use non - perturbative techniques to study ndc . do these results disqualify the possibility of ndc if contacts where the capacitances between left and right are comparable ? the answer is no , although ndc could be more difficult to detect in this case . while the global shape of the current voltage characteristics displays the staircase structure typically attributed to pdc , the onset of each step is followed by a small spike , thus displaying ndc behavior . the amplitude of this peak can be optimized with phonon coupling . for slightly asymmetric capacitances , the same effects are observed , but the staircase structure is modified as it involves two distinct periodicities . because background charges may shift the bare polaron level , and because of the possibility of using a voltage gate on the molecule / dot we also studied the possibility of ndc in this case . this suggests that when the level is shifted upwards , ndc is possible even when the capacitance of the injecting junction is small . finally , we have proposed that a half - shuttle mechanism may play a role in a stm experiment . it can be detected by the asymmetry of the current - voltage curves . yet , in our opinion , this mechanism can not account for ndc alone and only tends to increase the heights of pdc and ndc steps in the current - voltage characteristics . notice that approaching the tip closer to the molecule should strongly enhance the half - shuttle mechanism . in our numerical calculations , we set @xmath292 assuming the amplitude of the zero - point rbm oscillations to be of the order of @xmath261pm . estimating the zero - point amplitude , we have considered a single - wall carbon nanotube as an elastic hollow cylinder with diameter @xmath293 nm , with wall thickness @xmath294 , and with a length corresponding to the relevant breathing part @xmath295 nm . accordingly , the undeformed volume of the cylinder is @xmath296 . the elastic properties of the nanotube cylinder are characterized with the average young s modulus @xmath297 tpa ( see , for instance , ref . ) . assuming @xmath298 mev ( @xmath77 is the frequency of the rbm ) , one finds for the relative extension of the nanotube due to the breathing , @xmath299 thus , the estimated diameter change , @xmath300 , associated with the zero - point oscillations is of the order of 1 pm . overall , we have provided a rather complete account on the possible occurrence of ndc behavior in molecular electronic transport , due to phonon - assisted transport . this phenomenon is due to a distribution of the total spectral weight of the injected electron on the polaron levels included in the bias window . taking into account the correct microscopic expressions of the franck - condon factors associated to phonon transitions , we have demonstrated that ndc effects are a fingerprint of the polaronic nature of charge carriers . while we can not rule out that ndc could exist in molecular system with symmetric molecular junction , it was not observed within this model . on the other hand , we have clearly shown that for asymmetric tunneling rates , there exist a wide range of parameters which lead to ndc . a striking result of this work is the fact that a simple counting of the ndc steps in the current voltage characteristic ( or alternatively , a counting of the peaks in differential conductance ) provides direct information about how many phonon subbands contribute to transport . the question remains about whether this ndc behavior can be observed in experiments . our model could apply to ref . , if the tunnel couplings are indeed asymmetric . concerning ref . , ndc occurs but only for the first phonon sideband . this can be due to several factors : first , as a general rule , the height of pdc or ndc steps tends to decrease with increasing phonon sidebands number . furthermore , if the electron phonon coupling is weak , ndc is present but its manifestation is weak . finite temperature tends to smear these steps , so in order to observe several peaks in ndc , lower temperatures would be required , keeping in mind that the tunneling contacts should not be too large in order to `` avoid '' any kondo like regime . another aspect which we have not considered in this work is the possibility of phonon damping . the population of the phonon subbands may rearrange due to the coupling with the environment of the molecule @xcite . the environment could include electron - hole pairs generated in the leads , which should be reduced because of our weak coupling assumption . on the other hand , for a `` large '' molecular system , the optical phonon mode ( the breathing mode of the nanotube of ref . ) is also accompanied by other phonon modes such as acoustic - like modes , which have typically lower energies , and which therefore can play the role of a `` bath '' . because of the importance of the occupation of the phonon subbands for ndc in our present work , we expect that substantial coupling to this bath will tend to suppress ndc . 999 h. park , j. park , a.k.l . lim , e. h. andersen , a.p . alivisatos , and p.l . mceuen , nature ( london ) * 407 * , 57 ( 2000 ) . zhitenev , h. meng , and z. bao , phys . * 88 * , 226801 ( 2002 ) . l.h . yu and d. natelson , nanolett . * 4 * , 79 ( 2004 ) ; l.h . keane , j.w . ciszek , l. cheng , m.p . stewart , j.m . tour , and d. natelson , phys . * 93 * , 266802 ( 2004 ) . qiu , g.v . nazin , and w. ho , phys . 92 * , 206102 ( 2004 ) . pasupathy , j. park , c. chang , a.v . soldatov , s. lebedkin , r.c . bialczak , j.e . grose , l.a.k . donev , j.p . sethna , d.c . ralph , and p.l . mceuen , nanoletters * 5 * , 203 ( 2005 ) . s. sapmaz , p. jarillo - herrero , ya.m . blanter , c. dekker , and h.s.j . van der zant , cond - mat/0508270 ( unpublished ) . b. j. leroy , s. g. lemay , j. kong , and c. dekker , nature * 432 * , 371 ( 2004 ) . b. j. leroy , j. kong , v. k. pahilwani , c. dekker , and s. g. lemay , phys . b * 72 * , 075413 ( 2005 ) . dresselhaus , and p.c . eklund , adv . phys . * 49 * , 705 ( 2000 ) . v. perebeinos , j. tersoff and p. avouris , phys . lett . * 94*,086802 ( 2005 ) . d. boese and n. schoeller , eur . . lett . * 54 * , 668 ( 2001 ) . u. lundin and h. mckenzie , phys . b * 66 * , 075303 ( 2002 ) . k. d. mccarthy , n. prokofev and m. t. tuominen , phys . b * 67 * , 245415 ( 2003 ) . v. aji , j.e . moore , and c.m . varma , int . j. nanosci . * 3 * , 255 ( 2004 ) . k. flensberg , phys . b * 68 * , 205323 ( 2003 ) ; s. braig and k. flensberg , phys . rev . b * 68 * , 205324 ( 2003 ) . a. mitra , i. aleiner , and a. j. millis , phys . b * 69 * , 245302 ( 2004 ) . j. koch and f. von oppen , phys . lett . * 94*,206804 ( 2005 ) ; j. koch , m. semmelhack , f. von oppen , and a. nitzan , cond - mat/0504095 ( unpublished ) . k. c. nowack and m. r. wegewijs , cond - mat/0506552 ( unpublished ) g.a . kaat and k. flensberg , phys . b * 71 * , 155408 ( 2005 ) . m.h . hettler , h. schoeller , and w. wenzel , europhys . lett . * 57 * , 571 ( 2002 ) . m. galperin , m.a . ratner and a. nitzan , * 5 * , 125 ( 2005 ) . s. datta , nanotechnology * 15 * , s433 ( 2004 ) . glazman and r.i . shekhter , zh . * 94 * , 292 ( 1988 ) [ sov . jetp * 67 * , 163 ( 1988 ) ] ; n.s . wingreen , k. w. jacobsen , and j. w. wilkins , phys . lett . * 61 * , 1396 ( 1988 ) . b * 40 * , 11834 ( 1989 ) . h. ness and a. j. fisher , phys . rev . lett . * 83 * , 452 ( 1999 ) . e.g. emberly and g. kirczenow , phys . b * 61 * , 5740 ( 2000 ) . j. bonca and s.a . trugman , phys . * 75 * , 2566 ( 1995 ) ; ibid . 79 * , 4874 ( 1997 ) . y. meir and n. s. wingreen , phys . lett . * 68 * , 2512 ( 1992 ) . jian - xin zhu and a. v. balatsky , phys . b * 67 * , 165326 ( 2003 ) . pasupathy , r.c . bialczak , j. martinek , j.e . grose , l.a.k . donev , p.l . mceuen , and d.c . ralph , science * 306 * , 86 ( 2004 ) . b. j. leroy , s. g. lemay , j. kong , and c. dekker , appl . 84 * , 4280 ( 2004 ) . l. y. gorelik , a. isacsson , m.v . voinova , b. kasemo , r.i . shekhter , and m. jonson , phys . lett . * 80 * , 4526 ( 1998 ) ; a. isacsson , l. y. gorelik , m.v . voinova , b. kasemo , r.i . shekhter , and m. jonson , physica b * 255 * , 150 ( 1998 ) . d. fedorets , l. y. gorelik , r.i . shekhter , and m. jonson , phys . lett . * 92 * , 166801 ( 2004 ) . l. m. jonsson , l. y. gorelik , r.i . shekhter , and m. jonson , nano lett . * 5 * , 1165 ( 2005 ) . a. erbe , r. h. blick , a. tilke , a. kriele , and j. p. kotthaus , appl . * 73 * , 3751 ( 1998 ) . a. e. hanna and m. tinkham , phys . b * 44 * , 5919 ( 1991 ) . a. zazunov , v. s. shumeiko , e. n. bratus , and g. wendin , phys . b * 71 * , 214505 ( 2005 ) . v. n. popov , v. e. van doren , and m. balkanski , phys . b * 61 * , 3078 ( 2000 ) .

recently , there have been some interesting developments involving off - forward parton distributions of the nucleon , deeply virtual compton scattering , and hard diffractive vector - meson production . these developments are triggered by the realization that the off - forward distributions contain information about the internal spin structure of the nucleon and that diffractive electroproduction of vector mesons depends on these unconventional distributions . this paper gives a brief overview of the recent developments .

hep-ph9807358

parton distributions were first introduced by feynman as the phenomenological quantities describing the properties of the nucleon manifest in high - energy scattering@xcite . when a nucleon travels near the speed of light , it can be viewed as consisting of a beam of non - interacting partons that are massless and collinear . parton distributions are the densities of partons as a function of the fraction @xmath0 of the nucleon momentum they carry . with the advent of quantum chromodynamics ( qcd ) , feynman s parton model , modulo logarithms , can be justified in a rigorous field theoretical approach @xcite . meanwhile , parton distributions can be expressed as the matrix elements of the bilocal light - ray operators in the ground state of the nucleon @xcite . only in the light - like gauge , @xmath1 , and light - cone quantization , does their physical interpretation become apparent . when studying the physics of parton distributions themselves , one of course can no longer regard the nucleon as consisting of free partons . calculating these distributions requires a detailed knowledge of the nucleon wave function in quark and gluon degrees of freedom , and undoubtedly is one of the most challenging problems in nonperturbative qcd . at present , the only fundamental approach we know is the lattice qcd method @xcite . on the phenomenological side , however , the unpolarized parton distributions have been extracted with good precision from various high - energy scattering data taken in the last thirty years @xcite . the empirical parton distributions not only provide the necessary input for calculating high - energy scattering cross sections , e.g. , of top - quark production , but also contain valuable information about how the nucleon is made of its constituents . recently , there has been much theoretical activity surrounding the so - called off - forward parton ( or non - forward , off - diagonal ) distributions ( ofpd s ) . the new distributions generalize those introduced by feynman : they characterize certain properties of the nucleon exhibited in a class of high - energy scattering ; they reflect the low - energy internal structure of the particle . however , there are also important differences between the off - forward and forward distributions . the ofpd s in general can not be regarded as particle densities , but rather their physical interpretation is given in terms of probability amplitude . as we shall see , they do have simple physical significance in light - cone coordinates ( or the infinite momentum frame ) . the most natural appearance of the ofpd s is in _ non - forward _ high - energy processes . as far as we know , such processes were first considered by watanabe @xcite and by bartels and m. loewe @xcite . in ref . @xcite , a general virtual - photon compton scattering off a nucleon target was studied in operator product expansion ( ope ) , an integral representation of the amplitude was obtained , and the off - forward matrix elements of twist - two operators were defined . the study , however , mentioned little about physical motivations and the practicality of the process ; in particular , significance of the integral representation was obscure . in ref . @xcite , diffractive production of @xmath2 and photons was studied in perturbative qcd and the properties of the non - diagonal gluon ladder were investigated . however , no notion of an off - forward distribution of the nucleon target was apparent in the paper . following @xcite , a set of off - forward evolution kernels were presented by gribov , levin , and ryskin @xcite . these kernels were considered as those for the scattering amplitude . the role of off - forward distributions in hard diffractive processes had remained obscured in relatively recent studies . in ref . @xcite , diffractive electroproduction of @xmath3 was analyzed by ryskin in the small @xmath0 region . the production amplitude was related to the usual feynman gluon distribution . the off - forwardness was taken into account by a phenomenological form factor . in refs . @xcite , ryskin s analysis was generalized to light vector - meson production in which the meson structure is accounted for by the leading - twist light - cone wave function . to the author s knowledge , one of first examples of ofpd s was introduced by dittes et al . @xcite , and was called `` interpolating function . '' ironically , the main motivation for introducing such an object was not from studying any off - forward process ; rather , it was considered as an example of the physical quantities whose evolution kernel interpolates between the two well - known cases : the dglap and erbl evolutions @xcite . a serious study of the ofpd in a physical process was made by d. mller et al . @xcite in which they showed that it contributes to the general two virtual - photon compton scattering the same process considered earlier by watanabe . they also derived the evolution equations for the ofpd . note that in ref . @xcite a general off - forward amplitude ( or single - particle green s function ) was introduced in studying hadron helicity flip in hard scattering . however , because the correlation of two quark fields in the object is off the light cone , it does not have a simple parton interpretation as the ofpd s do . the recent wave of interest in the ofpd s was generated from two new developments . the first development is the recognition of the importance of the ofpd s in describing the internal structure of the nucleon @xcite . in studying the spin structure , it was found that the fractions of the nucleon spin carried by quarks and gluons can be determined by the form factors of the corresponding energy - momentum tensors . these form factors can be obtained from the moments of the ofpd s which were independently introduced in @xcite . in the same paper , deeply virtual compton scattering ( dvcs ) was proposed as a practical means to measure the ofpd s . in dvcs , a highly virtual photon scatters off a quark in a nucleon target , which then emits a real photon and returns to form a recoil nucleon . although the two - photon process studied earlier by watanabe and mller et al . contains dvcs as its special kinematic limit , the latter has important experimental advantages and a special theoretical status . indeed , dvcs is more delicate to treat theoretically because of an extra light - like vector among the external momenta . the second development is the realization that hard diffractive electroproduction of vector mesons requires the use of ofpd s @xcite . the first serious studies in this direction were made by radyushkin @xcite , hoodbhoy @xcite , and collins et al . @xcite . in ref . @xcite , an off - forward gluon distribution - most important in the low @xmath0 region - was introduced and used to calculate vector meson production in deep - inelastic electron scattering . the result is free of the assumptions made in ref . @xcite about the dominance of the absorptive contribution and the use of the forward gluon distribution . hoodbhoy @xcite used the off - forward gluon distribution to calculate heavy quarkonium production investigated earlier by ryskin @xcite . he obtained a similar result as radyushkin s . in a comprehensive paper , collins , frankfurt , and strikman showed that a factorization theorem exists for general hard diffractive meson production in deep - inelastic scattering @xcite . the factorization theorem explicitly involves the ofpd s . recently , many works have been produced in studying off - forward parton distributions , deeply virtual compton scattering , and diffractive vector meson production . it is the goal of this paper to review these interesting developments . in section ii , off - forward distributions are introduced from different perspectives and various definitions in the literature are compared . in section iii , we make a digression to discuss the spin structure of the nucleon . the role of the ofpd s in this interesting subject is exposed . in section iv , we summarize our present knowledge about the ofpd s at hadron mass scales and their evolution to high - energy scales . in section v , we consider processes in which the ofpd s may actually be measured : deeply virtual compton scattering and hard diffractive vector - meson production . section vi contains the summary and outlook . since the off - forward distributions contain rich information about the nucleon structure , it is instructive to examine their definitions and interpretations from different angles . in this section , we will first introduce the off - forward distributions from the form factors of twist - two operators and then we explore their significance in light - cone coordinates . finally we compare the ofpd s with other equivalent definitions in the literature . from the point view of the low - energy nucleon structure , it is perhaps most interesting to consider the off - forward distributions as the generating functions for the form factors of the so - called twist - two operators . recall that the matrix elements of the electromagnetic current in the equal momentum states are entirely determined by symmetry , whereas those in the unequal momentum states define the ( dirac and pauli ) form factors which contain such interesting information as charge radius and magnetic moment of the nucleon . the following tower of twist - two operators represents a generalization of the electromagnetic current @xmath4 where all indices are symmetrized and traceless ( indicated by @xmath5 ) and @xmath6 technically , these operators transform like @xmath7 under lorentz transformations . our interest in such operators is not accidental . they appear , for instance , in the operator production expansion of the two electromangetic currents in deep - inelastic scattering @xcite . thus although these generalized currents do not couple directly to any known fundamental interactions , they can nonetheless be studied indirectly in hard scattering processes . since the operators for @xmath8 are not related to any symmetry in the qcd lagrangian , their matrix elements between the equal momentum states , @xmath9 contain valuable dynamical information about the internal structure of the nucleon . in this paper , we use the covariant normalization @xmath10 for the nucleon state . the @xmath11 dependence of the above matrix elements signifies the dependence on renormalization scale and scheme @xcite . the quark distribution @xmath12 introduced by feynman has a simple connection to the above matrix elements : @xmath13 where @xmath14 is chosen to have support in @xmath15 . for @xmath16 , @xmath12 is the just density of quarks which carry the @xmath0 fraction of the parent nucleon momentum . the density of antiquarks is customarily denoted as @xmath17 , which in the above notation is @xmath18 @xcite . just like the form factors of the electromagnetic current , additional information about nucleon structure can be found in the form factors of the twist - two operators when the matrix elements are taken between the states of unequal momenta . using lorentz symmetry and parity and time reversal invariance , one can write down all possible form factors of the spin-@xmath19 operator @xmath20 } a_{qn,2i}(t ) \delta^{\mu_2}\cdots \delta^{\mu_{2i+1 } } \overline{p}^{\mu_{2i+2}}\cdots \overline{p}^{\mu_n ) } \nonumber \\ & & + ~ { \overline u}(p'){\sigma^{(\mu_1\alpha } i\delta_\alpha \over 2m}u(p ) \sum_{i=0}^{[{n-1\over 2 } ] } b_{qn,2i}(t ) \delta^{\mu_2}\cdots \delta^{\mu_{2i+1 } } \overline{p}^{\mu_{2i+2}}\cdots \overline{p}^{\mu_n ) } \nonumber \\ & & + ~ c_{qn}(t ) ~{\rm mod}(n+1,2)~{1\over m}\bar u(p ' ) u(p ) \delta^{(\mu_1 } \cdots \delta^{\mu_n ) } \ , \label{form}\end{aligned}\ ] ] where @xmath21 and @xmath22 are the dirac spinors and @xmath23 is 1 when @xmath19 even , 0 when @xmath19 odd . the four - momentum transfer is denoted by @xmath24 and its invariant by @xmath25 . the average nucleon momentum @xmath26 is also a useful variable . for @xmath27 , even or odd , there are @xmath28 form factors . @xmath29 is present only when @xmath19 is even . as the forward matrix elements @xmath30 , the above form factors can be used to define a new type of parton distributions _ off - forward distributions_. to accomplish this , we introduce a light - like vector @xmath31 , which is conjugate to @xmath32 in the sense that @xmath33 . write @xmath34 , where @xmath35 and @xmath36 is another light - like vector . contracting both sides of eq . ( [ form ] ) with @xmath37 , we have @xmath38 where @xmath39 } a_{qn,2i}(t ) ( -2\xi)^{2i } + { \rm mod}(n+1,2)~ c_{qn}(t ) ( -2\xi)^n \ , \nonumber \\ e_{qn } ( \xi , t ) & = & \sum_{i=0}^{[{n-1\over 2 } ] } b_{qn,2i}(t ) ( -2\xi)^{2i } - { \rm mod}(n+1,2)~ c_{qn}(t ) ( -2\xi)^n \ . \end{aligned}\ ] ] here we have defined a new variable @xmath40 . clearly , the @xmath41 dependence of @xmath42 and @xmath43 helps to distinguish among the different form factors of the same operator . now one can introduce the off - forward parton distributions ( ofpd s ) @xmath44 and @xmath45 through their moments @xmath46 since all form factors are real , the new distributions are consequently real . moreover , because of time - reversal and hermiticity , they are _ even _ functions of @xmath41 . for simplicity , we have not shown here the renormalization scale dependence . frequently we will also omit the @xmath47 dependence . the new distributions are more complicated than the feynman parton distributions because of their dependence on the momentum transfer @xmath48 . as such , the ofpd s contain two more scalar variables besides the feynman variable @xmath0 . the variable @xmath47 is the usual @xmath47-channel invariant which is always present in a form factor . the @xmath41 variable is a natural product of marrying the concepts of the feynman distribution and form factor : the former requires the presence of a prefered momentum @xmath49 along which the partons are predominantly moving and the latter requires a four - momentum transfer @xmath48 ; @xmath41 is just a scalar product of the two momenta . some insights about the dependence of the ofpd s on these two extra variables will be offered later in section iv.a . in qcd , there are five additional towers of twist - two operators besides that in eq . ( [ o1 ] ) : @xmath50 the corresponding off - forward distributions can be labelled by @xmath51 , @xmath52 , @xmath53 , @xmath54 , @xmath55 , @xmath56 , @xmath57 , @xmath58 , and @xmath59 , @xmath60 , respectively . the physical significance of parton distributions in high - energy processes is apparent only in light - cone coordinates and light - cone gauge . to see this , we sum up all the local twist - two operators into a light - cone bilocal operator and express the parton distributions in terms of the latter , @xmath61 the light - cone bilocal operator ( or light - ray operator ) arises frequently in hard scattering processes in which partons propagate along the light - cone . in fact , the taylor - expansion of this operator along the light - cone leads us immediately to the twist - two operators considered in the last subsection . the parton distributions are most naturally defined in terms of the matrix elements of the bilocal operator . in this context , the feynman @xmath0 is just the fourier variable of the light - cone distance . the lorentz structures in the second line in the above equation are independent and complete . in ordinary coodinates , the correlation of two quark fields in the light - ray operator is in both space and time . as such , the physical interpretation is quite complicated . however , in light - cone coordinates , the physical content of the bilocal operator becomes transparent . to see this , let us recall that the light - cone coordinates are defined as , @xmath62 in this new system of coordinates , the role of time is played by @xmath63 . in the same fashion , introduce the light - cone dirac matrices @xmath64 and the projection operators , @xmath65 . a dirac field @xmath66 can be decomposed into a sum of @xmath67 , the independent degrees of freedom ( good " ) , and @xmath68 , the dependent degrees of freedom ( bad " ) . in light - cone quantization , the independent components of the dirac field obey the following commutation relation @xcite @xmath69_+|_{x^+=y^+ } = { 1\over \sqrt{2 } } p_{+\alpha\beta}\delta(x^--y^-)\delta^2(x_\perp - y_\perp ) \ , \ ] ] where @xmath70_+$ ] denotes anticommutation . the above relation can be solved by a fourier expansion @xmath71 and similarly for @xmath72 . the quark ( antiquark ) creation and annihilation operators , @xmath73 ( @xmath74 and @xmath75 ( @xmath76 , obey the commutation relation @xmath77_+ & = & 2k^+ \delta_{\lambda\lambda'}(2\pi)^3 \delta(k^+-k'^+ ) \delta^2(k_\perp - k_\perp ' ) \ , \nonumber \\ \left(~\left[d_\lambda(k^+,\vec{k}_\perp ) , d^\dagger_{\lambda'}({k'}^+ , \vec{k'}_\perp ) \right]_+ \right . & = & \left . 2k^+ \delta_{\lambda\lambda'}(2\pi)^3 \delta(k^+-k'^+ ) \delta^2(k_\perp - k_\perp ' ) \ . \right ) \end{aligned}\ ] ] in the presence of a gauge field , the above is unchanged if the light - cone gauge , @xmath1 , is chosen . the same gauge choice eliminates the gauge link in the light - ray operator . [ fig1 ] now we can look at the physical content of the off - forward distributions in the light - cone coordinates and gauge . since they are even in @xmath41 , we first restrict ourselves to @xmath78 . substituting the light - cone fock expansion into eq . ( [ string ] ) , we have @xmath79 where @xmath80 is a volume factor . the distribution has different physical interpretations in the three different regions . in the region @xmath81 , it is the amplitude for taking a quark of momentum @xmath82 out of the nucleon , changing its momentum to @xmath83 , and inserting it back to form a recoiled nucleon . in the region @xmath84 , it is the amplitude for taking out a quark and antiquark pair with momentum @xmath85 . finally , in the region @xmath86 , we have the same situation as in the first , except the quark is replaced by an antiquark . the first and third regions are similar to those present in the ordinary parton distributions and hence will be called the dglap region . the middle region is similar to that in a meson amplitude and hence will be called the erbl region . the @xmath41 symmetry is an important property of the ofpd s . to illustrate the parton content of this symmetry , one can work out the equivalent of eq . ( [ fock ] ) by assuming @xmath41 is negative . the result is best shown in fig . 1 , where the quark and antiquark creations and annihilations are indicated explicitly . from eq . ( [ fock ] ) , the off - forward distribution is very much like a quark - nucleon scattering amplitude . use @xmath87 to label the helicity amplitudes , where @xmath88 are the helicities of the initial and final nucleons and @xmath89 are those of the created and annihilated quarks . time - reversal and parity invariance provide the constraints @xmath90 and @xmath91 . therefore , there are total of six independent helicity amplitudes which imply four additional twist - two off - forward distributions @xcite @xmath92 } \over m } u(p ) \ . \end{aligned}\ ] ] their parton interpretation is similar to that of @xmath93 , except that for @xmath94 there is a sign depending on the quark helicity and for @xmath95 there is a quark helicity flip @xcite . off - forward gluon distributions can be introduced in the similar fashion . consider first the gluon - nucleon helicity amplitudes . symmetries limit the number of independent ones to six , and hence there are six leading - twist gluon distributions . here we quote only the helicity - independent ones : @xmath96 where the gauge link is not shown . because of the bose symmetry , @xmath97 . the interested reader can find the other twist - two gluon distributions in @xcite . in light - front coordinates , only @xmath98 are the independent degrees of freedom . they obey the commutation relation @xcite @xmath99|_{x^+=y^+ } = { 1\over 2 } \delta_{ij } \delta^2(x_\perp - y_\perp ) \epsilon(x^--y^-)\ ] ] where @xmath100 . the commutation relation can be solved by the following expansion @xmath101 where @xmath102 and @xmath103 are the gluon creation and annihilation operators . using this expansion , the physical content of the gluon distributions can be worked out immediately @xmath104 in the region @xmath81 , it is the amplitude for taking a gluon of momentum @xmath82 out of the nucleon , changing its momentum to @xmath83 , and inserting it back to form a recoiled nucleon . in the region @xmath105 , it is the amplitude for taking out a gluon pair with momentum @xmath85 . in an actual calculation of the distributions , the vacuum subtraction in matrix elements is always implied . as such , it is convenient to insert a @xmath106 product between the two fields at light - cone separations . the difference between a @xmath106 product and an ordinary product is just a constant @xmath107_+ \ .\ ] ] the constant does not contribute to any physical matrix element . therefore , a light - cone @xmath106 product contains only dispersive contributions . the notion of ofpd s we discussed above are essentially that of refs . @xcite and @xcite . the variables @xmath0 and @xmath41 here are the same as the variables @xmath47 and @xmath108 in @xcite . other names ( non - forward distributions , off - diagonal distributions , double distributions ) and notations exist in the literature . radyushkin introduced the _ non - forward distributions _ @xmath109 in refs . @xcite . to see their relationship to the ofpd s , one can write eq . ( [ string ] ) for an arbitrary light - like @xmath19 @xmath110 one can also shift the argument of the string operator using heisenberg s equations of motion . now if one chooses @xmath111 the variables are related according to @xmath112 the matrix element of the string operator becomes @xmath113 compared with eqs . ( 9.1 ) and ( 9.2 ) in ref . @xcite , we have @xmath114 the advantages of @xmath115 over @xmath109 are : 1 ) it treats the initial and final nucleon symmetrically and thus @xmath41 symmetry is obvious ; 2 ) it has a simple connection with local hermitian operators . however , the @xmath116 variable in @xmath109 has the advantage of measuring one of the initial parton momenta according to the longitudinal momentum of the initial nucleon , analogous to the forward case . collins et al . introduced the off - diagonal distributions @xmath117 @xcite . they also chose the @xmath19 vector conjugate to the initial state momentum . the relations of their variables to ours are , @xmath118 here @xmath119 is the same as radyuskin s @xmath116 . the off - diagonal distributions are related to the ofpd s by @xmath120 on the other hand , for gluons , one has @xmath121 again , the @xmath41 symmetry is not obvious in these off - diagonal distributions . radyushkin also introduced what he called double distributions @xmath122 , where @xmath0 and @xmath123 are the fractions of the initial nucleon momentum @xmath124 and the momentum transfer @xmath48 carried by the active parton @xcite . it is interesting from the point of view of parametrizing the momentum flow out of a hadron vertex in a @xmath41-independent way . however , the physics of a double distribution is equivalent to that of the corresponding off - forward distribution . one of the principal motivations for studying off - forward distributions is that they provide a new class of observables for the internal structure of the nucleon . we believe that the physics of strong interactions is described by the fundamental theory quantum chromodynamics ( qcd ) . however , qcd in the low - energy region is strongly interacting and highly relativistic , involving an infinite number of degrees of freedom . no theory of this type has ever been solved before with a clear physical insight . the only available tool now for solving qcd bound states is lattice qcd @xcite . at this stage , the approach has not yet been developed to a degree of sophistication so that one can reliably calculate fundamental observables with respectable precisions . therefore , any experimental information on the structure of the nucleon is valuable in helping to understand how nature constructed this particle of extreme importance to our existence . since gell - mann and zweig s quark model , many nucleon models have been proposed to correlate the experimental data from low and high energy probes . although these models are not precisely qcd , they have been quite successful phenomenologically . however , the ultimate question confronting models is how the degrees of freedom used in them are related to those in qcd . this question is of high importance because experimental probes do not couple to model degrees of freedom . in practice , simple assumptions have been used in comparing model predictions and data , e.g. , the constituent quarks are identified as qcd quarks at low - energy scales although a precise notation of the latter is unclear . these assumptions have been seriously challenged in 1987 when the european muon collaboration ( emc ) measured , with an unprecedented precision , the proton s spin - dependent structure function @xmath125 in polarized deep - inelastic scattering . combining their data with the hyperon beta decay rates , augmented with the assumption of the flavor su(3 ) symmetry , emc extracted the fraction of the nucleon spin carried in the spin of quarks : @xmath126 this result is in flat contradiction with the cherished quark model prediction @xmath127 . this so - called spin crisis " has touched off an explosive activity in both theoretical and experimental communities . after ten more years of careful experimental studies , the original emc result has essentially been confirmed , although the small @xmath0 contribution to @xmath128 remains largely uncertain . a recent global analysis of the data can be found , for instance , in ref . @xcite . many theoretical solutions " to the spin crisis " have been proposed in the literature . apart from doubts about the experimental data and the interpretation of them , there have been serious attempts in explaining the emc result from the fundamental theory . for instance , lattice qcd calculations have produced numbers close to the emc result @xcite , although the systematic effects of quenched approximation , finite lattice sizes , and large quark masses remain uncertain . despite these , more fundamental approaches have not yet offered a complete picture about the spin structure of the nucleon . as to the naive quark model prediction , the key question is what has been alluded to abovel : namely , whether the quark model result can be directly compared with the experimental data , or whether the constituent quarks are the same as the qcd quarks probed in high - energy scattering . no convincing argument has been offered towards a positive answer . more recently , progress has been made in attempting to understand how qcd builds up the nucleon spin . the progress potentially has an important impact on other theoretical studies , and ultimately on our basic knowledge of the internal structure of the nucleon . on the one hand , it suggests a thorough examination of the spin structure in a fundamental way . on the other hand , they called for understanding the role of the gluon degrees of freedom in model building . as it turns out , a key element in unraveling the nucleon spin is the off - forward parton distributions . the history began with identifying the sources of the qcd angular momentum . jaffe and manohar wrote down the complete angular momentum operator in a free - field - theory form which contains naturally the quark and gluon spin and orbital contributions @xcite . the importance of the quark orbital angular momentum in qcd has been recognized before by sehgal @xcite and ratcliffe @xcite in different contexts . the emc result indicates , among others , that a large fraction of the nucleon spin is carried by other sources of angular momentum . in order to associate these contributions to possible observables , the qcd angular momentum operator has been rewritten in a gauge - invariant form @xcite , @xmath129 where @xmath130 \ , \nonumber \\ \vec{j}_g & = & \int d^3x ~\vec{x } \times ( \vec{e } \times \vec{b } ) \ . \label{ang}\end{aligned}\ ] ] the quark and gluon parts of the angular momentum are generated from the quark and gluon momentum densities @xmath131 and @xmath132 , respectively . @xmath133 is the dirac spin - matrix and the corresponding term is clearly the quark spin contribution . @xmath134 is the covariant derivative and the associated term is the gauge - invariant quark orbital contribution . with the above expression , one can easily construct a sum rule for the spin of the nucleon . consider a nucleon moving in the @xmath135-direction , and polarized in the helicity eigenstate @xmath136 . the total helicity can be evaluated as an expectation value of @xmath137 in the nucleon state @xmath138 where the three terms denote the matrix elements of three parts of the angular momentum operator in eq . ( [ ang ] ) . the physical significance of each term is obvious , modulo the scale and scheme dependence indicated by @xmath11 . the scale dependence in @xmath139 is generated from the u(1 ) axial anomaly , and there have been attempts to remove the scale dependence by subtracting a gluon contribution @xcite . unfortunately , such a subtraction is by no means unique . here we adopt the standard definition of @xmath140 as the matrix element of the multiplicatively renormalized quark spin operator . note that the individual term in the above equation is independent of the momentum of the nucleon . in particular , it applies when the nucleon is travelling with the speed of light ( the infinite momentum frame ) @xcite . the scale dependence of the quark and gluon contributions can be calculated in perturbative qcd . by studying renormalization of the nonlocal operators , one can show @xcite @xmath141 as @xmath142 , there exists a fixed point solution @xmath143 thus as the nucleon is probed at an infinitely small distance scale , approximately one - half of the spin is carried by gluons . a similar result has been obtained by gross and wilczek in 1974 for the quark and gluon contributions to the momentum of the nucleon @xcite . strictly speaking , these results reveal little about the nonperturbative structure of bound states . however , experimentally it was found that about half of the nucleon momentum is carried by gluons already at quite low energy scales @xcite . thus the gluon degrees of freedom not only play a key role in perturbative qcd , but also are a major component of nonperturbative states , as they should be . an interesting question is then , how much of the nucleon spin is carried by the gluons at low energy scales ? a solid answer from the fundamental theory is not yet available . recently , balitsky and this author have made an estimate using the qcd sum rule approach @xcite . we found @xmath144 which yields approximately 0.25 . if this calculation indicates anything of the truth , the spin structure of the nucleon roughly looks like @xmath145 a quark model calculation of @xmath146 has also been made recently by barone et al . @xcite . the result also indicates @xmath147 at low - energy scales . by examining carefully the definition of @xmath148 @xmath149 one realizes that they can be extracted from the form factors of the quark and gluon parts of the qcd energy - momentum tensor @xmath150 . specializing eq . ( [ form ] ) to ( @xmath151 ) @xmath152 u(p)\ .\ ] ] taking the forward limit in the @xmath153 component and integrating over 3-space , one finds that @xmath154 give the momentum fractions of the nucleon carried by quarks and gluons ( @xmath155 ) . on the other hand , substituting the above into the nucleon matrix element of eq . ( [ matrix ] ) , one finds @xcite @xmath156 \ .\end{aligned}\ ] ] therefore , the matrix elements of the energy - momentum tensor provide the fractions of the nucleon spin carried by quarks and gluons . there is an analogy for this . if one knows the dirac and pauli form factors of the electromagnetic current , @xmath157 and @xmath158 , the magnetic moment of the nucleon , defined as the matrix element of ( 1/2)@xmath159 , is @xmath160 . since the quark and gluon energy - momentum tensors are just the twist - two , spin - two , parton helicity - independent operators , we immediately have the following sum rule from the off - forward distributions @xmath161 = a_q(t ) + b_q(t ) \ , \end{aligned}\ ] ] where the @xmath41 dependence , or @xmath162 contamination , drops out . extrapolating the sum rule to @xmath163 , the total quark ( and hence quark orbital ) contribution to the nucleon spin is obtained . a similar sum rule exists for gluons . thus a deep understanding of the spin structure of the nucleon may be achieved by measuring the ofpd s from high - energy experiments . comparing with the ordinary parton distributions , off - forward distributions depend on two additional variables : the @xmath47-channel longitudinal momentum fraction @xmath41 and the invariant mass @xmath47 . because of these , the ofpd s contain much more information and are necessarily more difficult to model in practice . at present , there is little experimental data to constrain these distributions . in this section , we will describe a few of the recent studies of the distributions at low - energy scales where they reflect directly the nonperturbative structure of the nucleon . we will also discuss their evolution equations which have been computed to leading logarithmic order in perturbation theory . some qualitative features of the evolution to an asymptotically large energy scale will be highlighted . a few rigorous results about the ofpd s are known . first of all , in the limit of @xmath164 and @xmath165 , they reduce to the ordinary parton distributions . for instance , @xmath166 where @xmath167 and @xmath168 are the unpolarized and polarized quark densities . similar equations hold for gluon distributions . for practical purposes , in the kinematic region where @xmath169 an off - forward distribution may be approximated by the corresponding forward one . the first condition , @xmath170 , is crucial - otherwise there is a significant form - factor suppression which can not be neglected at any @xmath0 and @xmath41 . for a given @xmath47 , @xmath41 is restricted to @xmath171 therefore , when @xmath172 is small , @xmath41 is automatically limited and there is in fact a large region of @xmath0 where the forward approximation holds . the first moments of the off - forward distributions are constrained by the form factors of the electromagnetic and axial currents . indeed , by integrating over @xmath0 , we have @xcite @xmath173 the @xmath47 dependence of the form factors are characterized by hadron mass scales . therefore , it is reasonable to speculate that similar mass scales control the @xmath47 dependence of the off - forward distributions . in the small @xmath0 region , it has been argued that the @xmath47 dependence corresponds to a mass scale @xmath174 or something close to the inverse slope of the elastic @xmath175 cross section , @xmath176 gev@xmath177 @xcite . martin and ryskin made a very important observation that one can obtain an upper bound on the off - forward distributions by using the schwarz inequality @xcite . for instance , consider the quark distribution in eq . ( [ fock ] ) and take a ket @xmath178 . since @xmath179 , we have the following constraint @xmath180 for @xmath81 , where @xmath181 and @xmath182 which is different from that in eq . ( [ x2 ] ) . similarly , one has @xmath183 for @xmath86 . in the middle region , however , the vacuum subtraction ruins a useful constraint . an upper bound for the off - forward gluon distribution can be derived in a similar way : @xmath184 for @xmath81 . because of the different definitions of @xmath185 , the above inequality is slightly different from martin and ryskin s . however , they are practically the same in the small @xmath0 region . it must be cautioned that the above inequalities are derived by disregarding renormalization of the distributions , and hence are correct in the leading - logarithmic approximation only . a first model calculation of the quark off - forward distributions was reported in ref . the calculation was done in the mit bag model , whose parameters are adjusted so that the electromagnetic form factor and the feynman parton distributions are best produced . the shape of the distributions as a function of @xmath0 are rather similar at different @xmath47 and @xmath41 : peaked around @xmath186 @xmath187 0.4 , and tapering off rather quickly as @xmath188 , 0 and negative , where the calculation can not be trusted . the @xmath47 dependence of the energy - momentum form factors is controlled by a mass parameter between 0.5 and 1 gev@xmath177 . the @xmath41 dependence of the distributions is surprisingly weak ; a clear physical reason is unknown . the off - forward quark distributions were also studied in the chiral quark - soliton model by petrov et al . in contrast to the bag model results , the chiral soliton model yields a rather strong @xmath41 dependence of the off - forward distribution . the model also predicts qualitatively different behaviors in the regions @xmath189 and @xmath190 , with discontinuities at @xmath191 . although the discontinuities are believed to be an artifact of the calculation , the qualitatively different features in different regions are in line with the interpretation of the ofpd s . from eq . ( [ fock ] ) , we understand that the off - forward distributions have similar physics as the forward distributions in @xmath189 . when @xmath190 , however , the distributions reflect @xmath47-channel exchanges of meson and glueball quantum numbers . therefore , a physics - oriented modelling in this region may consider the convolution of the meson cloud of a nucleon with the leading light - cone wave function of the mesons . models can not be used yet to calculate off - forward gluon distributions in the small @xmath0 region , which are phenomenologically important @xcite . it is believed in the literature that the distributions in this region and negligible @xmath47 are independent of @xmath41 @xcite . as we mentioned earlier , this might be correct as long as @xmath192 . for @xmath193 , however , the best argument so far is offered by martin and ryskin @xcite . from the diffractive scattering phenomenology , they argue that the distribution @xmath194 at small @xmath0 should be bounded below by the forward distribution at @xmath195 , @xmath196 combining this with the upper bound in eq . ( [ gmr ] ) , they concluded phenomenologically when @xmath197 , @xmath198 at @xmath199 gev@xmath177 for the grv gluon distribution @xcite and at @xmath200 gev@xmath177 for the mrs(r2 ) distribution @xcite . parton distributions are renormalization scale and scheme dependent because the defining operators are . thus , a natural starting point for discussing scale evolution is to consider renormalization of twist - two operators . however , it is well known that the parton evolution has a simple physical interpretation in light - cone coordinates and gauge , as exemplified by the dglap evolution kernels . therefore , we shall also consider the off - forward parton evolution kernels . renormalization of the twist - two operators appearing in feynman parton distributions is a special example of a more general case . for a given twist - two operator of spin @xmath19 shown in eq . ( [ o1 ] ) , apart from its mixing with the gluon operator of same quantum numbers it also mixes with the following operators with total derivatives , @xmath201 consideration of such mixing is already necessary in studying the evolution of the leading - twist meson wave functions . the answer was first obtained by efremov and radyushkin @xcite , and by brodsky and lepage @xcite . actually , at the leading logarithmic order the answer may be guessed from the naive conformal symmetry which is broken by quantum effects only at the next - to - leading - logarithmic order @xcite . the following combination of the twist - two operators furnishes a representation of the special conformal symmetry group , @xmath202 where the @xmath203 are the gegenbauer polynomials of order 3/2 . its evolution takes exactly the diagonal form of that for the operator without the total derivatives @xcite @xmath204 where @xmath205 \ , \ ] ] and @xmath206 . @xmath207 for the su(3 ) color group and @xmath208 is the number of light quark flavors . for the leading - twist pion wave function , the above result leads to a general expansion in terms of gegenbauer polynomials . generalization to the singlet case is in principle straightforward . for the twist - two operators with natural parity , the evolution was first worked out by chase @xcite . for those with unnatural parity , it was worked out by ohrndorf @xcite , shifman and vysotsky @xcite , and baier and grozin @xcite . the conclusion of these studies is that the gluon towers of operators evolve in the leading - logarithmic approximation exactly like the forward case if they are combined according to the gegenbauer polynomials of order 5/2 . these results were confirmed later in a more elegant form of light - ray or string operators by geyer et al . @xcite , and by balitsky and braun @xcite . in light of these developments , evolution of off - forward distributions can simply be worked out by converting the anomalous dimensions of the composite operators into evolution kernels . however , as in the case of the dglap evolution @xcite , one can obtain the same results more directly through studying parton splitting processes . a first such study for the nonforward , helicity - independent parton evolution was made by bartels and loewe in the small @xmath0 limit @xcite . gribov et al . presented the complete leading - order kernels including the singlet mixing @xcite . mller et al . found the non - singlet evolution kernel by generalizing the brodsky - lepage result @xcite . more recently , the issue has been reexamined from different angles by radyushkin @xcite , ji @xcite , chen @xcite , frankfurt et . @xcite , blmlein et al . @xcite , and martin and ryskin @xcite . as an example of the off - forward evolution , we cite below the helicity - independent result in ref . @xcite . according to eq . ( [ fock ] ) , it is natural that the distributions in three different kinematic regions evolve differently . in the region @xmath81 where we have a quark creation and annihilation , the evolution equation is , @xmath209 where @xmath210 \ .\ ] ] the parton splitting function is @xmath211 the end - point singularity is cancelled by the divergent integrals in @xmath212 . obviously , when @xmath213 , the splitting function becomes the usual dglap evolution kernel @xcite . for @xmath214 where we have creation or annihilation of a quark - antiquark pair , the evolution takes the form , @xmath215 f_{ns}(y,\xi , q^2 ) \ , \ ] ] where @xmath216 when @xmath217 , the kernel reduces to the brodsky and lepage kernel @xcite . for @xmath86 where we have antiquark creation and annihilation , the evolution takes the same form as eq . ( [ ap ] ) , apart from the replacement @xmath218 . the helicity - dependent off - forward distributions were introduced in ref . @xcite and their evolution was also studied in that paper . subsequently , the kernels have been confirmed by the studies of blmlein et al . @xcite and balitsky and radyushkin @xcite . the helicity - flip off - forward quark distributions were introduced by collins et al . @xcite , and their evolution was studied by belitsky and mller @xcite . the helicity - flip gluon distributions were defined by hoodbhoy and ji @xcite who also derived the leading - logarithmic evolution . it is worthwhile to point out that the combinations of the helicity - flip quark and gluon operators according to the gegenbauer polynomials of order 3/2 and 5/2 respectively also form representations of the naive conformal symmetry group and hence have diagonal evolutions at the leading - logarithmic level . as in the forward case , the off - forward parton evolution equations can be solved in many different ways . a few approaches have already been suggested in the literature @xcite . one important aspect of the evolution is the asymptotic form of the off - forward distributions . radyushkin found @xcite ( also @xcite ) for the non - singlet quark distributions , @xmath219 where @xmath220 is the appropriate electromagnetic form factor . this asymptotic form has interesting implications . when @xmath213 , the asymptotic form is a @xmath221-function in @xmath0 . this reminds us of the well - known fact that as @xmath222 , parton distributions migrate to smaller and smaller @xmath0 . for @xmath217 , the evolution is that of a meson wave function with an asymptotic limit @xmath223 , which has a broad shape in @xmath0 . therefore , it is quite plausible that for an intermediate @xmath41 , one would see the migration of the distribution from large to small @xmath224 if @xmath189 . however , in the middle region @xmath190 , the total strength increases and at the same time , the distribution evens out toward @xmath223 . these features of evolution were observed in the numerical calculations by belitsky et al . @xcite and mankiewicz et al . @xcite . the evolution effects for the off - forward gluon distribution at small @xmath0 are especially interesting from the phenomenological point of view . these effects have been studied by frankfurt et al . @xcite , and martin and ryskin @xcite . in ref . @xcite , they assumed negligible off - diagonal effects at @xmath225 gev , and then evolved the off - diagonal distribution to @xmath226 , 17 , 41 , and 110 gev at @xmath227 , and @xmath228 . for @xmath229 to 1 , they found that the ratio of off - diagonal to diagonal distributions ranges from 1.2 to 1.7 . they pointed out that for a fixed @xmath230 , the off - diagonal distribution at large @xmath231 is obtained from that in the much larger @xmath0 region , however , with the same difference @xmath230 where the off - diagonal effect is very small . therefore at very small @xmath0 , unless @xmath231 is very close to hadron mass scales , the off - diagonal effects come mostly from the perturbative evolution . these effects are less than a factor of 2 for the kinematics region considered . similar studies have also been made in ref . the qualitative conclusion is the same although quantitatively there are differences . martin and ryskin pointed out that the off - diagonal evolution happens faster than the diagonal one because the difference in the splitting functions is positive . furthermore , the difference is only important when @xmath231 is close to the final target scale and @xmath0 becomes comparable to the size of @xmath41 . eventually , usefulness of the ofpd s depends on whether they can actually be measured in any experiment . the simplest , and possibly the most promising , type of experiments are hard electro- or deep - inelastic production of photon , mesons , and perhaps even lepton pairs . in this section , we will consider separately two experiments : deeply virtual compton scattering ( dvcs ) in which a real photon is produced , and diffractive meson production . there are practical advantages and disadvantages from both processes . real photon production is , in a sense , cleaner but the cross section is reduced by an additional power of @xmath232 . moreover , the dvcs amplitude interferes with the bethe - heitler one . meson production may be easier to detect , however , it has a twist suppression , @xmath233 . in addition , the theoretical cross section depends on the unknown light - cone meson wave function . at present , there exist already some data from hera on meson production @xcite . more experiments will be done in the future at hera , cern , and jefferson lab . deeply virtual compton scattering was proposed in ref . @xcite as a practical way to measure the off - forward distributions . the result was confirmed in ref . @xcite using the double distributions . the general two - photon processes had actually been studied earlier by watanabe @xcite and mller et al . @xcite , and later by chen @xcite . recently , there have been a number of new results obtained about this process , which will be described below . consider the virtual photon scattering shown in fig . 2a , where the momenta of the incoming ( outgoing ) photon and nucleon are @xmath234 and @xmath235 , respectively . the compton amplitude is defined as @xmath236 where @xmath237 . in the bjorken limit , @xmath238 and @xmath239 and their ratio remains finite , the scattering is dominated by the single quark process shown in fig . 2b in which a quark absorbs the virtual photon , immediately radiates a real one , and falls back to form the recoiled nucleon . in the process , the initial and final photon helicities remain the same . the leading - order compton amplitude is @xcite @xmath240 where @xmath19 and @xmath36 are the conjugate light - cone vectors defined according to the collinear direction of @xmath241 and @xmath242 , and @xmath243 is the metric tensor in the transverse space . @xmath41 is related to the bjorken variable @xmath244 by @xmath245 . [ fig2 ] the same dvcs final state can also be produced through the bethe - heitler process in which the initial or final state electron ( or muon ) radiates a real photon and at the same time scatters elastically off the target nucleon . this process can be calculated as accurately as the data on the nucleon elastic form factors . physically , the bethe - heitler process is a background to dvcs and can overshadow the latter signal if the former is too large . on the other hand , with an appropriate size of bethe - heitler amplitude , one may hope to measure its interference with the dvcs amplitude which then can be directly extracted from data . the complete leading - order dvcs cross sections and the interference with the bethe - heitler amplitude can be found in ref . @xcite . recently , guichon made some interesting estimates of the cross sections at compass and jefferson lab energies @xcite . he found that while at low scattering energy , the bethe - heitler background is quite large , it drops significantly at high energies when the virtual photon flux is large . in fact , if one is focused on production of real photons along the direction of the virtual photon , the dvcs cross section is quite significant at compass energy , and certainly far above the bethe - heitler background . as far as studying the orbital angular momentum of quarks is concerned , this is the most interesting kinematic region . guichon s estimate is encouraging for a real experiment . the one - loop corrections to dvcs have recently been studied by osborne and ji@xcite . they have also been studied by mller @xcite and by belitsky and mller @xcite using the constraints from conformal symmetry . results of these studies have been confirmed by mankiewicz et al . @xcite . besides their obvious use in precision analysis , the results indicate that dvcs is factorizable at the next - to - leading order . a higher - order calculation of the coefficient functions in the renormalon - chain approximation has been done by belitsky and schfer @xcite . an all order proof of the dvcs factorization was first given by radyushkin @xcite and recently in a different perspective by osborne and ji @xcite , and collins and freund @xcite . the simple dvcs scattering mechanism can be tested through scaling relations and selection rules , as in the case of deep - inelastic scattering . in ref . @xcite , diehl et al . considered the possibility of using @xmath246 dependence of the cross section to test the single quark scattering picture , where @xmath246 is the angle between the lepton and hadron scattering planes . they observed that through qcd radiative corrections , the photon helicity - flip amplitude may contribute to dvcs . a recent study by hoodbhoy and ji showed that this indeed happens @xcite . the key element here is the helicity - flip off - forward gluon distributions , which decouple in the forward limit due to angular momentum conservation . they pointed out that a nonvanishing leading - twist photon helicity - flip amplitude signals unequivocally the presence of gluons in the nucleon . there are other processes that are as clean as the dvcs , but more difficult to measure experimentally . for instance , @xmath2 production was considered in ref . since @xmath2 s are signaled through lepton pairs , one can also consider lepton - pair production through intermediate time - like virtual photons . theoretical treatment of these is exactly the same as for dvcs . heavy quarkonium production was first considered by ryskin as a way to measure the feynman gluon distributions at small @xmath0 @xcite . a leading - order diagram for the process is shown in fig 3a , where the virtual photon fluctuates into a @xmath247 pair which subsequently scatters off the nucleon target through two - gluon exchanges . in the process , the pair transfers away certain amount of its longitudinal momentum and reduces its invariant mass to that of a @xmath248 . the cross section is , @xmath249 ^ 2\ ] ] where @xmath250 , @xmath251 is the @xmath3 mass , and @xmath252 is the decay width into the lepton pair . the equation was derived in the kinematic limit @xmath253 and the fermi motion of the quarks in the meson was neglected . two other important approximations were used in the derivation . first , the contribution from the real part of the amplitude is neglected , which may be justifiable at small @xmath0 . second , the off - forward distributions are identified with the forward ones . as we have discussed in iv.a , even for small @xmath47 , the approximation is good only if the off - diagonal evolution effects can be neglected . a more refined treatment of the process and comparison with hera data can be found in ref . @xcite . [ fig3 ] the above result was extended to the case of light vector - meson production by brodsky et al . , who considered the effects of meson structure in perturbative qcd @xcite . they found a similar cross section , @xmath254 where the dependence on the meson structure is in the parameter @xmath255 and @xmath256 is the leading - twist light - cone wave function . evidently , the above formula reduces to ryskin s in the heavy - quark limit @xmath257 . more discussions on vector meson production along this line , including comparison with recent data , can be found in ref . @xcite . radyushkin was the first to calculate the amplitude of hard diffractive electroproduction in terms of off - forward gluon distributions @xcite . with the virtual photon and vector meson both polarized longitudinally , his result corresponds to @xmath258 where again @xmath259 . the above formula is valid for any @xmath260 and @xmath47 smaller or around hadron mass scales . hoodbhoy has also studied the effects of the off - forward distributions in the case of @xmath3 production @xcite . he found that ryskin s result needs to be modified in a similar way once the off - forward effects becomes important . collins , frankfurt , and strikman have made an extensive analysis of the high - order qcd contributions to hard diffractive meson production @xcite . using the standard tools described in @xcite , they showed that the leading - twist contributions are factorizable to all orders in perturbation theory . the factorization theorem provides a rigorous qcd basis for calculating the process in terms of the off - forward distributions . in fact , the paper outlined a recipe for calculating a whole class of processes in perturbation theory . mankiewicz , piller and weigl recently calculated the neutral pseudoscalar and vector meson production cross sections in terms of the non - forward quark and gluon distributions @xcite . shown in fig . 3b is a feynman diagram in which meson production depends on a quark distribution . they also investigated charged particle production in which the initial nucleon has a different charge from the final one @xcite . the charge - changing off - forward distributions can be obtained from the charge - conserving ones using isospin symmetry . the result can be generalized to the octet baryons if the su(3 ) flavor symmetry is good . one can also consider production of heavy - flavored baryons from the nucleon target . unlike the heavy - quark distributions in light hadrons , the corresponding off - forward distributions are not suppressed by heavy - quark masses . the off - forward parton distributions are a new type of nucleon observables which have recently generated much interest in hadron physics community . on the one hand , the ofpd s contain all the form factors of the twist - two operators which generalize the well - known electroweak currents . the form factors of the spin - two , twist - two operators ( energy - momentum tensor ) are especially interesting in unravelling the spin structure of the nucleon . on the other hand , they summarize succinctly the leading structural dependence in a class of high - energy processes involving the nucleon . the partonic interpretation of the ofpd s is quite simple it represents parton - nucleon scattering amplitudes on the light - cone . recent studies of the ofpd s can largely be classified into three categories : modelling the new distributions , studying their scale dependences , and investigating processes that are sensitive to them . in terms of model studies , our accomplishments so far are rather limited . even though there are bag and chiral soliton model calculations , these calculations need much improvement in the future . in particular , the physics in the @xmath190 region has not been entirely captured in these calculations . the models are yet to be extended to including gluon degrees of freedom which are known to be important even at low energy scales . indeed , a vanishing helicity - flip gluon distribution at low - energy scales will remain zero at other scales because it has no mixing with any quark distribution . a number of rigorous constraints discussed in section iv.a can be helpful in modelling : apart from the special limits and @xmath261 moments , there exist useful upper bounds for the ofpd s . there have been many studies in perturbative evolution . at the leading - logarithmic level , evolution equations now have been obtained for all leading - twist ofpd s . at next - to - leading order , one needs two - loop anomalous dimensions which exist only for the spin - independent non - singlet operators @xcite . calculations of additional two - loop anomalous dimensions will be needed as data on the ofpd s become available . recent works of mller and belitsky illustrate the power of exploiting the anomalous conformal symmetry to obtain these anomalous dimensions @xcite . numerical studies of evolution effects have just begun . qualitatively we can understand the evolution effects from its asymptotic behavior . quantitatively , one needs to establish high - precision , high - speed evolution codes which allow efficient data fitting in the future . the leading - order cross sections for dvcs and meson production can be found in the literature . for dvcs , the cross section has also been computed to the next - to - leading order in @xmath262 . a similar study for meson production can be made , and will be necessary once precision data become available . higher - twist effects are important at low @xmath263 and must be investigated thoroughly . generally speaking , they are more complex to study than that in deep - inelastic scattering because of the extra photon or meson , and the extra soft scale @xmath47 . additional processes may be explored to probe the ofpd s , including ones in which the final state baryon may contain strange or heavy flavors . polarization observables need to be seriously investigated . a solid experimental confirmation of the scattering mechanisms involving the ofpd s is itself very significant . it involves testing scaling relations and selection rules . finally , we notice that the ofpd s are functions of multi - variables . as such , it would be difficult to get a complete picture of them in a short time period and in a single type of experiments . moreover , experimental cross sections can not be used to extract the distributions directly ; they usually involve integrals over parton momenta . therefore , it will be necessary to parametrize the ofpd s and fit the parameters to data . in finding appropriate ways to make parametrizations and fits , model studies will play important role , unless of course lattice calculations can make a significant breakthrough in the near future . * note added : * after this paper was submitted for publication , the author has noticed some new works in the literature which are relevant to the subject of this paper . in ref . @xcite , radyushkin studied both hadron form factors and wide - angle compton scattering using nonforward parton densities . in ref . @xcite , diffractive dijet photoproduction was proposed as a probe of the off - diagonal gluon distribution . in ref . @xcite , certain positivity constraints for off - forward parton distributions were derived . in a different line of development , the orbital angular momentum distribution in the nucleon was investigated @xcite . in the last two years i have benefited considerably from discussions surrounding the topic of this article with s. brodsky , j. collins , p. hoodbhoy , a. mueller , j. osborne , a. radyushkin , g. sterman , m. strikman , and others , and i wish to thank them all . i also would like to thank a. martin for his encouragement in writing this topical review , k. golec - biernat , a. radyushkin and l. robinette for a critical reading of the manuscript , and j. osborne for drawing the figures . this work is supported in part by funds provided by the u.s . department of energy ( d.o.e . ) under cooperative agreement doe - fg02 - 93er-40762 . dittes and a. v. radyushkin , phys . lett . * 134b * ( 1984 ) 359 ; m. h. sarmadi , _ ibid . _ * 143b * ( 1984 ) 471 ; g. r. katz , phys . rev . * d31 * ( 1985 ) 652 ; s. v. mikhailov and a. v. radyushkin , nucl . b254 * ( 1985 ) 89 .

motivated by recent experimental progress in preparing encapsulated graphene sheets with ultra - high mobilities up to room temperature , we present a theoretical study of dc transport in doped graphene in the hydrodynamic regime . by using the continuity and navier - stokes equations , we demonstrate analytically that measurements of non - local resistances in multi - terminal hall bar devices can be used to extract the hydrodynamic shear viscosity of the two - dimensional ( 2d ) electron liquid in graphene . we also discuss how to probe the viscosity - dominated hydrodynamic transport regime by scanning probe potentiometry and magnetometry . our approach enables measurements of the viscosity of any 2d electron liquid in the hydrodynamic transport regime .

1508.00363

transport in systems with many particles ( such as gases and liquids ) undergoing very frequent inter - particle collisions has been studied for more than two centuries and is described by the theory of hydrodynamics @xcite . in the hydrodynamic regime , transport is described by @xcite three non - linear partial - differential equations the continuity , navier - stokes , and energy - transport equations reflecting the conservation of mass , momentum and energy , respectively . the navier - stokes equation contains two transport coefficients @xcite : the shear viscosity , @xmath0 , which describes friction between adjacent layers of fluid moving with different velocities , and the bulk viscosity , @xmath1 , which describes dissipation arising in a liquid due to homogeneous compression - like deformations . the energy transport equation contains the thermal conductivity @xmath2 , which describes dissipative heat flow between regions with different temperatures . these coefficients quantify the tendency of the liquid to restore a homogeneous state in response to a velocity or thermal gradient : they therefore control the magnitude of non - local contributions to the linear - response functions of the liquid . viscous flow of dilute classical gases attracted the attention of maxwell , who theoretically discovered @xcite a puzzling property of the shear viscosity of a dilute gas . using a molecular approach @xcite , he found that the shear viscosity of a dilute gas is independent of density @xmath3 ( and depends on temperature @xmath4 according to @xmath5 ) , a counterintuitive result that he felt needed immediate experimental testing @xcite . the importance of @xmath0 in the hydrodynamic behavior of dilute gases and liquids stems from the fact that this parameter controls the term of the navier - stokes equation that opposes turbulent flow @xcite . recent years have witnessed a tremendous interdisciplinary interest in the hydrodynamic flow of strongly interacting quantum fluids . this interest was sparked by a series of results @xcite , which were obtained via the anti - de sitter / conformal field theory ( ads / cft ) correspondence , for the shear viscosity of a large class of strongly interacting thermal quantum field theories . these efforts culminated in 2005 when it was conjectured @xcite that all quantum fluids obey the following universal lower bound : @xmath6 , where @xmath7 is the entropy density . note that this bound does not contain the speed of light , thereby explaining why the conjecture was extended also to non - relativistic quantum field theories . fluids that saturate this bound have been dubbed nearly perfect fluids " ( npfs ) @xcite , i.e. fluids that dissipate the smallest possible amount of energy and satisfy the laws of hydrodynamics at distances as short as the inter - particle spacing . currently , two laboratory systems come closest to saturating the ads / cft bound : i ) the quark - gluon plasma @xcite , which is created at brookhaven s relativistic heavy ion collider and at cern s large hadron collider by bashing heavy ( e.g. gold and lead ) ions together and ii ) ultracold atomic fermi gases @xcite ( such as @xmath8 ) close to a feshbach resonance . although mathematical counterexamples have appeared in the literature @xcite , there are no known experimental violations of the ads / cft bound . the present work is motivated by the following questions : do electron liquids display hydrodynamic behavior ? if so , how can it be experimentally proven that the electron system has entered the hydrodynamic transport regime ? once in the hydrodynamic regime , how can the shear viscosity of an electron liquid be measured in a solid - state device ? can an electron liquid in a solid - state device be a npf ? hydrodynamics has been used for a long time to describe transport of electrons in solid - state devices @xcite . however , since the ( bloch ) momentum of an electron in a solid is a poorly conserved quantity due to collisions against impurities , phonons , and structural defects in the crystal , experimental signatures of hydrodynamic electron flow are expected only in ultra - clean crystals , at sufficiently low temperatures . second , electron - electron ( e - e ) interactions need to be sufficiently strong to ensure that the mean free path @xmath9 for e - e collisions is the shortest length scale in the problem , i.e. @xmath10 . here , @xmath11 is the fermi velocity , @xmath12 is the mean free path for momentum - non - conserving collisions , @xmath13 is the sample size , and @xmath14 the frequency of the external perturbation . unfortunately , the low - temperature requirement ( needed to mitigate the breakdown of momentum conservation in a solid - state environment ) and the strong e - e interaction requirement are conflicting : at low temperatures , where @xmath12 is large , the mean free path @xmath9 for e - e collisions is also very large due to pauli blocking . indeed , normal fermi liquids at low temperatures @xcite display very large values of @xmath9 , i.e. @xmath15 for temperatures @xmath16 , with @xmath17 the fermi temperature . ( in two spatial dimensions there is a very well known @xcite logarithmic correction that has been dropped . ) these severe restrictions , imposed by a rigid fermi surface on the phase space for two - body e - e collisions , can be relieved by increasing temperature . indeed , @xmath9 quickly decreases for increasing @xmath4 . short e - e mean free paths therefore require to operate at sufficiently elevated temperatures . at such temperatures , strong scattering of electrons against optical phonons ( e.g. in polar cystals such as gaas ) often leads to the unwanted inequality @xmath18 . realizing hydrodynamic flow at `` high '' temperatures therefore requires not only ultra - clean crystals but also crystals where electron - phonon coupling is extremely weak . we note that it is in principle easier to reach the hydrodynamic transport regime in non - polar crystals such as graphene , where the dominating mechanism at large temperatures is scattering of electrons against acoustic phonons . in this case @xmath12 decays as @xmath19 , i.e. slower than @xmath9 . for these reasons , evidence of hydrodynamic transport in solid - state devices is , to the best of our knowledge , limited to early work carried out by molenkamp and de jong @xcite in electrostatically defined wires in gaas / algaas heterostructures . these authors measured a non - monotonic dependence of the four - point longitudinal resistivity @xmath20 on the electronic temperature @xmath4 , which is increased above the lattice temperature by using a large dc heating current . the decrease of @xmath20 with increasing temperature above a certain value of @xmath4 was attributed to e - e collisions @xcite . this is the so - called `` gurzhi effect '' and will be discussed extensively in this article . more recently , indirect evidence of hydrodynamic flow comes from an explanation @xcite of coulomb drag between two neutral graphene sheets @xcite , which differs from that offered by the authors of ref . . recent experimental progress @xcite , however , has made it possible to fabricate samples with ultra - high carrier transport mobilities _ up to room temperature_. these are graphene sheets encapsulated between thin hexagonal boron nitride ( hbn ) slabs , which display ultra - high mobilities reaching @xmath21 in a wide range of temperatures up to room temperature . these values can be achieved in gaas / algaas heterostructures only below @xmath22 because of polar phonon scattering @xcite . in addition , the finite electron mass and moderate doping required to achieve high mobilities limits the fermi energy to values below a few tens of mev , which makes the fermi level smearing important even at liquid nitrogen temperatures . encapsulated samples @xcite are special because electrons roaming in graphene suffer very weak scattering against acoustic phonons @xcite and because hbn provides an exceptionally clean and flat dielectric environment for graphene @xcite . furthermore , microscopic calculations based on many - body diagrammatic perturbation theory @xcite indicate that the e - e mean free path @xmath9 in graphene is shorter than @xmath23 in a wide range of carrier densities and temperatures @xmath24 . we therefore conclude that hbn / graphene / hbn stacks are ideal samples where the long - sought hydrodynamic regime can be unveiled and explored . indeed , recent non - local transport measurements @xcite carried out in high - quality encapsulated single - layer ( slg ) and bilayer ( blg ) graphene samples have demonstrated that this is the case . the authors of ref . have reported evidence of hydrodynamic transport , showing that doped graphene exhibits an anomalous ( negative ) voltage drop near current injection points , which has been attributed to the formation of whirlpools in the electron flow . from measurements of non - local signals , bandurin _ et al . _ @xcite extracted the viscosity of graphene s electron liquid and found it to be in quantitative agreement with many - body theory calculations @xcite . in this article , we present a fully - analytical theoretical study of non - local dc transport in the two - dimensional ( 2d ) electron liquid in a graphene sheet in the hydrodynamic regime . in sect . [ sect : linearizedtheory ] we present the theoretical framework that was used in ref . to interpret the experimental results , i.e. a linearized steady - state hydrodynamic approach based on the continuity and navier - stokes equations . suitable boundary conditions for these hydrodynamic equations are discussed in sect . [ subsect : bcs ] . in sect . [ sect : longitudinal ] we present analytical solutions for longitudinal transport in a rectangular hall bar and we discuss the dependence of the solutions on the boundary conditions , providing details on the gurzhi effect . in sect . [ sect : nonlocal ] we present analytical results for the spatial depedence of the 2d electrical potential , non - local resistance , and current - induced magnetic field , as obtained by solving the hydrodynamic equations with the free - surface boundary conditions , which we believe to be the appropriate boundary conditions for the linear - response regime . finally , in sect . [ sect : summary ] we present a summary of our main results and offer some perspectives . we remark that , in the present work , we focus only on doped slg and blg sheets , where the applicability of the fermi - liquid theory is granted . however , it is believed that the hydrodynamic behavior of the semimetals is particularly interesting @xcite when these are in the charge neutral state . in this case the fermi surface shrinks to a point and fermi liquid theory is not applicable . for example , the authors of ref . have found that the ratio @xmath25 for the 2d electron liquid in a graphene sheet at the charge neutrality point comes close to saturating the ads / cft bound . in nearly neutral semimetals @xmath9 is also short due to frequent collisions between thermally excited carriers ( @xmath26 ) . it is , however , very well known @xcite that any theory at the charge neutrality point must take into account the spatially inhomogeneous pattern of electron - hole puddles created by disorder @xcite . in the regime of sizeable doping we consider , we can safely ignore puddles in encapsulated graphene devices @xcite . we consider a two - dimensional electron liquid in a doped slg or blg sheet , deep in the hydrodynamic transport regime ( @xmath27 ) . for the sake of definiteness , we consider the hall bar geometry sketched in fig . [ fig : device ] . since the energy - momentum dispersion of electrons in these systems is particle - hole symmetric @xcite , we assume , without loss of generality , that the sample hosts a back gate - controlled equilibrium electron density equal to @xmath28 . ( the charge density is @xmath29 , @xmath30 being the electron charge . ) we neglect thermally - excited carriers and coupling between charge and heat flow @xcite , which is strong only at the charge neutrality point . finally , we consider the linear response regime and steady - state transport . in this framework of approximations , the hydrodynamic transport equations @xcite for the 2d electron liquid greatly simplify and reduce to @xmath31 and @xmath32 in eqs . ( [ eq : continuitylinearised ] ) and ( [ eq : navierstokeslinearised ] ) we have introduced the linearized steady - state particle current density @xmath33 , where @xmath34 is the linearized steady - state fluid - element velocity . ( [ eq : continuitylinearised ] ) is the continuity equation , while eq . ( [ eq : navierstokeslinearised ] ) is the navier - stokes equation . the latter contains three forces acting on a fluid element : i ) the electric force @xmath35 , written in terms of the electric potential @xmath36 in the 2d plane where electrons are moving , which is generated by the steady - state charge distribution @xmath37 in response to the drive current @xmath38 ; ii ) the internal force due to the shear viscosity @xmath39 of the 2d electron liquid , here written in terms of the kinematic viscosity @xcite @xmath40 and iii ) friction exerted on a fluid element by agents external to the electron liquid such as phonons and impurities , which dissipate the fluid - element momentum at a rate @xmath41 . the latter is a phenomenological parameter , which depends on @xmath28 and @xmath4 and is commonly used in modelling transport in semiconductor devices @xcite . in eqs . ( [ eq : navierstokeslinearised ] ) and ( [ eq : nudefinition ] ) @xmath42 is a suitable effective mass defined by : @xmath43 where @xmath44 is the 2d massless dirac fermion cyclotron mass @xcite , @xmath45 being the fermi wave number and @xmath46 the fermi velocity , and @xmath47 is the bare electron mass in vacuum . ( color online ) schematic representation of the non - local transport setup analyzed in this work . a dc current @xmath38 is injected ( red arrow ) into an encapsulated graphene hall bar of width @xmath13 . current injection occurs at a lateral contact located at @xmath48 and @xmath49 . the same current is drained ( blue arrow ) at a contact located at @xmath50 and @xmath49 . measurements of voltage drops @xmath51 `` near '' the current injection region are sensitive to the kinematic viscosity @xmath52 of the two - dimensional massless dirac fermion liquid . the notion of `` vicinity '' between voltage probe and current injector is defined by a crucial length scale , i.e. the vorticity diffusion length @xmath53 . here @xmath54 ( exceeding @xmath55 in high - quality encapsulated devices ) represents a phenomenological scattering time due to momentum - non - conserving collisions of a fluid element ( and _ not _ of single electrons ) . ] multiplying both members of eq . ( [ eq : navierstokeslinearised ] ) by @xmath54 , we obtain @xmath56 in eq . ( [ eq : navierstokeslinearised_2 ] ) we have introduced the following characteristic length scale of the problem : @xmath57 for @xmath58 ( as in high - quality hbn / graphene / hbn samples ) and @xmath59 ( see ref . ) we obtain @xmath60 . the physical significance of @xmath61 can be understood as follows . we first note that we can rewrite @xmath62 by using the following identity : @xmath63 - \nabla \times [ \nabla \times { \bm j}({\bm r})]~.\ ] ] because of ( [ eq : continuitylinearised ] ) , we can drop the first term on the right - hand side of eq . ( [ eq : identity ] ) . the second term is finite and related to the vorticity @xcite @xmath64 which in 2d is oriented along the @xmath65 axis . we can then rewrite eq . ( [ eq : navierstokeslinearised_2 ] ) as following @xmath66 taking the curl of eq . ( [ eq : intermediate ] ) and using the identities @xmath67 and @xmath68 ( the latter being valid because @xmath69 s only non - vanishing component is along @xmath65 , while @xmath70 acts only on the 2d @xmath71-@xmath72 plane ) , we obtain a damped - diffusion equation for the vorticity @xmath73 we therefore see that @xmath61 plays the role of a diffusion length for @xmath74 . in eq . ( [ eq : navierstokeslinearised_2 ] ) we have also introduced a `` drude - like '' conductivity , @xmath75 since we are in the hydrodynamic regime , @xmath76 should not be confused @xcite with the ordinary dc conductivity in the diffusive transport regime : once again , @xmath77 represents a phenomenological parameter that should be fit to experimental data , as we discuss below in sect . [ sect : longitudinal ] . this nave description of momentum - non - conserving collisions in the hydrodynamic transport regime can be relaxed by following similar arguments to those in ref . : this is however well beyond the scope of the present article and will be the topic of future studies . in the absence of viscosity , eq . ( [ eq : navierstokeslinearised ] ) reduces to a local version of ohm s law , i.e. @xmath78 . finally , we note that taking the divergence of eq . ( [ eq : navierstokeslinearised_2 ] ) and making use of eq . ( [ eq : continuitylinearised ] ) we obtain the laplace equation @xmath79 for the electric potential @xmath36 on the 2d plane . this should not be confused with the usual three - dimensional ( 3d ) poisson equation for the 3d electrostatic potential @xmath80 , @xmath81 the 2d potential in eq . ( [ eq : navierstokeslinearised ] ) is @xmath82 . on the right - hand side of eq . ( [ eq : poisson ] ) we note the steady - state charge density distribution @xmath83 which occurs in the sample in response to the drive current @xmath38 . ( [ eq : poisson ] ) needs to be solved in 3d space with suitable boundary conditions depending on the dielectric environment , gates , etc . surrounding the graphene sheet if one is interested in determining @xmath37 . in this article we will focus our attention on @xmath84 and @xmath36 . ( [ eq : continuitylinearised ] ) and ( [ eq : navierstokeslinearised_2 ] ) will be used to describe transport in the hall bar geometry pictorially represented in fig . [ fig : device ] . mathematically , it is convenient to work in a hall bar of infinite length in the longitudinal direction @xmath71 , since this allows us to use the fourier transform to solve the equations of motion see sect . [ sect : nonlocal ] . the width @xmath13 of the hall bar will be kept finite . in the next section we will describe a crucially important ingredient of the theory : boundary conditions . in order to find @xmath36 and @xmath84 in the hall bar geometry depicted in fig . [ fig : device ] , we need to solve eqs . ( [ eq : continuitylinearised ] ) and ( [ eq : navierstokeslinearised_2 ] ) in the rectangle @xmath85 $ ] , with appropriate boundary conditions ( bcs ) at the edges , i.e. at @xmath86 . lateral electrodes acting as current injectors / collectors are described through bcs on the component of the current _ perpendicular _ to the edges : @xmath87 here @xmath88 is a function that describes a distribution of current injectors and collectors on the upper ( lower ) edge of the multi - terminal hall bar . it is through eq . ( [ eq : normalbc ] ) that the total drive current @xmath38 injected into the system at the boundaries enters the problem . following abanin _ et al . _ @xcite , we model the electrodes as point - like ( i.e. delta - function ) sources and sinks . ( a more realistic modeling of electrodes has been carried out in ref . , where finite - width effects and metallic boundary conditions at extended electrodes have been taken into account in a fully numerical solution of eqs . ( [ eq : continuitylinearised ] ) and ( [ eq : navierstokeslinearised_2 ] ) . such details have essentially no impact on the physics we are going to highlight below . ) for example , for the setup depicted in fig . [ fig : device ] with a current injector at @xmath89 , a current collector at @xmath90 , and no injectors / collectors on the upper edge , we will use : @xmath91 and @xmath92 . in the presence of a finite shear viscosity @xmath52 , we need an additional bc on the _ tangential _ component of the current at the top ( @xmath93 ) and bottom ( @xmath94 ) edges of the hall bar . we use the following bc : @xmath95_{y = \pm w/2}= \mp \frac{j_x(x , y = \pm w/2)}{l_{\rm b}}~,\ ] ] where @xmath96 is a `` boundary slip length '' , i.e. a length scale describing friction at the physical boundaries of the sample . this bc can be explained as following . the left - hand side of eq . ( [ eq : generalbc ] ) is proportional to the off - diagonal component of the stress tensor @xcite , calculated at the edges of the hall bar . it represents the tangential component of the frictional force exerted by the boundaries of the hall bar on the 2d electron liquid @xcite . this force depends on the tangential velocity of the 2d electron liquid and boundary roughness : in the linear - response regime , it is natural to replace such unknown dependence with a linear law characterized by the single parameter @xmath96 , as in the right - hand side of eq . ( [ eq : generalbc ] ) . in the description of transport of molecular liquids in constrained geometries , like water in a pipe , where the interactions between the molecules of the fluid and the walls of the container are of the same nature as of those between molecules of the fluid , the most used bcs are the so - called `` no - slip '' bcs @xcite , in which the component of the current tangential to the boundary vanishes . the no - slip bcs can be obtained from eq . ( [ eq : generalbc ] ) by taking the limit @xmath97 . in the opposite limit of a free - surface geometry , like the surface of water in an open bucket , the tangential force applied from the boundary to the fluid element vanishes at the boundary . these `` free - surface '' bcs @xcite can be obtained from eq . ( [ eq : generalbc ] ) by taking the limit @xmath98 . which of these bcs should be used to model the experiments in ref . will become clear at the end of sect . [ sect : longitudinal ] . the validity of the linearized navier - stokes equation ( [ eq : navierstokeslinearised_2 ] ) relies on the smallness of the reynolds number @xcite @xmath99 . this is a dimensionless parameter ( which depends on the sample geometry ) that controls the smallness of the non - linear term @xmath100{\bm v}({\bm r},t)$ ] in the convective derivative with respect to the viscous term . in our case we can define the reynolds number as following : @xmath101 { \bm v}({\bm r},t)}{\nu \nabla^2 { \bm v}({\bm r},t)}\right| \simeq \frac{\bar{v}w}{\nu } = \frac{i}{e { \bar n } \nu } \equiv { \cal r}_w~,\ ] ] where @xmath102 is the typical value of the fluid - element velocity . for an injected current @xcite @xmath103 , a hall bar width @xmath104 , and an equilibrium density @xmath105 , we obtain @xmath106 . we note that @xmath107 is much smaller than the graphene fermi velocity @xmath108 and the flow is therefore `` non - relativistic . '' the corresponding value of the reynolds number is @xmath109 , obtained by using a kinematic viscosity @xmath110 of the 2d electron liquid in graphene @xcite . our linearized theory in eqs . ( [ eq : continuitylinearised ] ) and ( [ eq : navierstokeslinearised_2 ] ) is therefore fully justified . we first consider the situation in which no current is injected or extracted laterally at the hall bar edges , i.e. @xmath111 . in this case the local current @xmath84 does not depend on the longitudinal coordinate @xmath112 and all the spatial derivatives with respect to @xmath112 in eqs . ( [ eq : continuitylinearised ] ) , ( [ eq : navierstokeslinearised_2 ] ) , and ( [ eq : generalbc ] ) vanish . the continuity equation implies that @xmath113 does not depend on @xmath114 and vanishes identically because of eq . ( [ eq : normalbc ] ) . therefore also the @xmath114 component of the electric field must vanish . the @xmath112 component of the current respects the following equation : @xmath115 , where @xmath116 is the electric field . note that @xmath117 can not depend on @xmath114 because @xmath118 vanishes and @xmath119 . the solution of this equation that fulfils the bc ( [ eq : generalbc ] ) is @xmath120~,\ ] ] where we have introduced the length @xmath121 we can calculate the total longitudinal current @xmath38 carried by the flow by integrating eq . ( [ eq : solutionuniform ] ) in the transverse direction , i.e. @xmath122 where we have defined the dimensionless quantity @xmath123 measuring the longitudinal potential drop @xmath51 between two lateral contacts at positions @xmath112 and @xmath124 yields a four - point longitudinal conductivity @xmath125 of the form : @xmath126 eq . ( [ eq : viscousconductancecorrections ] ) is the most important result of this section . fig02a(2,140)(a ) fig02b(2,140)(b ) in the limit @xmath127 ( i.e. free - surface bcs ) @xmath128 . for this choice of bcs the longitudinal conductivity @xmath125 depends only on the rate of momentum - non - conserving collisions @xmath129 ( through @xmath76 ) and is independent of @xmath52 . on the other hand , in the limit @xmath97 ( i.e. no - slip bcs ) eq . ( [ eq : viscousconductancecorrections ] ) reduces to @xmath130~.\ ] ] we can easily understand two asymptotic limits of eq . ( [ eq : sigmanoslip ] ) . in the limit @xmath131 eq . ( [ eq : sigmanoslip ] ) yields @xmath132 : the small correction to the drude - like conductivity @xmath76 is due to a reduction of the fluid - element velocity in a thin region of width @xmath61 near the top and bottom edges of the hall bar . in the opposite limit , @xmath133 , we obtain @xmath134 . in this limit the problem is equivalent to that of poiseuille flow in a pipe @xcite , with a velocity profile @xmath135 that depends quadratically on the transverse coordinate @xmath114 and a resistance that is entirely due to viscosity . a summary of our main results for longitudinal electron transport in the presence of a finite viscosity is reported in fig . [ fig : longitudinal ] . we now would like to make a remark on the temperature dependence of @xmath125 in eq . ( [ eq : viscousconductancecorrections ] ) . for the sake of simplicity , we assume that @xmath96 does not depend on temperature . we observe that the derivative of @xmath125 with respect to @xmath4 , @xmath136 is the sum of two contributions with opposite signs . the first term on the right - hand side of eq . ( [ eq : t_dependence ] ) is negative , because @xmath137 and @xmath138 . the latter inequality holds because the scattering rate @xmath129 describing momentum non - conserving collisions is a monotonically increasing function of temperature @xcite . on the contrary , the second term is positive , because the @xmath139 and @xmath140 . the vorticity diffusion length @xmath61 decreases with increasing temperature because both @xmath52 and @xmath54 are decreasing functions of @xmath4 . we therefore conclude that , due to viscosity , @xmath125 ( @xmath20 ) can increase ( decrease ) upon increasing temperature . this is the so - called gurzhi effect @xcite . the existence of this effect relies crucially on the nature of bcs that are used to solve the hydrodynamic equations . in particular , it disappears for free - surface bcs . all previous experimental studies of transport in graphene and other 2d electron liquids we are aware of have reported monotonic temperature dependencies ( i.e. no evidence of the gurzhi effect ) in the ordinary longitudinal geometry in the linear - response regime . we therefore conclude that free - surface bcs are the most appropriate for a weak driving current @xmath38 . in this case , @xmath125 depends only on the unknown damping rate @xmath129 , which can therefore be determined from an ordinary four - point longitudinal transport measurement at every value of @xmath28 and @xmath4 , i.e. @xmath141 . in the next section , we will discuss another hydrodynamic phenomenon occurring in 2d electron liquids , i.e. the formation of whirlpools in electron flow @xcite , yielding a clear - cut experimental signal of hydrodynamic transport in weakly non - local linear - response transport measurements . since the experimental data in ref . do not show any gurzhi effect in the linear - response regime , in the next section we will utilize only the free - surface bcs ( @xmath127 ) . whirlpools in hydrodynamic electron flow , however , do exist also when no - slip bcs are used @xcite . in this sense whirpools are a much more robust phenomenon that the gurzhi effect in longitudinal transport . whirpools are also more dramatic in experimental appearance . we now present the solution of the problem posed by eqs . ( [ eq : continuitylinearised ] ) , ( [ eq : navierstokeslinearised_2 ] ) , ( [ eq : normalbc ] ) , and ( [ eq : generalbc ] ) in the rectangle @xmath85 $ ] . we use free - surface boundary conditions , corresponding to @xmath142 . to this end , we introduce the fourier transform with respect to the longitudinal coordinate @xmath112 in the three equations ( [ eq : continuitylinearised ] ) and ( [ eq : navierstokeslinearised_2 ] ) ( the latter for the two components @xmath143 and @xmath113 ) . the fourier transform of a function @xmath144 will be denoted by @xmath145 . these equations can be grouped into a linear system of three second - order ordinary differential equations ( odes ) with respect to the independent variable @xmath114 . it is convenient to rewrite this system in terms of four first - order odes . we find @xmath146 eq . ( [ eq : system ] ) can be solved by diagonalizing the @xmath147 matrix on the right - hand side , which has four distinct eigenvalues : @xmath148 and @xmath149 . the general solution will therefore be a linear combination of exponentials of the form @xmath150 where @xmath151 ( @xmath152 ) are the eigenvectors ( eigenvalues ) of the matrix . the four unknown coefficients @xmath153 can be found by enforcing the desidered bcs . these are found by taking the fourier transform of eqs . ( [ eq : normalbc ] ) and ( [ eq : generalbc ] ) with respect to @xmath112 : @xmath154 and @xmath155_{y = \pm w/2}=0~.\ ] ] the solution reads as follows : @xmath156~,\ ] ] @xmath157-\frac{2d_{\nu}^2}{w^2 } \partial_y { \hat}{f}_{3\alpha}\left(kw,\frac{y}{w},\frac{d_{\nu}}{w}\right)\right\}~,\ ] ] and @xmath158+\frac{2d_{\nu}^2}{w^2 } ik { \hat}{f}_{3\alpha}\left(kw,\frac{y}{w},\frac{d_{\nu}}{w}\right)\right\}~.\ ] ] in writing eqs . ( [ eq : potential_infinite_strip])-([eq : current_y_infinite_strip ] ) we have introduced the following functions of dimensionless arguments : @xmath159~,\ ] ] @xmath160~,\ ] ] and @xmath161~.\end{aligned}\ ] ] eqs . ( [ eq : potential_infinite_strip])-([eq : f3 ] ) are the most important results of this article . in general , it is not an easy task to inverse fourier transform eqs . ( [ eq : potential_infinite_strip])-([eq : current_y_infinite_strip ] ) to real space , after the functions @xmath162 have been specified . indeed , this requires to calculate a convolution which involves the bcs and the functions @xmath163 , @xmath164 , and @xmath165 in real space . we now introduce the inverse fourier transforms of the functions @xmath166 , @xmath167 , and @xmath168 , which read : @xmath169 \mp \frac{1}{4 \pi } \ln \left [ 1 + e^{-4 \pi |{\tilde{x}}|}+ 2 \cos ( 2 \pi { \tilde{y } } ) e^{-2 \pi |{\tilde{x}}| } \right ] \mp \frac{|{\tilde{x}}|}{2}~,\ ] ] and @xmath170e^{-2\pi |{\tilde{x}}|}\right\ } \nonumber\\ & & \pm 2 \pi e^{-2 \pi |{\tilde{x}}|}\left[\cos(2 \pi { \tilde{y}})(1 + e^{-4 \pi |{\tilde{x}}|})+2 e^{-2 \pi |{\tilde{x}}|}\right]\bigg\ } \times \left[1+e^{-4 \pi |{\tilde{x}}|}+2\cos(2 \pi { \tilde{y } } ) e^{-2 \pi |{\tilde{x}}|}\right]^{-2}~.\end{aligned}\ ] ] the functions @xmath165 do not have simple expressions in terms of elementary functions but can be cast in the form of exponentially converging series : @xmath171 e^{-|{\tilde{x}}|\sqrt{\lambda^{-2}+\pi^2 ( 2\ell+1)^2}}\nonumber\\ & \mp & \sum_{\ell=1}^{\infty } ( 2\ell)(-1)^{\ell } \sin(2\ell \pi { \tilde{y } } ) e^{-|{\tilde{x}}|\sqrt{\lambda^{-2}+\pi^2 ( 2\ell)^2 } } \bigg\}~.\end{aligned}\ ] ] .[tab : asymptoticbehaviours ] explicit expressions of the functions @xmath172 defined in the main text , evaluated at @xmath173 . we also summarize useful asymptotic behaviors in the limit @xmath174 and @xmath175 . similar expressions can also be obtained for the quantities @xmath176 by noting that @xmath177 . [ cols="^,^,^,^,^ " , ] in the following we will make use of a number of asymptotic behaviors of the functions @xmath163 , @xmath178 , and @xmath179 , which have been listed for the sake of convenience in table [ tab : asymptoticbehaviours ] . the task of calculating the potential and currents in real space simplifies substantially if the currents @xmath88 in eq . ( [ eq : normalbc ] ) can be represented by the sum of a finite number of delta - functions in real space . let us focus on the geometry of fig . [ fig : device ] , where the fourier transform of the bcs ( [ eq : bcpointlike ] ) reads @xmath180 and @xmath181 . in this case , we find that the steady - state current pattern can be written as @xmath182 where the potential @xmath36 and vorticity @xmath183 are given by : @xmath184\bigg\}\nonumber\\\end{aligned}\ ] ] and @xmath185~.\end{aligned}\ ] ] in eqs . ( [ eq : solutionsdevicegeometrypotential ] ) and ( [ eq : solutionsdevicegeometryvorticity ] ) we have introduce the shorthand @xmath186 , with @xmath187 ( @xmath188 ) representing the lateral separation between the observation point and the collector ( injector ) . from eq . ( [ eq : solutiondecomposition ] ) we clearly notice an important feature of the solution , i.e. for vanishing viscosity the current flow is irrotational . more precisely , the viscosity plays a twofold role : it modifies the irrotational contribution due to the electric potential @xmath36 _ and _ introduces a finite vorticity . it is noteworthy that these effects yield independent experimental signatures : the modification of the electrical potential can be detected by monitoring the resistances in a non - local configuration ( or by carrying out scanning probe potentiometry ) , while the vorticity generates a magnetic field , which can be detected by scanning probe magnetometry . these two effects are discussed in detail in the following sections . illustrative results for the spatial map of the 2d electrical potential @xmath36eq . ( [ eq : solutionsdevicegeometrypotential])and the charge current pattern @xmath189eq . ( [ eq : solutiondecomposition])are shown in fig . [ fig : nonlocal ] . for typical values of the drive current @xmath38 and conductivity , i.e. @xmath190 through a submicron constriction and @xmath191 , we find that the scale over which the 2d electrical potential changes is @xmath192 . we clearly see that in the case @xmath193panels ( b ) and ( c ) in fig . [ fig : nonlocal]whirlpools with a spatial extension @xmath194 develop in the spatial current pattern @xmath189 , to the right of the current injector and to the left of the current collector . once again , the spatial variations of the 2d electrical potential @xmath36 are amenable to experimental studies based on scanning probe potentiometry . in passing , we note that near the current injector at @xmath89 the potential is dominated by the singular parts of the functions @xmath172 . taking the limit @xmath195 in eq . ( [ eq : solutionsdevicegeometrypotential ] ) we find an extremely simple expression for the potential near the injector : @xmath196~,\ ] ] where @xmath197 is the distance from the injection point , @xmath198 is the angle measured from the injection direction @xmath72 , and @xmath199 is a length determined by bcs far from the contact . note that changing @xmath199 is equivalent to changing @xmath200 by an arbitrary additive constant . if dc transport is to be used as the main tool to detect hydrodynamic electron flow , it is pivotal to understand the spatial dependence of the non - local resistance @xmath201 , which we define in the following way : @xmath202 where the quantity @xmath203 does not depend on @xmath114 . because of ( [ eq : solutionsdevicegeometrypotential ] ) , we find that , at each point in space , @xmath204 is a quadratic function of @xmath61 : @xmath205 where @xmath206\ ] ] and @xmath207 to make contact with ref . , we now introduce the `` vicinity '' resistance , which is the non - local resistance measured on the edge where current is injected , at a distance @xmath208 from the current injector : @xmath209 using eqs . ( [ eq : lambdadependence ] ) , ( [ eq : a - expression ] ) , ( [ eq : b - expression ] ) , the asymptotic results in table [ tab : asymptoticbehaviours ] , and taking the limit @xmath210 , we find @xmath211~.\nonumber\\\end{aligned}\ ] ] here @xmath212 is the heaviside step function . for positive @xmath208 the two terms in eq . ( [ eq : vicinity - expression ] ) have opposite sign . for this reason , @xmath213 is expected to change sign as a function of @xmath61 . the change of sign of the vicinity resistance is a key signature of the viscous contribution to the electric potential . maximum sensitivity to viscosity is achieved when slightly non - local or `` vicinity '' voltage drops are measured outside the region @xmath214 $ ] where the current flux is maximum . the vicinity resistance ( [ eq : vicinity - expression ] ) rapidly decays for @xmath215 : it is therefore pivotal to measure @xcite the potential @xmath216 for a lateral separation @xmath208 from the current injection point which is of the order of the vorticity diffusion length @xmath61 . once the rate of momentum - non - conserving collisions @xmath217 is measured from an ordinary four - point longitudinal measurement of @xmath125 ( as explained in sect . [ sect : longitudinal ] ) , a measurement of the vicinity resistance @xmath213 yields a map @xmath218 of the kinematic viscosity of the 2d electron liquid . we hasten to stress that our all - electrical non - local protocol to measure the kinematic viscosity @xmath218 applies to any 2d electron liquid driven into the hydrodynamic transport regime ( and not only to doped slg and blg ) . for the sake of completeness , we note that the authors of ref . have proposed an ac corbino disk viscometer , which allows a determination of the hydrodynamic shear viscosity from the dc potential difference that arises between the inner and the outer edge of the disk in response to an _ oscillating _ magnetic flux . fig03a(2,65)(a ) fig03b(2,65)(b ) fig03c(2,65)(c ) because the steady - state current @xmath219 generates a magnetic field in the proximity of 2d electron system , whirlpools and viscosity - dominated hydrodynamic transport can also be detected by scanning probe magnetometry ( see , for example , refs . ) . as shown in appendix [ appendix : lia ] , in a sample in which a backgate is placed at a distance @xmath220 below the graphene sheet with @xmath221 , a local relation exists between the @xmath65-component @xmath222 of the magnetic field and the vorticity @xmath223 . in si units , this relation reads as following @xmath224 where @xmath225 is the magnetic constant and the vorticity has been introduced earlier in eq . ( [ eq : solutionsdevicegeometryvorticity ] ) . a typical 2d spatial map of @xmath222 is shown in fig . [ fig : magneticfield ] for different values of the vorticity diffusion constant @xmath61 . in this figure the magnetic field is plotted in units of @xmath226 . for @xmath227 , @xmath104 , and @xmath228 , we find @xmath229 . this value is well within reach of current technology @xcite . fig04a(2,65)(a ) fig04b(2,65)(b ) in summary , we have presented a theoretical study of dc transport in graphene driven into the hydrodynamic regime . as highlighted in ref . , this regime occurs in a wide range of temperatures and carrier concentrations . our theory , which applies only to the doped regime , relies on the continuity ( [ eq : continuitylinearised ] ) and navier - stokes ( [ eq : navierstokeslinearised_2 ] ) equations , augmented by suitable boundary conditions at hall bar edges . we have demonstrated analytically that a combination of ordinary four - point longitudinal transport measurements and measurements of non - local resistances in hall bar devices can be used to extract the hydrodynamic shear viscosity of the two - dimensional electron liquid in graphene @xcite . we have also discussed how to probe the viscosity - dominated hydrodynamic transport regime by scanning probe methods . indeed , we believe that it possible to observe hydrodynamic electron flow with spatial resolution by using available scanning probe potentiometry and magnetometry setups . spatial maps of the two - dimensional electrical potential @xmath36 and current - induced magnetic field @xmath230 for experimentally relevant parameters are shown in figs . [ fig : nonlocal ] and [ fig : magneticfield ] . we wish to emphasize that our theoretical approach is immediately applicable to any 2d electron liquid in the hydrodynamic transport regime . in the future , we plan to extend our theoretical approach to semimetals with linear and quadratic band touchings at the charge neutrality point , where viscosity is expected to be very low @xcite and the reynolds number ( [ eq : reynoldsnumber ] ) is expected to be very large , possibly enabling the observation of electronic turbulence . this will require to deal with thermally - excited carriers , coupling between charge and heat flow , and non - linear terms in the navier - stokes equation . we also would like to gain a fully microscopic understanding of momentum - non - conserving collisions in the hydrodynamic transport regime by treating smooth scalar _ and _ vector potentials due to disorder ( strain @xcite , charged impurities @xcite , etc . ) along the lines of what was done by the authors of ref . for the case of a smooth scalar potential . last but not least , we strongly believe that hydrodynamic flow and the shear viscosity of 2d electron liquids can also be accessed @xcite by scattering - type near - field optical spectroscopy ( see , for example , ref . and references therein to earlier work ) in the terahertz spectral range , since this technique measures the non - local conductivity @xmath231 . terahertz radiation is required a ) to make sure that the hydrodynamic inequality @xmath232 is satisfied ( with @xmath233 ) and b ) to have measurable non - local effects due to viscosity , since the latter decreases quickly @xcite as a function of the external probe frequency @xmath14 . in this appendix we present a derivation of eq . ( [ eq : lia ] ) . we use si units and the coulomb gauge @xmath234 for the vector potential . we assume that a bottom gate positioned at @xmath220 exists below the graphene sheet . this will be treated as a perfect conductor . the vector potential is related to the steady - state current pattern by the following 3d poisson equation : @xmath235 this is similar to the poisson equation in eq . ( [ eq : poisson ] ) for the scalar potential @xmath236 . the general solution of eq . ( [ eq : poissonvectorq ] ) is : @xmath240 the quantities @xmath241 must obey the condition @xmath242 to enforce the coulomb gauge and must be determined from bcs . requiring @xmath243 implies that @xmath244 must vanish . the corresponding @xmath245 component of the magnetic field is given by : @xmath246_z~.\ ] ] since @xmath247 must vanish at the gate position , i.e. for @xmath220 , we find : @xmath248 . in deriving the previous result we have assumed that @xmath249 . the fourier transform of the vector potential is therefore given by : @xmath250 now , if @xmath251 and @xmath252 are small with respect to the lateral lengthscales @xmath13 and @xmath61 over which the steady - state current pattern @xmath84 changes in the sample , i.e. @xmath253 , the above formula can be approximated for @xmath254 as : @xmath255 making use of eq . ( [ eq : bz ] ) and transforming back to real space we finally obtain the desired result , @xmath256_z = -e \mu_0 \bar{n}d \omega(\bm{r})~.\ ] ] this work was partially supported by miur through the program `` progetti premiali 2012 '' - project `` abnanotech '' . free software ( www.gnu.org , www.python.org ) was used . we thank m.f . crommie , l.s . levitov , l.a . ponomarenko , g. vignale , and a. yacoby for fruitful discussions . landau and e.m . lifshitz , _ course of theoretical physics : fluid mechanics _ ( pergamon , new york , 1987 ) . g.k . batchelor , _ an introduction to fluid dynamics _ ( cambridge university press , cambridge , 1967 ) . k. huang , _ statistical mechanics _ ( wiley , new york , 1987 ) . maxwell , http://www.jstor.org/stable/108948[philos . london * 156 * , 249 ( 1866 ) ] . for a review see e.g. d.t . son and a.o . starinets , http://dx.doi.org/10.1146/annurev.nucl.57.090506.123120[annu . nucl . part . sci . * 57 * , 95 ( 2007 ) ] . kovtun , d.t . son , and a.o . starinets , http://dx.doi.org/10.1103/physrevlett.94.111601[phys . lett . * 94 * , 111601 ( 2005 ) ] . t. schfer and d. teaney , http://dx.doi.org/10.1088/0034-4885/72/12/126001[rep . * 72 * , 126001 ( 2009 ) ] . jacak and b. mller , http://dx.doi.org/10.1126/science.1215901[science * 337 * , 310 ( 2012 ) ] . c. cao , e. elliott , j. joseph , h. wu , j. petricka , t. schfer , and j.e . thomas , http://dx.doi.org/10.1126/science.1195219[science * 331 * , 58 ( 2011 ) ] . e. elliott , j.a . joseph , and j.e . thomas , http://dx.doi.org/10.1103/physrevlett.113.020406[phys . * 113 * , 020406 ( 2014 ) ] . s. cremonini , http://dx.doi.org/10.1142/s0217984911027315[mod . b * 25 * , 1867 ( 2011 ) ] . r.n . gurzhi , sov . uspekhi * 11 * , 255 ( 1968 ) . m. dyakonov and m. shur , http://dx.doi.org/10.1103/physrevlett.71.2465[phys . rev . * 71 * , 2465 ( 1993 ) ] . a.o . govorov and j.j . heremans , http://dx.doi.org/10.1103/physrevlett.92.026803[phys . lett . * 92 * , 026803 ( 2004 ) ] . n. sai , m. zwolak , g. vignale , and m. di ventra , http://dx.doi.org/10.1103/physrevlett.94.186810[phys . * 94 * , 186810 ( 2005 ) ] . m. mller and s. sachdev , http://dx.doi.org/10.1103/physrevb.78.115419[phys . b * 78 * , 115419 ( 2008 ) ] . m. mller , j. schmalian , and l. fritz , http://dx.doi.org/10.1103/physrevlett.103.025301[phys . lett . * 103 * , 025301 ( 2009 ) ] . r. bistritzer and a.h . macdonald , http://dx.doi.org/10.1103/physrevb.80.085109[phys . rev . b * 80 * , 085109 ( 2009 ) ] . andreev , s.a . kivelson , and b. spivak , http://dx.doi.org/10.1103/physrevlett.106.256804[phys . 106 * , 256804 ( 2011 ) ] . m. mendoza , h.j . herrmann , and s. succi , http://dx.doi.org/10.1103/physrevlett.106.156601 [ * 106 * , 156601 ( 2011 ) ] . d. svintsov , v. vyurkov , s. yurchenko , t. otsuji , and v. ryzhii , http://dx.doi.org/10.1063/1.4705382[j . . phys . * 111 * , 083715 ( 2012 ) ] . m. mendoza , h.j . herrmann , and s. succi , http://dx.doi.org/10.1038/srep01052[sci . rep . * 3 * , 1052 ( 2013 ) ] . a. tomadin and m. polini , http://dx.doi.org/10.1103/physrevb.88.205426 [ * 88 * , 205426 ( 2013 ) ] . a. tomadin , g. vignale , and m. polini , http://dx.doi.org/10.1103/physrevlett.113.235901[phys . lett . * 113 * , 235901 ( 2014 ) ] . narozhny , i.v . gornyi , m. titov , m. schtt , and a.d . mirlin , http://dx.doi.org/10.1103/physrevb.91.035414[phys . b * 91 * , 035414 ( 2015 ) ] . a.a . abrikosov and i.m . khalatnikov , http://dx.doi.org/10.1088/0034-4885/22/1/310[rep . * 22 * , 329 ( 1959 ) ] . d. pines and p. nozires , _ the theory of quantum liquids _ benjamin , inc . , new york , 1966 ) . giuliani and g. vignale , _ quantum theory of the electron liquid _ ( cambridge university press , cambridge , 2005 ) . g.f . giuliani and j.j . quinn , http://dx.doi.org/10.1103/physrevb.26.4421 [ * 26 * , 4421 ( 1982 ) ] . molenkamp and m.j.m . de jong , http://dx.doi.org/10.1016/0038-1101(94)90244-5[solid-state electron . * 37 * , 551 ( 1994 ) ] . de jong and l.w . molenkamp , http://dx.doi.org/10.1103/physrevb.51.13389 [ * 51 * , 13389 ( 1995 ) ] . j.c.w . song and l.s . levitov , http://dx.doi.org/10.1103/physrevlett.109.236602 [ * 109 * , 236602 ( 2012 ) ] . gorbachev , a.k . geim , m.i . katsnelson , k.s . novoselov , t. tudorovskiy , i.v . grigorieva , a.h . macdonald , s.v . morozov , k. watanabe , t. taniguchi , and l.a . ponomarenko , http://dx.doi.org/10.1038/nphys2441[nature phys . * 8 * , 896 ( 2012 ) ] . mayorov , r.v . gorbachev , s.v . morozov , l. britnell , r. jalil , l.a . ponomarenko , p. blake , k.s . novoselov , k. watanabe , t. taniguchi , and a.k . geim , http://dx.doi.org/10.1021/nl200758b[nano lett . * 11 * , 2396 ( 2011 ) ] . mayorov , d.c . elias , i.s . mukhin , s.v . morozov , l.a . ponomarenko , k.s . novoselov , a.k . geim , and r.v . gorbachev , http://dx.doi.org/10.1021/nl301922d[nano lett . * 12 * , 4629 ( 2012 ) ] . l. wang , i. meric , p.y . huang , q. gao , y. gao , h. tran , t. taniguchi , k. watanabe , l.m . campos , d.a . muller , j. guo , p. kim , j. hone , k.l . shepard , and c.r . dean , http://dx.doi.org/10.1126/science.1244358[science * 342 * , 614 ( 2013 ) ] . t. taychatanapat , k. watanabe , t. taniguchi , and p. jarillo - herrero , http://dx.doi.org/10.1038/nphys2549[nature phys . * 9 * , 225 ( 2013 ) ] . a. woessner , m.b . lundeberg , y. gao , a. principi , p. alonso - gonzlez , m. carrega , k. watanabe , t. taniguchi , g. vignale , m. polini , j. hone , r. hillenbrand , and f.h.l . koppens , http://dx.doi.org/10.1038/nmat4169[nature mater . * 14 * , 421 ( 2015 ) ] . for a recent review see e.g. a.k . geim and i.v . grigorieva , http://dx.doi.org/10.1038/nature12385[nature * 499 * , 419 ( 2013 ) ] . l. pfeiffer and k.w . west , http://dx.doi.org/10.1016/j.physe.2003.09.035[physica e * 20 * , 57 ( 2003 ) ] . e.h . hwang and s. das sarma , http://dx.doi.org/10.1103/physrevb.77.115449[phys . b * 77 * , 115449 ( 2008 ) ] . a. principi , m. carrega , m.b . lundeberg , a. woessner , f.h.l . koppens , g. vignale , and m. polini , http://dx.doi.org/10.1103/physrevb.90.165408[phys . b * 90 * , 165408 ( 2014 ) ] . c .- h . park , n. bonini , t. sohier , g. samsonidze , b. kozinsky , m. calandra , f. mauri , and n. marzari , http://dx.doi.org/10.1021/nl402696q[nano lett . * 14 * , 1113 ( 2014 ) ] . t. sohier , m. calandra , c .- h . park , n. bonini , n. marzari , and f. mauri , http://dx.doi.org/10.1103/physrevb.90.125414[phys . b * 90 * , 125414 ( 2014 ) ] . c.r . dean , a.f . young , i. meric , c. lee , l. wang , s. sorgenfrei , k. watanabe , t. taniguchi , p. kim , k.l . shepard , and j. hone , http://dx.doi.org/10.1038/nnano.2010.172[nature nanotech . * 5 * , 722 ( 2010 ) ] . m. polini and g. vignale , _ the quasiparticle lifetime in a doped graphene sheet_. in _ no - nonsense physicist : an overview of gabriele giuliani s work and life _ ( eds . m. polini , g. vignale , v. pellegrini , and j.k . jain ) ( edizioni della normale , pisa , 2015 ) . also available as http://arxiv.org/abs/1404.5728[arxiv:1404.5728 ] . q. li and s. das sarma , http://dx.doi.org/10.1103/physrevb.87.085406 [ * 87 * , 085406 ( 2013 ) ] . a. principi , g. vignale , m. carrega , and m. polini , http://arxiv.org/abs/1506.06030[arxiv:1506.06030 ] . d. bandurin , i. torre , r.k . kumar , m. ben shalom , a. tomadin , a. principi , g.h . auton , e. khestanova , k.s . novoseiov , i.v . grigorieva , l.a . ponomarenko , a.k . geim , and m. polini , submitted . s. das sarma , s. adam , e.h . hwang , and e. rossi , http://dx.doi.org/10.1103/revmodphys.83.407 [ * 83 * , 407 ( 2011 ) ] . j. martin , n. akerman , g. ulbricht , t. lohmann , j.h . smet , k. von klitzing , and a. yacoby , http://dx.doi.org/10.1038/nphys781[nature phys . * 4 * , 144 ( 2008 ) ] . j. xue , j. sanchez - yamagishi , d. bulmash , p. jacquod , a. deshpande , k. watanabe , t. taniguchi , p. jarillo - herrero , and b.j . leroy , http://dx.doi.org/10.1038/nmat2968[nature mater . * 10 * , 282 ( 2011 ) ] . r. decker , y. wang , v.w . brar , w. regan , h .- z . tsai , q. wu , w. gannett , a. zettl , and m.f . crommie , http://dx.doi.org/10.1021/nl2005115[nano lett . * 11 * , 2291 ( 2011 ) ] . kotov , b. uchoa , v.m . pereira , f. guinea , and a.h . castro neto , http://dx.doi.org/10.1103/revmodphys.84.1067 [ * 84 * , 1067 ( 2012 ) ] . phan , j.c.w . song , and l.s . levitov , http://arxiv.org/abs/1306.4972[arxiv:1306.4972 ] . s. selberherr , _ analysis and simulation of semiconductor devices _ ( springer - verlag / wien , wien , 1984 ) ; d. vasileska and s.m . goodnick , _ computational electronics _ ( morgan & claypool , usa , 2006 ) . abanin , s.v . morozov , l.a . ponomarenko , r.v . gorbachev , a.s . mayorov , m.i . katsnelson , k. watanabe , t. taniguchi , k.s . novoselov , l.s . levitov , and a.k . geim , http://dx.doi.org/10.1126/science.1199595[science * 332 * , 328 ( 2011 ) ] . j. m. taylor , p. cappellaro , l. childress , l. jiang , d. budker , p.r . hemmer , a. yacoby , r. walsworth , and m.d . lukin , http://dx.doi.org/10.1038/nphys1075[nature phys . * 4 * , 810 ( 2008 ) ] . maze , p.l . stanwix , j.s . hodges , s. hong , j.m . taylor , p. cappellaro , l. jiang , m.v . gurudev dutt , e. togan , a.s . zibrov , a. yacoby , r.l . walsworth , and m.d . lukin , http://dx.doi.org/10.1038/nature07279[nature * 455 * , 644 ( 2008 ) ] . grinolds , s. hong , p. maletinsky , l. luan , m.d . lukin , r.l . walsworth , and a. yacoby , http://dx.doi.org/10.1038/nphys2543[nature phys . * 9 * , 215 ( 2013 ) ] . m. gibertini , a. tomadin , f. guinea , m.i . katsnelson , and m. polini , http://dx.doi.org/10.1103/physrevb.85.201405 [ * 85 * , 201405(r ) ( 2012 ) ] . i. torre , a. tomadin , r. krahne , v. pellegrini , and m. polini , http://dx.doi.org/10.1103/physrevb.91.081402[phys . b * 91 * , 081402(r ) ( 2015 ) ] .

we present a microscopic theory for superconductivity in a magnetic field based on a local approximation approach . we derive an expression for free energy density as a function of temperature and vector potential * a * , and two basic equations of the theory : the first is an implicit solution for energy gap parameter amplitude as a function of wave vector * k * , temperature and vector potential * a * ; and the second is a london - like relation between electrical current density * j * and vector potential * a * , with an `` effective superconducting electron density '' that is both- and * a*-dependent . the two equations allow determination of spatial variations of * a * and in a superconductor for given temperature , applied magnetic field and sample geometry . the theory shows the existence of a `` partly - paired state , '' in which paired electrons ( having ) and de - paired electrons ( having ) co - exist . such a `` partly - paired state '' exists even at when is above a threshold for a given sample , giving rise to a non - vanishing knight shift at for above the threshold . we expect the theory to be valid for highly - local superconductors for all temperatures and magnetic fields below the superconducting transition . in the low - field limit , the theory reduces to the local - limit result of bcs . as examples , we apply the theory to the case of a semi - infinite superconductor in an applied magnetic field parallel to the surface of the superconductor and the case of an isolated vortex in an infinite superconductor , and determine , in each case , spatial variations of quantities such as * a * and . we also calculate magnetic field penetration depth and lower critical magnetic field . the ratio ( where is the thermodynamic critical magnetic field ) is found to be only weakly-dependent for low temperatures and nearly-independent for intermediate and high temperatures , and quantitatively not very different from that of the ginzburg - landau theory for ginzburg - landau parameter .

0706.2394

currently accepted microscopic theories for superconductivity in a magnetic field include the microscopic derivation of the meissner effect in the bcs theory@xcite and the microscopic derivation of the phenomenological ginzburg - landau equations by gorkov@xcite from the bcs theory . however , the bcs derivation of the meissner effect is based on a linear - response approach ( i.e. , an externally applied magnetic field is treated as a weak perturbation ) , and thus , is valid only in the low - field limit . the gorkov derivation of the ginzburg - landau equations is based on the assumption that energy gap function @xmath15 is a small quantity , and thus , is valid only for temperatures just below critical temperature @xmath16 . various extensions of the gorkov theory also rely heavily on the assumption of @xmath15 being small , or the assumption of magnetic field being weak.@xcite it is desirable to have a microscopic theory for superconductivity in a magnetic field that is valid under more general conditions . as an effort along this line , we present in this paper a theory based on a local approximation approach . the theory allows microscopic description of the suppression of superconductivity by an externally applied magnetic field . the theory is expected to be valid for highly - local superconductors ( the high - temperature copper - oxide superconductors are examples of highly - local superconductors , for which ginzburg - landau parameter @xmath14 ) for all temperatures and magnetic fields below the superconducting transition ( except , perhaps , in the high - field limit when the effect of spin paramagnetism , which is neglected in the present work , may become important ) . in the low - field limit , the theory reduces to the local - limit result of bcs.@xcite in sec . [ sectheory ] , we present the details of the theory , and derive an expression for free energy density @xmath0 as a function of temperature @xmath1 and vector potential * a * , and two basic equations of the theory : the first is an implicit solution for energy gap parameter amplitude @xmath4 as a function of wave vector * k * , temperature @xmath1 and vector potential * a * ; and the second is a relation between electrical current density * j * and vector potential * a*. we also analyze solutions for @xmath17 . in sec . [ secappl ] , as examples , we apply the theory to the case of a semi - infinite superconductor in an applied magnetic field @xmath5 parallel to the surface of the superconductor and the case of an isolated vortex in an infinite superconductor , and determine , in each case , spatial variations of quantities such as * a * and @xmath4 . we also calculate magnetic field penetration depth @xmath10 and lower critical magnetic field @xmath11 . a brief summary is given in sec . [ secsummary ] . we consider a superconductor in an applied magnetic field . our starting point is the same bcs pairing hamiltonian@xcite @xmath18 where @xmath19 is the normal state single - electron energy , @xmath20 the fermi energy , @xmath21 the pairing interaction matrix element , and @xmath22 and @xmath23 are the fermi operators of an electronic state of wave vector * k * and spin @xmath24 in the normal state . single - electron energy @xmath19 and its corresponding single - electron wave function @xmath25 satisfy the schrdinger equation @xmath26 with single - electron hamiltonian @xmath27 ^ 2 + u({\bf x}),\ ] ] where @xmath28 is the vector potential , and @xmath29 a periodic scalar potential . in writing down hamiltonian @xmath30 , we have neglected , for simplicity , the effect of spin paramagnetism ( which may be important for high fields ) . we note that , if @xmath19 and @xmath25 for @xmath31 are known , @xmath30 can be diagonalized in essentially the same way as for @xmath32 ( i.e , by making the bcs pairing approximation , and then applying the bogoliubov transformation@xcite ) . however , since @xmath28 in the superconducting state is itself an unknown function , a simultaneous determination of @xmath19 , @xmath25 and @xmath33 does not seem possible . we therefore adopt a local approximation approach , which we describe next . we note that , in the superconducting state , vector potential @xmath28 varies spatially with the length @xmath34 , the magnetic field penetration depth , which is @xmath35 . in contrast , single - electron wave function @xmath25 oscillates spatially with a much shorter length @xmath36 ( here @xmath37 is a fermi wave vector),@xcite so that @xmath28 can be considered locally constant over many wavelengths of @xmath25 . our approach is based on this observation , and can be outlined as consisting the following three steps . _ step 1 : _ we focus on a small region around a local point , say @xmath38 . dimension @xmath39 of this small region satisfies @xmath40 , where @xmath41 is a crystal lattice constant of the superconductor . in this small region , we assume @xmath42 is a constant , and solve the schrdinger equation [ eq . ( [ scheq ] ) ] to obtain @xmath19 and @xmath25 . _ step 2 : _ based on the obtained @xmath19 and @xmath25 in the vicinity of @xmath38 , we diagonalize hamiltonian @xmath30 , which now becomes a local quantity , because of its dependence upon @xmath43 through @xmath19 and @xmath25 . _ step 3 : _ once hamiltonian @xmath30 is diagonalized , we move on to derive an expression for local free energy density @xmath0 and two basic equations of the theory . the equations allow determination of the spatial variations of vector potential @xmath44 and other quantities for given temperature , applied magnetic field and sample geometry . since the non - local effect ( or coherence effect)@xcite in the superconducting state is not accounted for , we expect this approach to be valid only for highly - local superconductors for which magnetic field penetration depth @xmath34 is much larger than coherence length @xmath45 ( i.e. , @xmath46 ) . we explain the details of this approach in the following subsections . as outlined above , we first focus on a small region around a local point , say @xmath38 . dimension @xmath39 of this small region satisfies @xmath47 , where @xmath48 is a crystal lattice constant of the superconductor , and @xmath34 is the magnetic field penetration depth . since vector potential @xmath28 varies spatially with the length @xmath34 , which is @xmath35 , whereas @xmath25 oscillates spatially with the length @xmath48 , which is @xmath49 , we can assume @xmath50 in this small region , and solve eq . ( [ scheq ] ) . for a constant @xmath42 , it is not difficult to solve eq . ( [ scheq ] ) . for free electrons , for which scalar potential @xmath51 , solutions of eq . ( [ scheq ] ) are easily obtained . namely , for @xmath32 , we have @xmath52 and @xmath53 and for @xmath31 , we have @xmath54 and @xmath55 here we have used @xmath56 and @xmath57 to denote solutions of eq . ( [ scheq ] ) for @xmath32 . in this paper , we will not consider the case of a general periodic scalar potential @xmath29 ( i.e. , we will not consider in this paper how the details of an electronic energy band structure may affect properties of the superconducting state ) . instead , for simplicity in presenting the theory , we will use the solutions for free electrons in the following . having obtained @xmath19 and @xmath25 , we can move on to diagonalize hamiltonian @xmath30 of eq . ( [ h0 ] ) , which now becomes a local quantity , because @xmath19 and @xmath25 are obtained locally at @xmath38 for @xmath58 . an important step in the diagonalization of hamiltonian @xmath30 is assuming that the pairing@xcite of electrons of opposite momenta and spins holds even for @xmath31 . i.e. , for a pair of @xmath59 and @xmath60 electrons , we assume @xmath61 when the electrons are superconducting ( here the angle brackets @xmath62 denote a thermal average ) . the diagonalization of hamiltonian @xmath30 is the same as in the case of @xmath32,@xcite except that we have @xmath63 for @xmath64 . the results of the diagonalization are as follows . the energy gap parameter is defined as@xcite @xmath65 the diagonalized hamiltonian is @xmath66 where @xmath67 @xmath68 is the single - electron energy in the normal state , measured relative to the fermi energy @xmath20 ; @xmath69 the quasi - particle excitation energy in the superconducting state ; @xmath70 the symmetric part of @xmath71 ; @xmath72 the fermi function ; and @xmath73 and @xmath74 are the fermi operators for quasi - particles in the superconducting state . the operators @xmath73 and @xmath74 are related to @xmath75 and @xmath23 via the bogoliubov transformation@xcite @xmath76 where the coefficients @xmath77 and @xmath78 satisfy the following relations : @xmath79 @xmath80 and @xmath81 after the diagonalization of hamiltonian @xmath30 , eq . ( [ gap - def ] ) can be expressed as @xmath82 with regard to the above - described diagonalization of hamiltonian @xmath30 , the following are worth emphasizing . \(i ) interaction matrix element @xmath83 is @xmath44-independent , because @xmath84 . \(ii ) for @xmath31 , since @xmath63 , we have @xmath85 , @xmath86 and @xmath87 . however , we still have @xmath88 , @xmath89 and @xmath90 , as one can see from eqs . ( [ uk ] ) , ( [ eks ] ) and ( [ gap - eq ] ) . \(iii ) fermi energy @xmath20 , relative to which energies such as @xmath91 and @xmath71 are measured , is a local quantity , i.e. , @xmath92 . this can be understood as follows . in the superconducting ground state , all electrons are paired . for each pair of electrons , we have @xmath93 for @xmath31 . thus , we expect @xmath94 for @xmath31 . we further expect that @xmath95 for @xmath96 in the superconducting state , where the increase in the fermi energy is due to a flow of the paired electrons . for an electronic energy spectrum of the form of eq . ( [ ek ] ) , fermi energy @xmath20 is determined by @xmath97 , which gives @xmath98 with the help of eqs . ( [ ek ] ) and ( [ ef ] ) , we also have @xmath99 and @xmath100 where @xmath101 from the above - described diagonalization of hamiltonian @xmath30 , we see the following : electronic states ( or quasi - particles ) in the superconducting state are each characterized by a wave - vector * k * and a spin @xmath24 , and are in one - to - one correspondence with those in the normal state ( this point is the same as in the case of @xmath102).@xcite the superconducting state is different from the normal state because ( i ) there exists an energy gap parameter @xmath103 for an electronic excitation in the superconducting state ; and ( ii ) fermi energy @xmath20 is vector potential * a * dependent in the superconducting state . both the existence of @xmath103 and the * a*-dependence of @xmath20 originate from the pairing of @xmath59 and @xmath104 electrons . from diagonalized hamiltonian @xmath30 [ eq . ( [ h ] ) ] , the following expression for free energy density @xmath0 at location * x * ( we can drop the prime in @xmath38 hereafter ) in the superconductor can be derived : @xmath105 \nonumber \\ & & \mbox { } + \,\,\ , \frac{ne^2}{2mc^2}a^2 \,\,\ , + \,\,\ , \frac { ( \nabla\!\times\!{\bf a})^2 } { 8 \pi } \ , . \label{f}\end{aligned}\ ] ] the first of the three terms in the above expression comes from @xmath106 $ ] , which is the usual statistical electronic free energy density.@xcite the second term comes from @xmath107 , where @xmath108 is the density of electrons , and we have used eq . ( [ ef ] ) . this term is added to @xmath0 because electronic energies in the expression for hamiltonian @xmath30 are measured relative to @xmath20 . the third term is the magnetic field energy density . note that the expression for @xmath0 involves @xmath4 ( through @xmath109 , @xmath71 and @xmath110 ) . as we will see in the next subsection , @xmath4 is a function of * k * , @xmath1 and * a * , i.e. , @xmath111 , and the function @xmath17 is determined by the self - consistency of the diagonalization of @xmath30 . thus , we see that @xmath0 is a function of @xmath1 and * a * , i.e. , @xmath112 . [ as can be shown , free energy density @xmath0 , excluding the magnetic field energy density , becomes * a*-independent when @xmath7 for all electronic states . ] the following equation is derived as a condition for the self - consistency of the diagonalization of hamiltonian @xmath30 : @xmath113 for @xmath32 , this equation was previously derived by the author in ref . . we present the details of the derivation of this equation for @xmath31 , which is similar to that for @xmath114 , in appendix [ derivehaoeq ] . by using @xmath115 and the condition that @xmath116 at @xmath117 , we can express the above equation as @xmath118^{-1 } \left\ { \tanh \left [ \left ( \sqrt { \left ( \xi^{(0)}_{\bf k } \right)^2 + |\delta_{\bf k}|^2 } + \frac{\hbar e}{mc}{\bf k}\!\cdot{\bf a } \,\ , \right ) / 2k_bt \right ] \right.\ ] ] @xmath119 \right\ } = \frac { \tanh \left ( |\xi^{(0)}_{\bf k}| / 2k_bt_c \right ) } { |\xi^{(0)}_{\bf k}| } \ , . \label{haoeq}\ ] ] this equation is an implicit solution for @xmath4 as a function of * k * , @xmath1 and * a * for given @xmath16 . note that interaction @xmath83 does not appear in eq . ( [ haoeq ] ) . instead , critical temperature @xmath16 is involved through the condition that @xmath7 at @xmath120 . namely , @xmath4 depends on @xmath83 only through @xmath16 . critical temperature @xmath16 and phase @xmath121 of @xmath103 ( i.e. , @xmath122 ) are determined by solving the eigenvalue problem @xmath123 for given interaction @xmath83 and electronic energy spectrum @xmath124 . equation ( [ tc - eq ] ) is the linearized form of eq . ( [ gap - eq ] ) in the limit of @xmath125 . a complete solution for @xmath103 is therefore a combination of the solution for @xmath4 of eq . ( [ haoeq ] ) and the solutions for @xmath16 and @xmath121 of eq . ( [ tc - eq ] ) . ( related discussions are also given in refs . and . ) having obtained an expression for @xmath127 [ i.e. , the expression given by eq . ( [ f ] ) , with @xmath128 being implicitly given by eq . ( [ haoeq ] ) ] , we are now ready to consider determination of @xmath33 . in thermodynamic equilibrium , the overall free energy , given by the volume integral of @xmath129 , must be stationary with respect to arbitrary variation of @xmath28 . this variational problem leads to @xmath130 where * j * is the electrical current density , and we have used the relations @xmath131 , with @xmath132 being the magnetic flux density . the first term on the right - hand side of eq . ( [ haoeq2 ] ) represents a uniform flow of all the electrons , which originates from the pairing of @xmath59 and @xmath104 electrons . the second term is due to quasi - particle excitations and de - paired electrons , and tends to cancel the first term . when all the electrons are in the superconducting ground state , the second term is zero . on the other hand , when @xmath7 for all the electrons ( i.e. , when the superconductor is in the normal state ) , the two terms cancel each other , and we have @xmath133 . equation ( [ haoeq2 ] ) was also derived by bcs.@xcite however , the bcs derivation of eq . ( [ haoeq2 ] ) is based on a linear - response approach ( i.e. , vector potential * a * is treated as a small perturbation ) , and thus , is valid only in the low - field limit . note that eq . ( [ haoeq2 ] ) is not gauge invariant ( this feature is the same for the london equation@xcite ) . this is because the pairing of @xmath59 and @xmath104 electrons fixes the total ( canonical ) momentum of the electrons , and thereby also the gauge of the vector potential . namely , the theory became not - gauge - invariant at the point when the pairing assumption [ eq . ( [ pairing ] ) ] was made . for the same reason , the expression for @xmath127 [ eq . ( [ f ] ) ] and the implicit solution for @xmath17 [ eq . ( [ haoeq ] ) ] also are not gauge invariant . since , as can be shown , @xmath134 ( which means that a larger value of @xmath135 is energetically less favorable ) , we see that the gauge of vector potential * a * in the present theory is such that @xmath136 deep inside a bulk superconductor , i.e. , the london gauge.@xcite . [ table1 ] units for forming dimensionless quantities [ cols="<,<",options="header " , ] it is convenient to introduce a set of units so that physical quantities involved in the theory become dimensionless . the units that we choose to use are listed in table 1 , where @xmath137 is the magnetic field penetration depth at @xmath8 in the limit of zero magnetic field , and satisfies @xmath138 ; and @xmath139 is the thermodynamic critical magnetic field at @xmath8 , and satisfies @xmath140 , where @xmath141 is the density of states at the fermi level ( @xmath142 for free electrons@xcite ) . the result for @xmath139 was previously obtained by the author in ref . . we also make the substitution @xmath143 , and , as usual , assume that @xmath144 so that the substitution @xmath145 and the approximation @xmath146 hold , where @xmath147 , with @xmath148 being the angle between * k * and * a*. by using the units shown in table 1 and the above - mentioned assumption and approximation , eq . ( [ f ] ) becomes @xmath149 + a^2 + ( \nabla\!\times\!{\bf a})^2 \label{f}\ ] ] with @xmath150 and @xmath151 and eqs . ( [ haoeq ] ) and ( [ haoeq2 ] ) become @xmath118^{-1 } \left\ { \tanh \left [ \left ( \sqrt { \left ( \xi^{(0)}_{\bf k } \right)^2 + |\delta_{\bf k}|^2 } + \frac{\pi}{\sqrt{2 } } az_{\bf k } \,\ , \right ) / 2 t \right ] \right.\ ] ] @xmath152 \right\ } = \frac { \tanh \left ( |\xi^{(0)}_{\bf k}| / 2 \right ) } { |\xi^{(0)}_{\bf k}| } \label{haoeq1}\ ] ] and @xmath153 respectively , where the `` effective superconducting electron density '' @xmath154 with @xmath155^{-1 } \!\!\!.\ ] ] note that eq . ( [ haoeq2 ] ) is the same as the london equation,@xcite except that our @xmath3 , given by eq . ( [ ns - eq ] ) , is both @xmath1- and * a*-dependent , whereas @xmath3 is only @xmath1-dependent in the london theory . the condensation energy density is defined as @xmath156 where @xmath157 is the superconducting state free energy density excluding the magnetic field energy density ; and @xmath158 is the normal state counterpart of @xmath159 . the quantity @xmath160 can be used as a measure of the difference between the normal and superconducting states . as can be shown , the following relation exists : @xmath161 where the second expression holds because @xmath158 is * a*-independent . for an isotropic superconductor ( as in the present case ) , @xmath3 and @xmath160 are functions of temperature @xmath1 and the magnitude of vector potential * a * , i.e. , @xmath162 and @xmath163 . we analyze the functions @xmath17 , @xmath164 and @xmath165 in the next subsection . we solve eq . ( [ haoeq1 ] ) to obtain @xmath17 by using an iterative method.@xcite [ note that the variables @xmath166 for the function @xmath17 appear in eq . ( [ haoeq1 ] ) in the forms of @xmath167 ; and remember that @xmath168 is the component of * a * along * k*. ] when solving eq . ( [ haoeq1 ] ) , it is important to note the following : \(i ) we define temperature @xmath169 as such that eq . ( [ haoeq1 ] ) has no @xmath170 solution for given @xmath171 , @xmath1 and @xmath172 if @xmath173 . note that @xmath169 is a function of @xmath171 and @xmath172 . only for @xmath32 is @xmath174 the same for all the electronic states . \(ii ) similarly , we define vector potential magnitude @xmath175 as such that eq.([haoeq1 ] ) has no @xmath170 solution for given @xmath171 , @xmath1 and @xmath172 if @xmath176 . note that @xmath175 is a function of @xmath171 , @xmath1 and @xmath177 . \(iii ) depending on @xmath171 , @xmath1 and @xmath172 , a @xmath6 solution of eq . ( [ haoeq1 ] ) may be an unstable solution , because the normal state solution @xmath116 may be energetically more favorable . point ( iii ) can be understood as follows . note that for given @xmath171 , @xmath1 and @xmath172 , there are always two possible solutions for @xmath4 if @xmath178 or @xmath179 : the @xmath6 solution of eq . ( [ haoeq1 ] ) and the @xmath7 solution . the free energy associated with a pair of @xmath180 excitations with @xmath6 is @xmath181 \left [ 1 + e^{-\left(e^{(s)}_{\bf k } - \frac{\pi}{\sqrt{2 } } az_{\bf k } \right)/t } \right ] , \label{fsk}\ ] ] where @xmath182 . for @xmath7 , it becomes @xmath183 \left [ 1 + e^{-\left(|\xi_{\bf k}^{(0)}| - \frac{\pi}{\sqrt{2 } } az_{\bf k } \right ) /t } \right ] . \label{fnk}\ ] ] both @xmath184 and @xmath185 are functions of @xmath171 , @xmath1 and @xmath172 . for given @xmath171 , @xmath1 and @xmath172 , the @xmath6 solution is the stable solution if @xmath186 . otherwise , the @xmath7 solution is the stable solution . we define temperature @xmath187 and vector potential magnitude @xmath188 as such that the @xmath7 solution becomes the stable solution for @xmath189 or @xmath190 . by definition , we have @xmath191 and @xmath192 . note that a pair of @xmath193 electrons becomes de - paired ( having @xmath7 ) for @xmath194 or @xmath195 . versus @xmath171 for different values of @xmath196 ; ( b ) @xmath197 versus @xmath171 for different values of @xmath1 ; ( c ) @xmath198 versus @xmath172 for different values of @xmath199 . the dotted curves show the corresponding results for @xmath169 or @xmath200 . ] versus @xmath171 for different values of @xmath1 and @xmath172 ; @xmath201 , @xmath202 , and @xmath203 in ( a ) , ( b ) , and ( c ) , respectively . the values of @xmath204 are indicated on the curves . unstable solutions are shown by dotted curves . ] versus @xmath1 for different values of @xmath205 and @xmath172 ; @xmath206 and @xmath207 in ( a ) and ( b ) , respectively . the values of @xmath208 are indicated on the curves . unstable solutions are shown by the dotted curves . ] versus @xmath172 for different values of @xmath171 and @xmath1 ; @xmath209 and @xmath207 in ( a ) and ( b ) , respectively . the values of @xmath204 are indicated on the curves . unstable solutions are shown by the dotted curves . ] versus @xmath1 for different values of @xmath135 as indicated on the curves ; ( b ) @xmath3 versus @xmath135 for different values of @xmath1 as indicated on the curves . the dotted curve in ( a ) shows mhlschlegel s result@xcite for @xmath210 , where @xmath211 . ] versus @xmath1 for different values of @xmath135 as indicated on the curves ; ( b ) @xmath160 versus @xmath135 for different values of @xmath1 as indicated on the curves . ] numerical results for @xmath198 and @xmath212 are shown in fig . 1(a)-1(c ) ] . figure 1(a ) shows @xmath198 versus @xmath199 for different values of @xmath196 ; fig . 1(b ) shows @xmath213 versus @xmath199 for different values of @xmath1 ; and fig . 1(c ) shows @xmath198 versus @xmath172 for different values of @xmath199 . the dotted curves in fig . 1 show corresponding results for @xmath169 and @xmath200 . as shown in figs . 1(a ) and 1(b ) , both @xmath198 and @xmath214 are increasing functions of @xmath199 , meaning that pairs of electrons with smaller @xmath199 become de - paired at lower values of @xmath1 and/or @xmath172 than those with larger @xmath199 . figure 1(c ) shows that @xmath215 is a decreasing function of @xmath172 , which also means that @xmath214 is a decreasing function of @xmath1 . for constant @xmath199 , the values of @xmath198 and @xmath212 are lowest when * k * is parallel to * a * , because * a * appears in eq . ( [ haoeq1 ] ) only in the form of @xmath172 , which is largest ( therefore , most effective as a de - pairing force ) when @xmath216 . for two electronic states with wave - vectors @xmath217 and * k * , respectively , if @xmath218 and * k * is parallel to * a * , then , the following relations hold : @xmath219 ; and @xmath220 . note that @xmath221 is the lowest value of @xmath222 , i.e. , @xmath223 . similarly , @xmath224 is the lowest value of @xmath225 , i.e. , @xmath226 . the function @xmath221 [ or its inverse function @xmath224 ] , which is shown by the @xmath228 curve in fig . 1(c ) ( for @xmath229 ) , is of particular importance . when @xmath230 [ or @xmath231 in a region in a superconductor [ remember that @xmath232 is a location - dependent quantity ] , this region is in a `` all - paired state , '' in which @xmath6 for all the electrons . on the other hand , when @xmath233 [ or @xmath234 in a region in a superconductor , this region is in a `` partly - paired state , '' in which electrons with @xmath235 in the range @xmath236 [ or with @xmath225 in the range @xmath237 become de - paired ( having @xmath7 ) , while electrons with @xmath238 [ or @xmath239 remain paired ( having @xmath170 ) . numerical results for @xmath4 versus @xmath199 for different values of @xmath1 and @xmath172 are shown in fig . 2(a)-2(c ) ] . figure 2(a ) shows the case of @xmath32 . in this case , @xmath4 is a monotonic decreasing function of @xmath199 for all temperatures below @xmath16 ; and @xmath4 vanishes at the same temperature @xmath16 for all values of @xmath171 . figure 2(b ) shows an example of the case of @xmath240 [ note that @xmath241 ( in dimensionless units ) ] . in this case , the @xmath4-versus-@xmath199 curve for @xmath8 is the same as in the case of @xmath32 . however , as @xmath1 increases , @xmath4 for a smaller @xmath199 is more strongly suppressed , and decreases faster , so that the @xmath2-versus-@xmath199 curve eventually becomes non - monotonic , with a maximum located away from @xmath242 . as @xmath1 increases further , @xmath4 for a smaller @xmath199 vanishes at a lower temperature [ namely , @xmath198 is smaller for smaller @xmath171 , a feature that is also shown by fig . the @xmath4-versus-@xmath199 curve then has two parts : a @xmath7 part for low energies , for which @xmath243 [ or @xmath244 , and a @xmath6 part for higher energies , for which @xmath245 [ or @xmath239 . figure 2(c ) shows an example of the case of @xmath246 . in this case , the @xmath2-versus-@xmath199 curve has a @xmath7 part even for @xmath8 . namely , at @xmath8 , @xmath7 for those electronic states with @xmath247 . the vertical rises in the @xmath2-versus-@xmath199 curves for low temperatures ( i.e. , the curves for @xmath248 0.3 and 0.5 ) in fig . 2(c ) indicate discontinuities . the doted curves in fig . 2(c ) show corresponding unstable solutions of eq . ( [ haoeq1 ] ) for @xmath1 in the range @xmath249 ( or for @xmath135 in the range @xmath250 ) . figures 3 and 4 show , respectively , the @xmath1-dependence and @xmath172-dependence of @xmath4 . the dotted curves in figs 3 and 4 show corresponding unstable solutions of eq . ( [ haoeq1 ] ) in the range @xmath251 or @xmath252 . as shown in the figures , @xmath4 is a monotonic decreasing function of @xmath1 and @xmath172 , except that at @xmath8 , @xmath6 is a constant for @xmath253 . note that , at @xmath254 or @xmath255 , @xmath4 may become zero continuously or discontinuously , depending on @xmath199 , @xmath1 and @xmath172 . the vertical drops in some of the curves shown in figs . 3 and 4 indicate discontinuities . based on numerical solutions for @xmath17 , we can compute @xmath164 and @xmath165 straightforwardly by numerically carrying out the integrals involved in the expressions for @xmath3 and @xmath160 . numerical results for @xmath164 and @xmath165 are shown in figs . 5 and 6 , respectively . the dotted curve in fig . 5(a ) shows mhlschlegel s numerical result@xcite for @xmath210 , which is based on a linear - response approach,@xcite and therefore , is valid only in the low - field limit , i.e. , @xmath256 . [ more discussion on magnetic field penetration depth @xmath10 is given below in sec . [ seclambda ] . ] note that @xmath210 corresponds to our @xmath257 for @xmath258 . the quantitative difference between mhlschlegel s @xmath210 and our @xmath259 , as shown in fig . 5(a ) , is due to the use of the cut - off approximation@xcite for solving the energy gap equation in mhlschlegel s work . the @xmath160-versus-@xmath1 curve for @xmath260 in fig . 6(a ) shows @xmath261 , which is the same as @xmath262 . this result was also previously obtained by the author in ref . . as shown in figs . 5 and 6 , @xmath164 and @xmath165 both are monotonic decreasing functions of @xmath1 and @xmath135 , except that @xmath263 is a constant for @xmath264 ( where @xmath265 ) at @xmath8 [ fig . note that , for @xmath8 and @xmath264 , we have @xmath266 [ which can be derived either from the expression for @xmath160 , or by solving eq . ( [ nsfc ] ) for @xmath267 and @xmath268 . this is shown in fig . 6(b ) , where we can see that the @xmath160-versus-@xmath135 curve for @xmath8 is parabolic for @xmath269 . namely , in this case , all the electrons are paired and no thermal excitations exist , so that the decrease in @xmath160 ( i.e. , the @xmath270 term ) is entirely due to the kinetic energy associated with the uniform flow of all the electrons . at @xmath8 , as @xmath135 increases further so that @xmath271 ( i.e. , @xmath272 ) , de - paired electrons ( having @xmath7 ) begin to appear , and we have @xmath273 . the present analysis provides a possible explanation for the experimentally observed non - vanishing knight shifts in the superconducting state near @xmath8 ( for example , refs . and ) . namely , at @xmath8 , electrons in a superconducting sample are all paired only when the applied magnetic field is weak so that we have @xmath274 everywhere in the sample ; this gives zero spin - polarization , and therefore , zero knight shift ( which is proportional to the density of spin polarization@xcite ) . however , we note that the applied magnetic fields that were used for the knight shift measurements@xcite are comparable to the thermodynamic critical magnetic fields of the samples . therefore , it is likely that we actually had @xmath275 over a significant portion of the sample , where a finite fraction of the electrons were de - paired and spin - polarized , giving rise to a non - zero knight shift . the previous theoretical prediction of a zero knight shift in the superconducting state for @xmath276 by yosida@xcite is valid only for weak magnetic fields . although the present theory is developed for highly - local superconductors , we expect qualitative conclusions of the theory , including the prediction for a non - vanishing knight shift near @xmath8 for a not - so - weak magnetic field , to be valid also for non - local superconductors . we present in this section a few examples of application of the theory . we consider the case of a semi - infinite superconductor in an applied magnetic field @xmath5 parallel to the surface of the superconductor and the case of an isolated vortex in an infinite superconductor . we determine , in each case , spatial variations of vector potential * a * , magnetic flux density * b * , electrical current density * j * , energy gap parameter amplitude @xmath4 , `` effective superconducting electron density '' @xmath3 and condensation energy density @xmath160 . we also calculate magnetic field penetration depth @xmath10 and lower critical magnetic field @xmath11 . we consider a semi - infinite superconductor in an applied magnetic field @xmath5 parallel to the surface of the superconductor . let the superconductor occupy the half space @xmath277 and @xmath5 be applied along the @xmath278-axis . in terms of the cartesian coordinates @xmath279 and the unit vectors @xmath280 , we can write @xmath281 , @xmath282 , and @xmath283 . then , eq . ( [ haoeq2 ] ) and the relation @xmath284 become @xmath285 and @xmath286 respectively , where a `` prime '' indicates a derivative with respect to @xmath287 ; @xmath288 is given by eq . ( [ ns - eq ] ) , where @xmath289 is determined by eq . ( [ haoeq1 ] ) ; and the relation @xmath290 has been used . this is a non - linear second - order boundary - value problem with boundary conditions @xmath291 and @xmath292 note that , for the convenience of numerical calculation , we have expressed this second - order boundary - value problem as a system of two first - order differential equations . the numerical method for solving this boundary - value problem is explained in appendix [ computeab ] . numerical results for @xmath293 , @xmath294 , @xmath295 , @xmath296 , @xmath297 and @xmath298 near the surface of the superconductor for different values of @xmath9 are shown in fig . 7(a)-7(c ) ] and fig . 8 [ figs . 8(a)-8(c ) ] for @xmath8 and 0.6 , respectively . as an example of @xmath299 , @xmath300 is shown in the figures . here @xmath301 is the energy gap parameter of an electronic state on the fermi surface with wave vector @xmath37 parallel to * a * , of which the value of @xmath222 , or @xmath225 , is the lowest among all the electronic states , i.e. , @xmath302 , and @xmath303 , as we mentioned earlier . figures 7(a ) and 8(a ) show examples of the case where @xmath9 is low so that @xmath304 at the surface of the superconductor . in this case , no de - paired electrons exist in the superconductor , i.e. , @xmath305 for all the electrons in the superconductor . at @xmath8 [ fig . 7(a ) ] , @xmath4 and @xmath3 are @xmath287-independent [ because @xmath4 and @xmath3 are @xmath135-independent for @xmath306 at @xmath8 , as shown in figs . 4 and 5 , respectively ] , and therefore , our results for @xmath293 , @xmath294 , @xmath295 and @xmath298 are the same as those of the london theory . namely , in this case , our eq . ( [ haoeq2 ] ) is the same as the london equation , which gives linear relations between @xmath293 , @xmath294 and @xmath295 , i.e. , @xmath307 . for @xmath308 [ fig . 8(a ) ] , @xmath4 and @xmath3 become @xmath287-dependent near the surface of the superconductor , and therefore , our results for @xmath293 , @xmath294 , @xmath295 and @xmath298 near the surface of the superconductor are no longer the same as those of the london theory . figures 7(b ) , 7(c ) , 8(b ) and 8(c ) show examples of the case where @xmath9 has been increased to a point such that @xmath309 at the surface of the sample . let @xmath310 be the coordinate of the location at which @xmath311 . we have @xmath312 in the region @xmath313 , and @xmath314 in the region @xmath315 . the region @xmath313 is in a `` partly - paired state , '' where paired electrons ( having @xmath316 ) and de - paired electrons ( having @xmath317 ) co - exist . the region @xmath315 remains `` all - paired , '' where @xmath316 for all the electrons . our results for @xmath293 , @xmath294 , @xmath295 and @xmath298 in the region @xmath313 are significantly different from those of the london theory , especially when @xmath9 is high [ figs . 7(c ) and 8(c ) ] . in particular , @xmath295 is no longer a monotonic decreasing function of @xmath287 , having a maximum located near @xmath310 . , @xmath294 , @xmath295 , @xmath296 , @xmath297 , and @xmath298 near the surface at @xmath8 and different values of @xmath9 : @xmath318 ( a ) , @xmath319 ( b ) , and @xmath202 ( c ) . the quantities @xmath293 , @xmath294 , @xmath295 , @xmath296 , @xmath297 , and @xmath298 are measured in units of @xmath320 , @xmath9 , @xmath321 , @xmath322 , @xmath323 , and @xmath324 , respectively . ] , @xmath294 , @xmath295 , @xmath296 , @xmath297 , and @xmath298 near the surface at @xmath325 and different values of @xmath9 : @xmath326 ( a ) , @xmath327 ( b ) , and @xmath328 ( c ) [ @xmath329 ( a ) , @xmath330 ( b ) , and @xmath331 ( c ) ] . the quantities @xmath293 , @xmath294 , @xmath295 , @xmath296 , @xmath332 , and @xmath298 are measured in units of @xmath333 , @xmath9 , @xmath334 , @xmath322 , @xmath323 , and @xmath324 , respectively , where @xmath335 . ] versus @xmath1 for different values of @xmath9 as indicated on the curves ; ( b ) @xmath336 versus @xmath9 for different values of @xmath1 as indicated on the curves . the dotted curve in ( a ) shows the mhlschlegel s result@xcite for @xmath337 . ] the solution for @xmath293 obtained in sec . [ secsemiinfinitesc ] for the case of a semi - infinite superconductor can be used to calculate magnetic field penetration depth @xmath10 , which , for a semi - infinite superconductor , is defined as@xcite @xmath338 where @xmath339 is the value of @xmath293 at the surface of the superconductor . numerical results for @xmath10 are shown as @xmath340 versus @xmath1 for different values of @xmath9 in fig . 9(a ) , and @xmath340 versus @xmath9 for different values of @xmath1 in fig . the dotted curve in fig . 9(a ) shows mhlschlegel s numerical result@xcite for @xmath210 . as we mentioned earlier , the mhlschlegel s result is based on a linear - response approach , and corresponds to our @xmath257 for @xmath260 , which is the same as @xmath341 for @xmath342 , i.e. , @xmath343 . the quantitative difference between mhlschlegel s @xmath210 and our @xmath257 for @xmath260 , or @xmath341 for @xmath342 , is due to the use of the cut - off approximation@xcite in mhlschlegel s work . as shown in fig . 9(b ) , for @xmath9 up to about @xmath344 , @xmath336 is @xmath9-independent or nearly @xmath9-independent for low temperatures , and only weakly @xmath9-dependent for higher temperatures . however , for @xmath9 above about @xmath344 , @xmath336 drops rapidly as @xmath9 increases . note that the case of @xmath9 below about @xmath344 corresponds to the case shown in figs . 7(a ) and 8(a ) , where @xmath304 at the surface of the superconductor so that no de - paired electrons exist in the superconductor , whereas the case of @xmath9 above about @xmath344 corresponds to the case shown in figs . 7(b ) , 7(c ) , 8(b ) and 8(c ) , where @xmath345 at the surface of the superconductor so that de - paired electrons exist in the region near the surface of the superconductor . as shown in figs . 9(a ) and 9(b ) , @xmath346 when @xmath347 , or @xmath348 [ here @xmath349 is the inverse function of @xmath350 . however , we should note that this is not true for a type - ii superconductor , for which superconductivity is completely suppressed only at the upper critical magnetic field @xmath351 , which is usually much higher than @xmath13 for a highly - local superconductor . this is because the definition for @xmath34 , i.e. , eq . ( [ lambda - eq ] ) , is valid only when the superconductor is in the meissner state . in the mixed state of a type - ii superconductor , in which vortices exist , this definition for @xmath34 is no longer valid . further , we note that the definition of eq . ( [ lambda - eq ] ) is valid only for bulk samples with dimensions much larger than the magnetic field penetration depth . we consider an isolated vortex in an infinite superconductor . let the vortex be centered on the @xmath278-axis . in terms of the cylindrical coordinates @xmath352 and the unit vectors @xmath353 , we can write @xmath354 , @xmath355 , and @xmath356 . then , eq . ( [ haoeq2 ] ) and the relation @xmath284 become @xmath357 and @xmath358 respectively , where a `` prime '' indicates a derivative with respect to @xmath359 ; @xmath360 is given by eq . ( [ ns - eq ] ) , where @xmath361 is determined by eq . ( [ haoeq1 ] ) ; and the relation @xmath362 has been used . this is a non - linear second - order boundary - value problem with boundary conditions @xmath363 and @xmath364 the last boundary condition comes from flux quantization , i.e. , @xmath365 in conventional units , where @xmath366 is the flux quantum . the parameter @xmath367 is defined as @xmath368 note the difference between the present definition for @xmath367 and the one in the ginzburg - landau ( gl ) theory:@xcite @xmath369 . besides the difference between @xmath370 for @xmath367 of eq . ( [ kappa ] ) and @xmath371 for @xmath372 , there is an extra factor @xmath373 in the expression for @xmath372 . in the gl theory,@xcite @xmath372 can also be expressed as a ratio between the magnetic field penetration depth and the coherence length . a similar expression for @xmath367 does not exist in the present theory , because the coherence effect ( or the non - local effect ) is not accounted for in the present theory . for the convenience of numerical calculation , the above - described second - order boundary - value problem has been expressed as a system of two first - order differential equations . the numerical method for solving this boundary - value problem is explained in appendix [ computevortex ] . numerical results for @xmath374 , @xmath375 , @xmath376 , @xmath377 , @xmath378 and @xmath379 near the vortex core are shown in fig . 10 for @xmath380 at several different temperatures as indicated in the figure . here , as an example of @xmath381 , @xmath378 is shown in the figure . let @xmath382 denote the location at which @xmath311 . we have @xmath383 for @xmath384 , and @xmath385 for @xmath386 . paired electrons ( with @xmath305 ) and de - paired electrons ( with @xmath7 ) co - exist in the region @xmath384 , whereas no de - paired electrons exist ( i.e. , @xmath6 for all the electrons ) in the region @xmath387 . , @xmath375 , @xmath376 , @xmath377 , @xmath378 , and @xmath379 near the vortex core at @xmath388 ( a ) , @xmath319 ( b ) , and @xmath389 ( c ) , for @xmath380 . the quantities @xmath374 , @xmath375 , @xmath376 , @xmath377 , @xmath378 , and @xmath379 are measured in units of @xmath390 , @xmath391 , @xmath392 , @xmath322 , @xmath323 , and @xmath324 , respectively . ] versus @xmath367 at different values of @xmath204 as indicated on the curves . the two curves for @xmath325 and @xmath389 are practically indistinguishable from each other . the dotted curve shows the ginzburg - landau results for @xmath393 in the high-@xmath367 limit . ] as shown in fig . 10 , as @xmath394 , we have @xmath395 , @xmath396 , @xmath397 and @xmath398 . it also appears that @xmath375 , @xmath376 and @xmath377 have zero slopes at @xmath399 , while @xmath379 has a finite slope at @xmath399 and is nearly linear for small @xmath359 . the electrical current density @xmath376 has a maximum located near @xmath382 . the numerical solutions obtained in the last subsection for an isolated vortex can be used to calculate lower critical magnetic field @xmath400 of the superconductor . by definition,@xcite at @xmath401 , the gibbs free energy must be the same whether the first vortex exists or not , i.e. , @xmath402 at @xmath401 . this condition leads to @xmath403 numerical results for @xmath400 are shown in fig . 11 as @xmath393 versus @xmath367 for several different temperatures . the dotted curve in fig . 11 shows the ginzburg - landau result for @xmath393 for high-@xmath367 superconductors:@xcite @xmath404 where we have used @xmath405 by replacing @xmath406 with @xmath370 in the expression for @xmath372 . as shown in fig . 11 , our result for @xmath393 is only weakly @xmath1-dependent for low temperatures , and nearly @xmath1-independent for intermediate and high temperatures , as indicated by the fact that the two @xmath393-versus-@xmath367 curves for @xmath407 and @xmath389 are practically indistinguishable from each other . this feature is to be compared with that @xmath408 is @xmath1-independent in the ginzburg - landau theory . figure 11 also shows that our result for @xmath393 and the ginzburg - landau result are quantitatively not very different for @xmath14 . for lower values of @xmath367 , the present theory underestimates the value of @xmath400 , because the coherence ( or non - local ) effect , which increases the energy associated with a vortex and thus leads to a larger @xmath400 , is not accounted for in the present theory . we have presented a microscopic theory for superconductivity in a magnetic field based on a local approximation approach . the theory allows microscopic description of the suppression of superconductivity by an externally applied magnetic field . in sec . [ sectheory ] , we presented the details of the theory . the main results derived in sec . [ sectheory ] include an expression for free energy density @xmath0 as a function of temperature @xmath1 and vector potential * a * , and two basic equations of the theory : the first is an implicit solution for energy gap parameter amplitude @xmath2 as a function of wave vector * k * , temperature @xmath1 and vector potential * a * ; and the second is a london - like relation between electrical current density * j * and vector potential * a * , with an `` effective superconducting electron density '' @xmath3 that is both @xmath1- and * a*-dependent . the two equations allow determination of the spatial variations of * a * and @xmath4 in a superconductor for given temperature @xmath1 , applied magnetic field @xmath5 and sample geometry . in the low - field limit , the theory reduces to the local - limit result of bcs . we also numerically analyzed the functions @xmath17 , @xmath164 and @xmath165 ( where @xmath160 is the condensation energy density ) . in sec . [ secappl ] , as examples , we applied the theory to the case of a semi - infinite superconductor in an applied magnetic field @xmath5 parallel to the surface of the superconductor and the case of an isolated magnetic vortex in an infinite superconductor , and determined , for each case , spatial variations of quantities such as * a * , @xmath4 , @xmath3 and @xmath160 . we also calculated magnetic field penetration depth @xmath10 and lower critical magnetic field @xmath11 . an important conclusion of the theory is that , depending on temperature @xmath1 , applied magnetic field @xmath5 and sample geometry , a `` partly - paired state '' can exist in which paired electrons ( having @xmath6 ) and de - paired electrons ( having @xmath7 ) co - exist . such a `` partly - paired state '' exists even at @xmath8 when @xmath9 is above a threshold for a given sample , giving rise to a non - vanishing knight shift in the superconducting state at @xmath8 for @xmath9 above the threshold . since the non - local effect ( or coherence effect ) in the superconducting state is not accounted for in the present theory , we expect the theory to be valid only for highly - local superconductors ( for which magnetic field penetration depth @xmath34 is much larger than coherence length @xmath45 ) . however , when a more complete theory is developed that is able to account for the non - local effect ( or coherence effect ) , we expect it to reduce to the present theory in the local limit . for @xmath32 , eq . ( [ haoeq ] ) was previously derived by the author in ref . ( related discussions are also given in refs . and ) . the derivation of eq . ( [ haoeq ] ) for @xmath31 is similar to that for @xmath32 . we now present the details of the derivation of eq . ( [ haoeq ] ) . for convenience , we define @xmath409 which is a real number . then , the self - consistency equation , eq . ( [ gap - eq ] ) , can be rewritten as @xmath410 in the presence of an applied magnetic field , we expect @xmath103 to be a function of temperature @xmath1 and vector potential @xmath411 , where @xmath412 ( @xmath413 ) are the components of * a*. let @xmath414 denote any one of @xmath1 , @xmath415 , @xmath416 and @xmath417 . we operate @xmath418 on both sides of eq . ( [ agapeq ] ) to obtain @xmath419 we next multiply both sides of the above equation by @xmath420 , and then take summation over * k * , i.e. , @xmath421 the quantity inside the first pair of parentheses on the right - hand side of the above equation equals to @xmath422 [ according to eq . ( [ agapeq ] ) ] so that the second of the two terms on the right - hand side is the same as the term on the left - hand side . thus , we have @xmath423 we want a @xmath170 solution . clearly , @xmath424 which is eq . ( [ haoeq ] ) and satisfies @xmath425 is a solution of eq . ( [ asum ] ) . however , since eq . ( [ ahaoeq ] ) is not the only possible solution of eq . ( [ asum ] ) [ as one can see , eq . ( [ asum ] ) actually can have an infinite number of solutions ] , we need to justify that eq . ( [ ahaoeq ] ) is the only physical solution . since the diagonalized hamiltonian , eq . ( [ h ] ) , describes a set of independent quasi - particle excitations , there should be no coupling ( except pair correlation ) between the quasi - particle excitations . therefore , we expect the thermal energy and entropy associated with each pair of @xmath426 excitations to be @xmath427 and @xmath428 , \label{ask}\end{aligned}\ ] ] respectively . similarly , we expect the contribution to the electrical current density from each pair of @xmath429 excitations to be @xmath430 [ here @xmath431 for free electrons , and @xmath432 corresponds to the second term on the right - hand side of eq . ( [ haoeq2 ] ) ] . however , as compared to the above expressions for @xmath433 , @xmath434 , and @xmath435 , those derived from the diagonalized hamiltonian contain additional terms involving @xmath436 , @xmath437 and @xmath438 ( where @xmath439 in the case of @xmath433 or @xmath434 ; and @xmath440 in the case of @xmath435 ) . letting the sum of the additional terms to be zero , one gets eq . ( [ partialck ] ) , and therefore eq . ( [ ahaoeq ] ) . the boundary - value problem of sec . [ secsemiinfinitesc ] , as specified by the system of eqs . ( [ bprime ] ) and ( [ aprime ] ) , with boundary conditions given by eqs . ( [ bcat0 ] ) and ( [ bcatinfty ] ) , can be solved by using the runge - kutta method.@xcite in order to use the runge - kutta method , we first need to know values of @xmath135 and @xmath441 at one point on the @xmath287-axis . we know that deep inside the superconductor both @xmath135 and @xmath441 become small , and @xmath3 becomes an @xmath287-independent constant , so that the london solutions for @xmath135 and @xmath441 hold , which give the following relation between @xmath135 and @xmath441 : @xmath442 let @xmath310 denote the coordinate of such a point located deep inside the superconductor . since both eqs . ( [ bprime ] ) and ( [ aprime ] ) do not involve coordinate @xmath287 explicitly , we can assign an arbitrary value to @xmath310 . we then assign a sufficiently small value for @xmath443 , and obtain @xmath444 from the above relation , where @xmath445 is obtained from eq . ( [ ns ] ) for @xmath260 ( since the @xmath135-dependence of @xmath3 is negligible when @xmath135 is small ) and given @xmath1 . once we know @xmath443 and @xmath444 at @xmath446 , we use the runge - kutta method to compute @xmath447 and @xmath448 for @xmath449 , where @xmath450 is a small positive interval and @xmath451 , until @xmath452 at @xmath453 . usually , @xmath454 is greater than @xmath9 . if the difference between @xmath454 and @xmath9 is small , we can simply use @xmath455 as coordinate @xmath456 of the surface of the superconductor . or , we can obtain @xmath456 by making a linear interpolation : @xmath457 the boundary - value problem of sec . [ secsinglevortex ] , as specified by the system of eqs . ( [ bprime2 ] ) and ( [ aprime2 ] ) , with boundary conditions given by eqs . ( [ bcatinfty ] ) and ( [ bcfluxquantum ] ) , can be solved by using the runge - kutta method.@xcite in order to use the runge - kutta method , we first need to know values of @xmath135 and @xmath441 at one point on the @xmath359-axis . we know that far away from the vortex core , both @xmath135 and @xmath441 become small and @xmath3 becomes a constant . for a constant @xmath3 , eqs . ( [ bprime2 ] ) and ( [ aprime2 ] ) can be solved analytically ( see , for example , ref . ) ): @xmath458 and @xmath459 where @xmath460 is a constant to be determined , and @xmath461 are the modified bessel functions of the second kind . let @xmath382 denote the @xmath359-coordinate of such a point located far away from @xmath399 , the center of the vortex . we choose a value for @xmath382 that is sufficiently large , and guess an initial value for constant @xmath460 , say @xmath462 , and obtain @xmath463 and @xmath464 from eqs . ( [ aatr0 ] ) and ( [ batr0 ] ) , respectively , where @xmath3 is obtained from eq . ( [ ns ] ) for @xmath260 ( since the @xmath135-dependence of @xmath3 is negligible when @xmath135 is small ) and given @xmath1 . once we know @xmath463 and @xmath464 at @xmath465 , the runge - kutta method allows us to compute @xmath374 and @xmath375 for any @xmath359 . we then compute total magnetic flux @xmath466 associated the vortex by numerically carrying out the integral on the right - hand side of eq . ( [ bcfluxquantum ] ) . if total magnetic flux @xmath467 for @xmath468 is , for example , greater than flux quantum @xmath366 , we reassign a smaller value for @xmath460 , say @xmath469 , and repeat the computation of @xmath374 , @xmath375 , and @xmath466 . the computation of @xmath374 , @xmath375 and @xmath466 is repeated until the difference between total magnetic flux @xmath466 and flux quantum @xmath366 is within a predetermined range . in practice , it usually involves only a few repetitions . v. l. ginzburg and l. d. landau , j. exptl . ( ussr ) * 20 * , 1064 ( 1950 ) [ emglish translation in _ men of physics : l. d. landau _ , vol . 1 , edited by d. ter haar , pergamon press , new york ( 1965 ) ] .

the carbon dimer , thec molecule , is ubiquitous in astronomical environments . experimental - quality rovibronic energy levels are reported forc , based on rovibronic transitions measured for and among its singlet , triplet , and quintet electronic states , reported in 42 publications . the determination utilizes the measured active rotational - vibrational energy levels ( marvel ) technique . the 23,343 transitions measured experimentally and validated within this study determine 5,699 rovibronic energy levels , 1,325 , 4,309 , and 65 levels for the singlet , triplet , and quintet states investigated , respectively . the marvel analysis provides rovibronic energies for six singlet , six triplet , and two quintet electronic states . for example , the lowest measurable energy level of the state , corresponding to the total angular momentum quantum number and the spin - multiplet component , is 603.817(5 ) . this well - determined energy difference should facilitate observations of singlet triplet intercombination lines which are thought to occur in the interstellar medium and comets . the large number of highly accurate and clearly labeled transitions that can be derived by combining marvel energy levels with computed temperature - dependent intensities should help a number of astrophysical observations as well as corresponding laboratory measurements . the experimental rovibronic energy levels , augmented , where needed , with _ ab initio _ variational ones based on empirically adjusted and spin - orbit coupled potential energy curves obtained using the duo code , are used to obtain a highly accurate partition function , and related thermodynamic data , forc up to 4,000 k. llll singlet & phillips & & + & mulliken & & + & _ herzberg f _ & f & + & bernath b & & + & bernath b & b & + & _ deslandres dazambuja _ & & + & _ messerle krauss _ & c & + & freymark & & + & _ goodwin cool a _ & 1 a & + & _ goodwin cool b _ & 1 b & + triplet & ballik ramsay & & + & swan & & + & fox herzberg & e & + & _ herzberg f _ & f & + & _ herzberg g _ & g & + & krechkivska schmidt & 4 a & + & duck & & + & _ kable schmidt _ & e & + quintet & radi bornhauser & & + intercombination & triplet - singlet & & + & quintet - triplet & & + & singlet - triplet & & + + [ tab : bands ] c showing the bands considered in this work . the dashed lines represent observed but unnamed intercombination bands.,height=604 ]

1605.01504

the rovibronic spectra of @xmath0c@xmath1 , involving singlet , triplet , and quintet electronic states , are rich and complex in features as well as in anomalies . this is true despite the fact that @xmath0c@xmath1 , hereafter simply called c@xmath1 , is a homonuclear diatomic molecule containing only 8 valence electrons . there are 62 electronic states corresponding to the six possible separated - atom limits formed by different pairs of the c(@xmath21p ) , c(@xmath22d ) , and c(@xmath22s ) atoms . 18 of these states , six singlet , six triplet , and six quintet ones , correspond to the c(@xmath21p ) + c(@xmath21p ) asymptote . as summarized in table 1 and figure 1 , the measured rovibronic spectroscopy of c@xmath1 is characterized by 19 known band systems , and a further three intercombination transitions , which cover the spectral range @xmath23 @xmath4 and now all three spin states . due to the lack of detailed transition data , see section 2.4 for details , only 14 band systems are considered in the present study . all the states considered , except the and the 4 @xmath24 states , share the same c(@xmath21p ) + c(@xmath21p ) separated - atom limit . the present study was initiated as , rather unusually , at least six of these bands have been used for astronomical observations , with a further one having been proposed for such use . there are numerous astronomical observations of c@xmath1 spectra with a multitude of important applications for astrophysics . c@xmath1 spectra have been observed in comets @xcite , in high - temperature stars @xcite , translucent clouds @xcite , and in the low - temperature interstellar medium @xcite . high - quality studies of rotational and spatial distributions of c@xmath1 in comets are available @xcite . the astronomical observations and their modelling are supported by a number of laboratory studies @xcite . to make maximum use of the observations one needs accurate rovibronic energy level information , the principal topic of the present paper . both the swan @xcite and the deslandres @xcite bands have been observed in fluorescence from comets . indeed , two singlet , and , and four triplet , , , , and d @xmath24 , electronic states are needed to model c@xmath1 emission observed in comets . two spin - forbidden transition systems , @xmath25 and @xmath25 , are needed to explain the observed intensities in the swan band @xcite . the astronomical observations include solar spectra , where c@xmath1 forms an important component of the photospheric carbon abundance @xcite . so , for example , c@xmath1 can be observed in the sun s photosphere at visible wavelengths using the swan band @xcite and in the infrared via the phillips and ballik ramsay bands @xcite . transitions in the swan band have also been observed in peculiar white dwarfs @xcite and the coronae borealis star v coronae australis @xcite , while the phillips band is prominent , for example , in the carbon star hd19557 @xcite . transitions in the ballik ramsay band have also been observed in carbon stars @xcite . interstellar c@xmath1 has been observed via the infrared phillips band , for example in the perseus molecular complex @xcite and toward the cyg ob2 association @xcite . emission features from the swan band have been observed in the red rectangle @xcite . @xcite observed absorbotion in the phillips , mulliken and herzberg f bands in their study of translucent clouds by the simultaneous use of observations from both space and ground - based telescopes . the rovibrational manifolds of the and states strongly overlap as the corresponding electronic excitation energy is less than half of the vibrational spacing in either state . @xcite suggested that long - wavelength transitions between levels in the band should provide a good method for monitoring interstellar c@xmath1 . @xcite attempted to observe such lines in the hale bopp comet , without success . the accurate rovibronic energy levels presented in this paper can be used to provide accurate transitions for such features . as for the other bands , at ultraviolet wavelengths the international ultraviolet explorer ( iue ) was used to observe c@xmath1 towards x persei via the herzberg f band @xcite . the hubble space telescope also has been used to monitor c@xmath1 absorption in the ultraviolet using both the mulliken and the herzberg f bands @xcite . finally , we note that all of the bands mentioned are important in laboratory plasmas @xcite and flames @xcite . for example , there are particular vibrational bands of the swan system known as the high pressure ( hp ) bands @xcite which can be prominent in such environments at atmospheric pressure . studies of the spectra , and thus the band systems , of c@xmath1 date all the way back to @xcite and the dawn of spectroscopy , when c@xmath1 emissions were first observed in flames . detailed studies by @xcite of the most prominent band of c@xmath1 also predate the development of the quantum mechanical tools required to interpret the spectroscopic results obtained . over the following century and a half , a number of other bands have been detected and studied . indeed , four band systems have been identified during the last decade @xcite . table [ tab : bands ] lists the known band systems of c@xmath1 relevant for the present study . a large number of spectroscopic measurements exist for the different band systems of c@xmath1 . all the studies which contain primary measured transitions at a reasonable level of accuracy are considered during the present analysis . the experimental papers found and analysed by us are listed here based on the band systems : phillips @xcite , mulliken @xcite , bernath b @xcite , bernath b@xmath26 @xcite , freymark @xcite , ballik ramsay @xcite , swan @xcite , fox herzberg @xcite , deslandres dazambuja @xcite , messerle krauss @xcite , + goodwin cool @xcite , duck @xcite , krechkivska - schmidt @xcite and radi - bornhauser @xcite , as well as various spin - changing intercombination bands @xcite . it is interesting to note that for a long time a triplet state , now named , was believed to be the lowest electronic state of c@xmath1 and not a singlet state , the true ground electronic state , . the source of the confusion , detailed in @xcite , is that the energy difference between the two electronic states is only about one half of the vibrational fundamental of either state so that the swan system can be seen in absorption in many sources . the spectroscopy and the spectroscopic constants of c@xmath1 have been the subject of several reviews @xcite . @xcite reported the analysis of seven singlet and seven triplet states . @xcite reviewed the then available experimental and theoretical results . the spectroscopic and kinetic properties of c@xmath1 , including 23 electronic states studied prior to 1992 , were reviewed by @xcite . @xcite reviewed the spectroscopy of small carbon clusters which , once again , covered the extensive spectroscopic literature available for c@xmath1 . the present paper surveys all rovibronically - resolved measurements made up to the end of 2015 . a large number of _ ab initio _ computations have been performed on c@xmath1 @xcite . notable among the sophisticated electronic structure computations are those of bacskay @xcite ; for example , they led to the observation of the duck band and enabled the identification of the krechkivska schmidt band system . @xcite pointed out that `` at the present state of the art , theoretical pecs that reproduce the rotational - vibrational levels to spectroscopic accuracy ( 1 @xmath4 ) or near spectroscopic accuracy ( 10 @xmath4 ) are considered highly accurate . '' reproduction of measured electronic excitation energies of c@xmath1 appears to be even more problematic , an accuracy of a few hundred @xmath4 seems to be the norm . thus , it is still of interest to perform accurate quantum chemical computations on c@xmath1 and its rovibronic states . results of a preliminary , first - principles analysis are reported here ( _ vide infra _ ) , used in particular for checking the experimental transitions and energy levels . the main body of the present work is a marvel ( measured active rotational - vibrational energy levels ) @xcite analysis of the measured rovibronic states of c@xmath1 . our analysis has been in progress for four years but a real breakthrough came with the study of @xcite , who observed 16 forbidden transitions between singlet and triplet states . this study coupled , for the first time , the singlet and triplet components of the observed spectroscopic network ( sn ) @xcite of c@xmath1 , allowing a much improved analysis of its rovibronic energy level structure . we analyze all the known bands of c@xmath1 with the exception of the five vuv bands due to @xcite and @xcite . these bands involve upper energy levels which arise from a single experiment and are too high for us to be able to independently validate them . finally , when a complete set of accurate energy levels are available for a molecule , they can be used , via the direct summation technique , to compute accurate ideal - gas thermodynamic functions , most importantly the high - temperature internal partition function , @xmath27 . we do this here for c@xmath1 , complementing and improving several previous efforts @xcite and arrive at very precise and accurate values for @xmath27 of c@xmath1 up to 4,000 k. . the blue , red , and green dots correspond to the singlet , triplet , and quintet rovibronic states , respectively . the dots represent energy levels and only some of the transitions are visible , especially those connecting the singlet and the triplet as well as the triplet and the quintet rovibrational states.,height=566 ] the marvel @xcite procedure and code is used in this study to obtain rovibronic energies of c@xmath1 by inverting all assigned experimental rovibronic transitions available in the literature for this molecule . marvel is based on the concept of spectroscopic networks ( sn ) @xcite : sns are large , finite , weighted , rooted graphs , where the vertices are discrete energy levels ( with associated uncertainties ) , the edges are transitions ( with measured uncertainties ) , and in a simple picture the weights are provided by the transition intensities . no weights are considered during the present study . within the experimental sn of a molecule there can be several rooted components and several floating ones . for many molecules the rooted components belong to ortho and para nuclear spin isomers . however , the nuclear spin of @xmath0c is zero ; thus , @xmath0c@xmath1 has only one nuclear spin isomer and one root for its lowest electronic state , . transitions between electronic states of different spin multiplicity are spin forbidden ; thus , until 2011 no lines were measured and assigned experimentally between the singlet and triplet and the triplet and quintet manifold of states of c@xmath1 . first , @xcite observed transitions linking triplet and quintet electronic states , leading to the first observation and characterization of a quintet band of c@xmath1 . later , @xcite managed to identify 16 spin - forbidden transitions between singlet and triplet states . these spin - forbidden transitions proved to be particularly important during the marvel analysis of the experimental spectra of c@xmath1 . using these forbidden transitions the experimental sn of c@xmath1 simplifies as now it contains only one principal component ( pc ) . the label of the root of the pc is \{0 + 0 @xmath3 @xmath28 } , for the notation employed for the label see subsection [ s : qn ] . the measured sn of c@xmath1 is shown pictorially in figure 2 , the singlet , triplet , and quintet energy levels are indicated with different colors . since figure 2 is a particular representation of a network ( @xmath29 , a graph ) , the arrangement of the nodes ( energy levels ) and links ( transitions ) is arbitrary , but clearly displays several important characteristics of the sn of c@xmath1 . figure 2 vividly shows , for example , how weakly the singlet rovibrational energy levels are connected to the triplet core . during a marvel analysis we simultaneously process all the available assigned and labelled experimental transitions . the energy levels are obtained from the set of transitions via a weighted linear least - squares inversion protocol . as the marvel technique has been employed to determine experimental - quality energy levels of nine isotopologues of water @xcite , of three isotopologues of h@xmath30 @xcite , as well as of ammonia @xcite and ketene @xcite , the interested reader is referred to these publications for details about the different stages of a marvel analysis . since c@xmath1 has both regular and inverted triplet states ( see the next subsection for details ) , for example the and of c@xmath1 are inverted , during the marvel analysis it was checked whether the labels of all experimental transitions follow the same convention . at the beginning , the marvel analysis of the spectra of c@xmath1 was complicated by the fact that there have been no truly high - accuracy variational or effective hamiltonian rovibronic energy levels available for this molecule at higher energies . this situation greatly improved by the first - principles duo @xcite analysis performed as part of this study ( _ vide infra _ ) , allowing at least a preliminary validation of the marvel levels up to 35,000 @xmath4 . for marvel to work properly one needs appropriate and unique labels . the label used in the present study for a rovibronic energy level is built up from information concerning the uncoupled electronic state , the vibrational and rotational quantum numbers , and the symmetry of the rovibronic state . the label finally chosen for each energy level , @xmath31 , fully characterizes the rovibronic states of a homonuclear diatomic molecule , such as c@xmath1 , but contains redundant information @xcite . here @xmath32 is the quantum number corresponding to the total angular momentum ( an integer value ) characterizing the state , @xmath33 labels the total state parity , @xmath34 , in terms of the laboratory - fixed inversion operator @xmath35 @xcite , ` state ' is the customary term symbol of the electronic state before spin - orbit coupling is taken into account ( _ e.g. _ , x @xmath36 and b @xmath37 ) , and @xmath38 is the vibrational quantum number . the @xmath39 , @xmath40 , label denotes the spin multiplet components : for singlet states @xmath41 , _ i.e. _ , @xmath42 , for triplet states @xmath43 , which corresponds to the standard spectroscopic notation @xmath44 , and @xmath45 ( @xmath3 , @xmath46 , @xmath45 refer to levels with @xmath47 , where @xmath48 is the quantum number corresponding to the angular momentum exclusive of nuclear and electron spin and it is usually not a good quantum number ) , and for quintet states @xmath49 @xcite . the convention @xcite is such that whether the state is regular ( @xmath50 ) or inverted ( @xmath51 ) , the energy order is always @xmath52 for the triplet states . the @xmath53 and @xmath54 subscript in the ` state ' label indicate if the symmetry of the electronic state is ` gerade ' ( positive ) or ` ungerade ' ( negative ) in terms of the molecule - fixed inversion operator , @xmath55 @xcite . the @xmath33 superscripts within the @xmath56 states indicate the parity of the electronic component with respect to vertical reflection , @xmath57 , in the molecular frame @xcite . the rotationless parity @xmath58 is a widely used alternative to the total parity @xmath33 , this redundant information is included in our marvel label to help experimentalists . for c@xmath1 the allowed combination of the parity @xmath33 and the label g / u is ( @xmath59,g ) and ( @xmath5,u ) . the rotationless parity @xmath58 is then obtained as follows : ( a ) for the @xmath28 states , the parity ( @xmath60 or @xmath61 ) can be recovered as @xmath62 , _ i.e. _ , for the even values of @xmath63 the parity of the @xmath28 states is @xmath60 , while for odd @xmath63s it is @xmath61 ; and ( b ) for the @xmath64 states , the parity is recovered as @xmath65 , _ i.e. _ , for even @xmath63s the parity is @xmath61 , and for odd @xmath63s it is @xmath60 . the rigorous electric dipole selection rules for the rovibronic transitions are @xmath66 as mentioned , the nuclear spin statistical weights @xmath67 of the nuclear spin - zero @xmath0c@xmath1 molecule for the @xmath68 and @xmath69 states is 1 , while the @xmath70 and @xmath71 states have @xmath72 . therefore , the latter states do not appear in spectroscopic experiments on @xmath0c@xmath1 . the marvel input set of measured transitions was checked for a corresponding labeling error . l l c l c l [ tab : exp ] + 15chkabeta&@xcite & @xmath73 & fts & 319/318 & + 88daabph & @xcite & @xmath74 & d - fts & 191/191 & + 88donibea & @xcite & @xmath75 & fts & 241/238 & + 77chmama & @xcite & @xmath76 & fts & 774/770 & + 63barab & @xcite & @xmath77 & & 574/532 & + 06pesi & @xcite & @xmath78 & & 8/8 & ( 2a ) + 04chyewoli & @xcite & @xmath79 & & 293/293 & ( 2b ) + 13naen & @xcite & @xmath80 & lif & 77/73 & + + + 39landsver & @xcite&@xmath81 & cae & 171/167 & ( 2c ) + 97sobllixu & @xcite & @xmath82 & lif & 179/165 & ( 2c ) + 95blliso & @xcite & @xmath83 & dl & 9/8 & + + + 88donibeb & @xcite & @xmath84 & fts & 507/507 & + 16chkabeta & @xcite & @xmath85 & fts & 1001/1001 & + + + 88donibeb & @xcite & @xmath86 & fts & 237/237 & + + + 97sobllixu & @xcite & @xmath87 & lif & 66/65 & ( 2c ) + 51freymark & @xcite&@xmath88 & & 376/354 & ( 2d ) + + + 15chkabeta & @xcite & @xmath89 & fts & 3510/3507 & + 85yacumeca & @xcite & @xmath90 & & 356/352 & ( 2e ) + 85rowamive & @xcite & @xmath91 & & 1309/1298 & ( 2f ) + 79amchma & @xcite & @xmath92 & & 2168/2139 & + 88daabsa & @xcite & @xmath93 & d - fts & 382/365 & + 06pesi & @xcite & @xmath94 & & 80/80 & + 11bosyknge & @xcite & @xmath95 & & 8/8 & + + + 13naen & @xcite & @xmath96 & lif & 168/168 & + 14naen & @xcite & @xmath97 & lif & 150/141 & + 13bosyknge & @xcite & @xmath98 & & 23/23 & ( 2 g ) + 13yechwa & @xcite&@xmath99 & hcd & 276/273 & ( 2h ) + 07tahiam & @xcite & @xmath100 & fts & 3853/3771 & + 48phillips & @xcite&@xmath101&discharge&1181/1128 & + 02taam & @xcite & @xmath102 & hcd & 356/352 & + 03kayaguyu & @xcite & @xmath103 & & 153/150 & ( 2i ) + 85cusa & @xcite & @xmath104 & doppler - free&217/217 & ( 2j ) + 94prbe & @xcite & @xmath105 & jet cooled&39/39 & ( 2j ) + 83amiot & @xcite & @xmath106 & fts & 347/346 & + 99llew & @xcite & @xmath107 & & 138/138 & + 85susahi & @xcite & @xmath108 & & 194/194 & ( 2j ) + 10boknge & @xcite & @xmath109 & & 23/23 & ( 2j ) + 11bosyknge & @xcite&@xmath110 & & 46/46 & + + + 86hawi & @xcite & @xmath111 & & 100/100 & ( 2k ) + 49phillips & @xcite&@xmath112&discharge&1833/1664 & + 98brhakocr & @xcite & @xmath113 & & 10/10 & ( 2l ) + + + 13chyewa & @xcite&@xmath114 & hcd & 221/210 & + 13naen & @xcite & @xmath115 & lif & 513/513 & + 14naen & @xcite & @xmath116 & lif & 205/196 & + 07jonare & @xcite&@xmath117 & lif & 235/234 & + + + 15krbatrna & @xcite & @xmath118 & rempi & 67/67 & + 16krbawena & @xcite & @xmath119 & & 1/1 & ( 2 m ) + + + 15chkabeta & @xcite&@xmath120 & fts & 32/32 & + + + 15chkabeta & @xcite & @xmath121 & fts & 1/1 & + + + 11bosyknge & @xcite&@xmath122 & fwm & 68/68 & + + + 15bomago & @xcite & @xmath123 & fwm & 57/57 & + l l c l c l [ tab : expnot ] + 82erlama & @xcite & & & no & ( 3a ) + + + 63cagi & @xcite&&&no & + 68meme & @xcite & & & no & ( 3b ) + 68phdaxx&@xcite & & & yes & ( 3c ) + 88urbaja&@xcite&21,280 - 25,930&&no + 90chma & @xcite & & & no + 97kahuew&@xcite & & & no & ( 3d ) + 94cado&@xcite & & & no & ( 3e ) + 99llew & @xcite & & & yes & ( 3f ) + 05smpasc & @xcite & & cl & no + + + 63baraa & @xcite & & & yes & ( 3 g ) + + @xmath5b@xmath26 @xmath124 + 91baurja&@xcite&28,030 - 28,555 & lif & no + + + 06koremona&@xcite&&&no + + + 09najopare&@xcite&&&no&(3h ) + + + 69helama&@xcite & 69,000 - 73,000 & fd & yes & ( 3i ) + + + 96blliso&@xcite&43,510 - 43,545&lif & no . + + + 85anbope&@xcite&23,800 - 31,250 & e & & ( 3j ) + 87vahe & @xcite & 21,000 - 24,500 & lif & no & ( 3k ) + 88urbaja&@xcite&21,280 - 25,930 & lif & no & ( 3l ) + + + 67mekr&@xcite&&&no&(3 m ) + + + 88gocoa & @xcite & & & no&(3n ) + 88gocob & @xcite & & & yes&(3o ) + 89goco & @xcite & & & yes & ( 3p ) + first , we make comments on those observed bands of c@xmath1 which were not used in the marvel analysis . the situation with the deslandres dazambuja band is a remarkable one . the band was originally observed more than a century ago @xcite and , as discussed in the introduction , has been observed in a number of astronomical objects . however , while a high - resolution line list is available for @xmath125c@xmath1 @xcite and there are a number of papers reporting laboratory observation of the band for @xmath0c@xmath1 @xcite , there are no high resolution line data for the band . a high resolution re - measurement of the band for @xmath0c@xmath1 would appear to be welcome . observation of the messerle krauss band is reported in a single , short paper @xcite , which provides no line data . similarly , there are so far no published lines for the recently detected kable schmidt band @xcite . the situation with the three herzberg and two goodwin cool vuv bands is somewhat different . there are papers reporting detailed spectra for each of these bands @xcite . however , each of them comes from a single uncorroborated measurement . as these bands probe upper states which are too high in energy for reliable , independent theoretical predictions , it was decided to leave their inclusion in a marvel analysis for future work . we note that omitting the energy levels associated with these bands is not critical for the partition sums and thermodynamic data determined as part of this study . we also note that papers by @xcite refer to the so - called leblanc band comprising weak @xmath5 b@xmath26 @xmath126 transitions ; however , we could find no line data on this band or any papers by leblanc reporting it . second , the set of comments below refer to data sources used in our marvel analysis and listed in table 2 . ( 2a ) 06pesi @xcite uncertainty assumed to be 0.01 @xmath4 ( petrova , private communication , 2015 ) . + ( 2b ) 04chyewoli @xcite not the original data are used but those data extracted from 15chkabeta @xcite . + ( 2c ) 97sobllixu @xcite and 39landsver @xcite uncertainty of these two sources is assumed to be 1.0 and 0.2 @xmath4 , respectively . + ( 2d ) 51freymark @xcite the stated uncertainty was doubled to 0.02 @xmath4 , as this seems to be a more adequate guess of the accuracy of these transitions . + ( 2e ) 85yacumeca @xcite results recorded using magnetic rotation . the data are provided by curl ( private communication , 2015 ) . the uncertainty was increased to 0.005 @xmath4 as values below this did not give consistent results during the marvel analysis . + ( 2f ) 85rowamive @xcite this source presents an analysis of measurements by @xcite ; the data were extracted from 15chkabeta @xcite . + ( 2 g ) 13bosyknge @xcite new assignments and reassignments of the measurements by 13yechwa @xcite . + ( 2h ) 13yechwa @xcite assignments corrected following 13bosyknge @xcite . + ( 2i ) 03kayaguyu @xcite 0.007 @xmath4 uncertainty ; includes 26 transitions with @xmath127 . + ( 2j ) 85cusa @xcite , 85susahi @xcite , 94prbe @xcite and 10boknge @xcite data extracted from 13brbescba @xcite . + ( 2k ) 86hawi @xcite . data provided by hardwick ( private communication , 2015 ) . + ( 2l ) 98brhakocr @xcite uncertainty assumed to be 0.02 @xmath4 . + ( 2 m ) 16krbawena @xcite give a single line position as part of their measurement of their multiphoton spectroscopic measurement of the ionization energy of c@xmath1 . + 0.2 cm third , the next set of comments refer to sources listed in table 3 ; these were not used in our marvel analysis for reasons listed below . ( 3a ) 82erlama @xcite this source contains lifetime data but no transition frequencies . + ( 3b ) 68meme @xcite no actual line data given in the paper but 11bosyknge @xcite presents some lines reassigned from this work which are included in our dataset , see table [ tab : exp ] . + ( 3c ) 68phdaxx @xcite a book with an extensive list of lines including higher bands but only 3607 of the 10910 lines could be be validated ; @xcite also found that these assignments did not match those of other work . it was therefore decided to omit these lines from the final compilation . + ( 3d ) 97kahuew @xcite a precursor to 99llew @xcite . + ( 3e ) 94cado @xcite so - called high pressure ( hp ) band . + ( 3f ) 99llew @xcite none of these data were selected by 13brbescba @xcite , so they were not considered in this study either . + ( 3 g ) 63baraa @xcite original observation by ballik and ramsay of their eponymous band . the work was re - assigned by @xcite , but his results are not available . + ( 3h ) 09najopare @xcite report of a new c@xmath1 band but no data are provided and no follow - up study exists . + ( 3i ) 69helama @xcite report of three new vuv bands with line data . upper states lie at too high energy for data to be validated . + ( 3j ) 85anbope @xcite report of extensive data for @xmath125c@xmath1 but with no dataset for @xmath0c@xmath1 . + ( 3k ) 87vahe @xcite report of a new c@xmath1 band : 25 transitions from one ( unknown ) band were recorded with an uncertainty of 0.2 @xmath4 and assigned @xmath63 quantum numbers ; these were supplied by heaven ( private communication , 2015 ) . the calculations of @xcite suggest that this band actually belongs to c@xmath128 rather than c@xmath1 . the band remains unassigned and the data were not included in our current analysis . + ( 3l ) 88urbaja @xcite no transition data reported in the paper . + ( 3 m ) 67mekr @xcite discovery paper giving many spectroscopic parameters but no primary transition data . + ( 3n ) 88gocoa @xcite discovery paper : follow - up work with data in 88gocob @xcite and 89goco @xcite . + ( 3o ) 88gocob @xcite new band 1 @xmath129 a @xmath11 with line data . upper states lie at too high energy for data to be validated . + ( 3p ) 89goco @xcite new band 1 @xmath129 b @xmath12 with line data . upper states lie at too high energy for data to be validated . in order to decide on their correctness , we have compared the experimental marvel rovibronic energies with their theoretical counterparts . the latter approximate but complete set of energy levels is based on empirical potential energy curves ( pec ) , spin - orbit curves ( soc ) , and electronic angular momentum curves ( eamc ) of c@xmath1 as given by @xcite . the theoretical rovibronic energies were computed using a new diatomic nuclear motion program called duo @xcite . duo solves the fully - coupled rovibronic schrdinger equation variationally using a combination of discrete variable representation ( dvr ) and rigid - rotor basis sets to represent the vibrational and the spin - rotational degrees of freedom in the hund s case ( a ) representation , respectively . the final pecs , socs , and eamcs were obtained by refining _ ab initio _ curves obtained at the ic - mrci / aug - cc - pvqz level of electronic structure theory for the nine lowest electronic states of c@xmath1 , , , , b@xmath26 @xmath124 , , , , , and @xmath20 by fitting to the marvel energies . a detailed account of these computations will be reported elsewhere . duo uses the following quantum numbers motivated by the hund s case ( a ) choice of the basis set : @xmath130 where @xmath131 , @xmath132 , and @xmath133 are the signed quantum numbers corresponding to projections of the electronic , spin , and total angular momenta , respectively ( the projection of the rotational angular momentum @xmath134 on the molecular axis is zero ) . the duo quantum numbers characterizing the computed rovibronic states are used to check and complete the marvel labels . it should be noted that duo uses an approximate assignment scheme based on the largest contribution to the wavefunction expansion @xcite . although this scheme is very robust , it can sometimes lead to ambiguous sets of quantum numbers , especially for states in strong resonance with other rovibronic states . comparison of the duo and marvel results helped us to identify problems in the experimental data , such as misassigned lines , duplicate transitions , and outliers . @xmath4 from the energy @xmath135 of the state.,height=472 ] lrrr|crrr [ tab : tes ] & calc.@xmath136 & empirical & duo@xmath137&empirical & marvel@xmath138 & @xmath63 & @xmath139 + & 0.0 & 0.0 & 0.0 & 0.0 & 0.0 & 0 & 1 + & 509 & 720.0083(21)@xmath140 & 722.58 & 603.828@xmath141 & 603.817(4 ) & 2 & 1 + & 6233 & 6439.08382(58)@xmath140 & 6437.00 & 6250.164@xmath141 & 6250.149(7 ) & 0 & 3 + & 8374 & 8391.4062(19)@xmath140 & 8393.63 & 8271.606@xmath141 & 8271.607(7 ) & 1 & 1 + & 9371 & 9124.2@xmath142 & 9172.30 & 9277@xmath143 & 9280.215(5 ) & 1 & 2 + & 11966 & 12082.34355(54)@xmath144 & 12092.45 & 11859@xmath143 & 11867.825(5 ) & 2 & 1 + b@xmath26 @xmath124&15261 & 15410.33(36)@xmath144 & 15401.52 & 15197@xmath143 & 15196.509(5 ) & 0 & 1 + d @xmath24 & 20092 & 20022.5@xmath142 & 20030.92 & 19992@xmath143 & 19983.953(8 ) & 2 & 1 + @xmath145 & & & & 29258.5922(48)@xmath146 & 29860.921(5 ) & 2 & 1 + & & 40796.7@xmath142 & & & 40422.691 ( 50 ) & 2 & 1 + & & 43239.8@xmath142 & & & 43230.499 ( 138 ) & 1 & 1 + @xmath147 & & & & 51 049.799@xmath148 & 51651.142(20 ) & 1 & 1 + @xmath149 & & & & & 52106.042(94 ) & 1 & 1 + e@xmath150 & & 55034.7@xmath142 & & & 54936.672(150 ) & 0 & 1 + in its ground electronic state c@xmath1 has a fairly strong bond @xcite . the equilibrium bond distance of the state at 1.2425 is short for a double bond ( considerably shorter than the c = c bond of ethylene at @xmath151(c = c ) = 1.331 ) , but long for a triple bond ( in c@xmath1h@xmath1 @xmath151(cc ) = 1.203 ) . this also means that the dissociation energy of the state is large , in fact more than 50,000 @xmath4 . compared to this , the energy differences among the several feasible asymptotes comprised by the low - lying @xmath21p , @xmath22d , and @xmath22s states of the c atom are relatively minor . furthermore , the structure of the molecular orbitals ( mo ) of c@xmath1 is such that a large number of low - energy singlet , triplet , and quintet valence states are feasible and many of them are part of experimentally measurable rovibronic transitions ( see table 1 ) . in the state the leading valence electron configuration is ( core)@xmath152 . by promoting electrons from the weakly antibonding @xmath153 and the strongly bonding @xmath154 mos and populating the @xmath155 mo a large number of electronic states arise . fortunately , for obtaining proper , temperature - dependent ideal - gas thermochemical quantities up to about 4,000 k , it is sufficient to consider 9 electronic states , four singlet , four triplet , and one quintet states ( see table 4 ) . as mentioned already , all these lowest - energy electronic states correlate with the c(@xmath21p ) + c(@xmath21p ) separated - atom limit . not too surprisingly for such a simple molecule , a large number of electronic structure computations are available for c@xmath1 in the literature @xcite . there are several issues which make the electronic structure computations extremely challenging for c@xmath1 . first , there is a quasi - degeneracy of the fully occupied @xmath156 and the empty @xmath157 mos , explaining some of the unusual characteristics of the excited electronic states of c@xmath1 . second , the existence of several low - lying excited electronic states leads to the occurrence of a considerable number of avoided crossings among the pecs as the cc distance is varied . third , as pointed out by @xcite based on full configuration interaction ( fci ) computations , at least in the cases of the x , b , and b@xmath26 states , methods based on an unrestricted hartree fock ( uhf ) reference provide correct but methods based on a restricted hf ( rhf ) reference provide incorrect results . fourth , one must account for the strong multireference character of the electronic states and the near degeneracies changing rapidly along the cc distance . fifth , rather large atom - centered , fixed - exponent gaussian basis sets are required for the correct and converged description of the valence states . these difficulties explain why this deceptively simple diatomic molecule is still one of the favorites of developers of modern wavefunction - based electronic structure techniques @xcite . using ci methods , 27 bound valence states of c@xmath1 were computed by @xcite . @xcite obtained results for all 62 electronic states in the valence manifold , including weakly bound and repulsive ones . @xcite obtained results for rydberg states , as well . electronic states with @xmath158 values up to 75,000 @xmath4 ( this is in fact the f @xmath11 state ) have been studied but for the present investigation the energy cut - off value was chosen to be 35,000 @xmath4 . this limits the number of singlet , triplet , and quintet states to 4 , 4 , and 1 , respectively , 9 states altogether . none of the higher - lying states will be considered in what follows . note that rkr potential curves are given for several singlet , triplet , and quintet states in @xcite . in this study , the so - called @xmath159 values obtained for the electronic states define directly the lowest measurable term energies of the states ( thus , they may not necessarily correspond to @xmath160 ) . it is not that simple to determine the @xmath158 values of the excited electronic states of c@xmath1 , as these are not measurable quantities . this can only be achieved if the zero - point vibrational energy ( zpve ) of all the states is determined . however , since all states are coupled in the duo computations , these do not come directly from our joint marvel and duo analysis . in particular , for the singlet @xmath161 and @xmath162 electronic states there are no transitions to @xmath160 upper rovibronic states . llrrrclcl [ tab : vib_singletfirst ] x @xmath126 & 1 & 1829.15 & 1827.4849(2 ) & 1827.486(5 ) + & 2 & 3630.35 & 3626.6835(2 ) & 3626.681(10 ) + & 3 & 5402.78 & 5396.6892(4 ) & 5396.686(9 ) + & 4 & 7145.40 & 7136.3507(6 ) & 7136.350(6 ) + & 5 & 8856.84 & 8844.1241(11 ) & 8844.124(7 ) + & 6 & 10536.37 & 10517.9659(39 ) & 10517.950(7 ) + & 7 & 12178.69 & 12154.9615(29 ) & 12154.961(6 ) + & 8 & 13783.76 & 13751.3944(38 ) & 13751.393(3 ) + & 9 & 15346.69 & 15302.8952(46 ) & 15302.893(7 ) + b@xmath163 & 1 & & 1420.4850(4 ) & 1420.488(9 ) + & 2 & & 2840.0048(4 ) & + & 3 & & 4261.0686(4 ) & 4261.071(1 ) + & 4 & & 5681.5113(6 ) & + e @xmath126 & 1 & & & 1592.316(200 ) + lrrrllrrr [ tab : vib_singletsecond ] a @xmath164 & 1 & 1 & 1584.008(4 ) & & b @xmath165 & 1 & 2 & 1384.440(2 ) + & 2 & 1 & 3143.805(5 ) & & & 2 & 2 & 2746.010(9 ) + & 3 & 1 & 4679.323(6 ) & & & 3 & 2 & 2746.010(9 ) + & 4 & 1 & 6190.503(7 ) & & & 4 & 2 & 5400.804(8 ) + & 5 & 1 & 7677.273(4 ) & & & 5 & 2 & 6694.148(10 ) + & 6 & 1 & 9139.523(9 ) & & & 6 & 2 & 7964.836(3 ) + & 7 & 1 & 10577.184(13 ) & & & 7 & 2 & 9214.047(3 ) + & 8 & & & & & 8 & 3 & 10446.668(5 ) + & 9 & 1 & 13378.377(1 ) & & & + & 10 & 1 & 14741.688(1 ) & & & + & 11 & 1 & 16079.978(7 ) & & & + & 12 & 1 & 17393.116(8 ) & & & + & 13 & 1 & 18680.944(7 ) & & & + & 14 & 1 & 19943.297(7 ) & & & + & 15 & 1 & 21179.968(7 ) & & & + & 16 & 1 & 22390.754(7 ) & & & + lrrr [ tab : vib_triplet ] a @xmath166 & 1 & 1617.985(10 ) + & 2 & 3212.620(9 ) + & 3 & 4783.940(5 ) + & 4 & 6331.973(1 ) + & 5 & 7855.893(100 ) + & 7 & 10835.820(1 ) + & 8 & 12290.393(1 ) + & 9 & 13721.623(5 ) + & 10 & 15129.564(1 ) + & 11 & 16514.369(1 ) + b @xmath167 & 1 & 1448.103(8 ) + & 2 & 2874.028(2 ) + & 3 & 4277.927(1 ) + c @xmath168 & 1 & 2031.833(8 ) + & 2 & 4034.776(7 ) + & 3 & 6007.745(8 ) + & 5 & 9859.060(1 ) + & 6 & 11734.338(10 ) + & 7 & 13573.601(1 ) + d @xmath24 & 1 & 1753.500(7 ) + & 2 & 3469.636(10 ) + & 3 & 5145.247(6 ) + & 4 & 6776.153(1 ) + & 5 & 8356.139(7 ) + & 6 & 9880.362(7 ) + & 7 & 11337.658(1 ) + & 8 & 12722.024(1 ) + & 9 & 14025.567(1 ) + & 10 & 15245.388(2 ) + figure [ fig : marvelels ] shows all marvel term values below 35,000 @xmath4 for four singlet and four triplet states , where @xmath169 @xmath4 has been subtracted from the computed energies to make the figure clearer : this means that near - horizontal sequences of levels for a particular electronic state are all associated with something one could call a single vibrational level . figure [ fig : marvelels ] shows that the largest total angular momentum quantum numbers , @xmath170 , are 74 , 75 , 86 , and 70 for the , , , and states , respectively . as expected , as the vibrational excitation increases , the @xmath171 value usually decreases . finally , note that the coverage of rovibronic levels up to 35,000 @xmath4 from experiment is not complete ; assuming rigid rotation , data up to about @xmath172 is needed to have full coverage of the energy levels required during the thermochemical analysis . this coverage is provided in this study by duo energy levels ( _ vide infra _ ) . due to the strength of the cc bond in all the electronic states studied , the vibrational fundamental is substantial in almost all the bound electronic states of c@xmath1 . in fact , for the ground electronic state the harmonic wavenumber is close to 2,000 @xmath4 , a high value for a relatively heavy molecule . thus , the number of vibrational states is not that high , despite the large dissociation energy . in particular , @xcite computed 57 , 54 , 49 , and 36 bound vibrational levels for the , , , and b@xmath26 @xmath124 states , respectively . the number of vibrational levels characterized by our marvel analysis is considerably smaller , only 9 , 16 , 8 , and 3 , respectively . the vibrational energies presented for the triplet electronic states given in table 7 can not be compared easily with existing literature values , since the marvel values are for a specific spin component of a rovibronic energy level . the largest `` vibrational fundamental '' ( @xmath160 ) corresponding to the electronic states studied here is that of the c @xmath168 state , at 2031.833 @xmath4 . consequently , this state must have the strongest cc bond . the rotational constant of c@xmath1 for the state is relatively small , about 1.81 @xmath4 . this results in a large number of rotational states for each vibrational level . as part of this study , duo results were obtained up to @xmath172 . they served to check whether a marvel energy level is viable or not as well as they were generated to help the thermochemical analysis of this study ( _ vide infra _ ) . the lowest missing marvel energy level is at 7,800 @xmath4 ; up to this energy the coverage is complete . as figure [ fig : marvelrovib ] shows , as the energy increases there are more and more experimentally unknown energy levels . the coverage drops below 10% at about 30,000 @xmath4 . it is interesting to note how close some of the rovibrational energy differences are to each other . take the state as an example . the @xmath173 energy difference for @xmath160 is 1617.985 @xmath4 . the highly similar energy differences for the different spin components ( @xmath174 ) , ( @xmath175 ) , ( @xmath176 ) , ( @xmath177 ) , and ( @xmath178 ) , are 1618.063 , 1617.941 , 1617.985 , 1617.902 , and 1617.831 , respectively . the reason behind this observation is that each spin component of each vibrational level of the state has a slightly different effective value for their rotational constant due to both electronic and vibrational effects , and these energy differences include rotational energies . the first - principles duo energies , where such interactions are taken explicitly into account , results in similar energy differences . @xcite and @xcite seem to be the first to address the thermochemical properties of the c@xmath1 molecule . they both based their analysis on spectroscopic constants available to them and included several , but not all necessary , electronic states in their study . more recent thermochemical studies include those by @xcite , @xcite , and @xcite . the most reliable results appear to have been given by @xcite . the internal partition function , @xmath27 , of c@xmath1 is computed here via the direct summation recipe @xcite using a mixture of experimental ( marvel ) and theoretical ( duo ) energy levels . duo provides the full set of energies for the nine electronic states considered during the determination of the ideal - gas thermochemistry of c@xmath1 . while this is a complete set , the energy levels are of limited accuracy . thus , whenever possible , the duo energy levels are replaced by the incomplete but accurate set of marvel rovibronic energies . the final energies are used to compute the internal partition sum , its first ( @xmath179 ) and second ( @xmath180 ) moments , and the specific heat ( @xmath181 ) as a function of temperature . inaccuracies in @xmath27 have three sources of origin . the first is the intrinsic uncertainty of the energy levels . the second is the lack of a complete set of bound rovibronic energy levels . the third is associated with the treatment , including the possible neglect , of the unbound states . the main source of uncertainty in @xmath27 can be estimated straightforwardly using the uncertainty of each energy level and an error propagation formula . all experimental ( marvel ) energy levels have an associated uncertainty , while an uncertainty of 5.0 @xmath4 was assumed for the uncertainties of all the duo levels . ( note that this uncertainty estimate is rather pessimistic . ) figure [ fig : qerror ] ( solid line ) shows the impact of the uncertainties of the energy levels on the uncertainty of @xmath27 . it can be seen that up to about 2,500 k this type of uncertainty is dominant , but its maximum value , occurring at the lowest temperatures , is still less than 0.01% . estimation of the second type of error is hard since ( a ) the exact number of the energy levels is unknown , and ( b ) the value of the partition function grows monotonically as more and more energy levels are considered in the direct sum . therefore , only an approximate convergence can be reached at higher temperatures during the direct summation . to check the convergence of the partition function we need a larger set of energy levels ; therefore , we computed approximate rovibronic energy levels for all the electronic states considered , using the spectroscopic constants published by @xcite up to 70,000 @xmath4 . using the spectroscopic constants of 27 electronic states we could determine 332 347 extra energy levels above 35,000 @xmath4 . figure [ fig : qerror ] ( dashed line ) shows the difference ( in % ) of the two data sets , ( marvel + duo ) and ( marvel + duo + approximate energy levels ) . although up to 2,000 k the partition function is fully converged ( the difference is less than @xmath182 % ) , above 2,000 k the difference begins growing appreciably . nevertheless , the maximum uncertainty is still less than 0.1% at 4,000 k , which is acceptable for probably all practical applications . in case of c@xmath1 , no consideration of unbound states @xcite is necessary , due to the large dissociation energy of c@xmath1 . the @xmath183 , @xmath179 , @xmath180 , and @xmath181 results are given in table 10 in 100 k intervals . the full set of results at 1 k increments is given in the machine readable version of the table in the electronic edition . figure [ fig : qdiff ] shows the result of the comparison of our @xmath27 with those of @xcite and @xcite . it can be seen that the earlier studies always yield smaller numbers for the partition function of c@xmath1 than the present study . a possible explanation is that we use a larger ( more complete ) set of energy levels . note that the temperature range of the two earlier works begins from 1,000 k. figure [ fig : cpdiff ] shows the difference between our @xmath181 results and the janaf @xcite , @xcite , esa @xcite , and @xcite and @xcite data . @xcite reported @xmath181 values from 0 k to 6,000 k , but @xcite corrected his values above 2,000 k. while there is good agreement with the esa and altman & clementi data at low temperatures , the difference begins to increase with the temperature . conversely , for gurvich and janaf , the differences , especially compared to those of gurvich & veyts , are surprisingly large at lower temperatures . @xcite also found this discrepancy cpncerning the gurvich & veyts ( nasa ) data at lower temperatures ; nevertheless , they could not explain this strange behavior . we believe that the problem originates from the incorrect usage of @xmath184 values in @xcite . the standard ( spectroscopic ) energy expansions use @xmath158 as a minimum - to - minimum excitation energy ; in this case the rovibronic energy levels of the upper electronic state will be shifted by the difference of the zero - point energies , by about 0.5 @xmath185 . to get the correct energies , the @xmath159 values should be used instead of the @xmath158 values . the first excited electronic state usually lies much above the ground state ; therefore , this relatively small shift does not cause a significant problem if left out of consideration . however , in the case of c@xmath1 , where the relative energy of the state is smaller than the vibrational fundamental of either state , the incorrect use of @xmath158 leads to wrong @xmath181 values at lower temperatures . table 8 gives coefficients of the least squares fit to our computed partition function using the traditional form of @xcite @xmath186 in order to get the best reproduction of the directly computed values , the fit had to be performed in two separate temperature ranges . the first range is @xmath187 k , the other is @xmath188 k. these fits can reproduce the values of log@xmath183 reasonably accurately , within 0.1% in either region . nevertheless , to take full advantage of the high accuracy of the present thermochemical reults for c@xmath1 , the numerical results of the supplementary information should be used . this supplementary information also contains the transitions file which forms the input for marvel , and which can be augmented with any future spectroscopic data on c@xmath1 and rerun , and the associated energies file which is the output from the marvel run . c@xmath1 up to 4000 k. the solid ( black ) line shows the uncertainty which arises from the uncertainties of the known energy levels . the dashed ( red ) curve represents the convergence error , in % , due to energy levels not included in the analysis.,height=472 ] results with janaf @xcite ( solid ) , @xcite ( dashed ) , esa @xcite ( dotted ) , and @xcite & @xcite ( dashed - dotted , see text ) values . , height=472 ] crr [ tab : qcoeff ] @xmath189 & 3.6577362306 & 186.8558092069 + @xmath190 & @xmath57.0625294443 & @xmath5139.9034556834 + @xmath191 & 5.4057108902 & 42.4445304374 + @xmath192 & @xmath52.0744319852 & @xmath56.6283715970 + @xmath193 & 0.4447612738 & 0.5612382751 + @xmath194 & @xmath50.0504276952 & @xmath50.0241572167 + @xmath195 & 0.0023610646 & 0.00040521942 + rrrrlcl[tab : q ] 100.0 & 19.3786(1 ) & 19.3839(8 ) & 40.242(8 ) & 29.7333(2 ) + 200.0 & 41.6629(1 ) & 55.1971(7 ) & 170.867(15 ) & 40.2915(2 ) + 300.0 & 78.0529(1 ) & 137.7708(7 ) & 453.195(28 ) & 43.1578(2 ) + 400.0 & 133.6964(1 ) & 260.3942(7 ) & 812.605(39 ) & 39.7816(2 ) + 500.0 & 207.5463(1 ) & 411.2436(7 ) & 1225.949(48 ) & 37.2546(2 ) + 600.0 & 297.5930(1 ) & 585.6218(7 ) & 1700.047(59 ) & 36.0862(2 ) + 700.0 & 402.3796(1 ) & 782.8055(7 ) & 2244.006(70 ) & 35.6865(2 ) + 800.0 & 521.0433(1 ) & 1003.3874(7 ) & 2863.969(87 ) & 35.6538(2 ) + 900.0 & 653.1204(1 ) & 1248.2596(7 ) & 3564.19(15 ) & 35.7888(2 ) + 1000.0 & 798.3910(1 ) & 1518.3519(7 ) & 4348.50(38 ) & 36.0006(2 ) + 1100.0 & 956.7878(1 ) & 1814.6138(7 ) & 5221.04(92 ) & 36.2499(2 ) + 1200.0 & 1128.3464(1 ) & 2138.0495(7 ) & 6186.5(19 ) & 36.5204(3 ) + 1300.0 & 1313.1770(1 ) & 2489.7472(7 ) & 7250.5(37 ) & 36.8053(5 ) + 1400.0 & 1511.4500(2 ) & 2870.8933(7 ) & 8418.9(65 ) & 37.1015(7 ) + 1500.0 & 1723.3857(2 ) & 3282.7725(13 ) & 9698(10 ) & 37.4068(10 ) + 1600.0 & 1949.2475(3 ) & 3726.7596(19 ) & 11095(15 ) & 37.7195(14 ) + 1700.0 & 2189.3369(4 ) & 4204.3059(28 ) & 12616(22 ) & 38.0375(17 ) + 1800.0 & 2443.9883(6 ) & 4716.9251(38 ) & 14269(30 ) & 38.3585(21 ) + 1900.0 & 2713.5652(8 ) & 5266.1783(51 ) & 16060(41 ) & 38.6803(25 ) + 2000.0 & 2998.4562(11 ) & 5853.6614(65 ) & 17996(53 ) & 39.0006(29 ) + 2100.0 & 3299.0714(15 ) & 6480.9933(83 ) & 20084(66 ) & 39.3173(33 ) + 2200.0 & 3615.8393(19 ) & 7149.806(11 ) & 22332(81 ) & 39.6285(36 ) + 2300.0 & 3949.2042(24 ) & 7861.740(17 ) & 24744(98 ) & 39.9326(40 ) + 2400.0 & 4299.6233(32 ) & 8618.434(34 ) & 27329(116 ) & 40.2281(43 ) + 2500.0 & 4667.5647(49 ) & 9421.521(74 ) & 30092(135 ) & 40.5139(50 ) + 2600.0 & 5053.5054(88 ) & 10272.63(15 ) & 33039(156 ) & 40.7893(63 ) + 2700.0 & 5457.929(17 ) & 11173.38(32 ) & 36178(178 ) & 41.0535(92 ) + 2800.0 & 5881.327(33 ) & 12125.37(63 ) & 39513(201 ) & 41.306(14 ) + 2900.0 & 6324.193(64 ) & 13130.2(12 ) & 43051(225 ) & 41.546(23 ) + 3000.0 & 6787.02(11 ) & 14189.4(21 ) & 46798(251 ) & 41.775(36 ) + 3100.0 & 7270.32(21 ) & 15304.7(36 ) & 50760(280 ) & 41.991(56 ) + 3200.0 & 7774.59(36 ) & 16477.4(60 ) & 54941(316 ) & 42.195(83 ) + 3300.0 & 8300.33(61 ) & 17709.2(98 ) & 59348(363 ) & 42.38(12 ) + 3400.0 & 8848.04(98 ) & 19001(15 ) & 63986(429 ) & 42.56(17 ) + 3500.0 & 9418.2(15 ) & 20355(23 ) & 68859(528 ) & 42.73(23 ) + 3600.0 & 10011.4(23 ) & 21773(35 ) & 73971(673 ) & 42.89(32 ) + 3700.0 & 10628.0(35 ) & 23256(52 ) & 79328(882 ) & 43.03(42 ) + 3800.0 & 11268.6(52 ) & 24805(74 ) & 84933(1172 ) & 43.16(56 ) + 3900.0 & 11933.7(75 ) & 26421(105 ) & 90789(1562 ) & 43.28(71 ) + 4000.0 & 12624(11 ) & 28106(146 ) & 96899(2076 ) & 43.39(91 ) + this study utilizes the marvel technique to accurately determine close to 6,000 experimental rovibronic energies of @xmath0c@xmath1 for six singlet , six triplet , and two quintet electronic states , including the eight lowest valence states which gives coverage up to 35 000 @xmath4 . we survey all available laboratory high - resolution spectroscopic studies to provide input data for this process , resulting in 23,343 transitions connecting the 14 electronic states . while there are many spectroscopic studies available , in fact 42 were analyzed to yield the transitions analyzed , and there has been significant recent activity including the identification of several new band systems , there are also surprising gaps . for example , there is a detailed , fully rovibronically - resolved study of the deslandres dazambuja ( @xmath5 ) band system for @xmath125c@xmath1 @xcite , but even a century after the original observation of this band @xcite there is no available high - resolution study for @xmath0c@xmath1 . the recent observation of singlet triplet intercombination bands by @xcite helped us to achieve to link all rovibronic levels of @xmath0c@xmath1 into a single huge component within its experimental spectroscopic network ; thus , individual intercombination lines can now be predicted accurately using the results of our study . this is a significant step towrd the astronomical detection of these transitions @xcite . to further aid this work and other astronomical studies involving c@xmath1 , a full rovibronic line list for @xmath0c@xmath1 is currently being constructed using the variational code duo by @xcite , as part of the exomol project @xcite . as to now , the full set of marvel results comprising a file of validated transition frequencies and a file containing the resulting rovibronic energy levels are given in the supplementary information to this paper . the highly accurate but limited set of experimental ( marvel ) energy levels augmented with the much less accurate but much more complete set of duo energy levels has been used to compute ideal - gas thermochemical functions for @xmath0c@xmath1 up to 4,000 k. the accuracy of the partition function is better than 0.1% even at the highest temperatures , considerably exceeding the accuracy of all previous studies . this assures that the accuracy of the present isobaric heat capacity of @xmath0c@xmath1 is significantly better than that of any previous study . we thank robert curl , john hardwick , michael heaven , and jian tang for supplying the ( unpublished ) data from their spectroscopic experiments , and tatiana petrova and peter radi for comments on their data . peter radi is also thanked for his comments on the manuscript . this work has received support from the european research council under advanced investigator project 267219 , and the scientific research fund of hungary ( grant otka nk83583 ) . collaboration of the ucl and elte groups has greatly benefited from the support of two cost actions , codecs ( cm1002 ) and molim ( cm1405 ) . some funding was provided by the nasa laboratory astrophysics program . capitelli , m. , colonna , g. , giordano , d. , et al . 2005 , tables of internal partition functions and thermodynamic properties of high - temperature mars - atmosphere species from 50 k to 50000 k , tech . rep . , esa scientific technical review

we consider the tunneling current through a double point - contact fabry - prot interferometer such as used in recent experimental studies of the fractional quantum hall plateau at filling fraction . we compare the predictions of several different models of the state of the electrons at this plateau : the moore - read , anti - pfaffian , su naf , strong pairing , and states . all of these predict the existence of charge quasiparticles , but the first three are non - abelian while the last two are abelian . we give explicit formulas for the scaling of charge and charge quasiparticle contributions to the current as a function of temperature , gate voltage and distance between the two point contacts for all three models . based on these , we analyze several possible explanations of two phenomena reported for recent experiments by willett _ et al . _ , namely halving of the period of the observed resistance oscillations with rising temperature and alternation between the same two observed periods at low temperatures as the area of the interference loop is varied with a side gate . we conclude that the most likely explanation is that the observed alternation is due to switching between even and odd numbers of charge quasiparticles enclosed within the loop as a function of side gate voltage , which is a clear signature of the presence of non - abelian anyons . however , there are important features of the data which do not have a simple explanation within this picture . we suggest further experiments which could help rule out some possible scenarios . we make the corresponding predictions for future tunneling and interference experiments at the other observed second landau level fractional quantum hall states . `` with luck , we might see a non - abelian interferometer within a year . '' attributed to kirill shtengel , april 16 , 2008 in _ quantum computation : the dreamweaver s abacus _ .

0903.3108

the observation @xcite of a fractional quantum hall ( fqh ) state at @xmath0 and suggestion @xcite that the moore - read pfaffian ( mr ) state @xcite might occur at this filling fraction gave the first real indication that non - abelian topological phases of matter might actually occur in nature . the striking feature of such new phases is that they possess quasiparticle excitations with exotic non - abelian braiding statistics @xcite . this property makes non - abelian topological phases appealing for their potential use as intrinsically fault - tolerant media for quantum information processing @xcite . recent experimental studies of transport through a point contact in fqh systems at @xmath0 gave evidence that there are charge @xmath4 quasiparticles in this state @xcite and found that the dependence of the current on voltage and temperature is most consistent @xcite with two particular non - abelian models : the anti - pfaffian ( @xmath6 ) state @xcite and the su@xmath1 naf ( non - abelian fqh ) state @xcite . however , these results are not conclusive because the @xmath3 state @xcite , which is abelian , also supports charge @xmath4 quasiparticles . it is also roughly consistent with the voltage and temperature dependence of tunneling found in ref . and , in any case , one might expect non - universal physics to have a significant effect on the observed dependence . thus , there is a glaring need for experiments which directly probe the braiding statistics of quasiparticles . in order to probe braiding statistics in fqh systems , one can use a double point - contact interferometer , as proposed in ref . for abelian states and later considered for the @xmath0 state in refs . . such interferometers can play a crucial role in properly identifying which phase a fqh state is in by providing information about the topological @xmath7-matrix , a mathematical quantity related to the braiding statistics that is strongly characteristic of the topological order ( for more details , see ref . ) . interferometers are also important for the implementation of topological quantum computation @xcite because they can be used for the topological charge measurements necessary for readout of qubits @xcite and , through adroit manipulation , can even be used to implement computational gates @xcite . fortunately , there have been recent advances in realizing quantum hall interferometers at integer filling @xcite and fractional filling in the lowest landau level @xcite . even more recently , double point - contact interferometers have been experimentally implemented for the @xmath0 fqh state @xcite . in this paper , we study the signatures of non - abelian statistics which can be seen in a double point - contact interferometer and discuss other effects which can mimic these signatures . we propose further experiments which can help disentangle the effects of non - abelian statistics from coulomb blockade and disorder physics . the paper is structured as follows : in section [ sec : experiment ] , we describe the basic features of the experiment of willett _ et al . _ @xcite . in section [ sec : interpretations ] , we explain three different interpretations of this experiment : ( a ) non - abelian interferometry , ( b ) coulomb blockade , and ( c ) possible explanations loosely grouped together because they depend on non - linear dependence of the interferometer area on the side gate voltage @xmath8 . in section [ sec : critical ] , we criticize each of these three interpretations . we argue that , while all three interpretations have problems , the problems with ( b ) and ( c ) are more serious and these explanations are less likely to be correct . in section [ sec : proposals ] , we propose further experiments which might further strengthen or rule out interpretation ( a ) . in section [ sec : discussion ] , we comment on the implications of this experiment for topological quantum computation , assuming that the non - abelian interference interpretation is correct . in the two appendices , we give predictions for interference experiments at other suspected non - abelian fractions and we argue that the bare backscattering amplitude for @xmath4 quasiparticles should be much larger than that for @xmath5 quasiparticles . in recent experiments , willett _ et al . _ @xcite measured the current through a double point - contact device , depicted schematically in fig . [ fig : interf ] . as a function of magnetic field @xmath9 , the longitudinal resistance @xmath10 of the device has prominent minima at roughly the @xmath9 values at which the @xmath11 and @xmath12 quantum hall states occur in the bulk ( near , but not at , the point contacts ) . at the minima corresponding to @xmath0 and @xmath13 , the longitudinal resistance is @xmath14 , while at @xmath15 and @xmath16 it is @xmath17 . there are small oscillations with @xmath9 on top of these large features , but these were not the focus of the experiment since changing the magnetic field can change both the flux enclosed and , possibly , the quasiparticle number , thereby making it difficult to isolate the effect of braiding statistics . instead , a side gate voltage is varied , as shown in fig . [ fig : interf ] . as the side gate voltage @xmath8 is varied , @xmath10 oscillates with an amplitude of roughly @xmath18 . the period of the oscillations , @xmath19 , is larger at @xmath20 and @xmath13 than at @xmath15 . this was interpreted in the following way : it was assumed that the principle effect of varying the side gate voltage is to change the area of the interference loop between the two point contacts and that they are related linearly by @xmath21 , where @xmath22 is essentially constant , even between different filling fractions . thus , the oscillations are hypothesized to be due to the aharonov - bohm ( ab ) effect , which implies a period @xmath23 , where @xmath24 is the charge of the tunneling quasiparticle and @xmath25 is the electron charge , and @xmath26 is the magnetic flux quantum . willett _ et al . _ @xcite analyze their data to find that the period at @xmath20 and @xmath13 , normalized by the corresponding magnetic fields , is three times larger than at @xmath15 : @xmath27 . thus , they interpret their findings as evidence that @xmath28 at @xmath29 , assuming that the oscillation period at @xmath15 reflects interference of ordinary electrons . at @xmath0 , two types of behavior are seen at @xmath30 mk . in some regions , which we will call type i , @xmath31 . in the regions of type ii , @xmath32 . at @xmath33 mk , only one behavior is seen : @xmath32 . the type of oscillations observed for a region of @xmath8 were found to be reproducible throughout multiple scans over the period of @xmath34 days @xcite . the type i oscillations in a given region sometimes exhibited a roughly @xmath35 phase shift from one scan to another . in the next section , we discuss several possible explanations for the occurrence of these two periods at @xmath30 mk and the disappearance of one of them at higher temperatures at @xmath0 . and @xmath36 , respectively . the interferometry area is changed by applying a voltage @xmath37 to a plunger gate @xmath38 that depletes the 2deg beneath it . quantum interference between the two paths manifests an observable signature of the aharonov - bohm effect and the braiding statistics ( of the edge quasiparticle with the bulk quasiparticles in the central interferometry region ) in the oscillation patterns of the tunneling current when the area is changed . ] at first glance , these experimental results appear to be dramatically consistent with the predicted behavior of the proposed non - abelian @xmath39 fqh states , particularly with that of the mr , @xmath6 , and su@xmath1 naf states , all of which have a non - abelian fundamental quasihole with charge @xmath4 . the basic assumption is that as one changes the area of the interferometry region , one also occasionally changes the number @xmath40 of charge @xmath4 quasiholes contained in the bulk within the interference loop . ( for the purposes of this counting , charge @xmath41 excitations , where @xmath42 , count as @xmath43 fundamental quasiholes . ) thus , changing the area will cause the edge current to exhibit interference behavior due to the ab effect , modulated by occasional changes in the number of quasiparticles in the loop and their concomitant braiding statistics . the interference term @xmath44 of the backscattered current due to lowest order tunneling of @xmath4 edge quasiholes is predicted to be @xcite @xmath45 where the @xmath46 corresponds to the mr and su@xmath1 states and the @xmath47 to the @xmath6 state ; and @xmath48 or @xmath49 , depending on whether the contained quasiparticles are in a collective state corresponding to the @xmath50 or @xmath51 fusion channel . this interference exhibits the usual ab oscillations with period @xmath52 corresponding to @xmath53 , but also a striking complete suppression of this term that results from the non - abelian braiding statistics of the edge quasiparticle with the bulk quasiparticles when @xmath40 is odd . thus , as the area of the interferometry region is changed , and bulk quasiparticles enter or exit the interference loop , the non - abelian states should see @xmath52 oscillations switch on and off , as they do in going from the type i regions to the type ii regions in the experiments of refs . . the observed reproducibility of oscillation type regions in multiple scans @xcite suggests that the bulk @xmath4 quasiparticles are pinned and do not move on the time scale of the experiment . the observation of the oscillations in a given type i region being shifted by @xmath35 from one scan to the next also agrees with the expected behavior of non - abelian states . specifically , the collective state of several quasiparticles , some of which are inside and some outside the interferometry loop , is decohered by the current of edge quasiparticles around the loop @xcite . hence , depending on the bulk quasiparticles entering or exiting the interferometry loop , the collective state of quasiparticles inside the interferometer may be randomized between @xmath54 and @xmath49 when @xmath40 is changed to an even value . ( this is the same randomization that gives rise to a non - abelian signature in the switching noise @xcite . ) there are two sources that could potentially contribute to @xmath55 oscillations in the non - abelian @xmath39 states . the first is tunneling of the abelian @xmath5 edge quasiparticles @xcite , which to lowest order gives the interference current @xcite @xmath56 the second possibility comes from higher - order tunneling processes where the interference path encircles the interferometry area twice . the resulting double pass interference term in the current coming from @xmath57nd order tunneling of @xmath4 edge quasiparticles is @xcite @xmath58 where the @xmath47 corresponds to the mr state and the @xmath46 to the @xmath6 and su@xmath1 states . of course , this @xmath57nd order contribution to the tunneling current will typically have much smaller amplitude , since it both incurs an additional tunneling probability factor and doubles the distance over which coherence must be maintained . for the interferometer of refs . , the quasiparticle tunneling probability at each point contact is approximately @xmath59 . this estimate is based on the relation @xcite @xmath60 for point - contact tunneling of the half - filling edge modes at @xmath0 , where @xmath61 here is roughly the sum of individual tunneling probabilities of the two point - contacts , and @xmath62 @xmath63 in refs . . furthermore , there will generally be a suppression of the interference oscillation amplitudes that results from the loss of coherence , as well as from having unequal tunneling probabilities at the two point - contacts ( which gives a suppression of @xmath64 ) . roughly speaking , we can define the suppression factor to be @xmath65 as long as this is a small quantity . the observed oscillations in @xmath66 have amplitude of approximately @xmath57 @xmath63 , indicating a suppression factor of @xmath67 . higher order interference terms will be suppressed by higher powers of @xmath38 and @xmath68 , so combining these we find that the amplitude of double pass interference oscillations is expected to be roughly @xmath69 times that of the lowest order oscillation amplitude . hence , the @xmath55 oscillations , for which the amplitudes are of the same order of magnitude as that of the @xmath52 oscillations , should be attributed almost entirely to the tunneling of @xmath5 edge quasiparticles . we emphasize that the @xmath55 oscillations ( from both sources ) have an amplitude that is independent of @xmath40 ( unlike the @xmath52 oscillations ) , but pick up phase shifts when @xmath40 changes . these two sources of @xmath55 oscillations were not discussed in refs . because it was assumed neither would have significant contributions to the tunneling current . for the double pass interference of @xmath4 quasiparticles , this appears to be a valid assumption , since higher - order tunneling processes are suppressed in the weak - backscattering regime . on the other hand , for interference of @xmath5 quasiparticles this assumption was based on such quasiparticles having less relevant tunneling operators than the @xmath4 quasiparticles . we will see in the following that there are several ways in which this line of reasoning can break down and permit the @xmath5 quasiparticles to have a contribution to the tunneling current oscillations that is comparable to that of the @xmath4 quasiparticles . combining these results , we see that tunneling of both non - abelian @xmath4 quasiparticles and abelian @xmath5 quasiparticles at the point contacts of the interferometer would produce a combined backscattered current with regions of type i , exhibiting a sum of both @xmath52 and @xmath55 oscillations , when @xmath40 is even , and regions of type ii , exhibiting only @xmath55 oscillations , when @xmath40 is odd . we also note that the bulk - edge coupling that occurs as a bulk @xmath4 quasiparticle approaches the edge gives the regions near transitions between type i and ii oscillations the most potential for exhibiting non - linear and/or noisy behavior . the behavior of the interference current in the weak to strong coupling crossover as the quasiparticle approaches the edge was recently studied in refs . , which found results not entirely inconsistent with the experimental data . in order for interference to be observed , it is necessary that the current - carrying excitations remain phase coherent . even if we neglect ( irrelevant ) interactions between the edge modes , coupling to localized excitations in the bulk , and phonons , there will still be thermal smearing of the interference pattern . consequently , as shown in ref . ( see also ref . ) , the amplitude of interference oscillation for double point - contact interferometers will be exponentially suppressed in temperature and in the average length @xmath70 between point contacts along each edge @xmath71 where the coherence length @xmath72 and temperature @xmath73 of edge excitations are given by @xmath74 we can use these expressions , together with estimates of the charge and neutral edge mode velocities ( @xmath75 m / s and @xmath76 m / s ) from numerical studies @xcite of @xmath0 with pure coulomb interactions on a disk states . ] ( the charged and neutral scaling exponents @xmath77 and @xmath78 are given in table [ table : qps ] ) , to estimate coherence lengths and temperatures for the charge @xmath4 and @xmath5 excitations in the various candidate states ( the states are all the same , as far as the charge @xmath5 quasiparticle is concerned ) . in table [ table : coherences ] , we give estimates of coherence lengths at @xmath79 mk , and coherence temperatures for @xmath80 m . the temperature @xmath81 mk is the lowest temperature at which the experiments of refs . were carried out , and @xmath80 m is the approximate interference path length on each side of the interferometry area determined in refs . to be @xmath82 . we note that the observation of only type ii oscillations at higher temperatures in refs . also excludes double pass interference of @xmath4 quasiparticles as the explanation for @xmath55 oscillations , whereas it fits very nicely with the @xmath5 quasiparticle tunneling explanation . @xmath83 if the region between the two point contacts in the experiment of willett _ et al . _ @xcite was nearly an isolated puddle , the coulomb charging energy of the puddle would dominate the behavior of the device . this might occur if the gates pinched off the point contact too strongly . due to its isolation , the puddle must contain an integer number of electrons . the electron number can change when the gate voltage is increased by enough to allow one additional electron into the puddle . at this point , there is a peak in the longitudinal conductance ( which is also a peak in the longitudinal resistance , since @xmath84 ) since it is only at this point ( or within @xmath85 of it ) that the charge on the puddle can fluctuate . the maxima of the oscillations seen in willett _ et al . _ s experiment @xcite would be these peaks . if the density in the puddle is fixed , then the spacing between peaks as a function of area is naively just the additional area required to allow one more electron into the puddle : @xmath86 where @xmath87 is the charge density inside the dot . however , in the case of a paired state , one would expect that it is easier to add an electron when the electron number is odd than when it is even since , in the latter case , an unpaired fermionic excitation is necessarily created . so one would expect that , instead of evenly - spaced peaks , the interval between an odd peak and the next even peak would be smaller than the interval between an even peak and the next odd peak because @xmath8 must also supply the energy needed to create an unpaired fermionic excitation . consequently , the peak spacing would alternate between @xcite @xmath88 as a result of this `` bunching '' effect , the periodicity would be the interval between two successive even peaks , i.e. twice what one might ordinarily expect . but when there is an odd number of charge @xmath4 quasiparticles in the mr , @xmath6 , or su@xmath1 states , the minimum energy to create a fermionic excitation is zero . thus , there is no bunching effect in this case , and the period is not doubled @xcite . in the case of the @xmath3 state , bunching generically occurs with either an even or odd number of quasiparticles in the puddle . however , when @xmath40 is odd , the bunching depends on the strength of the violation of @xmath89 conservation ( where @xmath89 is the @xmath90-component of the spin or , if one contemplates a bi - layer version of this experiment , the layer pseudospin ) so @xmath91 in eq . ( [ eq : cb_even ] ) is replaced by a different constant dependent on this violation . this is because the neutral sector of the edge theory of the @xmath3 state is a pair of majorana fermions or , equivalently , a dirac fermion ( which can be bosonized , leading to the standard @xmath92-matrix description ) . if @xmath89 is conserved ( or only weakly non - conserved ) , both majorana fermions will have edge zero modes for odd @xmath40 and the bunching will disappear . in this case , the bunching pattern would look just like that predicted for the mr , @xmath6 , and su@xmath1 states . however , if @xmath89 is more than weakly non - conserved , then the two majorana fermions zero modes will mix and split , leaving no zero modes . consequently , when @xmath89 is not conserved , there will be bunching even when there is an odd number of quasiparticles . thus , switching between bunching and non - bunching regions in coulomb blockade at @xmath93 is not necessarily an indication of a non - abelian state . furthermore , if @xmath94 , bunching will never be seen in any state . more generally , the switching between different bunching patterns in coulomb blockade described in refs . for non - abelian states may similarly be mimicked by corresponding abelian states ( for more details , see ref . where explicit examples are given for all the most physically relevant non - abelian states ) . the strongly - paired @xmath2 state @xcite always exhibits bunching , now with @xmath91 in eq . ( [ eq : cb_even ] ) replaced by a constant dependent upon the finite energy cost of having an unpaired electron . if this energy cost is small , it may not appear bunched . on the other hand , if it is large enough , it will be maximally bunched with @xmath95 corresponding to tunneling electron pairs . the assumptions that @xmath96 with only a single value of @xmath22 across a range of filling factors and a range of @xmath8 values are important for the two previous interpretations of this experiment . it is not clear that @xmath97 should be constant across an appreciable range of @xmath8 values because the density is not constant across the device . in fact , we expect @xmath8 to vary linearly with total charge in the central puddle . so long as the electron density is essentially fixed , apart from a small number of quasiparticles , @xmath8 will vary linearly with @xmath98 . however , if there are high - density and low - density regions , then we will have @xmath96 in some regions and @xmath99 in the others , with @xmath100 . ( this could lead , for instance , to @xmath101 puddles within the @xmath0 droplet . ) this would , in turn , lead to two different regions with different oscillation periods . however , it is difficult to see why one period would be twice the other or why there would be two periods only at @xmath0 and not at @xmath102 . one other possibility , which also depends on spatial inhomogeneity although still assuming a linear @xmath98 vs @xmath8 , is that there are regions in the sample in which the @xmath2 abelian state occurs . the rest of the state is assumed to be non - abelian , i.e. either the mr or @xmath6 state . then , when a @xmath2 region is at the edge of the system , varying @xmath8 does not change the area enclosed by the edge of the non - abelian part of the system , which would lead to @xmath4 oscillations . it does cause the total area to vary , but this only causes @xmath5 oscillations since these oscillations can move coherently along both @xmath2 abelian and non - abelian edges . thus , the two regions correspond to when the edge of the system near the side gate is a @xmath2 region or a non - abelian region . the data of refs . are broadly consistent with the hypothesis that the device is functioning as a quantum hall edge state interferometer . as the temperature is raised , the putative @xmath4 oscillations , which are observed at 30 mk , disappear while the @xmath5 oscillations persist even at 150 mk . this was anticipated in ref . , where it was noted that the coherence length will be substantially longer for @xmath5 quasiparticles than for @xmath4 quasiparticles since the former do not involve the slow neutral edge modes . thus , any of the proposed @xmath0 states ( apart from the strong - pairing state ) would be broadly consistent with the @xmath4 oscillations seen in refs . . however , there is no simple explanation of their absence in the type ii regions in the @xmath3 state , while the mr , @xmath6 , and su(2)@xmath103 states all provide a simple explanation , as described in the previous section . one however needs a more careful and detailed study of the temperature and voltage behavior before a favored candidate non - abelian state can be identified . perhaps the most serious challenge to the non - abelian interferometer hypothesis is that the same amplitude of @xmath5 oscillations should always be present ( i.e. in type i and ii regions ) , while @xmath4 oscillations should only be observed when the quasiparticle number contained within the interferometry region is even . this appears to be the case in figs . 3 and s2b of ref . and fig . 2a of ref . , and to a lesser extent in figs . 4b and s2a of ref . and fig . 2c of ref . . however , this appears not to be the case in fig . 4a of ref . and fig . 2b of ref . it is possible to generate some accidental destructive interference between the oscillations due to tunneling of @xmath5 quasiparticles given in eq . ( [ eq : i12e2 ] ) and that of double pass interference of @xmath4 quasiparticles given in eq . ( [ eq : i1212e4 ] ) , since the relative phase of these terms is not fixed . this could potentially result in the appearance and disappearance of @xmath5 oscillations , however , as previously mentioned , the magnitude of oscillations in eq . ( [ eq : i12e2 ] ) is so strongly suppressed in the experiments of refs . that it could not explain such behavior there . as we describe in appendix [ sec : matrix - elements ] , a simple model of quasiparticle tunneling predicts that the amplitude for @xmath4 quasiparticle backscattering , @xmath104 , is much larger than the amplitude for @xmath5 quasiparticle backscattering , @xmath105 . however , the magnitude of @xmath5 oscillations in the type ii regions is comparable to the magnitude of @xmath4 oscillations in the type i regions . it may be that @xmath106 is `` accidentally '' large , e.g. due to the presence of a tunneling resonance for @xmath5 quasiparticles at the point contact . ( for example , such a resonance might occur if there is an impurity or a region of different filling in the point contact acting like a dot or anti - dot through which resonant tunneling favors the @xmath5 charge . ) alternatively , as a result of the shorter coherence length for @xmath4 excitations , the corresponding oscillations are more strongly suppressed . this would require a coincidence that thermal smearing of @xmath4 excitations compensates for the smallness of the ratio @xmath107 . however , this could be tested by going to lower temperatures to reduce the suppression , and by increasing the separation between the point contacts to increase the suppression . at any rate , given that @xmath5 oscillations are observed in the type ii regions , it would be a problem for the non - abelian interferometer picture if they are not generically seen in the type i regions . however , it is worth noting in this context that the presence of charge @xmath5 quasiparticle tunneling is not manifest in the point contact experiments of refs . . in the former , the shot noise appears to indicate that only charge @xmath4 quasiparticles tunnel at the point contact ( although there is sufficient scatter in the data that one might argue that there could be a component due to @xmath5 quasiparticle , the scatter does not seem to be asymmetric in the direction of charges larger than @xmath4 as one might have expected ) . in the latter experiment , the best fit to the data is actually @xmath108 , so including any @xmath5 tunneling leads to a worse fit to the data @xcite . thus , the appearance and strength of @xmath5 quasiparticle tunneling remains a mystery in several different experiments . a conventional coulomb blockade picture seems inappropriate since @xmath109 indicates that the system is in the weak back - scattering limit . it is also unlikely that coulomb blockade could lead to two distinguishable periods since , for @xmath91 small ( as we expect it to be ) , the bunching will be difficult to resolve . numerical calculations of the edge velocities @xcite give @xmath110 , confirming this expectation . on the other hand , we note that coulomb blockade is capable of producing peaks that alternate between the @xmath4 and @xmath5 periodicities , with no @xmath5 background in the @xmath4 region . thus , if the two prior points against it were somehow incorrect , coulomb blockade could provide a consistent explanation of the periodicity issue . furthermore , coulomb blockade could be easily ruled out by measuring its temperature dependence and its dependence on asymmetry between the tunneling amplitudes at the two point contacts . in particular , the coulomb blockade peak widths are expected to scale linearly with temperature @xcite . however , a more general view of coulomb blockade has emerged @xcite ( see , also refs . ) , according to which coulomb blockade ( cb ) can be distinguished from ab interference by _ inter alia _ the dependence of @xmath111 on @xmath9 ( it should be inversely proportional for ab and independent for cb ) . this more general view of coulomb blockade is probably better described as `` coulomb dominated '' since it corresponds to a regime in which the charging energy of the puddle between the point contacts is the dominant energy scale . it does _ not _ rule out a simple interpretation of the backscattered current according to eq . ( [ eq : na_int_5_2 ] ) . at any rate , by this criterion as well , the data appears to be more consistent with ab interference since @xmath112 . however , it is worth keeping in mind that we do not know precisely how the area of the droplet changes with @xmath8 or with @xmath9 ( the @xmath9 dependence further distinguishes aharonov - bohm from coulomb blockade because @xmath98 should be independent of @xmath9 in the former case , but not in the latter case ) ; knowing this would enable us to cement an interpretation of the experiment . as previously mentioned , one might question the validity of the assumption that @xmath113 holds with the same value of @xmath22 across a range of filling fractions . however , the assumption that @xmath22 is independent of the filling fraction for nearby filling fractions is , in fact , reasonable . @xmath8 is several volts , and the oscillation periods are @xmath114 mv which are much higher energy scales than the weak energy gaps and correlation effects associated with the @xmath115 quantum hall states . thus , the details of these quantum hall states are probably unimportant and @xmath97 is probably determined by the electric potential due to the donor impurities and the electron density , which are not varying significantly . however , when there are filled landau levels beneath the quantum hall state of interest , their edges can screen the side - gate voltage , presumably weakening the dependence of @xmath98 on @xmath8 ( since @xmath98 is the area of the droplet of the fractional state in the partially - filled landau level ) . in particular , we would expect @xmath116 at @xmath117 but @xmath118 at @xmath101 , with @xmath119 . however , by the same reasoning , we expect that the relationship between @xmath98 and @xmath8 will be the same for @xmath115 ( if the @xmath20 edge is two filled landau levels with a backward propagating @xmath117 edge mode ) . as beautiful as the non - abelian anyon explanation of the results of ref . may be , it is clear from the preceding analysis that there are some significant gaps which need to be closed through further measurements . if it is , indeed , the case that @xmath10 is due to the weak backscattering of @xmath4 quasiparticles at the constrictions , then both the non - oscillatory and oscillatory parts of the current should have non - trivial temperature and voltage dependence . modeling the edge in the simplest way ( i.e. fully equilibrated neutral modes and no edge reconstruction ) using the `` natural '' conformal field theory inherited from the bulk , one can perform a more detailed analysis of the tunneling edge current @xcite , along the lines of that carried out in refs . for abelian states . the non - oscillatory part of the backscattered current the sum of the contributions from each point contact independently will behave as the power laws : @xmath120 where @xmath121 is the tunneling exponent combining charge and neutral ( abelian and non - abelian ) sectors of the quasiparticles tunneling operator . the tunneling operator is relevant for @xmath122 , and quasiparticles with smaller @xmath123 are more relevant , and are thus expected to dominate the tunneling current in the weak backscattering limit . @xmath124 from table [ table : qps ] , we see that the @xmath4 backscattering operator is a relevant perturbation of the edge effective theory for all of the candidate states . thus , the effective tunneling amplitude(s ) will decrease as the temperature is raised , as @xmath125 , @xmath126 , or @xmath127 in the mr , @xmath3 , or @xmath6 and su@xmath1 states , respectively . charge @xmath5 backscattering is also relevant in all of the candidate states . because @xmath5 excitations have @xmath128 and are entirely in the charge sector , their lowest order , single point - contact tunneling current contribution is the same in all of the candidate states . using the methods of ref . to explicitly compute its value , we find @xmath129 thus , in the linear response regime , the effective tunneling amplitude for @xmath5 backscattering decreases as @xmath127 . for the mr and abelian states , charge @xmath4 backscattering is more relevant than charge @xmath5 backscattering , so it is expected to dominate at lower temperatures . for the @xmath6 and su@xmath1 states , @xmath4 and @xmath5 backscattering are equally relevant ( with @xmath128 ) . as previously mentioned in section [ sec : introduction ] , the single point contact experiment of ref . found voltage and temperature dependence of the tunneling current to be most consistent with the @xmath6 and su@xmath1 states , providing a point of reference with which to compare . turning now to the oscillatory current , we note that for @xmath130 , where @xmath131 is the interference path length , it has the same voltage dependence as the non - oscillatory current . for larger voltages , it becomes apparent that there are oscillations with a period @xmath132 , but these are much larger voltages than are probed in the experiments of refs . we note that these oscillations can be turned around and interpreted as oscillations as a function of @xmath70 , which changes when the interferometry area is changed ( i.e. when the side gate voltage is changed ) . however , these oscillations have periodicity @xmath133 , where @xmath134 represents several characteristic velocities , which are all dominated by the slowest edge mode velocity ( which is expected to be @xmath135 ) . since @xmath136 v , these will only give rise to envelopes with periods much longer than that of the oscillations observed in the experiment . the temperature dependence of the oscillatory current includes a power - law prefactor of the form in eq . ( [ eqn : power - laws ] ) in addition to the exponential suppression @xmath137 which we discussed earlier . thus , the relative suppression of the @xmath4 contribution , compared to the @xmath5 contribution , must be due entirely to the shorter coherence length in the @xmath6 case but could be due to a combination of effects in the mr or @xmath3 case . to make the case for interference stronger , it would be helpful to disentangle the effects of the temperature dependence of the coherence length from the temperature dependence of the effective tunneling amplitude . one way to do this would be to carefully study the bias voltage dependence of the current backscattered by the interferometer of refs . at some fixed @xmath8 in the low-@xmath138 limit . if the behavior is similar to that observed in ref . ( and , especially , if it is the behavior expected for one of the possible @xmath0 states ) , then this is a strong indication that @xmath10 is due to the weak backscattering of charge @xmath4 quasiparticles . another useful way to do this would be to turn on the point - contacts one at a time and study their tunneling behavior individually . this would help determine which state occurs in the point - contact region ; it is important that it is at the same filling fraction as the rest of the bulk . furthermore , it would allow one to determine the relative tunneling amplitudes of @xmath4 and @xmath5 quasiparticles and confirm that the experiment is not in the cb regime . it is important to verify that the oscillation periodicities behave as expected . in addition to confirming the dominance of @xmath139 and @xmath140 oscillations in the fourier spectrum , one should examine the spectrum in the different regions more carefully . specifically , by using windowing techniques in the fourier analysis of the data , one should check that the type i regions have both @xmath139 and @xmath140 oscillations , that the type ii regions have only @xmath140 oscillations , and that the amplitude of @xmath140 oscillations are roughly the same in the type i and ii regions . it is also useful to know the relative oscillations amplitudes of the two frequencies in the type i regions . another aspect of the data worth examining more closely is the phase shifts observed in the oscillations of a given type . while an @xmath4 bulk quasiparticle entering or exiting the interference loop causes a switch between type i and ii oscillations , there could also be @xmath5 quasiparticles in the bulk that enter or exit the interference loop when the area is changed ( or pairs of @xmath4 quasiparticles that enter or exit nearly simultaneously ) , which would not switch the oscillations between type i and ii . from eq . ( [ eq : i12e2 ] ) we see that this would cause a shift of phase @xmath35 in the @xmath5 oscillations , while from eq . ( [ eq : na_int_5_2 ] ) we see that this would cause a shift of phase @xmath141 in the @xmath4 oscillations . to more firmly establish the origin of these phase shifts , as well as how reproducible are the fine details of the observed oscillations , one should immediately backtrack on the side gate voltage within an oscillation region of fixed type to see whether the oscillations ( including locations of phase shifts ) are nearly exactly reproduced . furthermore , we recall that if one is in a type i region and an @xmath4 bulk quasiparticle exits and then re - enters the interference loop , the value of @xmath142 in eq . ( [ eq : na_int_5_2 ] ) will be randomized with equal probability . hence , one should also increase the side gate voltage ( decreasing the interferometry area ) until one transitions from a type i region into a type ii region , and then soon afterward backtrack on the side gate voltage to return to the type i region . when this is done repeatedly , if the @xmath4 oscillations in the type i region are found to be shifted by a phase of @xmath35 half the time , it would be a direct observation of non - abelian statistics . doing this backtracking immediately would provide a more convincing observation of the @xmath35 phase shift due to the non - abelian statistics than backtracking after a complete run , which leaves plenty of time ( one day ) for small changes to occur that may give non - universal phase shifts . furthermore , it is more convenient to perform a statistical experiment for which each sampling takes one hour rather than one day to complete . it is important to establish that the periodicities with @xmath8 obtained in the experiment correspond directly to periodicities with @xmath98 . one way to attempt to do this would be to vary both @xmath8 and @xmath9 at @xmath15 and to use the periodicity in @xmath9 to determine the area for several different values of @xmath8 . one could , in this way , check that the assumed constant @xmath22 in @xmath143 is really constant . such a measurement would also determine whether the oscillation pattern corresponds to ab interference or coulomb blockade , as in ref . . one could also check that the oscillations are due to ab interference by turning down or off one of the point contacts ( and then the other ) and repeating the experiment , which should cause the oscillations to disappear . this will further exclude resonances at a single point contact as the source of oscillations and will give a better value of the tunneling amplitude for a single point contact . a more ambitious approach to measuring the area within the interference loop , which could simultaneously tackle the even more fundamental problem of determining directly whether the type i and ii regions correspond to even / odd quasiparticle numbers , would be to image the two dimensional electron gas ( 2deg ) in the interferometer using a scanning single electron transistor ( set ) , as in ref . . in this earlier experiment , a scanning set was used to image the compressibility of the electron liquid at @xmath144 . by measuring the compressibility , it should be possible to determine where the edge of the hall fluid is in the droplet and , hence , the area of the interferometry region . it should also be possible to find the localized states near the fermi energy where @xmath4 quasiparticles could be trapped . by imaging the charge @xmath4 quasiparticles , one might even be able to see these localized quasiparticles enter or leave the interference loop as @xmath8 is varied , and hence allow the most direct verification of the non - abelian interferometer interpretation . similarly , the possibility that the existence of two periodicities signals different regions with different electron densities and , therefore , two different possible relations @xmath143 and @xmath145 could be ruled in or out through a more detailed knowledge of the electron density in the sample . we note , as a consistency check on the data of ref . that one can use the area periodicity of the ab oscillations to estimate the density of bulk @xmath4 quasiparticles by attributing each observed switching between type i and ii regions to an @xmath4 quasiparticle entering or leaving the interferometry region . in this way , we estimate the density to be @xmath146 . this translates to @xmath114 charge @xmath4 quasiparticles in the interference loop . we can also estimate the number of @xmath5 quasiparticles in the bulk , though perhaps less reliably , by attributing phase disruptions observed within an oscillation type region to an @xmath5 quasiparticle crossing the interferometry region . using this to similarly estimate the density of bulk @xmath5 quasiparticles gives @xmath147 , or roughly the same number as charge @xmath4 quasiparticles in the interference loop . depending on how seriously one takes the @xmath5 contribution , this gives approximately @xmath148 depletion of the electron density in the bulk , which is the reported density variation in the device @xcite . with a scanning set setup , it may be possible to find the @xmath114 charge @xmath4 quasiparticles that are necessary for the non - abelian interferometer interpretation and observe them entering or exiting the interference loop . localized @xmath5 quasiparticles or , equivalently , closely - spaced pairs of @xmath4 quasiparticles should also be observable . a more crude , but also more easily implementable way to further strengthen the correlation between which oscillation type is observed and the localization of excitations in the bulk is to independently vary two or more plunger gates of the interferometer . by refining the ability to control how the interferometry area is changed beyond a single plunger variable , the changes between oscillation types can be more strongly associated with a particular area . if a region of one plunger s gate voltage exhibits type i oscillations , but then , after changing a separate plunger s position , the same voltage range in the first plunger exhibits type ii oscillations , this would demonstrate that a particular oscillation type is not associated with that particular voltage range of the first plunger , but rather that an ability to change between oscillation types is associated with a localized quantity in the area added or removed by the second plunger . this would greatly strengthen the evidence for non - abelian braiding statistics . a double point - contact interferometer may also be used to test whether the quantum hall states at @xmath101 , @xmath149 , @xmath150 , and @xmath151 are non - abelian . these filling fractions all have compelling abelian alternatives which almost certainly occur at their corresponding lowest landau level counterparts . while numerical studies strongly support the mr and @xmath6 states at @xmath0 @xcite and the ( particle - hole conjugate ) laughlin state at @xmath152 @xcite , they are far less conclusive for @xmath101 , @xmath149 , and @xmath150 @xcite , where several candidates seem plausible , including ones that are non - abelian . it is clearly important to also test these fqh states experimentally . the details of the plausible candidates experimental signatures are discussed in appendix [ sec : other - candidates ] . the signatures of non - abelian statistics in these states will again be dramatic , though not quite as much as for the mr , @xmath6 , or su@xmath1 naf states . we close this discussion by assuming , for a moment , that the experiments of refs . are , in fact , performing interferometry on the @xmath0 state of the sort envisioned in refs . and are detecting non - abelian quasiparticles . what forecast would these results give for topological quantum computation @xcite ? certainly , it would be encouraging that a non - abelian topological state , the _ sine qua non _ for topological quantum computing , would be found . one potential source of concern is the appearance of some seemingly unpredictable phase disruptions , which would make it difficult to distinguish the two states of a topological qubit , which differ by a @xmath35 phase shift in their ( @xmath4 oscillation ) interference patterns . however , if further investigation shows that they are @xmath141 phase shifts ( in the type i regions ) , then they may be attributable to @xmath5 quasiparticles or pairs of @xmath4 quasiparticles simultaneously entering or exiting the interference loop , and it would be a manageable problem . on the other hand , if they turn out to be @xmath35 phase shifts in the @xmath4 oscillations , then they could be attributed to either tunneling of a @xmath4 quasiparticle from one edge to the other between bulk quasiparticles within the interferometry region , or tunneling of an electrically neutral fermion between the edge and a bulk quasiparticle within the interferometry region . either of these @xmath35 phase shifting processes would cause errors in topological qubits , so if we attribute all the observed phase disruptions to such processes , this gives a crude estimate of about an hour for the time scale for such errors , or a contribution to the topological qubit error rate of @xmath153 . to better determine the rate of such errors , one could simply tune the voltage to a local @xmath4 oscillation maximum / minimum of the tunneling current in a type i region and time average duration it takes for the current to jump to a lower / higher value ( i.e. experiencing a @xmath35 phase shift to a minimum / maximum of the current ) . if the phase disruptions are neither @xmath141 nor @xmath35 phase shifts , then they would be a serious concern , as they would have no obvious explanation . finally , the apparent stability of the type i and type ii regions implies that thermally - activated charge @xmath4 quasiparticles do not move in and out of the interferometry region over the time scales of this experiment . indeed , these regions are stable on a time scale of a week , which would imply a topological qubit error rate from mobile bulk quasiparticles of @xmath154 . `` with luck , we might see a topological qubit within a year . '' attributed to kirill shtengel , january 9 , 2009 @xcite . we would like to thank r. willett for many important discussions and access to unpublished work , m. heiblum for a discussion of unpublished work , and s. das sarma , m. freedman , and y. gefen for helpful comments . wb , ks , and jks would like to acknowledge the support and hospitality of microsoft station q. pb , cn , and ks would like to acknowledge the hospitality of the aspen center for physics . cn and ks are supported in part by the darpa - quest program . ks is supported in part by the nsf under grants dmr-0748925 and phy05 - 51164 , and would like to acknowledge the hospitality of the kitp . jks is supported in part by the science foundation of ireland principal investigator grant 08/in.1/i1961 . in this section , we consider the other observed fqh states in the second landau level @xcite , i.e. @xmath155 , @xmath149 , @xmath150 , and @xmath151 . we provide the data of their prominent candidate descriptions that will be useful for interpreting tunneling and interference experiments , similar to earlier in this paper . as described earlier , the interference term of the tunneling current combines : ( 1 ) the ab effect , ( 2 ) the braiding statistics with encircled quasiparticles , and ( 3 ) the edge physics . the ab effect simply contributes a phase @xmath156 when the quasiparticle of charge @xmath157 encircles flux @xmath158 . the physics describing propagation of excitations on the edge gives rise to a temperature , bias voltage , and interference path length dependence of the tunneling edge current @xcite that we denote as @xmath159 . the details of this edge physics can generally be complicated , but the most significant aspect is the coherence length and temperature , which is given as before in eqs . ( [ eq : coh_length],[eq : coh_temp ] ) , with the appropriate scaling exponents for excitations of a given state ( which are given in the tables ) . for the lowest order tunneling interference process , the braiding statistics contributes the factor @xcite @xmath160 where @xmath161 is the topological @xmath7-matrix , and @xmath162 and @xmath163 are the topological charges of the tunneling edge excitation and the encircled bulk quasiparticle excitations , respectively . these combine to give the lowest order interference contribution to the tunneling current ( in the asymptotic limit where the state of the bulk quasiparticles is projected onto a definite value of @xmath163 ) @xmath164 if either @xmath162 or @xmath163 is an abelian charge , @xmath165 is simply a phase . more generally , when @xmath162 and @xmath163 are both non - abelian charges , @xmath165 is a complex number with @xmath166 . this leads to the potential for a suppression of the interference term [ eq . ( [ eq : na_int_gen ] ) ] resulting from non - abelian braiding statistics , similar to the non - abelian @xmath0 states . for the non - abelian fqh states considered here , the braiding statistics are essentially given by the su@xmath167 theories @xcite , up to abelian phase factors . these theories have topological charges @xmath168 and @xmath169 the @xmath170 , @xmath171 , and @xmath172 cases are the most pertinent to our discussion , so we write them out explicitly : @xmath173 , \ ] ] @xmath174 , \ ] ] where @xmath175 is the golden ratio , and @xmath176 .\ ] ] the braiding statistics of the mr , @xmath6 , and su@xmath1 naf @xmath39 states are all derived from su@xmath1 . the non - abelian quasiparticles in these states carry su@xmath1 charge @xmath177 . it follows that an odd number cluster of such quasiparticles will also carry a collective su@xmath1 charge of @xmath177 , while an even number cluster will carry either @xmath178 or @xmath49 . thus , looking at the ( @xmath179 ) middle column of eq . ( [ eq : m2 ] ) , we see exactly the source of the behavior described in eq . ( [ eq : na_int_5_2 ] ) . for the @xmath101 fqh plateau , the leading candidates are the laughlin ( l ) state @xcite , two types of bonderson - slingerland ( bs ) states @xcite , and a @xmath172-clustered read - rezayi ( rr ) state @xcite . ( the bar indicates particle - hole conjugation . ) the bs states considered here are hierarchically constructed over the mr and @xmath6 states , and so have similar non - abelian statistics derived from eq . ( [ eq : m2 ] ) using the fact that the non - abelian quasiparticles carry su@xmath1 charge @xmath177 . the rr@xmath180 state is related to su@xmath181 , and so has more complicated non - abelian statistics , derived from eq . ( [ eq : m4 ] ) . its fundamental @xmath182 quasiparticles carry su@xmath181 charge @xmath177 . @xmath183 we see in table [ table_7_3:qps ] that all of these states have an @xmath184 excitation with smallest scaling exponent @xmath185 , and so one expects these to dominate the tunneling . the @xmath186 state also has @xmath182 excitations with @xmath185 , which should give a comparable contribution to the tunneling current . the experiments of refs . , which appear to observe @xmath184 tunneling , but not @xmath182 tunneling at @xmath101 , seem to exclude the @xmath186 state , while they agree with the l@xmath187 , @xmath188 , and bs@xmath189 states . in fact , since the relevant excitations of these latter three states all have @xmath184 , and furthermore , the most relevant tunnelers are all abelian , it will likely be difficult to distinguish between l@xmath187 , @xmath188 , and bs@xmath189 using tunneling and interferometry experiments . thermal transport experiments are probably the best hope of distinguishing between these . for the @xmath191 fqh plateau , the leading candidates are the haldane - halperin ( hh ) state @xcite , two types of bs states @xcite , and a @xmath171-clustered rr state @xcite . these bs states again have non - abelian statistics derived from eq . ( [ eq : m2 ] ) using the fact that the non - abelian quasiparticles carry su@xmath1 charge @xmath177 . the rr@xmath192 state is related to su@xmath193 , and so has non - abelian statistics derived from eq . ( [ eq : m3 ] ) . its fundamental @xmath194 quasiparticles carry su@xmath193 charge @xmath177 . @xmath195 we see in table [ table_12_5:qps ] that all of these states have an abelian @xmath196 excitation with scaling exponent @xmath197 , so there should always be a background of such excitations in tunneling . the hh@xmath198 , bs@xmath198 , and @xmath199 states all have an abelian @xmath194 excitation with @xmath200 , so there should be a weaker background of these excitations in the tunneling . the smallest scaling exponent for the bs@xmath198 state belongs to the non - abelian @xmath194 excitation , which is therefore expected to dominate the tunneling in this state . the @xmath199 state has a non - abelian @xmath194 excitation which has slightly less relevant tunneling operator than the @xmath196 excitation . the @xmath201 state has a non - abelian @xmath194 excitation with the same scaling exponent @xmath197 as the @xmath196 excitation , so they should have roughly equal contribution to tunneling . in interferometry experiments , the bs@xmath198 , @xmath199 , and @xmath201 states will all exhibit @xmath194 oscillations that will sometimes be suppressed . however , there are important distinctions within this behavior that can distinguish between them . in particular , the bs states will exhibit an even - odd effect similar to eq . ( [ eq : na_int_5_2 ] ) , always returning to suppression for @xmath202 odd , where @xmath202 is the number of non - abelian @xmath194 fundamental quasiparticles . on the other hand , the rr state can exhibit both suppression and full amplitude oscillations for all values of @xmath40 , and it has a probability of switching between them when a given quasiparticle is taken in and out of the interferometry region . furthermore , when the oscillations are suppressed for the bs state , the smaller amplitude @xmath194 oscillations will be due to tunneling of the abelian @xmath194 excitations ( which will always be present ) , because the non - abelian excitation will have fully suppressed interference . the relative contribution to the tunneling of these excitations is not a fixed amount , and will change depending on temperature and voltage ( i.e. they have different scaling ) . in contrast to this , the suppression that would be observed in the rr state is due entirely to the braiding statistics of the non - abelian @xmath194 excitation , and the suppressed oscillation amplitude should always be a constant factor of @xmath203 smaller than the full oscillation amplitude . the candidates for the @xmath205 fqh plateau are , of course , similar to the @xmath101 candidates , since the filling fractions are particle - hole dual . we stress , however , that the physically observed states at these filling fractions need not be particle - hole dual to each other , since physical effects , such as landau level mixing , will tend to break particle - hole symmetry at these fillings . the leading candidates are the laughlin state @xcite , two types of bs states @xcite , and a @xmath172-clustered rr state @xcite . these bs states again have non - abelian statistics derived from eq . ( [ eq : m2 ] ) using the fact that the non - abelian quasiparticles carry su@xmath1 charge @xmath177 . the rr@xmath180 state again is related to su@xmath181 , with non - abelian statistics derived from eq . ( [ eq : m4 ] ) and fundamental @xmath182 quasiparticles carrying su@xmath181 charge @xmath177 . @xmath206 we see in table [ table_8_3:qps ] that all of these states have an abelian @xmath207 excitation with scaling exponent @xmath208 , so there should always be a background of such excitations in tunneling . the @xmath209 , bs@xmath210 , and @xmath211 states all have an @xmath184 excitation also with @xmath208 , so these two excitations are expected to have roughly equal contribution to the tunneling in these theories . however , the smallest scaling exponent for the bs@xmath210 and @xmath211 states belong to non - abelian @xmath184 excitations , which are therefore expected to dominate the tunneling in these states . the smallest scaling exponent for the rr@xmath180 state belongs to the non - abelian @xmath182 excitation , which should thus dominate tunneling . there are additional relevant tunnelers for rr@xmath180 that are non - abelian with different statistics than the fundamental quasiparticle , namely the @xmath184 and @xmath5 excitations which carry su@xmath181 charges @xmath49 and @xmath212 , respectively . the experiments of ref . , which observes only @xmath184 tunneling , appear to exclude the rr@xmath180 state and best agree with the bs@xmath210 and @xmath211 states . the @xmath152 fqh plateau is most likely the standard ( particle - hole conjugate ) laughlin state @xcite , but we include this filling fraction for completeness , and list bs states @xcite as ( unlikely ) alternative candidates . these bs states again have non - abelian statistics derived from eq . ( [ eq : m2 ] ) using the fact that the non - abelian quasiparticles carry su@xmath1 charge @xmath177 . @xmath214 we see in table [ table_14_5:qps ] that all of these states have abelian @xmath194 and @xmath215 excitations with relevant scaling exponents , so there should always be a background of such excitations in tunneling . the @xmath194 excitation is the most relevant tunneler for the @xmath216 state . for the bs@xmath217 and @xmath218 states , the non - abelian @xmath196 excitation has the most relevant tunneling . while a fully developed fqh plateau has not been observed at @xmath220 , there does appear to be a feature of a developing plateau there @xcite , and it is the next filling fraction in the sequence of bs states following @xmath191 , so we will list these states @xcite . these bs states again have non - abelian statistics derived from eq . ( [ eq : m2 ] ) using the fact that the non - abelian quasiparticles carry su@xmath1 charge @xmath177 . @xmath221 we see in table [ table_19_8:qps ] that for both of these states the two smallest tunneling exponents belong to a non - abelian @xmath222 excitation and an abelian @xmath223 excitation . the scaling exponent of the abelian @xmath223 excitation is @xmath224 in both theories , while the non - abelian @xmath222 excitation has tunneling exponent @xmath225 for the bs state and @xmath226 for the @xmath227 state . thus , the contribution of the non - abelian @xmath222 excitation will be slightly stronger than that of the abelian @xmath223 excitation for the bs state , whereas it will be the other way around for the @xmath227 state . we also mention that a bs type hierarchy could be built over the su@xmath1 naf state to produce candidates for all the filling fractions listed above @xcite . the relevant data could be read off the above tables for the non - abelian quasiparticle excitations in the bs states by simply adding @xmath228 to @xmath78 and @xmath123 , and multiplying the twist factors by @xmath229 . in this section , we examine the tunneling amplitudes of @xmath4 and @xmath5 quasiparticles in detail , and argue that generically @xmath230 . when they are small , the tunneling amplitudes @xmath104 and @xmath105 are the matrix elements for the transfer of charge from one edge of a hall device to the other . for simplicity and concreteness , let us suppose that the device is a hall bar with a single constriction . then , the transfer of charge @xmath231 from one edge to the other entails a momentum change @xmath232 , where the @xmath233-direction is along the hall bar . ( this is seen most easily in landau gauge , as we discuss below in the context of specific trial wavefunctions . ) however , in order to cause a momentum change of @xmath234 , the potential due to the gates must have weight at this wavevector , i.e. the matrix element is determined by the variation of the potential on a length scale @xmath235 . hence , in order to transfer charge @xmath231 from one edge to the other , we need the potential to vary on a length scale @xmath236 satisfying @xmath237 if the constriction were much smaller than this , then we would expect that the potential would have comparable weight at the wavevector necessary for charge @xmath4 transfer , @xmath238 , and at the larger wavevector necessary for charge @xmath5 transfer , @xmath239 . otherwise , we expect the weight to fall off rapidly with the wavevector , and to have @xmath240 . this can be made a little more precise by considering , for the sake of concreteness , the mr pfaffian state . we work in landau gauge on a cylinder @xcite : @xmath241 where @xmath242 , @xmath243 and @xmath244 are the coordinates around and along the cylinder , respectively , and @xmath245 is the radius of the cylinder . @xmath246 is a symmetric polynomial which deforms the shape of the hall droplet from a rotationally - symmetric band around the cylinder to one with a constriction . for instance , we could take @xmath247 where @xmath248 and @xmath249 are points _ outside _ the droplet with the same @xmath233-coordinate . the precise form of @xmath246 is not important at the present level of discussion , but we will assume that it is a polynomial of degree @xmath231 which is less than @xmath250 . then , the wavefunction @xmath251 has charge @xmath4 transferred from one edge to the other , while @xmath252 has charge @xmath5 transferred from one edge to the other . the tunneling matrix elements @xmath104 and @xmath105 for charge-@xmath4 and @xmath5 quasiparticles , respectively , are @xmath253 where @xmath254 and @xmath255 is the potential due to the gates which define the point contact . while we would need a detailed knowledge of @xmath255 and of the precise shape of the hall droplet in order to determine @xmath104 and @xmath105 quantitatively , we can make a few qualitative remarks which echo our earlier observations . first , let us ignore @xmath246 . then , @xmath256 are eigenstates of angular momentum around the cylinder with eigenvalues @xmath257 . thus , the tunneling matrix elements @xmath104 and @xmath105 are controlled by @xmath258 for @xmath259 and @xmath260 , respectively . these will be comparable if the scale @xmath236 over which the potential varies in the @xmath233-direction is smaller than @xmath261 . but the distance between the two edges @xmath262 is @xmath263 . hence , we need @xmath264 in order for the two tunneling matrix elements to be comparable . otherwise , both are determined by the tails of the ( fourier transform of the ) potential and @xmath240 . the presence of the constrictions , which is reflected in @xmath246 means that the wavefunctions are no longer angular momentum eigenstates . instead , @xmath265 has non - zero amplitude for a range of angular momenta @xmath266 while @xmath267 has non - zero amplitude for a range @xmath268 , and similarly for @xmath269 . here , @xmath270 is determined by @xmath246 ; the minimum distance between the two edges at the constriction is @xmath271 . thus , the tunneling matrix elements @xmath104 and @xmath105 are controlled by @xmath272 and @xmath273 . hence , we obtain the same requirement as above , but with @xmath262 now understood as the distance between the two edges at their point of closest approach .

the many similarities between the prompt emission pulses in gamma ray bursts ( grbs ) and x - ray flares during the fast decay and afterglow phases of grbs suggest a common origin . in the cannonball ( cb ) model of grbs , this common origin is mass accretion episodes of fall - back matter on a newly born compact object . the prompt emission pulses are produced by a bipolar jet of highly relativistic plasmoids ( cbs ) ejected in the early , major episodes of mass accretion . as the accretion material is consumed , one may expect the engine s activity to weaken . x - ray flares ending the prompt emission and during the afterglow phase are produced in such delayed episodes of mass accretion . the common engine , environment and radiation mechanisms ( inverse compton scattering and synchrotron radiation ) produce their observed similarities . flares in both long grbs and short hard gamma ray bursts ( shbs ) can also be produced by bipolar ejections of cbs following a phase transition in compact objects due to loss of angular momentum and/or cooling . optical flares , however , are mostly produced in collisions of cbs with massive stellar winds / ejecta or with density bumps along their path . in this paper we show that the master formulae of the cb model of grbs and shbs , which reproduce very well their prompt emission pulses and their smooth afterglows , seem to reproduce also very well the lightcurves and spectral evolution of the prominent x - ray and optical flares that are well sampled . -1 cm

0908.0650

a flaring activity during the afterglow ( ag ) phase of a gamma ray burst ( grb ) was first observed in the late - time ag of grb 970508 with the narrow field instrument ( nfi ) aboard the bepposax satellite in the x - ray band @xcite , and with ground based telescopes in the optical band @xcite . it was interpreted in the framework of the fireball ( fb ) model of grbs as a delayed burst from the central grb engine @xcite . alternatively , in the cannonball ( cb ) model of grbs it was interpreted as a synchrotron radiation ( sr ) flare from an encounter of the highly relativistic jetted ejecta from an underlying supernova ( sn ) explosion with a density jump in the interstellar medium @xcite . late - time flares were later discovered in the broadband ag of several other grbs that were localized by the bepposax satellite , most notably in grb 000301c at @xmath0 days after burst @xcite where the flare was attributed to gravitational microlensing @xcite and in grb 030329 @xcite where the flare was interpreted in the framework of the fb model as due to ` refreshed shocks ' generated by a late activity of the central engine @xcite . in the cb model , however , these flares were well reproduced by the emission of sr from encounters of the jetted ejecta from an underlying sn explosion in a star formation region with density jumps within or at the border of a super bubble created by the star formation region @xcite . early - time x - ray flares ending the prompt emission phase were also detected by the wide field camera ( wfc ) aboard bepposax in a few grbs such as grb 011121 . in the fb model they were attributed to the onset of the external shock in the circumburst material . in the cb model they were interpreted as being due to the last episodes of bipolar cb ejections from a shutting off central engine . shortly after the launch of the swift satellite in november 2004 , data collected with its x - ray telescope ( xrt ) showed that x - ray flares are quite common in all the phases of the emission from grbs . in more than 50% of the grbs observed with the swift x - ray telescope ( xrt ) , flares were observed at the end of the prompt emission and/or the early ag phase @xcite . in some cases x - ray flares were observed also at very late times , of the order of several days after the prompt emission . although the information on flares is much more sketchy compared to that on the prompt gamma ray pulses , their spectral and temporal behaviours show clearly that the x - ray flares during the prompt @xmath1-ray emission follow the pattern of the @xmath1-ray pulses , suggesting they are the low energy part of these pulses . the x - ray flares ending the prompt emission and those superimposed on the early - time afterglow have a fast spectral evolution . their peak intensities decrease with time , and their spectral and temporal behaviours are similar to those of the prompt x/@xmath1-ray pulses , except that they are progressively softer and last longer . in most cases their @xmath1-ray emission probably is below the detection sensitivity of the swift broad alert telescope ( bat ) . in some cases their fluence in the xrt band exceeded that of the prompt emission in the bat 15 - 150 kev band @xcite . late - time ( @xmath2 s ) x - ray flares , however , seem to exhibit temporal and spectral behaviours that are different from those of most of the early - time flares . their power - law decline is more moderate , they are `` achromatic '' with a power - law spectrum almost identical to that of the late - time ag , and they show very little spectral evolution ( see , e.g. , the late time broad band flares in grbs 060614 @xcite and 081028 @xcite , and the hardness ratio during late time flares in swift grbs reported in the swift - xrt lightcurve repository , @xcite ) . flares in the x - ray lightcurve of swift grbs were studied phenomenologically by various observer groups @xcite . modifications of previously suggested models and new theoretical models were proposed and discussed by several authors @xcite , but none of these proposed models was shown to actually derive the observed lightcurves and spectral evolution of either early - time or late - time flares from underlying physical assumptions . flares are a natural consequence of the cannonball ( cb ) model of grbs , which was motivated by a grb - microquasar analogy @xcite . in the cb model , _ long - duration _ grbs and their ags are produced by bipolar jets of highly relativistic plasmoids of ordinary matter ejected in accretion episodes on the newly formed compact stellar object @xcite in core - collapse supernova ( sn ) explosions @xcite . it is hypothesized that an accretion disk or a torus is produced around the newly formed compact object , either by stellar material originally close to the surface of the imploding core and left behind by the explosion - generating outgoing shock , or by more distant stellar matter falling back after its passage @xcite . as observed in microquasars @xcite , each time part of the accretion disk falls abruptly onto the compact object , two jets of cannonballs ( cbs ) made of _ ordinary - matter plasma _ are emitted with large bulk - motion lorentz factors in opposite directions along the rotation axis , wherefrom matter has already fallen back onto the compact object due to lack of rotational support . the entire radiation emitted from a grb is produced by the interaction of the jet of cbs with the environmemt along its path , as illustrated in fig . the prompt @xmath1-ray and x - ray emission is dominated by inverse compton scattering ( ics ) of photons of the sn light filling the cavity produced by the pre - supernova wind / ejecta blown from the progenitor star long before the grb . the cbs electrons compton up - scatter this ` glory ' light into a narrow conical beam of @xmath1 rays along the cbs direction of motion . each cb produces a single grb pulse . an x - ray ` flare ' coincident in time with a prompt @xmath1-ray pulse is simply its low - energy part . the natural explanation of flares ending the prompt emission and during the early time afterglow is the same : ics of glory photons by the electrons of cbs ejected in late accretion episodes of fall - back matter on the newly formed central object . early - time x - ray flares without an accompanying detectable @xmath1-ray emission are usually ic flares ( icfs ) produced by cbs with relatively smaller lorentz factors , due to weakening activity of the engine : as the accretion material is consumed , one may expect the ` engine ' to have a few progressively - weakening dying pangs . like the lightcurves of the prompt grb pulses , the lightcurves of icfs exhibit a rapid softening during their fast decline phase ( see , e.g. , the xrt hardness ratio reported for swift grbs in the swift lightcurve repository , @xcite ) . in the cb model , each ic flare ( icf ) is followed by the emission of sr flare ( srf ) from the encounter of the cb with the wind / ejecta , which was blown from the progenitor star long before the grb ( see fig . [ fig1 ] ) . because of time - aberration in the observer frame , these srfs appear to have only short lag - times relative to the corresponding icfs . below their peak energy , the spectral behaviour of the ics pulses / flares is roughly @xmath3 , while the sr emission from fast cooling electrons has typically @xmath4 . thus , while the prompt kev - mev pulses / flares are dominated by ics of glory light , the ` prompt ' optical emission is dominated by sr . as the glory extends into the wind , often the sr emission begins before the ics pulse / flare has ended . the initial expansion of the cbs and the slow - down of the leading ones by the circumburst matter may merge most of them into a single leading cb @xcite during the afterglow phase . its collimated beam of the prompt gamma rays ionizes the matter in front of it . the ions continuously impinging on a cb with a relative lorentz factor @xmath5 , where @xmath5 is the bulk motion lorentz factor of the cb , generate within it an equipartition turbulent magnetic field . the intercepted electrons are isotropized and fermi accelerated by these fields and emit isotropic synchrotron radiation in the cb s rest frame , which is doppler boosted and beamed relativistically into a narrow cone with a typical opening angle @xmath6 in the observer s rest frame . late srfs are produced mainly when the cbs encounter winds or density bumps along their path first from the progenitor star and later in the interstellar medium ( ism ) . the lightcurve of these flares depends on the unknown density profile of the encountered wind / density bump that can not be predicted a - priori . but , both the early - time and the late - time srfs have a typical sr spectrum and a weak spectral evolution that are quite different from those of the accretion induced icfs and can be used to identify their origin late ejection episodes from the central engine or density bumps . in the cb model , short hard bursts ( shbs ) are also produced by bipolar jets of plasmoids ejected in mergers of compact objects in close binary systems such as neutron stars merger @xcite or in mass accretion episodes on compact objects in close binary systems , or in phase transitions in compact stars ( neutron stars , hyper stars and strange quark stars ) @xcite . bipolar ejections in late accretion episodes or phase transitions after cooling and loss of angular momentum probably produce the observed icfs in shbs , and sr radiation from encounters with winds / density bumps produces srfs . flares were routinely included in the cb model description of the afterglows of grbs and shbs @xcite . they were calculated from the master formulae of the model , which describe well the prompt ics emission and the emission of sr at all times . it was shown that ics explains successfully both the prompt kev - mev pulses and the x - ray flares ending the prompt emission , while the sr emitted in the collision of the jet of cbs with the wind / ejecta from the progenitor star explains well the prompt optical flares measured with robotic telescopes in very bright grbs @xcite . however , attention was focused there on the general properties of the prompt emission and the smooth afterglow , rather than on flares during the afterglow phase . moreover , in many grb afterglows , flares are weak , or are blended , or are not well sampled , and their properties could neither be inferred reliably from the ag lightcurve nor used reliably to test theoretical models . the situation concerning prominent x - ray flares observed by the swift xrt is different . they are well resolved and their spectral properties are much better measured . in this paper we focus our attention on grb x - ray and optical lightcurves with prominent flares . in particular , we compare the cb model lightcurves and their spectral evolution with representatives set of x - ray and optical lightcurves measured with the swift xrt , and with ground - based robotic telescopes and swift uvo in space , respectively , which have prominent flares that are well sampled . such a comparison provides stringent tests of both the cb model and its interpretation of the origin of prompt emission pulses in grbs , their afterglows and the early and late time flares in their lightcurves . we show that the cb model correctly predicts their main observed properties , and reproduces well their entire lightcurves and spectral properties . for completeness , we first summarize the relevant master formula of the cb model and their simplified forms that we later use in our analysis of the x - ray and optical lightcurve of grbs . in this section we summarize the master formula of the cb model for the pulse shape and spectral evolution of ics pulses / flares ( see @xcite and references therein for their derivation ) . let @xmath7 denote the time in the observer frame after the beginning of a flare . the light - curve of a flare , produced by the electrons in the cbs by ics of thin bremsstrahlung photons filling the cavity formed by a wind blown by the progenitor star long before the grb , is generally well approximated by @xcite : @xmath8 where @xmath9 is a constant that depends on the cb s baryon number , lorentz and doppler factors , and on the density of the glory light and on the redshift and distance of the grb , and @xmath10 , the peak energy of @xmath11 at time @xmath7 , is given roughly by : @xmath12 with @xmath13 being the time ( after the beginning of the flare ) when the ics photon count - rate reaches its peak value . for @xmath14 , it satisfies @xmath15 where @xmath16 is the peak energy of the time - integrated spectrum of the flare . thus , in the cb model , each ics pulse in the grb lightcurve is described by four parameters , @xmath17 @xmath18 @xmath19 and the beginning time of the pulse when @xmath7 is taken to be 0 . ( [ icslc ] ) with @xmath16 given by eq . ( [ peake ] ) describes well the shape and the spectral evolution of grb pulses and of early - time x - ray flares . in particular , it correctly describes their rapid spectral softening during their fast decline as demonstrated in figs . [ fig2]c , f , [ fig3]b and [ fig7]d , and in @xcite for many more cases . if absorption in the cb is dominated by free - free transitions , then @xmath20 , and for @xmath14 the lightcurve of an icf is approximately a function of @xmath21 ( the ` @xmath22 ' law ) , with a peak at @xmath23 , a full width at half maximum , fwhm @xmath24 and a rise time from half peak value to peak value , @xmath25 independent of e. note that the approximate @xmath21 law makes the fast decline sensitive only to the product @xmath26 and not to their individual values . this degeneracy in the pulse shape can be removed by inferring @xmath16 from the broad band spectrum of the icf . the late - time decay of the energy flux of the prompt emission pulses and icfs in an energy band @xmath27 $ ] , which follows from eq . ( [ icslc ] ) , is given approximately by : @xmath28 . \label{inicslc}\ ] ] thus , for the swift xrt lightcurves where @xmath29 kev and @xmath30 kev , as long as @xmath31 , the energy flux in an ics pulse / flare decays like @xmath32 until it is taken over by the sr afterglow . when @xmath33 but @xmath34 the energy flux decays like @xmath35 and when @xmath36 the energy flux decays like @xmath37 . for a single icf beginning at @xmath7=0 , which is superimposed on a smooth sr afterglow , the effective photon spectral index @xmath38 is given approximately by @xcite : @xmath39 \theta(t_{ag}-t)\ , \theta(t ) + \gamma_{sr}\ , \theta(t - t_{ag})\ , , \label{geff}\ ] ] where @xmath40 is the time when the sr afterglow takes over the ics emission , @xmath41 for a glory with thin thermal bremsstrahlung spectrum , and @xmath42 is the best fit photon spectral index of the smooth x - ray ag . all the above properties are clear fingerprints of icfs produced by highly relativistic cbs fired in mass accretion episodes on the newly formed compact central object . moreover , the observations of x - ray flares indicate that their widths are proportional to their emission time @xmath43 after the grb trigger . in the cb model , this implies that their peak energy , equivalent isotropic gamma ray energy and peak luminosioty decrease like @xmath44 @xmath45 ^ 2 $ ] and @xmath46 ^ 3\,.$ ] in the cb model , the lightcurve in the observer frame of the sr emitted by a cb is given by the master formula @xcite : @xmath47 \propto a(\nu , t)\,{n(t)\ , r(t)^2\ \gamma(t)^2\ , \delta(t)^4 \over \nu_b(t)}\ , \left[{\nu\over \nu_b(t)}\right]^{-1/2}\ , \left[1+{\nu\over \nu_b(t)}\right]^{(1-p)/2}\ , , \label{fnusr}\ ] ] where @xmath7 is the time after the ejection of the cb , @xmath48 is its radius , @xmath49 is the density along its trajectory , @xmath50 is the attenuation of radiation along the line of sight to it , @xmath51 is the typical frequency in the observer frame of the sr emitted by the electrons that are swept into it at time @xmath7 with a relative lorentz factor @xmath5 , @xmath52 and @xmath53 is the spectral index of the fermi accelerated electrons in it . the sr radiation that is emitted from the encounter of a cb with the wind / ejecta of the progenitor star , with a density profile @xmath54 for @xmath55 and @xmath56 for @xmath57 , follows from eq . ( [ fnusr ] ) and is given approximately by @xcite : @xmath58 where @xmath59 with @xmath60 being the time after trigger and @xmath61 the time at the cb - wind encounter , @xmath62 is the typical slow - down time of the fast cb expansion , @xmath63 , and the exponent describes the decreasing attenuation of the emitted radiation when the cb penetrates the wind and/or the initial rise in the wind density due to an exponential cutoff of the wind ejection . note that for @xmath64 the asymptotic decline of an srf is a simple power law @xcite , @xmath47 \propto t^{-\gamma}\ , \nu^{-\gamma+1}\ , , \label{srtail}\ ] ] while that of an icf is , @xmath47\propto t^{-2}\ , e^{-\beta_g}\ , e^{-e / e_p(t ) } \sim t^{-2}\ , e^{-\beta_g}\ , e^{-e\,(t^2+t_p^2)/2\ , e_p\ , t_p^2}\ , . \label{icstail}\ ] ] thus , icfs and srfs may be distinguished by their different tempo - spectral evolution . in the x - ray band , early - time icfs are usually much brighter than their following srfs . but , due to their rapid late - time decay , occasionally the @xmath65 tail ( eq . [ srtail ] ) of the srf , which follows an icf , can be seen before the plateau / shallow decay phase of the early - time x - ray ag takes over ( see , e.g. , fig . 3 in @xcite ) . the early time sr has usually @xmath66 and @xmath67 that decreases strongly between the optical and the x - ray band . consequently , although ics dominates the prompt x - ray emission , in the optical band srfs are usually much brighter than their preceding icfs , which typically have @xmath68 around their peak time . consequently , optical flares are usually srfs . the evolution of the effective spectral index during an optical srf that follows from eq . ( [ fnusr ] ) has the simple form , @xmath69\over dlog\ , \nu}= 1/2+(\beta_x-1/2)\ , { \nu\over \nu + \nu_b}\ , . \label{betao}\ ] ] in early - time optical srfs , @xmath70 that is initially below the optical band can cross above it as @xmath49 of the wind / ejecta increases and then cross back as the density decreases . such a change in @xmath71 , from @xmath72 towards @xmath73 and back to @xmath72 , has been observed in some early time optical flares , e.g. , in grb 071031 @xcite . when the merged cbs coast through the constant - density ism , their sr in the x - ray band is well above the cooling frequency of the fermi accelerated electrons and their unabsorbed ag as given by eq . ( [ fnusr ] ) reduces to @xcite : @xmath74 \propto \gamma(t)^{3-\beta_x}\ , \delta(t)^{3\,\beta_x+1}\ , \nu^{-\beta_x } \label{fnu}\ ] ] where @xmath75 is their doppler factor with @xmath76 being the angle between the line of sight to the cbs and their direction of motion . for @xmath77 and @xmath78 , @xmath79 to an excellent approximation . in the cb model the canonical value of the spectral index well above the bend frequency @xmath70 has the value @xmath80 for a cb of a baryon number @xmath81 , a radius @xmath82 and an initial lorentz factor @xmath83 , relativistic energy - momentum conservation yields the deceleration law of the cb in an ism with a constant density @xmath49 @xcite : @xmath84^{1/2}}\ , , \label{goft}\ ] ] with @xmath85 as can be seen from eq . ( [ goft ] ) , @xmath1 and @xmath86 change little as long as @xmath87 ^ 2\,t_0\ , , $ ] and eq . ( [ fnu ] ) yields the _ ` plateau ' _ phase of ` canonical ags ' @xcite . for @xmath88 , @xmath1 and @xmath86 decrease like @xmath89 the transition @xmath90 around @xmath91 induces a bend , the so called ` jet break ' , in the synchrotron ag from a plateau to an asymptotic power - law decay , @xmath74 \propto t^{-\beta_x-1/2}\,\nu^{-\beta_x}\ , . \label{asymptotic}\ ] ] thus , the shape of the entire lightcurve of the sr afterglow from a cb that enters the constant density ism depends only on three parameters , the product @xmath92 , the deceleration parameter @xmath93 ( or the break time @xmath91 ) and the spectral index @xmath94 . the post break decline is given by the simple power - law ( eq . [ asymptotic ] ) independent of the values of @xmath92 and @xmath91 . in cases where @xmath91 is earlier than the beginning of the xrt observations or is hidden under the prompt emission , the entire observed lightcurve of the ag has this asymptotic power - law behaviour @xcite . the lightcurve of late - time srfs strongly depend on the density profile of the density bumps . for a wind - like density jump @xmath95 beyond @xmath96 , the lightcurve is given by eq . ( [ srp ] ) . such a density profile is expected for the boundaries of star formation regions where grbs usually take place . at late times both the optical and the x - ray bands are above the bend frequency , @xmath97 , and the asymptotic decline of the corresponding late - time srf is given by : @xmath47 \propto t^{-\gamma}\,\nu^{-\gamma-1}\ , , \label{asymptoticrm2}\ ] ] where @xmath59 is the observer - time after the cb has reached @xmath98 . at late times , the bend frequency is well below the optical band and remains so during the crossing of density bumps . thus , no detectable spectral variation is expected in late - time srfs . because of the large bulk motion lorentz factors of the jets of cbs , doppler boosting , relativistic beaming and time aberration yield strong dependence of their observed radiations on @xmath1 and @xmath86 . this dependence dominates the flare observables and can be used to correlate triplets of independent flare observables without knowing the exact values of @xmath1 and @xmath86 . moreover , many independent observables depend on the same combinations of @xmath1 and @xmath86 , which results in pair correlations . finally , due to selection effects , various observables depend strongly only on the lorentz factor or on the doppler factor , which also yields pair - correlations . these correlations between various radiations , between flare observables and between flare and grb observables are discussed in detail in @xcite . although in the cb model early - time optical srfs follow the x - ray icfs , they have a different origin and a different tempo - spectral evolution . thus , we shall compare separately the observational data on x - ray flares and optical flares and the cb model predictions . in the cb model , the x - ray emission in grbs , xrfs and shbs during the prompt emission phase is a sum of x - ray flares that are part of their prompt gamma ray pulses . that has been shown repeatedly in cb model publications ( see , e.g. , @xcite ) and independently by many other authors . we have also shown that the fast decline phase of their prompt emission with its rapid spectral softening is just the tail of the last prompt emission pulses , because the exponential factor in eq . ( [ icslc ] ) suppresses very fast the relative contribution of the earlier pulses by the time the data sample the later pulses or flares . this is demonstrated in figs . 2a , b , c where we compare between the ` canonical x - ray lightcurve ' @xcite of grb 060729 and the evolution of its spectral index as inferred by @xcite and the cb model descriptions of these lightcurves . this has already been demonstrated in detail in @xcite for many other grbs and xrfs . it is also the case for shbs as shown in figs . 2e,2f for shb 050724 and in @xcite for many other shbs . grbs with a non - canonical x - ray ag , are simply grbs where the emission of sr begins to dominate the x - ray lightcurve before the fast decline phase of the prompt emission . such a case is shown in fig . 3d where we compare the xrt lightcurve of grb 060418 and its cb model description . many other cases are shown in @xcite . in order to have a stringent test of the cb model interpretation of x - ray flares during the afterglow phase of grbs , we have selected a sample of 14 grbs with x - ray lightcurves that are reported in the swift / xrt grb lightcurve repository @xcite and have well sampled prominent flares during their ag phase . this sample includes grbs 050502b , 050916 , 060526 , 060929 , 070704 , 080506 , 080607 , 080810 , 080906 , 081102 , 090417b , 090621a , 090709a , and the shb 050724 . for all these grbs we have fitted the entire xrt lightcurve with the master formulae of the cb model . in order to minimize the number of adjustable parameters in the theoretical lightcurves , we adopted the standard simplifying cb model assumptions @xcite : the burst environment is a cavity full of thin bremsstrahlung optical photons ( glory light ) enclosed within a wind / ejecta that have a density profile @xmath99 beyond @xmath98 until the density is taken over by the constant density of the ism . the cbs were taken to be well separated during the prompt emission phase and to be well represented by a single effective cb during the afterglow phase . the latest one or two observed pulses / flares in the prompt emission were assumed to dominate its fast decay . this fast icf decay is overtaken by the tail of the sr emission from the encounter of the cbs with the wind / ejecta as given by eq . ( [ srp ] ) , or by the plateau phase of the sr afterglow emitted from the decelerating cb in the constant - density ism . the smooth afterglow was calculated from eqs . ( [ fnu ] ) , ( [ goft ] ) , using best fit values of the normalization , @xmath100 , @xmath93 , and @xmath101=@xmath94 with @xmath94 within the error range reported for @xmath102 + 1 in the swift x - ray repository @xcite . prominent x - ray flares during the afterglow with a rapid spectral softening during their decline phase were assumed to be ics pulses / flares . such flares were superimposed on the cb model smooth sr afterglow . flares with a constant hardness ratio similar to that of the smooth sr afterglow were assumed to be srfs . they were generated by introducing density bumps with a windy profile into the master formula of the sr lightcurve @xcite . the fitted parameters of the cb model descriptions of the 14 x - ray light curves of the above grbs are listed in tables [ t1],[t2 ] . when only the tail of the srf was visible we used its parameter - free asymptotic form , eq . ( [ srtail ] ) . because of the use of simplifying assumptions , the values of the parameters may be effective values and not true values . therefore , we refer to the cb model fits as ` descriptions ' rather than as best fit predictions . in order to avoid repetitions and an excessively long section , we limit our detailed discussion to three representative cases , grb 050502b representing long soft grbs , shb 050724 representing short hard bursts , and the recent peculiar grb 090709a with a suspected 8 s periodicity in its prompt emission . * grb 050502b * was studied in detail by @xcite . the xrt began taking data 63 s after the bat trigger and followed its x - ray lightcurve until 10.6 days after burst . the measured lightcurve in the 0.3 - 10 kev band is shown in figure 2a . following an initial low - flux level , the xrt detected a giant x - ray flare , which began at @xmath103 s , reached a peak value around 770 s , with intensity more than 500 times larger than that of the underlying afterglow . the fluence of the flare , @xmath104 in the 0.2 - 10.0 kev energy band , exceeded that of the prompt emission measured with the swift bat in the 15 - 350 kev energy band . after several hours , two weaker flares in the x - ray emission occurred consecutively beyond which the decay of the x - ray lightcurve became steeper . except for the giant flare , the spectrum of the afterglow and the two late flares was well fit by a power - law with a photon spectral index @xmath105=1.945 ( + 0.077 , -0.100 ) @xcite . the spectrum of the flare was well fit with a power - law with an exponential cutoff ; however , due to its non detection by the bat , its value could not be well determined from the spectrum measured only by the xrt . the photon spectral index of the giant flare before peak - time was @xmath106 , much harder than that measured before and after the flare . during the fast decline phase of the flare , a rapid spectral softening took place and the spectral index increased rapidly to a value well above the constant value of the smooth afterglow . cb model fits to the x - ray lightcurve of grb 050502b and the evolution of its spectral index are shown in figs . they show that the observed lightcurve and spectral evolution of the giant x - ray flare are well described by the master formula ( eq . ( [ icslc ] ) ) of the cb model for an icf . as can be seen in figs . [ fig3]a , b , the fast decay and spectral softening stopped simultaneously when the ag was taken over by the smooth sr from the decelerating cb in an ism of a constant density . the late - time afterglow shows two flares superimposed on the smooth ag , similar to those observed in many long grbs . their spectrum , which is similar to the smooth ag , suggests that they were produced by enhancement of the emitted sr when the cb encountered density bumps in its voyage through the host s ism . such late flares with a typical sr spectrum and little spectral evolution produced by density bumps along the cb trajectory in the ism add no specific information on the origin of grbs . * shb 050724 * at redshift z = 0.257 was studied in detail by @xcite , citetgrupe2006 and @xcite . the swift bat triggered on the burst at 12:34:09 ut on july 24 , 2005 . the burst had @xmath107=3.0@xmath108 s , but most of the energy of the initial shb was released in a hard spike with a duration of 0.25 s. the bulk of the burst energy was not emitted in the short initial spike but in an extended soft emission component that lasted @xmath109 s. swift xrt began observing the afterglow 74 s after the bat trigger . the chandra x - ray observatory performed two observations , two days and about three weeks after the burst . the complete x - ray lightcurve is shown in fig . it has a rapid decay with a fast spectral softening ending with a sharp transition to a shallower decay with a much harder power - law spectrum , @xmath110 ( swift repository , @xcite ) . the ag steepens gradually into a late power - law decay . a large flare superimposed on the canonical lightcurve occurred around 50 ks after burst with a fluence of @xmath1117% of that of the prompt burst . the flare has been detected also in the optical and nir bands , e.g. , @xcite . spectral analysis of the xrt data @xcite showed no evolution during the afterglow phase , including the large late flare . spectral analysis of the chandra observations from the fading tail of this flare confirmed this result @xcite . the burst took place 2.5 kpc ( in projection ) from the center of an elliptical host galaxy @xcite . cb model fits to the x - ray lightcurve of shb 050724 and the evolution of its spectral index are shown in figs . the initial fast decay with a rapid spectral softening is described in the cb model by the tail of an icf . as can be seen in figs . 2b,2c , the fast decay and spectral softening stopped simultaneously when the ag was taken over by the sr from the decelerating cb in an ism of a constant density . as expected , the late - time sr afterglow is similar in shape to the sr afterglow of long grbs . also the late - time flare superimposed on the canonical lightcurve is similar to those observed in many long grbs . its spectrum , which is similar to the smooth ag , suggests that it was produced by enhancement of the emitted sr when the cb encountered a density bump in its voyage through the host s ism . the chandra data show that the canonical ag continued to decay after the flare with the same slope and the same spectral index , @xmath112 , as that of the ag before the srfs . as shown in fig . 2e , the complete xrt lightcurve is well described by the cb model . moreover , the cb model relation for the asymptotic decline of the smooth ag is well satisfied : the temporal behaviour of the smooth ag was best fit with p=1.56 , implying an unabsorbed spectral index , @xmath94=p/2=0.78 , in agreement with that inferred from the xrt and chandra observations . the observed elliptical host galaxy of shb 050724 was argued to provide strong support for a neutron star merger origin of this shb . but , it was pointed out that neutron star mergers do not produce the late accretion episodes needed to power a late central activity that could produce the large flare around 50 ks after burst @xcite . in the cb model , a late flare with a typical sr spectrum and little spectral evolution is produced by density bumps along the cb trajectory in the ism . such flares neither rule out nor support any specific origin of the shb . * grb 090709a * triggered the swift bat on july 9 , 2009 at 07:38:34 ut @xcite . its prompt emission lightcurve within 100 seconds after trigger appeared to show a quasi periodic variation with a period of @xmath113 seconds @xcite , which may have also seen in independent measurements of its lightcurve with konus - wind @xcite , integral @xcite , and suzaku @xcite . the swift xrt began its follow - up observations 68 seconds after the bat trigger @xcite . however , analysis by @xcite of the xrt data during 79 - 469 seconds after the bat trigger did not reveal any significant periodicity . from deep optical observations that did not detect its host galaxy , it was concluded that grb 090709a took place either in the milky way or at a redshift between 8 and 10 @xcite . the large redshift and the fluence of @xmath114 measured with konus - wind in the 20 kev-3 mev energy range imply an isotropic equivalent gamma - ray energy between @xmath115 and @xmath116 @xmath117 , which makes grb 090709a more luminous than any grb with known redshift . [ fig6]e , f present a comparison between the xrt lightcurve of grb 090709a and its cb model description assuming it was an ordinary grb . the early time flare around @xmath7=90 seconds probably is the x - ray part of the prompt emission pulse with a peak around 87 seconds in the swift bat lightcurve that is slightly delayed relative to the gamma - ray peak , as expected from the @xmath21 law of the cb model . its fast decline and rapid softening are those expected from an icf . they are taken over by sr around 150 seconds after the bat trigger . the next 3 peaks probably are sr peaks as suggested by their shape and their hardness ratio that is roughly the same as that of the sr afterglow . the best fit cb model lightcurve of the smooth ag yields @xmath118=2.16 , which , in the cb model , implies @xmath119 , in good agreement with the best fit spectral index reported in the swift repository @xcite , @xmath1201.081 ( + 0.076 , -0.074 ) . all together , grb 090709a looks like a normal grb with a normal early time flaring activity that took place at a relatively very large redshift , and its x - ray lightcurve is well reproduced by the cb model . the cb model predictions for early - time and late - time flares in the optical lightcurves of grbs are compared with observations for a representative set of grbs in figs . [ fig7]-[fig11 ] . for all these grbs we show both their measured optical and x - ray lightcurves ( if available ) and their cb model descriptions with the parameters listed in tables [ t1]-[t5 ] . only in a very few bright grbs was the prompt optical emission resolved into separate flares . two such cases , grb 080319b @xcite and grb 071003 @xcite are shown in fig . [ fig7 ] . in most grbs and xrfs , the prompt optical emission that appears as a single extended flare , probably , is a sum of unresolved flares . cases where the prompt optical flare was partially resolved into a sum of flares or where there is clear evidence for overlapping flares are , e.g. , grb 080330 @xcite , xrf 071031 @xcite , grb 061007 @xcite , which are shown in figs . [ fig8],[fig9 ] . in most grbs where the prompt optical emission was detected with robotic telescopes from the ground or from space ( with the swift / uvo ) , the prompt emission appears like a single flare , probably as a result of either strongly overlapping flares , or insufficient temporal resolution due to low statistics , or because the prompt emission did consist of a single flare . examples of grbs with a ` single ' prompt optical flare are 990123 @xcite , 030418 @xcite , 050820a @xcite , 081203a @xcite , 090102 @xcite , 090618 @xcite and 091029 @xcite , which are shown in figs . [ fig9]-[fig11 ] . examples of bright grbs with well resolved late - time optical flares include grb 030329 @xcite and grb 060206 @xcite shown in fig . [ fig11]f . however , in order not to inflate this paper , only grb 080319b , xrf 071031 , grb 061007 and grb 990123 are discussed here in detail . * grb 080319b * at redshift @xmath121 , the brightest grb observed so far , was simultaneously detected by the swift - burst alert telescope ( bat ) and the konus gamma - ray detector aboard the wind satellite @xcite . the location of grb 080319b was fortuitously only 10@xmath122 away from grb 080319a , which was detected by swift less than 30 minutes earlier , and allowed several wide field telescopes to detect the optical emission of grb 080319b instantly . it started after the beginning of the prompt kev - mev emission and it peaked 26 s after the swift trigger at magnitude @xmath123 @xcite visible to the naked eye . the extreme brightness of the burst and its gamma - ray , x - ray and @xmath124 afterglows led to a flurry of follow - up observations with a variety of space- and ground - based telescopes , which were summarized by @xcite . its isotropic equivalent gamma - ray energy release was @xmath125 erg , similar to that of grb 990123 . the fast spectral variation of its hard x - ray and gamma ray emission was well parametrized with an exponentially cut - off power - law with a cut - off energy that was strongly correlated with the peak structure of the lightcurve and a low - energy photon spectral index , @xmath106 , which changed abruptly into @xmath126 after the fast decay phase of the prompt emission . the optical and gamma - ray lightcurves during the explosion were not correlated ( see , e.g. , fig . 1 in @xcite ) : the onset of the optical emission lagged behind the gamma ray emission by several seconds and decayed more slowly at the end of the prompt emission . the typical time scales of their temporal variability were entirely different . the extremely bright optical emission could not be reconciled with a single emission mechanism - extrapolating the gamma - ray spectrum to the optical band underestimates the optical flux by more than 4 orders of magnitude . their spectra were also quite different . contrary to fireball ( fb ) model expectations , the x - ray and @xmath127 afterglow lightcurves were also chromatic , with no obvious ` jet breaks ' and with spectral and temporal power - law decays that did not satisfy the closure relations expected in the fb model ( see , however , @xcite for attempts to reconcile the observations with the fireball model ) . the prompt @xmath1-ray and hard x - ray emission in grb 080319b is composed of many narrow peaks ( see fig . 1 in @xcite ) , most of which are not well resolved . its 0.3 - 10 kev x - ray lightcurve measured with the swift xrt @xcite and its cb model description assuming a constant ism density until around @xmath128 s , presumably when the cb escaped the star formation region into the halo of the host galaxy , are shown in fig . the best fit parameters are listed in table [ t1 ] . the late time temporal decay of the x - ray ag is well described by a power - law with @xmath129 , except around 40 ks , where the lightcurve is poorly sampled . this value of @xmath130 satisfies well the cb model closure relation @xmath130=@xmath131=@xmath132 . as expected for grbs with large measured @xmath16 , @xmath133 and @xmath134 @xcite , no ag break was observed in the xrt lightcurve before @xmath135 s. the wiggling of the measured lightcurve around a power - law decay is probably due to variations in the ism density along the cb trajectory , which we have not tried to parametrize . a late - time srf with a typical late - time decay @xmath136 probably was observed around @xmath135 s. in fig . 7b we compare the measured @xmath82-band ( and @xmath137 band renormalized to the @xmath82 band ) lightcurve of grb 080319b @xcite and its cb model description assuming that the initially expanding 3 cbs merged into a single cb by the end of the prompt ics emission of gamma - rays and hard x - rays around 300 s ( observer time ) , which decelerates in roughly a constant density ism . the afterglow parameters are listed in table [ t3 ] . the ` missing jet break ' probably is hidden under the prompt emission . shown also is the contribution to the @xmath82-band afterglow from an sn akin to sn1998bw @xcite displaced to the grb site . the early - time optical emission that was resolved into three prominent peaks and its cb model description in terms of 3 sr peaks , each one described by eq . ( [ srp ] ) with the parameters listed in table [ t3 ] , are compared in fig . the decay of the prompt emission can also be reproduced assuming a single cb crossing 3 wind shells , which were ejected by the progenitor star long before its sn explosion , rather than 3 cbs crossing a continuous pre - supernova wind blown by the sn progenitor . in fig . 7d , the observed evolution of the spectral index in the 15 - 150 kev band during the prompt emission and the afterglow @xcite are compared with that predicted by the cb model . the predicted sharp transition from the prompt emission , which is dominated by ics of thin bremmstrahlung ( @xmath106 ) , to sr ( @xmath138 ) , which dominates the emission once the cbs encounter the progenitor s wind / ejecta , is clearly observed . * xrf 071031 * was studied in detail by kruhler et al . the swift / xrt began follow - up observations of xrf 071031 103 s after the burst . the early xrt lightcurve ( xrt lightcurve repository , @xcite is dominated by bright flares at around 120 s , 150 s , 200 s , 250 s and 450 s. the late x - ray data exhibit re - brightenings at 5.5 ks , 20 ks and 55 ks superimposed on a smooth ag . the complete xrt lightcurve was reproduced with the cb model , assuming 7 early - time icfs plus two late time srfs superimposed on a smooth sr afterglow . their parameters are listed in table [ t2 ] . as shown in figs . [ fig8]c , d the cb model describes well the very complex xrt lightcurve ( @xmath139 ) . the values of the photon spectral index of the late time afterglow , @xmath140=1.86 ( + 0.14 , -0.15 ) , and the index of the late time power - law decay , @xmath141 , which were inferred from the swift / xrt observations ( swift xrt lightcurve repository , @xcite , satisfy well the cb model closure relation , @xmath142 - 1/2 . the emission in the optical band was detected and followed up from the ground in automated observations by grond @xcite with good temporal resolution , which began 225 s after burst and lasted nearly 7 h. the white lightcurve obtained by adding the various optical bands , in order to increase sensitivity , shows a broad peak with clear deviations from a power - law rise and decay . [ fig8]e shows the cb model description of this lightcurve in terms of strongly overlapping 4 srfs following their preceding icfs in the x - ray lightcurve . the parameters used in the cb model description are listed in table [ t3 ] . comparison between the optical spectral index lightcurve that was inferred by @xcite and the cb model prediction as given by eq . ( [ betao ] ) is shown in fig . [ fig8]f . * grb 061007 * at a redshift @xmath143=1.26 @xcite , which was detected by swift , had a t90 duration of 75.3 s in the swift bat 15 - 350 kev band , and consisted of three large peaks with a long faint tail @xcite . the swift xrt began observations 80 s after the start of the burst . the optical observations began 27 s after the start of the burst by the rotse - iiia robotic telescope @xcite the optical lightcurve shows a sharp rise , brightening by over a factor of 50 in less than 5 s , followed by two peaks and a steady power - law decline . the @xmath1-ray and x - ray lightcurves have multiple peaks that are not contemporaneous with the optical peaks @xcite , followed by a steady decline in the x - rays that tracks the optical decline . the three overlapping optical peaks could be the three srfs following the three large bat peaks . in fig . [ fig9]c we compare the 0.3 - 10 kev swift xrt lightcurve of grb 061007 ( swift xrt lightcurve repository , @xcite ) , and its cb model description as the tail of the prompt emission srfs overtaken around @xmath7=300 s by a standard ag with an early - time break hidden under the prompt emission . the best fit parameters are listed in table [ t1 ] . note that both the early - time and late - time decays satisfy well the cb model closure relations : @xmath144 for the early - time decay and @xmath145 for the late - time decay , respectively . in fig . [ fig9]d we compare the optical lightcurve of grb 061007 as measured by rotse - iiia @xcite and its cb model description in terms of three overlapping srfs and a standard ag taking over around @xmath7=400 s. note that the late - time optical ag also satisfies well the cb model prediction , @xmath146 . * grb 990123 * at redshift @xmath147 @xcite was for a long time the brightest known grb . it was also the first grb in which an optical emission was detected during the prompt @xmath1/x - ray emission . grb 990123 was detected and localized by the burst and transient source experiment ( batse ) on board the compton gamma ray observatory ( cgro ) in the kev - mev range and at higher energies by the comptel , osse and egret instruments @xcite . it was also detected and localized by the gamma ray burst monitor ( gbm ) aboard the bepposax satellite @xcite . the lightcurves of the prompt emission in the kev - mev range showed a complex structure of at least 9 pulses / flares ( see fig . [ fig9]a ) . the prompt optical emission that was detected by the robotic optical transient search experiment ( rotse ) at los alamos 22 s after the onset of the burst , brightened and peaked at magnitude @xmath148 , about 50 s after the grb onset , and decayed rapidly with time @xcite . the prompt optical flare was not resolved into separate flares . it was followed in the @xmath149 bands with large ground based telescopes @xcite and with the hubble space telescope until it faded to a magnitude @xmath150 , two months after burst @xcite . the broad band @xmath1-ray , x - ray , @xmath127 and @xmath151 lightcurves of grb 990123 were reanalyzed recently within the synchrotron fireball ( fb ) model by corsi et al . essentially they found that the spectral and temporal properties of the prompt optical emission are uncorrelated to the @xmath1 and x - ray emission , implying different physical origins , that the optical and x - ray afterglow lightcurves are chromatic contrary to expectations , and that their spectral and temporal power - law decays do not satisfy the closure relations of the fb model . in fig . [ fig9]a we compare the batse multi - peak lightcurve of grb 990123 in the 20 - 50 kev channel @xcite and its cb model description . the count - rate in the 20 - 50 kev energy band was calculated from the integral @xmath152 using eq . ( [ icslc ] ) with the best fit parameters , which are listed in table [ t4 ] for the 9 peaks suggested by the multichannel batse data and by the bepposax data @xcite . as shown in fig . [ fig9]a , the shape of the peaks and the entire lightcurve are well reproduced by eq . ( [ icslc ] ) . the 2 - 10 kev lightcurve of the x - ray afterglow of grb 990123 that was measured with bepposax @xcite for @xmath153 days ( not shown here ) was best fit by the cb model with @xmath154 , implying @xmath155 , consistent with @xmath156 that was inferred by @xcite from their data . the observed temporal power - law decay index of the late - time x - ray afterglow , @xmath157 @xcite , also obeys the cb model relation , @xmath158 in fig . [ fig9]b we compare the observations of the optical lightcurve of grb 990123 from onset @xcite until late time @xcite , normalized to the @xmath137-band , and its cb model description as given by eq . ( [ fnu ] ) with the afterglow parameters @xmath159 , @xmath160 s and @xmath161 due to a gap in the data between 500 s and 15,000 s , the expected transition from a circumstellar density profile @xmath162 to a constant ism density was not well determined . however , the gradual bending ( ` jet break ' ) of the optical ag to an asymptotic power - law decay , @xmath163 is well reproduced with the expected late - time spectral index @xmath164 in the cb model , grbs , xrfs and shbs and their afterglows are produced by the interaction of bipolar jets of highly relativistic plasmoids ( cannonballs ) of ordinary matter , which are ejected in mass accretion episodes on a newly formed compact stellar object , with the radiation and matter that they encounter along their path . as observed in microquasars , each time part of the accretion disk falls abruptly onto the compact object , two jets of cannonballs ( cbs ) made of _ ordinary - matter plasma _ with large bulk - motion lorentz factors are emitted in opposite directions along the rotation axis , wherefrom matter has already fallen back onto the compact object due to lack of rotational support . the prompt @xmath1-ray and x - ray emission is dominated by inverse compton scattering ( ics ) of photons of the glory - a quasi isotropic optical light emitted by the supernova and scattered by the wind / ejecta blown from the progenitor star long before the grb . the cbs electrons compton up - scatter the glory photons into a narrow conical beam of @xmath1 rays along the cbs direction of motion . an x - ray ` flare ' coincident in time with a prompt @xmath1-ray pulse is simply its low - energy part . the early - time x - ray flares without a detectable accompanying @xmath1-ray emission are usually icfs produced by cbs with relatively smaller lorentz factors due to a weakening activity of the central engine : as the accretion material is consumed , the ` engine ' has a few progressively - weakening dying pangs . the lightcurves of icfs , like those of the prompt emission pulses , exhibit a rapid softening during their fast decline phase . roughly , the lightcurves are a function of the product @xmath21 and not of the individual values of the photon energy @xmath165 and the time @xmath7 after the beginning of the flare . the peak energy , isotropic equivalent energy and peak luminosity of the icfs are correlated like those of the prompt grb pulses . in the cb model , each icf is followed by srf from the encounter of the cb with the wind / ejecta that was blown from the progenitor star long before the grb . because of time - aberration , in the observer frame , these srfs lag after their preceding icfs by a short time of the order of the ics pulse duration . optical flares are usually much wider than their corresponding gamma / x - ray pulses and overlap , which makes it difficult to associate the early time optical flares with their preceding gamma / x - ray pulses / flares and measure their lag - time . only in single - pulse grbs that are bright enough to be detected with robotic telescopes and/or swift uvo and in very bright grbs , such as 080319b and 071031 , where the optical flares were partially resolved with robotic telescopes , could the predicted association between early time optical flares following gamma / x - ray pulses / flares be tested . often the fast decay of an x - ray icf is taken over by sr of x - rays from the cb encounter with the wind enclosing the glory light before it disappears under the plateau / shallow decay phase of the ag ( see , e.g. , figs . 3a , b , c in @xcite ) . late - time flares are usually srfs produced by cb encounters with the bumpy boundary of the star formation region , or with density bumps within this region where the grb took place . their exact profile is not known apriori but a wind - like profile seems to be a good working hypothesis , unlike the empirical parametrizations ( such as band function @xcite , cut - off power - law , beuermann function @xcite , broken power - law , segmented power - law , etc . ) used in most of the published standard analyses of grb lightcurves , which have never been properly derived from underlying physical assumptions , the master formulae of the cb model were derived in fair approximations from its underlying physical assumptions . as shown in this paper , the lightcurves and spectral evolution of x - ray flares and optical flares in grbs , xrfs and shbs are well described by the master formulae of the cannonball model of grbs . so far , no new assumptions or modifications of these formulae were needed when applied to well sampled flares in the grb lightcurves . probably , in the future , when much more refined data and better sampled lightcurves of grbs and their ags will become available , the cb model with its current simplifying assumptions , which were introduced in order to avoid ` over parametrization ' and make it predictive and falsifyable , will have to be refined in order to reproduce such data with sufficient accuracy . llllc grb050502b & 674 & 0.73 & 0.96 & 0.945 ( + 0.077 , -0.100 ) + shb050724 & & & & 0.81 ( + 0.14 , -0.17 ) + grb050820a & 9545 & 1.19 & 1.10 & 0.966 ( + 0.051 , -0.050 ) + grb050916 & 2178 & 0.74 & 0.90 & + grb060206 & 9402 & 1.04 & 1.12 & 1.21 ( + 0.27 , -0.24 ) + grb060418 & 123 & 1.73 & 1.05 & 0.96 ( + 0.15 , -0.14 ) + grb060526 & 4828 & 1.35 & 1.02 & 0.931 ( + 0.086 , -0.084 ) + grb060729 & 32665 & 2.52 & 1.10 & 1.067 ( + 0.038 , -0.037 ) + grb060929 & 5383 & 1.27 & 0.91 & 1.22 ( + 0.25 , -0.25 ) + grb061007 & 40 & 0.15 & 1.08 & 1.018 ( + 0.087 , -0.083 ) + grb070704 & 5183 & 1.27 & 0.90 & 0.98 ( + 0.18 , -0.33 ) + grb071003 & 1214 & 1.94 & 1.08 & 0.984 ( + 0.107 , -0.059 ) + xrf071031 & 5451 & 2.56 & 0.74 & 0.86 ( + 0.14 , -0.15 ) + grb080319b & 86 & 0.14 & 1.08 & 1.03 ( + 0.064 , -0.063 ) + xrf080330 & 541 & 4.61 & 0.89 & 0.89 ( + 0.13 , -0.12 ) + grb080506 & 7801 & 1.76 & 0.91 & 0.990 ( + 0.122 , -0.077 ) + grb080607 & 2904 & 0.53 & 0.91 & 1.102 ( + 0.098 , -0.092 ) + grb080810 & 3452 & 0.63 & 1.25 & 1.156 ( + 0.099 , -0.089 ) + grb080906 & 5790 & 0.90 & 0.96 & 1.049 ( + 0.069 , -0.164 ) + grb081102 & 1891 & 0.47 & 0.92 & 0.921 ( + 0.079 , -0.114 ) + grb081203a & 436 & 0.91 & 1.15 & 1.096 ( + 0.089 , -0.081 ) + grb090102 & 348 & 0.80 & 0.95 & 0.858 ( + 0.078 , -0.076 ) + grb090417b & 1259 & 0.63 & 1.04 & 1.09 ( + 0.11 , -0.11 ) + grb090618 & 1540 & 1.10 & 1.04 & 1.008 ( + 0.047 , -0.046 ) + grb090621a & 1309 & 0.83 & 0.94 & 1.01 ( + 0.18 , -0.18 ) + grb090709a & 1098 & 0.51 & 1.08 & 1.081(+0.076 , -0.074 ) + grb090812 & 598 & 1.34 & 0.92 & 0.914 ( + 0.138 , -0.089 ) + grb091029 & 5753 & 2.32 & 1.01 & 1.197 ( + 0.077 , -0.070 ) + [ t1 ] lllllllc grb050502b & icf , srf , & 485.7 & 308.3 & 0.27 kev & 13750 & 28950 & 6013 s + & srf & 58950 & 18911 & 3350 s & & & + grb060418 & icf , icf & 60 & 12 & 0.13kev & 9.9 & 118 & 0.13kev + grb060206 & srf , srf & 1.19 & 32.2 & 235 s & 653 & 3427 & 3645 s + & srf & 609359 & 245909 & 1980979 s & & & + shb050724 & icf , icf & 1.58 & 114 & 1.08 kev & 469 & 1422 & 1.32 kev + & icf & 9757 & 49512 & 3.53 kev & & & + grb050916 & srf , icf & & & & 17836 & 1335 & 91 kev + & icf & 9757 & 49512 & 3.53 kev & & & + grb050916 & srf , icf & & & & 17836 & 1335 & 91 kev + grb060526 & srf , icf & & & & 233 & 42.5 & 0.15 kev + & icf , srf & 280 & 25.7 & 60.2 kev & 85 & 106 & 288 s + grb060929 & srf , icf & & & & 472 & 45 & 117 kev + grb071031 & icf , icf & 84.8 & 46.2 & 13.2 kev & 131 & 33.8 & 0.74 kev + & icf , icf & 181.7 & 30.5 & 4.3 kev & 228 & 40.6 & 14.63 kev + & icf , icf & 328 & 176 & 2.74 kev & 572 & 173.6 & 19,74 kev + & icf & 4351 & 1201 & 4.5 kev & & & + & srf , srf & 16743 & 5463 & 4.2 s & 33241 & ( 53238 ) & @xmath166 + grb070704 & srf , icf & ( 63 ) & ( 0.35 ) & ( 25 s ) & 472 & 45.22 & 117 kev + grb080506 & icf , srf & 130 & 36.7 & 14.4 kev & 211 & 43 & 171 s + & icf , icf & 433 & ( 41 ) & ( 159 kev ) & 684 s & ( 160 ) & ( 156 kev ) + grb080607 & ics , icf & 65.9 & 8.6 & 145 kev & 113.9 & ( 15.4 ) & ( 350 kev ) + & srf , srf & 222 & 332 & 0.6 s & 652 & 1018 & 335 s + grb081203a & srf & 79 & 10.8 & ( 0 ) & & & + grb080810 & icf , srf , & 74 & 16 & 25.5 kev & 98.6 & 4.6 & 34 s + & icf , srf & 196 & ( 14 ) & ( 203 kev ) & 265 & 208 & ( 7.2 s ) + grb080906 & srf , srf , & ( 49 ) & ( 1 ) & ( 47.7 s ) & 157 & 27.5 & 10.75 s + & srf , icf & 490 & 8.58 & 200 s & 836 & 78 & 145 kev + grb081102 & srf , icf & 63 & 11 & 42.7 s & 914 & 88 & 111 kev + grb090417b & srf , icf & 336.2 & 203 & 82.6 s & 1227.8 & 431 & 1.96 kev + grb090618 & icf , srf & 65.9 & 12.9 & & ( 0 ) & 884 & ( 0 ) + grb090621a & srf , icf & 150.6 & 2.6 s & 215 & 217.9 & 50.4 & 0.95 kev + grb090709a & icf , srf , & 70.3 & 20.61 & 2.38 kev & 75.5 & 14.3 & 116 s + & srf , icf & 165.8 & 110.2 & 149 & 324.6 & 37 & 187 s + grb091029 & icf , icf & 0 & 38.8 & 9.8 kev & 218.6 & 102.5 & 10.9 kev + & srf & 209360 & 245909 & & & & + [ t2 ] llllc -5.cm grb 080319b : & & & & + srf1 & 9.50 & 3.52 & 10.24 & 0.70 + srf2 & 32.37 & 3.52 & 5.83 & 0.70 + srf3 & 39.54 & 3.52 & 6.96 & 0.70 + grb 060206 : & & & & + srf1 & @xmath167 & 548 & @xmath166 & 0.50 + srf2 & 2526 & 1214 & 887 & 0.80 + srf3 & 4286 & 3113 & 31 & 0.87 + grb 071003 : & & & & + srf1 & 4.11 & @xmath168 & 16.7 & 0.79 + srf2 & 125 & 139 & 14.6 & 1.03 + srf3 & 441 & 348 & @xmath169 & 0.75 + grb 071031 : & & & & + srf1 & 90 & 97 & 1875 & 0.62 + srf2 & 540 & @xmath168 & 43 & 0.63 + srf3 & 2335 & 7900 & 158 & 0.63 + srf4 & 10318 & 13372 & 134 & 0.63 + grb 080330 : & & & & + srf1 & 0.29 & 32 & 946 & 0.67 + srf2 & 3.40 & 1716 & 235 & 0.67 + srf3 & 3416 & 1491 & 18337 & 0.67 + [ t3 ] llllc -2.cm 1 & -5.57 & 15.4 & 300 & @xmath170 + 2 & 19.42 & 4.28 & 1450 & @xmath171 + 3 & 29.88 & 0.87 & 500 & @xmath172 + 4 & 32.95 & 5.43 & 800 & @xmath173 + 5 & 43.67 & 4.99 & 500 & @xmath174 + 6 & 52.77 & 4.30 & 450 & @xmath175 + 7 & 61.09 & 4.10 & 450 & @xmath176 + 8 & 70.35 & 5.63 & 600 & @xmath177 + 9 & 85.50 & 1.86 & 200 & @xmath178 + [ t4 ] lllllc grb990123 & srf & 22 & @xmath179 & 1.67 & 0.50 + grb030418 & srf & 200 & 4097 & @xmath166 & 0.56 + grb050820a & srf & 109 & 479 & @xmath166 & + grb061007 & srf & 15.8 & 90.37 & @xmath166 & 0.60 + grb081203a & srf & 0 & 316 & 368 & 0.90 + grb090102 & srf & 1.65 & 0.13 & 2.16 & 0.53 + grb090618 & srf & 118 & @xmath167 & 61 & 0.51 + [ t5 ]

the recent reported 750 gev diphoton excess at the 13 tev lhc is explained in the framework of effective field theory assuming the diphoton resonance is a scalar ( pseudoscalar ) particle . it is found that the large production rate and the broad width of this resonance are hard to be simultaneously explained if only visible final states are considered . therefore an invisible decay channel to dark matter ( dm ) is strongly favored by the diphoton resonance with a broad width , given a large coupling of the new scalar to dm . we set constraints on the parameter space in this scenario using the results from lhc run 1 , dm relic density , and dm direct and indirect detection experiments . we find that the dm searches can exclude a large portion of the parameter regions accounting for the diphoton excess with a broad width .

1512.06787

a great success of lhc run 1 is the discovery of the 125 gev higgs boson in the standard model ( sm ) . with a higher collision energy of 13 tev , lhc run 2 is ideal for probing heavy exotic particles in new physics beyond the standard model ( bsm ) . recently , the cms and atlas collaborations have reported their results based on run 2 data with an integrated luminosity of @xmath0 @xcite . as the majority of run 2 searches have not detected exotic signatures , their results have been used to set limits on bsm models , most of which are stronger than those from run 1 searches . some anomalies found in run 1 , such as the @xmath1 diboson excess , have not yet been confirmed by the latest run 2 searches @xcite . however , some surprising results have been provided from diphoton resonance searches . based on a data set of @xmath2 , the atlas collaboration found an excess in the diphoton invariant mass distribution around 750 gev , with a local ( global ) significance of @xmath3 ( @xmath4 ) @xcite . a fit to data suggested that this excess could be due to a resonance with a width of about 45 gev . a similar excess at @xmath5 has also been reported by the cms collaboration based on @xmath6 data , with a lower local ( global ) significance of 2.6@xmath7 ( 1.2@xmath7 ) @xcite . both collaborations claimed that this excess was not excluded by run 1 data , because of the large uncertainties in this energy range . this excess may be just due to statistical fluctuation , as many other disappeared anomalies in high energy physics experiments . future lhc searches with more data are required to confirm its existence . undoubtedly , if it is true , it would become the beginning of an epoch of bsm physics . according to the landau - yang theorem @xcite , a particle decaying into two photons can not be of spin-1 , implying that this 750 gev resonance should be either a spin-0 or spin-2 particle . spin-2 tensor fields can be realized in bsm physics related to gravity . for instance , this particle may be an excitation of the graviton in the randall - sundrum model . however , this kind of interpretations suffer stringent constraints @xcite . many studies on the diphoton excess assuming a scalar or pseudoscalar resonance have appeared soon after the reports @xcite . this resonance , which is denoted as @xmath8 hereafter , can be realized as an axion or composite higgs boson in models with new strong dynamics , as well as an exotic higgs boson with weak coupling . the cross section of the @xmath9 signal is of @xmath10 . in general , higgs - photon couplings induced by top and @xmath11 loops are suppressed . therefore , heavy higgs bosons in ordinary two higgs doublet models ( including higgs bosons in the mssm ) could not account for this excess with such a large production rate @xcite . thus the excess may suggest the existence of new charged and colored particles coupled to @xmath8 . furthermore , the broad width of @xmath12 suggested by the atlas analysis is also a challenge for model building . on the other hand , the cms fit seems to favor a narrower width , but the signal significance is too low . nevertheless , current data with low statistics could not give a final conclusion . in this work , we attempt to provide an interpretation of the diphoton excess in the context of effective field theory . we find that in order to explain the diphoton production rate , large @xmath8 couplings to gluons and photons are required . considering the constraints from resonance searches of lhc run 1 , it is challenging to achieve a broad decay width , if only the visible decay final states , such as dijet , @xmath13 , @xmath14 , and @xmath15 , are taken into account . this implies that there may be some invisible decay channels . an appealing and interesting possibility is that @xmath8 can decay into dark matter ( dm ) particles , which compose @xmath16 of the universe energy but leave no energy deposit in the atlas and cms detectors . thus it is straightforward to assume a scenario where @xmath8 is a portal between dm and sm particles , i.e. , dm particles interact with sm particles through the @xmath17 scalar ( pseudoscalar ) particle . constraints on this scenario are investigated in this work . we consider the limits from the lhc monojet search @xcite , the dm relic density measurement @xcite , dm direct detection results from the lux experiment @xcite , and indirect detection results from the fermi - lat @xcite and hess @xcite @xmath18-ray observations , and the ams-02 measurement of the ratio of cosmic - ray antiproton and proton fluxes @xcite . the expected exclusion limit of xenon1 t @xcite is also used to show the sensitivity of future direct detection experiments . we find that these dm experiments set strong constraints on the model parameter space , and that most of the parameter regions accounting for the diphoton signal have been excluded or can be explored . this paper is organized as follows . in sec . [ sec : interpret ] , we provide an interpretation to the diphoton excess in the context of effective field theory and consider the constraints from lhc run 1 data . in sec . [ sec : dm ] , the constraints from dm relic density and direct / indirect detection experiments on the model parameter space are discussed . finally we give a conclusion in sec . [ sec : concl ] . as there is no measurement on the cp property of @xmath8 at this moment , @xmath8 can be either a cp - even scalar or cp - odd pseudoscalar . in principle , its production could be initiated by @xmath19 annihilation , @xmath20 fusion , and electroweak vector boson fusion . since lhc is a @xmath21 machine , @xmath20 fusion has higher luminosity than @xmath19 annihilation after considering parton distribution functions . @xmath8 production via @xmath20 fusion is most probably a loop process , like the sm higgs case , where the major contribution comes from the top quark loop . @xmath8 decay into diphoton can be mediated by loops of charged fermions and @xmath11 bosons . however , lhc run 1 data have given constraints on its direct couplings to sm particles @xcite , such as @xmath22 at 95% cl for @xmath23 . therefore , the contributions to @xmath24 from the top and @xmath11 loops are stringently constrained and unlikely to account for the diphoton excess with a cross section of @xmath25 . in order to increase the cross section , some extra vector - like charged fermions may be required to induce a large effective @xmath26 coupling @xcite . if these new fermions are heavy enough , they can be integrated out and the loop processes become contact interactions . consequently , the interactions between @xmath8 and gauge bosons can be described by some dimension-5 effective operators . assuming cp conservation , the effective operators can be given by @xmath27 for a cp - even @xmath8 , and @xmath28 for a cp - odd @xmath8 . @xmath29 , @xmath30 , and @xmath31 are parameters describing the effective couplings between @xmath8 and sm gauge fields of the @xmath32 , @xmath33 , and @xmath34 groups , respectively . @xmath35 is a cutoff energy scale set by uv theory , and will be taken as a typical value of @xmath36 for simplicity . in order to respect sm gauge symmetry , the lagrangians and are expressed by gauge eigenstates . in terms of the physical fields @xmath37 ( photon ) and @xmath38 , the effective interactions become @xmath39 where @xmath40 , @xmath41 , and @xmath42 with @xmath43 and @xmath44 . we can see that the @xmath45 and @xmath46 couplings generally accompany the @xmath47 coupling . under the narrow width approximation , the production cross section for @xmath48 can be estimated by @xcite @xmath49 where @xmath50 is the center - of - mass energy of the system and @xmath51 is the momentum of one of the initial particles . @xmath52 and @xmath53 are the polarization states of the initial particles . @xmath54 is the spin of the resonance . @xmath55 is the total decay width of the resonance . @xmath56 and @xmath57 are the partial widths for the resonance decaying into initial states and final states , respectively . for the process @xmath58 , the production cross section depends on a factor of @xmath59 . the partial widths of @xmath60 and @xmath61 are independent of the cp property of @xmath8 and given by @xmath62 as an illuminating case , we may set @xmath63 , i.e. , assume that the @xmath26 coupling only comes from the @xmath8 coupling to the @xmath32 gauge field . then the @xmath14 and @xmath64 partial widths can be estimated as @xmath65 compared with the 8 tev upper limits and , one can easily see that the diphoton excess signal would not be constrained by run 1 data in this case . the total decay width can be given by @xmath66 thus , for @xmath20 fusion production , @xmath67 is predominantly determined by a factor of @xmath68 as the diphoton production cross section through an @xmath69-channel @xmath8 is @xmath70 under the narrow width approximation , we can separate the cross section calculation into two parts , i.e. , 1-body production cross section and decay branching ratio . other channels can be dealt with in a similar way . we calculate the @xmath71 production cross section with the simulation code ` madgraph 5 ` @xcite , to which the effective lagrangian is added through ` feynrules ` @xcite . as @xmath13 fusion production would be important when the @xmath26 coupling is much larger than the @xmath72 coupling @xcite , we include both @xmath20 fusion and @xmath13 fusion to compute the production cross section , using the parton distribution function set ` nnpdf2.3 ` with qed corrections @xcite . in the rest of this section , we only consider the case of cp - even @xmath8 . the results for cp - odd @xmath8 are similar . we present the contours of @xmath73 at @xmath74 and @xmath55 for @xmath63 in the @xmath29-@xmath31 plane in the left panel of fig . [ fig : crosecvisdec ] . for large @xmath31 , the @xmath73 contours tend to be parallel with x - axis . this behavior can be easily understood from eq . . for @xmath75 , @xmath13 fusion dominates , and @xmath73 is also independent of @xmath31 . it can be seen that a wide region in the parameter space can satisfy the desired values , @xmath76 . however , in order to simultaneously obtain @xmath77 and @xmath78 , @xmath31 should be as large as @xmath79 , where the @xmath61 channel dominates in @xmath8 decays . 8 tev lhc diphoton resonance searches give the constraints @xcite @xmath80 and @xmath81 at 95% cl . they are also plotted in fig . [ fig : crosecvisdec ] with magenta solid lines , where the lower ( upper ) one corresponds to @xmath82 . it seems that the @xmath13 fusion dominant interpretation is incompatible with lhc run 1 data . in many uv models , the interaction between @xmath8 and gauge bosons are induced through loop diagrams containing new charged and colored vector - like fermions . hence the effective couplings @xmath83 and @xmath31 can be approximately expressed as @xmath84 where the @xmath85 functions denote loop factors , @xmath86 is a typical fermion mass , @xmath87 and @xmath88 are the flavor and color number of the new fermions . therefore , in order to have a large @xmath31 that is greater than @xmath89 , a very large @xmath87 is needed , which may not be reasonable . furthermore , a large @xmath31 coupling would induce a significant dijet resonance signal via @xmath90 , which could conflict with lhc run 1 data . the 8 tev dijet constraint is also plotted in the left panel of fig . [ fig : crosecvisdec ] , from which we can see that it excludes the possibility of simultaneously having @xmath77 and @xmath78 for @xmath63 . on the other hand , we can fixed @xmath91 and assume that the @xmath26 coupling solely comes from the @xmath8 coupling to the @xmath33 gauge field . the result is shown in the right panel of fig . [ fig : crosecvisdec ] . we find that the situation becomes worse , as the run 1 constraints from the @xmath92 , @xmath14 , and @xmath93 channels has already excluded the region for desired @xmath73 . at @xmath74 as functions of @xmath94 for different values of @xmath29 or @xmath31 assuming @xmath63 and @xmath95 in the case of cp - even @xmath8 . here @xmath55 has been set to be @xmath96 . the green band denotes the favored range @xmath76.,scaledwidth=45.0% ] therefore , we can conclude that the run 2 diphoton excess indicates that there should be a large branching ratio into invisible final states , whose most tempting candidate is the dm particle @xmath97 . for @xmath98 , the required branching ratio into dm particles is @xmath99 , with @xmath100 denoting the total width of visible channels . in fig . [ fig : invbranch ] we demonstrate @xmath101 as a function of @xmath94 for @xmath102 . we assume @xmath103 and plot two sets of lines with either @xmath29 or @xmath31 fixed . when @xmath29 ( @xmath31 ) is fixed , we can obtain the required @xmath55 by adjusting @xmath31 ( @xmath29 ) and @xmath104 and then derive the value of @xmath101 . it can be seen that the broad width @xmath102 can be accommodated for moderate values of @xmath105 when @xmath106 is dominant . at @xmath74 as functions of @xmath55 for different values of @xmath29 assuming @xmath63 and @xmath95 in the case of cp - even @xmath8 . here @xmath101 has been set to be @xmath107 . the shaded region is excluded by the 8 tev monojet search.,scaledwidth=45.0% ] the invisible channel is actually constrained by @xmath108 searches in run 1 , where @xmath109 denotes transverse missing energy . the 8 tev atlas monojet analysis @xcite with an integrated luminosity of @xmath110 is adopted to give this constraint . in order to take into account the acceptance and efficiency of the signal regions in the atlas analysis , we simulate the @xmath111 process with ` madgraph 5 ` , ` pythia 6 ` @xcite , and ` delphes ` @xcite and apply the same cut conditions in each signal region . it turns out that the signal region sr6 gives the most stringent constraint . thus we obtain a 95% cl upper limit on the @xmath112 cross section at @xmath23 : @xmath113 this 8 tev limit can be translated to an upper limit on @xmath114 at @xmath74 , which is @xmath115 when @xmath20 fusion is dominant . in fig . [ fig : monojet ] , we show @xmath116 at 13 tev as functions of the total decay width @xmath55 for different values of @xmath29 assuming @xmath63 . here @xmath101 has been set to be @xmath107 . we find that @xmath117 is basically proportional to @xmath55 , as @xmath118 is actually the dominant decay channel in these cases . besides , a smaller @xmath29 requires a larger @xmath119 to give a broad total width , but it is easier to be excluded by the monojet search . in the previous section , we have shown that a broad scalar resonance accounting for the diphoton excess should have a large branching ratio into invisible decay modes . an appealing assumption is that the invisible channel is into dm particles . in this scenario , the diphoton excess would have a strong correlation with dm phenomenology . in this section , we discuss constraints on the parameter space from dm searches . we consider three types of dm particles , i.e. , majorana fermion , real scalar , and real vector . they couple to @xmath8 , which serves as a dm portal to sm particles . below we study 4 simplified models with respect to cp conservation . in model m1 , we assume @xmath8 is cp - even and @xmath97 is a majorana fermion , and the lagrangian with interaction and mass terms is @xmath120 model m2 also contains a majorana fermion @xmath97 , but a cp - odd @xmath8 : @xmath121 in model s , we consider a real scalar @xmath97 with a cp - even @xmath8 : @xmath122 a real vector @xmath97 and a cp - even @xmath8 is assumed in model v : @xmath123 note that the coupling @xmath124 is dimensionless in models m1 and m2 , but has a mass dimension in models s and v. each model has 5 free parameters , @xmath29 , @xmath30 , @xmath31 , @xmath124 , and @xmath125 , as we adopt @xmath126 and @xmath95 without loss of generality . a @xmath127 symmetry is imposed to guarantee the stability of the dm particle . other possible interactions between @xmath97 and @xmath8 described by higher dimensional operators have been neglected . in the following analysis , we randomly scan the parameter space to investigate the regions where the diphoton excess can be well explained and study its implication for dm experiments . in the scan , we allow the free parameters varying in the ranges as @xmath128 we impose the requirement of @xmath129 . lhc run 1 bounds from @xmath14 , @xmath93 , @xmath64 , dijet , and monojet searches are taken into account . since the current statistics of the diphoton excess is quite low and could not allow a very precise measurement of @xmath55 , we adopt a wide range of @xmath130 , and would select some distinct points satisfying a broad resonance condition @xmath131 . to begin with , we attempt to find some parameter points that could provide a correct dm relic density . in these @xmath8-portal simplified models , dm particles can annihilate into a pair of gauge bosons ( see e.g. refs . @xcite ) , @xmath14 , @xmath64 , @xmath132 , @xmath93 , and @xmath20 , through the exchange of an @xmath69-channel @xmath8 and stay in the thermal equilibrium in the early universe . thus we assume dm particles are produced via the standard thermal freeze - out mechanism . the dm relic density measured by the planck experiment @xcite , @xmath133 , would set strong constraints on the parameter space of the models . in this analysis , we obtain the predicted dm relic density by numerically solving the boltzmann equation , as described in appendix [ app : anni ] , and require the relic density satisfies a loose criterion of @xmath134 . after imposing these conditions , we project the parameter points in the @xmath125-@xmath124 plane in fig . [ fig : dmgxmx ] . the notations of the parameter points in fig . [ fig : dmgxmx ] , as well as in figs . [ fig : dm2pho ] , [ fig : dmcont ] , and [ fig : dmsi ] below , have the following meaning : red circles represent the parameter points satisfying both a correct relic density and the broad resonance condition ; blue circles corresponds to a correct relic density but @xmath135 ; both purple and green crosses can not give a correct relic density ; purple crosses can satisfy the broad resonance condition , while green crosses lead to @xmath135 . from fig . [ fig : dmgxmx ] we can see that red circles favor large values of @xmath124 . this is because the broad resonance condition @xmath131 always requires a large invisible decay width , as discussed in sec . [ sec : interpret ] . for instance , if the invisible decay width is 45 gev in model m1 , @xmath124 should be larger than @xmath136 . a correct relic density can be achieved by a canonical value of annihilation cross section @xmath137 , which requires moderate values of @xmath138 ( @xmath139 ) . since @xmath140 would be constrained by the lhc bounds , a large @xmath124 is helpful for achieving a canonical annihilation cross section . however , if the resonance width condition is relaxed , a smaller @xmath124 may also provide a correct relic density as a result of the resonant annihilation effect . this is the case for many blue circle points corresponding to @xmath141 . for heavy @xmath97 with @xmath142 , the contribution from the annihilation process @xmath143 becomes important . in this region , a correct relic density requires a large @xmath124 to enhance this channel . since the @xmath144 decay is forbidden when @xmath145 , @xmath55 is quite narrow . this is the case for blue circles and green crosses in the heavy dm region . the distributions of parameter points in models m1 and m2 are similar , due to their similar kinematics at the lhc . however , the distributions in models s and v are very different . this is because model v is not a uv - complete model . the @xmath144 decay width would be significantly increased by the longitudinal polarization of @xmath97 for light dm . thus , the condition @xmath146 excludes a large portion of the parameter space for @xmath147 . for dm indirect searches , we consider the limits from the fermi - lat gamma - ray line spectrum observation and dwarf galaxy continuous spectrum observation , as well as the ams-02 @xmath148 ratio measurement . since dm annihilation in model m1 is velocity suppressed , it is irrelevant to indirect detection . therefore , we only consider the constraints for the rest models . firstly , we show the constraints from the gamma - ray line spectrum searches on the @xmath149 annihilation cross section @xmath150 in fig . [ fig : dm2pho ] . the corresponding gamma - ray signal is monochromatic at @xmath151 , usually considered as a `` smoking - gun '' signature for the discovery of the dm particle . since no significant line - spectrum photon signal was found , the fermi - lat collaboration set upper limits on @xmath150 up to @xmath152 based on 5.8 years of data @xcite . the fermi - lat limits at 95% cl from the regions r41 and r3 optimized for nfw profiles with @xmath153 and @xmath154 , respectively , are shown in fig . [ fig : dm2pho ] . the r41 limit is at the order of @xmath155 for @xmath156 , and becomes stricter for lighter dm . as the r3 limit is obtained using a denser dm profile around the galactic center , it is stricter than the r41 limit . however , it is not very reliable due to the indeterminate dm distribution in the galactic center region . for heavy dm with @xmath157 , 95% cl upper limits come from the hess observation of the central galactic halo region based on data with 112 hours effective time @xcite . from fig . [ fig : dm2pho ] , we can see that these searches have set stringent constraints for the parameter points satisfying the correct relic density and lhc bounds . the annihilation process @xmath158 could give rise to another kind of line spectrum signal at @xmath159 . published limits for this channel given by fermi - lat are based on their 2 years of data @xcite , and weaker than those from @xmath160 searches . dm annihilation channels into @xmath14 , @xmath93 , @xmath64 , and @xmath20 would produce photons with continuous energy distribution via final state radiation , hadronization , and decay processes . here we define an effective total cross section for the channels with continuous gamma - ray spectra as @xmath161 dwarf spheroidal satellite galaxies around the galaxy are ideal targets for probing such gamma - ray signals , since the corresponding astrophysical gamma - ray backgrounds are quite clean . as no signal has been detected , the fermi - lat collaboration set stringent constraints on the dm annihilation cross section from a combined analysis of 15 dwarf galaxies @xcite . this analysis is based on 6 years of data . for many benchmark points , the dominant contributions to the continuous gamma - ray spectrum are given by the @xmath20 final states . as the initial gamma - ray spectra induced by the @xmath20 , @xmath162 and light quark final states are similar , the corresponding upper limits from fermi - lat observations would also be similar . we show the 95% cl upper limit on @xmath163 in fig . [ fig : dmcont ] , as the typical continuous spectrum induced by @xmath162 would be analogous to the spectrum here . for pure @xmath93 and @xmath14 final states , the fermi - lat limits are weaker than that in the @xmath162 channel by a factor @xmath164 . however , these final states have small fractions in most cases , so we would not treat them separately here . since the @xmath14 , @xmath93 , @xmath64 , and @xmath20 channels would also produce antiprotons via hadronic decay and hadronization processes , an important constraint comes from the cosmic - ray antiproton measurement . the latest result of the @xmath148 ratio has been reported by the ams-02 collaboration @xcite . here we show the ams-02 antiproton limits on @xmath163 at 95% cl derived from ref . @xcite in fig . [ fig : dmcont ] . in that analysis , two cosmic - ray propagation models , namely the diffusion - convection ( dc ) and diffusion - reacceleration ( dr-2 ) , was adopted . the uncertainties from propagation processes was considered with great care . as can be seen , the constraints from the fermi - lat dwarf galaxy observations and the ams-02 @xmath148 ratio are comparable . compared with line - spectrum gamma - ray searches , lesser parameter points are excluded by these two searches . for models m2 and v , the constraints for @xmath165 are not very strong . for model s , the antiproton constraints can exclude some points with a correct relic density in the low @xmath125 region . finally , we discuss the constraints from dm direct detection . dm particles can interact with nuclei through the @xmath166 coupling induced by the @xmath72 coupling . since the scattering cross section in model m2 is spin - dependent and momentum suppressed , this model would not predict testable signals in direct detection experiments . in the rest models , dm - nucleus scattering is spin - independent ( si ) . the content of gluons in a nucleon is given by @xcite @xmath167 the dm - nucleon si scattering cross section for model m1 can be expressed as @xcite @xmath168 while that for models s and v is @xcite @xmath169 here for these three models , @xmath170 where @xmath171 are the nucleon form factors , whose values are adopted from ref . @xcite . in the calculation , the value of @xmath172 should be taken at the scale of @xmath173 @xcite . we obtain @xmath174 through rge running with an input value of @xmath175 @xcite . we demonstrate the si scattering cross section in fig . [ fig : dmsi ] . also shown are the current 90% cl upper limit from lux @xcite and the projected sensitivity of xenon1 t @xcite . a large portion of the parameter points with @xmath176 in model s have been excluded by the lux result . this is because these points basically correspond to large @xmath177 . in models m1 and v , the constraints would be much weaker . nevertheless , the parameter points with a correct relic density and a broad decay width can be further tested by xenon1 t . in model v , some points in the resonant annihilation region may remain undetectable in future direct detection experiments . finally , we show the viable parameters to explain the diphoton excess in the @xmath29-@xmath31 plane in fig . [ fig : k3k1 ] , with the color scale indicating @xmath55 . all the parameter points are required to satisfy the lhc constraints . the parameter points represented by black circles satisfy the correct relic density as well as all the dm direct and indirect constraints . for the fermi - lat line - spectrum gamma - ray constraints , we take the limit from the region r41 optimized for a nfw profile with @xmath178 . we can see that @xmath179 requires @xmath180 . as the @xmath181 production cross section depends on a factor of @xmath182 , the 8 tev lhc dijet search constrains @xmath31 below @xmath183 , which has been illustrated in fig . [ fig : crosecvisdec ] . for @xmath184 , @xmath31 is further constrained below @xmath185 due to the monojet search . just a few points with @xmath134 and @xmath179 in models m2 , s , and v evade the constraint from fermi line - spectrum search , which , however , can not constrain model m1 . note that if the stricter fermi limit based on the region r3 for a nfw profile with @xmath186 is adopted , there would be much less black circles . the recent 750 gev diphoton excess found in lhc run 2 data has stimulated great interests . if this signal is confirmed in future lhc searches , it will open a new era of new physics beyond the standard model . in this work we attempt to explain the data in the framework of effective field theory . we find that the spin-0 resonance at 750 gev with a broad width of @xmath187 requires quite large couplings to gluons and photons . imposing the constraints from the dijet , @xmath14 , @xmath64 , and @xmath15 resonance searches in lhc run 1 , we find that the resonance should have a significant branching ratio into an invisible decay mode , which has also been constrained by the monojet searches in run 1 . if the final states of the invisible decay mode are dm particles , the 750 gev scalar coupling to dm would also be constrained by dm detection experiments . thus we consider the requirement from dm relic density and the constraints from direct and indirect detection . we find that the fermi - lat line - spectrum gamma - ray searches provide strong constraints on the model parameter space . a large parameter region that can explain the diphoton excess has been excluded . the fermi - lat dwarf galaxy continuous spectrum gamma - ray observations and the ams-02 cosmic - ray measurement of the @xmath148 ratio can also constrain the parameter space , but their limits are weaker than those from the line - spectrum gamma - ray searches . direct detection experiments could constrain the scalar coupling to gluons and dm particles . the future xenon1 t experiment is expected to explore a large parameter region accounting for the diphoton excess . it is interesting to note that for majorana fermionic dm coupled to a cp - odd scalar resonance , the dm - nuclei elastic scattering cross section is momentum - suppressed and spin - dependent , and hence could not be constrained by direct detection experiments . however , such a scenario can be deeply explored in line - spectrum gamma - ray searches . in the case of majorana fermionic dm coupled to a cp - even scalar resonance , annihilation processes can not be observed due to velocity suppression , but dm - nuclei si scatterings could be probed in direct searches . therefore direct and indirect searches are complementary to each other . in summary , the lhc run 2 diphoton excess may reveal the tip of the new physics iceberg and may have close connection to dark matter in the universe . we note that the corresponding effective couplings of this resonance to sm gauge fields and dm may be quite large . it seems not trivial to explain such large couplings in a uv - complete model . it would be very interesting to further clarify the properties of the resonance and its potential coupling to dm particles by combining more upcoming lhc run 2 data , fermi - lat searches , ams-02 data , and xenon1 t searches . this work is supported by the nsfc under grant nos . 11475191 , 11135009 , 11475189 , 11175251 , and the 973 program of china under grant no . 2013cb837000 , and by the strategic priority research program `` the emergence of cosmological structures '' of the chinese academy of sciences under grant no . zhy is supported by the australian research council . this appendix lists the partial widths for @xmath8 decay channels , except for those have already been listed in the main text . in the case of a cp - even @xmath8 , we have the following partial width expressions : @xmath188 where @xmath189 and @xmath190 . in the case of a cp - odd @xmath8 , we can obtain @xmath191 below we write down the partial widths for @xmath192 in the four simplified models . in model m1 , @xmath193 in model m2 , @xmath194 in model s , @xmath195 in model v , @xmath196 by solving the boltzmann equation , dm relic density can be expressed as @xcite @xmath197 where @xmath198 with @xmath199 denoting the dm freeze - out temperature . @xmath200 is the effectively relativistic degrees of freedom at the freeze - out epoch . @xmath201 is the planck mass and @xmath202 is the present cmb temperature . @xmath203 and @xmath204 is the coefficients in the velocity expansion of annihilation cross section @xmath205 , including all open channels . in order to derive the predicted relic density in the four simplified models , we should firstly calculate the @xmath203 and @xmath204 coefficients in various annihilation channels . in model m1 , the leading contribution to @xmath206 is of @xmath207-wave and the @xmath203 coefficient in every channel vanishes . for @xmath208 , @xmath209}}.\ ] ] for @xmath210 , @xmath211}},\ ] ] where @xmath212 . for @xmath213 , @xmath214}}.\ ] ] for @xmath215 , @xmath216}},\ ] ] where @xmath217 . for @xmath218 , @xmath219}}.\ ] ] for @xmath220 , @xmath221 where @xmath222 . in model m2 , for @xmath208 , @xmath223 } } , \\ b & = & \frac{{k_{{\mathrm{aa}}}^2g_\chi ^2m_\chi ^4(m_\phi ^4 + m_\phi ^2\gamma _ \phi ^2 - 16m_\chi ^4)}}{{\pi { \lambda ^2}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2}}}.\end{aligned}\ ] ] for @xmath210 , @xmath224 } } , \\ b & = & \frac{{k_{{\mathrm{zz}}}^2g_\chi ^2m_\chi ^2{\rho _ z}}}{{2\pi { \lambda ^2}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times[2m_\chi ^2(m_\phi ^4 + m_\phi ^2\gamma _ \phi ^2 - 16m_\chi ^4 ) + m_z^2(m_\phi ^4 - 24m_\phi ^2m_\chi ^2 + m_\phi ^2\gamma _ \phi ^2 + 80m_\chi ^4)].\end{aligned}\ ] ] for @xmath213 , @xmath225 } } , \\ b & = & \frac{{k_{{\mathrm{az}}}^2g_\chi ^2{{(4m_\chi ^2 - m_z^2)}^2}}}{{64\pi { \lambda ^2}m_\chi ^2{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times[2m_\chi ^2(m_\phi ^4 + \gamma _ \phi ^2m_\phi ^2 - 16m_\chi ^4 ) + m_z^2(m_\phi ^4 - 12m_\phi ^2m_\chi ^2 + m_\phi ^2\gamma _ \phi ^2 + 32m_\chi ^4)].\end{aligned}\ ] ] for @xmath215 , @xmath226 } } , \\ b & = & \frac{{k_{\mathrm{2}}^2g_\chi ^2m_\chi ^2{\rho _ w}}}{{\pi { \lambda ^2}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times[2m_\chi ^2(m_\phi ^4 + m_\phi ^2\gamma _ \phi ^2 - 16m_\chi ^4 ) + m_w^2(m_\phi ^4 - 24m_\phi ^2m_\chi ^2 + m_\phi ^2\gamma _ \phi ^2 + 80m_\chi ^4)].\end{aligned}\ ] ] for @xmath218 , @xmath227 } } , \\ b & = & \frac{{8k_{\mathrm{3}}^2g_\chi ^2m_\chi ^4(m_\phi ^4 + m_\phi ^2\gamma _ \phi ^2 - 16m_\chi ^4)}}{{\pi { \lambda ^2}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2}}}.\end{aligned}\ ] ] for @xmath220 , @xmath228 in model s , for @xmath208 , @xmath229 } } , \nonumber\\ b & = & \frac{{4k_{{\mathrm{aa}}}^2g_\chi ^2m_\chi ^4(m_\phi ^2 - 4m_\chi ^2)}}{{\pi { \lambda ^2}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2}}}.\end{aligned}\ ] ] for @xmath210 , @xmath230 } } , \\ b & = & \frac{{k_{{\mathrm{zz}}}^2g_\chi ^2}}{{32\pi { \lambda ^2}m_\chi ^4{\rho _ z}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times [ 128m_\chi ^8(m_\phi ^2 - 4m_\chi ^2 ) + 8m_\chi ^4m_z^2(3m_\phi ^4 - 56m_\phi ^2m_\chi ^2 + 3m_\phi ^2\gamma _ \phi ^2 + 176m_\chi ^4 ) \nonumber\\ & & \quad - 4m_\chi ^2m_z^4(9m_\phi ^4 - 116m_\phi ^2m_\chi ^2 + 9m_\phi ^2\gamma _ \phi ^2 + 320m_\chi ^4 ) \nonumber\\ & & \quad + 3m_z^6(5m_\phi ^4 - 56m_\phi ^2m_\chi ^2 + 5m_\phi ^2\gamma _ \phi ^2 + 144m_\chi ^4)].\end{aligned}\ ] ] for @xmath213 , @xmath231 } } , \\ b & = & \frac{{k_{{\mathrm{az}}}^2g_\chi ^2{{(4m_\chi ^2 - m_z^2)}^2}}}{{256\pi { \lambda ^2}m_\chi ^4{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times[32m_\chi ^4(m_\phi ^2 - 4m_\chi ^2 ) + m_z^2(3m_\phi ^4 - 32m_\phi ^2m_\chi ^2 + 3m_\phi ^2\gamma _ \phi ^2 + 80m_\chi ^4)].\end{aligned}\ ] ] for @xmath215 , @xmath232 } } , \\ b & = & \frac{{k_{\mathrm{2}}^2g_\chi ^2}}{{16\pi { \lambda ^2}m_\chi ^4{\rho _ w}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times[128m_\chi ^8(m_\phi ^2 - 4m_\chi ^2 ) + 8m_\chi ^4m_w^2(3m_\phi ^4 - 56m_\phi ^2m_\chi ^2 + 3m_\phi ^2\gamma _ \phi ^2 + 176m_\chi ^4 ) \nonumber\\ & & \quad - 4m_\chi ^2m_w^4(9m_\phi ^4 - 116m_\phi ^2m_\chi ^2 + 9m_\phi ^2\gamma _ \phi ^2 + 320m_\chi ^4 ) \nonumber\\ & & \quad + 3m_w^6(5m_\phi ^4 - 56m_\phi ^2m_\chi ^2 + 5m_\phi ^2\gamma _ \phi ^2 + 144m_\chi ^4)].\end{aligned}\ ] ] for @xmath218 , @xmath233 } } , \\ b & = & \frac{{32k_{\mathrm{3}}^2g_\chi ^2m_\chi ^4(m_\phi ^2 - 4m_\chi ^2)}}{{\pi { \lambda ^2}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2}}}.\end{aligned}\ ] ] for @xmath220 , @xmath234 in model v , for @xmath208 , @xmath235 } } , \\ b & = & \frac{{2k_{{\mathrm{aa}}}^2g_\chi ^2m_\chi ^2(m_\phi ^4 - 2m_\phi ^2m_\chi ^2 + m_\phi ^2\gamma _ \phi ^2 - 8m_\chi ^4)}}{{9\pi { \lambda ^2}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2}}}.\quad\quad~\end{aligned}\ ] ] for @xmath210 , @xmath236 } } , \\ b & = & \frac{{k_{{\mathrm{zz}}}^2g_\chi ^2}}{{288\pi { \lambda ^2}m_\chi ^4{\rho _ z}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times [ 64m_\chi ^6(m_\phi ^4 - 2m_\phi ^2m_\chi ^2 + m_\phi ^2\gamma _ \phi ^2 - 8m_\chi ^4 ) - 8m_\chi ^4m_z^2(7m_\phi ^4 + 40m_\phi ^2m_\chi ^2 + 7m_\phi ^2\gamma _ \phi ^2 - 272m_\chi ^4 ) \nonumber\\ & & \quad - 4m_\chi ^2m_z^4(5m_\phi ^4 - 172m_\phi ^2m_\chi ^2 + 5m_\phi ^2\gamma _ \phi ^2 + 608m_\chi ^4 ) \nonumber\\ & & \quad + 3m_z^6(7m_\phi ^4 - 104m_\phi ^2m_\chi ^2 + 7m_\phi ^2\gamma _ \phi ^2 + 304m_\chi ^4)].\end{aligned}\ ] ] for @xmath213 , @xmath237 } } , \\ b & = & \frac{{k_{{\mathrm{az}}}^2g_\chi ^2{{(4m_\chi ^2 - m_z^2)}^2}}}{{2304\pi { \lambda ^2}m_\chi ^4{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times [ 16m_\chi ^2(m_\phi ^4 - 2m_\phi ^2m_\chi ^2 + m_\phi ^2\gamma _ \phi ^2 - 8m_\chi ^4 ) + m_z^2(5m_\phi ^4 - 64m_\phi ^2m_\chi ^2 + 5m_\phi ^2\gamma _ \phi ^2 + 176m_\chi ^4 ) ] . \nonumber\\*\end{aligned}\ ] ] for @xmath215 , @xmath238 } } , \\ b & = & \frac{{k_{\mathrm{2}}^2g_\chi ^2}}{{144\pi { \lambda ^2}m_\chi ^4{\rho _ w}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2 } } } \nonumber\\ & & \times [ 64m_\chi ^6(m_\phi ^4 - 2m_\phi ^2m_\chi ^2 + m_\phi ^2\gamma _ \phi ^2 - 8m_\chi ^4 ) - 8m_\chi ^4m_w^2(7m_\phi ^4 + 40m_\phi ^2m_\chi ^2 + 7m_\phi ^2\gamma _ \phi ^2 - 272m_\chi ^4 ) \nonumber\\ & & \quad - 4m_\chi ^2m_w^4(5m_\phi ^4 - 172m_\phi ^2m_\chi ^2 + 5m_\phi ^2\gamma _ \phi ^2 + 608m_\chi ^4 ) \nonumber\\ & & \quad + 3m_w^6(7m_\phi ^4 - 104m_\phi ^2m_\chi ^2 + 7m_\phi ^2\gamma _ \phi ^2 + 304m_\chi ^4)].\end{aligned}\ ] ] for @xmath218 , @xmath239 } } , \\ b & = & \frac{{16k_{\mathrm{3}}^2g_\chi ^2m_\chi ^2(m_\phi ^4 - 2m_\phi ^2m_\chi ^2 + m_\phi ^2\gamma _ \phi ^2 - 8m_\chi ^4)}}{{9\pi { \lambda ^2}{{[{{(m_\phi ^2 - 4m_\chi ^2)}^2 } + m_\phi ^2\gamma _ \phi ^2]}^2}}}. \quad\quad~\end{aligned}\ ] ] for @xmath220 , @xmath240 l. d. landau , dokl . akad . nauk ser . fiz . * 60 * , no . 2 , 207 ( 1948 ) . c. n. yang , phys . * 77 * , 242 ( 1950 ) . f. deramo , j. de vries and p. panci , arxiv:1601.01571 [ hep - ph ] . c. csaki , j. hubisz , s. lombardo and j. terning , arxiv:1601.00638 [ hep - ph ] . y. hamada , t. noumi , s. sun and g. shiu , arxiv:1512.08984 [ hep - ph ] . m. dhuria and g. goswami , arxiv:1512.06782 [ hep - ph ] . a. alves , a. g. dias and k. sinha , arxiv:1512.06091 [ hep - ph ] . y. bai , j. berger and r. lu , arxiv:1512.05779 [ hep - ph ] . a. falkowski , o. slone and t. volansky , arxiv:1512.05777 [ hep - ph ] . c. csaki , j. hubisz and j. terning , arxiv:1512.05776 [ hep - ph ] . j. chakrabortty , a. choudhury , p. ghosh , s. mondal and t. srivastava , arxiv:1512.05767 [ hep - ph ] . l. bian , n. chen , d. liu and j. shu , arxiv:1512.05759 [ hep - ph ] . d. curtin and c. b. verhaaren , arxiv:1512.05753 [ hep - ph ] . s. fichet , g. von gersdorff and c. royon , arxiv:1512.05751 [ hep - ph ] . w. chao , r. huo and j. h. yu , arxiv:1512.05738 [ hep - ph ] . s. v. demidov and d. s. gorbunov , arxiv:1512.05723 [ hep - ph ] . j. m. no , v. sanz and j. setford , arxiv:1512.05700 [ hep - ph ] . d. becirevic , e. bertuzzo , o. sumensari and r. z. funchal , arxiv:1512.05623 [ hep - ph ] . p. agrawal , j. fan , b. heidenreich , m. reece and m. strassler , arxiv:1512.05775 [ hep - ph ] . a. ahmed , b. m. dillon , b. grzadkowski , j. f. gunion and y. jiang , arxiv:1512.05771 [ hep - ph ] . p. cox , a. d. medina , t. s. ray and a. spray , arxiv:1512.05618 [ hep - ph ] . a. kobakhidze , f. wang , l. wu , j. m. yang and m. zhang , arxiv:1512.05585 [ hep - ph ] . q. h. cao , y. liu , k. p. xie , b. yan and d. m. zhang , arxiv:1512.05542 [ hep - ph ] . b. dutta , y. gao , t. ghosh , i. gogoladze and t. li , arxiv:1512.05439 [ hep - ph ] . c. petersson and r. torre , arxiv:1512.05333 [ hep - ph ] . b. bellazzini , r. franceschini , f. sala and j. serra , arxiv:1512.05330 [ hep - ph ] . m. low , a. tesi and l. t. wang , arxiv:1512.05328 [ hep - ph ] . s. d. mcdermott , p. meade and h. ramani , arxiv:1512.05326 [ hep - ph ] . e. molinaro , f. sannino and n. vignaroli , arxiv:1512.05334 [ hep - ph ] . r. s. gupta , s. jager , y. kats , g. perez and e. stamou , arxiv:1512.05332 [ hep - ph ] . j. ellis , s. a. r. ellis , j. quevillon , v. sanz and t. you , arxiv:1512.05327 [ hep - ph ] . d. buttazzo , a. greljo and d. marzocca , arxiv:1512.04929 [ hep - ph ] . s. knapen , t. melia , m. papucci and k. zurek , arxiv:1512.04928 [ hep - ph ] . r. franceschini _ et al . _ , arxiv:1512.04933 [ hep - ph ] . s. di chiara , l. marzola and m. raidal , arxiv:1512.04939 [ hep - ph ] . y. nakai , r. sato and k. tobioka , arxiv:1512.04924 [ hep - ph ] . a. angelescu , a. djouadi and g. moreau , arxiv:1512.04921 [ hep - ph ] . m. backovic , a. mariotti and d. redigolo , arxiv:1512.04917 [ hep - ph ] . y. mambrini , g. arcadi and a. djouadi , arxiv:1512.04913 [ hep - ph ] . k. harigaya and y. nomura , arxiv:1512.04850 [ hep - ph ] . g. aad _ et al . _ [ atlas collaboration ] , eur . j. c * 75 * , no . 7 , 299 ( 2015 ) [ eur . j. c * 75 * , no . 9 , 408 ( 2015 ) ] [ arxiv:1502.01518 [ hep - ex ] ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1502.01589 [ astro-ph.co ] . d. s. akerib _ et al . _ [ lux collaboration ] , phys . lett . * 112 * , 091303 ( 2014 ) [ arxiv:1310.8214 [ astro-ph.co ] ] . m. ackermann _ et al . _ [ fermi - lat collaboration ] , phys . rev . d * 86 * , 022002 ( 2012 ) [ arxiv:1205.2739 [ astro-ph.he ] ] . m. ackermann _ et al . _ [ fermi - lat collaboration ] , phys . lett . * 115 * , no . 23 , 231301 ( 2015 ) [ arxiv:1503.02641 [ astro-ph.he ] ] . m. ackermann _ et al . _ [ fermi - lat collaboration ] , phys . rev . d * 91 * , no . 12 , 122002 ( 2015 ) [ arxiv:1506.00013 [ astro-ph.he ] ] . a. abramowski _ et al . _ [ hess collaboration ] , phys . rev . lett . * 110 * , 041301 ( 2013 ) [ arxiv:1301.1173 [ astro-ph.he ] ] . ams-02 collaboration , talks at the ams days at cern , 15 - 17 april , 2015 . e. aprile _ et al . _ [ xenon collaboration ] , arxiv:1512.07501 [ physics.ins-det ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . lett . * 111 * , no . 21 , 211804 ( 2013 ) [ phys . * 112 * , no . 11 , 119903 ( 2014 ) ] [ arxiv:1309.2030 [ hep - ex ] ] . g. aad _ et al . _ [ atlas collaboration ] , arxiv:1507.05930 [ hep - ex ] . v. khachatryan _ et al . _ [ cms collaboration ] , jhep * 1510 * , 144 ( 2015 ) [ arxiv:1504.00936 [ hep - ex ] ] . g. aad _ et al . _ [ atlas collaboration ] , arxiv:1509.00389 [ hep - ex ] . g. aad _ et al . _ [ atlas collaboration ] , phys . d * 92 * , no . 3 , 032004 ( 2015 ) [ arxiv:1504.05511 [ hep - ex ] ] . cms collaboration [ cms collaboration ] , cms - pas - hig-14 - 006 . g. aad _ et al . _ [ atlas collaboration ] , phys . d * 91 * , no . 5 , 052007 ( 2015 ) [ arxiv:1407.1376 [ hep - ex ] ] . k. a. olive _ et al . _ [ particle data group collaboration ] , chin . phys . c * 38 * , 090001 ( 2014 ) . j. alwall _ et al . _ , jhep * 1407 * , 079 ( 2014 ) [ arxiv:1405.0301 [ hep - ph ] ] . a. alloul , n. d. christensen , c. degrande , c. duhr and b. fuks , comput . commun . * 185 * , 2250 ( 2014 ) [ arxiv:1310.1921 [ hep - ph ] ] . r. d. ball _ et al . _ [ nnpdf collaboration ] , nucl . b * 877 * , 290 ( 2013 ) [ arxiv:1308.0598 [ hep - ph ] ] . v. khachatryan _ et al . _ [ cms collaboration ] , phys . b * 750 * , 494 ( 2015 ) [ arxiv:1506.02301 [ hep - ex ] ] . t. sjostrand , s. mrenna and p. z. skands , jhep * 0605 * , 026 ( 2006 ) [ hep - ph/0603175 ] . j. de favereau _ et al . _ [ delphes 3 collaboration ] , jhep * 1402 * , 057 ( 2014 ) [ arxiv:1307.6346 [ hep - ex ] ] . x. chu , t. hambye , t. scarna and m. h. g. tytgat , phys . d * 86 * , 083521 ( 2012 ) [ arxiv:1206.2279 [ hep - ph ] ] . r. m. godbole , g. mendiratta and t. m. p. tait , jhep * 1508 * , 064 ( 2015 ) [ arxiv:1506.01408 [ hep - ph ] ] . s. j. lin , x. j. bi , p. f. yin and z. h. yu , arxiv:1504.07230 [ hep - ph ] . m. a. shifman , a. i. vainshtein and v. i. zakharov , phys . lett . b * 78 * , 443 ( 1978 ) . j. m. zheng , z. h. yu , j. w. shao , x. j. bi , z. li and h. h. zhang , nucl . phys . b * 854 * , 350 ( 2012 ) [ arxiv:1012.2022 [ hep - ph ] ] . z. h. yu , j. m. zheng , x. j. bi , z. li , d. x. yao and h. h. zhang , nucl . phys . b * 860 * , 115 ( 2012 ) [ arxiv:1112.6052 [ hep - ph ] ] . j. r. ellis , a. ferstl and k. a. olive , phys . b * 481 * , 304 ( 2000 ) [ hep - ph/0001005 ] . e. w. kolb and m. s. turner , front . phys . * 69 * , 1 ( 1990 ) . g. jungman , m. kamionkowski and k. griest , phys . rept . * 267 * , 195 ( 1996 ) [ hep - ph/9506380 ] .

we report on a computer simulation and integral equation study of a simple model of patchy spheres , each of whose surfaces is decorated with two opposite attractive caps , as a function of the fraction of covered attractive surface . the simple model explored the two - patch kern - frenkel model interpolates between a square - well and a hard - sphere potential on changing the coverage . we show that integral equation theory provides quantitative predictions in the entire explored region of temperatures and densities from the square - well limit down to . for smaller , good numerical convergence of the equations is achieved only at temperatures larger than the gas - liquid critical point , where however integral equation theory provides a complete description of the angular dependence . these results are contrasted with those for the one - patch case . we investigate the remaining region of coverage via numerical simulation and show how the gas - liquid critical point moves to smaller densities and temperatures on decreasing . below , crystallization prevents the possibility of observing the evolution of the line of critical points , providing the angular analog of the disappearance of the liquid as an equilibrium phase on decreasing the range for spherical potentials . finally , we show that the stable ordered phase evolves on decreasing from a three - dimensional crystal of interconnected planes to a two - dimensional independent - planes structure to a one - dimensional fluid of chains when the one - bond - per - patch limit is eventually reached .

1004.1064

spherically symmetric potentials have become a well - established paradigm of colloidal science in past decades . @xcite this is because , at a sufficiently coarse - grained level , colloidal surface composition can be regarded as uniform with a good degree of confidence , so that relevant interactions depend only on relative distances among the particles . recent advances in chemical particle synthesis @xcite have however challenged this view by emphasizing the fundamental role of surface colloidal heterogeneities and their detailed chemical compositions . this is particularly true for an important subclass of colloidal systems , namely proteins , where the presence of anisotropic interactions can not be neglected , even at the minimal level . @xcite directional interactions introduce novel properties in such systems . these properties depend both on the number of contacts ( _ i.e. _ , the valency ) and the amplitude of these interactions ( _ i.e. _ , flexibility of the bonds ) , a notable example of this class being hydrogen - bond interactions , ubiquitous in biological , chemical , and physical processes . @xcite as a reasonable compromise between the high complexity of interactions governing the above systems and the necessary simplicity required for a minimal model , patchy - sphere models stand out for their remarkable success in this rapidly evolving field . @xcite see ref . for a recent review on the subject . within this class of models , interactions are spread over a limited part of the surface , either concentrated over a number of pointlike spots @xcite or distributed over one or more extended regions . @xcite while the former have the considerable advantage of a simple theoretical scheme @xcite which allows a first semi - quantitative description , the latter can easily account for both the effect of the number of contacts and their amplitude , unlike `` spotty '' interactions which are always limited by the one - bond - per - site constraint . in this paper we consider a particular model due to kern and frenkel @xcite of this patchy - spheres class wherein short - range attractive interactions of the square - well ( sw ) form are distributed over circular patches on otherwise hard spheres ( hs ) . interactions between particles ( spheres ) are then attractive in the sw - sw interfacial geometry or purely hard - sphere repulsive under the hs - sw or hs - hs interfacial geometries , and can sustain more than one bond in fact , as many as the geometry allows even in the case of a single patch assigned to each sphere . a number of real systems ranging from surfactants to globular proteins can be described with simplified interactions of these particular forms , with well - defined solvophilic and solvophobic regions , and despite their simplicity patchy hard spheres have already shown a remarkable richness of theoretical predictions . @xcite notwithstanding the discontinous nature of the angular interactions , highly simplified integral equation approaches are possible , @xcite but only very recently has a complete well - defined scheme , within the framework of the reference hypernetted - chain ( rhnc ) integral equation , been proposed and solved for patchy spheres . @xcite this integral equation belongs to a class of approximate closures which have been extensively exploited in the field of molecular associating fluids . @xcite its main advantage over other available approximations ( other than its less - accurate parent hnc closure ) lies in the fact that it relies on a single approximation , for the bridge function appearing in the exact relation between pair potential and pair distribution function @xmath4 , @xcite to directly yield structural and thermodynamic properties that include the helmholtz free energy and the chemical potential with no further approximations . @xcite in addition , it can be made to display enhanced consistency among different thermodynamic routes . @xcite this is an important point when analyzing fluid - fluid phase diagrams such as we propose to do here . we thus build upon our previous work with the one - patch potential @xcite to study the two - patch case and its relationship with its one - patch counterpart . in addition to rhnc integral equation results , we provide dedicated monte carlo simulations which can assess the performance of rhnc . we find that rhnc provides a robust representation of both structural and thermophysical properties of the two - patch kern - frenkel model for a wide range of coverage @xmath0 ( the ratio between attractive and total hard - sphere surface ) , extending from an isotropic sw to a bare hs potential . the competition arising between phase separation and polymerisation is discussed in terms of the angular dependence of the pair correlation function and the structure factor . finally , a comparison between the one - patch and two - patch phase diagrams shows a strong impact on the different morphology and stable structures obtained in the two cases . we also report numerical simulation results of the model in the region where the rhnc integral equations do not numerically converge , to explore the low temperature , small @xmath0 limit . we find that for @xmath5 it becomes impossible to investigate the low - temperature disordered phases , since the system quickly transforms into an ordered structure , which itself depends on the coverage value . indeed , on decreasing @xmath0 one progressively enters the region where the maximum number of contacts per patch evolves from four to two and eventually reaches the one - bond - per - patch condition . when three or four bonds per patch are possible , the observed ordered structure is a crystal of interconnected planes , while when only two contacts are possible , particles order themselves into a set of disconnected planes . the patchy interaction model examined here can be regarded as a prototype of a special colloidal architecture where there exist competitive interactions on the colloidal surface that drive , by free energy minimization , the different colloidal particles through a spontaneous self - assembly process into complex superstructures whose final target can be experimentally probed and properly tuned . @xcite the possibility , discussed in the present study , of identifying the position of the gas - liquid coexisting lines and its relative interplay with different structures , opens up fascinating scenarios in material science , on the possibility of novel material design exploiting a bottom - up process not requiring human intervention . as a paradigmatic model for highly anisotropic interactions , we take the kern - frenkel @xcite two - patch model where two attractive patches are symmetrically arranged as polar caps on a hard sphere of diameter @xmath6 . each patch can be reckoned as the intersection of a spherical shell with a cone of semi - amplitude @xmath7 and vertex at the center of the sphere . consider spheres @xmath8 and @xmath9 and let @xmath10 be the direction joining the two sphere centers , pointing from sphere @xmath8 to sphere @xmath9 ( see fig . [ fig : fig1 ] ) . the orientation of sphere @xmath11 is defined by a unit vector @xmath12 passing outward through the center of one of its patches , to be arbitrarily designated as the `` top '' @xmath13 patch . the patch on the opposite , `` bottom '' @xmath14 pole is then identified with the outward normal @xmath15 . two spheres attract via a square - well potential of range @xmath16 and depth @xmath17 if any combination of the two patches on each sphere are within a solid angle defined by @xmath7 and otherwise repel each other as hard spheres . the pair potential then reads @xcite @xmath18 where @xmath19 and @xmath20 where @xmath21 or @xmath22 indicates which patch , top or bottom , is involved on each sphere . the unit vectors @xmath23 are defined by the spherical angles @xmath24 in an arbitrarily oriented coordinate frame and @xmath25 is identified by the spherical angle @xmath26 in the same frame . reduced units , temperature @xmath27 and density @xmath28 , will be used throughout . this model was introduced by kern and frenkel , @xcite patterned after a similar model studied by chapman _ et al . _ , @xcite as a minimal model where both the distributions and the sizes of attractive surface regions on particles can be tuned . in this sense , the model constitutes a useful paradigm lying between spherically symmetric models which do not capture the specificity of surface groups , not even at the simplest possible level , and models with highly localized interactions having the single - bond , single - site limitation . @xcite several previous studies have already examined potentials of the kern - frenkel form using numerical simulations , @xcite corresponding - state arguments , @xcite highly simplified integral equation theories , @xcite and pertubation theories . @xcite more recently , @xcite the single patch kern - frenkel potential was studied using a more sophisticated integral equation approach based on the rhnc approximation coupled with rather precise and extensive monte carlo simulations . in the present paper , we extend this last study to the two - patch kern - frenkel potential and provide new methodologies specific for the angular distribution analysis . we define the coverage @xmath0 as the fraction of the total sphere surface covered by attractive patches . thus @xmath29 corresponds to a fully symmetric square - well potential while @xmath30 corresponds to a hard - sphere interaction and the model smoothly interpolates between these two extremes in the intermediate cases @xmath31 . the two - patch potential is expected to present qualitative as well as quantitative differences with respect to its one - patch counterpart . one interesting question , for instance , concerns the subtle interplay between distribution and size of the attractive patches on the fluid - fluid phase separation diagram . it is now well established @xcite that as coverage decreases the fluid - fluid coexistence line progressively diminishes in width and height . indeed , this feature can be exploited to suppress phase separation altogether to enhance the possibility of studying glassy behavior @xcite and can not be accounted for with a simple temperature and density rescaling , @xcite although corresponding - state type of arguments can be proposed . @xcite on the other hand , the above mechanism can significantly depend on how the same reduced attractive region is distributed on the surface of particles . in ref . , for instance , it was suggested that lines of decreasing critical temperature as a function of decreasing coverage , for the one - patch and two - patch kern - frenkel models with very short - range interactions , could cross each other at a specific coverage : for low coverages , critical temperatures for the one - patch model lie above the two - patch counterpart whereas the opposite is true for larger coverages . this would have far - reaching consequences on the phase diagram , as phase separation would occur at higher or lower temperatures for fixed coverage , depending on the specific allotment of the coverage . another interesting issue regards micellization phenomena , present in the one - patch version of the model , @xcite which is expected to be replaced by polymerization ( or chaining ) in the two - patch version . @xcite the ornstein - zernike ( oz ) equation @xcite defines the direct correlation function @xmath32 in terms of the pair correlation function @xmath33 ; it is convenient for computation to write it using the indirect correlation function @xmath34 instead of @xmath35 . we have then @xmath36 c\left(32\right).\end{aligned}\ ] ] a second , or `` closure , '' equation coupling @xmath37 and @xmath32 is needed . the general form for this is @xcite @xmath38 - 1-\gamma\left(12\right),\end{aligned}\ ] ] where @xmath39 and a third pair function , the so - called `` bridge '' function @xmath40 , has also been introduced . while known in a formal sense as a power series in density , @xcite @xmath40 can not in fact be evaluated exactly and at this point an approximation is unavoidable . the rhnc approximation replaces the unknown @xmath40 with a known version @xmath41 from some `` reference '' system . in practice , only the hard - sphere model is today well - enough known to play the role of reference system . here we will use the verlet - weis - henderson - grundke parametrization @xcite for @xmath42 , where @xmath43 is the reference hard - sphere diameter . some computational details of the rhnc integral equation approach can be found in ref . ( see expecially appendix a ) , so only the most relevant equations will be repeated here . solution of the ornstein - zernike integral equation for molecular fluids @xcite seemingly requires expansions in spherical harmonics of the angular dependence of all pair functions , a need that would be very problematic in the case of the discontinuous angular dependence in the present @xmath44 . in fact , the integral equation algorithm allows @xmath44 to remain unexpanded . @xcite there is a potential problem however in evaluating the gauss - legendre quadratures used in the numerical solution , in that of the angles @xmath45 used for an @xmath46th - order quadrature , none is likely to coincide with the angle @xmath7 defining the semi - amplitude of a patch . thus the algorithm will not `` know '' the correct patch size . this problem is ameliorated in the following _ ad hoc _ fashion . from the interaction @xmath44 of eq . ( [ ie : eq3 ] ) , the total coverage @xmath0 can be computed in terms of @xmath7 as @xmath47,\end{aligned}\ ] ] where @xmath48 is the heaviside step function , equal to @xmath8 if @xmath49 and @xmath50 if @xmath51 . the integrals can be readily evaluated to give @xcite @xmath52 this quantity can also be numerically evaluated by gauss - legendre quadrature using the @xmath46 roots @xmath53 of the legendre polynomial @xmath54 and the computed result compared with the exact value ( [ rhnc : eq5 ] ) . we may then vary @xmath46 so as to find that number @xmath46 ( typically kept between 30 and 40 ) that minimizes the known error in computing @xmath0 . all gaussian quadratures for that @xmath0 value are then evaluated with the same number @xmath46 of points , thus ensuring that minimal error arises from the selected angular grid . in an axial * r * frame @xcite with @xmath55 , the internal energy per particle in units @xmath56 is obtained from @xmath57 where @xmath58 denotes an average over spherical angle @xmath59 and where we have written out @xmath60 . similarly , the pressure @xmath61 is computed from the compressibility factor @xmath62 \right \rangle_{\omega_1 \omega_2 } \right \},\end{aligned}\ ] ] where the cavity function @xmath63 has been introduced . the angular integrations in these expressions are evaluated with gauss - legendre and gauss - chebyshev quadratures . finally , the dimensionless free energy per particle @xmath64 and chemical potential @xmath65 can also be directly computed from the pair functions produced by the rhnc equation ; the overall calculation is optimized by choosing the reference hard sphere diameter @xmath43 so as to minimize the free energy functional . @xcite we solve the rhnc equations numerically on @xmath66 and @xmath67 grids of @xmath68 points , with intervals @xmath69 and @xmath70 , using a standard picard iteration method . @xcite the square - well width is set at @xmath71 as a reasonable value dictated by the availability of isotropic square well results . @xcite further details of these and other computations can be found in ref . . for an assessment of the performance of the rhnc integral equation , we also perform nvt , grand canonical , and gibbs ensemble monte carlo ( mc ) simulations @xcite following the path set in the one - patch case . @xcite standard nvt mc simulations of a system of 1000 particles are used to compute structural information ( pair correlation functions and structure factors ) for comparison with integral equations results , whereas grand canonical and gibbs ensemble mc ( gemc ) are used to locate critical parameters and coexisting phases . the exact locations of the critical points ( points connected by the thick dashed green line in fig . [ fig : fig2 ] ) have been obtained from the mc data assuming the ising universality class and properly matching the density fluctuations with the known fluctuations of the magnetization close to the ising critical point . @xcite for gemc , we use a system of 1200 particles , which partition themselves into two boxes whose total volume is @xmath72 , corresponding to an average density of @xmath73 . at the lowest temperature considered , this corresponds to roughly @xmath74 particles in the liquid box and 150 particles in the gas box ( of side @xmath75 ) . on average , the code attempts one volume change every five particle - swap moves and 500 displacement moves . each displacement move is composed of a simultaneous random translation of the particle center ( uniformly distributed between @xmath76 ) and a rotation ( with an angle uniformly distributed between @xmath77 radians ) around a random axis . we have studied systems of size @xmath78 up to @xmath79 to estimate the size dependence of the critical point , with an average of one insertion / deletion step every 500 displacement steps in the case of grand canonical monte carlo ( gcmc ) . we have also performed a set of gcmc simulations for different choices of @xmath80 and @xmath81 to evaluate @xmath82 . see ref . and additional references therein for details . locating coexistence lines is not an easy task within integral equation theory , given the fact that virtually all integral equations are unable to access the critical region with reliable precision due to significant thermodynamic inconsistencies among various possible routes to thermodynamics , a consequence of the approximation buried in the closure eq . ( [ rhnc : eq2 ] ) . the rhnc closure is no exception to this rule , but has the strong advantage of relying on a single approximation expressed by the choice of the reference bridge function @xmath41 , at odds with other available closures which require additional approximations in constructing various thermodynamic quantities such as the chemical potential . here we follow the protocol outlined in ref . for the isotropic square - well potential and ref . for the one - patch kern - frenkel potential , where both the well - known pseudo - solutions shortcoming @xcite and the numerical drawbacks @xcite can be conveniently accounted for . figure [ fig : fig2 ] depicts the location of the fluid - fluid coexistence line for the two - patch case upon varying the coverage @xmath0 . the limiting case @xmath29 corresponds to the square - well potential . both mc ( points ) and rhnc ( thick solid lines ) results are shown . as previously noted , rhnc is not able to approach the critical point close enough to provide a direct estimate of its location . however , since it provides a quite good description of the low - temperature part of the coexistence line , we have tried to use these data to approximately locate the critical point . visual inspection of the rhnc coexistence points reveals , in the cases where it is possible to go closer to the critical region , an unphysical change of curvature of the coexistence line moving from low to high temperature . for this reason , for each coverage we selected only data clearly consistent with a rectilinear diameter law . then we fitted those data with the following function ( corresponding to the first correction to the scaling @xcite ) : @xmath83 where the values of the exponents @xmath84 and @xmath85 are appropriate for the 3d ising model universality class . @xcite once @xmath86 and the amplitudes @xmath87 , @xmath22 have been determined , the critical density can be obtained from the rectilinear diameter best fit . the numerical results of such a procedure are compared with mc estimates of the critical points in table [ tab : tab1 ] . it is evident that even though in general the validity of eq . ( [ correction ] ) is deemed to be limited to a smaller neighborhood of the critical point , @xcite in the present case it provides an acceptable procedure for a quick first estimate of the critical point location . as the coverage decreases , the coexistence line shrinks and moves to lower temperature and density , as expected from an overall - decreasing attractive interaction . this trend can be tracked rather precisely by mc simulations down to remarkably low coverages ( @xmath88 ) and rhnc correctly reproduces this evolution down to @xmath89 coverage . below this value , more powerful algorithms are required to achieve good numerical convergence . a few remarks are here in order . the coexistence curves shown in fig . [ fig : fig2 ] are consistent with previous analogous results reported in ref . but extend the range of temperatures and , more importantly , the range of coverages ( @xmath0 values ) . this allows a quantitative measure of the significant deviation from the simple mean - field - like results which can be obtained from the simple scaling ( not shown ) @xmath90 as suggested by the second - virial coefficient @xmath91 for this model , @xcite @xmath92 @xmath93 being the hard - sphere result . this is also consistent with the breakdown of the above simple scaling at the level of the third virial coefficient , derived in ref . for the companion patchy sticky - hard - sphere model . as we shall see later on , the dependence of the critical temperature and density on _ both _ the coverage _ and _ the number of patches is one of the main results of the present work . as a final point , we note that all curves in fig . [ fig : fig2 ] collapse into a single master curve upon scaling @xmath94 , in agreement with ref . . below @xmath88 , it becomes impossible to properly estimate the location of the critical point or the density of the coexisting gas and liquid phases . indeed , the gas - liquid separation becomes pre - empted by crystallization into a structure that depends on the value of @xmath0 . hence , the liquid phase , as an equilibrium phase , ceases to exist for small @xmath0 . this is strongly reminiscent of the disappearence of the liquid phase using spherical potentials when the range of the interaction becomes smaller than about 10% of the particle diameter , @xcite thus providing the angular analog of the same phenomenon . interestingly enough , the crystal structure which is spontaneously observed during the simulation depends on the value of @xmath0 , since the @xmath0 value controls the maximum number of bonds per patch . in the range of @xmath0 values such that each patch can be involved in four bonds , the observed ordered structure is made by planes exposing the sw parts to their surfaces ( see fig . [ fig : fig3](d ) ) . particles in the plane are located on a square lattice and adjacent planes are shifted in each direction by a half lattice constant , resulting in a reduced energy of -4 per particle ( _ i.e. _ , eight bonded neighbors ) . on decreasing @xmath0 below @xmath95 , the region where only three bonds per particle are possible ( @xmath96^{-1 } < \sin\theta_0 < [ \sqrt{2}(1+\delta/\sigma)]^{-1}$ ] ) is entered and the crystal structure is made by interconnected planes of particles arranged in a triangular lattice ( see fig . [ fig : fig3](c ) ) . for @xmath97^{-1 } < \sin\theta_0 < [ \sqrt{3}(1+\delta/\sigma)]^{-1}$ ] only two bonds per patch are possible and the system organizes into independent planes ( see fig . [ fig : fig3](b ) ) , this time turning a hs surface to their neighboring planes . particles in the plane are now arranged on a triangular lattice and each patch is able to bind only to two different neighbors located in the same plane , resulting in a reduced energy per particle of @xmath98 . when @xmath99 becomes smaller than the value @xmath100^{-1}$ ] ( corresponding to @xmath101 ) , the one - bond - per - patch condition is reached and the system can form only isolated chains ( see fig . [ fig : fig3](a ) ) . in this limit , the system is expected to behave as the two single - bond - per - patch model . we turn our attention next to structural information , where the advantages of a reliable integral equation approach become evident . one has to keep in mind that gibbs ensemble and gcmc simulations are particularly painstaking , due to the combined effect of the required low temperatures and the aggregation properties of the fluid ( as detailed below ) , so that many of the state points examined here require several weeks of computer time . on the other hand , the rhnc integral equation , while rather demanding from an algorithmic point of view ( see _ e.g. _ appendix a of ref . ) is a rapidly convergent scheme yielding solutions on the order of minutes , depending on the temperatures considered . a more profound advantage stems from the fact that , within the approximation defined by the rhnc closure , all possible pair structural information is in fact exactly available , unlike mc calculations where , though available in principle , their statistics would be so limited as to make such calculations impractical . thus , only the pair correlation function @xmath4 averaged over angle @xmath25 is computed . by symmetry , the resulting pair function in this context depends only on @xmath102 and @xmath103 and will be denoted here as @xmath104 ; see appendix b in ref . for details . consider then the unaveraged pair correlation function @xmath60 from rhnc in an axial frame with @xmath55 . two noteworthy configurations occur when ( a ) all four patches lie along the same line ( we denote this as the parallel @xmath105 configuration , with @xmath106 , the actual labeling of each patch being unimportant ) and when ( b ) patches on sphere @xmath9 lie on an axis perpendicular to those of sphere @xmath8 ( in this case @xmath107 , which we denote as the crossed ( x ) configuration ) . figure [ fig : fig4 ] reports the results for the case @xmath71 , @xmath108 , and @xmath109 , which has been selected so that the fluid is above the coexistence line for all considered coverages , with configurations @xmath110 and x in the top and bottom panels respectively . values of coverages range from a full square - well potential ( @xmath29 ) to a hard - sphere potential ( @xmath30 ) . as coverage decreases , the contact value @xmath111 of the @xmath110 configuration has the unusual behavior of first a slight increase from @xmath29 to @xmath112 , followed by a more marked increase starting at @xmath113 up to the very small coverage @xmath114 limit which eventually backtracks to roughly the same value as at @xmath113 in the hard - sphere limit . at the opposite side of the well , @xmath115 , an even more erratic behavior is observed , with an increase in the range @xmath116 , then a decrease for @xmath117 , a new increase down to @xmath114 , and a final sudden decrease to the hard sphere value @xmath30 . a somewhat similar feature occurs in the x configuration where within the entire well region @xmath118 one observes a sudden decrease from @xmath29 to @xmath119 and a more gradual increase until reaching the highest value for the hs case . it is worth noting that in the x configuration there is no discontinuous jump at @xmath120 for any value @xmath121 . the reason for this has already been addressed in ref . for the one - patch case . outside the first shell , there is a very weak dependence on the coverage , with a slight shift in the location of the second peak from a value of @xmath122 at the sw @xmath29 to a value of @xmath123 for lower @xmath0 . we compare rhnc integral equation results with mc simulations in fig . [ fig : fig5 ] . as noted above , only the averaged pair function @xmath104 can be compared and this is done in the figure for different values of the coverage @xmath0 at the same state point ( @xmath124 and for the same @xmath110 and x configurations considered earlier . the good overall performance of rhnc in representing mc results is apparent as both contact values at the well edges and the jump discontinuities are very well reproduced . it is instructive to contrast these results with those of fig . [ fig : fig4 ] , as many of the abrupt changes appearing in the actual pair correlation function are smoothed out by the orientational average carried out here . for instance , the characteristic jump at @xmath120 of the @xmath110 configuration ( top panel ) progressively decreases as coverage is reduced and disappears in the hard - sphere limit . conversely , the jump is present also in the x configuration ( bottom panel ) unlike the corresponding case of the full @xmath4 . in addition , the strong increase of the @xmath110 configuration for low coverages is not present in this figure ; as remarked earlier , this level of detail in @xmath4 is one of the main advantages of an integral equation approach . an additional useful quantity to consider , in view of its direct experimental access through scattering experiments , @xcite is the structure factor , which will be denoted @xmath125 within our theoretical framework . @xcite this is also strongly related , via hankel transforms , to the radial distribution function @xmath126 , which is @xmath4 averaged over all orientations @xmath127 and @xmath128 of the patches and of the relative angular position @xmath26 . note that @xmath126 is also the simplest rotational invariant ( see ref . and appendix [ app : appb ] ) . the structure factor and the radial distribution function are reported in fig . [ fig : fig6 ] for two representative values of coverage , @xmath129 and @xmath130 , corresponding to almost fully attractive and almost fully repulsive limits . these values have been selected at the same state point previously considered ( @xmath109 , @xmath108 ) as having a very different behavior within the first shell @xmath131 . this high - density result is also contrasted with a low - density state point @xmath132 at the same temperature , a value which , in the temperature - density plane , lies symmetrically with respect to the coexistence curves in the single fluid phase for all coverages ( see fig . [ fig : fig2 ] ) . a few features are worth noting . for density @xmath108 there is a significant coverage dependence , where the contact value @xmath133 for @xmath130 coverage is larger than that for the corresponding @xmath129 case and , conversely , the jump present at the other extreme @xmath134 is much smaller in the former than in the latter case . a similar feature also occurs for the low - density state point @xmath132 . this results from an angular average of the results given in fig . [ fig : fig4 ] . likewise , there is a marked difference in the behavior of the structure factor @xmath125 for the high - density case @xmath108 , both in the height of the first peak ( related to the @xmath133 value ) and of the secondary peaks ( related to the behaviors of @xmath126 in the @xmath135 region and of the @xmath136 discontinuity ) . similarly , in the low - density branch @xmath132 , the large @xmath137 value for the @xmath129 coverage case is signaling the approach to a spinodal instability which is clearly not present in the corresponding @xmath130 coverage . one natural interpretation of the above results is the progressive rearrangement of the distribution within the first shell upon varying both the coverage and the density . to support this view , we consider the angular distribution within the first shell in the next subsection . the nonmonotonic dependence of @xmath4 in terms of the distance @xmath138 for decreasing coverage @xmath0 , as illustrated in fig . [ fig : fig4 ] , is rather intriguing and requires an explanation . a similar , albeit different , feature occurs even in the one - patch case , as shown in ref . . we have tackled this in two ways , illustrated in the following . all previous representations of @xmath4 have been depicted in the molecular axial frame , where @xmath139 , so that patches on sphere @xmath8 are parallel to the vector @xmath140 joining sphere @xmath8 with sphere @xmath9 . this is clearly preventing an understanding of the angular distribution of the patches around a given sphere @xmath8 , that is , as a function of @xmath25 . this is however a needless restriction , as one can start from the expression for @xmath4 in a general ( laboratory ) frame and study the dependence on the angle @xmath26 for fixed patch directions @xmath141 and @xmath142 . in figs . [ fig : fig4 ] , [ fig : fig5 ] and [ fig : fig6 ] , we notice that the main dependence on coverage @xmath0 stems from the region within the well , @xmath131 ; it sufficies therefore to investigate the _ average _ @xmath26 dependence by integrating over the radial variable @xmath66 within this region . we further note that there is azimuthal symmetry with respect to the @xmath143 variable , so that we can focus on the @xmath144 dependence . the details of the analysis are reported in appendix [ app : appa ] , where it is shown that the relevant quantity is @xmath145 , which is a function of the angle @xmath144 ( polar dependence of @xmath25 ) and of the polar angle @xmath146 of the patches on particle @xmath9 , given that the patches on particle @xmath8 lie along the @xmath147 axis . we report comparative calculations for both low ( @xmath132 ) and high ( @xmath108 ) density at identical temperature @xmath109 at two representative coverages , @xmath129 , representing a case with almost all attraction , and @xmath130 , as representative of an almost hard - sphere case . these are the same conditions considered in fig . [ fig : fig6 ] ; the results are reported in fig . [ fig : fig7 ] . let us consider first the high - density , @xmath108 , situation as depicted in the two left panels for @xmath129 ( a ) and @xmath130 ( c ) . here , the @xmath129 case yields a very well - defined pattern with a periodically modulated distribution of the patches in symmetrical fashion as indicated by the trimodal distribution as a function of the relative positional angle @xmath144 , so that @xmath148 are almost equally represented . ( note that @xmath149 are necessarily equivalent due to the up - down symmetry of the two - patch distribution . ) the two interstitial minima are a consequence of the reduced valency the corresponding fully symmetrical result under this condition would be a flat distribution around the value @xmath150 in between the two maxima and minima so this slightly favours perpendicular orientation of the patches along the forward ( or backward ) direction and parallel orientation along the transversal direction . under low coverage ( @xmath130 ) conditions , on the other hand , there is clear evidence of a parallel orientation of the patches along the forward ( or backward ) direction , the opposite being true for a perpendicular orientation of the patches . this confirms the tendency to filament formations previously alluded to . the situation is even more evident at low density , @xmath132 , as shown in the right two panels ( b ) and ( d ) . additional insights on the angular correlations of patch distributions can be obtained by considering other coefficients @xmath151 of the rotational invariants @xmath152 as defined in eqs . ( [ appa : eq1 ] ) and ( [ appa : eq2 ] ) ; they have proven to be of invaluable help in discriminating between parallel and antiparallel configurations occurring in different models such as dipolar hard spheres @xcite and heisenberg spin fluids . @xcite some relevant properties of these coefficients are also listed in appendix [ app : appb ] , where we display explicit expressions for the first few coefficients . in the two - patch case , we note that all coefficients with @xmath153 or @xmath154 odd vanish , so we depict the first nonvanishing coefficients @xmath155 , @xmath156 , @xmath157 in fig . [ fig : fig8 ] for the same state points as before . note that the left - most two curves , ( a ) and ( c ) , correspond to density @xmath108 , temperature @xmath109 , coverages @xmath129 ( a ) and @xmath130 ( c ) , and are plotted on the same scale . while qualitative trends are similar , the two cases have significantly different behavior . within a given coverange @xmath155 and @xmath156 are almost coincident , with positive correlation in the well region @xmath158 , whereas @xmath159 has decreasing positive correlation in the same region and negative correlation for @xmath160 . numerical values , on the other hand , differ among each other , with small values for @xmath155 and @xmath156 in the high coverage case @xmath129 and significantly higher values in the low coverage limit @xmath130 . likewise , for the low - density state point @xmath132 , @xmath109 , right - hand - side plots ( b ) and ( d ) can be unambiguously discriminated between high ( b ) and low ( d ) coverages . we shall return to this point in the comparison with the one - patch results , where the physical meaning of these coefficients will be discussed . figure [ fig : fig9 ] shows the critical parameters for the case of particles with one and two patches , both reported as a function of the total coverage . here only mc results are reported in view of their precision and reliability . the @xmath29 limit corresponds to the sw case and coincides for both models . an analogous figure , for the case of adhesive patchy spheres ( the limit of the present model for vanishing ranges ) , has been reported in ref . within a simplified integral equation scheme . with respect to the adhesive limit , the range of coverages which can be explored numerically is significantly wider ( for both one - patch and two - patch cases ) . in the case of two patches , crystallization pre - empts the possibility of exploring the smaller values . in the one - patch case , the process of micelle formation , also observed experimentally , @xcite suppresses the phase - separation process @xcite at small @xmath0 values . the critical parameters decrease on decreasing @xmath0 for both one - patch and two - patch models . the behavior of @xmath161 can be explained on the basis of a progressive reduction of the attractive surface . the decrease in the critical density becomes significantly pronounced only for the smallest @xmath0 values and can be attributed to the lower local density requested for extensive bonding . such behavior is analogous to the suppression of the critical density observed when the particle valence decreases . @xcite in the @xmath0 region where it is possible to evaluate the critical parameters , @xmath161 and @xmath162 for the two - patch case are always larger than the corresponding one - patch values , a trend which can be tentatively rationalized on the basis of the ability to form a larger number of contacts and higher local bonded densities for the case in which both poles of the particles can interact attractively with their neighbors . no evidence of a crossing between the two geometries is observed . such a crossing has been predicted by a theoretical approach based on a virial expansion up to third order in density and appropriate closures of the direct correlation function . @xcite it is worth emphasizing that the above dependence on the number of patches , at a given coverage , provides clear evidence of the impossibility of rationalizing the change of the critical line on the basis of a trivial decrease of the attractive strength of interactions due to the reduction in coverage , as alluded to in section [ subsec : coex ] . for the sake of completness , we also report rhnc integral equation results for the most relevant thermodynamic quantities , as a function of the coverage @xmath0 . these are shown in table [ tab : tab2 ] for the same high - density state , @xmath108 at @xmath109 , considered above for structural information . here we present the reduced internal energy per particle and the reduced excess free energy per particle , @xmath163 and @xmath164 , respectively , the reduced chemical potential @xmath165 , the compressibility factor @xmath166 , and the inverse compressibility @xmath167 . these results may be compared with those of table iv in ref . listing the same quantities for the one - patch counterpart . ( we ignore the tiny difference in densities between the two calculated states . ) the last two columns give the reference hs diameter @xmath43 stemming from the variational rhnc scheme ( see ref . for details ) and the average coordination number within the wells @xmath168 , whose one - patch counterparts are included in tables iv and v of ref . , respectively . note that @xmath168 here is systematically larger than in the one - patch case , in qualitative agreement with the mc results of fig . [ fig : fig9 ] . within the rhnc integral approach , the analysis of the angular distribution of patches within the first shell given in section [ subsec : angular_dist2p ] revealed that the cylindrical symmetry of a pair of patches ( 2p case ) on each particle was very effective in driving the system to morphologically different configurations in the low ( @xmath169 ) and high ( @xmath170 ) coverage limits , as illustrated in fig . [ fig : fig7 ] . it is natural to expect a very different situation in the single - patch case ( 1p case ) . this is indeed the case as further elaborated below . for the single patch with @xmath130 coverage ( fig . [ fig : fig10 ] , bottom panels ( c ) and ( d ) ) parallel patches ( @xmath171 ) are more likely in the perpendicular direction ( @xmath172 or @xmath173 ) , whereas antiparallel patches ( @xmath174 ) are more likely in the forward direction ( @xmath175 ) . the case of perpendicular patches ( @xmath176 ) are conversely more or less equally distributed along the whole angular region @xmath177 . there is no qualitative difference between the situation of high ( @xmath108 ) and low ( @xmath132 ) densities as shown by the contrast between the bottom left ( c ) and right ( d ) panels . note that the result significantly contrasts with the corresponding results of the two - patches case depicted in fig . [ fig : fig7 ] . consider now the opposite situation of very large coverage ( @xmath129 ) ( fig . [ fig : fig10 ] , ( a ) ) where there is a single well - defined peak for antiparallel orientations ( @xmath174 ) in the backward direction ( @xmath178 ) . again , this markedly differs from the two - patch case ( fig . [ fig : fig7 ] , top left panel ( a ) ) , where there is a triple peak for aligned patches ( @xmath179 ) in the forward ( @xmath180 ) , perpendicular ( @xmath173 ) , and backward ( @xmath178 ) orientations . this is a dense state point . under diluted conditions , @xmath132 , we find a qualitatively similar behavior as in the dense case , with antiparallel alignment in the forward direction ( which can not be physically distingushed from the backward one ) . clearly the predominant antiparallel alignment is reflecting the tendency to micellization rather than polymerization which is built into the single patch symmetry . it is also interesting to contrast the coefficients of rotational invariants for the one - patch case with those obtained in the two - patch counterpart in fig . [ fig : fig7 ] . at variance with the two - patch case , here all coefficients are nonvanishing so that we consider the first nonvanishing instances , that is @xmath181 , @xmath182 , and @xmath155 , which are particularly useful as giving the projections over the important invariants . @xcite we have evaluated these coefficients for the same state points considered in the two - patch case in fig . [ fig : fig11 ] for both dense or diluted conditions and small or large coverages . in contrast with the two - patch case , here the effect of coverage appears to be less significant , as can be inferred by inspection of the dense case @xmath108 ( left panels ( a ) and ( c ) ) . for the @xmath181 case the projection coefficient along the ferroelectic invariant @xmath183 in appendix [ app : appb ] we find a negative correlation within the well both for @xmath129 ( top left panel ( a ) ) and @xmath129 ( bottom left panel ( c ) ) , as expected from the tendency to form antiparallel alignments . likewise , the projection @xmath182 along the dipolar invariant @xmath184 is found to be negative and numerically similar to @xmath181 at both coverages , again indicating negative correlation to dipolar alignment . the only positive correlation is found for the @xmath155 component , which does not distinguish between up and down symmetry , in qualitative agreement with the two - patch analogue . this situation is replicated in the diluted case ( right two panels ( b ) and ( d ) ) with different numerical values , thus indicating that these correlations are signatures of robust orientational trends induced by the the particular one - side symmetry of the single - patch potential we have performed a detailed study of a fluid whose particles interact via a two - patch kern - frenkel potential that attributes a negative square - well energy whenever any two patches on the spheres are within a solid angle associated with a predefined coverage and within a given distance given by the well width , and a simple hard - sphere repulsion otherwise . this model can be reckoned as a paradigm of a unit system with incompatible elements ( _ e.g. _ , hydrophobic and hydrophilic ) that can self - assemble into different complex superstructures depending on the parameters of the original unit ( _ e.g. _ , coverage ) . we have exploited state - of - the - art numerical simulations ( standard metropolis , gibbs ensemble , and grand canonical monte carlo ) coupled with rhnc integral equation theory following the approach outlined in previous work on a single patch . @xcite on comparing rhnc integral equation with numerical simulations , we find the former to be quantitatively predictive in a large region of coverage , even close to the gas - liquid transition critical region , over a range of coverage which is significantly larger than the single patch counterpart . the reason for this is attributed to the fact that rhnc uses the approximated hard - sphere bridge function , which retains spherical information , as a unique approximation throughout the entire calculation , a feature which works better for the more symmetric two - patch case than the highly asymmetric one - patch kern - frenkel potential . having assessed the reliability of the rhnc integral - equation approach , we have fully exploited its capabilities in providing detailed angular information that is typically inaccessible to mc simulations , as already discussed in ref . . this has been done in two ways . first , by computing the orientational distribution probability of parallel and perpendicular alignment of patches within a spherical shell in the region @xmath135 . this methodology is able to account for the erratic coverage dependence of the pair correlation function @xmath4 by clearly discriminating between small and large coverages at all densities . the same approach also enlightens the characteristic symmetries of the patch distributions when the two - patch case result is contrasted with the one - patch analogue . second , by computing the rotational - invariant coefficients that are the projections of @xmath4 over rotational invariants . again , this can discriminate between small and large coverages ( at all densities ) and single and double patches . our monte carlo results extend those originally obtained by kern and frenkel @xcite for the two - patch case and can be contrasted with those of the corresponding single - patch counterpart @xcite and those obtained when the radial part of the potential is of the baxter type . @xcite the rhnc calculation presented here , along with the corresponding calculation carried out in our previous paper , @xcite together constitute the first attempt to apply a well - defined integral equation theory to such highly anisotropic potentials having sharp angular modulation . an important outcome of our calculations is the clarification of the combined effect that size and distribution of the patches have on the gas - liquid coexistence lines and critical parameters . the reduction of the bonding surface clearly decreases the critical temperature , an effect which can be related to the decrease in the bonding energy of the system . more interestingly , it also shows a suppression of the critical density , which can be interpreted along the same lines used in interpreting the valence dependence in patchy colloids . @xcite indeed , the maximally bonded structures require lower and lower local densities on decreasing @xmath0 . interestingly , while in the single - bond - per - patch condition the evolution of the critical parameters on decreasing valence can be followed down to the limit where clustering prevents phase separation , @xcite in the model studied here crystallization pre - empts the observation of the liquid - gas separation for @xmath185 . crystallization is here much more effective due to the analogy between the local fully - bonded configuration and the crystal structure . by contrast , crystallization is never observed for the one - patch case , where it has been shown that the lowest energy configuration is reached instead via the process of formation of large aggregates ( micelles and vesicles ) or via the formation of lamellar phases . @xcite this difference highlights the important coupling between the orientational part of the potential and the possibility of forming extended fully - bonded structures . our results indicate that , for a given coverage , in the two - patch case both the critical temperature and density are slightly higher then their corresponding one - patch counterparts , thus indicating that an increase of the valence favours the gas - liquid transition , in agreement with previous findings . a final important consequence of our study concerns the limit of very small coverages that is particularly interesting . indeed , it is possible to tune the structure of the system and control the topology of its ordered arrangement . by doing this we have observed a progression from the case where chains are stable ( in the one - bond - per - patch limit ) to the case where independent planes are found , evolving for slightly larger @xmath0 values into an ordered three - dimensional crystalline structure . each of these ordered structures is observed in a restricted range of @xmath0 values . this possibility of fine tuning the morphology by controlling the patterning of the particle surfaces may offer an interesting possibility for specific self - assembling structures . an additional perspective of our work should be stressed . several studies ( see _ e.g. _ ref . and references therein ) have exploited spherically - symmetric potentials to mimic effective protein - protein interactions , especially in connection with protein crystallization . @xcite this is clearly unrealistic for the majority of proteins where the distribution of hydrophobic surface groups is significantly irregular , a feature that can be captured , at the simplest level of description , by the model studied here . the specific location of the coexistence lines , such as those considered in the present study , have important consequences in the study of pathogenic events for sickle cell anemia @xcite and other human diseases . @xcite we thank philip j. camp and enrique lomba for useful suggestions . fs acknowledges support from noe softcomp nmp3-ct-2004 - 502235 , erc226207patchycolloids and itn - comploids . ag acknowledges support from prin - cofin 2007b57eab(2008/2009 ) . the expansion in spherical harmonics @xmath186 of @xmath4 in an arbitrary space frame reads @xcite @xmath187 where we have introduced the rotational invariants @xcite @xmath188 note that @xmath189 coincides with @xmath190 up to a normalization constant ( see appendix [ app : appb ] ) . we are free to set the origin of the coordinate frame at the center of particle @xmath8 and choose its orientation so that @xmath191 without loss of generality . we first note that @xmath192 clearly , the orientation of the @xmath193 axes is then irrelevant , so we may integrate out the angles @xmath194 and @xmath143 ; this leads to the average @xmath195 , where we note that @xmath196^{1/2 } p_{l_{2 } m_{2}}\left(\cos \theta_2\right ) \frac{1}{2\pi } \int_{0}^{2\pi } d\varphi_2 e^{\mathrm{i } m_2 \varphi_2 } \\ \nonumber & = & \left ( \frac{2l_2 + 1}{4\pi } \right)^{1/2 } p_{l_{2}}\left(\cos \theta_2\right ) \delta_{m_20}\end{aligned}\ ] ] and similarly for @xmath197 . here the @xmath198 are the usual legendre polynomials . we have then from eqs . ( [ appa : eq1 ] ) and ( [ appa : eq2 ] ) that @xmath199^{1/2 } c\left(l_{1 } l_{2 } l ; 0 0 0 \right ) p_{l_{2}}\left(\cos \theta_2\right ) p_{l}\left(\cos \theta\right ) . \label{appa : eq5}\end{aligned}\ ] ] we are interested in the angular behavior within the well , @xmath131 , so we finally integrate over the radial variable @xmath66 and define @xmath200 the result is then a function of the polar coordinate @xmath144 of @xmath10 and the polar orientation of the second patch , @xmath146 ; that is , @xmath201^{1/2 } c\left(l_{1 } l_{2 } l ; 0 0 0 \right ) p_{l_{2}}\left(\cos \theta_2\right ) p_{l}\left(\cos \theta\right),\end{aligned}\ ] ] given that the @xmath202 axis is aligned with the patches of particle 1 . in this appendix we consider the coefficients @xmath190 of rotational invariants that have proven to be particularly useful in discriminating among different orientational behaviors . in numerical simulations they are defined as follows ( see _ e.g. _ ref . ) , @xmath204 where the @xmath205 are rotational invariants . explicit expressions for the first few are @xcite @xmath206 , \\ \nonumber \delta^{220}\left(12 \right)&= & \frac{5}{2 } e\left(12\right)=\frac{5}{2 } \left[3\left ( \hat{\mathbf{n}}_1 \cdot \hat{\mathbf{n}}_2\right)^2 -1\right].\end{aligned}\ ] ] other expressions can be found in ref . . we note that the first expression in eqs . ( [ appb : eq2 ] ) yields @xmath207 , which coincides with the radial distribution function @xmath208 . here we have @xmath209 ; in all other cases @xmath210 . some of the coefficients have particularly interesting physical interpretations : the term @xmath181 is the coefficient of ferroelectric correlation , the term @xmath182 the coefficient of dipolar correlation , the term @xmath155 the coefficient of nematic correlation , and so on . it might be useful to show how these general expressions ( typically computed in monte carlo simulations ) connect with the corresponding ones typically evaluated in an integral equation approach . we do this for the representative case of @xmath211 , the others being similar . we define @xmath212 , where the proportionality constant is obtained through a particular prescription to be further elaborated below . the projections @xmath213 of @xmath4 on the rotational invariants @xmath214 as defined by eq . ( [ appa : eq1 ] ) are related to the values @xmath215 in the axial frame by @xcite @xmath216 where @xmath217 . consider as a representative example the quantity @xmath218 defined in eqs . ( [ appb : eq2 ] ) and note that in eq . ( [ appb : eq1 ] ) one has @xmath219 upon choosing the molecular frame ( @xmath220 ) , one finds @xmath221,\end{aligned}\ ] ] where the @xmath186 are spherical harmonics . @xcite the expansion ( [ appa : eq1 ] ) in a molecular frame reduces to @xmath222 using eqs . ( [ appb : eq5 ] ) , ( [ appb : eq6 ] ) , and the orthogonality relations for spherical harmonics , @xcite one finds easily that @xmath223,\end{aligned}\ ] ] so that , combining with @xmath182 given in eqs . ( [ appb : eq1 ] ) and ( [ appb : eq2 ] ) and using the symmetry property @xmath224 , one finds @xmath225 we can work out the first few projections by using tabulated values for the clebsch - gordan coefficients @xmath226 . using the symmetry properties of the clebsch - gordan coefficients and the @xmath227 , one finds @xmath228 , \\ \nonumber g\left(r;112\right)&= & \left ( 8\pi/15 \right)^{1/2 } \left [ g_{110}\left(r\right ) + g_{111}\left(r\right ) \right ] , \\ \nonumber g\left(r;220\right)&= & \left ( 4\pi/5 \right)^{1/2 } \left[g_{220}\left(r\right)-2 g_{221}\left(r\right)+2g_{222}\left(r\right ) \right ] , \\ \nonumber g\left(r;222\right)&= & -\left ( 8\pi/35 \right)^{1/2}\left[g_{220}\left(r\right ) -g_{221}\left(r\right)-2 g_{222}\left(r\right ) \right ] , \\ \nonumber g\left(r;224\right)&= & \left ( 8\pi/35 \right)^{1/2 } \left[g_{220}\left(r\right)+\frac{4}{3 } g_{221}\left(r\right)+ \frac{1}{3}g_{222}\left(r\right ) \right ] , \\ \nonumber g\left(r;011\right)&= & \left ( 4\pi/3 \right)^{1/2 } g_{010}\left(r\right ) = -g\left(r;101\right ) , \\ \nonumber g\left(r;022\right)&= & \left ( 4\pi/5 \right)^{1/2 } g_{020}\left(r\right)=g\left(r;202\right ) , \\ \nonumber g\left(r;121\right)&= & -\left ( 8\pi/15 \right)^{1/2 } \left[g_{120}\left(r\right)-\sqrt{3}\ , g_{121}\left(r\right ) \right]=-g\left(r;211\right ) , \\ \nonumber g\left(r;123\right)&= & \left ( 12\pi/35 \right)^{1/2}\left[g_{120}\left(r\right)+ \frac{2}{\sqrt{3 } } g_{121}\left(r\right ) \right]=-g\left(r;213\right ) . \label{appb : eq9}\end{aligned}\ ] ] as anticipated above , we now fix the proportionality constant in @xmath212 by dividing out the leading constants above so that the coefficient of @xmath229 is unity ; _ e.g. _ , @xmath230 . 0.5 cm and @xmath235 ( @xmath236 ) , where the orientation vectors @xmath141 , @xmath142 define angles @xmath237 , @xmath238 with the vector @xmath10 joining the centers of the two spheres and directed from sphere @xmath8 to sphere @xmath9 . the particular case shown corresponds to a @xmath239 fraction of attractive surface ( coverage @xmath0 ) . [ fig : fig1],title="fig:",width=529 ] 0.5 cm . points represent mc results while thick solid lines report rhnc values ( for @xmath240 ) . thinner solid lines are a guide for the eye whereas the thick dashed line shows the estimated gcmc critical point for any fixed coverage @xmath0 . the mc data for @xmath241 ( sw ) coincide within the numerical error with the ones reported in refs . and . [ fig : fig2],title="fig:",width=529 ] 0.5 cm . ( a ) the case of coverages such that each patch can be involved in only one interaction . in this case , the system forms polydisperse chains . the snapshot here refers to the case @xmath242 and @xmath243 . ( b ) values of @xmath0 such that each patch can be involved in only two interactions . in this case , at low @xmath80 the system forms bonded planes interacting with each other only via excluded volume interactions . the snapshot shows one such plane . ( c ) values of @xmath0 such that each patch can be involved in only three interactions . the crystal is now formed by interconnected planes , with a triangular arrangement of the particles in the plane . adjacent planes are shifted in such a way that each particle sits in correspondence to the center of a triangle of the previous and following planes . ( d ) values of @xmath0 such that each patch can be involved in at most four interactions . the crystal is now formed by interconnected planes , with a square arrangement of the particles in the plane . adjacent planes are shifted in such a way that each particle sits in correspondence to the center of a square of the previous and following planes . [ fig : fig3],title="fig:",width=529 ] from the rhnc equation for different coverages and two specific orientations of the patches : @xmath110 configuration corresponding to @xmath106 ( a ) and x configuration corresponding to @xmath107 ( b ) . all curves are for a state point with @xmath71 , @xmath108 , and @xmath109 . black and green lines show the limiting cases of square - well ( @xmath29 ) and hard - sphere ( @xmath30 ) potentials , respectively . other coverages are @xmath244 for both the @xmath110 and x configurations . [ fig : fig4],width=453 ] for different coverages and two specific orientations of the patches : @xmath110 configuration corresponding to @xmath106 ( a ) and x configuration corresponding to @xmath107 ( b ) . all curves are for a state point with @xmath71 , @xmath108 , and @xmath109 . both rhnc and mc results are depicted for @xmath245 . [ fig : fig5],width=453 ] ( ( a ) and ( c ) ) and the radial distribution function @xmath126 ( ( b ) and ( d ) ) for coverages @xmath129 and @xmath130 , respectively . here @xmath109 and @xmath71 as before , while densities are @xmath108 and @xmath132 . [ fig : fig6],width=529 ] as functions of @xmath246 for two different orientations of the patches on sphere @xmath9 , given that sphere @xmath8 is fixed with patches along the @xmath147 axis . results are reported for two different coverages , @xmath129 ( ( a ) and ( b ) ) and @xmath130 ( ( c ) and ( d ) ) , and two different densities , @xmath108 ( ( a ) and ( c ) ) and @xmath132 ( ( b ) and ( d ) ) , at the same temperature @xmath109 . again , the square - well width was set to @xmath71 . the colored arrows are cartoons of the orientations of sphere @xmath9 patches , corresponding to @xmath247 . note that these are the same state points considered in fig . [ fig : fig6 ] . [ fig : fig7],width=529 ] , @xmath159 , and @xmath156 as functions of @xmath138 . results are reported for two different coverages , @xmath129 ( ( a ) and ( b ) ) and @xmath130 ( ( c ) and ( d ) ) , and two different densities , @xmath108 ( ( a ) and ( c ) ) and @xmath132 ( ( b ) and ( d ) ) , at the same temperature @xmath109 and square - well width @xmath71 . again , these are the same state points considered in figs . [ fig : fig6 ] and [ fig : fig7 ] . [ fig : fig8],width=529 ] for the one - patch model as a function of @xmath246 for three different orientations of the patch on sphere @xmath9 , given that sphere @xmath8 is fixed with patch along the @xmath147 axis . this is the one - patch counterpart of fig . [ fig : fig7 ] . results are reported for two different coverages , @xmath129 ( ( a ) and ( b ) ) and @xmath130 ( ( c ) and ( d ) ) , and two different densities , @xmath108 ( ( a ) and ( c ) ) and @xmath132 ( ( b ) and ( d ) ) , at the same temperature @xmath109 and square - well width @xmath71 . the colored arrows are cartoons of the orientations of the sphere @xmath9 patch , corresponding to @xmath248 . [ fig : fig10],width=529 ] , @xmath182 , and @xmath155 for the one - patch model as functions of @xmath138 . results are reported for two different coverages , @xmath129 ( ( a ) and ( b ) ) and @xmath130 ( ( c ) and ( d ) ) , and two different densities , @xmath108 ( ( a ) and ( c ) ) and @xmath132 ( ( b ) and ( d ) ) , at the same temperature @xmath109 and square - well width @xmath71 . again , these are the same state points considered in fig . [ fig : fig8 ] . [ fig : fig11],width=529 ]

this paper establishes the existence of a _ gap _ for the stable length spectrum on a hyperbolic manifold . if is a hyperbolic-manifold , for every positive there is a positive depending only on and on such that an element of with stable commutator length less than is represented by a geodesic with length less than . moreover , for any such , the first accumulation point for stable commutator length on conjugacy classes is at least . conversely , `` most '' short geodesics in hyperbolic-manifolds have arbitrarily small stable commutator length . thus stable commutator length is typically good at detecting the thick - thin decomposition of , and can be thought of as a kind of _ homological margulis constant_. [ section ] [ theorem]lemma [ theorem]corollary [ theorem]conjecture [ theorem]question [ theorem]problem [ theorem]definition [ theorem]construction [ theorem]notation [ theorem]convention [ theorem]remark [ theorem]example h

math0605354

if @xmath0 is a closed hyperbolic @xmath1-manifold with @xmath7 , the mostow rigidity theorem says that the geometric structure on @xmath0 is uniquely determined by the algebraic structure of @xmath4 , up to isometry . this means that in principle , any kind of geometric information about @xmath0 ( e.g. length spectrum , volume , systoles , etc . ) can be recovered from group theory . on the other hand , with some notable exceptions , it is not easy to find concepts which have a straightforward interpretation on both the geometric and the algebraic side . one of the most important geometric structures on a hyperbolic manifold is the _ thick - thin _ decomposition . there is a universal positive constant @xmath8 in each dimension ( i.e. the _ margulis constant _ ; see @xcite ) such that the part of a hyperbolic @xmath1-manifold @xmath0 with injectivity radius less than @xmath2 ( i.e. the thin " piece ) consists of cusps and solid tubes . in the complement of these margulis pieces ( i.e. the thick " piece ) the geometry and topology is uniformly controlled by the volume . the existence of the margulis constant implies that in each dimension , there is a _ universal _ notion of what it means for a closed geodesic to be _ a careful analysis of the behavior of thin pieces under various operations is the source of many compactness and finiteness theorems . in this paper , we establish a relationship between the _ hyperbolic length _ of a closed geodesic @xmath9 , and the _ stable commutator length _ ( to be defined in the sequel ) of the corresponding conjugacy class in @xmath4 . parallel to the geometric relationship between hyperbolic length and margulis constant , we show that there is a _ universal _ notion of what it means for a conjugacy class to have small stable commutator length . the results in this paper show that when the stable commutator length of a conjugacy class in @xmath4 is sufficiently small , so is the length of the closed geodesic in the corresponding free homotopy class of loops in @xmath0 , and ( in the typical case ) , conversely . thus the thick - thin decomposition of @xmath0 is typically detected by stable commutator length . a precise definition of stable commutator length in groups is given in [ scl_definition_subsection ] . informally , if @xmath10 is a group and @xmath11 is an element of the commutator subgroup ( denoted @xmath12 $ ] ) , the _ commutator length _ of @xmath11 , denoted @xmath13 , is the minimum number of commutators in @xmath10 whose product is equal to @xmath11 . the _ stable commutator length _ of @xmath11 , denoted @xmath14 , is the liminf of the ratio of @xmath15 to @xmath1 as @xmath16 . by convention define @xmath17 if @xmath11 is not in @xmath12 $ ] , so that @xmath18 if and only if the image of @xmath11 in @xmath19 is nontrivial . note that @xmath20 is a _ characteristic _ function ; i.e. it is constant on orbits of @xmath21 . in particular , it is constant on conjugacy classes in @xmath10 . a detailed summary of the contents of this paper follows . sections [ scl_section ] and [ quasimorphism_section ] are introductory , and contain a brief exposition of the theory of stable commutator length and its relation to @xmath22-dimensional bounded cohomology and quasimorphisms via bavard s duality theorem . references for these sections are @xcite or @xcite . section [ surgery_section ] exhibits a relationship between stable commutator length and dehn surgery in hyperbolic @xmath6-manifolds . a fundamental observation is that elements with very small stable commutator length may be obtained in @xmath6-manifold groups by dehn filling a cusped manifold . let @xmath0 be a @xmath6-manifold with a single torus boundary @xmath23 . choose a meridian and a longitude @xmath24 for @xmath25 such that the curve @xmath26 is a longitude in @xmath25 ; i.e. its image in @xmath27 is trivial . let @xmath28 denote the result of @xmath29 surgery on @xmath0 . that is , @xmath28 is obtained from @xmath0 by dehn filling @xmath0 along the curve @xmath30 in @xmath25 . if @xmath28 is hyperbolic , let @xmath31 denote the core geodesic of the filling torus . since the image of @xmath26 in @xmath27 is trivial , some multiple of @xmath26 bounds a surface in @xmath0 . let @xmath32 be an orientable surface in @xmath0 which bounds @xmath1 times the longitude , and let @xmath33 . if @xmath34 is in the conjugacy class corresponding to @xmath31 , there is an inequality @xmath35 from this the following estimates can be derived . let @xmath28 be a hyperbolic @xmath6-manifold , obtained by @xmath29 surgery on a @xmath36-cusped hyperbolic @xmath6-manifold @xmath0 . when the core geodesic @xmath37 is contained in an embedded tube of radius @xmath25 at least @xmath22 then @xmath38 let @xmath0 be a @xmath36-cusped hyperbolic @xmath6-manifold . normalize the euclidean structure on the cusp to have area @xmath36 , and let @xmath39 be the shortest curve on the cusp which is homologically essential in @xmath0 . then @xmath40 similar estimates are implicit in the work of agol @xcite . from a theorem of neumann - zagier @xcite , it follows that these estimates are of the correct order of magnitude . the results in [ surgery_section ] may be summarized by saying that one can estimate the length of `` most '' sufficiently short geodesics in a hyperbolic @xmath6-manifold ( up to an order of magnitude ) and the geometry of a margulis tube directly from stable commutator length . section [ spectral_gap_section ] exhibits a relationship between stable commutator length and geodesic length which is converse to the relationship studied in [ surgery_section ] . the following theorem shows that an element with sufficiently small stable commutator length is represented by an arbitrarily short geodesic . for every dimension @xmath1 and any @xmath41 there is a constant @xmath42 such that if @xmath0 is a complete hyperbolic @xmath1-manifold , and @xmath43 has stable commutator length @xmath44 then if @xmath9 is the unique geodesic in the free homotopy class associated to the conjugacy class of @xmath11 , @xmath45 the dependence of @xmath3 on @xmath2 is not proper ; as @xmath46 the function @xmath47 approaches some limit . a uniform estimate for this upper bound , valid in any dimension , can be obtained , leading to a _ universal _ estimate on the first accumulation point for stable commutator length on conjugacy classes in a hyperbolic manifold . let @xmath0 be a closed hyperbolic manifold , of any dimension . let @xmath48 be the first accumulation point for stable commutator length on conjugacy classes . then @xmath49 this theorem implies that for any closed hyperbolic @xmath1-manifold , and any @xmath50 , there are only finitely many conjugacy classes of elements in @xmath51 with stable commutator length less than @xmath3 . this gives a precise sense in which a stable commutator length less than @xmath5 should be thought of as _ small_. the lower bound of @xmath5 is the homological analogue of margulis constant . in contrast with the geometric margulis constant , this homological constant can be estimated accurately , and is independent of dimension . the phenomena revealed in these theorems are quite robust ; in a forthcoming paper with koji fujiwara @xcite we prove similar theorems for word - hyperbolic groups and certain groups acting on @xmath3-hyperbolic spaces , such as amalgamated free products and mapping class groups . section [ de_rham_section ] gives an application of the results in [ spectral_gap_section ] . given an element @xmath52 where @xmath0 is hyperbolic , let @xmath9 denote the geodesic in the free homotopy class corresponding to the conjugacy class of @xmath11 . define the _ de rham length _ @xmath53 by the formula @xmath54 where the supremum is taken over all smooth @xmath36-forms @xmath55 on @xmath0 , and @xmath56 denotes the operator norm of the @xmath22-form @xmath57 . from results in [ quasimorphism_section ] and [ surgery_section ] there is an inequality @xmath58 . moreover , theorem a and theorem b imply that the values of @xmath14 and @xmath53 are usually comparable when @xmath9 is a short geodesic . by contrast , theorem c together with a construction due to thurston shows that the values of @xmath59 and @xmath60 are not always comparable when @xmath61 is large . let @xmath0 be a finite volume complete hyperbolic @xmath1-manifold with @xmath7 . then for any @xmath62 there are infinitely many conjugacy classes @xmath43 with @xmath63 finally , in [ geometry_section ] we briefly discuss @xmath59 in @xmath6-manifolds with other ( non - hyperbolic ) geometric structures . i would like to thank ian agol , iain aitchison , marc culler , nathan dunfield , koji fujiwara , jason manning , shigenori matsumoto and the anonymous referee for some useful comments on the material in this paper . i would also like to thank bill thurston for some substantial comments , for a number of key ideas , and for drawing my attention to @xcite , especially theorem [ injective_wrap_estimate ] , and its relevance to stable commutator length . while writing this paper , i was partially supported by a sloan research fellowship , and nsf grant dms-0405491 . the material in this section and in [ quasimorphism_section ] is standard , and is included for the benefit of the reader . a basic reference is @xcite or @xcite . let @xmath10 be a group , and @xmath64 an element . suppose @xmath11 is in the commutator subgroup @xmath65 $ ] . the _ commutator length _ of @xmath11 , denoted @xmath13 , is the word length of @xmath11 in the ( usually infinite ) generating set of @xmath12 $ ] consisting of all commutators . in words , @xmath13 is smallest number of commutators whose product is @xmath11 . the function @xmath66 can be extended to all of @xmath10 by setting @xmath17 by convention when @xmath11 is not in @xmath12 $ ] . the _ stable commutator length _ of @xmath11 , denoted @xmath14 is equal to @xmath67 the value of @xmath14 is infinite if and only if the image of @xmath11 in @xmath68 $ ] has infinite order . the image of a commutator under a homomorphism between groups is again a commutator . it follows that the functions @xmath66 and @xmath20 are non - increasing under homomorphisms between groups , and characteristic ; i.e. constant on orbits of @xmath21 . [ amenable_example_vanishes ] if @xmath10 is amenable , stable commutator length is identically zero on @xmath12 $ ] . see bavard @xcite cor . 2 , p. 139 . [ free_example_bound ] if @xmath69 is free , @xmath70 for all nontrivial @xmath71 . this follows from work of duncan - howie @xcite thm . 3.3 , as explained in @xcite . the same arguments apply also to fundamental groups of closed orientable surfaces of genus at least @xmath22 , so @xmath72 is a lower bound for @xmath59 on nontrivial elements in these groups too . a similar estimate @xmath73 was obtained earlier by culler @xcite . let @xmath32 be a closed orientable surface of genus @xmath74 . if @xmath11 in the mapping class group of @xmath32 is a product of @xmath75 right - handed dehn twists along essential disjoint simple closed curves in @xmath32 then @xmath76 see korkmaz @xcite thm . 2.1 for the case @xmath77 or kotschick @xcite for the general case . informally , a quasimorphism is a real - valued function on a group which is additive up to bounded error . let @xmath10 be a group . a _ quasimorphism _ on @xmath10 is a function @xmath78 for which there exists some least real number @xmath79 called the _ defect _ such that for all @xmath80 there is an inequality @xmath81 the number @xmath79 is sometimes called the _ error _ or the _ deficit _ of @xmath82 ( _ dfaut _ in french ) . [ product_inequality ] let @xmath32 be a ( possibly infinite ) generating set for @xmath10 . let @xmath83 be a word in the generators , representing an element of @xmath10 . let @xmath84 denote the length of @xmath83 in the generator set , and let @xmath85 denote the @xmath86th letter of @xmath83 . there is an inequality @xmath87 if @xmath88 where @xmath89 then @xmath90 by the definition of @xmath79 . now induct on the length of @xmath83 and apply the triangle inequality . a quasimorphism @xmath82 on @xmath10 is _ symmetric _ if @xmath91 for all @xmath64 . for any quasimorphism @xmath82 the linear combination @xmath92 is called the _ symmetrization _ of @xmath82 . [ symmetrize_lemma ] for any quasimorphism @xmath82 , the symmetrization @xmath93 is a symmetric quasimorphism and satisfies @xmath94 . for any @xmath95 , there is an equality @xmath96 a quasimorphism @xmath82 on @xmath10 is _ homogeneous _ if @xmath97 for all @xmath64 and all integers @xmath1 . for any quasimorphism @xmath82 the limit @xmath98 exists and is called the _ homogenization _ of @xmath82 . [ homogenize_lemma ] for any quasimorphism @xmath82 , the homogenization @xmath99 is a homogeneous quasimorphism and satisfies @xmath100 . see bavard @xcite lem . 3.6 , p. 142 or @xcite for a proof . [ conjugacy ] a homogeneous quasimorphism is symmetric , and is constant on conjugacy classes . symmetry is part of the definition . for any @xmath80 , any homogeneous quasimorphism @xmath82 and any integer @xmath1 , @xmath101 where the last inequality follows from the symmetry of @xmath82 and lemma [ product_inequality ] . the set of quasimorphisms on @xmath10 is denoted @xmath102 and the set of homogeneous quasimorphisms is denoted @xmath103 . for any group @xmath10 , the sets @xmath104 are vector subspaces of @xmath105 . the function @xmath106 defines a pseudo - norm on these vector spaces , and vanishes exactly on the set of quasimorphisms with defect @xmath107 . a quasimorphism with defect @xmath107 is a homomorphism to @xmath108 ; the set of such homomorphisms is naturally isomorphic to @xmath109 . the subspace @xmath109 is closed in @xmath102 , the quotient space @xmath110 is a normed vector space with norm @xmath106 , and @xmath111 is a relatively closed normed vector subspace . a function on @xmath10 such as an element @xmath112 extends by linearity to define a @xmath36-cochain . the coboundary of the @xmath36-cochain @xmath82 is denoted @xmath113 and is a @xmath22-cocycle on @xmath10 . the defining property of a quasimorphism implies that @xmath113 is a bounded @xmath22-cocycle with norm equal to @xmath79 . this defines a continuous map @xmath114 . there is an exact sequence @xmath115 see bavard @xcite prop . 3.3.1 , p. 135 or besson @xcite for a proof and @xcite for a more substantial discussion of bounded cohomology . there is a fundamental relationship between stable commutator length and quasimorphisms in any group . [ bavard_theorem ] let @xmath10 be any group . then for any @xmath65 $ ] there is an equality @xmath116 the proof is non - constructive . it uses the hahn - banach theorem , @xmath117@xmath118 duality , and the fact that @xmath119 is generated efficiently ( in the @xmath117 sense ) by surfaces for which one knows the exact values of the @xmath117 norm on homology classes . one direction of the theorem is elementary . a homogeneous quasimorphism is constant on conjugacy classes by lemma [ conjugacy ] . if @xmath120\cdots[a_n , b_n]$ ] for some @xmath121 then @xmath122\cdots[a_n , b_n ] ) - \sum_i\phi(a_ib_ia_i^{-1 } ) - \sum_i \phi(b_i^{-1})| \le ( 2n-1)d(\phi)\ ] ] by symmetry and lemma [ product_inequality ] . it is this direction of the theorem that is most relevant to this paper . [ rotation_example ] let @xmath10 be a group of orientation - preserving homeomorphisms of @xmath123 and let @xmath124 be the group of orientation - preserving homeomorphisms of @xmath108 which cover elements of @xmath10 under the covering map @xmath125 . for @xmath126 define @xmath127 by @xmath128 then @xmath127 is an element of @xmath129 with defect at most @xmath36 . let @xmath69 be a free group on a generating set @xmath32 . let @xmath83 be a reduced word in the generators @xmath32 of length at least @xmath22 . for any reduced word @xmath71 , define @xmath130 then one may verify directly that @xmath131 is a nontrivial quasimorphism with defect at most @xmath6 . more examples are given in [ derham_subsection ] . this section relates stable commutator length to hyperbolic geometry , especially in @xmath6-manifolds . the discussion is concrete and only makes use of the `` easy '' direction of theorem [ bavard_theorem ] . let @xmath132 denote hyperbolic @xmath1-space . let @xmath133 be a ( possibly ideal ) geodesic triangle with vertices @xmath134 and respective interior angles @xmath135 in @xmath132 . then by the gauss - bonnet formula , the area of @xmath133 satisfies @xmath136 in particular , @xmath137 with equality if and only if @xmath133 is ideal . the following construction is due to barge - ghys @xcite : let @xmath0 be a hyperbolic manifold . let @xmath138 be a basepoint . let @xmath139 , and let @xmath140 be a @xmath36-form . for each element @xmath64 let @xmath141 denote the oriented geodesic , based at @xmath142 , in the homotopy class of @xmath11 . for each @xmath64 , define @xmath143 by @xmath144 [ form_is_quasimorphism ] let @xmath145 be as above . the function @xmath82 is a symmetric quasimorphism on @xmath10 with @xmath146 where @xmath147 denotes the operator norm on the @xmath22-form @xmath57 . if @xmath80 then the geodesics @xmath148 form the three sides of a geodesic triangle @xmath133 . observe that @xmath149 and @xmath150 are the same geodesic with opposite orientations . by stokes theorem we get @xmath151 quasimorphisms @xmath82 of this kind are sometimes called _ barge - ghys _ quasimorphisms , or _ de rham _ quasimorphisms . by lemma [ homogenize_lemma ] , the homogenization @xmath99 satisfies @xmath152 . by theorem [ bavard_theorem ] , for any @xmath153 there is an inequality @xmath154 . the homogenization @xmath99 is obtained by setting @xmath155 where @xmath156 is the _ free _ geodesic corresponding to the conjugacy class of @xmath11 . we will not use this fact in this paper . given an element @xmath11 , let @xmath156 denote the free geodesic in the free homotopy class corresponding to the conjugacy class of @xmath11 . observe if @xmath157 then @xmath158 and @xmath159 . if @xmath156 is a short simple geodesic contained in an embedded tube of large radius , one can obtain lower bounds on @xmath143 . [ tube_estimate ] let @xmath43 and let the conjugacy class of @xmath11 be represented by a simple geodesic loop @xmath9 in @xmath0 which is the core of an embedded solid torus of radius @xmath25 . then there is a homogeneous quasimorphism @xmath160 with @xmath161 satisfying @xmath162 let @xmath32 be the solid torus of radius @xmath25 about @xmath9 . let @xmath163 be the function whose value at every point is the hyperbolic distance to @xmath9 . denote the radial projection map by @xmath164 . parameterize @xmath9 by @xmath165 , so that @xmath166 is the length form on @xmath9 , and @xmath167 . pulling back by @xmath142 extends @xmath165 and @xmath166 to all of @xmath32 . note that @xmath165 takes values in @xmath168 but @xmath166 is a real - valued @xmath36-form . define @xmath169 on @xmath32 , and extend it by @xmath107 outside @xmath32 . there is an equality @xmath170 on @xmath32 . by direct calculation , @xmath171 on @xmath32 , so that @xmath172 . the form @xmath55 is not smooth along @xmath173 , but it is lipschitz . let @xmath174 be a @xmath175 function on @xmath176 $ ] taking the value @xmath36 in a neighborhood of @xmath107 and the value @xmath107 in a neighborhood of @xmath177 , and with @xmath178 throughout , for some small @xmath2 . the product @xmath179 is @xmath175 and satisfies @xmath180 so @xmath181 . let @xmath182 be a homogeneous quasimorphism , obtained from @xmath183 by integrating along geodesics as in lemma [ form_is_quasimorphism ] , and then homogenizing . the forms @xmath183 converge in @xmath184 as @xmath185 and therefore @xmath182 converges in @xmath103 to some @xmath82 . multiplying @xmath82 by @xmath186 we obtain a homogeneous quasimorphism with defect at most equal to @xmath187 , whose value on @xmath11 is @xmath188 . to apply lemma [ tube_estimate ] to a sufficiently short geodesic , it is necessary to estimate the radius of an embedded tube in terms of the length of the core geodesic . the following estimate is due to hodgson - kerckhoff : [ hodgson_kerckhoff_estimate ] let @xmath32 be an embedded tube in a hyperbolic @xmath6-manifold . let @xmath25 be the radius of @xmath32 and @xmath61 the length of the core geodesic . then there is an estimate @xmath189 see @xcite for a proof . note for @xmath25 sufficiently large , we get @xmath190 . [ reznikov_remark ] for any dimension @xmath1 there exists a constant @xmath191 such that @xmath192 . see reznikov @xcite p. 478 for a proof . putting this together with lemma [ tube_estimate ] and the easy direction of theorem [ bavard_theorem ] , one obtains a lower bound on the stable commutator length of @xmath11 in terms of the geometry of a geodesic representative . conversely , one can sometimes obtain an upper bound on stable commutator length from topology . let @xmath0 be a @xmath6-manifold with a single torus boundary @xmath23 . choose a meridian and a longitude @xmath24 for @xmath25 such that the curve @xmath26 is a longitude in @xmath25 ; i.e. its image in @xmath27 is trivial . let @xmath28 denote the result of @xmath29 surgery on @xmath0 . that is , @xmath28 is obtained from @xmath0 by dehn filling @xmath0 along the curve @xmath30 in @xmath25 , where the additive notation denotes a particular primitive class in @xmath193 and therefore corresponds to a unique isotopy class of simple closed curve . if @xmath28 is hyperbolic , let @xmath31 denote the core geodesic of the filling torus . since the image of @xmath26 in @xmath27 is trivial , there is some @xmath1 so that @xmath194 ( with additive notation ) is trivial in @xmath195 . this implies that there is an orientable surface @xmath32 in @xmath0 which bounds @xmath1 times the longitude . let @xmath32 be such a surface of least genus , and for such a surface , let @xmath33 . [ seifert_estimate_scl ] with notation as above , if @xmath34 is in the conjugacy class corresponding to @xmath31 , there is an inequality @xmath35 let @xmath196 be a finite cover of @xmath32 with two boundary components , for which the degree of the covering @xmath197 is @xmath198 . such a cover exists for every integer @xmath199 . the composition of @xmath197 with the map @xmath200 wraps each boundary component @xmath201 times around @xmath26 . add a rectangle to @xmath196 to boundary connect sum the two boundary components together , and map this rectangle to a point in @xmath26 . the result is a surface @xmath202 with @xmath203 . since euler characteristic is multiplicative under coverings , @xmath204 . the boundary component of @xmath202 maps @xmath205 times around @xmath26 so @xmath206 taking the limit @xmath207 gives the estimate @xmath208 . under @xmath29 dehn surgery , the longitude @xmath26 is wrapped @xmath142 times around the core geodesic of the filling torus . if @xmath209 ( hereafter @xmath11 for brevity ) is in this conjugacy class , then @xmath210 is the image of @xmath26 . since @xmath59 can not increase under homomorphisms , @xmath211 . lemma [ seifert_estimate_scl ] says that away from finitely many lines in dehn surgery space , the core geodesics obtained by hyperbolic dehn surgery correspond to conjugacy classes in @xmath51 with arbitrarily small stable commutator length . since all sufficiently short geodesics arise by geometric dehn surgery on cusped manifolds , this strongly suggests that `` typical '' short geodesics have very small stable commutator length . using lemma [ hodgson_kerckhoff_estimate ] and lemma [ tube_estimate ] , we can make the relation between @xmath59 and length quantitative . for simplicity , we treat the case of knot complements in @xmath212 . let @xmath28 be a hyperbolic @xmath6-manifold , obtained by @xmath29 surgery on a @xmath36-cusped hyperbolic @xmath6-manifold @xmath0 . when the core geodesic @xmath37 is contained in an embedded tube of radius @xmath25 at least @xmath22 then @xmath38 for the sake of legibility , abbreviate @xmath31 to @xmath9 throughout this proof . by lemma [ hodgson_kerckhoff_estimate ] and lemma [ tube_estimate ] there is a homogeneous quasimorphism @xmath82 with @xmath213 and satisfying @xmath214 where @xmath215 . by theorem [ bavard_theorem ] and lemma [ seifert_estimate_scl ] there is an estimate @xmath216 and therefore @xmath217 for @xmath218 there is an inequality @xmath219 and therefore @xmath220 similarly , for @xmath218 there is an inequality @xmath221 so @xmath222 in @xcite , neumann and zagier define the following quadratic form @xmath223 : @xmath224 where @xmath173 is the horotorus boundary of the cusp on @xmath0 , and @xmath30 is a straight curve on the horotorus ( in the intrinsic euclidean metric ) representing @xmath142 times the meridian plus @xmath225 times the longitude . equivalently , we can scale the euclidean cusp to have area @xmath36 , and the form reduces to @xmath226 . proposition 4.3 from @xcite gives the following asymptotic formula for @xmath61 , the length of the core geodesic of @xmath28 : [ neumann_zagier_formula ] @xmath227 it follows that for @xmath225 fixed , we get @xmath228 where @xmath229 is the length of the meridian in the euclidean cusp , normalized to have area @xmath36 . putting this together with theorem a and observing that the radius @xmath25 of an embedded tube around @xmath9 goes to infinity as @xmath230 gives the estimate @xmath231 note that this estimate is most informative when @xmath232 , and implies that the genus of a hyperbolic knot can be bounded from below by a term depending on the geometry of the cusp , when some choice of meridian is much shorter than the longitude . with more work , this estimate can be made quadratic in @xmath39 . when @xmath39 is small , this estimate is substantially better than the crude estimate above . let @xmath0 be a @xmath36-cusped hyperbolic @xmath6-manifold . normalize the euclidean structure on the cusp to have area @xmath36 , and let @xmath39 be the shortest curve on the cusp which is homologically essential in @xmath0 . then @xmath40 let @xmath173 denote the maximal horotorus cusp . by jorgensen s inequality ( see @xcite ) , every essential slope on @xmath173 has length at least @xmath36 . it follows that if @xmath233 on the normalized cusp , then the area of the _ maximal _ cusp @xmath173 is at least @xmath234 . if we do @xmath29 surgery with very large @xmath142 , the area of @xmath173 is almost unchanged . on the other hand , an embedded solid torus of radius @xmath25 with core geodesic @xmath9 has area @xmath235 and therefore @xmath236 and therefore @xmath237 up to a multiplicative constant which goes to @xmath36 as @xmath238 . using the approximations @xmath239 and @xmath240 , valid up to a multiplicative constant which goes to @xmath36 as @xmath241 , it follows from lemma [ tube_estimate ] that there are quasimorphisms @xmath82 with @xmath161 and asymptotically satisfying @xmath242 since @xmath243 there is an estimate @xmath244 rearranging gives @xmath245 since @xmath246 , @xmath247 or , in other words , @xmath248 as claimed . this quadratic bound ( in @xmath229 ) is a significant improvement over the linear bound . interpret the condition @xmath249 topologically . the bound in theorem b is not sharp . it is only sensitive to the total area of the surface @xmath32 whereas we need only measure the area of that part of the surface contained in an embedded tube , and moreover , only that part which is almost parallel to the core geodesic . a circle - packing theorem due to boroczky @xcite says that at most @xmath250 of the area of a pleated surface representing @xmath32 is inside the embedded tube , so one can replace the constant @xmath187 by @xmath251 above with no more work . precisely , one gets the estimate @xmath252 nathan dunfield pointed out to me that the same estimate can be derived from theorem 5.1 in @xcite ( moreover , the observation that boroczky s theorem is relevant is also made in @xcite ) . the most interesting thing about the geometric estimates obtained in this section is that they give the correct order of magnitude relationships between @xmath61 , @xmath142 , @xmath229 and @xmath253 . it follows that the straightforward estimate of stable commutator length obtained in lemma [ seifert_estimate_scl ] is typically close to optimal . stable commutator length in the fundamental group of a manifold can be studied by probing the manifold topologically by maps of surfaces . under suitable geometric hypotheses it makes sense to take representative maps of surfaces which are tailored to the geometry of the target manifold . for @xmath0 hyperbolic , the class of maps of surfaces into @xmath0 we use are so - called _ pleated surfaces _ , as introduced by thurston @xcite , chapter 8 . let @xmath0 be a hyperbolic manifold . a _ pleated surface _ is a complete hyperbolic surface @xmath32 of finite area , together with an isometric map @xmath254 which takes cusps to cusps , and such that every point in @xmath32 is in the interior of a straight line segment in @xmath32 which is mapped by @xmath255 to a straight line segment in @xmath0 . note that `` isometric map '' here means that @xmath255 takes rectifiable curves in @xmath32 to rectifiable curves in @xmath0 of the same length . the set of points @xmath256 where the line segment through @xmath142 is unique is called the _ pleating locus_. the set @xmath257 has the structure of a _ geodesic lamination _ of @xmath32 ; i.e. a closed union of disjoint simple geodesics . the restriction of @xmath255 to each component of @xmath258 is totally geodesic . one may add geodesics to the pleating locus if necessary so that all complementary regions are ideal triangles . a map @xmath254 has a _ pleated representative _ if there is a pleated surface in the relative homotopy class of @xmath255 . if @xmath254 is a pleated surface , @xmath259 by the gauss - bonnet theorem . [ pleated_surface_exists ] let @xmath0 be a hyperbolic manifold . every homotopy class of map @xmath254 sending @xmath173 to a nontrivial hyperbolic element of @xmath4 can either be replaced by a map @xmath260 sending @xmath261 to the same element , where @xmath262 , or it admits a pleated representative . for a proof , see @xcite , especially 8.8 and 8.10 or @xcite chapter 3 . in general , the relation between hyperbolic length and stable commutator length is not straightforward , as the following example shows . [ example_in_surface ] stable commutator length is characteristic , and therefore constant on orbits of @xmath21 . let @xmath263 where @xmath32 is a hyperbolic surface . except for small examples , the orbit of every conjugacy class under @xmath264 is infinite . let @xmath265 be a conjugacy class represented by a geodesic @xmath9 in @xmath32 . if @xmath266 has a pseudo - anosov representative , the geodesic representatives of the conjugacy classes of elements @xmath267 have unbounded length in @xmath32 . on the other hand , @xmath268 is constant . however the first result of this section says that an element with _ sufficiently small _ stable commutator length necessarily has a geodesic representative with a small length . for every dimension @xmath1 and any @xmath41 there is a constant @xmath42 such that if @xmath0 is a complete hyperbolic @xmath1-manifold , and @xmath43 has stable commutator length @xmath44 then if @xmath9 is the unique geodesic in the free homotopy class associated to the conjugacy class of @xmath11 , @xmath45 let @xmath32 be a surface of genus @xmath269 with one boundary component , and @xmath254 a map wrapping @xmath173 homotopically @xmath39 times around @xmath9 . we have @xmath270 . by lemma [ pleated_surface_exists ] , after possibly reducing the genus of @xmath32 if necessary , we can assume without loss of generality that @xmath271 is a pleated surface . note that @xmath259 , by gauss - bonnet , and can be decomposed into @xmath272 ideal triangles . since the boundary of @xmath32 wraps @xmath39 times around @xmath9 , there is an equality @xmath273 . if @xmath173 is very long compared to the area of @xmath32 , the surface @xmath32 must become very thin along most of its boundary . choose @xmath2 which is smaller than the @xmath22-dimensional margulis constant @xmath274 . we defer the precise choice of @xmath2 for the moment . consider the thick - thin decomposition of @xmath32 with respect to @xmath2 . note that since @xmath32 is a manifold with boundary , it makes sense to take the thick - thin decomposition of the double @xmath275 and then restrict this decomposition to @xmath32 . each component of the thick region is relatively bounded in @xmath32 by proper arcs of length @xmath2 . think of the thick region as a hyperbolic surface , possibly with cusps which have been neutered " . we distinguish between the components of @xmath276 , which we call the _ sides _ of @xmath277 , and the components of @xmath278 ( each of which has length @xmath2 ) which we call the _ cusps_. it follows that for each such component @xmath277 there is an inequality @xmath279 this gives a lower bound on the length of @xmath280 . we would like to conclude that @xmath280 contains a long connected segment ; to do this , we need to bound the number of components of @xmath281 . the area of @xmath277 can be bounded from below from the genus of @xmath277 , and the number of boundary components . the worst case is when each @xmath277 is a neutered polygon with @xmath282 sides , in which case the euler characteristic of the corresponding un - neutered polygon is @xmath283 . the area of @xmath277 can be estimated from @xmath284 by subtracting area approximately @xmath2 for each neutered cusp . that is , @xmath285 from this we see that the number of components of @xmath281 is maximized when each @xmath277 is a neutered triangle , in which case there are exactly @xmath286 components , and the sum of the lengths of these components is at most @xmath287 . therefore , by the pigeonhole principle , it follows that there is a component @xmath288 of @xmath280 of length at least @xmath289 by construction , @xmath32 contains a rectangular strip @xmath290 of thickness @xmath291 with @xmath288 on one side . we denote the side opposite to @xmath288 by @xmath292 . we call @xmath288 and @xmath292 the _ long _ sides of @xmath290 . because @xmath32 is oriented , the orientations on opposite sides of @xmath290 are anti - aligned " . we lift @xmath290 to the universal cover @xmath132 , and by abuse of notation refer to the lifted rectangle as @xmath290 . the sides @xmath293 of @xmath290 are contained in geodesics @xmath294 which are axes for elements @xmath295 which are both in the conjugacy class @xmath11 . note that the translation lengths of @xmath296 on @xmath294 respectively are both equal to @xmath61 , and moreover these elements act by translating in nearly opposite directions . we let @xmath297 be such that @xmath298 . let @xmath142 be the midpoint of @xmath288 , and let @xmath225 be a point on the opposite side of @xmath290 with @xmath299 . suppose further that @xmath300 it follows that @xmath301 and there is @xmath302 with @xmath303 and therefore @xmath304 . since @xmath305 and @xmath306 are geodesics , @xmath307 and therefore @xmath308 and we can estimate @xmath309 similarly we estimate @xmath310 . see figure [ comparison_quad ] . in the figure , the axes @xmath26 and @xmath311 are both roughly vertical . the element @xmath312 translates points roughly downwards , and @xmath313 translates points roughly upwards . @xmath142 at 110 210 @xmath225 at 45 210 @xmath314 at 20 380 @xmath315 at -30 355 @xmath177 at 120 355 @xmath316 at 120 175 @xmath317 at 220 140 translates the midpoint @xmath142 a small distance , title="fig : " ] if we choose @xmath318 less than an @xmath1-dimensional margulis constant @xmath8 then @xmath319 and @xmath320 must commute . there are two possibilities , which break up into subcases . * case i : * @xmath321 and @xmath322 are hyperbolic with the same axis . in this case , since they are conjugate , they are either equal or inverse . * case ia : * @xmath323 . in this case @xmath312 and @xmath313 commute , and since they are conjugate , they are equal or inverse . but @xmath312 and @xmath313 translate their axes in almost opposite directions , so they can not be equal ; hence we must have @xmath324 , and therefore @xmath325 which implies that @xmath326 has order @xmath22 , which is impossible in a hyperbolic manifold group . * case ib : * @xmath327 . in this case @xmath328 and therefore @xmath329 which we have already seen is impossible in a hyperbolic manifold group . * case ii : * @xmath321 and @xmath322 are parabolic with the same fixed point @xmath330 . in this case , @xmath331 is parabolic with fixed point @xmath332 . but @xmath333 which has fixed point @xmath334 , so @xmath335 . since @xmath312 is hyperbolic , we have constructed a hyperbolic and a parabolic element in @xmath4 with a common fixed point ; this is well - known to violate discreteness , see e.g. page 19 , @xcite for details . in every case we obtain a contradiction , and therefore we must have @xmath336 putting this together with our earlier inequality , we obtain @xmath337 rearranging this gives @xmath338 the right hand side is a constant which depends only on the size of a margulis constant in dimension @xmath1 . if @xmath59 is sufficiently small , we can choose pleated surfaces @xmath32 for which the ratio @xmath339 is sufficiently large , and therefore we obtain an upper bound on @xmath61 which goes to @xmath107 as @xmath340 as claimed . [ optimal_estimate_remark ] by lemma [ seifert_estimate_scl ] and theorem a , we know that in dimension @xmath6 the dependence of the optimal @xmath3 on @xmath2 is given by @xmath341 for small @xmath2 . similarly , in any fixed dimension @xmath1 there are inequalities @xmath342 again for sufficiently small @xmath2 , but it is not clear what the optimal @xmath3 is . the first inequality arises here since a hyperbolic @xmath6-manifold group acts isometrically on @xmath132 by stabilizing a totally geodesic @xmath6-dimensional subspace . the second inequality follows from remark [ reznikov_remark ] . on the other hand , the dependence of @xmath3 on @xmath2 is not proper , even in dimension @xmath22 , as illustrated in example [ example_in_surface ] . in particular , the constant @xmath47 is bounded above by some universal finite bound , independent of the dimension @xmath1 . this universal bound is the `` homological '' margulis constant ; in the next subsection we give an explicit estimate for this constant , and show that it is contained between @xmath5 and @xmath72 . let @xmath0 be a closed hyperbolic manifold , of any dimension . let @xmath48 be the first accumulation point for stable commutator length on conjugacy classes . then @xmath49 we use the same setup and notation as in the proof of theorem c. since @xmath0 is a closed hyperbolic manifold , there are only finitely many conjugacy classes represented by geodesics shorter than any given length . so we suppose @xmath11 is a conjugacy class represented by a geodesic @xmath9 which is sufficiently long " ( in a sense to be made precise in a moment ) . we choose @xmath2 and find a segment @xmath288 , as in the proof of theorem c , and suppose we have @xmath343 ( note the missing factor of @xmath22 ) . we choose @xmath142 to be one of the endpoints of @xmath288 , so that @xmath309 since @xmath0 is fixed , there is some @xmath2 such that @xmath318 is smaller than the translation length of any nontrivial element in @xmath4 . hence @xmath344 . but this means @xmath325 , and @xmath326 has order @xmath22 , which is impossible in a hyperbolic manifold group . contrapositively , this means that we must have @xmath345 and therefore , just as in the proof of theorem c , we obtain @xmath346 in contrast to the case of theorem c , the right hand side definitely depends on the manifold @xmath0 . nevertheless , for fixed @xmath0 , it is a constant , and we see that for @xmath9 sufficiently long , @xmath347 can not be much smaller than @xmath5 . this establishes the lower bound in the theorem . to establish the upper bound , observe that since @xmath0 is a closed hyperbolic manifold , @xmath4 contains many nonabelian free groups . in fact , if @xmath95 are arbitrary noncommuting elements of @xmath4 , sufficiently high powers of @xmath11 and @xmath312 generate a free group , by the ping - pong lemma . this copy of @xmath348 is quasi - isometrically embedded , and by passing to a subgroup , one obtains quasi - isometrically embedded copies of free groups of any rank . let @xmath32 be a once - punctured hyperbolic surface of genus @xmath22 . then @xmath349 is free of rank @xmath350 . let @xmath351 be a quasi - isometric injection , as above . let @xmath25 be a once - punctured torus , embedded as a subsurface of @xmath32 , and let @xmath11 be the conjugacy class in @xmath349 corresponding to @xmath352 . by construction , @xmath353 . moreover , under the action of the mapping class group of @xmath32 , the orbit of @xmath11 is infinite . so there are infinitely many conjugacy classes @xmath354 in @xmath349 with @xmath355 . if we represent the @xmath354 by geodesics in @xmath32 , their lengths grow without bound . since @xmath82 is a quasi - isometric embedding , the geodesics in @xmath0 corresponding to the elements @xmath356 have unbounded length , and therefore they correspond to infinitely many conjugacy classes in @xmath4 . since @xmath59 is monotone under homomorphisms , we have constructed infinitely many conjugacy classes in @xmath4 with @xmath357 . if @xmath0 is a finite volume hyperbolic @xmath6-manifold , possibly with cusps , the first accumulation point @xmath48 still satisfies @xmath358 . for , if @xmath9 is a sufficiently long closed geodesic in @xmath0 , we can find a dehn filling @xmath359 of @xmath0 in which the image of @xmath9 is also homotopic to an arbitrarily long closed geodesic . the stable commutator length of a sufficiently long geodesic in @xmath359 is at least @xmath360 for any positive @xmath2 , by theorem d. moreover , the stable commutator length can only go down under dehn filling , since stable commutator length is non - increasing under homomorphisms ; the claim follows . if @xmath0 is a finite volume hyperbolic @xmath1-manifold , the argument of theorem c shows directly that @xmath361 . but it seems reasonable to expect that @xmath358 in this case too . it is a well - known conjecture that there is a uniform positive lower bound on the length of a closed geodesic in an arithmetic hyperbolic manifold . this is related to lehmer s conjecture in number theory ; c.f . this motivates the following for each @xmath1 there is a constant @xmath362 such that for any arithmetic hyperbolic @xmath1-manifold @xmath0 and any non - parabolic @xmath363 , there is an inequality @xmath364 compare with the following example , due to agol . in @xcite , agol shows that in any dimension @xmath1 , if there is an arithmetic hyperbolic @xmath1-manifold defined by a quadratic form which is gferf ( i.e. subgroup separable on geometrically finite subgroups ) then there exist finite volume hyperbolic @xmath1-manifolds with arbitrarily short geodesics . such subgroups exist for all @xmath365 . the examples are obtained from arithmetic manifolds with lots of immersed totally geodesic hypersurfaces by mutation . [ inverse_conjugate_vanishes ] if `` manifold '' is replaced by `` orbifold '' in the statement of theorem c or theorem d , then there are a number of places where the argument must be modified . in an orbifold , elliptic elements with a common fixed point do not necessarily commute . moreover , we must allow the possibility that @xmath366 in which case @xmath312 is conjugate to @xmath367 . in any group , if there is @xmath326 such that @xmath368 then @xmath369 $ ] and the stable commutator length of @xmath312 vanishes . in particular , an element @xmath11 in a hyperbolic orbifold group is either conjugate to its inverse or satisfies @xmath370 where the bound may be estimated in terms of the translation length of @xmath11 and the order of torsion elements in the group . theorem c and theorem d admit natural generalizations to ( word- ) hyperbolic groups , and to groups acting on @xmath3-hyperbolic spaces , such as mapping class groups and amalgamated free products . these results will appear in a forthcoming paper @xcite . the following discussion is included for the benefit of the reader , and is not logically necessary for the rest of the paper . if an @xmath1th power of an element in @xmath4 can be expressed as a product of @xmath39 commutators , one obtains an immersed @xmath22-complex @xmath371 obtained from wrapping the boundary of a genus @xmath39 surface @xmath1 times around the corresponding geodesic @xmath9 . the map of @xmath372 into @xmath0 is almost never @xmath51-injective , but when it is , one can derive a global relationship between @xmath373 and @xmath61 . [ injective_wrap_estimate ] let @xmath374 and let @xmath375 be the group defined in terms of generators and relations by @xmath376 \rangle\ ] ] let @xmath0 be a complete hyperbolic @xmath6-manifold with @xmath377 . then there is a constant @xmath378 such that if @xmath9 is the geodesic in @xmath0 corresponding to the conjugacy class @xmath379 , there is an estimate @xmath380 this is actually an effective version of a special case of thurston s broken windows only theorem " . we give a sketch of a proof ; for details , consult @xcite . let @xmath371 be an embedded @xmath22-complex built out of the image of a suitable pleated surface @xmath32 by wrapping the boundary @xmath39 times around the geodesic @xmath9 . by gauss - bonnet , @xmath381 in the universal cover @xmath382 the preimage @xmath383 is a tree of hyperbolic planes in its intrinsic metric , with valence @xmath384 along the branch locus , which is equal to the preimage @xmath385 of @xmath9 . a random walk on @xmath383 pushes down to a random walk on @xmath372 . since the geodesic curvature of @xmath372 is uniformly negative , a random walk crosses @xmath9 with frequency proportional to the length of @xmath9 . it follows that there is a constant @xmath378 independent of @xmath39 and @xmath1 such that for a random geodesic @xmath288 in @xmath383 of length @xmath25 , the number of intersections of @xmath288 with @xmath385 satisfies @xmath386 since @xmath383 has valence @xmath1 along each such sheet , this implies that the area of the ball of radius @xmath25 about a point @xmath387 satisfies @xmath388 on the other hand , since @xmath372 is immersed in @xmath0 , this growth rate must be less than the growth rate of the volume of hyperbolic @xmath6-space , or else @xmath383 would accumulate on itself . it follows that we can estimate @xmath389 and we are done . let @xmath32 be the connect sum of three projective planes . a presentation for @xmath349 is @xmath390 = c^2 \rangle\ ] ] the element @xmath326 is represented by a simple one - sided geodesic @xmath9 on @xmath32 . geometrically , @xmath32 is obtained from a once - punctured torus with geodesic boundary by mapping the boundary to @xmath9 by a double covering . a hyperbolic once - punctured torus exists with boundary of any given positive length ; it follows that the length of @xmath9 may be specified arbitrarily . this example shows that the restriction @xmath391 is necessary in theorem [ injective_wrap_estimate ] . the results in [ surgery_section ] show that de rham quasimorphisms give good lower bounds on @xmath59 for short geodesics contained in embedded tubes . it is natural to ask how rare de rham quasimorphisms are amongst all quasimorphisms in general . let @xmath0 be a hyperbolic manifold , and let @xmath43 . let @xmath9 be the unique geodesic in @xmath0 in the free homotopy class of the conjugacy class of @xmath11 . the _ de rham length _ of @xmath11 , denoted @xmath53 , is the supremum @xmath392 by theorem [ bavard_theorem ] and lemma [ form_is_quasimorphism ] there is an inequality @xmath58 for any @xmath393 . de rham length is naturally related to norms from geometric measure theory , especially the _ whitney flat norm _ ( see @xcite,@xcite ) . the following construction is due to thurston . [ small_area_example ] let @xmath394 be an ideal quadrilateral in @xmath132 , @xmath7 with the property that opposite vertices are very close together in @xmath395 , and the ideal simplex they span has simplex parameter whose imaginary part is much bigger than its real part ( this requires @xmath7 ) . then @xmath394 spans a disk with arbitrarily small area , made from a bisecting `` square '' @xmath32 together with four `` cusps '' @xmath396 . see figure [ small_disk ] . @xmath397 at 103 210 @xmath398 at 100 70 @xmath399 at 215 105 @xmath400 at 217 190 @xmath32 at 108 140 and the @xmath401 spans @xmath394,title="fig : " ] for any finite volume hyperbolic @xmath1-manifold @xmath0 we can immerse @xmath394 in @xmath0 by projecting @xmath402 . now , for generic choices of endpoints , the four boundary geodesics are almost perfectly equidistributed in @xmath403 , and therefore after an arbitrarily small perturbation , we can glue up opposite faces to product a once - punctured torus @xmath25 with arbitrarily small area . for generic translation lengths along the sides , the hyperbolic structure on the torus is incomplete , and can be completed by adding a geodesic boundary @xmath9 . since the matching can be performed with infinitely many translation lengths along the sides , @xmath61 can be chosen to be arbitrarily long . this construction produces infinitely many closed geodesics @xmath9 in @xmath0 bounding surfaces of arbitrarily small area . by stokes theorem , if @xmath43 is a conjugacy class corresponding to such a @xmath9 , then @xmath53 is arbitrarily small . on the other hand , by theorem c , the stable commutator length @xmath14 can not be too small , when @xmath61 is large . this proves the following theorem . let @xmath0 be a finite volume complete hyperbolic @xmath1-manifold with @xmath7 . then for any @xmath62 there are infinitely many conjugacy classes @xmath43 with @xmath63 give an algebraic characterization of conjugacy classes @xmath43 with small @xmath60 . for completeness , we discuss lower bounds on @xmath59 in @xmath4 for @xmath0 a @xmath6-manifold with one of the seven ( non - hyperbolic ) @xmath6-dimensional geometries . a typical sol manifold @xmath0 has a fundamental group @xmath407 where the conjugation action of the generator @xmath269 of @xmath408 on @xmath409 is given by an anosov matrix @xmath410 ( i.e. one with @xmath411 ) . in this case , the kernel of the map @xmath412 is equal to the normal @xmath409 subgroup . so suppose @xmath413 . let @xmath414 be two independent eigenvectors of @xmath410 over @xmath415 . we can express @xmath11 as a product of elements in @xmath409 @xmath416 where @xmath417 is almost a multiple of @xmath418 , @xmath419 is almost a multiple of @xmath420 , and there is an inequality @xmath421 for some constant @xmath277 . for suitable non - negative integers @xmath422 , the conjugates @xmath423 and @xmath424 are both words of uniformly bounded length in @xmath409 with respect to any fixed generating set , so @xmath11 can be written as a product of a bounded number of commutators times something of size definitely smaller than that of @xmath11 . by induction , we obtain an estimate @xmath425 , and therefore @xmath405 . the typical case is that @xmath426 where @xmath10 is a hyperbolic surface group . if @xmath43 has nontrivial image in the @xmath408 factor , then @xmath427 otherwise , @xmath11 is contained in @xmath10 . in an orientable hyperbolic surface group , @xmath428 by example [ free_example_bound ] . there is also an action of @xmath4 on @xmath108 which comes from the universal covering @xmath125 lifting the natural action of @xmath431 on @xmath432 . for this action , poincar s _ rotation number _ ( example [ rotation_example ] ) is a natural homogeneous quasimorphism with defect @xmath36 . this quasimorphism takes the value @xmath36 on the generator of the center of @xmath4 , and takes some value @xmath29 where @xmath433 is coprime to @xmath225 on any element which maps to a torsion element of order @xmath225 in @xmath10 . 99 i. agol , _ bounds on exceptional dehn filling _ , geom . * 4 * ( 2000 ) , 431449 i. agol , _ systoles of hyperbolic @xmath350-manifolds _ , eprint , math.gt/0612290 j. barge and e. ghys , _ surfaces et cohomologie borne _ , invent . * 92 * ( 1988 ) , 509526 c. bavard , _ longueur stable des commutateurs _ , lenseign . * 37 * ( 1991 ) , 109150 g. besson , _ sminar de cohomologie borne _ , ens lyon ( 1988 ) k. boroczky , _ packing of spheres in spaces of constant curvature _ , acta math . acad . hungaricae , * 32 * ( 1978 ) , 243261 r. brooks , _ some remarks on bounded cohomology _ , _ riemann surfaces and related topics : proceedings of the 1978 stony brook conference ( state univ . new york , stony brook , n.y . , 1978 ) _ , 5363 , ann . of math . , * 97 * , _ princeton univ . press , princeton , n.j . _ , 1981 d. calegari _ scl _ , monograph , in preparation d. calegari and k. fujiwara , _ stable commutator length in word - hyperbolic groups _ , eprint , math.gr/0611889 l. comerford and c. edmunds , _ genus of powers in a free group _ , in geometric group theory ( ed . r. charney , m. davis , m. shapiro ) ohio state university msri pub . * 3 * , 6771 , de gruyter berlin 1995 m. culler , _ using surfaces to solve equations in free groups _ , topology * 20 * ( 1981 ) , no . 2 , 133145 a. duncan and j. howie , _ the genus problem for one - relator products of locally indicable groups _ , math . z. * 208 * ( 1991 ) , 225237 h. federer , _ geometric measure theory _ , die grund . * 153 * springer - verlag new york 1969 m. gromov , _ volume and bounded cohomology _ , inst . hautes tudes sci . 56 , ( 1982 ) , 5 - 99 c. hodgson and s. kerckhoff , _ universal bounds for hyperbolic dehn surgery _ , ann . of math . ( 2 ) * 162 * ( 2005 ) , no . 1 , 367421 d. kazhdan and g. margulis , _ a proof of selberg s hypothesis _ , math . sb . * 75 * ( 117 ) ( 1968 ) , 163168 m. korkmaz , _ stable commutator length of a dehn twist _ michigan math . j. * 52 * ( 2004 ) , 2331 d. kotschick , _ quasi - homomorphisms and stable lengths in mapping class groups _ , * 132 * ( 2004 ) , 31673175 d. h. lehmer , _ factorizations of certain cyclotomic functions _ , ann . of math . * 34 * ( 1933 ) , 461479 b. maskit , _ kleinian groups _ , die grund . * 287 * springer - verlag new york 1988 w. neumann and a. reid , _ arithmetic of hyperbolic manifolds _ , in topology 90 ( columbus , oh , 1990 ) ohio state university msri pub . * 1 * , 273310 , de gruyter berlin 1992 w. neumann and d. zagier , _ volumes of hyperbolic @xmath6-manifolds _ , topology * 24 * ( 1985 ) , no . 3 , 307332 h. poincar , _ sur les courbes dfinies par les quations diffrentialles _ , j. de math . * 1 * ( 1885 ) 167 a. reznikov , _ the volume and the injectivity radius of a hyperbolic manifolds _ , topology * 34 * ( 1995 ) , no . 2 , 477479 w. thurston , _ geometry and topology of @xmath6-manifolds ( a.k.a . `` thurston s notes '' ) _ notes from princeton university , circa 1979 w. thurston , _ hyperbolic structures on @xmath6-manifolds iii : deformations of manifolds with incompressible boundary _ , eprint , math.gt/9801058 h. whitney , _ geometric integration theory _ , princeton university press , princeton n. j. , 1957

lattice networks with dissipative interactions can be used to describe the mechanics of discrete micro- or meso - structures of materials such as 3d - printed structures and foams , or more generally heterogeneous materials . this contribution deals with the crack initiation and propagation in such materials and focuses on an adaptive multiscale approach that captures the spatially evolving fracture . lattice networks naturally incorporate non - locality , large deformations , and dissipative mechanisms taking place directly inside fracture zones . because the physically relevant length scales for fracture are significantly larger than those of individual interactions , discrete models are computationally expensive . the quasicontinuum ( qc ) method is a multiscale approach specifically constructed for discrete models . this method reduces the computational cost by fully resolving the underlying lattice only in regions of interest , while coarsening elsewhere . in this contribution , an incremental energy potential corresponding to the full lattice model with damageable interactions is constructed for engineering - scale predictions . subsequently , using a variational qc approach the incremental energy along with governing equations for the upscaled lattice network are derived . in order to deal with the spatially evolving fracture zone in an accurate , yet efficient manner , an adaptive scheme is proposed that utilizes the underlying lattice . implications induced by the adaptive procedure are discussed from the energy - consistency point of view , and theoretical considerations are demonstrated on two examples . the first one serves as a proof of concept , illustrates the consistency of the adaptive schemes , and presents errors in energies compared to the fully - resolved solution . the second one demonstrates the performance of the adaptive qc scheme for a more complex problem for which the fully resolved solution is not accessible . lattice networks , quasicontinuum method , damage , adaptivity , variational formulation , multiscale modelling

1604.04754

lattice networks are frequently employed to describe the mechanical response of materials and structures that are discrete by nature at one or more length scales , such as 3d - printed structures , woven textiles , paper , or foams . for lattice networks representing fibrous microstructures for instance , individual fibres can be identified with one - dimensional springs or beams . further examples are the models of e.g. @xcite , @xcite , @xcite , and @xcite . the reason why lattice models may be preferred over conventional continuum theories and finite element ( fe ) approaches , is twofold . first , the meaning and significance of the physical parameters associated with individual interactions in the lattice networks is easy to understand , whereas the parameters in constitutive continuum models represent the small - scale mechanics only in a phenomenological manner . an example is the young s modulus or ultimate strength of a spring or beam ( fibre or yarn ) versus that of the network . second , the formulation and implementation of lattice models is generally significantly easier compared to that of alternative continuum models . large deformations , large yarn re - orientations , and fracture are for instance easier to formulate and implement ( cf . e.g. the continuum model of @xcite that deals with large yarn re - orientations ) . thanks to the simplicity and versatility of lattice networks , they are furthermore used for the description of heterogeneous cohesive - frictional materials . the reason is that discrete models can realistically represent distributed microcracking with gradual softening , implement material structure with inhomogeneities , capture non - locality of damage processes , and reflect deterministic or stochastic size effects . examples of the successful use of lattice models for such materials are given by @xcite , @xcite , @xcite , and @xcite . as lattice models are typically constructed at the meso- , micro- , or nano - scale , they require reduced - model techniques to allow for application - scale simulations . a prominent example is the quasicontinuum ( qc ) method , which specifically aims at discrete lattice models . the qc method was originally introduced for conservative atomistic systems by @xcite and extended in numerous aspects later on , see e.g. @xcite . subsequent generalizations for lattices with dissipative interactions ( e.g. plasticity and bond sliding ) were provided by @xcite . in principle , the qc is a numerical procedure that can deal with local lattice - level phenomena in small regions of interest , whereas the lattice model is coarse grained in the remainder of the domain . the aim of this contribution is to develop a qc framework that can deal with the initiation and subsequent propagation of damage and fracture in the underlying structural lattice model . because such a phenomenon tends to be a highly localized and rather unstable process , sensitive to local mesh details , the qc framework must fully refine in critical regions _ before _ any damage occurs in order to capture the physics properly , cf . [ sect : introduction : fig:1 ] . moreover , as the entire problem is evolutionary , the location of the fully resolved region must evolve as well , which requires an adaptive qc framework . several previous studies have also focused on adaptivity in qc methodologies , but they were always dealing with atomistic lattice models at the nano - scale , see e.g. @xcite , @xcite , @xcite . contrary to that , this contribution focuses on structural lattice networks for materials with discreteness at the meso - scale . the qc approach aimed for is schematically presented in fig . [ sect : introduction : fig:1 ] . the macro - scale fracture emerges as individual interactions failures at the lattice level . their damage leads to strain - softening and hence , the fracture process zone remains spatially localized . consequently , only the crack tip and the process zone have to be fully resolved . the displacement fluctuations elsewhere remain small , allowing for efficient interpolation and coarse graining . due to the spatial propagation of the crack front through the system of interest , available qc formulations need to be generalized to involve dissipation induced by damage , and an adaptive meshing scheme that includes a suitable marking strategy needs to be developed . the theoretical framework employed in this contribution is closely related to our variational qc formulation for hardening plasticity discussed in @xcite . the present work can be viewed as an extension towards lattices with localized damage and with an adaptive refinement strategy . in principle , the overall procedure is based on the variational formulation by @xcite , developed for _ rate - independent _ inelastic systems . this variational formulation employs at each time instant @xmath0 an incremental potential energy @xmath1 , that can be minimized with respect to the observable ( kinematic ) as well as the internal ( history , dissipative ) variables , i.e. @xmath2 where @xmath3 denotes a generalized state variable . hence , this formulation is different from the one employed in the virtual - power - based qc framework of @xcite , which is based on the virtual - power statement of the lattice model in combination with a coleman - noll procedure . note that earlier variational formulations for rate - independent systems can be found e.g. in @xcite , @xcite , @xcite , @xcite , and @xcite . the theoretical concepts of the variational formulation and its application to damaging lattice models are discussed in section [ sect : varform ] . after the incremental energy is presented for the full lattice system , two reduction steps can be applied to it in analogy to the standard qc framework , see e.g. @xcite , @xcite , @xcite . in the first step , _ interpolation _ constrains the displacements of all atoms according to the displacements of a number of selected representative atoms , or _ repatoms _ for short . this procedure reduces the number of degrees of freedom drastically . in the second step , only a small number of atoms is sampled to approximate the exact incremental energy @xmath1 , its gradients , and hessians , analogously to the numerical integration of fe technology . this step is referred to as _ summation _ , and it entails also a significant reduction of the number of internal variables . together , the two steps yield a reduced state variable @xmath4 and an approximate incremental energy @xmath5 . in section [ sect : qc ] , a more detailed discussion about qc techniques , adaptive modelling , marking strategy , and mesh refinement will be provided . consequences of the proposed mesh refinement strategy will be discussed also from the energetic point of view . the minimization problem provides governing equations presented in section [ sect : sol ] , where a suitable solution strategy is described . incremental energy minimization procedures often employ some version of the alternating minimization ( am ) strategy , see e.g. @xcite . the approach used here , however , minimizes the so - called _ reduced energy _ , i.e. the energy potential @xmath1 with eliminated internal variables , cf . @xcite or @xcite . as a result , the overall solution process simplifies and is more efficient compared to the am approach . in section [ sect : examples ] , the proposed theoretical developments are first applied to an l - shaped plate test . the force - displacement diagrams and crack paths predicted with the adaptive qc approach are compared to those predicted with full lattice computations . further , the energy consistency during the entire evolution is assessed and the errors in energies are discussed . the second numerical example focuses on the antisymmetric four - point bending test , described e.g. in @xcite . it demonstrates the ability of the adaptive qc scheme to predict nontrivial , curved crack paths . finally , this contribution closes with a summary and conclusions in section [ sect : conclusion ] . in this section , we recall the general variational theory for rate - independent systems , followed by the geometric setting , description of state variables , and by the construction of energies . the entire exposition will be confined to 2d systems , but the extension to 3d is straightforward . let us note in addition that throughout this work the term `` atoms '' will be used to refer to individual lattice nodes or particles , consistently with the original qc terminology developed for atomistic systems . the evolution of a system within a time horizon @xmath6 $ ] is considered to be _ quasistatic _ and _ rate - independent _ , so the ( pseudo- ) time @xmath7 $ ] can be arbitrarily rescaled without any influence on the results . the system of interest is fully specified by the total free ( helmholtz type ) energy @xmath8\times\mathscr{q}\rightarrow\mathbb{r}$ ] and by the dissipation distance @xmath9 , @xmath10 . the dissipation distance reflects the minimum energy dissipated by the continuous transition between two consecutive states @xmath11 and @xmath12 . both functionals are defined for states @xmath13 , where @xmath14 is a suitable state - space , @xmath15 is a set of observable kinematic variables , and @xmath16 is a set of internal variables describing the inelastic processes . a function @xmath17 \rightarrow \mathscr{q}$ ] is called an _ energetic solution _ of the energetic rate - independent system @xmath18 if it satisfies the following _ stability condition _ and _ energy balance _ for all @xmath19 $ ] @xmath20 along with the _ initial condition _ @xmath21 for further details see e.g @xcite . the free energy @xmath22 can be expressed as @xmath23 where @xmath24\times\mathscr{q}\rightarrow\mathbb{r}$ ] is the internal free energy , @xmath25 \rightarrow \mathscr{r}^ * \times \mathscr{z}^*$ ] represents an external loading ( which satisfies some additional conditions related to the boundedness of @xmath22 not specified here , for further details see @xcite ) , @xmath26 and @xmath27 are spaces dual to @xmath28 and @xmath29 , and @xmath30 denotes the corresponding duality pairing . in , we have introduced the time - integrated power of external forces @xmath31 and the dissipated energy @xmath32 where the supremum is taken over all @xmath33 and all partitions of the time interval @xmath34 $ ] . upon introducing a discretization of the time interval @xmath6 $ ] in the form @xmath35 , the time - discrete energetic solution can be constructed by an _ incremental problem _ defined as @xmath36 cf . also eq . . in , each step is realized as a minimization problem of the following _ incremental energy _ : @xmath37 a characteristic difficulty related to time - discrete energetic solutions is that they are constructed by a recursive global minimization , which is computationally cumbersome for non - convex energies . yet it is reasonable and standard to assume that solutions of are associated with local minima that satisfy the energy balance . such a strategy , in combination with indirect displacement solution control , is adopted also in this work . local minimization entails that any additional requirements for @xmath38 can be dropped , cf . eq . . for further discussion about local versus global minimization see e.g. @xcite note that within the energetic framework , the solution @xmath39 may be a discontinuous function of time , leading to jumps in internal variables and energy quantities . to specialize the abstract framework to lattice networks , we first introduce the geometric setting and notation , cf . [ disslatt : fig:1 ] . the domain in a reference configuration @xmath40 contains @xmath41 atoms , collected in an index set @xmath42 . the spatial position of each atom @xmath43 is specified by its position vector @xmath44 . since only regular networks with nearest neighbour interactions are considered , the atoms positions can be expressed as linear combinations of the primitive lattice vectors ( in analogy to the bravais lattices ) . all position vectors @xmath45 are collected in a column matrix @xmath46^\mathsf{t}$ ] , @xmath47 . throughout this contribution , greek indices refer to atom numbers whereas latin indices are reserved for spatial coordinates and other integer parametrizations . the nearest neighbours of an atom @xmath48 are furthermore stored in a set @xmath49 . in contrast to atomistic lattices , the nearest neighbours of each atom do not change in time . the initial distance between two neighbouring atoms @xmath48 and @xmath50 and the set of all initial distances between neighbouring atoms within the network are defined as [ subsect : disslatt : eq:1 ] @xmath51 where @xmath52 is the euclidean norm . since @xmath53 , the set @xmath54 in consists of @xmath55 components , where @xmath55 is the number of all interactions collected in an index set @xmath56 . the above introduced symbol @xmath57 will be employed below in two contexts . first , in the context of atoms , the symbol @xmath58 measures the distance in the reference configuration between two atoms @xmath59 , as used in eq . . second , in the context of interactions , the same symbol @xmath58 measures the length of the @xmath60-th interaction in the reference configuration , @xmath61 , @xmath62 , with end atoms @xmath63 . a similar convention holds also for other physical quantities . and current configuration @xmath64 . ] as the body deforms , the atoms in the reference configuration transform to a current configuration @xmath65 . the deformed locations of all atoms are specified by their position vectors @xmath66 , @xmath67 . in analogy to @xmath68 , they are collected in a column matrix @xmath69^\mathsf{t}$ ] , @xmath70 , that represents also the abstract observable variable . the distance measure between two atoms @xmath71 and the set of all distances @xmath72 are introduced in the same manner as for the reference configuration , cf . due to dirichlet boundary conditions , @xmath73 is a function of time and forms a manifold in @xmath74 . each interaction is endowed with one internal variable , a damage variable @xmath75 , @xmath76 . for brevity , all @xmath75 are collected in a column matrix @xmath77 , that represents the abstract dissipative variable . recall that @xmath78 meaning that @xmath29 is identified with @xmath79 . the elastic part of the incremental energy reads , cf . also eq . , @xmath80 where the column matrix @xmath81 collects the prescribed external forces acting on lattice atoms . note that in agreement with the notation used in eqs . and , the hatted variables @xmath82 represent arbitrary admissible values , whereas non - hatted variables represent the minimizers of . the interatomic potential energy @xmath83 in reflects the recoverable part of the energy stored in all interactions , and is expressed in terms of pair potentials @xmath84 , \label{subsect : disslatt : eq:3}\ ] ] where the first equality holds due to the principle of interatomic potential invariance . assuming @xmath85 , the two quantities @xmath86 and @xmath87 in eq . ensure that interactions undergo damage only under tension and not in compression , i.e. @xmath88 can grow only if @xmath89 . the first term in the square brackets of eq . , i.e. the pair potential @xmath90 weakened by @xmath91 , reflects the elastic portion of the energy stored in a single interaction stretched to a length @xmath92 and damaged to the level of @xmath93 . note that for @xmath94 , the interaction is completely intact ( stores the full amount of energy ) , whereas for @xmath95 the interaction is fully damaged ( no energy can be stored in it ) . the second term in the square brackets of eq . contributes only in compression , and the pair potential @xmath90 represents the elastic part of the energy stored in a single interaction compressed to a length @xmath96 , independently of the level of damage @xmath93 . the energy dissipated by a single interaction , @xmath97 , during the evolution between two consecutive states @xmath98 and @xmath99 , is defined as @xmath100 where @xmath101 reflects the amount of energy dissipated during a unidirectional damage process up to a given state @xmath102 . consequently , @xmath103 must be increasing with @xmath104 and @xmath105 , where @xmath106 represents the energy dissipated by the complete failure process . for a general derivation and further details see sections [ subsect : fullsol ] , [ sect : examples ] , and [ sect : a ] . in definition , the value @xmath107 restricts the internal variables @xmath88 to be only non - decreasing functions of time . the global dissipation distance then simply collects contributions of all interactions , i.e. @xmath108 based on the formulations of @xmath22 and @xmath109 , it may be clear that the total incremental energy @xmath1 can be expressed in terms of the incremental energies of each interaction , @xmath110 , or in terms of the incremental site - energies of each atom , @xmath111 , [ subsect : disslatt : eq:7 ] @xmath112 in the form @xmath113 both expressions in eq . will be used later on for the full - lattice , and in a slightly adjusted form also for the qc , computations . specifically , for the minimization of the incremental energy @xmath1 ( or its approximation @xmath114 in eq . ) with respect to the kinematic variables @xmath115 , the definition via the site - energies will be employed . for the minimization with respect to the internal variables @xmath116 , the definition using the interaction energies will be used . this section extends the previously discussed theory to an adaptive qc scheme . first , the two standard qc steps interpolation in section [ subsect : interpol ] and summation in section [ subsect : summ]are applied to the incremental energy @xmath1 at a fixed time step @xmath0 . subsequently , a heuristic marking strategy and a mesh refinement algorithm are presented in sections [ subsect : adaptivity ] and [ subsect : refinement ] . finally , adaptivity is discussed from an energetic point of view in section [ subsect : meshenergy ] . according to the standard qc theory , we introduce @xmath117 _ repatoms _ stored in an index set @xmath118 , that behave in analogy to fe nodes . the kinematic variables of the remaining atoms , @xmath119 , are interpolated using finite element shape functions constructed between repatoms . this can be expressed as @xmath120 where @xmath121 represents the column matrix with the positions of all repatoms in an arbitrary admissible configuration , and where the interpolation matrix @xmath122 stores the basis vectors spanning @xmath123 column - wise . substitution of eq . in eq . entails that the incremental energy becomes a function of @xmath124 and @xmath125 , i.e. @xmath126 which reduces the number of degrees of freedom associated with the kinematic variable , i.e. from @xmath127 to @xmath128 . the minimization in with respect to @xmath129 then becomes a minimization over some subspace @xmath123 , which reduces the computational effort substantially if @xmath130 . in order to specify @xmath131 in more detail , one introduces the standard fe triangulation @xmath132 ( with elements @xmath133 ) of @xmath134 with piecewise affine shape functions inside triangles . in the region of interest , the triangulation fully recovers the underlying lattice , whereas it coarsens elsewhere . the interpolation matrix then attains the form @xmath135 where @xmath136 denotes the @xmath137-th element of the set @xmath138 according to its ordering , and @xmath139 represents the shape function associated with a repatom @xmath50 that is evaluated in the undeformed position of an atom @xmath48 . note that higher - order shape functions can also be used , cf . e.g. @xcite . the summation step involves the selection of a limited number of site - energies to sample the contributions of all atoms . the expression for the total incremental energy can be written as @xmath140 where @xmath141 denotes a set of @xmath142 sampling atoms . these atoms are chosen carefully in order to estimate the energy of the entire lattice , in analogy to numerical integration in fe implementations . sampling atom @xmath143 represents the contributions of @xmath144 atoms , including its own ( @xmath145 ) . with respect to fe technology , @xmath144 is equivalent to the weight of gauss integration point @xmath48 . because @xmath146 , the computational effort associated with the assembly of the energy , gradients , and hessians is substantially reduced due to summation . explicit instructions on how to choose the sampling atoms and how to compute their weights @xmath144 can be found e.g. in @xcite . in the examples section [ sect : examples ] , the so - called _ central summation _ rule will be used , cf . @xcite , which selects one atom at the triangles centres plus their vertices . similarly to @xmath147 , we introduce a set of @xmath148 sampling interactions stored in an index set @xmath149 , defined as all interactions connected to all sampling atoms ; in analogy to eq . , any duplicity is removed . consequently , the summation in eq . can be again expressed as a sum over all sampling interactions @xmath150 , i.e. in analogy to eq . , we can write @xmath151 where @xmath152 denote the weight factors corresponding to bonds rather than to atom sites . since all internal variables associated with bonds in @xmath153 become irrelevant , a reduced dissipative internal variable ( associated only with the sampling interactions ) @xmath154 can be introduced , where @xmath155 is identified with @xmath156 . this , in combination with the interpolation step , gives rise to the reduced state variable @xmath157 and to the reduced abstract state space @xmath158 . in sections [ subsect : interpol ] [ subsect : summ ] , a fixed triangulation @xmath132 of the reference domain @xmath134 has been assumed , meaning that @xmath159 was a function of time only due to the evolving dirichlet boundary conditions . for the lattice of interest , however , the location of damage growth evolves during each computation . as the aim is to let the initiation of damage occur only in fully resolved regions , the triangulation has to evolve as well . for this purpose , the adaptive procedure summarized in alg . [ subsect : adaptivity : alg:1 ] changes the dimensionality of @xmath160 at each time step @xmath0 due to mesh refinement . consequently , @xmath161 , @xmath162 , @xmath163 , @xmath164 , and @xmath165 become functions of time too . 1 . initialize the system : apply initial condition @xmath166 and construct initial ( coarse ) mesh @xmath132 with required information , e.g. @xmath167 , @xmath168 , @xmath169 , @xmath170 . 2 . * for @xmath171 apply the boundary conditions at time @xmath0 , @xmath172 , @xmath173 , @xmath174 , @xmath175 , @xmath176 , etc . 2 . equilibrate the unbalanced system , i.e. solve for @xmath177 in using eq . substituted into the approximate incremental energy @xmath178 defined in eq . . 3 . for each element @xmath179 evaluate its _ indicator _ and decide for possible refinement , cf . condition ; all elements @xmath180 marked for refinement are collected in a set @xmath181 . if * @xmath182 , refine current mesh , update the system information @xmath183 , @xmath162 , @xmath163 , @xmath164 , @xmath165 , etc . , and return to ( ii ) since the refined system is unbalanced . + * else if * @xmath184 , the mesh has converged ; proceed to ( v ) . store relevant output data for @xmath185-th time step : @xmath186 , @xmath183 , @xmath162 , @xmath163 , @xmath164 , @xmath165 , etc . proceed to the next time step . * end * given a triangulation @xmath183 at a time step @xmath0 , a _ marking strategy _ decides in stage ( iii ) of alg . [ subsect : adaptivity : alg:1 ] which triangles should be refined . this procedure consists of the evaluation of a _ mesh indicator _ for each triangle that is not fully refined yet . as the damage phenomenon is sensitive to local mesh details , it is convenient to fully refine the triangulation in critical regions _ before _ any damage occurs there . to this end , each coarse element @xmath180 is endowed with a subset of sampling interactions @xmath187 . for each of these interactions that is moreover loaded in tension ( i.e. @xmath89 , since damage is assumed to evolve only in tension recall section [ subsect : energies ] ) , the elastic energy is computed . as the indicators have to be evaluated only for coarse triangles , the elastic energies usually reduce to pair potential evaluations , i.e. to @xmath188 . then , a triangle @xmath180 is marked for refinement if at least one interaction from @xmath189 has its pair potential evaluation @xmath188 higher than a given threshold @xmath190 , where @xmath191 specifies a certain safety margin and @xmath192 denotes the stored elastic energy threshold in the bond @xmath57 at which damage starts to evolve . the safety margin @xmath193 serves to control the accuracy of the qc method and the mesh refinement algorithm . furthermore , in order to make sure that no damage occurs prior to full refinement , the associated internal variables are verified : if @xmath194 for some interaction @xmath195 , this triangle is marked for refinement , irrespective of the energy condition . the above - described procedure is formalized as follows . for a coarse triangle @xmath196 , the set @xmath189 is defined as all sampling bonds that are situated at least partly inside the element @xmath180 , is considered as a closed set . consequently , the bonds and atoms lying on an element s edges or vertices are contained in that element . ] i.e. as the bonds that * are connected to the central sampling atom of @xmath180 , @xmath197 , if a central sampling atom exists , * are connected to the discrete sampling atoms of @xmath180 , @xmath198 i.e. vertex or edge atoms , and their intersections with given triangle @xmath180 have strictly positive lengths . the energy criterion is expressed as @xmath199 and the triangle indicator is evaluated as @xmath200 note that the proposed marking strategy is based on heuristic considerations and supposedly also performs well for diffuse phenomena such as hardening plasticity . in section [ sect : examples ] we will show that this methodology performs well for the lattice of interest if the safety margin @xmath193 is relatively small ( @xmath201 ) . as an alternative to mesh indicators , one could use _ error estimators _ such as the _ goal - oriented _ error estimator presented e.g. in @xcite , @xcite , or @xcite . this approach will not be pursued further in this work and is left as a possible future challenge . elaborating on stage ( iv ) of alg . [ subsect : adaptivity : alg:1 ] , the current mesh @xmath183 is refined based on the marked triangles @xmath202 . since the underlying lattice is regular , it is convenient to employ a right - triangulated initial mesh @xmath132 and a self - similar mesh refinement . this may be considered reasonable because right - angled triangles lead to significantly smaller summation errors , cf . @xcite , and because these triangles also naturally avoid artificial deformations in transition regions that would lead to spurious mesh refinements and non - physical evolution of internal variables . for the refinement , the standard @xcite algorithm conserving _ non - degeneracy _ , _ conformity _ , and _ smoothness _ , is used for each marked element ( @xmath203 ) . this algorithm tracks the so - called longest - edge propagation path associated with a triangle @xmath180 in a backward manner , denoted for brevity as lepp(@xmath180 ) . for any given conforming triangulation @xmath183 , the lepp(@xmath204 ) is defined as an ordered list of triangles @xmath205 , such that @xmath206 is a neighbour to @xmath207 by the longest edge of @xmath207 for each @xmath208 . the backward - longest - edge - bisection algorithm for a pair @xmath209 , recalled in alg . [ subsect : adaptivity : alg:2 ] , is then used for each @xmath210 ; for further details see e.g. @xcite , section 3 . in fig . [ subsect : adaptivity : fig:1 ] , the lepp(@xmath204 ) consists initially of four triangles , and the bisection proceeds from @xmath211 towards @xmath204 . 1 . choose @xmath212 scheduled for refinement . 2 . * while @xmath213 is not bisected * 1 . find / update the lepp(@xmath204 ) . if * @xmath214 , the last triangle of the lepp(@xmath204 ) , is a terminal boundary triangle ( its longest edge is a part of the physical boundary @xmath215 ) , bisect @xmath214 . + * else * bisect the last pair of terminal triangles of the lepp(@xmath204 ) , @xmath216 and @xmath214 . * end * in this section , the implications of the adaptive scheme for the energy evolutions are discussed . the main motivations are threefold : ( i ) the above presented theory is based on energy minimization , so the mesh refinement procedure should be consistent with these principles ; ( ii ) the energy balance must hold during the mesh refinement ; ( iii ) energy profiles obtained with the adjusted mesh refinement compare much better to those computed for the full lattice solutions , hence the accuracy of the adaptive variational qc method can be assessed from the energetic viewpoint . first , the reader is referred to fig . [ subsect : meshenergy : fig:1 ] . starting from a relaxed configuration at a time step @xmath217 ( for a converged mesh and an equilibrated system ) , the next load increment is applied ( using the same triangulation of the previous time step , i.e. @xmath218 ) . this yields _ physical _ energy increments , which are denoted by the subscript `` p '' , cf . [ subsect : meshenergy : fig:1 ] . after the system is again equilibrated at @xmath219 ( using the same triangulation ) , the indicator condition is violated ( assuming that the damage evolves ) for some elements and hence , the mesh needs to be refined in several steps until convergence is reached at @xmath220 . consequently , a new triangulation @xmath221 is obtained ( stages ( ii ) ( iv ) in alg . [ subsect : adaptivity : alg:1 ] ) . during this mesh refinement ( during the transition from @xmath219 to @xmath220 ) , some elastic energy is released , which is referred to as _ artificial energy of constraints _ ; associated changes in energies are denoted by the subscript `` a '' , see fig . [ subsect : meshenergy : fig:1 ] . the elastic energy released due to mesh refinement was before ( i.e. at @xmath219 ) used to enforce some of the geometric constraints due to interpolation . the gradients of the released elastic energy ( with respect to the kinematic variables ) represent artificial constraining forces . the projection of these and the physical forces via @xmath131 at @xmath219 was zero ( the system was in equilibrium ) . when the new triangulation @xmath222 is constructed , the system must be re - equilibrated because some of the geometric constraints are released . this takes place from stage @xmath223 in alg . [ subsect : adaptivity : alg:1 ] . the elastic part of the artificial energy released during mesh refinement is denoted @xmath224 . this energy is transformed in three contributions : ( i ) an additional dissipation increment @xmath225 , ( ii ) a change of the work performed by the external reactions , and ( iii ) work done due to the relaxation of the internal constraining forces . the latter two contributions are jointly denoted as @xmath226 . . for clarity , only the @xmath227 and @xmath228 energy components are shown . ] taking the above - described aspects into account and evaluating the internal elastic energy @xmath83 , dissipated energy @xmath229 , and the work done by the external forces @xmath230 at time instants @xmath220 provides energy profiles that we will call _ reconstructed_. note that whereas the reconstructed qc energies can be compared well with the energy profiles obtained for the fully - resolved system , the physical qc energies ( energies evaluated at @xmath219 ) are incomparable . this can be verified in section [ sect : examples ] ( figs . [ subsect : simpleex : fig:2b ] and [ subsect : complexex : fig:4b ] ) . finally , let us emphasize that the dissipation distance requires two states @xmath98 and @xmath99 ( see definition ) , that are inherently incompatible for @xmath219 and @xmath220 since they are associated with different triangulations @xmath231 and @xmath221 . in order to compute @xmath225 , first the internal variables required for the new sampling interactions need to be established . in section [ sect : examples ] , this is carried out such that the additional internal variables , @xmath88 , @xmath232 , are computed retrospectively by minimizing their incremental interaction energies , @xmath233 , @xmath234 including @xmath219 ( the entire evolution is required because @xmath77 is history - dependent ) , with respect to internal variables while keeping the kinematic variables fixed ( @xmath235 is computed from the previous results through eq . ) . this operation corresponds to a mapping @xmath236 ( not necessarily unique ) touched upon in @xcite , section 3 . nonetheless , as the mesh is fully refined before any damage occurs ( recall condition ) , all additionally required internal variables in all previous time steps are zero . an analogous procedure is applied also for the internal and external forces . let us return momentarily from qc techniques to the full system , whose evolution is governed by and discuss a suitable strategy for its solution . because indirect displacement load control is used in section [ sect : examples ] , it is convenient to adopt a minimization of the reduced incremental energy @xmath237 as proposed by @xcite , section 3.1.2 . problem then transforms to @xmath238 and can be solved by the standard newton algorithm if @xmath239 is sufficiently smooth in terms of @xmath115 . the taylor expansion of @xmath240 in eq . with respect to @xmath115 in the vicinity of @xmath241 provides the following stationarity condition @xmath242 where superscript @xmath243 indicates the @xmath243-th iteration , and where [ subsect : fullsol : eq:2 ] @xmath244 eq . presents a system of linear equations for the increments @xmath245 . iterating eqs . and until convergence of @xmath246 then yields @xmath247 . explicit expressions for the gradients and hessians can be found in @xcite , appendix a. note that the expressions in @xcite are derived for lattice structures with isotropic hardening plasticity and according to eq . they have to be multiplied by @xmath91 if a given bond experiences tension . moreover , the dependencies on the plastic deformations @xmath248 have to be dropped . for iteration @xmath243 in eq . the potential @xmath249 needs to be minimized with respect to @xmath116 before the gradients and hessians in eq . are evaluated . this strategy is analogous to the condensation of internal variables in fe methods , because the hessian ( @xmath250 in eq . ) effectively corresponds to the consistent tangent stiffness matrix . problem for arbitrary @xmath185 , @xmath243 , and configuration @xmath241 , is approached as follows : by rewriting the incremental energy into the interaction - wise form , cf . eq . , the minimization is split into separate one - dimensional problems that can be solved individually according to : @xmath251 the box - constrained minimization coincides with the karush tucker complementarity conditions for a uniformly stretched bar @xmath57 with homogeneous damage . to see this , we take the derivative in that provides us with first - order optimality conditions : @xmath252 see also @xcite . by recalling eqs . and , three states are distinguished : a. @xmath253 , i.e. elastic loading / unloading from a damaged state , implying @xmath254 . since @xmath255 is non - negative in this case , we obtain @xmath256 b. @xmath257 , i.e. damage evolves , implying @xmath258 . as @xmath255 can be either positive or negative , we have @xmath259 c. @xmath260 , i.e. fully damaged state , implying @xmath261 . in this case , @xmath255 can only be non - positive and hence , @xmath262 for brevity , we have denoted @xmath263 . if the interaction undergoes compression , the minimization in eq . is equivalent to @xmath264 which provides @xmath265 as @xmath266 is an increasing function of @xmath93 ; recall definition in eq . and the discussion therein . once the constitutive law is specified , i.e. @xmath267 is given , the internal variable @xmath88 can be determined for stretched interaction @xmath57 based on the three possible states ( i ) , ( ii ) , or ( iii ) . the general derivation of @xmath267 is presented in [ sect : a ] , where its specific form for an exponential softening rule is derived as well , cf . also section [ sect : examples ] . following the steps presented in section [ sect : qc ] , instead of directly minimizing the exact incremental energy @xmath239 , its approximation @xmath268 is minimized with respect to the reduced variable @xmath269 . a fixed triangulation @xmath183 is assumed , meaning that only stage ( ii ) of alg . [ subsect : adaptivity : alg:1 ] is addressed in this section ; the other ingredients of the algorithm have already been discussed and do not affect stage ( ii ) . using the chain rule in the taylor expansion of @xmath270 provides the following stationarity condition @xmath271 where [ subsect : qcsol : eq:2 ] @xmath272 and where the partial derivatives read [ subsect : qcsol : eq:3 ] @xmath273 cf . also eq . . note that in eqs . , reduced site energies @xmath274 are introduced in analogy to . the converged solution of and yields @xmath275 . in order to construct @xmath270 , the interaction - wise formulation of the incremental energy is used , recall eq . . the minimization then requires the solution of @xmath276 independent problems @xmath277 because the solutions of do not depend on the weights @xmath152 , each minimization problem can be solved independently , according to section [ subsect : fullsol ] . section [ sect : varform ] made clear that the energetic solution @xmath278 has to satisfy the energy equality , which serves as a selection criterion with respect to the numerous local minimizers of @xmath1 . usually , a certain variant of the _ backtracking _ algorithm helps to select the proper solution , see e.g. @xcite , @xcite , @xcite , or @xcite . in order to keep our exposition brief and simple , an alternative approach is followed . it can be shown that as long as the solution process remains continuous in time , the energy balance holds , see e.g. @xcite . for this purpose , indirect load - displacement control is used in this contribution ( see e.g. @xcite , section 22.2.3 ) , employing either crack mouth opening displacement ( cmod ) , crack mouth sliding dispacement ( cmsd ) , or a combination of both . for the full lattice system , dirichlet boundary conditions are imposed in the usual way , whereas the neumann boundary conditions enter the system via eq . , as the work performed by the force vector @xmath279 . the indirect displacement control is then introduced via an additional parameter @xmath280 that proportionally scales the dirichlet or the neumann boundary conditions . for example , in the case of neumann boundary conditions we have @xmath281 , where @xmath282 is a given reference load vector . the parameter @xmath283 is then determined from the following scalar condition @xmath284 which holds for prescribed step size @xmath285 at each newton iteration and time step . due to the substitution of @xmath281 in , the solution increment in becomes a function of @xmath283 , which can subsequently be eliminated using eq . . a similar procedure is also applied to the dirichlet boundary conditions . the column matrices @xmath286 and @xmath287 in eq . ( both in @xmath74 ) specify suitable displacement measures . the displacement difference between a pair of crack - mouth atoms in the direction perpendicular to a crack is specified by @xmath286 , i.e. @xmath288 , whereas @xmath287 specifies the difference along the crack direction , i.e. @xmath289 . the combination of both simply reads @xmath290 . in the case of qc systems , the controlled quantity is @xmath275 , i.e. eq . is enforced for @xmath291 , and the column matrices @xmath286 and @xmath287 are from @xmath292 . finally , the dirichlet boundary conditions are applied only on repatoms and , in accordance with eqs . , , and , the qc system is loaded with @xmath293 if external forces are present . the previously discussed theory is demonstrated on two benchmark problems in this section . a regular x - braced lattice with the following quadratic pair potential is used in both cases ( the superscripts @xmath57 are dropped for brevity ) @xmath294 where the interaction stiffness is @xmath295 according to standard truss theory , @xmath296 is the young s modulus , and @xmath297 the cross - sectional area . note that this definition corresponds to the rotated engineering deformation measure and an undamaged bond . if damage takes place , @xmath298 needs to be multiplied by @xmath299 , recall eq . . the mechanical behaviour is made geometry - independent by the introduction of the bond strain @xmath300 and stress @xmath301 ( see [ sect : a ] for more details ) . in both examples below , the exponential stress - strain softening law sketched in fig . [ sect : examples : fig:1 ] is employed . the strain softening branch @xmath302 takes the following form @xmath303 where @xmath304 is a measure for its initial slope , and @xmath305 is the limit elastic strain . as shown in eqs . of [ sect : a ] , function @xmath302 in eq . fully defines @xmath306 . the employed physical constants for both examples are specified in tab . [ sect : examples : tab:1 ] . and @xmath307 , cf . the elastically stored energy , @xmath308 , corresponds to the area of the dashed triangle , and the dissipated energy , @xmath309 , to the area of the upper triangle with the curved side . ] .dimensionless material and geometric parameters for both test examples . [ cols="<,>,<,>,<",options="header " , ] [ sect : examples : tab:1 ] denotes the applied vertical displacement . ] the first example considers an l - shaped plate modelled by a moderate number of atoms , serving therefore as a proof of concept . for a continuum - based analogy see e.g. @xcite , section 4.1 . the reference domain @xmath134 fills a @xmath310 lattice - spacing - sized square with cut out right - bottom quarter ( see fig . [ subsect : simpleex : fig:0 ] ) . it comprises @xmath311 atoms and @xmath312 interactions . the lattice spacing is @xmath313 unit length in both directions , and the material is homogeneous throughout the body . the specimen is fixed at the bottom part of its boundary , @xmath314 , while a vertical displacement @xmath315 is prescribed at @xmath316 , i.e. [ subsect : simpleex : eq1 ] @xmath317 where @xmath318 , @xmath319 denotes the deformed positions of all atoms lying on a line segment @xmath320 , and @xmath321 stands for the middle point of @xmath320 ; each component of @xmath322 is a vector , i.e. @xmath323^\mathsf{t}$ ] . in order to prevent any damage evolution under applied dirichlet boundary conditions , local stiffening is used in the vicinity of @xmath316 ( the young s modulus is @xmath324 times larger than elsewhere and the limit elastic strain @xmath305 is infinite ) . the overall evolution of the system is controlled by the cmod , recall eq . where we set @xmath325 , meaning that the indirect displacement control parameter @xmath326 is equal to @xmath327 . the numerical study has been performed for three systems . the first one is the fully - resolved lattice , whereas the other two correspond to the adaptive qc with different safety margins @xmath193 , cf . the first qc system uses a _ moderate _ ( @xmath328 ) and the second a _ progressive _ ( @xmath329 ) refinement strategy ; these systems are referred to as _ moderate qc _ and _ progressive qc _ for brevity . rather low values of @xmath193 are necessary because of the steep peaks of interaction ( and site ) energies occurring in the vicinity of the crack tip . the deformed configuration predicted by the full lattice computation at @xmath330 is depicted in fig . [ subsect : simpleex : fig:1a ] ; note that only the atoms are shown . it can be observed that the crack initiates at the inner corner ( @xmath331 ) , and propagates horizontally leftward . for all three systems , the crack paths are identical ( not shown ) . the reaction force @xmath332 as a function of @xmath327 is plotted in fig . [ subsect : simpleex : fig:1b ] . in the initial stages , the reaction forces increase linearly , with the smallest stiffness corresponding to the fully - resolved system . because both qc systems have the same initial triangulations @xmath132 ( cf . figs . [ subsect : simpleex : fig:3a ] and [ subsect : simpleex : fig:3e ] ) , their stiffnesses are identical . for an increasing applied displacement @xmath327 , the progressive qc starts to refine first , followed by the moderate one . as the refinement process entails the relaxation of the geometric interpolation constraints and hence , a decrease of stiffness , the reaction forces approach the values of the fully - resolved system . the peak force is nevertheless overestimated by approximately @xmath333 by both qc systems . the post - peak behaviour exhibits a mild structural snap - back , and the curves match satisfactorily with negligible error for the post - peak part of the diagram ( when the crack fully localizes ) . overall , it can be concluded that the results are more accurate for a lower safety margin @xmath193 . this entails , however , an increase in the number of repatoms and hence , decrease in efficiency . nevertheless , once the crack fully localizes , the differences between the results for various values of @xmath193 decrease , though the lower value is generally more accurate . the energy profiles are shown in fig . [ subsect : simpleex : fig:2a ] as functions of @xmath327 . the elastic energy @xmath83 increases quadratically until the peak load and then drops gradually as the fracture process occurs . near the peak load , the dissipated energy @xmath229 increases rapidly . it then continues to grow at a more moderate rate throughout the softening part of the load - displacement response . notice that the energy balance is satisfied along the entire loading path , since the thin dotted lines corresponding to the work performed by external forces @xmath230 lie on top of the thick dashed lines representing @xmath334 . the maximum relative unbalance between @xmath335 and @xmath230 is below @xmath336 . from the energy point of view , it can be concluded that both qc systems approximate the results of the full lattice system well . from the @xmath229-curves we deduce that the crack starts to propagate first according to the moderate qc approach ( because of its highest stiffness ) , and that the cracks corresponding to the progressive qc and to the full - lattice solution initiate almost at the same instant . let us recall section [ subsect : meshenergy ] and note that the energy profiles presented in fig . [ subsect : simpleex : fig:2a ] correspond to reconstructed energies evaluated at time instants @xmath220 , cf . [ subsect : meshenergy : fig:1 ] . the energy components that are exchanged ( the artificial energies ) during the moderate qc prediction are presented in fig . [ subsect : simpleex : fig:2b ] . we see that their magnitudes are large compared to the two physical energies ( @xmath337 and @xmath338 ) . this means that the energy - reconstruction procedure described in section [ subsect : meshenergy ] is essential and that the artificial energies can not be neglected . in fig . [ subsect : simpleex : fig:4 ] , the number of repatoms normalized by the total number of atoms ( i.e. @xmath339 ) as a function of @xmath327 is shown . because both curves are situated below @xmath340 , and the moderate qc below @xmath341 , appreciable computational savings are achieved . in terms of sampling interactions , the relative numbers of sampling interactions @xmath342 are slightly higher . namely , below @xmath343 and @xmath344 for the progressive and moderate approach . in the l - shaped plate simulations as a function of @xmath327 . ] + for completeness , fig . [ subsect : simpleex : fig:3 ] shows eight snapshots of the mesh evolution . although both initial meshes @xmath132 are similar , different safety margins @xmath193 cause the fully - resolved region of the progressive qc approach to be larger . consequently , the obtained results are more accurate , but at the price that also regions far from the crack path are refined ( e.g. along the @xmath345 part of the boundary , cf . [ subsect : simpleex : fig:3h ] ) . the mesh of the moderate approach remains more localized , at the expense of a minor loss of accuracy . in the second example , a rectangular domain @xmath134 is exposed to antisymmetric four - point bending , cf . [ subsect : complexex : fig:1 ] and , e.g. , @xcite . the homogeneous body is pre - notched from the top edge to initiate a crack , and stiffened locally where the dirichlet and neumann boundary conditions are applied ( again , the young s modulus is @xmath324 times larger than elsewhere and the limit elastic strain @xmath305 is infinite to prevent any damage evolution ) . the lattice spacing is of a unit length in both directions . the entire specimen consists of @xmath346 atoms connected by @xmath347 interactions . the ( vertical ) forces @xmath348 and @xmath349 are prescribed as , cf . [ subsect : complexex : fig:1 ] , @xmath350 where @xmath326 is the indirect displacement control parameter , cf . eq . and the discussion on it . in contrast to the previous example , the sum of cmod and cmsd is used to control the simulation ( recall eq . where @xmath351 and @xmath352 ) . this combination of the two measures is required because of the following reasons . initially , the cmod is close to zero or even negative whereas the cmsd drives the evolution . in the later stages , however , the cmsd is constant while cmod parametrizes the process . their sum , therefore , naturally switches between the two approaches , see also fig . [ subsect : complexex : fig:2a ] . due to a higher brittleness compared to the previous example , cf . tab . [ sect : examples : tab:1 ] , two loading rates for @xmath353 , as specified in fig . [ subsect : complexex : fig:2a ] , are used . the numerical example is studied only for the two qc systems : the moderate qc , @xmath328 , and the progressive qc , @xmath329 , with no validation against the fully - resolved solution . in order to achieve a higher accuracy using the progressive approach , a globally fine initial mesh is used , in which the maximum triangle edge length is restricted to @xmath354 lattice spacings . for the moderate approach , the mesh is as coarse as possible to describe the specimen geometry by a right - angled triangulation . the initial meshes are the top triangulations in fig . [ subsect : complexex : fig:5 ] ( figs . [ subsect : complexex : fig:5a ] and [ subsect : complexex : fig:5e ] ) . the deformed configuration predicted by the progressive qc approach at @xmath355 is presented in fig . [ subsect : complexex : fig:2b ] . in qualitative accordance with experimental data , see e.g. @xcite , section 4.1 , the crack path initiates at the right bottom corner of the notch , subsequently curves downwards and then approaches the bottom part of the boundary to the right side of the force @xmath348 . the crack paths predicted by both adaptive qc schemes are presented on the undeformed configuration in fig . [ subsect : complexex : fig:3a ] . here we notice that the results are almost identical . the total applied force @xmath356 , recall eq . , is plotted in fig . [ subsect : complexex : fig:3b ] against @xmath353 . although the initial triangulations differ significantly ( cf . figs . [ subsect : complexex : fig:5a ] and [ subsect : complexex : fig:5e ] ) , the initial slopes are practically identical . as the moderate qc refines later and less extensively , the peak force is overestimated by the moderate qc compared to the progressive qc approach by approximately @xmath333 . in the post - peak region , both curves are practically identical again . the reconstructed energy profiles of both refinement approaches are presented in fig . [ subsect : complexex : fig:4a ] . from there it may be concluded that the results match well . moreover , we see that both solutions satisfy the energy balance along the entire loading path . in fig . [ subsect : complexex : fig:4b ] , substantial energy exchanges ( i.e. artificial energies ) due to mesh refinement can again be observed ( similar to the first numerical example ) . because the size of the fully - resolved domain is small compared to the entire domain , both adaptive qc approaches achieve a substantial computational gain . this is supported by fig . [ subsect : complexex : fig:6 ] in which the relative numbers of repatoms are presented . the ratio remains below @xmath357 and even below @xmath358 for the moderate refinement strategy . in the case of the relative numbers of sampling interactions , the ratios remain below @xmath359 and @xmath360 for the progressive and moderate approaches . note also that the number of repatoms increases rapidly near the peak load for the progressive approach whereas it develops more gradually for the moderate one . finally , in fig . [ subsect : complexex : fig:5 ] several snapshots that capture the evolution of the triangulations @xmath183 are presented . it can again be noticed that the progressive approach refines until quite far from the crack tip , whereas the fully - refined region in the moderate qc remains localized . as a function of @xmath353 . eight chosen triangulations corresponding to @xmath361 , @xmath313 , @xmath362 , and @xmath363 are presented in fig . [ subsect : complexex : fig:5 ] . ] + + + in this contribution , we have developed an energy - based dissipative qc approach for regular lattice networks with damage and fracture . the study shows that the efficiency of the qc methodology applies also to brittle phenomena , and that together with an adaptive refinement strategy it provides a powerful tool to predict crack propagation in lattice networks . the main results can be summarized as follows : 1 . the general variational formulation for rate - independent processes by @xcite was rephrased for the case of lattice networks with damage . the two standard qc steps , interpolation and summation , were revisited from an adaptive point of view . for the interpolation , meshes with right - angled triangles were used because of their * ability to naturally refine to the fully - resolved underlying lattice * binary - tree structure that allows for fast and efficient data transfer * significant reduction of the summation part of the qc error . 3 . to determine the location of the critical region , a heuristic marking strategy with a variable parameter that controls the accuracy of the simulation was proposed . the mesh refinement procedure was discussed from an energetic standpoint , and the significance of the reconstruction procedure with respect to the energy consistency was shown . the numerical examples demonstrated that the introduced marking strategy is capable of satisfactorily predicting the evolution of the crack path and the load - displacement response , especially in the post - peak region . solutions obtained using indirect load displacement control satisfied the energy equality condition . let us note that as the crack tip propagates throughout the body , it would be convenient to include besides the mesh refinement ahead of it also mesh coarsening in its wake . furthermore , instead of the proposed heuristic marking strategy , techniques such as goal - oriented error estimators may be implemented to improve further the performance of the adaptive qc . both aspects enjoy our current interest and will be reported separately . this appendix provides the details on the dissipation function @xmath309 needed in the dissipation distance @xmath109 of eq . for the constitutive law with exponential softening shown in fig . [ sect : examples : fig:1 ] , cf . also eq . . because damage evolves only under tension , recall section [ subsect : energies ] , we assume in the remainder of this section that @xmath364 for the ease of notation . recall that @xmath365 , in accordance with eq . , denotes the initial length of the interaction , and @xmath366 its ( admissible ) deformed length . in order to make the interaction behaviour geometry - independent , we introduce the interaction strain , @xmath367 , and the interaction stress ( the superscripts @xmath57 are dropped for the sake of brevity ) @xmath368 where the normal force of the interaction is denoted as @xmath369 , and where we have used @xmath370 . remind that @xmath296 is the young s modulus , @xmath297 the cross - sectional area , and @xmath371 associated ( admissible ) damage variable . the pair potential @xmath298 from eq . can be rewritten as @xmath372 where we have emphasized that the admissible strain @xmath373 is a function of the admissible length @xmath366 . to construct a constitutive model that under monotonic damage evolution displays a specific stress - strain response , say @xmath374 where @xmath375 is a given target softening function ( recall eq . and fig . [ sect : examples : fig:1 ] ) , the damage variable is considered as a function of the current strain . using the constitutive relation in eq . while employing , one obtains @xmath376 i.e. the damage variable as a function of the strain @xmath373 . in the interval of growing damage , the function @xmath377 is invertible , providing @xmath378 eq . substituted into ( where @xmath298 is now considered as a function of @xmath373 rather than @xmath366 according to ) provides ; recall the first - order optimality conditions in eqs . , where the minimizer was denoted as @xmath379 , cf . also eqs . and . using hats in eq . is therefore a slight abuse of notation as @xmath371 is not entirely arbitrary . ] @xmath380 integrating this relation yields @xmath381 that can be expressed in a closed form for some special cases such as linear softening , cf . @xcite , eq . ( 85 ) . for the exponential softening law defined in eq . , the expression attains the following form @xmath382 and is shown in fig . [ sect : a : fig:1a ] . rewriting eq . as @xmath383 we can cast its inversion in terms of the lambert transcendental @xmath384 function ( recall the defining equation @xmath385 , cf . @xcite , eq . ( 1.5 ) ) @xmath386 , \label{sect : a : eq:2}\ ] ] which is the counterpart to eq . . upon substituting this inversion in eq . , we end up with the following integral @xmath387 ^ 2 \,\mathrm{d}\eta , \label{sect : a : eq:3}\ ] ] where @xmath388 have been introduced for brevity . let us further denote @xmath389 and write ( by the defining equation of the lambert function ) @xmath390 which can be differentiated on both sides to yield @xmath391 employing eq . in eq . , we obtain @xmath392 now , a change of variables in the integral according to eq . can be carried out , providing us with @xmath393 the relation on the right hand side of eq . can be expanded as @xmath394 \right . + w(c_2)e^{-w(c_2 ) } - w(c_2/(1-\widehat{\omega}))e^{-w(c_2/(1-\widehat{\omega } ) ) } \right\ } , \end{aligned } \label{sect : a : eq:9}\ ] ] see also fig . [ sect : a : fig:1b ] where a sketch of @xmath306 is shown . the energy dissipated by the complete failure process then reads @xmath395 . \label{sect : a : eq:10}\ ] ] financial support of this work from the czech science foundation ( gar ) under project no . 14 - 00420s is gratefully acknowledged . 52 natexlab#1#1[1]`#1 ` [ 2]#2 [ 1]#1 [ 1]http://dx.doi.org/#1 [ ] [ 1]pmid:#1 [ ] [ 2]#2 , , , . . , http://www.sciencedirect.com/science/article/pii/s0022509615000630 , . , , , , a. . , . http://www.sciencedirect.com/science/article/pii/s0045782514002047 , . , , , . . , http://dx.doi.org/10.1002/nme.3134 , . , , , b. . , . http://www.sciencedirect.com/science/article/pii/s0022509614001100 , . , , , http://www.sciencedirect.com/science/article/pii/s0022509613002445 , . , , , d. . , . http://www.sciencedirect.com/science/article/pii/s004578251300279x , . , , , , . . , http://dx.doi.org/10.1002/gamm.201510018 , . http://www.sciencedirect.com/science/article/pii/s1359835x13000134 , . , . http://dx.doi.org/10.1007/s10898-010-9560-6 , . , . . , . . , , , http://www.sciencedirect.com/science/article/pii/s0022509699000289 , . , , , . . , http://epubs.siam.org/doi/abs/10.1137/080741033 , . , , , . . , , , , , . . , , , , , . . , http://dx.doi.org/10.1007/bf02124750 , . , , , . . , http://www.sciencedirect.com/science/article/pii/s0045782505003956 , . , , , , . . , http://www.sciencedirect.com/science/article/pii/s0013794415000053 , . http://www.sciencedirect.com/science/article/pii/s0022509698000349 , . http://www.sciencedirect.com/science/article/pii/s0020768309004752 , . http://dx.doi.org/10.1007/s10704-012-9753-8 , . http://www.sciencedirect.com/science/article/pii/s0022509610002425 , . http://eu.wiley.com/wileycda/wileytitle/productcd-0471987166.html . , , . http://www.sciencedirect.com/science/article/pii/s0020768315002620 , . http://cvgmt.sns.it / paper/2832/. , , . . http://www.sciencedirect.com/science/article/pii/s0167663612000683 , . , , , . . , http://www.sciencedirect.com/science/article/pii/s016784421000039x , . http://journals.cambridge.org/article_s0962492913000068 , . , , , . http://dx.doi.org/10.1007/s00466-015-1127-4 , . http://www.sciencedirect.com/science/article/pii/s004578251400423x , . http://www.springer.com/us/book/9781493927050 , . , , , . . , http://www.sciencedirect.com/science/article/pii/s0045782509003181 , . , , , . http://dx.doi.org/10.1007/s002050200194 , . http://dx.doi.org/10.1023/a%3a1026098010127 , . http://www.sciencedirect.com/science/article/pii/s0021999102971834 , . http://www.sciencedirect.com/science/article/pii/s0045782598002199 , . http://www.sciencedirect.com/science/article/pii/s1359835x04002593 , . http://dx.doi.org/10.1007/s00161-011-0228-3 , . http://www.sciencedirect.com/science/article/pii/s002250961100055x , . , , , . . , , , , . . , http://www.sciencedirect.com/science/article/pii/s0022509610001341 , . , . . , . . , , , , http://arxiv.org/abs/1601.00625[arxiv:1601.00625 ] . , . ph.d . thesis . technische universiteit delft . . http://www.sciencedirect.com/science/article/pii/095894659290004f , . , , , , , , . . , http://www.sciencedirect.com/science/article/pii/s0022509698000519 , . http://dx.doi.org/10.1080/01418619608243000 , , http://arxiv.org/abs/http://dx.doi.org/10.1080/01418619608243000 [ arxiv : http://dx.doi.org/10.1080/01418619608243000 ] . , , . . http://www.sciencedirect.com/science/article/pii/s0045782514003545 , .

bl lacertae was the target of an extensive multiwavelength monitoring campaign in the second half of 2000 . the campaign had revealed optical and x - ray intraday variability on time scales of hours and evidence for significant spectral variability both at optical and x - ray frequencies . during the campaign , bl lacertae was observed in two different activity states : a quiescent state with relatively low levels of optical and x - ray fluxes and a synchrotron cutoff at energies below the x - ray regime , and a flaring state with high levels of optical and x - ray emission and a synchrotron cutoff around or even beyond kev . in this paper , we are using both leptonic and hadronic jet models to fit the broadband spectra and spectral variability patterns observed in both activity states in 2000 . we start out with global spectral models of both activity states . subsequently , we investigate various flaring scenarios for comparison with the observed short - term variability of bl lacertae in 2000 . for our leptonic jet model , we find that the short - term variability , in particular the optical and x - ray spectral variability , can be best represented with a flaring scenario dominated by a spectral - index change of the spectrum of ultrarelativistic electrons injected into the jet . based on this result , a detailed model simulation of such a flaring scenario , reproducing the observed optical and x - ray spectral variability and broadband sed of bl lacertae during the bepposax pointing around nov . 1 , 2000 , simultaneously , is presented . our leptonic modeling results are compared to fits using the hadronic synchrotron - proton blazar ( spb ) model . that model can reproduce the observed seds of bl lacertae in a scenario with-synchrotron dominated high - energy emission . it requires a significantly higher magnetic field than the leptonic model ( g vs. g in the leptonic model ) and a lower doppler factor associated with the bulk motion of the emission region ( vs. in the leptonic model ) . the hadronic model predicts a significantly larger gev flux than the leptonic models , well within the anticipated capabilities of veritas and magic .

astro-ph0403627

bl lacertae (= 1es 2200 + 420 ; z = 0.069 ) was historically the prototype of the bl lac class of active galactic nuclei ( agn ) . these objects are characterized by continuum properties similar to those of flat - spectrum radio quasars ( non - thermal optical continuum , high degree of linear polarization , rapid variability at all wavelengths , radio jets with individual components often exhibiting apparent superluminal motion ) , but do usually show only weak emission lines ( with equivalent width in the rest - frame of the host galaxy of @xmath8 ) , if any . in bl lacertae itself , however , h@xmath9 ( and h@xmath10 ) emission lines with equivalent widths in excess of 5 have been detected during a period of several weeks in 1995 @xcite , and in 1997 @xcite . superluminal motion of @xmath11 up to @xmath12 has been observed in this object @xcite . bl lacertae is classified as a low - frequency peaked bl lac object ( lbl ; * ? ? ? * ) . from an interpolation between the ghz radio spectrum and the ir optical spectrum , it can be inferred that its low - frequency spectral component typically peaks at mm to @xmath2 m wavelengths , while the high - frequency component seems to peak in the multi - mev gev energy range . bl lacertae has been studied in detail during various intensive multiwavelength campaigns ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? it is a particularly interesting object for detailed x - ray studies since it is in the x - ray regime where the two broad components of the multiwavelength seds of bl lacertae ( and other lbls ) are overlapping and intersecting . x - ray observations of this source at different epochs show significant flux and spectral variability , indicating that the x - ray emission is at times dominated by the high - energy end of the synchrotron emission , while at other occasions it is dominated by the low - frequency portion of the high - energy bump of the sed . in fact , bl lacertae has repeatedly shown a concave shape ( e.g. , * ? ? ? * ; * ? ? ? * ) , with rapid variability mainly restricted to the low - energy excess portion of the spectrum ( e.g. , * ? ? ? * ; * ? ? ? * ) . in the framework of relativistic jet models , the low - frequency ( radio optical / uv ) emission from blazars is interpreted as synchrotron emission from nonthermal electrons in a relativistic jet . the high - frequency ( x - ray @xmath13-ray ) emission could either be produced via compton upscattering of low frequency radiation by the same electrons responsible for the synchrotron emission ( leptonic jet models ; for a recent review see , e.g. , * ? ? ? * ) , or due to hadronic processes initiated by relativistic protons co - accelerated with the electrons ( hadronic models , for a recent discussion see , e.g. , * ? ? ? * ; * ? ? ? the lack of knowledge of the primary jet launching mechanism and the difficulty in constraining the jet composition from general energetics considerations ( for recent discussions see , e.g. , * ? ? ? * ; * ? ? ? * ) are currently leaving both leptonic and hadronic models open as viable possibilities . also , detailed simulations of particle acceleration at relativistic shocks or shear layers , which may be relevant for the acceleration of ultrarelativistic particles in blazar jets , show that a wide variety of particle injection spectra may result in such scenarios ( e.g. , * ? ? ? * ; * ? ? ? * ) , greatly differing from the standard spectral index of 2.2 2.3 , which was previously believed to be a universal value in relativistic shock acceleration scenarios ( e.g. , * ? ? ? * ; * ? ? ? thus , both the nature of the matter in blazar jets and the energy spectra of ultrarelativistic particles injected into the emission regions in blazar jets are difficult to constrain from first principles . for this reason , we are leaving these aspects as virtually free parameters in our models , and attempt to constrain them through the results of detailed time - dependent modeling of blazar emission . while simultaneous broadband spectra are very useful to constrain both types of blazar jet models , there still remain severe ambiguities in their interpretation w.r.t . the dominant electron cooling , injection , and acceleration mechanisms , as was recently illustrated for the case of w comae by @xcite . those authors have also demonstrated that a combination of broadband spectra with timing and spectral variability information , in tandem with time - dependent model simulations @xcite can help to break some of these degeneracies . for this reason , we had organized an intensive multiwavelength campaign to monitor bl lacertae in the second half of 2000 at radio , optical , x - ray , and very - high - energy ( vhe ) @xmath13-ray frequencies , putting special emphasis on detailed variability information . the results of this multiwavelength campaign were published in @xcite ; for more details on the optical and x - ray observations , see also @xcite and @xcite , respectively . in [ observations ] , we briefly highlight the main results of that campaign , emphasizing those aspects that we will specifically use here to constrain our model calculations . the purpose of this paper is to use both leptonic and hadronic jet models to fit the spectral energy distributions ( seds ) and spectral variability of bl lacertae observed in 2000 . following a brief description of both leptonic and hadronic models in [ model_descriptions ] , we will first present spectral fits to the seds of bl lac in [ sed_modeling ] . our code used to evaluate leptonic models allows us to make detailed simulations of the spectral variability resulting from different plausible flaring scenarios . in [ var_modeling ] , we will first do a general parameter study of various scenarios and compare the results qualitatively with the observed spectral variability trends in bl lacertae ( [ parameter_study ] ) . this will allow us to decide on a preferable flaring scenario , for which we subsequently optimize our choice of model parameters to fit simultaneously the sed and spectral variability of bl lacertae as observed in 2000 ( [ var_fit ] ) . possible physical scenarios triggering the inferred variability of the electron injection spectrum will be discussed in [ discussion ] . we summarize in [ summary ] . throughout this paper , we refer to @xmath9 as the energy spectral index , @xmath14 [ jy ] @xmath15 . a cosmology with @xmath16 and @xmath17 and @xmath18 km s@xmath19 mpc@xmath19 bl lacertae was observed in a co - ordinated multiwavelength campaign at radio , optical , x - ray , and vhe @xmath13-ray energies during the period mid - may 2000 , until the end of the year . results of the multiwavelength observing campaign have been published in @xcite . here , we briefly highlight the results that we will specifically concentrate on in our modelling effort . focusing on an originally planned core campaign period of july 17 aug . 11 , bl lacertae was the target of an intensive optical campaign by the whole earth blazar telescope ( webt ; * ? ? ? * ; * ? ? ? * see also http://www.to.astro.it/blazars/webt/ ) , in which 24 optical telescopes throughout the northern hemisphere participated . details of the data collection , analysis , cross - calibration of photometry from different observatories , etc . pertaining to the webt campaign have been published in @xcite . bl lacertae was in a rather quiescent state during the core campaign , in which the densest light curve sampling was obtained . however , the source underwent a state transition to an extended high state in mid - september 2000 , which lasted throughout the rest of the year . the webt campaign returned optical ( r - band ) light curves of unprecedented time coverage and resolution . brightness variations of @xmath20 , corresponding to flux variations of @xmath21 , within @xmath0 hr have been found . clear evidence for spectral variability at optical wavelengths was found , and the color changes were more sensitive to rapid variations than the long - term flux level . during well - sampled , short flares ( on time scales of a few hours ) , the color changes strictly follow the flux variability in the sense that the spectra are harder when the flux is higher . a plot of b - r vs. r ( see fig . [ b_r_fit ] ) reveals two separate regimes within which the r magnitudes are well correlated with the respective b - r colors . however , there seems to be a discontinuity at @xmath22 mag , separating a high - flux and a low - flux regime . within each regime , a similar range of b - r colors is observed @xcite . at x - ray energies , bl lacertae was observed with the _ bepposax _ narrow field instruments ( nfi ) in two pointings on july 26 27 and oct . 31 nov . 2 , 2000 @xcite . in addition , the source was monitored by the _ rossi x - ray timing explorer ( rxte ) _ proportional counter array ( pca ) in 3 short pointings per week @xcite . the details of the _ bepposax _ observations and the data analysis methods have been published in @xcite . the drastic change of the activity state of bl lacertae in mid - september observed in the optical range is accompanied by several large flares in the pca light curve over a @xmath4 months period , but not by a similarly extended high flux state as seen in the optical . in fact , while the average flux level increased only slightly , a higher level of activity was indicated by a higher degree of variability . during our second _ bepposax _ pointing around nov . 1 , 2000 , bl lacertae was in an exceptionally bright x - ray outburst state . interestingly , the r - band lightcurve indicates a relatively low optical flux , compared to the average flux level after mid - september 2000 , coincident with this x - ray outburst . in the following , we will concentrate on the data analysis results obtained using a neutral hydrogen column density of @xmath23 @xmath24 and an optical extinction coefficient of @xmath25 . during the july 26 27 _ bepposax _ observation , the source was in a low flux and activity state . the fit to the _ bepposax _ spectrum resulted in @xmath26 , confirming the low - activity state of the source at that time and indicating that the entire x - ray spectrum might have been dominated by the low - frequency end of the high - energy component of the broadband sed of bl lacertae . the short - term lecs ( [ 0.5 2 ] kev ) and mecs ( [ 2 10 ] kev ) lightcurves of bl lacertae during this observation ( see fig . 3 of * ? ? ? * ) display a large ( factor @xmath27 ) flare on a time scale of @xmath28 hr , while the source appears less variable at higher energies . this behavior has been noted in this source before ( e.g. , * ? ? ? * ) , and is even more obvious in the oct 31 nov . 2 observation . during the second _ bepposax _ pointing on oct . 31 nov . 2 , 2000 , the lecs + mecs 0.3 10 kev spectrum was well fitted with a power - law model with @xmath29 @xcite . in this observation , bl lacertae was also significantly detected by the pds up to @xmath30 kev , indicating a spectral hardening in this energy range . the soft shape of the lecs + mecs spectrum clearly indicates that it was dominated by the high - energy end of the low - energy ( synchrotron ) component in this observation . the spectral hardening evident in the pds spectrum might indicate the onset of the high - energy component beyond @xmath1 kev . @xcite extracted the lecs and mecs light curves in three different energy channels during the second _ bepposax _ pointing , along with the two hardness ratios : hr1 = mecs [ 2 - 4 ] / lecs [ 0.5 - 2 ] and hr2 = mecs [ 4 - 10 ] / mecs [ 2 - 4 ] . the lecs and mecs light curves show significant variability in all energy channels , with flux variations of factors of @xmath31 4 on time scales down to @xmath32 2 hr . the x - ray spectral variability on short ( intra - day ) time scales can be characterized through variations of the _ bepposax _ hardness ratios hr1 and hr2 as a function of the respective count rates . an example of such a hardness - intensity diagram ( hid ) is shown in fig . [ hid30_fit ] . a weak hardness - intensity anti - correlation at soft x - rays ( hr1 vs. lecs ) is generally found . occasionally , a positive hardness - intensity correlation at medium - energy x - rays ( hr2 vs. mecs ) can be found , but is not always apparent . generally , no significant cross - correlations with measurable time delays between different energy bands could be identified in this campaign . a possible correlation between the x - ray and optical light curves with an optical delay of 4 5 hr during the july 26 27 _ bepposax _ observation did not hold up to any statistical significance test ( for more details , see * ? ? ? however , if we assume that the optical lag of @xmath28 5 hr is real and can be interpreted as due to synchrotron cooling , it allows an independent magnetic field estimate , which is in good agreement with the independent estimate based on the measured synchrotron peak flux and a basic equipartition argument ( see [ parameter_estimates ] ) . @xcite constructed two simultaneous broadband seds for the times of the two _ beppo_sax pointings ( see fig . [ mw_combined ] ) . [ mw_combined ] illustrates the different activity states between the july 26/27 and the oct . 2 _ beppo_sax observations . in the july 26/27 sed , the synchrotron peak appears to be located at frequencies clearly below the optical range , and a synchrotron cutoff near or below the soft x - ray regime . in contrast , the sed of oct . 31 nov . 2 shows clear evidence for the presence of the synchrotron component out to at least 10 kev , and the synchrotron peak might be located in the optical range . the figure also shows the _ rxte _ pca spectrum of the observation a few hours before the beginning of the oct . beppo_sax pointing . this pca spectrum shows characteristics rather similar to the low - state spectrum , and illustrates the drastic nature of the short - term x - ray variability . @xcite have shown that the extrapolation of the optical spectrum towards higher frequencies does not connect smoothly with the contemporaneous soft x - ray spectrum ( see their fig . in addition to the various scenarios discussed by @xcite to possibly explain this misalignment , it seems also possible that it could be an artifact of the flux averaging over the @xmath0 days of the _ beppo_sax observations , including multiple short - term flares of only a few hours each . in order to test for this possibility , it will be essential to apply fully time - dependent agn emission models , as we do in this paper , and fit time - dependent spectral variability patterns rather than only time - averaged seds . in this section , we give a brief description of the leptonic and hadronic blazar jet models used for our spectral fitting and variability study ( [ leptonic_description ] and [ hadronic_description ] ) . here , we also briefly review some general parameter estimates derived previously @xcite which will be used as a starting point in our spectral modelling efforts ( [ parameter_estimates ] ) . the leptonic model adopted in this paper is a generic , fully time - dependent one - zone relativistic jet model . the details of the model as well as the numerical procedure adopted to solve the time - dependent electron continuity equation and the photon transport equations are given in @xcite . here , we briefly highlight the salient features of this model . a population of ultrarelativistic , non - thermal electrons ( and positrons ) is injected into a spherical emitting volume of co - moving radius @xmath33 ( the `` blob '' ) at a time - dependent rate . the injected pair population is specified through an injection power @xmath34 and the spectral characteristics of the injected non - thermal electron distribution . we assume that electrons are injected with a single power - law distribution with low and high energy cutoffs @xmath35 and @xmath36 , respectively , and a spectral index @xmath37 . the jet is powered by accretion of material onto a supermassive central object , which is accompanied by the formation of an accretion disk with a bolometric luminosity @xmath38 . the randomly oriented magnetic field @xmath39 in the emission region is parameterized through an equipartition parameter @xmath40 , which is the fraction of the magnetic field energy density , @xmath41 , compared to its value for equipartition with the relativistic electron population in the emission region . the blob moves with relativistic speed @xmath42 along the jet which is directed at an angle @xmath43 ( with @xmath44 ) with respect to the line of sight . the doppler boosting of emission from the co - moving to the observer s frame is determined by the doppler factor @xmath45^{-1}$ ] . as the emission region moves outward along the jet , particles are continuously injected , are cooling , primarily due to radiative losses , and may leak out of the system . we parametrize particle escape through an energy - independent escape time scale @xmath46 with @xmath47 . radiation mechanisms included in our simulations are synchrotron emission , compton upscattering of synchrotron photons ( ssc = synchrotron self compton scattering ; * ? ? ? * ; * ? ? ? * ) , and compton upscattering of external photons ( ec = external compton scattering ) , including photons coming directly from the disk @xcite as well as re - processed photons from the broad line region @xcite . the broad line region is modelled as a spherical shell between @xmath48 and @xmath49 , and a radial thomson depth @xmath50 . @xmath51 absorption and the corresponding pair production rates are taken into account self - consistently . while leptonic models deal with a relativistic e@xmath52 plasma in the jet , in hadronic models the relativistic jet consists of a relativistic proton ( @xmath53 ) and electron ( @xmath54 ) component . here we use the hadronic synchrotron - proton blazar ( spb- ) model of @xcite to model the spectral energy distribution ( sed ) of bl lacaerte in july and november 2000 . like in the leptonic model , the emission region , or `` blob '' , in an agn jet moves relativistically along the jet axis which is closely aligned with our line - of - sight . relativistic ( accelerated ) protons , whose particle density @xmath55 follows a power law spectrum @xmath56 in the range @xmath57 , are injected instantaneously into a highly magnetized environment ( @xmath58 const . within the emission region ) , and are subject to energy losses due to proton - photon interactions ( meson production and bethe - heitler pair production ) , synchrotron radiation and adiabatic expansion . the mesons produced in photonmeson interactions always decay in astrophysical environments . however , they may suffer synchrotron losses before the decay , which is taken into account in this model . if the relativistic electrons are accelerated together with the protons at the same site , their injection spectrum shows most likely the same spectral shape @xmath59 with @xmath60 . in the following we shall assume this as a working hypothesis . the relativistic primary @xmath61 radiate synchrotron photons which constitute the low - energy bump in the blazar sed , and serve as the target radiation field for proton - photon interactions and the pair - synchrotron cascade which subsequently develops . the spb - model is designed for objects with a negligible external target photon component , and hence suitable for bl lac objects . the cascade redistributes the photon power to lower energies where the photons eventually escape from the emission region . the cascades can be initiated by photons from @xmath62-decay ( `` @xmath62 cascade '' ) , electrons from the @xmath63 decay ( `` @xmath64 cascade '' ) , @xmath53-synchrotron photons ( `` @xmath53-synchrotron cascade '' ) , charged @xmath2- , @xmath65- and @xmath66-synchrotron photons ( `` @xmath67-synchrotron cascade '' ) and @xmath68 from proton - photon bethe - heitler pair production ( `` bethe - heitler cascade '' ) . because `` @xmath62 cascades '' and `` @xmath64 cascades '' generate rather featureless photon spectra @xcite , proton and muon synchrotron radiation and their reprocessed radiation turn out to be mainly responsible for the high energy photon output in blazars . the contribution from the bethe - heitler cascades is mostly negligible . the low energy component is dominanted by synchrotron radiation from the primary @xmath54 , with a small contribution of synchrotron radiation from secondary electrons ( produced by the @xmath53- and @xmath67-synchrotron cascade ) . a detailed description of the model itself , and its implementation as a ( time - independent ) monte - carlo code , has been given in @xcite and @xcite . @xcite have derived some model - independent parameter estimates based on the observational results of the bl - lacertae multiwavelength campaign of 2000 , which we will briefly summarize here . these estimates apply to both leptonic and hadronic models , unless specifically noted otherwise . the co - moving magnetic field can be estimated by assuming that the dominant portion of the time - averaged synchrotron spectrum is emitted by a quasi - equilibrium power - law spectrum of electrons . from the observed properties of the synchrotron spectrum , @xcite have derived a magnetic - field estimate of @xmath69 where @xmath70 and @xmath71 with @xmath72 the energy density of the relativistic electrons , and @xmath41 the magnetic field energy density . typically @xmath73 in leptonic models while @xmath74 in hadronic models since @xmath75 ( @xmath76 is the energy density of the relativistic protons ) and @xmath77 . from the modelling in the framework of the spb model we find @xmath78 g ( see sect . 4.2 ) which is in good agreement with the magnetic field values required for this hadronic model . although the tentatively identified time delay between the _ beppo_sax lecs [ 0.5 2 ] kev and the r - band light curves of @xmath79 5 hr was found not to be statistically significant , it is interesting to investigate which magnetic field could be derived if such a correlation was indeed real and the delay was caused by synchrotron cooling of high - energy electrons . this has been done in @xcite , resulting in @xmath80 where @xmath81 is the ratio of the external - photon - field energy density to the magnetic - field energy density in the co - moving frame . as pointed out by @xcite , eq . [ b_rx ] may , in fact , slightly overestimate the actual magnetic field since at least the optical synchrotron emitting electrons may also be affected by adiabatic losses and escape . depending on the details ( geometry and mechanism ) of the jet collimation , those processes can act on time scales as short as the dynamical time scale , which is constrained by the observed minimum variability time scale of @xmath82 hr ( in the observer s frame ) . another note of caution that needs to be kept in mind is that the rather large sampling time scale of the x - ray light curve of @xmath83 hr , precludes the estimation of magnetic fields larger than @xmath84 g from delays between the optical and x - ray light curves . it is remarkable that the two magnetic - field estimates are in good agreement if the doppler factor is slightly larger than 10 and/or the parameter @xmath85 is only slightly less than 1 . we thus conclude that a magnetic field of @xmath86 g might be a realistic value for bl lacertae . this is also in good agreement with magnetic - field estimates for this source based on earlier observations ( e.g. , * ? ? ? * ; * ? ? ? based on the magnetic - field estimate of 1.5 2 g for leptonic models , the approximate location of the synchrotron peak of the seds of bl lacertae at @xmath87 hz allows us to estimate that the electron energy distribution in the synchrotron emitting region should peak at @xmath88 , also in reasonable agreement with earlier estimates for this source @xcite . the location of the synchrotron cutoff in the quiescent state at @xmath89 hz then yields a maximum electron energy in the quiescent state of @xmath90 , while the synchrotron cutoff in the flaring state at @xmath91 hz yields @xmath92 . for hadronic models , the magnetic - field estimate is a factor of @xmath93 higher than in the leptonic case . consequently , the estimates for the co - moving energies of the synchrotron - emitting electrons will be lower by a factor of @xmath94 . the superluminal - motion measurements place a lower limit on the bulk lorentz factor @xmath95 , and we expect that the doppler boosting factor @xmath96 is of the same order . since , unfortunately , we only have an upper limit on the vhe @xmath13-ray flux during the campaign of 2000 , and no measurements in the mev gev regime , no independent estimate from @xmath51 opacity constraints can be derived . however , such an estimate was possible for the july 1997 @xmath13-ray outburst and yielded @xmath97 @xcite , which is a much weaker constraint than derived from the superluminal motion observations . from the optical and x - ray variability time scale , we find an upper limit on the source size of @xmath98 cm . if the electrons in the jet are efficiently emitting most of their co - moving kinetic energy before escaping the emission region ( fast cooling regime ) , then the kinetic luminosity of the leptonic component of the jet would have to be @xmath99 ergs s@xmath19 . if the electrons are in the slow - cooling regime ( i.e. they maintain a substantial fraction of their energy before escaping the emitting region ) and/or the jet has a substantial baryon load ( see , e.g. , * ? ? ? * ) , the kinetic luminosity of the jet would have to be accordingly larger . also , if the magnetic field required to reproduce the synchrotron emission is present continuously throughout the jet , the jet luminosity may actually be dominated by the poynting flux ( see caption of tab . [ equilibrium_fit ] ) . in order to estimate the energy density in the external photon field , an estimate of the average distance of the blr from the central engine is required . this can be achieved in the following way . the most recent determination of the mass of the central black hole in bl lacertae can be found in @xcite . they find a value of @xmath100 . then , if the width of the emission lines measured by @xcite , @xcite , and @xcite is interpreted as due to keplerian motion of the blr material around the central black hole , we find an estimate of the average distance of the line producing material of @xmath101 pc . with this value , we can estimate the co - moving energy density in the external radiation field from the blr compared to the magnetic - field energy density as @xmath102 where @xmath103 ergs s@xmath19 is the bolometric luminosity of the accretion disk , @xmath104 is the bulk lorentz factor , and @xmath50 is the reprocessing depth of the broad line region . the luminosity of the accretion disk is very hard to constrain since it has never been observed directly in bl lacertae . here , we use a standard value of @xmath105 erg s@xmath19 as a typical value for moderately luminous agn . using a value of the luminosity of the reprocessed emission from the blr of @xmath106 ergs s@xmath19 @xcite , this would imply a value of @xmath107 . then , with a magnetic field of @xmath108 g in the case of leptonic models , we find @xmath109 . unfortunately , the lack of a simultaneous mev gev @xmath13-ray observation with our data set prevents us from imposing a tighter constraint on the blr parameters . however , we point out that our basic model assumptions will not be severely affected by moderate variations in the parameters determining @xmath110 . for the case of the leptonic models , @xcite have demonstrated that the spectral and variability patterns observed at optical and x - ray frequencies are only very weakly dependent on an additional contribution from external compton scattering , as long as that contribution is not strongly dominant over other electron cooling mechanisms . in the case of hadronic models , with magnetic fields of order @xmath111 40 g ( see [ hadr_spectra ] ) the estimate on @xmath110 is lower by more than two orders of magnitude . under these circumstances , the a priori assumption of a negligible external photon field in the hadronic model used here is clearly justified . also , because @xmath112 in the latter models ( see [ hadr_spectra ] ) the ssc component does not noticeably contribute to the total flux . the egret data from the @xmath13-ray outburst in 1997 , the highest @xmath13-ray flux ever observed from this source , are included in our figures only as a guideline for an upper limit . those measurements had been accompanied by simultaneous optical and x - ray observations @xcite , which indicate that the source was in a markedly different activity state than during the 2000 campaign . in particular , the asca 2 10 kev x - ray spectrum showed an energy index of @xmath113 , indicating that it might have been entirely dominated by the low - energy portion of the high - energy ( x - ray @xmath13-ray ) spectral component . for this reason , we did not make any attempt to reproduce the 1997 egret data in our model fits . starting with the parameters derived in [ parameter_estimates ] , we have done a series of simulations with our leptonic jet code , letting the electron and photon spectra relax to an equilibrium state . since a moderate contribution from an external radiation field does not severely affect the sed and spectral variability signatures at optical and x - ray frequencies @xcite and we do not have a measurement of the mev gev flux simultaneous with our 2000 campaign data , we set the blr thomson depth to 0 in order to save cpu time . the solid curves in fig . [ mw_combined ] shows our best - fit leptonic models for the two simultaneous seds of july 26/27 , and nov . 1 , respectively . the relevant fit parameters are listed in tab . [ equilibrium_fit ] . the major change of parameters between the quiescent and the flaring state is given by a hardening of the electron spectrum , both through a significant change of the injection spectral index @xmath37 and the high - energy cutoff @xmath36 . in addition , slight changes in the doppler boosting factor @xmath96 and the injection luminosity @xmath114 are required . tab . [ vhe_fluxes ] lists the predicted gev tev fluxes from our spectral fits for threshold photon energies @xmath115 gev , @xmath116 gev , and @xmath117 gev , which have been corrected for @xmath51 absorption by the intergalactic infrared background absorption using the models of @xcite . since we have neglected any contribution from external compton scattering of blr photons , the values listed in tab . [ vhe_fluxes ] should be regarded as lower limits . the predicted flux levels indicate that bl lacertae should be detectable with the new generation of atmospheric erenkov telescope arrays like veritas only in an extreme flaring state . if magic reaches its design goals , it should be able to detect bl lacertae in any state of activity . we note that in all our leptonic and hadronic fits ( see next section ) , our model radio fluxes are far lower than the actual data . this is because our models only follow the evolution of the jet during the early phase of @xmath13-ray production during which radiative cooling is strongly dominant over adiabatic cooling . in this phase , the emission region is highly optically thick out to ghz radio frequencies . we do not follow the further evolution of the jet components through a possible phase of expansion in which they are expected to become gradually optically thin at radio frequencies , because this would necessitate the introduction of several additional , poorly constrained parameters . the evolutionary phase of the emission components followed in our model simulations happens on sub - pc scales , which are not resolveable even with vlbi ( see , e.g. , @xcite for a recent , detailed discussion of vlbi polarimetry of bl lacertae ) since an angular resolution of 1 mas corresponds to a linear scale of @xmath118 pc at the distance of bl lacertae . for this reason , our results are consistent with bl lacertae being a core - dominated radio source even in vlbi images . [ spbfitsnovall ] shows a summary of spb - models representing the data of october 31 november 2 , 2000 , best . the primary electron synchrotron spectrum shows a low - energy break at the synchrotron self - absorption turnover energy of @xmath119 ev , followed by the synchrotron radiation from the injection particle spectrum that is modified by synchrotron losses . the turnover at about a few 100 ev with a subsequent steep tail is due to the cutoff in the electron distribution at particle lorentz factor @xmath120 . this interpretation implies spectral breaks at a few 100 ev energies that are larger than 0.5 . a spectral break between the optical and x - ray band can in principle explain the finding of the optical flux lying significantly below the power - law extrapolation of the _ bepposax _ lecs+mecs spectrum @xcite . the observed color - flux diagram in the r - band ( fig . [ b_r_fit ] ) constrains the electron injection spectra to be not significantly harder than @xmath121 . our model fits ( fig . [ spbfitsnovall ] ) use @xmath122 @xmath123 . a steep spectral decline at soft x - rays is suggested by the _ bepposax _ lecs+mecs data . a high magnetic field of @xmath124 g in the emission region leads to a dominance of synchrotron losses throughout the emitted low - energy component ( the escape loss dominated energy range lies below the synchrotron - self absorption turnover frequency ) . with these magnetic field strengths the optical synchrotron emission is expected to lag the soft x - ray emission by @xmath125 minutes . the synchrotron radiation serves as the target photon field for photon - proton interactions and cascading which determines the radiative output at high energies . the high energy component of the sed is constrained by the rxte / pca data , the _ bepposax _ pds data and the @xmath126 upper limit from hegra . we have also included the egret data from the 1997 outburst ( the highest egret flux ever measured from this source ) as an upper limit in the mev - gev regime . the hardening of the pds spectrum seems to indicate the onset of the high energy component just below 10 kev . in the spb model the pds data can in general be explained by either direct proton synchrotron radiation or a strong reprocessed cascade component . the former possibility , however , requires extremely large doppler factors and/or very high magnetic field strengths which would increase the total jet power to @xmath127 erg / s . such high values are unlikely for low - luminosity bl lac objects . in the following we therefore concentrate on the second option . no variability has been observed with the pds within the exposure time of @xmath128sec in the jet frame . this constrains the parameter space . if the hard x - rays are due to reprocessed proton synchrotron radiation , the magnetic field is limited to values @xmath129 g. for a dominating reprocessed @xmath2/@xmath65-synchrotron component at hard x - rays only ( jet frame ) target photon densities @xmath130 ev @xmath131 are in agreement with no variability within the exposure time . both requirements favor models with dominating @xmath65-production loss rates as compared to proton synchrotron losses . indeed , all models that fit the november 2000 data exhibit strong @xmath2/@xmath65-synchrotron radiation and its reprocessed component while proton synchrotron radiation is almost negligible . as an example we show in fig . [ spbfitsnov ] the contributions of the various cascade spectra to the total emerging radiation for model 1 . the hegra upper limit at @xmath132 gev may potentially limit the maximum proton energy . the emerging high energy photon spectrum at source is , however , modified by photon - photon pair production during propagation through the cosmic background radiation field . the optical depth exceeds unity above 0.4 1.2 tev utilizing the two most extreme background models in @xcite . this absorption effect efficiently prevents photons of energy @xmath132 gev to arrive at earth . another method for estimating the maximum input proton energy is possible through the luminosity of the reprocessed component provided the target photon density allows sufficient reprocessing . in this case , the luminosity of the reprocessed component is dependent on the input proton energies . we find that a limit to the proton injection spectrum of @xmath133 ( due to @xmath65-production losses ) is in agreement with the observations in the x - ray regime . reasonable representations of the observed spectral energy distribution ( sed ) of november 2000 can be found for doppler factors @xmath134 10 ( leading to target photon densities of @xmath135 ev @xmath131 ) , magnetic field strengths between 20 and 40 g and electron and proton injection spectra with spectral indices of @xmath136 1.9 ( see table [ equilibrium_fit ] ) . equipartition is reached within a factor 2 . models with higher doppler factors usually violate the upper limit at tev energies . in all cases the hard x - ray / soft @xmath13-ray band up to @xmath32 mev is dominated by reprocessed @xmath2/@xmath65 synchrotron radiation , which is followed by a broad dip up to gev energies determined by the @xmath62-cascade ( see fig . [ spbfitsnov ] ) . tev photons are expected due to @xmath2/@xmath65 synchrotron radiation , and may be detectable by @xmath137 generation atmospheric cherenkov telescopes like veritas or magic . above @xmath138 gev the spectrum is noticeably modified by the photons interacting with the cosmic background radiation field during propagation . model fit 3 + 4 are in conflict with the hegra upper limit at energies @xmath139 tev only for an extremely thin cosmic background photon field . the model parameters representing the data are chosen to satisfy the following constraints : flux variability provides an upper limit for the size of the emission region . we therefore fix the comoving emission region to @xmath140 d cm for both activity states ( see sect.[parameter_estimates ] ) . the range of bulk doppler factors of @xmath141 10 considered for the fitting procedure is consistent with the superluminal motion of @xmath142 detected by @xcite , and imply viewing angles between @xmath143 and 8 degrees with bulk lorentz factors @xmath144 . these values are also in good agreement with the expectations from unification schemes ( e.g. , * ? ? ? furthermore , approximate equipartition between particles and fields is anticipated . this effectively constrains the magnetic field strengths through eq.([b_eb ] ) . the injection spectral index @xmath145 finds limits from the observed optical colors ( see fig . [ b_r_fit ] ) . the maximum electron ( and proton ; see above ) energy is well constrained by the x - ray observations . in addition , the maximum proton energy achievable by acceleration can never exceed the limit imposed by the larmor motion , which must fit into the space of the emission region . in comparison to november 2000 , bl lacertae was in a lower activity state in july 2000 . [ spbfitsjuly ] shows the simultanous broad band data together with a selection of spb - models representing this state . because both , electrons and protons , are assumed to be accelerated together , the maximum particle energy of each species reached in this process is expected to be correlated ( though not necessarily linearly ) the hard spectrum found by _ rxte _ and _ bepposax _ in july 2000 indicates that the radiation in this band belongs to a separate component from the optical emission detected by the webt campaign , implying a significantly lower cutoff energy of the primary electron population in july 2000 than in november 2000 if the magnetic field strength does not change significantly . it follows that also the maximum proton energies reached in july 2000 should be lower than in november 2000 . indeed , our modeling procedure requires injected proton spectra with a high - energy cutoff at lower energies in july 2000 ( see table [ equilibrium_fit ] ) . in addition , we find bl lacertae s sed in its lower activity state in agreement with doppler factors @xmath141 8 . a comparison with the fit parameters for the november 2000 sed suggests that the bulk lorentz factor might be a relevant parameter for explaing different activity states . for the modeling of fit 1 4 in july 2000 we use @xmath146 g , @xmath147 , @xmath148 and a primary electron - to - proton density ratio @xmath149 . with doppler factors @xmath141 8 the target photon energy density in the jet frame , @xmath150 , is @xmath151 ev @xmath131 . the models predict the high energy power output in the gev - to - tev regime due to @xmath67/@xmath64-synchrotron radiation altered by @xmath51 attenuation in the cosmic backgound radiation field , and a broad `` dip '' in the egret energy range determined by the @xmath65-cascades and extending into the hard x - ray band . the expected flux level at these energies lies close to egret s flux sensitivity ( for a typical exposure ) . the hard x - ray radiation is due to reprocessed @xmath67/@xmath64-synchrotron radiation . a spectral analysis at 0.01 1 mev may reveal a broad curvature in the spectrum . while in all these models proton synchrotron radiation plays only a minor role because of the rather thick target photon field for @xmath152-interactions , we note that also model fits are possible where proton synchrotron emission is the dominant radiation process from x - rays to gev @xmath13-rays . these models , however , require large doppler factors @xmath153 and magnetic field strengths @xmath154 g which leads to jet powers that are unreasonably high for bl lac objects . models involving meson production inevitably predict neutrino emission due to the decay of charged mesons . the spb - model for bl lacertae in 2000 predict a @xmath155-output of about @xmath156 gev s@xmath19 @xmath24 peaking at around @xmath157 gev . the neutrino power at @xmath158 gev is about @xmath159 gev s@xmath19 @xmath24 . neutrino flavor oscillations are not included in these estimates . in summary , the hadronic spb - model predicts tev - emission on a flux level near or below the detectability capabilities of celeste and stacee for bl lacertae , but clearly above the sensitivity limit of future instruments like veritas , magic and h.e.s.s . while leptonic models predict integral fluxes at @xmath160 gev for bl lacertae on a similar level as hadronic models do , ( sub-)tev emission detectable with very high - sensitivity instruments is only predicted for the hadronic emission processes , in contrast to leptonic models ( see tab.2 ) . interestingly , this finding is similar to the case of w comae in 1998 where a similar comparative study has been performed @xcite . high - sensitivity tev observations may therefore be useful as a diagnostic to distinguish between the hadronic and leptonic nature of the high - energy emission at least from some lbls , in addition to its possible neutrino emission . due to the multitude of parameters involved in our models , we may expect that our choice of parameters is not unique . the ambiguities in pure spectral modeling of blazar seds have been drastically demonstrated for the case of w comae by @xcite . in order to refine our choice of parameters for our leptonic model fit and investigate the source of variability of bl lacertae , we have done a detailed parameter study of various plausible flaring scenarios , starting from parameters of our quiescent - state fit . the results of this parameter study have been compared qualitatively with the observed trends in bl lacertae in 2000 in order to pin down the most likely flaring scenario at work in this source ( see [ parameter_study ] ) . based on this result , we have then resumed our fitting procedure to fit simultaneously the sed and optical and x - ray spectral variability patterns consistently in one complete model ( [ var_fit ] ) . the variability of blazars can in principle be initiated by a multitude of physical processes , all of which would imply specific changes in the fundamental modeling parameters of leptonic jet models . for the purpose of a qualitative comparison with the observed spectral variability patterns of bl lacertae , we have done a series of simulations , focusing on a fluctuation of one or 2 of the basic model parameters , leaving all other parameters unchanged : ( a ) the total injection luminosity of ultrarelativistic particles into the jet @xmath114 , ( b ) the injection spectral index @xmath37 , ( c ) the high - energy cutoff of the electron injection spectrum @xmath36 , and ( d ) a combination of electron spectral hardening from ( b ) and ( c ) . other scenarios like a fluctuation in the doppler factor @xmath96 or the magnetic field @xmath39 only can be ruled out immediately by virtue of the observed spectral variability . in our simulations , we have represented a parameter fluctuation by a change to a new parameter value over a time @xmath161 , and then switching back to the equilibrium value . in the case of simulations ( b ) ( d ) there is still an ambiguity concerning the choice of the normalization of the electron injection spectrum under spectral fluctuations . we have executed the suite of simulations ( b ) ( d ) for two extreme assumptions : ( 1 ) leaving the injection power @xmath162 constant between the equilibrium state and the simulated flare , and ( 2 ) leaving the total number of injected electrons per unit time constant . we have found that the optical and x - ray spectral variability patterns for those two cases do not differ substantially from each other . the results presented in the following paragraphs concerning fluctuating electron spectral parameters refer specifically to the case of unchanged @xmath162 . a typical set of simulation results is illustrated in figs . [ sed_34 ] and [ lc_34 ] . from the simulated , time - dependent spectra and light curves , we have calculated optical r - band magnitudes and color indices r. we have folded the simulated x - ray fluxes through the detector response of _ bepposax _ , using the exact same effective area curves as used in the data analysis of @xcite to evaluate the resulting _ bepposax _ count rates and hardness ratios as mentioned in [ observations ] . the simulated optical and x - ray spectral variability patterns from our flaring scenarios ( a ) ( d ) are compared in fig . [ sv_comparison ] . first , we note that a model with a fluctuation of only the injection luminosity ( a ) is predicting very limited x - ray spectral variability and does not lead to the characteristic , positive brightness hardness correlation observed at optical frequencies . such a scenario thus seems unlikely to be the driving mechanism behind the variability of bl lacertae . our model simulation with a fluctuation of the electron spectral index @xmath37 only ( b ) qualitatively reproduces the optical color magnitude relation and the hardness intensity anti - correlation at soft x - rays . it appears to be capable of reproducing a weak positive hardness intensity correlation at harder x - rays ( hr2 vs. mecs 4 10 kev count rate ) , which has occasionally been observed during our campaign . we conclude that such a scenario has a good potential to reproduce all the optical and x - ray spectral variability patterns observed during the 2000 campaign on bl lacertae . a scenario invoking primarily a fluctuation in @xmath36 ( c ) predicts a very small amplitude of optical variability , compared to the x - ray variability amplitude . it does predict a strong flux hardness anti - correlation at soft x - rays , as observed in bl lacertae , but fails to reproduce the optical color magnitude correlation . we therefore conclude that this mechanism is not consistent with the observed spectral variability of bl lacertae either . a scenario of combined changes of @xmath37 and @xmath36 ( d ) does qualitatively reproduce both the optical color magnitude correlation and the soft x - ray hardness intensity anti - correlation , but also predicts a strong hardness intensity anti - correlation at harder x - rays ( hr2 vs. mecs 4 10 kev count rate ) , which has not been observed by _ bepposax_. in summary , we find that our flaring scenario ( b ) , based on a hardening of the electron injection spectral index @xmath37 only , seems to be the most promising candidate for modeling the sed and spectral variability of bl lacertae . we are now ready to narrow down the parameter choices to model simultaneously the sed and spectral variability of bl lacertae in 2000 . for this purpose , we are first choosing parameters similar to the low state of july 26/27 , but with the higher doppler factor of @xmath163 to achieve approximate agreement with the average optical flux level around nov . 1 and the hard _ rxte _ pca spectrum measured just prior to the flaring episode caught during the second _ beppo_sax pointing on oct . 31 nov . 2 , 2000 . various scenarios of short - term fluctuations of the electron spectral index over @xmath164 were tested and compared with the observed sed , optical color magnitude correlation and the _ beppo_sax hardness intensity correlations for individual flares during the oct . 31 nov . 2 , 2000 , pointing . satisfactory agreement with all three of these observational results was achieved for the following choice of parameters : @xmath163 ; @xmath165 , @xmath166 , @xmath167 outside the flaring episode , changing to @xmath168 during the flare ; @xmath169 ergs s@xmath19 , @xmath170 , yielding a magnetic field of @xmath171 g. the broadband spectral evolution resulting from this flaring scenario is illustrated in fig . [ mw_fit ] . it indicates how this flaring scenario reproduces the hard x - ray spectrum seen by pca right before the flaring episode , and switches to the synchrotron - dominated soft high - flux spectrum during the flare . the light curves at optical , x - ray and @xmath13-ray frequencies resulting from this simulation are shown in fig . [ lightcurve_fit ] . the significantly larger flaring amplitude at x - rays compared to optical frequencies is clearly well reproduced . the flaring amplitude is largest at the highest @xmath13-ray energies , where the flux increases by almost 2 orders of magnitude , to reach levels well above the anticipated , nominal detection threshold of magic . the results of our leptonic fit simulation are compared to the observed optical color magnitude correlation and to the time averaged emission from our hadronic fits in fig . [ b_r_fit ] . we see that the hadronic fit for the low state lies well within the observed range of optical colors and r - band magnitudes , while the optical color predicted from the high - state fit is harder by @xmath172 ( corresponding to a difference in the local spectral index of @xmath173 ) than the observed b - r values . the leptonic flaring fit coincides reasonably well with the range of r magnitudes and b - r colors observed during the active phase after sept . 2000 , though the actual simulated spectral hysteresis curve lies slightly above the observed correlation . we have explored multiple attempts to remedy this slight discrepancy , but could not find a better representation of the data than the one shown in fig . [ b_r_fit ] which would still be consistent with both the sed and the x - ray variability patterns discussed below . however , the difference is minute even for the hadronic model fit , and may be explained by uncertainties in the adopted de - reddening and the subtraction of the host galaxy contribution ( for an in - depth discussion of these issues , see * ? ? ? * ; * ? ? ? [ hid30_fit ] compares our simulated x - ray spectral hysteresis curves to the observed hardness - intensity correlation during a well - resolved x - ray flare at 0.5 hr 6.5 hr ut on nov . 1 , 2000 . the figure illustrates that the overall flux levels and hardness ratio values are well within the observed range , and that the time evolution of those values is consistent with the results of our simulation . clearly , the statistical errors on the _ beppo_sax count rate and hardness - ratio measurements are too large to test for the existence of the actual spectral hysteresis phenomena predicted in our simulation . our fit predicts slight counterclockwise spectral hysteresis for our favoured sed + spectral variability fit . future observations using , e.g. , _ chandra _ or _ xmm - newton _ would be extremely useful to test this prediction . finally , we discuss a possible 4 5 hr delay of the optical fluxes with respect to x - ray flares for which a statistically not significant hint in the bl lac data of 2000 was found . all of our simulations discussed in this section did not lead to a systematic time delay with significant flux peak separation in time between the x - ray and optical flares . such a feature might be expected in a scenario where solely a high - energy population of electrons is injected into the jet , which subsequently cools due to radiative losses . we have run a simulation , similar to the ones described above , but injecting only a narrow distribution of ultrarelativistic electrons into the jet during the flare . such a scenario does reproduce the spurious 4 5 hr delay , but would predict a strong anti - correlation between optical flux and hardness , in contradiction with the observed color - magnitude relation . thus , we conclude that such a scenario can be ruled out . overall , our parameter choice for the leptonic models for bl lacertae are in reasonable agreement with those found by other authors based on earlier multiwavelength campaigns on this source ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in agreement with those authors , our best - fit doppler factor of @xmath163 is well within the range typically found for blazar modelling , and the magnetic field of @xmath174 g is intermediate between typical values found for leptonic modeling of flat - spectrum radio quasars ( fsrqs ) and high - frequency peaked bl lac objects ( hbls ) . while fsrqs are usually successfully modelled with @xmath175 a few g ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , typical values found for hbls are of the order of @xmath176 g ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in our analysis of the spectral variability patterns , we have found that those patterns can successfully be modelled with a fluctuation of the electron injection spectral index . remarkably , our time - dependent fits indicate that an injection index larger than @xmath177 , even during the peak of an individual short - term flare , is required . if the injection of ultrarelativistic electrons into the emitting volume is caused by fermi acceleration at relativistic shocks , detailed numerical studies have shown that with fully developed turbulence in the downstream region , a unique asymptotic index of @xmath178 2.3 should be expected ( e.g. , * ? ? ? * ; * ? ? ? however , recently @xcite have shown that fermi acceleration might lead to drastically steeper injection spectra if the turbulence is not fully developed . furthermore , depending on the orientation of the magnetic field at the shock front , an abrupt steepening of the injection spectra may result if the shock transits from a subluminal to a superluminal configuration . in this context , our leptonic fit results may indicate that such predominantly geometric effects , may be the cause of the rapid variability observed in bl lacertae . in their analysis of the _ bepposax _ + optical continuum spectra of the oct . 2 observations , @xcite have noticed that the time averaged optical and lecs + mecs x - ray spectra can not be connected smoothly using a single power - law or a smoothly connected broken power - law . they have suggested and investigated several possibilities how this discrepancy could be remedied , including a variable dust - to - gas ratio , the bulk compton process @xcite , a multi - component emission model , and a spectral upturn resulting from klein - nishina effects on the electron cooling rates . the results of our combined leptonic spectral + variability modeling of bl lacertae suggest that the discrepancy ultimately arises artificially as a result of the time averaging involved when producing the high - quality _ beppo_sax lecs + mecs spectrum . the fact that our time - dependent leptonic fits reproduce the observed ranges of optical and x - ray fluxes and spectral indices simultaneously resolves the issue of this optical x - ray spectral discrepancy , and removes the need for any additional assumptions concerning intergalactic absorption and/or electron populations in the jet . in contrast to leptonic models , the hadronic spb model required significantly larger magnetic field strengths ( of order 20 40 g ) on the length scale of the size of the emission region of @xmath179 cm . the range of doppler factors appears about a factor of 2 lower than in the leptonic models . the total jet power , which turns out to be below the estimated accretion disk luminosity , remains larger in hadronic models than in leptonic ones owing to the higher particle and field energy content . the state transition from low to higher activity in 2000 is well described by an increase of the co - moving particle energy and the bulk lorentz factor . in the picture of diffusive shock acceleration ( in the test particle limit ) the maximum particle energies are related to the magnetic turbulence spectrum ( see , e.g. , * ? ? ? * ; * ? ? ? * and references therin ) . the required maximum electron and proton energies in the presented models can be understood if the particles gain energy by diffusive shock acceleration in a @xmath180 turbulence spectrum where @xmath110 is the wave number in the turbulent magnetic field . spectral - index changes can not be ruled out , but they are not the dominant cause of the spectral variability of bl lacertae in the framework of the presented modeling using the spb model . an interesting diagnostic for the particle content in the jet in addition to any neutrino detections might be achieved through high sensitivity observations in the ( sub-)tev energy range by e.g. magic or veritas . while both , leptonic and hadronic models , predict a similar flux level in the gev - energy range , hadronic models predict about an order of magnitude higher flux values than leptonic ones do above 40 gev for bl lacerate in 2000 . furthermore , considering the results of our variability study , the predicted vhe @xmath13-ray flux from a leptonic jet only reaches the peak level mentioned above during short flares . it depends critically on the duty cycle of such flaring events whether a sufficient time - averaged level of vhe flux can be sustained for low - energy - threshold cherenkov telescopes like magic to accumulate a measurable signal . our leptonic sed + spectral variability fit predicted spectral hysteresis at soft x - ray energies which might serve as a confirmation of our fit results . the limited count statistics of our _ beppo_sax observations did not firmly establish nor rule out the existence of x - ray spectral hysteresis . more sensitive , dedicated observations by _ chandra _ and/or _ xmm - newton _ would be extremely helpful to test this prediction . according to our hadronic model fits presented in this paper , flares of bl lacertae were primarily caused by increasing particle energies . if this is indeed the dominant flaring mechanism and it is not accompanied by significant changes of the electron spectral index , then short - term x - ray spectral variability might be reasonably well represented by the patterns resulting from our leptonic models with increasing @xmath36 with at most very moderate spectral - index changes . these did not show significant spectral hysteresis . however , fluctuations of the electron injection spectral index could not be excluded in our hadronic fits . consequently , the presence of pronounced soft x - ray spectral hysteresis in bl lacertae may slightly favour the leptonic models , while its absence would favour hadronic models , but the discriminating power of such a measurement in the case of bl lacertae would be rather limited . in this paper , we have presented the results of detailed numerical modeling of the seds and spectral variability of bl lacertae in 2000 , using both leptonic and hadronic ( spb ) jet models . details of the data analyses and observational results have been published in three previous papers on this campaign @xcite . the main results of our modelling work are : * both leptonic and hadronic models are able to provide acceptable fits to the seds of bl lacertae in 2000 , both in the low activity state on july 26/27 and in the high activity state on oct . in addition to the naturally much higher overall jet power required for hadronic models ( @xmath181 ergs s@xmath19 vs. @xmath182 ergs s@xmath19 [ depending on the possible poynting - flux contribution ] ) , the hadronic spb model requires a factor of @xmath93 higher magnetic fields ( @xmath183 40 g vs. @xmath4 g ) and a significantly lower bulk lorentz factor ( @xmath184 9 vs. @xmath185 ) . * considering time - averaged emission during the two intensity states , hadronic models predict a sustained level of multi - gev tev emission which should be detectable with second - generation atmospheric cherenkov telescope systems like veritas , hess , or magic . in contrast , our leptonic model only predicts a peak flux exceeding the anticipated nominal magic sensitivity during short flares ; the accumulated fluence over observing time scales of several hours might not be sufficient for a significant detection . thus , a future vhe detection of bl lacertae would be a strong indication for hadronic processes being at work in this object . * a parameter study of various spectral variability scenarios in the framework of our leptonic jet model revealed that the observed optical and x - ray spectral variability in bl lacertae in 2000 can be reproduced through short - term fluctuations of only the electron injection spectral index , with all other parameters remaining unchanged . our simulation of this scenario predicts counter - clockwise spectral hysteresis at x - ray energies . such hysteresis was not predicted in the specific spb model fits presented in this paper , but could not clearly be ruled out either . thus , sensitive spectral - hysteresis measurements of bl lacertae could possibly serve as a test of our modeling results and a secondary diagnostic to distinguish between leptonic and hadronic models , though , by itself , it would not be sufficient as a model discriminant . * the previously noted discrepancy between the time - averaged optical and x - ray spectra may be resolved by considering the spectral variability . our successful modeling of the observed time - dependent flux and hardness values at optical and x - ray frequencies in the framework of a leptonic model effectively removes the need for additional assumptions concerning additional particle populations , extreme klein - nishina effects on the electron cooling rates , and/or anomalies in the intergalactic absorption . we thank m. ravasio for sending us the bepposax effective area curves used in the data analysis of the bl lac 2000 campaign , and the anonymous referee for a constructive report which has definitely helped to improve the clarity of the manuscript . ar thanks the bundesministerium fr bildung und forschung for financial support through desy grant verbundforschung 05ch1pca6 . marscher , a. p. , jorstad , s. j. , aller , m. f. , mchardy , i. m. , balonek , t. j. , terasranta , h. , & tosti , g. , 2004 , in x - ray timing 2003 : rossi and beyond , ed . p. kaaret , f. k. lamb , & j. h. swank , aip conf . proc . , in press cccccccccc lept . , july 26/27 & 1100 & @xmath190 & & & 2.4 & @xmath191 & 1.4 & @xmath192 & 16 + lept . , nov . 1 & 1100 & @xmath193 & & & 2.15 & @xmath194 & 1.4 & @xmath192 & 18 + + hadr . , july 26/27 & 1 & @xmath195 & @xmath196 & 2.7 & 1.9 & @xmath197 & 40 & @xmath198 & 7 + hadr , nov . 1 & 1 & @xmath199 & @xmath200 & 1.9 & 1.9 & @xmath201 & 40 & @xmath202 & 9 + [ equilibrium_fit ] cccc lept . , july 26/27 & @xmath203 & @xmath204 & @xmath205 + lept . , nov . 1 & @xmath206 & @xmath207 & @xmath208 + + hadr . , july 26/27 & @xmath209 & @xmath210 & @xmath211 + hadr . , nov . 1 & @xmath212 & @xmath213 & @xmath214 + [ vhe_fluxes ]

we investigate thermalization processes occurring at different time scales in the yang - mills - higgs system at high temperatures . we determine the largest lyapunov exponent associated with the gauge fields and verify its relation to the perturbatively calculated damping rate of a static gauge boson .

hep-th9608181

to describe phenomena like heavy ion collisions or the electroweak phase transition in the early universe , the methods of quantum field theory must be applied to finite densities and temperatures . the most reliable approach , especially for the case of qcd , are lattice calculations in euclidean space - time ( for a recent overview see e.g. @xcite ) . unfortunately the results are only applicable to equilibrium processes . dynamical processes in quantum field theory , on the other hand , are most commonly studied using perturbation theory . in theories with interacting massless degrees of freedom , however , naive perturbation theory in powers of the coupling constant breaks down at high temperature due to infrared divergences caused by the exchange of massless modes . to obtain finite results one must resum classes of feynman diagrams which lead to dynamical mass generation and thus regulate the infrared singularities . to do this systematically one must disentangle the influence of different momentum scales ; in qcd at finite temperature , these are characterized by @xmath0 , @xmath1 , and @xmath2 where the last two give the order of magnitude of the dynamical masses for the originally massless electric and magnetic gluonic modes , respectively . an improved one - loop perturbation scheme based on these principles , called hard thermal loop resummation , was developed by braaten and pisarski @xcite ( see also @xcite for an overview ) . using effective field theory methods , the above separation of the different momentum scales was achieved by braaten and nieto for static quantities at high temperature @xcite . at high temperature , it is reasonable to expect that phenomena which are dominated by long wavelength modes can be reliably described by classical field theory . this is the basic motivation for the work presented here where we study the classical dynamics of the coupled yang - mills - higgs system . it is well known that the classical non - abelian gauge field equations are nonintegrable and exhibit dynamical chaos @xcite . therefore one has to rely on numerical methods to study the evolution of classical fields . the real time evolution of classical su(@xmath3 ) yang - mills fields on a lattice , for @xmath4 , was recently studied in ref . starting with randomly chosen initial conditions for the gauge fields on each lattice link it was found that the system shows chaotic behavior , i.e. the distance in phase space between two adjacent configurations increases exponentially in time . for systems with sufficiently weak coupling , the time - averaged growth rate ( maximal lyapunov exponent ) @xmath5 was shown to scale linearly with the total energy of the system . the slope depends on the number @xmath3 . even if one starts with highly off - equilibrium configurations , the gauge fields thermalize very fast , i.e. the energy distribution over the lattice soon exhibits a thermal shape . therefore the total energy of the system can be related to the associated temperature , and the scaling of the maximal lyapunov exponent with the total energy can be reinterpreted as a scaling with the temperature @xmath0 at least in the weak coupling regime . precisely the relation @xmath6 which connects dimensionless quantities was established . here @xmath7 is the lattice spacing and @xmath8 the coupling constant . one surmises that the actual value of the proportionality factor @xmath9 survives in the continuum limit @xmath10 . for the classical su(2 ) and su(3 ) gauge field systems , it was found in @xcite that the maximal lyapunov exponent coincides numerically with twice the damping rate @xmath11 of a gauge boson at rest as calculated quantum field theoretically @xcite using thermal perturbation theory with hard thermal loop resummation : @xmath12 at first sight it is surprising that the lyapunov exponent which is deduced from the classical field dynamics by an exact ( i.e. non - perturbative ) numerical simulation should be related at all with the plasmon damping rate which is calculated perturbatively in quantum field theory . however , the close relation between linear response functions calculated in the hard thermal loop approximation and classical transport phenomena has been known for many years @xcite , indicating that the hard thermal loops are essentially classical ( see also @xcite ) . in addition it was argued in @xcite that the damping of a gluon and the exponential growth of the distance between two adjacent classical trajectories are related . so far there is , however , no rigorous proof for the relation ( [ scaling ] ) . an empirical approach to the question whether the coincidence between the lyapunov exponent and ( twice ) the damping rate is only accidental would be to simulate other systems on the lattice and also calculate perturbatively their damping rates . a natural extension of the pure su(@xmath3 ) system is to couple it to a matter field . in this article we study the real - time evolution of the coupled su(2 ) yang - mills - higgs system . again we are interested in the thermalization properties and the chaoticity of the fields . to compare with the high temperature results from hard thermal loop calculations we concentrate on the symmetric phase , i.e. we simulate the higgs behavior in the electroweak plasma above the phase transition in the @xmath13 limit . our work was partly motivated by the recent study presented in ref . @xcite in which the question of the connection between the damping rate and the lyapunov exponent in such a system was also addressed @xcite . this paper provides in particular the analytical results for the plasmon damping rates for the gluon and higgs fields in the hard thermal loop approximation . however , in ref . @xcite the thermalization properties of the classical fields were not taken into account , and thus the determination of the temperature of the system was not dealt with in a satisfactory manner . the correct extraction of the temperature is , of course , crucial to establish relations like ( [ lyapscal ] ) . we postpone a more detailed assessment of the conclusions of ref . @xcite until after presenting our own results . the organization of the paper is as follows : in sec . [ sec : form ] we formulate the problem . the thermalization properties of the fields are studied in sec . [ sec : therm ] . in sec . [ sec : lyap ] we compute the lyapunov exponents , and we discuss our results and compare them with previous work in sec . [ sec : sum ] . in this section we present the realization of the coupled su(2 ) yang - mills - higgs system on the lattice , describe the initialization of the field variables , discuss the allowed ranges of the relevant parameters in the weak coupling and continuum limits , define the lyapunov exponents and describe how we measure them . in the temporal gauge @xmath14 , the hamiltonian of the coupled su(2 ) yang - mills - higgs system in the continuum is given by @xmath15 \ , . \label{h - higgs}\ ] ] the dot denotes the time derivative , and @xmath16 is a charged higgs field in the fundamental representation of su(2 ) . the gauge field is described by vector potentials @xmath17 . @xmath18 denote the electric fields , @xmath19 the magnetic fields , and @xmath20 the covariant derivatives . as mentioned before , the equations of motion derived from the hamiltonian ( [ h - higgs ] ) are not integrable . one has to rely on numerical methods to study the time evolution of the fields . for numerical purposes we discretize the spatial coordinates by using a cubic lattice with @xmath21 sites . time is kept as a continuous variable . for details about the numerical simulation of the real - time evolution of the system we refer to @xcite . to realize the coupled su(2 ) yang - mills - higgs system on the lattice without losing its gauge symmetry properties we use link variables @xcite @xmath22 instead of vector potentials @xmath23 . here @xmath24 are the pauli matrices . the covariant derivative acting on the higgs field is replaced by the difference @xmath25 the electric field strength on the lattice is related to the link variables and their time derivatives by @xmath26 the magnetic energy is expressed in terms of the so - called plaquette operator @xmath27 , which is the product of all four link variables on an elementary plaquette with sites @xmath28 : @xmath29 with @xmath30 . the links are directed and hence the plaquettes are oriented . in the continuum limit the plaquette variable @xmath31 is related to the local magnetic field @xmath32 by @xmath33 it is useful to write the su(2 ) matrices in the form @xmath34 where the @xmath35 are four real numbers , which can be thought of as components of a quaternion . since @xmath36 , each link variable @xmath37 can be expressed in terms of three angles , @xmath38 , @xmath39 , @xmath40 , characterizing a three dimensional hypersphere : @xmath41 with a three dimensional unit vector @xmath42 similarly the four independent components of the higgs field can be represented as a quaternion : @xmath43 in contrast to the link variable @xmath37 the higgs quaternion @xmath44 has arbitrary length @xmath45 defined by @xmath46 again we can introduce three angles to get the representation @xmath47 using the quaternion representations for the link variables and the higgs fields , the hamiltonian of the coupled su(2 ) yang - mills - higgs system on the lattice reads @xmath48 where we define the electric energy @xmath49 the magnetic energy @xmath50 the higgs kinetic energy @xmath51 and the higgs potential energy @xmath52 and the @xmath53 s are energies associated with the respective site , link , or plaquette : [ epsilon ] @xmath54 \ , , \label{emag}\\ \epsilon_{\rm kin}(x ) & = & a^3 \left [ { \textstyle { 1\over 2 } } \ , { \rm tr } \left ( \dot \phi_x^\dagger \dot \phi_x \right ) \right ] \ , , \label{ekin}\\ \epsilon_{\rm pot,2}(x ) & = & 6a \left[{\textstyle { 1\over 2 } } \,{\rm tr}(\phi_x^\dagger \phi_x ) \right ] \ , , \label{ephi}\\ \epsilon_{\rm pot,4}(x ) & = & a^3 \lambda \left [ { \textstyle { 1\over 2}}\ , { \rm tr } \left ( \phi_x^\dagger \phi_x \right ) \right]^2 \ , , \ \label{elambda}\\ \epsilon_{\rm int}(x , i ) & = & - a \,{\rm tr } \left(\phi_x^\dagger u_{x , i } \phi_{x+i}\right ) \ , . \label{egh } \end{aligned}\ ] ] to derive the last equation we have exploited the unitarity of the link variable @xmath37 and the fact that the trace of a quaternion is real . the hamiltonian ( [ htot ] ) is scale invariant . to see this we define the dimensionless variables @xmath55 in terms of these we get @xmath56 \label{hscal}\\ & + & \sum\limits_{x } { \textstyle { 1\over 2 } } \ , { \rm tr } \left ( \dot { \bar\phi}\vphantom{\bar\phi}_x^\dagger \dot { \bar \phi}_x \right ) + \sum\limits_{x , i } { \rm tr } \left [ \bar\phi_x^\dagger \bar\phi_x - \bar\phi_x^\dagger u_{x , i } \bar\phi_{x+i } \right ] + \bar\lambda \sum\limits_x { \textstyle { 1\over 4 } } \left [ { \rm tr } \left ( \bar\phi_x^\dagger \bar\phi_x \right ) \right]^2 \nonumber \end{aligned}\ ] ] where the dot now denotes the derivative with respect to the scaled time @xmath57 . therefore the only parameters in the system are the scaled energies and the scaled higgs self coupling @xmath58 while the lattice spacing and the gauge coupling constant do not appear explicitly . it is easy to derive the equations of motion from the lattice hamiltonian ( [ hscal ] ) and the relations ( [ eudot ] ) and ( [ uquat ] ) : [ eom ] @xmath59 \ , , \label{udot}\\ \dot { \bar e}\vphantom{e}^a_{x , i } & = & i \sum\limits_j \ , { \rm tr } \left ( \tau^a\,u_{x , ij } \right ) + { \bar j}^a_{x , i } \ , , \label{eome}\\ \ddot { \bar \phi}_x & = & - \bar\lambda \ , { \rm tr } \left(\bar\phi_x^\dagger \bar\phi_x \right ) \bar\phi_x - 6 \ , \bar\phi_x + \sum\limits_i \left ( u_{x , i}\ , \bar\phi_{x+i } + u_{x - i , i}^\dagger \bar\phi_{x - i } \right ) \ , . \label{phiddot } \end{aligned}\ ] ] here @xmath60 is the lattice version of the gauge - covariant isospin current of the higgs field . in addition to satisfying these equations of motion the physical solutions must obey gauss law @xmath61 with the covariant derivative @xmath62 in the adjoint representation . the expression on the l.h.s . of ( [ gauss ] ) is a conserved quantity since it commutes with the hamiltonian of the system . thus to satisfy gauss law for all times we must only make sure that our initial conditions obey ( [ gauss ] ) . for a study of the thermalization properties of the different degrees of freedom the energy transfer between different sectors of the system is an important quantity . as we will see below the energy transfer from the gauge part of the system to the higgs part is of particular interest . it is given by @xmath63 as can be easily derived from the equations of motion . the local higgs isospin current @xmath64 is defined in ( [ hj ] ) . we initialize the system in the same way as in @xcite by choosing random initial configurations : * the isospin directions of the higgs fields , characterized by the angle variables @xmath65 , @xmath66 , @xmath67 are chosen randomly for each lattice site . to keep control over the initial potential energy of the higgs fields the length of each higgs quaternion is _ not _ randomly chosen . instead we take for simplicity the same length @xmath45 for all @xmath68 initially . obviously the initial higgs potential energy ( [ hpot ] ) increases with @xmath45 . * the isospin directions of the gauge fields are also randomly chosen . in this case , however , this does not mean that _ all _ the angles are arbitrary . the reason is that not only the isospin direction of the vector potential @xmath69 but also its amplitude enters the link variable @xmath37 as an angle variable . it is easy to see that @xmath38 is connected to the amplitude while @xmath39 and @xmath40 characterize the isospin direction . therefore the latter two angles are initialized with random values on each link . to limit the initial magnetic energy we restrict the gauge field amplitude . thus we introduce a parameter @xmath70 which controls the initial magnetic energy and choose @xmath38 on each link randomly but within the range @xmath71 $ ] . for @xmath72 the initial magnetic energy also vanishes . for small @xmath70 the relation @xmath73 holds . * to satisfy gauss law ( [ gauss ] ) in an easy way we choose initially vanishing field momenta on each link , i.e. @xmath74 . this implies that the initial electric energy and the higgs kinetic energy , ( [ hel ] ) and ( [ hkin ] ) respectively , vanish . note that with non - vanishing potential but vanishing kinetic energies the system is initially in a highly off - equilibrium state . when initialized in this way the system s further evolution is determined by the three parameters @xmath70 , @xmath45 , and @xmath58 , or by the initial magnetic energy and the initial higgs potential energy ( quadratic and quartic part ) , respectively . to make contact with perturbative calculations it is necessary to stay in the weak coupling regime . furthermore , weak coupling is required for the classical calculation to be a valid approximation to the full quantum treatment . this puts constraints on the range of the physical parameters @xmath75 . the scale invariance of the hamiltonian allows to fix the gauge coupling at @xmath76 and the lattice spacing at @xmath77 . weak coupling within the gauge sector is realized if the link variables are in the vicinity of unity ( cf . ( [ eq7 ] ) ) . in the initial state this is achieved by choosing @xmath70 small which implies that @xmath38 in ( [ rhog ] ) is small . on the other hand , as first noted in @xcite and recently worked out in @xcite , one encounters finite size artifacts in the pure su(2 ) system when using too small values of @xmath70 on a finite lattice . to explore the weak gauge coupling regime one is thus restricted to a window of not too small and not too large values for @xmath70 , or one must take very large lattices . with the latter option one soon runs into computer limitations . from the hamiltonian ( [ hscal ] ) one sees that weak coupling within the higgs sector is ensured if the quadratic terms dominate the quartic one . a system with weak higgs self coupling can thus be initialized by requiring @xmath78 to ensure weak coupling between the gauge and the higgs fields , the higgs isospin current @xmath79 in ( [ eome ] ) should be small compared to the magnetic energy contribution ) is @xmath80 . ] . estimating both contributions by the ( maximal ) field amplitudes we get the condition @xmath81 the continuum limit also puts constraints on the possible parameter ranges . first , as seen from eq . ( [ eq7 ] ) , to recover the continuum limit from the lattice formulation one needs @xmath82 . if the lattice simulations are supposed to yield a reasonable approximation to continuum behaviour , @xmath70 must therefore not be large . another lattice artifact shows up if the total energy in the system is too high . due to the compactness of the space of link variables the magnetic energy on the lattice ( [ hmag ] ) is limited from above by @xmath83 = 24 n^3 \ , . \label{magmax}\ ] ] if the total energy in the system is much larger than this value , equipartition of the energy among the magnetic , electric and higgs degrees of freedom may thus not be possible . all these considerations must be taken into account when selecting the initial parameters @xmath70 and @xmath45 as well as the value @xmath84 for the higgs self - coupling . in our calculations we have made sure to choose these parameters in such a way that we are simultaneously in the weak coupling regime and close to the continuum limit . in the next section we will study the thermalization properties of systems initialized as described above . the self - thermalization of the fields is strongly influenced by the chaoticity in its dynamics which is characterized by a set of positive lyapunov exponents . we will here concentrate on the largest lyapunov exponent associated with the most unstable dynamical mode . in principle one can define two such exponents , @xmath85 and @xmath86 , for the gauge and higgs fields , respectively , by using two different distance measures in the space of field configurations . the first measures the growth rate of the euclidean distance in the gauge field sector , @xmath87 = \sum\limits_{x , i } \sum\limits_{\alpha = 0}^3 \left [ ( ( u_\alpha)_{x , i } - ( u'_\alpha)_{x , i } ) ^2 + ( ( \dot u_\alpha)_{x , i } - ( \dot u'_\alpha)_{x , i } ) ^2 \right ] \ , , \label{gdist}\ ] ] where @xmath37 and @xmath88 denote the gauge fields associated with two solutions of the equations of motion ( [ eom ] ) resulting from two initially very close field configurations . the other lyapunov exponent measures similarly the growth rate in the higgs sector , @xmath89 = \sum\limits_{x } \sum\limits_{\alpha = 0}^3 \left [ ( ( \phi_\alpha)_{x } - ( \phi'_\alpha)_{x } ) ^2 + ( ( \dot \phi_\alpha)_{x } - ( \dot \phi'_\alpha)_{x } ) ^2 \right ] \ , , \label{hdist}\ ] ] where @xmath44 and @xmath90 are the corresponding higgs fields from the same two solutions . as we will see in the following , the two lyapunov exponents @xmath85 and @xmath86 turn out to be identical . as usual the lyapunov exponent is defined as @xmath91 to avoid the problem of distance saturation @xcite in directly measuring the lyapunov exponent , we use the rescaling method for the evaluation of the large - time limit required in ( [ lyap_def ] ) ( see @xcite for more details ) . in principle other definitions for the distance between field configurations are possible . in particular , one might wish to replace the higgs distance ( [ hdist ] ) by a gauge invariant definition like e.g. @xcite @xmath92 however , the euclidean distances defined above are more appropriate for the rescaling method . for the pure su(2 ) gauge system it was carefully checked that different distance measures give rise to the same lyapunov exponent @xcite . in this section we investigate various thermalization processes in the coupled su(2 ) yang - mills - higgs system . we will encounter different types of thermalization processes which operate on different time scales . in particular we find that , starting with the initialization described in sec . [ sec : form ] , the gauge and the higgs subsectors each thermalize rather rapidly , while the equipartition of energy between them requires a much longer time . in the following two subsections we first present results on the fast thermalization processes associated with the gauge and the higgs subsystems , and then discuss the system s long - time behaviour . for reasons given in sec . [ sec : form ] our study is confined to certain parameter ranges which are consistent with weak coupling and proximity to the continuum limit . thermalization processes in a large system can be studied by following the evolution of the energy distribution over the microscopic degrees of freedom . in general , the distribution should approach that of a gibbs ensemble upon equilibrium . as shown in the cases of pure su(2 ) and su(3 ) gauge fields @xcite , by measuring appropriate energy distributions , one can determine a parameter @xmath93 which we call the _ apparent temperature _ henceforth . the apparent temperature @xmath93 is related to the system s true temperature @xmath0 , but generally differs from @xmath0 , if the degrees of freedom of the system are not all independent , such as for a gauge theory . we will test the degree of thermalization in various subsectors of the system by comparing the distributions for the associated energies over the lattice with a boltzmann distribution @xmath94 here @xmath53 is the energy associated with the selected degrees of freedom ( e.g. the magnetic or electric energy of the gauge fields ) , and @xmath95 is an appropriate phase space factor , specified below . @xmath96 is a normalization constant . the distribution @xmath97 is obtained in the simulation by sampling , at a fixed time @xmath98 , the variable @xmath53 over all sites or plaquettes on the lattice ; it gives the probability , averaged over the whole lattice , to find the value @xmath53 at a given , though arbitrary , lattice point . if this distribution has the boltzmann shape ( [ distri ] ) the system has lost all initial order in the distribution of @xmath53 , and the entropy has reached a maximum under the constraint of fixed total energy associated with the selected degrees of freedom . we will present measurements for four different energy distributions , the distribution @xmath99 for the electric energy @xmath100 of the gauge fields , the distribution @xmath101 of the magnetic energy @xmath102 , as well as the distributions @xmath103 and @xmath104 of the kinetic and potential energies of the higgs field , @xmath105 and @xmath106 , respectively , as defined in eqs . ( [ epsilon ] ) . in the weak coupling limit the respective phase space factors @xmath95 are given by @xmath107 a detailed calculation of the phase space factors can be found in appendix [ sec : appendix ] . by plotting @xmath108 $ ] against @xmath53 one can read off the apparent temperature @xmath93 . it was discussed in @xcite that @xmath93 is not in all cases identical with the true temperature of the system . writing @xmath109 one finds @xmath110 @xmath111 is a number varying between @xmath112 and @xmath113 as @xmath0 rises from low to high values @xcite . in ref . @xcite it was argued that this can be understood in terms of the changing number of effective degrees of freedom which at low temperatures is only @xmath112 of that at high temperatures , due to the increasing importance of longitudinal plasma modes at high temperature . more details on the relation between the apparent temperature and the actual temperature can be found in the appendix [ sec : appendix ] . by numerically evolving the initial configurations forward in time , we can thus follow the system s thermalization . at each moment , we can measure its energy distributions and then read off its characteristic temperatures from the measured distributions , using ( [ tcorr ] ) for correct normalization . in fig . [ f1 ] several characteristic temperatures of the system are plotted as functions of time , for short times below @xmath114 ( in lattice units ) . to ensure weak coupling and closeness to the continuum limit we choose @xmath115 , @xmath116 , and @xmath117 . in this case it was found that @xmath118 . the curves in fig . [ f1 ] demonstrate a fast thermal equilibration within the gauge field sector between the electric ( solid line ) and magnetic ( dashed line ) energy distributions . after some transient oscillations they approach , on a typical time scale of 1 lattice unit , a common temperature @xmath119 . on roughly the same time scale the kinetic ( dotted line ) and quadratic potential ( dot - dashed line ) parts of the higgs field energies also equilibrate , but at a much lower temperature @xmath120 . this indicates that in the coupled su(2 ) higgs system the gauge and the higgs subsector each undergo a separate , fast equilibration , starting from a highly off - equilibrium initial state . the approximate equality of the two associated relaxation times , @xmath121 for the gauge sector and @xmath122 for the higgs sector , can be easily understood if one makes the following two hypotheses : 1 . thermalization in a chaotic system is driven by the most unstable modes and thus dominated by its largest lyapunov exponent . the gauge and the higgs sectors are associated with the same largest lyapunov exponent . while it is difficult to find a rigorous proof for the first hypothesis , such a behaviour is intuitively expected . the second hypothesis can be checked in our simulation and is found to be correct ( see sec . [ sec : lyap ] ) . at short times , each subsector settles down at a common temperature for all its degrees of freedom , but the temperatures of the two subsectors differ widely . this can be traced to the different initial energy content in the two subsectors . on average , the energy per degree of freedom in the gauge sector is initially 0.478 , while that in the higgs sector is only 0.060 . these specific values are , of course , a result of the parameters chosen in our initialization procedure . they would be different if other parameter values were selected . the larger total gauge energy per gauge degree of freedom compared to the total higgs energy per higgs degree of freedom results in the higher temperature of the gauge sector . equilibration between the two subsectors apparently takes much longer . since both subsectors are apparently equally `` chaotic '' , the relative stability of the two temperatures of the subsystems indicates that energy exchange _ between _ the gauge and higgs sectors is suppressed . this will be discussed quantitatively in the following subsection . as seen in fig . [ f1 ] , the coupled su(2 ) yang - mills - higgs system shows a very rapid thermalization within each subsector , with the relaxation times being of the order of @xmath113 lattice unit . however , it turns out that the mutual equilibration between the two sectors is rather slow and requires much longer time . to study the long time behavior of the coupled system , we follow the evolution of the energies . in fig . [ f2 ] , we plot as a function of time ( on a logarithmic scale ) the following energies : @xmath123 ( solid ) , @xmath124 ( dotted ) , @xmath125 ( short dash ) , @xmath126 ( long dash ) , and @xmath127 ( dot - dash ) . here @xmath123 is the electric energy per gauge degree of freedom : @xmath128 @xmath124 the magnetic energy per gauge degree of freedom : @xmath129 @xmath126 the total higgs energy per higgs degree of freedom : @xmath130 @xmath125 the total gauge energy per gauge degree of freedom : @xmath131 and @xmath127 the total energy per degree of freedom in the entire system : @xmath132 @xmath21 is the number of lattice sites . note that each lattice site is associated with @xmath133 degrees of freedom : @xmath134 gauge degrees of freedom and @xmath135 higgs degrees of freedom . the parameters used in the calculation are the same as in fig . [ f1 ] . in the weak coupling regime , @xmath126 is the temperature of the higgs sector as @xmath125 is that of the gauge sector , while @xmath127 will be the final temperature of the entire system after the coupled gauge and higgs sectors reach complete equilibrium . again , we notice that the energy transfer from the magnetic sector to the initially unpopulated electric sector occurs on a time scale of the order of unity , which agrees with fig . [ f1 ] . the same has been seen for the higgs kinetic energy and the higgs potential energy ( not shown here ) . however , it takes a very long time for the initially hotter gauge sector to transfer some of its energy to the cooler higgs sector . as shown in fig . [ f2 ] , @xmath125 ( short dash ) and @xmath126 ( long dash ) approach the flat line @xmath136 ( dot - dash ) only after times of many thousand lattice units . this means that the mutual equilibration process is associated with a very long relaxation time . assuming an exponential relaxation law for the gauge field energy @xmath137 , @xmath138 with an adjustable constant @xmath139 , we can determine the relaxation time @xmath140 for the mutual equilibration process from a linear fit to @xmath141 $ ] . in fig . [ f3 ] we plot @xmath142 $ ] as a function of time ; @xmath143 is taken as @xmath127 from eq . ( [ et ] ) . the solid line corresponds to the actual data while the dotted line represents the linear fit . from its slope we extract @xmath144 lattice units . as shown in ( [ rate ] ) , the rate of energy transfer from the gauge sector to the higgs sector is given by the sum @xmath145 . in a large system with no correlations between the signs of @xmath146 and @xmath147 , this sum would be a fluctuating quantity with a zero mean value . in our calculations , the sum turns out to be fluctuating around a small negative value , which means on average there is energy flowing from the gauge sector to the higgs sector although the flow is very slow . furthermore , from ( [ eudot ] ) and ( [ hj ] ) , we notice that @xmath145 is initially proportional to @xmath70 and to @xmath148 . both are restricted to small values if we want to work both in the weak coupling regime and close to the continuum limit . for larger couplings ( i.e. for increasing @xmath45 ) , keeping @xmath70 small to preserve the proximity to the continuum limit , one expects faster mutual equilibration . this is borne out numerically : increasing @xmath45 by a factor 10 at fixed @xmath70 , we found a much shorter relaxation time @xmath149 . this is , however , still much longer than the time scales for the thermalization of the gauge and the higgs fields individually , which did not change appreciably compared to the case with the smaller value for @xmath45 . our main goal in this section is to determine from the numerical integration of the equations of motion the lyapunov exponents associated with the gauge and the higgs fields , and to relate them to the damping rates obtained in perturbative calculations at finite temperature @xcite . let us begin with the discussion of two technical issues . first , when determining the lyapunov exponent from the exponential growth of the distance between two initially nearby configurations , one can in principle use different distance measures . however , for all possible choices the exponential growth in the distance will be dominated by the eigenmode associated with the largest lyapunov exponent , which is unique to a dynamical system . unless the chosen distance measure accidentally projects out this most unstable mode , different distance measures should thus yield the same lyapunov exponent . ( depending on the power with which the classical fields appear in the distance measure the actually measured growth rates can be different , but they are always related to the lyapunov exponent by a simple multiplicative constant . ) based on these remarks one expects the two distance measures given in ( [ gdist ] ) and ( [ hdist ] ) to yield the same lyapunov exponent . second , in order to relate results from numerical integration to those from perturbative calculations , we need to study the scaling behavior of the lyapunov exponents with the energy ( temperature ) of the system . for the pure su(2 ) gauge system the relation @xmath150 was established @xcite , with @xmath151 denoting the gauge energy per plaquette . in @xcite it was shown by a perturbative calculation with hard thermal loop resummation that , at leading order in the coupling constant , in the coupled su(2 ) yang - mills - higgs system the damping rate of a static gauge boson is the same as that in pure su(2 ) gauge theory : @xmath152 for the higgs damping rate the relation @xmath153 was found @xcite to hold for small values of the higgs self coupling @xmath84 . using the scaling law ( [ oneover6 ] ) together with the fact that in weak coupling the energy per plaquette @xmath154 can be linearly related to the temperature of the system , the authors of @xcite found that in the pure su(2 ) and su(3 ) gauge systems the scaled largest lyapunov exponent coincides with twice the gauge boson damping rate . what should one expect in our case ? in ref . @xcite it was speculated that here the gauge boson damping rate should be related with the lyapunov exponent extracted from the gauge distance measure ( [ gdist ] ) while the higgs boson damping rate could be connected with the lyapunov exponent extracted from the higgs distance measure ( [ hdist ] ) . since the numerical results did not support this expectation , the authors questioned the existence of a simple relationship between damping rates and lyapunov exponents in general , raising the possibility that the coincidence found in pure su(2 ) and su(3 ) gauge theory was a numerical accident . since , however , the largest lyapunov exponent is related to the most unstable mode in the entire coupled system and thus , as discussed above , is picked up by essentially every distance measure one can choose , we argue that one should rather expect both measures to yield the same lyapunov exponent . we will show promptly that the numerical computations confirm this expectation . if this is the case the lyapunov exponent can be related with at most one of the two damping rates . since the gluon damping rate ( [ scaling_1 ] ) is much larger than the higgs damping rate ( [ higgsdamp ] ) it is clear that if there is such a relation it should be between the _ gluon _ damping rate and the maximal lyapunov exponent . to which classically computable observable the smaller higgs damping rate can be related remains at this moment unclear . of course , it could also be that the lyapunov exponent is related to neither of the damping rates . to clarify this issue a careful study of the energy scaling of the lyapunov exponent is indispensible . in the upper part of fig . [ f4 ] we show that , within the numerical accuracy , the two distance measures ( [ gdist ] ) and ( [ hdist ] ) indeed yield the same lyapunov exponent . ( please note the suppressed zero on the vertical axis . ) thus they both measure the largest lyapunov exponent of the entire coupled system . we also see that the lyapunov exponent shows a slight , but characteristic long - time variation which is the same for both distance measures . this behaviour will be further discussed below . to understand the magnitude and physical meaning of the extracted lyapunov exponent , we studied its scaling behaviour with the energy contained in the system . to this end we varied the relative energy contents in the gauge and higgs subsectors as well as the total energy of the system , by changing @xmath70 and @xmath45 within the permitted ranges . we found that the lyapunov exponent ( which reflects the most unstable mode in the _ entire _ coupled system ) does not scale with the total energy of the entire system , but rather with the energy stored in the gauge subsector . these scaling tests were performed on the basis of numerical runs over a few hundred to a few thousand lattice time units . on these time scales the scaling of the largest lyapunov with the gauge field energy per plaquette was found to be accurate on the level of 5 10% . ( note that in fig . [ f4 ] the lyapunov exponent shows long - time variations over a similar range of about 10% . ) these findings indicate that the largest lyapunov exponent is associated which the gauge field dynamics , and that the chaoticity in the gauge sector dominates the dynamics of the entire system , including the equipartition between potential and kinetic energy contributions in the higgs subsystem . returning to fig . [ f4 ] we can now discuss the magnitude of the lyapunov exponent scaled by the energy per plaquette @xmath154 in the gauge field sector . the lower part in fig . [ f4 ] shows that for very large times of order ten thousand lattice units the lyapunov exponent seems to converge to a constant value slightly above 0.16 . this is close to the value of about 1/6 which , according to the findings of ref . @xcite , is expected ( see eq . ( [ oneover6 ] ) ) if the lyapunov exponent is related to the gluon damping rate . the observed long - time variation of the lyapunov exponent with final convergence only after many thousand time steps is apparently connected with the already observed long - time variation of the relative energy content of the higgs and gauge field subsectors seen in fig . [ f2 ] . as the gauge and higgs field sectors finally reach mutual equilibrium after several ten thousand lattice time units , the scaled lyapunov exponent @xmath155 settles down to a constant asymptotic value of approximately 1/6 . noting that @xmath156 after full equilibration and inserting @xmath76 , @xmath77 , we obtain @xmath157 which together with ( [ scaling_1 ] ) verifies ( [ scaling ] ) at the level of 10% . a more accurate verification is prohibited by the extremely slow equilibration between the gauge and higgs subsectors in our coupled system . the numerically determined lyapunov exponent is somewhat on the low side ; according to figs . [ f4 ] and [ f2 ] , the magnitude of this discrepancy is directly related to the deviation of the gauge and higgs field energies from equipartition . let us summarize the findings from this section : we determined the gauge field lyapunov exponent and verified that for the yang - mills - higgs system it is related to the static gauge boson damping rate in the same way as for the pure gauge theory . the quantity @xmath86 , as defined in ( [ hdist ] ) and ( [ lyap_def ] ) , was shown to give the same lyapunov exponent and therefore to be governed by the gauge degrees of freedom , too . therefore @xmath86 can not be related to any specific dynamics of the higgs field , and it should not be called `` higgs lyapunov exponent '' as suggested in @xcite . in particular , it is not possible to establish a relation between @xmath86 and the higgs damping rate ( [ higgsdamp ] ) . unfortunately , we have not been able to come up with an alternative observable in the dynamics of classical fields which could be related to @xmath158 . we close this paper with a discussion of our results in comparison with the recent work presented in @xcite . for small values of the higgs self coupling @xmath84 , bir and thoma @xcite found that the lyapunov exponents @xmath86 and @xmath85 deduced from the higgs and gauge fields , respectively , agree within the numerical accuracy . for very large values of the higgs self coupling , on the other hand , @xmath86 was found to decrease , thereby breaking the coincidence between @xmath86 and @xmath85 . for the case of small @xmath58 we have verified the coincidence of @xmath86 and @xmath85 . it has a natural explanation since , as we already discussed , in principle any distance measure between two neighboring trajectories in phase space simply yields the maximal lyapunov exponent of the entire system . in view of this argument , the disagreement between @xmath86 and @xmath85 at large @xmath58 at first seems surprising . if , however , the distance measure projects out the eigenmode with the largest lyapunov exponent , it can yield a different value for the growth rate . this is apparently what happens at very large values of @xmath58 for the higgs distance ( [ gidist ] ) used in @xcite : as already pointed out in @xcite , for large @xmath58 the dynamics of the higgs amplitude ( [ defhiggsamp ] ) becomes completely determined by the quartic self - interaction term ( [ elambda ] ) , thereby effectively freezing the higgs field amplitude @xmath45 . thus this degree of freedom starts to decouple from the rest of the system , and it may happen that the growth of perturbations in the higgs amplitude is no longer influenced by the maximal lyapunov exponent associated with the gauge field , at least , over the time scale used for the determination of the lyapunov exponent . we have not studied here the case of large @xmath58 , for the following reasons : first , if one increases @xmath58 without decreasing the higgs amplitude one leaves the perturbative regime ( [ weakcouph ] ) . then the computed lyapunov exponents can no longer be compared with the damping rates from hard thermal loop calculations . second , for very large @xmath58 the energy density of the quartic interaction term ( [ elambda ] ) and thus also the total energy density become very large . as discussed in sec . [ subsec : iwc ] , in this case one runs into lattice artifacts due to the compactness of the magnetic energy on the lattice . while we agree with bir and thoma @xcite about the coincidence of @xmath85 and @xmath86 , we disagree with their determination of the temperature . consequently we also differ in the interpretation of their results with respect to their connection with perturbative damping rates , believing that their rather negative conclusions are not justified . bir and thoma @xcite use the stefan - boltzmann law for 10 degrees of freedom @xmath159 to extract the temperature @xmath0 from the total energy density @xmath160 of the system . here we put in @xmath161 , @xmath162 , and @xmath163 explicitly ( elsewhere we take @xmath164 ) . as can be seen from the occurrence of planck s constant in ( [ stefbol ] ) , this law makes explicit use of the quantized nature of excitation energies in a gas of ( massless ) particles . as such it can not be applied to a classical field theory . the classical yang - mills - higgs fields on the lattice correspond to a dynamical system of coupled anharmonic oscillators which in thermal equilibrium and for weak coupling obeys the relation @xmath165 thus the energy per degree of freedom is directly proportional to the temperature , and not to its fourth power . even more crucial is our finding that the whole system equilibrates only after a very long time . since the authors of @xcite did not use the rescaling method for extracting the lyapunov exponents , they ran very soon into the problem of distance saturation due to the compactness of the gauge group @xcite , and thus their analysis was necessarily restricted to times which were short compared to the mutual equilibration time between the gauge and higgs sectors . if the lyapunov exponent is determined on such short time scales , the total energy of the system does not tell anything about the temperature of the relevant gauge subsector no matter whether it is determined from ( [ stefbol ] ) or ( [ harmos ] ) . for this reason bir and thoma missed the specific connection of the lyapunov exponent with the gauge subsystem rather than with the entire coupled yang - mills - higgs system . only an analysis like the one presented here can uncover the fact that after very short thermalization times one can cleanly define separate temperatures for the gauge and the higgs subsystems , and that the maximal lyapunov exponent actually scales with the temperature of the gauge sector already long before global equilibration between the gauge and higgs sectors sets in . in conclusion , we have studied the high temperature behavior of the coupled yang - mills - higgs system in the weak coupling regime close to the continuum limit . we found that the coupled system undergoes thermalization in stages : while the gauge and higgs subsectors self - thermalize very rapidly , equipartition of the energy between these two sectors takes several orders of magnitude longer . we expect and have qualitative evidence for a connection between the rapid thermalization time scale within each of the two subsectors and the magnitude of the largest lyapunov exponent ; to establish this relation quantitatively requires , however , a much more detailed and computer - intensive investigation . since we found that the largest lyapunov exponent is associated with the gauge field and scales with the gauge field energy , we conclude that thermalization of both the gauge and higgs subsectors is driven by the chaotic dynamics of classical non - abelian gauge fields . the higgs subsector thus thermalizes via its coupling to the chaotic gauge fields ( gluon heat bath ) and not due its own intrinsic non - linearity via the selfcoupling @xmath58 . the scaling of the lyapunov exponent with the gauge field energy was used to verify that , even in the presence of a coupling to matter fields , the largest lyapunov exponent still agrees with twice the damping rate of a static gauge boson . the analytical results of thoma and bir @xcite from thermal perturbation theory show that in the coupled yang - mills - higgs system also the higgs damping rate is proportional to @xmath2 and thus , from the general arguments given e.g. in @xcite , is expected to survive in the classical limit . at the moment it remains an open question whether it is possible to identify an observable in the dynamical evolution of classical yang - mills - higgs fields which can be associated with the higgs boson damping rate in a similar way as the largest lyapunov exponent is related to the gluon damping rate . the authors thank m. thoma and t. bir for fruitful discussions . this work was supported in part by the u.s . department of energy ( grant no . de - fg02 - 96er40945 ) and by the north carolina supercomputing center . u.h . and s.l . acknowledge support by the deutsche forschungsgemeinschaft ( dfg ) and the bundesministerium fr bildung , wissenschaft , forschung und technologie ( bmbf ) . in this appendix we derive the phase space factors presented in sec . [ subsec : short ] . in general it is difficult to calculate the energy distribution over the microscopic degrees of freedom of an arbitrary system evolving in time . it is equivalent to solving the equations of motion . however , as shown in the main part of the paper , the gauge and higgs sectors of the system equilibrate separately very rapidly ( albeit reaching different temperatures ) . in addition we are working in a regime where weak coupling is assumed between the gauge and higgs sectors as well as within each sector . this significantly simplifies the task of calculating the energy distributions of different degrees of freedom . to determine how the electric energy is distributed over the lattice sites for the thermalized gauge sector at temperature @xmath166 , we neglect its coupling to the higgs fields and write the partition function for the gauge sector as @xmath167 \ , \delta(d_{x , i}^{ab } \bar e_{x , i}^b ) \ , \delta(f(u_{x , i } ) ) \ , , \label{partition}\ ] ] where the first @xmath70-function takes into account the constraints imposed by gauss law , and the second one represents gauge fixing of the form @xmath168 , which involves link variables only . @xmath169 is the haar measure defined on the su(@xmath170 ) group mainfold . for the electric energy on a site , we define its distribution function @xmath99 as @xmath171 \ , \nonumber \\ & & \times\,\delta(d_{x , i}^{ab } \bar e_{x , i}^b ) \ , \delta(f(u_{x , i } ) ) \ , \delta(\epsilon - { \textstyle { 1\over 2 } } \sum\limits_{i , a } \bar e_{x_0,i}^a \bar e_{x_0,i}^a ) \nonumber \\ & = & z^{-1}\ , \int \prod\limits_{x , i , a } d\bar e_{x , i}^a \ , d\mu(u_{x , i } ) \ , d\lambda_x^a \ , d\kappa \ , \delta(f(u_{x , i } ) ) \nonumber \\ & & \times \exp\left[- { \bar h_{\rm el}+\bar h_{\rm mag } \over t_{\rm g } } + i \sum\limits_{x , i , a , b } \lambda_x^a d_{x , i}^{ab } \bar e_{x , i}^b + i \kappa \ , ( \epsilon - { \textstyle { 1\over 2 } } \sum\limits_{i , a } \bar e_{x_0,i}^a \bar e_{x_0,i}^a ) \right ] \ , , \label{pel } \end{aligned}\ ] ] where we have introduced lagrange multipliers @xmath172 and @xmath173 to account for the @xmath70-functions . the electric and magnetic energies are defined in ( [ hel ] ) and ( [ hmag ] ) , respectively . since the electric fields appear only up to quadratic order in ( [ pel ] ) , the integrations over @xmath174 can be performed with the result @xmath175 \nonumber \\ & & \times \exp\left[- { \bar h_{\rm mag } \over t_{\rm g } } - { t_{\rm g } \over 2 } \sum\limits_{x , i , a , b}^ { x \neq x_0 } ( d_{x , i}^{ab } \lambda_x^a)^2 - { t_{\rm g}\over { 2(1+i\kappa t_{\rm g } ) } } \sum\limits_{i , a , b } ( d_{x_0,i}^{ab } \lambda_{x_0}^a)^2 + i \kappa \epsilon \right ] \nonumber \\ & \sim & \int \prod\limits_{x , i , a } d\mu(u_{x , i } ) \ , d\lambda_x^a \ , d\kappa \ , \delta(f(u_{x , i } ) ) \ , \left ( 1 + i \kappa t_{\rm g}\right)^{-9/2 } \nonumber \\ & & \times \exp\left[- { \bar h_{\rm mag } \over t_{\rm g } } - { t_{\rm g } \over 2 } \sum\limits_{x , i , a , b } { { ( d_{x , i}^{ab } \lambda_x^a)^2 } \over{1+i\kappa t_{\rm g}\delta_{x , x_0 } } } + i \kappa \epsilon \right ] \ , . \end{aligned}\ ] ] the integration over the lagrange multipliers @xmath172 can be evaluated similarly . after that , the whole integral factorizes into integrations over the link variables and the lagrange multiplier @xmath173 . the integration over the link variables is @xmath53-independent and can be factored out as a constant . we thus obtain @xmath176 which yields the electric phase space factor @xmath177 . three points should be stressed here : first , the above result is exact for a pure su(2 ) gauge system and approximately valid for the weakly coupled su(2 ) higgs system . second , the apparent temperature agrees with the true temperature , i.e. @xmath178 . finally we mention that @xmath179 is replaced by @xmath180 for the general case of su(@xmath3 ) . in deriving the distribution for the magnetic energy on a plaquette , we again neglect the coupling to the higgs fields and write @xmath181 \ , \nonumber \\ & & \times\,\delta(d_{x , i}^{ab } \bar e_{x , i}^b ) \ , \delta(f(u_{x , i } ) ) \ , \delta\left ( \epsilon - 4 ( 1 - { \textstyle { 1\over 2}}{\rm tr } \ , u_{x_0,i_0\,j_0 } ) \right ) \ , . \end{aligned}\ ] ] assuming weak coupling within the gauge sector itself , we neglect the nonlinear term in the gauss law , which reduces @xmath182 to a linear form : @xmath183 . then the integration over the electric fields decouples from that over the link variables and can be factored out as a constant independent of @xmath53 . hence we find @xmath184 now the integration is over the link variables @xmath185 , while the energy is expressed in terms of the plaquettes @xmath31 . in order to carry out the integration , we use a technique described in @xcite to transform the integral over link variables into an integral over plaquettes @xcite : @xmath186 where the @xmath70-function @xmath187 imposes a constraint on each elementary cube @xmath163 . these constraints are the lattice version of the bianchi identities , which correspond to @xmath188 in the continuum limit . @xmath189 is an su(@xmath170 ) matrix defined in terms of the link variables associated with the cube @xmath163 . the transformation from links to plaquettes corresponds to the transformation from vector potentials to magnetic field strengths in the continuum limit . for details on the transformation and the exact form of @xmath189 , we refer the reader to @xcite and @xcite . now we consider ( [ pmag0 ] ) in two limiting cases : @xmath190 and @xmath191 . in the low temperature limit , i.e. when the field amplitudes are small , the non - abelian part of the constraints becomes negligible and hence the non - abelian bianchi identities @xmath188 reduce to their abelian equivalents @xmath192 . on the lattice , they take the form of @xmath193 , where the sum runs through the six plaquettes forming the surface of a cube . these are strong constraints which completely eliminate the longitudinal components of the @xmath194 fields and reduce the effective number of degrees of freedom by @xmath195 . so in this limit , the apparent temperature @xmath196 . in the high temperature limit , we expand the constraints in terms of the irreducible representations of the group su(@xmath170 ) @xcite : @xmath197 where @xmath198 is the trace of @xmath199 in the @xmath200-dimensional representation . for @xmath201 , the dominant contribution to ( [ pmag0 ] ) comes from @xmath202 . taking @xmath202 and neglecting those with @xmath203 is equivalent to ignoring the constraints imposed by the bianchi identities and treating the plaquettes as independent variables . then it is not surprising to find that the apparent temperature @xmath93 obtained in measuring @xmath101 coincides with the actual temperature @xmath166 . while it is easy to do the integrations in the limiting cases , it is difficult to calculate ( [ pmag0 ] ) for intermediate temperatures . noting that the major effect of the constraints @xmath187 on the distribution function is to change its logarithmic slope from the actual temperature @xmath166 to the apparent temperature @xmath93 , we assume that these constraints can be effectively taken into account by replacing @xmath166 by @xmath93 in ( [ pmag0 ] ) and write @xmath204 of course , the validity of this assumption is to be justified and the relation between @xmath93 and @xmath166 to be established by numerical calculations . the evaluation of ( [ pmag1 ] ) is straightforward . making use of the quaternion representation @xmath205 for a plaquette @xmath37 and the explicit form of the haar measure on the su(@xmath170 ) group manifold @xmath206 one can easily carry out the integration in ( [ pmag1 ] ) and obtain @xmath207 where @xmath208 . the relation @xmath209 can be determined numerically and @xmath210 varies from @xmath112 to @xmath113 as @xmath166 increases from zero to infinity @xcite . note that , for the sake of convenience , we have calculated the distribution of the electric energy on a site and that of the magnetic energy on a plaquette . hence the apparent electric and magnetic distributions do not coincide even at low temperature where , in principle , one can treat the electric and magnetic energy in exactly the same way and arrive at the same distributions . however , as seen in fig . [ f1 ] , the different results for @xmath93 relate to the same value of @xmath166 in the electric and magnetic sector . to derive how the higgs kinetic and potential energies are distributed over the lattice sites for the thermalized higgs sector at temperature @xmath211 , we neglect both the gauge - higgs coupling and the higgs self - coupling . then it is straightforward to derive @xmath212 from which we obtain @xmath213 . in principle , perturbative corrections to the results obtained in ( [ pelres ] ) , ( [ pmag2 ] ) , and ( [ phiggs ] ) can be calculated . for our case at hand , however , it turns out that the formulae derived above are in good agreement with the numerical results . this is not surprising because our numerical calculations are performed for weak coupling . permanent address : institut fr theoretische physik , universitt regensburg , d-93040 regensburg , germany also at yerevan physics institute , yerevan , armenia c. detar , in _ quark - gluon plasma 2 _ , edited by r.c . hwa , ( world scientific , singapore , 1995 ) , p. 1 . e. braaten and r. d. pisarski , phys . rev . d*42 * , 2156 ( 1990 ) . thoma , in _ quark - gluon plasma 2 _ , edited by r.c . hwa , ( world scientific , singapore , 1995 ) , p. 51 . e. braaten and a. nieto , phys . rev . d*53 * , 3421 ( 1996 ) . bir , s. matinyan , and b. mller , _ chaos and gauge field theory _ ( world scientific , singapore , 1994 ) . bir , c. gong , b. mller , and a. trayanov , int . j. mod . c*5 * , 113 ( 1994 ) . silin , zh . . fiz . * 38 * , 1577 ( 1960 ) [ sov . jetp * 11 * , 1136 ( 1960 ) ] . u. heinz , ann . ( n.y . ) * 168 * , 148 ( 1986 ) . kelly , q. liu , c. lucchesi , c. manuel , phys . d*50 * , 4209 ( 1994 ) . bir , c. gong , and b. mller , phys . rev . d*52 * , 1260 ( 1995 ) . bir and m.h . thoma , preprint hep - ph/9603339 , to appear in phys . d. k. wilson , phys . rev . d*10 * , 2445 ( 1974 ) . m. creutz , _ quarks , gluons , and lattices _ ( cambridge university press , cambridge , 1983 ) . b. mller and a. trayanov , phys . lett . * 68 * , 3387 ( 1992 ) . b. mller , duke university preprint duke - th-96 - 118 ( chao - dyn/9607001 ) . c. gong , phys.rev . d*49 * , 2642 ( 1994 ) . c. gong , ph.d . thesis , duke university , 1994 ( unpublished ) . batrouni , nucl . b*208 * , 467 ( 1982 ) .

we first remind the reader of a simple geometric description of the kirby - siebenmann invariant of a 4manifold in terms of a quadratic refinement of its intersection form . this is the first in a sequence of higher - order intersection invariants of whitney towers , studied by the authors , particularly for the 4ball . in the second part of this paper , a general theory of quadratic forms is developed and then specialized from the non - commutative to the commutative to finally , the symmetric settings . the intersection invariant for twisted whitney towers is shown to be the universal symmetric refinement of the framed intersection invariant . as a corollary we obtain a short exact sequence that has been essential in the understanding of whitney towers in the 4ball .

1207.0109

a beautiful consequence of mike freedman s disk embedding theorem is the existence of non - smoothable @xmath0-manifolds . in the easiest setting , his result can be stated as follows . [ thm : classification ] any odd unimodular symmetric form @xmath1 is realized as the intersection form of exactly _ two _ closed simply - connected oriented 4manifolds ( up to homeomorphism ) . these 4manifolds are homotopy equivalent and are distinguished by the following ( equivalent ) criteria : exactly one of the manifolds 1 . is smoothable after crossing with @xmath2 . is smoothable after connected sum with finitely many copies of @xmath3 . 3 . has a linear reduction of its micro normal bundle . has vanishing kirby - siebenmann invariant in @xmath4 . exhibits the following formula for a quadratic refinement @xmath5 of @xmath6 : @xmath7 by donaldson s theorem a @xcite , _ exactly _ the diagonalizable odd forms @xmath6 are realized by closed _ smooth _ diagonal forms are realized by connected sums of complex projective planes ( with varying orientations ) ; in fact , most such forms are now known to admit infinitely many smooth representatives ( all being homeomorphic by the above theorem ) , see e.g. @xcite . criterion ( v ) is the most elementary and will be explained in detail in section [ sec : ks ] . the key is the following geometric interpretation for the quadratic refinement @xmath8 . in a simply - connected closed 4manifold @xmath9 , any class in @xmath10 can be represented by a ( topologically generic ) immersed sphere @xmath11 . this means that @xmath12 looks locally like @xmath13 , except for finitely many double points around which @xmath12 looks like @xmath14 . one can add more local self - intersection points to @xmath12 until their algebraic sum is zero . this implies that one can choose _ whitney disks _ @xmath15 , pairing all these self - intersection points . these are ( topologically generic ) immersed disks @xmath16 whose boundary consists of two arcs , each going between the two intersection points but on different sheets , see figure [ fig : whitney - move ] . we will explain in lemma [ lem : c ] why @xmath8 equals an intersection invariant @xmath17 , computed by summing the ( topologically generic ) intersections between an immersed sphere @xmath12 , representing the characteristic @xmath18 , and ( the interiors of ) framed whitney disks @xmath15 for @xmath12 : @xmath19 in @xcite , this invariant was called the _ kervaire - milnor invariant _ because these authors first proved rohlin s formula below @xcite for the case where @xmath9 is smooth and @xmath20 is represented by an embedded sphere , implying the properties @xmath21 . [ rem : framing ] the figure above shows a _ framed _ whitney disk @xmath15 in the sense that there are two _ disjoint _ parallel copies of @xmath15 , as needed for the whitney move on the right hand side . in general , a whitney disk comes with a framing of its boundary and hence admits a well defined euler number in @xmath22 , its _ twist_. the operation of _ boundary twisting _ @xcite allows to assume that all whitney disks are framed , i.e. have twist zero . moreover , one can also assume that the @xmath15 are ( disjointly ) embedded disks , by pushing all ( self)-intersections off the boundary . a generalization of rohlin s theorem @xcite says that this geometric invariant determines the kirby - siebenmann invariant of a closed oriented 4manifold @xmath9 by the formula @xmath23 explaining the equivalence of criteria ( iv ) and ( v ) above . in section [ sec : ks ] we ll recall a definition of @xmath24 which makes the above formula hold for all closed oriented 4manifolds @xmath9 ( without assuming that @xmath20 is spherical ) . the 2-complex @xmath25 in @xmath9 is referred to as a _ whitney tower _ of order 1 , built on @xmath12 , with order 1 whitney disks @xmath15 . the invariant @xmath26 used above is the first _ intersection invariant _ of such whitney towers . it has again _ order 1 _ , the order zero intersection invariants being given by the intersection form @xmath27 . in a sequence of papers , the authors generalized this invariant to higher orders , see for example our survey @xcite . the idea is that if @xmath28 vanishes then all intersections between @xmath12 and @xmath15 can be paired by _ order 2 _ whitney disks @xmath29 and there should be a second - order intersection invariant @xmath30 measuring the obstruction for finding order 3 whitney disks , and so on . in @xcite we worked out this higher - order intersection theory in detail for whitney towers built on immersed disks in the @xmath0ball bounded by framed links in the @xmath31sphere . in this simply connected setting the invariant @xmath32 of an order @xmath33 ( framed ) whitney tower @xmath34 takes values in an abelian group @xmath35 ( where @xmath36 is number of link components ) , and the vanishing of @xmath32 implies that the link bounds and order @xmath37 whitney tower . for links bounding _ twisted _ whitney towers there is an analogous obstruction theory and intersection invariant @xmath38{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ k({\mathcal{w}})\in{\mathcal{t}}^{{\mathchoice { \begin{minipage}{.15in}\includegraphics[width=.15in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ k(m)$ ] , and in the second part of this paper we develop an algebraic theory of quadratic forms , leading to a beautiful relation between these framed and twisted obstruction groups , spelled out in theorem [ thm : exact ] . this result is used in the computation of the whitney tower filtration on classical links described in @xcite . * acknowledgments : * the main part of this paper was written while the first two authors were visiting the third author at the max - planck - institut fr mathematik in bonn . they all thank mpim for its stimulating research environment and generous support . the third author was also supported by nsf grants dms-0806052 and dms-0757312 . in the notation of the introduction , recall that a class @xmath39 is called _ characteristic _ if @xmath40 the set @xmath41 of characteristic elements is a @xmath42-torsor via the action @xmath43 . a closed oriented 4manifold @xmath9 with @xmath44 defines a _ quadratic refinement _ @xmath45 of @xmath6 in the sense that @xmath46 this means that @xmath24 is completely determined by _ one _ of its values ( and that value is determined by the kirby - siebenmann invariant of @xmath9 via rohlin s theorem above ) . if @xmath6 is even then @xmath47 is characteristic and a simple geometric argument shows that @xmath48 , see below . that s why @xmath24 is only interesting for odd intersection forms . freedman and kirby @xcite defined @xmath49 by representing @xmath20 by an embedded surface @xmath50 such that @xmath51 has a spin structure that does not extend across @xmath52 . let @xmath53 be the projection map of the boundary of a normal disk bundle for @xmath52 . then there is a quadratic refinement @xmath54 of the intersection form on @xmath55 defined by taking the inverse image torus @xmath56 for a circle @xmath57 in @xmath52 and observing that it comes equipped with a canonical spin structure . then @xmath58 is the arf invariant of this spin torus , aka its spin bordism class , and finally @xmath49 is the arf invariant of @xmath54 . this definition can be drastically simplified if @xmath20 is represented by an immersed sphere @xmath12 whose self - intersection points are paired by framed whitney disks @xmath59 . ( this is always the case if @xmath9 is simply - connected as in the introduction . ) [ lem : c ] define @xmath60 to be the sum of all intersections between the immersed sphere @xmath12 and the interiors of the whitney disks @xmath15 . then @xmath61 . in @xcite the following simplification of @xmath49 is already explained , in fact , this was the original definition ( and only later it was realized that one needs a more general approach because caps do nt always exist ) : assume that @xmath20 is represented by a surface @xmath52 with ( immersed , framed ) _ caps_. these are ( immersed , framed ) disks @xmath62 in @xmath9 bounding a hyperbolic basis @xmath63 of curves in @xmath55 . freedman - kirby show that the quadratic form @xmath64 is determined by @xmath65 number of intersections between the interior of the cap @xmath66 and @xmath52 , similarly for @xmath67 . by definition of the arf invariant , one gets that @xmath68 assume now that @xmath20 is represented by an immersed sphere @xmath12 whose self - intersection points are paired by ( immersed , framed ) whitney disks @xmath69 . we can get into the capped surface situation as follows : for each pair of self - intersection points of @xmath12 , add a tube @xmath70 on one sheet going from one self - intersection to the other . that turns @xmath12 into an embedded surface @xmath52 with half of the caps @xmath66 given by small normal disks to @xmath52 that bound the generating circles on @xmath70 . moreover , the whitney disks @xmath15 can serve as the dual caps @xmath71 , preserving the framing , as illustrated in figure [ fig : framings - preserved ] . by construction , @xmath72 since each normal disk @xmath66 intersects @xmath52 in a single point . therefore , the required formula follows : @xmath73 [ rem : tau well - defined ] in the simply - connected case , it is not hard to see that @xmath74 is well - defined exactly on characteristic elements . one thing to check is that it does not depend on the choices of the whitney disks @xmath15 . once we fix the boundary , any two such choices differ by a connected sum into a sphere @xmath75 . if we require the whitney disks to be ( stably ) framed then @xmath75 needs to be ( stably ) framed and hence it intersects a characteristic sphere in an even number of points , leaving our count @xmath76 unchanged modulo two . all these considerations can be found in chapter 10 of the book @xcite by freedman and quinn . unfortunately , the results do nt hold as stated for 4manifolds with fundamental groups that contain 2-torsion elements . the problem arises from different choices of pairings of intersection points , as observed by richard stong in @xcite . in @xcite , the last two authors gave a complete discussion of the invariant @xmath76 in the presence of fundamental groups . all trees considered in this paper are _ unitrivalent , oriented and labelled_. this means that they are equipped with _ vertex orientations _ , i.e. cyclic orderings of the edges incident to each trivalent vertex . moreover , the univalent vertices of a tree are labeled by elements of the index set @xmath77 , except for the unlabeled _ root vertex _ if a given tree happens to be rooted . ( there is at most one root , even though any other index can appear several times ) . all trees are considered up to label - preserving isomorphism . the _ order _ of such a tree is the number of trivalent vertices . given rooted trees @xmath78 and @xmath79 , the _ rooted product _ @xmath80 is the rooted tree gotten by identifying the two roots to a vertex and adjoining a rooted edge to this new vertex , with the orientation of the new trivalent vertex given by the ordering of @xmath78 and @xmath79 in @xmath80 . the _ inner product _ @xmath81 of two rooted trees @xmath78 and @xmath79 is defined to be the unrooted tree gotten by identifying the two rooted edges to a single edge . we observe that the two products interact well in the sense of figure [ fig : rotate ] . let @xmath82 be the free abelian group generated by ( isomorphism classes of ) rooted trees as above . it is graded by order and the rooted product can be extended linearly to a pairing @xmath83 this is grading preserving on @xmath84 $ ] , i.e. it preserves the grading when shifted up by one ( so order is replaced by the number of univalent non - root vertices ) . on the other hand , the inner product @xmath85 is grading preserving via order . here @xmath86 is the free abelian group generated by unrooted trees as above . note that rotating the relevant planar trees by 180 respectively 120 degrees shows that the inner product is both _ symmetric _ and _ invariant _ : @xmath87 and @xmath88 , see figure [ fig : rotate ] for the proof of invariance . [ def : tree - groups ] the graded abelian groups @xmath89 respectively @xmath90 are defined as quotients of @xmath82 respectively @xmath86 by the as and ihx relations as in figure [ fig : relations ] . and @xmath90 . all trivalent orientations come from an orientation of the plane , and univalent vertices extend to subtrees which are fixed in each equation . ] it is well known that @xmath89 is the free ( quasi ) lie algebra over @xmath22 on @xmath36 generators with lie bracket induced by the rooted product . here the word _ _ refers to the fact that we only require the antisymmetry relations @xmath91= - [ y , x]$ ] . so @xmath92 $ ] is not necessarily zero in these lie algebras . in our previous papers , we needed to consider both versions of lie algebras and used the notation @xmath93 for @xmath94 ( recall that one gets a _ graded _ lie algebra only when shifting the order by one ) . in this paper we will only study one type of lie algebras and usually omit the adjective ` quasi ' . the inner product extends uniquely to a bilinear , symmetric , invariant pairing @xmath95 this follows simply from observing that the as and ihx relations hold on both sides and are preserved by the inner product . we will show in lemma [ lem : inner product ] that this inner product is in fact _ universal_. [ def : twisted - tree - groups ] the group @xmath96{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } } _ { 2n}(m)$ ] is gotten from @xmath97 by adding order @xmath33 @xmath98{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } $ ] -trees as new generators . these are rooted trees of order @xmath33 as above , except that the root carries the label @xmath98{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } $ ] . in addition to the ihx- and as - relations on unrooted trees in @xmath97 , we introduce the following new _ symmetry _ , _ interior twist _ and _ twisted ihx _ relations : @xmath99{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } = ( -j)^{{\mathchoice { \begin{minipage}{.15in}\includegraphics[width=.15in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } \quad\quad 2\cdot j^{{\mathchoice { \begin{minipage}{.15in}\includegraphics[width=.15in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } = \langle j , j\rangle \quad\quad i^{{\mathchoice { \begin{minipage}{.15in}\includegraphics[width=.15in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } = h^{{\mathchoice { \begin{minipage}{.15in}\includegraphics[width=.15in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } + x^{{\mathchoice { \begin{minipage}{.15in}\includegraphics[width=.15in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } - \langle h , x\rangle\ ] ] as their names suggest , these new relations arose from geometric considerations for twisted whitney towers in @xcite . they will be explained algebraically in our last section via the theory of universal quadratic refinements . roughly speaking , the universal symmetric pairing @xmath100 will be shown to admit a universal quadratic refinement @xmath101{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } ( m)$ ] defined by @xmath102{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } $ ] . in particular , with the right algebraic notion of ` quadratic refinement ' , the group @xmath103{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } ( m)$ ] is completely determined by the pairing @xmath100 . the rest of this paper is devoted to finding this notion . as a consequence , we will prove the following exact sequence at the very end of this paper . it was used substantially in @xcite for the classification of whitney towers in the 4-ball . [ thm : exact ] for all @xmath104 , the maps @xmath105 respectively @xmath106{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } \mapsto 1\otimes j$ ] give an exact sequence : @xmath107{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ { 2n}(m ) { \longrightarrow}{\mathbb{z}}_2 \otimes { \mathcal{l}}_{n}(m ) { \longrightarrow}0\ ] ] in this section we explain an algebraic framework into which our groups @xmath90 and @xmath108{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } ( m)$ ] fit naturally . in lemma [ lem : inner product ] we show that the @xmath90-valued inner product @xmath109 on the free lie algebra is universal . then a general theory of quadratic refinements is developed and specialized from the non - commutative to the commutative to finally , symmetric settings . in corollary [ cor : quadratic ] we show that @xmath103{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } ( m)$ ] is the home for the universal quadratic refinement of the @xmath97-valued inner product @xmath110 . we work over the ground ring of integers but all our arguments go through for any commutative ring . we also only discuss the case of finite generating sets @xmath111 , even though everything holds in the infinite case . [ lem : inner product ] let @xmath112 be a lie algebra together with a bilinear , symmetric , invariant pairing @xmath113 into some abelian group @xmath9 . if @xmath114 is a lie homomorphism ( given by @xmath36 arbitrary elements in @xmath112 ) there exists a _ unique _ linear map @xmath115 such that for all @xmath116 @xmath117 the uniqueness of @xmath118 is clear since the inner product map is onto . for existence , we first construct a map @xmath119 as follows . given a tree @xmath120 pick an edge in @xmath121 to split @xmath122 for rooted trees @xmath123 . then set @xmath124 if we split @xmath121 at an adjacent edge , this expression stays unchanged because of the symmetry and invariance of @xmath6 . however , one can go from any given edge to any other by a sequence of adjacent edges , showing that @xmath125 does not depend on the choice of splitting . it is clear that @xmath126 can be extended linearly to the free abelian group on @xmath86 and since @xmath127 preserves as and ihx relations by assumption , this extension factors through a map @xmath118 as required . [ rem : grading ] recall that @xmath128 $ ] is actually a graded lie algebra , i.e. the lie bracket preserves the grading when shifted up by one ( so order is replaced by the number of univalent non - root vertices ) . let s assume in the above lemma that the groups @xmath129 are @xmath22-graded , @xmath130 $ ] is a graded lie algebra and that @xmath131 preserve those gradings . then the proof shows that the resulting linear map @xmath118 also preserves the grading . the rest of this section describes a general setting for relating our groups @xmath108{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ { 2n}(m)$ ] that measure the intersection invariant of twisted whitney towers to a universal ( symmetric ) quadratic refinement of the @xmath97-valued inner product . we first give a couple of definitions that generalize those introduced by hans baues in @xcite and @xcite , and ranicki in @xcite . these will lead to the most general notion of quadratic refinements for which we construct a universal example . later we shall specialize the definitions from _ non - commutative _ to _ commutative _ and finally , to _ symmetric _ quadratic forms and construct universal examples in all cases . @xmath139 will serve as the range of the ( non - commutative ) quadratic forms defined below . we will write both groups additively since in most examples @xmath137 turns out to be commutative . a morphism @xmath140 between quadratic groups is a pair of homomorphisms @xmath141 such that both diagrams involving @xmath142 commute . [ ex : m ] the example motivating the notation comes from homotopy theory , see e.g.@xcite . for @xmath143 , let @xmath144 , @xmath145 , @xmath146 be the hopf invariant and @xmath136 be given by post - composing with the whitehead product @xmath147:s^{2n-1}\to s^n$ ] . this quadratic group satisfies @xmath148 which is part of the definition used in @xcite , where @xmath137 is also assumed to be commutative . as we shall see , these additional assumptions have the disadvantage that they are not satisfied for the universal example [ ex : universal ] . another important example comes from an abelian group with involution @xmath149 . then we let @xmath150):=x+x^*\ ] ] and @xmath136 be the natural quotient map . for example , if @xmath9 is a ring with involution @xmath151 , then we get two possible involutions on the abelian group @xmath9 : @xmath152 . the choice of sign determines whether we study symmetric respectively skew - symmetric pairings . [ lem : involutions ] given a quadratic group , the endomorphism @xmath155 gives an involution on @xmath135 ( which we will denote by @xmath156 ) . moreover , the formula @xmath157 defines an anti - involution on @xmath137 . these satisfy the proof of lemma [ lem : involutions ] is straightforward and will be left to the reader . to show that @xmath161 is an anti - homomorphism one uses that @xmath162 is central and that @xmath163 is an anti - homomorphism . [ def : quadratic group refinement ] a quadratic group @xmath139 is a _ quadratic refinement _ of an abelian group with involution @xmath149 if @xmath164 it follows from ( i ) in lemma [ lem : involutions ] that in this case , the image of @xmath146 lies in the fixed point set of the involution : @xmath165 . it follows from ( ii ) in lemma [ lem : involutions ] that the additional condition @xmath148 used in @xcite is satisfied if and only if @xmath167 , or equivalently , if @xmath136 factors through the cofixed point set of the involution : @xmath168 it follows that the notion in @xcite is equivalent to that in @xcite , except that @xmath135 is assumed to be the ground ring @xmath169 in the former . in that case , our involution is simply @xmath170 , where @xmath171 and @xmath151 is the given involution on the ring @xmath169 . then @xmath172-symmetric forms in the sense of ranicki become hermitian forms in the sense defined below . in particular , ranicki s @xmath173-symmetric forms are different from the notion of _ symmetric _ form in this paper : we reserve it for the easiest case where both involutions , @xmath156 and @xmath161 , are trivial . [ def : quadratic ] a _ ( non - commutative ) quadratic form _ on an abelian group @xmath174 with values in a ( non - commutative ) quadratic group @xmath175 is given by a bilinear map @xmath176 and a map @xmath177 satisfying we say that @xmath180 is a _ quadratic refinement _ of @xmath6 : property ( i ) says that @xmath180 is quadratic and property ( ii ) means that it `` refines '' @xmath6 . the notation @xmath137 and @xmath135 was designed ( by baues ) to reflect the number of variables ( * e*ntries ) of the maps @xmath180 and @xmath6 respectively . he also writes @xmath181 and @xmath182 , however , we decided not to follow that part of the notation . we write @xmath183 for such quadratic forms and we always assume that the quadratic group @xmath139 is part of the data for @xmath184 . this means that the morphisms in the category of quadratic forms are pairs of morphisms @xmath185 such that both diagrams involving @xmath186 commute . [ lem : symmetries ] let @xmath183 be a quadratic form as above . then @xmath6 is _ hermitian _ with respect to the involution @xmath187 on @xmath135 : @xmath188 and @xmath180 is _ hermitian _ with respect to the anti - involution @xmath189 on @xmath137 : @xmath190 starting with a hermitian form @xmath6 with values in a group with involution @xmath149 , the first step in finding a quadratic refinement for @xmath6 is to find a quadratic refinement @xmath139 of @xmath149 in the sense of definition [ def : quadratic group refinement ] , motivating our terminology . [ ex : universal ] given a hermitian form @xmath195 , one gets a quadratic refinement @xmath196 of @xmath6 as follows . set @xmath197 and define the universal target @xmath198 to be the group consisting of pairs @xmath199 with @xmath200 and @xmath201 and multiplication given by @xmath202 in other words , @xmath203 is the central extension @xmath204 & m_{ee } \ar[r ] & m_{ee } \times_\lambda a \ar[r ] & a \ar[r ] & 1 } \ ] ] determined by the cocycle @xmath6 , compare section [ sec : presentations ] . it follows that @xmath137 is commutative if and only if @xmath6 is _ symmetric _ in the naive sense that @xmath205 . set @xmath206 we claim that @xmath207 is a quadratic group as in definition [ def : quadratic group ] . it is clear that @xmath208 is a homomorphism with image in the center of @xmath137 . the homomorphism property of @xmath209 follows from the fact that @xmath6 is bilinear and hermitian : @xmath210 condition ( iii ) of a quadratic group is also checked easily : @xmath211 we also see that @xmath212 which means that @xmath213 `` refines '' ( in the sense of definition [ def : quadratic group refinement ] ) the group with involution @xmath149 . finally , setting @xmath214 , we claim that @xmath215 is a quadratic refinement of @xmath6 . we need to check properties ( i ) and ( ii ) of a quadratic form ( definition [ def : quadratic ] ) : ( i ) is simply @xmath216 , and ( ii ) explains why we used a sign in front of @xmath6 in our central extension : @xmath217 the following result will show that @xmath196 is indeed a _ universal _ quadratic refinement of @xmath6 . this is the content of the first statement in the theorem below . it follows from the second statement because for any quadratic refinement @xmath180 of @xmath6 it shows that forgetting the quadratic data gives canonical isomorphisms @xmath218 where @xmath219 respectively @xmath220 are ( the morphisms in ) the categories of quadratic respectively hermitian forms . since @xmath221 and the morphisms in the category @xmath222 of quadratic refinements of @xmath6 by definition all lie over the identity of @xmath6 , the set @xmath223 contains a unique element , namely the required universal morphism @xmath224 . [ thm : universal ] the quadratic form @xmath225 is initial in the category of quadratic refinements of @xmath6 . in fact , the forgetful functor @xmath226 from the category of quadratic forms to the category of hermitian forms has a left adjoint @xmath227 given by @xmath228 . we have to construct natural isomorphisms @xmath229 for any quadratic form @xmath230 and hermitian form @xmath6 . recall that the morphisms in @xmath219 are pairs @xmath231 and @xmath232 such that the relevant diagrams commute . this implies that forgetting about the quadratic datum @xmath233 gives a natural map from the left to the right hand side above . given a morphism @xmath234 consisting of homomorphisms @xmath235 and @xmath236 such that @xmath237 we need to show that there is a _ homomorphism @xmath238 such that the following 3 diagrams commute : @xmath239 |{(1 ) } m_{e } \ar[r]^{h } \ar[d]_{\beta_e } & m_{ee } \ar[d]^{\beta_{ee } } & \ar @ { } [ dr ] |{(2 ) } m_{ee } \ar[r]^{p } \ar[d]_{\beta_{ee } } & m_{e } \ar[d]^{\beta_{e } } & \ar @ { } [ dr ] |{(3 ) } a \ar[r]^{\mu_{\lambda } } \ar[d]_{\alpha } & m_{e } \ar[d]^{\beta_{e } } \\ m_{e } ' \ar[r]^{h ' } & m_{ee } ' & m_{ee } ' \ar[r]^{p ' } & m_{e } ' & a ' \ar[r]^{\mu ' } & m_{e } ' } \ ] ] we will now make use of the fact that @xmath240 because @xmath196 is given as in example [ ex : universal ] . in this case , diagrams ( 2 ) and ( 3 ) are equivalent to @xmath241 because @xmath242 and @xmath243 . this implies directly the uniqueness of @xmath233 . for existence , we only have to check that the formula @xmath244 gives indeed a group homomorphism @xmath245 that makes diagram ( 1 ) commute . note that the image of @xmath246 is central in @xmath247 and hence the order of the summands does not matter . we have @xmath248 to get to the fourth line , we used property ( ii ) of a quadratic form to cancel the term @xmath249 . for the commutativity of diagram ( 1 ) we use property ( i ) of a quadratic form , as well as the fact that @xmath250 preserves the involution @xmath156 : @xmath251 this finishes the proof of left adjointness of @xmath227 . if the bilinear form @xmath6 happens to be _ symmetric _ , or more precisely , if it takes values in a group @xmath135 with _ trivial _ involution @xmath156 , then the above construction still gives a quadratic refinement @xmath196 . its target quadratic group @xmath213 has the properties that @xmath137 is abelian and @xmath252 . it is not hard to see that our construction above leads to the following result . [ thm : universal ] for any symmetric form @xmath6 one can functorially construct a quadratic form @xmath225 that is initial in the category of quadratic refinements of @xmath6 with trivial involution @xmath156 . in fact , the forgetful functor @xmath226 from the category of quadratic forms with trivial involution @xmath156 to the category of symmetric forms has a left adjoint @xmath253 . the case where @xmath156 is non - trivial but the anti - involution @xmath161 on @xmath137 is trivial is even more interesting . in this case , @xmath6 is still hermitian with respect to @xmath156 but one is only interested in quadratic refinements @xmath180 that are symmetric in the sense that @xmath257 . this case deserves its own definition : in fact , a commutative quadratic group is the same thing as a non - commutative quadratic group with trivial anti - involution @xmath161 . this comes from the fact that the squaring map @xmath259 is a homomorphism if and only if @xmath137 is commutative . our universal example @xmath213 is in general _ not _ commutative because one gets in this case @xmath260 however , we shall see in theorem [ thm : commutative universal ] that we can just divide by these relations @xmath261 to obtain another universal quadratic refinement of a given hermitian form @xmath6 but this time with values in a _ commutative _ quadratic group . before we work this out , let us mention the essential example from topology . consider a manifold @xmath262 of dimension @xmath263 and let @xmath139 be as in ( m ) from example [ ex : m ] with @xmath264 $ ] . in particular , we have @xmath265 but in general the involution @xmath156 is non - trivial . on group elements , it is given by @xmath266 with @xmath267 ( induced by ) the first stiefel - whitney class of @xmath262 . then the equivariant intersection form @xmath268 on @xmath269 is bilinear and hermitian as required . moreover , the self - intersection invariant @xmath270 defined by wall @xcite gives a quadratic refinement of @xmath271 , at least on the subgroup @xmath174 of elements represented by immersed @xmath33-spheres with vanishing normal euler number . since the involution @xmath274 is trivial by assumption , we already know that @xmath257 from lemma [ lem : symmetries ] . thus it suffices to prove the claim for positive @xmath275 by induction : @xmath276 here we again used the fact that @xmath277 . [ thm : commutative universal ] any hermitian bilinear form @xmath6 has a universal commutative quadratic refinement . in fact , the forgetful functor @xmath226 from the category @xmath278 of commutative quadratic forms to the category @xmath220 of hermitian forms has a left adjoint @xmath279 . as hinted to above , we will force the anti - involution @xmath161 to be trivial in the universal construction of theorem [ thm : universal ] . this means that we should define the universal ( commutative ) group @xmath280 as the quotient of our previously used group @xmath281 by the relations @xmath282 by setting @xmath57 respectively @xmath36 to zero , these relations imply @xmath283 vice versa , these two types of equations imply the general ones and hence we see that @xmath280 is the quotient of the centrally extended group @xmath204 & m_{ee}/(m^*=m ) \ar[r ] & m_{ee}/(m^*=m ) \times_\lambda a \ar[r ] & a \ar[r ] & 1 } \ ] ] by the relations @xmath284 . we write elements in @xmath280 as @xmath285 $ ] with the above relations understood . it then follows that @xmath286 $ ] is a homomorphism @xmath287 ( which is in general not any more injective ) . moreover , our original formula leads to a homomorphism @xmath288 given by @xmath289:= h_\lambda(m , a)= m+m^ * + \lambda(a , a)\ ] ] to see that this is well defined , observe @xmath290 and @xmath291 finally , we set @xmath292 $ ] to obtain a commutative quadratic refinement of @xmath6 which is proven exactly as in theorem [ thm : universal ] . to show that @xmath293 is universal , or more generally , that @xmath294 is a left adjoint of the forgetful functor @xmath169 , we proceed as in the proof of theorem [ thm : universal ] : we are given a morphism @xmath234 consisting of homomorphisms @xmath235 and @xmath236 such that @xmath295 we need to show that there is a _ unique _ homomorphism @xmath296 such that the three diagrams from the proof of theorem [ thm : universal ] commute . we can use the same formulas as before , if we check that they vanish on our new relations in @xmath280 . for this we ll have to use that the given quadratic group @xmath297 is _ commutative_. recall the formula @xmath298 splitting our relations into two parts as above , it suffices to show that @xmath299 the first equation follows from part ( iii ) of lemma [ lem : involutions ] and the fact that we are assuming that @xmath300 : @xmath301 for the second equation we compute : @xmath302 we used lemma [ lem : square ] for @xmath303 and hence the commutativity of @xmath139 . the simplest case of a quadratic group is where both @xmath156 and @xmath161 are trivial . let s call such a quadratic group @xmath175 _ symmetric_. equivalently , this means that @xmath304 ( and hence @xmath137 is commutative ) . then a quadratic form @xmath305 will automatically be _ symmetric _ in the sense that @xmath306 we call @xmath180 a _ symmetric quadratic refinement _ of @xmath6 and obtain a category of symmetric quadratic forms with a forgetful functor @xmath169 to the category of symmetric forms . it is not hard to show that the construction in theorem [ thm : commutative universal ] gives a universal symmetric quadratic refinement @xmath293 for any given symmetric bilinear form @xmath6 . more precisely , [ thm : symmetric universal ] any symmetric bilinear form @xmath6 has a universal symmetric quadratic refinement . in fact , the forgetful functor @xmath226 from the category @xmath307 of symmetric quadratic forms to the category @xmath308 of symmetric forms has a left adjoint @xmath279 . [ rem : injective ] we observe that the map @xmath309 is a monomorphism in this easiest , symmetric , case , just like it was in the hardest , non - commutative , case . this can be seen by noting that the first set of relations @xmath310 is redundant if the involution @xmath156 is trivial . therefore , if @xmath311 $ ] then @xmath312 must come from the second set of relations , i.e. it must be of the form @xmath313 this implies that @xmath314 and hence @xmath315 which in turn means @xmath316 . [ ex : quadratic ] if @xmath317 then @xmath318 and @xmath319 is a canonical choice for which @xmath180 is determined by @xmath6 . another canonical choice is @xmath320 and @xmath321 . then a quadratic refinement of @xmath322 with this choice exists exactly for even forms , at least for free groups @xmath174 . moreover , if @xmath135 has no 2-torsion then a quadratic refinement is uniquely determined by the given even form . at the other extreme , consider @xmath323 . if @xmath174 is a finite dimensional @xmath324-vectorspace then non - singular symmetric bilinear forms @xmath6 are classified by their rank and their _ parity _ , i.e. whether they are even or odd , or equivalently , whether they admit a quadratic refinement or not . in the even case , quadratic forms @xmath184 are classified by rank and _ arf invariant_. this additional invariant takes values in @xmath324 and vanishes if and only if @xmath180 takes more elements to zero than to one ( thus the arf invariant is sometimes referred to as the `` democratic invariant '' ) . if @xmath6 is odd then the following trick allows one to still define arf invariants and it motivates the introduction of @xmath203 . let again @xmath174 be a finite dimensional @xmath324-vectorspace , @xmath325 and @xmath326 with the unique nontrivial homomorphisms @xmath327 . then any non - singular symmetric bilinear form @xmath6 has a quadratic refinement @xmath180 and quadratic forms @xmath184 are classified by rank and an arf invariant with values in @xmath328 . if @xmath6 is even , this agrees with the previous arf invariant via the linear inclusion @xmath329 . consider a central group extension @xmath330 and assume that @xmath9 and @xmath174 have presentations @xmath331 respectively @xmath332 . to avoid confusion , we write groups multiplicatively for a while and switch back to additive notation when returning to hermitian forms . it is well known how to get a presentation for @xmath333 : pick a section @xmath334 with @xmath335 which is not necessarily multiplicative . write a relation in @xmath174 as @xmath336 , where @xmath337 are generators of @xmath174 or their inverses , then @xmath338 where @xmath339 is a word in the generators of @xmath9 . this equation follows from the fact that the projection @xmath340 is a homomorphism and for simplicity we have identified @xmath9 with its image in @xmath333 . we obtain the presentation @xmath341 , \beta_\ell \ , w_\ell \rangle\ ] ] where @xmath342 and @xmath343 is the same word in the @xmath344 as @xmath345 is in the @xmath346 . the commutators @xmath347 $ ] arise because we are assuming that the extension is central , in a more general case one would write out the action of @xmath174 on @xmath9 . it will be useful to rewrite this presentation as follows . observe that the section @xmath348 satisfies @xmath349 for a uniquely determined _ cocycle _ @xmath350 . by induction one shows that @xmath351 comparing this expression with the definition of the word @xmath352 in the presentation of @xmath333 , it follows that @xmath353 so that the above presentation of @xmath333 is entirely expressed in terms of the cocycle @xmath20 ( and does not depend on the section @xmath348 any more ) . now assume that @xmath354 is a hermitian form with respect to an involution @xmath156 on @xmath9 . then the universal ( non - commutative ) quadratic group @xmath137 from example [ ex : universal ] is a central extension as above with cocycle @xmath355 . reverting to additive notation , we see that @xmath356 where the ordering of the summands is irrelevant because @xmath9 is central in @xmath137 . summarizing the above discussion , we get . [ lem : presentation ] the universal ( non - commutative ) quadratic group @xmath137 corresponding to the hermitian form @xmath6 has a presentation @xmath357 , \beta_\ell + \sum_{1\leq i < j \leq r } \lambda(a'_i , a'_j ) \rangle\ ] ] where the generators @xmath358 and words @xmath359 are defined as above . moreover , the universal quadratic refinement @xmath360 is a ( in general non - multiplicative ) section of the central extension and hence @xmath361 for the generators @xmath346 of @xmath174 . as discussed in theorem [ thm : commutative universal ] , we get the universal _ commutative _ quadratic group @xmath362 for @xmath6 by adding the relations @xmath363 and @xmath364 . the latter can be rewritten in the form @xmath365 . in the current notation , where @xmath312 is identified with @xmath366 , we obtain the relations @xmath367 recalling that @xmath368 and @xmath362 are _ commutative _ groups , we can write our presentation in that category to obtain [ lem : commutative presentation ] the universal ( commutative ) quadratic group @xmath280 corresponding to the hermitian form @xmath354 has a presentation @xmath369 here @xmath331 is a presentation of @xmath9 and @xmath346 are generators of @xmath174 . moreover , for every relation @xmath370 in @xmath174 , we use the word @xmath371 . if we apply this construction to the universal inner product on order @xmath33 rooted trees @xmath372 we obtain a universal symmetric quadratic refinement @xmath373 let us compute the presentation from lemma [ lem : commutative presentation ] in this case . recall that the generators of @xmath374 are rooted trees @xmath79 of order @xmath33 and the relations are the as and ihx relations from figure [ fig : relations ] . similarly , @xmath97 is generated by unrooted trees @xmath121 of order @xmath263 , modulo the same relations . putting these together , we see that @xmath375 is generated by unrooted trees @xmath121 of order @xmath263 and elements @xmath376 , one for each rooted tree @xmath79 of order @xmath33 . the three types of relations from lemma [ lem : commutative presentation ] are : * relations in @xmath377 are ordinary as and ihx relations for unrooted trees @xmath121 , * every relation @xmath345 in @xmath378 is an as - relation @xmath379 or an ihx - relation @xmath380 . we obtain the following _ twisted _ as- respectively ihx - relations : @xmath381 * @xmath382 the last relation @xmath20 builds in the commutativity of the universal group as discussed above because we are in the easiest , symmetric , setting where the involution @xmath156 is trivial . using relation @xmath20 , the twisted as relation simply becomes @xmath383 which was expected since we are in the symmetric case . this relation means that the orientation of @xmath79 is irrelevant when forming @xmath376 and in fact , with some care one can see that the twisted ihx - relation makes sense for unoriented trees . [ lem : quadratic - form ] this is a presentation for the target group @xmath384{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ { 2n}(m)$ ] of twisted whitney towers from definition [ def : twisted - tree - groups ] . the translation comes from setting @xmath106{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } = q(j)$ ] for rooted trees @xmath79 ( and keeping unrooted trees unchanged ) . we need to show that the twisted ihx - relations in the original definition of @xmath385{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ { 2n}(m)$ ] are equivalent to the twisted ihx - relations above , all other relations were already shown to agree . this is very easy to see in the presence of the interior - twist relations : together with the ( untwisted ) ihx - relations , they imply that @xmath386 this last expression is exactly the difference between the two versions of the twisted ihx - relations . [ cor : quadratic ] there is an isomorphism of symmetric quadratic groups @xmath387{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ { 2n}(m)\ ] ] which is the identity on @xmath388 and takes @xmath376 to @xmath106{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } $ ] for rooted trees @xmath79 . the quadratic group structure on @xmath108{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ { 2n}(m)$ ] is given by the homomorphisms @xmath389{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } _ { 2n}(m)\overset{h}{\to}{\mathcal{t}}_{2n}(m)$ ] which are uniquely characterized ( for unrooted trees @xmath121 and rooted trees @xmath79 ) by @xmath390{infty2.pdf}\end{minipage } } { \begin{minipage}{.12in}\includegraphics[width=.12in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.10in}\includegraphics[width=.10in]{infty2.pdf}\end{minipage } } { \begin{minipage}{.08in}\includegraphics[width=.08in]{infty2.pdf}\end{minipage } } } } ) = \langle j , j \rangle\ ] ]

recent results from the pierre auger observatory ( pao ) indicate that the composition of ultra - high - energy cosmic rays ( uhecrs ) with energies above ev may be dominated by heavy nuclei . an important question is whether the distribution of arrival directions for such uhecr nuclei can exhibit observable anisotropy or positional correlations with their astrophysical source objects despite the expected strong deflections by intervening magnetic fields . for this purpose , we have simulated the propagation of uhecr nuclei including models for both the extragalactic magnetic field ( egmf ) and the galactic magnetic field ( gmf ) . we find that the gmf is particularly crucial for suppressing the anisotropy as well as source correlations . assuming that only iron nuclei are injected steadily from sources with equal luminosity and spatially distributed according to the observed large scale structure in the local universe , at the number of events published by the pao so far ( 69 events above ev ) , the arrival distribution of uhecrs would be consistent with no auto - correlation at 95% confidence if the mean number density of uhecr sources mpc , and consistent with no cross - correlation with sources within 95% errors for mpc . on the other hand , with 1000 events above ev in the whole sky , next generation experiments can reveal auto - correlation with more than 99% probability even for mpc , and cross - correlation with sources with more than 99% probability for mpc . in addition , we find that the contribution of centaurus a is required to reproduce the currently observed uhecr excess in the centaurus region . secondary protons generated by photodisintegration of primary heavy nuclei during propagation play a crucial role in all cases , and the resulting anisotropy at small angular scales should provide a strong hint of the source location if the maximum energies of the heavy nuclei are sufficiently high . ultra - high - energy cosmic rays

1202.2874

the origin of ultra - high - energy cosmic rays ( uhecrs ) with energies @xmath7 ev is an intriguing mystery in modern astrophysics . their sources are generally believed to be extragalactic objects , although some galactic objects may also be viable . in either case , the maximum energy of @xmath8 ev can only be achieved in extreme environments @xcite . prominent source candidates suggested so far include active galactic nuclei ( agn ) @xcite , gamma - ray bursts ( grbs ) @xcite , neutron stars or magnetars @xcite and clusters of galaxies @xcite . if uhecrs with energies above @xmath9 ev are mainly protons , their propagation distance should be limited by interactions with cosmic microwave background ( cmb ) photons @xcite , the so - called greisen - zatsepin - kuzmin ( gzk ) mechanism , so that they are observable only from sources sufficiently nearby ( typically @xmath10 mpc ) . since all known astrophysical objects are distributed inhomogeneously at such distances in the local universe , anisotropies in the distribution of uhecr arrival directions are expected as long as cosmic magnetic fields are weak enough to allow quasi - rectilinear propagation of the uhecrs . searches for correlations between the arrival directions and the celestial positions of potential source candidates should then provide valuable clues to reveal their origin @xcite . the pierre auger observatory ( pao ) operating in the southern hemisphere has reported possible evidence of a gradual increase in the average mass of uhecr particles with energies above @xmath11 ev by analyzing the average depth of the shower maximum @xmath12 and the root mean square of the shower - to - shower fluctuations of @xmath13 ( @xmath14 ) , assuming that current hadronic interaction models are realistic at these energies @xcite . this is in contrast to measurements of similar quantities for the northern sky by the high resolution fly s eye ( hires ) experiment that are consistent with an uhecr composition dominated by protons @xcite . if uhecrs consisted mainly of heavy nuclei rather than protons as indicated by the pao results , their deflections during propagation in magnetized extragalactic and galactic environments would be much larger than those for protons , and the hopes for observing anisotropy may be considerably weakened . on the other hand , pao has seen statistically significant correlations between the arrival directions of uhecrs with energies above @xmath9 ev and the projected positions of agn with @xmath15 within an angular scale of @xmath16 @xcite . this does not necessarily point to agn as the true sources , since they may only be tracers of the more numerous population of galaxies that make up the anisotropic , local cosmic matter distribution , and there are also considerable uncertainties in the angular displacement due to deflections by both the extragalactic magnetic field ( egmf ) and the galactic magnetic field ( gmf ) . nevertheless , this does imply that uhecr sources are associated with the matter distribution . independent analyses of the pao data have confirmed the correlations of uhecr arrival directions with various kinds of nearby astrophysical objects @xcite . such correlations were predicted in several theoretical studies under the assumption of protons as the primary uhecr particles @xcite . we must beware that at the moment , the positional correlations ( or anisotropy ) are observed only _ above _ @xmath9 ev , whereas the pao results on the composition are available only _ up to _ @xmath9 ev due to the lack of event statistics . we can not rule out the possibility that protons become dominant above this energy and produce the anisotropy . however , there are presently no hints of such a sudden change in the composition , and the simplest extrapolation of the measured trends of @xmath12 and @xmath14 to higher energies would entail a composition dominated by heavy nuclei . since the appearance of anisotropies or cross correlations with astrophysical objects in the case of uhecr nuclei is not trivial , deeper , relevant studies of uhecr propagation are warranted in order to achieve a consistent interpretation of the current data set as well as to quantify the prospects for future observations . propagation of uhe nuclei has been studied previously by many authors from various perspectives @xcite . many have studied the photodisintegration process and consequent spectra in detail without accounting for magnetic fields , but only a few have focused on the effects of the magnetic fields . ref . @xcite calculated the propagation of uhe nuclei in a uniform turbulent egmf model and examined their trajectories and photodisintegration interactions in intergalactic photon fields , concluding that a characteristic feature in the spectrum may result . @xcite discussed the propagation of nuclei in a structured egmf model obtained through a numerical simulation of cosmological structure formation , showing that while heavy nuclei can be strongly deflected ( @xmath17 ) by the egmf , nuclei ( and their secondaries ) with relatively small deflections can generate anisotropy on intermediate angular scales . in this paper , we study the propagation of uhecr nuclei in cosmic radiation and magnetic fields under different assumptions for the source properties and the fields , focusing on the resulting distribution of arrival directions . in addition to a detailed treatment of the photopair , photomeson and photodisintegration interactions of uhe nuclei with background photons , deflections in gmf and egmf are taken into account for both primary nuclei as well as secondary nuclei arising from photodisintegration . by statistically analyzing the calculated arrival distributions , we discuss the implications for current and future observations of anisotropy and cross correlations with sources . motivated by the recent pao result , we adopt a model for the uhecr source distribution that follows the density of large - scale structure actually observed in the local universe . this is in contrast to refs . @xcite who took the sources to be related to the matter distribution in a numerical simulation , but not reflecting the actual universe . because photodisintegration limits the propagation distance of heavy nuclei in a way similar to the gzk mechanism for protons , a realistic account of local source inhomogeneity is essential . unlike most previous work on the subject , here we also account for the gmf as well as the egmf , since the former unavoidably affects all uhecrs arriving at earth . concerning the uhecr source composition , a wide variety of possibilities are currently allowed that are consistent with both the observed spectrum and the composition indicators @xmath13 and @xmath14 ( e.g. , @xcite ) . this study assumes a pure iron composition , which can also reproduce the highest - energy spectrum , and which would be the most pessimistic situation for producing anisotropy or source correlations in view of their large magnetic deflections . this paper is laid out as follows . in section [ method ] we describe our models and methods of calculation and statistical analysis . parametrization of photodisintegration cross sections to calculate the mean free paths of nuclei in cosmic background radiation fields is summarized in section [ method_pd ] . section [ method_mf ] is dedicated to our models for the gmf and egmf . a method of calculation for nuclear propagation is introduced in section [ method_prop ] , and that for uhecr arrival distribution is presented in section [ method_ad ] . the contents of these two subsections are essentially based on ref . @xcite that dealt with uhe proton propagation in intergalactic space , and here we extend their methods to heavy nuclei . we describe our statistical methods for studying anisotropy and cross correlations with sources in section [ method_stat ] . in section [ results ] , we present our simulation results , compare with the current data @xcite , and discuss the implications thereof for the sources of uhe nuclei . we also give special attention to centaurus ( cen ) a , for which several pao events positionally correlate @xcite . we summarize in section [ summary ] . a nucleus emits its constituent nucleons by interactions with photons ( _ photodisintegration _ ) . since the number of emitted nucleons per interaction is stochastically determined following branching ratios dependent on interaction energy , a lot of disintegration paths are realized during propagation in photon fields . however , all the branches in photodisintegration are not well known yet . this study adopts a simple disintegration track on the nuclear chart proposed by ref . there is only one nuclear species for each nuclear mass number @xmath18 in this treatment . photodisintegration can be phenomenologically classified into 4 processes : giant dipole resonance ( gdr ) , quasi - deuteron ( qd ) process , baryonic resonance ( br ) , and photofragmentation ( pf ) . since the importance of the latter two processes appears above @xmath19 ev for fe nuclei in the cmb because of their very high threshold energy ( fig . [ fig : mfp ] , and see also section [ method_prop ] ) , our assumption of @xmath20 ev allows to neglect these processes . this study adopts the parametrization of photodisintegration cross - sections developed by ref . the mean free paths of nuclear photodisintegration in cosmic background photon fields can be calculated from the cross - sections ( explained in section [ method_prop ] ) . gdr has the lowest threshold energy in the photodisintegration processes , typically @xmath21 mev in the nucleus rest frame . a nucleus emits several nucleons or @xmath22 particles , but one nucleon emission is dominated . the treatment of ref . @xcite approximated gdr as a process emitting one nucleon . the cross - section of gdr is parametrized by a lorenzian function @xmath23 as @xmath24 for @xmath25 and 0 otherwise . here @xmath26 is photon energy in the nuclear rest frame , @xmath27 mbarn , @xmath28 mev , and @xmath29 mev . the threshold energy of this process is also scaled as a function of @xmath18 , @xmath30 mev . the quasi - deuteron process is treated as a two nucleon emission process . the parametrization of its cross - section is @xmath31 for @xmath32 where @xmath33 mev . the contribution of individual processes of photodisintegration is also shown for comparison . ] while the scaling laws on @xmath18 can well reproduce experimental data for large @xmath18 nuclei , the behavior of small nuclei ( @xmath34 ) is different because of their small number of nucleons . thus , other analytical functions of the cross - sections are used for deuteron , trinucleon ( triton or @xmath35 ) , helium , and @xmath36be instead of @xmath37 , @xmath38 . @xmath36be is assumed to decay into two @xmath39he and one nucleon . we also assume that a nucleus with @xmath40 which is produced by qd of a particle with @xmath41 immediately decay into two @xmath39he because there is no stable particle with @xmath40 . therefore , nuclei between @xmath42 and @xmath40 are not observed at the earth in this treatment . the photoabsorption cross - section of a deuteron was derived by ref . @xcite as @xmath43 where @xmath44 is a photoabsorption cross - section for a zero range nuclear force @xcite . @xmath45 , @xmath46 , @xmath47 , @xmath22 , @xmath48 mev , @xmath49 fm , and @xmath50 mbarn are the planck constant , the proton mass , the speed of light , the fine structure constant , the nuclear binding energy of a deuteron , the effective range of nuclear force , and the cross - section of thomson scattering for proton mass , respectively . after this reaction , a deuteron is broken to a proton and neutron . a trinucleon has two interaction channels ; it emits a nucleon and is changed into a deuteron , or three body decay . the cross - sections of these processes can be well fitted by using @xmath51 , @xmath52 the cross - sections of one or two particle emission of @xmath22-particle are parametrized as @xmath53 respectively , where @xmath54 here , @xmath55 is a step function and @xmath56 . the cross - section of photo - absorption of @xmath36be is parametrized as @xmath57 where @xmath58 we should also treat the propagation of protons as secondaries . for protons , photopion production plays an important role above @xmath59 ev . the cross - section and energy - loss rate of protons are calculated by a photomeson production event generator _ sophia _ @xcite . based on the cosmological structure formation theory , density fluctuations produced by inflation have gravitationally grown and form large - scale structure of the current universe . galaxy surveys have confirmed through galaxy distribution that matter distribution is highly structured . since magnetic fields has been also amplified through the compression and dynamo mechanism of astrophysical plasmas simultaneously , they could be also structured following the matter distribution . several simulations of cosmological structure formation with magnetic fields have confirmed this expectation qualitatively @xcite , but egmf structures highly depend on modeling due to few observational constraints . the difference appears , for example , in the volume filling factor of strong egmfs ( see ref . ) . therefore , a variety of egmf modelings gives us different results on the deflection of uhecrs and source identification possibility even for protons @xcite . in addition to the uncertainty , it is technically difficult to calculate the arrival distribution of uhecrs with a complex egmf structure because only a very small fraction of uhecrs injected from sources can arrive at the earth . @xcite developed a numerical method to solve this problem for uhe protons , but this method can not apply to the cases of heavy nuclei owing to the stochastic nature of photodisintegration . in order to avoid the uncertainty of egmf modeling and the numerical difficulty , this study assumes a uniform turbulent egmf model with the strength of @xmath60 , the correlation length of @xmath61 , and the kolmogorov power spectrum @xcite . the averaged property of egmfs has been constrained by faraday rotation measurements of distant quasars as @xmath62 @xcite . a recent analysis of cmb and matter power spectra derives an upper limit of the primordial magnetic field at present , which is regarded as a magnetic field in voids , of @xmath63 ng ( 95 % c.l . ) for the coherent length of @xmath64 mpc @xcite . several simulations of cosmological structure formation with magnetic fields indicate @xmath65 ng @xcite . on the other hand , pair - halo and/or pair - echo emission from tev blazars may provide lower bounds , suggesting @xmath66 - @xmath67 g from the data of _ fermi _ large area telescope @xcite . even magnetic fields in voids are negligibly weak , the effective magnetic fields of cosmic structures such as filaments and clusters of galaxies may have @xmath68 ng mpc@xmath69 @xcite . in this study @xmath60 is treated as a parameter , while @xmath70 mpc is fixed which is motivated by the average separation distance between galaxies . the gmf has been relatively well observed compared to egmfs through faraday rotation measurements of radio pulsars and extragalactic radio sources . it has a regular component with a few @xmath71 g on average in the galactic disk , whose shape is similar to the spiral structure of the milky way . there are also turbulent fields with the magnetic strength 0.5 - 2 times as large as that of the spiral component @xcite . its correlation length has been lower - limited to @xmath72 pc due to the finite resolution of radio observations . both the components significantly contribute to the total deflections of uhe nuclei @xcite . however , at present , even the newest data are difficult to discriminate the regular structure between the bisymmetric and axisymmetric models @xcite , and the difference produces significantly different propagation of uhecrs especially in the case of heavy nuclei , e.g. , nuclear propagation is sensitive to gmf models @xcite . in order to avoid the model dependence , we simply assume that the deflection angles of uhecrs do not depend on directions . in ref . @xcite , a gmf model with a weak dipole field predicted a typical deflection angle of @xmath73-@xmath74 for irons with @xmath9 ev on average in the whole sky . assuming a rigidity scaling additionally , we set the typical deflection angles of uhecrs in gmf to be @xmath75 where @xmath76 is the energy of uhecrs . the calculation of nuclear propagation can be divided into two parts ; propagation in extragalactic space and in galactic space . the size of the galactic space ( assumed to be 40 kpc ) is much smaller than the energy - loss length of bethe - heitler pair creation . on the other hand , the mean free path of photodisintegration is close to the size of galactic space above @xmath77 ev for @xmath78fe . however , it does not affect the propagation in the galaxy because particles traveled for longer distances than the mean free path in extragalactic space before entering the milky way . thus , the nuclear reactions can be neglected during the propagation in the galaxy . this allows us to treat the effect of the propagation as modifications of the arrival directions of uhecrs . the modifications will be described in section [ method_ad ] . below , we introduce a method of the propagation of uhecrs in extragalactic space . we calculate the propagation trajectories of uhecrs including secondary particles in extragalactic space , and estimate the nuclear species , energies , deflection angles of nuclei arriving at earth . in our method , the source of primary nuclei ( @xmath78fe ) is located at the center of coordinates in extragalactic space . the primary nuclei are ejected from the source isotropically , and then their trajectories are calculated step by step including secondary particles taking magnetic deflections , photodisintegration , and bethe - heitler pair creation into account . adiabatic energy - loss due to cosmic expansion is neglected because the energy - loss rate is much lower than the others . since this study focuses on the highest energy range , @xmath79 ev , the calculation of the trajectories is finished at 500 mpc from the source . we confirmed that nuclei with the energy above @xmath79 ev can not arrive at the earth from more distant sources even for rectilinear propagation . the nuclear species , energy , and deflection angles of the particles are recorded every 1 mpc . we divide a logarithmic energy bin into 10 per decade , i.e. , @xmath80-@xmath81 ev , @xmath81-@xmath82 ev , @xmath83 , @xmath84-@xmath85 ev , and inject 10,000 fe nuclei in each bin . the record allows us to make the distribution of the nuclear species , energy and deflection angles of the particles at 500 distances . for instance , when we consider uhecrs injected from a source with the distance of 100 mpc , the 100th distributions of them are regarded as their distributions at the earth . in the calculation of propagation of nuclei , photo - disintegration processes are treated as a stochastic process , i.e. , by a monte - carlo method . the mean free path of photo - disintegration process @xmath86 ( @xmath87 gdr , qd ) of a nucleus with the nuclear mass number @xmath18 and the energy @xmath76 can be calculated by using the expressions of cross - sections described in section [ method_pd ] as @xcite , @xmath88 where @xmath89 , @xmath90 , @xmath91 , and @xmath92 are nuclear velocity normalized by @xmath47 , the energy of photons in the laboratory frame , the lorentz invariant mass squared , and nucleus mass , respectively . @xmath93 is cross - section of photo - disintegration process considered , @xmath94 , @xmath95 , and @xmath96^{-1}$ ] . @xmath92 is simply defined as @xmath97 , where @xmath98 mev is nucleon mass . @xmath99 is the number density of extragalactic background light ( ebl ) photons including cmb photons . this study adopts the low - ir model of ref . @xcite . since the bethe - heitler pair creation process is treated in a continuous energy - loss approximation ( see below ) , the occurrence of reactions is judged by the method discussed in ref . gdr and qd are treated independently . we approximate that the lorentz factor of nuclei is unchanged by photodisintegration . for the energy - loss length of the bethe - heitler pair creation , we adopt an analytical fitting formula developed by ref . @xcite , @xmath100 where @xmath101 is the classical electron radius , @xmath102 is electron mass , and @xmath103 . @xmath104 is an integral similar to the first integral of eq . [ eq : mfp ] , but is parametrized as @xmath105^{-1 } & \kappa \leq 25 \\ \kappa \left [ \sum_{i=0}^3 d_i ( \ln \kappa)^i \right ] \left [ 1 - \sum_{i=1}^3 f_i \kappa^{-i } \right]^{-1 } & \kappa > 25 , \end{array } \right.\end{aligned}\ ] ] where @xmath106 , @xmath107 , @xmath108 , @xmath109 , @xmath110 , @xmath111 , @xmath112 , @xmath113 , @xmath114 , @xmath115 , and @xmath116 . the arrival distribution of uhecrs can be calculated using the distributions obtained by the method written in section [ method_prop ] with an assumed source model . we adopt a source model developed in ref . @xcite , in which uhecr sources are distributed following galaxy distribution actually observed . the source distribution of this model is based on the galaxy catalog of the infrared astronomical satellite ( iras ) point source redshift survey ( pscz ) @xcite . this galaxy catalog covers @xmath117 of the whole sky , and is enough to construct the model of matter distribution in local universe ( e.g. , ref . selection effects of galaxies ( e.g. , unobserved sky and galaxies fainter than the flux limit ) are compensated by a luminosity function @xcite , and then the model of galaxy distribution following matter distribution is constructed within 200 mpc . outside 200 mpc , we assume isotropic distribution due to the lack of completeness of galaxies in the catalog . this assumption does not change the results of this study , since distant sources do not contribute to anisotropy . we select galaxies randomly from the galaxy distribution following @xmath118 , which is the number density of uhecr sources as a parameter , and regard a set of the selected galaxies as a source distribution . although , @xmath118 has been estimated as @xmath119 mpc@xmath3 @xcite on the assumption of pure proton composition , it is not clear in the case of heavy nuclei . a smaller @xmath118 may be allowed owing to larger deflections . in this study , we survey over @xmath120-@xmath121 mpc@xmath3 . corresponding objects are shown in table [ tab : nd ] . all the analyses in this study are limited to steady sources . we assume that all the sources emit uhecrs with the same power for simplicity . the injection spectrum of the particles is assumed to be a power - law @xmath122 with @xmath123 ev . we confirmed that this spectrum can reproduce observed spectrum above @xmath59 ev within uncertainties . in table [ tab : nd ] , fanaroff - riley ( fr ) i galaxies and bl lac objects are in different categories . however , following a unification theory of radio - loud agns , these originate from the same population with the only difference due to the viewing angle @xcite . thus , if the deflection angles of uhecrs during propagation in extragalactic space is much larger than a typical opening angle of their jets ( @xmath124 radian ) , we can see these objects as the essentially same objects in the viewpoint of uhecr sources . an assumed @xmath118 constrains the energy budget of uhecr emission per source . the observed spectrum of uhecrs requires @xmath125 erg mpc@xmath3 yr@xmath126 @xcite for uhecr emission above @xmath0 ev . thus , the energetics of uhecr emission required per source is @xmath127 erg s@xmath126 . the flux contribution of each source is inversely proportional to the distance squared . following the injection spectrum weighted by this factor and the distribution of uhecrs in deflections , energies , and nuclear mass number ( see section [ method_prop ] ) , we can simulate the arrival distribution of uhecrs . in addition , we can add the contribution of the gmf in the arrival distribution . once the arrival directions of uhe nuclei are realized following the method above ( considering only egmf ) , the arrival directions are modified following the two dimensional gaussian distribution with the zero mean and the standard deviation of @xmath128 . .local number densities of active objects . [ cols="<,<,<",options="header " , ] [ tab : nd ] anisotropy in the arrival distribution of uhecrs can be quantified by a cumulative auto - correlation function @xcite , @xmath129 where @xmath76 and @xmath130 denote uhecr events and events randomly distributed following the aperture of a uhecr detector , respectively . @xmath131 is the number of uhecr event pairs with the separation angle less than @xmath132 divided by the total number of pairs . @xmath133 and @xmath134 are defined similarly to @xmath131 . this function is essentially equivalent to the quantity that the number of uhecr event pairs with the separation angle within @xmath132 is divided by the total number of pairs and by a solid angle within @xmath132 if the aperture of uhecr observatories is uniform . however , since , in general , the aperture is not uniform in the cases of ground - based detectors , the non - uniformity should be corrected . @xmath130 can correct this in this formula . this function was originally proposed by ref . @xcite as a differential auto - correlation function . this correction simplifies the interpretation of the cumulative auto - correlation function ; @xmath135 shows the positive excess of uhecr distribution within @xmath132 compared to isotropic distribution . many tests of the correlation between the arrival directions of uhecrs and the positions of their source candidates are based on the excess of the number of events within a circle centered by the source candidates over randomly distributed events . this study also adopts a similar method to study the correlation , using a cumulative cross - correlation function , which is defined similarly to the cumulative auto - correlation function , @xmath136 where @xmath137 and @xmath138 denote uhecr sources and sources randomly distributed following selection effects of the catalog of the source candidates . @xmath138 is isotropically distributed in the entire sky in this study because of the correction of selection effects ( see section [ method_ad ] ) . @xmath139 is the normalized number of pairs between uhecr events and sources with the separation angle less than @xmath132 . @xmath140 , @xmath141 , and @xmath142 are similarly defined . the interpretation of this quantity is also simple ; @xmath143 means positive correlation within @xmath132 . in order to calculate eqs.[eq : acor ] and [ eq : ccor ] , the distribution of @xmath130 should be understood . the aperture of a ground array depends on the declination of observed directions reflecting the daily rotation of the earth . the declination dependence of the aperture can be analytically estimated as @xcite @xmath144 where @xmath145 is given by @xmath146 and @xmath147 when observation time is sufficiently larger than a day . here , @xmath148 is the celestial latitude of the ground array and @xmath149 is the zenith angle for a data quality cut because of experimental reasons . for the pao , @xmath150 and @xmath151 @xcite . the @xmath130 sample is generated following this equation . we set the number of @xmath130 events to be 200000 in order that the distribution of random events reflects the pao aperture sufficiently . also , we checked that this particular choice of the number of random events does not affect results . fig.[fig : acor ] shows the cumulative auto - correlation functions of simulated events with 68% errors in the cases of @xmath120-@xmath121 mpc@xmath3 . they are divided into two panels for visibility . the simulated event set imitates the recent pao data @xcite , e.g. , 69 events above @xmath1 ev are distributed following the aperture geometry of the pao . @xmath152 ng and gmf are taken into account . the points and error bars of these cumulative auto - correlation functions are calculated as follows . first of all , 69 events are generated from a source distribution with the source number density of @xmath118 , and then a cumulative auto - correlation function is calculated from this event set . next , the former step is repeated over 1000 source distributions with the same @xmath118 . then , the averages and errors are calculated from the 1000 cumulative auto - correlation functions . when the number of events is small , the distribution of the values of the auto - correlation functions in a bin are asymmetric and therefore the standard deviation of the 1000 auto - correlation functions is not a good indicator of errors . in order to estimate anisotropic error bars , we arrange the values of the 1000 cumulative auto - correlation functions in each bin in order , calculating the range in which the central 68% of the values are included , and then this range is regarded as a 68% ( @xmath153 ) error . this error includes not only an error due to the finite number of events ( poisson error ) but also an error due to the uncertainty originating from the selection of uhecr sources . finally , the averages and 68% errors are plotted in the figure . all the figures in this paper follow this rule , as long as there is not exceptional description . ev on the assumption of the pao aperture for @xmath154 mpc@xmath3 and @xmath152 ng . the gmf is considered ( _ green _ ) and is not considered ( _ red _ ) for comparison . ] all the auto - correlation functions except for @xmath155 mpc@xmath3 are consistent with zero within 68% errors in this angular range , i.e. , the arrival distributions are consistent with isotropic distribution . note that the case of @xmath156 mpc@xmath3 corresponds to only 4 sources within nearby 100 mpc . in spite of such a small number , the predicted arrival distribution is consistent with isotropy because of the strong effects of the gmf and egmf . these magnetic fields make even the case of @xmath155 mpc@xmath3 become consistent with isotropy within 95% error at small angular scale ( @xmath157 ) . the gmf significantly contributes to the arrival distribution of uhecrs . [ fig : gmf ] shows the cumulative auto - correlation functions of 69 simulated events in the case of @xmath154 mpc@xmath3 and @xmath152 ng . in order to see the effect of the gmf , one ( _ green _ ) takes the gmf into account , but the other ( _ red _ ) does not consider the gmf . only in this figure , the medians of 1000 cumulative auto - correlation functions are plotted instead of the averages for visibility because the very large values of cumulative auto - correlation functions are realized in some realizations by very nearby sources at small angular scale and the averaged values of the auto - correlation functions become larger than the upper edge of the 68% error bars when the gmf is not taken into account . note that the medians will be close to the averages when the number of events increases . focusing on the first bin , we can see that the medians in both the cases are close to the lower edges of the error bars . this means that source distributions including nearby sources accidentally make strong anisotropy whereas no event pair appears in the first bin in many realizations because of the small solid angle within @xmath158 and the small number of simulated events . thus , the distribution of the values of the cumulative correlation functions in the first bin has a long tail to large values . although both the cumulative auto - correlation functions are consistent each other ( and also with isotropic distribution ) within 68% errors , the error bars are suppressed when the gmf is considered , that is , the probability that large values of the auto - correlation function are realized becomes smaller . the gmf strongly weakens anisotropy produced by nearby sources . furthermore , we can find that the gmf can dominantly contribute to suppressing anisotropy . [ fig : gmfsup ] shows the dependence of the cumulative auto - correlation functions on @xmath60 in the case of @xmath159 mpc@xmath3 and considering the gmf . for comparison , @xmath160 , @xmath161 , and @xmath162 ng are adopted . the three cumulative auto - correlation functions are quite similar , and therefore are controlled by the gmf . this situation is the same down to the case of @xmath156 mpc@xmath3 . small difference of the averages among the cases of the three egmf strengths appears only for @xmath155 mpc@xmath3 ; @xmath163 ng predicts a bit higher values of the averages , but the difference is still within the error bars . thus , the effect of the gmf is essential and should be inevitably considered when the anisotropy of uhecr arrival distribution is tested . ev simulated on the assumption of the pao aperture for different @xmath60 , @xmath164 , @xmath161 , and @xmath162 ng for comparison . @xmath159 mpc@xmath3 and the gmf are considered . ] + here , we are interested in how often a positive value of the auto - correlation function is realized in each bin . this provides the probability that the positive excess of events appears in each angular scale . this probability is calculated as follows . we simulate a set of uhecr events from a source distribution and calculate the cumulative auto - correlation function of them . this calculation is repeated over 1000 source distributions with the same @xmath118 . then , we focus on the values of the 1000 cumulative auto - correlation functions in an angular bin , counting the number of the auto - correlation functions which are positive , and the number is divided by 1000 . this step is repeated over all the angular bins . since only 1000 source distributions are considered ( because of the computational limitation ) , it can not resolve the probability less than @xmath165 , but it is enough for this study . by definition , isotropic distribution predicts this value of @xmath166 due to statistical fluctuation . this probability also can be regarded as a confidence level in the one - side statistical test to rule out isotropic distribution . for visibility , we plot one minus this probability throughout the paper , which is the probability that the positive excess of events is not realized . [ fig : cprob_auto69 ] shows such probabilities for 69 simulated events above @xmath1 ev . the aperture geometry of the pao is applied . the upper two panels and the lower two panels are calculated under the conditions of @xmath167 ng and @xmath168 ng , respectively . the gmf is considered for only the right two panels . in the case of no magnetic field ( _ upper left _ ) , the strong positive excess of events appears at small angular scale . this is natural because the propagation trajectories are not deflected . in general , the probability that the positive excess of events is realized in a certain bin is higher for a smaller @xmath118 at small to intermediate angular scale because each source has stronger power and stronger clusterings of uhecr events are expected in the directions of their sources . the positive anisotropy is produced at more than 99% for @xmath169 mpc@xmath3 at small angular scale . note that this does not mean that the positive excess signal of a set of observed events compared to isotropic distribution is larger than 99% confidence level at the angular scale . magnetic fields change the shape of the probability curves . when the gmf is taken into account ( _ upper right _ ) , the angular scale at which the value of a probability curve is minimized moves from the first angular bin to larger angular scale for all the @xmath118 . the possibility of positive excess is maximized at this angular scale . this angular scale reflects the deflection angles of uhecr trajectories by the gmf . since these are independent of the distances of sources , it does not depend on @xmath118 . although the egmf gives a similar effect to the gmf ( _ lower left _ ) , the angular scale of the minimum of a probability curve is different and depends on @xmath118 because the deflections of uhecrs by the egmf are dependent on their propagation distance and the nearest sources , which contribute to anisotropy most strongly , are more distant in the case of a smaller @xmath118 . in both cases only sources with @xmath155 mpc@xmath3 produce positive excess in intermediate angular scale with more than 95 % probability . the result when both the gmf and egmf are taken into account ( _ lower right _ ) can be interpreted as combination of the upper right and lower left panels . the contribution of the two magnetic fields reduces the probability that positive excess of events appears and moves the angular scale of the maximum of the probability to larger angular scale . at the end of this section we consider the case with the number of events expected by future uhecr experiments such as the northern pierre auger observatory @xcite and extreme universe space observatory on board japanese experiment module ( jem - euso ) @xcite . [ fig : cprob_auto1000 ] is the same figure as that in the lower right panel of fig . [ fig : cprob_auto69 ] ( the gmf and egmf are considered ) , but for 1000 events above @xmath1 ev on the assumption of the uniform aperture . all the cases of @xmath118 predict the positive excess of events at intermediate angular scale at more than 99 % . we notice that the case of @xmath120 mpc@xmath3 shows a larger probability of the positive excess than that of @xmath154 mpc@xmath170 at around @xmath73 . the extremely close sources included for @xmath120 mpc@xmath3 ( within a few mpc ) causes this feature . when sources within 5 mpc are artificially neglected for example to check this , the probability curves for @xmath171 mpc@xmath3 change significantly and the inversion disappears . in this case , the probability for @xmath120 mpc@xmath3 becomes a bit smaller to 90% . , but for 1000 events above @xmath1 ev simulated on the assumption of a uniform aperture . ] fig.[fig : ccor1 ] shows the cumulative cross - correlation functions between the arrival directions of 69 uhecrs above @xmath1 ev simulated on the assumption of the pao aperture and the celestial positions of their sources within 75 mpc with 68% errors in the cases of @xmath120-@xmath172 mpc@xmath3 . the 75 mpc is motivated by the first report of the correlation between uhecrs and nearby extragalactic sources by the pao @xcite . @xmath173 ng and the gmf are taken into account . the error bars and the averaged points of the cross - correlation functions are estimated similarly to those of cumulative auto - correlation functions . note that we do not plot the cumulative cross - correlation functions for @xmath156 and @xmath121 mpc@xmath3 because there are source distributions not having sources within 75 mpc in 1000 source realizations due to the small number densities . the cumulative cross - correlation functions for @xmath120 and @xmath174 mpc@xmath3 are fully consistent with zero within 68% errors in this angular scale , i.e. , are consistent with no correlation . the case of @xmath159 mpc@xmath3 is positive at small angular scale with 68% errors , but is consistent with no correlation within 95% errors . the error bars are longer for smaller @xmath118 because the fluctuation of source positions is larger . , @xmath174 , and @xmath172 mpc@xmath3 within 75 mpc . the simulated event sets imitate the published events of the pao , i.e. , 69 events with energies above @xmath1 ev simulated on the assumption of the pao aperture . both egmf ( @xmath152 ng ) and the gmf are taken into account . the points are the averaged values of the cumulative cross - correlation functions over 1000 source realizations for each @xmath118 . the error bars correspond to 68% errors . ] ev simulated under the pao aperture and sources within 75 mpc for different @xmath60 , @xmath164 , @xmath161 , and @xmath168 ng for comparison . @xmath154 mpc@xmath3 and the gmf are considered . ] the cross - correlation between uhecrs and their sources is also dominantly affected by the gmf . fig.[fig : ccor2 ] is similar to fig . [ fig : gmfsup ] , but the cumulative cross - correlation functions between the 69 uhecrs and their sources with @xmath154 mpc@xmath3 within 75 mpc . @xmath160 ( _ red _ ) , @xmath161 ( _ green _ ) , and @xmath162 ng ( _ blue _ ) are considered with the gmf for comparison . similarly to fig . [ fig : gmfsup ] , the three cumulative cross - correlation functions are very similar . the situation is the same even for @xmath120 and @xmath172 mpc@xmath3 . even in this figure , we can notice the importance of the gmf on the discussions of the arrival distribution of uhecrs . estimating the probability that positive correlation between uhecrs and their sources appears is also useful . we can calculate this probability by a similar method to calculate fig . [ fig : cprob_auto69 ] . following that figure , we plot the possibility that the values of the cross - correlation function are not positive for visibility in fig.[fig : ccorprob69 ] similarly to fig . [ fig : cprob_auto69 ] . we consider 69 uhecrs above @xmath1 ev on the assumption of the pao aperture . the sources used for calculating the cross - correlation function are within 75 mpc from the galaxy . the basic tendency of the probability curves is similar to fig.[fig : cprob_auto69 ] . in the case of no magnetic field , strong correlation is predicted at the smallest angular bin because of the absence of the deflections of uhecrs . taking the gmf and/or egmf into account , the probability curves have a minimum at intermediate angular scale . for the cross - correlation function , the angular scale at which gives the minimum depends not only the deflection angles of uhecrs but also the typical separation angle between sources . the potential minimum is at @xmath175 for @xmath159 mpc@xmath3 if the gmf and/or egmf are considered , but the probability of the positive correlation is less than 95% . + again , we demonstrate the prospects of future uhecr observatories . fig.[fig : ccorprob1000 ] shows cumulative cross - correlation functions ( _ left _ ) and the corresponding probability curves ( _ right _ ) . we consider 1000 uhecrs above @xmath1 ev on the assumption of the uniform aperture and both the gmf and egmf ( @xmath152 ng ) . in the left panel , the cumulative cross - correlation functions in the cases of @xmath154 and @xmath172 mpc@xmath3 are clearly separated from zero within 68% errors , in comparison to fig . [ fig : ccor1 ] . although it is difficult to see , even the cases of @xmath120 mpc@xmath3 is also inconsistent with no correlation within 68% errors . the probability that the positive correlation appears is shown in the right panel . the cases of @xmath154 and @xmath172 mpc@xmath3 produce the correlation with more than 99 % at small angular scale . since stronger correlation is expected for the smaller @xmath118 , next generation uhecr experiments can observe positive correlation between uhecrs and their sources at small angular scale with more than 99 % probability if @xmath6 mpc@xmath3 . the recent pao data with energies above @xmath1 ev show the overdensity of the arrival directions in the direction of cen a and that 2 events positionally correlate with the nucleus position of cen a @xcite . on the other hand , cen a has been expected as nearby production site of uhecrs @xcite . in this section , we specially focus on cen a , and discuss the possibility that cen a is the nearest uhecr source from the viewpoint of the arrival distribution of uhecrs . for this purpose , we use two source models which are slightly modified versions of the source model introduced in section [ method_ad ] . in the first model , we artificially neglect sources within 5 mpc from the original source model and add cen a. in this source model , all the source distributions include cen a as the nearest uhecr source . we assume that the positions and distance of cen a are ( @xmath176 , @xmath177 ) @xmath87 ( @xmath178 , @xmath179 ) and 4.1 mpc following parameters listed in the iras pscz catalog @xcite . the second source model is for reference , i.e. , we artificially neglect sources within 5 mpc from the original model but do not add cen a. comparing the results from these two source models , we can investigate the effect of cen a to uhecr arrival distribution . fig . [ fig : map ] demonstrates the arrival distributions of uhecrs above @xmath1 ev from a specific source distribution including cen a with @xmath154 mpc@xmath3 ( _ lower right _ ) . this number density is comparable with the local number density of fr i galaxies , which cen a is classified into @xcite . the source distribution is shown as circles in the galactic coordinates . the radii of the circles are inversely proportional to the distances of sources . although we use all the sources up to 500 mpc in the calculations , only sources within 75 mpc are plotted as circles for visibility . the position of cen a can be identified as the center of the largest circle . the simulated events are 69 under the pao aperture ( _ upper left _ ) and 1000 under the uniform aperture ( _ lower left _ ) . the colors represent the composition of uhecrs ; protons ( _ black _ ) , helium ( _ blue _ ) , light nuclei with @xmath180 ( _ light blue _ ) , intermediate mass nuclei with @xmath181 ( _ green _ ) , and heavy nuclei with @xmath182 ( _ red _ ) , where @xmath183 is the nuclear number . even the 69 events remind you that a significant fraction of the whole events is distributed around cen a. in this case , about 30% of uhecrs originates from cen a. nevertheless , the arrival directions of the uhecrs are scattered by gmf ( see also ref . @xcite ) . several black points are secondary protons . since protons are minimally affected by the gmf , they can point out their sources . three protons correlate with the position of cen a very well . if cen a is a true source , it is possible that the two pao events correlating with the position of cen a are protons , because several plausible gmf models predict very small deflection angles ( @xmath184 ) in the direction of cen a @xcite . note that secondary protons above @xmath1 ev require the maximum acceleration energy of irons at sources above @xmath185 ev . the reproducibility of the auto - correlation function of the observed data by the source model including cen a should be checked . [ fig : cenaornot1 ] shows the cumulative auto - correlation functions of the arrival directions of 69 uhecrs above @xmath1 ev simulated on the assumption of the pao aperture . the source models with @xmath154 mpc@xmath3 explained above are applied . the error bars represent 68% ( _ thick _ ) and 95% ( _ thin _ ) errors . both the gmf and egmf ( @xmath152 ng ) are taken into account . the cumulative auto - correlation function of the pao events is also shown for reference . this figure shows that the predictions of both source models are consistent with the recent pao data within 95% errors , even in the case that cen a is the nearest uhecr sources . on the other hand , the arrival distribution of the recent pao data is also consistent with isotropic distribution above @xmath1 ev at 95% confidence level @xcite . thus , both possibilities , cen a is a source or isotropic , are allowed at present . note that the meaning of the 95% in the pao analysis is different from that of the 95% in this analysis . the observed arrival distribution is consistent with the predictions of our source model at 95% confidence level in this study , while the distribution is consistent with isotropic distribution at 95% confidence level in the analysis of the pao . ev simulated under the pao aperture . the source models including ( _ red _ ) and neglecting ( _ green _ ) cen a with @xmath154 mpc@xmath3 are used . both egmf ( @xmath152 ng ) and gmf are taken into account . the error bars represent 68% ( _ thick _ ) and 95% ( _ thin _ ) errors . the cumulative auto - correlation function calculated from the recent published pao data @xcite is also shown for comparison . ( _ black _ ) . ] the pao data also reported significantly positive excess of uhecrs in the arrival distribution around the nucleus position of cen a @xcite . this excess gives information to test whether cen a significantly contributes to the total uhecr flux or not . [ fig : cenadist ] shows the normalized cumulative numbers of events used in fig . [ fig : cenaornot1 ] . the average of the cumulative numbers is represented as _ blue _ lines with 68% ( _ green _ ) and 95% ( _ light blue _ ) errors . the event distribution of the 69 pao events ( _ black _ ) and that of isotropic events ( _ red _ ) are also shown for comparison . the source model without cen a can well reproduce the event distribution of the pao events above @xmath186 within 95% error bars ( _ left _ ) . the prediction is also consistent with the distribution of isotropic events . however , this source model can not reproduce the excess around cen a despite that our source model includes the overdensity of the centaurus region . on the other hand , the source model including cen a reproduces the excess around cen a ( _ right _ ) . although the predicted distribution can not reproduce the observed event distribution at large angular scales , it indicates that the contribution of cen a is important to reproduce this excess . in order to reproduce this observed distribution over all the angular scale , the contribution of cen a has to be reduced or the deficit of events should exist at large angular scale from cen a. possible solutions of the former idea are ( i ) uhecr flux of cen a is weaker than that of the other fr i galaxies , ( ii ) there is an additional source population with @xmath171 mpc@xmath3 like seyfert galaxies @xcite and the relative contribution of cen a is reduced , ( iii ) gmf can work more efficiently than what we consider in this study . for the latter idea a possibility is cen a mainly contributes to uhecrs in the whole sky ( at least southern sky ) but the uhecrs are not completely isotropized with keeping consistency with the observed isotropy mentioned above . in any case , scans over large parameter space are needed , which is beyond the scope of this paper . here , we only claim that the absence of cen a can not reproduce the observed excess . the increasing number of events reveals anisotropy by nearby sources more clearly , as seen in the lower left panel of fig . [ fig : map ] . if cen a significantly contributes to the total flux of uhecrs , anisotropy in the arrival distribution at angular scale of @xmath187 clearly appears and the order of tens secondary protons correlates with the nucleus position of cen a. in this case , @xmath188 of the events originates from cen a. in addition , a significant fraction of secondary protons positionally correlates with nearby sources other than cen a. these secondary protons are an indicator of nearby sources even if only irons are accelerated at sources . [ fig : cenaornot2 ] shows the cumulative auto - correlation functions ( _ left _ ) and cumulative cross - correlation functions ( _ right _ ) of 1000 events above @xmath1 ev simulated on the assumption of the uniform aperture . the source models including cen a ( _ red _ ) and neglecting cen a ( _ green _ ) with @xmath154 mpc @xmath3 are applied , and both egmf ( @xmath152 ng ) and gmf are taken into account . the two source models are well separated in the cumulative auto - correlation functions , i.e. , anisotropy in events detected in the future will be a good clue to understand the contribution of nearby uhecr sources . on the other hand , although the two source models are not separated in the cumulative cross - correlation functions as good as in the auto - correlation functions , both the cases predict significantly positive correlation between uhecrs and their sources at small angular scale , clearly . even if primary uhecrs are purely irons , information on the location of nearby sources can be extracted from the future uhecr data . we have investigated the anisotropy in the arrival directions of uhecr nuclei and their positional correlations with source objects , on the basis of propagation calculations conducted under various assumptions for the source properties and cosmic magnetic fields . although @xmath20 ev was fiducially assumed in the previous section , the results depend somewhat on @xmath189 . since only iron nuclei above @xmath185 ev can produce secondary protons above @xmath1 ev ( @xmath190 [ @xmath191 ev ] ev ) , the threshold energy for the published pao events and also adopted in this study , secondary protons do not contribute to the total flux of uhecrs above it if @xmath192 ev . [ fig : acoremax ] shows the probability curves as discussed in fig . [ fig : cprob_auto1000 ] , but for @xmath193 , @xmath194 , and @xmath185 ev , in order to check the dependence on @xmath189 . the probability for the appearance of a positive excess of events is the highest for @xmath20 ev at small angular scales , and the angular scale of the peak is smaller for higher @xmath189 . these reflect the fact that secondary protons contribute only in the case of @xmath20 ev . furthermore , the probability is slightly higher at small angular scales for @xmath195 ev than for @xmath193 ev because light and/or high energy nuclei are included above the threshold in the former case . therefore , positive anisotropy at small angular scales is weakened when @xmath192 ev , owing to the lack of secondary protons above threshold , but the probability at intermediate angular scales is still greater than 95% . we have considered a rather extreme case where only iron nuclei are emitted from sources . however , other nuclei and/or protons may also plausibly constitute the composition of uhecr sources ( e.g. , a mixed composition model @xcite or a wolf - rayet star model @xcite ) . the other elements can affect the spectral shape as mentioned in section [ introduction ] due to different energy losses and magnetic deflections . although the source composition is highly uncertain at the moment , further studies of anisotropy and correlations under more realistic assumptions for the source composition are in order . for instance , assuming a mixed composition and a simple rigidity scaling for the maximum energies , protons can produce anisotropy at low energies comparable to or stronger than the anisotropy due to heavy nuclei at the highest energies , because a proton with energies @xmath183 times smaller than an iron nucleus propagates along the same trajectory @xcite . thus , if cen a is an uhecr source , anisotropy at low energies may also be expected . however , pao has reported the lack of significant anisotropy at low energies toward the direction of the excess highest energy events @xcite . this result implies that 1 ) a significant fraction of uhecrs may be nuclei even at lower energies , 2 ) the anisotropy at the highest energies is actually weaker than reported , 3 ) the composition at high energies is not as heavy nuclei - dominant as indicated from the xmax measurements , or some combination of these three possibilities . furthermore , strong magnetic fields could modify the theoretical expectation through the extension of the propagation distance as mentioned in @xcite , so magnetic fields may also be important in this context . although we have adopted simple models for the gmf and egmf to avoid the uncertainties with their structure , the details of these fields can actually play important roles in the propagation of uhecrs and their arrival directions at earth , as has been well studied for uhe protons @xcite . in particular , as discussed in ref . @xcite for protons , the structures actually observed near the galaxy are also relevant when searching for the evidence of nearby uhecr sources . whether the positional information in the sky is lost or not during propagation depends on the egmf strength and structures in local universe . the effect of the gmf is inevitable , but the deflection angles and directions of uhecrs strongly depend on the region of the sky , reflecting the structure of the gmf , as studied in refs . @xcite for iron nuclei . thus more realistic gmf models should be applied for correlation studies , although such gmf models also have significant uncertainties . the dependence of the anisotropy signals on different models for the gmf and egmf will be investigated in future work . in addition to cen a , m87 is also an interesting object as an uhecr source candidate . since it is more distant than cen a ( @xmath196 mpc ) , the contribution of m87 to the anisotropy is smaller , depending on the egmf strength . if the power of uhecr injection from m87 is comparable to that of typical fr i galaxies , it is not expected to contribute strongly to the anisotropy . however , if its power is an order of magnitude stronger or @xmath118 is smaller , anisotropy can appear in the direction of m87 even for @xmath152 ng . secondary protons should also be a good source indicator in this case . this study has assumed the sources of uhecrs to be steady . on the other hand , it is also very much possible that transient phenomena like grbs or agn flares produce uhecrs . transient scenarios are especially relevant for accelerating protons up to @xmath8 ev in outflows because not many steady sources satisfy the luminosity requirement for uhecr acceleration @xcite . for transient sources , the arrival distribution of uhecrs will be different from that in the case of steady source scenarios . the time delay of uhecrs by the gmf and egmfs leads to apparent durations of uhecr bursts longer than their intrinsic duration . if the apparent duration is longer than the observation timescale of human beings , we would misperceive an uhecr burst as a steady source @xcite . however , since the arrival time of uhecrs depends on the uhecr energies , features of anisotropy may be strongly energy - dependent @xcite . this will be evidence for the transient generation of uhecrs . even in transient scenarios , anisotropy of ( secondary ) protons at small angular scale should still be a strong hint of uhecr sources . to summarize , we have calculated the propagation of uhe nuclei above @xmath1 ev , taking into account magnetic fields in the universe on the assumption that only iron nuclei are injected by steady sources , and examined the anisotropy in the distribution of uhecr arrival directions and positional correlation between uhecrs and their sources in local universe . we found that both the anisotropy and correlation are dominantly suppressed by the gmf rather than by egmfs . assuming the current pao status ( 69 events above @xmath1 ev ) , the arrival distribution of uhecrs is consistent with isotropy within 95% errors for @xmath2 mpc@xmath3 and is consistent with no correlation within 95% errors for @xmath4 mpc@xmath3 when the gmf and egmf are taken into account . we also discussed whether future experiments improve these situation . 1000 events above @xmath1 ev in the whole sky would reveal positive anisotropy with more than 99% probability at intermediate angular scales even for @xmath197 mpc@xmath3 , and correlation between uhecrs and their sources in local universe would appear with more than 99% at small angular scales for @xmath6 mpc@xmath3 . in addition , we found that the contribution of cen a is required to reproduce the observed uhecr excess around the centaurus region . secondary protons from primary heavy nuclei are working positively to produce anisotropy at small angular scales , which would provide a strong hint of the source location , depending on the maximum energy at the uhecr source . we thank d. allard , g. medina - tanco , and f. takahara for useful comments . this work is supported by grant - in - aid no . 22540278 ( si ) from the ministry of education , culture , sports , science and technology of japan . 121 natexlab#1#1[2]#2 , ( ) . , , ( ) ( ) . , , ( ) . , , , ( ) . , ( ) . , , , ( ) . , , ( ) . , , , , available from arxiv:1107.5576 , ( ) . , ( ) . , , , , ( ) . , , , , ( ) . , , , ( ) . , ( ) . , , , ( ) . , ( ) . , , , ( ) . , , , ( ) . , , , , available from astro - ph/0701167 , ( ) . , , ( ) . , ( ) . , , ( ) . , , , , ( ) . , , , ( ) . , , , , , , , , , , , ( ) . , et al . , ( ) . , et al . , ( ) . , et al . , ( ) . , et al . , ( ) . , et al . , ( ) . , , ( ) . , , , , , ( ) . , , , , , ( ) . , , , , ( ) . , , , , ( ) . , , ( ) . , , ( ) . , , , , ( ) . , , , , , ( ) . , , , ( ) . , , , , , ( ) . , , , ( ) . , , ( ) . , , , , , ( ) . , , , , , , , , ( ) . , , , ( ) . , , , ( ) . , , , ( ) . , , , , , ( ) . , , , , , ( ) . , , , ( ) . , , , ( ) . , , , available from arxiv:0802.4452 , , , , , ( ) . , , , available from arxiv:1006.2484 , , , , ( ) . , , , , , , , , , , , ( ) . , , , , , ( ) . , , , , ( ) . , , , ( ) . , , , ( ) . , , , , ( ) . , , , , ( ) . , , ( ) . , , , ( ) . , , , , ( ) . , , , ( ) . , , , ( ) . , , , , ( ) . , , , , ( ) . , , , , , , ( ) . , , , , ( ) . , , available from arxiv:1110.3245 , ( ) . , , , , ( ) . , , , , ( ) . , , , , ( ) . , , ( ) . , , , , , ( ) . , , , ( ) . , et al . , ( ) . , , , ( ) . , , ( ) , , , , , ( ) . , , ( ) . , ( ) . , , , ( ) . , , , ( ) . , , ( ) . , , , , ( eds . ) , , . , , , ( ) . , , ( ) . , ( ) . , , , , , , , , , , , available from arxiv:1009.1855 , , ( ) . , in : proc . 30th icrc ( melida ) , 2007 , 10451048 , ( ) . , , , , ( ) . , , , , ( ) . , , , ( ) . , , ( ) . , , , ( ) . , , ( ) . , , , , , , , , , , , ( ) . , ( ) . , , ( ) .

we numerically construct the bps and non - bps wavefunctions of an quiver quantum mechanics with two abelian nodes and a single arrow . this model captures the dynamics of a pair of wrapped d - branes interacting via a single light string mode . a dimensionless parameter , which is inversely proportional to the the fayet - iliopoulos parameter , controls whether the bulk of the wavefunctions are supported on the higgs branch or the coulomb branch . we demonstrate how the bps and excited states morph as is tuned . we also numerically compute the energy gap between the ground state and the first excited states as a function of . an expression for the the gap , computed on the coulomb branch , matches nicely with our numerics at large but deviates at small where the higgs branch becomes the relevant description of the physics . in the appendix , we provide the schrdinger equations fully reduced via symmetries which , in principle , allow for the numerical determination of the entire spectrum at any point in moduli space . for the ground states , this numerical determination of the spectrum can be thought of as the first _ in silico _ check of various witten index calculations . mit - ctp-4731 susy in silico : numerical d - brane bound state spectroscopy tarek anous _ center for theoretical physics , massachusetts institute of technology , cambridge , ma 02139 , usa _

1511.01450

the low - energy effective theory describing the dynamics of a collection of wrapped d - branes at small separations is given by quiver quantum mechanics @xcite . these d - branes form bound states and , depending on parameters , can either be well separated or on top of each other . when well separated , the theory is on the coulomb branch , parametrized by a macroscopic vev @xmath2 for the fields describing the separation between the branes . conversely , when the branes are on top of each other , the theory is on the higgs branch , parametrized by a large vev @xmath3 for the field representing the light stretched string mode between the branes . in this paper we will consider the simplest example of quiver quantum mechanics , describing the dynamics of a pair of d - branes with a single light stretched string mode between them . being a supersymmetric theory , the relative separation between the branes lives in a vector multiplet @xmath4 and the lightest stretched string mode lives in a chiral multiplet @xmath5 . these models have proven fruitful playgrounds for the study of supersymmetric black hole bound states at weak coupling @xcite . however , one obvious question remains unsolved : what is the structure of the bound state wavefunctions ? while supersymmetry allows us to make some statements about the structure of the ground states , the full ground state wavefunctions remain unknown @xcite . this is because , while simpler than the full schrdinger equation , the bps equations remain too difficult to solve analytically . the situation is worse for the excited states , where we have to abandon the crutch of supersymmetry altogether . it is our interest in this paper to construct the bps and excited states of this model . we will not be able to do so analytically , but there exist numerical methods to compute the eigenspectra of differential operators on a finite domain , see e.g. @xcite . using these techniques we numerically solve the schrdinger equation , plot bound state wavefunctions , and determine their energies . in our analysis , we find that the physics is governed by a dimensionless quantity @xmath1 defined in ( [ nu ] ) , which is inversely proportional to the fayet - iliopoulos parameter @xmath6 . the quantity @xmath1 dictates whether the higgs branch or the coulomb branch is the dominant description of the dynamics . in studying the wavefunctions and their dependence on @xmath1 , we uncover that the two branches never quite decouple and the wavefunction has nonzero support in both branches for any value of @xmath1 . since this model is supersymmetric , much work has gone into studying features of the ground states . for example , the ground state wavefunction was previously studied in a born - oppenheimer approximation @xcite , which involves splitting the hamiltonian @xmath7 into a vector part @xmath8 and a chiral part @xmath9 and putting the chiral degrees of freedom in the harmonic oscillator ground state of @xmath9 . one then ` integrates out ' the chiral multiplet and solves the effective supersymmetric quantum mechanics on the vector multiplet degrees of freedom . this approximation is self - consistent if we are on the coulomb branch , that is if the ground state is such that the branes are well separated , and is only valid for large values of @xmath10 . one easy way to see this is that integrating out the chiral degrees of freedom generates a nontrivial metric on the moduli space of the vector multiplet , which appears in the effective lagrangian as @xmath11 the moduli space metric is singular at the origin where the approximation breaks down . the structure of the ground state wavefunction near the origin has previously been out of reach , and herein we fill this gap . interestingly the effective quantum mechanics on the coulomb branch of this model has an enhanced symmetry , allowing for a full determination of the spectrum on the coulomb branch @xcite . in computing excited state wavefunctions we numerically compute the energy gap between the supersymmetric ground state and the first excited state as a function of @xmath1 . we verify that the gap approaches the analytically determined value on the coulomb branch as we increase @xmath1 . recently @xcite the witten index @xmath12 , twisted by a combination of global symmetries , has been computed for various quivers ; and for the model of interest in this paper , @xmath13 ( depending on the sign of the fayet - iliopoulos parameter @xmath6 ) . we indeed find a single bosonic ground state in the correct representation of the global symmetry , meaning our result can be thought of as the first _ in silico _ experimental confirmation of the mathematically predicted bps spectrum of the model . moreover , using the methods described in this note we have access to more than the count of bps states . this includes plots of the ground state and excited state wavefunctions , the gap with the first excited state as well as various field expectation values . the numerical methods we describe will hopefully make the study of the quantum nature of weakly coupled black hole bound states more tractable , and can potentially provide a useful experimental testing ground for other witten index calculations . the organization of this paper is as follows : in section [ setup ] we present the supercharges and symmetry generators of the theory . these fix the wavefunctions up to their dependence on radial variables . we also give a quick review of the stringy interpretation of this model . in section [ groundstatesec ] we provide the bps equations obeyed by the ground state wavefunction , suitably reduced via symmetries . we also plot numerical solutions ( bps and non - bps ) for the full schrdinger problem in the singlet sector of the bosonic symmetry group . we study the dependence of the bps and non - bps wavefunctions on @xmath1 . we conclude in section [ outlook ] . we have collected formulae in appendices [ sphereapp ] and [ wavefunceqs ] and review the born - oppenheimer approximate ground state wavefunction in appendix [ boapp ] . in this paper we focus on the @xmath0 supersymmetric quantum mechanics defined by the following four supercharges : @xmath14_\alpha^{~\,\gamma}\left(\bar{\psi}\epsilon\right)_\gamma+i\left[\nabla_{\vec{\mathbf{x}}}\cdot\vec{\boldsymbol{\sigma}}-\left(|\phi|^2+\theta\right)\mathds{1}\right]_\alpha^{~\,\gamma}\lambda_\gamma~,\label{q}\\ \bar{q}^\beta&\equiv-\sqrt{2}\left(\epsilon\psi\right)^\gamma\left[\mathds{1}\,\partial_{\phi}+\bar{\phi}\ , \vec{\mathbf{x}}\cdot\vec{\boldsymbol{\sigma}}\right]_\gamma^{~\,\beta}+i\,\bar{\lambda}^\gamma\left[\nabla_{\vec{\mathbf{x}}}\cdot\vec{\boldsymbol{\sigma}}+\left(|\phi|^2+\theta\right)\mathds{1}\right]_\gamma^{~\,\beta}~.\label{qb } \ ] ] we will discuss the stringy origin of this quantum mechanics in section [ stringy ] , and refer the reader to @xcite for a more in depth treatment . the lagrangian of the model can also be found in @xcite . in the above expressions , the @xmath15 are the usual pauli matrices and @xmath16 . this quantum mechanics arises as the dimensional reduction of a @xmath17 , @xmath18 , @xmath19 gauge theory coupled to chiral matter@xcite . the matter content therefore includes a vector multiplet @xmath4 and a chiral multiplet @xmath5 , along with its complex conjugate . nor @xmath20 appear in the supercharges . this is because they are replaced by their on - shell values , e.g. @xmath21 and @xmath22 , when going to the operator formalism . the model under consideration has no superpotential and hence @xmath23 . ] the constant @xmath6 is a fayet - iliopoulos parameter . the triplet @xmath24 transforms as a vector under @xmath25 rotations . the @xmath26 @xmath19 gauge connection @xmath27 is non - dynamical and thus neither appears in the supercharges nor in the hamiltonian . however , as is manifestly the case in the lagrangian formulation of this theory , the chiral fields are charged under @xmath27 , and therefore @xmath27 generates a gauss - law constraint , which will appear in the algebra below . the two - component chiral fermion @xmath28 and gaugino @xmath29 obey the canonical quantization conditions : @xmath30 we have borrowed the notation of @xcite where spinors with an index down transform in the @xmath31 of @xmath25 and spinors with an index up transform in the @xmath32 . indices are raised and lowered using the levi - civita symbol @xmath33 with @xmath34 . thus in our conventions : @xmath35 whenever spinor indices are suppressed it indicates that they are summed over . the supercharges ( [ q]-[qb ] ) satisfy the supersymmetry algebra @xmath36 with hamiltonian @xmath37 the operator @xmath38 generates the gauss - law constraint , and must vanish on gauge - invariant states . appears in the supersymmetry algebra with the normal ordering constant @xmath39 , which is not present in the classical generator . ] thus , the proper supersymmetric states of the theory satisfy the following bps equations @xmath40 let us briefly list the dimensions of the parameters and fields in units of the energy @xmath41 $ ] . they are @xmath42=-1/2 $ ] , @xmath43=1 $ ] , @xmath44=0 $ ] , @xmath45={3/2}$ ] , @xmath46=-3 $ ] and @xmath47=-1 $ ] , where the right hand sides of these expressions are shorthand for powers of @xmath41 $ ] . an important role will be played by the dimensionless quantity @xmath48 which we note here for later use . the symmetry group of this theory is @xmath49 @xcite and the states fall into representations of this symmetry group . we will use this to our advantage . the generator of @xmath50 is the angular momentum operator : @xmath51 with @xmath52 the usual orbital angular momentum , which is not independently conserved . the components of @xmath53 satisfy the following algebra : @xmath54=i\,\epsilon_{ijk}\,\mathbf{j}^k~ , & & \left[\mathbf{j}^i , q_\alpha\right]=-\frac{1}{2}\boldsymbol{\sigma}^{i~\gamma}_\alpha q_\gamma~,\nonumber\\ & \left[\vec{\mathbf{j}}^2,\mathbf{j}^i\right]=0~ , & & \left[\mathbf{j}^i,\bar{q}^\alpha\right]=\frac{1}{2}\bar{q}^\beta\,\boldsymbol{\sigma}^{i~\alpha}_\beta ~.\label{jalg}\end{aligned}\ ] ] since @xmath55 , it is easy to verify that ( [ jalg ] ) implies @xmath56=0 $ ] as promised . the @xmath57 generator identitcal chiral multiplets labeled by the index @xmath58 then the @xmath59-charge operator would take the form @xmath60with the @xmath61 an arbitrary set of numbers . since herein we are considering the case @xmath62 , the term proportional to @xmath63 is simply @xmath64 and shifts the @xmath59-charge operator by the constant @xmath63 on gauge - invariant states , hence we set @xmath65 . ] @xmath66 satisfies @xmath67=-q_\alpha~,\quad\quad\quad\quad\left[r,\bar{q}^\alpha\right]=+\bar{q}^\alpha~,\quad\quad\quad\quad\left[r,\mathbf{j}^i\right]=0~.\label{ralg}\ ] ] again we may deduce that @xmath68=0 $ ] . notice that the conserved @xmath59-charge counts the difference in the number of gauginos and chiral fermions , while the fermion number @xmath69 is not conserved as a result of the mixing term in ( [ ham ] ) . the hamiltonian is also invariant under the following discrete symmetry : @xmath70 which takes @xmath71 . as a result the spectrum is degenerate under the interchange @xmath72 , and we verify this explicitly in appendix [ wavefunceqs ] . this simple quantum mechanical model can be understood as the dimensional reduction of the worldvolume gauge theory living on a pair of wrapped d - branes in a calabi - yau compactification down to four dimensions . we consider here only the relative gauge group between the branes and ignore the `` center of mass '' degrees of freedom as they play no role in the study of the bound states . the vector muliplet scalars @xmath10 parametrize the separation between the d - branes in the non - compact space . furthermore , in our particular setup , the two branes intersect at a single point in the calabi - yau , where there exists a stretched string between them @xcite . the lightest string excitation is parametrized by the bifundamental charged chiral multiplet and the chiral scalar mass can be read off from the classical potential @xmath73 that is @xmath74 . the quantity @xmath75 should be understood as the difference between the phases of the complexified masses of the d - branes in the language of the calabi - yau compactification . thus , having neglected the tower of higher string modes , this model is only valid as an approximation to the d - brane system when @xmath76 and @xmath77 where @xmath78 is the the string length ( see @xcite for an in - depth discussion as well as @xcite for earlier references ) . an important role is played by the fayet - iliopoulos parameter @xmath6 , which determines whether or not supersymmetry is broken , as is evidenced in ( [ potential ] ) . for @xmath79 there are two types of classical minima : the higgs branch with @xmath80 ( modulo @xmath19 gauge transformations ) , which preserves supersymmetry ; and the coulomb branch with @xmath81 , where supersymmetry is classically broken . the potential has a flat direction when @xmath82 for which @xmath83 . thus for @xmath79 the model is expected to have bound states with energies below this threshold value of @xmath84 . for @xmath85 there is no higgs branch and the coulomb branch is metastable for all values of @xmath86 . this picture changes drastically ( for @xmath79 ) when quantum effects are taken into consideration . if we are interested in the large distance behavior of the wavefunctions of this model , that is @xmath87 , then the chiral multiplet can be integrated out ( see appendix [ boapp ] ) generating an effective potential on the vector multiplet @xmath88 the new effective potential has a zero energy minimum at @xmath89 and supersymmetry is restored on the coulomb branch . for the coulomb branch description to be consistent we thus need @xmath90 or , restated in the language of the dimensionless parameter @xmath1 given in ( [ nu ] ) : @xmath91 for @xmath92 the bps wavefunction is localized on the higgs branch and the coulomb branch description is no longer valid . for @xmath1 sufficiently large we expect the quiver description to break down and for supergravity to become the valid description @xcite . thus , in this model @xmath1 plays the role of the closed string coupling @xmath93 . since this model is a dimensionally reduced worldvolume gauge theory , the mass @xmath94 of the vector multiplet is related to the closed string coupling @xmath95 . it is our interest in this paper to study how the quantum states of this quiver model depend on @xmath1 . in quantum mechanics the higgs and coulomb branches are not fully separate , and the wavefunction will always have support in both directions . we aim to answer how the majority of the wavefunction s support moves from the @xmath96 direction to the @xmath86 direction as @xmath1 is increased . in order to study this we resort to numerically solving the full schrdinger equation and obtain supersymmetric and non - supersymmetric wavefunctions . this also serves as a numerical check of known supersymmetric index calculations @xcite . in doing so we also gain insight into the excited state spectrum of the theory and show how the gap with the first excited state depends on @xmath1 . we wish to diagonalize the maximal set of mutually commuting operators : @xmath97 and @xmath98 . before doing so , it is convenient to introduce the following parametrization for the chiral bosons : @xmath99 derivatives with respect to @xmath100 and @xmath101 are now given by @xmath102 and the gauss law generator becomes @xmath103 we will also find it convenient to work in spherical coordinates for the vector multiplet degrees of freedom , that is @xmath104 . details on how operators are represented in spherical coordinates are given in appendix [ sphereapp ] . let us first attempt to solve the bps equations ( [ bpseqs ] ) for the ground state wavefunctions of this model . usually this would involve applying the supercharges to a generic state and looking for the normalizable solutions . it is to our advantage , however , that the refined supersymmetric index @xcite @xmath105 has been computed for this system and evaluates to @xmath106 when @xmath79 . as @xmath6 goes from negative to positive , @xmath12 jumps to zero and provides the simplest example of wall crossing @xcite . since @xmath12 is independent of @xmath107 there exists a single bosonic ground state with @xmath108 and @xmath109 . the most general gauge - invariant , @xmath59-symmetric wavefunctions take the form :- charge sectors to appendix [ wavefunceqs ] . ] @xmath110 where @xmath111 , @xmath112 and @xmath20 are functions of @xmath113 . we now further restrict to an @xmath50 highest weight state , which satisfies @xmath114 and @xmath115 , yielding in @xmath111 and @xmath20 as well as the factor of @xmath116 in @xmath20 for later convenience . with this choice the schrdinger equation is manifestly real . ] @xmath117 & \tilde{e}_{12}+\cos\vartheta\,\tilde{e}_{22}\\&\tilde{e}_{21}+\cos\vartheta\,\tilde{e}_{22 } & e^{i\varphi}\sin\vartheta\,\tilde{e}_{22}\end{pmatrix}~,\end{aligned}\ ] ] where the @xmath118 are also functions of @xmath119 . the ground state has spin @xmath120 , meaning that @xmath121 also satisfies @xmath122 , which leads to the further simplification @xmath123 thus when @xmath120 , the matrix @xmath112 is simply @xmath124 demanding @xmath125 leads to the following set of coupled differential equations @xmath126 we can solve for @xmath127 and @xmath128 algebraically in favor of @xmath129 using ( [ e12eq1 ] ) and ( [ e12eq2 ] ) . explicitly @xmath130 where @xmath131 is the classical potential , as in ( [ potential ] ) . making these substitutions , we find that @xmath129 statisfies the following schrdinger equation:-[e11eq2 ] ) is overdetermined but not inconsistent . substitution of ( [ frel ] ) and ( [ e11rel ] ) into both ( [ e11eq1 ] ) and ( [ e11eq2 ] ) gives rise to the same equation ( [ hardschro ] ) . ] @xmath132 as we will see in ( [ j0ham ] ) , this is precisely what we get out of the full schrdinger equation , but here the bps equations have provided us with expressions for @xmath127 and @xmath128 in terms of @xmath129 , and therefore a single un - coupled partial differential equation . were we able to solve ( [ hardschro ] ) , we would have an analytic expression for the bps wavefunction . this has proven too difficult , so we instead solve for the ground state numerically . before moving on we note that ( [ hardschro ] ) is very much akin to the ` deprolongation ' of the bps equations noticed in @xcite , namely we can rewrite ( [ hardschro ] ) as @xmath133 where we have implicitly defined the differential operator @xmath134 and @xmath135 to construct the ground state wavefunction numerically , we will input the full schrdinger problem into the mathematica s ndeigensystem command . the coupled differential equations of interest ( see appendix [ r0eqsapp ] ) obtained from @xmath136 are : @xmath137 with @xmath138 as in ( [ laplaceop ] ) . we have defined @xmath139 . by virtue of the bps equations , @xmath140 in the ground state , but it does not necessarily vanish in the excited states . it turns out that @xmath141 satisfies the same wave equation as an @xmath142 state ( see for example ( [ hrm2 ] ) and ( [ hrp2 ] ) ) , hence states with vanishing @xmath141 have no superpartners in the @xmath142 sectors . the excited states with @xmath140 thus live in short multiplets , as described in appendix [ r0eqsapp ] . we also note here that the norm of the @xmath143 wavefunctions is given by @xmath144 however , we will have to restrict the domain of @xmath145 and @xmath146 when computing the eigenfunctions of ( [ j0ham ] ) numerically , and the wavefunctions will be normalized on this restricted domain . we are only interested in the case @xmath79 , since this is the phase with unbroken supersymmetry and relegate the study of @xmath6 positive to later work . the procedure is as follows , we input the differential operator ( [ j0ham ] ) into mathematica s ndeigensystem command , which uses a finite element approach to solve for the eigenfunctions of a coupled differential operator on a restricted domain . we vary the size of the domain and the refinement of the mesh until the numerics are stable . before moving on we pause here to discuss the the subtlety of diagnosing supersymmetry breaking in the presence of numerical errors . that is , it is naively quite difficult to tell the difference between an _ actual _ zero energy eigenstate and one whose energy is lifted by supersymmetry breaking effects , especially when the algorithm s resolution of the energy is @xmath147 ( as it can be in certain parameter regimes ) . this is where being able to numerically determine the excited state spectrum comes in handy . to diagnose if supersymmetry is broken , it suffices to check that the state of interest lives in a supersymmetric multiplet and to find its susy partners . if the state in question is a singlet under supersymmetry , then supersymmetry is unbroken . ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector with @xmath148 and @xmath149 , corresponding to @xmath150 . the minimum of the potential is at @xmath151 . the energies of the first three eigenstates are respectively @xmath152 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] the radial probability density of the first three eigenstates in the @xmath153 sector are shown in figure [ wfhiggs ] for a set of paratmeters such that @xmath154 . we also plot the components @xmath155 of the ground state wavefunction individually . let us summarize some results : 1 . the ground state wavefunction passes several consistency checks . note , for example , that it is peaked near the classical minimum @xmath156 . furthermore , the ground state is nodeless , unlike the excited states , and has energy @xmath157 within numerical error ( that is , for a quick enough run we find @xmath158 ) . 2 . to verify that we have actually found the supersymmetric ground state , it suffices to check that it is not part of a supersymmetric multiplet . that means we should check that it is not paired with any states in the @xmath159 sectors . this is indeed the case . furtheremore , we were able to numerically find all the supersymmetric partners of the first excited state for the parameters shown in figure [ wfhiggs ] . the multiplet consists of 8 states , 4 bosonic and 4 fermionic , which split into a bosonic @xmath153 singlet , two fermionic doublets with @xmath160 , and a bosonic triplet with @xmath161 . in the @xmath153 sector , the first two excited state indeed satisfy @xmath140 . generically in the ground state we find @xmath162 , to a very good approximation ( see figure [ expectationvals ] ) . this is true independent of whether @xmath1 is less than or greater than 1 . we fit the gap between the ground state and the first excited state and find that it is @xmath163 , for @xmath164 ( see figure [ expectationvals ] ) . notice that the components of the wavefunction are all on equal footing . that is , they are all more or less peaked at the classical minimum , save for @xmath128 which vanishes at @xmath165 and is peaked slightly off the axis . it is worth mentioning however that their qualititative features are similar . we will find that this changes significantly on the coulomb branch . the radial probability density for the first three eigenfunctions in the @xmath153 sector are shown in figure [ wfcoulomb ] for a set of paratmeters such that @xmath167 . we again summarize some results : ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] ( after integrating over the angular variables ) in the @xmath108 , @xmath120 sector for @xmath168 , @xmath169 corresponding to @xmath170 . the minimum of the potential is at @xmath171 . the minimum of @xmath172 is at @xmath173 and the chiral fields become tachyonic for @xmath174 . note the difference in scales between the @xmath145-axis and the @xmath96-axis , particularly for the excited states . the energies of the first three eigenstates are respectively @xmath175 . _ bottom row _ : components of the ground state wavefunction.,title="fig:",scaledwidth=32.0% ] 1 . the main difference between the case at hand where @xmath166 and the previous case where @xmath92 is that the wavefunction has considerable support on the coulomb branch , that is near @xmath82 and @xmath176 . recall that the mass of @xmath100 is given by @xmath177 , so @xmath178 is precisely where @xmath100 becomes tachyonic . the wavefunction s support on the coulomb branch grows as @xmath1 gets larger ( see figure [ probplot ] ) . notice how the ground state width in the @xmath96 direction increases as @xmath145 decreases . this can be understood from the born - oppenheimer analysis , where the wavefunction has a factor @xmath179 . for @xmath180 the higgs branch becomes important and the born - oppenheimer analysis breaks down . we see that the width of the wavefunction in the @xmath96 direction grows greatly at this point . 2 . the excited state wavefunctions have considerably less support on the higgs branch than the ground state , and significantly more support on the coulomb branch than the ground state . the component @xmath127 of the wavefunction only has support near @xmath165 . this is expected from the born - oppenheimer analysis given in appendix [ boapp ] . both @xmath20 and @xmath111 in ( [ r0wf ] ) can be understood as excited states of the chiral hamiltonian @xmath181 in ( [ hamdecomp1 ] ) . the exact energy spectrum on the coulomb branch was calculated in @xcite and is given by @xmath182 for @xmath183 . we confirm this by noting that the gap between the ground state and the first excited state in our numerically obtained spectrum fits well with @xmath184 for @xmath1 sufficiently large ( see figure [ expectationvals ] ) . as @xmath1 increases the gap with the first excited state approaches zero . 5 . in keeping with the born - oppenheimer analysis , we find that to a good approximation @xmath185 for @xmath1 sufficiently large . again @xmath186 . as a function of @xmath1 . we conclude that @xmath186 in both the higgs and coulomb branch . _ top right _ : expectation value of @xmath187 as a function of @xmath1 . as @xmath1 increases the approximate coulomb branch formula @xmath188 becomes more exact . _ bottom row _ : gap between the ground state and the first excited state as a function of @xmath1 . as @xmath1 increases we go from a regime where the gap is given by @xmath163 to a regime where the gap is @xmath189.,title="fig:",scaledwidth=48.0% ] as a function of @xmath1 . we conclude that @xmath186 in both the higgs and coulomb branch . _ top right _ : expectation value of @xmath187 as a function of @xmath1 . as @xmath1 increases the approximate coulomb branch formula @xmath188 becomes more exact . _ bottom row _ : gap between the ground state and the first excited state as a function of @xmath1 . as @xmath1 increases we go from a regime where the gap is given by @xmath163 to a regime where the gap is @xmath189.,title="fig:",scaledwidth=48.0% ] as a function of @xmath1 . we conclude that @xmath186 in both the higgs and coulomb branch . _ top right _ : expectation value of @xmath187 as a function of @xmath1 . as @xmath1 increases the approximate coulomb branch formula @xmath188 becomes more exact . _ bottom row _ : gap between the ground state and the first excited state as a function of @xmath1 . as @xmath1 increases we go from a regime where the gap is given by @xmath163 to a regime where the gap is @xmath189.,title="fig:",scaledwidth=48.0% ] as a function of @xmath1 . we conclude that @xmath186 in both the higgs and coulomb branch . _ top right _ : expectation value of @xmath187 as a function of @xmath1 . as @xmath1 increases the approximate coulomb branch formula @xmath188 becomes more exact . _ bottom row _ : gap between the ground state and the first excited state as a function of @xmath1 . as @xmath1 increases we go from a regime where the gap is given by @xmath163 to a regime where the gap is @xmath189.,title="fig:",scaledwidth=48.0% ] we can now see that that the the ground state wavefunction morphs smoothly from having the bulk of its support on the higgs branch to being supported mostly on the coulomb branch as we increase @xmath1 . this is accompanied by an increasing vev for @xmath190 and a decreasing vev for @xmath191 . since the ground state is nodeless , it must be the case that as @xmath1 is tuned larger ( @xmath192 tuned smaller ) , the peak of the wavefunction moves towards the origin and the tails spread out in the l - shape seen in figure [ wfcoulomb ] . this smooth metamorphosis is accompanied by the bulk of the probability being localized at large values of @xmath145 . to make this quantitative , we compute the integrated probability @xmath193 where @xmath194 is the naive value of @xmath145 where the higgs branch becomes important , i.e. when the @xmath100 field becomes tachyonic . the function @xmath195 is a diagonostic which tells us what fraction of the ground state wavefunction has support on the coulomb branch . we have plotted it in figure [ probplot ] . . , scaledwidth=52.0% ] as expected @xmath195 increases monotonically as a function of @xmath1 . for @xmath1 sufficiently large , the born - oppenheimer analysis , and the wavefunction given in appendix [ boapp ] capture the bulk of the large @xmath145 physics . in particular the energy gap and the expectation of @xmath190 match quite precisely with the coulomb branch formulae for @xmath196 , as shown in figure [ expectationvals ] . it is also of note that tuning the fayet - illiopoulos parameter near the wall of marginal stability at @xmath197 forces us into the large @xmath1 regime . hence to study wall crossing , one need only focus on the coulomb branch @xcite . of course , the higgs branch never fully decouples and the wavefunction will always have support there , it simply becomes less relevant as @xmath1 gets large as evidenced by the decreasing vev of @xmath198 . for @xmath199 , there is no such decoupling , and we need to consider all degrees of freedom of the model . this case can be very rich . in more complicated quiver models for example , where there can be multiple vector multiplets and a large number of chiral multiplets in play , one can consider tuning parameters such that some subset of chiral multiplets and vector multiplets are relevant . it was advocated in @xcite that such a mixed higgs - coulomb problem would be a toy model for studying a d - particle falling into a black hole . in that case , where we have little analytic control , numerics seems to be one avenue toward approaching this problem . this seems exciting , as it would provide a numerical avenue towards testing the fuzzball proposal as advocated in @xcite . in this brief note we numerically constructed the bound state wavefunctions of a simple d - brane quiver quantum mechanics . in doing so we demonstrated how the ground state and excited states shuffle around as parameters are tuned . we have also given a numerical - experimental verification of the witten index computed for this model @xcite . this is an example of what may someday be commonplace an _ in silico _ approach to studying supersymmetric quantum systems . apart from numerically verifying complicated witten index computations , one can also imagine , for example , revisiting old problems related to d0-brane matrix quantum mechanics @xcite such as the structure of the d0-brane bound state at small distance for fixed @xmath200 or a nonabelian quiver quantum mechanics at fixed but nonzero @xmath200 . these systems have threshold bound states in the absence of fayet - iliopoulos parameters and we hope to report on whether numerics are useful in this scenario in a subsequent publication . another avenue is to study the states of mass deformed quiver theories recently described in @xcite , which represent the effective dynamics of wrapped branes in ads . these questions are within reach using numerics . another interesting future direction is to study the entanglement structure between the chiral and vector degrees of freedom . typically when studying these models we have recourse to the born - oppenheimer approximation in which the chiral degrees of freedom are integrated out . as we mentioned , this integrating out generates a metric on the moduli space of the coulomb branch and it would be interesting to understand if this metric can be linked to the entanglement between the degrees of freedom . left untouched of course is what happens for quiver models with more fields . it has been shown for example @xcite that a quiver with three nodes and a closed loop has an exponential number of ground states ( in the number of arrows ) . it was then argued in @xcite that these states are ` pure - higgs ' , in the sense that they have no analogue on the coulomb branch . this seems to be in some tension with the notion that the ground states should morph smoothly as parameters are tuned , and that the wavefunction always has support in both the higgs and coulomb branch directions . one can envision studying the three - node quiver numerically in order to get a handle of the relation between higgs branch and coulomb branch states in more complicated models . it is a pleasure to thank dionysios anninos , arjun bagchi , nikolai bobev , frederik denef , daniel z. freedman , jaehoon lee , hong liu , daniel mayerson , antonin rovai , edgar shaghoulian , shu - heng shao , julian sonner , yifan wang and bert vercnock for helpful discussions . furthermore i am indebted to matt headrick s grassman.m package . i am also indebted to greg moore for sharing some of his unpublished notes on the subject . i would like to especially thank dionysios anninos and frederik denef for stimulating discussions and comments on the draft . this work is supported by the u.s . department of energy under grant contract number de - sc0012567 and by nsf grant phy-0967299 . it will be most convenient to work in spherical coordinates @xmath201 in which the matrix @xmath202 furthermore we use the identity @xmath203 to write @xmath204\\ & = \frac{1}{r}\left(\vec{\mathbf{x}}\cdot\vec{\boldsymbol{\sigma}}_\alpha^{~\,\omega}\right)\left[\delta_\omega^{~\,\gamma}\partial_r-\frac{1}{r}\,\vec{\mathbf{l}}\cdot\vec{\boldsymbol{\sigma}}_\omega^{~\,\gamma}\right]~.\end{aligned}\ ] ] and the different components of the angular momentum are : @xmath205 the supercharge operators as well as the angular momentum and @xmath59-charge operators are now easy to put into mathematica using matt headrick s grassman.m package . annihilation operators for fermions are implemented as grassman derivatives on the @xmath206 and @xmath207 . for completeness we also include the well - known expression for the laplacian in spherical coordinates @xmath208 in this appendix , we provide the differential equations one has to solve in order to obtain the energy eigenstates in each @xmath59-charge sector . to do so we will construct the @xmath50 highest weight states , which satisfy @xmath209 and @xmath210 , where @xmath211 . we can fill out the remaining states in the representation by acting with @xmath212 . without loss of generality , we can write down the full set of gauge invariant wavefunctions , labeled by their respective @xmath59-charges : @xmath213 where @xmath214 are functions of @xmath215 and @xmath216 is the fermionic ground state annihilated by @xmath28 and @xmath29 . these wavefunctions span the full 16-dimensional fermionic hilbert space . we now provide the schrdinger equations in order of difficulty . in the hopes of being self - contained , we recall here certain definitions that appear throughout the main text : @xmath217 and @xmath218 for the sake of keeping expressions compact , we define the operator @xmath219 . the highest weight state in the @xmath221 sector takes the form @xmath222 the schrdinger equation @xmath223 applied to this state then reduces to @xmath224 the highest weight state in the @xmath226 sector takes the form @xmath227 the schrdinger equation @xmath228 applied to this state then reduces to @xmath229 this is the same equation as in the @xmath221 case , and we see that their spectra coincide as a result of the discrete symmetry ( [ discretesymm ] ) . the highest weight state in the @xmath230 sector takes takes the form @xmath231 with @xmath232 and @xmath233 where we have included the factor of @xmath234 in @xmath235 to ensure that the restricted schrdinger equation , represented as a matrix on the @xmath236 and @xmath235 is manifestly real and symmetric . plugging @xmath237 into @xmath238 gives rise to a set of coupled equations for @xmath239 and @xmath240 which can be conveninently expressed in the following way @xmath241 it is worth noting that there are no @xmath120 solutions save for the trivial solution that satisfy @xmath242 . hence the states in this sector have @xmath243 . the highest weight state satisfying @xmath244 and @xmath245 in the @xmath246 sector takes the form @xmath247 with @xmath248 and similarly @xmath249 where we have again introduced the factor of @xmath250 in @xmath251 to simplify the form of the restricted equations . interestingly we find that the restricted schrdinger equation in this sector can be expressed as @xmath252 which is the same as ( [ schrorm1 ] ) , which again is a result of the symmetry ( [ discretesymm ] ) . for @xmath121 a highest weight state , we repeat the expressions found in the main text , that is @xmath253 with @xmath117 & \tilde{e}_{12}+\cos\vartheta\,\tilde{e}_{22}\\&\tilde{e}_{21}+\cos\vartheta\,\tilde{e}_{22 } & e^{i\varphi}\sin\vartheta\,\tilde{e}_{22}\end{pmatrix}~,\end{aligned}\ ] ] before writing down the schrdinger equation for the restricted set of functions , let us mention that @xmath254 satisfies the same differential equation as @xmath255 . unlike in the bps sector , where the equations implied strict equality , we can not immediately come to the same conclusion here . in fact , away from zero energy it is not always possible to argue that the functions should be equal . the schrdinger equation implies that @xmath127 satisfies @xmath256 and similarly for @xmath254 . let us assume that @xmath257 , then @xmath141 satisfies @xmath258 which is the restricted schrdinger equation of the @xmath142 sectors , which is expected to have non - zero solutions for certain @xmath259 and @xmath260 . we are now ready to write the coupled equations in the @xmath108 sector @xmath261 when solving these equations , we will encounter systems where @xmath141 is nonzero . when this is the case the state @xmath121 will come in a multiplet that includes states with @xmath59-charge @xmath262 with the same spin @xmath260 . when @xmath140 , the multiplet containing @xmath121 does not contain states with @xmath263 and thus must be annihilated by the pair @xmath264 and @xmath265 . when this happens we refer to it ( in a slight abuse of terminology ) as a short multiplet . previous treatments of this model have assumed a separation of scales between the chiral multiplet degrees of freedom and those of the vector multiplet @xcite . this is natural given the d - brane interpretation of the degrees of freedom , where the chiral fields represent stretched strings at the intersection points between the branes . when the branes are well separated , the chiral fields become effectively heavy and can be integrated out . in practice , we can study the effective interactions of the remaining degrees of freedom in a born - oppenheimer approximation . this was explicitly done in @xcite and we repeat the analysis here . the idea is to split the hamiltonian @xmath7 and the supercharges in powers of @xmath266 ( recall from ( [ ccrm ] ) that the gauginos come with an effective power of @xmath266 ) . thus we can write @xmath267 with @xmath268\left(\bar{\psi}\epsilon\right)_\gamma~,\\ q_\alpha^{(1)}&\equiv i\left[\nabla_{\vec{\mathbf{x}}}\cdot\vec{\boldsymbol{\sigma}}_\alpha^{~\,\gamma}-\left(|\phi|^2+\theta\right)\delta_\alpha^{~\,\gamma}\right]\lambda_\gamma~,\end{aligned}\ ] ] and similarly @xmath269 : @xmath270 we assume that the ground state wavefunction of the system can be approximated as @xmath271 where @xmath272 is the ground state of @xmath181 , in which @xmath10 appears as a parameter , and the @xmath273 are the excited states of @xmath181 . it is easy to show that @xmath274 where we have expressed @xmath10 in spherical coordinates , and have normalized @xmath272 . the fermionic vacuum @xmath216 is annihilated by @xmath28 and @xmath29 . there is some subtlety in the choice of @xmath275 , which we will come to shortly . notice however that @xmath276 . we now must solve for @xmath277 . in order to do so , we need to find the effective hamiltonian which acts on it . since this theory is supersymmetric , we can instead define the effective supercharges : @xmath278 ; these are @xmath279\lambda_\gamma~,\\ \bar{q}^{\beta}_{\rm eff}&=i\,\bar{\lambda}^\gamma\left[\left(\nabla_{\vec{\mathbf{x}}}-i\vec{\mathbf{a}}\right)\cdot\vec{\boldsymbol{\sigma}}_\gamma^{~\,\beta}+\left(\frac{1}{2r}+\theta\right)\delta_\gamma^{~\,\beta}\right]~,\end{aligned}\ ] ] where @xmath280 is the berry s vector potential that arises from integrating out the chiral degrees of freedom . a different choice of @xmath275 leads to a different gauge for this potential . notice that @xmath281 is the vector potential of a dirac monopole sitting at the origin , with a dirac string along the negative @xmath282-axis . the effective supercharges satisfy an algebra in order to show this . ] with non - vanishing commutator @xmath283 , with @xmath284 in summary , integrating out the chiral matter has generated an effective hamiltonian which represents the behavior of a supersymmetric particle moving in the background of a dyon ! dhoker and vinet @xcite studied this model in depth and algebraically determined its spectrum . later avery and michelson @xcite managed to write down the wavefunctions of this system for all energy levels . has an enhanced @xmath285 symmetry , and the states organize themselves into representations of this symmetry . ] to approximate the supersymmetric ground state , @xmath277 must be annihilated by @xmath286 and @xmath287 . this yields @xmath288 which is only normalizable for @xmath79 , as expected , in which case @xmath289 . the wavefunction @xmath290 in ( [ bototwv ] ) can be viewed as an expansion of the exact bps wavefunction in powers of @xmath266 @xcite . we would like estimate how close the first order term @xmath291 is to the actual ground state of the theory . let us do so by computing the expectation value of the total hamiltonian ( [ ham ] ) in this state . writing @xmath292 and integrating over the angular variables reveals @xmath293 this quantity is logarithmically divergent at small @xmath145 . the excited states of @xmath181 , which have energies @xmath294 , become relevant near the origin and can not be neglected . let us address some well warranted objections after all , the born - oppenheimer approximation is typically phrased as a variational problem whereby one tries to minimize the functional @xmath295 . this is done in two steps , one first defines @xmath296 , with @xmath272 given by the ground state of @xmath181 , and then minimizes the functional @xmath297 over @xmath277 . instead of following this procedure , we computed the effective supercharges , and used their algebra to find @xmath298 given in ( [ hameff ] ) . the only way for a discrepancy to arise is if @xmath299 , which is indeed the case . notice that @xmath300 , so terms proportional to the gauginos in ( [ hameff ] ) must come from higher order terms in perturbation theory , such as@xcite @xmath301 in conclusion , we have incorrectly dropped higher order terms in ( [ bototwv ] ) whose contributions to the total energy ( [ hameff ] ) come in at the same order in @xmath266 . this is nicely explained in @xcite . to obtain ( [ hameff ] ) from the born - oppenheimer approximation without first computing the supercharges , one must include the higher order corrections in ( [ bototwv ] ) . this will lead to coupled equations of the form @xmath302 , with @xmath303 some set of differential operators . one then back - substitutes these expressions for the @xmath304 into the schrdinger equation for @xmath277 to find @xmath298 of ( [ hameff ] ) acting on @xmath277 . this indicates that the born - oppenheimer approximation of the supersymmetric ground state will include terms proportional to the excited states of @xmath181 . these can be thought of as @xmath111 and @xmath20 in ( [ r0wf ] ) . f. denef , `` quantum quivers and hall / hole halos , '' jhep * 0210 * , 023 ( 2002 ) [ hep - th/0206072 ] . f. denef and g. w. moore , `` split states , entropy enigmas , holes and halos , '' jhep * 1111 * , 129 ( 2011 ) [ hep - th/0702146 ] . i. bena , m. berkooz , j. de boer , s. el - showk and d. van den bleeken , `` scaling bps solutions and pure - higgs states , '' jhep * 1211 * , 171 ( 2012 ) [ arxiv:1205.5023 [ hep - th ] ] . m. r. douglas and g. w. moore , `` d - branes , quivers , and ale instantons , '' hep - th/9603167 . s. -j . lee , z. -l . wang and p. yi , `` quiver invariants from intrinsic higgs states , '' jhep * 1207 * , 169 ( 2012 ) [ arxiv:1205.6511 [ hep - th ] ] . j. manschot , b. pioline and a. sen , `` from black holes to quivers , '' jhep * 1211 * , 023 ( 2012 ) [ arxiv:1207.2230 [ hep - th ] ] . j. manschot , b. pioline and a. sen , `` on the coulomb and higgs branch formulae for multi - centered black holes and quiver invariants , '' jhep * 1305 * , 166 ( 2013 ) [ arxiv:1302.5498 [ hep - th ] ] . s. sethi and m. stern , `` d - brane bound states redux , '' commun . * 194 * , 675 ( 1998 ) [ hep - th/9705046 ] . s. sethi and m. stern , `` the structure of the d0 - d4 bound state , '' nucl . b * 578 * , 163 ( 2000 ) [ hep - th/0002131 ] . a. v. smilga , `` vacuum structure in the chiral supersymmetric quantum electrodynamics , '' sov . jetp * 64 * , 8 ( 1986 ) [ zh . eksp . teor . * 91 * , 14 ( 1986 ) ] . a. v. smilga , `` perturbative corrections to effective zero mode hamiltonian in supersymmetric qed , '' nucl . b * 291 * , 241 ( 1987 ) . a. v. smilga , `` born - oppenheimer corrections to the effective zero mode hamiltonian in sym theory , '' jhep * 0204 * , 054 ( 2002 ) [ hep - th/0201048 ] . l. n. trefethen , _ spectral methods in matlab _ ( oxford , oxford university press , 2000 ) c. cordova and s. h. shao , `` an index formula for supersymmetric quantum mechanics , '' arxiv:1406.7853 [ hep - th ] . k. hori , h. kim and p. yi , `` witten index and wall crossing , '' jhep * 1501 * , 124 ( 2015 ) [ arxiv:1407.2567 [ hep - th ] ] . k. ohta and y. sasai , `` exact results in quiver quantum mechanics and bps bound state counting , '' jhep * 1411 * , 123 ( 2014 ) [ arxiv:1408.0582 [ hep - th ] ] . h. kim , `` scaling behaviour of quiver quantum mechanics , '' jhep * 1507 * , 079 ( 2015 ) [ arxiv:1503.02623 [ hep - th ] ] . m. berkooz , m. r. douglas and r. g. leigh , `` branes intersecting at angles , '' nucl . phys . b * 480 * , 265 ( 1996 ) [ hep - th/9606139 ] . s. kachru and j. mcgreevy , `` supersymmetric three cycles and supersymmetry breaking , '' phys . d * 61 * , 026001 ( 2000 ) [ hep - th/9908135 ] . s. g. avery and j. michelson , `` mechanics and quantum supermechanics of a monopole probe including a coulomb potential , '' phys . d * 77 * , 085001 ( 2008 ) [ arxiv:0712.0341 [ hep - th ] ] . d. anninos , t. anous , p. de lange and g. konstantinidis , `` conformal quivers and melting molecules , '' jhep * 1503 * , 066 ( 2015 ) [ arxiv:1310.7929 [ hep - th ] ] . i. bena , a. puhm and b. vercnocke , `` non - extremal black hole microstates : fuzzballs of fire or fuzzballs of fuzz ? , '' jhep * 1212 * , 014 ( 2012 ) [ arxiv:1208.3468 [ hep - th ] ] . yi , `` witten index and threshold bound states of d - branes , '' nucl . b * 505 * , 307 ( 1997 ) [ hep - th/9704098 ] . g. w. moore , n. nekrasov and s. shatashvili , `` d particle bound states and generalized instantons , '' commun . phys . * 209 * , 77 ( 2000 ) [ hep - th/9803265 ] . j. plefka and a. waldron , `` on the quantum mechanics of m(atrix ) theory , '' nucl . b * 512 * , 460 ( 1998 ) [ hep - th/9710104 ] . m. porrati and a. rozenberg , `` bound states at threshold in supersymmetric quantum mechanics , '' nucl . b * 515 * , 184 ( 1998 ) [ hep - th/9708119 ] . j. frohlich , g. m. graf , d. hasler , j. hoppe and s. t. yau , `` asymptotic form of zero energy wave functions in supersymmetric matrix models , '' nucl . b * 567 * , 231 ( 2000 ) [ hep - th/9904182 ] . y. h. lin and x. yin , `` on the ground state wave function of matrix theory , '' arxiv:1402.0055 [ hep - th ] . c. t. asplund , f. denef and e. dzienkowski , `` massive quiver matrix models for massive charged particles in ads , '' arxiv:1510.04398 [ hep - th ] . e. dhoker and l. vinet , `` spectrum ( super)symmetries of particles in a coulomb potential , '' nucl . b * 260 * , 79 ( 1985 ) . e. dhoker and l. vinet , `` constants of motion for a spin 1/2 particle in the field of a dyon , '' phys . * 55 * , 1043 ( 1985 ) . l. vinet and e. dhoker , `` spectrum supersymmetry for a multiplet of two spin 0 and one spin 1/2 particles in a coulomb potential , '' in * seoul 1985 , proceedings , group theoretical methods in physics * , 569 - 575 e. dhoker and l. vinet , `` hidden symmetries and accidental degeneracy for a spin 1/2 particle in the field of a dyon , '' lett . * 12 * , 71 ( 1986 ) . e. dhoker and l. vinet , `` supersymmetries of the dyon , '' in * de vega , h.j . ( ed . ) , sanchez , n. ( ed . ) : field theory , quantum gravity and strings , vol . 2 * , 156 - 173

frequent episode discovery framework is a popular framework in temporal data mining with many applications . over the years many different notions of frequencies of episodes have been proposed along with different algorithms for episode discovery . in this paper we present a unified view of all such frequency counting algorithms . we present a generic algorithm such that all current algorithms are special cases of it . this unified view allows one to gain insights into different frequencies and we present quantitative relationships among different frequencies . our unified view also helps in obtaining correctness proofs for various algorithms as we show here . we also point out how this unified view helps us to consider generalization of the algorithm so that they can discover episodes with general partial orders .

1007.0690

temporal data mining is concerned with finitely many useful patterns in sequential ( symbolic ) data streams @xcite . frequent episode discovery , first introduced in @xcite , is a popular framework for mining patterns from sequential data . the framework has been successfully used in many application domains , e.g. , analysis of alarm sequences in telecommunication networks @xcite , root cause diagnostics from faults log data in manufacturing @xcite , user - behavior prediction from web interaction logs @xcite , inferring functional connectivity from multi - neuronal spike train data @xcite , relating financial events and stock trends @xcite , protein sequence classification @xcite , intrusion detection @xcite , text mining @xcite , seismic data analysis @xcite etc . the data in this framework is a single long stream of events , where each event is described by a symbolic event - type from a finite alphabet and the time of occurrence of the event . the patterns of interest are termed episodes . informally , an episode is a short ordered sequence of event types , and a _ frequent _ episode is one that occurs often enough in the given data sequence . discovering frequent episodes is a good way to unearth temporal correlations in the data . given a user - defined frequency threshold , the task is to efficiently obtain all frequent episodes in the data sequence . an important design choice in frequent episode discovery is the definition of frequency of episodes . intuitively any frequency should capture the notion of the episode occurring many times in the data and , at the same time , should have an efficient algorithm for computing the same . there are many ways to define frequency and this has given rise to different algorithms for frequent episode discovery @xcite . in the original framework of @xcite , frequency was defined as the number of fixed - width sliding windows over the data that contain at least one occurrence of the episode . another notion for frequency is based on the number of _ minimal _ occurrences @xcite . two frequency definitions called _ head frequency _ and _ total frequency _ are proposed in @xcite in order to overcome some limitations of the windows - based frequency of @xcite . in @xcite , two more frequency definitions for episodes were proposed , based on certain specialized sets of occurrences of episodes in the data . many of the algorithms , such as the winepi of @xcite and the occurrences - based frequency counting algorithms of @xcite , employ finite state automata as the basic building blocks for recognizing occurrences of episodes in the data sequence . an automata - based counting scheme for minimal occurrences has also been proposed in @xcite . the multiplicity of frequency definitions and the associated algorithms for frequent episode discovery makes it difficult to compare the different methods . in this paper , we present a unified view of algorithms for frequent episode discovery under all the various frequency definitions . we present a generic automata - based algorithm for obtaining frequencies of a set of episodes and show that all the currently available algorithms can be obtained as special cases of this method . this viewpoint helps in obtaining useful insights regarding the kinds of occurrences tracked by the different algorithms . the framework also aids in deriving proofs of correctness for the various counting algorithms , many of which are not currently available in literature . our framework also helps in understanding the anti - monotonicity conditions satisfied by different frequencies which is needed for the candidate generation step . our general view can also help in generalizing current algorithms , which can discover only serial or parallel episodes , to the case of episodes with general partial orders and we briefly comment on this in our conclusions . the paper is organized as follows . [ sec : overview ] gives an overview of the episode framework and explains all the currently used frequencies in literature . [ sec : algorithms ] presents our generic algorithm and shows that all current counting techniques for these various frequencies can be derived as special cases . [ sec : proof - of - correctness ] gives proofs of correctness for the various counting algorithms utilizing this unified framework . [ sec : candgen ] discusses the candidate generation step for all these frequencies . in sec . [ sec : discussion ] we provide some discussion and concluding remarks . in this section we briefly review the framework of frequent episode discovery @xcite . the data , referred to as an _ event sequence _ , is denoted by @xmath0 @xmath1 @xmath2 , where each pair @xmath3 represents an _ event _ , and the number of events in the event sequence is @xmath4 . each @xmath5 is a symbol ( or _ event - type _ ) from a finite alphabet , @xmath6 , and @xmath7 is a positive integer representing the time of occurrence of the @xmath8 event . the sequence is ordered so that , @xmath9 for all @xmath10 . the following is an example event sequence with 10 events : @xmath11 an @xmath12-node episode , @xmath13 , is defined as a triple , @xmath14 , where @xmath15 , is a collection of @xmath12 nodes , @xmath16 is a partial order on @xmath17 and @xmath18 is a map that associates each node in @xmath13 with an event type from @xmath6 . thus an episode is a ( typically small ) collection of event - types along with an associated partial order . when the order @xmath16 is total , @xmath13 is called a serial episode , and when the order is empty , @xmath13 is called a parallel episode . in this paper , we restrict our attention to serial episodes . without loss of generality , we can now assume that the total order on the nodes of @xmath13 is given by @xmath19 . for example , consider a 3-node episode @xmath20 , @xmath21 , @xmath22 , @xmath23 , with @xmath24 . we denote such an episode by @xmath25 . an occurrence of episode @xmath13 in an event sequence @xmath26 is a map @xmath27 such that @xmath28 for all @xmath29 , and for all @xmath30 with @xmath31 we have @xmath32 . in the example event sequence @xmath33 , the events @xmath34 , @xmath35 and @xmath36 constitute an occurrence of @xmath37 while @xmath35 , @xmath38 and @xmath36 do not . we use @xmath39 $ ] to refer to the @xmath40 event - type in @xmath13 . this way , an @xmath12-node episode @xmath13 can be represented using @xmath41 { \ensuremath{\rightarrow}}\alpha[2 ] { \ensuremath{\rightarrow}}\ldots { \ensuremath{\rightarrow}}\alpha[n])$ ] . an episode @xmath42 is said to be a _ subepisode _ of @xmath13 ( denoted @xmath43 ) if all the event - types in @xmath42 also appear in @xmath13 , and if their order in @xmath42 is same as that in @xmath13 . for example , @xmath44 is a 2-node subepisode of the episode @xmath37 while @xmath45 is not . the _ frequency _ of an episode is some measure of how often it occurs in the event sequence . a frequent episode is one whose frequency exceeds a user - defined threshold . the task in frequent episode discovery is to find all frequent episodes . given an occurrence @xmath46 of an @xmath12-node episode @xmath13 , @xmath47 is called the _ span _ of the occurrence . in many applications , one may want to consider only those occurrences whose span is below some user - chosen limit . ( this is because , occurrences constituted by events that are widely separated in time may not represent any underlying causative influences ) . we call any such constraint on span as an _ expiry - time constraint_. the constraint is specified by a threshold , @xmath48 , such that occurrences of episodes whose span is greater than @xmath48 are not considered while counting the frequency . one popular approach to frequent episode discovery is to use an apriori - style level - wise procedure . at level @xmath49 of the procedure , a ` candidate generation ' step combines frequent episodes of size @xmath50 to build candidates ( or potential frequent episodes ) of size @xmath49 using some kind of anti - monotonicity property ( e.g. frequency of an episode can not exceed frequency of any of its subepisodes ) . the second step at level @xmath49 is called ` frequency counting ' in which , the algorithm counts or computes the frequencies of the candidates and determines which of them are frequent . there are many ways to define the frequency of an episode . intuitively , any definition must capture some notion of how often the episode occurs in the data . it must also admit an efficient algorithm to obtain the frequencies for a set of episodes . further , to be able to apply a level - wise procedure , we need the frequency definition to satisfy some anti - monotonicity criterion . additionally , we would also like the frequency definition to be conducive to statistical significance analysis . in this section , we discuss various frequency definitions that have been proposed in literature . ( recall that the data is an event sequence , @xmath51 ) . [ def : windowsbased ] @xcite a window on an event sequence , @xmath26 , is a time interval @xmath52 $ ] , where @xmath53 and @xmath54 are positive integers such that @xmath55 and @xmath56 . the _ window width _ of @xmath52 $ ] is given by @xmath57 . given a user - defined window width @xmath48 , the * windows - based frequency * of @xmath13 is the number of windows of width @xmath48 which contain at least one occurrence of @xmath13 . for example , in the event sequence ( [ eq : example - sequence ] ) , there are @xmath58 windows with window width @xmath58 which contain an occurrence of @xmath37 . @xcite the time - window of an occurrence , @xmath46 , of @xmath13 is given by @xmath59 $ ] . a _ minimal window _ of @xmath13 is a time - window which contains an occurrence of @xmath13 , such that no proper sub - window of it contains an occurrence of @xmath13 . an occurrence in a minimal window is called a minimal occurrence . the * minimal occurrences - based frequency * of @xmath13 in @xmath26 ( denoted @xmath60 ) is defined as the number of minimal windows of @xmath13 in @xmath26 . [ def : minimal ] in the example sequence ( [ eq : example - sequence ] ) there are 3 minimal windows of @xmath37 : @xmath61 $ ] , @xmath62 $ ] and @xmath63 $ ] . @xcite given a window - width @xmath49 , the * head frequency * of @xmath13 is the number of windows of width @xmath49 which contain an occurrence of @xmath13 starting at the left - end of the window and is denoted as @xmath64 . [ def : head ] @xcite given a window width @xmath49 , the * total frequency * of @xmath13 , denoted as @xmath65 , is defined as follows . @xmath66 [ def : total ] for a window - width of @xmath67 , the head frequency @xmath68 of @xmath69 in ( [ eq : example - sequence ] ) is @xmath70 . the total frequency of @xmath71 , @xmath72 , in ( [ eq : example - sequence ] ) is @xmath73 because the head frequency of @xmath74 in ( [ eq : example - sequence ] ) is @xmath73 . @xcite two occurrences @xmath75 and @xmath76 of @xmath13 are said to be _ non - overlapped _ if either @xmath77 or @xmath78 . a set of occurrences is said to be non - overlapped if every pair of occurrences in the set is non - overlapped . a set @xmath79 , of non - overlapped occurrences of @xmath13 in @xmath26 is _ maximal _ if @xmath80 , where @xmath81 is any other set of non - overlapped occurrences of @xmath13 in @xmath26 . the * non - overlapped frequency * of @xmath13 in @xmath26 ( denoted as @xmath82 ) is defined as the cardinality of a maximal non - overlapped set of occurrences of @xmath13 in @xmath26 . [ def : nonoverlapped ] two occurrences are non - overlapped if no event of one occurrence appears in between events of the other . the notion of a maximal non - overlapped set is needed since there can be many sets of non - overlapped occurrences of an episode with different cardinality@xcite . the non - overlapped frequency of @xmath71 in ( [ eq : example - sequence ] ) is @xmath83 . a maximal set of non - overlapped occurrences is @xmath84 and @xmath85 . @xcite two occurrences @xmath75 and @xmath76 of @xmath13 are said to be _ non - interleaved _ if either @xmath86 or @xmath87 . a set of occurrences @xmath79 of @xmath13 in @xmath26 is _ non - interleaved _ if every pair of occurrences in the set is non - interleaved . a set @xmath79 of non - interleaved occurrences of @xmath13 in @xmath26 is * maximal * if @xmath80 , where @xmath81 is any other set of non - interleaved occurrences of @xmath13 in @xmath26 . the * non - interleaved frequency * of @xmath13 in @xmath26 ( denoted as @xmath88 ) is defined as the cardinality of a maximal non - interleaved set of occurrences of @xmath13 in @xmath26 . [ def : noninterleaved ] the occurrences @xmath89 and @xmath90 are non - interleaved ( though overlapped ) occurrences of @xmath37 in @xmath26 . together with @xmath91 , these two occurrences form a set of maximal non - interleaved occurrences of @xmath37 in ( [ eq : example - sequence ] ) and thus @xmath92 . @xcite two occurrences @xmath75 and @xmath76 of @xmath13 are said to be _ distinct _ if they do not share any two events . a set of occurrences is distinct if every pair of occurrences in it is distinct . a set @xmath79 of distinct occurrences of @xmath13 in @xmath26 is * maximal * if @xmath80 , where @xmath81 is any other set of distinct occurrences of @xmath13 in @xmath26 . the * distinct occurrences - based frequency * of @xmath13 in @xmath26 ( denoted as @xmath93 ) is the cardinality of a maximal distinct set of occurrences of @xmath13 in @xmath26 . [ def : distinct ] the three occurrences that constituted the maximal non - interleaved occurrences of @xmath37 in ( [ eq : example - sequence ] ) also form a set of maximal distinct occurrences in ( [ eq : example - sequence ] ) . + the first frequency proposed in the literature was the windows based count @xcite and was originally applied for analyzing alarms in a telecommunication network . it uses an automata based algorithm called winepi for counting . candidate generation exploits the anti - monotonicity property that all subepisodes are at least as frequent as the parent episode . a statistical significance test for frequent episodes based on the windows - based count was proposed in @xcite . there is also an algorithm for discovering frequent episodes with a maximum - gap constraint under the windows - based count @xcite . the minimal windows based frequency and a level - wise procedure called minepi to track minimal windows were also proposed in @xcite . this algorithm has high space complexity since the exact locations of all the minimal windows of the various episodes are kept in memory . nevertheless , it is useful in rule generation . an efficient automata - based scheme for counting the number of minimal windows ( along with a proof of correctness ) was proposed in @xcite . the problem of statistical significance of minimal windows was recently addressed in @xcite . an algorithm for extracting rules under a maximal gap constraint and based on minimal occurrences has been proposed in @xcite . in the windows - based frequency , the window width is essentially an expiry - time constraint ( an upper - bound on the span of the episodes ) . however , if the span of an occurrence is much smaller than the window width , then its frequency is artificially inflated because the same occurrence will be found in several successive sliding windows . the head frequency measure , proposed in @xcite , is a variant of the windows - based count intended to overcome this problem . based on the notion of head frequency , @xcite presents two algorithms minepi+ and emma . they also point out how head frequency can be a better choice for rule generation compared to the windows - based or the minimal windows - based counts . under the head frequency count , however , there can be episodes whose frequency is higher than some of their subepisodes ( see @xcite for details ) . to circumvent this , @xcite propose the idea of total frequency . currently , there is no statistical significance analysis based on head frequency or total frequency . an efficient automata - based counting algorithm under the non - overlapped frequency measure ( along with a proof of correctness ) can be found in @xcite . a statistical significance test for the same is proposed in @xcite . however , the algorithm in @xcite does not handle any expiry - time constraints . an efficient automata - based algorithm for counting non - overlapped occurrences under expiry - time constraint was proposed in @xcite though this has higher time and space complexity than the algorithm in @xcite . no proofs of correctness or statistical significance analysis are available for non - overlapped occurrences under an expiry - time constraint . algorithms for frequent episode discovery under the non - interleaved frequency can be found in @xcite . no proofs of correctness are available for these algorithms . another frequency measure we discuss in this paper is based on the idea of distinct occurrences . no algorithms are available for counting frequencies under this measure . the unified view of automata - based counting that we will present in this paper can be readily used to design algorithms for counting distinct occurrences of episodes . in this section , we present a generic algorithm for obtaining frequencies of episodes under the different frequency definitions listed in sec . [ sec : frequencies - of - episodes ] . the basic ingredient in all the algorithms is a simple finite state automaton ( fsa ) that is used to recognize ( or track ) an episode s occurrences in the event sequence . the fsa for recognizing occurrences of @xmath37 is illustrated in fig . [ fig : automaton ] . in general , an fsa for an @xmath12-node serial episode @xmath94{\ensuremath{\rightarrow}}\alpha[2]{\ensuremath{\rightarrow}}\dots{\ensuremath{\rightarrow}}\alpha[n]$ ] has @xmath95 states . the first @xmath12 states are represented by a pair @xmath96)$ ] , @xmath97 . the @xmath98 state is @xmath99 where @xmath100 is a null symbol . intuitively , if the fsa is in state @xmath101)$ ] , it means that the fsa has already seen the first @xmath102 event types of this episode and is now waiting for @xmath103 $ ] ; if we now encounter an event of type @xmath103 $ ] in the data it can accept it ( that is , it can transit to its next state ) . the start ( first ) state of the fsa is @xmath104)$ ] . the @xmath98 state is the accepting state because when an automaton reaches this state , a full occurrence of the episode is tracked . \(0 ) @xmath105 ; ( 1 ) [ right = of 0 ] @xmath106 ; ( 2 ) [ right = of 1 ] @xmath107 ; ( 3 ) [ right = of 2 ] @xmath108 ; ( 0 ) edge node @xmath109 ( 1 ) edge [ loop above ] node @xmath6\@xmath110 ( ) ( 1 ) edge node @xmath111 ( 2 ) edge [ loop above ] node @xmath6\@xmath112 ( ) ( 2 ) edge node @xmath113 ( 3 ) edge [ loop above ] node @xmath6\@xmath114 ( ) ( 3 ) edge [ loop above ] node @xmath6 ( ) ; we first explain how these fsa can be used for obtaining all the different types of frequencies of episodes before presenting the generic algorithm . while discussing various algorithms , we represent any occurrence @xmath46 by @xmath115 $ ] , which is the vector of times of the events that constitute the occurrence . for the discussion of all algorithms in this section , we consider the example of tracking occurrences of @xmath116 in the data stream @xmath117 given by @xmath118 there is a ` natural ' lexicographic order on the set of * all * occurrences @xmath119 , of any episode , @xmath13 , defined below . this is a total order on @xmath119 and it will be useful in our analysis . the lexicographic ordering , @xmath120 , on the set of all occurrences of @xmath13 is defined as follows : @xmath121 if the least @xmath122 for which @xmath123 is such that @xmath124 . the simplest of all automata - based frequency counting algorithms is the one for counting non - overlapped occurrences @xcite which uses only @xmath125-automata per episode . ( we call it algorithm no here ) . at the start , one automaton for each of the candidate episodes is initialized in its start state . each of the automata make a state transition as soon as a relevant event - type appears in the data stream . whenever an automaton reaches its final state , frequency of the corresponding episode is incremented , the automaton is removed from the system and a fresh automaton for the episode is initialized in the start state . as is easy to see , this method will count non - overlapped occurrences of episodes . under the no algorithm , we denote the occurrence tracked by the @xmath40 automaton initialized for @xmath13 as @xmath126 . in our example , algorithm no tracks the following two occurrences of the episode @xmath13 : ( i ) @xmath127 $ ] and ( ii ) @xmath128 $ ] , and the corresponding non - overlapped frequency is @xmath83 . in this paper we introduce the concept of _ earliest transiting _ occurrence of an episode which is useful for analyzing different frequency counting algorithms . _ an occurrence @xmath46 of @xmath13 is called earliest transiting if @xmath129 is the first occurrence of @xmath39 $ ] after @xmath130 @xmath131 . [ def : earliest transiting ] _ it is easy to see that all occurrences tracked by algorithm no are earliest transiting . let @xmath132 denote the set of all earliest transiting occurrences of a given episode . we denote the @xmath40 occurrence ( as per the lexicographic ordering of occurrences ) in @xmath132 as @xmath133 . there are 6 earliest transiting occurrences of @xmath13 in @xmath117 . they are @xmath134 $ ] , @xmath135 $ ] , @xmath136 $ ] , @xmath137 $ ] , @xmath138 $ ] and @xmath139 $ ] . the earliest transiting occurrences tracked by the no algorithm are @xmath140 and @xmath141 . while the algorithm no is very simple and efficient , it can not handle any expiry - time constraint . recall that the expiry - time constraint specifies an upper - bound , @xmath48 , on the span of any occurrence that is counted . suppose we want to count with @xmath142 . both the occurrences tracked by no have spans greater than @xmath143 and hence the resulting frequency count would be zero . however , @xmath144 is an occurrence which satisfies the expiry time constraint . algorithm no can not track @xmath144 because it uses only one automaton per episode and the automaton has to make a state transition as soon as the relevant event - type appears in the data . to overcome this limitation , the algorithm can be modified so that a new automaton is initialized in the start state , whenever an existing automaton moves out of its start state . all automata make state transitions as soon as they are possible . each such automaton would track an earliest transiting occurrence . in this process , two automata may reach the same state . in our example , after seeing @xmath145 , the second and third automata to be initialized for @xmath13 , would be waiting in the same state ( ready to accept the next @xmath111 in the data ) . clearly , both automata will make state transitions on the same events from now on and so we need to keep only one of them . we retain the newer or most recently initialized automaton ( in this case , the third automaton ) since the span of the occurrence tracked by it would be smaller . when an automaton reaches its final state , if the span of the occurrence tracked by it is less than @xmath48 , then the corresponding frequency is incremented and all automata of the episode except the one waiting in the start state are retired . ( this ensures we are tracking only non - overlapped occurrences ) . when the occurrence tracked by the automaton that reaches the final state fails the expiry constraint , we just retire the current automaton ; any other automata for the episode will continue to accept events . under this modified algorithm , in @xmath117 , the first automaton that reaches its final state tracks @xmath146 which violates the expiry time constraint of @xmath147 . so , we drop only this automaton . the next automaton that reaches its final state tracks @xmath144 . this occurrence has span less than @xmath147 . hence we increment the corresponding frequency count and retire all current automata for this episode . since there are no other occurrences non - overlapped with @xmath144 , the final frequency would be 1 . we denote this algorithm for counting the non - overlapped occurrences under an expiry - time constraint as no - x . the occurrences tracked by both no and no - x would be earliest transiting . note that several earliest transiting occurrences may end simultaneously . for example , in @xmath117 , @xmath148 , @xmath149 and @xmath150 all end together at @xmath151 . both @xmath152 and @xmath153 form maximal sets of non - overlapped occurrences . sometimes ( e.g. when determining the distribution of spans of occurrences for an episode ) we would like to track the _ innermost _ one among the occurrences that are ending together . in this example , this means we want to track the set of occurrences @xmath153 . this can be done by simply omitting the expiry - time check in the no - x algorithm . ( that is , whenever an automaton reaches final state , irrespective of the span of the occurrence tracked by it , we increment frequency and retire all other automata except for the one in start state ) . we denote this as the no - i algorithm and this is the algorithm proposed in @xcite . in no - i , if we only retire automata that reached their final states ( rather than retire all automata except the one in the start state ) , we have an algorithm for counting minimal occurrences ( denoted mo ) . in our example , the automata tracking @xmath150 , @xmath154 and @xmath155 are the ones that reach their final states in this algorithm . the time - windows of these occurrences constitute the set of all minimal windows of @xmath13 in @xmath117 . expiry time constraints can be incorporated by incrementing frequency only when the occurrence tracked has span less than the expiry - time threshold . the corresponding expiry - time algorithm is referred to as mo - x . the windows - based counting algorithm ( which we refer to as wb ) is also based on tracking earliest transiting occurrences . wb also uses multiple automata per episode to track minimal occurrences of episodes like in mo . the only difference lies in the way frequency is incremented . the algorithm essentially remembers , for each candidate episode , the last minimal window in which the candidate was observed . then , at each time tick , effectively , if this last minimal window lies within the current sliding window of width @xmath48 , frequency is incremented by one . this is because , an occurrence of episode @xmath13 exists in a given window @xmath156 if and only @xmath156 contains a minimal window of @xmath13 . it is easy to see that head frequency with a window - width of @xmath48 is simply the number of earliest transiting occurrences whose span is less than @xmath48 . thus we can have a head frequency counting algorithm ( referred to here as hd ) that is similar to mo - x except that when two automata reach the same state simultaneously we do not remove the older automaton . this way , hd will track all earliest transiting occurrences which satisfy an expiry time - constraint of @xmath48 . for @xmath157 and for episode @xmath13 , hd tracks @xmath146 , @xmath144 , @xmath158 and @xmath159 and returns a frequency count of @xmath70 . the total frequency count for an episode @xmath13 is the minimum of the head frequencies of all its subepisodes ( including itself ) . this can be computed as the minimum of the head frequency of @xmath13 and the total frequency of its @xmath160-suffix subepisodes which would have been computed in the previous pass over the data . ( see @xcite for details ) . the head frequency counting algorithm can have high space - complexity as all the time instants at which automata make their first state transition need to be remembered . the non - interleaved frequency counting algorithm ( which we refer to as ni ) differs from the minimal occurrence algorithm in that , an automaton makes a state transition only if there is no other automaton of the same episode in the destination state . unlike the other frequency counting algorithms discussed so far , such an fsa transition policy will track occurrences which are not necessarily earliest transiting . in our example , until the event @xmath161 in the data sequence , both the minimal and non - interleaved algorithms make identical state transitions . however , on @xmath145 , ni will not allow the automaton in state @xmath162 to make a state transition as there is already an active automaton for @xmath13 in state @xmath163 which had accepted @xmath161 earlier . eventually , ni tracks the occurrences @xmath164 $ ] , @xmath165 $ ] , @xmath166 $ ] and @xmath167 $ ] . while there are no algorithms reported for counting distinct occurrences , we can construct one using the same ideas . such an algorithm ( to be called as do ) differs from the one for counting minimal occurrences , in allowing multiple automata for an episode to reach the same state . however , on seeing an event @xmath3 which multiple automata can accept , only one of the automata ( the oldest among those in the same state ) is allowed to make a state transition ; the others continue to wait for future events with the same event - type as @xmath5 to make their state transitions . the set of maximal distinct occurrences of @xmath13 in @xmath117 are @xmath168 , @xmath169 $ ] , @xmath170 $ ] , @xmath171 $ ] and @xmath172 $ ] which are the ones tracked by this algorithm . we can also consider counting _ all _ occurrences of an episode even though it may be inefficient . the algorithm for counting _ all _ occurrences ( referred to as the ao ) allows all automata to make transitions whenever the appropriate events appear in the data sequence . however , at each state transition , a copy of the automaton in the earlier state is added to the set of active automata for the episode . from the above discussion , it is clear that by manipulating the fsa ( that recognize occurrences ) in different ways we get counting schemes for different frequencies . the choices to be made in different algorithms essentially concern when to initiate a new automaton in the start state , when to retire an existing automaton , when to effect a possible state transition and when ( and by how much ) to increment the frequency . we now present a unified scheme incorporating all this in _ algorithm [ algo : unified ] _ for obtaining frequencies of a set of serial episodes . this algorithm has five boolean variables , namely , transit , copy - automaton , join - automaton , increment - freq and retire - automaton . the counting algorithms for all the different frequencies are obtained from this general algorithm by suitably setting the values of these boolean variables ( either by some constants or by values calculated using the current context in the algorithm ) . tables [ tab : transit ] [ tab : retire - automata ] specify the choices needed to obtain the algorithms for different frequencies . ( a list of all algorithms is given in table [ tab : various - counts ] ) . as can be seen from our general algorithm , when an event type for which an automaton is waiting is encountered in the data , the the automaton can accept it only if the variable transit is true . hence for all algorithms that track earliest transiting occurrences , transit will be set to true as can be seen from table [ tab : transit ] . for algorithms ni and do where we allow the state transition only if some condition is satisfied . the condition copy - automaton ( table [ tab : copy - automaton ] ) is for deciding whether or not to leave another automaton in the current state when an automaton is transiting to the next state . except for no and ao , we create such a copy only when the currently transiting automaton is moving out of its start state . in no we never make such a copy ( because this algorithm uses only one automaton per episode ) while in ao we need to do it for every state transition . as we have seen earlier , in some of the algorithms , when two automata for an episode reach the same state , the older automaton is removed . this is controlled by join - automaton , as given by table [ tab : join - automaton ] . increment - frequency ( table [ tab : increment - freq ] ) is the condition under which the frequency of an episode is incremented when an automaton reaches its final state . this increment is always done for algorithms that have no expiry time constraint or window width . for the others we increment the frequency only if the occurrence tracked satisfies the constraint . retire - automata condition ( table [ tab : retire - automata ] ) is concerned with removal of all automata of an episode when a complete occurrence has been tracked . this condition is true only for the non - overlapped occurrences - based counting algorithms . apart from the five boolean variables explained above , our general algorithm contains one more variable , namely , inc , which decides the amount by which frequency is incremented when an automaton reaches the final state . its values for different frequency counts are listed in table [ tab : inc ] . for all algorithms except wb , we set @xmath173 . we now explain how frequency is incremented in wb . to count the number of sliding windows that contain at least one occurrence of the episode , whenever a new minimal occurrence enters a sliding window , we can calculate the number of consecutive windows in which this new minimal occurrence will be found in . for example , in @xmath117 , with a window - width of @xmath174 , consider the first minimal occurrence of @xmath175 , namely , the occurrence constituted by events @xmath145 , @xmath176 , @xmath177 and @xmath151 . the first sliding window in which this occurrence can be found is @xmath178 $ ] . the occurrence stays in consecutive sliding windows , until the sliding window @xmath179 $ ] . when this first minimal occurrence enters the sliding window @xmath178 $ ] , we observe that there is no other ` older ' minimal occurrence in @xmath178 $ ] , and hence , as per the _ else _ condition in _ table [ tab : inc ] _ , the @xmath180 is incremented by @xmath181 . similarly , when the second minimal occurrence enters the sliding window @xmath182 $ ] , we increment @xmath180 by @xmath183 . the third minimal occurrence ( constituted by the events @xmath184 , @xmath185 , @xmath186 and @xmath187 ) first enters the sliding window @xmath188 $ ] , with the second minimal window still occurring within this window . this third minimal occurrence remains in consecutive sliding windows until @xmath189 $ ] . as per the _ if _ condition of _ table [ tab : inc ] _ , @xmath180 is incremented by @xmath190 . we note that such an implementation of the windows - based algorithm removes the need for the @xmath191 list of @xcite which was used to store all automata whose first state transition occurred at time - tick @xmath192 . set @xmath193 of @xmath12-node serial episodes , event stream @xmath194 , @xmath195 , @xmath196 , frequencies of episodes in @xmath193 add automaton of @xmath13 waiting in the start state . initialize frequency of @xmath13 to zero . @xmath13=candidate associated with @xmath197 ; @xmath102 = state which @xmath197 is ready to transit into ; add copy of @xmath197 to collection of automata . transit @xmath197 to state @xmath102 retain @xmath197 and retire earlier automaton retire @xmath197 . increment frequency of @xmath13 by inc . retire all automaton of @xmath13 and create a state 0 automaton . .values taken by inc [ cols="^,^",options="header " , ] [ remark:3 - 1 ] even though we included ao ( for counting all occurrences of an episode ) for sake of completeness , this is not a good frequency measure . this is mainly because it does not seem to satisfy any anti - monotonicity condition . for example , consider the data sequence @xmath198 . there are @xmath199 occurrences of @xmath200 but only 4 occurrences of each of its @xmath83-node subepisodes . also , its space complexity can be high . [ remark:3.2 ] : the quantitative relationships between the different frequency counts for a given episode can be described as follows : @xmath201 where @xmath202 denotes the frequency of an episode under ao , while @xmath203 and @xmath204 denote the corresponding head and total frequencies defined with a window - width exceeding the total time - span of the event sequence . for a large sliding window width , the head frequency @xmath203 is same as the number of earliest transiting occurrences of an episode . in general , the inequality @xmath205 holds only for injective episodes ( an episode @xmath13 is injective if it does not contain any repeated event - types ) . all other inequalities are true for any serial episode . the first inequality is obvious . the second inequality follows directly from _ equation [ eq : total - frequency ] _ in _ definition [ def : total]_. given a set of @xmath206 maximal distinct occurrences of an episode @xmath13 in a data stream @xmath26 , one can extract that many earliest transiting occurrences of not only @xmath13 but also of all its subepisodes in @xmath26 . hence the third inequality is also true . also , it is easy to verify that a set of non - interleaved occurrences of an injective episode are also distinct , which validates the fourth inequality . we will show the correctness of the remaining two inequalities in the next section . in this section , we present proofs of correctness of the different frequency counting algorithms presented in sec . [ sec : algorithms ] ( all of which are specific instances of _ algorithm [ algo : unified ] _ ) . in our proofs , we consider the case of event sequences with distinct occurrence - times for events . when we are not considering expiry - time constraints , the actual values of times of occurrences of different events are not really important ; only the time ordering of the events is important in deciding on the occurrences of episodes . hence , in this section we will use @xmath207 interchangeably with @xmath208 , the time of the first event in the occurrence @xmath46 and so on . modifications needed in the case of data having multiple events with the same time of occurrence , are discussed at the end of the section . first , we analyze the minimal occurrences counting algorithm ( mo ) . our proof methodology is different from the one presented in @xcite , where , the algorithm is viewed as computing a table @xmath209 $ ] , where , @xmath210 $ ] is the largest value @xmath211 such that @xmath212 contains an occurrence of @xmath213{\ensuremath{\rightarrow}}\ldots \alpha[j]$ ] , using dynamic programming . the algorithm , after processing @xmath5 , stores the @xmath40 row of this matrix . the dynamic programming recursion helps compute the @xmath40 row of this matrix from its @xmath214 row . whenever @xmath215>s[i-1,n]$ ] , the count is incremented since a new minimal occurrence is recognized . viewed from an automata perspective , the @xmath40 row of the matrix essentially stores the first state transition times of the currently active automata . our analysis of the minimal occurrence algorithm also leads to an analysis and proof for counting non - overlapped occurrences ( no and no - x ) as well . another advantage of our proof strategy is that it may be generalized to the case of episodes with general partial orders . ( we briefly discuss this in section [ sec : discussion ] ) . [ lemma : power ] suppose @xmath46 is an earliest transiting occurrence of an @xmath12-node episode @xmath13 . if @xmath216 is any general occurrence such that @xmath217 , then @xmath218 @xmath219 . this lemma follows easily from the definition of the lexicographic ordering , @xmath120 , and the definition of earliest transiting occurrence . [ remark:4 - 1 ] recall that @xmath220 is the @xmath40 earliest transiting ( et ) occurrence of an episode . thus , by definition , @xmath221 and @xmath222 whenever @xmath223 . hence , from the above lemma , we have @xmath224 for all @xmath49 and @xmath223 . in particular , we have , @xmath225 and @xmath226 , for an @xmath12-node episode . the main idea of our proof is that to find all minimal windows of an episode , it is enough to capture a certain subset of earliest transiting occurrences . [ lemma : minimal_et ] an earliest transiting ( et ) occurrence @xmath220 , of an @xmath12-node episode , is not a minimal occurrence if and only if @xmath227 . the ` if ' part follows easily from remark [ remark:4 - 1 ] . for the ` only if ' part , let us denote by @xmath228=[h_i^e(v_1),h_i^e(v_n)]$ ] the window of @xmath220 . given that @xmath156 is not a minimal window , we need to show that @xmath227 . since @xmath156 is not a minimal window , one of its proper sub - windows contains an occurrence , say , @xmath46 , of this episode . that means if @xmath46 starts at @xmath229 then it must end before @xmath230 . but , since @xmath220 is earliest transiting , any occurrence starting at the same event as @xmath220 can not end before @xmath220 . thus we must have @xmath231 . this means , by lemma [ lemma : power ] , since @xmath220 is earliest transiting , we can not have @xmath232 . since the window of @xmath46 has to be contained in the window of @xmath220 , we thus have @xmath233 . by definition , @xmath234 will start at the earliest possible position after @xmath220 . since there is an occurrence starting with @xmath207 we must have @xmath235 . now , since @xmath234 is earliest transiting , it can not end after @xmath46 . thus we must have @xmath236 . also , @xmath234 can not end earlier than @xmath220 because both are earliest transiting . thus , we must have @xmath237 . this completes proof of lemma . [ remark:4 - 2 ] this lemma shows that any et occurrence @xmath220 such that @xmath238 is a minimal occurrence . the converse is also true . consider a minimal window @xmath239 $ ] . since this is a minimal window , there is an occurrence ( and hence an et occurrence ) starting at @xmath229 . denote this et occurrence by @xmath220 . we know @xmath240 because @xmath156 is a minimal window . then the next et occurrence @xmath234 has to start after @xmath229 and has to end beyond @xmath230 because @xmath156 is minimal . thus we have @xmath241 . now we are ready to prove correctness of the mo algorithm . consider _ algorithm [ algo : unified ] _ operating in the mo(minimal occurrence ) mode for tracking occurrences of an @xmath12-node episode @xmath13 . since transit is always true in the mo mode , all automata would be tracking et occurrences . since copy - automaton is true in mo mode whenever an automaton transits out of start state , we will always have an automaton in the start state . _ this , along with the fact that transit is always true , implies that the @xmath40 initialized automaton would be tracking @xmath220 , the @xmath40 et occurrence . _ let us denote by @xmath242 the @xmath40 initialized automaton . however , since join - automaton is also always true , not all automata ( initialized for this episode ) would result in incrementing the frequency ; some of them would be removed when one automaton transits into a state already occupied by some other automaton . in view of lemma [ lemma : minimal_et ] and remark [ remark:4 - 2 ] , if we show that the automaton @xmath242 results in increment of frequency if and only if @xmath220 , the occurrence tracked by it , is such that @xmath243 , then , the proof of correctness of mo algorithm is complete . [ lemma : minimal_algorithm ] in the mo algorithm the @xmath40 automaton that was initialized for @xmath13 , referred to as @xmath244 , contributes to the frequency count iff @xmath245 . @xmath246 the last step follows because both @xmath220 and @xmath234 are et occurrences and hence @xmath247 implies @xmath248 . conversely , we have @xmath249 the first step follows because , if @xmath244 contributes to the frequency then no automaton initialized after it would ever come to the same state occupied by it and since all occurrences tracked are earliest transiting , this must mean @xmath250 , @xmath251 . this completes proof of the lemma . another interesting observation is that if @xmath133 is minimal , then it is non - interleaved with @xmath252 . suppose @xmath133 is minimal and @xmath133 is not non - interleaved with @xmath252 . since @xmath133 is minimal , we have @xmath253 . if @xmath220 is not non - interleaved with @xmath234 , there exists a @xmath254 such that @xmath255 . thus we must have @xmath256 . but this can not be because @xmath257 is the earliest @xmath103 $ ] after @xmath258 and if it is also after @xmath259 then the fact that both @xmath220 and @xmath234 are et occurrences should mean @xmath260 which contradicts that @xmath220 is minimal . hence @xmath220 and @xmath252 are non - interleaved . thus , given the sequence of minimal windows , the earliest transiting occurrences from each of these minimal windows gives a sequence of ( same number of ) non - interleaved occurrences . this leads to @xmath261 as stated earlier in ( [ eq : all - f - relationships ] ) . the no - x algorithm can be viewed as a slight modification to the mo algorithm . as in the mo algorithm , we always have an automaton in the start state and all automata make transitions as soon as possible and when an automaton transits into a state occupied by another , the older one is removed . however , in the no - x algorithm , the increment - freq variable is true only when we have an occurrence satisfying @xmath48 constraint . hence , to start with , we look for the first minimal occurrence which satisfies the expiry time constraint and increment frequency . at this point , ( unlike in the mo algorithm ) we terminate all automata except the one in the start state since we are trying to construct a non - overlapped set of occurrences . then we look for the next earliest minimal occurrence ( which will be non - overlapped with the first one ) satisfying expiry time constraint and so on . since minimal occurrences locally have the least time span , this strategy of searching for minimal occurrences satisfying expiry time constraint in a non - overlapped fashion is quite intuitive . let @xmath262 denote the sequence of occurrences tracked by the no - x algorithm ( for an @xmath12-node episode ) . then the following property of @xmath263 is obvious . [ property : algo3 ] @xmath264 is the earliest minimal occurrence satisfying expiry time constraints . @xmath265 is the first minimal occurrence ( satisfying expiry time constraint ) that starts after @xmath266 . there is no minimal occurrence satisfying expiry time constraint which starts after @xmath267 . [ theorem : maximality non - overlap constraints ] _ @xmath263 is a maximal non - overlapped sequence satisfying expiry time constraint @xmath48 . _ consider any other set of non - overlapped occurrences satisfying expiry constraints , @xmath81 = @xmath268 such that @xmath269 . let @xmath270 then we first show @xmath271 suppose @xmath272 . consider the earliest transiting occurrence @xmath273 starting from @xmath274 . this ends at or before @xmath275 by lemma [ lemma : power ] . among all et occurrences that end at the same event as @xmath273 , the last one ( under the lexicographic ordering ) is a minimal occurrence by lemma [ lemma : minimal_et ] . its window is contained in that of @xmath276 which satisfies the expiry time constraint . hence we have found a minimal occurrence satisfying expiry constraint ending before @xmath264 which contradicts the first statement of property [ property : algo3 ] . hence @xmath277 . now applying the same argument to the data stream starting with the first event after @xmath278 , we get @xmath279 and so on and thus can conclude @xmath280 . this shows that no other set of non - overlapped occurrences can have more number of occurrences than those in @xmath263 . hence , @xmath263 is maximal . if we choose @xmath48 equal to the time span of the data stream , the no - x algorithm reduces to the no - i algorithm because every occurrence satisfies expiry constraint . hence proof of correctness of no - i algorithm is immediate . we now explain the relation between the sets of occurrences tracked by the no and no - i algorithms . as proved in @xcite the no algorithm ( which uses one automaton per episode ) , tracks a maximal non - overlapped sequence of occurrences , say , @xmath281 . since the no - i algorithm has no expiry time constraint , it also tracks a maximal set of non - overlapped occurrences . among all the et occurrences that end at @xmath282 , let @xmath283 be the last one ( as per the lexicographic ordering ) . then the @xmath40 occurrence tracked by the no - i algorithm would be @xmath283 as we show now . since @xmath284 would be the first et occurrence , it is clear from our discussion in the previous subsection that the first occurrence tracked by the mo algorithm would be @xmath285 . as is easy to see , the mo and no - i algorithms would be identical till the first time an automaton reaches the accepting state . hence @xmath285 would be the first occurrence tracked by the no - i algorithm . now the no - i algorithm would remove all automata except for the one in the start state . hence , it is as if we start the algorithm with data starting with the first event after @xmath286 . now , by the property of no algorithm , @xmath287 would be the first et occurrence in this data stream and hence @xmath288 would be the first minimal window here . hence it is the second occurrence tracked by no - i and so on . the above also shows that each occurrence tracked by the no - i algorithm is also tracked by the mo algorithm and hence we have @xmath289 as stated in ( [ eq : all - f - relationships ] ) . @xmath290 is also a maximal set of non - overlapping minimal windows as discussed in @xcite . the algorithm ni which counts non - interleaved occurrences is different from all the ones discussed so far because it does not track et occurrences . here also we always have an automaton waiting in the start state . however , the transitions are conditional in the sense that the @xmath40 created automaton makes a transition from state @xmath291 to @xmath102 provided the @xmath214 created automaton is past state @xmath102 after processing the current event . this is because we want the @xmath40 automata to track an occurrence non - interleaved with the occurrence tracked by @xmath214 automaton . let @xmath292 be the sequence of occurrences tracked by ni . from the above discussion it is clear that it has the following property ( while counting occurrences of @xmath13 ) . [ property : algo1 ] _ @xmath293 is the first or earliest occurrence ( of @xmath13 ) . for all @xmath294 and @xmath295 , @xmath296 is the first occurrence of @xmath297 $ ] at or after @xmath298 ; and @xmath299 is the earliest occurrence of @xmath300 $ ] after @xmath301 . there is no occurrence of @xmath13 beyond @xmath302 which is non - interleaved with it . _ the proof that @xmath303 is a maximal non - interleaved sequence is very similar in spirit to that of the no - x algorithm . as earlier , we can show that given an arbitrary sequence of non - interleaved occurrences @xmath81 = @xmath268 , we have @xmath304 and hence get the correctness proof of ni algorithm . it is easy to verify the correctness of the do algorithm also along similar lines . it appears difficult to extend both the ni and do algorithms to incorporate expiry time constraints . for this we should track a set of occurrences @xmath305 of @xmath13 , where @xmath75 is the first occurrence satisfying @xmath48 and @xmath76 is the next earliest occurrence satisfying @xmath48 that is non - interleaved with ( or distinct from , in case of do ) @xmath75 and so on . note that this @xmath76 need not have to be the earliest occurrence non - overlapped with @xmath75 . at present , there are no algorithms for counting non - interleaved or distinct occurrences satisfying an expiry time constraint . before ending this section , we briefly outline what needs to be done when the data stream contains multiple events having the same time of occurrence . an important thing to note is that two events having the same time of occurrence can not be a part of a serial episode occurrence . hence , each automata can at most accept one event from a set of events having the same occurrence time . with this condition , the do , ao and hd algorithms go through as before . one would need to process the set of events having the same occurrence time together and allow all the permissible automata to make a one step transition first as done using @xmath306 list in @xcite . after this , before processing the set of events with the next occurrence time , we would need to do the multiple automata check for the various candidate episodes and delete the appropriate older automata for algorithms mo , mo - x , no - i and no - x . for the non - interleaved algorithm , one needs to actually back track the transitions which resulted in two automata to coalesce . in this section , we discuss the anti - monotonicity properties of the various frequency counts , which in - turn are exploited by their respective candidate generation steps in the apriori - style level - wise procedure for frequent episode discovery . it is well known that the windows - based@xcite , non - overlapped@xcite and total@xcite frequency measures satisfy the anti - monotonicity property that _ all subepisodes of a frequent episode are frequent_. one can verify that the same holds for the distinct occurrences based frequency too . it has been pointed out in @xcite that the head frequency does not satisfy this anti - monotonicity property . for an episode @xmath13 , in general , only the subepisodes involving @xmath213 $ ] are as frequent as @xmath13 under the head count . in a level - wise apriori - based episode discovery , the candidate generation for the head frequency count would exploit the condition that if an @xmath12-node episode is frequent , then all @xmath160-node subepisodes that include @xmath213 $ ] have to be frequent . the head frequency definition has some limitations in the sense that the frequency of the @xmath160-node suffix subepisode - node episode @xmath213 \rightarrow \alpha[2 ] \rightarrow \cdots \rightarrow \alpha[n]$ ] , its @xmath307-node _ prefix subepisode _ is @xmath213 \rightarrow \alpha[2 ] \rightarrow \cdots \rightarrow \alpha[k]$ ] and its @xmath308-node suffix subepisode is @xmath309 \rightarrow \alpha[k+2 ] \rightarrow \cdots \rightarrow \alpha[n]$ ] for @xmath310 . ] can be arbitrarily low . consider the event stream with @xmath311 @xmath109s followed by a @xmath111 and @xmath113 . suppose all occurrences of @xmath312 satisfy the expiry constraint @xmath48 . even though there are @xmath311 occurrences of @xmath313 , there is only one occurrence of @xmath314 . this can be a problem when one desires that the frequent episodes capture repetitive causative influences . like the head frequency , the minimal occurrences ( windows ) and the non - interleaved occurrences also do not satisfy the anti - monotonicity property that all subepisodes are at least as frequent as the corresponding episode . however , the @xmath160-node prefix and suffix subepisodes are at least as frequent as the episode as we show below . for an example , consider a data stream where successive events are given by @xmath315 . even though there are two minimal windows ( and two non - interleaved occurrences ) of @xmath316 , there is only one minimal window ( and one non - interleaved occurrence ) of each of the non - prefix and non - suffix subepisodes @xmath317 and @xmath318 . however , the situation here is not as bad as that for head frequency because all such subepisodes will have at least as many distinct occurrences as the number of minimal or non - interleaved occurrences of the episode , at least in case of injective episodes . ( note that this example is that of an injective episode ) . this is because , in case of injective episodes , the number of distinct occurrence is always greater than the non - interleaved count , which in - turn is greater than the minimal windows count . hence , given that there are @xmath206 non - interleaved or minimal occurrences of an injective episode @xmath13 , there are at least @xmath206 distinct occurrences of @xmath13 too . since the distinct occurrences based frequency satisfies the original anti - monotonicity property , all subepisodes of @xmath13 too will have at least @xmath206 distinct occurrences . we now formally prove the anti - monotonicity property for minimal and non - interleaved occurrences based frequencies . if a n - node serial episode @xmath13 has a frequency @xmath206 in the minimal or the non - interleaved sense , then its @xmath160-prefix subepisode @xmath319 and suffix subepisode @xmath320 have a frequency of at least @xmath206 . consider a minimal window of the episode @xmath13 , @xmath321 $ ] . consider the earliest occurrence @xmath322 of the prefix subepisode starting from @xmath323 and let @xmath324 be its window . any proper sub - window of @xmath324 starting at @xmath323 and containing an occurrence of @xmath325 contradicts lemma [ lemma : power ] . a proper sub - window of @xmath324 containing an occurrence of @xmath325 starting after @xmath323 would contradict the minimality of @xmath156 itself . hence @xmath324 is a minimal window of @xmath325 starting at @xmath323 . we hence conclude that @xmath325 has a frequency of at least @xmath206 . a similar proof works for the suffix subepisode by considering the window of the last occurrence @xmath326 of the suffix subepisode ending at @xmath327 . let @xmath328 be a maximal non - interleaved sequence . from each occurrence @xmath329 , we choose the sub - occurrence @xmath330 $ ] , of @xmath325 . it is easy to see that this new sequence of occurrences @xmath331 forms a non - interleaved sequence . hence the frequency of @xmath325 is at least @xmath206 . a similar argument works for the suffix episode . hence , for every episode @xmath13 , we extract its @xmath160 suffix , go down the candidate list and search for a block of episodes whose @xmath332 prefix matches this suffix . we form candidates as many as the number of episodes in this matching block . this kind of candidate generation has already been reported in the literature in @xcite , @xcite and @xcite in the context of sequences under inter - event time constraints . the framework of frequent episodes in event streams is a very useful data mining technique for unearthing temporal dependencies from data streams in many applications . the framework is about a decade old and many different frequency measures and associated algorithms have been proposed over the last ten years . in this paper we have presented a generic automata - based algorithm for obtaining frequencies of a set of candidate episodes . this method unifies all the known algorithms in the sense that we can particularize our algorithm ( by setting values for a set of variables ) for counting frequent episodes under any of the frequency measures proposed in literature . as we showed here , this unified view gives useful insights into the kind of occurrences counted under different frequency definitions and thus also allows us to prove relations between frequencies of an episode under different frequency definitions . our view also allows us to get correctness proofs for all algorithms . we introduced the notion of earliest transiting occurrences and , using this concept , are able to get simple proofs of correctness for most algorithms . this has also allowed us to understand the kind of anti - monotonicity properties satisfied by different frequency measures . while the main contribution of this paper is this unified view of all frequency counting algorithms , some of the specific results presented here are also new . the relationships between different frequencies of an episode ( cf . eqn [ eq : all - f - relationships ] ) , is proved here for the first time . the distinct - occurrences based frequency and an automata - based algorithm for it are novel . the specific proof of correctness presented here for minimal occurrences is also novel . also , the correctness proofs for non - overlapped occurrences based frequency counting under expiry time constraint has been provided here for the first time . in this paper we have considered only the case of serial episodes . this is because , at present , there are no algorithms for discovering general partial orders under the various frequency definitions . however , all counting algorithms explained here for serial episodes can be extended to episodes with a general partial order structure . we can come up with a similar finite state automata(fsa ) which track the earliest transiting occurrences of an episode with a general partial order structure @xcite . for example , consider a partial order episode @xmath333 which represents @xmath109 and @xmath111 occurring in any order followed by a @xmath113 . in order to track an occurrence of such a pattern , the initial state has to wait for either of @xmath109 and @xmath111 . on seeing an @xmath109 it goes to state-1 where it waits only for a @xmath111 ; on the other hand , on seeing a @xmath111 first it moves to state-2 where it waits only for an @xmath109 . then on seeing a @xmath111 in state-1 or seeing a @xmath109 in state-2 it moves into state-3 where it waits for a @xmath113 and so on . thus , in each state in such a fsa , in general , we wait for any of a set of event types ( instead of a single event for serial episodes ) and a given state will now branch out into different states on different event types . with such a fsa technique it is possible to generalize the method presented here so that we have algorithms for counting frequencies of general partial order episodes under different frequencies . the proofs presented here for serial episodes can also be extended for general partial order episodes . while it seems possible , as explained above , to generalize the counting schemes to handle general partial order episodes , it is not obvious what would be an appropriate candidate generation scheme for general partial order episodes under different frequency definitions . this is an important direction for future work . in this paper , we have considered only expiry time constraint which prescribes an upper bound on the span of the occurrence . it would be interesting to see under what other time constraints ( e.g. , gap constraints ) , design of counting algorithms under this generic framework is possible . also , some unexplored choice of the boolean conditions in the proposed generic algorithm may give rise to algorithms for new useful frequency measures . this is also a useful direction of research to explore . bouchra bouqata , christopher d. caraothers , boleslaw k. szymanski , and mohammed j. zaki . vogue : a novel variable order - gap state machine for modeling sequences . in _ proc . european conf . principles and practice of knowledge discovery in databases ( pkdd06 ) _ , pages 4254 , sep 2006 . iwanuma k. , takano y. , and nabeshima h. on anti - monotone frequency measures for extracting sequential patterns from a single very - long sequence . in _ proc . ieee conf . cybernetics and intelligent systems _ , pages 213217 , dec 2004 . srivatsan laxman , p. s. sastry , and k. p. unnikrishnan . a fast algorithm for finding frequent episodes in event streams . in _ proc . acm sigkdd intl conf . knowledge discovery and data mining ( kdd07 ) _ , pages 410419 , aug 2007 . srivatsan laxman , vikram tankasali , and ryen w. white . stream prediction using a generative model based on frequent episodes in event sequences . in _ proc . acm sigkdd intl conf . knowledge discovery and data mining ( kdd09 ) _ , pages 453461 , jul 2008 . nicolas meger and christophe rigotti . constraint - based mining of episode rules and optimal window sizes . in _ proc . european conf . principles and practice of knowledge discovery in databases ( pkdd04 ) _ , september 2004 . anny nag and ada fu , wai - chee . mining freqeunt episodes for relating financial events and stock trends . in _ proc . pacific - asia conf . knowledge discovery and data mining , ( pakdd 2003 ) _ , pages 2739 , 2003 . min - feng wang , yen - ching wu , and meng - feng tsai . exploiting frequent episodes in weighted suffix tree to improve intrusion detection system . in _ proc . intl conf . advanced information networking and applications(aina08 ) _ , pages 12461252 , mar 2008 .

semi - discrete runge - kutta schemes for nonlinear diffusion equations of parabolic type are analyzed . conditions are determined under which the schemes dissipate the discrete entropy locally . the dissipation property is a consequence of the concavity of the difference of the entropies at two consecutive time steps . the concavity property is shown to be related to the bakry - emery approach and the geodesic convexity of the entropy . the abstract conditions are verified for quasilinear parabolic equations ( including the porous - medium equation ) , a linear diffusion system , and the fourth - order quantum diffusion equation . numerical experiments for various runge - kutta finite - difference discretizations of the one - dimensional porous - medium equation show that the entropy - dissipation property is in fact global .

1506.07040

evolution equations often contain some structural information reflecting inherent physical properties such as positivity of solutions , conservation laws , and entropy dissipation . numerical schemes should be designed in such a way that these structural features are preserved on the discrete level in order to obtain accurate and stable algorithms . in the last decades , concepts of structure - preserving schemes , geometric integration , and compatible discretization have been developed @xcite , but much less is known about the preservation of entropy dissipation and large - time asymptotics . entropy - stable schemes were derived by tadmor already in the 1980s @xcite in the context of conservation laws , thus without ( physical ) diffusion . later , entropy - dissipative schemes were developed for ( finite - volume ) discretizations of diffusion equations in @xcite . in @xcite , a finite - volume scheme which preserves the gradient - flow structure and hence the entropy structure is proposed . all these schemes are based on the implicit euler method and are of first order ( in time ) only . higher - order time schemes with entropy - dissipating properties are investigated in very few papers . a second - order predictor - corrector approximation was suggested in @xcite , while higher - order semi - implicit runge - kutta ( dirk ) methods , together with a spatial fourth - order central finite - difference discretization , were investigated in @xcite . in @xcite , multistep time approximations were employed , but they can be at most of second order and they dissipate only one entropy and not all functionals dissipated by the continuous equation . in this paper , we remove these restrictions by investigating time - discrete runge - kutta schemes of order @xmath0 for general diffusion equations . we stress the fact that we are interested in the analysis of entropy - dissipating schemes by `` translating '' properties for the continuous equation to the semi - discrete level , i.e. , we study the stability of the schemes . however , we will not investigate convergence , stiffness , or computational issues here ( see e.g. @xcite ) . more precisely , we consider time discretizations of the abstract cauchy problem @xmath1 = 0 , \quad t>0 , \quad u(0)=u^0,\ ] ] where @xmath2 is a ( differential ) operator defined on @xmath3 and @xmath4 is a banach space with dual @xmath5 . in this paper , we restrict ourselves to diffusion operators @xmath6 $ ] defined on some sobolev space with solutions @xmath7 , which may be vector - valued . a typical example is @xmath6={\operatorname{div}}(a(u)\na u)$ ] defined on @xmath8 with domain @xmath9 , where @xmath10 is a smooth function ( see section [ sec.de ] ) . equation often possesses a lyapunov functional @xmath11=\int_\omega h(u)dx$ ] ( here called _ entropy _ ) , where @xmath12 , such that @xmath13 = \int_\omega h'(u)\pa_t u dx = -\int_\omega h'(u)a[u ] dx \le 0,\ ] ] at least when the _ entropy production _ dx$ ] is nonnegative , here , @xmath15 is the derivative of @xmath16 and @xmath17 $ ] is interpreted as the inner product of @xmath18 and @xmath6 $ ] in @xmath19 . furthermore , if @xmath16 is convex , the convex sobolev inequality @xmath14dx \ge \kappa h[u]$ ] for some @xmath20 may hold @xcite , which implies that @xmath21 and hence exponential convergence of @xmath11 $ ] to zero with rate @xmath22 . the aim is to design a higher - order time - discrete scheme which preserves this entropy - dissipation property . to this end , we propose the following semi - discrete runge - kutta approximation of : given @xmath23 , define @xmath24 , \quad i=1,\ldots , s,\ ] ] where @xmath25 are the time steps , @xmath26 is the uniform time step size , @xmath27 approximates @xmath28 , and @xmath29 denotes the number of runge - kutta stages . since the cauchy problem is autonomous , the knots @xmath30 are not needed here . in concrete examples ( see below ) , @xmath27 are functions from @xmath31 to @xmath19 . if @xmath32 for @xmath33 , the runge - kutta scheme is explicit , otherwise it is implicit and a nonlinear system of size @xmath34 has to be solved to compute @xmath35 . we assume that scheme is solvable for @xmath36 . given @xmath12 , we wish to determine conditions under which the functional @xmath37 = \int_\omega h(u^k(x))dx\ ] ] is dissipated by the numerical scheme , @xmath38 + \tau\int_\omega a[u^k]h'(u^k)dx \le h[u^{k-1 } ] , \quad k\in{{\mathbb n}}.\ ] ] in many examples ( see below ) , @xmath39h'(u^k)dx\ge 0 $ ] and thus , the function @xmath40 $ ] is decreasing . such a property is the first step in proving the preservation of the large - time asymptotics of the numerical scheme ( see remark [ rem.exp ] ) . our main results , stated on an informal level , are as follows : a. we determine an abstract condition under which the discrete entropy - dissipation inequality holds for sufficiently small @xmath41 . this condition is made explicit for special choices of @xmath42 and @xmath16 , yielding entropy - dissipative implicit or explicit runge - kutta schemes of any order . b. numerical experiments for the porous - medium equation indicate that @xmath43 may be chosen independent of the time step @xmath44 , thus yielding discrete entropy dissipation for all discrete times . c. we show that for runge - kutta schemes of order @xmath45 , the abstract condition in ( i ) is exactly the criterion of liero and mielke @xcite to conclude geodesic 0-convexity of the entropy . in particular , it is related to the bakry - emery condition @xcite . let us describe the main results in more detail . we recall that the runge - kutta scheme is consistent if @xmath46 and @xmath47 . furthermore , if @xmath48 , it is at least of order two ( * ? ? ? we introduce the number @xmath49 which takes only three values : @xmath50 the _ first main result _ is an abstract entropy - dissipation property of scheme for entropies of type . [ thm.ed ] let @xmath51 , let @xmath2 be frchet differentiable with frchet derivative @xmath52:x\to x'$ ] at @xmath53 , and let @xmath54 be the runge - kutta solution to . suppose that @xmath55(a[u^k])+h''(u^k)(a[u^k])^2\big)dx > 0.\ ] ] then there exists @xmath41 such that for all @xmath56 , @xmath57 + \tau\int_\omega a[u^k]h'(u^k)dx \le h[u^{k-1}].\ ] ] we assume that the solutions to are sufficiently regular such that the integral can be defined . in the vector - valued case , @xmath58 is the hessian matrix and we interpret @xmath59)^2 $ ] as the product @xmath60^\top h''(u^k)a[u^k]$ ] . for runge - kutta schemes of order @xmath45 ( for which @xmath61 ) , the integral corresponds exactly to the second - order time derivative of @xmath62 $ ] for solutions @xmath63 to the _ continuous _ equation . observe that the entropy - dissipation estimate is only _ local _ , since the time step restriction depends on the time step @xmath44 . for implicit euler schemes ( and convex entropies @xmath16 ) , it is known that @xmath43 can be chosen independent of @xmath44 . for general runge - kutta methods , we can not prove rigorously that @xmath43 stays bounded from below as @xmath64 . however , our numerical experiments in section [ sec.num ] indicate that inequality holds for sufficiently small @xmath65 uniformly in @xmath44 . [ rem.exp ] if the convex sobolev inequality @xmath39h'(u^k)dx\ge \kappa h[u^k]$ ] holds for some constant @xmath20 and if there exists @xmath66 such that @xmath67 for all @xmath68 , we infer from that for @xmath69 , @xmath70 \le ( 1+\kappa\tau)^{-k}h[u^0 ] = \exp(-\eta\kappa t^k)h[u^0 ] , \quad\mbox{where } \eta = \frac{\log(1+\kappa\tau)}{\kappa\tau } < 1,\ ] ] which implies exponential decay of the discrete entropy with rate @xmath71 . this rate converges to the continuous rate @xmath22 as @xmath72 and therefore , it is asymptotically sharp . theorem [ thm.ed ] can be generalized to a larger class of entropies , namely to so - called _ first - order entropies _ @xmath73 = \int_\omega |\na f(u^k)|^2 dx,\ ] ] where , for simplicity , we consider only the scalar case with @xmath74 . an important example is the fisher information with @xmath75 . [ thm.ed2 ] let @xmath76 , let @xmath2 be frchet differentiable , and let @xmath54 be the runge - kutta solution to . assume that the boundary condition @xmath77 on @xmath78 is satisfied . furthermore , suppose that @xmath79|^2 - c_{\rm rk}\delta f(u^k)f'(u^k)da[u^k](a[u^k ] ) \\ & { } - \delta f(u^k)f''(u^k)(a[u^k])^2\big)dx > 0 . \end{aligned}\ ] ] then there exists @xmath41 such that for all @xmath56 , @xmath80 + \tau\int_\omega a[u^k]f'(u^k)dx \le f[u^{k-1}].\ ] ] the key idea of the proof of theorem [ thm.ed ] ( and similarly for theorem [ thm.ed2 ] ) is a concavity property of the difference of the entropies at two consecutive time steps with respect to the time step size @xmath81 . to explain this idea , let @xmath82 be fixed and introduce @xmath83 . clearly , @xmath84 . a formal taylor expansion of @xmath85-h[v(\tau)]$ ] yields @xmath70-h[u^{k-1 } ] = g(\tau ) = g(0 ) + \tau g'(0 ) + \frac{\tau^2}{2}g''(\xi^k),\ ] ] where @xmath86 . a computation , made explicit in section [ sec.meth ] , shows that @xmath87h'(u^k)dx$ ] and @xmath88 . now , if @xmath89 , there exists @xmath41 such that @xmath90 for @xmath91 $ ] and in particular @xmath92 . consequently , @xmath93 , which equals . the definition of @xmath94 assumes implicitly that is _ backward _ solvable . we prove in proposition [ prop.back ] below that this property holds if the operator @xmath42 is a smooth self - mapping on @xmath4 . [ rem.tauk]since @xmath54 is expected to converge to the stationary solution , @xmath95 . thus , in principle , for larger values of @xmath44 , we expect that @xmath43 becomes smaller and smaller , thus restricting the choice of time step sizes @xmath81 . however , practically , the situation is better . for instance , for the implicit euler scheme , if @xmath16 is convex , we obtain @xmath70 - h[u^{k-1 } ] \le \int_\omega h'(u^k)(u^k - u^{k-1})dx = -\tau\int_\omega h'(u^k)a[u^k]dx\ ] ] for _ any _ value of @xmath65 . moreover , for other ( higher - order ) runge - kutta schemes , the numerical experiments in section [ sec.num ] indicate that there exists @xmath66 such that @xmath90 holds for all @xmath96 $ ] uniformly in @xmath68 . in this situation , inequality holds for all @xmath97 , and thus our estimate is global . in fact , the function @xmath98 is numerically even nonincreasing in some interval @xmath99 $ ] but we are not able to prove this analytically . the _ second main result _ is the specification of the abstract conditions and for a number of examples : a quasilinear diffusion equation , porous - medium or fast - diffusion equations , a linear diffusion system , and the fourth - order derrida - lebowitz - speer - spohn equation ( see sections [ sec.de]-[sec.dlss ] for details ) . for instance , for the porous - medium equation @xmath100 we show that the runge - kutta scheme scheme satisfies @xmath70 + \tau\beta\int_\omega(u^k)^{\alpha+\beta-2}|\na u^k|^2 dx \le h[u^{k-1 } ] , \quad\mbox{where } h[u]=\frac{1}{\alpha(\alpha+1)}\int_\omega u^{\alpha+1 } dx,\ ] ] for @xmath56 and all @xmath101 belonging to some region in @xmath102 ( see figure [ fig.0th ] below ) . for @xmath103 , we write @xmath11=\int_\omega u(\log u-1)dx$ ] . in one space dimension and for runge - kutta schemes of order @xmath45 , this region becomes @xmath104 , which is the same condition as for the continuous equation ( except the boundary values ) . furthermore , the first - order entropy is dissipated for runge - kutta schemes of order @xmath45 , in one space dimension , @xmath80 + \tau c_{\alpha,\beta}\int_\omega ( u^k)^{\alpha+\beta-2}(u^k)_{xx}^2 dx \le f[u^{k-1 } ] , \quad\mbox{where } f[u]=\int_\omega ( u^{\alpha/2})_x^2 dx,\ ] ] for @xmath56 and all @xmath101 belonging to the region shown in figure [ fig.1st ] below , and @xmath105 is some constant . this region is smaller than the region of admissible values @xmath101 for the continuous entropy . the borders of that region are indicated in the figure by dashed lines . the proof of the above results , and namely of @xmath89 , is based on systematic integration by parts @xcite . the idea of the method is to formulate integration by parts as manipulations with polynomials and to conclude the inequality @xmath89 from a polynomial decision problem . this problem can be solved directly or by using computer algebra software . our _ third main result _ is the relation to geodesic 0-convexity of the entropy and the bakry - emery approach when @xmath61 ( runge - kutta scheme of order @xmath45 ) . liero and mielke formulate in @xcite the abstract cauchy problem as the gradient flow @xmath106dh[u ] , \quad t>0 , \quad u(0)=u^0,\ ] ] where the onsager operator @xmath107 $ ] describes the sum of diffusion and reaction terms . for instance , if @xmath6={\operatorname{div}}(a(u)\na u)$ ] , we can write @xmath6={\operatorname{div}}(a(u)h''(u)^{-1}\na h'(u))$ ] and thus , identifying @xmath18 and @xmath108 $ ] , we have @xmath107\xi={\operatorname{div}}(a(u)h''(u)^{-1}\na\xi)$ ] . it is shown in @xcite that the entropy @xmath109 is geodesic @xmath110-convex if the inequality @xmath111k[u]\xi\rangle - \frac12\langle\xi , dk[u]a[u]\xi\rangle \ge \lambda\langle\xi , k[u]\xi\rangle\ ] ] holds for all suitable @xmath112 and @xmath113 . we will prove in section [ sec.meth ] that @xmath114 hence , if @xmath115 then with @xmath116 is satisfied for @xmath117 and @xmath118 , yielding geodesic @xmath119-convexity for the semi - discrete entropy . moreover , if @xmath120 then we obtain geodesic @xmath110-convexity . since @xmath121/dt$ ] and @xmath122/dt^2 $ ] in the continuous setting , the inequality @xmath120 can be written as @xmath123 \ge -\lambda \frac{dh}{dt}[u],\ ] ] which corresponds to a variant of the bakry - emery condition @xcite , yielding exponential convergence of @xmath11 $ ] ( if @xmath67 for all @xmath44 ) . thus , our results constitute a first step towards a _ discrete bakry - emery approach_. the paper is organized as follows . the abstract method , i.e. the proof of backward solvability and of theorems [ thm.ed ] and [ thm.ed2 ] , is presented in section [ sec.meth ] . the method is applied in the subsequent sections to a scalar diffusion equation ( section [ sec.de ] ) , the porous - medium equation ( section [ sec.pme ] ) , a linear diffusion system ( section [ sec.sys ] ) , and the fourth - order derrida - lebowitz - speer - spohn equation ( section [ sec.dlss ] ) . finally , section [ sec.num ] is devoted to some numerical experiments showing that @xmath98 is negative in some interval @xmath99 $ ] . in this section , we show that the runge - kutta scheme is backward solvable if @xmath42 is a self - mapping and we prove theorems [ thm.ed ] and [ thm.ed2 ] . [ prop.back ] let @xmath124 , where @xmath4 is some banach space , and let @xmath125 be a self - mapping . then there exists @xmath126 , a neighborhood @xmath127 of @xmath128 , and a function @xmath129 such that holds for @xmath130 . moreover , @xmath131 , \quad\mbox{and}\quad v''(0)=c_{\rm rk}da[u](a[u]).\ ] ] the self - mapping assumption is strong for differential operators @xmath42 but it is somehow natural in the context of runge - kutta methods and valid for smooth solutions . the idea of the proof is to apply the implicit function theorem in banach spaces ( see ( * ? ? ? * corollary 15.1 ) ) . to this end , we set @xmath82 and define the mapping @xmath132 by @xmath133 , \quad i=1,\ldots , s.\end{aligned}\ ] ] the frchet derivative of @xmath134 in the direction of @xmath135 , where @xmath136 , reads as @xmath137 \bigg(v_h + \tau_h\sum_{j=1}^s a_{ij}k_{j } + \tau\sum_{j=1}^s a_{ij}k_{hj}\bigg ) , \end{aligned}\ ] ] where @xmath138 . let @xmath139 and @xmath140,\ldots ,- a[u],u)$ ] . then @xmath141 and @xmath142(v_h ) , \quad i=1,\ldots , s.\ ] ] the mapping @xmath143 is clearly an isomorphism from @xmath144 onto @xmath144 . by the implicit function theorem , there exist an interval @xmath145 , a neighborhood @xmath146 of @xmath147 , and a function @xmath148 such that @xmath149,\ldots ,- a[u],u)$ ] and @xmath150 for all @xmath151 . implicit differentiation of @xmath150 yields @xmath152 \bigg(v'(\tau ) + \sum_{j=1}^s a_{ij } k_j(\tau ) + \tau\sum_{j=1}^s a_{ij } k_j'(\tau)\bigg),\end{aligned}\ ] ] where @xmath138 and @xmath151 . using @xmath47 and @xmath46 , we infer that @xmath153 = a[u ] , \nonumber \\ k_i'(0 ) & = -da[u]\bigg(a[u ] - \sum_{j=1}^s a_{ij}a[u]\bigg ) = -(1-c_i)da[u](a[u ] ) . \label{aux1}\end{aligned}\ ] ] differentiating @xmath154 twice leads to @xmath155 because of , this reads at @xmath156 as @xmath157(a[u ] ) = c_{\rm rk}da[u](a[u]).\ ] ] this finishes the proof . we prove now theorems [ thm.ed ] and [ thm.ed2 ] . we set @xmath82 . by proposition [ prop.back ] , there exists a backward solution @xmath158 such that @xmath84 , @xmath159 $ ] , and @xmath160(a[u])$ ] . furthermore , the function @xmath161 satisfies @xmath162 , @xmath163dx , \\ g''(0 ) & = -\int_\omega\big(h'(v(0))v''(0 ) + h''(v(0))v'(0)^2\big)dx \\ & = -\int_\omega\big(c_{\rm rk}h'(u)da[u](a[u ] ) + h''(u)(a[u])^2\big)dx = -i_0^k < 0,\end{aligned}\ ] ] using the assumption . by continuity , there exists @xmath164 such that @xmath165 for @xmath166 . then the taylor expansion @xmath167 concludes the proof . following the lines of the previous proof , it is sufficient to compute @xmath168 and @xmath169 , where now @xmath170 . using integration by parts and the boundary condition @xmath171 on @xmath78 , we compute @xmath172dx,\ ] ] since @xmath84 and @xmath159 $ ] . furthermore , again integrating by parts , @xmath173 since @xmath160(a[u])$ ] , this reduces at @xmath156 to @xmath174)|^2 - \delta f(u)f''(u)(a[u])^2 - c_{\rm rk}\delta f(u)f'(u)da[u](a[u])\big)dx.\end{aligned}\ ] ] this expression equals @xmath175 , and the result follows . finally , we show that @xmath169 for entropies is related to the geodesic convexity condition of @xcite . let @xmath6=k(u)dh[u]$ ] for some symmetric operator @xmath176 and frchet derivative @xmath108 $ ] , let @xmath177 be defined as in the proof of theorem [ thm.ed ] for a solution @xmath27 to the runge - kutta scheme of order @xmath45 , and let @xmath178 be given by . then @xmath179).\ ] ] the proof is just a ( formal ) calculation . recall that for runge - kutta schemes of order @xmath45 , we have @xmath61 . set @xmath82 and identify @xmath108 $ ] with @xmath180 . inserting the expression @xmath52(v)=dk[u](v)h'(u)+ k[u]h''(u)v$ ] into the definition of @xmath169 , we find that @xmath181(a[u])\rangle + \langle a[u],h''(u)a[u]\rangle \\ & = \big\langle\xi , dk[u](a[u])\xi + k[u]h''(u)a[u]\big\rangle + \langle a[u],h''(u)a[u]\rangle \\ & = \langle\xi , dk[u](k[u]\xi)\xi\rangle + \langle\xi , k[u]h''(u)k[u]\xi\rangle + \langle k[u]\xi , h''(u)k[u]\xi\rangle \\ & = \langle\xi , dk[u](k[u]\xi)\xi\rangle + 2\langle\xi , k[u]h''(u)k[u]\xi\rangle,\end{aligned}\ ] ] since @xmath107 $ ] is assumed to be symmetric . rearranging the terms , we obtain @xmath182(k[u]\xi)\xi\rangle + 2\langle\xi , k[u]h''(u)k[u]\xi\rangle - \langle\xi , dk[u](k[u]\xi)\rangle \\ & = 2\langle\xi , da[u](k[u]\xi)\xi\rangle - \langle\xi , dk[u](a[u])\rangle = 2m(u,\xi),\end{aligned}\ ] ] which proves the claim . in this section , we analyze time - discrete runge - kutta schemes of the diffusion equation @xmath183 with periodic or homogeneous neumann boundary conditions . this equation , also including a drift term , was analyzed in @xcite in the context of geodesic convexity . our results are similar to those in @xcite but we consider the time - discrete and not the continuous equation and we employ systematic integration by parts @xcite . setting @xmath184 , we can write the diffusion equation as a formal gradient flow : @xmath185 : = { \operatorname{div}}(\mu(u)\na h'(u ) ) , \quad t>0.\ ] ] we prove that the runge - kutta scheme dissipates all convex entropies subject to some conditions on the functions @xmath186 and @xmath16 . [ thm.de ] let @xmath187 be convex with smooth boundary . let @xmath54 be a sequence of ( smooth ) solutions to the runge - kutta scheme of the diffusion equation . let @xmath68 be fixed and @xmath27 be not equal to the constant steady state of . we suppose that for all admissible @xmath112 , it holds that @xmath188 , @xmath189 , @xmath190 then there exists @xmath41 such that for all @xmath191 , @xmath70 + \tau\int_\omega h''(u^k)a(u^k)|\na u^k|^2 dx \le h[u^{k-1}].\ ] ] conditions - correspond to ( 4.12 ) in @xcite . condition is satisfied for concave functions @xmath186 , except for the explicit euler scheme ( @xmath192 ) for which we need additionally @xmath193 . for the implicit euler scheme , we may allow even for nonconcave mobilities @xmath186 , e.g. @xmath194 for @xmath195 . according to theorem [ thm.ed ] , we only need to show that @xmath196 . to simplify , we set @xmath82 . first , we observe that the boundary condition @xmath197 on @xmath31 implies that @xmath198\cdot\nu$ ] on @xmath78 . using @xmath52(a[u ] ) = { \operatorname{div}}(a'(u)a[u]\na u+a(u)\na a[u])=\delta(a(u)a[u])$ ] , the abbreviation @xmath180 , and integration by parts , we compute @xmath199 ) + h''(u)\big({\operatorname{div}}(\mu(u)\na h'(u))\big)^2\big)dx \\ & = \int_\omega\big(c_{\rm rk}\na h'(u)\cdot\na(a(u)a[u ] ) - h''(u)\big(\mu'(u)\na u\cdot\na h'(u)+\mu(u)\delta h'(u)\big)^2\big)dx \\ & = -\int_\omega\bigg(c_{\rm rk}\delta\xi a(u)a[u ] + h''(u)\left(\frac{\mu'(u)}{h''(u)}|\na\xi|^2 + \mu(u)\delta\xi\right)^2 \bigg)dx.\end{aligned}\ ] ] the boundary integrals vanish since @xmath200\cdot\nu=0 $ ] on @xmath78 . replacing @xmath6 $ ] by @xmath201 and expanding the square , we arrive at @xmath202 where we have employed the identity @xmath203 and the abbreviations @xmath204 and @xmath205 . we apply now the method of systematic integration by parts @xcite . the idea is to identify useful integration - by - parts formulas and to add them to @xmath169 without changing the sign of @xmath169 . the first formula is given by @xmath206 where @xmath207 is an arbitrary ( smooth ) scalar function which still needs to be chosen , and @xmath208 is the unit matrix in @xmath209 . the left - hand side can be expanded as @xmath210 where we have set @xmath211 and @xmath212 . the boundary integral in becomes @xmath213 since @xmath207 , @xmath214 on @xmath78 , and it holds that @xmath215 on @xmath78 for all smooth functions satisfying @xmath214 on @xmath78 ( * ? ? ? 4.2 ) . here we need the convexity of @xmath31 . thus , the first integration - by - parts formula becomes @xmath216 the second formula reads as @xmath217 where @xmath218 is an arbitrary scalar function . the goal is to find functions @xmath207 and @xmath219 such that @xmath220 . according to @xcite , the computations simplify if we introduce the variables @xmath221 and @xmath222 satisfying @xmath223 the existence of @xmath221 follows from the inequality @xmath224 which is proven in ( * ? ? ? * lemma 2.1 ) . then @xmath225 where @xmath226 the aim now is to determine conditions on @xmath227 such that the polynomial @xmath228 is nonnegative as this implies that @xmath115 . in the general case , this leads to nonlinear ordinary differential equations for @xmath229 and @xmath218 which can not be easily solved . a possible solution is to require that the coefficients of the mixed terms vanish , i.e. @xmath230 , and that the remaining coefficients are nonnegative . the case @xmath231 being simpler than the general case ( since @xmath232 is not necessary ) , we assume that @xmath233 . then @xmath234 implies that @xmath235 . replacing @xmath236 by @xmath237 in @xmath238 gives @xmath239 on the other hand , replacing @xmath219 by @xmath240 in @xmath238 , we find that @xmath241 or , after integration , @xmath242 these functions have to satisfy the conditions @xmath243 note that @xmath244 and @xmath245 correspond to and , respectively . this shows that @xmath246 for all @xmath247 and @xmath115 . if @xmath248 , the nonnegative polynomial @xmath249 , which depends on @xmath250 via @xmath113 , has to vanish . in particular , @xmath251 in @xmath31 . as @xmath252 by assumption , @xmath253 for @xmath250 . this contradicts the hypothesis that @xmath112 is not a steady state . consequently , @xmath89 , and we finish the proof by setting @xmath254 . the results of the previous section can be applied in principle to the runge - kutta scheme for the porous - medium or fast - diffusion equation @xmath255 where @xmath256 . it can be seen that conditions - are not optimal for particular entropies . this is not surprising since we have neglected the mixed terms in the polynomial in ( i.e. @xmath230 ) which is not optimal . in this section , we make a different approach by making an ansatz for the functions @xmath229 and @xmath218 , considering both zeroth - order and first - order entropies . we prove the following result . [ thm.pm ] let @xmath187 be convex with smooth boundary . let @xmath54 be a sequence of ( smooth ) solutions to the runge - kutta scheme for . let the entropy be given by @xmath11=\alpha^{-1}(\alpha+1)^{-1}\int_\omega u^{\alpha+1}dx$ ] with @xmath257 , let @xmath68 , and let @xmath27 be not the constant steady state of . there exists a nonempty region @xmath258 and @xmath41 such that for all @xmath259 and @xmath56 , @xmath70 + \tau\beta\int_\omega ( u^k)^{\alpha+\beta-2}|\na u^k|^2 dx \le h[u^{k-1 } ] , \quad k\in{{\mathbb n}}.\ ] ] in one space dimension , we have @xmath260 for the implicit euler scheme , the theorem shows that any positive values for @xmath101 is admissible which corresponds to the continuous situation . for the runge - kutta case with @xmath61 , our condition is more restrictive . as expected , the explicit euler scheme requires the most restrictive condition . the set @xmath261 is illustrated in figure [ fig.0th ] for @xmath262 and @xmath263 . of all @xmath101 for which the zeroth - order entropy is dissipating . left column : @xmath262 , right column : @xmath263 . top row : explicit euler scheme with @xmath192 , middle row : implicit euler scheme with @xmath61 , bottom row : runge - kutta scheme of order @xmath45 with @xmath264.,title="fig:",width=245,height=207 ] of all @xmath101 for which the zeroth - order entropy is dissipating . left column : @xmath262 , right column : @xmath263 . top row : explicit euler scheme with @xmath192 , middle row : implicit euler scheme with @xmath61 , bottom row : runge - kutta scheme of order @xmath45 with @xmath264.,title="fig:",width=245,height=207 ] of all @xmath101 for which the zeroth - order entropy is dissipating . left column : @xmath262 , right column : @xmath263 . top row : explicit euler scheme with @xmath192 , middle row : implicit euler scheme with @xmath61 , bottom row : runge - kutta scheme of order @xmath45 with @xmath264.,title="fig:",width=245,height=207 ] of all @xmath101 for which the zeroth - order entropy is dissipating . left column : @xmath262 , right column : @xmath263 . top row : explicit euler scheme with @xmath192 , middle row : implicit euler scheme with @xmath61 , bottom row : runge - kutta scheme of order @xmath45 with @xmath264.,title="fig:",width=245,height=207 ] of all @xmath101 for which the zeroth - order entropy is dissipating . left column : @xmath262 , right column : @xmath263 . top row : explicit euler scheme with @xmath192 , middle row : implicit euler scheme with @xmath61 , bottom row : runge - kutta scheme of order @xmath45 with @xmath264.,title="fig:",width=245,height=207 ] of all @xmath101 for which the zeroth - order entropy is dissipating . left column : @xmath262 , right column : @xmath263 . top row : explicit euler scheme with @xmath192 , middle row : implicit euler scheme with @xmath61 , bottom row : runge - kutta scheme of order @xmath45 with @xmath264.,title="fig:",width=245,height=207 ] since @xmath68 is fixed , we set @xmath82 . we choose the functions @xmath265 it holds @xmath266 and @xmath267 . then the coefficients in are as follows : @xmath268 introducing the variables @xmath269 for @xmath270 , we can write as @xmath271 with coefficients @xmath272 we need to determine all @xmath101 such that there exist @xmath273 , @xmath274 such that @xmath275 for all @xmath276 . without loss of generality , we exclude the cases @xmath277 and @xmath278 since they lead to parameters @xmath101 included in the region calculated below . thus , let @xmath279 and @xmath280 . these inequalities give the bound @xmath281 . thus , we may introduce the parameter @xmath282 by setting @xmath283 . the polynomial @xmath284 can be rewritten as @xmath285 where @xmath286 is a quadratic polynomial in @xmath287 with the nonpositive leading term @xmath288 . the polynomial @xmath286 is nonnegative for some @xmath287 if and only if its discriminant @xmath289 is nonnegative . here , @xmath290 is a quadratic polynomial in @xmath110 . in order to derive the conditions on @xmath101 such that @xmath291 for some @xmath282 , we employ the computer - algebra system mathematica . the result of the command _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ .... resolve[exists[lambda , s[lambda ] > = 0 & & lambda > 0 & & lambda < 1 ] , reals ] .... _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ gives all @xmath292 such that there exist @xmath273 , @xmath274 such that @xmath275 . the interior of this region equals the set @xmath261 , defined in the statement of the theorem . this shows that @xmath115 for all @xmath259 . if @xmath248 , the nonnegative polynomial @xmath293 has to vanish . in particular , @xmath294 . if @xmath295 in @xmath31 , the boundary conditions imply that @xmath112 is constant , which contradicts our assumption that @xmath112 is not the steady state . thus @xmath296 . similarly , @xmath297 . this gives a system of four inhomogeneous linear equations for @xmath298 which is unsolvable . consequently , @xmath89 . the set @xmath261 is nonempty since , e.g. , @xmath299 . indeed , choosing @xmath300 and @xmath301 , we find that @xmath302 . in one space dimension , the situation simplifies since the laplacian coincides with the hessian and thus , the integration - by - parts formula is not needed . then ( see ) @xmath303 where @xmath304 the polynomial @xmath305 with @xmath306 is nonnegative if and only if @xmath244 and @xmath307 , which is equivalent to @xmath308 this inequality has a solution @xmath274 if and only if the quadratic polynomial has real roots , i.e. if its discriminant is nonnegative , @xmath309 the polynomial @xmath310 with @xmath311 is always nonnegative if @xmath264 ( implicit euler ) . for @xmath61 and @xmath192 , this property holds if and only if @xmath312 . this concludes the proof . we consider the one - dimensional case and first - order entropies with @xmath313 , @xmath257 . [ thm.pm1 ] let @xmath314 be a bounded interval . let @xmath54 be a sequence of ( smooth ) solutions to the runge - kutta scheme of order @xmath45 for in one space dimension . let the entropy be given by @xmath315=\int_\omega ( u^{\alpha/2})_x^2 dx$ ] with @xmath257 , let @xmath68 be fixed , and let @xmath27 be not the constant steady state of . there exists a nonempty region @xmath316 and @xmath41 such that for all @xmath317 , there is a constant @xmath105 such that for all @xmath56 , @xmath80 + \tau c_{\alpha,\beta}\int_\omega ( u^k)^{\alpha+\beta-3}(u^k_{xx})^2 dx \le f[u^{k-1 } ] , \quad k\in{{\mathbb n}}.\ ] ] figure [ fig.1st ] illustrates the set @xmath318 . the set of admissible values @xmath101 for the continuous equation is given by @xmath319 ( the borders of this set are depicted in the figure by the dashed lines ) . for which the discrete first - order entropy for solutions to the one - dimensional porous - medium equation is dissipating . the continuous first - order entropy is dissipated for @xmath320 . the borders of this set is indicated in the figure by dashed lines.,width=321 ] first , we compute @xmath168 according to theorem [ thm.ed2 ] : @xmath321 we show that @xmath168 is nonpositive in a certain range of values @xmath101 . we formulate @xmath168 as @xmath322 where @xmath323 , @xmath324 . we employ the integration - by - parts formula @xmath325 therefore , @xmath326 where @xmath327 this polynomial is nonnegative if and only if @xmath328 which is equivalent to @xmath329 the maximizing value @xmath330 , obtained from @xmath331 , yields @xmath332 and consequently @xmath333 if @xmath334 . this condition is the same as in ( * ? ? ? * theorem 13 ) for the continuous equation . next , we turn to the proof of @xmath89 . the proof of theorem [ thm.ed2 ] shows that @xmath335 we integrate by parts in the last term and use @xmath336 on @xmath78 : @xmath337 where @xmath323 , @xmath324 , @xmath338 , and @xmath339 we employ three integration - by - parts formulas : @xmath340 then @xmath341 and the coefficients are given by @xmath342 choosing @xmath343 , we eliminate the cubic term @xmath344 . furthermore , setting , @xmath345 and @xmath346 , we can write the polynomial @xmath249 as a quadratic polynomial in @xmath347 : @xmath348 the following lemma is a consequence of the proof of lemma 2.2 in @xcite . the polynomial @xmath349 with @xmath350 is nonnegative for all @xmath351 if and only if @xmath352 note that in case @xmath353 and @xmath354 , we may replace @xmath355 by the condition @xmath356 or ( since @xmath350 ) @xmath357 . the first inequality in case ( i ) , @xmath358 is linear in @xmath287 and provides a lower bound for @xmath287 : @xmath359 the second inequality in case ( i ) becomes @xmath360 where @xmath361 and @xmath362 are some polynomials in @xmath363 , @xmath364 , and @xmath287 . this quadratic expression in @xmath365 is nonnegative if and only if its discriminant is nonnegative , @xmath366 where @xmath367 and @xmath368 are some polynomials in @xmath363 and @xmath364 . the factor @xmath369 is positive , so we have to ensure that @xmath370 for some @xmath371 . therefore we must ensure that the rightmost root of @xmath372 is larger or equal than the lower bound for @xmath287 , i.e. , @xmath373 . for @xmath61 , the values @xmath101 for which there exists @xmath371 such that @xmath374 is depicted in figure [ fig.1st ] . in case ( ii ) , we may immediately calculate @xmath287 and @xmath365 but this results in a region which is already contained in the first one . this shows that @xmath115 . if @xmath248 , the polynomial @xmath293 vanishes . thus , either @xmath375 or @xmath376 in @xmath31 . the first case is impossible since @xmath112 is not constant in @xmath31 . as @xmath377 , the second case @xmath376 implies that @xmath378 . hence , @xmath112 is a quadratic polynomial . in view of the boundary conditions , @xmath112 must be constant , but this contradicts our assumption . hence , @xmath89 . we consider the following linear diffusion system : @xmath379 with initial and homogeneous neumann boundary conditions , @xmath380 , @xmath381 , @xmath382 , and the entropy @xmath383 = \int_\omega h(u)dx = \int_\omega\sum_{i=1}^2 u_i(\log u_i-1)dx,\ ] ] where @xmath384 . if the initial data is nonnegative , the maximum principle shows that the solutions to are nonnegative too . [ thm.sys ] let @xmath54 be a sequence of ( smooth ) nonnegative solutions to the runge - kutta scheme for with @xmath61 and @xmath385 . let the entropy @xmath109 be given by . let @xmath68 be fixed and let @xmath27 be not the steady state of . then there exists @xmath41 such that for all @xmath191 , @xmath70 + \tau\int_\omega\bigg(\rho\sum_{i=1}^2\frac{|\na u_i^k|^2}{u_i^k } + \mu(\log u_1^k-\log u_2^k)(u_1^k - u_2^k)\bigg)dx \le h[u^{k-1}].\ ] ] note that we need equal diffusivities @xmath386 and higher - order schemes ( @xmath61 ) . these conditions are in accordance of @xcite , where the continuous equation was studied . in order to highlight the step where these conditions are needed , the following proof is slightly more general than actually needed . we fix @xmath68 and set @xmath82 . let @xmath6=(a_1[u],a_2[u ] ) = ( \rho_1\delta u_1+\mu(u_2-u_1),\rho_2\delta u_2+\mu(u_1-u_2))$ ] . since @xmath42 is linear , @xmath52(h)=a[h]$ ] . thus , @xmath387 + a[u]^\top h''(u)a[u]\big)dx = -g_{1 } - g_{2}.\ ] ] in the following , we set @xmath388 for @xmath389 . we integrate by parts twice , using the boundary conditions @xmath390 and @xmath391\cdot\nu=0 $ ] on @xmath78 , and collect the terms : @xmath392 + \mu(a_2[u]-a_1[u])\big ) \\ & \phantom{xx } { } + \pa_2 h(u)\big(\rho_2\delta a_2[u ] + \mu(a_1[u]-a_2[u])\big)\big)dx \\ & = c_{\rm rk}\int_\omega\big(\rho_1\delta\pa_1 h(u)a_1[u ] + \rho_2\delta\pa_2 h(u)a_2[u ] \\ & \phantom{xx}{}+ \mu(\pa_1 h(u)-\pa_2 h(u))(a_2[u]-a_1[u])\big)dx \\ & = c_{\rm rk}\int_\omega\big(\rho_1\big(\pa_1 ^ 2 h(u)\delta u_1 + \pa_1 ^ 3 h(u)|\na u_1|^2\big)\big(\rho_1 \delta u_1+\mu(u_2-u_1)\big ) \\ & \phantom{xx}{}+ \rho_2\big(\pa_2 ^ 2 h(u)\delta u_2 + \pa_2 ^ 3 h(u)|\na u_2|^2\big)\big(\rho_2 \delta u_2+\mu(u_1-u_2)\big ) \\ & \phantom{xx}{}+ \mu(\pa_2 h(u)-\pa_1 h(u))\big(\rho_1\delta u_1-\rho_2\delta u_2 + 2\mu(u_2-u_1)\big)\big)dx \\ & = c_{\rm rk}\int_\omega\big(\rho_1 ^ 2\pa_1 ^ 2h(u)(\delta u_1)^2 + \rho_2 ^ 2\pa_2 ^ 2h(u)(\delta u_2)^2 + \rho_1 ^ 2\pa_1 ^ 3h(u)\delta u_1|\na u_1|^2 \\ & \phantom{xx}{}+ \rho_2 ^ 2\pa_2 ^ 3h(u)\delta u_2|\na u_2|^2 + \rho_1 \mu\big(\pa_1 ^ 2h(u)(u_2-u_1 ) + \pa_2h(u)-\pa_1h(u)\big)\delta u_1 \\ & \phantom{xx } { } + \rho_2 \mu\big(\pa_2 ^ 2h(u)(u_1-u_2 ) + \pa_1h(u)-\pa_2h(u)\big)\delta u_2 + \rho_1 \mu \pa_1 ^ 3h(u)(u_2-u_1)|\na u_1|^2 \\ & \phantom{xx}{}+ \rho_2 \mu \pa_2 ^ 3h(u)(u_1-u_2)|\na u_2|^2 + 2\mu^2(\pa_2h(u)-\pa_1h(u))(u_2-u_1)\big)dx.\end{aligned}\ ] ] furthermore , @xmath393 adding @xmath394 and @xmath395 , we arrive at @xmath396 the idea of @xcite is to show that each integral @xmath397 , involving only derivatives of order @xmath398 , is nonnegative . in contrast to @xcite , we employ systematic integration by parts , which allows for a simpler and more general proof in our context . for the term @xmath399 , we use the following integration - by - parts formula : @xmath400 then , for @xmath401 , @xmath402 the integrand defines a quadratic polynomial in the variables @xmath403 and @xmath404 and is nonnegative if its discriminant satisfies @xmath405 . it turns out that this inequality holds true for @xmath406 if we choose @xmath407 and @xmath401 sufficiently small . when @xmath192 , we can show only that @xmath408 which is not sufficient to prove that @xmath89 ( see below ) . we conclude that @xmath409 integrating by parts in @xmath410 in order to obtain only first - order derivatives , we find after some rearrangements that @xmath411 the integrand is nonnegative if and only if @xmath412 for all @xmath413 . we compute : @xmath414 thus , @xmath412 is possible only if @xmath386 and @xmath61 . finally , we see immediately that the remaining term @xmath415 is nonnegative . this shows that @xmath115 . if @xmath248 , we infer from that @xmath416 , but this contradicts our hypothesis that @xmath417 is not a steady state . consider the one - dimensional fourth - order equation @xmath418 with periodic boundary conditions . this equation appears as a scaling limit of the so - called ( time - discrete ) toom model , which describes interface fluctuations in a two - dimensional spin system @xcite . the variable @xmath112 is the limit of a random variable related to the deviation of the spin interface from a straight line . the multi - dimensional version of models the eectron density @xmath112 in a quantum semiconductor , und the equation is the zero - temperature , zero - field approximation of the quantum drift - diffusion model @xcite . for existence results for , we refer to @xcite and references therein . to simplify our calculations , we analyze only the logarithmic entropy @xmath11=\int_\omega u(\log u-1)dx$ ] . it is possible to verify condition also for entropies of the form @xmath419 , but it turns out that only sufficiently small @xmath257 are admissible ( about @xmath420 ) and the computations are very tedious . therefore , we restrict ourselves to the case @xmath103 . let @xmath54 be a sequence of ( smooth ) solutions to the runge - kutta scheme with @xmath61 for . let the entropy be given by @xmath11=\int_\omega u(\log u-1)dx$ ] , let @xmath68 be fixed , and let @xmath27 be not a steady state . then there exists @xmath41 such that for all @xmath191 , @xmath70 + \tau q\int_\omega u(\log u)_x^8 dx + \tau\int_\omega u(\log u)_{xx}^2 dx \le h[u^{k-1 } ] , \quad q\approx 0.0045.\ ] ] first , we observe that @xmath421 . with @xmath6=(u(\log u)_{xx})_{xx}$ ] and @xmath52(h ) = \big(h_{xx } - 2(\log u)_xh_x + ( \log u)_x^2 h\big)_{xx}$ ] , we can write @xmath88 according to as @xmath422_{xx } - 2(\log u)_xa[u]_x + ( \log u)_x^2a[u]\big)_{xx } + \frac{1}{u}a[u]^2\big)dx \\ & = -\int_\omega\big((\log u)_{xx}\big(a[u]_{xx } - 2(\log u)_xa[u]_x + ( \log u)_x^2a[u]\big ) + \frac{1}{u}a[u]^2\big)dx \\ & = -\int_\omega\big(\big(v_{xxxx } + 2(v_x v_{xx})_x + v_x^2 v_{xx}\big ) a[u ] + \frac{1}{u}a[u]^2\big)dx,\end{aligned}\ ] ] where we have integrated by parts several times and have set @xmath423 . then @xmath6=u(v_x^2 v_{xx } + 2v_x v_{xxx } + v_{xx}^2 + v_{xxxx})$ ] and , with the abbreviations @xmath424 , @xmath425 we employ the following integration - by - parts formulas : @xmath426 then @xmath427 where @xmath428 next , we eliminate all terms involving @xmath429 by formulating the following square : @xmath430dx.\end{aligned}\ ] ] we eliminate all terms involving @xmath431 and set the corresponding coefficients to zero . from @xmath432 we conclude that @xmath433 . furthermore , @xmath434 by these choices , we obtain @xmath435 this quadratic polynomial in @xmath436 admits its maximal value at @xmath437 with value @xmath438 . the integral can now be written as @xmath439 where @xmath440 if @xmath441 , we can write the integal as the sum of two squares , noting that @xmath442 is positive , @xmath443 the expression @xmath444 defines a polynomial in @xmath445 which is linear in @xmath446 . solving it for @xmath446 gives @xmath447 it remains to show that @xmath448 , which is a polynomial of fourth order in @xmath365 , is positive . choosing @xmath449 , we find that @xmath450 . this shows that @xmath451 finally , if @xmath248 , we infer that @xmath112 is constant which is excluded . therefore , @xmath89 , which ends the proof . the aim of this section is to explore the numerical behavior of the second - order derivative of the function @xmath452 , defined in the introduction , for the porous - medium equation in one space dimension . the equation is discretized by standard finite differences , and we employ periodic boundary conditions . the discrete solution @xmath453 approximates the solution @xmath454 to with @xmath455 , @xmath456 , and @xmath457 , @xmath81 are the space and time step sizes , respectively . we choose the barenblatt profile @xmath458 where @xmath459 as the initial datum . its support is contained in @xmath460 $ ] ; see figure [ fig.evol ] ( left ) . we choose the exponent @xmath461 . the semi - logarithmic plot of the discrete entropy @xmath462=\sum_{i=0}^n ( u_i^k)^\alpha \triangle x$ ] with @xmath463 versus time is illustrated in figure [ fig.evol ] ( right ) , using the implicit euler scheme with parameters @xmath464 and the number of grid points @xmath465 . the decay is exponential for `` large '' times . the nonlinear discrete system is solved by newton s method with the tolerance @xmath466 . we have highlighted four time steps @xmath467 at which we will compute numerically the function @xmath452 for the following runge - kutta schemes : @xmath468 , \\ & \mbox{implicit euler scheme : } & u^k - u^{k-1 } & = -\tau a[u^{k } ] , \\ & \mbox{second - order trapezoidal rule : } & u^k - u^{k-1 } & = -\frac{\tau}{2}(a[u^k]+a[u^{k-1 } ] ) , \\ & \mbox{third - order simpson rule : } & u^k - u^{k-1 } & = -\frac{\tau}{6}(a[u^k]+4a[(u^k+u^{k-1})/2]+a[u^{k-1}]).\end{aligned}\ ] ] at various time steps @xmath467 , @xmath469 . right : semi - logarithmic plot of the discrete entropy @xmath462 $ ] versus time.,title="fig:",width=283 ] at various time steps @xmath467 , @xmath469 . right : semi - logarithmic plot of the discrete entropy @xmath462 $ ] versus time.,title="fig:",width=283 ] we set as before @xmath82 , @xmath83 and compute @xmath470-h_d[v(\tau)]$ ] and the discrete second - order derivative @xmath471 of @xmath177 ( using central differences ) . the result is presented in figure [ fig.g ] . as expected , the discrete derivative @xmath472 is negative on a ( small ) interval for all times @xmath467 , @xmath473 . we observe that @xmath471 is even slightly decreasing , but we expect that it becomes positive for sufficiently large values of @xmath81 . clearly , the values for @xmath472 tend to zero as we approach the steady state ( see remark [ rem.tauk ] ) . this experiment indicates that @xmath43 from theorem [ thm.ed ] is bounded from below by @xmath474 , for instance . for various runge - kutta schemes at the time steps @xmath467 . top left : explicit euler scheme ; top right : implicit euler scheme ; bottom left : implicit trapezoidal rule ; bottom right : simpson rule.,width=529 ] in order to understand the behavior of @xmath452 in a better way , it is convenient to study the discrete version of the quotient @xmath475 indeed , the analysis in section [ sec.pme ] gives an estimate of the type @xmath476 for some constant @xmath477 . thus , we expect that for sufficiently small @xmath65 , @xmath478 is bounded from above by some negative constant . this expectation is confirmed in figure [ fig.quot ] . in the examples , @xmath478 is a decreasing function of @xmath81 , and @xmath479 is decreasing with increasing time . , defined in , for various runge - kutta schemes at the time steps @xmath467 . top left : explicit euler scheme ; top right : implicit euler scheme ; bottom left : implicit trapezoidal rule ; bottom right : simpson rule.,width=529 ] c. chainais - hillairet , a. jngel , and s. schuchnigg . entropy - dissipative discretization of nonlinear diffusion equations and discrete beckner inequalities . to appear in _ esaim math _ , 2015 . arxiv:1303.3791 .

we study long time behavior of a discrete time weakly interacting particle system , and the corresponding nonlinear markov process in , described in terms of a general stochastic evolution equation . in a setting where the state space of the particles is compact such questions have been studied in previous works , however for the case of an unbounded state space very few results are available . under suitable assumptions on the problem data we study several time asymptotic properties of the-particle system and the associated nonlinear markov chain . in particular we show that the evolution equation for the law of the nonlinear markov chain has a unique fixed point and starting from an arbitrary initial condition convergence to the fixed point occurs at an exponential rate . the empirical measure of the-particles at time is shown to converge to the law of the nonlinear markov process at time , in the wasserstein-1 distance , in , as , _ uniformly _ in . several consequences of this uniform convergence are studied , including the interchangeability of the limits and and the propagation of chaos property at . rate of convergence of to is studied by establishing uniform in time polynomial and exponential probability concentration estimates . * ams 2010 subject classifications : * primary 60j05 , 60k35 , 60f10 . * keywords : * weakly interacting particle system , propagation of chaos , nonlinear markov chains , wasserstein distance , mckean - vlasov equations , exponential concentration estimates , transportation inequalities , metric entropy , stochastic difference equations , long time behavior , uniform concentration estimates . [ section ] [ theorem]*lemma * [ section ] [ section ] [ theorem]*corollary * [ section ] [ section ] [ theorem]*proposition * [ section ]

1401.3423

stochastic dynamical systems that model the evolution of a large collection of weakly interacting particles have long been studied in statistical mechanics ( cf . @xcite and references therein ) . in recent years such models have been considered in many other application areas as well , some examples include , chemical and biological systems ( e.g. biological aggregation , chemotactic response dynamics@xcite ) , mathematical finance ( e.g. mean field games@xcite , default clustering in large portfolios@xcite ) , social sciences ( e.g. opinion dynamics models @xcite ) , communication systems ( @xcite ) etc . starting from the work of sznitman@xcite there has been an extensive body of work that studies law of large number behavior ( propagation of chaos ) , central limit theory ( normal fluctuations from the mean ) and large deviation principles for such models . all of these results concern the behavior of the system over a finite time horizon . in many applications the time asymptotic behavior of the system is of central concern . for example , stability of a communication system , steady state aggregation and self organization in biological and chemical systems , long term consensus formation mechanisms in opinion dynamics modeling , particle based approximation methods for invariant measures all rely on a careful analysis of the time asymptotic behavior of such systems . such behavior for special families of weakly interacting particle systems has been considered by several authors . in @xcite the authors give general sufficient conditions for a family of discrete time systems for uniform in time exponential probability concentration estimates to hold . these conditions formulated in terms of dobrushin s coefficient are not very restrictive when the state space of the particles is compact , however they are hard to verify for settings with an unbounded state space . in @xcite a discrete time model with a compact state space for chemotactic cell response dynamics was studied . several time asymptotic results , including uniform in time law of large numbers , exponential stability of the associated nonlinear markov process and uniform in time convergence of a particle based simulation scheme , were established . for the setting of an unbounded state space and in continuous time , there have been several recent interesting works on granular media equations @xcite which establish uniform in time propagation of chaos , time uniform convergence of simulation schemes and uniform in time exponential concentration estimates . in the current work we consider a discrete time weakly interacting particle system and the corresponding nonlinear markov process in @xmath0 , described in terms of a general stochastic evolution equation . denoting by @xmath9 the state of the @xmath10-th particle @xmath11 at time instant @xmath3 , the evolution is given as @xmath12 here @xmath13 is the empirical measure of the particle values at time instant @xmath3 , @xmath14 is a @xmath15 matrix , @xmath16 is a small parameter , @xmath17 is an i.i.d array of @xmath18 valued random variables with common probability law @xmath19 and @xmath20 is a measurable function . also , @xmath21 are taken to be exchangeable with common distribution @xmath22 . as will be seen in section [ sec : mainresults ] , the following nonlinear markov chain will correspond to the @xmath23 limit of . @xmath24 where throughout we denote by @xmath25 the probability distribution of a random variable @xmath26 with values in some polish space @xmath27 . under conditions on @xmath28 and @xmath14 we study several long time properties of the @xmath1-particle system and the associated nonlinear markov chain . our starting point is the evolution equation for the law of the nonlinear markov chain given in . we show in theorem [ th2 ] that under conditions , that include a lipschitz property of @xmath29 with the wasserstein-1(@xmath30 ) distance on the space of probability measures ( assumptions [ assuma ] and [ assumb ] ) , contractivity of @xmath14 ( assumption [ assumc ] ) and @xmath31 being sufficiently small , has a unique fixed point and starting from an arbitrary initial condition convergence to the fixed point occurs at an exponential rate . using this result we next argue in theorem [ th3 ] that under an additional integrability condition ( assumption [ assumd ] ) , as @xmath6 , the empirical measure @xmath2 of the @xmath1-particles at time @xmath3 converges to the law @xmath4 of the nonlinear markov process at time @xmath3 , in the w1 distance , in @xmath5 , _ uniformly _ in @xmath3 . this result in particular shows that the @xmath30 distance between @xmath2 and the unique fixed point @xmath32 of converges to zero as @xmath7 and @xmath33 in any order . this result is key in developing particle based numerical schemes for approximating the fixed point of the evolution equation . we next show that under an irreducibility condition on the underlying markovian dynamics ( assumption [ assume ] ) the unique invariant measure @xmath34 of the @xmath1-particle dynamics is @xmath32-chaotic , namely as @xmath6 , the projection of @xmath34 on the first @xmath35-coordinates converges to @xmath36 for every @xmath37 . this propagation of chaos property all the way to @xmath8 crucially relies on the uniform in time convergence of @xmath2 to @xmath32 . the next three results study the rate of this uniform convergence by developing suitable probability concentration estimates . the first result ( theorem [ th5 ] ) , under an assumption of polynomial moments on the initial data and noise sequence ( assumption [ assumd ] ) establishes a corresponding uniform in time polynomial concentration bound . the proof relies on an idea of restricting measures to a compact set and estimates on metric entropy introduced in @xcite ( see also @xcite ) . the basic idea is to first obtain a concentration bound for the @xmath38 distance between the truncated law and its corresponding empirical law in a compact ball of radius @xmath39 along with an estimate on the contribution from the region outside the ball and finally optimize suitably over @xmath39 . the last two results are concerned with exponential concentration . these impose much stronger integrability conditions on the problem data ( assumption [ assumg ] ) . the first considers the setting where the initial random variables form a general exchangeable sequence and gives a concentration bound with an exponential decay rate of @xmath40 . the second result uses exponential concentration estimates for empirical measures of i.i.d . sequences based on transportation inequalities from @xcite ( see also @xcite ) and considers the setting where the initial data is i.i.d . in this case the concentration bound gives an exponential decay rate of order @xmath1 . the following notation will be used in this work . @xmath41 will denote the @xmath42 dimensional euclidean space with the usual euclidean norm @xmath43 . the set of natural numbers ( resp . whole numbers ) is denoted by @xmath44 ( resp . @xmath45 ) . cardinality of a finite set @xmath27 is denoted by @xmath46 . for a measurable space s , @xmath47 denotes the space of all probability measures on @xmath27 . for @xmath48 , @xmath49 is the dirac delta measure on @xmath41 that puts a unit mass at location @xmath50 . the space of real valued bounded measurable functions on @xmath27 is denoted as @xmath51 . borel @xmath52 field on a metric space will be denoted as @xmath53 . @xmath54 denotes the space of all bounded and continuous functions @xmath55 . the supremum norm of a function @xmath56 is @xmath57 . when @xmath27 is a metric space , the lipschitz seminorm of @xmath29 is defined by @xmath58 where @xmath42 is the metric on the space @xmath27 . for a bounded lipschitz function @xmath29 on @xmath27 we define @xmath59 . @xmath60 ( resp . @xmath61 ) denotes the class of lipschitz ( resp . bounded lipschitz ) functions @xmath62 with @xmath63 ( resp . @xmath64 ) bounded by 1 . occasionally we will suppress @xmath27 from the notation and write @xmath65 and @xmath66 when clear from the context . for a polish space @xmath27 , @xmath47 is equipped with the topology of weak convergence . a convenient metric metrizing this topology on @xmath47 is given as @xmath67 for @xmath68 . for a signed measure @xmath69 on @xmath41 , we define @xmath70 whenever the integral makes sense . let @xmath71 be the space of @xmath72 such that @xmath73 the space @xmath71 will be equipped with the wasserstein-1 distance that is defined as follows : @xmath74 where the infimum is taken over all @xmath41 valued random variables x , y defined on a common probability space and where the marginals of x , y are respectively @xmath22 and @xmath75 . from kantorovich - rubenstein duality ( cf . @xcite ) one sees the wasserstein-1 is same as @xmath76 for a signed measure @xmath77 on @xmath78 , the total variation norm of @xmath77 is defined as @xmath79 . convergence in distribution of a sequence @xmath80 of @xmath27 valued random variable to @xmath26 will be written as @xmath81 . a finite collection @xmath82 of @xmath27 valued random variables is called exchangeable if @xmath83 for every permutation @xmath84 on the @xmath1 symbols @xmath85 . let @xmath86 be a collection of @xmath27 valued random variables , such that for every @xmath1 , @xmath87 is exchangeable . let @xmath88 . the sequence @xmath89 is called @xmath90 -chaotic ( cf . @xcite ) for a @xmath91 , if for any @xmath92 , @xmath93 one has @xmath94 denoting the marginal distribution on first @xmath35 coordinates of @xmath95 by @xmath96 , equation ( [ chaotic ] ) says that , for every @xmath97 @xmath98 . recall the system of @xmath1 interacting particles in @xmath41 introduced in . throughout we will assume that @xmath21 is exchangeable with common distribution @xmath22 where @xmath99 . assumptions on @xmath28 and @xmath14 will be introduced shortly . note that in the notation we have suppressed the dependence of the sequence @xmath100 on @xmath1 . given @xmath101 define a transition probability kernel @xmath102 $ ] as @xmath103 } \theta(dz),\quad \quad ( x , c)\in { { \mathbb{r}}}^{d } \times \mathcal{b}({{\mathbb{r}}}^{d}).\ ] ] with an abuse of notation we will also denote by @xmath104 the map from @xmath105 to itself , defined as @xmath106 for @xmath107 , let @xmath108 be defined as @xmath109 note that @xmath110 when @xmath111 . under assumptions [ assuma ] and [ assumb ] introduced in the next section it will follow that , for @xmath112 @xmath113 as well . under these conditions , one can define @xmath114 as @xmath115 then the evolution of the law of the nonlinear markov chain given in is given by the equation @xmath116 using the above notation we see that @xmath117 is a @xmath118 valued discrete time markov chain defined recursively as follows . let @xmath119 and let @xmath120 then , for @xmath92 @xmath121 recall that @xmath21 is assumed to be exchangeable with common distribution @xmath22 where @xmath99 . we now introduce our assumptions on the nonlinearity . [ assuma ] @xmath122 , where @xmath123 note that the assumption [ assuma ] implies that @xmath124 where @xmath125 . we impose the following condition on @xmath126 [ assumb ] @xmath127 . our first result is a law of large numbers for @xmath128 as @xmath33 . note that under assumptions [ assuma ] and [ assumb ] , @xmath129 for all @xmath130 . [ th1 ] suppose assumptions [ assuma ] and [ assumb ] hold and suppose that @xmath131 as @xmath33 . then , as @xmath33 , @xmath132 for all @xmath133 . [ rem1 ] note that theorem [ th1 ] says that for all @xmath134 @xmath135 which in particular implies that @xmath136 in probability , in @xmath137 ( with the topology of weak convergence ) as @xmath33 . next we state a propagation of chaos " result which is an immediate consequence of remark [ rem1 ] and exchangeability of of @xmath138 . suppose assumptions [ assuma ] and [ assumb ] hold . then for any @xmath92 and @xmath139 @xmath140 as @xmath33 . for a @xmath15 matrix b we denote its norm by @xmath141 i.e. @xmath142 . [ assumc ] @xmath143 for some @xmath144 . a measure @xmath145 is called a fixed point for the evolution equation in ( [ nls ] ) , if @xmath146 let @xmath147 . [ th2 ] suppose assumptions [ assuma],[assumb ] and [ assumc ] hold and that @xmath148 then there exists a unique fixed point @xmath32 of equation ( [ recursion ] ) . furthermore , denoting for @xmath149 @xmath150 = \underbrace{\psi \circ \psi ... \circ \psi } _ { \mbox{n \text{times}}}(\gamma)$ ] , we have @xmath151,\mu_{\infty } ) < 0,\ ] ] namely @xmath150 $ ] converges to @xmath32 as @xmath152 , at an exponential rate . next , we study uniform in time ( i.e @xmath3 ) convergence of @xmath128 to @xmath153 as the number of particles @xmath33 . for this , we will strengthen assumptions [ assuma ] and [ assumb ] as follows . [ assumd ] for some @xmath154 @xmath155 [ th3 ] suppose that assumptions [ assumc ] and [ assumd ] hold . also suppose that @xmath156 then 1 . given @xmath157 there exist @xmath158 such that @xmath159 2 . suppose @xmath131 as @xmath33 . then @xmath160 as @xmath6 . [ cor1 ] suppose assumptions [ assumc ] and [ assumd ] hold and suppose @xmath148 then @xmath161 the interchangeability of the limits given in corollary [ cor1 ] allows one to characterize the large @xmath1 limit of the steady state behavior of the particle system . we need the following assumption on the markov chain @xmath162 where recall @xmath163 the assumption is essentially a communicability condition on the underlying markovian dynamics . [ assume ] for every @xmath164 , the markov chain @xmath162 has at most one invariant measure . [ th4 ] suppose assumptions [ assuma ] , [ assumb ] , [ assumc ] and [ assume ] hold and suppose @xmath165 . then for every @xmath166 , the markov chain @xmath162 has a unique invariant measure @xmath167 . suppose in addition assumption [ assumd ] holds . then @xmath167 is @xmath32- chaotic , where @xmath32 is as in theorem [ th2 ] . theorem [ th3 ] gives conditions under which @xmath168 converges to @xmath169 as @xmath6 , in @xmath5 , uniformly in @xmath3 . the next three theorems show that under additional conditions , one can provide concentration bounds uniformly in @xmath3 which give estimates on the rate of convergence . recall the measure @xmath170 introduced at the beginning of section [ sec : mod - desc ] . with @xmath171 defined in assumption [ assumd ] and @xmath172 as in assumption [ assumc ] , let @xmath173 [ th5 ] suppose assumptions [ assumc ] and [ assumd ] holds . fix @xmath174 and suppose that @xmath175 . . then there exists @xmath177 and @xmath178 such that for all @xmath179 and for all @xmath180 @xmath181 for all @xmath182 . [ cor2 ] \(i ) since @xmath183 , we have that @xmath184 and so the above theorem gives the following uniform concentration estimate : @xmath185 for all @xmath186 . \(ii ) under additional conditions on @xmath187 one can give concentration bounds for the first term on the right side of the above inequality . for example , when @xmath188 are i.i.d . such concentration bounds can be found in theorem 2.7 of @xcite . also , although not pursued here , the bound obtained in theorem [ th5 ] can be used to give an alternative proof of theorem [ th3](2 ) . next we obtain exponential concentration bounds . the bounds depend in particular on our assumptions on the initial condition . our first result ( theorem [ th6 ] ) treats the case where the initial random vector has a general exchangeable distribution while the second result ( theorem [ thm6 ] ) considers a more restrictive setting where the initial random vector is i.i.d . in the second case the probabilities will decay exponentially in @xmath1 whereas in the first case the exponent will be some dimensional dependent power of @xmath1 . we start with our main assumption for theorem [ th6 ] . [ assumg ] \(i ) for some @xmath189 , @xmath190 for @xmath19 a.e . @xmath191 . \(ii ) there exists @xmath192 such that @xmath193 and @xmath194 . [ th6 ] suppose that assumptions [ assumc ] and [ assumg ] hold . fix @xmath174 and suppose that @xmath195 . then there exists @xmath196 and @xmath178 such that for all @xmath197 @xmath198\leq p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0})>2\sigma\gamma_0 \vartheta^n\varepsilon ] + e^{-c_1 { \varepsilon}n^{1/d+2}},\ ] ] for all @xmath134 , @xmath199 , if @xmath200 ; and @xmath198\leq p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0})>2\sigma\gamma_0 \vartheta^n\varepsilon ] + e^{-c_1 ( { \varepsilon}\wedge 1 ) n^{1/d+2}},\ ] ] for all @xmath201 , @xmath202 , if @xmath203 . here @xmath204 is as in theorem [ th5 ] . finally we consider the case where the initial distribution of the @xmath1 particles is i.i.d . the proof relies on various estimates from @xcite . [ thm6 ] suppose that @xmath205 are i.i.d . with common distribution @xmath22 for each @xmath1 . suppose that assumptions [ assumc ] and [ assumg ] hold . fix @xmath206 . suppose that @xmath207 then there exist @xmath208 and a nonincreasing function @xmath209 such that @xmath210 as @xmath211 and for all @xmath197 and @xmath212 @xmath213 \le a_1e^{-n a_2 ( { \varepsilon}^2 \wedge { \varepsilon } ) } \ ] ] [ rem : vsconstant ] ( i ) one can describe the function @xmath214 in the above theorem explicitly . define for @xmath215 , @xmath216 as @xmath217 , where @xmath218 is as in assumption [ assumg ] . then @xmath219 where @xmath220 is defined by the right side of with @xmath221 replaced by @xmath222 where @xmath222 is as in corollary [ cor : cor621 ] . + + ( ii ) if assumption [ assumg ] is strengthened to @xmath223 for some @xmath224 then one can strengthen the conclusion of theorem [ thm6 ] as follows : for @xmath31 sufficiently small there exist @xmath208 and a nonincreasing function @xmath225 such that @xmath226 as @xmath211 and for all @xmath197 and @xmath227 @xmath213 \le a_1e^{-n a_2 { \varepsilon}^2 } .\ ] ] the following elementary lemma gives a basic moment bound that will be used in our analysis . [ lem1 ] suppose assumptions [ assuma ] and [ assumb ] hold . then , for every @xmath228 , @xmath229 in addition , if assumption [ assumc ] holds and @xmath165 then @xmath230 * proof : * we will only prove the second statement in the lemma . proof of the first statement is similar . note that , for @xmath228 and @xmath231 @xmath232 thus @xmath233 it follows that @xmath234 = e\frac{1}{n}\sum_{k=1}^{n}|x_{n-1}^{k}|= e|x_{n-1}^{1}|.\ ] ] taking expectation in ( [ eq1.1.1 ] ) and using independence between @xmath235 and @xmath236 , we have @xmath237 the assumption on @xmath31 implies that @xmath238 a recursive application of ( [ eq1.1.2 ] ) now shows that @xmath239 the result follows . recall the map @xmath240 defined in . [ cor2 ] under assumptions [ assuma ] and [ assumb ] , for every @xmath241 and @xmath228 , there exists a compact set @xmath242 such that @xmath243 * proof : * note that for any non - negative @xmath244 , @xmath245 and @xmath246 to get the desired result from above equalities it suffices to show that @xmath247 we will prove ( [ e1 ] ) by induction on @xmath3 . once more we suppress @xmath1 from the super - script . clearly by our assumptions @xmath248 is uniformly integrable . now suppose that the statement ( [ e1 ] ) holds for some @xmath3 . note that @xmath249 + \delta d_{1}(\epsilon_{n+1}^{i}).\label{eq1.2.2}\end{aligned}\ ] ] from exchangeability it follows @xmath250 $ ] , where @xmath251 . combining this with the induction hypothesis that @xmath252 is uniformly integrable , we see that @xmath253 is uniformly integrable . here we have used the fact that if @xmath254 is a uniformly integrable family and @xmath255 is a collection of @xmath52-fields where @xmath256 are arbitrary index sets , then @xmath257 is a uniformly integrable family . also from assumptions [ assuma ] and [ assumb ] the families @xmath258 , @xmath259 are uniformly integrable . these observations along with independence between @xmath260 and @xmath261 yield that the family @xmath262 is uniformly integrable . the result follows . we now proceed to the proof of theorem [ th1 ] . we will argue via induction on @xmath133 . by assumption ( [ pointwise ] ) holds for @xmath263 . assume now that it holds for some @xmath264 . note that , @xmath265 consider the last term in ( [ eq1.3.1 ] ) . using assumption [ assuma ] we see that if @xmath266 is lipschitz then @xmath267 is lipschitz and @xmath268 thus , almost surely @xmath269 taking expectations we obtain , @xmath270 consider now the second term in ( [ eq1.3.1 ] ) . using assumption [ assuma ] again , we have , @xmath271\theta(d\xi)\nonumber\\ & \leq & \delta \sigma\mathcal{w}_{1}(\mu_{n}^{n},\mu_{n}).\nonumber\end{aligned}\ ] ] taking expectations we get @xmath272 now we consider the first term of the right hand side of ( [ eq1.3.1 ] ) . we will use lemma [ cor2 ] . fix @xmath273 and let @xmath274 be a compact set in @xmath41 such that @xmath275 let @xmath276 . then , @xmath277 we will now apply lemma [ app1 ] in the appendix . note that for any @xmath278 , @xmath279 thus with notation as in lemma [ app1 ] @xmath280 where we have denoted the restrictions of @xmath281 and @xmath282 to @xmath274 by the same symbols . using the above inequality in ( [ 1.3.4 ] ) , we obtain @xmath283 using lemma [ app2 ] we see that the second term on the left hand side can be bounded by @xmath284 . combining this estimate with , ( [ eq1.3.2 ] ) and ( [ eq : ab428 ] ) we now have @xmath285 sending @xmath33 in ( [ 1.3.6 ] ) and using induction hypothesis , we have @xmath286 since @xmath241 is arbitrary , the result follows . [ sec - int ] we begin with the following estimate . [ lem2.1 ] under assumptions [ assuma],[assumb ] and [ assumc ] @xmath287 for any choice of @xmath288 . * proof : * given @xmath289 let @xmath290 fix @xmath291 and let @xmath292 be @xmath293 valued random variables with distribution @xmath77 . also , let @xmath294 be an iid sequence of random variables with common law @xmath19 independent of @xmath295 define for @xmath133 , @xmath296 where @xmath297 and @xmath298 . then clearly @xmath299 . for @xmath300 denote @xmath301 then @xmath302 also , @xmath303 where the second inequality in the display follows from assumptions [ assuma ] and [ assumc ] . combining and we have @xmath304 we now have , on taking infimum on the right hand side of the above display over all @xmath291 , that @xmath305 . the result follows . we now complete the proof of theorem [ th2 ] . observe that under our assumption on @xmath31 , @xmath306 . the first part of the theorem now follows from lemma [ lem2.1 ] and banach s fixed point theorem . furthermore @xmath307 second part of the theorem is now immediate . we start with the following moment bound . [ lem2 ] suppose that assumptions [ assumc ] and [ assumd ] hold and suppose that @xmath308 where @xmath309 is as in . then @xmath310 * proof : * by holder s inequality , for any four nonnegative real numbers @xmath311 @xmath312 using assumption [ assumc ] and ( [ fbound ] ) @xmath313 ) , we have , from assumption [ assumd ] @xmath314\nonumber + 4^{\alpha}c_{1}\nonumber\\ & \leq & 4^{\alpha}e^{-\omega(1+\alpha ) } e|x_{n}^{i}|^{1+\alpha } + 4^{\alpha } \delta^{(1+\alpha)}\sigma_{1}[2e|x_{n}^{i}|^{1+\alpha } ] + 4^{\alpha}c_{1}\quad \label{4.3.1.1}\end{aligned}\ ] ] where the last line in the display follows from jensen s inequality : @xmath315 and @xmath316 is as in assumption [ assumd ] . note that under our condition on @xmath31 @xmath317 < 1.\ ] ] thus @xmath318 where @xmath319 . the result follows . we now complete the proof of theorem [ th3 ] . note that @xmath320 + [ \psi^{n}(\mu_{0}^{n } ) - \psi^{n}(\mu_{0})].\ ] ] it then follows using lemma [ lem2.1 ] that with @xmath321 almost surely @xmath322 taking expectations , @xmath323 since @xmath324 as @xmath325 and @xmath326 , we can find @xmath327 such that @xmath328 . from lemma [ lem2 ] we then have that @xmath329 and consequently the family @xmath330 is uniformly integrable . similar to the proof of corollary [ cor2 ] ( cf . the argument below ) using ( [ sum ] ) and ( [ eq1.2.1 ] ) it follows that , for some compact @xmath331 @xmath332 < \epsilon . \label{4.3.3}\end{aligned}\ ] ] now for every @xmath37 @xmath333 when @xmath334 is as introduced above ( [ 1.3.4 ] ) . applying lemmas [ app1 ] and [ app2 ] as in the proof of theorem [ th1 ] we now see that @xmath335 where @xmath336 . thus @xmath337 given @xmath157 choose @xmath338 sufficiently small and @xmath339 sufficiently large such that @xmath340 @xmath341 choose @xmath342 large enough so that @xmath343 , @xmath344 . combining the above estimates we have @xmath345 and @xmath346 @xmath347 this proves the first part of the theorem . second part is immediate from the first part and theorem [ th1 ] . * proof : * note that @xmath348 combining this with we have @xmath349 the result now follows on using theorem [ th2 ] . for @xmath164 and @xmath130 , define @xmath350 as @xmath351 where @xmath352 are as defined in ( [ nlps ] ) . from lemma [ lem1 ] it folows that , for each @xmath353 the sequence @xmath354 is relatively compact and using assumption [ assuma ] it is easy to see that any limit point of @xmath355 ( as @xmath152 ) is an invariant measure of the markov chain @xmath356 . combining this with assumption [ assume ] we have that there exists a unique invariant measure @xmath357 for this markov chain and , as @xmath7 , @xmath358 this proves the first part of the theorem . define @xmath359 as @xmath360 let @xmath361 and @xmath362 . in order to prove that @xmath34 is @xmath32-chaotic , it suffices to argue that ( cf . @xcite ) @xmath363 we first argue that as @xmath152 @xmath364 it suffices to show that @xmath365 for any continuous and bounded function @xmath366 . but this is immediate on observing that @xmath367 the continuity of the map @xmath368 and the weak convergence of @xmath355 to @xmath34 . next , for any @xmath369 @xmath370 fix @xmath371 . for every @xmath372 there exists @xmath373 such that for all @xmath374 @xmath375 thus for all @xmath376 @xmath377 finally @xmath378 where the first equality is from , the second uses and the third is a consequence of corollary [ cor1 ] . since @xmath371 is arbitrary , we have and the result follows . we will first develop a concentration bound for @xmath379 for each fixed @xmath3 and then combine it with the estimate in in order to obtain the desired result . the first step is carried out in the lemma below , the proof of which is given in section [ sec : prooflem-4.20 ] . [ lem : fixn-4.10 ] suppose assumptions [ assumc ] and [ assumd ] hold . then , there exist @xmath380 such that for all @xmath381 and @xmath382 , @xmath383 & \leq & a_3 \left(e^{-a_2 \frac{n \varepsilon^{2}}{r^{2 } } } + \frac{r^{-\alpha}}{\varepsilon}\right ) \end{aligned}\ ] ] for all @xmath384 . we now complete the proof of theorem [ th5 ] using the lemma . we will make use of . recall that @xmath385 and by our assumption @xmath386 . let @xmath387 . note that @xmath388 . then @xmath389 & \leq & p[\cup_{i=1}^{n}\ { \chi^{n- i}\mathcal{w}_{1}(\mu_{i}^{n},\psi(\mu_{i -1}^{n } ) ) > \gamma ( 1-\gamma)^{n - i}\varepsilon \ } \cup \nonumber\\ & & \{\chi^{n}\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0 } ) > \gamma ( 1-\gamma)^{n}\varepsilon\ } ] \nonumber\\ & \leq & \sum_{i = 1}^{n } p[\mathcal{w}_{1}(\mu_{i}^{n},\psi(\mu_{i -1}^{n } ) ) > \gamma(\frac{1-\gamma}{\chi})^{n - i}\varepsilon ] \nonumber\\&+ & p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0 } ) > \gamma ( \frac{1-\gamma}{\chi})^{n}\varepsilon].\label{unif}\end{aligned}\ ] ] let @xmath390 . note that @xmath391 and from our choice of @xmath31 , @xmath184 . therefore @xmath392 thus from lemma [ lem : fixn-4.10 ] , for all @xmath393 and @xmath394 @xmath395 \leq a_3 \left(e^{-a_2 \frac{n \beta^2\vartheta^{2(n - k)}}{r^{2 } } } + \frac{r^{-\alpha}}{\beta \vartheta^{n - k}}\right).\ ] ] using the above estimate in @xmath389 \le & a_3 \sum_{i=0}^{n-1 } \left(e^{-a_2 \frac{n \beta^2\vartheta^{2i}}{r^{2 } } } + \frac{r^{-\alpha}}{\beta \vartheta^{i}}\right ) + p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0 } ) > \gamma\varepsilon ] \nonumber \\ \le & a_3 \sum_{i=0}^{\infty } e^{-a_2 \frac{n \beta^2\vartheta^{2i}}{r^{2 } } } + \frac{a_3 r^{-\alpha}\vartheta}{\beta ( \vartheta - 1 ) } + p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0 } ) > \gamma\varepsilon].\label{eq : eq421}\end{aligned}\ ] ] since @xmath396 we can find @xmath397 such that @xmath398 thus @xmath399e^{-a_2 \frac{n \beta^{2}\vartheta^{2}}{r^{2}}}.\label{eq : eqab644}\end{aligned}\ ] ] now for fixed @xmath400 choose @xmath401 . then holds for all such @xmath402 . let @xmath403 be large enough so that for all @xmath404 @xmath405 then letting @xmath406 we have for all @xmath404 @xmath407 \le a_4 e^{-a_5n^{\frac{d}{d+2 } } } + a_6 { \varepsilon}^{-(\alpha+1)}n^{-\frac{\alpha}{d+2 } } + p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0 } ) > \gamma\varepsilon ] .\ ] ] choose @xmath408 such that for all @xmath409 , @xmath410 . also let @xmath411 . then for all @xmath412 @xmath407 \le ( a_4 + a_6 ) { \varepsilon}^{-(\alpha+1 ) } n^{-\frac{\alpha}{d+2 } } + p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0 } ) > \gamma\varepsilon].\ ] ] the result follows . we now complete the proof of lemma [ lem : fixn-4.10 ] . the proof uses certain truncation ideas from @xcite . fix @xmath413 . for @xmath414 , @xmath415 and @xmath416 , where @xmath417 , define @xmath418 as @xmath419 for @xmath420 and @xmath415 , let @xmath421 . let @xmath422 be @xmath423 valued random variables which , conditionally on @xmath424 are mutually independent and also independent of @xmath425 , and @xmath426 define @xmath427 it is easily checked that @xmath428 for all @xmath14 and conditionally on @xmath424 , @xmath429 are mutually independent . define @xmath430 . using triangle inequality we have @xmath431 consider first the middle term on the right side of ( [ 4.6.1.2 ] ) . recall @xmath432 then @xmath433 where @xmath434 from lemma [ app1](a ) there exists a finite subset @xmath435 of @xmath436 such that @xmath437 thus @xmath438 & \leq & e\left[\ p \left[\max_{f \in \mathcal{f}_{r,1}^{\frac{\varepsilon}{4}}(\mathbb{b}_r(0))}|\frac{1}{n } \sum_{i=1}^{n } z^{f}_{i , n}| > \frac{\varepsilon}{2}\mid \mathcal{f}_{n-1}^{n}\right ] \right ] \nonumber\\ & \leq & e\sum_{f \in \mathcal{f}_{r,1}^{\frac{\varepsilon}{4}}(\mathbb{b}_r(0 ) ) } p\left [ |\frac{1}{n } \sum_{i=1}^{n } z^{f}_{i , n}| > \frac{\varepsilon}{2 } \ , \big\vert \ , \mathcal{f}_{n-1}^{n } \right].\end{aligned}\ ] ] since @xmath439 , @xmath440 . so by the azuma - hoeffding inequality , @xmath441 & \leq & |\mathcal{f}_{r,1}^{\frac{\varepsilon}{4}}(\mathbb{b}_r(0))| \max _ { f \in \mathcal{f}_{r,1}^{\frac{\varepsilon}{2}}(\mathbb{b}_r(0))}e\left ( p [ |\frac{1}{n } \sum_{i=1}^{n } z^{f}_{i , n}| > \frac{\varepsilon}{2 } \ , \big\vert \ , \mathcal{f}_{n-1}^{n}]\right)\nonumber\\ & \leq & |\mathcal{f}_{r,1}^{\frac{\varepsilon}{4}}(\mathbb{b}_r(0))|2 e^{-\frac{n \varepsilon^{2}}{32 r^{2}}}\nonumber\\ & \leq & 2\ |\mathcal{f}_{r,1}^{\frac{\varepsilon}{4 } } ( [ -r , r]^{d } ) |e^{-\frac{n \varepsilon^{2}}{32r^{2}}}.\end{aligned}\ ] ] from lemma [ app1](b ) we have the following estimate @xmath442 & \leq & \max\left \{2 , \frac{16r}{3\varepsilon}(2\sqrt{d}+1)3^{[\frac{8r}{\varepsilon}(\sqrt{d}+1)]^{d } } \right\ } e^ { -\frac{n \varepsilon^{2}}{32 r^{2}}}.\end{aligned}\ ] ] thus there exist @xmath443 such that for all @xmath444 , @xmath445 @xmath446 \leq k_2 [ e^{k_1(r/{\varepsilon})^d}\vee 1 ] e^ { -\frac{n \varepsilon^{2}}{32 r^{2}}}.\ ] ] for the first term in the right hand side of ( [ 4.6.1.2 ] ) we make use of the observation that if for @xmath447 , @xmath448 are @xmath449 valued random variables distributed according to @xmath450 respectively then @xmath451 thus @xmath452.\end{aligned}\ ] ] using the definition of @xmath453 we see @xmath454 & = & \frac{1}{n } \sum_{i=1}^{n } e[|x_{n}^{i , n } -y_{n}^{i}|1_{|x_{n}^{i , n}| > r } \mid \mathcal{f}_{n-1}^{n}]\nonumber\\ & \leq & \frac{2}{n } \sum_{i=1}^{n } e[|x_{n}^{i , n}|1_{|x_{n}^{i , n}| > r } \mid \mathcal{f}_{n-1}^{n}]\label{eq : eq1036 } \end{aligned}\ ] ] from we have that @xmath455 satisfies @xmath456 thus @xmath457 & \leq & \frac{1}{{\varepsilon } } e\frac{1}{n } \sum_{i=1}^{n } e[|x_{n}^{i , n } -y_{n}^{i}|1_{|x_{n}^{i , n}| > r } \mid \mathcal{f}_{n-1}^{n}]\nonumber\\ & \leq & \frac{2}{\varepsilon } e \{|x_{n}^{i , n}| 1_{|x_{n}^{i , n}| > r } \}\nonumber\\ & \leq & \frac{2 r^{-\alpha}}{\varepsilon } b_{n}^{\alpha } \leq \frac{2 r^{-\alpha}}{\varepsilon } b(\alpha).\label{first}\end{aligned}\ ] ] the third term in ( [ 4.6.1.2 ] ) can be treated similarly . indeed , note that @xmath458 thus using the bound for the right side of the first line in we have that @xmath459\nonumber\\ \leq & \frac{2 r^{-\alpha}}{\varepsilon } b(\alpha).\label{eq : eq1051 } \end{aligned}\ ] ] using , and in we have @xmath460 & \leq & k_2 [ e^{k_1(3r/{\varepsilon})^d}\vee 1 ] e^ { -\frac{n \varepsilon^{2}}{288 r^{2 } } } + \frac{12 r^{-\alpha}}{\varepsilon } b(\alpha).\end{aligned}\ ] ] letting @xmath461 , @xmath462 and @xmath463 , we have that @xmath464 \leq k_5 \left ( e^ { -\frac{k_4 n \varepsilon^{2 } } { r^{2 } } } + \frac { r^{-\alpha}}{\varepsilon } \right)\ ] ] for all @xmath465 . this completes the proof of the lemma . we will proceed as in section [ sec : secprfth5 ] by first first giving a concentration bound for @xmath379 for each fixed @xmath3 and then combining it with in order to obtain a uniform in @xmath3 estimate . we begin by observing that from assumption [ assumg ] it follows that there is a @xmath466 $ ] and @xmath467 such that for all @xmath468 $ ] @xmath469 [ lem:733 ] suppose assumptions [ assumc ] and [ assumg ] hold . let @xmath470 be as in theorem [ th6 ] . then for all @xmath195 and @xmath468 $ ] @xmath471 * proof . * note that for @xmath472 @xmath473 using holder s inequality and taking expectations , for all @xmath468 $ ] @xmath474 where the last inequality is from jensen s inequality . thus for all @xmath468 $ ] @xmath475 where by our assumption @xmath476 . iterating the above inequality we have for all @xmath472 @xmath477 where the second inequality is a consequence of . the result follows . the following lemma is proved in a manner similar to lemma [ lem : fixn-4.10 ] so only a sketch is provided . [ lem : lem352 ] there exist @xmath478 and , for each @xmath479 , @xmath480 such that for all @xmath381 and @xmath382 , @xmath383 & \leq & { \tilde}a_3 \left(e^{-{\tilde}a_2 \frac{n \varepsilon^{2}}{r^{2 } } } + { \tilde}b(\alpha_1)\frac{e^{-\alpha_1 r}}{\varepsilon}\right ) \end{aligned}\ ] ] for all @xmath481 . * proof . * from lemma [ lem:733 ] we have that for @xmath468 $ ] @xmath482 next , as in the proof of lemma [ lem : fixn-4.10 ] , we will use . for the middle term on the right side of we use the same bound as in . now consider the first term in . from we have that @xmath483 & \leq & \frac{2}{\varepsilon}e\left(|x_{n}^{1,n}| 1_{|x_{n}^{1,n}|>r}\right).\label{eq : eq603}\end{aligned}\ ] ] from it follows that for every @xmath479 @xmath484 applying markov s inequality we now have for @xmath479 @xmath483 & \leq & \frac{2}{\varepsilon } e^{-\alpha_1 r}{\tilde}b(\alpha_1).\label{eq : eq613}\end{aligned}\ ] ] the third term in is bounded similarly . indeed , as in we get for @xmath485 @xmath486 \leq \frac{2}{\varepsilon } e^{-\alpha_1 r}{\tilde}b(\alpha_1).\label{eq : eq616}\ ] ] using , and in we now have for @xmath479 @xmath460 & \leq & k_2 [ e^{k_1(3r/{\varepsilon})^d}\vee 1 ] e^ { -\frac{n \varepsilon^{2}}{288 r^{2 } } } + \frac{12 e^{-\alpha_1 r}}{\varepsilon } { \tilde}b(\alpha_1).\end{aligned}\ ] ] thus with @xmath487 as in the proof of lemma [ lem : fixn-4.10 ] and @xmath488 we have @xmath464 \leq k_5 \left ( e^ { -\frac{k_4 n \varepsilon^{2 } } { r^{2 } } } + { \tilde}b(\alpha_1)\frac { e^{-\alpha_1 r}}{\varepsilon } \right)\ ] ] for all @xmath465 . the result follows . we now complete the proof of theorem [ th6 ] . fix @xmath479 . following the steps in the proof of , with @xmath489 as in theorem [ th5 ] , we have from lemma [ lem : lem352 ] , for all @xmath490 and @xmath394 @xmath389 \le { \tilde}a_3 \sum_{i=0}^{\infty } e^{-{\tilde}a_2 \frac{n \beta^2\vartheta^{2i}}{r^{2 } } } + \frac{{\tilde}a_3 { \tilde}b(\alpha_1 ) e^{-\alpha_1 r}\vartheta}{\beta ( \vartheta - 1 ) } + p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0 } ) > \gamma\vartheta^n\varepsilon].\label{eq : eq635}\end{aligned}\ ] ] as before for fixed @xmath400 choose @xmath491 . then holds for all such @xmath402 with @xmath492 replaced by @xmath493 . let @xmath403 be large enough so that for all @xmath404 , holds with @xmath494 replaced by @xmath495 . then letting @xmath496 we have for all @xmath404 @xmath407 \le { \tilde}a_4 e^{-{\tilde}a_5n^{\frac{d}{d+2 } } } + { \tilde}a_6 { \varepsilon}^{-1}\exp ( -{\tilde}a_7 { \varepsilon}n^{\frac{1}{d+2 } } ) + p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0 } ) > \gamma\vartheta^n\varepsilon ] .\ ] ] note that @xmath497 if @xmath498 . consider now the case @xmath200 . then , taking @xmath499 , @xmath500 , @xmath501 , we have for all @xmath502 @xmath407 \le l_2 e^{-l_3 { \varepsilon}n^{1/d+2}}+ p[\mathcal{w}_{1}(\mu_{0}^{n},\mu_{0})>\gamma\vartheta^n\varepsilon].\ ] ] this proves the theorem for the case @xmath503 . finally for @xmath504 , with the same choice of @xmath505 , we have for all @xmath506 @xmath407 \le l_2 e^{-l_3 ( { \varepsilon}\wedge 1 ) n^{1/d+2}}.\ ] ] the result follows . in order to prove the theorem we will introduce an auxiliary sequence @xmath507 such that for each @xmath3 , @xmath508 are i.i.d . we will then employ results from @xcite and @xcite in order to give a uniform ( in @xmath35 ) concentration bound for @xmath509 , where @xmath510 is the empirical measure @xmath511 . finally we will obtain the desired concentration estimate on @xmath512 by making use of lemma [ lem : lemgron ] below . we begin by introducing our auxiliary system . consider the collection of @xmath513 valued random variables @xmath514 defined as follows . @xmath515 note that for each @xmath3 , @xmath508 are i.i.d . in fact , since @xmath516 , we have @xmath517 let @xmath518 the following lemma will give a useful relation between @xmath519 and @xmath520 [ lem : lemgron ] suppose assumptions [ assumc ] and [ assumg ] hold . let @xmath521 . then for every @xmath134 and @xmath400 @xmath522 * proof : * since by assumption [ assumg ] @xmath523 , we have for each @xmath524 @xmath525 thus @xmath526 now note that @xmath527 using triangle inequality @xmath528 applying lemma [ app4 ] with @xmath529 where @xmath530 , we have @xmath531\label{gron1}. \end{aligned}\ ] ] simplifying ( [ gron1 ] ) one gets @xmath532\nonumber\\ & = & \delta m\sum_{k=0}^{n } ( \chi_{1}+\delta m)^{n - k}\mathcal{w}_{1}(\eta_{k}^{n},\mu_{k}).\nonumber\end{aligned}\ ] ] the result now follows by an application of triangle inequality . proof of theorem [ thm6 ] is based on certain results from @xcite and @xcite which we summarize in this section . we begin with the definition of a ` transportation inequality ' . recall that given @xmath533 , the relative entropy of @xmath90 with respect to @xmath77 is defined as @xmath534 define @xmath535 as @xmath536 . [ def : transport ] let @xmath537 be a convex , increasing left continuous function such that @xmath538 . we say @xmath539 satisfies a @xmath540 inequality if for all @xmath541 @xmath542 the following result is established in @xcite . [ thm : thmboiss ] suppose that @xmath539 satisfies a @xmath540 inequality and suppose that there is @xmath543 such that @xmath544 let @xmath545 where @xmath546 are i.i.d . with common distribution @xmath90 . then for @xmath547 @xmath548 where @xmath549 @xmath550 is defined as @xmath551 @xmath552 @xmath553 , @xmath554 and @xmath555 is a positive scaler depending only on @xmath42 . the following result is from @xcite . [ th : thboll ] let @xmath556 . suppose that @xmath557 for some @xmath558 . then @xmath90 satisfies @xmath540 inequality with @xmath559 for any @xmath560 transportation inequalities presented in section [ sec : sectransp ] require exponential integrability of the underlying measure . in this section we show that under assumption [ assumg ] the desired integrability properties hold . [ lem : expinteg ] suppose that assumption [ assumg ] holds and that @xmath561 . then @xmath562 and for all @xmath468 $ ] @xmath563 * proof . * the property that @xmath564 is an immediate consequence of assumptions on @xmath31 . let @xmath565 . from ( [ eq1.1.1 ] ) and the condition @xmath523 we have @xmath566}\nonumber\\ & = & \mathcal{e}_{1}(\alpha_1 \delta ) ee^{\alpha_1 ( e^{-\omega } + \delta m)|x_{n}| + \delta m||\mu_{n}||_{1}}. \label{genu}\end{aligned}\ ] ] from jensen s inequality we have @xmath567 . applying jensen s inequality again to the function @xmath568 we have @xmath569e[e^{\alpha_1 \delta m|x_{n}| } ] . \label{genu1}\end{aligned}\ ] ] note that for any two non - decreasing , non - negative functions @xmath570 on @xmath571 and any @xmath572 , @xmath573 using this inequality in the above display yields the following recursion @xmath574 = \mathcal{e}_{1}(\alpha_1 \delta ) f_{n}(\alpha_1 \kappa_1).\end{aligned}\ ] ] iterating the above inequality we have , for all @xmath134 , @xmath575 thus using we see @xmath576 the result follows . in this section we will give , using results of sections [ sec : sectransp ] and [ sec : expinteg ] , uniform concentration bounds for @xmath578 as @xmath6 . [ cor : cor621 ] suppose that assumption [ assumg ] holds and @xmath561 . then the following hold . \(1 ) there exists a @xmath579 such that @xmath580 and for all @xmath581 , @xmath4 satisfies a @xmath540 inequality with @xmath582 as in and with @xmath583 \(2 ) for all @xmath547 and @xmath581 @xmath584 where @xmath585 is defined by the right side in with @xmath586 replaced with @xmath587 , @xmath588 is defined by the right side of with @xmath589 replaced by @xmath590 and @xmath220 is as in with @xmath221 replaced with @xmath222 . + \(3 ) there exist @xmath591 and @xmath592 such that for all @xmath593 , @xmath594 and @xmath409 @xmath595 \(4 ) there exist @xmath596 such that for all @xmath597 $ ] and all @xmath598 . @xmath599 * proof . * ( 1 ) suppose that the statement in fails to hold for any @xmath600 . then there exist sequences @xmath601 and @xmath602 such that @xmath603 from lemma [ lem : expinteg ] it follows that @xmath604 is tight . suppose along a further subsequence @xmath605 converges to some measure @xmath170 . then sending @xmath606 along this subsequence in and using lemma [ lem : expinteg ] once again we arrive at a contradiction . this proves the first statement in ( 1 ) . the second statement in ( 1 ) is an immediate consequence of theorem [ th : thboll ] . + ( 2 ) this is immediate from part ( 1 ) and theorem [ thm : thmboiss ] . + ( 3 ) it is easy to check that for all @xmath547 , @xmath372 ( see proof of corollary 2.5 in @xcite ) @xmath607 thus recalling the expression for @xmath220 in we see that @xmath608 . choose @xmath591 such that for all @xmath409 and @xmath609 @xmath610 then for all @xmath409 and @xmath609 @xmath611 where the second inequality follows on using the inequality @xmath612 combining this with ( 2 ) completes the proof of ( 3 ) . \(4 ) from the proof of corollary 2.5 of @xcite it follows that for @xmath613 @xmath614 where @xmath615 , @xmath616 and @xmath617 . from note that if @xmath618 @xmath619 thus for all such @xmath620 , @xmath621 . thus if additionally @xmath622 , the right side of is bounded above by @xmath623 . the result follows . in this section we complete the proof of theorem [ thm6 ] . fix @xmath624 . from , for any @xmath197 , @xmath389 & \leq & p[\mathcal{w}_{1}(\eta_{n}^{n},\mu_{n } ) > \gamma \varepsilon ] + \sum_{i=0}^{n-1 } p[\mathcal{w}_{1}(\eta_{i}^{n},\mu_{i } ) \ge \frac{\gamma\varepsilon}{\delta m } ( \frac{1-\gamma}{\chi_{2}})^{n - i}]\quad\quad\nonumber\\ & = & p[\mathcal{w}_{1}(\eta_{n}^{n},\mu_{n } ) > \gamma \varepsilon ] + \sum_{i=1}^{n } p[\mathcal{w}_{1}(\eta_{n - i}^{n},\mu_{n - i } ) \ge \frac{\gamma\varepsilon}{\delta m } \vartheta^{i}]\nonumber\\ & \equiv & t_1 + t_2,\label{unifexp1}\end{aligned}\ ] ] where @xmath625 , which , in view of our assumption on @xmath31 , is strictly larger than @xmath626 . let @xmath627 . then @xmath628 + \sum_{i = i^{{\varepsilon}}+ 1}^{n } p[\mathcal{w}_{1}(\eta_{n - i}^{n},\mu_{n - i } ) \ge \frac{\gamma\varepsilon}{\delta m } \vartheta^{i}].\ ] ] note that since @xmath184 and @xmath629 is non - increasing , @xmath630 implies @xmath631 for all @xmath632 where @xmath633 is as introduced in remark [ rem : vsconstant ] . therefore from lemma [ cor : cor621](4 ) , for all such @xmath1 @xmath634 \le \sum_{i=1}^{i^{{\varepsilon } } } \exp\{-l_2 n m^2_{\gamma}({\varepsilon}\vartheta^i)\}.\ ] ] also , from lemma [ cor : cor621](3 ) , for all @xmath409 , @xmath635 \le \sum_{i = i^{{\varepsilon}}+ 1}^{n } \exp\{-l_1 n m_{\gamma}({\varepsilon}\vartheta^i)\}.\ ] ] combining these estimates and letting @xmath636 and @xmath637 , we have for all @xmath638 @xmath639 where @xmath640 . let @xmath641 be such that for all @xmath642 , @xmath643 . then @xmath644 noting that @xmath645 whenever @xmath646 , we see that with @xmath647 and @xmath648 @xmath649 also from lemma [ cor : cor621 ] , for all @xmath650 @xmath651 where @xmath652 . using and in we now get the desired result with @xmath653 , @xmath654 , @xmath655 and @xmath656 . the first part of the following lemma is an immediate consequence of ascoli - arzela theorem where as the second follows from lemma 5 in @xcite . [ app1 ] ( a ) for a compact set @xmath657 in @xmath41 let @xmath658 be the space of functions @xmath659 such that @xmath660 and @xmath661 for all @xmath662 . then for any @xmath371 there is a finite subset @xmath663 of @xmath658 such that for any signed measure @xmath77 @xmath664 ( b ) if @xmath665^d$ ] for some @xmath415 , then @xmath666 can be chosen such that @xmath667^{d } } , 1 \right\}.\ ] ] [ app2 ] let @xmath668 $ ] be a transition probability kernel . fix @xmath164 and let @xmath669 . let @xmath670 be independent random variables such that @xmath671 let @xmath672 and let @xmath673 , @xmath674 . then @xmath675 jos a carrillo , robert j mccann , and cdric villani . kinetic equilibration rates for granular media and related equations : entropy dissipation and mass transportation estimates . , 19(3):9711018 , 2003 .

let be a simply connected semisimple algebraic group over an algebraically closed field of characteristic 0 and let be a rational simple-module . if is a spherical orbit and if is its closure , then we describe the orbits of and those of its normalization . if moreover the wonderful completion of is strict , then we give necessary and sufficient combinatorial conditions so that the normalization morphism is a homeomorphism . such conditions are trivially fulfilled if is simply laced or if is a symmetric subgroup .

0907.1177

let @xmath0 be a simply connected semisimple algebraic group over an algebraically closed field @xmath1 of characteristic 0 ; all @xmath0-modules considered in the following will be supposed to be rational . an algebraic @xmath0-variety is said to be _ spherical _ if it is normal and if it contains an open @xmath10-orbit , where @xmath11 is a borel subgroup ; a subgroup @xmath12 is said to be _ spherical _ if the homogeneous space @xmath7 is so : any spherical variety can thus be regarded as an open embedding of a spherical homogeneous space , namely its open @xmath0-orbit . important classes of spherical varieties are that of toric varieties and that of symmetric varieties : toric varieties are those spherical varieties whose open orbit is an algebraic torus ; symmetric varieties are those spherical varieties whose generic stabilizer @xmath9 is such that @xmath13 , where @xmath14 is an algebraic involution and where @xmath15 is the set of its fixed points . other important classes of spherical varieties are that of flag varieties and the more general one of wonderful varieties : a _ wonderful variety _ ( of _ rank _ @xmath16 ) is a smooth projective @xmath0-variety having an open @xmath0-orbit which satisfies following properties : * the complement of the open @xmath0-orbit is the union of @xmath16 smooth prime divisors having a non - empty transversal intersection ; * any orbit closure equals the intersection of the prime divisors containing it . a spherical subgroup @xmath9 is said to be _ wonderful _ if @xmath7 possesses a wonderful completion ( which is unique , if it exists ) . by @xcite every self - normalizing symmetric subgroup is wonderful ; more generally every self - normalizing spherical subgroup is wonderful by @xcite . however many natural examples of embeddings of a spherical homogeneous space do not need to be normal . for instance , consider a simple @xmath0-module @xmath2 ( in which case we will call @xmath17 a _ simple projective space _ ) possessing a line @xmath18 $ ] fixed by a spherical subgroup . then consider the orbit @xmath19 \subset { \mathbf p}(v)$ ] , which is spherical , and take its closure @xmath20 } \subset { \mathbf p}(v)$ ] , which generally is not normal ; denote @xmath21 its normalization . the aim of this work is the study of the orbits of compactifications which arise in such a way , and as well the study of the orbits of their normalizations . in @xcite it has been proved that any spherical subgroup which occurs as the stabilizer of a point in a simple projective space is wonderful . if @xmath22 is the wonderful completion of @xmath19 $ ] , then the morphism @xmath19 \to x$ ] extends to @xmath22 and thus we get a morphism @xmath23 : examining such morphism we get a description of the set of orbits of @xmath5 and of @xmath21 . moreover this leads to a combinatorial criterion to establish whether or not two orbits in @xmath22 map onto the same orbit in @xmath5 , which in particular implies that different orbits in @xmath5 are never @xmath0-equivariantly isomorphic . our main theorem ( theorem [ teorema caso stretto ] ) is a combinatorial criterion for @xmath24 to be bijective ; this is done under the assumption that @xmath22 is _ strict _ , i.e. that all isotropy groups of @xmath22 are self - normalizing : strict wonderful varieties , introduced in @xcite , are those wondeful varieties which can be embedded in a simple projective space ; they form an important class of wonderful varieties which generalize the symmetric ones of @xcite . the condition of bijectivity involves the double links of the dynkin diagram of @xmath0 and it is trivially fulfilled whenever @xmath0 is simply laced or @xmath22 is symmetric ; it is easily read off by the _ spherical diagram _ of @xmath22 , which is a useful tool to represent a wonderful variety starting from the dynkin diagram of @xmath0 . main examples of strict wonderful varieties where bijectivity fails arise from the context of _ wonderful model varieties _ introduced in @xcite ; the general strict case is substantially deduced from the model case . a _ model space _ for a connected ( possibly non - simply connected ) semisimple algebraic group @xmath25 is a quasi - affine homogeneous space whose coordinate ring contains every simple @xmath25-module with multiplicity one ; model spaces were classified in @xcite , where it is introduced the _ wonderful model variety _ @xmath26 , whose orbits naturally parametrize up to isomorphism the model spaces for @xmath25 : every orbit of @xmath26 is of the shape @xmath27 , where @xmath28 is a model space , and conversely this correspondence gives a bijection up to isomorphism . in order to illustrate the above mentioned criterion of bijectivity in the case of a wonderful model variety , let s set up some further notation . if @xmath29 is a dominant weight ( w.r.t . a fixed maximal torus @xmath30 and a fixed borel subgroup @xmath31 ) , define the @xmath32 of @xmath29 as the set @xmath33 where @xmath34 is the set of simple roots w.r.t . if @xmath36 is a simple factor of type @xmath37 or @xmath38 , number the associated subset of simple roots @xmath39 starting from the extreme of the dynkin diagram of @xmath40 which contains the double link ; define moreover @xmath41 as the subsets whose element index is respectively even and odd . if they are defined , set @xmath42 @xmath43 or set @xmath44 ( resp . @xmath45 ) otherwise . finally , if @xmath40 is of type @xmath46 , number the simple roots in @xmath47 starting from the extreme of the dynkin diagram which contains a long root . suppose that @xmath19\subset { \mathbf p}(v)$ ] is the open orbit of a wonderful model variety @xmath26 , where @xmath25 is isogenous with @xmath0 ; denote @xmath29 the highest weight of @xmath2 and set @xmath20}$ ] . then the normalization @xmath48 is bijective if and only if the following conditions are fulfilled , for every connected component @xmath49 : * if @xmath50 is of type @xmath37 , then either @xmath51 or @xmath52 ; * if @xmath50 is of type @xmath38 , then @xmath53 ; * if @xmath50 is of type @xmath46 and @xmath54 , then @xmath55 as well . when the generic stabilizer @xmath9 is a self - normalizing symmetric subgroup , compactifications in simple projective spaces were studied in @xcite . under this assumption , an explicit description of the orbits of @xmath5 was given and it was proved that these orbits are equal to those of the normalization of @xmath5 . thus our results generalize those contained in @xcite . in the case of a compactification of the adjoint group @xmath56 ( regarded as a @xmath57-symmetric variety ) obtained as the closure of the orbit of the line generated by the identity in a projective space @xmath58 ( where @xmath2 is a simple @xmath0-module ) , a complete classification of the normality and of the smoothness has been given in @xcite . the paper is organized as follows . in section 1 , we set notations and preliminaries ; in section 2 we give some general results about spherical orbit closures in projective spaces ; in section 3 we recall some results from @xcite about stabilizers of points in simple projective spaces . in section 4 , we describe the orbits of the compactifications @xmath5 and @xmath21 ; in section 5 , we prove the criterion of bijectivity of the normalization map in the strict case ; in section 6 , we briefly consider the non - strict case giving some sufficient conditions of bijectivity and non - bijectivity of the normalization map . + _ aknowledgements . _ i want to thank a. maffei , who proposed me the problem , for all his precious help , and p. bravi for many useful discussions on the subject . as well , i want to thank the referees for their careful reading and useful comments . spherical diagrams have been made with the package _ lunadiagrams _ , made by p. bravi and available at http://www.mat.uniroma1.it/@xmath59bravi/lunadiagrams . fix a borel subgroup @xmath11 and a maximal torus @xmath35 ; denote @xmath60 the corresponding root system and @xmath61 the corresponding set of simple roots . if @xmath62 is any subgroup , denote @xmath63 its character group ; if @xmath2 is a @xmath0-module , denote @xmath64 the set of @xmath9-eigenvectors of @xmath2 and , if @xmath65 , denote @xmath66 the subset of @xmath64 where @xmath9 acts by @xmath67 . if @xmath68 is a dominant weight , we will denote @xmath69 the simple @xmath0-module with highest weight @xmath29 . if @xmath70 is a lattice ( i.e. a finitely generated free @xmath71-module ) , then @xmath72 denotes the dual lattice and @xmath73 denotes the rational vector space generated by @xmath70 . if @xmath74 is a cone contained in some vector space @xmath2 , then @xmath75 denotes the dual cone in the dual vector space @xmath76 . let @xmath5 be a spherical @xmath0-variety and fix a base point @xmath77 in such a way that @xmath78 is an open subset ; denote @xmath79 . let s introduce some data associated to @xmath5 : * @xmath80 . * @xmath81 , its elements are called the _ colors _ of @xmath5 . if @xmath82 is an orbit , then @xmath83 denotes the set of colors which contain @xmath84 . * if @xmath85 is a rational discrete valuation of @xmath86 , then @xmath87 defines an element @xmath88 by @xmath89 where @xmath90 is any @xmath10-semiinvariant function of weight @xmath67 : since @xmath5 possesses an open @xmath10-orbit , such definition does not depend on the function , but only on the weight . if @xmath91 , by abuse of notation we will denote @xmath92 the image of the respective valuation @xmath93 . + a rational discrete valuation @xmath87 is said @xmath0-invariant if @xmath94 , for any @xmath95 and for any @xmath96 ; set @xmath97 the set of @xmath0-invariant rational valuations of @xmath86 . the map @xmath98 identifies @xmath99 with a convex cone which generates @xmath100 as a vector space ( * ? ? ? 4.1 ) ; together with such embedding , @xmath97 is called the @xmath0-_invariant valuation cone of _ @xmath5 . both @xmath101 and @xmath102 , as well as the map @xmath103 , depend only on the open orbit @xmath104 and they are the main objects of the _ luna - vust theory _ ( see @xcite ) , which classifies normal equivariant embeddings of a given spherical homogeneous space . a spherical variety is said to be _ simple _ if it possesses only one closed orbit ; it is said to be _ toroidal _ if no color contains a closed orbit . if a spherical homogeneous space @xmath7 possesses a complete , simple and toroidal embedding , then this is uniquely determined and it is called the _ canonical embedding _ of @xmath7 ; we will denote it @xmath105 and it dominates any simple complete embedding of @xmath7 . in general , a canonical embedding of @xmath7 exists if and only if the index of @xmath9 in its normalizer is finite , in which case @xmath9 is said to be _ if @xmath5 is a simple spherical variety with complete closed orbit @xmath84 , then the picard group @xmath106 is freely generated by the classes @xmath107 $ ] , with @xmath108 ; moreover , a divisor is generated by global sections ( resp . ample ) if and only if it is equivalent to a linear combination of such colors with non - negative ( resp . positive ) coefficients ( * ? ? ? 2.6 and thm . wonderful varieties are always spherical ( see @xcite ) and a spherical variety is wonderful if and only if it is complete , toroidal , simple and smooth . a spherical subgroup which appears as the generic stabilizer of a wonderful variety is said _ wonderful_. if @xmath9 is a spherical subgroup , then the normalizer @xmath109 acts on the right on @xmath7 by @xmath110 . consider the induced action of @xmath109 on @xmath111 : the kernel of such action is called the _ spherical closure _ of @xmath9 ; if @xmath9 coincides with its spherical closure , then it is called _ spherically closed_. spherically closed subgroups are always wonderful ( * ? ? ? 7.6 ) ; a wonderful variety is said to be _ spherically closed _ if its generic stabilizer is so . suppose now @xmath22 is a wonderful variety with open @xmath10-orbit @xmath112 and set @xmath79 ; suppose moreover that the center of @xmath0 acts trivially on @xmath22 . denote @xmath113 the set of colors and @xmath114 the closed orbit ; denote @xmath115 the @xmath116-fixed point ( where @xmath116 denotes the opposite borel subgroup of @xmath10 with respect to @xmath117 ) . since @xmath0 is semisimple and simply connected , @xmath118 is identified with a sublattice of @xmath119 , while @xmath120 is identified with @xmath63 : if @xmath121 , then @xmath122 will be identified with the character of @xmath116 acting on the fiber of @xmath122 over @xmath123 , while if @xmath124 , then @xmath122 will be identified with the character of @xmath9 acting on the fiber over @xmath125 . let s introduce some more data attached to a wonderful variety @xmath22 , together with some results which can be found with more details and references in @xcite and in @xcite . * @xmath126 ; its elements are called the _ spherical roots _ of @xmath22 . spherical roots form a basis of the lattice @xmath127 ; they also coincide with the minimal set of generators of the free semigroup @xmath128 . the cardinality of @xmath129 coincides with the _ rank _ of @xmath22 , i.e. with the number of @xmath0-stable prime divisors of @xmath22 , which are naturally in correspondence with spherical roots . if @xmath130 , denote @xmath131 the corresponding @xmath0-stable prime divisor : it is a wonderful @xmath0-subvariety whose set of spherical roots is @xmath132 . spherical roots are either positive roots or a sum of two positive roots ; the _ support _ of a spherical root @xmath133 is the set of simple roots where @xmath133 is supported . * the _ cartan pairing _ of @xmath22 is the natural pairing @xmath134 between colors and spherical roots defined by the equality @xmath135 = \sum_{d \in { \delta } } c(d,{\sigma } ) [ d]$ ] in the picard group @xmath136 . if @xmath137 and @xmath130 , then @xmath138 . * @xmath139 = @xmath140 is the set of colors _ moved by _ @xmath141 , where @xmath142 and where @xmath143 is the minimal parabolic subgroup associated to @xmath141 . for every @xmath142 it holds @xmath144 . * @xmath145 ; it coincides with the set of simple roots associated to the stabilizer of the open @xmath10-orbit , which is a parabolic subgroup . as well , @xmath146 coincides with the set of simple roots associated with the stabilizer of the @xmath116-fixed point @xmath123 in the closed orbit @xmath84 . * @xmath147 is the set of simple spherical roots ; correspondingly , set @xmath148 the _ set of colors of type a_. if @xmath149 , set @xmath150 ; then for every spherical root @xmath133 it holds @xmath151 * @xmath152 ; correspondingly , set @xmath153 the _ set of colors of type 2a_. if @xmath154 , set @xmath155 ; then for every spherical root @xmath133 it holds @xmath156 * @xmath157 ; correspondingly , set @xmath158 the _ set of colors of type b_. if @xmath159 , set @xmath155 ; then for every spherical root @xmath133 it holds @xmath160 * @xmath161 is the _ spherical system _ of @xmath22 , where @xmath162 has to be thought of as an abstract set together with the pairing @xmath163 . this is the combinatorial datum which expresses a wonderful variety : each wonderful variety is uniquely determined by its spherical system ( see @xcite ) . there is also an abstract combinatorial definition of spherical system ( see @xcite or @xcite ) , introduced in order to obtain the classification of wonderful varieties ( _ luna s conjecture _ ) : the geometrical realizability of spherical systems has been checked in many cases ( see @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ) and recently a general proof which avoids a case - by - case approach has been proposed in @xcite . a very useful tool to represent spherical systems are _ spherical diagrams _ , obtained adding information to the dynkin diagram of @xmath60 ( see @xcite ) . * denote @xmath164 and @xmath165 the restrictions to the closed and to the open orbit ; then we get a commutative diagram @xmath166 \ar@{->}_{{\omega}}[d ] & { \mathcal x}(h ) \ar@{->}[d ] \\ { \mathcal x}(b ) \ar@{->}[r ] & { \mathcal x}(b\cap h ) } \ ] ] which identifies @xmath167 with the fiber product ( * ? ? ? 2.2.1 ) @xmath168 on the combinatorial level , the map @xmath169 is described on colors as follows ( ( * ? ? ? * thm . 2.2 ) ) : @xmath170 where @xmath171 is the fundamental dominant weight associated to @xmath142 . * if @xmath172 is a subset of spherical roots , then the _ localization at @xmath173 _ of @xmath22 is the @xmath0-stable subvariety @xmath174 it is a wonderful variety whose spherical system is @xmath175 , where @xmath176 . denote @xmath177 the set of colors of @xmath178 ; if @xmath179 and @xmath180 set @xmath181 and @xmath182 . let @xmath183 be the pullback map ; then @xmath169 factors through @xmath184 and by its combinatorial description it follows that : * * if @xmath185 then @xmath186 ( resp . @xmath187 ) is supported on @xmath188 ( resp . on @xmath189 ) with multiplicity one , while it is not supported on @xmath189 ( resp . on @xmath188 ) ; * * if @xmath190 then @xmath186 and @xmath187 are supported on @xmath191 with multiplicity one ; * * if @xmath192 , then @xmath193 ; * * if @xmath194 , then @xmath195 ; * * if @xmath159 , then @xmath196 is supported on @xmath191 with multiplicity one and on at most one more color . * @xmath22 is said to be _ strict _ if the stabilizer of any point is self - normalizing ; equivalently , we will say also that @xmath9 is strict . a wonderful variety is strict if and only if it can be embedded in a simple projective space ( see @xcite ) . * consider the following sets of spherical roots @xmath197 set @xmath198 the _ set of loose spherical roots_. loose spherical roots of the first kind are easily described , they are those of the following types ( where @xmath199 and simple roots are labelled as in bourbaki ) : * * spherical roots @xmath200 with support of type @xmath201 and with @xmath202 ; * * spherical roots @xmath203 with support of type @xmath204 . + for every @xmath205 , it is defined a @xmath0-equivariant automorphism @xmath206 of order 2 which fixes pointwise the @xmath0-stable divisor @xmath131 associated to @xmath133 . if @xmath207 , then @xmath208 acts trivially on @xmath113 , while if @xmath209 , then @xmath208 exchanges @xmath210 and @xmath211 and acts trivially on @xmath212 . such automorphisms commute and generate @xmath213 ( see @xcite ) . + by the natural identification @xmath214 , it follows that * * @xmath9 is self - normalizing if and only if @xmath215 ; * * @xmath9 is spherically closed if and only if @xmath216 ; * * @xmath9 is strict if and only if @xmath217 and @xmath215 . + in particular , if @xmath218 , then @xmath9 is self - normalizing if and only if it is spherically closed if and only if it is strict . * @xmath219 denotes the set of the _ spherical roots of _ @xmath0 : its elements are the spherical roots of all possible rank one @xmath0-wonderful varieties . following the classification of the latter ( see @xcite ) , such set is classified for any @xmath0 ( see @xcite and @xcite ) . let @xmath2 be a @xmath0-module and let @xmath220 be a projective variety with an open @xmath10-orbit ; denote @xmath221 the open orbit . since @xmath5 contains finitely many @xmath10-orbits ( * ? ? ? 2.6 ) , every @xmath0-orbit in @xmath5 is spherical . let @xmath222 be the normalization ; then @xmath21 is a complete spherical variety with the same open orbit of @xmath5 whose orbits are naturally in bijection with those of @xmath5 : [ @xcite prop . 1 ] [ timashev ] the normalization morphism @xmath222 induces a bijection between @xmath0-orbits . if @xmath223 is an orbit , set @xmath224 the corresponding orbit . denote @xmath225 and @xmath226 the open @xmath10-orbits ; fix base points @xmath227 and @xmath228 so that we have isomorphisms @xmath229 with @xmath230 . let s recall a result which will be useful in the following : [ @xcite prop 5.1 and cor . 5.2 ] [ brion ] let @xmath9 be a spherical subgroup of @xmath0 . * the algebraic group @xmath231 is diagonalizable ; moreover , if @xmath232 is the identity component of @xmath9 , then @xmath233 . * if @xmath10 is any borel subgroup such that @xmath234 is open in @xmath0 , then @xmath109 equals the right stabilizer of @xmath234 . combing back to our situation , then we obtain : [ stabilizzatori sobri ] @xmath235 is a normal subgroup with finite index ; in particular @xmath236 is a finite diagonalizable group . since @xmath237 is a finite morphism , it preserves dimensions of orbits : so we have @xmath238 . then @xmath235 implies @xmath239 and we obtain @xmath240 this shows @xmath241 and the claim follows . [ indice ] let @xmath230 be two spherical subgroups of @xmath0 with @xmath242 normal in @xmath243 ; fix a borel subgroup @xmath10 such that @xmath244 is open in @xmath0 and consider the projection @xmath245 . then @xmath246 and @xmath247 identifies @xmath248 with a sublattice of @xmath249 such that @xmath250 first claim follows by the equality @xmath251 , which stems immediately from theorem [ brion ] ii ) . if @xmath252 , denote @xmath253 the kernel of the restriction @xmath254 , which is surjective by the following argument : if @xmath255 is the unipotent radical , then @xmath256 and @xmath257 and the restriction @xmath258 is surjective since @xmath259 is a diagonalizable subgroup of the torus @xmath260 . by definition , we have isomorphisms @xmath261 and @xmath262 ; thus the restriction gives a surjective homomorphism @xmath263 whose kernel is @xmath248 . on the other hand @xmath264 and @xmath265 , hence the equality @xmath251 implies @xmath266 therefore we get @xmath267 going back to our situation , since @xmath236 is a finite diagonalizable group , it is isomorphic to its character group and we get the following corollary . [ indice2 ] let @xmath268 be an orbit and let @xmath269 , with @xmath230 ; then @xmath270 fix a closed orbit @xmath82 . since parabolic subgroups are self - normalizing , @xmath84 and @xmath271 are isomorphic ; from now on , we will denote both of them with the same letter @xmath84 . let @xmath272 \in y^{b^-}$ ] be the unique fixed point by @xmath116 ( where @xmath273 is a lowest weight vector ) and let @xmath274 be a highest weight vector such that @xmath275 . if @xmath276 is the stabilizer of @xmath277 $ ] , then @xmath276 and @xmath278 are opposite parabolic subgroups ; denote @xmath279 the associated levi subgroup . consider the affine @xmath276-stable open subset @xmath280 defined by the non - vanishing of @xmath281 and recall that there exists an affine closed @xmath282-stable subvariety @xmath283 containing @xmath284 and possessing an open @xmath285-orbit such that the multiplication morphism @xmath286 ( where @xmath287 denotes the unipotent radical of @xmath276 ) is a @xmath276-equivariant isomorphism ( * ? ? ? 1.2 ) . since @xmath288 = k[s_x]^l = k$ ] , we get that @xmath289 possesses a unique closed @xmath282-orbit , namely the fixed point @xmath284 . if @xmath290 , denote @xmath291 and @xmath292 its closure respectively in @xmath5 and in @xmath21 . [ chiusura colori ] let @xmath290 , let @xmath223 be an orbit and set @xmath293 . then @xmath294 if and only if @xmath295 . by lemma [ indice ] , @xmath296 is the open @xmath10-orbit of @xmath297 . suppose that @xmath298 and fix @xmath227 ; if @xmath299 , then we obtain @xmath300 , which implies @xmath301 . suppose conversely that @xmath295 : then @xmath302 . [ colori xl ] let @xmath290 ; then @xmath303 if and only if @xmath304 . notice that @xmath305 is closed and that it contains @xmath84 if and only if @xmath291 contains @xmath84 . if @xmath304 , then @xmath306 , thus @xmath307 is non - empty , @xmath282-stable and closed in @xmath289 : hence @xmath308 , which implies that @xmath309 . if conversely @xmath310 , then @xmath311 and we get @xmath312 . combining previous lemmas it follows that the set of colors @xmath313 whose closure in @xmath21 contains the closed orbit @xmath84 is @xmath314 if @xmath5 possesses a unique closed orbit @xmath84 , then previous lemmas allow us to compute the _ colored cone _ of the normal embedding @xmath315 : this is the couple @xmath316 , where @xmath317 is the cone generated by the elements @xmath318 , where @xmath319 is any @xmath10-stable ( possibly @xmath0-stable ) prime divisor which contains @xmath84 ( see @xcite ) . in fact , since @xmath21 is simple and complete , then @xmath320 contains the @xmath0-invariant valuation cone @xmath321 ( ( * ? ? ? * thm . 5.2 ) ) : therefore @xmath320 is the cone generated by @xmath321 together with @xmath322 . let @xmath223 be an orbit . set @xmath323 , @xmath324 and @xmath325 ; then @xmath326 and @xmath327 . considering the rings of functions , we get a commutative diagram @xmath328 \ar@{^{(}->}[r ] \ar@{->>}[d ] & k[{\widetilde}{x}_0 ] \ar@{->>}[d ] \\ k[z_0 ] \ar@{^{(}->}[r ] & k[z'_0 ] } \ ] ] [ @xcite thm . 2.3 ] [ suriettivita semiinvarianti ] every @xmath10-semiinvariant function @xmath329^{(b)}$ ] ( resp . @xmath330^{(b)}$ ] ) can be extended to a @xmath10-semiinvariant function @xmath331^{(b)}$ ] ( resp . @xmath332^{(b)}$ ] ) . if @xmath70 is a finitely generated free @xmath71-module and @xmath333 is a submonoid , then the _ saturation _ of @xmath334 in @xmath70 is the submonoid @xmath335 , where @xmath336 is the cone generated by @xmath334 . if @xmath337 is a submonoid containing @xmath334 , then we will say that @xmath334 is _ saturated in _ @xmath338 if @xmath339 . [ reticoli saturi ] fix an orbit @xmath223 and set @xmath293 . * @xmath340 is the saturation of @xmath341 in @xmath127 . * if @xmath342 , then @xmath343 is the normalization of @xmath344 . set @xmath345^{(b)}/k^*$ ] and @xmath346^{(b)}/k^*$ ] ; in terms of colored cones , such monoids are described as follows ( ( * ? ? ? * thm . 3.5 ) ) : @xmath347 since every @xmath10-semiinvariant function is uniquely determined by its weight up to a scalar factor , by previous theorem restriction gives an isomorphism of multiplicative monoids @xmath348^{(b ) } { \ , : \,}f{\bigr|}_{z'}\not \equiv 0 \right\ } \stackrel{\sim}{{\longrightarrow } } k[z'_0]^{(b)}:\ ] ] since the first one is saturated in @xmath349^{(b)}$ ] , we may then identify @xmath350 with a saturated submonoid of @xmath351 . hence we may as well identify @xmath340 with a sublattice of @xmath127 : in fact every @xmath10-semiinvariant rational function on @xmath21 ( resp . on @xmath297 ) can be written as a quotient of two @xmath10-semiinvariant regular functions on @xmath352 ( resp . on @xmath353 ) . since @xmath21 is normal , @xmath354 is saturated ; therefore , being saturated in @xmath351 , we see that @xmath355 as well is saturated . since the colored cone of a spherical variety is strictly convex ( ( * 4.1 ) ) , @xmath356 has maximal dimension : thus the equality @xmath357 implies that @xmath358 is a saturated sublattice . hence by @xmath359 = [ k : k ' ] < \infty$ ] we get i ) . finally , ii ) stems from the fact that a @xmath0-stable subvariety of a spherical variety is normal ( * ? ? ? 3.5 ) together with the fact that the restriction @xmath360 is finite and birational . let @xmath2 be a simple @xmath0-module and suppose @xmath361 is a spherical orbit . a necessary and sufficient condition so that a spherical subgroup @xmath9 arises in such a way has been given in @xcite : [ sferically closed ] a spherical subgroup occurs as the stabilizer of a point in a simple projective space if and only if it is spherically closed . let @xmath22 be the wonderful completion of @xmath7 : then the embedding @xmath362 extends to a morphism @xmath363 and it determines a @xmath10-stable effective divisor @xmath364 which is generated by global sections . conversely , it is possible to start with an arbitrary spherically closed wonderful variety @xmath22 together with an effective @xmath10-stable divisor @xmath365 generated by global sections and to consider then the associated morphism @xmath366 , where @xmath367 is the dual of the simple module generated by the canonical section @xmath368 . in @xcite , there have been given necessary and sufficient conditions on such a divisor @xmath365 so that @xmath369 restricts to an embedding of the open orbit @xmath7 ; the aim of this subsection is to recall some facts which lead to such conditions . let @xmath22 be a wonderful variety with base point @xmath370 and set @xmath79 . set @xmath161 its spherical system and @xmath371 its set of colors . recall that a subset @xmath372 is said to be _ distinguished _ if there exists @xmath373 such that @xmath374 , for every @xmath130 . if @xmath375 is a sober subgroup and if @xmath376 is the projection , then the subset of colors @xmath377 is distinguished ; conversely , if @xmath378 is a distinguished subset , then there exists a unique wonderful subgroup @xmath379 with @xmath380 connected such that @xmath381 . this is the content of the following theorem : [ @xcite thm . 5.4 , @xcite prop . 3.3.2 , @xcite thm . 3.1.1 ] [ corrispondenza distinta ] @xmath382 + there is an inclusion - preserving bijection as follows @xmath383 moreover , if @xmath375 is a wonderful subgroup with @xmath380 connected and if @xmath378 is the corresponding distinguished subset , then * the projection @xmath384 identifies @xmath385 with @xmath386 ; * the spherical system of the wonderful completion of @xmath387 is @xmath388 defined as follows : * * @xmath389 is the set of indecomposable elements of the free semigroup @xmath390 * * @xmath391 ; * * @xmath392 , and the pairing is obtained by restriction . in the notations of previous theorem , the wonderful completion of @xmath387 is denoted @xmath393 and it is called the _ quotient wonderful variety _ of @xmath22 by @xmath394 , while @xmath395 is called the _ quotient spherical system _ of @xmath396 by @xmath394 . [ distinti saturi ] by ( * ? ? ? * lemma 5.3 and thm . 5.4 ) together with ( * ? ? ? * lemma 3.3.1 ) , there is an inclusion - preserving bijection between distinguished subsets @xmath378 and sober subgroups @xmath375 such that @xmath380 is connected . in ( 5.6.2 ) it was proved that , in case @xmath0 is of type @xmath397 , then such a subgroup @xmath398 is necessarily wonderful ; although this was claimed in general in @xcite , a general proof ( which stems from the classification of spherical systems ) appeared only recently in ( * ? ? ? 3.3.1 ) . suppose that @xmath375 is a sober subgroup such that @xmath380 is connected and denote @xmath378 the distinguished subset of colors which map dominantly to @xmath387 . denote @xmath399 the canonical embedding and extend the projection @xmath400 to a morphism @xmath401 ; consider @xmath402 as a sublattice of @xmath127 . set @xmath403 : it is a linear subspace which contains @xmath404 and which intersects the valuation cone @xmath321 in a face ; as a cone , @xmath405 is generated by this face together with @xmath404 ( * ? ? ? * lemma 3.3.1 ) . while the valuation cone @xmath406 is the image of @xmath321 under the quotient map @xmath407 by @xmath405 , the lattice @xmath402 is identified with a sublattice of @xmath127 as follows @xmath408 ( * ? ? ? * lemma 5.3 and thm . 5.4 ) ; as a consequence , @xmath402 is saturated in @xmath127 . if @xmath177 is the set of colors of @xmath409 and if @xmath410 , then @xmath411 ; since the fibers of @xmath412 are complete and connected , it follows that @xmath413 = k[\phi^{-1}(m'_0)]$ ] . considering the @xmath10-semiinvariant functions , we get then the identification of semigroups @xmath414^{(b ) } } } \ ! \raisebox{-.60ex}{\ensuremath{k^ * } } } } \simeq - { \ensuremath{\raisebox{.40ex}{\ensuremath{{\mathbf n}{\sigma } } } \ ! \big / \ ! \raisebox{-.60ex}{\ensuremath{{\delta}^*}}}}.\ ] ] if @xmath409 is smooth , then such semigroup is free ; conversely , given any distinguished subset @xmath372 , the semigroup @xmath415 is free ( * ? ? ? 3.1.1 ) , which means that @xmath409 is necessarily smooth . recall the restrictions to the closed and to the open orbit @xmath164 and @xmath416 and let @xmath417 with @xmath418 for every @xmath419 . if @xmath420 is the canonical section , then the submodule @xmath421 generated by @xmath422 is identified with the simple module @xmath423 ( which contains a unique @xmath9-invariant line where @xmath9 acts by @xmath424 ) and we get a morphism @xmath425 define the _ support _ of @xmath365 as @xmath426 as a consequence of theorem [ corrispondenza distinta ] , we get the following corollary . [ distinti stabilizzatori ] let @xmath22 be a wonderful variety and let @xmath427 be a divisor generated by global sections and consider the associated morphism @xmath428 . then the correspondence of theorem [ corrispondenza distinta ] gives an inclusion - preserving bijection as follows @xmath429 let @xmath375 be a wonderful subgroup with @xmath380 connected and set @xmath372 the corresponding distinguished subset . if @xmath430 is the wonderful embedding , then the projection @xmath384 extends to a morphism @xmath431 and pullback identifies @xmath432 with the submodule @xmath433 \subset { \mathbf z}{\delta}= \operatorname{pic}(m)$ ] . thus the map @xmath434 factors through a map @xmath435 if and only if @xmath436 . [ @xcite ] [ coppia fedele ] let @xmath22 be a spherically closed wonderful variety and denote @xmath113 its set of colors . a divisor generated by global sections @xmath437 is called _ faithful _ if it satisfies the following conditions : ( fd1 ) : : every non - empty distinguished subset of @xmath113 intersects @xmath438 ; ( fd2 ) : : if @xmath439 is a loose spherical root , then @xmath440 . [ @xcite prop . 2.4.3 ] [ fedelta ] let @xmath22 be a spherically closed wonderful variety and let @xmath427 . then the morphism @xmath441 restricts to an embedding @xmath442 if and only if @xmath365 is faithful . fix @xmath443 a representative of the line @xmath444 and suppose that @xmath445 $ ] ; then previous corollary implies * ( fd1)*. suppose by absurd that * ( fd2 ) * fails and let @xmath446 be a loose spherical root such that @xmath447 . if @xmath448 is the corresponding automorphism , then @xmath449 exchanges @xmath450 and @xmath451 and fixes every other color @xmath452 : therefore @xmath449 fixes @xmath365 . the action of @xmath213 on @xmath453 is defined extending by linearity the right action of @xmath231 on @xmath113 , i.e. by the action of @xmath109 on @xmath63 . therefore , if @xmath454 is a representative of @xmath449 , then @xmath455 , i.e. @xmath456 moves the line @xmath457 $ ] in a line where @xmath9 acts by the same character : since @xmath9 is spherical , such a line is unique , thus @xmath458 $ ] which is absurd . suppose conversely that @xmath365 is a faithful divisor . by * ( fd1 ) * it follows that @xmath459 $ ] , therefore by theorem [ brion ] we get @xmath460 \subset n_g(h)$ ] . suppose by absurd that there exists @xmath461 { \smallsetminus}h$ ] . then @xmath455 and the equivariant automorphism corresponding to the coset @xmath462 fixes @xmath365 : therefore by * ( fd2 ) * we get that every color @xmath463 is fixed by @xmath456 . on the other hand , since @xmath9 is spherically closed , every element in @xmath464 acts non - trivially on @xmath113 . take @xmath142 such that @xmath456 moves @xmath465 , then we get @xmath446 and @xmath466 : therefore @xmath467 , which contradicts * ( fd2)*. [ sistema sferico stabilizzatore ] in the same hypotheses of corollary [ distinti stabilizzatori ] , suppose moreover that every distinguished subset of @xmath113 intersects @xmath438 and set @xmath468 then the spherical system of @xmath469 is @xmath175 , where @xmath470 for every loose spherical root @xmath205 , the quotient @xmath471 is easily proved to be a wonderful variety , whose spherical system is @xmath472 , where @xmath473 and where @xmath474 . if @xmath454 is a representative of the coset corresponding to @xmath208 , then @xmath475 , where @xmath476 is the subgroup generated by @xmath9 together with @xmath456 . by the first part of the proof of previous theorem it follows that @xmath476 fixes @xmath444 , thus we get a commutative diagram @xmath477 \ar@{->}[drr ] & & \\ { \ensuremath{\raisebox{.40ex}{\ensuremath{m } } \ ! \raisebox{-.60ex}{\ensuremath{{\gamma}({\sigma } ) } } } } \ar@{->}[rr ] & & { \mathbf p}(v^*_{{\omega}({\delta } ) } ) } \ ] ] consider now the quotient variety @xmath478 , where @xmath479 is the subgroup generated by the elements @xmath208 , with @xmath480 : then , by previous discussion and by proposition [ fedelta ] , it follows that @xmath478 is a spherically closed wonderful variety endowed with a faithful divisor whose associated characters are the same of @xmath365 . in the hypotheses of previous corollary , the assumption that every distinguished subset of colors intersects @xmath438 ( which is equivalent to assume that @xmath9 and @xmath481 have the same dimension ) involves no loss of generality : we can always reduce to that case considering , instead of @xmath22 , the quotient wonderful variety @xmath482 , where @xmath483 is the maximal distinguished subset which does not intersect @xmath438 . let @xmath22 be a spherically closed wonderful variety with base point @xmath370 and set @xmath79 ; set @xmath161 its spherical system and @xmath371 its set of colors . if @xmath364 is a faithful divisor , set @xmath486 and consider the morphism @xmath366 . set @xmath487 and set @xmath488 the normalization ; then we get a commutative diagram @xmath489_{\phi_{\delta } } \ar@{->>}[rr]^{\tilde{\phi}_{\delta } } & & { \widetilde{x}_{\delta}}\ar@{->>}[d]^{p } \\ & & { x_{\delta } } & \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\ ! \subset { \mathbf p}(v ) } \ ] ] if @xmath490 is the closed orbit , then by lemma [ chiusura colori ] and lemma [ colori xl ] we get that @xmath491 is canonically identified with @xmath492 . if @xmath493 is an orbit , in the following @xmath494 will denote the pullback of @xmath364 . notice that if @xmath142 then @xmath495 where @xmath496 denotes the set of simple roots associated to the closed orbit @xmath84 . if @xmath22 is strict , then the variety @xmath484 depends only on the support of @xmath365 ( * ? ? ? * lemma 1 ) : if @xmath497 are faithful divisors on @xmath22 , then @xmath498 as will be shown by lemma [ inclusioni s(d ) ] , this is not true if @xmath22 is not strict . [ sistema sferico orbita xlt ] let @xmath499 be an orbit and let @xmath269 ; let @xmath500 be any orbit which maps on @xmath501 and choose the stabilizers so that @xmath502 . then @xmath242 is the maximal subgroup such that @xmath503 in particular , @xmath342 if and only if @xmath504 is connected . set @xmath505 the maximal subgroup of @xmath243 containing @xmath506 such that @xmath507 is connected . since @xmath508 and since @xmath509 , by theorem [ brion ] we get that @xmath510 is a normal subgroup ; thus by lemma [ indice ] it follows that @xmath511 if and only if @xmath512 . consider the inclusions @xmath513 : since @xmath514 is saturated in @xmath127 , proposition [ reticoli saturi ] shows that @xmath340 is the saturation of @xmath341 in @xmath514 . on the other hand , by remark [ distinti saturi ] it follows that @xmath515 is saturated in @xmath514 : since @xmath516 = [ k : k^ * ] < \infty$ ] , we get the equality @xmath512 . combining previous proposition together with ( 3.3.1 ) and corollary [ sistema sferico stabilizzatore ] we get the following corollary . let @xmath499 be an orbit and let @xmath517 with @xmath235 . then @xmath242 is a wondwerful subgroup . if moreover @xmath22 is strict , then @xmath243 is the spherical closure of @xmath242 . if @xmath518 is an orbit and if @xmath519 is the corresponding orbit , denote @xmath520 the sets of spherical roots of the respective wonderful completions . by corollary [ sistema sferico stabilizzatore ] there exists a bijection between @xmath521 and @xmath522 , which associates to @xmath523 the unique @xmath524 which is proportional to @xmath525 : more precisely , if @xmath526 , then @xmath527 . if @xmath528 , then we say that : * @xmath133 is of _ type _ @xmath529 if @xmath530 has support of type @xmath531 ; * @xmath133 is of _ type _ @xmath532 if @xmath533 has support of type @xmath531 ; * @xmath133 is of _ type _ @xmath534 if @xmath203 has support of type @xmath535 ; * @xmath133 is of _ type _ @xmath536 if @xmath537 has support of type @xmath535 . consider a spherical root @xmath528 such that @xmath538 : following the explicit description of @xmath219 , such a root either is a simple root , or it is of type @xmath529 or it is of type @xmath534 . if @xmath539 is any orbit and if @xmath519 is the corresponding orbit , define @xmath540 to be the subset of spherical roots which have to be doubled to get the spherical roots of @xmath501 . [ orbite non isomorfe e radici sferiche cong ] an orbit @xmath518 is not isomorphic to its corresponding orbit @xmath541 if and only if @xmath501 possesses a spherical root @xmath525 of the shape @xmath542 , where @xmath543 are pairwise distinct elements ( and where @xmath544 ) . by corollary [ sistema sferico stabilizzatore ] , @xmath501 and @xmath297 are not isomorphic if and only if @xmath545 ; suppose @xmath546 . by proposition [ sistema sferico orbita xlt ] the wonderful completion of @xmath297 is the quotient of a wonderful subvariety @xmath547 ; therefore we can write @xmath548 , where @xmath549 are spherical roots of @xmath409 . since @xmath550 , by the discussion preceeding the lemma @xmath551 is either a simple root , or it is of type @xmath529 or it is of type @xmath534 . if @xmath551 is a simple root or if it is of type @xmath529 then it follows immediately that every @xmath552 is equal to one . suppose instead that @xmath551 is of type @xmath534 ; in order to show the thesis it is enough to consider the case wherein @xmath409 is a wonderful variety whose spherical roots are all supported on a subset @xmath553 of type @xmath204 . an easy computation shows that , if @xmath554 and if @xmath394 is any distinguished subset of colors of @xmath409 , then the quotient @xmath555 never possesses @xmath556 as a spherical root . therefore , if @xmath557 , it must be either @xmath558 or @xmath559 and the claim follows . as exemplified in the following sections ( example [ ex2 ] and example [ ex3 ] ) , proposition [ sistema sferico orbita xlt ] together with corollary [ sistema sferico stabilizzatore ] allow to compute explicitly the set of orbits of @xmath484 and that of @xmath485 in terms of their spherical systems . this is further simplified by the following proposition , which shows that , given an orbit @xmath518 , there exists a minimal orbit @xmath560 mapping on @xmath501 . if @xmath561 , define @xmath562 its support over @xmath129 ; define @xmath563 [ orbita minimale ] let @xmath539 be an orbit and let @xmath560 the orbit whose closure has @xmath564 as set of spherical roots . then @xmath565 maps on @xmath501 and and every other orbit which maps on @xmath501 contains @xmath565 in its closure . let @xmath493 be an orbit mapping on @xmath501 and let @xmath566 be the associated set of spherical roots . since @xmath567 , we get @xmath568 ; this shows @xmath569 , i.e. @xmath570 . in order to prove that @xmath571 it is enough to notice that @xmath572 . [ orbite massimali ] unlike the symmetric case ( see @xcite ) , in the general spherical case there does not need to exist a maximal orbit in @xmath22 mapping on a fixed orbit @xmath518 : for instance this is shown by example [ ex2 ] and by example [ ex3 ] . since @xmath564 depends only on @xmath521 ( or equivalently on @xmath522 ) , we get the following corollaries . two orbits @xmath573 map to the same orbit in @xmath484 if and only if @xmath574 where @xmath575 is the pullback of @xmath365 to @xmath576 and where @xmath577 is the maximal distinguished subset of colors of @xmath578 not intersecting the support of @xmath575 . [ orbite non isomorfe ] two orbits in @xmath484 ( resp . in @xmath485 ) have different sets of spherical roots ; in particular two orbits in @xmath484 ( resp . in @xmath485 ) are never isomorphic . if @xmath217 , then by its combinatorial description it follows that the restriction map to the closed orbit @xmath164 is injective : this means that the generic stabilizer @xmath9 never fixes two different lines in the same simple module . however , if @xmath579 , it could happen that a simple module @xmath17 contains two different orbits both isomorphic to the open orbit @xmath7 : previous corollary shows then that there does not exist any spherical orbit in @xmath17 containing both of them in its closure . for instance , this occurs in the following example . [ ex orbite sferiche doppie ] consider @xmath580 , which is a wonderful variety for @xmath581 , and fix the base point @xmath582,[0,1])$ ] so that the generic stabilizer is the maximal torus @xmath117 of diagonal matrices . consider the simple module @xmath583_5 $ ] formed by the homogeneous polynomials of degree 5 : then @xmath584 $ ] and @xmath585 $ ] are distinct orbits in @xmath17 both isomorphic to the open orbit @xmath586 . keeping the notations of previous section , suppose that @xmath22 is strict . following lemma is a stronger version of lemma [ orbite non isomorfe e radici sferiche cong ] . [ orbite non isomorfe e radici sferiche ] let @xmath22 be a strict wonderful variety and let @xmath365 be a faithful divisor on it ; let @xmath518 be an orbit . then @xmath587 if and only if there exists a spherical root @xmath523 of type @xmath532 and a spherical root @xmath588 of type @xmath589 . by lemma [ orbite non isomorfe e radici sferiche cong ] , we may assume that @xmath297 possesses a spherical root @xmath525 of type @xmath529 or of type @xmath534 . since @xmath218 , it is uniquely determined a spherical root @xmath588 which is of type @xmath590 ( with @xmath591 ) in the first case and of type @xmath534 in the second case . since @xmath22 is strict , the latter can not happen ; thus we are in the first case . suppose that @xmath592 and @xmath593 ; let @xmath594 be the short root in the support of @xmath133 . since @xmath22 is strict , @xmath595 moves a color @xmath596 , while @xmath592 implies @xmath597 for every @xmath598 : therefore @xmath599 is distinguished and by the faithfulness of @xmath365 we get @xmath600 , which implies @xmath601 . but this is a contradiction since @xmath593 implies @xmath602 . if @xmath130 is a spherical root of type @xmath589 , write @xmath603 , where @xmath604 are respectively the long simple root and the short simple root in the support of @xmath133 . since @xmath22 is strict , both @xmath605 and @xmath606 move exactly one color ; set @xmath607 and @xmath608 . * if @xmath609 , then no orbit @xmath539 possesses a spherical root @xmath523 of type @xmath532 with @xmath588 . * if @xmath610 , then there exists an orbit @xmath539 such that @xmath611 ; in particular @xmath612 and the normalization @xmath613 is not bijective . ii ) . consider the rank one orbit @xmath493 whose unique spherical root is @xmath133 . if @xmath616 and @xmath617 , then @xmath618 set @xmath619 and @xmath293 ; set @xmath620 the maximal distinguished subset not intersecting the support of @xmath621 . since @xmath622 and since @xmath623 is the unique color @xmath624 such that @xmath625 , we get @xmath626 which shows @xmath627 . on the other hand @xmath628 , thus @xmath297 is not spherically closed and @xmath629 . * if @xmath22 is a strict wonderful symmetric variety and if @xmath365 is a faithful divisor on it , then the normalization @xmath488 is bijective . * suppose that the dynkin diagram of @xmath0 is simply laced . if @xmath22 is any strict wonderful variety for @xmath0 and if @xmath364 is any faithful divisor , then the normalization @xmath488 is bijective . * if @xmath609 for every @xmath130 of type @xmath589 , then the normalization morphism @xmath613 is bijective . by the classification of symmetric varieties , we deduce that a strict wonderful symmetric variety never possesses a spherical root of type @xmath589 . then all of the claims above follow straightforward by previous lemmas . consider the wonderful model variety @xmath22 of @xmath630 , whose spherical system is expressed by the spherical diagram @xmath631 then the divisor @xmath632 is faithful . consider the codimension one orbit @xmath493 having spherical root @xmath633 ; following proposition [ sistema sferico orbita xlt ] and corollary [ sistema sferico stabilizzatore ] , we get the following sequence of spherical diagrams @xmath634 where the first one represents the orbit @xmath635 , the second one represents the orbit @xmath636 and the third one represents the orbit @xmath637 . suppose that @xmath493 is an orbit with set of spherical roots @xmath172 and set of colors @xmath177 . if @xmath185 , set @xmath642 ; then by the description of the pullback map @xmath643 it follows that @xmath644 where @xmath645 . i ) . let @xmath518 be an orbit possessing @xmath646 as a spherical root ; let @xmath647 and let @xmath493 be an orbit which maps on @xmath501 . then by proposition [ reticoli saturi ] we get that @xmath648 , while by corollary [ sistema sferico stabilizzatore ] together with theorem [ corrispondenza distinta ] we get @xmath649 ; by the remark at the beginning of the proof this implies the thesis . ii ) . consider the rank one orbit @xmath650 whose unique spherical root is @xmath141 , set @xmath619 and @xmath293 . then @xmath648 is a loose spherical root and by the remark at the beginning of the proof we get @xmath651 , where @xmath652 is the pullback of a hyperplane section and where @xmath653 is identified with a subset of @xmath654 . then by corollary [ sistema sferico stabilizzatore ] we get that @xmath640 . suppose that @xmath639 . as shown by example [ ex3 ] , if @xmath655 , then it may not exist any orbit @xmath518 possessing @xmath646 as a spherical root ; conversely , if there exists such an orbit , it may be as well @xmath655 . * if there exists @xmath639 such that @xmath447 is non - zero , then the normalization @xmath656 is not bijective . * if the dynkin diagram of @xmath0 is simply laced and if @xmath440 for every @xmath639 , then the normalization @xmath656 is bijective . reasoning as in lemma [ d1d2 ] and in corollary [ esempi ] , other sufficient conditions of bijectivity can be obtained imposing further conditions on the support of @xmath365 on the multiple links of the dynkin diagram of @xmath0 and on the simple spherical roots of @xmath22 . m. brion , , notes de la session de la s. m. f. `` oprations hamiltoniennes et oprations de groupes algbriques '' , grenoble ( 1997 ) , http://www-fourier.ujf-grenoble.fr/@xmath59mbrion/spheriques.pdf , 59 pp .

interference is a major performance bottleneck in heterogeneous network ( hetnet ) due to its multi - tier topological structure . we propose almost blank resource block ( abrb ) for interference control in hetnet . when an abrb is scheduled in a macro bs , a resource block ( rb ) with blank payload is transmitted and this eliminates the interference from this macro bs to the pico bss . we study a two timescale hierarchical radio resource management ( rrm ) scheme for hetnet with _ dynamic abrb _ control . the long term controls , such as dynamic abrb , are adaptive to the large scale fading at a rrm server for co - tier and cross - tier interference control . the short term control ( user scheduling ) is adaptive to the local channel state information within each bs to exploit the _ multi - user diversity_. the two timescale optimization problem is challenging due to the exponentially large solution space . we exploit the sparsity in the _ interference graph _ of the hetnet topology and derive structural properties for the optimal abrb control . based on that , we propose a _ two timescale alternative optimization _ solution for the user scheduling and abrb control . the solution has low complexity and is asymptotically optimal at high snr . simulations show that the proposed solution has significant gain over various baselines . heterogeneous network , dynamic abrb control , two timescale rrm

1305.5884

as hetnet provides flexible and efficient topology to boost spectral efficiency , it has recently aroused immense interest in both academia and industry . as illustrated in fig . [ fig : hetnet_2tiers ] , a hetnet consists of a diverse set of regular macro base stations ( bs ) overlaid with low power pico bss . since this overlaid structure may lead to severe interference problem , it is extremely critical to control interference via rrm in hetnet . there has been much research conducted on rrm optimization for traditional cellular networks . in @xcite , the authors considered power and user scheduling in single - carrier cellular networks . in @xcite , the game theoretical approaches are proposed for distributed resource allocation . in @xcite , the authors proposed a dynamic fractional frequency reuse scheme to combat the inter - sector interference under a game - based optimization by each sector . the coordinated multipoint transmission ( comp ) @xcite is another important technique to handle the inter - cell interference . for example , in @xcite , the authors exploited the uplink - downlink duality to do joint optimization of power allocation and beamforming vectors . in @xcite , a wmmse algorithm is proposed to find a stationary point of the weighted sum - rate maximization problem for multi - cell downlink systems . while the above algorithms achieve comparably good performance , they require global channel state information ( csi ) for centralized implementation @xcite or over - the - air iterations and global message passing for distributed implementation @xcite . it is quite controversial whether comp is effective or not in lte systems due to large signaling overhead , signaling latency , inaccurate csit , and the complexity of the algorithm . on the other hand , solutions for traditional cellular networks can not be applied directly to hetnet due to the unique difference in hetnet topology . first , the inter - cell interference in hetnet is more complicated , e.g. , there is _ co - tier interference _ among the pico bss and among the macro bss as well as the _ cross - tier interference _ between the macro and pico bss . furthermore , due to load balancing , some of the mobiles in hetnet may be assigned to a pico bs which is not the strongest bs @xcite and the mobiles in the pico cell may suffer from strong interference from the macro bss . to solve these problems , some eicic techniques , such as the abs control @xcite , have been proposed in lte and lte - a @xcite . in @xcite , the authors analyzed the performance for abs in hetnet under different cell range extension ( re ) biases . however , they focused on numerical analysis for the existing heuristic eicic schemes , which are the baselines of this paper . in @xcite , the authors proposed an algorithm for victim pico user partition and optimal synchronous abs rate selection . however , they used a universal abs rate for the whole network , and as a result , their scheme could not adapt to dynamic network loading for different macro cells . in this paper , we focus on the resource optimization in the downlink of a hetnet without comp . we consider _ dynamic abrb _ for interference control and dynamic user scheduling to exploit _ multi - user diversity_. the abrb is similar to the abs but it is scheduled over both time and frequency domain . unlike @xcite , we do not restrict the abrb rate to be the same for all macro bss and thus a better performance can be achieved . however , this also causes several new technical challenges as elaborated below . * * exponential complexity for dynamic abrb : * optimization of abrb patterns is challenging due to the combinatorial nature and exponentially large solution space . for example , in a hetnet with @xmath0 macro bss , there are @xmath1 different abrb pattern combinations . hence , brute force solutions are highly undesirable . * * complex interactions between dynamic user scheduling and dynamic abrb * : there is complex coupling between the dynamic user scheduling and abrb control . for instance , the abrb pattern will affect the user sets eligible for user scheduling . furthermore , the optimization objective of abrb control depends on user scheduling policy and there is no closed form characterization . * * challenges in rrm architecture : * most existing solutions for resource optimization of hetnet requires global knowledge of csi and centralized implementations . yet , such designs are not scalable for large networks and they are not robust with respect to ( w.r.t . ) latency in backhaul . to address the above challenges , we propose a two timescale control structure where the long term controls , such as dynamic abrb , are adaptive to the large scale fading . on the other hand , the short term control , such as the user scheduling , is adaptive to the local csi within a pico / macro bs . such a multi - timescale structure allows _ hierarchical rrm _ design , where the long term control decisions can be implemented on a rrm server for inter - cell interference coordination . the short - term control decisions can be done locally at each bs with only local csi . such design has the advantages of low signaling overhead , good scalability , and robustness w.r.t . latency of backhaul signaling . while there are previous works on two timescale rrm @xcite , those approaches are heuristic ( i.e. the rrm algorithms are not coming from a single optimization problem ) . our contribution in this paper is a formal study of two timescale rrm algorithms for hetnet based on optimization theory . to overcome the exponential complexity for abrb control , we exploit the sparsity in the _ interference graph _ of the hetnet topology and derive structural properties for the optimal abrb control . based on that , we propose a _ two timescale alternative optimization _ solution for user scheduling and abrb control . the algorithm has low complexity and is asymptotically optimal at high snr . simulations show that the proposed solution has significant performance gain over various baselines . _ notations _ : let @xmath2 denote the indication function such that @xmath3 if the event @xmath4 is true and @xmath5 otherwise . for a set @xmath6 , @xmath7 denotes the cardinality of @xmath6 . consider the downlink of a two - tier hetnet as illustrated in fig . [ fig : hetnet_2tiers ] . there are @xmath0 macro bss , @xmath8 pico bss , and @xmath9 users , sharing @xmath10 ofdm subbands . denote the set of the macro bss as @xmath11 , and denote the set of the pico bss as @xmath12 . a two - tier heterogeneous network with macro and pico base stations , width=332 ] the hetnet topology ( i.e. , the network connectivity and csi of each link ) is represented by a topology graph as defined below . [ hetnet topology graph]define the _ topology graph _ of the hetnet as a bipartite graph @xmath13 , where @xmath14 denotes the set of all macro and pico bs nodes , @xmath15 denotes the set of all user nodes , and @xmath16 is the set of all edges between the bss and users . an edge @xmath17 between bs node @xmath18 and user node @xmath19 represents a wireless link between them . each edge @xmath17 is associated with a csi label @xmath20 , where @xmath21 represents the channel coefficient between bs @xmath22 and user @xmath23 on subband @xmath24 . for each bs node @xmath22 , let @xmath25 denote the set of associated users . for each user node @xmath23 , define @xmath26 as the set of neighbor macro bss and @xmath27 as the set of neighbor pico bss . in the topology graph , @xmath28 means that the path gain between user @xmath23 and bs @xmath22 is sufficiently small compared to the direct link path gain , and thus the interference from bs @xmath22 will have negligible effect on the data rate of user @xmath23 . we have the following assumption on the channel fading process @xmath29 . [ asm : fory]the channel fading coefficient has a two timescale structure given by @xmath30 . the small scale fading process @xmath31 is identically distributed w.r.t . the subframe and subband indices ( @xmath32 ) , and it is i.i.d . user and bs indices ( @xmath33 ) . moreover , for given @xmath34 , @xmath31 is a continuous random variable . the large scale fading process @xmath35 is assumed to be a slow ergodic process ( i.e. , @xmath36 remains constant for several _ _ super - frames _ _ subframes ] . ) according to a general distribution . the two timescale fading model has been adopted in many standard channel models . the large scale fading @xmath36 is usually caused by path loss and shadow fading , which changes much slowly compared to the small scale fading . we consider the following _ biased cell selection _ mechanism to balance the loading between macro and pico bss @xcite . let @xmath37 denote the serving bs of user @xmath23 . let @xmath38 denote the cell selection bias and let @xmath39 denote the transmit power of bs @xmath22 on a single sub - band . let @xmath40 and @xmath41 respectively be the strongest macro bs and pico bs for user @xmath23 . if @xmath42 , user @xmath23 will be associated to pico cell @xmath43 , i.e. , @xmath44 ; otherwise @xmath45 . if a user only has a single edge with its serving bs , it will not receive inter - cell interference from other bss and thus its performance is noise limited ; otherwise , it will suffer from strong inter - cell interference if any of its neighbor bss is transmitting data and thus its performance is interference limited . this insight is useful in the control algorithm design later and it is convenient to formally define the interference and noise limited users . [ interference / noise limited user][def : inuser]if a user @xmath23 has a single edge with its serving bs @xmath37 only , i.e. , @xmath46 , then it is called a _ noise limited user _ ( n - user ) ; otherwise , it is called an _ interference limited user _ ( i - user ) . fig . [ fig : abs_intro ] illustrates an example of the hetnet topology graph . in fig . [ fig : abs_intro](a ) , an arrow from a bs to a user indicates a direct link and the dash circle indicates the coverage area of each bs . an i - user which lies in the coverage area of a macro bs is connected to this macro bs , while a n - user does not have connections with the neighbor macro bss in the topology graph as illustrated in fig . [ fig : abs_intro](b ) . we consider a two timescale hierarchical rrm control structure where the control variables are partitioned into _ long - term _ and _ short - term _ control variables . the long - term control variables are adaptive to the large scale fading @xmath47 and they are implemented at the radio resource management server ( rrms ) . the short - term control variables are adaptive to the instantaneous csi @xmath48 and they are implemented locally at each macro / pico bs . abs is introduced in lte systems @xcite for interference mitigation among control channels in hetnet . it can also be used to control the _ co - tier _ and _ cross - tier interference _ among the data channels . in lte systems , abs is only scheduled over time domain . in this paper , we consider dynamic abrb control for interference coordination . the abrb is similar to abs but it is scheduled over both time and frequency domain . it is a generalization of abs and enables more fine - grained resource allocation . when an abrb is scheduled in a macro bs , a rb with blank payload will be transmitted at a given frequency and time slice and this eliminates the interference from this macro bs to the pico bss and the adjacent macro bss . hence , as illustrated in fig . [ fig : abs_intro ] , scheduling abrb over both time and frequency domain allows us to control both the _ macro - macro _ bs and _ macro - pico _ bs interference . we want to control the abrb dynamically w.r.t . the large scale fading because the optimal abrb pattern depends on the hetnet topology graph . for example , when there are a lot of pico cell i - users , we should allocate more abrbs at the macro bs to support more pico cell i - users . on the other hand , when there are only a few pico cell i - users , we should allocate less abrbs to improve the spatial spectrum efficiency . for any given subframe , define @xmath49 to indicate if abrb is scheduled ( @xmath50 ) for subband @xmath24 at macro bs @xmath22 . let @xmath51^{t}\in\mathcal{a}$ ] be the abrb pattern vector for subband @xmath24 and @xmath52 is the set of all possible abrb patterns macro bss can either schedule an abrb or not for the @xmath53-th rb ( i.e. , subframe @xmath54 and subband @xmath24 ) , there are @xmath1 possible abrb patterns . hence , the size of @xmath52 is @xmath1 . ] . in the proposed dynamic abrb control , each macro bs is allowed to dynamically change the ratio of abrb transmission on each subband and this ratio can be any positive real number . to facilitate implementation , we consider randomized abrb control policy as defined below . [ randomized abrb control policy ] an abrb control policy of the @xmath24-th subband @xmath55 is a mapping from the abrb pattern space @xmath52 to a probability in [ 0,1 ] . at any subframe , the instantaneous abrb pattern vector for subband @xmath24 is stochastically determined according to the probabilities @xmath56 , where @xmath57 denote the probability that the @xmath58 subband is in abrb pattern @xmath59 . to facilitate structural abrb design , we partition the users into two types . [ partitioning of user set][def : user - set - partition ] the mobile user set is partitioned into two subsets @xmath60 , where @xmath61 denotes the set of _ type a users _ and is defined as @xmath62 and @xmath63 denotes the set of _ type b users_. the type a users include all pico cell users and macro cell n - users , while the type b users include all macro cell i - users . for type a users , it will not lose optimality by imposing a _ synchronous abrb structure _ where the transmissions of the abrb at all macro bss are aligned as much as possible . the formal definition of the synchronous abrb structure is given in theorem [ thm : absreduce ] . as will be shown in theorem [ thm : absreduce ] , if there is only type a users , imposing the synchronous abrb structure can dramatically reduce the number of abrb control variables from exponential large ( @xmath1 ) to only @xmath0 and this complex reduction is achieved without loss of optimality . on contrast , the performance of the macro cell i - users is very poor under the synchronous abrb structure because aligning the data transmissions of all macro bss will cause strong inter - cell interference for macro cell i - users . motivated by these observations , we partition the @xmath10 subbands into two groups , namely @xmath64 and @xmath65 , and use different abrb control policies for type a and type b users on these two groups of subbands respectively . the variable @xmath66 controls the fraction of _ type a _ subbands . at each subframe , each bs @xmath22 dynamically selects a user from @xmath25 for each subband @xmath24 based on the knowledge of current abrb pattern @xmath59 and channel realization @xmath67 to exploit _ multi - user diversity_. let @xmath68 be the user scheduling variable ( of user @xmath23 at bs @xmath37 ) of subband @xmath24 and @xmath69 $ ] be the associated vectorized variable . the set of all feasible user scheduling vectors at bs @xmath22 for the @xmath24-th subbands with abrb pattern @xmath59 is given by @xmath70 where @xmath71 is the set of users that can not be scheduled on a type a subband under abrb pattern @xmath59 ; and @xmath72 . the physical meaning of @xmath73 is elaborated below . first , if a macro bs is transmitting abrb , none of its associated users can be scheduled for transmission . moreover , due to large cross - tier interference from macro bss , a pico cell i - user can not be scheduled for transmission if any of its neighbor macro bss @xmath74 is transmitting data subframe ( i.e. , @xmath75 ) . as will be seen in section [ sub : structural - properties - ofpa ] , explicitly imposing this user scheduling constraint for the pico cell i - users is useful for the structural abrb design . the user scheduling policy @xmath76 of the @xmath24-th sub - band is defined below . [ user scheduling policy][def : randomized - user - scheduling ] a user scheduling policy of the @xmath22-th bs and @xmath24-th sub - band @xmath77 is a mapping : @xmath78 , where @xmath79 is the csi space . specifically , under the abrb pattern @xmath59 and csi realization @xmath67 , the user scheduling vector of bs @xmath22 is given by @xmath80 . let @xmath81 denote the overall user scheduling policy on sub - band @xmath24 . assuming perfect csi at the receiver ( csir ) and treating interference as noise , the instantaneous data rate of user @xmath23 is given by : @xmath82 where @xmath83 , @xmath84 ; @xmath85 is the mutual information of user @xmath23 contributed by the @xmath24-th subband ; and @xmath86 is the interference - plus - noise power at user @xmath23 on subband @xmath24 . for a given policy @xmath87 and large scale fading state @xmath88 , the average data rate of user @xmath23 is given by : @xmath89=\sum_{m\in\mathcal{m}\left(k\right)}\overline{\mathcal{i}}_{m , k}\left(q_{m},\pi_{m}\right),\ ] ] where the average mutual information on subband @xmath24 is @xmath90 and @xmath91 $ ] . for conciseness , the abrb pattern @xmath59 for a specific subband @xmath24 is denoted as @xmath92 $ ] when there is no ambiguity . the performance of the hetnet is characterized by a utility function @xmath93 , where @xmath94 $ ] is the average rate vector . we make the following assumptions on @xmath93 . [ asm : ufun]the utility function can be expressed as @xmath95 , where @xmath96 is the weight for user @xmath23 , @xmath97 is assumed to be a concave and increasing function . moreover , for any @xmath98 such that @xmath99 and @xmath100 belongs to the domain of @xmath97 , @xmath101 satisfies @xmath102 where @xmath103 and @xmath104 are some scalar functions of @xmath105 . the above assumption is imposed to facilitate the problem decomposition in section [ sub : problem - decomposition ] . this utility function captures a lot of interesting cases below . * * weighted sum throughput : * the utility function is @xmath106 . * * @xmath107-fair @xcite : * @xmath107-fair can be used to compromise between the fairness to users and the utilization of resources . the utility function is * * @xmath108 * * proportional fair @xcite : * this is a special case of @xmath107-fair when @xmath109 . due to the statistical symmetry of the @xmath10 subbands , there is no loss of optimality to consider symmetric policy @xmath110 , where @xmath111 ( @xmath112 ) and @xmath113 ( @xmath114 ) if @xmath115 ( @xmath65 ) . [ optimality of symmetric policy][lem : optimality - of - symmetric]there exists a symmetric policy @xmath116 such that it is the optimal solution of the following optimization problem : @xmath117 please refer to appendix [ proof - of - lemma symp ] for the proof . moreover , we have @xmath118 and @xmath119 . as a result , the utility function under a symmetric policy @xmath120 can be expressed as : @xmath121 where @xmath122 , @xmath123 , and @xmath124 ( @xmath125 ) can be any type a ( type b ) subband . finally , for a given hetnet topology graph @xmath13 , the two timescale rrm optimization is given by : @xmath126 where ( [ eq : qeccon1a ] ) and ( [ eq : qeccon2a ] ) ensure that @xmath127 and @xmath128 satisfy the definition of probability mass function ( pmf ) . using primal decomposition , problem @xmath129 can be decomposed into the following subproblems . * subproblem a * ( * cross - tier interference control * ) : optimization of abrb @xmath130 and user scheduling @xmath131 . @xmath132 * subproblem b * ( * co - tier interference control * ) : * * optimization of abrb @xmath112 and user scheduling @xmath114 . @xmath133 * subproblem c * ( * subband partitioning * ) : optimization of subband partitioning @xmath134 . @xmath135 note that the solution of @xmath136/@xmath137 is independent of the value of @xmath134 because both @xmath138 and @xmath139 are independent of @xmath134 . after solving @xmath136 and @xmath137 , the optimal @xmath134 can be easily solved by bisection search . on the other hand , the optimization of @xmath131/@xmath114 is a stochastic optimization problem because the @xmath140/@xmath141 involves stochastic expectation over csi realizations and they do not have closed form characterization . furthermore , the number of abrb control variables in @xmath136/@xmath137 is exponential w.r.t . the number of macro bss @xmath0 . we shall tackle these challenges in section [ sec : solution - of - pa ] and [ sub : solution - of - subproblempb ] . in this section , we first derive structural properties of @xmath136 and reformulate @xmath136 into a simpler form with reduced solution space . then , we develop an efficient algorithm for @xmath136 . we require the following assumption to derive the results in this section . [ asm : picoiedge]for any @xmath142 , let @xmath143 denote the set of all pico cell i - users in pico cell @xmath22 . then we have @xmath144 and define @xmath145 as the set of neighbor macro bss of pico cell @xmath22 . the above assumption states that a macro bs will interfere with all the i - users in a pico cell as long as it interferes with any user in the pico cell . this is reasonable since the coverage area of a macro bs is much larger than that of a pico bs . we exploit the interference structure in the hetnet topology @xmath146 to derive the structural properties of @xmath136 . throughout this section , we will use the following example problem to illustrate the intuition behind the main results . [ exm - problema]consider @xmath147 for the hetnet in fig . [ fig : abs_intro ] with @xmath148 macro bss and @xmath149 pico bs . the set of type a users is @xmath150 and the objective function is specified as @xmath151 ( i.e. , we consider sum - rate utility ) . for illustration , we focus on the case when the marginal probability that a macro bs is transmitting abrb is transmitting abrb is @xmath152 . ] is fixed as @xmath153^{t}=[0.7,0.5]^{t}$ ] . define two sets of abrb patterns @xmath154 and @xmath155 for each bs @xmath22 . for macro bs , @xmath154 is the set of abrb patterns under which macro bs @xmath22 is transmitting data . for pico bs , @xmath154 is the set of abrb patterns under which all of its neighbor macro bss is transmitting abrb . using the configuration in example [ exm - problema ] , we have @xmath156,\left[1,1\right]\right\ } $ ] , @xmath157,\left[1,1\right]\right\ } $ ] and @xmath158,\left[1,0\right]\right\ } $ ] ( the formal definition of @xmath154 and @xmath159 for general cases is in appendix [ sub : spageneral ] ) . in observation [ clm : effect - of - abrb - mi ] , we find that the abrb pattern @xmath160 only affects the data rate of a type a user in cell @xmath22 by whether @xmath161 or not . based on that , we find that the policy space for both the abrb control @xmath130 and user scheduling @xmath131 can be significantly reduced in observation [ clm : qareduction ] and [ clm : policy - space - reductionpia ] . while these observations are made for the specific configuration in example [ exm - problema ] , they are also correct for general configurations and are formally stated in lemma [ lem : sysman ] , theorem [ thm : absreduce ] and theorem [ thm : policy - space - reductionpia ] in appendix [ sub : spageneral ] . finally , using the above results , we transform the complicated problem @xmath136 to a much simpler problem with @xmath162 as the optimization variables . [ clm : effect - of - abrb - mi]for given csi @xmath163 and a feasible user scheduling vector @xmath164 , the mutual information @xmath165 of a type a user @xmath23 in cell @xmath22 only depends on whether the abrb pattern @xmath161 or not , i.e. , @xmath166,@xmath167 ( or @xmath168 ) . moreover , we have @xmath169 for any @xmath170 . let us illustrate the above observation using the configuration in example [ exm - problema ] . there are 4 type a users : @xmath150 . for user 1 in macro cell 1 ( n - user ) , it is scheduled for transmission whenever macro bs 1 is not transmitting abrb ( i.e. , @xmath171 ) . hence the mutual information is @xmath172 , which only depends on whether @xmath171 ( i.e. , @xmath173 ) or not . moreover , the mutual information is higher if @xmath173 because @xmath174 . for user 4 in pico cell 3 ( i - user ) , if @xmath175 , its neighbor macro bs 2 is transmitting abrb and the mutual information is @xmath176 ; otherwise ( @xmath177 ) , macro bs 2 is transmitting data and we have @xmath178 and @xmath179 . similar observations can be made for user 2 and 3 . based on observation [ clm : effect - of - abrb - mi ] , the abrb control policy space can be significantly reduced . [ clm : qareduction]consider @xmath147 for the configuration in example [ exm - problema ] . the optimal abrb control of @xmath147 conditioned on a given marginal probability vector @xmath180^{t}$ ] , denoted by @xmath181 , has the _ _ synchronous abrb structure__. ] where the transmissions of abrb at the macro bss are aligned as much as possible . as a result , there are only @xmath182 active abrb patterns @xmath183,[0,1],[1,1]\right\ } $ ] and the corresponding pmf @xmath184 is given by a function of @xmath185 as illustrated in fig . [ fig : abs_reduction ] . illustration of the structure of the optimal @xmath186 conditioned on a given @xmath185,width=332 ] in general , for a hetnet with @xmath0 macro bss , there are only @xmath187 active abrb patterns under synchronous abrb structure , which is significantly smaller than the number of all possible abrb patterns @xmath1 . as a result , the optimization @xmath147 w.r.t @xmath130 ( with @xmath1 variables ) can be reduced to an equivalent optimization w.r.t . @xmath185 ( with @xmath0 variables ) with much lower dimensions . observation [ clm : qareduction ] can be understood as follows . by observation [ clm : effect - of - abrb - mi ] , a higher average mutual information can be achieved for user @xmath188 under the abrb patterns @xmath161 . hence , for given marginal probabilities @xmath180^{t}$ ] , the average mutual information region will be maximized if we can simultaneously maximize @xmath189 for all bss @xmath190 . for macro bss @xmath191 , we have @xmath192 , which is fixed for given @xmath185 . for pico bss @xmath193 , we have @xmath194 , and the equality holds if and only if @xmath130 has the synchronous abrb structure in fig . [ fig : abs_reduction ] . similarly , we can reduce the user scheduling policy space using observation [ clm : effect - of - abrb - mi ] . [ clm : policy - space - reductionpia]consider @xmath147 for the configuration in example [ exm - problema ] . for given csi @xmath163 and abrb pattern @xmath160 , the optimal user scheduling at bs @xmath22 is given by @xmath195 by observation [ clm : effect - of - abrb - mi ] , if @xmath196 solves the maximization problem ( [ eq : argmaxruo ] ) for certain @xmath161 , it solves ( [ eq : argmaxruo ] ) for all @xmath161 . hence , it will not loss optimality by imposing an additional constraint on the user scheduling such that @xmath197 ( or @xmath168 ) . for convenience , let @xmath198 denote the set of all feasible user scheduling policies satisfying the above constraint in observation [ clm : policy - space - reductionpia ] ( the formal definition of @xmath198 is given in theorem [ thm : policy - space - reductionpia ] ) . then for given @xmath185 , @xmath199 and under the synchronous abrb , the corresponding objective function of @xmath136 can be rewritten as @xmath200 , where @xmath201 and @xmath202 is the corresponding average mutual information given in ( [ eq : ibarak ] ) of appendix [ sub : spageneral ] . as a result , the subproblem @xmath147 can be transformed into a simpler problem with @xmath203 solution space . [ equivalent problem transformation of @xmath147][cor : equivalent - problem - pa]let @xmath205 denote the optimal solution of the following joint optimization problem . @xmath206 where @xmath207,\forall j\right\ } $ ] . then @xmath208 , @xmath209 is the optimal solution of problem @xmath147 . furthermore , problem ( [ eq : plare ] ) is a bi - convex problem , i.e. , for fixed @xmath131 , problem ( [ eq : plare ] ) is convex w.r.t . @xmath185 , and for fixed @xmath185 , problem ( [ eq : plare ] ) is also convex w.r.t . @xmath131 . please refer to appendix [ sub : proof - of - corollaypa ] for the proof . by corollary [ cor : equivalent - problem - pa ] , we only need to solve the equivalent problem of @xmath136 in ( [ eq : plare ] ) . since problem ( [ eq : plare ] ) is bi - convex , we propose the following _ two timescale alternating optimization ( ao ) _ algorithm . for notation convenience , time index @xmath54 and @xmath210 are used to denote the subframe index and super - frame index respectively , where a super - frame consists of @xmath211 subframes . _ algorithm ao_a _ ( two timescale ao for @xmath147 ) : * * initialization * * : choose proper initial @xmath212,@xmath213 . set @xmath214 . * * step 1 * * ( short timescale user scheduling optimization ) : for fixed @xmath215 , let @xmath216 , where @xmath217 is given by @xmath218 where @xmath219 is the average data rate of user @xmath23 under @xmath215 and user scheduling policy @xmath220 . for each subframe @xmath221 $ ] , the user scheduling vector of bs @xmath22 is given by @xmath222 , where @xmath163 and @xmath223 are the csi and abrb pattern at the @xmath54-th subframe . * step 2 * ( long timescale abrb optimization ) : * * find the optimal solution @xmath224 of problem ( [ eq : plare ] ) under fixed @xmath220 using e.g. , ellipsoid method . let @xmath225 . * * return to step 1 until @xmath226 * * or the maximum number of iterations is reached . while ( [ eq : plare ] ) is a bi - convex problem and ao algorithm is known to converge to local optimal solutions only , we exploit the hidden convexity of the problem and show below that algorithm ao_a can converge to the global optimal solution under certain conditions . [ global convergence of algorithm ao_a][thm : global - convergence - la]let @xmath227=\mathcal{f}\left(\left[\mathbf{q}_{a}^{(t)},\pi_{a}^{(t-1)}\right]\right)$ ] denote the iterate sequence of algorithm ao_a began at @xmath228 , and denote the set of fixed points of the mapping @xmath229 as @xmath230\right.$ ] @xmath231\right)\right\ } $ ] . for any @xmath232 that is not a fixed point of @xmath229 , assume that @xmath233 , where @xmath234=\mathcal{f}\left(\left[\mathbf{q}_{a},\pi_{a}\right]\right)$ ] . then : 1 . algorithm ao_a converges to a fixed point @xmath235\in\delta$ ] of @xmath229 . any fixed point @xmath235\in\delta$ ] is a globally optimal solution of problem ( [ eq : plare ] ) . please refer to appendix [ sub : proof - of - theoremla ] for the proof . step 1 of algorithm ao_a requires the knowledge of the average data rate @xmath236 under @xmath215 and @xmath220 . we adopt a reasonable approximation on @xmath236 using a moving average data rate @xmath237 $ ] given by @xcite @xmath238 where @xmath239 is the data rate delivered to user @xmath23 at subframe @xmath54 . if we replace @xmath236 in step 1 of algorithm ao_a with the approximation @xmath240 in ( [ eq : rbarappr ] ) , the global convergence result in theorem [ thm : global - convergence - la ] no longer holds . however , it has been shown in @xcite that @xmath240 converges to @xmath236 as @xmath241 . hence , with the approximation @xmath240 , algorithm ao_a is still asymptotically optimal for large super frame length @xmath211 . in step 2 of algorithm ao_a , the average mutual information @xmath242 in the optimization objective contains two intermediate problem parameters @xmath243 and @xmath244 defined under ( [ eq : ibarak ] ) in appendix [ sub : spageneral ] . the calculation of @xmath243 s and @xmath244 s requires the knowledge of the distribution of all the channel coefficients , which is usually difficult to obtain offline . however , these terms can be easily estimated online using the time average of the sampled data rates delivered to user @xmath23 under abrb patterns @xmath245 and @xmath246 respectively . the abrb control @xmath224 is then obtained by solving the long timescale problem in step 2 with say , the ellipsoid method based on these estimates . the number of abrb control variables @xmath247 is exponentially large w.r.t . @xmath0 . to simplify @xmath137 , we introduce an auxiliary variable called the _ abrb profile _ and decompose @xmath137 . we first define the abrb profile . [ abrb profile ] the abrb profile @xmath248 is a subset of abrb patterns for type b subbands . using the notion of abrb profile and primal decomposition , @xmath249 can be approximated by two subproblems : * optimization of @xmath112 and @xmath114 for a given abrb profile @xmath250 . * @xmath251 * optimization of abrb profile @xmath250 . * @xmath252 in @xmath253 , we restrict the size of @xmath250 to be no more than @xmath254 . in appendix [ sub : proof - of - theoremequpb ] , we prove that at the asymptotically optimal solution of @xmath249 as snr becomes high , the number of active abrb patterns is indeed less than or equal to @xmath254 . let @xmath256 denote the @xmath257 abrb pattern in @xmath250 . then the average mutual information @xmath141 can be rewritten as @xmath258 where @xmath259 $ ] with @xmath260 denoting the probability that the abrb pattern @xmath256 is used . hence @xmath261 can be reformulated as @xmath262 similar to ( [ eq : plare ] ) , we propose a two timescale ao to solve for problem ( [ eq : plbequ ] ) w.r.t . @xmath114 and @xmath263 . _ algorithm ao_b _ ( two timescale ao for @xmath261 ) : * * initialization * * : choose proper initial @xmath264,@xmath265 . set @xmath214 . * * step 1 * * ( short timescale user scheduling optimization ) : for fixed @xmath266 , let @xmath267 , where @xmath268 is given by @xmath269 where @xmath270 is the average data rate of user @xmath23 under @xmath266 and user scheduling policy @xmath271 . for each subframe @xmath221 $ ] , the user scheduling vector of bs @xmath22 is given by @xmath272 , where @xmath273 and @xmath274 are the csi and abrb pattern at the @xmath54-th subframe . * step 2 * ( long timescale abrb optimization ) : * * find the optimal solution @xmath275 of problem ( [ eq : plbequ ] ) under fixed @xmath276 s using e.g. , interior point method . let @xmath225 . * * return to step 1 until @xmath277 * * or the maximum number of iterations is reached . using similar proof as that in appendix [ sub : proof - of - theoremla ] , it can be shown that algorithm ao_b converges to the global optimal solution for problem ( [ eq : plbequ ] ) under similar assumption as in theorem [ thm : global - convergence - la ] . the average data rate @xmath236 and the mutual information @xmath276 can be estimated online in a similar way as in algorithm ao_a . problem @xmath253 is a difficult combinatorial problem and the complexity of finding the optimal abrb profile is extremely high . in this section , we illustrate the top level method for finding a good abrb profile . the detailed algorithm to solve @xmath253 is given in appendix [ sub : proof - of - theoremequpb ] . a good abrb profile can be found based on an _ interference graph _ extracted from the topology graph . [ interference graph][def : igraph]for a hetnet topology graph @xmath13 , define an undirected interference graph @xmath278 , where @xmath279 is the vertex set and @xmath280 is edge set with @xmath281 denoting the edge between @xmath282 and @xmath283 . for any @xmath284 , if @xmath285 , @xmath286 , otherwise , @xmath287 , where @xmath288 is the void set . [ fig : ifc_graph1 ] illustrates how to extract the interference graph from the topology graph using an example hetnet . given an interference graph @xmath289 for the hetnet , any two links @xmath290 having an edge ( i.e. , @xmath286 ) should not be scheduled for transmission simultaneously . on the other hand , we should `` turn on '' as many `` non - conflicting '' links as possible to maximize the spatial reuse efficiency . this intuition suggests that the optimal abrb profile is highly related to the _ maximal independent set _ of the interference graph @xmath289 . an example of extracting the interference graph from the hetnet topology graph , width=332 ] [ maximal independent set ( mis)]a subset @xmath291 of @xmath292 is an _ independent set _ of @xmath278 if @xmath293 . maximal independent set _ ( mis ) is an independent set that is not a proper subset of any other independent set . for any mis @xmath291 , define @xmath294 as the _ maximal independent macro bs set _ corresponding to @xmath291 . let @xmath295 denote the set of all miss of @xmath289 . for example , the set of all miss of the interference graph in fig . [ fig : ifc_graph1 ] is @xmath296 . define a set @xmath297 then the top level flow of finding a good abrb profile is summarized in table . [ tab : flowpb ] . in step 2 , we need to find a set of miss @xmath298 . for the example in fig . [ fig : ifc_graph1 ] , @xmath299 has a unique element given by @xmath300 and thus @xmath301 . in step 3 , the mapping from @xmath302 to the abrb profile @xmath303 is @xmath304 , where @xmath305 $ ] with @xmath306 and @xmath307 . for example , for the hetnet in fig . [ fig : ifc_graph1 ] , we have @xmath301 and thus @xmath308 , where @xmath309 $ ] , @xmath310 $ ] and @xmath311 $ ] . since @xmath312 in this case , @xmath303 can be reduced to @xmath313,\left[0,1,0\right]\right\ } $ ] . .[tab : flowpb]top level flow of finding a good abrb profile [ cols= " < " , ] fig . [ fig : conva ] shows the utility @xmath314 in ( [ eq : plare ] ) of the type a users versus the number of super - frames . the utility increases rapidly and then approaches to a steady state after only @xmath315 updates . the figure demonstrates a fast convergence behavior of algorithm ao_a . similar convergence behavior was also observed for algorithm ao_b and the simulation result is not shown here due to limited space . utility of the type a users versus the number of super - frames , width=321 ] we compare the complexity of the baselines and proposed rrm algorithms . the complexity can be evaluated in the following 3 aspects . \1 ) for the short term user scheduling , the proposed scheme and baseline 1 - 3 have the same complexity order of @xmath316 , while the baseline 4 has a complexity of @xmath317 @xcite , where @xmath318 is a proportionality constant that corresponds to some matrix and vector operations with dimension @xmath319 , and @xmath320 is the number of bss in each cooperative cluster . \2 ) for the long term abrb control variables @xmath185 and @xmath263 , as they are updated by solving standard convex optimization problems in step 2 of algorithm ao_a and ao_b respectively , the complexities are polynomial w.r.t . the number of the associated optimization variables . specifically , for control variable @xmath185 , the complexity is polynomial w.r.t . the number of macro bss @xmath0 . for control variable @xmath321 , the complexity is polynomial w.r.t . the size of the abrb profile @xmath303 : @xmath322 . in addition , they are only updated once in each super - frame . \3 ) the abrb profile @xmath303 is computed using algorithm b2 in every several super - frames to adapt to the large scale fading . in step 1 of algorithm b2 , the complexity of solving the convex problem ( [ eq : fixthetapdet ] ) is polynomial w.r.t . @xmath323 . in step 2 of algorithm b2 , if the mwis algorithm in @xcite is used to solve problem ( [ eq : mwisp ] ) , the complexity is @xmath324 , where @xmath325 is the number of edges in the interference graph @xmath278 . we propose a two - timescale hierarchical rrm for hetnet with dynamic abrb . to facilitate structural abrb design for cross - tier and co - tier interference , the @xmath10 subbands are partitioned into type a and type b subbands . consequently , the two timescale rrm problem is decomposed into subproblems @xmath136 and @xmath137 which respectively optimizes the abrb control and user scheduling for the type a and type b subbands . both subproblems involve non - trivial multi - stage optimization with exponential large solution space w.r.t . the number of macro bss @xmath0 . we exploit the sparsity in the hetnet interference graph and derive the structural properties to reduce the solution space . based on that , we propose two timescale ao algorithm to solve @xmath136 and @xmath137 . the overall solution is asymptotically optimal at high snr and has low complexity , low signaling overhead as well as robust w.r.t . latency of backhaul signaling . define the average rate region as @xmath326\in\mathbb{r}_{+}^{k}:\ : x_{k}\le\overline{r}_{k}\left(\lambda\right),\forall k\right\ } .\ ] ] for any utility function that is concave and increasing w.r.t . to the average data rates @xmath327 , the optimal policy of @xmath328 must achieve a pareto boundary point of @xmath329 . hence , we only need to show that any pareto boundary point of @xmath329 can be achieved by a symmetric policy . define the average rate region under fixed @xmath134 as @xmath330\in\mathbb{r}_{+}^{k}:\\ & & x_{k}\le\overline{r}_{k}\left(\left\ { q_{s},\left\ { q_{m}\right\ } , \left\ { \pi_{m}\right\ } \right\ } \right),\forall k\bigg\}.\end{aligned}\ ] ] then we only need to show that any pareto boundary point of @xmath331 can be achieved by a symmetric policy @xmath110 . define the average mutual information region _ _ for subband @xmath115 as : @xmath332_{\forall k\in\mathcal{u}_{a}}\in\mathbb{r}_{+}^{\left|\mathcal{u}_{a}\right|}:\nonumber \\ & & x_{k}\le\overline{\mathcal{i}}_{m , k}\left(q_{m},\pi_{m}\right),\forall k\in\mathcal{u}_{a}\bigg\}.\label{eq : ratereg1g-1}\end{aligned}\ ] ] define the average mutual information region _ _ for subband @xmath333 as : @xmath332_{\forall k\in\mathcal{u}_{b}}\in\mathbb{r}_{+}^{\left|\mathcal{u}_{b}\right|}:\nonumber \\ & & x_{k}\le\overline{\mathcal{i}}_{m , k}\left(q_{m},\pi_{m}\right),\forall k\in\mathcal{u}_{b}\bigg\}.\label{eq : ratereg2g-1}\end{aligned}\ ] ] it can be verified that @xmath334 is a convex region in @xmath335 and @xmath336 is a convex region in @xmath337 . moreover , due to the statistical symmetry of the subbands , we have @xmath338 let @xmath339 and @xmath340 . from the convexity of @xmath341 and ( [ eq : ra]-[eq : rb ] ) , we have @xmath342 hence , for any pareto boundary point @xmath343^{t}$ ] of @xmath331 , @xmath344_{k\in\mathcal{u}_{a}}\in\mathbb{r}_{+}^{\left|\mathcal{u}_{a}\right|}$ ] is a pareto boundary point of @xmath345 and @xmath346_{k\in\mathcal{u}_{b}}\in\mathbb{r}_{+}^{\left|\mathcal{u}_{b}\right|}$ ] is a pareto boundary point of @xmath347 . due to the statistical symmetry of the subbands , there exists an abrb control policy and a user scheduling policy @xmath348 such that @xmath349 can be achieved for all subbands @xmath115 . similarly , there exists an abrb control policy and a user scheduling policy @xmath350 such that @xmath351 can be achieved for all subbands @xmath333 . hence , @xmath352 can be achieved using the symmetric policy @xmath353 . this completes the proof . [ [ structural - properties - of - mathcalp_a - for - general - casessubspageneral ] ] structural properties of @xmath136 for general cases[sub : spageneral ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 1 . @xmath167 , we have @xmath358 and @xmath359 @xmath360 . the same is true if we replace @xmath154 with @xmath159 . 2 . for given abrb patterns @xmath170 and user scheduling vector @xmath361 , there exists @xmath362 such that @xmath363 . by definition [ def : inuser ] , there is no inter - cell interference for macro cell n - users . by the definition of @xmath364 , a pico cell i - user @xmath357 can not be scheduled for transmission if any of the neighbor macro bss in @xmath365 is transmitting data subframe ( i.e. , the current abrb pattern @xmath366 ) . on the other hand , if all of the neighbor macro bss is transmitting abrb ( i.e. , the current abrb pattern @xmath161 ) , the interference from the macro bss is negligible . finally , by definition [ def : inuser ] , there is no inter - cell interference for pico cell n - users . then lemma 1 follows straightforwardly from the above analysis and the definition of @xmath367 . [ rem : usefulassumptions]note that in the proof of lemma [ lem : sysman ] , we have used the fact that the inter - cell interference seen at a n - user is negligible . we have also used assumption [ asm : picoiedge ] , which states that the sets of neighbor macro bss of the pico cell i - users belonging to the same pico cell are identical . [ policy space reduction for @xmath130][thm : absreduce]given a marginal probability vector that each macro bs is transmitting abrb @xmath369 , the optimal abrb control policy of @xmath147 conditioned on has the following synchronous abrb structure : by lemma [ lem : sysman ] , for given marginal probabilities @xmath185 , the average mutual information region will be maximized if we maximize @xmath189 for all @xmath382 . for @xmath383 , we have @xmath192 . for @xmath384 , we have @xmath385 , where the equality holds if and only if @xmath130 has the structure in theorem [ thm : absreduce ] . this completes the proof . define the _ achievable mutual information region _ for subband @xmath390 as @xmath391_{\forall k\in\mathcal{u}_{a}}:x_{k}\le\overline{\mathcal{i}}_{m_{a},k}\left(q_{a},\pi_{a}\right)\right\ } , \label{eq : ratereg1g}\ ] ] where @xmath392 . it can be verified that @xmath393 is a convex region in @xmath335 . since the utility function @xmath138 is concave and increasing w.r.t . @xmath394 , the optimal policy @xmath395 must achieve a pareto boundary point of @xmath393 . for given abrb pattern @xmath160 and bs @xmath22 , define a region as @xmath396_{\forall k\in\mathcal{u}_{a}\cap\mathcal{u}_{n}}:\ : x_{k}\le i_{m_{a},k}\left(\pi_{a},\mathbf{a}\right)\right\ } .\ ] ] it can be verified that @xmath397 is a convex region . from lemma [ lem : sysman ] , we have @xmath398 for convenience , define @xmath399 then @xmath400 , we have @xmath401 . from ( [ eq : rma]-[eq : rmabar ] ) and the fact that @xmath402_{\forall k\in\mathcal{u}_{a}}$ ] is a pareto boundary point of @xmath393 , it follows that @xmath403_{\forall k\in\mathcal{u}_{a}\cap\mathcal{u}_{n}}$ ] is a pareto boundary point of @xmath404 and @xmath405_{\forall k\in\mathcal{u}_{a}\cap\mathcal{u}_{n}}$ ] is a pareto boundary point of @xmath406 . hence , there exists user scheduling policy @xmath407 satisfying @xmath408 and @xmath409 for all @xmath410 . then it follows that @xmath402_{\forall k\in\mathcal{u}_{a}}$ ] can be achieved by the control policy @xmath411 . [ [ proof - of - corollary - corequivalent - problem - pa - subproof - of - corollaypa ] ] proof of corollary [ cor : equivalent - problem - pa ] [ sub : proof - of - corollaypa ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ the first part of the corollary follows straightforward from theorem [ thm : absreduce ] and [ clm : policy - space - reductionpia ] . we only need to prove that problem ( [ eq : plare ] ) is bi - convex . the average mutual information in ( [ eq : uaobj ] ) can be expressed as @xmath412 where @xmath413 , @xmath414 , @xmath415 is the set of macro cell n - users , @xmath416 is the set of pico cell n - users , and @xmath417 is the set of pico cell i - users . it is easy to verify that @xmath418 is a concave function w.r.t . @xmath185 for fixed @xmath131 . using the vector composition rule for concave function @xcite , the objective in ( [ eq : plare ] ) is also concave w.r.t . @xmath185 and thus problem ( [ eq : plare ] ) is convex w.r.t . @xmath185 for fixed @xmath131 . for fixed @xmath185 , @xmath418 is a linear function of the user scheduling variables @xmath419 . hence problem ( [ eq : plare ] ) is also convex w.r.t . @xmath131 for fixed @xmath185 . [ [ proof - of - theorem - thmglobal - convergence - lasubproof - of - theoremla ] ] proof of theorem [ thm : global - convergence - la][sub : proof - of - theoremla ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ it is clear that @xmath420 . by the assumption in theorem [ thm : global - convergence - la ] , we have @xmath421 if @xmath422 is not a fixed point of @xmath229 . combining the above and the fact that @xmath314 is upper bounded , ao_a must converge to a fixed point @xmath235\in\delta$ ] . the rest is to prove that any @xmath235\in\delta$ ] is globally optimal for problem ( [ eq : plare ] ) . note that problem ( [ eq : plare ] ) is equivalent to the problem @xmath423_{\forall k\in\mathcal{u}_{a}}\in\mathcal{r}_{m_{a}}.\label{eq : equplapre}\ ] ] since the objective in ( [ eq : equplapre ] ) is a concave function w.r.t . @xmath424 , and @xmath393 is a convex region , the following lemma holds . [ optimality condition for ( [ eq : plare])][lem : opt - cond - pa]a solution @xmath425 , @xmath209 is optimal for problem ( [ eq : plare ] ) if and only if its average mutual information @xmath426 satisfies @xmath427_{\forall k\in\mathcal{u}_{a}}\in\mathcal{r}_{m_{a}}\ ] ] according to the step 1 of ao_a , for any abrb pattern @xmath428 and csi realization @xmath355 , the user scheduling vector @xmath429 of bs @xmath22 under @xmath430 is the optimal solution of @xmath431 where @xmath432 and @xmath433 is the average mutual information under @xmath434 . combining ( [ eq : optpiatuta ] ) and the fact that @xmath435 is the optimal solution of problem ( [ eq : plare ] ) with fixed @xmath430 , we have @xmath436 . this implies that @xmath434 satisfy the optimality condition in lemma [ lem : opt - cond - pa ] , and thus is the globally optimal solution . * * step 1 * * ( update the coefficients @xmath442 ) : for fixed @xmath443 , obtain the optimal solution @xmath444_{j=1, ... ,\left|\theta^{(i)}\right|}$ ] of the following convex optimization problem @xmath445,\forall j\:\textrm{and}\:\sum_{j=1}^{\left|\theta^{(i)}\right|}\check{q}_{j}=1,\nonumber \end{aligned}\ ] ] where @xmath446_{j=1, ... ,\left|\theta^{(i)}\right|}$ ] , and @xmath447 is the @xmath257 mis in @xmath443 . [ asymptotically optimal abrb profile][thm : asymptotic - equivalence - ofpb]algorithm b2 always converges to an abrb profile @xmath303 with @xmath322 . furthermore , the converged result is asymptotically optimal for high snr . i.e. @xmath456 , where @xmath457 for some positive constants @xmath458 s , and @xmath459 is the optimal objective value of @xmath249 . consider problem @xmath460 which is the same as @xmath249 except that there are two differences : 1 ) the fading channel @xmath461 is replaced by a _ deterministic channel _ with the channel gain between bs @xmath22 and user @xmath23 given by the corresponding large scale fading factor @xmath462 ; 2 ) an additional constraint is added to the user scheduling policy such that any two links @xmath290 having an edge ( i.e. , @xmath286 ) in the interference graph @xmath289 can not be scheduled for transmission simultaneously . it can be shown that the optimal solution of problem @xmath460 is asymptotically optimal for @xmath249 at high snr . moreover , using the fact that the achievable mutual information region in the deterministic channel is a convex polytope with @xmath463 as the set of _ pareto boundary _ vertices , it can be shown that @xmath460 is equivalent to the following problem @xmath464 where @xmath465 is the @xmath257 mis in @xmath448 . to complete the proof of theorem [ thm : asymptotic - equivalence - ofpb ] , we only need to further prove that algorithm b2 converges to the optimal solution of problem ( [ eq : pdet ] ) . using the fact that any point in a @xmath466-dimensional convex polytope can be expressed as a convex combination of no more than @xmath254 vertices , it can be shown that there are at most @xmath254 non - zero elements in @xmath467 in step 1 of algorithm b2 . hence @xmath468 . moreover , it can be verified that @xmath469 if @xmath443 is not optimal for ( [ eq : pdet ] ) . combining the above and the fact that @xmath470 is upper bounded by @xmath471 , algorithm b2 must converge to the optimal solution of ( [ eq : pdet ] ) . this completes the proof . in step 2 of algorithm b2 , problem ( [ eq : mwisp ] ) is equivalent to finding a _ maximum weighted independent set _ ( mwis ) in the interference graph @xmath289 with the weights of the vertex nodes given by @xmath472 . the mwis problem has been well studied in the literature @xcite . although it is in general np hard , there exists low complexity algorithms for finding near - optimal solutions @xcite . although the asymptotic global optimality of algorithm b2 is not guaranteed when step 2 is replaced by a low complexity solution of ( [ eq : mwisp ] ) , we can still prove its monotone convergence . d. gesbert , s. kiani , a. gjendemsj _ et al . _ , `` adaptation , coordination , and distributed resource allocation in interference - limited wireless networks , '' _ proceedings of the ieee _ , vol . 95 , no . 12 , pp . 23932409 , 2007 . a. l. stolyar and h. viswanathan , `` self - organizing dynamic fractional frequency reuse in ofdma systems , '' in _ infocom 2008 . the 27th conference on computer communications . ieee_.1em plus 0.5em minus 0.4emieee , 2008 , pp . 691699 . r. irmer , h. droste , p. marsch , m. grieger , g. fettweis , s. brueck , h. mayer , l. thiele , and v. jungnickel , `` coordinated multipoint : concepts , performance , and field trial results , '' _ ieee communications magazine _ , vol . 49 , no . 2 , pp . 102111 , 2011 . q. shi , m. razaviyayn , z .- q . luo , and c. he , `` an iteratively weighted mmse approach to distributed sum - utility maximization for a mimo interfering broadcast channel , '' _ ieee trans . signal processing _ 59 , no . 9 , pp . 43314340 , sept . 2011 . y. wang and k. i. pedersen , `` performance analysis of enhanced inter - cell interference coordination in lte - advanced heterogeneous networks , '' in _ vehicular technology conference ( vtc spring ) , 2012 ieee 75th_. 1em plus 0.5em minus 0.4emieee , 2012 , pp . j. pang , j. wang , d. wang , g. shen , q. jiang , and j. liu , `` optimized time - domain resource partitioning for enhanced inter - cell interference coordination in heterogeneous networks , '' in _ wireless communications and networking conference ( wcnc ) , 2012 ieee_.1em plus 0.5em minus 0.4emieee , 2012 , pp . 16131617 . m. hong , r .- y . sun , h. baligh , and z .- q . luo , `` joint base station clustering and beamformer design for partial coordinated transmission in heterogenous networks , '' 2012 . [ online ] . available : http://arxiv.org/abs/1203.6390 r. ghaffar and r. knopp , `` fractional frequency reuse and interference suppression for ofdma networks , '' in _ proceedings of the 8th international symposium on modeling and optimization in mobile , ad hoc and wireless networks _ , 2010 , pp . 273277 . o. somekh , o. simeone , y. bar - ness , a. haimovich , and s. shamai , `` cooperative multicell zero - forcing beamforming in cellular downlink channels , '' _ ieee trans . inf . theory _ 55 , no . 7 , pp . 32063219 , 2009 .

using a similarity hamiltonian renormalization procedure , we determine an effective spin-1/2 representation of the bose - hubbard model at half - integer filling and at a finite on - site interaction energy . by means of bosonization , we are able to recast the effective hamiltonian as that of a spin-1/2 xxz magnetic chain with pertinently renormalized coupling and anisotropy parameters . we use this mapping to provide analytical estimates of the correlation functions of the bose - hubbard model . we then compare such results with those based on dmrg numerical simulations of the bose - hubbard model for various values of and for a number of lattice sites as low as . we find an excellent agreement up to between the output of analytical and numerical computations , even for relatively small values of . our analysis implies that , also at finite , the 1d bose - hubbard model with suitably chosen parameters may be seen as a quantum simulator of the xxz chain .

1210.2685

the study of magnetic systems is one of the most active fields of research in condensed matter physics @xcite : the variety of emerging ground - states , as well as the rich phase diagram of magnetic lattices , makes these systems an optimal testbed to probe the competition between various orders and frustration effects @xcite . from this perspective , it would be very useful to be able to engineer synthetic physical systems effectively describing magnetic model hamiltonians , with tunable geometry and parameters . a promising route is provided by cold atomic setups : for instance , itinerant magnetism in bulk ultracold fermi systems with repulsive interactions has been experimentally studied @xcite , while small spin networks have been simulated with ion chains @xcite . effective nearest - neighbour spin - spin interactions for atoms in neighbour wells of an optical lattice may result from super - exchange couplings : the corresponding second - order tunneling has been observed in array of double wells @xcite . furthermore , using fast oscillations of the optical lattice , it is possible to control the sign of the nearest - neighbour tunneling @xcite , which has been recently used to simulate classical frustrated magnetism in triangular lattices @xcite . one may also use two - component gases where the two internal degrees of freedom correspond to the simulated ( pseudo)spins . spin interactions can be tuned by adjusting the external potential @xcite . the recent realization of controllable bose - bose mixtures @xcite paves the way towards the experimental simulation of spin hamiltonians , in which the atomic counterpart of magnetic phases , like antiferromagnetic nel and xy ferromagnetic phases ( respectively corresponding to the checkerboard and the supercounterfluid phases @xcite ) may be detected and studied . a key tool in the manipulation of ultracold atomic systems is the possibility to superimpose and control optical lattices @xcite . the low - energy properties of ultracold bosons in deep optical lattices are well captured by the bose - hubbard ( bh ) hamiltonian : @xcite @xmath4 + \frac{u}{2 } \sum_{i } n_i ( n_i-1 ) \ , . \label{ham_gen}\ ] ] in eq . ( [ ham_gen ] ) , @xmath5 stands for any pair of nearest neighbouring sites , while the operators @xmath6 ( @xmath7 ) , with @xmath8 = \delta_{i , j}$ ] and @xmath9 , create ( annihilate ) a boson in the site @xmath10 . the parameter @xmath11 denotes the hopping strength , and @xmath0 ( @xmath12 ) is the interaction energy of two particles at the same site ( at two nearest neighbouring sites ) . the use of optical lattices in ultracold atomic systems is also central in other proposals to simulate spin hamiltonians , such as the quadratic - biquadratic spin model @xcite , or antiferromagnetic spin chains @xcite . following the latter suggestion , by means of a tilted 1d optical lattice , the ising chain in a transverse field was experimentally simulated @xcite . the paramagnetic , as well as the antiferromagnetic phase ( and the corresponding quantum phase transition ) , were detected by measuring the probability to have an odd occupation of sites , while the formation of magnetic domains was observed using in - situ site - resolved imaging and noise correlation measurements @xcite . for very large values of @xmath0 , i.e. for @xmath13 , the bh model can be mapped into the heisenberg xxz spin-1/2 hamiltonian : @xmath14 where @xmath15 are the @xmath16 spin operators , @xmath17 being the pauli matrices , @xmath18 is the nearest - neighbour coupling , and @xmath19 is the anisotropy parameter ( @xmath20 respectively correspond to the antiferromagnetic and the ferromagnetic isotropic heisenberg model ) . the use of lattice spin systems for interacting bosons traces back to the classical papers by matsubara and matsuda in the @xmath21 s , where the properties of helium ii were studied assuming that each atom can occupy one of the lattice points @xcite . the further assumption that two atoms can not simultaneously occupy the same lattice site ( due to the hard - core part of the interparticle interaction between helium atoms @xcite ) leads to an effective spin model in a magnetic field @xcite . to qualitatively understand the emergence of a spin representation of the one - component bh model one may say that , for @xmath22 and _ if _ two states per site give a dominant contribution to the energy , an xxz hamiltonian is retrieved : this is exactly what happens when the filling @xmath23 , defined as the average number of bosons per lattice site , is half - integer . indeed , for @xmath24 , with @xmath25 integer , the relevant states in the fock space are given by @xmath26 and @xmath27 ( deviations from half - integer fillings would result in a magnetic term in the xxz hamiltonian ) . for half - integer @xmath23 , at the leading order in @xmath28 one has @xmath29 and @xmath30 ( see the discussion in sec . [ building ] ) . the xxz model is a paradigmatic spin hamiltonian which has been the object of many investigations and that in 1d is exactly solvable by bethe ansatz @xcite ; this provides an ideal arena to test different analytical and numerical techniques , from bosonization @xcite to density matrix renormalization group ( dmrg ) @xcite . the study of ( static and dynamical ) correlation functions in this model is currently an active area of research @xcite and exact analytical results for the correlation functions at small distance ( both at zero and finite temperature ) are by now available @xcite . the asymptotic form of the ground - state correlation functions in the thermodynamic limit is power - law with an exponent that has been obtained by comparing the result of abelian bosonization with the bethe ansatz solution @xcite : for an open chain in the region @xmath31 , the numerical findings for correlation functions obtained with dmrg were compared with the results of a low - energy field theory , showing a very good agreement and allowing for precise estimates of the amplitudes of the correlation functions @xcite . in turn , the obtained amplitudes were found in agreement with the analytical expressions given by lukyanov and zamolodchikov @xcite . finally , exact results for the xxz chain in a special scaling limit were used to compute the local correlations of a continuous lieb - liniger 1d bose - gas @xcite . in this paper we determine a correspondence between the bh chain at half - integer filling for _ finite _ @xmath0 and a 1d xxz spin-1/2 model . this enables us to provide analytical expressions for the bh correlation functions , which we compare with numerical results obtained with dmrg , showing that there is a very good agreement both at large and small distances and also for @xmath32 as low as @xmath33 and for a number of sites @xmath34 . as a consequence , the numerical determination of the superfluid to charge - density - wave and superfluid to mott - insulator phase transitions ( respectively corresponding , in the effective xxz chain , to @xmath35 and @xmath36 ) well agrees with the analytical results for the xxz chain . using our approach , we are able not only to provide analytical expressions for the 1d bh correlation functions , but also to show that the bh chain at half - integer filling provides a reliable quantum simulator of the xxz chain . in the following we derive an effective spin-1/2 hamiltonian for the bh chain at half - integer filling as a power series of @xmath37 . following refs . , we perform a continuous unitary transformation @xmath38 which block - diagonalizes the hamiltonian in the basis of the eigenvectors of @xmath39 with @xmath40 and determine @xmath38 perturbatively to the order @xmath41 ( a similar technique has been used in ref . for the fermionic hubbard model ) . we finally show that , using bosonization , this hamiltonian can be recast in the xxz form with pertinent coupling and anisotropy parameters . we observe that , while to the first order in @xmath37 one finds a xxz model with @xmath29 and @xmath30 , to the next order in @xmath37 one gets an effective spin hamiltonian which is not of the xxz form , since it also contains next - nearest neighbours and @xmath42-spin terms ( this is the bosonic counterpart of a similar computation done for the 1d , as well as for the 2d , fermi - hubbard model @xcite , where @xmath43-spin terms appear ) . however , in 1d it is possible to proceed further using bosonization : introducing a luttinger liquid description of the effective hamiltonian , we are able to incorporate the long - wavelength behaviour of non - xxz terms in the effective coupling and anisotropy parameters , @xmath44 and @xmath45 , which are now function of @xmath11 , @xmath12 , @xmath23 and @xmath0 . the plan of the paper is the following : after introducing the bh and the xxz models and recalling some useful properties and results ( sec . [ models ] ) , we employ the continuous unitary transformation introduced by glazek and wilson @xcite to approximate the bh chain at half - integer filling with an effective spin-1/2 hamiltonian ( sec . [ building ] ) . in sec . [ equivalence ] we use bosonization to recast this effective hamiltonian as an xxz hamiltonian , with coupling @xmath44 and anisotropy @xmath45 , while in sec . [ correco ] we establish the correspondence between the correlation functions of the bh model and the ones of the xxz chain . we then proceed in comparing the analytical results obtained for the bh correlation functions with the numerical findings obtained by dmrg numerical simulations ( sec . [ results ] ) , both for the correlation functions and the phase transition points . section [ concl ] is devoted to our conclusions , while more technical details are contained in the appendices . let us start by reviewing the basic properties of the bh and of the spin-1/2 xxz hamiltonians , in particular focusing on known analytical results about the real - space spin correlations in the xxz chain . the low - energy properties of interacting bosons in a one - dimensional deep optical lattice are in general well described by the bose - hubbard hamiltonian ( [ ham_gen ] ) , which , in 1d and with open boundaries , reads : @xmath46 we denote with @xmath47 the total number of particles in the @xmath1-site chain , so that the filling @xmath23 , that is , the average number of particles per site , is given by @xmath48 . for alkali atoms usually @xmath49 , but with dipolar gases ( or polar molecules ) @xmath12 could be comparable with @xmath0 : experiments with dipolar gases @xcite and long - lived ground - state polar molecules @xcite in optical lattices have been already performed ( see also the review in ref . ) . a large amount of experiments investigated the properties of the bh model : the main reason for this interest lies on the fact that this model exhibits a quantum phase transition between a superfluid phase ( for @xmath50 ) and a mott insulator ( for @xmath13 ) @xcite . a finite @xmath12 generally favours charge - density - wave phases : e.g. , for half - integer filling @xmath51 , a large @xmath52 will result in a ground - state of the type @xmath53 ( where in general @xmath54 is an eigenfunction of @xmath39 with @xmath40 ) . the ground - state of the bh model has been studied in the seminal paper in ref . using the grand - canonical ensemble , where the chemical potential @xmath55 is introduced to enforce the constraint on the number of particles . the phase diagram in the @xmath56 plane shows the characteristic lobes : for a pertinently fixed value of @xmath55 , the half - integer fillings correspond to the `` basis '' of the lobes ( i.e. where the lobes touch ) and , for @xmath57 , one has a superfluid for each finite value of @xmath11 , while a finite and positive value of @xmath12 gives rise to a charge - density - wave region among the mott lobes . the mott - insulator / superfluid transition was first observed in 3d @xcite and subsequently in 1d @xcite and 2d @xcite . the effect of a superimposed external potential ( typically a parabolic one ) has been also considered : the so - called wedding - cake - like density has been studied both theoretically @xcite and experimentally @xcite . the coherence properties of ultracold bosons in optical lattices have been studied , as well , showing that phase coherence on short length scales still persists deep in the insulating phase @xcite . the bh model in a 1d geometry can be obtained either by tightly confining the bosonic cloud in two radial directions in presence of a periodic potential in the transverse direction , or by creating many ( eventually uncoupled ) tubes with a 2d optical lattice . the properties of strongly correlated phases across the superfluid to mott - insulator phase transition have been analyzed in 1d by means of bragg spectroscopy @xcite . the excitation spectrum in the strongly interacting regime has been also studied in presence of a tunable disorder , created by a bichromatic optical lattice , showing a broadening of the mott - insulator resonances @xcite . the finite-@xmath12 1d bh model has been studied with a number of analytical and numerical techniques : in particular in ref . the phase boundaries of the mott insulators and charge - density - wave phases were determined by dmrg . the zero - temperature phase diagram both of the bh model and of a spin - s heisenberg model was constructed and their relation investigated @xcite . the role of @xmath12 in inducing supersolid phases in the bh chain was also studied @xcite . bosonization techniques have been applied as well to bh chains , providing a very effective way to compute the correlation functions and their decay at large distance @xcite ) . finally , we mention that the effect of intersite interactions was considered since the 90 s in the related quantum phase model , describing josephson junction arrays @xcite : this can be obtained from the bh model for large filling per site when the number fluctuations are negligible in the kinetic term . the chemical potential term in the bh model corresponds to the so - called `` offset charge '' @xmath58 , which are external charges present in the superconducting network @xcite : the lobes in the quantum phase model are equal , since there is an invariance for @xmath59 ( @xmath60 being the charge of the cooper pairs ) , and an half - integer value of the filling @xmath23 corresponds to half - integer values of the offset charges @xmath61 . the study of intersite interactions is relevant in josephson junction arrays since the interaction term depends on the capacitance matrix @xmath62 , which is in general not diagonal , resulting in terms of the form @xmath63 , where @xmath64 : as a mean - field analysis shows @xcite , for a diagonal capacitance matrix one has that at @xmath65 the superconducting phase is obtained for each value of the josephson energy @xmath66 ( @xmath67 in the mapping ) and that at @xmath68 one has a finite critical temperature for the mott - insulator / superfluid transition for each finite value of @xmath66 ( unlike @xmath69 , where a critical value of @xmath66 is required ) . non - diagonal terms of the capacitance matrix favour charge density waves @xcite : the role of the intersite terms was considered for superconducting chains and the corresponding phase diagram investigated @xcite , revealing that in 1d a ( superconducting ) repulsive luttinger liquid phase exists . the opening of luttinger liquid phases with tunable parameters also allows for designing josephson junction networks supporting emerging two - level quantum systems with a high level of quantum coherence @xcite . to conclude this section let us mention that , in the rest of the paper , we will mostly deal with half - integer fillings , @xmath70 , with @xmath71 . the reason for such a choice is that in this case the relevant states for the description of system for @xmath22 are just @xmath26 and @xmath72 . simple arguments , reviewed in sec . [ xxzmodel_0 ] , then show that , to first order in @xmath37 , the bh hamiltonian is mapped into an xxz spin-1/2 hamiltonian which is integrable in 1d . within the xxz - model framework , it is also possible to consider small deviations from the half - filled regime , which mainly give rise to a uniform magnetic field in the @xmath73-direction . even though we will not consider large fluctuations in @xmath23 ( of order 1 ) , it is possible to take them into account , by keeping , as relevant states for @xmath22 , @xmath26 , @xmath74 , @xmath72 . in this case , an effective spin-1 xxz effective model ( in general not integrable ) is expected @xcite . spin-1 models exhibit a gapped ( haldane ) insulator phase @xcite , which has been investigated in the context of the 1d bh model @xcite . for a chain with @xmath1 sites and open boundaries , the hamiltonian of a spin-1/2 xxz model given in eq . ( [ xxz_gen ] ) particularizes to : @xmath75 the global minus sign in the couplings has been introduced in order to more easily perform the comparison with the bh model , and it can be readily gauged away by implementing the canonical mapping to the spin-1/2 operators @xmath76 defined as @xmath77 , @xmath78 . therefore the chain is antiferromagnetic ( ferromagnetic ) for @xmath19 positive ( negative ) . following ref . , one can derive the hamiltonian in eq . ( [ xxz_1 ] ) from the bh hamiltonian ( [ ham ] ) at half - integer filling @xmath23 and for @xmath22 . to do so , let us define @xmath79 ( so that the eigenvalues of @xmath80 are @xmath81 ) . since for @xmath40 the energy per particle is ( for @xmath82 ) @xmath83 , it follows that @xmath84 , i.e. @xmath85 similarly , for @xmath86 , one gets @xmath87 as one can see by putting @xmath88 and mapping the obtained result in the xxz spin-1/2 language @xcite : for finite values of @xmath23 one gets ( see sec . [ building ] ) @xmath89 eqs . ( [ v_0 ] , [ j_0 ] ) provide the desired mapping between the bh model and the xxz hamiltonian to lowest order in @xmath37 . however , as we are going to see in sec . [ correco ] , to get a quantitative agreement between the bh and the xxz correlation functions even for @xmath37 relatively small ( as low as @xmath90 for @xmath51 ) one has to go to the next order in @xmath37 : the corresponding hamiltonian is determined in sec . [ building ] and recast in xxz form via a luttinger representation in sec . [ equivalence ] . we remark that , since our result are obtained at half - integer filling , we may omit the addition of a magnetic field term of the form @xmath91 to eq . ( [ xxz_1 ] ) . indeed such a term is proportional to the total spin @xmath92 in the @xmath73 direction and , since the system is half - filled , only eigenstates of @xmath93 with @xmath94 are physically meaningful - notice that in the following analytical results based on the xxz hamiltonian ( [ xxz_1 ] ) are compared with numerical dmrg simulations of the bh chain in the canonical ensemble , where @xmath95 is conserved and equal to @xmath47 . the hamiltonian @xmath93 is exactly solvable by means of standard bethe ansatz techniques @xcite : however , explicitly computing the real - space spin - spin correlation functions is quite a difficult task . exact analytical results for short - range correlators in a range of up to seven lattice sites were reported for the isotropic heisenberg model in ref . , in the thermodynamic limit ( @xmath82 ) and at arbitrary finite temperature , and for finite chains of arbitrary length @xmath1 in the ground - state . results for short - range correlation functions are also available for the xxz chain @xcite . for large distances , using the standard bosonization approach @xcite to spin-1/2 xxz model @xcite , one may find out all the spin - spin correlation functions in terms of two - point correlators of pertinent conformal operators @xcite : in the thermodynamic limit one finds the asymptotic forms @xmath96 where @xmath97 is the ground - state of @xmath93 and we set @xcite @xmath98 analytical expressions for the correlation amplitudes @xmath99 , @xmath100 and @xmath101 entering eqs . ( [ therm_1 ] , [ therm2 ] ) were presented in refs . and further discussed in ref . ( see also the discussion in sec . v of ref . ): @xmath102 with @xmath103 } - \eta e^ { - 2 t } \biggr ) , \ ] ] @xmath104 } \\ & + \frac{1 } { \sinh ( \eta t ) } - \frac { \eta^2 + 1 } { \eta } e^ { - 2 t } \biggr ) , \end{split}\ ] ] @xmath105 } { \sinh ( \eta t ) \cosh [ ( 1 - \eta ) t ] } - \frac{2 \eta -1 } { \eta } e^ { - 2 t } \biggr ) , \ ] ] and @xmath106 and @xmath107 being the euler s gamma function . analytical expressions ( in the large-@xmath1 limit ) for the subsequent prefactors of the correlation functions are reported in refs . . for chains of finite size @xmath1 with open boundary conditions , one obtains @xcite : @xmath108 - \frac { ( -1)^j } { f_{\frac{1}{2 \eta } } ( 2 j ) } [ g ( i - j)- g ( i+j ) ] \biggr\ } \label{xxz_2}\end{aligned}\ ] ] and @xmath109 \nonumber \\ & & - \frac{bc}{2 } \ : { \rm sgn } ( i - j ) \left [ \frac { ( -1)^i } { f_{\frac{1}{2 \eta } } ( 2 j ) } - \frac { ( -1)^j } { f_{\frac{1}{2 \eta } } ( 2 i ) } \right ] \biggr\ } , \label{xxz_4}\end{aligned}\ ] ] where @xmath110 is the sign function and @xmath111^\alpha , \label{xxz_3_a } \\ g ( x ) & = & \frac { \pi}{2 ( l + 1 ) } \cot \left ( \frac { \pi x } { 2 ( l + 1 ) } \right ) , \label{xxz_3_b}\end{aligned}\ ] ] with @xmath112 ( here and in the following all the distances are in units of the lattice constant ) . the agreement between exact numerical calculations of the xxz correlation functions and analytical expressions in ( [ xxz_2 ] , [ xxz_4 ] ) is very good , and it becomes excellent with @xmath113 for @xmath114 @xcite . thus one may readily assume that eqs . ( [ xxz_2 ] , [ xxz_4 ] ) provide quite an accurate analytical expression for the spin - spin correlation functions in the xxz model @xcite . as a consequence , constructing a rigorous mapping between the bh and the xxz spin-1/2 hamiltonian and expressing correlation functions of one model in terms of the ones of the other model gives an efficient and straightforward way to provide accurate analytic expressions for real - space correlation functions in the bh model at half - integer filling . we finally observe that the only system - dependent parameter determining the spin - spin correlation functions is the coefficient @xmath115 : thus , in tracing out the mapping between the two models , this is the key quantity to be calculated as a function of the bh parameters . in particular , one may distinguish between the regions in parameter space with @xmath116 and @xmath117 : while the former one corresponds to an antiferromagnetic spin chain , the latter one ( which may be realized for pertinently chosen values of the parameters of @xmath39 , as we shall show below ) corresponds to a ferromagnetic chain . as reviewed in the previous section , for @xmath22 , the bh hamiltonian maps onto the xxz model in eq . ( [ xxz_1 ] ) , with the parameters @xmath118 given in eqs . ( [ v_0 ] , [ j_0 ] ) . this may be seen as a first - order term in an expansion ( in powers of @xmath37 ) aimed at computing the effective hamiltonian : in this section we compute this effective hamiltonian to the next order . as we shall show in the following , this is enough to fit quite well the numerical data for the correlation functions of the bh model using the analytical results obtained for the correlators of the xxz chain . to approach the large-@xmath0 limit one may either proceed by performing a strong coupling expansion to the second or higher - order of perturbation theory , or by deriving effective hamiltonians using alternative techniques , based on canonical transformations or continuous unitary transformations @xcite . at integer filling , for instance , it is possible to evaluate the energy of the mott insulator and of the superfluid state in higher - order perturbation theory and determine the phase diagram in the @xmath56 plane @xcite . since we are rather interested to the bh at half - integer filling , i.e. , in the region of the phase diagram where the lobes touch and the superfluid phase persists also at very small @xmath0 ( with @xmath57 ) , we found it convenient to use an approach based on continuous unitary transformations @xcite . we follow the notation and the method presented in the paper by glazek and wilson ( gw ) @xcite : systematically using the gw renormalization procedure , we work out an effective description of the dynamics of the bh model , restricted to the low - energy subspace determined by the constraint on the total number of particles and by the large-@xmath0 assumption . as a result , the low - energy subspace is spanned by states with either @xmath25 or @xmath119 particles per site , with the total number of particles being fixed to @xmath47 . thus , the space of physically relevant states at each site is in one - to - one correspondence with the hilbert space of states of a quantum spin-1/2 degree of freedom ; we shall see that , at half - integer filling , even for finite @xmath0 the bh model may be replaced by an effective spin-1/2 hamiltonian , with pertinently determined parameters . the method amounts to an iterative block - diagonalization of the bh hamiltonian on the space of eigenfunctions of @xmath39 with @xmath40 . to illustrate the procedure , we start from the explicit construction of the `` low - energy '' hilbert space of physically relevant states , in the large-@xmath0 limit . neglecting excitations with energy @xmath120 amounts to truncating the hilbert space to a subspace @xmath121 , defined as @xmath122 with @xmath123 taking the values @xmath124 and @xmath125 . in eq . ( [ spinsemi1 ] ) @xmath126 labels the state in the hilbert space with @xmath127 particles on site @xmath10 . to implement the gw approach , one splits the hamiltonian ( [ ham ] ) as @xmath128 , with @xmath129 from eqs . ( [ model_3_a ] , [ model_3 ] ) one sees that @xmath130 is diagonal with respect to the partition of the hilbert space into @xmath121 plus its orthogonal complement , since @xmath131 | n_1 , \ldots , n_l \rangle \label{model_5}\ ] ] with @xmath132 = ( u/2 ) \sum _ { i } n_i ( n_i - 1 ) + v \sum_i n_i n_{i+1}$ ] , while @xmath133 exhibits off - diagonal ( with respect to the partition of the hilbert space ) matrix elements which are @xmath134 . in order to block - diagonalize @xmath39 , one needs to perform a _ transformation @xcite @xmath135 with @xmath38 unitary . upon setting @xmath136 , the unitarity of @xmath38 implies the optical theorem @xmath137 setting @xmath138 , with @xmath139 one finds that eq . ( [ simi_1 ] ) yields @xmath140 eq . ( [ simi_3 ] ) shows that @xmath141 is always `` higher order '' than @xmath142 . following ref . , it is most convenient to define the new interaction hamiltonian @xmath143 as @xmath144 so that the new `` free '' hamiltonian is the same as the old one ( @xmath130 ) . to further proceed and determine @xmath38 , one has to require that the matrix elements of @xmath143 between states with energy difference @xmath145 are equal to zero , amounting to state that @xmath143 is block - diagonal with respect to the partition of the hilbert space into @xmath121 plus its orthogonal complement , i.e. @xmath146 where @xmath147 is the projector onto @xmath148 and @xmath149 the projector onto its complementary subspace @xcite . one sees that eq . ( [ simi_5 ] ) implies that @xmath150 using eqs . ( [ model_6 ] , [ simi_2 ] , [ simi_4 ] ) , one may write @xmath143 as @xmath151 and eq . ( [ simi_6 ] ) then becomes @xmath152 + { \bf t}^\dag h_i + h_i { \bf t } \\ & + { \bf t}^\dag h_i { \bf t } \ } \left ( { \bf i } - { \cal p } \right ) = 0 . \label{simi_7 } \end{split}\ ] ] eq . ( [ simi_7 ] ) , together with the identity @xmath153 and with eq . ( [ simi_3 ] ) , is all what one needs in principle to fully determine @xmath142 and @xmath141 ( and , therefore , the operator @xmath154 ) . however , except for some simple cases @xcite , an explicit solution for @xmath154 can not be exhibited . for this reason we proceed by writing the solution for @xmath154 iteratively , in a series in @xmath133 : in particular , we use eq . ( [ simi_7 ] ) to determine @xmath142 to first order ( @xmath155 ) in @xmath133 . we provide the details in appendix [ appgw ] and the result for @xmath155 in eq . ( [ simi_12 ] ) . using eq . ( [ simi_12 ] ) and setting @xmath156 , we find that eq . ( [ model_6 ] ) reads @xmath157 + [ h_i , { \bf a_1 } ] . \label{ex_1}\ ] ] the gw procedure may be readily iterated to determine , in principle , @xmath154 to any desired order in @xmath133 . however , since keeping only second - order contributions in @xmath133 provides already quite an excellent estimate for the real - space correlation functions of operators in the bh model ( as explicitly shown by the numerical calculations we report in sec . [ results ] ) , setting @xmath156 already provides quite a good approximation to the exact @xmath154 . since the approach we are implementing is perturbative in @xmath133 , one should enforce eq . ( [ simi_7 ] ) , as well as eq . ( [ simi_ter ] ) , to each order in @xmath133 ; moreover , since @xmath158 { \cal p } = 0 $ ] , one may neglect the term @xmath159 $ ] in eq . ( [ ex_1 ] ) and approximate the effective hamiltonian acting within @xmath148 as @xmath160 \right\ } { \cal p } \equiv h_{\rm xxz}^{(0 ) } + h^{(1)}. \label{ex_final}\ ] ] the first term in the right hand side of eq . ( [ ex_final ] ) yields a spin-1/2 hamiltonian which is actually the spin-1/2 xxz chain introduced in sec . [ xxzmodel_0 ] and having the anisotropy and the coupling given by eqs . ( [ v_0 ] , [ j_0 ] ) : @xmath161 with @xmath162 ( constant terms have been omitted ) . the effective spin-1/2 operators are defined as @xmath163 the boson number eigenstates at site @xmath10 correspond to the eigenstates of @xmath164 according to @xmath165 , and @xmath166 . therefore , the result in eq . ( [ spinsemi2 ] ) corresponds to the `` naive '' large-@xmath0 limit for the bh model at half - integer filling discussed in sec . [ xxzmodel_0 ] , in which off - diagonal matrix elements of relevant operators ( including the hamiltonian itself ) are set to zero from the very beginning . corrections to @xmath167 arising from virtual transitions involving states outside of @xmath148 may be properly accounted for within gw procedure , allowing to get the effective spin-1/2 hamiltonian to the next order in @xmath37 . summing over all virtual transitions outside of @xmath148 induced by @xmath133 , one finds @xmath168{\cal p } = - t^2 \sum _ { j , \ell } { \cal p } \left ( b_j^\dag b_{j + 1 } + b_{j + 1}^\dag b_j \right ) \times \nonumber \\ & & \times ( { \bf i } - { \cal p } ) \left ( h_{\rm bh } \right)^{-1 } \left ( { \bf i } - { \cal p } \right ) \left ( b_\ell^\dag b_{\ell + 1 } + b_{\ell + 1}^\dag b_\ell \right ) { \cal p}. \label{spinsemi5}\end{aligned}\ ] ] in particular , when computing @xmath169 , one has to consider intermediate states with either one of the @xmath170 in eq . ( [ spinsemi1 ] ) being equal to @xmath171 , or to @xmath172 ( all these states have energy @xmath120 , with respect to states in the subspace @xmath121 ) , or states with one of the @xmath170 equal to @xmath171 ( @xmath172 ) , and the other equal to @xmath172 ( @xmath171 ) ( all these states have energy @xmath173 , with respect to states in the subspace @xmath121 ) . thus , one eventually finds out that @xmath169 can be written as the sum of two terms : @xmath174 , with @xmath175 being the part of @xmath169 having 1- and 2-nearest - neighbour spin terms , while @xmath176 contains 2-next - nearest - neighbour and 3-spin terms . omitting constant terms , their expression are given by : @xmath177 as we shall see in the next section , using a luttinger liquid representation , @xmath169 may be recast in the xxz form with coupling and anisotropy coefficients depending on @xmath0 . the effective spin hamiltonian in eq . ( [ ex_final ] ) is not in the xxz form : in this section we show how the contribution coming from @xmath169 may be accounted for by a pertinent redefinition of the parameters of the spin-1/2 xxz - hamiltonian @xmath167 . the first contribution to @xmath175 in the right - hand side of eq . ( [ map1.2 ] ) describes an effective magnetic field in the @xmath73 direction @xcite , while the second term simply shifts the value of the xxz anisotropy . at variance , the term @xmath176 in eq . ( [ map1.3 ] ) contains @xmath42-spin , as well as non - nearest neighbour , couplings . to show how these terms can be accounted for via a redefinition of @xmath167 , it is most convenient to introduce the jordan - wigner ( jw ) fermions @xmath178 @xcite , in terms of which one gets @xmath179 where @xmath180 are jw fermionic operators in momentum space and @xmath181 denotes normal ordering with respect to the fermionic ground - state . in terms of jw fermions , one writes @xmath176 as a sum of a bilinear ( @xmath182 ) , plus a quartic ( @xmath183 ) term , that is @xmath184 with @xmath185 since @xmath182 is bilinear in the jw fermions , it merely modifies the single - fermion dispersion relation , yielding the quadratic hamiltonian in the jw fermions reading @xmath186 with @xmath187 . setting @xmath188 , one finds that the fermi points , defined by @xmath189 , are given by @xmath190 upon linearizing the dispersion relation around @xmath191 and setting @xmath192 , one gets @xmath193 \ , p. \label{map1.8}\ ] ] from eq . ( [ map1.8 ] ) one sees that , since @xmath194 , @xmath182 implies a nonzero effective magnetic field @xmath195 @xcite , as well as a redefinition of the fermi velocity @xmath196 . this yields a redefined coupling given by @xmath197 . since @xmath198 one obtains @xmath199 the quartic term @xmath183 can be dealt with by noticing that , in the low - energy , long - wavelength limit , one can write @xmath200 \biggr\ } \ , , \label{map1.12}\end{aligned}\ ] ] where the chiral fermion fields @xmath201 , @xmath202 are defined from the long - wavelength expansion of @xmath203 as @xmath204 with @xmath205 , and the chiral fermion densities given by @xmath206 and @xmath207 . as a result , @xmath183 may be written as @xmath208 comparing eq . ( [ map1.13 ] ) to eq . ( [ adiz.1 ] ) , one sees that @xmath183 takes the same form as the term @xmath209 in the spin-1/2 xxz hamiltonian in eq . ( [ xxz_1 ] ) . _ vs. _ @xmath210 for different values of @xmath211 and @xmath25 . the top ( bottom ) dotted line corresponds to the value of @xmath45 for @xmath212 and @xmath213 ( @xmath214 ) . the other lines are for @xmath215 ( top ) and @xmath216 ( bottom ) , with @xmath217 ( solid black lines ) , @xmath218 ( dashed red lines ) and @xmath219 ( dot - dashed blue blue lines ) . inset : same as in the main panel , but for @xmath115 _ vs. _ @xmath210 . ] collecting together all the above results allows to write an effective xxz hamiltonian , describing the bh model to the order @xmath41 , as : @xmath220 with @xmath44 defined in eq . ( [ map1.10 ] ) and @xmath221 with @xmath222 since @xmath44 acts just as an effective over - all scale of @xmath223 , then @xmath45 is the only parameter determining the behavior of spin - spin correlations in the xxz model . substituting eq . ( [ map1.17 ] ) in eq . ( [ eta_gen ] ) one gets @xmath224 which provides an explicit formula for the effective luttinger parameter for the bh model at half - integer filling . in fig . [ figure1 ] we plot both @xmath45 and @xmath115 versus @xmath210 , for different values of @xmath211 and @xmath25 . one sees that @xmath218 and @xmath219 are almost indistinguishable , and that the limit of the quantum phase model for josephson junction arrays ( @xmath225 ) at offset charge @xmath68 is practically reached at @xmath226 . furthermore , one sees that the dependence of @xmath115 upon @xmath25 is rather small . from fig . [ figure1 ] one also sees that @xmath45 may be tuned by varying the ratio @xmath210 : in particular @xmath45 can be different from @xmath227 even if @xmath57 ( as it is typical for alkali atoms ) . fig . [ figure1 ] also suggests the possibility of describing the whole phase diagram of the xxz spin-1/2 chain using bh model for a single species of bosons with pertinently chosen parameters , see also sec . [ results ] @xcite . finally we notice that , since the sign of @xmath228 may be changed by a pertinent choice of @xmath210 and @xmath12 , the luttinger liquid effectively describing the xxz - hamiltonian may be repulsive or attractive . as noticed in the context of 1d josephson junction arrays @xcite , the transition between the repulsive and the attractive side may be monitored by inserting a weak link ( i.e. , a nonmagnetic impurity @xcite ) : it would be then interesting to analyze the effects of a weak link introduced in a bosonic system described by the bh hamiltonian . the mapping between @xmath39 and @xmath223 derived in sec . [ equivalence ] enables to select the ground - states on which to compute the pertinent vacuum expectation values . indeed if @xmath229 is the ground - state of the bh hamiltonian given in eq . ( [ ham ] ) , and @xmath230 is the ground - state of @xmath231 , the gw approach requires @xmath232 \phi_0 \rangle & = & \langle \psi_0 | { \bf s}^\dag { \cal o}_{\rm bh } \left [ \left\ { b , b^\dag \right\ } \right ] { \bf s } | \psi_0 \rangle \nonumber \\ & \equiv & \langle \psi_0 | { \cal o}_{\rm xxz } \left [ \left\ { s^a \right\ } \right ] | \psi_0 \rangle , \label{corco.1}\end{aligned}\ ] ] where @xmath233 $ ] ( @xmath234 $ ] ) denotes a generic bh ( xxz ) operator . of course , eq . ( [ corco.1 ] ) is exact only if @xmath38 is the exact solution of the gw equation ( [ simi_ter ] ) : by computing it perturbatively at a given order , one recovers the correspondence between ground - state expectation values of bh and spin-1/2 operators at the chosen order . in the rest of the paper , we will be interested in correlation functions of the following bh operators : @xmath235 using the results of appendix [ appop ] one has @xmath236 $ ] , so that @xmath237 more generally , if the operator @xmath238 satisfies @xmath239 , then @xmath240 , with @xmath241 obtained from @xmath242 by substituting @xmath7 , @xmath243 and @xmath244 respectively with @xmath245 , @xmath246 and @xmath164 . at variance , for @xmath247 one obtains a more involved expression ( see appendix [ appop ] for details ) : @xmath248 s_{j}^+ + \left [ s_{j + 1}^+ + s_{j - 1}^+ \right ] s_{i}^- | \psi_0 \rangle \\ \nonumber & & + \frac{t \bar{n } ( \bar{n } + 1 ) } { 2 u } \langle \psi_0 | s_{i } ^- [ s_{j -1}^+ + s_{j + 1}^+ ] + s_{j } ^+ \left [ s_{i -1}^- + s_{i + 1}^- \right ] | \psi_0 \rangle + \delta _ { | i - j | , 1 } \ : \langle \psi_0 | \left ( \frac{1}{2 } - s_{i + 1}^z \right ) \left ( \frac{1}{2 } + s_{i } ^z \right ) | \psi_0 \rangle , \end{aligned}\ ] ] where again we neglected contributions arising to @xmath249 . in this section we compare the numerical results obtained by means of dmrg for the correlation functions and the phase diagram of the bh model with the analytical predictions for the correlators from the effective hamiltonian @xmath223 given by eq . ( [ map1.18 ] ) . let us focus on the bh correlation functions . since dmrg simulations are performed on a finite number of sites @xmath1 and for open boundary conditions , we may use eqs . ( [ xxz_2 ] , [ xxz_4 ] ) yielding the @xmath250 and @xmath251 correlation functions of the xxz model . we evaluate the values of the non - universal constants @xmath252 , @xmath253 , @xmath254 defined in eq . ( [ xxz_5 ] ) both numerically and analytically , by using the expressions presented in refs . and reported in sec . [ xxzmodel_0 ] . as confirmed in ref . , the values of @xmath252 , @xmath253 , @xmath254 obtained in the two ways are in excellent agreement . we show that the analytical expressions for the xxz correlations are well confirmed by the numerical bh correlations also for small @xmath1 ( e.g. , for @xmath255 ) and for @xmath210 relatively large ( as large as @xmath256 ) . it should be stressed that , at variance , the agreement is not very good by setting @xmath257 , i.e. by using the hamiltonian @xmath167 obtained for @xmath22 neglecting contributions arising from the gw procedure . the correlators @xmath258 and @xmath259 are evaluated from the corresponding xxz quantities using respectively eqs . ( [ corco.12 ] ) and ( [ corco.17 ] ) . they are plotted in figs . [ figure2]-[figure9 ] as a function of @xmath260 , with @xmath10 and @xmath261 such that @xcite @xmath262 , @xmath263 for odd @xmath264 , and @xmath265 , @xmath266 for even @xmath264 ( for instance , for @xmath267 sites , @xmath268 corresponds to @xmath269 ; @xmath270 corresponds to @xmath271 ; @xmath272 corresponds to @xmath273 , and so on ) . the meaning of the various symbols is summarized in the following table : [ cols="<,<",options="header " , ] where for simplicity @xmath274 ( @xmath275 ) denotes the average error for the xxz correlators at finite-@xmath0 ( infinite-@xmath0 limit ) with ( without ) the gw procedure . we see that , for the density - density @xmath250 correlations , the average error increases when the size @xmath1 decreases . _ vs. _ @xmath260 for different sizes : @xmath276 ( @xmath277 , @xmath278 , @xmath279 ) . black squares denote the bh results , red triangles the xxz results with @xmath252 , @xmath253 , @xmath254 numerically determined , and blue stars the infinite-@xmath0 xxz results ( we do not report the xxz results with @xmath252 , @xmath253 , @xmath254 analytically determined , since they practically coincide with the red triangles ) . ] @xmath260 for the sizes and the parameters of fig . [ figure6 ] . ] in figs . [ figure3 ] and [ figure5 ] we also plotted ( magenta circles and lines ) the results obtained according to eq . ( [ corco.1 ] ) , where we took @xmath280 and not @xmath281 . indeed , as stressed in ref . , solving the equation for @xmath38 amounts to perturbatively find a transformation enabling to block - diagonalize @xmath39 . the ground - state of @xmath39 changes accordingly : if one wants to compute expectation values of certain operators in the bh model , one has to rotate the chosen operator according the @xmath38 transformation - in other words , physical quantities in the effective theory are not simply the expectation values of the operators in the projected subspace : this guarantees the unitarity of the procedure . an example is already provided in ref . for the computation of the staggered magnetization in the 2d fermi - hubbard model with large-@xmath0 effective spin models . in appendix [ appop ] we give details on the the explicit computation of the gw rotation for the operators @xmath282 and @xmath283 . we remark that , while for density - density correlation functions ( @xmath250 correlations in the xxz model ) magenta circles coincide with black squares , this is not the case for @xmath284 ( @xmath251 planar correlations in the xxz model ) . in figs . [ figure8]-[figure9 ] we plot the @xmath250 and @xmath251 correlation functions for different values of @xmath0 : in these plots the ratio @xmath210 ranges from @xmath90 to @xmath285 . as expected , one sees that for @xmath286 the relative error made by the infinite-@xmath0 results is not very large ( @xmath287 for @xmath284 correlations ) , but , as soon as @xmath288 , it is already well visible . the relative error made by using the effective @xmath223 turns out to be rather small even for @xmath289 , where the relative error on density - density correlations is only @xmath290 , while for @xmath284 correlations it is @xmath291 @xcite . _ vs. _ @xmath260 for different values of @xmath292 , corresponding , respectively , to @xmath293 ( with @xmath278 , @xmath279 , @xmath294 ) . ] @xmath260 for the @xmath0 values and the parameters of fig . [ figure8 ] . ] the xxz model is gapless and critical for @xmath295 , antiferromagnetic for @xmath296 and ferromagnetic for @xmath297 : in the latter ferromagnetic phase , all the spins are aligned . however , the bh at half - integer filling maps into the effective xxz chain ( [ map1.18 ] ) supplemented by the condition that the total spin is vanishing : therefore we expect that , in the bh model at @xmath298 , domain walls form separating regions with `` up '' spins ( i.e , with @xmath299 particles per site ) and regions with `` down '' spins ( i.e. , with @xmath300 particles per site ) . at variance , at @xmath301 the staggered magnetization becomes non vanishing : in the bosonic bh language the antiferromagnetic state corresponds to the `` charge - checkerboard ordered state '' @xmath302 . this shows that , consistently with the xxz representation of the bh model at half - integer filling , at finite @xmath0 one can realize the transition between the spin - liquid and the nel - ising antiferromagnetic phase of the xxz model ( superfluid to charge - density - wave phase transition of the bh model ) , as well as the transition between the spin - liquid and the domain ferromagnetic ising phase of the xxz model ( superfluid to domain mott - insulating phase transition of the bh model ) @xcite . since the former transition sets in at @xmath303 and the latter one at @xmath304 , using eq . ( [ map1.17 ] ) for @xmath45 allows to determine the corresponding phase boundaries in terms of the parameters of the bh hamiltonian . a complete discussion of the phase diagram of the bh chain in presence of nearest - neighbour interactions is provided in ref . : here we just focus on the half - integer bh chain with parameters chosen so as to lie close to @xmath305 , in order to show that the effective xxz representation given in this paper also provides a good description of these transitions . for the spin - liquid / ferromagnetic transition , we studied the bh chain with open boundary conditions varying @xmath12 ( similar results are obtained varying @xmath11 ) and we plot in fig . [ figure10 ] the expectation value of @xmath306 as a function of the position @xmath10 along the chain . we observe that , as a consequence of the open boundary conditions , a magnetic field proportional to @xmath12 on the two boundaries ( i.e. , at @xmath307 and @xmath308 ) appears , whose effect close to the boundaries is clearly visible in the figure . computing the quantity @xmath309 , one sees that it significantly increases around a critical value @xmath310 . from the numerical data for the bh model shown in fig . [ figure10 ] one may estimate @xmath311 , in good agreement with the analytical value @xmath312 @xcite . we notice that a better estimate of @xmath310 could be performed by adding a magnetic field in the boundaries to compensate the boundary magnetic fields arising from the open boundary conditions . _ vs. _ @xmath10 numerically computed in the bh chain for @xmath277 , @xmath279 and @xmath294 for different values of @xmath12 : from top to bottom @xmath313 ( black circles ) , @xmath314 ( red squares ) , @xmath315 ( green diamonds ) and @xmath316 ( blue stars ) corresponding , respectively , to @xmath317 . ] regarding the domain ferromagnet / superfluid transition , we performed numerical simulations on the bh model with parameters chosen such that @xmath45 is close to @xmath172 ( see figs . [ figure11]-[figure12 ] ) . in fig . [ figure11 ] we plot @xmath318 as a function of the position @xmath10 : one sees that the expectation value of the spin is constant and it changes sign close to the edges of the chain in order to satisfy the constraint on the number conservation . for this reason we then plot the modulus of the same quantity in fig . [ figure12 ] : since the average of the @xmath164 expectation values is of course zero , to determine the transition point from bh numerical data we consider the averaged quantity @xmath319 ( e.g. , for the different values of @xmath12 shown in fig . [ figure12 ] , such quantity is reported in the caption ) . from these data one can estimate that the domain ferromagnet is occurring at @xmath320 , with an error of few percent with respect to the analytical result @xmath321 @xcite . notice that the error made by using @xmath93 with @xmath30 in the infinite-@xmath0 limit is @xmath322 : as expected , the errors made in using the infinite-@xmath0 results are generally smaller when one deals with global quantities . _ vs. _ @xmath10 numerically computed in the bh chain for @xmath277 , @xmath279 and @xmath294 for different values of @xmath12 : @xmath323 , corresponding , respectively , to @xmath324 . ] _ vs. _ @xmath10 computed in the bh chain for the same parameters of fig . [ figure11 ] : @xmath323 . the corresponding averages are @xmath325 . ] in this paper we studied an xxz representation of the bose - hubbard chain at half - integer filling for finite on - site interaction energy @xmath0 . the effective xxz model is obtained in two steps : first , we used a similarity renormalization group procedure amounting to solve perturbatively up to the order @xmath41 the exact equation for the operator block - diagonalizing the bose - hubbard model . the resulting spin-1/2 effective hamiltonian is then recast as a xxz spin-1/2 hamiltonian with pertinently redefined coupling and anisotropy parameters . we use this mapping to provide analytical estimates of the correlation functions of the bose - hubbard model at half - integer filling and finite @xmath0 . we then compared these analytical results with the outcomes of the numerical dmrg evaluation of the bose - hubbard correlation functions . we found that the agreement is very good , also for @xmath210 rather large ( @xmath256 ) and for small number of sizes ( @xmath2 ) . such a good agreement is not achieved , even for @xmath210 relatively small ( @xmath326 ) , if one uses the xxz hamiltonian @xmath167 with @xmath29 and @xmath30 corresponding to the infinite - coupling limit of the bose - hubbard model . the transitions predicted at @xmath327 for the xxz chain are as well compared with bose - hubbard results , and a good agreement is found . since the bh model at half - integer filing is not integrable or exactly solvable , it is quite valuable to have analytical estimates for its correlation functions . besides its mathematical interest , we stress out that our results can be viewed from a two - fold point of view : on one side , we use known results from the ( integrable ) xxz model to construct with high accuracy correlation functions of the bose - hubbard model . on the other side , the bose - hubbard chain at half - filling and at finite @xmath0 may be seen as a quantum simulator of the xxz chain . finally , our results could be relevant towards extending to the bh model the analysis of nonequilibrium steady state in the xxz chain performed in ref . . in our approach , the effect of an harmonic trap results in a locally varying magnetic field : we feel that it would be interesting to compare the results stemming from an xxz - based approach with the ones known in literature for hard- and soft - core bosons in harmonic traps in the scaling limit @xcite . in this paper we focused on the half - integer filling bose - hubbard model , but deviations from such filling could be easily accounted with the introduction of a magnetic field . we stress that the similarity hamiltonian renormalization procedure could also be applied to bosonic ladders @xcite and at integer filling , where a spin-1 model is found in the infinite-@xmath0 limit . the large-@xmath12 effects of edge magnetic field could also be studied , following the results known for the xxz chain @xcite : we observe that , for open boundary conditions and finite @xmath12 , two boundary magnetic field terms @xmath328 , with @xmath329 , emerge in the xxz effective hamiltonian @xcite . since a magnetic field at the edge induces corrections to the average value of @xmath164 decreasing as a power law @xcite , these corrections are not only expected , but could be also worth the effort of future investigation . we would like to thank f. becca , l. campos - venuti , f. essler , a. ferraz , v. korepin , f. minardi , m. mller , g. santoro and a. smerzi for very useful discussions . a.t . acknowledges kind hospitality from iip - ufrn ( natal ) , where part of this work was performed . d.r . acknowledges financial support from eu through the project solid . in this appendix we show how use eq . ( [ simi_7 ] ) to determine @xmath142 to _ first order _ in @xmath133 , that is @xmath155 . to this order , one gets @xmath330 \right\ } \left ( { \bf i } - { \cal p } \right ) = 0 , \label{simi_9}\ ] ] which may be solved by setting @xmath331 \nonumber \\ & & \times \big\ { - { \cal p } h_0 { \cal p } + \left ( { \bf i } - { \cal p } \right ) h_0 \left ( { \bf i } - { \cal p } \right ) \big\}^{-1}. \label{simi_10}\end{aligned}\ ] ] up to term that are second order in @xmath332 , we may make the approximation @xmath333 { \bf i}$ ] , with @xmath334 = l \left\ { \frac{u}{2 } \bar{n } ( \bar{n } - 1 ) + v \bar{n}^2 \right\}$ ] , which implies @xmath335 = 0 $ ] . as a result , we get @xmath336 using the fact that @xmath142 is antihermitean , from eq . ( [ simi_11 ] ) one obtains @xmath337 an advantage of the gw procedure is that it may be easily applied to single - boson operators : in particular , we are interested in the average values of the operators @xmath247 and @xmath338 defined in eqs . ( [ corko.a ] ) and ( [ corko.b ] ) . since @xmath155 is fully off - diagonal and @xmath339 , if one approximates @xmath340 with @xmath155 , one obtains @xmath341 . instead , acting onto @xmath342 gives raise to a more complicated expression : expressing the final result in terms of spin-1/2 variables , one obtains we observe that due to the constraint on the fixed total particle number @xmath47 , the total magnetization in any of the physical states of @xmath344 is zero : since @xmath345 contains no terms breaking the parity symmetry ( @xmath346 ) , its ground - state @xmath347 is nondegenerate and , thus , it must be parity invariant . as a consequence , the average of any product of three spin-1/2 operators must necessarily give @xmath227 , greatly simplifying the calculation of the ground - state average of the operator . since only eigenstates with zero total spin along the @xmath73 axis are physically meaningful ( due to the constraint on the total number of particles ) , in the following we use the formulas of ref . at vanishing applied magnetic field . although the contribution to the total energy arising from magnetic fields vanishes on the physical states due to the constraint on the particle number , in the intermediate calculations one has to retain it when computing the redefined xxz parameters . notice that , as discussed in the following , the constraint on the total number ( implying that only states of the xxz model with the z component of the total spin equal to zero may be realized within the bh model in the canonical ensemble ) does not affect the spin liquid / nel - ising phase transition since it just implies the existence of ( at least two ) ferromagnetic domain walls , in the ferromagnetic- ising phase of the xxz model . in passing we point out that , to obtain more accurate results on the phase transition points , a finite - size scaling of the data with @xmath1 is needed . namely , one should estimate the value of @xmath354 for different sizes @xmath1 , and then perform a fit of that quantity with @xmath1 , such to extrapolate the thermodynamic limit . this analysis lies beyond our present purposes , which are to provide estimates of the transition within a precision of @xmath355 . notice that computing the renormalization of the effective xxz parameters via the luttinger treatment done in sec . [ equivalence ] with periodic boundary conditions is enough to give excellent agreement with bh numerical data , provided that , of course , the correlation functions of the xxz model are computed for open boundary conditions according to eqs . ( [ xxz_2 ] , [ xxz_4 ] ) .

in this paper the possibility of generating large scale curvature perturbations induced from the entropic perturbations during the waterfall phase transition of standard hybrid inflation model is studied . we show that whether or not appreciable amounts of large scale curvature perturbations are produced during the waterfall phase transition depend crucially on the competition between the classical and the quantum mechanical back - reactions to terminate inflation . if one considers only the classical evolution of the system we show that the highly blue - tilted entropy perturbations induce highly blue - tilted large scale curvature perturbations during the waterfall phase transition which dominate over the original adiabatic curvature perturbations . however , we show that the quantum back - reactions of the waterfall field inhomogeneities produced during the phase transition dominate completely over the classical back - reactions . the cumulative quantum back - reactions of very small scales tachyonic modes terminate inflation very efficiently and shut off the curvature perturbations evolution during the waterfall phase transition . this indicates that the standard hybrid inflation model is safe under large scale curvature perturbations during the waterfall phase transition .

1005.2934

inflation proved to be very successful both theoretically @xcite and observationally @xcite as a theory of early universe . the simplest models of inflation consist of a scalar field which is minimally coupled to gravity . a period of acceleration expansion is obtained if the potential is flat enough to allow for the inflaton field to slowly roll towards its minimum . with sufficient tunings in the parameters of the model , one can achieve 60 number of e - foldings or more to solve the horizon and the flatness problems of the standard cosmology . hybrid inflation @xcite is an interesting model of inflation containing two scalar fields , the inflaton field and the waterfall field . in linde s original hybrid inflation , the energy density during inflation is dominated by the vacuum while the inflaton is slowly rolling . the waterfall field is very heavy compared to the hubble expansion rate during inflation , @xmath0 , and it quickly rolls to its instantaneous minimum . the potential has the property that once the inflaton field reaches a critical value , @xmath1 , the waterfall field becomes tachyonic triggering an instability and inflation ends quickly thereafter and the systems settles down into its global minimum . usually it is assumed that the waterfall field does not play any role in curvature perturbations during inflation and during phase transition . in this picture one basically borrows the technics and the results of single field inflationary models . that is , the super - horizon curvature perturbations , once they leave the hubble radius , are frozen and remain unchanged until they re - enter the hubble radius at a later time , such as at the time of cmb decouplings . here we would like to examine this picture more closely . we would like to see if hybrid inflation is safe under large scale curvature perturbations during the waterfall phase transition . if one considers only the classical evolution of the system , we show that during the phase transition the highly - blue tilted entropy perturbations can induce large blue - tilted curvature perturbations on super - horizon scales which can completely dominate over the original adiabatic curvature perturbations . however , we show that the quantum back - reactions of the waterfall field inhomogeneities produced during the phase transition become important before the classical back - reactions become relevant . we demonstrate that the cumulative quantum back - reactions of the short - wavelength inhomogeneities are so strong that they uplift the tachyonic instability of the entropy perturbations and the curvature perturbations freezes . ideas similar to this line of thought , studying the amplifications of large scale curvature perturbations during preheating , were studied in @xcite-@xcite and more recently in @xcite . the paper is organized as follows . in section [ hybrid ] we review the basics of hybrid inflation and obtain the background evolutions of the inflaton and the waterfall fields . in section [ entropy ] the entropy perturbations and in section [ curvature ] their effects on curvature perturbations are studied . in section [ back ] the classical non - linear back - reactions as well as the quantum mechanical back - reactions are calculated and are compared to each other . brief conclusions and discussions are followed in section [ conclusions ] . while our work was finished the work by lyth @xcite appeared which has overlaps with our results . see also @xcite which appeared shortly after our work . here we study the basics of hybrid inflation @xcite and the background field dynamics . the potential in standard hybrid inflation has the form @xmath2 where @xmath3 is the inflaton field , @xmath4 is the waterfall field and @xmath5 and @xmath6 are dimensionless couplings . the system has a global minimum given by @xmath7 and @xmath8 . inflation takes place for @xmath9 where @xmath10 is the initial value of the inflaton field and @xmath11 is the critical value of @xmath3 where the waterfall field becomes instantaneously massless . during inflation @xmath4 is very heavy and is stuck to its instantaneous minimum @xmath12 . for @xmath13 the waterfall becomes tachyonic triggering an instability in the system which ends inflation abruptly . soon after phase transition , the systems settles down to its global minimum and inflation is followed by the ( p)reheating phase . as in linde s realization of hybrid inflation @xcite , we consider the limit where the inflation is dominated by the vacuum . for this condition to hold one requires that @xmath14 to solve the flatness and the horizon problem , we assume that inflation proceeds at least for about 60 number of e - foldings . in the vacuum dominated limit the number of e - foldings is given by @xmath15 where @xmath16 with @xmath17 being the newton s constant . we assume that @xmath18 is few times @xmath19 so one can basically neglect the logarithmic contribution above . to get the correct amplitude of density perturbations , one has to satisfy the cobe normalization for the curvature perturbations @xmath20 . the power spectrum of curvature perturbations is = ~ , where the relevant quantities are calculated at the time of hubble radius crossing ( @xmath21 ) at 60 e - folds before the end of inflation . in this picture , it is assumed that the curvature perturbations are frozen once the modes of interest leave the hubble radius , as have been treated in conventional analysis of hybrid inflation so far . our main goal in this paper is to examine the validity of this assumption more closely . we are interested in the limit where the waterfall field rolls rapidly to its global minimum once the instability is triggered . for this to happen , the absolute value of the @xmath4 mass should be much bigger than the hubble expansion rate during phase transition so @xmath22 here we study the classical evolutions of background fields @xmath3 and @xmath4 during inflation and phase transition , see also @xcite where somewhat similar analysis were carried out too . in subsection [ back - reaction ] we study the quantum back - reactions to the the system in the hartree approximation . the equations of motion for @xmath3 and @xmath4 are + 3h + ( m^2+g^2 ^2 ) = 0 + + 3h + ( -m^2+g^2 ^2+^2 ) = 0 . with the assumption of vacuum dominated potential , the hubble expansion rate is nearly constant during inflation , @xmath23 . it is more convenient to use the number of e - foldings as the clock , @xmath24 and the background fields equations are now written as [ phi - c ] + 3 + ( + g^2 ) = 0 + [ psi - c ] + 3 + ( -+g^2 + ) = 0 , where the dimensionless parameters @xmath25 and @xmath26 are defined as = , = , and the prime denotes the differentiation with respect to the number of e - foldings . using the rapid waterfall condition ( [ waterfall - condition ] ) combined with the expression for the total number of e - foldings eq . ( [ ne ] ) one obtains [ vd - ne ] n_e . similarly , one can check that @xmath27 . to simplify the notation , we take the critical point as the reference point and define @xmath28 . we use the convention that at the start of inflation for @xmath29 , @xmath30 , at the time of phase transition @xmath31 and at the end of inflation @xmath32 . with this convention @xmath33 before phase transition whereas @xmath34 afterwards . as assumed , the @xmath4 field is much heavier than @xmath35 during inflation so it rapidly rolls down to @xmath12 and one can simply solve eq . [ phi - c ] [ phif ] ( n ) _ c ( -r n ) with r = ( - ) . equivalently , one also has [ nc ] n _ c ( ) . ) with @xmath36 , @xmath37 and @xmath38 in the units where @xmath39 . the upper solid blue curve shows the amplitude of curvature perturbations , the middle solid green curve shows the background @xmath3 field evolution and the lower solid red curve shows the behavior of background @xmath4 field , all three curves are obtained from full numerical analysis . the dashed dark blue curve shows our analytical solution for @xmath4 field . the vertical axis is logarithmic and the horizontal axis is the number of e - foldings . ] [ bkd ] let us now turn to the dynamics of @xmath4 field . as it can be confirmed from our full numerical results we are in the limit where @xmath40 so the equation for @xmath4 simplifies to [ psi - substituted ] + 3 + ( e^-2rn -1)=0 . we divide the solution into two regions . first , we solve this equation for the period before phase transition , @xmath41 . the solution of this equation during this period is = e^-3n/2 , where @xmath42 and @xmath43 are the bessel functions , @xmath44 and @xmath45 are constants of integrations and @xmath46 . as explained previously , we are in the vacuum dominated limit so @xmath47 and @xmath48 . using the approximations for the bessel functions with large arguments one finds [ psi - nc1 ] e^-(3-r)n /2 . this solution means that @xmath4 has oscillatory behavior with exponentially decaying frequency and exponentially decaying amplitude . this behavior can be seen in * fig . [ bkd]*. now , we solve the @xmath4 evolution for the period after phase transition till end of inflation , @xmath49 . since inflation ends in few e - folds after @xmath3 reaches the critical point , it is more appropriate to use the small @xmath50 approximation in eq . ( [ psi - substituted ] ) and [ psi - eq - water ] + 3 - ( 2rn)=0 , which has the following solution e^-3n /2 , in which _ , and ai(x ) and bi(x ) are airy functions of first and second kind respectively . for @xmath51 , which is the case in our analysis to satisfy the water - fall condition , the second term in the argument of airy function can be ignored . since ai(x ) is a damping function for @xmath52 we just keep the term containing bi(x ) . in a good approximation the solution can be read as [ orpsi ] ( n ) ( n_c ) ( ^2/3 _ n ) e^-3n/2 . where @xmath53 . by using the asymptotic behavior of @xmath54 for large @xmath55 , @xmath56 , bi(z ) , and keeping just the exponential dependence one finds that for @xmath57 [ psi-2 ] _ i ( - n_c ) ( -n + n^3/2 ) , where to get the final result eq.([phif ] ) have been used . * are shown in the last few e - foldings after the phase transition at @xmath58 . here we see that , considering only the classical fields dynamics , @xmath59 grows rapidly at the end of inflation until the condition ( [ back - phi ] ) is met when @xmath60 at @xmath61 . this is followed by a short period of inflation when condition ( [ back - psi ] ) is met and the non - linear corrections to @xmath4 dynamics settles it down to the global minimum ending inflation . ] here we look into entropy perturbation following @xcite closely . in the field space of @xmath62 one can perform local fields rotations such that = + , s = - + , such that @xmath63 and @xmath64 . in this picture , @xmath65 and @xmath66 represent , respectively , the adiabatic and the entropic perturbations . one can check that the evolution of the curvature perturbation for a mode with momentum @xmath67 is [ dotr ] = + s , where [ dot - theta - eq ] = -v_s / with @xmath68 . ( [ dotr ] ) indicates that in the presence of large entropy perturbations or sharp turns in field space , the curvature perturbations can change on super - horizon scales . the equation of entropy perturbations is @xcite [ deltas - eq ] + 3h + ( + v_ss+3 ^ 2 ) s = , in which @xmath69 . field space with the same parameters as in * fig . [ bkd ] * : the upper solid black curve and the lower solid blue curve , respectively , show the full numerical solutions of @xmath70 and @xmath71 . the upper dashed green curve and the lower dashed red curve , respectively , show the evolution of the corresponding quantities obtained from our analytical solutions , eqs . ( [ theta ] ) and ( [ thetap ] ) . inflation ends when @xmath72 indicated in the graph by the vertical line . ] we are interested in solving this equation for @xmath73 . as we shall see later , these conditions are true until the end of inflation which can also be seen from * fig . [ bkt]*. for the later references , from eq . ( [ phif ] ) and eq.([psi-2 ] ) , one can easily find that [ theta ] = = - ( ) ( - n_c ) ( ^5/6 _ n^1/4 ) ( n^3/2 - n ) , which for @xmath34 or @xmath57 ( after phase transition ) results in : [ thetap ] = - ( ) ( - n_c ) ( ^11/6 _ n^3/4 ) ( n^3/2 - n ) . neglecting @xmath74 in eq . ( [ deltas - eq ] ) , the equation governing the evolution of entropy fluctuations is the same as the equation of background @xmath4 field , eq . ( [ psi - substituted ] ) . this indicates that , although the entropy perturbations are suppressed before the phase transition , but they becomes highly tachyonic during the phase transition . the subsequent tachyonic enhancement of the entropy perturbations overcome their initial suppression prior to phase transition . this is the key effect to obtain large curvature perturbations induced from the entropy perturbations at the classical level . however , as we shall see in [ back - reaction ] , one should also take into account the quantum back - reactions which can uplift the tachyonic instability of the background @xmath4 fields and modify the evolution significantly . the differences between @xmath66 and @xmath4 evolutions therefore are only in the initial conditions . for a given mode , we consider the evolution of the entropy fluctuation when it leaves the hubble radius at the moment of hubble radius crossing @xmath75 , so for @xmath76 , similar to eq . ( [ psi-2 ] ) , one obtains [ delpsi - n ] s(n ) s _ ( ) ^1/2 ( n _ ) ( -n_c ) ( - + n^3/2 ) . we note that the contribution @xmath77 represents the tachyonic enhancement of the entropy perturbations during the waterfall phase transition whereas the factors @xmath78 indicates the suppression of the entropy perturbations from the time of horizon crossing till the onset of phase transition . both of these competitive behaviors can be seen in * fig . [ bkd]*. to find the final amplitude of entropy perturbations , we need to find their amplitude at the time of hubble radius crossing , @xmath79 . to do this we note that the entropy perturbations , which are basically the waterfall field perturbations , are very heavy during inflation . as explained previously , this causes their suppression before the phase transition which should be taken into account . during inflation @xmath80 so @xmath81 . rewriting the mode equation in conformal time @xmath82 and defining @xmath83 the equation of entropy perturbation is + v_k = 0 in which ( -1 ) . since we are interested in momenta which exit the horizon during first few e - folds of inflation the above ratio is nearly constant and ( -1 ) ~ . by these considerations and noting that @xmath47 one has + v_k = 0 . as the frequency of this equation changes adiabatically , one can use the wkb approximation with the bunch - davis vacuum for initial times or @xmath84 and obtains [ qem ] v_k e^i . using this solution the amplitude of the entropy perturbation at the time of hubble radius crossing , @xmath85 , is obtained to be s _ _ . the extra factor @xmath86 represents the suppression of the entropy perturbations before the phase transition . assuming that the scale factor at the start of inflation , @xmath87 e - folds before the end of inflation , is unity , @xmath88 , one also has @xmath89 . the final amplitude of entropy perturbations at the end of inflation therefore is [ delsf ] s_f ( ) _ ^(3-r)/2 ( ) ^1/2 e^- 3/2 n_c ( - n_f + n_f^3/2 ) , where @xmath90 denotes the number of e - foldings from the start of phase transition till end of inflation . for a vacuum dominated potential with a quick phase transition , we have @xmath91 . the final amplitude of the entropy perturbation @xmath92 is [ ps ] p_s = a_s ( ) ^(3-r ) , with = ( ) ^2 ( ) e^- 3 n_c ( - 3 n_f + n_f^3/2 ) . from eq . ( [ ps ] ) one observes that the entropy perturbations are highly blue - tilted with @xmath93 . in next section we show that at the classical level these highly blue - tilted entropy perturbation induce large blue - tilted spectrum on super - horizon curvature perturbations . we now have all the materials to calculate the final power spectrum of curvature perturbations . for this purpose we need to know the amplitude of adiabatic curvature perturbations at horizon crossing as the initial conditions and integrate the evolutions of curvature perturbation from the time of horizon crossing till end of inflation . the final amplitude of curvature perturbation , therefore , is _ f = r_0 + _ 0^n_f r n , where @xmath94 represents the adiabatic curvature perturbations in the absence of entropy perturbations . starting with eq . ( [ dotr ] ) , the evolution of curvature perturbations for the super - horizon modes , induced by the entropy perturbations , can be written as [ r - eq ] r= s . as one can see from above equation both @xmath95 and @xmath66 can source the curvature perturbations . we also note that @xmath95 represents the acceleration of the field @xmath4 , specially during the phase transition . as can be seen from the full numerical analysis , our classical background is such that during inflation and phase transition , @xmath96 . however , shortly after phase transition @xmath4 rises quickly from its value during inflation @xmath12 to its final value at the global minimum @xmath97 when inflation ends . the rapid rise of @xmath4 and @xmath98 cause inflation to end when @xmath99 become large which can also be seen in bf fig . equivalently , this can be interpreted as when the classical as well as quantum back - reactions from @xmath100 and @xmath101 interactions induce large masses for @xmath3 and @xmath4 such that they roll rapidly to the global minimum , ending inflation . therefore , in the analysis below we work in the limit where @xmath96 and consider the end of inflation when @xmath102 . to calculate the evolution of curvature perturbation from eq . ( [ r - eq ] ) we need to estimate @xmath95 and @xmath66 . the derivative of @xmath103 in field space is [ theta - eq ] = = - . since @xmath104 , the first term is much larger than the second term by a factor of @xmath105 and @xmath106 . to calculate @xmath66 we observe that its equation has the same form as the background @xmath4 equation and therefore s ( n ) = _ s ( n ) , in which [ omega ] _ s = ( ) _ ^ e^i . the term @xmath107 represents the phase difference between the oscillations of @xmath4 and the entropy perturbations . this phase difference vanishes after time averaging when we find the final curvature power spectrum . combining the above expressions for @xmath95 and @xmath66 , the final curvature perturbation is integrated to [ rne ] r ( n_e)= r_0 -2 _ 0^n_f n . as the function @xmath108 grows more rapidly than the linear exponential , from eq . [ orpsi ] one obtains ( n ) ( n_c ) e^-3/2 n ( ^2/3 _ n ) . using eq . ( [ sdaf ] ) and noting that @xmath109 from eq . ( [ phif ] ) , the final curvature perturbations is obtained to be [ rn ] r = r_0 - 2 ( n_c)^2 _ s _ 0^n_f n e^-(3 - 2r)n ^2(^2/3 _ n ) n . the amplitude of quantum fluctuations of adiabatic perturbations at the moment of horizon crossing is q _ = _ , where @xmath110 represents the sasaki - mukhanov variables for the adiabatic perturbations @xcite . using the form of @xmath111 given in eq . ( [ omega ] ) , the curvature perturbation calculated from eq . ( [ rn ] ) is [ rnf ] r(n_f ) = r_0 ( 1- c ( ) _ ^ i(n_f ) ) , where @xmath94 is the initial value of adiabatic curvature perturbations at the moment of horizon crossing _ 0 = _ , and = e^(-3 + 3r ) n_c . furthermore , the integral @xmath112 has the following form [ in1 ] i(n_f ) = _ 0^n_f n e^-(3 - 2r)n ^2(^2/3 _ n ) n . ( [ rnf ] ) has some interesting features . if the second term in the big bracket in eq . ( [ rnf ] ) is larger than unity , then the induced curvature perturbations from the entropy perturbations dominate over the adiabatic curvature perturbations . furthermore , the dominant momentum dependence in the big bracket comes from the initial amplitudes of entropy perturbations and the integral @xmath112 is nearly constant for all momenta . this implies that the induced curvature perturbations from the entropy perturbations are highly blue - tilted . using the integral approximation given by eq . ( [ mixint ] ) one can calculate the integral approximately and [ in2 ] i(n_f ) e^-(3 - 2r)n_f ^2(^2/3 _ n ) . plugging this into eq . ( [ rnf ] ) , and using the asymptotic behavior of airy function of second kind , the curvature perturbations at the end of inflation is calculated to be [ rnf2 ] r(n_f ) = r_0 to get the final curvature power spectrum we need to know @xmath113 , the time of end of inflation . to determine this we proceed as follows . the exponential growth of the @xmath4 modes ( background as well as quantum fluctuations ) can violate our background solutions for @xmath3 and @xmath114 . the exponential growth of the background @xmath114 field have two important effects . first , through the interaction term @xmath115 , it can induce large mass for @xmath3 field which speeds up its rolling toward the global minimum and violate the slow - roll conditions . the second important effect is that the back - reaction from @xmath116 term induces large positive mass for @xmath4 which uplifts the tachyonic mass of @xmath114 field leading to the deceleration of this field . these two effects jointly terminate both inflation and the growth of the super - horizon curvature perturbations . to see the latter effect we start from eqs . ( [ theta - eq ] ) and ( [ r - eq ] ) where we observe that the deceleration of the @xmath4 field as well as the fast - rolling of the @xmath3 field jointly cause the termination of the super - horizon curvature perturbations as explained above . in this section we study the back - reactions of classical ( zero momentum ) mode of @xmath4 field as well as the quantum back - reactions of @xmath4 inhomogeneities produced during phase transition on the dynamics of the system . we examine which of the above two mechanisms dominate sooner to terminate inflation . we will demonstrate that the quantum mechanical back - reactions dominate completely over the classical back - reactions and hence the end - point of inflation is determined by quantum back - reactions . however , as we shall see later , the variance of the quantum fluctuations ( on all scales ) after phase transition has the same time dependence ( @xmath117-dependence ) as the classical trajectory and the difference is just in a proportionality factor . in order to demonstrate why the classical back - reactions are secondary we start the analysis with the classical back - reactions . due to smallness of background @xmath4 field before phase transition our analysis in [ background ] concentrated only on the linear level . now we add the back - reactions of the non - linear terms @xmath115 and @xmath118 on the dynamics of the system at the classical level . the correction from the interaction term @xmath115 becomes important in @xmath3 evolution , eq.([phi - c ] ) , when [ back - phi ] g^2 . on the other hand , the non - linear self - interaction term in background @xmath4 evolution , eq . ( [ psi - c ] ) , becomes important when [ back - psi ] r . however , @xmath119 as a consequence of the vacuum domination condition eq.([vd - ne ] ) . this indicates that the back - reaction of the @xmath4 field on the inflaton dynamics becomes important sooner before its self - interaction corrections affect its own dynamics . putting it another way , during hybrid inflation @xmath120 where @xmath121 and @xmath122 are the conventional slow - roll parameters . due to back - reactions of @xmath4 on inflaton mass the condition @xmath123 is met sooner before @xmath124 find the chance to become order of unity due to its initial smallness during inflation . we have numerically checked that when the condition eq . ( [ back - phi ] ) is satisfied then @xmath125 . this followed by a very short period of inflation when condition eq . ( [ back - psi ] ) is met and inflation ends . combining eq.([back - phi ] ) and eq.([psi-2 ] ) one finds the end - point of inflation to be [ c - back ] ( _ n_f^3/2 -3 n_f ) ^1/3 _ ( ) ( ) ^2 e^3 n_c . using this in eq . ( [ rnf2 ] ) , the amplitude of curvature perturbation at the end of inflation is [ rnf3 ] r_f _ 0 . consequently , the final power spectrum of curvature perturbations , @xmath126 , is calculate to be [ final - power ] p _ r = p_r_0 , where @xmath127 is the power spectrum of the adiabatic curvature perturbations . as explained previously , during inflation @xmath4 is very heavy so @xmath4 rolls to its instantaneous minimum very quickly and @xmath128 . below we demonstrate that @xmath129 such that the curvature perturbations induced from the entropy perturbations dominate completely over the adiabatic curvature perturbations . to see this note that for the @xmath4 field to be nearly zero such that its fluctuations do not contribute to the curvature perturbation around @xmath87 e - foldings before the end of inflation , that is @xmath130 at the start of inflation , one requires that 1 , 1 . using eq.([psi - substituted ] ) and eq.([phi - c ] ) one can easily find that _ i this in turn results in [ psi - phi ] . now compare the second term in the big bracket in eq . ( [ final - power ] ) to the initial curvature perturbations ( ) _ ^(3 - 3r ) ( ) ( ) ^2 . noting that @xmath131 with @xmath132 for super - horizon modes , and using the inequality eq . ( [ psi - phi ] ) and @xmath133 one obtains [ deltar ] ( ) n_e . noting that @xmath134 and @xmath135 , one concludes that @xmath136 . this indicates that , in the limit where only the classical back - reactions are considered to determine the endpoint of inflation , the curvature perturbations induced from the entropy perturbations dominate completely over the original adiabatic curvature perturbations . we shall see in next subsection that this conclusion is not stable against quantum back - reactions . here the final power spectrum of curvature perturbations , at the classical level , as a function of comoving momentum for the parameters used in * fig . [ bkd]*. are shown . the momenta exited approximately in first 6 e - folds . blue squares shows the full numerical results whereas the solid red curve shows the best fit . the best power law fit is acquired for @xmath137 with @xmath138 at @xmath139 . ] with @xmath140 , the spectral index of curvature perturbations , @xmath141 , is n_r -1 = n^0_r -1 + 3 - 3r , in which @xmath142 is the spectral index calculated from the adiabatic curvature perturbations using @xmath143 , @xmath144 , where @xmath124 and @xmath122 are the standard slow - roll parameters . using the approximation @xmath145 one has [ nr ] n_r -1 = 3 - . this indicates that the curvature perturbations receive a large blue - tilted spectrum from the entropy perturbations and @xmath146 . finally , to calculate the time of end of inflation , @xmath113 , from eq . ( [ c - back ] ) we obtain n_f ^2/3 for @xmath147 , the second term in the big bracket above can be ignored and one approximately has n_f ( n_c ) ^2/3 . it is also interesting to calculate the angle in phase space at the end of inflation , @xmath148 . using eqs . ( [ theta ] ) and ( [ c - back ] ) we obtain _ f 3 e^- r n_c ~1 . this verify our previous claim that inflation ends classically when @xmath149 . this can also be seen from * fig * [ bkt ] . as an example consider the vacuum dominated hybrid inflation with parameters @xmath150 , @xmath151 and @xmath152 . from eq . ( [ nr ] ) the spectral index is @xmath153 which is in good agreement with @xmath154 obtained from the full numerical analysis . also from * fig . * [ spectral ] one can see that there are good agreements between our analytical results and the results obtained from the full numerical analysis at the classical level . finally , @xmath155 which is in agreement with the numerical results ( see * fig . * [ stf ] , @xmath156 and @xmath157 so @xmath158 ) . our analysis so far concentrated on the classical evolution of the system . due to tachyonic instability during the phase transition quanta of @xmath4 particles inhomogeneities , @xmath159 , are produce @xcite which can back - react on the classical backgrounds as in preheating models @xcite . if the mechanism of particle creation due to tachyonic instability is very efficient , the back - reaction of the produced particles can induce large mass for inflaton field in the form of @xmath160 where @xmath161 is the expectation value of @xmath162 in the hartree approximation . this violates the slow - roll conditions and @xmath3 rapidly rolls towards the global minimum . furthermore , the large back - reactions of @xmath162 also induce a large mass for the background @xmath4 which can uplift its tachyonic mass . therefore , one has to take into account the quantum back - reactions and see which of the classical or quantum back - reactions dominate first to terminate inflation . to handle the quantum back - reactions , we use the hartree approximation and calculate the effects of all modes which become tachyonic during the phase transition . let us compute which modes become tachyonic after the phase transition . the equation of the @xmath163 fluctuations at the linear order is _ k + 3 _ k + ( - 2 r n ) _ k = 0 therefore modes which satisfy the inequality = _ become tachyonic soon after the phase transition . here we defined @xmath164 as the critical mode which leaves the horizon at the time of waterfall phase transition . we divide the tachyonic modes into two categories : first , large modes @xmath165 , corresponding to @xmath166 , which exit the hubble radius sometime before the phase transition and second , the small modes @xmath167 , corresponding to @xmath168 , which do not exit the hubble radius till time of phase transition . ( [ qem ] ) , the amplitude of both modes at the time of critical point @xmath169 is |v_k(n=0)| , and therefore for large modes one has [ delpsi - ls ] |_k^l ( n=0)| . this is in agreement with our previous result , eq.([delsf ] ) , for @xmath170 . on the other hand , for the small modes which remain sub - horizon till the time of phase transition one has [ delpsi - ss ] |_k^s ( n=0)| . with this division of the modes , and using eq . ( [ delsf ] ) , @xmath161 is calculated to be ^2 & & ( -3 n + n^3/2 ) + & & ^2 _ l + ^2 _ s , [ quantum - psi ] in which we have ignored factors of order unity . the first integral , representing @xmath171 , comes from the large modes which leave the horizon sometimes prior to phase transition whereas the second integral , representing @xmath172 , indicates the contributions from the very small scales modes which are sub - horizon till phase transition . we observe that both @xmath173 and @xmath174 have similar @xmath117-dependence which becomes important in our discussion below when we compare the classical and quantum back - reactions . however , [ stol ] ^2 _ s ~_^2 ^2 _ l . as @xmath175 the second integral in eq . ( [ quantum - psi ] ) is much bigger than the first one . this means that the cumulative contributions of the modes which becomes tachyonic but remained sub - horizon till the time of phase transition are more important in quantum back - reactions and [ q - back ] ~ ( n^3/2 -3n ) . as explained before , there are two competitive corrections which terminate inflation and the evolution of large scale curvature perturbations during the phase transition . the first one is the classical corrections from the @xmath176 interactions which induce large mass for the inflaton , violating the slow - roll condition during the phase transition . this is the effect which was studied in subsection ( [ classical ] ) , specifically eq . ( [ c - back ] ) . the second contribution is the quantum back - reaction corrections into inflaton mass via @xmath177 . we need to see which of the above two corrections dominate first . comparing eq . ( [ c - back ] ) and eq . ( [ q - back ] ) , one finds that these two terms have similar time - dependence ( @xmath117-dependence ) and [ qtoc ] _ ^2 ( ) ^2 e^3n_c . using eq.([vd - ne ] ) and eq.([psi - phi ] ) , one finds that [ q - c ] ( g^2 ^3/2 ) e^3n_c . as @xmath178 , one concludes that for any reasonable value of coupling @xmath6 the quantum back - reactions dominate completely over the classical back - reactions to terminate inflation . therefore , the end - point of inflation is determined by the quantum back - reaction effects from @xmath179 and [ eoi ] ( _ n_f^3/2 -3 n_f ) . it s now instructive to calculate @xmath113 , the time when the quantum back - reactions terminate inflation . if we follow the estimation of @xcite performed in flat backgrounds , we obtain n_f ~ . for the parameters of our numerical investigations such as in * fig [ bkd ] * one has @xmath180 . however , taking into account the background cosmological expansion , from our eq . ( [ eoi ] ) , one finds [ nf - final ] n_f ~ ( ) ^2/3 , which for the parameters of our numerical studies this gives @xmath181 . this clearly indicates that quantum back - reactions dominate sooner than the classical back - reaction which happens at @xmath158 . another important point which should be considered is that the quantum fluctuations , @xmath161 , can change the effective classical trajectory . as we demonstrated in eq . ( [ qtoc ] ) the expectation value of the quantum fluctuations dominates completely over the zero momentum " mode and they will induce new effective trajectory in the field space . the fact that @xmath161 dominates over the background classical field contributions in waterfall dynamics also indicates that using the @xmath114 `` zero momentum '' as the classical trajectory is not reliable . quantum modes with tachyonic mass become highly occupied soon after the phase transition and in some senses they become classical . this suggests that we may take @xmath182 as the effective classical field trajectory . below we justify this proposal so one can introduce an effective classical trajectory defined by @xmath183 . to justify our proposal , we start with the conventional method for the evolution of super - horizon curvature perturbations @xcite . in the study of curvature perturbations in hybrid inflation this method was pioneered in @xcite and @xcite ( see also @xcite ) . in this method , the change in the comoving curvature perturbation on super - horizon scales is given by [ r - dot - eq ] _ c = h , where @xmath184 is the pressure perturbations on comoving slices . in our formalism we relate @xmath184 to the entropy perturbations @xmath66 or @xmath162 . in the analysis here it is assumed that the @xmath4 field is frozen at the background so @xmath185 classically . our goal is to demonstrate that on super - horizon scales eq . ( [ r - dot - eq ] ) reduces to our starting equation for the evolution of curvature perturbation , eq . ( [ dotr ] ) , obtained in @xcite for the two field inflationary system , with the appropriate definition of @xmath186 and replacing @xmath187 . one can simply check that @xmath188 , where taking into account the quantum effects , one also has @xcite @xmath189 . however , as in our previous analysis , the velocity along the inflation trajectory in the field space can be well approximated by @xmath190 till end of inflation . therefore [ rdot - sim ] _ c = . as the potential is vacuum dominated we can also neglect the gravitational back - reactions on @xmath162 . neglecting the self interaction term , and noting that the comoving slice coincides with the @xmath191 surface @xcite , the contribution of @xmath4-field in the pressure becomes p_c = ^2 + ( m^2 - g^2 ^2 ) ^2 . using eq . [ delpsi - n ] for the time evolution of @xmath66 mode ( which is the same as @xmath162 mode ) one finds that the contribution of the kinetic term above is approximately equal to the potential term and [ delta - p ] p_c ( m^2 - g^2 ^2 ) ( ^2 ) _ . below we would like to find an expression for @xmath192 to calculate @xmath193 . one can show that @xcite ( ^2 ) _ ( ^2 ) _ = 2 ^3 p ^2 _ p ^2 _ - ^3(+ ) . most of the contributions to the momentum @xmath194 integration above comes from the small scales modes with the amplitude _ ( n ) = _ ( n=0 ) f(n ) , in which @xmath195 is defined via f(n ) ( n^3/2 -3 n ) , and @xmath196 as given by eq . ( [ delpsi - ss ] ) . performing the integral and setting the uv cutoff @xmath197 yields [ a1 ] ( ^2 ) _ ( ^2 ) _ &= & ( 2 ) ^3 ^3(+ ) p p^2 ^2 _ p f^4(n ) + & = & ( 2 ) ^3 ^3(+ ) p p ^2 _ p f^4(n ) + & = & ( 2 ) ^3 ^3(+ ) _ h_0 f^4(n ) . on the other hand , for the large scale quantum fluctuations we have [ a2 ] _ ^l _ ^l= ( 2 ) ^3 ^3(+ ) f(n)^2 , where eq . ( [ delpsi - ls ] ) have been use for their amplitudes at the time of waterfall . furthermore , as we showed in eq . [ q - back ] [ a3 ] ^2 = _ ^2 f(n)^2 . combining eqs . ( [ a1 ] ) , ( [ a2 ] ) and ( [ a3 ] ) we obtain ( ^2 ) _ ( ^2 ) _ ^2 _ ^l _ ^l . this equation suggests that , as long as we are interested in two - point functions at the linear perturbation theory , one can make the following identification ( ^2 ) _ _ ^l . having obtained this identification for @xmath198 we plug it into @xmath193 expression in eq . ( [ delta - p ] ) and eq . ( [ rdot - sim ] ) to obtain [ rdot - subs ] _ c = ^l_. now we can cast eq . ( [ rdot - subs ] ) in the form of our starting formula eq . ( [ dotr ] ) for @xmath199 as prescribed in@xcite . noting that @xmath200 and @xmath201 on super - horizon scales , eq . ( [ rdot - subs ] ) can be written as [ rdot - subs2 ] _ = _ s _ where similar to eq . ( [ dot - theta - eq ] ) _ = - . therefore we have justified our proposal in using @xmath202 as the effective trajectory when the background @xmath4 field is zero . compared to our starting formula eq . ( [ dotr ] ) , there is an extra factor @xmath203 in eq . ( [ rdot - subs2 ] ) which originated from the uv cutoff imposed in eq . ( [ a1 ] ) . however , this extra factor is not significant and it does not affect our analysis below . for instance , in our numerical example we have @xmath204 . the quantum back - reactions therefore have two crucial effects . first they determine the end of inflation given by eq . ( [ qtoc ] ) and eq . ( [ eoi ] ) . second , as demonstrated above , they provide the effective classical trajectory via @xmath183 . to see how the replacement @xmath183 changes our results in previous sections based on classical analysis we note that the only place in which we need the classical trajectory is in the calculation of @xmath205 in the eq . ( [ r - eq ] ) . however , @xmath4 and @xmath206 have the same time dependence and they only differ by a normalization which is given by eq . ( [ qtoc ] ) . one can simply adopt our previous results , eq . ( [ rnf2 ] ) , for the final curvature perturbations but with @xmath206 as the classical trajectory and take into account the extra modifying factor @xmath203 mentioned above . using eq . ( [ qtoc ] ) and eq . ( [ eoi ] ) in eq . ( [ rnf2 ] ) , the amplitude of curvature perturbation therefore is [ rnf4 ] r_f _ 0 , in which we considered @xmath207 , @xmath208 and neglected the @xmath209 dependence in the exponent . similar to the analysis in classical case , the induced curvature perturbations becomes [ final ] ~ ( ) ^3 . as in @xcite , eq . ( [ final ] ) shows that the induced curvature perturbations from the entropy perturbations has the power spectrum @xmath210 and is suppressed compared to the primordial curvature perturbations by the factor @xmath211 . as @xmath212 and @xmath213 one concludes that @xmath214 for any reasonable values of the coupling @xmath6 . this indicates that there would be no large scale curvature perturbations once the quantum back - reactions are turned on during the phase transition . physically this means that the cumulative quantum back - reactions of very small scale modes , modes which become tachyonic during the phase transition but remained sub - horizon during entire inflation , shuts off the background @xmath4 instability , forcing @xmath3 and @xmath4 to their global minima ending inflation quickly . the tachyonic instability of entropy perturbations are lifted and they can not produce appreciable large scale curvature perturbations during the phase transition . our results can be compared with the findings of @xcite , see also @xcite , although these works concern about the second order curvature perturbations effects . we can also compare our results for @xmath215 with the result obtained in @xcite using the @xmath216 formalism . noting that @xmath217 , we obtain @xmath218 whereas in @xcite they have @xmath219 . the agreements between these two results are good ( the two estimations of @xmath220 differ by one order of magnitude in @xmath221 orders of magnitude ) . one concern may be the overproduction of primordial black holes in this model . this question was studied in @xcite . as they showed , with @xmath47 , the standard hybrid inflation model is safe under overproduction of primordial black holes . in this paper the possibility of producing large scale curvature perturbations induced from the entropy perturbations during the waterfall phase transition in hybrid inflation are studied . we have shown that whether or not appreciable amounts of large scale curvature perturbations are produced depend crucially on the competition between classical and quantum mechanical back - reactions to terminate inflation . if one considers only the classical back - reaction effects , one obtains a significant large scale curvature perturbations which completely dominate over the initial curvature perturbations . the induced large scale curvature perturbations would be highly blue - tilted with @xmath222 as in @xcite . however , we have shown that the quantum - mechanical back - reactions of the waterfall field inhomogeneities produced during the phase transition dominate before the classical - back - reaction becomes important . in the hartree approximation we found that the quantum back - reactions shuts off the classical tachyonic instability very efficiently terminating inflation as well as curvature perturbations evolutions quickly after phase transition we have shown that the main contribution to quantum back - reactions comes from the cumulation of the very small scales inhomogeneities , modes which are tachyonic during the phase transition but remain sub - horizon during entire inflationary period . we also made the interesting observation that in standard hybrid inflation where the waterfall field rapidly freezes to @xmath185 at the background level , one can use @xmath223 as the effective classical trajectory . in summary , the quantum back - reactions have two crucial effects . first they determine the end of inflation given by eq . ( [ qtoc ] ) and eq . ( [ eoi ] ) . second , as mentioned above , they provide the effective classical trajectory via @xmath183 . although we have presented the analysis here only for the standard hybrid inflation , but we believe that this picture also holds for other models of inflation where there are sharp phase transitions at the end of inflation . this includes models of brane inflation where inflation ends abruptly due to tachyon formation once the distance between the brane and anti - brane reaches a critical value . however , it would be interesting to see what happens in models of inflation , such as in double inflation @xcite where there is a mild phase transition in fields evolution during early stages of inflation . since the difference @xmath224 in eq . ( [ final ] ) is not very large , an appreciable amount of curvature perturbations can be created even when the quantum mechanical back - reactions are taken into account . this in turn can produce features in curvature perturbations such as in models @xcite . it would be interesting to see the observational effects of the phase transitions during inflation as considered e.g. in @xcite . * acknowledgments * we would like to thank razieh emami , jinn - ouk gong , david lyth , mohammad hossein namjoo , david wands and specially m. sasaki for useful discussions and correspondences . we specially thank bruce bassett for many insightful discussions and for bringing refs . @xcite into our attention which initiated this work . a. a. a. would like to thank ipm and `` bonyad nokhbegan iran '' for partial support . here we summarize some important properties of the airy functions which are used in the main text . to the second leading order the airy functions of second kind and their derivatives have the following asymptotic expansion for large arguments ( z ) ~^-1/2 z^-1/4 e^ ( 1 + c_1 ^-1 ) + (z ) ~^-1/2 z^1/4 e^ ( 1 + d_1 ^-1 ) + in which = z ^3/2 and @xmath225 and @xmath226 . by using the above equations one can estimate the derivative of airy functions to the second leading order as (z ) ( z ) ( 1- ) . by using the above approximation for the derivative of airy function one can find ( az)^2 z ( z)^-1/2 ( az)^2 , and also z ( az)^2 z ( z)^1/2 ( az)^2 . similarly , one can show that in the leading order the following relation also holds z^n ( az)^2 z ( z)^n-1/2 ( az)^2 . also one can find the following useful relation [ mixint ] z^n e^ z ( az)^2 z ( z)^n-1/2 ( az)^2 . which was used to estimate the integral in eq . ( [ in1 ] ) . finally there is another simple but important relation for second derivative of airy functions [ sdaf ] ( z)=z ( z ) . 99 a. h. guth , `` the inflationary universe : a possible solution to the horizon and flatness problems , '' phys . d * 23 * , 347 ( 1981 ) ; + a. d. linde , `` a new inflationary universe scenario : a possible solution of the horizon , flatness , homogeneity , isotropy and primordial monopole problems , '' phys . b * 108 * , 389 ( 1982 ) ; + a. albrecht and p. j. steinhardt , `` cosmology for grand unified theories with radiatively induced symmetry breaking , '' phys . lett . * 48 * , 1220 ( 1982 ) . e. komatsu _ et al . _ , `` seven - year wilkinson microwave anisotropy probe ( wmap ) observations : cosmological interpretation , '' arxiv:1001.4538 [ astro-ph.co ] . a. d. linde , `` hybrid inflation , '' phys . d * 49 * , 748 ( 1994 ) [ arxiv : astro - ph/9307002 ] . e. j. copeland , a. r. liddle , d. h. lyth , e. d. stewart and d. wands , `` false vacuum inflation with einstein gravity , '' phys . d * 49 * , 6410 ( 1994 ) [ arxiv : astro - ph/9401011 ] . j. garcia - bellido , a. d. linde and d. wands , `` density perturbations and black hole formation in hybrid inflation , '' phys . d * 54 * , 6040 ( 1996 ) [ arxiv : astro - ph/9605094 ] . a. taruya and y. nambu , `` cosmological perturbation with two scalar fields in reheating after inflation , '' phys . b * 428 * , 37 ( 1998 ) [ arxiv : gr - qc/9709035 ] . b. a. bassett , d. i. kaiser and r. maartens , `` general relativistic preheating after inflation , '' phys . b * 455 * , 84 ( 1999 ) [ arxiv : hep - ph/9808404 ] . f. finelli and r. h. brandenberger , `` parametric amplification of gravitational fluctuations during reheating , '' phys . lett . * 82 * , 1362 ( 1999 ) [ arxiv : hep - ph/9809490 ] ; f. finelli and r. h. brandenberger , `` parametric amplification of metric fluctuations during reheating in two field models , '' phys . d * 62 * , 083502 ( 2000 ) [ arxiv : hep - ph/0003172 ] . j. p. zibin , r. h. brandenberger and d. scott , `` backreaction and the parametric resonance of cosmological fluctuations , '' phys . d * 63 * , 043511 ( 2001 ) [ arxiv : hep - ph/0007219 ] . b. a. bassett and f. viniegra , `` massless metric preheating , '' phys . rev . d * 62 * , 043507 ( 2000 ) [ arxiv : hep - ph/9909353 ] . b. a. bassett , c. gordon , r. maartens and d. i. kaiser , `` restoring the sting to metric preheating , '' phys . d * 61 * , 061302 ( 2000 ) [ arxiv : hep - ph/9909482 ] . k. jedamzik and g. sigl , `` on metric preheating , '' phys . d * 61 * , 023519 ( 2000 ) [ arxiv : hep - ph/9906287 ] . a. r. liddle , d. h. lyth , k. a. malik and d. wands , `` super - horizon perturbations and preheating , '' phys . d * 61 * , 103509 ( 2000 ) [ arxiv : hep - ph/9912473 ] . p. ivanov , `` on generation of metric perturbations during preheating , '' phys . d * 61 * , 023505 ( 2000 ) [ arxiv : astro - ph/9906415 ] . t. tanaka and b. bassett , `` application of the separate universe approach to preheating , '' arxiv : astro - ph/0302544 . s. tsujikawa and b. a. bassett , `` when can preheating affect the cmb ? , '' phys . b * 536 * , 9 ( 2002 ) [ arxiv : astro - ph/0204031 ] . s. tsujikawa , d. parkinson and b. a. bassett , `` correlation - consistency cartography of the double inflation landscape , '' phys . d * 67 * , 083516 ( 2003 ) [ arxiv : astro - ph/0210322 ] . j. r. bond , a. v. frolov , z. huang and l. kofman , `` non - gaussian spikes from chaotic billiards in inflation preheating , '' phys . lett . * 103 * , 071301 ( 2009 ) [ arxiv:0903.3407 [ astro-ph.co ] ] . k. kohri , d. h. lyth and c. a. valenzuela - toledo , `` on the generation of a non - gaussian curvature perturbation during preheating , '' jcap * 1002 * , 023 ( 2010 ) [ arxiv:0904.0793 [ hep - ph ] ] . l. p. levasseur , g. laporte and r. brandenberger , `` analytical study of mode coupling in hybrid inflation , '' arxiv:1004.1425 [ hep - th ] . d. h. lyth , `` issues concerning the waterfall of hybrid inflation , '' arxiv:1005.2461 [ astro-ph.co ] . j. fonseca , m. sasaki and d. wands , `` large - scale perturbations from the waterfall field in hybrid inflation , '' arxiv:1005.4053 [ astro-ph.co ] . l. randall , m. soljacic , a. h. guth , `` supernatural inflation : inflation from supersymmetry with no ( very ) small parameters , '' nucl . * b472 * , 377 - 408 ( 1996 ) . [ hep - ph/9512439 ] . c. gordon , d. wands , b. a. bassett and r. maartens , `` adiabatic and entropy perturbations from inflation , '' phys . d * 63 * , 023506 ( 2001 ) [ arxiv : astro - ph/0009131 ] . g. n. felder , j. garcia - bellido , p. b. greene , l. kofman , a. d. linde and i. tkachev , `` dynamics of symmetry breaking and tachyonic preheating , '' phys . * 87 * , 011601 ( 2001 ) [ arxiv : hep - ph/0012142 ] . g. n. felder , l. kofman and a. d. linde , `` tachyonic instability and dynamics of spontaneous symmetry breaking , '' phys . d * 64 * , 123517 ( 2001 ) [ arxiv : hep - th/0106179 ] . l. kofman , a. d. linde and a. a. starobinsky , `` reheating after inflation , '' phys . lett . * 73 * , 3195 ( 1994 ) [ arxiv : hep - th/9405187 ] ; + l. kofman , a. d. linde and a. a. starobinsky , `` towards the theory of reheating after inflation , '' phys . d * 56 * , 3258 ( 1997 ) [ arxiv : hep - ph/9704452 ] . j. h. traschen and r. h. brandenberger , `` particle production during out - of - equilibrium phase transitions , '' phys . d * 42 * , 2491 ( 1990 ) . y. shtanov , j. h. traschen and r. h. brandenberger , `` universe reheating after inflation , '' phys . d * 51 * , 5438 ( 1995 ) [ arxiv : hep - ph/9407247 ] . n. barnaby and j. m. cline , `` nongaussianity from tachyonic preheating in hybrid inflation , '' phys . d * 75 * , 086004 ( 2007 ) [ arxiv : astro - ph/0611750 ] ; + n. barnaby and j. m. cline , `` nongaussian and nonscale - invariant perturbations from tachyonic preheating in hybrid inflation , '' phys . rev . d * 73 * , 106012 ( 2006 ) [ arxiv : astro - ph/0601481 ] . a. mazumdar and j. rocher , `` particle physics models of inflation and curvaton scenarios , '' arxiv:1001.0993 [ hep - ph ] . k. enqvist , a. jokinen , a. mazumdar , t. multamaki and a. vaihkonen , `` non - gaussianity from preheating , '' phys . rev . lett . * 94 * , 161301 ( 2005 ) [ arxiv : astro - ph/0411394 ] ; + k. enqvist , a. jokinen , a. mazumdar , t. multamaki and a. vaihkonen , `` non - gaussianity from instant and tachyonic preheating , '' jcap * 0503 * , 010 ( 2005 ) [ arxiv : hep - ph/0501076 ] . k. enqvist and a. vaihkonen , `` non - gaussian perturbations in hybrid inflation , '' jcap * 0409 * , 006 ( 2004 ) [ arxiv : hep - ph/0405103 ] . j. silk and m. s. turner , `` double inflation , '' phys . d * 35 * , 419 ( 1987 ) ; + d. polarski and a. a. starobinsky , `` spectra of perturbations produced by double inflation with an intermediate matter dominated stage , '' nucl . b * 385 * , 623 ( 1992 ) ; + d. polarski and a. a. starobinsky , `` isocurvature perturbations in multiple inflationary models , '' phys . rev . d * 50 * , 6123 ( 1994 ) [ arxiv : astro - ph/9404061 ] ; + d. langlois , `` correlated adiabatic and isocurvature perturbations from double inflation , '' phys . d * 59 * , 123512 ( 1999 ) [ arxiv : astro - ph/9906080 ] . m. joy , v. sahni and a. a. starobinsky , `` a new universal local feature in the inflationary perturbation spectrum , '' phys . d * 77 * , 023514 ( 2008 ) [ arxiv:0711.1585 [ astro - ph ] ] . d. battefeld , t. battefeld and a. c. davis , `` staggered multi - field inflation , '' jcap * 0810 * , 032 ( 2008 ) [ arxiv:0806.1953 [ hep - th ] ] . d. battefeld , t. battefeld , h. firouzjahi and n. khosravi , `` brane annihilations during inflation , '' arxiv:1004.1417 [ hep - th ] . d. parkinson , s. tsujikawa , b. a. bassett and l. amendola , `` testing for double inflation with wmap , '' phys . d * 71 * , 063524 ( 2005 ) [ arxiv : astro - ph/0409071 ] . m. joy , a. shafieloo , v. sahni and a. a. starobinsky , `` is a step in the primordial spectral index favored by cmb data ? , '' jcap * 0906 * , 028 ( 2009 ) [ arxiv:0807.3334 [ astro - ph ] ] . j. o. gong and m. sasaki , `` waterfall field in hybrid inflation and curvature perturbation , '' arxiv:1010.3405 [ astro-ph.co ] . d. wands , k. a. malik , d. h. lyth _ et al . _ , `` a new approach to the evolution of cosmological perturbations on large scales , '' phys . rev . * d62 * , 043527 ( 2000 ) . [ astro - ph/0003278 ] .

non - relativistic reduction of the s - matrix for the quasifree electron scattering process is studied in order to understand the source of differences between non - relativistic and relativistic models . we perform an effective pauli reduction on the relativistic expression for the s - matrix in the one - photon exchange approximation . the reduction is applied to the nucleon current only ; the electrons are treated fully relativistically . an expansion of the amplitude results in a power series in the nuclear potentials . the series is found to converge rapidly only if the nuclear potentials are included in the nuclear current operator . the results can be cast in a form which reproduces the non - relativistic amplitudes in the limit that the potentials are removed from the nuclear current operator . large differences can be found between calculations which do and do not include the nuclear potentials in the different orders of the nuclear current operator . in the high missing momentum region we find that the non - relativistic calculations with potentials included in the nuclear current up to second order give results which are close to those of the fully relativistic calculation . this behavior is an indication of the importance of the medium modifications of the nuclear currents in this model , which are naturally built into the relativistic treatment of the reaction .

nucl-th9501036

electromagnetic probes provide invaluable information about nuclear structure . the quasifree process @xmath1 has been used extensively to study proton hole states and to determine single particle spectroscopic factors @xcite . this reaction is advantageous because the electromagnetic interaction is known ; the relative weakness of the reaction permits the probe to interact almost uniformly through the entire nucleus , and the first order of perturbation theory should provide an adequate description of the process . coincidence measurements of the @xmath1 reaction can provide detailed information about the single particle structure of the nucleus over a wide range of momentum transfer . the @xmath1 reaction has been widely studied both non - relativistically @xcite and relativistically @xcite , and there are some discrepancies between the results of these investigations . both analyses begin with a lagrangian which allows for the interaction of the photon with both electrons and protons . non - relativistic analyses involve the reduction of the free electron - proton interaction to a form involving two - component spinors for the nucleon . this results in a hamiltonian which is expanded in powers of @xmath2 where @xmath3 is the nucleon mass @xcite . the resulting interaction hamiltonian is sandwiched between schrdinger wave functions describing the nucleons in order to form the nuclear current . relativistic analyses are based on the feynman diagram for one - photon exchange between the projectile electron and a proton which is imbedded in the nucleus . the electrons and nucleons are all described relativistically as spin 1/2 objects via the dirac equation containing appropriate potentials @xcite . a long - standing problem in quasifree electron scattering has been that the spectroscopic factors extracted from non - relativistic analyses are smaller than expected from shell model calculations . spectroscopic factors which are found on the basis of the relativistic approach are generally larger than those found via the non - relativistic approach @xcite . several groups have attempted to understand the underlying differences between these two approaches . this mainly involved looking at the sensitivity of quasifree electron scattering calculations to different optical potentials and renormalizations of the continuum wave function @xcite . this concentration on optical potentials was largely a result of the improvement in the description of proton elastic scattering observables via dirac phenomenology over the standard non - relativistic optical model description . boffi et al . @xcite have multiplied the non - relativistic continuum wave function by a potential - dependent factor @xmath4 / \left ( e + m \right ) \right\}^{1/2}$ ] , where @xmath5 is the dirac scalar potential and @xmath6 is the vector potential . this modification essentially changes the two - component schrdinger wave function into the upper component of the dirac wave function , while no other change is made in the non - relativistic calculation . they find that extracted occupation probabilities are larger than those obtained from the unmodified non - relativistic analysis . the analysis of udias et al . @xcite replaces the non - relativistic bound state wave function with the upper component of a dirac wave function , and the non - relativistic continuum wave function is modified by factors of the same shape as the factor used by boffi et al . the continuum wave function in this case is generated from schrdinger - equivalent potentials @xcite . the nuclear current operators are obtained in the standard way by expansion to order @xmath7 . their `` non - relativistic '' calculations then involve non - relativistic nuclear current operators surrounded by the upper components of dirac wave functions . with these choices little difference is found between the relativistic and `` non - relativistic '' calculations . their conclusion is that differences in observed cross sections are not due to non - relativistic reduction , but to the choice of optical potential . jin and onley @xcite have presented a model which can take either relativistic or non - relativistic optical potentials while keeping other aspects of the calculation the same . they find that different optical potentials can change the results by as much as 14% . these results clearly demonstrate the variability due to final state interactions , however , the issue is clouded by the occasional use of upper components of dirac wave functions in a non - relativistic calculation . we believe that the essential difference between relativistic and non - relativistic approaches are not just in the changes in the optical potentials ; these are usually phenomenological and equivalent potentials can always be found . rather the essential difference is in the appearance of the nuclear potentials in the nuclear current operators when the relativistic amplitude is reduced to a non - relativistic form . such medium effects on the nuclear currents are absent in the standard non - relativistic calculations . in this paper we study the differences between the relativistic and non - relativistic approaches in calculating the amplitude for the @xmath1 reaction . we do this through an effective pauli reduction of the relativistic transition amplitude @xcite . an expansion of the amplitude in powers of @xmath8 allows us to recover a non - relativistic limit , which matches the standard non - relativistic calculations , with the difference that optical potentials used to generate the distorted waves are exactly equivalent to those used in the relativistic calculations . the main difference , as mentioned above is that the nuclear currents are potential dependent . we compare the two approaches and thus explain why they can still give different values for the extracted spectroscopic factors , even when equivalent optical potentials are used . we introduce the relativistic amplitude for quasifree electron scattering in section [ rel - amp ] . section [ pauli ] outlines the pauli reduction of the amplitude and some of its relevant features are discussed in section [ e - results ] . the non - relativistic limit is discussed in section [ nonrel ] . in section [ n - results ] we compare our non - relativistic calculations with and without nuclear potentials in the nuclear current operators , to the results of the fully relativistic calculations . our conclusions are given in section [ concl ] . we consider the one photon exchange model for the @xmath1 process @xcite , in which a photon is exchanged between the incident electron and a target proton . the struck proton is detected in coincidence with the final state electron . in this paper we are interested in the differences between the relativistic and non - relativistic treatment of the hadronic part the @xmath1 reaction . in the course of this discussion we do not include the coulomb interaction in the leptonic part of the s - matrix since this will only be important for heavy nuclei @xcite . we will therefore not discuss any nuclei heavier than @xmath9 in this work . the relativistic expression for the s - matrix describing the quasifree electron scattering process @xmath1 in the distorted wave born approximation ( dwba ) is @xcite @xmath10^{1/2 } \sum_{j_b m_b } { \left(j_f , j_b;m_f , m_b| j_i , m_i \right ) } \nonumber \\ & & \times { \left [ { \cal s}_{j_i j_f } ( j_b ) \right ] } ^{1/2 } \int d^4x \ , d^4y \ , d^4q \ , j_{e\mu}(y ) \frac{e^{-iq\cdot ( x - y ) } } { q^2 + i\epsilon } j^\mu_n(x ) , \label{f1}\end{aligned}\ ] ] where @xmath11 is the spectroscopic factor and @xmath12 and @xmath13 are electron and nuclear currents respectively . @xmath3 is the nucleon mass and @xmath14 is the energy of the outgoing proton . the electron current is given by @xmath15 where @xmath16 and @xmath17 are the initial and final dirac spinors for electrons . the electron wave functions are taken to be free dirac spinors and the integration at the electron vertex can then be done analytically . this also allows the momentum integration for the photon propagator to be done , leaving one four - dimensional integration at the nucleon vertex . the nuclear current is similarly given by @xmath18 where the nuclear current operator @xmath19 is the choice @xmath20 discussed by de forest @xcite @xmath21 @xmath22 and @xmath23 are nucleon form factors and are functions of the four - momentum squared of the exchanged photon , which couples to the nucleons . we have @xmath24 where @xmath25 and @xmath26 are the momenta of the initial and final electrons respectively . the matrix @xmath27 is formed from the dirac @xmath28-matrices in the standard way as @xcite @xmath29 the integration over @xmath30 in eq . ( [ f1 ] ) is associated with the propagator of the exchanged photon . the electrons are described by positive energy dirac spinors , and the integration over coordinates at the electron vertex , in the s - matrix of equation in eq . ( [ f1 ] ) , can be done analytically to yield a dirac @xmath31-function giving energy and momentum conservation at the vertex so @xmath32 . the resulting @xmath31-function fixes the momentum of the intermediate photon so the integration over that momentum is done trivially . the integration over the time coordinate at the nucleon vertex can then be done to yield a @xmath31-function providing overall energy conservation . the s - matrix can then be cast in the form @xmath33^{1/2 } \delta \left ( e_c + e_f - e_b - e_i \right ) \nonumber \\ & & \times \sum _ { j_b m_b } { \left ( j_f , j_b ; m_f , m_b| j_i , m_i \right ) { \left [ { \cal s}_{j_i j_f } ( j_b ) \right ] } ^{1/2 } z_{\nu_{f } \nu_{i}}^{\mu m_b } } , % \label{z_def}\end{aligned}\ ] ] where @xmath34 is a function of the initial and final spin projections , momenta , etc . specifically we have at this point @xmath35 where the @xmath36 matrix operating on the nucleon spinors is @xmath37 .\end{aligned}\ ] ] the four - vector which comes from the electron vertex is @xmath38^{1/2 } \nonumber \\ & & \times \left < 1/2 , \nu_{f } \right| \left [ 1 , \matrix { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$k$}}_{f } \cr \overline{e_{f } + m_{e } } } \right ] \gamma_{0 } \gamma^{\alpha } \left [ \matrix { 1 \cr \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$k$}}_{i } \cr \overline{e_{i } + m_{e } } } \right ] \left| 1/2 , \nu_{i } \right > .\end{aligned}\ ] ] and this depends on the energies and momenta of the initial and final electrons as well as their spin projections . useful details of the fock space calculations and the expansion of the dirac wave function in partial waves can be found in the paper by johansson and sherif @xcite . when the appropriate factors of @xmath39 and @xmath40 are included , the relativistic expression for the triple differential cross section is related to @xmath34 by @xmath41 } \nonumber \\ & & \times \sum _ { j_b m_b \mu \nu_{f } \nu_{i } } { \frac { { \cal s}_{j_i j_f } ( j_b ) } { 2j_b + 1 } } { | z_{\nu_{f } \nu_{i}}^{\mu m_b } | } ^2 , \label{cross}\end{aligned}\ ] ] where @xmath42 and @xmath43 are the spin projections of the incoming and outgoing electrons respectively , while @xmath44 and @xmath45 are the spin projections of the bound and continuum protons . the cross section for quasifree electron scattering in the plane wave impulse approximation can be written in a factorized form as the product of three parts @xcite : a kinematic factor , the cross section for the elementary process @xmath46 , which is evaluated off - shell , and finally a function of the energy and momentum of the nucleon inside the nucleus referred to as the spectral function . in the following we will discuss results of calculations of the spectral function , proton polarization and an asymmetry parameter . the spectral function is obtained from the cross section given above by dividing by a kinematic factor and the cross section for the elementary process for @xmath47 . we write @xcite : @xmath48 where @xmath14 and @xmath49 are the energy and momentum of the final state proton , and @xmath50 is the missing momentum , i.e. the momentum of the bound nucleon in the initial state . the free cross section is calculated using the nucleon current operator of eq . ( [ f4 ] ) , and is evaluated using the kinematics of the quasifree process , i.e. off - shell . note that the experimental data are divided by the elementary cross section @xmath51 of de forest @xcite , while we use the nuclear current operator which leads to his cross section @xmath20 , throughout this work . we are not concerned with detailed comparison with experimental data in this work , so we retain a consistent approach by using the same form for the current operator in the calculation of the quasifree s - matrix and the elementary process . the polarization of the final state proton is given by @xmath52 } ^{\ast } } } { \sum_{m_b \mu \nu_{f } \nu_{i } } { { \left| z_{\nu_{f } \nu_{i}}^{\mu m_b } \right| } ^{2 } } } . % \label{}\end{aligned}\ ] ] we also define an asymmetry parameter in the missing momentum which is calculated from the differential cross sections of eq . ( [ cross ] ) as @xmath53 this asymmetry is similar to the parameter defined by bianconi , boffi and kharzeev @xcite . we now perform the effective pauli reduction @xcite on the hadronic part of the amplitude only ; the electrons are treated relativistically throughout . consider the nuclear current of eq . ( [ f3 ] ) above . ( [ f4 ] ) one can write : @xmath54 \psi_{n_b}\left ( x \right ) . \label{f6}\end{aligned}\ ] ] the dirac spinors may be written in terms of their upper components as @xmath55 u , \label{f7}\end{aligned}\ ] ] where @xmath5 and @xmath6 are the scalar and vector potentials , respectively , for either the bound or final state nucleons . the energy of the nucleon is @xmath56 , and the associated momentum operator is @xmath57 . the upper component of the dirac spinor @xmath58 is related to a schrdinger - like wave function @xmath59 by @xcite @xmath60 note that the two - component wave function @xmath59 is a solution of the schrdinger equation used in the ordinary non - relativistic calculations , i.e. containing central and spin - orbit potentials , but these potentials are `` equivalent potentials '' meaning that they are functions of the vector and scalar potentials of the original dirac equation , as well as containing explicit energy dependence . for the continuum nucleon the dirac potentials result in an improved description of nucleon - nucleus elastic scattering data @xcite , while for the bound state the dirac potentials offer a slightly better description of the spin - orbit splitting than those used in earlier non - relativistic calculations @xcite . the relativistic nuclear current of eq . ( [ f6 ] ) can , with the help of eq . ( [ f8 ] ) , be written in the form @xmath61 \right . \nonumber \\ & & \hspace{18.5 mm } \times \;\ ; \gamma^o \left [ f_1 \gamma^\mu + f_2 \frac{i \kappa } { 2 m } \sigma^{\mu\nu}q_\nu \right ] \label{f9 } \\ & & \hspace{18.5 mm } \times \;\;\;\;\ ; \left . \left [ \matrix { 1 \cr \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$ p$ } } \vspace{0 mm } \cr \overline{m + e_b + s_b\left ( r \right ) - v_b\left ( r \right ) } \cr } \right ] d_b^{1/2}\left ( r \right ) \right\ } \psi_{sch , b } . \nonumber\end{aligned}\ ] ] we now perform an expansion of the object between the braces of eq . ( [ f9 ] ) . the usual representation of the dirac @xmath28-matrices is used @xcite to write the @xmath36 operator in terms of @xmath62 pauli matrices . the matrix multiplications are performed and a @xmath62 operator results . the radial function @xmath63 from eq . ( [ f8 ] ) and the factor @xmath64^{-1}$ ] coming from both the bound and continuum wave functions are then expanded in powers of @xmath8 . this procedure leads to a sum of reduced nuclear current operators for each of the contributing orders : @xmath65 \psi_{sch , b}\left ( x \right ) . \label{f10}\end{aligned}\ ] ] the reduced current operators can be written in terms of time - like and space - like components as @xmath66 , \nonumber \\ j^{0 ^ { ( 2 ) } } = & & ef_1 \left [ \frac{q_c q_b}{4 } - \frac{q_c^2 + q_b^2}{8 } + \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m + e_c } \frac{\mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m + e_b } \right ] \nonumber \\ & + & \frac{\kappa f_2}{2 m } \left [ \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$q$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m + e_b } - \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$q$ } } } { m + e_c } \right ] , \nonumber \\ { \bf j}^{(0 ) } = & & 0 , \nonumber \\ { \bf j}^{(1 ) } = & & \frac{ie\kappa f_2 } { 2 m } \ ; \mbox{\boldmath{$\sigma$ } } \times \mbox{\boldmath{$q$ } } + ef_1 \left [ \frac { \mbox{\boldmath{$\sigma$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m+e_b } + \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } } { m + e_c } \right ] , \nonumber \\ { \bf j}^{(2 ) } = & & \frac { ie \kappa f_2}{4 m } \left[q_b + q_c \right ] \mbox{\boldmath{$\sigma$ } } \times \mbox{\boldmath{$q$ } } + \frac { e f_1 q_c } { 2 } \left [ \frac { \mbox{\boldmath{$\sigma$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m+e_b } + \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } } { m+e_c } \right ] \nonumber \\ & + & \frac { e f_1 } { 2 } \left [ \frac { \mbox{\boldmath{$\sigma$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m + e_b } + \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } } { m + e_c } \right ] q_b \nonumber \\ & - & e f_1 \left [ q_b \frac{\mbox { \boldmath{$\sigma$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m + e_b } + \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } } { m + e_c } q_c \right ] \nonumber \\ & + & \frac { \kappa f_2 q_o } { 2 m } \left [ \frac { \mbox{\boldmath{$\sigma$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m + e_b } - \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } \;\ ; \mbox{\boldmath{$\sigma$ } } } { m + e_c } \right ] , \label{f11}\end{aligned}\ ] ] where we have defined @xmath67 and the labels @xmath68 and @xmath40 refer to the bound and continuum states , respectively . using the current operators from eq . ( [ f11 ] ) up to first order in @xmath8 , in the s - matrix ( [ f1 ] ) , we find that for the @xmath1 reaction the s - matrix to first order in @xmath8 reduces to @xmath69^{1/2 } \nonumber \\ & & \times \sum_{j_b m_b } { ( j_f , j_b;m_f , m_b| j_i , m_i ) } { \left [ { \cal s}_{j_i j_f } ( j_b ) \right ] } ^{1/2 } \nonumber \\ & & \times \int d^4x \ ; d^4y \ ; d^4q \ ; \frac{e^{-iq \cdot \left ( x - y \right ) } } { q^2 + i \epsilon } \ ; \psi_{sch , c}^\dagger \left ( x \right ) \nonumber \\ & & \times \left\ { j_e^0 \left ( y \right ) f_1\left ( q^2 \right ) \left [ 1 + \frac{1}{2 } ( q_c + q_b ) \right ] \right . \nonumber \\ & & \hspace{5 mm } - \mbox{\boldmath{$j$}}_e \left ( y \right ) \cdot \left [ i \kappa \mu_n f_2\left ( q^2 \right ) \frac { \mbox{\boldmath{$\sigma$ } } \times \mbox { \boldmath{$q$ } } } { 2 m } \right . \nonumber \\ & & \left . \hspace{20 mm } + f_1\left ( q^2 \right ) \left ( \frac { \mbox{\boldmath{$\sigma$ } } \ ; \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } } { m + e_b } + \frac { \mbox{\boldmath{$\sigma$ } } \cdot \mbox{\boldmath{$p$ } } \ ; \mbox{\boldmath{$\sigma$ } } } { m + e_c } \right ) \right ] \right\ } \psi_{sch , b } \left ( x \right ) . \label{f13}\end{aligned}\ ] ] note that we have written this equation in a form in which the integrations over the electron coordinate and the intermediate photon momentum have not been done . the expansion method does not depend on the plane wave approximation for the electrons , and electron distortions could be included if desired . the s - matrix to second order in @xmath8 is similarly found by including the second - order nuclear current as well . note the dependence of the nuclear current operators on the dirac vector and scalar potentials ( through the functions @xmath70 ) . this dependence appears in all orders of the reduction scheme . thus as we go to a description in terms of the schrdinger - like wave function for the nucleon , the currents undergo a medium modification affected via the nuclear potential . this point is central to the present work . we shall discuss the traditional non - relativistic limit of the amplitude in section [ nonrel ] ; but will concentrate in the following section on clarifying the role of the nuclear potentials in the convergence properties of the pauli expansion of the s - matrix . in this section we discuss the convergence of the expansion obtained above to the fully relativistic calculation . in these convergence calculations , all of the factors in the expansion are the original relativistic ones . _ this is not yet equivalent to a standard non - relativistic calculation _ ! the non - relativistic calculations are discussed below . the calculations of the relativistic s - matrix requires knowledge of the dirac wave functions for the bound and continuum states . for the bound state hartree bound state wave functions are used @xcite . the continuum wave functions for the knocked out proton are obtained using the energy and a dependent optical potential of cooper _ et al_. @xcite . we restrict our discussion to the case of parallel kinematics @xcite . in the diagrams referred to in the following discussion the curves are labelled according to their order in @xmath8 for the expansion calculations , and whether or not the dirac potentials are included in the nuclear current operators : dotted curve first order in @xmath8 without dirac potentials ; dashed curve first order with potentials ; dot - dashed curve second order without potentials ; dot - dot - dashed curve second order with potentials ; solid curve fully relativistic calculation . in doing these comparisons we are attempting to clarify the convergence of the expansion and the role of the nuclear potentials ( as they appear in the nuclear currents ) in the rate of convergence of this expansion . figure 1 shows observables as a function of missing momentum for the reaction @xmath71 , 1(a ) is the spectral function while 1(b ) is the proton polarization . the ground state of the residual nucleus , @xmath72 , is assumed to be a @xmath73 hole . the energy of the incident electron is 456 mev , and the kinetic energy of the detected proton is fixed at 90 mev with parallel kinematics . the relativistic calculations of the spectral function are fitted to the peak of the data @xcite ; the resulting spectroscopic factor is then used in all the other calculations for that particular state . ( we adopt this simple fitting procedure because our main concern here is comparison between the different calculations , rather than a judicious determination of the spectroscopic factors . ) note that the calculations with potentials included converge rapidly toward the fully relativistic results in this case , with the curve for the second order calculations being very close the relativistic results over the range of momentum transfers shown . in calculating the spectral function , the inclusion of the potential in the first order interaction terms bring the results closer to the fully relativistic calculation than the second order without potentials . it must be stressed that the inclusion of potentials in the interaction brings the results close to the fully relativistic results , while the calculations without potentials are quite far from the relativistic results and do not show a strong indication for convergence to the relativistic result . we have also done similar calculations for the same target but leaving the residual nucleus in an excited state , as well as using different targets , namely @xmath74 and @xmath9 , with the residual nucleus left in both ground and excited states . these calculations show the same behavior as the calculations shown in fig . 1.the above results are of course expected on simple mathematical grounds . the essential point however , is to shed light on the role of the appearance of the potentials in the nuclear currents . we have seen no evidence that expansions that are based on free vertices ( i.e. no nuclear potentials ) , will converge to the fully relativistic results , even if calculations are done to higher order in the inverse of the nucleon mass @xcite . this will have implications for the comparisons with the standard non - relativistic calculations , which we discuss next . the expansion of the s - matrix in powers of @xmath8 discussed above does not quite yield the amplitude used in standard non - relativistic calculations . some care must be taken at this point in the discussion to differentiate between the correct non - relativistic limit , and the standard operator used in non - relativistic calculations . there are three things that must be done in order to obtain the proper non - relativistic limit from the relativistic amplitude : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ \i ) _ the bound state wave function must be normalized to unity . _ in the expansion obtained above , the dirac bound state wave function is normalized to unity and the related schrdinger - equivalent wave function is not . in the non - relativistic calculations it is the schrdinger - equivalent wave function that must be normalized . + ii ) _ the continuum wave function must be normalized correctly_. the factors arising from the dirac field and the normalization of the dirac wave function result in a factor of @xmath75 being set equal to one to obtain the non - relativistic expression for the cross section , ( this is equivalent to multiplying the right - hand - side of eq . ( [ cross ] ) by the inverse of this factor ) . + iii ) finally , to obtain non - relativistic expressions for the nuclear current operators from the relativistic expressions of eq . ( [ f11 ] ) , _ the nucleon energies ( both continuum and bound ) are set equal to the nucleon mass , i.e. @xmath76 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is important to note that these changes still have not yielded the standard non - relativistic amplitudes because the nuclear current operators at this stage contain the dirac potentials explicitly . this is an essential difference between the relativistic and non - relativistic approaches , and the presence of these potentials can lead to large differences in the observables obtained via relativistic and non - relativistic approaches . in order to obtain the usual non - relativistic expression , the dirac potentials must be removed from the nuclear current operators . when this is done , the non - relativistic equivalent of the s - matrix of eq . ( [ f13 ] ) yields the usual first order non - relativistic transition amplitude used by many authors @xcite . when terms to second order are included in the non - relativistic s - matrix , and in the limit of no nuclear potentials , there are some differences between our expression and the usual non - relativistic second order s - matrix , which is obtained via a foldy - wouthuysen transformation of the interaction between electrons and free nucleons @xcite . fearing , poulis and scherer @xcite have compared foldy - wouthuysen and pauli reductions of a dirac hamiltonian containing a generic potential with harmonic time dependence . they found that differences do occur beyond first order in @xmath2 . detailed calculations show that these differences between the pauli and foldy - wouthuysen reductions are small when the nuclear potentials are ignored in the nuclear current operators . this seems the only consistent way to compare the operators since we use two different hamiltonians for the pauli calculations . we discuss below the effects that the presence of the potentials in the non - relativistic current operators have on calculated observables . we now discuss results of numerical calculations using the proper non - relativistic reduction presented above . in fig . we show results for the reaction on an @xmath77 target with the same kinematics as in fig . 1 . figure 2(a ) shows the spectral function while 2(b ) shows the proton polarization . the non - relativistic calculations show the same effects due to the inclusion of the nuclear potentials in the interaction operators that we saw in the corresponding convergence calculations of fig . 1 . the first and second order calculations without potentials in the nuclear currents ( dotted and dot - dashed curves respectively ) yield very similar results . this is generally true in the cases we have considered ; going from first to second order in @xmath2 does little to move the results in the direction of the relativistic calculations . when potentials are included in the nuclear currents , a large change is noticeable in going from first order to second order calculations , particularly at larger values of missing momentum . note that the non - relativistic calculations for the spectral function converge to a lower value than the simple expansion in powers of @xmath8 ( i.e. below the relativistic calculation ) . this is because the normalization of the schrdinger - equivalent bound state wave function to unity . this results in the non - relativistic expansion converging at a point which is not the relativistic one , but a factor of the square of the inverse normalization constant lower than the fully relativistic result . this amounts to a reduction of the spectral function from the relativistic result by a factor typically in the range 1.2 to 1.4 . spin observables are not affected by changes in overall normalization , so the proton polarization calculations shown in fig . 2(b ) are very similar to those shown in fig . 1(b ) , with slight differences coming from the replacement @xmath76 in the non - relativistic nuclear current operators . figure 3 emphasizes the behavior of the spectral function for the high missing momentum region of fig . 2(a ) , with the missing momentum in the range 150 mev / c to 300 mev / c . in this region the first and second order calculations without potentials lie above the relativistic calculations , while the inclusion of potentials in first and second order moves the results to lie below the relativistic results . note that the the relativistic calculations were fitted to the data in the low missing momentum region , but still do rather well for high missing momenta . similar results are obtained for @xmath74 and @xmath9 targets , whether the residual nucleus is left in the ground or excited state . when the potentials are not included in the nuclear currents the results diverge from the relativistic calculations as the magnitude of the missing momentum is increased . on the other hand , calculations which include potentials in the nuclear current operators remain close to the relativistic results over a wide range of missing momenta . note that we are only including terms to second order in the inverse mass . figure 4 shows non - relativistic calculations of the spectral function and proton polarization for the same reaction discussed in the previous figures , however in this case the energy of the incident electron is 2000 mev , and the kinetic energy of the detected proton is fixed at 400 mev . the larger energies allow for a much larger range of missing momenta than considered previously . it is important to note that the first and second order calculations of the spectral function , without potentials included in the nuclear current operators , differ from the relativistic calculations by up to an order of magnitude for large missing momenta , while inclusion of the nuclear potentials results in convergence to the fully relativistic results in the high missing momentum region . in addition we see that for low missing momenta the convergence point is lower than the relativistic ( see insert ) . the @xmath78 reaction shows behavior consistent with these observations for @xmath1 at high missing momentum @xcite . the momentum transfer in the @xmath78 reaction is generally in the range 400 mev / c to 600 mev / c so these two reactions can both probe this part of the single particle bound state wave function . proton polarization is shown in fig . 4(b ) . in the region of large missing momentum there are large differences between the polarization calculated with and without nuclear potentials in the current operator . the polarization calculated with first and second order currents containing nuclear potentials yields results close in magnitude and shape to the results of the fully relativistic calculations . note in particular , that in the region of the minimum and maximum in the relativistic calculations close to @xmath79 and @xmath80 respectively , the calculations without potentials included do not reproduce the shape of the relativistic calculations at all . the potentials must be included in the nuclear current operators in order to get close to the relativistic results . in particular a measurement of the proton polarization near @xmath79 provides a clear opportunity to differentiate between relativistic and non - relativistic models . calculation of the asymmetry parameter of eq . ( [ f15 ] ) , for small values of angular momentum @xmath81 of the bound nucleon ( @xmath82 ) , yields similar results for all the calculations whether fully relativistic ; or non - relativistic first or second order , with or without potentials in the nuclear current operator . when the angular momentum of the bound nucleon is increased , the differences between these calculations of the asymmetry become larger , as is evident in fig . the asymmetry is calculated for a @xmath9 target , with the residual state in @xmath83 assumed to be a @xmath84 proton hole . the incident electron has an energy of 461 mev , and the kinetic energy of the detected proton is fixed at 100 mev . in this case the differences are particularly apparent for missing momenta in the neighborhood of 20 mev / c . in order to clarify the differences arising from relativistic and non - relativistic descriptions of quasifree electron scattering @xcite , we have discussed an expansion of the s - matrix for the reaction @xmath1 in powers of @xmath8 through the effective pauli scheme . the resulting s - matrix depends on schrdinger - like wave functions for the bound and continuum nucleons , and nuclear current operators which contain the strong dirac potentials at the different orders . when the dirac potentials are included in the nuclear current operators , the series essentially converges to the fully relativistic results at second order for light- to medium - mass nuclei we have considered . this indicates the importance of the role played by the nuclear potentials in the modification of the currents . when the potentials are not included in the nuclear currents , the calculations can be far from the relativistic results particularly for larger missing momenta . these points were further studied in setting up a comparison between relativistic and non - relativistic calculations . a proper non - relativistic calculation is obtained through several steps : normalization of the bound schrdinger - like wave function to unity , proper normalization of the continuum schrdinger - like wave function , and in the nuclear current operators the energy of the nucleons is set equal to the nucleon mass ( i.e. take the limit @xmath76 ) . an additional step of removing the dirac potentials from the resulting nuclear current operators yields the standard non - relativistic amplitude . this results in a consistent and fair comparison between the relativistic and non - relativistic calculations . the potentials used for the bound and continuum protons yield both the relativistic and non - relativistic wave functions , with normalizations handled appropriately . the non - relativistic calculations we have shown for first and second order nuclear current operators without potentials give the same results that a standard non - relativistic calculation would give if provided with the schrdinger equivalent wave functions derived from the dirac equation . inclusion of the nuclear potentials in the non - relativistic nuclear current operator results in a large change in the calculated observables . in particular , calculations of the spectral function and final proton polarization using second order nuclear current operators which include the dirac potentials , can reproduce the magnitude and shape of the fully relativistic calculations . this is true even at large missing momenta where the non - relativistic calculations without potentials in the nuclear current operators yield very different results than the fully relativistic calculations . the polarization of the final proton is particularly sensitive to differences in the calculations , and measurements of this observable at large missing momenta could assist in the choice between the relativistic and non - relativistic approaches . we have also calculated the asymmetry defined in the text for the different orders , with and without potentials , and found that in cases in which the angular momentum of the bound nucleon is less than 2 , there are no noticeable differences between these calculations and the full relativistic . when the orbital angular momentum of the bound nucleon is greater than 2 differences between the resulting asymmetry in missing momentum of these calculations will appear . this observable thus will be useful in differentiating between relativistic and non - relativistic models only for nuclear states with large orbital angular momentum . other groups have examined the sensitivity of the models to changes in the optical potentials and modifications of the wave functions @xcite , and have found sensitivities at the level of 15% . however , the essential differences between the relativistic and non - relativistic approaches do not lie in modifications of the wave functions . the essential difference comes from the appearance of the nuclear potentials in the nuclear current operators ; a result of the reduction of the relativistic amplitude . we emphasize that _ these nuclear medium effects , characteristic of the present model , will not appear through a non - relativistic impulse description of the process_. they are , however , inherent in the relativistic description . we would like to thank j.m . udias and l. lapiks for useful correspondence regarding quasifree electron scattering . 99 s. frullani and j. mougey , _ advances in nuclear physics _ , edited by j.w . negele and e. vogt , * 14 * ( 1984 ) 1 . s. boffi , c. giusti and f.d . pacati , nucl . phys . * a336 * ( 1980 ) 416 ; nucl . phys . * a336 * ( 1980 ) 427 . mcvoy and l. van hove , phys . * 125 * ( 1962 ) 1034 . mcdermott , phys . * 65 * ( 1990 ) 1991 . y. jin , d.s . onley and l.e . wright , phys . c * 45 * ( 1992 ) 1311 . udias , p. sarriguren , e. moya de guerra , e. garrido and j.a . caballero , phys . c * 48 * ( 1993 ) 2731 . s. boffi , c. giusti , f.d . pacati and f. cannata , il nuovo cimento * 98a * ( 1987 ) 291 . udias , p. sarriguren , e. moya de guerra , e. garrido and j.a . caballero , nikhef - k preprint numbers nikhef-93-p12 and fniem-13 . yanhe jin and d.s . onley , phys . c * 50 * ( 1994 ) 377 . sherif , r.i . sawafta and e.d . cooper , nucl . phys.*a449 * ( 1986 ) 709 . m. hedayati - poor and h.s . sherif , phys . lett . * b326 * ( 1994 ) 9 . c. giusti and f.d . pacati , nucl . phys . * a473 * ( 1987 ) 717 . t. de forest jr . , nucl . phys . * a392 * ( 1983 ) 232 . bjorken and s.d . drell , _ relativistic quantum mechanics _ , mcgraw - hill book company ( 1964 ) . j.i . johansson and h.s . sherif , nucl . phys . * a575 * ( 1994 ) 477 . clark , s. hama , r.l . mercer , aip conf . * 97 * , ed : h.o . mayer ( 1982 ) 260 ; j. raynal , aust . * 43 * ( 1990 ) 9 ; + g.q . li , j. phys . * g19 * ( 1993 ) 1841 . e.d . cooper , s. hama , b.c . clark and r.l . mercer , phys . c * 47 * ( 1993 ) 297 . c.g . horowitz and b.d . serot , nucl . phys . * a368 * ( 1986 ) 503 . fearing , g.i . poulis and s. scherer , nucl . a570 * , ( 1994 ) 657 . a. bianconi , s. boffi and d.e . kharzeev , phys . c * 49 * ( 1994 ) 1243 . sherif , r.i . sawafta and e.d . cooper , nucl . a449 * ( 1986 ) 709 . l. lapiks , nucl . phys . * a553 * ( 1993 ) 297 . 1 . observables for the reaction @xmath71 where @xmath72 is in the @xmath73 state . the energy of the incident electron is 456 mev , and the kinetic energy of the detected proton is fixed at 90 mev with parallel kinematics . hartree bound state wave functions are used @xcite and the proton optical potentials are from @xcite . the data are from reference @xcite . ( a ) spectral function and ( b ) proton polarization . curves labelled according to their order in @xmath8 and whether or not the dirac potentials are included in the nuclear current operators : dotted curve first order in @xmath8 without dirac potentials ; dashed curve first order with potentials ; dot - dashed curve second order without potentials ; dot - dot - dashed curve second order with potentials ; solid curve fully relativistic calculation . observables for the reaction @xmath71 where @xmath72 is in a @xmath73 proton hole state . the kinematics are those of fig . curves labelled according to their order in @xmath2 and whether or not the dirac potentials are included in the nuclear current operators : dotted curve non - relativistic calculations , first order in @xmath8 without dirac potentials ; dashed curve first order with potentials ; dot - dashed curve second order without potentials ; dot - dot - dashed curve second order with potentials ; solid curve fully relativistic calculation . potentials and data from the sources of fig . 1 . fig . observables for @xmath85 where @xmath72 is in the @xmath73 state . the energy of the incident electron is 2000 mev , and the kinetic energy of the detected proton is fixed at 400 mev with parallel kinematics . curves labelled as in fig . potentials from the source of fig . 1 . asymmetry in missing momentum for @xmath86 where @xmath83 is in the @xmath84 state . the energy of the incident electron is 461 mev , and the kinetic energy of the detected proton is fixed at 100 mev with parallel kinematics . curves labelled as in fig . potentials from the source of fig . 1 .

measurements of diffractive phenomena observed at hera and the tevatron are reviewed . a short introduction to the theoretical background is presented where colour singlet exchange reactions are discussed and the diffractive contribution and its interpretation via pomeron exchange outlined . the review focuses on the current experimental directions at hera and discusses exclusive production of vector mesons , the dissociation of real photons and the deep inelastic structure of diffraction . complementary information obtained from hadronic final states and jet structures is also discussed . the experimental signatures for diffractive jet and production observed at the tevatron are described and the rates compared with those from the hera experiments . _ talk presented at the 25th slac summer institute , _ _ slac , august 1997 . _ 17.0 cm -1.0 in -42pt plus 2 mm minus 2 mm epsf glas - ppe/971319 december 1997

hep-ex9712027

_ prologue : _ in 1992 , h. fritzsch summarised the status of qcd 20 years on " @xcite with the words a very large amount of data in strong - interaction physics is described by the pomeron singularity ... the fact that the physics of the pomeron is very simple needs to be explained in equally simple terms in qcd ... using hera the experimentalists will be able to study the region of very low @xmath2 in deep inelastic scattering ... furthermore , both in @xmath3-scattering at high energies and at hera the structure functions of the pomeron can be studied in more detail . " now , 25 years on from the first developments of qcd , at the 25th slac summer institute we can discuss a series of diffractive measurements that have been made in the intervening period at hera and the tevatron . _ hera kinematics : _ the diffractive processes studied at hera are of the form : @xmath4 where the photon dissociates into the system @xmath5 and the outgoing proton , @xmath6 , remains intact , corresponding to single dissociation , as illustrated in fig . the measurements are made as a function of the photon virtuality , @xmath7 the centre - of - mass energy of the virtual - photon proton system , @xmath8 , the invariant mass of the dissociated system , @xmath5 , denoted by @xmath9 and the four - momentum transfer at the proton vertex , given by @xmath10 . = 8.cm _ signatures of diffraction : _ the processes studied build upon the basic elastic scattering process @xmath11 which is constrained through the measurement of the corresponding total cross - section @xmath12 anything , via the optical theorem . the diffractive processes measured are then @xmath13 @xmath14 where the systems @xmath5 and @xmath15 have a limited mass compared to the overall energy available , @xmath16 . the process is thought of as being mediated by the exchange of an object with vacuum quantum numbers ( in particular , no colour is exchanged in the process ) . a signature for diffraction is via the momentum fraction carried by the exchanged colour singlet state . when this fraction is less than 1% of the momentum of , say , particle @xmath17 ( in the infinite momentum frame of @xmath17 ) we can interpret this as being largely due to pomeron exchange . the kinematics of producing two low - mass outgoing states ( @xmath5 and @xmath17 or @xmath5 and @xmath15 ) with a small momentum fraction exchanged between them therefore leads to a rapidity gap . the high energy available at hera provides a large rapidity span , @xmath18 , of @xmath19 12 units ( here , @xmath20 where @xmath21 and @xmath22 for a typical @xmath23 gev ) . a colour singlet exchange of reggeons ( dominating at lower @xmath16 ) and the pomeron ( dominating at higher @xmath16 ) can be used to interpret the data on single dissociation and double dissociation reactions . fluctuations from processes where colour is exchanged may also generate low - mass states . however , these will be exponentially suppressed : the ( poisson ) probability of @xmath24 producing a given particle in the rapidity gap @xmath25 when the two systems are colour connected is @xmath26 , where @xmath27 is the mean particle density for a given @xmath25 interval . this exponential fall - off reflects the plateau in the corresponding ( non - diffractive ) multiplicity distribution as a function of @xmath28 which increases relatively slowly ( logarithmically ) with increasing @xmath16 @xcite . in fig . [ tsps ] , the @xmath29 distributions of @xmath30 elastic scattering data are illustrated as a function of the longitudinal momentum ( @xmath31 ) of the outgoing proton transformed to a fixed - target rest frame . the patterns are similar to the diffraction patterns observed when light is scattered from an aperture and exhibit an exponential fall - off for values of @xmath29 below @xmath32 1 gev@xmath33 . this characteristic fall - off increases with increasing energy , a property known as shrinkage , and differs for different incident and outgoing systems . in order to characterise the @xmath29-dependence , a fit to the diffractive peak is performed . in the most straightforward approach , a single exponential fit to the @xmath29 distribution , @xmath34 for @xmath35gev@xmath33 is adopted . physically , the slope of the @xmath29 dependence in diffractive interactions tells us about the effective radius of that interaction , @xmath36 : if d@xmath37 , then @xmath38 @xmath39 . = 10.cm _ regge trajectories : _ the subject of diffraction is far from new : diffractive processes have been measured and studied for more than thirty years @xcite . their relationship to the corresponding total cross - sections at high energies has been successfully interpreted via the optical theorem and regge theory . at lower energies the colour singlet exchange of virtual mesons , called reggeons , contribute to the fall of the cross - section with increasing energy . at higher energies , the introduction of an additional trajectory , known as the pomeron trajectory , with a characteristic @xmath40 and @xmath29 dependence is necessary @xcite . the energy behaviour of the total cross - sections can then be described by the sum of two power - law dependences on the centre - of - mass energy , @xmath41 @xmath42 where @xmath16 is measured in gev , @xmath43 and @xmath28 is defined to be positive such that @xmath44 . here , @xmath45 and @xmath46 are the pomeron and reggeon intercepts ( i.e. the values of the parameters at @xmath29 = 0 gev@xmath33 ) , respectively . a wide range of total cross - section data are used to determine the parameters @xmath47 and @xmath28 . the fall - off at low energy due to reggeon exchange constrains the value of @xmath48 . the slow rise of hadron - hadron total cross - sections with increasing energy indicates that the value of @xmath49 i.e. the total cross - sections increase as @xmath50 , although the @xmath3 data from cdf at two @xmath51 values indicate @xmath52 @xcite . recent fits using all @xmath30 and @xmath53 data are consistent with a value of @xmath54 which will be used here to characterise this behaviour @xcite . = 8.cm in a regge analysis , the diffractive data are interpreted via exchanges with spin @xmath55 and @xmath56 , with @xmath57 . at lower energies , these correspond to reggeon ( i.e. approximately degenerate @xmath58 , @xmath59 , @xmath60 and @xmath61 ) exchanges . at the highest energies , where the pomeron contribution dominates , the optical theorem relates the total cross - sections to the elastic , and hence diffractive , scattering amplitude at the same @xmath40 . in fig . [ regge ] , the trajectories , @xmath62 , are shown as a function of @xmath9 . the diffractive data probe the region of negative @xmath29 . given the dominance of the pomeron contribution at large @xmath16 and an approximately exponential behaviour of the @xmath63 distribution with slope @xmath64 , whose mean @xmath65 value is given by @xmath66 at the mean @xmath67 of a given data sample , the diffractive cross - section rise is moderated from @xmath68 to @xmath69 where @xmath70 and @xmath71gev@xmath72 reflects the shrinkage of the diffractive peak as a function of @xmath29 with increasing @xmath40 . the observed shrinkage of the diffractive peak therefore corresponds to a relative reduction of the diffractive cross - section with increasing energy . this value may be compared with the corresponding parameter @xmath73 gev@xmath72 for reggeon exchanges . _ maps of the pomeron : _ whilst these regge - based models gave a unified description of pre - hera diffractive data , this approach is not fundamentally linked to the underlying theory of qcd . it was anticipated that at hera and tevatron energies if any of the scales @xmath74 , @xmath9 or @xmath29 become larger than the qcd scale @xmath75 , then it may be possible to apply perturbative qcd ( pqcd ) techniques , which predict changes to this power law behaviour . qualitatively , the @xmath16 dependence could be ascribed to the rise of the gluon density with decreasing @xmath2 determined from the large scaling violations of @xmath76 , where @xmath2 is the bjorken scaling variable , @xmath77 . qcd factorisation into a long - timescale and short - timescale process , where this timescale is characterised by @xmath78 or @xmath79 or @xmath80 , leads to the following approaches . * for exclusive final states , e.g. vector meson production , with a hard scale the approach is very simple @xmath81 i.e. two - gluon exchange where @xmath82 is the square of the gluon density of the proton at a representative value of @xmath2 and @xmath83 represents the hard scattering process . the rise of @xmath84 with decreasing @xmath2 , which constrains the gluon density , corresponds to an increase in the effective value of @xmath47 . this brings us from the regime of dominance of the slowly - rising soft " pomeron to the newly emergent hard " behaviour and the question of how a transition may occur between the two . the qcd expectation is that the cross - sections should approximately scale as a function of @xmath29 , corresponding to a weak dependence of @xmath47 as a function of @xmath29 and therefore a decrease of @xmath85 for the perturbative pomeron . * for inclusive diffraction with a hard scale @xmath86 i.e. leading - gluon exchange where @xmath87 is the gluon density of the proton , @xmath83 represents the hard scattering process and @xmath88 represents the hadronisation process . here , the final state with a leading proton is seen as a particular hadronisation process @xcite . for processes involving one incoming hadron , the above approaches can be tested and compared with experimental data . * finally , one can break the process down further and invoke regge factorisation where a flux of pomerons , @xmath89 lead to partons from the pomeron , @xmath90 , which interact and hadronise @xmath91 for processes involving two incoming hadrons , this approach has been generalised to @xmath92 where @xmath93 represents the partons from the hadron which has not diffractively dissociated . this regge factorisation approach can therefore be experimentally tested when diffractive data from hera and the tevatron are compared . precisely where the regge - based approach breaks down or where pqcd may be applicable is open to experimental question . in addition , once we observe hard " diffractive phenomena , we can ask whether the pqcd techniques applied to inclusive processes also apply to these exclusive colour singlet exchange reactions . the emphasis is therefore on the internal ( in)consistency of a wide range of measurements of diffractive and total cross - sections . as an experimentalist navigating around the various theoretical concepts of the pomeron , it is sometimes difficult to see which direction to take and what transitions occur where ( fig . [ map](a ) ) . however , from an experimental perspective , the directions are clear , even if the map is not yet complete ( fig . [ map](b ) ) . _ outline : _ the hera collider allows us to observe a broad range of diffractive phenomena at high @xmath40 . what is new is that we have the ability to observe the variation of these cross - sections at specific points on the @xmath9 scale , from the @xmath94 up to the @xmath95 system as discussed in section 2 . similarly , the production cross - section can be explored as a function of @xmath74 , using a virtual photon probe . the observation of a significant fraction of events ( @xmath96 ) with a large rapidity gap between the outgoing proton and the rest of the final state in deep inelastic scattering ( dis ) has led to measurements of the internal structure of the pomeron . in addition , the leading proton spectrometer data , where the diffracted proton is directly measured , enable the @xmath29 distribution as well as the structure function to be determined simultaneously . these results are discussed in section 3 . studies of the hadronic final state in events with a large rapidity gap , including transverse energy flows , event shape distributions and high-@xmath97 jets , have been used to provide complementary information on this structure . also , the observation of rapidity gaps between jets , corresponding to large-@xmath29 diffraction , are presented in section 4 . many of these hadronic final state investigations were initiated at @xmath3 colliders . in section 5 , the latest results from the tevatron on diffractive dijet and @xmath0 production , rapidity gaps between jets and first observations of hard double - pomeron exchange are presented . a comparison of hard diffractive event rates at hera and the tevatron is given and the interpretation of the observed differences is discussed . the experimental signals are the exclusive production of the vector mesons in the following decay modes @xmath98 first results on @xmath99 and higher vector mesons ( @xmath100 , @xmath101 and @xmath102 ) have also been presented . a recent review of the hera data can be found in ref @xcite . the relevant components of the h1 and zeus detectors are the inner tracking chambers which measure the momentum of the decay products ; the calorimeters which allow identification of the scattered electron and are used in the triggering of photoproduced vector mesons ; and the outer muon chambers used to identify muonic decays of the @xmath103 and @xmath95 . the clean topology of these events results in typical uncertainties on the measured quantities ( @xmath29 , @xmath9 , @xmath40 and @xmath74 ) , reconstructed in the tracking chambers , of order 5% . containment within the tracking chambers corresponds to a @xmath16 interval in the range @xmath104 gev . photoproduction processes have been extensively studied in fixed - target experiments , providing a large range in @xmath16 over which to study the cross - sections . the key features are the weak dependence of the cross - section on @xmath16 , an exponential dependence on @xmath29 with a @xmath64 slope which shrinks with increasing @xmath16 and the retention of the helicity of the photon by the vector meson . in fig . [ trho ] , the zeus results for exclusive @xmath94 production as a function of @xmath29 are shown . an exponential fit to the zeus data in three @xmath16 intervals yields @xmath64-slopes which are fitted to the form @xmath105 expected from regge theory . the fitted value of @xmath106 gev@xmath72 is consistent with a shrinkage of the @xmath29 slope with increasing @xmath40 expected from soft diffractive processes . = 10.cm the interaction radius , @xmath36 , can be approximately related to the radii of the interacting proton and vector meson , @xmath107 . the variation of these @xmath64 values is shown in fig . [ bmq2](a ) as a function of vector meson mass @xmath108 . in fig . [ bmq2](b ) , these slopes are presented as a function of increasing virtuality of the photon for @xmath94 production data . in each case , the range of measured @xmath64-slopes varies from around 10 gev@xmath72 ( @xmath109 fm ) at low @xmath108 or @xmath74 to 4 gev@xmath72 ( @xmath110 fm ) at the highest @xmath108 or @xmath74 so far measured . given @xmath111 fm , this variation in @xmath64-slopes corresponds to a significant decrease in the effective radius of the interacting vector meson from @xmath112 fm to @xmath113 fm as @xmath108 ( at fixed @xmath114 gev@xmath33 ) or @xmath74 ( at fixed @xmath115 ) increase . = 12.cm integrating over the measured @xmath29 dependence , the @xmath16 dependence of the results on exclusive vector meson photoproduction cross - sections are shown in fig . there is generally good agreement between the experiments on the measured cross - sections . the @xmath116 total cross - section is also shown in fig . [ wvm ] , rising with increasing energy as in hadron - hadron collisions and consistent with a value of @xmath49 i.e. the total cross - section increases as @xmath50 . given the dominance of the pomeron trajectory at high @xmath16 and a @xmath63 distribution whose mean value @xmath65 is given by @xmath66 , the diffractive cross - section rise is moderated from @xmath117 to @xmath118 here @xmath119 characterises the effective energy dependence after integration over @xmath29 for @xmath120 gev@xmath72 ( which is appropriate for @xmath94 exchange as observed in fig . [ trho ] ) . the observed shrinkage of the diffractive peak therefore corresponds to a relative reduction of the diffractive cross - section with increasing energy . such a dependence describes the general increase of the @xmath94 , @xmath59 and @xmath121 vector meson cross - sections with increasing @xmath16 . however , the rise of the @xmath103 cross - section is clearly not described by such a @xmath16 dependence , the increase being described by an effective @xmath122 dependence . whilst these effective powers are for illustrative purposes only , it is clear that in exclusive @xmath103 production a new phenomenon is occurring . for example , fits to the zeus data yield results for @xmath123 which are inconsistent with soft pomeron exchange @xmath124 the @xmath103 ( charm ) mass scale , @xmath125 , is larger than the qcd scale , @xmath75 , and it is therefore possible to apply pqcd techniques . the theoretical analysis predicts that the rise of the cross - section is proportional to the square of the gluon density at small-@xmath2 ( a pair of gluons with no net colour is viewed as the perturbative pomeron ) @xmath126 ^ 2 \simeq w^{4\bar{\epsilon } } \simeq w^{0.8}\ ] ] where @xmath127 gev@xmath33 for the photoproduction data ) is chosen as the scale in the ryskin model @xcite . this approach enables discrimination among recent parametrisations of the proton structure function as shown in fig . [ jpsiwq2](a ) . here , the measured cross - sections are compared to the ryskin model for various choices of parton densities which describe recent @xmath84 data . the approach is therefore very promising as an independent method to determine the gluon distribution at low-@xmath2 from the hera data . currently , however , there are model uncertainties and the calculations are only possible in leading order . the normalisation is therefore uncertain by up to a factor of two . we also know from measurements of the dis @xmath128 total cross - section that application of formula ( 1 ) results in a value of @xmath47 which increases with increasing @xmath74 , with @xmath129 0.2 to 0.25 at @xmath130 gev@xmath33 @xcite . the fact that the corresponding relative rise of @xmath84 with decreasing @xmath2 can be described by pqcd evolution @xcite points towards a predicted function @xmath131 for @xmath132 . the current data indicate that this transition occurs for @xmath133 gev@xmath33 @xcite . @xmath103 electroproduction results are also available ; the @xmath134 dependence of the data is shown in fig . [ jpsiwq2](b ) . the cross - sections are fitted to @xmath135 where the fitted line corresponds to @xmath136 for the combined h1 and zeus data . this compares to the prediction of the vector dominance model ( vdm ) , applicable to soft photoproduction processes , where @xmath137 = 1 ( shown as the dashed line ) and the ryskin model where @xmath138 . also shown are the lower-@xmath16 emc data where @xmath139 . the @xmath103 electroproduction cross - section is of the same order as the @xmath94 data . this is in marked contrast to the significantly lower photoproduction cross - section for the @xmath103 , even at hera energies , also shown in fig . further results in this area will allow tests of the underlying dynamics for both transverse and longitudinally polarised photons coupling to light and heavy quarks in the pqcd calculations . = 14.cm = 8.cm one contribution to the dis @xmath128 total cross - section is the electroproduction of low mass vector mesons , here typified by the @xmath94 data . the decay angle distributions of the pions in the @xmath94 rest frame with respect to the virtual photon proton axis from the e665 fixed - target experiment ( @xmath140 gev ) are shown in fig . [ q2ang ] @xcite . the measurements of this helicity angle of the vector meson decay determines @xmath141 for the ( virtual ) photon , assuming @xmath142-channel helicity conservation , i.e. that the vector meson preserves the helicity of the photon . the decay angular distribution can be written as @xmath143 , $ ] where the density matrix element @xmath144 represents the probability that the @xmath94 was produced longitudinally polarised by either transversely or longitudinally polarised virtual photons . the value of @xmath145 is then obtained from @xmath146 , where @xmath147 is the fractional energy loss of the electron in the proton rest frame and @xmath148 . the kinematic factor @xmath149 is typically close to unity . this variation of @xmath145 with @xmath74 is summarised in fig . [ rplot ] . the photoproduction measurements for the @xmath94 ( not shown ) are consistent with the interaction of dominantly transversely polarised photons and hence @xmath150 . the electroproduction data are consistent with a universal dependence on @xmath74 independent of @xmath16 and show a transition from predominantly transverse to predominantly longitudinal photons with increasing @xmath74 . this increase of @xmath151 is due to an increased flux of longitudinal photons , @xmath152 . at higher @xmath74 values , the cross - section due to longitudinal exchange is determined in leading - log pqcd @xcite where the underlying interaction of the virtual photon with the constituent quarks of the @xmath94 is calculated . as noted previously ( see fig . [ bmq2](b ) ) , the measured @xmath64-slope decreases by about a factor of two from the photoproduction case to values comparable to that in the photoproduced @xmath103 case . the basic interaction is probing smaller distances , which allowed a first comparison of the observed cross - section with the predictions of pqcd @xcite . the @xmath16 dependence of the ( virtual-)photon proton @xmath94 cross - sections for finite values of @xmath74 are shown in fig . [ rhoq2](a ) , compared to the corresponding photoproduction cross - sections ( the @xmath121 data , not shown , exhibit similar trends ) . there is a significant discrepancy between the zeus and h1 measured cross - sections at @xmath153 gev@xmath33 as well as a smaller discrepancy between the e665 and nmc measurements at @xmath154 gev@xmath33 . this is illustrated by comparison with the @xmath122 ( dashed ) and @xmath155 dotted lines for @xmath154 gev@xmath33 and @xmath156 gev@xmath33 . at each @xmath74 value , a simple dependence can not account for all the data . one of the key problems in obtaining accurate measurements of these exclusive cross - sections and the @xmath29 slopes is the uncertainty of the double dissociation component , where the proton has also dissociated into a low mass nucleon system @xcite . at hera , the forward calorimeters will see the dissociation products of the proton if the invariant mass of the nucleon system , @xmath157 , is above approximately 4 gev . a significant fraction of double dissociation events produce a limited mass system which is typically not detected . one expects that the dissociated mass spectrum will fall as @xmath158 and integrating over the @xmath29-dependence cdf obtains @xmath47 = 0.100 @xmath159 0.015 at @xmath160 gev @xcite . precisely how the proton dissociates and to what extent the proton can be regarded as dissociating independently of the photon system is not @xmath161 known . currently , this uncertainty is reflected in the cross - sections by allowing the value of @xmath123 to vary from around 0.0 to 0.5 , a choice which covers possible variations of @xmath123 as a function of @xmath157 and @xmath16 . combining all uncertainties , the overall systematic errors on the various cross - sections are typically @xmath162 for both the photoproduction and electroproduction measurements . the estimation of the double dissociation contribution has , however , historically been one of the most significant experimental problems with these measurements . whether this is the source of the h1 and zeus discrepancy is not yet known . the combined @xmath16 dependence of the @xmath94 electroproduction data are , therefore , currently inconclusive . however , taking the zeus data alone , shown in fig . [ rhoq2](b ) there are indications of a transition from the soft to the hard intercept with @xmath123 varying from @xmath163 to @xmath164 as indicated by the fitted lines in fig . [ rhoq2 ] . these data are therefore consistent with a @xmath155 ( @xmath123 = 0.05 ) dependence at the lowest @xmath74 values and the @xmath122 ( @xmath123 = 0.2 ) dependence at the highest @xmath74 values . an important point to emphasise here is that the relative production of @xmath121 to @xmath94 mesons approaches the quark model prediction of 2/9 at large @xmath16 as a function of @xmath74 . similar observations have been made on the @xmath29 dependence of this ratio for photoproduction data , as shown in the upper plot of fig . [ tratio ] . here , the ratio of the @xmath165 cross - sections approaches the su(4 ) flavour prediction of @xmath166 ^ 2 : [ \frac{1}{\sqrt{2}}(u\bar{u } + d\bar{d})]^2 : [ s\bar{s}]^2 : [ c\bar{c}]^2 = 9:1:2:8 $ ] . the restoration of this symmetry indicates that the photon is interacting via quarks , rather than as a vector meson with its own internal structure . this therefore indicates the relevance of a gluonic interpretation of the pomeron and the applicability of pqcd to these cross - sections . similarly , the relative production of @xmath103 to @xmath94 mesons is shown with the asymptotic prediction of 8/9 from the quark model in the lower plot of fig . [ tratio ] . in this case , it is evident that threshold effects for the heavy charm quark are still significant in the measured @xmath29 range , however the ratio climbs by almost two orders of magnitude from @xmath167 to @xmath168 gev@xmath33 . exclusive @xmath95 photoproduction has also been observed @xcite . the @xmath32 20 events observed in the @xmath169 channel corresponds to a cross - section for predominantly ( 1s ) production , as well as higher @xmath95 states , of @xmath170 nb for @xmath171 where @xmath172 4 gev and @xmath173 gev , @xmath174 gev@xmath33 . this is about 1% of the @xmath103 cross - section , as shown in fig . [ upsm ] , emphasing the importance of the mass of the heavy quark in the production of exclusive vector mesons . in conclusion , there is an accumulating body of exclusive vector meson production data , measured with a systematic precision of @xmath175 , which exhibit two classes of @xmath40 behaviour : a slow rise consistent with that of previously measured diffractive data for low @xmath108 photoproduction data but a significant rise of these cross - sections above a finite value of @xmath108 , @xmath29 or @xmath74 . in general , the cross - sections at large @xmath40 can be compared to pqcd when either @xmath108 , @xmath29 or @xmath74 become larger than the scale @xmath176 . precisely how the transition from the non - perturbative to the perturbative regime is made is currently being determined experimentally . _ diffractive event selection : _ the study of the vector meson resonances enables specific points on the @xmath9 scale to be investigated . the inclusive dissociation into any low - mass state , @xmath5 , from the ( virtual ) photon provides additional information : one of the major advances in the subject of diffraction has been the observation of large rapidity gap events in dis and their subsequent analysis in terms of a diffractive structure function @xcite . in addition , the relationship between these dis measurements and those in the photoproduction regime provide insight into the transition of the diffractive structure function in the @xmath177 limit . in these analyses , the typical signature of diffraction is a rapidity gap , defined by measuring the maximum pseudorapidity of the most - forward going particle with energy above 400 mev , @xmath178 , and requiring this to be well away from the outgoing proton direction . a typical requirement of @xmath179 corresponds to a low mass state measured in the detectors of @xmath180 units and a large gap of @xmath181 units with respect to the outgoing proton ( nucleon system ) . in order to increase the lever arm in @xmath182 , the h1 analysis has extended the @xmath178 cuts to 3.2 . this is achieved by combining the calorimetry information with the forward muon system and proton remnant taggers . these extensions enable a cross - section beyond that due to simple diffractive processes to be determined at the expense of a significant non - diffractive contribution ( up to @xmath183 ) . as illustrated in fig . [ h1c ] , the monte carlo description of the @xmath178 distributions shows a clear excess over the non - diffractive models ( labelled mc django ) and a monte carlo to describe higher reggeon exchanges ( labelled mc pion ) . the data is well - described by a mixture ( labelled mc mix ) including an additional contribution due to the pomeron exchange , both in the @xmath178 distribution and the corresponding observed @xmath184 distribution . = 8.cm = 8.cm one of the major uncertainties comes from the estimation of the various contributions to the cross - section which depends on monte carlo techniques . this problem has been addressed in a different way in the zeus analysis @xcite . here , the mass spectrum , @xmath182 , is measured as a function of @xmath16 and @xmath74 , as shown in fig . [ zeusm ] for a representative interval , where the measured mass is reconstructed in the calorimeter and corrected for energy loss but not for detector acceptance , resulting in the turnover at large @xmath182 . the diffractive data are observed as a low mass shoulder at low @xmath182 . which becomes increasingly apparent at higher @xmath16 . also shown in the figure are the estimates of the non - diffractive contribution based on a direct fit to the data , discussed below . the probability of producing a gap is exponentially suppressed as a function of the rapidity gap , and hence as a function of @xmath185 ) , for non - diffractive interactions . the slope of this exponential is directly related to the height of the plateau distribution of multiplicity in the region of rapidity where the subtraction is made . the data can thus be fitted to functions of the form @xmath186 , in the region where the detector acceptance is uniform , where @xmath64 , @xmath187 and @xmath188 are determined from the fits . here , @xmath188 represents a first - order estimate of the diffractive contribution which is approximately flat in @xmath185 ) . the parameter which determines the background is @xmath64 . in general the measured value of @xmath64 is incompatible with that of the ariadne monte carlo . this result in itself is interesting , since the fact that ariadne approximately reproduces the observed forward @xmath189 ( @xmath19 multiplicity ) flow but does not reproduce the measured value of @xmath64 suggests that significantly different correlations of the multiplicities are present in non - diffractive dis compared to the monte carlo expectations . this method enables a diffractive cross - section to be determined directly from the data at the expense of being limited in the range of large masses that can analysed . finally , the advent of the leading proton spectrometers ( lps ) at hera is especially important in these diffractive measurements , since internal cross - checks of the measurements as a function of @xmath29 , @xmath9 , @xmath40 and @xmath74 can be performed and underlying assumptions can be studied experimentally . only in these measurements can we positively identify the diffracted proton and hence substantially reduce uncertainties on the non - diffractive and double dissociation backgrounds . this is illustrated in fig . [ mxl ] , where the @xmath190 ( where @xmath191 ) distribution includes a clear diffractive peak for @xmath192 . it should be noted , however , that the contribution from other reggeon exchanges can not be neglected until @xmath193 ( in fact the result at lower @xmath190 can be simply interpreted via reggeon ( approximated by pion ) exchange , as discussed below . ) however , new experimental uncertainties are introduced due to the need for precise understanding of the beam optics and relative alignment of the detectors . reduced statistical precision also results due to the limited geometrical acceptance of the detectors ( @xmath32 6% ) . = 8.cm _ photoproduction results : _ zeus has measured the photon dissociation @xmath29 distribution using the lps , as shown in fig . an exponential fit to the data yields a @xmath64-slope parameter , @xmath194 gev@xmath72 . a comparison of the data with lower @xmath16 data from chapin et al . shows that the result is consistent with shrinkage , as previously discussed in relation to exclusive @xmath94 production . results on the photon dissociation cross - sections as a function of @xmath182 in two @xmath16 intervals are shown in fig . [ h1phot ] . regge theory predicts the form of the cross - section as a function of @xmath184 and @xmath16 , as discussed with respect to proton dissociation . the cross - section is therefore fitted to the form @xmath195 the contributions due to reggeon exchange are fixed using the lower energy data . integrating over the @xmath29 dependence of @xmath123 yields a value of @xmath196(sys)@xmath197(model ) , again , consistent with soft pomeron exchange . = 10.cm _ deep inelastic structure of diffraction : _ a new era for diffraction was opened with the study of the dissociation of @xmath198 photons . here , the photon can be considered as probing the structure of the exchanged colourless object mediating the interaction . the deep inelastic structure of colour singlet exchange is therefore being studied . in the presentation of the results , the formalism changes @xcite , reflecting an assumed underlying partonic description , and two orthogonal variables are determined @xmath199 where @xmath200 is the momentum fraction of the pomeron within the proton and @xmath201 is the momentum fraction of the struck quark within the pomeron . the structure function is then defined by analogy to that of the total @xmath202 cross - section @xmath203 where the contribution of @xmath204 and radiative corrections are neglected and an integration over the @xmath29 variable has been performed . = 10.cm in addition to the structure of the pomeron , corresponding to large @xmath190 , it is also possible to study the structure of the reggeons that contribute at lower @xmath190 . h1 has analysed the leading proton data at lower @xmath190 ( @xmath205 ) and employed the formalism noted above to measure the structure of the exchange for reasonably forward protons , as shown in fig . [ lp3 ] @xcite . the data are consistent with a flat @xmath206 dependence in all intervals of @xmath201 and @xmath74 i.e. @xmath207 . this is consistent with a factorisable ansatz of @xmath208 where @xmath209 measures the flux of reggeons in the proton and @xmath210 is the probed structure of these reggeons . the exponent of @xmath206 is identified as @xmath211 , where @xmath212 measures the effective @xmath206 dependence ( @xmath213 dependence at fixed @xmath182 and @xmath74 ) of the cross - section , integrated over @xmath29 . the data are consistent with @xmath214 , corresponding to @xmath215 . these data involve colour singlet exchange and need to be explained in terms of qcd , but they are clearly not of a diffractive nature . the area of interest for diffraction is in the behaviour of small values of @xmath216 , where @xmath206 is now identified as @xmath200 . here , the data fall approximately as @xmath217 ( equivalent to a flat cross - section with increasing @xmath16 ) and therefore the data are plotted as @xmath218 in fig . [ h1f2d3 ] . in the h1 case , the measurement is presented with no explicit subtraction for the non - diffractive contribution and quoted for limited masses of the dissociated proton system ( @xmath219 gev ) . the measurement relies upon a good understanding of the various contributions to the cross - section in and around the measured region : the control plots in fig . [ h1c ] illustrate how well this is achieved by combining the different monte carlo contributions . = 12.cm fits of the form @xmath220 are performed where the normalisation constants @xmath221 are allowed to differ in each @xmath222 interval . the fits are motivated by the factorisable ansatz of @xmath223 where @xmath224 measures the flux of pomerons in the proton and @xmath225 is the probed structure of the pomeron . the exponent of is identified as @xmath226 , where @xmath123 measures the effective dependence ( @xmath213 dependence at fixed @xmath182 and @xmath74 ) of the cross - section , integrated over @xmath29 , as discussed in relation to exclusive vector meson production . however , the data are now sufficently precise that a single value for @xmath137 is not sufficient . in fact , the contribution from reggeon exchange can not be excluded since the data extend to @xmath200 beyond approximately 0.01 and therefore a full fit to the pomeron and reggeon contributions is required . a fit to the pomeron and reggeon contributions has therefore been performed @xmath227 the fit provides a very good overall description of the data with a @xmath228/dof = 165/156 . here , the function @xmath229 is taken from the grv parametrisation of the pion structure function and it is assumed that the interference between the pomeron and reggeon exchanges is maximal . this is illustrated for example bins of @xmath201 at fixed @xmath153 gev@xmath33 in fig . [ f2dbits ] . the lower curve represents the pomeron contribution , the middle curve corresponds to the sum of the two contributions without taking into account the ( positive ) interference and the upper curve is the full fit result . it should be noted that the quark - like reggeon structure function falls like @xmath230 at these @xmath201 values whereas the pomeron structure function emerging after integration over @xmath200 is rather flat : the reggeon contributions thereforefore play a significant role at smaller values of @xmath201 . also , for the smallest values of @xmath201 , the form of the pion ( and hence reggeon ) structure function is less well known and the contribution of non - colour singlet exchange contributions starts to become significant . assuming @xmath231 and integrating over @xmath29 with @xmath232 gev@xmath72 using @xmath233 gev@xmath72 yields @xcite @xmath234 @xmath235 = 10.cm the zeus virtual - photon proton cross - sections measured at fixed @xmath182 and @xmath16 , measured using the @xmath184 method ( and therefore explicitely subtracting the non - diffractive contribution ) can be converted to @xmath218 at fixed @xmath201 and @xmath200 . these results are shown in fig . [ zeusf2d3 ] as the zeus(@xmath184 ) @xcite analysis , compared to the zeus lps analysis of @xmath236 , integrated over the measured @xmath29 range , in comparable intervals of @xmath201 and @xmath74 as a function of @xmath200 . the double dissociation contribution is estimated with similar uncertainties to the vector meson case . common systematic errors are similar to those for the @xmath84 analyses ( @xmath237 ) with additional acceptance uncertainties due to variations of the input diffractive monte carlo distributions . the lps data therefore provide an excellent calibration to cross - check the background subtraction methods . the overall cross - sections in overlapping @xmath201 and @xmath74 intervals are in good agreement . in the region of overlap , the h1 and zeus lps data points also agree well . the zeus lps data beyond of 0.01 again tend to turn over , as observed in the h1 data . = 10.cm the corresponding zeus lps measurement of the @xmath29 distribution is shown in fig . [ tdis ] , measured in the range @xmath238 , @xmath239 gev@xmath33 , @xmath240 and @xmath241 gev@xmath33 . the slope can be characterised by a single exponential fit with @xmath242 gev@xmath72 @xcite . this is very similar to the slope obtained in the photoproduction case and is somewhat high compared to the value of @xmath243 expected for a predominantly hard pomeron but lies within the range of expectations of @xmath244 . the zeus @xmath184 data are largely restricted to the small @xmath245 and @xmath246 region and therefore purely pomeron contributions are significant . a direct fit to the data in fixed @xmath184 intervals yields a similar value for @xmath45 . this is illustrated in fig . [ alphapom ] where the the @xmath74 dependence of the extracted @xmath45 is compared to the h1 result as well as the soft pomeron prediction . the results for @xmath45 , obtained using different experimental methods , are compatible . these values are however incompatible with the predicted soft pomeron behaviour of @xmath247 . as noted previously in relation to exclusive @xmath94 production ( corresponding to the large @xmath201 region ) , the contribution of longitudinal photons is significant . an upper estimate of the effect of @xmath151 made by assuming @xmath248 rather than @xmath249 increases the measured value of @xmath45 by about 0.05 . similarly , if no shrinkage were assumed , the value of @xmath45 would increase by @xmath250 . = 5.cm = 10.cm the values can be compared with @xmath251 0.2 obtained from the exclusive photoproduction of @xmath103 mesons and the zeus exclusive @xmath94 electroproduction data at large @xmath74 . these values are also compatible with fits to the inclusive dis @xmath84 ( i.e. the dependence of the corresponding total cross - sections ) which yield @xmath129 0.2 to 0.25 in the measured @xmath74 range @xcite . in the model of buchmller and hebecker @xcite , the effective exchange is dominated by one of the two gluons . in terms of @xmath47 , where the optical theorem is no longer relevant , the diffractive cross - section would therefore rise with an effective power which is halved to @xmath129 0.1 to 0.125 . given the uncertainties , the measured values are within the range of these estimates . the overall cross - sections in each @xmath201 , @xmath74 interval are similar and one can integrate over the measured dependence in order to determine @xmath252(@xmath253 ) , a quantity which measures the internal structure of the pomeron up to an arbitrary integration constant . in fig . [ h1f2dt ] the h1 data are compared to preliminary qcd fits @xcite . the data are fitted in the range @xmath254 using the functional form shown in fig . [ h1f2d3 ] . the general conclusions from the @xmath201 dependence are that the pomeron has a rather flat structure as a function of @xmath201 . this is typically characterised by a symmetric @xmath255 dependence , but an additional significant contribution at low @xmath201 is required which has been fitted in the zeus analysis @xcite . the @xmath74 behaviour is broadly scaling , consistent with a partonic structure of the pomeron . probing more deeply , however , a characteristic logarithmic rise of @xmath252 is observed in all @xmath201 intervals . most significantly , at large @xmath201 a predominantly quark - like object would radiate gluons resulting in negative scaling violations as in the case of the large-@xmath2 ( @xmath256 ) behaviour of the proton . the question of whether the pomeron is predominantly quarks or gluons , corresponding to a quarkball " or a gluemoron " @xcite , has been tested quantitatively by h1 using qcd fits to @xmath252 @xcite . a flavour singlet quark density input of the form @xmath257 , where @xmath258 is the momentum fraction carried by the quark , yields a high @xmath228/dof = 95/39 , since the characteristic @xmath74 behaviour is not reproduced . adding a gluon contribution of similar form gives an excellent description of the data with @xmath228/dof = 36.8/37 , as shown in fig . [ h1f2dt ] . in fig . [ h1pdfs ] , the corresponding parton distributions are shown . in general , the fits tend to favour inputs where the gluon carries a significant fraction , @xmath19 70 to 90% , of the pomeron s momentum . the measurements of the scaling violations of the structure function of the pomeron provide a method to determine the parton distributions of the pomeron . the question of whether such an approach is useful can be addressed by applying these parton distributions to calculations for other processes which are directly sensitive to this partonic structure . historically , the measurements of @xmath259 , the mean transverse momentum - squared of the outgoing hadrons , as a function of @xmath260 , the scaled longitudinal momentum distribution , provided insight into the structure of the proton . here , the variables are measured in the hadronic centre of mass frame and with respect to the virtual photon - proton axis which is equivalent to the virtual photon - pomeron axis for small values of @xmath29 . in fig . [ seathr](a ) , the h1 @xmath261 data ( full circles ) are compared to the emc @xmath262 data at similar @xmath263 values . the data are also compared to the rapgap ( rg ) monte carlo predictions incorporating quarks and gluons ( -qg ) and quarks only ( -q ) @xcite . ( meps ) and ( cdm ) refer to the matrix elements plus parton showers and colour dipole model fragmentation schemes , respectively . the h1 data are approximately symmetric about @xmath264 with a relatively large @xmath259 peaking around 0.6 gev@xmath33 . the symmetry and relatively large @xmath265 values reflect the underlying boson - gluon fusion process where a leading " gluon from the pomeron interacts with the virtual photon . this behaviour is in contrast to the emc @xmath262 data where qcd radiation is suppresed in the negative-@xmath266 ( proton remnant ) region . quantitatively the rapgap monte carlo which incorporates the pomeron parton densities ( -qg ) gives a good description of the data , provided that quarks and gluons are incorporated whereas a model with only quark ( -q ) fails to describe the data . these conclusions are relatively independent of the fragmentation scheme , but the colour dipole model tends to give a better description of the data . similarly , event shape variables have been used at @xmath267 colliders in order to establish the existence of gluon bremsstrahlung radiation . in this case , the measurement of e.g. mean thrust ( the mean value of the scaled longitudinal momentum with respect to the axis which maximises this value ) is sensitive to the gluon - induced diagrams . a comparison of @xmath268thrust@xmath269 with @xmath267 annihilation experiments as a function of the reciprocal of hadronic centre of mass is shown in fig . [ seathr](b ) . the diffractive data exhibit lower thrust values compared to @xmath267 data for all values of @xmath184 . this additional broadening is due to the boson gluon fusion process which has no analogue in @xmath267 annihilation continuum region . the general increase in thrust with increasing @xmath184 ( decreasing @xmath270 ) is indicative of jet production . the question of the constituent content of the pomeron can also be addressed via measurements of diffractively produced jets in the photoproduction data @xcite . jets are reconstructed at large @xmath16 ( @xmath271 gev ) using the cone algorithm with unit cone radius and two jets with @xmath272 gev . the diffractive contribution is identified as a tail in the @xmath178 distribution of these events above the pythia 5.7 @xcite monte carlo expectation . in fig . [ betaxg ] the measured cross - section is compared to various model predictions as a function of @xmath273 , an estimator of the fraction of the pomeron momentum transferred to the dijet system . the non - diffractive contribution estimated from pythia ( not shown ) is significantly lower than the data . here , standard photon and proton parton distributions are adopted and the overall scale , which agrees with the non - diffractive data normalisation , is set by @xmath274 . also shown are the predicted diffractive cross - sections from the lo qcd calculation plus parton showers of pompyt , using a hard ( @xmath275 ) quark combined with either a hard , soft ( ( @xmath276 ) or singular gluon where a donnachie - landshoff flux factor is adopted . sampling low - energy ( soft ) gluons corresponds to a small cross - section and can be discounted , whereas high - energy ( hard ) gluons and/or quarks can account for the cross - section by changing the relative weights of each contribution . the shape of the @xmath273 distribution is clearly sensitive to the shape of the input gluon distribution . the @xmath277 distribution for these events , where @xmath277 is the corresponding estimator of the fraction of the photon momentum transferred to the dijet system , is peaked around 1 , indicating that at these @xmath274 values a significant fraction of events is due to direct processes where the whole photon interacts with the pomeron constituents . so far we have only considered the case of small-@xmath29 diffraction with respect to the outgoing proton . further insight into the diffractive exchange process can be obtained by measurements of the rapidity gap between jets . here , a class of events is observed with little hadronic activity between the jets @xcite . the jets have @xmath272 gev and are separated by a pseudorapidity interval ( @xmath25 ) of up to 4 units . the scale of the momentum transfer , @xmath29 , is not precisely defined but is of order @xmath278 . a gap is defined as the absence of particles with transverse energy greater than 300 mev between the jets . the fraction of events containing a gap is then measured as a function of @xmath25 , as shown in fig . [ zeuscs ] . the fit indicates the sum of an exponential behaviour , as expected for non - diffractive processes and discussed in relation to the diffractive dis data , and a flat distribution expected for diffractive processes . at values of @xmath279 , an excess is seen with a constant fraction over the expectation for non - diffractive exchange at @xmath280 . this can be interpreted as evidence for large-@xmath29 diffractive scattering . in fact , secondary interactions of the photon and proton remnant jets could fill in the gap and therefore the underlying process could play a more significant rle . the size of this fraction is relatively large when compared to a similar analysis by d and cdf where a constant fraction at @xmath281 is observed @xcite , as discussed below . = 5.cm = 5.cm the study of diffraction benefits by looking at information from other types of interaction . in @xmath267 where there is no complex colour state in the interacting beams , various searches for rapidity gaps yield results which can be explained in terms of exponentially - suppressed colour exchange and these data do not require the introduction of a pomeron " @xcite . in contrast , the earliest work on diffractive processes focussed on @xmath30 collisions where two complex colour states interact . the latest studies at the tevatron complement those at hera and allow tests of factorisation in @xmath202 compared to @xmath53 . the studies of colour singlet exchange between jets at hera was inspired by earlier studies at the tevatron . these studies determined percentage gap - fractions of @xmath282% for d @xcite and @xmath283% for cdf @xcite . the behaviour of this colour singlet fraction as a function of average dijet energy is shown in fig . [ d0cs](a ) for jet @xmath274 thresholds of 15 gev ( low ) , 25 gev ( medium ) and 30 gev ( high ) @xcite . in fig . [ d0cs](b ) , the gap - fraction is examined as a function of @xmath25 . a simple two - gluon exchange model with no additional qcd dynamics would tend to produce a flat gap - fraction , but the tendency towards an increasing gap - fraction with @xmath25 indicates an additional dynamical mechanism is necessary to describe the data . the differences in the overall gap - fractions observed at hera near 10% , compared to those at fermilab of approximately 1% , may reflect the higher @xmath16 values of the tevatron compared to hera . but the fact that this difference is so large indicates differences in the underlying high-@xmath284 @xmath116 interactions compared to the relatively low-@xmath2 @xmath3 interactions where spectator interactions are more likely to fill in the gap . = 12.cm diffractive dijet production has been studied at the tevatron . the presence of a diffracted proton may be identified either by a large rapidity gap on either the proton or antiproton side or by directly detecting the leading proton ( cdf roman pots ) . in the rapidity gap analyses , d measure uncorrected dijet rates with @xmath285 gev and @xmath286 in coincidence with the multiplicity in the electromagnetic calorimeter ( @xmath287 ) opposite the dijet system , as shown in fig . [ tevdijet](a ) . similarly , cdf measure the multiplicity in the forward part of the calorimeter ( @xmath288 ) in coincidence with the number of hits in the bbc scintillator counter close to the beampipe ( @xmath289 ) , as shown in fig . [ tevdijet](b ) . here the jets are measured for @xmath290 gev and @xmath291 the diffractive events concentrate in the region @xmath292 . the ratio of diffractive dijet events is measured by cdf to be @xmath293% and by d to be @xmath294% . these preliminary figures are therefore in good agreement and can be used to constrain the gluon content of the pomeron . in addition , @xmath295% has been measured using the cdf roman pots in region of large @xmath206 ( @xmath296 , @xmath297 gev and @xmath298 gev@xmath33 ) . this is a region where reggeon ( quark - like ) contributions are presumably important . cdf have also tagged diffractive @xmath16 production using high-@xmath97 electrons / positrons and missing @xmath97 to tag the @xmath16 and then searching for a rapidity gap on the opposite side as in the dijet gap analysis @xcite . the corrected ratio for diffractive / non - diffractive @xmath16-production is measured to be @xmath299 an important question when relating the various diffractive measurements which involve a hard scale is whether regge factorisation , in terms of a pomeron flux and parton densities within the pomeron , is applicable . in particular , if this approach is to be useful , a @xmath300 flux is required ( or a qcd description of the non - universal correction to the flux ) . in this context , the rates for diffractive processes at hera are compared to those observed at the tevatron below . we have two sets of cdf data probing the pomeron structure at similar momentum scales , @xmath274 and @xmath301 . each probes the large @xmath258 structure of the pomeron with the dijet and @xmath16 data predominantly sampling the ( hard ) gluon and quark distributions , respectively . in addition , we have the corresponding dis @xcite and jet hera data sampling the ( hard ) gluon and quark distributions , respectively . in fig . [ cdfglu ] the momentum fraction carried by the ( hard ) gluon , @xmath302 , is plotted versus the momentum fraction of partons in the pomeron assuming a donnachie - landshoff flux . the cdf data are consistent with a momentum fraction carried by the gluons of @xmath303 , in agreement with the zeus measurements of @xmath304 , taking into account the systematic uncertainties due ( mainly ) to the estimation of the non - diffractive background . this in turn can be compared with the h1 nlo parton distributions ( see fig . [ h1pdfs ] ) which indicate @xmath305 in the high @xmath74 region . there is therefore reasonable agreement on the parton content of the pomeron . however , the overall diffractive rates are significantly higher at hera compared to the tevatron . this is reflected in the difference in the overall level of the zeus and cdf data in fig . [ cdfglu ] . this corresponds to a significantly different flux of pomerons ( i.e. breaking of regge factorisation ) which has been predicted in terms of qcd @xcite to reduce the diffractive cross - sections for processes which have two strongly - interacting initial state hadrons . these effects are not apparent in the hera data , where a virtual photon or a ( predominantly ) direct photon participate in the hard scattering process . = 10.cm double pomeron exchange , where the @xmath306 and @xmath307 remain intact , has also been studied by cdf and d for their dijet samples . this process should be directly sensitive to regge - factorisation breaking effects . both experiments find a ratio of hard double pomeron exchange events to non - diffractive events of @xmath308 . this is consistent with independent dissociation of the @xmath306 and @xmath307 , with probabilities @xmath309 , but further studies are required to establish these rates and determine whether these can be explained by the factorisation - breaking calculations @xcite . the soft pomeron does not describe @xmath310 diffractive data measured at hera . as the photon virtuality or the vector meson mass increase a new dependence on @xmath40 emerges . as we investigate the pomeron more closely , a new type of dynamical pomeron may begin to play a rle ; a dynamical pomeron whose structure is being measured in dis . these data are consistent with a partonic description of the exchanged object which may be described by pqcd . the cross - sections for hard diffractive processes at hera can be compared to those observed at the tevatron . all data are consistent with a significant gluon contribution of the partons within the pomeron i.e. qcd factorisation appears to be observed . however , a universal pomeron flux does not describe the observed rates i.e. regge factorisation does not apply . this is qualitatively expected from qcd corrections to the rates , but the effects are large and non - perturbative . the experimental work focuses on extending the lever arms and increasing the precision in @xmath29 , @xmath9 , @xmath40 and @xmath74 in order to explore this new structure . before more precise tests can be made , further theoretical and experimental input is required to reduce the uncertainties due to non - diffractive backgrounds and proton dissociation as well as the treatment of @xmath204 and radiative corrections . the results presented in this talk are a summary of significant developments at hera and the tevatron in the study of diffraction . the financial support of the desy directorate and pparc allowed me to participate in this research , whilst based at desy , for which i am very grateful . many thanks to halina abramowicz , nick brook , allen caldwell , nicolo cartiglia , malcolm derrick , john dainton , martin erdmann , elisabetta gallo , peppe iacabucci , michael kuhlen , aharon levy , jason mcfall , paul newman , juan puga , david saxon , laurel sinclair , ian skillicorn , juan terron and robert waugh for their encouragement , enthusiasm , help and advice . it is a pleasure to thank the organisers for their additional financial support and a very enjoyable summer school . 99 h. fritzsch , aachen workshop on qcd - 20 years later " , world scientific , ( 1992 ) 827 . n. cartiglia ( ssi 96 ) , hep - ph/970325 . k. goulianos , phys . rep . 101 ( 1983 ) 169 ; nucl . phys . suppl . ) 12 ( 1990 ) 110 . a. donnachie and p.v . landshoff , nucl . b244 ( 1984 ) 322 . cdf collab . , f. abe et al . , phys . d50 ( 1994 ) 5550 . cudell , k. kang and s.k . kim , hep - ph/9601336 . wa91 collab . , s. albatzis et al . , phys . b321 ( 1994 ) 509 . w. buchmller and a. hebecker , desy 95 - 077 , w. buchmller , dis97-aip conference proceedings 147 ( 1997 ) 542 . crittendon , hep - ex/9704009 ( and references therein ) . d. aston et al . b209 ( 1982 ) 56 . m. ryskin , z. phys . c57 ( 1993 ) 89 . a. levy , desy 95 - 204 . m. glck , e. reya and a. vogt , phys . b306 ( 1993 ) 391 . zeus collab . , j. breitweg et al . , jerusalem eps 97 - 646 . e665 collab . , m. adams et al . , mpi - phe-97 - 03 . s. brodsky et al . , phys . rev d50 ( 1994 ) 3134 . zeus collab . , m. derrick et al . b356 ( 1995 ) 601 . h. holtmann et al . , hep - ph/9503441 . cdf collab . , f. abe et al . , d50 ( 1994 ) 5535 . e. gallo , hep - ex/9710013 . h1 collab . , t. ahmed et al . b348 ( 1995 ) 681 . zeus collab . , m. derrick et al . , z. phys . c68 ( 1995 ) 569 . zeus collab . , j. breitweg et al . , jerusalem eps 97 - 638 . ( final results for the measurement of @xmath311 are available in desy 97 - 184 . ) g. ingelman and k. jansen - prytz , z. phys . c58 ( 1993 ) 285 . h1 collab . , c. adloff et al . , jerusalem eps 97 - 379 . h1 collab . , c. adloff et al . , jerusalem eps 97 - 337 . ( final results are available in z. phys . c76 ( 1997 ) 613 . ) l. lnnblad , comp . 71 ( 1992 ) 15 . p. bruni and g. ingelman , desy 93 - 187 ; proceedings of the europhysics conference on hep , marseille 1993 , 595 . nikolaev and b.g . zakharov , z. phys . c53 ( 1992 ) 331 ; m. bertini et al . , hep - ph/9710547 . f. close and j. forshaw , hep - ph/9509251 . c. cormack , dis97-aip conference proceedings 147 ( 1997 ) 585 . ( final results for the measurement of @xmath312 are available in desy 97 - 210 . ) zeus collab . , j. breitweg et al . , jerusalem eps 97 - 648 ; zeus collab . , m. derrick et al . b356 ( 1995 ) 129 . bengtsson and t. sjstrand , comp . 46 ( 1987 ) 43 ; t. sjstrand , cern - th.6488/92 . zeus collab . , m. derrick et al . b369 ( 1996 ) 55 . sld collab . , k. abe et al . , 76 ( 1996 ) 4886 . d collab . , s. abachi et al . 72 ( 1994 ) 2332 ; d collab . , s. abachi et al . , fermilab - pub-95 - 302-e ( 1995 ) . cdf collab . , f. abe , et al . , 74 ( 1995 ) 855 . j. perkins , dis97-aip conference proceedings 147 ( 1997 ) 533 . cdf collab . , f. abe , et al . 78 ( 1997 ) 2698 . d. soper , dis97-aip conference proceedings 147 ( 1997 ) 147 . j. collins et al . , phys . b307 ( 1993 ) 161 ; j. collins , hep - ph/9709499 .

due to the gravitational influence of density fluctuations driven by magneto - rotational instability in the gas disk , planetesimals and protoplanets undergo diffusive radial migration as well as changes in other orbital properties . the magnitude of the effect on particle orbits can have important consequences for planet formation scenarios . we use the local - shearing - box approximation to simulate an ideal , isothermal , magnetized gas disk with vertical density stratification and simultaneously evolve numerous massless particles moving under the gravitational field of the gas and the host star . we measure the evolution of the particle orbital properties , including mean radius , eccentricity , inclination , and velocity dispersion , and its dependence on the disk properties and the particle initial conditions . although the results converge with resolution for fixed box dimensions , we find the response of the particles to the gravity of the turbulent gas correlates with the horizontal box size , up to 16 disk scale heights . this correlation indicates that caution should be exercised when interpreting local - shearing - box models involving gravitational physics of magneto - rotational turbulence . based on heuristic arguments , nevertheless , the criterion , where is the horizontal box size and is the distance to the host star , is proposed to possibly circumvent this conundrum . if this criterion holds , we can still conclude that magneto - rotational turbulence seems likely to be ineffective at driving either diffusive migration or collisional erosion under most circumstances .

1103.3268

the gas disk surrounding a young stellar object is usually argued to be a turbulent accretion disk when the star is still in the process of accumulating materials ( see , e.g. , * ? ? ? while the molecular viscosity of the gas is negligibly small , shear stress resulting from turbulence can effectively drive angular momentum transport in the disk . at the same time , the dynamics of protoplanetary objects embedded in the disk is significantly affected by their interactions with the turbulent gas , at a stage when these objects are still amassing solids to grow in size ( e.g. , * ? ? ? * and references therein ) . understanding particle dynamics in a turbulent gas disk , therefore , will provide insight into the paths along which new planets may eventually form . one of the most promising mechanisms to drive the turbulence is through the magneto - rotational instability ( mri ; see , e.g. , * ? ? ? * and references therein ) . a weakly magnetized , differentially rotating gas disk is unstable to linear perturbations , and after nonlinear effects set in , the gas presumably reaches a statistically steady , sustained , turbulent state . the underlying processes that determine the properties of this saturated state are still under active research ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? nevertheless , it is known that the strength of this magneto - rotational turbulence depends on the net magnetic flux threading the disk ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * hereafter ymm09 ) , and thus can be controlled to the desired level and treated as a parameter . the gravitational torques exerted on an embedded solid object due to density fluctuations in magneto - rotational turbulence have been shown to be stochastic , rendering the motion of the object a random walk ( @xcite ; ymm09 ; @xcite ) on top of the ordered migration expected in a laminar disk @xcite . note that the ordered migration may be either inward or outward in different regions of the disk @xcite . in addition to inducing this radial diffusive migration , the same gravitational force drives the diffusion of orbital eccentricity of the solid objects , increasing their velocity dispersion ( @xcite ; ymm09 ; @xcite ) . the effect on the orbital inclination , however , has not been investigated yet , as that requires the inclusion of the vertical component of the gravity from the host star . overall , the gravitational influence of the turbulent gas on particle orbital properties may play an important role in the course of planet formation . to assess the significance of this process in planet formation scenarios , numerical simulations simultaneously evolving magneto - rotational turbulence and particle dynamics can be conducted and the resulting particle orbital evolution can be statistically measured . however , a consistent picture of the actual magnitude of this effect has been elusive . using the local - shearing - box simulations , we reported in ymm09 that the response of test particles to the gravity of the turbulent gas is systematically lower than what was previously reported in global disk models . we also found in the same study that the effect significantly depends on the horizontal box size of the numerical models . recently , @xcite have found similar behavior in their local models and argued that large box size might be necessary to see convergence in the stochastic torque and forcing generated by the turbulence . extending the local unstratified disk models of ymm09 , we now consider disks with vertical density stratification by including the linearized vertical gravity from the central star . we examine whether a stratified disk model gives results consistent with a comparable unstratified model and whether convergence in particle orbital evolution , if any , can be achieved with local - shearing - box simulations . in section [ s : model ] , we describe in detail our numerical models . we report the properties of our modeled magneto - rotational turbulence in section [ s : turbulence ] and the statistical evolution of particle orbital properties in section [ s : particles ] . we analyze the gravitational field generated by the turbulence in section [ s : force ] and discuss the convergence issues of our local models in section [ s : box_size ] . we conclude our discussion in section [ s : summary ] . directly extending the work we reported in ymm09 , we continue to use the pencil code to model particles moving in magneto - rotational turbulence . we describe in detail the equations and the relevant numerical techniques we implemented in the code . we use the local shearing box approximation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? a local shearing box is a small cartesian box at a large distance @xmath2 from the host star such that the center of the box revolves at the keplerian angular speed @xmath3 . the box is always oriented with the @xmath4-axis directed radially and the @xmath5-axis azimuthally . in contrast to ymm09 , we include the linearized vertical gravity from the host star so that the disk is vertically stratified . we again impose a vertical , external magnetic field @xmath6 to maintain a non - zero magnetic flux . the mhd equations then become @xmath7 where @xmath8 is the gas density , @xmath9 is the gas velocity relative to the background shear flow , @xmath10 is the gas pressure , @xmath11 is the electric current density , @xmath12 , and @xmath13 is the vacuum permeability . the terms @xmath14 , @xmath15 , and @xmath16 are numerical dissipation terms , including both hyper - diffusion and shock diffusion , that are needed to stabilize the scheme . the reader is referred to ymm09 for their description . to close the system of equations ( [ e : mhd_cont])([e : mhd_ind ] ) , we assume an isothermal `` equation of state '' , @xmath17 , where @xmath18 is the isothermal speed of sound . this is an ideal limit of the energy equation , in which heat diffusion is effectively instantaneous . gas buoyancy should thus be absent and the vertical mixing in the resulting magneto - rotational turbulence might be artificially enhanced @xcite . nevertheless , @xcite showed that vertical motions are not completely suppressed even in the case of adiabatic gas , where no heat diffusion occurs , and the differences in the mean kinetic and the magnetic energies of a vertically stratified gas disk between the isothermal and the adiabatic limits may only amount to about a factor of two . we set up the gas density so that the gas is in vertical hydrostatic equilibrium initially : @xmath19 where @xmath20 is the mid - plane gas density and @xmath21 is the vertical disk scale height . difference between our definition of disk scale height @xmath22 ( equation ( [ e : rho_0 ] ) ) and some other authors in the literature.[scale_height ] ] we fix the vertical dimension of the computational domain at @xmath23 . the initial magnetic vector potential @xmath24 is set to zero , while the external magnetic field is fixed at such a level that the corresponding plasma @xmath25 is @xmath26 in the mid - plane . gaussian noise of magnitude @xmath27 in each component of the gas velocity @xmath9 is generated to seed the mri , where @xmath28 is the orbital period at the center of the shearing box . our assumption of a net vertical magnetic field allows us to compare our models directly to the unstratified models presented in ymm09 . the strength of this field controls the strength of the mri - driven turbulence , so imposing it is an effective way to adjust the viscous stress generated by the mri . a similar procedure has been used in other recent publications including @xcite , @xcite , @xcite , and @xcite . some justification for this assumption can be found in the argument that at high altitude where the gas is tenuous , the magneto - convective motions tend to turn horizontal magnetic fields into vertical ones ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? therefore , we take the limit that @xmath29 at the @xmath30-boundary while allowing the vertical component ( @xmath31 ) to change freely . in terms of the vector potential @xmath24 , this boundary condition translates to @xmath32 . in the horizontal directions , we use the standard sheared periodic boundary conditions for the magnetic field , i.e. , the field is periodic after a shift compensating for the keplerian shear flow ( @xcite ; see also section [ ss : poisson ] ) . we explored several different vertical boundary conditions for the gas and the field . we found that outflow boundary conditions for both led to numerical instability during the initial growth of the mri . although @xcite and @xcite did publish models using outflow boundary conditions , their disk lost significant amounts of mass through the boundary during their run . we also found that periodic boundary conditions on both gas and field led to random crashes after tens of orbits of saturated turbulence for reasons that remain unclear . on the other hand , the combination of the vertical field boundary condition with periodic boundary conditions in the gas leads to a stable numerical result for the hundreds of orbits required for us to accurately measure the evolution of particle orbital properties ( see also * ? ? ? the horizontal boundary conditions are again sheared periodic . in this setup , the total mass is conserved while saturated turbulent motion can be maintained near the vertical boundaries at a steady level . while it is interesting in itself to study the coronal structure and possibly the net flow of the gas @xcite , we specifically exclude the modeling of the coronal regions to maintain the steadiness of the magneto - rotational turbulence and thus improve the statistical measurement of the particle orbital properties . the gravitational acceleration due to the host star across the boundary does not present any numerical difficulty . even though the vertical component of the gravitational acceleration changes sign when a fluid element is wrapped around the vertical boundary , the vertical velocity component of the element retains sign such that its velocity and acceleration always have the same sign after reemerging from the opposite boundary , resembling material ballistically falling back down onto the disk . numerically , no z - derivative of the gravitational acceleration due to the host star is ever evaluated ( see equation ( [ e : mhd_mom ] ) ) , so the discontinuity of the acceleration across the vertical boundary is not an issue here . turbulence properties near the disk mid - plane , where the gravitational influence of the turbulent gas on the particles is strongest @xcite , appear insensitive to the choice of vertical boundary conditions @xcite , as we also demonstrate in section [ ss : bce ] . for large box simulations , concerns have been raised that artificial numerical diffusion varying with @xmath4 may occur due to the radial dependence of the shear velocity and thus corresponding truncation errors on a fixed grid @xcite . explicit treatment of the shear advection may be needed to eliminate these truncation errors . in the pencil code , such a technique , dubbed shear advection by fourier interpolation ( safi ) , has been implemented @xcite . in our work , we implement safi for all of our simulations to remove this undesirable numerical effect . to find the gravitational influence of the turbulent gas on the movement of solid particles , we need to solve the poisson equation for the fluctuation potential @xmath33 : @xmath34 where @xmath35 is the gravitational constant and @xmath36 is the density fluctuation with respect to the basic state of the gas stratification @xmath37 , which is given by equation ( [ e : rho_0 ] ) . note that we have neglected the self - gravity of the gas in equation ( [ e : mhd_mom ] ) on the assumption that the disk is gravitationally stable ( see ymm09 ) , so the solution to equation ( [ e : poisson ] ) does not affect the gas dynamics . although the vertical boundary conditions for the gas density are assumed to be periodic in the mhd equations ( see section [ ss : mhd ] ) , we find it inappropriate to assume the same in solving the poisson equation ( [ e : poisson ] ) . consider , for example , density perturbations near the top edge of the computational domain . if periodic boundary conditions in the vertical direction were adopted , particles near the mid - plane would experience the gravity of the same density perturbations at almost equal distance from above and below , resulting in cancellation of the gravitational force exerted by these perturbations on the particles . this would amount to undesirable , potentially large , numerical errors for calculating the gravity of density perturbations at high altitude . therefore , we instead adopt isolated or open boundary conditions in the vertical direction to solve equation ( [ e : poisson ] ) . with these boundary conditions , we assume that there are no fluctuations in density outside the vertical boundary , i.e. , @xmath38 for @xmath39 . the boundary conditions in the horizontal directions remain sheared periodic . to implement approximate isolated vertical boundary conditions with minimal computational cost , we use a variation of the @xcite fast algorithm to solve the poisson equation ( [ e : poisson ] ) , combining it with the fourier interpolation technique to achieve sheared periodicity @xcite . the essence of this algorithm is to double the computational domain in the @xmath30-direction by appending regions with zero density fluctuation for @xmath40 so the vertical domain now covers @xmath41 . ] and apply fast fourier transforms to this expanded domain . although the vertical boundary conditions remain periodic with fourier transforms , the @xmath42 nature of gravity means that influence from the periodic copies , now at least @xmath43 away , of the original computational domain of size @xmath44 onto itself is significantly reduced . we now describe the algorithm to solve equation ( [ e : poisson ] ) in detail . after expanding the vertical dimension of the computational domain and initializing the expanded region with zero density fluctuations , @xmath45 is fourier transformed in the @xmath5-direction into @xmath46 , where @xmath47 is the @xmath5-wavenumber . the result is phase shifted to recover periodicity in @xmath4-direction : @xmath48 $ ] , where @xmath49 is the time step . then @xmath50 is fourier transformed in the @xmath4-direction into @xmath51 , where @xmath52 is the @xmath4-wavenumber . for the modes @xmath53 , @xmath51 is fourier transformed in @xmath30-direction to find @xmath54 , where @xmath55 is the @xmath30-wavenumber . the fluctuation potential in fourier space @xmath56 is calculated using the usual convolution theorem : @xmath57 it is then inverse fourier transformed in @xmath55 to find @xmath58 . in principle , the mode @xmath59 can similarly be treated by fourier transforms in the vertical direction and convolved by equation ( [ e : convol ] ) as above . however , this mode has an analytical solution that we can employ with marginal computational cost to improve the solution to equation ( [ e : poisson ] ) . ) would yield a 50% relative error in the vertical acceleration due to a horizontal uniform slab at a distance @xmath60 away ( after domain doubling ) when compared to the exact acceleration derived from equation ( [ e : sheet ] ) . ] it represents a horizontally infinite , vertically thin mass layer of constant density at a given altitude @xmath30 , where the density is equal to the horizontal average of the density fluctuation at @xmath30 . ( the layer can be considered as either a slab of finite size @xmath61 and constant volume density @xmath62 or as an infinitely thin sheet of surface density @xmath63 ; the solutions outside the layer are the same by symmetry and gauss law . ) the gravitational acceleration due to this mode is constant above and below @xmath30 . therefore , the resultant fluctuation potential can be calculated by @xmath64 where we have arbitrarily defined the reference potential to be zero at each altitude . the discretized version of equation ( [ e : sheet ] ) reads @xmath65 equation ( [ e : dis_sheet ] ) remains exact for this mode given that we do not consider reconstruction of the density field within each cell . finally , we reverse the process to inverse fourier transform @xmath58 in @xmath52 and @xmath47 to derive the fluctuation potential in real space @xmath66 , incorporating the corresponding phase shifts with opposite sign . in practice , it is not necessary to allocate storage space for the whole extended domain in the @xmath30-direction . the appended zero density is only involved in the calculation between the forward and inverse fourier transforms in @xmath4 and @xmath5 . in addition , the convolutions in @xmath30 ( equations ( [ e : convol ] ) and ( [ e : dis_sheet ] ) ) for different modes @xmath67 are independent of each other . therefore , these convolutions can be distributed among processors and performed sequentially by allocating only a single one - dimensional working array along the @xmath30-direction . we continue to use the zero - mass approximation to model our solid particles as in ymm09 . in this approximation , particles behaves as test particles and only respond to the gravity of the host star and the gas . we ignore the viscous drag forces between particles and gas . this remains a good approximation for kilometer - sized planetesimals ( e.g. , * ? ? ? we also ignore mutual gravitational interactions between particles , which helps us isolate the net effect induced by hydromagnetic turbulence . the equations of motion for each particle , therefore , become [ e : eomp ] @xmath68 the vector @xmath69 is the position of the particle in the shearing box , while @xmath70 is the velocity of the particle relative to the background shear flow . in equation ( [ e : eomp2 ] ) , the terms inside the parentheses stem from the linearized gravity of the host star and the coriolis and centrifugal forces in the co - rotating frame of the shearing box . the remaining terms on the right - hand side represent the acceleration attributed to the gas : @xmath71 is the gravitational acceleration due to the basic state of the gas stratification and has the analytical expression @xmath72 while @xmath33 is the gravitational potential due to density fluctuations with respect to the basic state and is the solution of the poisson equation ( [ e : poisson ] ) ( section [ ss : poisson ] ) . note that by separating the density perturbation from the equilibrium density stratification and treating the gravity of the equilibrium state exactly , we improve the accuracy of the gravitational potential resulting from the turbulence . the position @xmath73 and velocity @xmath74 of each particle is evolved in time by integrating the equations of motion ( [ e : eomp ] ) simultaneously with the third - order runge - kutta steps advancing the mhd equations . in addition to the courant conditions set by the mhd equations , the time - step is limited by the absolute maximum of velocity @xmath75 so that no particles can cross more than half a grid zone each timestep . we calculate the gradient of the fluctuation potential @xmath76 on the grid and then quadratically interpolate it onto each particle . we uniformly distribute @xmath77 particles in a horizontal plane . we do not allow these particles to move until after 20 orbital periods , when the turbulence has reached a statistically steady state . the particles can have an initial eccentricity @xmath78 by setting an initial velocity @xmath75 so that they are at the apogee of their orbits ( see ymm09 ) . they can also have an initial inclination @xmath79 by placing the initial particle plane at a distance to the mid - plane . particles are wrapped around when crossing any of the six boundary planes of the shearing box . in this section , we present several properties of the magneto - rotational turbulence in our numerical models and discuss their convergence with the box size and the resolution . figure [ f : mri_t ] shows the horizontally averaged density fluctuation , inverse plasma @xmath80 , and @xmath81 parameter in the mid - plane ( @xmath82 ) as a function of time , for resolutions up to 64 points per @xmath22 and horizontal box dimensions up to @xmath83 . the magnitude of the density fluctuation is indicated by the rms value relative to the equilibrium density @xmath84 , the plasma @xmath85 is the ratio of the equilibrium thermal pressure to the magnetic pressure , and @xmath81 is calculated by ( e.g. , * ? ? ? * ) @xmath86 which includes both reynolds and magnetic stresses . denotes the horizontal average of the enclosed variable at any given altitude @xmath30 and thus is a function of @xmath30 . ] the magneto - rotational turbulence in the mid - plane of our models reaches a statistically steady state at about @xmath87 . in this saturated state , we see in general that the larger the horizontal size of the box , the smaller the amplitude of oscillation in the turbulence properties . little difference exists between an 8@xmath888@xmath884@xmath22 box and a 16@xmath8816@xmath884@xmath22 box at the same resolution , indicating convergence with horizontal box dimension . on the other hand , the amplitude of oscillation at saturation level increases with resolution for fixed box dimensions . there exists some evidence that the magnetic activity and thus the @xmath81 value increase with resolution , especially for the case of 2@xmath882@xmath884@xmath22 box , but this effect seems to become smaller with larger box sizes . we also note that there exists a curious trend of increase in density perturbation with time for the case of an 8@xmath888@xmath884@xmath22 box at the resolution of 64 points/@xmath22 . in spite of these trends , the time average of each turbulence properties remains roughly the same for different resolutions and thus shows convergence . we note that the turbulence properties at the saturated state in the mid - plane of our vertically stratified disks agrees with those in our unstratified disks of ymm09 ( see also * ? ? ? the means and standard deviations of these properties for each set of resolution and box dimensions are reported in table [ t : mri ] , which includes the magnetic stress and the reynolds stress as well . figure [ f : mri_z ] shows the time - averaged vertical profiles of density perturbation , inverse plasma @xmath80 , and @xmath81 parameter at saturation level over a period of @xmath89 . each of the three properties increases with @xmath90 , indicating stronger turbulence at higher altitude . the increasing activity with altitude is related to the increasing importance of the external magnetic field , which can be quantified by @xmath91 ( section [ ss : mhd ] ) . this agrees with the trend found in models of unstratified disks ( @xcite ; ymm09 ) . notice that @xmath92 at our highest altitude , and thus the corona is not modeled in our simulations . given that our vertical height only amounts to 2@xmath22 , the gas flow in our computational domain remains marginally stable against magnetic buoyancy during all stages of the development of the mri ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? as shown in figure [ f : mri_z ] , the time - averaged inverse plasma beta near the vertical boundary amounts to only about 0.4 , and thus on average , thermal pressure still dominates magnetic pressure there . we now report the response of zero - mass particles to the gravity of the density fluctuations of the statistically steady , numerically convergent magneto - rotational turbulence described in section [ s : turbulence ] . the reference time @xmath93 in the following discussion is the time at which the turbulent gas reaches its saturated state and the particles are allowed to move . as reported in section [ s : turbulence ] . see figure [ f : mri_t ] . ] the evolution of the mean orbital radius of one particle can be found by averaging the radial position @xmath4 of the particle over each epicycle motion . for the case of ideal unstratified disks , the distribution of particles in terms of the orbital radius change can be described by a time - dependent normal distribution centered at zero : @xmath94,\ ] ] where @xmath95 is the orbital radius change from the initial radius @xmath96 at @xmath93 and @xmath97 is the time - dependent standard deviation . we reported in ymm09 that @xmath97 depends on the properties of the gas disk and can be concisely expressed by @xmath98 where @xmath99 is a dimensionless proportionality constant as well as other dimensionless constants introduced in the following discussions may depend on the @xmath81 parameter . this dependency is not investigated in this work . ] and @xmath100 is a dimensionless quantity indicating the strength of the gas gravity . equations ( [ e : dist_dx ] ) and ( [ e : sigma_x ] ) demonstrate the diffusive nature of particle radial migration driven by magneto - rotational turbulence . for the case of ideal stratified disks studied here , we again find the evolution of particle orbital radius can be described by equations ( [ e : dist_dx ] ) and ( [ e : sigma_x ] ) . the dimensionless quantity @xmath101 is now defined by @xmath102 in terms of the mid - plane gas density @xmath20 , where @xmath103 is the toomre stability parameter for the gas . in table [ t : orbit ] , we report the measured values of the constant @xmath99 from our simulations at different resolutions and horizontal box sizes for the case of particles with zero initial inclination . comparing the value from the 2@xmath882@xmath884@xmath22 stratified model at a resolution of 64 points/@xmath22 with that from the unstratified model at the same resolution and horizontal box size ( see equation ( 15 ) of ymm09 ) , we find excellent agreement in the two @xmath99 values . as can be seen from table [ t : orbit ] and figure [ f : orbit ] , the value of @xmath99 remains roughly the same with different resolutions for fixed box dimensions ( except the anomaly shown by the 2@xmath882@xmath884@xmath22 box at a resolution of 64 points/@xmath22 ) . conversely , it is significantly dependent on the horizontal box size @xmath104 . the larger the size , the stronger the influence of the turbulence on the particle orbital radius . this relationship can be represented by the following power - law fit : @xmath105 we find no evidence of convergence with horizontal box size up to @xmath106 , the largest size we have investigated . equation ( [ e : sigma_x ] ) can be transformed into the diffusion coefficient @xmath107 introduced by @xcite for describing the radial random walks of orbiting particles induced by turbulent torques , in terms of the keplerian orbital angular momentum @xmath108 . the reader is referred to ymm09 for a detailed description of this procedure . we emphasize here that this transformation involves no assumption about the correlation time of the stochastic torques and is thus a direct measurement of @xmath107 . using a heuristic choice of dimension for the diffusion coefficient , @xcite defined a dimensionless parameter @xmath109 to represent the magnitude of @xmath107 . we report our measured values of @xmath109 in table [ t : orbit ] . the dependence of @xmath109 on the horizontal box size @xmath1 in our models for @xmath110 is shown in figure [ f : epsilon_gamma ] and can be written as @xmath111 the amplitude of each epicyclic oscillation of a particle gives the instantaneous eccentricity of the particle orbit . we reported in ymm09 that in ideal unstratified disks , the distribution of particles in terms of the eccentricity deviation should be a time - dependent normal distribution centered at zero , as long as the particles have non - negligible initial eccentricity ( c.f . equation ( [ e : dist_dx ] ) ) : @xmath112,\ ] ] where @xmath113 is the eccentricity deviation from the initial eccentricity @xmath78 at @xmath93 and the time - dependent standard deviation @xmath114 can be written as @xmath115 where @xmath116 is a dimensionless proportionality constant . we emphasize that the time - dependent rayleigh distribution . ] found for particles with zero ( or negligible ) initial eccentricity @xmath117\ ] ] is a manifestation of equation ( [ e : dist_de ] ) since the eccentricity @xmath118 is a positive definite quantity . equations ( [ e : dist_de ] ) and ( [ e : dist_e ] ) share the same time - dependent parameter @xmath114 , and there exists no evidence that @xmath114 depends on the initial eccentricity @xmath78 . we find that the same evolution of particle distribution in eccentricity deviation also holds for the case of ideal stratified disks . the measured values of the constant @xmath116 for different resolutions and horizontal box sizes when particles have zero initial inclination are listed in table [ t : orbit ] . the same comparison performed in section [ ss : radius ] for the orbital radius evolution indicates that the eccentricity evolution in a stratified disk again agrees very well with that in an unstratified disk ( see equation ( 17 ) of ymm09 ) . these comparisons demonstrate that a local unstratified disk model gives results consistent with the mid - plane of a local stratified disk model that has the same physical conditions except vertical stratification . as in the case of orbital radius discussed in section [ ss : radius ] , the eccentricity evolution of the particles does not noticeably depend on the resolution for given box dimensions , while it is strongly affected by the horizontal box size ( table [ t : orbit ] and figure [ f : orbit ] ) . a power - law regression gives @xmath119 which is close to a linear relation to the horizontal box size . we discuss the box - size effect on both the orbital radius and the eccentricity in section [ s : box_size ] . the eccentricity driven by magneto - rotational turbulence enhances orbital crossing among planetesimals and thus increases the chance of collisions between them . @xcite defined a dimensionless parameter @xmath120 to represent the strength of this effect . equation ( [ e : sigma_e ] ) can be used to estimate the value of @xmath120 , and the reader is referred to ymm09 for a detailed description of this procedure . we report in table [ t : orbit ] our measured values of @xmath120 . as shown in figure [ f : epsilon_gamma ] , the dependence of @xmath120 on the horizontal box size @xmath1 in our models for @xmath110 can be described by @xmath121 the only orbital property of a particle that can not be measured in an unstratified disk model is the inclination @xmath122 . in a stratified disk , vertical linear gravity from the host star ( the third term in the parentheses on the right - hand side of equation ( [ e : eomp2 ] ) ) provides a restoring force such that particles oscillate in the @xmath30-direction about the mid - plane , as demonstrated in figure [ f : zpt1 ] . note that because the particles also experience the gravity of the gas , the period of the oscillation is @xmath123 in the linear limit ( derived from equations ( [ e : eomp2 ] ) and ( [ e : g_0 ] ) ) , which is slightly shorter than the orbital period @xmath124 we can calculate the induced inclination ( in radians ) for a single particle by @xmath125 , in which @xmath126 and @xmath127 are the maximum and the minimum vertical positions in one oscillation , respectively , provided that @xmath128 . figure [ f : inc](a ) shows the histograms of particles with zero initial inclination sorted in bins of instantaneous inclination at three different times . similar to the eccentricity distribution for particles with @xmath129 described by equation ( [ e : dist_e ] ) , the inclination distribution resembles a time - dependent rayleigh distribution @xmath130,\ ] ] where @xmath131 is a time - dependent parameter denoting the width of the distribution . one might expect the inclination distribution of particles with nonzero initial inclination would evolve as a time - dependent normal distribution with fixed center and increasing width , similar to the eccentricity distribution of particles with nonzero initial eccentricity . however , figure [ f : inc](b ) shows different behavior : the distribution increasingly deviates from a normal distribution with time . the peak leans toward the mid - plane and asymmetry develops with more particles on the left side of the peak ( i.e. , smaller inclination ) . this indicates the diffusion is stronger near the mid - plane , which in fact is consistent with vertical stratification of the gas density . it would be interesting to measure the dependence of the diffusion coefficient on vertical height , but given our limited sample of initial inclinations , this remains to be investigated . we therefore focus our attention on the case of particles with zero initial inclination . figure [ f : sigma_i ] shows @xmath131 for disks with varying gravity parameter @xmath101 and particles with varying initial eccentricity @xmath78 but zero initial inclination @xmath132 . note that when @xmath131 is normalized by @xmath133 , all the curves roughly coincide . this indicates that @xmath131 is linearly dependent on both @xmath133 , but independent of @xmath78 , similar to what we reported for the evolution of the mean radius and the eccentricity . the results can thus be summarized by the following expression : @xmath134 where @xmath135 is a dimensionless proportionality constant . table [ t : orbit ] lists our measured values of the constant @xmath135 for different resolutions and box dimensions when particles have zero initial inclination . we note that in contrast to the orbital radius and the eccentricity , the evolution of the orbital inclination is not significantly affected by the horizontal box size . the inclination of the particle orbits is mainly affected by the vertical structures of the density field , which in turn is related to the vertical scale height of the density stratification . the horizontal structures in the magneto - rotational turbulence may have little correlation with the vertical structures so that the horizontal box size has little effect on the vertical motions of the particles . finally , all three components of the velocity dispersion @xmath136 among the particles as a function of time can be measured in our stratified disk models . similarly to what we reported in ymm09 , each component assumes the same form : @xmath137 where the index @xmath122 is either @xmath4 , @xmath5 , or @xmath30 and @xmath138 is the corresponding dimensionless proportionality constant . we emphasize that @xmath139 always holds because of the fixed ratio of amplitudes in @xmath4 and @xmath5 directions in the epicycle motions of the particles , which has been verified in our simulations . the values of @xmath140 and @xmath141 at different resolutions and box dimensions for the case of particles with zero initial inclination are listed in table [ t : orbit ] . note that the value of @xmath140 for the 2@xmath882@xmath884@xmath22 box at a resolution of 64 points/@xmath22 is consistent with that from the unstratified disk with the same horizontal size and resolution we reported previously ( see equation ( 18 ) of ymm09 ) , a further confirmation of the consistency between stratified and unstratified models discussed in sections [ ss : radius ] and [ ss : ecc ] . as shown by figure [ f : vel_disp ] , we do not see evident dependence of velocity dispersion on resolution of our models . on the other hand , the horizontal component @xmath140 ( and thus @xmath142 ) significantly depends on the horizontal box size @xmath1 while the vertical component @xmath141 does not . this is in accordance with the dependence of the three orbital properties found in previous subsections . from our measured values for @xmath143 , we quantify @xmath140 and @xmath141 with the following expressions : @xmath144 in order to confirm that our numerically convenient but perhaps physically inconsistent assumption of vertical field and periodic gas boundary conditions in the @xmath30-direction does not change our results significantly , we compare turbulence properties and particle dynamics from models with different vertical boundary conditions for the magnetic fields run at a lower resolution of 16 points per disk scale height @xmath22 in large boxes with size @xmath145 . the gas density and velocity remain periodic . figure [ f : mri_bcz ] shows that within about one disk scale height , the vertical structure of the rms density perturbation , the plasma @xmath80 , and the @xmath81 parameter are in quantitatively close agreement between the models . note that the model with periodic boundary conditions creates a coronal region as low as @xmath146 , where @xmath147 . more importantly , figure [ f : orbit_bcz ] shows from top to bottom the time evolution in the standard deviations of particle orbital radius , eccentricity and inclination for the two models . the orbital radius and eccentricity agree very well irrespective of the vertical boundary conditions , and the effect on inclination differs by only @xmath14830% , likely due to increased activity at high altitudes for the model with periodic magnetic fields . thus , we believe that our results on particle dynamics are robust . we have demonstrated in section [ s : turbulence ] that various properties of the saturated magneto - rotational turbulence in our simulations converge with both resolution and box dimensions . however , while converging with resolution , the response of massless particles to the gravity of the turbulent gas does not similarly converge with the horizontal box size , as presented in section [ s : particles ] . this result raises serious questions about the validity of using the local - shearing - box approximation to simulate the dynamics of any particles under gravitational influence of magneto - rotational turbulence . in order to understand the lack of convergence with box dimension in the orbital evolution of massless particles , we further study the properties of the gravitational force exerted by the turbulent gas . table [ t : force ] lists the radial and the azimuthal components of the gravitational force exerted by the turbulent gas , @xmath149 and @xmath150 , respectively . since the force is stochastic , we ensemble average the root mean square of each force component over all particles , and then report it as time average and @xmath151 variation scaled by @xmath152 , where @xmath153 is the mass of a particle . note that in our models , particles only follow the gravitational potential generated by the gas and the host star , so the acceleration of each particle is independent of its mass . this is an advantage in that the measurements reported in this paper can be used to gauge the gravitational influence of magneto - rotational turbulence on solid particles or planetary objects of any mass and thus its significance compared with other relevant interactions . as can be seen in table [ t : force ] , both the radial and the azimuthal components of the force generally increase with the horizontal box size . it is not clear if there exists any convergence in the radial component towards larger box size , since the 16@xmath8816@xmath884@xmath22 box at a resolution of 32 points per @xmath22 is the only model that contradicts the general increasing trend . certainly , the azimuthal component ( i.e. , the torque ) shows no sign of convergence up to the largest box size we have investigated , with a regression power - law index of about 0.6 . @xcite also reported in their recent local models a slight trend of increasing torque with increasing horizontal box size ( see their figure 8 and the corresponding discussion ) . since the density structure of the turbulent gas determines its gravitational force on the particles , it is informative to consider how the structure changes with resolution and horizontal box size . the density structure of the gas is best analyzed in fourier space , using the fourier amplitudes @xmath154 defined in section [ ss : poisson ] . is strictly periodic in sheared coordinates and the wavenumber @xmath155 is slightly different from that of a direct fourier decomposition of @xmath45 ; see section [ ss : poisson].[f : wavenumber ] ] figure [ f : fourier ] plots the time - averaged fourier amplitudes of the gas density in the mid - plane @xmath156 along either the radial or the azimuthal direction . the top - left panel shows the amplitude as a function of @xmath52 for @xmath157 . in general , the largest amplitude resides at the longest wavelength while monotonically decreasing with increasing wavenumber @xcite . for any given box dimensions , increasing resolution has little effect on each amplitude , apart from extending the profile toward higher wavenumber . flattening of the profile for wavelengths @xmath1588@xmath22 is hinted for 16@xmath8816@xmath884@xmath22 boxes . we find no evidence of a downward turn toward even longer wavelength . on the other hand , increasing the horizontal box size @xmath1 lowers the overall profile of the fourier amplitudes . interestingly , the amplitude of the longest wavelength mode @xmath159 remains roughly constant . the bottom - left panel of figure [ f : fourier ] plots the summation of fourier amplitudes over all @xmath47 at any given @xmath52 . in contrast to the amplitudes for @xmath157 , the horizontal box size @xmath1 has little effect on the amplitudes for most of the wavelengths . nevertheless , increasing resolution indeed increases the inertial range and resolves more power toward shorter wavelength . the right - hand column of figure [ f : fourier ] plots the azimuthal counterpart of the left - hand column . we find similar features in the azimuthal amplitudes as those found in the radial ones . the only difference is that the flat part of the spectrum in the azimuthal direction is short compared with that in the radial direction , as can be seen in the bottom - right panel . power that is not captured in the long - wavelength range of the spectrum in the 2@xmath882@xmath884@xmath22 boxes is probably the cause for artificially higher amplitude at @xmath160 ( the top - right panel ) , which may in turn be responsible for the anomaly found in the particle dynamics shown in section [ s : particles ] . the fourier amplitudes shown in figure [ f : fourier ] help us understand the relationship between the gravitational force exerted by the gas and the horizontal box size . the radial and the azimuthal force components for any given ( horizontal ) fourier mode @xmath161 are related to the corresponding fourier amplitude @xmath162 by ( see equation ( [ e : convol ] ) ) @xmath163 where we have used the facts that the dominant modes in the radial and the azimuthal directions are the longest wavelength ones @xmath164 and @xmath165 , and both are about constant irrespective of resolution and box size . therefore , both components should be at most linearly proportional to the horizontal box size in our simulations . although involving some simplifications , equations ( [ e : fx ] ) and ( [ e : fy ] ) at least qualitatively explain the general trend of increasing stochastic force with horizontal box size . we next evaluate the correlation time @xmath166 of the stochastic torque . this quantity is in fact not well defined in the literature , and different authors have different preferences for its evaluation . for the purpose of consistency with the framework of radial diffusive migration of particles driven by random torque , we adopt the definition of @xcite : @xmath167 where @xmath107 is the radial diffusion coefficient as a function of the keplerian orbital angular momentum @xmath108 , @xmath168 is the fluctuating part of the torque , and the overline denotes an ensemble average . we repeat the relationship between @xmath107 and the dimensionless parameter @xmath109 here ( see * ? ? ? * ) : @xmath169 to calculate the correlation time @xmath166 with our definition , both the diffusion coefficient @xmath107 and the magnitude of the stochastic torque are needed . as emphasized in ymm09 , we have all the information about the particle distribution against each orbital property as a function of time , and thus we can directly evaluate the diffusion coefficient in the context of a diffusion equation , which is reported in section [ ss : radius ] and table [ t : orbit ] . furthermore , we also have the ensemble - averaged rms value of the stochastic torque , which is reported in section [ ss : force ] and table [ t : force ] . equations ( [ e : ct1 ] ) and ( [ e : diff_coeff ] ) can then be used to calculate the correlation time @xmath166 , and the result is listed in table [ t : force ] . as indicated in table [ t : force ] , the correlation time of the stochastic torque @xmath166 does not appreciably depend on the resolution but it is significantly affected by the horizontal box size . the larger the horizontal box size , the longer the correlation time of the torque ; this trend was also reported by @xcite . up to the largest box size we have investigated ( @xmath170 ) , it is not clear if @xmath166 will converge with @xmath1 . this behavior is similar to that of the torque magnitude , as discussed in section [ ss : force ] . as a side note , one may want to evaluate the correlation time of the stochastic torque @xmath166 without referring to stochastic particle orbital evolution and invert equation ( [ e : ct1 ] ) to estimate the radial diffusion coefficient @xmath107 indirectly . we have shown in ymm09 , for the case of ideal unstratified disks , that the following formula , motivated by the definition of @xmath107 under the framework of the fokker - planck formalism , serves this purpose : @xmath171 where @xmath172 is the autocorrelation function of the stochastic torque as a function of the time lag @xmath173 . here we repeat the exercise and use equation ( [ e : ct2 ] ) to estimate @xmath166 for the case of ideal stratified disks . the result is listed in table [ t : force ] and shows remarkable consistency with the values obtained directly from the evolution of particle orbital distribution . therefore , equation ( [ e : ct2 ] ) remains a useful estimator of the correlation time @xmath166 . to further understand the trend of increasing correlation time with increasing horizontal box size , we plot the autocorrelation function of the stochastic torque in figure [ f : acf ] . as also demonstrated by @xcite , the larger the box , the less oscillatory the autocorrelation function is and the less pronounced is the negative contribution to the radial diffusion of the particles . this behavior in turn is consistent with increasing values of the correlation time and the radial diffusion coefficient . in addition to the stochastic torque , a similar analysis may be conducted for the radial component of the gravitational force exerted by the gas . we find that the correlation time of the radial force also increases with the horizontal box size . however , our models indicate a much longer correlation timescale , which may amounts to several tens of orbital periods . given that our simulations only lasted for about 100@xmath124 , the magnitude of the correlation time is not well calculated and much longer simulations may be required to cover the full spectrum of the autocorrelation function for the radial force . @xcite have discussed the dominant role of the longest wavelength , purely radial fourier mode @xmath174 in great detail and suggested that these kind of persistent structures are similar to the zonal flows found in many other astrophysical systems . as noted by the authors , these axisymmetric structures can not generate torques that affect the orbital radius of the embedded particles . while the radial forces due to the zonal flows can still affect the eccentricity of the particles , their long correlation time indicates that they may not be responsible for the diffusive evolution of the particle eccentricity , as shown in ymm09 and section [ ss : ecc ] . on the other hand , the longest wavelength , purely azimuthal fourier mode @xmath175 in the gas structure is independent of time ; see footnote [ f : wavenumber ] and section [ ss : poisson ] . ] is significant enough that both the radial diffusive migration and the eccentricity evolution of the particles are affected by the large - scale azimuthal density perturbations and thus the box dimensions . as described in section [ s : force ] , the magnitude and the correlation time of the gravitational force exerted by the turbulent gas correlates with the horizontal size of our local shearing box . this behavior essentially explains the correlation of the stochastic evolution of the particle orbital radius and eccentricity with the horizontal box size reported in section [ s : particles ] . these results challenge the validity of the local shearing box to describe large - scale density structures in magneto - rotational turbulence . by measuring the two - point correlation functions of density , velocity , and magnetic field in magneto - rotational turbulence , @xcite found extended density features in contrast to well localized magnetic structures . the authors suggested that the density features are propagating acoustic waves excited by the turbulence ( see also * ? ? ? * ; * ? ? ? it is not clear yet what the dissipation length scale for these waves is , which is crucial to understanding whether a local shearing box can capture the largest - scale structures in the density fluctuations excited by magneto - rotational turbulence . on the other hand , a global disk model may require high resolution in order to self - consistently produce large - scale structures . for example , @xcite argued that the inverse cascade of magnetic energy from small scales to large scales might be responsible for ultimately launching zonal flows . to confirm this mechanism in a global context , a model that is capable of resolving at least the correlation lengths in the magnetic structures might be necessary . given the correlation with horizontal box size reported in this work , it is not clear how a local shearing box can be connected to a global disk model so that both models give consistent results on the particle dynamics under the gravitational influence of density fluctuations in magneto - rotational turbulence . nevertheless , we conjecture at this point that the criterion @xmath0 might provide a physical length scale for a local shearing box . at this scale , the local - shearing - box approximation formally breaks down since it assumes @xmath176 . the curvature terms become important and this might trigger turbulent eddies to damp propagating waves . if this conjecture could be verified , equations ( [ e : cx ] ) , ( [ e : ce ] ) and ( [ e : sx ] ) would prove useful in evaluating the gravitational influence of the turbulent gas on particle dynamics . substituting @xmath2 for @xmath1 in these equations , our models predict that @xmath177 @xcite presented the most recent work on stochastic planetesimal orbital evolution in a global disk model . their measurement of the orbital radius and radial velocity dispersion can be directly compared with ours . these authors reported in their global model g3 , whose @xmath178 was the closest to ours , that @xmath179 and @xmath180 . in this specific model , a disk aspect ratio . ] of @xmath181 and a strength of gas gravity of @xmath182 at 5 au were adopted . in comparison , equations ( [ e : sigma_x ] ) , ( [ e : sigma_u ] ) , and ( [ e : cs_predict ] ) give @xmath183 and @xmath184 . our results are only about a factor of two lower than those of @xcite . therefore , the criterion @xmath185 for a local shearing box seems to offer some consistency between global and local models . these comparisons are not meant to be accurate and conclusive , though , since the proposed criterion @xmath0 invites significant uncertainty . further investigation into the large - scale structures of magneto - rotational turbulence and the discrepancy between global and local disk models is still needed . if the criterion @xmath185 holds , equations ( [ e : epsilon ] ) and ( [ e : gamma ] ) imply @xmath186 consider a region at @xmath187 au in a minimum - mass solar nebula ( mmsn ; * ? ? ? * ) , where @xmath188 . equations ( [ e : eg_predict ] ) give @xmath189 and @xmath190 . this value of the @xmath109 parameter is about one order of magnitude higher than was reported in ymm09 for the case of a 2@xmath882@xmath882h unstratified box , while the value of the @xmath120 parameter is only about a factor of three larger . note that according to equations ( [ e : eg_predict ] ) , both @xmath109 and @xmath120 decrease with increasing disk aspect ratio @xmath191 , which often increases with increasing radial distance @xmath2 . we suggested in ymm09 that in a typical protoplanetary disk , the radial diffusive migration of protoplanets induced by magneto - rotational turbulence may be unimportant compared to secular migration . according to @xcite , for the diffusive migration to be able to dominate over type i migration , the @xmath109 parameter should be greater than or about 0.11 for the mmsn . our new measurement remains orders of magnitude smaller than this transition value . therefore , our previous conclusions on the insignificance of planetary diffusive migration may still hold even though @xmath109 is increased by an order of magnitude . however , note that recent calculations for type i migration in non - isothermal disks ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) may negate the dominance of secular migration over stochastic migration driven by magneto - rotational turbulence . we also argued in ymm09 that kilometer - sized planetesimals moving in magneto - rotational turbulence survive mutual collisional destruction , except in the inner region of a young protoplanetary disk . this was based on the results of @xcite for the cases of @xmath192 and @xmath193 . since our new measurement does not fall outside this range , the same conclusion still applies . in contrast , @xcite argued that the velocity dispersion of kilometer - sized planetesimals excited by magneto - rotational turbulence might be so large that these planetesimals should be erosive to each other . before a consistent scenario for the survivability of planetesimals could be assembled , however , several factors remain to be accounted for . first , eccentricity damping due to tidal interaction between a planetesimal and its surrounding gas acts to circularize the orbits of the planetesimals ( e.g. , * ? ? ? this mechanism was absent in the models of @xcite and thus their measurement of the velocity dispersion of planetesimals should be considered as an upper limit . on the other hand , using analytical arguments to estimate the equilibrated eccentricity of planetesimals , @xcite took three possible eccentricity damping effects into account tidal interaction , gas drag , and inelastic collisions and thus their estimate of the velocity dispersion of planetesimals might be more realistic . even though significant uncertainty was involved in the assumed strength of turbulent excitation of eccentricity in the models of @xcite , it could be improved by more accurate measurement of the @xmath120 parameter , as provided by the present work . other interesting progress in the study of mutual collisions of planetary objects includes different outcomes from different impact angles ( head - on vs. hit - and - run scenarios ; see , e.g. , * ? ? ? * ; * ? ? ? * ) and improved calculations on material properties ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? given all these new developments , a reconsideration of the survivability of kilometer - sized planetesimals moving in magneto - rotational turbulence seems warranted . directly extending our previous publication ( ymm09 ) , we continue to study massless particles moving under the gravitational influence of density fluctuations due to saturated magneto - rotational turbulence in a local , isothermal , keplerian gas disk . we include linearized vertical gravity from the host star and thus vertical stratification of the gas disk . for comparison , the conditions in the mid - plane of the vertically stratified disks are exactly the same as those in the unstratified disks of ymm09 . in order to accurately measure the gravitational effect of the turbulent gas , we separate the gas density @xmath8 into two components : the basic state @xmath37 for the vertical hydrostatic equilibrium ( equation ( [ e : rho_0 ] ) ) and the density deviation @xmath194 from this basic state . we use the exact gravitational acceleration due to the basic state ( equation ( [ e : g_0 ] ) ) and only solve the poisson equation for the gravitational potential due to the density deviation ( equation ( [ e : poisson ] ) ) . we emphasize that since the poisson equation is linear in density , this approach does not assume small density fluctuations . furthermore , we implement isolated boundary conditions in the vertical direction and thus any density fluctuation outside the vertical computational domain is neglected . by imposing a weak , uniform external magnetic field , we maintain a constant level of saturated magneto - rotational turbulence in the disk mid - plane . several turbulence properties demonstrate convergence with both resolution up to 64 points per disk scale height @xmath22 and horizontal box size up to 16@xmath22 . the @xcite @xmath81 parameter in the mid - plane of our models is controlled at the level of @xmath14810@xmath195 . however , even though the properties of the turbulent gas appear numerically convergent , the dynamics of massless particles moving under the gravity of this turbulent gas does not converge with the horizontal box size @xmath1 . the larger the horizontal box size , the stronger the gravitational effect of the gas on the particles . specifically , the evolution of the orbital radius , the eccentricity , and the horizontal velocity dispersion of the particles is roughly linearly dependent on @xmath1 up to 16@xmath22 . this trend was also found in our unstratified models ( ymm09 ) , and we find consistency between the unstratified models and the mid - plane of the stratified models . in contrast to the horizontal components of the particle movement , we find that the evolution of the inclination and the vertical velocity dispersion is not significantly affected by @xmath1 . the dependence of particle dynamics on the horizontal box size can be traced back to the density structure of the gas . consistent with the large - box models studied by @xcite , the longest wavelength fourier mode dominates the density spectrum along the radial direction in our models . furthermore , we find that the longest wavelength fourier mode in the azimuthal direction also strongly influences the particle dynamics , leading to the diffusive evolution of both the orbital radius and the eccentricity of the particles . the spectral amplitudes of these longest wavelength modes are roughly constant against the horizontal box size . using a simple single - mode analysis , we show that the linear dependence of the particle response is a natural outcome of these findings for the density spectrum . correlation of particle dynamics with box size poses a major difficulty for the interpretation of local - shearing - box models involving gravitational physics of magneto - rotational turbulence . we can nevertheless conjecture that @xmath185 , where @xmath2 is the distance of the box center to the host star , might be a natural scale of choice for a local model to approach reality . if this conjecture holds , we find that our previous conclusions in ymm09 on the unimportance of radial diffusive migration for protoplanets as well as the survivability of kilometer - sized planetesimals under collisional destruction may still be valid . ultimately , high - resolution global disk models and detailed comparisons with large - box local models might be necessary to settle this issue . we thank the anonymous referee for critical comments that significantly improved the rigor and clarity of this paper , and thank charles gammie , anders johansen , roman rafikov , clment baruteau , martin pessah , richard nelson , and oliver gressel for their insightful discussion on this research . this research was supported in part by the perimeter institute for theoretical physics . resources supporting this work were provided by the nasa high - end computing ( hec ) program through the nasa advanced supercomputing ( nas ) division at ames research center . partial support of this work was provided by the nasa origins of solar systems program under grant nnx07ai74 g . rcccccc 2@xmath882@xmath884 & 16 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 4@xmath884@xmath884 & 16 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 8@xmath888@xmath884 & 16 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 16@xmath8816@xmath884 & 16 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 2@xmath882@xmath884 & 32 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 4@xmath884@xmath884 & 32 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 8@xmath888@xmath884 & 32 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 16@xmath8816@xmath884 & 32 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 2@xmath882@xmath884 & 64 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 4@xmath884@xmath884 & 64 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 8@xmath888@xmath884 & 64 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] rcccccccc 2@xmath882@xmath884 & 16 & 2.4(-4 ) & 2.4(-4 ) & 2.0(-4 ) & 3.6(-4 ) & 2.1(-4 ) & 8.3(-5 ) & 1.2(-4 ) + 4@xmath884@xmath884 & 16 & 4.4(-4 ) & 3.5(-4 ) & 2.9(-4 ) & 5.0(-4 ) & 3.0(-4 ) & 2.9(-4 ) & 1.7(-4 ) + 8@xmath888@xmath884 & 16 & 1.3(-3 ) & 7.4(-4 ) & 2.8(-4 ) & 1.1(-3 ) & 2.9(-4 ) & 2.6(-3 ) & 3.7(-4 ) + 16@xmath8816@xmath884 & 16 & 3.0(-3 ) & 1.6(-3 ) & 2.4(-4 ) & 2.4(-3 ) & 3.3(-4 ) & 1.4(-2 ) & 8.0(-4 ) + 2@xmath882@xmath884 & 32 & 2.6(-4 ) & 2.7(-4 ) & 2.6(-4 ) & 3.8(-4 ) & 2.2(-4 ) & 1.0(-4 ) & 1.3(-4 ) + 4@xmath884@xmath884 & 32 & 3.7(-4 ) & 3.0(-4 ) & 2.5(-4 ) & 4.3(-4 ) & 2.3(-4 ) & 2.1(-4 ) & 1.5(-4 ) + 8@xmath888@xmath884 & 32 & 1.1(-3 ) & 6.4(-4 ) & 1.7(-4 ) & 9.3(-4 ) & 2.4(-4 ) & 1.7(-3 ) & 3.2(-4 ) + 16@xmath8816@xmath884 & 32 & 2.3(-3 ) & 1.3(-3 ) & 2.2(-4 ) & 1.9(-3 ) & 3.0(-4 ) & 7.8(-3 ) & 6.4(-4 ) + 2@xmath882@xmath884 & 64 & 4.0(-4 ) & 4.1(-4 ) & 4.5(-4 ) & 6.1(-4 ) & 4.7(-4 ) & 2.4(-4 ) & 2.0(-4 ) + 4@xmath884@xmath884 & 64 & 4.2(-4 ) & 3.3(-4 ) & 2.6(-4 ) & 4.7(-4 ) & 2.8(-4 ) & 2.7(-4 ) & 1.6(-4 ) + 8@xmath888@xmath884 & 64 & 1.1(-3 ) & 6.6(-4 ) & 2.3(-4 ) & 9.6(-4 ) & 3.0(-4 ) & 1.9(-4 ) & 3.2(-4 ) rccccc 2@xmath882@xmath884 & 16 & [email protected] & [email protected] & 0.016 & 0.020 + 4@xmath884@xmath884 & 16 & [email protected] & [email protected] & 0.019 & 0.014 + 8@xmath888@xmath884 & 16 & [email protected] & [email protected] & 0.067 & 0.067 + 16@xmath8816@xmath884 & 16 & [email protected] & [email protected] & 0.130 & 0.120 + 2@xmath882@xmath884 & 32 & [email protected] & [email protected] & 0.023 & 0.029 + 4@xmath884@xmath884 & 32 & [email protected] & [email protected] & 0.020 & 0.014 + 8@xmath888@xmath884 & 32 & [email protected] & [email protected] & 0.070 & 0.065 + 16@xmath8816@xmath884 & 32 & [email protected] & [email protected] & 0.119 & 0.112 + 2@xmath882@xmath884 & 64 & [email protected] & [email protected] & 0.033 & 0.022 + 4@xmath884@xmath884 & 64 & [email protected] & [email protected] & 0.020 & 0.022 + 8@xmath888@xmath884 & 64 & [email protected] & [email protected] & 0.068 & 0.070

the lieb - mattis theorem about antiferromagnetic ordering of energy levels on bipartite lattices is generalized to finite - size two - leg spin-1/2 ladder model frustrated by diagonal interactions . for reflection - symmetric model with site - dependent interactions we prove exactly that the lowest energies in sectors with fixed total spin and reflection quantum numbers are monotone increasing functions of total spin . the nondegeneracy of most levels is proved also . we also establish the uniqueness and obtain the spin value of the lowest - level multiplet in the whole sector formed by reflection - symmetric ( antisymmetric ) states . for a wide range of coupling constants , we prove that the ground state is a unique spin singlet . for other values of couplings , it may be also a unique spin triplet or may consist of both multiplets . similar results have been obtained for the ladder with arbitrary boundary impurity spin . some partial results have also been obtained in the case of periodical boundary conditions .

cond-mat0702148

nowadays , the frustrated spin systems are the subject of intensive study . @xcite the interest on them is stimulated by recent progress in synthesizing corresponding magnetic materials . @xcite in these models , due to competing interactions , the classical ground state can not be minimized locally and usually possesses a large degeneracy . the frustration can be caused by the geometry of the spin lattice or by the presence of both ferromagnetic and antiferromagnetic interactions . in contrast , the ground states of classical models on bipartite lattices are nel ordered and are unique up to the global spin rotations . the bipartitness means that the lattice can be divided into two sublattices @xmath0 and @xmath1 , so that all interactions within the same sublattice are ferromagnetic while the interactions between different sublattices are antiferromagnetic . for the nel state , all spins of @xmath0 have same direction while all spins of @xmath1 are aligned in opposite direction . the quantum fluctuations destroy nel state and the quantum model , in general , has more complicated ground state . however , for bipartite spin systems , the quantum ground state inherits some properties of its classical counterpart . in particular , lieb and mattis proved that the quantum ground state of finite - size system is a unique multiplet with total spin @xmath2 , which coincides with the spin of the classical nel state . here , @xmath3 and @xmath4 are the highest possible spins on corresponding sublattices . @xcite this feature of bipartite systems looks natural , since in the limit of large spin values the quantum model approaches to the classical one . moreover , a simple general rule that the energy increases with increasing spin , which is conventional for classical antiferromagnets , makes a sense in this case , too . more precisely , the lowest energy @xmath5 in the sector , where the total spin is equal to @xmath6 , is a monotone increasing function of the spin for any @xmath7 . @xcite this property is known as lieb - mattis theorem about antiferromagnetic ordering of energy levels . lieb - mattis theorem is very important because it provides information about the ground state and spectrum of bipartite spin systems without exact solution or numerical simulation . recently , it has been generalized to su(@xmath8 ) symmetric quantum chain with defining representation and nearest - neighbor interactions . @xcite a ferromagnetic ordering of energy levels has been formulated and proven also for xxx heisenberg chains of any spin . @xcite more limited number of exact results is available for frustrated quantum spin systems . it is difficult to treat even their classical ground states as was mentioned above . the usual lieb - mattis theorem is no longer valid for frustrated systems . recently , however , lieb and schupp proved rigorously that a reflection - symmetric spin system with antiferromagnetic crossing bonds possesses at least one spin - singlet ground state . @xcite moreover , under certain additional conditions , all ground states become singlets . a lot of frustrated spin systems satisfy the conditions of lieb - schupp approach . two - dimensional pyrochlore antiferromagnet having checkerboard lattice structure is an example of such type of system . @xcite another example is well - known two - dimensional antiferromagnetic @xmath9 model , namely , the heisenberg model on square lattice with diagonal interactions , which has been studied recently . @xcite note that this approach does not provide any information about the degeneracy of ground states . thus , for frustrated spin systems , this question still remains open . we mention also that the method used in refs . and is restricted to the systems , which do not have a spin on their symmetry axis . at the same time , recently , certain signs in favor of lieb - mattis theorem for frustrated spin systems have appeared . for @xmath9 model , the number of arguments based on exact diagonalization and spin - wave approximation had been presented in support of the fact that the antiferromagnetic ordering of energy levels is preserved under weak frustration caused by diagonal couplings . @xcite recently , for the same model , the condition @xmath2 for the ground state spin has been tested numerically in ref . . the authors came to the conclusion that it remains true provided that the frustration is sufficiently weak in order to destroy ground - state nel order . the lieb - mattis property had been observed also at finite size spectrum of the heisenberg model on the triangular and kagome lattices . @xcite all these investigations suggest that in many cases , even a relatively strong frustration can not destroy the antiferromagnetic ordering of energy levels . on the other hand , recently , the spin ladder systems have attracted much attention . the interest to them has been generated by significant progress made within the last years in fabrication of such type of compounds having a structure similar to the structure of two - dimensional high - temperature superconductors . @xcite ladder systems are simpler for study than their more complex two - dimensional counterparts . moreover , the powerful theoretical and numerical methods , elaborated for one - dimensional models , can be applied to study them . the two - leg ladder frustrated by diagonal interactions is one of the simplest frustrated spin systems . it can be viewed as a quasi - one - dimensional analog of @xmath9 model . together with various generalizations , it has been investigated intensively during the last decade . @xcite the ground - state phase diagram consists of two topologically distinct haldane phases separated by the curve ( or surface ) of phase transition . inspired by aforementioned activities , in this paper we generalize the lieb - mattis ordering theorem to two - leg spin-1/2 ladder model with diagonal interactions . we consider the system , which rests invariant under the reflection with respect to the longitudinal symmetry axis of the ladder . up to our knowledge , this is the first exact result related to the energy level ordering of a frustrated spin system . in sec . [ sec : open ] , we formulate and prove an analog of lieb - mattis theorem for frustrated spin-1/2 ladder with free boundaries . the reflection symmetry spits the total space of states into two invariant sectors , composed correspondingly from reflection - symmetric and reflection - antisymmetric states . we establish the antiferromagnetic ordering of energy levels in symmetric and antisymmetric sectors separately for wide range of the interaction constants . the nondegeneracy of most levels is proved also . in sec . [ sec : gs ] , using the ordering rule and the results of lieb and schupp @xcite on reflection - symmetric spin systems , the total spin and reflection quantum numbers of the ground state are derived . the results are tested and compared with numerical simulations . in sec . [ sec : comparison ] , the validity of obtained results is checked in some particular cases , for which exact results are already known . section [ sec : impurity ] is devoted to the frustrated ladder with boundary impurity spin . similar ordering of energy levels is established in this case , too . some partial results are obtained for the model with periodic boundary conditions in sec . [ sec : periodic ] . in the last section , we briefly summarize the results obtained in this paper . in appendix , we apply the general results for reflection - symmetric spin systems obtained in refs . and to the frustrated ladder model . in this paper we consider two finite identical antiferromagnetic spin-1/2 chains coupled by rung and diagonal interactions . @xcite the hamiltonian reads ( see fig . [ fig : ladder ] ) : ladder.eepic @xmath10 where @xmath11 and @xmath12 are the spin operators of the first and second chains respectively . @xmath13 and @xmath14 are rung and diagonal interchain couplings , while @xmath15 are intrachain couplings . the open boundary conditions are applied here . all couplings depend on site . their choice corresponds to a model possessing the reflection symmetry @xmath16 with respect to the longitudinal symmetry axis . the reflection just permutes the spins of two chains : @xmath17 . we do not put any restriction on the rung couplings @xmath18 and consider antiferromagnetic intrachain couplings only . we suppose also that the diagonal spin interactions are weaker than intrachain interactions , i.e. , @xmath19 note that the condition above is not too restrictive for antiferromagnetic values of @xmath14 because the system remains invariant under the exchange of rung and diagonal couplings @xmath20 belonging to @xmath21th box . the resulting model corresponds to a ladder obtained by permutation @xmath22 of two spins in all rungs positioned on the right from @xmath21th rung ( i.e. , for all @xmath23 ) . it is topologically equivalent to the original model . therefore , one can consider the couplings subjected to @xmath24 only without loss of generality . note that the condition excludes the values of diagonal couplings @xmath14 equal to @xmath25 . one must mention that we call the model a frustrated ladder , but , in fact , it becomes bipartite for some couplings . in particular , it is bipartite for ferromagnetic diagonal or rung interactions . these cases will be discussed in sec . [ sec : comparison ] . the hamiltonian preserves the total spin @xmath26 of the system . since @xmath16 and @xmath27 are compatible , the eigenstates of ( [ h ] ) can be chosen to be parametrized by the spin ( @xmath28 ) , spin projection ( @xmath29 ) , and reflection ( @xmath30 ) quantum numbers . in order to make easier the use of the reflection symmetry , we introduce the symmetric and antisymmetric superpositions of two spins on each rung : @xmath31 the symmetrized spin @xmath32 describes the total spin of @xmath21th rung . it remains unchanged under reflection , i.e. @xmath33 , while the antisymmetrized rung spin acquires minus sign , i.e. , @xmath34 . now we express the hamiltonian ( [ h ] ) in terms of these operators . after omitting nonessential scalar term , the hamiltonian takes the following simple form : @xmath35 here , we have introduced the symmetrized and antisymmetrized couplings , which are antiferromagnetic due to the condition imposed above on the intrachain and diagonal couplings : @xmath36 note that does not contain terms , which mix symmetrized and antisymmetrized spin operators . this fact is a consequence of the reflection symmetry of the model . we mention also that similar decomposition for frustrated ladder hamiltonian was applied also in refs . and . in terms of lowering and rising operators @xmath37 and @xmath38 the hamiltonian reads @xmath39 further , in this section , we will show that for @xmath40 the minimal energy levels @xmath41 in the symmetric ( @xmath42 ) as well as in the antisymmetric ( @xmath43 ) spin-@xmath6 sectors are nondegenerate and ordered antiferromagnetically , i.e. , are increasing functions of @xmath6 . this is an extension of lieb - mattis ordering theorem @xcite to the frustrated ladder model . here , we briefly outline the steps of the proof . first , we construct a basis , in which all nonzero off - diagonal matrix elements of the hamiltonian become negative . in the next step , we show that the matrix of hamiltonian being restricted to the subspace of states with fixed values of spin projection @xmath44 and reflection @xmath45 quantum numbers is connected . then according to perron - frobenius theorem , the lowest - energy state in every such subspace , called a _ relative ground state _ , is nondegenerate . as we will show , the relative ground state in most cases has total spin value equal to absolute value of its @xmath46 projection ( @xmath47 ) . together with the total spin symmetry of the hamiltonian this implies that the multiplet , to which the relative ground state belongs , possesses the lowest energy value among all spin-@xmath6 multiplets . the antiferromagnetic ordering between lowest energy levels @xmath41 follows then directly from their nondegeneracy . the existence of a basis , where all off - diagonal elements of the hamiltonian are nonpositive , is not obvious as it is for nonfrustrated models . the spin flip applied on one sublattice @xcite is adopted for bipartite systems and does not lead to the desired basis for frustrated systems . in this subsection , we construct nonpositive basis for the frustrated ladder model . the first step is the use of the basis consisting of combined spin states on each rung instead of the standard ising basis consisting of on - site spin - up and spin - down states . then we apply a unitary shift to the hamiltonian , which makes all nonvanishing off - diagonal elements arising from the symmetric part of negative . in the final step , all basic states are multiplied by a sign factor making all nonvanishing off - diagonal matrix elements negative . it is rather difficult to trace out the sign of matrix elements of @xmath48 in the ising basis , consisting of on - site spin states . instead , we start with a more suitable basis consisting of combined spin states on every rung . any rung state can be expressed as a superposition of three symmetric triplet states , @xmath49\uparrow\end{subarray}\right\rangle } } , \qquad { \left|-1\right\rangle}:={\left|1,-1\right\rangle}={{\left|\begin{subarray}{c}\downarrow\\[4pt]\downarrow\end{subarray}\right\rangle } } , \qquad { \left|\tilde{0}\right\rangle}:= { \left|1,0\right\rangle}=\frac1{\sqrt{2 } } \left({{\left|\begin{subarray}{c}\uparrow\\[4pt]\downarrow\end{subarray}\right\rangle}}+{{\left|\begin{subarray}{c}\downarrow\\[4pt]\uparrow\end{subarray}\right\rangle}}\right),\ ] ] and one antisymmetric singlet state , @xmath50\downarrow\end{subarray}\right\rangle}}-{{\left|\begin{subarray}{c}\downarrow\\[4pt]\uparrow\end{subarray}\right\rangle}}\right).\ ] ] we have used above standard notations @xmath51 for on - site spin - up and spin - down states . for convenience , we mark the triplet states shortly by labels @xmath52 and @xmath53 . the total space of states is spanned by the basic states @xmath54 here @xmath55 is one of the four rung states defined above . the reflection operator @xmath16 is diagonal in this basis . its quantum number is @xmath56 , where @xmath57 is the number of singlets in . now define unitary operator , which rotates the odd - rung spins around the @xmath46 axis on angle @xmath58 , as follows : @xmath59}s^{(s)z}_{2l-1}\right),\ ] ] where by @xmath60 $ ] we have denoted the integer part of @xmath61 . under the action of @xmath62 , the odd - rung lowering - rising operators change the sign ( @xmath63 ) , while the others remain unchanged . recall that for bipartite models , a similar unitary shift applied to the spins of one sublattice makes all off - diagonal elements of the hamiltonian nonpositive . @xcite in our case , the hamiltonian ( [ h - spm ] ) transforms to @xmath64 it is easy to see that @xmath65 part of the hamiltonian , presented by the last sum in ( [ h - u ] ) , is diagonal in the basis because it consists of the squares of rung spin operators . the next observation is that @xmath66 part of the hamiltonian @xmath67 gives rise to negative off - diagonal elements . indeed , the symmetrized spin operators describe the spin of rung state and act separately on singlet and triplet states in the usual way . all their off - diagonal matrix elements are positive : @xmath68 taking into account the fact that the coefficients @xmath69 are antiferromagnetic [ see ] , it is easy to see that all nonvanishing off - diagonal elements of the shifted hamiltonian , which are generated by terms containing symmetrized spin operators , are negative . finally , we consider the matrix elements produced by the antisymmetric local terms of hamiltonian . in contrary to symmetric case , the antisymmetrized spin operators mix triplet and singlet states . all their nonzero matrix elements are off - diagonal and have the following values : @xcite @xmath70 using the equations above , we can obtain all nontrivial matrix elements generated by @xmath71 part of the hamiltonian . the nontrivial action of terms with operators @xmath72 on two adjacent spins is [ spm ] @xmath73 the action of terms in containing @xmath74 reads [ sz ] @xmath75 the subscripts 1 and 2 indicate the rung , on which the spin operator acts . taking into account the positivity of antisymmetrized couplings @xmath76 [ see ] , we conclude that the elements generated by acquire an overall minus sign in the shifted hamiltonian matrix . at the same time , the elements corresponding to enter in the hamiltonian with the same sign . hence , only the contribution from gives rise to negative matrix elements in the hamiltonian @xmath67 , whereas and are responsible for unwanted positive off - diagonal elements . in order to alter their sign , we make the following observation . due to the reflection symmetry , the parity of singlet states , i.e. , @xmath56 , is conserved quantity under the action of each term in . the matrix elements and are the only ones , which are responsible for the creation and annihilation of a singlet pair . other elements rest singlet number unchanged . thus , multiplying the basic states on sign factor @xmath77}$ ] , one can make the elements arising from and negative . the sign factor does not affect on other matrix elements . the remaining action just permutes the singlet and @xmath78 triplet states . in order to make this term negative , introduce the ordering between pair of rung states @xmath79 and @xmath80 inside multirung state . we say that a pair is ordered if @xmath80 is located on the left - hand side from @xmath79 . denote by @xmath81 the number of disordered pairs in , i.e. , the number of pairs @xmath82 , where @xmath79 is on the left - hand side from @xmath83 . note that the nearest - neighbor actions change the ordering of only one pair . therefore , if we multiply a basic state by another sign factor @xmath84 , all matrix elements generated by will change the sign and become negative . at the same time , other matrix elements will hold unchained . indeed , it is easy to see , that the nearest - neighbor permutations @xmath85 and @xmath86 do not change the number of disordered pairs @xmath81 while pair creations and annihilations ( @xmath87 and @xmath88 ) change its value on _ even _ number . note that similar type of sign factor has been used in order to prove uniqueness of relative ground states for heisenberg chains with higher symmetries . @xcite finally , the basis , in which all off - diagonal matrix elements of the hamiltonian are nonpositive , is @xmath89+n_{0\tilde0 } } { \left|m_1\right\rangle}\otimes{\left|m_2\right\rangle}\otimes\ldots\otimes{\left|m_{n}\right\rangle}.\ ] ] due to the spin projection and reflection symmetries , the hamiltonian is invariant on each subspace with the definite values of spin projection and reflection operators : @xmath44 , @xmath45 , where @xmath29 and @xmath30 . keeping the terminology , we call it @xmath90 subspace . below , we outline the proof that the matrix of the hamiltonian in basis being restricted on every @xmath90 subspace is connected . note that due to all local actions considered in , contribute in the hamiltonian with nonvanishing coefficients . it is easy to verify using , , and that any two adjacent states @xmath91 and @xmath92 are connected by the @xmath21th local terms of the hamiltonian provided that their quantum numbers are subjected to the conservation laws . in other words , both states must possess the same spin projection and reflection quantum numbers . this rule can be generalized by induction to any two basic states . in fact , any symmetric ( @xmath42 ) basic state after successive applications of the local terms in can be transformed to state @xmath93 , where plus ( minus ) sign holds for positive ( negative ) values of @xmath94 . similarly , any antisymmetric ( @xmath43 ) state is connected to state @xmath93 . this finishes the proof of the connectivity . now all conditions of perron - frobenius theorem @xcite are fulfilled and one comes to the following result . the state @xmath95 must have a definite value @xmath96 of total spin quantum number . otherwise , it could be presented as a superposition of independent states with different spins but the same energy . this would be in contradiction with the uniqueness condition established above . it is evident that if some spin-@xmath6 state overlaps with the relative ground state @xmath95 then @xmath97 . below , we use this property in order to determine @xmath96 . due to the spin reflection symmetry @xmath98 . so , one can consider only non - negative values of @xmath94 . first , we suppose that @xmath99 . if @xmath100 , then the state @xmath101 , which is the highest weight state for a spin @xmath102 multiplet , contributes in the sum . therefore , it overlaps with the relative ground state . according to the arguments above , @xmath103 for this case . else if @xmath104 , then both states @xmath105 and @xmath106 are also presented in the decomposition of @xmath95 with _ positive _ coefficients . according to the definition of basic states , their sum , up to a nonessential sign factor , can be presented as @xmath107 and , of course , overlaps with . in fact , @xmath108 has a definite spin value , which we will determine now . remember we are currently working with the hamiltonian , obtained from the original one by unitary shift . turning back to @xmath48 representation , we have to shift the states back . thus , the original state is @xmath109 up to an overall sign factor . the first term of the product is the highest state of the triplet in the decomposition of two triplets , while the second term is the highest state of @xmath110 multiplet . together they form an @xmath102 multiplet . therefore , @xmath103 in the case of @xmath104 , too . consider now the two subspaces with @xmath111 . if @xmath112 , the state constructed from rung singlets participates in the sum , so the relative ground state is a spin singlet too : @xmath113 . if @xmath114 , the triplet state @xmath115 enters in the sum , and , therefore , the relative ground state is a triplet , i.e. , @xmath116 . summarizing the results , we arrive at the following conclusion : consider now the restriction of the hamiltonian to the sector containing states with fixed values of the total spin and reflection quantum numbers . apparently , the relative ground state @xmath95 is also the lowest - energy state in the sector characterized by @xmath117 and @xmath45 . denote by @xmath41 the lowest energy level in that sector . according to the previous subsection , the relative ground state is unique on @xmath90 subspace . therefore , the level @xmath41 with @xmath117 must be nondegenerate , i.e. , it must contain only one spin-@xmath6 multiplet with parity @xmath118 . this is just the multiplet , which encloses the state @xmath95 itself . using , it is easy to see that any level @xmath41 is nondegenerate in this regard , except perhaps the one with @xmath119 and @xmath114 . moreover , every @xmath90 subspace contains a representative from any multiplet with parity @xmath118 and spin @xmath120 . therefore , the state @xmath95 has the minimum energy for all these multiplets . according to , its spin has the value @xmath121 provided that @xmath122 . the uniqueness of relative ground state then implies that all levels @xmath41 with spin @xmath123 are higher than the level with @xmath117 . consequently , @xmath41 is a monotone increasing function of @xmath6 . if @xmath114 , then the last equation is true for @xmath124 only and , hence , @xmath41 increases in the range @xmath125 finally , we arrive at the following conclusion . the ordering rule above enables to determine the total spin of the minimum - energy states in the symmetric and antisymmetric sectors . in other words , for frustrated ladder with even number of rungs , the ground state in the symmetric sector is a singlet state , while in the antisymmetric sector , it is formed by the three states of a triplet . in contrary , for odd number of rungs , it is a singlet in the antisymmetric sector , while in the symmetric sector , it consists of three triplet states . the ordering rule does not compare energy levels in symmetric ( @xmath42 ) sector with the levels in antisymmetric ( @xmath43 ) sector . it is clear that the total ground state coincides with the minimum - energy state in either symmetric or antisymmetric sector . if the lowest levels in both sectors coincide , the total ground state becomes degenerate . using also the results of previous section , we come to the following conclusion . the frustrated ladder in certain parameter region belongs to a large class of reflection - symmetric models considered recently by lieb and schupp . @xcite in appendix , the rigorous results obtained in ref . are applied to the frustrated spin ladder case . it appears that _ all _ ground states of the hamiltonian are spin _ singlets _ if the couplings satisfy @xmath126 , where @xmath127 is supposed . the nonstrict inequality sign can be set instead of strict one for antiferromagnetic values of diagonal couplings . the details are given in appendix . note that the aforementioned result is true for any values of intrachain couplings @xmath15 including ferromagnetic ones . from the other side , for @xmath40 our results established in previous section become valid too . setting together both lieb - schupp and our results , we come to the conclusion that _ only _ case ( a ) above can take place . this means that the ground state is a unique @xmath122 singlet , provided that the couplings satisfy @xmath128 consider separately the most familiar case of ladder with site - independent couplings . according to the discussions above and in appendix , we come to the following conclusion . it must be emphasized that this result is rigorous for finite size ladders only , which are the main objects of study in this paper . the ground state and lowest excitation of frustrated ladder with 20 sites ( @xmath129 ) obtained by exact diagonalization for different values of rung coupling . both states have the same quantum numbers . the level crossing does not happen in agreement with our exact results . ] the ground state and lowest excitation of frustrated ladder with 20 sites ( @xmath129 ) obtained by exact diagonalization for different values of rung coupling . both states have the same quantum numbers . the level crossing does not happen in agreement with our exact results . ] the results obtained above are in a good agreement with the ground - state properties of frustrated ladder model , which have been investigated intensively in the literature so far . from our results , it is unclear whether or not the singlet - triplet degeneracy takes place for finite - size ladders . moreover , in the thermodynamic limit @xmath130 , the additional degeneracy of energy levels can occur and , in principle , the strict inequality in must be replaced by nonstrict one . the investigations carried out by different methods suggest that all three possibilities for the ground state described in the previous subsection may take place . in fact , the ground - state degeneracy happens at critical points only , while in most cases , the ground state is a unique @xmath122 singlet or @xmath114 triplet . this property remains true in the thermodynamic limit . the rest of this section is devoted to the testing of our exact results and more detailed comparison with data obtained by other methods . the weak interchain coupling analysis ( @xmath131 ) based on conformal field theory approach shows that for @xmath132 , the ground state has a tendency to form singlets along the rungs and triplets along the diagonals , while for @xmath133 , it has a tendency to form triplets along the rungs and singlets along the diagonals . @xcite the former ground state corresponds to the usual ladder or rung - dimer phase , while the last one corresponds to spin-@xmath134 haldane chain . in both cases , it is nondegenerate . the two phases belong to the same universality class because a continuous path connecting them exists and it does not contain any critical point . @xcite the distinction between them has a topological character . @xcite for @xmath135 , both types of dimerized ground states exist leading to twofold degeneracy . @xcite numerical simulations @xcite strongly support this picture and extend the phase transition curve out of the weak coupling region . the transition curve obeys the relation @xmath133 and approaches to the line @xmath135 at the weak coupling limit . the properties of the ground state differ for ladders with odd and even values of the rung number . first , we consider odd - rung ladders . small values of rung coupling @xmath65 give rise to haldane - type ground state , which is a spin triplet with @xmath42 [ case ( b ) in the previous subsection ] while large values lead to rung - singlet - type ground state with @xmath43 [ case ( a ) ] . at some intermediate point , the level crossing happens , which corresponds to aforementioned case ( c ) with degenerate singlet - triplet ground state . the level crossing point approaches to the critical line in the thermodynamic limit . the numerical investigation of small size ladder hamiltonians supports this picture . in fig . [ fig : crossing18 ] , the two lowest levels of frustrated ladder with 18 sites ( @xmath136 ) obtained by exact diagonalization are plotted for different values of rung coupling . the figure clearly indicates that both levels cross at some point . on the contrary , for even - rung ladders , both haldane and rung - dimmer states are singlets and located in @xmath42 sector . the level crossing is forbidden in this case because our results exclude the existence of degenerate double - singlet ground state for finite - size systems . in fig . [ fig : crossing20 ] , we present the two lowest levels obtained by numerical diagonalization of the system with 20 sites ( @xmath129 ) close to the composite spin point @xmath137 . both levels approach each other but do not intersect . the level crossing is absent . it occurs at the composite spin point , which is out of the parameter space considered in this paper . this model will be discussed in the next section . in the thermodynamic limit , the closest point of both levels approaches to the level crossing point of odd - rung ladders and the level crossing picture recovers . similar behavior is observed for ferromagnetic interchain couplings ( @xmath138 and @xmath139 ) . large values of @xmath140 correspond to haldane phase and its small values put the system in rung - dimmer phase . the phase transition curve in the thermodynamic limit is remarkably close to the line @xmath135 ( ref . ) and coincides with it in the weak interchain interaction limit . @xcite this curve has been traced out with high precision by applying different boundary conditions for each phase . @xcite a slight deviation from it is observed close to the ferromagnetic phase only . in this section , we compare the properties of the generalized ladder hamiltonian , established in the previous section with those obtained earlier for some particular values of couplings . if we set all rung couplings to be ferromagnetic and others to be antiferromagnetic , the ladder model becomes _ bipartite_. the two sublattices @xmath0 and @xmath1 , forming the ladder , consist of spins of even and odd rungs correspondingly . then the conditions of standard lieb - mattis ordering theorem @xcite are fulfilled . the ising basis after the unitary shift becomes negative . thus the relative ground state @xmath141 is unique in whole subspace @xmath44 . apparently , it is an eigenstate of the reflection operator . the corresponding eigenvalue is @xmath42 because @xmath141 is a positive superposition of shifted basic states @xcite and the shift operator remains invariant with respect to the reflection . thus , in this case the minimal energy in the symmetric sector is lower than the corresponding one in antisymmetric sector : @xmath142 . as was mentioned in the introduction , the ground state is unique and has spin @xmath143 , @xcite i.e. , is a singlet for even values of @xmath144 and triplet for odd values of @xmath144 . in both cases , @xmath42 . this is in agreement with our results , which claim that @xmath122 for singlet ground state and @xmath114 for triplet one . all mentioned properties are similar to the properties of spin-1 haldane chain because due to ferromagnetic rung couplings , which make preferable rung triplet states , the low - energy properties are well described by haldane chain . @xcite consider now the ladder with ferromagnetic diagonal couplings and antiferromagnetic rung and intrachain couplings . the low - energy properties of this system have been investigated recently in ref . . the model contains standard antiferromagnetic ladder as a particular case . the frustration is lost in this case , too . the lattice becomes bipartite of checkerboard type and the lieb - mattis theorem can be applied also . the sublattice @xmath0 consists of the odd sites of the first chain and the even sites of the second chain , while the sublattice @xmath1 consists of the even sites of the first chain and the odd sites of the second one . the spin - rotation operator leading to a nonpositive basis is given by @xmath145 , where @xmath146 is the spin projection operator of @xmath0 sublattice . @xcite as in the previous case , the relative ground state @xmath141 in @xmath94 subspace is expressed as a positive superposition of the shifted basic states @xmath147m_{2,1},\dots , m_{2,n}\end{subarray}\right\rangle}}= u'\left({{\left|\begin{subarray}{c}m_{1,1}\\[4pt]m_{2,1}\end{subarray}\right\rangle}}\otimes\dots\otimes{{\left|\begin{subarray}{c}m_{1,n}\\[4pt]m_{2,n}\end{subarray}\right\rangle}}\right ) $ ] , where @xmath148 labels on - site spins . now using the definition of @xmath16 and @xmath149 after a simple algebra , we obtain @xmath150m_{2,1},\dots , m_{2,n}\end{subarray}\right\rangle}}= e^{i\pi ( s^z-2 s^z_a ) } { { \left|\begin{subarray}{c}m_{2,1},\dots , m_{2,n}\\[4pt]m_{1,1},\dots , m_{1,n}\end{subarray}\right\rangle}}= ( -1)^{n - m}{{\left|\begin{subarray}{c}m_{2,1},\dots , m_{2,n}\\[4pt]m_{1,1},\dots , m_{1,n}\end{subarray}\right\rangle}}.\ ] ] hence , we have @xmath151 . thus , among two @xmath90 subspaces with different values of @xmath118 , the minimal energy level in the subspace with @xmath100 is the lowest : @xmath152 . the total ground state is a spin singlet with @xmath122 reflection quantum number in accordance with our results in sec . [ sec : open ] . the equal values of diagonal and intrachain couplings ( @xmath153 ) correspond to so - called composite spin model . @xcite these values of coupling parameters are out of the range , where the results of sec . [ sec : open ] are valid . however , due to continuity , some results still remain true in this limiting case . in particular , the ordering rule is fulfilled after replacing the strict inequality sign in by the nonstrict one like in the thermodynamic limit . in contrast , the results related to the uniqueness of relative ground states can not be applied here . indeed , in this case , @xmath154 and the antisymmetrized term , which is responsible for the exchanges between singlet and triplet rung states , is absent in the hamiltonian . the spins of individual rungs are now conserved quantities . @xcite therefore , all singlets remain frozen at their sites and the hamiltonian is _ not connected _ on @xmath90 subspaces anymore . for site - independent values of couplings , the ground state has been obtained explicitly . @xcite for small values of @xmath65 , it is reduced to the ground state of haldane chain ( @xmath119 for even @xmath144 , @xmath155 for odd @xmath144 and @xmath42 always ) , while the large values lead to the rung - singlet state @xmath156 [ @xmath119 and @xmath122 ] . this agrees with our results in sec . [ sec : gs ] . both states are eigenstates of the composite spin hamiltonian , which makes the level crossing at some critical point @xmath158 inevitable . for even values of @xmath144 , the degeneracy disappears under a weak deviation from the composite spin point , as has been discussed in sec . [ sec : gs ] . however , the two lowest levels still approach closely , as is shown in fig . [ fig : crossing20 ] . consider frustrated spin-1/2 ladder with an impurity spin @xmath159 coupled antiferromagnetically to two spins of the first rung , as was shown on fig . [ fig : bounladder ] . the original hamiltonian is supplemented by boundary term @xmath160 , @xmath161 , which preserves the reflection symmetry . bounladder.eepic after the rotation of odd - rung spins on angle @xmath58 around @xmath46 axis by means of unitary operator , the impurity - ladder interaction term acquires the form . @xmath162 we will work here in a basis , which is a natural extension of the basis , used before for the ladder with free boundary conditions : @xmath163+n_{0\tilde0}}{\left|m_0\right\rangle}\otimes{\left|m_1\right\rangle}\otimes \ldots\otimes{\left|m_n\right\rangle}.\ ] ] here @xmath164 ( @xmath165 ) is the usual basis of spin-@xmath159 multiplet . all nonvanishing matrix elements of @xmath166 are positive : @xmath167 . it is easy to see that the spin exchanges due to the boundary term do not affect on the sign factor @xmath168+n_{0\tilde0}}$ ] . this together with implies that @xmath169 has only nonpositive off - diagonal elements in the basis . the bulk part of the hamiltonian is also nonpositive in this basis as was proved already in sec . [ sec : open ] . thus , the entire hamiltonian @xmath170 has no positive off - diagonal elements , too . it is easy to ensure that its restriction to any ( @xmath171 ) subspace with fixed spin projection @xmath94 and reflection @xmath118 quantum numbers is connected . this can be shown using the arguments similar to those applied before for the ladder without impurity spin . applying again perron - frobenius theorem , we come to the conclusion that the relative ground state of @xmath172 of each ( @xmath171 ) subspace is unique and is a positive superposition of _ all _ basic states : @xmath173 now turn to the determination of the spin @xmath96 of this state . as before , in order to do this , we will construct a state with certain value of spin and positive or vanishing coefficients in its decomposition over the basic states . in this case , it will overlap with the relative ground state . then the uniqueness of the relative ground state would indicate that both states have the same spin value . due to the spin reflection symmetry , both states @xmath174 have the same spin quantum number . thus , without any restriction , we can consider @xmath175 values only . first , we consider the values @xmath176 and will prove below that @xmath103 . in @xmath177 sector , there is a simple highest state @xmath178 of a multiplet with spin @xmath102 . it coincides up to an unessential sign factor with one of the basic states . therefore , @xmath103 for this case . for the opposite values of the reflection quantum number [ @xmath179 , we take as an indicator the state @xmath180 which can be considered as a generalization of the state @xmath108 used before for bulk ladder . again , @xmath181 is a _ positive _ superposition of two basic states from . as has been done before for state @xmath108 , its spin can be detected going back to the original @xmath48 representation : @xmath182 . as a result , just the sign of the second term in the brackets is changed . it is easy to check that @xmath183 is the highest state of spin-@xmath159 multiplet , which appears in the decomposition of the tensor product of two multiplets with spins @xmath159 and @xmath134 . thus , the spin of the entire shifted state @xmath184 is again @xmath94 . next , we consider the values @xmath185 , which are essential for @xmath186 . we will show that in this case the relative ground state in @xmath90 subspace belongs to a spin @xmath187 multiplet if @xmath112 ; otherwise it belongs to a spin @xmath188 multiplet . the first case is easy to prove using for the test the basic state @xmath189 of spin @xmath159 . for the second case we choose as a test the state @xmath190 , where @xmath191 it is easy to see that @xmath192 is a positive superposition of three states from the basic set . the action of @xmath193 on @xmath192 changes only the sign of the first and last terms of @xmath194 giving rise to a state of spin @xmath195 multiplet , as is easy to verify by applying @xmath196 to it . finally , summarizing the results obtained above , we arrive at the following conclusion . note that for @xmath197 value of impurity spin , only the first line holds and we have simply @xmath121 . using the spin value of the relative ground state and its uniqueness proved above , one can compare the minimal energy levels @xmath41 of the sectors with fixed spin and reflection quantum numbers and study their degeneracy just in the same way as was done above for the open ladder . for @xmath122 , it is clear from that the energy level @xmath41 is nondegenerate for any @xmath198 . the uniqueness of the relative ground state suggests that all levels with spin @xmath123 are above the level with @xmath117 , which contains the state @xmath95 itself . thus , @xmath41 is a monotone increasing function of spin in the range @xmath198 . for @xmath199 , the level degeneracy and ordering are not clear . all relative ground states @xmath95 with @xmath200 constitute a spin-@xmath159 multiplet having the lowest - energy value among all states with the reflection parity @xmath122 . for @xmath114 , the level @xmath41 is nondegenerate for any @xmath201 . the lowest energy @xmath41 is a monotone increasing function of spin in the range @xmath201 . in the case of @xmath186 , all relative ground states @xmath95 with @xmath202 are combined into @xmath195 multiplet having the minimal energy among the states with @xmath114 . for @xmath197 , the lowest level increases with spin everywhere . the following statement sums up the discussions above . the absolute value sign is used in order to take into consideration the impurity with @xmath197 also . in that case , the ordering rule is simplified because in that case @xmath203 is the lowest value of the total spin and the relation @xmath204 holds for any @xmath118 and any two spins obeying @xmath205 . the lowest - energy states in both symmetric and antisymmetric sectors are unique spin doublets . as for the ladder model without impurity , one can not compare energy levels related to different reflection quantum numbers . nevertheless , one can gain information about the total ground state of the model from the ordering rule established above . the total ground state is nondegenerate and either belongs to @xmath122 sector and , hence , has spin value @xmath206 or belongs to @xmath114 and has spin @xmath207 . if the lowest energy levels in both sectors coincide , which can occur , for example , due to some additional symmetry presented in the model , then the ground state becomes degenerate and is the superposition of two multiplets with spin values @xmath159 and @xmath208 . therefore , the ground state could be _ at most _ doubly degenerate . in contrast to the model with free boundary conditions studied in sec . [ sec : open ] , here we can not apply the lieb - schupp results @xcite in order to choose the valid quantum numbers for the ground state . the reason is that their approach can not be used for a reflection - symmetric spin system if some spins are positioned on the symmetry axis . the translationally invariant periodic spin-1/2 chain with _ odd _ @xmath209 number of spins is a particular case of the general class of frustrated ladders with boundary spin , considered in this section . it corresponds to @xmath197 and @xmath210 with vanishing values for other coupling coefficients . the model is integrable by bethe ansatz @xcite like its more familiar even - site counterpart , but many of its properties differ from those of even - site chain . @xcite in particular , the even - site chain is bipartite and lieb - mattis theorem is valid in this case , while the odd - site chain is a frustrated system . recently , the classical ground state of the last model has been constructed and the lowest energy has been obtained exactly . @xcite the system possesses an additional symmetry : it remains invariant with respect to the cyclic translation by one site @xmath211 . the translation and reflection operators satisfy the commutation relation @xmath212 . thus , if @xmath108 is an eigenstate of @xmath211 with eigenvalue @xmath213 then the reflected state @xmath214 is also an eigenstate with eigenvalue @xmath215 . hence , for all values of momentum except @xmath216 , the energy levels of periodic chain are at least twofold degenerate . the relative ground states of even - site translationally invariant chain have just the exceptional values of momenta ( @xmath217 or @xmath58 ) , which is in agreement with their uniqueness . in contrast , for odd - site chain , the exact solution shows that relative ground state in @xmath44 subspace is _ exactly _ doubly degenerate with two opposite momenta @xmath218 . @xcite this is in agreement with our results , which assert that it should be _ at most _ doubly degenerate . therefore , the relative ground states in symmetric and antisymmetric @xmath94 subspaces have the same energy levels . the reason of the degeneracy is the translation symmetry , which mixes these two states . hence , the lowest levels in symmetric and antisymmetric sectors coincide ( @xmath219 ) except , of course , @xmath220 subspaces , where only one ferromagnetic state presents . moreover , antiferromagnetic ordering rule ( @xmath221 if @xmath205 ) is fulfilled like for even - site chains . the total ground state is the combination of two spin doublets with different reflection quantum numbers . consider now the frustrated ladder with periodic boundary conditions . the bulk hamiltonian is supplemented by the term @xmath222 which describes the interactions between the spins of the first and last rungs . following we set the condition @xmath223 on the boundary coupling constants . in the expression of the hamiltonian , in terms of symmetrized and antisymmetrized spin operators , the boundary term contributes as @xmath224 , where the coefficients @xmath225 , @xmath226 are given in . both are antiferromagnetic . under the action of unitary shift operator , the total hamiltonian acquires the form @xmath227 , where @xmath67 is derived already in and @xmath228 the boundary term @xmath229 produces positive off - diagonal elements in the basis , in which the bulk part @xmath67 of the total hamiltonian is negative . one reason is the sign factor @xmath230 in due to which positive off - diagonal elements appear for even values of @xmath144 . another more important reason is that the sign factor @xmath84 in front of basic states is essentially nonlocal . remember that @xmath81 counts the number of such @xmath82 pairs that @xmath79 is positioned on the left - hand side from @xmath80 . the exchange between the first and last rung spins can produce an uncertain sign factor , which depends on intermediate rung states . below , we will derive the conditions , under which the basis nevertheless remains nonpositive . the term @xmath231 is diagonal in the basis . the other terms produce off - diagonal elements . first , consider the matrix elements , which are generated by the terms with @xmath74 . they do not depend on the parity of @xmath144 . using and it is easy to check that @xmath232 where the dots indicate all intermediate rung states , which have to be the same for the bra and ket states to ensure that the matrix element is nonzero . here , @xmath57 and @xmath233 are the numbers of rung - singlet and @xmath78 triplet states correspondingly . next , we consider the matrix elements generated by lowering - rising spin operators in . taking into account , , and , we obtain @xmath234 the matrix elements and together with their transpositions are the only off - diagonal elements generated by @xmath229 . they will be negative if @xmath235 because the values of couplings @xmath225 and @xmath226 are positive . using the relations @xmath236 for reflection quantum number @xmath118 and @xmath237 for the spin projection quantum number @xmath94 , we can rewrite the last equations as @xmath238 remember now that the bulk part of the hamiltonian has only nonpositive off - diagonal elements in the basis for any @xmath94 and @xmath118 ( see sec . [ sec : open ] ) . therefore , the entire hamiltonian is nonpositive only on those @xmath90 subspaces ( where @xmath44 and @xmath45 ) , which are subjected to the condition . moreover , in any such subspace , its matrix is connected because @xmath67 is connected as was already proven . thus , according to perron - frobenius theorem , the relative ground state in each @xmath90 subspace with quantum numbers obeying is nondegenerate and is a positive superposition of all basic states . the last fact has been used in sec . [ sec : open ] in order to determine the spin @xmath96 of the relative ground state for the open ladder . so , here , the formula can be applied too , and , actually , it is reduced just to @xmath121 since the equations exclude the exceptional case of @xmath111 and @xmath114 . the outcome is as follows . as a consequence , the lowest energy levels @xmath41 among all states with fixed spin @xmath6 and reflection @xmath118 quantum numbers are nondegenerate for @xmath42 and even values of @xmath6 as well as for @xmath43 and odd values of @xmath6 . moreover , we get a partial antiferromagnetic ordering of the lowest energy levels @xmath41 . for odd @xmath144 , the lowest state in the antisymmetric may be a unique spin triplet , a spin singlet(s ) , or a singlet(s)@xmath239triplet representation . remember that in this paper we consider the ladders with coupling obeying @xmath40 . the ladder with periodic boundary term under certain supplementary condition on coupling constants becomes a part of more general class of reflection - symmetric models investigated recently by lieb and schupp . @xcite this fact has been used in sec . [ sec : open ] in the case of free boundaries and is discussed in detail in appendix . for the coupling values obeying latexmath:[$j^\perp_l > and among them there is one with @xmath122 . moreover , for antiferromagnetic diagonal interactions ( i.e. , if all @xmath241 ) the nonstrict inequality sign can be used in the last inequality instead of strict one [ see in appendix ] . in particular , the translationally invariant frustrated ladder with the antiferromagnetic couplings obeying @xmath242 and @xmath243 has only spin - singlet ground states as well as is subjected to the ordering rule proved in this section . for the ladders with even rungs , the aforementioned properties are in good agreement with our results obtained in this section . for the ladders with odd rungs , they even strengthen the partial ordering rule . the relation @xmath244 , holding for any odd @xmath245 and any @xmath246 such that @xmath205 , becomes valid for @xmath247 , too . the question of the degeneracy of the level @xmath248 still remains open . unlike in open ladder case , here we did not obtain definite exact results related to the degeneracy degree and spin quantum number of the total ground state . the main reason is the absence of exact results related to @xmath114 sector in the case when the periodic boundary term presents . we have generalized lieb - mattis energy level ordering rule for spin systems on bipartite lattices @xcite to the frustrated spin-1/2 ladder model . the model consists of two coupled antiferromagnetic chains frustrated by diagonal interactions and possesses reflection symmetry with respect to the longitudinal axis . the spin exchange coupling constants depend on site and are subjected to the relations . we have considered the finite - size system with free boundaries , the system with any impurity spin attached to one boundary as well as the ladder with periodic boundary conditions . below , we describe briefly the results obtained in this paper . the total spin @xmath6 and reflection parity @xmath30 are good quantum numbers . so , the hamiltonian remains invariant on individual sectors with fixed values of both quantum numbers . for the open ladder and ladder with impurity , we have established that the lowest - energy levels @xmath41 of these sectors with the same value of @xmath118 are ordered antiferromagnetically . more precisely , the relation @xmath204 holds for any two spins satisfying @xmath249 . here , @xmath250 is the spin value of the _ unique _ lowest level multiplet among all multiplets with the reflection parity @xmath118 : @xmath251 where @xmath144 is the number of rungs and @xmath159 is the impurity spin value . the case @xmath252 corresponds to the open ladder without impurity . we have also proven that all levels @xmath41 with @xmath253 are nondegenerate , which means that only one multiplet exists on the corresponding level . note that @xmath250 coincides for @xmath42 with the ground - state spin @xmath254 of the haldain chain obtained after replacing each rung with the spin @xmath255 . the last model is bipartite and the lieb - mattis formula @xmath143 , mentioned in the introduction , gives just the value of @xmath256 . in contrast to bipartite spin systems , the lowest levels in the symmetric ( @xmath42 ) and antisymmetric ( @xmath43 ) sectors can not be compared with each other , at least by means of our approach . therefore , in general , the total ground state of the model is a unique multiplet with spin @xmath159 or @xmath208 . if the lowest levels in both sectors coincide , then the degeneracy occurs and the ground state is the superposition of both multiplets . in this regard , our results claim that the ground state can be _ at most _ doubly degenerate . for more restrictive values of couplings given in appendix , the frustrated ladder with free boundaries ( @xmath252 ) fits the class of reflection - symmetric spin systems , all ground states of which are spin singlets . @xcite combining with our results , this implies that the ground state is a _ unique _ spin singlet in this case . this property is true , in particular , for frustrated ladder with site - independent antiferromagnetic couplings obeying @xmath257 and @xmath243 . for ladder with periodic boundary conditions , we have proven a weaker ordering rule . namely , for the ladder with even ( odd ) number of rungs the ordering @xmath204 if @xmath205 is established exactly for @xmath42 ( @xmath43 ) and even ( odd ) values of spin @xmath245 only . the degeneracy of the ground state and its total spin value remain open questions in this case . the author expresses his gratitude to a. a. nersesyan for very useful discussions . he is thankful also to the abdul salam international center for theoretical physics ( ictp ) , where this work was started , for hospitality . the work was supported by the volkswagen foundation of germany , grants intas-03 - 51 - 4000 and intas-05 - 7928 . recently , lieb and schupp @xcite proved exactly that a reflection - symmetric spin system with antiferromagnetic crossing bonds has at least one spin - singlet ground state . moreover , under certain additional conditions , all ground states become singlets . frustrated spin-1/2 ladder is an example of reflection - symmetric system ( see fig . [ fig : ladder ] ) . the hamiltonian in the lieb - schupp form is @xmath258 here , @xmath259 are arbitrary real coefficients . the boundary conditions are @xmath260 for open ladder and @xmath261 ( @xmath262 ) for periodic ladder . there are two kinds of bond in : the first sum consists of rung bonds while the second sum contains square bonds . comparing with one can express the spin coupling constants in in terms of lieb - schupp coefficients : @xmath263 in the second equation above the boundary conditions @xmath264 and @xmath265 are used for open and periodic ladders correspondingly . note that the value of interchain rung coupling is always antiferromagnetic . exploiting the reflection symmetry , lieb and schupp proved that there is a ground state of , which overlaps with rung - singlet state @xmath266 ( ref . ) and , more generally , with the canonical singlet state . @xcite this means that the spin ladder hamiltonian with diagonal interactions possesses a spin - singlet ground state with @xmath122 reflection symmetry for any value of @xmath15 provided that the remaining two couplings are subjected to the relation @xmath267 for some @xmath268 . setting several restrictions on auxiliary parameters @xmath269 , one can simplify and significantly and get the constraints containing interaction couplings only . of course , those restrictions narrow the total region in phase space , where lieb - schupp result is valid . consider the simplest and most familiar case of ladder with site - independent couplings . we set @xmath270 , @xmath271 , @xmath272 for periodic ladder , where @xmath273 are some real parameters . for open ladder , we choose the boundary values of @xmath274 parameter as @xmath275 . it is easy to verify that both for open and periodic boundary conditions , all values of interchain couplings satisfying @xmath276 , @xmath277 lie inside the general region in phase space defined by . hence , there is a @xmath122 singlet among the ground states if two interchain coupling constants satisfy @xmath278 this relation can be generalized to the ladder model with site - dependent couplings . setting @xmath279 , one reduces the general region to @xmath280 here , as before , the boundary conditions @xmath281 and @xmath282 are applied for open and periodic ladders correspondingly . another important property of reflection - symmetric spin systems established exactly in refs . and is that for any crossing bond , the expectation of the spin of its all sites , weighted by their coefficients , vanishes for _ any _ ground state @xmath283 . using this feature of ground states , it was proved that in a system with sufficient symmetry , so that every spin can be considered to be involved in a crossing bond , _ all _ ground states are spin singlets . as was already mentioned , for frustrated ladder , there are rung bonds and square bonds , which are described correspondingly by the first and second term in . the aforementioned condition means that @xmath284 if @xmath285 and @xmath286 , where @xmath32 is the spin operator of @xmath21th rung . therefore , if @xmath287 , then the expectation of total spin of each square bond vanishes also , i.e. , @xmath288 . applying this to the ladder with site - independent spin exchange , one can ensure that all its ground states are singlets provided that the interchain couplings satisfy @xmath289 indeed , using aforementioned parametrization of @xmath290 and taking sum over all square bonds , we get @xmath291 for periodic ladder and any ground state @xmath283 . for the open ladder , one must add two boundary rung bonds with coefficients @xmath292 and @xmath293 in order to obtain the last equation . so , if @xmath294 , then any ground state is a singlet . consider now the ladder with site - dependent spin exchange and set again @xmath279 in . note that if all @xmath295 , which makes strict the inequality sign in , then due to the discussions above the mean value of the spin of every rung vanishes for any ground state . therefore , all ground states are singlets if @xmath296 . then the ground - state mean value of the total spin of each such bond vanishes as has been mentioned above . taking the sum over all square bonds for the periodic ladder and the sum over half of square bonds in checkerboard order for the open ladder with even number of rungs , we obtain the zero expectation for the total spin of the model provided that @xmath297 for the open ladder with odd number of rungs , one of the boundary rung spins must be added to that sum . the ground state then is a singlet if the corresponding coefficient [ @xmath298 or @xmath299 in ] is nonzero . this means that the inequality in one of the two boundary relations in must be strict , i.e. , @xmath300 or @xmath301 . c. lhuillier and g. misguich , in _ high magnetic fields : applications in condensed matter physics and spectroscopy _ , lecture notes in physics vol . 595 , edited by c. berthier , l. p. levy , and g. martinez , ( springer , berlin , 2002 ) , pp . 161 - 190 , arxiv : cond - mat/0109146 . m. p. gelfand , phys . b * 43 * , 8644 ( 1991 ) ; i. bose and s. gayen , phys . rev . b * 48 * , 10653 ( 1993 ) ; h. kitatani and t. oguchi , j. phys . . jpn . * 65 * , 1387 ( 1996 ) ; a. honecker , f. mila , and m. troyer , eur . j. b * 15 * , 227 ( 2000 ) ; g. fth , o. legeza , and j. slyom , phys . b * 63 * , 134403 ( 2001 ) ; e. chattopadhyay and i. bose , phys . rev . b * 65 * , 134425 ( 2002 ) ; m. nakamura , physica b * 329 * , 1000 ( 2003 ) ; h. h. hung , c. d. gong , y. c. chen , and m. f. yang , phys . b * 73 * , 224433 ( 2006 ) ; t. vekua and a. honecker , phys . b * 73 * , 214427 ( 2006 ) .

studying quasars at the highest redshifts can constrain models of galaxy and black hole formation , and it also probes the intergalactic medium in the early universe . optical surveys have to date discovered more than 60 quasars up to , a limit set by the use of the-band and ccd detectors . only one quasar has been discovered , namely the quasar ulas j1120 + 0641 , using near - infrared imaging . here we report the discovery of three new quasars in 332deg of the visible and infrared survey telescope for astronomy kilo - degree infrared galaxy ( viking ) survey , thus extending the number from 1 to 4 . the newly discovered quasars have redshifts of , , and . the absolute magnitudes are between 26.0 and 25.5 , 0.61.1mag fainter than ulas j1120 + 0641 . near - infrared spectroscopy revealed the emission line in all three objects . the quasars are powered by black holes with masses of . in our probed redshift range of we can set a lower limit on the space density of supermassive black holes ofmpc . the discovery of three quasars in our survey area is consistent with the quasar luminosity function when extrapolated to . we do not find evidence for a steeper decline in the space density of quasars with increasing redshift from to .

1311.3666

the cosmic microwave background reveals the earliest observable structure in the universe , which dates back to the epoch of recombination at @xmath17 . the next most distant observable structure represents the population of galaxies and quasars at @xmath18 . the interval between @xmath19 and @xmath16 contains many landmark events , such as the formation of the first stars and galaxies , the growth of the first massive black holes and the reionization of the neutral hydrogen in the intergalactic medium ( igm ) . measurements by the _ wilkinson microwave anisotropy probe _ give a @xmath20 lower bound to the epoch of reionization of @xmath21 @xcite but do not pin down when reionization began or ended . another probe of the ionized state of the igm are spectra of high redshift quasars . already , the analysis of spectra of @xmath22 quasars suggest that the epoch of reionization ended around @xmath23 @xcite . while in recent years the evidence is pointing toward galaxies providing the required uv flux to reionize the universe ( e.g. * ? ? ? * ; * ? ? ? * ) , the identification and study of quasars up to the highest redshifts is very important for several reasons : ( 1 ) absorption line studies of cosmologically distributed intervening material allow the determination of the baryonic content and physical conditions ( metallicity , temperature , ionization state ) of the universe during the epoch of reionization ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ; ( 2 ) the number density of high redshift quasars provide constraints on the mechanisms that are required to seed and grow @xmath24@xmath10 supermassive black holes less than a gyr after the big bang ( e.g. , * ? ? ? * ) ; and ( 3 ) high redshift quasars are generally located in luminous , massive host galaxies . by studying the host galaxies we get insight into the formation of galaxies in the early universe ( e.g. , * ? ? ? * ; * ? ? ? * ) . in the last decade more than sixty @xmath25 quasars have been discovered in various different surveys ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the vast majority of these quasars have been discovered in optical surveys , in particular the sloan digital sky survey ( sdss ; e.g. , * ? ? ? the canada france high@xmath1 quasar survey ( cfhqs ; * ? ? ? * ) . the limitation of optical surveys that use silicon based ccd detectors is that they can not find quasars beyond @xmath26 as the sources become too faint in the reddest optical band of the survey ( @xmath1-band ) due to absorption by the intervening ly@xmath27 forest . to overcome this limit large area surveys in near - infrared bandpasses are necessary . one of the first wide - field near - infrared surveys reaching a depth that allowed a search for the highest redshift quasars was the uk infrared telescope infrared deep sky survey ( ukidss ) large area survey ( las ; * ? ? ? * ) . using the ukidss las , @xcite discovered the first quasar beyond @xmath28 , ulas j1120 + 0641 at @xmath29 . this one @xmath15 quasar already provides constraints on the neutral fraction of hydrogen up to @xmath29 @xcite , on the metallicity of the igm @xcite , and on the formation of dust and stars in the host galaxy @xcite . this shows that more quasars at @xmath28 are needed . here we present the first @xmath28 quasars discovered in the visible and infrared survey telescope for astronomy ( vista ) kilo - degree infrared galaxy ( viking ) survey . in section [ surveydata ] we give a short description of the viking survey . in section [ candidateselection ] we discuss the selection of high - redshift quasar candidates in the viking data , followed by the details of the imaging and spectroscopic observations of quasar candidates in section [ followup ] . we continue with an overview of the results of the follow - up observations in section [ results ] . in section [ properties ] we characterize the newly discovered quasars . in sections [ catcomp][colcomp ] we assess various sources of incompleteness and the effect on our quasar search . we discuss the implications of our results on the quasar luminosity function at @xmath15 in section [ qsolf ] . we conclude with a summary in section [ discussion ] . all magnitudes are given in the ab system . for magnitudes obtained with vista we use the vega to ab conversions provided on the cambridge astronomical survey unit ( casu ) website : @xmath30 , @xmath31 , @xmath32 , @xmath33 and @xmath34 . an @xmath35-dominated cosmology with @xmath36 , @xmath37 and @xmath38kms@xmath39mpc@xmath39 @xcite is adopted . the viking survey is one of six public surveys @xcite ongoing since late 2009 on vista , a 4 m wide field survey telescope located at eso s cerro paranal observatory in chile @xcite . the telescope is equipped with the near - infrared camera vista infrared camera @xcite , which contains sixteen 2048@xmath402048pixel@xmath4 raytheon virgo hgcdte infrared detectors and has a field of view of 1.65deg in diameter . with a pixel scale of 034pixel@xmath39 , the instantaneous field of view ( called a `` pawprint '' ) is 0.6deg@xmath4 . because the detectors sparsely sample the field of view , a mosaic of six `` pawprints '' carried out in a six - step pattern covers a continuous area of 1.5deg@xmath4 of the sky ( a `` tile '' ) . the viking survey aims to cover 1500deg@xmath4 spread over three extragalactic areas in five broad - band filters @xmath41 , @xmath42 , @xmath43 , @xmath44 , and @xmath45 . the details of the five broad - band filters and the limiting magnitudes of the viking survey are listed in table [ vikingfilters ] . the three areas and the r.a . and decl . limits are as follows : lcccc @xmath41 & 0.877 & 0.097 & 480 & 23.1 + @xmath42 & 1.020 & 0.093 & 400 & 22.3 + @xmath43 & 1.252 & 0.172 & 400 & 22.1 + @xmath44 & 1.645 & 0.291 & 300 & 21.5 + @xmath45 & 2.147 & 0.309 & 480 & 21.2 + \1 . sgp : 22@xmath4600@[email protected].@xmath4803@xmath4630@xmath47 , @xmath49@xmath48decl.@xmath48@xmath50 ; + 2 . ngp : 10@xmath4630@[email protected].@xmath4815@xmath4630@xmath47 , @xmath51@xmath48decl.@xmath48@xmath52 ; + 3 . gama09 : 08@xmath4636@[email protected].@xmath4809@xmath4624@xmath47 , @xmath53@xmath48decl.@xmath48@xmath54 . the data are processed by casu and archived at the wide field astronomy unit at the royal observatory in edinburgh as part of the vista data flow system @xcite . at casu all the vista data are processed to science products , including astrometric and photometric calibration , for each night of data @xcite . the stacking of pawprints into a tile is done if a suitable set of pawprints has been taken within a single observing block ( ob ) . for the viking survey , the observations of a single tile in the five broad - band filters are spread over two obs : in one ob , the @xmath41- , @xmath42- and half the @xmath43-band are taken , while the second ob consists of the @xmath44- , and @xmath45-band images and the second half of the @xmath43-band observations . after the reduction at casu , we thus have @xmath41- , @xmath42- , @xmath44- and @xmath45-band images and catalogs to the full depth , while we have two separate @xmath43-band images with half the total exposure time . the viking data used in this article were obtained between 2009 november and 2011 march . we downloaded the processed images and catalogs directly from casu and band - merged the catalogs ourselves . when we had two separate @xmath43-band catalogs for the same tile , we merged the two catalogs . a search radius of 15 was used to merge the catalogs . the search radius is roughly three times that of the seeing measured in the viking data , which has a medium fwhm of 099 . if no entry was found in a catalog , a @xmath55 limit was assigned to the object . the total number of tiles observed up to 2011 march for which we had at least @xmath41- , @xmath42- and @xmath43-band catalogs was 44 in the gama09 region , 59 in the ngp and 121 in the sgp . the total unique area covered by these tiles was 331.6deg@xmath4 ( 61.2deg@xmath4 in gama09 , 89.6deg@xmath4 in the ngp and 180.8deg@xmath4 in the sgp ) . the area as function of @xmath56 depth in the @xmath42-band is shown in figure [ areadepth ] . the median @xmath56 @xmath42-band depths are 21.67 , 21.55 , and 21.51 for the gama09 , ngp , and sgp regions respectively . we selected candidate high redshift ( @xmath57 ) quasars from the band - merged catalogs by applying the following steps : 1 . objects should be detected in the @xmath42-band with a signal - to - noise ratio ( s / n ) of at least ( s / n)@xmath58 . we also required that the object is detected in at least one other band . for all bands we use aperture - corrected magnitudes measured in an aperture with a radius of 10 ( labeled apermag3 in the catalogs ) . 2 . we only consider point sources that we define as objects having a probability of being a galaxy ( pgalaxy ) of pgalaxy@xmath59 ( see * ? ? ? * ; * ? ? ? * for details on the morphological parameters ) . this selection makes sure that most real point sources are selected with a manageable fraction of galaxies ( see also section [ pscomp ] and * ? ? ? 3 . the aperture - corrected magnitudes in the @xmath42-band had to be independent of the size of aperture used to within 0.2mag . this was forced by applying apermag1@xmath60apermag3@xmath61 and apermag3@xmath60apermag4@xmath61 . removing this criterion did not increase the number of candidate quasars , but did add more artifacts to our candidate lists . 4 . because @xmath28 quasars occupy a unique region in color space , we can isolate potential quasars in the catalogs by introducing color criteria . we applied two sets of color criteria to select @xmath28 quasar candidates , a conservative color selection and an extended color selection . the conservative color selection minimizes the number of foreground objects wrongly selected as candidate quasars and is based on the work of @xcite . the extended color criteria maximize the completeness of the high redshift quasar selection , without purposely selecting foreground objects . the conservative color criteria are as follows : + @xmath62 and @xmath63 and ( @xmath64 or @xmath65 ) and @xmath66 and undetected in @xmath67 , @xmath68 , @xmath69 and @xmath70 ( when optical imaging is available ) . + the extended color criteria are : + @xmath71 and @xmath63 and ( @xmath72 or @xmath73 ) and @xmath74 and @xmath75 and undetected in @xmath67 , @xmath68 , @xmath69 , @xmath70 ( when optical imaging is available ) . + our color criteria are illustrated in figure [ cc_zyj ] . detector 16 has a time - varying quantum efficiency . as a result the flat fielding is not accurate and the detector produces a large number spurious detections . we therefore ignore objects that originated from detector 16 in one pawprint and were not independently confirmed in another pawprint . 6 . to check the magnitudes listed in the catalogs and especially nondetections in the @xmath41-band , we performed aperture photometry in each of the viking images at the location of the candidate . if the colors measured did not fulfill the color criteria listed above , the object was removed from our candidate list . the final step was visually inspecting all candidates in order to remove any remaining spurious objects . most of the eliminated objects were near the edge or just outside the @xmath41-band image , or on diffraction spikes . this step approximately halved the number of candidates on our lists . with the conservative color criteria we selected 43 objects in the three different regions observed by viking : 7 objects in gama09 , 18 in the ngp , and 18 in the sgp . we applied our extended color selection to the ngp region and further selected another 42 objects . we loosely ranked the objects according to their @xmath76 color , i.e. , objects with a higher @xmath76 had a higher priority for follow - up observations . on 2011 june 2629 we observed quasar candidates from viking with the european southern observatory s faint object spectrograph and camera 2 ( efosc2 ; * ? ? ? * ) on the 3.58 m eso new technology telescope ( ntt ) . we observed in filters gunn @xmath70 ( filter # 705 , hereafter @xmath77 ) and gunn @xmath1 ( filter # 623 , hereafter @xmath78 ) . images taken with the @xmath78 filter were calibrated using stars within the efosc2 field of view . using the filter transmission curves , the ccd efficiency and stellar spectra from the bpgs spectroscopic atlas @xcite , we established a relation between the @xmath78 and the viking @xmath41 and @xmath42 magnitudes of @xmath79 . in fields with sdss coverage ( the ngp and gama09 regions ) , the @xmath77 images were calibrated using the relation ( derived from synthesized colors of the bpgs stars ) @xmath80 . observations in the @xmath77 filter of candidates in the sgp , for which no optical photometry of nearby stars was available , were calibrated using the spectrophotometric standard stars ltt1020 , ltt6248 , and ltt9239 @xcite that were observed in the evening and morning twilight . during the ntt run we observed 18 quasar candidates that were selected with the conservative color criteria and 20 sources that fulfilled the extended color criteria . to fill gaps in the nights , we also targeted seven objects that fell just outside the color selection and/or had a @xmath42-band detection just below the ( s / n)@xmath81@xmath82 limit . individual images had an exposure time of either 600s or 900s , and the total exposure time per source varied from 600s to 8100s . we observed 45 objects in the @xmath77 filter , and 8 of those were also observed in the @xmath78 filter . generally , if an object showed a clear detection in the initial @xmath77 images , we stopped observing it . objects not detected in @xmath77 were prioritized for deeper @xmath77 and @xmath78 imaging . the results of the efosc2 photometry are listed in table [ efoscobs ] and are shown in figure [ nttccplot ] . lcccc j01093047 & 01@xmath4609@[email protected] & 30@xmath8447263 & @xmath8525.27 & [email protected] + j01253016 & 01@xmath4625@[email protected] & 30@xmath8416274 & [email protected] & + j02213139 & 02@xmath4621@[email protected] & 31@xmath8439494 & [email protected] & + j02233208 & 02@xmath4623@[email protected] & 32@xmath8408377 & [email protected] & + j02233144 & 02@xmath4623@[email protected] & 31@xmath8444266 & [email protected] & + j02593100 & 02@xmath4659@[email protected] & 31@xmath8400544 & [email protected] & + j03053150 & 03@xmath4605@[email protected] & 31@xmath8450560 & @xmath8524.97 & [email protected] + j03053400 & 03@xmath4605@[email protected] & 34@xmath8400019 & [email protected] & + j03223350 & 03@xmath4622@[email protected] & 33@xmath8450252 & [email protected] & + j0914 + 0155 & 09@xmath4614@[email protected] & + 01@xmath8455520 & [email protected] & + j1146 + 0145 & 11@xmath4646@[email protected] & + 01@xmath8445488 & [email protected] & + j1147 + 0126 & 11@xmath4647@[email protected] & + 01@xmath8426346 & [email protected] & + j11490046 & 11@xmath4649@[email protected] & 00@xmath8446008 & [email protected] & + j1151 + 0118 & 11@xmath4651@[email protected] & + 01@xmath8418053 & [email protected] & + j1154 + 0141 & 11@xmath4654@[email protected] & + 01@xmath8441533 & @xmath8524.18 & @xmath8522.81 + j11560007 & 11@xmath4656@[email protected] & 00@xmath8407581 & @xmath8524.49 & @xmath8523.22 + j1211 + 0007 & 12@xmath4611@[email protected] & + 00@xmath8407312 & [email protected] & + j1213 + 0159 & 12@xmath4613@[email protected] & + 01@xmath8459457 & [email protected] & + j1213 + 0200 & 12@xmath4613@[email protected] & + 02@xmath8400327 & [email protected] & + j1218 + 0051 & 12@xmath4618@[email protected] & + 00@xmath8451129 & [email protected] & + j1218 + 0106 & 12@xmath4618@[email protected] & + 01@xmath8406242 & [email protected] & + j1219 + 0003 & 12@xmath4619@[email protected] & + 00@xmath8403264 & [email protected] & + j1412 + 0128 & 14@xmath4612@[email protected] & + 01@xmath8428100 & [email protected] & + j1413 + 0139 & 14@xmath4613@[email protected] & + 01@xmath8439555 & [email protected] & + j1414 + 0126 & 14@xmath4614@[email protected] & + 01@xmath8426197 & [email protected] & + j14180018 & 14@xmath4618@[email protected] & 00@xmath8418531 & [email protected] & + j1418 + 0109 & 14@xmath4618@[email protected] & + 01@xmath8409339 & [email protected] & + j14190100 & 14@xmath4619@[email protected] & 01@xmath8400153 & [email protected] & + j14250054 & 14@xmath4625@[email protected] & 00@xmath8454295 & [email protected] & + j1429 + 0142 & 14@xmath4629@[email protected] & + 01@xmath8442569 & [email protected] & + j1431 + 0029 & 14@xmath4631@[email protected] & + 00@xmath8429000 & [email protected] & [email protected] + j14340003 & 14@xmath4634@[email protected] & 00@xmath8403454 & [email protected] & + j1447 + 0113 & 14@xmath4647@[email protected] & + 01@xmath8413464 & [email protected] & + j14500023 & 14@xmath4650@[email protected] & 00@xmath8423457 & [email protected] & + j1450 + 0100 & 14@xmath4650@[email protected] & + 01@xmath8400221 & [email protected] & + j1451 + 0021 & 14@xmath4651@[email protected] & + 00@xmath8421334 & [email protected] & + j22143134 & 22@xmath4614@[email protected] & 31@xmath8434390 & [email protected] & + j22153145 & 22@xmath4615@[email protected] & 31@xmath8445446 & [email protected] & + j22183154 & 22@xmath4618@[email protected] & 31@xmath8454300 & @xmath8524.77 & + j22223129 & 22@xmath4622@[email protected] & 31@xmath8429424 & [email protected] & + j22233158 & 22@xmath4623@[email protected] & 31@xmath8458356 & [email protected] & [email protected] + j22263124 & 22@xmath4626@[email protected] & 31@xmath8424453 & [email protected] & + j22293057 & 22@xmath4629@[email protected] & 30@xmath8457228 & [email protected] & [email protected] + j23483054 & 23@xmath4648@[email protected] & 30@xmath8454100 & @xmath8525.13 & [email protected] + j23583355 & 23@xmath4658@[email protected] & 33@xmath8455211 & [email protected] & + of the 45 objects we observed , 6 remained undetected in the @xmath77-band ( see section [ results ] ) . these sources were targets for our follow - up spectroscopy . between 2011 july 22 and 2011 august 20 we obtained spectra for four of the six that remained undetected in @xmath77 images . we used the focal reducer / low dispersion spectrograph 2 ( fors2 ; * ? ? ? * ) on the 8.2 m very large telescope ( vlt ) antu . the candidates were observed through the 600z holographic grism with a 13 wide longslit . the pixels were 2@xmath402 binned to decrease the readout time and noise , giving a spatial scale 025 pixel@xmath39 and a dispersion of 1.62pixel@xmath39 . between exposures the pointing of the telescope was shifted by @xmath810 along the slit for more accurate sky subtraction . the position angle of the slit was set to include a nearby star that could provide the trace of the spectrum . on 2011 july 22 , 23 , and 29 candidate viking j234833.34305410.0 ( hereafter j23483054 ) was observed for a total of 6750s under varying conditions . on july 22 and 23 the seeing varied between 15 and 25 with clear sky . on july 29 the seeing was 1113 while there were some thin clouds . candidate viking j010953.13304726.3 ( hereafter j01093040 ) was observed for 2280s on 2011 july 22 . conditions were clear with a seeing around 12 . six exposures totaling 4500s were taken of candidate viking j030516.92315056.0 ( hereafter j03053150 ) on 2011 august 13 and 20 . the seeing was around 07 , and there were some thin clouds . finally candidate viking j221857.36315430.0 ( hereafter j22183153 ) was observed on 2011 august 20 for 1350s through thin clouds and with a seeing of 06 . the data reduction , which included bias subtraction , flat fielding using lamp flats and sky subtraction , was done using custom written routines in idl . for the wavelength calibration exposures of he , hgcd , and ne arc lamps were obtained . the wavelength calibration was checked using sky emission lines . the typical rms of the wavelength calibration was better than 0.6 . the spectra cover roughly the wavelength range 710010300 . observations with a 5 slit of the spectrophotometric standard star ltt1020 @xcite were used for the flux calibration . the flux calibration was found to be highly uncertain above 10100 due to a lack of flux density values for ltt1020 in the literature above this value . we therefore ignore the spectra above this wavelength . the fors2 spectra are shown in figure [ forsspectra ] . to increase the wavelength coverage of our spectroscopy to the near - infrared , we observed two of our spectroscopic targets , j23483054 and j01093040 , with the medium resolution spectrograph x - shooter @xcite on the cassegrain focus of the 8.2 m vlt kueyen ( ut2 ) on 2011 august 1921 and on 2011 november 24 . x - shooter consists of three spectrographs , each covering a different wavelength region . the three arms and wavelength coverage of the instrument are as follows : uvb ( 30005595 ) , vis ( 559510240 ) and nir ( 1024024800 ) . given the extreme red colors of our targets , we ignored the uvb arm . the slit length in the vis and nir arm was 11 , and the pixel scales are 016pixel@xmath39 and 021pixel@xmath39 , respectively . the slit width used for the observations ranged between 0915 in the vis arm and between 0912 in the nir arm , with the width adjusted to counter periods of bad seeing . the resulting resolution varied between @xmath87 in the vis arm and between @xmath88 in the nir arm . observing conditions on 2011 august 19 were variable with the seeing in the optical ranging from 09 to @xmath813 and some thin clouds . for flux calibration , observations of the spectrophotometric standard stars ltt7987 and feige110 were obtained @xcite . we observed j01093040 for 14,400s in august and for 7200s in november . j23483054 was observed in august only for 8783s . we made use of the x - shooter pipeline version 1.3.7 @xcite to reduce the data to two - dimensional rectified wavelength - calibrated spectra . the extraction of the spectra and the flux calibration were performed with custom - made programs in idl . we scaled the one - dimensional spectra to match the observed infrared colors from viking ( listed in table [ qsocolors ] ) . we do note that the @xmath89 color of the quasars derived from the x - shooter spectra is 0.2mag redder as compared to the catalog colors ( table [ qsocolors ] ) . a possible reason is that the light that the @xmath42-band is measuring ends up partly in the vis arm and in the nir arm . this causes additional uncertainties in the flux calibration around 1@xmath90 m . in the top two panels of figure [ irspecs ] we show the reduced x - shooter spectra . lccc[t ] ra ( j2000 ) & 23@xmath4648@[email protected] & 01@xmath4609@[email protected] & 03@xmath4605@[email protected] + dec ( j2000 ) & 30@xmath8454100 & 30@xmath8447263 & 31@xmath8450560 + @xmath91 & [email protected] & [email protected] & [email protected] + @xmath92 & @xmath8523.42 & @xmath8522.98 & [email protected] + @xmath93 & [email protected] & [email protected] & [email protected] + @xmath94 & [email protected] & [email protected] & [email protected] + @xmath95 & [email protected] & [email protected] & [email protected] + @xmath96 & [email protected] & & [email protected] + @xmath97 & @xmath98 & @xmath99 & @xmath100 + @xmath101 & [email protected] & [email protected] & [email protected] + @xmath102 ( @xmath10 ) & @xmath103@xmath86@xmath104@xmath40@xmath105 & @xmath106@xmath86@xmath107@xmath40@xmath105 & @xmath108@xmath86@xmath109@xmath40@xmath105 + we observed j03053150 using the folded - port infrared echellette ( fire ; * ? ? ? * ; * ? ? ? * ) on the magellan baade telescope on four separate observing runs between 2011 november and 2012 january . fire produces spectra with @xmath110 ( @xmath111kms@xmath39 ) over the full @xmath112 m wavelength range . the data were taken in generally clear conditions with a seeing of 08 fwhm or better , with a total integration time of 26,400s . we reduced each echelle frame using a customized idl suite ( firehose ) designed for fire data . the pipeline produces an order - by - order two - dimensional spectral model of the sky using the principles outlined in @xcite and subtracts this from the flattened data frames . wavelength solutions are obtained using tabulated values of the oh sky emission lines imprinted on the same frames . we interspersed observations of a0v stars for telluric correction , which was performed using the algorithms of @xcite as implemented in the spextool software package @xcite . finally , the corrected and flux calibrated orders for all exposures were combined into a single one - dimensional spectrum , with each individual sampling weighted by its squared s / n for an optimum s / n composite ( figure [ irspecs ] ) . we obtained @xmath77 imaging of 45 high redshift quasar candidates with the ntt and @xmath78 imaging for more promising candidates . based on the colors ( figure [ nttccplot ] ) , 39 objects could be rejected as quasar candidates after the ntt imaging as they are likely to be late m or early l dwarfs . this is much as expected , since these objects are numerous and a small fraction of them are expected to scatter into our selection region due to random photometric errors . only six quasar candidates were undetected in the @xmath77 images ( figure [ nttccplot ] and table [ efoscobs ] ) and could not immediately be identified as a foreground object . five of these satisfied the conservative color criteria , while the sixth source , j22183153 , was a filler object with a ( s / n)@xmath113 . four candidates , including j22183153 , were spectroscopically observed with fors2 ( figure [ forsspectra ] ) . the spectrum of j22183153 shows a smoothly rising spectrum and the source is most likely an early l dwarf . the fors2 spectra of the three other objects , j23483054 , j01093047 and j03053150 , show a strong continuum decrement between 9200 and 9600 , and were identified as quasars at @xmath114 . for these three sources , we obtained near - infrared spectroscopy ( figure [ irspecs ] ) to more accurately determine the redshift and estimate the properties of the quasars . in section [ method ] we describe how we derive the various quasar properties . these results are discussed in section [ properties ] and summarized in table [ qsocolors ] . based on the available photometry , the two remaining candidates , j115439.84 + 014153.3 and j115642.26000758.1 , could either be high redshift quasars or late - type stars ( m , l , or t dwarfs ) . spectroscopic observations of these sources are needed to determine their nature . the absolute magnitude of the quasars at a rest - frame wavelength of 1450 ( @xmath115 ) can be derived by measuring the flux density in the near - infrared spectra at the redshifted wavelength 1450@xmath40@xmath116 . the redshifts of the quasars of @xmath114 shift the rest - frame wavelength of 1450 to an observed wavelength of 1.101.15@xmath90 m , which is roughly between the @xmath42- and @xmath43-band . because the @xmath89 colors measured from the x - shooter spectra do not agree with the quasar colors from the catalogs ( see sections [ xshooter ] ) , we scale the spectra of j23483054 and j01093040 to either the catalog @xmath42 or @xmath43 magnitude and compute @xmath115 twice from the flux density that we measure in the spectra at the appropriate wavelength . the absolute magnitudes of these two quasars listed in table [ qsocolors ] are the average of the two values , and the uncertainties are the standard deviation of the mean . this way the uncertainty in the absolute magnitude includes the issues with the flux calibration . for j03053150 the @xmath89 color measured from the fire spectrum was consistent with the one from the catalog , and we could measure the flux density at 1450 in the rest - frame directly in the spectrum . _ black hole mass . _ in @xcite the infrared spectra of the quasars were fitted to estimate the black hole mass and the redshift of the @xmath1172799 line . for the details of the fitting we refer to @xcite . here we will give a short summary of the fitting procedures . @xcite modeled the quasar continuum as the sum of the power - law emission from the active galactic nucleus , the balmer continuum emission @xcite and the and emission line forest @xcite . the probability of the continuum model parameters was estimated using a grid - based approach . to fit the line , @xcite sampled the obtained continuum probability distribution @xmath118 times by using a monte carlo rejection method . for each sample , the corresponding continuum was subtracted from the observed spectrum and the emission line was fitted , assuming a gaussian profile , using a @xmath119 minimization routine . the resulting distribution of black hole masses was estimated by combining the @xmath118 continuum luminosities at @xmath120 with the fwhm measured for the line : @xmath121 @xcite where @xmath122 is the fwhm of the line in kms@xmath39 and @xmath123 is the nuclear continuum luminosity at @xmath120 in ergs@xmath39@xmath39 . the uncertainties concerning the black hole mass estimates listed in section [ properties ] only include the uncertainties from the continuum modeling and from the line fitting . systematic uncertainties , such as the scatter in the scaling relation of 0.55dex , are not taken into account . _ redshift . _ we use the peak of the emission line as measured by @xcite to obtain the redshift of the quasars . the redshifts derived this way have been corrected for the velocity shift of @xmath124kms@xmath39 as found between and the systemic redshift of lower redshift quasars as traced by the [ ] @xmath125 emission line @xcite . the uncertainty in this shift is added in quadrature to the statistical redshift uncertainty of the gaussian fit of the emission line . this source was selected with @xmath126 and @xmath127 ( see table [ qsocolors ] ) . with the ntt we observed the source for 5400s in @xmath77 and for 3600s in @xmath78 . the source was not detected in the @xmath77 image , implying @xmath128 and @xmath129 . combined with a detection in the @xmath78 image of @xmath130 , we could identify this object as a probable high redshift quasar with @xmath131 and rule out a foreground identification of the source ( figure [ nttccplot ] ) . the discovery spectrum taken with vlt / fors2 ( figure [ forsspectra ] ) shows a source with continuum detected in the red part and a sharp break between 9500 and 9600 . we interpreted this feature as the onset of absorption in the igm shortward of the ly@xmath27 line in the quasar and estimated a redshift of @xmath132 . the vlt / x - shooter spectrum ( figure [ irspecs ] ) shows several strong emission lines , such as @xmath1171397+]@xmath1171402 , @xmath1171549 , ] @xmath1171909 , and @xmath1172799 . a fit of the line produced a more accurate redshift of @xmath133 . the continuum blueward of the emission line is partly absorbed , suggesting that this quasar is a broad absorption line ( bal ) quasar . extrapolating the continuum above the absorption , we measure an absolute magnitude at 1450 in the rest - frame of @xmath134 , which is approximately 1mag fainter than ulas j1120 + 0641 at @xmath29 @xcite . the black hole mass that @xcite derive from the width of the emission line and the strength of the continuum is @xmath102=@xmath135@xmath10 . in our catalogs , this source was selected with colors @xmath136 and @xmath137 ( table [ qsocolors ] ) . we observed the source for 8100s in @xmath77 and 3600s in @xmath78 with the ntt . similar to j23483054 , j01093040 remained undetected in the deep @xmath77 image , suggesting a very red @xmath138 . the @xmath139 was slightly ( @xmath80.2mag ) bluer then that of j23483054 , again ruling out a foreground interpretation of the optical infrared colors . the fors2 spectrum revealed an object with a strong continuum break around 94009500 ( figure [ forsspectra ] ) , and we identified the source as a quasar at @xmath140 . the x - shooter spectrum ( figure [ irspecs ] ) shows several strong emission lines , including + , ] , and , from which we derive a redshift of @xmath141 . the ly@xmath27 emission line is completely absorbed . despite the lack of ly@xmath27 emission , the continuum at wavelengths shortward of is not absorbed and , unlike j23483054 , j01093040 is not a bal . from the x - shooter spectrum we can directly measure the flux density at 1450 in the rest frame . we find that the quasar has @xmath142 . the uncertainty in the absolute magnitude includes the moderate s / n of the @xmath42 and @xmath43 viking magnitudes that were used to derive the absolute flux calibration of the spectrum . the absolute magnitude of @xmath143 makes it the faintest quasar in our sample and 1.52.5mag fainter than the @xmath22 quasars found in the main sdss survey @xcite . although the quasar is fainter in the continuum , the black hole mass of @xmath102=@xmath144@xmath10 is similar to that of j23483054 @xcite . j03053150 was the only quasar in our sample that was detected in the viking @xmath41 image . the @xmath76 color of @xmath145 was red enough to fulfill the conservative color selection . over the three observing nights with the ntt we exposed the source for a total of 3600s through the @xmath77 filter and for 2700s through @xmath78 . while the @xmath146 color we measured was similar to that of the other two quasars ( @xmath147 ; table [ qsocolors ] ) , the @xmath139 was 0.20.4mag bluer . this was an indication that the redshift is lower than that of the other quasars . of the three new quasars , this object had the brightest ( observed ) magnitudes , being around 0.5mag brighter in @xmath43 than the other two quasars . the fors2 observations showed a quasar spectrum with a break around @xmath89200 , corresponding to a ly@xmath27 redshift of @xmath148 . the redshift was more accurately determined using the fire spectrum of the quasar . the infrared spectrum showed the same emission lines as in the other two viking quasars . a fit to the emission line gives a redshift of @xmath149 . the absolute magnitude of this quasar is @xmath150 making it the brightest of the new quasars , but still @xmath80.6mag fainter than ulas j1120 + 0641 . the wavelength coverage of the fire spectrum allowed @xcite to estimate the mass of the black hole . based on the fit to the continuum and on the gaussian modeling of the emission line , they derive a @xmath102=@xmath151@xmath10 . in this section we will compare our discovery of three quasars at @xmath28 in 331.6deg@xmath4 that was covered by viking with the expected number of quasars based on the quasar luminosity function at @xmath22 . we first estimate the completeness of the catalogs and the efficiency of the point source selection ( sections [ catcomp ] and [ pscomp ] ) . we then test the efficiency and completeness of our conservative and extended color selections ( section [ colcomp ] ) . in section [ qsolf ] we apply the completeness calculations to the number density of viking quasars and compare those with the predictions based on the @xmath14 quasar luminosity function extrapolated to @xmath28 . to assess how complete the viking catalogs are as a function of limiting s / n , we need to compare the detected number of sources in the viking catalogs with the expected number of sources . to determine the number of objects that are expected in the viking images , we make use of the first data release of the vista deep extragalactic observations ( video ) survey @xcite . the video data overlap with test data from the viking survey in the canada france hawaii - telescope - legacy survey deep-1 field . since the video survey is 2.22.6mag deeper than viking in all five vista bands and highly complete down to the limits of the viking survey ( see table 3 in * ? ? ? * ) , we can take the video catalogs to determine which fraction of the sources is recovered in the viking catalogs . the video area is partly covered by two different viking pointings , with the total overlap roughly 1deg@xmath4 . the two viking tiles that overlap the video region have similar characteristics ( average limiting magnitudes , seeing , etc . ) as those of a typical viking tile . our calculation of the completeness of the viking catalogs as a function of s / n , was done in the following way . first , we computed for a given s / n what the corresponding magnitude range was . we then counted the number of objects within that magnitude range in the video catalogs and compared that to the number of sources with a match in the viking catalogs . in figure [ sncompleteness](a ) we show the fraction of objects in the video @xmath42-band catalog that are recovered in the viking catalog . the viking catalogs are over 95% complete down to a s / n@xmath152 . the completeness as function of limiting s / n is plotted in figure [ sncompleteness](b ) . we computed the number of video sources above the magnitude corresponding to the viking s / n limit and counted the fraction of sources that had a match above the s / n limit in the viking catalog . we have six independent measurements of the completeness ( two viking pointings with three filters each ) , and the completeness above a s / n@xmath153 was in the range 94.4%97.9% with an average of 95.7% . based on the counts in the deeper video catalogs , we recover around 96% of the objects in the viking catalogs above our s / n cut of 7 . therefore , the catalog completeness correction is very small and will not have a strong influence on the results presented in this section . to remove foreground galaxies from our quasar color selection , we only selected point sources by forcing a pgalaxy@xmath59 ( section [ candidateselection ] ) . we tested the efficiency and completeness of this criterion in two different ways . first , we matched the viking data in the gama09 region with objects from the sdss database ( data release 8 ; * ? ? ? * ) . in a second test we used the video data to select a clean , photometric sample of stars in the region for which viking data exist . in the gama09 region , there are 3653 objects that were classified as point sources in the sdss imaging catalogs and had a spectroscopic classification `` star '' . in the same region , we selected 12618 objects that were classified as galaxies in both the imaging and spectroscopic sdss data . using our adopted point - source classifier , pgalaxy@xmath59 , we selected 3397 of the sdss point sources and only 9 of the galaxies ( 0.07% ) . based on these numbers , we estimate that our point - source completeness is 93.3% , and our sample has a purity of 99.7% . to obtain a clean sample of stars in the video data , we selected objects for which the morphological information from the five video images ( @xmath41 , @xmath42 , @xmath43 , @xmath44 , and @xmath45 ) indicate that the source is stellar ( i.e. , the catalog parameter `` mergedclass '' set to 1 or 2 ) . in addition , we required that the sources have optical infrared colors close to the stellar locus ( see e.g. figure 1 in * ? ? ? * ) . this selection resulted in 2446 stellar sources in the video catalog that had a match in the viking data . in the viking catalogs 2287 of the 2446 point sources had pgalaxy@xmath59 . this comparison implies that the viking catalogs have a point - source completeness of 93.5% . we find that the fraction of point sources misclassified in the viking catalogs is small ( 6%7% ) . combined with the catalog completeness , we estimate that we are missing 10.7% of the point sources in our search . to estimate which fraction of quasars we are missing due to our color criteria , we follow the approach by @xcite to clone sdss spectra of lower redshift quasars and shift them to higher redshifts . an alternative method is to create artificial quasar spectra based on an empirical model for the spectral energy distribution ( e.g. , * ? ? ? * ; * ? ? ? the main assumption of both methods is that the rest - frame uv spectral properties of quasars redward of the ly@xmath27 line do not evolve with redshift ( e.g. , * ? ? ? * ; * ? ? ? although the cloning of lower redshift quasar spectra could suffer from incompleteness due to the sdss quasar selection function ( e.g. , * ? ? * ; * ? ? ? * ) , we decided to mimic the procedure of @xcite as our goal is to compare our results to the @xmath22 quasar luminosity function presented in @xcite that was derived using the same color completeness correction . for the cloning we selected optical spectra of quasars from the fourth sdss quasar catalog @xcite . we needed to model colors of quasars in a redshift interval covering at least @xmath154 in the filters @xmath41 , @xmath42 and @xmath43 ( figure [ cc_zyj ] ) . given the wavelength range of the sdss spectra ( @xmath155 ) , we chose to use sdss quasars in the redshift range @xmath156 for the cloning . the sdss archives the spectra of 428 quasars over this redshift interval ; after culling for quality issues such as missing data and incorrect redshift estimates , this number is reduced to 405 spectra , which can be artificially redshifted . the absolute magnitudes of these quasars range from @xmath157 to @xmath158 with a median of @xmath159 , close to the average of our newly discovered quasars . the most significant difference between quasar spectra at @xmath160 and @xmath161 is the spectral energy distribution blueward of the ly@xmath27 transition due to absorption by neutral hydrogen . broadly speaking , as the volume averaged content of universe increases toward higher redshifts , absorption of photons in the line of sight redshifted to the wavelengths of ly@xmath27 , ly@xmath162 , and higher order transitions becomes stronger . when cloning high redshift quasar spectra from a lower redshift sample this difference must be accounted for . we applied artificial absorption to the sdss spectra at @xmath163 following the procedure of @xcite . in short , @xcite corrected for ly@xmath27 absorption already present in the spectra of sdss quasars using the igm transmission as a function of redshift as measured by @xcite . the transmission was sampled in large redshift intervals of @xmath164 , which acted to smooth out inhomogeneity of the igm and reduce the difference between the average optical depth per pixel and the effective optical depth . the spectra were then further corrected for ly@xmath162 absorption where the ly@xmath162 optical depth is given by @xmath165 @xcite . higher order absorption is negligibly small and was therefore left uncorrected . once these corrections were applied to the low - redshift spectra they were then artificially redshifted and artificial absorption was added blueward of ly@xmath27 and ly@xmath162 according to the same relation from @xcite . complete absorption of the spectra was applied blueward of the lyman limit ( 912 ) . the only difference between the procedure of @xcite and our procedure is that we have updated the transmission redshift relation to reflect new measurements recently published by @xcite for redshifts up to @xmath166 . the transmission functions from @xcite and @xcite are in good agreement at @xmath167 , so we have not attempted to join the two functions . each of the artificial spectra was redshifted to @xmath168 in steps of @xmath169 and multiplied by the transmission curves of the @xmath41 , @xmath42 , and @xmath43 filters . the average fluxes in the filters were converted to observed magnitudes . to take photometric errors into account in the color completeness , we added random magnitude errors as a function of the s / n . since each tile can have different relative depths between the @xmath41 , @xmath42 , and @xmath43 images , we computed the color completeness for each tile . an example of the resulting color completeness as a function of redshift is shown in figure [ tilecomp ] . the completeness in other tiles looks very similar . at high s / n ( i.e. , negligible magnitude errors ) , the completeness for the conservative color criteria rises steeply between @xmath170 and @xmath171 and remains above @xmath895% up to @xmath172 . at higher redshifts the completeness drops quickly to 10% at @xmath173 . this is because at @xmath174 the igm absorption cuts into the @xmath42 bandpass , and the model quasars move redder in @xmath89 . at @xmath175 the average @xmath89 color of the model quasars is @xmath176 , outside our selection region . at lower s / n the peak completeness drops to 75%80% for artificial quasars with ( s / n)@xmath177 , while the redshift distribution broadens slightly . at the high redshift end , the completeness of the extended color criteria is identical to the one of the conservative color criteria . due to the lower @xmath76 threshold in the extended color criteria , a higher fraction of artificial quasars at @xmath178 are selected . around a completeness of 30%40% the color completeness is almost independent of s / n . we therefore take the redshift range where the completeness is above 30% as the redshift range we probe with the viking data . for the conservative color criteria , the redshift range is @xmath179 , and for the extended criteria it is @xmath11 . the extended criteria are therefore sensitive for quasars down to the redshift where the highest redshift quasars are found in optical surveys ( e.g. , * ? ? ? the surveyed area of 331.6deg@xmath4and the probed redshift range of @xmath180 ( @xmath181 ) for the conservative ( extended ) color criteria translates into a survey volume of 2.7comoving gpc@xmath182 ( 2.9comoving gpc@xmath182 ) . to take the varying depth of the viking tiles ( figure [ areadepth ] ) into account , we computed the color completeness for each tile as a function of absolute magnitude @xmath115 of the model quasars . in figure [ avecomp ] we show the completeness averaged over all tiles as a function of both redshift and absolute magnitude . the color completeness at @xmath183 drops to 50% ( 30% ) for quasars with an absolute magnitude of @xmath184 ( 25.07 ) . we will now apply all the completeness corrections to our data and compare the results with the expected number of quasars based on the extrapolated luminosity function of @xmath14 quasars . the quasar luminosity function , @xmath185 ) , is often parameterized by a double power - law ( e.g. * ? ? ? * ) : @xmath186 where @xmath27 is the faint end slope , @xmath162 the bright end slope , @xmath187 the absolute magnitude , @xmath188 the break magnitude and @xmath189 the space density at @xmath14 of quasars with a magnitude @xmath188 . the factor @xmath190 accounts for the observed decline with redshift in the space density of quasars between @xmath191 and @xmath192 ( e.g. , * ? ? ? @xcite compiled a sample of 40 quasars at @xmath22 from various optical surveys to derive the @xmath22 quasar luminosity function . fixing the faint end slope to @xmath193 , based on the quasar luminosity function at lower redshift ( e.g. * ? ? ? * ) , @xcite determined the following best - fit parameters for the quasar luminosity function at @xmath14 : @xmath194mpc@xmath13mag@xmath39 , @xmath195 and @xmath196 . a value of @xmath197 was adopted by @xcite which fits the evolution of the ( bright end of the ) luminosity function over the redshift range @xmath198 @xcite . based on a study of @xmath167 quasars , @xcite derived a steeper decline in the space density of quasars with @xmath199 between @xmath200 and @xmath14 of a factor @xmath201 per unit redshift ( @xmath202 ) . if we assume that the quasar luminosity function evolves with constant parameters @xmath203 , @xmath162 and @xmath27 and pure density evolution @xmath204@xmath190 , then we can extrapolate the @xmath14 luminosity function to @xmath205 and derive the expected number of high - redshift quasars in the viking survey . assuming @xmath197 we expect 0.0124 quasardeg@xmath206 down to @xmath184 in the redshift range @xmath179 , while the extrapolated luminosity function predicts 0.0139 quasardeg@xmath206 in the redshift range @xmath11 . in an area of 331.6deg@xmath4 we therefore expect on average @xmath84.1 quasars between @xmath179 and @xmath84.6 quasars at @xmath11 if we ignore any incompleteness . the larger volume probed by the extended color criteria and the higher predicted space density of quasars at the lower end of our redshift range results in an increase of 13% in the expected number of quasars . with a sharper decline in the normalization of @xmath202 , the expected numbers are a factor of @xmath81.75 smaller . in figure [ lumfunct ] we show the cumulative luminosity function normalized over our surveyed volume and the completeness - corrected space density of @xmath15 quasars derived from our newly discovered viking quasars . although the follow - up of all quasar candidates has not been completed , the observed density of @xmath207 quasars is already fully consistent with the one predicted from the extrapolated @xmath14 luminosity function . a steeper decline in space density at @xmath192 as obtained by @xcite underpredicts the density of quasars at @xmath15 as found by our study . if we take the various sources of incompleteness into account , then with @xmath197 we expect to find @xmath83.1 @xmath207 quasars in our volume down to @xmath184 . assuming @xmath202 we expect 1.8 quasars in the same volume . therefore , given the large ( poisson ) error on these estimates , we can not rule out the steeper decline . to estimate for which values of @xmath208 we would expect to find three quasars in our survey volume , we extrapolated the @xmath14 luminosity function to @xmath11 with @xmath208 ranging from @xmath603 to 0 in steps of 0.01 . for each value of @xmath208 , we integrated the luminosity function down to @xmath184 and , including the various sources of incompleteness , computed the expected number of quasars in our probed volume . the best - fit value of @xmath208 for which the expected number is three quasars and the 1@xmath209 confidence interval is @xmath210 . the best - fit value of @xmath211 is close to the @xmath197 @xcite adopts . the best - fit value obtained by @xcite , @xmath202 , is nonetheless well within the range of @xmath208 that fit our results . we used the viking near - infrared imaging survey to search for quasars above a redshift @xmath57 . we analyzed 332deg@xmath4 of viking data down to a median 7@xmath209 @xmath42-band depth of @xmath212 . we employed two sets of color criteria to select potential quasars at high redshift . our conservative color criteria is sensitive to quasars in the redshift range @xmath179 and with these criteria we selected 43 quasar candidates . our extended color criteria probed an increased redshift range of @xmath11 and selected an additional 42 objects . after we obtained optical imaging of 45 quasar candidates with the ntt , 6 objects remained having colors consistent with high redshift quasars . spectroscopic follow - up with vlt / fors2 of four of these sources identified three previously unknown quasars with redshifts of @xmath114 . the fourth object was a foreground star . the newly discovered quasars have the following characteristics : 1 . the absolute magnitudes of the new quasars are in the range of @xmath213 , which is about 12mag fainter than the @xmath22 quasars discovered in the sdss main survey @xcite and roughly similar to the quasars at @xmath22 found in the deep stripe of the sdss southern survey @xcite . the only other @xmath15 quasar published so far , j1120 + 0641 at @xmath29 , has an absolute magnitude of @xmath214 @xcite , 0.61.1mag brighter than the viking quasars presented in this article . 2 . near - infrared spectroscopy revealed that the new viking quasars exhibit emission lines commonly seen in lower redshift quasars , including @xmath1171397+]@xmath1171402 , @xmath1171549 , ] @xmath1171909 , and @xmath1172799 . furthermore , j23483054 shows continuum absorption blueward of the emission line and could be a bal quasar . 3 . using the infrared spectra , @xcite determined the mass of the central black hole powering the quasars . by fitting a gaussian to the line and measuring the continuum luminosity , they estimate central black hole masses of @xmath215@xmath10 . these black hole masses are very similar to those powering quasars at @xmath22 that were discovered in the sdss main survey ( e.g. , * ? ? ? * ; * ? ? ? * ) , but generally more massive than those of the @xmath22 quasars discovered in the cfhqs @xcite . ignoring any incompleteness in our quasar search we can set a lower limit to the density of supermassive black holes in the early universe . the discovery of three black holes with masses @xmath216@xmath10 in the 2.9comoving gpc@xmath182 probed in this study implies a minimum density of @xmath12mpc@xmath13 . to derive the space density of quasars in the volume probed by the viking data , we computed the efficiency of our selection criteria in order to correct for the incompletenesses in our quasar search . the volume density of quasars at @xmath11 that we derive agrees well with the extrapolated @xmath14 quasar luminosity function . this includes a decline in the space density of quasars of a factor 3 per unit redshift , which was previously obtained by measuring the density of bright quasars between @xmath198 . our results do not support a steeper decline in the space density of quasars with redshift . these findings imply that we should discover a significant number of quasars beyond @xmath57 in the remainder of the viking area . the total area that will be covered by viking will be 1500deg@xmath4 . if we ignore any incompleteness , we expect 1819 quasars between @xmath179 and @xmath821 quasars at @xmath11 in the viking area down to @xmath217 . with a larger number of quasars we will be able to constrain the evolution of the quasar luminosity function at @xmath15 . these new quasars are prime targets for various follow - up studies . in future articles we will study the emission lines that can be seen in the near - infrared spectra and examine the implications on the chemical enrichment near the black holes . the optical spectra will be used to derive constraints on the column density of neutral hydrogen along the line of sight . furthermore , the southern declination of the sources makes them ideal targets for alma observations to study the galaxies that host these quasars . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy office of science . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . , g. b. , caldwell , m. , ward , a. k. , et al . 2006 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 6269 , society of photo - optical instrumentation engineers ( spie ) conference series , m. j. , lewis , j. , hodgkin , s. , et al . 2004 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 5493 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . p. j. quinn & a. bridger , 411 , j. r. , irwin , m. , & bunclark , p. 2010 , in astronomical society of the pacific conference series , vol . 434 , astronomical data analysis software and systems xix , ed . y. mizumoto , k .- morita , & m. ohishi , 91 , j. r. , irwin , m. j. , hodgkin , s. t. , et al . 2005 , in astronomical society of the pacific conference series , vol . 347 , astronomical data analysis software and systems xiv , ed . p. shopbell , m. britton , & r. ebert , 599 , a. , goldoni , p. , royer , f. , et al . 2010 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 7737 , society of photo - optical instrumentation engineers ( spie ) conference series , r. a. , burgasser , a. j. , bernstein , r. a. , et al . 2008 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 7014 , society of photo - optical instrumentation engineers ( spie ) conference series , r. a. , burgasser , a. j. , bochanski , j. j. , et al . 2010 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 7735 , society of photo - optical instrumentation engineers ( spie ) conference series

the circuit recently proposed by murali , lakshmanan and chua ( mlc ) is one of the simplest non - autonomous nonlinear electronic circuits which shows a variety of dynamical phenomena including various bifurcations , chaos and so on . in this paper we study the spatio - temporal dynamics in one and two dimensional arrays of coupled mlc circuits both in the absence as well as in the presence of external periodic force . in the absence of any external force , the propagation phenomena of travelling wave front and its failure have been observed from numerical simulations . we have shown that the propagation of travelling wave front is due to the loss of stability of the steady states ( stationary front ) via subcritical bifurcation coupled with the presence of necessary basin of attraction of the steady states . we also study the effect of weak coupling on the propagation phenomenon in one dimensional array which results in the blocking of wave front due to the existence of a stationary front . further we have observed the spontaneous formation of hexagonal patterns ( with penta - hepta defects ) due to turing instability in the two dimensional array . we show that a transition from hexagonal to rhombic structures occur by the influence of external periodic force . we also show the transition from hexagons to rolls and hexagons to breathing ( space time periodic oscillations ) motion in the presence of external periodic force . we further analyse the spatio - temporal chaotic dynamics of the coupled mlc circuits ( in one dimension ) under the influence of external periodic forcing . here we note that the dynamics is critically dependent on the system size . above a threshold size , a suppression of spatio - temporal chaos occurs , leading to a space - time regular ( periodic ) pattern eventhough the single mlc circuit itself shows a chaotic behaviour . below this critical size , however , a synchronization of spatiotemporal chaos is observed .

patt-sol9708003

coupled nonlinear oscillators in the form of arrays arise in many branches of science . these coupled arrays are often used to model very many spatially extended dynamical systems explaining reaction - diffusion processes . one can visualize such an array as an assembly of a number of subsystems coupled to their neighbours , or as coupled nonlinear networks ( cnns ) [ chua & yang , 1988 ; chua , 1995 ] . theoretically one can consider a system of coupled ordinary differential equations ( odes ) to represent a macroscopic system , in which each of the odes corresponds to the evolution of the individual subsystem . specific examples are the coupled array of anharmonic oscillators [ marn & aubry , 1996 ] , josephson junctions [ watanabe _ et al . _ , 1995 ] , continuously stirred tank reactors ( cstr ) exhibiting travelling waves [ dolnik & marek , 1988 ; laplante & erneux , 1992 ] , propagation of nerve impulses ( action potential ) along the neuronal axon[scott , 1975 ] , the propagation of cardiac action potential in the cardiac tissues [ allesie _ et al . _ , 1978 ] and so on . for the past few years several investigations have been carried out to understand the spatio - temporal behaviours of these coupled nonlinear oscillators and systems . the studies on these systems include the travelling wave phenomena , turing patterns , spatio - temporal chaos and synchronization [ feingold _ et al . _ , 1988 ; muuzuri _ et al . _ , 1993 ; prez - muuzuri _ et al . _ , 1993 ; sobrino _ et al . _ , 1993 ; prez - muuzuri _ et al . _ , 1994 , prez - muuzuri _ et al . _ , 1995 ; kocarev & parlitz , 1996 ] . one of the interesting behaviours of these coupled oscillators is that they can exhibit a failure mechanism for travelling waves which can not be observed in the homogeneous continuous models [ keener , 1987 ] . similar failure phenomena in the nerve impulse propagation have been observed in biological experiments [ balke _ et al . _ , 1988 ; cole _ et al . _ , 1988 ] . recently various spatio - temporal patterns have been found in discrete realizations of the reaction - diffusion models by means of arrays of coupled nonlinear electronic circuits [ prez - muzuri _ et al . _ , 1993 ] . for example prez - muuzuri and coworkers studied the various spatio - temporal patterns in a model of discretely coupled chua s circuits . this system is a _ three _ variable model and coupled to its neighbours by means of a linear resistor . in this paper we consider similar arrays of coupled nonlinear electronic circuits consisting of murali - lakshmanan - chua ( mlc ) circuits as a basic unit , which is basically a _ two _ variable model . the dynamics of the driven mlc circuit including bifurcations , chaos , controlling of chaos and synchronization has been studied by murali _ [ 1994a ; 1994b ; 1995 ; lakshmanan & murali , 1996 ] . of particular interest among coupled arrays is the study of _ diffusively coupled driven systems _ as they represent diverse topics like faraday instability [ lioubashevski _ et al . _ , 1996 ] , granular hydrodynamics [ umbanhawar _ et al . _ , 1996 ; kudrolli _ et al . _ , 1997 ] , self organized criticality [ bak & chen , 1991 ] and so on . identification of localized structures in these phenomena has been receiving considerable attention very recently [ fineberg , 1996 ] . under these circumstances , study of models such as the one considered in this paper can throw much light on the underlying phenomenon in diffusively coupled driven systems . in the following by considering one and two dimensional arrays of coupled mlc circuits , we have studied the various spatio - temporal patterns in the presence as well as in the absence of external force . we present a brief description on the one and two dimensional arrays of mlc circuits in sec . then , in sec . [ sec3 ] , in the absence of periodic external force , we study the propagation phenomenon of travelling wave front and its failure below a critical diffusive coupling coefficient . we show that the propagation phenomenon occurs due to a loss of stability of the steady states via subcritical bifurcation coupled with the presence of necessary basin of attraction of the steady states for appropriate diffusive coupling coefficient . we also study the effect of weak coupling on the propagation phenomenon which causes a blocking in the propagation due to the presence of a stationary front at the weakly coupled cell . in addition we also study the onset of turing instability leading to the spontaneous formation of hexagonal patterns in the absence of any external force . further as mentioned earlier it is of great physical interest to study the dynamics of the coupled oscillators when individual oscillators are driven by external forces . in sec . [ sec4 ] , we study the spatio - temporal patterns in the presence of periodic external force and investigate the effect of it on the propagation phenomenon and turing patterns . depending upon the choice of control parameters , a transition from hexagons to regular rhombic structures , hexagons to rolls and then to breathing oscillations from hexagons are observed . the presence of external force with sufficient strength removes the penta - hepta defect pair originally present in the spontaneously formed hexagonal patterns leading to the formation of regular rhombic structures . the inclusion of external periodic force can also induce a transition from hexagons to rolls provided there are domains of small roll structures in the absence of force . we further show that in the region of hopf - turing instability , the inclusion of external periodic force with sufficiently small amplitude induces a type of breathing oscillations though the system shows a regular hexagonal pattern in the absence of any external force . finally , we also study the spatio - temporal chaotic dynamics of the one dimensional array of mlc circuits in sec . [ sec5 ] when individual oscillators oscillate chaotically . in this case , the emergence of spatio - temporal patterns depends on the system size . for larger size , above a critical number of cells , we observe a controlled space - time regular pattern eventhough the single mlc circuit itself oscillates chaotically . however , synchronization occurs for a smaller system size , below the threshold limit . finally , in sec . [ sec6 ] we give our conclusions . as mentioned in the introduction the circuit proposed by murali , lakshmanan and chua ( mlc ) is one of the simplest second order dissipative nonautonomous circuit having a single nonlinear element [ murali _ et al_. , 1994a ] . brief details of its dynamics are given in the appendix a for convenience . here we will consider one and two dimensional arrays of such mlc circuits , where the intercell couplings are effected by linear resistors . [ fig1](a ) shows a schematic representation of an one dimensional chain of resistively coupled mlc circuits . extending the analysis of single mlc circuit as given in the appendix a , the dynamics vs. @xmath0 for the parametric choice @xmath1 @xmath2 in eq . ( c ) graph showing @xmath3 vs. @xmath0 for the parametric choice @xmath4 @xmath5 in eq . ( 2 ) . ] of the one dimensional chain can be easily shown to be governed by the following system of equations , in terms of the rescaled variables ( see appendix a ) , [ eqn1all ] @xmath6 where @xmath7 is the diffusion coefficient , @xmath8 is the chain length and @xmath3 is a three segment piecewise linear function representing the current voltage characteristic of the chua s diode , @xmath9 in ( [ eqn2 ] ) @xmath10 , @xmath11 and @xmath12 are the three slopes . depending on the choice of @xmath10 , @xmath11 and @xmath12 one can fix the characteristic curve of the chua s diode . here @xmath13 corresponds to the dc offset . of special importance for the present study are the two forms of @xmath3 shown in figs . [ fig1](b ) and [ fig1](c ) . [ fig1](b ) corresponds to the choice of the parameters @xmath14 while fig . [ fig1](c ) corresponds to the parameters @xmath15 . as noted in the appendix a , the significance of the characteristic curve in fig . [ fig1](b ) is that it admits interesting bifurcations and chaos , while fig . [ fig1](c ) corresponds to the possibility of bistable states with asymmetric nature which is a prerequisite for observing wave front propagation and its failure . in the following studies we will consider a chain of @xmath16 mlc circuits . as in the case of one dimensional array one can also consider a two dimensional array with each cell in the array being coupled to four of its nearest neighbours with linear resistors . the model equation can be now written in dimensionless form as [ eqn3all ] @xmath17 this two dimensional array has @xmath18 cells arranged in a square lattice . in our numerical study we will again take @xmath16 . in the following sections we present some of the interesting dynamical features of the array of mlc circuits such as propagation of wave front and its failure , effect of weak coupling in the propagation , turing patterns , effect of external periodic forcing in the turing patterns and spatio - temporal chaotic dynamics . we have used zero flux boundary conditions for the study of propagation phenomenon and turing patterns and periodic boundary conditions for the study of spatio - temporal chaos . transport processes in living tissues , chemical and physiological systems have been found to be associated with special types of waves called travelling waves . earlier , continuous models were created to describe the wave propagation phenomenon in these systems , but these failed to cover all the important aspects of travelling wave behaviour . one of the most important of them is the so called travelling wave propagation failure , occurring at weak coupling between cells . it has been proved by keener [ 1987 ] that propagation failure can not be observed in a continuous , one variable , homogeneous reaction - diffusion system . therefore to study this kind of phenomenon , one has to use discrete models . recently travelling wave phenomenon and its failure have been studied in arrays of discretely coupled chua s circuits [ prez - muuzuri _ et al . _ , wave propagation and its failure have also been observed even in one variable models like discrete nagumo equation [ leenaerts , 1997 ] . in this section we carry out a study of such wave propagation phenomenon in one and two dimensional arrays of murali - lakshmanan - chua s ( mlc ) circuit oscillators , without forcing , and investigate the mechanism by which such a failure occurs . in order to observe wave front propagation in the one dimensional array of mlc circuits , we numerically integrated the eqs . ( 1 ) using fourth order runge - kutta method . in this analysis we fix the parameters at @xmath19 @xmath20 so that the system admits bistability and this choice leads to the asymmetry characteristic curve for the chua s diode as shown in fig . [ fig1](c ) . the existence of bistable state in the asymmetric case is necessary to observe a wave front . zero flux boundary conditions were used in the numerical computations , which in this context mean setting @xmath21 and @xmath22 at each integration step ; similar choice has been made for the variable @xmath23 also . to start with , we will study in this section the force - free case , @xmath24 , and extend our studies to the forcing case @xmath25 in the next section . the choice of the values of the parameters guarantees the existence of two stable equilibrium points @xmath26 , @xmath27 and @xmath28 , @xmath29 for each cell . in the particular case , corresponding to the above parametric choice , each cell in the array has three equilibrium points at @xmath30 , @xmath31 and @xmath32 . out of these three equilibrium points , @xmath33 and @xmath34 are stable while @xmath35 is unstable . due to the asymmetry in the function @xmath3 for the present parametric choices , the basin of attraction of the point @xmath33 is much larger than that of @xmath34 and it is harder to steer a trajectory back into the basin @xmath34 once it is in the basin of @xmath33 . now we choose an initial condition such that the first few cells in the array are excited to the @xmath36 state ( having a large basin of attraction compared to that of @xmath37 ) and the rest are set to @xmath37 state . in other words a wave front in the array is obtained by means of the two stable steady states . on actual numerical integration of eqs.([eqn1all ] ) with @xmath16 and with the diffusion coefficient and ( b ) propagation failure for @xmath38 . ] chosen at a higher value , @xmath39 , a motion of the wave front towards right ( see fig . [ fig2](a ) ) is observed , that is a travelling wave front is found . after about 80 time units the wave front reaches the @xmath40 cell so that all the cells are now settled at the more stable state @xmath41 as demonstrated in fig . [ fig2](a ) . when the value of @xmath7 is decreased in steps and the analysis is repeated , the phenomenon of travelling wavefronts continues to be present . however , below a critical value of the diffusion coefficient @xmath42 a failure in the propagation has been observed , which in the present case turns out to be @xmath43 . this means that the initiated wavefront is unable to move as time progresses and fig . [ fig2](b ) shows the propagation failure for @xmath38 . to understand the failure mechanism , we have carried out a detailed case study of an array of five coupled mlc circuits , represented by the following set of equations we have verified our conclusions for the case of 3 , 4 , 6 and 7 oscillators also . the analysis can be in principle extended to arbitrary but finite number of oscillators . from eq . ( [ eqn1all ] ) , the present array can be represented by the following set of ten coupled first order odes : @xmath44 where @xmath3 is as given in eq . ( [ eqn2 ] ) and the parameters are fixed at @xmath45 , @xmath46 , @xmath47 , @xmath48 , @xmath49 and @xmath50 . by carrying out a numerical analysis of the system ( [ eqsta ] ) with zero flux boundary conditions , we identify the following main results . [ [ casea ] ] we solve the above system of equations ( [ eqsta ] ) for the chosen initial condition @xmath51 @xmath52 @xmath53 @xmath54 @xmath55 @xmath56 @xmath57 @xmath58 @xmath59 @xmath59 @xmath60 and @xmath61 , @xmath62 for different values of the diffusion coefficient @xmath7 in the region @xmath63 . note that the values @xmath64 and @xmath65 correspond to the stable equilibrium points @xmath36 and @xmath37 , respectively , of the uncoupled mlc oscillators . when @xmath7 is decreased downwards from @xmath66 we find that in the region @xmath67 propagation of wavefront occurs as in fig . [ fig2](a ) ( of the @xmath16 cells system ) . at the critical value @xmath38 , the wavefront fails to propagate and asymptotically ends up in a stationary wavefront which is nearer to the initial condition ( as in fig . [ fig2](b ) ) . the wavefront failure phenomenon persists for all values of @xmath68 . [ [ caseb ] ] on the other hand , if we choose the initial condition as @xmath51 @xmath52 @xmath53 @xmath54 @xmath55 @xmath56 @xmath69 @xmath70 @xmath71 @xmath72 @xmath73 and @xmath61 , @xmath74 then the critical value @xmath7 , where wavefront propagation failure occurs , turns out to be at a slightly higher value @xmath75 . in order to understand the failure mechanism of wavefront propagation we consider all the possible steady states ( equilibrium states ) associated with eqs.([eqsta ] ) for nonzero values of @xmath7 . it is easy to check that eq . ( [ eqsta ] ) possesses a maximum of @xmath76 steady states . first we consider the three trivial steady states which can be easily obtained by considering @xmath77 and @xmath78 , namely , @xmath79 @xmath56 @xmath80 , @xmath81 @xmath56 @xmath82 and @xmath83 @xmath56 @xmath84 . ( here the suffix @xmath85 indicates that the steady states are the same as that of the @xmath86 case , namely @xmath36 , @xmath87 and @xmath37 ) . from a linear stability analysis ( details are given in appendix b ) we can easily conclude that the steady states @xmath79 and @xmath83 are stable while @xmath81 is unstable irrespective of the value of the diffusion coefficient @xmath7 . then any initial condition in the neighbourhood of @xmath88 will evolve into @xmath88 asymptotically . also one has to note that , due to asymmetry in the function @xmath3 , the basin of @xmath79 is much larger than that of @xmath83 . since we are interested here to understand the propagation failure mechanism , we concentrate on those steady states which possess stationary wavefront structures _ near _ to the two types of chosen initial conditions in the numerical analysis discussed in sec . [ secnum ] . out of the 243 available steady states , we select a subset of them , @xmath89 , which is nearer to the numerically chosen initial state . these three steady states are defined in eqs.([steady ] ) of appendix b. they are identified such that the components @xmath90 and @xmath91 are in the right extreme segment of the characteristic curve , fig . [ fig1](c ) , while @xmath92 and @xmath93 are in the left extreme segment . on the other hand @xmath94 is allowed to be in any one of the three segments . the reason for such a choice is that we are looking for the formation of a wavefront which is closer to the chosen initial conditions in sec . [ secnum ] . the corresponding forms of @xmath95 as given in eq . ( [ eqn2 ] ) are then introduced in eqs.([eqsta ] ) to find the allowed steady states . then @xmath96 corresponds to the case in which @xmath94 as well as @xmath90 and @xmath91 are in the right most segment while the rest are in the left most segment and @xmath97 and @xmath98 correspond to the cases in which @xmath94 is , respectively , in the middle and the left extreme segments of the characteristic curve while @xmath90 , @xmath91 and @xmath92 , @xmath93 are chosen as in @xmath96 . these steady states may be explicitly obtained as a function of the diffusion coefficient @xmath7 , whose forms are given in appendix b. we now consider the linear stability of these steady states @xmath96 , @xmath97 and @xmath98 . the corresponding stability ( jacobian ) matrices are obtained from the linearized version of eq . ( [ eqsta ] ) , whose form is given in eq . ( [ jacob ] ) . the eigenvalues are evaluated by numerical diagonalization of ( [ jacob ] ) as a function @xmath7 and the stability property analysed . it has been found that out of these three steady states both @xmath96 and @xmath98 are stable for @xmath99 in which all of the eigenvalues of ( [ jacob ] ) are having negative real parts , while @xmath97 is unstable for all values of @xmath100 . however , the real part of atleast one of the eigenvalues associated with each of @xmath96 and @xmath98 changes its sign to positive value at @xmath101 and thereby both the stationary fronts , @xmath96 as well as @xmath98 , also lose their linear stability at this critical value . thus for @xmath102 , all the three steady states @xmath96 , @xmath97 and @xmath98 are linearly unstable . first let us consider the second of the initial conditions chosen in sec . [ secnum ] . at and above the critical value @xmath101 , all the three steady states are unstable and so the system transits to the other nearby steady states . these are also found to be unstable by a similar analysis , ultimately ending up in the only nearby available stable state which is @xmath79 @xmath56 @xmath103 @xmath104 @xmath104 @xmath104 @xmath105 , thereby explaining the propagation of the wavefront starting from the initial condition @xmath106 @xmath56 @xmath107 @xmath108 @xmath109 @xmath110 @xmath111 @xmath112 ( which is the second of the numerical results given in sec . [ secnum ] ) . however for @xmath113 , since @xmath96 is stable and is also quite closer to the initial state , the system settles down in this state itself and no propagation occurs , thereby explaining the phenomenon of propagation failure as a subcritical bifurcation when the diffusion coefficient @xmath7 is increased from @xmath114 upwards . note that eventhough @xmath98 is also stable , it is far away from the initial state and so no transition to this state will occur . a table of these steady states @xmath96 and @xmath98 as a function of @xmath7 ( table [ table2 ] ) is given confirming that @xmath96 is always nearer to the initial state compared to @xmath98 . now let us consider the first of the initial conditions of our numerical analysis ( see sec . [ secnum ] ) , where wavefront propagation occurs at @xmath38 itself and not at @xmath101 , eventhough the steady states @xmath96 and @xmath98 are still linearly stable in the region @xmath115 . in order to understand this aspect , we start analysing the nature of the basin of attraction associated with the steady state @xmath98 for different @xmath7 values numerically ( note that for the present set of initial conditions @xmath98 is nearer to it than @xmath96 ) . we find that the basin of @xmath98 shrinks as @xmath7 is increased from @xmath114 upwards and vanishes completely at @xmath101 . we also find that the chosen initial condition , @xmath106 @xmath56 @xmath103 @xmath104 @xmath116 @xmath116 @xmath117 falls fully within the basin of @xmath98 as long as @xmath118 . however due to the for various values of the diffusive coupling coefficient . in the figure each line corresponds to an initial condition . the figure is drawn by fixing the initial state of the first oscillator at @xmath119 , @xmath120 and for clarity decreasing the rest downwards in steps of 0.2 . ] shrinking nature of the basin as @xmath7 increases ( as given in figs . ( [ figbas ] ) ) , a part of the initial state falls outside the basin for @xmath67 ( figs . [ figbas](b)-(d ) ) . as a result the chosen initial condition does not end up in the stationary front @xmath98 for @xmath121 indicating a global instability ( eventhough the state is linearly stable ) . therefore propagation starts to occur as soon as @xmath67 , while it fails for @xmath122 and it finally ends in the stable state @xmath79 described in the previous case . however , if an initial condition which happens to fall completely within the basin of attraction of @xmath123 , the propagation starts to occur only for values of @xmath102 . in fact this is what happens for the second chosen initial condition ( @xmath106 @xmath56 @xmath124 @xmath109 @xmath110 @xmath111 @xmath112 , as explained above ( note that the basin structure shown in fig . [ figbas ] does not cover this case ) . at various time units ( a ) @xmath125 , ( b ) @xmath126 , ( c ) @xmath127 and ( d ) @xmath128 for @xmath129 and @xmath130 and ( e ) the same plot for propagation failure case @xmath131 and @xmath130 at @xmath132 . ] in conclusion , the propagation failure of wave fronts in arrays of coupled nonlinear oscillators is essentially due to the existence of stationary fronts ( stable steady states ) near to the initial states for low diffusion coefficients , below a certain critical value , which lose stability via subcritical bifurcation . this coupled with the existence of necessary basin of attraction ( for a global stability ) of the steady state gives rise to the specific critical value of @xmath133 above which propagation begins for a given initial condition . for studying the propagation phenomenon in two dimensional array of coupled mlc circuits , we have considered the case in which @xmath24 , @xmath134 and @xmath130 . the propagation of wave front and its failure have been observed as in the case of one dimensional array . a two dimensional array of @xmath135 coupled mlc circuits has been considered for the numerical simulation with initial conditions chosen such that the first few cells in the array are excited to the @xmath33 state and the other cells are set at the @xmath34 state . figs . [ fig3](a)-[fig3](d ) show the propagation of the wave front in the two dimensional array for @xmath129 and @xmath130 at various time units . again , below a certain critical diffusion coefficient , @xmath131 , propagation failure occurs . one such case is illustrated in fig . [ fig3](e ) for @xmath131 and @xmath130 . the phenomenon can again be explained through a linear stability analysis as in the case of the one dimensional array . in the above , the investigation has been made by considering the system as an ideal one ( as far as the circuit parameters are concerned ) . but from a practical point of view , there are defects in the coupling parameters which may result in a weak coupling at any of the cell in the array . in the following we study the effect of such weak coupling on the propagation of wave front . let us consider a weak coupling at the @xmath136 cell . by this we mean that the @xmath136 cell in the array is coupled to its nearest neighbour @xmath137 and @xmath138 cells by resistors with slightly higher values than that of the others . we have studied the effect of this defect on the propagation of wave fronts . numerical simulations have been carried out by considering an one dimensional array with 100 cells , where the initial conditions are chosen as in the case of propagation phenomenon in regular one dimensional arrays ( see sec . [ sec3a ] ) . from the numerical simulation results , we find that there is an abrupt stop in the propagation when the wave front reaches the weakly coupled cell . this happens when the coupling coefficient on either side of the @xmath136 cell in the array has a value even above the critical value @xmath139 for propagation failure discussed in sec . [ sec3a ] . cell in the one dimensional array ( eq . ( 3 ) ) . ( a ) three dimensional space - time plot and ( b ) density plot . ] [ fig4 ] shows a blocking in the propagation of the wave front when the coupling coefficients on either side of the @xmath136 cell @xmath140 , which we call as @xmath141 , is set to @xmath142 with the rest of the coupling coefficients set to @xmath143 @xmath144 . we observe that the actual blocking occurs when the wave front reaches the @xmath136 cell . we can say this is a kind of failure in the propagation ( because the wavefront will never reach the last cell ( that is the @xmath40 cell in the array ) by means of blocking . as far as the initial conditions are concerned , since the system is in the propagation region @xmath145 any choice of initial condition of wavefront structure will lead to uniform propagation upto the weakly coupled cell . now our motivation is to find the property of the above uniform wavefront when it reaches the weakly coupled cell . one can analyse the above phenomenon by means of a linear stability analysis of the steady states as in the case of propagation phenomenon in sec . [ sec3b ] . we consider again as an example the system of five coupled mlc circuits as in eqs.([eqsta ] ) in which the middle cell ( @xmath146 ) is now taken as an example to be coupled weakly to its neighbours , respectively @xmath147 and @xmath148 . this set up can be represented by the following set of ten coupled odes : @xmath149 here @xmath141 corresponds to the coupling coefficient of the weakly coupled cell and other parameters are the same as in ( [ eqsta ] ) . as in sec . [ sec3b ] , we calculated the steady states associated with ( [ eqweak ] ) and analysed the linear stability properties . let us now consider the three steady states of the system ( [ eqweak ] ) , namely , @xmath89 which possess wavefront structures near the chosen initial conditions . these steady states can be found as discussed in sec . [ sec3b ] . by fixing @xmath150 , we analyse the stability of @xmath151 as a function of @xmath141 . we find that , the steady states , @xmath123 are linearly stable for values of @xmath141 , @xmath152 and become unstable for @xmath153 via subcritical bifurcation . however @xmath97 state continues to be unstable irrespective of the value of @xmath141 . at this point , one may think of an array with large number of mlc circuits . since all the cells , except the @xmath136 cell , have diffusion coefficient @xmath150 which is in the propagation region , the appropriately chosen initial condition will evolve as a propagating front upto the weakly coupled cell . when this wave front reaches the weakly coupled cell , the existence of the stationary front @xmath98 for the array due to the presence of the weakly coupled cell will block the propagating front and hence it will end up in the stationary front @xmath98 asymptotically for @xmath154 . for @xmath155 , the stationary state @xmath98 is no longer stable and hence the propagation occurs without any blocking . thus the blocking of the wave front can also be attributed to the existence of a stable stationary front when weak coupling is present in the system . another interesting dynamical phenomenon in the coupled arrays is the formation of turing patterns . these patterns are observed in many reaction diffusion systems when a homogeneous steady state which is stable due to small spatial perturbations in the absence of diffusion becomes unstable in the presence of diffusion[turing , 1952 ] . to be specific , the turing patterns can be observed in a two variable reaction - diffusion system when one of the variables diffuses faster than the other and undergo turing bifurcation , that is , diffusion driven instability[turing , 1952 ; murray , 1989 ] . treating the coupled array of mlc circuits as a discrete version of a reaction - diffusion system , one can as well observe the turing patterns in this model also . for this purpose , one has to study the linear stability of system ( 3 ) near the steady state . in continuous systems , the linear stability analysis is necessary to arrive at the conditions for diffusion driven instability . a detailed derivation of the general conditions for the diffusion driven instability can be found in murray[1989 ] . for discrete cases one can follow the same derivation as in the case of continuous systems by considering solutions of the form @xmath156 [ murray 1989 ; muuzuri _ et al . _ , 1995 ] . here @xmath157 and @xmath158 are considered to be independent of the position @xmath159 @xmath160 . for eq . ( 3 ) , the criteria for the diffusion driven instability can be derived by finding the conditions for which the steady states in the absence of diffusive coupling are linearly stable and become unstable when the coupling is present . one can easily derive from a straightforward calculation that the following conditions should be satisfied for the general reaction - diffusion system of the form given by eq . ( [ eqn3all ] ) : @xmath161 the critical wave number for the discrete system ( 3 ) can be obtained as @xmath162 combining eqs . ( [ cond:1])-([cond:2 ] ) , one obtains[muuzuri _ et al . _ , 1995 ] the condition for the turing instability such that @xmath163 we apply these conditions to the coupled oscillator system of the present study . for this purpose , we fix the parameters for the two - dimensional model ( 3 ) as @xmath164 @xmath165 and verify that this choice satisfies the conditions ( [ cond:1 ] ) to ( [ cond:3 ] ) . the numerical simulations are carried out using an array of size @xmath166 and random initial conditions near the steady states are chosen oscillators for the parameters @xmath167 , @xmath46 , @xmath168 , @xmath169 , @xmath170 , @xmath171 , @xmath172 , @xmath173 , @xmath174 and @xmath175 in eq . ( 3 ) at various time units ( a ) @xmath126 , ( b ) @xmath127 , ( c ) @xmath176 and ( d ) @xmath177 ( e ) the same figure ( d ) in the fourier space . ] for the @xmath0 and @xmath23 variables . [ fig5](a)-[fig5](d ) show how the diffusion driven instability leads to stable hexagonal pattern ( fig . [ fig5](d ) ) after passing through intermediate stages ( figs . [ fig5](a)-[fig5](c ) ) . [ fig5](e ) shows the two dimensional fourier spectrum of the hexagonal pattern ( fig . [ fig5](d ) ) . further , the spontaneously formed patterns are fairly uniform hexagonal patterns having a penta - hepta defect pair . these defects are inherent and very stable . we note that such turing patterns have already been observed in an array of coupled chua s circuits[muuzuri _ et al _ , 1995 ] . however , our aim here is to investigate the effect of external forcing on these patterns , which is taken up in the next section . the effect of external fields on a variety of dynamical systems has been studied for a long time as driven systems are very common from a practical point of view . for example , in a large number of dynamical systems including the duffing oscillator , van der pol oscillator and the presently studied mlc circuits , temporal forcing leads to a variety of complex dynamical phenomena such as bifurcations , quasiperiodicity , intermittency , chaos and so on[guckenheimer & holmes , 1983 ; hao bai - lin , 1984 ; lakshmanan & murali,1996 ] . also the studies on the effect of external fields in spatially extended systems have been receiving considerably increasing interest in recent times [ pismen , 1987 ; walgraef , 1988 ; 1996 ] . particularly , with the recent advances in identifying localized and oscillating structures and other spatio - temporal patterns in driven nonlinear dissipative systems such as granular media , driven ginzburg - landau equations and so on , it is of considerable interest to study the effects of forcing on array of coupled systems such as ( 3 ) . motivated by the above , we investigate the effect of external forcing in the propagation of wave front and formation of turing patterns in the coupled mlc circuits in one and two dimensions . in this subsection we study the effect of external forcing on the propagation of wave fronts . for this purpose we consider an one dimensional array of coupled mlc circuits with initial and boundary conditions as discussed in sec . [ sec3a ] . now we perform the numerical integration by the inclusion of external periodic force of frequency @xmath173 in each cell of the array ( see eq . ( 1 ) ) . by varying the strength , @xmath178 , of the external force we study the behaviour of the propagating wave front in comparison with the force free case ( @xmath24 ) as discussed in sec . [ sec3a ] . we find that in the propagation region ( @xmath179 ) , the effect of forcing is just to introduce temporal oscillations and the propagation continues without and @xmath180 and ( b ) the partial propagation observed for the @xmath181 and @xmath180 . ] any disturbance ( see fig . [ fig6](a ) ) as in the case @xmath24 ( fig . [ fig2](a ) ) . of course this can be expected as the external force is periodic in time . however , interesting things happen in the propagation failure region discussed in sec . [ sec3a ] . in this region , beyond a certain critical strength of the external forcing , the wavefront tries to move a little distance and then stops , leading to a partial propagation . [ fig6](b ) shows such a partial propagation observed for @xmath180 and @xmath181 ( this may be compared with fig . [ fig2](b ) ) . the phenomenon can be explained by considering the propagation failure mechanism discussed in sec . [ sec3b ] in which one may look for a spatially stationary and temporally oscillating wavefront . the initial wavefront tries to settle in the nearby stationary state . however , the system will take a little time and space to settle due the effect of forcing combined with the transient behaviour of the system . thus the inclusion of external forcing in the propagation failure region can induce the wavefront to achieve a partial propagation . it is well known that the defects are inherent in very many natural pattern forming systems . in most of the pattern forming systems , the observed patterns are not ideal . for example , the patterns are not of perfect rolls or hexagons or rhombs . a commonly observed defect in such systems is the so called penta - hepta defect ( phd ) pair which is the bound state of two dislocations [ tsimring , 1996 ] . experiments on spatially extended systems often show the occurrence of phd in spontaneously developed hexagonal patterns[pantaloni & cerisier , 1983 ] . in the present case also , the existence of phd pair can be clearly seen from fig . [ fig5](d ) . in such a situation , it is important to study the effect of external periodic force in the coupled arrays of mlc circuits . now we include a periodic force with frequency @xmath182 and amplitude @xmath178 in each cell of the array and we numerically integrate eqs.([eqn3all ] ) using fourth order runge - kutta method with zero flux at boundaries . by fixing the frequency of the the external periodic force as @xmath173 and varying the amplitude ( @xmath178 ) we analyse the pattern which emerges spontaneously . interestingly for @xmath183 , the defects ( phd pair ) which are present in the absence of external force ( fig . [ fig5](d ) ) , gets removed resulting in the and ( b ) the same figure in fourier space . ] transition to a regular rhombic array . [ fig7](a ) shows the gray scale plot of the pattern observed for @xmath183 and the corresponding plot in fourier domain ( fig . [ fig7](b ) ) . thus , from the above we infer that the inclusion of external periodic force can cause a transition from hexagonal pattern to rhombic structures . we note that prez - muuzuri _ et al_. have observed , in arrays of coupled chua s circuits , similar transition from hexagons to rhombs by means of _ sidewall _ forcing [ prez - muuzuri _ et al . _ , 1995 ] . however in our case we apply forcing simultaneously to all the cells , in order to mimic situations such as application in vertically vibrated granular media[umbanhowar _ et al . _ , 1996 ] . in addition to the transition from hexagons to rhombs by the influence of external periodic force , there are also other possible effects due to it . to realize them , we consider a different set of parametric @xmath184 @xmath185 @xmath10 , @xmath186 @xmath187 @xmath56 @xmath188 @xmath189 @xmath190 @xmath191 @xmath192 @xmath193 with @xmath174 and @xmath194 . ( a ) @xmath24 , ( b ) @xmath195 , ( c ) @xmath196 and ( d ) @xmath197 . ] choice @xmath198 @xmath56 @xmath188 @xmath189 @xmath191 @xmath199 @xmath193 with @xmath200 , @xmath201 and @xmath202 . for this choice the system shows hexagonal patterns with defects including domains of small roll structures ( fig . [ fig8](a ) ) . now when the external periodic force is included a transition in the pattern from hexagonal structure to rolls starts appearing . by fixing the frequency of the external periodic force again at @xmath173 , we observed the actual transition from hexagons to rolls as we increase the forcing amplitude ( @xmath178 ) . [ fig8](b)-[fig8](d ) show the gray scale plots for @xmath195 , @xmath203 and @xmath204 , respectively . obviously the transition is due to the existence of small roll structures in the pattern for @xmath24 which nucleates the formation of rolls in the presence of forcing . in the above , we have shown that the inclusion of the external periodic force can make a transition from one stationary pattern to another stationary pattern like the transition from hexagons to rolls . besides these , are there any time varying patterns ? as mentioned above , patterns such as localized and breathing oscillations have considerable and @xmath205 @xmath184 @xmath185 @xmath10 , @xmath186 @xmath187 @xmath56 @xmath188 @xmath189 @xmath190 @xmath191 @xmath192 @xmath193 with @xmath206 and @xmath194 for various time units starting from @xmath207 . ] physical interest . in this regard , we considered the parameters @xmath208 @xmath56 @xmath188 @xmath189 @xmath209 @xmath191 @xmath199 @xmath193 with @xmath129 and @xmath202 such that a regular hexagonal pattern is observed in the _ absence _ of external periodic force . from numerical simulations , we observed that a space - time periodic oscillatory pattern ( breathing motion ) sets in for a range of low values of @xmath178 . [ fig9 ] shows the typical snapshots of the oscillating pattern at various instants for the specific choice @xmath210 . we have integrated over 10000 time units and the figure corresponds to the region @xmath211 . typically we find that the breathing pattern repeats itself approximately after a period @xmath212 in the range of our integration . one may conclude that the emergence of such breathing oscillations is due to the competition between the turing and hopf modes in the presence of external periodic force . next we move on to a study of the spatio - temporal chaotic dynamics of the array of coupled mlc circuits when individual cells are driven by external periodic force . the motivation is that over a large domain of ( @xmath178 , @xmath182 ) values the individual mlc circuits typically exhibits various bifurcations and transition to chaotic motion ( see appendix a ) . so one would like to know how the coupled array behaves collectively in such a situation , for fixed values of the parameters . for this purpose , we set the parameters at @xmath213 @xmath214 . in our numerical simulations , we have mainly considered the one dimensional array specified by eq . ( 1 ) and assumed periodic boundary conditions . the choice of periodic boundary conditions here makes the calculation of lyapunov exponents easier so as to understand the spatio - temporal chaotic dynamics of ( 1 ) in a better way . numerical simulations were performed by considering random initial conditions using fourth order runge - kutta method for seven choices of @xmath178 values . the coupling coefficient in eq . ( 1 ) was chosen as @xmath150 . out of these , the first three lead to period-@xmath215 , period-@xmath216 , period-@xmath217 oscillations , respectively and the remaining choices oscillators for the external periodic forcing strength ( a ) @xmath210 , ( b ) @xmath218 , ( c ) @xmath219 , ( d ) @xmath220 , ( e ) @xmath221 , ( f ) @xmath222 and ( g ) @xmath195 ] correspond to chaotic dynamics of the single mlc circuit . [ fig10](a)-[fig10](g ) show the space - time plots for @xmath210 , @xmath218 , @xmath219 , @xmath220 , @xmath221 , @xmath222 and @xmath195 , respectively . from the figs . [ fig10](a)-[fig10](c ) , it can be observed that for @xmath210 , @xmath223 and @xmath224 the mlc array also exhibits regular periodic behaviour with periods @xmath215 , @xmath216 and @xmath217 respectively in time alone as in the case of the single mlc circuit . however , for @xmath220 and @xmath225(figs . [ fig10](d ) and [ fig10](e ) ) one obtains _ space - time periodic _ oscillations eventhough each of the individual uncoupled mlc circuits for the same parameters exhibits chaotic dynamics . we may say that a kind of controlling of chaos occurs due to the coupling , though the coupling strength is large here . from the above analysis it can be seen that the macroscopic system shows regular behaviour in spite of the fact that the microscopic subsystems oscillate chaotically . finally for @xmath195 the coupled system shows spatio - temporal chaotic dynamics ( fig . [ fig10](g ) ) and this was confirmed by calculating the lyapunov exponents using the algorithm given by wolf _ et al_. [ wolf _ et al . _ , 1985 ] . for example , we calculated the lyapunov exponents for @xmath226 coupled oscillators ( for @xmath16 , it takes enormous computer time relatively which we could not afford at present ) and we find the highest three exponents have the values @xmath227 , @xmath228 , @xmath229 and the rest are negative ( see next subsection for further analysis ) . we also note that for @xmath222 , the system is in a transitional state as seen in fig . [ fig10](f ) . since the above study of spatio - temporal chaos involves a large number of coupled chaotic oscillators it is of great interest to analyse the size dependence of the dynamics of these systems . to start with , we consider the case of 10 coupled oscillators with periodic boundary conditions and numerically solve the system using fourth order runge - kutta method with the other parameters chosen as in sec . [ sec5a ] . the value of @xmath178 is chosen in the range @xmath230 . we find that this set up shows different behaviour as compared to the 100 cell case . actually the system gets synchronized to a chaotic orbit rather than showing periodic behaviour as in the case of 100 cells described above . first we analyse the dynamics for @xmath220 by slowly increasing the system size from @xmath231 . we noted that for the system size , @xmath232 it gets synchronized to a chaotic orbit as mentioned above . we have also calculated the corresponding lyapunov spectrum and found that there is one positive maximal lyapunov exponent as long as @xmath232 . for example , for @xmath233 , @xmath234 , with the rest cell for the synchronized state @xmath235 and ( b ) the periodic orbit in the @xmath236 cell for the controlled state @xmath237 ] of the exponents being negative . fig . [ fig11](a ) shows the dynamics of the @xmath236 cell in the array and [ fig12](a ) depicts the space - time plot of the synchronized spatio - temporal chaos . the system shows entirely different behaviour when we increase the system size to @xmath238 . as noted in the previous subsection , [ sec5a ] , there occurs kind of suppression of spatio - temporal chaos . [ fig11](b ) shows the resultant periodic orbit in the @xmath236 cell of the array and fig . [ fig12](b ) shows the space - time plot of the spatio - temporal periodic behaviour of the array . the maximal lyapunov exponent is found to be negative in this case @xmath239 . we observed the same phenomenon for @xmath221 also . and ( b ) @xmath238 . ] next we consider the case of @xmath195 in which the coupled system in the previous subsection showed spatio - temporal chaos . from numerical simulations , we again observed a synchronized motion for @xmath240 and the corresponding lyapunov spectrum shows one positive exponent only with all other exponents being negative . but for @xmath241 , the coupled system shows spatio - temporal chaos . the lyapunov spectrum in this case ( for @xmath241 ) possesses multiple positive exponents . for example , for @xmath238 , @xmath242 , @xmath243 , @xmath244 . from the above analysis we see definite evidence that the spatio - temporal chaotic dynamics in coupled mlc circuits depends very much on the size of the system . a similar analysis can be performed to investigate the nature of the spatio - temporal dynamics on the strength of the coupling coefficient @xmath7 . preliminary analysis shows similar phenomenon at critical values of @xmath7 . further work is in progress on these lines . in this paper we have discussed various spatiotemporal behaviours of the coupled array of murali - lakshmanan - chua circuits in one and two dimensions . in the first part we have reported the propagation phenomenon of the travelling wave fronts and failure mechanism in the absence of external forcing . we have shown that the propagation phenomenon is due to the loss of stability of the steady states via subcritical bifurcation coupled with the presence of necessary basin of attraction of the steady states . the effect of weak coupling on the travelling wave phenomenon has also been studied . in the two dimensional array we studied the onset of turing instability leading to the spontaneous formation of hexagonal patterns . introduction of external periodic force has been shown to lead to a partial propagation in the failure region of one dimensional array , while in the two dimensional array a transition from hexagons to rhombs take place . we have also showed that a transition from hexagons to rolls can occur by the influence of external periodic force provided domains of small roll structures already present in the absence of external force . space - time breathing oscillations as a result of the competition between hopf and turing modes in the presence of external periodic force have also been observed . we have also studied the spatio - temporal chaotic dynamics and showed that the mlc array exhibits both periodic and chaotic dynamics as in the case of single oscillator . in the chaotic regime the dynamics is also critically dependent on the system size . for a large array , the collective behaviour shows periodic oscillations both in space and time eventhough the individual subsystems oscillate chaotically as the coupling brings the macroscopic system into a regular motion . on the other hand , below a critical size of the system , it gets synchronized to a chaotic orbit . the study of array of diffusively coupled driven nonlinear oscillators is of intrinsic interest as they represent many natural phenomena such as faraday instability , patterns in granular media and so on . of particular interest will be to look for localized structures in these systems . whether such excitations exists in the present arrays of oscillators is a question to be answered in the near future . work is in progress along these lines . p.m. acknowledges with gratitude the financial support of council of scientific and industrial research in the form of a senior research fellowship . the work of m.l . has been supported by the department of science and technology , government of india and national board for higher mathematics ( department of atomic energy ) in the form of research projects . the murali - lakshmanan - chua circuit , fig . [ fig13](a ) , is the simplest second order dissipative nonautonomous circuit , consisting of chua s diode as the only nonlinear element[murali _ et al_. , 1994a ] . this circuit contains a capacitor @xmath245 , an inductor @xmath246 , a linear resistor @xmath247 , an external periodic forcing @xmath248 and a chua s diode . by applying the kirchoff s laws to this circuit , the governing equations for the voltage @xmath249 across the capacitor @xmath250 and the current @xmath251 through the inductor @xmath252 are represented by the following set of two first order nonautonomous differential equations : the piecewise nature of the characteristic curve of chua s diode is obvious from eq . ( [ app : eq2])[murali _ et al_. , 1994a ] . the slopes of left , middle and right segments of the characteristic curve are @xmath257 , @xmath258 and @xmath259 , respectively . @xmath260 and @xmath261 are the break points and @xmath262 corresponds to the dc offset in the chua s diode . rescaling eq . ( [ app : eq1 ] ) as @xmath263 , @xmath264 , @xmath265 , @xmath266 @xmath267 and @xmath268 and then redefining @xmath269 as @xmath270 the following set of normalized equations are obtained : @xmath271 with @xmath272 where @xmath273 , @xmath274 and @xmath275 . obviously @xmath3 takes the form as in eq . ( [ eqn2 ] ) with @xmath276 , @xmath277 and @xmath278 . the dynamics of eq . ( [ app : eq3 ] ) depends on the parameters @xmath279 , @xmath280 , @xmath10 , @xmath11 , @xmath12 , @xmath281 , @xmath182 and @xmath178 . the rescaled parameters corresponding to the experimental observations reported in murali _ et al_.[1994a ] correspond to @xmath282 , @xmath283 , @xmath284 , @xmath285 , @xmath286 , @xmath50 and @xmath173 . by varying @xmath178 one can observe the familiar period doubling bifurcations leading to chaos and several periodic windows in the mlc circuit . [ fig14](a ) shows the one parameter bifurcation diagram in the @xmath178-@xmath0 plane [ @xmath292 . a summary of the bifurcations that occur in this case for different @xmath178 values is given in table [ table1 ] . further it is of great interest to consider the parametric choice @xmath293 @xmath294 which corresponds to the function @xmath3 having the form as shown in fig . [ fig1](c ) . this choice of parameters provides the possibility of bistability nature in the asymmetric case in the absence of periodic forcing . in this case one can easily observe from numerical simulations that the mlc circuit admits only limit cycles for @xmath295 . the bifurcation diagram in the @xmath178-@xmath0 plane is depicted in fig . [ fig14](b ) . the eigenvalues of the above matrix ( [ jacob1 ] ) as a function of @xmath7 can be obtained by solving the corresponding characteristic equation . we have verified by numerical evaluation that all the eigenvalues of ( [ jacob1 ] ) have negative real parts for @xmath296 . similarly all the eigenvalues corresponding to the steady state @xmath83 have negative real parts irrespective of the values of @xmath7 @xmath297 . we have also verified that atleast one of the eigenvalues corresponding to the state @xmath81 has positive real part for @xmath296 . thus @xmath88 are stable , while @xmath81 is unstable for all values of @xmath296 . similarly one can obtain the stability properties of the set of steady states @xmath298 , by substituting the appropriate expressions from ( [ eqnb2 ] ) into ( [ jacob ] ) . numerical diagonalization of ( [ jacob ] ) for these steady states show that @xmath97 is unstable for all values of @xmath7 , while @xmath96 and @xmath98 are stable for @xmath299 and unstable for @xmath102 for the chosen initial conditions . the numerical values of @xmath123 in the stable region are given in table [ table2 ] as a function of @xmath7 . 41 allesie , m. a. , bonke , f. i. m. & scopman , t. y. g. [ 1973 ] circus movement in rabit atrial muscle as a mechanism in tachycardia , " _ circ . * 33 * , 54 - 62 . bak , p. , & chen , k. [ 1991 ] self - organized criticality " _ sci . _ * 264 * , 46 . balke , c. w. , lesh , m. d. , spear , j. f. , kadish , a. , levine , j. h. & moore , e. n. [ 1988 ] effects of cellular uncoupling on conduction in anisotropic canine ventricular myocardium , " _ circ . res . _ * 63 * , 879 - 892 . chua , l. o. and yang , l. [ 1988 ] cellular neural networks : theory , " _ ieee trans . * 35 * , 1257 - 1272 . chua , l. o. ( ed . ) [ 1995 ] special issue on nonlinear waves , patterns and spatio - temporal chaos in dynamic arrays , " _ ieee trans . syst . _ * 42 * , 557 - 823 . cole , w. c. , picolne , j. b. & sperelakis , n. [ 1988 ] gap junction uncoupling and discontinuous propagation in the heart , " _ biophys . j. _ * 53 * , 809 - 818 . dolnik , m. & marek , m. [ 1988 ] extincion of oscillations in forced and coupled reaction cells , " _ j. phys . _ * 92 * , 2452 - 2455 . feingold , m. , gonzlez , d. l. , piro , o. & viturro , h. [ 1988 ] phase locking , period doubling and chaotic phenomena in externally driven excitable systems , " _ phys . a _ * 37 * , 4060 - 4063 . fineberg , j [ 1996 ] physics of jumping sandbox , " _ nature _ * 382 * , 763 - 764 guckenheimer , j. and holmes , p. [ 1983 ] _ nonlinear oscillations , dynamical systems and bifurcation of vector fields _ ( springer verlag , new york ) hao bai - lin [ 1984 ] _ chaos _ ( world scientific , singapore ) keener , j. p. [ 1987 ] propagation and its failure in coupled systems of excitable cells " _ siam j. appl . math _ * 47 * , 556 kocarev , l. & parlitz , u. [ 1996 ] synchronizing spatiotemporal chaos in coupled nonlinear oscillators , " _ phys . lett . _ * 77 * , 2206 - 2209 . kudrolli , a. , wolpert , m & gollub , j.p . , [ 1997 ] cluster formation due to collisions in granular material , " _ phys . lett . _ * 78 * , 1383 - 1386 . lakshmanan , m. & murali , k. [ 1995 ] experimental chaos from non - autonomous electronic circuits , " _ phil . a _ * 353 * , 33 - 46 . lakshmanan , m. & murali , k. [ 1996 ] _ chaos in nonlinear oscillators : controlling and synchronization _ ( world scientific , singapore ) . laplante , j. p. & erneux , t. [ 1992 ] propagation failure and multiple steady states in an array of diffusion coupled flow reactors , " _ physica a _ * 188 * , 89 - 98 . leenaerts , d. m . w. [ 1997 ] wave propagation and its failure in piecewise linear nagumo equation " proc . lioubasheski , o. , arbell , h. & fineberg , j. [ 1996 ] dissipative solitary states in driven surface waves , " _ phys lett . _ * 76 * , 3959 - 3962 . marn , j. l. & aubry , s. [ 1996 ] breathers in nonlinear lattices : numerical calculation from the anticontinuous limit , " _ nonlinearity _ * 9 * , 1501 - 1528 muuzuri , a. p. , prez - muuzuri , v. , prez - villar , v. & chua , l. o. [ 1993 ] spiral wave in 2d array of nonlinear circuits , " _ ieee transactions circuits and systems _ * 40 * , 872 - 877 . muuzuri , a. p. , prez - muuzuri , v. , gmez - gesteria , m. , chua , l. o. & prez - villar , v. [ 1995 ] spatiotemporal structures in discretely - coupled arrays of nonlinear circuits : a review , " _ int . j. bifurcation and chaos _ * 5 * , 17 - 50 . murali , k. , lakshmanan , m. & chua , l. o. [ 1994a ] the simplest dissipative non - autonomous chaotic circuit , " _ ieee trans . sys . _ * i41 * , 462 - 463 murali , k. , lakshmanan , m. & chua , l. o. [ 1994b ] bifurcation and chaos it the simplest dissipative non - autonomous chaotic circuit , " _ int . j. bifurcation and chaos _ * 4 * , 1511 - 1524 murali , k. , lakshmanan , m. & chua , l. o. [ 1995 ] controlling and synchronization of chaos in the simplest dissipative nonautonomous circuit " _ int . j. bifurcation and chaos _ * 5 * , 6498 - 6510 murray , j. d. [ 1989 ] _ mathematical biology _ ( springer - verlag , newyork ) pantaloni , j. & cerisier , p. [ 1983 ] structure defects in bnaed - marangoni convection , " _ cellular structures and instabilities , leccture notes in physics _ * 210 * , eds . j. e. & zaleski , s. ( springer - verlag , berlin ) prez - muuzuri , v. , gmez - gesteria , m. , prez - villar , v. & chua , l. o. [ 1993 ] travelling wave propagation in one dimensional fluctuating medium , " _ int . j. bifurcation and chaos _ * 3 * , 211 - 215 . prez - muuzuri , v. , alonso , m. , chua , l. o. & prez - villar , v. [ 1994 ] resonance patterns in 1d arrays of coupled nonlinear excitable systems , " _ int . j. bifurcation and chaos _ * 4 * , 1631 - 1638 . prez - muuzuri v. , gmez - gesteria , m. , muuzuri , a. p. , chua , l. o. & prez - villar [ 1995 ] sidewall forcing of hexagonal turing patterns : rhombic patterns , " _ physica d _ * 82 * , 195 - 204 perraud , j. j. , de wit , a. , dulos , e. , de kepper , p. , dewel , g. & borkmans , p. [ 1993 ] one - dimensional spirals : a novel asynchronous chemical wave source " , _ phys . * 71 * , 1272 - 1275 pismen , l. [ 1987 ] bifurcations of quasi - periodic and nonstationary pattern under external forcing " , _ phys . lett . _ * 59 * , 2740 scott , a. c. [ 1975 ] the electrophysics of nerve fiber , " _ rev . _ * 47 * , 487 - 533 . sobrino , t. , alonso , m. , prez - muuzuri , v. & prez - villar , v. [ 1993 ] reaction - diffusion process in a one - dimensional array of active nonlinear circuits , " _ eur . j. phys . _ * 14 * , 74 - 79 . tsimring , l. s. [ 1996 ] dynamics of penta - hepta defects in hexagonal patterns " , _ physica _ * d89 * , 368 - 380 . turing , a. m. [ 1952 ] the chemical basis of morphogenesis , " _ phil . . london _ * b237 * , 37 - 72 . umbanhowar , p. b. , melo , f. & swinney , h. l. [ 1996 ] localized excitations in a vertically vibrated granular layers " _ nature ( london ) _ * 382 * , 793 - 796 walgraef , d. [ 1988 ] external forcing of spatio - temporal pattern " _ europhys . lett . _ * 7 * , 485 walgraef , d. [ 1996 ] _ spatio - temporal pattern formation _ ( springer , new york ) pp . 127 - 167 watanabe , s. , strogatz , s. h. , van der zant , h. s. j. & orlando , t. p. [ 1995 ] whirling modes and parametric instablilities in the discrete sine - gordon equation : experimental tests in josephson rings , " _ phys . * 74 * , 379 - 382 . wolf , a. , swift , j. b , swinney , h. l. , and vastano , j. a. [ 1985 ] determining lyapunov exponents from a time series " _ physica _ * 16d * 285 - 317 .

based on a careful analysis of functional models for contractive multi - analytic operators we establish a one - to - one correspondence between unitary equivalence classes of minimal contractive liftings of a row contraction and injective symbols of contractive multi - analytic operators . this allows an effective construction and classification of all such liftings with given defects . popescu s theory of characteristic functions of completely non - coisometric row contractions is obtained as a special case satisfying a szeg condition . in another special case of single contractions and defects equal to all non - zero schur functions on the unit disk appear in the classification . it is also shown that the process of constructing liftings iteratively reflects itself in a factorization of the corresponding symbols .

1402.4220

functional models from analytic functions were developed by sz.nagy-foias @xcite and used for classifying contractive operators on hilbert spaces . a similar approach was used by popescu in @xcite for classifying row contractions by certain multi - analytic operators . these classifying objects were called characteristic functions . in @xcite and more explicitly in @xcite characteristic functions of liftings of row contractions were introduced and it was shown that they are complete invariants for unitary equivalence in a certain class of liftings . here we present an approach which is based on a systematic use of associated functional models which on the one hand exhibits popescu s characteristic functions as special cases of characteristic functions of liftings and which on the other hand fully discloses the additional potentials of our generalization . + let us immediately introduce the two main players . \(1 ) let @xmath1 , the _ full fock space _ over @xmath2 , and let @xmath3 be the standard basis of @xmath4 . denote the left creation operator w.r.t . @xmath5 on @xmath6 by @xmath7 , that is @xmath8 for @xmath9 . let @xmath10 and @xmath11 be hilbert spaces . a linear operator @xmath12 is called _ multi - analytic _ ( cf . @xcite ) if it intertwines @xmath13 for all @xmath14 . a multi - analytic operator @xmath15 is determined by its symbol @xmath16 defined by @xmath17 for all @xmath18 , here @xmath19 denotes the standard basis vector of @xmath20 . we call @xmath15 _ contractive _ if @xmath21 . note that for @xmath22 and @xmath23 a contractive multi - analytic operator corresponds exactly to multiplication with a function in the schur class , i.e. , a bounded analytic function in the open unit disk with supremum norm at most @xmath0 . \(2 ) a @xmath24-tuple @xmath25 of operators on a hilbert space @xmath26 is called a _ row contraction _ if it is a contraction from @xmath27 to @xmath26 or , equivalently , if @xmath28 . if a @xmath24-tuple @xmath29 on a hilbert space @xmath30 can be written in the form @xmath31 for suitable @xmath24-tuples @xmath32 and @xmath33 then @xmath34 is called a _ lifting _ of @xmath35 . the lifting is called _ contractive _ if @xmath34 is still a row contraction and it is called _ minimal _ if @xmath36 is the smallest @xmath34-invariant subspace containing @xmath26 . ( by @xmath34-invariance we mean invariance for all @xmath37 . ) we remark that it presents no particular difficulties to include sequences of operators ( @xmath38 ) but we write all formulas as if @xmath24 is finite . to establish a correspondence between the two main players we start with a detailed discussion of functional models . we use the generalized setting introduced by popescu in @xcite to study row contractions . the case @xmath22 ( single contractions ) is of course of special interest and our results are new also for @xmath22 but these results work just as well for general @xmath24 . this observation is important because from the case @xmath39 there are promising applications to the dynamics of open quantum systems , see @xcite for an introduction to this topic and further references along these lines . the impact of our results on these applications will be worked out elsewhere . in section 2 we prove properties of functional models to be used later . the results about functional models which are new depend on observations about the geometry of an invariant subspace for a row isometry which we derive from a geometric lemma proved in an appendix to this paper . they establish relations between the positions of certain subspaces on the one hand and properties of the symbol @xmath40 of the multi - analytic operator on the other hand . section 3 is the core of the paper . based on the results about the functional model and extending ideas from @xcite , we work out a mapping @xmath41 which , for any given row contraction @xmath35 , maps contractive multi - analytic operators @xmath42 , where @xmath43 is the defect space of @xmath35 , to contractive liftings @xmath34 of @xmath35 . in the converse direction we make use of the theory of the minimal isometric dilation for the contractive lifting @xmath34 to construct a mapping @xmath44 from contractive liftings @xmath34 of a given row contraction @xmath35 to contractive multi - analytic operators @xmath42 . suitably restricted , the maps @xmath41 and @xmath44 become inverses of each other and we obtain a one - to - one correspondence between unitary equivalence classes of minimal contractive liftings and equivalence classes of injective symbols . this justifies to call the corresponding multi - analytic operators characteristic functions of liftings , as has been done already in @xcite . with the theory developed here we provide a complete answer to the open problem posed at the end of section 3 in @xcite , namely to classify the multi - analytic operators which can occur as characteristic functions of liftings . the surprisingly simple answer , all that is needed is the easily checked property of injectivity of the symbol , shows that the connection with liftings is a very natural application of multi - analytic operators and makes it now much easier to develop the applications . we can always study minimal contractive liftings via the corresponding symbols . as an example of such an application we study the factorization of liftings with the help of the corresponding factorization of the characteristic functions ( compare @xcite , @xcite ) . in section 4 we revisit popescu s work in @xcite where he defined a characteristic function for a completely non - coisometric row contraction as a certain multi - analytic operator . we show that this can be considered as a special case of our theory in the sense that popescu s characteristic functions appear as characteristic functions of a special type of liftings and that the property of being a complete invariant for unitary equivalence follows from the corresponding result about liftings in section 3 . in section 5 we look at a class of examples : minimal contractive liftings @xmath45 of a single contraction @xmath46 such that both @xmath46 and @xmath45 have defect equal to @xmath0 . by our theory the unitary equivalence classes of these liftings are in one - to - one correspondence with non - zero schur functions on the unit disk ( up to unimodular complex factors ) . this gives us an opportunity to illustrate many of the previous results by easily computable examples . already in this case in the future there is a lot more to find out about these liftings by making a more systematic use of what is known about schur functions . for example , operator - valued schur functions have been used for commutant lifting , see @xcite and more recently @xcite . related work in different directions is done in @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite and @xcite . but the explicit parametrization of all minimal contractive liftings achieved here is new and provides an excellent basis for studying applications , for example the dynamics of open quantum systems mentioned earlier . in fact in the special case of the minimal isometric dilation it is long known and well studied how it describes the embedding of open into closed quantum systems , see for example @xcite . let us finish this introduction with a short reminder of the well known theory of the _ minimal isometric dilation _ @xmath47 of a row contraction @xmath48 which appears in several places in this paper . it was first presented in @xcite , with small modifications we use the notation from @xcite . in both papers a lot of additional details can be found . recall that a @xmath24-tuple @xmath49 on a hilbert space @xmath50 is called a _ row isometry _ if @xmath51 or , equivalently , the @xmath52 are isometries with orthogonal ranges . for any row contraction @xmath48 on a hilbert space @xmath53 there exists a row isometry @xmath54 on a bigger hilbert space such that @xmath55 for all @xmath56 . here we use , as in similar cases , the notation @xmath57 for the orthogonal projection onto @xmath58 and the notation @xmath59 for any word @xmath60 with letters @xmath61 to stand for the operator @xmath62 . we call @xmath63 the length of the word . for all @xmath56 means here and later : for all such words of all possible lengths including the empty word @xmath64 of length @xmath65 which corresponds to the identity operator . a row isometry @xmath66 with this property is called an isometric dilation of @xmath67 . it is easily seen that this is actually a very specific example of a contractive lifting in the sense introduced earlier . if we require minimality for an isometric dilation in the sense that the bigger hilbert space is the smallest closed @xmath66-invariant space containing @xmath53 then this determines @xmath66 up to unitary equivalence and we denote it by @xmath47 , on the hilbert space @xmath68 . there is a construction of the minimal isometric dilation analogous to the schffer construction for a single contraction : recall that the operator @xmath69 given by @xmath70 is called the _ defect operator _ , @xmath71 is called the _ defect space _ and the ( hilbert space ) dimension of @xmath72 is called the _ defect _ of @xmath67 . the minimal isometric dilation @xmath47 is then realized in a canonical way on the hilbert space @xmath73 . again @xmath1 is the full fock space over @xmath2 and we note that @xmath47 restricted to @xmath74 is nothing but @xmath75 , with @xmath76 , the @xmath24-tuple of creation operators on @xmath6 . we refer to the row contraction @xmath75 as the _ canonical row shift _ and to @xmath77 as its _ generating wandering subspace_. ( as usual , the subspace is called wandering for @xmath75 because its translates @xmath78 over all words @xmath56 , including the empty word @xmath64 , are orthogonal to each other . ) the minimal isometric dilation @xmath47 is obtained from any isometric dilation @xmath66 by restriction . the unitary equivalence to the canonical construction above is expressed by a canonical unitary from the defect space @xmath72 onto the @xmath66-wandering subspace @xmath79 given by @xmath80 . it can be extended to a canonical unitary from @xmath81 onto @xmath82 ( over all words @xmath56 , including the empty word @xmath64 ) intertwining the two versions of the minimal isometric dilation . compare @xcite and @xcite . as already defined in the introduction , let @xmath6 be the full fock space over @xmath2 and @xmath3 be the standard basis of @xmath4 . denote the left creation operator w.r.t . @xmath5 on @xmath6 by @xmath7 . let @xmath12 be a contractive multi - analytic operator where @xmath10 and @xmath11 are hilbert spaces . we now define the model space @xmath83 where @xmath84 and an isometric map @xmath85 by @xmath86 since @xmath87 is isometric , its range @xmath88 is closed . note that by this construction we can think of @xmath15 as the orthogonal projection onto @xmath89 restricted to @xmath90 . further define the @xmath24-tuple @xmath49 on the hilbert space @xmath91 by @xmath92 it is easy to see that @xmath66 is a row isometry on @xmath93 . observe that @xmath94 for @xmath95 . thus @xmath90 is an invariant subspace for @xmath66 . define @xmath96 for @xmath97 and denote the row isometric @xmath24-tuple @xmath98 on @xmath90 by @xmath99 . because for @xmath100 @xmath101 for @xmath102 we conclude that @xmath103 is a unitary operator intertwining between the canonical row shift on @xmath104 and @xmath99 . restriction to the generating wandering subspaces yields a unitary operator @xmath105 further we define @xmath106 the use of the subscript @xmath45 will become clear in the context of the next section . the following geometric lemma helps to clarify the positions of these spaces . [ lem : geo ] * @xmath107 . * @xmath108 . because this follows from a general argument about adjoints we have postponed this argument to an appendix . in lemma [ inv ] we choose @xmath109 to be the row isometry @xmath110 and we choose @xmath111 and @xmath112 , so that @xmath113 because @xmath90 is an invariant subspace . then @xmath114 and we get ( a ) from lemma [ inv](ii ) and ( b ) from lemma [ inv](iii ) . now from lemma [ lem : geo](b ) we have @xmath115 . further @xmath116 which implies @xmath117 , the last inclusion is lemma [ lem : geo](a ) . hence also @xmath118 and we conclude that @xmath119 in the following we determine how equalities in these inclusions depend on properties of the symbol @xmath40 . recall that the symbol @xmath40 of @xmath15 is defined by @xmath120 for @xmath18 . we also write @xmath121 as a shorthand for the formally correct @xmath122 . since @xmath123 , we have @xmath124 [ lem : fm ] for @xmath125 for some @xmath126 we have * @xmath127 . * @xmath128 . \(a ) from lemma [ lem : geo](b ) it follows that for @xmath129 we have @xmath130 if and only if @xmath131 . this is the case if and only if for all @xmath132 and all @xmath14 @xmath133 but @xmath134 for all @xmath135 and all @xmath14 if and only if @xmath136 , and hence ( a ) follows . \(b ) @xmath137 if and only if @xmath130 and @xmath138 because for @xmath139 @xmath140 we infer that @xmath141 if and only if @xmath142 from this and part ( a ) we obtain @xmath143 . [ non - constant ] we say that a symbol @xmath16 has no constant directions if @xmath144 only for @xmath145 . the following result relating properties of certain subspaces in the functional model to properties of the symbol is crucial for our applications to characteristic functions later . [ fm ] suppose @xmath12 is a contractive multi - analytic operator where @xmath10 and @xmath11 are hilbert spaces . then the following statements hold : * @xmath146 $ ] for all @xmath18 . * @xmath147 is injective . * @xmath148 has no constant directions . \(a ) is immediate from lemma [ lem : fm ] . \(b ) assume that @xmath149 . let @xmath18 such that @xmath150 . then @xmath151 but from lemma [ lem : fm](b ) also @xmath152 so @xmath153 and this implies @xmath154 because @xmath155 is unitary . thus @xmath40 is an injective map . conversely , assume that @xmath40 is injective . let @xmath156 . then there exist @xmath18 such that @xmath157 and @xmath137 . from lemma [ lem : fm](b ) we infer that @xmath158 . since @xmath159 is injective , we obtain @xmath160 and @xmath161 . thus @xmath162 . finally ( c ) follows by combining ( a ) and ( b ) . note that having no constant directions clearly implies injectivity . the traditional motivation for functional models is the study of the compression @xmath33 of @xmath66 to @xmath163 , that is @xmath164 with @xmath165 for @xmath166 . in the following propositions we state some properties which will be used later . note that by construction @xmath66 is an isometric dilation of @xmath33 and the following criterion for minimality is our first application of theorem [ fm ] . [ min - dil ] @xmath66 is a minimal isometric dilation of @xmath33 if and only if @xmath167 which happens if and only if @xmath40 has no constant directions . restriction of @xmath66 to the smallest closed @xmath66-invariant subspace containing @xmath163 , with the wandering subspace @xmath168 , provides a minimal isometric dilation for @xmath33 . hence @xmath66 itself is a minimal isometric dilation of @xmath33 if and only if @xmath167 . the second half is a restatement of theorem [ fm](c ) . recall that @xmath33 is called _ completely non - coisometric _ if @xmath169 satisfies @xmath170 for all @xmath171 only if @xmath172 . [ model ] the following statements hold : * @xmath173 . * @xmath33 is completely non - coisometric . if @xmath174 is orthogonal to @xmath175 then it can be written in the form @xmath176 with @xmath177 . if further @xmath169 then it is also orthogonal to @xmath178 for all @xmath179 . with @xmath180 we find @xmath181 and hence @xmath172 . this proves ( a ) . if @xmath169 satisfies @xmath170 for all @xmath171 then @xmath182 must be orthogonal to @xmath175 because @xmath183 and @xmath66 acts on @xmath175 as a row shift . hence ( b ) follows from ( a ) . note that ( b ) has already been observed in @xcite , in the proof of theorem 5.1 there . we discuss the characteristic function of @xmath33 in section 4 and then compare our concept of having no constant directions with the traditional concept of a purely contractive function . in this section we state and prove the main result of this paper , theorem [ char ] , which gives a complete classification of minimal contractive liftings by multi - analytic operators ( or their symbols ) . the correspondence is constructive and can be used to analyze the structure of the set of minimal contractive liftings . to be able to state this result in an optimal way as a one - to - one correspondence we need to introduce suitable equivalence classes . recall that liftings and the concepts of contractivity and minimality have already been defined in the introduction . we also use the notation which is given there . [ def : equivalent ] let @xmath34 and @xmath184 be liftings of a row contraction @xmath35 . if there exists a unitary @xmath185 which intertwines @xmath34 and @xmath184 and restricts to the identity on the subspace @xmath26 corresponding to @xmath35 then we say that the liftings @xmath34 and @xmath184 are unitarily equivalent . if @xmath186 and @xmath187 are multi - analytic operators with symbols @xmath188 and @xmath189 ( with the same @xmath11 ) and there exists a unitary @xmath190 such that @xmath191 then we say that @xmath186 and @xmath187 ( also @xmath40 and @xmath192 ) are equivalent . note that if @xmath193 and @xmath194 are multi - analytic operators then even if @xmath195 it only makes sense to say that they are equivalent after fixing an identification between the spaces @xmath196 and @xmath197 , for example by specifying unitaries @xmath198 and @xmath199 to identify both @xmath196 and @xmath197 with a space @xmath11 . in this sense the concept of equivalence is slightly different from the concept of coincidence used in @xcite . this is further discussed in section 4 . we are now able to state the main result . [ char ] let @xmath35 be a row contraction on a hilbert space @xmath26 . then there is a one - to - one correspondence between unitary equivalence classes of minimal contractive liftings @xmath34 of @xmath35 and equivalence classes of injective symbols @xmath200 . + the dimension of @xmath10 is equal to the defect of @xmath34 . the correspondence is given explicitly by two mappings @xmath41 from contractive multi - analytic operators to contractive liftings and @xmath44 in the opposite direction which are constructed below . by this one - to - one correspondence it is justified , in the case of minimal contractive liftings @xmath34 , to call the contractive multi - analytic operator @xmath201 the _ characteristic function _ of the lifting @xmath34 . this terminology has been introduced in @xcite . with the theory developed here we have completely solved the open problem formulated at the end of section 3 in @xcite , namely to classify the symbols which can occur as characteristic functions of minimal contractive liftings : exactly the injective symbols can and do occur . we give a precise statement as follows . [ char2 ] the characteristic function of a minimal contractive lifting has always an injective symbol . conversely , if @xmath186 is any contractive multi - analytic operator with an injective symbol then it is the characteristic function of the minimal contractive lifting @xmath202 . let us also note here the following immediate consequence of theorem [ char ] which shows a surprising correspondence between liftings of different row contractions . [ c change ] let @xmath203 and @xmath204 be row contractions with the same defect @xmath205 . then there is a one - to - one correspondence between their sets of unitary equivalence classes of minimal contractive liftings . this is obvious because both sets are in one - to - one correspondence with the classes of injective symbols described in theorem [ char ] . note that to fix a specific correspondence you have to fix an identification between @xmath43 and @xmath206 , compare the remarks after definition [ def : equivalent ] . we now construct the mappings @xmath41 and @xmath44 and determine their properties in a sequence of propositions and lemmas , more and more closing in on a proof of theorem [ char ] . let @xmath203 be any row contraction on a hilbert space @xmath26 , fixed once and for all . we first define the mapping @xmath41 from contractive multi - analytic operators @xmath207 to contractive liftings @xmath34 of @xmath35 . we introduce the hilbert space @xmath208 where the notation comes from the functional model construction in section 2 , with @xmath209 . note that on the subspace @xmath93 we have the row isometry @xmath66 introduced in section 2 . it restricts to the canonical row shift on @xmath210 which is a reducing subspace for @xmath66 . now we also have the minimal isometric dilation @xmath211 on the subspace @xmath212 , in the canonical version stated in the final paragraph of the introduction , which also restricts to the canonical row shift on @xmath210 . hence there exists a row isometry on @xmath213 , which we call @xmath214 , which restricts to @xmath66 on @xmath93 and restricts to @xmath211 on @xmath212 . if we examine the decomposition @xmath215 with @xmath216 then we obtain on @xmath36 a contractive lifting of @xmath35 as follows : @xmath217 we now define the mapping @xmath41 by setting @xmath218 equal to the contractive lifting @xmath34 of @xmath35 defined in this way . we also write @xmath219 or @xmath220 if we want to include the dependence on the original row contraction @xmath35 in the notation . note further that @xmath214 is an isometric dilation of @xmath34 and we also have @xmath221 for all words @xmath56 . the notation @xmath222 introduced in section 2 is consistent with the notation for isometric dilations of @xmath34 in the sense that @xmath223 hence we find a version of the minimal isometric dilation @xmath224 of @xmath34 by restricting @xmath214 to @xmath225 . as usual we identify this space ( by a canonical unitary ) with @xmath226 . now from theorem [ fm](b ) we immediately conclude that under a rather weak assumption , injectivity of the symbol @xmath40 , our construction of @xmath214 already provides us with a minimal isometric dilation of the lifting @xmath34 : [ prop : injective ] for the lifting @xmath227 of @xmath35 @xmath228 this means that for an injective symbol @xmath40 the minimal isometric dilation of the lifting @xmath229 can be built in a straightforward way from the minimal isometric dilation of @xmath35 together with the functional model for @xmath40 . + now we construct the mapping @xmath44 in the opposite direction . we still have the row contraction @xmath203 on a hilbert space @xmath26 , fixed once and for all . suppose that we are further given a contractive lifting @xmath230 on a hilbert space @xmath216 . we can now construct the minimal isometric dilation @xmath224 on the hilbert space @xmath226 . instead of @xmath231 we can also write @xmath222 . clearly @xmath224 is an isometric dilation of @xmath35 on @xmath26 and hence we can find a version of the minimal isometric dilation @xmath211 by restricting @xmath224 to @xmath232 , where @xmath233 is a subspace with dimension equal to the defect of @xmath35 and hence can be identified canonically with @xmath43 . both @xmath222 and @xmath234 are wandering subspaces of @xmath224 and we can think of the orthogonal projection onto @xmath235 restricted to @xmath236 as a contractive multi - analytic operator @xmath237 . that this is indeed a multi - analytic operator can be directly verified from @xmath238 , compare @xcite , theorem 1.2 , for a systematic study of the construction of multi - analytic operators from pairs of wandering subspaces . we set @xmath239 which defines a map @xmath44 from contractive liftings @xmath34 of a given row contraction @xmath35 to contractive multi - analytic operators @xmath240 . we also write @xmath241 or @xmath242 if we want to include the dependence on the original row contraction @xmath35 in the notation . recall from section 2 that within the functional model we can in fact always think of @xmath15 as such a restriction of an orthogonal projection . by inspection we observe that if and only if in the construction of @xmath41 we end up with @xmath243 the application of @xmath44 after @xmath41 just reconstructs the original @xmath15 . thus from proposition [ prop : injective ] we conclude : [ me1 ] @xmath244 to make the equality on the left explicit recall that in the functional model space @xmath93 for @xmath245 with injective symbol @xmath40 we have @xmath246 see section 2 and theorem [ fm](b ) . now @xmath222 in the dilation space of @xmath247 , which is provided by the functional model space @xmath93 together with @xmath26 , is related by the canonical unitary to @xmath248 . with this canonical identification of @xmath10 and @xmath248 we have the equality in proposition [ me1 ] . + it remains to determine and to examine the class of contractive liftings which correspond to multi - analytic operators in this way . it is here where the minimality of liftings comes into the game . we restate the definition of minimality in a more explicit way . [ def : minimal ] a lifting @xmath34 of @xmath35 is called minimal if @xmath249 here @xmath250 with @xmath61 if @xmath251 and @xmath252 if @xmath253 . equivalently , the lifting @xmath34 is minimal if and only if @xmath36 is the smallest @xmath34-invariant subspace containing @xmath26 . hence definition [ def : minimal ] is consistent with the definition of minimality given in the introduction . [ prop : minimal ] for a contractive lifting @xmath230 of a row contraction @xmath35 the following are equivalent : * @xmath34 is minimal . * @xmath254 . let @xmath255 . with @xmath256 denoting the orthogonal complement in @xmath36 we have @xmath257^\perp \\ & \leftrightarrow & y \perp e_\alpha x \;(= p_{{\mathcal}h_e } v^e_\alpha x ) \;\text{for all}\ ; x \in { \mathcal}h_c , \;\text{all words}\;\alpha \\ & \leftrightarrow & y \perp p_{{\mathcal}h_e } \big [ { \mathcal}h_c \oplus ( \bigoplus_\alpha v^e_\alpha { \mathcal}l_c ) \big ] = { \mathcal}h_c + p_{{\mathcal}h_e } \bigoplus_\alpha v^e_\alpha { \mathcal}l_c \\ & \leftrightarrow & y \in { \mathcal}h_a \cap ( \bigoplus_\alpha v^e_\alpha { \mathcal}l_c)^\perp\end{aligned}\ ] ] we conclude that @xmath258^\perp = \{0\}$ ] if and only if @xmath254 , which implies the proposition . remark : comparing ( b ) with lemma 3.5(iii ) of @xcite shows that what we call a minimal lifting in this paper is the same as what was called a reduced lifting in @xcite . we prefer the terminology ` minimal ' because definition [ def : minimal ] above is simpler and because this is consistent with the terminology ` minimal isometric dilation ' which is the most important example of a minimal contractive lifting . a number of additional results about minimal or reduced liftings can be found in @xcite , we only include the following basic property which foreshadows the connection between minimal liftings and functional models established later . [ a cnc ] the right lower corner @xmath33 of a minimal contractive lifting @xmath34 of @xmath35 is completely non - coisometric . taking adjoints in the definition of minimal lifting we see that @xmath34 is minimal if and only if there is no @xmath259 such that @xmath260 for all @xmath56 . in other words , if @xmath34 is minimal then for all @xmath261 we find @xmath56 such that @xmath262 and because @xmath34 is contractive it follows that @xmath33 is completely non - coisometric . alternatively we could use corollary [ cor : minimal ] below to establish that @xmath33 arises from a functional model and then quote proposition [ model](b ) to get the result . we now proceed to obtain minimal liftings from multi - analytic operators . [ e minimal ] if @xmath263 is any contractive multi - analytic operator then the lifting @xmath264 of @xmath35 is always minimal . in the construction of @xmath265 using a functional model of @xmath186 it is the subspace @xmath210 in the model which becomes the subspace @xmath235 in the dilation space of @xmath34 . hence by proposition [ prop : minimal ] minimality of @xmath34 is satisfied if and only if in the model we have @xmath266 . but by proposition [ model](a ) this is always the case . [ minimal to injective1 ] if @xmath40 is an injective symbol then there exists a minimal lifting @xmath34 of @xmath35 such that @xmath267 . put @xmath268 . then @xmath34 is minimal by proposition [ e minimal ] and @xmath269 by proposition [ me1 ] . [ minimal to injective2 ] if a contractive lifting @xmath34 of @xmath35 is minimal then the symbol of @xmath270 is injective . suppose the symbol of @xmath270 is not injective . by the definition of @xmath270 this means that there exists @xmath271 which is orthogonal to @xmath272 . further by definition @xmath273 . hence there exists @xmath274 such that @xmath275 . [ in fact , assume that on the contrary @xmath276 for all @xmath14 . then @xmath277 is orthogonal to @xmath278 for all @xmath279 . but @xmath280 , hence @xmath277 is also orthogonal to @xmath36 . it follows that @xmath277 is orthogonal to @xmath222 and hence that @xmath281 , contradicting our assumption above . ] note that @xmath255 and @xmath282 for all @xmath283 , by the assumption about @xmath277 and because the @xmath284 leave this subspace invariant . hence @xmath285 and @xmath34 is not minimal by proposition [ prop : minimal ] . but injectivity of the symbol of @xmath270 does not imply minimality of @xmath34 . for example take a lifting @xmath34 with @xmath286 so that @xmath34 is a direct sum of @xmath35 and @xmath33 . if @xmath287 this is clearly not minimal . in this case @xmath288 and if @xmath289 , that is if @xmath33 is chosen to be isometric , then @xmath270 is the identity . [ me2 ] if @xmath186 is any contractive multi - analytic operator then the symbol of @xmath290 is injective . this follows from proposition [ e minimal ] and proposition [ minimal to injective2 ] . to obtain the one - to - one correspondence stated in theorem [ char ] we need to study the equivalence classes introduced in definition [ def : equivalent ] . [ prop : equivalent ] let @xmath34 and @xmath184 be minimal contractive liftings of a row contraction @xmath35 . then @xmath34 and @xmath184 are unitarily equivalent liftings if and only if @xmath242 and @xmath291 are equivalent . the proof is easier to understand by always using the canonical identifications with full fock spaces and defect spaces . then @xmath292 etc . if @xmath34 and @xmath184 are unitarily equivalent liftings then we can extend the unitary @xmath293 to a unitary @xmath294 which intertwines the minimal isometric dilations @xmath224 and @xmath295 and restricts to the identity on a space @xmath212 contained in both . by restricting @xmath294 we obtain a unitary @xmath296 providing the equivalence of @xmath242 and @xmath291 . ( note that this direction is true even without assuming minimality . ) conversely assume that @xmath242 and @xmath291 are equivalent via @xmath297 . again we identify the space of the minimal isometric dilation of @xmath35 with @xmath212 . then by minimality of @xmath34 we have @xmath266 from proposition [ prop : minimal ] , with @xmath298 and using the identifications announced in the beginning of the proof . taking orthogonal complements in the space @xmath299 of the minimal isometric dilation we note that @xmath300 and hence find @xmath301 . similarly @xmath302 . because @xmath242 resp . @xmath291 are the orthogonal projections onto @xmath212 restricted to @xmath303 resp . @xmath304 we can extend the unitary @xmath305 to a unitary @xmath306 which is the identity on @xmath212 and intertwines the minimal isometric dilations @xmath224 and @xmath295 . because @xmath307 and @xmath308 this restricts to a unitary @xmath185 intertwining @xmath34 and @xmath184 and being the identity on @xmath26 . [ cor : minimal ] for a contractive lifting @xmath34 of a row contraction @xmath35 : @xmath309 by proposition [ e minimal ] we know that @xmath310 is always minimal . because minimality is preserved by unitary equivalence , @xmath34 and @xmath310 can only be unitarily equivalent if @xmath34 is minimal . conversely , if @xmath34 is minimal then by proposition [ minimal to injective2 ] the symbol of @xmath201 is injective , hence by proposition [ me1 ] @xmath311 and now proposition [ prop : equivalent ] implies that @xmath34 and @xmath310 are unitarily equivalent . we remark that there is no canonical identification of @xmath312 and @xmath313 and this unitary equivalence is the best we can expect . + by combining the results obtained so far we are now able to complete the proof of theorem [ char ] and corollary [ char2 ] as follows . let us denote by @xmath314 respectively @xmath315 the mappings between equivalence classes which are given by @xmath44 respectively @xmath41 on representatives . we have to prove that these are well defined and inverse to each other . from proposition [ prop : equivalent ] we find that @xmath314 is well defined and it maps into classes of injective symbols by proposition [ minimal to injective2 ] . conversely , assume that @xmath186 and @xmath187 have injective symbols which are equivalent to each other . the liftings @xmath316 and @xmath317 are both minimal by proposition [ e minimal ] . by proposition [ me1 ] we have @xmath318 and @xmath319 and we conclude from proposition [ prop : equivalent ] that @xmath34 and @xmath184 are unitarily equivalent . hence @xmath315 is well defined and maps into classes of minimal liftings . we have @xmath320 from proposition [ me1 ] and we have @xmath321 by corollary [ cor : minimal ] . proposition [ prop : injective ] shows that the dimension of @xmath10 is equal to the defect of @xmath34 . note further that corollary [ char2 ] is nothing but a restatement of proposition [ minimal to injective2 ] and proposition [ me1 ] established above . we have now completely proved theorem [ char ] and corollary [ char2 ] . let us have a look at two easy examples . first , if @xmath322 then we have the trivial lifting @xmath323 . this is a minimal lifting and the characteristic function @xmath242 is the identity ( here @xmath324 ) . second , if @xmath34 is the minimal isometric dilation of the row contraction @xmath35 then this is a minimal lifting and the characteristic function @xmath242 is the zero function . here @xmath325 , so it is the zero function on the zero space and hence injective : no contradiction to theorem [ char ] . more complicated examples will appear in the following sections . we finish this section with an application . the one - to - one correspondence established in theorem [ char ] very naturally leads to the possibility to examine the structure of the set of minimal contractive liftings via the corresponding characteristic functions . along these lines we obtain a result about the factorization of the characteristic function of a minimal lifting . [ factor ] let @xmath35 be a row contraction on a hilbert space @xmath26 . if @xmath34 is a minimal contractive lifting of @xmath35 on a hilbert space @xmath326 and @xmath327 is a minimal contractive lifting of @xmath34 on a hilbert space @xmath328 then @xmath184 is a minimal contractive lifting of @xmath35 and for the characteristic function we have @xmath329 conversely , if @xmath184 is a minimal contractive lifting of @xmath35 on a hilbert space @xmath330 and the characteristic function @xmath331 can be written as @xmath332 where @xmath333 and @xmath334 are contractive multi - analytic operators , for a hilbert space @xmath10 and both with injective symbols , then there exists a minimal contractive lifting @xmath34 of @xmath35 such that @xmath184 is a minimal contractive lifting of @xmath34 , and @xmath335 and @xmath336 are equivalent to @xmath242 and @xmath337 respectively . the first half is theorem 4.1 of @xcite . its proof can be simplified in the present setting . it is easy to check that contractivity and minimality are both preserved if we iterate liftings . further let @xmath338 the spaces of minimal isometric dilations for which we have @xmath339 if we iterate liftings . we can think of @xmath340 as the orthogonal projection @xmath341 from @xmath342 onto @xmath343 restricted to @xmath304 because this maps to @xmath210 . similar for @xmath344 and @xmath337 . with this observation the factorization @xmath345 follows from the obvious factorization @xmath346 . let us now prove the converse direction . we use the map @xmath41 , in particular corollary [ char2 ] and proposition [ prop : injective ] . from @xmath335 with its injective symbol @xmath347 we can build the minimal contractive lifting @xmath348 of @xmath35 which has characteristic function @xmath335 and defect equal to @xmath349 . then from @xmath336 with its injective symbol @xmath350 we can build the minimal contractive lifting @xmath351 of @xmath352 which has characteristic function @xmath336 . we can think of @xmath353 as a minimal contractive lifting of @xmath35 with characteristic function @xmath354 , by the first part above ( or theorem 4.1 in @xcite ) . hence with the assumption that @xmath184 has characteristic function @xmath354 we can use proposition [ prop : equivalent ] to conclude that @xmath184 and @xmath353 are unitarily equivalent as liftings of @xmath35 . if we use this unitary to rotate the lifting @xmath352 then we obtain a minimal contractive lifting @xmath34 of @xmath35 with the properties required . recall the notion of a characteristic function of a completely non - coisometric row contraction @xmath33 on @xmath163 from @xcite : the space @xmath355 of the minimal isometric dilation @xmath356 not only contains the wandering subspace @xmath357 but also the wandering subspace @xmath358 arising in the wold decomposition of @xmath356 into the direct sum of a row unitary part and a row shift part . the space @xmath359 is canonically identified with the @xmath360-defect space @xmath361 , the closure of @xmath362 . here @xmath363 and the canonical unitary from @xmath361 onto @xmath359 is given by @xmath364 ( where @xmath365 ) . then the orthogonal projection onto the space of the row shift part restricted to @xmath366 can be thought of as a contractive multi - analytic operator @xmath367 with a symbol @xmath368 . this is called the _ characteristic function _ of @xmath33 . it is verified in @xcite , theorem 4.1 , that if we construct the functional model for @xmath369 and then form the compression to @xmath163 ( as we did more generally in section 2 ) then we recover the original @xmath33 we started from , up to unitary equivalence . our proposition [ model ] implies that this is only possible if @xmath33 is completely non - coisometric and the functional model constructed from @xmath369 adds the insight that all completely non - coisometric row contractions can be obtained in this way . recall that a contractive multi - analytic operator @xmath370 with symbol @xmath188 is called _ purely contractive _ if @xmath371 for all @xmath372 . we attach a name , motivated in section 5 , to another property which already appears in @xcite as ( 5.1 ) . [ sz ] we say that a contractive multi - analytic operator @xmath373 satisfies the szeg condition if @xmath374 where @xmath375 . recall that multi - analytic operators @xmath186 and @xmath187 with symbols @xmath188 and @xmath376 are said to _ coincide _ ( compare @xcite ) if there are unitaries @xmath377 and @xmath378 so that @xmath379 . with this terminology we can now restate one of the main results obtained in sections 4 and 5 of @xcite , as follows . [ @xcite ] [ popescu ] a contractive multi - analytic operator coincides with the characteristic function of a completely non - coisometric row contraction if and only if it is purely contractive and satisfies the szeg condition . we want to discuss the relationship between this result and our theory of characteristic functions of liftings . [ pure ] let @xmath380 be a symbol . if @xmath40 has no constant directions ( see definition [ non - constant ] ) then it is purely contractive . if the szeg condition for @xmath15 holds then , conversely , purely contractive implies having no constant directions . recall that having no constant directions means that @xmath144 only for @xmath145 ( compare definition [ non - constant ] ) . this implies purely contractive , in fact in the remaining case @xmath381 it is obvious that @xmath382 . the converse direction , going from purely contractive to having no constant directions , only fails if there exists @xmath372 such that @xmath383 but @xmath384 . we show that this contradicts the szeg condition . in fact , from @xmath385 we conclude that @xmath386 for all @xmath14 , hence @xmath387 for all @xmath274 and @xmath388 . we conclude that @xmath389 and @xmath390 for all @xmath274 . because @xmath391 we have @xmath392 , so the szeg condition can not hold . [ pop - min ] the characteristic function of a completely non - coisometric row contraction has no constant directions and ( hence ) it is injective . the isometric dilation provided by the functional model is minimal . from theorem [ popescu ] together with lemma [ pure ] we conclude that we have no constant directions and then by proposition [ min - dil ] the isometric dilation provided by the functional model is minimal . the second statement is also part of @xcite , theorem 5.1 , where an alternative proof for it can be found . [ cnc ] let @xmath203 be a row contraction on a hilbert space @xmath26 and @xmath230 be a minimal contractive lifting on a hilbert space @xmath30 with characteristic function @xmath393 . then @xmath242 is purely contractive and satisfies the szeg condition if and only if @xmath394 and @xmath395 . in this case , with the canonical identifications , the characteristic function @xmath393 of the lifting @xmath34 is equal to the characteristic function @xmath367 of the row contraction @xmath33 . the characteristic function @xmath369 of any completely non - coisometric row contraction @xmath33 can be written in this form for any row contraction @xmath35 with defect equal to @xmath396 . from @xcite : two characteristic functions @xmath369 and @xmath397 coincide if and only if @xmath33 and @xmath398 are unitarily equivalent . note that because @xmath34 is minimal the symbol @xmath40 of @xmath242 is injective by proposition [ minimal to injective2 ] . restricted to @xmath399 the minimal isometric dilation @xmath224 is given by the functional model row isometry @xmath66 corresponding to @xmath242 , as described in section 2 . with theorem [ fm](b ) and proposition [ min - dil ] we see that here @xmath400 is equivalent to @xmath66 on @xmath93 being the minimal isometric dilation of @xmath33 . recall that @xmath66 can be represented by @xmath401 for @xmath402 . the szeg condition says exactly that the ranges of the maps @xmath403 for all @xmath279 have as their closed linear span the whole space @xmath404 , hence , under the condition @xmath400 , it is equivalent to @xmath405 . we can now verify the first part of the theorem . note that if we have @xmath394 and @xmath395 then @xmath406 follows from the fact that both are described by the same restriction of an orthogonal projection . assume @xmath394 . then @xmath40 has no constant directions by theorem [ fm](b),(c ) and hence @xmath242 is also purely contractive by lemma [ pure ] . assume additionally @xmath395 . then the szeg condition holds , as shown above . conversely from the szeg condition together with @xmath242 purely contractive we conclude by lemma [ pure ] that @xmath40 has no constant directions which implies @xmath394 , again by theorem [ fm](b),(c ) . as shown above , under this condition the szeg condition also implies @xmath395 . the first part of the theorem is proved . if we start with any completely non - coisometric row contraction @xmath33 then by corollary [ pop - min ] the functional model for @xmath369 provides a minimal isometric dilation for @xmath33 . if we choose any @xmath35 with defect equal to @xmath396 then we can build @xmath407 ( as shown in section 3 ) based on any ( unitary ) identification of @xmath361 and @xmath43 . then @xmath408 . note that also @xmath400 from proposition [ min - dil ] . we conclude that @xmath406 by the first part . this shows that indeed for any completely non - coisometric row contraction @xmath33 the characteristic function @xmath369 appears in suitable liftings by @xmath33 as the characteristic function of the lifting . if @xmath33 and @xmath398 are unitarily equivalent then clearly @xmath369 and @xmath397 coincide ( because the minimal isometric dilations are unitarily equivalent ) . conversely assume that @xmath369 and @xmath397 coincide , so we have unitaries @xmath409 and @xmath410 so that @xmath411 . now for any @xmath35 with defect equal to @xmath412 , we can choose any unitary @xmath413 and use it to identify the spaces @xmath414 and @xmath43 . then the unitary @xmath415 can be used to identify @xmath361 and @xmath43 . based on these identifications we construct @xmath416 and @xmath417 and we can check that , with these identifications , @xmath406 and @xmath418 are actually equivalent . by proposition [ prop : equivalent ] it follows that the corresponding liftings @xmath34 and @xmath184 are unitarily equivalent and this implies , by restriction , the unitary equivalence of @xmath33 and @xmath398 . we finish this section with a few comments . the final part of the proof of theorem [ cnc ] illustrates the relationship between the concepts of coincidence and equivalence , compare the comments after definition [ def : equivalent ] . note that in the proof of theorem [ cnc ] the only thing we used from the theory of characteristic functions of completely non - coisometric row contractions ( from @xcite ) was the fact that the functional model constructed from the characteristic function provides a minimal isometric dilation . given an independent proof of this fact we have obtained a new proof of theorem [ popescu ] because clearly theorem [ popescu ] follows from theorem [ cnc ] . we claim that the new context of liftings simplifies some of the original arguments . while in the theory of characteristic functions of completely non - coisometric row contractions the concept of a purely contractive function is fundamental it seems that in the wider class of characteristic functions of liftings the concept of having no constant directions is more fundamental . theorem [ cnc ] does not solve all natural questions about the relationship between the different notions of characteristic functions : whenever we have a characteristic function of a lifting then we always have associated the characteristic function of the completely non - coisometric row contraction in the right lower corner . only in the very special situation of theorem [ cnc ] the two functions are equal . there are some computations in the general case in @xcite , section 4 , see also @xcite , section 2 , but we leave a more systematic treatment of this topic to future work . we finally remark that in the original theory of sz .- nagy and foias @xcite for @xmath22 the more general case of a completely non - unitary contraction was treated and that ball and vinnikov in @xcite provided a generalization of it for @xmath39 . the relationship of our setting with the theory in @xcite is another promising field of investigation . we are content in this section to illustrate our results by examples and do not develop a complete theory . recall that an analytic function @xmath40 in the open unit disk which satisfies @xmath419 is called a _ schur function . _ we refer to @xcite for further information about schur functions . we can think of @xmath40 as a symbol of a contractive multiplication operator @xmath15 on the _ hardy space _ @xmath420 , the hilbert space of analytic functions in the open unit disk with square summable taylor coefficients and isomorphic in this way to the unilateral sequence space @xmath421 . this is the special case @xmath22 and @xmath23 in our general scheme for contractive multi - analytic operators . note that , to simplify notation , we use @xmath40 in two different ways here : the schur function @xmath422 for @xmath423 with @xmath424 has a sequence of taylor coefficients which is the image of @xmath425 under @xmath40 if used in the way we introduced the symbol for multi - analytic operators . note also in this respect that identifying a one - dimensional space with @xmath426 is not canonical but involves the choice of a unimodular factor . we call the schur function the _ spectral representation _ of @xmath40 . note further that a schur function @xmath40 represents an injective symbol ( in the sense used in previous sections ) if and only if it is non - zero . two schur functions @xmath40 and @xmath192 are equivalent in the sense of definition [ def : equivalent ] if and only if @xmath427 with @xmath428 and @xmath429 . let @xmath46 be a contraction with defect equal to @xmath0 . then by theorem [ char ] we have a one - to - one correspondence between unitary equivalence classes of minimal contractive liftings @xmath45 ( of @xmath46 ) with defect equal to @xmath0 and non - zero schur functions ( up to unimodular complex factors ) . applying theorem [ fm ] we confirm that for all non - zero schur functions @xmath430 ( equal to the defect of @xmath45 ) and find in addition that always @xmath431 and that @xmath432 if and only if ( the spectral representation of ) @xmath40 is not a constant function . + let @xmath433 be the contraction on a hilbert space @xmath434 . the defect is equal to @xmath0 . consider the mbius transformation @xmath435 because @xmath40 is an inner function , the multiplication operator @xmath436 is an isometry and we note that @xmath437 . the model space is @xmath438 . in this case @xmath439 consists of all functions @xmath440 with @xmath441 . hence @xmath442 is one - dimensional and equal to the scalar multiples of the function @xmath443 . we want to construct the lifting @xmath444 , compare section 3 for the procedure and for the notation . we find @xmath445 with the isometry @xmath446 so that @xmath447 ( the second summand is a constant function in @xmath420 ) and the restriction of @xmath446 to @xmath420 is the multiplication operator @xmath448 with the variable @xmath182 . we compute @xmath449 and @xmath450 and conclude that the minimal contractive lifting of @xmath433 with the mbius transformation @xmath451 as its characteristic function is the ( scalar ) @xmath452-matrix @xmath453 it is easy to check that the defect of @xmath45 is indeed equal to @xmath0 . as a non - constant inner function the mbius transformation @xmath451 clearly satisfies the conditions in theorem [ cnc ] and indeed it is well known that @xmath40 is the characteristic function of the ( completely non - coisometric ) contraction @xmath454 . let us see what theorem [ cnc ] tells us in general about schur functions as characteristic functions . for this consider @xmath455 , the hilbert space of square integrable functions on the unit circle with lebesgue measure . this can be achieved by considering boundary values . for a schur function @xmath40 and @xmath440 in general @xmath456 ( the latter function is not always in @xmath420 ) , but because @xmath457 we can replace @xmath458 and @xmath459 by the multiplication with @xmath460 and the subspace @xmath461 ( using the unitary defined by @xmath462 ) . [ cnc schur ] let @xmath40 be a non - zero schur function . * @xmath15 is purely contractive if and only if ( in the spectral representation ) @xmath463 . * @xmath15 satisfies the szeg condition ( see definition [ sz ] ) if and only if ( the spectral representation of ) @xmath40 satisfies the _ spectral szeg condition _ : @xmath464 * a schur function @xmath40 is the characteristic function of a completely non - coisometric contraction if and only if it satisfies @xmath463 and the spectral szeg condition . \(a ) is obvious because @xmath465 is equal to the zeroth taylor coefficient of @xmath40 . with the considerations preceding this proposition we see that the szeg condition ( from definition [ sz ] ) holds if and only if the constant polynomial @xmath0 can be approximated in norm by analytic polynomials @xmath466 with @xmath467 in the hilbert space @xmath468 . here the non - negative function @xmath469 plays the role of a density for lebesgue measure . by a classical theorem of szeg this is the case if and only if the spectral szeg condition holds , see @xcite , chapter 4 . this proves ( b ) . now ( c ) is just the reformulation of theorem [ cnc ] in the setting of schur functions . it is a classical result that a schur function @xmath40 satisfies what we have called the spectral szeg condition if and only if it is an extremal schur function , see @xcite , chapter 9 . generalizations of this result for the case of operator - valued schur functions ( still with @xmath22 ) are given in @xcite , generalizations for @xmath39 are discussed in @xcite , section 1.4 , in connection with the notion of prediction entropy . note at this point that any non - zero schur function is the characteristic function of a lifting but only a very special subclass consists of characteristic functions of completely non - coisometric contractions . we give an example outside of this special subclass . again we start with the contraction @xmath433 on a one - dimensional hilbert space @xmath26 but this time we look at the schur function @xmath470 . clearly this does not satisfy the spectral szeg condition . again we want to construct the minimal contractive lifting @xmath471 . in this case the model space is @xmath472 . hence @xmath473 and the isometry @xmath446 maps @xmath474 to @xmath475 and acts as @xmath476 on @xmath477 . further @xmath478 if and only if @xmath479 which is satisfied if and only if @xmath480 . so we have @xmath481 here @xmath482 for @xmath483 ( note that @xmath484 applied to a constant function yields zero ) and for @xmath485 we find @xmath486 the last equality can be checked by verifying that the difference @xmath487 is indeed orthogonal to @xmath163 . we note that any factorization @xmath488 among non - zero schur functions leads to a corresponding factorization of liftings , by theorem [ factor ] . in particular the lifting in the previous example can be factorized in many ways . but we postpone a detailed investigation of this phenomenon to another place . we finally remark that the general case of minimal contractive liftings for a single contraction ( the general case @xmath22 ) can be handled in a similar way . in fact , for @xmath22 a multi - analytic operator has a spectral representation by an operator - valued schur function , i.e. , a bounded analytic function @xmath40 on the open unit disk such that @xmath489 and @xmath490 for all @xmath182 in the open unit disk . then theorem [ char ] takes the form that there is a one - to - one correspondence between unitary equivalence classes of minimal contractive liftings @xmath45 with defect @xmath349 ( of a given contraction @xmath46 ) and operator - valued schur functions with values in @xmath491 , up to a unitary on @xmath10 ( the same for all @xmath182 ) , and with the following injectivity property : if @xmath372 then @xmath492 is not the zero function . in our analysis of functional models in section 2 we need a few technical results about how the kernel of the adjoint changes if we go to a restriction of an operator . we provide them in the following lemma . these are quite general observations , useful in particular to describe the geometry of invariant subspaces . \(i ) for @xmath503 @xmath504 ( ii ) decompose @xmath505 as @xmath506 such that @xmath507 and @xmath508 . since @xmath509 it follows that @xmath510 . thus @xmath511 . we conclude that @xmath512 by ( i ) . finally because @xmath513 we obtain @xmath514 which is ( ii ) . for the opposite inclusion let @xmath524 . because @xmath525 we have @xmath526 and @xmath527 by ( i ) . but from @xmath528 and @xmath529 we also get , using the definition of @xmath11 , that @xmath530 . hence @xmath531 and @xmath532 . we have now established the second equality in ( iii ) . finally it is clear that @xmath533 , hence @xmath534 . on the other hand we have seen above that @xmath521 and we have @xmath535 by ( ii ) . hence we also have @xmath536 and we have proved the first equality in ( iii ) . dey , s. ; haria , k.j . : _ generalized repeated interaction model and transfer functions _ , operator theory in harmonic and non - commutative analysis , series : operator theory : advances and applications , vol . 240 , birkhuser ( 2014 ) , 111 - 135 . gohm , r. : _ transfer functions for pairs of wandering subspaces _ , spectral theory , mathematical system theory , evolution equations , differential and difference equations , series : operator theory : advances and applications , * 221 * , birkhuser ( 2012 ) , 385 - 398 . treil , s.r . : _ geometric methods in spectral theory of vector - valued functions : some recent results _ , in : toeplitz operators and spectral function theory : essays from the leningrad seminar on operator theory ( ed . n. nikolski ) , pp . 209 - 280 , oper . theory adv . 42 , birkhuser ( 1989 )

the ensemble of euclidean gluon field configurations represented by the domain wall network is considered . a single domain wall is given by the sine - gordon kink for the angle between chromomagnetic and chromoelectric components of the gauge field . the domain wall separates the regions with self - dual and anti - self - dual fields . the network of the domain wall defects is introduced as a combination of multiplicative and additive superpositions of kinks . the character of the spectrum and eigenmodes of color - charged fluctuations in the presence of the domain wall network is discussed . the concept of the confinement - deconfinement transition in terms of the ensemble of domain wall networks is outlined . conditions for the formation of a stable thick domain wall junction ( the chromomagnetic trap ) during heavy ion collisions are discussed , and the spectrum of color charged quasiparticles inside the trap is evaluated . an important observation is the existence of the critical size of the trap stable against gluon tachyonic modes , which means that deconfinement can occur only in a finite region of space - time in principle . the size is related to the value of gluon condensate .

1403.0415

in general , diffusion of the relativized versions of ideas born in condensed matter and solid state physics to the quantum field theory has been proven to be extremely fruitful . it was realised long time ago that a complex of problems associated with investigation of the qcd vacuum structure appeared as particularly suitable object in this respect . this paper is focused on the further development of approach to qcd vacuum as a medium describable in terms of statistical ensemble of domain wall networks . this concept plays important role in description of condensed matter systems with rival order and disorder but has been insufficiently explored in application to qcd vacuum . the identification of the properties of nonperturbative gauge field configurations relevant to a coherent resolution of confinement , chiral symmetry breaking , @xmath2 and strong cp problems is an overall task pursued by most approaches to the qcd vacuum structure . as a rule , analytical as well as lattice qcd studies of qcd vacuum structure are focused on localized topological configurations ( instantons , monopoles and dyons , vortices ) which _ via _ condensation could be seen as appropriate gauge field configurations responsible for confinement of static color charges and other nonperturbative features of strong interactions . in recent years , three - dimensional configurations akin to domain walls became popular as well @xcite . first of all , these are the @xmath3 domain walls related to the center symmetry of the pure yang - mills theory @xcite and double - layer domain wall structures in topological charge density @xcite . lattice qcd serves as a main source of motivation and verification tool for these studies in pair with the theoretically appealing scenario of static quark confinement in the spirit of the dual meissner effect equipped with the wilson and polyakov loop criteria . the localized configurations are characterized by the vanishing ratio of the action to the 4-volume in the infinite volume limit . in this sense , instantons , monopoles , vortices and double - layer domain walls are localized configurations . a complementary treatment of the above mentioned overall task is based on the investigation of the properties of quantum effective action of qcd . as in other quantum systems with infinitely many degrees of freedom , the global minima of the effective action define the phase structure of qcd . the identification of global minima in different regimes ( high energy density , high baryon density , strong external electromagnetic fields ) has highest priority for understanding the phase transformations in hadronic matter . in general , a nontrivial global minimum corresponds to a gauge field with the strength not vanishing at space - time infinity and , hence , extensive action proportional to the four dimensional space - time volume of the system , unlike the localized configurations . a variety of essentially equivalent statements of the problem in the context of qcd can be found , for instance , in @xcite . global minima related by discrete symmetry transformations like cp , weyl symmetry in the root space of @xmath4 , center symmetry in particular , is a reason to look for field configurations interpolating between them . first of all , these are domain wall configurations , but also lower dimensional topological defects . there is among others one difference between this treatment and approaches based on localized objects : the last one intends to merge the initially isolated objects ( e.g. , instanton gas or liquid ) while the former collects defects in an initially homogeneous background . at first sight , both ways seem to lead to a similar outcome - a class of nonperturbative gluon field configurations with a self - consistent balance of order and disorder which can be characterized , in particular , by nonzero gluon condensate and topological charge density . however , essential disparity can arise since a superposition of localized objects inherits the properties of isolated objects while the superposition of defects in the initially homogeneous ordered background brings some disorder and merely refines the overall properties of the background . for instance , the superposition of infinitely many instantons and anti - instantons is not a configuration with a finite classical action but it maintains the property to have integer - valued topological charge . on the contrary , the configuration obtained by implanting infinitely many domain wall defects into the abelian covariantly constant ( anti-)self - dual field can have any real value of the mean topological charge density as well as any real value of topological charge fraction per domain @xcite . both configurations can be seen as lumps of the topological charge density distributed in the euclidean space - time like in fig.[fig : kink_network ] or in the lattice configurations @xcite . however , in the instanton picture each lump carries an integer charge while in the treatment of global minima the charge is any real , irrational , for instance , number . this can have dramatic consequences for the fate of @xmath5 parameter in qcd and the natural resolution of the strong cp - problem @xcite . in the euclidean formulation , the statement of the problem starts with the very basic symbol of the functional integral @xmath6\ } , \end{aligned}\ ] ] where the functional space @xmath7 is subject to the condition @xmath8 the constant @xmath9 is not equal to zero in the general case , which is equivalent to nonzero gluon condensate @xmath10 . condition ( [ cond0 ] ) singles out fields @xmath11 with the strength which is constant almost everywhere in @xmath12 . it is a necessary requirement to allow gluon condensate to be nonzero . it does not forbid also fields with a finite action since the case @xmath13 has to be also studied . the dynamics chooses the value of @xmath9 . however , the phenomenology of strong interactions has already required nonzero gluon and quark condensates . hence , they must be allowed in the qcd functional integral from the very beginning . separation of the long range modes @xmath11 responsible for gluon condensate and the local fluctuations @xmath14 in the background @xmath11 , must be supplemented by the gauge fixing condition . the background gauge condition for fluctuations @xmath15 is the most natural choice . further steps include integration over the fluctuation fields resulting in the effective action for the long - range fields and identification of the minima of this effective action ( for more details see @xcite ) which dominate over the integral in the limit @xmath16 and define the phase structure of the system . as soon as minima are identified , this setup defines a principal scheme for self - consistent identification of the class of gauge fields which almost everywhere coincide with the global minima of the quantum effective action . a treatment of these `` vacuum fields '' in the functional integral @xmath17 \\ & & \times\delta[d(b)q ] \exp\{-s[b+q]+s[b]\}.\end{aligned}\ ] ] must be nonperturbative . the fields @xmath18 are subject to condition ( [ cond0 ] ) with the fixed vacuum value of the condensate @xmath19 . the condensate plays the role of the scale parameter of qcd to be identified from the hadron phenomenology . the fluctuations @xmath20 in the background of the vacuum fields can be seen as perturbations . the homogeneous fields with the domain wall defects are the most natural and simplest example of gluon configurations which are homogeneous almost everywhere in @xmath12 and satisfy the basic condition eq.([cond0 ] ) . basic argumentation in favour of the abelian ( anti-)self - dual homogeneous fields as global minima of the effective action originates from papers @xcite . within the ginzburg - landau approach to the effective action the domain wall is described simply by the sine - gordon kink for the angle between chromomagnetic and chromoelectric components of the gluon field @xcite . this kink configuration can be seen as either bloch or nel domain wall separating the regions with self - dual and anti - self - dual abelian gauge fields . on the domain wall the gluon field is abelian with orthogonal to each other chromomagnetic and chromoelectric fields . we shall not repeat here arguments leading to this conclusion but just refer to papers @xcite where a more detailed discussion can be found . group theoretical analysis of the weyl symmetry and subgroup embeddings behind the domain wall formation in the effective gauge theories is given in a recent paper @xcite . it should be also mentioned that the model of confinement , chiral symmetry breaking and hadronization based on the dominance of the gluon fields which are ( anti-)self - dual abelian almost everywhere demonstrated high phenomenological performance @xcite . the purpose of the present paper is to evolve the approach outlined in article @xcite in two respects : explicit analytical construction of the domain wall network in @xmath12 through a combination of additive and multiplicative superpositions of kinks , and refining the spectrum and eigenmodes of the color charged scalar , spinor and vector fields in the background of a domain wall . in particular the spectrum of quasiparticles inside the thick domain wall junction is evaluated . it is shown that the standard methods of the sine - gordon model @xcite allow one to generate various domain wall networks . the eigenvalues and eigenfunctions are found for the laplace operator in the background of a single infinitely thin domain wall . in this case , the eigenvalue problem has to be solved separately in the bulk of @xmath12 and on the 3-dimensional hyperplane of the wall . the continuity of the charge current through the wall is required together with the square integrability of the bulk eigenfunctions . for the infinitely thin wall the bulk eigenfunction possesses the purely discrete spectrum which coincides with the spectrum for the case of a homogeneous ( anti-)self - dual field without a kink defect . the eigenfunctions differ in a certain way but have the same harmonic oscillator type as the ones in the absence of the kink defect . these modes describe confined color charged fields . the eigenfunctions localized on the wall have continuous spectrum with the dispersion relation of charged quasi - particles . this confirms qualitative conjectures of @xcite . it is argued that thick domain wall junction may be formed during heavy ion collisions and play the role of a trap for charged quasi - particles . confinement is lost inside the trap of a finite size . there exists a critical size of the stable trap , beyond which the emerging tachyonic gluon modes destroy it . the paper is organized as follows . section ii is devoted to the domain wall network construction . the spectrum of scalar color charged field in the background of infinitely thin domain wall is discussed in section iii . in the fourth section we discuss the chromomagnetic trap formation and evaluate the spectrum and eigenmodes of the color charged scalar , vector and spinor quasiparticles inside the trap . the calculation of the effective quantum action for the abelian ( anti-)self - dual homogeneous gluon field within the functional renormalization group approach @xcite has indicated that this configuration is a serious candidate for the role of global minimum of qcd effective action and has enhanced the older one - loop results @xcite . the functional rg result also supported conclusions of @xcite based on the ginzburg - landau type effective lagrangian of the form @xmath21 where @xmath22 is a scale of qcd related to gluon condensate , @xmath23 , and @xmath24 detailed discussion of this expression can be found in @xcite . here it should be noted that all symmetries of qcd are respected and the signs of the constants are chosen so that the action is bounded from below and its minimum corresponds to the fields with nonzero strength , i.e. @xmath25 at the minimum . thus , an important input is the existence of the nonzero gluon condensate . by inspection , one gets as an output twelve ( for @xmath26 ) global degenerate discrete minima . the minima are achieved for covariantly constant abelian ( anti-)self - dual fields @xmath27 where the matrix @xmath28 belongs to the cartan subalgebra of @xmath29 @xmath30 the values @xmath31 correspond to the boundaries of the weyl chambers in the root space of @xmath29 . the minima are connected by the discrete parity and weyl transformations , which indicates that the system is prone to existence of solitons ( in real space - time ) and kink configurations ( in euclidean space ) . below we shall concentrate on the simplest configuration kink interpolating between self - dual and anti - self - dual abelian vacua . if the angle @xmath32 between chromoelectric and chromomagnetic fields is allowed to deviate from the constant vacuum value and all other parameters are fixed to the vacuum values , then the lagrangian takes the form @xmath33 with the corresponding sine - gordon equation @xmath34 and the standard kink solution @xmath35 interpolating between @xmath36 and @xmath37 . here @xmath38 stays for one of the four euclidean coordinates . the kink describes a planar domain wall between the regions with almost homogeneous abelian self - dual and anti - self - dual gluon fields . chromomagnetic and chromoelectric fields are orthogonal to each other on the wall , see fig.[fig : single_kink ] . far from the wall , the topological charge density is constant , its absolute value is equal to the value of the gluon condensate . the topological charge density vanishes on the wall . the upper plot shows the profiles of the components of the chromomagnetic and chromoelectric fields corresponding to the bloch domain wall the chromomagnetic field flips in the direction parallel to the wall plane . ( lower plot ) . here @xmath32 is the angle between the chromomagnetic and chromoelectric fields , @xmath39 . the three - dimensional planar domain wall separates the four - dimensional regions filled with the self - dual ( blue color ) and anti - self - dual ( red color ) abelian covariantly constant gluon fields . the chromomagnetic and chromoelectric fields are orthogonal to each other inside the wall ( green color ) . , width=283 ] ( lower plot ) . here @xmath32 is the angle between the chromomagnetic and chromoelectric fields , @xmath39 . the three - dimensional planar domain wall separates the four - dimensional regions filled with the self - dual ( blue color ) and anti - self - dual ( red color ) abelian covariantly constant gluon fields . the chromomagnetic and chromoelectric fields are orthogonal to each other inside the wall ( green color ) . , width=264 ] ) . , width=264 ] the domain wall network can be now constructed by the standard methods @xcite . let us denote the general kink configuration as @xmath40 where @xmath41 is the inverse width of the kink , @xmath42 is a normal vector to the plane of the wall , @xmath43 with @xmath44 - coordinates of the wall . the topological charge density for the multiplicative superposition of two kinks with the normal vectors anti - parallel to each other @xmath45 is shown in fig.[multsupkink ] . the additive superposition of infinitely many pairs @xmath46 gives a layered topological charge structure in @xmath12 , fig.[fig : layered ] . formally , one may try to go further and consider the product @xmath47 for an appropriate choice of normal vectors @xmath48 this superposition represents a lump of anti - self - dual field in the background of the self - dual one , in two , three and four dimensions for @xmath49 , respectively . the case @xmath50 is illustrated in fig.[fig : asd_lump ] . the general kink network is then given by the additive superposition of lumps ( [ asd_lump ] ) @xmath51 the correponding topological charge density is shown in fig . [ fig : kink_network ] . this figure as well as the lhs of fig . [ fig : layered ] represents the configuration with infinitely thin domain wall defects , that is the abelian homogeneous ( anti-)self - dual field almost everywhere in @xmath12 characterized by the nonzero absolute value of the topological charge density which is constant and proportional to the value of the action density almost everywhere . the most rhs plots in figs . [ fig : layered ] and [ fig : kink_network ] show the opposite case of the network composed of very thick kinks . green color corresponds to the gauge field with an infinitesimally small topological charge density . study of the spectrum of colorless and color charged fluctuations indicates that the lhs configuration is expected to be confining ( only colorless hadrons can be excited as particles ) while the rhs one ( crossed orthogonal field ) supports the color charged quasiparticles as the elementary excitations . it is expected that the rhs configuration can be triggered by external electromagnetic fields @xcite . strong electromagnetic fields emerge in relativistic heavy ion collisions @xcite . even after switching off the external electromagnetic field the nearly pure chromomagnetic vacuum configuration ( rhs fig.[fig : kink_network ] ) can support strong anisotropies @xcite and , in particular , influence the chiral symmetry realization in the collision region @xcite . more detailed consideration of the spectrum of elementary color charged excitations at the domain wall junctions ( the green regions ) is given in the section [ trap ] . a comment on representation of the domain wall network in terms of the vector potential is in order . the domain wall network constructed in this section relies on the separation of the abelian part from the general gauge field . the vector potential representation can be easily realized for the planar bloch domain wall and their layered superposition , fig . [ fig : layered ] . the same is true also for the interior of a thick domain wall junction , where field is almost homogeneous . the description of the domain walls in the general network fig . [ fig : kink_network ] in terms of the vector potential requires application of the gauge field parametrization suggested in a series of papers by y.m . cho @xcite , s. shabanov @xcite , l.d . faddeev and a. j. niemi @xcite and , recently , by k .- kondo @xcite . in this parameterization the abelian part @xmath52 of the gauge field @xmath53 is separated manifestly , @xmath54\hat{n } ( x)=b_\mu(x)\hat{n}(x ) , \nonumber \\ { \hat c}_\mu ( x ) & = & g^{-1}\partial_\mu \hat{n}(x)\times \hat{n}(x ) , \nonumber\\ { \hat x}_\mu ( x ) & = & g^{-1}{\hat n}(x ) \times \left ( \partial_\mu { \hat n}(x ) + g { \hat a}_\mu ( x ) \times { \hat n}(x ) \right ) , \nonumber\end{aligned}\ ] ] where @xmath55 , @xmath56 , @xmath57 , and @xmath58=if^{abc}t^c.\end{aligned}\ ] ] the field @xmath59 is seen as the abelian field in the sense that @xmath60=0 $ ] . the color vector field @xmath61 may be used for detailed description of the thin domain wall junctions in general case . this issue is beyond the scope of the present paper and will be considered elsewhere . . , title="fig:",width=94 ] . , title="fig:",width=94 ] . , title="fig:",width=94 ] in this section we study the spectrum of color charged field fluctuations in the background of a single planar domain wall of the bloch type . the best thing to do would be to solve the eigenvalue problem for the kink of the finite width . however , the problem turns out to be not that simple . let us consider the problem for the scalar field in the adjoint representation , that is just the faddeev - popov ghost field in the background gauge . the quadratic part of the action for the scalar field in the background field of a planar kink with the finite width placed at @xmath62 looks like @xmath63&=&-\int d^4x ( d_\mu\phi)^\dagger(x)d_\mu\phi(x ) \\ \nonumber & = & \int d^4x \phi^\dagger(x)d^2\phi(x ) , \\ \nonumber d_\mu&=&\partial_\mu+i\breve b_\mu , \ , \breve b_\mu=-\breve n b_\mu(x).\end{aligned}\ ] ] here @xmath64 is the constant color matrix , @xmath65 is the vector potential for the planar bloch domain wall . for our purposes the most convenient gauge for @xmath65 is @xmath66 a kink with the finite width is a regular everywhere in @xmath12 function , the scalar field is assumed to be a continuous square integrable function . integration by parts in eq.([action_sc ] ) does not generate surface terms either at infinity or at the location of the kink . however , there is a peculiarity related to the chosen gauge of the background field . according to eq.([gauge1 ] ) , @xmath67 the action can be written as @xmath68&=&\int d^4x \phi^\dagger(x)\tilde d^2\phi(x ) \label{surf_gauge}\\ & -&i\int d^4x \phi^\dagger(x)\breve n\phi(x ) \left[h^\prime_2(x_1)x_3+h^\prime_3(x_1)x_2\right ] . \nonumber\end{aligned}\ ] ] it should be noted that the integral in the second line is equal to zero if @xmath69 is an even function of @xmath70 and @xmath71 . the structure of @xmath72 in eq.([covder1 ] ) is quite complicated . in the eigenvalue problem the variables can hardly be separated in the case of the finite width of the kink . the problem becomes much simpler and tractable in the limit of the infinitely thin domain wall @xmath73 . this limit brings discontinuity into the background field and thus creates a sharp boundary the hyperplane of the domain wall . in such a situation one has to solve the problem in the bulk and on the wall and match the solutions according to some appropriate conditions . for our choice of the kink location there are three regions to be studied : @xmath74 with the self - dual field @xmath65 , @xmath75 with the anti - self - dual field , and @xmath62 with the chromomagnetic and chromoelectric fields orthogonal to each other . conditions imposed onto the eigenmodes of color charged fields on the sharp wall can be obtained from the requirement of preservation of the properties of eigenmodes for finite @xmath76 as far as they can be identified . the continuity of the normal to the wall component of the total ( through the whole hypersurface of the wall ) charged current offers a reliable guiding principle for identification of the matching conditions . continuity of the total current means that the surface terms do not appear under integration by parts in the action , @xmath77=0 , \\ \nonumber & & j_\mu(x_1)=\int d^3x \phi^\dagger(x)d_\mu\phi(x ) , \\ \nonumber & & d^3x = dx_2dx_3dx_4.\end{aligned}\ ] ] moreover , this requirement restricts the form of the eigenfunctions in such a way that the surface terms associated with the gauge dependent delta - function singularuties in @xmath78 , eq.([sing_gauge ] ) vanish as well . .the coordinate @xmath79 is given in units of @xmath80 . in the limit of the infinitely thin domain wall ( @xmath81 ) the derivatives develop the delta - function singularities at the location of the wall . , width=283 ] let us consider the eigenvalue problem @xmath82 for the functions square integrable in @xmath12 and satisfying the condition ( [ current_cont_sc ] ) . for all @xmath83 the operator @xmath84 takes the form @xmath85 where plus corresponds to the anti - self - dual configuration and minus is for the self - dual one . by inspection one can see that the eigenfunctions satisfy the relation @xmath86 where @xmath87 denotes the duality of the background field for a given @xmath79 . respectively , the square integrable solutions are @xmath88\right ) \nonumber\\ \chi_k(x_3,x_4)&=&\int dp_4 g(p_4)e^{ip_4x_4-\frac{1}{2}|\breve n|b(x_3+p_4/|\breve n|b)^2 } \nonumber\\ & \times & h_l\left(\sqrt{|\breve n|b}\left[x_3+\frac{p_4}{|\breve n|b}\right]\right ) , \nonumber\end{aligned}\ ] ] where @xmath89 are the hermite polynomials . the eigenvalues are @xmath90 the amplitudes @xmath91 and @xmath92 have to provide square integrability of the eigenfunctions in @xmath79 and @xmath93 . in order to satisfy condition ( [ current_cont_sc ] ) one has to restrict the amplitude @xmath91 additionally . the integral current through the domain wall is continuous if both @xmath94 and @xmath95 are odd or even functions simultaneously under the combined change @xmath96 and @xmath97 @xmath98 this property also guarantees the absence of the gauge specific contribution to the action related to the derivative of @xmath99 in eqs.([sing_gauge],[surf_gauge ] ) . a combination of ( [ cont_cond_sc ] ) and ( [ conditions ] ) obviously leads to the relation @xmath100 where @xmath87 denotes the duality of the background field for a given @xmath79 . this identity allows one to show that the eigenfunctions @xmath101 form a complete orthogonal set in the space of square integrable functions which are even with respect to simultaneous reflection @xmath102 and @xmath97 . the eigenfunctions are of the bound state type with the purely discrete spectrum . field fluctuations of this type can be seen as confined . it should be noted that the eigenvalues coincide with those for the purely homogeneous ( anti-)self - dual abelian field . in this sense , the domain wall defect does not destroy dynamical confinement of color charged fields . the eigenfunctions are restricted by the correlated evenness condition ( [ cont_cond_sc ] ) , while in the case of the homogeneous field the properties of the amplitude @xmath91 and the polynomial @xmath95 are mutually independent . let us now consider the eigenvalue problem on the domain wall , i.e. for the region @xmath62 . on the wall the chromomagnetic and chromoelectric fields are orthogonal to each other ( see fig.[fig : single_kink ] ) . in conformity with ( [ current_cont_sc ] ) the absence of the charged current off the infinitely thin domain wall requires @xmath103 and the eigenvalue problem on the wall takes the form @xmath104\phi=\lambda\phi \nonumber\end{aligned}\ ] ] with the solution @xmath105 , \nonumber\\ \lambda_k(p^2_2,p_4 ^ 2)&=&\sqrt{2}|\breve n|b(2k+1)+\frac{p_4 ^ 2}{2}+\frac{p_2 ^ 2}{2 } , \nonumber\\ & & k=0,1,2,\dots \nonumber\end{aligned}\ ] ] the spectrum of the eigenmodes on the wall is continuous , it depends on the momentum @xmath106 longitudinal to the chromomagnetic field and euclidean energy @xmath107 , the corresponding eigenfunctions are oscillating in @xmath70 and @xmath93 . in the direction @xmath71 transverse to the chromomagnetic field the eigenfunctions are bounded and the eigenvalues display the landau level structure . the continuation @xmath108 leads to the dispersion relation @xmath109 this can be treated as the lack of confinement - the color charged quasiparticles with masses @xmath110 and momentum @xmath111 parallel to the chromomagnetic field @xmath112 can be excited on the wall . the case of the planar domain wall configuration ( two infinite parts of the space - time separated by a three - dimentional hypersurface like in fig.[fig : single_kink ] ) is rather artificial . its weight in the whole ensemble of the gluon field configurations with the constant scalar condensate @xmath113 and the lumpy structured distribution of the topological charge density @xmath114 is negligible . the entropy - energy balance implies that the typical configuration should be highly disordered ( see fig.[fig : kink_network ] ) . moreover , in the case of the planar domain wall the eigenvalue problem for the square integrable vector gauge fields @xmath115q_\nu=\lambda q_\mu\end{aligned}\ ] ] leads to the negative eigenvalues and corresponding tachyonic modes on the wall where @xmath116 . this is a well - known instability of the nielsen - olesen type @xcite . the presence of the tachyonic mode is due to the three infinite dimensions of the planar domain wall hypersurface . one can expect that finite size of boundaries between lumps in the typical kink network configuration , fig.[fig : kink_network ] , removes the tachyonic modes . this is manifestly exemplified in the next section where the color charged field eigenvalues and modes are studied for thick cylindrical domain wall junction . the relatively stable defect of this type can occur in the ensemble of confining gluon fields due to the influence of the strong electromagnetic fields on the qcd vacuum structure . it has been observed that the strong electromagnetic fields generated in relativistic heavy ion collisions can play the role of a trigger for deconfinement @xcite . the mechanism discussed in @xcite is as follows . the electric @xmath117 and magnetic @xmath118 fields are practically orthogonal to each other @xcite : @xmath119 . for this configuration of the external electromagnetic field the one - loop quark contribution to the qcd effective potential for the homogeneous abelian gluon fields is minimal for the chromoelectric and chromomagnetic fields directed along the electric and magnetic fields respectively . the orthogonal chromo - fields are not confining : color charged quasiparticles can move along the chromomagnetic field . it has been noted also that this mechanism assumes the strong azimuthal anisotropy in momentum distribution of color charged quasiparticles . deconfined quarks as well as gluons will move preferably along the direction of the magnetic field but this will happen due to the gluon field configuration even after switching the electromagnetic field off . a detailed and systematic analytical one - loop calculation of the qcd effective potential for the pure chromomagnetic field was performed recently in @xcite and confirmed the result that the chromomagnetic field prefers to be parallel ( or anti - parallel ) to the external magnetic field . another important source of verification of the basic observations of paper @xcite is due to the recent lattice qcd studies of the response of the qcd vacuum to external electromagnetic fields @xcite . in particular , in qualitative agreement with @xcite lattice qcd study @xcite has demonstrated that in the presence of external magnetic field the gluonic action develops an anisotropy : the chromomagnetic field parallel to the external field is enhanced , while the chromo - electric field in this direction is suppressed . the results of @xcite indicated that the magnetic field can affect the azimuthal structure of the expansion of the system during heavy ion collisions . . blue and red regions represent self - dual and anti - self - dual lumps . confinement is lost in the green region where @xmath120 . the scalar condensate density @xmath121 is nonzero and homogeneous everywhere.,title="fig:",width=143 ] . blue and red regions represent self - dual and anti - self - dual lumps . confinement is lost in the green region where @xmath120 . the scalar condensate density @xmath121 is nonzero and homogeneous everywhere.,title="fig:",width=143 ] within the context of the confining domain wall network these observations mean that a flash of the strong electromagnetic field during heavy ion collisions produces a kind of defect in the form of the thick domain wall junction in the confining gluon background exactly in the region where collision occurs ( see fig.[fig : chromo_bag ] ) . the electromagnetic flash can act as one of the preconditions for conversion of the high energy density and baryon density to the thermodynamics of color charged degrees of freedom . + since topological charge density is zero in the interior of the trap ( @xmath120 ) there exists a specific reference frame where one can use the pure chromomagnetic field for description of the gluon background inside the trap . for simplicity we take cylindrical geometry of the trap and study the properties of scalar and vector ( gluon ) color charged field eigenmodes . extension of the present consideration to more realistic form of the trap is straightforward . consider the eigenvalue problem for the massless scalar field @xmath122 @xmath123 in the cylindrical region @xmath124 with the homogeneous dirichlet condition at the boundary @xmath125 here @xmath126 stays for adjoiunt representaion of the homogeneous chromomagnetic field @xmath127 with the vector potential taken in the symmetric gauge @xmath128 the eigenvalues of the matrix @xmath64 are @xmath129 . \label{vadj}\end{aligned}\ ] ] for any value of the angle @xmath130 there are two zero eigenvalues @xmath131 . two additional zero elements occur in @xmath132 if the angle takes values @xmath31 ( see eq . ) minimizing the effective potential and corresponding to the boundaries of the weyl chambers . by inspection one can check that nonzero eigenvalues @xmath132 take values @xmath133 with @xmath134 . below we use notation @xmath135 for example , if @xmath136 then the nonzero values of @xmath137 correspond to @xmath138 and @xmath139 it has to be noted that the effective lagrangian leads to the kink configuration ( for details see @xcite ) @xmath140+\frac{\pi k}{3 } , \ k=0,\dots,5,\end{aligned}\ ] ] interpolating between boundaries @xmath31 and @xmath141 of the @xmath142-th weyl chamber . superposition of these `` color '' domain walls can be arranged in a complete analogy with the `` duality '' domain walls . the only new feature of the `` color '' domains is that there are six different types interrelated by the weyl reflections instead of two types as in the case of duality domains . solution of the problem is straightforward . we give it below just for completeness . it is convenient to introduce dimensionless variables using the strength of the chromomagnetic field as a basic scale . below all quantities are assumed to be measured in terms of this scale , for instance @xmath143 after diagonalization with respect to color indices and transformation to the cylindrical coordinates eq . takes the form @xmath144\phi^a=\lambda^2\phi^a,\ \ \\end{aligned}\ ] ] where it has been used that @xmath145 the variables in eq . are separated by substitution @xmath146 periodicity of the solution in angle @xmath147 $ ] requires integer values of parameter @xmath148 . the radial part @xmath149 should satisfy equation @xmath150\phi=\mu^2\phi,\ ] ] where @xmath76 is related to the original eigenvalue @xmath151 , @xmath152 by means of the substitution @xmath153 one arrives at the kummer equation ( @xmath154 ) @xmath155\chi=0.\ ] ] the complete solution can be chosen in the form @xmath156 where @xmath157 is kummer function . general solution of equation takes the form @xmath158\ ] ] the first term is regular at @xmath159 provided @xmath160 while the second one is well - defined for @xmath161 . therefore , the solution regular inside the cylinder is @xmath162 where the color index @xmath163 has been explicitly indicated . the color matrix elements @xmath164 can be negative . in this case one has to apply kummer transformation @xcite @xmath165 dirichlet boundary condition defines the infinite discrete set of eigenvalues as the solutions @xmath166 ( @xmath167 ) of the equations @xmath168 if @xmath166 satisfies equation , than @xmath169 is a solution of . finally the complete orthogonal set of eigenfunctions for the problem and reads @xmath170 where functions @xmath171 are defined by with @xmath172 solving the boundary condition . unlike landau levels in the infinite space the eigenvalues @xmath173 are not equidistant in @xmath142 and non - degenerate in @xmath148 as it is illustrated in fig.[fig : eigen_set1 ] ) . the dependence of several low - lying eigenvalues @xmath173 on the dimensionless size parameter @xmath174 is shown in fig.[fig : eigen_flow ] . for pure chromomagnetic field the adjoint representation vector field eq . takes the form @xmath175q_\nu=\lambda q_\mu,\end{aligned}\ ] ] and the boundary conditions are @xmath176 in terms of the eigenvectors @xmath177 of matrices @xmath178 and @xmath64 eqs . and take the form @xmath179^a\breve{q}^{a}_\mu = \lambda_{a\mu}\breve{q}^{a}_\mu , \\\nonumber \breve v \breve{q}_{\mu}(x)=0 , \ x\in \partial\mathcal{t}.\end{aligned}\ ] ] omitting obvious well - known details we just note that equation describes sixteen charged with respect to @xmath64 spin - color polarizations of the gluon fluctuations with @xmath180 and @xmath181 as well as sixteen `` color neutral '' with respect to @xmath182 modes @xmath183 neutral mode @xmath184 is a zero mode of @xmath64 , and it is insensitive to the boundary condition . we shall briefly discuss the possible role of the neutral modes in the last section . equations for the color charged modes have the same form as the scalar field equation in the previous subsection . the only essential difference is that the eigenvalues @xmath185 for nonzero @xmath137 have an addition @xmath186 to the eigenvalues @xmath187 of the scalar case : @xmath188 where @xmath187 are the same as in the scalar case . if we were considering the square integrable solutions in @xmath12 then the lowest mode @xmath189 with @xmath190 would be tachyonic . in the finite trap the lowest eigenvalue is @xmath191 the dependence of @xmath192 on dimensionless size parameter @xmath174 is strongly nonlinear . few lowest eigenvalues @xmath173 as functions of @xmath174 are shown in fig . [ fig : eigen_flow ] . one concludes that if the dimensionless size @xmath174 of the trap is sufficiently small @xmath193 then there are no unstable tachyonic modes in the spectrum of color charged vector fields . to estimate the critical size one may use the mean phenomenological value of the gluon condensate ( gauge coupling constant @xmath194 is included into the field strength tensor ) @xmath195 equation leads to the critical radius @xmath196 thus the tachyonic mode is absent if the diameter of the cylindrical trap is less or equal to @xmath197 . for the scalar field problem , @xmath198 and @xmath199 , for @xmath200 . eigenvalues are denoted by asterisks in the case of positive @xmath201 and by circles in the case of negative @xmath201 . , width=283 ] as functions of @xmath174 . the critical radius @xmath202 corresponds to @xmath203 . for large @xmath174 eigenvalues approach correct landau levels , the degeneracy in @xmath148 is restored.,width=283 ] in this subsection we address the eigenvalue problem for dirac operator in the cylindrical region in the presence of chromomagnetic background field ( [ purechrmag ] ) @xmath204 euclidean dirac matrices are taken in the anti - hermitian representation . the angle @xmath130 is assumed to take one of the vacuum values @xmath31 , and according to eq . the following forms of the matrix @xmath205 can occur @xmath206 below we use notation @xmath207 the boundary conditions are @xmath208 where @xmath209 is a unit vector normal to the cylinder surface @xmath210 , see eq . . these are simply the bag boundary conditions . this choice appears to be rather natural . indeed , inside the thick domain wall junction one expects an existence of the color charged quasiparticles ( quarks ) being the carriers of the color current , but outside the junction gluon configurations are confining ( see fig . [ fig : chromo_bag ] ) ) and the current has to vanish at the boundary . unlike the adjoint representation of color matrix the matrix @xmath205 in fundamental representation has no zero eigenvalues for any value of the angle @xmath130 corresponding to the boundaries of weyl chambers , see eq . . boundary condition restricts all three color components of the quark field . substitution @xmath211 leads to the equation @xmath212 where it has been used that in the pure chromomagnetic field @xmath213 equation is essentially the same as . its solution in cylindrical coordinates ( @xmath214-periodic in @xmath215 and regular at @xmath159 ) is given by four independent components @xmath216 ( @xmath217 , @xmath218 ) : @xmath219 with @xmath220 for the case @xmath160 and @xmath221 for @xmath222 . here @xmath223 denotes the sign of the quark spin projection on the direction of chromomagnetic field , and therefore @xmath224 the variable @xmath76 is related to the dirac eigenvalues as @xmath225 finally the dirac operator eigenfunction @xmath226 can be obtained by means of relation with @xmath227 where @xmath228 \\ d_1-id_2=e^{-i\vartheta}\left(\frac{\partial}{\partial r}-\frac{i}{r}\frac{\partial}{\partial \vartheta}-\frac12\hat{u}r\right).\end{gathered}\ ] ] four solutions are for @xmath160 @xmath229 @xmath230 @xmath231 @xmath232 and for @xmath222 @xmath233 @xmath234 @xmath235 @xmath236 all four spinors are eigenfunctions , @xmath237 of the total momentum projection operator onto @xmath71 . @xmath238 only two of these solutions at given @xmath148 are linearly independent . we select @xmath239 as a general solution to equation for the reason that @xmath240 and @xmath241 remain linearly independent in the limit @xmath242 . the limit will be used in the next section for the solving the dirac equation in minkowski space - time . boundary condition with @xmath243 leads to the equation defining the values of the parameter @xmath76 as well as the ratio of @xmath244 and @xmath245 . for @xmath160 one gets @xmath246 this system has a nontrivial solution for @xmath244 and @xmath245 if the determinant of the matrix composed of the coefficients in front of them is equal to zero @xmath247 ^ 2=\left[\frac{\mu}{2(1+l)}\phi^{\uparrow\downarrow } _ { l+1}(r)\right]^2.\end{gathered}\ ] ] this equation defines the spectrum of @xmath248 . states with definite spin orientation with respect to the chromomagnetic field are mixed in the boundary condition , and the spin projection onto the direction of the field is not a good quantum number unlike the projection of the total momentum @xmath249 as it is taken into account in fig.[fig : q_eigenv ] . as is illustrated in fig.[fig : q_eigenv ] there is a discrete set of solutions @xmath250 which depend also on the color orientation @xmath251 ( @xmath252 with @xmath218 , @xmath253 , @xmath254 ) . as a rule one can omit the color index @xmath255 assuming that @xmath256 is a diagonal color matrix for any @xmath257 . the values @xmath258 has to be used to find the relation between @xmath244 and @xmath245 @xmath259 here @xmath256 is taken to be positive , and @xmath260 takes both positive and negative values . equation has been used in combination with observation ( by inspection ) that the sign of the ratio @xmath261 and @xmath262 depends on @xmath253 as it is indicated in irrespectively to @xmath148 and color orientation . analogous consideration for the case @xmath222 leads to the equation for @xmath76 @xmath263 ^ 2=\left[\frac{\mu}{2l}\phi^{\uparrow\uparrow } _ { l}(r)\right]^2,\end{gathered}\ ] ] and for the ratio of coefficients @xmath264 the orthogonal normalized set of solutions has the form for @xmath160 @xmath265 and for @xmath222 @xmath266 the spinors @xmath267 and @xmath268 correspond to the positive and negative eigenvalues @xmath260 in respectively , they are eigenfunctions of @xmath269 with @xmath270 . normalization constants are @xmath271 \nonumber\\ b^{-2}_{jlk}(r)&=&\int_0^r drr\left[\left(\frac{\mu_{jlk}}{2l}\phi_{l+1}^{\uparrow\uparrow}(r)\right)^2+\left(\phi_{l+1}^{\uparrow\downarrow}(r)\right)^2\right ] \label{normconst}\end{aligned}\ ] ] the same procedure applied to the equation @xmath272 leads to the solutions for @xmath160 @xmath273 and for @xmath222 @xmath274 solving eq . for @xmath200 . here @xmath270 is the projection of the total momentum on the direction of the chromomagnetic field . eigenvalues are denoted by asterisks in the case of positive @xmath275 and by circles in the case of negative @xmath275.,width=302 ] to get insight into the physical treatment of above - considered euclidean eigenmodes one has to solve the minkowski space klein - gordon and dirac equations in the presence of chromomagnetic field inside the cylinder with the bag - like boundary conditions . solutions describe the elementary quasiparticle excitations inside the thick cylindrical domain wall junction . quite detailed analysis of the notion of quasiparticles in relativistic quantum field theory can be found in @xcite . unlike the fundamental elementary and composite particles the characteristic properties of quasiparticles ( for instance the specific form of the dispersion relation ) need not be necessarily lorentz invariant or even gauge invariant . the overall statement of the problem under consideration necessarily assumes that space direction along the chromomagnetic field is singled out by underlining experimental setup as it coincides with the direction of the strong magnetic field generated for short time in heavy ion collision . in generic relativistic frame both chromoelectric and chromomagnetic fields are present inside the domain wall junction . however since the topological charge density vanishes in the region ( see fig.[fig : chromo_bag ] ) there exists specific frame where chromoelectric field is absent . this frame is the most convenient for our purposes . in minkowski space - time the problem and turns to the wave equation @xmath276 for color charged adjoint spin zero field inside a cylindrical wave guide . as it follows from and futher discussion the charged components of the adjoint field of the color matrix @xmath64 comes in complex conjugate pairs . for instance if @xmath277 then there are two pairs @xmath278 and @xmath279 . thus @xmath280 is a complex scalar field , the corresponding solution of satisfying boundary condition takes the form @xmath281e^{il\vartheta}\phi_{alk}(r ) , \label{seigenf_mink } \\ & & \phi^{a\dagger}(x)=\sum_{lk}\int\limits_{-\infty}^{+\infty}\frac{dp_3}{2\pi}\frac{1}{\sqrt{2\omega_{alk } } } \left[b^{+}_{akl}(p_3)e^{-ix_0\omega_{akl}+ip_3x_3 } + a_{akl}(p_3)e^{ix_0\omega_{akl}-ip_3x_3}\right]e^{-il\vartheta}\phi_{alk}(r ) , \nonumber \\ & & p_0 ^ 2=p_3 ^ 2+\mu_{akl}^2 , \nonumber\\ & & p_0=\pm\omega_{akl}(p_3 ) , \ \omega_{akl}=\sqrt{p_3 ^ 2+\mu_{akl}^2 } , \label{mass_mink } \\ & & k=0,1,\dots,\infty , \ \ l\in z , \nonumber\end{aligned}\ ] ] with @xmath282 defined in but here it is assumed to be normalized @xmath283 equation can be treated as the dispersion relation between energy @xmath284 and momentum @xmath285 for the quasiparticles with masses @xmath286 . these quasiparticles are extended in @xmath79 and @xmath70 directions and are classified by the quantum numbers @xmath257 . the orthogonality , normalization and completeness of the set of functions @xmath287 guarantees the standard canonical commutation relations for the field @xmath280 and its canonically conjugated momentum if @xmath288 , @xmath289 , @xmath290 and @xmath291 are assumed to satisfy the standard commutation relations for creation and annihilation operators . the fock space of states for the quasiparticles with masses @xmath286 can be constructed by means of the standard qft methods . this treatment provides one with a suitable terminology and formalism for discussion of the confining properties of various gluon field configurations in the context of qft : unlike the chromomagnetic field the ( anti-)self - dual fields characteristic for the bulk of domain network configuration ( see the lhs plot in fig . [ fig : kink_network ] ) lead to purely discrete spectrum of eigenmodes in euclidean space and do not possess any quasiparticle treatment in terms of dispersion relation between energy and momentum for elementary color charged excitations . if there is a reason for long - lived defect in the form of thick domain wall junction then its boundary defines a shape and a size for the space region which can be populated by color charged quasiparticles . the vector adjoint field can be elaborated in the similar to the scalar case way . a modification relates just to the inclusion of polarization vectors . as it has already been mentioned the most important feature is the absence of tachyonic mode of the vector color charged field if @xmath292 . disappearance of the tachyonic mode for subcritical size of the trap is one of the most important observations of this paper . neither the background field nor the boundary condition involve the time coordinate . the solution of the dirac equation @xmath293 satisfying condition can be obtained from euclidean solutions ( unnormalized solutions have to be used ) by the analytical continuation @xmath294 , @xmath295 and the requirement @xmath296 , which leads to the energy - momentum relation for the solutions with definite @xmath249 , @xmath142 and color @xmath255 @xmath297 finally the solution of the dirac equation takes the form @xmath298 , \\ \bar\psi^{j}(x)=\sum_{lk}\int\limits_{-\infty}^{+\infty}\frac{dp_3}{2\pi}\frac{1}{\sqrt{2\omega_{jlk } } } \left[b^{\dagger}_{jlk}(p_3)\bar\chi_{jlk}(p_3|r,\vartheta ) e^{-ix_0\omega_{jlk}+ix_3p_3}+a_{jlk}(p_3){\bar\upsilon } _ { jlk}(p_3|r,\vartheta)e^{ix_0\omega_{jlk}-ix_3p_3}\right].\end{aligned}\ ] ] here the pair of spinors for positive @xmath299 and negative @xmath300 energy solutions are @xmath301 for @xmath302 and @xmath303 for @xmath222 . the spinors are normalized as @xmath304 the dirac conjugated spinors are @xmath305 as usual . the fock space can be constructed by means of the creation and annihilation operators @xmath306 satisfying the standard anticommutation relations . the one - particle state is characterized by a color orientation @xmath255 , momentum @xmath285 , projection @xmath307 of the total angular momentum and the energy @xmath308 . since the boundary condition mixes the states with spin parallel and anti - parallel to the chromomagnetic field the spin projection is not a good quantum number unlike the half - integer valued projection of the total angular momentum @xmath249 . an ensemble of confining gluon configurations has been constructed explicitly as a domain wall networks representing the almost everywhere homogeneous abelian ( anti-)self - dual gluon fields . confinement is understood here as the absence of the color charged wave - like elementary excitations . the dynamical quark confinement occurs in the ( four - dimensional ) bulk of the domain wall network . inside the ( three - dimensional ) domain walls topological charge density vanishes and the color charged quasiparticles can be excited . under extreme conditions , in particular under the influence of the strong electromagnetic field specific for relativistic heavy ion collisions , a relatively stable defect in the confining ensemble , a thick domain wall junction , can be formed . though the scalar gluon condensate is nonzero everywhere @xmath309 , the region of defect is characterized by the vanishing topological charge density @xmath310=0 unlike the rest of the space , which indicates the lack of confinement in the junction . the quark field excitations inside the junction are represented by the color charged quasiparticles . the spectrum of gluon excitations besides the trapped color charged modes contains also the color neutral with respect to the background field modes . almost obvious but important observation is that there exists a critical size @xmath311 of the junction beyond which the tachyonic gluon modes emerge in the excitation spectrum and destabilize the defect . the critical size can be related to the value of the gluon condensate @xmath1 and in the case of the considered in the paper cylindrical trap @xmath312fm for the standard value of the condensate , see . the specific value of the critical size depends on the geometry of the trap but its very existence and its commensurability with a distance of order of @xmath313fm is a generic feature . this observation underlines the generic necessity of accounting for the essentially finite size of the space - time region in which deconfinement may occur . the reason is that thermodynamic limit does not exist as the system under consideration disappears as soon as the typical size of the space volume exceeds the critical value . excess of the internal pressure of the trap filled by many charged quasiparticles leads to its expansion and breakdown of stability followed by its disintegration to many smaller traps ( or bags ) , which is reminiscent of the heterophase fluctuations studied in @xcite as well as the dynamics and statistical mechnaics of bags with a surface tension @xcite . the dynamics of the color charged quasiparticles as it is described above is strictly one - dimentional in space . this feature can be a source of the azimuthal asymmetries in heavy ion collisions , similarly to the approach of paper @xcite upto a substitution of the magnetic field by the abelian chromomagnetic field ( see also @xcite ) . however it should be noted that the one - dimensional dynamics is a property of the zero - th order approximation based on the quadratic part of the action . taking into account interactions between the quasiparticles according to the interaction terms in the action should certainly dither the direction of the quasiparticle momenta , leaving just some degree of azimuthal asymmetry . we acknowledge fruitful discussions with v.toneev , s. molodtsov , j. pawlowski , m.ilgenfritz , a.dorokhov , k.bugaev s.vinitsky , g.efimov , v.yukalov , a.efremov , a.titov . 100 e. -m . ilgenfritz , k. koller , y. koma , g. schierholz , t. streuer and v. weinberg , phys . d * 76 * , 034506 ( 2007 ) [ arxiv:0705.0018 [ hep - lat ] ] . p. j. moran and d. b. leinweber , arxiv:0805.4246 [ hep - lat ] . p. j. moran and d. b. leinweber pos * lat2007 * ( 2007 ) 383 [ arxiv:0710.2380 [ hep - lat ] ] . de forcrand , a. kurkela and a. vuorinen phys . d * 77 * ( 2008 ) 125014 . de forcrand aip conf . ( 2007 ) 29 [ arxiv : hep - lat/0611034 ] . a. r. zhitnitsky , arxiv:1301.7072 [ hep - ph ] . p. minkowski , phys . b 76 ( 1978 ) 439 . h. pagels , and e. tomboulis , nucl . phys . b * 143 * ( 1978 ) 485 . p. minkowski , nucl . phys . b**177 * * ( 1981 ) 203 . h. leutwyler , nucl . b * 179 * ( 1981 ) 129 ; _ ibid _ phys . b96 ( 1980 ) 154 . kalloniatis and s.n . nedelko , phys . d * 64 * ( 2001 ) 114025 ; l. d. faddeev , [ arxiv:0911.1013 [ math - ph ] ] . a.c . kalloniatis and s.n . nedelko , phys . d * 73 * ( 2006 ) 034006 . galilo and s.n . nedelko , phys . d84 ( 2011 ) 094017 . galilo and s.n . nedelko , phys . * 8 * ( 2011 ) 67 [ arxiv : hep - ph/1006.0248v2 ] . h. d. trottier and r. m. woloshyn , phys . * 70 * ( 1993 ) 2053 . a. eichhorn , h. gies and j. m. pawlowski , phys . d * 83 * , 045014 ( 2011 ) [ erratum - ibid . d * 83 * , 069903 ( 2011 ) ] [ arxiv:1010.2153 [ hep - ph ] ] . d. p. george , a. ram , j. e. thompson and r. r. volkas , phys . d * 87 * , 105009 ( 2013 ) [ arxiv:1203.1048 [ hep - th ] ] . efimov , and s.n . nedelko , phys . d * 51 * ( 1995 ) 176 ; j. .v . burdanov , g. v. efimov , s. n. nedelko , s. a. solunin , phys . d * 54 * ( 1996 ) 4483 . kalloniatis and s.n . nedelko , phys . d * 69 * ( 2004 ) 074029 ; _ erratum - ibid . d * 70 * ( 2004 ) 119903 ; _ ibid _ , phys . d * 71 * ( 2005 ) 054002 ; t. vachaspati , kinks and domain walls , cambridge university press , 2006 . m. delia , m. mariti and f. negro , phys . * 110 * , 082002 ( 2013 ) [ arxiv:1209.0722 [ hep - lat ] ] . g. s. bali , f. bruckmann , g. endrodi , f. gruber and a. schaefer , jhep * 1304 * , 130 ( 2013 ) [ arxiv:1303.1328 [ hep - lat ] ] . d. e. kharzeev , l. d. mclerran , and h. j. warringa , nucl . phys . a * 803 * * 227 * ( 2008 ) v. voronyuk , v. d. toneev , w. cassing , e. l. bratkovskaya , v. p. konchakovski and s. a. voloshin , phys . rev . c * 83 * , 054911 ( 2011 ) [ arxiv:1103.4239 [ nucl - th ] ] . v. skokov , a. y. illarionov and v. toneev , int . j. mod a * 24 * ( 2009 ) 5925 [ arxiv:0907.1396 [ nucl - th ] ] . k. tuchin , adv . high energy phys . * 2013 * , 490495 ( 2013 ) [ arxiv:1301.0099 ] . k. fukushima and y. hidaka , phys . lett . * 110 * , 031601 ( 2013 ) [ arxiv:1209.1319 [ hep - ph ] ] . cho , phys . rev . lett . 44 ( 1980 ) 1115 . y. m. cho , j. h. kim and d. g. pak mod . lett . a 21 ( 2006 ) 2789 . shabanov , j. math . 43 ( 2002 ) 4127 [ hep - th/0202146 ] . s. v. shabanov phys . 326 ( 2000 ) 1 [ arxiv : hep - th/0002043 ] ; s. v. shabanov and j. r. klauder phys . lett . b 456 ( 1999 ) 38 . _ l. v. prokhorov _ yad . fiz . , 35 ( 1982 ) 229 . l. d. faddeev , a. j. niemi // b. 776 . 2007 ; _ ibid _ , phys . b 449 ( 1999 ) 214 . kei - ichi kondo , toru shinohara , takeharu murakami , prog . ( 2008 ) 1 [ arxiv:0803.0176 [ hep - th ] ] . n. k. nielsen and p. olesen , nucl . b * 144 * , 376 ( 1978 ) . s. ozaki , arxiv:1311.3137 [ hep - ph ] . g. s. bali , f. bruckmann , g. endrodi and a. schafer , arxiv:1311.2559 [ hep - lat ] . c. bonati , m. delia , m. mariti , f. negro and f. sanfilippo , arxiv:1312.5070 [ hep - lat ] . m. abramowitz , i.a . stegun , handbook of mathematical functions with formulas , graphs , and mathematical tables , dover ( 1964 ) , new york . d. arteaga , annals phys . * 324 * , 920 ( 2009 ) [ arxiv:0801.4324 [ hep - ph ] ] . v. i. yukalov and e. p. yukalova , pos ishepp * 2012 * , 046 ( 2012 ) [ arxiv:1301.6910 [ hep - ph ] ] ; v.i . yukalov , phys . rep . * 208 * ( 1991 ) 395 ; v.i . yukalov , int . j. mod . b * 17 * , ( 2003 ) 2333 . k. a. bugaev , phys . c * 76 * , 014903 ( 2007 ) [ hep - ph/0703222 ] ; k. a. bugaev , v. k. petrov and g. m. zinovjev , phys . c * 79 * , 054913 ( 2009 ) [ arxiv:0807.2391 [ hep - ph ] ] ; _ ibid _ phys . atom . 76 ( 2013 ) 341 [ arxiv:0904.4420 [ hep - ph ] ] .

we review and extend the formalism introduced by peliti , that maps a markov process to a path - integral representation . after developing the mapping , we apply it to some illustrative examples : the simple decay process , the birth - and - death process , and the malthus - verhulst process . in the first two cases we show how to obtain the exact probability generating function using the path integral . we show how to implement a diagrammatic perturbation theory for processes that do not admit an exact solution . analysis of a set of coupled malthus - verhulst processes on a lattice leads , in the continuum limit , to a field theory for directed percolation and allied models . 1em email : [email protected]

cond-mat0205321

it is often noted that nonequilibrium statistical mechanics lacks the comprehensive formalism of ensembles that has proved so useful in equilibrium . the reason is that equilibrium statistical mechanics treats stationary states for a special subclass of systems , that possess detailed balance . this allows one to bypass the dynamics , and study stationary properties directly . for systems out of equilibrium , we generally do not have such a shortcut , and must deal with the full dynamical problem , even if our goal is only to obtain stationary properties . in the case of stochastic systems with a discrete state space , the fundamental description is given by the master equation , which governs the evolution of the probability distribution . this class of problems includes a wide range of systems of current interest , that exhibit phase transitions or scale invariance far from equilibrium : driven lattice gases , birth - and - death processes such as directed percolation or the contact process , sandpile models , and interface growth models . one of the more powerful tools for studying stochastic models is a formalism that maps the process to a path - integral representation @xcite . the mapping generates an _ effective action _ that can be studied using the tools of equilibrium statistical physics , for example , the renormalization group . several methods for mapping a stochastic process to an equilibrium - like action have been proposed @xcite . in these notes we review the method developed by peliti @xcite , and apply it to some simple stochastic processes . this method has several advantages . with it , one can map any birth - and - death type process to a path - integral representation without ambiguity . in particular , the step of writing a langevin equation , and of postulating noise autocorrelations , does not arise in this formalism . thus it provides a direct path from the model of interest to an effective action , and ( in the continuum limit ) , to its field theory , without the uncertainties that often attend the specification of the noise term @xcite . a second advantage , which we explore in some examples , is that it leads to a systematic perturbative analysis for markov processes . the principal aim of this article is to acquaint the reader with the formalism and provide a set of worked examples whose mastery will allow one to apply the method to problems at the frontier of research . while peliti s article @xcite provides an excellent exposition of the mapping , we include , for completeness , a derivation of the central formulas . our development of the perturbation theory differs somewhat from peliti s . most of the applications discussed are also new . the balance of this article is structured as follows . in sec . ii we derive the path - integral representation , starting from the master equation . sec . iii presents an application to the simple decay process , and expressions for two - time joint probabilities . in sec . iv we begin our discussion of diagrammatic perturbation theory for the probability generating function , which is illustrated with a pedagogical example . this is extended in sec . v where we analyze the birth - and - death process using perturbation theory . in sec . vi a perturbation expansion for moments of the distribution is developed , which turns out to be much simpler than that for the full generating function . this method is applied to the malthus - verhulst process in sec . vii . in sec . viii we illustrate another application of the formalism , showing how the path - integral description for a lattice of coupled malthus - verhulst processes leads , in the continuum limit , to a field theory for directed percolation . sec . ix presents a brief summary . in this section we recapitulate peliti s derivation of the path integral mapping . we consider markov processes in continuous time , and with a discrete state space @xmath1 . ( we may think of @xmath2 as the size of a certain population . ) the probability @xmath3 of state @xmath2 at time @xmath4 is governed by the master equation @xcite : @xmath5 where @xmath6 is the rate for transitions from @xmath2 to @xmath7 . ( we study stationary _ stationary _ markov processes , i.e. , time - independent transition rates . ) we now associate a vector @xmath8 in a hilbert space with each state @xmath2 , @xcite and for convenience define the inner product so : @xmath9 the identity may then be written as @xmath10 ( all sums run from zero to infinity unless otherwise specified . note that while we make use of many pieces of notation familiar from quantum mechanics , we are dealing with c - number functions . ) a probability distribution @xmath11 ( @xmath12 , @xmath13 ) , may be represented as a linear combination of basis states : @xmath14 the hilbert space formalism is useful because it provides a simple way to express the evolution in terms of creation ( @xmath15 ) and annihilation ( @xmath16 ) operators , which we define via : @xmath17 these relations imply that @xmath18 = 1 $ ] . here again we note a fundamental difference from quantum mechanics : expected values are _ linear _ , not bilinear , in @xmath19 . the mean population size , for instance , is given by @xmath20 = \langle \;| a |\phi(t ) \rangle \ ; , \label{expect}\ ] ] where @xmath21 is the the projection onto all possible states . central to our analysis will be the _ probability generating function _ ( pgf ) , @xmath22 we denote the pgf corresponding to state @xmath23 as @xmath24 . ( note that @xmath25 by normalization . ) next consider the inner product between states @xmath23 and @xmath26 : @xmath27 we can write this in terms of the corresponding pgfs if we note the identity @xmath28 which is readily proved , integrating by parts . ( unless otherwise specified , all integrals are over the real axis . ) then we have @xmath29 where we used the integral representation of the @xmath30 function . for birth - and - death processes , it is always possible to write the master equation in terms of an evolution operator @xmath31 composed of creation and annihilation operators ( specific examples are considered below ) . the master equation then takes the form @xmath32 and has the formal solution @xmath33 this evolution has its analog in the space of probability generating functions ; it is for the analog of the operator @xmath34 in the pgf representation that we shall develop a path - integral expression . to do this we define , for any operator @xmath35 in the hilbert space , a function called its kernel : @xmath36 where @xmath37 are the matrix elements of @xmath35 . suppose @xmath38 . the pgf corresponding to @xmath26 is @xmath39 we shall also require an expression for the kernel of a product of a pair of operators , @xmath35 and @xmath40 : @xmath41_{m , n } \frac{\zeta^n}{n ! } \\ \nonumber & = & \sum_{n , m , r , s } \frac{z^m}{m ! } \langle m |a| r\rangle \langle s |b| n\rangle \frac{\zeta^n}{n ! } \frac{\delta_{r , s}}{r ! } \\ \nonumber & = & \sum_{n , m , r , s } \int d\eta \frac{z^m}{m ! } a_{m , r}b_{s , n } \frac{\zeta^n}{n ! } \frac{1}{(r!)^2 } \eta^r \left ( - \frac{d}{d\eta } \right)^s \delta(\eta ) \\ \nonumber & = & \int \frac{d\eta d\eta'}{2 \pi } \sum_{n , s } a(z,\eta ) b_{s , n } \frac{\zeta^n}{n ! } \frac{(i\eta')^s}{s ! } ( i\eta')^s e^{-i\eta \eta ' } \\ & = & \int \frac{d\eta d\eta'}{2 \pi } a(z,\eta ) b(i\eta',\zeta ) e^{-i\eta \eta ' } \;. \label{kerpr}\end{aligned}\ ] ] given an operator @xmath35 , we may put it in _ normal form _ by commuting all creation operators to the left of all annihilation operators ; it will then have the form : @xmath42 with this we associate the _ normal kernel _ : @xmath43 the ordinary and normal kernels are related via : @xmath44 [ to prove eq.([kerrel ] ) we show that the coefficients of @xmath45 on the right- and left - hand sides are equal . the coefficient on the r.h.s . is : @xmath46 } \frac{1}{r ! } { \cal a}_{m - r , n - r } \;,\ ] ] while on the l.h.s . it is simply @xmath47 . the matrix element , however , may be written so : @xmath48 } { \cal a}_{m - t , n - t } \frac{m!n!}{t!}\end{aligned}\ ] ] which establishes the identity . ] we may now develop a path - integral representation for @xmath49 . to begin , recall trotter s formula , which allows us to write : @xmath50 each factor in the product has a corresponding kernel , which , using eq . ( [ kerrel ] ) , can be written @xmath51 with @xmath52 the normal kernel of the evolution operator . now using eq . ( [ kerpr ] ) , we have @xmath53 \right\ } \ ; , \label{uexp1}\ ] ] or , rearranging , @xmath54 \right . \\ & + & \left . \frac{t}{n } { \cal l } ( z,\eta_{n-1 } ) + z \eta_{n-1 } \right\ } \;. \label{uexp2}\end{aligned}\ ] ] finally , we let @xmath55 with @xmath56 , @xmath57 and @xmath58 , to obtain a path - integral expression for @xmath59 : @xmath60 + z \psi(t ) \right\ } \ ; , \label{ut}\ ] ] where the dot denotes a time derivative . the functional integrals over @xmath61 and @xmath62 are for @xmath63 , with boundary conditions @xmath64 and @xmath65 ; @xmath66 and @xmath67 are real . the symbol @xmath68 is defined by the limiting process in eq . ( [ uexp2 ] ) . [ the reader may wonder at this point what has become of the factors of @xmath69 in the denominator of eq . ( [ uexp2 ] ) . the answer is that the prefactor , which for the moment is undefined , will be fixed via normalization . ] note also that the first term in the argument of the exponential could be written more precisely as @xmath70 , i.e. , the time derivative is evaluated at @xmath71 with @xmath72 from above . while the function @xmath73 has no obvious physical significance , we shall see that @xmath66 is closely related to the random variable @xmath74 in the birth - and - death process . the kernel @xmath59 has two principal uses . first , from eqs . ( [ pgfz ] ) and ( [ kerpr ] ) , we see that @xmath34 provides a mapping between pgfs at different times : @xmath75 we make considerable use of this relation in the examples that follow . evaluating the integral is particularly simple if the initial distribution is poisson . then @xmath76 , and @xmath77 another simple case is @xmath78 ( exactly @xmath79 individuals initially ) , corresponding to @xmath80 , which yields @xmath81 setting @xmath82 in the above expressions , we obtain the probability of the state @xmath83 , which in many cases is absorbing , so that the survival probability is given by @xmath84 . independent of its probability interpretation , the kernel has a second utility : we may treat the argument of the exponential as an effective action . in the continuum limit of a system with many degrees of freedom , the function @xmath85 becomes the classical field @xmath86 ( similarly for @xmath87 ) , leading to a field theory for a markov process originally described by transition rates for particles on a lattice . ( an example of this , for the problem of directed percolation , is discussed in sec . viii . ) with such a field theory in hand , we can apply methods such as the renormalization group to study critical behavior . the effective action is known once we construct the evolution operator @xmath31 ; at no point do we need to write a langevin equation or stipulate noise properties . as a simple example we consider exponential decay , i.e. , the markov process with transition rates @xmath88 . the evolution operator is @xmath89 since this is in normal order , we have @xmath90 and thus , @xmath91 \;. \label{ut1}\ ] ] it is convenient to transform away the linear term in the action via the change of variable @xmath92 . the first term in the integral then contributes the additional ( boundary ) terms @xmath93 , and using @xmath64 we have @xmath94 \psi + ( z-1)\psi(t ) + \zeta \right ] . \label{ut1a}\ ] ] this can be evaluated exactly . recalling that @xmath95 , we see that the functional integral @xmath96 \psi \right]\ ] ] imposes the condition @xmath97 for @xmath98 , and so @xmath99 . in other words the functional integral over @xmath100 yields a product of @xmath30-functions : @xmath96 \psi \right ] = { \mbox const . } \times \prod_{0<s \leq t } \delta(\psi(s)-\zeta e^{-ws } ) , \label{delta}\ ] ] which when inserted in eq . ( [ ut1a ] ) gives @xmath101\ } , \label{ut1b}\ ] ] where @xmath102 represents the as yet undetermined normalization factor . using this in eq . ( [ map ] ) results in @xmath103\ } \phi_0(i\zeta ' ) \\ & = & { \cal c } \int d \zeta ' \delta[i\zeta ' - ( 1+(z-1)e^{-wt } ) ] \phi_0(i\zeta ' ) \\ & = & \phi_0 [ 1+(z-1)e^{-wt } ] , \end{aligned}\ ] ] where in the final line we set @xmath104 to satisfy the normalization condition , @xmath105 . if there are exactly @xmath2 particles at time zero , @xmath106 , and @xmath107^n\ ] ] which on expanding yields @xmath108 as expected . for a poisson initial distribution we find @xmath109 corresponding to a poisson distribution whose mean decays exponentially : @xmath110 . it is useful to extend the formalism to joint probabilities , i.e. , for the values of the process at different times . for @xmath111 , let @xmath112 be the probability of state @xmath113 at time @xmath114 _ and _ @xmath115 at @xmath116 , _ given _ @xmath79 at time 0 . the generating function for the joint probability is @xmath117 where in the second line we used the markov property and in the third , stationarity . let us modify slightly our notation for the one - time generating function , to include the initial condition , @xmath118 : @xmath119 then we have @xmath120 where , in the second line , we used eq . ( [ map ] ) . now , noting that @xmath121 , we have @xmath122 thus we have a formula analogous to eq . ( [ map ] ) , for the two - time generating function . the generalization to @xmath2 times is straightforward . for the decay process , eq . ( [ 2time ] ) reads : @xmath123 \nonumber \\ & \times & e^{-i\xi\xi ' } \exp\left[\xi\{(iz_2\zeta'-1)e^{-wt_2 } + 1\}\right ] ( i\xi')^{n_0 } \;,\end{aligned}\ ] ] which , after integrations over @xmath30-functions , yields : @xmath124 \right\}^{n_0 } \;.\ ] ] from this we readily obtain the expectation : @xmath125 and @xmath126 the covariance is then @xmath127 \;.\ ] ] for @xmath128 we find @xmath129 $ ] , which is the variance of a binomial random variable [ see eq . ( [ binom ] ) ] . the preceding example illustrates the general procedure for calculating probabilities , but no perturbative treatment was needed as the action was purely bilinear in the fields . an example of perturbation theory is afforded by analyzing the malthus - verhulst process , in which each individual has a rate @xmath130 to reproduce , and a rate of @xmath131 to die , if the total population is @xmath2 . ( the term @xmath132 represents saturation , so that the population does not grow without limit even if @xmath133 . ) for this process the kernel is @xmath134 since the corresponding evolution operator has the property @xmath135 + \mu n[|n-1\rangle -|n\rangle ] + \nu n(n-1)[|n-1\rangle -|n\rangle ] , \label{opmv}\ ] ] which evidently reproduces the rates defining the process . using eq . ( [ map ] ) we have ( with @xmath136 ) , @xmath137 . \label{ut2}\end{aligned}\ ] ] as before , we eliminate the linear term by changing variables , @xmath138 , yielding @xmath139\psi \right . \right . \ ! & - & \ ! \left . \lambda \hat{\psi}^2 \psi \!+\ ! \nu \hat{\psi}(1\!+\!\hat{\psi } ) \psi^2 \right ) \\ & + & \ ! \left . \zeta + ( z\!-\!1)\psi(t)\right ] , \label{ut2a}\end{aligned}\ ] ] where @xmath140 . separating the bilinear part of the action from the cubic and quartic terms , we have @xmath141\psi + \ ! \zeta + ( z\!-\!1)\psi(t)\right ] \exp[-s_i ] , \label{ut2b}\ ] ] where @xmath142 \equiv \int_0^t dt'{\cal l}_i ( t ' ) . \label{si}\ ] ] evidently @xmath143 must be treated perturbatively . it is interesting to note that while the process with @xmath144 admits an exact solution , the cubic term generates a perturbation series in the present formalism . when we expand @xmath145 we obtain a series of functional integrals over @xmath66 and @xmath146 of various products of these fields times the gaussian " factor @xmath147 . consider the basic contraction : @xmath148 = \int \!{\cal d } \ ! { \psi } \ ! \int \ ! { \cal d } \hat{\psi } \psi(t_1 ) \hat{\psi}(t_2 ) \exp \!\left[\ ! -\ ! \int_0^t\ ! \!dt ' \hat{\psi}[\partial_{t ' } \!+\ ! w]\psi + \ ! \zeta + ( z\!-\!1)\psi(t)\right ] . \label{brac}\ ] ] let @xmath149 & = & \exp \!\left[\ ! -\ ! \int_0^t\ ! \!dt ' \hat{\psi}[\partial_{t ' } \!+\ ! w]\psi + \ ! \zeta + ( z\!-\!1)\psi(t)\right ] \nonumber \\ & = & \exp \!\left[\ ! -\!\int_0^t\ ! \!dt ' \psi[-\partial_{t ' } \!+\ ! w]\hat{\psi } + \zeta[\hat{\psi}(0 ) + 1]\right ] , \end{aligned}\ ] ] where in the second line we integrated by parts and used the boundary conditions @xmath64 and @xmath150 . since the contraction @xmath151 $ ] is the functional integral of @xmath152 , it is convenient to introduce the operator @xmath153 via the relation @xmath154 = \hat{\psi}(t_2 ) { \cal f}[\psi,\hat{\psi } ] . \label{defk}\ ] ] the fact that @xmath155 = -(-\partial_{t '' } + w ) \hat{\psi}(t '' ) { \cal f}[\psi,\hat{\psi } ] , \label{fderiv}\ ] ] suggests that @xmath153 take the form @xmath156 where @xmath157 is a boundary term . ( [ fderiv ] ) and integrating by parts yields @xmath158 { \cal f}.\end{aligned}\ ] ] consistency with eq . ( [ defk ] ) then requires that @xmath159 which has the ` causal ' solution @xmath160 which leads to @xmath161 as may be verified directly . to evaluate @xmath148 = \int \!{\cal d } \ ! \psi \ ! \int \ ! { \cal d } \hat{\psi } \psi(t_1 ) { \cal k}(t_2 ) { \cal f}(\psi,\hat{\psi})~ , \label{brac2}\ ] ] we make use of our earlier result , eq . ( [ delta ] ) , to obtain : @xmath162 & = & \int \!{\cal d } \ ! \psi \psi(t_1 ) \left\ { - \int_{t_2}^t dt '' e^{-w(t''-t_2 ) } \frac{\delta}{\delta \psi(t '' ) } + ( z\!-\!1 ) e^{-w(t - t_2 ) } \right\ } \nonumber \\ & \;\;\;\;\;\;\;\;\;\ ; \times & e^{\zeta + ( z\!-\!1)\psi(t ) } \prod_{0<s \leq t } \delta(\psi(s)-\zeta e^{-ws } ) . \label{brac3}\end{aligned}\ ] ] now perform a functional integration by parts so that @xmath163 operates on @xmath164 . this yields @xmath162 \!\!\!\ ! & = & \!\ ! \int \!{\cal d } \ ! \left [ \int_{t_2}^t dt '' e^{-w(t''-t_2 ) } \delta(t_1-t '' ) + ( z\!-\!1 ) e^{-w(t - t_2 ) } \psi(t_1 ) \right ] \nonumber \\ & \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \times & e^{\zeta + ( z\!-\!1)\psi(t ) } \prod_{0<s \leq t } \delta(\psi(s)-\zeta e^{-ws } ) . \label{brac4}\end{aligned}\ ] ] we may now integrate over @xmath66 to obtain @xmath162 & = & \left [ ( z\!-\!1)e^{-w(t - t_2 ) } \zeta e^{-wt_1 } + \theta(t_1-t_2)e^{-w(t_1-t_2 ) } \right ] u_t^0(z,\zeta ) \nonumber \\ & \equiv & \langle \psi(t_1 ) \hat{\psi}(t_2 ) \rangle u_t^0(z,\zeta ) \ ; , \label{brac5}\end{aligned}\ ] ] where @xmath165 } . \label{defu0}\ ] ] in light of the discussion following eq . ( [ ut ] ) , we see that in eq . ( [ defk ] ) and the subsequent expressions , @xmath116 should be interpreted as @xmath166 , which means that in the expectation , eq . ( [ brac5 ] ) , we should take @xmath167 . to see what our results mean , consider the very simple situation of @xmath168 , which is equivalent to exponential decay with a rate @xmath169 . then , @xmath170,\ ] ] and @xmath171 , \label{ex1}\end{aligned}\ ] ] which leads to the exact result : @xmath172 } = e^{\zeta[1 + ( z\!-\!1)e^{-wt}]}[1 - ut\zeta(z\!-\!1)e^{-wt } + { \cal o}(u^2)].\ ] ] ( note that obtaining the correct result depends on using @xmath173 . ) the preceding analysis exposes a curious feature of this formalism . normally in a gaussian model , the expectation of the field @xmath174 is zero . here , by contrast , we have @xmath175 = \int \!{\cal d } \ ! { \psi } \ ! \int \ ! { \cal d } \hat{\psi } \psi(t_1 ) { \cal f } = \zeta e^{-wt_1 } u_t^0 , \label{exppsi}\ ] ] while from eq . ( [ brac3 ] ) we have @xmath176 = \int \!{\cal d } \ ! { \psi } \ ! { \cal d } \hat{\psi } { \cal k}(t_1 ) { \cal f } = ( z\!-\!1)e^{-w(t - t_1 ) } u_t^0 . \label{exppsih}\ ] ] this suggests that it would simplify matters if we were to introduce new fields : @xmath177 and @xmath178 which by construction have expectation zero . the boundary conditions on @xmath66 and @xmath146 imply that @xmath179 . noting that @xmath180 annihilates @xmath181 , we have @xmath182 ( \partial_{t ' } + w ) \varphi \nonumber \\ & = & -\int_0^t dt ' \varphi ( -\partial_{t ' } + w ) \hat{\varphi } \ ; -(z\!-\!1 ) \varphi(t),\end{aligned}\ ] ] where in the second line we integrated by parts and used the fact that @xmath183 annihilates @xmath184 . integrating by parts once again gives @xmath185 , \ ] ] which , when inserted in eq . ( [ ut2b ] ) , yields @xmath186 e^{-s_i } \\ & \equiv & u_t^0(z,\zeta ) \tilde{u}_t(z,\zeta ) \;. \label{ut2c}\end{aligned}\ ] ] the change of variable , then , yields the unperturbed solution , @xmath187 , automatically , and eliminates the boundary term from the exponential . the normalization is such that the functional integral is unity if @xmath188 . we may now repeat the analysis of the basic contraction by letting @xmath189 \equiv \exp \!\left[\ ! -\!\int_0^t\ ! \!dt ' \hat{\varphi}(\partial_{t ' } \!+\ ! w)\varphi \right ] , \label{defg}\ ] ] and defining @xmath190 such that @xmath191 one readily verifies that @xmath192 which leads directly to @xmath193 & = & u_t^0(z,\zeta ) \theta(t_1-t_2 ) e^{-w(t_1-t_2 ) } \nonumber \\ & \equiv & u_t^0(z,\zeta ) \ ; \langle \varphi(t_1 ) \hat{\varphi } ( t_2 ) \rangle . \label{conphi}\end{aligned}\ ] ] then an average of @xmath2 fields ( with equal numbers @xmath194 s and @xmath195 s ) may be written @xmath196 = u_t^0(z,\zeta)\ ; \langle \varphi(t_1 ) \hat{\varphi } ( t_2 ) \cdots \varphi(t_{n-1 } ) \hat{\varphi } ( t_n ) \rangle , \label{contn}\ ] ] with @xmath197 given by the sum of all distinct products of pairwise contractions of the form of eq . ( [ conphi ] ) . let us analyze the simple quadratic perturbation " using the @xmath194 " representation . introducing the new fields we have @xmath198 [ \varphi + \zeta e^{-wt ' } ] , \end{aligned}\ ] ] which gives us @xmath199 \exp \!\left[\ ! -u\!\int_0^t\ ! \!dt ' \left(\hat{\varphi}\varphi + b \varphi + c\hat{\varphi } \right ) \right ] , \label{uqpert}\ ] ] where @xmath200 and @xmath201 \;.\ ] ] let @xmath202 . the latter factor has the expansion : @xmath203 \nonumber \\ & \;\;\;\ ; \times & \left[\hat{\varphi}_2 \varphi_2 + b_2 \varphi_2 + c_2 \hat{\varphi}_2 \right ] \cdots \left[\hat{\varphi}_n \varphi_n + b_n \varphi_n + c_n \hat{\varphi}_n \right ] { \cal g}[\varphi,\hat{\varphi } ] , \end{aligned}\ ] ] where @xmath204 , etc . the @xmath205 term is unity ( by normalization ) , while the @xmath206 term vanishes , since @xmath207 . for @xmath208 , of the nine possible terms in @xmath209 , only one survives the functional integral , giving @xmath210 next consider the @xmath211 contribution . once again there is but a single nonzero term , since we must choose @xmath212 for the first factor ( there is no @xmath194 with time @xmath213 , with which to contract @xmath214 ) , @xmath215 for the final factor ( there is no @xmath195 with time @xmath216 , i.e. , no way to contract @xmath217 ) , while the middle factor must be @xmath218 . writing this contribution in terms of functional derivatives we have @xmath219 \label{u3}\end{aligned}\ ] ] upon performing the two functional integrations by parts , @xmath220 may act on @xmath221 and @xmath222 on @xmath223 , or vice - versa . in the former case we obtain the factor @xmath224 the @xmath30-functions can both be satisfied since @xmath225 $ ] and @xmath226 $ ] . on the other hand , the second alternative gives the factor @xmath227 since @xmath228 does not fall within the range of integration . combining this result with the other factors the third - order term becomes @xmath229 . we now introduce a diagrammatic notation which will prove essential in this and subsequent analyses . the three terms in @xmath230 are represented by vertices , so : @xmath231 a contraction between @xmath232 and @xmath233 ( with @xmath234 due to the factor @xmath235 ) is represented by joining the line exiting vertex @xmath236 with that entering vertex @xmath237 . thus the @xmath208 and @xmath211 terms correspond to the diagrams : @xmath238 respectively . associated with the line connecting vertices @xmath237 and @xmath236 ( with @xmath234 ) , is the factor @xmath239 . each diagram constructed according to the following rules corresponds to a term in the expansion of @xmath240 . \1 ) draw @xmath241 vertices of type @xmath242 and an equal number of type @xmath243 . add @xmath244 vertices of type @xmath245 , for a total of @xmath246 vertices . \2 ) form all possible _ unlabelled _ diagrams , i.e. , distinct connections in which each outgoing line is joined with an ingoing line . \3 ) generate all distinct labellings of each unlabelled diagram by assigning an index ( time ) to each vertex , such that the arrows always point from the larger to the smaller index . \4 ) a diagram consists of one or more connected parts ; its contribution to @xmath240 is the product of factors associated with these connected parts . in the example under consideration , the factor associated with the @xmath236-th connected part ( having @xmath247 vertices ) is @xmath248 we now make use of a well known theorem in diagrammatic analysis ( see , e.g. , 8.3 of ref . @xcite ) , which in the present context states that @xmath249 is given by the sum of all contributions associated with _ connected diagrams only_. in the present case there is exactly one connected diagram at each order @xmath250 , yielding @xmath251 which , when inserted in eq . ( [ uqpert ] ) , gives the exact result , @xmath252 . \label{uqpex}\ ] ] while the expansion of @xmath253 in terms of connected graphs represents a considerable simplification , it is possible , in principle , to evaluate @xmath254 directly , without transforming to the fields @xmath194 and @xmath255 . consider , for example , the second order term in the expansion of @xmath254 , eq . ( [ ex1 ] ) : @xmath256 \\ & \ ; & \times \left[\int_{t_2}^t d\tau ' e^{-w(\tau'-t_1 ) } \frac{\delta}{\delta \psi(\tau ' ) } + b(t_2 ) \right ] \psi(t_1 ) \psi(t_2 ) \;. \label{u2a}\end{aligned}\ ] ] noting that the functional derivative w.r.t . @xmath257 gives zero , the above expression is seen to be @xmath258 \ ; , \label{u2b}\ ] ] which is precisely the @xmath259 contribution in the expansion of @xmath260 , eq . ( [ uqpex ] ) . clearly , this analysis is considerably more cumbersome than the connected diagram expansion . next we consider the more complex , but still exactly soluble problem of a birth - and - death process without saturation , i.e. , eq . ( [ si ] ) with @xmath144 . our first step , as before , is to rewrite the perturbation in terms of the variables @xmath194 and @xmath195 : @xmath261 ^ 2 [ \varphi + c(t ' ) ] \\ & = & \lambda \int_0^t dt ' [ \hat{\varphi}^2 \varphi+ 2b \hat{\varphi } \varphi + b^2 \varphi + c\hat{\varphi}^2 + 2 b c \hat{\varphi } + b^2c ] \;. \label{lmu0}\end{aligned}\ ] ] in the second line , the final term is independent of @xmath194 and @xmath195 . integrating it , and combining it with the usual prefactor @xmath187 , we have for this problem @xmath262 \;. \label{u0pmu0}\ ] ] the remaining terms will again be analyzed using diagrams . there are five vertices , as shown in fig . 1 ( note that the three we have encountered before carry different factors here ) . the first vertex in fig . 1 ( terminal " ) must bear the lowest index of a given branch , since no lines exit from it . the second and fifth can only appear in the interior of a diagram , while the remaining two can only appear as the @xmath2th of an @xmath2-vertex diagram , since no lines enter . thus the expansion is not as complicated as it might at first appear . the lowest - order diagram is again @xmath263 which now corresponds to @xmath264 at third and higher orders there are graphs with bifurcations , the simplest being the one shown in fig . 2 , corresponding to @xmath265 the factor of 2 arises because there are two ways to contract the @xmath195-lines exiting vertex 3 . thus all the vertices in this problem , except for the terminal @xmath266 , carry a factor of 2 , either explicitly , or due to the combinatorial factor . the other third - order connected diagram is @xmath267 the two diagrams differ only by their numerical prefactors , the sum of which just cancels the @xmath268 that comes from the integrations . we show in the appendix that this pattern continues at higher orders , i.e. , that the sum of all numerical factors due to @xmath2-vertex diagrams is @xmath269 , so that @xmath270^n \;.\ ] ] combining this with eq . ( [ u0pmu0 ] ) , we obtain the exact result @xcite , @xmath271 \;.\ ] ] while the transformation to variables @xmath194 and @xmath195 , with expectation zero , yields a certain simplifcation , it also causes a proliferation in the number of vertices , so that in many cases it may be advantageous to work with the original variables @xmath66 and @xmath146 . we have seen that in this case the expansion of @xmath34 becomes quite complicated . fortunately , it is still possible to derive relatively simple expressions for the moments @xmath272 of the distribution . note that from the definition of @xmath273 we have @xmath274 and that , in general , the @xmath275-th _ factorial moment _ , @xmath276 , is given by : @xmath277 using eq . ( [ map ] ) we then have @xmath278 if @xmath34 is of the form of eq . ( [ ut2b ] ) , then @xmath279 \exp[-s_i ] \;. \label{drudzr}\ ] ] ( note that the prefactor @xmath280 is just @xmath281 for @xmath282 . ) thus the expectation of the variable @xmath66 is closely related to that of @xmath74 itself . in case @xmath283 ( so that @xmath80 ) , we have , using eq . ( [ mapn0 ] ) , @xmath284 evaluation of the above expression is facilitated by noting that @xmath285 if the initial distribution is poisson with parameter @xmath286 , then eq . ( [ mappoi ] ) implies that @xmath287 it is instructive to evaluate the above expressions first for the simple decay process , @xmath188 . in this case @xmath288_{z=1 } = e^\zeta [ \zeta e^{-wt}]^r \ ; , \label{drudzrdec}\ ] ] and a short calculation shows that for the initial condition @xmath283 , @xmath289 and @xmath290 = n_0 e^{-wt } ( 1 - e^{-wt } ) \;,\ ] ] as expected . for the poisson initial distribution , one finds that @xmath291 = p e^{-wt } \;,\ ] ] consistent with the probability distribution @xmath3 being poisson with parameter @xmath292 . next we consider the moments for the birth - and - death process . in this case @xmath293 corresponding to a vertex that branches to the left . to evaluate @xmath294 , we expand @xmath145 , generating diagrams consisting of vertices and a single sink " , with @xmath275 lines entering , to the left of all vertices . each line entering a node ( i.e. , a vertex or the sink ) , corresponds to a variable @xmath295 ; if not contracted , it contributes a factor of @xmath296 . ( we call such uncontracted lines external " . recall that for the birth - and - death process , @xmath297 . ) on the other hand , lines _ exiting _ a vertex correspond to @xmath146 variables , and _ must _ be contracted with a @xmath66 line , because @xmath298 = 0 $ ] when @xmath282 . with @xmath282 , the basic contraction is : @xmath299 thus the perturbation series for a given moment is much simpler than the full series for @xmath59 . consider the case @xmath300 . the sink has only a single line entering , so diagrams with @xmath301 vertices all give zero , and the only nonzero contribution to @xmath302 is for @xmath303 ( the sink itself ) , which is @xmath304 . for @xmath305 the sink has two lines . there is then the @xmath83 contribution , @xmath306 , and the 1-vertex diagram shown in fig . 3 . including the combinatorial factor of 2 , its contribution is @xmath307 using these results , we readily find that @xmath308 for @xmath283 , the variance is @xmath309 = n_0 \frac{\mu \!+\ ! \lambda } { w } e^{-wt}(1 - e^{-wt})\ ] ] for @xmath310 , while for @xmath311 we have @xmath309 = 2 \lambda n_0 t \;.\ ] ] for a poisson initial distribution , @xmath309 = p e^{-wt } \left [ 1 + \frac{2 \lambda } { w } ( 1 - e^{-wt } ) \right ] \;,\ ] ] which becomes @xmath312 in case @xmath311 . we see that at long times , the variance decays ( grows ) exponentially , for @xmath313 ( @xmath314 ) , and that for @xmath315 the variance grows @xmath316 , reflecting the diffusive character of the process in this case . higher - order moments may be evaluated similarly ; only diagrams with @xmath317 vertices contribute to @xmath318 . we now return to the malthus - verhulst process ; the three vertices associated with @xmath319 [ see eq . ( [ si ] ) ] are shown in fig . the series for @xmath320 begins with the sink " term , @xmath321 , and then includes diagrams with @xmath301 vertices . ( note that the @xmath2-th order contributions carry factors of @xmath322 , with @xmath323 variable . ) based on our diagrammatic analysis of the birth - and - death process , we can formulate the following simple rules for evaluating the @xmath2-th order contribution to @xmath324 : \1 ) draw all diagrams consisting of @xmath2 vertices and a single @xmath275-line sink to the left of all vertices . each line exiting a vertex must be contracted with a line entering another vertex to the left . \2 ) the sink has time variable @xmath4 ; assign time variables @xmath114, ... ,@xmath325 to the vertices from left ( nearest the sink ) to right . \3 ) for each external line include a factor of @xmath326 , where @xmath327 is the time variable of the associated vertex ( if the external line is attached to the sink , @xmath328 ) . for each internal line from vertex @xmath236 to vertex @xmath237 , ( @xmath329 ) include a factor of @xmath330 . \4 ) include the factors ( @xmath130 or @xmath331 ) associated with each vertex , and the combinatorial factor associated with the number of ways of realizing the contractions . finally , integrate over the time variables @xmath114, ... ,@xmath325 , with @xmath332 ( note that by fixing the time ordering we effectively include the factor @xmath333 that comes from expanding @xmath145 . ) let us consider some examples in the evaluation of @xmath334 ; the low - order diagrams are shown in fig . 5 . for @xmath335 we have only diagram ( a ) , whose contribution is : @xmath336 at second order we have the following contributions . diagram ( b ) : @xmath337.\end{aligned}\ ] ] diagram ( c ) : @xmath338.\end{aligned}\ ] ] diagram ( d ) : @xmath339 ^ 2 .\end{aligned}\ ] ] for fixed @xmath340 this yields the expansion : @xmath341 [ e^{-wt } \!-\ ! 1 \!+\ ! wt ] - \nu ( n_0 \!-\ ! 1)(n_0 \!-\ ! 2 ) ( 1 \!-\ ! e^{-wt})^2 \right)+ \cdots \right ] . \label{denmv}\end{aligned}\ ] ] for @xmath342 a similar calculation results in @xmath343 \rangle & = & n_0 ( n_0 \!-\!1 ) e^{-2wt } - 2\nu n_0 ( n_0 \!-\!1 ) t e^{-2wt } \\ & \;\;\;\ ; + & \frac{2}{w^2 } n_0 e^{-wt } ( 1 \!-\ ! e^{-wt } ) \left [ \lambda - \nu ( n_0 \!-\ ! 1)(n_0 \!-\ ! 2 ) e^{-wt } \right ] + \cdots . \label{fm2mv}\end{aligned}\ ] ] the evaluation of diagrams can be simplified somewhat by considering the laplace transform of @xmath344 : @xmath345 we have seen that each line entering a node with time variable @xmath327 carries a factor of @xmath346 , and each line exiting the node carries a factor of @xmath347 . letting @xmath348 be the number of lines entering node @xmath237 less the number exiting , we can write the time integration factor for the general diagram so : @xmath349 .\ ] ] inverting the order of the integrations , the laplace transform of the above expression becomes : @xmath350\ ] ] @xmath351 the diagramatic expansion for the malthus - verhulst process can be extended , and simplified ( by identifying the so - called _ reducible _ diagrams @xcite ) , but such analysis is beyond the scope of this article and will be defered to a future work . we close this section with an application of the second - order expansion for the mean population , eq . ( [ denmv ] ) . note that for short times , the factor multiplying @xmath352 may be written as @xmath353 with @xmath354 and @xmath355 $ ] . evidently , our analysis generates an expansion in powers of @xmath4 , whose convergence would seem to require that @xmath356 . if we wish to extend the range of validity of the expansion , we must transform to a new variable that remains finite as @xmath357 ; a glance at eq . ( [ denmv ] ) suggests that we use @xmath358 as the new variable . ( this transformation is very useful in analysis of series , for example in the study of random sequential adsorption @xcite . ) in fact , @xmath359 is readily expressed in terms of @xmath360 : @xmath361 with @xmath362 and @xmath363 to proceed , we form a pad approximant @xcite to the power series in @xmath360 : @xmath364 equating coefficients of @xmath360 and of @xmath365 , we find @xmath366 and @xmath367 ( here we must note that a pad approximant to a series of three terms can only serve as a very rough approximation ! ) we apply this expression to the malthus - verhulst process with parameters @xmath368 , @xmath369 , and @xmath370 . in fig . 6 we compare the numerical solution of the master equation with the perturbation theory expression , @xmath371 ( with @xmath372 represented by the pad approximant ) , and with the simple exponential decay , @xmath352 , for @xmath373 and @xmath374 . the perturbation expansion , using the pad expression , yields a good approximation to the correct value , despite the very short series used , and the fact that for @xmath374 we have @xmath375 , so that the time series has only a small radius of convergence . in this section we show how the path integral formalism can be applied to processes on a lattice , leading to a field theory in the continuum limit . the process of interest is a lattice of coupled malthus - verhulst processes , with diffusive exchange of particles between neighboring cells . this process describes the dynamics of a population distributed in space , with the individuals performing random walks , allowing the population to spread . in the continuum limit , the process is of great interest as a field theory for directed percolation @xcite . at each site * r * of the lattice , there is a process @xmath376 whose evolution is given by eq . ( [ opmv ] ) . the new feature in this model is diffusion , represented by the operator : @xmath377 where @xmath378 is the diffusion rate and @xmath379 is over the vectors from a given site to its nearest neighbors . the evolution kernel @xmath34 is a function of the variables @xmath380 and @xmath381 defined at each site ; we use @xmath382 and @xmath383 to denote these sets of variables . to write @xmath34 for this process we add the diffusive contribution to the malthus - verhulst part , eq . ( [ ut2 ] ) , to find @xmath384 , \label{utmvl}\end{aligned}\ ] ] where @xmath385 is the discrete laplacian , and the functional integrals are now understood to include the funcions @xmath386 and @xmath387 at each site . eliminating the linear term as usual , by letting @xmath388 , and introducing @xmath389 , we obtain @xmath390 \right ] . \label{utmvl2}\end{aligned}\ ] ] up to this point , our expression for @xmath34 is exact . we now make a number of simplifications , leading to an effective action for the lattice of coupled malthus - verhulst processes . ( each deserves a careful justification , but we shall not enter into such questions here . ) \i ) we drop the boundary term , which should not influence stationary properties . \ii ) we take the continuum limit , so that @xmath391 , and @xmath392 . \iii ) we discard the term @xmath393 , which turns out to be _ irrelevant _ to the scaling behavior near the critical point @xcite . under these approximations the argument of the exponential in eq . ( [ utmvl2 ] ) becomes the effective action @xmath394 this is the action corresponding to directed percolation @xcite or reggeon field theory ( a particle physics model with the same formal structure as the continuum theory of spatially extended population ) @xcite . this theory has been analyzed using renormalization group techniques , to show that the upper critical dimension @xmath395 , and to derive expressions for critical exponents in an expansion in @xmath396 @xcite . while such developments lie beyond the scope of this article , we can get a qualitative understanding of the physics represented by @xmath397 by ignoring , for the moment , the term @xmath398 . functional integration over @xmath146 then imposes the following partial differential equation as a constraint on @xmath66 : @xmath399 which is , in fact , the mean - field theory of directed percolation and allied models @xcite . we see that this equation correctly predicts an _ absorbing state _ , @xmath400 . for @xmath401 the solution flows to this state , regardless of the initial condition . for @xmath314 , however , another stationary state appears : the uniform solution @xmath402 . thus @xmath311 marks a continuous phase transition . the field @xmath403 may be interpreted as a local population density . this mean - field description is readily shown to yield the critical exponents @xmath404 ( for the order parameter @xmath174 ) , @xmath405 ( for the relaxation time ) , and @xmath406 ( for the correlation length ) @xcite . in this case , the effect of the neglected term in the action , @xmath407 , can be represented by a noise term @xmath408 in eq . ( [ dpmft ] ) , which now becomes a langevin equation or stochastic partial differential equation : @xmath409 where @xmath410 is a gaussian noise with @xmath411 and autocorrelation proportional to the local density @xcite : @xmath412 ( note that in this way the noise respects the absorbing state ) . in the presence of noise , the critical point is renormalized from its mean - field value of @xmath413 , and , more significantly , the critical exponents take non - mean - field values for @xmath414 . as we have noted , the transcription of a stochastic model to a langevin equation is not always straightforward . for this reason , the path - integral formalism discussed in this article is especially valuable : it allows on to construct an action ( that is , the starting point for a renormalization group analysis ) , without having to postulate noise properties . in addition to the analysis of directed percolation discussed above , recent applications of the method include the annihilation reactions @xmath415 @xcite , and branching and annihilating random walks @xcite . we have reviewed the formalism , based on the work of peliti , that maps a markov process to a path integral representation , and have presented detailed examples of its application to birth - and - death processes . we show , in particular , how the exact solutions for the decay and simple birth - and - death process can be recovered , and derive a perturbation expansion for moments in the malthus - verhulst process . finally , we show how the evolution kernel for a lattice of coupled malthus - verhulst processes leads , in the continuum limit , to a field theory for directed percolation . as the study of nonequilibrium processes grows , we expect the methods discussed here to attract ever - greater interest . we aim to show that the sum of all numerical weights associated with @xmath2-vertex diagrams in the expansion of @xmath253 for the birth - and - death process is @xmath269 . to see this , consider an arbitrary @xmath416-diagram , i.e. , one having @xmath2 vertices , @xmath417 of them bifurcations . there is a single factor of @xmath418 , associated with vertex @xmath2 . such a diagram will have @xmath419 terminal vertices ( carrying factors @xmath420 ) , and so @xmath421 nonterminal vertices that are not bifurcations , which carry factors @xmath422 . each @xmath423 carries a factor of @xmath424 for a total of @xmath425 such . next consider the factors @xmath426 for each vertex @xmath237 . if @xmath237 is terminal , there are two such factors ( from @xmath420 ) , and a factor @xmath427 due to the line entering the vertex . if @xmath428 is neither terminal nor a bifurcation , there is a factor @xmath426 from @xmath429 , while the exponential factors associated with the lines entering and leaving the vertex cancel . if @xmath428 is a bifurcation , there is no factor @xmath429 , but there is again a net factor of @xmath426 as two lines exit , while only one enters vertex @xmath237 . consider vertex @xmath2 . if it is a bifurcation , then the net factor is @xmath430 ( two lines exiting , factor @xmath431 in @xmath432 ) . the same holds if vertex @xmath2 is not a bifurcation ( single line exiting , factors in @xmath429 and @xmath433 cancel ) . in summary , there is a factor @xmath426 associated with _ each _ vertex . finally , all vertices carry a factor of 2 , except for the terminal ones , leading to an overall factor of @xmath434 . combining all of these observations , the contribution due to a given labelled @xmath416-diagram is @xmath435^n e^{-(n+1)t } \int_0^t dt_1 \cdots \int_0^{t_{n-1 } } dt_n e^{w(t_1+\cdots + t_n ) } \\ & = & \frac{2^{n - b-1}}{n ! } \left(\frac{\lambda}{w}\right)^n \zeta ( z-1)^{n+1 } e^{-wt } ( 1-e^{-wt})^n \;.\end{aligned}\ ] ] to find the contribution @xmath436 , it remains to evaluate the sum of all such terms , that is , to determine @xmath437 where the sum is over all distinct labelled diagrams of @xmath2 vertices . a labelled @xmath416-diagram has a _ weight _ of @xmath438 . let @xmath439 be the sum the weights of all @xmath416-diagrams . define the _ degree _ of a vertex as the number of lines that exit from it , so that a terminal vertex has degree 0 , bifurcations have degree 2 , and all other vertices have degree 1 . ( in this model there are no vertices with degree @xmath440 . ) it is easy to see that ( 1 ) each choice for attaching the new vertex generates a different diagram ; ( 2 ) the sets generated by different @xmath2-vertex labelled diagrams are mutually disjoint ; ( 3 ) applied to the complete set of @xmath2-vertex labelled diagrams , the procedure generates the complete set of @xmath441-vertex diagrams . when we attach the new vertex to one of degree 0 , the number of bifurcations does not change , so the new diagram has an additional factor of 2 in its weight . if on the other hand we attatch the new vertex to one of degree 1 , we generate a new bifurcation , and the weight remains unchanged . recalling that an @xmath416-diagram has @xmath442 vertices of degree zero , and @xmath443 of degree 1 , we have for @xmath444 the following recurrence relation : to solve for @xmath453 in general , we introduce a generating function @xmath454 ( since @xmath453 appears to grow like @xmath269 we need the factorial for @xmath455 to have a nonzero radius of convergence . without it , we get a function with an essential singularity at @xmath456 , as the reader may verify ! ) for purposes of analysis , it is convenient to define @xmath457 and @xmath458 for @xmath448 , which is completely consistent with the recurrence relation for @xmath208 . ( there is , of course , no diagram with @xmath206 . ) for @xmath205 , @xmath459 vanishes identically , so we must add a source term in eq . ( [ recrel ] ) . the modified recurrence relation , valid for @xmath460 and @xmath461 , is fig . mean population size in the malthus - verhulst process with @xmath368 , @xmath369 , @xmath370 ; @xmath475 ( upper ) ; @xmath476 ( lower ) . in each panel , the middle curve represents the exact ( numerical ) solution , the lower , the pad approximant to the second - order perturbation series , and the upper , simple exponential decay , @xmath477 .