article
stringlengths 0
745k
| abstract
stringlengths 0
92.7k
|
|---|---|
of late years , incommensurate ( ic ) stripe structures have been interesting subjects in various physical phenomena . as the typical examples , we may list an alloy @xmath2 which contains heavy rare earth metals , er , tm,@xcite the incommensurate phase of dielectric material such as @xmath3 and @xmath4,@xcite and the stripe structure in @xmath5 planes of oxide superconductors.@xcite in @xmath2 , the longitudinal incommensurate oscillatory phase appears between the paramagnetic phase and the ordered phase .
the aligned holes ( domain walls ) separate antiferromagnetic stripes in @xmath5 planes of oxide superconductors , then the spin and charge are modulated . in such systems ,
cooperative effects of fluctuation and frustration are considered to play important roles .
thus , we sometimes treat them with the axial next - nearest - neighbor ising ( annni ) model as the simplified theoretical model .
for instance , when a uniaxial anisotropy is strong in the dielectric material , the hamiltonian is equivalent to the annni model if we only consider the dipole interactions up to the next - nearest - neighbor distance .
the phase diagram of @xmath3 obtained by experiments@xcite agrees with that of the three - dimensional annni model obtained by the mean - field approximation.@xcite in copper oxide materials , @xmath6 and @xmath7 , the cu - o - cu chains with ferromagnetic nearest - neighbor and antiferromagnetic next - nearest - neighbor interactions are aligned on two - dimensional planes whose interchain interactions are antiferromagnetic .
furthermore , the spins on this plane are predicted to have a strong ising anisotropy.@xcite consequently , we may treat these copper oxide planes as the two - dimensional annni model , between which the conduction electron planes exist . in the annni model , there are exchange interactions up to the next - nearest - neighbor pairs along one axis , while there are only the nearest - neighbor interactions along the other axes .
most commonly , we take a convention that the nearest - neighbor interactions are ferromagnetic and the next - nearest - neighbor ones are antiferromagnetic , which cause frustration . when the next - nearest - neighbor interactions ( frustration ) are weak , a ferromagnetic state is the ground state , and a paramagnetic - ferromagnetic phase transition occurs . on the other hand , when frustration is strong , the ground state is the _ antiphase _ state ( abbreviated by @xmath8 ) , which is a commensurate(c ) stripe structure of two up - spins and two down - spins like @xmath9 .
it is widely accepted that the incommensurate stripe phase exists between the paramagnetic phase and the antiphase , and that the successive phase transitions ( paramagnetic - ic - antiphase ) take place ; there is a `` finite '' incommensurate phase where the up - spins and down - spins are aligned with a period longer than two .
although only a topology of the phase diagram is known by the mean - field approximation in the three dimensional model,@xcite estimates for the phase transition temperatures in two dimensions have been done by several approximation theories @xcite and numerical simulations.@xcite however , values of the phase transition temperatures scatter much depending on the methods employed .
one of the reasons that make the estimate difficult is a lack of a reliable numerical simulation . because of frustration
, the correlation time of monte carlo ( mc ) simulations becomes very long near the critical point in large systems .
thus , we are hardly able to reach thermal equilibrium states within reasonable time steps . a powerful method called a cluster heat bath ( chb ) method@xcite was developed to reduce the correlation time , however , the system size accessible by this method is restricted up to about @xmath10 lattice sites .
recently , a new method using a mc simulation has been developed : the non - equilibrium relaxation ( ner ) method.@xcite we are able to understand phase transitions from the differences of behaviors in non - equilibrium relaxation processes , which have been discarded in conventional mc simulations .
since we do not wait until the equilibrium is realized , we can use the cpu time to enlarge the system sizes .
therefore , it makes possible to treat large systems that can not be possible by other methods .
accordingly , there expected to be little finite - size effect in the obtained data , and thus we can regard them as those of the infinite systems . by reasons stated above , the ner method can be effective especially in systems with a slow dynamics which a very long equilibration is necessary .
@xcite in this paper , we study the two - dimensional annni model by the ner method in order to determine the successive phase transition temperatures .
however , the obtained results suggest that the ic phase may disappear within limits of numerical errors .
moreover , we found a very exotic phase transition , which seems to be the kosterlitz - thouless ( kt ) transition@xcite if it is observed from the high temperature side , and seems to be the second - order transition if observed from the low temperature side .
this evidence suggests that the frustration parameter does not serve as an asymmetric parameter@xcite which explicitly favors the ic structure .
we describe the model hamiltonian and the ner method in sec .
[ sec2 ] , and present simulational results in sec .
section [ sec4 ] is devoted to conclusions .
the two - dimensional annni model is described by the following hamiltonian , @xmath11 where @xmath12 is the nearest - neighbor interaction along the chain direction , which is the direction that has no frustration . @xmath13 and @xmath14 are respectively the nearest - neighbor and the next - nearest - neighbor interactions along a direction perpendicular to the chain direction ( axial direction that has frustration ) , and @xmath15 . in this paper , we fix @xmath16 for simplicity , and impose the open boundary conditions along the axial ( @xmath17 ) direction , while we use the periodic boundary conditions along the chain ( @xmath18 ) direction .
we define a ratio between the nearest and the next - nearest - neighbor interactions along the axial direction as @xmath19 .
our interest is restricted to the region @xmath20 , where the successive phase transitions ( @xmath21 ) has been considered to occur .
the fermion approximations assumed that the system has straight domain walls along the chain direction at the intermediate temperatures.@xcite that is , domains whose periods are longer than two appear among the commensurate antiphase domains . in this case , the spin structure becomes incommensurate along the axial direction , while the spins along the chain direction order ferromagnetically . on the other hand
, it was postulated that domain walls run along the axial direction in the interface free energy method of @xmath22-hartmann and zittartz(mhz).@xcite namely , the correlation of spins ordered ferromagnetically along the chain direction is destroyed by these domain walls .
sato and matsubara@xcite discussed using the chb simulation that the transition temperature ( @xmath23 ) between the antiphase and the ic phase agrees with that of the free - fermion approximation while the transition temperature ( @xmath24 ) between the ic phase and the paramagnetic phase is close to @xmath23 obtained by mhz .
thus , they considered that the domain walls penetrate into the system along the chain direction at @xmath23 as the temperature increases from the ground state .
the ferromagnetic correlations along the chain direction remain until they are destroyed at @xmath24 and the paramagnetic state is realized .
this idea means that the ic phase exists between the penetration of incommensurate domains in the _ axial _ direction and the disappearance of the ferromagnetic _ chain _ correlation .
therefore , the spin structure along the axial direction is incommensurate , while it is ferromagnetic along the chain direction in the ic phase .
it has been considered that the phase transition between the paramagnetic phase and the ic phase is the kt transition and that between the ic phase and the antiphase is the second - order transition .
we consider following two quantities to clarify the phase transitions .
one is the _ antiphase magnetization _ defined by @xmath25 where @xmath26 is the total number of spins in the system , @xmath27 denotes the spin value of the site @xmath28 at time @xmath29 , and @xmath30 represents the antiphase ordered state .
the antiphase magnetization takes a finite value in the antiphase , but vanishes in the ic phase and the paramagnetic phase , since the domain walls destroy the antiphase state at the lower transition temperature , @xmath23 .
therefore , this parameter is employed to estimate @xmath23 . here
, we should note that the antiphase magnetization is not a relevant order parameter that decays algebraically at the transition temperature .
since the elementary excitation is a domain wall which readily percolates the chain direction , the antiphase magnetization decays exponentially once the domain wall penetrate into the system .
usually , the density of the domain walls has been used as an order parameter .
this parameter is considered to diverge algebraically at the transition temperature as @xmath31 , which is equivalent to that the correlation length diverges algebraically as @xmath32 . in this point , the phase transition between the ic phase and the antiphase is the second - order transition . in the monte carlo simulations ,
the correlation length is related to the characteristic time as @xmath33 , where @xmath34 is called the dynamic exponent .
therefore , the characteristic time also diverges at @xmath23 as @xmath35 .
we can extract it from the relaxation of the antiphase magnetization : the time that the antiphase magnetization begins to decay exponentially is the characteristic time that the domain wall penetrates into the system .
we execute the finite - time scaling to obtain the characteristic time at each temperature and estimate @xmath23 as its diverging temperature .
a concrete procedure of the finite - time scaling is explained in sec .
[ sec2_3 ] .
the antiphase magnetization is an extensive variable so that it shows better accuracy as the system size is enlarged because of the self averaging .
thus , we use the antiphase magnetization to estimate the lower transition temperature , @xmath23 .
the other quantity is the _ layer magnetization _ defined as @xmath36 where @xmath37 and @xmath38 are the length of the system along the chain and the axial direction , respectively .
the layer magnetization reflects the spin order along the chain direction . in the ic phase ,
the incommensuration is realized in the axial direction .
the spin arrangement along the chain direction should order ferromagnetically or at least be critical .
the free energy of this state , which is of the order of @xmath39 , is always lower than that of the state with the spins disordered only along the chain direction , which is of the order of @xmath40 , if @xmath41 is positive .
accordingly , the layer magnetization vanishes exponentially in the paramagnetic phase , but takes a finite value or behaves in a power law in the ic phase . therefore
, this order parameter is used to estimate the upper transition temperature , @xmath24 .
we determine the transition temperatures by using these two quantities in this paper .
phase transitions occur in the infinite - size limit(@xmath42 ) and in the equilibrium limit(@xmath43 ) .
because we can not take both limits at the same time in the simulation , the equilibrium limit has been taken first conventionally , and then , we take the infinite - size limit by using the finite - size scaling . however , the dynamics of the simulations are very slow in the frustrated systems , which causes a very large correlation time . from a time - space relation ,
@xmath33 , this means that the correlation length is also very large . as will be mentioned in sec .
[ sec3 ] , we estimate the correlation length , @xmath44 , for @xmath45 and @xmath46 , which resides in the ic phase of previous phase diagram .
in this situation , reliability of the finite - size scaling might become doubtful . in this paper , we follow a completely alternative approach to the thermodynamic limit , i.e. , we observe the relaxation of the infinite - size system to the equilibrium state . in order to extract the equilibrium properties , the finite - time scaling analysis is utilized instead of the conventional finite - size scaling .
this approach is known as the non - equilibrium relaxation ( ner ) method .
@xcite actually , we prepare a very large lattice and observe the relaxation of physical quantities .
the simulations are stopped before the finite - size effect appears .
accordingly we can regard the systems as the infinite systems . using the ner method
, we can estimate the phase transition temperature and the critical exponents by examining behaviors of the relaxation processes to the thermal equilibrium state ( non - equilibrium relaxation processes ) .
the analysis is based on the dynamic finite - size scaling hypothesis of the free energy that @xmath47 where @xmath48 are the relative temperature , the symmetry breaking field , the system size and time , respectively . @xmath41 and @xmath49 denote static exponents , while @xmath34 is a dynamic exponent , and @xmath50 is a dimension of space.@xcite the order parameter is given by a derivative of the free energy with the field : @xmath51 where we have set @xmath52 at the transition point ( @xmath53 ) because the characteristic time scale @xmath54 and the correlation length @xmath55 should scale as @xmath56 .
equation ( [ dfss_ord ] ) describes that the order parameter decays in a power law with time at the transition temperature . on the other hand ,
the order parameter behaves exponentially at temperatures away from the critical point .
thus , we are able to estimate @xmath57 by this difference .
actually , we measure the order parameter @xmath58 at each mc step @xmath59 started from an ordered spin configuration at a given temperature .
we repeat this mc run by changing the random number seeds , and @xmath58 is averaged over these independent mc runs .
the temperature at which the @xmath58 curve decays in a power law is the transition temperature .
we use the local exponent to ascertain whether the order parameter decays in a power law or exponentially , which is defined by @xmath60 when we plot the local exponent against @xmath61 , it diverges to infinity for @xmath62 , it converges to zero for @xmath63 and to a finite value ( @xmath64 ) at @xmath65 in the limit of @xmath66 .
the upper bound of @xmath57 is the lowest temperature that @xmath67 diverges , and the lower bound is the highest temperature that @xmath67 decays to zero .
the convergent value of the local exponent at @xmath65 is the critical exponent @xmath68 from eq .
( [ dfss_ord ] ) . here
, we describe the ner of fluctuation . @xcite the susceptibility is written by differentiating eq .
( [ dfss_free ] ) with the symmetry breaking field twice , @xmath69 where we have used the scaling relations : @xmath70 the susceptibility diverges in a power law at the transition temperature .
thus , we are able to estimate the transition temperature and the critical exponents from the susceptibility .
it is also noticed that the ner of fluctuation does not require us to start with a symmetry - broken ordered state .
the quantity of fluctuation always takes a definite value and diverges at @xmath57 even though the symmetry is not spontaneously broken .
therefore , we can start from a paramagnetic state or any state in this scheme .
this is especially useful when an ordered state is not known yet , or it is difficult to realize . the layer magnetization defined by eq .
( [ layer_mag ] ) is equivalent to the second derivative with respect to a local field along a single chain in the paramagnetic phase .
when the simulation starts from the antiphase , the first derivative term remains finite in the ner process even though the temperature is in the paramagnetic phase .
therefore , the layer magnetization is not regarded as the susceptibility .
the ner of the layer magnetization is within a scheme of the ner of fluctuation , only when the simulation is started from the paramagnetic state . in case when the ner function does not begin to decay algebraically within a reasonable time , it is difficult or almost impossible to estimate the transition temperature directly by the local exponent .
even in such a situation , we are able to determine it by using the finite - time scaling analysis , which is a direct interpretation of the finite-``size '' scaling by a relation @xmath56 .
we present the finite - time scaling relation @xcite as follows , @xmath71 where @xmath68 , and @xmath72 denotes a relaxation time at the relative temperature @xmath73 . since the relaxation time diverges algebraically in the case of the second - order phase transitions , the relation between the relaxation time and the relative temperature is described by @xmath74 on the other hand , in the case of the kt transition,@xcite it is considered that the relaxation time diverges exponentially , which we assume that @xmath75 now , we describe how we actually estimate the transition temperature by using eq .
( [ ft_scaling])@xmath76([kt_trans_sc ] ) .
we use only data which is clearly in the paramagnetic phase ; @xmath62 .
first , we plot @xmath77 at various temperature against @xmath78 by using eq .
( [ ft_scaling ] ) to determine @xmath79 and @xmath72 so that all data points fall on a single curve .
next , we plot @xmath72 against @xmath73 , and fit the points to a smooth curve as we change @xmath57 and @xmath80 by using eq .
( [ 2_trans_sc ] ) for the second - order transition . in the kt - transition case
, we use eq .
( [ kt_trans_sc ] ) instead .
the temperature at which the least - square fitting error becomes minimum is the most probable estimate for @xmath57 .
the phase transition between the paramagnetic phase and the ic phase is the kt transition and that between the ic phase and the antiphase is the second - order transition in the two - dimensional annni model .
therefore , we use eq .
( [ kt_trans_sc ] ) for the upper transition temperature , and eq .
( [ 2_trans_sc ] ) for the lower one .
we examine the phase transition temperature for @xmath81 and @xmath82 .
the temperature is measured in a unit of @xmath83 . here , the ner of two quantities are presented by the simulation started from the antiphase state .
first , we observe the ner of the layer magnetization to check the relation between the finite size , @xmath37 , and the correlation length , @xmath84 , along the chain direction .
we change the size along the chain direction from @xmath85 to @xmath86 , while the length along the axial direction is fixed to @xmath87 .
figure1 shows the ner of the layer magnetization , eq .
( [ layer_mag ] ) , for @xmath45 and @xmath46 .
this temperature belongs to the ic phase ( kt phase ) in previous investigations as summarized in table [ tc_comparison ] .
when the system size is small ( @xmath88 ) , the relaxation roughly looks like a power - law decay , which misled us to the kt phase .
as the system becomes larger , the relaxation exhibits an exponential decay confirming us that the system is in the paramagnetic phase .
note that the convergence to a finite value is due to the finite - size effect . by its definition
, the equilibrium value of the layer magnetization is roughly estimated as @xmath89 in the paramagnetic phase .
as shown in fig.1 , the convergence value takes a half - value if the system size is doubled .
thus , we can estimate the correlation length @xmath90 in this system . from this figure
, we can also understand the relation between the finite size - effect and the time - effect ; the effective time that the system behaves as the infinite size .
for example , it is about @xmath91 mcs in the system with @xmath92 . after this time scale
, finite size effect appears in dynamics . in this subsection
, we use the lattice of @xmath93 and determine the observing time at each temperature until which the relaxation curve do not begin to bend to an equilibrium value .
we average @xmath94 independent mc runs at each temperature .
figure2 is a raw data of the ner of the antiphase magnetization , eq .
( [ anti_mag ] ) , for @xmath81 .
at all the temperatures , the antiphase magnetization clearly decays exponentially , which guarantees that the transition temperature must be lower than 0.98 . as mentioned in sec .
[ sec2_0 ] , the antiphase magnetization is not a relevant order parameter , and so it decays exponentially as soon as domain walls penetrate into the system .
therefore , we estimate the transition temperature by using the finite - time scaling . at first , we determine the exponent , @xmath79 , and the relaxation time , @xmath95 , at each temperature so that the scaled data , @xmath96 , fall on a single scaling function when plotted against @xmath97 ( fig.3(a ) ) . here , the relaxation times are normalized by the value of @xmath98 , and are listed in table [ tau_0.6 ] .
the exponent , @xmath79 , that gives best scaling is @xmath99 .
excellence of the scaling shown in fig.3 may yield the validity of the finite - time scaling hypothesis .
next , we estimate @xmath23 by eq .
( [ 2_trans_sc ] ) because it is predicted that the phase transition between the antiphase and the ic phase is the second - order .
we plot @xmath95 against @xmath100 as changing @xmath23 and find a @xmath23 that gives the best linearity in the log - log scale as shown in fig.3(b ) .
we obtain @xmath101 and @xmath102 .
the error @xmath103 is a range of temperature in which the data points clearly fall on the fitting line .
the exponent has a range within which we can perform a good scaling as @xmath104 .
the transition temperature takes the same value irrespective of the choice of @xmath79 .
evidences such as the algebraic divergence of @xmath95 and excellence of the finite - time scaling support that this transition is the second - order .
this is a clear distinction from the three dimensional model , where the lower transition is considered as the first - order .
we estimate the upper transition temperature , @xmath24 , using the layer magnetization , eq .
( [ layer_mag ] ) , in the same way as mentioned above .
figure4 shows raw data of the ner of the layer magnetization , @xmath105 , for @xmath81 . here
, the layer magnetization clearly decays exponentially , and thus the @xmath23 must be less than 0.98 . the finite - time scaling is shown in fig.5(a ) .
the obtained relaxation time , @xmath106 , is also presented in table . [ tau_0.6 ] . since it is predicted that the phase transition between the ic phase and the paramagnetic phase is the kt transition , we fitted @xmath106 by eq .
( [ kt_trans_sc ] ) in fig.5(b ) , by which we obtain @xmath107 and @xmath108 .
if we assumed that the phase transition is the second - order transition , and performed the fitting by eq .
( [ 2_trans_sc ] ) , the estimated @xmath24 becomes far from a physically meaningful value . in consequence ,
we confirm that the upper phase transition is of the kt type .
actually , sato and matsubara@xcite performed the finite - size scaling of the layer magnetization supposing @xmath109,\ ] ] which gave @xmath110 and @xmath111 for @xmath81 .
this scaling form is equivalent to the finite - time scaling , eq .
( [ ft_scaling ] ) and ( [ kt_trans_sc ] ) , if we admit @xmath112 and @xmath113 .
the difference of the obtained transition temperatures can be attributed to the difference of the system sizes .
we estimate @xmath99 by minimizing the normalized residual , however , this value does not correspond to @xmath114 .
because we start simulations from the antiphase state , the layer magnetization contains contributions from the antiphase magnetization , @xmath115 .
we have obtained the @xmath23 and the @xmath24 for @xmath45 in the same way . in the finite - time scaling plot
, we used the data at five temperature ranging @xmath116 , and obtained @xmath117 and @xmath118 .
the phase transition temperatures , @xmath24 and @xmath23 , coincide within limits of error both for @xmath81 and @xmath82 .
this means the ic phase may vanish in the thermodynamic limit .
because this is a rather daring conclusion , we must confirm it from another point of view .
actually , it might be dangerous to estimate the transition temperature between the paramagnetic phase and the ic phase started from the ground state .
we start the simulation from the paramagnetic state and observe the ner of fluctuation ; the layer magnetization . since the observable is a quantity of fluctuation , it is necessary to take much more sample averages compared to the ner of antiphase magnetization
. therefore , the system size is restricted to @xmath119 at most .
the ner function of fluctuation diverges algebraically at @xmath65 , diverges exponentially at @xmath63 and remains finite at @xmath62 .
we find the phase transition temperature from these differences .
it is noticed that the ner function finally converges to a finite value because the system is finite .
therefore , we must be aware of the range of time not to observe the finite - size effect .
this is a time scale that the correlation length reaches the finite system size .
we compare the ner functions of two different sizes ( @xmath120 and @xmath121/@xmath122 and @xmath123/@xmath124 and @xmath125 et al . ) for every data point , and estimate this crossover time until which two curves fall on the same line and the finite - size effect does not appear .
an example of this comparison is shown in fig.6 .
the ner functions of the layer magnetization from the paramagnetic phase , @xmath126 , are plotted for three sizes , @xmath119 , @xmath127 and @xmath128 at @xmath129 and @xmath81 .
three curves fall on the same line until @xmath130 mcs .
after this crossover time , however , the curve of @xmath128 bends down from the other two curves , probably because the correlation length reaches @xmath131 at this temperature .
so , we have to discard the data of @xmath132 after @xmath130 mcs . comparing two curves of @xmath119 and @xmath127
, we can use the data of @xmath127 until at least @xmath133 mcs .
table [ size_effect ] shows the crossover mcs at each temperature for @xmath81 and @xmath82 .
thus , we use a system of a proper size for a proper observing time at each temperature .
we take averages over @xmath134 independent mc runs .
the ner of the layer magnetization , @xmath126 , from the paramagnetic state are shown in fig.7 for ( a)@xmath81 and ( b)@xmath45 .
figure8 shows the corresponding local exponents . in fig.8(a )
, the exponent decreases for @xmath135 and diverges for @xmath136 . at @xmath137
, it converges to a finite value . in consequence
, we predict that the phase transition temperature between the paramagnetic phase and the ic phase is @xmath138 . for @xmath45
, we obtained @xmath139 as shown in fig.8(b ) .
furthermore , we also analyze the upper transition temperature , @xmath24 , using the finite - time scaling as shown in figs.9 and 10 .
we show the scaling in fig.9 for ( a)@xmath81 and ( b)@xmath45 .
since the layer magnetization diverges algebraically as @xmath140 at the transition temperature , we plot @xmath141 against @xmath142 .
all the curves excellently fall on the same line , by which we obtain the relaxation time as shown in fig.10 . here
, we adopt @xmath143 that is a value which the local exponent converges to ( see fig.8 ) . if we admit the kt - criterion @xmath144 , the dynamic exponent is estimated as @xmath145 .
next , we fit the relaxation time using eq .
( [ kt_trans_sc ] ) .
figure10 shows the best least - square fitting for ( a)@xmath81 and ( b)@xmath45 .
we obtain the transition temperature @xmath146 for @xmath81 and @xmath147 for @xmath45 .
these transition temperatures are again consistent with the values estimated from the local exponents of the ner from the paramagnetic state , and those by the scaling from the ground state
. therefore , we are able to conclude that @xmath24 and @xmath23 coincide with each other within limits of error in both cases of @xmath81 and @xmath82 .
the phase transition temperature between the paramagnetic phase and the ic phase is @xmath148 and that between the antiphase and the ic phase is @xmath101 for @xmath81 , and @xmath149 and @xmath117 for @xmath45 .
since these temperatures are very close to each other , it is suggested that the ic phase does not exist or it is very narrow even if it exists .
we need to pay attention to a fact that monte carlo simulation is not able to exclude an very tiny temperature region .
it has been considered that the successive phase transitions with a finite ic phase take place for @xmath150 in two - dimensional annni model . in this paper , we estimated the phase transition temperatures by applying the ner method , and found that the @xmath24 is equal to @xmath23 within limits of errors .
this is a very exotic phase transition , which is the kt type if approached from the high - temperature side , and is the second - order if approached from the low - temperature side .
therefore , we speculate successive phase transitions with an _ infinitesimal _ ic phase may occur in this system . in the studies of the c - ic transitions in two - dimensional systems , it has been investigated the systems with an asymmetric parameter which explicitly favors the ic structure .
@xcite for a finite value of the parameter , there exists the finite ic phase between the commensurate phase and the paramagnetic phase . in the limit of vanishing the parameter
, the ic phase shrinks to the infinitesimal .
the frustration parameter , @xmath151 , of the annni model has been considered as the asymmetric parameter based on the approximate theory @xcite valid only at low temperatures , and the small - scaled monte carlo simulations.@xcite however , what is actually favored by frustration is the creation of the domain wall . between the creation of the domain wall and the realization of the ic structure ,
there are many conditions to satisfy , which have been supposed in the free - fermion approximation.@xcite one of these is the spin correlation along the ferromagnetic chain . in the two - dimensional annni model ,
the non - frustrated direction is only one dimension , and thus the spin correlation along the ferromagnetic chain can be easily destroyed .
therefore , we question regarding the frustration parameter as the asymmetric parameter .
if we consider that these two are not related with each other and the asymmetric parameter is zero in the present model , the width of the ic phase becomes infinitesimal .
this is what we observed in this paper . here
, we show the comparison of the obtained transition temperatures with the previous ones in table [ tc_comparison ] . it is recognized that our @xmath24 is lower than any other ones , though @xmath23 is consistent with each other .
this can be explained by the difference of the finite - size effect .
the phase transition between the ic phase and the paramagnetic phase is confirmed to be the kt transition while the one between the antiphase and the ic phase is the second - order .
the correlation length diverges algebraically against the temperature in the latter , while it diverges exponentially in the former case .
if a system size is small compared with the correlation length and the accuracy of the numerical data is not enough , even the finite - size scaling analysis may mislead to a wrong @xmath57 , where the finite but very large correlation length reaches the finite system size .
therefore , it is very likely that the kt transition temperature obtained previously is over - estimated . we may rather easily obtain the phase transition temperature accurately in the second - order transition , because the divergence is algebraically .
in the two - dimensional annni model , the correlation length along the chain direction is @xmath90 at @xmath46 and @xmath45 which is the temperature a little higher than the kt transition temperature and the correlation length is far larger than the system sizes ever simulated previously ( @xmath152).@xcite we actually used data with system sizes , @xmath153 in this paper , and checked the finite - size effect and the range of observing time in which the system behaves as an infinite system at each temperature . thus , our data can be considered as closed to the thermodynamic value .
this is a reason why we could detect the kt transition accurately . in this paper
, we have supposed that the ferromagnetic correlation along the chain direction remains finite or at least it is critical in the ic phase .
thus , the @xmath24 we have obtained is the temperature at which the ferromagnetic correlation becomes critical . above this temperature , each ferromagnetic chain exhibits paramagnetism , even though the correlation length is very large near the @xmath24 .
the present results only exclude the finite ic phase of this type . in this study
, it is clarified that the domain walls penetrate into the system along the chain direction at the lower transition temperature , @xmath23 .
the spin configuration along the chain direction changes drastically from the ferromagnetic ordered state to the paramagnetic disordered state at the same temperature within limits of error .
we can neglect the ic phase where the ferromagnetic correlation has been expected to be critical .
therefore , the domain walls mostly run straight along the chain direction .
we consider this is why the naive free - fermion approximation of villain and bak@xcite gives the best quantitative agreements with our estimate of @xmath23 , and is the best approximation .
the analyses for other regions of @xmath151 and determination of the critical exponents will be a task in the future . the ner of the binder parameter , the specific heat and the spin correlation with high accuracy is necessary.@xcite as for the dielectrics , @xmath154 , the phase diagram@xcite in the low concentration of ti is very similar to that of the annni model if we interpret the paraelectric , the ferroelectric and the antiferroelectric phases as the paramagnetic , the ferromagnetic and the antiphase phases .
there are two controversial explanations for the existence of the incommensurate phase observed in the experiment .
et al_.@xcite concluded that the appearance of the ic phase is due to the surface effect by two experiments using the powder neutron diffraction and a transmission electron microscope . on the other hand , viehland _
et al_.@xcite considered it a bulk effect by directly observing the high resolution image of a transmission electron microscope .
in addition watanabe _
et al_.@xcite also described that it is not because of the surface effect by examining the stability of the ic phase in the bulk .
if this compound can be explained by the annni model , the existence of the ic phase is only possible in three dimensions .
therefore , we predict that the appearance of the ic phase in @xmath3 is the effect of a bulk . in this work ,
we have presented that the ner analysis is very effective for systems with the kt transition and/or with slow - dynamics which has been difficult by numerical methods .
furthermore , a phase transition is studied using the finite - time scaling if we know a proper quantity that can probe characteristic time , @xmath54 , or a characteristic length , @xmath55 , even though it is not a relevant order parameter .
it is considered that images of the phase transition which has been believed as standard may change in some systems .
the authors would like to thank professor fumitaka matsubara , professor kazuo sasaki , professor nobuyasu ito and dr .
yukiyasu ozeki for fruitful discussions and comments .
they also thank professor nobuyasu ito and professor yasumasa kanada for providing us with a fast random number generator rndtik .
99 s. kawano and n. achiwa , j. magn .
mater * 52 * , 464 ( 1985 ) .
elliott , phys.rev . * 124 * , 346 ( 1961 ) .
x. dai , z. xu , and d. viehland , j. am .
* 78 * , 2815 ( 1995 ) .
j. ricote , d.l .
corker , r.w .
whatmore , s.a .
impey , a.m. glazer , j. dec , and k. roleder , j. phys .
* 10 * , 1767 ( 1998 ) .
d. viehland , phys . rev .
b * 52 * , 778 ( 1995 ) .
s. watanabe and y. koyama , kotai butsuri ( solid state physics ) * 35 * , 880 ( 2000 ) ( in japanese ) .
v. massidda and c.r .
mirasso , phys .
b * 40 * , 9327 ( 1989 ) .
tranquada , b.j .
sternlieb , j.d .
axe , y. nakamura , and s. uchida , nature * 375 * , 561 ( 1995 ) .
p. bak and j. von boehm , phys .
b * 21 * , 5297 ( 1980 ) .
m. matsuda , k. katsumata , t. yokoo , s.m .
shapiro , and g. shirane , phys .
b * 54 * , r15 626 ( 1996 ) .
fong , b. keimer , j.w .
lynn , a. hayashi , and r.j .
cava , phys .
b * 59 * , 6873 ( 1999 ) .
e. m@xmath155ller - hartmann and j. zittartz , z. phys .
b * 27 * , 261 ( 1977 ) .
j. villain and p. bak , j. phys .
( paris ) * 42 * , 657 ( 1981 ) .
j. kroemer and w. pesch , j. phys .
a * 15 * , l25 ( 1982 ) .
m. d. grynberg and h. ceva , phys .
b * 36 * , 7091 ( 1987 ) .
saqi and d.s .
mckenzie , j. phys .
a * 20 * , 471 ( 1987 ) .
y. murai , k. tanaka , and t. morita , physica a * 217 * , 214 ( 1995 ) .
w. selke and m. e. fisher , z. phys .
b * 40 * , 71 ( 1980 ) .
w. selke , z. phys .
b * 43 * , 335 ( 1981 ) .
a. sato and f. matsubara , phys .
b * 60 * , 10 316 ( 1999 ) .
n. ito , physica a * 192 * , 604 ( 1993 ) .
n. ito and y.ozeki , intern .
j. mod .
1495 ( 1999 ) . n. ito , k. hukushima , k. ogawa , and y. ozeki , j. phys .
. jpn . * 69 * , 1931 ( 2000 ) .
n. ito , t. matsuhisa , and h. kitatani , j. phys .
. jpn . * 67 * , 1188 ( 1998 ) .
y. ozeki and n. ito , j. phys .
a * 31 * , 5451 ( 1998 ) .
n. ito , y. ozeki , and h. kitatani , j. phys .
* 68 * , 803 ( 1999 ) .
k. ogawa and y. ozeki , j. phys .
. jpn . *
69 * , 2808 ( 2000 ) .
y. nonomura , j. phys .
jpn . * 67 * , 5 ( 1998 ) .
y. nonomura , j. phys .
a * 31 * , 7939 ( 1998 ) .
y. ozeki , n. ito , and k. ogawa , issp supercomputer center activity report 1999 , 37 ( the university of tokyo , 2000 ) .
y. ozeki , k. ogawa , and n. ito , in preparation .
a. jaster , j. mainville , l. sch@xmath155lke , and b. zheng , j. phys .
a * 32 * , 1395 ( 1999 ) .
kosterlitz and d.j.thouless , j. phys .
c * 6 * , 1181 ( 1973 ) . h.j .
schulz , phys .
b * 28 * , 2746 ( 1983 ) .
w. selke and j.m .
yeomans , z. phys .
b * 46 * , 311 ( 1982 ) .
s. ostlund , phys .
b * 24 * , 398 ( 1981 ) .
m. suzuki , prog .
* 58 * , 1142 ( 1977 ) .
a. sadiq and k. binder , j. stat .
* 35 * , 517 ( 1984 ) . | an axial next - nearest - neighbor ising ( annni ) model is studied by using the non - equilibrium relaxation method .
we find that the incommensurate stripe phase between the ordered phase and the paramagnetic phase is negligibly narrow or may vanish in the thermodynamic limit .
the phase transition is the second - order transition if approached from the ordered phase , and it is of the kosterlitz - thouless type if approached from the paramagnetic phase .
both transition temperatures coincide with each other within the numerical errors .
the incommensurate phase which has been observed previously is a paramagnetic phase with a very long correlation length ( typically @xmath0 ) .
we could resolve this phase by treating very large systems ( @xmath1 ) , which is first made possible by employing the present method . |
one of the exciting developments in radio galaxy research in recent years has been the recognition of a restarting of nuclear activity following a quiescent period .
the usual extended twin - lobe radio morphology of radio galaxies are in some cases found to be accompanied by a second pair of lobes sharing the same central radio core ( roettiger et al .
1994 ; subrahmanyan , saripalli & hunstead 1986 ; schoenmakers et al .
the ` double - double ' structures were interpreted by these authors as radio galaxies that were experiencing a second epoch of activity .
detailed studies of some of the more striking examples of this class of radio galaxies ( e.g. lara et al .
1999 ; schoenmakers et al .
2000b ; saripalli , subrahmanyan & udayashankar 2002 ; saripalli , subrahmanyan & udayashankar 2003 ) have estimated the ages of the second activity epoch to be few times @xmath2 yr suggesting that the restarted activity is fairly recent in comparison with the ages of radio galaxies in general ( scheuer 1995 ; mack et al .
1998 ) .
the nuclear activity that manifests itself as extended twin lobe radio structures in radio galaxies is believed to be the result of gas being accreted on to a super - massive black hole located at the centre .
there is evidence that gas at the centres of active galaxies forms an accretion disk via which it gets transported to the vicinity of the schwarzschild radius of a few thousandths of a parsec .
an important component of the central region of an active galaxy is a torus of dust and gas situated at much larger scales from several hundred parsec to a few kiloparsec .
observations have targeted this region for clues to the presence of a fuel reservoir powering the active galaxies ( e.g. mazzarella et al . 1993 ; lim et al .
the fuel searched for at millimetre wavelengths is the co gas since next to molecular hydrogen , that has only weak emission lines , the co molecules are the most abundant .
the causes behind the restarting of nuclear activity are far from clear .
suggestions have been speculative and have included ` long term variability ' in central engine output ( roettiger et al .
1994 ) and accretion disk instabilities ( e.g. schoenmakers et al .
2000a ) . in this paper
we explore this question of why radio galaxies become inactive and then restart their nuclear activity after several million years .
we look at the causes from the point of view of fuel availability to the central engine . starting from the basic premise that all agn require fuel for their activity we enquire whether the inactivity is caused by an interruption to the fuelling process or a depletion of the fuel store that sustained the activity .
are the manifestations we see ( of nested , inner double lobe structures ) a result of a new accumulation of fuel after its depletion in the earlier epoch or a continuation of the old fuelling process after an interruption ?
.the sample [ cols="<,<,<,>,^,^ " , ] references : 1 : evans et al .
( 2005 ) , 2 : braine et al .
( 1997 ) , 3 : elfhag et al .
we have presented results of a search program for molecular gas in a sample of powerful radio galaxies that are thought to be undergoing a second epoch of nuclear activity .
the search was carried out using the iram 30-m telescope on pico veleta in the co j(10 ) and j(21 ) line transitions of the co molecule , a known effective tracer of molecular hydrogen .
the motivation for the search for molecular gas in these radio galaxies was to throw light on the possible causes of the restarting of nuclear activity after a period of quiescence .
one possible scenario is depletion of a previous tranche of fuel followed by recent accumulation of fuel after a period of time and another possibility is the continuation of the original fuelling process after an interruption .
we expect that the former possibility , where a recent fuel accumulation has restarted the nuclear activity in these sources would increase their detectability in co. we discuss the co detection rate for a representative sample of nine restarting radio galaxies , seven of which were observed for the first time .
we report relatively sensitive upper limits that suggest there is no indication of molecular gas mass larger than few @xmath3 @xmath4 .
abundant molecular gas content therefore does not necessarily seem to accompany the nuclear restarting phenomenon .
the restarting radio galaxies are found to be as deficient in molecular gas as the edge brightened fr - ii radio galaxies suggesting that there has been no recent accumulation of molecular gas .
the interruption to the jet production in these sources may have been a result of instabilities in the earlier fuel flow rather than due to a depletion of fuel . for comparison
, we also discuss the molecular gas properties of five normal giant radio galaxies , three of which were observed by us using sest .
these sources are thought to form the parent sample of restarting sources .
all five have only upper limits suggesting a lack of large ( @xmath5 ) molecular gas content in these largest and among the oldest of radio galaxies .
we are grateful to the referee whose comments have helped in improving the paper significantly .
khm thanks the staff of the paul wild observatory in narrabri for their warm hospitality during a work stay .
this work was supported by a bilateral science agreement between cnr and csiro ( 132.46.1/006297 ) and the distinguished visitor programme of atnf - csiro and has benefited from research funding from the european union s sixth framework programme under radionet r113ct2003 5058187 .
part of the observations were carried out with the iram 30-m telescope on pico veleta .
iram is supported by insu / cnrs ( france ) , mpg ( germany ) and ign ( spain ) .
ls is grateful for the support provided by the staff at sest during the observations .
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 . | several radio galaxies are known that show radio morphological signatures that are best interpreted as restarting of nuclear activity after a period of quiescence .
the conditions surrounding the phenomenon of nuclear recurrence are not understood . in this paper
we have attempted to address this question by examining the nuclear fuelling characteristics in a sample of restarting radio galaxies .
we have examined the detection rate for molecular gas in a representative sample of nine restarting radio galaxies , for seven of which we present new upper limits to the molecular gas mass derived from co line observations we made with the iram 30-m telescope .
we derive a low co detection rate for the relatively young restarted radio galaxies suggesting that the cessation of the nuclear activity and its subsequent restarting may be a result of instabilities in the fuelling process rather than a case of depletion of fuel followed by a recent fuel acquisition .
it appears that abundant molecular gas content at the level of few @xmath0 @xmath1 does not necessarily accompany the nuclear restarting phenomenon . for comparison
we also discuss the molecular gas properties of five normal giant radio galaxies , three of which we observed using sest . despite obvious signs of interactions and nuclear dust disks
none of them has been found to host significant quantities of molecular gas .
[ firstpage ] galaxies : active radio lines : galaxies galaxies : ism galaxies : nuclei |
the gottesmann - knill theorem for continuous variables ( cv ) states that quantum computing based only on components described by one- or two - mode quadratic hamiltonians , gaussian states input , and measurements on canonical variables can be efficiently simulated by a classical computation .
that means , although some of the algorithms are of fundamentally quantum nature , they do not provide any speedup over classical processes @xcite .
it has also been proved that construction of a cv universal quantum computer for transformations that are polynomial in those variables requires cubic or higher non - linear operations @xcite .
therefore , investigation of the non - gaussian transformations and states generation is crucial for a nontrivial quantum computation .
more recently it has been noted that kerr - like nonlinearities , in a variety of systems , enable high precision quantum metrology that would otherwise require entanglement to achieve it @xcite . over the last decade
ion trap experiments have led the emerging technologies of coherent quantum control , especially in quantum information theory @xcite and quantum computation @xcite .
those applications require efficient creation and precise manipulation of both electronic and motional trapped ion state .
various theoretical proposals on how to produce nonclassical arbitrary states of ion motion have been discussed . in experiment , fock number states @xcite , coherent states @xcite , vacuum squeezed states
@xcite , and schrdinger cat states @xcite have been realized . in this latter case
, the state is an entangled state of the vibrational and electronic degrees of freedom . on the contrary ,
in this paper we give a deterministic way to prepare the vibrational degree of freedom in a non - gaussian state that is not entangled with the electronic states . according to our knowledge , neither non - gaussian state (
other than an entangled cat state ) nor a superposition of more than two coherent states has been observed so far .
in the case of photons there is no practical method of non - gaussian state generation so far .
efforts have been made to explore a class of the @xmath0 non - gaussian states produced using a photon coherent state @xmath1 interacting with @xmath0 kerr nonlinearity in an optical fiber @xmath2 this class of non - gaussian states , parametrized by the unitless evolution parameter @xmath3 @xcite , is the most popular one .
the state ( [ eq : kerr ] ) is known also as the kerr state . in general , this is a highly nonclassical state and after a certain time of evolution @xmath3 in the fiber its wigner function would take negative values in the phase space @xcite .
however , the nonlinearity in a fiber , or any other experimentally achievable kerr medium , is too small , @xmath4 , to reach a highly nonlinear regime and thus produce the negativity in an experimentally reasonable time , before it is destroyed by dissipation @xcite .
although microstructured fibers seem to be more promising with @xmath5 , their length does not exceed @xmath6 with current technology .
the most known examples of the kerr state are the cat states @xmath7 corresponding to @xmath8 , which have been found useful for studies of quantum decoherence and quantum - classical boundary @xcite . the larger cat states for which the two components @xmath9 and @xmath10 are almost orthogonal ( @xmath11 ) find their application in quantum information processing @xcite and quantum computation @xcite .
recently , there has been introduced a probabilistic method of non - gaussian state generation relying on a conditional photon subtraction from a squeezed vacuum state @xcite .
such a state is a good approximation for the cat state if the amplitude is small @xmath12 .
however , it is neither practical to subtract more than two photons in the experiment nor to produce a state with @xmath13 @xcite .
moreover , the state is not pure .
another approach is suggested by recent achievements in the the polzik group @xcite based on their demonstration of teleportation of a quantum state from optical to matter degrees of freedom .
if this process was reversed , so that a non - gaussian state is either teleported or mapped from matter to light , optical kerr states might be robustly generated and thus overcome obstacles which are met with in trying to generate optical non - gaussian states directly .
of course this requires us to demonstrate a way to make non - gaussian states in matter degrees of freedom .
the proposals for transferring an arbitrary motional quantum state of an atom to a cavity field already exist @xcite .
the effort to integrate ion traps and optical fields @xcite might offer a path based on the extraordinary level of coherent quantum control one has over the vibrational degrees of freedom for trapped ions . in this paper
we discuss two efficient and deterministic methods of @xmath0 non - gaussian state generation for light using a single trapped ion .
both methods require ion motional state transfer to the cavity mode .
the first method is based on the well - known protocol @xcite for an ion finite motional superposition state generation .
it allows to produce the @xmath0 non - gaussian states with arbitrary good approximation .
we give criteria , based on wigner function comparison and its measurement precision , which quantify the error resulting from the approximation .
the second method is novel and it enables an exact non - gaussian state generation using one laser pulse only .
we point out that a wigner function measurement of ion motional state can be performed using currently available technology and already existing experimental schemas .
we also suggest a quantum metrology application , based on the work of caves and co workers @xcite .
this paper is organized as follows . in section [ sec1 ]
we show that , applying a well - known protocol , one can produce an approximated @xmath0 non - gaussian state of motion for a trapped ion .
the method relies on using a series of laser pulses to couple electronic and vibrational degrees of freedom to effect the desired state preparation for the motional degree of freedom .
we give the criteria for quantifying the extent to which the prepared state approximates the desired non - gaussian state , and discuss the technical limitations of the method . in section [ sec2 ]
we present an alternative , and novel , method for generation of an exact @xmath0 non - gaussian state .
we also discuss the range of application of this method .
we finish the paper with conclusions and a brief discussion of possible applications to quantum metrology .
an ion in a paul trap @xcite may be prepared in an arbitrary state of the form @xmath14 using the method proposed in @xcite , where @xmath15 are finite superpositions of ion motional states ( @xmath16 ) , @xmath17 is a fock state of a harmonic oscillator potential in the trap , @xmath18 and @xmath19 are the ion electronic ground and excited states respectively . parameters @xmath20 and @xmath21 are complex numbers obeying the normalization constraint @xmath22 and are set at the start of the experiment .
the method is based on applying a series of alternating laser pulses tuned , first to the carrier , and then the red sideband , transition of the trapped ion .
it works both within and beyond the lamb - dicke regime .
the ion is initially prepared in its electronic and vibronic ( motional ) ground state @xmath23 . adjusting the rabi frequencies and duration of each pulse properly , one could achieve @xmath24 equal to @xmath25 these coefficients correspond to the coefficients of a quantum non - gaussian state resulting from the unitary evolution of a self - kerr interaction ( [ eq : kerr ] ) decomposed in the fock basis , up to the normalization factor ( which results from the fact that we cut off the infinite sum in ( [ eq : kerr ] ) and take into account only the first @xmath26 terms ) .
this method produces an approximated @xmath0 non - gaussian state @xmath27 corresponding to the state reached via a kerr interaction with an arbitrary value of evolution parameter @xmath3 .
furthermore the effective value of @xmath3 can be much larger than can be achieved via unitary interaction under a realistic optical kerr interaction were the bosonic degree of freedom an optical field mode . a proper choice of the cut - off value @xmath26 enables one to approximate the state ( [ eq : kerr ] ) arbitrarily closely .
if we set @xmath28 for simplicity , the ion state is given by @xmath29 the number of required pulses for preparing the state ( [ eq : approx_kerr ] ) is equal to @xmath30 latexmath:[\[|\psi^{(m)}(\alpha,\tau)\rangle = r_m c_{m-1 } \cdot ... \cdot r_1 c_0 carrier and a red sideband laser pulse respectively
. therefore @xmath26 should be as small as possible @xcite .
the hint for existence of a good non - gaussian state approximation comes from a simple observation that only a finite amount of coefficients @xmath32 contribute to the sum ( [ eq : kerr ] ) significantly and the exact number of them is strongly @xmath33 dependent .
the coefficients , evaluated for exemplary values of the amplitude , @xmath34 and @xmath35 , are depicted on fig .
[ fig : significant_coeff_a1 ] . for a given value of an amplitude @xmath33 the choice of the cut - off value @xmath26
is based on three criteria .
all criteria rely on comparing the wigner function @xmath36 of an approximated state ( [ eq : approx_kerr ] ) with the untruncated one @xmath37 evaluated for ( [ eq : kerr ] ) .
please note that the comparison is made for @xmath38 .
we chose this particular value of the evolution parameter because for this value the original wigner function is given by a simple analytic formula , which can be computed directly .
the first criterion derives from an investigation of the approximated wigner function isolines .
the smaller the value of @xmath26 the more the function deviates from an ideal gaussian function for @xmath38 : the isolines are no longer circles .
the isolines of interest can be chosen arbitrarily .
we however chose @xmath39 , @xmath40 , @xmath41 and calculate , for each isoline , the ratio between the most and the least distant point with respect to the point @xmath42 separately .
for an ideal circle the ratio is always equal to @xmath43 . in order to take into account the interference resulting from the approximate state , which takes the values around zero
, we calculate the maximal and average ratio of the difference @xmath44 to the value of the ideal wigner function , for all points @xmath45 in a phase space .
therefore , the second criterion gives the maximal and average relative error of the approximation respectively .
the third criterion reveals the percentage of points @xmath45 in a phase space for which @xmath46 for a given precision .
we assume a precision of order of either @xmath47 or @xmath48 to be good enough , since it relates to the accuracy of both the wigner function computer visualisation using the density plots and the wigner function reconstruction using quantum tomography @xcite .
the above criteria turn out to be more subtle than the well - known fidelity @xmath49 which measures the overlap of the approximated and the original state in the phase space @xmath50 obviously , the fidelity approaches unity in the limit of @xmath51 .
we would like to point out that having the ion already prepared in state ( [ eq : approx_kerr ] ) one could measure the wigner function of its vibronic state using either the standard method of quantum tomography @xcite or the method of direct measurement developed in @xcite .
this would allow for detailed investigation of @xmath0 non - gaussian states , which has never been verified experimentally so far .
the direct wigner function measurement method is especially interesting .
it relies on the fact that the wigner function of a displaced vibronic state of an ion is related to the probability of finding the ion in the ground and the excited state .
the probabilities are measured by detecting a fluorescence signal .
such a measurement has been done for a light in a cavity with accuracy of @xmath52 @xcite .
estimation of the minimal number of laser pulses for @xmath53 generation , which approximates the original state @xmath54 for any value of the evolution parameter @xmath3 well , requires comparison of the approximated wigner functions @xmath36 evaluated for a few different cut - off values @xmath26 , with the original wigner function @xmath37 .
the selection of the best possible cut - off values is based on analysis of significant coefficients @xmath32 . for @xmath34
only the coefficients ranging from @xmath55 to @xmath56 are greater than @xmath57 .
the coefficients @xmath58 to @xmath59 are of order of @xmath60 .
therefore , we test @xmath61 .
we next investigate the isolines .
table [ tabela_a2 ] shows the ratios of the most to the least distant point , with respect to the point of @xmath62 for the isolines ( @xmath39 , @xmath40 , @xmath41 ) of the approximated wigner functions @xmath36 . the minimal value of @xmath26 for which the ratios evaluated for @xmath36 are equal to the ratios evaluated for the original gauss function using the grid ( discretized phase space ) with a step equal to @xmath63 is equal to @xmath64 .
it means that for the given grid and greater values of @xmath26 , the numerical simulations of the wigner function will not differ .
therefore , the value of @xmath65 could be regarded as an appropriate cut - off value . [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] these quantities
were evaluated for the following rabi frequencies @xmath66 , @xmath67 corresponding to carrier transition and red sideband respectively , and lamb - dicke parameter @xmath68 .
the total time of state build - up is equal to @xmath69 .
the wigner function of the approximated compass state for @xmath70 is is presented on fig .
[ fig : wigner_a2_t0.5pi_comparem ] .
we also present the wigner functions for @xmath71 and the original one for @xmath72 for comparison .
+ + + at the end of this section we estimate the maximum value of the amplitude @xmath33 for which the method works .
the maximum value seems to be @xmath73 .
this limitation does not result from the decoherence of the ion state .
both the time of single pulse duration and the total duration of state built - up remain within the coherence of the ion for the vibronic state @xmath74 .
the coherence time for the electronic state is equal to @xmath75 .
the number of laser pulses is limited due to the finite binding energy in the trap .
the cut - off value @xmath26 corresponds to the maximal ion excitation @xmath76 .
the @xmath77 is the upper limit for trapping the ion so far @xcite .
the one laser pulse method of @xmath0 non - gaussian state generation requires one carrier resonance pulse applied to the ion cooled down to a lamb - dicke regime .
the vibronic state of the ion is initially prepared in a coherent state @xmath1 .
the advantage of this method over the first one is that it allows for generation of the original kerr state ( [ eq : kerr ] ) without any approximations .
however , not all values of the evolution parameter are accessible : the pulse duration is limited due to the ion decoherence .
the interaction between an ion and a laser pulse of rabi frequency @xmath78 is governed by the following hamiltonian @xmath79 where @xmath80 is an electronic state rising operator , @xmath81 is a vibronic state annihilation operator , @xmath82 is a trapping frequency , and @xmath83 is a lamb - dicke parameter . using the expansion of the exponens function into a taylor series @xmath84 in the rotating wave approximation and in an interaction picture the hamiltonian is approximated by the first five terms of the expansion @xmath85 the terms in the first line in the above formula govern the free ion evolution .
the last term in ( [ rozwiniecie ] ) we associate with a self - kerr interaction hamiltonian , up to the electronic state operators .
therefore , the unitary evolution operator resulting from the nonlinear part of the hamiltonian is given by @xmath86 it allows for reading out the effective value of the evolution parameter @xmath87 in the further discussion we neglect the electronic state evolution since it may be made to take no part in the dynamics . to do this
one first needs to prepare the electronic state in an eigenstate of @xmath88 using a @xmath89 pulse tuned to the carrier transition , but the pulse must not excite the vibrational degree of freedom in any way .
this can be done by making the pulse propagate orthogonal to the vibrational axis .
please note that by way of contrast to the method described in the previous section this method is independent of the initial amplitude @xmath33 .
the laser pulse duration required for the kerr state generation is the same for all values of the amplitude . in other words ,
the time required for generation e.g. @xmath90 and @xmath91 is the same .
however , accessibility of the evolution parameter @xmath92 is limited instead .
the greater rabi frequency @xmath78 and lamb - dicke parameter value are the shorter pulse duration @xmath93 must be to obtain @xmath92 . on the other hand , the ion has to remain within the lamb - dicke regime in order to hold the expansion ( [ rozwiniecie ] ) true .
figure [ fig : one_pulse_duration ] shows the time of the pulse duration assuming its rabi frequency @xmath94 and the lamb - dicke parameters @xmath95 , @xmath96 , @xmath68 , and @xmath97 .
for the first two lamb - dicke parameters it is less than @xmath98 , which is within the coherence time of vibronic ion state ( @xmath99 ) , to obtain @xmath100 and therefore produce the cat state .
for the two remaining @xmath83 values the duration of the pulse is of order of tenth of second .
the duration of the pulse required for @xmath101 generation is equal to @xmath102 for @xmath95 and @xmath103 for @xmath96 ( @xmath104 for @xmath68 and @xmath105 for @xmath97 ) .
for @xmath106 the first negativities in the wigner function become visible @xcite .
this method allows for obtaining also the other coherent state superpositions easily .
for example , for @xmath107 one achieves superposition of six states and if @xmath108 one achieves superposition of four states ( the compass state ) @xcite . for @xmath96 and @xmath109
the pulse duration is equal to @xmath110 , and for @xmath111 it is equal to @xmath112 .
these values are experimentally feasible .
+ the one laser pulse method of @xmath0 non - gaussian state generation is limited due to introduced cut - off in the expansion ( [ eq : rozwiniecie ] ) .
it is valid only if we take the expansion for a small parameter . to estimate the small parameter we approximate the quantum operators by the classical amplitudes @xmath113 @xmath114 and read out the small parameter in the expansion to be @xmath115 .
the above series is decreasing if @xmath116 .
this constraint is enough to ensure that the contribution from the higher order terms is negligible .
for example , for @xmath35 and @xmath95 , obeying @xmath117 ( @xmath118 and @xmath119 ) the coefficients @xmath120 are equal to : @xmath41 ( @xmath121 ) for @xmath122 , @xmath123 ( @xmath124 ) for @xmath125 , @xmath126 ( @xmath127 ) for @xmath128 . for coefficients obeying @xmath119
the sixth order coefficient is of two orders of magnitude smaller than the forth order coefficient .
we now show how the cat - state weak force detection protocol proposed in munro et al @xcite can be implemented in an ion trap given the ability to engineer an approximate cat state as we have described .
it must be admitted that there is no compelling case to use the vibrational motion of a trapped ion as a weak force detector .
however , the method we propose here would be a nice demonstration of how non - gaussian states can beat the standard quantum limit for weak force detection .
suppose one was able to prepare the vibrational degree of freedom in the non - gaussian state @xmath129 with the amplitude @xmath33 real .
when a weak force acts it can be described by the action of the unitary displacement operator @xmath130 acting on the initial state @xmath131 to give the output state @xmath132 . using the result that @xmath133 we find that for @xmath134 , and @xmath33 real that @xmath135 where the even and odd parity states are given by @xmath136 in other words , the weak force is well approximated by a rotation in the two dimensional parity subspace . as in @xcite
it then follows that the minimum detectable force is then given by @xmath137 which beats the standard quantum limit by a factor of @xmath138 .
ion trap provide a simple way to reach this lower bound .
suppose that after the weak force has acted , the total vibrational and electronic state is given by eq .
( [ roatated ] ) .
the first step is to make a @xmath89 rotation of the electronic state , followed by the conditional rotation @xmath139 which can easily be done @xcite .
finally another @xmath89 rotation gives the state @xmath140 the electronic state can now be readout , and the probability to find the ion in , say , the excited state is @xmath141 sampling this distribution gives the estimation for the force with a minimum detectable force that is inversely proportional to @xmath33 .
for example , the minimum force required to shift the interference distribution by one fringe is @xmath142 .
in the paper we have presented two experimentally feasible methods of @xmath0 non - gaussian states generation for light using a trapped ion .
the first method is based on a well - known protocol ( a series of laser pulses ) and allows for generation of an arbitrarily well approximated non - gaussian state . the approximation is quantified by three criteria based on wigner function analysis .
this method is only limited by technical parameters of the trap : the binding energy .
the second method enables an exact non - gaussian state generation using one laser pulse .
it is limited by the decoherence time of the motional ion state . however , adjusting the laser pulse and trap parameters properly one is able to produce a cat state .
based on the proposed protocols @xcite we believe that non - gaussian state transfer from ion motion to a light beam is possible in a foreseeable future .
it will enable application of a nontrivial quantum computing protocols based for example on coherent states .
this is also a step towards exploring so far unknown branch of quantum optics such as non - gaussian states .
m. s. acknowledges support by the alexander von humboldt foundation and by men grant no .
n202 021 32/0700 .
m. s. also thanks i. cirac for the discussion .
m. keller , b. lange , k. hayasaka , w. lange , and h. walther , appl .
b * 76 * , 125 ( 2003 ) ; a. b. mundt , a. kreuter , c. russo , c. becher , d. leibfried , j. eschner , f. schmidt - kaler , and r. blatt , appl . phys .
b * 76 * , 117 ( 2003 ) . | according to the gottesmann - knill theorem the non - gaussian states are necessary component for a nontrivial quantum computation .
we show two efficient and deterministic methods of @xmath0 non - gaussian state generation for a cavity mode using a single trapped ion . both require ion motional state transfer to the cavity field .
the methods are experimentally feasible .
the first is based on the well - known protocol for an ion finite motional superposition state generation .
it allows for an arbitrary good approximation of @xmath0 non - gaussian states .
we give criteria based on the wigner function which quantify the error resulting from the approximation .
the second and novel method enables an exact non - gaussian state generation using one laser pulse only . |
the apparent metallic - like temperature dependence of the resistivity is observed in various two - dimensional ( 2d ) carrier systems @xcite and remains a focus of a broad interest because it challenges the conventional theory of the metallic conduction .
the effect manifests itself in a strong drop of the resistivity over a limited range of temperatures , @xmath0 , from a high temperature value ( call it ` drude ' value ) @xmath1 to a low temperature one @xmath2 , here @xmath3 is the fermi temperature . upon lowering the temperature further , a strong ` metallic ' drop in @xmath4
was found to cross over to the conventional weak localization type dependence @xcite .
recently , it was demonstrated @xcite that the ` metallic ' drop is not related to quantum interference and in this sense should have a semiclassical origin .
magnetic field applied in the plane of the 2d system causes a dramatic increase in resistivity @xcite .
it was proven experimentally @xcite that the magnetoresistance ( mr ) in si - mos samples is mainly caused by spin - effects , though certain contribution of orbital effects @xcite is noticeable at very large fields @xcite , bigger than the field of the spin polarization .
the two effects , _ strong metallic drop in resistivity _ and _ parallel field magnetoresistance _ remain _
puzzling_. in this paper we demonstrate that ( i ) the strength of both effects is well matched and ( ii ) the action of magnetic field on conduction is similar to that of disorder and to some extent to that of temperature
. in particular , in the presence of in - plane magnetic field the strong temperature dependence of the resistivity for a high mobility sample transforms into a weaker one and shifts to higher temperatures ; both these features are typical for low mobility samples .
next , ( iii ) the effect of the magnetic field on the resistivity is continuous and shows no signatures of a threshold . finally , ( iv ) increase in either temperature or parallel magnetic field restores the same high - temperature ( ` drude ' ) resistivity value while transport remains non - activated .
matching of the actions of the above controlling parameters sets substantial constraints on the choice of theoretical models .
we performed measurements on a number of different mobility ( 100 ) @xmath5si - mos samples , because this system demonstrates the most dramatic appearance of the discussed effects .
table [ samples ] below shows the relevant parameters for the four most intensively studied samples whose mobilities differ by a factor of 25 ; parameters for other samples were reported in refs .
@xcite .
.relevant parameters of the five samples .
@xmath6 [ m@xmath7/vs ] is the peak mobility at @xmath8k , @xmath9 is in units of @xmath10 , and @xmath11 is in @xmath12@xmath13 . [
cols="^,^,^,^",options="header " , ] for the curve _ 3 _ , the fit is not perfect because it oscillates vs temperature , which is beyond the frameworks of the model .
the only steep feature may be found in the field dependence of @xmath14 at the transition from insulating to metallic behavior ; this however may be caused by a minor mismatching of the oversimplified model curves in the metallic and insulating ranges .
the ` activation energy ' @xmath15 depends on the choice of the model and on average is 2 - 3 times larger than the values estimated from fig .
the right columns in table 2 ) .
this discrepancy is presumably caused by the admixture to the data of a non - exponential temperature dependence which is essentially strong for the considered sample and which can not be separated at low temperatures from the exponential one .
for example , we were able to achieve a reasonable fit of the data with eq .
( 2 ) , using the _ calculated _ in table 2 values @xmath16 when we modeled the ` tilted separatrix ' in eq .
( [ tilted ] ) by a power low factor , @xmath17 , rather than by the polynomial factor . in ref .
@xcite it was stated that `` the effect of magnetic field can not be ascribed solely to a field - induced change in the critical electron density '' .
we verified this possibility and arrive at the opposite conclusion .
the success of our fitting confirms that the action of the parallel field , at least , to the first approximation consists in progressive increase of disorder .
this is described by increasing the @xmath18-independent scattering rate @xmath19 , decreasing the magnitude of the resistivity drop , @xmath20 , and increasing the critical carrier density @xmath21 ; the three parameters are not independent and within the same material system may be reduced only to the single parameter , e.g. @xmath21 .
to summarize , the results reported in this paper demonstrate that the effect of the parallel magnetic field on resistivity in high mobility si - mos samples ( though via zeeman coupling ) is similar to that of disorder and , to some extent , to that of temperature . in other words , _ parallel magnetic field increases disorder_. the temperature dependence of resistivity for various disorder and magnetic fields may be reduced to the dependences of the ` critical ' density @xmath11 on magnetic field and on disorder ; the changes in the resistivity drop with temperature , @xmath22 , caused by disorder and magnetic field may be also mapped onto each other .
we find a similarity between the action of the parallel field and temperature for the region of carrier density and field where conduction remains ` non - activated ' .
these findings set constraints on the choice of the developed microscopic models . the analogy between the parallel field and disorder points to the existence of a sub - band of localized carriers
. such band of localized carriers should lift in energy as a function of the parallel field , gets ` undressed ' after passing through the fermi energy and cause the increase in the scattering rate .
the strong evidence for this mechanism is the variation in the hall voltage observed in the vicinity of the metal - insulator transition .
we can not exclude also the contribution of the inter spin - subband scattering , as a complementary feature to the above mechanism .
however , our shubnikov - de haas data @xcite indicates that the mobility in different spin and valley sub - bands of conducting electrons are almost equal and the intersubband scattering does not play a major role in the themperature- and parallel field- dependence of the resistance . applying the model of a field dependent disorder
, we find a qualitative explanation of the whole set of presented results .
one of the authors ( v.p . ) acknowledges hospitality of the lorenz centre for theoretical physics at leiden , where part of this work was done and reported at the workshop in june , 2000 .
the work was supported by nsf dmr-0077825 , austrian science fund ( fwf p13439 ) , intas ( 99 - 1070 ) , rfbr ( 00 - 02 - 17706 ) , the programs physics of solid state nanostructures , statistical physics , integration and ` the state support of the leading scientific schools ' .
s. v. kravchenko , g. v. kravchenko , j. e. furneaux , v. m. pudalov , and m. diorio , phys .
b * 50 * , 8039 ( 1994 ) ; s. v. kravchenko , g. e. bowler , j. e. furneaux , v. m. pudalov , and m. diorio , phys . rev .
b * 51 * , 7038 ( 1995 ) .
d. popovi , a. b. fowler , s. washburn , phys .
lett . * 79 * , 1543 ( 1997 ) .
t. coleridge , r. l. williams , y.feng , p. zawadszki , phys .
b. * 56 * , r12764 ( 1997 ) .
a. r. hamilton , m. y. simmons , m. pepper , e. h. linfield , p. h. rose , and d. a. ritchie , phys .
* 82 * , 1542 ( 1999 ) .
the magnetic field driven transition from non - activated to activated conduction regime causes a strong exponential magnetoresistance @xmath23 , where @xmath24 ; therefore , @xmath25 scales as @xmath26 . a theoretical consideration
see in ref . | we compare the effects of temperature , disorder and parallel magnetic field on the strength of the metallic - like temperature dependence of the resistivity .
we found a similarity between the effects of disorder and parallel field : the parallel field weakens the metallic - like conduction in high mobility samples and make it similar to that for low - mobility samples .
we found a smooth continuous effect of the in - plane field on conduction , without any threshold .
while conduction remains non - activated , the parallel magnetic field restores the same resistivity value as the high temperature does .
this matching sets substantial constraints on the choice of the theoretical models developed to explain the mechanism of the metallic conduction and parallel field magnetoresistance in 2d carrier systems .
we demonstrate that the data for magneto- and temperature dependence of the resistivity of si - mos samples in parallel field may be well described by a simple mechanism of the magnetic field dependent disorder . 2 |
the last decade has witnessed to significant experimental progresses on the spectrum and decay products of the hadrons containing heavy quarks .
these progresses have been stimulated the theoretical interests on the spectroscopy of these baryons via various methods ( for some of them see @xcite and references therein ) .
for a better understanding of the heavy flavor physics , it is also necessary to gain deeper insight into the radiative , strong and weak decays of the baryons containing a heavy quark .
for some of the related studies , see @xcite and references therein .
the strong coupling constants are the main ingredients of the strong interactions of the heavy baryons . to improve our understanding on the strong interactions among the heavy baryons and other hadrons and gain knowledge about the nature and structure of the participated particles
, one needs the accurate determinations of these coupling constants . in the present study
, we calculate the strong coupling constants @xmath4 and @xmath5 within the framework of the qcd sum rule @xcite as one of the most powerful and applicable tools to hadron physics .
these coupling constants are relevant in the bottom and charmed mesons clouds description of the nucleon which may be used to explain exotic events observed by different collaborations . besides , in order to exactly determine the modifications in the masses , decay constants and other parameters of the @xmath1 and @xmath3 mesons in nuclear medium we should immediately consider the contributions of the baryons @xmath9}$ ] and @xmath10}$ ] in the medium produced by the interactions of @xmath1 and @xmath3 mesons with the nucleon , viz .
@xmath11 hence , we need to know the exact values of the strong coupling constants @xmath4 , @xmath5 , @xmath12 and @xmath13 entering the born term in the calculations @xcite . note that , among these couplings , we have only one approximate prediction for the strong coupling @xmath5 in the literature calculated at zero transferred momentum square taking the borel masses in the initial and final channels as the same @xcite . we shall also refer to a pioneering work @xcite , which estimates the strong coupling constant @xmath14 . here
we should also stress that our work on the calculation of the strong coupling constants @xmath12 and @xmath13 is in progress .
the layout of this article is as follows .
the next section presents the details of the calculations of the strong coupling constants under consideration . in section 3
, we numerically analyze the sum rules obtained and discuss the results .
the purpose of the present section is to give the details of the calculations of the coupling form factors @xmath15 and @xmath16 .
the values of these form factors at @xmath17}^2 $ ] give the strong coupling constants among the participating particles . to fulfill this aim ,
the starting point is the usage of the following three - point correlation function : @xmath18}(0)~ \bar{j}_{\lambda_b[\lambda_c]}(x)\right)|0{\rangle},\end{aligned}\ ] ] where @xmath19 denotes the time ordering operator and @xmath20 is transferred momentum .
the three - point correlation function contains interpolating currents that can be written in terms of the quark field operators as : @xmath21}(x)&=&\varepsilon_{abc}u^{a^t}(x)c\gamma_5d^{b}(x)b[c]^{c}(x ) , \nonumber \\
j_{n}(y)&=&\varepsilon_{ijk}\big(u^{i^t}(y)c\gamma_{\mu}u^{j}(y)\big)\gamma_{5}\gamma_{\mu}d^{k}(y ) , \nonumber
\\ j_{b[d]}(0)&=&\bar{u}(0)\gamma_5b[c](0),\end{aligned}\ ] ] where @xmath22 is the charge conjugation operator .
the calculation of the three - point correlation function is made via following two different ways . in the first way , which is called as hadronic side , one calculates it in terms of the hadronic degrees of freedom . in the second way , which is called as ope side , it is calculated in terms of quark and gluon degrees of freedom using the operator product expansion in deep euclidean region .
these two sides are then matched to obtain the qcd sum rules for the coupling form factors .
we apply a double borel transformation with respect to the variables @xmath23 and @xmath24 to both sides to suppress the contributions of the higher states and continuum .
the calculation of the hadronic side of the correlation function requires its saturation with complete sets of appropriate @xmath25 $ ] , @xmath26 $ ] and @xmath27 hadronic states having the same quantum numbers as their interpolating currents .
this step is followed by performing the four - integrals over @xmath28 and @xmath29 , which leads to @xmath30}\mid b[d](q)\rangle \langle \lambda_{b}[\lambda_{c}](p ) \mid \bar{j}_{\lambda_{b}[\lambda_{c}]}\mid 0\rangle } { ( p^2-m_{\lambda_{b}[\lambda_{c}]}^2)(p^{\prime^2}-m_{n}^2)(q^2-m_{b[d]}^2 ) } \nonumber \\ & \times&\langle n(p^{\prime})b[d](q)\mid \lambda_{b}[\lambda_{c}](p)\rangle+\cdots~,\end{aligned}\ ] ] where @xmath31 represents the contributions coming from the higher states and continuum .
the matrix elements in this equation are parameterized as @xmath32}\mid 0\rangle&=&\lambda_{\lambda_b[\lambda_c]}\bar{u}_{\lambda_b[\lambda_c]}(p , s ) , \nonumber \\
\langle 0 \mid j_{b[d]}\mid b[d](q)\rangle&=&i\frac{m_{b[d]}^2f_{b[d]}}{m_u+m_{b[c ] } } , \nonumber \\
\langle n(p^{\prime})b[d](q)\mid \lambda_b[\lambda_c](p)\rangle&=&g_{\lambda_bnb[\lambda_c nd]}\bar{u}_n(p^{\prime},s^{\prime})i \gamma_5 u_{\lambda_b[\lambda_c]}(p , s),\end{aligned}\ ] ] where @xmath33 and @xmath34}$ ] are the residues ; and @xmath35 and @xmath36}$ ] are the spinors for the nucleon and @xmath37 $ ] baryon , respectively . in the above equations , @xmath38}$ ] is the leptonic decay constant of @xmath26 $ ] meson and @xmath39}$ ] is the strong coupling form factor among @xmath37 $ ] , @xmath27 and @xmath26 $ ] particles . the use of eqs .
( [ matriselement1 ] ) in eq .
( [ physide1 ] ) is followed by summing over the spins of the @xmath27 and @xmath37 $ ] baryons , i.e. @xmath40}(p , s)\bar{u}_{\lambda_b[\lambda_c]}(p , s)&= & \not\!p+m_{\lambda_b[\lambda_c]}~.\end{aligned}\ ] ] as a result , we have @xmath41}^2f_{b[d]}}{m_{b[c]}+m_u}\frac{\lambda_n\lambda_{\lambda_b[\lambda_c ] } g_{\lambda_bnb[\lambda_c nd]}}{(p^2-m_{\lambda_b[\lambda_c]}^2)(p^{\prime^2}-m_n^2)(q^2-m_{b[d]}^2 ) } \nonumber \\ & \times&\big\{(m_nm_{\lambda_b[\lambda_c]}- m_{\lambda_b[\lambda_c]}^2)\gamma_5+(m_{\lambda_b[\lambda_c]}-m_n)\not\!p\gamma_5+\not\!q\not\!p\gamma_5-m_{\lambda_b[\lambda_c ] } \not\!q\gamma_5\big\}\nonumber\\ & + & \cdots~.\end{aligned}\ ] ] the final form of the hadronic side of the correlation function is obtained after the application of the double borel transformation with respect to the initial and final momenta squared , viz .
@xmath42}^2f_{b[d]}}{m_{b[c]}+m_u}\frac{\lambda_n\lambda_{\lambda_b[\lambda_c ] } g_{\lambda_bnb[\lambda_cnd]}}{(q^2-m_{b[d]}^2)}e^{-\frac{m_{\lambda_b[\lambda_c]}^2}{m^2 } } e^{-\frac{m_n^2}{m^{\prime^2 } } } \nonumber \\ & \times&\big\{(m_nm_{\lambda_b[\lambda_c]}- m_{\lambda_b[\lambda_c]}^2)\gamma_5+(m_{\lambda_b[\lambda_c]}-m_n)\not\!p\gamma_5+\not\!q\not\!p\gamma_5-m_{\lambda_b[\lambda_c ] } \not\!q\gamma_5\big\}\nonumber\\&+&\cdots~,\end{aligned}\ ] ] where @xmath43 and @xmath44 are borel mass parameters .
the ope side of the correlation function is calculated in deep euclidean region , where @xmath45 and @xmath46 .
to proceed , the explicit expressions of the interpolating currents are inserted into the correlation function in eq .
( [ correlationfunc1 ] ) . after contracting out all quark pairs via wick s theorem
we get @xmath47}^{h\ell}(-x ) \nonumber \\ & -&\gamma_5\gamma_{\mu}s^{cj}_{d}(y - x)\gamma_{5}cs_{u}^{ai^t}(y - x)c\gamma_{\mu}s^{bh}_{u}(y ) \gamma_{5}s_{b[c]}^{h\ell}(-x ) \bigg\}~,\end{aligned}\ ] ] where @xmath48}(x)$ ] represents the heavy quark propagator which is given by @xcite @xmath49}^{i\ell}(x)&=&\frac{i}{(2\pi)^4}\int d^4k e^{-ik \cdot x } \left\ { \frac{\delta_{i\ell}}{\!\not\!{k}-m_{b[c ] } } -\frac{g_sg^{\alpha\beta}_{i\ell}}{4}\frac{\sigma_{\alpha\beta}(\!\not\!{k}+m_{b[c]})+ ( \!\not\!{k}+m_{b[c]})\sigma_{\alpha\beta}}{(k^2-m_{b[c]}^2)^2}\right.\nonumber\\ & & \left.+\frac{\pi^2}{3 } \langle \frac{\alpha_sgg}{\pi}\rangle \delta_{i\ell}m_{b[c ] } \frac{k^2+m_{b[c]}\!\not\!{k}}{(k^2-m_{b[c]}^2)^4 } + \cdots\right\ } \ , , \end{aligned}\ ] ] and @xmath50 and @xmath51 are the light quark propagators and are given by @xmath52 + \cdots \ , .\end{aligned}\ ] ] the substitution of these explicit forms of the heavy and light quark propagators into eq .
( [ correlfuncope1 ] ) is followed by the usage of the following fourier transformations in @xmath53 dimensions : @xmath54^n}&=&\int\frac{d^dt}{(2\pi)^d}e^{-it\cdot(y - x)}~i~(-1)^{n+1}~2^{d-2n}~\pi^{d/2}~ \frac{\gamma(d/2-n)}{\gamma(n)}\big(-\frac{1}{t^2}\big)^{d/2-n } , \nonumber \\
\frac{1}{[y^2]^n}&=&\int\frac{d^dt^{\prime}}{(2\pi)^d}e^{-it^{\prime}\cdot y}~i~(-1)^{n+1}~2^{d-2n}~\pi^{d/2}~ \frac{\gamma(d/2-n)}{\gamma(n)}\big(-\frac{1}{t^{\prime^2}}\big)^{d/2-n}.\end{aligned}\ ] ] then , the four-@xmath28 and four-@xmath29 integrals are performed in the sequel of the replacements @xmath55 and @xmath56 . as a result , these integrals turn into dirac delta functions which are used to take the four - integrals over @xmath57 and @xmath58 .
finally the feynman parametrization and @xmath59^{\alpha-\beta-2}},\end{aligned}\ ] ] are used to perform the remaining four - integral over @xmath60 . the correlation function in ope side is obtained in terms of different structures as @xmath61 where each @xmath62 function includes the contributions coming from both the perturbative and non - perturbative parts and can be written as @xmath63 the imaginary parts of the @xmath64 functions give the spectral densities @xmath65 appearing in the last equation , viz . @xmath66 $ ] . as examples ,
we present only the explicit forms of the spectral functions @xmath67 and @xmath68 corresponding to the dirac structure @xmath69 , which are obtained as @xmath70}m_us^{\prime^2}}{64\pi^4(q^2-m_{b[c]}^2 ) } \theta[l_1(s , s^{\prime},q^2)]+\int_{0}^{1}dx \int_{0}^{1-x}dy \frac{1}{64\pi^4u^3 } \nonumber \\ & \times&\big[2m_{b[c]}^4x^2\big(1 + 3x^2-y+6xy-4x)\big)+m_{b[c]}^3x\big(3m_d u(2x-1 ) + m_u(3 + 2x^2 \nonumber \\ & -&3y-5x-2xy)\big)+2m_{b[c]}^2x\big(s(12x^4+y^2-y-30x^3 + 36x^3y-6x+20xy \nonumber \\ & -&13xy^2 + 24x^2 -55x^2y+24x^2y^2)+q^2xy(18x-24xy+7y-12x^2 - 6 ) \nonumber \\ & + & s^{\prime}y(12x^3 + 7y-4y^2 - 27x^2 + 36x^2y+18x-43xy+24xy^2 - 3 ) \big ) \nonumber \\ & + & 2s^2u^2x\big(10x^3 + 6x-15xy+2y-16x^2 + 20x^2y\big)+2q^4x^2y^2\big(10x^2 - 7y \nonumber \\ & -&16x+20xy+6\big)+2s^{\prime^2}y^2u^2 \big(10x^2 - 3y-12x+20xy\big)-4q^2s^{\prime}xy^2\big(10x^3 \nonumber \\ & + & 9y-5y^2 - 24x^2 + 30x^2y+18x-39xy+20xy^2 - 4\big)+2suy \big(q^2x(32x^2 \nonumber \\ & -&40x^2y-20x^3 - 2y-13x+22xy+1)+s^{\prime}(20x^4 - 48x^3 + 60x^3y - y+y^2 \nonumber \\ & -&8x+27xy-18xy^2 + 36x^2 - 86x^2y+40x^2y^2)\big)+3m_{b[c]}m_u u\big(q^2x(x+2y \nonumber \\ & -&3xy-1)+sux(3x-1)+s^{\prime}u(3xy - x - y)\big ) -m_{b[c]}m_u\big(q^2x(3x^2y-3x^2 \nonumber \\ & + & 7y-4y^2 + 6x-10xy-3xy^2 - 3)-sux(3x^2-y-6x-6xy+3 ) \nonumber \\ & -&3s^{\prime}u(x^2y - x^2+y -y+x-3xy - xy^2)\big ) \big]\theta[l_2(s , s^{\prime},q^2 ) ] \bigg\},\end{aligned}\ ] ] and @xmath71}^2-q^2)}\big[2m_{b[c]}m_dm_u\langle \bar{d}d\rangle+\big(m_{b[c]}(3m_u^2 - 3m_dm_u-2s^{\prime } ) \nonumber \\ & + & m_d(4m_u^2+s - s^{\prime})+2m_us^{\prime } \big)\langle \bar{u}u\rangle\big]-\langle \alpha_s\frac{g^2}{\pi}\rangle\bigg[\frac{m_{b[c]}m_uq^2s^{\prime^2}}{192\pi^2(q^2-m_{b[c]}^2)^4 } \nonumber \\ & -&\frac{9m_{b[c]}s^{\prime}(m_d+m_u)+2s^{\prime}(q^2 - 2s+5s^{\prime})}{1152\pi^2(q^2-m_{b[c]}^2)^2}- \frac{m_{b[c]}(m_d-3m_u)}{128\pi^2(q^2-m_{b[c]}^2)}\bigg ] \nonumber \\ & -&m_0 ^ 2\langle \bar{d}d\rangle \frac{3m_{b[c]}+4m_d}{96\pi^2(m_{b[c]}^2-q^2)}+m_0 ^ 2\langle \bar{u}u\rangle \frac{9m_{b[c]}+3m_d-7m_u}{96\pi^2(m_{b[c]}^2-q^2)}\bigg\}\theta[l_1(s , s^{\prime},q^2 ) ] \nonumber \\ & + & \int_{0}^{1}dx \int_{0}^{1-x}dy\bigg\{\frac{1}{8\pi^2u}\big[\langle \bar{d}d\rangle\big(m_{b[c]}-2m_{b[c]}x - m_uu+m_d(3x-1)(y+u)\big ) \nonumber \\ & + & \langle \bar{u}u\rangle\big(m_{b[c]}-2m_{b[c]}x-4m_du-2m_u(y-3xy-3xu)\big)\big ] + \langle \alpha_s\frac{g^2}{\pi}\rangle \frac{1}{96\pi^2u^3 } \nonumber \\&\times&\big[3u^2(3x-1)(y+u)+xy(1-y+x(3x+6y-4))\big ] \bigg\}\theta[l_2(s , s^{\prime},q^2 ) ] , \nonumber \\\end{aligned}\ ] ] where @xmath72}^2x+sx - sx^2+s^{\prime}y+q^2xy - sxy - s^{\prime}xy - s^{\prime}y^2 , \nonumber \\ u&=&x+y-1,\end{aligned}\ ] ] with @xmath73 $ ] being the unit - step function .
as we previously mentioned , the qcd sum rules for the strong form factors are obtained by matching the hadronic and ope sides of the correlation function . as a result , for @xmath69 structure
, we get @xmath74}(q^2)=-e^{\frac{m_{\lambda_b[\lambda_c]}^2}{m^2}}e^{\frac{m_n^2}{m^{\prime^2}}}~ \frac{(m_{b[c]}+m_u)(q^2-m_{b[d]}^2)}{m_{b[d]}^2f_{b[d]}\lambda_{\lambda_{b}[\lambda_c]}^{\dag}\lambda_n(m_nm_{\lambda_{b}[\lambda_c]}- m_{\lambda_b[\lambda_c]}^2 ) } \nonumber \\ & \times & \bigg\{\int^{s_0}_{(m_{b[c]}+m_{u}+m_{d})^2}ds\int^{s'_0}_{(2m_u+m_d)^2}ds^{\prime}e^{-\frac{s}{m^2 } } e^{-\frac{s^{\prime}}{m^{\prime^2 } } } \big[\rho_1^{pert}(s , s^{\prime},q^2)+\rho_1^{non - pert}(s , s^{\prime},q^2)\big]\bigg\}~,\nonumber\\\end{aligned}\ ] ] where @xmath75 and @xmath76 are continuum thresholds in @xmath37 $ ] and @xmath27 channels , respectively .
this section contains the numerical analysis of the obtained sum rules for the strong coupling form factors including their behavior in terms of @xmath77 . for the analysis
, we use the input parameters given in table 1 .
[ table1 ] .input parameters used in calculations . [ cols="^,^",options="header " , ] the analysis starts by the determination of the working regions for the auxiliary parameters @xmath43 , @xmath78 , @xmath75 and @xmath76 .
these parameters , which arise due to the double borel transformation and continuum subtraction , are not physical parameters so the strong coupling form factors should be almost independent of these parameters .
being related to the energy of the first excited states in the initial and final channels , the continuum thresholds are not completely arbitrary .
the continuum thresholds @xmath79 and @xmath76 are the energy squares which characterize the beginning of the continuum .
if we denote the ground states masses in the initial and final channels respectively by @xmath80 and @xmath81 , the quantities @xmath82 and @xmath83 are the energies needed to excite the particles to their first excited states with the same quantum numbers .
the @xmath82 and @xmath83 are well known for the states under consideration @xcite , where they lie roughly between @xmath84 and @xmath85 .
these values lead to the working intervals of the continuum thresholds as @xmath86~\mbox{gev$^2$}\leq s_0\leq34.5[6.7]~\mbox{gev$^2$}$ ] and @xmath87 for the strong vertex @xmath88 $ ] . in the determination of the working regions of borel parameters @xmath43 and @xmath78 ,
one considers the pole dominance as well as the convergence of the ope . in technique language ,
the upper bounds on these parameters are obtained by requiring that the pole contribution exceeds the contributions of the higher states and continuum , i.e. the condition [ nolabel ] ^_s_0ds^_s_0ds^e^- e^-_i(s , s^,q^2 ) ^_s_minds^_s_minds^e^- e^-_i(s , s^,q^2 ) < 1/3 , should be satisfied , where for each structure @xmath89 , @xmath90}+m_{u}+m_{d})^2 $ ] and @xmath91 .
the lower bounds on @xmath43 and @xmath78 are obtained by demanding that the contribution of the perturbative part exceeds the non - perturbative contributions .
these considerations lead to the windows
@xmath92~\mbox{gev$^2$}\leq m^2\leq 20[6]~\mbox{gev$^2$}$ ] and @xmath93 for the borel mass parameters corresponding to the strong vertex @xmath88 $ ] in which our results have weak dependencies on the borel mass parameters ( see figures 1 - 2 ) . as a function of the borel mass @xmath43 at average values of continuum thresholds .
* right : * @xmath94 as a function of the borel mass @xmath44 at average values of continuum thresholds .
, title="fig:",width=302 ] as a function of the borel mass @xmath43 at average values of continuum thresholds . *
right : * @xmath94 as a function of the borel mass @xmath44 at average values of continuum thresholds .
, title="fig:",width=302 ] . ,
title="fig:",width=302 ] . , title="fig:",width=302 ] as a function of @xmath95 at average values of the continuum thresholds and borel mass parameters . *
right : * @xmath96 as a function of @xmath95 at average values of the continuum thresholds and borel mass parameters . , title="fig:",width=302 ] as a function of @xmath95 at average values of the continuum thresholds and borel mass parameters . *
right : * @xmath96 as a function of @xmath95 at average values of the continuum thresholds and borel mass parameters .
, title="fig:",width=302 ] now , we use the working regions of auxiliary parameters as well as values of other input parameters to find out the dependency of the strong coupling form factors on @xmath95 .
our numerical calculations reveal that the following fit function well describes the strong coupling form factors in terms of @xmath95 : @xmath97}(q^2)=c_1\exp\big[-\frac{q^2}{c_2}\big]+c_3,\end{aligned}\ ] ] where the values of the parameters @xmath98 , @xmath99 and @xmath100 for different structures are presented in tables [ fitparam ] and [ fitparam1 ] for @xmath101 and @xmath6 , respectively . in figure 3
, we depict the dependence of the strong coupling form factors on @xmath95 at average values of the continuum thresholds and borel mass parameters for both the qcd sum rules and fitting results . from this figure
, we see that the qcd sum rules are truncated at some points at negative values of @xmath95 and the fitting results coincide well with the sum rules predictions up to these points .
the values of the strong coupling constants obtained from the fit function at @xmath102}^2 $ ] for all structures are given in table [ couplingconstant ] .
the errors appearing in the results are due to the uncertainties of the input parameters and those coming from the calculations of the working regions for the auxiliary parameters . from table 4 , we see that all structures except that @xmath69 lead to very close results .
we also depict the average of the coupling constants under consideration , obtained from all the structures used , in table 4 .
@xmath103 @xmath104 @xmath105 at this stage , we compare our result of the coupling constant @xmath5 obtained at @xmath7 with that of ref .
@xcite for the dirac structure @xmath106 . at @xmath7
, we get the result @xmath107 for this structure , which is consistent with the prediction of @xcite , i.e. , @xmath108 within the errors . to summarize , we have calculated the strong coupling constants @xmath4 and @xmath5 in the framework of the three - point qcd sum rules .
our results can be used in the bottom and charmed mesons clouds description of the nucleon which may be used to explain exotic events observed by different experiments .
the obtained results can also be used in analysis of the results of heavy ion collision experiments like @xmath109 at fair .
these results may also be used in exact determinations of the modifications in the masses , decay constants and other parameters of the @xmath1 and @xmath3 mesons in nuclear medium .
this work has been supported in part by the scientific and technological research council of turkey ( tubitak ) under the research project 114f018 .
d. w. wang , m. q. huang , phys .
d 68 , 034019 ( 2003 ) . z. g. wang , eur . phys .
j. c 54 , 231 ( 2008 ) .
f. o. duraes , m. nielsen , phys .
b 658 , 40 ( 2007 ) .
x. liu , h. x. chen , y. r. liu , a. hosaka , s. l. zhu , phys . rev .
d 77 , 014031 ( 2008 ) .
d. w. wang , m. q. huang , c. z. li , phys .
d 65 , 094036 ( 2002 ) .
n. mathur , r. lewis , r. m. woloshyn , phys .
d 66 , 014502 ( 2002 ) .
d. ebert , r. n. faustov , v. o. galkin , phys .
d 72 , 034026 ( 2005 ) .
m. karliner , h. j. lipkin , phys .
b 660 , 539 ( 2008 ) .
m. karliner , b. keren - zur , h. j. lipkin , j. l. rosner , arxiv:0706.2163 ( 2007 ) .
j. l. rosner , phys .
d 75 , 013009 ( 2007 ) . t. m. aliev , k. azizi , a. ozpineci , nucl .
b 808 , 137 ( 2009 ) . b. julia - diaz , d. o. riska , nucl . phys . a 739 , 69 ( 2004 ) .
s. scholl , h. weigel , nucl .
phys . a 735 , 163 ( 2004 ) .
a. faessler et .
al , phys . rev .
d 73 , 094013 ( 2006 ) .
b. patel , a. k. rai , p. c. vinodkumar , frascati physics series vol .
xlvi ( 2007 ) , arxiv : 0803.0221 .
c. s. an , nucl .
phys . a 797 , 131 ( 2007 ) , erratum - ibid ; a 801 , 82 ( 2008 ) . t. m. aliev , a. ozpineci , m. savci , phys . rev .
d 65 , 096004 ( 2002 ) .
t. m. aliev , k. azizi , a. ozpineci , phys .
d 79 , 056005 ( 2009 ) .
f. s. navarra , m. nielsen , phys .
b 443 , 285 ( 1998 ) .
huang , h .- x .
chen , s .- l .
zhu , phys .
d 80 , 094007 ( 2009 ) .
wang , eur .
j. a 44 , 105 ( 2010 ) ; phys .
d 81 , 036002 ( 2010 ) .
k. azizi , m. bayar , a. ozpineci , phys . rev .
d 79 , 056002 ( 2009 ) .
t. m. aliev , k. azizi , m. savci , phys .
b 696 , 220 ( 2011 ) . h .- y .
cheng , c .- k .
chua , phys .
d 75 , 014006 ( 2007 ) .
a. khodjamirian , ch .
klein , th .
mannel , y .- m .
wang , jhep 1109 , 106 ( 2011 ) .
e. hernandez , j. nieves , phys . rev .
d 84 , 057902 ( 2011 ) .
k. azizi , m. bayar , y. sarac and h. sundu , phys .
d 80 , 096007 ( 2009 ) .
t. gutsche , m. a. ivanov , j. g. korner , v. e. lyubovitskij , p. santorelli , arxiv:1410.6043 ( 2014 ) .
m. a. shifman , a. i. vainshtein , v. i. zakharov , nucl .
b147 , 385 ( 1979 ) ; nucl .
b 147 , 448 ( 1979 ) . l. j. reinders , h. rubinstein and s. yazaki , phys .
127 , 1 ( 1985 ) .
k. a. olive et al .
( particle data group ) , chin .
c , 38 , 090001 ( 2014 ) .
a. khodjamirian , `` b and d meson decay constant in qcd '' , lecture 1 at 3rd belle analysis school " , sept . 22 , 2010 , kek .
b. i. eisenstein et al .
( cleo collab . ) , phys .
d78 , 052003 ( 2008 ) .
k. azizi , n. er , eur .
j. c74 , 2904 ( 2014 ) .
b. l. ioffe , prog .
56 , 232 ( 2006 ) . v. m. belyaev , b. l. ioffe , sov .
jetp 57 , 716 ( 1982 ) ; phys .
b 287 , 176 ( 1992 ) . | we investigate the strong vertices among @xmath0 , nucleon and @xmath1 meson as well as @xmath2 , nucleon and @xmath3 meson in qcd .
in particular , we calculate the strong coupling constants @xmath4 and @xmath5 for different dirac structures entered the calculations . in the case of @xmath6 vertex , the result is compared with the only existing prediction obtained at @xmath7 . # 1#2#3 @xmath8 # 1#2#3 @xmath8 0^*0 5_5 o _ ^0 _ pacs number(s ) : 13.30.-a , 13.30.eg , 11.55.hx |
an accurate determination of the ages and distances of globular clusters ( gcs ) is an important constraint for the age of the universe , and for the theory of galaxy formation . in particular
it is important to compute very accurate relative ages to understand if there is a spread in ages among the galactic gcs or not .
the use of the stellar luminosity function ( lf ) to compute ages of gcs was first proposed by paczynski ( 1984 ) .
later on , jimenez & padoan ( 1996 ) and padoan & jimenez ( 1996 ) developed a method to determine _ the age and the distance of a gc simultaneously _ , using the lf .
the method is described in detail in padoan & jimenez ( 1996 ) , where it is concluded , on the basis of artificial data , that an uncertainty of about 0.6 gyr in the age and 0.06 mag in the distance modulus can be achieved , if the number of stars , in 1 mag - wide luminosity bins , is known with an uncertainty of 3% . in this paper
we use recent observations of the galactic globular clusters m5 ( sandquist et al .
1996 ) and m55 ( desidera & ortolani , private communication ) to apply the lf method and compute accurate ages and distance module .
these two clusters are very adequate since m55 is a metal - poor one and m5 has intermediate metallicity , so we can investigate the spread in ages ( if any ) in the formation of the gc system . in this letter
we apply for the first time the lf method to real data and we show that the method is much more superior to traditional methods ( isochrone fitting to the main sequence turn off point , @xmath4 method , or any other methods that involve the fitting of the main sequence turn off ) .
the method is superior because it allows to determine the age and the distance simultaneously and independently and because the errors in computing the age and distance are straightforward to calculate .
furthermore , it gives age determinations with sufficient accuracy to make cosmological predictions .
this first application of our lf method to real data shows that our previous theoretical predictions were correct .
in order to apply our lf method it is necessary to obtain the complete lf of the globular cluster from almost the tip of the red giant branch ( rgb ) , down to the upper main sequence .
the number of observed stars should be very large , in order to keep statistical errors sufficiently low .
recently , two lfs that fulfill these requirements have been obtained by sandquist et al .
( 1996 ) for m5 , and by desidera & ortolani ( private communication ) for m55 .
m5 is a massive globular cluster , with an average metallicity of @xmath5=-1.17\pm0.01 $ ] , according to sneden et al .
( 1992 ) , and @xmath5=-1.4 $ ] , according to zinn & west ( 1984 ) .
since it is a high altitude cluster ( @xmath6 ) , it is not seriously affected by contamination and interstellar reddening . for the metallicity of this cluster we used @xmath5=-1.3 $ ] , and adopt @xmath7 .
the completeness of the lf is discussed in detail in sandquist et al .
( 1996 ) .
m55 is among the poor - metal clusters .
the main advantage of m55 is that it is not very concentrated and therefore it is possible to resolve its core into stars .
again due to its high galactic latitude ( @xmath8 ) , interstellar reddening and contamination are negligible .
we adopt a metallicity of @xmath5=-1.9 $ ] ( briley et al .
1993 ) and @xmath7 .
the lf was provided to us by desidera & ortolani ( private communication . ) , who have performed extensive tests with artificial stars , in order to compute the completeness of the sample . in this letter
we use the data in the filter band @xmath9 for m55 and in @xmath10 for m5 .
this allows us to illustrate the robustness of the method in two very different bands .
the other colours were used in both cases to remove the horizontal and asymptotic giant branch ( agb ) stars . in the case of the horizontal branch
it is quite easy to distinguish them from the rgb .
even if this is not the case for the agb stars , the number of stars used in the lf method is so large in that bin that a small mistake in distinguishing rgb and agb stars does not contribute significantly to the errors in the final age and distance determination . in producing lfs to study both gcs
we have used @xmath11=0.4 $ ] and compute this effect in our solar - scaled tracks using the approach by chieffi , straniero & salaris ( 1991 ) .
this value of alpha enhancement is well justified by spectroscopic observations of giants in gcs ( minniti et al 1996 ) , and this enhancement is valid for a metallicity range ( @xmath5=-1.5 $ ] to @xmath12 ) .
to determine the age and the distance of a gc , two independent constraints are needed from the lf , that is to say the number of stars in two different bins .
one more bin is needed for the normalization , and a fourth bin is useful to estimate the completeness of the data , but it is not used in this work , since the completeness was previously estimated performing experiments with artificial stars ( nevertheless , we have checked that the fourth bin gives the same completeness estimation as the artificial star tests ) .
we use a bin positioned at the rgb to normalize the lf , and two bins around the sub - giant region to constrain age and distance modulus . as discussed in padoan & jimenez ( 1996 ) ,
the precise position of the two bins at the sub - giant region is extremely important .
in fact , our lf method is based on the careful optimization of those two bins .
the result of the optimization process is a contour plot of the quantity @xmath13 , on the plane @xmath14 , where : @xmath15 ^ 2 \nonumber \\ + [ n_{3,\rm th}(t)-n_{3,\rm obs}(t , m - m)]^2 \end{aligned}\ ] ] where @xmath16 is the age , @xmath17 the distance modulus , @xmath18 the normalized ratio @xmath19 , @xmath20 the number of stars in the @xmath21th observational bin and @xmath22 the corresponding theoretical ratio .
the values of @xmath22 are only functions of the age of the gc , since the shape of the theoretical lf depends only on the age ( for a given chemical composition ) , while the values of @xmath18 are also functions of the distance modulus , because the observational lf depends on the distance modulus , if the bins are defined in absolute magnitudes .
if the set of bins is not optimal , the contour plot of @xmath23 shows only open lines , which define a relation between age and distance modulus .
once the set of bins is optimized , the contour plot shows also closed lines , that define both the age and the distance modulus of the gc at the same time .
if the lines start to become closed only for @xmath24 , the degeneracy age - distance modulus is broken only if stellar counts are available with uncertainty smaller than 10% .
this is the reason why the method requires lf with a large number of stars and excellent photometry .
the optimization process , applied to m55 and m5 gave the following set of bins : @xmath25 mag \nonumber \\ m_{i , optimal , m5}=(t_{gyr}-8.0)\times0.05 \nonumber \\
+ [ 2.87,2.07,1.27,-3.13 ] mag \nonumber \end{aligned}\ ] ] the optimal bins shift by @xmath26 mag / gyr , as determined in our previous work ( padoan & jimenez 1996 ) .
this is an essential point in the attempt to obtain the contour plot of @xmath23 .
note that for m5 we could use two bins narrower than 1.0 mag , in order to improve the sensitivity of the method , thanks to the very large number of stars in the bins and to the good quality of the photometry . our results for m55 is shown in fig . 1 ,
and for m5 in fig . 2 .
in both cases , closed contour lines are obtained for @xmath27 , and the uncertainty in age and @xmath17 , obtained at any given level of @xmath23 , is comparable for the two clusters , and similar to what we previously predicted with artificial lfs ( padoan & jimenez , 1996 ) given the photometric uncertainty and the statistical uncertainty in the stellar counts ( @xmath28 , where n is the number of stars ) , we estimate a global uncertainty of 6% for the case of m55 , and of 4% for the case of m5 . entering fig . 1 and fig . 2 with @xmath29 and @xmath30 respectively ,
one gets the results listed in table 1 .
the results listed in table 1 show that the age and the distance modulus of m55 are in good agreement with previous determinations by mandushev et al .
( 1996 ) and alcaino et al .
( 1992 ) .
.the table gives the values for the age and distance modulus for m5 and m55 .
these values have been determined _ simultaneously _ using the luminosity function method described in the text . [ cols="^,^,^",options="header " , ] in the case of m5 the results agree with sanquist et al . conclusions .
they estimate in fact an age of @xmath31 gyr , for @xmath5=-1.17 $ ] , and they state that the age would be 11.5 gyr , for @xmath5=-1.4 $ ] .
we use @xmath5=-1.3 $ ] , and get an age of @xmath32 gyr .
they also estimate @xmath33 mag for @xmath5=-1.17 $ ] , and @xmath34 mag , for @xmath5=-1.4 $ ] , using the sub - dwarf fitting of the main sequence .
we get @xmath35 mag , for @xmath5=-1.3 $ ] .
note that in padoan & jimenez ( 1996 ) we estimated a variation of 0.02 mag in @xmath17 , for a shift of 0.1 in metallicity ; so we would predict @xmath36 mag , for @xmath5=-1.4 $ ] .
= 1.0 the distance modulus measured with the lf method is therefore in excellent agreement with the distance modulus determined with the sub - dwarf fitting .
the uncertainty of our estimates is very small ( @xmath37 mag ) , and no assumption on the age is required .
the lf method is superior to the main sequence turn - off ( msto ) method ( chaboyer , demarque & sarajedini 1996 ) , to determine the absolute age of globular clusters , because it is not affected by the three largest sources of theoretical uncertainty affecting the msto method , that is to say the determination of the value of the mixing length parameter , the morphology of the msto and the color-@xmath38 calibration ( see jimenez et al .
1996 for a detailed discussion of the main uncertainties in the msto method ) .
furthermore , the msto method needs to know the distance in order to determine the age , and it is unable to break this degeneracy .
the absolute ages , determined in this work for m55 and m5 , seem to indicate that the oldest gcs are not older than 14 gyr .
the lf method is a very powerful tool to investigate relative ages , since most uncertainties of stellar evolution theories are in that case avoided . from the comparison of the ages of m5 and m55
we can conclude that the age of the two gcs is not significantly different .
we conclude by remarking that most methods to determine age and distance module of gcs share two common problems : some degree of dependence of age on distance modulus ( or vice - versa ) , and a somewhat fuzzy procedure to estimate the uncertainty of the final result .
our lf method , instead , gives constraints for both age and distance modulus independently , and estimates both most probable values and uncertainties in a straightforward way .
= 1.0 on the basis of the present work , we think that very high quality data for gcs , together with the lf method , may shed new light on the problems of the age of the oldest stars in the universe and the formation of the galaxy .
we are grateful to s. desidera % s. ortolani for providing us with unpublished data of m55 .
alcaino , g. , liller , w. , alvarado , f. , wenderoth , e. , 1992 , aj , 104 , 190 briley , m.m . ,
smith , g.h . ,
hesser , j.e . ,
bell , r.a .
, 1993 , apj , 306,142 chaboyer , b. , demarque , p. , & sarajedini , a. 1996 , apj , 459 , 558 chieffi , a. , straniero , o. , salaris , m. , 1991 , in asp conf .
13 , the formation and evolution of star clusters , ed .
k. janes ( san francisco : asp ) , 219 jimenez , r. , thejll , p. , jrgensen , u.g . , macdonald , j. , pagel , b. , 1996 , mnras , 282 , 926 jimenez , r. , padoan , p. , 1996 ,
apj , 463 , l17 mandushev , g.i . ,
fahlman , g.g . , richer , h.b . ,
thompson , i.b .
, 1996 , aj , 112 , 1536 minniti , d. , petterson , r.c . , geisler , d. , claria , j.j . , 1996 , apj , 470 , 953 paczynski , b. , 1984 , apj , 284 , 670 padoan , p. & jimenez , r. 1997 , apj , 475 , 580 sandquist , e.l . ,
bolte , m. , stetson , p.b . ,
hesser , j.e . , 1996 ,
apj , 470 , 910 sneden , c. , kraft , r.p . , prosser , c.f .
, langer , g.e . , 1992 ,
aj , 104 , 2121 zinn , r. & west , m. j. 1984 , apjs , 55 , 45 | we present new age and distance determinations for the galactic globular clusters m55 and m5 , using the luminosity function method ( jimenez & padoan 1996 , padoan & jimenez 1997 ) .
we find an age of @xmath0 gyr for m55 and @xmath1 gyr for m5 .
this confirms previous results ( jimenez et al .
1996 , sandquist et al . 1996 ) and allows to conclude that * the oldest stars in the universe are not older than 14 gyr*. we also find @xmath2 for m55 , and @xmath3 for m5 .
these values agree with the ones obtained using the tip of the red giant branch ( jimenez et al .
1996 ) and the sub - dwarf fitting method ( sandquist et al . 1996 ) .
epsf globular clusters : general globular clusters : individual ( m5 , m55 ) |
well into its fifth decade ( e.g. lynden - bell 1969 , shakura & sunyaev 1973 ) , classical thin disc accretion theory has withstood the test of time reasonably intact .
while it is clear that accretion is considerably more complex than allowed for in classical disc theory ( hereafter cdt ) , there are many examples in which its computed disc spectrum , a superposition of blackbody rings over the surface of the disc , is realised with some fidelity in observations .
even when there are temporal changes in the spectra , there is generally a rather well - defined thin disc state to be found ( e.g. davis et al . 2005 ) .
a detailed theoretical examination of the disc spectrum is both useful and revealing . in its classical and simplest formulation ( lynden - bell 1969 , pringle 1981 , frank , king , & raine 2002 ) ,
the computation of the disc spectrum involves an integral that must in general be evaluated numerically . in this _ letter _ , we show that the integral in question may , however , be performed analytically in the astrophysically relevant limit @xmath5 , where @xmath6 is the photon energy , @xmath7 the boltzmann constant , and @xmath8 the maximum disc temperature .
the resulting spectra are similar to , but depart from , a wien spectrum .
the computed frequency dependence is both robust and insensitive to the precise temperature profile .
the form of the departure from a wien spectrum changes ( nearly discontinuously ) beyond a moderate threshold value of the stress at the inner disc edge .
it is this result that may prove useful to the analysis of the dynamical conditions at the inner disc edge
. the calculational technique , however , should find application under any conditions in which the disc has a well - defined temperature maximum . in principle
, the effect of several isolated maxima may be superposed , but in general the spectrum will be dominated at high @xmath0 by the highest temperature peak .
the calculations are presented in the next section , which is followed by a brief discussion of our results .
both relativistic and nonrelativistic discs are considered .
in nonrelativistic cdt , the emitted flux per unit frequency @xmath0 from a thin disc is given by the expression ( e.g. frank , king , & raine 2002 ) : [ fnu ] f_= 4h^3ic^2 r^2 ^_r _ * r dr- 1 where @xmath9 is planck s constant , @xmath10 the speed of light , @xmath11 the distance to the source , @xmath12 the inclination angle ( from a face - on orientation ) , @xmath13 the inner disc radius , and @xmath14 the temperature profile .
the integral is over all radii @xmath15 and formally extends to infinity .
the time - steady surface temperature profile of a thin keplerian disc is given by ( pringle 1981 , balbus & hawley 1998 ) : [ t1 ] t^4(r ) = 3gmm8r^3 where @xmath16 is gravitational constant , @xmath17 the stefan - boltzmann constant , @xmath18 the central mass , @xmath19 the steady mass accretion rate .
this result is calculated by assuming that the stress vanishes at the inner edge @xmath20 . but
@xmath13 might equally well be regarded as an integration constant ( arising from angular momentum flux conservation ) whose value is determined by specifying the stress at some radius . to avoid confusion ,
we will write [ t2 ] t^4(r ) = 3gmm8r^3 where @xmath21 is an arbitrary constant with dimensions of a length whose value we leave unspecified .
the choice @xmath22 corresponds to what we shall view as the special case of the stress vanishing at the inner edge of the disc .
other notational conventions we will use are ^2 = gmr^3 , for the keplerian angular velocity ( s@xmath23 ) , @xmath24 for the height - integrated disc surface density , and the characteristic temperatures [ t0 ] t_0 ^ 4 = 3gmm8r_0 ^ 3 , t_*^4 = 3gmm8r_*^3 .
care should be taken to distinguish @xmath25 and @xmath26 , which are quite distinct .
the dominant @xmath27 component of the density - weigted velocity stress tensor ( dimensions of velocity@xmath28 ) is denoted @xmath29 .
the temperature reaches a maximum , @xmath8 , when @xmath30 ( kubota et al .
this is near the inner edge if @xmath21 also is , but there may be no temperature maximum if @xmath31 .
this corresponds to the case of a significant stress at the inner edge .
`` significant '' means a value close to the local @xmath32 ( balbus & hawley 1998 ) .
we will show that the two cases of an interior temperature maximum and a boundary maxium lead to distinct observational signatures .
the condition of angular momentum conservation may quite generally be written ( balbus & hawley 1998 ) [ wrp ] -mr^22 + r^2 w_r = constant = -mr_*^2_*2 + _ * r_*^2 w where @xmath33 is the selected value of @xmath29 at @xmath20 .
( both @xmath34 and @xmath35 are evaluated at @xmath20 . )
this value of @xmath29 is to be distinguished from a fiducial characteristic value w_char = m _ * 2 _ * , which will appear ( as a normalization for @xmath33 ) in the equations .
if we now solve equation ( [ wrp ] ) for @xmath29 , we obtain w_r = m2 where [ r0 ] r_0 = r _ * ( 1- ww_char)^2 equating the energy radiated by ( each side of ) the disc to the energy extracted from the differential rotation , we have ( balbus & hawley 1998 ) : [ ss ] 2 t^4 = - w_r ddr = - m4 d^2dr , leading immediately to ( [ t2 ] ) . equation ( [ r0 ] ) tells us precisely how the @xmath21 constant is related to the imposed stress @xmath33 and inner boundary @xmath13 , and is for that reason very useful . as noted , the formal location of the temperature maximum from ( [ t2 ] ) is @xmath30 .
the question is , at what value of @xmath33 does this radius move from within the disc proper to the inner edge ?
setting @xmath36 and using ( [ r0 ] ) leads to [ wsmall ] w = w_char7 thus , when @xmath33 exceeds @xmath37 , the temperature maximum lies on the inner disc boundary , and when the stress drops below this , the temperature maximum moves off the boundary to within the disc inerior . as we shall now see
, the location of the temperature maximum makes a signficant difference to the emitted spectrum .
consider the integral [ ii ] i = ^_r _ * r dr- 1 with @xmath14 given by ( [ t2 ] ) .
when @xmath39 we may safely ignore the @xmath40 in the denominator across the entire domain of integration , since it leads only to exponentially small corrections .
thus , [ i ] i = ^_r _ * r dr the function @xmath41 has a sharp _ minimum _ at @xmath42 , which renders the integral an ideal candidate for an asymptotic expansion based on laplace s method ( bender & orszag 1978 ) . under these conditions , the entire contribution to the integral comes from a small region near @xmath42 . expanding @xmath43 and remembering the first derivative vanishes at @xmath42 : [ beta1 ] = 1kt_max + _ min(r - r_max)^22 +
... where _ min = d^2 dr^2 _ r = r_max the integral ( [ i ] ) transforms to i r_max(-h
/ kt_max ) ^_- ( -h_minx^2/2 ) dx where @xmath44 , and we set @xmath42 since only this neighborhood contributes .
extending the limits of integration introduces only exponentially small corrections .
hence , [ ias ] i r_max ( 2_minh)^1/2 ( -h / kt_max ) for the distribution ( [ t2 ] ) , @xmath45 works out to r_0 ^ 2 _
min = 3 ^ 7/2/(7 ^ 5/42 ^ 1/2kt_0 ) = 2.90426/(kt_0 ) this leads to an emission spectrum [ f1 ] f_= 2.002 4ic^2 ( hk t_0)^1/2 r_0 ^ 2r^2 ^5/2 ( -h / kt_max ) where @xmath46 . the solution for a classical zero - stress inner boundary is obtained by setting by using the inner edge of the disc @xmath13 for @xmath21 and @xmath25 for @xmath47 .
the key point is that the frequency dependence differs from a wien spectrum by a factor of @xmath48 .
[ h ] , renormalised for display , with the large @xmath0 asymptotic result for the case of vanishing stress .
solid line is from numerical evaluation ; dotted line is large @xmath0 asymptotic form ( eq . [ [ ias]]).,title="fig:",width=453 ] if the stress @xmath33 exceeds @xmath49 , the temperature maximum moves to the boundary . in that case
, equations ( [ t2 ] ) and ( [ r0 ] ) may be combined to yield the temperature at the inner disc edge , @xmath26 : t(r _ * ) = t _ * w^1/4 , ww / w_char .
another quantity of interest we shall require is the temperature gradient at @xmath20 .
this is most conveniently expressed in the form ( drd)_r = r _ * = - ( drdt)_r = r _ * = w2(7w-1 ) .
the high frequency behaviour of ( [ i ] ) is obtained by a simple integration by parts .
now the first derivative @xmath50 term dominates and one finds [ ias2 ] i ( r_*h ) ( (r _ * ) ) = r_*^2 ( kt_*h ) w^5/42(7w-1 ) ( -h/[kt _ * w^1/4 ] ) , where _ * = - 1kt_*^2(dtdr)_r = r_*. this gives a spectral flux of [ f2 ] f_= ( r_*r)^2 2ic^2 ( kt _ * ) w^5/47w-1 ^2 ( -h/[kt_*w^1/4 ] ) this is a less steep frequency dependence than is present in ( [ f1 ] ) .
the case of small stress and an interior maximum corresponds to a flatter region of high temperature , and a correspondingly larger disc area is able to contribute . a boundary maximum is more steeply cut - off as one moves outward , and less of the disc is able to contribute , reducing the high @xmath0 emission relative to the interior maximum case . figures ( 1 ) and ( 2 ) show two representative examples illustrating the region of validity of our approximation .
deviations from the leading asymptotic behavior are expected to be @xmath51 in both cases ( bender & orszag 1978 ) .
quantitatively , the agreement between asymptotic and exact integrals is excellent when @xmath52 . when the ratio is @xmath53 , the results are indistinguishable , at which point the flux is about @xmath54 of its peak value .
.,width=453 ] the axisymmetric form of the emission integral ( [ ii ] ) of classical disc theory is not preserved when relativistic physics is included .
the most important deviation is due to beaming from the portion of the disk approaching the observer . to extract the asymptotic behaviour of the spectrum , however
, we need not explicitly invoke the full machinery of general relativity . it will suffice to note that the emission integral will be of the form i = ^ 3 _ s f(s_1 , s_2 ) ds-1 where @xmath55 is the effective `` working surface , '' @xmath56 and @xmath57 surface coordinates , @xmath58 an unspecified function of coordinates ( but , importantly , _ not _ @xmath0 ) and @xmath59 an area element . if @xmath60 now has a maximum within the disc interior localised at some particular location @xmath61 , the high @xmath0 contribution is dominated by a small two - dimensional neighbourhood of this point .
even if the disc shape is globally complicated , locally it can always be represented as a flat plane with cartesian @xmath62 coordinates . near the temperature maximum @xmath8
we proceed as in subsection 2.3.1 , expanding the @xmath43 function around the coordinates @xmath63 : = 1kt_max + _ xx(x - x_max)^22 + _ yy ( y - y_max)^22 + ... where the subscript @xmath64 or @xmath65 denotes partial differentiation with the other variable held fixed .
the first order partial derivatives vanish at the maximum , as does the mixed derivative @xmath66 for the proper choice of coordinates .
the second order ( nonmixed ) derivatives are understood to be evaluated at @xmath63 .
exactly the same reasoning as before leads to the emission ( double ) integral i = ^3 f_max _ -^ _ -^ ( -h_xxx^2/2 ) ( -h_yyy^2/2 ) dx dy where @xmath67 is @xmath58 evaluated at @xmath63 .
this yields i = 2f_max h ( _ xx_yy)^1/2 ^2 the frequency dependence is exactly that of a classical axisymmetric disc with a temperature maximum at the inner boundary . on the other hand , for a disc with a localised hot spot at the inner boundary , exactly the same techniques we have been using show that the large @xmath0 asymptotic form is i ~^3/2 ( -h / kt_max )
we have shown that at large photon energies , cdt yields a mathematically simple and possibly observationally interesting difference in the frequency dependence evinced by a thin disc and a true wien spectrum . for a novikov - thorse relativistic disk ,
the difference is yet more pronounced . of greater astrophysical significance , perhaps , is our finding that a spectral changes of comparable magnitude and simplicity occur in going from zero to moderate stress at the disc s inner boundary .
this arises because , unless the stress is very small ( cf eq.[[wsmall ] ] ) , the maximum disc temperature is reached on this inner boundary ) .
whether magnetic stresses behave in this manner is currently being investigated .
( i thank j. krolik for drawing my attention to this important point . ) ] . by constrast
, a zero stress constraint always results in a temperature peak within the disc interior .
these different locations of the maxima cause measurably different high energy spectra : a @xmath68 ( @xmath3 ) power law multiplying an exponential for the case of an interior ( edge ) maximum .
( recall that a wien spectrum has a @xmath2 power law prefactor . ) for a relativistic disc , or any other disc in which the maximum is a localised hot spot , the scalings are @xmath3 ( @xmath69 ) .
these findings stand on their own , but since the question of the presence or absence of an inner stress , which is likely to be magnetic in origin ( e.g. agol & krolik 2000 ) , is a lively and contested issue ( beckwith , hawley , & krolik 2008 ) , the current results may not be devoid of practical significance . at the very least , our findings are useful benchmarks for numerical calculations of disc spectra .
real discs , onthe other hand , live in messy accretion environments , often with many spectral components .
principal sources of confusion at high frequencies include emission from a hot corona and comptonisation of soft photons .
even here , there is some utility in knowing the precise frequency dependence of a thermal disc , if only as a baseline from which to mark differences .
moreover , there are discs with minimal coronal components , and the models studied by shimura & takahara ( 1995 ; see also the discussion of davis et al .
2005 ) indicate that the effects of comptonisation can be well - modelled by replacing the planck function in equation ( [ fnu ] ) by a `` dilute blackbody '' form .
this modification would not change the frequency dependence of our formulae .
asymptotic expansions in the high frequency limit of spectral integrals , together with data that promises to be ever more accurate , will both strengthen and deepen our understanding of compact x - ray sources .
development and observational applications of these findings , as well as detailed comparisons with relativistic disc models ( novikov & thorne 1973 ) , are currently being pursued .
it is a pleasure to thank shane davis , julian krolik , and chris done for extended correspondence and important advice .
i would also like to acknowledge detailed conversations with omer blaes , mari kolehmainen , will potter ( who also kindly prepared figures 1 and 2 ) , and helpful comments from an anonymous referee .
support from the royal society in the form of a wolfson research merit award is gratefully acknowledged . | we derive simple and explicit expressions for the high frequency spectrum of a classical accretion disc . both stress - free and finite stress
inner boundaries are considered .
a classical accretion disc spectrum with a stress - free inner boundary departs from a wien spectrum at large @xmath0 , scaling as @xmath1 ( as opposed to @xmath2 ) times the usual exponential cut - off .
if there is finite stress at the inner disc boundary , the maximum disc temperature generally occurs at this edge , even at relatively modest values of the stress . in this case
, the high frequency spectrum is proportional to @xmath3 times the exponential cut - off .
if the temperature maximum is a local hot spot , instead of an axisymmetric ring , then an interior maximum produces a @xmath3 prefactor while an edge maximum yields @xmath4 .
because of beaming effects , these latter findings should pertain to a classical relativistic disc .
the asymptotics are in general robust and independent of the detailed temperature profile , provided only that the liberated free energy of differential rotation is dissipated locally , and may prove useful beyond the strict domain of classical disc theory .
as observations continue to improve with time , our findings suggest the possibility of using the high energy spectral component of black hole candidates as a signature prediction of classical theory , as well as an diagnostic of the stress at the inner regions of an accretion disc . |
soft gamma repeaters ( sgrs ) are a rare type of stellar object characterized by their transient emission of bursts of hard x - rays and soft @xmath2-rays .
bursts have been detected from three such sources from 1979 ( mazets et al .
1981 ) until early 1998 ; two are in the galactic plane ( sgr 180620 , sgr 1900 + 14 ) and one is in the large magellanic cloud ( sgr 052666 ) .
one of the first sgr bursts detected , the famous 1979 march 5 burst from sgr 052666 ( mazets et al . 1979 ) , provided a wealth of information about these sources ( thompson & duncan 1995 ) .
this flare started with a short initial spike followed by a 3 minute train of coherent 8 s pulsations .
a precise location of this burst was consistent with a young ( @xmath3 10@xmath4 year ) supernova remnant ( snr ) n49 ( cline et al . 1982 ) .
the train of pulsations and the positional coincidence with the snr indicated that the burst source is a young , magnetized neutron star .
pointed x - ray observations of sgr burst location regions have shown that each sgr has associated with it a persistent x - ray source ( murakami et al .
1994 , hurley et al .
1999a , rothschild et al . 1994 ) within or near a young snr ( kulkarni & frail 1993 , hurley et al .
1999a , cline et al .
furthermore , the persistent sources associated with the two galactic sgrs are x - ray pulsars ( kouveliotou et al
. 1998a , hurley et al
. 1999a ) which show secular spin down at a rate @xmath3 10@xmath5 s s@xmath6 ( kouveliotou et al . 1998a , 1999 ) .
as argued by kouveliotou et al .
( 1998a ) , this spin down is likely caused by magnetic dipole radiation which implies a neutron star dipole magnetic field of @xmath3 10@xmath7 gauss .
these results have provided strong observational evidence in support of the idea that sgrs are strongly magnetized neutron stars or ` magnetars ' ( duncan & thompson 1995 ) .
the majority of sgr bursts have durations less than 200 msec and are well characterized by optically thin thermal bremsstrahlung ( ottb ) spectra with temperatures @xmath0 @xmath3 30 40 kev ( kouveliotou 1995 ) . with the exception of the much more luminous march 5@xmath8 event and a similar bright flare detected recently from sgr 1900 + 14 ( hurley et al .
1999c ) , sgr bursts reach peak luminosities up to @xmath3 10@xmath9 ergs sec@xmath6 , far exceeding the eddington luminosity for a 1.4 @xmath10 neutron star .
a statistical study of bursts from sgr 180620 has shown that no correlation exists between the energy released in a burst and the time until the next burst ( laros et al .
also , it was found that both burst peak fluxes and time intervals between bursts resemble truncated log - normal and log - normal distributions , respectively ( laros et al .
1987 , hurley et al .
a differential energy distribution of events follows a gutenberg - richter power law ( 1.66 exponent ; gutenberg & richter 1956 ) with a maximum energy e@xmath11 ergs ( cheng et al .
each of these statistical properties are consistent with characteristics of earthquakes , which suggests the sgr bursts may be triggered by neutron star crustquakes ( thompson & duncan 1995 ) . here , we report the discovery of sgr 162741 , the first new sgr to be detected since 1979 .
we provide information on general burst characteristics and the persistent x - ray emission and draw comparisons to other sgrs . like the other sgrs
, this sgr is associated with a persistent x - ray source and the young snr , g337.00.1 .
during a period of intense burst activity from sgr 1900 + 14 , the burst and transient source experiment ( batse ; fishman et al .
1989 ) trigger criteria were optimized to detect sgr burst events . on 1998
june 15 , three consecutive batse triggered bursts short in duration and having soft spectra , originated from a region of the sky which was inconsistent with the three known sgr locations .
two days later , batse detected another 17 events from the same region , confirming the existence of a new sgr , sgr 162741 ( kouveliotou et al .
1998b ; source name based upon initial batse location ) . over the course of the next month and a half
, a total of 99 bursts from this source were detected with batse , 39 of which triggered the instrument .
figure 1 shows the observed burst rate as seen with batse .
the bursts from sgr 162741 last between 25 msec and 1.8 sec with most burst durations clustering near 100 msec . in figure 2
, we show some representative burst profiles .
this small sample shows the diverse temporal variability observed in bursts from this source .
the longest event ( figure 2d ) is similar to two bursts seen from sgr 1900 + 14 with respect to both spectrum and temporal structure .
these bursts are much longer than typical sgr events and have very smooth temporal profiles with abrupt @xmath2-ray emission start and end points . due to the rapid succession of bursts on june 17 and 18 , only limited
fine spectral data for a given trigger were read out from the spacecraft before the next trigger , making detailed spectral reconstruction impossible for most bursts .
of the 39 triggered bursts , fine spectral resolution data were available for only 8 events , including two very bright bursts ( figures 2c and 2d ) .
in fact , these two bright bursts have the two highest peak count rates of any extra - solar event ever observed with batse .
we fit power law , blackbody and ottb spectral models to those bursts , and find the ottb model best represents the time - integrated burst spectra of the six dim events and one of the two bright bursts ( figure 2d ) .
the measured @xmath0 values of the six dim events range between 25 35 kev ; having a weighted mean of 27 kev .
the spectral form of these six events agrees well with previous modeling of burst spectra from the other three sgrs ( see e.g. fenimore et al .
1994 ) . due to dead - time problems at the peak of the two bright bursts , we fit spectra taken at the tail of each event .
for the longest event ( figure 2d ) , we find the tail spectrum is best fit by an ottb model with a @xmath12 kev .
two spectra taken from the tail of the brightest event ( figure 2c ) are significantly harder than any other sgr 162741 burst spectrum measured with batse .
furthermore , they are not consistent with one another , which shows spectral evolution exists for this event
. the ottb and power law spectral models can not fit the first spectrum separately , but a combination of the two yields an acceptable fit . for this fit , the power law ( photon ) index @xmath13 and the ottb @xmath14 kev .
the following spectrum taken is at a lower flux level and is much harder .
we find this spectrum can be fit by a simple power law with an index @xmath15 .
evidence for similar hard burst emission from sgr 180620 was found by strohmayer & ibrahim ( 1997 ) .
a detailed discussion of this topic is beyond the scope of this @xmath16 , but it will be presented elsewhere ( woods et al .
1999a ) .
in order to estimate the peak fluxes of a larger sample of bursts , we applied the ottb model to bursts with coarse spectral resolution ( 4 channels ) and fair temporal resolution ( 64 msec ) .
we assumed a fixed @xmath0 corresponding to the measured weighted mean value for the five dim events ( 27 kev ) and allowed only the normalization ( energy flux ) to vary .
one drawback to this method is that many bursts reach their peak flux for only a short time , less than 64 msec , so these peak flux measurements will underestimate the true peak flux for some events . given the limited data availability , however , this time scale provided the largest sample of events .
figure 3 shows the cumulative peak flux distribution on the 64 msec time scale for 57 events .
the observed peak fluxes range over 3 orders of magnitude between @xmath17 and @xmath18 ergs @xmath19 sec@xmath6 .
dead - time effects for the two brightest events were excessive , so these peak flux measurements ( 1.1 and 0.51 @xmath20 ergs @xmath19 sec@xmath6 ) can be treated as lower limits . the dashed line represents a power law fit to this distribution which has an exponent @xmath21 .
no turnover is seen for this distribution out to @xmath18 ergs @xmath19 sec@xmath6 .
we also constructed a cumulative burst fluence distribution for these events , which has a slightly flatter slope of @xmath22 .
the differential fluence ( energy ) distribution then has an exponent equal to 1.5 , which agrees well with the the gutenberg - richter power law index ( 1.66 ) . using the batse triggers ,
the burst source was coarsely located ( kouveliotou et al .
1998b ) at @xmath23 = 16@xmath24 27@xmath25 and @xmath26 = 41@xmath27 ( j2000 ) with an error circle of radius 2@xmath27 .
detection of sgr events by both batse and the ulysses spacecraft provided a narrow location annulus 1.7@xmath28 wide ( hurley et al .
1998a ) . using batse earth occultation constraints
, we limited the allowable range along the annulus to 1.5@xmath27 ( woods et al .
1998 ; figure 4 ) .
a more detailed account of the localization of this sgr is reported in hurley et al .
( 1999d ) and smith et al .
( 1999 ) . in view of the association of sgrs with young snrs
, we searched the whiteoak & green ( 1996 ) catalogue of snrs near the refined error box .
a single snr , g337.00.1 ( sarma et al .
1997 ) , was found ( woods et al . 1998 ) within the 1.5@xmath29 error box . with hopes of detecting an x - ray counterpart for this sgr , a too observation of this snr
was initiated using the bepposax ( boella et al .
1997a ) narrow field instruments ( nfis ) .
two observations of snr g337.00.1 were performed on 1998 august 7 and again on september 16 .
these observations revealed a previously undetected x - ray source ( sax j1635.84736 ) at @xmath23 = 16@xmath24 35@xmath25 49.8@xmath30 and @xmath26 = 47@xmath27 35@xmath28 44@xmath31 ( j2000 ) with an error circle of radius 1@xmath28 ( 95% confidence ; figure 4 ) , consistent with the snr location .
a known source , 4u1630472 , is also seen near the edge of the field of view for each observation .
a light curve of the new source for each observation does not show any burst activity which is consistent with batse observations for those time periods ( see figure 1 )
. using the low energy concentrator spectrometer ( lecs ; parmar et al .
1997 ) and two medium energy concentrator spectrometers ( mecs ; boella et al .
1997b ) , we fit the spectrum of sax j1635.84736 from 0.1 10 kev .
the spectrum is well represented by a power law with interstellar absorption . under the assumption that the spectral form ( i.e. the power law index and hydrogen column density ) remains constant between observations
, we fit the observations simultaneously allowing only the normalization to vary between the two .
we get an acceptable fit with a reduced @xmath32 value of 0.92 for 160 degrees of freedom and find a power law ( photon ) index @xmath23 = 2.5 @xmath33 0.2 and a column density n@xmath34 = ( 7.7 @xmath35 @xmath19 .
the unabsorbed flux ( 2 10 kev ) declines between the observations ( 40.3 days ) from @xmath36 ergs @xmath19 sec@xmath6 to @xmath37 ergs @xmath19 sec@xmath6 . assuming this source is located within the snr , the distance is 11 kpc ( sarma et al . 1997 ) .
the source luminosity is then 9.7 @xmath38 10@xmath39 ergs sec@xmath6 and 7.6 @xmath38 10@xmath39 ergs sec@xmath6 for the two observations . using standard sax analysis techniques ,
source counts were extracted from the combined mecs units for sax j1635.84736 and binned at 0.5 sec time resolution .
we then performed a fast fourier transform ( fft ) of the 1998 august light curve searching frequencies from 0 1 hz and found the largest value in the power density spectrum was at 0.156 hz .
although not very significant by itself , the corresponding period falls within the tight range of observed periods ( 5 8 sec ) for the other sgrs . using the barycenter corrected time tags , we ran an epoch fold search about the period corresponding to the highest power , which revealed a marginally significant peak ( 6 @xmath40 chance probability taking into account the number of trials ; 1500 between 6.38 and 6.44 sec ) at 6.41318(3 ) sec ( jd = 2451032.5 ) .
the nearly sinusoidal pulse profile has an r.m.s .
pulse fraction of 10.0 @xmath33 2.6 % .
we performed the same analysis on the 1998 september observation , but did not find any significant peak in the power density spectrum near 0.156 hz or anywhere else between 0 1 hz .
however , given the weak signal found in august and the fact that 45% fewer source counts were recorded in the mecs units during the september observation , we would not expect to find this pulsed signal .
we propose that sax j1635.84736 is the x - ray counterpart to sgr 162741 .
there are a number of observations discussed here which support this claim .
first , the position of sax j1635.84736 is mutually consistent with the narrow error box for sgr 162741 ( hurley et al . 1999d , smith et al .
1999 ) and the snr g337.00.1 .
second , its spectrum is very similar to those found for the other sgr x - ray counterparts ( hurley et al .
1999a ) . also , this x - ray source is variable near a burst active period for sgr 162741 which has also been found for the x - ray counterpart of sgr 1900 + 14 ( hurley et al .
1999a , murakami et al .
1999 , woods et al .
finally , the marginal detection of pulsations at 6.4 sec , if confirmed , would agree well with the known spin periods of the three other sgrs which fall within a tight range of 5 8 sec ( kouveliotou et al .
1998a , hurley et al
. 1999a , mazets et al . 1979 ) . for the distance of 11 kpc
, g337.00.1 has a small diameter of @xmath3 5 pc ( sarma et al .
there are only a few snrs this small ( see case & bhattacharya 1998 ) , and a large fraction of them are very young ( e.g. tycho , kepler , cas a ) .
this suggests that g337.00.1 is also very young .
the association of sgr 162741 with g337.00.1 strengthens the connection between sgrs and young snrs .
given the distance to the snr g337.00.1 and assuming isotropic emission , the burst peak luminosities ( @xmath41 25 kev ) vary from 10@xmath42 to 10@xmath9 ergs sec@xmath6 ( figure 3 ) .
paczynski ( 1992 ) suggested that sgr 0526 - 66 may have a critical luminosity @xmath3 @xmath43 ergs sec@xmath6 .
he calculated the relation between the dipole magnetic field ( b@xmath44 ) of a neutron star and the critical luminosity ( l@xmath45 ) the magnetosphere will allow to escape in the limit where l@xmath46 l@xmath47 ( l@xmath47 is the standard eddington luminosity for a 1.4 @xmath10 neutron star ) .
this relation is given by where @xmath49 is the surface acceleration due to gravity .
for sgr 052666 , paczynski found a dipole magnetic field of @xmath3 @xmath50 gauss , which agrees with independent estimates made by duncan & thompson ( 1992 ) .
for sgr 162741 , we do not detect a turnover in the cumulative peak luminosity distribution , but we can place a lower limit on this value of @xmath51 ergs sec@xmath6 based upon the highest observed peak luminosity .
this , in turn , places a lower limit on the magnetic field of the presumed magnetar associated with sgr 162741 of b@xmath52 gauss .
this estimate may be confirmed with the definitive measurement of a pulse period and its derivative for sgr 162741 . | we report the discovery of a new soft gamma repeater ( sgr ) , sgr 162741 , and present batse observations of the burst emission and bepposax nfi observations of the probable persistent x - ray counterpart to this sgr .
all but one burst spectrum are well fit by an optically thin thermal bremsstrahlung ( ottb ) model with @xmath0 values between 25 and 35 kev .
the spectrum of the x - ray counterpart , sax j1635.84736 , is similar to that of other persistent sgr x - ray counterparts .
we find weak evidence for a periodic signal at 6.41 s in the light curve for this source . like other sgrs
, this source appears to be associated with a young supernova remnant g337.00.1 . based upon the peak luminosities of bursts observed from this sgr
, we find a lower limit on the dipole magnetic field of the neutron star b@xmath1 gauss . |
the feshbach resonance @xcite , as experimentally realized in ultracold atoms and molecules in optical lattices , has made it possible to investigate the many - body physics of multicomponent quantum degenerate mixtures of fermions and/or bosons with interspecies interactions @xcite .
operationally , a magnetic field near resonance can tune the energy splitting between different hyperfine configurations of atoms , yielding a tunable scattering amplitude with a magnitude that depends on the mismatch of the magnetic moments @xcite . in this
context , theoretical studies have introduced two primary interaction vertices : a short - ranged , one - channel density - density type interaction and a two - channel interaction that couples open - channel atoms to a molecular bound - state ( mb ) particle @xcite . recently , heteronuclear fermionic feshbach molecules composed of bosonic @xmath0na and fermionic @xmath1li @xcite and of bosonic @xmath2rb and fermionic @xmath3k @xcite have been observed experimentally and attracted the attention of theoretical studies @xcite focusing on the competition between the condensed state of unpaired bosons and the degenerate mb particles with an additional fermi surface .
it has been argued that there can be depletion transitions @xcite where one or more of the atomic or molecular species can be exhausted by driving the formation or disassociation of mb particles .
furthermore , if bosons are condensed , the spectrum can be directly diagonalized , yielding mb particles that are dressed by free atomic fermions , which form low - energy quasiparticles in a fermi - liquid theory @xcite .
additionally , the superfluidity of a paired state of a fermionic atom and a fermionic molecule , which is formed through attractive interactions mediated by the condensed and/or uncondensed bosons , has been predicted to occur @xcite . however , it is questionable as to whether such features obtained by a mean - field approach can persist when strong quantum fluctuations are present , especially for atoms trapped in one - dimensional ( 1d ) tubes .
there are many reliable analytical and numerical methods available for 1d systems @xcite .
in particular , the bosonization technique has been applied to one - channel systems with density - density type interactions , showing pairing and density - wave instabilities @xcite , polaronic phases @xcite , and competing orders @xcite .
the dominant phases exhibit variants of `` paired '' order parameters with algebraic decay or quasi - long - range order ( qlro ) @xcite .
systematic analysis has also been performed for a two - channel type model arising from atom - molecule mixture , expected for narrow resonances @xcite ; however , these investigations were primarily focused on the bosonic mb particles for two - component fermions in the context of the bec - bcs crossover .
the possibility of more complex pairing and superfluid orders that couple the open and closed fermionic channels has not been observed experimentally or discussed theoretically in detail . in this paper
, we study a general two - channel model of fermionic and bosonic atoms near a narrow feshbach resonance where bosons , fermions , and molecules can coexist .
using a renormalization - group ( rg ) method based on the bosonization formalism , we obtain a low - energy theory and attempt to clarify the ground - state phase diagram , with an emphasis on the conditions that allow the pairing of the fermionic atoms and molecules across the feshbach resonance . in doing so , we make use of the analogy to the two - chain problem of coupled tomonaga - luttinger liquids ( tll ) . the paper is organized as follows . in sec .
[ sec:2 ] , we introduce the model and examine the condition for ternary mixed phases of bosonic atoms , fermionic atoms , and fermionic molecules . in sec .
[ sec:3 ] , the ternary mixed phase is studied and possible order parameters are introduced to characterize qlro .
we determine the phase diagram for the case of an incommensurate density regime of fermions and molecules . in sec .
[ sec : rg ] , the rg method is applied to analyze the low - energy properties , and in sec .
[ sec:5 ] , the phase diagram is determined for the commensurate density regime of fermions and molecules .
lastl , in sec .
[ sec:6 ] we summarize our results in the conclusion .
it so happens that , given the mathematical form of the resonant interaction , we can draw on an rg approach applied to the spinless two - coupled chain , which is revisited in appendix [ sec : app - a ] . finally , as a supplement , we present an alternative approach based on a gauge transformation procedure in appendix [ sec : app - b ] .
our starting point is a coupled , two - channel model that describes a resonant scattering process , where free bosonic ( @xmath4 ) and fermionic ( @xmath5 ) atoms resonate into fermionic feshbach molecules ( @xmath6 ) .
the model hamiltonian is given by @xmath7 where @xmath8 } , \label{eq : hamilt - g3}%\end{aligned}\ ] ] [ eq : hamilt ] and we have set @xmath9 .
the density operators are @xmath10 , ( @xmath11 ) , where the field operators @xmath12 obey the usual commutation and anticommutation relations for bosons ( @xmath13 ) and fermions ( @xmath14 ) .
the hamiltonian @xmath15 ( @xmath11 ) consists of a kinetic energy term and an intraspecies density - density interaction term .
the coupling @xmath16 in eq .
( [ eq : hamilt - g3 ] ) induces the conversion of bosonic ( @xmath4 ) and fermionic ( @xmath5 ) atoms into fermionic mb particles ( @xmath6 ) and vice versa ( disassociation ) @xcite .
the individual particle numbers are not conserved ; instead , the total numbers of bosonic and fermionic atoms , @xmath17 , \\ \mathcal{n}_f \!\ !
& = & \!\!\ ! \int dx \left [ \rho_f(x ) + \rho_\psi(x ) \right ] , \end{aligned}\ ] ] [ eq : n ] are conserved quantities .
it follows that the masses ( @xmath18 ) and the chemical potentials ( @xmath19 ) obey the sum rules for mass conservation and chemical equilibrium , @xmath20 and the detuning parameter @xmath21 in eq .
( [ eq : hamilt - psi ] ) defines the energy splitting between the open and closed channels .
the fermionic intraspecies couplings , @xmath22 , and @xmath23 are assumed to be short - ranged , while the @xmath4 atoms interact with each other through the coupling @xmath24 . at strong repulsion ,
the boson system is described by an ordinary tonks - girardeau ( tg ) gas which behaves as free fermions . as a function of the detuning parameter @xmath21 . in this figure
, we choose @xmath25 and @xmath26 . at @xmath27 ,
the densities of fermions and molecules become equal , @xmath28 . , width=302 ] before proceeding to the many - body features of the model described by eq .
( [ eq : fullham ] ) , it is important to first establish the range of physical parameters that allow the ternary coexistence of all atoms and molecules . for simplicity
, we will consider the limit @xmath29 , with tonks - girardeau bosons [ @xmath30 with @xmath31 , and noninteracting fermions and molecules [ @xmath32 .
as noted in ref .
@xcite , we can construct a set of dimensionless parameters @xmath33 , @xmath34 , and @xmath35 , where @xmath36 is the `` fermi '' degeneracy temperature for hard - core bosons : @xmath37 , with @xmath38 being the system size .
let us introduce the average particle density @xmath39 and the corresponding normalized quantity @xmath40 .
the conditions for the conserved total numbers of atoms [ eqs .
( [ eq : n ] ) ] are expressed as @xmath41 and @xmath42 , respectively .
for hard core bosons , free fermions , and free molecules , the chemical potentials are given by @xmath43 , @xmath44 , and @xmath45 , where the `` fermi momenta '' for each species are given by @xmath46 and the fermion number @xmath47 for the case of equal masses @xmath25 .
the ternary mixed state of bosonic atoms , fermionic atoms , and feshbach molecules is realized in the region denoted by `` @xmath4+@xmath5+@xmath6 . ''
the regions denoted by `` @xmath5+@xmath6 , '' `` @xmath4+@xmath6 , '' and `` @xmath4+@xmath5 '' represent the fermion - molecule , boson - molecule , and boson - fermion binary mixed phases , respectively . along the dashed line ,
the densities of fermions and molecules become equal , @xmath48 .
, width=302 ] the particle densities can be determined from the equilibrium condition of eq .
( [ eq : eqmu ] ) . in the ternary mixed phase of @xmath4 , @xmath5 , and @xmath6 particles ( @xmath4+@xmath5+@xmath6 phase ) , the density of molecules @xmath49
is determined by the following equation : @xmath50 where @xmath51 , @xmath52 , and @xmath53 .
the densities for @xmath4 atoms and @xmath5 atoms are determined by @xmath54 and @xmath55 , respectively , and the expected @xmath21 dependence is shown in fig . [
fig : density - limit ] . notice that in the case of sufficiently strong positive detuning , the @xmath6 particle is completely depleted and only the @xmath4 and @xmath5 atoms remain .
we thus label this binary mixture the `` @xmath4+@xmath5 '' phase ; effects of possible heteroatomic interactions in this regime have been analyzed in the literature , where it has been pointed out that the excitation spectrum can have a gap and the pairing fluctuations are enhanced when the particle densities of two kinds of atoms become equal . on the other hand ,
the mb particles become stable for sufficiently strong negative detuning ; for @xmath56 @xmath57 , either @xmath4 or @xmath5 atoms coexist with the @xmath6 particles and the resulting binary mixtures are labeled `` @xmath4+@xmath6 '' and `` @xmath5+@xmath6 '' phases , respectively .
the phase diagram in terms of the detuning parameter @xmath21 and the total fermion number @xmath47 is shown in fig . [
fig : pd - limit ] , which can be contrasted with the corresponding phase diagram in the three - dimensional ( 3d ) case ( see fig . 3 in ref .
@xcite , fig . 3 in ref .
@xcite , and also fig . 1 in ref .
@xcite ) , where the bose - einstein condensate ( bec ) proliferates everywhere except for the small @xmath21 and large @xmath58 region corresponding to the @xmath5+@xmath6 phase in fig .
[ fig : pd - limit ] . in the present 1d case
, no bec can occur in any parameter region , but a `` fermi surface '' of the @xmath4 atoms can be observed instead . with this in mind , we find qualitative agreement with our phase diagram for 1d mixtures and that for 3d mixtures .
the densities of fermions and molecules become identical ( @xmath59 ) in both ternary and binary mixed phases for a particular @xmath60 , satisfying @xmath61 which is represented by the dashed line in fig .
[ fig : pd - limit ] . the analysis given in ref .
@xcite may be applied to the case @xmath62 in the @xmath5+@xmath6 phase .
however , the spectrum for the case @xmath62 in the @xmath4+@xmath5+@xmath6 phase has not yet been analyzed so far . in the following sections , we study phases realized inside the @xmath4+@xmath5+@xmath6 phase upon turning on the @xmath16 coupling .
the dominant low - energy behavior of the model defined by eqs .
( [ eq : hamilt ] ) can be studied by using a harmonic fluid representation , where the single - particle dispersion relations are linearized near the `` fermi '' points . in the problem of bec - bcs crossover in one dimension , a two - channel model of two - component fermions that dimerize into bosonic molecules has been previously analyzed by means of the bosonization method in refs .
@xcite and @xcite .
because of the different statistics of particles , the bosonization analysis of the present model will reveal different phases . in terms of the bosonic phase fields @xmath63 , the density operators can be expressed as @xcite @xmath64 } , \label{density}\ ] ] where @xmath65 is the equilibrium density and the summation is over nonzero integer @xmath66 .
the field operators for the respective particles are represented as @xcite @xmath67 + i\theta_b ( x)},\qquad \label{eq : psi_b } \\
\psi_f^{l / r } ( x ) \!\ ! & = & \!\ ! \frac{\xi_f}{\sqrt{2\pi\alpha } } e^{\mp i [ { \pi \rho^0_f x - { \phi _ f } ( x ) } ] + i{\theta_f } ( x ) } , \\
\psi_\psi^{l / r } ( x ) \!\ ! & = & \!\ ! \frac{\xi_\psi}{\sqrt{2\pi\alpha } } e^{\mp i [ { \pi \rho^0_\psi x - { \phi _ \psi}(x ) } ] + i{\theta_\psi } ( x ) } , \qquad\end{aligned}\ ] ] [ eq : psi ] where @xmath68 is a short - distance cutoff .
the field operators @xmath69 and @xmath70 ( @xmath14 ) represent the left - moving and right - moving chiral branches of fermionic particles , respectively .
the klein factors @xmath71 and @xmath72 , satisfying @xmath73 and @xmath74 , are introduced in order to retain the anticommutation relation between @xmath5 and @xmath6 particles .
@xmath75 are dual fields to @xmath63 and obey @xmath76 = i\pi\delta_{s , s'}\theta(-x + x')$ ] , where @xmath77 is the heaviside step function , i.e. , @xmath78 for @xmath79 , @xmath80 , and @xmath81 for @xmath82 . by introducing the conjugate momenta @xmath83 , a generic tll hamiltonian for each component
is expressed as @xmath84 ^ 2 + \frac{1}{k_{s}}[\partial_x \phi_{s } ( x)]^2 \right\}. \label{eq : h0}\ ] ] the parameters @xmath85 and @xmath86 are velocities and tll parameters , respectively , which depend on the precise forms of microscopic intraspecies interactions .
we will consider the general case where @xmath87 .
the noninteracting limits @xmath88 and @xmath89 correspond to @xmath90 and @xmath91 , respectively . by tuning @xmath92 ,
the system enters the tg regime at @xmath93 @xcite . for specific realizations of optical lattice systems ,
the commensurability of the bose - hubbard interaction allows the possibility of tuning into the regime @xmath94 , when @xmath95 is long ranged @xcite . after substituting the bosonized form of @xmath12 defined in eqs .
( [ eq : psi ] ) into eq .
( [ eq : hamilt - g3 ] ) and keeping only the @xmath96 term for @xmath97 , we obtain @xmath98 \nonumber \\ & & { } \qquad\qquad \times \sin[\phi_f(x ) - \phi_\psi ( x ) - \delta k_f x ] , \quad \label{eq : hg3p}\end{aligned}\ ] ] where @xmath99 and @xmath100 . in deriving eq .
( [ eq : hg3p ] ) we have discarded terms like @xmath101 , which are strongly irrelevant in the rg sense because they have spatial oscillations with the wave number @xmath102 .
furthermore , we have replaced @xmath103 with @xmath104 , because the product of the two klein factors is a constant of motion [ the identity @xmath105 implies either @xmath106 or @xmath107 , and we have chosen the former ] .
we will use the same sign convention when we derive bosonized form of order parameters . in the incommensurate case ( @xmath108 ) , the @xmath16 interaction [ eq .
( [ eq : hg3p ] ) ] is irrelevant in the rg sense .
the analysis of the previous section is then applicable , and the phase diagram therefore can be determined as in the previous section , with various depletion transitions occurring between binary and ternary mixture phases . on the other hand , eq .
( [ eq : hg3p ] ) has a dramatically different effect for the commensurate case ( @xmath109 ) , which is satisfied along the dashed line in fig .
[ fig : pd - limit ] . in this case
, sinusoidal potentials can lock a particular phase variable ( @xmath110 or @xmath111 ) , and a competition of various orders due to the phase locking must be studied by performing a rg analysis . in the context of quantum mixtures ,
composite `` pairing '' correlations have been previously introduced in the literature and will be extended here to a more comprehensive list of possible order parameters .
first , the conventional @xmath112 density - wave ( dw ) order parameters are given by @xmath113 here ( and below ) we have dropped unimportant numerical prefactors . in analogy with order parameters in the spinless two - coupled chain system [ see eq .
( [ eq:2chain - o - cdw ] ) ] , we introduce the out - of - phase dw state of @xmath5 and @xmath6 particles , @xmath114 next , the order parameters for the superfluidity ( sf ) of bosons , @xmath115-wave - paired fermions , and @xmath115-wave - paired molecules are given by @xmath116 we will also consider the @xmath115-wave - paired sf state composed of @xmath5 and @xmath6 particles , @xmath117 which is odd under the parity transformation , @xmath118 , and therefore can be classified as @xmath115-wave pairing . moreover ,
this order parameter can be identified with the interchain pairing sc@xmath119 state in the two - coupled spinless chain problem [ see eq .
( [ eq:2chain - o - scd ] ) ] , where @xmath5 and @xmath6 can be replaced by the two - chain indices . earlier work in ref .
@xcite investigated the phase diagram of interacting `` @xmath4+@xmath5 '' binary mixtures near the commensurate point @xmath120 , where the composite @xmath115-wave superfluidity @xmath121 was shown to have dominant qlro correlations . in the ternary system
studied here , similar orders can persist , @xmath122 which describes the @xmath115-wave pairing of two fermionic ( @xmath5 or @xmath6 ) particles combined with a single @xmath4 atom .
note that the two composite operators in eq .
( [ eq : obff ] ) , @xmath123 and @xmath124 , annihilate equal numbers of fermionic and bosonic atoms ( including the ones forming a molecule ) , as seen from the commutation relations @xmath125= -\mathcal{o}^{\mathrm{sf}}_{bff+b^\dagger\psi\psi}$ ] and @xmath126= -2\mathcal{o}^{\mathrm{sf}}_{bff+b^\dagger\psi\psi}$ ] .
this order parameter corresponds to the intrachain sc@xmath127 pairing in the two - coupled chain problem [ see eq .
( [ eq:2chain - o - scs ] ) ] .
in addition , we consider other composite order parameters defined by @xmath128[eq : op ] which represent the particle - hole combinations of @xmath5 and @xmath6 fermions .
these operators are composed of the products of three field operators , @xmath129 and @xmath130 , which are similar in form to the @xmath16 term of eq .
( [ eq : hamilt - g3 ] ) but asymmetrical in the @xmath131 branches . the second bosonized contribution in eq .
( [ eq : obfp1 ] ) , coming from the @xmath132 contribution in eq .
( [ eq : psi_b ] ) , can become a dominant order parameter for some parameter regime , as will be shown later .
we also note that the order parameter in eq .
( [ eq : obfp2 ] ) corresponds to the `` orbital antiferromagnetic state '' in the two - coupled chain problem , in which circulating currents flow between the two chains , if the @xmath5 and @xmath6 indices are regarded as chain indices [ see eq .
( [ eq:2chain - o - oaf ] ) ] . when @xmath133 , the @xmath16 interaction of eq .
( [ eq : hg3p ] ) oscillates in space and does not affect the low - energy spectrum .
thus we can set @xmath134 in the low - energy limit , and the system is described as a three - component tll , in which the @xmath4 , @xmath5 , and @xmath6 particles are decoupled and the correlation functions exhibit algebraic decay .
for example , the correlators for the @xmath4 particles are given by @xmath135 [ eq : cor0b ] we find that the superfluidity correlation dominates over the density - wave correlation when @xmath136 .
similarly , the correlation functions for the @xmath115-wave superfluidity and the density - wave of the @xmath5 and @xmath6 particles exhibit algebraic decay , @xmath137 [ eq : cor0f ] where @xmath14 .
the dominant correlation for fermions changes between the superfluidity and density - wave orders at @xmath138 . in fig .
[ fig : pd](a ) we show the phase diagram in the parameter space of @xmath86 ( @xmath139 ) , which is obtained by identifying the dominant qlro among those in eqs .
( [ eq : cor0b ] ) and ( [ eq : cor0f ] ) .
when @xmath140 , the effects of the sinusoidal potential ( [ eq : hg3p ] ) can be analyzed using rg techniques @xcite . apparently , the form of eq .
( [ eq : hg3p ] ) contains dual fields which do not commute @xmath141\neq 0 $ ] .
this type of interaction has been analyzed in the context of two tll chains coupled by one - particle interchain hopping @xcite , where it has been confirmed that higher - order corrections are crucial to determine the low - energy spectrum of the two tll chains @xcite .
we thus can expect that interactions generated by rg transformation should similarly be taken into account in our model . in order to properly derive the rg equations and to determine the ground - state phase diagram for the hamiltonian including the potential as eq .
( [ eq : hg3p ] ) , we have to pay special attention to the commutative properties of the phase fields , besides the klein factors . in appendix [ sec : app - a ] , we analyze the two - coupled chain system on the basis of the present bosonization scheme and verify that the correct results @xcite can be derived . in ref .
@xcite , the interchain hopping term was treated nonperturbatively and the phase diagram was determined . in appendix
[ sec : app - b ] , we analyze the present model ( [ eq : fullham ] ) using the method of ref .
@xcite and observe that the consistent results can be obtained . in this section we set @xmath142 for simplicity .
the euclidean action of the system is given by @xmath143 with @xmath144 [ eq : action ] where @xmath145 , @xmath146 , @xmath147 , and @xmath148 .
although the extra terms @xmath149 , @xmath150 , @xmath151 , @xmath152 , and @xmath153 , are absent in the original hamiltonian , they are generated through the rg process @xcite . in this paper
, we adopt the momentum - space rg method @xcite by introducing the momentum space cutoff @xmath154 .
the rg equations can be obtained by integrating out the high - momentum components @xmath155 , where @xmath156 is the reduced cutoff ( @xmath157 ) and @xmath158 with the frequency @xmath159 .
accordingly , the phase fields @xmath160 are split into two components @xmath161 @xcite , where @xmath162 is the field having components in lower momentum @xmath163 and @xmath164 has higher momentum components @xmath165 .
the free propagators for these fields are given by @xmath166 , \nonumber \\ \label{eq : hcorre}\end{aligned}\ ] ] where @xmath167 and @xmath168 is the bessel function of the first kind .
with the smooth cutoff function @xmath169 @xcite , we have the correlation functions @xmath170 ^ 2\rangle = k_s \ln(e^\gamma c\lambda ' |\bm r|/2)$ ] and @xmath171 for @xmath172 , where @xmath173 is the modified bessel function .
the constant @xmath174 is taken as @xmath175 in order to reproduce the asymptotic form @xmath176 ^ 2\rangle = k_s \ln(|\bm r|/\alpha)$ ] for @xmath177 .
the derivation of one - loop rg equations proceeds similarly to the case of the two - coupled chain system explained in appendix [ sec : app - a ] . by exploiting the commutation relation @xmath178=i\pi \delta_{s , s'}\theta(-x+x')$ ] and the normal ordering procedure for the operator - product expansion @xcite
, we eventually obtain the following one - loop rg equations : @xmath179 [ eq : rg ] where we have defined @xmath180 } , \\
a_2(\beta ) dl \!\ ! & \equiv & \!\ ! 2\beta \int_0^\infty \frac{dr}{\alpha}\frac{r^3}{\alpha^3 }
\delta g(r ) e^{-2\beta[\bar g(0 ) - \bar g(r ) ] } .
\qquad\end{aligned}\ ] ] [ eq : a ] the exponential factors in the rhs of eqs .
( [ eq : a ] ) appear as a result of normal ordering in operator - product expansions @xcite ; for example , @xmath181 \nonumber \\ & = & \!\ ! : \
! \cos\!\left [ p\phi'_{s}(\bm r_1 ) \ ! + \
! q\phi'_{s } ( \bm r_2 ) \right ] \ ! : \
, e^{- \frac{1}{2}(p^2+q^2 ) \langle { \phi'_{s}}^2\rangle - pq \langle \phi'_{s}(\bm 1)\phi'_{s}(\bm 2)\rangle } \nonumber \\ & \approx & \!\ ! : \ !
\cos\!\left [ ( p \ ! + \ !
q)\phi'_{s}(\bm r ) \right ] \ ! : e^{- \frac{1}{2}(p^2+q^2 ) \langle { \phi'_{s}}^2\rangle - pq \langle \phi'_{s}(\bm 1)\phi'_{s}(\bm 2)\rangle } \nonumber \\ & = & \!\ !
\cos\!\left [ ( \ !
p + \ ! q)\phi'_{s}(\bm r ) \right ] \ , e^{\frac{1}{2}[(p+q)^2 - ( p^2+q^2 ) ] \langle { \phi'_{s}}^2\rangle - pq \langle \phi'_{s}(\bm 1)\phi'_{s}(\bm 2)\rangle } \nonumber \\ & = & \!\ !
\cos\!\left [ ( p+q)\phi'_{s}(\bm r ) \right ] \ , e^{\frac{1}{2 } pq k [ \bar g(0 ) - \bar g ( r_{12 } ) ] } , \label{eq : ope}\end{aligned}\ ] ] where @xmath182 and @xmath183 .
we have used the short - hand notations @xmath184 and @xmath185 .
we note that @xmath186 for small @xmath187 , and @xmath188 , where @xmath189 is the euler - mascheroni constant .
one can neglect the velocity renormalization up to one - loop order .
the initial values of the rg equations are given by @xmath190 , @xmath191 , and @xmath192 .
, @xmath193 , and @xmath150 terms , and low - order contributions to @xmath193 and @xmath150 .
the dashed , sold , and double lines represent the boson , fermion , and molecule propagators , respectively .
the wavy line represents the intraspecies density - density interaction .
, width=283 ] diagrammatic representations for the @xmath194 , @xmath193 , and @xmath150 terms are shown in fig .
[ fig : vertex ] .
the @xmath193 coupling is a four - point vertex representing interactions between @xmath5 and @xmath6 particles , while the @xmath150 coupling is a six - point vertex for a two - molecule conversion from two @xmath4 and two @xmath5 particles .
low - order contributions to @xmath193 and @xmath150 are also shown in fig .
[ fig : vertex ] .
the lowest - order contribution to the @xmath193 coupling comes from the effective interaction mediated by @xmath4 atoms .
pairing between fermions ( @xmath5 ) and molecules ( @xmath6 ) induced by such boson- ( @xmath4 ) mediated interaction has been suggested in ref .
we will contrast this paper with our work in more detail later . since @xmath195\neq 0 $ ] , the phase variables @xmath196 and @xmath197 can not be locked simultaneously .
this means that there should be two distinct phases separated by a quantum phase transition , a phase where @xmath196 is locked by the @xmath193 term and a phase where @xmath197 is locked by the @xmath150 term , in addition to a three - component tll phase where none of the phase fields are locked .
here we obtain the phase diagram by comparing the scaling dimensions , which we denote by dim [ ] , of the operators for the couplings @xmath194 , @xmath193 , and @xmath150 .
we ignore renormalization of @xmath198 , @xmath199 , and @xmath200 for weak @xmath16 , because the right - hand side of eqs .
( [ eq : rg-4])@xmath201([eq : rg-6 ] ) are of order @xmath202 .
the scaling dimensions of the sinusoidal potential operators are found from eqs .
( [ eq : rg-1])@xmath201([eq : rg-3 ] ) as @xmath203\!\!&=&\!\ ! \frac{1}{4}\!\left(\frac{1}{k_b}+\frac{1}{k_f}+\frac{1}{k_\psi
} + k_f+k_\psi\right ) , \nonumber \\
! k_f+k_\psi , \\
! \frac{1}{k_b}+\frac{1}{k_f}+\frac{1}{k_\psi}. \nonumber \end{aligned}\ ] ] in the case when three inequalities , @xmath204>2 $ ] , @xmath205>2 $ ] , and @xmath206>2 $ ] , are simultaneously satisfied , all the locking potential operators are irrelevant , and consequently , we have a three - component tll phase .
this is the case for large @xmath199 and @xmath200 and small @xmath198 .
otherwise , either the coupling @xmath193 or @xmath150 becomes relevant and flows to strong coupling at low energy .
we observe from eqs .
( [ eq : rg-2 ] ) and ( [ eq : rg-3 ] ) that the condition @xmath207 defines the particular case where the scaling dimensions @xmath205 $ ] and @xmath206 $ ] become identical and the factor @xmath208 in the second terms of the right - hand side of eqs .
( [ eq : rg-2 ] ) and ( [ eq : rg-3 ] ) vanishes . thus , eq .
( [ condition ] ) determines the phase boundary between the phase where the @xmath193 operator is relevant and the phase where the @xmath150 operator is relevant . in the case where @xmath209 , the coupling @xmath193 is relevant and renormalized to strong coupling with @xmath210 .
we note that the positive @xmath193 coupling implies _ repulsive _ density - density interactions between @xmath5 and @xmath6 particles . on the other hand , in the opposite case where @xmath209 , the coupling @xmath150 is relevant and renormalized to strong coupling with @xmath211 . the resulting phase diagram is shown in fig .
[ fig : pd](b ) , for which the nature of the ground state in each phase is discussed in the next section .
in the preceding section we determined the phase boundaries in the phase diagram that admit quantum phase transitions . therefore , in a given region of relevance where a particular phase variable is locked , the properties of the resulting phase that may exhibit dominant qlro can be understood by analyzing the exponents of the order parameter correlations . for this purpose ,
the analysis based on the rg equations given by eqs.([eq : rg ] ) is not simple since the @xmath194 term contains both @xmath111 and @xmath110 ( @xmath14 ) fields .
when treating this type of term , one often encounters subtleties in determining ground - state phases , especially in the case that the @xmath194 term becomes relevant .
thus it is necessary to make transformation to a suitable basis .
we perform the following canonical transformation : @xmath212 where @xmath213 the transformation matrices @xmath214 and @xmath215 are generally nonorthogonal , but the commutation relations of @xmath216 and @xmath217 , @xmath218 = i\delta_{a , b}\delta(x - y)$ ] , are preserved as long as the relation @xmath219 is satisfied @xcite . a simplification of eq .
( [ eq : hg3p ] ) follows from the following choice of the matrices : @xmath220 substituting the phase variables @xmath221 and @xmath222 , we rewrite the cosine terms in eqs .
( [ eq : action ] ) as @xmath223 [ eq : hg3p - rotated ] we note that the phase variables @xmath224 and @xmath225 are under the influence of the @xmath193 and @xmath150 cosine potentials , respectively . in terms of the phase variables @xmath221 and @xmath222 , the tll hamiltonian ( [ eq : h0 ] ) is rewritten as @xmath226 , \nonumber\\ & & \label{eq : h0-rotated}\end{aligned}\ ] ] where @xmath227 and @xmath228 are real symmetric matrices defined by @xmath229 [ eq : matrixnm ] the order parameters introduced in sec .
[ section : op ] can now be expressed in terms of the new phase variables @xmath230 and @xmath231 .
the order parameters for the @xmath4 particles are given by @xmath232 the order parameters for the @xmath115-wave - pairing sf and out - of - phase dw states of the @xmath5 and @xmath6 particles are written as @xmath233 from which it follows that correlations of sf@xmath234 and dw@xmath234 are enhanced when the phase field @xmath224 is locked at @xmath235 .
finally , the order parameters for the composite particles are expressed as @xmath236 we see that the correlations of these order parameters are enhanced when the phase field @xmath225 is locked at @xmath237 , except for the first contribution in eq .
( [ eq : oph1 - 2 ] ) . the sinusoidal potentials of eq .
( [ eq : hg3p - rotated ] ) take on forms similar to those of the spinless two - coupled chain system @xcite ( see appendix [ sec : app - a ] ) . in the two - chain system ,
operators generated in rg transformations become relevant in the low - energy limit .
similarly , we expect that either the @xmath193 or @xmath150 term can become relevant and renormalized to strong coupling , as we have discussed below eq .
( [ condition ] ) .
the relevant @xmath210 leads to locking of the phase field @xmath224 at @xmath238 , whereas the relevant @xmath211 leads to the locking of the phase field @xmath225 at @xmath239 . when either @xmath224 or @xmath225 is locked , the remaining phase fields @xmath240 and @xmath241 ( @xmath242 ) remain gapless , and then the system is effectively described by a two - component tll and a massive sine - gordon model .
however , in contrast to the simple forms of sinusoidal potentials , the quadratic hamiltonian in eq .
( [ eq : h0-rotated ] ) is complicated by the presence of many cross terms .
one approach that we will implement here is to integrate out the massive mode @xmath243 in a manner similar to ref .
@xcite , thereby reducing the problem to a two - band system which can be exactly diagonalized . to be more precise , when @xmath244 in the rg analysis , the quantum fluctuations of the @xmath224 field are suppressed , and we can make the approximation @xmath245 . moreover ,
since the cosine potentials can be ignored for the strongly fluctuating @xmath225 field , @xmath225 can be integrated out by completing the square for @xmath246 in the quadratic hamiltonian , as described in ref .
the same approach can be used for @xmath247 .
consequently , the system can be described effectively by the two - component tl liquid with the effective low - energy hamiltonian @xmath248 where @xmath249 and @xmath250 . in the case when @xmath224 is locked ( @xmath251 ) , the renormalized coefficients are given by @xmath252 and @xmath253 ( @xmath254 ) .
similarly , when @xmath225 is locked ( @xmath255 ) , the coefficients are given by @xmath256 and @xmath257 .
the hamiltonian ( [ eq : heff ] ) can be diagonalized sequentially @xcite , yielding @xmath258 \nonumber \\ & & { } + \frac{u_2}{2\pi } \int { dx } \left [ ( \partial_x \tilde \varphi_2)^2 + ( \partial_x \tilde \vartheta_2)^2 \right ] .
\label{eq : h^eff}\end{aligned}\ ] ] the canonical transformation between the phase variables @xmath259 and @xmath260 are given by @xmath261 where the transformation matrices @xmath262 and @xmath263 are defined as @xmath264 and @xmath265 with @xmath266 and @xmath267 being diagonal matrices . here the rotation matrix @xmath268 diagonalizes the matrix @xmath269 as @xmath270 , and the rotation matrix @xmath271 diagonalizes the matrix @xmath272 .
the velocities @xmath273 and @xmath274 are diagonal elements of @xmath275 . in this section
we calculate correlation exponents for order parameters characterizing the phases in fig .
[ fig : pd](b ) .
for the gaussian model ( [ eq : h^eff ] ) , the correlation functions of vertex operators , @xmath276 and @xmath277 with real parameters @xmath278 , show power - law decay , @xmath279 where the exponents are given by @xmath280 [ eq : eta ] these results can be applied to the cases of interest . in the case when @xmath193 is renormalized to strong coupling ( @xmath251 ) ,
the fluctuations in the @xmath225 field diverge , and , consequently , the order parameters that contain the vertex operator of @xmath225 exhibit short - range correlations or exponential decay at large distances . on the other hand ,
the locked field @xmath224 can be replaced by its average @xmath281 in the order parameters that contain @xmath224 .
the correlation functions for the boson order parameters are then given by @xmath282 we note that the exponents are unchanged from those in the @xmath134 case [ see eqs .
( [ eq : cor0b ] ) ] and that the correlation functions of @xmath4 particles are controlled by the tll parameter @xmath198 .
( to be precise , @xmath198 should be replaced by its renormalized value @xmath283 , whose difference from @xmath198 is on the order of @xmath202 . ) the sf@xmath284 is dominant for @xmath136 , while the dw@xmath284 becomes dominant for @xmath285 . for @xmath5 and @xmath6 particles ,
slowly decaying correlation functions are given by @xmath286 where @xmath287^{-1/2}. \label{eq : corre_case1_exp}%\]][eq : corre_case1 ] the most dominant order for @xmath5 and @xmath6 particles is determined by @xmath288 : the sf@xmath234 state for @xmath289 and the dw@xmath234 state for @xmath290 .
in the phase diagram shown in fig .
[ fig : pd](b ) , the region of relevant @xmath193 is classified into four regions according to the most slowly decaying correlation for the bosonic ( @xmath4 ) and fermionic ( @xmath5 , @xmath6 ) particles . here
we briefly discuss the correspondence to the results obtained in ref .
@xcite , in which the @xmath5-@xmath6 paired state is predicted within a mean - field analysis of a 3d model .
it is pointed out in ref .
@xcite that the molecular conversion term induces a _ repulsive _ density - density interaction between a fermionic atom and a molecule through a lowest - order virtual process .
this effective interaction is consistent with the interaction vertex @xmath210 generated in our perturbative rg analysis .
furthermore , it is argued in ref .
@xcite that , if the bosons are _ condensed _ , the effective interaction between a fermionic atom and a molecule can become _ attractive _ , thereby yielding the sf order of `` @xmath127-wave '' @xmath5-@xmath6 pairing state . in the present 1d case ,
the mean - field theory is invalid ( bosons can not condense ) , and the effective interaction @xmath193 is repulsive . therefore the @xmath127-wave @xmath5-@xmath6 pairing can not be stabilized . instead
, we obtain a `` @xmath115-wave '' @xmath5-@xmath6 pairing ( or out - of - phase dw state of @xmath5 and @xmath6 particle ) which can be stabilized due to the induced _ repulsive _ interaction between @xmath5 and @xmath6 particles .
next we consider the case where the phase field @xmath225 is locked .
the fluctuations of the @xmath224 field are divergent , and its order parameters exhibit short - range correlations .
the order parameters of our interest are those involving @xmath225 , which can be simplified by replacing @xmath291 with its expectation value @xmath292
. the correlation functions of these leading order parameters exhibit algebraic decay , @xmath293 [ eq : corre - theta ] the correlation functions of the order parameters @xmath294 and @xmath295 also exhibit algebraic decay .
however , these orders can not dominate over those given in eqs .
( [ eq : corre - theta ] ) , since their exponents are always greater than those in eqs .
( [ eq : corre - theta ] ) .
when @xmath296 and @xmath297 , the hamiltonian ( [ eq : heff ] ) takes a diagonal form , and the exponents are simplified to @xmath298 [ exponents ] in the parameter region in fig .
[ fig : pd](b ) where @xmath150 flows to strong coupling , the exponent @xmath299 is always smaller than the others in eqs . ( [ exponents ] ) .
hence , the sf@xmath284 state is designated as the most dominant state .
we also note that the sf@xmath284 correlation is enhanced as compared with the case of @xmath134 where @xmath300 .
in summary , we have carried out a comprehensive study of a two - channel bose - fermi mixture , for which the analysis and results presented here can possibly be applied towards more general many - body problems involving interacting multicomponent quantum liquids .
when the densities of the fermionic atoms and fermionic mb molecules are identical , the feshbach molecule conversion and disassociation , the @xmath16 term , can become relevant and induce an excitation gap , while the system retains two gapless modes .
one appealing feature of the phase diagram in particular is the existence of a dominant composite @xmath115-wave pairing state @xmath301 , which occurs for fermions in both the open and closed hyperfine channels , induced by an effective interaction mediated by @xmath4 atoms . ultimately , we hope that the phase diagram presented here should demonstrate more general features of composite orders and indirect scattering processes that will manifest in higher dimensions .
although we have established the qualitative behavior of the phase diagram for a wide range of interaction couplings , a better comparison with experiments will require microscopic determination of the tll parameters using numerical methods .
since our model contains specific order parameters that couple different atomic species , a direct experimental probe must be sensitive to interspecies density correlations .
time - of - flight spectroscopy is the most promising method , as it can directly image an atomic cloud s density profile , which should demonstrate specific commensurability in the presence of density wavelike order @xcite .
a possible experimental realization within the cold atoms systems would involve a magnetic trapping technique developed on atom chips @xcite .
recently , the tll signatures have been confirmed by observing certain quasi - long - range order within the noise correlations between two independent 1d bosonic atomic condensates created on an atom chip @xcite . as discussed in ref .
@xcite , the analysis of the noise correlations would be also useful to detect the composite pairing states proposed in the present paper , since this measurement would be sensitive not only to density - wave fluctuations but also to pairing fluctuations . in order to make a proper comparison of the results obtained in this paper with actual experiments in trapped cold atom systems
, we have to take into account the density inhomogeneity arising from the harmonic trap . for this purpose
, we can apply the local density approximation ( lda ) @xcite when the range of the density variation is much larger than the average interparticle distance . in the incommensurate case ( @xmath302 ) , where the system is described as the three - component tll in the homogeneous limit
, the low - energy properties can be analyzed by the bosonization scheme based on the lda @xcite .
on the other hand , in the commensurate case ( @xmath59 ) , the extension of the rg analysis would not be so straightforward .
the numerical studies on the trapped boson system in an optical lattice @xcite have shown the transition from a superfluid to a mott insulating state in the so - called `` wedding cake '' structure with density plateaus of the mott state , which was indeed observed experimentally @xcite . such a structure can be ascribed to the commensurability effect which is present when the number of bosons per site becomes integer .
since the commensurability effects can be represented as the sinusoidal potentials in the bosonization scheme , we expect that similar commensurate - incommensurate transitions should be realized when a trapping potential is taken into account in the present system .
we thank t. giamarchi , e. orignac , and masahiro sato for important discussions .
s.a . acknowledges helpful conversations with a. m. tsvelik and support from the riken fpr program .
the model which we consider in the present paper has a close connection to the model of spinless two chains coupled by the one - particle interchain hopping @xcite .
the model hamiltonian for the two - coupled chains is given by @xmath303,\end{aligned}\ ] ] where @xmath304 refers to the left- ( right- ) moving particle and @xmath242 is the chain index .
the couplings @xmath305 and @xmath306 represent the intrachain and interchain interactions , respectively @xcite . in earlier works , the interchain hopping term is diagonalized by introducing the bonding and antibonding band basis of the field operators , and then the bosonization and rg methods are applied to the field operators on the band basis @xcite . in this appendix , we verify that the same results can be obtained by directly applying the bosonization to the field operators on the original chain basis .
the bosonized forms of the field operators are given by @xmath307 where @xmath242 is the chain index and @xmath308 is the klein factor satisfying @xmath309 .
the commutation relation of the phase variables is @xmath178 = i\pi \delta_{s , s ' } \theta(-x+x')$ ] . since
a dominant phase can be determined by the locking position of @xmath111 or @xmath110 , we have to carefully apply the fusion rules for vertex operators . with the symmetric and antisymmetric combinations of phase variables , @xmath310 and @xmath311 , the bosonized hamiltonian
is written as @xmath312 \nonumber \\ & & { } + \frac{u_-}{2\pi } \int dx \left [ \frac{1}{k_- } ( \partial_x \phi_-)^2 + k_- ( \partial_x \theta_-)^2 \right ] \nonumber \\ & & { } + i\frac{u_- g_\perp}{\pi \alpha^2 } \int dx \cos \sqrt{2}\theta_-
\sin \sqrt{2}\phi_- \nonumber \\ & & { } + \frac{u_-\widetilde{g}_\phi}{\pi \alpha^2}\int dx \cos 2\sqrt{2}\phi_- \nonumber \\ & & { } + \frac{u_-\widetilde{g}_\theta}{\pi\alpha^2}\int dx \cos 2\sqrt{2}\theta_- , \label{eq : ap - h}\end{aligned}\ ] ] where @xmath313 , @xmath314 , and @xmath315 .
the coupling constants @xmath316 and @xmath317 are initially zero but generated through the rg transformation . only the asymmetric fields @xmath318 are subject to the sinusoidal potentials , and the symmetric fields @xmath319 remain free .
the order parameters characterizing the ground state are written in the bosonized form as @xcite @xmath320 where cdw@xmath321 , oaf , sc@xmath322 , and sc@xmath323 stand for charge - density wave , orbital antiferromagnetic , @xmath119-wave superconducting , and @xmath127-wave superconducting states , respectively . in order to analyze the low - energy behavior of the @xmath324 mode ,
we apply the momentum - shell renormalization - group method @xcite .
first , we split the phase variable as @xmath325 and @xmath326 , where @xmath327 and @xmath328 are the phase fields containing low - momentum and high - momentum components , respectively , @xmath329 where @xmath330 , @xmath331 , @xmath332 @xcite , and @xmath333 is the fourier transform of @xmath334 .
the fields @xmath335 and @xmath336 are defined similarly as low- and high - momentum components of the conjugate fields @xmath110 .
the rg equations are derived by integrating out the @xmath337 and @xmath338 fields with the help of eq .
( [ eq : hcorre ] ) .
we perform rg transformations of the action @xmath339 by treating the interchain hopping part , @xmath340 as a weak perturbation , where @xmath341 . in doing so
, we have to pay special attention to the commutative properties . the equal - time commutation relation between @xmath111 and @xmath110
is given by @xmath342= i\pi \delta_{s , s'}\theta(-x+x ' ) , \label{eq : commutation}\ ] ] and their correlation functions are given by @xcite @xmath343 [ eq : correlationfn ] where @xmath344 with @xmath345 .
the last terms @xmath346 in eqs .
( [ eq : correlationfn ] ) are added in order to reproduce the commutation relation ( [ eq : commutation ] ) .
integrating out the @xmath328 fields yields the @xmath347 contribution to the action @xmath339 , @xmath348 where @xmath349 and @xmath350 stand for @xmath351 and @xmath352 , respectively , and @xmath353 is the cumulant expectation with respect to the @xmath337 and @xmath338 fields .
the cumulant expectations can be evaluated as @xmath354 .
\quad \,\,\end{aligned}\ ] ] we note that the integrand becomes nonzero only for small @xmath355 since the function @xmath356 decays rapidly in @xmath357
. we can rewrite the product of the vertex operators as @xmath358 where we have used @xmath359 + [ \phi'_s(\bm r_2 ) , \theta'_s(\bm r_1 ) ] = \langle [ \phi'_s(\bm r_1 ) , \theta'_s(\bm r_2 ) ] \rangle + \langle [ \phi'_s(\bm r_2 ) , \theta'_s(\bm r_1)]\rangle = i\pi$ ] , together with the relation ( [ eq : correlationfn ] ) . using eq .
( [ eq : ope ] ) , we can perform the operator - product expansion [ @xmath360 and @xmath182 ] : @xmath362 } e^{-k_-[\bar g(0)+\bar g(r_{12 } ) ] } \nonumber \\ & \approx & \ , : e^ { + i\epsilon ' 2\sqrt{2}\phi_-'(\bm r ) } : \ , \
, e^{-k_-^{-1}[\bar g(0)-\bar g(r_{12 } ) ] } e^{-k_-[\bar g(0)+\bar g(r_{12 } ) ] } \nonumber \\ & = & e^ { + i\epsilon ' 2\sqrt{2}\phi_-'(\bm r ) } e^{-(k_-^{-1}-k_-)[\bar g(0)-\bar g(r_{12})]},\end{aligned}\ ] ] where @xmath363 is given in eq .
( [ eq : phipcorre ] ) , and we have used eqs .
( [ eq : correlationfn ] ) .
thus we find @xmath364 where @xmath365 is defined in eqs .
( [ eq : a ] ) .
the first ( second ) term renormalizes the @xmath366 ( @xmath367 ) term .
the full rg equations for the coupling constants and the tll parameter @xmath368 are given by @xmath369 [ eq : ap - rg ] we see from eqs .
( [ dg_phi / dl ] ) and ( [ dg_theta / dl ] ) that the one - loop rg processes yield contributions of order @xmath370 to @xmath316 and @xmath317 , respectively .
consequently , when @xmath371 , the coupling @xmath316 is relevant and renormalized to strong coupling ( @xmath372 ) . in this case , the phase field @xmath324 is locked at @xmath373 . on the other hand , if @xmath374 , the coupling @xmath317 is relevant and renormalized to strong coupling ( @xmath375 ) , and then the phase field @xmath376 is locked at @xmath377 . for @xmath371 ( i.e. , @xmath378 )
, the relevant order parameters , cdw@xmath321 and sc@xmath322 , are reduced to @xmath379 , @xmath380 , as @xmath324 is locked at @xmath381 .
these correlation functions show qlro , @xmath382 the dominant correlation is determined by the value of the tll parameter @xmath383 ; the cdw ( sc@xmath322 ) state becomes most dominant for @xmath384 ( @xmath385 ) , i.e. , @xmath386 ( @xmath387 ) . for @xmath374 ( i.e. , @xmath388 )
, the relevant order parameters are given by @xmath389 and @xmath390 , and their correlation functions are @xmath391 the dominant correlation is the oaf ( sc@xmath323 ) state when @xmath384 ( @xmath385 ) , i.e. , @xmath386 ( @xmath387 ) .
since the rg analysis described above correctly reproduces the phase diagram obtained in ref .
@xcite , the validity of our method is confirmed . as we noted earlier ,
the sinusoidal potentials of the two - chain hamiltonian ( [ eq : ap - h ] ) have forms similar to those of eqs .
( [ eq : hg3p - rotated ] ) .
we can thus study the phase diagram of our model using the same rg method ( with straightforward generalization ) , as described in secs .
[ sec : rg ] and [ sec:5 ] , where the rg equations ( [ eq : rg ] ) are indeed similar to eqs .
( [ eq : ap - rg ] ) .
the model hamiltonian ( [ eq : fullham ] ) can be mapped to the hamiltonian for the two - coupled chain with gauge field . by applying the phase representation only for the boson [ eq . ( [ eq : psi_b ] ) ] , and by expressing @xmath392 and @xmath393 , the effective hamiltonian is expressed as @xmath394,\quad \\
h_{3p } \!\ !
& = & \!\ !
-t_\perp \sum_{p = l , r } \int dx \left(\psi_{1}^{p\dagger } \psi_{2}^p e^{-i\theta_b}+\mathrm{h.c.}\right ) , \qquad\end{aligned}\ ] ] where @xmath395 , and we have set @xmath396 . we assumed the short - range interaction @xmath397 and @xmath398 , where @xmath399 is the small quantity , and we set @xmath400 for simplicity .
we focus on the commensurate case @xmath401 .
this model can be interpreted as the two - coupled chain model with a gauge field on the _ rung_. in this section , we first eliminate the effect of the gauge field by gauge transformation and diagonalize the @xmath402 term . in the next step ,
we apply the bosonization , as performed in the two - chain problem @xcite . by the gauge transformation
@xmath403 the @xmath402 term becomes @xmath404 . since the @xmath402 term is expressed in the form of conventional interchain hopping , we can follow the approach of ref .
@xcite in which the relevant @xmath402 term was treated nonperturbatively
. the interchain hopping term can be diagonalized by introducing the bonding and antibonding operators : @xmath405 the @xmath402 term is given by @xmath406 , and the intrachain kinetic terms are given by @xmath407 in contrast to the @xmath402 term , which is given in a diagonalized form , the intrachain kinetic terms contain the gauge field and the field operators are given in the nondiagonalized form .
now we bosonize the fields @xmath408 : @xmath409 + i\theta_\pm ( x)},\ ] ] where @xmath410 for @xmath411 . in order to simplify the notation
, we further apply the simple transformation , @xmath412 and then the hamiltonians @xmath413 and @xmath414 are expressed as @xmath415 \nonumber \\ & & { } \!\ !
+ \frac{u_-}{2\pi } \int dx \left [ \frac{1}{k_-}(\partial_x \tilde\phi_-)^2 + k_- ( \partial_x \tilde\theta_-)^2 \right ] \nonumber \\ & & { } \!\ ! - \frac{ig_0}{\pi \alpha } \int dx \ , ( \partial_x \theta_b ) \sin \sqrt{2}\tilde\theta_- \cos \sqrt{2}\tilde\phi_- \nonumber \\ & & { } \!\ ! + \frac{1}{2 \pi^2 \alpha^2 } \int dx \bigl [ -g_\phi \cos 2\sqrt{2}\tilde\phi_- + g_\theta \cos 2\sqrt{2}\tilde\theta_- \bigr ] , \nonumber \\
h_{3p } \!\ ! & = & \!\ ! t_\perp \int dx \frac{\sqrt{2}}{\pi } \partial_x \tilde \phi_-,\end{aligned}\ ] ] where @xmath416,\end{aligned}\ ] ] and @xmath417 and @xmath418 depend on @xmath305 and @xmath419 .
the microscopic determination of the parameters @xmath417 and @xmath418 are given in ref .
@xcite for weak - coupling region , while it requires numerical analysis in the wide range of interactions . from the scaling analysis , the rg equation for @xmath420
is given by @xmath421 , implying that the @xmath420 term is marginal for @xmath422 and becomes irrelevant for @xmath423 .
the presence of the marginal or irrelevant @xmath420 term would give rise to slight renormalization of other quantities like @xmath198 , @xmath368 , @xmath424 , @xmath425 , and @xmath402 .
however , we argue below that , up to such relatively unimportant corrections , we can safely neglect the @xmath420 term so this hamiltonian takes on the same form as two - coupled chains , derived in ref .
first , we observe that the @xmath402 term suppresses the potential @xmath426 , since the former favors the incommensurate state while the latter favors the commensurate state .
depending on the value of @xmath368 , the @xmath427 field can either remain massless or develop a gap .
we concentrate here on the case where @xmath427 develops a gap and acquires a nonzero expectation value determined by minimizing the ground - state energy .
the average value of the massive field @xmath428 depends on the sign of @xmath429 , where @xmath429 is the fixed - point value of @xmath425 . the average value of @xmath428 and the corresponding order parameters are summarized in table [ table : ap2 ] .
the order parameters of the interest are @xmath432 [ eq .
( [ eq : o_dw_fpsi ] ) ] , @xmath433 [ eq . ( [ eq : fpsi ] ) ] , and @xmath434 [ eq . ( [ eq : obff ] ) ] . after the gauge transformation and the bosonization , these order parameters are written as this behavior is consistent with eqs .
( [ eq : corre_case1 ] ) if we equate @xmath383 with @xmath288 .
we note that , in the simplified case where @xmath296 and @xmath297 , the exponent @xmath288 is given by @xmath439 , as seen from eq .
( [ eq : corre_case1_exp ] ) .
on the other hand , when @xmath440 , the correlation function of @xmath434 exhibits algebraic decay , @xmath441 this behavior is consistent with eq .
( [ eq : corre_bff ] ) by noting that the exponent @xmath442 is given by @xmath443 [ eqs .
( [ exponents ] ) ] in the case of @xmath296 and @xmath297 . | we study the ground - state properties of one - dimensional mixtures of bosonic and fermionic atoms resonantly coupled to fermionic feshbach molecules . when the particle densities of fermionic atoms and feshbach molecules differ , the system undergoes various depletion transitions between binary and ternary mixtures , as a function of the detuning parameter .
however , when the particle densities of fermionic atoms and feshbach molecules are identical , the molecular conversion and disassociation processes induce a gap in a sector of low - energy excitations , and the remaining system can be described by a two - component tomonaga - luttinger liquid . using a bosonization scheme
, we derive the effective low - energy hamiltonian for the system , which has a similar form as that of the two - chain problem of coupled tomonaga - luttinger liquids . with the help of improved perturbative renormalization group analysis of the latter problem
, we determine the ground - state phase diagram and find that it contains a phase dominated by composite superfluid or pairing correlations between the open and closed resonant channels . |
intensive multiwavelength monitoring campaigns have shown that variability studies provide an excellent tool to investigate the innermost region of active galactic nuclei ( agns ) .
detailed studies of the continuum and emission - line variations have revealed new insights about the size , structure , and dynamics of the broad - line region ( blr ) in these sources ( see peterson 1993 for a review ) . over the last decade , a number of large space - based and ground - based agn monitoring programs have been undertaken .
our own group , the international agn watch consortium ( alloin et al .
1994 ) , has undertaken monitoring programs on several seyfert galaxies , including programs on ngc 5548 ( clavel et al .
1991 ; peterson et al . 1991 , 1993 , 1994 ;
maoz et al .
1993 ; dietrich et al . 1993
; korista et al .
1995 ) , ngc 3783 ( reichert et al . 1994 ;
stirpe et al . 1994 ; alloin et al .
1995 ) , ngc 4151 ( crenshaw et al .
1996 ; kaspi et al . 1996 ; warwick et al .
1996 ; edelson et al . 1996 ) , fairall 9 ( rodrguez - pascual et al . 1997
; santos - lle et al .
1997 ) , and ngc 7469 ( wanders et al . 1997 ; collier et al .
1997 ) . in late 1994 , the international agn watch began a multiwavelength monitoring campaign on the broad - line radio galaxy ( blrg ) 3c390.3 , a prominent nearby ( @xmath13 ; osterbrock et al . 1975 )
agn with broad double - peaked emission - line profiles ( sandage 1966 ; lynds 1968 ) .
3c390.3 has a well - known variability history ( e.g. selmes , tritton , & wordsworth 1975 ; barr et al .
1980 ; penston & prez 1984 ; veilleux & zheng 1991 ; zheng 1996 ; wamsteker et al . 1997 ) .
zheng ( 1996 ) and wamsteker et al . ( 1997 ) analysed the iue spectra of 3c390.3 which have been taken from 1978 until 1992 .
the variable broad ly@xmath5 and iv civ emission are delayed by @xmath1060 days with respect to the uv continuum variations .
3c390.3 is an extended double - lobed fr ii radio source ( leahy & perley 1995 ) with a 60one - sided narrow jet at @xmath14 and it is one of the rare lobe - dominated radio galaxies which show superluminal motions , with @xmath15 ( alef et al .
this is the first time that variations in a radio - loud agn have been studied over a broad energy range ( from radio to x - ray energies ) for an extended period ( one year ) the observations obtained for this program cover over eight decades in frequency .
the results of the x - ray and ultraviolet monitoring campaigns performed with rosat and the iue satellite will be presented in leighly et al .
( 1997 ) and obrien et al .
( 1998 ) , respectively . in this contribution , we present the optical photometric and spectroscopic observations that were obtained as part of this monitoring program ; observations in other wavelength bands will be presented elsewhere . in 2 , we describe the optical observations and outline intercalibration procedures by which a homogeneous set of the photometric and spectroscopic measurements is achieved . in 3 , we highlight emission - line profile variations based on mean and rms spectra and present the results of some preliminary time - series analysis .
we summarize our results in 4 .
spectra and broad - band photometric measurements of 3c390.3 were obtained by a large number of observers between 1994 october and 1995 october .
table 1 gives a brief overview of the various sources of the data we report here .
each group ( column 1 ) was assigned an identification code given in column ( 2 ) which will be used throughout this paper .
column ( 3 ) gives the aperture of the telescope used .
columns ( 4 )
( 7 ) list the focal - plane apertures used in various observations ; generally , fixed instrument apertures were used , although two groups ( f , n ) adjusted their aperture to compensate for changes in seeing . in most cases
the size of the apertures corresponds to three times the average seeing at the observing site .
those large apertures cover the agn as well as the entire galaxy . in column ( 8) ,
the spectrograph slit widths ( in the dispersion direction ) and extraction widths ( cross - dispersion dimension ) , respectively , of the spectra are listed .
complete logs of the photometric and spectroscopic observations ( table 2 and 3 ) are available as electronic files on the www at the url given in the following .
the photometric observations were made through different combinations of broad - band filters .
generally , the brightness of 3c390.3 was scaled with respect to standard stars in the photometric sequence ( stars a , b , and d ) defined by penston , penston , & sandage ( 1971 ) , plus a hst guide star located at 208 from 3c390.3 at @xmath16 ( hst gs 4591:731 ) . in some cases , star a could not be used because of saturation .
the colors of the comparison stars ( b , d , and hst gs 4591:731 ) differ by up to 0.1 mag . while star a is @xmath172.5 mag .
brighter than the mean values of these three stars .
the mean colors of 3c390.3 are similiar to the mean colors of the comparison stars within 0.2 mag .
since the stars and the agn are within the same field of view of the ccd images , effects of different spectral energy distributions and airmasses on the internal calibration can be neglected . in the following
, the procedure we use to derive the @xmath2-band magnitudes will be described in some detail .
the light curves of the other broad - band measurements were obtained in a similar way .
first , a subsample was selected which covered the entire monitoring campaign with a good temporal sampling rate .
for the @xmath2-band measurements , the observations recorded with the 0.7-m telescope of the landessternwarte heidelberg ( l ) provide an appropriate data set .
the used aperture was a 15 rectangle to ensure that the light loss can be neglected even in the case of bad seeing .
the frames were bias and flat - field corrected in the standard way . in the next step ,
the brightness of 3c390.3 was measured with respect to the comparison stars in the same field , yielding the difference in apparent magnitude between 3c390.3 and the comparison stars . in the @xmath2-band ,
stars b and d from penston et al .
( 1971 ) and hst gs 4591:731 mentioned above were used for calibration .
the @xmath2-band magnitudes that were derived for these stars are presented in table 4 together with the @xmath0 and @xmath1 magnitudes from penston et al.(1971 ) .
finally , the @xmath2-band light curves of the other subsamples were shifted to the @xmath2-band light curve derived from the heidelberg sample by applying a constant additive magnitude offset to all of the measurements in a given subset .
the additive factor was derived by comparison of @xmath2-band magnitudes which were observed within @xmath183 days .
the additive factors for the individual subsamples are presented in table 5 .
the additive scaling factors given in table 5 are quite large in some cases .
generally , the brightness variations of 3c390.3 were provided in magnitudes using the standard stars in the field for calibration .
but a few groups ( m , r , u ) provided the brightness of 3c390.3 as differential magnitudes relative to the calibration stars in the field .
the @xmath0-band magnitudes are scaled with respect to the measurements taken at wise observatory ( o ) .
the offset for the steward observatory sample ( f ) was derived by comparison of epochs which were simultaneous to within @xmath183 days .
this restriction could not be used for the special astrophysical observatory ( j ) ; in this case , the offset is based on epochs separated by no more than 6 days .
the resulting additive scaling factors are given in table 5 .
the sample recorded at behlen observatory ( q ) was used as the standard in the @xmath1-band because the temporal sampling is good and the measurements were made through a large aperture ( 13 .
@xmath197 ) so that uncertainties due to seeing variations are small .
the @xmath1-band measurements are made relative to star b , which is close in brightness to the nucleus of 3c 390.3 . in order to scale the @xmath1-magnitudes to a common flux level , we compared measurements at epochs separated by no more than three days . in table 5 ,
the differential magnitudes are given which were applied to the individual light curves to produce a common light curve .
for the calibration of the @xmath3-band magnitudes we used observations taken at calar alto observatory .
the @xmath3-band frames were recorded in december 1994 .
the calibration was based on the globular cluster ngc2419 using the stars given by christian et al.(1985 ) .
the resulting brightnesses of star a , b , d , and hst gs 4591:731 are given in table 4 .
the additive factors for the individual subsamples are presented in table 5 .
the measured brightnesses of the stars used for calibration are in good agreement with the values provided by penston et al .
however , star a deviates from this trend in @xmath2 and @xmath3 in comparison to stars b and d. it might be that this is due to the brightness of star a causing non - linearity effects and an underestimate of the brightness of star a. after combining the measurements of the individual groups to common broad - band light curves , the apparent magnitudes were transformed into flux .
the conversion has been performed using the following equations ( allen 1973 ; wamsteker 1981 ) : @xmath20 the resulting time - binned light curves from the @xmath0 , @xmath1 , @xmath2 , and @xmath3 broad - band measurements in units of @xmath21ergs s@xmath12@xmath22@xmath12 are displayed in figure 1 .
the flux calibration of agn spectra can be accomplished in several ways . in variability studies , it is common practice to normalize the flux scale to the fluxes of the narrow emission lines , which are assumed to be constant over time scales of at least several decades ( peterson 1993 ) .
this assumption is justified by the large spatial extent and low gas density of the narrow - line region ( nlr ) , since light travel - time effects and the long recombination time scale ( @xmath23years for @xmath24@xmath25 ) damp out short time - scale variability .
however , 3c390.3 is apparently a special case in this regard , as narrow - line variability has been reported in this object ( clavel & wamsteker 1987 ) . on the basis of spectra obtained between 1974 and 1990 , zheng et al .
( 1995 ) present evidence that the [ oiii]@xmath26 , 5007 fluxes follow the variations in the continuum , although on a longer time scale than the broad emission lines . during the period of decreasing and increasing [ oiii ]
flux there might also be periods of several months or one year of nearly constant [ oiii ] flux .
however , we need to examine the data closely to test for variability of the [ oiii ] lines before we use them for flux calibration . two data sets , the ohio state sample ( a ) and the lick observatory sample ( b ) , were selected to measure the absolute flux in the [ oiii]@xmath27 line .
the spectra were photometrically calibrated by comparison with the broad - band photometric measurements .
the spectra were convolved with the spectral response curve of the johnson @xmath1 and @xmath2 filters ( schild 1983 ) , and the flux of the convolved spectra was measured for the wavelength range of the filter curves and compared with photometric data points .
photometric and spectroscopic data obtained no more than 3 days apart were used in this comparison .
the spectra were then scaled by multiplicative factors to achieve the same total flux ratio as the intercalibrated photometric measurements in @xmath1 and @xmath2 . for the lick data
( b ) , the @xmath2-band measurements were used to scale the entire spectrum , and the [ oiii]@xmath27 flux was then measured .
spectra from the ohio state sample ( a ) do not cover the entire wavelength range of the @xmath1-band filter , and therefore the missing contribution to the @xmath1-band measurement was estimated from the lick spectra , which cover the entire wavelength range of the @xmath1 bandpass .
the ratio of @xmath1-band flux measured from the ohio state spectra to that measured from the lick spectra was found to be @xmath28 , i.e. , a constant contribution .
the fluxes from the ohio state spectra were thus corrected by this constant factor .
the [ oiii]@xmath27 flux was then measured from the photometrically scaled ohio state ( a ) and lick observatory ( b ) spectra by integrating the spectrum over the range 52585320 .
therefore , a linear pseudo - continuum was fitted beneath the [ oiii]@xmath27emission line . in fig . 2
, we show the [ oiii]@xmath27 flux measured from the ohio state and lick subsets normalized to a mean value of unity and displayed as a function of time .
no time - dependent trend is detected , and the rms variation about the mean is @xmath29% .
we thus conclude that it is safe to assume that the [ oiii]@xmath27 flux is constant within a few percent over the duration of this monitoring program , and that the [ oiii]@xmath27 flux can be used to calibrate all of the spectra .
the [ oiii]@xmath27 flux is taken to be @xmath30\,\lambda5007 ) = 1.44 \times 10^{-13}$]ergs s@xmath12@xmath22 , the mean value of the data points plotted in fig . 2 .
it is also important to take aperture effects into account ( peterson & collins 1983 ) .
the seeing - dependent uncertainties which are introduced by the aperture geometry can be minimized by using large apertures .
it has been shown that apertures of 5 @xmath31 7 .
@xmath195 can reduce seeing - dependent photometric errors to no more than a few percent in the case of nearby agns ( peterson et al .
1995 ) . in the case of 3c390.3
, the blr and the nlr can be taken to be point sources ( cf .
baum et al . 1988 ) , which means that no aperture correction needs to be made for the agn continuum / narrow - line flux ratios or the broad - line / narrow - line ratios , since seeing - dependent light losses at the slit will be the same for each of these components . however , the amount of host - galaxy starlight that is recorded is still aperture dependent , and systematic corrections need to be employed . in order to estimate how sensitive the measurements are to seeing - dependent light loss from the host galaxy , we carried out simulated aperture photometry on the @xmath2-band frame obtained at calar alto with the 2.2-m telescope under good seeing conditions ( @xmath32 ) .
the host galaxy has been separated from the point - like agn component by fitting a de vaucouleurs @xmath33 profile to the observed surface - brightness distribution .
this image was convolved with gaussians of various width to simulate various seeing conditions up to 4 . @xmath190 .
the flux of the host galaxy and of the point - like agn were measured for a fixed aperture of 10 .
@xmath195x10 .
the flux ratio of the point - line agn to the host galaxy for the @xmath2-band is @xmath34 .
this is similar to the result of smith & heckman ( 1989 ) , who found @xmath35 for @xmath1-band measurements . by using different aperture geometries in these simulations
, we find that the ratio of agn light to starlight from the host galaxy changes over the full range of observed seeing values by less than 1% for the larger apertures ( i.e. , slit width greater than 4 ) . for intermediate slit widths ( 2 .
@xmath195 3 .
@xmath196 ) , seeing variations introduce uncertainties of @xmath36% . for the smallest slit widths ( @xmath37 ) ,
seeing effects can alter the nucleus to starlight ratio by as much as @xmath38% , with the largest uncertainties occurring for seeing worse than @xmath39 .
since the data that constitute the various samples were taken with different instruments in different configurations , the spectra have to be intercalibrated to a common flux level . as we have shown above , the [ oiii]@xmath27 line flux was constant to better than 3% during this campaign , so we can safely use the narrow emission lines as flux standards . in order to avoid any wavelength - dependent calibration errors ,
each spectrum was scaled in flux locally over a limited wavelength range prior to measurement .
the h@xmath6 spectral region was scaled with respect to the [ oiii]@xmath26 , 5007 line fluxes , while the h@xmath5 region was scaled with respect to the fluxes of the [ oi]@xmath40 and [ nii]@xmath41 , 6584 emission lines .
the spectra were intercalibrated using the method described by van groningen & wanders ( 1992 ) .
this procedure corrects the data for different flux scales , small wavelength shifts , and different spectral resolutions by minimizing the narrow - line residuals in difference spectra formed by subtracting a `` reference spectrum '' from each of the observed spectra .
the rescaled spectra are used to derive integrated emission - line fluxes as well as the optical continuum flux .
the continuum fluxes are then adjusted for different amounts of host - galaxy contamination ( see peterson et al .
1995 for a detailed discussion ) through the relationship @xmath42}\,\lambda5007 ) } \right]_{\rm obs } - g,\ ] ] where @xmath43 is the adopted absolute [ oiii]@xmath27 flux , the quantity in brackets is the observed continuum to [ oiii]@xmath27 flux ratio measured from the spectrum , and @xmath44 is an aperture - dependent correction for the host - galaxy flux . the ohio state sample ( a ) , which uses a relatively large aperture ( 5@xmath31 7 .
@xmath195 ) , was adopted as a standard ( i.e. , @xmath45 by definition ) , and other data sets were merged progressively by comparing measurements based on observations made no more than @xmath183 days apart .
this means that any real variability that occurs on time scales this short tends to be somewhat suppressed by the process that allows us to combine the different data sets . the additive scaling factor @xmath44 for the various samples are given in table 6 .
the average interval between measurements is about @xmath46days for @xmath47 @xmath48 and @xmath3 while the @xmath0-band variations have been measured with an average sampling interval of @xmath49days ( table 7 ) .
the time - binned broad - band continuum light curves are shown in fig . 1 . the variations can be characterized as a nearly monotonic increase of the flux with smaller - scale variations superposed on this general trend . from jd2449760 to jd2449800 , the flux rose in the v , r , i bands faster than during the previous interval .
after this period , the flux level stays nearly constant for nearly 3 months and then a second strong increasing episode follows .
the @xmath1-band light curve appears to have more complicated structure , and we note in particular apparently rapid variations during the interval jd2449800 to jd2449920 . figure 3 shows two spectra of 3c390.3 that represent the low ( jd2449636 ) and high ( jd2450007 ) flux states observed during this campaign .
the most obvious variation is the strong increase of the small blue bump , the very broad feature shortward of 4000 , which is usually ascribed to balmer continuum emission and a blend of several thousand feii emission lines ( wills , netzer , & wills 1985 ) .
the broad balmer lines show also evidence for increasing flux . underneath the low - state spectrum in fig .
3 we show the integration ranges for the various features that have been measured in these spectra .
the optical emission - line fluxes were integrated over a common range of @xmath50 ^ -1 km s@xmath12 for the strong balmer emission lines ( h@xmath5 , h@xmath6 , and h@xmath7 ) and i5876 hei@xmath8 .
the ii4686 heii@xmath9 line flux was integrated over a range corresponding @xmath51 ^ -1 km s@xmath12 to @xmath52 ^ -1 km s@xmath12 to reduce contamination by h@xmath6(cf.table 8) .
a local linear continuum fit was interpolated under each emission line . in the case of the h@xmath6 region ,
the continuum was defined by the flux measured in two narrow ( 10 width ) windows at 4400 and 5475 in the observed frame .
although the region of the long wavelength window might be contaminated by several weak emission lines and the mgb absorption feature , this is still the most line - free region near h@xmath6 and is thus an appropriate place to estimate the optical continuum flux . the continuum window at 4400 is displaced from the h@xmath7 line center by @xmath53 ^ -1 km s@xmath12 , and thus possible contamination by broad emission line flux can be safely ignored ( see fig . 3 ) .
contamination of h@xmath6 by [ oiii ] emission has been corrected by subtracting the constant [ oiii]@xmath27 flux given earlier plus the [ oiii]@xmath54 flux which we account for by assuming a [ oiii]@xmath27/[oiii]@xmath54 flux ratio of 3 . in the case of the h@xmath5 region ,
two narrow ( 10 in width ) windows at 5960 and 7495 were used to define the continuum underlying the lines .
no attempt has been made to correct any of the measured emission - line fluxes for their respective narrow - line contributions . in each spectrum ,
the optical continuum flux measured is the average value in the range 54605470 .
the extracted range of the emission lines and of the optical continuum is given in table 8 .
the time - binned light curves for the strong emission lines and the optical continuum flux @xmath4(5177 ) measured from the calibrated spectra are plotted in fig .
the corresponding flux measurements are available on the www at the url given above ( table 9 ) .
a final check of the uncertainty estimates was performed by examining the ratios of all pairs of photometric and spectroscopic observations which were separated by 2 days or less .
the error estimates were also calculated using maximum intervals of 3 days and 4 days , and the resulting errors are identical to within 3% . generally , there are at least a few hundred independent pairs of measurements within 2 days of one another for the photometric measurements .
only the @xmath0-band curve yields fewer independent pairs . for the spectroscopic data
the number of independent pairs is of the order of several dozen with the exception of h@xmath7 .
the dispersion about the mean ( unity ) , divided by @xmath55 , provides an estimate of the typical uncertainty in a single measurement ( @xmath56 ) .
the observational uncertainties ( @xmath57 ) assigned to the spectral flux measurements were estimated from the error spectra which were calculated within the intercalibration routine , as well as from the signal - to - noise ratio within the spectral range near the individual emission lines . for the continuum the mean fractional error ( @xmath57 ) in a given measurement is 0.041 .
the average fractional uncertainty from the quoted estimate ( @xmath56 ) for the same measurements ( table 10 ) is 0.040 which implies that the error estimates are probably quite good .
generally , the estimated errors ( @xmath58 ) are of the same order as the observational uncertainties ( @xmath59 ) derived directly from the measurements .
a comparison of the flux of the broad band measurements and the broad emission lines is given in table 11 .
the mean spectroscopic continuum flux is lower than the mean broad band @xmath1- and @xmath2-flux .
this can be explained by the fact that the broad band flux measurements contain in addition to the continuum flux emission line contributions .
the variability parameter @xmath60 and @xmath61 have been calculated for the broad - band flux variations ( cf .
clavel et al . 1991 ; rodrguez - pascual et al .
the quantity @xmath61 is simply the ratio of the maximum to the minimum flux .
the quantity @xmath60 is an estimation of the fluctuations of the intrinsic variations relative to the mean flux .
therefore , the rms of the light curves has been corrected with respect to the uncertainties introduced by the observations .
@xmath61 and @xmath60 of the broad - band variations are given in table 11 .
we calculated the mean and root - mean - square ( rms ) spectra from the flux - scaled spectra , and the h@xmath5 and h@xmath6 regions are shown in fig .
the velocity scale is set by adopting a redshift @xmath62 given by the narrow components of h@xmath5 and h@xmath6 and the [ oiii]@xmath63 emission lines , i.e. the restframe of the nlr .
the rms spectrum is useful for isolating the variable parts of the line profile .
the full - width at zero intensity ( fwzi ) of the mean h@xmath5 and h@xmath6profiles is ( 25000@xmath182000)^-1 km s@xmath12 , while the fwzi of the rms profiles is only ( 14000@xmath18500)^-1 km s@xmath12 ( cf.fig.5 ) .
thus , the variations in the broad emission line profiles are strong at lower radial velocities while the flux originating in high radial - velocity gas varies little , if at all .
this behavior is similar to what has been seen in mrk 590 ( ferland , korista , & peterson 1990 ; peterson et al .
1993 ) .
the line profiles of h@xmath5 and h@xmath6 show clear asymmetric structure .
at least three substructures can be identified in both line profiles . a blue hump in the mean spectra
is located near @xmath64 ^ -1 km s@xmath12 with respect to the line peak , and a red hump is seen near @xmath65 ^ -1 km s@xmath12 .
the red and blue humps can be seen clearly in the rms spectra , and in addition a broad central component appears .
if these three components are modeled as gaussians , the fwhm of the components in the wings is of order @xmath66 ^ -1 km s@xmath12 , while for the central component the fwhm is @xmath67 ^ -1 km s@xmath12 .
the components in the profile wings are nearly symmetrically located with respect to the line center . in the rms - spectrum of the h@xmath6 emission line a strong broad feature at @xmath1010000
^ -1 km s@xmath12 is clearly visible .
the fwhm of the structure can be estimated to @xmath103500 ^ -1 km s@xmath12 .
the residuals of the narrow [ oiii ] lines are located at @xmath106000 ^ -1 km s@xmath12 and @xmath108900 ^ -1 km s@xmath12 .
since this feature is also visible in the rms spectrum of h@xmath5 and of h@xmath7 it can not be caused entirely by inaccurate scaling of the nlr contribution of the narrow [ oiii ] lines .
furthermore , in the difference spectrum of the low and high state of 3c390.3 during this monitoring campaign ( cf . fig .
3 ) a strong broad feature is clearly visible at the red side of h@xmath5 ( @xmath107150 ) and a weaker structure at the red side of h@xmath6 ( @xmath10 5300 ) .
the analysis of the simultaneous ultraviolet campaign reveals the existence of a feature at @xmath108500 ^ -1 km s@xmath12 in the outer red wing of the iv civ@xmath681548 emission line ( obrien et al.1998 ) . in order to quantify any possible time delay between the various light curves shown in figs . 1 and 4
, we perform a simple cross - correlation analysis .
three methods that are commonly used in agn variability studies have been used to compute cross - correlation functions ( ccfs ) the interpolated cross - correlation function ( iccf ) of gaskell & sparke ( 1986 ) and gaskell & peterson ( 1987 ) , the discrete correlation function ( dcf ) of edelson & krolik ( 1988 ) , and the @xmath69-transformed discrete correlation function method ( zdcf ) of alexander ( 1997 ) .
the iccf and dcf algorithms and the limitations of the methods have been discussed in detail by robinson & prez ( 1990 ) and by white & peterson ( 1994 ) , and the specific implementation of the iccf and dcf used here are as described by white & peterson ( 1994 ) .
the emission lines are expected to change in response to variations in the far - uv continuum , primarily to the unobservable wavelengths just shortward of 912 .
we must therefore assume that the observable continuum can approximate the behavior of the ionizing continuum .
it is generally assumed in this type of analysis that the shortest observed uv wavelength provides the best observable approximation to the `` driving '' ( ionizing ) continuum .
for the first time , however , we have a well - sampled simultaneous soft x - ray light curve ( leighly et al .
1997 ) , so we can compare the x - ray and uv continua ( obrien et al . 1998 ) directly . in this analysis
, we used the x - ray light curve as the driving continuum ( fig.6 ) since the observational uncertainties are smaller than for the uv light curve .
as will be discussed elsewhere ( obrien et al .
1998 ) , the relative amplitudes of the ultraviolet variations at a wavelength of 1370 are nearly identical to those in the _ rosat _ hri light curve , so the results presented here do not depend critically on our choice of the x - ray light curve in preference to the uv light curve .
since the sampling of the @xmath0-band light curve is very poor , we have excluded the @xmath0-band data from the cross - correlation analysis .
operationally , the temporal coverage of the optical light curves was restricted to the period covered by the x - ray measurements .
all of the light curves were binned into time intervals of 0.5 days to avoid unnecessary structure in the iccf . for each light curve
, we also computed the sampling - window autocorrelation function ( acf@xmath70 ) , which is a measure of how much of the width of the acf is introduced by the interpolation process rather than by real correlation of the continuum values at different times .
the acf@xmath70 is computed by repeatedly sampling white - noise light curves in exactly the same way as the real observations and then computing the autocorrelation function ( acf ) .
the average of many such autocorrelations , the acf@xmath70 , has a peak at zero lag whose width depends on how much interpolation the iccf has to do on short time scales .
the width ( fwhm ) of the acf@xmath70 is @xmath71 days for the emission lines and broad - band flux measurements for the x - ray restricted time period .
these values are negligible compared to the widths of any of the emission - line acfs or ccfs found here ( see figs .
79 ) , and thus interpolation of the light curves is justified .
the general sampling characteristics of each of the restricted and rebinned light curves are given in table 12 .
the name of the feature is given in column ( 1 ) , and column ( 2 ) gives the total number of points @xmath72 in the light curve that are used in computing the cross - correlation functions .
the width ( fwhm ) of the acf is given in column ( 3 ) , and the width of iccf computed by cross - correlation with the x - ray continuum is given in column ( 4 ) .
column ( 5 ) gives the width ( fwhm ) of the corresponding acf@xmath70 .
uncertainties in the iccf results for the cross - correlation maxima and centroids , @xmath73 and @xmath74 , respectively , were computed through monte carlo techniques as follows : for both time series , each flux value in the light curves was modified with gaussian deviates based on the quoted uncertainty for that point .
each light curve of @xmath72 points was then randomly resampled @xmath72 times in a `` bootstrap '' fashion , specifically allowing points to appear more than one time ( the effect of which is only to remove at random certain points from the light curve ) .
the iccf was computed , and the values of @xmath75 and @xmath76 were recorded if they were statistically significant at a confidence level higher than 95% and not clear outliers ( e.g. , lags larger than 100 days ) . by repeating this process 5001000 times , distributions of @xmath75 and @xmath76
were built up .
the means of these distributions were always very close to the values of @xmath75 and @xmath76 obtained from the original series , and the standard deviations of these distributions are taken to be the uncertainty associated with a single realization , i.e. , the quoted uncertainties @xmath77 and @xmath74 .
the acf , acf@xmath70 , iccf , and zdcf are shown for the broad - band variations in @xmath1 , @xmath2 , and @xmath3 ( fig .
7 ) , the broad balmer emission lines h@xmath5 ,
h@xmath6 , and h@xmath7 ( fig . 8) , the helium lines i5876 hei@xmath8 and ii4686 heii@xmath9 , and the optical continuum @xmath4(5177 ) ( fig
the acfs of the light curves are broad since the shape is dominated by the nearly monotonic increase in the light curves .
the fwhms of the acfs of the well - sampled light curves are of the order of 50100days ( table 12 ) . within the uncertainties ,
the dcfs and zdcfs are identical ; therefore , to avoid confusion only the iccfs and the corresponding zdcfs are displayed in figs .
the time delay derived from the centroid of a cross - correlation function provides a more robust estimate of the lag than does the peak , as evidenced by the consistently smaller widths of the monte carlo distributions for the centroid compared to the peak .
also , in the case of the emission lines , the centroid is readily identified with a physically meaningful quantity , the luminosity - weighted radius of the line - emitting region ( koratkar & gaskell 1991 ) .
the delays expressed by @xmath78 of the three methods used here are nearly identical within the uncertainties , although the dcf method tends to yield smaller delays than the iccf and zdcf method .
the results of the cross - correlation analysis are given in table 13 .
column ( 1 ) indicates the `` responding '' light curve ( i.e. , the light curve that is assumed to be responding to the driving light curve ) , and column ( 2 ) gives the peak value of the correlation coefficient @xmath79 for the iccf .
the position of the peak of the cross - correlation functions @xmath80 was measured by fitting a gaussian curve to the upper 85% of the iccf , zdcf , and dcf ; these values are given in columns ( 3 ) , ( 4 ) , and ( 5 ) , respectively , and column ( 6 ) gives the error estimate for the position of the cross - correlation peak @xmath77 .
the centroids @xmath78 of the iccf , zdcf , and dcf , in each case computed using the points in the cross - correlation function with values greater than @xmath81 , are given in columns ( 7 ) , ( 8) , and ( 9 ) , respectively , and the uncertainty in the iccf centroid @xmath82 is given in column ( 10 ) . the broad - band ( @xmath1 , @xmath2 , and @xmath3 ) and the optical continuum flux variations appear to be delayed by a few days , relative to the x - ray or ultraviolet continuum variations
however , the measured lags are in no case different from zero at any reasonable level of statistical significance if the monte - carlo based error estimates are reliable . somewhat smaller , but marginally statistically significant , wavelength - dependent continuum lags have been reported in the case of ngc 7469 ( wanders et al .
1997 ; collier et al .
1997 ) , although in the case of ngc 7469 the mean spacing between observations of the driving continuum is much smaller than for the observations reported here .
the emission - line time delays are similar for all of the optical emission lines , about 20 days , although in each case the uncertainties are somewhat larger than we have obtained in similar experiments on account of the vagaries of the continuum behavior and sampling . within the uncertainties , the measured time delays for h@xmath5 , h@xmath6 , h@xmath7 , and
i5876 hei@xmath8 are indistinguishable
. however , the uncertainties are sufficiently large that ionization stratification , as detected in other well - studied sources , also can not be ruled out . only ii4686 heii@xmath9 appears to respond more rapidly than the other lines , but the relatively low value of @xmath79 and relatively narrow width of the cross - correlation function ( about 30 days as compared to @xmath83 days for the other lines ) cast some doubt on the significance of this result .
in contrast to the agn studied before , the delay of the ly@xmath5 and iv civ line variations of the current 3c390.3 campaign is significantly larger than the delay of the optical lines .
the analysis of the ultraviolet spectra yields a delay of 35 70 days for ly@xmath5 and iv civ ( obrien et al . 1998 ) . in order to study the response of individual parts of the line profile to the x - ray continuum variations and thus search for evidence of an organized radial - velocity field in the blr , we have divided the profiles of the strongest and least contaminated lines , h@xmath5 and h@xmath6 , into three parts : the blue wing ( @xmath51 to @xmath84kms@xmath12 ) , the core ( @xmath85 to @xmath86kms@xmath12 ) , and the red wing ( @xmath87 to @xmath88kms@xmath12 ) .
the light curves of the different profile sections ( cf.figs .
10,11 ) were rebinned onto 0.5-day intervals ( available in electronic form on the www at the url given above as table 14 ) and again cross - correlations were performed restricting the data to the x - ray monitoring period .
since we are concerned at this point with _ differential _ lags between different sections of the line profiles , we cross - correlate the light curves for the different line sections against each other rather than against the driving continuum ; we arbitrarily chose the line cores as the first ( driving ) series , and computed the lags of the red and blue wings of the lines relative to the core .
these results are shown in table 15 and in figs .
12 and 13 .
neither h@xmath5 nor h@xmath6 shows any evidence for any differences in the response of the wings relative to the core . in the case of h@xmath5 ,
the uncertainties are especially large on account of the low amplitude of variability , but the h@xmath6 results seem to exclude the possibility that the blr velocity field is characterized by primarily radial motions , either infall or outflow .
the results of a year - long ( 1994 october to 1995 october ) optical monitoring campaign on the blrg 3c390.3 are presented in this paper .
the principal findings are as follows : 1 . the broad - band ( @xmath0 , @xmath1 , @xmath2 , and @xmath3 ) fluxes , the optical continuum measured from spectrophotometry @xmath4(5177 ) , and the integrated emission - line fluxes of h@xmath5 , h@xmath6 , h@xmath7 , i5876 hei@xmath8 , and ii4686 heii@xmath9 showed significant variations of order 50% in amplitude .
2 . the parameter @xmath60 , which is essentially the rms variation about the mean , increased with decreasing wavelength for the broad - band measurements as well as for the balmer emission lines . 3 .
the variations of the broad - band and emission - line fluxes are delayed with respect to the x - ray variations .
cross - correlation functions were calculated with three different methods ( iccf , dcf , and zdcf ) .
the time delays of the optical continuum variations expressed by the centroid of the cross - correlation functions are typically about 5 days , but with uncertainties of @xmath105 days .
therefore , zero - time delay between the high - energy and low - energy continuum variability can not be ruled out .
the delays of the balmer lines h@xmath5 , h@xmath6 , and h@xmath7 and of i5876 hei@xmath8are typically around 20 days @xmath18 8 days .
there is some evidence that ii4686 heii@xmath9 responds somewhat more rapidly with a time delay of @xmath1010 days , but again the uncertainties are quite large ( @xmath108 days ) .
4 . the simple cross - correlation analysis of the line core with the line profile wings of the h@xmath5 and h@xmath6 emission might indicate that the wings vary in the same way with respect to the line core .
but there might be a weak indication that the variations of the blue wing are delayed by @xmath10 4 days with respect to the red wing .
the mean and rms h@xmath5 and h@xmath6 line profiles reveal the existence of at least three substructures a central component , plus strong blue and red components at about @xmath11 ^ -1 km s@xmath12 relative to line center .
the rms spectra show that the broad - line variations are much stronger at line center than in the outer wings .
alef , w. , wu , s.y . , preuss , e. , kellermann , k.i . , & qui y.h .
1996 , a&a , 308 , 376 alexander , t. 1997 , in astronomical time series , ed d.maoz , a.sternberg , & e.leibowitz ( dordrecht : kluwer ) , in press allen , c.w .
1973 , astrophysical quantities , the athlone press alloin , d. , clavel , j. , peterson , b.m . ,
reichert , g.a .
, & stirpe , g.m . , 1994 , in frontiers of space and ground - based astronomy , ed .
w.wamsteker , m.s .
longair , & y. kondo ( dordrecht : kluwer ) , p. 423 alloin , d. , et al .
1995 , a&a , 293 , 293 barr , p. , et al . 1980 ,
mnras , 193 , 549 baum , s.a . , heckman , t. , bridle , a. , van breugel , w. , & miley , g. 1988 , apjs , 68 , 643 christian , c.a .
, et al . 1985 , pasp , 97 , 363 clavel , j. , et al . 1991 , apj , 366 , 64 clavel , j. , & wamsteker , w. 1987 , apj , 320 , l9 collier , s. , et al .
1997 in prep .
crenshaw , d.m .
, et al . 1996 , apj , 470 , 322 dietrich , m. , et al .
1993 , apj , 408 , 416 edelson , r.a . , &
krolik , j.h .
1988 , apj , 333 , 646 edelson , r.a . ,
1996 , apj , 470 , 364 ferland , g.j , korista , k.t . , & peterson , b.m .
1990 , apj , 363 , l21 gaskell , c.m .
, & sparke , l.s .
1986 , apj , 333 , 646 gaskell , c.m . , &
peterson , b.m .
1987 , apjs , 65 , 1 kaspi , s. et al .
1996 , apj , 470 , 336 koratkar , a.p . , &
gaskell , c.m .
1991 , apjs , 75 , 719 korista , k.t . , et al .
1995 , apjs , 75 , 719 leahy , j.p . , &
perley , r.a .
1995 , mnras , 277 , 1097 leighly , k.m . ,
1997 , apj , 483 , 767 lynds , c.r .
1968 , aj , 73 , 888 maoz , d. , & netzer , h. 1989 , mnras , 236 , 21 maoz , d. , et al .
1993 , apj , 404 , 576 obrien , p.t . , et al .
1998 , apjs , submitted osterbrock , d.e . ,
koski , a.t . , & phillips , m.m .
1975 , apj , 197 , l41 penston , m.j . ,
penston , m.v . , & sandage , a. 1971 , pasp , 83 , 783 penston , m.v . , & prez , e. 1984 , mnras , 211 , 33 peterson , b.m . , & collins , g.w . , ii .
1983 , apj , 270 , 71 peterson , b.m . , et al .
1991 , apj , 368 , 119 peterson , b.m .
, et al . 1993 , apj , 402 , 469 peterson , b.m .
1993 , pasp , 105 , 247 peterson , b.m . , et al .
1994 , apj , 425 , 622 peterson , b.m . , pogge , r.w .
, wanders , i. , smith , s.m . , & romanishin , w. 1995 , pasp , 107 , 579 reichert , g.a . , et al .
1994 , apj , 425 , 582 robinson , a. , & prez , e. 1990 , mnras , 244 , 138 rodrguez - pascual , p.m. , et al .
1997 , apjs , 110 , 9 sandage , a.r . 1966 ,
apj , 145 , 1 santos - lle , m. , et al .
1997 , apjs , in press schild , r.e . 1983 ,
pasp , 95 , 1021 selmes , r.a . ,
tritton , k.p . , & wordsworth , r.w .
1975 , mnras , 170 , 15 smith , e.p . , &
heckman , t.m .
1989 , apj , 341 , 658 stirpe , g.m . ,
1994 , apj , 425 , 609 van groningen , e. , & wanders , i. 1992 , pasp , 104 , 700 veilleux , s. , & zheng , w. 1991 , apj , 377 , 89 wanders i. , et al .
1997 , apj , submitted wamsteker , w. 1981 , a&a , 97 , 329 wamsteker , w. , ting - gui , w. , schartel , n. , & vio , r. 1997 , mnras , 288 225 warwick , r. , et al .
1996 , apj , 470 , 349 wills , b.j . ,
netzer , h. , & wills , d. 1985 , apj , 288 , 94 white , r.j . , & peterson , b.m .
1994 , pasp , 106 , 879 zheng , w. , prez , e. , grandi , s.a . , & penston , m.v .
1995 , aj , 109 , 2355 zheng , w. 1996 , aj , 111 , 1498 lccccccc perkins reflector , lowell obs . &
a&1.8&&&&&5x7.5shane reflector , lick obs.&b&3.0&&&&&4x10,2x5.6mt .
hopkins observatory & c & 1.6&&&&&3x3,2x3,1x3shajn reflector , crimean obs . &
d & 2.6&&&&&3x11beijing observatory & e & 2.2&&&& & 4x7.2 , 4x8.4 , & & & & & & & 4x12 , 4x10.3 & & & & & & & 4x3.6 , 2.5x9.6 & & & & & & & 3.5x9.6steward observatory & f&2.3&2.5 - 3.3 & 2.5 - 3.3 & 2.5 - 3.3&&2.5x12calar alto observatory & g1 & 1.2&&&25.0&25.0&calar alto observatory & g2 & 2.2&14.0&14.0&14.0&14.0&2x14.1,2x10mmt observatory & h & 4.5&&&&&2x10,1x7.2,2x6isaac newton telescope & i & 2.5&&&&&1.62x6.5special astrophysical obs . & j & 1.0&4.0,4.4&5.3,4.8&5.4,8.0&&special astrophysical obs . & j & 6.0&&&&&3x3.6,3.6x3.6mcdonald observatory & k & 2.7&&&&&2x7.2landessternwarte heidelberg&l & 0.7&&&15.0&15.0&james gregory tel . , st . andrews & m & 0.9&&47.0&47.0&47.0&roboscope , indiana univ . &
n & 0.4& & var.&&&wise observatory&o & 1.0&10.0&10.0&12.0&12.0&vainu bappu observatory & p1&2.3&&7.85&&&vainu bappu observatory & p2&1.0&&6.06&&&behlen observatory&q & 0.8&&13.7&&¢er for basement astrophys .
& r & 0.7& & 6.0 & 6.0 & 6.0 & hoher list obs . ,
bonn & s1 & 0.6&&3.0 & 3.0&&hoher list obs . , bonn & s2 & 0.4&&5.0 & 5.0&&sara telescope , kitt peak & t & 0.9&&&&20.0 & shanghai observatory & u & 1.6&&6 - 7.5&6 - 7.5&6 - 7.5& lcccc
n & & @xmath89 & & f & @xmath90&@xmath91 & @xmath92&g1 & & & @xmath93&@xmath94g2 & & & @xmath95&0.305j & @xmath96 & @xmath97&@xmath98 & m & & @xmath99&@xmath100&@xmath101o & & @xmath102&@xmath103 & @xmath104p1& & @xmath105 & & p2& & @xmath106 & & r & & @xmath107&@xmath108&@xmath109s1& & @xmath110&@xmath111&s2& & @xmath112&@xmath113&t & & & & @xmath114u & & @xmath115&@xmath116&@xmath117 lcclcc @xmath0 & 21&[email protected]&h@xmath7 & 55 & [email protected] @xmath1 & 244 & 1.4@xmath18 2.5&heii@xmath9&100 & 4.8@xmath18 9.1 @xmath2 & 206 & 1.8@xmath18 2.7&h@xmath6 & 104 & [email protected] @xmath3 & 149 & 2.4@xmath18 3.4&hei@xmath8 & 60 & [email protected] @xmath4(5177 ) & 97 & [email protected]&h@xmath5 & 84 & [email protected] lcc @xmath0 & 0.036 & 0.027 @xmath1 & 0.048 & 0.043 @xmath2 & 0.020 & 0.015 @xmath3 & 0.019 & 0.018 h@xmath7 & 0.062 & 0.055 heii@xmath9 & 0.145 & 0.137 h@xmath6 & 0.025 & 0.031 @xmath4(5177 ) & 0.040 & 0.041 hei@xmath8 & 0.106 & 0.099 h@xmath5 & 0.028 & 0.029 lcccc @xmath0 & @xmath133 & @xmath134&@xmath135&@xmath136@xmath1 & @xmath137 & @xmath138&@xmath139&@xmath140@xmath2 & @xmath141 & @xmath142&@xmath143&@xmath144@xmath3 & @xmath145 & @xmath146&@xmath147&@xmath148h@xmath7 & @xmath149&@xmath150&@xmath151&@xmath152heii@xmath9 & @xmath153&@xmath154&@xmath155&@xmath156h@xmath6 & @xmath157&@xmath158&@xmath159&@xmath160@xmath4(5177 ) & @xmath161&@xmath162&@xmath163&@xmath164hei@xmath8 & @xmath165&@xmath166&@xmath167&@xmath168h@xmath5 & @xmath169&@xmath170&@xmath171&@xmath172 lcccc @xmath1 & 126 & 89 & 53 & 1.5@xmath2 & 104&121 & 50 & 2.0@xmath3 & 82&132 & 53 & 2.5@xmath4(5177 ) & 55 & 69 & 53 & 1.8h@xmath7 & 42 & 36 & 49 & 1.7 heii@xmath9 & 58 & 8 & 33 & 1.8h@xmath6 & 59 & 64 & 52 & 1.8hei@xmath8 & 40 & 6 & 55 & 1.5h@xmath5 & 54&105 & 56 & 1.7 lccccccccc @xmath1 & 0.62 & 3.5 & 5.3 & 4.9&4.5 & 3.5 & 5.0 & 5.0&3.0 @xmath2 & 0.63 & 3.5 & 4.8 & 3.8&5.6 & 5.5 & 5.8 & 4.0&3.9@xmath3 & 0.61 & 3.5 & 4.5 & 3.7&6.2 & 7.9 & 7.0 & 4.7&4.2 @xmath4(5177 ) & 0.73&@xmath173 & 8.3&11.4&6.2 & 5.3 & 8.5 & 8.5&2.8 h@xmath7 & 0.66 & 24.5 & 23.7&24.1&6.2 & 16.8&13.8&20.8&5.5 heii@xmath9&0.37 & 13.0 & 5.9&10.4&8.3 & 8.6 & 4.6&17.8&7.8 h@xmath6 & 0.64 & 24.5 & 24.4&12.5&5.3 & 22.9&23.2&15.4&3.9 hei@xmath8 & 0.69 & 25.0 & 23.4&11.4&9.5 & 18.6&22.6&17.2&8.5 h@xmath5 & 0.53 & 29.0 & 18.7&11.2&10.3 & 19.3&20.6&15.9&9.5 llccccccccc h@xmath174 & h@xmath175 & 0.89 & 0.0 & 1.0 & -4.0 & 5.7 & 5.6 & 3.1 & 0.0 & 5.6 h@xmath174 & h@xmath176 & 0.84 & 0.5 & -1.2 & 0.0 & 8.6 & 0.3 & 0.4 & 8.8 & 8.7 h@xmath177 & h@xmath178 & 0.86 & 0.0 & 4.4 & -4.0 & 18.4 & 3.9 & 2.1 & -4.0 & 17.8 h@xmath177 & h@xmath179 & 0.87 & 0.0 & 1.2 & 0.0 & 22.6 & @xmath180 & -0.9 & -1.9 & 19.0 | results of a ground - based optical monitoring campaign on 3c390.3 in 199495 are presented . the broad - band fluxes ( @xmath0 , @xmath1 , @xmath2 , and @xmath3 ) , the spectrophotometric optical continuum flux @xmath4(5177 ) , and the integrated emission - line fluxes of h@xmath5 , h@xmath6 , h@xmath7 , i5876 hei@xmath8 , and
ii4686 heii@xmath9 all show a nearly monotonic increase with episodes of milder short - term variations superposed .
the amplitude of the continuum variations increases with decreasing wavelength ( 4400 9000 ) .
the optical continuum variations follow the variations in the ultraviolet and x - ray with time delays , measured from the centroids of the cross - correlation functions , typically around 5 days , but with uncertainties also typically around 5 days ; zero time delay between the high - energy and low - energy continuum variations can not be ruled out . the strong optical emission lines h@xmath5 ,
h@xmath6 , h@xmath7 , and i5876 hei@xmath8respond to the high - energy continuum variations with time delays typically about 20 days , with uncertainties of about 8 days . there is some evidence that ii4686 heii@xmath9 responds somewhat more rapidly , with a time delay of around 10 days , but again , the uncertainties are quite large ( @xmath108 days ) . the mean and rms spectra of the h@xmath5 and h@xmath6 line profiles provide indications for the existence of at least three distinct components located at @xmath11 and 0 ^ -1 km
s@xmath12 relative to the line peak .
the emission - line profile variations are largest near line center . |
humans are the basic constituents of the society , and every individual can interact with his / her family , friend and peers .
these interactions among individuals can induce some interesting collective behavior , such as , spontaneous formation of a common language or culture , emergence of consensus about a specific issue , and the adoptions of innovation , healthy or microfinance behavior . understanding the mechanisms or regularities behind these collective behavior
has led to a booming subfield of research in complex network science
social contagions , which has attracted much attention in recent years @xcite .
statistical physics approaches were widely used to investigate social contagions . on the one hand
, scientists used these methods to analyse large databases of social contagions , and revealed that reinforcement effect widely exists @xcite .
the reinforcement effect means that individual adopting a behavior is based on the memory of the cumulative behavioral information that he / she received from his / her neighbors .
centola established the artificially structured online communities to study health behavior spreading , and found that the reinforcement effect significantly increases the adoption of a new health behavior @xcite .
the reinforcement effect also exists in the adoptions of facebook @xcite and skype @xcite services . on the other hand , researchers proposed some novel models with reinforcement effect to describe the dynamics of social contagions . among these models ,
linear threshold model @xcite is a famous one , and it is a deterministic model ( i.e. , a trivial case of markovian process ) once the network topology and initial seeds are fixed . in this model , an individual will adopt the behavior once the current fraction of his / her adopted neighbours is larger than a static threshold .
the linear threshold model induces that the final behavior adoption size first grows continuously and then decreases discontinuously with the increasing of mean degree for vanishing small fraction of seeds .
another more realistic way to incorporate the reinforcement effect is whether an individual adopts the behavior should take his / her cumulative pieces of behavioral information into consideration @xcite . in this case , the dynamics is a non - markovian process , which makes it more difficult to develop an accurate theory .
et al _ proposed a non - markovian behavior spreading model , and found that the dependence of final behavior adoption size on information transmission probability can change from being discontinuous to being continuous under dynamical or structural parameters perturbation @xcite .
recently , scholars found that individuals exhibit limited contact capacity ( i.e. , individuals can only communicate or interact with a finite number of neighbors during a short time ) since the inelastic resources ( e.g. , time , funds , and energy ) restrict them to dedicate to social interaction from empirical analysis @xcite . in facebook communication networks , golder _ et al _ revealed that users only communicate with a small number of people even though they have many declared friends @xcite .
in scientific cooperation networks , a scientist exchanges knowledge with only a fraction of his / her cooperators in a paper @xcite . in sexual contact networks , individuals can not have sexual intercourse with his / her all sexual partners in a very short time due to the limitation of morality and physiology @xcite .
researchers have studied the effects of contact capacity on some markovian dynamics ( i.e. , epidemic spreading ) @xcite .
they found that the epidemic outbreak threshold increases when the contact capacity is limited @xcite .
meanwhile , each connection ( edge ) has distinct effective spreading probability ( to be defined in sec .
[ sec : theory ] ) , which makes the theoretical prediction deviate from simulation results more easily , especially in the case of strong structural heterogeneity . for the dynamics of social contagions , whether an individual adopts a behavior behavior or not is determined by the cumulative pieces of behavioral information that he / she has received from neighbors @xcite . once the contact capacity is limited , the behavioral information transmission will be limited , thus further affects the dynamics of social contagions .
however , a systematic study to understand the effects of contact capacity on dynamics of social contagions is still lacking . in this paper
, we try to address how the contact capacity affects the behavior spreading dynamics .
we first propose a non - markovian behavior spreading model with limited contact capacity , in which each adopted individual tries to transmit the behavioral information to a finite number of his / her neighbors . in order to understand ,
quantitatively , the effects of contact capacity on behavior spreading , we develop a heterogeneous edge - based compartmental theory .
we find that the final behavior adoption size increases with the contact capacity .
more interestingly , the crossover phenomenon in which the dependence of the final adoption size on the information transmission probability can change from being continuous to being discontinuous . by enlarging the contact capacity ,
the crossover phenomenon can be induced only when the degree exponent is above a critical critical degree exponent .
however , the final adoption size always grows continuously for any contact capacity when degree exponent is below the critical degree exponent .
the theoretical results from the suggested method can accurately predict the above results .
the paper is organized as follows . in sec .
[ sec : model ] , we describe the behavior spreading model with limited contact capacity .
we develop the heterogeneous edge - based compartmental theory in sec .
[ sec : theory ] . in sec .
[ sec : sim ] , we verify the effectiveness of the theory through large number of simulations . finally , we present conclusions and discussions in sec .
[ sec : dis ] .
we consider the behavior spreading on uncorrelated configuration networks @xcite with @xmath0 individuals ( nodes ) and degree distribution @xmath1 .
we use a generalized model sar ( susceptible - adopted - recovered ) model @xcite to describe behavior spreading on networks . at each time step , each individual can be in one of the three different states : susceptible , adopted , or recovered . in the susceptible state , an individual does not adopt the behavior . in the adopted state
, an individual adopts the behavior and tries to transmit the information to his / her selected neighbors . in the recovered state ,
an individual loses interest in the behavior and will not transmit the information further .
each individual holds a static adoption threshold @xmath2 , which reflects the criterion ( wills ) of an individual to adopt the behavior . initially , a fraction of @xmath3 individuals ( nodes ) are randomly selected to be in the adopted state ( seeds ) , while other individuals are in the susceptible states .
all susceptible individuals do not know any information about this behavior , in other words , the cumulative pieces of information is zero initially for all susceptible individuals . at each time step , each adopted individual @xmath4 with @xmath5 neighbors randomly chooses @xmath6 number of neighbors due to the contact capacity is limited , and tries to transmit the information to each selected neighbors with probability @xmath7 .
note that the function @xmath6 represents the contact capacity of @xmath4 , the larger value of @xmath6 , the more neighbors can receive the information from him / her . if @xmath8 , the contact capacity of individual @xmath4 is limited .
once the contact capacity of @xmath4 is larger than his / her degree , we let he / she transmit information to his / her all neighbors .
if @xmath4 transmits the information to @xmath9 successfully , the cumulative pieces of information @xmath10 that @xmath9 ever received will increase by @xmath11 , and the information can not be transmitted between @xmath9 and @xmath4 in the following spreading process ( i.e. , redundant information transmission on the edge is forbidden ) . if @xmath10 is larger than the adoption threshold @xmath2 , individual @xmath9 becomes adopted in the next time step . from the mentioned procedures of susceptible individuals becoming adopted , we learn that the dynamics of social contagion is a non - markovian process . the adopted individuals then lose interest in the behavior and enters into recovered with probability @xmath12 .
individuals in the recovered state do not take part in the spreading process .
the dynamics terminates once all adopted individuals become recovered .
the non - markovian behavior spreading model with limited contact capacity described in sec .
[ sec : model ] makes theoretical prediction from the classical theory ( e.g. , heterogeneous mean - field theory ) deviate from simulation results easily . on the one hand , in this proposed model , whether a susceptible individual adopts the behavior or not is dependent on the cumulative pieces of information he / she ever received . in this case , the memory effect of non - markovian process is induced . on the other hand
, the heterogeneity of effective spreading probability for edges increases with the heterogeneity of degree distribution , and further enhances the difficulty in developing an accurate theory .
the effective spreading probability of an edge includes two aspects : ( @xmath11 ) an edge is randomly selected with probability @xmath13 , where @xmath5 is the degree of adopted individual @xmath4 ; ( @xmath14 ) the information is transmitted through the selected edge with probability @xmath7 .
thus , the effective spreading probability of an edge for individual @xmath4 is @xmath15 . to describe this process
, we develop a novel theory heterogeneous edge - based compartmental theory , which is inspired by refs .
the theory is based on the assumption that behavior spreads on uncorrelated , large sparse networks .
we denote @xmath16 , @xmath17 and @xmath18 as the density of individuals in the susceptible , adopted and recovered states at time @xmath19 , respectively .
denoting @xmath20 as the probability that an individual @xmath4 with degree @xmath5 has not transmitted the information to individual @xmath9 along a randomly selected edge up to time @xmath19 . for simplicity
, we assume that individuals with identical degrees are the same in statistics . in the spirit of the cavity theory ,
we let individual @xmath9 in the cavity state ( i.e. , individual @xmath9 can not transmit information to his / her neighbors but can receive information from his / her neighbors ) . considering all possible degrees of individual @xmath4 , the average probability that individual @xmath9 has not received the information from his / her neighbors by time @xmath19 @xmath21 where @xmath22 represents the probability that an edge from @xmath9 connects to @xmath4 with degree @xmath5 in uncorrelated network , and @xmath23 is the mean degree . it is straightforward to get the probability that individual @xmath9 with @xmath24 neighbors has @xmath10 cumulate pieces of information by time @xmath19 @xmath25^{k - m}[1-\theta(t)]^m.\ ] ] the formula @xmath26 represents that only individuals in the susceptible state initially can get the information . from sec .
[ sec : model ] , we know that only when @xmath9 s cumulate pieces of information are less than @xmath2 , he / she can be susceptible at time @xmath19 .
thus , individual @xmath9 is susceptible by time @xmath19 with probability @xmath27 taking all possible values of @xmath24 into consideration , we can get the fraction ( density ) of susceptible individuals at time @xmath19 @xmath28 similarly , we can get the fraction of individuals who have received @xmath10 pieces of information at time @xmath19 @xmath29 according to the definition of @xmath20 , one can further divide it as @xmath30 the value of @xmath31 , @xmath32 , and @xmath33 represents that the probability of individual @xmath4 with degree @xmath5 is susceptible , adopted , and recovered and has not transmitted information to its neighbors ( e.g. , individual @xmath9 ) , respectively . an initial susceptible neighbor individual @xmath4 of @xmath9 can only get the information from the other @xmath34 neighbors , since individual @xmath9 is in the cavity state .
similar to eq .
( [ phi_k_t ] ) , one can get the probability that @xmath4 has @xmath10 cumulate pieces of information by time @xmath19 @xmath35^{k^\prime - m-1}[1-\theta(t)]^m.\ ] ] we further get the probability of individual @xmath4 in the susceptible @xmath36 if the adopted neighbor individual @xmath4 with degree @xmath5 transmits the information via an edge , this edge will not meet the definition of @xmath20 .
the conditions of individual @xmath4 transmits information to @xmath9 are : ( @xmath11 ) the edge connecting them is selected with probability @xmath13 and ( @xmath14 ) the information is transmitted through this edge with probability @xmath7 .
thus , the evolution of @xmath20 is @xmath37 if @xmath6 is larger than @xmath5 , we restrict that @xmath4 transmits the information to his / her all neighbors [ i.e. , @xmath38 .
according to information spreading process described in sec . [
sec : model ] , the growth of @xmath39 should simultaneously satisfy : ( @xmath11 ) the adopted individual @xmath4 does not transmit the information to @xmath9 through the edge between them and ( @xmath14 ) @xmath4 moves into recovered state with probability @xmath12 . for the first condition , there are two possible cases : the edge between @xmath9 and @xmath4 is selected with probability @xmath13 and the information is not transmitted through it with probability @xmath40 ; the edge between @xmath9 and @xmath4 is not selected with probability @xmath41 . from the analyses above ,
the evolution of @xmath39 is @xmath42.\ ] ] now , combining eqs .
( [ d_theta_k])-([d_xi_r ] ) and the initial situations [ i.e. , @xmath43 and @xmath44 , we obtain the expression of @xmath33 in terms of @xmath20 as @xmath45[\frac{k^\prime}{\lambda f(k^\prime)}-1].\ ] ] utilizing eqs .
( [ theta_k ] ) , ( [ neighbour_s ] ) , ( [ d_theta_k ] ) and ( [ xi_r ] ) , we obtain that @xmath46\\ & + \gamma[1-\theta_{k^\prime}(t)][1-\frac{\lambda f(k^\prime)}{k^\prime } ] .
\end{split}\ ] ] according to the model described in sec .
[ sec : model ] , the densities of individuals in adopted and recovered individuals evolve as @xmath47 and @xmath48 respectively .
( [ s_t ] ) and ( [ d_a_t])-([d_r_t ] ) give us a complete description of the social contagions with limited contact capacity .
the evolution of each type of density versus time can be obtained .
the densities of susceptible , adopted and recovered individuals do not change when @xmath49 .
we denote @xmath50 as the final behavior adoption size . to obtain the value of @xmath50
, one can first solve @xmath51 from eq .
( [ d_theta_k_con ] ) , that is @xmath52 + \gamma[1-\theta_{k^\prime}(\infty)][\frac{k^\prime}{\lambda f(k^\prime)}-1 ] .
\end{split}\ ] ] iterating eq .
( [ d_theta_k_steady ] ) to obtain @xmath51 .
then , inserting @xmath51 into eqs .
( [ theta_mean])-([s_t ] ) to get the values of @xmath53 and @xmath54 .
another important aspect we mainly focus on is the condition under which the global behavior adoption occurs .
the global behavior adoption means that a finite fraction of individuals adopted the behavior , and the corresponding local behavior adoption represents that only a vanishingly small fraction of individuals adopted the behavior .
similar to biological contagions , we define a critical transmission probability @xmath55 . when @xmath56 , the behavior can not be adopted by a finite fraction of individuals ; when @xmath57 , the global behavior adoption occurs .
now , we discuss @xmath55 for several different values of @xmath3 and @xmath2 . for @xmath58 ( i.e. , only a vanishing small fraction of seeds ) and @xmath59
, @xmath60 is the trivial solution of eq .
( [ d_theta_k_steady ] ) .
if we change the values of other dynamical parameters , such as information transmission probability @xmath7 , a global behavior adoption may occur .
the global behavior adoption occurs only when a nontrivial solution of eq .
( [ d_theta_k_steady ] ) emerges [ i.e. , @xmath61 .
note that the corresponding fraction of @xmath51 should be taken into consideration .
linearizing eq .
( [ d_theta_k_steady ] ) at @xmath60 @xcite , and summing all possible values of @xmath5 one can get the critical information transmission probability @xmath62 where @xmath63 note that @xmath55 is tightly correlated with the network topology [ i.e. , degree distribution @xmath1 ] and dynamical parameters [ i.e. , contact capacity @xmath6 and recover probability @xmath12 ] .
hubs in heterogeneous networks adopt the behavior with large probability .
thus , the value of @xmath55 decreases with degree heterogeneity .
the value of @xmath55 increases with @xmath6 , in other words , increasing the contact capacity of individuals makes the network more fragile to the behavior spreading .
we should emphasize that @xmath12 also affects @xmath66 ( i.e. , the effective critical information transmission probability ) , which has been neglected in previous studies .
the value of @xmath66 increases with @xmath12 .
if @xmath67 for every value of @xmath5 ( i.e. , adopted individual transmits the information to his / her all neighbors ) , eq .
( [ lambda_c ] ) is the epidemic outbreak threshold @xcite .
if every adopted individual only transmits information to his / her @xmath68 neighbors , we can get the critical transmission probability @xmath55 of the model in ref . @xcite . for @xmath58 and @xmath69 , we find that @xmath60 is the solution of eq .
( [ d_theta_k_steady ] ) . however , the left and right hands of eq .
( [ d_theta_k_steady ] ) can not be tangent to each other at @xmath70 , which indicates that vanishingly small seeds can not trigger the global behavior adoption @xcite . with the increase of @xmath3 , different dependence of @xmath50 on @xmath7 occurs for different @xmath2 .
that is , the growth pattern of @xmath50 versus @xmath7 can be continuous or discontinuous . through bifurcation analysis @xcite of eq .
( [ d_theta_k_steady ] ) , we find that @xmath50 grows continuously for @xmath59 , while a discontinuous growth may be induced for @xmath69 .
in this section , we verify the effectiveness of the heterogeneous edge - based compartmental theory developed in sec .
[ sec : theory ] by lots of simulations . for each network
, we perform at least @xmath71 times for a dynamic process and measure the final fraction of individuals in the recovered [ @xmath50 ] and subcritical state [ @xmath73 . these results are then averaged over 100 network realizations .
to built the network topology , we use the uncorrelated configuration model @xcite according to the given degree distribution @xmath74 with maximal degree @xmath75 .
there is no degree - degree correlations when @xmath0 is very large .
the heterogeneity of network increases with the decrease of @xmath76 . for
the sake of investigating the effects of heterogeneous structural properties on the social contagions directly , the network sizes and mean degree are set to be @xmath77 and @xmath78 , respectively .
all individuals with different degrees have the same contact capacity @xmath79 and recover probability @xmath80 .
we first study the effects of the adoption threshold @xmath2 and contact capacity @xmath68 on the final behavior adoption size @xmath50 for strong heterogeneous networks in fig .
we find that @xmath50 decreases with the increase of @xmath2 , since individuals adopting the behavior need to expose more information .
once the contact capacity increases ( i.e. , @xmath68 increases ) , individuals in adopted state will have more chances to transmit the information to susceptible individuals , thus , the values of @xmath50 increases . obviously , the theoretical predictions from heterogeneous edge - based compartmental theory agree well with the simulation results . another important issue we concern is the dependence of @xmath50 on @xmath7 . as shown in fig .
[ fig1 ] , for strong heterogeneous networks the dependence of @xmath50 on @xmath7 is continuous for any values of @xmath2 and @xmath68 , and we verify this claim by the bifurcation analysis of eq .
( [ d_theta_k_steady ] )
. we can also understand this phenomenon by discussing the fraction of individuals in the subcritical state from an intuitive perspective ( see fig .
[ fig2 ] ) .
an individual in the subcritical state means that he / she is in the susceptible state , and the @xmath10 cumulative pieces of information is just one smaller than his / her adoption threshold @xmath2 . from ref .
@xcite , we know that a discontinuous dependence of @xmath50 on @xmath7 will occur only when a large number of those subcritical individuals adopt the behavior simultaneously at some information transmission probability . fig .
[ fig2 ] shows the final fraction of individuals in the subcritical state @xmath81 versus @xmath7 .
we find that @xmath81 first increases and then decreases gradually with @xmath7 , since the existence of strong degree heterogeneity makes individuals in the subcritical state adopt the behavior consecutively . in these cases , a continuous growth of @xmath50 versus @xmath7 occurs on strong heterogeneous networks .
we now study behavior spreading on weak heterogeneous networks , such as @xmath82 in fig .
similar with the case of @xmath83 , increasing @xmath2 leads to the decrease of @xmath50 ; and the value of @xmath50 increases with @xmath68 , that is the network will become more fragile to the behavior spreading once the contact capacity increases . once again , our theory can predict the social dynamics very well . for the dependence of @xmath50 on @xmath7 ,
a crossover phenomenon transition is observed .
a crossover phenomenon means that the dependence of @xmath50 on @xmath7 can change from being continuous to being discontinuous .
more specifically , as shown in fig .
[ fig3](b ) , the dependence of @xmath50 on @xmath7 is continuous for small values of @xmath68 ( e.g. , @xmath84 ) , while the dependence is discontinuous for larger @xmath68 ( e.g. , @xmath85 ) .
we justify this claim by the bifurcation analysis of eq .
( [ d_theta_k_steady ] ) from the theoretical view , which is also verified through analyzing @xmath81 from an intuitive perspective in fig .
[ fig4 ] . for weak heterogeneous networks ,
most individuals adopt the behavior with the same probability since they have similar degrees .
when @xmath84 , @xmath81 increases continuously with @xmath7 , which leads to a continuous growth in the value of @xmath50 .
when @xmath85 , @xmath81 first increases with @xmath7 , and reach a maximum at some values @xmath55 , and a slight increment of @xmath7 induces a finite fraction of @xmath86 to adopt the behavior simultaneously , which leads to a discontinuous jump in the value of @xmath50 .
we further study the effects of @xmath76 and @xmath7 in fig .
[ fig5 ] for different values of @xmath68 . for small contact capacity [ i.e. , @xmath84 in figs . [ fig5](a ) and ( b ) ] , the dependence of @xmath50 on @xmath7 is always continuous for any value of @xmath76 . in other words ,
this dependence is irrelevant to the network topology . for large contact capacity [ i.e. , @xmath85 in figs .
[ fig5](c ) and ( d ) ] , there is a crossover phenomenon in which the dependence of @xmath50 on @xmath7 can change from being continuous to being discontinuous .
more particularly , there is a critical degree exponent @xmath87 below which the dependence is continuous [ see region i in figs .
[ fig5](c ) and ( d ) ] , while above @xmath87 the dependence is discontinuous [ see region ii in figs .
[ fig5](c ) and ( d ) ] .
the value of @xmath87 can be gotten by bifurcation analysis of eq .
( [ d_theta_k_steady ] ) . in region
ii , we also find that the discontinuous information transmission probability @xmath88 increases with @xmath76 , since the fraction of hubs decreases with @xmath76 .
the theoretical predictions of @xmath88 can be gotten by bifurcation analysis of eq .
( [ d_theta_k_steady ] ) , and the simulation results of @xmath88 are predicted by noi ( number of iterations ) method @xcite .
regardless of network heterogeneity , our theoretical predictions about the behaviors of @xmath50 have a good agreement with numerical calculations .
the average relative error @xcite between the two predictions of @xmath50 for all the values of @xmath7 and @xmath76 is less than @xmath89 . finally , we study the effects of network topology on the final behavior adoption size @xmath50 in fig .
[ fig6 ] for @xmath90 and @xmath84 .
we find that increasing @xmath76 can promote ( suppress ) behavior adoption at large ( small ) value of @xmath7 .
this phenomenon can be qualitatively understood in the in following ways @xcite : for strong heterogeneous networks , the more hubs and a large number of individuals with small degrees are coexisted .
since those hubs have more chances to expose the information , they adopt the behavior more easily even when @xmath7 is small .
however , the situations for individuals with small degree are just opposite .
that is , the large number of individuals with small degrees hinder the behavior adoption for large value of @xmath7 , thus , cause a smaller value of @xmath50 . through bifurcation analysis of eq .
( [ d_theta_k_steady ] ) , the dependence of @xmath50 on @xmath7 is always continuous for different @xmath76 .
to study social contagion dynamics in human populations is an extremely challenging problem with broad implications and interest . for social contagions on networks , some inelastic resources
( e.g. , time , funds , and energy ) restrict individuals to dedicate to social interaction , which have always been neglected in previous studies . in this paper
, we first proposed a non - markovian behavior spreading model with limited contact capacity , in which each adopted individual transmits the information to a fraction of his / her neighbors .
we then developed a heterogeneous edge - based compartmental theory to describe this model .
the average relative error between the theoretical predictions and numerical calculations is less than @xmath89 . through theory and simulations
, we found that increasing the contact capacity @xmath68 promotes the final behavior adoption size @xmath50 . with the help of bifurcation theory
, we found a crossover phenomenon in which the dependence of @xmath50 on @xmath7 can change from being continuous to being discontinuous . more specifically , we uncovered a critical degree exponent @xmath87 above which the crossover phenomenon can be induced by enlarging @xmath68 .
however , @xmath50 always grows continuously for any value of @xmath68 when degree exponent is below @xmath87 .
here we developed an accurate theoretical framework for non - markovian social contagion model with limited contact capacity , which could be applied to other analogous dynamical processes ( e.g. , information diffusion and cascading failure ) .
further more , how to design an effective strategy to control the behavior spreading with limited contact capacity is an interesting research topic .
this work was partially supported by the national natural science foundation of china under grant nos . 11105025 , 91324002 and the program of outstanding ph .
d. candidate in academic research by uestc under grand no .
yxbszc20131065 .
g. miritello , r. lara , and e. moro , _ time allocation in social networks : correlation between social structure and human communication dynamics _ in , _ temporal networks _ edited by p. holme , and j. saramki , springer complexity ( 2013 ) .
s. a. golder , d. wilkinson , and b. a. huberman , _ rhythms of social interaction : messaging within a massive online network .
_ in , c. steinfield , b. pentland , m. ackerman , n. contractor ( eds ) proceedings of third international conference on communities and technologies , pp 41 - 66 ( 2008 ) . | individuals are always limited by some inelastic resources , such as time and energy , which restrict them to dedicate to social interaction and limit their contact capacity .
contact capacity plays an important role in dynamics of social contagions , which so far has eluded theoretical analysis . in this paper , we first propose a non - markovian model to understand the effects of contact capacity on social contagions , in which each individual can only contact and transmit the information to a finite number of neighbors .
we then develop a heterogeneous edge - based compartmental theory for this model , and a remarkable agreement with simulations is obtained . through theory and simulations , we find that enlarging the contact capacity makes the network more fragile to behavior spreading .
interestingly , we find that both the continuous and discontinuous dependence of the final adoption size on the information transmission probability can arise .
and there is a crossover phenomenon between the two types of dependence .
more specifically , the crossover phenomenon can be induced by enlarging the contact capacity only when the degree exponent is above a critical degree exponent , while the the final behavior adoption size always grows continuously for any contact capacity when degree exponent is below the critical degree exponent . *
recent empirical studies of facebook communication networks , scientific cooperation networks and sexual contact networks suggest that individuals activities are limited by the time , funds , energy and other inelastic resources .
thus , individuals exhibit limited contact capacity ( i.e. , individuals can only communicate or interact with a finite number of neighbors during a short time ) in the dynamics of epidemic and behavior .
previous studies have proven that limited contact capacity enlarges the epidemic outbreak threshold and makes the theoretical prediction deviate from simulation results more easily .
unfortunately , a systematical investigation the effects of contact capacity on the dynamics of social contagions is still lacking . to fill this gap ,
we first propose a non - markovian behavior spreading model with limited contact capacity .
then , we develop a novel heterogeneous edge - based compartmental theory for this proposed model , and verify the effectiveness of this suggested theory based on large number of simulations . through theory and simulations , we find that the final behavior adoption size increases with the contact capacity . strikingly , we uncover a crossover phenomenon in which the dependence of the final adoption size on the information transmission probability can change from being continuous to being discontinuous .
we find a critical degree exponent above which the crossover phenomenon can be induced by enlarging the contact capacity .
however , the final adoption size always grows continuously for any contact capacity when degree exponent is below this critical degree exponent .
our results help us to have a deeper understanding of the effects of contact capacity on social contagions , and the developed theory could be applied to other analogous dynamical processes ( e.g. , information diffusion and cascading failure ) . * |
in this talk , i review the work collaborated with covi _
@xcite on the possibility of the cdm axino .
modern cosmology needs dark matter and dark energy in the universe : @xmath0 .
there are several particle candidates for the cdm : heavy neutrino in the gev range , axion in the @xmath1 ev range , the lightest superpartner(lsp ) in the 100 gev range , wimpzilla in the @xmath2 gev range , axino , and other hypothetical heavy particles .
the particle mass and the interaction strength are provided by particle physics . in the standard big bang cosmology ,
this information leads to the decoupling temperature(@xmath3 ) below which the number of particles in the comoving volume is preserved .
but the needed inflationary period reheats the universe , after a period of supercoolong , to the reheating temperature @xmath4 .
if @xmath5 , the particles are mostly diluted out and the first guess on its number density is that it can not serve as dark matter(dm ) . but even after the reheating phase , the thermal production can be effective enough to create sufficient number of heavy particles to close the universe .
it has been most extensively studied for the case of o(100 gev ) gravitino@xcite . because of the gravitino problem , we consider the reheating temperature below @xmath6 gev . for @xmath7 , the dm candidates are * neutrinos : @xmath8 is the interaction strength .
* lsp : @xmath9 is the interaction strength . for @xmath5 , the dm candidates are * axion : @xmath10 is the interaction srength
. collective motion of axions below the qcd phase transion contributes .
* axino : @xmath10 is the interaction strength .
thermal and nonthermal productions contribute .
before discussing the axino cdm possibility , let us briefly review the cold axions , collectively moving in the universe .
the basic difference of the axion candidate from the other ones is that axion is boson and it is possible to have a collective motion .
the classical axion potential is extremely flat due to the very tiny invisible axion mass@xcite .
because of the extremely weak interaction strength , the axion vacuum stays at a point @xmath11 for a long time .
the vacuum starts to oscillate when the hubble time ( @xmath12 ) is larger than the oscillation period ( @xmath13),@xmath14 which occurs when the temperature is about 1 gev@xcite .
we understand that the axion is the pseudo - scalar field , having the interaction only through the anomaly @xmath15 where @xmath16 is the dual field strength @xmath17 .
we stress that there should not exist any other term in the axion potential .
this kind of nonrenormalizable interaction can arise in several different ways : * from fundamental interaction such as in superstring@xcite : @xmath18 planck mass . * from composite models@xcite : @xmath18 compositeness scale . * from renormalizable theories@xcite : a global symmetry can have a gluon anomaly .
if this global symmetry is spontaneously broken , there arises a goldstone boson coupling to the gluon anomaly . @xmath18
the global symmetry breaking scale .
then the current axion energy density is estimated to be@xcite @xmath19 if @xmath20 is large(@xmath21 gev ) , the axion energy dominates the current energy density in the universe .
since the energy density is proportional to the number density , it behaves like a cdm . on the other hand ,
if @xmath20 is small then the axion interaction is relatively strong and too much axions are produced in the core of stars .
sn1987a restricts its lower bound around @xmath6 gev , leading to the axion window @xmath22 the gravitino produced thermally after inflation decays very late in the cosmic time scale(@xmath23
s ) , and can dissociate the light nuclei by its decay products@xcite .
inflation was used to dilute the primordial gravitinos , but thermal production of gravitinos are troublesome in cosmology if it exceeds@xcite @xmath24 therefore , in supersymmetric theories we must consider relatively small reheating temperature after inflation .
therefore , in supersymmetric extension of the axion model , axion can be a cdm candidate but it is produced very late .
its supersymmetric partner is axino , and the reheating temperature @xmath4 must be smaller than @xmath3 .
the axion @xmath25 and axino @xmath26 are gathered in a supermultiplet , @xmath27 where @xmath28 is the scalar partner of axion , the saxion .
the axion couples to the gluon anomaly through @xmath29 .
it is known that the potential arising from this gluon interaction settles @xmath30 at zero , which is the peccei - quinn mechanism .
for axino to be cdm , it must be stable or practically stable . without the @xmath31-parity conservation
, this can not happen .
thus , we require a practical _ r - parity conservation_. for axino to be the lsp , it must be lighter than the lightest neutralino whose mass is expected to be around 100 gev . thus , an estimate of the axino mass is of prime importance for a cdm axino .
note that there is no theoretical upper bound on the axino mass .
since axion is almost massless , one might expect that the axino and saxion are almost massless in the first approximation .
however , the saxion obtains a soft mass term below susy breaking scale .
it is like the sm susy scalars .
so the axino mass is intimately reated to the susy breaking scenario also .
actually , it is known that the axino obtains a substantial mass@xcite . for a specific model , including the soft susy breaking terms , consider @xmath32 where @xmath33 and @xmath34 are the sm singlets , and the supersymmetric term comes from the superpotential @xmath35 . since @xmath36 is of order @xmath20 , @xmath37 is of order the @xmath38 term .
thus , the fermionic partners have the mass matrix of the form , @xmath39 where @xmath40 .
one eigenvalue is the axino mass @xmath41 , and the other masses are of order @xmath20 . as seen from this example
, the axino mass is basically a free parameter , but is expected to be somewhat smaller than the naive susy breaking scale due to the coupling @xmath42@xcite .
but in some models , the axino mass can be much smaller than the susy breaking scale .
take a superpotential @xmath43 with one more singlet @xmath44 which carries a vanishing pq charge .
this superpotential is much more complicated to analyze , but still we can show that @xmath45 . for the standard pattern of soft terms , @xmath46 .
then , the axino mass is of order kev .
thus , even the tree level axino mass needs the knowledge on the full superpotential@xcite .
therefore , we can consider the axino mass a free parameter , and we restrict our discussion in the region where axino is the lsp , which is the most probable choice with the pq symmetry .
the universal axion coupling is to the anomaly term .
therefore , the axino coupling to the gauge multiplet is the most important couplings . for the fermions , only the coupling to the top quark
is important .
we consider all these couplings when they become appropriate .
the axino decoupling temperature is @xmath47 where @xmath48 is the domain wall number in axion models and @xmath49 is the axion decay constant in units of @xmath50 gev . for @xmath7 , at @xmath3 the number density is determined . for axinos not to close the universe , it should not exceed 12.8 ev @xmath51 where @xmath52 , which was used for the o(@xmath53 kev ) warm dm axino by rajagopal _ et al._@xcite .
but in our @xmath54 case , due to the gravitino problem@xcite , we do not consider this region seriously . on the other hand
, we can consider much heavier axinos .
the estimation of the axino density in the universe is considered for the thermal case and for the non - thermal case(by the neutralino decay ) . for the thermal production ,
we solve the boltzmann equation , @xmath55 the axino yield , the number density divided by the entropy @xmath56 , is split into two pieces one from the scattering process and the other by the decay , @xmath57 where @xmath58 for the scattering , we consider ten important processes shown in table 1 .
.the cross sections for the different axino thermal production channels involving strong interactions .
masses are neglected except for the plasmon mass @xmath59 . [ cols="<,^,<,^,^,^ " , ] [ tab1.1a ] on the other hand , the non - thermal production is basically by the decay process of the very long lived neutralino@xcite .
these contributions are shown in fig .
1 . for a low reheating temperature ,
the thermal production is not effective . in fig .
1 , the bulge at the lower left corner is the non - thermal production . as a function of @xmath4 for @xmath60 gev and @xmath61 tev . ] in fig . 1
, we show for @xmath60 gev so that the cold axions are not the dominant component of cdm . for the calculation of the non - thermal production , we consider losp and nlsp where losp is the lightest superpartner among ordinary sm particles and nlsp is the next - to - lsp or the second lightest superpartner .
it is most likely that the losp is the lightest bino - like neutralino and nlsp can be losp or gravitino or something else . for a bino - like losp @xmath62 ,
the decoupling temperature of the losp is @xmath63 . at @xmath64 and @xmath65 ,
the losp decays during the radiation dominated era to produce axinos through @xmath66 to yield@xcite @xmath67 therefore , we have a simple expression , @xmath68 the neutralino decays at the cosmic time @xmath69 depending on the models , the decay modes can be different . if there are more channels for the @xmath62 decay , then it will give an even more efficient implementation of the non - themal production of axino through the neutralino decay .
we must impose the conditions that ( 1 ) not too much axino energy density at the time of big bang nucleosynthesis(bbn ) , ( 2 ) the bbn not spoiled by @xmath62 decay producing the sm particles , and ( 3 ) axinos becoming cold before the galaxy formation era . note in passing that if @xmath70 , i.e. for the gravitino nlsp then the gravitino problem disappears since it decays to axion and axino@xcite . on the other hand ,
if @xmath62 is the nlsp , the gravitino problem is present@xcite . in this case , the non - thermally produced axinos are estimated to give eq .
( [ number ] ) . and axino mass . ] in fig .
2 , we include all these constraints and plot the restriction on the reheating temperature versus axino mass .
this figure is not changed even if there is no gravitino problem because of the gravitino nlsp .
the shaded region is where the axino can be a cdm possibility .
this happens o(gev ) axino .
note that for a high reheating temperature the thermal production contribution dominates . in the figure
, there is the solid line denoting the critical energy density by axinos .
even if the reheating temperature is below the critical energy density line , there still exists a axino cdm possibility by non - thermally produced axinos@xcite .
note that non - thermally produced axinos can be cdm for relatively low reheating temperature(@xmath71 tev ) for which axino mass should be @xmath72 the shaded region corresponds to the mssm models with the constraint @xmath73 but still allowable axino energy density .
one can choose 30 % cdm in this region .
however , if we restrict further that all susy mass parameters are below 1 tev , then we have @xmath74 . for a sufficient axino energy density for the axino cdm
, then we have @xmath75 but less than the neutralino mass .
if @xmath31-parity is exact , then the cdm axino can not decay , and there is no way to prove the cdm axino possibility . if there is any chance to prove the cdm axino by observation , then the axino must decay@xcite .
for it to constitute most of cdm in the universe , its lifetime must be of order the age of the universe . since it is assumed to be the lsp , the @xmath31-parity must be feebly broken .
then the cdm axino can decay .
for the simplicity of discussion , let us consider the bilinear terms only @xmath76 where @xmath77 is the flavor index , @xmath78(ev ) is the constraint from the neutrino mass bound .
this bound applies to the heaviest neutrino , presumably @xmath79 . in this case
@xmath80 is bounded as @xmath81 with the @xmath31-parity violation , we expect the decays @xmath82 in fact , the @xmath83 mode is most important in cosmology . for the cosmic importance of the decay , we need to know the lifetime .
let us try phenomenological lagrangians , @xmath84\psi,\ ] ] for the final pseudo - scalar and photon , respectively .
then , for each case the lifetime can be estimated .
in particular , to the photon mode @xmath85\ ] ] where @xmath86 is the phase space factor .
@xmath87 is known if the @xmath31-parity violation is given . since the most stringent bound comes from the diffuse gamma ray , we focus on this mode .
the observed gamma ray flux is @xmath88.\ ] ] if @xmath89 where @xmath90 is the age of the universe , most axinos have decayed by now and the flux has a peak at @xmath91 .
since the photons from axino decay should not exceed the observed flux , we obtain @xmath92 which is in conflict with @xmath93 s. if @xmath94 , a similar consideration gives the axino lifetime larger than @xmath95 s. this translates to the @xmath80 bound , @xmath96 if @xmath80 in ( [ mu3 ] ) is taken , the idea of neutrino mass generation by @xmath31-parity violation is not enough .
thus , the unstable cdm axino is not viable with the neutrino mass generation by @xmath31-parity violation .
still , the cdm axino decay can be detected if the parameters lie in the region given above .
in conclusion , we discussed * supersymmetry is the solution of the gauge hierarchy problem , and the pq symmetry with a very light axion with @xmath97 gev is the solution of the strong cp problem .
this leads to the axino which can be the lsp . *
the gravitino problem must be considered in this case , with @xmath98 gev . * due to the much more stronger axino production than the gravitino production , @xmath4 must be much smaller than the one given by the gravitino consideration .
it can be in a multi tev region . *
thermal and nonthermal production of axinos are possible . in this case , o(gev ) axino cdm possibility exists . * finally , the attempt to explain the neutrino mass via the @xmath31-parity violation mechanism is inconsistent with the cdm axino .
however , in this case there is a room for the detection of axino decay debris .
j. ellis , j. e. kim , and d. v. nanopoulos , phys . lett . * b145 * , 181 ( 1984 ) ; t. moroi , h. murayama , and m. yamaguchi , phys . lett . * b303 * , 289 ( 1993 ) ; m. bolz , w. buchm@xmath99ller , and m. pl@xmath99macher , phys . lett . * b443 * , 209 ( 1998 ) . r. d. peccei and h. r. quinn , phys . rev . lett .
* 38 * , 1440 ( 1977 ) ; s. weinberg , phys .
* 40 * , 223 ( 1978 ) ; f. wilczek , phys .
40 * , 279 ( 1978 ) ; j. e. kim , phys .
* 43 * , 103 ( 1979 ) . | the possibility of the cdm axino is presented . from the early days of supersymmetry
, axino has been considered as hot dark matter and warm dark matter .
but the cdm axino has been considered quite recently .
it is because the low reheating temperature of the universe is gaining momentum recently from various considerations in particle cosmology .
i present a general introduction on the cdm s , in particular the axino , the calculation on the axino density in the universe , and the constraints imposed on the axino parameters . |
study of the flow structure in astrophysics is important , because the results could be used both for consideration of the evolution status of binary stars and for the interpretation of observational data .
therefore , when investigating close components , it is necessary to include physical essence of the flow dynamics response to the interaction processes . in the astrophysics of binary stars , one of the most investigated phenomena are vortexes .
vortices and spiral configurations play a key role in the accretion disc dynamics , because they are considered as an efficient mechanism of angular momentum transportation ( baranco & marcus 2005 ) in regions where the magneto - rotational instability ( balbus & hawly 1998 ) does not operate . at present , there are many hydrodynamics studies in finding the way of vortices appear and behave in the flow .
( 2000 , 2001 ) have shown that vortices are formed by rossby waves instability and moving radially and thus transport mass through the disc .
vortices can be generated by a globally unstable radial entropy gradient ( klahr & bodenheimer 2003 ) that may result in local outward transport of angular momentum . in astrophysics , the problems of structures development have been investigated mainly numerically ( e.g. lithwick 2009 ; godon & livio 1999 ) .
bracco et al .
( 1998 ) by using two - dimensional , incompressible fluid dynamics , show that anticyclonic vortices shift out and that smaller vortices merge to form larger vortices .
godon & livio ( 1999 , 2000 ) confirm this result with two - dimensional , compressible , barotropic simulations .
shen et al . ( 2006 ) examine the formation of 2d vortices starting from 2d turbulence in fully compressible simulations .
barranco & marcus ( 2005 ) compute the evolution of 3d vortices and show that part of the vortical formations could be destroyed , but the other part survive for several hundreds of orbits .
they use an `` inelastic code '' with vertical stratification and examine the influence of 3d perturbations over the vortices in the middle disc s zone . by performing a series of runs with zero initial vorticity and perturbation wavelengths ,
johnson and gammie ( 2006 ) give a very realistic way of the initial vorticity generation .
they have noted that the remaining vorticity can be generated from `` finite - amplitude compressive perturbations '' .
johnson and gammie ( 2005 ) argue that the vortices are `` long - lived '' assuming they could not be conserved in 3d calculations .
lesur and papaloizou ( 2009 ) have obtained that most anticyclonic vortices are unstable .
their simulations show that only the vortex core is stable and that the instability appears at the vortex boundary .
formation of vortices could be accompaned with spirals and vice - versa .
fridman ( 2007 ) confirms that for one arm spiral waves in the disc , based on the spiral wave theory .
sawada , in his paper ( sawada et al . 1986 ) , finds that a two - armed spiral structure developed in the result of 2-d simulations of mass transfer via roche lobe overflow .
their research has been followed by a series of investigations ( e.g. rozyczka & spruit 1993 ) , finding that during the stage of spirals development the appearance of vortices are observed , after the interaction event . in our previous studies
we have found that presence of one - armed spiral structure could be arised as an effect of tidal interaction of two stars in binary system ( boneva 2010(1 ) , boneva 2010(2 ) ) .
the same spiral wave structures are observed in the calculations of ( bisikalo et al 2008 ; steeghs et al .
currently , the question of how the structures in accretion discs arise is still open for discussion , along with numerous unresolved problems concerning the formation mechanism and the duration of vortices existence in discs . by applying numerical calculations on gas - dynamical flow
, we suggest modeling of patterns formation and explanation of supporting physical processes in interacting flows .
the paper is organized as follows : section 2 presents some base equations and the methods of analysis on it . in section 3
we introduce the main results that reveal the variations in disc s and vicinity s density .
the development of vortical structures formation is examined and described in section 4 .
the nature of the interaction between a flow of matter and an envelope of two star components requires employment of gas - dynamics equations .
therefore , to obtain solutions of the above stated problem a system of equations is needed .
herein , the basic equations are presented in a form that have been suggested and affirmed by many authors : ( shore 2007 ; clark & carswell 2007 ; frank et al .
2002 ; graham 2001 ; shu 1992 ) .
we present the equations in their vector form .
the equation of mass conservation is : @xmath0 the existence of viscous processes in the accretion flow , as well as influence of forces and rotation could be performed by the following useful form of the navier - stokes equations , suggested by ( thorn 2004 ) and simplified here : @xmath1 where the basic notations are : @xmath2 is the mass density of the flow , @xmath3 - is the velocity of the flow ; @xmath4- is the pressure ; @xmath5- is the kinematic viscosity ; @xmath6 - is the angular velocity ; @xmath7 - is the centrifugal acceleration of the centrifugal force ; and @xmath8 - is the coriolis acceleration in the mean of the coriolis force . in the current analysis @xmath9 and @xmath10 .
@xmath11 is the gravitational potential and it depends on the density distribution inside each of the star s component ( boyarchuk et al .
then the gravitational potentials @xmath12 and @xmath13 could be defined from the poisson s equations : @xmath14 and @xmath15 the energy balance equation for a viscous non - ideal fluid is : @xmath16+\nabla .\left[\rho v\left(\frac{1}{2 } v^{2 } + h+\phi \right)-2\eta \sigma .v\right]=0;\end{aligned}\ ] ] where @xmath17 $ ] is the total energy density , where the first term on the left denotes the kinetic energy , the second is the internal energy and the third expresses again the full potential of the gravitational fields . and @xmath18 $ ] is the total energy flux , where @xmath19 is the enthalpy , @xmath20 is the shear ( or dynamical ) viscosity of the flow , and @xmath21 is the rate of shear . the equation of state for compressible flow is : @xmath22 where @xmath23is the sound speed .
we present the equations in the above system in their common form and we can easily transform them into quantities for each of the posted problems . next to explore is the vortical transport equation , because it is related to the examination of transfer s mechanisms in the flow .
this equation could be derived in the following commonly used way , how it has been done by nauta ( 2000 ) , lithwick ( 2007 ) , godon ( 1997 ) .
if the curl of navier - stokes equations is considered , and use the expressions : @xmath24 - expressing the vorticity in the flow ; @xmath25 and @xmath26 ; then the following expression is obtained : @xmath27 here @xmath28- is the vorticity ; @xmath29- is the diffusion coefficient ( or matrix of the transport coefficient ) .
this equation expresses the relation between the transport coefficient , which takes part in the angular momentum transfer , evolution of the vorticity with time and the non - conserve relationship between density and pressure in the flow .
non - conservancy of specific vorticity by the each fluid element is observed in the right - hand side of eq .
( [ 5 ] ) , which gives a condition for baroclinicity in the flow ( lovelace et all .
1999 ) , and enables the development of vortices in the accretion matter .
the baroclinicity of the general flow is given by the baroclinic term ( klahr 2004 ; petersen 2007 ) : @xmath30 .
the importance for this instability is the non - axisymetric deviations from the mean state which can lead to the rise of the baroclinic term even in two dimensions : @xmath31 and vorticity can be generated .
this instability , known as baroclinic ( klahr & bodenhiemer 2000 , 2003 ) and arising vortices are discussed in section 4 of the current survey .
the complexity of the studied problem and analysis on the corresponding equations in hydrodynamical matter , require applying of numerical codes .
we use `` pdetools '' and `` pdesolve '' packages , implicated in maple 5 and maple 12 ( maple tutorials of monnagan et al . 1998 ; heal et al .
we chose to insert into calculations the methods , which are employed in these codes , known as : the runge - kutta ( implicit part ) method ( further referred to as rk ) , alternating direction implicit method ( adi ) , centeredtime1space(cts)[forward / backward ] , backwardtime1space ( bts)[forward / backward ] , based on finite difference scheme or godunov s algorithm .
more detail description for rk and adi methods could be found in ( autar & egwu 2008 ; chang et al .
1991 ) and in maple tutorials again for cts and bts .
all they are implicit methods , which are general in their application .
it is suitable to use them in the solutions of partial differential equations , because of their high stability .
we give here a short description of their operation .
the method of runge - kutta is widely used in physical calculations .
the method treats every step in a sequence of steps in identical manner .
that fact makes it easy to add in runge - kutta into relatively simple schemes ( forsythe et al .
this is mathematically proper , since any point along the trajectory of an ordinary differential equation can serve as an initial point .
euler method is a part of rk family methods and it s common expression is : @xmath32 , which is unsymmetrical .
it advances the solution through an interval @xmath33 , from @xmath34 to @xmath35 and uses derivative information only at the beginning of that interval .
this means that the step s error is only one power of @xmath33 smaller than the correction , i.e. the member @xmath36 is added to that expression .
the symmetrization in the error term could cancels out the first - order error term , making the method second order .
( if the error term of the method is @xmath37 , a method is known as @xmath38-th order ) .
adding up the right combination of these , we can eliminate the error terms order by order ( cash & karp 1990 ) . that is the basic idea of the runge - kutta method and we applied its schemes to equations posted in the current survey .
it is common that numerical codes need some control during the processing .
adaptive step - size control is used here ( cash & karp 1990 ; hairer & soderling 2005 ) , with the purpose of achieving some predetermined correctness in the solution with minimum computational effort .
implementation of adaptive step - size control requires that the stepping algorithm returns information about its performance and estimation of its truncation error .
adi method belongs to the group of finite difference methods and follows the idea to split the finite difference equations in two , in relation to the derivatives in coordinates taken implicitly .
the system of equations then becomes symmetric and tridiagonal and is usually solved with tridiagonal matrix solver . the centered time forward / backward space method and backward time forward / backward space method are implicit single stage methods that can be used to find solutions to pdes containing the derivatives v , v[x],v[t],v[xt ] . in the both methods , pdes that describe right - traveling waves should use the backward method ( specified as [ backward ] ) , and specify the boundary condition at the left boundary , while left - traveling waves require the forward method ( specified as [ forward ] ) , and the boundary condition at the right boundary .
numerical boundary conditions are not required .
the points of discretization that are used to compute the values of the grid are diferent : ( t+k , x[i ] ) is in bts ; ( t+k/2,x[i ] ) in cts .
we test the equations with the `` pdetest '' tool , which checks if the solution is correct .
to reduce the pdes to a simpler problem , we applied an `` ansatz '' tool .
the specifying initial and boundary conditions are assigned in sects . 3 and 4 in accordance with the presented model .
box - framed scheme has been suggested and applied .
we could perform the calculations in limited regions of all disc s areas by configuring the scheme for each problem , then , we make the calculations inside of the box , or frame with different measurement .
we have made the code validation , applying it to the problem of shock wave distribution in a tube .
the box - framed model has been applyed again .
we depict most of the results by graphic simulations , which present some full picture of the studied processes .
then , based on the images decipherment , we have to identify the fine forms for better interpretations of the results . for the purpose of the study
the high - pass : gaussian filters , which are adequate to enhance the structure of the graphical results and to obtain spatial filtered image ( gonzalez & woods 1992 ) , are applied . they can be used to enhance edges between different regions .
the gaussian filters are part of convolution filters which remove the low frequency components of an image while retaining the high frequency ( haddad & akansu 1991 , shapiro & stockman 2001 ) local variations .
this is accomplished using a kernel with a high central value , typically surrounded by negative weights .
high pass filter with [ 3 x 3 ] kernel is used with a value of 8 for the central pixel and values of -1 for the exterior pixels ; high pass filters can only have odd kernel dimensions .
we employ this method in section 4 .
the convolution filters produce output images in which the brightness value at a given pixel is a function of some weighted average of the brightness of the surrounding pixels .
convolution with the image array returns a new , spatially filtered image ( nixon & aguado 2008 ) . different types of filters can be generated by selecting the kernel size and values , producing ( vandevenne 2004 - 2007 ) .
we have found the above combination of methods suitable for the exploring here processes : mass transfer , interaction of the flows and concomitant structures formation .
application of these methods and results are presented in the next two sections .
all used equations from sect .
2.1 are transformed further in cylindrical frame of reference ( @xmath39 ) .
this system allows applying 2d and 3d numerical modeling of structure of the flowing matter .
the modeling is performed in non - inertial frame of reference and quadratic 2d - 3d set .
the gas - dynamical analysis of the flow structure in the close binary star system has revealed tidal interaction between the out - flowing matters from donor through the point of libration @xmath40 and the flow around the accretor ( boyarchuk et al .
2002 ; bisikalo et al .
2003 ; frank 2008 ; pringle 1985 , 1992 ) .
bisikalo et al .
( 2001 ) have shown that even a small variation in mass transfer rate of the binary system , could disturb the equilibrium state of the hot accretion disc .
this may cause the appearance of area with increased density , called `` blob . using this result as a point
, we obtained a similar dense formation , called ' ' thickened zone .
we used the above numerical methods in the calculations ( rk & adi ) and reported the solution in paper boneva ( 2010 ) firstly .
the graphical result there was obtained only for a moment of time in one period of rotation to mean an orbital period : @xmath41 ( see fig.1a and fig . 6 in appx . ) .
where @xmath42 denotes the moment of time when the mass transfer rate begins to change and @xmath43 is the orbital period of the binary system . in the current study , the simulations are performed for three additional moments of rotational period , until one orbital period of the system is fulfilled , see fig . 1 ( b ) , ( c ) , ( d ) : @xmath44 .
we repeated numerical runs for @xmath45 .
we apply free boundary conditions at the outer disc edge , where the density is defined to be constant : @xmath46 , where @xmath47 is the density of the inner lagrangian point @xmath48 . in the inner regions , where the mass transfer and the interaction of streams take place
, the values of the density , as was shown , could not remain constant .
a spherical form of the location of interaction was adopted , with a free radius . the limitations of the software imposed some employment restrictions and shortening the necessary time for these calculations , but it does not affect the final results .
the results show that the thickened zone could exists for a long period of time and it does not change its mean characteristics over the time of our calculations .
this is close to the result , reported by bisikalo et al .
( 2001 ) who have showed this for several tens of orbital periods of the binary star system . in the current paper
, we stopped our examinations before the mass transfer rate begins to subside .
[ figure 1 ] .the mesh grid was taken in a ( x , y , z ) calculation frame of coordinates , corresponded to the density distribution in r direction over the time t[days ] .
, title="fig:",width=453,height=377 ] in the next calculations , we can see that the changes in the mass transfer rate are in a close relation both with disturbances in the density and in the velocity .
we put values to the rate of accretion ranging : @xmath49 . to study this , we apply the modified perturbation function , described in appendix and in ( boneva 2009 ) , on the navier - stokes equations ( eq . 2 ) , and obtain the `` term of instability '' : @xmath50 .
this term gives the relation of velocities , including perturbation value and angular velocity . in this expression @xmath51
is the full quantity of velocity , which is a sum of the perturbation value and the time averaged velocity ( see app . , eq .
the behavior of perturbed velocity is calculated for different values of the dimensionless parameter @xmath52 , which is the mode number along @xmath53 direction ( see appendix ) .
after the calculations , we detected sharp decreasing in velocity values and it is showed in fig .
5 ( in the appx . ) and in boneva ( 2009 ) .
let s see , how the appearance of our `` thickened zone '' is provoked , related to the variability in density and velocity . in the places where the velocity values are close to their minimum , the density starts to increase and it pulls a matter there .
thereafter , this means that the matter from a disc could be pumped out or concentrated within given places , causing the density s dilution in close areas .
we express this graphically and the thickening place in the disc can be viewed now in form of the pillar ( see fig.2 ) .
the result points for three periods of rotation . and on variations of the perturbed density value @xmath54.,width=340,height=340 ] the pillars , seen in fig .
2 , indicate matter accumulation in the places with increasing density , defined by the boundary conditions area , for several periods of rotation .
the image in fig.2 corresponds to the appearance of dense zone in fig.1 .
this result is the second stage of the case presented by boneva ( 2010 ) .
it was already mentioned in sect . i and it is confirmed by many studies vortices
play a vital role in accretion disc dynamics .
they are considered to be an effective mechanism of angular momentum transport ( barranco & marcus 2005 ) .
herein , we perform a computational analysis to reveal one possible way of their appearance by visual simulation of their development in the flow . our calculations are based on the vortical transport equation ( eq .
5 ) , because it includes the condition that could provokes baroclinic character of the flow , ( klahr & bodenhiemer 2003 ) .
the box - frame model is used once again .
the introduced boundary conditions are of dirichlet- and cauchy type : @xmath55 ; @xmath56 is the radius of the vortex ; @xmath57 is the boundary area of equations activity .
we place the cylindrical coordinates @xmath58 frame for the equations and quadratic @xmath59 set for the numerical scheme .
we perform a series of runs with zero initial vorticity , but different from zero the initial turbulence values : @xmath60 , @xmath61 , @xmath62 , @xmath63 , and @xmath64 .
results of the simulations show a vortex type growth in @xmath65 plane of the disc zone .
the box - frame values range from about @xmath66 au and from @xmath67 to @xmath68 au , corresponding to the above values of @xmath69 and @xmath70 , referred to as @xmath71 and @xmath72 .
the development of this kind of vortices passes through three stages , resulting from the calculations .
firstly , a distortion of the flow laminarity is observed ( fig .
3a ) . in the next series of calculations , the velocity values step to ( @xmath73 ) and the layers in the examined area undergo a weak undulation ( fig.3b ) .
this means that the variations of velocity and density have significant impact on the flow s behavior . for the final round of calculations
the density and velocity accepted values @xmath74 are used as an input .
then , we consider the stage of vortex evolution in some steady period of their development , when they are " ready for the angular momentum transport ( fig .
in contrast to sect .
3 , where it was shown that the thickening patterns are formed inside the disc area , here in this case the development of vortices is more frequently observed along the outer sides , close to the disc s edges . in conclusion , we suppose that , when this kind of vortical formations leave the disc zone , they could crumble and merge into the matter of the curcumdisc hallo , influenced by the conditions of low density there .
the solution of the last assumption is still under investigation and the results will be presented in another report .
to @xmath75 au and @xmath76 to @xmath77 au , reffered to the boundary frame @xmath78 of the calculation performance .
the light blue and dark blue colours ( light and dark in a grey scale for the printed version ) show the difference in density in the interacting flow layers .
the density values are increasing from dark to light zone.,width=566,height=491 ] we have received the view of the fig.3 by image processing stages shown in fig .
4 . figure 4a shows contour profile of vortex obtained by the result of numerical simulations , which we already described in this section .
we apply on it the convolution method by high pass gaussian filter , discussed in sect . 2.2.2 , and then depict the same vortexes in fig.4b .
the final model is obtained by removing the outliers and eliminating the inessential elements from the image map ( fig.4c ) .
we made this in an attempt to better understand the results of these simulations and distinguish the main structure components .
the image map gives a summary of the data , including the set of projection of all examined parameters ; each pixel of the image consists of some information about the numerical graphical simulations . the baroclinicity conditions of ( klahr & bodenhiemer 2003 ) , misalignment of pressure gradient and density gradient is required wherever there is azimuthal density gradient .
this misalignment acts as a source term for the generation of vorticity in the region of outer edge of the disc .
the latter processes have initially been presented in a similar way by boneva & filipov ( 2009 ) and boneva ( 2010 ) , where the received images are of an isolated vortex , which is a part of the examined region , defined above .
these vortices may propagate throughout the disc , according to the baroclinicity global character , but they are local , temporal formations .
how long these vortices would live and what would destroy them is another unanswered question and unresolved problem .
the point we want to make is that two - dimensional simulations have been used to identify physical processes that would likely play a role in a full three - dimensional simulation .
if vortices primarily form on the upper and lower surfaces of accretion discs , then the radiative cooling rates could be rapid , as was shown in the numerical results by barranco & marcus ( 2005 ) .
therefore , it could be expected that vortices confined to the surface of a disc could have longer lifetimes due to their ability to transport heat radially outwards .
the investigation in this paper concerns the morphology of the flow in the area of accretion disc during tidal interaction in the binary star .
numerical methods were employed in the analysis of the gas - dynamical equations , to study the processes and resulting effects , during mass transfer between the components of the binary stars and corresponding tidal waves .
we present a modeling that reveal of how the accreting flow structure could be organized after interaction processes in the binary .
the change of the rate in mass transfer causes variations in the flow density , which are expressed as accumulation of matter in some places and dilution in others .
this situation leads to the appearance of the thickened zone that remains stable despite of the consequently tidal influences and dissipation processes . besides the description in sect .
2 , the analysis show us that the construction of vortical transport equation ( eq .
5 ) is some kind of the whole disc structure s analog .
it gives the relation between the angular momentum transport part ( the last term in the right hand side ) and describes the sources of the transport mechanism in sense of the vorticity function , and the patterns developed . in the studied interaction parts of the binary star flow , the incoming speed of flow is close to the speed of sound
then , this tide may causes some perturbations in quantities , which changed the dynamics of the gas flow .
after applying the numerical methods via numerical code the result shows the presence of two - dimensional vortical patterns .
vortices are usually local formations , but under the conditions assumed here , they could propagate globally throughout the disc . this type of vortices has often been observed in 2-d simulations of accretion discs .
vortices can be generated by two - dimensional instability as the rossby wave instability or baroclinic instability , which have been investigated in recent years . in this study
we showed once again that under the influence of tidal wave , the accretion flow could not remain stable and the conditions of patterns development are generated .
then , their movement throughout the accretion flow could support the transportation of angular momentum .
the presence of a gravity - connected companion of the star can affect the physical processes in the star and appreciably change its evolution .
studying the accretion disc s structure in binary stars by using numerical methods proves that the gravitational effect of the second component may cause the appearance of spirals shocks .
tidal waves from the donor star , which usually cause a development of spiral density structure , could also be responsible for vorticity in the accretion area
. however the modeling of such effects will be the topic of another report .
we know from the perturbation analysis of papaloizou & pringle ( 1984 ) that in most hydrodynamical flows the flow could change its state , when some of the parameters characterizing it ( e.g. the velocity , density or pressure ) are perturbed
. then , the accretion flow could remain stable , become instable , turbulent , or to form a vorticity .
we describe here how these perturbations are quantified and for this purpose we employ the equations of navier - stockes .
the first step is to introduce the perturbation quantities into the parameters and to obtain an equation for hydrodynamic instability , such as : where @xmath80 are the total quantities of velocity , density and pressure , respectively ; @xmath81 are the time averaged values ; @xmath82 are the perturbations in time .
balbus ( 2003 ) has made a full and detailed analysis of the disc s stability , using similar expression for the perturbation function , but under different conditions of the disc flow compared to those , considered here .
the perturbations are set to be in an exponent form with a power of second order and for the velocity they have the form : @xmath83 .
this form is applied also to the other quantities . here
@xmath84 is the wave number and @xmath52 is the mode number of @xmath85 direction . for the initial values of these quantities , denoted by `` zero '' subscripts
, we may write : @xmath86 ( we apply@xmath87 ) , and @xmath88 . the second step is to express the navier - stokes equation in cylindrical coordinates ( @xmath39 ) .
the newly introduced quantities from eq.(a1 ) are taken into account , which give us the form of equations transformed with perturbation values : because of the time averaging and the presence of perturbations in the quantities , some of the terms in the equations may vanish or reform their expression .
the expression in z - direction brings no new information and after averaging over z , eq .
( a2c ) can be ignored . the second and third term in the right hand site of eqs . a2a and a2b , express turbulence activity .
this term would allow detection of sharp changes in the instability and is called : `` term of instability '' ( further shortly call : `` inst term '' ) .
now , following the perturbed equations ( a2a , a2b ) , and the effect of perturbation function , next form of this `` term '' have been extracted : @xmath92 and @xmath93 .
they contain the quantity of angular velocity @xmath6 that is found to be of importance for studying the differential flow in accretion discs .
it appears that the `` inst term '' in the above equation is significant only in @xmath94 and @xmath53 directions , as should be expected in accordance with the initial conditions for this case .
then , using the mode number @xmath52 from the perturbation expression we examine how the velocity and density variations develop . for the graphical visualization of the disturbances effects
, we use a grid - scale measurement of values [ 12 , 12 , 12 ] , which is more suitable to detect the results .
the next initial values are applyed : @xmath95 , @xmath96 , @xmath63 , @xmath64 .
it results in occurrence of velocity excess during the period of disturbance ( fig .
the velocity state is in a quiet ( fig .
variations start at @xmath97 ( fig .
5b , 5c ) . the critical value in fig .
5(d ) is at @xmath98 , when some fission and sharply decrease of the velocity are observed.,width=377,height=377 ] figures 5(a - d ) show how the process changes when different values of @xmath52 are applied .
it is observed that in a range of @xmath99 there is no velocity exchange . when the values of @xmath52 start to decrease ( @xmath100 ) the process reverses .
this leads to a fission in the solution and a suddenly drop - off in the velocity values is observed .
this could be caused by some local unstable activity in the accretion discs zone ( e.g. the change of mass transfer rate of the inflow matter in close binary , studied in sect.3 ) .
the scales of the axes are constrained by the parametrization of the code . in the next figure
we show the result of applying the perturbation analysis ( explained above ) on the matter of interaction flow after variation in mass transfer rate .
the calculations are made only for one orbital period of the rotating system .
figure 6 shows the appearing of dense pattern , called `` thickened zone '' .
the analysis of the perturbation functions and the expression of navier - stokes equations presented in this appendix help us to understand the behavior of interacting flows in close binary stars , as the processes studied in sect.3 and 4 in the current paper .
lovelace , r.v.e .
, li h. , colgate , s.a . ,
nelson , a.f . 1999 ,
, 513 , 805 - 810 monnagan , m.b . , geddes , k.o .
, heal , k.m . , labhan , g. , vorkoetter , s.m .
1998 , maple v : programming guide , 2nd edn .
( for maple v realease 5 ) , springer , new york nauta , .
d. 2000 , two - dimensional vortices and accretion disks , university utrecht shen , y. , stone , j. m. , gardiner , t. a. 2006 , apj , 653 , 513 steeghs , d. , harlaftis , e.t .
, horne , k. 1997 , mnras , 290 , l28 .
shore , n.s .
2007 , astrophysical hydrodynamics , 2-nd ed .
, wiley - vch verlag gmbh & co. kgaa , isbn : 978 - 3 - 527 - 40669 - 2 | we investigate the problem of structures formation in accretion disc zone , resulting from the tidally interaction in close binary star system .
we aim to examine the area where the incoming flow meets the matter around secondary star and the resulting effects throughout the accretion disc .
the research is based on employment of the fluid dynamics equations in conjunction with numerical simulations leading to the design of graphical models of accretion processes . for the simulation
we propose box - framed sharing schemes .
( i ) the tidal transfer of matter through the inner lagrangian point in close binary stars disturbs the flow in disc s zones and outer disc area .
calculations on the perturbed parameters of density and velocity reveal formation of a thickened zone in the contact area of interacted flows .
it appears to be stable for a number of periods and is unaffected by rotation and dissipation processes .
( ii ) the results also show development of undulations , which grow to vortical patterns in the accretion disc s zone .
it is confirmed that under the influence of tidal wave the conditions of reconfiguring of accretion flow structure are generated .
keywords : accretion ; accretion disks ; hydrodynamics ; waves ; methods : numerical ; ( stars ) : binaries : close ; |
the space of positive definite symmetric matrices has been studied extensively in statistics as a means of understanding dependence in multivariate data along with the accompanying problems in statistical inference .
many books and papers , for example , @xcite and @xcite , have been written on this subject , and there has been considerable interest recently in high - dimensional random matrices with particular emphasis on the distribution of certain eigenvalues @xcite and on graphical models @xcite.=1 in this paper we consider the problem of estimating the mixing density of a continuous mixture of wishart distributions .
we construct a nonparametric estimator of that density and obtain minimax rates of convergence for the estimator . throughout this work ,
we adopt , as a guide , results developed for the classical problem of deconvolution density estimation on euclidean spaces ; see , for example , @xcite and @xcite . much of the difficulty with the space of positive definite symmetric matrices is due to the fact that mathematical analysis on the space is technically demanding .
helgason @xcite and terras @xcite provide much insight and technical innovation , however , and we make extensive use of these methods .
we summarize the paper as follows . in section [ secwishart ]
we discuss and set up the notation for wishart mixtures . in section [ secdecon ]
we begin by reviewing the necessary fourier methods which allow us to construct a nonparametric estimator of the mixing density , and then we provide the estimator . the minimax property of our nonparametric estimator is stated in section [ secmain ] along with supporting results .
section [ secns ] presents simulation results as well as an application to finance examining real financial data .
finally , sections [ secproof - upper ] and [ secproofs - lower ] present the proofs .
throughout the paper , for any square matrix @xmath0 , we denote the trace and determinant of @xmath0 by @xmath1 and @xmath2 , respectively ; further , we denote by @xmath3 the @xmath4 identity matrix .
we will denote by @xmath5 the space of @xmath4 positive definite symmetric matrices . for @xmath6 with @xmath7 , @xmath8 , the multivariate gamma function
is defined as @xmath9 where @xmath10 denotes the classical gamma function .
we denote by @xmath11 the general linear group @xmath12 of all @xmath4 nonsingular real matrices , by @xmath13 the group @xmath14 of @xmath4 orthogonal matrices and by @xmath15 the group of diagonal positive definite matrices .
the group @xmath11 acts transitively on @xmath5 by the action @xmath16 @xmath17 , @xmath18 , where @xmath19 denotes the transpose of @xmath20 . under this group action , the isotropy group of the identity in @xmath11 is @xmath13 ; hence the homogeneous space @xmath21 can be identified with @xmath5 by the natural mapping from @xmath22 that sends @xmath23 . in
distinguishing between left and right cosets , we place the quotient operation on the left and right of the group , respectively . for @xmath24 ,
define the measure @xmath25 it is well known that the measure @xmath26 is invariant under action ( [ gp - action ] ) .
relative to the dominating measure @xmath26 , the probability density function of the standard wishart distribution with @xmath27 degrees of freedom is latexmath:[\[\label{wishart } w(y ) = \frac{1}{2^{nm/2}\gamma_m(0,\ldots,0,n/2 ) } @xmath18 . consequently , for @xmath29
, we note that @xmath30 and @xmath31 .
it then follows that , relative to the dominating measure @xmath26 , the density of the general wishart distribution , with covariance parameter @xmath32 , is @xmath33 , @xmath18 .
suppose next that @xmath32 is a random matrix and , relative to the dominating measure @xmath34 , has a continuous mixing density , @xmath35 , that is invariant under the action ( [ gp - action ] ) . by integration with respect to @xmath32 , the continuous wishart mixture density is given by @xmath36 @xmath18 .
for the case in which @xmath37 , the standard wishart density is essentially a chi - square density , in which case ( [ marginal ] ) is a continuous mixture of chi - square densities . in general , ( [ marginal ] )
is a convolution operation for functions on @xmath5 .
we denote by @xmath38 any matrix with @xmath39 , where @xmath40 and denote @xmath41 . define @xmath42 , the convolution of two random matrices @xmath43 and @xmath44 which are distributed on @xmath5 by @xmath45 and @xmath46 , the convolution of @xmath47 and @xmath48 by @xmath49 where @xmath50 is the space of integrable functions raised to the @xmath51th power on @xmath5 for @xmath52 .
if @xmath43 and @xmath44 with densities @xmath35 and @xmath53 , respectively , are independent , then @xmath54 has the density @xmath55 since @xmath56 .
finally , ( [ marginal ] ) can be transformed into a scalar multiplication ; see section [ ssechfc ] .
in this section we review the fourier methods needed to transform the convolution product ( [ marginal ] ) and to construct a nonparametric estimator of the mixing density @xmath35 . for @xmath18 ,
denote by @xmath57 the principal minor of order @xmath58 , @xmath8 . for @xmath59 , the _ power function _
@xmath60 is @xmath61 @xmath18 .
let @xmath62 denote the haar measure on @xmath13 , normalized to have total volume equal to one ; then @xmath63 @xmath18 , is the _ zonal spherical function _ on @xmath5 .
it is well known that the functions @xmath64 are fundamental to harmonic analysis on symmetric spaces @xcite .
if @xmath65 are nonnegative integers then , up to a constant factor , ( [ zonalsphfunct ] ) is an integral formula for the zonal polynomials which arise in many aspects of multivariate statistical analysis @xcite , pages 231 and 232 .
let @xmath66 denote the space of infinitely differentiable , compactly supported , complex - valued functions @xmath35 on @xmath5 ; also , let @xmath67 for @xmath59 and @xmath68 , the _ helgason fourier transform _
( @xcite , page 87 ) of a function @xmath69 is @xmath70 where @xmath71 denotes complex conjugation . for the case
in which @xmath72 , we make the change of variables @xmath73 in ( [ hf - transform ] ) , @xmath74 , and integrate with respect to the haar measure @xmath75 .
applying the invariance of @xmath35 and formula ( [ zonalsphfunct ] ) , we deduce that @xmath76 does not depend on @xmath77 . specifically , @xmath78 where @xmath79 @xmath59 , is the _ zonal spherical transform _ of @xmath35 . in the case of the standard wishart density ( [ wishart ] ) , which is a @xmath13-invariant function ,
the zonal spherical transform is well known ( muirhead @xcite , page 248 ; terras @xcite , pages 85 and 86 ) : @xmath80 the following result shows that the convolution operation can be transformed into a scalar multiplication .
[ prophfconvc ] suppose @xmath43 and @xmath44 with densities @xmath81 and @xmath82 , respectively , are independent , and @xmath44 is @xmath13-invariant .
let @xmath83 be the density of @xmath84 .
then @xmath85 note @xmath86 using the @xmath87-iwasawa decomposition of @xmath88 ( terras @xcite , page 20 ) , we have @xmath89 for @xmath90 and @xmath91 , an upper triangular matrix .
observe @xmath92 where proposition 1 of terras @xcite , page 39 , is used for the first equality .
for @xmath93 with @xmath94 , let @xmath95 denote the classical beta function .
for @xmath59 such that @xmath96 for all @xmath97 , the _ harish chandra _ @xmath98-_function _ is @xmath99 let @xmath100 , and set @xmath101 and @xmath102 let @xmath103 be the set of @xmath4 diagonal matrices with entries @xmath104 on the diagonal ; then @xmath105 is a subgroup of @xmath13 and is of order @xmath106 . by factorizing the haar measure @xmath62 on @xmath13 ,
it may be shown ( @xcite , page 88 ) that there exists an invariant measure @xmath107 on the coset space @xmath108 such that @xmath109 for @xmath24 , we define the @xmath4 matrix of partial derivatives , @xmath116 where @xmath117 denotes kronecker s delta . the laplacian , @xmath118 , on @xmath5 can be written ( @xcite , page 106 ) in terms of the local coordinates @xmath119 as @xmath120 the power function @xmath121 in ( [ powerfunct ] ) is an eigenfunction of @xmath118 ( see @xcite , page 229 , @xcite , page 283 , @xcite , page 49 ) .
indeed , let @xmath122 , @xmath8 , and define @xmath123 then @xmath124 . since @xmath125 then each @xmath126 , @xmath8 , is purely imaginary ; hence , @xmath127 , @xmath128 . the operator @xmath110 changes the effect of invariant differential operators on functions to pointwise multiplication : if @xmath111 , then @xmath129 @xmath59 , @xmath68 ( @xcite , page 88 ) . for @xmath130 , we therefore define the fractional power , @xmath131 , of @xmath118 , as the operator such that @xmath132 @xmath111 .
having constructed @xmath131 , we define the _ sobolev class _ , @xmath133 where for @xmath134 , @xmath135 denotes the @xmath136-norm with respect to the measure @xmath26 . for @xmath137 , we also define the _ bounded sobolev class _ , @xmath138
in this section we will present the main result .
we do so by applying the helgason
fourier transform to the mixture density ( [ marginal ] ) so that @xmath139 @xmath59 , @xmath68 ; see proposition [ prophfconvc ] . having observed a random sample @xmath140 from the mixture density , @xmath141 , in ( [ marginal ] ) , we estimate @xmath142 by its _ empirical helgason
fourier transform _
, @xmath143 on substituting ( [ emp ] ) in ( [ hc - convoprop - prior ] ) , together with the assumption that @xmath144 , @xmath59 , we obtain @xmath145 @xmath59 , @xmath68 . analogous with classical euclidean deconvolution , we introduce a smoothing parameter @xmath146 where @xmath147 as @xmath148 , and then we apply the inversion formula ( [ hf - inversion ] ) using a spectral cut - off based on the eigenvalues of @xmath118 .
first , we introduce the notation @xmath149 where @xmath150 is defined in ( [ cmrho ] ) .
we now define @xmath151 @xmath18 , and take this as our nonparametric estimator of @xmath35 .
we now state the minimax result for the estimator ( [ emp - hf - inversion ] ) .
let @xmath152 denote a generic positive constant independent of @xmath153
. for two sequences of real numbers @xmath154 and @xmath155 , we use the notations @xmath156 and @xmath157 to mean @xmath158 and @xmath159 , respectively , as @xmath148 .
moreover , @xmath160 means that @xmath156 and @xmath157 .
[ thmwishart1 ] suppose @xmath35 is a density on @xmath5 and @xmath161 . then , for the wishart mixture ( [ marginal ] ) , @xmath162 and for any estimator @xmath163 of @xmath35 , @xmath164 we now provide some comments about this result . in the situation
where ( [ marginal ] ) is a finite sum , so that @xmath165 we have the finite mixture model .
methods for recovering the mixing coefficients can be covered by the techniques employed in @xcite .
we note that the continuous mixture model is a generalization of this approach .
it is noted that the condition @xmath35 is a density and seems to be mild . the upper bound property of ( [ mainresult ] ) is established in @xcite , theorem 3.3 , with @xmath166 . in the latter , the moment condition @xmath167 on the principal minors @xmath168 of @xmath18 is assumed . in our theorem
, we did not impose this moment condition as condition ( [ moment ] ) is automatically satisfied .
this is pointed out and commented upon below in the proof . to derive the lower bound for estimating @xmath35 in the @xmath136-norm
, we shall follow the standard euclidean approach .
thus we choose a pair of functions @xmath169 , @xmath170 , and , with @xmath53 denoting the wishart density ( [ wishart ] ) , we shall show that , for some constants @xmath171 , @xmath172 and @xmath173 where @xmath174 precisely , let us suppose we can choose @xmath175 and a perturbation @xmath176 , and , for @xmath177 , let @xmath178 be a scaling of @xmath179 such that @xmath180 . define @xmath181 if @xmath182 can be chosen so that @xmath183 then the lower bound rate of convergence is determined by @xmath184 .
we shall develop such a construction and , moreover , do so in a way such that @xmath185 as @xmath148 .
the profound influence of charles stein on covariance estimation originates largely from his rietz lecture ; see @xcite .
the idea is that for certain loss functions over @xmath5 , the usual estimator of the covariance matrix parameter is inadmissible . through an unbiased estimation of the risk function over covariance matrices ,
stein was able to improve upon the usual estimator by pooling the observed eigenvalues of the sample covariance matrix .
subsequent to this , through a series of papers , improvements were obtained in haff @xcite .
other related works include takemura @xcite , lin and perlman @xcite and loh @xcite , to name a few . in this paper , we contribute to the case in which one observes data from a continuous mixture of wishart distributions , not merely a sample from a single distribution .
therefore , the parameter of interest would be the mixing density of the covariance parameters . and
the nonparametric estimator of the mixing density ( [ emp - hf - inversion ] ) is an attractive candidate because of its minimax property .
based on this procedure , one could consider the moment , or mode , of @xmath186 , as a possible estimator of the corresponding population parameters .
alternatively , one could take a nonparametric empirical bayes approach as in pensky @xcite .
this section presents numerical aspects for the @xmath187 case with an application to finance .
suppose @xmath43 and @xmath44 are independentwith @xmath44 having a wishart distribution .
let @xmath188 where @xmath189 is upper triangular . for visualization , we display estimators of the marginal density for @xmath190 where @xmath191 with @xmath90 and @xmath192 .
let @xmath193 where @xmath194 denotes the diagonal matrix of eigenvalues of @xmath195 , @xmath196 .
denote by @xmath197 the density of eigenvalues of @xmath43 .
then , the estimator for @xmath197 is given by @xmath198 consider the computation of @xmath199 when @xmath200 so that @xmath125 . from pages 90 and 91 of @xcite , the spherical function is given by @xmath201 where legendre function @xmath202 can be computed using ` conicalp_0(t , x ) ` in the ` gsl ` library in ` r ` .
the mean integrated squared error ( mise ) of @xmath203 is defined by @xmath204 it is reasonable to choose @xmath205 which minimizes @xmath206 or equivalently @xmath207 one can find an unbiased estimator @xmath208 of @xmath209 ; see @xcite .
we choose @xmath210 by @xmath211 monte carlo approximation is used for integration of ( [ eqfxhat2 ] ) and @xmath208 .
denote by @xmath212 the wishart density with degrees of freedom @xmath27 and covariance matrix @xmath32 .
we generate data as follows . for @xmath196 :
* generate @xmath213 ; * generate @xmath214 ; * do a cholesky decomposition of @xmath215 , and calculate @xmath216 .
, upper right shows an estimate with @xmath217 , lower left with @xmath218 and lower right with @xmath219 . ] as examples , we consider a unimodal mixing density @xmath220 and a bimodal density @xmath221 .
figure [ fig1 ] show the results for the unimodal case whereas figure [ fig2 ] show the results for the bimodal case . in each of these plots
the , upper right shows an estimate with @xmath217 , lower left with @xmath218 and lower right with @xmath222 . ]
domain consists of the two eigenvalues starting with the largest .
one can see that the general shapes of the estimators become closer to that of the true density as @xmath153 increases .
stochastic volatility using the wishart distribution is of much interest in finance ; see , for example , @xcite and @xcite .
in particular , this entails a situation precisely of the form ( [ marginal ] ) .
let us apply this to the situation where we are interested in estimating the mixing density .
although our methods can be applied to a portfolio of many assets , let us restrict ourselves to two assets since this would be the smallest multivariate example .
indeed , let @xmath223 and @xmath224 denote the daily closing stock prices of samsung electronics ( 005930.ks ) and lg - display ( 034220.ks ) , respectively , traded on the korea stock exchange ( ksc ) for 2010 , where the data can be easily accessed on public financial websites .
we will assume as usual that @xmath225 follows a bi - variate normal distribution for @xmath226 .
we transform the daily data to weekly data and compute the weekly @xmath227 covariance matrix @xmath228 for @xmath229 . in case a week
has a holiday , we repeat the last previous observation . under the usual assumptions this would constitute observations from a mixture model ( [ marginal ] ) with a standard wishart distribution with four degrees of freedom .
figure [ fig3 ] plots the mixing density estimator corresponding to the two eigenvalues .
one can see that there are two peaks , suggesting a possible bimodal stochastic volatility mixing density in the eigenvalues .
the strategy here is , first , to decompose the integrated mean - squared error into its variance and bias components , @xmath230\\[-8pt ] & = & { \mathbb{e}}\|f_n - { \mathbb{e}}f_n\|^2 + \|{\mathbb{e}}f_n - f\|^2 , \nonumber \ ] ] and , last , to estimate each component separately using estimates based on the plancherel formula and the inversion formula for the helgason fourier transform .
[ lemma - bias ] suppose that @xmath231 and @xmath232 .
then @xmath233 we have for @xmath234 @xmath235 applying the plancherel formula , we obtain @xmath236 where we use the fact that @xmath237 where the equality follows from the plancherel formula . by assumption , @xmath231 , the latter integral
is bounded above by @xmath238 , so we obtain @xmath239 and the proof is complete . to obtain bounds for the integrated variance , several preliminary calculations are needed . in particular , we begin with the variance calculation of the empirical helgason fourier transform , which has similarities to the usual empirical characteristic function . [ lememp - helgason ] for @xmath240 and @xmath241 , @xmath242 by ( [ emp ] ) ,
@xmath243 observe also that @xmath244 since @xmath245 .
applying this result to ( [ var-1 ] ) and taking expectations , we obtain @xmath246 where the last equality follows from the fact that @xmath247 are independent and identically distributed as @xmath248 , and because @xmath249 .
following terras @xcite , pages 34 and 35 , let @xmath250 denote the positive weyl chamber in @xmath15 . for @xmath251 ,
let @xmath252 and set @xmath253 and define the normalizing constant @xmath254 by @xmath255 .
denote @xmath256 .
[ lemma - var ] as @xmath257 , @xmath258 by the plancherel formula , latexmath:[\[\begin{aligned } & & { \mathbb{e}}\|f_n - { \mathbb{e}}f_n\|^2 \\ & & \qquad= \int_{{\mathbb{c}}^m(\rho , t ) } \int_{k / m } { \mathbb{e}}|{{\mathcal h}}_n f(s , k ) - { \mathbb{e}}{{\mathcal h}}_n f(s , k)|^2 \,\mathrm{d}\bar{k } \,{\mathrm{d}_*}s \\ & & \qquad= \int_{{\mathbb{c}}^m(\rho , t ) } \int_{k / m } { \mathbb{e}}|{{\mathcal h}}_n r(s , k ) - { \mathbb{e}}{{\mathcal h}}_n r(s , k)|^2 \,\mathrm{d}\bar{k } |{\hat{w}}_{z}(s)|^{-2 } \,{\mathrm{d}_*}s \\ & & \qquad\leq \frac1n \sup_{\lambda_s < t , { \operatorname{re}}(s)=-\rho}|{\hat{w}}_{z}(s)|^{-2 } \int_{k / m } { \mathrm{d}_*}s \\ & & \qquad\ll{\sup_{\lambda_s < t,{\operatorname{re}}(s)=-\rho } } as @xmath257 .
choose @xmath260 .
observe that @xmath261 since @xmath262 is a continuous function of @xmath77 on a compact set @xmath13 , @xmath262 is uniformly bounded on @xmath13 such @xmath263 on @xmath13 .
since @xmath35 is a density so that @xmath141 is also a density , we have @xmath264 hence , it follows from continuity and the compactness of @xmath108 @xmath265 which has been used in the above calculation .
in addition , we use the fact that , as @xmath257 , @xmath266 a result which follows from proposition 7.2 of helgason @xcite , page 450 . the proof of the upper bound can now be obtained by applying lemmas [ lemma - bias ] and [ lemma - var ] to ( [ ims ] ) and setting @xmath267 .
we need to provide some detailed calculations , and the essence of the proof is contained for the case @xmath268 ; hence we will keep this assumption for the remainder of this paper .
the generalization to @xmath269 may be obtained by using higher order hyperbolic spherical coordinates . in this section
, we assume that @xmath179 is a @xmath13-invariant function defined on @xmath270 . for @xmath271 ,
let @xmath272 with @xmath273 , @xmath68 so that @xmath274 .
let @xmath275 with @xmath276 for @xmath277 , and write @xmath278 by a change of variables , @xmath279 where @xmath280 and @xmath281 [ lemmatrixequation ] for @xmath283 , @xmath284 and @xmath285 $ ] , the matrix equation @xmath286 has a solution @xmath287 and @xmath288 .
further , @xmath289 has minimum and maximum values @xmath290 and @xmath291 , respectively , and @xmath292 and @xmath293 can be defined uniquely . in general , if both @xmath35 and @xmath20 are @xmath13-invariant functions on @xmath5 , then @xmath294 is also @xmath13-invariant and @xmath295 .
hence , @xmath296 is @xmath13-invariant and @xmath297 due to @xmath13-invariance of @xmath179 and @xmath53 . from this and lemma [ lemmatrixequation ]
, we can define @xmath298 for @xmath277 , @xmath283 , @xmath284 . denoting by @xmath299 the distribution function corresponding to the standard wishart density @xmath300 with respect to the measure @xmath301
, we have that for @xmath302 , @xmath303 the laplacian for @xmath13-invariant functions in polar coordinate is given by @xmath304 , where @xmath305 and the spherical function is given by @xmath306 with @xmath307 , the legendre function ; see terras @xcite .
it can be seen that @xmath308 so that @xmath309 with @xmath310 .
the helgason fourier transform of @xmath179 is given by @xmath311 from ( [ eqy - to - polar ] ) and ( [ eqpolar - hs ] ) .
suppose @xmath179 is separable so that @xmath312
. then , ( [ eqpolar - hf ] ) implies @xmath313 where @xmath314 and @xmath315 denote , respectively , the laplace transform and the mehler
fock transform ; see terras @xcite .
let @xmath327 be a density on @xmath328 such that @xmath327 is sufficiently smooth and satisfies @xmath329 , @xmath330 , where @xmath331 is a normalizing constant .
define the function @xmath332 where @xmath333 for @xmath334 .
suppose @xmath357 .
observe that @xmath358 and that for @xmath359 , @xmath360 now , the plancherel formula ( [ hf - plancherel ] ) and change of variable @xmath361 gives @xmath362 a suitable choice of @xmath355 gives the desired result .
consider @xmath387 .
suppose @xmath388 .
if @xmath389 and @xmath390 , then @xmath391 .
note that @xmath392 since @xmath393 then we have @xmath394 and by lemma [ lemw - tail ] with @xmath395 , @xmath396 and @xmath397 @xmath398 hence , if @xmath399 , then @xmath400 as @xmath401 .
if @xmath390 and @xmath389 , then @xmath402 since @xmath403 , then we have @xmath404 and by lemma [ lemw - tail ] , with @xmath405 , @xmath406 and @xmath407 , @xmath408 hence , for @xmath409 , @xmath410 as @xmath411 .
consider @xmath412 .
let @xmath413 and @xmath414 .
suppose @xmath388 .
on @xmath415 , @xmath416 on @xmath415 . note that @xmath417 if follows from these and lemma [ lemw - tail ] that @xmath418 and if @xmath419 , @xmath420 from ( p5 ) , one can choose @xmath355 sufficiently small such that @xmath423 is a density . by ( [ eqy - to - polar ] ) and ( [ eqconv - polar - delta ] ) , @xmath424 let @xmath425 and for @xmath385 , @xmath426 by ( [ eqconv - polar - delta ] ) , lemmas [ lempsic ] , [ lemg0 ] and the plancherel formula ( [ hf - plancherel ] )
, we obtain @xmath427 lemma [ lemmatrixequation ] and ( p4 ) imply @xmath428 let @xmath429 .
it follows from ( [ eqpsieta ] ) , lemmas [ lemg0 ] and [ lembound - munu ] that @xmath430 | the space of positive definite symmetric matrices has been studied extensively as a means of understanding dependence in multivariate data along with the accompanying problems in statistical inference .
many books and papers have been written on this subject , and more recently there has been considerable interest in high - dimensional random matrices with particular emphasis on the distribution of certain eigenvalues . with the availability of modern data acquisition capabilities , smoothing or nonparametric techniques
are required that go beyond those applicable only to data arising in euclidean spaces .
accordingly , we present a fourier method of minimax wishart mixture density estimation on the space of positive definite symmetric matrices . , , . |
quantum statistics is an important aspect of quantum mechanics and it lays down the rules for identifying different classes of particles . in three dimensions , the exchange of two particles in a quantum system can result in a phase change of either 0 or @xmath6 for the wave function , leading to bosons and fermions .
the scenario is very different in two - dimensions ( 2d ) and it is well known that such quantum systems give rise to anyonic quasiparticle excitations.@xcite these anyons are observed as excitations , localized disturbances of the ground state in systems with topological order .
rich behavior emerges especially for non - abelian anyons , for which the exchange of two anyons is described by a unitary matrix.@xcite models of interacting non - abelian anyons draw motivation from quantum spin models , and in particular the heisenberg model , which has been studied for a wide range of lattices including one - dimension ( 1d ) chains and ladders of @xmath2 spins .
non - abelian anyons can be described as so - called quantum deformations of the usual @xmath2 spins .
formally , they are described by @xmath7 chern - simons theories .
these @xmath7 theories are obtained by ` deforming ' the @xmath2 algebra so as to retain only the first @xmath8 representations .
the most widely studied models are the @xmath9 and @xmath10 theories that describe ` ising ' and ` fibonacci ' anyons respectively .
non - abelian anyons are expected to be present in topologically ordered systems including certain fractional quantum hall states,@xcite @xmath11-wave superconductors,@xcite spin models,@xcite solid state heterostructures@xcite and rotating bose - einstein condensates.@xcite in particular , fibonacci anyons occur as quasi holes in the @xmath12 read - rezayi state which might be able to describe the @xmath13 fractional quantum hall state.@xcite while several experiments have shown evidence for emergent majorana modes,@xcite experimental evidence of fibonacci anyons is yet to come . in the absence of interactions ,
non - abelian anyons have an exponentially degenerate ground state manifold . on this ground state manifold exchange ( braiding ) of non - abelian anyons
acts in a non - commutative way , by bringing about a non - trivial transformation in the degenerate manifold of the many - quasiparticle hilbert space .
this has generated interest since it can be used for topological fault - tolerant quantum computation.@xcite interactions between non - abelian anyons can be modeled by generalizations of the spin-1/2 heisenberg model .
these models have been studied for chains of fibonacci anyons@xcite including nearest neighbour couplings@xcite and also longer range couplings@xcite have been studied .
chains of higher spin quasiparticles have also been explored,@xcite critical phases have been identified and the corresponding conformal field theory ( cft ) has been obtained from numerical simulations.@xcite the effect of disorder for chains of fibonacci anyons has been investigated.@xcite as a step towards understanding the collective behavior of itinerant non - abelian anyons , these chains can be doped with mobile holes , inspired by the electronic @xmath0@xmath1 model.@xcite electrons confined in 1d can exhibit the phenomenon of `` spin - charge separation '' , where the spectrum can be interpreted in terms of two independent pieces , one arising from electric charge without spin , and the other from a spinon without any charge.@xcite analogously low - energy effective @xmath0@xmath1 models have been analyzed for the case of doped chains of fibonacci ( and ising ) anyons .
these models exhibit a fractionalization of the spectrum into charge and anyonic degrees of freedom,@xcite an extension of a phenomenon that exists in luttinger liquids.@xcite as a step towards two dimensions , anyon models have been investigated on chains coupled to form so - called quantum _
ladders _ of non - abelian anyons , which provide anyonic generalizations of the 2d quantum magnets.@xcite in this paper , we combine the models mentioned above , by studying _ doped _ quantum ladders of fibonacci anyons consisting of two or three chains .
the structure of the rest of the paper is as follows : section [ model ] begins with a brief introduction of non - abelian anyons and introduces our model . in section [ isolated_rungs ] , we discuss the strong rung coupling limit drawing analogy with the spin ladder models , listing out all the possible quantum states and the phase diagram for the model under study . in section [ coupled_rungs ] , we list all the phases that arise for weakly coupled rungs .
the subsequent sections are dedicated to the discussion of all these phases .
section [ sec : effpaired ] discusses the hard - core boson models and section [ sec : effgolden ] describes the golden chain phases . in section [ sec : efftj ] , we analyse the effective @xmath14 models for two and three - leg ladders , presenting the numerical results for the same .
we extend the phenomenon of spin - charge separation to ( doped ) ladders of fibonacci anyons . in section [ fm_rungs ] , we introduce a new model of heavy and light fibonacci anyons that is obtained for certain density regimes when the rung couplings are fm .
we conclude by summarising our results in section [ conclude ] .
in this paper we focus on fibonacci anyons that are described by the @xmath3 theories . @xmath7 chern - simons theories@xcite are so - called quantum deformations of the @xmath2 algebra .
their degrees of freedom are encoded by @xmath15topological charges @xmath16 , which are generalized angular momenta .
in contrast to @xmath2 , in @xmath7 theories the total ` spin ' @xmath16 is limited to be @xmath17 .
akin to the tensor product of spins , non - abelian anyons can be @xmath15fused according to fusion rules given by @xmath18 for example , for the fusion of two anyons with @xmath19 , these rules would mean @xmath20 ( for @xmath21 ) .
similarly , for the case of fibonacci anyons @xmath22 , when @xmath23 the fusion rule reads as @xmath24 . note that this is different from what one would obtain under a tensor product of @xmath2 spin-1 particles . in the limit @xmath25 , however , we recover the @xmath2 algebra and the rule simply describes the tensor product of two ordinary @xmath2 spins .
( that can be either or @xmath26 ) and bond labels @xmath27 .
( b ) the flat version of the fusion tree .
( c ) a basis change to a different fusion tree using an @xmath28-move .
, title="fig : " ] 0.4truecm ( that can be either or @xmath26 ) and bond labels @xmath27 . ( b ) the flat version of the fusion tree .
( c ) a basis change to a different fusion tree using an @xmath28-move .
, title="fig : " ] in the rest of this manuscript , we focus only to the fibonacci theory with @xmath10 , unless otherwise stated . there
the allowed values for the topological charges are @xmath29 .
but if we look closely at the fusion rules , we can make the identification @xmath30 and @xmath31 .
thus , the fibonacci theory has two distinct types of _ particles _ which we denote as @xmath26 for the trivial particle with @xmath32 and @xmath5 for the fibonacci anyon with @xmath33 respectively . using these
, the fusion rules read @xmath34 we represent a system of @xmath35 anyons by means of a _ fusion tree _ as shown in fig .
[ fig:3figs](a ) , where the anyonic charges of the individual anyons are labelled by @xmath36 .
the fusion outcome of successive fusion of the anyons are encoded by the ` bond ' labels ( links ) in the fusion tree , labelled by @xmath27 in fig .
[ fig:3figs](a ) .
the constraints on the bond labels due to fusion rules which must be satisfied at each vertex significantly reduce the size of the internal hilbert space . for @xmath35 fibonacci anyons ( @xmath5 )
the hilbert space grows asymptotically as @xmath37 where @xmath38 is the golden ratio . from now on we draw a flat version of the fusion tree , as shown in fig .
[ fig:3figs](b ) . to perform an operation on nearest neighbor anyons ,
it is advantageous to change to a different basis in which the two - particle fusion outcome is explicit .
this is done via the @xmath28-move shown schematically in fig .
[ fig:3figs](c ) .
the @xmath28-move depends on the two site labels @xmath36 and @xmath39 and the bond labels @xmath40 and @xmath41 .
if at least one of these four labels is @xmath26 , then there is only one choice of bond labels that satisfies the fusion algebra and the @xmath28-move is trivial .
a non - trivial matrix is obtained only when all the four labels are anyons .
specialising to the case where @xmath42 , the labels for the three bonds @xmath43 allowed by the fusion rules are @xmath44 which transforms to a new basis @xmath45 after the @xmath28-move : @xmath46 using these bases the @xmath28-matrix is represented as , @xmath47 where as mentioned above we have a non - trivial @xmath48 submatrix only when also @xmath49 .
another operation that we need to perform on nearest neighbor anyons is that of exchanging ( or braiding ) them . in fig .
[ fig : r ] , we show our convention for a right - handed braid .
the left - hand braiding is the inverse of the process shown here . under a right - handed braid anyons @xmath50 and @xmath51 pick up a phase @xmath52 depending on the anyon types , @xmath50 and @xmath51 , that are undergoing an exchange and their fusion outcome @xmath53 .
note that whenever @xmath50 or @xmath51 are @xmath54 , the phase is trivial .
non - trivial phases are only obtained are for @xmath55 : @xmath56 in order to implement a braid on the standard fusion tree , we first have to change basis using an @xmath28-move to make the fusion outcome of the two anyons explicit , and then braid .
this process is shown schematically in fig .
[ fig : frf ] .
this is represented by a braid matrix @xmath57 acting on the bond labels @xmath43 .
the only non - trivial braid matrix is obtained when both the sites are occupied by anyons . in the basis of eq .
( [ eq : basis ] ) we obtain : @xmath58 note that when the two site labels are a @xmath59 and a @xmath5 , the @xmath28-moves and the exchange phases are all trivial .
the braid matrix is effectively the hopping of the anyon to the adjacent site .
when both the site labels are @xmath59 , the braid matrix is simply given by the identity matrix . in this section
we review the so - called ` golden chains ' , consisting of 1d arrays of localized fibonacci anyons with pairwise interactions between nearest neighbors.@xcite in this model the hamiltonian for the magnetic interactions between anyons is defined in analogy to the heisenberg exchange interaction .
we assign an energy @xmath60 if the fusion outcome of two interacting anyons is trivial . for afm couplings
@xmath61 , this favours the fusion outcome of two neighbouring anyons to be trivial , while for fm couplings @xmath62 , the fusion of two anyons is preferred to be @xmath5 .
this interaction between nearest neighbor anyons is depicted schematically in fig .
[ fig : tj](a ) and is implemented by projecting on the identity fusion channel @xmath63 where @xmath28 is the operator corresponding to the @xmath28-move ( see eq .
( [ eq : f ] ) ) and @xmath64 is an operator that projects onto the @xmath26 state . in the basis of eq .
( [ eq : basis ] ) , the matrix representation for the ( dimensionless ) magnetic interaction can be written explicitly as : @xmath65 .
( b ) the kinetic hopping ( of amplitude @xmath0 ) of an anyon to its nearest neighbor vacant site .
the blue circles represent anyons while the white circles denote vacant sites or holes . ] golden chains with afm couplings are described by the @xmath10 restricted solid on solid ( rsos ) model which is a cft with central charge @xmath66 . for ferromagnetic couplings the corresponding cft is that of the critical 3-state potts model with @xmath67.@xcite to model itinerant anyons we introduce holes , _
i.e. _ sites with a trivial anyon @xmath68 on some of the sites . the holes and @xmath5 anyons
are labelled by different @xmath69 ( electric ) charges and anyonic ( non - abelian ) charges .
the @xmath5 anyons ( referred to simply as ` anyons ' or ` particles ' hereafter ) can move on the chain which results in an additional kinetic energy contribution .
this kinetic process is schematically shown in fig .
[ fig : tj](b ) .
it involves the hopping of a particle , along with its electric and anyonic charge to a neighbouring site .
our specific model of itinerant anyons is a generalization of the electronic @xmath0@xmath1 model.@xcite assuming a large on - site charging energy , we eliminate the possibility of doubly occupied sites and allow anyons to only hop to empty sites . as in the case of electrons , the low - energy effective @xmath0@xmath1 model allows hopping of the anyons to nearest neighbor vacant sites and an exchange interaction between nearest neighbor anyons analogous to the heisenberg interactions explained above in sec . [ goldenchain ] . the kinetic term
can then be written as @xmath70 , where @xmath71 is the ( dimensionless ) operator corresponding to the nearest neighbor hopping process shown in fig .
[ fig : tj](b ) .
a @xmath0@xmath1 chain of itinerant anyons was studied for ising and fibonacci anyons,@xcite revealing a separation of excitations into charge and anyonic excitations , similar to spin - charge separation in its electronic counterpart .
ladders of fibonacci anyons are formed by coupling chains of localized anyons.@xcite anyons interact with their nearest neighbors along the leg and rung directions via and @xmath72 .
these interactions are shown schematically in fig .
[ fig : ladder](a ) . as fusion path
we choose the _ zig - zag path _ shown in fig .
[ fig : ladder](b ) , since it minimizes the effective range of interactions on the fusion path .
we choose periodic boundary conditions along the leg direction and open boundary conditions along the rungs . with this choice of fusion path ,
nearest neighbor interactions on the rungs are also nearest neighbor along the fusion path , while those between anyons on the same leg are longer range along the fusion path .
nearest neighbor rung interactions can be implemented in exactly the same way as for nearest neighbor on a chain ( see sec .
[ goldenchain ] ) . in order to evaluate the interactions between particles on the same leg
, we have to implement a change of basis , this time by braiding them in a clockwise manner until they are neighbours along the fusion path .
this braiding is performed by the unitary braid matrix @xmath57 ( see eq . ( [ eq : braid ] ) ) .
once the particles are nearest neighbours along the path they can interact with the same term as discussed above in sec .
[ goldenchain ] . after carrying out the interaction ,
the anyons have to be braided back to their original positions . more specifically , for a two - leg ladder , adjacent anyons along the leg direction are next nearest neighbours along the fusion path ( see fig . [
fig : ladder](b ) ) .
thus , one needs to implement one braid operation .
the hamiltonian for the magnetic interactions between nearest neighbor rungs @xmath73 and @xmath74 on the upper leg is given by @xmath75 and on the lower leg as @xmath76 where @xmath77 has been defined in eq .
( [ eq : hmag ] ) and @xmath73 labels the rungs ( so that @xmath78 labels the diagonal bonds along the path ) . in fig .
[ fig : braid](a ) , we summarize the magnetic interactions between nearest neighbor along the leg direction for a two - leg ladder . for a three - leg ladder , the leg interactions are longer ranged interactions , since adjacent anyons on a leg are third neighbours along the fusion path .
thus , nearest neighbor leg interactions on a three - leg ladder ladder require two braids before particles are nearest neighbors on the fusion path.@xcite one gets for interaction between rungs @xmath73 and @xmath74 on the upper leg , @xmath79 on the middle leg , @xmath80 and on the lower leg , @xmath81 where @xmath82 labels the diagonal bonds along the path .
the full magnetic hamiltonian on the legs is obtained by adding all contributions , @xmath83 , where @xmath84 is the number of legs .
as an implementation detail we want to mention that the action of the operator @xmath85 can generate up to @xmath86 states for each bond interaction .
the operator mixes spin labels , thus generates multiple images for each two - body interaction .
this exponential increase in the number of resulting states leads to denser hamiltonian matrices as one increases the width @xmath84 of the ladder and restricts us from exploring larger system sizes . for ladders of fibonacci anyons
it was shown that similar odd - even effects as seen for @xmath2 spins continue to exist in the limit of strong afm rungs.@xcite afm coupled ladders of fibonacci anyons with even number of legs are gapped while those with odd number of legs are critical and are described by the same cft as the golden chain . on the other hand , fibonacci ladders with fm rung couplings are quite different from their @xmath2 counterparts.@xcite fibonacci ladders with a width that is a multiple of three are gapped since the rungs form singlets @xmath87 .
other widths are gapless since the isolated rungs effectively form @xmath5 s , thereby yielding a gapless chain as the effective low - energy model .
this is different from @xmath2 spins , for which an even number form a singlet ground state , and where thus all even width ladders are gapped . in this paper
we focus on itinerant ladders of fibonacci anyons . as schematically represented in fig .
[ fig : ladder](a ) we denote the interaction strengths along the leg direction by and for the magnetic and kinetic terms respectively . along the perpendicular direction ,
the couplings and denote respectively the magnetic and kinetic terms .
the magnetic interactions for the ladder have already been described in the section [ undoped ] and we thus only need to discuss the kinetic terms . once again , the two sites on a rung are adjacent on the fusion path and the hopping is thus implemented as for the 1d @xmath0@xmath1 chain ( see section [ tj ] ) . for a particle and hole lying on adjacent sites on the same leg of the ladder ,
we need to , once again , braid the sites to bring them to adjacent positions on the fusion tree .
for a two - leg ladder , the kinetic term between rungs @xmath73 and @xmath74 on the upper leg is given by @xmath88 and on the lower leg as @xmath89 where @xmath71 has been defined above and @xmath73 labels the rungs . in fig .
[ fig : braid](b ) , we summarize the kinetic process between nearest neighbor sites along the leg direction for a two - leg ladder . for a three - leg ladder ,
a particle and hole lying on adjacent sites on a leg are third neighbours along the fusion path .
thus , nearest neighbor leg interaction on a three - leg ladder requires two braids before the particle and the hole are nearest neighbors on the fusion path .
one gets for kinetic terms between rungs @xmath73 and @xmath74 on the upper leg , @xmath90 on the middle leg , @xmath91 and on the lower leg , @xmath92 where @xmath82 labels the diagonal bonds along the path .
note that , in the above equations , some of the braids may be trivial , in contrast to the case of the magnetic interactions .
the full kinetic hamiltonian on the legs is obtained by adding all contributions , @xmath93 . here
, the action of the operator @xmath94 can generate up to @xmath95 states for each bond interaction .
we also consider models with an additional rung charging term @xmath96 where @xmath97 is the number of anyons on a rung and @xmath98 is determined by the implicit chemical potential .
[ fig : parabola ] shows the energy profile for a given rung composition on the ladder .
for the three - leg ladder under consideration , this term acts pairwise between all the three possible pairs of particles that can exist on the three - site rung . assuming a charging energy which is much larger compared to the exchange energy
, we can consider the limit @xmath99 . in this limit
we can restrict our calculations to just two values of the occupation on the rung , @xmath100 and @xmath101 , with @xmath100 ranging from @xmath102 to @xmath103 .
this reduces the hilbert space and thus allows us to perform simulations of larger ladders .
our numerical simulations have been performed by exact diagonalization using the lanczos algorithm .
we exploit periodic boundary conditions along the leg direction to implement translation symmetry .
this allows us to block - diagonalize the hamiltonian into @xmath104 blocks labeled by the total momentum @xmath105 ( @xmath106 being an integer ) , which reduces the size of the hilbert space , which is the major limiting factor in the simulations , by a factor @xmath104 .
s while white circles represent vacant sites .
the @xmath69 and topological charges for the different configurations are denoted in the parenthesis .
the superscript @xmath107 refers to the lying on the upper ( lower ) leg . ]
analogous to standard electronic @xmath0@xmath1 ladders the physics of anyonic @xmath0@xmath1 ladders can be understood starting from the strong rung coupling limit.@xcite we thus begin by identifying the low - lying states of isolated rungs . in fig .
[ fig : eff_particles_2 ] , we show the five rung configurations on isolated rungs for a two - leg ladder , and the total @xmath69 and anyonic charges that are possible for these rung configurations . when there are two holes on a rung , both of these charges are trivially zero .
when the rung is occupied by two particles , the net @xmath69 charge is 2 , however the topological charge may be either @xmath26 or @xmath108 , giving rise to two different quantum states @xmath109 ( named heavy @xmath5 " ) and @xmath110 ( named heavy hole " an empty rung being a light hole " ) . in the case
when there is a single on the rung , it can be either on the upper or on the lower leg with charges denoted as @xmath111 and @xmath112 respectively .
the corresponding quantum states are respectively @xmath113 and @xmath114 .
the bonding and anti - bonding states @xmath115 ( named light @xmath5 " ) are formed by linear superpositions of the configurations with charges @xmath116 and @xmath117 given by @xmath118 s while white circles represent vacant sites .
the first labels in the parenthesis signify the @xmath69 charge , corresponding to the number of @xmath5 s present on each rung and the second refers to their fusion outcome .
the superscripts @xmath119 refer to the positions of the on the different legs ( upper , middle , lower ) of the ladder . ] for a three - leg ladder , many more states are possible , as shown in fig .
[ fig : eff_particles_3 ] . when there is a single on the rung @xmath120 , the
light @xmath5 " quantum states formed by linear superpositions of the three different positions of the particle are given by latexmath:[\ ] ] in addition to the above cases , we can also have a kinetic and potential terms when the difference in the @xmath69 charges on neighboring rungs is larger than 1 .
such process occurs e.g. ( i ) between a charge-0 ( light ) hole and a charge-2 ( heavy ) hole ( marked by blue arrows in table . [ table ] ) in the @xmath342 paired phase or ( ii ) between a charge-0 ( light ) hole and a charge-3 ( super heavy ) @xmath5 ( marked by orange arrows in table .
[ table ] ) in the @xmath148 phase . in case ( i )
, one needs to hop twice to be able to come back to a configuration @xmath343 that has the same @xmath69 charges as the initial configuration @xmath344 , so the effective hamiltonian ( leaving in the relevant subspace ) is obtained in second - order perturbation , in @xmath275@xcite @xmath345 where the sum is on the intermediate states corresponding to two ( light ) @xmath5 s on neighboring sites .
the energy denominator is given by @xmath346 and matrix elements for all the intermediate states are ( @xmath347 ) so that one gets for @xmath348 a potential energy @xmath349 and for @xmath350 a hopping term @xmath351 in case ( ii ) , one needs to hop three times to be able to come back to a configuration that has the same @xmath69 charges as the initial configuration , so the effective hamiltonian is now obtained in third - order perturbation in @xmath275:@xcite where the intermediate states @xmath352 and @xmath353 carry a heavy hole and a light on neighboring rungs .
the energy denominators given by the difference in energy between the initial ( degenerate with the final ) state and the intermediate states thus take the value @xmath354 starting from an initial state @xmath355 a simple hopping yields the intermediate state @xmath356 with the matrix elements given by for hopping on the upper leg @xmath357 on the middle leg , @xmath358 and on the lower leg , @xmath359 subsequently , a second hopping yields the intermediate state @xmath360 with hopping matrix elements on the upper leg @xmath361 on the middle leg , @xmath362 and on the lower leg , @xmath363 finally to come back to a state carrying the same @xmath69 charges on the rungs , the hopping amplitudes for the three legs are just the complex conjugate of those described in eqs .
( [ eq : t1u])-([eq : t1l ] ) , thus giving for @xmath348 a potential energy @xmath364 and for @xmath350 a hopping term @xmath365 r. s. k. mong , d. j. clarke , j. alicea , n. h. lindner , p. fendley , c. nayak , y. oreg , a. stern , e. berg , k. shten- gel , and m. p. a. fisher , universal topological quantum computation from a superconductor / abelian quantum hall heterostructure , phys . rev .
x * 4 * , 011036 ( 2014 ) .
v. mourik , k. zuo , s. m. frolov , s. r. plissard , e. p. a. m. bakkers , and l. p. kouwenhoven , signatures of majorana fermions in hybrid superconductor - semiconductor nanowire devices , science * 336 * , 1003 ( 2012 ) .
a. das , y. ronen , y. most , y. oreg , m. heiblum , and h. shtrikman , zero - bias peaks and splitting in an al - inas nanowire topological superconductor as a signature of majorana fermions , nat .
* 8 * , 887 ( 2012 ) .
h. o. h. churchill , v. fatemi , k. grove - rasmussen , m. t. deng , p. caroff , h. q. xu , and c. m. marcus , superconductor - nanowire devices from tunneling to the multichannel regime : zero - bias oscillations and magnetoconductance crossover , phys . rev .
b * 87 * , 241401 ( 2013 ) . c. gils , e. ardonne , s. trebst , a. w. w.ludwig , m. troyer , and z. wang , collective states of interacting anyons , edge states , and the nucleation of topological liquids , phys . rev .
lett . * 103 * , 070401 ( 2009 ) .
r. n. c. pfeifer , o. buerschaper , s. trebst , a. w. w. ludwig , m. troyer , and g. vidal , translation invariance , topology , and protection of criticality in chains of interacting anyons , phys .
b * 86 * , 155111 ( 2012 )
. f. d. m. haldane , luttinger liquid theory of one - dimensional quantum fluids . i. properties of the luttinger model and their extension to the general 1d interacting spinless fermi gas , j. phys .
c * 14 * , 2585 ( 1981 ) . i. de p. r. moreira , n. suaud , n. guihry , j. p. malrieu , r. caballol , j. m. bofill , and f. illas , derivation of spin hamiltonians from the exact hamiltonian : application to systems with two unpaired electrons per magnetic site , phys . rev .
b * 66 * , 134430 ( 2002 ) . jean paul malrieu , rosa caballol , carmen j. calzado , coende graaf , and nathalie guihry , magnetic interactions in molecules and highly correlated materials : physical content , analytical derivation , and rigorous extraction of magnetic hamiltonians , chem . rev . * 114 * , 429 - 492 ( 2014 ) . | quantum spin models have been studied extensively in one and higher dimensions .
furthermore , these systems have been doped with holes to study @xmath0@xmath1 models of @xmath2 spin-1/2 .
their anyonic counterparts can be built from non - abelian anyons , such as fibonacci anyons described by @xmath3 theories , which are quantum deformations of the @xmath2 algebra .
inspired by the physics of @xmath2 spins , several works have explored ladders of fibonacci anyons and also one - dimensional ( 1d ) @xmath0@xmath1 models . here
we aim to explore the combined effects of extended dimensionality and doping by studying ladders composed of coupled chains of interacting itinerant fibonacci anyons .
we show analytically that in the limit of strong rung couplings these models can be mapped onto effective 1d models .
these effective models can either be gapped models of hole pairs , or gapless models described by @xmath0@xmath1 ( or modified @xmath0@xmath1@xmath4 ) chains of fibonacci anyons , whose spectrum exhibits a fractionalization into charge and anyon degrees of freedom .
the charge degrees of freedom are described by the hardcore boson spectra while the anyon sector is given by a chain of localized interacting anyons . by using exact diagonalizations for two - leg and three - leg ladders ,
we show that indeed the doped ladders show exactly the same behavior as that of @xmath0@xmath1 chains . in the strong ferromagnetic rung limit ,
we can obtain a new model that hosts two different kinds of fibonacci particles - which we denote as the _ heavy _ @xmath5 s and _ light _
@xmath5 s .
these two particle types carry the same ( non - abelian ) topological charge but different ( abelian ) electric charges . once again , we map the two - dimensional ladder onto an effective chain carrying these heavy and light @xmath5 s .
we perform a finite size scaling analysis to show the appearance of gapless modes for certain anyon densities whereas a topological gapped phase is suggested for another density regime . |
the hygro - thermal behavior of concrete plays a crucial role in the assessment of the reliability and lifetime of concrete structures .
the heat and mass transfer processes become particularly important at high temperatures , where the increased pressure in pores may lead to catastrophic service failures . since
high - temperature experiments are very expensive , predictive modeling of humidity migration and pore pressure developments can result in significant economic savings .
the first mathematical models of concrete exposed to temperatures exceeding @xmath0c were formulated by baant and thonguthai in @xcite . since then , a considerable effort has been invested into detailed numerical simulations of concrete structures subject to high temperatures . however , much less attention has been given to the qualitative properties of the model , as well as of the related numerical methods .
in particular , the only related work the authors are aware of is due to dalk _
et al . _
@xcite , who analyzed the numerical solution of the kiessl model @xcite for moisture and heat transfer in porous materials .
they proved some existence and regularity results and suggested an efficient numerical approach to the solution of the resulting system of highly non - liner equations .
however , the kiessl model is valid for limited temperature range only and as such it is inappropriate for high - temperature applications . in this contribution
, we extend the work @xcite by proving the existence and uniqueness of an approximate solution for the baant - thonguthai model , arising from the semi - implicit discretization in time . a fully discrete algorithm
is then obtained by standard finite element discretization and its performance is illustrated for a model problem of a concrete segment exposed to transient heating according to the standard iso fire curve . here , the focus is on the short - term pore pressure build up , which is decisive for the assessment of so - called thermal spalling during fire . at this point
, it is fair to mention that the baant - thonguthai model was later extended towards more detailed multi - phase description , see e.g. the works of gawin _ et al . _
@xcite , tenchev _ et al . _
@xcite and davie _ et al . _
@xcite for specific examples .
when compared to the original version , these advanced models provide better insight into physical and chemical processes in concrete ( such as influence of gel water , pore water , capillary water , chemical reactions at elevated temperatures , etc . ) . such potential increase in accuracy
, however , comes at the expense of increased number of parameters , which typically reflect complex multi - scale nature of concrete .
hence , their experimental determination is rather complicated and the parameters can often only be calibrated by sub - scale simulations .
therefore , in this work we adopt a pragmatic approach and consider the single - phase baant - thonguthai model with parameters provided by heuristic relations , obtained from regression of reliable macroscopic experiments .
the paper is organized as follows . in section [ model ] , we present the general single - phase , purely macroscopic , model for prediction of hygro - thermal behavior of heated concrete . in section
[ notation ] , we introduce basic notation , the appropriate function spaces and formulate the problem in the strong and variational sense . in section [ assumptions ] , we specify our assumptions on data and modify structure conditions to obtain a reasonably simple but still realistic model of hygro - thermal behavior of concrete at high temperatures due to baant and thonguthai @xcite .
an application of the rothe method of discretization in time leads to a coupled system of semilinear steady - state equations , which ( together with the appropriate boundary conditions ) form a semilinear elliptic boundary value problem , formulated in the form of operator equation in appropriate function spaces .
the existence result for this problem in space @xmath1 with @xmath2 is proven in section [ existence_discrete ] using the general theory of coercive and pseudomonotone operators in banach spaces .
next , the problem is resolved using the finite element method as presented in section [ fem_impl ] . in section
[ example ] , numerical experiments are performed to investigate the moisture migration , temperature distribution and pore pressure build up in the model of concrete specimen exposed to fire , including a simple engineering approach to study the spalling phenomenon .
the heat and mass transport in concrete is governed by the following system of conservation laws : _ energy conservation equation : _ @xmath3 _ water content conservation equation : _ @xmath4 the primary unknowns in the balance equations are the temperature @xmath5 and the water content @xmath6 ; @xmath6 represents the mass of all evaporable water ( free , i.e. not chemically bound ) per m@xmath7 of concrete .
further , @xmath8 and @xmath9 represent the amount of ( internal ) energy and the amount of free water , respectively , in @xmath10 m@xmath7 of concrete , @xmath11 is the heat flux , @xmath12 the isobaric heat capacity of bulk ( liquid ) water and @xmath13 the humidity flux .
following @xcite , the heat flux @xmath11 arises due to the temperature gradient ( fourier s law ) and due to the water content gradient ( dufour flux ) @xmath14 and the flux of humidity @xmath13 consists of the flux due to the humidity gradient ( fick s law ) and due to the temperature gradient ( soret flux ) @xmath15 where @xmath16 , @xmath17 , @xmath18 and @xmath19 are continuous diffusion coefficient functions depending non - linearly on @xmath5 and @xmath6 . to complete the introduction of the model ,
let us specify the boundary and initial conditions on @xmath5 and @xmath6 .
the humidity flux across the boundary is quantified by the newton law : @xmath20 where the right hand side represents the humidity dissipated into the surrounding medium with water content @xmath21 , specified in terms of the film coefficient @xmath22 . as for the heat flux
, we shall distinguish the convective and radiation boundary conditions given by @xmath23 respectively , in which @xmath24 designates the film coefficient for the heat transfer , and @xmath25 is temperature of the environment .
the last expression in eq . expresses the radiative contribution to the heat flux , quantified by the stefan - boltzmann law in terms of the relative surface emissivity @xmath26 and the stefan - boltzmann constant @xmath27 and the temperature difference @xmath28 . with @xmath29
is essential later in the proof of theorem 2 . ]
the initial conditions are set as follows : @xmath30 here , @xmath31 and @xmath32 represent the initial distributions of the primary unknowns @xmath5 and @xmath6 , respectively .
vectors , vector functions and operators acting on vector functions are denoted by boldface letters . throughout the paper , we will always use positive constants @xmath33 , @xmath34 , @xmath35 , @xmath36 , which are not specified and which may differ from line to line .
for an arbitrary @xmath37 $ ] , @xmath38 denotes the usual lebesgue space equipped with the norm @xmath39 , and @xmath40 , @xmath41 , @xmath42 $ ] , denotes the usual sobolev space with the norm @xmath43 .
let @xmath44 be a banach space . by @xmath45,x)$
] we denote the space of all continuous functions @xmath46\rightarrow x$ ] . throughout the paper @xmath47 , @xmath48 ,
denotes the conjugate exponent to @xmath49 .
@xmath50 indicates the partial derivative with respect to time ; we also write @xmath51 .
we consider a mixed initial boundary value problem for a general model of the coupled heat and mass flow in a two - dimensional domain @xmath52 with a lipschitz boundary @xmath53 , which consists of non - intersecting pieces @xmath54 and @xmath55 , @xmath56 .
@xmath54 re presents the part of the boundary which is exposed to fire , whereas the other part denoted by @xmath55 is exposed to atmosphere .
let @xmath57 be the fixed value of the time horizon , @xmath58 , @xmath59 and @xmath60 .
the strong formulation of our problem is as follows : @xmath61 here we assume that all functions are smooth enough .
now we can formulate the problem in the variational sense .
suppose that @xmath62\in c(0,t)^2 $ ] and @xmath63\in w^{1,r}(\omega)^2 $ ] , @xmath64 .
find a pair @xmath65\in c([0,t];w^{1,r}(\omega)^2)$ ] such that @xmath66 holds for all test functions @xmath67\in c_0^{\infty } ( [ 0,t];c^{\infty}(\overline{\omega})^2)$ ] and @xmath68 such pair @xmath65 $ ] is called the variational solution to the system . to the best of our knowledge ,
there are no existence results for the presented model available .
* in @xcite , baant and thonguthai expressed the time variation of amounts @xmath8 and @xmath9 as follows : @xmath69 here @xmath70 and @xmath71 , respectively , are the mass density and the isobaric heat capacity of solid microstructure ( excluding hydrate water ) , @xmath72 represents the total mass of the free water released in the pores by drying .
@xmath73 denotes the enthalpy of evaporation per unit mass and @xmath74 denotes the enthalpy of dehydration per unit mass . * we assume that the parameters @xmath70 , @xmath12 , @xmath75 , @xmath24 , @xmath76 , @xmath27 and @xmath26 are real positive constants . *
the functions @xmath77 and @xmath78 are positive continuous functions , @xmath79 is positive increasing function belonging to @xmath80 , @xmath81 and @xmath82 are given continuous functions of time and @xmath83 , @xmath64 . *
water content @xmath6 is connected with temperature @xmath84 and pore pressure @xmath85 via a so - called sorption isotherm @xmath86 , which has to be determined experimentally for each type of concrete .
we assume @xmath87 to be a continuous function such that @xmath88 for @xmath89 and @xmath90 otherwise . * following @xcite
, we consider the cross effects to be negligible .
this leads to simple phenomenological relations introduced by baant and thonguthai in the form @xmath91 where the thermal conductivity @xmath92 and permeability @xmath93 are assumed to be positive continuous functions of their arguments and @xmath94 is the gravitational acceleration . incorporating the relations into the system
we get the modified baant - thonguthai model with primary unknowns @xmath6 , @xmath5 and @xmath85 consisting of _ conservation laws : _
@xmath95 _ state equation of pore water : _
@xmath96 _ radiation boundary conditions : _
@xmath97 _ neumann boundary conditions : _
@xmath98 and _ initial conditions : _
@xmath99 let @xmath100 be an equidistant partitioning of time interval @xmath101 $ ] with step @xmath102
. set a fixed integer @xmath103 such that @xmath104 . in what follows we abbreviate @xmath105 by @xmath106 for any function @xmath107 .
the time discretization of the continuous model is accomplished through a semi - implicit difference scheme @xmath108 @xmath109 here , we assume that the functions @xmath110 , @xmath111 and @xmath112 are known . in what follows we study the problem of existence of the solution @xmath113 , @xmath114 and @xmath115 . incorporating the relation into the system
we can eliminate the unknown field @xmath6 and consider the problem with only two unknowns @xmath5 and @xmath85 .
consequently , the existence of @xmath114 follows from the existence of @xmath113 and @xmath115 by the relation . for the sake of simplicity
we denote @xmath116:=[p_{n+1},\theta_{n+1}]$ ] .
let us put @xmath117 , @xmath118 , @xmath119 and introduce the functions @xmath120 obviously , we have to solve , successively for @xmath121 , the following semilinear system with primary unknowns @xmath116 $ ] @xmath122 and with the boundary conditions @xmath123 the pair @xmath116\in
w^{1,r}(\omega)^2 $ ] , @xmath124 , is called a variational solution to the system
iff @xmath125 holds for every @xmath126\in w^{1,r'}(\omega)^2 $ ] , @xmath127 .
the main result of this section is the following theorem [ existence_thm ] and corollary [ main_cor ] .
[ existence_thm ] assume that @xmath128\in w^{1,p}(\omega)^2 $ ] with some fixed @xmath129 is known and let @xmath130@xmath131 be satisfied .
then there exists the variational solution @xmath116\in w^{1,p}(\omega)^2 $ ] to the system . in order to prove theorem [ existence_thm ] ,
it is convenient to define the operator @xmath132 by @xmath133),[v_{\pi},v_{\tau}]\rangle&=&\int_{\omega } ( { \widetilde{\kappa}({\mbox{\boldmath{$x$ } } } ) } \nabla \pi ) \cdot \nabla v_{\pi } \,{\rm d}{\mbox{\boldmath{$x$}}}+\int_{\omega } r_1({\mbox{\boldmath{$x$}}},\pi,\tau ) \ ; v_{\pi } \,{\rm d}{\mbox{\boldmath{$x$}}}\nonumber \\ & & + \int_{\omega } ( \widetilde{\lambda}_c({\mbox{\boldmath{$x$ } } } ) \nabla \tau ) \cdot \nabla v_{\tau } \,{\rm d}{\mbox{\boldmath{$x$}}}+\int_{\omega } r_2({\mbox{\boldmath{$x$}}},\pi,\tau ) \ ; v_{\tau } \,{\rm d}{\mbox{\boldmath{$x$}}}\nonumber\\ & & + \ !
\beta_c \pi v_{\pi } { \rm d}{\mbox{\boldmath{$s$}}}+ \!\int_{\gamma_r\cup\gamma_n } \!\!\!\!\!\!\alpha_c\tau v_{\tau } { \rm d}{\mbox{\boldmath{$s$}}}+ \!\!\int_{\gamma_r}\!\!\!e\sigma |\tau|^3\tau v_{\tau } \,{\rm d}{\mbox{\boldmath{$s$}}}\nonumber\\\end{aligned}\ ] ] and the functional @xmath134 by @xmath135\rangle&= & \int_{\omega}\ ! f_1({\mbox{\boldmath{$x$ } } } ) v_{\pi}{\rm d}{\mbox{\boldmath{$x$}}}+ \int_{\omega}\ !
f_2({\mbox{\boldmath{$x$ } } } ) v_{\tau}{\rm d}{\mbox{\boldmath{$x$}}}+ \int_{\gamma_r\cup\gamma_n}\!\!\ !
\beta_c p_{\infty , n}v_{\pi}{\rm d}{\mbox{\boldmath{$s$}}}\nonumber \\ & & + \int_{\gamma_r\cup\gamma_n}\!\!\ !
\alpha_c \theta_{\infty , n } v_{\tau }
{ \rm d}{\mbox{\boldmath{$s$}}}+ \int_{\gamma_r } e\sigma \theta^4_{\infty , n } v_{\tau } { \rm d}{\mbox{\boldmath{$s$}}}\end{aligned}\ ] ] for all @xmath136\in w^{1,p'}(\omega)^2 $ ] .
since we assume @xmath128\in w^{1,p}(\omega)^2 $ ] with some fixed @xmath129 , we have the following estimate for the convective term @xmath137 @xmath138 for all @xmath139 .
one can prove in the similar way that the other integrals in are finite and the functional @xmath140 is well - defined .
the variational problem can now be treated as a single operator equation @xmath141)={\mbox{\boldmath{$f$}}}$ ] .
obviously , @xmath142 implies @xmath143 .
first of all we prove that for a given @xmath144 there exists @xmath116\in w^{1,2}(\omega)^2 $ ] : the solution of the equation @xmath145)={\mbox{\boldmath{$h$}}}$ ] with @xmath146 defined by ( substituting @xmath147 instead of @xmath148 ) for all @xmath136\in w^{1,2}(\omega)^2 $ ]
. [ t_bounded ] @xmath149 is bounded .
test by @xmath116\in w^{1,2}(\omega)^2 $ ] .
take into account @xmath130@xmath150 to get @xmath151),[\pi,\tau]\rangle & \leq & c_1\|\pi\|_{w^{1,2}(\omega)}^2+c_2\|\tau\|_{w^{1,2}(\omega)}^2 \\ & & + c_3\|\pi\|_{l^{2}(\omega)}^2+c_4\|\tau\|_{l^{2}(\omega)}^2 \\ & & + \beta_c\|\pi\|_{l^{2}(\partial\omega)}^2 + \alpha_c\|\tau\|_{l^{2}(\partial\omega)}^2 + e\sigma\|\tau\|_{l^{5}(\partial\omega)}^5 \\
& \leq & c_5 \|[\pi,\tau]\|_{w^{1,2}(\omega)^2}^2 + e\sigma\|\tau\|_{l^{5}(\partial\omega)}^5 .\end{aligned}\ ] ] due to the trace theorem @xcite there exists a constant @xmath152 such that @xmath153 hence @xmath154 is bounded .
@xmath155 [ t_coercive ]
@xmath149 is coercive .
@xmath156 , @xmath150 and the young inequality yield @xmath157 for every @xmath158 and arbitrary @xmath159 .
further , @xmath156 , @xmath150 and the young inequality yield the existence of a positive function @xmath160 and a non - negative function @xmath161 ( both of spatial variable @xmath162 ) such that @xmath163\in \mathbb{r}^2.\end{aligned}\ ] ] now , , , the embedding @xmath164 and the friedrichs inequality imply @xmath165),[\pi,\tau]\rangle & = & \int_{\omega } { \widetilde{\kappa}({\mbox{\boldmath{$x$ } } } ) } |\nabla \pi|^2 \,{\rm d}{\mbox{\boldmath{$x$}}}+ \int_{\gamma_r\cup\gamma_n } \!\!\ ! \beta_c
|\pi|^2 \,{\rm d}{\mbox{\boldmath{$s$}}}\nonumber \\ & & + \int_{\omega } \widetilde{\lambda}_c({\mbox{\boldmath{$x$ } } } ) |\nabla \tau|^2 \,{\rm d}{\mbox{\boldmath{$x$}}}+ \!\ ! \int_{\gamma_r\cup\gamma_n } \!\!\!\!\!\ ! \alpha_c|\tau|^2 \,{\rm d}{\mbox{\boldmath{$s$}}}+ \!\ ! \int_{\gamma_r } \!\!\ ! e\sigma|\tau|^3\tau^2\,{\rm d}{\mbox{\boldmath{$s$}}}\nonumber\\ & & + \int_{\omega } r_1({\mbox{\boldmath{$x$}}},\pi,\tau)\;\pi \,{\rm d}{\mbox{\boldmath{$x$}}}+\int_{\omega } r_2({\mbox{\boldmath{$x$}}},\pi,\tau)\;\tau\,{\rm d}{\mbox{\boldmath{$x$}}}\nonumber \\ &
\geq & \ ! c_1\|\pi\|_{w^{1,2}(\omega)}^2\!+\!c_2\|\tau\|_{w^{1,2}(\omega)}^2 \!-\ ! \eta
\|\pi\|_{l^{2}(\omega)}^2 \!+\ ! c_3\|\tau\|_{l^{2}(\omega)}^2 \!-\ !
c_4 \nonumber \\ &
\geq & c_5 \|[\pi,\tau]\|_{w^{1,2}(\omega)^2}^2 - c_6\end{aligned}\ ] ] with some positive constants @xmath166 and choosing @xmath167 sufficiently small .
[ t_pseudomonotone ] @xmath149 is pseudomonotone .
obviously , since @xmath168 and @xmath169 in @xmath170 , the inequality @xmath171 holds for all @xmath172 and for all @xmath173,[\xi'_1,\xi'_2]\in \mathbb{r}^2 $ ] , @xmath173\neq[\xi'_1,\xi'_2]$ ] .
@xmath155 [ existence_p2 ] the smoothness assumptions on @xmath174 , @xmath175 , @xmath176 and @xmath177 ( the smoothness of @xmath176 and @xmath177 follows from @xmath130@xmath150 and and ) and lemma [ t_bounded][t_pseudomonotone ] imply that @xmath154 is continuous , bounded , coercive and pseudomonotone .
now ( * ? ? ?
* theorem 3.3.42 ) yields the existence of the solution @xmath116\in w^{1,2}(\omega)^2 $ ] to the equation @xmath178)={{\mbox{\boldmath{$h$}}}}$ ] for every @xmath179 .
to get higher regularity results go back and consider @xmath180 with some @xmath129 and rewrite the system in the form @xmath181 it is easy to verify that for the functional @xmath180 defined by and for the weak solution @xmath116\in w^{1,2}(\omega)^2 $ ] ( whose existence is ensured by corollary [ existence_p2 ] ) , the functional @xmath182 given by @xmath183\rangle=\int_{\omega}\ !
\left(f_1-r_1\right ) v_{\pi } + \left(f_2-r_2\right ) v_{\tau}{\rm d}{\mbox{\boldmath{$x$}}}+ \int_{\gamma_r\cup\gamma_n}\!\!\ !
\alpha_c(\tau -\theta_{\infty , n } ) v_{\tau}{\rm d}{\mbox{\boldmath{$s$}}}\\ + \int_{\gamma_r\cup\gamma_n}\!\!\ !
\beta_c(\pi -p_{\infty , n})v_{\pi}{\rm d}{\mbox{\boldmath{$s$}}}+ \int_{\gamma_r } \left(e\sigma ( |\tau|^3\tau-\theta^4_{\infty , n})\right ) v_{\tau}{\rm d}{\mbox{\boldmath{$s$}}}\end{gathered}\ ] ] for every @xmath136\in w^{1,p'}(\omega)^2 $ ] is well defined .
it is known ( see @xcite ) that for given @xmath182 defined by with @xmath129 the neumann problem for the elliptic system ( where the right hand side is represented by @xmath184 ) possess the solution @xmath116\in w^{1,p}(\omega)^2 $ ] .
this completes the proof of theorem [ existence_thm ] .
[ main_cor ] following theorem [ existence_thm ] , @xmath185\in w^{1,p}(\omega)^2 $ ] yields @xmath186\in w^{1,p}(\omega)^2 $ ] with any @xmath129 .
since we suppose @xmath187\in w^{1,r}(\omega)^2 $ ] with @xmath188 , we can conclude , that @xmath189\in w^{1,p}(\omega)^2 $ ] successively for @xmath121 for any @xmath190 if @xmath191 and @xmath129 if @xmath192 .
note that this solution needs not to be unique .
consider a polygonal approximation @xmath193 to @xmath170 , defined by an admissible quadrilateral partition @xmath194 such that @xmath195 , @xmath196 and every element @xmath197 has diameter at most @xmath198 .
@xmath199 is used to denote the number of nodes of the mesh and @xmath200 or @xmath201 stand for the part of the approximate boundary @xmath202 where the radiation and convection boundary conditions are prescribed .
we associate with @xmath203 a finite - dimensional space of piecewise bi - linear basis functions ( recall that @xmath204 ) @xmath205 where @xmath206 denotes the set of polynomials of degree @xmath207 , cf .
* section 5 ) . from the implementation point of view
, it is more convenient to derive the numerical scheme by considering all three unknowns @xmath208 instead of the reduced version ; the approximate solution @xmath209 ^ 3 $ ] is thus provided by the weak form of balance equations and , tested by @xmath210 and @xmath211 , constrained by the isotherm relation @xmath150 enforced at the nodes .
this leads to a system of non - linear algebraic equations @xmath212 where e.g. @xmath213 stores the unknown nodal values of water content , temperature and pore pressure at time @xmath214 , respectively .
the constant matrices in exhibit a block structure @xmath215 and the non - linear term reads as @xmath216 the sub - matrices in and are obtained by the assembly of element contributions , e.g. @xmath217 here , @xmath218 is the assembly operator ( * ? ? ?
* section 2.8 ) and @xmath219 stores the temperature values at the nodes of the @xmath26-th element at time @xmath220 , related to the local approximation @xmath221 in which @xmath222 denotes the operator of linear basis functions .
analogous relations hold for the remaining fields .
the individual symmetric positive - definite matrices @xmath223 can now be expressed as @xmath224 whereas the symmetric positive - semidefinite blocks @xmath225 attain the form @xmath226 the non - linear terms @xmath227 are expressed as @xmath228 and the right hand - side - blocks @xmath229 are provided by @xmath230 in the following example , the integrals were approximated using the @xmath231-point gauss quadrature and the non - linear system was solved iteratively using the newton method with the residual tolerance set to @xmath232 .
our model problem deals with a square concrete specimen of cross - section @xmath233 , which is exposed to fire on the part @xmath54 of the boundary and insulated on the remaining portion @xmath55 ( see fig .
[ concrete_column ] ) .
we assume that the ambient temperature in the vicinity of @xmath54 increases according to the iso fire curve , @xmath234 , with @xmath235 given in minutes ( see fig . [ iso fire curve ] ) .
the uniform initial conditions are set to @xmath236 and @xmath237 and the constants appearing in boundary conditions on @xmath54 are taken as @xmath238 , @xmath239 , @xmath240 , @xmath241 and @xmath242 .
.material constants of concrete [ cols="<,<,>,<,^,^",options="header " , ] in order to assess this behavior more quantitatively , we adopt a conservative engineering approach and assume that the spalling of concrete occurs when the ( appropriately reduced ) pore pressure exceeds the temperature - dependent tensile strength of concrete . in particular
, the spalling at @xmath243 and time @xmath220 occurs when ( * ? ? ?
* section 3.3 ) @xmath244 where @xmath245 is the porosity of concrete , @xmath246 and @xmath247 denote the finite element approximations of pore pressure and temperature for mesh size @xmath248 and the @xmath103-th time step and @xmath249 as a function of @xmath5 is provided by a piecewise linear relation ( * ? ? ?
* section 4.4 ) @xmath250 where @xmath251 designates the initial tensile strength of concrete . the spatial distribution of the spalling damage for @xmath252 and @xmath253 mpa appears in fig .
[ fig_spalling ] . here
, three different zones can be distinguished .
the first region ( a ) , highlighted in black color , corresponds to elements failed due to spalling damage as predicted by the criterion for the element center .
the second zone ( b ) corresponds to the part of the structure in which the local strength is sufficiently high to sustain the pore pressure , but its stability is lost due the explosive spalling of the former region .
finally , the light gray zone ( c ) indicates the portion of the cross - section still capable of transmitting stresses due to mechanical loading , which is thereby responsible for the structural safety during fire .
this outcome has been achieved with the financial support of the ministry of education , youth and sports of the czech republic , project no . 1m0579 , within activities of the cideas research centre ( the first author ) and project no .
msm6840770001 ( the second author ) . additional support from the grants 201/09/1544 ( the first author ) ,
103/08/1531 and 201/10/0357 ( the third author ) provided by the czech science foundation is greatly acknowledged .
davie , c.j .
pearce and n. bicanic , coupled heat and moisture transport in concrete at elevated temperatures - effects of capillary pressure and adsorbed water , numer .
heat transfer part a 49 ( 2006 ) 733763 .
mazya , j. rossmann , weighted @xmath254 estimates of solutions to boundary value problems for second order elliptic systems in polyhedral domains .
zamm z. angew .
83 ( 2003 ) , no . 7 , 435467 | in this paper , we study a non - linear numerical scheme arising from the implicit time discretization of the baant - thonguthai model for hygro - thermal behavior of concrete at high temperatures . existence and uniqueness of the time - discrete solution in two dimensions
is established using the theory of pseudomonotone operators in banach spaces .
next , the spatial discretization is accomplished by the conforming finite element method .
an illustrative numerical example shows that the model reproduces well the rapid increase of pore pressure in wet concrete due to extreme heating .
such phenomenon is of particular interest for the safety assessment of concrete structures prone to thermally - induced spalling .
heat and moisture transfer phenomena in concrete at high temperatures , rothe method , pseudo - monotone operators , finite element discretization , spalling |
confinement of quarks is a very old problem that until now has not been solved . at
low energies perturbative qcd can not be applied because the coupling constant is large .
effective theories and lattice qcd have given important contributions to the understanding of this problem .
a very useful approach to the confinement problem is the study of large n gauge theories because of the possibility of obtaining analytical and numerical results for large ( t hooft ) coupling in a simple way via the ads / cft correspondence @xcite .
this correspondence relates large @xmath4 gauge theories on a @xmath5 dimensional manifold @xmath6 to 10-d string theories on @xmath7 being @xmath6 the boundary of @xmath3 and @xmath8 some compact space . in particular , in the context of ads / cft correspondence witten showed that @xmath9 super yang - mills ( sym ) theory in @xmath10 at strong coupling experiments a confinement / deconfinement transition that corresponds holographically to a gravitational transition between @xmath11 and black hole @xmath11 defined in global coordinates @xcite .
this gravitational transition was studied before by hawking and page @xcite .
the confinement criteria stated in @xcite is the behavior of the free energy of large n @xmath12 sym theory in @xmath13 .
the theory is confining at low temperatures because the free energy is of order 1 ( color singlet contribution ) while is deconfining at high temperatures because the free energy is or order @xmath14 ( gluons contribution ) .
this criteria can be extended to the cases @xmath0 .
finite temperature qcd lives in @xmath15 which is the boundary of @xmath11 in poincar coordinates .
sym theory in @xmath15 is not confining as can be seen using the wilson loop prescription @xcite .
this prescription relates a wilson loop of the gauge theory to the world - sheet area of a static string living in @xmath16 in poincar coordinates . in particular
, the free energy of a quark anti - quark pair in @xmath15 can be calculated from the correlator of two polyakov loops and exhibits a coulombian behavior .
this behavior can be interpreted as deconfining because the free energy vanishes when the quark anti - quark distance goes to infinity .
the wilson loop prescription was generalized to other curved spaces and a confinement criteria was established @xcite .
the quark anti - quark free energy at high temperatures for the space @xmath15 was calculated in @xcite by considering an @xmath11 black hole in poincar coordinates .
other wilson loop calculations involving gauge / string duality can be found at @xcite .
there are some phenomenological models that introduce confinement in @xmath15 .
a well succeeded one is the hard wall model which consists on introducing a hard wall on poincar @xmath11 space interpreted holographically as an infrared cut - off for the gauge theory .
this model was motivated on the calculation of scattering amplitudes @xcite and leads to confinement at low temperatures @xcite . at high temperatures
there is a confinement / deconfinement transition corresponding to a hawking - page transition between ( poincar ) @xmath3 and ( poincar ) black hole @xmath3
. other calculations of quark anti - quark potential using phenomenological models can be found at @xcite . in spite of being a little far from real world ,
large @xmath4 gauge theories on @xmath17 are physically interesting .
for instance , it is possible to use perturbation theory in the small radius regime so that thermal transitions can be studied and connect them with the hawking - page and hagedorn transitions which are important in string theory and quantum gravity @xcite .
the mean of this work is to calculate the free energy of a quark anti - quark pair living in @xmath17 space where @xmath18 represents the imaginary time coordinate with a period @xmath19 identified with the inverse of the temperature . for this purpose
we consider a wilson loop containing two temporal lines and two spatial lines .
the quark anti - quark free energy can be extracted from the correlator of two polyakov loops . in order to relate our wilson loop to the polyakov loop correlator
, the size of the temporal lines must be @xmath19 .
the dual description consists in a static string with the quark and anti - quark at the end points .
this string configuration gives the dominant contribution to the connected part of the polyakov loop correlator ( see for example @xcite ) .
then what we mean by potential in this article is the quark anti - quark free energy obtained from the connected part of the polyakov loop correlator .
we follow the procedure of @xcite to calculate this potential .
the space @xmath17 is the boundary of two spaces : global @xmath3 and black hole @xmath3 in @xmath20 dimensions .
we first calculate explicitly the @xmath21 case where asymptotic limits for the potential are studied and then we generalize our results to the cases @xmath22 .
we find a coulomb - like behavior of the free energy in the case of global @xmath1 and show that this behavior can be interpreted as confining . at high temperatures we calculate the quark anti -
quark free energy for the two possible black hole solutions : the big black hole and small black hole in @xmath1 . in the case
@xmath23 our results for the big black hole agree with those obtained before by landsteiner and lopez @xcite where it was defined a screening length for the quark anti - quark free energy . here
, we calculate this screening length as a function of the temperature and the horizon position . for the big black hole
the screening length never reaches the maximal distance while for the small black hole this could happen
. this will imply ( for @xmath24 ) that the big black hole @xmath1 is indeed deconfining at any ( high ) temperature while the small black hole could give confinement in the limit @xmath25 .
the hawking - page criteria states that the small black hole solution is instable , so deconfinement is guaranteed at high temperatures .
our results indicate a confinement / deconfinement transition for the quark anti - quark free energy in @xmath17 corresponding to a hawking - page transition from @xmath1 to black hole @xmath1 .
in this section let us calculate the quark anti - quark free energy on the compact space @xmath26 .
this space is the boundary of the @xmath27 space in global coordinates .
the metric of global @xmath27 with euclidean signature is @xmath28 where @xmath29 , @xmath30 and @xmath31 . the time coordinate is compact : @xmath32 with @xmath33 . for simplicity
we consider @xmath34 and define a new coordinate @xmath35 with dimension of length .
the metric then reads @xmath36 where @xmath37 .
we consider the rectangular wilson loop of fig [ fig1 ] living in the @xmath3 boundary ( @xmath38 ) .
the quark and anti - quark are localized at @xmath39 and @xmath40 .
the distance between the quarks @xmath41 has a maximum value @xmath42 because of the compactness of the @xmath43 coordinate .
the dual configuration consists on a static string living on @xmath3 with the quarks at the end points .
the nambu - goto action for this string is @xmath44 where @xmath45 and we parameterized the string configuration by @xmath46 and @xmath47 valid for @xmath48 ( the other region should be parameterized in other way but it is not necessary to work there because of the symmetry of the problem ) .
the solutions for the static string configuration ( corresponding to a minimum action ) represent geodesics of the ads space where the string reaches a minimum for the radial coordinate @xmath49 ( see fig 2 ) . using the prescription of @xcite we find @xmath50^{1/2 }
\nonumber\\ & = & \ , \int_{1}^{\infty } \frac{2 r \sqrt{1+\lambda^2}dy}{y^4\sqrt{\lambda^2+\frac{1}{y^2}}\sqrt{1-\frac{1}{y^2 } } \sqrt{\frac{1+\lambda^2}{y^2}+\lambda^2}}\end{aligned}\ ] ] where @xmath51 and @xmath52 .
the regulated free energy is given by @xmath53\end{aligned}\ ] ] where the second integral of the first line represents a regulator interpreted as the sum of the quark anti - quark masses .
the integrals for @xmath41 and @xmath54 are elliptical so we calculate them numerically .
we find a free energy with coulombian behavior at small distances and that goes to zero only when @xmath41 goes to @xmath55 .
we will explain in the last section why this behavior for the free energy can be interpreted as confining .
our results are shown in figs .
[ fig4 ] and [ fig5 ] .
it is interesting to analyze the behavior of the free energy in the asymptotic limits @xmath56 and @xmath57 . for this purpose
it is convenient to define the variable @xmath58 .
then @xmath59 @xmath60\!\ ! -\ ! { \lambda r \over \pi \alpha'}\ ] ] now we can evaluate the asymptotic limits . in this case
@xmath61 is very large , then we have @xmath62 where @xmath63 .
the integral for the free energy f is @xmath64- { \lambda r \over \pi \alpha'}= -{a \over l } \label{coulomb}\ ] ] where @xmath65 with @xmath66 .
this result is similar as the one obtained working in the poincar @xmath3 metric @xcite . for small distances @xmath43
this metric can be thought as the asymptotic limit @xmath67 of our global @xmath3 metric , which is @xmath68 but note that in this case the coordinate @xmath43 has to be small .
the behavior of this free energy is analogous to that of the phenomenological cornell potential for a heavy quark anti - quark pair at small distance . in this case
@xmath69 is small then the integral for l is @xmath70 \label{l}\end{aligned}\ ] ] the integral for the free energy is given by @xmath71 \ ! \!- \ !
\!{\lambda r \over \pi \alpha ' } \nonumber \\ & = & { \lambda r \over 4 \pi \alpha ' } [ { l \over r } - \pi ] \end{aligned}\ ] ] where we approximated the term @xmath72 from equation ( [ l ] ) we have that @xmath73 so we finally obtain @xmath74 this result could be obtained considering the asymptotic behavior @xmath75 of the metric : @xmath76 with the quarks in a brane localized at @xmath77 . note that when @xmath41 reaches @xmath78 the free energy takes its maximal value @xmath79 .
the black hole @xmath27 metric in global coordinates is defined by @xmath80 where @xmath29 , @xmath30 and @xmath32 .
the factor @xmath81 is included so that @xmath82 is the mass of the black hole . also , the spacetime is restricted to the region @xmath83 , with @xmath84 the largest solution of the equation @xmath85 the region @xmath86 is the horizon of the black hole metric .
the temperature of the black hole is related to @xmath87 by @xmath88 note that there is a minimum temperature @xmath89 for the existence of this black hole that corresponds to a critical horizon @xmath90 .
this critical horizon divides two possible black hole solutions : the big black hole ( bbh ) with an horizon that grows with temperature and the small black hole ( sbh ) with an opposite behavior .
the hawking - page criteria states that for temperatures @xmath91 the big black hole is stable only for @xmath92 while the small black hole is always instable @xcite .
we again consider @xmath34 and define @xmath93 so the black hole metric reads @xmath94 then @xmath95 curve @xmath96 , correspond a u - type geodesic and @xmath97 a box - type geodesic .
, width=207 ] now , we put a quark at @xmath39 and an anti - quark at @xmath98 . in this space
we have two possible solutions for the static string : the first one is a u - type solution similar to the one obtained before for the @xmath3 where the string reaches a minimum for the radial coordinate @xmath49 , the second solution is a string reaching the horizon ( see fig [ graficocaixote ] ) .
depending on the value of @xmath41 one of those solutions will have the minimum free energy .
the distance and free energy for the quark anti - quark pair are in this case @xmath99^{1/2 } } \\ & & \nonumber\\ & & \nonumber \\ f & = & \frac{r\lambda}{\pi \alpha ' } \int_0^{1 } dt \
, t^{-2}\big [ \frac{\sqrt{\lambda^2+t^2-\frac{\gamma}{\lambda } t^3}}{\sqrt{1-t}[(1+t)(\lambda^2 + ( 1+\lambda^2)t^2)-\frac{\gamma}{\lambda } t^3]^{1/2}}-1 \big ] - { r\over \pi \alpha ' } \big(\lambda -\lambda_+\big ) \ , \label{eqenergybh}. \end{aligned}\ ] ] where @xmath100 , @xmath51 and we defined vs @xmath101 for black hole @xmath27 at different temperature values .
[ energiabh],width=340 ] @xmath102 with @xmath103 .
note that @xmath104 because @xmath105 .
we also used the mass regulator @xmath106 from eq ( [ eqenergybh ] ) it is not difficult to show that the box - type solution has always zero free energy .
the free energy of the u - type solution is negative for small @xmath41 and exhibits the same coulombian behavior found in eq .
( [ coulomb ] ) .
when we increase the distance the free energy increases and reaches the zero value for some length @xmath107 that depends on @xmath108 . when @xmath109 the free energy of the u - type solution becomes positive and is inestable compared with the box - type solution ( which has zero free energy for every @xmath41 ) .
so we find a transition for the free energy corresponding to a transition between the u - type and the box - type solutions being @xmath107 a screening length for which the free energy reaches the zero value .
we show in fig .
[ energiabh ] the free energy as a function of the distance for the two black hole solutions at different values of the temperature .
note that the big black hole solution has a screening length that is always lower than the maximal distance @xmath110 while the small black hole has a confining limit corresponding to @xmath111 .
vs @xmath112 for black hole @xmath27 .
[ tlcrit],width=264 ] this limit is achieved when @xmath25 that corresponds to @xmath113 when the small black hole @xmath27 becomes @xmath27 ( which is confining ) . in fig .
[ tlcrit ] we show the temperature as a function of the screening length .
it is interesting to obtain the screening length dependence on @xmath108 for the big black hole . from fig .
[ lcritlambdamais ] we see that for large @xmath108 the product @xmath114 is constant and approximately equal to @xmath115 .
this behavior was also obtained for the poincar black hole @xcite .
vs @xmath108 for black hole @xmath27 .
[ lcritlambdamais],width=321 ]
the quark anti - quark potential in @xmath0 can be calculated in a similar way as the @xmath117 case . at low temperatures we have to deal with the global @xmath1 space .
the metric of this space for @xmath118 is given by @xmath119 where @xmath120 @xmath121 , @xmath30 , @xmath31 and @xmath32 .
if we choose @xmath122 and define @xmath123 we arrive at the same metric of eq.([metricaadsx ] ) so we find the same free energy found before working in @xmath27 ( fig [ fig5 ] ) . at high temperatures , we use the black hole @xmath1 ( @xmath124 ) with metric @xmath125 where @xmath126 .
the temperature is now related to @xmath87 by @xmath127 now the critical horizon is @xmath128 , the hawking - page horizon is always @xmath129 which is always above @xmath130 so the big black hole ( @xmath131 ) is stable while the small black hole is always instable @xcite . choosing again @xmath122 and defining @xmath123 we obtain @xmath132 so we find in this case @xmath133 leading to @xmath134^{1/2 } } \nonumber\\ & & \nonumber \\ & & \nonumber \\ \ ! \ ! \ ! \ ! \ ! f \ ! \ ! \ ! & = & \ ! \ ! \!\frac{r \!\lambda}{\pi \alpha ' } \ ! \ ! \ ! \ !
\int_0^{1}\ ! \ ! \ ! \ ! \ ! dt \ ,
t^{^{\ ! -2 } } \!\left[\ !
\!\frac{\sqrt{\ ! \lambda^2\ ! + \ !
t^2 \ ! - \ !
\frac{\gamma}{\lambda^{n-2 } } t^n } } { [ ( 1-t^2 ) ( ( 1\ !
+ \lambda^2 ) t^2+\lambda^2)+\frac{\gamma}{\lambda^{n-2}}(t^4-t^n)]^{1/2}}\ ! -\ ! 1\ !
\right ] \ ! \ ! -\ ! \frac{r } { \pi \alpha ' } \big(\lambda - \lambda_+ \big ) \!\nonumber \\ & & \label{eqenergybh2}\end{aligned}\ ] ] where @xmath135 @xmath136 vs @xmath137 for black hole in @xmath27,@xmath11 and @xmath138 + ( @xmath139 ) . [ energiabh2],width=8 ] these integrals are slightly different from the integrals found before but the result for the free energy is very similar .
we find ( again ) a transition between a u - type and a box - type solution and a screening length @xmath107 so that the free energy is negative below @xmath107 and zero above .
we compare in fig [ energiabh2 ] the potential energies as a function of the distance for @xmath140 and @xmath141 at a fixed horizon position @xmath142 .
we see that the energies coincide at small distances and there is a little diminution of the screening length when the dimension grows up
. then the screening length will never reach the maximum distance @xmath143 in the big black hole .
this means that the corresponding gauge theory in @xmath0 is deconfining at high temperatures for any @xmath24 . for the special case @xmath23 which corresponds to sym theory in @xmath144 our results agree with those obtained before by landsteiner and lopez @xcite where this screening length was defined
we calculated the free energy of a quark anti - quark pair in a compact space @xmath0 with the period of @xmath145 being the inverse of the temperature . for the global @xmath3
we find a quark anti - quark free energy that has a coulomb - like behavior and that goes to zero only at the maximum distance @xmath55 in the circle of fig [ fig2 ] .
so if the quark is in a fixed position the only way to obtain a zero free energy is to put the anti - quark at a distance @xmath146 .
this means that the anti - quark does not have the freedom to move in the @xmath43 coordinate without feeling the potential so this corresponds to a confinement scenario .
this scenario is in accordance with the one obtained in @xcite . at high temperatures the static string in the black hole
metric leads to a coulomb - like free energy that goes to zero at a screening length @xmath107 .
we calculate the dependence of @xmath107 on the horizon position and temperature and show that it only reaches @xmath55 for the small black hole solution which is instable by the hawking - page criteria .
for the ( stable ) big black hole we found that @xmath107 is always lower than @xmath55 so considering a quark in a fixed position the anti - quark has some freedom to move in the @xmath43 coordinate with corresponds to a deconfinement scenario .
we conclude that the transition of the quark anti - quark free energy in @xmath0 when one goes from @xmath3 to black hole @xmath3 is indeed a confinement / deconfinement transition when included the hawking - page criteria of stability .
this confirms the interpretation of the confinement / deconfinement transition in @xmath0 as the dual of the hawking - page transition supported by the entropy jump of the spaces .
* acknowledgments * : we would like to thank nelson braga and henrique boschi for important suggestions and comments .
we also thank the cbpf for the hospitality .
the authors are partially supported by cnpq and claf .
h. boschi - filho , n. r. f. braga and c. n. ferreira , phys .
d * 73 * , 106006 ( 2006 ) [ erratum - ibid .
d * 74 * , 089903 ( 2006 ) ] [ arxiv : hep - th/0512295 ] .
s. j. rey , s. theisen and j. t. yee , nucl .
b * 527 * , 171 ( 1998 ) [ arxiv : hep - th/9803135 ] .
a. brandhuber , n. itzhaki , j. sonnenschein and s. yankielowicz , phys .
b * 434 * , 36 ( 1998 ) [ arxiv : hep - th/9803137 ] .
h. boschi - filho , n. r. f. braga and c. n. ferreira , phys .
d * 74 * , 086001 ( 2006 ) [ arxiv : hep - th/0607038 ] .
o. andreev and v. i. zakharov , jhep * 0704 * , 100 ( 2007 ) [ arxiv : hep - ph/0611304 ] .
h. dorn and h. j. otto , jhep * 9809 * , 021 ( 1998 ) [ arxiv : hep - th/9807093 ]
. s. a. hartnoll and s. prem kumar , phys .
d * 74 * , 026001 ( 2006 ) [ arxiv : hep - th/0603190 ] .
o. aharony , j. marsano , s. minwalla , k. papadodimas and m. van raamsdonk , phys . rev .
d * 71 * , 125018 ( 2005 ) [ arxiv : hep - th/0502149 ] .
o. aharony , j. marsano , s. minwalla , k. papadodimas and m. van raamsdonk , adv .
* 8 * , 603 ( 2004 ) [ arxiv : hep - th/0310285 ] .
d. bak , a. karch and l. g. yaffe , jhep * 0708 * , 049 ( 2007 ) [ arxiv:0705.0994 [ hep - th ] ] . | we investigate the finite temperature quark anti - quark problem in a compact space @xmath0 by considering static strings in global @xmath1 space with @xmath2 . for high temperatures we work in the black hole metric where two possible solutions show up : the big black hole and the small black hole .
using the ads / cft correspondence , we calculate the quark anti - quark potential ( free energy ) as a function of the distance .
we show that this potential can be intrepeted as confining for the @xmath3 space and deconfining for the big black hole .
we find for the small black hole a confining limit for the potential but this solution is instable following the hawking - page criteria .
our results for the free energy reinforce the witten interpretation of the confinement / deconfinement transition as the dual of the well - known hawking - page transition . |
a good knowledge and understanding of the chromosphere is essential for making progress in the questions of chromospheric and coronal heating . studying the chromosphere
is made challenging by the fact that only few spectral lines in the visible and ir spectral ranges accessible from the ground are formed there . of these even fewer sample the middle and upper chromosphere ( hei 10830 , h@xmath2 , caii lines ) .
therefore , it is of considerable interest to explore further avenues .
besides lines in the vacuum ultraviolet , e.g. the ly alpha and neutral carbon and oxygen lines , observed by e.g. sumer @xcite or vault @xcite , another source of information on the chromosphere are the h and k lines of mgii .
these lines are expected to sample a large height range , reaching from the lower to the upper chromosphere , but observations made in these lines are rare due to their inaccessibility from ground . in the mid-1970s ,
space - based observations of mgii h&k spectra were carried out with the lpsp instrument onboard the oso-8 mission @xcite , followed in the early 1980s by observations with the ultraviolet spectrometer and polarimeter ( uvsp ) onboard the solar maximum mission @xcite . a somewhat higher spatial resolution ( one or more arcsec )
was reached with the 30 cm telescopes of the balloon - borne spectrograph rasolba @xcite in 1986 as well as with the sounding - rocket experiments hrts-9 in 1995 @xcite and the solar ultraviolet magnetograph investigation ( sumi ) in 2010 @xcite .
mgii h&k spectra and slit jaw images are also being recorded with the 20 cm telescope of the interface region imaging spectrograph ( iris ; * ? ? ?
* ) , which was launched on 2013 june 27 , i.e. after the images discussed here were taken .
in this paper we present the first high - resolution mgiik images of quiet and active solar regions which were recorded with the sunrise filter imager ( sufi ; * ? ? ?
* ) during the second science flight of the balloon - borne observatory sunrise @xcite in june 2013 .
its 1 m primary mirror makes sunrise the largest solar telescope to have left earth s surface .
the second science flight of the ballon - borne solar observatory sunrise started on 2013 june 12 from esrange near kiruna in northern sweden .
it reached a mean float altitude of @xmath3 km , allowing seeing - free observations in the visible and near - ultraviolet spectral range .
after a total flight time of 127 hours , the balloon landed safely in boothia , a peninsula in northern canada .
technical details of the 1 m aperture gregory telescope are described by @xcite .
image stabilization and feature tracking were achieved by the gondola s pointing system in conjunction with a six - element shack@xmath4hartmann correlating wavefront sensor ( cws ) which controlled the telescope s focus mechanism , and a tip / tilt mirror @xcite .
the observations presented here were taken by the sunrise filter imager ( sufi ; * ? ? ?
some of the filters from the 2009 flight , described by @xcite , were replaced to put more emphasis on the chromosphere .
of greatest interest for the purposes of the present paper is the addition of a filter with a full - width - half - maximum ( fwhm ) of @xmath1 centered on @xmath0 .
figure [ fig1 ] displays the filter profile overplotted on the solar spectrum ( average plage ; * ? ? ?
the filter covers the core of the mgiik line , but also gets a minor contribution from mgiih and part of the wings of the two lines , as well as some weak photospheric absorption features . due to the strength of the emission core ,
this provides much of the signal in the filter , although the wings also contribute .
note that in the quiet sun the contribution of photospheric features ( reversed granulation ) dominates for broad - band filters ( e.g. * ? ? ?
* ; * ? ? ?
in plage regions the chromospheric contribution is enhanced due to the increased strength of the mgii h&k emission peaks .
another important change was the inclusion of a second wavelength band centered on the core of the caiih @xmath5 , in addition to the @xmath6 wide filter that already flew on the 2009 flight .
the new @xmath7 wide filter gets more contribution from the h@xmath8 and h@xmath9 features . here
we present first results from the mgiik channel .
further descriptions of instrumentation updates relative to the 2009 flight as well as of the flight and the full data set obtained will be given elsewhere .
the mgiik line lies in the hartley absorption band of ozone , close to the maximum of absorption around @xmath10 . even at the float altitude of sunrise
, there is sufficient stratospheric ozone to significantly attenuate the sunlight reaching the telescope ( we estimate roughly by @xmath11 orders of magnitude , based on the modelling by * ? ? ?
* ) . to reduce the influence of this absorption , we restricted observations of mgiik to times around local noon , when the solar elevation was above @xmath12 , and scheduled ballast drops to the morning hours to ensure that the payload was at maximum altitude around local noon .
since sunrise s flight trajectory lay to the north of the arctic circle , the sun s elevation never exceeded about @xmath13 .
even after these measures , the mgii images needed to be integrated for @xmath14s and the s / n rarely exceeded 25 - 30 in the quiet sun .
we concentrate here on the first presentation of the mgii images and compare them with sufi s caiih and @xmath15 observations .
the @xmath15 filter was unchanged compared to the 2009 flight and we refer to @xcite for further details .
sufi is equipped with a phase - diversity ( pd ) prism .
after the images were dark - current corrected and flat - fielded , they were pd reconstructed using individual wavefront errors ( level-2 data , see * ? ? ?
* ; * ? ? ?
* for more details ) .
all intensity images were normalized to the intensity level of the mean quiet sun , @xmath16 , defined as the average of the whole image .
sufi s plate scale is 0@xmath170207 pixel@xmath18 and the field of view ( fov ) was about 15@xmath1939 .
the flight altitude of on average @xmath20 km provided seeing - free observations , nonetheless the pointing accuracy was changing due to varying shear wind strength , so that the cws control loop could not be closed all the time . in total we got 105 focussed magnesium images during the 2013 flight from which we show one of the highest resolution ones obtained in a quiet part of the sun , at solar disk center , in figure [ fig2 ] .
it looks qualitatively similar to high - resolution images recorded in caiih , e.g. during the first sunrise flight @xcite : it is criss - crossed by brightenings roughly on the granular scale . in between are bright points present at various locations in the internetwork . these appear to be similar to caiih bright points ( see * ? ? ?
* and references therein ) .
a concentration of bright features is visible near the top of the frame in a strong network element .
the resolution of the image is approximately 0@xmath172 .
this value is deduced from both , measuring the sizes of the smallest visible structures and from the cut - off of the noise filter in the pd reconstruction , which is self - consistently computed from the data @xcite .
this resolution , although considerably higher than previous images taken in mgii h or k , is below the diffraction limit .
it is _ a priori _ unclear if the low resolution is mainly due to accumulated jitter during the @xmath14s exposure , or if it is caused by the evolution of the chromospheric structures during the exposure .
note that gondola - jitter was largest around local noon , when the telescope was pointing upward at a slanting angle , a less stable configuration than at times when the sun was closer to the horizon and the telescope pointed more horizontally .
the rms intensity contrast of the displayed region is 24.2% .
this value is higher than the contrasts obtained by @xcite in caiih ( 21 - 22% ) .
the comparison is not so clear - cut , however , due to the higher activity level during the 2013 flight , and possibly a difference in spatial resolution . in order to judge the effect of having an increased number and size of network features ( possibly enhanced network ) in the field of view of the mgiik data , we additionally determined the rms contrast exclusively of clearly internetwork features , e.g. the region between 12 and 30 in the y direction of fig .
this gave a value of 19.1% which is at the lower end of the caiih intensity contrasts found by @xcite , who did not explicitly exclude network features from their data .
therefore , disregarding the different filter widths for mgiik and caiih observations , the resulting rms intensity contrasts are very similar .
the grey scale in the right panel of figure [ fig2 ] is limited to three times the rms range of the image which leads to a better visibility of reverse granulation ( or possibly shock waves ) that we find at the mg wavelength .
note that in the lower part of the figure a similar structure is present as had first been noticed by @xcite in caiih images obtained during the 2009 flight of sunrise : there are often 2 parallel bright stripes , between which a dark lane is located ( see the encircled example ) .
obviously the mgiik line displays a similar structure .
this structure may have to do with the width of the filter , since the monochromatic synthetic mgii images computed by @xcite do not display this feature .
we detected 42 bright points ( bps ) with the manual method used by @xcite and determined their peak intensities .
the mean bp peak intensity in mgiik was found to be @xmath21 , the highest value being @xmath22 .
these values are considerably higher than the contrasts found by @xcite in the caiih core , with an average contrast of approximately @xmath23 and a maximum value of @xmath24 .
note , however , that the caiih contrasts were obtained in the broad @xmath6 filter ; as we shall see below the mgiik images are more strongly reminiscent of the images recorded with the narrower @xmath7 caiih filter .
the bps in mgii have a mean width of 0@xmath1755 .
this value is larger than the mean found in caii images by @xcite ( 0@xmath1735 ) and 0@xmath172 by @xcite .
the latter authors restricted themselves to particularly small bps ( only those with diameters @xmath25@xmath173 were considered ) . .
the grey scale covers the full intensity range in the left panel , while it covers only three times the rms range of the image in the right panel , so that the brightest features are saturated .
this view enhances the weaker structures .
the oval marks a feature discussed in the text . ]
nearly simultaneous sufi observations at the wavelengths of mgii @xmath0 , the narrow ( @xmath7 ) band caii @xmath5 , and @xmath15 are displayed in figure [ fig3 ] .
the fov exhibits , in addition to quiet granulation , two intrusions of plage around y=10 and y=30 .
while the @xmath15 wavelength is mainly formed in the deep photosphere and shows normal granulation in the quiet regions , the mgii @xmath0 and caii @xmath5 wavelengths obviously look more chromospheric , i.e. the granulation pattern is reversed , although some of the bright structures may be signatures of shock waves .
the contrast is higher and the bright features are more diffuse than in the photosphere .
furthermore , the bright features are also more diffuse in mgiik than in caiih . in the lower part , below 10 , of the two chromospheric panels we find first stirrings of loop - like structures .
these are clearly more pronounced in the mgii line than in caii .
since the solar scene is not totally quiet , the rms intensity contrast calculated over the entire mgii image , 29.8% , is distinctly higher than in figure [ fig2 ] . for the caii @xmath5 and @xmath15 image
we find 22.5% and 20.4% , respectively .
interestingly , these last two values are not higher than what @xcite found in the quiet sun . for @xmath15
this is acceptable since the density of bps is relatively low when averaged over the whole fov .
however , for caii it is surprising since the wavelength band of this image is narrower than at the time of the 2009 flight .
one implication may be that either the features become more diffuse at the greater heights sampled by caii in the narrower filter , or that this particular image is less sharp than some of the sharpest ones from the first flight of sunrise .
wavelength bands observed by sufi on june 12 at 12:25 ut . the telescope pointed to a region at disk center with patches of quiet sun alternating with bands of plage . ]
fibrils become clearly visible when observed nearer to the limb in regions of high magnetic activity .
this is displayed in fig .
[ fig4 ] , in which a part of ar11768 is depicted , observed at @xmath26 .
the field - of - view was located between the main sunspots of the active region . while the photospheric @xmath15 image shows numerous faculae that are typical of active regions at higher heliocentric angles , the two chromospheric images clearly reveal a forest of bright chromospheric fibrils .
the fibrils in this figure are more diffuse in mgii than in caii as demonstrated by the intensity cuts across fibrils in fig .
note that the structure along the fibrils ( e.g. due to finite length of the fibrils ) looks rather similar in both the mgii and caii . but observed on june 16 at 13:13 ut while the telescope pointed to a plage region between the two sunspots of ar11768 ( @xmath26 ) .
the white arrow in the top right corner of the image points towards the disk center .
the white lines in the bottom part mark the cuts shown in fig .
[ fig5 ] . ] : mgii - dotted line , caii - solid line , @xmath15 - dashed line . ]
we have presented the first high - resolution images of the solar atmosphere taken in the mgiik line .
these were obtained with the sufi instrument on the sunrise balloon - borne solar observatory during its second science flight that took place on 12 - 17 june 2013 .
the atmospheric absorption by stratospheric ozone is still very strong , even at altitudes above @xmath20 km , so that long integration times were needed , and only images with relatively low s / n were obtained .
the images display a pattern that may be due to reversed granulation or shock waves in the quiet - sun internetwork , very similar to interference - filter images taken in the core of caiih .
figure 3 of @xcite , who computed the mgii h&k lines in a snapshot from a radiation mhd simulation carried out with the bifrost code @xcite , displays such a pattern at the mgiik2v and k2r wavelengths , but not in the mgiik3 image .
this suggests that the sunrise / sufi mgiik filter samples the k2 peaks much more than k3 , which is not surprising given the greater strength of the peaks and their larger spectral coverage .
the rms contrast in mgiik of these very quiet regions is 19.1% , obtained after masking out network areas in quiet sun images .
this is very close to the rms contrast of the quiet sun obtained by @xcite in caiih , also with sunrise / sufi during its first science flight in 2009
. at disk center mgiik also shows bright points at the same locations at which such features are seen in the photospheric @xmath15 line and in the core of caiih .
these bps are considerably brighter in mgii than at other wavelengths .
mgiik provides a mean contrast of @xmath27 , which is considerably higher than the values of @xmath28 and @xmath29 found by @xcite in caiih and at @xmath30 , respectively , the two wavelengths displaying the largest bp contrasts during the first sunrise flight .
this makes the mgiik line very promising for identifying and tracking small - scale magnetic features , although the structures may be more diffuse than at photospheric heights ( see below ) .
note , however , that the computations of @xcite , do not display any particular brightening at the footpoints of the emerging bipole in k2v and k2r .
one of the footpoints is bright in k3 , but the picture is quite different from the observations .
we speculate that the difference has to do with the width of the sufi filter , which samples also considerable portions of the inner wings of the mgiik line .
these are formed mainly in the upper photosphere , where kilo - gauss magnetic features appear as bright point - like structures .
the mg images also display fibrils emanating from plage regions .
the presence of such fibrils had been predicted by @xcite .
not unexpectedly , the density of fibrils increases very strongly from k2v and k2r to k3 .
the density of fibrils in our images is closer to that in k2v and k2r , although only a very qualitative comparison can be carried out , since we find that the density of fibrils depends very strongly on the magnetic activity level ( i.e. on how strong the plage is ) .
this is in line with the behaviour of caiih . also , possibly the fibril density is larger closer to the limb ( this will have to be studied in future investigations ) . at disk center
the fibrils seen in mgii are more pronounced than in caiih ( see , e.g. * ? ? ?
* for a study of the fibrils seen in caiih ) .
features in mgiik are more diffuse than in caiih .
e.g. , the bps have an average size of 0@xmath1755 , which is significantly larger than in caii @xcite .
we expect the following possible reasons for this : because a ) mgii is formed higher in the chromosphere than caii @xcite or b ) because of the much longer exposure time of mgii images , which allows the features to evolve during the exposure , or c ) due to residual jitter not completely compensated by the image stabilization system
. a study of these causes as well as a more detailed and quantitative comparison between mgii and caii will be the topic of a separate paper ( danilovic et al . in preparation ) . in summary , a very first analysis of some of the images recorded by sunrise during its 2013 flight
has shown that observations of the mgiik line with a @xmath1 interference filter provide a qualitatively similar picture as obtained from the narrowest interference filter observations of the caiih line core .
however , there are considerable quantitative differences , partly due to the different formation of the lines , partly due to observing constraints
. it will be interesting to continue such an analysis using data from both sunrise and the iris mission .
the german contribution to the sunrise reflight has been funded by the max planck foundation , the innovations fund of the president of the max planck society ( mpg ) and private donors , which is gratefully acknowledged .
the spanish contribution has been funded by the spanish micinn under the project aya2012 - 39636-c06 .
the hao contribution has been partly funded through nasa grant number nnx13ae95 g .
this work has been partly supported by the wcu grant ( no .
r31 - 10016 ) funded by the korean ministry of education , science & technology . | we present the first high - resolution solar images in the mgiik @xmath0 line .
the images , taken through a @xmath1 broad interference filter , were obtained during the second science flight of sunrise in june 2013 by the sufi instrument .
the mgiik images display structures that look qualitatively very similar to images taken in the core of caiih .
the mgii images exhibit reversed granulation ( or shock waves ) in the internetwork regions of the quiet sun , at intensity contrasts that are similar to those found in caiih .
very prominent in mgii are bright points , both in the quiet sun and in plage regions , particularly near disk center .
these are much brighter than at other wavelengths sampled at similar resolution .
furthermore , mgiik images also show fibril structures associated with plage regions .
again , the fibrils are similar to those seen in caiih images , but tend to be more pronounced , particularly in weak plage . |
solar flares exhibit a multitude of observed dynamic phenomena arising from magnetic energy release ( e.g. , * ? ? ?
* ; * ? ? ?
these phenomena constitute a set of constraints that must be naturally reproduced by any physical model that aims at being generic . in the case of eruptive flares , many observed phenomena have been interpreted in the framework of the cshkp flare model @xcite both in its 2d version and in its realizations in 3d @xcite .
s - shaped x - ray or euv loops seen within so - called sigmoids prior to eruptions ( e.g. * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) as well as thick coronal loop - like structures erupting from the loci of sigmoids ( e.g. * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) have both been interpreted as typical signatures of a flux rope that is formed and eventually erupts .
observations of slipping and hot ( typically 10mk ) flare loops with their footpoints moving along j - shaped euv ribbons have been reported @xcite , and narrow photospheric current ribbons have been measured along the latter @xcite .
these observations are consistent with flux - rope related quasi - separatrix layers ( qsls ) , and with qsl reconnection @xcite .
inflows of nearly vertical loops towards each other , below expanding cmes have been long - since reported .
these motions have been interpreted as motions towards the vertical reconnecting current sheet that is formed below the expanding flux rope ( e.g. * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) . @xcite and
@xcite have shown that the disappearance of the inflowing loops is followed by the appearance of hot flare loops , suggesting that these loop motions are reconnection - related inflows .
these authors have also noticed that the inflows tend to accelerate towards the reconnection site with inflow velocities that range from less than 10 kms@xmath0to few hundreds of kms@xmath0@xcite . in the last decade
several observations of contractions of closed coronal loops located at the periphery of active regions towards the flare and/or eruption site have also been reported ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
these contracting loops are seen during all possible flare phases ; early , impulsive , and gradual .
firstly , the most detailed observational analysis have highlighted that strong contractions ( with velocities of several tens or up to a hundred kms@xmath0 ) occur in the impulsive phase and are usually followed by coronal dimming and loop oscillations .
secondly , it has been shown that these contractions are either preceded by , or simultaneous with the beginning of the eruption ( e.g. , * ? ? ?
* ; * ? ? ?
@xcite have shown that the strong coronal loop contractions occur only after the onset of the fast eruption .
thirdly , @xcite and @xcite have shown that the onset of the contraction depends on the location of the loops with respect to the flare : loops located progressively further away from the flare contract later .
finally , @xcite presented a common model for loop contraction and accompanying oscillations .
these coronal inflows and contractions have been interpreted as an observational confirmation of the conjecture proposed by @xcite , that is , during a transient ( in a low plasma-@xmath1 environment and with negligible gravity ) , `` the coronal field lines must contract in such a way as to reduce @xmath2 . ''
this conjecture is based on the premise that a decrease of @xmath3 must be accompanied by a decrease of @xmath4 . in the case of an eruptive flare
the decrease of volume must be stronger than for a confined flare of the same energy in order to compensate the eruption - related increase in @xmath4 . in this picture
, the decreasing volume should be observable as the coronal volume contracting ( imploding ) towards the flare site . to date , this conjecture is the only attempt to explain the observations of contracting loops .
the aim of this paper is to show that contracting and inflowing loops naturally exist in the 3d mhd models of eruptive flares , and to investigate which magneto - hydrodynamics processes generate the observed loop s behavior . to this purpose
we analyze the relation between the magnetic forces and fluid dynamics effects in a zero-@xmath1 three - dimensional mhd simulation of a torus unstable erupting flux rope .
the plan of our paper is as follows . in the next section
we outline the evolution of different sets of field lines present in the numerical simulation .
section [ sec : t - vortex ] is devoted to the analysis of the 3d distribution of the flows that are generated during the outward propagation of the magnetic flux rope , while in section [ sec : t - inflow ] we describe the flows towards the current sheet . finally , in section [ sect : discussion ] we discuss our findings and their relation with the implosion conjecture .
the dynamics of the magnetic field during the occurrence of erupting flares is modeled using the ohm - mpi code @xcite . in particular , for this paper we analyze the later stages , i.e. , from the onset of the torus instability onward , of the simulation labeled as ` run d2 ' in @xcite . in this simulation
a magnetic flux rope is formed through flux cancellation at the polarity inversion line ( pil ) of a previously sheared magnetic arcade .
the details of the numerical setup have been extensively discussed in section 2 and section 4.5 of @xcite .
however , we retain the important information that the flux rope becomes torus unstable at around @xmath5 , and undergoes a full eruption . during this phase
the pre - eruptive photospheric drivers are reset to zero , i.e. , the system is allowed to evolve only under the effect of the lorentz forces already present in the system .
the simulation ends at @xmath6 . as already discussed in @xcite the simulation
is performed on a non - uniform grid that expands starting from the center of the domain .
while the flux rope moves upwards and sideways also the flare current sheet eventually reaches a location in the domain where the grid resolution is not sufficient to properly resolve the local gradients , so a numerical instability sets in eventually halting the simulation .
while this has no impact on the simulation apart from the last alfvn time , it prevents to follow the longer - term development of the simulated eruption .
the evolution of the coronal magnetic field surrounding the flux rope as driven by the eruption constitutes the subject of this paper .
this dynamics is only related to the presence of moving current - carrying magnetic fields and it is , in principle , independent from the exact trigger of the eruption . as a consequence of the outward motion of the magnetic flux rope ( magenta field lines in figure [ fig : evol ] ) ,
the top part of the magnetic envelope that surrounds the flux rope expands upward and outward in all the directions , while the bottom part of the legs are seen to converge towards the current sheet that is formed below the erupting flux rope and above the pil ( see figure [ fig : evol ] , top - panels and accompanying movie 1 ) .
this inflow of the legs of the overlying field towards the current sheet is more visible following the green field lines in figures [ fig : evol](a ) and [ fig : evol](b ) , i.e. , between @xmath7 $ ] , and the gray field lines in figures [ fig : evol](b ) and [ fig : evol](c ) , i.e. , between @xmath8 $ ] , when more and more field lines are advected towards the current sheet .
figure [ fig : evol ] also shows the change of connectivity of the green field lines . while the flux rope propagates upward overlying magnetic field lines are advected towards the current sheet , where they eventually reconnect . at time
@xmath6 , all the green field lines that originally were part of the overlying magnetic field , have now reconnected at the current sheet and close below it , displaying the typical cusp shaped structure often observed in post flare loops ( e.g. , * ? ? ?
* ; * ? ? ?
therefore , the simulation predicts that when the eruption is seen from a plane that is ( almost ) perpendicular to the strongest portion of the pil , the side of the magnetic envelope , which surrounds the flux rope , initially moves apart to let the erupting flux rope pass , but eventually the legs of the envelope are seen to move back toward the current sheet that is formed below the erupting field .
the evolution of the system as seen from the side is shown in figure [ fig : evol ] ( bottom - panels ) and accompanying movie 1 .
the blue / cyan magnetic field lines highlight the part of the external background magnetic field that is rooted on the weaker peripheral magnetic field of the active region .
as soon as the flux rope begins to expands , the cyan / blue field lines start to move as well .
these magnetic field lines are seen to undergo an expansion and a decrease in altitude before contracting towards the flanks of the flux rope ( the flanks are defined as the external sides of the flux rope s legs ) .
the analysis of figure [ fig : evol ] ( bottom - panels ) and movie 1 shows that while the cyan / blue field lines all display a similar behavior , the importance of the different kind of motions is different for different field lines .
in particular : field line 1 ( fl 1 ) shows a minor expansion and rotation before contracting ; fl 2 shows a more pronounced rotation and expansion and starts to contract only towards the end of the simulation ; fl 3 shows a contraction that is much more important than its expansion .
an analysis of movie 1 shows that fl 3 is seen to contract while fl 1 and fl 2 are still expanding .
in particular , fl 3 is the first that starts to contract , followed by fl 1 while fl 2 is still expanding and finally fl 2 starts to contract as well .
an analysis of the cyan field lines reveals a similar behavior .
in particular , the neighboring fl 4 and fl 5 both show an initial expansion . however , after about @xmath9 fl 5 inverts its behavior and starts to contract , while fl 4 continues to expand .
hereafter , we investigate the dynamic of the system to determine the physical mechanism that is responsible for the observed inflow and expansion / contraction motions .
in this section we describe the generation and propagation of vortex flows during the eruption of the magnetic flux rope .
analysis of the plasma flow field around the selected moving field lines ( as described in section [ sec : t - loops ] ) reveals the presence of rotating flows in the form of two large - scale vortices ( see arrows in figure [ fig:3d - rot_u ] and accompanying movie 2 ) .
these vortex - flows are located above the pil and at each flank of the flux rope .
the plasma velocities inside them correspond to a fraction of the upward - directed velocity around the central part of the erupting flux rope .
each of these two large - scale vortices is associated with an extended source of vorticity located at its center , that is most visible when only its component orthogonal to the flow field is plotted ( see the semi - transparent gray surface in figure [ fig:3d - rot_u ] ) .
movie 2 shows that the general circulation of these flows is established very early in the eruption .
this can be seen in the bottom - right part of the domain ( around the dark blue loops ) , in the time - interval [ @xmath10 . there
, the velocity arrows progressively change direction , from an initial overall expansion towards a vortex - like circulation , firstly close to the flux rope and later away from it .
this evolution and its duration roughly correspond to the transit of an alfvn wavefront from the location of the flux rope ( around @xmath11 ) towards the boundaries of the numerical domain ( at @xmath12 and @xmath13 ) .
the same behavior can be seen on the other side of the flux rope ( around the cyan loops ) , but there it is not as straightforward , primarily because the deflection of the erupting flux rope in this direction alters the velocity pattern ( see * ? ? ? * for more detail on the flux rope s deflection ) . at later times , figure [ fig:3d - rot_u ] at @xmath14 and movie 2 show that the vortices extend in size both horizontally and vertically .
firstly , the associated flow pattern eventually fills the whole domain , including up to altitudes far above the apex of the erupting flux rope .
secondly , the rotating velocities increase in magnitude , co - temporally with the accelerating flux rope , which in our model results from the development of the torus instability .
thirdly , the vorticity concentrations move away from the erupting flux rope , and they rise in altitude .
finally , as the boundary effect of the photosphere diminishes with time , the flow pattern around each vorticity concentration becomes more and more circular with time . in spite of the associated smooth rotational pattern , these quickly - established and slowly - strengthening vortex flows only lead to an incomplete revolution of plasma and loops around their centers .
the analysis of the modeled ( dark blue and cyan ) loop dynamics shows that their rotation is not greater than @xmath15 during the time - interval [ @xmath16 . also , the field line motion along the rotational flow is continuous from the beginning of the eruption up to the end of the calculation .
so , the expansion and contraction of the field lines is just a manifestation of their location within the vortex flow at a given time .
finally , there are two other concentrations of vorticity in the domain ( figure [ fig:3d - rot_u ] ) .
one region is located below the erupting flux rope associated with the flare current sheet , due to the reconnection - driven deflection of the flows around it ( see section [ sec : t - inflow ] ) .
another elongated region is located close to the photosphere above the pil , due to a boundary layer that is created by the compression of the vortex flows by the asymmetric eruption .
the sudden formation of the vortex - flows and the continuity of these flows are reminiscent of the same behavior as observed for the generation of vortex rings from a vortex cannon , or during a thermal explosion , and as observed for the generation of vortex pairs around a fluid element that moves within a non - moving environment as modeled by @xcite .
this surprising phenomenological similarity may be challenged a - priori for the four coupled reasons .
first , the modeled system is zero-@xmath1 and therefore does not involve any plasma pressure .
second , the magnetic and flow field geometries involve complex couplings between the magnetic pressure force @xmath17 , the tension force @xmath18 , and the non - linear term @xmath19 in the momentum equation that governs the acceleration of the frozen - in plasma .
third , the magnetic forces could resist the vortex flows .
finally , the vortices strengthen in time , which does not occur in vortex - cannon experiments where the vortex are readily generated by the sudden outflows .
these arguments can actually be subjected to an analysis of the forces at work in the simulation . such an analysis is provided in the following subsections .
@xcite have shown that in 2.5d an initially - confined low mach number velocity pulse results in the generation of two vortices on its sides ( see their figures .
9.7 - 9.8 in pages 367 - 368 ) .
these authors have shown that two pressure wave fronts are suddenly generated and propagates into the domain .
the important finding is that the boundary of the region filled with the vortices coincides with the pressure wave front at all times during the propagation of this wavefront .
they concluded that this is an evidence that vortices are indeed accelerated by the advancing pressure wave fronts .
the same authors ( r. grappin , private communication ) have shown that the initial flow can actually be decomposed into a solenoidal flow ( @xmath20 ) and a compressible flow .
the solenoidal flow has two vortex flows centered on two @xmath21 concentrations aside of the moving fluid .
the compressible flow has two components related with a @xmath22 sink located ahead of the moving fluid element and a @xmath23 source behind it .
the compressible component of the flow is evacuated at the characteristic ( sonic or magnetosonic ) speed away from the slowly moving fluid element ( hence the aforementioned pressure wave ) , while the solenoidal component is left behind with its two vortices slowly moving the fluid in the whole domain ( this component has no associate wave as it is incompressible : @xmath20 ) .
figure [ fig : force_early ] shows the forces , velocities and @xmath24 contours on one side of the erupting flux rope during the early formation phase of the vortex - flows .
the link with the vortex generation in @xcite is not straightforward since the analyzed systems are different .
there are at least four main differences between our model and the above hydrodynamic case .
first , the length of the wave packet is much larger in our model than in the simple hydrodynamic case ( probably because the flux rope is continuously accelerated , unlike the fluid element in the hydrodynamic case ) .
second , the wave generates a magnetic tension force in our model while it generates a thermal pressure gradient in the hydrodynamic case .
third , the vortex flows are generated in a fluid that is initially at rest in the hydrodynamic case while in our model the forces first need to deflect pre - existing coronal flows that are present because of the preceding flux rope formation phase .
fourth , the erupting magnetic configuration is anchored to the photospheric boundary . a global expansion , @xmath25 , of the plasma is present in front of the erupting flux rope , already before the eruption starts .
this presumably masks part of the @xmath22 compressible component formed in front of the flux rope ( i.e. the moving fluid of belmont et al .
2013 ) , which in our simulation only manifests itself as a sudden diminution of the extended @xmath25 region . however , the @xmath22 compressible component is directly visible on the flanks of the erupting flux rope , where the global expansion is weaker . as in the 2.5 d simulation of @xcite
, we interpret the formation of the vortices by the solenoidal flow component left behind the propagating compressible flow component . indeed , the @xmath26 region ( resp .
the sudden decrease of @xmath25 ) is initially located on the side of the flux rope s flank ( resp . on the top of the flux rope ) , then it travels at the characteristic alfvn speed .
it generates magnetic tension force and accelerates the vortex - flow . in front of the flux rope , the tension force associated with the passage of the compressible wave
is then replaced by the pressure gradient force associated with the expansion of the flux rope ( see section [ vortex - long ] and left columns of figures [ fig : force_early ] and [ fig : force_late ] ) .
another interesting behavior is that the @xmath22 component amplifies a relatively strong and narrow region of magnetic tension , which is located aside of the flux rope s flanks .
this tension force originally points downwards towards the photosphere . while the wave propagates , it extends farther and farther away from the flux rope , and it induces a component of the force that points towards the edge of the flux rope s flank .
careful inspection of figure [ fig : force_early ] shows that this force does not accelerate the flow towards the flux rope . instead
, this force is orthogonal to the velocities of the vortex .
it is therefore arguable that this centripetal tension force balances the centrifugal acceleration of the vortex flow , and thus maintains the vortex integrity , just like thermal pressure radial gradients do in the hydrodynamic case .
the difference is that in the mhd model , the centrifugal effect of the flow does not lead to a pressure deficit .
it rather tends to stretch field lines .
so a curvature - related tension force at each field line apex is generated , until it halts the centrifugal growth of the vortex .
finally , movie 3 shows that the arcades move in the direction of the flow , so their internal magnetic forces are obviously not strong enough to brake the vortex .
the reason is probably that , since the velocity gradients along streamlines are small , and since the magnetic field locally points orthogonally to the plane of the flows , no significant magnetic pressure gradient is generated along the streamlines ( as also noted by * ? ? ?
* in their 2.5d simulations ) .
an exception to this is close to the photospheric plane , which acts as a wall that confines the vortex .
this boundary effect is probably responsible for the asymmetry between the top and the bottom of the vortex at these early times when the vortex has still not moved far above the photosphere . in summary
, the key result of this analysis is that , during the early stages of the eruption , the vortices are formed neither by a stationary magnetic pressure increase ahead of the central part of the flux rope , nor by a stationary magnetic pressure deficit in the wake of the flux rope .
the vortices are rather formed by a fairly broad propagating compressible wave packet that generates magnetic forces in its passage .
the vortices eventually stop expanding because a strong tension in the magnetic arcades works against their centrifugal effect . during the later phase , figure [ fig : force_late ]
shows how the vortices are amplified and how they move , while the erupting flux rope is globally accelerated upwards and starts to expand in all directions .
both motions lead to an increase in magnetic pressure in front of the erupting flux rope .
then , the magnetic pressure gradient eventually replaces the formed magnetic tension force that had been induced by the initial alfvn wave . during this phase
this magnetic pressure gradient accelerates the upper half of the vortex . and
since the over - pressure is continuously strengthened because of the expansion of the erupting flux rope , the vortex flow is continuously amplified . as a response to this strengthening of
the vortex flows , the magnetic tension also increases in the lower half of the vortex , so as to balance the increasing centrifugal force . and the area in which the tension acts gradually moves upward and away from the flux rope , along with the vortex itself .
this vortex displacement away from the flux rope is apparently due to another noticeable difference with the early stage of the vortex formation .
this difference is that a strong pressure gradient eventually develops on the external side of the flux rope , pointing towards the vortex center .
it is most likely that the latter is a reaction of the tendency of the centrifugally growing vortex to push against the flux rope s flank , and therefore to compress it . eventually , this pressure gradient balances the centrifugal acceleration on this side of the vortex , just like the magnetic tension does on its other side , and it eventually pushes the vortex away from the flux rope . in summary ,
the motions of the arcades towards the flux rope are merely due to the early - generated and later - amplified vortices , and is completely independent of the flare reconnection and of the diminishing magnetic pressure behind the rope . )
3d view of the system showing an iso - contour of @xmath27 with a semi - transparent grey surface .
( @xmath28 ) 2d projected view showing the distribution of @xmath24 along the @xmath29-plane passing through @xmath30 .
( @xmath31 ) same view as ( b ) , but showing @xmath32 . in all panels
the arrows highlight the velocity field along the xz - plane passing through @xmath30 .
they are displayed using a random distribution , color - coded with @xmath33 for panel ( @xmath34 ) and in grey for panel ( @xmath35 ) . in all panels the magenta lines outline the erupting flux rope .
( an animation of panel ( a ) , movie 4 , is available in the online version of the article . ) ] + the flux rope eruption tends to generate a vortex ring that is equivalent to the hydrodynamic case .
however , the presence of the magnetic field induces an anisotropy that is not observed in the hydrodynamic case . as a result of this
, the vortex ring is actually divided in two arcs located around the flanks of the erupting flux rope ( see figure [ fig:3d - rot_u ] ) .
fig : rot_u - xzview ] shows that the vortex motions are present along the plane almost parallel to the flux rope axis , but they are not present along the plane almost perpendicular to the axis .
let s first consider the dynamic in the plane almost parallel to the axis of the flux rope ( figure [ fig : rot_u - xzview1 ] ) . in this case
the largest component of the flow velocity is approximately perpendicular to the apex of the cyan / blue magnetic field lines .
thus the flow encounters no significant opposition , neither by the magnetic tension that acts along the axis of curvature of the field line itself , nor by the magnetic pressure gradient because the neighboring field lines are also subjected to a similar flow . as a consequence the magnetic field moves following the flow .
different field lines move in different ways because they are advected by different portions of the flow field .
in particular , while some field lines expand some other field lines located in the returning part of the vortex flows contract because they are advected towards the central part of the active region ( figure [ fig : rot_u - xzview1 ] and movie 2 ) .
this is clearly the case for fl 2 and fl 3 discussed in section [ sec : t - loops ] . on the plane almost perpendicular to the axis of the flux rope ( figure [ fig : rot_u - xzview2 ] ) ,
the behavior is significantly different . in this case
the largest component of the vortex flows would be lying on the plane that contains the field lines and any motion acts against the magnetic tension .
therefore , in this case the magnetic tension force affects the dynamics of the flow , eventually inhibiting it .
the above result confirms and extends the earlier finding of @xcite . in their 2d low-@xmath1 simulation
the vortex flows develops along a plane that is perpendicular to the axis of the flux rope ( because of the 2d geometry this is the only plane where they could develop ) and did not affect significantly the evolution of the magnetic field . in our simulation the low-@xmath1 regime is brought to its extreme , i.e. , @xmath36 , and the flows in a plane perpendicular to the flux rope s axis do not even develop .
we expect that the behavior of the solar corona should be some how in between the two cases . to summarize , vortex flows are present only on a plane aligned with the axis of the erupting flux rope and localized around the flanks of it , while they are absent on a plane perpendicular to the flux rope . as a result ,
magnetic field lines that are located in the top part of the vortices will display expansion and rotation , and contraction if / when they approach the returning part of the vortex - flow .
of course , if a peripheral field line happens to be located in the returning part of the vortices then it will essentially be seen as a contracting one , without displaying the initial expansion .
furthermore , field lines that are highly inclined and thus closer to the photosphere , are ( 1 ) more probable to be located in the return part of the vortex flow ( figure [ fig : evol ] and movie 1 ) and ( 2 ) have a higher density and therefore are the ones preferably visible in observations .
in this section we describe the inflows towards the current sheet that develop below the magnetic flux rope .
the presence of these flows can be deduced from figure [ fig : evol ] ( top - panels ) and accompanying movie 1 that show the motion of the legs of the overlying field towards the region that lies below the expanding flux rope .
an analysis of movie 4 , shows that during the initial phase of the vortex formation , i.e. , between [ @xmath37 , the flows below the erupting flux rope essentially display an upward directed motion .
this is true up to about @xmath9 , when a deflection downwards starts to be visible . at time @xmath38
, the @xmath39 iso - surface appears and grows in size until the end of the simulation ( figure[fig : div_u_3d]a ) .
the snapshot of movie 4 at @xmath38 shows that the flow approaches from both sides of this region , then the flow splits into upward and downward directed flows .
this is true until the end of the simulation , with the extra effect that the flow actually approaches the growing @xmath26 region at faster and faster speeds both as a function of time and as a function of the distance from the @xmath26 layer . a comparison between figure [ fig : div_u_3d](b ) and [ fig : div_u_3d](c ) shows how the @xmath26 region ( below the most right magenta field line ) is spatially correlated with the current sheet that exists below the erupting flux rope .
the center of this current sheet coincides with the thinnest region of @xmath26 . this region becomes broader at the two extremities of the current sheet .
figure [ fig:1d - rot_u ] ( bottom panels ) shows a one - dimensional cut along a direction roughly perpendicular to the planar @xmath26 region .
this figure shows that ( 1 ) the flow accelerates towards the current sheet , ( 2 ) the flow divergence slowly becomes more and more negative while the simulation develops ( during tens of alfvn times ) , and ( 3 ) the flow approaches this region faster and faster while the time passes . in order to test
if the observed flow resembles the flow around a hydrodynamic sink , we consider the flow generated by a two - dimensional , planar source of @xmath24 .
we choose this kind of source because the thickness of the iso - surface of @xmath26 is much smaller than its other two linear dimensions ( figure [ fig : div_u_3d]a and movie 4 at @xmath6 ) .
the theoretical flow profile along a direction orthogonal to center of the planar sink ( i.e. , along @xmath40 ) can be computed by using an analogy with the electric field ( see appendix 1 for the derivation ) and it is : @xmath41 where @xmath42 $ ] is a parameter related to the thickness of the ( almost ) planar sink .
the parameters @xmath43 can be derived from the simulation as follows .
we approximated the profile of @xmath24 along @xmath40 to a triangle of base @xmath44 .
for the purpose of this analysis we assume two different possible basis for this triangle : one where @xmath45 , i.e. , @xmath46 , and one just outside the two maxima of @xmath24 ( see figure [ fig:1d - rot_u](i ) ) , i.e. , @xmath47 . to determine the parameters @xmath48 and @xmath49 , we consider an iso - surface of @xmath24 equal to its average value withing the current sheet , i.e. , @xmath50+\text{max}[\nabla \cdot \mathbf{u}])/2 $ ] .
we find @xmath51 and @xmath52 .
the theoretical model reproduces the simulated flow reasonably well ( see figure [ fig:1d_analyt ] ) , suggesting that the flow is indeed compatible with a sink - flow . to clarify the physical origin of this sink - like flow
, we studied the evolution of the lorentz force and of its decomposition in magnetic pressure gradient and magnetic tension terms across the current sheet .
figure [ fig:1d - rot_u](a - c ) shows that a double - peaked lorentz force directed towards the current sheet itself exists at the edges of the current sheet ( maximum in the current density ) .
this is the result of an imbalance between the magnetic pressure gradient and the magnetic tension terms ( panels ( d - f ) ) .
if the magnetic field at the current sheet is reconnected fast enough the inward - directed magnetic pressure gradient term is maintained . this leads to a quasi - stationary situation where a @xmath26 region is sustained by the magnetic pressure deficit within the current sheet itself .
eventually , a current - sheet collapse occurs in our model .
it halts the simulation .
however , the collapse takes several tens of alfvn time units of the simulation , which correspond to hundreds of alfvn times across the thickness of the modeled current sheet .
this relatively slow evolution of the sink ( i.e. , the @xmath26 layer ) is the property of the system that permits to establish , through the launching of quickly - propagating alfvn waves away from the sink ( as in * ? ? ?
* ) , a large - scale sink - flow that can quasi - statically adjust to the local conditions met within the current sheet at a given time .
so our model implies that the large - scale coronal inflow that brings distant magnetic field lines towards the reconnection site is controlled by the magnitude of the negative flow divergence inside the current sheet which , in our zero-@xmath1 assumption , is directly due to the double - peaked lorentz force that is primarily caused by the local magnetic depression within the sheet .
this interesting result implies that , although ( the flare ) reconnection ( and therefore energy release ) does occur in our model , it is not directly responsible for the inflow . and
it also predicts that finite-@xmath1 effects should influence these inflows .
indeed , the development of a local over - pressure within the current sheet should diminish the magnitude of the local sink , hence should make the large - scale inflows slower than in the zero-@xmath1 case .
an alternative possible explanation for the reported inflows towards the current sheet could be that the large scale pressure drop left behind the erupting flux rope forces the legs of the line - tied envelope towards the current sheet , resulting in the observed inflows .
however , there is essentially no lorentz force outside the current sheet , apart from at late times when the flux rope legs start to move inward .
also the modeled flow is remarkably compatible with the analytic solution around a sink region that has the same characteristics as the one in our simulation .
moreover , the presence of a lorentz force directed towards the current sheet does not depend on the flux rope eruption itself because the same occurs during the relaxation of quasi - force free fields ( e.g. * ? ? ?
then , we argue that it is the sink - flow generated at the current sheet that drives the modeled loops ( green / gray field lines in figure [ fig : evol ] ) towards the reconnection region , rather than the large scale pressure drop left by the erupting flux rope .
in this paper we investigated the hydro - magnetic mechanisms at the origin of recently - observed flare - related active - region loop contractions , and of well - known coronal - inflows towards flare current - sheets .
this study was conducted through the analysis of a three - dimensional line - tied visco - resistive mhd simulation of an eruptive flare , achieved with the ohm - mpi code .
the modeled eruption was triggered by the onset of the torus instability of a weakly - twisted flux rope , which was previously formed by flux cancellation @xcite .
a key property of this simulation is that the corona was treated as a zero-@xmath1 plasma , so that thermal pressure and gravity were neglected and the only forces at work were magnetic pressure and tension .
typical observations as well as a first visual inspection of our model tend to suggest that loop contractions would be consistent with a mere relaxation of previously expanding loops , which would eventually shrink towards the place where the pre - erupting current - carrying field was located before the flare , and where the flare reconnection eventually occurs . in principle , such a relaxation would be natural since our model clearly shows that both the evacuation of the flux rope as well as the reconnection do lead to a magnetic energy density decrease in this region ( see figure [ fig : b_mag ] ) , so the related inward magnetic - pressure gradient may induce a global implosion .
this interpretation was already put forward with observational analyses ( e.g. * ? ? ?
* ; * ? ? ?
* ) building upon a previously proposed conjecture linking magnetic energy and coronal implosions @xcite .
instead , the model reveals that coronal loops are advected by a pair of vortices that develop on both sides of the flux rope legs and that eventually reach an active - region scale in a characteristic alfvn time .
their rotational motions do not imply a volume decrease , and they are unrelated with the magnetic energy decrease that develops behind the erupting flux rope .
the flows are firstly accelerated by a magnetic tension force that develops in the wake of the alfvn wave ( which would be magneto - acoustic in a finite-@xmath1 regime ) that is launched from the flux rope as soon as its starts moving .
this wave evacuates away the compressible component of the flow , leaving behind two vorticity arcs located on the flanks of the flux rope legs .
this behaviour is reminiscent of experimental generation of vortex rings when a fast flow of liquid is injected from a hole into a tank @xcite , of modeled high-@xmath1 vortices that form around flux ropes that move in the solar convection zone @xcite , and more generally of modeled vortex pairs that form around a moving fluid element ( * ? ? ?
* figures 9.7 - 9.8 pages 367 - 368 ) .
some differences exist between our model and these purely hydrodynamical or high-@xmath1 examples , as follows . in our model , the centrifugal acceleration of the flows
is not ( and can not be ) balanced by thermal pressure .
it is rather and mostly ensured by magnetic tension , although a magnetic pressure gradient that points away from the flux rope also contributes on the side of the vortex that is closest to the flux rope . also , it is not a full vortex ring that forms around the eruption , but rather two vorticity arcs .
the reason is that the magnetic field in the legs of the arcades that overlay the flux rope prevents rotational motions to develop in a plane perpendicular to the flux rope s axis with our zero-@xmath1 assumption .
2d finite-@xmath1 simulations still show that vorticity can develop in this direction too , but with seldom field line deformation @xcite .
finally , our modeled vortices are not left to evolve on their own , once formed .
instead they gradually strengthen and move away from each other .
the reason is that the flux rope acceleration and expansion generate new magnetic pressure gradients , which accelerate the vortex flows , and which lead them to push against the flux rope legs as their increasing centrifugal acceleration tends to make them grow .
in addition to the vortices and loop contractions , we analyzed the nature of the coronal inflows towards the flare current - sheet .
we found that those are consistent with a sink flow , perpendicular to the flux rope axis and to the vortex flows .
the relatively good match found between the simulation and an idealized analytical model of a sink flow suggests that the inflows are not due the large - scale depression left behind the flux rope as it erupts away from the active region ( see the large - scale color change from dark to bright orange at low altitudes in figure [ fig : b_mag ] ) , but instead it is due to the very narrow depression that develops within the elongated current sheet ( see white region in figure [ fig : b_mag]b ) .
the origin of this sink is merely due to the change in sign of the vertical component of the magnetic field across the current sheet .
this naturally produces a local minimum in magnetic pressure inside the sheet .
this local depression can cause the fast collapse of current sheets , when neither of the internal thermal pressure or the magnetic pressure from another component are strong enough to halt the inflows driven by this local magnetic - pressure deficit @xcite .
our theoretical results on loop contractions and coronal inflows thus far have only been analyzed in one simulation .
so they will have to be investigated in other ones , to test whether or not they are specific to our physical and numerical settings .
however , as such behavior exist also in the hydrodynamic case , it should in principle be unrelated to the exact topology of the pre - erupting current - carrying field , nor with the eruption trigger mechanism .
so it is arguable that they should be present in general . but
this will have to be checked against other solar eruption models that are neither based on flux ropes nor on the torus instability .
in addition , our results imply several observational predictions that should be tested in the future .
in particular the sink flow interpretation predicts a flow magnitude that asymptotically scales inversely proportional to the square of the distance from the current sheet ( albeit for the possible role of magnetic forces from distant loops that could brake them ) .
also since these sink flows are a property of current sheets , the model predicts that they should develop both in confined and eruptive flares . on the other side ,
the vortex flows are caused by the flux rope eruption , so according to the model they should not be observed in confined flares , but only in eruptive flares and filament eruptions .
however , since these vortices are merely initiated by the early displacement of current - carrying fields , they should also occur in failed eruptions .
finally , vortices are mere rotational motions .
so they are neither associated with a volume decrease nor with a global coronal expansion followed by an implosion sequence .
so both expanding and contracting loops should be observed co - temporally .
in section [ sec : t - curr ] we discussed how the flow of the simulation could be interpreted as the flow due to the presence of an hydrodynamic sink . to show this
, we use the analogy between a sink flow and an electric field .
let us consider the electric field in a point @xmath53 at a distance @xmath40 from the center of a 2d planar distribution of charge located on the @xmath54-plane such that its surface charge density is @xmath55 within the region @xmath56 $ ] , and zero outside .
the @xmath57component of the electric field at the point @xmath58 is then given by @xcite : @xmath59 if the charge distribution is not actually planar , but has a given thickness @xmath60 $ ] we have : @xmath61 where @xmath62 is an integration variable along the direction perpendicular to the charged plane . in our simulation at @xmath6 and along the selected line
, @xmath24 reaches a minimum value of @xmath63 and the thickness of the region where @xmath45 is @xmath64 . however , as shown in figure [ fig:1d - rot_u](i ) , @xmath24 has two local maxima very close to the region of negative divergence .
the distance between the two closest grid points such that the two maxima are included within them is @xmath65 . if we assume a triangular profile for the distribution of @xmath24 along @xmath40 , the integral in equation ( [ eq_ef2 ] ) can be evaluated as the area of a triangular surface of base @xmath66 and height min@xmath67 $ ] ( neglecting the positive @xmath24 of the local maxima ) . with these assumptions equation ( [ eq_ef2 ] ) becomes : @xmath68 where @xmath69 $ ] .
moreover , the system undergoes an asymmetric expansion and the actual flow is the combined result of the expansion of the flux rope and flow towards the current sheet . to account for this the theoretical profile is shift by a value that is half of the net expansion flow , i.e. , @xmath70
this is the analytically velocity profile displayed in figure [ fig:1d_analyt ] .
the authors wish to thank the anonymous referee for her / his constructive comments that led to the improvement of the manuscript .
z. and g. a. thank s. antiochos , r. grappin and j. klimchuk for the useful discussions .
the work of f. p. z. was funded by a contract from the axa research fund .
f. p. z. is a fonds wetenschappelijk onderzoek ( fwo ) research fellow .
j. d. acknowledges the grant 17 - 16447s of the grant agency of the czech republic , the institutional support rvo:67985815 of the czech academy of sciences , and the hospitality provided by the observatoire de paris during his 1-month stay . s.a.g . acknowledges the financial support of the dim acav and rgion ile de france .
this work was granted access to the hpc resources of mesopsl financed by the rgion ile de france and the project equip@meso ( reference anr-10-eqpx-29 - 01 ) of the programme investissements d avenir supervised by the agence nationale pour la recherche . | eruptive flares are sudden releases of magnetic energy that involve many phenomena , several of which can be explained by the standard 2d flare model and its realizations in three - dimensions .
we analyze a three - dimensional magnetohydrodynamics simulation in the framework of this model that naturally explains the contraction of coronal loops in the proximity of the flare sites , as well as the inflow towards the region above the cusp - shaped loops .
we find that two vorticity arcs located along the flanks of the erupting magnetic flux rope are generated as soon as the eruption begins .
the magnetic arcades above the flux - rope legs are then subjected to expansion , rotation or contraction depending on which part of the vortex - flow advects them .
in addition to the vortices , an inward - directed magnetic pressure gradient exists in the current sheet below the magnetic flux rope .
it results in the formation of a sink that is maintained by reconnection .
we conclude that coronal loop apparent implosions observed during eruptive flares are the result of hydro - magnetic effects related to the generation of vortex- and sink - flows when a flux rope moves in a magnetized environment . |
the structure of a group which acts freely and cocompactly on a tree is well understood .
any such a group is a finitely generated free group . by way of contrast
, a group which acts in a similar manner on a product of trees can have remarkably subtle properties .
for example , m. burger and s. mozes @xcite have proved rigidity and arithmeticity results analogous to the theorems of margulis for lattices in semisimple lie groups .
this article will consider a discrete subgroup @xmath3 of @xmath4 where @xmath0 , @xmath1 are homogeneous trees of finite degree .
in addition , we require that @xmath3 is torsion free and acts freely and transitively on the vertex set of @xmath5 .
for simplicity , refer to such a group as a bm group .
bm groups were used in @xcite to exhibit the first known examples of finitely presented , torsion free , simple groups . a product of two trees may be regarded as the 1-skeleton of an affine building whose 2-cells are euclidean squares .
a bm group @xmath3 acting freely and transitively on the vertex set of @xmath5 defines a 2-dimensional tiling system .
associated to this tiling system there is a @xmath9-algebra @xmath10 , which is called a rank-2 cuntz - krieger algebra in @xcite .
this algebra is isomorphic to a crossed product @xmath9-algebra @xmath11 arising from a boundary action of @xmath3 .
it follows from the results of @xcite that @xmath11 is purely infinite , simple , unital and nuclear , and is therefore itself classified by its k - theory .
this provides the motivation for us to examine the k - theory of these examples in some detail . in theorem [ maink ]
we obtain an explicit expression for the k - theory of @xmath11 analogous to that of @xcite for algebras associated with @xmath12 buildings .
in particular @xmath7 for this algebra . in proposition [ orderof1 ] , the class of the identity in @xmath13 is shown to be a torsion element . in section [ exs
] , these issues are examined for several explicit groups . in section [ bm2by2 ] a complete list is given of all bm groups acting on @xmath5 , where @xmath0 , @xmath1 are homogeneous trees of degree four .
the abelianizations and k - groups are also computed .
after this article was submitted , we became aware of the work of diego rattaggi @xcite , which undertakes a detailed analysis of bm groups , including extensive computations with explicit presentations .
we are grateful to him for several helpful comments on this article .
given a homogeneous tree @xmath14 , there is a type map @xmath15 defined on the vertices of @xmath14 and taking values in @xmath16 . to see this , fix a vertex @xmath17 and define @xmath18 where @xmath19 denotes the usual graph distance between vertices of the tree .
the type map partitions the set of vertices into two classes so that two vertices are in the same class if and only if the distance between them is even .
thus the type map is independent of @xmath20 , up to addition of @xmath21 .
since any automorphism of the tree preserves distances between vertices this observation proves [ lem1 ] for each automorphism @xmath22 of @xmath14 there exists @xmath23 such that , for every vertex @xmath24 , @xmath25 suppose that @xmath26 is the @xmath27 dimensional cell complex associated with a product @xmath28 of homogeneous trees .
then @xmath26 is an affine building of type @xmath29 in a natural way @xcite .
write @xmath30 for a generic vertex of @xmath26 .
there is a type map @xmath15 on the vertices of @xmath26 where @xmath31 we say that an automorphism @xmath22 of @xmath26 is * type - rotating * if there exists @xmath32 such that , for each vertex @xmath24 , @xmath33 the chambers of @xmath26 are geometric squares and each chamber has exactly one vertex of each type .
we denote by @xmath34 the group of type rotating automorphisms of @xmath26 .
an automorphism @xmath22 of @xmath26 is type - rotating if and only if it is a cartesian product of automorphisms of the two trees .
if we have @xmath35 for some type rotating automorphisms @xmath36 of @xmath37 then it follows from lemma [ lem1 ] that @xmath22 is type rotating .
conversely suppose that @xmath22 is type rotating .
let @xmath38 and @xmath39 be neighbouring vertices in @xmath26 .
then @xmath40 and @xmath41 are neighbouring vertices in @xmath26 and the type - rotating assumption on @xmath22 means that @xmath42 . since neighbouring vertices in @xmath43 have distinct types we must have @xmath44 . by induction on @xmath45 , we see that the first coordinate of @xmath46 is independent of @xmath47 .
similarly , the second coordinate of @xmath46 is independent of @xmath48 . thus there
exist maps @xmath49 of @xmath43 and @xmath50 of @xmath51 such that @xmath52 . since @xmath22 is an automorphism of @xmath26 it follows that each @xmath36 is an automorphism of @xmath37 .
thus @xmath53 for some automorphisms @xmath36 of @xmath37 .
@xmath54 .
an * apartment * in @xmath26 is a subcomplex isomorphic to the plane tessellated by squares .
see @xcite for some comments on this and alternative ways of looking at apartments .
denote by @xmath55 the vertex set of @xmath26 .
any two vertices @xmath56 belong to a common apartment .
the convex hull , in the sense of buildings , between two vertices @xmath57 and @xmath24 is the subset of an apartment containing @xmath57 and @xmath24 depicted in figure [ convex hull ] .
[ convex hull ] units < 1 cm , 1 cm > x from -6 to 6 , y from -1 to 2.2 from -3 -1 to 3 -1 from -3 0 to 3 0 from -3 1 to 3 1 from -3 2 to 3 2 from -3 -1 to -3 2 from -2 -1 to -2 2 from -1 -1 to -1 2 from 0 -1 to 0 2 from 1 -1 to 1 2 from 2 -1 to 2 2 from 3 -1 to 3 2 [ r ] at -3.2 -1 at -3 -1 [ l ] at 3.2 2 at 3 2 [ t ] at 0 -1.2 [ l ] at 3.2 0.5 the convex hull of @xmath57 and @xmath24 is contained in every apartment of @xmath26 which contains @xmath57 and @xmath24 .
define the * distance * , @xmath19 , between @xmath57 and @xmath24 to be the graph theoretic distance on the one - skeleton of @xmath26 .
any path from @xmath57 to @xmath24 of length @xmath19 lies in their convex hull , and the union of the vertices in such paths is exactly the set of vertices in the convex hull .
we define the * shape * @xmath58 of the ordered pair of vertices @xmath59 to be the pair @xmath60 as indicated in figure [ convex hull ] . note that @xmath61 .
the components of @xmath58 indicate the relative contributions to @xmath19 from the two factors .
if @xmath62 and @xmath63 are vertices of @xmath26 , the * shape * from @xmath24 to @xmath64 is @xmath65 where @xmath66 denotes the usual graph - theoretic distance on a tree .
an edge in @xmath26 connects the vertices @xmath24 and @xmath64 if @xmath67 or @xmath68 .
[ shape decomposition for vertices ] suppose @xmath69 and @xmath70 for vertices @xmath71 .
then there is a unique vertex @xmath72 such that @xmath73 such a @xmath72 satisfies @xmath74 so it must lie in the convex hull of @xmath57 and @xmath64 . inside the convex hull existence and uniqueness of @xmath24
are clear .
it is a direct consequence of the definitions that every type - rotating automorphism @xmath75 preserves shape in the sense that @xmath76 for all @xmath56 .
suppose that @xmath77 acts freely and transitively on the vertex set @xmath78 .
fix any vertex @xmath79 and let @xmath80 the * cayley graph * of @xmath8 constructed via right multiplication with respect to the set @xmath81 has @xmath8 itself as its vertex set and has @xmath82 as its edge set .
there is a natural action of @xmath8 on its cayley graph via left multiplication . using the convention that an undirected edge between vertices @xmath57 and @xmath24 in a graph represents the pair of directed edges @xmath83 and @xmath84 , it is immediate that the @xmath8-map @xmath85 from @xmath8 to @xmath26 is an isomorphism between the cayley graph of @xmath8 and the one - skeleton of @xmath26 . in this way we identify @xmath8 with the vertex set @xmath55 of the @xmath26 .
connectivity of the @xmath26 implies that @xmath81 is a generating set for @xmath8 .
it is traditional to label the directed edge @xmath86 with the generator @xmath87 . more generally , to the pair @xmath88 we assign the label @xmath89 .
equivalently , to the pair @xmath90 we assign the label @xmath91 .
suppose this label is written as a product of generators ; @xmath92 .
then there is a path @xmath93 from @xmath94 to @xmath95 whose successive edges are labelled @xmath96 .
the left translate of @xmath90 by @xmath97 is @xmath98 and also carries the label @xmath66 .
conversely , any pair @xmath99 which carries the label @xmath66 is the left translate by @xmath100 of @xmath90 .
thus two pairs carry the same label if and only if one is the left translate of the other .
we define a shape function on @xmath8 by @xmath101 for @xmath97 . the pair @xmath102 has label @xmath103 and its shape , defined via the identification of the cayley graph and the one - skeleton of @xmath26 , is @xmath104 a different choice of @xmath20 leads to a shape function on @xmath8 which differs from the first by an inner automorphism of @xmath8 .
@xcite suppose that the group @xmath77 acts freely and transitively on @xmath55 .
then @xmath8 is called an * @xmath105 group*. consider an @xmath106 group @xmath8 . fix a vertex @xmath107 and suppose that @xmath108
recall that @xmath109 consider the generating set @xmath110 of @xmath8 .
let @xmath111 [ normal form ] each element @xmath112 has a unique reduced expression of the form @xmath113 and of the form @xmath114 for some @xmath115 and @xmath116 .
moreover @xmath117 .
this follows immediately from lemma [ shape decomposition for vertices ] . in ( * ? ? ?
* section 1 ) , m. burger and s. mozes constructed a class of groups which act freely and transitively on the vertices of a product of trees .
it is convenient to refer to these groups as bm groups .
our aim now is to show that the class of bm groups coincides with the class of torsion free @xmath29 groups .
[ def bm group ] ( * ? ? ?
* section 1 ) a * bm group * is defined as follows .
choose sets @xmath118 , with @xmath119 where @xmath120 are even integers .
choose fixed point free involutions @xmath121 , @xmath122 on @xmath118 respectively and a subset @xmath123 with the following properties . * if @xmath124 then each of @xmath125 , and @xmath126 belong to @xmath127 . * all four @xmath128-tuples in ( i ) are distinct .
equivalently @xmath129 for @xmath130 .
* each of the four projections of @xmath127 to a subproduct of the form @xmath131 or @xmath132 is bijective .
equivalently at least one such projection is bijective .
a * bm square * is defined to be a set of four distinct tuples as in ( ii ) , that is four element subsets of @xmath133 of the form @xmath134 if @xmath124 then write @xmath135 . [ geometric square ] units < 1 cm , 1 cm > x from -6 to 6 , y from -1 to 1.2 from -1 -1 to 1 -1 from -1 1 to 1 1 from -1 -1 to -1 1 from 1 -1 to 1 1 [ r ] at -1.2 0 [ l ] at 1.2 0 [ t ] at 0 -1.2 [ b ] at 0 1.2 [ .3,.67 ] from 0.8 -1 to 1 -1 [ .3,.67 ] from 0.8 1 to 1 1 [ .3,.67 ] from -1 0.8 to -1 1 [ .3,.67 ] from 1 0.8 to 1 1 a bm group @xmath8 has presentation @xmath136 in a subsequent article @xcite , a set of objects @xmath137 satisfying the above conditions is called a * vh - datum*. [ bma ] a group @xmath8 is a bm group if and only if it is a torsion free @xmath29 group .
suppose that @xmath8 is a torsion free @xmath29 group with generating set @xmath138 , where @xmath139 and @xmath140 are described by equations ( [ ab ] ) .
the map @xmath141 on @xmath81 is fixed point free since @xmath8 is torsion free . define @xmath127 to be the set of @xmath128-tuples @xmath142 such that @xmath143 . condition
( i ) for a bm group is clearly satisfied . to verify condition ( ii )
note that if @xmath144 then @xmath145 , contradicting the assumption that @xmath8 is torsion free . condition ( iii ) follows immediately from lemma [ normal form ] .
we now prove the converse . given a bm group @xmath8 we may construct as in @xcite a cell complex @xmath146
whose fundamental group is @xmath8 .
the complex @xmath146 has one vertex @xmath24 and the cells are geometric squares as in figure [ geometric square ]
. there are @xmath147 such cells whose four edges form a bouquet of four loops meeting at @xmath24 .
the boundary labels of the directed edges are elements of @xmath148 and edges with the same label are identified in the complex .
definition [ def bm group](ii ) says that none of the cells of @xmath146 is a projective plane . by definition , the * link * of the vertex @xmath24 in @xmath146
is the graph @xmath149 whose vertices are in 1 - 1 correspondence with the half - edges incident with @xmath24 and whose edges are in 1 - 1 correspondence with the corners incident at @xmath24 . in our setup
the link @xmath149 is a complete bipartite graph with vertex set @xmath148 and an edge between each element of @xmath139 and each element of @xmath140 .
intuitively , completeness of this bipartite graph means that there are no `` missing corners '' .
it follows from ( * ? ? ?
* theorem 10.2 ) that the universal cover of @xmath146 is a product of homogeneous trees @xmath150 , where @xmath0 has valency @xmath151 and @xmath1 has valency @xmath152 .
( in the terminology of @xcite , @xmath146 is said to be a complete @xmath153 complex . )
elements of @xmath8 correspond to edge paths in @xmath146 . by (
* lemma 4.3 ) each @xmath154 can be expressed uniquely in each of the normal forms @xmath155 , for some @xmath115 and @xmath116 .
the 1-skeleton of the universal covering space @xmath28 may therefore be identified with the cayley graph of @xmath8 with respect to the generating set @xmath148 .
thus @xmath156 , and @xmath157 .
let @xmath8 be a bm group with presentation ( [ presentation ] ) .
in view of the preceding discussion we need only show that @xmath8 is torsion free . the argument for this
is well known ( * ? ? ?
* vi.5 , p.161 , theorem ) , and it was shown to us by donald cartwright , in the context of @xmath12 groups .
suppose that @xmath158 with @xmath159 for some integer @xmath160 .
let @xmath161 denote the cyclic group generated by @xmath162 . fix a vertex @xmath20 of the @xmath163 skeleton of @xmath150 .
then @xmath164 is the set of vertices of @xmath26 .
now the set @xmath165 is a bounded @xmath161-stable subset of @xmath26 .
since the complete metric space @xmath26 satisfies the negative curvature condition of ( * ? ? ?
* vi.3b ) , it follows from the bruhat - tits fixed point theorem that there is a @xmath161-fixed point @xmath166 .
since the action of @xmath8 is free on the vertices of @xmath26 , @xmath167 can not be a vertex .
thus @xmath167 lies in the interior of an edge @xmath168 or a square @xmath169 in @xmath26 , and either @xmath168 or @xmath169 is invariant under @xmath161 . by considering @xmath170 for suitable @xmath154
, we may suppose that one vertex of @xmath168 [ respectively @xmath169 ] is @xmath20 .
* @xmath168 is invariant under @xmath161 .
it follows that @xmath168 has endpoints @xmath20,@xmath172 and that @xmath173 satisfies @xmath174 , contradicting the definition of a bm group .
we therefore reduce to * case 2 . * @xmath169 is invariant under @xmath161 , where @xmath169 is the square illustrated in figure [ square s ]
. then @xmath175 ( since the action is free ) and so @xmath172 is one of the other three vertices of @xmath169 . thus @xmath176 or @xmath177 or @xmath178 , where @xmath179,@xmath180 .
if @xmath176 then @xmath181 is a vertex of @xmath169 , which is impossible . similarly @xmath182
thus @xmath178 .
again @xmath183 is a vertex of @xmath169 . according to lemma [ normal form ] ,
the only way this can happen is if @xmath184 .
however this contradicts condition ( ii ) in the definition [ def bm group ] .
this completes the proof of theorem [ bma ] .
[ square s ] units < 1 cm , 1 cm > x from -6 to 6 , y from -1 to 1.2 from -1 -1 to 1 -1 from -1 1 to 1 1 from -1 -1 to -1 1 from 1 -1 to 1 1 [ r ] at -1.2 0 [ l ] at 1.2 0 [ t ] at 0 -1.2 [ b ] at 0 1.2 at 1 1 at 1 -1 at -1 1 at -1 -1 [ l ] at 1.2 -1 [ r ] at -1.2 1 [ r ] at -1.2 -1 [ l ] at 1.2 1 the fact that bm groups are torsion free is an immediate consequence of ( * ? ? ?
* theorem 4.13(2 ) p.201 ) . that much more general result applies to fundamental groups of spaces of non - positive curvature .
they are always torsion free .
identify elements of @xmath8 with vertices of @xmath26 .
the set @xmath127 may be identified with the set of @xmath8-equivalence classes of oriented basepointed squares ( chambers ) in @xmath26 .
we refer to such an equivalence class of squares as a tile .
we now construct a 2-dimensional shift system associated with @xmath3 .
the transition matrices are defined as follows .
if @xmath185 then define horizontal and vertical transition matrices @xmath186 as indicated in figure [ trans ] : that is @xmath187 if @xmath188 and @xmath189 represent the labels of tiles in @xmath26 which lie as shown in figure [ trans ] , and @xmath190 otherwise .
the @xmath191 matrices @xmath186 are nonzero @xmath192-matrices .
[ trans ] units < 1 cm , 1 cm > x from -6 to 6 , y from -1 to 2.2 from -3 0 to -1 0 from -3 1 to -1 1 from -3 0 to -3 1 from -2 0 to -2 1 from -1 0 to -1 1 from 2 0 to 3 0 from 2 1 to 3 1 from 2 2 to 3 2 from 2 0 to 2 2 from 3 0 to 3 2 [ ] at -2 -1 at 2.5 -1 [ ] at -2.5 0.5 at -1.5 0.5 [ ] at 2.5 0.5 at 2.5 1.5 it follows that @xmath193 if and only if @xmath194 and @xmath195 . (
the condition @xmath196 is redundant , because two adjacent sides of a square uniquely determine it . )
see figure [ trans1 ] .
it follows that each row or column of @xmath197 has precisely @xmath198 nonzero entries .
a diagram similar to figure [ trans1 ] applies to vertical transition matrices , with the result that each row or column of @xmath199 has precisely @xmath200 nonzero entries .
[ trans1 ] units < 2 cm , 2 cm > x from -1.5 to 1.5 , y from -0.2 to 1.2 from -1 0 to 1 0 from -1 1 to 1 1 from -1 0 to -1 1 from 0 0 to 0 1 from 1 0 to 1 1 [ ] at -0.5 0.5 at 0.5 0.5 at -0.5 -0.1 at 0.5 -0.1 at -0.5 1.1 at 0.5 1.1 at -0.1 0.5 at 1.1 0.5 at -1.1 0.5 at 0.1 0.5 we now use @xmath127 as an alphabet and @xmath186 as transition matrices to build up two dimensional words as in @xcite . let @xmath201 $ ] denote @xmath202 , where @xmath203 are integers . if @xmath204 , say that @xmath203 if @xmath205 for @xmath206 , and when @xmath203 , let @xmath201 = [ m_1,n_1 ] \times [ m_2,n_2]$ ] . in @xmath207 , let @xmath208 denote the zero vector and let @xmath209 denote the @xmath210 standard unit basis vector . if @xmath211 , let @xmath212 \to { { \mathcal r}};\ m_j(w(l+e_j),w(l ) ) = 1\
\text{whenever}\ l , l+e_j \in [ 0,m ] \}\ ] ] and call the elements of @xmath213 words .
let @xmath214 .
we say that a word @xmath215 has shape @xmath216 , and we identify @xmath217 with @xmath127 in the natural way via the map @xmath218 .
define the initial and final maps @xmath219 and @xmath220 by @xmath221 and @xmath222 . in order to apply the theory of
@xcite we need to show that the matrices @xmath197,@xmath199 satisfy the following conditions .
( h0 ) : : each @xmath223 is a nonzero @xmath192-matrix .
( h1a ) : : @xmath224 .
( h1b ) : : @xmath225 is a @xmath192-matrix .
( h2 ) : : the directed graph with vertices @xmath226 and directed edges @xmath227 whenever @xmath228 for some @xmath229 , is irreducible .
( h3 ) : : for any nonzero @xmath230 , there exists a word @xmath231 which is not _ @xmath167-periodic _ , i.e. there exists @xmath232 so that @xmath233 and @xmath234 are both defined but not equal .
[ m_1m_2 ] the matrices @xmath197 , @xmath199 satisfy conditions ( h0),(h1a ) , ( h1b ) , and ( h3 ) . * ( h0 ) : * by definition @xmath197 and @xmath199 are @xmath192-matrices and they are clearly nonzero . * ( h1a , b ) : * consider the configuration of figure [ w1 ] consisting of chambers lying in some apartment of the building . given the tiles @xmath235 , @xmath236 , and @xmath188 , there is exactly one tile @xmath237 which completes the picture . therefore if @xmath238 then @xmath239
. likewise ,
if @xmath240 , then @xmath241 . conditions ( h1a ) and ( h1b ) follow .
[ w1 ] units < 1 cm , 1 cm > x from -1 to 1 , y from -1 to 1 from -1 -1 to 1 -1 from -1 0 to 1 0 from -1 1 to 1 1 from -1 -1 to -1 1 from 0 -1 to 0 1 from 1 -1 to 1 1 at -0.5 -0.5 at 0.5 -0.5 at -0.5 0.5 at 0.5 0.5 * ( h3 ) : * fix any nonzero @xmath242 .
choose @xmath243 large enough that the rectangle @xmath244 $ ] contains a point @xmath232 and its @xmath167-translate @xmath245 .
we can construct @xmath215 which is not @xmath167-periodic , as follows .
let @xmath246 be chosen arbitrarily .
now for @xmath206 there are at least two choices of @xmath247 such that @xmath248 [ respectively @xmath249 .
thus one can begin to extend the domain of definition of @xmath64 in any one of four directions so that there are at least two choices of @xmath250 , @xmath206 . by induction , one can extend @xmath64 in many ways to an element of @xmath213 , at each step choosing a particular direction for the extension . in order to do this , first choose arbitrarily a shortest path from @xmath232 to @xmath245 , and then extend step by step along the path .
it is important to note that at each step , @xmath64 extends uniquely to be defined on a complete rectangle in @xmath207 , as illustrated in figure [ extension ] . in that figure
, we assume that @xmath64 is defined on the rectangle @xmath251 $ ] , and then define @xmath252 , where @xmath253 . by conditions ( h1a ) and ( h1b ) , there is a unique choice of @xmath254 which is compatible with the values of @xmath255 and @xmath256 .
continue the process inductively until @xmath64 is defined uniquely on the whole rectangle @xmath257 $ ] .
[ extension ] units < 1 cm , 1 cm > x from -6 to 6 , y from -1 to 2.2 from -3 -1 to 2 -1 from -3 1 to 2 1 from -3 -1 to -3 1 from 2 -1 to 2 2 from -3 2 to 2 2 from -3 2 to -3 1 from 1 2 to 1 1 at -0.5 0 [ l ] at 2.2 1 [ r ] at -3.2 -1 [ l ] at 2.2 2 [ t ] at 1 0.8 [ b ] at 1 2.2 at 2 1 at 2 2 at 1 1 at 1 2 at -3 -1 figure [ path ] illustrates how @xmath64 is defined on @xmath258 $ ] ( where @xmath259 ) , by moving along a certain path from @xmath232 to @xmath245 .
the values of @xmath64 on this path are given by a sequence of tiles .
the values of @xmath64 up to a certain point on the path determine the values on a rectangle which contains the corresponding initial segment of the path . at the end of the process
, the values of @xmath64 on the path have completely determined the values on @xmath258 $ ] . the final extension from @xmath258 $ ] to the complete rectangle
@xmath244 $ ] is done similarly .
[ path ] units < 0.7 cm , 0.7 cm > x from -6 to 6 , y from -2 to 2 corners at -3 -2 and -2 -1 corners at -2 -2 and -1 -1 corners at -1 -2 and 0 -1 corners at 0 -2 and 1 -1 corners at 0 -1 and 1 0 corners at 0 0 and 1 1 corners at 1 1 and 2 1 corners at 1 0 and 2 1 corners at 1 1 and 2 2 corners at 2 1 and 3 2 corners at -3 -2 and 3 2 at -2.5 -1.5 at 2.5 1.5 at each step , there are at least two different choices for the extension in any direction for which @xmath64 is not already defined . in particular , one can ensure that @xmath260 .
therefore @xmath64 is not @xmath167-periodic .
[ l2 ] consider the directed graph which has a vertex for each @xmath226 and a directed edge from @xmath188 to @xmath189 for each @xmath229 such that @xmath228 .
this graph is irreducible .
i.e. condition ( h2 ) holds .
given @xmath261 we need to find a directed path starting at @xmath235 and ending at @xmath262 .
there are @xmath198 letters @xmath263 such that @xmath264 . for each such @xmath236
there are @xmath200 letters @xmath265 such that @xmath266 . since @xmath267 is a @xmath192-matrix ( equivalently
the paths @xmath268 are distinct ) it follows that the set @xmath269 contains @xmath270 elements .
see figure [ w1 ] .
similarly the set @xmath271 contains @xmath270 elements . since @xmath272 , we have @xmath273 and so there exists @xmath274 .
it follows that there is a directed path from @xmath235 ( to @xmath188 ) to @xmath262 , as required .
associated with the 2-dimensional shift system constructed above there is a finitely generated abelian group defined as follows .
the block @xmath275 matrix @xmath276 defines a homomorphism @xmath277 .
define @xmath278 .
thus @xmath279 can be defined as an abelian group , in terms of generators and relations : @xmath280 as we shall see , this group plays an important role in classifying the @xmath9-algebra @xmath11 which is studied in the next section .
the next observation will be needed there .
[ mtranspose ] there exists a permutation matrix @xmath281 such that @xmath282 and @xmath283 in particular @xmath284 .
define @xmath285 by @xmath286 .
( this corresponds to a rotation of the square in figure [ geometric square ] through the angle @xmath287 . )
then @xmath187 if and only if @xmath288 .
that is @xmath289 .
let @xmath281 be the corresponding permutation matrix defined by @xmath290 , where @xmath291 is the standard basis of @xmath292 .
a * sector * in @xmath26 is a @xmath293-angled sector in some apartment .
two sectors are * equivalent * ( or parallel ) if their intersection contains a sector .
see figure [ equivalent sectors ] , where the equivalent sectors with base vertices @xmath162 , @xmath294 do not necessarily lie in a common apartment , but the shaded subsector is contained in them both .
[ equivalent sectors ] units < 1 cm , 1 cm > x from -6 to 6 , y from -1 to 2.5 from 0 0 to 4 0 from 0 0 to 0 2 from 1 -0.5 to 4.5 -0.5 from 1 -0.5 to 1 1.5 from 2 1 to 4 1 from 2 1 to 2 2 [ r ] at -0.1 0 [ r ] at 0.9 -0.5 span < 4pt > 2 1 2 4 1 2 / the boundary @xmath295 of @xmath26 is defined to be the set of equivalence classes of sectors in @xmath26 . fix a vertex @xmath162 . for any @xmath296
there is a unique sector @xmath297 in the class @xmath298 having base vertex @xmath162 , as illustrated in figure [ repsec ] ( * ? ? ?
* theorem 9.6 ) .
[ repsec ] units < 1 cm , 1 cm > x from -6 to 6 , y from -1 to 2.2 from -3 -1 to 2 -1 from -3 -1 to -3 2 at -3 -1 [ t , r ] at -3.2 -1.1 [ l ] at 0 1 @xmath295 is a totally disconnected compact hausdorff space with a basis for the topology given by sets of the form @xmath299 where @xmath24 is any fixed vertex of @xmath26 .
it is easy to see that @xmath295 is ( non canonically ) homeomorphic to @xmath300 .
recall from section [ subshift ] that the alphabet @xmath127 is identified with the set of @xmath8-equivalence classes of basepointed chambers in @xmath26 .
we refer to such an equivalence class as a tile .
each tile has a unique representative labelled square based at a fixed vertex of @xmath26 , where each edge label is a generator of @xmath8 . for each vertex @xmath301 the convex hull @xmath302 is a rectangle @xmath303 in some apartment .
associated with the rectangle @xmath303 there is therefore a unique word @xmath231 defined by the labellings of the constituent squares of @xmath303 .
conversely , by construction of the bm group , every word @xmath231 arises in this way .
there is thus a natural bijection between the set of rectangles @xmath303 in @xmath26 based at @xmath162 and the set of words @xmath231 .
denote by @xmath304 the basepointed final chamber ( square ) in the rectangle @xmath303 .
thus @xmath304 has edge labelling corresponding to @xmath305 .
the square @xmath304 is oriented , with basepoint chosen to be the vertex closest to the origin @xmath162 .
it is worth recalling that the terminology has been set up so that @xmath306 .
that is , tiles are words of shape @xmath307 .
if @xmath231 , denote by @xmath308 the set of all @xmath309 such that the sector @xmath310 contains the rectangle in @xmath26 based at @xmath162 , corresponding to the word @xmath64 .
denote by @xmath311 the indicator function of this set .
it is clear from the definition of the topology on @xmath312 that @xmath313 .
[ word ] units < 0.7 cm , 0.7 cm > x from -6 to 6 , y from -1 to 5 from -3 -1 to 7 -1 from -3 0 to 2 0 from -3 1 to 2 1 from -3 -1 to -3 5 from -2 -1 to -2 1 from -1 -1 to -1 1 from 0 -1 to -0 1 from 1 -1 to 1 1 from 2 -1 to 2 1 [ l ] at 2.2 1 [ r ] at -3.2 -1 [ l ] at 5 3 at 1.5 0.5 the group @xmath8 acts on @xmath295 and hence on @xmath314 via @xmath315 , where @xmath316 , for @xmath317 , @xmath318 . the algebraic crossed product relative to this action is the @xmath319-algebra @xmath320 of functions @xmath321 of finite support , with multiplication and involution given by @xmath322 the full crossed product algebra @xmath323 is the completion of the algebraic crossed product in an appropriate norm .
there is a natural embedding of @xmath324 into @xmath323 which maps @xmath325 to the function taking the value @xmath326 at the identity of @xmath8 and @xmath208 elsewhere .
the identity element @xmath327 of @xmath323 is then identified with the constant function @xmath328 .
there is a natural unitary representation @xmath329 , where @xmath330 is the function taking the value @xmath327 at @xmath331 and @xmath208 otherwise .
it is convenient to denote @xmath330 simply by @xmath331 .
thus a typical element of the dense @xmath319-algebra @xmath332 can be written as a finite sum @xmath333 , where @xmath334 , @xmath318 .
the definition of the multiplication implies the covariance relation @xmath335 [ main01 ] let @xmath336 .
then @xmath11 is isomorphic to the rank-2 cuntz - krieger algebra @xmath10 associated with the alphabet @xmath127 and transition matrices @xmath186 , as described in @xcite .
the proof of this result is essentially the same as that given in ( * ? ? ?
* section 7 ) , in the case of a group of automorphisms of a building of type @xmath337 . here is how the isomorphism is defined .
the @xmath9-algebra @xmath10 is defined as the universal @xmath9-algebra generated by a family of partial isometries @xmath338 satisfying the relations [ rel1 * ] @xmath339 we refer to ( * ? ? ?
* section 1 ) for details , in particular for the meaning of the product of words used in ( [ rel1c * ] ) .
the isomorphism @xmath340 is defined as follows .
if @xmath341 with @xmath342 , let @xmath343 be the unique element such that @xmath344 .
the condition @xmath345 means that @xmath346 , @xmath347 lie in the same @xmath3-orbit , so that @xmath348 exists .
moreover @xmath348 is unique , since @xmath3 acts freely on @xmath26 .
now define @xmath349 the proof of that @xmath350 is an isomorphism is exactly the same as the corresponding result for @xmath337 groups given in ( * ? ? ?
* section 7 ) . here are the essential details .
the equation ( [ isomorphism ] ) does define a @xmath319-homomorphism of @xmath10 because the operators of the form @xmath351 are easily seen to satisfy the relations ( [ rel1 * ] ) . since the algebra @xmath10 is simple ( * ? ? ?
* theorem 5.9 ) , @xmath350 is injective .
now observe that @xmath352 .
it follows that the range of @xmath350 contains @xmath314 . for the sets @xmath308 , @xmath231 , form a basis for the topology of @xmath295 , and
so the linear span of @xmath353 is dense in @xmath314 . to show that @xmath350 is surjective
, it therefore suffices to show that the range of @xmath350 contains @xmath8 .
it is clearly enough to show that @xmath354 contains the generating set @xmath148 for @xmath8 .
suppose that @xmath179 . then @xmath355 similarly @xmath356 . in view of lemmas
[ m_1m_2 ] , [ l2 ] , the following is an immediate consequence of ( * ? ? ?
* proposition 5.11 , theorem 5.9 , corollary 6.4 and remark 6.5 ) .
[ main1 ] the @xmath9-algebra @xmath11 is purely infinite , simple and nuclear .
moreover it satisfies the universal coefficient theorem .
it also follows from @xcite that @xmath11 satisfies the u.c.t .
, hence it is classified by its k - theory , together with the class @xmath357 $ ] of its identity element in @xmath13 .
it is therefore of interest to determine the k - theory of @xmath11 .
the matrices @xmath276 and @xmath358 define homomorphisms @xmath277 . the k - theory of @xmath11 can be expressed as follows @xcite , where @xmath359 denote the torsion part of a finitely generated abelian group @xmath360 , and @xmath361 denotes the rank of @xmath360 . @xmath362 recall that we defined @xmath363 .
the next result therefore follows from lemma [ mtranspose ] .
[ maink ] if @xmath8 is a bm group then @xmath364 the identity element in @xmath11 is denoted by @xmath327 . as is the case for similar algebras (
* proposition 5.4 ) , @xcite , the class @xmath357 $ ] has torsion in @xmath365 . in the present setup we can be much more precise . for notational convenience , let @xmath366 and let @xmath367 .
[ orderof1 ] let @xmath357 $ ] be the class in @xmath365 of the identity element of @xmath11 , where @xmath8 is a bm group .
then @xmath368=0 $ ] .
\(a ) if @xmath369 is odd , then the order of @xmath357 $ ] is precisely @xmath369 .
\(b ) if @xmath369 is even , then the order of @xmath357 $ ] is either @xmath369 or @xmath370 .
the proof of this result depends upon an examination of explicit projections in @xmath11 .
if @xmath94 is an oriented basepointed square in @xmath26 with base vertex @xmath20 , let @xmath371 denote the clopen subset of @xmath312 consisting of all boundary points with representative sector having initial square @xmath94 and initial vertex @xmath20 .
the indicator function @xmath372 of the set @xmath371 is continuous and so lies in @xmath373 .
see figure [ bdpt ] .
the covariance relation ( [ covar ] ) implies that the class of @xmath372 in @xmath365 depends only on the @xmath8-equivalence class of the oriented basepointed square @xmath94 .
recall that we identify such a @xmath8-equivalence class with a tile @xmath265 .
it is therefore appropriate to denote the class of @xmath372 in @xmath365 by @xmath374 $ ] .
[ bdpt ] units < 1 cm , 1 cm > x from -6 to 6 , y from -1 to 2.2 from -3 -1 to 2 -1 from -3 -1 to -3 2 from -3 0 to -2 0 from -2 -1 to -2 0 [ .3,.67 ] from -2.5 -1 to -2.3 -1 [ .3,.67 ] from -3 -0.5 to -3 -0.3 at -3 -1 [ l ] at 1 1 [ t ] at -2.5 -1.2 [ r ] at -3.2 -0.5 at -2.5 -0.5 [ t , r ] at -3.1 -1.1 similarly , to each @xmath179 and @xmath375 we can associate elements @xmath376 , [ b ] \in k_0({{\mathcal a}}({{\gamma}}))$ ] .
for example , if @xmath179 , fix a directed edge labelled by the element @xmath179 and consider the set of all boundary points @xmath377 with representative sector having initial square @xmath94 containing that edge , as in figure [ bdpt ] . as
above , the class in @xmath365 of the characteristic function of this set depends only on the label @xmath378 , and may be denoted by @xmath376 $ ] .
the class @xmath379\in k_0({{\mathcal a}}({{\gamma}}))$ ] for @xmath375 is defined similarly .
recall now the following result .
[ ronan ] ( * ? ? ?
* lemma 9.4 ) given any chamber @xmath380 and any sector @xmath169 in @xmath26 , there exists a sector @xmath381 such that @xmath382 and @xmath94 lie in a common apartment .
it follows by considering parallel sectors in an appropriate apartment that if @xmath383 $ ] is a directed edge in @xmath26 and if @xmath384 , then @xmath298 has a representative sector @xmath169 that lies relative to @xmath385 in one of the two positions in figure [ chambersector ] , in some apartment containing them both .
[ chambersector ] units < 0.8 cm , 0.8 cm > x from -6 to 6 , y from -1 to 2.2 from -3 -1 to 1.5 -1 from -3 -1 to -3 2 from -3 0 to -2 0 from -2 -1 to -2 0 at -3 -1 at 0 1 [ t ] at -2.5 -1.2 [ t , r ] at -3.1 -1.1 [ .3,.67 ] from -2.5 -1 to -2.3 -1 units < 0.8 cm , 0.8 cm > point at -7 0 x from -6 to 6 , y from -1 to 2.2 from -3 -1 to 1.5 -1 from -3 -1 to -3 2 from -3 0 to -2 0 from -2 -1 to -2 0 at -3 -1 at 0 1 [ t ] at -2.5 -1.2 [ t , r ] at -3.1 -1.1 [ .3,.67 ] from -2.5 -1 to -2.7 -1 let @xmath386 denote the characteristic function of the set of points @xmath296 such that @xmath385 is contained in @xmath387 , as in the left hand diagram .
let @xmath388 denote the characteristic function of the set of points @xmath296 such that @xmath385 is contained in @xmath389 , as in the right hand diagram .
it follows that @xmath390 are idempotents in @xmath11 and @xmath391 .
if the edge @xmath385 has label @xmath179 [ respectively @xmath375 ] then in @xmath365 , @xmath393= [ a]\ , \text{and}\ , [ \overline p_e]=[a^{-1}]$ ] [ respectively @xmath393= [ b]\ , \text{and}\ , [ \overline p_e]=[b^{-1}]$ ] ] . we therefore obtain @xmath394 & = [ a ] + [ a^{-1 } ] , \qquad a\in a\\ & = [ b ] + [ b^{-1 } ] , \qquad b\in b. \end{split}\ ] ] the relations ( [ relationsp ] ) imply that @xmath395 & = \displaystyle\sum_{a\in a}[a],\\ \beta[{{\bf 1 } } ] & = \displaystyle\sum_{a\in b}[b ] .
\end{split}\ ] ] also , each boundary point @xmath298 has a unique representative sector based at a fixed vertex @xmath20 with initial edges as in figure [ bdpt ] .
it follows that @xmath396 = \displaystyle\sum_{a\in a}[a ] = \displaystyle\sum_{b\in b}[b].\ ] ] therefore @xmath396 = \alpha[{{\bf 1}}]=\beta[{{\bf 1}}],\ ] ] from which it follows that @xmath368=0 $ ] , thus proving the first assertion in proposition [ orderof1 ] . in order to obtain a lower bound for the order of the class @xmath357 $ ] in @xmath365 we need to use the fact ( proved in @xcite ) that the map @xmath397 $ ] is a monomorphism from the abstract group @xmath398 onto a direct summand of @xmath365
the class @xmath357 $ ] is the image of the element @xmath399 under this map .
moreover each column of the matrix @xmath197 [ respectively @xmath199 ] has @xmath401 [ respectively @xmath402 nonzero terms .
( see section [ subshift ] . )
let @xmath403 , and define a map @xmath404 by @xmath405 .
the map @xmath350 is well defined since each relation in the presentation of @xmath279 expresses a generator @xmath188 as a sum of @xmath401 or @xmath406 other generators .
also @xmath407 , since @xmath408 .
there are now two cases to consider .
\(a ) suppose that @xmath369 is odd .
then @xmath409 has order @xmath369 in @xmath410 .
therefore the order of @xmath385 in @xmath279 is divisible by @xmath369 and hence equal to @xmath369 .
\(b ) suppose that @xmath369 is even .
then @xmath409 has order @xmath370 in @xmath410 .
therefore the order of @xmath385 in @xmath279 is divisible by @xmath370 .
[ conjectures ] it is tempting to conjecture that the order of the class @xmath357 $ ] is always precisely @xmath369 . as we shall see below
, there is some supporting evidence for this .
there is also computational evidence that @xmath411 , that is @xmath412 .
[ tpuct ] recall from theorem [ main1 ] that @xmath11 is a p.i.s.u.n .
@xmath9-algebra satisfying the uct . furthermore , by theorem [ maink ] , @xmath413 where @xmath14 is a finite abelian group .
it follows from ( * ? ? ?
* proposition 7.3 ) that @xmath11 is stably isomorphic to @xmath414 , where @xmath415 , @xmath416 are simple rank one cuntz - krieger algebras and @xmath417 .
in this section we consider some examples of bm groups @xmath8 and the results of the computations for the group @xmath418 .
it is useful to relate our results to the euler - poincar characteristic @xmath419 , which is the alternating sum of the ranks of the groups @xmath420 .
the finite cell complex @xmath421 is a @xmath422 space and @xmath8 has homological dimension at most two , so that @xmath423 is free abelian and @xmath424 for @xmath425 .
it follows that @xmath419 coincides with the usual euler - poincar characteristic of the cell complex @xmath426 .
explicitly , @xmath427 , where @xmath428 and @xmath429 .
suppose that @xmath8 is a direct product of free groups of ranks @xmath430 and @xmath431 , acting on a product @xmath26 of homogeneous trees of degrees @xmath428 and @xmath429 respectively .
then direct computation shows that @xmath432 where @xmath433 is the order of the class @xmath357 $ ] . for this group , @xmath434 , and
the conjectures of remark [ conjectures ] are verified . in this example
@xmath11 is actually isomorphic in a natural way to a tensor product of cuntz - krieger algebras .
in fact , by a result of j. spielberg ( * ? ? ? * section 1 ) , the action of a free group @xmath435 on the boundary of its cayley graph gives rise a cuntz - krieger algebra @xmath436 .
it is easy to check that @xmath437 .
( compare with remark [ tpuct ] . ) the formula for @xmath365 can thus also be verified using the knneth theorem for tensor products .
consider some specific examples studied in ( * ? ? ?
* section 3 ) . if @xmath438 are two distinct primes , mozes constructs a lattice subgroup @xmath439 of @xmath440 .
the building @xmath26 of @xmath360 is a product of two homogeneous trees @xmath0 , @xmath1 of degrees @xmath441 and @xmath442 respectively .
the group @xmath439 is a bm group which acts freely and transitively on the vertex set of @xmath26 , but @xmath439 is * not * a product of free groups . in fact
@xmath439 is an irreducible lattice in @xmath360 . here
is how @xmath439 is constructed @xcite .
let @xmath443 , the ring of integer quaternions .
let @xmath444 be a square root of @xmath445 in @xmath446 and define @xmath447 by @xmath448 let @xmath449 .
then @xmath450 ) is a torsion free cocompact lattice in @xmath360 .
let @xmath451 then @xmath139 contains @xmath452 elements and @xmath140 contains @xmath453 elements .
the images @xmath454 , @xmath455 of @xmath118 in @xmath439 generate free groups @xmath456 , @xmath457 of orders @xmath458 , @xmath459 respectively and @xmath439 itself is generated by @xmath460 .
the product @xmath28 is the cayley graph of @xmath439 relative to this set of generators .
the group @xmath439 is a bm group and @xmath461 is even .
explicit computations , using the formula ( [ kth ] ) and the magma computer algebra package , show that the order of @xmath357 $ ] is @xmath369 in each of the 28 groups @xmath439 where @xmath438 are two distinct primes @xmath462 . the normal subgroup theorem ( * ? ? ?
* theorem 4.1 ) can be applied to @xmath439 , if the legendre symbol @xmath463 .
for , using the notation of ( * ? ? ?
* section 2.4 ) and ( * ? ? ?
* remarks following proposition 1.8.1 ) , the group @xmath464 has finite index in @xmath465 .
moreover @xmath466 is locally @xmath467 transitive .
thus the hypotheses of [ bm2 : theorem 4.1 ] are satisfied . applying ( * ? ? ?
* theorem 4.1 ) to the commutator subgroup @xmath468 $ ] of @xmath439 shows that the abelianization @xmath469 $ ] is finite .
the euler - poincar characteristic of @xmath439 is @xmath470 .
thus @xmath471 and so @xmath472 .
explicit computations for the same range of values of @xmath473 as above shows that the second conjecture of remark [ conjectures ] is also verified in these cases .
another experimental observation is the following . checking through a large number of values of the pairs of primes @xmath474 congruent to 1 mod 4
, one sees that the abelianization of @xmath439 seems to depend only the greatest common divisor @xmath475 .
in fact we conjecture that @xmath476 the validity of this formula was checked for all values of the pair @xmath474 up to @xmath477 , and for several other values .
a bm group acts on a product of homogeneous trees @xmath28 , where @xmath0 has degree @xmath479 and @xmath1 has degree @xmath480 .
we now examine the simplest case @xmath478 .
recall that a bm square is a set of four distinct tuples @xmath134 contained in @xmath133 , as in definition [ def bm group](i),(ii ) . in the presentation ( [ presentation ] ) ,
a bm group has @xmath481 generators and a set of relations which we call a set of bm relations .
possible sets of bm relations correspond to those unions of @xmath482 bm squares whose images under a projection chosen from definition [ def bm group](iii ) are disjoint .
choose once and for all such a projection , say @xmath483 .
consider the graph @xmath484 whose vertices are the bm squares and whose edges join bm squares having disjoint images under the chosen projection .
possible sets of bm relations correspond to cliques ( complete subgraphs ) of size @xmath482 in the graph @xmath484 .
the computer algebra package magma contains a dynamic programming algorithm due to @xcite for finding all cliques of a given size .
for @xmath485 there are 541 cliques in @xmath484 of size @xmath128 .
say that two of the corresponding presentations are equivalent if one may be obtained from the other by permuting the generators .
there are precisely 52 equivalence classes of presentations . in the tables
below we list 52 representative presentations , together with the structure of the groups @xmath486 and @xmath487 . for bm groups
@xmath8 with semi - degrees @xmath485 , we have @xmath488 . thus @xmath489 .
therefore the tables show that the conjectures of remark [ conjectures ] are verified for all such groups . in the tables ,
each of the 52 groups arising from these presentations is assigned a name @xmath490 . using an algorithm due to c. c. sims @xcite , which is implemented in the magma package
, we have found the indices and lengths of all conjugacy classes of subgroups with index at most 8 in each of these 52 groups . from these data
we have checked that the only possible isomorphisms between @xmath491 and @xmath492 occur if both @xmath493 and @xmath494 lie in one of the following sets : the quotient complexes @xmath504 corresponding to the groups @xmath505 with @xmath493 in any of these sets are not homeomorphic , as can be seen by considering the fundamental groups of one point deletion subspaces of @xmath146 .
one therefore obtains 52 pairwise non - homeomorphic complexes @xmath146 . of course
, this does not imply that the groups @xmath8 are non - isomorphic .
in fact , we have discovered tietze transformations showing that , if @xmath506 is contained in any one of the listed sets , except for @xmath507 or @xmath498 , then the corresponding groups are isomorphic .
there are thus at most 43 non - isomorphic groups .
we do not know whether the groups @xmath491 and @xmath492 are isomorphic in the two cases @xmath508 and @xmath509 .
note that the shift system group @xmath487 depends only on @xmath8 and not on the presentation of @xmath8 .
the following notation is used in the tables .
the group @xmath3 is generated by @xmath148 where @xmath510 and @xmath511 .
the fixed point free involution on each of these sets is given by @xmath512 .
thus @xmath513 is denoted @xmath514 . moreover the relations are written in the form @xmath515 rather than the form @xmath516 used in definition [ def bm group ] .
for example , the first relation for the presentation of group @xmath517 is @xmath518 .
the tables are simplified by use of the abbreviations : @xmath523 $ & 2$\left [ \left ( 2\right ) 4\right ] $ \\\hline 2x2.02 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+1&-4 \\+2&+3&+2&-3\qquad&+2&+4&-2&+4 \end{matrix}$ & 1$\left [ \left ( 3\right ) 2\right ] $ & 1$\left [ \left ( 3\right ) 4\right ]
$ \\\hline 2x2.03 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+1&-4 \\+2&+3&+2&-3\qquad&+2&+4&-2&-4 \end{matrix}$ & 2$\left [ \left ( 2\right ) 2\right ] $ & 2$\left [ \left ( 2\right ) 4\right ] $ \\\hline 2x2.04 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+1&-4 \\+2&+3&+2&+4\qquad&+2&-3&+2&-4 \end{matrix}$ & 1$\left [ 2,4\right ] $ & 1$\left [ 2,4,8\right ] $ \\\hline 2x2.05 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+1&-4 \\+2&+3&-2&+3\qquad&+2&+4&-2&+4 \end{matrix}$ & 1$\left [ \left ( 3\right ) 2\right ] $ & 1$\left [ \left ( 3\right ) 4\right ]
$ \\\hline 2x2.06 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+1&-4 \\+2&+3&-2&+3\qquad&+2&+4&-2&-4 \end{matrix}$ & 2$\left [ \left ( 2\right ) 2\right ] $ & 2$\left [ \left ( 2\right ) 4\right ]
$ \\\hline 2x2.07 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+1&-4 \\+2&+3&-2&-3\qquad&+2&+4&-2&-4 \end{matrix}$ & 3$\left [ 2\right ] $ & 3$\left [ 4\right ] $ \\\hline 2x2.08 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+1&-4 \\+2&+3&-2&+4\qquad&+2&+4&-2&+3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ 2,4\right ] $ \\\hline 2x2.09 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+1&-4 \\+2&+3&-2&+4\qquad&+2&+4&-2&-3 \end{matrix}$ & 1$\left [ \left ( 2\right ) 2\right ] $ & 1$\left [ \left ( 2\right ) 2,4\right ] $ \\\hline 2x2.10 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&+4 \\+2&+3&+2&-3\qquad&+2&+4&-2&-4 \end{matrix}$ & 1$\left [ \left ( 3\right ) 2\right ] $ & 1$\left [ \left ( 3\right ) 4\right ] $ \\\hline 2x2.11 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&+4 \\+2&+3&+2&+4\qquad&+2&-3&+2&-4 \end{matrix}$ & 0$\left [ \left ( 2\right ) 2,4\right ] $ & 0$\left [ \left ( 2\right ) 4,8\right ] $ \\\hline 2x2.12 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&+4 \\+2&+3&-2&+3\qquad&+2&+4&+2&-4 \end{matrix}$ & 0$\left [ \left ( 4\right ) 2\right ] $ & 0$\left [ \left ( 2\right ) 2,\left ( 3\right ) 4\right ] $ \\\hline 2x2.13 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&+4 \\+2&+3&-2&+3\qquad&+2&+4&-2&-4 \end{matrix}$ & 1$\left [ \left ( 3\right ) 2\right ] $ & 1$\left [ \left ( 3\right ) 4\right ] $ \\\hline 2x2.14 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&+4 \\+2&+3&-2&-3\qquad&+2&+4&-2&-4 \end{matrix}$ & 2$\left [ \left ( 2\right ) 2\right ] $ & 2$\left [ \left ( 2\right ) 4\right ]
$ \\\hline 2x2.15 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&+4 \\+2&+3&-2&+4\qquad&+2&+4&-2&+3 \end{matrix}$ & 1$\left [ \left ( 2\right ) 2\right ] $ & 1$\left [ \left ( 2\right ) 4\right ] $ \\\hline 2x2.16 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&+4 \\+2&+3&-2&+4\qquad&+2&+4&-2&-3 \end{matrix}$ & 1$\left [ \left ( 2\right ) 2\right ] $ & 1$\left [ \left ( 2\right ) 4\right ]
$ \\\hline 2x2.17 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&-4 \\+2&+3&+2&-3\qquad&+2&+4&-2&-4 \end{matrix}$ & 2$\left [ \left ( 2\right ) 2\right ] $ & 2$\left [ \left ( 2\right ) 4\right ] $ \\\hline 2x2.18 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&-4 \\+2&+3&+2&+4\qquad&+2&-3&+2&-4 \end{matrix}$ & 1$\left [ 2,4\right ] $ & 1$\left [ 2,4,8\right ]
$ \\\hline 2x2.19 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&-4 \\+2&+3&-2&-3\qquad&+2&+4&+2&-4 \end{matrix}$ & 2$\left [ \left ( 2\right ) 2\right ] $ & 2$\left [ \left ( 2\right ) 4\right ] $ \\\hline 2x2.20 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&-4 \\+2&+3&-2&-3\qquad&+2&+4&-2&+4 \end{matrix}$ & 2$\left [ \left ( 2\right ) 2\right ] $ & 2$\left [ \left ( 2\right ) 4\right ] $ \\\hline \end{tabular}\ ] ] @xmath524 $ & 3$\left [ 4\right ] $ \\\hline 2x2.22 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&-4 \\+2&+3&-2&+4\qquad&+2&+4&-2&+3 \end{matrix}$ & 2$\left [ 2\right ]
$ & 2$\left [ 2,4\right ] $ \\\hline 2x2.23 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&-1&-4 \\+2&+3&-2&+4\qquad&+2&+4&-2&-3 \end{matrix}$ & 1$\left [ \left ( 2\right ) 2\right ] $ & 1$\left [ \left ( 2\right ) 2,4\right ] $ \\\hline 2x2.24 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+2&+4 \\+1&-4&+2&-4\qquad&+2&+3&+2&-3 \end{matrix}$ & 1$\left [ 2,4\right ] $ & 1$\left [ \left ( 2\right ) 2,8\right ]
$ \\\hline 2x2.25 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+2&+4 \\+1&-4&+2&-4\qquad&+2&+3&-2&-3 \end{matrix}$ & 1$\left [ 2,4\right ] $ & 1$\left [ \left ( 2\right ) 2,8\right ] $ \\\hline 2x2.26 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+2&-4 \\+1&-4&+2&+4\qquad&+2&+3&+2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ \left ( 2\right ) 2\right ] $ \\\hline 2x2.27 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+2&-4 \\+1&-4&+2&+4\qquad&+2&+3&-2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ \left ( 2\right ) 2\right ]
$ \\\hline 2x2.28 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+2&-4 \\+1&-4&-2&+4\qquad&+2&+3&+2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ \left ( 2\right ) 2\right ]
$ \\\hline 2x2.29 & $ \begin{matrix } + 1&+3&+1&-3\qquad&+1&+4&+2&-4 \\+1&-4&-2&+4\qquad&+2&+3&-2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ \left ( 2\right ) 2\right ]
$ \\\hline 2x2.30 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+1&-4 \\+2&+3&+2&+4\qquad&+2&-3&+2&-4 \end{matrix}$ & 1$\left [ 2,4\right ] $ & 1$\left [ 2,4,8\right ] $ \\\hline 2x2.31 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+1&-4 \\+2&+3&+2&-4\qquad&+2&-3&+2&+4 \end{matrix}$ & 0$\left [ 2,\left ( 2\right ) 4\right ] $ & 0$\left [ \left ( 2\right ) 2,\left ( 2\right ) 8\right ] $ \\\hline 2x2.32 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+1&-4 \\+2&+3&-2&-3\qquad&+2&+4&-2&-4 \end{matrix}$ & 2$\left [ 4\right ] $ & 2$\left [ 2,8\right ] $ \\\hline 2x2.33 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+1&-4 \\+2&+3&-2&+4\qquad&+2&+4&-2&+3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ 2,4\right ] $ \\\hline 2x2.34 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+1&-4 \\+2&+3&-2&+4\qquad&+2&+4&-2&-3 \end{matrix}$ & 1$\left [ \left ( 2\right ) 2\right ] $ & 1$\left [ \left ( 2\right ) 2,4\right ] $ \\\hline
2x2.35 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+1&-4 \\+2&+3&-2&-4\qquad&+2&+4&-2&-3 \end{matrix}$ & 1$\left [ 2,4\right ] $ & 1$\left [ \left ( 2\right ) 2,8\right ] $ \\\hline
2x2.36 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+2&-3 \\+1&-4&+2&-4\qquad&+2&+3&+2&+4 \end{matrix}$ & 0$\left [ 4,8\right ] $ & 0$\left [ \left ( 2\right ) 2,4,8\right ] $ \\\hline 2x2.37 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+2&-3
\\+1&-4&-2&-4\qquad&+2&+3&+2&-4 \end{matrix}$ & 0$\left [ \left ( 2\right ) 2,\left ( 2\right ) 3\right ] $ & 0$\left [ \left ( 2\right ) 4,\left ( 2\right ) 3\right ] $ \\\hline 2x2.38 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+2&-4 \\+1&-4&+2&-3\qquad&+2&+3&+2&+4 \end{matrix}$ & 1$\left [ 8\right ] $ & 1$\left [ \left ( 2\right ) 2,8\right ]
$ \\\hline 2x2.39 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+2&-4 \\+1&-4&-2&-3\qquad&+2&+3&-2&+4 \end{matrix}$ & 1$\left [ 2\right ] $ & 1$\left [ \left ( 3\right ) 2\right ] $ \\\hline 2x2.40 & $ \begin{matrix } + 1&+3&+1&+4\qquad&+1&-3&+2&-4
\\+1&-4&-2&-3\qquad&+2&+3&-2&-4 \end{matrix}$ & 0$\left [ \left ( 2\right ) 2,3\right ] $ & 0$\left [ \left ( 2\right ) 2,4,3\right ] $ \\\hline 2x2.41 & $ \begin{matrix } + 1&+3&-1&-3\qquad&+1&+4&-1&-4 \\+2&+3&-2&-3\qquad&+2&+4&-2&-4 \end{matrix}$ & 4$\left [ \right ] $ & 4$\left [ \right ]
$ \\\hline 2x2.42 & $ \begin{matrix } + 1&+3&-1&-3\qquad&+1&+4&-1&-4 \\+2&+3&-2&+4\qquad&+2&+4&-2&+3 \end{matrix}$ & 3$\left [ \right ] $ & 3$\left [ 2\right ] $ \\\hline 2x2.43 & $ \begin{matrix } + 1&+3&-1&-3\qquad&+1&+4&-1&-4 \\+2&+3&-2&+4\qquad&+2&+4&-2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ \left ( 2\right ) 2\right ]
$ \\\hline 2x2.44 & $ \begin{matrix } + 1&+3&-1&+4\qquad&+1&+4&-1&+3 \\+2&+3&-2&+4\qquad&+2&+4&-2&+3 \end{matrix}$ & 3$\left [ \right ] $ & 3$\left [ 2\right ] $ \\\hline 2x2.45 & $ \begin{matrix } + 1&+3&-1&+4\qquad&+1&+4&-1&+3 \\+2&+3&-2&+4\qquad&+2&+4&-2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ \left ( 2\right ) 2\right ]
$ \\\hline 2x2.46 & $ \begin{matrix } + 1&+3&-1&+4\qquad&+1&+4&-1&+3 \\+2&+3&-2&-4\qquad&+2&+4&-2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ \left ( 2\right ) 2\right ]
$ \\\hline 2x2.47 & $ \begin{matrix } + 1&+3&-1&+4\qquad&+1&+4&-1&-3 \\+2&+3&-2&+4\qquad&+2&+4&-2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ \left ( 2\right ) 2\right ]
$ \\\hline 2x2.48 & $ \begin{matrix } + 1&+3&-1&+4\qquad&+1&+4&+2&+3 \\+1&-4&+2&-3\qquad&+2&+4&-2&+3 \end{matrix}$ & 2$\left [ \right ] $ & 2$\left [ \left ( 2\right ) 2\right ] $ \\\hline
2x2.49 & $ \begin{matrix } + 1&+3&-1&+4\qquad&+1&+4&+2&+3 \\+1&-4&+2&-3\qquad&+2&+4&-2&-3 \end{matrix}$ & 1$\left [ 2\right ] $ & 1$\left [ 2,4\right ] $ \\\hline 2x2.50 & $ \begin{matrix } + 1&+3&-1&+4\qquad&+1&+4&+2&-3 \\+1&-4&+2&+3\qquad&+2&+4&-2&+3 \end{matrix}$ & 1$\left [ 4\right ] $ & 1$\left [ \left ( 2\right ) 2,4\right ]
$ \\\hline 2x2.51 & $ \begin{matrix } + 1&+3&+2&+4\qquad&+1&-3&+2&-4 \\+1&+4&+2&+3\qquad&+1&-4&+2&-3 \end{matrix}$ & 2$\left [ 2\right ] $ & 2$\left [ 2,4\right ] $ \\\hline 2x2.52 & $ \begin{matrix } + 1&+3&+2&+4\qquad&+1&-3&+2&-4 \\+1&+4&+2&-3\qquad&+1&-4&+2&+3 \end{matrix}$ & 1$\left [ \left ( 2\right ) 2\right ] $ & 1$\left [ \left ( 2\right ) 2,4\right ] $ \\\hline \end{tabular}\ ] ] w. myrvold , t. prsa , and n. walker . a dynamic programming approach for timing and designing clique algorithms .
_ algorithms and experiments ( alex 98 ) : building bridges between theory and applications _ ,
88 - 95 , 1998 .
_ http://www.csr.uvic.ca/@xmath526wendym/pubs.html _
g. robertson , boundary actions for affine buildings and higher rank cuntz - krieger algebras .
_ @xmath9-algebras : proceedings of the sfb workshop on @xmath9-algebras _ ( mnster , march 812 , 1999 ) , 182202 , springer - verlag , 2000 . | let @xmath0 and @xmath1 be homogeneous trees of even degree @xmath2 .
a bm group @xmath3 is a torsion free discrete subgroup of @xmath4 which acts freely and transitively on the vertex set of @xmath5
. this article studies dynamical systems associated with bm groups .
a higher rank cuntz - krieger algebra @xmath6 is associated both with a 2-dimensional tiling system and with a boundary action of a bm group @xmath3 .
an explicit expression is given for the k - theory of @xmath6 .
in particular @xmath7 . a complete enumeration of possible bm groups @xmath8 is given for a product homogeneous trees of degree 4 , and the k - groups are computed . |
the casimir effect @xcite was studied in various specific cases and geometries @xcite .
a new geometry that recently attracted attention in the theory of the casimir effect is the piston geometry .
a piston plate is perpendicular to the walls of a semi - infinite cylinder and moves freely inside it , this geometry was first investigated in a @xmath0 dirichlet model @xcite . an exact solution for a perfectly conducting square piston at zero temperature
was found in a @xmath1 model in the electromagnetic and scalar cases @xcite by making use of a geometric optics approach ; the limit of short distances between the piston and the base of a cylinder was found in @xcite for an arbitrary cross section of a piston ; rectangular geometries and finite temperatures were considered in @xcite . in this paper and our previous related papers @xcite we considered a slightly different geometry - two piston plates inside an infinite cylinder , which yielded the same results for rectangular pistons as in the case of a semi - infinite cylinder due to perfectly conducting boundary conditions . in @xcite
an exact solution for arbitrary cross sections and arbitrary distances between piston plates was found at zero and finite temperatures in the electromagnetic @xmath1 case .
rectangular and circular cross sections are special cases of our general solution .
a dilute circular piston and cylinder were studied perturbatively in @xcite . in this case
the force on two plates inside a cylinder and the force in a piston geometry differ essentially .
the force in a piston geometry can change sign in this approximation for thin enough walls of the material .
different examples of pistons in a scalar case were investigated in @xcite .
the case when the piston s cross section differs from that of a cylinder was recently studied numerically @xcite and by means of a geometric optics approach @xcite .
everywhere in this paper we study casimir energies of an electromagnetic field with perfectly conducting boundary conditions imposed .
first we study the energy of a rectangular cavity by making use of pistons interactions .
then we generalize the formulas for the case of a 3d piston with an arbitrary cross section and consider several special and limiting cases .
our formulas can be applied in every case when the two dimensional dirichlet and neumann boundary problems for helmholtz equation can be solved analytically or numerically .
we take @xmath2 .
the casimir energy can be regularized as follows : @xmath3 where @xmath4 is large enough to make ( [ f23 ] ) convergent
. then it should be continued analytically ( [ f23 ] ) to the value @xmath5 , this procedure yields the renormalized finite casimir energy .
the regularized electromagnetic casimir energy for the rectangular cavity @xmath6 can be written in terms of epstein @xmath7 and riemann @xmath8 zeta functions : @xmath9^{-s/2 } - \\ & \quad - 2 \bigl(\frac{1}{a } + \frac{1}{b } + \frac{1}{c } \bigr ) \sum_{n=1}^{+\infty } \frac{1}{n^s } \biggr ) \end{split}\ ] ] @xmath10^{-s/2 }
\\ \zeta_r ( s ) & = \sum_{n=1}^{+\infty } \frac{1}{n^s}\end{aligned}\ ] ] the prime means that the term with all @xmath11 should be excluded from the sum .
the reflection formulas for an analytical continuation of zeta functions exist : @xmath12 the renormalized casimir energy for a perfectly conducting rectangular cavity can therefore be written as @xcite : @xmath13 the expression ( [ f24 ] ) can be rewritten in a different mathematical form @xcite : @xmath14\biggr ) .
\label{box}\end{gathered}\ ] ] here the casimir energy for a unit length of a rectangular waveguide is ( @xmath15 or @xmath16 ) : @xmath17 let us discuss different terms appearing in ( [ box ] ) and thus clarify the physical meaning of a zeta function regularization in this case .
imagine that a piston is large in two dimensions with sides @xmath18 and @xmath19 ( plates @xmath20 and @xmath21 on fig.@xmath20 ) . in this case
the contribution to the energy in this geometry is given by the casimir result for two parallel plates with the surface area @xmath22 and the edge term @xmath23 : @xmath24 where one of the terms in @xmath25 ( @xmath26 ) is taken into account .
the next step is to move other pistons ( plates @xmath27 and @xmath28 on fig.@xmath20 ) that have a large side @xmath18 between these already existing parallel plates .
the energy change is equal to @xmath29 where @xmath30 in ( [ ch2 ] ) another term in @xmath31 was taken into account ( @xmath32 is a distance between the two piston plates @xmath27 and @xmath28 , @xmath33 ) . from the energy change ( [ ch2 ] ) it is straightforward to obtain the force on the piston plates @xmath27 and @xmath28 , and in the limit @xmath34 one
immediately obtains the exact force on a unit length of stripes @xmath27 and @xmath28 from @xmath35 .
the expression for the energy change ( [ ch1 ] ) is valid only when @xmath36 is much smaller than the sizes of the plates @xmath20,@xmath21 , and ( [ ch2 ] ) is valid when @xmath32 is much smaller than @xmath18 and of the order @xmath36 or less .
some comments are needed to clarify the meaning of the terms @xmath37 in the energy expression ( [ box ] ) .
these terms appear due to edges of the piston .
they are precisely equal to next to leading order terms in the expansion ( [ r23 ] ) , which means that they account for @xmath28 rectangular edges ( @xmath38 ) of the finite piston in two different expansions ( for small @xmath36 and small @xmath32 ) . the next possible step is to insert pistons @xmath39 and @xmath40 from the opposite sides of the existing waveguide with sides @xmath36 and @xmath32 and move them towards each other .
the term @xmath41\biggr ) \label{lm3}\end{aligned}\ ] ] yields the interaction energy of two plates distance @xmath42 apart inside an infinite rectangular cylinder with sides @xmath36 and @xmath32 ( the term @xmath43 is excluded from the sum due to a cancellation with the term @xmath44 in ( [ box ] ) ) .
the force on the pistons @xmath39 and @xmath40 is straightforward . in summary ,
the expression @xmath45 yields the energy change inside a cavity volume with sides @xmath46 during a construction of the following system : step @xmath20 : : : pistons @xmath20 and @xmath21 are being moved inside a waveguide with large sides @xmath47 from a large distance between them towards each other until the distance @xmath36 between them is achieved . for a validity of the energy change ( [ ch1 ] )
it is necessary to assume @xmath48 .
step @xmath21 : : : pistons @xmath27 and @xmath28 with sides @xmath18 , @xmath36 are being moved inside the existing box @xmath49 between the existing pistons @xmath20 and @xmath21 towards each other until the distance @xmath32 between them is achieved . for a validity of the energy change ( [ ch2 ] )
it is necessary to assume @xmath50 , also @xmath51 .
step @xmath27 : : : pistons @xmath39 and @xmath40 with sides @xmath52 are being moved inside the existing box @xmath53 towards each other until the distance @xmath42 between them is achieved .
it is assumed here that @xmath54 .
however , the formula ( [ lm3 ] ) itself is exact for arbitrary values of @xmath42 in the limit @xmath34 , i.e. for plates inside an infinite waveguide . during each step the energy in the system decreases . in the limit of infinite plates
@xmath55 and stripes @xmath56 ( @xmath57 ) the total energy change ( steps @xmath58 ) inside the cavity volume with sides @xmath46 can be written in the form : @xmath59
suppose there are two plates inside an infinite cylinder of an arbitrary cross section @xmath60 ( fig.@xmath21 ) . to calculate the force on each plate imagine that @xmath28 parallel plates are inserted inside an infinite cylinder and then @xmath21 exterior plates are moved to spatial infinity .
this situation is exactly equivalent to @xmath27 perfectly conducting cavities touching each other . from the energy of this system one has to subtract the casimir energy of an infinite cylinder without plates inside it .
doing so we obtain the energy of interaction between the interior parallel plates and the attractive force on each interior plate inside the cylinder : @xmath61 the sum here is over all @xmath62 and @xmath63 eigenfrequencies @xmath64 for a cylinder with the cross section @xmath60 and an infinite length .
in fact , the casimir energy of our electromagnetic system is proportional to the sum of free energies for two boson scalar fields ( with dirichlet and neumann boundary conditions imposed at the boundary of an infinite cylinder with a cross section @xmath60 and zero neumann eigenvalue excluded ) at finite temperature @xmath65 if we make a substitution @xmath66 .
free energies have a well defined finite part , their sum up to a factor @xmath67 coincides with ( [ r7 ] ) .
one can rewrite ( [ r7 ] ) in a different form : @xmath68 here @xmath69 the results ( [ r7 ] ) , ( [ p10 ] ) are our main zero temperature results .
our results are _
exact for an arbitrary curved geometry of a cylinder_. for a rectangular cylinder with sides @xmath36 and @xmath32 the exact casimir energy of two plates inside it can be written as : @xmath70 the term @xmath71 is omitted in the sum . for a circular cylinder
the eigenvalues of the two dimensional helmholtz equation @xmath72 are determined by the roots of bessel functions and derivatives of bessel functions .
the exact casimir energy of two circular plates of the radius @xmath36 separated by a distance @xmath42 inside an infinite circular cylinder of the radius @xmath36 is given by : @xmath73 the sum is over positive @xmath74 and @xmath75 .
the leading asymptotic behaviour of @xmath76 for long distances @xmath77
, @xmath78 is determined by the lowest positive eigenvalues of the two dimensional dirichlet and neumann problems @xmath79 : @xmath80 so the casimir force between the two plates in a cylinder is exponentially small for long distances .
this important property of the solution is due to the gap in the frequency spectrum or , in other words , it is due to the finite size of the cross section of the cylinder . due to this property
one needs a finite number of the eigenvalues of the helmholtz equation for @xmath81 dirichlet and neumann boundary problems ( [ helm1]-[helm4 ] ) to obtain the casimir energy at a specific distance @xmath42 between the plates with a desired accuracy . the free energy at a temperature @xmath65 describing the interaction of two parallel perfectly conducting plates inside an infinite perfectly conducting cylinder with the cross section @xmath60 has the form : @xmath82 this is our central finite temperature result .
note that @xmath83 .
we used the standard notation @xmath84 . in the long distance limit
@xmath85 one has to keep only @xmath86 term in @xmath87 .
thus the free energy of the plates inside a cylinder in the high temperature limit is equal to : @xmath88 one can check that the limit @xmath89 , @xmath90 in ( [ r24 ] ) immediately yields the known high temperature result for two parallel perfectly conducting plates separated by a distance @xmath42 @xcite . for @xmath91 and @xmath89 , @xmath90 one can use the heat kernel expansion @xcite and properties of the zeta function @xcite to obtain the leading terms for the free energy : @xmath92 where @xmath93 here @xmath94 is an area of the cross section @xmath60 , @xmath95 is the interior angle of each sharp corner at the boundary @xmath96 and @xmath97 is the curvature of each boundary smooth section described by the curve @xmath98 .
the force calculated from ( [ r23 ] ) coincides with the zero temperature force @xmath99 in @xcite , ( eq.7 ) .
it is a pleasure to thank michael bordag for hospitality in leipzig during @xmath100 , also all colleagues for fruitful discussions in leipzig .
this work has been supported by a cnrs grant anr-06-nano-062 and grants rnp @xmath101 , ss @xmath102 , rfbr @xmath103 .
bordag m , mohideen u , mostepanenko v m 2001 _ phys.rept .
_ * 353 * 1 ; nesterenko v v , lambiase g , scarpetta g 2004 _ riv.nuovo cim . _ * 27 * ( n6 ) 1 ; milton k a 2004 _ j.phys . a _ * 37 * r209 ; + lambrecht a , maia neto p a and reynaud s 2006 _ new j.phys._ * 8 * 243 zaheer s , rodriguez a w , johnson s g and jaffe r l 2007 optical - approximation analysis of sidewall - spacing effects on the force between two squares with parallel sidewalls _ preprint _
quant - ph/0709.0699 | the energy of a perfectly conducting rectangular cavity is studied by making use of pistons interactions .
the exact solution for a 3d perfectly conducting piston with an arbitrary cross section is being discussed . |
field theory on noncommutative spaces ( for a review see @xcite ) are a interesting arena to study the properties of noncommutative geometry @xcite .
we will not review here the motivations for a noncommutative geometry . among the applications of noncommutative geometry are fuzzy spaces .
the original fuzzy space is the fuzzy sphere @xcite , but other spaces have been built ( for a review see @xcite ) .
field theory on fuzzy spaces are a way to study field theory on noncommutative spaces on a finite setting .
the fact that the approximation is based on matrices makes them ideal for numerical studies .
the most studied case is the fuzzy sphere , for a review see @xcite .
most of the interest for these investigation is the presence of the limit for which the rank of the matrices grow , and at the same time the radius of the sphere increases , thus recovering in the limit a noncommutative plane . here and in the following by noncommutative plane we mean the algebra generated by the noncommuting variables @xmath5 and @xmath6 with commutator = i[commrel ] this sort of plane is usually described by a noncommutative @xmath7 product among the fields , such as the grnewold - moyal product @xcite .
other product are possible , and a comparison among them is given in @xcite .
for our purposes it is however better to think of the `` quantization '' of the filed as a map from functions @xmath8 to operators @xmath9 . as we shall see
the fuzzy disc approximation will render the matrices finite , and hence amenable to numerical simulations .
we will study then a quantized @xmath0 scalar field theory approximating field with @xmath10 matrices , and perform three different limits , giving rise to the noncommutative plane , the commutative plane and the commutative disc .
we are interested in particular to the phase transitions @xcite of the theory as we change the parameters of the action .
the quantity of interest are susceptibility and specific heat , as well as other order parameters , which we will describe below .
is worthwhile compare the present method with the simulations of such theories on the lattice . in general the simulations of scalar theories on fuzzy spaces
are slower than in their lattice counterparts since the fuzzy models due to the self - interaction term @xmath0 are intrinsically non local and for higher power of the self - interacting the number of operations to calculate each monte carlo step @xmath11 grows even faster .
however , we can expect some advantages of the fuzzy approach with the simulations of other field theories , in particular the fermionic theories or super symmetric models @xcite .
beside another recent field of application the matrix models are the yang - mills actions @xcite in which appear the concept of the emerging geometry .
the paper is organized as follows . in the next section
will be introduce the fuzzy disc and its properties . the real scalar field theory on the fuzzy disc
will be presented in section [ se : scafieldtheo ] .
the simulation and the result are in sect .
[ se : simulations ] , where also the phase boundaries between the three phases present and their scaling properties are derived .
finally , the results are summarized and discussed in the conclusions .
the construction of the fuzzy disc @xcite is based on a quantization of the function on the plane , i.e. the association to a function of an operator on an infinite dimensional hilbert space @xmath12 .
consider the hilbert space generated by the countable basis @xmath13 with @xmath14 positive or zero .
consider the operator @xmath15 and its hermitian conjugate @xmath16 acting on the basis as &= & , + ^ & = & [ aad ] , with the commutation rule : = 11 .
one recognizes of course the usual creation and annihilation operator of single harmonic oscillator , with a slightly unusual normalization ( in the following we will revert to the usual one ) . in terms of the variable of ( [ commrel ] ) .
consider a function on the plane in terms of the conjugate variables @xmath17 with @xmath18 we define a quantization map which associates to a function an operator @xmath9 according to the following rule : f(z,|z)=_m , n=0^f_mn^tay|z^m z^n = _ m , n=0^f_mn^tay^m^n .
[ map1 ] here and in the following we will not discuss the issues of convergence of the series , we assume the coefficients are of rapid decrease . a more rigorous construction , based on the coherent states of the heisenberg group and on a weyl - wigner map can be found in @xcite . this quantization associates an operator to a function , the inverse map is f=|f |z . with @xmath19
the usual coherent states @xmath20 . the function associated to the operator
is called the _ symbol _ of the operator .
more generally we can consider operators written in a density matrix notation : = _ m , n=0^f_mn .
[ densitymatrixbasis ] to this operator is associated the function : f(z,|z)=e^-_m , n=0^f_nm .
[ simden ] where the relation between the taylor coefficients @xmath21 and the @xmath22 is f_lk=_q=0^min(l , k)f_l - q k - q^tay .
after quantization if the plane using the map and its inverse we define the subalgebras of finite @xmath23 matrices considering the functions with a truncated expansion , more precisely taking in account only the expansion terms if both @xmath14 and @xmath24 are smaller than a given integer @xmath25 . keeping @xmath26
fixed the limit @xmath27 the truncated algebra tends to the noncommutative plane .
this is obtained with the help of the projection operator ^(n)_=_n=0^n . to this operator
corresponds the function p_^(n)(|z , z)=_n=0^n = e^-_n=0^n = e^-_n=0^n .
this sum can be expressed in terms of incomplete gamma function and gamma function obtaining a radial function : p_^(n)(r,)=.[gamma projectors ] in the limit @xmath28 if @xmath26 is fixed and nonzero the symbol converges pointwisely to the constant function @xmath29 and recovering the noncommutative plane . otherwise if the product @xmath30 is fixed equals to a real constant @xmath31 the limit for @xmath28 of the become : p_^(n)(r , ) = 11_d ( r ) . where the convergence still pointwise .
this limit converges to a step function in the radial coordinate @xmath32 in other words the characteristic function of the disc on the plane .
given the algebra @xmath33 generated by @xmath15 and @xmath16 , a sequence of sub algebras @xmath34 can be defined by : _ ^(n)=p_n a p_n . the effect of this projection on a generic function is : f_^(n)=p_^(n ) p_^(n ) = e^-_m , n=0^n f_nm , namely a truncation of the series expansion .
the full algebra @xmath35 is isomorphic to an algebra of operators however the previous relations shows that @xmath34 is isomorphic to @xmath36 the algebra of @xmath37 rank matrices , on each sub algebra @xmath34 the symbol is then the identity matrix in every @xmath36 .
the name fuzzy disc derives from the cutoff ; for a fixed @xmath25 every function with the same first @xmath38 terms are mapped in the same matrix losing all the informations of the higher orders . in addiction the symbols of the fuzzyfied function are still defined outside the disc of radius @xmath39 , but they are exponentially damped outside the disc .
it s convenient to use a related way is to define the fuzzy disc using an association that maps directly a function defined on the disc of radius @xmath40 to a @xmath38 matrix . is well know that in polar coordinates @xmath41 a base for @xmath42 are the bessel functions of integer order which are the eigenfunctions of the laplacian on the disc with dirichlet boundary conditions .
if @xmath8 is square integrable with respect to the standard measure on the disc @xmath43 it can be expanded in terms of bessel functions : f(,)=_n=-^+_k=1^f^bes_nk e^i n j_n ( ) .
[ bessel ] there is also a fuzzy version of the angle variable in terms of the phase operator @xcite .
it is possible to define a fuzzy laplacian the eigenstates of this fuzzy laplacian will be the analogous of the standard bessel in the sense that the fuzzy bessel are a base for @xmath36 and their symbols tends to the standard bessel for @xmath27 .
the first step is to define the derivations in @xmath44 using the projectors . in @xmath35
we have : _ z f & = & , + _ |z
f & = & - . therefore we define _ z f_^(n ) & & p_^(n)p_^(n ) , + _ |z
f_^(n ) & & -p_^(n ) p_^(n ) .
these operators satisfy the leibniz rule and are a derivations on each @xmath45 . in the same way for the definition of the laplacian operator
we start from the exact expression on @xmath35 : ^2f(z,|z)=4_z_|zf(z,|z)=. we define in each @xmath45 : ^2f_^(n)(z,|z)=p_^(n)p_^(n ) , and then ^2=p_^(n)p_^(n ) .
can be proved that the eigenvalues of this laplacian are the solution of this equation : _ k=1^n+1-n(-1)^n+1-n+k ( ) ^k=0 .
[ lam ] there is a sequence of eigenvalues labeled by @xmath46 where @xmath14 is the order of the correspondent bessel function and @xmath47 indicates the @xmath47-th zero .
this eigenvalues can be calculated numerically from and they appears to converge for @xmath27 to the spectrum of the standard laplacian defined on a disc with dirichlet homogeneous boundary conditions on the edge .
the eigenstates of the standard laplacian are : ^(n)_n , k= ( ) ^n/2_a=0^n - n.where the eigenstates with the negative @xmath14 are hermitian conjugate of @xmath48 . in the fuzzy approximation the continuum eigenfunctions @xmath49 are represented by matrices @xmath50 but the max possible @xmath14 are fixed by the dimension of the fuzzyfication @xmath51 .
the previous construction allows us to define a fuzzyfied version of a field theory on a disc . for the real scalar case with euclidean signature , with the @xmath0 potential ,
the action is given by : s ( ) = ( ^2 + ^2 + ^4 ) ^2z , [ s - com ] with @xmath52 the laplacian on the disc , @xmath53 and @xmath54 are respectively a mass and an interaction parameters .
we choose this particular model because for the plane case it is well studied .
it is known @xcite that the diagrammatic expansion of this theory has only one divergent diagram , the so called tadpole diagram and it is renormalizable . using the fuzzyfication procedure
the action can be approximated by : _ n()= ( ^2 + ^2 + ^4 ) = ( + ^2 + ^4 ) . in this action @xmath55 are finite dimensional hermitian matrices and the product between the field became the standard matrix multiplication , being a finite matrix model can be approached numerically using monte carlo techniques a method currently in use for other fuzzy spaces like the sphere .
note that the fuzzy approximation provides both an infrared and an ultraviolet regularization of the theory .
we can rescale the operators @xmath56 as @xmath57 in order the extract the @xmath26 dependence and recast the action making @xmath26 manifest : _
n()= \{4+(^2 + ^4 ) } .[s - ncom ] it will be useful later to define the potential part of the action given by ( from now to simplify notations we will omit the hats and the subscript @xmath25 , as all fields are matrices ) : s_v ( ) = ( ^2 + ^4 ) [ par-1 ] . in the simulation we will use another equivalent choice of the parameters often used for the potential part namely : s^_v ( ) = r(^^2 + ) [ par-2 ] .
the two choices are related by r&= & , + r&=&. since the scalar field on either commutative or fuzzy disc is defined on a finite volume it is impossible to have a phase transition , however phase transitions may be found when the matrix dimension or the radius of the sphere become infinite . as already states the fuzzy disc arise from the relation @xmath58 where @xmath26 is the parameter of non - commutativity and @xmath40 is the radius of the disc . introducing in the non - commutativity parameter a dependence on the matrix size @xmath25 and performing different limits we have three different cases : * @xmath59 with @xmath39 fixed . in this case
we recover the commutative disc of radius @xmath40 . in the following
we will take @xmath60 . * @xmath59 with @xmath61 and @xmath26 fixed . in this case
we recover the noncommutative plane
. we will take @xmath62 and therefore @xmath63 .
* @xmath59 , @xmath64 and @xmath61 . in this case
we recover the commutative plane . in the following we will the @xmath65 which implies @xmath66 , in the limit @xmath67 we recover the commutative plane , where @xmath68 is a real number varying between @xmath69 , our choice is just dictated by simplicity . ] . in the simulation we expect at least one critical line corresponding to the critical behavior of the continuous scalar field on the plane .
indeed one of the main aim of the simulation will be to find the phase diagrams for the fuzzy field theory .
using monte carlo method we will produce a sequence of configurations @xmath70 and evaluate the average of the observables over the set of configurations .
these sequences of configurations , called monte carlo chain , are representatives of the configuration space at given parameters . in this framework
the expectation value is approximated as o_j=1^t_mco_j [ expet ] , where @xmath71 is the value of the observable @xmath72 evaluated in the @xmath73-sampled configuration , @xmath74 , @xmath75 $ ] .
the internal energy is defined as e(n,,)= s , it is also very useful to compute separately the average values of the two contributions : @xmath76 where @xmath77 is defined as : s_d ( ) = 4 \{}. the specific heat takes the form : c(n,,)= s^2- s^2 .
these quantities correspond to the usual definitions for energy e(n , , ) = - and specific heat @xmath78 where @xmath79 is the partition function .
another important observable used to look for a phase transition is finite - volume susceptibility defined as : = ( ) ^2- |()|^2 .
[ sus ] the phase transitions are located at the values the previous quantities show a peak when varying the parameters .
note that in comparing the with the usual susceptibility the second term is contains @xmath80 , otherwise due to the fact that symmetry causes @xmath81 there will be no peak even in presence of a phase transition .
the simulations were conducted following the standard way using a metropolis algorithm with the jackknife and binning methods @xcite in order to evaluate the errors on the expectation values .
the usual update algorithm , used to generate a new configuration , changes all the entries of the independent matrices : @xmath82 where @xmath25 is the matrix size and @xmath83 are random real numbers .
the amplitude of variation of each matrix entry is tuned dynamically during the run time , requiring an acceptance rate between @xmath84 and @xmath85 .
the new proposed configurations are judged by the metropolis algorithm , which requires to compute the difference between the action for the new proposed configuration and the action for the configuration at the current time step @xmath86 .
the probability of accepting the new configuration @xmath87 is calculated as @xmath88 . in order to reduce the computation time , instead a global matrix change ,
we have changed just one matrix entry for each monte carlo step . the coefficient to update
is not chosen randomly but following an order used to vary all the coefficients after @xmath89 time steps . the interaction term proportional to @xmath0 of the potential in the action
couples a matrix entry @xmath90 to all the other elements on its line and column .
this means that using the previous described update procedure , the computation of the variation of the action @xmath11 ( and consequently the metropolis check ) needs a number of operations which increases linearly with the matrix size .
the drawback of this approach are mainly two : the complication of the code and the dependence of the correlation time @xmath91 in @xmath25 .
in fact two configurations will share at least one coefficient until @xmath89 monte carlo steps , thus this correlations introduce an increase of the correlation time increasing @xmath25 and presumably @xmath91 is proportional to @xmath89 .
however , using this optimization and the implementation of parallel computing we were able to execute our simulation for @xmath92 with a good precision and in reasonable time despite our limited computation resource .
the initial conditions of the the markov chain are chosen randomly in the configuration space although we used two general types ; hot initial conditions , which are configurations far from the minimum and cold start conditions , which correspond to configurations close to the minimum .
as usual we have excluded all the time step before the thermalization process take place .
the previous introduced quantities are not sufficient if we want to measure the various contributions of different modes of the fields to the configuration @xmath93 .
we need some control parameters , usually called order parameters . recalling the fuzzyfication procedure we associate to a function on the disc a matrix given by coefficients @xmath94 in the density matrix base expansion of ( [ densitymatrixbasis ] ) : ( z,|z)=e^-_m , n=0^_nm = e^-_m , n=0^_nm,[dens - r ] due to the linearity of the expansion and since the action is even
, the coefficients @xmath95 have the property to average to zero . as a first idea
, we can think about a quantity related to the square norms of the fields , for example the sums @xmath96 .
this quantity is called the full - power - of - the - field ; it can be computed as the trace of the square : ^2_a = ( ^2 ) [ vara ] the quantity @xmath97 alone is not a good order parameter because it does not distinguish contributions from the different modes , but we can use it as a reference to define the quantity : @xmath98 it is easy to see that the quantity is connected with the purely radial contribution , thus can be used to analyze the pure radial contribution to the full - power - of - the - field .
we will see that @xmath99 defines a disordered phase while @xmath100 defines a kind of ordered regime . in order to describe
the type of ordering it is necessary to study the contribution from higher modes to @xmath101 .
we can generalize the previous quantity and define parameters @xmath102 in such a way that they form a decomposition of @xmath101 : ^2_a=^2_0+_l>0 ^2_l;. following this prescription , the other quantities for @xmath103 can be defined as : @xmath104 if the contribution is dominated by the radial symmetric parameter we expect to have @xmath105 . in the next simulations we will evaluate , apart from @xmath106 , the quantity with @xmath107 as representatives of those contributions where the rotational symmetry is broken . according to we have ^2_1 = ^n-1_n=0(|_n , n+1|^2+|_n+1,n|^2 ) .
using higher @xmath108 in we could analyses the contributions of the remaining modes , but it turns out that the measurements of the first two modes are enough to characterize the behavior of the system .
before we start let us remind of some aspects of the 2-dimensional @xmath0 model on a fuzzy sphere .
many simulations have focused on scalar fields on fuzzy spheres and they lead show the presence of new phases among the classical ones , characterized as nonuniform ( or matrix or striped ) ordered phase @xcite .
numerical studies show three different phases : a disordered phase and an ordered regime , which splits into phases of uniform and nonuniform order . in the nonuniform phases , which is not present in the commutative planar @xmath109 theory ,
the rotational symmetry is spontaneously broken .
the three transition curves intersect at the so called triple point . analyzing the limit @xmath67 limit ( which for the fuzzy sphere leads to the noncommutative plane ) is been found that the three curves and the triple point collapse using the same scaling function of @xmath25 .
therefore , the triple point , all three phases and in particular the new nonuniform phase , survive in the limit .
this means that the naive scalar field action can not be used as an approximation of the scalar field theory on the commutative 2-sphere . using a perturbative approach @xcite
it is possible to prove that in the limit of the commutative sphere the two point function is non affected by the uv / ir mixing but again the fuzzyfication does not commute with the commutative limit . taking a different limit to the noncommutative 2-plane the uv / ir mixing reappears .
so we can see the appearance of the nonuniform phase as a finite version of the uv / ir mixing .
these nonuniform phases are closely connected with spontaneous symmetry breaking @xcite , this appearance of the new phase seems in contradiction with the coleman - mermin - wagner theorem which states that there can be no spontaneous symmetry breaking of continuous symmetry on 2-dimensional commutative spaces
. however the theorem is strongly based on the locality of interactions therefore is no direct generalization of the theorem for the noncommutative case , where the theory is nonlocal . in the work of gubser and sondhi @xcite ,
where those new phases were originally conjectured , it is argued that this translation non - invariant phase is possible only in dimensions @xmath110 .
nevertheless , subsequent numerical simulation , in particular of ambjorn and catterall @xcite , show the existence of such a phase even in @xmath111 .
now we are ready to discuss the result of the monte carlo simulations and the various phases which arise in our model . in the following , for the first phase transitions
, we use parameters @xmath54 or @xmath53 of . fig .
[ figure1 ] shows the susceptibility plots for a fixed values of @xmath112 , varying the matrix size and for all the limits described .
the location of a maximum in the susceptibility is used to identifies the critical value related to the phase transition for the continuous model . in fig .
[ figure1 ] the maximum of the three susceptibility in the @xmath92 case can be easily extrapolated for @xmath113 negative equals to @xmath114 respectively for the noncommutative plane ( * ncp * ) , commutative disc ( * cd * ) , commutative plane ( * cp * ) limit . in general
we find the critical parameter @xmath113 for negative values and for @xmath115 . ) , for @xmath112 and @xmath116 .
_ , title="fig : " ] ) , for @xmath112 and @xmath116 .
_ , title="fig : " ] ) , for @xmath112 and @xmath116 .
_ , title="fig : " ] this critical behavior shows up in the specific heat density as well , in fact the specific heat can be used as an alternative in order to determine where the phase transition take place .
actually both quantities displays the same critical value but , comparing fig .
[ figure1 ] and fig .
[ figure2 ] , we find a small difference in the critical value of @xmath113 , likely due to a finite volume effect . in the following
we adopt the the susceptibility criterion in order to find the transitions curves because it is easier to estimate the maximum of susceptibility respect to estimate the critical parameter through the specific heat step . beside using the same statistic
the errors are smaller for the susceptibility than in the case of the specific heat since in general more statistics is required for the specific heat to have the same precision . from the slopes of the specific heats and of the the susceptibility we can infer information about the order of the phase transition . in fact this phase transitions turn out to be of the second order for all the three cases . and @xmath92 _ , title="fig : " ] and @xmath92 _
, title="fig : " ] and @xmath92 _ ,
title="fig : " ] in order to gain some informations on the nature of the phases involved in the transitions we look at the order parameters defined in the previous section . in fig .
[ figure3 ] are plotted the expectation values of the order parameters .
looking for values greater than the critical point @xmath117 we find the @xmath118 and @xmath119 much smaller than the full power of the field @xmath120 with a weak dependence on @xmath53 . and
@xmath92 _ , title="fig : " ] and @xmath92 _ , title="fig : " ] and @xmath92 _ , title="fig : " ] the behavior suggest that all the modes contributes effectively to the full power of the field as random fluctuations .
this type of phase is usually named disordered phase due to the contribution of the complete composition of the field by random fluctuations around the null constant average field . for smaller values of @xmath121 with respect to respect the critical values
@xmath122 we find a phase characterized by @xmath123 , the pure radial contribution @xmath118 is dominant while @xmath124 thus only one mode ( the zero one ) is mainly present in the field decompositions governing its behavior . in contrast with the previously described phase , where the full powers of the fields are basically a sums of random fluctuations , this ones is named uniform phase . in this regime
the our fields are essentially diagonal matrices and the kinetic action part of the action by the regions around its minima at @xmath125 .
these results are completely consistent with what one would expect for the usual plane so we can state that all the three limits possess the disordered - uniform phase transition observed for a @xmath0 scalar field theory on the commutative plane . _ ,
title="fig : " ] _ , title="fig : " ] _ , title="fig : " ] repeating the previous analysis for different values of @xmath54 and finding the corresponding critical values @xmath113 we can plot the transitions curves disorder - uniform ; the scaled plots are shown in fig .
[ figure4 ] .
all the graphs possess good scaling properties even for small matrix sizes without strong finite dimensional effects .
each plot has a different scaling coefficient for the noncommutative plane limit the lines collapse for a scaling like the square root of the matrix size , for the commutative disc limit the best collapse take place for @xmath126 and for the commutative plane for @xmath127 .
furthermore the phase transition line is well approximated by a linear function therefore performing a fit of the data we obtains the approximations : ^*ncp*_c(n,)-0.29 n^-0.5 , ^*cd*_c(n,)-0.49 n^-0.33 , ^*cp*_c(n,)-0.43 n^-0.3 .
the previous linear regression has been done on the highest matrix size data set @xmath128 , which is our best approximation to the large @xmath25 limits . , for @xmath129 and @xmath116 .
_ , title="fig : " ] , for @xmath129 and @xmath116 .
_ , title="fig : " ] , for @xmath129 and @xmath116 .
_ , title="fig : " ] we see that the two commutative case , * cd * and * cp * behave in a similar way among themselves , and are different from the noncommutative plane limit . beside the uniform and disorder phases
the simulations show the appearance of a new phase , in this case it is convenient to adopt the other choice of parameters .
we find this new phase transition for large @xmath32 , fig .
[ figure5 ] show an example of this phase transition for fixed @xmath130 and varying @xmath53 . again
the critical points can be easy located using the maximum of the susceptibility . as we can see the locations of the peaks for all the plots
are pushed to the left increasing @xmath25 in particular if we compare the plot concerning the non - commutative limit and the other two , we observe a bigger shift of the locations of the critical points to the left of the two commutative limit plots respect the noncommutative one .
we will see this difference clearly in the transition curves . looking at fig .
[ figure6 ] and @xmath92 _ , title="fig : " ] and @xmath92 _ , title="fig : " ] and @xmath92 _ , title="fig : " ] the new phase appears characterized , for @xmath131 , by @xmath132 and @xmath133 , this is true for all the limits we have investigated , therefore we are dealing with the same phase .
in particular we notice that in the regions @xmath131 the full power of the fields becomes almost linear , after various fits on the data we obtain for all limits : ^2_a ( -2 ) .
[ full - fit ] the factor @xmath134 is a normalization .
we can generalize this result introducing another parametrization analogous to in witch we reintroduce @xmath54 but just as fixed parameter : s_v ( ) = r(^2 + ^4 ) . in this case
we have for the correspondent full power of the field the fit : ^2_a ( - ) .
[ full - fit - lam ] we recast our model in this way in order to show the analogies with the fuzzy sphere @xcite , for which equation is very similar .
the behavior of the order parameters suggests that in the new phase the eigenvalues of the field are in a minimum of the potential .
following @xcite the hint that the neighborhood of the minima of the potential mainly contribute to the expectation values .
the minima of the potential @xcite are @xmath25 disjoint orbits of the form o_n=
u^(1_n -1 _
n -n ) u | u u(n)/[u(n ) u(n - n ) ] , where @xmath135 and @xmath136 are the unit matrices . in the case of a negligible kinetic term
, the orbit with the largest phase space volume will dominated the field composition . in the previous case of the uniform phase ,
the minimum of the potential is clearly found minimizing the kinetic term as well , but in the new phase there is not a direct hint that the kinetic term might be negligible . to test this assertion
we have studied the partial contributions of the kinetic part @xmath137 and the potential part @xmath138 to the total energy . and @xmath137 of the internal energy @xmath139 in the different limits for for @xmath130 and @xmath140 .
_ , title="fig : " ] and @xmath137 of the internal energy @xmath139 in the different limits for for @xmath130 and @xmath140 .
_ , title="fig : " ] and @xmath137 of the internal energy @xmath139 in the different limits for for @xmath130 and @xmath140 .
_ , title="fig : " ] as fig
. [ figure7 ] shows , our results confirm that for @xmath131 the transition the average kinetic contribution is much smaller than the modulus of the potential .
however , this behavior is more marked for more negative @xmath53-values in the uniform order regime as well , a fact that does not allow to interpret the rising of the this new phase caused only from the potential dominance .
this new phase was already found for the same scalar field theory on the fuzzy sphere and is usually referred as non - uniform phase or matrix phase @xcite . .
_ , title="fig : " ] .
_ , title="fig : " ] .
_ , title="fig : " ] fig .
[ figure8 ] shows the scaled non - uniform - disordered critical lines , each curves scale in the matrix size with a different scaling factor ; for the noncommutative plane limit the curves collapse for a scaling like for @xmath141 , for the commutative disc limit the best collapse take place for @xmath142 scaling behavior and for the commutative plane for @xmath143 .
actually the scalings of of the commutative limits are not true scalings , in the sense that the collapse takes place for @xmath32 large enough , in fact looking closely at the data set fig .
[ figure9 ] we notice that the point towards the origin does not collapse .
we will discuss this issue later . and @xmath144 _ , title="fig : " ] and @xmath144 _ , title="fig : " ] the asymptotic behavior for large @xmath32
can be estimated algebraically @xcite from the scaled susceptibility curves and the critical point for the pure potential model , in fact for @xmath145 and @xmath146 _ c ( r ) . furthermore using the scaling factors the asymptotic phase transition lines we have the approximations : ^*ncp*_c(n , r ) n^-0.8 , ^*cd*_c(n , r ) n^-1.2 , ^*cp*_c(n , r ) n^-0.9 .
this large @xmath32 behavior has an asymptote in zero and fits the scaled critical lines well for @xmath145 . _ ,
title="fig : " ] _ , title="fig : " ] in fig .
[ figure10 ] we compare the non scaled transition curves of the noncommutative limit case and the other commutative ones , in both cases the difference became bigger for large @xmath25 commutative transition curves , the difference is much more evident for the commutative disc .
the reason of this discrepancy is the explicit @xmath26 dependence of the potential term in the action .
the noncommutative limit is characterized by @xmath62 making the model very similar to the one obtained using the fuzzy sphere in which the potential part has no dependence on the noncommutative parameter .
switching to the other to cases we introduce in the pre factor a dependence on the matrices size , increasing @xmath25 for fixed @xmath32 the effect is to lower the potential dominance and graphically to lower the transition curve . for each @xmath25
we will find the previous @xmath147 shifted towards lower values and since we have for all the cases an finite asymptotic behavior , in the commutative limits we can lower any point of the transition curves just increasing @xmath25 .
the scalar field action on the fuzzy disc , in analogy with equivalent simulations on the fuzzy sphere , shows the occurrence of a new nonuniform phase , peculiar of the noncommutative models .
we found the existence of three behaviors in the model described by which arise for all the limit performed .
we studied the scaling properties of the transition curves disordered - ordered obtaining a scaling coefficient in each case , the noncommutative case has the same coefficient as the simulations on the fuzzy sphere scaling like the square root of the matrix dimension , the other two cases show different scaling behavior .
nevertheless looking at the order parameters in all three cases we can recognize the two phases as the disordered an the ordered ones .
the critical line for the non - uniform - disordered phase transition is also found and the noncommutative limit scales as @xmath141 and grows asymptotically for large @xmath53 parameters . in the other two commutative limits it is possible to consider a good data collapse using our matrix size approximations only for @xmath53 large enough .
the asymptotic behaviors of are well approximated for large @xmath32 by a pure potential model .
we have confirmed that even for this models that the non - uniform phase is connected to the dominance of the minimized potential over the kinetic term via the phase space volume of the largest orbit .
the geometry for a scalar field theory , defined on a fuzzy space , is defined by the laplacian operator which correspond to a particular kinetic term of the action , thus the new phase should be insensitive on the underlying geometry . in particular , this argument justifies the similar behavior of the noncommutative limit comparing to the results found for the fuzzy sphere studied in @xcite .
a substantial difference from the fuzzy sphere case is in the commutative limits , in fact the @xmath26 dependence of the potential part in each two cases modify the data collapse of the transition curves disordered non - uniform making a good collapse possible only for large @xmath32 respect @xmath25 . in the fuzzy sphere models
the usual scalar field action can not be used as an approximation of the scalar field theory on the commutative sphere due the persistence of the non ordered phase in the continuous limit .
the problem is patched with an additional term used to increase the weight of the kinetic part or equivalently to decrease the influence of the potential term in order to suppress the anomalous phase @xcite . in our model ,
considering the commuting limits , the action of the @xmath26 on the potential term act naturally reducing the ratio between the potential part and the kinetic one , so we can surmise that in the continuous limits the non - uniform phase does not exist .
the ability to differentiate the commutative and noncommutative cases is one of the main advantages of the method described in this paper .
f. lizzi acknowledges support by cur generalitat de catalunya under project fpa2010 - 20807 and the _ faro _ project _ algebre di hopf , differenziali e di vertice in geometria , topologia e teorie di campo classiche e quantistiche _ of the universit di napoli _ federico ii_. g. landi , _ an introduction to noncommutative spaces and their geometries _ , _ springer lecture notes in physics 51 _ , springer verlag ( berlin heidelberg ) 1997 .
arxiv : hep - th/9701078 .
gracia - bondia , j.c .
varilly , h. figueroa , _ elements of noncommutative geometry _ , birkhauser , 2000 .
j. madore , `` the fuzzy sphere , '' class .
* 9 * ( 1992 ) 69 .
a. p. balachandran , s. kurkcuoglu and s. vaidya , `` lectures on fuzzy and fuzzy susy physics , '' singapore , singapore : world scientific ( 2007 ) 191 p. [ hep - th/0511114 ] .
j. medina , `` fuzzy scalar field theories : numerical and analytical investigations , '' arxiv:0801.1284 [ hep - th ] .
j. medina , w. bietenholz , f. hofheinz and d. oconnor , pos lat * 2005 * ( 2006 ) 263 [ hep - lat/0509162 ] . h. grnewold , `` on the principles of quantum mechanics '' , physica * 12 * ( 1946 ) 405 .
j. e. moyal , `` quantum mechanics as a statistical theory , '' proc .
cambridge phil .
* 45 * ( 1949 ) 99 .
s. galluccio , f. lizzi and p. vitale , `` twisted noncommutative field theory with the wick - voros and moyal products , '' phys .
d * 78 * ( 2008 ) 085007 [ arxiv:0810.2095 [ hep - th ] ] .
j. zinn - justin , _ quantum field theory and critical phenomena _ , oxford university press , second edition ( 1993 ) .
b. ydri , `` impact of supersymmetry on emergent geometry in yang - mills matrix models ii , '' int .
j. mod .
a * 27 * ( 2012 ) 1250088 [ arxiv:1206.6375 [ hep - th ] ] .
r. delgadillo - blando , d. oconnor and b. ydri , `` geometry in transition : a model of emergent geometry , '' phys .
* 100 * ( 2008 ) 201601 [ arxiv:0712.3011 [ hep - th ] ] .
b. spisso , r. wulkenhaar,``a numerical approach to harmonic non - commutative spectral field theory , '' int .
j. mod .
a * 27 * ( 2012 ) 1250075 [ arxiv:1111.3050v4 [ math - ph ] ] .
b. spisso , `` a numerical approach to harmonic non - commutative spectral field theory '' [ arxiv:1111.2871v1 [ hep - th ] ] .
f. lizzi , p. vitale and a. zampini , `` the fuzzy disc , '' jhep * 0308 * ( 2003 ) 057 [ hep - th/0306247 ] .
a. p. balachandran , k. s. gupta and s. kurkcuoglu , `` edge currents in noncommutative chern - simons theory from a new matrix model , '' jhep * 0309 * ( 2003 ) 007 [ hep - th/0306255 ] .
f. lizzi , p. vitale and a. zampini , `` the beat of a fuzzy drum : fuzzy bessel functions for the disc , '' jhep * 0509 * ( 2005 ) 080 [ hep - th/0506008 ] .
f. lizzi , p. vitale and a. zampini , `` from the fuzzy disc to edge currents in chern - simons theory , '' mod . phys
. lett .
a * 18 * ( 2003 ) 2381 [ hep - th/0309128 ] .
f. lizzi , p. vitale and a. zampini , `` the fuzzy disc : a review , '' j. phys .
* 53 * ( 2006 ) 830 .
a. zampini , `` applications of the weyl - wigner formalism to noncommutative geometry , '' hep - th/0505271 .
s. kobayashi and t. asakawa , `` angles in fuzzy disc and angular noncommutative solitons , '' arxiv:1206.6602 [ hep - th ] . s. minwalla , m. van raamsdonk and n. seiberg , `` noncommutative perturbative dynamics , '' jhep * 0002 * ( 2000 ) 020 [ hep - th/9912072 ] . m. chaichian , a. demichev and p. presnajder , `` quantum field theory on noncommutative space - times and the persistence of ultraviolet divergences , '' nucl .
b * 567 * ( 2000 ) 360 [ hep - th/9812180 ] .
m. e. newman and g. t. barkema , _ monte carlo methods in statistical physics _ , oxford university press ( 2002 ) .
w. janke , `` statistical analysis of simulations : data correlations and error estimation '' , published in quantum simulations of complex many - body systems : from theory to algorithms , lecture notes john von neumann institute for computing , jlich , nic series , vol .
10 , isbn 3 - 00 - 009057 - 6 ( 2002 ) 423 - 445 .
x. martin , `` a matrix phase for the @xmath0 scalar field on the fuzzy sphere , '' jhep * 0404 * ( 2004 ) 077 [ hep - th/0402230 ] .
m. panero , `` numerical simulations of a non - commutative theory : the scalar model on the fuzzy sphere , '' jhep * 0705 * ( 2007 ) 082 [ hep - th/0608202 ] .
f. garcia flores , x. martin and d. oconnor , `` simulation of a scalar field on a fuzzy sphere , '' int .
j. mod .
a * 24 * ( 2009 ) 3917 [ arxiv:0903.1986 [ hep - lat ] ] .
m. panero , `` quantum field theory in a non - commutative space : theoretical predictions and numerical results on the fuzzy sphere , '' sigma * 2 * ( 2006 ) 081 [ hep - th/0609205 ] .
b. p. dolan , d. oconnor and p. prenajder , `` matrix @xmath0 models on the fuzzy sphere and their continuum limits '' , jhep * 03 * ( 2002 ) 013 [ hep - th/0109084 ] .
das , s. digal , t.r .
govindarajan , `` spontaneous symmetry breakdown in fuzzy spheres '' , mod.phys.lett.a24 [ arxiv:0801.4479v2 [ hep - th ] ] | the fuzzy disc is a discretization of the algebra of functions on the two dimensional disc using finite matrices which preserves the action of the rotation group .
we define a @xmath0 scalar field theory on it and analyze numerically for three different limits for the rank of the matrix going to infinity .
the numerical simulations reveal three different phases : uniform and disordered phases already the present in the commutative scalar field theory and a nonuniform ordered phase as a noncommutative effects .
we have computed the transition curves between phases and their scaling .
this is in agreement with studies on the fuzzy sphere , although the speed of convergence for the disc seems to be better .
we have performed also three the limits for the theory in the cases of the theory going to the commutative plane or commutative disc . in this case
the theory behaves differently , showing the intimate relationship between the nonuniform phase and noncommutative geometry .
iccub-12 - 312 * noncommutative field theory : + numerical analysis with the fuzzy disc * fedele lizzi@xmath1 and bernardino spisso@xmath2 + @xmath2_dipartimento di scienze fisiche , universit di napoli _ federico ii _ _ + @xmath3_infn , sezione di napoli _ + _ monte s. angelo , via cintia , 80126 napoli , italy _
+ @xmath4 _ departament de estructura i constituents de la matria , + institut de cincies del cosmos , universitat de barcelona , + barcelona , catalonia , spain _
+ [email protected] , [email protected] 2 cm |
graphene , a two - dimensional honeycomb lattice of carbon atoms , has attracted intense attention in the past few years.@xcite its properties bear some similarities with , and some striking differences from , conventional 2d electron gas ( 2deg ) systems found in semiconductor heterostructures .
it is well - known that the latter has a rich phase diagram in the quantum hall regime .
when @xmath5 , the average inter - electron distance measured in units of bohr radius , is not very big , there are integer and fractional quantum hall liquid states , as well as charge density waves ( cdws ) of various forms , including wigner crystals of quasi - electrons , bubbles and stripes at fillings around these liquid states , and analogous particle - hole conjugates of these states @xcite . in high magnetic fields , the particular state is essentially determined by the filling factor @xmath3 , defined as the ratio of the electron density to the density of magnetic flux quanta penetrating the plane . when @xmath5 is increased , these quantum hall phases undergo transitions to wigner crystal ( wc ) states with a single electron per unit cell .
( for very small @xmath3 , there may also be wigner crystals of composite fermions @xcite . )
if the phase diagram is plotted in the @xmath6 ( density ) - @xmath2 ( magnetic field ) plane , away from the origin , there is a fan of quantum hall phases , but the origin is expected to be completely surrounded by wigner crystal states @xcite [ see fig .
[ fig:2deg ] ] .
the integer quantized hall effect has been observed in graphene @xcite , and , except for a well - understood shift in the precise values of the plateaus @xcite , the hall conductance appears rather similar to that found in the conventional 2deg .
nevertheless , the behavior of clean and cold graphene in the low doping limit is likely to be different than that of the conventional 2deg . unlike the latter , non
- interacting electrons in graphene to a good approximation obey a massless dirac equation @xcite , with two inequivalent dirac points in two different valleys ( denoted as @xmath7 and @xmath8 ) in the brillouin zone .
when the system is undoped the fermi energy passes directly through these dirac points . with interactions , continuum @xcite and tight - binding @xcite studies of the this system in mean - field theory indicate that the system remains in a liquid state in zero magnetic field even at arbitrarily low doping . on the other hand
, hartree - fock calculations @xcite and exact diagonalization studies @xcite suggest that cdws are possible in a large magnetic field where states are restricted to a single or two @xcite landau levels ( lls ) and that the phase diagram is similar to that of the conventional 2deg . in this paper
we address the question of how the system passes from these strong - field states into the liquid state as the field and density are lowered to small values . for the conventional 2deg , the quantum hall states give way to the wc in the low - field , low density limit due to landau level mixing ( llm ) .
this allows the electrons to form wavepackets that are more localized than is possible within a single landau level , thereby lowering the interaction energy @xcite .
the degree of llm is determined by a coupling constant @xmath9 , the ratio of the typical coulomb interaction energy @xmath10 to the scale of the ll separation . for both graphene and the conventional 2deg ,
@xmath10 is given by @xmath11 , where @xmath12 is the magnetic length .
however , the ll separations are different in the two cases . in the conventional 2deg ,
it is given by @xmath13 , so that @xmath14 and in the large @xmath2 limit where @xmath9 is small , llm is negligible . in graphene ,
the lls are not equally spaced @xcite , so we instead characterize it by the gap between the @xmath15 and @xmath16 lls divided by @xmath17 , @xmath18 , where @xmath19 is the fermi velocity . then @xmath20 is a field - independent constant @xcite , typically estimated to be of order 1 or smaller .
nevertheless , even though @xmath9 is field - independent , the degree of llm can change with @xmath2 even for fixed filling factor , and we shall see below that it in fact does , albeit by a small amount . this is because in addition to positive energy levels , the dirac equation in a magnetic field supports negative energy landau level states , as well as a zero energy ll , for each spin , provided the zeeman energy is neglected .
moreover , the low energy theory of graphene involves two such dirac points ( @xmath7 and @xmath8 valleys ) , so there are two copies of these energy levels in the spectrum . when undoped , all the negative energy states are filled , as well as one of the two zero - energy states @xcite .
added electrons interact with the electrons in these filled levels , which changes the _ effective _ energy of the higher lls .
because the landau level structure of these filled levels varies with field , the splitting between the @xmath15 and @xmath16 energy levels does not precisely the follow the @xmath21 behavior discussed above . in a continuum description ,
the filling of the negative levels is characterized by a ( negative ) integer @xmath22 , which denotes the lowest ll which must be filled to accommodate one electron per atom , the density of mobile electrons of undoped graphene @xcite .
an extra field dependence thus enters the problem through @xmath22 , and is given by @xmath23 where @xmath24 is the area of the sample , @xmath25 is the lattice constant of the triangular ( bravais ) lattice and the factor of 4 in the denominator comes from the spin and valley degeneracies .
the interaction of electrons with those in the negative levels is most easily described in the hartree - fock ( hf ) approximation , where it appears as a contribution to the exchange self - energy . for uniform liquid states ,
we find that the coulomb energy decreases faster with decreasing b than the difference in effective energy between the highest occupied level and the lowest unoccupied level , so that the ratio between kinetic and potential energy actually _ increases _ with decreasing @xmath2 , because @xmath26 increases . we will demonstrate a similar effect for stripe states , and believe it should be ubiquitous for charge - ordered and liquid quantum hall states . because this effect is a result of filling @xmath26 negative lls
, it is a concern that it may be an artifact of the cutoff procedure used in our hf calculations . to check this
, we performed an analogous calculation for interacting electrons in a tight - binding model , where no artificial cutoff needs to be introduced .
we obtain results from this model that are very similar to those of the continuum model .
the consequence of this is that , for states where llm is small at large values of @xmath2 , we expect it remain small , and even _ decrease _ , with decreasing @xmath2 . while it is not immediately obvious that with @xmath27 one should find weak llm in these quantum hall states , this does appear to be the case for wc and bubble states @xcite , and as we demonstrate below , for stripe and uniform liquid states .
the surprising result is that , within the hartree - fock approach , one expects these states to persist to arbitrarily small field .
thus , many different states persist down to the origin of the phase diagram in @xmath6 - @xmath2 plane [ see fig .
[ fig : graphene ] ] .
because these states follow trajectories of fixed @xmath3 in the plane , the density of electrons participating these cdw states decreases with @xmath2 , such that their amplitude scales roughly as @xmath2 , and the wavelength as @xmath28 .
the stripe states , and by analogy other cdw states , disappear continuously as @xmath4 , eventually becoming indistinguishable from a uniform liquid state in the low field limit .
nevertheless , in principle , for a clean , undeformed graphene system , this implies that in principle many states emanate from the @xmath29 point in the phase diagram .
we note that this behavior is very specific to the @xmath30 form for coulomb interactions that is natural in this system . for shorter range interactions , where a length scale other than the magnetic length is involved in the interaction range , the effective value of @xmath9 will increase with decreasing field as in the standard 2deg , at low densities and fields llm should destabilize the high field states , and a wc state
should result .
such a situation could arise if a metallic gate is sufficiently close to the graphene plane to effectively screen the long - range component of the coulomb interaction .
more generally , the behavior discussed here may be understood as being a consequence of the marginal nature of the @xmath30 coulomb interaction in undoped graphene . as has been shown by us
elsewhere @xcite , the energy difference between landau levels near the fermi energy is increased by the filled fermi sea , by an amount proportional to @xmath31 .
this logarithmic increase of the ll spacing with increasing @xmath26 can be reinterpreted as the fermi velocity being renormalized upwards as the high - energy cutoff of the theory is increased @xcite . that the ll spacing
_ increases _ slightly with increasing @xmath26 as the doping is decreased is consistent with interactions being marginally irrelevant in this system @xcite .
had it decreased instead , the interactions would be marginally relevant , one would expect to find a wc state near the origin of the @xmath32 phase diagram .
this paper is organized as follows . in section
[ sec : continuum ] , we describe the continuum limit hamiltonian , and the hartree - fock approximation used to study quantum hall states in the presence of llm . in section [ sec : result ] , we discuss the results of these continuum calculations . in section
[ sec : hubbard ] , we introduce a tight - binding model with hubbard interactions , and demonstrate that the suppression of llm found in the continuum calculations is not an artifact of our cutoff procedure .
finally , we conclude with a summary in section [ sec : sum ] .
in standard 2deg s , it is known that the hartree - fock approximation is quite reliable for electronic states in high landau levels @xcite .
the situation should be similar for graphene , particularly if one can show that llm is small for states in high lls , as we will indeed find self - consistently below .
we thus adopt the hartree - fock approximation for the states we study .
more specifically , our hartree - fock approach to the dirac equation description of uniform and stripe quantum hall phases in graphene is adapted from a procedure developed for electrons in a standard 2deg @xcite ; in what follows we briefly outline the method , and highlight the ( largely technical ) differences .
the hamiltonian for the system in a magnetic field is @xmath33 where the numbers denote composite indices for the different quantum numbers specifying the states [ e.g. , @xmath34 ( ll index , guiding center coordinate , spin , pseudospin ( valley ) index ) ] , @xmath35 is the ll spectrum plus the zeeman energy , and @xmath36 are matrix elements for the coulomb interaction .
@xmath37 is related to standard matrix elements @xcite @xmath38 with @xmath39 and @xmath40f_{n_1,n_4}(\bm{q})\delta_{x_1,x_4+q_y\ell^2},\end{aligned}\ ] ] where @xmath41 for @xmath42 , where @xmath43 is the generalized laguerre polynomial .
note that @xmath44^{\ast}$ ] . in terms of @xmath45
, @xmath46 takes the form @xmath47.\end{aligned}\ ] ] note that the guiding center coordinates ( @xmath48 ) have been suppressed in the subscripts in eqs .
( [ eq : vtilde ] ) and ( [ eq : w ] ) .
the density matrix operators are defined as @xmath49 this relation may be inverted to obtain the expectation value of an arbitrary single - particle operator in terms of density operator expectation values , @xmath50\delta_{x , x'-p_y \ell^2}.\end{aligned}\ ] ] for states with discrete translational symmetry , the sum over @xmath51 is restricted to reciprocal lattice vectors @xmath52 .
the interaction part of the hf hamiltonian @xmath53 may now be written as @xmath54,\ ] ] where @xmath55 with @xmath56,\end{aligned}\ ] ] @xmath57,\end{aligned}\ ] ] where @xmath58 the single - particle green s function is defined by @xmath59 and its fourier - transform by @xmath60 within the hf approximation , the equation of motion ( eom ) for @xmath61 is given by @xmath62 where @xmath63e^{i\bm{g}\times\bm{g'}\ell^2/2}.\end{aligned}\ ] ] because llm could be important , we retain several `` active '' lls ( with ll indices between @xmath64 and @xmath65 ) around the chemical potential ( see fig . [ division ] ) ; i.e. , we solve the eom explicitly for the green s function matrix allowing off - diagonal elements in the ll index for values @xmath6 satisfying @xmath66 .
however , it would be incorrect to completely neglect the filled lls below the active lls ( @xmath67 ) .
these levels can enter the calculations through @xmath68 and @xmath69 . however if sufficiently below the chemical potential , we expect ll mixing to be negligible for these states .
we thus treat these levels as `` inactive '' , and fix their density matrix elements to be @xmath70 .
( for self - consistency , we verify numerically that ll mixing for the lowest active level is very small , justifying the dividing point between active and inactive levels . ) with this form the inactive levels do not contribute to @xmath68 due to the @xmath71 in the hartree term ; i.e. , it is precisely cancelled by an interaction with a uniform neutralizing background.@xcite however , these levels do contribute a non - vanishing exchange energy @xmath69 , @xmath72 where @xmath73 stands for `` inactive '' and @xmath74 we can rewrite eq .
( [ ag ] ) as @xmath75 \delta_{t_1t}\delta_{s_1s}\delta_{\bm{g'}\bm{g}}\\ \nonumber & & + [ u_h^{act}(n , n_3;\bm{g'-g})\delta_{t_1t}\delta_{s_1s}\\ \nonumber & & -u_x^{act}(s_1,s , t_1,t , n , n_3;\bm{g'-g})]e^{i\bm{g}\times\bm{g'}\ell^2/2},\end{aligned}\ ] ] where the superscripts @xmath76 means now the summations in @xmath68 and @xmath69 are restricted to the active lls .
division of lls into active and inactive lls . in this example , @xmath77 and @xmath78 .
the upper and lower cutoffs of the active lls are determined by the self - consistency conditions @xmath79 and @xmath80 .
chemical potential @xmath81 for the case of filling @xmath82 is indicated as an example . in general several active levels
are retained both above and below @xmath81 for all calculations reported here . ]
the calculations involve solving eq .
( [ eq : eom ] ) to obtain the green s function , from which we obtain the density operator matrix elements .
finally , the hartree - fock energy is given by @xmath83\left<\hat{\rho}^{n_2st}_{n_3st}(0)\right > \\ \nonumber & & \quad + \frac{1}{2}\sum_{\bm{g}}[u_h^{act}(n_2,n_3;\bm{g})\left<\hat{\rho}^{n_2st}_{n_3st}(\bm{g})\right > \\ \nonumber & & \quad -\sum_{s_1,t_1}u_x^{act}(s_1,s_2,t_1,t_2,n_2,n_3;\bm{g } ) \left<\hat{\rho}^{n_2st}_{n_3s_1t_1}(\bm{g})\right>]\rbrace \\ & & + \frac{s}{\pi\ell^2}\sum_{n\atop{(inact)}}[\sum_s \varepsilon_n^s-\frac{e^2}{4\ell}\sigma_x(n , n)],\end{aligned}\ ] ] where the last line is a constant for given @xmath2 and @xmath64 .
table [ occupation ] details some typical results for the occupations of the various landau levels near the fermi energy .
in this example @xmath84 ; i.e. , the ll with @xmath85 is half - filled . note that in this notation we denote the valley index as a pseudospin , with two values @xmath86 and @xmath87 denoting the @xmath7 and @xmath8 valleys respectively .
we present results for different coupling constants in the range @xmath88 , consistent with previous estimates of its appropriate value @xcite .
our qualitative results are very similar for different values of @xmath9 , even for ( unphysical ) values well above 1 .
we can see that the occupations immediately become very small above the half - filled ll , and very close to 1 below it , indicating that llm is indeed small .
this small level of mixing , in spite of the small non - interacting energy gap between lls where the fermi energy is located , may be understood as being a consequence of the large exchange enhancement of the gap due to the filled lls .
furthermore , for smaller @xmath2 , deviations of the occupations from a step function decreases ( albeit just slightly ) , which means for decreasing field llm becomes even _ less _ important .
this unintuitive result occurs because of the large sea of negative energy ll states . with smaller field the degeneracy of each of these decreases , and so that more inert lls need to be filled to obtain the correct density of electrons [ see eq .
( [ cutoff ] ) ] . in units of @xmath89 ,
the exchange interaction increases with increasing @xmath26 , and the lls effectively become slightly more separated .
.[occupation]the diagonal density matrix elements @xmath90 , indicating the occupation of the spin and pseudospin split ll with quantum numbers @xmath91 .
the occupations are very close to a step function , indicating that llm is weak .
the deviation from a step function decreases as @xmath2 decreases , indicating that llm becomes less important . here
@xmath9 is set to 1 . [
cols="^,^,<,<,<",options="header " , ] in fig .
[ fig : tb ] we illustrate the stripe amplitude for states generated for three values of @xmath92 , corresponding to three different magnetic fields , but with the ratios of the unit cell sizes and magnetic length the same , and with a relatively small value of @xmath93 ( 0.1 ) . in this case one
may fit a straight line through these points , and find that it extrapolates to the origin rather accurately .
this is consistent with the stripe amplitude continuously vanishing in the @xmath94 limit , as was found in the continuum approach .
we have examined the stability of liquid and charge - ordered states for graphene ( focusing on stripes as a paradigm for the latter ) in the quantum hall regime against the effects of landau level mixing . because the coupling constant @xmath9 is independent of field ,
we find the llm does not increase with decreasing field , and that , counterintuitively , it decreases , albeit by a small amount .
this latter effect is due to a large exchange enhancement of the ll gaps from the filled negative energy lls , which increase in number as the field decreases . within mean - field theory
, this implies that clean and cold graphene at small fields and densities should support many different phases , determined solely by the filling factor .
this contrasts with the conventional 2deg , where a wigner crystal state is believed to reside throughout this regime . in graphene ,
the liquid phase thought to exist in the absence of doping is reached in the @xmath94 limit at fixed filling factor by an `` evaporation '' of the cdw , in which the amplitude vanishes linearly with @xmath2 .
the authors thank m. fogler , r. ct and i. herbut for helpful discussions .
this work is supported by nsf grant no .
dmr-0704033 and mat2006 - 03741(spain ) .
computer time was provided by indiana university . | we investigate integer and half - integer filling states ( uniform and unidimensional stripe states respectively ) for graphene using the hartree - fock approximation . for fixed filling factor
, the ratio between the scales of the coulomb interaction and landau level spacing @xmath0 , with @xmath1 the magnetic length , is a field - independent constant .
however , when @xmath2 decreases , the number of filled negative landau levels increases , which surprisingly turns out to decrease the amount of landau level mixing .
the resulting states at fixed filling factor @xmath3 ( for @xmath3 not too big ) have very little landau level mixing even at arbitrarily weak magnetic fields .
thus in the density - field phase diagram , many different phases may persist down to the origin , in contrast to the more standard two dimensional electron gas , in which the origin is surrounded by wigner crystal states .
we demonstrate that the stripe amplitudes scale roughly as @xmath2 , so that the density waves `` evaporate '' continuously as @xmath4 .
tight - binding calculations give the same scaling for stripe amplitude and demonstrate that the effect is not an artifact of the cutoff procedure used in the continuum calculations . |
particle statistics of a collection of indistinguishable particles is a genuinely quantum mechanical concept without any classical analog . in spatial dimensions three and above ,
pairwise interchange of particle coordinates in a many - body system is equivalent to a simple permutation of the coordinates .
consequently , each interchange has the effect of either a change of sign ( fermion ) or no change at all ( boson ) on the many - body quantum wave function . in @xmath1 dimensions
, however , exchanges and permutations are not necessarily equivalent .
@xcite in this case , under simple interchange of the particle coordinates , the corresponding space - time trajectories can form non - trivial braids in the @xmath1-dimensional space - time .
@xcite consequently , in ( @xmath2 ) dimensions , particles can have quantum statistics strikingly different from the statistics of bosons and fermions . a straightforward extension of the statistics of bosons and fermions is the abelian anyonic statistics , in which the many body wave - function , under pairwise exchange of the particle coordinates , picks up a phase @xmath3 , which can take any value between @xmath4 ( bosons ) and @xmath5 ( fermions ) . since a phase factor is only a one - dimensional representation of the braid group in @xmath6 , the statistics is still abelian . on the other hand ,
if the many body ground state wave function happens to be a linear combination of states from a degenerate subspace , a pairwise exchange of the particles can unitarily _ rotate _ the wave function in the ground state subspace . in this case , the effect of exchanging the particle positions is an operation on the wave function vector by a unitary matrix representing this rotation .
consequently , the statistics is non - abelian , @xcite and the corresponding system is a non - abelian quantum system .
it has been proposed that such systems , if the ground state subspace is concurrently protected by an energy gap , can be used as a fault - tolerant platform for topological quantum computation ( tqc ) .
@xcite one important class of non - abelian quantum systems , sometimes referred to as the ising topological class , @xcite is characterized by quasiparticle excitations called majorana fermions , which involve no energy cost ( when the mutual separation among the excitations is large ) .
the second quantized operators , @xmath7 , corresponding to these zero energy excitations are self - hermitian , @xmath8 , which is in striking contrast to ordinary fermionic ( or bosonic ) operators for which @xmath9 .
however , since @xmath7 and @xmath10 anticommute when @xmath11 , they retain some properties of ordinary fermion operators as well .
the majorana fermions , which are actually more like half - fermions , were envisioned @xcite by e. majorana in 1935 as fundamental constituents of nature ( _ e.g. _ neutrinos are thought to be majorana , rather than dirac , fermions ) .
majorana modes are intriguing @xcite because each majorana particle is its own anti - particle unlike dirac fermions where electrons and positrons ( or holes ) are distinct .
although the emergence of majorana excitations , which are effectively fractionalized objects ( anyons ) obeying non - abelian anyonic statistics , in solid state systems is by itself an extraordinary phenomenon , a great deal of attention has also been focused on them because of the possibility of carrying out fault tolerant tqc in two dimensional systems using these majorana particles .
tqc , in contrast to ordinary quantum computation , would not require any quantum error correction since the majorana excitations are immune to local noise by virtue of their non - local topological nature .
@xcite the direct experimental observation of majorana modes in solid state systems would therefore be a remarkable breakthrough both from the perspective of fundamental physics of fractional statistics in nature and the technological perspective of building a working quantum computer .
it is therefore not surprising that there has been recent resurgence of immense interest for the experimental realization ( and detection ) of majorana fermions in solid state systems .
recently , some exotic ordered states in condensed matter systems , such as the pfaffian states in fractional quantum hall ( fqh ) systems , @xcite @xmath12-wave superconductors / superfluids , @xcite theoretical models that can be potentially simulated in cold atom optical lattice systems , @xcite as well as the surface state of a topological insulator ( ti ) or related systems @xcite have been discussed as systems which can support majorana fermions as the lowest energy excitations . in the context of optical lattice systems , it has also been proposed that a 2d @xmath13 superfluid can be realized using only @xmath0-wave feshbach resonance modified by the topological berry phases arising from artificially generated spin - orbit coupling .
@xcite it has been shown recently @xcite that even a regular semiconducting film with a sizable rashba - type spin - orbit coupling , such as ingaas thin films , can host , under suitable conditions , majorana fermions as low energy excitations . since the basic effects behind the emergence of the majorana fermion excitations
spin - orbit coupling , @xmath0-wave superconductivity , and zeeman splitting are physically well - understood and experimentally known to occur in many solid state materials , the proposed semiconductor heterostructure @xcite is possibly one of the simplest condensed matter systems supporting majorana quasiparticles and non - abelian quantum order . by an analysis of the real - space bogoliubov - de gennes ( bdg ) equations for a vortex in the semiconductor thin film , in which @xmath0-wave superconductivity and a zeeman splitting are proximity - induced ( fig .
( 1a ) ) , it has been shown that the lowest energy quasiparticle excitation in the vortex core is a zero - energy majorana fermion mode .
this real space analysis has also been supported by a momentum space analysis in the form of an index theorem @xcite analogous to such a treatment in the context of one - dimensional dirac theory .
@xcite . here
a comment about the various means to induce a zeeman splitting in the semiconductor thin film is in order .
note that when the spin - orbit coupling is of the rashba type , we require a zeeman splitting which is perpendicular to the plane of the film ( zeeman splitting parallel to the film does not produce a gap in the one - electron band - structure , a firm requirement of our non - abelian state ) . @xcite inducing such a splitting by applying a strong perpendicular magnetic field is not convenient , because the magnetic field will give rise to unwanted order parameter defects such as vortices .
it is for this reason that we propose to induce the zeeman splitting by the exchange proximity effect of an adjacent magnetic insulator ( we ignore the small coupling of the spins in the semiconductor with the actual magnetic field of the magnetic insulator ) .
more recently , it has been shown that , when the spin - orbit coupling also has a component which is of the dresselhaus type , the appropriate zeeman splitting can also be induced by applying an in - plane magnetic field .
@xcite the majorana mode is separated by a finite energy gap ( so - called mini - gap ) from the other conventional fermionic excited states in the vortex core .
thus , for a collection of well - separated vortices , the resulting degenerate ground state subspace is protected from the environment by the mini - gap .
this enables the potential use of the semiconductor heterostructure in fig .
( 1a ) in tqc .
one of the main goals of the present paper is to provide the important mathematical details relevant to our solutions of the bdg equations in the semiconductor heterostructure .
these mathematical details , which were left out in ref .
[ ] , are given in sec .
[ ii ] and sec .
[ iii ] below .
it is important to note that , unlike the case of the surface of a @xmath14 strong topological insulator adjacent to an @xmath0-wave superconductor , @xcite the bdg equations in the spin - orbit coupled semiconductor are not exactly solvable .
we therefore only show that , in a specified region of the parameter space , a single non - degenerate solution of the bdg equations , which is spatially localized around the vortex core , is allowed .
we also show that the second quantized operator corresponding to such a solution is indeed a majorana fermion operator .
( in a subsequent section ( sec . [ vi ] ) we confirm the existence of such zero - energy majorana fermion states localized at the vortex cores by a full numerical solution of the bdg equations set up on a sphere . ) in the next few sections ( secs .
[ iv - ix ] ) we provide a comprehensive discussion of the interesting physics of non - abelian topological order arising via the complex interplay of spin - orbit coupling , zeeman splitting , and @xmath0-wave superconductivity externally induced in a host system .
we also deduce the parameter space needed for the establishment of the non - abelian order , as well as the associated tqpt accessed by tuning the system in or out of this parameter space . in sec .
[ x ] we study the superconducting and magnetic proximity effects in the host semiconductor thin film by a microscopic model tight - binding calculation . in the last part of the paper ( sec .
[ xi ] ) , we consider a one - dimensional version of our proposed structure a semiconducting nanowire with proximity induced @xmath0-wave superconductivity and a zeeman splitting .
we emphasize that the zeeman splitting can now be induced by a magnetic field parallel to the length of the wire , because such a field does produce a gap in the one - electron band structure without producing unwanted excitations in the adjacent superconductor .
this obviates the need for a nearby magnetic insulator .
for the zeeman splitting above a critical value , the wire is in a non - abelian topological phase with zero - energy majorana excitations at the ends .
we propose a scanning tunneling experiment from the ends of the semiconducting nanowire as possibly the most realistic experiment proposed so far to detect a majorana fermion state in a condensed matter system .
we use the terminology majorana particle or majorana fermion or majorana state or majorana excitation or majorana mode interchangeably in this article , all of them meaning precisely the same entity , namely , the non - degenerate zero - energy eigenstate ( _ i.e. _ a solution of the bdg equations ) at the vortex core of a spin - less chiral @xmath12-wave or other such topological superconductor .
we emphasize that this object obeys the non - abelian braiding statistics rather than ordinary fermionic statistics , and the majorana particle is its own anti - particle in contrast to the ordinary dirac fermions where electrons and holes ( positrons ) are distinct particle - hole conjugates of each other .
a part of the results presented in this paper approximate solutions of the bdg equations in the semiconductor heterostructure has been published elsewhere .
@xcite in secs .
[ ii , iii ] we provide all the mathematical details relevant to the solutions of the bdg equation which were left out in ref . [ ] .
most of the results contained in the subsequent sections are new .
some additional mathematical details related to the solution of the bdg equations are relegated to the appendix .
-wave superconductor ( sc ) and a magnetic insulator ( mi ) . in this geometry
, the semiconducting film can support non - abelian topological order .
( b ) : single - particle band - structure in the semiconducting film with and without the zeeman splitting induced by the mi . without the zeeman splitting , the two spin - orbit shifted bands touch at @xmath15 ( red lines ) .
then , for any value of the chemical potential , the system has two fermi surfaces . with a finite zeeman splitting ,
the bands have an energy gap near @xmath15 ( blue lines ) .
if the chemical potential lies in the gap , the system just has one fermi surface ( indicated by the dotted circle).,scaledwidth=47.0% ]
the single - particle effective hamiltonian @xmath16 for the conduction band of a spin - orbit coupled semiconductor in contact with a magnetic insulator is given by ( we set @xmath17 henceforth ) @xmath18 here , @xmath19 , @xmath20 and @xmath21 are the conduction - band effective mass of an electron , effective zeeman coupling induced by proximity to a magnetic insulator ( we neglect the direct coupling of the electrons with the magnetic field from the magnetic insulator ) , and chemical potential , respectively .
the coefficient @xmath22 describes the strength of the rashba spin - orbit coupling and @xmath23 are the pauli matrices .
the proximity - induced superconductivity in the semiconductor can be described by the hamiltonian , @xmath24 where @xmath25 are the creation operators for electrons with spin @xmath26 and @xmath27 is the proximity - induced gap .
the pairing term @xmath28 and the non - interacting part @xmath16 can be combined to obtain the bcs mean - field hamiltonian @xmath29 .
the excitation spectrum of this hamiltonian is defined in terms of the bogoliubov quasiparticle operators @xmath30 which satisfy @xmath31=e\hat{\gamma}^\dagger.\ ] ] such a quasiparticle operator @xmath32 can be used to construct excited states @xmath33 with energy @xmath34 from the ground state @xmath35 with energy @xmath36 .
the ground state @xmath35 is defined as the lowest energy state of the bcs hamiltonian satisfying @xmath37 .
the equation for the quasiparticle operator , eq .
[ eq : qpeqn ] , can be re - written as the bdg equations in the nambu basis , @xmath38 here , @xmath39 is the wave function in the nambu spinor basis , @xmath40 . introducing the pauli matrices @xmath41 in the nambu space the hamiltonian on the left hand side in eq .
( [ eq : h5 ] ) can be written as @xmath42\tau_z+[\delta(\bm r)\tau_++h.c]\ ] ] where @xmath43 .
the single - particle hamiltonian @xmath16 can be written in polar coordinates as @xmath44 where @xmath45 , @xmath46 and @xmath47 and @xmath48 .
the full bdg hamiltonian for an @xmath49-fold vortex can be written conveniently in the nambu space as @xmath50.\ ] ] in order to diagonalize the above hamiltonian it is convenient to note that the bdg hamiltonian has a combined spin - orbit - pseudospin rotational symmetry .
this symmetry can be expressed compactly by noting that @xmath51 commutes with the operator @xmath52 therefore , the eigenspinors of the bdg hamiltonian can be taken to be @xmath53 eigenstates with eigenvalue @xmath54 of the form @xmath55 the above equation can be used to eliminate the angular degree of freedom @xmath3 from the bdg equations as follows : @xmath56 here @xmath57 is @xmath3 independent .
more specifically @xmath58 under the action of the particle - hole transformation operator , @xmath59 , the @xmath60 spinor eigenstate with energy @xmath61 transforms into a @xmath62 eigenstate with energy @xmath63 because @xmath64 therefore , a necessary condition for a non - degenerate , @xmath65 , majorana state solution is that @xmath66 . from here onwards we will write @xmath67 and @xmath68 . single - valuedness of the spinor wave - functions in eq .
( [ eq : theta ] ) requires that @xmath69 must be an integer .
therefore , only odd vortices can have non - degenerate majorana eigenstates . from here onwards , for the simplicity of discussion
, we will consider only zero energy solutions at the cores of single - flux - quantum vortices @xmath70 .
the bdg matrix @xmath71 may be reduced to a real hamiltonian by applying the @xmath72 rotation @xmath73 as @xmath74 .
the solutions of the resulting @xmath65 bdg equation @xmath75 must come in complex conjugate pairs @xmath76 and @xmath77 . therefore the solutions @xmath76 can be required to be real without loss of generality . for such real solutions , it follows from the particle - hole symmetry of the bdg equations that @xmath78 is also a solution .
thus , any non - degenerate @xmath65 solution must be real and satisfy the property @xmath79 .
moreover , because @xmath80 , the possible values of @xmath81 are @xmath82 . using the relation @xmath83 , which follows from @xmath84 , the bdg hamiltonian for a given value of @xmath81 is of the form @xmath85 the hamiltonian in this limit does not couple the @xmath86 sectors ( electron and hole ) .
this allows one to write the bdg differential equation in terms of only the electron - sector @xmath87 of the spinor @xmath88 .
the corresponding reduced bdg equations for a single vortex @xmath70 take the form of a @xmath89 matrix differential equation : @xmath90 the hole part of the spinor is not independent and is constrained by the value of @xmath81 such that @xmath91 and @xmath92 and the majorana spinor has the form @xmath93 .
we now approximate the radial dependence of @xmath94 by @xmath95 for @xmath96 and @xmath97 for @xmath98 where @xmath99 is of the order of the radius of a vortex core . in view of the topological stability of the putative majorana zero - energy solution to local changes in the hamiltonian , @xcite
such an approximation can be made without any loss of generality . inside the vortex core @xmath100 , which is the non - superconducting region @xmath101 ,
it is possible to construct explicit analytic solutions @xmath102 to these equations in terms of the bessel functions @xmath103 and @xmath104 as , @xmath105 by substituting eq .
( [ eq : bessel ] ) into eq .
( [ eq : zeroenergy ] ) we find that @xmath106 and @xmath107 satisfy @xmath108 which implies @xmath109 existence of solutions in terms of @xmath110 requires that @xmath107 satisfies the characteristic equation of the matrix in eq .
( [ eq : matrix ] ) , which is given by @xmath111 this is a quadratic equation in @xmath112 with real roots
. therefore , the solutions for @xmath107 are either purely real or purely imaginary and come in pairs with opposite signs .
a real root @xmath113 of this equation corresponds to a crossing of some band at the fermi level . for purely real roots @xmath107 ,
only the solution @xmath102 corresponding to positive @xmath107 is normalizable at the origin and therefore physically acceptable . on the other hand , the purely imaginary roots @xmath114 lead to a single real solution @xmath115 .
thus we can see that in general there are @xmath116 linearly independent solutions .
if all @xmath117 solutions of @xmath107 are real then these correspond to the @xmath116 fermi surfaces obtained from the intersection of the bands with the fermi level .
if only @xmath118 pair of solutions is real then the imaginary pair corresponds to a decaying state .
the bdg equations describing the proximity - induced supercondutivity at a ti / sc interface @xcite follow from the bdg equations for the present system , eq .
( [ eq : zeroenergy ] ) , by taking @xmath119 . in this case
the matrix equation reduces to @xmath120 this equation only has @xmath118 pair of real solutions @xmath121 and therefore has only @xmath118 linearly independent solution in the core of the vortex .
the solution outside the vortex does not have a simple analytic form as the solution inside .
motivated by the large @xmath122 asymptotic expansion for bessel functions , for @xmath123 we can consider a series expansion of the form @xmath124 where @xmath125 are analytic power series in @xmath126 .
we expect to be able to close such a series of equations since the matrix in eq .
( [ eq : zeroenergy ] ) only has derivatives and powers of @xmath127 . @xmath128
as shown in the appendix , the last equation has a simple solution as a power - series in @xmath127 which can be determined numerically .
moreover in this power series expansion we can determine the equation for the @xmath4-th order term by formally setting @xmath129 as below @xmath130 setting @xmath131 one can see that the matrix appearing in the above equation is related to the one determining the quasiparticle bandstructure from the bdg equations .
the values of @xmath107 consistent with the above equation are determined by setting @xmath132 the two families of solution for @xmath82 are related simply by flipping the sign of @xmath107 .
the sign of the product of the roots @xmath133 of eq .
[ eq : char_eq ] is given by @xmath134 where @xmath135 is the polynomial evaluated at @xmath136 .
the parity of the number of normalizable solutions with @xmath137 is given by @xmath138 . since eq .
[ eq : char_eq ] is real , complex roots @xmath133 occur in conjugate pairs
. therefore , complex roots can not affect the sign of either @xmath139 or @xmath140 .
it follows that @xmath141 .
of the characteristic equation eq .
[ eq : char_eq ] shown in the complex plane . in the non - abelian phase @xmath142 , 3 roots with positive real parts and 1 root with negative real part for @xmath143 while there are only 2 roots on either side of the imaginary axis for @xmath144 for @xmath143 .
the roots in the @xmath145 channel are the negative of the roots in the @xmath145 channel.,scaledwidth=47.0% ] therefore , the condition @xmath146 implies that there are an odd number of roots with positive real parts .
specifically , as seen in fig .
( [ fig : roots ] ) , if @xmath147 , there are @xmath148 roots on one side of the imaginary axis and @xmath118 root on the other side . similarly for @xmath144 we must have @xmath116 solutions on each side of the imaginary axis . a slightly different version of this argument has previously been presented .
@xcite as discussed before , for @xmath149 , one of the channels labeled by @xmath150 has a solution with @xmath148 decaying ( negative real parts ) solutions and @xmath118 growing solution . out of these @xmath148 decaying solutions ,
one is purely real and negative and the other @xmath116 are complex conjugate with negative real parts .
an observation that can be made by considering eq .
( [ eq:64 ] ) is that if @xmath151 corresponds to a value @xmath107 then @xmath152 corresponds to an eigenvalue @xmath153 .
thus from one pair of complex conjugate decaying eignevalues we can construct a pair of real solutions @xmath154 where @xmath155 and @xmath156 with the solution 1 corresponding to @xmath157 and 2 to @xmath158 . the non - degenerate real eigenvalue already corresponds to a real eigenvector @xmath159 on the other hand , for @xmath96 we expect a 2 parameter family with the general solution given by @xmath160 matching the gradient and the wave - function at @xmath161 we get @xmath162 together with the normalization constraint on the global wave - function , this leads to 5 equations in 5 variables , which leads to a unique solution for the majorana mode in the case @xmath147 .
however , for the other case with @xmath144 , there are only 2 decaying modes outside the vortex core .
the existence of a majorana mode would then require us to satisfy 5 equations with 4 variables .
such a problem in general is over constrained and no majorana solutions exist in this case .
for a planar system , the rashba spin - orbit term @xmath163 in the hamiltonian we considered can also be written as @xmath164 .
these 2 terms are simply related to each other by a @xmath72 spin rotation and a more general rashba - type spin - orbit term can be written as @xmath165.\ ] ] a recent proposal @xcite has considered a defect in such a spin - orbit coupling where the angle of the spin - orbit @xmath166 varies in space to form a vortex @xmath167 .
the full bdg hamiltonian for such a vortex can be written in nambu space as @xmath168 where the anti - commutation must be introduced in the spin - orbit term to preserve hermiticity . substituting the circular - polar form for the derivatives we note that the hamiltonian becomes @xmath169 which in turn simplifies to a @xmath3 independent form @xmath170 therefore , @xmath171 commutes with the above hamiltonian and
the spinor form is @xmath172 as before only the @xmath66 channel can lead to a non - degenerate majorana solution , and the bdg equation in this channel is given by @xmath173 since the above bdg equation is real , it can be reduced to a @xmath89 matrix differential equation : @xmath174 as before , considering a step function vortex profile @xmath175 for @xmath96 and @xmath176 for @xmath123 , one notices that the reduced bdg equation outside the spin - orbit vortex core resembles the reduced bdg equation in the same region for a regular vortex in the large @xmath122 limit ( eq . ( [ eq : zeroenergy ] ) with @xmath97 ) . inside the vortex core @xmath175 , and the bdg equations , as before , are analytically solvable via bessel functions as below @xmath177 where @xmath178 as before , this leads to 2 solutions inside and 3 solutions outside the vortex core , with 5 constraints at the interface .
this leads to a single non - degenerate majorana solution at the interface .
contrary to the result in ref .
[ ] for the asymptotic @xmath179 behavior of the zero energy wave function , @xmath180 , we find that the asymptotic zero energy wave function behaves as @xmath181 for @xmath182 and small @xmath183 .
therefore , according to our result , the decay length of the zero energy wave function diverges in the limit of vanishingly weak superconductivity @xmath184 and the majorana mode disappears by delocalizing over the entire system .
this is in contrast to the result for the wave function in ref .
[ ] where the zero energy majorana solution remains localized for arbitrarily small values of the superconducting gap .
now we apply a similar argument to the vortex in proximity - induced @xmath0-wave superconductivity on a ti surface @xcite which is obtained from our rashba model by setting @xmath119 . the equation for the allowed values of @xmath107 in the superconductor for @xmath123
are then @xmath185 therefore , in each of the @xmath150 channels , for small @xmath20 , there are a pair of complex conjugate eigenvalues on the same side of the imaginary axis . for the @xmath157 channel both the eigenvalues are to the right of the imaginary axis and therefore are acceptable decaying solutions .
thus there are 2 linearly independent solutions @xmath186 and @xmath187 for @xmath123 . from our previous discussion it is now clear that there is only one such solution @xmath188 for @xmath96 .
since the hamiltonian is linear in the derivative , the boundary conditions only require us to match the wave - function @xmath189 at @xmath161 and not the derivative .
the boundary conditions that the zero energy solution must satisfy at @xmath161 are given by , @xmath190 together with the normalization condition for the zero energy wave function , the above equations provide @xmath148 constraints for the three variables @xmath191 .
this yields a unique zero energy majorana wave - function for a vortex on a ti surface .
in previous sections we calculate and show the existence of a majorana mode in a vortex at the interface of an @xmath0-wave superconductor and a spin - orbit coupled semiconductor .
the most important information missing from these analytical calculations is the excitation gap above the zero - energy majorana state , the so - called minigap .
a proper calculation of this requires a numerical solution of the vortex problem which can be done by considering the system on a sphere with a vortex - antivortex pair @xcite as shown in the inset of fig .
( [ fig : minigap](a ) ) .
the bdg hamiltonian of this problem can be written as @xmath192\tau_z+ v_z \sigma \cdot \hat{r}+\delta(\mathbf{r})\tau_x\ ] ] where @xmath193 $ ] is the non - hermitean gradient operator restricted to the surface of the sphere and @xmath194 .
the above hamiltonian takes a more familiar form in angular coordinates as @xmath195\tau_z+v_z \sigma\cdot \hat{r}+\delta(\theta)\{\cos{\phi}\tau_x+\sin{\phi}\tau_y\}\nonumber\\ & = [ \frac{\eta}{r^2 } l^2-\frac{\alpha}{r } \{l_z\sigma_z+\frac{1}{2}l_+\sigma_-+\frac{1}{2}l_-\sigma_+\}-\mu]\tau_z+v_z \{r_z \sigma_z+\frac{1}{2}r_+\sigma_-+\frac{1}{2}r_-\sigma_+\}\nonumber\\ & + \frac{1}{2}\frac{\delta(\theta)}{\sin{\theta}}\{r_+\tau_-+r_-\tau_+\}\end{aligned}\ ] ] where @xmath196 and @xmath197 . in these equations @xmath198 and @xmath199 .
the spectrum of excitations of this system is found by solving the eigenvalue problem @xmath200 for the zero energy majorana state at the north pole .
the components of @xmath201 satisfy @xmath202 confirming the majorana nature of the wave - function .
we also show the semiconductor heterostructure on the surface of a sphere with a vortex and an antivortex ( with reduced superconducting amplitudes at the vortex cores ) situated at the north and the south poles .
( b)numerical results for the vortex mini - gap @xmath203 ( solid line ) and bulk gap ( dashed line ) plotted against the spin - orbit coupling strength @xmath22 on the semiconductor . in these plots
we have used @xmath204 , @xmath205 and @xmath206 in the units where @xmath207 .
the spin - orbit coupling strength @xmath208 in ( a ) and varies for @xmath209 . in these units ,
the size of the vortex core has been taken to be unity . ]
similar to the vortex in the planar geometry , the bdg hamiltonian has a combined spin - orbit - pseudospin rotational symmetry .
this symmetry can be expressed compactly by noting that the hamiltonian commutes with @xmath210 where we have used the identity @xmath211=\pm r_{\pm}$ ] .
the @xmath212 dependence of the eigenstates with @xmath213 can be written as @xmath214 .
the @xmath212 independent part of the eigenstate @xmath215 then satisfies a 1 dimensional bdg equation @xmath216 where @xmath217 which can be explicitly checked to be @xmath212 independent , and @xmath218 . to solve for the @xmath3 dependent part of @xmath215
it is necessary to convert @xmath219 to a discrete matrix by expanding @xmath220 where @xmath221 are the associated legendre polynomials which are the @xmath212 independent parts of the spherical harmonics . in the associated legendre polynomial
basis the kinetic energy term @xmath222 has the simple diagonal form @xmath223 . under the transformation in eq .
( [ eq : hm ] ) , the terms @xmath224 in @xmath225 transform into @xmath226 in @xmath219 .
therefore , its matrix elements in the associated legendre polynomial basis can be calculated using the spherical harmonic addition theorem .
a similar procedure can be used to calculate the matrix elements of the @xmath3 dependent vortex . for a vortex , we take @xmath227 where @xmath228 is taken to be the length - scale of the vortex . from symmetry properties it is clear that @xmath229 is an even polynomial in @xmath230 and can be written in terms of associated legendre polynomials as @xmath231 where the associate legendre polynomial can be written as @xmath232 and @xmath233 .
as with the @xmath99 operators , the angular momentum matrix elements can be calculated from the above expansion by using the spherical harmonic addition theorem . as in the analytic solution for the vortex , the angular momentum index @xmath234 transforms from @xmath235 under the particle - hole transformation @xmath236 and we expect non - degenerate @xmath65 majorana solutions of eq .
[ eq : spherebdg ] only in the @xmath237 channel .
this is confirmed by our numerical solution of eq .
[ eq : spherebdg ] where we find that only in the topological phase @xmath147 , are there a pair of states in the @xmath237 angular momentum channel whose eigenvalues approach 0 exponentially with increasing radius @xmath99 .
the non - zero energy eigenvalue of the majorana fermion is a result of the presence of 2 vortices in our calculation with a finite distance between them .
the wave - function of the @xmath65 eigenvalue of the @xmath237 angular momentum channel , that is localized at the north - pole is plotted in fig .
( [ fig : minigap](a ) ) .
the components of the wave - function are seen to satisfy @xmath202 confirming the majorana character of these states . in the figure ,
the wave - functions of the majorana modes is seen to decay and oscillate away from the north pole .
the splitting between the majorana modes into a pair of exponentially small oscillating eigevalues is a result of the overlap between the majorana modes at the two poles @xcite . aside from the @xmath65 eigenvalue in the @xmath237 angular momentum channel , an isolated vortex confines a set of non - zero eigenvalues in other @xmath238 angular momentum channels . of these
, the eigenvalue with the smallest absolute value occurs in the @xmath239 angular momentum channel and has an eigenvalue equal to the so - called mini - gap of the vortex . as mentioned before
, the superconductivity in the non - abelian superconducting phase is re - entrant with a bulk gap that is proportional to the spin - orbit coupling strength . as seen from fig .
( [ fig : minigap](b ) ) both the bulk and mini - gap are proportional to the spin - orbit coupling strength @xmath22 .
for spin - orbit coupling @xmath240 and chemical potential @xmath182 , both the mini - gap and the bulk gap are of order the induced pairing potential @xmath241 .
therefore for the semiconductor structure where superconductivity is proximity induced , the mini - gap of a vortex can be tuned to be of order @xmath241 if the chemical potential @xmath21 can be tuned to be less than order @xmath242 , the zeeman potential applied to the semiconductor .
this is different from the conventional case of a regular ( not proximity induced ) where the chemical potential is of order @xmath243 and the mini - gap is of order @xmath244 which is much smaller than @xmath241 .
thus the ability to independently control the chemical potential in the semiconductor heterostructure provides us with a powerful tool that can increase the mini - gap of the vortex in the semiconductor heterostructure shown in fig .
( [ fig1 ] ) by orders of magnitude from the values in chiral p - wave superconductors to of the order of 1 k. this leads to the possiblility of performing tqc with the majoranas trapped in vortices in the heterostructure at temperatures which are as large as a fraction of a 1k .
we found that the majorana modes exist for a spin - orbit coupled semiconductor system only in the parameter regime @xmath245 .
this in turn was related to the parity of the roots in one half of the complex wave - vector plane of solutions outside the vortex core . as pointed out before in the context of eq .
( [ eq : char_eq ] ) , these roots are indeed properties of the reduced bulk superconducting hamiltonian in the absence of a vortex .
we also expect such a connection between the bulk properties and the existence of majorana modes on general topological grounds @xcite . to show explicitly the connection between the condition for the existence of majorana modes @xmath246 and the bulk properties , we note that even though the gap in the bulk superconducting state prevents the existence of propagating states at @xmath65 , it allows evanescent states . since the states at @xmath65 are particle - hole symmetric eigenstates @xmath247 of a real hamiltonian , we can apply an argument analogous to eq.([eq : zeroenergy ] ) to obtain a bulk bdg equation in a @xmath81 channel @xmath248 considering an evanescent state of the form @xmath249 leads to a constraint on @xmath107 which was previously written as eq .
( [ eq : char_eq ] ) .
therefore , the condition on @xmath250 , which determines whether the phase supports a majorana solution or not is precisely related to the parity of decaying evanescent modes in a given @xmath81 channel in the bulk superconductor at @xmath65 .
a change of the parity of the decaying evanescent modes requires an @xmath65 mode to become propagating , which can only exist if the bulk superconductor is gapless .
therefore , a change of the sign of @xmath250 , which determines the topological nature of the phase , must be accompanied by a closing of the bulk spectrum .
this is determined by the full bdg hamiltonian for a state with momentum @xmath251 and can be written in the nambu space as @xmath252 the spectrum is obtained by considering @xmath253 which can be simplified to the equation , @xmath254 where @xmath255 . setting @xmath256 , it can be seen that @xmath257 which vanishes as @xmath250 becomes zero , as expected .
recent work @xcite has shown that the quantity @xmath250 is the pfaffian of the bdg hamiltonian at @xmath256 , @xmath258 . the sign of @xmath250 , which determines whether the phase of the superconductor is non - abelian or not has been shown to be related @xcite to the parity of the first chern number topological invariant describing time - reversal broken topological superconductors . @xcite .
the strength of the spin - orbit coupling in the inset is such that @xmath208 corresponds to @xmath259 ev - .
the gap vanishes at the critical value @xmath260 .
the spin - orbit coupling has negligible effect below this critical point and the superconducting gap is of a conventional @xmath0-wave type .
above the critical value and in the absence of spin - orbit coupling , the superconducting gap is destroyed by the zeeman coupling .
spin - orbit coupling opens up a gap in this phase leading to re - entrant superconductivity with is topological.,scaledwidth=47.0% ] the phase diagram of the spin - orbit coupled semiconductor system can be understood from fig .
( [ fig : phasediag ] ) , which gives the variation of the quasiparticle gap versus the zeeman splitting .
one knows from topological stability of the majorana fermion mode that , due to its non - degeneracy , the majorana state is protected as long as the bulk gap does not close as one moves through the parameter space . in fig .
( [ fig : phasediag ] ) , the gap closes ( at the wave vector @xmath256 ) for the zeeman splitting corresponding to @xmath261 .
the phase with @xmath262 supports the non - degenerate majorana state while the phase with @xmath263 does not .
these two regions are separated by a gapless point in the parameter space , which signifies a topological quantum phase transition .
the quantum phase transition is topological since the superconducting order on both sides is explicitly @xmath0-wave and the two phases differ only by the topological properties such as majorana modes in defects and boundaries .
the proposal to realize majorana fermion modes in spin - orbit coupled semiconductor system involves the introduction of a large zeeman potential . in general , a zeeman splitting is known to compete with and eventually destroy superconductivity . to understand better the competition between the zeeman splitting and superconductivity in a spin - orbit coupled semiconductor , we first consider the case without spin - orbit coupling .
this case is described by the bdg hamiltonian @xmath264 the dispersion relation of this hamiltonian is @xmath265 . in this case , with @xmath266 , we obtain a conventional proximity induced @xmath0-wave superconductor with no majorana phase . as @xmath20 increases above @xmath267 the quasiparticle gap of the system closes and one obtains a metal with a fermi momentum @xmath268 given by @xmath269 .
this is the well - known chandrasekhar - clogston limit @xcite where strong zeeman splitting suppresses the superconducting quasiparticle gap .
this suppression is due to the fact that , in the spin - polarized regime @xmath270 , a small pairing potential can not open a @xmath0-wave superconducting gap since the latter couples opposite spins .
the bdg hamiltonian at @xmath268 is doubly degenerate and is given by @xmath271 the degeneracy of the above hamiltonian at the gapless point , which arises from the particle - hole symmetry , is lifted by the rashba spin - orbit coupling term @xmath272 in the semiconductor .
this yields a topological superconductor with a gap given by @xmath273 considering the gap as a function of the zeeman splitting ( fig .
( [ fig : phasediag ] ) ) , it is clear that , for zeeman splitting below the critical value @xmath274 , the superconductivity is non - topological in nature .
the topological superconducting phase that supports majorana fermions is created by the application of a zeeman splitting to _ suppress the conventional pairing potential_. in this regime , the spin - orbit coupling can open up a gap resulting in a re - entrant superconducting phase .
however , as is evident from the previous discussions , the re - entrant superconductivity is unconventional ( topological ) in the sense that it supports majorana fermions .
one of the signatures of a topological phase is the existence of gapless edge states which are inextricably linked to bulk topological properties such as majorana modes in vortices .
@xcite the spin - orbit coupled semiconductor structure introduced in sec .
ii can be shown to have gapless edge states using methods similar to the ones described in the previous sections .
furthermore it turns out that this approach to analyze the existence of majorana edge modes does not impose additional requirements such as rotational invariance that were critical for the demonstration of a majorana solution in a vortex .
therefore , this method can be used to examine the question of the existence of majorana edge modes even in the heterostructures with more general forms of spin - orbit coupling proposed by alicea @xcite where the zeeman splitting can be introduced by an in - plane magnetic field .
edges can be created in the semiconductor heterostructure by varying a parameter of the hamiltonian such as @xmath21 , @xmath242 or @xmath241 perpendicular to the edge of a surface . without loss of generality we can consider an edge that is perpendicular to the direction @xmath275 .
because of translational symmetry along the edge , the resulting edge bdg hamiltonian has @xmath276 as a parameter .
the momentum parameter @xmath276 transforms as @xmath277 under the particle - hole transformation .
therefore , a non - degenerate majorana mode can only exist for @xmath278 .
fixing @xmath278 reduces the two dimensional edge problem to a @xmath118-dimensional bdg hamiltonian for a single band semiconductor with spin - orbit coupling ( assumed to be linear in the momentum @xmath279 ) , which in general can be written as @xmath280 where the unit vector @xmath281 is the direction of the effective zeeman field and the unit vector @xmath282 characterizes the spin - orbit coupling . using the spin rotation transformations on @xmath51
, we can choose @xmath283 without loss of generality .
this yields the hamiltonian @xmath284 which is invariant under spin - rotations about the @xmath285-axis . therefore without loss of generality
we can reduce the above hamiltonian to @xmath286 non - degenerate majorana spinor solutions are of the form @xmath287 and are completely determined by the 2-spinor @xmath288 .
this fact was used to obtain the majorana solutions for vortices to reduce the bdg equation from a @xmath289 system of equations to a @xmath89 system of equation . however , this reduction procedure required the bdg hamiltonian to be real which is not the case for general forms of spin - orbit coupling and zeeman splitting .
the bdg equation for the zero energy mode @xmath290 may be reduced to an equation for @xmath288 as @xmath291u+\delta(y)e^{\imath\phi}(\imath\sigma_y)u^*=0.\ ] ] this equation is not real but may be reduced to a system of real equations by writing @xmath292 and taking the real and imaginary parts of the resulting equation giving a pair of equations of the form @xmath293u_r-[\delta(y)\sin{\phi}-v_z\sin{\nu}](\imath\sigma_y)u_i=0\\ & \left[(-\partial_y^2-\mu(y ) ) + v_z\cos{\nu}\sigma_z+\alpha(\imath \sigma_y)\partial_y-\delta(y)\cos{\phi}(\imath\sigma_y)\right]u_i+[\delta(y)\sin{\phi}-v_z\sin{\nu}](\imath\sigma_y)u_r=0.\end{aligned}\ ] ] the above pair of equations is similar to the pair of equations obtained for the two @xmath82 channels except that earlier the two channels were decoupled . for @xmath294 independent of @xmath285 , the 2 channels
can also be decoupled by choosing @xmath212 such that @xmath295 . in
what follows , we will make this choice and also replace @xmath296 and @xmath297 .
this results in a reduced bdg equation for the @xmath65 reduced spinor @xmath298 @xmath299 where @xmath82 .
an edge in a two dimensional system of the type considered above is defined by requiring some parameter of the hamiltonian to vary across the edge situated at @xmath300 .
we take this parameter to be constant for @xmath301 and @xmath302 . in this case
, our previous approach can be applied in a way even simpler than the application to the vortex problem , since the solutions on both sides of the interface at @xmath300 can be approximated as a sum @xmath303 where , as in the interior of the vortex ( but far from the vortex core ) , ( eq.[eq : largermode_vortex ] ) , @xmath304 similar to the vortex case , in the topological phase @xmath305 , there are 3 values of @xmath133 such that @xmath306 , while in the non - topological phase @xmath144 , there are only 2 solutions in a given @xmath81 channel .
the coefficients @xmath307 in the solution are determined by matching the boundary conditions on @xmath298 at @xmath300 .
the coefficient @xmath250 , written in terms of the original parameters of the wire , reduces to @xmath308 and is not affected by the @xmath212 and @xmath309 parameters that were introduced to make the bdg hamiltonian real .
the procedure of reducing the bdg hamiltonian to a real hamiltonian only introduces the additional constraint @xmath310 .
based on analogy with fqhe and chiral @xmath12-wave superconductors , one expects a chiral gap - less state confined to the edge of the semiconductor heterostructure .
an edge can be created in such structures by raising the chemical potential @xmath21 towards the edge such that electrons stay confined inside the system .
therefore , an edge of a system confined to @xmath301 is defined by @xmath311 for @xmath301 and @xmath312 for @xmath302 . in these structures
we assume that @xmath313 is independent of @xmath285 .
the bdg equation now reduces to a @xmath314 system of equations , @xmath315\psi(x)=0,\ ] ] where @xmath316 in order to solve a semi - infinite system we make a plane - wave trial solution ansatz @xmath317 where @xmath107 must now satisfy @xmath318 as in the case of the vortex , for @xmath319 there are 3 solutions on the right half of the complex @xmath107 plane and 1 solution on the right half for @xmath143 .
the situation is opposite for @xmath145 .
solutions with @xmath320 are physical on the left - edge of the system while @xmath321 is physical on the right edge of the system .
thus for @xmath143 there are 3 physical solutions on the left - edge of the system which is the exact number needed to make the 2-component spinor vanish at the left - edge .
consequently , there is a localized zero mode at the left edge of the system in the @xmath143 channel .
similarly there is a localized zero mode at the right edge of the system in the @xmath322 channel .
finally , for @xmath323 , there are no zero energy solutions at either edge since there are only 2 physical solutions at each edge which is insufficient to match the boundary conditions .
since the wave - function is confined to the edge , we expect the boundary conditions @xmath324 , which together with normalization lead to 3 constraints .
as mentioned before , in the topological phase we obtain 3 @xmath307 coefficients corresponding to the 3 normalizable solutions in the interior .
therefore , there is a unique zero energy state resulting from a matching of the boundary conditions .
this state is a majorana mode for the end - point of the nanowire in the topological phase@xmath246 , which disappears when we tune the wire through a phase transition to @xmath144 .
the majorana modes at the edges discussed above only occured at @xmath278 .
the complete spectrum of the edge is obtained by considering the bdg hamiltonian at small @xmath325 using the @xmath326 perturbation theory .
to lowest order in @xmath276 , the nambu spinor wave - function can be approximated as @xmath327 with an energy @xmath328 where @xmath329 .
a similar chiral majorana wire is obtained by considering an edge between the topological phase @xmath147 and @xmath144 where @xmath330 is constant but @xmath331 for @xmath302 . in that case
there are 5 constraints to match as in the vortex case , and there are 5 coefficients , 3 arising from the topological phase at @xmath301 and 2 from the non - topological phase @xmath302 .
now we consider the junction of a pair of topological superconducting islands with phases @xmath212 and @xmath332 which is a geometry that is of particular interest to tqc architechtures @xcite .
for such a choice of phases , the effective pairing potential in @xmath333 is a step function given by @xmath334 for @xmath335 and @xmath336 for @xmath337 .
as before we then replace @xmath297 . focusing on the @xmath278 particle - hole symmetric momentum for the edge , solutions for @xmath335 and @xmath337 can be expanded in terms of spinor functions given in eq .
[ eq : psix ] which is written in terms of eigenvalues @xmath107 and eigenvectors satisfying eq .
[ eq : z2 ] .
normalizable solutions must now be composed of superpositions of exponentials with @xmath338 for @xmath301 and @xmath339 for @xmath337 .
we note at this point that the equations for @xmath337 and @xmath335 differ by a change in sign of @xmath340 across the interface which corresponds to a change in sign of @xmath107 .
thus , as before for @xmath143 , in the topological phase we have 3 values of @xmath107 such that @xmath339 for @xmath337 and 3 values such that @xmath338 for @xmath335 .
following the boundary condition matching argument of the last section , for the @xmath5 junction there are 6 states at @xmath341 to compose wave - functions at @xmath341 which need to satisfy 5 constraints .
therefore , generically there will be a pair of zero energy modes satisfying these equations .
it might appear that , unlike the case for the chiral edge states , the pair of majorana states can not be topologically protected . in the case of the ti / sc interface , [ ]
the existence of such a pair of non - chiral majorana edge modes at a phase difference @xmath5 was a consequence of time - reversal symmetry which is broken here .
in our calculation we find that because of this time - reversal breaking , the pair of degenerate zero modes may occur at a phase difference of @xmath342 .
in fact , by considering the evolution of the andreev bound state spectrum in the junction as a function of phase difference , @xcite it is possible to show that even though the majorana nature of the pair of non - chiral majorana modes is not topologically protected , it is not possible to eliminate the zero - modes all together .
the zero - crossing of the non - chiral majorana modes may only be shifted to different values of phase by time - reversal breaking perturbations .
this is a result of the fact that the 2 particle - hole symmetry related branches of the andreev bound state spectrum differ by fermion parity .
therefore , even though an infinitesimal perturbation can lift the degeneracy of the 2 zero modes at some value of phase difference @xmath343 , it can only do so by shifting the crossing to a neighboring value of @xmath343 .
similar to the chiral edge modes one can use the @xmath344 perturbation theory to construct a pair of linearly dispersing modes from the pair of zero energy states at @xmath345 .
in this section we study , starting from a microscopic tight binding model , the excitation spectrum of a semiconductor thin film sandwiched between an s - wave superconductor ( sc ) and a ferromagnetic insulator ( mi ) .
we determine the dependence of the effective sc gap and zeeman splitting induced by proximity effects on the parameters that characterize the heterostructure model .
we also calculate the dynamical contributions to a low - energy effective theory of the proximity effect and identify parameter regimes suitable for the experimental implementation of the semiconductor - based proposal of a platform for topological quantum computation .
to study the proximity effect in the sc - semiconductor - mi heterostructure , we consider the minimal microscopic model defined by the hamiltonian @xmath346 the @xmath16 term describes the semiconductor thin film , @xmath347 + \sum_{i,\sigma } v(z_i)c_{i\sigma}^{\dagger}c_{i\sigma } , \label{h0}\end{aligned}\ ] ] where the first contribution describes hopping on a cubic lattice , while the second represents a lattice model of the rashba spin - orbit interaction .
the hopping matrix elements are non vanishing for nearest neighbors , @xmath348 , and next nearest neighbors , @xmath349 , and we also include an on - site contribution @xmath350 that shifts the bottom of the semiconductor spectrum to zero energy .
the parameter @xmath22 represents the rashba coupling constant and @xmath351 are nearest neighbor displacements in the xy - plane of a cubic lattice with lattice parameter @xmath352 .
the system is assumed to be infinite in the x- and y - directions and contains n planes perpendicular to the z - direction .
the quantities @xmath353 are pauli matrices and @xmath354 are electron creation and annihilation operators , respectively .
the last term represents an external bias potential that modify the on - site energies along the z - direction . for the superconductor we use
a simple mean - field model defined by the hamiltonian @xmath355 where @xmath356 represents the nearest neighbor hopping on a cubic lattice and @xmath241 is the mean - field s - wave sc order parameter .
the sc and the semiconductor thin film have a planar interface perpendicular to the z - direction . for clarity , the electron creation and annihilation
operators inside the sc were denoted @xmath357 and @xmath358 , respectively . the third term in eq .
( [ htot ] ) represents the ferromagnetic insulator which , again , is modeled at the mean field level by the hamiltonian @xmath359 the nearest neighbor hopping is spin dependent and we consider the case @xmath360 . the spin - dependent on - site energy was divided into a contribution @xmath361 that places the top of the valence band and the bottom of the conduction band at the same energy @xmath362 and a term proportional to the insulating gap @xmath363 .
the last two terms in eq .
( [ htot ] ) describe the coupling at the two interfaces , semiconductor - sc and semiconductor - mi , respectively : @xmath364 the parameters @xmath365 and @xmath366 characterize the transparencies of the two interfaces and provide the energy scales for the coupling between the semiconductor thin film and the sc and mi , respectively .
these energy scales are crucial for determining the strength of the proximity effects . we diagonalize eq .
( [ htot ] ) numerically for a system with periodic boundary conditions along the x- and y - directions and a finite size along the z - direction .
the semiconductor , sc and mi contain @xmath367 , @xmath368 and @xmath369 planes , respectively .
for the semiconductor thin film we consider @xmath370 , while @xmath368 and @xmath369 are typically of the order of several hundreds . for large systems ,
the proximity effects are independent on @xmath368 and @xmath369 and we explicitly checked that the values used in the calculations are within that regime .
the hopping matrix elements for the semiconductor are @xmath371 and @xmath372 , which , for a cubic lattice with @xmath373 generate in the low wavelength limit a small effective mass @xmath374 characteristic of semiconductors such as inas .
the value of the rashba coupling constant is @xmath375 mev , corresponding to a strong spin - orbit coupling of the order 100 mev . for the sc ,
the hopping parameter is @xmath376 and @xmath377 , while the magnetic insulator is described by @xmath378 and a gap value @xmath379 .
the coupling parameters @xmath365 and @xmath366 are varied and typically range within several hundreds mev . ) , ( [ h0 ] ) , ( [ hmi ] ) and ( [ htilm ] ) with @xmath380 and @xmath381mev .
the red points represent semiconductor states , while the magenta bands are ferromagnetic insulator states .
the lower panel shows the low - energy behavior around @xmath382 .
notice the spitting of the semiconductor band at @xmath382 due to the effective zeeman term induced by ferromagnetic proximity effect.,scaledwidth=47.0% ] first , we turn on the coupling at the interface between the semiconductor and the mi .
[ figssm1 ] shows the spectrum of a slab containing a semiconductor thin film with @xmath380 atomic layers in contact with a mi .
the coupling at the interface is @xmath383mev .
the semiconductor spectrum is characterized by 10 strongly dispersive bands represented by red curves in the upper panel of fig .
[ figssm1 ] . for an isolated semiconductor ,
these bands are weakly split by the spin - orbit interaction and are double degenerate at @xmath384 .
however , as a result of the magnetic proximity effect , an effective zeeman splitting removes this degeneracy .
this is shown in the lower panel of fig .
[ figssm1 ] for the lowest energy mode that has the minimum inside the insulating gap .
the induced zeeman splitting is entirely a result of the exchange interaction between the mi and the sm layers and is not related to the magnetic field produced by the ferromagnetic mi .
in fact , effective zeeman splittings have been known to induced by anti - ferromagnetic insulators as well through a phenomenon commonly known as exchange bias @xcite . a natural question of practical importance concerns the dependence of the induced zeeman splitting on the parameters of the model .
to get a better intuition of the physics behind the proximity effect , we start with an analysis of the structure of the wave functions of the relevant states .
consider a semiconductor state with an energy within the insulating gap @xmath363 .
when the semiconductor and the mi are coupled , the wave function describing this state will partly penetrate into the magnetic insulator , where it decays exponentially .
these components of the wave functions inside the mi will acquire ferromagnetic correlations equivalent to an effective zeeman field .
hence , the strength of the proximity effect is related to the fraction of the wave function that penetrates the mi .
this qualitative picture of the proximity effect is illustrated in fig .
[ figssm2 ] . in general , the fraction of the wave function that penetrates the insulator depends on three types of factors : i ) the properties of the quasi - two dimensional parent system , in this case the semiconductor film , ii ) the properties of the insulating host system , and iii ) the coupling strength at the interface . ) and @xmath385ev ( blue ) , second semiconductor band ( @xmath386 ) and @xmath385ev ( black ) , and @xmath386 , @xmath387ev ( red ) . for a given set of parameters , a certain fraction of each low - energy wave function
penetrates the insulator leading to ferromagnetic correlations in the semiconductor thin film , .
i.e. , to an effective zeeman field induced by proximity effect . ] the first and last types of factors are qualitatively illustrated in fig .
[ figssm2 ] .
for example , by controlling the value of the on - site energy @xmath388 , i.e. , chemical potential of the semiconductor , one can bring the minima of different semiconductor bands within the insulating gap .
the profiles of the wave functions in the transverse direction ( i.e. , perpendicular to the film ) depend on which band is inside the gap , as shown in fig .
[ figssm2 ] for the first two bands , @xmath389 and @xmath386 .
furthermore , the fraction of the wave function that penetrates the insulator depends on its amplitude at the interface .
this amplitude is significantly different for @xmath389 and @xmath386 , hence one expects significantly different strengths of the proximity effect .
of course , the amplitude at the interface can be also modified by changing the size of the system ( n ) , or by applying a bias potential @xmath390 that can tilt the spectral weight toward or away from the interface .
in addition to the amplitude at the interface , which is determined by the parameters of the semiconductor , the fraction of the wave function that penetrates the mi also depends on the slope at the interface .
in essence , this slope is controlled by the coupling parameter @xmath391 , as illustrated in fig .
[ figssm2 ] . finally , the parameters of the exponential decay inside the insulator , as well as the strength of the ferromagnetic correlations depend on the properties of the insulating host . .
the reference gap is @xmath392 mev .
the wave function amplitude in the absence of an applied bias , @xmath393 , is determined using eq .
( [ psi2 ] ) , while for a linear bias @xmath394 with @xmath395 ev the amplitude is determined numerically.,scaledwidth=47.0% ] next we proceed with a quantitative analysis of the induced zeeman splitting and derive an effective low - energy theory for the ferromagnetic proximity effect .
we consider the situation when the states of the nth semiconductor band with wave vectors near @xmath384 have energies inside the insulating gap .
the states @xmath396 corresponding to vanishing spin - orbit coupling , @xmath397 , form a convenient basis for the low - energy hilbert subspace of interest .
projecting the hamiltonian @xmath16 onto this subspace we obtain @xmath398 where @xmath19 is an effective mass with a value slightly different from the effective mass corresponding to the 3d semiconductor model .
the difference stems from the quasi 2d geometry of the system .
the insulating degrees of freedom can be integrated out and replaced by an interface self - energy .
when projected onto the low - energy subspace , this contribution becomes @xmath399 where @xmath400 is the green function of the magnetic insulator and @xmath401 the amplitude of the semiconductor wave function , both at the interface . from eq .
( [ sigmam ] ) one immediately notice that , neglecting dynamical effects , i.e. , setting @xmath402 , any induced zeeman splitting has to be proportional with the amplitude of the wave function at the interface times the square of the interface transparency . for a vanishing bias potential , @xmath393 ,
the wave function amplitude on the j atomic layer is proportional to @xmath403 $ ] and the amplitude at the interface becomes @xmath404 note that , for clarity , we dropped the spin label , as the amplitude is spin independent .
also note that for large n the amplitude becomes @xmath405 .
the dependence of the induced zeeman splitting on the wave function amplitude is shown in fig .
[ figssm3 ] . and different model parameters ( dots ) .
the straight lines are guides for the eye .
the reference interface coupling is @xmath406 mev .
the rather small deviations from a linear dependence indicate that dynamical effects are negligible , i.e. , neglecting the frequency dependence in eq .
( [ sigmam ] ) is a good approximation.,scaledwidth=47.0% ] as the film thickness @xmath407 is increased , the value of the of the wave function amplitude at the interface drops rapidly , as expected from eq .
( [ psi2 ] ) . due to the proximity effect
, an effective zeeman splitting creates a gap in the nth semiconductor band at @xmath384 , @xmath408 where @xmath409 are the energies of the nth semiconductor band within the magnetic insulator gap ( see lower panel of fig .
[ figssm1 ] ) .
the proportionality between the induced zeeman gap @xmath410 ( dots in fig .
[ figssm3 ] ) and the amplitude of the wave function ( lines in fig .
[ figssm3 ] ) reveal the absence of significant dynamical effects eq .
( [ sigmam ] ) .
this conclusion is further supported by the linear dependence @xmath410 on the square of the interface transparency , @xmath411 , shown in fig .
[ figssm4 ] .
we emphasize that the effective coupling constant that determines the strength of the ferromagnetic proximity effect , @xmath412 , where @xmath413 is a characteristic bandwidth for the magnetic insulator , can be tuned by : a ) modifying the semiconductor film thickness ( see fig .
[ figssm3 ] ) , b ) applying a bias voltage ( see figs
. [ figssm3 ] and [ figssm4 ] ) , and c ) changing the semiconductor - mi coupling ( fig .
[ figssm4 ] ) . ) .
notice the remarkable agreement between the low - energy theory and the exact results and the difference in energy scales between the insulating gap and the induced gap.,scaledwidth=47.0% ] so far we have discussed the dependence of the proximity effect on the properties of the quasi - two dimensional parent system and on the coupling strength at the interface .
next we briefly investigate the role of the insulating host system .
we note that a full analysis would require a treatment of the magnetic insulator beyond the simple mean - field picture used here , but this would not change the results obtained so far . within our mean - field approximation , the mi green function from eq .
( [ sigmam ] ) can be easily evaluated , @xmath414 where @xmath415 are the energies of the insulator bands and @xmath416 the values of the corresponding eigenstates at the interface .
it is convenient to express the green function in terms of the partial density of states @xmath417 . within our model
, this partial density of states becomes @xmath418 with @xmath419 being half of the bandwidth in the insulating phase and @xmath420 the energy values at the middle of the valence and conduction bands for a given wave vector parallel to the interface . at @xmath384
we have @xmath421 .
note that @xmath422 vanishes for values of @xmath423 outside the bandwidth .
( [ rho ] ) , we obtain for the green function the expression @xmath424 . \label{gss}\ ] ] ) for a semiconductor thin film with @xmath380 sandwiched between a mi ( @xmath425ev , @xmath426mev , @xmath427mev ) and an s - wave sc ( @xmath428mev , @xmath429mev ) .
the red points represent states that reside ( mostly ) within the semiconductor , magenta designates the mi bands , while the sc states are blue . due to
the effective zeeman filed , the semiconductor bands are split and only one crosses the chemical potential ( see upper panel ) . as a result of the superconductor proximity effect , a small gap opens at the crossing points ( see lower panel).,scaledwidth=47.0% ] note that the imaginary part of the green function vanishes for values of @xmath423 within the insulating gap . also ,
because the energies of interest are much smaller than the insulator bandwidth , @xmath430 , we can neglect the frequency dependence in eq .
( [ gss ] ) .
finally , within the static approximation , we obtain for the induced zeeman splitting the expression @xmath431 , \nonumber\end{aligned}\ ] ] where the effective coupling constant is @xmath432 . to test the accuracy of this effective low - energy theory
, we compare the values of the induced zeeman splitting predicted by eq .
( [ delgtil ] ) with the numerical calculations .
the results for various values of the insulating gap shown excellent agreement ( see fig . [ figssm5 ] ) . finally , we note that away from @xmath384 the dispersion of the low - energy bands @xmath409 can be obtained by adding the self - energy contribution ( [ sigmam ] ) to the effective theory described by eq .
( [ h0eff ] ) . within the static approximation we have @xmath433 with @xmath410 given by eq .
( [ delgtil ] ) .
explicit calculations for various sets of parameters show that this low - energy theory represents an excellent approximation for all @xmath434 values of interest .
( red ) , @xmath429mev ( black ) , @xmath435 ( yellow ) .
a proximity effect - induced gap opens at finite @xmath434 . as @xmath436
is increased , the minimal gap shifts to lower wave vectors and first increases , then decreases and eventually vanishes at a critical coupling before opening again ( see also fig .
[ figssm8 ] ) . , scaledwidth=47.0% ] next , we turn our attention to the effects induced by the proximity of an s - wave superconductor on the semiconductor - mi heterostructure .
the parameters of the semiconductor - mi interface are fixed , with @xmath426 mev .
we consider a semiconductor thin film with @xmath380 and create a new interface at the free surface of the semiconductor by coupling it to a sc with an s - wave gap @xmath241 , i.e. , we add the terms given by equations ( [ hsc ] ) and ( [ htils ] ) to the total hamiltonian ( [ htot ] ) . the corresponding spectrum is shown in fig .
[ figssm6 ] .
the semiconductor band is split due to the ferromagnetic proximity effect and the chemical potential is tuned so that it crosses only the lower energy mode . when the coupling to the sc is turned on , a small gap opens at low energies due to the sc proximity effect ( see fig .
[ figssm6 ] lower panel ) . before a quantitative analysis ,
let us illustrate qualitatively the behavior of the proximity - induced sc gap .
[ figssm7 ] shows the low - energy spectrum for three different values of the coupling constant @xmath436 . for @xmath437 ( red line )
the bdg spectrum is gapless .
a non - vanishing interface coupling opens a small gap at the crossing points .
the value of this finite wave vector gap increases with @xmath436 , but the gap at @xmath384 decreases .
( black line )
. eventually , at a critical value @xmath438 the gap vanishes at @xmath384 , before opening again for larger couplings ( yellow curve ) .
this closing of the induced gap signals the presence of a quantum phase transition @xcite . ) ( lines ) and the numerical solution of hamiltonian ( [ htot ] ) ( dots ) for two values of the chemical potential . only positive energies are shown .
lower panel : dependence of the induced minimum gap on the interface transparency .
the vanishing of the gap at a critical value @xmath439 reflects a quantum phase transition between a topologically non - trivial sc ( at small values of @xmath440 ) and a trivial s - wave sc ( at large couplings ) [ reference ] .
, scaledwidth=47.0% ] in order to understand this behavior , it is useful to develop an effective low - energy theory for the sc proximity effect .
as shown previously , @xcite dynamical corrections are crucial in capturing the low - energy physics in this case , in contrast to the ferromagnetic proximity effect . using the results obtained in refs .
[ ] , the green function describing the low - energy physics of the heterostructure can be written as @xmath441 where @xmath442 , @xmath410 is given by eq .
( [ delgtil ] ) , @xmath443 , and the effective coupling is @xmath444 .
the wave function amplitude is evaluated at the semiconductor - sc interface and @xmath445 is half of the sc bandwidth . the low - energy spectrum can be obtained by solving the corresponding bdg equation , @xmath446 .
explicitly , we have @xmath447 a comparison between the solution of eq .
( [ effbdg ] ) and the numerical calculations is shown in the upper panel of fig .
[ figssm8 ] .
the good agreement between the two calculations indicates that all the relevant ingredients have been incorporated into the effective low - energy theory .
the dependence of the minimum gap on the interface coupling is shown in the lower panel of fig .
[ figssm8 ] .
note that the critical value of @xmath440 at which the gap vanishes can be obtained from eq .
( [ effbdg ] ) by setting @xmath402 .
explicitly we have @xmath448 semiconductor band has the minimum in the vicinity of the chemical potential .
the low - energy physics in the vicinity of @xmath384 ( upper panel ) is described by the effective theory given by eq .
( [ effbdg ] ) with a coupling constant @xmath449 that includes the amplitude of the @xmath386 state at the interface .
the @xmath389 semiconductor bands cross the chemical potential at larger values of @xmath434 ( lower panel ) and are gaped due sc proximity effect .
note that the gap at large wave vectors does not vanish at a critical coupling.,scaledwidth=47.0% ] the last question that we address in this section concerns the situation when a higher energy semiconductor band , @xmath450 , has the minimum in the vicinity of the chemical potential .
for concreteness , we consider the case @xmath386 .
the physics in the vicinity of @xmath384 is similar with the case studied above .
in addition , the @xmath389 bands cross the chemical potential at some large value of @xmath434 . nonetheless , assuming that the partial density of states of the superconducting metal @xmath451 does not vary significantly with the wave vector , i.e.
, the effective mass of the metal is much grater than the effective mass of the semiconductor , the @xmath389 bands will be gaped and the induced gap is typically grater than the minimum gap near @xmath384 .
hence , the general conclusions of our analysis of the @xmath389 case hold for @xmath450 . to illustrate this point
, we show in fig .
[ figssm9 ] the relevant details of the low - energy spectrum for the case @xmath386 .
in the previous sections , we discussed how majorana states may appear at vortices and edges of various two dimensional spin - orbit coupled semiconductor heterostructures . in this section , we show that majorana fermions can also be realized in the much simpler one - dimensional nanowire set up ( fig .
[ fig : nanowire ] ) . in this
set - up we propose to study a semiconducting nanowire with a sizeable spin - orbit coupling ( _ e.g. _ , inas ) placed on an @xmath0-wave superconductor ( _ e.g. _ , al ) .
an in - plane magnetic field is used to create a zeeman splitting in the nanowire , @xcite which gives rise to the band - structure shown in fig .
[ fig : nanowire](b ) .
the direction of the magnetic field ( parallel or perpendicular to the length of the wire ) that is required to open a gap in the nanowire depends on the exact direction of the spin - orbit coupling in the wire .
the chemical potential in the wire @xmath21 is assumed to be controlled by external gate voltages .
we argue below that , for the zeeman splitting satisfying @xmath452 , a single non - degenerate zero energy state exists at the end of the wire as the only low - energy bound state .
the second quantized operator for this state is again a majorana fermion operator .
this mode can be detected as a zero - bias conductance peak in the stm tunneling spectrum .
on reducing the zeeman splitting so that @xmath453 , the majorana mode , and hence the zero - bias conductance peak , should disappear from the tunneling spectrum .
the bdg hamiltonian for a one - dimensional single band semiconductor with spin - orbit coupling ( which is linear in the momentum ) in proximity to an @xmath0-wave superconductor ( fig .
[ fig : nanowire ] ) , can be written as @xmath454 here the unit vector @xmath281 gives the direction of the effective zeeman field and the unit vector @xmath282 characterizes the spin - orbit coupling . from inspection
it is now clear that this hamiltonian is formally identical to the hamiltonian for the edge states , eq .
[ eq : bdgedge ] , at @xmath278 .
therefore , from the solution of the chiral edge state at @xmath278 for the semiconductor thin film ( sec . x b ) , we conclude that in the topological phase of the wire @xmath246 , the ends of the nanowire support localized zero - energy majorana states . in the previous section , we showed that the end of a nanowire in the topological phase is characterized by a majorana mode .
the majorana modes at the ends of a one dimensional @xmath12-wave superconductor have been shown to lead to distinct signatures in the stm tunneling spectrum .
@xcite in what follows , we analytically and numerically calculate the stm conductance spectrum from one end of our semiconducting nanowire .
we find that , for the zeeman coupling satisfying @xmath455 , this spectrum has a zero - bias conductance peak .
the zero - bias peak disappears as the zeeman splitting is reduced to satisfy @xmath456 .
the tunneling spectrum of a superconducting system , similar to the tunneling conductance of normal systems , can be calculated using the keldysh formalism of non - equilibrium green functions @xcite where the superconducting system is coupled to a tip which is initially in thermal equilibrium at a chemical potential @xmath457 , by an adiabatically increasing tunneling amplitude perturbation @xmath458 . to calculate the current
, we consider an stm tip state @xmath459 where @xmath460 is restricted to the tip .
the tunneling hamiltonian between the stm tip and the superconducting wire can be written as @xmath461.\ ] ] the tunneling current can be related to the keldysh green function of the combined system as @xcite @xmath462\label{eq : current0}.\ ] ] the keldysh green function @xmath463 can be evaluated using the dyson equation @xmath464 where @xmath465 are the unperturbed green functions and @xmath16 is the unperturbed hamiltonian . since the initial systems are in equilibrium , @xmath466 .
substituting these into eq .
( [ eq : current0 ] ) , we get @xmath467\ ] ] where @xmath468 and the tip spectral function at energy @xmath423 is given @xmath469.\ ] ] the exact green functions in the superconductor can be calculated by integrating out the tip using dyson s equation through the relation @xmath470 where the tip induced self - energy is given by @xmath471 the self - energy discussed above is for the normal green function , which is relevant for the tip .
however to describe the superconductor , we need to consider the nambu spinor green function at complex frequency @xmath423 . to be consistent with particle - hole symmetry , the self - energy in nambu spinor notation is given by @xmath472 the retarded and advanced self - energy are given at frequencies @xmath473 respectively .
the dyson equations for the full green function @xmath474 in terms of the self - energy can be decomposed component - wise @xcite as @xmath475 since the starting system is in equilibrium , @xmath476 $ ] and writing @xmath477 where @xmath16 is the unperturbed hamiltonian , the latter equation can be reduced to @xmath478 finally using the equilibrium constraint on the tip , one obtains the expression for the current @xcite @xmath479 to proceed further we must make assumptions about the green function for the stm tip . for a simple stm tip tunneling to the end of the nanowire
, the tunneling can be assumed to occur between one point on the tip and the end of the wire .
furthermore , we can assume that the local density of states of the tip is a lorentzian function @xmath480 where @xmath481 is the band - width about the tip . with these assumptions @xmath363
is given by @xmath482 where @xmath457 is the chemical potential of the tip .
the corresponding self - energy is given by @xmath483 where @xmath484 is the sign of the imaginary part of @xmath423 .
the expression for the current with the above spectral density is given by @xmath485 using the identity @xmath486 and the fact @xmath487 and @xmath488 are analytic in the upper and lower half complex frequency planes respectively , the @xmath423 integral can be replaced by a discrete sum over imaginary matsubara frequencies as @xmath489 the second term in the expression for @xmath490 regulates the singularity in the first term for @xmath491 . using the dyson equations ( eq .
[ eq : dyson ] ) , the expression for the green function @xmath492 can be written in terms of the unperturbed nanowire green function @xmath493 at complex frequency @xmath423 as @xmath494^{-1}|_{\sigma\sigma}\label{eq : gfull}.\ ] ] the unperturbed nanowire green @xmath495 at @xmath496 satisfies the bdg equation @xmath497g(x , x'=a;\omega)=\delta(x - a)\ ] ] for @xmath498 with the boundary condition @xmath499 . here
the end of the wire has been taken to be at @xmath500 . away from the boundary @xmath500 and the contact point of the tip , @xmath501 , the green function
can be expanded @xmath502 where @xmath503\psi_n=0\label{eq : qep}\ ] ] and @xmath504 are a set of vectors that are determined from boundary conditions . the quadratic eigenvalue problem in eq .
( [ eq : qep ] ) can be reduced to a linear generalized eigenvalue problem by defining @xmath505 as @xmath506\psi.\ ] ] the coefficient vectors @xmath504 are determined numerically by solving the boundary conditions @xmath507=\bm 1\\ & g(x=0_-,x'=0;\omega)=g(x=0_+,x'=0;\omega).\end{aligned}\ ] ] as a function of voltage @xmath508 .
for the plot we have taken @xmath509 where @xmath510 for inas , @xmath204 mev for nb , @xmath182 , @xmath511 ev - , and @xmath512 mk .
the different values of @xmath242 in mev are given in the inset .
the distance of the stm tip from the end of the nanowire has been taken as @xmath513 nm .
the plots for @xmath514 mev ( corresponding to @xmath515 t for inas with @xmath516 ) are in the topological phase and show a zero bias peak while the plots for @xmath517 mev do not .
( b ) position dependence of the zero - bias conductance in the topological phase .
the amplitude of the zero - bias conductance is seen to be localized near the end of the wire and corresponds to the localized wave - function of the majorana mode at the end of the wire .
, scaledwidth=80.0% ] the current @xmath518 is calculated by using the green function @xmath519 in eqs .
( [ eq : imat_full ] ) and ( [ eq : gfull ] ) . the conductance @xmath520 obtained by numerical differentiation of the current
is shown in fig .
( [ fig : conductance ] ) . as can be seen from panel ( a ) of the figure , the majorana fermion mode at the end of the wire ( in the non - abelian phase , _
i.e. _ @xmath521 mev ) gives rise to a peak in the conductance at zero - bias @xmath522 .
apart from the zero - bias peak , the bulk states in the nanowire also contribute to the stm conductance at bias voltages above the bulk energy gap .
it is important to emphasize that , unlike @xmath12-wave superconducting nanowires , the tunneling spectrum of the present system ( fig .
( [ fig : conductance ] ) ) has no state ( other than the zero - energy majorana state at the end ) below the bulk energy gap @xmath241 .
therefore the effective mini - gap at the ends is order the bulk energy gap @xmath523 .
as the zeeman splitting is lowered towards the critical value @xmath524 mev , the bulk energy gap closes before re - opening for @xmath525 mev .
for @xmath242 in this regime , the semiconductor is in a regular @xmath0-wave superconducting phase and there is no peak at zero bias . instead there are 2 peaks which are associated with van hove singularities of the bulk quasiparticle spectrum .
these peaks become sharper as the zeeman potential is gradually lowered to zero .
stm spectroscopy can be used to not only verify the existence of the zero energy majorana states localized around the ends but also to study the spatial structure of the zero energy wave - functions .
panel ( b ) of fig .
( [ fig : conductance ] ) shows the zero - bias tunneling conductance as a function of distance @xmath352 from one end of the wire .
the calculation shows that the conductance vanishes at the end of the wire ( the majorana wave - function must vanish at the end of the wire to satisfy the physical boundary condition ) .
the zero - bias conductance then rises to a peak and oscillates with an envelope that decays away from the end of the wire .
the decay length becomes longer as the in - plane magnetic field is tuned towards the critical value @xmath526 mev .
no such zero - bias conductance should be observable for @xmath527 .
thus , the stm experiment from the semiconducting nanowire provides one of the most experimentally feasible probes of majorana fermions in condensed matter systems .
the numerical results discussed in the previous paragraphs ( fig .
( [ fig : conductance ] ) , can be understood analytically in the limit that the stm tunneling strenght is smaller than the thermal broadening @xmath528 . in this limit
the current can be approximated as @xmath529\label{eq : imat0}\ ] ] which can be cast into the more familiar expression for stm current @xmath530 where the stm spectral function is given by @xmath531@xcite .
the corresponding conductivity @xmath532 from the previous subsection , it is clear that the topological phase is characterized by a single majorana state localized at the end .
this zero energy majorana mode is expected to lead to a zero - bias peak in the tunneling spectrum . to see this we consider the @xmath289 nambu green function near @xmath402 where it is dominated by the zero energy pole as @xmath533 where @xmath534 .
the corresponding spectral function is @xmath535 using this in eq .
[ eq : conductance ] , the contribution of the zero - energy state to the conductance becomes @xmath536 where @xmath537 .
therfore stm spectroscopy provides a multi - facted tool to study the properties of the topological quantum phase transition in the wire as the magnetic field is tuned from @xmath538 to @xmath515 t ( which is significantly below the parallel critical field of thin - film nb @xcite ) .
firstly the stm spectra away from the ends provides information about the induced superconducting gap in the wire , which diminishes as a function of applied magnetic field , goes to 0 at the transition and then increases . for magnetic fields above the critical value
, a zero - bias peak appears in the stm spectrum near the ends of the wire .
finally , stm can also be used to probe the spatial structure of the wave - function of this majorana mode .
thus , the stm experiment from the semiconducting nanowire provides one of the most experimentally feasible probes of majorana fermions in condensed matter systems .
let us first recapitulate the most important results contained in each section in this paper . in secs .
[ ii , iii ] we analyze in detail the hamiltonian and the bdg equations for a spin - orbit coupled semiconductor in the proposed heterostructure geometry in the presence of a vortex .
here we provide all the mathematical details , left out in ref .
[ ] , which are necessary to conclusively establish the presence of a non - degenerate majorana mode at the vortex core . in sec .
[ iv ] , we use the same formalism to establish the existence of a majorana mode localized in a `` vortex - like '' defect in the spin - orbit coupling that can be artificially created in a spin - orbit coupled atomic system potentially realizable in an optical lattice . here
we find that , contrary to a previous treatment of the same problem , @xcite the decay length of the zero - energy majorana mode localized in the spin - orbit vortex is inversely proportional to the superconducting gap .
therefore , in the limit of vanishing superconducting gap , the majorana mode will delocalize over the entire system .
we then show in sec .
[ v ] how the same formalism can be used to demonstrate the existence of a majorana fermion mode in a vortex on the surface of a @xmath14 topological insulator , even though in this simpler case an exact solution of the bdg equations is already available .
@xcite in sec .
[ vi ] , we confirm our approximate analytical calculations , which are indicative of the existence of the zero - energy modes in the vortices in the spin - orbit coupled semiconductor , by a full numerical solution of the bdg equations set up on a sphere .
@xcite here we obtain the full bound - state excitation spectrum for the bdg hamiltonian relevant for a vortex - antivortex pair placed at the poles of a sphere .
@xcite in the non - abelian phase of the semiconductor film , this calculation produces a pair of lowest energy states whose energy eigenvalues approach zero exponentially with the radius of the sphere .
this indicates the presence of one exact zero - energy state on each vortex in the limit of infinite inter - vortex separation .
the numerical calculations also show that the excitation gap above the zero energy state , the mini - gap , can be made comparable to the induced @xmath0-wave gap @xmath241 in the semiconductor film .
this surprising result , which was obtained previously for the proximity - induced superconducting state on the surface of a ti , @xcite is now extended to the spin - orbit coupled semiconductor in this paper .
the enhancement of the mini - gap increases the regime of temperature in which any non - abelian quasiparticle is accessible in experiments by many folds ( from @xmath539 to @xmath540 ) . in sec .
[ vii ] we briefly discuss the parameter regime in which the non - abelian topological state is the ground state in the semiconductor film .
the associated topological quantum phase transition ( tqpt ) , which can be accessed by varying any one of the three parameters zeeman coupling ( @xmath20 ) , chemical potential ( @xmath21 ) , and the proximity - induced @xmath0-wave gap ( @xmath241 ) is a transition at which the excitation gap vanishes at wavevector @xmath256 . in sec .
[ viii ] , we analyze the interplay of the zeeman coupling and the proximity - induced @xmath0-wave superconductivity in the presence of spin - orbit coupling . we show that , even though the zeeman coupling can eliminate @xmath0-wave superconductivity in the absence of spin - orbit coupling , the latter can give rise to a _ re - entrant _ non - abelian superconducting state despite the fact that @xmath541 .
apart from the zero - energy majorana states in order parameter defects such as vortices , the non - abelian topological state in the semiconductor film is also characterized by gapless majorana modes at the sample edges . in sec .
[ ix ] , we use the same techniques employed in the earlier sections to demonstrate the existence of these edge modes , which turn out to be _ chiral _ majorana modes because of the explicit breakdown of the time reversal symmetry . in sec .
[ x ] we study the proximity effects in superconductor - semiconductor - magnetic insulator heterostructures starting from a microscopic tight - binding model .
the superconductor and the magnetic insulator are described at the mean - field level .
we determine the excitation spectrum of a slab containing a semiconductor thin film sandwiched between an @xmath0-wave superconductor and a ferromagnetic insulator and identify the dependence of the induced gaps on the parameters of the model . in particular , we study the dependence of the effective zeeman splitting and induced superconducting gap on the thickness of the semiconductor film , the applied bias potential and the strength of the coupling at the interfaces . finally , in sec .
[ xi ] we demonstrate the existence of zero - energy majorana modes at the ends of a one dimensional version of the spin - orbit coupled semiconductor system a semiconducting nanowire .
it has been shown that it may be far simpler to experimentally realize majorana fermions in the one dimensional nanowire system because the zeeman splitting can be induced by a modest in - plane magnetic field , obviating the need for a proximate magnetic insulator .
@xcite we find that the majorana modes at the two ends of the nanowire can be probed by scanning tunneling microscope experiments .
we show by explicit calculations that the majorana modes at the ends of the nanowire give rise to zero - bias conductance peaks in the tunneling spectrum at the ends .
these peaks disappear on lowering the zeeman coupling so that the system settles into the non - topological superconducting state .
furthermore these zero - bias conductance peaks are found to be the only features at bias voltages below the induced superconducting gap in the nanowire .
we believe that the observation of this zero - bias tunneling peak in the semiconductor nanowire is the simplest experiment proposed so far to unambiguously detect a majorana fermion state in a condensed matter system .
we note here that the majorana fermions , being non - abelian particles belonging to the @xmath542 conformal field theory ( i.e. the so - called `` ising anyon '' universality class ) , can not directly be used for universal fault - tolerant topological quantum computation ( tqc ) .
@xcite they can serve as topologically protected quantum memory or can be used in quantum computation along with supplementary unprotected quantum gates requiring only small amounts of error corrections . @xcite since the topological protection for the semiconductor heterostructures in our work is very robust , with the energy gap providing the protection being of the order of the superconducting gap ( 1 - 10k ) itself , our proposed system could serve as an excellent quantum memory in tqc applications . in a recent development , bonderson et al . have shown @xcite that certain dynamic - topology - changing operations , which are feasible in our proposed semiconductor heterostructures , dubbed ising sandwich heterostructures ( ish)in ref .
@xcite , could allow fully fault - tolerant tqc to be carried out using our proosed systems .
thus , in addition to the fundamental significance of the possible non - abelian quantum order and the existence of topological majorana excitations , these semiconductor - superconductor structures may have future application as the basic components of a topological quantum computer .
our proposed non - abelian system is possibly one of the simplest to study experimentally since it involves neither special materials nor exceptional purity nor the application of a high magnetic field .
it is encouraging to note that proximity - induced @xmath0-wave superconductivity has already been realized in a host inas semiconductor film,@xcite which additionally has a sizable spin - orbit coupling .
experimentally , the only new effect that must be introduced to the system is a strong enough zeeman splitting of the spins .
we argue how this can be achieved also via the proximity effect due to a nearby magnetic insulator .
it is important to emphasize that when the spin - orbit coupling is of the rashba type , we require a zeeman splitting which is perpendicular to the plane of the film .
this is because a zeeman splitting parallel to the film does not produce a gap in the one - electron band - structure .
@xcite inducing a perpendicular splitting by applying a strong perpendicular magnetic field is not convenient , because the magnetic field will give rise to unwanted order parameter defects such as vortices .
it is for this reason that we propose to induce the zeeman splitting by the exchange proximity effect of an adjacent magnetic insulator ( we ignore the small coupling of the spins in the semiconductor with the actual magnetic field of the magnetic insulator ) .
more recently , it has been shown that , when the spin - orbit coupling also has a component which is of the dresselhaus type , the appropriate zeeman splitting can also be induced by applying an in - plane magnetic field .
@xcite note that an in - plane magnetic field does not produce unwanted vortex excitations .
the one dimensional version of our system a semiconducting nanowire is also a non - abelian system in the presence of proximity induced _
s_-wave superconductivity and a zeeman coupling .
it is quite exciting that the superconducting proximity effect on an inas nanowire has already been realized in experiments.@xcite in this case , the zeeman coupling can be more easily introduced by applying an external magnetic field parallel to the length of the wire because such a field does produce a gap in the one - electron band structure without producing unwanted excitations in the adjacent superconductor .
this obviates the need for a nearby magnetic insulator .
@xcite in the topological superconducting state of the nanowire ( _ i.e. , _ the zeeman coupling is above a critical value ) , there are zero - energy majorana states localized around the two ends .
such zero energy states can be revealed by zero - bias conductance peaks in stm tunneling experiments at the ends of the wire .
all other contributions to the conductance occur at energies higher than the bulk gap @xmath241 .
there will be no such zero - bias peak when the wire is in the topologically trivial _
s_-wave superconducting state ( _ i.e. , _ the zeeman coupling is below the critical value ) .
such an stm experiment from the semiconducting nanowire will serve as an unambiguous probe for the condensed matter manifestation of a majorana fermion mode .
this work is supported by darpa - quest , jqi - nsf - pfc , and lps - nsa .
st acknowledges doe / epscor grant # de - fg02 - 04er-46139 and clemson university start up funds for support .
even though we found an analytic solution for the null vectors of the matrix for @xmath95 in the region @xmath96 , we could not find such a solution for @xmath543 for @xmath123 . in a previous section we claimed without explicit proof that the solution to this equation can be written in terms of a power - series expansion for @xmath544 .
since we are interested in the solution at large @xmath99 we expect a power - series in @xmath545 to converge . to generate the equation for the power - series for @xmath546 it is convenient to shift to a basis where the @xmath129 part of the matrix is diagonal .
@xmath547 where @xmath548 if @xmath139 is the matrix of eigenvectors of @xmath549 then @xmath550 where @xmath551 is a diagonal matrix such that @xmath552 and @xmath553 .
thus the relevant system of differential equations becomes @xmath554\left(\begin{array}{c}\rho_{\uparrow}(1/r)\\\rho_{\downarrow}(1/r)\end{array}\right)=0\\ & \left(\begin{array}{c}\rho_{\uparrow,1}(1/r)\\\rho_{\downarrow,1}(1/r)\end{array}\right)=s^{-1}\left(\begin{array}{c}\rho_{\uparrow}(1/r)\\\rho_{\downarrow}(1/r)\end{array}\right)\\ & [ \left(\begin{array}{cc } \eta & 0\\ 0 & 0\end{array}\right ) + \eta ( -\partial_r^2+\frac{1}{4 r^2}-\frac{1}{2 r^2}s^{-1}\sigma_z s+2 z\partial_r)+\alpha s^{-1}(\imath \sigma_y)s \partial_r + \frac{\alpha}{2 r}s^{-1}\sigma_x s]\left(\begin{array}{c}\rho_{\uparrow,1}(1/r)\\\rho_{\downarrow,1}(1/r)\end{array}\right)=0\end{aligned}\ ] ] where in the above we note that it is sufficient to provide a series expansion for @xmath555 since @xmath139 is independent of r. the matrix @xmath139 may be written explicitly as @xmath556 where @xmath557 , @xmath558 and @xmath559 .
it is easy to see that our equation is satisfied to order @xmath560 by the choice @xmath561 the upper and lower components of the above solutions are different orders in @xmath127 .
therefore we redefine our spinor as @xmath562 the above solution at lowest order motivates us to modify our ansatz so that the upper and lower component are of the same order in @xmath127 as below @xmath563 where the matrix differential operator @xmath564 is given by @xmath565\nonumber\\ & \left(\begin{array}{cc}1&0\\0&r\end{array}\right).\end{aligned}\ ] ] the terms in the matrix part of the above equation can be separated into 2 categories .
those that preserve the order of a term @xmath566 and those that increase the order of a term to @xmath567 .
the terms that preserve the order are contained within the matrix below @xmath568\left(\begin{array}{cc}1&0\\0&r\end{array}\right)+\textrm{higher order}\\ & = \left(\begin{array}{cc } \eta & 0\\ 0 & 0\end{array}\right)+(2\eta z + \alpha s^{-1}(\imath \sigma_y)s)\left(\begin{array}{cc } 0 & 0\\ 0 & 1\end{array}\right)+\alpha\frac{\sqrt{b^2-c^2}-b } { c}\left(\begin{array}{cc } 0 & \frac{1}{2}+\frac{b r}{\sqrt{b^2-c^2}}\partial_r \\ 0 & 0\end{array}\right)+\textrm{higher order}.\end{aligned}\ ] ] the matrix written explicitly here preserves the order of @xmath566 while the rest of the terms generate terms of order @xmath567 or higher .
we can check this by applying the above matrix to a spinor proportional to @xmath566 .
the resulting spinor is @xmath569\frac{1}{r^n}\left(\begin{array}{c}\rho_{\uparrow}\\\rho_{\downarrow}\end{array}\right)\\ & = \frac{q_n}{r^n}\left(\begin{array}{c}\rho_{\uparrow}\\\rho_{\downarrow}\end{array}\right)\end{aligned}\ ] ] where @xmath570.\ ] ] from this it is clear that the action of the matrix differential operator @xmath564 on the spinor @xmath571 is the non - singular matrix that acts on the spinor @xmath572 in the above equation .
this fact can be used to create a procedure for generating the power series for @xmath573 iteratively . to see how this is the case
consider the stage after the @xmath574 iteration , where the power series has been approximated to @xmath575 to calculate @xmath576 , we note that @xmath577 solving the equation to @xmath578 yields the iterative relation @xmath579 as discussed previously , we already have a starting guess for the solution for @xmath389 so that the residue is of order @xmath560 . using the non - singular matrix @xmath580
we can continue to solve the for the higher - order coefficient by inverting the matrix over the residue .
the above argument provides a procedure for how to construct the power - series for @xmath544 referred to in eq .
22 of the main body of the text .
j. d. sau , r. m. lutchyn , s. tewari , s. das sarma , phys .
lett . * 104 * , 040502 ( 2010 ) .
s. tewari , j. d. sau and s. das sarma , ann . phys . * 325 * , 219 , ( 2010 ) .
r. jackiw and c. rebbi , phys .
d * 13 * , 3398 ( 1976 ) .
r. jackiw and j. r. schrieffer , nuclear physics b * 190 * , 253 ( 1981 ) . | we show that an ordinary semiconducting thin film with spin - orbit coupling can , under appropriate circumstances , be in a quantum topologically ordered state supporting exotic majorana excitations which follow non - abelian statistics .
the key to the quantum topological order is the coexistence of spin - orbit coupling with proximity - induced @xmath0-wave superconductivity and an externally - induced zeeman coupling of the spins . for the zeeman coupling below a critical value ,
the system is a non - topological ( proximity - induced ) @xmath0-wave superconductor .
however , for a range of zeeman coupling above the critical value , the lowest energy excited state inside a vortex is a zero - energy majorana fermion state .
the system , thus , has entered into a non - abelian @xmath0-wave superconducting state via a topological quantum phase transition ( tqpt ) tuned by the zeeman coupling . in the topological phase ,
since the time reversal symmetry is explicitly broken by the zeeman term in the hamiltonian , the edge of the film constitutes a _ chiral _ majorana wire . just like the @xmath0-wave superconductivity
, the zeeman coupling can also be proximity - induced in the film by an adjacent magnetic insulator .
we show this by an explicit model tight - binding calculation for both types of proximity effects in the heterostructure geometry . here
we show that the same tqpt can be accessed by varying the interface transparency between the film and the superconductor . for the transparency below ( above ) a critical value , the system is a topological ( regular ) @xmath0-wave superconductor . in the one - dimensional version of the same structure and for the zeeman coupling above the critical value , there are localized majorana zero - energy modes at the two ends of a semiconducting quantum nanowire . in this case , the zeeman coupling can be induced more easily by an external magnetic field parallel to the wire , obviating the need for a magnetic insulator .
we show that , despite the fact that the superconducting pair potential in the nanowire is explicitly @xmath0-wave , tunneling of electrons to the ends of the wire reveals a pronounced zero - bias peak .
such a peak is absent when the zeeman coupling is below its critical value , _
i.e. , _ the nanowire is in the non - topological @xmath0-wave superconducting state .
we argue that the observation of this zero - bias tunneling peak in the semiconductor nanowire is possibly the simplest and clearest experiment proposed so far to unambiguously detect a majorana fermion mode in a condensed matter system . |
the special supersymmetries known as brst symmetry @xcite and brst - antibrst symmetry @xcite provide a basis of the modern quantization methods for gauge theories @xcite .
these symmetries feature the presence of a grassmann - odd parameter @xmath10 and two grassmann - odd parameters @xmath11 , respectively .
the latter parameters in the @xmath12-covariant schemes of generalized hamiltonian @xcite and lagrangian @xcite quantization ( see @xcite as well ) @xmath13 form an @xmath14-doublet .
these infinitesimal odd - valued parameters may be regarded both as constants and field - dependent functionals , used , respectively , to derive the ward identities and to establish the gauge - independence of the partition function in the path integral approach . in @xcite ,
the brst transformations with a finite field - dependent parameter ( ffdbrst ) in the gauge theories with a gauge group , i.e. , yang mills ( ym ) theories , whose quantum action is constructed by the faddeev
popov ( fp ) rules @xcite , were first introduced by means of a functional equation for the parameter in question and used to provide the path integral with such a change of variables that would allow one to relate the quantum action given in a certain gauge with the one given in a different gauge , however , without solving it in the general setting ( see @xcite as well ) .
the problem of establishing a relation of the fp action in a certain gauge with the action in a different gauge by using a change of variables induced by a ffdbrst transformation was solved in @xcite , thereby providing an exact relation between a finite parameter and a finite change of the gauge - fixing condition in terms of the gauge fermion . in particular , this result implies the conservation of the number of physical degrees of freedom in a given ym theory with respect to ffdbrst transformations , which means the impossibility of relating the yang
mills theory to another theory , whose action may contain , in addition to the fp action , certain non - brst invariant terms ( such as the gribov horizon functional @xcite in the gribov zwanziger theory @xcite , as a consequence of singer s result @xcite ) in the same configuration space .
notice that the solution of a similar problem for arbitrary constrained dynamical systems in the generalized hamiltonian formalism @xcite has been recently proposed in @xcite , whereas for general gauge theories ( possessing reducible gauge symmetries and/or an open gauge algebra ) an exact jacobian generated by ffdbrst transformations in the path integral constructed by the bv procedure @xcite has been obtained in @xcite ( see @xcite as well ) , with a solution of the consistency of _ soft brst symmetry breaking _ @xcite .
recently , we have proposed an extension of brst - antibrst transformations to the case of finite ( both global and field - dependent ) parameters in yang
mills @xcite and general gauge theories @xcite , using the lagrangian and generalized hamiltonian brst - antibrst quantization methods ; also see @xcite . here , we review the origin of finite brst - antibrst transformations and use their properties to study their influence on the quantum structure of ym theoris in the framework of the brst - antibrst setting .
we use the conventions introduced in @xcite . unless otherwise specified by an arrow , derivatives with respect to the fields are taken from the right , and those with respect to the corresponding antifields are taken from the left . the raising and lowering of @xmath15 indices , @xmath16 , @xmath17 ,
is carried out by a constant antisymmetric metric tensor @xmath18 , @xmath19 , subject to the normalization @xmath20 .
the generating functional of green s functions for irreducible gauge theories with a closed algebra in the brst - antibrst lagrangian quantization @xcite is given by @xmath21 \right\ } \label{z(j)}\ ] ] and depends on the sources @xmath22 , with the brst - antibrst - invariant quantum action @xmath23 defined in the same total configuration space @xmath24 as in the fp method , and parameterized by the respective classical fields , the @xmath25-doublet of ghost - antighost fields , and the nakanishi
lautrup fields , @xmath26 , with the grassmann parities @xmath27 , @xmath28 @xmath29 , using the condensed notation .
the quantities @xmath30 , @xmath31 are the respective classical gauge - invariant action and an admissible gauge - fixing bosonic functional , chosen here in the quadratic approximation for ym theories , with @xmath32 defined in a @xmath33-dimensional minkowski space and taking values in the lie algebra of the @xmath34 gauge group , with @xmath35 , @xmath36 in terms of the strength @xmath37 n}+ f^{nop}a^{\mu 0}a^{\nu p}$ ] and @xmath38 which corresponds to the @xmath8-family of gauges with @xmath39 in the fp rules for ym theories .
the remaining terms in ( [ s(a , b , c ) ] ) , i.e. , the gauge - fixing term @xmath40 , the ghost term @xmath41 , and the interaction term @xmath42 , quartic in @xmath43 , are given by @xmath44 b^{m},\nonumber \\
\hspace{-0.9em}&&\hspace{-0.9em}s_{\mathrm{gh } } = \frac{1 } { 2}\textstyle\int d^{d}x \left ( \partial^{\mu}c^{ma}\right ) d_{\mu}^{mn}c^{nb}\varepsilon_{ab } , \label{sadd } \\
\hspace{-0.9em}&&\hspace{-0.9em } s_{\mathrm{add } } = -\textstyle\frac{\xi}{48}\int d^{d}x\ \ f^{mnl}f^{lrs}c^{sa}c^{rc}c^{nb}c^{md}\varepsilon_{ab}\varepsilon_{cd}. \nonumber\ ] ] the action ( [ ymaction ] ) is invariant with respect to the infinitesimal gauge transformations @xmath45 with arbitrary functions @xmath46 , @xmath47 , defied in @xmath48 , whereas the infinitesimal brst - antibrst transformations @xmath49 for ym theories are given in terms of anticommuting generators @xmath50 , @xmath51 leaving the action @xmath52 and the integrand @xmath53 in @xmath54 invariant only in the @xmath55-st order in powers of @xmath56 . to restore the total brst - antibrst invariance of @xmath52 and @xmath53 to all orders in @xmath56 , we have introduced @xcite finite transformations of @xmath57 with a doublet @xmath0 of anticommuting parameters , @xmath58,@xmath59 as a solution of the functional equation @xmath60 for any regular functional @xmath61 invariant under infinitesimal brst - antibrst transformations .
the general solution of ( [ funceq ] ) allows one to restore _
finite brst - antibrst transformations _ in a unique way @xmath62 , @xmath63 where the set of elements @xmath64 forms an abelian two - parametric supergroup with odd - valued generating elements @xmath0 .
the prescription ( [ z ] ) , ( [ funceq ] ) for obtaining finite ( group ) brst - antibrst transformations ( [ finite1 ] ) works perfectly well in the case of changing the configuration space @xmath24 to another representation space in which the action of superalgebra of @xmath65 is realized .
thus , the finite brst - antibrst transformations have also been constructed in the generalized hamiltonian @xcite and lagrangian @xcite brst - antibrst quantization schemes , which is also supported by the frobenius theorem .
the brst - antibrst invariance of @xmath53 implies the validity of the relation @xmath66 with allowance for the fact established in @xcite that the global finite transformations , corresponding to @xmath67 , respect the integration measure .
as we have already mentioned , the jacobian of the change of variables corresponding to the global finite transformations is equal to 1 : @xmath68 we have also examined @xcite the finite field - dependent transformations in the particular case of functionally - dependent parameters @xmath69 , @xmath70 , with a certain even - valued potential , @xmath71 , inspired by infinitesimal field - dependent brst - antibrst transformations with the parameters @xmath72 which respect the gauge independence of the integrand , and therefore also of the vacuum functional @xmath73 , with accuracy up to the terms linear in @xmath74 : @xmath75 . in the case of finite field - dependent transformations with the group elements @xmath76 , which now form a non - abelian 2-parametric supergroup , the superdeterminant of a change of variables takes the form @xmath77^{-2 } , \label{superjaux}\\ \hspace{-0.5em } & & \hspace{-1em } d\phi^{\prime } = \textstyle d\phi\ \exp\left\ { \frac{i}{\hbar}\left [ i\hbar\,\mathrm{\ln}\left ( 1-\frac{1}{2}s^2\lambda\right ) ^{2}\right ] \right\ } \ . \label{superj1}\ ] ]
in view of the invariance of the quantum action @xmath78 under ( [ finite1 ] ) , the change @xmath79 induces in ( [ z(j ) ] ) the transformation of the integrand @xmath53@xmath80 \hspace{-0.1em } \right\}. \nonumber \ ] ] due to the explicit form of the initial quantum action @xmath81 , the brst - antibrst - exact contribution @xmath82 to the action @xmath52 , resulting from the transformation of the integration measure , can be interpreted as a change of the gauge - fixing functional made in the original integrand @xmath53,@xmath83 for a certain @xmath84 , whose relation to @xmath85 is established by ( [ superj3 m ] ) , referred to in @xcite as the _ compensation equation _ for an unknown parameter @xmath86 , which provides the gauge independence of the vacuum functional , @xmath87 . an explicit solution to ( [ superj3 m ] ) , which obeys the solvability condition due to the brst - antibrst exactness of its both sides ,
is given , up to brst - antibrst - exact terms , by the relations @xmath88 \nonumber \\ \hspace{-0.7em}&&\hspace{-0.5em } = \frac{1}{2i\hbar } \delta f\sum_{n=0}^{\infty}\frac{1}{\left ( n+1\right ) ! } \left ( \frac { 1}{4i\hbar}s^{a}s_{a}\delta f\right ) ^{n}\ .
\label{lambda - fsol1}\ ] ] conversely , having considered the equation ( [ superj3 m ] ) for an unknown @xmath74 with a given @xmath4 , we obtain @xmath89 thus , the field - dependent transformations with the parameters @xmath3 amount to a precise change of the gauge - fixing functional . for instance , in order to relate @xmath90 to @xmath91 in the @xmath92-family of gauges , one has to carry out ffdbrst - antibrst transformations with the parameters @xmath93\hspace{-0.2em } ^{n}\hspace{-0.4em}. \label{lamaxi}\ ] ]
in @xcite , the property ( [ superj4 ] ) leads to a so - called modified ward identity , depending on field - dependent parameters , @xmath69 , and therefore also on a finite change of the gauge , in view of ( [ lambda - fsol1 ] ) , @xmath94-\frac{1}{4}\left ( \frac{i}{\hbar}\right ) { } ^{2}\varepsilon_{ab } \right.\right.\nonumber \\
\hspace{-1.3em}&&\hspace{-0.9em } \textstyle\left.\left .
\times j_{a}x^{aa } j_{b}x^{bb}\lambda^{2}(\lambda)\right\ } \left ( 1-\frac{1}{2}\lambda\overleftarrow { s}^{2}\right ) { } ^{-2}\right\rangle _ { f , j } = 1 , \label{mwiclalg}\ ] ] for @xmath95 .
the property ( [ superj4 ] ) also implies a relation describing the gauge dependence of @xmath96 for a finite change of the gauge @xmath97:@xmath98 - \left .
( -1)^{\varepsilon_{b}}\left ( \frac{i}{2\hbar}\right ) ^{2}j_{b}j_{a}\right .
\nonumber\\ \hspace{-0.9em}&&\hspace{-0.5em } \textstyle \left .
\times\left ( x^{aa}x^{bb}\right ) \varepsilon_{ab } \lambda^{2}\left ( \phi|-\delta{f}\right ) \right\rangle _ { f , j } , \label{gdinew1}\ ] ] which means the on - shell ( @xmath99 ) gauge independence of the conventional @xmath9-matrix due to equivalence theorem @xcite . in ( [ mwiclalg ] ) , ( [ gdinew1 ] ) , the symbol @xmath100 for a quantity
@xmath101 denotes the source - dependent average expectation value for a gauge - fixing @xmath102 @xmath103 \right\ } , \ ] ] with @xmath104 . for constant @xmath0 , the relation ( [ mwiclalg ] )
implies an @xmath105-doublet of the usual ward identities at the first order in @xmath0 : @xmath106 and a derivative identity at the second order in @xmath0 : @xmath107 \right\rangle _ { f , j}=0\ .
\label{wilag3}\ ] ]
since the gauge - fixing functional @xmath108 corresponds to the landau gauge , we introduce the gribov horizon functional in the same manner as in @xcite for the fp procedure in the euclidian space : @xmath109 with @xmath110 , @xmath111 being the gribov mass parameter determined from the gap equation and the fp matrix .
a proper brst - antibrst - non - invariant action in the @xmath108 reference frame has the form @xmath112 we determine the gribov zwanziger theory in any @xmath113 gauges ( @xmath92-gauges ) in a way compatible with the gauge - independence of the generating functional of green s functions in @xmath108 , where gribov horizon in the gauge @xmath113 should be determined as @xmath114 ^{2}\big ) , \label{gribovhxi}\ ] ] where @xmath115 is readily determined with account taken of ( [ lamaxi ] ) ; for details , see @xcite .
the construction of the gribov horizon functional @xmath7 in the gauge @xmath113 , starting from @xmath116 in the gauge @xmath108 , may be considered as a generalization of the result @xcite obtained in the brst setting of the problem .
considering , instead of @xmath117 , a functional calculated , e.g. , in the coulomb gauge @xcite , we may use the above recipe to construct the form of @xmath7 in any @xmath118-gauges , including the landau and feynman gauges , and compare them with those obtained by the non - perturbative zwanziger s
recipe @xcite , in order to verify the consistency of zwanziger s recipe with the requirement of gauge independence for the corresponding vacuum functionals .
we have proposed the concept of finite brst - antibrst and ffdbrst - antibrst transformations for yang
mills theories in the @xmath105-covariant lagrangian quantization .
the jacobian of the change of variables generated by ffdbrst - antibrst transformations with functionally dependent parameters is exactly calculated .
it is established that quantum ym actions in different gauges are related to each other by means of ffdbrst - antibrst transformations with functionally dependent parameters obtained as solutions of the compensation equation .
a new ward identity and the gauge dependence problem for finite changes of the gauge for the generating functional of green s functions are derived and studied . the gribov - zwanziger theory in a brst - antibrst formulation is suggested and the gribov horizon functional , @xmath119 , in the feynman and arbitrary gauges , starting from @xmath120 given in landau gauge , is suggested in a way compatible with the gauge independence of the @xmath9-matrix .
concluding , note that gribov s problem for ym theories in brst - antibrst setting may be elaborated with help of composite fields technique following to the results @xcite obtained within brst setting of the problem .
the study has been supported by the rfbr grant , project no . 12 - 02 - 00121 , and by the grant for lrss , project no .
it has also been partially supported by the ministry of science of the russian federation , grant no . | following our recent study [ p.yu .
moshin , a.a .
reshetnyak , nucl . phys .
b 888 ( 2014 ) 92 ] , we discuss the notion of finite brst - antibrst transformations , with a doublet @xmath0 , @xmath1 , of anticommuting ( both global and field - dependent ) grassmann parameters .
it turns out that the global finite brst - antibrst transformations form a 2-parametric abelian supergroup .
we find an explicit jacobian corresponding to this change of variables in the theories with a gauge group .
special field - dependent brst - antibrst transformations for the yang mills path integral with @xmath2-potential ( functionally - dependent ) parameters @xmath3 generated by a finite even - valued functional @xmath4 and the anticommuting generators @xmath2 of brst - antibrst transformations , amount to a precise change of the gauge - fixing functional .
this proves the independence of the vacuum functional under such brst - antibrst transformations and leads to the presence of modified ward identities .
the form of transformation parameters that induces a change of the gauge in the path integral is found and is exactly evaluated for connecting two arbitrary @xmath5-like gauges .
the finite field - dependent brst - antibrst transformations are used to generalize the gribov horizon functional @xmath6 , in the landau gauge of the brst - antibrst setting in the gribov zwanziger model , and to find @xmath7 corresponding to general @xmath8-like gauges in the form compatible with a gauge - independent @xmath9-matrix . 2 |
galaxies with massive star formation are the primary source that ejects metals into galactic halos and the intergalactic medium .
the properties of outflows driven by sn explosions and stellar winds are fundamental to the understanding of feedback processes of galaxies . in previous studies , the energy and metals of outflows
are generally inferred from fitting observed spectra with thermal models .
the x - ray line emission , however , can arise not only from hot plasma , but also from its interaction with neutral cool gas .
for example , highly ionised ions in the solar wind can readily pick up electrons from neutral species of a comet .
these electrons , captured in excited states of the ions , will cascade down , leading to x - ray line emission .
this process is called charge exchange x - ray emission ( cxe , also called charge transfer ) and explains the bright cometary x - ray emission ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
the process can be represented by : @xmath1 where a highly ionised ion @xmath2 ( like ) picks up an electron from a neutral species
@xmath3 ( like h ) and produces an excited ion @xmath4 . for historical studies and recent developments of the cxe , we refer the readers to @xcite and references therein .
the charge - exchange cross - sections are quite large compared to the cross - sections of electron collisional excitation . for example , at collision velocities of a few hundred km / s , the charge - exchange cross - sections of and are @xmath5 @xmath6 , compared to those of electron collisional excitation ( @xmath7@xmath6 ) .
this makes the cxe important even though it takes place in a very thin region compared to the tenuous hot gas .
different from the thermal emission , the cxe contributes only emission lines .
if the x - ray line emission , or part of it , is due to the cxe , the measurement of the thermal and chemical properties of the hot outflow based on thermal - only models will be misleading .
to correctly understand the hot outflow , it is crucial to determine the origin of the x - ray line emission and to separate the contributions from the cxe and the thermal emission . with the unprecedented spatial resolution of , the extra - planar diffuse x - ray emission of a number of edge - on disk galaxies has been found to be correlated with h@xmath0 emission ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) ; in less - inclined galaxies , x - ray emission is also found to be correlated morphologically with h@xmath0 emission ( e.g. , * ? ? ?
@xcite speculated on the importance of the cxe between the hot outflow and cool halo gas for starburst galaxies . in the scenario of the cxe , h@xmath0 emission represents entrained cool gas clouds / filaments , while the x - ray emission arises from their interfaces with the surrounding hot plasma .
the spatial correlation between the h@xmath0 and x - ray emissions is thus naturally explained .
a marginal detection of an emission line at 0.459 kev is reported for the x - ray cap above the disk of the starburst galaxy of m82 , which may be due to the cxe of ( @xmath8 ) @xcite .
the direct evidence for the cxe occurring in m82 is from the line ratios of the k@xmath0 triplet @xcite .
the k@xmath0 triplet of he - like ions is a powerful diagnostic that can be used to reveal the origin of the x - ray line emission ( for a recent review , see * ? ? ?
it consists of a resonance line , two inter - combination lines , and a forbidden line . for a thermal plasma in ionization equilibrium ,
the electron collisional excitation is efficient and favours the resonance line , while for the cxe , the electron downward cascade favours the triplet states and thus the forbidden line .
therefore , the line ratios of the triplet can be used to determine the origin of the x - ray line emission . in a previous paper @xcite
, we analysed the k@xmath0 triplets of he - like ions of , , and of m82 observed by the reflection grating spectrometers ( rgss ) .
the triplet is found to be dominated by the forbidden line , and the intensity ratio of the triplet is consistent with an origin of the cxe .
the flux contributions of the cxe are 90% , 50% , and 30% to the , , and triplets , respectively .
the triplet was also used to show the possible presence of charge exchanges between the hot gas and the cool gas spiral in the inner bulge of m31 @xcite . in this paper , we study the triplet for a sample of additional eight star - forming galaxies , in comparison with m82 . in all these galaxies ,
the forbidden lines are comparable to or stronger than the resonance lines , which is in contrast to the thermal prediction .
we discuss the other possible mechanisms that can produce a relatively strong forbidden line , such as a collisional non - equilibrium - ionization recombining / ionizing plasma and resonance scattering , which are found to be not favoured for our sample galaxies .
it shows that the cxe may be a common process in galaxies with massive star formation .
the paper is structured as follows .
we describe the observation sample in 2 .
the analysis results are presented in 3 .
the conclusion and discussion are given in 4 .
errors are given at 1 @xmath9 confidence level .
.list of the sample [ cols="<,^,^,^,<",options="header " , ] : : note : for the meanings of the symbols , see eq ( 2 ) .
@xmath10@xmath11 and @xmath12 are in units of @xmath13photons / s/@xmath6 .
@xmath14@xmath15 and @xmath16 are the wavelength shift and dispersion , respectively .
@xmath17@xmath18 and @xmath19 refer to the continuum - subtracted fluxes within 18.5 19.5 and 21.4 22.4 , respectively .
@xmath20 the two data points between 21.3 and 21.5 are neglected when fitting the triplet of m82 a ( see discussion in 3.2 ) .
the observed triplets of the galaxies are plotted in fig .
3 . for the edge - on galaxies of ngc 253 and ngc 4631 , the triplets are heavily blended .
the forbidden lines are relatively well separated for others .
except for the c region of m82 , fig .
3 shows that for all galaxies the forbidden lines are comparable to or stronger than the resonance lines .
to study the line ratios of the triplets quantitatively , we fit a model of three gaussians and a constant continuum to each of the observed triplets , which is written as @xmath21 + f_{\rm con},\ ] ] where @xmath22 ( 21.6 , 21.8 and 22.1 ) are the wavelengths of the resonance , inter - combination , and forbidden lines , respectively , @xmath23 and @xmath12 the corresponding fluxes , @xmath16 the dispersion , and @xmath15 the wavelength shift . in total , we have 15 bins of data , with 6 parameters .
the errors are estimated using the bootstrap method .
the wavelength dispersions are largely determined by the spatial extent , while the wavelength shifts are mainly determined by the spatial offset between the peak x - ray line emission region and the adopted nominal center .
as the rgs is slit - less , we can not remove point sources from the rgs spectra .
therefore , the continua of the rgs spectra are dominated by point sources and instrument background . as a result
, we choose to fit the triplets with a local continuum within a fitting region extending beyond the triplets .
the fitted continua are generally close to the fluxes at the wings of the triplets .
the fitting results are plotted in fig . 3 and listed in table 2 .
the often - used line ratio @xmath24 which is a diagnostic of temperature of thermal plasmas , is calculated and listed in table 2 . because the inter - combination lines are poorly constrained due to blending
, the ratio @xmath25 is not included . for a thermal plasma in ionization equilibrium ,
the ratio @xmath26 is smaller than 1.2 for temperatures higher than 0.1 kev .
all the fitted @xmath26 ratios are larger than 1.2 , except for m82 c , which is discussed in the next section .
therefore , the majority of the triplets of the sample galaxies can not come from the ionization - equilibrium hot gas directly , but may originate from the cxe . for the cxe , the experiment measured ratio @xmath27 @xcite . for galaxies of m51 , m83 , and m82 b ,
the triplets are relatively well resolved , and the fitted @xmath26 ratios @xmath28 are in good agreement with the predicted value of the cxe . for m94 and ngc 2903 ,
the triplets are also relatively resolved , but their @xmath26 ratios are @xmath29 , which indicate multiple origins . except for m82 c and ngc 253 b ,
the @xmath26 ratios of other galaxies are consistent with the value of the cxe within 1 @xmath9 errors .
lines and the triplets of m82 in three different regions . for clarity ,
errors are not plotted.,title="fig : " ] lines and the triplets of m82 in three different regions . for clarity ,
errors are not plotted.,title="fig : " ] the observed triplet of m82 a in fig .
3 shows a feature around 21.4 , which can not be fitted with the model of three gaussians . as shown in fig .
1 , for m82 , the rgs cross - dispersion direction is not parallel to the outflow direction , and the triplet will have a different wavelength shift for different cross - dispersion position .
therefore , in the top panel of fig .
4 , we plot the cross - dispersion distributions of m82 for photons between 21.25 and 21.55 and between 22.1 and 22.4 . the photons between 22.1 and 22.4
are peaked around the center , while the photons between 21.25 and 21.55 have a plateau from -30@xmath30 to 90@xmath30 .
thus , we divide the cross - dispersion direction into three regions : b ( -90@xmath30 -30@xmath30 ) , a ( -30@xmath30 30@xmath30 ) , and c ( 30@xmath30 90@xmath30 ) , which are denoted in fig .
the triplets of the three regions are plotted in the bottom panel of fig .
the triplets of the three regions , however , are not simply shifted according to the spatial offsets , but have different line intensity ratios .
the peak around 21.5 of the c region can not be identified as the forbidden line of the triplets , otherwise , it will have a shift of 0.6 , which is larger than the shift provided by the spatial offset .
instead , it is identified as the resonance line , as evidenced by the relative shift from the line at 18.8 . for the b region , the forbidden line is dominant ; for the c region , however , the resonance line is dominant .
the three lines peak at 21.4 , 21.9 , and 22.2 of the a region can not be fitted with a single triplet lines , but are a mixture of features of b and c regions . as a result , we have neglected the two data points between 21.3 and 21.5 when fitting the triplet of m82 a. the different ratios of the triplets of different regions indicate complicated processes and multiple origins of the x - ray emission of m82 .
while the origin of triplets can be tested through their line ratios , we can not directly tell the origin of other lines such as the @xmath31line .
however , we can still get some hints by comparing the spatial distribution of such lines with that of the triplet in the cross - dispersion direction . in fig . 5 , we plot the cross - dispersion distributions of the triplet ( 21.422.4 ) , the @xmath31line ( 18.519.5 ) , and the emission between 14 and 18 for six relatively bright galaxies
we see that the profiles of the @xmath31line are similar to those of the triplet , except for m94 , the @xmath31line of which is more centered .
the similarity between the distributions of the @xmath31line and the triplet indicates that they may have a similar origin .
the distributions of the emission between 14 and 18 of the bright galaxies ( m82 , ngc 253 , m51 , and m94 ) are more centered compared to the corresponding profiles of the triplet .
it may be due to the concentration of the point sources of power - law continuum , as suggested by the increasing continua at 14 18 in the spectra of fig .
2 . nevertheless , for ngc 253 and m51 , the profiles of the emission between 14 and 18 and the triplet appear to be similar outside the center .
if most of the @xmath31line emission is indeed primarily from the cxe , it is then interesting to calculate the observed flux ratios of the @xmath31line to the triplet ( @xmath32 ) , which are listed in table 2 .
the charge exchange cross sections are similar for and ions , and the efficiencies of producing a @xmath33 photon are also similar for the resulting excited and ions @xcite .
if only a single charge exchange collision occurs ( collisionally thin case ) , the flux ratio @xmath32 reflects the composition ratio of / of the interacting hot gas . the observed flux ratios of @xmath32 are @xmath34 , which means the ratios of / are also @xmath34 for the interacting hot gas .
these ratios correspond to a temperature of @xmath35 k in a thermal equilibrium plasma .
we have analysed the triplets of a sample of nine nearby star - forming galaxies .
all the observed @xmath26 ratios ( eq.3 ) of the galaxies are larger than 1.2 , which is the upper limit for a thermal plasma with temperatures higher than 0.1 kev . for ngc 253 , m51 , m83 , m61 , ngc 4631 , and the antennae , the observed @xmath26 ratios are consistent with the laboratory measured ratio of the cxe , while for m94 and ngc 2903 , the observed @xmath26 ratios indicate multiple origins . for m82 ,
the triplets of different regions show different line ratios .
the cross - dispersion profiles of the @xmath31line are similar to those of the triplet , indicating that the @xmath31lines may have a similar origin .
below we discuss the cxe and other possible mechanisms for the triplet emission of a high @xmath26 ratio . in our sample , three galaxies ( m51 , m61 , and m94 )
show low level agn activities , which can complicate the interpretation of the triplets .
a high @xmath26 ratio is also expected from the recombining ions due to the photoionization by an agn . however , at least for m61 and m94 , the analysis of the images indicates that the agn is weak and unlikely to be important .
the image of the nucleus of m61 reveals that the soft x - ray emission traces an ultraviolet star - forming spiral down to an unresolved core ( 1.5@xmath30 ) , and the soft x - ray emission of the core accounts for only 15% of the total soft x - ray emission in the central 8@xmath30 region @xcite . for m94
, the x - ray emission is resolved into a diffuse component and a cluster of stellar x - ray sources , which appear to follow a recent episode of starburst , and shows no evidence of an agn @xcite . a collisionl non - equilibrium - ionization ( nei ) recombining / ionizing plasma can also produce a relatively strong forbidden line ( e.g. , * ? ? ?
in a nei recombining plasma , the recombination lags the cooling ( e.g. , * ? ? ? * ) , and the delayed recombination can produce a strong forbidden line . for the star - forming galaxies studied here , however ,
the x - ray emission is spatially correlated with the filamentary h@xmath0 emission .
this fact is consistent with the scenario that x - ray emission arises from the interfaces between the hot outflow and the entrained cold gas , but would be difficult to explain for the nei model in which the emission should come from the bulk of the outflow , where the low density regions will be the most over - ionized . furthermore , most of the triplet emission arises from the central region of the galaxies . according to @xcite , @xmath36 in an adiabatic free - flow .
the initial temperature is generally estimated to be around @xmath37 k ( e.g. , * ? ? ?
if we adopt a star - forming region of 100 pc ( as in the case of m82 ) , when it extends to 0.5 kpc , the temperature decreases a factor of 10 , still far above the fully stripping temperature of ion ( @xmath38 k ) .
a mass - loading factor of 10 will help the plasma to cool to about @xmath39 k , the peak ion temperature .
but the mass - loading leads to a high density , hence an equilibrium condition .
thus , especially in the central regions where most of the triplet emission arises from , a nei recombining plasma seems unlikely to be important .
it is also possible to produce a high @xmath26 ratio in an ionizing nei plasma ( e.g. , * ? ? ?
* ) , such as supernovae remnants ( snr ) , in which the heating generally precedes the ionization .
the inner - shell collisional ionization of ions can lead to excited ions and to the enhanced forbidden line emission .
however , this happens only in a short transition region when the fraction is large .
in fact , the observed triplets of snr generally show a thermal - like ratio ( e.g. , * ? ? ? * ; * ? ? ? * ) . for the observed ( accumulated plasma ) spectra of the sample galaxies , the @xmath31emission is stronger than the triplet , which means that the ion fraction is negligible .
another process that may have effect on the line ratios is resonance scattering .
the resonance line has a larger optical depth due to its large oscillator strength compared to the forbidden line .
if the optical depth is large enough , the resonance line intensity can then be re - distributed from the central optical thick region to the outer optical thin region @xcite , and the ratio @xmath26 is increased in the central region and reduced outside . for the off - disk triplets of m82 b and ngc 253 b , the @xmath26 ratios are higher than the ratio of thermal models , which can not be due to resonance scattering .
furthermore if resonance scattering is important for the resonance line , so is for the @xmath31line . for thermal models ,
the ratio of @xmath32 of @xmath40 corresponds to an ion ratio of /@xmath41 , which means the scattering optical depth of the @xmath31line is @xmath42 times larger than that of the resonance line .
thus , if resonance scattering is important , the profiles of the @xmath31line should be flatter than those of the triplet , which is not observed in fig .
5 . for the cxe , we can estimate the required property of the hot gas assuming that all the oxygen ions are in state and can participate in the exchange .
then , the triplet flux is @xmath43 , where @xmath44 is the emission radius and assumed to be 0.5 kpc , @xmath45 the velocity ( 1000 km s@xmath46 ) , @xmath47 the distance ( 4 mpc for m82 ) , and @xmath48 the ion density .
taking the solar oxygen abundance ( @xmath49 ) , we find that an electron density of @xmath50 @xmath51 is enough to produce the observed central triplet flux of m82 ( @xmath52 photons @xmath53s@xmath46 )
. the effective interacting surface area can , in principle , be substantially greater than @xmath54 , depending on the topology of the cool gas and the dynamics of the turbulent mixing of the hot and cool gases .
the lack of spatial information limits further estimates , but the details of the cxe process can be better constrained with future spatially - resolved high - resolution x - ray spectroscopy .
since the majority of the triplets are consistent with the cxe and the emission between 14 and 20 is likely to have a similar origin , one can not measure the temperature and metal abundances of the hot gas assuming thermal models alone .
instead , one needs to incorporate the cxe in the spectral modelling .
the relative abundances of different ions can be obtained by comparing the observed spectra with the cxe models , as has been done for the solar wind based on x - ray spectra of comets ( e.g. , * ? ? ?
* ; * ? ? ?
* ) . for m82 , the relative abundances of o , ne , and mg were found to be similar to the solar values based on the cxe fluxes of their k@xmath0 triplets @xcite . in some galaxies , one may need to consider the cxe and the thermal emission simultaneously , as shown by the @xmath26 ratios of m94 and ngc 2903 and the triplets of m82 .
for such a task , a detailed modelling of the cxe is needed .
the cxe cross sections of heavy elements of ne , mg , and fe need to be calculated and/or experimentally measured .
we appreciate our referee s critical and constructive report . this research has made use of archival data .
xmm - newton is an esa science mission with instruments and contributions directly funded by esa member states and the usa ( nasa ) .
we also use images from dss-2 ( poss - ii ) , which was made by the california institute of technology with funds from the national science foundation , the national geographic society , the sloan foundation , the samuel oschin foundation , and the eastman kodak corporation .
jrl and sm thank the chinese academy of sciences for financial support and qdw acknowledges the support by nasa via the grant nnx10ae85 g . | properties of hot gas outflows from galaxies are generally measured from associated x - ray line emission assuming that it represents atomic transitions in thermally excited hot gas .
x - ray line emission , however , can also arise from the charge exchange between highly ionised ions and neutral species .
the k@xmath0 triplet of he - like ions can be used as a powerful diagnostic , because the charge exchange x - ray emission ( cxe ) favours the inter - combination and forbidden lines , while the thermal emission favours the resonance line .
we analyse the triplet of a sample of nine nearby star - forming galaxies observed by the reflection grating spectrometers . for most galaxies ,
the forbidden lines are comparable to or stronger than the resonance lines , which is in contrast to the thermal prediction . for ngc 253 , m51 , m83 , m61 ,
ngc 4631 , and the antennae galaxy , the observed line ratios are consistent with the ratio of the cxe ; for m94 and ngc 2903 , the observed ratios indicate multiple origins ; for m82 , different regions show different line ratios , also indicating multiple origins .
we discuss other possible mechanisms that can produce a relatively strong forbidden line , such as a collisional non - equilibrium - ionization recombining / ionizing plasma , which are not favoured .
these results suggest that the cxe may be a common process and contribute a significant fraction of the soft x - ray line emission for galaxies with massive star formation . atomic processes plasmas
ism : jets and outflows ism : abundances galaxies : starburst galaxies : individual : ( m82 , ngc 253 , m51 , m61 , m83 , m94 , ngc 2903 , ngc 4631 , the antennae ) x - rays : galaxies |
marcy et al . ( 2001 ) have discovered two planets about the nearby m dwarf star gj@xmath0876 .
a preliminary fit of the stellar radial velocity ( rv ) variations due to two unperturbed kepler orbits implies that the orbital periods of the two planets are nearly in the ratio 2:1 .
this resonance is an analog to the orbital resonances among the satellites of jupiter and saturn ( e.g. , peale 1999 ) , but it is the first to be discovered among extrasolar planetary systems .
table [ table1 ] shows the system parameters from the marcy et al .
analysis based on data taken at both the keck and lick observatories and an adopted stellar mass of @xmath26 .
the orbital periods are approximately 30 and 60 days .
the reduced chi - square statistic of @xmath27 indicates that the two - kepler system is an adequate fit to the data .
however , the rather large minimum masses of the planets of 0.56 and 1.89 @xmath28 ( jupiter masses ) mean that the two - kepler fit may not be a good determination of the system characteristics because the large mutual perturbations will ensure that the orbits deviate from kepler orbits .
in fact , substitution of the marcy et al . parameters from table [ table1 ] ( with @xmath29 for both planets , where @xmath30 is the inclination of the orbital plane to the plane of the sky ) into a calculation of the perturbed orbital motions beginning at the specified initial epoch leads to large variations in the orbital elements that were assumed constant in the fit . on the other hand , projecting the orbital parameters in table [ table1 ] to those at a different epoch within the time span of the data set leads to much less variations in the orbital elements and apparent long term stability of the system ( marcy et al .
for some choices of epoch , the system is not in the 2:1 resonance , and it eventually becomes unstable ( laughlin & chambers 2001 ) . clearly , a dynamical fit of system characteristics to the rv observations is necessary to constrain the parameters that define an rv curve for gj@xmath0876 . the publication of the data set by marcy et al .
( 2001 ) allowed laughlin & chambers ( 2001 ) to perform such a fit .
they assumed that the two planets are on coplanar orbits , and used two methods to minimize @xmath31 as a function of the initial system parameters . starting with the 2-kepler fit , a levenberg - marquardt minimization scheme driving an @xmath32-body integrator was used to find a _
local _ minimum in @xmath31 .
a second method uses a genetic algorithm combined with a simple model for the variations of the orbital elements due to the 2:1 orbital resonance to search for the _ global _ minimum in @xmath31 .
both methods converged to similar solutions .
the dynamical fits are sensitive to @xmath33 since the determined masses ( and perturbations ) grow as @xmath30 is decreased .
the best - fit solutions obtained using the levenberg - marquardt method are included in table [ table1 ] for both the keck@xmath34lick data and keck data alone .
the keck data is much more accurate than the lick data and the solution for this data alone has a minimum @xmath35 with an rms scatter of @xmath36 for @xmath37 .
although the lick data is less accurate than the keck data , the longer time span of observation that includes the lick data may give the better solution .
the solution for the keck@xmath34lick data has a minimum @xmath38 with an rms scatter of @xmath39 for @xmath40 , but the @xmath31 minimum is broad , with several nearby solutions giving nearly as good a fit .
the conclusion of laughlin & chambers is that the broad minimum in @xmath31 allows probable values of @xmath41 .
( rivera & lissauer 2001 examined mutually inclined orbits and got similarly small @xmath31 values for several different configurations , but they confirmed the laughlin & chambers results for coplanar orbits . ) figure [ figure1 ] shows the trajectories of the motions for 3100 days ( the average periapse precession period for both planets ) for the two laughlin - chambers best - fit solutions in plots of @xmath42 versus @xmath43 , where @xmath44 is the eccentricity of the @xmath45th planet ( with @xmath46 and @xmath47 for the inner and outer planets , respectively ) , @xmath48 are the two lowest order , eccentricity - type mean - motion resonance variables at the 2:1 commensurability , and @xmath49 and @xmath50 are the mean longitude and longitude of periapse of the @xmath45th planet .
the two laughlin - chambers solutions place the system deep in both mean - motion resonances , with @xmath11 and @xmath7 librating about @xmath4 with remarkably small amplitudes .
the simultaneous librations of @xmath11 and @xmath7 about @xmath4 mean that the secular resonance variable @xmath51 also librates about @xmath4 .
although the parameters will be more tightly defined as more data is acquired , the almost singular nature of the low amplitude librations indicates that the real system is most likely indeed locked in multi - resonance librations at the 2:1 mean - motion commensurability .
this means that the line of apsides of the two orbits are nearly aligned with conjunctions of the two planets always occurring very close to the periapse longitudes .
the resonance configuration ensures that the planets can never make close approaches in spite of their large masses , and barring external perturbations or an unreasonably high dissipation of tidal energy in the planets , the system should be stable for the lifetime of the star .
in fact , we find that the librations are stable even for amplitudes of the inner planet resonance variable @xmath11 approaching @xmath14 if we induce larger amplitudes of libration of the resonance variables by changing the initial value of the mean anomaly of the inner planet .
the existence of the mean - motion resonances means that the assumption that the orbits are nearly coplanar should be correct , and such coplanarity is consistent with the accretion of planetary bodies in a nebular disk surrounding the forming star .
we shall assume that the orbits are coplanar hereinafter .
furthermore , there is not much to distinguish the two laughlin - chambers solutions in figure [ figure1 ] , and we shall assume the parameters appropriate to the solution based on both the keck and lick data .
figure [ figure2 ] shows the variations in the semi - major axes and eccentricities of both planets for @xmath52 days for the laughlin - chambers keck@xmath34lick solution .
the small amplitude librations of the resonance variables about @xmath4 ensure that both the eccentricities and semi - major axes have little variation in spite of the large mutual perturbations .
this is why the 2-kepler fit of marcy et al .
( 2001 ) could produce a fit that was not too bad .
the fact that both 2:1 mean - motion resonance variables librate about @xmath4 in figure [ figure1 ] contrasts with the geometry of the io - europa 2:1 orbital resonances at jupiter , where the resonance variable involving io s longitude of periapse , @xmath11 , librates about @xmath4 but the resonance variable involving europa s longitude of periapse , @xmath7 , librates about @xmath9 ( e.g. , peale 1999 ) .
this means that in the io - europa case the lines of apsides are anti - aligned , with the periapses @xmath9 apart ( i.e. , @xmath8 librates about @xmath9 ) .
conjunctions occur when io is near periapse and europa is near apoapse . in
2 we explain why the gj@xmath0876 and io - europa systems have different resonance configurations by considering a condition for stable simultaneous librations of the two mean - motion resonance variables .
in particular , the longitudes of periapse should precess in the same direction at the same average rate , so that the relative alignment of the lines of apsides is maintained . in the io - europa case ,
the orbital eccentricities are sufficiently small that the precession rate @xmath53 is dominated by a single term lowest order in eccentricities and containing @xmath54 .
the overall sign of the coefficient for this term is @xmath55 for @xmath56 ( inner satellite ) but @xmath57 for @xmath58 ( outer satellite ) , and retrograde precessions of both satellites require @xmath11 to librate about @xmath4 and @xmath7 to librate about @xmath9 . in the case of gj@xmath0876 ,
the eccentricities are sufficiently large that there are large contributions to the precession rates from higher order terms whose cosine arguments are linear combinations of the resonance variables . summing over all contributing terms , coincident retrograde precessions of the gj@xmath0876 planets require both @xmath11 and @xmath7 to librate about @xmath4 .
if the orbits of the two planets about gj@xmath0876 were originally much further apart , with the ratio of their mean motions considerably greater than 2:1 , any process that drives the orbits toward each other could lead to capture into the 2:1 orbital resonances that we observe . in
3 we describe a particular migration process due to the gravitational interaction between the planets and the nebular disk surrounding the young gj@xmath0876 .
we note the simulations by bryden et al .
( 2000 ) and kley ( 2000 ) , which show that two planets that are massive enough to open gaps individually in the gas disk can rather quickly clear out the disk material between them , if they are not separated too far .
disk material outside the outer planet exerts torques on the planet that are not opposed by disk material on the inside , and the outer planet migrates toward the star on the disk viscous time scale . any disk material left on the inside of the inner planet exerts torques on the inner planet that push it away from the star .
thus the condition of approaching orbits necessary to form the resonances is established .
we discuss the effect of planet - nebula interaction on orbital eccentricities and note that the gj@xmath0876 system is in the interesting regime where it is uncertain whether eccentricity damping or growth is expected because of the rather large planet - star mass ratio of the outer planet .
( although the mass of the outer planet , @xmath59 , for the laughlin - chambers keck@xmath34lick solution is only @xmath60 , @xmath61 because the stellar mass @xmath62 . ) in 4 we present the results of a series of numerical orbit integrations in which the orbits of the gj@xmath0876 planets , initially far from the 2:1 mean - motion commensurability , are forced to approach each other .
( the numerical methods are described in the appendix . )
we show that capture into the two 2:1 mean - motion resonances and the secular resonance is certain if the orbital eccentricities start reasonably small and the rate of migration is not too fast .
if there is no eccentricity damping , the eccentricities of both planets increase rapidly after resonance capture and exceed the observed values after a very short migration of the resonantly locked planets .
eccentricity damping of the form @xmath63 , where a dot over a symbol denotes @xmath64 and @xmath23 is the forced migration rate , leads to a termination in the eccentricity growth , and the eccentricities reach equilibrium values that remain constant for arbitrarily long migration in the resonances . we find that significant eccentricity damping with latexmath:[$|{\dot e}_i / e_i| \gg the gj@xmath0876 system . in
5.1 we show that alternative damping of the eccentricities by tidal dissipation within the star or planets is insignificant , and in 5.2 we discuss other related studies .
our conclusions are summarized in 6 .
as shown in figure [ figure1 ] , the gj@xmath0876 system has both lowest order , eccentricity - type mean - motion resonance variables at the 2:1 commensurability , @xmath66 and @xmath67 , and hence the secular resonance variable , @xmath68 , librating about @xmath4 .
this resonance configuration is different from that of io and europa , where @xmath7 and @xmath8 librate about @xmath9 .
it should be noted that the differences are not due to the additional resonances involving ganymede in the io - europa case . in the scenario where the resonances among the inner three galilean satellites of jupiter are assembled by differential tidal expansion of the orbits ( yoder 1979 ; yoder & peale 1981 ) , io is driven out most rapidly and the resonance variables involving io and europa only are captured into libration first .
these resonance variables have the same centers of libration before and after the 2:1 commensurability between europa and ganymede is encountered .
the differences in the resonance configurations of gj@xmath0876 and io - europa are instead due to the magnitudes of the eccentricities involved and can be understood from a condition for stable simultaneous librations of the two mean - motion resonance variables .
this condition for stable simultaneous librations of @xmath11 and @xmath7 is that the longitudes of periapse , @xmath69 and @xmath70 , on average precess at the same rate . for coplanar orbits ,
the equations for the variation of @xmath71 , @xmath19 , and @xmath12 in jacobi coordinates are ( e.g. , peale 1986 ) @xmath72 where @xmath73 , @xmath74 , @xmath75 is the mass of the star ( or jupiter ) , @xmath76 is the mass of the @xmath30th planet ( or satellite ) , the hamiltonian @xmath77 and the disturbing potential @xmath78 if we neglect terms of order @xmath79 and higher and assume that @xmath80 , @xmath81 can be expanded to the form @xmath82 where @xmath83 and the summation is over the range @xmath84 , @xmath85 , and @xmath86 . the coefficient @xmath87 , where @xmath88 , can be written in terms of the laplace coefficient @xmath89 and its derivatives ( brouwer & clemence 1961 ) . for a given cosine argument @xmath90 , the term lowest order in eccentricities is of order @xmath91 . within the 2:1 mean - motion resonances and the secular resonance ,
the perturbations are dominated by the terms whose arguments are nearly fixed because of the resonances , i.e. , those terms involving the resonance variables @xmath11 , @xmath7 , @xmath8 , and their linear combinations . since the terms in equation ( [ dadt ] ) for @xmath92 and equation ( [ dedt ] ) for @xmath93 are proportional to @xmath94 and @xmath95 and @xmath96 and @xmath71 appear in the cosine arguments only , the cosines are changed into sines and there is no secular change in @xmath19 and the forced @xmath12 if the resonance variables @xmath11 , @xmath7 , and @xmath8 librate about either @xmath4 or @xmath9 .
( note , however , that if the eccentricities are not small and @xmath92 and @xmath93 are not dominated by a single term , there is also the possibility that the sums of all contributing terms are zero for resonance variables having values other than @xmath4 and @xmath9 . )
we have calculated the contributions to the precession rate @xmath53 from secular terms in equation ( [ phiexp ] ) with argument of the form @xmath97 ( including the @xmath98 non - resonant secular term ) , up to fourth order in eccentricities , and from mean - motion resonance terms with argument of the form @xmath99 ( @xmath100 ) , up to third order in eccentricities , and they are listed in table [ table2 ] .
we set the resonance variables to their average values , i.e. , the libration center values .
in addition , we ignore the small deviation of @xmath101 from @xmath102 ( since the commensurability is not exact ) and evaluate @xmath87 at @xmath103 . for io and europa ,
we adopt @xmath104 and @xmath105 , which are the equilibrium eccentricities before the resonances with ganymede are encountered in the tidal scenario ( yoder & peale 1981 ) .
the numerical values in table [ table2 ] are for the current semi - major axis of io , but they can be scaled to other values of the semi - major axis since the precession rates are proportional to @xmath106 . as we can see in table [ table2 ] , because the orbital eccentricities of io and europa are small , the precession rate @xmath53 is dominated by a single term lowest order in the eccentricities .
this is the term with argument @xmath107 and proportional to @xmath12 in the disturbing potential ( eq . [ [ phiexp ] ] ) , and whose contribution to @xmath53 is proportional to @xmath108 ( eq . [ [ precess ] ] ) .
since the coefficients of the @xmath109 and @xmath110 terms are @xmath111 and @xmath112 , respectively , retrograde precessions of both io and europa require @xmath113 and @xmath114 .
furthermore , the requirement that the regression rates are identical implies a simple relationship between the eccentricities : @xmath115 where @xmath116 is the mean motion and the secular motion @xmath117 includes contributions from secular terms in the disturbing potential and the additional secular motion induced by the oblateness of jupiter . as we can see in table [ table2 ] , the contributions from the secular terms in the disturbing potential and even the secular motion induced by the oblateness of jupiter ( which is @xmath118 for io s orbit ) are small compared with the @xmath119 precession induced by the first order mean - motion resonance terms with arguments @xmath11 and @xmath7 .
thus @xmath120 , where the last equality is for the masses of io and europa . for gj@xmath0876 ,
we know from numerical orbit integration of the laughlin - chambers keck@xmath34lick solution that on average @xmath121 .
the numerical values in table [ table2 ] are obtained using @xmath122 , @xmath123 , and @xmath124 , which are the average values for the laughlin - chambers keck@xmath34lick solution ( see fig . [ figure2 ] ) .
as we would expect from the analysis above for the io - europa case , with @xmath125 , the contributions to @xmath126 and @xmath127 from the @xmath109 and @xmath110 terms , respectively , have opposite signs .
however , because the eccentricities are large , these terms no longer dominate the precession rates .
in particular , the largest contribution to @xmath127 comes from terms second order in eccentricities the largest of which being the @xmath128 term , which is of order @xmath13(@xmath129 ) lower than the @xmath110 term , but whose coefficient is large and negative ( @xmath130 ) .
we can also see from table [ table2 ] that both @xmath126 and @xmath127 converge only slowly with the order of the @xmath131 terms . with the inclusion of terms up to @xmath132
, the precession rates have the correct , negative signs when @xmath133 , but the magnitude of the precession rate for the inner ( outer ) planet is significantly larger ( smaller ) than that for both planets from numerical orbit integration .
based on the magnitudes and the alternating signs of the contributions from the lower order terms , the expected contributions from the @xmath134 ( and higher order ) terms are consistent with their bringing the results into agreement with the actual precession rates .
there is no simple analytic expression relating @xmath13 to @xmath135 when the eccentricities are large , but as in the small eccentricity limit , the eccentricities are related by the requirement that the periapses precess at the same rate .
although the analysis in this section is able to explain why the gj@xmath0876 and io - europa systems have different resonance configurations , the slow convergence of the series for moderate to large eccentricities means that it has limited usefulness if one is interested in the more general question of what stable configurations are possible for different periapse precession rates and masses . a practical way to investigate this latter question is through numerical migration calculations like those in 4.1 , which drive a system through a sequence of configurations with different precession rates . as we shall see , for systems with masses like those in gj@xmath0876 , there are stable configurations with @xmath11 , @xmath7 , and @xmath8 librating about @xmath4 for @xmath136 ( see fig . [ figure3 ] ) .
we have also found that configurations with resonance variables librating about angles other than @xmath4 and @xmath9 are possible when the masses are different from those in gj@xmath0876 ; these configurations will be discussed in a subsequent paper .
we now turn to the question of the origin of the two 2:1 mean - motion resonances and the secular resonance in the gj@xmath0876 system . a scenario for the origin of this resonance configuration
is that it was assembled by differential migration of orbits that were initially much further apart .
we shall see in 4 that libration of all three resonance variables is easily established by any process that drives the orbits toward each other .
although the results presented in 4 are quite general , they will be interpreted in the context of a particular differential migration process , namely via the gravitational interaction between the planets and the nebula from which they form .
it is generally accepted that planets form in a disk of gas and dust that surrounds a young star .
since the two planets around gj@xmath0876 are of order jupiter mass and are presumably gas giants , they must have formed within the lifetime of the gas disk .
if a planet is sufficiently massive , the torques exerted by the planet on the gas disk can open an annular gap in the disk about the planet s orbit .
the conditions for gap formation are that the roche radius of the planet , @xmath137 , exceeds the scale height @xmath138 of the disk , or equivalently , @xmath139 and that the viscous condition @xmath140 is satisfied ( e.g. , lin & papaloizou 1993 ) . in the above equations ,
@xmath141 and @xmath75 are the masses of the planet and the star , @xmath142 is the semi - major axis of the planet s orbit , @xmath143 is the angular kepler speed at @xmath142 , and the kinematic viscosity @xmath144 is expressed using the shakura - sunyaev @xmath145 prescription : @xmath146 . since the planet - star mass ratios of the planets around gj@xmath0876 are @xmath147 and @xmath61 , these planets are expected to open gaps individually during their growth to their final masses if their orbits are sufficiently far apart , unless @xmath148 or @xmath149 . the numerical values @xmath150 and @xmath151 are typical of models of protoplanetary disks ( e.g. , bryden et al . 2000 ; kley 2000 ; papaloizou et al .
hartmann et al .
( 1998 ) have also inferred from observed properties of t tauri disks that @xmath152 .
the most promising source of an effective viscosity in accretion disks is magnetohydrodynamic ( mhd ) turbulence initiated and sustained by the magnetorotational instability , which is capable of producing an effective @xmath145 as large as @xmath153 ( e.g. , stone et al .
2000 ) . however , protoplanetary disks are probably too weakly ionized for mhd turbulence to develop fully , except at small radii ( @xmath154 ) and possibly in a layer near the surface of the disk at larger radii ( gammie 1996 ) . another possible source of effective viscosity is damping of density waves excited by many small ( of order earth mass ) planets , which is capable of producing an effective @xmath155 ( goodman & rafikov 2001 ) .
bryden et al . ( 2000 ) and kley ( 2000 ) have performed hydrodynamic simulations of a system consisting of a central star , a gas disk and two planets .
the planets are massive enough ( @xmath156 ) to open gaps individually , and the ratio of their initial semi - major axes is @xmath157 .
they found that the planets clear out nearly all of the nebular material between them in a few hundred orbital periods .
then the torques exerted by the nebular material outside the outer planet drives that planet toward the star , whereas any nebular material left on the inside of the inner planet drives that planet away from the star .
the depletion of the inner disk means that the inner planet may not move out very far , but the net effect is always to drive the orbits of two massive planets toward each other .
the time scale on which the planets migrate is the disk viscous time scale , whose inverse is ( ward 1997 ) @xmath158 where a dot over a symbol denotes @xmath64 .
in addition to its effect on the semi - major axis , planet - nebula interaction can also affect the orbital eccentricity of a planet ( e.g. , goldreich & tremaine 1980 ; artymowicz 1992 , 1993 ; papaloizou et al .
a planet interacts with a disk in the vicinity of lindblad and corotation resonances .
the leading contribution to @xmath159 is due to lindblad resonances with the @xmath160 fourier components of the planet s perturbation potential .
( the indices @xmath161 and @xmath162 for the fourier series in time and azimuthal angle in this context should not be confused with @xmath163 and @xmath162 in eq .
[ [ phiexp ] ] . ) for a planet orbiting in a disk gap , the so - called co - orbital lindblad and corotation resonances are not important , and the leading contributions to @xmath164 are damping due to @xmath165 corotation resonances and excitation due to @xmath166 outer and @xmath167 inner lindblad resonances .
the net effect of the corotation resonance damping and the lindblad resonance excitation depends on the distribution of the nebular material , which is itself determined by the interaction with the planet .
if the gap is not too wide and many resonances of both types are present , previous calculations indicate that there is net eccentricity damping .
for example , if we consider an outer disk of constant surface mass density @xmath168 that extends radially from @xmath169 to @xmath170 , with @xmath171 , then integration of equations ( 30 ) and ( 31 ) of goldreich & tremaine ( 1980 ) yields @xmath172 ( note that the above equations are identical to eqs . [ 109 ] and [ 110 ] of goldreich & tremaine where they set @xmath173 . ) in the case of gj@xmath0876 , since the planets clear out nearly all of the nebular material between them and the inner disk is likely to be depleted , the dominant resonant interactions are expected to be those between the outer disk and the outer planet .
however , the estimate ( [ gtdedt ] ) for @xmath174 is not adequate because , in addition to the fact that the mass distribution near the inner edge of the outer disk is not modeled properly , the outer planet is sufficiently massive that only a few low-@xmath162 resonances are likely to be present in the disk .
in fact , if the outer planet is able to open a gap out to the 2:1 commensurability , there would be no corotation resonances and only one lindblad resonance with @xmath175 at the 3:1 commensurability , and there would be eccentricity growth instead of damping ( artymowicz 1992 ) .
( in addition , the disk can become eccentric and the growth of the planet s orbital eccentricity can be enhanced by the interaction with the eccentric disk , if the planetary mass is comparable to a characteristic mass of the disk ; papaloizou et al .
2001 . ) from a comparison of the resonant and viscous torques , an approximate condition for opening a gap out to the 2:1 commensurability is ( artymowicz 1992 ; lin & papaloizou 1993 ) @xmath176 since the planet - star mass ratio of the outer planet around gj@xmath0876 is @xmath177 and close to the critical value in equation ( [ 2to1cond ] ) , without knowing the exact values of the disk parameters , it is unclear whether one should expect eccentricity damping or growth
. however , as we shall see in 4 , significant eccentricity _ damping _ with @xmath178 is required to produce the observed eccentricities of the gj@xmath0876 system , unless the migration after resonance capture is severely limited .
therefore , at least the condition ( [ 2to1cond ] ) must not be satisfied , implying that @xmath179 for the outer disk of the young gj@xmath0876 system .
a more detailed analysis using hydrodynamic simulations is necessary to determine whether sufficient eccentricity damping can be produced by such an outer disk .
in this section we present the results of a series of numerical orbit integrations designed to determine the conditions under which the dynamical properties of the current gj@xmath0876 system could be produced by any process ( such as the planet - nebular interaction discussed in 3 ) that drives the orbits of the two planets toward each other .
we consider a system consisting of a central star and two planets , where the stellar and planetary masses are those for the laughlin - chambers keck@xmath34lick solution ( table [ table1 ] ) . unless stated otherwise , the planets are initially on coplanar , circular orbits , with the mean longitudes differing by @xmath9 and the ratio of the semi - major axes @xmath180 , far from the 2:1 mean - motion commensurability .
in addition to the mutual gravitational interactions of the star and the planets , we force the osculating semi - major axis of planet @xmath30 to migrate at a rate @xmath23 . in most cases , we assume that only the outer planet is forced to migrate inward , with a migration rate of the form @xmath181 constant .
the effects of adopting a more general form of the migration rate ( e.g. , @xmath182 being a function of @xmath183 ) or forcing the inner planet to migrate outward are discussed in 4.2 . for the calculations in 4.2
, we also damp the osculating eccentricity at a rate @xmath184 .
we adopt a damping rate of the form @xmath185 , where @xmath25 is a positive constant .
as we shall see , this form of eccentricity damping has the convenient property that the eccentricities reach equilibrium values after capture into the 2:1 resonances .
we note that the relation @xmath186 is valid for the simple estimates ( [ gtdedt ] ) and ( [ gtdadt ] ) for the interaction of a massive planet with an outer disk if @xmath187 constant and that the same relation between the damping and migration rates also holds for a planet that is too small to open a gap in a gas disk with constant @xmath188 and undergoes the so - called type i migration ( artymowicz 1993 ; ward 1997 ) .
the numerical orbit integrations were performed using the symplectic integrator symba ( duncan , levison , & lee 1998 ) modified to include the orbital migration and eccentricity damping terms .
symba is based on a variant of the wisdom - holman ( 1991 ) method and employs a multiple time step technique to handle close encounters .
( the latter feature is not essential for the integrations presented here . )
we confirmed the results by repeating some of the integrations using a bulirsch - stoer integrator that has also been modified to include migration and damping .
we also checked that both modified integrators give the correct exponential decays in @xmath142 and @xmath189 when they are used to integrate the orbit of a single planet with @xmath190 constant and @xmath191 constant .
we describe how the integrators were modified in the appendix .
we consider first inward migration of the outer planet without eccentricity damping . in figure [ figure3 ]
we show the results of a calculation with @xmath192 and @xmath193 initially and @xmath194 .
the migration rate is consistent with that due to planet - nebular interaction at @xmath195 ( eq .
[ [ vistime ] ] ) . figure [ figure3]_a _ shows the time evolution of the semi - major axes and eccentricities , and figure [ figure3]_b _ shows the evolution of the two 2:1 mean - motion resonance variables , @xmath11 and @xmath7 , in plots of @xmath196 versus @xmath197 .
initially , only the outer planet migrates inward at the prescribed rate .
when the 2:1 mean - motion commensurability is encountered , both @xmath11 and @xmath7 ( and hence the secular resonance variable @xmath8 ) are captured into libration about @xmath4 .
we do not see libration of @xmath7 and @xmath8 about @xmath9 , which is expected for small eccentricities ( see 2 ) , because the planetary masses are so large that fairly large eccentricities ( with @xmath13 reaching about @xmath198 ) are generated before the 2:1 commensurability is encountered .
because of the forced migration , the centers of libration are actually slightly offset from @xmath4 .
the resulting resonant interaction slows down the migration of the outer planet and forces the inner planet to migrate inward while keeping the ratio of the semi - major axes nearly constant ( @xmath199 ) ; it also causes the eccentricities to increase rapidly .
the centers of libration remain near @xmath4 and the amplitudes of libration remain small as the eccentricities increase .
the sequence of configurations with increasing eccentricities that the system is driven through after resonance capture is in fact a sequence with increasingly less negative periapse precession rates .
as we discussed in 2 , the forced eccentricities ( and the libration centers ) for a system with stable simultaneous librations of @xmath11 and @xmath7 are determined by the requirement that the longitudes of periapse on average precess at the same rate ( which in turn is determined by @xmath200 ) .
although a longer integration indicates that the system does eventually become unstable when @xmath201 , the existence of stably librating configurations with @xmath13 up to @xmath202 is remarkable . in particular , the configurations with
@xmath203 have _ prograde _ periapse precessions .
we have integrated the configurations at @xmath204 ( with @xmath205 and retrograde periapse precessions ) and @xmath206 ( with @xmath207 and prograde periapse precessions ) forward with migration turned off and confirmed that the system remains stable as seemingly indefinitely repeating configurations .
as we can see in figure [ figure3]_a _ , without eccentricity damping , the eccentricities exceed the observed values for the gj@xmath0876 planets ( dashed lines ) when the semi - major axes of the resonantly locked planets have decreased by only @xmath208 after capture into the resonances .
this result is insensitive to the adopted parameters .
if the migration rate is not too fast ( see below ) , the evolution of a system with different initial @xmath183 ( but for convenience , the same @xmath209 ) or @xmath182 is essentially identical to that shown in figure [ figure3 ] if we plot the semi - major axes in units of initial @xmath183 and time in units of @xmath210 .
therefore , unless by coincidence resonance capture occurs just before migration stops because of , e.g. , nebula dispersal , eccentricity damping is necessary to produce the observed eccentricities of the gj@xmath0876 system . from a set of calculations similar to that shown in figure [ figure3 ] , with initial @xmath180 , @xmath211 , @xmath47 , and @xmath212 , and different @xmath182
, we find that certain capture of both 2:1 mean - motion resonance variables requires @xmath213 where @xmath183 is the semi - major axis of the outer planet when the 2:1 commensurability is encountered . for migration rate within a factor of a few of , but below , the above limit
, both resonance variables are captured into libration , but the centers of libration can be significantly different from @xmath4 and the amplitudes of libration can be large .
the condition ( [ dadtlim ] ) is satisfied by a migration rate due to planet - nebular interaction with the nominal parameter values in equation ( [ vistime ] ) by almost 3 orders of magnitude .
to examine the effects of orbital eccentricities on capture into the 2:1 resonances , we performed a series of calculations with non - zero initial eccentricities .
four types of initial conditions were considered : the outer planet was initially at either periapse or apoapse , and either the initial @xmath13 or @xmath135 was fixed at @xmath214 and the other initial eccentricity was varied .
the inner planet was always started at periapse , with its mean longitude differing from that of the outer planet by @xmath9 , and the remaining parameters were identical to those used in the calculation shown in figure [ figure3 ] .
since gravitational interaction between the planets causes the eccentricities to fluctuate even when @xmath215 and @xmath183 are close to their initial values , the eccentricities quoted below are the maximum eccentricities when @xmath215 and @xmath183 are close to their initial values and not the initial eccentricities .
we find that certain capture of both 2:1 mean - motion resonance variables requires @xmath216 for eccentricities above these limits , there is non - zero probability for the planets to be captured into higher order resonances ( e.g. , 5:2 ) encountered before the 2:1 commensurability .
figure [ figure4 ] shows the results of a calculation similar to that shown in figure [ figure3 ] , but with eccentricity damping of the form @xmath217 , where @xmath218 .
the capture of the resonance variables @xmath11 and @xmath7 into libration about @xmath4 and the initial evolution after resonance capture are similar to the case without damping .
however , the eccentricity growth eventually terminates when the damping balances the excitation due to resonant interaction between the planets .
the eccentricities reach equilibrium values that remain constant for arbitrarily long migration in the resonances . with @xmath218 ,
the equilibrium eccentricities are close to the observed eccentricities of the gj@xmath0876 system . at the end of the calculation shown in figure [ figure4 ] , when @xmath219 , the semi - major axes are also similar to those of the gj@xmath0876 planets .
we have integrated the configuration at the end of the calculation shown in figure [ figure4 ] forward with migration and damping turned off .
the system remains stable , with almost no change in the amplitudes of libration ( compare figs .
[ figure4]_b _ and [ figure5 ] ) , which are somewhat smaller than those of the gj@xmath0876 system ( compare figs .
[ figure1 ] and [ figure5 ] ) .
although the libration amplitudes are fairly similar in the two laughlin - chambers best - fit solutions , we may find in the future , as more data is acquired , that improved best - fit solutions will have smaller libration amplitudes .
alternatively , larger amplitudes could be generated after termination of migration and damping by encounters with remaining planetesimals . in figure [ figure6 ]
we show the time evolutions of the eccentricities for a set of calculations with @xmath220 , @xmath217 , and different @xmath25 .
the equilibrium eccentricities decrease with increasing @xmath25 and are significantly different from the observed eccentricities of the gj@xmath0876system if @xmath25 is more than a factor 23 larger or smaller than @xmath218 . as long as @xmath217 , where @xmath25 is a positive constant ,
the equilibrium eccentricities are determined by the value of @xmath25 only and are insensitive to either the magnitude or functional form of @xmath182 .
this is demonstrated in figure [ figure7 ] , where we plot the results of a calculation with the same @xmath25(@xmath221 ) as that shown in figure [ figure4 ] but with a different form of the migration rate : @xmath222 constant or @xmath223 .
the equilibrium eccentricities are identical in figures [ figure4 ] and [ figure7 ] .
if @xmath224 is not exactly proportional to @xmath18 , the eccentricities would decrease slowly ( after reaching maxima ) or increase slowly as the resonantly locked planets migrate , but the ratio of damping to migration rates , @xmath225 , just before migration and damping stop should be close to @xmath226 for the final eccentricities to be close to the observed values of the gj@xmath0876 system .
thus far we have considered forced inward migration of the outer planet only .
this is the most likely situation in the planet - nebular interaction scenario discussed in 3 , since the inner disk is likely to be depleted and the dominant interactions are expected to be those between the outer disk and the outer planet .
however , if the inner disk is not depleted , at least initially , or migration and damping are due to another process , the inner planet may also be forced to migrate outward . in figure [ figure8 ]
we show the results of a calculation where , for simplicity , the inner planet is forced to migrate outward at the same rate that the outer planet is forced to migrate inward ( @xmath227 ) and @xmath24 .
the inner ( outer ) planet migrates outward ( inward ) at the prescribed rate . after resonance capture , resonant interaction overcomes the forced outward migration of the inner planet , and both planets migrate inward slowly .
the equilibrium eccentricities are close to the observed eccentricities of the gj@xmath0876 system when @xmath228 . therefore , even with eccentricity damping of both planets , significant eccentricity damping with @xmath229 is required to produce the observed orbital eccentricities of the gj@xmath0876planets .
we have found in 4 that significant eccentricity damping with @xmath229 is required to produce the observed eccentricities of the gj@xmath0876 system if the migration has been at all extensive after resonance capture . as we discussed in
3 , it is as yet unclear whether sufficient damping could be produced by planet - nebula interaction , even if the condition ( [ 2to1condii ] ) is satisfied and eccentricity damping is expected ( see also 5.2 ) . in this subsection
we show that alternative eccentricity damping by tidal dissipation within the star and planets during planet migration is completely negligible
. the star gj@xmath0876 would most likely still be in its pre - main - sequence contracting phase during disk evolution and planet migration . to estimate the rate of circularization of an orbit due to tidal dissipation in the star , we use the stellar radius when the time @xmath230 since initial contraction is approximately the migration time scale of @xmath231 ( eq .
[ [ vistime ] ] ) and keep the planets at their current distances from the star .
the planets would have migrated for longer than the migration time scale ( see , e.g. , fig .
[ figure4 ] ) and it would take time for the planets to form .
but by adopting this minimum time to yield the maximum probable stellar radius and by keeping the planets at their current distances , we maximize our estimate of the tidal rate of circularization of an orbit . since the mass of gj@xmath0876 is only @xmath26 , we assume that gj@xmath0876 follows the nearly vertical hayashi track in the hr diagram and remains fully convective during its contracting phase .
hence the star remains a polytrope of index @xmath232 , with effective temperature @xmath233 ( e.g. , dantona & mazzitelli 1994 ) . from the virial theorem , the stellar luminosity @xmath234 where @xmath235 is the stefan - boltzmann constant
, @xmath236 is the gravitational constant , @xmath75 and @xmath237 are the stellar mass and radius , and the form of the gravitational energy for a polytrope of index @xmath238 is given by chandrasekhar ( 1939 ) .
integration of equation ( [ eq : virial ] ) , with @xmath239 for a polytrope of index @xmath232 , gives @xmath240 after @xmath241 , leading to a ratio of the stellar radius to the orbital semi - major axis @xmath242 for the inner planet and @xmath243 for the outer planet .
since the present orbital period of the inner planet of the gj@xmath0876system is about 30 days , which is long compared to the periods of free oscillation of the star , we can use the rate for circularizing an orbit corresponding to dissipation dominated by the equilibrium tide for the star , which is given by ( zahn 1989 ) @xmath244 where @xmath245 is the planet - star mass ratio and @xmath246 .
we neglect the reduction of the turbulent viscosity when the tidal period is short compared to the turnover time of the largest convective eddies and adopt the maximum @xmath247 for a fully convective star .
substitution of @xmath248 from the previous paragraph , along with @xmath249 for the laughlin - chambers keck@xmath34lick solution ( table [ table1 ] ) , @xmath250 , and @xmath247 , into equation ( [ eq : tcirc ] ) gives @xmath251 since we have assumed that the planets are at their closest proximity to the star for the whole time the tidal effects are damping the eccentricities and that the star is inflated to a maximum size , these values are extreme upper bounds on the rate of eccentricity damping by tides raised on the star . even in its inflated state ,
the dissipation in the star can have essentially no effect on the orbital eccentricities . the theoretical circularization rate in equation ( [ eq : tcirc ] )
has been tested by comparing with the observed circularization rates of binaries .
it is in good agreement with the observed rates for binaries containing giant stars ( verbunt & phinney 1995 ) , but even with the maximum @xmath252 , it is about @xmath253@xmath226 times slower than the observed rates for binaries containing main - sequence solar - type stars , which have radiative cores ( claret & cunha 1997 ; goodman & oh 1997 ) .
it is not yet clear why there is a discrepancy in the latter case or that this discrepancy is relevant for a fully convective pre - main - sequence star
. nevertheless , even if we increase the circularization rates in equation ( [ eq : tcirc2 ] ) by a factor @xmath226 , they are still much smaller than the migration rate due to planet - nebula interaction . tidal dissipation within a gaseous planet damps its orbital eccentricity at a rate given by ( e.g. , peale , cassen , & reynolds 1980 ) @xmath254 where @xmath255 , @xmath256 , and @xmath257 are the radius , the potential love number , and the dissipation function of the planet . if we adopt values for @xmath255 , @xmath256 , and @xmath257 similar to those of jupiter , with @xmath258 , @xmath259 ( gavrilov & zharkov 1977 ) , and @xmath260 ( which is approximately the lower bound on @xmath257 for jupiter ; yoder & peale 1981 ) , @xmath261 for the inner planet of the gj@xmath0876 system .
the rate for damping the outer planet s orbital eccentricity from dissipation within itself is of course even smaller .
the planets would most likely be larger and contracting during their migration . however , unless the planetary radii are unrealistically large and comparable to the roche radii , the eccentricity damping rate due to tidal dissipation within the planets is smaller than the migration rate due to planet - nebula interaction by several orders of magnitude . after completing our calculations ( lee & peale 2001 ) , two papers with complementary calculations came to our attention ( snellgrove , papaloizou , & nelson 2001 ; murray , paskowitz , & holman 2001 ) .
snellgrove et al .
( 2001 ) also find from numerical orbit integrations with forced migration and eccentricity damping of the outer planet that the orbital eccentricities of the gj@xmath0876 planets require a short time scale for eccentricity damping compared to the migration time scale .
note , however , that they adopt the minimum planetary masses from the two - kepler fit by marcy et al .
( 2001 ) and , not surprisingly , have difficulties matching both @xmath13 and @xmath135 from their calculations to those from the same fit , since the eccentricities from the two - kepler fit are not consistent with small - amplitude simultaneous librations of @xmath11 and @xmath7 .
snellgrove et al . also develop an analytic resonance theory that is first order in the eccentricities .
as we showed in 2 , a first order theory is inadequate for understanding the current gj@xmath0876 resonance configuration , which has @xmath11 , @xmath7 , and @xmath8 all librating about @xmath4 .
furthermore , as we found in 4 ( see , e.g. , fig .
[ figure3 ] ) , because the eccentricities are excited to large values in the gj@xmath0876 evolution before resonance capture , there is no time during the evolution when a first order theory can be useful for this system .
thus all predictions coming from this theory should be viewed with caution .
snellgrove et al . also present a hydrodynamic simulation of the planet - nebula interaction , where both planets are inside a cavity with almost no disk material . at the end of this simulation , @xmath262 and the semi - major axes of the planets
have decreased by about @xmath263 since resonance capture .
we have performed a numerical orbit integration _ without _ eccentricity damping similar to that shown in figure [ figure3 ] but with the planetary masses adopted by snellgrove et al . and find that @xmath264 for the same reduction in semi - major axes .
thus it is not clear that the eccentricities have reached equilibrium values at the end of this simulation ; even if they have , these equilibrium values are too large for the gj@xmath0876 system .
it appears that the disk model used in this hydrodynamic simulation is not very effective in damping the eccentricities , which emphasizes the uncertainty in such damping .
murray et al .
( 2001 ) consider mainly an alternative scenario in which the migration and eccentricity damping of the outer planet are due to scattering of planetesimals in the disk population .
their numerical simulations confirm our results ( and those of snellgrove et al .
2001 ) of easy capture of an inner planet into resonance , the dual migration of both planets in the resonance , and the growth of the eccentricities , although they limit their numerical and analytical studies to an outer planet of jupiter mass and inner planets of several earth masses and devote much of the discussion to resonances of higher order than those at the 2:1 mean - motion commensurability .
thus their results are not directly applicable to the gj@xmath0876 system .
radial velocity measurements by marcy et al .
( 2001 ) have revealed two planets in resonant orbits about the star gj@xmath0876 .
the remarkable orbital fit obtained by laughlin & chambers ( 2001 ) , which finds both lowest order , eccentricity - type mean - motion resonance variables at the 2:1 commensurability librating with small amplitudes , means that the resonances are almost certainly real and indefinitely stable .
the existence of the eccentricity - type resonances implies that the assumed coplanarity of the orbits is probably close to reality .
the gj@xmath0876 planetary system has revealed properties of the 2:1 orbital resonances that have not been observed nor analyzed before .
the libration of both lowest order mean - motion resonance variables , @xmath66 and @xmath67 , and the secular resonance variable , @xmath68 , about @xmath4 was not anticipated , as the familiar io - europa 2:1 resonance has @xmath11 librating about @xmath4 , but @xmath7 and @xmath8 librating about @xmath9 a configuration that would persist in the absence of ganymede .
thus conjunctions for the jovian satellites occur when io is near periapse and europa is near apoapse , whereas conjunctions of the two planets about gj@xmath0876 occur when both planets are near periapse .
we understood this to be mainly a function of the eccentricities of the orbits , where the resonance configuration with @xmath265 and @xmath266 must obtain when the eccentricities are small , but the resonance configuration with @xmath267 prevails for a system with masses like those in gj@xmath0876 when the eccentricities are large .
a necessary condition for stable simultaneous librations of both mean - motion resonance variables is that @xmath268 on average , so that the relative alignment of the lines of apsides of the two orbits is maintained .
the periapse precessions are dominated by resonant terms in the disturbing potential whose arguments are @xmath11 , @xmath7 , @xmath8 , and their linear combinations .
the dominance of the lowest order terms when the eccentricities are small allows equal precession rates only if the lines of apsides are anti - aligned ( @xmath269 , @xmath266 ) , whereas dominance of higher order terms when the eccentricities are large results in equal precession rates for a system with masses like those in gj@xmath0876 when @xmath267 .
the equality of the precession rates also determines a relationship between the eccentricities of the two orbits , although there is no simple analytic expression for this relationship when the eccentricities are large , since the first order theory is not a good representation .
the stability of the gj@xmath0876 resonance configuration for values of @xmath13 up to 0.86 was also a surprise .
any process that drives the two originally widely separated orbits toward each other can result in capture of the planets into orbital resonances at the 2:1 commensurability .
the naturally occurring situation where nebular disk material is cleared between two planets sufficiently massive to individually open gaps in the disk ( bryden et al .
2000 ; kley 2000 ) leads to inward migration of the outer planet and possibly outward migration of the inner planet .
thus the likely origin of the resonances in the gj@xmath0876 system is this differential planet migration due to torques induced by the planet - nebula interaction .
we have shown that forced inward migration of the outer planet of the gj@xmath0876system results in certain capture of @xmath11 , @xmath7 and hence @xmath8 into libration if initially @xmath15 and @xmath16 and @xmath270 .
the latter rate is three orders of magnitude higher than the likely rate of @xmath271 due to planet - nebular interaction .
the bounds on the eccentricities result not so much from the transition from certain to probabilistic capture at the 2:1 resonances but from likely capture into higher order resonances such as 5:2 before the 2:1 commensurability is encountered . continued migration of the planets while locked in the 2:1 resonances leads to rapid growth in the orbital eccentricities that exceed the observed eccentricities of the gj@xmath0876 system after only a further decrease in the semi - major axes of about 7% if there is no eccentricity damping .
so unless resonance capture occurred near the end of migration , the observed values of the eccentricities require eccentricity damping . with damping of the form @xmath20 , where @xmath25 is a positive constant
, eccentricity growth is terminated at values of the eccentricities that increase with decreasing @xmath25 , and the eccentricities remain constant for indefinite duration of the migration .
the observed eccentricities result for @xmath21 if there is forced migration and eccentricity damping of the outer planet only , but for @xmath22 if there is also forced migration and eccentricity damping of the inner planet .
this result is independent of the magnitude or functional form of @xmath272 as long as @xmath20 is preserved .
relaxing the last condition leads to a slow drift in the eccentricities during migration and would require the migration to terminate as the eccentricities pass through the observed values .
existing analytic estimates of the effects of planet - nebular interaction are consistent with eccentricity damping of the form @xmath20 , if the planet - star mass ratio is not too large ( e.g. , goldreich & tremaine 1980 ; artymowicz 1992 , 1993 ; lin & papaloizou 1993 ; ward 1997 ) . however , the planet - star mass ratio of the outer planet of the gj@xmath0876system is sufficiently close to the critical value separating eccentricity growth from damping for nominal values of the disk parameters that it is uncertain whether such damping would occur . even if the disk parameters are such that eccentricity damping would occur , it is not clear that the magnitude of @xmath25 would be sufficiently large to constrain the eccentricities in the gj@xmath0876 system to the observed values .
we have shown that the alternative eccentricity damping mechanism involving the dissipation of tidal energy within the star and the planets is completely negligible .
further long - term hydrodynamic simulations with different physical assumptions and parameters are required to determine whether planet - nebular interaction could produce sufficient eccentricity damping to allow arbitrary migration of the planets within the resonances in the young gj@xmath0876 system while preserving eccentricities comparable to those observed . if not
, the migration must have been finely tuned to stop when the system had progressed to its observed state , although this latter constraint is too ad hoc to be believable .
it is a pleasure to thank greg laughlin for furnishing the details of the dynamical fit by laughlin & chambers before publication and lars bildsten for pointing out the enhancement of tidal dissipation in a pre - main - sequence star .
we also thank d.n.c .
lin , j.j .
lissauer , and w.r .
ward for informative discussions .
this research was supported in part by nasa grants pgg nag5 - 3646 and oss nag5 - 7177 .
in this appendix we describe how the symplectic integrator symba and the bulirsch - stoer integrator used for the numerical orbit integrations presented in 4 were modified to include the forced orbital migration and eccentricity damping terms .
a second - order symplectic integrator for a hamiltonian of the form @xmath273 , where @xmath274 and @xmath275 are separately integrable , can be represented as @xmath276 the three operators in equation ( [ sia2 ] ) represent a single step of an algorithm that starts with evolving the system under the influence of @xmath274 only for half a time step @xmath277 , then evolving it for a full time step @xmath278 under the influence of @xmath275 , and then evolving it for another half a time step @xmath277 under @xmath274 .
for example , in the wisdom - holman ( 1991 ) method for the gravitational @xmath32-body problem , the hamiltonian in jacobi coordinates is divided into @xmath274 that describes the keplerian motion of the planets around a central star and @xmath275 that describes the perturbation of the planets on one another and is a function of the positions @xmath279 only .
a step of the wisdom - holman method is thus : ( 1 ) each planet evolves along a kepler orbit for time @xmath277 ; ( 2 ) each planet receives a kick to its momentum of the amount @xmath280 while @xmath279 is unchanged ( since @xmath275 does not involve the canonical momenta ) ; ( 3 ) each planet evolves along a kepler orbit for time @xmath277 , starting with the new momentum after the kick .
recursive application of the basic algorithm ( [ sia2 ] ) allows one to construct symplectic integrators for hamiltonians that consist of more than two integrable parts .
for example , a single step of an algorithm for a hamiltonian of the form @xmath281 is @xmath282 .
the symplectic integrator symba ( duncan et al . 1998 ) is based on a variant of the wisdom - holman method , with the gravitational @xmath32-body hamiltonian written in terms of positions relative to a central star and barycentric momenta , and employs a multiple time step technique to handle close encounters . in the symba algorithm ,
the hamiltonian is divided into more than two parts and the recursive application of the algorithm ( [ sia2 ] ) discussed in the previous paragraph is utilized .
although the additional forced orbital migration and eccentricity damping terms are not hamiltonian , they can be incorporated in an analogous manner .
our modified algorithm is @xmath283 where @xmath284 denotes a complete time step for the conservative gravitational @xmath32-body problem using the symba algorithm , and @xmath285 and @xmath286 denote changing the canonical variables according to the imposed @xmath23 and @xmath184 terms , respectively , for time @xmath277 . during the application of the @xmath23 term , all of the other orbital elements are constant , and @xmath287 , where @xmath288 and @xmath289 are @xmath19 at the beginning and end of the step , if @xmath290 constant ( this can be easily generalized for , e.g. , @xmath291 ) .
note that we do not use truncated approximation such as @xmath292 .
similarly , the @xmath286 step changes the eccentricities according to @xmath293 if @xmath294 . by modifying the algorithm in a symmetric manner and using exact solutions in the @xmath285 and @xmath286 parts
, there should be little ( if any ) secular growth in the energy error ( e.g. , mikkola 1998 ) . for the bulirsch - stoer integrator , additional terms must be included in the equations of motion in cartesian coordinates to account for the forced orbital migration and eccentricity damping . in the following
, we simplify the notation by considering a specific planet and dropping the subscript .
let @xmath295 be the cartesian coordinates of the planet with respect to the star and @xmath296 be the distance of the planet from the star .
the osculating orbital elements @xmath142 , @xmath189 , @xmath30 , @xmath297 , @xmath298 , and @xmath143 are the semi - major axis , eccentricity , inclination , true anomaly , argument of periapse , and longitude of the ascending node on the @xmath299 plane , respectively .
the additional terms in the equations of motion due to the forced migration @xmath159 and eccentricity damping @xmath164 terms are @xmath300 with similar expressions for the other coordinates . to evaluate the partial derivatives in equations ( [ eq : dxdt ] ) and ( [ eq : ddotxdt ] ) and similar expressions for the other coordinates , we need to express the position and velocity in terms of the osculating orbital elements : @xmath301 and @xmath302 \nonumber\\ & & \quad -\cos{i}\sin{\omega}[{\dot r}\sin{(\omega+f ) } + r{\dot f}\cos{(\omega+f ) } ] , \nonumber\\ { \dot y } & = & \sin{\omega}[{\dot r}\cos{(\omega+f ) } -r{\dot f}\sin{(\omega+f ) } ] \label{eq : dotx}\\ & & \quad + \cos{i}\cos{\omega}[{\dot r}\sin{(\omega+f ) } + r{\dot f}\cos{(\omega+f ) } ] , \nonumber\\ { \dot z } & = & \sin{i}[{\dot r}\sin{(\omega+f ) } + r{\dot f}\cos{(\omega+f ) } ] , \nonumber\end{aligned}\ ] ] where @xmath296 , @xmath303 , and @xmath304 are in terms of @xmath142 , @xmath189 , and @xmath297 ( e.g. , murray & dermott 1999 ) .
we also need @xmath305,\nonumber\\ & & \nonumber \\
\frac{\partial\dot r}{\partial a } & = & -\frac{\dot r}{2a},\nonumber\\ & & \nonumber \\
\frac{\partial\dot r}{\partial e } & = & \frac{\dot r}{e(1-e^2 ) } , \label{eq : drdtae}\\ & & \nonumber \\ \frac{\partial ( r\dot f)}{\partial a } & = & -\frac{r\dot f}{2a},\nonumber\\ & & \nonumber \\ \frac{\partial ( r\dot f)}{\partial e } & = & \frac{r\dot f(e+\cos{f } ) } { ( 1-e^2)(1+e\cos{f})}.\nonumber\end{aligned}\ ] ] from equations ( [ eq : dxdt ] ) , ( [ eq : x ] ) , and ( [ eq : drdtae ] ) , we find that @xmath306\frac{x}{e}{\dot e } ; \label{eq : dxdtf}\ ] ] the additional terms for @xmath307 and @xmath308 are similar .
the additional terms for each of @xmath309 , @xmath310 , @xmath311 are distinct for variations in @xmath189 , and we have @xmath312{\dot e}\nonumber\\ & & \nonumber\\ & & \quad -\cos{i}\sin{\omega}\left[\frac{\partial\dot r}{\partial e } \sin{(\omega+f)}+\frac{\partial(r\dot f)}{\partial e}\cos{(\omega+f ) } \right]{\dot e},\nonumber\\ & & \nonumber\\ \frac{d\dot y}{dt}\bigg|_{\dot a } + \frac{d\dot y}{dt}\bigg|_{\dot e } & = & -\frac{\dot y}{2a } { \dot a}+\sin{\omega}\left [ \frac{\partial \dot r}{\partial e}\cos{(\omega+f)}-\frac{\partial ( r\dot f)}{\partial e}\sin{(\omega+f)}\right]{\dot e}\label{eq : ddotxdtf}\\ & & \nonumber\\ & & \quad + \cos{i}\cos{\omega}\left[\frac{\partial \dot r}{\partial e } \sin{(\omega+f)}+\frac{\partial ( r\dot f)}{\partial e}\cos{(\omega+f ) } \right]{\dot e},\nonumber\\ & & \nonumber\\ \frac{d\dot z}{dt}\bigg|_{\dot a } + \frac{d\dot z}{dt}\bigg|_{\dot e } & = & -\frac{\dot z}{2a } { \dot a}+\sin{i}\left [ \frac{\partial \dot r}{\partial e}\sin{(\omega+f)}+\frac{\partial ( r\dot f ) } { \partial e}\cos{(\omega+f)}\right]{\dot e},\nonumber \end{aligned}\ ] ] where equation ( [ eq : drdtae ] ) should be used to get the functional forms .
unfortunately , the orbital elements must be calculated at each call to the differential equations when there is eccentricity damping .
artymowicz , p. 1992
, , 104 , 769 artymowicz , p. 1993
, , 419 , 166 brouwer , d. , & clemence , g. m. 1961 , methods of celestial mechanics ( new york : academic ) bryden , g. , ryczka , m. , lin , d. n. c. , & bodenheimer , p. 2000 , , 540 , 1091 chandrasekhar , s. 1939 , an introduction to the study of stellar structure ( chicago : univ
of chicago press ) claret , a. , & cunha , n. c. s. 1997 , , 318 , 187 dantona , f. , & mazzitelli , i. 1994 , , 90 , 467 duncan , m. j. , levison , h. f. , & lee , m. h. 1998 , , 116 , 2067 gammie , c. f. 1996 , , 457 , 355 gavrilov , s. v. , & zharkov , v. n. 1977 , icarus , 32 , 443 goldreich , p. , & tremaine , s. 1980 , , 241 , 425 goodman , j. , & oh , s. p. 1997 , , 486 ,
403 goodman , j. , & rafikov , r. r. 2001 , , 552 , 793 hartmann , l. , calvet , n. , gullbring , e. , & dalessio , p. 1998
, , 495 , 385 kley , w. 2000 , , 313 , l47 laughlin , g. , & chambers , j. e. 2001 , , 551 , l109 lee , m. h. , & peale , s. j. 2001 , , 33 , in press lin , d. n. c. , & papaloizou , j. c. b. 1993 , in protostars and planets iii , ed . e. h. levy & j. i. lunine ( tucson : univ .
arizona press ) , 749 marcy , g. w. , butler , r. p. , fischer , d. , vogt , s. s. , lissauer , j. j. , & rivera , e. j. 2001 , , 556 , 296 mikkola , s. 1998 , celest .
astron . , 68 , 249 murray , c. d. , & dermott , s. f. 1999 , solar system dynamics ( cambridge : cambridge univ . press ) murray , n. , paskowitz , m. , & holman , m. 2001 , preprint ( astro - ph/0104475 ) papaloizou , j. c. b. , nelson , r. p. , & masset , f. 2001 , , 366 , 263 peale , s. j. 1986 , in satellites , ed .
j. a. burns & m. s. matthews ( tucson , univ .
arizona press ) , 159 peale , s. j. 1999 , , 37 , 533 peale , s. j. , cassen , p. , & reynolds , r. t. 1980 , icarus , 43 , 65 rivera , e. j. , & lissauer , j. j. 2001 , , 558 , 392 snellgrove , m. d. , papaloizou , j. c. b. , & nelson , r. p. 2001
, , 374 , 1092 stone , j. m. , gammie , c. f. , balbus , s. a. , & hawley , j. f. 2000 , in protostars and planets iv , ed . v. mannings , a. p. boss & s. s. russell ( tucson : univ .
arizona press ) , 589 verbunt , f. , & phinney , e. s. 1995 , , 296 , 709 ward , w. r. 1997 , icarus , 126 , 261 wisdom , j. , & holman , m. 1991 , , 102 , 1528 yoder , c. f. 1979 , , 279 , 747 yoder , c. f. , & peale , s. j. 1981 , icarus , 47 , 1 zahn , j .-
1989 , , 220 , 112 ccccccc @xmath141 ( @xmath28)&@xmath313&@xmath314&@xmath315&@xmath316&@xmath317&@xmath318 + @xmath319 ( day)&30.12&61.02&29.995&62.092&30.569&60.128 + @xmath142 ( au)&0.130&0.208&0.1294&0.2108&0.1309&0.2061 + @xmath189&0.27&0.10&0.314&0.051&0.244&0.039 + @xmath320 ( deg)&330&333&51.8&40.0&159.1&163.3 + @xmath321 ( jd ) & & & + @xmath322 ( deg)&0.0&@xmath323&289&340&356&173 + lccccccc @xmath324 & @xmath325 & & 0.0034 & 0.0092 & & 0.0374 & @xmath326 + & @xmath134 & & 0.0000 & 0.0000 & & 0.0032 & @xmath327 + @xmath328 & @xmath329 & & @xmath330 & @xmath331 & & @xmath332 & 0.1677 + & @xmath325 & & 0.0190 & 0.0820 & & 0.1456 & @xmath333 + & @xmath132 & & @xmath334 & @xmath335 & & @xmath336 & 0.2471 + + total & & & @xmath337 & @xmath338 & & @xmath339 & @xmath340 + actual & & & & & & + | the discovery by marcy et al .
( 2001 ) of two planets in 2:1 orbital resonance about the star gj@xmath0876 has been supplemented by a dynamical fit to the data by laughlin & chambers ( 2001 ) which places the planets in coplanar orbits deep in three resonances at the 2:1 mean - motion commensurability .
the selection of this almost singular state by the dynamical fit means that the resonances are almost certainly real , and with the small amplitudes of libration of the resonance variables , indefinitely stable .
several unusual properties of the 2:1 resonances are revealed by the gj@xmath0876system .
the libration of both lowest order mean - motion resonance variables and the secular resonance variable , @xmath1 , @xmath2 , and @xmath3 , about @xmath4 ( where @xmath5 are the mean longitudes of the inner and outer planet and @xmath6 are the longitudes of periapse ) differs from the familiar geometry of the io - europa pair , where @xmath7 and @xmath8 librate about @xmath9 . by considering the condition that @xmath10 for stable simultaneous librations of @xmath11 and @xmath7 , we show that the gj@xmath0876 geometry results because of the large orbital eccentricities @xmath12 , whereas the very small eccentricities in the io - europa system lead to the latter s geometry .
surprisingly , the gj@xmath0876 configuration , with @xmath11 , @xmath7 , and @xmath8 all librating , remains stable for @xmath13 up to 0.86 and for amplitude of libration of @xmath11 approaching @xmath14 with the current eccentricities further supporting the indefinite stability of the existing system .
any process that drives originally widely separated orbits toward each other could result in capture into the observed resonances at the 2:1 commensurability .
we find that forced inward migration of the outer planet of the gj@xmath0876system results in certain capture into the observed resonances if initially @xmath15 and @xmath16 and the migration rate @xmath17 .
larger eccentricities lead to likely capture into higher order resonances before the 2:1 commensurability is reached .
the planets are sufficiently massive to open gaps in the nebular disk surrounding the young gj@xmath0876 and to clear the disk material between them , and the resulting planet - nebular interaction typically forces the outer planet to migrate inward on the disk viscous time scale , whose inverse is about three orders of magnitude less than the above upper bound on @xmath18 for certain capture . if there is no eccentricity damping , eccentricity growth is rapid with continued migration within the resonance , with @xmath12 exceeding the observed values after a further reduction in the semi - major axes @xmath19 of only 7% . with eccentricity damping @xmath20 , the eccentricities reach equilibrium values that remain constant for arbitrarily long migration within the resonances .
the equilibrium eccentricities are close to the observed eccentricities for @xmath21 if there is migration and damping of the outer planet only , but for @xmath22 if there is also migration and damping of the inner planet .
this result is independent of the magnitude or functional form of the migration rate @xmath23 as long as @xmath24 .
although existing analytic estimates of the effects of planet - nebula interaction are consistent with this form of eccentricity damping for certain disk parameter values , it is as yet unclear that such interaction can produce the large value of @xmath25 required to obtain the observed eccentricities .
the alternative eccentricity damping by tidal dissipation within the star or the planets is completely negligible , so the observed dynamical properties of the gj@xmath0876 system may require an unlikely fine tuning of the time of resonance capture to be near the end of the nebula lifetime . |
spreading of opinions , frauds , behavioural patterns , and product adoptions are all examples of social contagion phenomena where collective patterns emerge due to correlated decisions of a large number of individuals . although these choices are personal , they are not independent but potentially driven by several processes such as social influence @xcite , homophily @xcite , and information arriving from external sources like news or mass media @xcite .
social contagion evolves over networks of interconnected individuals , where links associated with social ties transfer influence between peers @xcite .
several earlier studies aimed to identify the dominant mechanisms at play in social contagion processes @xcite .
one key element , termed behavioural threshold by granovetter @xcite , is defined as _
`` the number or proportion of others who must make one decision before a given actor does so''_. following this idea various network models have been introduced @xcite to understand the threshold - driven spreading , commonly known as _ complex contagion _ @xcite .
although these models are related to a larger set of collective dynamics , they are particularly different from _ simple contagion _ where the exposure of nodes is driven by independent contagion stimuli @xcite .
in addition , collective adoption patterns may appear as a consequence of homophilic structural correlations , where connected individuals adopt due to their similar interests and not due to direct social influence .
distinguishing between the effects of social influence and homophily at the individual level remains as a challenge @xcite .
furthermore , in real social spreading phenomena all these mechanisms are arguably present . however , while in the case of homophily the adoption behaviour is only seemingly correlated , and for simple contagion only the number of exposures matters , in complex contagion the fraction of adopting neighbours relative to the total number of partners determines whether a node adopts or not , capturing the natural mechanisms involved in individuals decision makings @xcite . due to this additional complexity ,
threshold models are able to emulate system - wide adoption patterns known as global cascades .
behavioural cascades are rare but potentially stupendous social spreading phenomena , where collective patterns of exposure emerge as a consequence of small initial perturbations .
some examples are the rapid emergence of political and grass - root movements @xcite , fast spreading of information @xcite or behavioural patterns @xcite , etc . the characterisation @xcite and modelling
@xcite of such processes have received plenty of attention and provide some basic understanding of the conditions and structure of empirical and synthetic cascades . however
, these studies commonly fail in addressing the temporal dynamics of the emerging cascades , which may vary considerably between different cases of social contagion .
moreover , they have not answered why real - world cascades can evolve through various dynamic pathways ranging from slow to rapid patterning , especially in systems where the threshold mechanisms play a role and social phenomena spread globally . besides the case of rapid
cascading mentioned above , an example of the other extreme is the propagation of products in social networks @xcite , where adoption evolves gradually even if it is driven by threshold mechanisms and may cover a large fraction of the total population @xcite .
this behaviour characterises the adoption of online services such as facebook , twitter , linkedin and skype ( fig.[fig:1]a ) , since their yearly maximum relative growth of cumulative adoption @xcite ( for definition see material and methods ( mm ) ) is lower than in the case of rapid cascades as suggested e.g. by the watts threshold ( wt ) model . to fill this gap in the modelling of social diffusion , here we will analyse and model real - world examples of social contagion phenomena .
our aim is to identify the crucial mechanisms necessary to consider in models of complex contagion to match them better with reality , and define a model that incorporates these mechanisms and captures the possible dynamics leading to the emergence of real - world global cascades .
we follow the adoption dynamics of the skype paid service `` buy credit '' for @xmath0 months since 2004 , which evolves over the social network of one of the largest voice over internet providers in the world .
data includes the time of first payment of each user , an individual and conscious action that tracks adoption behaviour .
in addition we follow the subscription " service over @xmath1 months since 2008 ( for results see supplementary information [ si ] ) .
in contrast to other empirical studies where incomplete knowledge about the underlying social network leads to unavoidable bias @xcite , we use here the largest connected component of the aggregated free skype service as the underlying structure , where nodes are skype users and links confirmed contacts between them . this is a good approximation since it maps all connections in the skype social network without sampling , and the paid service is only available for individuals already enrolled in the skype network .
the underlying structure is an aggregate from september 2003 to november 2011 ( i.e. over @xmath2 months ) and contains roughly 4.4 billion links and 510 million registered users worldwide @xcite .
the data is fully anonymised and considers only confirmed connections between users ( for more data details see si ) . in what follows
we first provide empirical evidence of the distribution of individual adoption thresholds and other structural and dynamical features of a worldwide adoption cluster .
we incorporate the observed structural and threshold heterogeneities into a dynamical threshold model where multiple nodes adopting spontaneously ( i.e. firstly among their neighbours ) are allowed @xcite .
we find that if the fraction of users who reject to adopt the product is large , the system enters a quenched state where the evolution and structure of the global adoption cluster is very similar to our empirical observations .
model calculations and the analysis of the real social contagion process suggest that the evolving structure of an adoption cluster differs radically from what has been proposed earlier @xcite , since it is triggered by several spontaneous adoptions arriving at a constant rate , while stable adopters who are initially resisting exposure , are actually responsible for the emergence of global social adoption ( fig . [ fig:1]b and c ) .
social contagion phenomena can be modelled as binary - state processes evolving on networks and driven by threshold mechanisms . in these systems individuals
are represented by nodes , each being either in a susceptible ( 0 ) or adopter ( 1 ) state and influencing each other by transferring information via social ties @xcite .
nodes are connected in a network with degree distribution @xmath3 and average degree @xmath4 .
in addition , each node has an individual threshold @xmath5 $ ] drawn from a distribution @xmath6 with average @xmath7 .
this threshold determines the minimum fraction of exposed neighbours that triggers adoption and captures the resistance of an individual against engaging in spreading behaviour .
once a node reaches its threshold , it switches state from @xmath8 to @xmath9 and keeps it until the end of the dynamics . in his seminal paper about threshold dynamics , watts @xcite classified nodes into three categories based on their threshold and degree .
he identified _
nodes that spontaneously change state to @xmath9 , thus starting the process .
such nodes have a trivial threshold @xmath10
. then there are nodes with threshold @xmath11 , called _
vulnerable _ , which need one adopting neighbour before their own adoption . finally , there are more resilient nodes with threshold @xmath12 , denoted as _ stable _ , referring to individuals in need of strong social influence to follow the actions of their acquaintances . in the wt model @xcite , small perturbations ( like the spontaneous adoption of a single seed node )
can trigger global cascading patterns .
however , their emergence is subject to the so - called _ cascade condition _ : the innovator seed has to be linked to a percolating vulnerable cluster , which adopts immediately afterwards and further triggers a global cascade ( i.e. a set of adopters larger than a fixed fraction of the finite network ) .
the cascade condition is satisfied if the network is inside a bounded regime in @xmath13-space @xcite .
this regime depends on degree and threshold heterogeneities @xcite and may change its shape if several innovators start the process @xcite .
degree and threshold heterogeneities are indeed present in the social network of skype .
the degree distribution @xmath3 is well approximated by a lognormal function @xmath14 ( @xmath15 ) with parameters @xmath16 , @xmath17 and @xmath18 ( fig . [ fig:1]d ) , giving an average degree @xmath19 ( for goodness of fit see si ) .
moreover , at the time of adoption we can measure the threshold @xmath20 of a user by counting the number @xmath21 of its neighbours who have adopted the service earlier .
we then group users by degree and calculate the distribution @xmath22 of the integer threshold @xmath21 @xcite ( fig .
[ fig:1]e ) . by using the scaling relation @xmath23 all distributions collapse to a master curve well approximated by a lognormal function @xmath24 , with parameters @xmath25 and @xmath26 as constrained by the average threshold @xmath27 ( see mm and si ) .
note that we observe qualitatively the same scaling and lognormal shape of the threshold distribution for another service ( see si ) .
these empirical observations , in addition to the broad degree distribution , provide quantitative evidence about the heterogeneous nature of adoption thresholds .
since we know the complete structure of the online social network , as well as the first time of service usage for all adopters , we can follow the temporal evolution of the adoption dynamics . by counting the number of adopting neighbours of an ego
, we identify innovators ( @xmath28 ) , and vulnerable ( @xmath29 ) or stable ( @xmath30 ) nodes .
the adoption rates for these categories behave rather differently from previous suggestions @xcite ( fig . [
fig:1]f ) .
first , there is not only one seed but an increasing fraction of innovators in the system who , after an initial period , adopt approximately at a constant rate .
second , vulnerable nodes adopt approximately with the same rate as innovators suggesting a strong correlation between these types of adoption .
third , the overall adoption process accelerates due to the increasing rate of stable adoptions induced by social influence . at the same time a giant adoption cluster grows and percolates through the whole network ( fig .
[ fig:3]a , main panel ) . despite of this expansion dynamics and connected structure of the service adoption cluster ,
the service reaches less than @xmath31 of the total number of active skype users over a period of @xmath32 years @xcite .
therefore we ask whether one can refer to these adoption clusters as cascades .
they are not triggered by a small perturbation but induced by several innovators ; their evolution is not instantaneous but ranges through several years ; and although they involve millions of individuals , they reach only a reduced fraction of the whole network .
to answer we incorporate the above mentioned features into a dynamical threshold model @xcite with a growing group of innovators and investigate their effect on the evolution of global social adoption .
note that we also perform a null model study to demonstrate , on the system level , that social influence dominates the contagion process , but not homophily ( see section s3 of the si , together with another empirical spreading scenario in s7.1 ) .
our modelling framework is an extension to conventional threshold dynamics on networks studied by watts , gleeson , singh , and others , where all nodes are initially susceptible and innovators are only introduced as an initial seed of arbitrary size @xcite . apart from the threshold rule discussed above , our model considers two additional features : ( i ) a fraction @xmath33 of ` immune ' nodes that never adopt , indicating lack of interest in the service ; ( ii ) due to external influence , susceptible nodes adopt the innovation spontaneously ( i.e. become innovators ) throughout time with constant rate @xmath34 , rather than only at the beginning of the dynamics . in this way
, the dynamical evolution of the system is completely defined by the online social network , the distribution @xmath6 and the parameters @xmath33 , @xmath34 . for the sake of simplicity
we consider a configuration - model network , i.e. , we ignore correlations in the social network and characterise it solely by its degree distribution @xmath3 .
furthermore , node degrees and thresholds are considered to be independent @xcite .
our threshold model , which has also been introduced in @xcite , can be studied analytically by extending the framework of approximate master equations ( ames ) for monotone binary - state dynamics recently developed by gleeson @xcite , where the transition rate between susceptible and adoption states only depends on the number @xmath35 of network neighbours that have already adopted .
we describe a node by the property vector @xmath36 , where @xmath37 is its degree and @xmath38 its type , i.e. @xmath39 is the type of the fraction @xmath33 of immune nodes , while @xmath40 is the type of all non - immune nodes that have threshold @xmath41 . in this way @xmath6 is substituted by the discrete distribution of types @xmath42 ( for @xmath43 ) .
the integer @xmath44 is the maximum number of degrees ( or non - zero types ) considered in the ame framework , which can be increased to improve the accuracy of the analytical approximation at the expense of speed in its numerical computation ( see s4.2 ) . under these conditions ,
the ame system describing the dynamics of the threshold model is reduced to the pair of ordinary differential equations ( see si ) , [ eq : reducedames ] @xmath45 where @xmath46 is the fraction of adopters in the network , @xmath47 is the probability that a randomly chosen neighbour of a susceptible node is an adopter , and the initial conditions are @xmath48 . here , @xmath49,\ ] ] and , @xmath50,\ ] ] where @xmath51 , @xmath52 , and @xmath53 is the binomial distribution
. the fraction of adopters @xmath54 is then obtained by solving eq .
( [ eq : reducedames ] ) numerically .
since susceptible nodes adopt spontaneously with rate @xmath34 , the fraction of innovators @xmath55 in the network is given by ( see s4.3 ) , @xmath56 in @xmath13-space , for @xmath57 and @xmath58 .
contour lines signal parameter values for which @xmath59 of non - immune nodes have adopted , for fixed @xmath33 and varying time ( a ) , and for fixed time and varying @xmath33 ( b ) .
the continuous contour line and dot indicate parameter values in the last observation of skype s3 .
a regime of maximal adoption ( @xmath60 ) grows as time goes by , and shrinks for larger @xmath33 .
* ( c ) * time series of the fraction of adopters @xmath54 for fixed @xmath61 and varying @xmath33 ( main ) , and for fixed @xmath62 and varying @xmath34 ( inset ) .
these curves are well approximated by the solution of eq .
( [ eq : reducedames ] ) for @xmath63 , @xmath64 and @xmath65 ( dashed lines ) .
the dynamics is clearly faster for larger @xmath34 .
as @xmath33 increases , the system enters a regime where the dynamics is slowed down and adopters are mostly innovators . *
( d ) * final fraction of innovators @xmath66 and time @xmath67 when @xmath68 of non - immune nodes have adopted as a function of @xmath33 , both simulated and theoretical .
the crossover to a regime of slow adoption is characterised by a maximal fraction of innovators and time @xmath67 . unless otherwise stated , @xmath69 and we use @xmath70 , @xmath71 , @xmath17 , @xmath18 , @xmath25 , and @xmath26 to obtain @xmath19 and @xmath27 as in skype s3 .
the difference in @xmath72 between data and model is due to finite - size effects ( see materials and methods ) .
numerical results are averages over @xmath73 ( a - b ) and @xmath74 ( c - d ) realisations .
[ fig:2],scaledwidth=60.0% ] we also implement the threshold model numerically via a monte carlo simulation in a network of size @xmath75 , with a lognormal degree distribution and a lognormal threshold distribution as observed empirically .
thus , we can explore the behaviour of @xmath54 and @xmath76 as a function of @xmath77 , @xmath78 , @xmath34 and @xmath33 , both in the numerical simulation and in the theoretical approximation given by eqs .
( [ eq : reducedames ] ) and ( [ eq : innovfrac ] ) . for @xmath79
some nodes adopt spontaneously as time passes by , leading to a frozen state characterised by a final fraction @xmath80 of adopters . however , the time needed to reach such state depends heavily on the distribution of degrees and thresholds , as signalled by a region of large adoption ( @xmath60 ) that grows in @xmath13-space with time ( contour lines in fig .
[ fig:2]a ) .
if we fix a time in the dynamics and vary the fraction of immune nodes instead , this region shrinks as @xmath33 increases ( contour lines in fig .
[ fig:2]b ) . in other words ,
the set of networks ( defined by their average degree and threshold ) that allow the spread of adoption is larger at later times in the dynamics , or when the fraction of immune nodes is small .
when both @xmath81 and @xmath33 are fixed , the normalised fraction of adopters @xmath82 gradually decreases for less connected networks with larger thresholds ( surface plot in fig .
[ fig:2]a and b ) .
for @xmath83 the critical fraction of innovators necessary to trigger a cascade of fast adoption throughout all susceptible nodes may be identified as the inflection point in the time series of @xmath54 ( fig .
[ fig:2]c , inset ) .
the adoption cascade appears sooner for larger @xmath34 , since this parameter regulates how quickly the critical fraction of innovators is reached .
yet as we increase @xmath33 above a threshold @xmath84 , the system enters a regime where rapid cascades disappear and adoption is slowed down .
the crossover between these regimes is gradual , as seen in the shape of @xmath54 for increasing @xmath33 ( fig .
[ fig:2]c , main panel ) .
we may identify @xmath84 in various ways : by the maximum in both the final fraction of innovators @xmath66 and the critical time @xmath67 when @xmath85 ( fig .
[ fig:2]d ) , or as the @xmath33 value where the inflection point in @xmath54 disappears .
these measures indicate @xmath86 for the chosen parameters .
all global properties of the dynamics ( like the functional dependence of @xmath54 and @xmath76 ) are very well approximated by the solution of eqs .
( [ eq : reducedames ] ) and ( [ eq : innovfrac ] ) ( dashed lines in fig . [ fig:2]c and d ) . indeed , the ame framework is able to capture the shape of the @xmath54 time series , the crossover between regimes of fast and slow adoption , as well as the maximum in @xmath66 and @xmath67 . to better understand how innovation spreads throughout real social networks ,
we take a closer look at the internal structure of the service adoption process . by taking into account individual adoption times
we construct an evolving adoption network with links between users who have adopted the service before time @xmath81 and are connected in the social structure . in order to avoid the effect of instantaneous group adoptions ( evidently not driven by social influence ) ,
we only consider links between nodes who are neighbours in the underlying social network and whose adoption did not happen at the same time .
this way links in the adoption graph indicate ties where social influence among individuals could have existed .
the size distribution @xmath87 of connected components in the adoption network shows the emergence of a giant percolating component over time ( fig .
[ fig:3]a ) , along with several other small clusters .
moreover , after decomposition we observe that the giant cluster does not consist of a single innovator seed and percolating vulnerable tree @xcite , but builds up from several innovator seeds that induce small vulnerable trees locally ( fig . [ fig:3]b ) , each with small depth ( fig .
[ fig:3]d ) @xcite . at the same time
the stable adoption network ( considering connections between all stable adopters at the time ) has a giant connected component , indicating the emergence of a percolating stable cluster with size comparable to the largest adoption cluster ( fig .
[ fig:3]a , inset ) .
these observations suggest a scenario for the evolution of the global adoption component different from earlier threshold models @xcite .
it appears that here multiple innovators adopt at different times and trigger local vulnerable trees ( fig.[fig:1]b ) , which in turn induce a percolating component of connected stable nodes that holds the global adoption cluster together ( fig.[fig:1]c ) .
consequently , in the structure of the adoption network primary triggering effects are important only locally , while external and secondary triggering mechanisms seem to be responsible for the emergence of global - scale adoption .
, main panel ] and stable adoption [ @xmath88 , inset ] networks , with @xmath89 and @xmath90 relative to system size . *
( b ) * empirical connected - component size distribution @xmath91 for the relative size of innovator - induced vulnerable trees at different times . *
( c ) * average size of the largest ( @xmath92 ) and 2nd largest ( @xmath93 ) components of the model network ( ` net ' ) , adoption network ( ` casc ' ) , stable network ( ` stab ' ) , and induced vulnerable trees ( ` vuln ' ) as a function of @xmath33 .
dashed lines show the observed relative size of the real @xmath92 of the adopter network in @xmath94 [ see main panel in ( a ) ] and the predicted @xmath33 value . *
( d ) * distribution @xmath95 of depths of induced vulnerable trees in both data and model for several @xmath33 values , showing a good fit with the data for @xmath96 .
the difference in the tail is due to finite - size effects . *
( e ) * correlation @xmath97 between innovator degree and average size of vulnerable trees in both data and model with the same @xmath33 values as in ( e ) .
model calculations for ( d ) and ( e ) correspond to networks of size @xmath98 and are averaged over @xmath73 realisations .
[ fig:3],scaledwidth=60.0% ] to model the observed dynamics and explore the effect of immune nodes , we perform extensive numerical simulations of the threshold model with parameters determined directly from the data ( see mm and si ) .
we use a network structure with empirical degree and threshold distributions and fix @xmath69 as the constant rate of innovators , implying that the time scale of a monte carlo iteration in the model is 1 month .
we measure the average size of the largest ( @xmath92 ) and second largest ( @xmath93 ) connected components of the background social network , and of the stable , vulnerable and global adoption networks , as a function of the fraction of immune nodes @xmath33 .
after @xmath99 iterations ( matching the length of the real observation period ) we identify three regimes of the dynamics ( fig .
[ fig:3]c ) : if @xmath100 ( dark - shaded area ) the spreading process is very rapid and evolves in a global cascade , which reaches most of the nodes of the shrinking susceptible network in a few iteration steps .
about @xmath101 of adopters are connected in a percolating stable cluster , while vulnerable components remain very small in accordance with empirical observations . in the crossover regime @xmath102 ( light - shaded area ) , the adoption process slows down considerably ( fig .
[ fig:2]d , lower panel ) , as stable adoptions become less likely due to the quenching effect of immune nodes .
the adoption process becomes the slowest at @xmath103 ( fig .
[ fig:2]d , lower panel ) when the percolating stable cluster falls apart , as demonstrated by a peak in the corresponding @xmath93 curve ( fig .
[ fig:3]c , lower panel ) .
finally , around @xmath104 the adoption network becomes fragmented and no global diffusion takes place .
we repeat the same calculations for another service and find qualitatively the same picture , but with the crossover regime shifted towards larger @xmath33 values due to the different parametrisation of the model process .
note that another possible reason for the slow adoption could be the time users wait between their threshold has been reached and actual adoption .
we test for the effect of this potential scenario on the empirical curves but find no qualitative change in the dynamics ( see si ) .
we can use these calculations to estimate the only unknown parameter ( the fraction @xmath33 of immune nodes in skype ) by matching the size of the largest component ( @xmath105 ) between real and model adoption networks at time @xmath106 .
empirically , this value is the relative size corresponding to the last point on the right - hand side of the distribution for @xmath94 ( fig . [ fig:3]a , main panel ) .
the corresponding value in the model is @xmath57 ( dashed lines in fig .
[ fig:3]c ; also fig .
[ fig:2]a and b ) , suggesting that the real adoption process lies in the crossover regime .
the other analysed service turns out to lie right of the crossover regime , which explains its large innovator adoption rate and reduced size of stable and vulnerable adoption clusters ( see si ) . to test the validity of the prediction of @xmath33 we perform three different calculations .
first we measure the maximum relative growth rate of cumulative adoption and find a good match between model and data ( skype s3 and model skype s3 in fig .
[ fig:1]a ) . in other words , the model correctly estimates the speed of the adoption process .
second , we measure the distribution @xmath95 of depths of induced vulnerable trees ( fig .
[ fig:3]d ) . finally ,
in order to verify earlier theoretical suggestions @xcite , we look at the correlation @xmath97 between the degree of innovators and the average size of vulnerable trees induced by them ( fig .
[ fig:3]e ) .
we perform the last two measurements on the real data and in the model process for @xmath107 and @xmath108 , as well as for the predicted value @xmath96 . in the case of @xmath97 , we find a strong positive correlation in the data , explained partially by degree heterogeneities in the underlying social network , but surprisingly well emulated by the model .
more importantly , although both quantities appear to scale with @xmath33 , measures for the estimated @xmath33 value fit the empirical data remarkably well , confirming our earlier validation based on the matching of relative component sizes ( for further discussion see si ) .
although some products and innovations diffusing in society may cover a large fraction of the population , their spreading tends to follow slow cascading patterns , the dynamics of which have been modelled before by simple diffusion models like that of bass @xcite .
however , this approach neglects threshold mechanisms that arguably drive the decision making of single individuals . on the other hand , threshold models study the conditions for cascades in global diffusion but do not address their temporal evolution , which is clearly a relevant factor in real - world adoption processes .
these models are commonly used to predict rapid cascading patterns of adoption , which is a more realistic scenario for the spreading of information , opinions , or behavioural patterns but are not observed in the case of product or innovation diffusion where adoption requires additional efforts , e.g. , free or paid registration . here
we provide a solution for this conundrum by analysing and modelling the worldwide spread of an online service in the techno - social communication network of skype . beyond the novel empirical evidence about heterogeneous adoption thresholds and non - linear dynamics of the adoption process
, we identify two additional components necessary to introduce in the modelling of product adoption , namely : ( a ) a constant flow of innovators , which may induce rapid adoption cascades even if the system is initially out of the cascading regime ; and ( b ) a fraction of immune nodes that forces the system into a quenched state where adoption slows down .
these features are responsible for a critical structure of empirical adoption components that radically differs from previous theoretical expectations .
we incorporate these mechanisms into a threshold model controlled by the rate of innovators and the fraction of immune nodes .
the model is able to reproduce several pathways ranging from cascading behaviour to more realistic dynamics of innovation adoption . by constraining the model with empirically determined parameters ,
we provide an estimate for the real fraction of susceptible agents in the social network of skype , and validate this prediction through correlated structural features matching empirical observations .
our aim in this study was to provide empirical observations as well as methods and tools to model the dynamics of social contagion phenomena with the hope it will foster thoughts for future research .
one possible direction would be the observation of the reported structure and evolution of the global adoption cluster in other systems similar to the ones studied in @xcite .
other promising directions could be the consideration of homophilic or assortative structural correlations , the evolving nature of the underpinning social structure as studied in @xcite , interpersonal influence , or the effects of leader - follower mechanisms on the social contagion process .
finally , we hope that the reported results may improve efficiency in the strategies of enhancing the diffusion of products and innovations , by shifting attention from the creation of short - lived perturbations to the sustenance of external input .
the authors have no competing interests .
, r.k . , k.k , and j.k designed the research and participated in writing the manuscript .
r.k . and m.k .
analysed the empirical data , g.i .
made the analytical calculations , and m.k . and
g.i . performed the numerical simulations .
the authors gratefully acknowledge the support of m. dumas , a. saabas , and a. dumitras from stacc and microsoft / skype labs as well as constructive comments by j. saramki and t. nsi .
g.i . acknowledges the academy of finland , and j.k .
the cimplex fet open h2020 eu project for support .
this research was partly funded by microsoft / skype labs .
we use a static representation of the skype social network aggregated over 99 months between september 2003 and november 2011 .
we follow the adoption of the `` buy credit '' paid service for @xmath0 months starting from 2004 , and the paid service `` subscription '' for @xmath1 months starting from 2008 ( for further details about the network and service see si ) . by considering the online social structure and adoption times , we identify users as innovator , vulnerable , or stable nodes based on the number @xmath21 of adopting neighbours at the time of exposure .
thresholds are calculated as @xmath20 for users with @xmath109 contacts .
the adoption network is constructed by considering confirmed social links between users who adopted the service earlier than @xmath81 . in order to avoid the effect of instantaneous group adoptions ( evidently not driven by social influence ) ,
we only consider links between nodes who are neighbours in the underlying social network and whose adoption did not happen at the same time .
note that for the categorisation of nodes we use only the adoption time and the state of their peers , and thus real categories may differ slightly .
for example , an innovator may appear as a vulnerable or stable node , even if its decision was not driven by social influence but some of its peers adopted earlier . to consider this bias we measure effective rates of adoption for the model process as well , just like for the empirical case ( fig.[fig:1 ] ) and section s3 .
this measure is obtained by taking the maximum of the yearly adoption rate ( yearly count of adoptions ) normalised by the final observed adoption number of a given service .
it characterises the maximum speed of adoption a service experienced during its history and takes values between 0 ( no cascade ) and 1 ( instantaneous cascade ) .
we repeat this measurement for the estimated number of registered users of facebook , twitter , and linkedin @xcite , as well as for the number of active users of skype and three paid skype services . adoption rates for facebook , twitter , and
linkedin correspond to the period between 2006 and 2012 , and for skype and its services to the interval from release date until 2011 .
we use the skype data to directly determine all model parameters , apart from the fraction @xmath33 of immune nodes .
to best approximate the degree distribution of the real network , after testing different candidate functions ( see si ) we select a lognormal function @xmath110 with parameters @xmath16 and @xmath17 and minimum degree @xmath111 , leading to the average degree @xmath19 . to account for finite - size effects in the model results for low @xmath75 ( fig .
[ fig:2 ] ) , we decrease @xmath72 slightly to obtain the same value of @xmath77 as in the real network .
the threshold distribution of each degree group collapses to a master curve after normalisation by using the scaling relation @xmath112 .
this master curve can be well approximated by the lognormal distribution @xmath113 , with parameters @xmath25 and @xmath26 as determined by the empirical average threshold @xmath27 and standard deviation @xmath114 ( for further details see si ) .
we estimate a rate of innovators @xmath61 by fitting a constant function to @xmath115 for @xmath116 ( fig .
[ fig:1]f ) .
the fit to @xmath117 also matches the time scale of a monte carlo iteration in the model to 1 month .
model results ( fig .
[ fig:3]d and e ) are calculated with @xmath57 and @xmath61 .
simulation results in fig .
[ fig:3]c ( d and e ) are averaged over @xmath118 configuration - model networks of size @xmath119 ( @xmath120 ) after @xmath99 iterations , matching the length of the observation period in skype .
we characterise the static social network by the extended distribution @xmath121 , where @xmath122 for @xmath39 and @xmath123 for @xmath43 .
non - immune , susceptible nodes with property vector @xmath124 adopt spontaneously at a constant rate @xmath117 , else they adopt only if a fraction @xmath41 of their @xmath109 neighbours have adopted before .
these rules are condensed in the probability @xmath125 that a node will adopt in a small time interval @xmath126 , given that @xmath35 of its neighbours are already adopters , @xmath127 with @xmath128 @xmath129 and @xmath130 @xmath131 ( for immune and isolated nodes , respectively ) .
the dynamics of adoption is well described by an ame for the fraction @xmath132 of @xmath124-nodes that are susceptible at time @xmath81 and have @xmath133 adopting neighbours @xcite , @xmath134 where @xmath135 . to reduce the dimensionality of eq .
( [ eq : amesthres ] ) we consider the ansatz @xmath136 for @xmath137 , leading to the condition @xmath138 . with @xmath139 and some algebra , this condition
is reduced to eq .
( [ eq : reducedames ] ) ( see si ) .
10 centola , d. .
, 11941197 ( 2010 ) .
mcpherson , m. , smith - lovin , l. & cook , jm . . , 415 - 444 ( 2001 ) .
j. l. , cha .
m. & gonzlez , m. c. .
, e29528 ( 2012 ) .
castellano , c. , fortunato , s. & loreto , v. .
, 591646 ( 2009 ) .
rogers , e. m. .
( simon & schuster ) , 5th edition ( 2003 ) .
granovetter , m. .
, 14201443 ( 1978 ) .
schelling , t. c. .
, 488493 ( 1969 ) .
axelrod , r. .
, 203226 ( 1997 ) .
watts , d. j. .
, 57665771 ( 2002 ) .
handjani , s. .
, 737746 ( 1997 ) .
valente , t. w. .
, 6989 ( 1996 ) . watts , d. j. & dodds , p. s. .
, 441458 ( 2007 ) .
melnik , s. , ward , j. a. , gleeson , j. p. & porter , m. a. .
, 013124 ( 2013 ) .
gmez , v. , kappen , h. j. , & kaltenbrunner , a. .
( ht11 , acm , new york , ny , usa ) , pp . 181190 ( 2010 ) .
centola , d. & macy , m. .
, 702734 ( 2007 ) .
barrat a. , barthlemy , m. & vespignani , v. .
( cambridge university press , 2008 ) .
bass , f. m. .
, 215227 ( 1969 ) .
aral , s. , muchnika , l. , sundararajana , a. .
, 2154421549 ( 2009 ) .
shalizi , c. r. , thomas , a. c. .
, 211239 ( 2011 ) .
holt , c. a. .
( addison wesley , 2006 ) .
bikhchandani , s. , hirshleifer d. & welch , i. .
, 9921026 ( 1992 ) .
karsai , m. , iiguez , g. , kaski , k. & kertsz , j. .
, 20140694 ( 2014 ) .
ruan , z. , iiguez , g. , karsai , m. & kertsz , j. .
, 218702 ( 2015 ) .
gonzlez - bailn , s. , borge - holthoefer , j. , rivero , a. & moreno , y. . , 197 ( 2011 ) .
borge - holthoefer , j. , et .
, e23883 ( 2011 ) .
ellis , c. j. & fender , j. .
, 763792 ( 2011 ) .
dow , p. a. , adamic , l. a. & friggeri , a. .
( icwsm , aaai , boston , ma , usa ) , pp . 145154 ( 2013 ) .
gruhl , d. , guha , r. , nowell , d. l. & tomkins , a. .
( www 04 , acm , new york , ny , usa ) , pp .
491501 ( 2004 ) .
baos , r. a. , borge - holthoefer , j. & moreno , y. .
, 6 ( 2013 ) . hale , h. e. . , 331353 ( 2013 ) .
leskovec , j. , singh , a. & kleinberg , j. .
( pakdd 06 , singapore ) , pp .
380389 ( 2006 ) .
leskovec , j. , adamic , l. a. & huberman , b. a. .
( tweb , acm , new york , ny , usa ) , vol .
1 , pp . 5 ( 2007 ) .
goel , s. , watts , d. j. & goldstein , d. g. .
( ec 12 , acm , new york , ny , usa ) , pp . 623638 ( 2012 ) .
fowler , j. h. & christakis , n. a. .
, 53345338 ( 2009 ) .
borge - holthoefer , j. , baos , r. a. , gonzlez - bailn , s. & moreno , y. .
, 122 ( 2013 ) .
hackett , a. & gleeson , j. p. .
, 062801 ( 2013 ) .
gleeson , j. p. .
, 046117 ( 2008 ) .
brummitt , c. d. , dsouza , r. m. & e. a. leicht , .
, e680e689 ( 2011 ) .
ghosh , r. & lerman , k. , wsdm 11 .
( wsdm 11 , acm , new york , ny , usa ) , pp . 665674 ( 2010 ) .
hurd , t. r. & gleeson , j. p. .
, 25 - 43 ( 2013 ) .
singh , p. , sreenivasan , s. , szymanski , b. k. & korniss , gy . .
, 2330 ( 2013 ) .
gleeson , j. p. & cahalane , d. j. .
, 050101(r ) ( 2007 ) .
white , d. s. ( 2013 ) .
date of access : 2015.01.29 .
morrissey , r. c. , goldman , n. d. & kennedy , k. p. ( 2011 ) . date of access : 2014.10.14 .
porter , m. a. & gleeson , j. p. .
gleeson , j. p. .
, 021004 ( 2013 ) .
gleeson , j. p. .
, 068701 ( 2011 ) .
bakshy , e. , hofman , j. m. , mason , w. a. & watts , d. j. .
( wsdm 11 , acm , new york , ny , usa ) , pp . 6574 ( 2011 ) . [ [ section ] ]
this study has been conducted on a dataset of the social network of skype .
the centrepiece of the dataset is the _ contact network _ , where nodes represent users and edges between pairs of users exist if they are in each other s contact lists . a user s contact list is composed of _
friends_. if user @xmath140 wants to add another user @xmath141 to his / her contact list , @xmath140 sends @xmath141 a contact request , and the edge is established at the moment @xmath141 approves the request ( or not , if the contact request is rejected ) .
each edge is labelled with a time stamp indicating the moment the contact request was approved . as the underpinning social structure we consider the static representation of the skype social network , aggregated for @xmath2 months between september 2003 and november 2011 .
the largest connected component of this structure includes roughly 510 million users and 4.4 billion edges .
as the chosen service evolving on the skype network , we follow how users purchase `` credits '' for calling phones . for each user ,
the dataset includes the date when he / she first adopted the paid product `` buy credit '' ( first credit purchase , for all purposes ) .
we select this service since its lifetime of @xmath0 months is considerably long ( it was introduced in 2004 ) , and it can be adopted by registered skype users only . this way the aggregated skype network provides a complete description of the mediating social structure , which allows us to calculate the correct degree and adoption threshold for all individuals . to make additional observations and to further test our model ,
we perform calculations on a second paid service called `` subscription '' , which was introduced in april 2008 , lasts for over @xmath1 months , and can also be adopted by registered skype users only .
results regarding this service are presented in section [ sec : addserv ] . by considering the online social structure and the adoption times
we identify users as innovator , vulnerable , or stable nodes based on the number @xmath21 of adopting neighbours at the time of exposure .
thresholds are calculated as @xmath20 for users with @xmath109 contacts .
the adoption network is constructed by considering confirmed social links between users who adopted the service earlier than the time of observation @xmath81 . in order to avoid the effect of instantaneous group adoptions ( evidently not driven by social influence ) ,
we only consider links between nodes who are neighbours in the underlying social network and whose adoption did not happen at the same time .
note that for the categorization of nodes we use only the adoption time and the state of their peers , and thus ` real ' categories may differ slightly .
for example , an innovator may appear as a vulnerable or stable node , even if its decision was not driven by social influence but some of its peers adopted earlier . to consider this bias we also measure ` effective ' rates of adoption for the model process , just like for the empirical case ( fig.1 , main text ) and section s3
. the dataset does not include identity information .
all usernames are anonymized and there is no way of inferring a user s identity solely from the profile .
the dataset does not contain any information about interpersonal interactions , apart from the contact list .
parameters in the model are the rate of innovators @xmath142 , the degree distribution @xmath3 , the threshold distribution @xmath6 , and the fraction of immune nodes @xmath33 . other than @xmath33 , all of them can be estimated from the data as follows .
as discussed in the main text , the rate of spontaneous adoption saturates approximately to a constant value after an initial transition period , which allows us to determine the rate of innovators by fitting a constant function on the curve after time @xmath143 .
we estimate this rate to be @xmath144 , as demonstrated in fig .
s[fig : nullmodelrate]a where the dashed line assigns the fitted constant function .
degrees in the aggregated skype network are broadly distributed with a fat tail corresponding to strong degree heterogeneities . to characterize this distribution
analytically we select two candidate distribution functions .
the first is a shifted power - law distribution function of the form , @xmath145 where @xmath109 denotes the degree , @xmath146 is the power - law exponent scaling the tail of the distribution , and @xmath147 is the minimum degree ( in our case 1 ) .
@xmath148 is a constant scaling the shift of the distribution , which can be determined as @xmath149 since we know the average degree @xmath19 of the empirical network .
this way our only free parameter during the fit is the degree exponent @xmath146 . after fitting this function by using the non - linear least - square method
, we obtain a relatively good match with the empirical distribution ( fig .
s[fig : degdistr]a ) for exponent @xmath150 .
our second candidate function is a lognormal distribution function of the form , @xmath151 where @xmath72 and @xmath152 are the scaling parameters . after fitting this function by using the non - linear least square method with two free parameters ( @xmath72 and @xmath152 )
, we obtain an excellent fit with the empirical distribution for parameters @xmath16 and @xmath17 . to select the best candidate function , we calculate the corresponding jensen - shannon ( @xmath153 ) divergence values @xcite between the empirical and fitted distributions . as a result
we find that for the shifted power - law function the best fit provides @xmath154 , while for the lognormal distribution we get @xmath155 .
thus we select the lognormal distribution as the best analytical function describing the degree distribution of the empirical network .
the adoption threshold @xmath156 of a node is defined as @xmath157 , i.e. the fraction of adopting neighbours that trigger the adoption of the central node
. therefore it can only take certain fractional values determined by the degree @xmath109 .
although thresholds are defined as a fraction , by considering nodes of the same degree we can focus on the integer threshold @xmath21 , defined as the number of a node s neighbours who have adopted the service earlier . in our method
we first group nodes of the same degree , record their integer thresholds , and then calculate the threshold distribution for each degree group , as shown in the main text ( fig . 1e , inset ) .
these distributions collapse to a master curve after normalization by using the scaling relation @xmath112 ( fig . 1e , main panel ) .
moreover , this master curve can be well approximated by a lognormal distribution of the form , @xmath158 where @xmath25 and @xmath26 , as determined by the empirical average threshold @xmath27 and standard deviation ( std ) @xmath114 .
these findings indicate that although individual thresholds are strongly determined by degree , their distribution is degree - invariant , suggesting that the fraction of adopting friends rather than its absolute number is relevant during the service adoption process .
the estimated empirical values of parameters are summarized in table [ table : pars ] . ) and 2nd largest ( @xmath93 ) components of the model network ( ` net ' ) , model adoption network ( ` casc ' ) , model stable network ( ` stab ' ) , and induced vulnerable trees ( ` vuln ' ) as a function of @xmath33 . dashed lines show the observed relative size of the real @xmath92 of the adopter network in 2011 ( fig .
s[fig : srv4adclust ] , main panel ) and the predicted @xmath33 value .
the lower panel depicts the time @xmath159 when the adoption process has reached @xmath68 of the susceptible network as a function of @xmath33 .
we use @xmath118 realizations of configuration - model networks with size @xmath119 and lognormal degree distribution parametrized as described in section [ sec : degdistrfit ] .
model calculations correspond to the parameters of table [ table : parssrv4 ] for @xmath1 iteration steps ( matching the length of the observation period).,width=415 ] we repeat all model calculations with the parameters of the `` subscription '' service to see whether we can recover its adoption dynamics by using the dynamical threshold model introduced in the main text and in section [ sec : thresmodel ] .
we check the dependence on @xmath33 of the average size of the largest connected component of the network ( @xmath92 ) of susceptible nodes available for the adoption process , the adoption network , the stable adoption network , and of vulnerable trees ( fig .
s[fig : srv4model ] , upper panel ) .
in addition we record the average size @xmath93 of the second largest connected component ( fig .
s[fig : srv4model ] , middle panel ) .
finally we show the time when the adoption process has reached the @xmath68 of available susceptible nodes in the adoption network ( fig .
s[fig : srv4model ] , lower panel ) .
the @xmath33 dependence of the adoption process appears to be qualitatively similar to our earlier calculations on the `` buy credit '' service , but there are remarkable differences .
firstly , the crossover regime ( depicted by the light grey area in fig .
s[fig : srv4model ] ) is shifted towards larger @xmath33 values due to the different threshold distribution and innovator adoption rate . secondly , after matching the relative size of the largest connected component of the empirical adoption network ( last point on the right - hand side of fig .
s[fig : srv4adclust ] , main panel ) , the predicted @xmath160 is out of the crossover regime .
at this point the background social network is still not fragmented ( as evidenced by the black line in fig .
s[fig : srv4model ] , which has not reached its maximum yet ) and it allows for the emergence of large connected adoption clusters .
it is very sparse , however , which explains : ( a ) the dominating innovator adoption rate observed empirically ; ( b ) the reduced size of the giant component of the adoption and stable adoption networks ; and ( c ) the relatively large innovator trees as compared to the stable adoption network components .
we observe that the largest vulnerable trees are smaller than the largest stable clusters in the empirical data , while the opposite is true for the model .
a possible explanation of this difference is the assumption in the model that the network is degree - uncorrelated .
this is a necessary approximation in order to treat the model analytically , but it might not hold for the empirical network . all in all , this picture suggests that the `` subscription '' service is out of the rapid and even the crossover cascading regimes , and that its dynamics is mostly driven by independent innovators rather than social influence , on a network of which a large majority is not susceptible to innovation . | adoption of innovations , products or online services is commonly interpreted as a spreading process driven to large extent by social influence and conditioned by the needs and capacities of individuals .
to model this process one usually introduces behavioural threshold mechanisms , which can give rise to the evolution of global cascades if the system satisfies a set of conditions .
however , these models do not address temporal aspects of the emerging cascades , which in real systems may evolve through various pathways ranging from slow to rapid patterns .
here we fill this gap through the analysis and modelling of product adoption in the world s largest voice over internet service , the social network of skype .
we provide empirical evidence about the heterogeneous distribution of fractional behavioural thresholds , which appears to be independent of the degree of adopting egos .
we show that the structure of real - world adoption clusters is radically different from previous theoretical expectations , since vulnerable adoptions induced by a single adopting neighbour appear to be important only locally , while spontaneous adopters arriving at a constant rate and the involvement of unconcerned individuals govern the global emergence of social spreading .
* keywords : * cascading behaviour , social spreading phenomena , complex contagion , adoption thresholds |
geometrically and optically thick belts of matter can be seen around many active galactic nuclei ( antonucci 1993 ) . although the evidence is best for nearby seyfert galaxies ( e.g. , the direct detection by jaffe et al .
2004 of the torus in ngc 1068 ) , there is also strong indirect evidence that this picture applies to radio galaxies ( barthel 1989 , di serego alighieri et al .
1994 ) and quasars ( e.g. , the spectropolarimetry presented in zakamska et al .
the ratio of observed numbers of obscured agn to unobscured gives a measure of the fraction of opaque solid angle ; it is generally @xmath2 ( hao et al . 2005 ) .
x - ray column densities to obscured nuclei range from @xmath3 @xmath5 to in excess of @xmath6 @xmath5 ( risaliti et al . 1999 ,
treister et al .
2004 ) ; applying a galactic dust / gas ratio would imply a dust extinction @xmath7@xmath8 mag , easily ample for blocking the entire optical and ultraviolet continuum .
one of the principal puzzles about these tori is how they stand up against gravity .
orbital speeds in galactic potentials are generically @xmath9 km s@xmath10 , and the nearby black hole can only increase this characteristic speed . to stretch upward enough to block a significant fraction of solid angle
requires a vertical velocity component that is a sizable fraction of the orbital speed ; if the velocities are interpreted as thermal , temperatures of at least @xmath11 k are implied .
if the gas is this hot , how can dust survive ?
numerous ideas have been proposed to answer this question , but none has proved entirely satisfactory .
the first suggestion ( made by krolik & begelman 1988 and elaborated by beckert & duschl 2004 ) was that the gas and dust are highly clumped , and the clumps have highly supersonic motions . to support these motions against the inevitable losses in cloud - cloud collisions ,
orbital shear energy can be randomized if magnetic fields make the clouds sufficiently elastic .
the central problem with this scheme is that , although physically possible , the required magnetic field strengths are not terribly plausible .
another idea , almost as old , is that the gas and dust are always _ locally _ geometrically thin , but the plane in which they orbit varies as a function of radius ( sanders et al . 1989 ) . countering this suggestion
are both the indirect evidence of well - formed ionization cones " and polarized reflection regions collimated not far from the innermost part of the torus , and the direct evidence provided by detection of a geometrically thick structure right at the torus s inner edge . once again
, one obtains the clearest view of all three sorts of data from ngc 1068 ( capetti et al .
1997 , kishimoto 1999 , jaffe et al .
still another suggestion was made by knigl & kartje ( 1994 ) , who argued that the large vertical motions required by geometrical thickness are best explained as arising in a magneto - centrifugal wind .
the principal drawbacks to this idea are the unknown origin of the large - scale magnetic field and the large energy source needed to drive the wind .
still another idea is that magnetic fields alone support a static equilibrium ( lovelace et al . 1998 ) .
the last notional explanation to list is that the large optical / ultraviolet radiation flux of the agn is converted to mid - infrared by dust at the inner edge of the torus , and the large opacity of dust in that band couples the radiation so strongly to the torus matter that the radiation force is comparable to gravity ( pier & krolik 1992a ) .
when first proposed , this scheme was only an order of magnitude estimate because its authors did not attempt to find a self - consistent solution of both the infrared transfer problem and the force balance problem .
it is the object of this paper to show , via an idealized model , that self - consistent equilibrium solutions in which the torus is supported by infrared radiation force do exist .
the model presented is admittedly highly simplified and rests on a number of rough approximations .
however , it does contain all the zeroth - order physics of the problem , and , as will be shown , it is completely solvable analytically .
moreover , the solutions found demonstrate certain qualitative features that should apply to any more realistic description .
obscuring tori wrap around a central active nucleus . in their central hole " , the matter must be largely transparent , so that viewers located along the axis can see the nucleus clearly ( they then see a type 1 " agn ) .
on the other hand , on oblique and equatorial lines of sight , there is so much opacity to wavelengths from the near - infrared to soft x - rays that viewers in those directions see nothing but infrared radiation from the warm dust in the torus or very hard x - rays . the very fact that these tori are , indeed , obscuring " means that they must act as light reprocessing machines which are heated from the inside and cooled from the outside .
radiation flux from the agn passes through the torus ( although suffering some drastic changes in spectrum _ en route _ ) , entering through its inner edge and departing through its upper and outer surfaces .
the key point behind the idea of supporting the geometric thickness of the torus by infrared radiation pressure is that the acceleration due to a given radiation flux @xmath12 is @xmath13 , where @xmath14 is the opacity per unit mass .
the luminosity @xmath15 capable of balancing the gravity of a mass @xmath16 in spherical symmetry ( the effective eddington luminosity " ) is therefore @xmath17 , inversely proportional to the opacity .
because the opacity of dust per unit mass of gas is an order of magnitude greater than the thomson opacity per unit mass when the radiation temperature is in the range 1001000 k ( semenov et al . 2003 ) , @xmath18 for predominantly infrared light passing through dusty gas ( this argument can be rephrased more precisely in terms of the divergence of the radiation pressure tensor , but its essential grounding still lies in the relatively high opacity of dusty gas ) .
enhanced heavy - element abundances likely lead to larger than local dust / gas ratios ; where that is the case , @xmath19 might be even smaller .
luminosities @xmath20 are commonly expected in agn , so if a significant part of the flux striking the inner surface of the torus can be converted to mid - ir wavelengths , radiation forces comparable to gravity can easily result .
rotational support against gravity is so common in astrophysical contexts that it is entirely plausible to suppose that it is important here , too .
however , the matter of the torus must have an effective collision rate at least as large as the orbital frequency .
if it is fluid , this follows by definition .
if it is highly clumped , this condition is compelled by the requirement that the torus be consistently opaque ( krolik & begelman 1988 ) .
when the average line of sight through the torus has at least one clump on it , the mean collision rate of clumps is at least the orbital frequency .
if these collisions are at all dissipative , one might expect the torus matter to settle into the plane normal to the total angular momentum and no longer be geometrically thick .
adding radiation to the picture changes this conclusion .
radiation diffusing through the equatorial plane of a geometrically thin but optically thick annular structure quickly develops a large vertical flux in the region just outside its inner edge because that is the most direct path out of the opaque matter . at the order of magnitude level ,
this vertical flux is @xmath21 , where the inner radius of the annulus is @xmath22 and its half - thickness is @xmath23 .
the upward acceleration it creates , @xmath24 , competes with the downward acceleration of gravity , @xmath25 .
because the same factor of @xmath26 enters the expressions for both the acceleration due to radiation and the acceleration due to gravity , if the magnitude of the flux is comparable to or greater than the effective eddington flux , the annulus expands vertically . greater thickness leads , of course , to interception of more light in a manner exactly balancing the growing magnitude of the vertical gravity .
a corollary of oblique flux producing a force comparable to gravity is that the radial component of the radiation force is also comparable to gravity .
if this is so , radial force balance demands a rotation rate that is sub - keplerian . because the radial flux diminishes outward faster than @xmath27 as flux
is diverted to the vertical direction , in equilibrium the specific angular momentum of the torus must increase outward , approaching keplerian . in a real obscuring torus ,
the optical through soft x - ray continuum of the active nucleus is absorbed in a thin layer along its inner edge , where warm dust reprocesses the nuclear luminosity into the infrared .
this inner edge is a complicated place , as we will discuss below .
we therefore begin with the simpler problem of finding a self - consistent description of dynamics and radiation transfer in the torus interior .
a more precise definition of the boundaries of this region , and therefore the domain of applicability of these results , will be given in the following subsection .
the simplest non - trivial geometry in which this picture can be explored is 2-d axisymmetry .
adopting cylindrical coordinates @xmath28 and @xmath29 , we write the equation of hydrostatic equilibrium for radiation and rotation balancing gravity as @xmath30 where @xmath31 is the local orbital frequency and the gas s specific angular momentum is @xmath32 .
note that we are supposing that gas pressure gradients are entirely negligible .
we also adopt three simplifying assumptions , all appropriate to flattened geometries .
first , we take @xmath31 at all heights @xmath29 to be the rotation rate of a circular orbit in the torus midplane at cylindrical radius @xmath28 .
second , we follow only the component of angular momentum parallel to the torus axis , and @xmath33 is assumed to be a function of @xmath28 alone .
third , we approximate the vertical component of the gravity by @xmath34 . this last approximation would be exact if we evaluated @xmath31 at the actual local value of @xmath29 , rather than at @xmath35 . ideally , to find the flux
, one would solve a complete transfer problem at all relevant frequencies for all photon directions .
here we take a much simpler approach : the gray diffusion approximation , using a thermally - averaged opacity . in this approximation ,
the flux is given by @xmath36 where @xmath37 is the gas density and @xmath38 is the radiation energy density .
under the assumption of hydrostatic balance ( eqn .
[ eq : hydrostat ] ) , the radiation energy density and the dynamics are related by @xmath39 if the only source of infrared radiation is the conversion via dust reradiation of optical and ultraviolet photons at the inner edge of the torus , then in the body of the torus @xmath40 it can also be of interest to explore the effect of distributed sources of infrared photons .
these may be created , for example , by local heating due to compton recoil when hard x - rays penetrate deep in the torus material ( as discussed , e.g. , in chang et al .
2006 ) or by the absorption of locally - generated starlight . when there are distributed sources , the right - hand - side of equation [ eq : fluxcons ] may be non - zero . for the purposes of this paper ,
however , we adopt the simple assumption of no internal sources . with that assumption
, the diffusion equation becomes @xmath41 combining equation [ eq : fluxconsdiff ] with equation [ eq : eraddyn ] gives @xmath42\right\ } = 0.\ ] ] detailed radiation transfer studies of obscuring tori consistently find that their interior temperatures are in the range 1001000 k ( pier & krolik 1992b , efstathiou & rowan - robinson 1995 , granato et al .
1997 , nenkova et al .
according to the most recent dust opacity models ( e.g. , semenov et al .
2003 ) , the rosseland mean opacity for gas of solar abundances and normal dust content is @xmath4330 times greater than thomson and has no consistent trend within this temperature range , instead exhibiting only a few mild local maxima and minima ; the ratio of the largest opacity to the smallest is no more than @xmath44 . on this ground , we approximate @xmath14 as exactly constant .
equation [ eq : fluxhydrostat ] then reduces to @xmath45 + { \partial \over \partial z}\left [ z \omega^2\right ] = 0.\ ] ] although the black hole mass may be larger than the stellar mass enclosed within the torus , it is easy to allow for potentials more general than a simple point - mass by writing @xmath46
. then , for any particular potential described by @xmath47 , equation [ eq : reducedhydrostat ] determines the unique @xmath48 permitting a self - consistent solution . written in terms of @xmath49 , equation
[ eq : reducedhydrostat ] becomes @xmath50 which has the solution @xmath51\left({{r \over r_{\rm in}}}\right)^{2(\alpha-1 ) } - f(\alpha),\ ] ] where @xmath22 is the radius at which the higher - energy photons are converted to infrared , @xmath52 , and @xmath53 .
when @xmath54 ( a point - mass potential ) , @xmath55 and @xmath56 .
when @xmath57 ( a logarithmic potential ) , a logarithmic dependence replaces the power - law : @xmath58 . assuming that the point where @xmath59 marks the outer edge of the torus , we can use this solution to find the span of radii over which the torus exists .
for @xmath60 , it stretches from @xmath22 to @xmath61^{1/[2(\alpha-1)]};\ ] ] in the special case of the logarithmic potential , @xmath62 . as figure
[ fig : rmax ] illustrates , for fixed @xmath63 , the breadth of these tori stretches moderately as the slope of the potential steepens : for @xmath64 , @xmath65 rises from @xmath66 for @xmath57 to @xmath67 for @xmath68 .
not surprisingly , @xmath65 increases with diminishing @xmath63 at fixed @xmath49 . for fixed @xmath63 .
the case illustrated is for @xmath64 .
[ fig : rmax ] ] with @xmath69 known , equation [ eq : eraddyn ] may be separated into two equations for @xmath37 , which must be consistent with each other : @xmath70}\frac{\partial e}{\partial r}.\ ] ] it is instructive to rewrite in characteristic form the partial differential equation for @xmath38 implied by the second equality : @xmath71 with @xmath72}.\ ] ] from the characteristic form , we see that @xmath38 is constant along contours parameterized by the pair of characteristic equations
@xmath73 = -s + \lambda,\ ] ] where @xmath74 is arbitrary and we set the integration constant for @xmath75 to zero by choosing @xmath35 at @xmath76 . the radius at which a contour labelled by @xmath77 passes through @xmath35 is given implicitly by the second characteristic equation . equating the two expressions for @xmath78
gives an explicit definition of the contours of constant @xmath38 : @xmath79 = \lambda.\ ] ] evaluating the integral , we find : @xmath80 \left({{r \over r_{\rm in}}}\right)^{2\alpha } = \lambda,\ ] ] where we have redefined @xmath77 so as to absorb a number of terms that depend on the ( arbitrary ) @xmath74 .
because @xmath48 depends only on @xmath47 , the shapes of these contours depend only on the shape of the gravitational potential and on the boundary condition @xmath63 .
generically , they are closed curves with principal axes parallel to the @xmath28 and @xmath29 axes , and elongated in the @xmath29 direction .
for example , when @xmath54 , the contours are given by @xmath81 that is , the larger @xmath63 is , the more the contours stretch in the @xmath29 direction .
we may solve an ordinary differential equation for @xmath82 along any path on which @xmath77 varies monotonically .
for example , if we choose the path to run outward along the @xmath28 axis from @xmath22 , we have @xmath83 in other words , knowledge of @xmath84 on this path is a prerequisite for accomplishing such a solution .
physically , this should come as no surprise : the run of radiation energy density with position must certainly depend on how much matter there is with which to interact and how it is distributed .
somewhat arbitrarily , we choose to supply @xmath85 and integrate @xmath86.\ ] ] it is convenient in this context ( in which we have already written @xmath87 ) to consider a density boundary condition that is also a power - law in radius : @xmath88 . with this choice of @xmath89 , every term on the right - hand - side of equation
[ eq : ode ] is a power - law and the equation may be integrated exactly .
the result is @xmath90 + \frac{j_{\rm in}^2 + f(\alpha)}{\gamma } \left[\left({{r \over r_{\rm in}}}\right)^{-\gamma } -1\right]\right\}.\ ] ] note that if the power - law for @xmath89 were to extend to @xmath91 , securing a finite optical depth would require @xmath92 .
however , because we cut off the power - law at a finite maximum radius , this restriction on @xmath93 is eliminated .
once @xmath94 in the equatorial plane has been found , one may directly determine @xmath38 at _ all _ @xmath28 and @xmath29 by extending that solution into the plane following the characteristic curves that define its constant - value contours .
the density @xmath37 then follows from either of the equations [ eq : rhoconsist ] .
so far we have identified three parameters that govern the character of these solutions : @xmath49 , @xmath63 , and @xmath93 .
there are two more .
one of these is @xmath95 , which sets the optical depth scale .
if the density declines outward , it must be at least several , or there will be nowhere where the diffusion approximation is valid .
the other parameter is @xmath96 , whose physical meaning is most clearly seen when rewritten in terms of more familiar quantities : @xmath97 where @xmath98 is the total mass interior to @xmath22 , and @xmath99 is the mass of the central black hole alone ( which is what determines @xmath100 in the usual definition ) .
the factor @xmath101 is the amount by which the optical depth of the torus enhances @xmath102 over the value it would have if the radiation could stream out freely .
because we expect the torus to be optically thick in the mid - infrared , but radiation can escape freely through the axial hole , @xmath103 $ ] , where @xmath104 is the fraction of solid angle that is open as viewed from the position @xmath105 . from the statistics of type 1 versus type 2 agn
, we might suppose that @xmath106 .
if @xmath10730 , then @xmath108 .
reasonable values of @xmath109 would then be @xmath11030 .
our approximations do not hold at all locations , so it is important to mark off carefully the region of the @xmath28-@xmath29 plane in which we can apply them . because the phenomenology points us strongly toward a density distribution with an inner hole , we need to define an inner radial edge , within which the matter is optically thin . unfortunately , we do not know the shape of this inner edge _ a priori _ ; indeed , that shape is likely the result of some rather complicated dynamics , as will be discussed in the next several paragraphs . in this paper ,
whose goal is merely to demonstrate the possibility of thick structures supported by radiation pressure , we simply choose a vertical inner edge , i.e. , @xmath111 .
even if @xmath112 were a function of @xmath29 , the solution we have derived in the previous subsection would still hold wherever @xmath113 .
the extension of @xmath114 into the full @xmath28@xmath29 plane is entirely independent of the exterior boundary conditions because the hydrostatic balance condition substitutes , in effect , for boundary conditions in determining the interior diffusion solution .
reconciliation of our interior solution with the exterior boundary conditions will also be discussed later in this section . when we know more about @xmath115 , that knowledge will change the size and shape of the region where our hydrostatic diffusion solution applies , but will not not alter its nature in the torus interior .
unfortunately , a proper determination of the position of the inner edge is far beyond the scope of this initial effort ( pier & voit 1995 present a simplified model ) .
although the problem is superficially similar to the one treated here ( simultaneously solving the equations of 2-d radiation transfer and 2-d hydrostatic equilibrium ) , several new effects become important at the edge and drastically complicate the problem .
there is a large outward force from absorption and scattering of the optical / uv continuum arriving directly from the nucleus ; there is also a comparably large , but inward , force due to the infrared flux emerging from the torus . because the most important effects for both occur across photospheres ( at different places for the different wavelengths , of course ) the diffusion approximation is wholly inadequate for both .
gas pressure gradients , although ( by assumption in this model ) unimportant in the bulk of the torus , can also become significant locally . in one respect , evaluating
the force exerted by the optical / uv continuum is relatively simple : it has a single point - like source at the nucleus , so its transfer problem is essentially one - dimensional along radial rays ( departures from true one - dimensionality arise only to the degree that the albedo of dust permits scattering to spread the beam ) . on the other hand ,
the opacity , far from being roughly constant , is a strong function of the radiation intensity , making this transfer problem highly nonlinear . at the inner edge of the torus
, the gas can be photoionized and , in some cases , radiative heating can warm the dust above its sublimation temperature .
in addition , photoionization heating may raise the gas s temperature high enough for it to destroy the dust by sputtering .
the only simple aspect of the infrared transfer problem is its qualitative behavior in two extreme geometric limits : in the limit of infinitesimal opening angle , the infrared intensity would be identically constant across the central hole ; in the limit of very large opening angle , it would be considerably smaller along the axis than at the torus s inner edge .
however , because its principal opacity is also due to dust , many of the same difficulties that apply to the optical / uv continuum at the torus s inner edge also apply to the infrared .
in addition , because its sources are distributed , the solution is thoroughly 2-d and global . at the order of magnitude level , the infrared intensity in the central hole is likely to be comparable to or greater than the intensity of the optical / uv continuum precisely because of the blanketing provided by the optically thick torus
; we have hidden this ratio in the @xmath116 fudge factor already introduced .
the total force it exerts on the gas in the inner edge region is determined by the contrast in intensity between the main body of the torus and the axis . in rough terms
, we expect this fractional contrast to increase from the midplane upward because the fraction of open sky " seen from a position on the axis increases upward , but to be quantitative about this dependences requires a proper 2-d global transfer solution .
further complications can be caused by strong local gas pressure gradients , which are entirely absent in our model for the torus interior .
because the opacity of dust in the optical / uv is considerably greater than in the infrared , the radiation force due to this band is expressed across a much narrower zone than that due to the infrared . as an immediate consequence ,
sharp gas pressure gradients are likely to be created across these short lengthscales , even though the total gas pressure contrast from the torus body to the outside may not be that large .
gas pressure effects introduce further nonlinearity into the transfer problem because the gas s equation of state is also strongly dependent on the radiation intensity . summarizing this qualitative discussion
, we expect the location and shape of the inner edge to be the immediate result of balancing the opposing radiation and gas pressure gradient forces in the context of a rapidly changing physical state for the gas .
over longer timescales , mass flux balance will also come into play .
as matter is ionized and heated at the extreme inner edge of the torus , it rushes away toward the lower pressures found at higher altitudes in the torus hole and beyond ( krolik & begelman 1986 , balsara & krolik 1993 ) . to achieve a steady state
, matter must accrete through the torus in order to replace the evaporated matter .
the accretion rate is controlled by angular momentum transport ; if , as is common in other disks , this is due to mhd turbulence , the problem is further complicated .
thus , determining the position of the inner edge is a much more difficult problem than solving for the static structure of the torus interior .
the outer radial edge is determined by the requirement that @xmath117 ; greater @xmath33 would make hydrostatic equilibrium impossible .
although there is no automatic physical inconsistency created by placing the outer radial edge where @xmath118 , doing so begs the question of the dynamical state of the matter beyond : how does it make the transition from partial radiation force support to rotational support ?
the diffusion approximation is valid only within those regions whose optical depth to infinity is @xmath119 .
consequently , our solution should not be extended beyond the surface on which the vertical optical depth @xmath120 .
even if our approximations remained valid at small optical depth , one might define the torus edge as its infrared photospheric surface in any case . at the photosphere we have another boundary condition , but one that can be applied only approximately : the flux as estimated by the diffusion approximation should roughly match the flux as estimated on the basis of free - streaming . because we equate the diffusive flux with the flux necessary for hydrostatic balance
, this condition amounts to requiring that @xmath121 defined by @xmath122 . in other words , we find the photosphere using the density distribution in exactly that region where we know it least well . a further uncertainty is introduced by the fact that we can estimate to an accuracy of only a factor of @xmath44 the radiation energy density required to carry the flux in the optically thin regime .
for all these reasons , we require equation [ eq : opthinbc ] to be satisfied only to within a factor of 3 . there is also one additional constraint on acceptable solutions : if at anywhere along the radial axis @xmath123 , the solution is obviously unphysical . in practise
, we find that the photospheric boundary condition is best matched at the smallest @xmath93 such that @xmath124 everywhere in the range @xmath125 .
with these thoughts in mind , consider the typical " parameters @xmath64 , @xmath68 , @xmath126 , and @xmath127 .
that is , rotational support is substantially depressed at the inner edge , the potential is that of a pure point - mass , the column density in the midplane is @xmath4 @xmath5 , and @xmath1281 .
an acceptable solution requires @xmath129 , that is , the density declines slowly outward in the equatorial plane .
if the matter density declines more steeply , there is too little optical depth in the torus , and the energy density outside the photosphere predicted by the diffusion approximation is substantially larger than what is required to carry the flux in the optically thin regime ; if the matter density declines more slowly , the optical depth is too great , forcing @xmath38 to go negative .
the successful solution that results from a compromise between these two extremes ( i.e. , @xmath130 ) is illustrated in figure [ fig : typresult ] .
full keplerian support is reached at @xmath131 .
as expected , the contours of radiation energy density inside the torus are extended upward ; this is exactly what one would expect when the radiation finds it easier to move vertically than radially .
note that , by assumption , the local temperature @xmath132 , so the greatest temperature is found at @xmath105 , and it declines upward and outward from there .
contours of constant density , on the other hand , are extended radially .
this , too , is entirely in line with expectations , given the difficulty of vertical support against gravity .
in both panels of figure [ fig : typresult ] , a white curve shows the location of the photospheric surface on the top of the torus .
formally , our solution is invalid outside this white curve , as the diffusion approximation does not well describe the relation between energy density and flux in optically thin regions .
although much of the volume shown is in the optically thin region , most of its mass is within the optically thick portion of the torus .
consequently , while not taking too seriously the details of the solution in the optically thin zone , we can also be assured that they will not have serious impact on the issue of greatest concern here : the mass distribution within the torus . , @xmath68 , @xmath133 , @xmath126 , and @xmath127 .
left panel : radiation energy density .
right panel : matter density . in both , the scale is logarithmic , and the white curve shows the surface on which @xmath134 , the photosphere on the top of the torus . [
fig : typresult],title="fig : " ] , @xmath68 , @xmath133 , @xmath126 , and @xmath127 . left panel : radiation energy density .
right panel : matter density . in both , the scale is logarithmic , and the white curve shows the surface on which @xmath134 , the photosphere on the top of the torus . [
fig : typresult],title="fig : " ] having located the outer boundaries within which this solution is physical , it is now time to consider its behavior on the inner boundary . as discussed earlier , we are not prepared to present a proper global infrared transfer solution that includes the central axial hole .
qualitatively , however , we might expect that the radiation intensity inside the torus hole would fall with height somewhat faster than @xmath135 because the confinement due to the torus walls diminishes as its surface is neared .
in fact , in our fiducial model , @xmath136 falls slowly with increasing @xmath29 , reaching @xmath137 where the photosphere intersects the inner edge .
this sort of behavior therefore appears to be at least loosely consistent with what one might guess about transfer solutions inside the torus hole .
the density profile in the equatorial plane is a strong function of @xmath109 when all other parameters are held fixed .
larger @xmath109 demands a steeper profile : as it changes over the factor of 10 from 0.6 to 3.0 to 6.0 , the density power - law demanded rises from -2.58 to 0.50 to 3.05 .
in other words , the density must fall more steeply outward when @xmath138 is smaller .
when @xmath139 , of course , the integrated optical depth is dominated by the outermost radius ; solutions in this regime must involve very sharp cut - offs in the density profile at the outer radius .
greater rotational support at the inner edge means that full keplerian angular momentum is reached at a smaller radius .
for example , if @xmath140 but the other parameters are held fixed at their fiducial values , @xmath141 , and the matter density must rise outward quite steeply ( @xmath142 ) for there to be enough optical depth across this diminished radial thickness ; otherwise , the boundary condition matching the diffusion approximation energy density and the free - streaming energy density can not be satisfied .
conversely , if the fractional angular momentum support at the inner edge is much smaller , @xmath65 is much larger and the density falls steeply outward : for @xmath143 , @xmath144 and @xmath145 .
it is possible that distributed stellar mass may contribute significantly to the gravitational potential in the torus region .
if so , orbital speeds will decline more slowly with increasing radius than in the keplerian prediction . when this is the case , the equatorial density profile able to produce an equilibrium switches from one that gradually falls outward to one that gradually rises .
for example , for @xmath64 and @xmath127 , @xmath93 drops from 0.5 to -0.19 as @xmath49 falls from 1.5 to 1.1 .
the optical depth parameter @xmath146 does not appear explicitly in the differential equations , so the shapes of the matter and energy density contours , as well as the @xmath93 required by the radiation energy density boundary condition at the photosphere , do not depend on it . however ,
the outline of the torus that results _ does _ change with @xmath146 because the photosphere ( not surprisingly ) rises higher and higher with increasing optical depth .
we have found that hydrostatic radiation - supported solutions can be found for plausible parameters , particularly when @xmath109 is within a factor of several of unity . for these solutions to be found in real agn , two additional conditions must be satisfied : the matter s angular momentum must be redistributed so that the equilibrium @xmath48 is achieved ; and the radial matter density profile in the equatorial plane must be adjusted to the right shape .
it is possible that both may be achieved as a result of the magneto - rotational instability and the magnetic torques created by the mhd turbulence it drives .
this is because @xmath63 may be viewed as a function of @xmath93 : a matter density profile in the equatorial plane that does not fall as rapidly , or even rises outward ( i.e. , a smaller @xmath93 ) is consistent with dynamical equilibrium when the fractional rotational support at the inner edge is greater .
whether this actually happens in practise is uncertain . in conventional global disk simulations ,
mhd turbulence very efficiently redistributes angular momentum until it reaches a nearly - keplerian radial profile ( de villiers et al .
if that happened here , hydrostatic balance would be possible only if accompanied by a substantial positive @xmath147 .
this might perhaps be supplied by local sources of heat , such as stars or compton - heating due to hard x - rays that penetrate the bulk of the torus , but if sources such as these are inadequate , the disk would be forced to collapse to a thin configuration . on the other hand , in this context the equilibrium angular momentum distribution reached as a result of mhd turbulence might actually be the one required by this model ( eqn .
[ eq : jsqsoln ] ) because the net gravity " has been effectively reduced by the outward radiation force . in this context , it is notable that the velocity profile of the maser spots in ngc 1068 is @xmath148 , rather than @xmath149 as one might expect from circular orbits in a point - mass potential ( greenhill et al .
our model predicts a rotational speed that always declines more slowly outward than simple circular orbits in the gravitational potential would dictate ; for example , the mean rotational speed is constant as a function of radius when @xmath68 .
thus , a shallow rotation curve may not signal a stellar contribution to the gravitational potential it could instead be a symptom of radiation support .
maser kinematics can also be used to estimate the central mass , either from the magnitude of the circular speeds or from the acceleration @xmath150 seen in maser emission on the direct line of sight to the nucleus ( although the circular speed is a more easily - observed quantity , @xmath150 has been measured in four examples : henkel et al .
2002 and references therein ) . in the former case ,
the mass inferred on the basis of keplerian orbits is @xmath151 , in the latter , @xmath152 .
the sub - keplerian rotation that directly follows from the presence of radial radiation forces means that this inferred mass is an underestimate of the true mass by a factor @xmath153 in both cases . if the maser emission is driven by x - ray excitation ( neufeld et al .
1994 ) , it takes place very close to @xmath22 , so the relevant value of @xmath33 is @xmath63 .
interestingly , the shape of the angular momentum profile required for equilibrium depends only on the underlying gravitational potential , and in this respect is a comparatively robust prediction of the model . on the other hand , because @xmath63 is related to @xmath93 through the detailed equilibrium solution , which in turn depends on approximations like the photospheric boundary condition , their quantitative relationship is dependent on the quality of the several approximations made here .
we have also seen that a relatively small change in @xmath109 requires a large change in density profile : @xmath93 falls from @xmath154 to @xmath155 when @xmath109 falls from 6 to 0.6 .
if the optical depth is held fixed , @xmath109 is primarily dependent on @xmath138 , to which it is inversely proportional . a consequence of this model
, therefore , is that geometrically thick tori may be associated only with agn having @xmath138 within a range not much greater than a factor of 10 , logarithmically centered on @xmath156 . although the measurement of eddington luminosity ratios is still very difficult , this range is easily consistent with the data in hand ( e.g. , the maser - based measurement of the black hole mass in ngc 1068 : greenhill et al .
1996 , gallimore et al . 1996
; the somewhat shakier black hole masses from reverberation - mapping : metzroth et al .
2006 and references therein ; or the still shakier inferences from photoionization - scaling : mclure & dunlop 2004 ) .
more precisely , if stellar contributions to the gravitational potential are negligible , the fact that @xmath157@xmath158 for warm dust implies that tori should be puffed up by radiation whenever @xmath159@xmath160 .
the ratio @xmath161 , so it may vary over a modest range , from @xmath2 to a few , depending on the total optical depth through the torus and the shape of the inner hole .
infrared radiation pressure can not provide predominantly vertical support against gravity unless the torus is optically thick in the mid - infrared ; if it were optically thin , the flux would emerge more or less radially .
when @xmath92 , this requirement means that @xmath146 can not be less than a few .
however , when @xmath139 , as is often the case , most of the optical depth is found near the outer edge of the torus , so the lower limit on @xmath146 can be smaller . at the low end of the permitted optical depth range
, we would also expect @xmath101 to be no more than a few .
larger optical depth can not lead to much greater values of @xmath101 , however , because the axial hole through the torus creates an escape channel .
this is why , although solutions can be found for arbitrarily large values of @xmath146 , they are unrealistic unless somehow @xmath138 can be @xmath162 . measurements of the total optical depth in the equatorial plane of obscuring tori are difficult to come by .
the column density of matter on the line of sight can be measured directly by soft x - ray absorption ( or at least a lower bound placed when the compton depth is greater than unity ) . however , in most cases , we do not have any direct evidence of the inclination angle of the torus to our line of sight ; even if we did , it would be difficult to constrain directly the optical depth along the equatorial plane , in general a direction oblique to our line of sight . on the other hand , we can measure the statistical distribution of column densities to obscured agn and compare it to the predictions of this model , although this calculation is somewhat sensitive to the looseness in our location of the torus inner edge .
the predicted probability of seeing a given column density is simply proportional to the solid angle associated with the polar angle producing that column .
typically , these solutions predict a wide range of associated column densities because lines of sight farther from the equatorial plane pass through densities exponentially lower than those closer to the plane . because we are primarily interested in @xmath163 , which corresponds to @xmath164@xmath165 , the range of expected thomson depths is from @xmath166 to @xmath167 , in good correspondence with observations ( risaliti et al .
1999 , treister et al .
however , in contrast to the observational results , which tend to show a flatter distribution , there is a tendency for most of solid angle to be associated with the higher column densities .
for example , if we take @xmath168 , in our fiducial case the number of systems per logarithm of column density with @xmath169 is predicted to be @xmath170 the number seen with @xmath171 ( fig .
[ fig : coldensdist ] ) .
qualitatively , the shape of this distribution has two sources : the solid angle per unit polar angle is , of course , greatest near the equatorial plane , where the densities are also greatest ; and the high density region stretches away from the equatorial plane because the vertical component of gravity is relatively small there ( @xmath172 ) .
given the approximate character of our model , we expect that the prediction of a broad range of observed column densities should be fairly robust , but the exact shape of the predicted distribution is subject to significant uncertainty .
a different comparison with observations relates to the most fundamental reason why we believe the obscuration is geometrically thick : if the number of obscured agn is comparable to or greater than the number of unobscured agn having the same bolometric luminosity , then much of the sky surrounding the nucleus must be optically thick in the optical and ultraviolet . in principle , the solutions we have described also predict the fraction of solid angle obscured by the torus . for this question
, however , the arbitrariness of our inner edge introduces an especially large systematic uncertainty .
because only small column densities are required to stop uv photons if there is a normal dust / gas ratio , any curvature of the inner edge could be significant .
the following numbers should therefore be taken more in the way of examples than as serious predictions .
nonetheless , if we temporarily leave aside these considerations , we can still estimate the obscured solid angle by taking these models at face - value .
if the wavelength of interest has an opacity @xmath173 ( corresponding to @xmath174 : draine & lee 1984 ) , @xmath175 of solid angle is opaque in our fiducial model .
this corresponds to a half - opening angle of @xmath176 .
not surprisingly , larger @xmath109 ( lower luminosity relative to eddington ) leads to a smaller obscured fraction , but only slightly smaller : @xmath177 ( for all other parameters fixed ) makes a torus that blocks @xmath178 of the sky ( @xmath179 half - opening angle ) . in summary , we have constructed a simple , and entirely analytic , model of how infrared radiation pressure can , in principle , support obscuring tori around agn .
the model employs numerous approximations and simplifications that would undoubtedly be improved in a more complete and realistic picture .
most notably , a true transfer solution , rather than one adopting the diffusion approximation , would allow a proper connection to the boundary conditions at the photosphere of the obscuring matter .
frequency - dependent opacities would also improve its realism , although probably not as dramatically . in its treatment of the gas , a more realistic model might allow for clumping , both in regard to the opacity and to permit the introduction of supersonic random motions .
however , we also wish to point out that , to the degree vertical support is provided by radiation pressure , the necessity of supersonic motions and therefore clumping is diminished .
this model also raises a number of questions .
for example , as already discussed , the required angular momentum profile may or may not be achieved . it is also unclear whether this equilibrium is stable to a variety of perturbations
smooth motions in the gas , clumping in the gas , departures from the equilibrium angular momentum profile to name a few possibilities .
it is similarly uncertain whether the equilibrium , even if stable , can be reached from a wide range of initial conditions . despite these questions ,
the complete analytic solvability of this model means that we can learn from it a number of interesting qualitative facts about radiation forces in this context .
specifically , we have shown that a density distribution for the obscuration can be found in which _ both _ radiation diffusion and dynamics are in equilibrium .
this distribution , as demanded by the phenomenology of agn , can obscure a sizable solid angle for reasonable parameters . we have also found that in order for the equilibrium to be possible , several conditions must be met : the luminosity of the nucleus must be within a factor of several of @xmath180 ; because there is always a significant radial radiation force , the matter in the torus must orbit more slowly than in a keplerian orbit ; the thomson depth of matter in the equatorial plane must be not too far from @xmath181 ; and the level of rotational support relative to keplerian is linked to the radial profile of matter density in the equatorial plane , @xmath138 , and the total optical depth of the matter .
antonucci , r.r.j .
1993 , araa , 31 , 473 balsara , d.s . & krolik , j.h .
1993 , ap j , 402 , 109 barthel , p. 1989 , apj , 336 , 606 beckert , t. & duschl , w. 2004 , a&a 426 , 445 capetti , a. , macchetto , f.d . & lattanzi , m.g .
1997 , apj letts , 476 , l71 chang , p. , quataert , e. & murray , n. 2006 , in preparation de villiers , j .-
p . , hawley , j.f . &
krolik , j.h .
2003 , apj , 599 , 1238 di serego alighieri , s. , cimatti , a. , & fosbury , r.a.e .
1994 , apj , 431 , 123 draine , b.t . &
lee , h.m .
1984 , apj , 285 , 89 efstathiou , a. & rowan - robinson , m. 1995 , mnras , 273 , 649 gallimore , j.f . ,
baum , s.a . ,
odea , c.p . ,
brinks , e. & pedlar , a. 1996 , apj , 462 , 740 granato , g.l . , danese , l. & franceschini , a. 1997 , apj , 486 , 147 greenhill , l.j . , gwinn , c.r . ,
antonucci , r.r.j . & barvainis , r. 1996 , apjletts , 472 , l21 hao , l. et al . 2005 , aj , 129 , 1795 henkel , c. , braatz , j.a . ,
greenhill , l.j . & wilson , a.s .
2002 , a & a 394 , l23 jaffe , w. et al .
2004 , nature , 429 , 47 kishimoto , m. 1999 , apj , 518 , 676 knigl , a. & kartje , j.f .
1994 , apj , 434 , 446 krolik , j.h . , & begelman , m.c .
1986 , apj letts , 308 , l55 krolik , j.h . ,
& begelman , m.c .
1988 , apj , 329 , 702 lovelace , r.v.e . , romanova , m. & biermann , p.l .
1998 , a&a , 338 , 856 mclure , r.j . &
dunlop , j.s .
2004 , mnras , 352 , 1390 metzroth , k.g . ,
onken , c.a . & peterson , b.m .
2006 , apj , 647 , 901 nenkova , m. , ivezi , z. & elitzur , m. 2002 , apjletts , 570 , l9 neufeld , d.a . ,
maloney , p.r , & conger , s. 1994 , apj letts .
436 , l127 pier , e.a . &
krolik , j.h .
1992 , apjletts 399 , l23 pier , e.a . &
krolik , j.h .
1992 , apj , 401 , 99 pier , e.a .
& voit , g.m .
1995 , apj , 450 , 628 risaliti , g. , maiolino , r. & salvati , m. 1999 , apj , 522 , 157 sanders , d.b . ,
phinney , e.s . ,
neugebauer , g. , soifer , b.t . & matthews , k. 1989 , apj , 347 , 29 semenov , d. , henning , t. , helling , c. , ilgner , m. & sedlmayr , e. 2003 , a&a , 410 , 611 thompson , t. , quataert , e. & murray , n. 2005 , apj , 630 , 167 treister , e. et al .
2004 , apj , 616 , 123 zakamska , n. et al .
2005 , aj , 129 , 1212 | explicit 2-d axisymmetric solutions are found to the hydrostatic equilibrium , energy balance , and photon diffusion equations within obscuring tori around active galactic nuclei .
these solutions demonstrate that infrared radiation pressure can support geometrically thick structures in agn environments subject to certain constraints : the bolometric luminosity must be roughly @xmath0@xmath1 the eddington luminosity ; and the compton optical depth of matter in the equatorial plane should be @xmath2 , with a tolerance of about an order of magnitude up or down .
both of these constraints are at least roughly consistent with observations .
in addition , angular momentum must be redistributed so that the fractional rotational support against gravity rises from the inner edge of the torus to the outer in a manner specific to the detailed shape of the gravitational potential .
this model also predicts that the column densities observed in obscured agn should range from @xmath3 to @xmath4 @xmath5 . |
physics of the bilayer quantum hall ( qh ) system is enormously rich owing to the intralayer and interlayer phase coherence controlled by the interplay between the spin and the layer ( pseudospin ) degrees of freedom@xcite . at the filling factor @xmath2 there arises a unique phase , the spin - ferromagnet and pseudospin - ferromagnet phase , which has well been studied both theoretically and experimentally .
one of the most intriguing phenomena is the josephson - like tunneling between the two layers predicted in refs.@xcite , whose first experimental indication was obtained in ref.@xcite .
other examples are the anomalous behavior of the hall resistance reported in counterflow experiments@xcite and in drag experiments@xcite .
they are triggered by the supercurrent within each layer@xcite .
quite recently , careful experiments @xcite were performed to explore the condition for the tunneling current to be dissipationless .
these phenomena are driven by the goldstone mode describing an interlayer phase coherence .
it exhibits the linear dispersion relation in the zero tunneling - interaction limit ( @xmath1 ) . on the other hand , at @xmath0
the bilayer qh system has three phases , the spin - ferromagnet and pseudospin - singlet phase , the spin - singlet and pseudospin ferromagnet phase , and a canted antiferromagnetic phase@xcite ( abridged as the caf phase ) , depending on the relative strength between the zeeman energy @xmath3 and the tunneling energy @xmath4 .
the pattern of the symmetry breaking is su(4)@xmath5u(1)@xmath6su(2)@xmath6su(2 ) , associated with which there appear four complex goldstone modes@xcite .
a part of them has been studied in refs.@xcite .
we have recently analyzed the full details of these goldstone modes in each phase@xcite .
the caf phase is the most interesting , where the spins are canted coherently and making antiferromagnetic correlations between the two layers .
moreover , one of the goldstone modes becomes gapless and has a linear dispersion relation@xcite as @xmath1 .
it is an urgent and intriguing problem what kind of phase coherence this goldstone mode develops . in this paper , we show that it is the entangled spin - pseudospin phase coherence , and we explore associated phase coherent phenomena .
we employ the grassmannian formalism@xcite , where the basic field is the grassmannian field consisting of two complex - projective ( @xmath7 ) fields .
the cp@xmath8 field emerges when composite bosons undergo bose - einstein condensation@xcite .
the formalism provides us with a clear physical picture of the spin - pseudospin phase coherence in the caf phase .
furthermore , it enables us to analyze nonperturbative phase coherent phenomena , where the phase field @xmath9 is essentially classical and may become very large .
we show that the supercurrent flows within the layer when there is inhomogeneity in @xmath9 .
this is precisely the same as in the @xmath2 bilayer qh system .
indeed , the supercurrent leads to the same formula@xcite of the anomalous hall resistivity for the counterflow and drag geometries as the one at @xmath2 .
what is remarkable is that the total current flowing the bilayer system is a supercurrent carrying solely spins and not charges in the counterflow geometry .
we note that the supercurrent flows both in the balanced and imbalanced systems at @xmath2 but only in imbalanced systems at @xmath0 .
in the bilayer system an electron has two types of indices , the spin index @xmath10 and the layer index @xmath11 .
they can be incorporated into 4 types of isospin index @xmath12 f@xmath13,f@xmath14,b@xmath13,b@xmath14 .
the electron field @xmath15 has four components , and the bilayer system possesses the underlying algebra su(4 ) with the subalgebra @xmath16(2)@xmath17(2 ) .
we denote the three generators of the @xmath16(2 ) by @xmath18 , and those of @xmath19(2 ) by @xmath20 .
there are remaining nine generators @xmath21 , which are the generators of the r - spin operators .
their explicit forms are given in appendix d in ref.@xcite .
all the physical operators required for the description of the system are constructed as the bilinear combinations of @xmath22 and @xmath23 .
they are 16 density operators @xmath24 , @xmath25 , @xmath26 , and @xmath27 , where @xmath28 describes the total spin , @xmath29 measures the electron - density difference between the two layers .
the operator @xmath30 transforms as a spin under @xmath31 and as a pseudospin under @xmath32 .
it is @xmath30 that plays the key role in the entangled spin - pseudospin phase coherence in the caf phase .
the kinetic hamiltonian is quenched , since the kinetic energy is common to all states in the lowest landau level ( lll ) .
the coulomb hamiltonian is decomposed into the su(4)-invariant term @xmath33 and the su(4)-noninvariant term @xmath34 .
the additional potential terms are the zeeman , tunneling , and bias terms , @xmath35 , where @xmath36 is the bias voltage which controls the density imbalance between the two layers .
the total hamiltonian is @xmath37 .
we project the density operators to the lll .
what are observed experimentally are the classical densities , which are expectation values such as @xmath38 , where @xmath39 represents a generic state in the lll .
we may set @xmath40 , @xmath41 , @xmath42 , and @xmath43 for the study of goldstone modes , where @xmath44 is the density of states . taking the nontrivial lowest order terms in the derivative expansion
, we obtain the su(4 ) effective hamiltonian density@xcite @xmath45 , \label{su4hamil}\end{aligned}\ ] ] where @xmath46 with @xmath47 and @xmath48 the intralayer and interlayer stiffness , @xmath49 the capacitance energy , @xmath50 the exchange coulomb energy due to @xmath34 : their explicit formulas are given in appendix a in ref.@xcite .
this effective hamiltonian is valid at @xmath51 .
the ground state is obtained by minimizing the effective hamiltonian ( [ su4hamil ] ) for homogeneous configurations of the classical densities .
the order parameters are the classical densities for the ground state .
they are explicitly given in ref.@xcite for the @xmath0 system .
in the limit @xmath1 , they read @xmath52 and all others being zero .
here , @xmath53 is the imbalance parameter with @xmath54 being the electron density in the front ( back ) layer .
both the spin and the pseudospin are polarized into the @xmath55-axis in this limit .
we have analyzed the excitations around the classical ground state@xcite .
there emerge four complex goldstone modes associated with the spontaneous symmetry breaking su(4)@xmath5u(1)@xmath6su(2)@xmath56su(2 ) . when @xmath57 and @xmath58 , the su(4 ) symmetry is exact and all of them are gapless , but they get gapped by these interactions .
we are interested in the limit @xmath1 since we expect the enhancement of the interlayer phase coherence just as in the @xmath59 system .
we have already shown that there exists one gapless goldstone mode with a linear dispersion relation in a perturbation theory@xcite . in this paper
we employ the grassmannian formalism@xcite to make the physical picture of this goldstone mode and its phase coherence clearer , and to construct a nonperturbative theory in terms of the density difference field @xmath60 and its conjugate phase field @xmath61 .
the grassmannian field @xmath62 consists of two @xmath7 fields @xmath63 and @xmath64 at @xmath0 , since there are two electrons per one landau site . due to the pauli exclusion principle
they should be orthogonal one to another .
hence , we require @xmath65 with @xmath66 . using a set of two @xmath7 fields subject to this normalization condition
we introduce a @xmath67 matrix field , the grassmannian field given by @xmath68 obeying @xmath69 .
the dimensionless su(4 ) isospin densities are given by @xmath70 = \frac{1}{2}\sum_{i=1}^{2}\boldsymbol{n}_{i}^{\dagger}\tau_{a}^{\text{spin}}\boldsymbol{n}_{i } , \notag \\
\mathcal{p}_{a}(\boldsymbol{x } ) & = \frac{1}{2}\text{tr}\left [ z^{\dagger } \tau_{a}^{\text{ppin}}z\right ] = \frac{1}{2}\sum_{i=1}^{2 } \boldsymbol{n}_{i}^{\dagger}\tau_{a}^{\text{ppin}}\boldsymbol{n}_{i } , \label{su4isospin1 } \\
\mathcal{r}_{ab}(\boldsymbol{x } ) & = \frac{1}{2}\text{tr}\left [ z^{\dagger } \tau_{a}^{\text{spin}}\tau_{b}^{\text{ppin}}z\right ] = \frac{1}{2 } \sum_{i=1}^{2}\boldsymbol{n}_{i}^{\dagger}\tau_{a}^{\text{spin } } \tau_{b}^{\text{ppin}}\boldsymbol{n}_{i } , \notag\end{aligned}\ ] ] where @xmath71 consists of the basis @xmath72 .
it is a straightforward task to carry out the perturbative analysis of the effective hamiltonian ( [ su4hamil ] ) in terms of the grassmannian field and obtain the same results as given in ref.@xcite .
we concentrate solely on the gapless mode in the limit @xmath1 .
we parametrize the @xmath7 fields as @xmath73 for @xmath74 , and @xmath75 for @xmath76 .
the isospin density fields are expressed in terms of @xmath60 and @xmath61 , @xmath77 with all others being zero .
the ground - state expectation values are @xmath78 , @xmath79 , with which the order parameters are reproduced from ( [ isospin ] ) .
it is notable that the fluctuations of the phase field @xmath61 affect both spin and pseudospin components of the @xmath80-spin .
this is very different from the spin wave in the monolayer qh system or the pseudospin wave in the bilayer qh system at @xmath2 .
hence we call it the entangled spin - pseudospin phase field @xmath61 . by substituting ( [ isospin ] ) into ( [ su4hamil ] ) ,
apart from irrelevant constant terms the resulting effective hamiltonian is @xmath81 where @xmath82 , @xmath83 , and @xmath84 is the capacitance parameter at @xmath2 .
the effective hamiltonian is correct up to @xmath85 as @xmath1 .
when we require the equal - time commutation relation , @xmath86 = i\delta ( \boldsymbol{x}-\boldsymbol{y } ) , \label{cr}\ ] ] the hamiltonian ( [ effechamil ] ) is second quantized , and it has the linear dispersion relation , @xmath87 this agrees with eq.(136 ) of ref.@xcite . it should be emphasized that the effective hamiltonian ( [ effechamil ] ) is valid in all orders of the phase field @xmath61 .
it may be regarded as a classical hamiltonian as well , where ( [ cr ] ) should be replaced with the corresponding poisson bracket .
the effective hamiltonian ( [ effechamil ] ) for @xmath61 and @xmath60 reminds us of the one that governs the josephson - like effect at @xmath2 .
the main difference is the absence of the tunneling term , as implies that there exists no josephson - like tunneling . nevertheless , the supercurrent is present within the layer , which is our main issue .
-axis in the counterflow geometry . ( a ) all spins are polarized into the positive z axis due to the zeeman effect at @xmath88 .
no spin current flows .
( b ) all electrons belong to the front layer at @xmath89 .
no spin current flows .
( c ) in the caf phase for @xmath90 , some up - spin electrons are moved from the back layer to the front layer by flipping spins .
there appears a goldstone mode associated with this charge - spin transfer .
the interlayer phase difference @xmath91 is created by feeding a charge current @xmath92 to the front layer , which also drives the spin current .
electrons flow in each layer as indicated by the dotted horizontal arrows , and the spin current flows as indicated by the solid horizontal arrow .
( d ) in the caf phase for @xmath93 , similar phenomena occur but the direction of the spin current becomes opposite.,scaledwidth=48.0% ] by using the hamiltonian ( [ effechamil ] ) and the commutation relation , we obtain the equations of motion , @xmath94
we now study the electric supercurrent carried by the gapless mode @xmath61 .
the electron densities are @xmath95 on each layer . taking the time derivative and using we find @xmath96 the time derivative of the charge
is associated with the current via the continuity equation , @xmath97 .
we thus identify @xmath98constant , where @xmath99 consequently , the current @xmath100 flows when there exists inhomogeneity in the phase @xmath61 .
it is a supercurrent because the coherent mode exhibits a linear dispersion relation .
it is intriguing that the current does not flow in the balanced system since @xmath101 at @xmath102 .
let us inject the current @xmath92 into the @xmath103 direction of the bilayer sample , and assume the system to be homogeneous in the @xmath104 direction ( fig.[spincurrentfigure ] ) .
it creates the electric field @xmath105 so that the hall current flows into the @xmath103-direction .
a bilayer system consists of the two layers and the volume between them .
the coulomb energy in the volume is minimized@xcite by the condition @xmath106 .
we thus impose @xmath107 .
the current is the sum of the hall current and the supercurrent , @xmath108 with @xmath109 the von klitzing constant .
we obtain the standard hall resistance when @xmath110 .
namely , the emergence of the supercurrent ( @xmath111 ) is detected if the hall resistance becomes anomalous .
we apply these formulas to analyze the counterflow and drag experiments since they occur without tunneling . in the counterflow experiment ,
the current @xmath92 is injected to the front layer and extracted from the back layer at the same edge . since there is no tunneling we have @xmath112 .
hence , it follows from ( [ totalcurrent ] ) that @xmath113 , or @xmath114 all the input current is carried by the supercurrent , @xmath115 .
it generates such an inhomogeneous phase field that @xmath116 . on the other hand , in the drag experiment , since the interlayer coherent tunneling
is absent , no current flows on the back layer , or @xmath117 .
hence , it follows from ( [ totalcurrent ] ) that @xmath118 , or @xmath119 a part of the input current is carried by the supercurrent , @xmath120 .
the standard hall resistance is given by @xmath121 at @xmath0 .
we thus predict the anomalous hall resistance ( [ counterflowanomalous ] ) and ( [ draganomalous ] ) in the caf phase at @xmath0 by carrying out similar experiments@xcite due to kellogg _ et al_. and tutuc _
et al_. in imbalanced configuration ( @xmath122 ) .
the phase field @xmath61 describes the entangled spin - pseudospin coherence according to the basic formula ( [ isospin ] ) in the caf phase .
the spin density in each layer is defined by @xmath123 , where @xmath124 for @xmath125 and @xmath126 for @xmath127 .
we note the relation @xmath128 up to @xmath129 , we obtain @xmath130 , and @xmath131\partial_{x}^{2}\vartheta ( \boldsymbol{x } ) , \\
\partial_{t}\rho_{\text{f}\uparrow}^{\text{spin } } & = \partial_{t } \rho_{\text{b}\downarrow}^{\text{spin}}=-\frac{j_{\vartheta}}{4}[1-\text{sgn}(\sigma_{0})]\partial_{x}^{2}\vartheta ( \boldsymbol{x } ) .
\label{rhotderivative}\end{aligned}\ ] ] the time derivative of the spin is associated with the spin current via the continuity equation , @xmath132 for each @xmath133 .
we thus identify @xmath134 the spin current @xmath135 flows along the @xmath103-axis , when there exists an inhomogeneous phase difference @xmath61 . in the counterflow experiment ,
the total charge current along the @xmath103-axis is zero , @xmath136 .
consequently , the input current generates a pure spin current , @xmath137 this current is dissipationless since the dispersion relation is linear .
it is appropriate to call it a spin supercurrent .
it is intriguing that the spin current flows in the opposite directions for @xmath138 and @xmath139 , as illustrated in fig.[spincurrentfigure ] .
a comment is in order : the spin current only flows within the sample , since spins are scattered in the resistor @xmath80 and spin directions become random outside the sample .
we have explored the entangled spin - pseudospin phase coherence in the caf phase , governed by the goldstone mode @xmath61 describing the @xmath80-spin according to the formula ( [ isospin ] ) .
we have predicted anomalous hall resistivity in the counterflow and drag experiments in the imbalanced regime ( @xmath122 ) at @xmath0 .
in particular , there flows a spin supercurrent in the counterflow geometry .
this research was supported in part by jsps research fellowships for young scientists , and a grant - in - aid for scientific research from the ministry of education , culture , sports , science and technology ( mext ) of japan ( no . 21540254 ) .
l. tiemann , w. dietsche , m. hauser , and k. von klitzing , new .
j. phys .
* 10 * , 045018 ( 2008 ) ; l. tiemann , y. yoon , w. dietsche , k. von klitzing , and w. wegscheider , phys .
b * 80 * , 165120 ( 2009 ) ; y. yoon , l. tiemann , s. schmult , w. dietsche , k. von klitzing , and w. wegscheider phys . rev .
* 104 * , 116802 ( 2010 ) . v. pellegrini , a. pinczuk , b. s. dennis , a. s. plaut , l. n. pfeiffer , and k. w. west , phys .
lett . * 78 * , 310 ( 1997 ) ; v. pellegrini , a. pinczuk , b. s. dennis , a. s. plaut , l. n. pfeiffer , and k. w. west , science * 281 * , 779 ( 1998 ) .
m. f. yang and m. c. chang , phys .
b * 61 * , r2429 ( 2000 ) ; e. demler and s. das sarma , phys . rev . lett . * 82 * , 3895 ( 1999 ) ; k. yang , phys . rev .
b * 60 * , 15578 ( 1999 ) ; e. demler , e. h. kim , and s. das sarma , phys.rev.b * 61 * , r10567 ( 2000 ) ; a. lopatnikova , s. h. simon , and e. demler phys .
b * 70 * , 115326 ( 2004 ) . | the spin and layer ( pseudospin ) degrees of freedom are entangled coherently in the canted antiferromagnetic phase of the bilayer quantum hall system at the filling factor @xmath0 .
there emerges a complex goldstone mode describing such a combined degree of freedom . in the zero tunneling - interaction limit ( @xmath1 )
, its phase field provokes a supercurrent carrying both spin and charge within each layer .
the hall resistance is predicted to become anomalous precisely as in the @xmath2 bilayer system in the counterflow and drag experiments .
furthermore , it is shown that the total current flowing in the bilayer system is a supercurrent carrying solely spins in the counterflow geometry .
it is intriguing that all these phenomena occur only in imbalanced bilayer systems . |
this work is supported by national natural science foundation of china under grant nos . 10235030 , 10235020 , 10175093 , 10175089 and major state basic research development program under contract , no .
g20000774 . `
fig.1 ` : : the time evolution of binding energies and root - mean - square charge radii for @xmath46zr and @xmath51pb calculated by the imqmd - ii model with parameter set of iq1 . `
fig.2 ` : : the static coulomb barriers of @xmath49ca+@xmath48ca , @xmath49ca+@xmath46zr , @xmath50o+@xmath50o and @xmath50o+@xmath51pb .
the solid and dashed curves denote the results of imqmd - ii with parameters of iq1 and those of proximity potential , respectively . ` fig.3 ` : : the static coulomb barriers of @xmath60i+@xmath60i , @xmath61cr+@xmath51pb , @xmath62s+@xmath63th and @xmath64c+@xmath65fm .
the solid and dashed curves denote the results of imqmd - ii with parameters of iq1 and those of proximity potential , respectively . `
fig.4 ` : : the probability of fusion reaction @xmath69 as a function of impact parameter for @xmath49ca+@xmath87ca at @xmath88 .
` fig.5 ` : : the probability of fusion reaction @xmath69 for @xmath49ca+@xmath87ca at @xmath71 . `
fig.6 ` : : the fusion excitation functions of a)@xmath49ca+@xmath48ti @xcite , b)@xmath75ti+@xmath75ti@xcite , c ) and d ) @xmath49ca+@xmath2zrtimm98 , e ) and f ) @xmath76s+@xmath77y@xcite , and g ) and h)@xmath78si,@xmath79cl+@xmath80zr@xcite at energies near the coulomb barrier .
the open circles denote the experimental data , and the filled triangles denote the results of imqmd - ii with parameters of iq1 .
the crosses denote the results of one - dimension wkb approximation ` fig.7 ` : : the fusion excitation function for @xmath49ca+@xmath48ca .
the stars denote the experimental data .
the solid circles and triangles denote the results without and with the surface - symmetry term taken into account , respectively . `
fig.8 ` : : the fusion excitation function of @xmath44sn+@xmath83ni@xcite .
the solid circles denote the experimental data .
the filled triangles denote the results of imqmd - ii with parameters of iq1 .
the dashed and solid curves denote the results of coupled - channel calculations including inelastic excitation ( ie ) of the projectile and target and ie plus neutron transfer ( n&ie ) , respectively . `
fig.9 ` : : the charge distribution of fragments for ( a ) ca+ca at @xmath89 and ( b ) xe+sn @xmath86 .
the open circles denote the experimental data , the stars and solid circles denote the results of imqmd - ii model without and with the @xmath90 term taken into account , respectively .
the calculation results are averaged over the impact parameter range of @xmath91 . `
table.1 ` : : parameters used in the qmd model and the corresponding values obtained from various skyrme parametrizations .
` table.2 ` : : the parameters of iq1 . `
table.3 ` : : the binding energies and root - mean - square charge radii of a series ground state nuclei calculated by the imqmd - ii model with iq1 interaction . | the improved quantum molecular dynamics model is further developed by introducing new parameters in interaction potential energy functional based on skyrme interaction of skm@xmath0 and sly series .
the properties of ground states of selected nuclei can be reproduced very well .
the coulomb barriers for a series of reaction systems are studied and compared with the results of the proximity potential .
the fusion excitation functions for a series of fusion reactions are calculated and the results are in good agreement with experimental data .
pacs numbers : 25.70.-z , 24.10.-i * i. introduction * recently , the study of the mechanism for heavy - ion fusion reactions at energies near the coulomb barrier , especially , the mechanism for the enhancement of fusion cross sections for neutron - rich systems has attracted a lot of attention .
the knowledge of the mechanism of the enhancement of fusion cross sections for heavy and neutron - rich systems is useful in the synthesis of superheavy elements . in heavy - ion fusion reactions ,
the excitation and deformation of projectile and target , neck formation and nucleon transfer strongly influence the dynamics of fusion processes .
these effects are more pronounced at near and below the fusion barrier . for heavy systems , in order to consider these effects a very large number of degrees of freedom of motion is involved and the situation becomes very complicated .
thus one will meet great difficulty by macroscopic dynamics models@xcite in which only few degrees of freedom of motion are included .
the difficulty is also encountered by the fusion coupled channel model@xcite for it is quite difficult even impossible to include such a large number of possible channels in practical calculations .
therefore , it is highly requisite to develop a microscopic dynamical model suitable for studying heavy ion fusion reactions by which one can consistently take account of the dynamical deformation , particle transfer , isospin and mass asymmetry effects , etc . in our previous work@xcite , an improved quantum molecular dynamics ( imqmd ) model was proposed .
main improvements in the imqmd model are as follows : the surface and surface symmetry energy terms are introduced in the potential energy part ; a system size dependent wave packet width is introduced in order to consider the evolution of the wave packet width ; an approximate treatment of anti - symmetrization , namely , a phase space occupation constraint is adopted @xcite . with this model we have studied the dynamical evolution of the fusion barrier as well as the development of the neck in fusion reactions of @xmath1ca+@xmath2zr . however , more tests of our model are needed .
before that one of the most urgent problem has to be solved is to increase the time of keeping an individual nucleus to be stable and in a good shape ( close to ground state shape ) without emission of nucleons in order to study the dynamical process for fusion reactions of heavy nuclei . as is well know that the time scale for the formation process of a compound system in a fusion reaction of heavy nuclei is about thousands fm / c , and also the time scale of quasi - fission which reduces the fusion probability substantially in fusion processes of heavy nuclei is about several thousands fm / c or longer . to meet this requirement , in this paper we develop an updated version of improved quantum molecular dynamics model named imqmd - ii based on our previously work@xcite . in this paper
we mainly devote to modify the parameters of the potential energy functional in the model . as is well known that the parameters of the potential energy functional in the qmd model are obtained based on skyrme forces .
there are quite a lot of new versions of skyrme forces having been developed following the development of the knowledge of nuclear equation of state .
the newly developed skyrme forces such as the @xmath3 series are designed to study the properties of nuclei away from the @xmath4-stability line in addition to nuclei along the @xmath4-stability line .
it has been shown that these modern skyrme forces can describe the properties of the nuclei away from @xmath4-stability line better than the old skyrme forces@xcite
. therefore it seems to us to be worthwhile to try new parameters for the imqmd model based on the modern parametrizations of skyrme forces such as the @xmath3 series as well as the other popular versions .
the paper is organized as follows . in sec.ii , we briefly introduce the formalism of the imqmd - ii model and the improvements . the applications of this model to the fusion reactions at energies near the coulomb barrier are reported in sec.iii .
finally , the summary and discussion are given in sec.iv .
* ii .
further development of the improved qmd model * in this section we introduce the updated imqmd model(imqmd - ii ) in more details .
first , a brief introduction to the imqmd model is presented .
then , we will give the new development of imqmd .
finally , the calculation results about the properties of selected nuclei with the imqmd - ii model are given .
* a. brief introduction of the imqmd model * for readers convenience , let us first briefly introduce the imqmd model . in the imqmd model , the same as in the original qmd model@xcite , each nucleon is represented by a coherent state of a gaussian wave packet @xmath5,\ ] ] where , @xmath6 , are the centers of i - th wave packet in the coordinate and momentum space , respectively .
@xmath7 represents the spatial spread of the wave packet .
the total n - body wave function is assumed to be the direct product of these coherent states . through a wigner transformation ,
the one - body phase space distribution function for n - distinguishable particles is given by : @xmath8.\ ] ] for identical fermions , the effects of the pauli principle were discussed in a broader context by feldmeier and schnack @xcite . the approximate treatment of anti - symmetrization is adopted in the imqmd model by means of the phase space occupation constraint method @xcite .
the density and momentum distribution functions of a system read @xmath9 @xmath10 respectively , where the sum runs over all particles in the system .
@xmath11 and @xmath12 are the density and momentum distributions of nucleon i : @xmath13 , \label{5}\ ] ] @xmath14,\ ] ] where @xmath15 and @xmath16 are the widths of wave packets in coordinate and momentum space , respectively , and they satisfy the minimum uncertainty relation : @xmath17 the propagation of nucleons under the self - consistently generated mean field is governed by hamiltonian equations of motion : @xmath18 the hamiltonian h consists of the kinetic energy and the effective interaction potential energy : @xmath19 @xmath20 the effective interaction potential energy includes the nuclear local interaction potential energy and the coulomb interaction potential energy : @xmath21and@xmath22where @xmath23 is potential energy density .
the potential energy density @xmath23 in the imqmd model reads @xmath24where @xmath25 .
the first three terms in above expression can be obtained from the potential energy functional of skyrme forces directly .
the fifth term is the symmetry potential energy part where both the bulk and the surface symmetry energy are included .
in addition , we introduce an extra small correction term @xmath26 ( named tau term ) in the potential energy functional . inserting expression ( 13 ) into ( 12 ) , we obtain the local interaction potential energy omitting self - energies : @xmath27where @xmath28 , \label{15}\]]@xmath29and @xmath30 for protons and @xmath31 for neutrons .
one should notice that the third term in ( 14 ) comes from both surface term and the correction to the second term of ( 13)(see@xcite ) , and thus @xmath32 is actually treated as a parameter in the model .
the coulomb energy can be written as a sum of the direct and the exchange contribution , and the latter being taken into account in the slater approximation @xcite : @xmath33where @xmath34 is the density distribution of protons of the system .
the collision term and phase space occupation constraint can also readjust the momenta , but the former plays a very small role in low energy heavy ion collisions and the latter only happens occasionally . the phase space occupation constraint method@xcite and the system - size - dependent wave - packet width
are adopted as that did in the previous version of imqmd@xcite .
* b. the new development in imqmd(imqmd - ii ) model * the new development of the imqmd model mainly are reconsidering the parameters in the potential energy functional ( see expression ( 13 ) ) . in the expression
( 13 ) , the first three terms can be obtained from the potential energy functional of a standard skyrme interaction directly .
the parameters @xmath35 , @xmath4 , @xmath36 and @xmath37 can be related to the parameters of skyrme interactions by @xmath38@xmath39 and @xmath40where the parameter @xmath36 is taken the same value as in the skyrme interaction .
the linear density dependence of the symmetry energy term is taken and the parameter is fixed by the symmetry energy coefficient .
in table.1 we list the @xmath35 , @xmath4 , and @xmath36 parameters used in the qmd modelai91 and the corresponding values obtained from various versions of skyrme interaction@xcite . from the table one can find the parameters of qmd(hard ) is close to that of siii , and the soft one is close to skp . [
cols="<,<,<,^,<",options="header " , ] @xmath41 in table 3 we list the binding energies and the root - mean - square charge radii of @xmath42pb , @xmath43ce , @xmath44sn , @xmath45sn , @xmath46zr , @xmath47ni , @xmath48ca , @xmath49ca , and @xmath50o calculated by the imqmd - ii model with parameter set iq1 .
the experimental data are also listed for comparison and it is shown that the calculated results are in good agreement with experimental data . in fig.1 , we present the time evolution of binding energies and root - mean - square charge radii for @xmath46zr and @xmath51pb calculated by the imqmd - ii model with iq1 parameters .
one can see that their binding energies , and root mean square charge radii remain constants with a very small fluctuation and the bound nuclei evolve stably without spurious emission for a period of time of about 3000fm / c , which is essential for applications to fusion reactions of heavy nuclei as is discussed in the introduction .
* iii . applications to fusion reactions near the coulomb barrier * in this section , we show the calculation results of coulomb barriers and fusion excitation functions for a series of fusion systems by means of the imqmd - ii model with parameters of iq1 .
* a. the coulomb barrier * the interaction potential @xmath52 is defined by @xmath53here @xmath54 is the distance between the centers of mass of projectile and target .
@xmath55 is the total energy of whole system , while @xmath56 and @xmath57 are the energies of projectile(like ) and target ( like ) part , respectively . for kinetic energies ,
the thomas - fermi approximation is adopted as mentioned in ref.@xcite . by using the imqmd model , both
the static and dynamic coulomb barrier can be calculated . for the static coulomb barrier case ,
the static density distribution which is the same as the initial density distribution of projectile and target is adopted , while for the dynamic coulomb barrier case the density distribution of the system changes dynamically due to the interaction between the reaction partners .
we show the static coulomb barriers calculated by the imqmd - ii model with parameter set of iq1 for @xmath49ca+@xmath48ca , @xmath49ca+@xmath46zr , @xmath50o+@xmath50o and @xmath50o+@xmath51pb in fig.2 .
the results calculated from proximity potential @xcite are also shown in the figures .
one can see that the coulomb barriers calculated with imqmd - ii are in good agreement with those from proximity potential@xcite .
@xmath58 @xmath59 in addition , the effects of the mass asymmetry of projectile and target on the static coulomb barrier are studied through calculating the static coulomb barriers for @xmath60i+@xmath60i , @xmath61cr+@xmath51pb , @xmath62s+@xmath63th and @xmath64c+@xmath65fm fusion systems which can form the same compound nucleus @xmath66sg . in fig.3 ,
the solid curves denote the static coulomb barriers calculated by imqmd - ii with parameters iq1 and the dashed curves denote the results from the proximity potential . from fig.3
one can get two points : the first one is that the results from imqmd - ii are in good agreement with those from proximity potential when two nuclei do not overlap too much in space .
the proximity potential may not be able to give an accurate result at the overlapping region where the imqmd model is applicable .
clearly , the results for the overlapping region are more interesting , especially for the cases of heavy systems .
the second one is that with the increase of the mass asymmetry of projectile and target ( from @xmath60i+@xmath60i to @xmath64c+@xmath65fm ) the height of the coulomb barrier decreases gradually and the capture probability should be enhanced consequently .
@xcite . * b. fusion cross sections * in the imqmd model , we first create certain reaction events for each incident energy @xmath67 and impact parameter @xmath68 ( in tis work the number is 100 ) and then counting the number of fusion events , we obtain the probability of fusion reactions , @xmath69 , then the cross section is calculated by using the expression@xcite : @xmath70at a certain incident energy , the probability of fusion reactions , @xmath69 , decreases with the increase of impact parameter as shown in fig.4 .
this is because the interaction between two nuclei decreases gradually from central collisions to peripheral collisions , and following it the probability for fusion reactions decreases and that for elastic scattering processes increases .
in addition , the @xmath69 decreases quickly with the decrease of incident energies within the energy range interested in this work .
fig.5 presents the evolution of the fusion probability with impact parameter at @xmath71 , respectively .
one can see from the figure that the fusion probability falls when the incident energy decreases from 60mev to 52mev and for energies below the coulomb barrier , fusion events generally only occur at central collisions .
@xmath72@xmath73@xmath74 in fig.6 , we show the fusion excitation functions for @xmath49ca+@xmath48ti @xcite , @xmath75ti+@xmath75ti@xcite , @xmath49ca+@xmath2zrtimm98 , @xmath76s+@xmath77y@xcite , and @xmath78si,@xmath79cl+@xmath80zr @xcite at energies near the coulomb barrier , and the experimental data are also presented for comparison . in the figure
the triangles denote the results of imqmd - ii with parameters of iq1 and the circles denote the experimental data and crosses denote the results of one - dimension wkb approximation@xcite .
one can see from the figure that calculation results of imqmd - ii for the fusion excitation functions are generally in good agreement with experimental data , which implies that our model is quite reasonable .
now let us study the influence of the surface - symmetry energy term on the fusion cross sections , we show the calculation results of the fusion excitation functions of @xmath49ca+@xmath48ca @xcite at energies near the coulomb barrier without and with the surface - symmetry energy term taken into account in fig.7 .
one can find from the figure that at incident energies above the coulomb barrier the fusion cross sections calculated under two cases are approximately equal , while at incident energies below the barrier the difference between two cases become obvious , and for this case the fusion cross sections without surface - symmetry energy term taken into account are obviously larger than the experimental data and those with surface - symmetry energy term taken into account can reproduce the experimental data well .
we have found that the n / z ratio at neck region is enhanced at the early stage for neutron - rich fusion process@xcite , which is driven by the symmetry potential .
it lowers the fusion barrier and consequently enhances the fusion cross section .
if only the bulk term is taken in account , the effect of the symmetry energy term becomes too strong at surface region and therefore a surface - symmetry energy term should be taken into account in order to reduce the effect of the bulk symmetry energy term at surface region .
it is especially important for having a correct neck dynamics in neutron - rich nuclear fusion reactions .
@xmath81@xmath82 for further testing the reliability of imqmd , we calculate the excitation function for fusion reactions of the neutron - rich radioactive beam @xmath44sn bombarding on the neutron - rich target @xmath83ni at energies near the coulomb barrier , which was recently measured @xcite .
fig.8 shows the comparison of our calculation results and the experimental data as well as the results from coupled - channel calculations @xcite .
the solid circles denotes the experimental data , the triangles denote the results of imqmd - ii with parameters of iq1 .
the dashed and solid curves denote the results of coupled - channel calculations with including inelastic excitation ( ie ) of the projectile and target and ie plus neutron transfer ( n&ie ) , respectively ( see ref.@xcite and references therein ) . from fig.8 one can find that the coupled - channel calculations significantly under - predicted the sub - barrier fusion cross sections@xcite . this is because with the coupled - channel model it is difficult to consider all degrees of freedom of motion , the number of which is extremely large in sub - barrier fusion reactions for heavier systems , while with the imqmd model , all degrees of freedom of motion are self - consistently included .
thus , this model may possibly provide us with a useful approach to explore the mechanism of the capture process in the synthesis of superheavy elements .
the work concerning this aspect is in progress and the results will be reported in the future work .
@xmath84 as a test we also make applications of our model to describe the charge distributions of fragments in multifragmentation processes . in fig.9
we show the charge distribution of fragments by using the imqmd - ii model with parameter set of iq1 for ca+ca at @xmath85@xcite and xe+sn at @xmath86@xcite .
one can find that the charge distribution of fragments , especially the number of intermediate mass fragments is in good agreement with experimental data .
it is well known that the number of intermediate mass fragments is usually being under - predicted in qmd model calculations .
our results are encouraging and it seems to us that the model can be used for heavy ion collisions at both low energies and intermediate energies . * iv . conclusions and discussion * in this work , we have made further improvements in the imqmd model and proposed a new version , namely , the imqmd - ii model in which a new small correction term and a surface symmetry energy term are introduced in the potential energy functional in addition to the terms adopted in the normal isospin dependent qmd model . a parameter set for the potential energy functional based on skyrme interaction of skm @xmath0 and sly series is introduced . by using the new version of the imqmd model , the ground state properties of a series selected nuclei can be described very well .
the time evolution of individual nuclei can keep stable for about several thousands fm / c which roughly fits the requirement for study of fusion reactions of heavy nuclei .
we have shown that with this model both the coulomb barriers and the fusion excitation functions for a series of fusion systems ( including neutron - rich radioactive beam @xmath44sn + @xmath83ni fusion reactions ) at energies near the barrier can be reproduced very well .
our study shows that the microscopic dynamical model such as the imqmd model has an advantage of taking account of the dynamical effects such as the excitations of projectile and target ( deformation and vibration ) , neck formation , isospin and mass asymmetry , etc . simultaneously and
thus offers a useful way to study fusion reactions of heavy nuclei .
furthermore , we have also shown that this model seems to work well on the charge distribution of fragments in multifragmentation processes . |
the analogy between linear lattices and the atom - field interaction @xcite or ion - laser interactions @xcite has been a fundamental step for the emulation , via classical interactions , of quantum mechanical systems .
this is important due to both pure scientific interest and possible applications to quantum computing . in the latter ,
the properties of classical systems have been used to realize quantum computational operations by quantum - like systems and , in particular , it has been show how a controlled - not gate may be generated in nonhomogeneous optical fibers @xcite . at the fundamental level , _
e.g. _ , it has been possible to the emulate the most basic atom - field interaction , the jaynes - cummings model , theoretically @xcite and experimentally @xcite with arrays of photonic waveguides and , just to give another example , it has been proposed to model non - linear coherent states @xcite in waveguide arrays @xcite ; linear coherent states have also been modelled in linear arrays of photonic waveguides @xcite . in what follows
we will take advantage of the simplest composite array of waveguides , _
i.e. _ , binary arrays , to introduce the most basic mathematical methods used to study photonic lattices . gives an introduction to the two kinds of binary photonic lattices considered ; those with either alternating self - energy or coupling .
then , we proceed to study the propagation of classical electromagnetic fields through binary lattices of infinite size by use of coupled mode theory for lattices with alternating self - energy and by an operator approach to floquet - bloch theory in the case of an array with alternating couplings . in
we use coupled mode theory ( with a twist given by the use of orthogonal polynomials ) to solve a finite binary self - energies lattice in order to exemplify how the presented methods are valid , changing what needs to be changed , in all cases . in this section
we also find the dynamics of a finite binary couplings array by basic matrix theory methods .
curiously , we find that the dynamics for both finite size binary lattices can be written in terms of fibonacci polynomials evaluated at the roots of a fibonacci polynomial which order is related to the size of the system . in we introduce some quantities of interest when studying photon transport in arrays of photonic waveguides .
we show that initial states with a gaussian distribution of amplitudes and linear coherent states , that is , initial states with poisson - like distributions , reconstruct in binary lattices
. finally , we present our conclusions .
a binary photonic super - lattice is composed by the repetition of a primitive unit cell where two different elements are characterized by one parameter .
one type of such binary waveguide arrays , shown in ( a ) , is composed by identical waveguides where nearest neighbour coupling between them alternates between two values ; hereby , we will refer to this type as binary coupling ( bc ) lattice . in the other type , shown in ( b ) , waveguides with two different refraction indexes alternate position and are homogeneously coupled ; which we will call binary index ( bi ) lattice from here on .
the propagation of a classical light field in an infinite bc lattice is ruled by the scalar differential equation set for the field amplitude at the @xmath0th waveguide , @xmath1 where the constant @xmath2 is the refraction index of each waveguide and a total coupling has been defined as @xmath3 with the difference as @xmath4 .
we have used the shorthand notation @xmath5 to express the partial derivative with respect to @xmath6 .
an infinite bi lattice it is given by , @xmath7 e_{j } + g \left ( e_{j+1 } + e_{j-1 } \right ) , \quad j = -\infty , \ldots , \infty,\end{aligned}\ ] ] where a base refraction index has been defined as @xmath8 such that it is halfway between the refraction index at each waveguide , @xmath9 and @xmath10 , _ i.e. _ , @xmath11 , and the tight binding coupling is given by the real constant @xmath12 .
infinite @xcite and semi - infinite @xcite bc and bi lattices , with the addition of a non - linearity to each waveguide , are well known in the non - linear optics community where the existence and stability of diverse types of solitons have been studied along the years . in linear optics ,
it is possible to borrow from condensed matter theory a floquet - bloch result for quasiparticle motion on a chain in order to solve the evolution of a classical field , or a single photon , through these lattices @xcite ; _ e.g. _ , the dispersion relation for a bc lattice , shown in ( a ) , @xmath13,\end{aligned}\ ] ] results of the infinite bc lattice has been used to discuss the existence of two propagation modes with opposite transverse velocities @xcite . for a bi lattice
the dispersion relation is , @xmath14 the band gap structure of this dispersion relation , shown in ( b ) , near the edge of the brillouin zone , @xmath15 , suggests the use of bloch waves close to @xmath16 to simulate one - dimensional dirac equations @xcite . under such a condition ,
a photonic analogue of _ zitterbewegung _ has been theoretically proposed @xcite and experimentally realized @xcite ; more complex approaches of such a classical simulation include atoms in bichromatic optical lattices @xcite . and ( b ) bi lattices . ]
we are going to implement a coupled mode theory method on the infinite bc lattice described by . for starters , let us define the field amplitudes with a rotation proportional to the refractive index function , @xmath17 , _
i.e. _ , @xmath18 ; in this way , we can get rid of the term involving the refractive index function , @xmath19 now , we can define two auxiliary proper modes , say @xmath20 and @xmath21 , such that our differential set is now given by a coupled modes set @xmath22 it is trivial to find a three term recurrence relation for the mode @xmath23 from this coupled set of equations , @xmath24 solved by @xmath25\end{aligned}\ ] ] under the restriction @xmath26 from the expression above , we straightforwardly recover the dispersion relation @xmath27.\end{aligned}\ ] ] notice that we can also write the normal mode components and dispersion relation as , @xmath28 under this treatment we can recover the field amplitude propagation at the @xmath0th waveguide for a field entering the lattice at the @xmath29th waveguide , @xmath30 and we have found all the necessary information to study the propagation of any given initial classical field impinging a bc lattice . it is possible to use an operator approach to find floquet - bloch waves . in order to demonstrate it ,
let us write the differential equation set for an infinite bi lattice in matrix form as @xmath31 with solution @xmath32 the expression is identical to the whole differential set up to a unitary transformation equivalent to a change into a frame rotating at a frequency proportional to the refractive index function @xmath17 and a change of units in terms of the homogeneous coupling @xmath12 .
we used the unitary ladder operators , @xmath33 , defined as @xmath34 where we substituted @xmath35 , _
i.e. _ , the state @xmath36 represents the field localized at the @xmath37th waveguide .
these ladder operators fulfil the commutation relations @xmath38
= -\hat{v } , \qquad \left [ \hat{n},\hat{v}^{\dagger } \right ] = \hat{v}^{\dagger } , \qquad \left [ \hat{v}^{\dagger},\hat{v } \right ] = 0.\end{aligned}\ ] ] by using this operator representation , we can define a phase state as @xmath39 which is the fourier transform of the localized field .
in other words , these operators allow us to do floquet - bloch theory , @xmath40 in this phase state basis , the action of the operator @xmath41 over some useful phase states reduces to : @xmath42 from the last two equations it is possible to infer the dispersion relation , @xmath43 in order to recover the rest of the information given by floquet - bloch theory , let us calculate the evolution of our phase state , @xmath44
\vert \phi \rangle , \\ & = & \cos \omega_{\phi } z \vert \phi \rangle + \rmi \left
( \frac{\sin \omega_{\phi } z}{\omega_{\phi } } \right ) \left ( \beta \vert \phi + \pi \rangle + 2 \cos \phi \vert \phi \rangle \right).\end{aligned}\ ] ] now , if the field started at the @xmath29th waveguide , @xmath45 , one can write the field amplitude at the @xmath0th waveguide as @xmath46 and , by use of @xmath47 , recover the field amplitude evolution from standard floquet - bloch theory @xmath48
in some cases one has to consider the evolution of a quantum field in an array of coupled photonic waveguides ; in a general case , it could be arrays of coupled microring resonators , coupled cavities , or capacitively coupled strip - line resonators .
when such a system is presented it is usually described by a hamiltonian which , in the case of finite bc and bi lattices of size @xmath49 , is given by @xmath50 \left ( \hat{a}_{j } \hat{a}_{j+1}^{\dagger } + \hat{a}_{j}^{\dagger } \hat{a}_{j+1 } \right),\\ \hat{h}_{bi } = \sum_{j=0}^{n } \left ( -1 \right)^{j } \beta ~\hat{a}_{j}^{\dagger } \hat{a}_{j } + \sum_{j=0}^{n-1 } \left ( \hat{a}_{j } \hat{a}_{j+1}^{\dagger } + \hat{a}_{j}^{\dagger } \hat{a}_{j+1 } \right),\end{aligned}\ ] ] where we have moved into an adequate reference frame rotating at a frequency defined by the refractive index function @xmath17 and set units in terms of @xmath51 and @xmath52 , respectively . the operator @xmath53 ( @xmath54 ) creates ( annihilates ) a photon in the @xmath37th waveguide . in the heisenberg picture
, the equations of motion for these systems are @xmath55 these sets with @xmath56 are the finite equivalent , by making the substitutions @xmath57 alongside @xmath58 , of the differential sets ruling the propagation of a classical field through the corresponding binary photonic lattice in and , respectively .
let us start with the finite bc lattice and introduce the transformation @xmath59 such that a general states is defined by @xmath60 and leads to the schrrodinger equation in units of @xmath61 @xmath62 with equations of motion in the heisenberg picture @xmath63 now , let us define the coupled modes as @xmath64 and @xmath65 .
notice that we have used @xmath66 in the argument of the exponential instead of @xmath67 , thus we are looking for a purely imaginary @xmath68 .
this allows us to write an eigenequation in the form , @xmath69 where it is possible to recover a recurrence relation for the coefficients of the normal modes , @xmath70 that is solved by fibonacci polynomials , @xmath71 the last of the recurrence relations gives a boundary condition that translates into the expression @xmath72 that is solved by @xcite @xmath73 and we recover the purely imaginary eigenvalue that we were looking for .
so , the normal modes @xmath74 ( up to a normalization constant ) with eigenvalues @xmath75 , diagonalize the hamiltonian @xmath76 and we have found the dynamics of the system . it is trivial to go back into the original frame .
now , let us consider a finite bc lattice .
heisenberg equations of motion for this system may be written as the matrix differential equation , @xmath77 the solution to this system may be found in a number of ways @xcite and we do it by finding the system , @xmath78 , of the matrix @xmath79 @xcite ; where the eigenvalues matrix @xmath80 is a matrix which diagonal elements are the @xmath49 eigenvalues of the matrix @xmath81 , @xmath82 , and the eigenvector matrix @xmath83 is a matrix which rows , @xmath84 , are the @xmath49 eigenvectors of m. the characteristic polynomial , @xmath85 , of tridiagonal matrix @xmath81 is found via the method of minors @xcite as @xmath86 where @xmath87 is the @xmath2th fibonacci polynomial @xcite , @xmath88 and @xmath89 are @xmath2th morgan - voyce polynomials @xcite .
the roots of @xmath90 are well known @xcite and yield the eigenvalues , @xmath91 these proper values , alongside @xmath92 , deliver recurrence relations fulfilled by eigenvector components , @xmath93 where orthogonal polynomials , @xmath94 , are defined as : @xmath95 thus the hamiltonian is diagonalized , @xmath96 , in terms of delocalized normal modes @xmath97 , giving a time evolution @xmath98 .
for the sake of space and simplicity we will only discuss typical quantities of interest when studying the propagation of a quantum electromagnetic field in an array of photonic waveguide lattices .
it is trivial to show that the time evolution of an arbitrary initial state can be obtained from the evolution of the annihilation operator found in the last section . in order to generalize ,
let us write such an evolution as @xmath99 where it is trivial to construct the matrix @xmath100 from the methods given in the last section .
the most basic and visually appealing quantity of interest is the transport of @xmath29 photons impinging a single - waveguide at time zero , @xmath101 the mean photon number at the @xmath102th waveguide for such an initial state after propagation is given by @xmath103 in the case of single photon input , @xmath104 , the expression for the mean photon number is identical to the normalized intensity at the @xmath102th waveguide for a propagating classical electromagnetic field .
superposition states of single photons are also interesting , @xmath105 and we can easily calculate the probabilities of finding the photon in the @xmath102th waveguide , @xmath106 here , we are interested in two peculiar single - photon distributions .
one that we will call a _
gaussian - like _ input distribution , @xmath107 which could be thought as the result of a gaussian beam whose intensity peak is aligned with the center of the lattice impinging parallel , @xmath108 , or at an angle , @xmath109 , with the propagation axis @xcite .
the other is similar to a quantum coherent state and is produced by a single - photon entering a glauber - fock lattice at the zeroth waveguide @xcite , we will call this a _ poisson - like _ input distribution , @xmath110 in these gaussian- and poisson - like distributions , the maximum probability to find the single - photon is given at the @xmath111th and @xmath112th waveguides ( where @xmath113 has been used to express @xmath6 rounded up ) , in that order , and the variance is given by @xmath114 and @xmath111 , respectively .
both initial states have a _ momentum _ in the direction perpendicular to propagation given by @xmath102 and @xmath115 in the sense that the center of mass for an input with @xmath109 or @xmath116 will drift to the right or left side of the lattice depending on the sign of @xmath102 or @xmath115 .
notice that a gaussian - like distribution will always be symmetric with respect to the center of the lattice , which is not the case for the coherent - like distribution .
propagation of gaussian - like input distributions have been studied in infinite bc and bi lattices . in the former ,
the lattice splits the @xmath108 input in two components propagating in opposite directions @xcite . in the latter ,
bi lattices allow the clasical simulation of the dirac equation when the parameter q is close to the edge of the brillouin zone @xcite . to our knowledge there
exist no record in the literature of propagation of poisson - like input distributions through finite binary lattices .
shows the mean photon probability for two kinds of superposition states impinging a bi and bc lattice , respectively .
it is possible to see that the states reconstruct periodically in both cases through the fidelity function @xmath117 and @xmath118 , in that order , and ( b , d , f , h ) their respective fidelities in a ( a - d ) bc and ( e - h ) bi with @xmath119 lattice of size @xmath120 .
time in units of @xmath121 for bc and @xmath12 for bi lattice . ]
it is also possible to study the time evolution of the center of mass , @xmath122 for gausian- and poisson - like input distributions with complex parameters .
time evolution of the fidelity for gaussian - like distributions heavily fluctuates with values below @xmath123 , see top - right insets in ( b , f ) , and its center of mass localizes at the central waveguide after a long time ; after @xmath124 in the case shown in ( f ) .
only poisson - like distributions partially reconstruct , see top - right insets in ( d , h ) .
of course , well - known results regarding classical simulation of dirac equation @xcite are reproduced ; for instance optical _ zitterbewegun _ by using a gaussian - like distribution , ( f ) .
notice that the center of mass of the coherent - like distribution , ( h ) , wobbles with an almost negligible amplitude compared to that of the gaussian - like distribution , ( f ) , and it appears to reconstruct with low fidelity after the second edge reflection ; top - right inset in ( h ) . and
@xmath125 , in that order , and ( b , d , f , h ) their respective center of mass ( with top - right insets showing the fidelities and top - bottom a longer - time evolution of the center of mass ) in a ( a - d ) bc and ( e - h ) bi with @xmath119 lattice of size @xmath120 .
time in units of @xmath121 for bc and @xmath12 for bi lattice . ] another interesting set of states are product states , @xmath126 where @xmath127 with @xmath128 $ ] and @xmath129 for any @xmath130 .
the evolution of the photon number at the @xmath102th waveguide for product states is given by @xmath131 an example of these states is the two - photon product state given by @xmath132 that gives a mean photon number evolution and two - photon correlation function @xmath133 for the sake of space we will finish this section presenting how to deal with noon states .
a higher order noon state , their mean - photon evolution at the @xmath102th waveguide , and two - photon correlation are given by the expression , @xmath134
we presented a review on mathematical methods used to study infinite and finite photonic lattices and the quantities of interest for quantized electromagnetic fields .
each method is presented in a particular context but any of the methods can be used in all case by making the necessary alterations . alongside this review , we introduced a novel result , up to our knowledge , in the form of the exact spectra and proper modes for bc and bi lattices of size @xmath49 ( bi latices include the single - type lattice when the characteristic parameter is zero ) in terms of the roots of the @xmath135 fibonacci polynomial and the fibonacci polynomials evaluated at these roots , in that order . in order to illustrate our results we chose to focus on the evolution of multiple - waveguide single - photon inputs in the form of
what we called gaussian- and poisson - like distributions impinging odd lattices ; these distributions are related to gaussian beams and to the output from glauber - fock lattices , respectively . due to their spectral decompositions ,
gaussian- and poisson - like distributions with real parameters partially reconstruct quasi - periodically when they impinge a binary super - lattice with their intensity peak aligned with the middle of the lattice .
hmc acknowledges l. moya - rosales involvement in the first stage of this work .
bmrl is grateful to miguel bandres for useful discussion and comments .
99 hoggatt v e and bicknell m 1973 _ fibonacci quart . _ * 11 * 271274 .
dodonov v v and manko v i , invariants and evolution of nonstationary quantum systems , in : proceedings of the lebedev physical institute ( nova science , new york , 1989 ) vol . 183 .
rodrguez - lara b m 2011 _ phys .
a _ * 84 * 053845 . | we present a review on the mathematical methods used to theoretically study classical propagation and quantum transport in arrays of coupled photonic waveguides .
we focus on analysing two types of binary photonic lattices where self - energies or couplings are alternated . for didactic reasons
, we split the analysis in classical propagation and quantum transport but all methods can be implemented , _ mutatis mutandis _ , in any given case . on the classical side , we use coupled mode theory and present an operator approach to floquet - bloch theory in order to study the propagation of a classical electromagnetic field in two particular infinite binary lattices . on the quantum side ,
we study the transport of photons in equivalent finite and infinite binary lattices by couple mode theory and linear algebra methods involving orthogonal polynomials .
curiously the dynamics of finite size binary lattices can be expressed as roots and functions of fibonacci polynomials . |
in this work we consider the numerical analysis of mathematical models arising in subdiffusion .
one simple example is the single - term subdiffusion , for which the governing equation is given by @xmath0 under the initial - boundary conditions @xmath1 where @xmath2 is a bounded domain in @xmath3 which for simplicity we assume to be polygonal , and where @xmath4 is a given function on @xmath2 . here
@xmath5 denotes the caputo fractional derivative of order @xmath6 with respect to @xmath7 , defined by the convolution ( see @xcite ) @xmath8 where @xmath9 is the gamma function . by continuity
we may include the standard heat equation for @xmath10 .
the fractional derivative is often used to describe anomalous diffusion , in which the mean square variance grows sublinearly with time @xmath7 , at a rate slower than that in a gaussian process .
it occurs in many applied disciplines , e.g. , electron transport in xerox copiers , molecule transport in membranes , and thermal diffusion on fractal domains . by the maximum principle
, the solution @xmath11 of the heat equation satisfies the nonnegativity preservation property @xmath12 it is natural to ask to what extent this property extends to numerical approximations , and for piecewise linear finite element methods over regular triangulations this question has attracted some attention .
fujii @xcite showed that for the lumped mass ( lm ) method , with backward euler time - stepping , nonnegativity is preserved when the angles in the triangulation are acute .
thome and wahlbin @xcite showed that for the semidiscrete standard galerkin ( sg ) method , nonnegativity is not preserved for small @xmath13 , but for the lm method , it is preserved if and only if the triangulation is of delaunay type .
later , schatz , thome and wahlbin @xcite analyzed several fully discrete methods and showed that , in particular , these properties carry over to the backward euler method .
recently , thome @xcite and chatzipantelidis , horvth and thome @xcite extended these results to cover also the finite volume element ( fve ) method and further discussed the existence of a threshold @xmath14 , such that nonnegativity is preserved for @xmath15 .
our purpose in this paper is to discuss the nonnegativity preservation property for a class of piecewise linear finite element approximations , including the sg , lm and fve methods described briefly in section [ sec : semidis ] , for the more general initial
boundary value problem @xmath16 where @xmath17 denotes a fractional order differential operator of the form @xmath18 with @xmath19 a positive measure on @xmath20 $ ] .
these more general models , including in addition to multi - term and distributed order models , are reviewed briefly in section [ sec : prelim ] .
there we also introduce the solution operator @xmath21 of and discuss its analytic properties , including asymptotic behavior . in a series of interesting works , luchko @xcite established that the solution @xmath11 of problem also satisfies the property .
numerical methods for the model and their analysis have received considerable interest , cf .
, e.g. , @xcite .
the weak formulation of is to find @xmath22 for @xmath13 such that @xmath23 where @xmath24 and @xmath25 is the inner product in @xmath26 .
the spatially semidiscrete finite element approximation of is based on a family of shape regular quasi - uniform triangulations @xmath27 of @xmath2 into triangles , with @xmath28 the maximal length of the sides of the triangulation @xmath29 , and the associated finite element space @xmath30 of continuous piecewise linear functions over @xmath31 , @xmath32 the spatially semidiscrete approximation of is then to find @xmath33 for @xmath34 such that @xmath35}+ a(u_h(t),\chi)= { 0 } , \quad \forall \chi\in x_h,\ t > 0 , { \quad\text{with}\ } u_h(0)=v_h,\ ] ] where @xmath36 is an approximation of the initial data @xmath4 , and @xmath37 $ ] is a suitable inner product on @xmath30 . with @xmath38
the interior nodes of @xmath29 , and @xmath39 the corresponding nodal basis , we may write @xmath40 setting @xmath41 and @xmath42 , the semidiscrete problem reads @xmath43 here @xmath44 , with @xmath45 $ ] , and @xmath46 , with @xmath47 , denote the mass and stiffness matrices , respectively , and are both symmetric and positive definite . the case of the semidiscrete sg method
was first developed and analyzed by jin et al .
@xcite for the single - term problem on a convex polygonal domain , and then extended to multi - term and distributed - order cases in @xcite , where optimal error estimates with respect to the regularity of initial data @xmath4 were established .
following the analysis in the case of the heat equation , we discuss nonnegativity preservation for the spatially semidiscrete methods in section [ sec : semidis ] .
introducing the solution matrix @xmath48 by writing @xmath49 , nonnegativity preservation may be expressed as @xmath50 , elementwise . in theorem
[ thm : failsg ] we show that if the mass matrix @xmath51 is nondiagonal , which holds for the sg and fve methods , then this can not happen for small @xmath13 .
however for the lm method this happens if and only if @xmath29 is delaunay . in theorem [ theorem : sg ] , we show that if @xmath52 , then there exists @xmath14 such that @xmath53 for all @xmath54 , if @xmath55 . in section [ sec :
fully ] we discuss positivity preservation for fully discrete time stepping methods with a time step @xmath56 , for , based on convolution quadrature generated by the backward euler method , or @xmath57 here @xmath58 denotes the convolution quadrature approximation to @xmath59 ; see section [ sec : fully ] .
this gives the fully discrete solution in the form @xmath60 this method was developed and analyzed in @xcite for the sg approximation , and the convergence of the fully discrete solution to the corresponding spatially semidiscrete one , which is needed below , is shown in the general case in appendix a. we show time discrete analogues of the results in section [ sec : semidis ] , and also , in theorem [ thm : threshold ] , we give an upper bound of the positivity threshold for special triangulations . in the final section 5 we present numerical examples for various domains @xmath61 , with delaunay and non - delaunay triangulations @xmath29 , to illustrate and complement our theoretical results . as a result of the maximum principle , in addition to
, it also follows that the solution operator @xmath21 is a contraction in the maximum - norm , i.e. , @xmath62 .
the question about analogues of this for the numerical methods was also discussed in @xcite for the heat equation , and in appendix b we show that the analogue of this holds in the present case for semidiscrete and fully discrete lm approximations , provided that the stiffness matrix @xmath63 is diagonally dominant , which is not equivalent to @xmath29 being delaunay .
in this section we present preliminary material concerning the subdiffusion model and its analytical properties .
in particular we introduce the solution operator @xmath21 for @xmath34 and the associated scalar kernel function @xmath64 , and study its asymptotic behavior for small and large @xmath7 , which will be needed later . as indicated in section [ sec : intro ] , in addition to the single - term model , several more complex models for subdiffusion have been proposed , with replaced by a weighted linear combination @xmath65 , defined in .
we will specifically study two cases , the discrete case ( denoted by case i below ) , with point masses @xmath66 , where @xmath67 , and the distributed case ( denoted by case ii below ) , in which @xmath68 is completely continuous , i.e. , @xmath69 , with @xmath70 smooth and nonnegative .
for the characteristic function of @xmath71 , the complex function @xmath72 , we then have @xmath73 in the discrete case , the constants @xmath74 are positive , with @xmath75 , which ensures the well - posedness of the problem , cf .
it reduces to the single - term subdiffusion model when @xmath76 . in the distributed case ,
the nonnegative weight function @xmath77 is assumed to be hlder continuous , with @xmath78 .
these conditions are sufficient for the positivity of the solution and are different from the ones used in the existing literature , cf .
@xcite , the general model was proposed in order to extend the flexibility of the single - term model , in that the underlying stochastic process may contain a host of different hurst exponents , cf .
@xcite . for the convenience of the reader
we prove the positivity preservation property for problem , using arguments in @xcite .
we first show an extremal principle for the fractional differential operator @xmath79 .
[ lem : ext ] assume @xmath80\cap c^1(0,t]$ ] , and let @xmath81 attain its minimum at @xmath82 $ ]
. then @xmath83 . setting @xmath84 , we have by integration by parts , since @xmath85 as @xmath86 , @xmath87 since @xmath88 we thus have @xmath89 , and hence @xmath90
. the solution operator @xmath21 for is nonnegative , i.e. , @xmath91 for @xmath92 , if @xmath93 .
assume that the minimum of @xmath94 is negative and achieved at @xmath95 $ ] .
with @xmath96 we set @xmath97 .
then , for @xmath98 or @xmath99 , @xmath100 hence @xmath101 can not attain its minimal value at @xmath99 of for @xmath98 . now let the minimum of @xmath102 be taken at @xmath103 $ ] .
then by lemma [ lem : ext ] , @xmath104 , and clearly @xmath105 . noting the identity @xmath106 , we find @xmath107 contradicting the fact that @xmath11 is a solution of .
next we derive an expression for the solution operator @xmath21 for problem by means of laplace transformation , denoting the laplace transform of @xmath108 by @xmath109 .
recall that , @xmath110 and consequently @xmath111 after laplace transformation of , with @xmath112 , the negative laplacian with a zero dirichlet boundary condition , and the initial condition @xmath113 , we therefore obtain @xmath114 or @xmath115 where @xmath116 is the identity operator .
hence we find for the solution operator , defined by @xmath94 , @xmath117 where @xmath118 .
letting @xmath119 be the eigenvalues in increasing order and @xmath120 the corresponding @xmath26 orthonormal eigenfunctions of the operator @xmath121 , we may also write @xmath122 where @xmath123 the scalar kernel function @xmath124 solves the fractional order initial - value problem @xmath125 since @xmath126 is analytic in the complex plane , cut along the negative real axis , we obtain , after deforming the contour @xmath127 in , and noting that , for small @xmath128 , @xmath129 and hence we find that @xmath130 where we have used @xmath131 since @xmath132 in is positive , @xmath64 is completely monotone , and , in particular , monotone for @xmath34 .
we now study the asymptotic behavior of the function @xmath133 defined in .
our main tool will be a special case of the karamata - feller tauberian theorem ( see ( * ? ? ? * theorems 24 , pp .
445466 ) ) , which we state as the following lemma . recall that a positive function @xmath134 defined on @xmath135 varies slowly at infinity if for every fixed @xmath136 , @xmath137 as @xmath138 .
[ thm : karamata - feller ] let @xmath11 be a monotone function with its laplace transform @xmath139 defined for @xmath140 , @xmath134 be slowly varying at infinity , and @xmath141 .
then @xmath142 the statement is also valid when the roles of the origin and infinity are interchanged .
we can now determine the asymptotic behavior for the kernel function @xmath143 .
[ lem : cpl_mt ] let @xmath144)$ ] and @xmath145 . then @xmath146 and @xmath147 since @xmath133 is monotone , lemma [ thm : karamata - feller ] applies .
first , we show the assertion as @xmath148 , which is determined by the behavior of @xmath149 for large @xmath150 . in case
i ( recall @xmath75 ) , we have @xmath151 by lemma [ thm : karamata - feller ] this shows the first part of assertion . in case
ii , we have @xmath152 since @xmath153)$ ] we obtain , after integration by parts , @xmath154 hence , @xmath155 and @xmath156 as @xmath157 thus @xmath158 and now the second part of assertion follows from lemma [ thm : karamata - feller ] .
next we show the asymptotic behavior as @xmath138 . in case i we have for @xmath159 @xmath160 the function @xmath134 is slowly varying at infinity .
hence , by lemma [ thm : karamata - feller ] , @xmath161 which shows the first part of assertion .
we finally consider the large time asymptotic behavior in case ii .
it follows from the splitting and using the hlder continuity of @xmath162 that @xmath163 ( see also ( * ? ? ?
* proposition 2.2 ) for a related estimate ) .
consequently , @xmath164 thus @xmath165 with @xmath166 slowly varying at infinity , and lemma [ thm : karamata - feller ] completes the proof of the lemma .
in this section we describe a general class of spatially semidiscrete methods for , and review three specific examples of such methods mentioned in the introduction .
we then analyze their nonnegativity preservation properties . as described in section [ sec : intro ] , we consider semidiscrete approximations for problem of the form , where @xmath37 $ ] is an inner product in @xmath30 , approximating the usual @xmath167 inner product @xmath168 .
we denote the stiffness matrix by @xmath169 , with @xmath47 , and the mass matrix by @xmath170 , with @xmath45 $ ] . the semidiscrete problem
may then be written in matrix form as @xmath171 or , equivalently , after multiplication by @xmath172 , @xmath173 in analogy with , we find , for the solution matrix of , @xmath174 where @xmath175 is the identity matrix . since both the stiffness matrix @xmath176 and the mass matrix @xmath177 are symmetric positive definite , the eigenvalue problem @xmath178 has a complete system of eigenvectors @xmath179 with positive eigenvalues @xmath180 , so that @xmath181 , where the rows of @xmath182 are the eigenvectors @xmath179 and the diagonal elements of the diagonal matrix @xmath183 are the eigenvalues @xmath180
. thus , corresponding to the eigenfunction expansion , we have @xmath184 we now briefly review our three examples of finite element methods and the corresponding inner products @xmath37 $ ] in @xmath30 .
the first example is the standard galerkin ( sg ) method , with the standard @xmath26 inner product , i.e. , we choose @xmath37=(\cdot,\cdot)$ ] .
our second example is the lumped mass ( lm ) method , using @xmath185 = ( w , \chi)_h = \sum_{\k \in \t_h } q_{\k , h}(w \chi ) , { \quad\text{with}\ } q_{\k , h}(f ) = \tfrac13\,|\k|\sum_{j=1}^{3 } f(\zk_j ) \approx \int_\k f dx,\ ] ] where @xmath186 , @xmath187 , are the vertices of the triangle @xmath188 and @xmath189 is its area . in this case
the mass matrix @xmath190 is diagonal with positive diagonal elements .
our third example is the finite volume element ( fve ) method , cf .
@xcite , which is based on a discrete version of the local conservation law @xmath191 valid for any @xmath192 with a piecewise smooth boundary @xmath193 , with @xmath194 the unit outward normal to @xmath193 .
the discrete method then requires to be satisfied for @xmath195 , which are disjoint so called _ control volumes _ associated with the nodes @xmath196 of @xmath29
. it can be recast as a galerkin method , by letting @xmath197 introducing the interpolation operator @xmath198 by @xmath199 , @xmath200 , and then defining the inner product @xmath201 for all @xmath202 .
the fve method then corresponds to with @xmath37=\langle\cdot,\cdot\rangle$ ] .
we recall that an edge @xmath203 of the triangulation @xmath204 is called a delaunay edge if the sum of the angles @xmath205 and @xmath206 opposite @xmath203 is @xmath207 , and that @xmath204 is a _ delaunay triangulation _ if all interior edges are delaunay .
a node of @xmath29 is said to be _ strictly interior _
if all its neighbors are interior nodes , and @xmath29 is _ normal _ if it has a strictly interior node , @xmath196 say , such that any neighbor of @xmath196 has a neighbor which is not a neighbor of @xmath196 . a symmetric , positive definite matrix with non - positive off diagonal entries
is called a _ stieltjes _ matrix .
we now turn to our main goal , to determine whether the nonnegativity preservation property remains valid for the semidiscrete problem , i.e. , if @xmath50 for @xmath34 .
we shall first discuss the general case , and then the lm method .
our first result states that the semidiscrete method does not satisfy in general , if the mass matrix @xmath177 is _ nondiagonal _ , in the sense that @xmath208 for all neighbors @xmath209 .
[ thm : failsg ] assume that the triangulation @xmath29 is normal and that the mass matrix @xmath177 is nondiagonal .
then the solution matrix @xmath48 can not be nonnegative for all @xmath13 .
assume that @xmath50 for @xmath34 . by and lemma [ lem : cpl_mt ]
, we have @xmath210 and , since @xmath211 for small @xmath7 , the nonnegativity of @xmath48 implies @xmath212 for all @xmath213 .
let @xmath196 be a strictly interior node as in the definition of @xmath29 being normal .
we shall show that @xmath214 for @xmath215 .
consider first the case that @xmath216 is not a neighbor of @xmath196 , so that @xmath217 .
since @xmath218 , @xmath219 since @xmath220 for @xmath221 , we have @xmath222 , @xmath221 , and thus @xmath223 for @xmath221 , and in particular , @xmath214 . when @xmath216 is a neighbor of @xmath196
, it has a neighbor @xmath224 which is not a neighbor of @xmath196 , and hence @xmath225 , which implies @xmath214 since @xmath226 , in view of the assumption that @xmath177 is nondiagonal .
thus , @xmath227 i.e. , the @xmath228th columns of @xmath176 and @xmath177 are proportional , which contradicts the facts that @xmath229 for a strictly interior node @xmath196 and @xmath230 .
next we show a nonnegativity preservation result in the general case , for large time .
[ theorem : sg ] suppose @xmath231 .
then there exists a @xmath14 such that @xmath232 for all @xmath233 . as before
, it follows from the asymptotic behavior of the function @xmath133 , cf .
lemma [ lem : cpl_mt ] , that @xmath234 since by assumption @xmath231 , we find that @xmath48 is positive for large @xmath7 .
theorem [ thm : failsg ] covers the sg and fve methods , and theorem [ theorem : sg ] applies also to the lm case .
we recall from @xcite that @xmath235 if @xmath63 is a stieltjes matrix , and that @xmath63 is a stieltjes matrix if and only if @xmath29 is a delaunay triangulation . in particular ,
if @xmath29 is delaunay , then @xmath236 , and hence also @xmath237 , but @xmath238 may be positive also for some non - delaunay triangulations .
in the case of the lm method we may prove the following sharper result . [ thm : semi - lm ] let @xmath48 be the solution matrix of the lm approximation .
then @xmath239 for all @xmath240 if and only if the triangulation @xmath29 is delaunay .
if @xmath239 for all @xmath240 we conclude as in the proof of theorem [ thm : failsg ] that @xmath212 for all @xmath213 , consequently , since @xmath51 is a positive diagonal matrix , @xmath241 for all @xmath213 , so that @xmath63 is stieltjes and hence @xmath29 delaunay .
the proof of the converse statement requires the corresponding result for the fully discrete solution matrix @xmath242 , and the convergence of the fully discrete solution matrix to the semidiscrete one .
these two results are shown in theorem [ thm : lm - fully ] and lemma [ 4conv ] , respectively , and assuming their validity we find @xmath243 .
we remark that in the single - term case , the converse part of theorem [ thm : semi - lm ] also follows from the representation @xmath244 of the solution shown in @xcite , where @xmath245 is the mittag - leffler function .
since , by @xcite , @xmath246 is completely monotone , bernstein s theorem implies that @xmath247 , with @xmath248 a positive measure . recalling from @xcite that @xmath249 for @xmath34
when @xmath29 is delaunay , we conclude @xmath250
now we discuss the preservation of nonnegativity for a fully discrete method for the subdiffusion model , developed in @xcite , based on applying convolution quadrature @xcite to the semidiscrete problem . using the definition of the caputo fractional derivative ,
we deduce that @xmath251 , and hence problem may then be rewritten as @xmath252 for the discretization in time of we introduce a time step @xmath253 , and set @xmath254 , @xmath255 . for a smooth function @xmath256 ,
the approximation @xmath257 , of @xmath258 , is then given by @xmath259 the weights @xmath260 are generated by the characteristic polynomial @xmath261 of the backward euler ( be ) method , i.e. , @xmath262 the fully discrete scheme is then given by @xcite : @xmath263 noting that @xmath264 , the fully discrete solution can be explicitly represented by @xmath265 clearly this relation defines a sequence of rational functions @xmath266 such that @xmath267 , and analogously to we may write , with @xmath268 the eigenvalues of @xmath269 , @xmath270 we now turn to nonnegativity preservation property for the method .
we first show some properties of the quadrature weights @xmath260 .
[ lem : quad1prop ] let the weights @xmath260 be generated by the backward euler method .
then @xmath271 it suffices to show the assertion for the single - term model ( @xmath76 and @xmath272 ) , since the more general case follows by linearity .
property ( i ) follows directly from the explicit formula @xmath273 for ( ii ) , we claim that the partial sum satisfies the relation @xmath274 we show the claim by mathematical induction .
first , we note that @xmath275 satisfies with @xmath276 .
next , assuming holds up to @xmath277 , we find @xmath278 thus the relation holds for all @xmath279 , which completes the proof of the lemma .
[ lem : fullyequiv1 ] let @xmath280 , where @xmath281 .
then @xmath282 for @xmath283 .
let @xmath284 and let @xmath285 be defined by for @xmath283
. for a proof by induction , assume that @xmath286 for @xmath287 .
by assumption , @xmath288 , and by lemma [ lem : quad1prop ] we have @xmath289 hence @xmath290 , which completes the proof .
next we state the following fully discrete analogue of theorem [ thm : failsg ] .
assume that the triangulation @xmath29 is normal and the mass matrix @xmath177 is nondiagonal .
then the solution matrix @xmath291 can not be nonnegative for small @xmath292 . from the proof of lemma [ lem : cpl_mt ] we find as @xmath293 @xmath294 and hence @xmath295 since @xmath296 ,
the argument in the proof of theorem [ thm : failsg ] then completes the proof .
we now show some results concerning positivity thresholds for the fully discrete method .
[ thm : sg - full ] the following statements hold . * if @xmath231 , then there exists a @xmath297 such that @xmath282 for @xmath298 . *
if @xmath299 , then @xmath300 for @xmath301 , and @xmath302 .
if @xmath231 , then by continuity , @xmath303 for @xmath304 small , i.e. , for @xmath56 large . in particular , there exists @xmath305 such that @xmath306 for @xmath307 , and hence @xmath282 for @xmath308 , by lemma [ lem : fullyequiv1 ] . for part ( ii ) , we note that if @xmath309 , then @xmath310 , with @xmath311 .
we show that then @xmath312 for @xmath313 $ ] .
in fact , with @xmath314 , we may write @xmath315 note that @xmath316 by assumption , and if @xmath317 is so small that for some matrix norm @xmath318 , @xmath319 , then @xmath320 , and therefore , @xmath321 .
but if @xmath321 for @xmath322 $ ] , with @xmath323 , then @xmath324 .
hence , by repeating the argument , we deduce @xmath312 for some @xmath325 , and thus the smallest such @xmath326 has to be 0 .
since @xmath327 is monotone , this means that @xmath306 for @xmath307 , and thus shows the first part of ( ii ) , by lemma [ lem : fullyequiv1 ] . the last part of ( ii ) now follows from @xmath328 .
we note that if @xmath237 and @xmath329 is the largest number such that @xmath330 , then in the single - term case the positivity threshold equals @xmath331 .
for @xmath332 , this number increases with @xmath333 $ ] , with its largest value for @xmath334 , i.e. , for the heat equation , and tends to zero as @xmath335 .
the following result gives a somewhat more precise estimate for the positivity threshold @xmath304 , in the case of strictly " delaunay triangulations .
[ thm : threshold ] if @xmath336 for all neighbors @xmath209 and @xmath177 is nondiagonal , then @xmath337 for all @xmath283 if @xmath338 where @xmath339 .
if the assumptions hold , we have @xmath340 for @xmath341 , so that @xmath342 is a stieltjes matrix .
hence @xmath343 , and thus @xmath337 for @xmath283 , by lemma [ lem : fullyequiv1 ] . in particular , for the single - term model , in the case of a quasiuniform family of triangulation @xmath204 , with @xmath344 for all edges @xmath345
, we have @xmath300 for @xmath283 , if @xmath346 , with @xmath347 .
we finally show the following fully discrete analogue of theorem [ thm : semi - lm ] for the lm method .
[ thm : lm - fully ] let @xmath242 be the solution matrix of the lm method .
then @xmath337 for all @xmath292 , @xmath283 , if @xmath29 is delaunay , and , conversely , if @xmath306 for all @xmath292 , then @xmath29 is delaunay .
if @xmath204 is delaunay , then the stiffness matrix @xmath176 is stieltjes , and so is @xmath348 , for any @xmath349 .
hence @xmath350 . and , by lemma [ lem : fullyequiv1 ] , @xmath282 for all @xmath351 .
conversely , if @xmath352 for all @xmath292 , then holds for @xmath56 small .
hence @xmath353 and thus @xmath241 for @xmath354 neighbors , so that @xmath176 is stieltjes and @xmath29 is delaunay .
in this section we present numerical experiments to illustrate and to complement our theoretical findings .
we consider the following five domains with different triangulations , cf . fig [ fig : meshes ] : * the unit square @xmath355 , partitioned by uniform delaunay triangulations ; * the unit square @xmath355 , partitioned into a structured family of non - delaunay triangulations ; * the l - shaped domain @xmath356)}$ ] , partitioned by unstructured delaunay triangulations ; * the unit disk @xmath357 , partitioned by unstructured delaunay triangulations ; * the unit square @xmath355 , partitioned into non - delaunay triangulations , obtained from the uniform triangulations in example ( a ) with an additional subdivision of one boundary triangle .
our numerical experiments cover single - term ( @xmath76 , @xmath358 ) , multi - term ( with @xmath359 , @xmath360 ) , and distributed order cases of problem , to be made specific below . for each example , we investigate the preservation of nonnegativity for the spatially semidiscrete and the fully discrete sg , fve and lm methods . in the third case , of course , calculations are only needed for the non - delaunay triangulations in ( b ) and ( e ) . for our numerical calculations of the solution matrices
@xmath48 we use the eigenvalue decomposition , where the @xmath361 are defined as inverse laplace transforms in . for the latter
we employ a numerical algorithm from @xcite , deforming the contour @xmath127 to the left branch of a hyperbola in the complex plane with the parametric representation @xmath362 , @xmath363 , @xmath364 and @xmath365 .
the resulting integral is then approximated by a truncated trapezoidal rule , which , for an appropriate choice of the parameters @xmath366 and @xmath367 , is exponentially convergent .
_ example ( a ) .
_ as in @xcite we begin by studying a family of uniform triangulations @xmath31 , constructed as follows . for given @xmath368
we divide the sides of @xmath61 into @xmath369 equal subintervals of length @xmath370 , thus dividing @xmath2 into @xmath371 small squares . by means of parallel diagonals in the small squares
we then obtain a uniform triangulation , with @xmath372 .
note that for such triangulations , the stiffness matrix @xmath46 has @xmath373 and @xmath374 if @xmath375 are vertical or horizontal neighbors , with @xmath376 otherwise .
in particular @xmath63 is stieltjes , with @xmath235 , and hence @xmath377 for the sg , lm and fve methods .
thus the results in sections [ sec : semidis ] and [ sec : fully ] concerning nonnegativity preservation for large @xmath7 and @xmath56 apply . in fig .
[ fig : semi_a ] , we plot the smallest entry of the solution matrix @xmath48 for the semidiscrete sg , lm and fve methods with @xmath378 .
the lm method preserves nonnegativity , and in table [ tab : semi - fully_a ] we present threshold values @xmath379 and @xmath380 for the semidiscrete sg and fve methods , respectively , for @xmath381 , @xmath382 and @xmath383 .
their behavior is in agreement with theorems [ thm : failsg ] , [ theorem : sg ] and [ thm : semi - lm ] . like in @xcite
the thresholds are smaller for the fve methods than for the sg methods and decrease with the fractional order @xmath384 in the single - term case .
we also observe that the thresholds decrease with @xmath28 .
for the fully discrete method , the positivity thresholds @xmath385 and @xmath386 of @xmath291 are also given in table [ tab : semi - fully_a ] .
note that the triangulations do not satisfy the condition in theorem [ thm : threshold ] .
nevertheless , following the argument in ( * ? ? ?
* section 5.1 ) , lower bounds for the positivity thresholds may be derived in the fully discrete case . by table [ tab : semi - fully_a ]
it appears that , for the single - term model , this is of order @xmath387 , and the multi - term model shows a similar behavior .
.positivity thresholds for example ( a ) , for the semidiscrete and fully discrete sg and fve methods .
[ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^,^ " , ]
in the proof of theorem [ thm : semi - lm ] and also in appendix b we need the convergence of the fully discrete solution matrix to the semidiscrete one as the time step tends to zero , as shown in the following lemma .
[ 4conv ] let @xmath48 and @xmath242 be the semidiscrete and fully discrete solution matrices , defined by and , respectively .
then , with the norm defined by the inner product @xmath388 , we have @xmath389 by eigenvector expansion , cf . and , and parseval s formula we have @xmath390 where @xmath391 with @xmath392 , are the rational functions for which @xmath267 , and the @xmath268 are the eigenvalues of @xmath269 .
so it remains to show the inequality @xmath393 we recall from that @xmath394 to determine a corresponding representation of @xmath395 we consider the fully discrete solution @xmath396 of the scalar problem for @xmath64 , which is given by @xmath397 multiplying both sides by @xmath398 and summing over @xmath194 from @xmath399 to @xmath400 , we obtain @xmath401 upon adding @xmath304 to both sides , the summation in the double sums may start at @xmath276 , so that they are both discrete convolutions , and hence , with @xmath402 the generating function of the @xmath403 ( i.e. , @xmath404 ) , @xmath405 that of the backward euler method , and @xmath406 that of the sequence @xmath407 , we obtain @xmath408 or @xmath409 consequently , we have @xmath410 hence , the @xmath194th term in @xmath402 is given by @xmath411 and changing variables with @xmath412 yields , with small @xmath413 , @xmath414 since @xmath415 is analytic in the complement of the negative real axis we may deform the contour of integration into @xmath127 and , with the change of variables @xmath416 , we obtain the error representation @xmath417 we now deform @xmath127 into @xmath418 , with @xmath419\}$ ] a circular arc and @xmath420 , a pair of rays in the left half - plane , where @xmath421 . on @xmath422
, we have @xmath423 and @xmath424 and hence @xmath425 next , we write @xmath426 where @xmath427 \}$ ] and @xmath428 . with @xmath429 and @xmath430 large enough such that @xmath431 on @xmath432 , using also @xmath433 on @xmath434 , we have @xmath435 further , with @xmath436 given , we have @xmath437 for @xmath438 and @xmath429 small enough , and hence @xmath439 we also note that @xmath440 for @xmath441 with @xmath442 .
in fact , if @xmath443 , with @xmath444 , say , then the same holds for @xmath445 if @xmath446 $ ] and @xmath447 , and it hence suffices to show @xmath448 for @xmath449 . for @xmath450 we have @xmath451 , and for @xmath452 , @xmath453 . together
these inequalities yield @xmath454 this completes the proof of the lemma .
we observe from that by cauchy s integral formula , @xmath455 .
the maximum principle for ensures the maximum - norm contractivity , i.e. , @xmath456 it is natural to ask if this property remains valid for discrete methods , and in @xcite this problem was discussed for the sg and lm methods for the heat equation .
it turned out that , in contrast to the continuous case , the discrete analogue of is not equivalent to the preservation of nonnegativity . in this appendix
, we discuss maximum - norm contractivity for the lm method . in this case
the desired property holds when the stiffness matrix @xmath63 is diagonally dominant .
we first show a contraction property for the fully discrete method , and write @xmath457 [ thm : lm - contraction - discrete ] the solution matrix @xmath242 of the fully discrete lm method is contractive in the maximum - norm if the matrix @xmath176 is diagonally dominant . if @xmath291 is contractive in the maximum - norm , then @xmath63 is diagonally dominant .
assume that @xmath176 is diagonally dominant . since @xmath177 is positive and diagonal , @xmath458 is row diagonally dominant .
let @xmath459 , set @xmath460 , and let @xmath461 .
then @xmath462 and hence @xmath463 , i.e. , @xmath464 .
we now prove that @xmath465 is a contraction , by showing by induction that , for all @xmath459 , @xmath466 . to this end , assume that this holds for all @xmath467 . using lemma [ lem : quad1prop ] , we then find @xmath468 and hence , cf . , we get @xmath469 which shows our claim . now , suppose @xmath291 is a contraction . in view of , we have @xmath470 if @xmath471 , by taking @xmath56 small we find @xmath472 for @xmath473 . hence @xmath474 is row diagonally dominant .
since @xmath475 is a positive diagonal matrix , @xmath476 is also row diagonally dominant and hence , since it is symmetric , diagonally dominant .
the next result gives the contractivity of the semidiscrete lm method .
[ thm : lm - contraction ] the semidiscrete lm solution matrix @xmath48 is contractive in the maximum norm if @xmath176 is diagonally dominant . if @xmath48 is contractive for small @xmath7 then @xmath63 is diagonally dominant . by theorem [ thm : lm - contraction - discrete ] , @xmath242 is a contraction in the maximum - norm , and , by lemma [ 4conv ] , converges to the semidiscrete solution matrix in the sense that @xmath477 .
hence also @xmath48 is a contraction .
conversely , if @xmath48 is a contraction in @xmath478 for small @xmath7 , then by we have @xmath479 the rest of the proof is then identical to that of theorem [ thm : lm - contraction - discrete ] .
the work of the first author ( b. jin ) is partly supported by uk engineering and physical sciences research council grant ep / m025160/1 .
p. chatzipantelidis , z. horvth , and v. thome .
on positivity preservation in some finite element methods for the heat equation . , 15(4 ) : 417437 , 2015 .
p. chatzipantelidis , r. lazarov , and v. thome .
some error estimates for the finite volume element method for a parabolic problem . , 13(3):251279 , 2013 .
h. fujii . some remarks on finite element analysis of time - dependent field problems .
in . theory and practice in finite element structural analysis ( y. yamada , r. h. gallagher & n. k. kyokai eds ) .
tokyo , japan : university of tokyo press , pp .
91106 . , 1973 .
j. r. shewchuk .
triangle : engineering a 2d quality mesh generator and delaunay triangulator . in m.
c. lin and d. manocha , editors , _ applied computational geometry : towards geometric engineering _ , pages 203222 .
springer , 1996 .
department of computer science + university college london + gower street , london wc1e 6bt , uk + e - mail : ( [email protected],[email protected] ) + department of mathematics + texas a&m university + college station , tx 77843 - 3368 , usa + e - mail : [email protected] + mathematical sciences + chalmers university of technology and the university of gothenburg + se-412 96 gteborg , sweden + e - mail : [email protected] + department of applied physics and applied mathematics + columbia university + 500 w. 120th street , new york , ny 10027 , usa + e - mail : [email protected] + | we consider three types of subdiffusion models , namely single - term , multi - term and distributed order fractional diffusion equations , for which the maximum - principle holds and which , in particular , preserve nonnegativity .
hence the solution is nonnegative for nonnegative initial data .
following earlier work on the heat equation , our purpose is to study whether this property is inherited by certain spatially semidiscrete and fully discrete piecewise linear finite element methods , including the standard galerkin method , the lumped mass method and the finite volume element method .
it is shown that , as for the heat equation , when the mass matrix is nondiagonal , nonnegativity is not preserved for small time or time - step , but may reappear after a positivity threshold .
for the lumped mass method nonnegativity is preserved if and only if the triangulation in the finite element space is of delaunay type .
numerical experiments illustrate and complement the theoretical results .
+ * keywords * : subdiffusion , finite element method , nonnegativity preservation , caputo derivative # 1h^#1 ( ) |
light propagation in a nonlinear dielectric may be used to model various subtle effects involving quantum theory and gravity .
these include lightcone fluctuations @xcite and the effects of quantum stress tensor expectation values in semiclassical gravity @xcite .
a fluctuating electric field in a nonlinear material causes fluctuations of the effective speed of light of probe pulses , and is analogous to the effects of spacetime geometry fluctuations on light propagation .
this analogy was developed in ref .
@xcite , where the source of the fluctuations was a squeezed state of the electromagnetic field , and in ref .
@xcite , where the effects of vacuum fluctuations of the electric field were investigated . in both cases , linear fluctuations of the electric field were treated , which models the lightcone fluctuations produced by active gravitational field fluctuations .
these are the fluctuations of the dynamical degrees of freedom of gravity itself , as opposed to the passive fluctuations of gravity , driven by quantum stress tensor fluctuations .
one of the purposes of the present paper will be to develop a model for passive spacetime geometry fluctuations .
this will involve a study of the fluctuations of the time averaged squared electric field , which is of interest in its own right .
a second purpose of this paper will be a further study of switched fluctuations of quantum fields .
the vacuum fluctuations of quantum field operators are only meaningful if the operators have been averaged in time or in spacetime with a smooth sampling function . in the case of linear fields , such as the electric field
, the associated probability distribution is gaussian .
some effects of the time averaged electric field were discussed in refs .
@xcite . in the latter paper
, it was shown that simple arguments may be used to estimate the one loop qed corrections to potential scattering by electrons .
the fluctuations of quadratic field operators , such as the stress tensor or the squared electric field , are more subtle , and are associated with non - gaussian probability distributions @xcite .
these distributions typically fall more slowly than a gaussian function , increasing the probability of large fluctuations , and depend sensitively upon the choice of sampling function . it was argued in ref .
@xcite that the sampling function for vacuum fluctuations in a dielectric can depend upon the geometry of the material .
this idea will be further developed here , where we will consider a broader class of functions than the lorentzian function used in ref .
@xcite .
the outline of this paper is as follows : section [ sec : flight ] will first briefly review classical light propagation in a nonlinear material , and then address the effects of switched vacuum fluctuations of the electric field and the squared electric field .
a convenient choice of switching function will be introduced in sec .
[ sec : switch ] .
some numerical estimates of the magnitude of the flight time fluctuations will be given in sec .
[ sec : estimates ] .
the probability distribution for the flight time fluctuations will be discussed in sec .
[ sec : prob ] .
our results will be summarized and discussed sec .
[ sec : sum ] . throughout this paper
we use lorentz - heaviside units with @xmath0 .
a nonlinear dielectric material is one where the electric polarization vector is a nonlinear function of the electric field , and can be written as @xcite @xmath1 here repeated indices are summed upon , and @xmath2 , @xmath3 , and @xmath4 are the first , second , and third order susceptibility tensors , respectively .
the second and higher order susceptibilities lead to a nonlinear wave equation for the electric field .
we wish to investigate the flight time of a probe pulse propagating through a slab of optical material when second and third order coefficients of the susceptibility tensor are included .
these nonlinearities of the medium couple to an external applied electric field @xmath5 , here called the background field . the electric field
associated with the probe pulse is denoted by the vector @xmath6 , which we choose to be polarized in the @xmath7-direction and propagating in the @xmath8-direction , i.e. , @xmath9 .
furthermore , we assume that the probe field is smaller in magnitude than the background field , but more rapidly varying .
that is , @xmath10 but @xmath11 in this case , @xmath12 obeys a linearized wave equation @xcite , @xmath13 here @xmath14 is the phase velocity of the wave , which is given by @xmath15^{-1 } , \label{eq : vph}\ ] ] where @xmath16 is the refractive index of the medium measured by the probe pulse when only linear effects take place , and we define the coefficients @xmath17 equation ( [ eq : vph ] ) shows that the background field couples to the nonlinearities of the medium , affecting the velocity of the waves propagating through it .
we will assume that dispersion can be ignored , so that the group velocity of a wavepacket is approximately equal to the phase velocity . in this case ,
the flight time of a probe pulse traveling a distance @xmath18 in the x - direction will be given by @xmath19\,dx , \label{td}\end{aligned}\ ] ] with @xmath20 here the parenthesis enclosing two indices denotes symmetrization , i.e. , @xmath21 .
in writing eq .
( [ td ] ) , we have assumed that the nonlinear effects are small , so that we may taylor expand @xmath22 from eq .
( [ eq : vph ] ) to first order in @xmath23 and second order in @xmath24 .
in addition , we take the integrand in eq .
( [ td ] ) to be evaluated at @xmath25 , which is the worldline of a pulse traveling at speed @xmath26 , that determined by the linear susceptibility . in this paper
, we will follow ref .
@xcite and study the effects of vacuum electric field fluctuations as the background field . in this case , @xmath27 becomes the quantized electric field operator , and @xmath28 defined in eq .
( [ td ] ) becomes an operator , where the term quadratic in @xmath27 is understood to be normal ordered , @xmath29 .
this leads to a finite mean flight time , which in the vacuum state is , to leading order , @xmath30 our primary interest in this paper will be in the variance of the flight time , @xmath31 note that this quantity is independent of the choice of vacuum state with respect to which normal ordering is performed .
a change in the state has the effect of adding a c - number , @xmath32 , to the operator @xmath28 , so that @xmath33 .
it is easily verified that the right - hand - side of eq .
( [ eq : var ] ) is unchanged .
a change of vacuum state can slightly change the mean time delay , @xmath34 , but does not change the variance of the flight time , which is our primary concern .
however , this quantity is only finite if the field operators have been averaged with a test function . in the present context
, the density profile of the slab of dielectric naturally defines a suitable function .
let @xmath35 be a profile function satisfying @xmath36 now the time delay operator may be written as @xmath37\,f(x)dx \ , .
\label{tdf}\end{aligned}\ ] ] the flight time variance now becomes @xmath38 .
\label{dtdf}\end{aligned}\ ] ] the definition of normal ordering , @xmath39 and the use of wick s theorem lead to @xmath40 thus the flight time variance can be expressed as an integral involving the correlation functions of the electric field .
this double integral is over the spacetime volume of the worldtube of the probe pulse wavepacket .
this worldtube is centered upon the worldline of the middle of the wavepacket , described by @xmath41 .
we will assume that the wavepacket is sufficiently localized around this worldline so that integrations over the spatial directions transverse to the @xmath8-direction may be neglected .
in this limit , we are averaging the electric field and the squared electric field along the worldline of an observer comoving with the probe pulse .
in the rest frame of this observer , the field operators are being averaged in time alone .
once we take the coincidence limit in the transverse spatial directions , the needed electric field correlation functions for a nondispersive , isotropic material become @xcite @xmath42 ^ 2}\ , , \label{eq : exx } \\
\langle e^0_y(x , t)e^0_y(x',t')\rangle&= \langle e^0_z(x , t)e^0_z(x',t')\rangle = \frac{(\delta x)^2 + ( \delta t)^2/n_b^2}{\pi^2 \ , n_b^3\ , \left [ ( \delta t)^2/n_b^2 - ( \delta x)^2\right]^3}\ , , \label{eq : eyy } \\ \langle e_i^0({x},t)e_j^0({x}',t')\rangle & = 0,\ \ i\neq j,\end{aligned}\ ] ] where @xmath43 and @xmath44 , with @xmath45 , and @xmath46 is the refractive index measured by the background field @xmath47 . using the above results and recalling that the integrations in eq . ( [ dtdf ] )
are performed along the path of the probe pulse , given by @xmath48 , we obtain @xmath49 , \label{final2}\end{aligned}\ ] ] where @xmath50 is understood to have a small negative imaginary part . here
we have defined the parameters @xmath51 and @xmath52 as @xmath53 , \\
\alpha_2 & = \frac{2{n_b}^2{n_p}^2}{\pi^4\left({n_p}^2-{n_b}^2\right)^4}\left[\mu_{xx}^2 + \left(\mu_{yy}^2+\mu_{zz}^2 + 2\mu_{zy}^2\right)\frac{\left({n_p}^2+{n_b}^2\right)^2}{\left({n_p}^2-{n_b}^2\right)^2 } % \right.-\nonumber\\&-\left .
+ 2\left(\mu_{xy}^2+\mu_{xz}^2\right)\frac{\left({n_p}^2+{n_b}^2\right)}{\left({n_p}^2-{n_b}^2\right)}\right ] .
\label{an}\end{aligned}\ ] ] this result generalizes previous work @xcite by including the contribution from the third order nonlinear susceptibility , and by giving an expression for the flight time for a general profile function @xmath35 .
we wish to choose a suitable smooth switching function that represents the transitions which occur as the probe pulse enters and exits the medium . it will be useful to have two parameters , one ( @xmath18 ) which describes the width of the slab and another ( @xmath54 ) which describes the effective length over which the nonlinearity changes smoothly as the pulse enters and exits .
there are several choices for such a function . here
we use a function @xmath55 defined by @xmath56 .
\label{fbd}\end{aligned}\ ] ] . ]
the derivative of this function with respect to @xmath8 is a sum of two lorentzian functions .
figure [ figa ] presents some plots of @xmath55 for a few values of the ratio @xmath57 .
the parameter @xmath58 describes the distance over which @xmath59 changes from its minimum to its maximum values , and vice versa .
note that when @xmath60 we recover a step function , as expected .
the integrals appearing in eq .
( [ final2 ] ) may be evaluated by contour integration , with @xmath61 , where @xmath45 .
the results , and their asymptotic forms when @xmath62 , are @xmath63 if we assume @xmath62 , and use the above asymptotoic forms , we obtain @xmath64 we define the squared fractional variance in flight time of the probe field as @xmath65 the modulus of the fourier transform of @xmath55 is given by @xmath66 and its behavior is depicted in fig .
[ figb ] , where we defined the dimensionless variable @xmath67 and function @xmath68 .
note that @xmath69 and that @xmath70 falls exponentially as @xmath7 increases .
the plot was done with the particular choice @xmath71 , for which more than @xmath72 of total area under the solid curve occurs in the range @xmath73 .
recall that our approximations require ( i ) eq . ( [ eq : c1 ] ) , the dominance of the vacuum field over the probe field , ( ii ) eq . ( [ eq : c2 ] ) , which is equivalent to @xmath74 , ( iii ) a range of frequencies in which the material can be assumed free of dispersion , and ( iv ) a material which is approximately isotropic , at least for the frequencies which give the primary contribution to the background field .
the rate of decay of the fourier transform @xmath75 allows us to test approximations ( ii ) and ( iii ) .
the exponentially decreasing behavior of the fourier transform of this function , depicted in fig .
[ figb ] , suppresses the high energy modes of the background field .
is illustrated .
specifically , the function @xmath68 is plotted as a function of @xmath67 for the case @xmath76 for the case @xmath71 , at least @xmath72 of the effect will occur in the range @xmath77 , which means that only wave lengths such that @xmath78 will significantly contribute . for a slab with @xmath79 ,
the dominant wavelengths of the background field are those with @xmath80 .
shorter wavelength modes are naturally suppressed by the time averaging .
furthermore , the larger contribution occurs arises from @xmath81 , which for @xmath71 , corresponds to a wavelength of @xmath82 .
thus if the material is relatively free of dispersion when @xmath80 , then our assumption that @xmath46 is independent of frequency is justified .
we may choose @xmath83 to satisfy eq .
( [ eq : c2 ] ) .
we may also justify the assumption of the dominance of the vacuum field over the probe field , eq .
( [ eq : c1 ] ) , using essentially the same argument as was given in sec . 3.2 of ref . @xcite .
the only difference is that in ref .
@xcite , @xmath84 is the expectation value of the square of the averaged electric field .
here it is the square root of the expectation value of the square of the averaged squared electric field , which can be obtained from eq .
( [ eq : b6 ] ) divided by @xmath85 , and is proportional to @xmath86 for @xmath87 .
thus , if @xmath88 , the two quantities are of the same order , and we obtain eq .
( 30 ) of ref .
@xcite as the condition that the vacuum field dominate the probe field .
if @xmath89 , then the vacuum field is enhanced by the shorter switch - on and switch - off times , and it becomes easier to satisfy eq .
( [ eq : c1 ] ) .
the physical reason for vacuum dominance is that many more modes contribute to the vacuum field than to the probe field .
the first example we wish to study is the crystal of cadmium selenide ( cdse ) , which is a hexagonal one , point group 6 mm .
this system was already investigated @xcite in the case of a lorentzian sampling function .
cdse is an optical medium with nonzero second order nonlinear dielectric susceptibilities and satisfies the conditions discussed at the end of last section .
this crystal has an index of refraction @xmath90 and a second order coefficient @xmath91 at a wavelength @xmath92 @xcite . now setting the wavelength of the probe field as @xmath93 , for which @xmath94 , and setting the parameter @xmath95
, we obtain from eq .
( [ final3 ] ) a fractional variance of the flight time , @xmath96 compared to the model where an idealized lorentzian distribution @xcite is used , this result shows that in the situation described by @xmath55 , with @xmath97 , the predicted effect is about 100 times stronger .
this enhancement arises because the contribution to @xmath98 due to linear electric field fluctuations is proportional to @xmath99 , as may be seen from the first term on the right - hand - side of eq .
( [ frac ] ) .
now we investigate a third order nonlinear optical material .
silicon ( si ) is a centrosymmetric crystal ( point group m3 m ) , which means that the second order nonlinear dielectric susceptibilities are identically zero .
this crystal has a third order coefficient @xmath100 at a wavelength @xmath101 @xcite , and an index of refraction @xmath102 at the same wavelength @xcite .
suppose the probe wave packet has a peak wavelength of @xmath103 , for which @xmath104 @xcite .
as before , using @xmath95 , we find the dominant contribution to the fractional variance of the flight time , @xmath105 note that the contribution to @xmath98 due to quadratic electric field fluctuations is proportional to @xmath106 , as may be seen from the second term on the right - hand - side of eq .
( [ frac ] ) . as expected , the effect produced by the third order coefficient tends to be smaller than that related to second order nonlinearities .
it may be possible to increase the effects of quadratic fluctuations if new materials with larger third order susceptibilities can be found . in the next section
, we will discuss a different type of enhancement .
in the previous sections , we have been concerned with the variance of the flight time , which is in turn determined by the variance of the sampled electric field or of the squared electric field . here
we wish to estimate the probability of much larger fluctuations than those described by the variance . in the case of effects produced by the second order polarizability
, this probability will be very small , as the probability distribution for fluctuations of the electric field is gaussian , and hence falls very rapidly .
however , flight time variations due to the third order polarizability will be associated with a more slowly decreasing probability distribution .
the distributions for quadratic quantum operators have been discussed in refs .
in particular , the asymptotic form for the probability distribution of the lorentzian average of the squared electric field was given in ref .
the sampling function used in this paper , eq .
( [ fbd ] ) , is not lorentzian , but the magnitude of its fourier transform , eq .
( [ fourier ] ) , has the same exponential decay as in the lorentzian case .
furthermore , it was argued in ref .
@xcite that the decay rate of the fourier transform of the sampling function determines the asymptotic form for the probability distribution .
thus it is reasonable to extrapolate the lorentzian results to the present case .
here we briefly summarize the needed results from ref .
let @xmath107 be the lorentzian time average of the normal ordered squared electric field operator at a given point in space , or more generally along a timelike worldline . in our problem
, this will be the path of the probe wave packet .
define the dimensionless variable @xmath108 where @xmath109 is the characteristic averaging time .
let @xmath110 be the probability distribution for finding a given value of @xmath8 in a measurement in the vacuum state , which is normalized by @xmath111 here @xmath112 is the lower bound , the smallest value of @xmath8 which could ever be observed .
note that this lower bound is negative , so measurements of @xmath113 in the vacuum state can return negative values , just as expectation values of the squared electric field in more general states can be negative .
in fact , one expects most measurements in the vacuum to result in a negative value , but when the outcome is positive , it is likely to be larger in magnitude .
note that a negative value of @xmath107 results in a time _ advance _ compared to the mean flight time in the material , just as positive values result in time delays .
our primary interest is in the asymptotic form of @xmath110 when @xmath114 , which describes the probability of finding especially large values of the squared electric field .
this asymptotic form is approximately @xmath115 where @xmath116 and @xmath117 .
a striking feature of this result is the one - third power in the exponential , which causes @xmath110 to fall much more slowly than a gaussian or an exponential function .
given @xmath110 , we can define the cumulative probability distribution by @xmath118 which gives the probability of finding any value greater than or equal to @xmath119 in a given measurement .
if @xmath120 , we can directly integrate eq .
( [ eq : p ] ) to find @xmath121 it is shown in ref .
@xcite that the second moment of @xmath110 for the squared electric field is @xmath122 so the root mean square of @xmath8 is @xmath123 .
now we may use eq .
( [ eq : pcumm ] ) to find the probability of a result which exceeds a large multiple of @xmath124 .
some examples are given in table [ table : probs ] .
.probabilities of large squared electric field fluctuations . [ cols="^,^,^,^,^,^",options="header " , ] the same probabilities apply to the flight time delay due to vacuum squared electric field fluctuations .
thus there is a probability of about @xmath125 that a given pulse will suffer a delay which is 1000 times larger than the root mean square value , given for example by eq .
( [ si ] ) . note that our discussion is rather heuristic , and these are order of magnitude estimates . in particular
, we have not made a clear distinction between the squared electric field in the rest frame of the probe pulse , and that in the rest frame of the dielectric material .
however , for @xmath126 , so @xmath127 to leading order , these quantities will be of the same order .
another important point is that both the lorentzian function and the function @xmath55 defined in eq .
( [ fbd ] ) have tails in both directions .
a more realistic choice is a function of compact support , which is strictly zero before the measurement process begins .
such functions lead to even slower decrease of the probability distribution for large arguments @xcite .
in this paper , we have extended previous work @xcite on analog models for lightcone fluctuations .
here we have been concerned with vacuum fluctuations of the quantized electric field , and especially with fluctuations of the squared electric field .
our model studies the flight time of a probe pulse through a material with a non - zero third order polarizability .
vacuum fluctuations of the squared electric field lead to fractional flight time variation which can be of order @xmath128 in the example given in sec .
[ sec : estimates ] .
these flight time variations model the effects of the passive spacetime geometry fluctuations driven by quantum stress tensor fluctuations .
we have also extended the study of the effects of temporal switching functions on the fluctuations of quadratic quantum operators . in sec .
[ sec : switch ] , we discussed a specific choice of switching function , which can be relatively constant over a finite interval , and can model the density profile of the nonlinear material .
the fourier transform of this function falls exponentially at a rate determined by the parameter @xmath58 , which controls the rate of rise and fall at the ends of the plateau of this function . as this parameter
is decreased , increasingly higher frequency modes contribute and increase the fractional flight time variation .
we were able to use the fourier transform to estimate the range of vacuum modes which contribute to the flight time variation and to test our approximation of ignoring dispersion . in sec .
[ sec : prob ] , we discussed the probability of especially large fluctuations in flight time
. this analog model may provide a means to study the probability of large stress tensor fluctuations , which tend to fall more slowly than a gaussian function @xcite .
we estimated , for example , a probability of @xmath129 for finding a flight time delay which is at least @xmath130 times the typical delay .
there are then two distinct signatures of squared electric field fluctuations .
the first is the fractional flight time variation @xmath98 estimated , for example , in eq .
( [ si ] ) .
the second is the pulses which undergo an especially long time delay due to a very large squared electric field fluctuation .
the extent to which either can be observed in a realistic experiment is a topic for future work .
one aspect of this work will be an exploration of finite duration switching , which is more realistic than functions with tails extending into the past and the future .
such functions with compact support were treated in ref .
@xcite , where it was shown that quadratic quantum operators averaged in time with such functions are associated with a probability distribution which falls more slowly than that for the lorentzian , eq .
( [ eq : p ] ) .
this raises the possibility that large fluctuations can be more likely than was estimated in sec .
[ sec : prob ] .
this work was supported in part by the national science foundation under grant phy-1506066 , and by the brazilian research agencies cnpq ( grants 304486/2012 - 4 and 168274/2014 - 0 ) , fapemig ( grant etc-00118 - 15 ) , and capes . f. charra and g. gurzadyan , _ nonlinear dielectric susceptibilities _ , in springermaterials , d. f. nelson , ed .
landolt - brnstein - group iii condensed matter volume 30b , sec .
6.5.2.2 , ( springer - verlag berlin heidelberg 2000 ) . | we study a model for quantum lightcone fluctuations in which vacuum fluctuations of the electric field and of the squared electric field in a nonlinear dielectric material produce variations in the flight times of probe pulses .
when this material has a non - zero third order polarizability , the flight time variations arise from squared electric field fluctuations , and are analogous to effects expected when the stress tensor of a quantized field drives passive spacetime geometry fluctuations .
we also discuss the dependence of the squared electric field fluctuations upon the geometry of the material , which in turn determines a sampling function for averaging the squared electric field along the path of the pulse .
this allows us to estimate the probability of especially large fluctuations , which is a measure of the probability distribution for quantum stress tensor fluctuations . |
the double - differential scattering cross - section per scattering center ( assuming @xmath89 atoms in the sample ) can be written as @xmath90 where @xmath91 are the initial and final states of the crystal with energies @xmath92 , and @xmath93 is the coulomb interaction potential between the nuclei and electrons of the beam and sample .
the delta - function selects the transitions with energy - difference equal to energy - loss @xmath94 . because of the form of the initial state of the probe wavefunction @xmath95 + \nonumber \\ & + & \sum_{\mathbf{k } \in \omega_2 } ce^{i\phi_\mathbf{k+g}}e^{i(\mathbf{k+g})\cdot\mathbf{r } } \left [ 1 + i t_\mathbf{g } e^{i\mathbf{g}\cdot\mathbf{r } } \right],\end{aligned}\ ] ] the double - differential scattering cross - section contains quite a number of terms and following an explicit derivation for all of them is too lengthy and cumbersome .
however , the derivation follows the same pattern for all terms . therefore we illustrate it on a particular cross - term from the initial probe wave - function : a cross - term between @xmath96 and @xmath97 for @xmath98 . essentially we will perform a similar procedure as a textbook derivation of inelastic electron scattering cross - section of a plane wave on an atom ( see , e.g. , sakurai s _ modern quantum mechanics _
@xcite ) into another plane - wave @xmath99 : @xmath100 where we introduced momentum transfer vectors @xmath28 and @xmath101 , which carry on the @xmath102-dependences .
the crystal wavefunctions are denoted @xmath103 for the initial and final state , respectively .
since the @xmath104 integrations are done over the whole space , we can substitute @xmath105 and analogically for the primed variables .
this shift introduces phase factors @xmath106 and @xmath107 in the two integrals .
integration over @xmath108 is then trivial and leads to @xmath109 focusing on core - level excitations , the initial states are localized around the excited atom .
considering for simplicity crystals with only one atom per unit cell , the sum over all initial states @xmath110 can be written as @xmath111 , i.e. , a double sum over lattice vectors @xmath112 and over initial states on a single atom @xmath113 . shifting the local coordinates for each @xmath114 by @xmath112 we obtain @xmath115
the final state @xmath116 is an unoccupied bloch state with wavevector @xmath117 ( not to be mistaken with the outgoing beam wave - vector @xmath118 ) , thus we can use a bloch theorem stating @xmath119 .
that gives @xmath120 where the bloch - wave phase factor cancels out . introducing a notation @xmath121 the formula above
can be written as @xmath122 since @xmath8 is a reciprocal lattice vector and @xmath112 is a lattice vector , in the second row @xmath123 . furthermore , sum over @xmath112 vanishes , unless @xmath124 is a reciprocal lattice vector .
[ more precisely , only @xmath4 and @xmath125 components of @xmath124 must be components of a reciprocal lattice vector , because the crystal has a small finite thickness .
a more detailed treatment would include two bloch waves with slightly different @xmath34-components of their wavevectors , leading to thickness dependent prefactors of individual mdffs in eq .
( [ eq : ddscs ] ) below , causing , e.g. , _ pendellsung _ effects . however , this does not qualitatively influence any of our conclusions . ]
that is equivalent to a condition requiring @xmath126 to be a reciprocal lattice vector .
because of our partitioning of the cbed disks , this condition can only fulfilled when @xmath127 .
in such case the expression simplifies to @xmath128 where @xmath26 and we have introduced the mixed dynamical form - factor ( mdff ) @xmath129 taking a cross - term with swapped order of the terms from initial probe wave - function would lead to @xmath130 utilizing the hermitian property of mdff @xmath131 , we obtain for their sum @xmath132 \\ & = & ( 4\pi c)^2 \sum_{\mathbf{k } } 2\mathrm{re } [ e^{-i\delta\phi_\mathbf{k , g } } s(\mathbf{q},\mathbf{q - g},e ) ] \\ & \equiv & ( 4\pi c)^2 \sum_{\mathbf{k } } 2\big [ \cos\delta\phi_\mathbf{k , g } \mathrm{re}[s(\mathbf{q},\mathbf{q - g},e ) ] \\ & + & \sin\delta\phi_\mathbf{k , g } \mathrm{im}[s(\mathbf{q},\mathbf{q - g},e ) ] \big]\end{aligned}\ ] ] which is one of the terms in eq .
( 2 ) of the main text . chosen partitioning of the cbed disc
means that nonzero cross - terms only happen , when @xmath102 are both from the same region
either @xmath5 or @xmath9 , respectively .
the final result considering all cross - terms [ and dropping the constant factor @xmath133 is then @xmath134 \big ] \nonumber \\ & + & \sum_{\mathbf{k } \in \omega_2 } c^2 \big [ [ 1 + 2t_{\mathbf{g}}\sin(\delta\phi_{\mathbf{k , g}})]s(\mathbf{q - g},\mathbf{q - g } ) \nonumber \\ & + & t^2_{\mathbf{g}}s(\mathbf{q-2g},\mathbf{q-2 g } ) \nonumber \\ & + & 2\cos(\delta\phi_{\mathbf{k , g } } ) \mathrm{re}[s(\mathbf{q},\mathbf{q - g } ) ] \nonumber \\ & + & 2\sin(\delta\phi_{\mathbf{k , g } } ) \mathrm{im}[s(\mathbf{q},\mathbf{q - g } ) ] \nonumber \\ & - & 2t_{\mathbf{g } } \sin(\delta\phi_{\mathbf{k , g } } ) \mathrm{re}[s(\mathbf{q},\mathbf{q-2 g } ) ] \nonumber \\ & + & 2t_{\mathbf{g } } \cos(\delta\phi_{\mathbf{k , g } } ) \mathrm{im}[s(\mathbf{q},\mathbf{q-2 g } ) ] \nonumber \\ & + & 2t_{\mathbf{g}}^2 \cos(\delta\phi_{\mathbf{k , g } } ) \mathrm{re}[s(\mathbf{q - g},\mathbf{q-2 g } ) ] \nonumber \\ & + & 2[t_{\mathbf{g}}^2 \sin(\delta\phi_{\mathbf{k , g } } ) + t_\mathbf{g } ] \mathrm{im}[s(\mathbf{q - g},\mathbf{q-2 g } ) ] \big ] , \end{aligned}\ ] ] equal to eq .
( 2 ) in the main text .
a notable feature of this expression is a lack of cross - terms between different @xmath7-vectors within a cbed disc .
this is , as was shown above , due to the sum over lattice vectors @xmath135 , which appears due to shifts of local coordinate systems for each @xmath114 .
it is important to note that the phase distribution in the reciprocal space , @xmath16 , can consist of several contributions .
the aberrations of electron optics , such as defocus , spherical aberrations , comas or star aberrations , are among the most common contributions to the non - trivial phase distribution .
they can be to a large extent removed by probe aberration correctors .
alternatively , as is suggested here , probe correctors can be utilized to deliberately set - up a desired phase distribution in the beam .
there is , however , another source of the phase variation , which can not be principially removed in a stem experiment the phase ramp due to shift of the probe .
a probe shifted from origin ( e.g. , from an investigated atomic column ) by position vector @xmath75 will introduce a phase ramp @xmath76 in the reciprocal space probe wavefunction , as a consequence of the fourier shift theorem @xcite .
therefore , if the stem probe does not pass directly through an investigated atomic column , the phase distribution will be non - trivially modified , including modification of the symmetry of the phase - distribution .
this will , for example , violate the relation @xmath62 valid for a vortex beam passing through an atomic column as discussed in relation to eq .
( 10 ) of the main text .
it is not easy to qualitatively estimate the consequence of shift on the strength of emcd , but explicit calculations have been performed for vortex beams with oam=1 in ref .
@xcite . those can be qualitatively summarized in the following way : 1 ) for small shifts emcd monotonously decreases with increasing shift @xmath136 from an atomic column ; 2 ) for @xmath137fwhm of a beam with the same convergence angle , but with @xmath138 , the emcd will be reduced to less than 1/2 of the intensity for a beam passing directly through an atomic column .
edge of iron integrated over an on - axis aperture of diameter 8.7 mrad as a function of @xmath19 and sample thickness from 10 to 40 nm .
overlap of cbed discs onsets at @xmath139 a.u.@xmath69.,width=325 ] in the main text , the fig .
3 illustrated the dependence of the emcd signal detectable by electron vortex beams of three different diameters .
in the first column , the cbed disks did not overlap ( @xmath68 a.u.@xmath69 ) , in the second column there was a slight overlap ( @xmath140 a.u.@xmath69 ) giving rise to a small but nonzero emcd at transmitted beam , and in the third column the cbed disks ovelap strongly ( @xmath141 a.u.@xmath69 ) leading to a substantial emcd at transmitted beam direction . here
we extend the analysis by showing the dependence of the emcd signal as a function of cbed disk diameter @xmath19 in the range from @xmath142 to @xmath143 a.u.@xmath69 , see fig .
[ fig : qmaxdep ] .
as mentioned in the main text , the overlap onsets when @xmath81 , where @xmath82 is the smallest allowed reflection in a bcc structure .
thus @xmath19 must be larger than @xmath83 a.u.@xmath69 so that the cbed discs overlap and @xmath9 is not empty . integrating the distribution of the magnetic signal over a circular aperture of diameter @xmath84 mrad we indeed observe a zero magnetic signal for @xmath85 a.u.@xmath69 and nonzero above , fig .
[ fig : qmaxdep ] .
note that the sign of emcd signal is not fixed by oam of the beam and direction of the magnetic moment .
particularly , for sample thickness of 30 nm the emcd signal is negative for a rather wide range of @xmath19 values .
this is not surprising , when one considers the origin of the vortex induced emcd , as discussed in the main text .
dynamical diffraction effects , resulting in pendellsung oscillations , determine the relative size and _ sign _ of the factors @xmath15 .
sign of this factor depends on the thickness of the sample and on extinction distance of the @xmath8 beam .
thus one can not _ a priori _ interpret the sign of observed emcd in terms of direction of the magnetic moment and oam of the beam alone . | the calculations presented here reveal that an electron probe carrying orbital angular momentum is just a particular case of a wider class of electron beams that can be used to measure electron magnetic circular dichroism ( emcd ) with atomic resolution .
it is possible to obtain an emcd signal with atomic resolution by simply breaking the symmetry of the electron probe phase distribution using the aberration - corrected optics of an scanning transmission electron microscope .
the required phase distribution of the probe depends on the magnetic symmetry and crystal structure of the sample .
the calculations indicate that emcd signals utilizing the phase of the electron probe are as strong as those obtained by nanodiffraction methods .
development of quantitative magnetic characterization techniques goes hand - in - hand with progress in nano - technology .
a terabit per square inch recording density @xcite means that the area available for one bit is not larger than a square of size @xmath0 nm@xmath1 , assuming bits arranged laterally .
this pushes demands for magnetic measurements down to few nm scale @xcite , approaching atomic resolution .
an attractive option to measure magnetism at such high spatial resolutions is an experimental technique based on electron magnetic circular dichroism @xcite ( emcd ) .
particularly , a great promise came recently from utilizing electron vortex beams @xcite ( evbs ) within an electron microscope .
with evbs it should be possible to measure emcd in the direction of the transmitted beam @xcite , which brings a substantial increase in signal to noise ratio compared to intrinsic emcd measured in between bragg spots @xcite .
however , obtaining isolated atomic - size evbs that can be used for emcd measurements have not yet been possible , although different electron optical setups have been proposed @xcite . in this letter , we show how emcd signals can be measured with atomic resolution in the electron microscope at the transmitted beam without the necessity of producing electron probes carrying orbital angular momentum ( oam ) .
the calculations presented here reveal that evbs carrying oam are just a particular case of a wider class of electron beams that can be used to measure emcd signals .
the key feature to obtain magnetic dichroism with atomic resolution in an electron microscope is the relation between the crystal structure and magnetic symmetry of the sample , and the distribution of the phase in the electron beam .
the calculations indicate that the strength of the emcd signal is only about half of what it was reported in the first emcd experiment on an iron crystal using a parallel beam @xcite , but with the main difference that it achieves atomic spatial resolution . in consequence ,
electron beams that can be obtained by aberration - corrected scanning transmission electron microscopes ( stems ) without additional apertures are predicted to lead to a nonzero emcd signal at transmitted beam .
-vector is marked , together with its mirror image @xmath2 .
the mirror axis is marked as the dashed horizontal line.,width=325 ] the theoretical prediction is based in a two - beam condition model for a convergent beam electron diffraction ( cbed ) in the stem , as shown schematically in fig .
[ fig : scheme ] .
the two disks represent a transmitted beam and elastically scattered beam with bragg vector @xmath3 . for simplicity
the model assumes only one bragg - scattered beam , a situation with a single symmetry plane _ i.e. _ , the @xmath4-axis .
however , the results obtained here can easily be generalized to a situation with more bragg - scattered beams and different symmetries .
overlap of the two cbed disks means that two regions describing the elastically scattered beam wavefunction need to be considered .
region @xmath5 is such that for wavevectors @xmath6 there is no other wavevector _ within the same transmitted disk _ , which would differ from @xmath7 by @xmath8 .
contrary to @xmath5 , the shaded lens - shaped region @xmath9 in fig .
[ fig : scheme ] contains the wavevectors @xmath7 , for which @xmath10 lies within the same transmitted disk .
the reasons for this distinction will become obvious below . here however , we note that it is related to the necessary condition for atomic resolution in stem , which requires overlap of discs to achieve a coherent interference of beam components @xcite . under these assumptions
the complete wavefunction of the elastically scattered incoming probe can be written as @xmath11 \nonumber \\ & + & \sum_{\mathbf{k } \in \omega_2 } ce^{i\phi_\mathbf{k+g}}e^{i(\mathbf{k+g})\cdot\mathbf{r } } \left [ 1 + i t_\mathbf{g } e^{i\mathbf{g}\cdot\mathbf{r } } \right],\end{aligned}\ ] ] where @xmath12 is a real - valued normalization constant .
bragg - scattered beam is phase shifted by @xmath13 and thus its relative amplitude can be written as @xmath14 with real - valued @xmath15 . both @xmath12 and @xmath15
are assumed to be @xmath7-independent , which is a good approximation for thin samples usually studied in aberration - corrected stem and spectrum imaging experiments @xcite .
the @xmath16 represents the phase of the beam component with wavevector @xmath7 .
nonzero @xmath16 can originate for example from aberrations or probe displacement . for an evb with oam @xmath17 one can write @xmath18 .
the radius of the cbed disks is @xmath19 which is related to the convergence semiangle @xmath20 via @xmath21 , where @xmath22 is the de broglie wavelength of electrons accelerated by voltage @xmath23 . for the outgoing wave
the elastic scattering of the probe will be neglected .
additionally , the detector will be considered to be far away , observing a single plane - wave @xmath24 .
the double - differential scattering cross - section can be then evaluated as ( see supplementary information ) : @xmath25 \big ] \nonumber \\ & + & \sum_{\mathbf{k } \in \omega_2 } c^2 \big [ [ 1 + 2t_{\mathbf{g}}\sin(\delta\phi_{\mathbf{k , g}})]s(\mathbf{q - g},\mathbf{q - g},e ) \nonumber \\ & + & t^2_{\mathbf{g}}s(\mathbf{q-2g},\mathbf{q-2g},e ) \nonumber \\ & + & 2t_\mathbf{g } \mathrm{im}[s(\mathbf{q - g},\mathbf{q-2g},e ) ] \nonumber \\ & + & 2\mathrm{re}[e^{-i\delta\phi_{\mathbf{k , g } } } s(\mathbf{q},\mathbf{q - g},e ) ] \nonumber \\ & + & 2t_{\mathbf{g } } \mathrm{im}[e^{-i\delta\phi_{\mathbf{k , g } } } s(\mathbf{q},\mathbf{q-2g},e ) ] \nonumber \\ & + & 2t_{\mathbf{g}}^2 \mathrm{re}[e^{-i\delta\phi_{\mathbf{k , g } } } s(\mathbf{q - g},\mathbf{q-2g},e ) ] \big],\end{aligned}\ ] ] where @xmath26 , and @xmath27 is the mixed dynamical form - factor ( mdff ) , with momentum transfer @xmath28 carrying the @xmath7-dependence of the terms in the sum in eq .
( [ eq : ddscs ] ) .
the @xmath29 denote initial and final states of crystal of energy @xmath30 , respectively .
note that the coulomb factors have been included directly into the definition of the mdff .
the magnetic signal in the electron scattering originates from the imaginary part of mdff according to the dipole model @xcite @xmath31 where @xmath32 stands for isotropic non - magnetic white - line component and @xmath33 is a vector representing the magnetic component .
assuming magnetization only along @xmath34 direction and having @xmath3 one obtains @xmath35 = \frac{(n - m)gm_z(e)q_y}{|\mathbf{q}-m\mathbf{g}|^2|\mathbf{q}-n\mathbf{g}|^2}.\ ] ] in the simplest case , when elastic scattering is neglected ( @xmath36 ) and the convergence angle is small , _
i.e. _ , @xmath37 and thus @xmath9 is empty , eq .
( [ eq : ddscs ] ) reduces to a simple sum of dynamical form - factors @xmath38 . in other words ,
eq .
( [ eq : ddscs ] ) becomes an incoherent summation over all components @xmath7 of the convergent electron probe .
no magnetic signal can arise from such condition because the dynamical form - factor is real [ see eq .
( [ eq : mdffmz ] ) for @xmath39 .
notice that emcd is defined as the result of subtracting two sets of electron energy - loss spectra collected with different electron phases @xcite .
if there is not a magnetic signal in the inelastic scattering , there would not be an emcd signal either .
assuming non - negligible elastic scattering ( nonzero @xmath15 ) for a probe with a small enough convergence angle , so that there is no overlap of the diffracted discs ( empty @xmath9 , see fig .
[ fig : scheme]a ) , the eq .
( [ eq : ddscs ] ) reduces to the first sum only .
the third term in the first sum explicitly contains an imaginary part of mdff , _
i.e. _ , one can expect a magnetic signal to be present at some scattering angles .
however , there is no dependence of the scattering cross - section on the phase @xmath16 , which means that the oam , or in fact , any @xmath7-space distribution of the phase in the probe does not matter . in other words , if the convergence angle is small enough such that @xmath37 , the beam vorticity does not influence the inelastic scattering cross - section . yet , one can still observe a magnetic signal in the setting described above .
however , the distribution of the magnetic signal is antisymmetric with respect to the mirror axes .
because the mirror axes necessarily pass through the transmitted beam , an emcd signal can not be observed by a detector centered on the transmitted beam , regardless of how large the collection angle is .
an example of a cbed diffraction pattern calculated for an evb with @xmath37 is shown below in fig .
[ fig : qmaxmaps ] , left column .
the proof of an antisymmetry of emcd signal proceeds in the following way : let s consider a wavevector @xmath40 , which is a mirror image of the wavevector @xmath41 , see fig .
[ fig : scheme]a .
their combined contribution to the magnetic signal at @xmath42 is evaluated using eqn .
( [ eq : mdffmz ] ) as @xmath43,\ ] ] where @xmath44 and @xmath45 . moving the detector orientation to its mirror image @xmath46 leads to @xmath47 and similarly for @xmath48 .
the @xmath49 and @xmath50 swap and change sign , _
i.e. _ , the magnetic signal in eqn .
( [ eq : add ] ) changes sign as well .
this holds true for all @xmath7 from the lower half - circles of the cbed disks , thus an emcd signal is indeed distributed _ antisymmetrically _ with respect to the mirror axis . in particular , it vanishes right at the symmetry axis .
all observations of an emcd signal done so far , possibly except for ref . , are of this nature so called intrinsic emcd , caused by coherence of elastically scattered beam components @xcite .
-edge energy - filtered diffraction pattern of an evb scattering on a single fe atom .
the individual panels show the non - magnetic and magnetic components ( evbs with oam @xmath51 ) of the inelastic scattering cross - section assuming the fe magnetic moment along the @xmath34-axis .
the emcd signal results from subtracting the oam @xmath52 from oam @xmath53 magnetic components.,width=325 ]
next we consider a situation with no elastic scattering ( @xmath36 ) , but @xmath54 .
this can occur either for ultra - thin samples and sufficiently large convergence angles , or when the unit cell is large ( _ i.e. _ , the reciprocal lattice vectors @xmath55 are small ) .
a limiting case is a single atom in a cell with infinite lattice constant .
then for any nonzero @xmath20 the @xmath56 will be larger than @xmath57 .
the inelastic scattering cross - section can be written as @xmath58.\end{aligned}\ ] ] there are two key findings : 1 ) the scattering cross - section depends on the distribution of the phase in the beam wave front , 2 ) an emcd signal can be observed , despite that the elastic scattering of the probe was neglected .
this time , however , the imaginary part of mdff is multiplied by a @xmath59 function of @xmath60 .
a combined contribution of the probe component @xmath7 and its mirror image @xmath2 is @xmath61.\ ] ] note that for a vortex beam passing directly through an atomic column @xmath62 , which allows to take the @xmath59 function out of the brackets together with a change of the plus sign into a minus sign in between the two terms . moving the detector from @xmath63 to its mirror image @xmath64 transforms the momentum transfer vectors as in eqns .
( [ eq : qqpr ] ) and ( [ eq : qprq ] ) both two terms change sign and then swap their order .
but because of the minus sign in between them , the resulting contribution is the same at both detector orientations .
this inelastic electron diffraction situation is very different from the previous case , because here there is a _
symmetric _ distribution of emcd with respect to the mirror axis .
a symmetric distribution of the magnetic signal allows to detect emcd at the transmitted beam .
this result is in agreement with predictions of refs . and recent simulations @xcite .
the prediction can be illustrated by a simulation of the distributions of the non - magnetic and magnetic contributions to the scattering cross - section of a single fe atom , fig .
[ fig : singlefe ] . the fe atom was placed in a cell of size @xmath65 nm@xmath66 .
the simulations were done using a combined multislice / bloch - waves approach described in ref . , with the outgoing beam described as a single plane wave and the incoming beam was an evb with @xmath67 and @xmath68 a.u.@xmath69 , which at @xmath70 kv means @xmath71 mrad . direction .
magnetic signal components of fe-@xmath72 diffraction patterns for oam of @xmath73 are shown in the middle rows .
( bottom row ) difference of magnetic signals for the two vorticities with oam of @xmath73 .
, width=325 ] the calculations indicate a symmetry of the distribution of magnetic signal in diffraction plane , which is a consequence of @xmath62 for all pairs of @xmath7 and @xmath2 connected by a mirror symmetry . if instead @xmath74 , the imaginary part term in eqn . (
[ eq : noelast ] ) would not change sign and the resulting distribution of emcd would be antisymmetric with respect to the mirror axis .
thus the key element here to detect an emcd signal with atomic resolution is the @xmath7-space distribution of the phase @xmath16 in the electron probe .
basically , in stem one simply needs to set a phase distribution of the electron probe that maximizes the symmetric component of emcd in the diffraction plane , which happens when the phase differences @xmath60 are antisymmetric .
note that shifting the stem probe from the atomic column by @xmath75 introduces a phase factor @xmath76 , which modifies the phase distribution and , in general , also its symmetry . in consequence , the emcd signal intensity
is reduced @xcite if the electron probe is not at the center of an atomic column ; see also the supplementary information . for a single atom
, there is a continuum of mirror axes passing through the atom . as a consequence ,
the optimal beam shape for emcd is a vortex beam passing through the atom because it has an antisymmetric phase difference with respect to all mirror axes passing through its core . for a crystal with a discrete set of mirror symmetries
there is a wider range of phase distributions in the probe wave - front that are antisymmetric with respect to all mirror axes .
this is illustrated below , where all the terms of eq .
( [ eq : ddscs ] ) are considered .
when considering both an overlap of cbed disks ( @xmath54 ) and elastic scattering of the incoming electron probe ( @xmath77 ) , the inelastic scattering cross - section contains several terms with imaginary part of mdff .
some of the terms are strictly antisymmetric and do not depend on @xmath60 , but other terms depend on a phase difference via the real and imaginary parts of the phase factor , @xmath78 , multiplying them .
thus an optimum phase distribution in the wave front may be rather complicated and will depend on the particular crystal structure and magnetic symmetry via dynamical diffraction effects . with evbs ,
when imaginary part of mdff is multiplied by the real ( imaginary ) part of the phase factor , @xmath78 , it leads to an antisymmetric ( symmetric ) emcd distribution , respectively .
it is interesting to observe how the symmetric contribution develops as the convergence angle increases .
this is illustrated in figs .
[ fig : qmaxmaps ] and in supplementary fig . 1 for bcc iron crystal with @xmath79 , beam direction along @xmath80 zone axis , and using an acceleration voltage of 200 kv .
the overlap onsets when @xmath81 , where @xmath82 is the smallest allowed reflection in a bcc structure .
thus @xmath19 must be larger than @xmath83 a.u.@xmath69 so that the cbed discs will overlap and @xmath9 becomes non - empty . integrating the distribution of the magnetic signal over a circular aperture of diameter @xmath84 mrad leads to a zero magnetic signal for @xmath85 a.u.@xmath69 and nonzero above . clearly , when the cbed discs do not overlap ( fig .
[ fig : qmaxmaps ] , left column ) , the emcd is antisymmetrically distributed with respect to all mirror axes
horizontal , vertical and two diagonal ones as anticipated .
above the onset of overlap this antisymmetry is broken .
mm ; @xmath86 a.u.@xmath69 , @xmath70 kv .
b ) non - magnetic and c ) magnetic component of the energy - filtered diffraction pattern at @xmath72 edge of iron , @xmath87.,width=325 ] note that for a discrete set of mirror symmetries , as in the case of a bcc crystal , it is easy to construct a beam with a phase distribution , which has antisymmetric phase differences @xmath60 with respect to all of the four mirror axes , but which is not a vortex beam carrying oam .
for example , a four - fold astigmatism @xcite has the required symmetry , because for any @xmath7 the @xmath16 changes sign under mirror symmetry , see fig .
[ fig : star]a . an explicit calculation of the inelastic electron scattering reveals an emcd signal distribution , which is indeed not antisymmetric .
the fe l@xmath88 signal for bcc iron integrated over a finite collection angle centered on a transmitted beam presents an emcd of relative intensity of about 1% .
the strength of this emcd signal is only about half of what it was reported in the first emcd experiment @xcite , but with the main difference that it has atomic spatial resolution .
moreover , with this approach one can use the full intensity of the electron beam because there is no need of a spiral @xcite or fork aperture @xcite , which should result in better signal to noise ratio emcd measurements than with evbs @xcite . in conclusion , we show that a finite emcd signal with atomic resolution can be observed at the transmitted beam without the necessity of using evbs carrying oam .
the conditions necessary for emcd are : 1 ) convergence angle large enough to cause an overlap of cbed disks and , 2 ) a phase distribution of the probe in reciprocal space that is not invariant under mirror symmetries of the crystal . as a corollary we propose a simple beam shape , which can be readily obtained in current aberration - corrected stem instruments , and which should allow to measure emcd at the transmitted beam . the emcd signal obtained with the method presented can achieve atomic resolution and be as strong as emcd nanodiffraction experiments . in a more general context
, our work opens up new ways of utilizing aberration - correction electron optics to design atomic - size electron probes with tailored phase distributions for specific applications and/or crystal symmetries .
these electron probes could be utilized to probe magnetic dichroism , optical dichroism , and the valley polarization of materials with unmatched spatial resolution .
j.r . acknowledges swedish research council and swedish national infrastructure for computing ( nsc center ) .
j .-
c.i . acknowledges support by the center for nanophase materials sciences ( cnms ) , which is sponsored at oak ridge national laboratory by the scientific user facilities division , office of basic energy sciences , u.s .
department of energy .
comments by peter m. oppeneer and by the referee during the review process of this manuscript are gratefully acknowledged . |
* rheological hysteresis + in soft glassy materials * five movies are provided as supplemental material to show the full data sets corresponding to figs . 1 and 2 in the main text .
the files names indicate the system under investigation ( laponite suspension or carbopol microgel ) and the value of the waiting time per point @xmath1 .
velocity profiles are shown only for @xmath54 s@xmath10 due to the limitation in the ultrasonic pulse repetition frequency to 20 khz , which sets an upper bound on the measurable velocities @xcite .
at very small shear rates noise becomes too large for velocities to remain reliable .
indeed , the amount of velocity data recorded at each shear - rate step of fixed duration @xmath1 increases with @xmath3 @xcite . therefore , at small @xmath3 , the lower statistics result in significant noise in the velocity profiles . in the movies , the data corresponding to the downward ( upward resp .
) sweep are shown in red ( blue resp . ) .
the velocities profiles measured during the downward sweep are replotted with smaller symbols together with the upward sweep data , in order to allow for a direct comparison of the two velocity profiles obtained at the same shear rate .
black squares at @xmath55 and @xmath56 mm indicate respectively the rotor velocity ( @xmath57 ) and the stator velocity ( @xmath40 ) .
figure [ suppfig.1 ] shows the curves @xmath5 vs @xmath1 obtained on a laponite suspension for different values of @xmath58 , the number of measurement points per decade in the shear rate sweeps .
the data all nicely collapse on a single curve when plotted as a function of @xmath29 . in the experiments reported in the main text @xmath58
is fixed arbitrarily to 15 and only @xmath1 is varied . the collapse shown in fig .
[ suppfig.1](b ) indicates that the true control parameter is @xmath29 .
therefore , when interpreting the time interval @xmath59 at which @xmath5 reaches a maximum , one has to keep in mind that the corresponding physical timescale is rather @xmath60 s in the case of fig .
[ suppfig.1 ] .
we checked for the influence of the boundary conditions on both observables @xmath5 and @xmath6 .
figure [ suppfig.2 ] reports results obtained on two different laponite suspensions with smooth and rough boundary conditions .
`` smooth '' walls refer to polished plexiglas , as used throughout the main text .
the typical roughness of smooth walls is 15 nm .
`` rough '' boundary conditions refer to a sandblasted plexiglas couette cell ( roughness of about 1 @xmath20 m ) with the same dimensions as the smooth cell .
we could not detect any systematic influence of the boundary conditions on @xmath5 and @xmath6 .
in particular , the positions of the maxima @xmath59 in fig .
[ suppfig.2](a ) and ( b ) are not affected by the wall roughness although the values of the hysteresis area @xmath5 can be up to 30% larger or smaller over the whole range of investigated @xmath33 depending on the boundary conditions . in fig .
[ suppfig.2](c ) , @xmath6 is seen to be about twice as large in the rough cell as in the smooth cell .
a close inspection of the velocity profiles indicates that the material tends to stick at the stator in a rough cell so that fully arrested profiles ( with @xmath40 everywhere ) persist for longer times in the upward shear rate sweeps , leading to larger values of @xmath6 . the larger scatter in the @xmath6 data of fig .
[ suppfig.2](d ) for rough boundary conditions also seems to arise from more erratic slip phenomena than in smooth conditions .
still the general evolution of the velocity profiles during shear - rate sweeps with rough walls remains otherwise very close to that observed with smooth walls .
laponite suspensions are known to exhibit strong aging properties due to slow chemical and physical processes @xcite .
we tested the influence of the sample age on rheological hysteresis by preparing a laponite suspension that was left at rest for a time @xmath61 before part of the sample was loaded into the smooth couette cell and submitted to series of shear rate sweeps as described in the main text .
figure [ suppfig.3 ] confirms that aging processes over more than one month do not significantly affect the quantitative behaviour of @xmath5 . | the nonlinear rheology of a soft glassy material is captured by its constitutive relation , shear stress vs shear rate , which is most generally obtained by sweeping up or down the shear rate over a finite temporal window . for a huge amount of complex fluids ,
the up and down sweeps do not superimpose and define a rheological hysteresis loop . by means of extensive rheometry
coupled to time - resolved velocimetry , we unravel the local scenario involved in rheological hysteresis for various types of well - studied soft materials .
we introduce two observables that quantify the hysteresis in macroscopic rheology and local velocimetry respectively , as a function of the sweep rate @xmath0 .
strikingly , both observables present a robust maximum with @xmath1 , which defines a single material - dependent timescale that grows continuously from vanishingly small values in simple yield stress fluids to large values for strongly time - dependent materials . in line with recent theoretical arguments , these experimental results hint at a universal timescale - based framework for soft glassy materials , where inhomogeneous flows characterized by shear bands and/or pluglike flow play a central role .
when submitted to external stress , soft glassy materials such as colloidal gels , clay suspensions , concentrated emulsions , and foams , display a fascinating variety of behaviors because the applied strain may disrupt and rearrange the microstructure over a wide range of spatial and temporal scales leading to heterogeneous flow properties @xcite . for more than a decade ,
flow dynamics have been probed by combining standard rheology , e.g. through the determination of the `` constitutive relation '' between the shear stress @xmath2 and the shear rate @xmath3 , and local structural or velocity measurements @xcite .
while much progress has been made on _ steady - state _ local flow properties @xcite , _ transient _ regimes upon shear start - up have been addressed only recently @xcite .
still , in practice , it can be argued that any experimental determination of the flow curve @xmath4 is effectively transient since it is obtained by sweeping up or down @xmath3 over a finite temporal window . in other words
the measured flow curve coincides with the steady - state relation @xmath4 _ only if _ the sweep rate is slow enough compared to any intrinsic timescale of the fluid .
otherwise one expects hysteresis loops in @xmath4 measurements performed by sweeping up then down the shear rate ( or vice versa ) .
this phenomenon , known as `` rheological hysteresis , '' has been commonly observed in a host of complex fluids for about 70 years @xcite .
however , to date , this ubiquitous signature of the interplay between timescales in complex fluids has not been quantitatively studied by means of local measurements . in this letter , we use time - resolved velocimetry to unveil the local scenario involved in rheological hysteresis in various types of well - studied soft materials . building upon a systematic experimental protocol
, we introduce two observables , @xmath5 and @xmath6 , that quantify the amplitude of the hysteresis phenomenon as a function of the sweep rate @xmath0 in macroscopic rheology and local velocity respectively . both @xmath5 and @xmath6
go through a maximum with @xmath1 , pointing to the existence of a characteristic timescale @xmath7 for the microstructure dynamics . in thixotropic ( laponite ) suspensions and ( carbon black ) gels , @xmath7 is of the order of several hundreds of seconds , while it becomes hardly measurable for simple yield stress fluids such as carbopol and concentrated emulsions .
velocity profiles allow us to understand this evolution by clearly differentiating a succession of homogeneous , shear - banded , and arrested flows depending on the fluid and on the sweep rate , thus providing a local interpretation of rheological hysteresis . _
experimental set - up and protocol.- _ experiments are performed in a polished plexiglas couette geometry ( typical roughness 15 nm , height 28 mm , rotating inner cylinder of radius 24 mm , fixed outer cylinder of radius 25 mm , gap @xmath8 mm ) equipped with a home - made lid to minimize evaporation .
rheological data are recorded with a stress - controlled rheometer ( mcr 301 , anton paar ) .
two flow curves are successively recorded , first by decreasing the shear rate @xmath3 from high shear ( @xmath9 s@xmath10 ) to low shear ( @xmath11 s@xmath10 ) through @xmath12 successive logarithmically - spaced steps of duration @xmath1 each , and then by immediately increasing @xmath3 back from @xmath13 up to the initial value @xmath14 following the same @xmath15 steps in reverse order . in general the downward and upward flow curves , @xmath16 and @xmath17
do not coincide and define a _
hysteresis loop_. simultaneously to the flow curves , the azimuthal velocity @xmath18 is measured as a function of the radial distance @xmath19 to the rotor , at about 15 mm from the cell bottom , and with a spatial resolution of 40 @xmath20 m by means of ultrasonic velocimetry @xcite .
velocity data are then averaged over the duration @xmath1 of each shear - rate step at @xmath3 and the corresponding velocity profiles @xmath21 and @xmath22 are normalized by the rotor velocity @xmath23 to allow for a direct comparison of flow properties at widely different shear rates @xcite .
we checked that at @xmath14 velocity profiles are all linear with no fluctuation so that the flow reaches a steady state within the time interval @xmath1 , even for the fastest sweep rates .
therefore starting from high enough shear rates ensures a well - defined and reproducible initial condition .
our choice of @xmath13 and @xmath15 results from a compromise between good sampling and reasonable total duration . by monitoring the viscoelastic moduli prior to and after each experiment
, we checked that evaporation and/or irreversible ( chemical ) aging of the sample were negligible even for the slowest sweeps ( @xmath24 s , @xmath25 h ) .
finally , approximating our step - like protocol by a continuous sweep , the equivalent sweep rate is @xmath26 , where @xmath27 is the number of steps per decade . in the present protocol
we fix @xmath28 and identify the sweep rate with @xmath0 , while keeping in mind that @xmath29 is the actual control parameter ( see fig . 1 in supplemental material ) .
laponite suspension.- _ clay suspensions are well known to exhibit large rheological hysteresis @xcite . here
we focus on laponite samples prepared by mixing ultrapure water with 3% wt . of laponite powder ( rockwood , grade rd ) and 0.3%
wt . of hollow glass spheres ( sphericel , potters ) acting as acoustic contrast agents @xcite .
figure [ fig.1 ] shows the hysteresis loops for @xmath30 , 10 , and 100 s. to quantify their areas , we introduce the following observable : @xmath31 where @xmath32 . note that the logarithmic sampling of @xmath33 gives an equal weight to low and high shear rates .
strikingly , when the sweep rate is decreased , i.e. when @xmath1 is increased , @xmath34 goes through a maximum at @xmath35 s [ fig .
[ fig.1](d ) ] . to uncover the local scenario underlying this non - monotonic behavior of @xmath34
, we turn to time - resolved velocity profiles . in a way similar to eq .
( [ eq.asigma ] ) , we consider @xmath36 , and integrate it over the gap and then over the range of explored shear rates : @xmath37 as shown in fig . [ fig.1](e ) , the evolution of this local observable @xmath6 with @xmath1 directly reflects that of the area @xmath5 extracted from the sole global rheology and its maximum is reached for a similar value of @xmath35 s. this suggests that the behavior of @xmath34 results from the bulk flow properties at the mesoscopic scale . more precisely , for small values of @xmath1 , the laponite suspension is quickly `` quenched '' from high shear rates to lower ones .
velocity profiles ( see also supplementary movies ) are all linear both on the way down and on the way up even at the smallest shear rates [ fig .
[ fig.1](a ) ] : the laponite suspension is not given enough time to restructure and remains fluid throughout the whole cycle , which leads to vanishingly small values of @xmath6 .
for intermediate values of @xmath1 , the velocity profiles along the two flow curves now strongly differ [ fig .
[ fig.1](b ) ] : while they remain linear during the downward sweep , they become inhomogeneous during the upward sweep and exhibit shear banding over a large range of shear rates . here , the suspension is given enough time to restructure leading to an arrested band subject to physical aging close to the fixed wall at @xmath38 , that progressively disappears as the shear rate is increased .
this contributes to increase @xmath6 .
finally , for large values of @xmath1 , flow arrest also occurs along the downward sweep , which tends to decrease @xmath6 [ fig .
[ fig.1](c ) ] : shear banding is observed for @xmath391 s@xmath10 until the system experiences total slippage at the rotor ( i.e. @xmath40 everywhere in the bulk ) for @xmath41 s@xmath10 .
this fully arrested state persists on the upward sweep up to much higher shear rates ( @xmath42 s@xmath10 ) and gives way to a homogeneously sheared state above a small interval of shear rates that corresponds to decreasing shear stresses and to inhomogeneous flows .
this scenario is robust and does not significantly depend on the boundary conditions or on sample age since preparation ( see figs . 2 and 3 in supplemental material ) .
as discussed below , we suggest that the characteristic time @xmath43 s results from the competition between physical aging ( restructuration ) and shear rejuvenation ( structure break up ) @xcite .
_ carbopol microgel.- _ let us turn to a simple yield stress fluid , namely a carbopol microgel , where restructuration is expected to be fast @xcite .
results are reported in fig .
[ fig.2 ] .
the global observable @xmath5 is of the same order as for the laponite suspension @xcite .
still , in the case of carbopol , @xmath5 is monotonically decreasing over the range of explored @xmath1 [ fig .
[ fig.2](c ) ] . here
again the local observable @xmath6 follows the exact same trend [ fig .
[ fig.2](d ) ] .
velocity profiles reveal that the local scenario remains qualitatively the same whatever the sweep rate : homogeneous shear flow is observed along most of the downward sweep together with an ever - increasing amount of wall slip . for @xmath44 s@xmath10 the local shear rate vanishes and the flow becomes pluglike at roughly half the rotor velocity . on the way up ,
pluglike flow gives way to shear - banded profiles right after the stress maximum and eventually to linear profiles at high shear rates [ see insets in fig .
[ fig.2](a , b ) and movies in the supplemental material ] , consistently with previous reports on startup experiments at constant shear rate @xcite .
however , as @xmath1 is increased , the range of shear rates over which shear banding is observed gets narrower , leading to smaller values of @xmath6 .
_ discussion.- _ to further establish the generality of our results we explore a broader range of soft glassy materials in fig .
[ fig.3 ] . for three different concentrations , thixotropic laponite suspensions [ fig .
[ fig.3](a ) ] and carbon black gels [ fig .
[ fig.3](b ) ] display a bell - shaped @xmath5 .
the case of a concentrated emulsion [ commercial mayonnaise , fig .
[ fig.3](c ) ] seems intermediate between carbopol and the two previous materials : the loop area @xmath5 exhibits a maximum at the faster end of the accessible range of sweep rates corresponding to @xmath45 s i.e. @xmath46 s. these bell - shaped curves are well captured by lognormal fits , leading to the master curve shown in fig [ fig.3](d ) . the above results hint at a unified picture in which the behaviors of both @xmath5 and @xmath6 are interpreted in terms of a single characteristic timescale @xmath47 . for @xmath48 ,
the bulk material remains fluidized during the downward sweep while flow heterogeneity during the upward sweep keeps increasing with @xmath29 [ fig .
[ fig.1](a , b ) ] , leading to increasing @xmath6 and @xmath5 .
note that @xmath6 displays a sharper increase than @xmath5 .
the lower sensitivity of @xmath5 may be attributed to microstructural changes in the _ fluidized _ suspension ( e.g. local fluctuations in the concentration or in the size of colloidal aggregates ) responsible for differences in viscosity only , so that downward and upward flow curves may not superimpose ( @xmath49 ) although velocity profiles are all linear ( @xmath50 ) . making a rigorous connection between @xmath5 and @xmath6 from theoretical approaches and/or material - dependent
modeling is a challenging task that must be addressed in the future together with time - resolved structural characterization during the sweeps . as for the maximum of @xmath5 and @xmath6 ,
velocity data show that @xmath7 corresponds to the point where flow heterogeneity first appears during the downward sweep .
@xmath51 can thus be interpreted as the sweep rate for which restructuration balances shear - induced fluidization over the cycle . for @xmath52
the decrease of @xmath5 is easily explained if one notices that the shear stress becomes independent of the sweep direction as soon as fully arrested states are reached [ figs .
[ fig.1](c ) and [ fig.2](a , b ) ] . whatever the material , arrested states span over larger portions of the flow curve as @xmath29 is increased at the expense of transient shear - banded flows .
this tends to decrease the low shear rate contribution to the loop area , hence the decreasing behavior of @xmath5 and @xmath6 . in other words ,
the slower the sweep rate , the closer the flow gets to its steady state , and thus the smaller the areas @xmath5 and @xmath6 as expected from first intuition @xcite .
we emphasize that wall slip , i.e. the presence of lubricating layers at the boundaries , does not seem to bring any significant contribution to @xmath5 and @xmath6 . indeed using roughened walls does not affect the shape of the areas nor the value of @xmath7 ( see supplementary figure 2 ) .
total slippage thus appears as a mere consequence of flow arrest in the bulk and the nonmonotonic behavior of @xmath5 and @xmath6 originates only from the bulk flow heterogeneity .
still potential effects of the flow geometry on the shape of the hysteresis loops and/or on @xmath7 should be investigated .
interestingly , in both carbopol microgels and laponite suspensions , inhomogeneous shear - banded flows are observed concomitantly with sections of the upward flow curves where @xmath2 presents a plateau [ fig .
[ fig.1](b ) ] or decreases with increasing @xmath33 [ fig .
[ fig.1](c ) and [ fig.2](a , b ) ] .
this is in agreement with a universal criterion proposed recently in ref .
@xcite which predicts transient shear banding to arise just after any stress overshoot during startup .
finally fig .
[ fig.3 ] suggests that the characteristic time @xmath7 continuously grows when going from simple yield stress fluids ( carbopol , emulsions ) to highly time - dependent materials ( laponite , carbon black ) . in carbopol microgels @xmath7
is too short to be measured and only the decreasing parts of @xmath5 and @xmath6 are observed . for thixotropic materials @xmath7
could be linked to the history - dependent timescales for structure breakdown and build - up inferred from shear rate jumps @xcite . however , here , @xmath7 is protocol - independent as long as @xmath13 and @xmath14 are well separated .
we may therefore speculate that it relates to the intrinsic material `` restructuring time '' invoked in recent theoretical arguments and numerical simulations @xcite , e.g. , the ( very short ) duration of rearrangement events in simple yield stress fluids ( known as t1 events in emulsions and foams @xcite ) and the ( much longer ) aggregation time of colloidal particles in other systems @xcite .
the observed decrease of @xmath7 with laponite concentration would thus result from faster aggregation dynamics while the inverse trend in carbon black gels could be attributed to sedimentation of large clusters which gets slower as the concentration is increased . to conclude our study provides experimentalists with observables that quantify the distance to steady state in measurements of the constitutive relation : whatever the soft glassy system , sweep rates much larger than @xmath51 should be used to ensure that effects of long - lived transients and inhomogeneous shear - banded or arrested flows are minimized .
exploring the hysteresis maximum through faster sweeps provides interesting physical insights on how local flow properties drive rheological hysteresis and on the type of material , with an apparently continuous transition from strong thixotropy ( large @xmath7 ) to simple yield stress fluids ( vanishingly small @xmath7 ) .
clearly , spatially - resolved models are needed to further explain and complete our experimental findings .
soft - glassy rheology approaches @xcite and local fluidity models @xcite appear as ideal candidates to check whether rheological hysteresis could be envisioned in a universal framework as suggested by the present work .
we thank y. forterre and v. trappe for providing the carbopol and the carbon black , and c. barentin and g. ovarlez for fruitful discussions .
sm acknowledges funding from the european research council under the european union s seventh framework programme ( fp7/2007 - 2013 ) / erc grant agreement n@xmath53 258803 .
g. ovarlez _
et al . _ , rheol .
acta * 48 * , 831844 ( 2009 ) .
p. schall and m. van hecke , ann .
rev .
fluid mech .
* 42 * , 6788 ( 2010 ) .
p.c.f .
mller _
et al .
_ , phys .
rev .
e * 77 * , 041507 ( 2008 ) .
j. goyon _
et al . _ ,
nature * 454 * , 84 - 87 ( 2008 ) ; g. ovarlez _
et al . _ ,
phys .
rev .
e * 78 * , 036307 ( 2008 ) .
k. n. nordstrom _
et al . _ ,
phys .
rev .
lett . * 105 * , 175701 ( 2010 ) ; v. chikkadi _ et al .
_ , phys .
rev .
lett . * 107 * , 198303 ( 2011 ) ; p. jop _
et al . _ ,
phys .
rev .
lett . * 108 * , 148301 ( 2012 ) t. divoux _
et al . _ ,
phys .
rev .
lett . * 104 * , 208301 ( 2010 ) .
r. l. moorcroft _
et al . _ ,
phys .
rev .
lett . * 106 * , 055502 ( 2011 ) .
j. d. martin and y. t. hu , soft matter , doi : 10.1039/c2sm25299f ( 2012 ) .
m. siebenburger _
et al .
_ , phys .
rev .
lett . * 108 * , 255701 ( 2012 ) .
j. mewis and n.j .
wagner , j. colloid interface sci .
* 147 - 148 * , 214227 ( 2009 ) and references therein .
h. green and r.n .
weltmann , ind .
eng .
chem .
, anal .
ed . * 15 * , 201 - 206 ( 1943 ) ; d.c .- h .
cheng , nature * 245 * , 9395 ( 1973 ) ; l.b .
chen _
et al .
_ , phys .
rev .
lett . * 69 * , 688 - 691 ( 1992 ) ; d. perret _
et al . _ ,
. geol . * 43 * , 3144 ( 1996 ) , f. da cruz _ et al .
_ , phys rev .
e * 66 * , 051305 ( 2002 ) .
s. manneville , l. bcu , and a. colin , eur .
phys .
j. ap * 28 * , 361373 ( 2004 ) .
since @xmath3 is varied by 6 orders of magnitude for each flow curve , the technique described for a fixed shear rate in @xcite was slightly modified : the analog output of the mcr 301 rheometer is used to monitor the rotor velocity in real - time and to adjust the repetition frequency of the ultrasonic pulses to the current value of @xmath3 . the amount of velocity data recorded at each shear - rate step of fixed duration @xmath1 increases with @xmath3 @xcite . at small @xmath3 , the lower statistics result in significant noise in the velocity profiles [ see leftmost insets in figs .
[ fig.1](a , b ) and [ fig.2](b ) ] .
p.c.f .
mller _
et al . _ , phil .
trans .
r. soc .
a * 367 * , 51395155 ( 2009 ) ; t. gibaud _
et al . _ ,
soft matter * 5 * , 30263037 ( 2009 ) .
p. coussot _
et al . _ ,
phys .
rev .
lett . * 88 * , 175501 ( 2002 ) .
p. coussot and g. ovarlez , eur .
phys .
j. e * 33 * , 183188 ( 2010 ) .
t. divoux _
et al . _ , soft matter * 8 * , 41514164 ( 2012 ) .
the apparently smaller hysteresis in carbopol ( fig .
[ fig.2 ] ) than in laponite ( fig .
[ fig.1 ] ) is due to the much larger vertical scale in the flow curves .
r. l. moorcroft and s. m. fielding , preprint arxiv:1201.6259v1 ( 2012 ) .
k. dullaert and j. mewis , j. rheol .
* 49 * , 12131230 ( 2005 ) ; j. colloid interface sci . *
287 * , 542551 ( 2005 ) .
k. martens , l. bocquet and j .- l .
barrat , soft matter * 8 * , 41974205 ( 2012 ) . a .-
l .
biance _
et al . _ ,
soft matter * 5 * , 46724679 ( 2009 ) .
mongondry _ et al .
_ , j. colloid interface sci . * 283 * , 397405 ( 2005 ) .
s.a .
rogers _
et al .
_ , phys .
rev .
lett . * 100 * , 128304 ( 2008 ) . |
this paper reports about research conducted on the edge between algebraic graph theory ( briefly agt ) and computer algebra .
association schemes are one of the traditional areas of investigation in agt . for a good decade ,
catalogues of small association schemes have been available from the web site @xcite .
it is known that all association schemes of order up to 14 are schurian , that is , they are coming from a suitable transitive permutation group in the standard manner .
first examples of non - schurian association schemes exist on 15 , 16 and 18 vertices .
in particular , there are just two classes of non - schurian association schemes of order 18 .
d. pasechnik explained in evident form in @xcite how the non - schurian rank 3 antisymmetric association scheme on 15 points appears , and determined its full automorphism group .
the famous shrikhande graph generates non - schurian rank 3 association schemes on 16 points .
in fact , there are many other non - schurian schemes of order 16 ( see @xcite )
. their clever computer free explanation might be of definite interest .
surprisingly , such an explanation has not been given for the schemes on 18 points yet .
we are here filling this gap , providing an interpretation in terms of finite geometries .
moreover , we are introducing a possible generalisation of these schemes which is leading to four infinite families of non - schurian association schemes on @xmath3 points , where @xmath4 is a prime .
this paper originated from computer aided experiments , which were fulfilled by the author .gy . under the guidance of m.k .
the manner in which the computer search helped to reach the presented results is , in our eyes , of independent interest .
the suggested structure of our paper was designed intentionally not only to reflect the obtained scientific results , but also to pursue additional pedagogical , expository and even philosophical goals .
preliminaries and the used computer tools are briefly discussed in sections 2 and 3 , respectively .
our starting geometrical object is a biaffine coherent configuration @xmath5 , which appears via the intransitive permutation group @xmath6 of order @xmath7 and degree @xmath3 having two orbits of length @xmath8 .
object @xmath5 is considered in sections 48 , along with its four related color graphs @xmath9 , @xmath10 , which turn out to be association schemes . in particular , an outline of a proof that these association schemes are non - schurian is given in section 8 .
the presentation in sections 48 is of a definite geometric nature , based on a consideration of the classical biaffine plane of order @xmath0 with @xmath8 points and @xmath8 lines . in section 9 ,
all discussed structures are developed from the scratch with the aid of an independent second model of @xmath5 , which is of a more algebraic nature .
here our starting structure is the finite heisenberg group of order @xmath7 .
sections 10 and 11 are devoted to the use of algebraic groups of coherent configurations . here
we suggest some innovative methodological elements , which might be of an independent interest for experts in agt .
sections 12 and 13 are aimed to extend the scope of our consideration , paying special attention to some extra features of our results , as well as to a number of well known classes of graphs which appear to be relevant to new association schemes discovered by the authors .
some of the considered graphs play a significant role in extremal graph theory .
section 14 serves as a research announcement of another portion of achieved results regarding new non - schurian association schemes of constant rank 6 and 5 , and some of their properties .
the level of rigour here is different from the main body of the text . in particular
, no attempt is made to provide full formulations and outlines of proofs .
finally , the last section 15 is written in a very specific style , quite typical of other publications co - authored by m.k .
it is , in a sense , a mosaic of diverse topics , implicitly or explicitly related to the content of the paper , however not previously touched in evident form .
special attention is paid to the research paradigm `` computer experiment , plausible reasonings , theoretical proof '' , comparing views , experience , and tastes of the authors with modern trends in science , which are correlated with the exploitation of computer tools .
appendices 13 contain some routine data about the investigated schemes .
these data will play an essential role in portions of the theoretical proofs .
below we provide a brief outline of the most significant concepts that will be used throughout the text . we refer to @xcite and @xcite for a more detailed background . by a _ color graph _
@xmath11 we will mean an ordered pair @xmath12 , where @xmath13 is a set of vertices and @xmath14 a partition of @xmath15 into binary relations .
the elements of @xmath14 will be called _
colors _ , and the number of colors will be the _ rank _ of @xmath11 . in other words ,
a color graph is an edge - colored complete directed graph with loops , whose arcs are colored by the same color if and only if they belong to the same binary relation .
a _ coherent configuration _ is a color graph @xmath16 , @xmath17 , such that the following axioms are satisfied : * the diagonal relation @xmath18 is a union of relations @xmath19 , for a suitable subset @xmath20 .
* for each @xmath21 there exists @xmath22 such that @xmath23 , where @xmath24 is the relation transposed to @xmath25 . * for any @xmath26 , the number @xmath27 of elements @xmath28 such that @xmath29 and @xmath30 is a constant depending only on @xmath31 , and independent on the choice of @xmath32 .
the numbers @xmath27 are called _ intersection numbers _ , or sometimes _ structure constants _ of @xmath33 . assume that @xmath34 , and let us put @xmath35 . to each _ basic graph _
@xmath36 we associate its adjacency matrix @xmath37 .
then the set of _ basic matrices _
@xmath38 may be regarded as a basis of a matrix algebra @xmath39 which contains the identity matrix , the all - ones matrix @xmath40 , and is closed under transposition and schur - hadamard multiplication of matrices .
such an algebra is called a _ coherent algebra _ , and we refer to the set @xmath38 as its _ standard basis_. the concepts of coherent configuration and coherent algebra were introduced by d. higman ( see e.g. @xcite ) .
similar concepts were introduced independently by b.ju .
weisfeiler and a.a .
leman , see @xcite and also @xcite for an historical discussion .
a significant source of coherent configurations appears as follows .
assume that @xmath41 is a permutation group acting on the set @xmath42 .
for @xmath43 the set @xmath44 , where @xmath45 , is called a _
2-orbit _ of @xmath46 , specifically the _ 2-orbit _ of @xmath46 _ corresponding to _ @xmath47 .
( note that when @xmath41 is a transitive permutation group , many authors prefer the term _ orbital _ for this set . ) denoting by 2-@xmath48 the set of 2-orbits of a permutation group @xmath49 , it is easy to check that @xmath502-@xmath51 is a coherent configuration .
coherent configurations that arise in this manner are called _ schurian _ , otherwise we call them _ non - schurian_. an _ association scheme _ @xmath52 ( also called a _ homogeneous coherent configuration _ ) is a coherent configuration in which the diagonal relation @xmath53 belongs to @xmath14 .
thus , schurian association schemes are coming from transitive permutation groups .
a coherent configuration @xmath33 is called _ commutative _ if for all @xmath26 we have @xmath54 .
we call @xmath33 _ symmetric _ if @xmath55 for all @xmath21 .
it is a well known fact that a symmetric coherent configuration is also commutative , but the converse is not true in general . to each coherent configuration @xmath33
we may assign three groups : @xmath56 and @xmath57 .
the ( combinatorial ) _ group of automorphisms _
@xmath58 consists of the permutations @xmath59 which preserve the relations , i.e. @xmath60 for all @xmath61 .
the _ color automorphisms _ are permitted to permute the relations from @xmath14 , i.e. for @xmath59 we have @xmath62 if and only if for all @xmath21 there exists @xmath63 such that @xmath64 .
algebraic automorphism _ is a bijection @xmath65 that satisfies @xmath66 .
it is easy to verify that @xmath58 is a normal subgroup of @xmath67 , and that the quotient group @xmath68 embeds naturally in @xmath57 . for each group @xmath69 of algebraic automorphisms of @xmath16 one
can define an _ algebraic merging _ of @xmath14 in the following way .
let @xmath70 denote the set of orbits of @xmath69 on @xmath14 .
for each @xmath71 define @xmath72 to be the union of all relations from @xmath73 .
then the set of relations @xmath74 forms a coherent configuration on @xmath42
. we will call it an _ algebraic merging _ of @xmath14 with respect to @xmath69 . note that if @xmath75 and @xmath33 is schurian , then the resulting merging is schurian as well .
in contrast , a merging with respect to a subgroup of @xmath69 not contained in @xmath68 may lead to a non - schurian coherent configuration .
it is clear from the definitions that an association scheme @xmath33 is schurian if and only if its rank coincides with the rank of its group of automorphisms @xmath58 .
this project heavily depends on the use of computer tools . moreover ,
in a sense it can serve as a pattern for the use of computers in agt .
indeed , we made a lot of fast computations ; investigated and organized the obtained results ; analyzed all relevant data ; made conjectures based on these data ; checked these conjectures for higher values of parameters of the considered series ; transformed these conjectures to a suitable analytical form ; transformed the obtained results from numerical to symbolic mode ; created pictures and tables ; and so on .
here we are mainly working with coherent configurations and association schemes , as well as with the permutation groups related to them .
for this purpose , in 199092 a computer package was created in moscow as a result of the activities of i.a .
faradev and the author m.k .
this package goes by the name coco , and was introduced in @xcite ; see also @xcite for deeper consideration of the used methodology and algorithms .
coco is still very helpful for performing initial computational experiments .
nevertheless , nowadays the mainstream of our computer aided activities is based on the use of the free software gap @xcite ( groups , algorithms and programming ) , in particular its share package grape @xcite which works in conjunction with nauty @xcite .
in addition , we strongly benefitted from the kind permission granted by sven reichard to use his unpublished package coco iir @xcite ( still under development ) which operates under the gap platform .
the foremost goal of coco iir is to extend the scope of algorithms from coco , relying on many new developments and fresh ideas adopted from modern computer algebra .
last but not least , we are pleased to acknowledge a package of programs for computing with association schemes , written by hanaki and miyamoto .
it is open - source software that may be freely downloaded from the homepage of the authors @xcite .
this package also works under gap , and contains some tools that have not yet been implemented in coco iir .
let @xmath0 be an odd prime , and let @xmath76 be the cyclic group of order @xmath0 . throughout this text ,
the set of nonzero elements in @xmath76 will be denoted by @xmath77 .
take two copies @xmath78 and @xmath79 of @xmath80 .
the first copy @xmath78 is none other than the point set of the classical ( desarguesian ) affine plane of order @xmath0 .
each element @xmath81 may be identified uniquely with a pair of coordinates of the form @xmath82 $ ] .
thus , we refer to the elements of @xmath78 as _ points_. let @xmath79 be the set of _ `` non - vertical '' lines _ in the affine plane , i.e. @xmath83 if and only if the equation for @xmath84 may be expressed as @xmath85 for some @xmath86 .
each line @xmath84 is determined uniquely by a pair @xmath87 . in order to distinguish points and lines we will use square brackets for points and parentheses for lines .
this geometry , in which one parallel class of the affine plane has been removed ( in our case , the class of `` vertical '' lines ) goes by the well established name _
biaffine plane_. we denote it by @xmath88 ( see figure [ fig.1 ] depicting @xmath89 ) . for more details about biaffine planes and their relation to other combinatorial structures ,
we recommend @xcite ; see also additional comments at the end of this paper .
.,width=491 ] given a point @xmath82 $ ] and a line @xmath87 of the biaffine plane , we define a _ quasidistance _ @xmath90 by the formulas : @xmath91 and @xmath92 .
note that @xmath93 does not define a metric ( indeed , it is not symmetric and non - negativity does not make sense in @xmath76 ) .
thus , it is just a vague analogue .
recall that a natural source of coherent configurations is coming from permutation groups .
if we take any permutation group @xmath41 , then @xmath94 2-@xmath95 is its associated ( schurian ) coherent configuration . in particular ,
if @xmath41 is transitive , then @xmath33 is an association scheme .
let us now consider an action of the group @xmath96 on the set @xmath97 , most conveniently described in terms of generators .
to each pair @xmath98 we associate a _ translation _ @xmath99 acting naturally on @xmath78 as @xmath100\mapsto [ x+a , y+b]$ ] , while the induced action on @xmath79 is @xmath101 .
of course , the set of all translations forms an abelian group of order @xmath8 under composition of permutations ( denoted @xmath102 ) and is isomorphic to @xmath103 .
further , let @xmath104 be defined by @xmath105\mapsto [ x , y - x]$ ] .
the corresponding permutation on lines is @xmath106 .
clearly , @xmath107 is a permutation of order @xmath0 and it is immediate that @xmath108 .
our group @xmath109 above is generated by all translations together with the permutation @xmath107 .
note that all elements of @xmath109 may be expressed in the form @xmath110 , where @xmath111 are suitable elements of @xmath76 .
moreover , to distinct triples of @xmath111 correspond distinct elements of @xmath109 .
in other words , @xmath112 and @xmath113 . now set @xmath114 .
then multiplication in @xmath109 is given by @xmath115 .
observe that the action of @xmath109 is intransitive on @xmath42 with two orbits @xmath78 and @xmath79 . in
what follows , the introduced permutation group @xmath116 will be called the _ heisenberg group modulo @xmath0 _ ; see section 9 for more details .
[ prop1 ] let @xmath0 be an odd prime .
then * the rank of @xmath116 is @xmath117 , * the @xmath118-orbits of @xmath116 may be divided into six different types of classes @xmath119 and @xmath120 , which are characterized by suitable relations between coordinates of objects in @xmath42
. * proof .
* let @xmath121 $ ] , @xmath122\in\mathcal p$ ] and @xmath123 , @xmath124 .
then the types of classes are the following : * @xmath125 and @xmath126 , where @xmath127 ( i.e. , those pairs of points whose first coordinates are equal and second coordinates differ by @xmath128 in the given order ; note that @xmath129 is the diagonal relation on @xmath78 ) , * @xmath130 , where @xmath131 ( i.e. , those pairs of points whose first coordinates differ by @xmath128 in the given order ) , * @xmath132 and @xmath133 , where @xmath127 ( i.e. , those pairs of lines whose first coordinates are equal and second coordinates differ by @xmath128 in the given order ; note that @xmath134 is the diagonal relation on @xmath79 ) , * @xmath135 , where @xmath131 ( i.e. , those pairs of lines whose first coordinates differ by @xmath128 in the given order ) , * @xmath136 , where @xmath127 ( i.e. , those point - line pairs whose quasidistance @xmath137 is @xmath128 ) , * @xmath138 , where @xmath127 ( i.e. , those point - line pairs whose quasidistance @xmath139 is @xmath128 ) . to complete the proof , one must verify two things : that the @xmath117 relations introduced above indeed form a partition of the set @xmath140 , and that each such relation is in fact a 2-orbit of @xmath116 .
* remark 1 .
* one can easily check from its definition that the permutation @xmath107 has exactly @xmath0 fixed points in its action on @xmath78 , as well as @xmath0 fixed points in its action on @xmath79 . relying on the bijection between 2-orbits of a transitive permutation group and orbits ( 1-orbits ) of the stabilizer of an arbitrary point ( see e.g. @xcite ) , the reader can easily deduce that there exists exactly @xmath141 2-orbits of the transitive action @xmath142 , and similarly for the action @xmath143 .
thus , we obtain in this manner @xmath144 2-orbits of @xmath116 . observing that there are @xmath0 2-orbits of type @xmath145 and @xmath0 of type @xmath120
, we arrive at the desired amount of @xmath117 .
* remark 2 . * reflexive 2-orbits @xmath129 and @xmath134 are obviously symmetric , however all remaining 2-orbits are antisymmetric .
namely , we obtain that @xmath146 , @xmath147 , @xmath148 , @xmath149 , @xmath150 and @xmath151 . here and below ,
operations on subscripts are performed as in @xmath76 .
this description provides a quite nice geometrical interpretation of the relations . in the sections to follow
, we will define certain families of color graphs as suitable mergings of these relations .
we will prove that these color graphs are in fact association schemes , and we will investigate them further . note that at the end of the paper we will once again justify a portion of our obtained results , this time relying on a more purely algebraic approach .
denote by @xmath5 the coherent configuration corresponding to the group @xmath109 in our construction above .
for obvious reasons , we shall refer to @xmath5 as a _ biaffine coherent configuration_.
the biaffine coherent configuration @xmath5 was constructed with the aid of a group , that is , it is schurian . presently , we are interested in the intersection numbers of @xmath5 . in this section
we display these numbers with the aid of tables . in each individual table
the superscript is fixed , the symbol in the row indicates the first subscript , and the symbol in the column indicates the second subscript .
faradev i.a . ,
, muzichuk m.e . : _ cellular rings and groups of automorphisms of graphs _ , in : faradev i.a . ,
ivanov a.a . ,
klin m.h . , and woldar a.j . , eds .
, `` investigations in algebraic theory of combinatorial objects , '' kluwer acad . publ . ,
dordrecht , 1994 , 1152 .
fiedler f. , klin m. , pech ch .
: _ directed strongly regular graphs as elements of coherent algebras _ , in : _ general algebra and discrete mathematics : proceedings of the conference on general algebra and discrete mathematics _
, potsdam 1998 , k. denecke & h.j.vogel ( eds . ) , shaker verlag , aachen 1999 , 6987 .
gyrki . , klin m. : _ sporadic examples of directed strongly regular graphs obtained by computer algebra experimentation _
, in gerdt , vladimir p. ( ed . ) et al . , _
computer algebra in scientific computing _ , lecture notes in computer science * 8660 * ( 2014 ) , 155170 .
klin m.h . ,
rcker ch .
, rcker g. , tinhofer g. : _ algebraic combinatorics in mathematical chemistry .
methods and algorithms .
i. permutation groups and coherent ( cellular ) algebras . _ match no .
40 ( 1999 ) , 7138 .
soicher l.h . :
_ grape : a system for computing with graphs and groups _ , groups and computation ( new brunswick , 1991 ) , dimacs ser
. discrete math .
, 11 , 287291 , amer .
soc . , providence , ri , 1993 . on construction and identification of graphs .
with contributions by a. lehman , g. m. adelson - velsky , v. arlazarov , i. faragev , a. uskov , i. zuev , m. rosenfeld and b. weisfeiler . edited by boris weisfeiler , lecture notes in mathematics , vol . 558 .
springer - verlag , berlin - new york , 1976 .
xiv+237 pp . | let @xmath0 be an odd prime . in this paper we provide a construction which gives four non - schurian association schemes for every @xmath1 and two for @xmath2 .
this construction is explained using incidences between points and lines of a biaffine plane and we also provide a pure algebraic model for it with the aid of finite heisenberg groups .
the obtained results are discussed in a more wide framework .
* keywords . *
coherent configuration , association scheme , biaffine plane , computer algebra , heisenberg group , non - schurian scheme , plausible reasonings .
* msc . * 05e30 , 51e15 . ' '' '' ' '' '' |
spin transport represents a new branch in mesoscopic physics with several technological applications@xcite , e.g. information storage , magnetic sensors and potentially quantum computation@xcite .
while most theoretical models are based on fermi liquid theory , some work has been done on strongly correlated 1d systems using luttinger liquid theory@xcite .
this work has focused in the weak tunneling regime between the ferromagnet and the 1d system and found that spin transport may provide experimental evidence of spin charge separation , one of the main predictions of luttinger liquid theory that remains to be observed experimentally in an unambiguously accepted way . despite the possible technological applications and contributions to the study of spin charge separation in strongly correlated systems ,
very little experimental work has been carried out on spin transport in 1d systems.@xcite this work is complicated by the use of multi - walled carbon nanotubes , and explored only situations with ferromagnetic contacts with parallel or antiparallel magnetizations .
early experimental work with nanotube devices was limited by poor contacts between the electrodes and the nanotube , and accordingly theoretical models focused in the weak tunneling regime .
recently , liang _ et al . _
@xcite have succeeded in fabricating single - walled carbon nanotube devices with near - perfect ohmic contacts to the electrodes . a schematic representation of their experiment is presented in fig .
[ scheme ] .
these devices are characterized by values of the conductance as high as @xmath0 , close to the theoretically predicted higher limit@xcite of @xmath1 . at temperatures below 10k , the measured conductance exhibits approximately periodic oscillations as a function of the gate voltage .
these oscillations are due to fabry - perot interference
i.e. quantum interference between propagating electron waves inside the resonant cavity defined by the two nanotube - electrode interfaces . experimental geometry ( from ref . ) .
a single - walled carbon nanotube is located on a silicon gate and oxide layer .
the electrodes , which may be ferromagnetic , are grown on top of the nanotube .
the doped silicon is used as a gate electrode to modulate the charge density .
the electronic transport properties of the nanotube devices were characterized as a function of bias ( @xmath2 ) and gate ( @xmath3 ) voltages .
, title="fig : " ] in order to explain their result , ref . considered a model of non - interacting electrons and used the multi - channel landauer - bttiker formalism to calculate the differential conductance as a function of the bias and gate voltages .
they have found qualitative agreement between the calculated conductance and their experimental data , especially with regard to the variation of the low - bias conductance with gate voltage . on the other hand , transport experiments on carbon nanotubes@xcite
have shown that electrons in nanotubes are strongly correlated and are better described by a luttinger liquid model@xcite .
this implies the electrons in these systems do not exhibit fermi liquid properties but instead form collective excitations better described by charge and spin like density waves that propagate with different velocities .
this luttinger liquid behavior drastically changes the charge conductance for these systems and it is interesting to know how this affects the results obtained for the particular setup used in ref . .
furthermore , this setup can be generalized to the use of ferromagnetic electrodes , in order to study both charge and spin transport in 1d electron systems . in this paper
we investigate the spin and charge transport properties in 1d electron systems with near - perfect contacts to ferromagnetic electrodes ( the normal metal electrodes correspond to the particular case of zero magnetization ) .
we consider both the case of quantum wires , i.e. single - channel electron systems , and single - walled carbon nanotubes , but mainly focus on the latter one .
we use a non - interacting stoner model to treat the ferromagnetic leads and a luttinger liquid model for the nanotube and consider the case of near - perfect contacts to the leads , therefore treating backscattering at the contacts perturbatively . in order to introduce the effect of a finite bias voltage
, we use the non - equilibrium keldysh formalism . following this procedure ,
we obtain the conductance , spin and spin current as functions of the gate and bias voltages , the external magnetic field and the orientation of the magnetization in each lead .
we study how the strong coulomb interactions affect these transport properties and find some features in the fabry - perot interference patterns that are related to spin charge separation .
a single - walled carbon nanotube with long - range coulomb interactions is well described by a forward - scattering model@xcite . in this model the hamiltonian density is given by @xmath4 ^ 2 , \end{aligned}\ ] ] where the right / left moving electron operators @xmath5 have the labels @xmath6 for the band and @xmath7 for the spin projection of the electrons in the nanotube and @xmath8 is the interaction strength .
the term in the square brackets corresponds to the electron density .
we also consider the same problem for a single - channel quantum wire , for which there is no sub - band degeneracy and the band index can be dropped . due to the similarities between the two cases ,
we give explicit analytical formulas throughout the paper only for the nanotube , but will present results for the quantum wire where appropriate .
the metallic leads are modeled as two semi - infinite 1d - non - interacting systems@xcite , which is obtained with a position dependent @xmath8 , constant in the wire and zero in the leads .
schematic representation of the model .
+ the leads are modeled as 1d non - interacting electron systems , the luttinger liquid parameter is therefore @xmath9 . the nanotube ( bold line ) , on the other hand , is described by a 1d interacting system , in this case @xmath10 , which corresponds to repulsive interactions .
the contacts to the leads are modeled as two weak backscattering barriers .
the two backscatterers generate fabry - perot interference.,title="fig : " ] we allow for ferromagnetism in the leads using a non - interacting stoner model@xcite ( mean - field treatment of the magnetization ) .
the hamiltonian density is @xmath11 with @xmath12 and @xmath13 where @xmath14 are the pauli matrices and @xmath15 is the `` exchange field '' , which is proportional to the magnetization .
this is constant in each ferromagnetic lead , i.e. @xmath16 for @xmath17 and @xmath18 for @xmath19 and in ordinary paramagnetic leads @xmath20 . in this case , the total system corresponding to a nanotube between two ferromagnetic leads is described by the hamiltonian @xmath21 the total hamiltonian @xmath22 can take a form identical to the hamiltonian in the case of normal metal leads by applying the following transformation to the electron field operators separately in the left and right leads respectively @xmath23 this transformation leaves @xmath24 invariant and @xmath25 transforms into @xmath26 .
we apply the usual bosonization procedure to study this model@xcite .
the four electron modes are associated to four bosonic modes described by the fields @xmath27 and their duals @xmath28 via the bosonization transformation @xmath29 where @xmath30 is a short - distance cutoff .
it is convenient to consider the following linear combinations of the fields@xcite : @xmath31 and @xmath32 , with @xmath33 and @xmath34 .
this allows us to define the new fields @xmath35 , which corresponds to the total charge mode and is the only interacting mode , and @xmath36 , @xmath37 and @xmath38 ; with similar transformations for the @xmath39 fields . in terms of these new fields the luttinger liquid hamiltonian density ( [ h ] ) is diagonalized @xmath40 \nonumber \\ & + \sum_{i=2}^4 \frac{{v_{\mathrm{\scriptscriptstyle f}}}}{2\pi } \left [ \left(\partial_x \varphi_i \right)^2 + \left(\partial_x \theta_i \right)^2 \right]\ , \end{aligned}\ ] ] where @xmath41 is the fermi velocity , @xmath42 is the renormalized velocity due to the interactions and @xmath43 is the luttinger liquid parameter . in the inhomogeneous model @xmath43 and @xmath42
are functions of the position : @xmath9 and @xmath44 in the leads and @xmath45 and @xmath46 in the nanotube .
the contacts between the leads and the nanotube are modeled by weak backscattering at the contact points , the corresponding hamiltonian density has the form @xmath47 \\ & = \sum_{m , a , b=1}^2 \sum_{\alpha=\pm 1 } \delta(x - x_m ) \left\ { \left ( u_m^{a b } + \alpha v_m^{a b } m_m^z \right ) e^ { i \left[\theta_1+\alpha\theta_2+(-1)^{a+1 } \delta_{a b } \left ( \theta_3+\alpha\theta_4 \right)+ ( -1)^{a+1 } \left ( 1-\delta_{a b } \right ) \left ( \varphi_3+\alpha \varphi_4 \right)\right ] } \right .
\nonumber \\ & \
\ + v_m^{a b } \left ( m_m^x+i\alpha m_m^y \right ) e^ { i \left[\theta_1 + ( -1)^{a+1 } \delta_{a b } \theta_3 + ( -1)^{a+1 } \left ( 1 - \delta_{a b } \right ) \alpha \theta_4 + \alpha \varphi_2 + ( -1)^{a+1 } \left ( 1-\delta_{a b } \right ) \varphi_3 + ( -1)^{a+1 } \delta_{a b } \alpha \varphi_4 \right ] } + \ , { \mathrm{h.c . } } \left .
\vphantom{e^ { i \theta_1 ^ 2 } } \right\ } \ , \nonumber\end{aligned}\ ] ] where @xmath48 labels the position of the contacts : @xmath49 ( @xmath50 is the nanotube length ) and @xmath51 and @xmath52 are constants proportional to the strength of the backscattering , @xmath53 and the same for @xmath52 .
the backscattering terms are restricted by symmetry according to charge conservation and spin rotational symmetry around the axis of magnetization of the ferromagnetic contact .
we consider only terms of the form @xmath54 because these are the most relevant in the renormalization group sense ( the scaling dimension in real space of these terms is @xmath55 , while the scaling dimension of terms of the form @xmath56 is @xmath57 ) .
hence if all scattering terms are weak , these terms will dominate .
it is straightforward to extend the present treatment to include the neglected interactions , though we do not attempt this here .
the effect of the magnetization appears only on the backscattering term . in the case of near - perfect contacts to the electrodes
, we can treat the backscattering hamiltonian @xmath58 as a perturbation to the hamiltonian @xmath59 .
this procedure is described in section [ greens ] in the context of the keldysh formalism that we use in order to account for the effects of the finite bias voltage .
the gate voltage introduces a term in the hamiltonian density proportional to @xmath60 , where @xmath61 is the electron density .
the constant of proportionality relates the voltage applied at the gate with the voltage felt by the nanotube and therefore depends on the sample .
the hamiltonian density @xmath62 becomes @xmath63 , after applying the following transformation to the @xmath64 field : @xmath65 , where we absorbed a constant of proportionality into the definition of @xmath3 for simplicity .
this transformation needs to be applied to the total hamiltonian , including the backscattering term @xmath58 , which means that the gate voltage after this transformation will only contribute to the perturbation hamiltonian .
the effect of the external magnetic field is introduced via a zeeman coupling term in the hamiltonian @xmath66 the contribution of the magnetic field can be transfered to the perturbation hamiltonian @xmath58 using a similar procedure to the one described above for the gate voltage . taking the z - direction as the direction of the magnetic field , i.e. @xmath67
, the zeeman hamiltonian density becomes @xmath68 and applying @xmath69 ( with @xmath70 ) , to the hamiltonian @xmath71 , it transforms as @xmath72 .
the results for non - zero magnetic field are presented in appendix [ ap : magn ] .
due to the finite bias voltage the distribution in this system is not in thermal equilibrium .
this non - equilibrium situation is studied using the keldysh formalism ( for a review , see ref . ) . to define a non - equilibrium initial state
, we assume that until some initial time @xmath73 , the system has reached quasi - equilibrium _ in the absence of impurity scattering _ @xmath74 . without impurity scattering ,
the total number of right- and left - moving carriers , @xmath75 , @xmath76 , are separately conserved , so that a partial equilibrium can be established with well - defined separate chemical potentials for the right and left movers .
hence , the system can be described , up to this time , by a thermal distribution governed by the grand canonical hamiltonian @xmath77 where the chemical potentials in each lead are taken to be @xmath78 and @xmath79 .
the right and left moving particle densities are given in the bosonization procedure by @xmath80 .
then @xmath81 we emphasize that the appearance of the voltage @xmath2 in @xmath82 does not represent a physical force on the electrons , but rather parameterizes their non - equilibrium distribution . after the initial time @xmath73 the evolution of the system
is governed by a different hamiltonian @xmath22 , which as deduced in section [ model ] is @xmath83 , with @xmath84 given in ( [ hll ] ) and @xmath58 in ( [ hbs ] ) .
we expect on physical grounds that introducing localized scattering at the ends of the wire or nanotube reduces the current , but can not affect the non - equilibrium distribution in the reservoirs .
hence , we believe that the prescription of defining the voltage @xmath2 from the initial distribution as is done using ( [ truthfull ] ) gives a faithful description of ideal leads . according to this prescription , a physical observable , represented by an operator @xmath85 ,
is then calculated from @xmath86 where @xmath87 the difficulty in evaluating such an expectation value is that , unlike in a conventional equilibrium calculation , the hamiltonian @xmath82 governing the initial distribution is different from @xmath22 , which governs the time evolution .
thus such an expectation value can not be evaluated by equilibrium green s function techniques . instead , we take advantage of the special property of @xmath82 that the voltage couples only to @xmath88 , which are decoupled `` zero mode '' degrees of freedom .
this technique has been applied a number of times before to related problems@xcite , but to our knowledge the details of its derivation have never been published . for completeness , pedagogical value , and to highlight the physical assumptions of the method , we include a thorough derivation in appendix [ ap : derivation ] .
the correction to @xmath89 due to the backscattering is given by @xmath90 where @xmath91\ ] ] is the evolution operator for a system with the time - dependent hamiltonian @xmath92 . here
@xmath93 is the hamiltonian in the frame co - moving with the ideal current , defined by @xmath94_{\theta_1 \rightarrow \theta_1 + v t } \ , \ ] ] with @xmath95 .
note that ( because @xmath96=0 $ ] ) all the time dependence in @xmath93 is in the backscattering term , and is hence easy to handle when working perturbatively in @xmath58 .
( [ eq : expect4]-[eq : hi ] ) provide a reformulation of the transport problem which is particularly convenient for a perturbative treatment of the backscattering .
note that because the voltage @xmath2 appears only within @xmath58 a direct expansion of eq .
( [ eq : expect4 ] ) in @xmath58 will involve _ equilibrium _ real - time propagators calculated with respect to @xmath84 .
we develop this perturbation theory using the keldysh path integral formulation .
this involves the usual trotterization of the two evolution operators @xmath97 in eq .
( [ eq : expect4 ] ) using coherent - state fields denoted @xmath98 for @xmath99 ( `` forward branch '' ) and @xmath100 ( `` backward branch '' ) for @xmath101 . further noting that @xmath84 is quadratic and @xmath58 acts only at the ends of the nanotube / wire , the fields away from @xmath102 can be integrated out to obtain the keldysh integral @xmath103 { \mathcal o}_k e^{i s_0 - i \int\ ! { \mathrm{d}}t h_{\mathrm{pert}}(t ) } \ , \ ] ] with @xmath104 - h_{\mathrm{bs } } [ \varphi^-_{i},\theta^-_{i}+\delta_{i1 } v t ] \ .\ ] ] here @xmath105 is an appropriate keldysh representation of the operator @xmath106 , which can be chosen as usual from fields lying on either the forward or backward moving branch , or any linear combination thereof
see below for convenient choices .
the quadratic action @xmath107 is a functional of @xmath108 , which can be determined from the fact that it must reproduce the _ equilibrium _ correlation and response functions for these fields .
indeed , we do not require an explicit expression for @xmath107 , but instead give the response and correlation functions , defined by @xmath109 \rangle \ , \end{aligned}\ ] ] where we have applied the standard keldysh rotation to the fields @xmath110 . by construction @xmath111 .
the green s functions involving the @xmath39 fields are defined in a similar way , replacing @xmath112 by @xmath39 in the above equations .
there are also green s functions that involve both @xmath112 and @xmath39 , these are defined by @xmath113 \rangle \ , \end{aligned}\ ] ] and similar definitions for @xmath114 and @xmath115 . again , by construction @xmath116 . using the above procedure
we obtain ( up to additive constants that will not contribute to the final result ) the green s functions for the @xmath64 fields : @xmath117 \nonumber \\
r^{\theta i}_{12}(t ) & = -\frac{\pi}{2 } ( 1-\alpha^2 ) \sum_{k \geq 0 } \alpha^{2k}\ , \theta ( t- ( 2k+1 ) t_v ) \ , \nonumber \\
c^{\theta i}_{11}(t ) & = -\frac{1-\alpha}{4 } \left [ \log t^2 + \phantom{\sum_{k \geq 1 } } \right .
\nonumber \\ & \qquad\qquad + \left .
\frac{1+\alpha}{\alpha } \sum_{k \geq 1 } \alpha^{2k}\ , \log \left|t^2-(2 k t_v)^2 \right| \right ] \ , \nonumber \\
c^{\theta i}_{12}(t ) & = -\frac{1-\alpha^2}{4 } \sum_{k \geq 0 } \alpha^{2k}\ , \log\left|t^2- [ ( 2k+1 ) t_v ] ^2 \right| \ , \end{aligned}\ ] ] where the subscripts label the position of the contacts @xmath118 ( e.g. @xmath119 ) and @xmath120 we also need the green s functions for the non - interacting modes @xmath121 , @xmath122 and @xmath123 .
these are obtained from ( [ thetagreens ] ) by taking @xmath124 and replacing @xmath125 by @xmath126 , @xmath127 the green s functions for the @xmath39 fields can be obtained from those for the @xmath112 fields given in ( [ thetagreens ] ) , by replacing @xmath43 by @xmath128 , i.e. by replacing @xmath129 by @xmath130 . on the other hand , the only @xmath39 green s functions that contribute to the transport properties studied in the following sessions are those that correspond to the non - interacting modes @xmath131 , and therefore they are identical to the functions given in ( [ freegreens ] ) . in order to compute the spin transport properties in section [ spint ]
, we also need the following functions for the @xmath132 and @xmath133 fields @xmath134 @xmath135 and @xmath136 .
in this section we study the charge transport properties of 1d electron systems and how these are affected by the magnetization of the leads and , more importantly , the presence of the strong coulomb interactions .
we use the procedure described in sections [ model ] and [ greens ] to calculate the differential conductance for these systems .
this is obtained from the expectation value of the current in a nanotube , i.e. a four mode 1d electron system with the hamiltonian given in ( [ hll ] ) , as @xmath137 after a lengthy but straightforward calculation we obtain that the differential conductance @xmath138 to second order in perturbation theory is given by @xmath139\cos(v t ) \!\right\}\ ] ] with @xmath140 \ , \\ u_2= & 2 \cos ( { v_{\!g}}l ) \sum_{a b } \left [ u^{a b}_1\ , u^{a b}_2 + v^{a b}_1\ , v^{a b}_2 \ , \vec{m}_1 \cdot \vec{m}_2 \right ] \ , \nonumber\end{aligned}\ ] ] and @xmath141 and similarly for @xmath142 . for a quantum wire ( i.e. a single - channel electron system )
these are replaced by @xmath143 and the global normalization is divided by a factor of two ( since the quantum wire has two modes instead of four ) .
( [ g2 ] ) and ( [ uu ] ) are valid for zero external magnetic field , which is the case considered in this section , the equations for non - zero magnetic field are presented in appendix [ ap : magn ] . equation ( [ g2 ] ) can be easily generalized to arbitrary order in perturbation theory , but the time integrals need to be computed numerically .
we present the calculated conductance to second order for three different physical models in fig .
[ fig : g ] .
this models correspond to ( a ) a nanotube with non - interacting electrons , i.e.taking @xmath9 , ( b ) a quantum wire with @xmath144 and ( c ) a nanotube with @xmath145 , which is a physically relevant value for single - walled carbon nanotubes@xcite .
the effect of the interactions is visible in the dependence of the conductance with bias voltage , at constant gate voltage .
the conductance is a quasi - periodic function of the bias voltage . at @xmath146 , for the non - interacting case ,
see fig .
[ fig : g].(a ) , this dependence is a cosine function with period @xmath147 . for a quantum wire ,
[ fig : g].(b ) , there are clearly two different `` periods '' in the oscillations , these are related to the two time scales @xmath126 and @xmath148 .
the existence of these two different time scales is due to the two bosonic excitations in this system : the spin excitation with velocity @xmath41 and the charge excitation with velocity @xmath42 , and is therefore an effect of spin charge separation .
the same effect appears in fig .
[ fig : g].(c ) , but since for the nanotube there are three non - interacting modes with velocity @xmath41 and only one mode , the total charge , with velocity @xmath42 , it is less visible than in the previous example .
the most visible effect of the interactions in the nanotube , is the enhancement of the amplitude in the conductance around @xmath149 .
this effect is observable in the experimental data presented in ref . .
the calculated conductance at @xmath150 as a function of the bias voltage for a nanotube with different interaction strengths corresponding to @xmath151 and @xmath152 , is presented in fig .
[ fig : pi2 ] .
it can be seen using eqs .
( [ g2 ] ) and ( [ uu ] ) that the conductance for this value of the gate voltage only depends on the green s functions @xmath153 and @xmath154 , which depends only on @xmath125 . as a result , we can clearly see in fig .
[ fig : pi2 ] that the period of the oscillations is @xmath155 , and therefore depends strongly on the interaction strength .
the amplitude of this oscillations is very small except for the first oscillation , which is enough to identify this effect . at constant gate voltage :
@xmath156 , the period of oscillations is @xmath157 , i.e. depends strongly on the interaction strength . in order for this effect to be clearly visible , we scaled and shifted the functions differently , therefore the values of the @xmath158 axes are not meaningful .
, title="fig : " ] as for the dependence with the gate voltage , the conductance is a periodic function , which is modulated by @xmath159 , and this is the main effect of the magnetization in the leads on the conductance . in particular
if @xmath160 , there is an angle between the two magnetizations for which @xmath161 vanishes for any value of the gate voltage , in this case the conductance is given also in fig .
[ fig : pi2 ] .
in this section we study spin transport properties , i.e. the spin density in the nanotube and the spin current generated by the magnetization in the leads . the spin density expectation value in the nanotube , calculated using bosonization and the keldysh perturbation formalism as described in sections [ model ] and [ greens ] , is given by @xmath162 for zero magnetic field , it is technically simplest to calculate @xmath163 from the bosonized form @xmath164 and then obtain the other two components by rotational invariance . for non - zero magnetic field the calculation as well as
the final results are much more involved , therefore and for the sake of clarity we only present the results in appendix [ ap : magn ] .
the result is @xmath165\sin(v t ) \right .
\nonumber \\ & + \sin ( { v_{\!g}}l ) \left\{\vec
u_2 \int\!\!{\mathrm{d}}t\ , e^{{\bm c}_{12}(t ) } \sin\!\left [ \frac{1}{2 } { \bm r}_{12}(t)\right]\sin(v t ) \right .
\nonumber \\ & \left .
+ \vec u_3 \int\!\!{\mathrm{d}}t\ , e^{{\bm c}_{12}(t ) } \sin\!\left [ \frac{1}{2 } { \bm r}_{12}(t)\right]\cos(v t ) \right\ } \right ) \ , \end{aligned}\ ] ] with @xmath166 notice that the spin density does not depend on the position in the nanotube , hence the total spin is @xmath167 .
the first term , proportional to @xmath168 , is the known non - equilibrium spin accumulation effect@xcite .
it is maximum for @xmath169 , when , in the case of identical contacts , the other terms vanish .
this term does not couple two backscatterers , is independent of the gate voltage , and is an increasing function of the bias voltage .
it is depicted in fig .
[ fig : spin](a ) , since it is the only term term that corresponds to the component of the spin in the direction of @xmath170 at @xmath146 .
the second term corresponds to the component of the spin perpendicular to the plane of the magnetizations and is depicted in fig .
[ fig : spin](c ) , as function of the bias and gate voltages .
the third term is the only one that survives in equilibrium , i.e. at zero bias , it is due to the fact that the backscattering strengths depend on the spins of the incoming and outgoing electron relative to the direction of the magnetizations .
it is maximum for @xmath171 , when again for identical contacts the other terms vanish .
this is depicted in fig .
[ fig : spin](b ) , at @xmath172 .
these terms that couple the two backscatterers , and hence depend on the gate voltage , vary with bias voltage in a manner approximately described by a sum of two periodic functions , with `` periods '' given by @xmath147 and @xmath173 , as discussed in the previous section for the conductance .
the spin current @xmath174 is as the spin density calculated from the @xmath175 component in its bosonized form ( again see result for non - zero magnetic field in appendix [ ap : magn ] ) @xmath176 it is not well - defined at the contact points because the backscattering term in the hamiltonian ( [ hbs ] ) does not conserve spin , and therefore it has different expressions in the nanotube and the leads .
the spin current in the left ( @xmath177 ) and right ( @xmath178 ) leads is given by @xmath179\sin(v t ) \nonumber \\ & + \left[\pm \sin ( { v_{\!g}}l)\ \vec u_2 + \cos ( { v_{\!g}}l)\ , \vec u_3 \right ] \nonumber\\ & \qquad \times \int\!\!{\mathrm{d}}t\ , e^{{\bm c}_{12}(t ) } \sin\!\left [ \frac{1}{2 } { \bm r}_{12}(t)\right]\sin(v t ) \end{aligned}\ ] ] and in the nanotube by : @xmath180\sin(v t ) \\ & + \cos ( { v_{\!g}}l)\,\left\ { \
\vec u_2 \int\!\!{\mathrm{d}}t\ , e^{{\bm c}_{12}(t ) } \sin\!\left [ \frac{1}{2 } { \bm r}_{12}(t)\right]\cos(v t ) \right .
\nonumber \\ & \quad\quad\qquad + \left .
\vec u_3 \int\!\!{\mathrm{d}}t\ , e^{{\bm c}_{12}(t ) } \sin\!\left [ \frac{1}{2 } { \bm r}_{12}(t)\right]\sin(v t ) \right\ } \ . \nonumber\end{aligned}\ ] ] with @xmath181 and @xmath182 defined in ( [ us ] ) and @xmath183 similarly to the results for the spin discussed above , the first term , which only involves one backscatterer and is independent of the gate voltage , corresponds to the usual spin injection effect .
it is an increasing function of the gate voltage and is maximum for @xmath184 .
this can be seen in fig .
[ fig : spincurr](b ) , since it is the only term that does not vanish in the direction of @xmath185 at @xmath186 . at @xmath146 ,
the terms proportional to @xmath187 and @xmath182 contribute equally to the component of the current in the direction of @xmath170 , the result for this case is presented in fig .
[ fig : spincurr](a ) . the second term , proportional to @xmath188 corresponds to an exchange interaction between the magnetizations of the leads , mediated by the nanotube .
it has opposite signs in the two leads and it is shown in fig .
[ fig : spincurr](c ) .
we studied the charge and spin transport properties of 1d systems , e.g. quantum wires and carbon nanotubes , focusing on the latter .
we considered the case of nearly perfect ohmic contact between the 1d system and the electrodes and included the strong coulomb interaction via a luttinger liquid model .
we found important effects on the transport properties of these systems that are due to the coulomb interactions .
these appear in the dependence with bias voltage .
in particular , the conductance is enhanced at low bias voltage , furthermore it is an oscillatory function where we can distinguish two quasi - periodic components , with periods that are related to the two velocities of the excitations of a luttinger liquid , @xmath42 and @xmath41 .
this effect is therefore a direct consequence of spin and charge separation .
it is clearly visible in single band quantum wires . in nanotubes ,
the amplitude of the higher period component is reduced by the presence of three ( as opposed to one ) neutral modes .
still , we can find evidence of the two velocities @xmath42 and @xmath41 by comparing the dependence of the conductance with bias voltage for two different gate voltages ( @xmath189 and @xmath190 ) .
it is perhaps worth noting that , for the case of non - magnetic leads with symmetric contacts , the conductance formula involves only two unknown parameters : the overall amplitude of the backscattered current , and the luttinger parameter @xmath43 , both of which can be simply estimated . nevertheless , a non - trivial functional dependence upon bias voltage is predicted .
the spin and spin current have one component in the plane of the magnetization that does not couple the two leads and is therefore independent of the gate voltage .
this term should be understood as arising from incoherent spin injection at each contact .
it is a monotonic function of the bias voltage , and corresponds to the known spin accumulation ( for the spin ) and spin injection ( for the spin current ) effects .
the other components that couple the two leads , and therefore depend on the gate voltage , are backscattering processes occurring with coherence between the two contacts .
these oscillate with the bias voltage , in a manner approximately described as a sum of two periodic components , with periods related to the two velocities of the excitation of the luttinger liquid .
the amplitude of the higher - period component is largest in a single - channel quantum wire , and somewhat suppressed in nanotubes by the sub - band degeneracy .
c.s.p . was supported by fct and fse through grant praxis / bd/18554/98 .
l.b . was supported by nsf through grant dmr9985255 , and by the sloan and packard foundations .
thanks the deutsche forschungsgemeinschaft for support through heisenberg grant wi/1932 1 - 1 .
in this appendix , we derive eq . ( [ eq : expect4 ] ) . in particular , we consider a large periodic system of size @xmath191 , where ultimately @xmath192 .
we define the right / left - moving combinations @xmath193 in the system of size @xmath191 we can decompose into finite wavevector and `` zero mode '' components .
in particular , for the total charge fields , we define @xmath194 where @xmath195 contains the non - zero momentum modes of the @xmath196 fields . with these definitions ,
the zero - mode variables form two canonically conjugate pairs : @xmath197 & = & [ n_l,\phi_l ] = i \ , \\
\nonumber [ n_r , \phi_l ] & = & [ n_l,\phi_r ] = [ n_r , n_l]=[\phi_r,\phi_l ] = 0 \ .
\end{aligned}\ ] ] moreover , @xmath198 commute with @xmath195 and all fields associated with channels @xmath199 .
since the interactions which transform the system from a fermi liquid into a luttinger liquid ( eq . ( [ h ] ) ) exist only for @xmath200 , they do not affect the zero - mode terms in the hamiltonian .
hence one may separate @xmath201 where @xmath202 is the luttinger liquid hamiltonian , eq .
( [ hll ] ) , with the zero - mode terms subtracted , i.e. with @xmath203 and @xmath204 .
we then see , using the independence of the zero mode variables , that the unitary operator @xmath205 generates the following transformation @xmath206 , hence @xmath207 where @xmath208 is an unimportant constant . inserting this into eqs .
( [ eq : expect]-[eq : pf ] ) , one obtains @xmath209 with @xmath210 and @xmath211 for the operators of interest , @xmath212 where @xmath213 is the current which would flow in an ideal nanotube in the absence of backscattering . defining @xmath214 , with @xmath215 for @xmath216 and @xmath217 for @xmath218 , one
has then @xmath219 we then apply the formula @xmath220 \ , \ ] ] to arrive at eqs .
( [ eq : expect4 ] ) ff . given in the main text .
the differential conductance including the magnetic field is still given by eqs .
( [ g2 ] ) and ( [ uu ] ) with only the following change in @xmath161 @xmath221\right\ } \nonumber \\ & + \sin ( b l ) \sin ( { v_{\!g}}l ) \sum_{ab } \left [ u_1^{ab } v_2^{ab } ( \vec m_2 \cdot \hat h ) + u_2^{ab } v_1^{ab } ( m_1\cdot \hat h ) \right ] \ , \end{aligned}\ ] ] where @xmath222 .
the total spin of the nanotube is @xmath223 \sin { v_{\!g}}l \nonumber \\ & - u_1^{ab } u_2^{ab } \hat h l \sin b l \cos { v_{\!g}}l - v_1^{ab } v_2^{ab } \left [ \left ( ( \vec m_1 \cdot \hat h ) \vec m_2 +
( \vec m_2 \cdot \hat h ) \vec m_1 \right ) \tfrac{1}{b } \left(1 - \cos b l \right ) \right .
\nonumber \\ & \qquad \qquad \qquad \left .
+ ( \vec m_1 \cdot \hat h)(\vec m_2 \cdot \hat h ) \hat h \left ( l \sin b l - \tfrac{2}{b } \left ( 1 - \cos b l \right ) \right ) \right ]
\cos { v_{\!g}}l \biggr\ } \int_t e^{{\bm c}_{12 } } \sin \left ( \tfrac{1}{2 } { \bm r}_{12 } \right)\cos ( v t ) \nonumber \\ + & \biggl\ { \left ( u_1^{ab } v_2^{ab } ( \vec m_2 \times \hat h ) - u_2^{ab } v_1^{ab } ( \vec m_1 \times \hat h ) \right ) \tfrac{1}{b } \left ( 1 - \cos b l \right ) \cos { v_{\!g}}l + v_1^{ab } v_2^{ab } \left [ \left ( ( \vec m_1\cdot\hat h ) ( \vec m_2 \times \hat h ) - ( \vec m_2 \cdot \hat h ) ( \vec m_1 \times \hat h ) \right ) \right .
\nonumber \\ & \qquad \qquad \times \left .
\left ( l-\tfrac{1}{b } \sin b l \right ) + \vec m_1 \times \vec m_2 l \right ] \sin { v_{\!g}}l \biggr\ } \int_t e^{{\bm c}_{12 } } \sin \left ( \tfrac{1}{2 } { \bm r}_{12 } \right)\sin ( v t ) \biggr ) \ .\end{aligned}\ ] ] the spin current in the nanotube ( eq . [ jsnt ] ) and the leads ( eq . [ jsl ] ) is @xmath224 + u_2^{ab } v_2^{ab } ( \vec m_2 \cdot \hat h ) \hat h \left [ 1 - \cos b ( x - x_2 ) \right ] \nonumber \\ & + u_1^{ab } v_1^{ab } ( \vec m_1 \times \hat h ) \sin b ( x - x_1 ) - u_2^{ab } v_2^{ab } ( \vec m_2 \times \hat h ) \sin b ( x - x_2 ) \biggr\ } \int_t e^{{\bm c}_{11 } } \sin \left ( \tfrac{1}{2 } { \bm r}_{11 } \right ) \sin ( v t ) \nonumber
\\ + \sum_{ab } & \biggl\ { \bigl [ u_1^{ab } v_2^{ab } ( \vec m_2 \cdot \hat h ) \hat h \left [ \cos b l - \cos b ( x - x_1 ) \right ] + u_2^{ab } v_1^{ab } ( \vec m_1 \cdot \hat h ) \hat h \left [ \cos b l - \cos b ( x - x_2 ) \right ] \nonumber \\ & \qquad + u_1^{ab } v_2^{ab } \vec m_2 \cos b ( x - x_1 ) + u_2^{ab } v_1^{ab } \vec m_1 \cos b ( x - x_2 ) \bigl ] \cos { v_{\!g}}l \nonumber \\ & + v_1^{ab } v_2^{ab } \bigl [ ( \vec m_1 \cdot \hat h)(\vec m_2 \cdot \hat h ) \hat h \left [ \sin bl - \sin b(x - x_1)+ \sin b(x - x_2 ) \right ] \nonumber \\ & \qquad + ( \vec m_1 \cdot \hat h ) \vec m_2 \sin b(x - x_1 ) - ( \vec m_2 \cdot \hat h ) \vec
m_1 \sin b(x - x_2 ) \nonumber \\ & \qquad + u_1^{ab } u_2^{ab } \hat h \sin bl \bigr ] \sin { v_{\!g}}l \biggr\ } \int_t e^{{\bm c}_{12 } } \sin \left ( \tfrac{1}{2 } { \bm r}_{12 } \right ) \sin ( v t ) \nonumber \\ + \sum_{ab } & \biggl\ { \left [ u_1^{ab } v_2^{ab } ( \vec m_2 \times \hat h ) \sin b ( x - x_1 ) - u_2^{ab } v_1^{ab } ( \vec m_1 \times \hat h ) \sin b ( x - x_2 ) \right ] \sin { v_{\!g}}l \nonumber \\ & - v_1^{ab } v_2^{ab } \left [ ( \vec m_1\cdot\hat h ) ( \vec m_2 \times \hat h ) \left [ 1-\cos b(x - x_1 ) \right ] - ( \vec m_2 \cdot \hat h ) ( \vec m_1 \times \hat h ) \left [ 1-\cos b(x - x_2 ) \right ] \right . \nonumber \\ & \qquad \left .
+ \vec m_1\times \vec m_2 \right ] \cos { v_{\!g}}l \biggr\ } \int_t e^{{\bm c}_{12 } } \sin \left ( \tfrac{1}{2 } { \bm r}_{12 } \right ) \cos ( v t ) \ , \end{aligned}\ ] ] @xmath225 + u_2^{ab } v_2^{ab } ( \vec m_2 \cdot \hat h ) \hat h \left [ 1 - \cos b ( x - x_2 ) \right ] \nonumber \\ & \mp \left [ u_1^{ab } v_1^{ab } ( \vec m_1 \times \hat h ) \sin b ( x - x_1 ) + u_2^{ab } v_2^{ab } ( \vec m_2 \times \hat h ) \sin b ( x - x_2)\right ] \biggr\ } \int_t e^{{\bm c}_{11 } } \sin \left ( \tfrac{1}{2 } { \bm r}_{11 } \right ) \sin ( v t ) \nonumber
\\ + \sum_{ab } & \biggl\ { \bigl [ u_1^{ab } v_2^{ab } ( \vec m_2 \cdot \hat h ) \hat h \left [ \cos b l - \cos b ( x - x_1 ) \right ] + u_2^{ab } v_1^{ab } ( \vec m_1 \cdot \hat h ) \hat h \left [ \cos b l - \cos b ( x - x_2 ) \right ] \nonumber \\ & \qquad + u_1^{ab } v_2^{ab } \vec m_2 \cos b ( x - x_1 ) + u_2^{ab } v_1^{ab } \vec m_1 \cos b ( x - x_2 ) \bigl ] \cos { v_{\!g}}l \nonumber \\ & + v_1^{ab } v_2^{ab } \bigl [ ( \vec m_1 \cdot \hat h)(\vec m_2 \cdot \hat h ) \hat h \left [ \sin bl - \sin b(x - x_1)+ \sin b(x - x_2 ) \right ] \nonumber \\ & \qquad + ( \vec m_1 \cdot \hat h ) \vec m_2 \sin b(x - x_1 ) - ( \vec m_2 \cdot \hat h ) \vec
m_1 \sin b(x - x_2 ) \bigr ] \sin { v_{\!g}}l \nonumber \\ & \mp \left [ u_1^{ab } v_2^{ab } ( \vec m_2 \times \hat h ) \sin b ( x - x_1 ) + u_2^{ab } v_1^{ab } ( \vec m_1 \times \hat h ) \sin b ( x - x_2 ) \right ] \cos { v_{\!g}}l \nonumber \\ & \mp v_1^{ab } v_2^{ab } \left [ ( \vec m_1\cdot\hat h ) ( \vec m_2 \times \hat h ) \left [ 1-\cos b(x - x_1 ) \right ] - ( \vec m_2 \cdot \hat h ) ( \vec m_1 \times \hat h ) \left [ 1-\cos b(x - x_2 ) \right ] \right . \nonumber \\ & \qquad \left .
+ \vec m_1\times \vec m_2 \right ] \sin { v_{\!g}}l + u_1^{ab } u_2^{ab } \hat h
\sin bl \sin { v_{\!g}}l \biggr\ } \int_t e^{{\bm c}_{12 } } \sin \left ( \tfrac{1}{2 } { \bm r}_{12 } \right ) \sin ( v t ) \ , \end{aligned}\ ] ] where the @xmath226 sign correspond to lead 1,2 respectively . | we study the two - terminal transport properties of a metallic single - walled carbon nanotube with good contacts to electrodes , which have recently been shown [ w. liang et al , nature 441 , 665 - 669 ( 2001 ) ] to conduct ballistically with weak backscattering occurring mainly at the two contacts .
the measured conductance , as a function of bias and gate voltages , shows an oscillating pattern of quantum interference .
we show how such patterns can be understood and calculated , taking into account luttinger liquid effects resulting from strong coulomb interactions in the nanotube .
we treat back - scattering in the contacts perturbatively and use the keldysh formalism to treat non - equilibrium effects due to the non - zero bias voltage . going beyond current experiments , we include the effects of possible ferromagnetic polarization of the leads to describe spin transport in carbon nanotubes .
we thereby describe both incoherent spin injection and coherent resonant spin transport between the two leads .
spin currents can be produced in both ways , but only the latter allow this spin current to be controlled using an external gate . in all cases , the spin currents , charge currents , and magnetization of the nanotube exhibit components varying quasiperiodically with bias voltage , approximately as a superposition of periodic interference oscillations of spin- and charge - carrying `` quasiparticles '' in the nanotube , each with its own period .
the amplitude of the higher - period signal is largest in single - mode quantum wires , and is somewhat suppressed in metallic nanotubes due to their sub - band degeneracy . |
understanding the generic conditions for biodiversity and species extinction remains a challenging problem in evolutionary and population dynamics @xcite .
whereas real world ecological systems are composed of tens or hundreds of species , theoretically well understood cases remain restricted to systems with only very few species that interact in rather simple ways .
but even these very simple cases have revealed a very rich behavior , due to the nonlinearity inherent to this type of systems .
further progress in this field can be expected through the use of well established methods from nonlinear dynamics and statistical physics @xcite .
the simplest few species models , like the three - species rock - paper - scissors model or its four species variant , have been the subject of a range of in - depth studies that have unveiled many generic properties of systems with cyclic competition @xcite .
similar studies of more complicated systems composed of multiple species interacting in less trivial ways have been scarce until recently @xcite .
these few investigations of the more complicated cases have been largely restricted to the numerical exploration of the most prominent features . however , in order to develop a better understanding of these cases , analytical approaches are needed . as a first step in that direction we present in this paper some analytical results for a large class of systems that display complicated interaction schemes .
we thereby discuss different dynamics : ( 1 ) lotka - volterra dynamics where the number of individuals is conserved , ( 2 ) may - leonard dynamics where this number is no longer constant , as well as ( 3 ) a mixture of both lotka - volterra and may - leonard dynamics @xcite .
for very general cases we investigate the deterministic mean - field rate equations and determine the space of coexistence fixed points on which species coexist and therefore biodiversity prevails . for cases with a single coexistence fixed point we determine the invariant manifold and study the dynamics around this point . for two - dimensional lattices this is done through a complex ginzburg - landau approach that allows to derive expressions for various quantities of interest .
our paper is organized in the following way . after having introduced our model in section 2 ,
we determine in section 3 the space of coexistence fixed points for very general cases .
we thereby obtain that the dimensionality of that space depends on the chosen dynamics . in section 4
we discuss in more detail cases with a single coexistence fixed point and derive for two - dimensional systems the complex ginzburg - landau equations that allow to describe the dynamics in close vicinity to that fixed point . in order to do
so we allow the particles to be mobile and diffuse on the lattice .
these results are then applied to some selected cases .
finally , in section 5 we discuss some consequences of our results and conclude .
some more technical aspects are discussed in the appendices .
= 3.75 in we consider systems composed of @xmath0 species living on a lattice where every lattice site is either occupied by a single individual or is empty .
individuals on neighbouring sites interact in the following way : @xmath1 where @xmath2 is one individual from species @xmath3 and @xmath4 is an empty site .
in addition we allow particles to be mobile and diffuse on the lattice by hopping to empty neighbouring sites . in this work we only consider the case of species independent reaction rates . whereas the first reaction describes the birth of an off - spring with rate @xmath5 , the other two reactions describe predation events .
reaction ( [ eq : reaction2 ] ) is a may - leonard type reaction where a predator simply removes a prey from the system , thereby changing the total number of individuals . allowing for the possibility of empty sites yields the formation of spiral patterns , see @xcite for a discussion of this point .
every species is thereby preying on @xmath6 other species in a cyclic way , i.e. species @xmath3 is preying on species @xmath7 , @xmath8 , @xmath9 , @xmath10 ( modulo @xmath0 ) .
the second type of predation ( [ eq : reaction3 ] ) is of lotka - volterra type and keeps the total number of individuals constant as a prey is immediately replaced by a predator .
this predation happens again in a cyclic way with each species preying on @xmath11 other species .
for @xmath12 respectively @xmath13 we have a system with may - leonard respectively lotka - volterra dynamics . for the general case , where both @xmath6 and @xmath11 are non zero , we allow for the presence of both types of dynamics .
we call this model the @xmath14 model ( see figure [ fig1 ] for a schematic illustration of the case @xmath15 ) .
the @xmath16 model discussed in @xcite corresponds to the @xmath17 model in this notation . in a spatial setting
this rather simple looking interaction scheme yields a plethora of different space - time patterns .
figures [ fig2 ] and [ fig3 ] show two typical examples in two space dimensions ( see @xcite for other examples ) in absence of mobility .
the ( 3,2,0 ) scheme in figure [ fig2 ] provides an example of coarsening of pure domains : as every species attacks every other species , each individual wants to be surrounded by individuals of the same species , such yielding the situation of complete segregation .
the ( 5,2,0 ) scheme shown in figure [ fig3 ] yields a more complex space - time pattern which results from two different types of spirals that take place in the system at the same time .
these spirals are not permanent and break up easily .
we come back to these different cases later in the paper .
we also note that other types of space time pattern can be realized ( coarsening pattern where each domain contains multiple mutually neutral species that ally in order to fend off other alliances or coarsening pattern where inside the domains non - trivial dynamics emerges due to the fact that the allying species are in a predator - prey relationship ) and refer the interested reader to the paper @xcite for a discussion of these cases .
all these coexistence states are quasi - stationary states : any finite system will eventually end up in an absorbing state where the time needed to enter this final state diverges with the system size @xcite .
= 3.75 in = 3.75 in in the following we first neglect any spatial degrees of freedom . we will consider the spatial dependence later when discussing the complex ginzburg - landau equations .
introducing the time - dependent population densities @xmath18 , the mean - field rate equations for the above reaction scheme read : @xmath19~,\end{aligned}\ ] ] where the first term describes the birth of off - springs in presence of empty sites , whereas the other terms result from the predation events .
the index @xmath20 means @xmath21 and takes care of the cyclic nature of our reaction scheme ( we use this notation for indices throughout the paper ) .
in addition , we have also introduced the shorthand notations @xmath22 $ ] and @xmath23 $ ] , where @xmath24 is the discrete heaviside step function , thus indicating the possible preys for each species .
using the identities @xmath25 and introducing the population density vector @xmath26 allows us to cast the rate equations in the compact form @xmath27 where the elements of the matrix @xmath28 are given by @xmath29 equations ( [ adotvec ] ) and ( [ fij ] ) form the starting point for the following discussion .
in this section we discuss the set of coexistence fixed points as a function of the total number of species @xmath0 and of the number of preys @xmath6 and @xmath11 each species has .
we denote as coexistence fixed point those fixed points for which none of the species is extinct .
the reader should note that for this section it is not needed to have mobile particles .
diffusing particles will be used when we discuss in section 4.3 the complex ginzburg - landau equation for spatial systems . in general ,
as we allow for empty sites , our system evolves within a @xmath30 simplex where each vertex represents the case where one species ( or empty sites ) completely fills the system .
these vertices are absorbing points from which the system described by the equations ( [ adotvec ] ) and ( [ fij ] ) can not escape . needless to say , that these absorbing points represent the complete loss of biodiversity .
the coexistence fixed points , on the other hand , are the steady states where biodiversity is maximal as all species remain present in the system . setting the left hand side of equation ( [ adotvec ] ) to zero ,
these points are given by the solutions of the equation ( @xmath31 being the vector which has the birth rate @xmath5 as each element ) @xmath32 where the vector @xmath33 has only non - zero elements .
as the matrix @xmath28 is circulant , its ( unnormalized ) eigenvectors can be given as @xmath34 with @xmath35 .
in addition , its eigenvalues are @xmath36 after some algebraic manipulations ( see appendix a ) , the last equation can be written in the following form : @xmath37 \right\ } ( 1-\delta_{k , n})\end{aligned}\ ] ] where @xmath38 is the kronecker delta . inspecting the expression ( [ lambda - k ] ) for the eigenvalues , one sees that ( a ) the eigenvalue @xmath39 is always larger than zero and that ( b ) the eigenvalue @xmath40 with @xmath41 is zero if and only if one of the following three conditions is fulfilled : @xmath42 where in the last condition the parentheses @xmath43 are to be understood in the sense of convention ( [ parentheses ] ) .
one immediately remarks that this last condition can not be satisfied for an odd number of species @xmath0 . for the case
@xmath44 , where the dynamics is exclusively of may - leonard type , the expression ( [ lambda - k ] ) for the eigenvalues reduces to @xmath45 so that the condition for vanishing @xmath40 is given by the condition that @xmath0 is a divisor of @xmath46 or , equivalently , that @xmath47 introduction of the matrix @xmath48 and its inverse @xmath49 with elements @xmath50 allows us to rewrite equation ( [ eqf ] ) as @xmath51 and therefore @xmath52 where @xmath53 is the diagonal matrix with the eigenvalues @xmath40 of @xmath28 .
it follows from equation ( [ lambda ] ) that the dimension @xmath54 of the space of coexistence fixed points is determined by the rank of @xmath53 ( we give some additional details regarding this space of coexistence fixed points in appendix b ) : @xmath55 the rank of @xmath53 is readily obtained from the condition ( [ eqk ] ) under which the eigenvalues @xmath40 vanish : @xmath56~,\end{aligned}\ ] ] and we finally obtain that for @xmath44 and @xmath57 the dimension of the space of coexistence fixed points is @xmath58 a similar analysis as for the previous case immediately yields the following result for the dimension of the space of coexistence fixed points for the case @xmath60 and @xmath61 where we have only lotka - volterra type dynamics : @xmath62 -1~.\end{aligned}\ ] ] as mentioned previously , for the general case where @xmath61 and @xmath57 an eigenvalue @xmath40 vanishes if anyone of the three conditions ( [ cond1 ] ) , ( [ cond2 ] ) , or ( [ cond3 ] ) is fulfilled .
the conditions ( [ cond1 ] ) and ( [ cond2 ] ) are dealt with in a very straightforward way , following the same steps as for the previous cases where one of the rates is zero , and one obtains that the dimension of the space of coexistence fixed points is then given by @xmath63^{\equiv n}),(n,[r , r'+1]^{\equiv n})\right\}-1\end{aligned}\ ] ] where @xmath64 \equiv\ ; \text{lcm } ( a , b)$ ] and @xmath64^{\equiv n } \equiv\ ; \theta\left[(n-1)-[a , b]\right]\cdot [ a , b ] + \theta\left[[a , b]-n\right]$ ] .
the condition ( [ cond3 ] ) , which can not be fulfilled if @xmath0 is odd , needs to be treated separately . assuming @xmath0 even , close inspection of ( [ cond3 ] ) reveals that none of the proper divisors @xmath65 of @xmath0 ( which are greater than 1 ) can satisfy this condition .
thus , if ( [ cond1 ] ) and ( [ cond2 ] ) are not fulfilled , the space of coexistence fixed points is zero - dimensional .
this case therefore is also captured by the expression ( [ d_sigma_nu ] ) . summarizing this part of the paper , we find that the dimensionality of space of coexistence fixed points depends on the way the predation events are implemented .
this result highlights the importance of the chosen dynamics and further illustrates that conserved and non - conserved dynamics can yield very different results in systems that are otherwise identical .
in systems with a single coexistence fixed point interesting theoretical insights can be gained by studying the motion around this fixed point .
for that the invariant manifold , i.e. the subspace in phase space which is left invariant by the deterministic rate equations , has to be identified .
adding then the spatial degrees of freedom neglected until now yields a system of complex ginzburg - landau equations that contain information on the spreading velocity of traveling waves or the wavelength and frequency of spiral waves . in @xcite
this calculation has been performed for the three - species cyclic may - leonard model which corresponds to the @xmath66 case in the notation adopted in our paper ( see @xcite for some additional details ) . in the following
we expand this calculations to large classes of systems with rather general interaction schemes , provided the space of coexistence fixed points is zero - dimensional .
we start our discussion with the case of an odd number of species @xmath0 that interact in a cyclic way such that @xmath67 .
it then follows that @xmath68 , so that there exists only a single coexistence fixed point , see the previous section .
the steady state equation ( [ eqf ] ) readily yields the reactive fixed point @xmath69 where all species equally coexist . introducing the coordinates @xmath70 we can rewrite the rate equations as @xmath71 with the matrix @xmath72 and @xmath73 the matrix @xmath74 is diagonalized by means of the matrix @xmath75 , see ( [ eqmatomega ] ) : @xmath76 with the eigenvalues @xmath77 we note that @xmath78 , whereas the other eigenvalues form pairs that are complex conjugate : @xmath79 . defining the new complex coordinates @xmath80 allows to recast equation ( [ eqx ] ) as @xmath81 with @xmath82 where we used that @xmath83 . for the @xmath0-th component
, this expression reduces to @xmath84 as @xmath85 . for the cyclic model under investigation the ( unstable )
invariant manifold is locally spanned by the directions of eigenvalues with positive real parts ( the _ positive _ or _ unstable _ directions ) , i.e. to lowest order the invariant manifold is the plane normal to the eigendirections corresponding to eigenvalues with negative real parts ( the _ negative _ or _ stable _ directions ) . whereas for the previously studied case @xmath86 @xcite as well as for the case @xmath87 the invariant manifold is two - dimensional as only one pair of conjugate eigenvalues has a negative real part , the situation is more complicated for larger values of @xmath0 where multiple eigenvalues have negative real parts . in order to determine the invariant manifold we seek functions of the form @xmath88 where @xmath65 indicates a negative direction whereas the set @xmath89 is the set of coordinates in the positive directions : @xmath90 the summation boundaries are discussed in appendix c. the full problem is obviously of a formidable nature .
luckily , we only need expressions up to second order in the @xmath91 s .
we therefore make the ansatz @xmath92 inserting this ansatz together with the expressions ( [ dotz ] ) and ( [ h_i ] ) into the equation ( [ eqinvman ] ) for the invariant manifold yields @xmath93 from which follows that @xmath94 putting this back into ( [ eqman ] ) leads to the equations @xmath95 that specify the invariant manifold up to quadratic order . in general ,
whenever the space of coexistence fixed point is zero - dimensional we have that @xmath96 , since by symmetry all components of the coexistence fixed point are identical : @xmath97 therefore for the general case @xmath14 with a single coexistence fixed point the eigenvalue @xmath98 takes the form @xmath99 \right\ } \left ( 1 - \delta_{k , n } \right)\end{aligned}\ ] ] following the steps in the previous subsection , one can then write down for the @xmath100 expressions similar to those obtained for the case with @xmath67 ( see equations ( [ h_i ] ) and ( [ h_n ] ) ) : @xmath101 more problematic are the steps that deal with the invariant manifold and the dynamics on this manifold
. indeed , although the @xmath98 s are all non - zero , their real part might be zero depending on the combination of the parameters @xmath0 , @xmath6 , @xmath11 , and @xmath65 as well as on the values of the rates @xmath102 and @xmath103 .
in such cases one needs to perform a further center manifold reduction and go beyond the linear order for the determination of the converging or diverging behavior of those specific directions near the coexistence fixed point . from now on we exclusively focus on those cases for which all @xmath98 s have a nonvanishing real part . under this additional constraint
the local specification of the invariant manifold is straightforward . as before we project the dynamics onto the unstable manifold by expressing the negative or stable directions ( i.e. @xmath65 s with @xmath104 ) , up to quadratic order , in terms of the positive or unstable directions .
this leads to the expression @xmath105 where the shorthand notation @xmath106 means the set of unstable ( positive ) directions , i.e. the sum is over terms where both @xmath107 and @xmath108 label unstable directions . in equation ( [ z_k_gen ] ) the parameters @xmath109
are given by @xmath110 with @xmath111 in vicinity of the unstable fixed point , this brings the rate equations on the unstable manifold to the following form ( @xmath112 ) @xmath113 the next step is to suppress the quadratic terms in the above equations .
this is achieved by means of a family of near identity transformations as @xmath114 where @xmath115 are quadratic polynomials in the coordinates of the unstable directions @xmath116 @xcite : @xmath117 this transformation will dismiss the quadratic terms ( with suitable choice of @xmath118 ) at the price of introducing further cubic terms .
actually , a subset of these cubic terms are welcome as we aim at ending up with a family of stuart - landau normal forms .
the rest , however , is redundant and can again be suppressed by another family of near identity transformations @xmath119 where @xmath120 is cubic in coordinates corresponding to the unstable directions : @xmath121 here the bar indicates that the @xmath122 terms , needed for the stuart - landau normal forms , are preserved in the process of transforming the rate equations ( [ rate - uns ] ) .
careful performance of transformations ( [ hr2 ] ) and ( [ hr3 ] ) then results in the following set of stuart - landau normal forms _ on the unstable manifold _ ( @xmath112 ) : @xmath123 restricting ourselves to the directions with negative imaginary part of the eigenvalue , this can be brought in the standard form @xmath124 with @xmath125 recalling that we allow for mobile particles , we want to incorporate at this stage the mobility of particles that can diffuse by jumping to an empty neighbouring site .
following @xcite we then obtain in the continuum limit the following reaction - diffusion equations for the space- and time - dependent particle densities : @xmath126~,\end{aligned}\ ] ] that differ from the rate equations ( [ adot ] ) by the diffusion term @xmath127 , where @xmath128 is the diffusion constant that depends on the rate for jumps into unoccupied neighbouring sites .
note that the realization of mobility through jumps into empty sites yields nonlinear diffusive terms in addition to the usual linear term @xcite . as we expect the dynamics to be dominated by the long wavelength modes , we only keep the leading order gradient term in equation ( [ rd ] ) . when redoing the calculation of the previous subsection with this equation , one notes the appearance of nonlinear diffusive terms due to the nonlinearities of the coordinate transformations .
these nonlinear terms are expected to be subleading when the dynamics is dominated by the long wavelength modes @xcite .
we therefore ignore these additional terms and end up with the following system of partial differential equations : @xmath129 where @xmath116 labels the unstable directions with negative imaginary parts .
this is a set of complex ginzburg - landau equations similar to those that have been studied extensively in the past in a large variety of different physical situations @xcite .
it should be stressed that neglecting the nonlinear diffusive terms is an uncontrolled approximation and that _ a posteriori _ tests are needed to check how reliable this approximation really is .
for the three - species case ( 3,1,0 ) , where the same assumptions have been made , it was checked through numerical simulations that the resulting complex ginzburg - landau equation faithfully describes some of the main features of the corresponding lattice model @xcite .
similar tests for checking the reliability of this approximation should also be done for the larger class of models studied in our paper , and we plan to come back to this important aspect in the future . from these equations a variety of quantities can be computed .
first we note that in the case @xmath130 and @xmath131 the complex ginzburg - landau equation reduces to the time - dependent real ginzburg - landau equation used commonly for the description of coarsening in ferromagnets @xcite ( see @xcite for a comparison between the dynamics in the complex and the real ginzburg - landau equations ) . for the case @xmath132 and @xmath133 the corresponding complex ginzburg - landau equation allows to describe spiral waves and derive analytical expressions for a range of relevant quantities @xcite .
for example for the linear spreading velocity @xmath134 one obtains @xmath135 whereas the wavelength @xmath136 and the frequence @xmath137 of the spiral are given by @xmath138 let us now apply this formalism to the two examples given in section 2 . for the case ( 3,2,0 ) , see figure [ fig2 ] , we have only one unstable direction .
a straightforward calculation then yields the expressions @xmath139 for the four parameters in equation ( [ diff ] ) . as @xmath140
the complex equation reduces to the real one .
consequently , one expects for this interaction scheme the appearance of coarsening domains , in complete agreement with the pattern showing up in figure [ fig2 ] .
the case ( 5,2,0 ) , see figure [ fig3 ] , represents a more complicated situation with two unstable directions @xmath141 and @xmath142 and two complex ginzburg - landau equations .
applying our formalism to this case yields the following expressions : @xmath143 one notes that @xmath144 but @xmath145 .
this describes a situation of two different types of spirals with the same spreading velocities but different wavelengths .
close inspections of snapshots like that shown in figure [ fig3 ] reveals indeed the presence of two different types of spirals , characterized by different thicknesses of their spiral arms ( which corresponds to different wavelengths ) , that interfere continually .
in this paper we have derived complex ginzburg - landau equations for @xmath0 species models in two dimensions with a large range of different interaction schemes given by the reactions ( [ eq : reaction1])-([eq : reaction3 ] ) . depending on the values of the parameters ,
a range of scenarios can be realised , as for example coarsening of pure domains , single spiral waves , or interacting multiple spiral waves .
numerical simulations of the corresponding lattice gas models show a quantitative agreement with the predictions that follow from the values of the parameters in the complex ginzburg - landau equations .
however , as neglecting the nonlinear diffusive terms that emerge when applying the nonlinear transformations is an uncontrolled approximation , more advanced checks through the quantitative comparison of the predictions from the complex ginzburg - landau equations and the results from lattice model simulations are needed in order to fully assess the reliability of this approximation .
it should be noted that recently an alternative approach was proposed that allows to derive in a more controlled way , through a perturbative expansion , a set of complex ginzburg - landau equations for systems with cyclic competition and an additional mutation process @xcite .
this method has been used successfully for the three - species case ( 3,1,0 ) .
the same approach should also work for the larger class of models discussed in our manuscript , and it would be interesting to compare our results with results obtained from this alternative approach .
there are two standard ways to implement mobility in systems where at most one particle is allowed at any lattice site : diffusion , realised through the hopping of particles to empty sites , and particle swapping , where particles on neighbouring sites exchange places . in this paper
we restricted ourselves to the first case .
however , the space - time pattern may change depending on the chosen way to implement mobility . as an example we show in figure [ fig4 ] the ( 5,2,0 ) case with swapping , which should be compared to figure [ fig3 ] which shows a snapshot of the same system in absence of mobility . using swapping yields pronounced and very stable spirals , characterised by wave number and frequency that differ from those obtained from the case with diffusion .
we will address the cases with swapping in a separate publication .
= 3.75 in obviously our numerical simulations are of stochastic nature , but we exclusively focused on deterministic equations in our theoretical treatment .
one may then wonder how the inclusion of noise , that yields a system of coupled stochastic partial differential equations , would change our results . a first , rather trivial remark , is that in simulating a finite system one eventually ends up with the extinction of some of the species @xcite .
we therefore focus on earlier times that are much shorter than any extinction times .
for the three - species may - leonard model , the model ( 3,1,0 ) in our notation , it was found @xcite that the numerically determined spreading velocity quantitatively agrees with the prediction from the complex ginzburg - landau equation , and this for a large range of values of the reaction rates .
the functional dependence of the wavelength was also found to agree between numerics and theory .
however , other interaction schemes , especially for more than three species , have not been studied in the same way .
it is therefore an important open question whether a similar good quantitative agreement can be achieved for more complicated cases .
we plan to come back to this question in the future .
another interesting question concerns the cases where at least one of the eigenvalues @xmath98 ( [ lambdak ] ) has a vanishing real part . in that case
an additional center manifold reduction has to be performed in order to go beyond the linear order .
this is a very demanding calculation that we leave for the future .
although our study deals with a large class of systems , the derivation of the complex ginzburg - landau equations remains restricted to cases where the space of coexistence fixed points is zero - dimensional .
it is an open problem whether similar theoretical insights can be gained for cases where that space is of higher dimension . + * acknowledgements : * + + this work is supported by the us national science foundation through grant dmr-1205309 .
we thank uwe c. tuber and darka labavi for useful discussions .
for the explicit calculation of the eigenvalues ( [ lambda - k ] ) we start from the expression @xmath146 writing out the element of the matrix @xmath28 , we obtain @xmath147 ) } + \tilde{\nu}_{(1-[n - j+1 ] ) } - \tilde{\nu}_{([n - j+1]-1 ) } \notag\\ = & \ ; \mu + \tilde{\sigma}_j + \tilde{\nu}_j - \tilde{\nu}_{(-j ) } \notag\\ = & \ ; \mu + \sigma\ ; \theta[r - j ] + \nu\ ; \theta[r'-j ] - \nu\ ; \theta [ r ' - ( n - j)]\end{aligned}\ ] ] and therefore @xmath148 + \nu \sum_{j=1}^n \omega^{jk } \theta\,[r'-j ] - \nu \sum_{j=1}^n \omega^{jk } \theta\,[r ' -(n - j ) ] \notag\\ = & \ ; n \mu\ , \delta_{k , n } + \sigma \sum_{j=1}^r \omega^{jk } + \nu \sum_{j=1}^{r ' } \omega^{jk } - \nu \sum_{j = n - r'}^{n-1 } \omega^{jk}\end{aligned}\ ] ] where @xmath38 is the kronecker delta . inserting @xmath149 into equation ( [ lambda - k- ] )
finally yields the expression ( [ lambda - k ] ) for the eigenvalues .
looking back at equation ( [ lambda2 ] ) we can see that for the @xmath65 s of ( [ eqk ] ) the corresponding components of the vector @xmath151 in ( [ lambda ] ) are arbitrary .
the number of these components is clearly @xmath152 and , as a result , so is the dimension of the space of coexistence fixed points .
as for other components of this vector , the last component is simply @xmath153 , and all the remaining components vanish since their corresponding @xmath40 s are nonvanishing in ( [ lambda ] ) . but
what does this mean for the vector @xmath154 itself ? to answer this question we focus on the equation @xmath155 one can show from the properties of the matrix @xmath156 that , ignoring the last row @xmath157 as well as the rows @xmath65 such that @xmath158 , with @xmath128 a proper divisor of @xmath0 , the most general solution of the remaining rows in the above equation is of the form @xmath159 for a set of _ arbitrary _ constants @xmath160 , where @xmath161 .
as for the rows that we just ignored , this solution produces nonvanishing constants for the right hand side of ( [ omega - a ] ) ( so long as the constants @xmath160 are nonvanishing ) . but
this is exactly the freedom we have for the @xmath65 s of ( [ eqk ] ) , and therefore we can exploit this .
all we need to do is to choose suitable proper divisors @xmath128 from the condition @xmath158 such that the spectrum of admissible @xmath65 s fits into that of ( [ eqk ] ) .
the possible values for @xmath128 are then @xmath162 where @xmath107 is any natural number that satisfies @xmath163 . for each such @xmath107
then we have a family of solutions of ( [ lambda ] ) as @xmath164 where @xmath165 the restriction to @xmath166 is due to the _ coexistence _ condition , enforcing the positivity of all densities .
the subspace @xmath167 in the codomain @xmath168 is defined via @xmath169 and reflects the constraint on the @xmath170 s imposed by the last row of ( [ lambda ] ) .
@xmath167 is removed from @xmath171 to , again , ensure coexistence .
considering the stable directions @xmath65 of ( [ eqinvman ] ) and ( [ eqman ] ) , in order to identify the quadratic contributions of unstable directions in the expansion @xmath172 of ( [ h_i ] ) , we need to pick only those combinations @xmath173 for which both @xmath107 and @xmath108 correspond to eigenvalues of ( [ lambda_k ] ) with positive real parts . noting that the latter equation carries an overall negative sign and that we are looking at the case @xmath174 with odd @xmath0 , it is then straightforward to recognize the corresponding indices @xmath107 and @xmath108 as those belonging to the interval @xmath175 $ ] which is characterized as the union of quadrants @xmath176 and @xmath177 in figure [ fig5 ] .
= 3.75 in this also implies that the set of indices that correspond to negative eigenvalues belong to the union of quadrants @xmath178 and @xmath179 ( except for @xmath157 which exhausts the entire interval @xmath175 $ ] ) .
one can show that for @xmath180 the range of @xmath107 such that both @xmath107 and @xmath108 belong to @xmath181 is @xmath182 $ ] , whereas for @xmath183 this range is @xmath184 $ ] .
these two intervals can then be combined into one , namely , @xmath185 $ ] for @xmath186 via the following identifications ( @xmath24 is the discrete heaviside step function ) : @xmath187 note that the special case @xmath157 is also captured by the above bounds .
99 may r m 1974 _ stability and complexity in model ecosystems _
( cambridge , england : cambridge university press ) maynard smith j 1974 _ models in ecology _ ( cambridge , england : cambridge university press ) sole r v and basecompte j 2006 _ self - organization in complex ecosystems _ ( princton , nj : princeton university press ) .
szab g and fth g 2007 _ phys .
* 446 * 97 frey e 2010 _ physica a _ * 389 *
4265 frachebourg l , krapivsky p l , and ben - naim e 1996 2125 frachebourg l , krapivsky p l , and ben - naim e 1996
6186 kobayashi k and tainaka k 1997 38 frachebourg l and krapivsky p l 1998 l287 provata a , nicolis g , and baras f 1999 _ j. chem .
phys . _ * 110 * 8361 tsekouras g a and provata a 2001 016204 sato k , yoshida n , and konno n 2002 _ appl .
* 126 * 255 kerr b , riley m a , feldman m w , and bohannan b j m 2002 _ nature _ * 418 * 171 szab g and sznaider g a 2004 031911 kirkup b c and riley m a 2004 _ nature _ * 428 * 412 he m , cai y , and wang z 2005 _ int . j. mod . phys .
c _ * 16 * 1861 reichenbach t , mobilia m , and frey e 2006 051907 szab g , szolnoki a , and sznaider g a 2007 051921 reichenbach t , mobilia m , and frey e 2007 238105 reichenbach t , mobilia m , and frey e 2007 _ nature _ * 448 * 1046 szab g and szolnoki a 2008 011906 claussen j c and traulsen a 2008 058104 peltomki m and alava m 2008 031906 reichenbach t and frey e 2008 058102 reichenbach t , mobilia m , and frey e 2008 _ j. theor .
* 254 * 368 berr m , reichenbach t , schottenloher m , and frey e 2009 048102 venkat s and pleimling m 2010 021917 case s o , durney c h , pleimling m , and zia r k p 2010 _ epl _ * 92 * 58003 shi h , wang w - x , yang r , and lai t - c 2010 030901(r ) andrae b , cremer j , reichenbach t , and frey e 2010 218102 wang w - x , lai y - c , and grebogi c 2010 046113 mobilia m 2010 _ j. theor .
biol . _ * 264 * 1 he q , mobilia m , and tuber u c 2010 051909 winkler a a , reichenbach t , and frey e 2010 060901(r ) noble a e , hastings a , and fagan w f 2011 228101 durney c h , case s o , pleimling m , and zia r k p 2011 051108 he q , mobilia m , and tuber u c 2011 _ eur . phys .
j. b _ * 82 * 97 rulands s , reichenbach t , and frey e 2011 _ j. stat .
wang w - x , ni x , lai y - c , and grebogi c 2011 011917 nahum j r , harding b n , and kerr b 2011 _ pnas _ * 108 * 10831 jiang l l , zhou t , perc m , and wang b h 2011 021912 platkowski t and zakrzewski j 2011 _ physica a _ * 390 * 4219 demirel g , prizak r , reddy p n , and gross t 2011 _ eur .
j. b _ * 84 * 541 zia r k p 2011 arxiv:1101.0018 .
he q , tuber u c , and zia r k p 2012 _ eur .
j. b _ * 85 * 141 dong l and yang g 2012 _ physica a _ * 391 * 2964 dobrinevski a and frey e 2012 051903 juul j , sneppen k , and mathiesen j 2012 061924 lamouroux d , eule s , geisel t , and nagler j 2012 021911
jiang l - l , wang w - x , lai y - c , and ni x 2012 _ phys .
lett . a _ * 376 * 2292 adamson m w and morozov a y 2012 _ bull . math .
biol . _ * 74 * 2004 juul j , sneppen k , and mathiesen j , 2013 042702 durney c h , case s o , pleimling m , and zia r k p 2012 _ j. stat .
p06014 roman a , konrad d , and pleimling m 2012 _ j. stat .
mech . _ p07014 rulands s , zielinski a , and frey e 2013 052710 szczesny b , mobilia m , and rucklidge a m 2013 _ epl _ * 102 * 28012 intoy b and pleimling m 2013 _ j. stat .
_ p08011 guisoni n c , loscar e s , and girardi m 2013 022133 park j , do y , huang z g , and lai y c 2013 _ chaos _ * 23 * 023128 schreiber s j and killingback t p 2013 _ theor .
biol . _ * 86 * 1 szab g and czrn t 2001 061904 szab g and czrn t 2001 042902 szab g 2005 6689 szab p , czrn t , and szab g 2007 _ j. theor .
biol . _ * 248 * 736 perc m , szolnoki a , and szab g 2007 , 052102 szab g , szolnoki a , and borsos i 2008 041919 vandermeer j and yitbarek s 2012 _ j. theor .
* 300 * 48 ltz a f , risau - gusman s , and arenzon j j 2013 _ j. theor .
biol . _ * 317 * 286 avelino p p , bazeia d , losano l , and menezes j 2012 031119 avelino p p , bazeia d , losano l , menezes j , and de oliveira b f 2012 036112 roman a , dasgupta d , and pleimling m 2013 032148 knebel j , krger t , weber m f , and frey e 2013 168106 avelino p p , bazeia d , menezes j , and de oliveira b f 2014 _ physics letters a _ * 378 * 393 vukov j , szolnoki a , and szab g 2013 022123 kang y b , pan q h , wang x t , and me h f 2013 _ physica a _ * 392 * 2652 avelino p p , bazeia d , losano l , menezes j , and de oliveira b f 2013 arxiv:1312.1859 dobrinevski a , alava m , reichenbach t , and frey e 2014 arxiv:1401.1755 wiggins s 2003 _ introduction to applied nonlinear dynamical systems and chaos _ ( new york : springer verlag ) cross m c and hohenberg p c 1993 _ rev . mod . phys . _ * 65 * 851 aranson i s and kramer l 2002 _ rev .
phys . _ * 74 * 99 bray a j 2002 _ adv .
phys . _ * 51 * 481 patra s and das s k 2013 _ eur .
j. e _ * 36 * 130 | in order to model real ecological systems one has to consider many species that interact in complex ways .
however , most of the recent theoretical studies have been restricted to few species systems with rather trivial interactions .
the few studies dealing with larger number of species and/or more complex interaction schemes are mostly restricted to numerical explorations . in this paper
we determine , starting from the deterministic mean - field rate equations , for large classes of systems the space of coexistence fixed points at which biodiversity is maximal . for systems with a single coexistence fixed point
we derive complex ginzburg - landau equations that allow to describe space - time pattern realized in two space dimensions . for selected cases
we compare the theoretical predictions with the pattern observed in numerical simulations . |
over the last fifteen years , the dynamics of glassforming liquids under nanoscale confinement has attracted much attention .
a great variety of molecular systems under many different types of confinement has been subjected to virtually all available experimental techniques , while at the same time extensive studies of simple model systems using molecular dynamics simulations were undertaken @xcite .
the main reason for this rapidly growing interest is that confinement is considered as a potential tool to investigate the concept of cooperativity , a key ingredient of many glass transition theories @xcite . indeed , there are now many evidences that the dynamics of deeply supercooled liquids is inhomogeneous and that dynamically correlated groups of molecules play a crucial role in the slowing - down of the dynamics when the temperature is decreased .
but , up to now , many questions pertaining , for instance , to the shape , size or temperature evolution of these dynamical heterogeneities remain essentially unanswered ( for recent progress , see , however , ref .
@xcite ) . in confinement ,
geometric constraints associated with the pore shape are imposed to the adsorbed fluid and new characteristic length scales , like the pore size , come into play .
thus , by looking for alterations in the dynamics under confinement compared to the bulk , one can hope to pinpoint some of the elusive characteristic features of the dynamical heterogeneities .
for instance , in the simplest scenarios , deviations from the bulk behavior are expected to occur due to finite size effects , when the typical size of the dynamical heterogeneities in the bulk would become larger than the pore size .
confinement effects would thus provide a ruler to measure dynamical heterogeneities .
the situation actually turns out more complex .
indeed , a direct comparison between the bulk and confined fluids is only meaningful if the physical phenomena which are specific to confinement have a weak impact on the properties of the imbibed fluid or at least if their influence is sufficiently well understood that it can be corrected for .
this is usually not so and strong confinement effects are often observed , for instance , the formation of structured layers of almost immobile molecules at the fluid - solid interface .
so , in fact , dynamics in confinement should be addressed as a problem of its own , not necessarily with reference to the bulk .
the variety of the systems to consider is immense .
porous media can differ in the size , shape and topology of their pore space .
they can be made of various materials or receive different surface treatments , leading to a wide range of fluid - solid interactions which adds to the already great variability of the intermolecular interactions met with usual glassformers .
thus , owing to the complexity of the field , a reasonable microscopic theory , able to catch at least some of these aspects , could be very helpful .
indeed , applied to various models , it would allow to explore thoroughly the phenomenology of confined glassforming liquids and maybe to disentangle the contributions of the different physical phenomena which interplay in these systems . many porous media , like controlled porous glasses and aerogels , are disordered . in the past few years ,
a very useful and quite successful model to deal with this kind of systems has been the so - called `` quenched - annealed '' ( qa ) binary mixture , first introduced by madden and glandt @xcite . in this model , sketched in fig .
[ figsketch ] , the fluid molecules ( the annealed component ) equilibrate in a matrix of particles frozen in a disordered configuration sampled from a given probability distribution ( the quenched component ) . the matrix is assumed to be statistically homogeneous , so that , while for any single realization , the system lacks translational and rotational invariance , all expectation values computed with the matrix probability distribution will have the same properties as in a truly translationally and rotationally invariant system . a common , but not unique ,
prescription is to take the equilibrium distribution of some simple fluid system , so that the various samples of the matrix can be thought of as the results of instantaneous thermal quenches of this original equilibrium system , hence the denomination `` quenched '' for the matrix component .
thanks to the assumption of statistical homogeneity , as far as the computation of matrix averaged quantities is concerned , qa mixtures can be studied with great ease using simple extensions of standard liquid state theoretical methods .
this has been put to good use to derive equations describing the structure and the thermodynamics of these systems , either via diagrammatic techniques @xcite or by application of the replica trick @xcite .
sketch of a qa system . in black , the immobile matrix particles . in white , with arrows symbolizing their movement , the fluid particles . ]
the aim of the present work is now to develop a dynamical theory for qa mixtures , able to deal with the problem of the glass transition in confinement .
this will take the form of an extension to qa systems of the ideal mode - coupling theory ( mct ) for the liquid - glass transition @xcite .
the mct occupies a central place in the study of the dynamics of supercooled liquids in the bulk , as one of the very few available microscopic theories in this field .
it has well known deficiencies , in particular the fact that its predicted sharp ergodicity breaking transition , the so - called ideal glass transition , is always located in the regime of weak to mild supercooling , rather far from the calorimetric glass transition point .
but , quite remarkably , it has been able to correctly predict novel nontrivial relaxation patterns which develop in systems like colloidal suspensions with short - ranged attractions @xcite or fluids of symmetric dumbbells @xcite , when the parameters of these models are varied .
it seems thus sensible to turn to this theoretical framework for a systematic investigation of confined glassforming liquids , aiming at understanding , at least qualitatively , how the microscopic details of the fluid - solid system impact its dynamics .
moreover , it is encouraging that recent computer simulation data for confined fluids could be interpreted with the universal predictions of the mct @xcite , showing that the mode - coupling glass transition scenario might indeed be of some relevance in confinement as well .
note that ref .
@xcite precisely deals with a qa system , as do refs .
@xcite , where other simulation studies are reported .
theories related to the one to be derived have already been discussed in the literature .
the mode - coupling approach to the diffusion - localization transition in the classical random lorentz gas @xcite is of particular relevance , since this system is actually a qa mixture taken in the limit of a vanishing density of the annealed component . besides
, the present theory will borrow some ideas from this approach .
other works have dealt with the conductor - insulator transition of quantum fluids in random potentials , originally neglecting the fluid - fluid interactions @xcite which were later reintroduced by thakur and neilson @xcite .
they appear as special cases of the present theory , with additional implicit and uncontrolled approximations in the treatment of the effect of randomness on the static fluid correlations @xcite .
finally , a similar nonlinear feedback mechanism has been derived for the freezing of a polymer chain in a quenched random medium using the self - consistent hartree approximation for the langevin dynamics of the system @xcite .
the paper is organized as follows . in sec .
ii , the mct equations for the collective dynamics of a qa mixture are derived and discussed , while sec .
iii explores the relation between the dynamics in a self - induced glassy phase and in a quenched random environment . in sec .
iv , dynamical phase diagrams for two simple models are computed , which show two basic types of ideal liquid - glass transitions discussed in more details in sec . v. section vi is devoted to concluding remarks .
preliminary reports on the present work can be found in refs .
@xcite and @xcite .
in this section , the mode - coupling equations for the qa binary mixture are derived and discussed .
but , before proceeding with the dynamical theory , a few static quantities have to be defined .
as mentioned in the introduction , in a qa system , the disordered porous medium is represented by a collection of @xmath0 rigorously immobile point particles , randomly placed in a volume @xmath1 at positions denoted by @xmath2 , according to a given probability distribution @xmath3 @xcite .
its overall density is @xmath4 and the fourier components of its frozen microscopic density , or , in short , its frozen density fluctuations , are given by @xmath5 where @xmath6 denotes the wavevector .
their disorder - averaged correlation functions , which , because of the assumed statistical homogeneity , are diagonal in @xmath6 and only depend on its modulus @xmath7 , define the matrix structure factor @xmath8 where @xmath9 denotes an average over the matrix realizations .
the fluid component consists of @xmath10 point particles ( density @xmath11 ) of mass @xmath12 , which equilibrate at a temperature @xmath13 in the random potential energy landscape created by the frozen matrix particles . as in the bulk , for the present theory
, we will be interested in the dynamics of the fluid density fluctuations @xmath14 where @xmath15 is the position of the fluid particle @xmath16 at time @xmath17 . at equal times
, they allow to define the fluid structure factor @xmath18 and the fluid - matrix structure factor @xmath19 where @xmath20 denotes a thermal average _ taken for a given realization of the matrix _ ,
the disorder average @xmath9 being performed _
subsequently_. for an ordinary binary mixture , the knowledge of the above three structure factors would be enough to fully characterize the structure at the pair level .
this is not the case for a qa mixture . indeed , because the matrix component is quenched , for any single realization , the system lacks translational and rotational invariance .
it results that , at variance with a bulk fluid , non - zero average density fluctuations exist at equilibrium , i.e. , @xmath21 .
it is only after averaging over disorder that the symmetry is restored , so that @xmath22 .
thus , one is led to consider relaxing and non - relaxing fluid density fluctuations , corresponding to @xmath23 and @xmath24 , respectively , and to define the connected fluid structure factor @xmath25 and the disconnected or blocked fluid structure factor @xmath26 such that @xmath27 .
this splitting of the fluid pair correlations is well known from the replica theory of qa systems , where it leads to the peculiar structure of the so - called replica ornstein - zernike ( oz ) equations @xcite , which are given in appendix [ app.oz ] for reference .
non - zero average fluid density fluctuations at equilibrium mean that , even without any dynamical ergodicity breaking , they will have time - persistent correlations . indeed , defining the normalized total density fluctuation autocorrelation function @xmath28
one expects using standard arguments that @xmath29 this is a general consequence of the fact that the fluid evolves in an inhomogeneous environment and we stress that this is a true static phenomenon . usually , mode - coupling theories are derived assuming that the statics of the problem is solved . for this reason , since the calculation of the contribution from the blocked correlations is actually a static problem , the theory has been developed using the relaxing part of the fluid density fluctuations as the central dynamical variable .
attempts to derive a dynamical theory starting from the full density fluctuations have resulted in complicated equations , which appeared unfaithful to the statics of the problem , and thus were abandoned .
the theory is derived using standard projection operator methods , as shown in ref .
@xcite for bulk systems . as usual , in a first step , one obtains a generalized langevin equation for the time evolution of the normalized connected autocorrelation function of the density fluctuations @xmath30 it is formally the same as for the bulk , i.e. , @xmath31 with initial conditions @xmath32 , @xmath33 , and @xmath34 the second step involves the calculation of the slow decaying portion of the memory kernel @xmath35 with a mode - coupling approach .
we will assume that the slow dynamics is dominated by three types of quadratic variables , which can be separated in two classes .
in the first class , we have variables quadratic in the relaxing density fluctuations , @xmath36 , in close analogy with bulk mct @xcite . in the second class , inspired by previous studies on the lorentz gas @xcite
, we consider variables expressing couplings of the relaxing density fluctuations to the two frozen density fluctuations present in the problem , @xmath37 and @xmath38 .
the last variable was omitted in all previous works , including a recent account of the present theory @xcite , with consequences to be discussed below .
the calculation is outlined in appendix [ app.derivation ] and we only quote the result here .
it reads @xmath39 , where @xmath40 is a friction coefficient associated with fast dynamical processes , and @xmath41,\ ] ] with [ vertices ] @xmath42 ^ 2 s^{c}_k s^{c}_{|\mathbf{q - k}|},\ ] ] and @xmath43 ^ 2 s^{c}_k s^{b}_{|\mathbf{q - k}|},\ ] ] where @xmath44 , the fourier transform of the connected direct correlation function , has been introduced .
based on these equations , a few general comments are in order .
first of all , the derived expressions retain the mathematical structure of the typical mode - coupling equations which have been extensively studied in ref . @xcite .
thus , all known properties of the solutions of mct equations , in particular near the transition , apply in the case of the qa mixture .
a significant addition compared to the bulk is however the presence of a linear term in the memory function , which opens the possibility of continuous ideal glass transitions .
like in the theory for the bulk , which is recovered in the limit @xmath45 , the slow dynamics is fully determined by smoothly varying static quantities .
interestingly , no explicit reference to the matrix is visible in the equations .
indeed , the only required information is @xmath46 , @xmath47 , @xmath48 , and @xmath44 , i.e. , quantities characterizing the fluid component of the qa system .
these functions and the relations between them are generically meaningful for the description of fluids evolving in statistically homogeneous random environments and they are by no means restricted to the model of the qa mixture ( for the case of non - particle - based random fields , see refs .
therefore , one can expect that the present dynamical theory shares the same degree of generality and is applicable in its present form out of the strict context of the qa binary mixture .
in fact , this shows up nicely in the process of deriving the equations , since one finds that the contributions resulting from the coupling of the random forces to @xmath37 are identically zero ( see appendix [ app.derivation ] ) .
thus , it is enough to consider only two types of quadratic variables , @xmath49 and @xmath50 , precisely those which are not specific to qa systems , to obtain the same dynamical equations .
further evidence of the generic character of the present equations is given in appendix [ app.pspin ] , where the dynamics of a mean - field spin - glass model in a random magnetic field is shown to obey equations with exactly the same structure .
along the same line , it is noteworthy that the total fluid structure factor @xmath51 does not appear in the derived dynamical equations .
only @xmath47 and @xmath48 do .
it results that , in the framework of the mct , the global fluid correlations are of limited relevance to discuss the relation between the statics and the dynamics .
a much more important aspect is rather the balance between the connected and disconnected contributions to these correlations .
this result is very welcome , since similar differences in the roles played by these three types of correlations are known to be crucial in the physics of qa systems .
this is best illustrated by the compressibility sum rule @xcite , which precisely involves @xmath47 and not @xmath51 , as one would naively expect from the relation for the bulk .
it is thus reassuring that this feature has been preserved despite the uncontrolled approximations involved in the derivation of the theory .
moreover , this finding allows one to clarify the issues raised in refs .
@xcite about the possibility of a description of the dynamics in confinement with theories based on structural quantities only .
indeed , it was observed there that systems with identical global fluid correlations could have significantly different dynamics , a result which appeared as a challenge to such approaches .
the present mct shows that this is not necessarily so and that one should also consider the connected and blocked correlations before a conclusion can be reached .
about the form of the vertices , one can note the familiar expression of @xmath52 , which measures the coupling of the random forces to @xmath49 .
it is the same as in a bulk system , with connected quantities simply replacing the fluid structure factor and direct correlation function .
this is true for @xmath53 , given by eq .
, as well .
thus , we find that the relaxing part of the density fluctuations in a qa system just behaves dynamically like the corresponding degree of freedom in the bulk .
this is not really surprising , since a similar correspondence is already visible in the statics , for instance in the oz equation or in the convolution approximation , which are both used in the calculation of @xmath52 .
such a correspondence was missing in all our attempts to derive a mct starting from the total density fluctuations and this is one of the reasons for which the resulting equations were considered unsatisfactory .
@xmath54 is less obvious and combines features of @xmath52 and of the vertex for the tagged particle dynamics in the bulk @xcite . in particular , like the latter , it diverges when @xmath55 . finally , the above equations differ slightly from those reported in ref .
@xcite , where a first account of the theory was given .
there , @xmath56 in eq .
is replaced by @xmath57 , where @xmath58 is the fourier transform of the blocked direct correlation function .
the only difference between the two approaches is that @xmath59 was not included as a slow variable in the earlier mode - coupling scheme .
a priori , this is an unjustified approximation . indeed ,
as shown in appendix [ app.derivation ] , the coupling to this variable is actually so strong , that , when both @xmath60 and @xmath50 are considered , the contribution of the former becomes identically zero , while the effect of the random environment is integrally transferred to the latter .
there are nevertheless circumstances in which the absence of @xmath59 looks perfectly well motivated , for instance when a single particle is moving in the porous matrix , corresponding to the limit @xmath61 , or when the adsorbed fluid is an ideal gas @xcite . in both cases
, the present mct reduces to @xmath62 and @xmath63 ^ 2 \hat{h}^{b}_{|\mathbf{q - k}| } \phi_{k}(t),\ ] ] where @xmath64 is the fourier transform of the blocked total pair correlation function .
but , since the only forces exerted on the fluid are those due to the random matrix , on physical grounds , the only mode - coupling contributions to the relaxation kernel are expected to come from @xmath65 .
all previous theories of the lorentz gas have been based on this insight , which leads to the same equations as above , except that @xmath66 in eq .
is replaced with @xmath67 @xcite .
note that , for an adsorbed ideal gas , both theories correctly predict that the dynamics is independent of @xmath46 .
it seems thus that there is a subtle interplay between the approximations involved in the derivation of the mct and the peculiar structure of the static correlations in the qa mixture . at present , it is not clear what to conclude from this observation , but it is interesting that the difference between the two mct schemes involves the blocked direct correlation function @xcite .
indeed , this is a rather delicate object , not easily captured by simple approximations .
this is best understood in the replica framework , where @xmath68 is obtained as the zero replica limit of the direct correlation function between two non - interacting fluid replicas only correlated through their common interaction with the matrix @xcite .
it results that @xmath68 has a highly non - additive character and that , for instance , most simple closures of the replica oz equations fail to provide expressions for this function which do not vanish identically .
thus , it is a rather nontrivial finding that the mct approximation scheme is actually sensitive to the existence of this function .
from the previous discussion , it is clear that the separation of the fluid density fluctuations into relaxing and frozen parts plays a crucial role in the derivation of the mct for qa systems . in that case , the freezing is of static origin , due to the random external field generated by the quenched matrix .
but , as it is well known from the mct for bulk fluids , freezing of the density fluctuations may also occur dynamically , when the system enters in the ideal glassy state .
it seems thus interesting to compare both situations .
this can be seen as a case of self - induced versus quenched disorder , similar to what has been discussed many times in the literature @xcite .
the residual relaxation of a bulk fluid in its ideal glassy state has been studied in ref .
it is described by mode - coupling equations of the same form as above , except that the characteristic frequency is given by @xmath69 and the vertices are [ verticesresidual ] @xmath70 ^ 2 \left\{(1-f_k ) s_k\right\ } \left\{(1-f_{|\mathbf{q - k}| } ) s_{|\mathbf{q - k}|}\right\}\ ] ] and @xmath71 ^ 2 \left\{(1-f_k ) s_k \right\ } \left\ { f_{|\mathbf{q - k}| } s_{|\mathbf{q - k}| } \right\},\ ] ] where @xmath46 is the density of the fluid , @xmath72 its structure factor , @xmath73 the fourier transform of its direct correlation function , and @xmath74 the debye - waller factor of the glass .
accordingly , @xmath75 corresponds to the dynamically frozen part of the density fluctuations , while @xmath76 corresponds to their relaxing part . the analogies between these equations and eqs . and are striking .
there is almost a perfect correspondence between @xmath76 and @xmath77 on the one hand , and @xmath75 and @xmath78 on the other hand .
thus , we find that , irrespective of the mechanism of freezing of the fluid density fluctuations , the resulting relaxing and frozen contributions essentially play the same role in both systems .
but there remains one significant difference between both sets of equations .
indeed , one of the fourier transformed direct correlation functions in eq . , the one which carries the same wavevector as the frozen part of the structure factor , is replaced by a simple constant factor @xmath79 in eq . .
the origin of this constant term should probably be traced back to the asymmetric nature of the qa system , where the fluid reacts to the matrix but not the other way around , resulting in a non - equilibrium character of its static and dynamical correlations , in the sense that the matrix is not equilibrated with the fluid .
it would then reflect the lack of dynamical self - consistency between the glassy dynamics in a qa system and the frozen background on top of which it develops .
it is tempting to try and obtain the mct for the qa binary mixture as a limiting case of the mct for the ordinary binary mixture @xcite , where one component would become the quenched matrix .
the above discussion shows that this is not possible .
indeed , based on heuristic considerations , one can get close to eqs .
and , for instance by canceling all terms which would result in a time dependence of the fluid - matrix and matrix - matrix correlations , but the non - equilibrium density factor identified above seems impossible to generate from the theory for the fully annealed system .
thus , as far as the development of theoretical approaches is concerned , and this statement is probably not restricted to the mct framework , the dynamics of a qa system should definitely be considered from the start as different from an infinite mass ( for newtonian systems ) or a zero bare diffusivity ( for brownian dynamics ) limit of the dynamics of a fully annealed mixture .
this is actually not so unexpected since , already in the statics , similar difficulties were met in early attempts to derive the equations valid for the qa mixture starting from those describing fully equilibrated systems ( see the discussion of refs .
@xcite in ref .
@xcite ) .
thus , the present theory is not a special case of earlier mode - coupling studies of the dynamics of particles moving in a glassy matrix @xcite . there ,
the mct for binary mixtures was used in a regime where one component , made of big particles , was completely glassy , i.e. , both the collective and tagged particle correlators did not relax to zero , while the other component , made of small spheres , was not necessarily localized , i.e. , the tagged particle correlators could relax to zero .
in fact , both theories are complementary . in the early approach
, the glassy matrix has to be the product of the self - consistent mode - coupling dynamics , but the model incorporates thermal fluctuations of the solid and aspects of its response to the presence of the fluid , while in the present approach , one has much more freedom to choose the structure of the confining medium , including realistic models of porous solids @xcite , but the disordered matrix is rigorously inert .
there are anyway qualitative analogies between both models .
for instance , there is a clear link between the fact that the collective motion of the small particles in a binary mixture necessarily becomes non ergodic at the same point as that of the big particles and the unavoidable existence of static blocked correlations for a fluid in a random matrix .
we now move to the quantitative predictions of the theory , which require numerical solutions of the mct equations . in this section
, we report dynamical phase diagrams , obtained by mapping , in the parameter space of a given model , the domain where @xmath80 , corresponding to the ergodic fluid phase , and the one where @xmath81 , corresponding to the non - ergodic ideal glassy state .
the interface between the two domains forms the ideal liquid - glass transition manifold ; in the present work , it will always be a line , since only systems with a two - dimensional parameter space are considered .
details of the dynamical changes when crossing this line are discussed in the next section . as already mentioned in sec .
ii , a significant difference between the mct equations for qa systems and those for bulk glassformers is the presence of a linear term in the memory kernel .
it results that the discontinuous or type b transition scenario known from the bulk , where the infinite time limit of @xmath82 jumps discontinuously from zero to a finite value when going from the liquid to the glass , is not the only possibility anymore .
continuous or type a transition scenarios , where @xmath74 grows continuously from zero when entering the glassy phase , indeed become possible .
this is best understood by reference to the so - called @xmath83 ( @xmath84 ) schematic models @xcite , in which the time evolution of a single correlation function @xmath85 is ruled by a two parameter memory kernel of the form @xmath86 it can be readily shown that , when the second term dominates ( @xmath87 small ) , these models have a line of type b transitions starting at @xmath88 , and , when the first term dominates ( @xmath89 small ) , they have a line of type a transitions starting at @xmath90 .
the study of the @xmath83 models also gives us information on the possible topologies of the dynamical phase diagrams , which depend on the way the transition lines meet .
two simple cases are found @xcite . if @xmath91 , the two lines join smoothly at a common endpoint , where a topologically stable degenerate @xmath92 singularity is located .
if @xmath93 , the two lines intersect and , in the glassy domain , only the extension of the type b transition line subsists beyond the intersection , forming a glass - glass transition line terminated by an ordinary @xmath92 singularity .
it turns out that these two prototypical shapes of phase diagrams are obtained with two very simple , closely related qa mixture models , to which the present study is restricted . in both ,
the fluid - fluid and fluid - matrix interactions are pure hard core repulsions of the same diameter @xmath94 .
the only difference lies in the matrix correlations . in model
i , the matrix configurations are assumed to be quenched from an equilibrium fluid of hard spheres of diameter @xmath94 , so that the matrix particles do not overlap , while in model ii the matrix particles are completely uncorrelated and overlap freely . in the following
, both systems will be parametrized by the two dimensionless densities @xmath95 and @xmath96 , and the percus - yevick ( py ) approximation @xcite will be used to compute the required structural quantities .
note that , in this approximation , @xmath97 , so that the difficulties mentioned at the end of sec .
ii are irrelevant .
the non - ergodicity parameter @xmath74 is a solution of the nonlinear set of equations @xmath98,\ ] ] which has to be solved numerically in order to locate the liquid and glassy phases when @xmath99 and @xmath100 are varied .
all computations in the present work have been achieved using the method of ref .
@xcite , to which the interested reader is referred for technical details , and we only provide the quantitative information needed to reproduce the present results , i.e. , that the wavevector integrals have been discretized to points on a grid of @xmath101 equally spaced values with step size @xmath102 , starting at @xmath103 @xcite .
we have checked by two means that this discretization is not too coarse , in particular with respect to the small @xmath7 divergence of the memory kernel which is cut off .
first , test calculations at various fluid and matrix densities have been performed on a finer @xmath7 grid , hence with a smaller @xmath104 .
second , in the @xmath61 limit , we have compared our prediction for the diffusion - localization point of model i with leutheusser s analytic result obtained within an additional hydrodynamic approximation which allows to integrate exactly over the full @xmath7 range @xcite .
in all cases , only modest quantitative differences of a few percents on the location of the transition points were found . when dealing with complex transition scenarios , including higher - order singularities and glass - glass transition lines , a useful quantity to consider is the largest eigenvalue @xmath105 of the stability matrix of the set of equations .
indeed , there holds @xmath106 , and @xmath105 goes to @xmath107 when a transition is approached from the strong coupling side @xcite .
thus , @xmath108 can be used as a convergence criterion for the determination of the transition points @xcite . in this work , @xmath109 was assured for ordinary transition points and @xmath110 was required in the regions where the transition lines of types a and b meet . the dynamical phase diagrams of models i and ii are reported in fig . [ figtrans ] .
in the left panel , they are plotted in the @xmath111 plane .
this is the obvious parameter space of the problem , but , with this choice of variables , no account is given of the differences in structure between the two matrix models .
this clearly limits the possibilities of a meaningful comparison between the two systems .
thus , in the right panel , both qa mixtures have been tentatively parametrized by the same physical constant , their henry constant @xmath112 . for systems with hard core interactions ,
@xmath112 is equal to the fraction of the total volume accessible to the center of an adsorbate particle in a matrix of density @xmath100 and is probably the most simple and generic scalar quantity characterizing the confining effect of a solid matrix on an adsorbed fluid .
it is given by @xmath113 for model ii @xcite , and by @xmath114 $ ] for model i , where @xmath115 is the excess chemical potential of the equilibrium hard sphere fluid with volume fraction @xmath100 @xcite .
the latter has been estimated following the compressibility route within the py approximation @xcite . as it can be readily seen in fig .
[ figtrans ] , both phase diagrams essentially have the same overall shape , especially when they are plotted as functions of @xmath112 . their main qualitative difference at this global scale , which is the difference in concavity of their upper parts visible in fig .
[ figtrans](a ) , turns out to be rather insignificant , as it is the simple and direct consequence of the fact that the strength of confinement increases more slowly with @xmath100 in model ii than in model i , because of the overlapping matrix particles .
it completely disappears in fig .
[ figtrans](b ) .
the nature of the ideal glass transitions met in the phase diagrams is just as expected from the study of schematic models .
starting from the type b liquid - glass transition point for the bulk ( at @xmath116 ) , where the memory kernel is purely quadratic , a line of type b transitions develops when increasing @xmath100 , and , starting from the type a diffusion - localization point ( at @xmath117 ) , where the memory kernel is purely linear , a line of type a transitions emerges when increasing @xmath99 . the same is true of the way the two lines meet in the phase diagrams . on the one hand , for model i , the type a and b lines have a common endpoint , denoted by e , where they smoothly join and form a degenerate @xmath92 singularity , like in the @xmath118 model . on the other hand , for model ii ,
the two lines intersect at a crossing point c and the extension of the type b transition line in the glassy domain becomes a glass - glass transition line ending with an ordinary @xmath92 singularity , denoted by e as well .
this is the scenario obtained with the @xmath83 models , for @xmath93 .
note that , to our knowledge , this is the first time that these widely studied one equation toy models find physical realizations as fluid systems .
the glass - glass transition line of model ii , located between points c and e , is barely visible at the scale of fig .
[ figtrans ] .
it is short and not well separated from the type a liquid - glass transition line .
taking into account the idealized nature of the predictions of the mct , it would probably be impossible to detect it unambiguously in computer simulations of this system . only the very specific features of the dynamics in the vicinity of a higher - order singularity , be it degenerate or not , should be visible , like logarithmic decay laws and subdiffusive behaviors @xcite .
but , at least , the present calculation shows that this scenario can actually be realized and that this does not require any exotic physical ingredient .
now that a suitable starting point is available , by playing with the parameters of the model , one can try and obtain a system for which this glass - glass transition line would be extended enough for its signatures to be observable in simulations .
a final requirement for a complete characterization of the transitions studied in this work is the knowledge of the so - called exponent parameter @xmath119 @xcite .
it determines many aspects of the dynamics near a transition ( see the next section ) and , for this reason , plays a crucial role in the theory .
one has @xmath120 and @xmath121 for type a and b transitions , respectively .
also , @xmath119 reaches @xmath107 , its maximum value , at endpoint singularities and jumps discontinuously at crossing points .
this is thus a useful parameter to follow during the computation of the phase diagrams .
it is plotted in fig .
[ figlambda ] as a function of @xmath99 at the transition . for model
ii , points c@xmath122 and c@xmath123 mark the discontinuity associated with the crossing point c in fig .
[ figtrans ] and the existence of glass - glass transitions is clearly visible , with values of @xmath119 given by the line between points c@xmath122 and e. for the same model , a nonmonotonic variation of @xmath119 along the type b line can be noted as well .
exponent parameter @xmath119 along the transition lines of models i and ii .
the lowest part , where @xmath119 goes to zero with @xmath99 , has been omitted for readability .
for both models , e denotes the @xmath92 singularity , with @xmath124 .
c@xmath122 and c@xmath123 delimit the discontinuity associated with the crossing point between the type a and b transition lines in model ii . ] in addition to these transition scenarios , and formally not related to them , another remarkable prediction of the present theory is a reentry phenomenon for matrix densities higher than the localization threshold ( obtained for @xmath117 ) . indeed , as shown in fig .
[ figtrans](a ) , for a given , not too high @xmath100 in this domain , ergodicity can be broken either by an increase or a decrease of the fluid density . in hard core systems , freezing by an increase of the fluid density can be qualitatively understood from simple free volume arguments , which provide a direct explanation for the decrease of @xmath99 at the transition as a function of @xmath100 in the upper part of the phase diagrams .
because of the volume excluded by the matrix particles , the larger the matrix density is , the smaller the fluid density has to be for structural arrest to occur . from the results for model
i , we might further note that this decrease is such that the total compacity @xmath125 at the transition is a decreasing function of @xmath100 as well , reflecting the fact that the inclusion of immobile matrix particles in the system slows down the dynamics more efficiently than the inclusion of the same amount of mobile fluid particles .
such a behavior , which is hardly surprising , has already been observed in molecular dynamics simulations @xcite .
the possibility of an ergodicity breaking transition by a decrease of the fluid density , which is reflected in the bottom part of the phase diagrams by the increase with @xmath100 of the transition @xmath99 , is more unexpected .
we interpret this prediction as the signature of a delocalization phenomenon induced by fluid - fluid interactions .
more precisely , we propose that the occasional collisions between the fluid particles at low @xmath99 can destroy the dynamical correlations responsible for the localization of individual particles in dense enough matrices . for this process to provide an efficient ergodicity restoring relaxation channel
, a reasonable criterion is that the localization domains should overlap to allow the fluid particles to interact .
thus , the localization length computed at @xmath117 , which decreases when @xmath100 increases , should be comparable to the average distance between two fluid particles , which decreases when @xmath99 increases . it is then immediate that , starting in the localized state , the larger @xmath100 is , the higher @xmath99 has to be in order to restore ergodicity .
the physical implications of this result will be discussed in more details in the last section .
dynamical scenarios involving reentrant glass transition lines , higher - order singularities , and glass - glass transition lines , have already been found for colloidal suspensions with short - ranged attractions @xcite . by analyzing various contributions to the memory kernel
, these features were shown to result from the interplay of two well defined phenomena , cage effect and bond formation , driven by the hard core and attractive parts of the interaction , respectively .
it seems thus interesting to attempt such an analysis for the present problem . guided by the results of sec .
iii , we propose that a contribution represented by a memory kernel @xmath126 derived from @xmath127 by replacing @xmath79 in eq . with @xmath128 should be isolated .
indeed , the resulting expression then coincides with the one describing the residual dynamics of a bulk ideal glass .
therefore , one can reasonably expect that , for a qa system , @xmath126 will provide a fair representation of the mechanism of caging by fluid particles in the presence of permanent density fluctuations , which is precisely the one at work in bulk glassy systems .
the remaining linear kernel @xmath129 can then be attributed to the confinement - specific phenomena , i.e. , the localization effect of the matrix combined with the unusual decorrelation mechanism due to fluid - fluid collisions discussed above . in principle , one could try to separate these two processes , using the fact that , for an ideal gas , only the localization effect is present and leads to a matrix density at the transition which is independent of the fluid density .
however , already with the present limited separation , the somewhat artificial character of the procedure shows up in the form of negative values of @xmath130 , which restrict the density domain where stable solutions of the mct equations can be found , so no further decomposition of @xmath130 was attempted .
the hypothetical phase diagrams computed with the partial kernels @xmath126 and @xmath130 for model i are reported in fig .
[ figfictitious ] , where they are compared to the one obtained with the full @xmath127 . clearly
, the reentry phenomenon observed in the complete phase diagram can be explained by the interplay of the two contributions discussed above .
but , at variance with the colloidal systems , the higher - order singularity is not located in the domain where they cross over .
the interpretation of this finding is ambiguous . on the one hand , since a higher - order singularity is found on the transition line computed with @xmath126 as well , one might argue that the singularity should be considered as an integral part of the scenario of caging by fluid particles .
then , in the relevant domain , confinement would simply appear as a modifier of the cage structure , progressively changing the nature of the ideal jamming transition from discontinuous to continuous . on the other hand
, the bifurcation analysis of the mct scenario shows that a higher - order singularity is necessarily formed when two lines of type a and b transitions meet .
then , since the type a and b lines arise from points representative of systems ruled by confinement and bulk caging , respectively , one might consider that the singularity is the product of the interplay of these two phenomena .
in favor of the latter interpretation , it has been recently suggested that higher - order singularities generically result from a competition between different arrest mechanisms @xcite .
note however that the conclusions of ref .
@xcite are drawn from the consideration of bulk systems only , which necessarily enter into the glassy state through type b transitions and for which there is no mathematical constraint in the theory imposing a priori the existence of a singularity .
it is not immediate that the proposed statement is valid or even needed for systems where type a and b liquid - glass transitions coexist .
real and hypothetical dynamical phase diagrams of model i , computed with the total and partial memory kernels @xmath127 , @xmath126 , and @xmath130 . the curves are labeled accordingly and the degenerate @xmath92 singularities are denoted by e. ]
it results from the previous section that , if one stays away from the higher - order singularities , the crossing points , and the glass - glass transitions , which require fine tuning of the parameters of the models , the present mct for qa mixtures predicts two generic liquid - glass transition scenarios .
one is discontinuous or type b , the other is continuous or type a. in this section , we discuss the different features of these transitions which are relevant for comparisons of experimental or simulation data with the predictions of the theory .
the analytic results will be quoted without their proofs , which can be found in ref .
they will be illustrated with detailed computations for model i at two matrix densities , @xmath131 and @xmath132 .
for the former value , a type b transition occurs at @xmath133 , with @xmath134 . for the latter , a type a transition is found at @xmath135 , with @xmath136 . for this matrix density , we consider the type a transition on the upper branch of the phase diagram , as it is of greater relevance for the problem of the glass transition in confinement . in both cases
, keeping @xmath137 allows one to consider that the transitions are far enough from the higher - order singularity . here
, a comment on the method of solution of the mode - coupling equations is in order . indeed , as mentioned in the previous section , it involves a cutoff of the low @xmath7 divergence of the memory kernel . since it has been demonstrated that this divergence can change the qualitative properties of the solutions in the asymptotic regime near the transition @xcite , this approximation and the use of the results of ref .
@xcite , which have been obtained under the assumption of nonsingular vertices , might look problematic .
in fact , this is not the case for physical reasons .
indeed , in the lorentz gas limit , the low @xmath7 singularity of the memory function has been shown to be an ill feature of the mode - coupling approximation @xcite .
thus , from a physical point of view , the discretized equations considered in the present calculation are actually more satisfactory than the continuous ones and the results reviewed in ref .
@xcite can legitimately be applied .
we first consider the density dependence of the non - ergodicity parameter @xmath74 , shown in fig .
[ fignonergvsphif ] . as it should , @xmath74 takes a finite value at the ideal glass transition point of type b , while it grows continuously from zero at the type a transition . in the glassy state , the bifurcation analysis of eq . to leading order yields two universal power law behaviors , @xmath138 for a type b scenario , and @xmath139 near a type a transition
accordingly , in fig . [ fignonergvsphif ] , the curves corresponding to the type b and a transitions start with infinite and finite slopes , respectively .
the wavevector dependence of @xmath74 , though not universal , is an important prediction of the mct as well .
it is reported in fig .
[ fignonergvsq ] . in fig .
[ fignonergvsq](a ) , @xmath74 is shown at the type b transition , while in fig .
[ fignonergvsq](b ) , since @xmath140 at a type a transition , the results at @xmath141 are plotted . for reference ,
the two relevant structure factors @xmath77 and @xmath78 at @xmath142 are also given .
note that in both cases , @xmath78 shows maxima both where @xmath77 has maxima or minima . at the type b transition , except for the peak at @xmath143 where @xmath74 reaches 1 as a consequence of the diverging kernel , the non - ergodicity parameter is very similar to the one found for a bulk hard sphere fluid and oscillates with @xmath77 , which precisely represents bulk - like correlations .
the overall amplitude is smaller than in the bulk , reflecting the fact that , when @xmath100 increases , the system evolves towards a continuous transition scenario . in comparison , the non - ergodicity parameter near the type a transition appears rather featureless .
@xmath74 simply decreases from @xmath107 at @xmath143 , with , as @xmath99 is increased , a small shoulder developing in the wavevector regime where @xmath77 has its main peak and @xmath78 its second peak . with the present models where the fluid and matrix particles have the same size , it is not clear which changes in the static correlations are actually responsible for the growth of this contribution . beside these results for the infinite time limit of the density correlation functions ,
the full dynamics is of great interest as well . for this computation , which uses the algorithm described in ref .
@xcite , we follow ref . @xcite and reduce the generalized langevin equation to its form valid for brownian dynamics , @xmath144 with @xmath145 and the initial condition @xmath32 . this simplification affects the short time transient part of the dynamics , but not its long time properties . in the following
, the unit of time shall be chosen such that @xmath146 .
the time evolution of the density correlation function @xmath147 at @xmath148 , corresponding to the main peak of @xmath77 , is reported in fig .
[ figcorrel ] for state points in the vicinity of the two transitions discussed above .
the curves for other values of @xmath7 are qualitatively similar . in fig .
[ figcorrel](a ) , the two step dynamics typical of the discontinuous ideal glass transition scenario and well known from the study of bulk systems is easily recognized in the curves corresponding to the liquid state .
the second step , associated to the decay from the plateau where @xmath149 , obeys the so - called superposition principle , which states that , in this regime and for a given @xmath7 , the shape of the dynamics is independent of the state point .
the relaxation functions only differ through the characteristic time scale @xmath150 , which displays a power law divergence , @xmath151 when the transition is approached .
one shows that @xmath152 where the exponents @xmath153 and @xmath154 ( @xmath155 , @xmath156 ) are related to @xmath119 through @xmath157 @xmath158 denoting euler s gamma function . in the glassy state , only the first relaxation step remains and , when @xmath17 goes to infinity , @xmath147 reaches @xmath74 , which increases with @xmath99 as shown in fig .
[ fignonergvsphif](a ) .
the dynamics near the type a transition visible in fig .
[ figcorrel](b ) looks significantly different , with a single step relaxation scenario , both in the liquid and glassy phases .
the slowing - down of the dynamics manifests itself through a weak long time tail which extends to longer times when @xmath99 is increased and turns above @xmath142 into a finite asymptote which grows as shown in fig .
[ fignonergvsphif](b ) .
there are nevertheless strong similarities between the two dynamics , provided one concentrates on the time domain where @xmath159 ( i.e. , where @xmath147 is small for a type a transition ) . for type
b dynamics , this corresponds to the so - called fast @xmath160 relaxation regime and we first specialize the discussion to this case . then , close enough to the transition , a reduction theorem holds , according to which the wavevector and time dependencies of @xmath147 factorize , yielding @xmath161 at the critical point , the critical decay law @xmath162 is obtained , where @xmath153 is given by eq . and
@xmath163 is a time scale obtained by matching the short and long time dynamics . for finite values of @xmath164 , one finds the scaling laws @xmath165 where the master functions @xmath166 are the solutions of @xmath167 and the scaling variables obey @xmath168 for large @xmath169 , in the non - ergodic phase , @xmath170 goes to a constant and eq .
results from the expression of @xmath171 . in the same time regime , in the ergodic phase , another power law behavior sets in for @xmath172 , yielding the so - called von schweidler decay law , @xmath173 where @xmath154 is given by eq . and @xmath174 is a positive constant which can be determined by matching eq . with eq . for @xmath175 .
by combining the power law behaviors of @xmath171 and @xmath176 in the resulting expression of @xmath177 , one recovers the divergence of @xmath150 , eq . .
moving now to the type a transitions , one finds that , both in the ergodic and non - ergodic states , @xmath147 essentially behaves near zero as it does when it approaches @xmath178 in the glassy phase in a type b scenario . in particular , the critical decay law remains valid at the transition . for finite values of @xmath164 and independently of its sign , eq .
has to be modified in order to properly define @xmath177 ( see ref .
@xcite for details ) , then one finds that a scaling law holds , of the form @xmath179 where @xmath170 is the same function as above .
however , as in eq .
, the exponents characterizing the scaling variables @xmath180 and @xmath181 are twice those for a type b transition , i.e. , @xmath182 , where the time evolution of @xmath183 ( simply @xmath82 for the type a transition ) at @xmath148 is plotted in a log - log scale in order to evidence the power laws and . in this graphs , an interesting consequence of the scaling laws for small
@xmath169 is clearly visible , which is the symmetric departure from the critical decay law at long times for points located in the liquid and glassy phases at the same distance from the transition .
this is a particularly important feature of the dynamics in the type a scenario .
the above results do not apply in the vicinity of higher - order singularities or crossing points , where a refined mathematical analysis is required @xcite .
a detailed discussion is beyond the scope of the present overview of the theory and we shall only mention that such liquid - glass transition points signal themselves in the dynamics through the appearance of logarithmic decay laws @xcite . finally , already for bulk systems , it is often quite difficult to unambiguously demonstrate that the above features are actually present in some experimental or simulation data , since , as mentioned in the introduction , the mct offers an idealized picture of the glass transition phenomenon and there are always significant alterations to the theoretical scenario .
an additional difficulty can be anticipated in the case of qa systems .
indeed , one usually has access to the total density correlation function @xmath184 and not to the connected function @xmath82 .
both are related through @xmath185 thus , the glassy dynamics is modulated by static factors and in particular develops itself on top of a state dependent static background .
the separation in the long time behavior of @xmath184 of the evolutions which are of purely static origin from those which characterize the glassy dynamics might then be delicate
. this could be especially critical for type a transitions , where the signatures of the continuous transition have to be identified just on top of the slowly drifting static contribution .
figures demonstrating the problem can be found in ref .
in this paper , a mode - coupling theory for the slow dynamics of fluids adsorbed in disordered porous solids made of spherical particles frozen in random positions has been developed .
derived by properly taking into account the peculiar structure of the correlations in these systems and by including a contribution which had been forgotten in a previous work @xcite , its equations are found to display many appealing features .
for instance , they show universality , in the sense that they do not contain any explicit reference to the precise nature of the random environment in which the fluid evolves . also , they compare favorably with previous mode - coupling equations derived in other contexts , for the residual dynamics in the glassy phase of a bulk fluid ( sec .
iii ) or for the equilibrium dynamics of a mean - field spin - glass in a random magnetic field ( appendix [ app.pspin ] ) . thus , from a formal point of view
, the theory appears rather satisfactory .
nevertheless , a few difficulties remain . first , there is the fact that , in the limit of vanishing fluid - fluid interactions , the present theory does not coincide with the mct which can be derived by assuming from the start that there are no such interactions .
second , there is the divergence of the memory kernel for small wavevectors and the resulting spurious long time anomalies @xcite .
we do not believe that these issues are really harmful , even if their handling requires ad hoc approximations , but they are worth stressing , since their solutions would probably teach us something on the nature of the approximations underlying the mct scheme and on possible extensions of the theory .
for instance , it has been argued by leutheusser that the inclusion of vertex corrections within a kinetic theory approach would solve the second problem , but no operational scheme was proposed for this calculation @xcite .
the numerical solution of the mct equations for two simple fluid - matrix models leads to a variety of transition scenarios , which are either discontinuous for dilute matrices or continuous for dense matrices .
depending on the model , in the intermediate region where the nature of the transition changes , degenerate or genuine higher - order singularities and glass - glass transition lines are found .
another remarkable prediction of the theory is the possibility of a reentry phenomenon for high matrix densities above the localization threshold , which has been interpreted as the signature of a decorrelation process induced by fluid - fluid collisions . before going further
, one should note that , strictly speaking , this prediction of the theory can not be correct in the case of hard core fluid - matrix interactions .
indeed , as recently confirmed by extensive computer simulations of the lorentz gas @xcite , the localization transition is driven by the percolation transition of the matrix void space , i.e. , localization occurs because , above a certain critical matrix density @xmath186 , the void space only consists of finite disconnected domains .
in such a scenario , it is obvious that , whatever the fluid density , it is impossible to have an ergodic system above @xmath186 , since fluid - fluid interactions will never change the geometry of the matrix .
this contradiction between the percolation theory and the mct clearly raises the issue of the relation between the two approaches , for which we propose the following simple argument .
the mct applied to the problem of the diffusion - localization transition attempts to capture the onset of percolation in an indirect way , by giving an account of the increasingly correlated nature of the fluid - matrix collision events as the threshold is approached . in this respect , one should note that none of the static structure functions on which it is based does show a sensitivity to the phenomenon of percolation .
following leutheusser , the theory works at the level of a self - consistent treatment of ring collision processes @xcite .
this turns out to be enough to predict a diffusion - localization transition , but , clearly , the infinite sequences of correlated collisions which would really reflect the permanent trapping of the fluid particle in a finite domain above the percolation threshold are missing . from this incomplete characterization of the dynamical processes escorting the percolation phenomenon
, it results that the mct diffusion - localization transition is actually an ideal version of the true percolation transition , in the usual sense that the mct predicts ideal glass transitions , and that the theory is not able to detect that , in an exact treatement , the percolation threshold fixes an absolute limit to diffusive behavior .
when fluid - fluid collisions come into play at finite fluid densities , this leaves room for the prediction of a reentry phenomenon in contradiction with percolation theory .
we believe that it is for the same reason that the mct also misses the fact that , at any matrix density , there is always a non - vanishing probability that particles will be trapped in a finite domain disconnected from the rest of the void space , so that the exploration of the available void space is never completely ergodic @xcite .
it is thus clear that the prediction of a reentrant behavior of the ergodicity breaking transition line in the low fluid density domain should not be taken too literally .
in fact , a reasonable expectation based on this finding is that , below the localization threshold , but in the regime where transient trapping effects are important , there might be an acceleration of the dynamics due to fluid - fluid collisions .
interestingly , such a behavior has already been observed in a computer simulation study of a two dimensional lattice gas model with fixed randomly placed scatterers @xcite .
indeed , it was found that , starting from the zero fluid density limit , the diffusion coefficient of a tagged particle first increases with the fluid density .
an interpretation in terms of a decorrelation process similar to the one discussed in sec .
iv was then proposed and validated by varying the parameters of the model . as a possible source of the difficulties of the theory
, one might blame the fact that it works at the level of disorder averaged quantities .
indeed , the procedure of averaging over disorder is equivalent to an averaging over the volume of a macroscopic system , an operation in which many microscopic details of the statics and dynamics become blurred .
this might confer a mean - field character to the theory , where the contribution of the matrix would actually be taken into account at the level of a diffuse effective localizing potential , with a possible loss of important local constraints .
unfortunately , this is a necessary step in order to develop a theory which is comparable in complexity with the one for bulk systems , since it allows one to consider the system as homogeneous .
some progress has recently been made on the application of the mct to inhomogeneous situations @xcite , so it should be possible to relax the condition of homogeneity in order to study the dynamics of fluids confined in pores of simple geometry which are often preferred in simulation works , but there is no doubt that the more complex wavevector dependence of the resulting theory will make it harder to obtain numerical solutions of the equations .
moreover , beyond this purely technical aspect , the present formulation in terms of disorder averaged quantities has a practical interest as well .
indeed , many real porous media are disordered and most experimental techniques measure quantities which are averaged over the volume of a macroscopic sample and thus equivalent to the disorder averages considered by the mct .
so , the current theoretical setup seems well suited for direct comparisons with experimental results . for molecular dynamics simulations , however ,
since rather small systems are usually considered , it might be necessary to explicitly perform the disorder average over a representative sample of matrix configurations before a comparison with the theory can be done .
altogether , in spite of the above merely technical issues , we believe that the present mode - coupling theory represents a valuable step towards a better understanding of the slow dynamics of confined glassforming liquids .
indeed , it remains rather simple and , since it is a microscopic approach , it allows one to study in detail the effect on the dynamics of changes in the different ingredients of a fluid - matrix model ( fluid - fluid and fluid - matrix interactions , structure of the matrix ) .
thus , this provides us with a tool to efficiently and thoroughly explore the phenomenology of dynamics in confinement .
this is clearly illustrated by our findings for two very simple systems with pure hard core interactions , which already display new and nontrivial glass transition scenarios . then remains the question of the validation of the theoretical predictions . because the model of the qa mixture is quite simple and the theory makes detailed predictions
, molecular dynamics studies should be able to give clear - cut answers .
the presently available results look rather encouraging , but more simulation work is definitely needed .
it is a pleasure to thank g. tarjus , w. gtze , and w. kob , for useful comments , and f. hfling for a valuable discussion and the communication of unpublished results .
for reference , we quote the replica oz equations relating the various pair correlation functions in qa binary mixtures @xcite .
they read , in fourier space , @xmath187 with @xmath188 and @xmath189 .
as usual , @xmath190 and @xmath180 denote total and direct correlation functions , respectively .
@xmath191 denotes the fourier transform of @xmath192 and the superscripts have the same meaning as for the structure factors ( see sec .
ii ) . using the relations
@xmath193 ( remember that @xmath194 , hence @xmath195 ) the oz equations can be formally solved for the structure factors , leading to @xmath196 \frac{1}{(1 - n_f \hat{c}^{c}_q)^2}.\end{aligned}\ ] ]
in this appendix , the derivation of the mode - coupling part of the memory kernel is outlined .
this calculation is a direct extension of the one for bulk systems which is described in its most minute details in ref .
@xcite . the memory function in eq .
is defined as @xmath197 where @xmath198 i \mathcal{q}_1 \mathcal{l } g^f_\mathbf{q}$ ] is the projected random force obtained from the longitudinal momentum density fluctuation @xmath199 @xcite .
@xmath200 is the liouville operator of the system and @xmath201 is the complementary operator of the projector @xmath202 which projects any dynamical variable onto the subspace spanned by @xmath203 and @xmath204 .
the calculation of the mode - coupling part of the kernel amounts to replacing @xmath205 in eq . by its projection @xmath206 onto the subspace spanned by @xmath207 , @xmath208 , and @xmath209 , where we introduce the projection operator @xmath210 such that @xmath211 in this expression , we have anticipated that , within the mode - coupling approximation , @xmath210 is diagonal in @xmath212 and the subspaces spanned by the @xmath174s and @xmath213s are orthogonal .
the prime in the first sum indicates that , to avoid double - counting , the wavevectors are assumed to be ordered somehow and the sum is restricted to @xmath214 .
@xmath215 and @xmath216 are normalization matrices insuring that @xmath217 and @xmath218 . in the course of this calculation , four - point density correlation functions are generated .
thus , in order to eventually obtain closed dynamical equations , a factorization approximation is needed to express them as products of two - point density correlation functions .
we follow the usual mode - coupling prescription and find that , within this approximation , the only non - vanishing four - point functions are given by @xmath219 specializing to @xmath220 and computing @xmath221 and @xmath222 , it results that @xmath223^{-1 } , \\ % h^{(11)}_{\mathbf{q},\mathbf{k } } = & \left [ n_f n_m s^{c}_k \right]^{-1 } \frac{s^{b}_{|\mathbf{q - k}|}}{s^{mm}_{|\mathbf{q - k}| } s^{b}_{|\mathbf{q - k}| } - ( s^{fm}_{|\mathbf{q - k}|})^2 } , \\ % h^{(22)}_{\mathbf{q},\mathbf{k } } = & \left [ n_f^2 s^{c}_k \right]^{-1 } \frac{s^{mm}_{|\mathbf{q - k}|}}{s^{mm}_{|\mathbf{q - k}| } s^{b}_{|\mathbf{q - k}| } - ( s^{fm}_{|\mathbf{q - k}|})^2 } , \\ % h^{(12)}_{\mathbf{q},\mathbf{k } } = & - \left [ n_f \sqrt{n_f n_m } s^{c}_k \right]^{-1 } \frac{s^{fm}_{|\mathbf{q - k}|}}{s^{mm}_{|\mathbf{q - k}| } s^{b}_{|\mathbf{q - k}| } - ( s^{fm}_{|\mathbf{q - k}|})^2}.\end{aligned}\ ] ] it remains to express @xmath224 and @xmath225 .
first , using yvon s theorem , one finds @xmath226,\\ % \overline{\langle i\mathcal{l}g^f_\mathbf{q } c^{(1)}_{\mathbf{-q ,- k}}\rangle } & = i q
k_b t \sqrt{n_f n_m } \frac{\mathbf{q}\cdot \mathbf{k}}{q^2 } s^{fm}_{|\mathbf{q - k}|},\\ % \overline{\langle i\mathcal{l}g^f_\mathbf{q } c^{(2)}_{\mathbf{-q ,- k}}\rangle } & = i q k_b t n_f \frac{\mathbf{q}\cdot \mathbf{k}}{q^2 } s^{b}_{|\mathbf{q - k}|}.\end{aligned}\ ] ] then , with the help of the extension of the convolution approximation @xcite to qa systems , @xmath227 it comes @xmath228 the desired results are obtained by substracting the matching equations .
we might now complete the explicit calculation of @xmath206 .
indeed , if we define @xmath229 such that @xmath230 the above results immediately lead to @xmath231,\\ % w^{(1)}_\mathbf{q , k } & = 0 , \\ % w^{(2)}_\mathbf{q , k } & = \frac{i q k_b t}{n_f } \left [ \frac{\mathbf{q}\cdot \mathbf{k}}{q^2 } \frac{1}{s^{c}_k } - 1 \right].\end{aligned}\ ] ] note the vanishing of the fluid - matrix contribution . eventually , injecting the resulting expression of @xmath206 into @xmath232 one obtains eqs . and after a few elementary steps .
exactly the same calculation can be done with subsets of the above three quadratic variables .
one then easily demonstrates that , if only @xmath233 and @xmath234 are considered , the same equations are obtained , while , if one works with @xmath233 and @xmath235 , the equations of ref .
@xcite are reproduced , which reduce to those of refs .
@xcite in the zero fluid density limit .
a fruitful source of new theoretical developments on the physics of glassy systems has been the finding that the mct provides an exact description of the equilibrium dynamics of a certain class of mean - field spin - glass models with multispin interactions @xcite . in this appendix , based on ref .
@xcite ( the interested reader is referred to this paper for details ) , we show that the equations describing the dynamics of such a mean - field spin - glass model in a random magnetic field reproduce the structure of those of the mct for a fluid in a random environment .
thus , the correspondence between the two approaches still holds in the presence of an external source of disorder .
we consider the fully connected mean - field spherical spin - glass model with three spin interactions in a random magnetic field , whose hamiltonian is @xmath236 = - \sum_{1\le i < j <
k\le n}j^{(3)}_{ijk } s_i s_j s_k - \sum_{1\le i\le n } j^{(1)}_{i } s_i,\ ] ] where the spins @xmath237 are @xmath238 real variables subject to the constraint @xmath239 , and the random coupling constants @xmath240 and fields @xmath241 are uncorrelated zero mean gaussian variables with variances @xmath242 and @xmath243 , respectively . in zero field
( @xmath244 ) , this model is known to generate the quadratic memory kernel introduced in refs .
@xcite and @xcite as a one wavevector approximation to the bulk mct functional .
thus , it is especially suitable for our purpose , since it displays nonlinearities of the same degree as in the mct . in the presence of a random magnetic field ,
just like qa mixtures are characterized by static frozen density fluctuations , the present spin - glass model is characterized by static frozen local magnetizations @xmath245 which vanish when the average over disorder ( both on couplings and fields ) is performed .
accordingly , we might define @xmath246 and introduce three correlation functions , which are the analogues of @xmath48 , @xmath47 , and @xmath82 .
they are , respectively , the two overlap functions @xmath247 and @xmath248 which obey @xmath249 , and the normalized connected spin correlation function , @xmath250 the exact solution of the langevin dynamics of the model using standard methods @xcite provides us with an equation for the time evolution of @xmath85 which reads , for a high enough temperature @xmath13 , a direct comparison of these equations with those of sec .
ii immediately shows that both sets of equations exactly have the same formal structure . in particular , beyond the simple fact that the memory functions are similar polynomials of the dynamical correlation functions , the same parametrization in terms of connected and disconnected static correlations is found .
we conclude with two short remarks .
first , as for qa mixtures , we note that the external random field does not enter explicitly in the above equations ( no @xmath253 is present ) .
because the model is exactly soluble , it is easy to understand how this happens .
in fact , the details of the random field are only needed for the computation of the static correlations , through the equality @xmath254 once this calculation is done , they can be forgotten .
the same is probably true for fluids in random environments and this supports the assumption that the mct equations for qa systems actually have a wider domain of applicability , provided they are expressed in terms of fluid quantities only .
second , it can be shown that eq . also describes the residual relaxation of the present spin - glass model in zero field , when the system is equilibrated in its low temperature phase and the total spin correlation function has a nonvanishing infinite time limit @xmath7 @xcite .
then , @xmath255 and @xmath256 simply have to be replaced by @xmath7 and @xmath257 .
thus , in the present model , the difference discussed in sec .
iii between self - induced glassiness and the effect of a quenched random field does not seem to exist .
this could have been anticipated , since by construction , fully connected models are unable to capture phenomena which only show up in the wavevector dependence of the coupling constants in fluid systems .
l. fabbian , w. gtze , f. sciortino , p. tartaglia , and f. thiery , phys .
e * 59 * , r1347 ( 1999 ) ; * 60 * , 2430 ( 1999 ) ; j. bergenholtz and m. fuchs , * 59 * , 5706 ( 1999 ) ; k. dawson , g. foffi , m. fuchs , w. gtze , f. sciortino , m. sperl , p. tartaglia , t. voigtmann , and e. zaccarelli , phys .
e * 63 * , 011401 ( 2001 ) .
similar discrepancies appear generically when different routes are followed to derive the mct equations for the tagged particle dynamics in a qa system .
they will be discussed in more details , with their quantitative consequences , in a forthcoming paper ( v. krakoviack , in preparation ) .
the possible values of @xmath261 are limited by the method of solution of the oz / py equations on a real space grid of equally spaced points with a step size commensurate to the hard core diameter @xmath94 . for our calculation
, the present choice of @xmath261 offers the closest match with the value @xmath262 used in ref .
@xcite , where analytic expressions for the structure functions were readily available . | we derive a mode - coupling theory for the slow dynamics of fluids confined in disordered porous media represented by spherical particles randomly placed in space .
its equations display the usual nonlinear structure met in this theoretical framework , except for a linear contribution to the memory kernel which adds to the usual quadratic term .
the coupling coefficients involve structural quantities which are specific of fluids evolving in random environments and have expressions which are consistent with those found in related problems .
numerical solutions for two simple models with pure hard core interactions lead to the prediction of a variety of glass transition scenarios , which are either continuous or discontinuous and include the possibility of higher - order singularities and glass - glass transitions .
the main features of the dynamics in the two most generic cases are reviewed and illustrated with detailed computations .
moreover , a reentry phenomenon is predicted in the low fluid - high matrix density regime and is interpreted as the signature of a decorrelation mechanism by fluid - fluid collisions competing with the localization effect of the solid matrix . |
throughout history , the scientific community has had the task of studying the properties of different dynamical systems and their importance to the environment where they are developed .
for example , the fact that some dynamical systems are chaotic and have a critical dependence on initial conditions is known since late last century .
this important characteristic has been correlated with the coexistence of several possible final stable states for a given set of parameters @xcite . in dissipative systems , this property is called generalized multistability " so it can be distinguished from the ordinary coexistence of stationary solutions @xcite .
the term multistability was first used with respect to visual perception in 1971 @xcite ; nevertheless , the occurrence of multistability is very common in various fields of science , such as chemistry @xcite , optics @xcite , physics @xcite and biological systems @xcite .
nevertheless , the importance of generating multistable structures resides in the wide variety of applications that exist : synchronization , complex networks , communication , climate , etc . in this work
it is taken advantage of the properties of the hybrid dynamical systems @xcite , such as the unstable dissipative systems theory based on piecewise linear ( pwl ) systems whose solution presents multi - scroll attractors @xcite .
the method proposed here consists on designing systems taken from the jerky equation which present the coexistence of multistable state if the following conditions are met : i ) if the distance between the commutation law and the equilibrium point satisfies a specific ratio , and ii ) if the commutation law changes automatically regarding the values given to the nearest integer function or @xmath0 , dividing the space in equally distributed subdomains given along an axis depending on the initial state applied to the system .
although there have been several reports on how to generate multistability , this is the first in our knowledge in which the values of the eigenspectra presented in the systems are being affected directly by means of the @xmath0 function .
it will be shown that regarding these values , the system changes from multi - scroll attractors into a mono - stable state to the coexistence of several single - scroll attractors into a multistable state .
this paper is organized as follows : in section 2 , the unstable dissipative system theory is introduced to generate scrolls along the @xmath1 axis and the function for the commutation surface displacement . in section 3 ,
the nearest integer or @xmath0 function is analyzed as a basis for the commutation surface and equilibria displacement .
section 4 contains the results about the transition from multi - scroll attractors to multistability . in order to explain the behavior of the system , bifurcation analysis of the parameters involved
are studied along with the variation of the eigenspectra of the system .
the results are displayed numerical depicting the multistable structures obtained .
finally conclusions are drawn in section 5 .
consider the dynamical systems which are defined by a class of linear affine system given by : @xmath2 where @xmath3 is the state vector , @xmath4 stands for a real vector and @xmath5 denotes a linear operator that is not singular with entries @xmath6 , @xmath7 .
the equilibrium point of the system results in @xmath8 .
the class of linear affine systems considered here , are those that present oscillations around the equilibria due to the stable and unstable manifolds @xmath9 and @xmath10 , respectively .
these manifolds are defined in a way that @xmath11 is a set of column eigenvectors such that @xmath12 with @xmath13 ; @xmath14 and @xmath15 . in order to present such oscillations and following a similar mechanisms as in @xcite ,
two types of dissipative systems with unstable dynamics have been studied which will be called unstable dissipative systems ( uds ) , however only one type of both will be considered here .
this type is defined in the following way : [ def_usd3 ] a linear system @xmath16 , where @xmath17 is the state vector , @xmath5 is a linear operator and @xmath18 , are the eigenvalues of @xmath19 .
if @xmath20 , and one @xmath21 is real negative @xmath22 , and two @xmath23 are complex conjugated with positive real part @xmath24 then the linear system will be called a uds of the type i. if the linear affine system given by eq .
( [ ec_sistema ] ) satisfies the definition [ def_usd3 ] with @xmath25 then it is possible to generate an attractor @xmath26 by means of a pwl system under the following considerations for the vector @xmath27 : @xmath28 the affine vector @xmath27 must be a switching function that changes depending on which domain @xmath29 with @xmath30 the trajectory is located .
the equilibria of system are given by @xmath31 , with @xmath32 , and each vector @xmath33 of the system is considered in order to generate a multiscroll attractor of system .
the idea of the method lies on defining vectors @xmath33 in order to assure stability of a class of dynamical systems in @xmath34 with oscillations within the attractor @xmath26 . in a way that for any initial condition @xmath35 , where @xmath36 is the basin of attraction , the system given by eq .
induces in the phase space @xmath37 the flow @xmath38 .
thus , each initial condition @xmath39 generates a trajectory given by @xmath40 which is trapped in an attractor @xmath26 after defining at least two vectors @xmath41 and @xmath42 , as it is described in @xcite .
this class of systems can display various multi - scroll attractors as a result of a combination of several unstable `` one - spiral '' trajectories .
the number of scrolls depends on the number of vectors @xmath43 , @xmath32 introduced in the system , thus the equilibrium points are given by @xmath44 and which trajectories oscillate around them .
this is a consequence of an important feature of this kind of uds , where they can result in one scroll attractor for each equilibrium point appropriately added in the domains @xmath45 in which the system is divided @xcite .
a convenient approach to build the matrix @xmath19 and vector @xmath27 is based on the linear ordinary differential equation ( ode ) given by the jerky form : @xmath46 .
the location where the scrolls are positioned can be understood from the following example which implies the coefficient matrix @xmath19 from the jerky equation @xcite and the affine vector @xmath27 as follows : @xmath47 according with definition [ def_usd3 ] the following entries of the matrix @xmath19 will be considered : @xmath48 and for the affine vector @xmath27 , the value of @xmath49 will commute according to the value of @xmath1 as follows : @xmath50 here @xmath51 determine the values of the equilibrium points since @xmath52 , and @xmath53 stands for the location of the commutation surface given along the @xmath1 axis regarding the position that it takes with respect to the positive or negative axis and with @xmath54 . considering @xmath55 and @xmath56 the equilibria
are located at @xmath57 and @xmath58 displacing only along the positive @xmath1 axis . in order to consider an
equally distributed scrolling around the equilibria , the distance between the equilibrium points is calculated with the euclidean distance @xmath59 , resulting in @xmath60 . therefore , the commutation surface that generates two equally distributed scroll trajectories is given by @xmath61 resulting in a surface located at the @xmath62 plane .
the eigenvalues and corresponding eigenvectors result in : projections of the attractor given by eqs . with and onto the @xmath63 plane for a ) and
c ) , and the @xmath64 plane for b ) and d ) .
the commutation law given by eq . for a ) and b ) , and eq . for c ) and d ) . marked with red asterisks the equilibria of the system and with black dashed line the commutation surface .
the direction of the eigenvectors are marked with red and blue lines for the stable and unstable manifolds @xmath9 and @xmath65 , respectively.,title="fig:",scaledwidth=48.0% ] projections of the attractor given by eqs . with and onto the @xmath63 plane for a ) and
c ) , and the @xmath64 plane for b ) and d ) . the commutation law given by eq . for a ) and
b ) , and eq . for c ) and d ) . marked with red asterisks the equilibria of the system and with black dashed line the commutation surface .
the direction of the eigenvectors are marked with red and blue lines for the stable and unstable manifolds @xmath9 and @xmath65 , respectively.,title="fig:",scaledwidth=48.0% ] + projections of the attractor given by eqs . with and onto the @xmath63 plane for a ) and
c ) , and the @xmath64 plane for b ) and d ) . the commutation law given by eq . for a ) and
b ) , and eq . for c ) and d ) . marked with red asterisks the equilibria of the system and with black dashed line the commutation surface .
the direction of the eigenvectors are marked with red and blue lines for the stable and unstable manifolds @xmath9 and @xmath65 , respectively.,title="fig:",scaledwidth=48.0% ] projections of the attractor given by eqs . with and onto the @xmath63 plane for a ) and
c ) , and the @xmath64 plane for b ) and d ) . the commutation law given by eq . for a ) and
b ) , and eq . for c ) and d ) . marked with red asterisks the equilibria of the system and with black dashed line the commutation surface .
the direction of the eigenvectors are marked with red and blue lines for the stable and unstable manifolds @xmath9 and @xmath65 , respectively.,title="fig:",scaledwidth=48.0% ] + @xmath66 the commutation surface must be located taking into consideration the manifolds @xmath67 and between @xmath68 , otherwise the trajectory of the system can escape and the scrolls are not formed ( this is described in figure 2 and figure 3 of the references @xcite , respectively ) .
the trajectory of the system given by the initial condition @xmath69 and eqs . with and presents a double scroll attractor as it is depicted in figure [ fig : bi ] a ) and b ) .
marked with red asterisks the equilibria of the system , and with red and blue lines the eigenvectors corresponding to the stable @xmath9 and the unstable @xmath65 manifolds ( _ i.e. _ @xmath70 , @xmath71 ) , respectively .
it is important to mention that the eigendirection determined by the two complex conjugate eigenvalues has been represented only as the projection of a line , however this manifold corresponds to a plane , also these manifolds end at the commutation surface but they were projected to the other domains in order to clarify their directions with respect to the other manifolds .
function @xmath49 given by .
the distance between each step is given by @xmath72.,width=415 ] between both scrolls , the commutation surface at the plane @xmath62 is marked with a black dashed line , dividing the space in two domains @xmath73 .
notice two important facts about the system , first that the scrolls are increasing their size due to the unstable manifold , this can be better appreciated at the projection of the attractor onto the @xmath64 plane from figure [ fig : bi ] b ) .
second , that the trajectory of the system escapes from the domain @xmath74 with @xmath75 located in the left side of the commutation surface .
this occurs near the @xmath65 manifold where it crosses the surface in the lower part of the figure [ fig : bi ] b ) and it is attracted by @xmath9 towards the equilibrium point in the domain @xmath76 located at the right side of the commutation surface .
the process is repeated in the inverse way forming scrolls around each equilibrium point .
this property of the uds can be easily extended for the generation of any number of scrolls along any of the axes if the above considerations are made when designing the system and locating equilibria along the axes in the following way @xcite .
first , start considering @xmath55 ( in case that the first equilibrium point is located at the origin ) and the value of @xmath77 .
thus the commutation surface @xmath78 that can result in two symmetric equilibrium points @xmath79 is given as @xmath80 , resulting in the two symmetrical scrolls .
now in order to extend it along the @xmath1 axis , the distance @xmath81 between the equilibria must be considered between adjacent equilibrium points in the system .
this idea of introducing more equilibria to the system can be done by considering the values @xmath82 , resulting in the consecutive equilibria along the @xmath1 axis @xmath83 with @xmath84 ( @xmath85 ) .
the number of scrolls that are introduced in the 1d - grid along @xmath1 is @xmath86 .
the commutation surfaces also should be located according to the value of the distance , in this case they are at @xmath87 .
therefore the commutation law for a @xmath88 scroll attractor is given by projection of the trajectory of the system given by eqs . and onto the @xmath63 plane with @xmath89 and @xmath60 .
marked with red asterisk the equilibria of the system , and with gray line the commutation surfaces generated by the function .,width=415 ]
@xmath90 the resulting attractor can be appreciated in figures [ fig : bi ] c ) and d ) .
both figures present the commutation surfaces @xmath91 marked with black line , along with the equilibria @xmath92 .
each equilibrium point also depicts the @xmath9 and @xmath10 manifolds regarding their eigenvectors which are parallel among them similar as in the two scroll attractor .
adding more equilibria to the system can be easily implemented by using a step function instead of generating commutation surfaces manually , i.e. , using an automatically step generating function as it has been applied in @xcite . here , the @xmath0 function will be implemented to simplify and automate this process .
the function will be defined as follows : @xmath93 for example , in order to have similar commutation surfaces and locations of equilibrium points that the ones described in eq . , consider the commutation of the vector @xmath94 given by the following function : @xmath95 where @xmath96 corresponds to the amplitude of the function which is similar to the variable @xmath97 , and @xmath72 corresponds to the length of the step given by the round function centered in the origin , this is depicted in the graph of figure [ fig : round ] .
notice that the value of @xmath72 has the same representation as @xmath98 .
if the function is considered as the commutation law of , then the commutation surfaces are located at every change of the steps , _
i.e. _ , @xmath99 with @xmath100 and @xmath85 . the commutation surfaces and equilibrium points introduced by the function are located when the corresponding domain @xmath101 is being visited by the trajectory .
it is important to mention that if @xmath72 corresponds to the distance between two continuous equilibrium points given a value of @xmath102 , then , each equilibrium point is located exactly at the middle of two consecutive commutation surfaces . besides
, the equilibria of the system and the domains are located from @xmath103 due to the fact that the function given by is not bounded as the commutation law in or are .
this can be understand in the following way , if the system increases the size of its scroll , eventually it crosses to the next or previous @xmath101 changing the value of @xmath102 and increasing the number of scrolls in the system .
this number of scrolls continues increasing according with @xmath104 . in figure
[ fig : atractor_round_1 ] a projection of the trajectory of the system given by eqs . and into the plane @xmath63 is displayed , presenting each commutation surface and equilibrium point between @xmath105 generated from the signal considering @xmath89 , @xmath60 and @xmath106 .
notice that the oscillating behavior is similar as the one based on the commutation surfaces given by eq . , but @xmath107 scrolls are presented in this case for 50,000 iterations using the fourth order runge kutta method .
if the number of iterations are increased so will the number of scrolls .
the relationship between these two parameters satisfies @xmath108 which is equal to the entry @xmath109 of the matrix @xmath19 , resulting in equilibrium points equally located between the commutation surfaces .
now based on the previous method for generating multi - scroll attractors , the following system is considered : @xmath110 where @xmath111 is a constant parameter . with this vector @xmath112 the displacement of the equilibria is also along the @xmath1 axis , where @xmath49 commutes according to the @xmath113 function given by eq . .
a ) bifurcation diagram of the system given by eqs . with and for the value of @xmath114 for figure
b ) depicts the bifurcation for the range @xmath115 for @xmath116 iterations .
figures c ) and d ) show the number of domains @xmath101 visited by the trajectory of the systems for the same values of the bifurcation parameters above .
the initial condition considered for both diagrams is @xmath117.,title="fig:",scaledwidth=45.0% ] a ) bifurcation diagram of the system given by eqs . with and for the value of @xmath114 for figure
b ) depicts the bifurcation for the range @xmath115 for @xmath116 iterations .
figures c ) and d ) show the number of domains @xmath101 visited by the trajectory of the systems for the same values of the bifurcation parameters above .
the initial condition considered for both diagrams is @xmath117.,title="fig:",scaledwidth=45.0% ] + a ) bifurcation diagram of the system given by eqs . with and for the value of @xmath114 for figure
b ) depicts the bifurcation for the range @xmath115 for @xmath116 iterations .
figures c ) and d ) show the number of domains @xmath101 visited by the trajectory of the systems for the same values of the bifurcation parameters above .
the initial condition considered for both diagrams is @xmath117.,title="fig:",scaledwidth=45.0% ] a ) bifurcation diagram of the system given by eqs . with and for the value of @xmath114 for figure
b ) depicts the bifurcation for the range @xmath115 for @xmath116 iterations .
figures c ) and d ) show the number of domains @xmath101 visited by the trajectory of the systems for the same values of the bifurcation parameters above .
the initial condition considered for both diagrams is @xmath117.,title="fig:",scaledwidth=45.0% ] + by using this approach , the equilibrium points and the position of the commutation surfaces are preserved .
the parameter @xmath118 is used to change the eigenspectra of the linear part of systems , mainly the directions of the stable and unstable manifolds , so @xmath118 can be taken as a bifurcation parameter .
figures [ fig : bif ] a ) and b ) show the bifurcation diagram of the maximum local at every @xmath101 of each scroll depicted at @xmath1 for the range of the parameter @xmath119 and a zoomed area at @xmath120 .
both diagrams were calculated by the same initial condition @xmath117 , the difference is that figure [ fig : bif ] a ) was calculated for @xmath121 iterations while figure [ fig : bif ] b ) for @xmath116 iterations .
additionally figures [ fig : bif ] c ) and d ) show the number of domains @xmath101 that the system visited for the same values of @xmath118 , respectively .
notice that , if the range of @xmath122 approximately , the number of domains @xmath123 visited remain at the constant value of 3 , because the system presents only one attractor located in the inside domain as depicted in figure [ fig : multistability_1 ] .
this can be better appreciated in the zoomed area in figure [ fig : bif ] d ) .
projection of the trajectory of the system with onto the @xmath63 plane with @xmath89 , @xmath60 and @xmath124 . marked with red asterisk the equilibria of the system , and with gray line the commutation surfaces generated by the function
. the initial condition of the system is @xmath106.,width=415 ] with these parameter values the equilibria of the system is given by @xmath125 .
therefore @xmath89 and @xmath126 are assigned to the function in eq . .
the eigenspectrum of the matrix @xmath127 depends now on the bifurcation parameter @xmath118 , which for the value of @xmath124 is given by : @xmath128 proving that the definition [ def_usd3 ] is satisfied . considering these values , the system results in an interesting multistable state phenomena due to the round function and the direction of their eigenvectors as depicted in figure [ fig : multistability_1 ] for the initial condition @xmath106 .
the attractor is oscillating near the equilibria in the origin marked in red asterisk due to the initial condition given .
however there is no oscillation near the adjacent equilibrium points , the reason of this resides on the eigenvectors , as they are not located in the same way as in the previous examples , _
i.e. _ , the stable manifold of the oscillating domain , does nt match and cross with the stable manifolds of the adjacent domains .
this can be easily observed in figure [ fig : variedades ] as it is next explained .
figures [ fig : variedades ] a ) and b ) present a projection of a trajectory onto the plane @xmath63 for the systems with and , both initialized with the same initial condition near the origin @xmath106 , but considering the values in the bifurcation parameter of @xmath129 and @xmath124 , respectively .
notice in the projections that as the time increases the oscillating clockwise trajectory on both systems attractor grows larger , until eventually the trajectories on the scrolls cross the commutation surface close to the unstable manifold of @xmath10 marked in blue lines .
apparently both systems trajectory cross the commutation surface plane @xmath78 marked with black lines in a neighbourhood near to the intersection of the unstable and stable manifolds marked with blue and red lines ( depending if the trajectory is escaping or entering the domain ) , respectively .
however , by observing different projections of the attractors , for example the projection onto the plane @xmath64 in figures [ fig : variedades ] c ) and d ) , it can be appreciated that the direction of the manifold has been slightly changed due to the variation of the parameter from @xmath129 to @xmath124 .
take a closer look at the graph in figure [ fig : variedades ] c ) between @xmath130 . in this domain
@xmath131 the trajectory of the system is entering from the upper part close to @xmath132 where the black arrow depicts the direction of the crossing in the intersection of the unstable manifold with the commutation surface ( @xmath133 ) .
after entering to @xmath131 the trajectory is directed below the stable manifold @xmath9 in this domain marked with the red line , and then crosses to the domain @xmath74 near the intersection of the next stable manifold in @xmath74 in @xmath134 .
notice that some of the trajectories entering this domain end up oscillating into the scroll as the arrow in the lower part depicts ( this phenomena can also be seen in the projection of the multi - scroll attractor in figures [ fig : bi ] b ) and d ) ) .
nevertheless , some of the trajectories instead of reaching the scrolling plane in @xmath74 are redirected again to @xmath131 and oscillate in the scroll in the unstable manifold @xmath10 .
this behaviour is completely different in the projection of the attractor in figure [ fig : variedades ] d ) for @xmath135 . here , when the trajectory escapes the domain @xmath74 near @xmath136 and enters @xmath131 , the trajectory
is directed towards the location of the stable manifold and crosses near the intersection @xmath137 at approximately @xmath138 . after crossing the trajectory
is redirected towards the scroll in @xmath74 as the arrows depicts the direction .
this process repeats continuously from the three consecutive domains in which the trajectory of the system lies ( notice this also from figure [ fig : bif ] b ) ) .
now , in order to understand the phenomenon and visualize the exact location of the intersection of @xmath9 and @xmath10 along with the points belonging to the attractor @xmath26 at the commutation surface @xmath78 , a poincar plane was implemented exactly at the commutation surface .
first , the poincar plane is defined as @xmath139 , where @xmath140 are the coefficients of an hyperplane equation whose values are considered depending on the location under study , which in this case it will be in the commutation surfaces @xmath91 with @xmath141 guaranteeing @xmath142 . the crossing events of interest are @xmath143 with @xmath144 . where @xmath145 corresponds to the total of crossing events in @xmath146 , and @xmath147 correspond to the @xmath148-th intersection of @xmath149 in the @xmath150 direction .
the sub - index @xmath151 corresponds to trajectories that cross @xmath146 with @xmath152 , and @xmath153 corresponds to trajectories that cross @xmath146 with @xmath154 .
figures [ fig : poincare_sections ] a ) and b ) present the projections of @xmath155 for the values of @xmath156 and @xmath135 , respectively .
three points can be noticed from these projections .
trajectory projections for the system given by eq . , with @xmath89 and @xmath60 . for @xmath129 a ) onto the plane @xmath63
, c ) onto the plane @xmath64 . for @xmath124
b ) onto the plane @xmath63 , d ) onto the plane @xmath64 . marked with green line the commutation surfaces generated by the function , with red line the complex eigenvector and with black line the real eigenvector .
the black arrows show the trajectory direction .
, title="fig:",scaledwidth=45.0% ] trajectory projections for the system given by eq . , with @xmath89 and @xmath60 . for @xmath129 a ) onto the plane @xmath63 ,
c ) onto the plane @xmath64 . for @xmath124
b ) onto the plane @xmath63 , d ) onto the plane @xmath64 . marked with green line the commutation surfaces generated by the function , with red line the complex eigenvector and with black line the real eigenvector .
the black arrows show the trajectory direction .
, title="fig:",scaledwidth=45.0% ] + trajectory projections for the system given by eq .
, with @xmath89 and @xmath60 . for @xmath129 a ) onto the plane @xmath63 , c ) onto the plane @xmath64 . for @xmath124
b ) onto the plane @xmath63 , d ) onto the plane @xmath64 . marked with green line the commutation surfaces generated by the function , with red line the complex eigenvector and with black line the real eigenvector .
the black arrows show the trajectory direction . ,
title="fig:",scaledwidth=45.0% ] trajectory projections for the system given by eq .
, with @xmath89 and @xmath60 . for @xmath129 a ) onto the plane @xmath63 , c ) onto the plane @xmath64 . for @xmath124
b ) onto the plane @xmath63 , d ) onto the plane @xmath64 . marked with green line the commutation surfaces generated by the function , with red line the complex eigenvector and with black line the real eigenvector .
the black arrows show the trajectory direction .
, title="fig:",scaledwidth=45.0% ] when the trajectory of @xmath26 is increasing its scroll size it exits the current domain @xmath101 to @xmath157 ( or @xmath101 to @xmath158 ) through the commutation surface near the intersection of the unstable manifold with the poincar plane ( @xmath159 ) , as it can be appreciated with the intersection points of the trajectory @xmath160 marked in blue circles .
the blue triangle corresponds to the intersection of @xmath161 .
the blue line appearing in both figures [ fig : poincare_sections ] a ) and b ) , corresponds to the intersection of the unstable manifold with the poincar plane @xmath162 . notice that most of the trajectories are crossing near this section due to the scrolling behaviour in or near the unstable manifold
nevertheless , figure [ fig : poincare_sections ] a ) presents a region of escaping intersection points @xmath160 not near the unstable manifold , which comes out as one of the main difference between the escaping points in figure [ fig : poincare_sections ] b ) for a larger value of the bifurcation parameter .
also notice that the entering intersection points @xmath163 , marked with orange asterisks for @xmath156 are located around the intersection of the stable manifold @xmath164 marked with the red triangle in figure [ fig : poincare_sections ] a ) , but in figure [ fig : poincare_sections ] b ) , the events are located below this intersection in an apparently ranked way .
intersections of the trajectory of the system with and with the commutation surface @xmath78 for a ) @xmath129 , b ) @xmath124 . marked with blue circles the trajectories exiting @xmath165 near @xmath166 . the orange asterisk represent the trajectories entering @xmath165 .
the red triangle stands for @xmath167 , and the blue line corresponds to the intersection of unstable manifold and commutation surface.,title="fig:",width=340 ] intersections of the trajectory of the system with and with the commutation surface @xmath78 for a ) @xmath129 , b ) @xmath124 .
marked with blue circles the trajectories exiting @xmath165 near @xmath166 .
the orange asterisk represent the trajectories entering @xmath165 .
the red triangle stands for @xmath167 , and the blue line corresponds to the intersection of unstable manifold and commutation surface.,title="fig:",width=340 ] to determine the relationship among the bifurcation parameter and the multi - scroll or multistable solutions , the distance between these crossing trajectory events and the crossing of the corresponding manifold ( whether the trajectory is displacing in or out ) was calculated as follows : @xmath168 where @xmath169 corresponds to the @xmath148-th crossing event of the trajectory in the @xmath170 direction for the @xmath171 manifold intersection .
the triplet @xmath172 represents the coordinate of the @xmath148-th corresponding crossing event of the trajectory , and @xmath173 the coordinate of the intersection @xmath174 ( which were depicted with blue and red triangles in figure [ fig : poincare_sections ] ) .
the results of the distances for @xmath175 and @xmath176 are depicted in figure [ fig : distancia ] in orange dotted line and in blue continuous line , respectively .
the range considered in the experiment consists of values of the bifurcation parameter @xmath177 with a spacing of @xmath178 and an initial condition @xmath179 for figure [ fig : distancia ] a ) and @xmath180 for figure [ fig : distancia ] c ) .
distance calculated from eq . between the crossing trajectory events and the crossing of the corresponding manifold whether the trajectory is displacing inwards @xmath176 ( continuous blue line ) or outward @xmath175 ( dotted orange line ) .
the initial conditions considered are : a ) @xmath179 , c ) @xmath180 .
figures b ) and d ) depict the standard deviation @xmath181 from eq . between the distances for the each corresponding set of initial conditions.,title="fig:",scaledwidth=45.0% ]
distance calculated from eq . between the crossing trajectory events and the crossing of the corresponding manifold whether the trajectory is displacing inwards @xmath176 ( continuous blue line ) or outward @xmath175 ( dotted orange line ) .
the initial conditions considered are : a ) @xmath179 , c ) @xmath180 .
figures b ) and d ) depict the standard deviation @xmath181 from eq . between the distances for the each corresponding set of initial conditions.,title="fig:",scaledwidth=45.0% ] + distance calculated from eq .
between the crossing trajectory events and the crossing of the corresponding manifold whether the trajectory is displacing inwards @xmath176 ( continuous blue line ) or outward @xmath175 ( dotted orange line ) .
the initial conditions considered are : a ) @xmath179 , c ) @xmath180 .
figures b ) and d ) depict the standard deviation @xmath181 from eq . between the distances for the each corresponding set of initial conditions.,title="fig:",scaledwidth=45.0% ] distance calculated from eq . between the crossing trajectory events and the crossing of the corresponding manifold whether the trajectory is displacing inwards @xmath176 ( continuous blue line ) or outward @xmath175 ( dotted orange line ) .
the initial conditions considered are : a ) @xmath179 , c ) @xmath180 .
figures b ) and d ) depict the standard deviation @xmath181 from eq . between the distances for the each corresponding set of initial conditions.,title="fig:",scaledwidth=45.0% ] both distances present a common behaviour for @xmath182 , _
i.e. _ , the distance @xmath175 loses its spiking variational behaviour and becomes smoother , while @xmath176 starts to increment in comparison with @xmath175 . to quantify this variations among the intersections points the standard deviation of the distances is represented in figure [ fig : distancia ] b ) and d ) for the initial conditions previously mentioned , and it is descried as follows : @xmath183 where @xmath184 corresponds to the number of values of @xmath118 evaluated ( for the range discussed above @xmath185 ) , and @xmath186 is the mean of @xmath187 .
these values were calculated by the @xmath188 matlab function . for both initials conditions figure b ) and d ) present a relation between @xmath175 and @xmath176 depicted by the values of @xmath181 in the range of 0 and @xmath189 for @xmath190 .
however , @xmath191 the standard deviation among the distances increases .
this correlates with the previous result of the bifurcation diagram in figure [ fig : bif ] for the same values of @xmath192 .
the last point to address is that the distance between the stable and unstable manifold crossings in the poincar plane ( marked with red and blue triangles for @xmath193 and @xmath194 , respectively ) is smaller in figure [ fig : poincare_sections ] a ) than the one in figure [ fig : poincare_sections ] b ) .
intersections of the stable and unstable manifolds @xmath9 and @xmath195 with the commutation surface @xmath78 for @xmath196.,width=415 ] to study in more detail this displacement in space of the manifolds , an analysis of the system s eigenspectra due to the variation of the bifurcation parameter was implemented .
the manifolds direction and their crossing position with the commutation surface was analysed with the poincar plane .
the range of the bifurcation parameter considered is of @xmath197 , from which for a step of @xmath178 the set of eigenvectors were calculated .
figure [ fig : poincare_eingevector ] displays the crossing locations due to this variation . and it can be appreciated that the intersection of the manifolds @xmath198 with @xmath199 ( marked with asterisk and circles , respectively ) are increasing their separation regarding the variation of @xmath118 shown in the color gradient .
this result correlates with the previous analysis of the distance and the escaping locations of the system s trajectory . in the same context , in figure [ fig : eigenvalores ]
are depicted the eigenvalues of the system @xmath200 with @xmath201 for the same range of @xmath197 resulting in an interesting contrast in their values .
the complex conjugate eigenvalues start to decrement their real part until they become negative for values of @xmath202 ( not satisfying definition 2.1 ) , and at the same time the imaginary part of the complex conjugate increases ( ensuring an augment in the frequency of oscillation ) . on the other hand , @xmath21 ,
remains negative with a decreasing magnitude .
this results in unstable uds of the type i equilibria ( for @xmath203 ) , that present a stretch oscillation in the scroll with a stronger attraction towards @xmath9 . summarizing ,
the variation of the number of scrolls and domains visited due to the values in the bifurcation parameter ( figure [ fig : bif ] ) , the distance of the intersecting locations of the system s trajectory in the commutation surfaces with respect to the location of the intersection of manifolds ( figure [ fig : poincare_sections ] ) , are caused by the variation of the eigenvalues of the system ( figure [ fig : eigenvalores ] ) , resulting in the transition between unstable multi - scroll structures and the multistable structure .
eigenvalues of the system with for @xmath204 .
figure a ) depicts the real eigenvalue @xmath21 , while b ) depicts the complex conjugated @xmath23 .
, width=415 ] this comes as a mayor advantage in the proposed system as one may consider different initial conditions for values of @xmath205 , and for each initial location given the system will oscillate near those continuous domains along the @xmath1 axis from @xmath103 .
consider the different sets of initial conditions @xmath206 , @xmath207 , @xmath208 and @xmath209 for @xmath135 . the resulting trajectory of the system given these conditions are depicted in figure [ fig : multistability_2 ] , each scroll presented is a unique experiment for a given number of iterations in time .
notice that the orbit does nt form attractors on the neighborhood domains .
resulting in a multistable system for the @xmath1 axis from @xmath103 which oscillates around the equilibria in which the initial condition is set , and the trajectory is bounded by the dynamics of the left and right equilibrium points although there are no oscillations around them .
additionally , the largest lyapunov exponent of the attractor was calculated throughout the algorithm proposed by rosenstein _ et .
_ in @xcite , taking a value of @xmath210 demonstrating a chaotic behaviour in the system .
projections of the trajectories of the system with onto the @xmath63 plane with @xmath124 , @xmath89 and @xmath60 for the initial condition sets @xmath211 , @xmath207 , @xmath208 and @xmath209.,width=415 ]
using the nearest integer or @xmath0 function it has been proved that multiple final states can be acquired in pwl systems considered in the uds of the type i theory .
the multistability phenomena is acquired by means of the changes in the bifurcation parameter @xmath118 , which correlate directly with the direction and location of their stable and unstable manifolds as well as the values of their eingenvalues .
it has been proven that with specific @xmath118 values the system may present multi - scroll attractors if the eigenspectra satisfies the specifications discussed here .
otherwise the systems trajectory will be trapped between the adjacent equilibria regarding on the initial condition given to the system . thus resulting in a multistable system along the @xmath1 axis .
the methodology discussed here may be applied to further axes in order to obtain multistability along @xmath212 or @xmath34 .
the results of this may be reported elsewhere .
h.e.g.v is a doctoral fellow of the conacyt in the graduate program on control and dynamical systems at dmap - ipicyt . l.j.o.g . acknowledges the uaslp for the financial support through c16-fai-09 - 46.46 .
e.c.c . acknowledges the conacyt financial support for sabbatical at department of mathematics , university of houston .
aihara , k. & suzuki , h. [ 2010 ] _
`` theory of hybrid dynamical systems and its applications to biological and medical systems''_. philosophical transactions of the royal society a : mathematical , physical and engineering sciences , * 368*(1930 ) , 48934914 .
campos cantn , e. , barajas - ramrez , j. g. , solis - perales , g. & femat , r. [ 2010 ]
_ `` multiscroll attractors by switching systems''_. chaos : an interdisciplinary journal of nonlinear science , * 20*(1 ) , 013116 .
huerta - cuellar , g. , jimnez - lpez , e. , campos - cantn , e. & pisarchik , a.n .
[ 2014 ] _ `` an approach to generate deterministic brownian motion''_. communications in nonlinear science and numerical simulation , * 19*(8 ) , 2740 - 2746 .
jimnez - lpez , e. , gonzlez
salas , j. , ontan
garca , l. j. , campos - cantn , e. & pisarchik , a.n .
[ 2013 ] _ `` generalized multistable structure via chaotic synchronization and preservation of scrolls''_. journal of the franklin institute , * 350*(10 ) , 2853 - 2866 .
marmillot , p. , kaufman , m. & hervagault , j.f .
[ 1991 ] _ `` multiple steady states and dissipative structures in a circular and linear array of three cells : numerical and experimental approaches''_. the journal of chemical physics , * 95*(2 ) , 1206 - 1214 .
ontan garca , l. j. & campos - cantn , e. [ 2015 ] _ `` displacement in space of the equilibria of unstable dissipative systems''_. 7th international scientific conference on physics and control , 19 - 22 august , 2015 , istanbul , turkey .
zhusubaliyev , z. t. , mosekilde , e. , churilov , a. n. , & medvedev , a. [ 2015 ] _
`` multistability and hidden attractors in an impulsive goodwin oscillator with time delay''_. the european physical journal special topics , * 224*(8 ) , 1519 - 1539 . | a multistable system generated by a piecewise linear ( pwl ) system based on the jerky equation is presented .
the systems behaviour is characterised by means of the nearest integer or @xmath0 function to control the switching events and to locate the corresponding equilibria among each of the commutation surfaces .
these surfaces are generated by means of the switching function dividing the space in regions equally distributed along one axis .
the trajectory of this type of system is governed by the eigenspectra of the coefficient matrix which can be adjusted by means of a bifurcation parameter .
the behaviour of the system can change from multi - scroll attractors into a mono - stable state to the coexistence of several single - scroll attractors into a multi - stable state .
numerical results of the dynamics and bifurcation analyses of their parameters are displayed to depict the multi - stable states .
multistability ; piecewise linear systems ; eigenspectra ; chaos ; multi - scroll attractor ; round function ; bifurcation . |
it is well known that in an @xmath2-tree the intersection of any two metric balls is itself a metric ball . in this paper
we will show that this is actually a characterization of @xmath0-trees , and that , more generally , the geometry of the intersection of balls encodes information about the curvature of a geodesic metric space .
recall from @xcite , @xcite , @xcite or @xcite that a geodesic metric space is _ hyperbolic _ ( in the sense of gromov ) if there is a constant @xmath3 such that for any geodesic triangle , any one side is contained in the @xmath4-neighborhood of the union of the two other sides .
we prove the following characterization of hyperbolicity .
[ qb]a geodesic metric space @xmath5 is hyperbolic if and only if the intersection of any two metric balls is at uniformly bounded hausdorff distance from a ball . studying curvature in terms of the geometry of the intersection of metric balls turns out to be very natural and both gromov hyperbolicity and the notion of cat(@xmath1 ) geometry may be characterised in theses terms ( see section [ ext ] ) .
tracking constants in the proof of theorem [ qb ] , one can show that the hyperbolicity constant depends only on the eccentricity constant .
as pierre pansu pointed out to us it is a then an easy observation to deduce that the hyperbolicity bound varies linearly with the eccentricity bound so we obtain : [ rtreeschar ] a geodesic metric is an @xmath0-tree if and only if the intersection of any two balls is a ball .
this characterisation of @xmath2-trees was first conjectured in an early version of this paper and wenger has independently established the conjecture using very different methods , see @xcite .
our approach is entirely self - contained .
the following notion is crucial for our purposes .
[ eccentricity]we say that a set @xmath6 has _ eccentricity less than _ @xmath4 ( for some @xmath3 ) if there is @xmath7 such that @xmath8 for some @xmath9 . by convention
the eccentricity of the empty set is 0 .
we shall see that the intersection of balls having uniformly bounded eccentricity is also equivalent to hyperbolicity ( proposition [ hyp=>qb ] and lemma [ qb=>hyp ] ) .
the paper is organized as follows . in section [ bigons ]
we discuss the geometry of @xmath10-quasigeodesics following an idea of papasoglu in @xcite and pomroy in @xcite , which is an important step in the proof . section [ div ] discusses divergence functions and a quantitative version of a theorem in @xcite and a classical argument implying hyperbolicity .
section [ proofs ] collects the proofs of theorem [ qb ] and corollary [ rtreeschar ] .
the idea is to show that hyperbolicity is equivalent to intersections of balls having uniformly bounded eccentricity .
one technical difficulty lies in the fact that the centre and radius of a ball are not , in general , well defined . when they are an elementary proof can be given as in section [ ext ] .
pomroy s work appeared in his warwick university masters dissertation but has never been published . in the appendix
we take this opportunity to place his main theorem on the record with our own variation on the proof .
the figures may be found at the back of the paper , inserted prior to the bibliography . _
acknowledgements : _ the authors are extremely indebted to pierre pansu for pointing out to us that our bound on the hyperbolicity constant implied a linear bound via scaling .
we would also like to thank theo buehler and deborah ruoss for pointing out an error in an early draft of this paper and chris hruska for the reference @xcite .
finally we would like to thank hamish short for his comments on the classical argument proving that non - linear divergence of geodesics implies hyperbolicity .
in this section we establish preliminary results concerning the geometry of geodesics and quasi - geodesics in a geodesic metric space .
we start with the simple observation that if two geodesics are synchronously far apart , then they are asynchronously at least half as far apart as well .
[ synchronous ] let @xmath11 and @xmath12 be geodesics with @xmath13 .
if there exists @xmath14 such that @xmath15 then @xmath16 for all @xmath17 .
suppose that there is an @xmath18 such that @xmath19 . then since @xmath11 is a geodesic we have @xmath20 . hence @xmath21 .
but by hypothesis and the triangle inequality @xmath22 this is a contradiction .
similarly , if there is an @xmath23 such that @xmath19 .
then since @xmath11 is a geodesic we have @xmath24 . hence @xmath25 .
but by hypothesis and the triangle inequality @xmath26 this is again a contradiction . for a constant @xmath27
a @xmath10-quasigeodesic is a continuous map @xmath28\to x$ ] such that @xmath29 , @xmath30 and for all @xmath31 $ ] @xmath32 the points @xmath33 and @xmath34 are said to be the endpoints of @xmath11 .
a @xmath10 quasi - geodesic bigon is a pair @xmath35 of @xmath10 quasi - geodesic bigons which have the same end points .
the images of the quasigeodesics @xmath11 and @xmath12 in @xmath36 are called the _ sides _ of the @xmath37-bigon . for @xmath38 , we say that a @xmath37-bigon is @xmath39-_fat _ if there are @xmath40 such that @xmath41 and that it is @xmath39-_thin _ if it is not @xmath39-fat .
the following remark provides an easy mechanism for constructing @xmath10 quasi - geodesics .
let @xmath38 and @xmath42 in @xmath36 .
a _ @xmath39-path _ from @xmath43 to @xmath44 is a continuous path @xmath45 from @xmath43 to @xmath44 such that , for any @xmath46 for some @xmath47 @xmath48 given any point on @xmath49 on a @xmath50-quasi - geodesic with end points @xmath42 we obtain a @xmath39-path by taking a broken geodesic @xmath51 .
the main result in this section is that if two geodesics start at the same point and travel `` almost parallel '' for long enough , then this ensures the existence of @xmath52-bigons for any @xmath53 whose fatness depends on the distance between the two geodesics .
the proof of this follows the outline in @xcite .
more precisely we prove the following . [ meat]let @xmath5 be a geodesic metric space , and choose integers @xmath54 with @xmath55 , @xmath56 .
assume that there are two geodesics @xmath11 , @xmath12 and a real number @xmath57 such that @xmath29 and @xmath58 $ ] for all @xmath59 .
if @xmath36 does not contain a @xmath60-fat @xmath61-bigon then @xmath62 in words , there is an upper bound on the length for which two geodesics can travel at a controlled distance . for the purpose of the proof
we introduce a few definitions .
take @xmath63 $ ] , we call an element @xmath64 a @xmath65-_point _ if @xmath66 .
we say that @xmath47 is an integral @xmath65-point if @xmath47 is a positive integer and a @xmath65-point .
we define the @xmath65-_distance _ between two integral @xmath65-points @xmath67 by @xmath68 and set @xmath69 if and only if @xmath70 .
note that since we are assuming that @xmath71 are integers @xmath72 .
the interval @xmath73 $ ] is covered by @xmath74 disjoint half open intervals of the form @xmath75 , where @xmath76 $ ] . given that @xmath36 does not contain a @xmath60-fat @xmath61-bigon we claim the following : _ claim : _ for any two integral @xmath65-points @xmath47 and @xmath77 , if @xmath78 , then @xmath79 we prove the above claim by induction over @xmath80 : first , for @xmath81 we have @xmath70 and @xmath82 because @xmath70 is a @xmath65-point .
so we assume the claim is true for @xmath83 and show it for @xmath84 .
given integral @xmath65-points @xmath71 with @xmath85 we let @xmath86 be the integral @xmath65 points such that @xmath87 note that @xmath88 because @xmath89 , and hence both @xmath90 and @xmath91 are greater than or equal to @xmath92 .
set @xmath93 , as shown in figure 1 .
suppose for a contradiction that @xmath94 .
we will show that this implies the broken geodesics @xmath95 and @xmath96 form a @xmath60-fat @xmath61-bigon .
since @xmath36 contains no such bigons we will conclude that in fact @xmath94 . to see that @xmath95 is a @xmath61-path we use the fact that @xmath97 and that , by our induction hypothesis , @xmath98 : @xmath99 the last inequality comes from our supposition that @xmath94 .
a similar argument shows that the broken geodesic @xmath96 is also a @xmath61-path if @xmath94 .
now let @xmath100 be a point on the path @xmath96 which minimises the distance to @xmath101 .
in particular @xmath102 .
if @xmath100 lies on the arc @xmath103 then , by the triangle inequality , we compute : @xmath104 it follows that @xmath105 , which is a contradiction since we assumed @xmath56 and @xmath106 .
it follows that @xmath100 must lie on the arc @xmath107 . but
applying lemma [ synchronous ] to the geodesics @xmath35 with @xmath108 we see that @xmath109 and hence the bigon is @xmath60-fat as required . hence @xmath110 completing the induction . now by the triangle inequality @xmath111 hence @xmath112 , so @xmath113 .
so , @xmath114 , and @xmath115 because there are at least @xmath116 integer points in @xmath117 and each of these is a @xmath65-point for one of the @xmath74 possible values of @xmath65 .
recall that for a geodesic metric space @xmath5 , a _ divergence function _ is a map @xmath118 such that for all @xmath119 , all @xmath120 and all geodesics @xmath121 $ ] , @xmath122 $ ] such that @xmath123 , if @xmath124 is such that @xmath125 and @xmath126 is a path in the closure of @xmath127 from @xmath128 to @xmath129 , then the length of @xmath126 is at least @xmath130 .
we say that _ geodesics diverge in @xmath36 _ if there is a divergence function @xmath131 so that @xmath132 .
papasoglu showed in @xcite[corollary 1.3 ] that a geodesic metric space @xmath5 is hyperbolic if and only if geodesics diverge in @xmath36 . here
we provide a quantitive version of this result in order to relate the hyperbolicity constant for a space to the eccentricity bound .
first , following @xcite we provide candidates for a divergence function .
let @xmath133 and for @xmath124 define @xmath134 since @xmath36 is a geodesic space any path joining two points @xmath135 must have length at least @xmath136 , and setting @xmath137 it is easy to see that the function @xmath138 is a divergence function for @xmath36 .
[ thinbigonsquant]let @xmath5 be a metric space such that any @xmath37-bigon is @xmath139-thin .
then for @xmath140 and @xmath141 , any @xmath142 such that @xmath143 satisfies @xmath144 in words , @xmath145 and hence geodesic diverge . before starting with the proof
we show an intermediate result .
suppose that @xmath5 is a metric space such that @xmath37-bigons are @xmath146-thin for some constant @xmath147 and some @xmath53 .
let @xmath148 and @xmath149 .
if two geodesics @xmath11 and @xmath12 starting at the same point are such that @xmath150 for some @xmath7 and @xmath151 for some @xmath142 , then @xmath152\ ] ] for any @xmath153 $ ] .
let us denote by @xmath154 . take @xmath155 to be the midpoint on a geodesic @xmath156 from @xmath128 to @xmath129 .
then the broken geodesics @xmath157 and @xmath158 form the sides of an @xmath159-bigon , which by assumption is @xmath160-thin . according to lemma [ synchronous ] , this bigon is at least @xmath161-fat since the distance between the two geodesics is more than @xmath162 at time @xmath163 .
hence @xmath164 .
the upper bound is obtained in a similar way by looking at the broken geodesics @xmath165 and @xmath166 .
they form a @xmath167-bigon , which by assumption is @xmath168-thin .
but by lemma [ synchronous ] , this bigon is at least @xmath159-fat since the distance between the two geodesics is @xmath169 at time @xmath170 .
hence @xmath171 as required .
we now can easily prove proposition [ thinbigonsquant ] .
since @xmath143 there are two geodesics @xmath11 and @xmath12 and @xmath120 such that * @xmath29 * @xmath150 * @xmath172
. we can apply theorem [ meat ] with @xmath173 , @xmath174 and @xmath175 which can be done using the previous lemma with @xmath176 .
the assumptions on @xmath162 and @xmath177 ensure that @xmath178 . from now on
we assume that @xmath140 so that the function @xmath138 satisfies the conclusions of proposition [ thinbigonsquant ] for appropriate constants .
the next step is to replace the divergence function @xmath138 by a divergence function @xmath179 of exponential growth . given a rectifyable path @xmath126 , let us denote by @xmath180 its length .
for @xmath124 we set @xmath181 where @xmath35 are geodesics , @xmath182 and the infimum is taken over all geodesics @xmath35 and all points @xmath119 and all @xmath183 .
it is clear that if we define @xmath184 then @xmath179 is a divergence function on @xmath36 and that @xmath185 for all @xmath186 .
the following shows that this divergence function has exponential growth .
[ exponentialdivergence]for any @xmath187 and @xmath188 , then @xmath189 where @xmath190 and @xmath191 .
in particular , @xmath179 has exponential growth .
we start with an intermediate result .
[ smalllemma ] if @xmath11 and @xmath12 are geodesics with @xmath192 and @xmath57 satisfies @xmath150 then @xmath193 , where @xmath194 is as in proposition [ exponentialdivergence ] .
let @xmath195 be a geodesic joining @xmath196 to @xmath197 .
suppose for a contradiction that @xmath195 has length less than @xmath198 so no point on @xmath195 lies in the interior of the ball @xmath199 , where @xmath200 .
concatenate @xmath195 with the terminal subarcs of @xmath35 of length @xmath198 to obtain a path joining @xmath201 to @xmath202 .
no point on this path lies in the interior of the ball @xmath199 so it must have length at least @xmath203 .
on the other hand we see that the path has length @xmath204 which by assumption is less than @xmath205 so we get @xmath206
. however @xmath207 by choice of @xmath194 and this is a contradiction .
arguing by induction on @xmath208 it is enough to show that @xmath209 for any @xmath210 .
if @xmath211 we are done so we can assume that @xmath212 is finite and therefore there are geodesics @xmath35 with @xmath182 and @xmath57 such that @xmath150 such that there is an arc @xmath126 in the closure of the complement of the ball @xmath213 which joins @xmath128 and @xmath129 and which has length less than @xmath214 .
let @xmath215\mid \alpha(t)\in b(x , r+r+n)\}$ ] and @xmath216\mid \alpha(t)\in b(x , r+r+n)\}$ ] .
note that @xmath217 and @xmath218 so the subarc @xmath219}$ ] must have length less than or equal to @xmath220 .
let @xmath221 and @xmath222 be geodesics from @xmath43 to @xmath223 and @xmath224 respectively . by combining the triangle inequality lemma [ smalllemma ]
we see that @xmath225 it follows that at least one of the three terms in the above sum must be greater than or equal to @xmath162
. it can not be the middle term for the following reason : if @xmath226 then , by definition of @xmath179 , any path joining @xmath227 to @xmath228 which lies in the closure of the complement of the ball @xmath229 must have length at least @xmath212 so in particular the subarc @xmath219}$ ] must have length at least @xmath212 .
on the other hand we observed that this subarc has length at most @xmath220 so we see that @xmath230 .
since @xmath231 we have that @xmath232 and since @xmath233 this is a contradiction .
so we may assume ( interchanging @xmath11 and @xmath221 with @xmath12 and @xmath222 if necessary ) that @xmath234 .
we construct a path @xmath235 in the complement of the ball @xmath236 joining @xmath128 to @xmath237 by concatenating the subarc @xmath238}$ ] with the subarc @xmath239 of @xmath221 joining @xmath237 to @xmath223 . since @xmath223 is within @xmath240 of @xmath43 we see that @xmath239 has length at most @xmath194 , and so @xmath235 has length at most @xmath241 . on the other hand
the length of @xmath235 is bounded below by @xmath242 , so we see that @xmath243 but recall that @xmath231 so we see that @xmath244 .
finally , since @xmath245 and by definition of @xmath246 , we have that @xmath247 .
it follows that @xmath248 and so @xmath249 it follows from lemma [ exponentialdivergence ] that for any affine function @xmath250 there is some @xmath251 such that @xmath252 for all @xmath253 .
the value @xmath251 depends only on the function @xmath138 and the constants @xmath254 and @xmath255 , though there may be no closed formula to compute it .
it is clear that the value of @xmath251 may be bounded in terms of the values of the constants @xmath194 and @xmath246 appearing in lemma [ exponentialdivergence ] , and those depend only on the constant @xmath162 and the eccentricity bound @xmath52 as shown in the next lemma .
this will enable us to show that there is an upper bound on the hyperbolicity constant for @xmath36 which is a function of @xmath52 alone .
[ constants ] let @xmath256 be a geodesic metric space such that for each @xmath37 the @xmath37-bigons are @xmath257-thin .
let @xmath140 , @xmath258 and @xmath259 as in proposition [ exponentialdivergence ] .
then @xmath194 and @xmath260 are bounded above by functions of @xmath52 and @xmath162 . more precisely , 1 .
@xmath261 2 .
@xmath262 \(1 ) apply proposition [ thinbigonsquant ] with @xmath263 .
\(2 ) apply proposition [ thinbigonsquant ] with @xmath264 .
[ thinbigons]let @xmath256 be a geodesic metric space such that for each @xmath37 the @xmath37-bigons are @xmath257-thin .
then @xmath36 is @xmath265-hyperbolic , for some function @xmath4 depending on @xmath52 alone . to ensure that @xmath140 we set @xmath266
let @xmath267 and choose geodesics @xmath268 , \alpha_x=[y , z ] , \alpha_y=[z , x]$ ] in @xmath36 .
we denote by @xmath269 the reverse of the geodesic @xmath270 .
we wish to estimate the thickness @xmath4 of this geodesic triangle @xmath271 using the exponential divergence function @xmath179 defined above .
the following argument is adapted from that given in @xcite .
let @xmath272 and set @xmath273 and @xmath274 .
similarly we define @xmath275 , @xmath276 , @xmath277 , @xmath277 , @xmath278 and @xmath279 .
now set @xmath280 , @xmath281 and @xmath282 case 1 .
at least one of the values , say @xmath283 , is non - positive . in this case
any point on @xmath284 is within @xmath162 of the other two sides , while any point on the subarcs @xmath285 or @xmath286 is within @xmath162 of the other two sides .
there is @xmath287 on @xmath284 between @xmath277 and @xmath288 that is within @xmath162 of both @xmath289 and @xmath290 .
hence the broken paths @xmath291 and @xmath292 form a @xmath162-bigon , consequently the triangle is @xmath293-thin .
all three of the values @xmath294 are positive . in this case
we will show that we can bound all three of these values by some uniform value @xmath295 given in terms of @xmath162 and @xmath52 , and hence , setting @xmath266 , @xmath295 depends on @xmath52 alone .
once this is done it is clear that any point on any side of the triangle is within @xmath296 of some point on one of the other two sides , so it remains to find this bound @xmath295 .
we can assume that @xmath297 .
first note that if @xmath298 we can run the argument from case 1 with @xmath299 in place of @xmath162 , so we may assume that @xmath300 .
we claim that the interior of @xmath301 does nt intersect the geodesic @xmath302 .
indeed , since @xmath303 , the interior of the balls @xmath301 and @xmath304 have disjoint intersection .
similarly , the balls @xmath301 and @xmath305 have disjoint interiors .
but since @xmath306 , the arc @xmath302 is contained in the union of @xmath305 and @xmath304 . now let @xmath307 and @xmath308 .
the arcs @xmath309 , @xmath310 , @xmath311 , @xmath312 and @xmath313 are in the complement of the ball @xmath301 ( since they are either in the ball @xmath304 or @xmath305 ) .
so concatenating those arcs we obtain a path in the closure of the complement of the ball @xmath314 of length @xmath315 . applying the divergence function to the geodesics @xmath316 emanating from the point @xmath43
we see that @xmath317 .
now choose an integer @xmath318 so that @xmath319 $ ] where @xmath320 are the constants estimated in lemma [ constants ] . if @xmath321 then @xmath322 and , since @xmath246 , @xmath194 depend only on @xmath52 , we are done . if @xmath323 we can apply theorem [ exponentialdivergence ] to show that @xmath324 .
it follows that @xmath208 can also be bounded above in terms of @xmath52 and @xmath162 . as noted before we may choose @xmath266 to obtain a bound on @xmath208 in terms of @xmath52 alone .
we denote this bound by @xmath325 .
this gives a bound on @xmath326 since @xmath327 .
this in turn bounds the fatness of the triangle as less than or equal to @xmath328 we have shown that any geodesic triangle is either @xmath329-thin , which since @xmath266 means that the triangle is @xmath330-thin , or it is @xmath331-thin , or it is @xmath332-thin . in the second case and third case the constants @xmath333 can all be written in terms of @xmath52 alone . if we take @xmath334 then we see that the space @xmath5 is @xmath265-hyperbolic as required .
the proof of theorem [ qb ] is a sequence of simple observations , combined with theorem [ thinbigons ] . recall that
if @xmath335 are subsets of a metric space @xmath5 , then the hausdorff distance between @xmath336 and @xmath337 is given by @xmath338 where for @xmath186 , @xmath339 is the @xmath260-neighborhood of @xmath336 .
it is not clear how having eccentricity less than a constant @xmath4 and being at hausdorff distance less than @xmath4 to a ball are related in general , but in case of intersection of balls in a metric space those notions are equivalent .
our first observation in this section holds for any geodesic metric space and gives the interior radius of the intersection of two balls .
[ general]let @xmath5 be a geodesic metric space .
* for any @xmath340 with @xmath341 and @xmath342 , if the balls @xmath343 and @xmath344 are neither disjoint nor nested , then @xmath345 where @xmath346 and @xmath347 is any point on any geodesic between @xmath43 and @xmath44 , at distance @xmath348 from @xmath43 . *
if @xmath349 and @xmath350 , then @xmath351 .
\(1 ) if the balls @xmath343 and @xmath344 are neither disjoint nor nested , then we have @xmath352 and @xmath353 , and hence given any geodesic @xmath11 from @xmath43 to @xmath44 we may take a point @xmath347 on @xmath11 at distance @xmath348 from @xmath43 and a ball of radius @xmath260 around @xmath347 . then for @xmath354 @xmath355 and similarly @xmath356 .
\(2 ) first notice that our assumptions on @xmath17 and @xmath47 being strictly smaller than @xmath357 show that @xmath43 and @xmath44 do not belong to @xmath358 .
let @xmath359 and @xmath360 , so that @xmath361 , @xmath362 and @xmath363 . take a geodesic @xmath364 from @xmath43 to @xmath100 and a point @xmath365 on this geodesic , at distance @xmath163 from @xmath100 .
such a point exists because @xmath43 does not belong to @xmath366 .
since @xmath367 we have @xmath368 similarly , take a geodesic @xmath369 from @xmath44 to @xmath100 and a point @xmath287 on this geodesic , at distance @xmath163 from @xmath100 .
since @xmath370 we have @xmath371 combining the 2 inequalities gives @xmath372 , hence @xmath373
. we can now prove one implication of theorem [ qb ] and corollary [ rtreeschar ] .
[ hyp=>qb ] if a geodesic metric space @xmath5 is @xmath4-hyperbolic with hyperbolicity constant less than or equal to @xmath3 , then both the eccentricity of the intersection of any two balls and the hausdorff distance from the intersection to a ball are both uniformly bounded by @xmath374 .
take @xmath340 with @xmath375 and @xmath376 , with @xmath377 .
we will show that the eccentricity of @xmath358 is less than @xmath374 .
this implies the statement about hausdorff distance as well , by definition of hausdorff distance ( recalled in the beginning of this section ) . according to lemma [ general ] part ( 1 ) either @xmath378 , @xmath379 , @xmath380 or @xmath381 , where @xmath347 and @xmath260 are as defined in that lemma . in the first three cases
the eccentricity is clearly bounded by @xmath382 . in the remaining case it suffices to show that there is a constant @xmath52 independent of @xmath383 such that @xmath358 is contained in some ball of radius @xmath384 .
we will show that in fact @xmath385 .
now , for @xmath386 , let us estimate the distance to @xmath347 .
since @xmath347 lies on a geodesic from @xmath43 to @xmath44 it is within @xmath4 of a point @xmath49 which lies on a geodesic from @xmath44 to @xmath365 or on a geodesic from @xmath43 to @xmath365 .
we first assume that @xmath49 lies on a geodesic from @xmath43 to @xmath365 . by the triangle inequality
, we have that @xmath387 . since @xmath49 lies on a geodesic from @xmath43 to @xmath365 this yields @xmath388 .
now we have , as required : @xmath389 if @xmath49 lies on a geodesic from @xmath44 to @xmath365 instead , then we use the same argument switching the roles of @xmath42 and of @xmath40 .
the following says that in a geodesic metric space such that the intersection of any two balls has uniformly bounded eccentricity , then the set of points on @xmath39-paths is uniformly close to a geodesic .
[ dpathtravelclose]suppose that @xmath5 has a uniform bound @xmath53 on the eccentricity of the intersection of any two balls .
then for any @xmath27 , given any two points @xmath340 , any point on a @xmath37-path from @xmath43 to @xmath44 is contained in the @xmath390-neighbourhood of any geodesic from @xmath43 to @xmath44 .
let @xmath365 be a point on a @xmath37-path @xmath45 from @xmath43 to @xmath44 with @xmath391 , @xmath392 so that @xmath393 . if @xmath394 then @xmath395 and @xmath365 is within @xmath37 of @xmath44 .
similarly if @xmath396 then @xmath397 and @xmath47 is within @xmath37 of @xmath43 so we may assume that both @xmath398 .
now let @xmath399 , so that
@xmath400 so by lemma [ general ] point ( 2 ) we see that any ball contained in @xmath401 has radius at most @xmath37 .
it follows from the bounded eccentricity hypothesis that @xmath402 for some point @xmath403 .
let @xmath11 be a geodesic from @xmath43 to @xmath44 and @xmath347 any point on @xmath11 at distance less than @xmath17 to @xmath43 and less than @xmath47 to @xmath44 ( such a point exists because @xmath404 ) .
then @xmath405 , hence @xmath406 . _
remark : taking @xmath407 in the lemma above shows that in a geodesic metric space such that the intersection of any two metric balls has eccentricity less than or equal to @xmath4 , any geodesic between two points is contained in a @xmath4-neighbourhood of any other geodesic between those two points .
_ an analogous result holds in terms of hausdorff distance .
[ dpathtravelclose2]suppose that @xmath5 has a uniform bound @xmath53 on the hausdorff distance from the intersection of any two balls to a ball .
then for any @xmath27 , given any two points @xmath340 , any point on a @xmath37-path from @xmath43 to @xmath44 is contained in the @xmath408-neighbourhood of any geodesic from @xmath43 to @xmath44 .
let @xmath365 be a point on a @xmath37-path @xmath45 from @xmath43 to @xmath44 with @xmath391 , @xmath392 so that @xmath393 . if @xmath409 then @xmath410 and @xmath365 is within @xmath411 of @xmath44 .
similarly if @xmath412 then @xmath413 and @xmath47 is within @xmath411 of @xmath43 so we may assume that both @xmath414 .
now let @xmath399 , so that @xmath400 . by assumption , @xmath415 , for some @xmath403 and some @xmath7 , which implies that @xmath416 and that @xmath417 .
so by lemma [ general ] point ( 2 ) we see that any ball contained in @xmath418 has radius at most @xmath419 .
it follows that @xmath420 .
let @xmath11 be a geodesic from @xmath43 to @xmath44 and @xmath347 any point on @xmath11 at distance less than @xmath17 to @xmath43 and less than @xmath47 to @xmath44 ( such a point exists because @xmath404 ) .
then @xmath421 , hence @xmath422 .
we can now prove the other implication in theorem [ qb ] , namely that ( b ) implies ( a ) . [ qb=>hyp]there is a function @xmath423 such that : 1 .
if @xmath5 is a geodesic metric space with the property that the intersection of any two balls has eccentricity bounded by @xmath52 then @xmath36 is @xmath265-hyperbolic .
if @xmath5 is a geodesic metric space with the property that the intersection of any two balls is at hausdorff distance less than @xmath424 then @xmath36 is @xmath265-hyperbolic .
according to lemma [ dpathtravelclose ] or lemma [ dpathtravelclose2 ] , in such a metric space all @xmath10 bigons are @xmath139-slim .
we conclude using theorem [ thinbigons ] above .
as pierre pansu pointed out to us it immediately follows that we can take @xmath4 to be linear in @xmath52 . for any @xmath425
the space @xmath5 is @xmath426 hyperbolic . scaling the metric
we see that @xmath427 has the property that the intersection of any two balls has eccentricity bounded by @xmath428 and so is @xmath429-hyperbolic .
rescaling we see that @xmath256 is @xmath426-hyperbolic .
we conclude this section with the proof of corollary 2 which asserts that @xmath0-trees are characterised by the property that the intersection of any two metric balls is a metric ball .
one implication is given by proposition [ hyp=>qb ] .
conversely , if the space @xmath5 has the property that the intersection of any two balls has eccentricity 0 then for any @xmath425 the intersection of any two balls has eccentricity bounded by @xmath52 and so the space is @xmath426 hyperbolic for all @xmath425 .
it is therefore @xmath382-hyperbolic and hence must be an @xmath2-tree .
_ remark : notice that in fact we do not need to assume that the intersection of any two balls is a ball to carry out the proof , only that the eccentricity of such an intersection is @xmath382 .
a priori this is a weaker condition , however in an @xmath2-tree the intersection of two balls is always a ball and therefore , as a consequence of the theorem , the two conditions are equivalent . _
in theorem [ qb ] and corollary [ rtreeschar ] , the assumption that the metric space @xmath5 be geodesic might not be needed .
the notion of hyperbolic spaces extends to non - geodesic metric spaces via the gromov product ( see e.g. dfinition 3 , page 27 of @xcite ) and it would be interesting to find an appropriate generalisation of these results to that context .
in particular recall that a @xmath4-ultrametric space is a metric space @xmath5 which satisfies the following strengthened version of the triangle inequality , @xmath430 for all @xmath431 in @xmath36 .
it is easy to see that at least two of @xmath432 , @xmath433 and @xmath434 differ by at most @xmath4 , meaning that any triangle is almost isoceles .
these are examples of @xmath374-hyperbolic spaces in the sense of gromov , see @xcite section 1.2 on page 90 .
it would be interesting to know if those spaces do satisfy the property that any intersection of two balls is almost a ball ( in some sense ) .
we say that a geodesic metric space @xmath256 has the geodesic extension property if any geodesic arc @xmath435\rightarrow x$ ] extends to a geodesic @xmath436 , i.e. , @xmath437}=\gamma$ ] .
the most important feature of a space with the geodesic extension property is that centres and radii of balls are well defined ( this is easily checked ) .
more precisely we will use the following .
[ outeradius]let @xmath5 be a space with the geodesic extension property . for any @xmath340 and @xmath342 , with @xmath438 .
then @xmath439 is the biggest ball that fits in @xmath358 , where @xmath346 and @xmath347 is a point on any geodesic between @xmath43 and @xmath44 , at distance @xmath348 from @xmath43 .
let @xmath399 . if @xmath440 then @xmath441 and there is nothing to prove .
so let us assume that @xmath442 .
take @xmath186 and @xmath443 such that @xmath444 .
let @xmath445 and @xmath446 , so that @xmath447 . since @xmath448 , we deduce that @xmath449 , and similarly , since @xmath450 , we deduce that @xmath451 ( here we use the geodesic extension property ) .
combining those two inequalities shows that @xmath452 if @xmath453 is the symmetric space of constant curvature @xmath454 there is a single triangle @xmath43 , @xmath44 , @xmath365 ( up to isometry ) with side lengths @xmath391 , @xmath392 and @xmath375 , so we define @xmath455 where @xmath347 is the point on the geodesic between @xmath43 and @xmath44 at distance @xmath456 from @xmath43 .
( this point exists because the triangle inequality ensures that @xmath457 so the distance is positive , and because @xmath458 implies that @xmath459 . )
we shall see that there is an analogue to theorem [ qb ] which characterises cat(@xmath1 ) geometry for @xmath454 [ cat0 ] given @xmath454 , a geodesic metric space @xmath5 with geodesic extension property is @xmath460 if and only if the eccentricity of the intersection of any two balls of respective radii @xmath17 and @xmath47 and at distance @xmath357 is bounded by @xmath461 .
one implication is clear so we suppose by contradiction that the eccentricity of the intersection of any two balls of respective radii @xmath17 and @xmath47 and at distance @xmath357 is bounded by @xmath461 but that @xmath36 is not cat(@xmath1 ) .
then there is a geodesic triangle @xmath462 in @xmath36 and a point @xmath463 on a geodesic @xmath11 from @xmath43 to @xmath44 such that @xmath464 .
let @xmath465 where @xmath466 denotes the geodesic @xmath467 in the comparison triangle .
similarly let @xmath468 .
in other words , @xmath251 and @xmath469 are the nearest points left and right of @xmath49 that satisfy the cat(@xmath1 ) inequality .
so for any @xmath470 in the open interval @xmath471 we have @xmath472 .
notice that @xmath473 since the metric varies continuously with points .
set @xmath474 , and @xmath475 .
let @xmath476 . now consider a geodesic triangle @xmath477 where the geodesic from @xmath478 to @xmath479 is taken to be the restriction of @xmath11 to the closed interval @xmath480 $ ] .
clearly the geodesic triangle @xmath481 is a comparison triangle in @xmath482 .
take the point @xmath483 on the above geodesic from @xmath478 to @xmath479 at distance @xmath484 from @xmath478 .
( this point exists because of the triangle inequality . ) first notice that @xmath483 has to be equal to either @xmath478 or @xmath479 , otherwise we would have @xmath485 and the intersection of @xmath486 with @xmath487 exceeds the allowed eccentricity .
assume that @xmath488 ( if @xmath489 the argument is similar and we omit it ) .
then @xmath490 so @xmath491 .
it follows that the path given by concatenating the geodesics from @xmath365 to @xmath478 and from @xmath478 to @xmath479 is itself a geodesic .
now let @xmath492 be the midpoint of the geodesic from @xmath478 to @xmath479 .
this point is at distance @xmath493 from @xmath365 . inspecting the ( degenerate ) comparison triangle
we see that @xmath494 is also at distance @xmath493 from @xmath495 but this contradicts our assumption that every point between @xmath478 and @xmath479 is further from @xmath365 than we see in the comparison triangle . _
remark : following the arguments in the proof of theorem 1 we note that there is an alternative notion of a `` based '' eccentricity function which measures eccentricity from the defined centre @xmath347 for both the inscribed and circumscribed balls . in these terms
it is easy to see that gromov hyperbolicity is equivalent to the existence of a uniform bound on the based eccentricity function , while the proof of theorem [ cat0 ] shows that cat(@xmath1 ) geometry is characterised by bounding the based eccentricity function in terms of the function @xmath496 .
hence both notions of non - positive curvature may be naturally expressed in terms of eccentricity bounds . _
in @xcite papasoglu showed that for a graph hyperbolicity was equivalent to a bound on the thinness of geodesic bigons .
as remarked before the same statement is not true for general geodesic metric spaces ( any non - hyperbolic cat(0 ) space furnishes a counter example since uniqueness fo geodesics gives a bound of @xmath382 on the fatness of geodesic bigons . ) the point is that the bound on the fatness of geodesic bigons in a graph gives an automatic bound on the fatness of @xmath497 quasi - geodesic bigons and papasoglu remarks that there is a natural generalisation of the result as follows .
the theorem appears in the masters dissertation of pomroy @xcite but to the best of our knowledge no proof exists in the literature .
we offer a proof of the result in order to place it on the record .
we argue by contradiction , similarly to corollary [ thinbigons ] and using those particular divergence functions . according to theorem [ thinbigonsquant ] for any @xmath500 , the functions @xmath138 as defined above never tend to infinity , and hence there is @xmath501 so that @xmath502 .
this means that for any @xmath503 , there are two geodesics @xmath11 and @xmath12 and @xmath120 such that let us fix @xmath56 so that @xmath506 and again let @xmath507 and @xmath508 be points on a geodesic from @xmath509 to @xmath510 and at distance less than @xmath499 from each other ( we can choose @xmath511 ) . for @xmath512 , we see a @xmath499-bigon as follows : one side is a geodesic from @xmath33 to @xmath513 , and the other is a broken geodesic @xmath514 .
so our assumptions say that it is @xmath52-thin , and hence @xmath515 .
now let @xmath516 points on a geodesic from @xmath128 to @xmath129 and at distance less than @xmath499 from each other ( we can choose @xmath517 ) .
again we construct a @xmath499-bigon as follows : one side is a geodesic from @xmath33 to @xmath518 , and the other is a broken geodesic @xmath519 .
moreover , since @xmath520 , it means that @xmath521 , and hence @xmath522 , so that @xmath523 .
this means that the geodesics @xmath11 and @xmath12 fulfill the assumptions of theorem [ meat ] for any @xmath162 big enough ( i.e. , @xmath162 so that @xmath524 , and hence , taking @xmath162 big enough ( i.e. , so that @xmath525 ) , theorem [ meat ] contradicts our assumption that @xmath499-bigons are uniformly @xmath52-thin .
_ remark : this may seem close to lemma 7.2 in @xcite which states that if @xmath526-quasigeodesics stay uniformly close to any geodesic between the endpoints , then the space is hyperbolic .
however the proof of papasoglu s or pomroy s result is considerably more elaborate . _
bridson and a. haefliger .
_ metric spaces of non - positive curvature .
_ grundlehren der mathematischen wissenschaften , * 319*. springer - verlag , berlin , 1999 .
m. coornaert , t. delzant and a. papadopoulos .
_ gomtrie et thorie des groupes .
_ les groupes hyperboliques de gromov .
lecture notes in mathematics , * 1441*. springer - verlag , berlin , 1990 .
e. ghys , p. de la harpe ed .
_ sur les groupes hyperboliques daprs mikhael gromov .
_ papers from the swiss seminar on hyperbolic groups held in bern , 1988 .
progress in mathematics * 83 * , birkhuser .
m. gromov .
_ hyperbolic groups . _ s. gersten , ed .
, msri publications , * 4 * ( 1987 ) . h. masur and y. minsky .
_ geometry of the complex of curves .
i. hyperbolicity . _
invent . math . * 138 * ( 1999 ) , no .
1 , 103149 .
p. papasoglu , _ strongly geodesically automatic groups are hyperbolic _ , invent .
( 1995 ) * 2 * pp 323334 .
j. pomroy , master thesis , warvick .
et al _ notes on word hyperbolic groups in _ group theory from a geometrical viewpoint _ ( e. ghys , a. haefliger , a. verjovsky , ed ) proc .
ictp trieste 1990 , worldscientific , singapore , 1991 , 364 .
s. wenger .
_ on gromov hyperbolicity and a characterization of real trees _ , preprint | we show that a geodesic metric space is hyperbolic in the sense of gromov if and only if intersections of balls have bounded eccentricity .
in particular , @xmath0-trees are characterized among geodesic metric spaces by the property that the intersection of any two balls is always a ball .
both gromov hyperbolicity and cat(@xmath1 ) geometry can be characterised in terms of the geometry of the intersection of balls . |
coding for cooperative wireless relay networks has attracted considerable attention recently . distributed space time coding was proposed as a coding strategy to achieve full cooperative diversity in @xcite assuming that the signals from all the relay nodes arrive at the destination at the same time . but this assumption is not close to practicality since the relay nodes are geographically distributed . in @xcite , a transmission scheme based on orthogonal frequency division multiplexing ( ofdm ) at the relay nodes
was proposed to combat the timing errors at the relays and a high rate space time code ( stc ) construction was also provided . however , the maximum likelihood ( ml ) decoding complexity for this scheme is prohibitively high especially for the case of large number of relays .
several other works in the literature propose methods to combat the timing offsets but most of them are based on decode and forward at the relay node and moreover fail to address the decoding complexity issue . in @xcite , a simple transmission scheme to combat timing errors at the relay nodes was proposed .
this scheme is particularly interesting because of its associated low ml decoding complexity . in this scheme
, ofdm is implemented at the source node and time reversal / conjugation is performed at the relay nodes on the received ofdm symbols from the source node .
the received signals at the destination after ofdm demodulation are shown to have the alamouti code structure and hence single symbol maximum likelihood ( ml ) decoding can be performed .
however , the alamouti code is applicable only for the case of two relay nodes and for larger number of relays , the authors of @xcite propose to cluster the relay nodes and employ alamouti code in each cluster .
but this clustering technique provides diversity order of only two and fails to exploit the full cooperative diversity equal to the number of relay nodes .
the main contributions of this report are as follows .
* the li - xia transmission scheme is extended to a more general transmission scheme that can achieve full asynchronous cooperative diversity for any number of relays . *
the conditions on the stc structure that admit its application in the proposed transmission scheme are identified . the recently proposed full diversity four
group decodable distributed stcs in @xcite for synchronous wireless relay networks are found to satisfy the required conditions for application in the proposed transmission scheme . *
it is shown how differential encoding at the source node can be combined with the proposed transmission scheme to arrive at a transmission scheme that can achieve full asynchronous cooperative diversity in the absence of channel knowledge and in the absence of knowledge of the timing errors of the relay nodes . moreover ,
an existing class of four group decodable distributed differential stcs @xcite for synchronous relay networks with power of two number of relays is shown to be applicable in this setting as well . in section [ sec2 ] , the basic assumptions on the relay network model
are given and the li - xia transmission scheme is briefly described .
section [ sec3 ] describes the transmission scheme proposed in this report and also provides four group decodable codes for any number of relays .
section [ sec4 ] briefly explains how differential encoding at the source node can be combined with the proposed transmission scheme and four group decodable distributed differential stcs applicable in this scenario are also proposed .
simulation results and discussion on further work comprise sections [ sec5 ] and [ sec6 ] respectively .
+ + * notation : * vectors and matrices are denoted by lowercase and uppercase bold letters respectively .
@xmath0 denotes an @xmath1 identity matrix and @xmath2 denotes an all zero matrix of appropriate size . for a set @xmath3
, the cardinality of @xmath3 is denoted by @xmath4 .
a null set is denoted by @xmath5 . for a matrix , @xmath6 , @xmath7 and
@xmath8 denote transposition , conjugation and conjugate transpose operations respectively . for a complex number , @xmath9 and @xmath10 denote its in - phase and quadrature - phase parts respectively .
in this section , the basic relay network model assumptions are given and the li - xia transmission scheme in @xcite is briefly described .
the transmission scheme in @xcite is based on the use of ofdm at the source node and the alamouti code implemented in a distributed fashion for a @xmath11 relay system .
essentially , the transmission scheme in @xcite is applicable mainly for the case of @xmath11 relays but by forming clusters of two relay nodes , it can be extended to more number of relays at the cost of sacrificing diversity benefits .
consider a network with one source node , one destination node and @xmath12 relay nodes @xmath13 .
this is depicted in fig .
[ fig_network ] .
every node is assumed to have only a single antenna and is half duplex constrained .
the channel gain between the source and the @xmath14-th relay @xmath15 and that between the @xmath16-th relay and the destination @xmath17 are assumed to be quasi - static , flat fading and modeled by independent and complex gaussian distributed with mean zero and unit variance .
the transmission of information from the source node to the destination node takes place in two phases . in the first phase
, the source broadcasts the information to the relay nodes using ofdm .
the relay nodes receive the faded and noise corrupted ofdm symbols , process them and transmit them to the destination . the relay nodes are assumed to have perfect carrier synchronization .
the overall relative timing error of the signals arrived at the destination node from the @xmath14-th relay node is denoted by @xmath18 . without loss of generality , it is assumed that @xmath19 , @xmath20 .
the destination node is assumed to have the knowledge of all the channel fading gains @xmath21 and the relative timing errors @xmath22 .
the source takes @xmath23 complex symbols @xmath24 and forms two blocks of data denoted by @xmath25^t , j=1,2 $ ] .
the first block @xmath26 is modulated by @xmath27-point inverse discrete fourier transform ( idft ) and @xmath28 is modulated by @xmath27-point discrete fourier transform ( dft ) .
then a cyclic prefix ( cp ) of length @xmath29 is added to each block , where @xmath29 is not less than the maximum of the overall relative timing errors of the signals arrived at the destination node from the relay nodes .
the resulting two ofdm symbols denoted by @xmath30 and @xmath31 consisting of @xmath32 complex numbers are broadcasted to the two relays using a fraction @xmath33 of the total average @xmath34 consumed by the source and the relay nodes together . if the channel fade gains are assumed to be constant for @xmath35 ofdm symbol intervals , the received signals at the @xmath14-th relay during the @xmath16-th ofdm symbol duration is given by @xmath36 where , @xmath37 is the additive white gaussian noise at the @xmath14-th relay node during the @xmath16 the ofdm symbol duration .
the two relay nodes then process and transmit the resulting signals as shown in table [ table_alamouti ] using a fraction @xmath38 of the total power @xmath34 .
the notation @xmath39 denotes the time reversal operation , i.e. , @xmath40 .
.alamouti code based transmission scheme [ cols="^,^,^",options="header " , ] this code is @xmath41 real symbol decodable and achieves full diversity for appropriately signal sets @xcite .
example [ eg_5relay ] illustrates how the proposed transmission scheme can be extended to odd number of relays as well .
in this section , it is shown how differential encoding can be combined with the proposed transmission scheme described in section [ sec3 ] and then the codes in @xcite are proposed for application in this setting . for the proposed transmission scheme in section [ sec3 ] , at the end of one transmission frame , we have in the @xmath42-th sub carrier @xmath43 .
note that the channel matrix @xmath44 depends on @xmath45 .
thus the destination node needs to have the knowledge of these values in order to perform ml decoding .
now using differential encoding ideas which were proposed in @xcite for non - coherent communication in synchronous relay networks , we combine them with the proposed asynchronous transmission scheme . supposing the channel remains approximately constant for two transmission frames , then differential encoding can be done at the source node in each sub carrier @xmath46 as follows : @xmath47^t,~ \mathbf{s_k^t}=\frac{1}{a_t-1}\mathbf{c_t}\mathbf{s_k^{t-1 } } , \mathbf{c_t}\in\mathscr{c}\ ] ] where , @xmath48 denotes the vector of complex symbols transmitted by the source during the @xmath14-th transmission frame in the @xmath42-th sub carrier and @xmath49 is the codebook used by the source which consists of scaled unitary matrices such that @xmath50=1 $ ] .
if for all @xmath51 , @xmath52 then we have : @xmath53 from which @xmath54 can be decoded as in each sub carrier @xmath46 . note that this decoder does not require the knowledge of @xmath55 at the destination .
it turns out that the four group decodable distributed differential space time codes constructed in @xcite for synchronous relay networks with power of two number of relays meet all the requirements for use in the proposed transmission scheme as well .
the following example illustrates this fact .
let @xmath56 .
the codebook at the source is given by + where @xmath57 , @xmath58 , @xmath59 , @xmath60 and .
differential encoding is done at the source node for each sub carrier @xmath46 as follows : @xmath47^t,~ \mathbf{s_k^t}=\frac{1}{a_t-1}\mathbf{c_t}\mathbf{s_k^{t-1 } } , \mathbf{c_t}\in\mathscr{c}.\ ] ] once we get @xmath61 from the above equation , the @xmath27 length vectors @xmath62 can be obtained .
then idft / dft is applied on these vectors as shown below and broadcasted to the relay nodes .
@xmath63 , @xmath64 , @xmath65 and @xmath66 . the relay nodes process the received ofdm symbols as given in table [ table_4relay ] for which @xmath67 , @xmath68 , , @xmath69 $ ] and .
it has been proved in @xcite that @xmath70 and for all @xmath51 . at the destination node ,
decoding for @xmath57 , @xmath58 , @xmath59 and @xmath71 can be done separately in every sub carrier due to the four group decodable structure of @xmath49 . relay system with and without channel knowledge , width=5 ]
in this section , we study the error performance of the proposed codes using simulations .
we take @xmath56 , @xmath72 and the length of cp as @xmath73 .
the delay @xmath18 at each relay is chosen randomly between @xmath74 to @xmath75 with uniform distribution .
two cases are considered for simulation : ( 1 ) with channel knowledge at the destination and ( 2 ) without channel knowledge at the destination . for the case of no channel information ,
differential encoding at the source as described in section [ sec4 ] is done using the distributed differential space time in @xcite .
when channel knowledge is available at the destination , rotated qpsk is used as the signal set @xcite .
the transmission rate for the both the schemes is @xmath76 bit per channel use ( bpcu ) if the rate loss due to cp is neglected .
the error performance curves for both the cases is shown in fig .
[ fig_simulation ] .
it can be observed from fig .
[ fig_simulation ] that the error performance of the no channel knowledge case performs approximately @xmath77 db worser than that with channel knowledge at the destination .
this is due to the differential transmission / reception technique in part and also in part because of the change in signal set from rotated qpsk to some other signal set @xcite in order to comply with the requirement of scaled unitary codeword matrices .
the change in signal set for the sake of scaled unitary codeword matrices results in a reduction of the coding gain .
a general transmission scheme for arbitrary number of relays that can achieve full cooperative diversity in the presence of timing errors at the relay nodes was proposed .
it was then pointed out that the four group decodable distributed space time codes in @xcite can be applied in the proposed transmission scheme for any number of relay nodes .
finally it was shown how the proposed scheme can be combined with differential encoding at the source node to end up with a transmission scheme that is robust to timing errors and also does not require the knowledge of the channel fading gains as well as the timing errors at any of the nodes . for this differential scheme ,
it was pointed out that the four group decodable distributed differential space time codes in @xcite are applicable for power of two number of relays .
a drawback of the proposed transmission scheme is that it requires a large coherence interval spanning over multiple ofdm symbol durations .
moreover there is a rate loss due to the use of cp , but this loss can be made negligible by choosing a large enough @xmath27 . some of the interesting directions for further work are listed below : 1 . constructing single symbol decodable distributed space time codes for the proposed transmission scheme .
2 . the codes in @xcite are applicable only for power of two number of relay nodes . constructing four
group decodable distributed differential space time codes for all even number of relay nodes that are applicable in asynchronous relay networks without channel knowledge is an important direction for further work .
3 . in this work
, we have assumed that there are no frequency offsets at the relay nodes .
extending this work to asynchronous relay networks with frequency offsets is an interesting direction for further work .
this problem has been addressed in @xcite for the case of two relay nodes .
the authors sincerely thank prof .
xiang gen xia and prof .
hamid jafarkhani for sending us preprints of their recent works @xcite .
g. susinder rajan and b. sundar rajan , `` algebraic distributed space - time codes with low ml decoding complexity , '' proceedings of _ ieee international symposium on information theory _ , nice , france , june 24 - 29 , 2007 , pp .
1516 - 1520 .
- , `` a non - orthogonal distributed space - time coded protocol , part - ii : code construction and dm - g tradeoff , '' proceedings of _ ieee information theory workshop _ , chengdu , china , october 22 - 26 , 2006 , pp.488 - 492 .
kiran t. and b. sundar rajan , `` partially - coherent distributed space - time codes with differential encoder and decoder , '' _ ieee journal on selected areas in communications _ , vol . 25 , no .
2007 , pp .
426 - 433 .
frdrique oggier , babak hassibi , " cyclic distributed space - time codes for wireless relay networks with no channel information , submitted for publication .
available online http://www.systems.caltech.edu/~frederique/submitdstcnoncoh.pdf zheng li and x .- g .
xia , `` an alamouti coded ofdm transmission for cooperative systems robust to both timing errors and frequency offsets , '' to appear in _ ieee transactions on wireless communications_. private communication . | recently li and xia have proposed a transmission scheme for wireless relay networks based on the alamouti space time code and orthogonal frequency division multiplexing to combat the effect of timing errors at the relay nodes .
this transmission scheme is amazingly simple and achieves a diversity order of two for any number of relays .
motivated by its simplicity , this scheme is extended to a more general transmission scheme that can achieve full cooperative diversity for any number of relays .
the conditions on the distributed space time code ( dstc ) structure that admit its application in the proposed transmission scheme are identified and it is pointed out that the recently proposed full diversity four group decodable dstcs from precoded co - ordinate interleaved orthogonal designs and extended clifford algebras satisfy these conditions .
it is then shown how differential encoding at the source can be combined with the proposed transmission scheme to arrive at a new transmission scheme that can achieve full cooperative diversity in asynchronous wireless relay networks with no channel information and also no timing error knowledge at the destination node .
finally , four group decodable distributed differential space time codes applicable in this new transmission scheme for power of two number of relays are also provided . |
[ cols="^,^,^ " , ]
next we study polarization properties of 3d models . motivated by 3d simulations of neutrino - driven explosions , where various sizes of complex structure appear , we set up 3d models by randomly placing different numbers of spherical clumps with different sizes .
here we introduce two parameters to depict the model : the size parameter of the clumps @xmath3 , i.e. , the radius of the clump normalized by the photospheric radius ( @xmath4 ) and the photospheric covering factor ( @xmath5 ) . since the optical depth near
the photosphere is the most important for line formation , the covering factor is evaluated by taking into account the clumps only in a shell between @xmath6 and @xmath7 km s@xmath8 .
note that , as in the 2d cases , the line optical depth in our models has a spherical component and it is enhanced by a factor of @xmath9 within the clumps .
the top panels of figure [ fig:3d ] show the polarization properties of the 3d model with the clump size of @xmath10 and the covering factor of @xmath11 . in the polarization spectrum , both stokes @xmath12 and @xmath13 parameters vary across the lines ( middle panel ) , and polarization shows a loop in the @xmath0 diagram ( right panel ) , as also found by @xcite . the @xmath0 loop in the 3d clumpy models can be understood as follows . in the 3d clumpy models , depending on the doppler velocities ,
different parts of the photospheric disk are hidden by the clumps .
since the distribution of the clumps does not have a common symmetric axis , the position angle of the polarization can change depending on the doppler velocities .
in general , the change in the position angle across the line can be arbitrary large , i.e. , the polarization in the @xmath0 diagram can be scattered around .
but for the relatively large size of the clumps as in the case of @xmath10 , the same clump keeps contributing to the absorption even for different doppler velocities , and thus the change in the position angle tends to be smooth as a function of doppler velocities .
therefore , the polarization tends to show a loop in the @xmath0 diagram in the 3d clumpy distribution with relatively large clumps .
note that the @xmath0 loop can also be produced by other geometries , e.g. , a combination of the ellipsoidal photosphere and ellipsoidal line scattering shell whose symmetric axes are misaligned with each other @xcite , but even in such a case , it is required that the axisymmetry of the system is broken . ] ( top , red ) , @xmath14 ( middle , blue ) , and @xmath15 ( bottom , green ) . for each panel , four different colors ( lighter to darker colors from left to right ) represent models with four different line strengths at the photosphere ( @xmath16 = 3.0 , 10.0 , 30.0 , and 100.0 , respectively ) . for each model , results of 100 lines of sight are shown .
the black points with error bars are observational data of the ( filled ) and ( open ) lines for 6 type ib and ic sne analyzed in @xcite : type ib sne 2005bf @xcite , 2008d @xcite , 2009jf @xcite , type ic sne 2002ap @xcite , 2007gr @xcite , and 2009mi @xcite .
the solid line shows @xmath17 $ ] ( see section [ sec : methods ] ) .
[ fig : pobs_fd],title="fig : " ] + ( top , red ) , @xmath14 ( middle , blue ) , and @xmath15 ( bottom , green ) . for each panel , four different colors ( lighter to darker colors from left to right ) represent models with four different line strengths at the photosphere ( @xmath16 = 3.0 , 10.0 , 30.0 , and 100.0 , respectively ) . for each model
, results of 100 lines of sight are shown .
the black points with error bars are observational data of the ( filled ) and ( open ) lines for 6 type ib and ic sne analyzed in @xcite : type ib sne 2005bf @xcite , 2008d @xcite , 2009jf @xcite , type ic sne 2002ap @xcite , 2007gr @xcite , and 2009mi @xcite .
the solid line shows @xmath17 $ ] ( see section [ sec : methods ] ) .
[ fig : pobs_fd],title="fig : " ] + ( top , red ) , @xmath14 ( middle , blue ) , and @xmath15 ( bottom , green ) . for each panel , four different colors ( lighter to darker colors from left to right ) represent models with four different line strengths at the photosphere ( @xmath16 = 3.0 , 10.0 , 30.0 , and 100.0 , respectively ) . for each model ,
results of 100 lines of sight are shown .
the black points with error bars are observational data of the ( filled ) and ( open ) lines for 6 type ib and ic sne analyzed in @xcite : type ib sne 2005bf @xcite , 2008d @xcite , 2009jf @xcite , type ic sne 2002ap @xcite , 2007gr @xcite , and 2009mi @xcite .
the solid line shows @xmath17 $ ] ( see section [ sec : methods ] ) .
[ fig : pobs_fd],title="fig : " ] .
one characteristic polarization degree is assigned for each object by taking the average of the corrected polarization ( @xmath18 ) of the and lines .
color lines show the cumulative distribution of polarization degree for 100 lines of sight .
three models with @xmath10 ( red ) , @xmath14 ( blue ) , and @xmath15 ( green ) are shown . in this plot
, we use the models with @xmath19 since these models approximately give @xmath20 ( figure [ fig : pobs_fd ] ) , where the corrected polarization is defined .
[ fig : pdist_size ] ] the size of the clumps is of interest to study the origin of the 3d structure in the sne .
we show the first attempt to quantify the size of the clumps by comparing the results of the modelling and the observed polarization degrees , i.e. , the maximum polarization level at the absorption line .
we calculate the polarization spectra with different sizes of the clumps by keeping the covering factor of @xmath21 and other parameters to be the same .
the middle and bottom panels in figure [ fig:3d ] show the results for the 3d models with @xmath14 and @xmath22 , respectively . as shown in the figures , for a given covering factor
, models with smaller clumps show a lower polarization . in such models ,
the photospheric disk is hidden by many small clumps , and polarization vectors tend to be cancelled out ( figure [ fig : schematic ] ) .
this behavior was also pointed out by @xcite in the context of type ia sne .
since stripped - envelope sne generally show non - zero line polarization , the typical size of the clumps should not be too small .
since the polarization degree depends not only on the geometry but also on the absorption strength , it is important to compare the models and observations for similar absorption strengths .
therefore , in figure [ fig : pobs_fd ] , we compare models and observations in the plane of the polarization degree and the fractional absorption depth .
the black points with error bars are observational data of the ( filled ) and ( open ) lines for 6 type ib and ic sne analyzed in @xcite .
the small dots show the polarization degree of the models for 100 lines of sight . in each panel , we show four sets of the models with the same size and distribution of the clumps but with the different line optical depth at the photosphere ( @xmath16 = 3.0 , 10.0 , 30.0 , and 100.0 from left to right ) .
when the clump is as small as @xmath15 ( bottom panels of figures [ fig:3d ] and [ fig : pobs_fd ] ) , the polarization degree can not be @xmath23 for any line of sight .
for the larger sizes of the clumps , a higher polarization can be obtained .
when the size of the clumps is @xmath24 ( middle panels ) , the polarization degrees of these models are still short of some of the observed polarization .
when the size of the clumps is relatively large , @xmath10 ( top panels ) , the polarization degree can be as high as @xmath25 for the fractional depth of 0.5 .
ideally the polarization properties of the models should be compared with the statistical distribution of the observed polarization .
although the number of objects with good data is still small , figure [ fig : pdist_size ] shows a cumulative distribution of polarization properties of 6 type ib and ic sne . to define one characteristic polarization for each object , we take the average of the corrected polarization ( @xmath18 ) for the and lines .
color lines show the cumulative distribution of the modelled polarization for 100 lines of sight .
we choose models with @xmath19 , which approximately give @xmath20 ( figure [ fig : pobs_fd ] ) .
the comparison in the cumulative distribution clearly shows that the model with too small clumps ( @xmath15 ) is not consistent with the observations . the @xmath26 value for a kolmogorov - smirnov ( ks ) test is @xmath27 .
since the number of objects is so small , we can not distinguish the model with the clump size of @xmath14 ( @xmath28 ) and @xmath10 ( @xmath29 ) .
nevertheless , the model with @xmath14 is already short to explain the polarization level of @xmath30 , and seems to close to the lower limit for the clump size . here
it is noted that our models adopt an enhancement factor of @xmath31 .
for a higher enhancement factor , the polarization degree is not largely affected because models with @xmath31 already give an optically thick absorption in the clumps near the photosphere . on the other hand , for a smaller enhancement factor
, the polarization degree decreases for a given fd . in such cases , even larger clumps
is required to reproduce a high polarization degree .
therefore , we conclude that a typical size of the 3d clumps should be @xmath32 of the photospheric radius to reproduce observed polarization degrees . to obtain possible constraints on the number or the covering factor of the clumps in the ejecta , we vary the covering factors of clumps keeping their size to be @xmath10 .
figure [ fig : pdist_fc ] shows the model input ( left ) and cumulative distributions of the resultant polarization ( right ) . for the models , we choose the line strength at the photosphere to have @xmath20 , i.e. , @xmath16 = 30 for the models with @xmath33 , 0.2 , and 0.3 .
@xmath16 = 10 for the models with @xmath34 and 0.7 .
the observed distribution is the same as in figure [ fig : pdist_size ] . for the model with a smaller covering factor ( @xmath33 ) ,
the probability to have a high polarization is also low .
then , by increasing the covering factor of the clumps , a higher polarization can be more frequently observed ( @xmath35 ) . however , if the covering factor of the clumps is too large ( @xmath36 ) , the distribution of resultant polarization shifts toward a lower value again since the system restores the symmetry again .
since the observational samples are small , it is difficult to draw a firm conclusion on the covering factor of the clumps .
however , the models with @xmath33 and @xmath36 are already at the edge of the distribution . by taking into account the fact that models with @xmath31 tend to give an upper limit of the polarization level ( see section [ sec:3d_size ] )
, it seems that current data do not support models with too small covering factors ( @xmath37 ) and too large covering factors ( @xmath38 ) .
we have modelled line polarization of stripped - envelope core - collapse sne .
the results of modelling are summarized as follows .
( 1 ) the observed @xmath0 loop can not be explained by the 2d axisymmetric models , but can be explained by the 3d clumpy models .
( 2 ) by comparing the results of the 3d clumpy models with the observed degrees of line polarization , it is found that a typical size of the clumps is relatively large , i.e. , @xmath32 of the radius , and a covering factor of the clumps in the ejecta is not too small and not too large ( @xmath39 ) .
it is intriguing that such a large - scale clumpy structure is also seen in the element distribution of cassiopeia a ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) , which is a supernova remnant produced by a type iib sn @xcite .
the similarity suggests that the element distribution as seen in cassiopeia a may also be able to reproduce the polarization properties observed in early phase of sne .
here we discuss possible origins for the clumpy structure suggested by observations and modelling .
one scenario is the rayleigh - taylor ( rt ) instability , causing matter mixing in the sn ejecta . by the rt instability ,
many clumps are produced and metal - rich ejecta inside are delivered toward the outer layers ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
however , the rt instability alone usually produces small fingers in many directions .
this is similar to the case of @xmath15 in figure [ fig : schematic ] and not consistent with the observations .
the clumpy structure suggested by observations is more in favor of large - scale convection or sasi developed in the initial stage of the explosion .
when the large - scale convection or sasi takes place , the subsequent evolution of the shock becomes asymmetric , which produces the large - scale asymmetry in the element distribution ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
also , results of long - term simulations show that the ejecta structure near the shock breakout still keeps an imprint of the large - scale asymmetry generated by neutrino - driven convection and sasi , with the small - scale structures by the rt instability added on top of it @xcite .
note that such long - term simulations for neutrino - driven explosion also nicely reproduce the geometry of cassiopeia a @xcite .
it is worth noting that , although the loop in the @xmath0 diagram does not support a purely axisymmetric element distribution ( figure [ fig:2d ] ) , spectropolarimetric data do not rule out the presence of an overall bipolar structure or a dominant axis in the sn ejecta . as long as some large - scale ,
non - axisymmetric components exist , they can produce a large enough polarization level and the loop in the @xmath0 diagram .
in fact , analysis of the [ ] line profiles in the late - phase spectra suggest a torus - like distribution of oxygen , which is consistent with a bipolar explosion ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) .
since polarization at the early phase and nebular line profile are sensitive to the outer and inner ejecta , respectively , the combination of early and late phase observations may indicate that global 2d structure exists more in the inner ejecta and 3d clumpy structure is added in the outer ejecta .
it is noted that , even by the late phase observations , presence of clumpy structure has also been suggested by the studies of line profiles ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) , ionization states ( e.g. , * ? ? ?
* ; * ? ? ?
* ) , and dust ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
thus , the transition from the inner 2d to the outer 3d structures may be somewhat gradual .
interestingly , our studies suggest that the shape of the loop in the @xmath0 diagram can be used as a probe of such a combined ( 2d @xmath40 3d ) geometry .
as expected from the results of 2d ( figure [ fig:2d ] ) and 3d models ( figure [ fig:3d ] ) , if the ejecta has an overall 2d geometry + 3d clumpy structure , it tends to produce an elongated loop in the @xmath0 diagram .
although current observational data do not allow us to extract such information , detailed studies will be possible in future with more observational samples with high signal - to - noise ratio .
it is emphasized that our modelling includes many simplifications .
for example , we parameterize the line optical depth and enhancement factor , but they must be determined by the combination of element abundance , temperature , and ionization states .
thus , our models shown in the left panels of figures [ fig:2d ] and [ fig:3d ] are not readily connected with the element distribution .
full radiation transfer modelling using 3d hydrodynamic models is required to obtain a closer link between the explosion models and observations .
also , the comparison with observed polarization degree is done by averaging the polarization degrees of different lines .
since polarization at different absorption lines reflects the distribution of each element and ion , the direct comparison for each element is necessary when larger observed samples and full transfer calculations are available .
we have performed 3d radiation transfer simulations for the analysis of line polarization in stripped - envelope sne .
we demonstrate that a purely axisymmetric , 2d structure always produces a straight line in the stokes @xmath0 diagram , and can not explain the commonly observed loop in the @xmath0 diagram . on the contrary
, 3d clumpy structures naturally reproduce the loop .
comparison of the results of the modelling and the observed polarization degrees enables to constrain a typical size of the clumps from polarization data for the first time . to reproduce the distribution of the observed polarization degrees ( 0.5 - 2.0 % ) ,
a typical size of the clump should be relatively large , i.e. , @xmath41 of the photospheric radius ( or the radius where the clump is located ) .
the covering factor of the clump in the ejecta is only weakly constrained i.e. , to @xmath42 % .
such a large - scale clumpy structure inferred by polarization is similar to that seen in the sn remnant cassiopeia a. the large - scale clumpy structure is unlikely to be produced only by the rt instability as it tends to produce small fingers in many directions .
instead , the presence of the large - scale clumpy structure in the ejecta suggests that large - scale convection or sasi takes place at the onset of the explosion .
polarization properties do not necessarily exclude the presence of a dominant axis in the sn ejecta since non - axisymmetric structure on top of the 2d axisymmetric structure can also reproduce the loop in the @xmath0 diagram .
in fact , the analysis of the nebular spectra supports a bipolar geometry in the innermost layer .
these observational constraints suggest that sn ejecta may have an overall 2d bipolar structure inside and 3d clumpy structure outside .
we speculate that such a hybrid structure could be produced by sasi . in order to obtain further constraints on the explosion mechanism ,
polarization modelling using realistic sn models will be worthwhile as more and more long - term realistic simulations from core collapse to the shock breakout are becoming available .
we thank takashi hattori , kentaro aoki , masanori iye , elena pian , toshiyuki sasaki , and masayuki yamanaka for their contribution to the spectropolarimetric observations with the subaru telescope , and the referee for valuable comments .
mt thanks thomas janka , takashi moriya , and takaya nozawa for fruitful discussion .
numerical simulations presented in this paper were carried out with cray xc30 at center for computational astrophysics , national astronomical observatory of japan .
this research has been partly supported by the grant - in - aid for scientific research from jsps ( 24740117 , 26800100 , 15h02075 ) and mext ( 25103515 , 15h00788 ) , and by world premier international research center initiative ( wpi initiative ) , mext , japan .
135 natexlab#1#1 , a. , & barlow , m. j. 2016 , , 456 , 1269 , j. m. , mezzacappa , a. , & demarino , c. 2003 , , 584 , 971 , j. c. , & mclean , i. s. 1977 , , 57 , 141 , s. w. , et al .
2013 , , 767 , l6 , m. , sim , s. a. , pakmor , r. , kromer , m. , taubenberger , s. , roepke , f. k. , hillebrandt , w. , & seitenzahl , i. r. 2015 , arxiv:1510.04128 , r. , rampp , m. , janka , h .-
t . , & kifonidis , k. 2006 , , 447 , 1049 , a. 2013 , reviews of modern physics , 85 , 245 , a. , dessart , l. , livne , e. , ott , c. d. , & murphy , j. 2007 , , 664 , 416 , a. , hayes , j. , & fryxell , b. a. 1995 , , 450 , 830 , j. i. 1970 , , 149 , 111 , s. 1960 , radiative transfer ( new york : dover , 1960 ) , j. , et al .
2014 , , 790 , 120 , r. , & filippenko , a. v. 2008 , , 136 , 2227 , r. , filippenko , a. v. , li , w. , & silverman , j. m. 2010 , , 713 , 1363 , a. d. , & whitney , b. a. 1995 , , 441 , 400 , s. m. , & oconnor , e. p. 2014
, , 785 , 123 , j. y. 1980 , , 86 , 198 , t. , et al .
2010 , , 725 , 2038 , l. , & hillier , d. j. 2011 , , 415 , 3497 , e. , & arendt , r. g. 2015 , , 810 , 75 , a. , danziger , i. j. , cappellaro , e. , della valle , m. , gouiffes , c. , phillips , m. m. , & turatto , m. 2004 , , 426 , 963 , b. , barlow , m. j. , & sugerman , b. e. k. 2007 , , 375 , 753 , r. 2010 , , 725 , 1563 , t. , galletti , p. , scheck , l. , & janka , h .- t .
2007 , , 654 , 1006 , b. , arnett , d. , & mueller , e. 1991 , , 367 , 619 , s .- i . , kotake , k. , hashimoto , m .- a . , ono , m. , & ohnishi , n. 2011 , , 738 , 61 , b. w. , et al .
2014 , , 506 , 339 , i. , matsuda , t. , nomoto , k. , & shigeyama , t. 1990 , , 358 , l57 , n. j. , janka , h .-
, & mller , e. 2010 , , 714 , 1371 , f. , marek , a. , mller , b. , & janka , h .-
2012 , , 755 , 138 , f. , mller , b. , wongwathanarat , a. , marek , a. , & janka , h .- t .
2013 , , 770 , 66 , m. , benz , w. , hix , w. r. , fryer , c. l. , & colgate , s. a. 1994 , , 435 , 339 , m. , & woosley , s. e. 1994 , , 425 , 814 , d. j. 1991 , , 247 , 455 , j. l. , leonard , d. c. , chornock , r. , filippenko , a. v. , barth , a. j. , & matheson , t. 2008 , , 688 , 1186 , p. 1991
, , 246 , 481 , p. , wheeler , j. c. , hines , d. c. , & trammell , s. r. 1996 , , 459 , 307 , k. t. , kasen , d. , & nordsieck , k. h. 2010 , , 720 , 1500 , d. a. , hflich , p. , wang , l. , & wheeler , j. c. 2001 , , 556 , 302 , u. , et al .
2004 , , 615 , l117 , k. , rudnick , l. , delaney , t. , smith , j. d. , rho , j. , reach , w. t. , kozasa , t. , & gomez , h. 2010 , , 725 , 2059 , w. , kotake , k. , ohnishi , n. , yamada , s. , & sawada , k. 2008 , , 678 , 1207 , k. , et al .
2000 , , 534 , 660 , h .- t . 2012 , annual review of nuclear and particle science , 62 , 407 , h .- t . , & mueller , e. 1996 , , 306 , 167 , h .- t . , mller , b. , kitaura , f. s. , & buras , r. 2008 , , 485 , 199 , d. j. 1989 , , 71 , 951 , c. c. , almgren , a. , & woosley , s. e. 2010 , , 723 , 353 , d. , et al .
2003 , , 593 , 788 , d. , thomas , r. c. , & nugent , p. 2006
, , 651 , 366 , k. s. , et al .
2002 , , 580 , l39 , k. , plewa , t. , janka , h .- t . , & mller , e. 2003 , , 408 , 621 , k. , plewa , t. , scheck , l. , janka , h .-
t . , & mller , e. 2006 , , 453 , 661 , f. s. , janka , h .-
t . , & hillebrandt , w. 2006 , , 450 , 345 , r. , et al .
2009 , , 704 , 306 , k. , sawai , h. , yamada , s. , & sato , k. 2004 , , 608 , 391 , k. , takiwaki , t. , suwa , y. , iwakami nakano , w. , kawagoe , s. , masada , y. , & fujimoto , s .- i .
2012 , advances in astronomy , 2012 , 428757 , o. , birkmann , s. m. , usuda , t. , hattori , t. , goto , m. , rieke , g. h. , & misselt , k. a. 2008 , science , 320 , 1195 , m. , & sim , s. a. 2009 , , 398 , 1809 , e. j. , et al . 2015 , , 807 , l31 , d. c. , filippenko , a. v. , chornock , r. , & foley , r. j. 2002 , , 114 , 1333 , m. , mezzacappa , a. , thielemann , f .- k . , messer , o. e. , hix , w. r. , & bruenn , s. w. 2001 , , 63 , 103004 , l. b. 1999 , , 345 , 211 , k. , et al .
2008 , science , 319 , 1220 , k. , mazzali , p. a. , deng , j. , nomoto , k. , yoshii , y. , tomita , h. , & kobayashi , y. 2003 , , 593 , 931 , k. , nakamura , t. , nomoto , k. , mazzali , p. a. , patat , f. , & hachisu , i. 2002 , , 565 , 405 , a. , & janka , h .- t .
2009 , , 694 , 664 , t. , filippenko , a. v. , ho , l. c. , barth , a. j. , & leonard , d. c. 2000 , , 120 , 1499 , c. d. , & mckee , c. f. 1999 , , 510 , 379 , j. c. , et al .
2016 , arxiv:1611.07930 . 2015 , , 453 , 4467 , j. r. , wheeler , j. c. , baade , d. , patat , f. , hflich , p. , wang , l. , & clocchiatti , a. 2009 , , 705 , 1139 , j. r. , wheeler , j. c. , patat , f. , baade , d. , wang , l. , & hflich , p. 2007
, , 381 , 201 , j. r. , wheeler , j. c. , patat , f. , wang , l. , baade , d. , & hflich , p. a. 2007 , , 671 , 1944 , j. i. , et al .
2010 , , 402 , 161 , p. a. , et al .
2007 , , 661 , 892 , p. a. , iwamoto , k. , & nomoto , k. 2000 , , 545 , 407 , p. a. , et al .
2007 , , 670 , 592 . 2005 , science , 308 , 1284 , p. a. , & lucy , l. b. 1993 , , 279 , 447 , p. a. , nomoto , k. , patat , f. , & maeda , k. 2001 , , 559 , 1047 , t. , janka , h .- t .
, & marek , a. 2015 , , 801 , l24 , d. , & fesen , r. a. 2015 , science , 347 , 526 , d. , et al . 2015 , , 799 , 51 , p. a. , williams , g. g. , porter , a. , smith , p. s. , smith , n. , leising , m. d. , jannuzi , b. t. , & green , e. m. 2016 , arxiv:1611.07070 , m. , kirshner , r. p. , blondin , s. , challis , p. , & matheson , t. 2008 , , 687 , l9 , p. , et al . 2014 , , 785 , l29 , b. 2016 , pasa , 33 , e048 , b. , janka , h .-
t . , & heger , a. 2012 , , 761 , 72 , s. , hashimoto , m .- a . , sato , k. , & yamada , s. 1997 , , 486 , 1026 , s. , mizuta , a. , & sato , k. 2006 , , 647 , 1255 , s. , shimizu , t. m. , & sato , k. 1998 , , 495 , 413 , m. , cerd - durn , p. , mller , e. , & aloy , m. a. 2009 , , 498 , 241 , n. , kotake , k. , & yamada , s. 2006 , , 641 , 1018 , m. , nagataki , s. , ito , h. , lee , s .- h . , mao , j. , hashimoto , m .- a . , & tolstov , a. 2013 , , 773 , 161 , c. d. , burrows , a. , dessart , l. , & livne , e. 2008 , , 685 , 1069 , f. , baade , d. , hflich , p. , maund , j. r. , wang , l. , & wheeler , j. c. 2009 , , 508 , 229 , a. l. , et al .
2016 , arxiv:1605.03994 , m. , & janka , h .-
2000 , , 539 , l33 , l. f. , ott , c. d. , haas , r. , oconnor , e. p. , diener , p. , & schnetter , e. 2016 , arxiv:1604.07848 , r. , et al .
2013 , , 434 , 2032 , h. , kotake , k. , & yamada , s. 2005 , , 631 , 446 , l. , plewa , t. , janka , h .-
t . , kifonidis , k. , & mller , e. 2004 , physical review letters , 92 , 011103 , r. e. , nordsieck , k. h. , nook , m. a. , magalhaes , a. m. , taylor , m. , bjorkman , k. s. , & anderson , c. m. 1990 , , 365 , l19 , p. r. , & sutherland , p. g. 1982 , , 263 , 902 , i. , et al .
2013 , , 436 , 3614 , j. , leibundgut , b. , & spyromilio , j. 1998 , , 337 , 207 , j. 1994 , , 266 , l61 + , n. , de la chevrotire , a. , & moffat , a. f. j. 2012 , in american institute of physics conference series , vol . 1429 , american institute of physics conference series , ed .
j. l. hoffman , j. bjorkman , & b. whitney , 176180 , h. f. , et al .
2016 , arxiv:1606.05465 , b. e. k. , et al .
2006 , science , 313 , 196 , k. , yamada , s. , suzuki , h. , shen , h. , chiba , s. , & toki , h. 2005 , , 629 , 922 , y. , kotake , k. , takiwaki , t. , whitehouse , s. c. , liebendrfer , m. , & sato , k. 2010 , , 62 , l49 + , t. , kotake , k. , & sato , k. 2009 , , 691 , 1360 , t. , kotake , k. , & suwa , y. 2012 , , 749 , 98 . 2014 , , 786 , 83 , m. , et al .
2012 , , 754 , 63 , m. , kawabata , k. s. , maeda , k. , hattori , t. , & nomoto , k. 2008 , , 689 , 1191 , m. , et al . 2009 , , 699 , 1119 . 2010 , , 714 , 1209 . 2009 , , 700 , 1680 , s. , et al .
2009 , , 397 , 677 , t. a. , burrows , a. , & pinto , p. a. 2003 , , 592 , 434 , n. 2009 , , 690 , 526 , s. , et al .
2011 , , 416 , 3138 , l. , et al .
2003 , , 591 , 1110 , l. , baade , d. , hflich , p. , & wheeler , j. c. 2003 , , 592 , 457 , l. , & wheeler , j. c. 2008 , , 46 , 433 , r. , barlow , m. j. , matsuura , m. , & ercolano , b. 2015 , , 446 , 2089 , b. a. 2011 , bulletin of the astronomical society of india , 39 , 101 , b. a. , & hartmann , l. 1992 , , 395 , 529 , a. , janka , h .-
mueller , e. , pllumbi , e. , & wanajo , s. 2016 , arxiv:1610.05643 , a. , janka , h .-
t . , & mller , e. 2013 , , 552 , a126 , a. , mller , e. , & janka , h .-
2015 , , 577 , a48 , s. , & sawai , h. 2004 , , 608 , 907
we have developed a new 3d radiation transfer code to compute polarization spectrum of one line from arbitrary 3d distribution of the line optical depth .
the code uses the monte carlo method , which is a common method to compute polarization by scattering processes ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
for the application to sne , see @xcite .
we set up the 3d cartesian spatial mesh with the @xmath43 meshes .
the velocity is used as a spatial coordinate because the sn ejecta are homologously expands ( @xmath44 ) , the outer velocity of the grid is @xmath45 25000 km s@xmath8 , and thus , the resolution is @xmath46 500 km s@xmath8 .
this spatial resolution gives the wavelength resolution of @xmath47 , which is sufficient to make comparison with observed data .
since the code computes only one ( arbitrary ) line , the wavelength range used in the computation is very small .
if the rest wavelength of the line is @xmath48 , we compute the spectrum only at the wavelength range between @xmath49 and @xmath50 . within this wavelength range
, the energy spectrum is assumed to be constant ( @xmath51 = const ) .
our code assumes a sharply defined inner boundary , and solves radiation transfer above the boundary by tracking the photon packets in the expanding ejecta .
every photon packet has assigned energy , wavelength , and stokes parameters .
especially each photon packet in the simulation has a constant energy , irrespective of the wavelength of the packet . because of this treatment , any photon is not lost during the simulation , which results in the accurate energy conservation @xcite .
the position of the inner boundary is determined so that the electron scattering optical depth from the inner boundary to infinity is @xmath52 . in the simulations used in the main text of the paper
, we always adopt @xmath53 ( table [ tab : param ] ) as in @xcite and @xcite .
the radiation from the inner boundary is assumed to be thermalized , and thus , to be unpolarized ; @xmath54 the direction of the photon is determined by @xmath55 @xcite ( hereafter we use @xmath56 to denote a random number , @xmath57 ) , where @xmath58 is cosine of the angle between the radial and photon direction .
the azimuthal angle around the radial direction @xmath59 is uniformly distributed , i.e. , @xmath60 .
the emitted photon packets experience the electron scattering and the line scattering , which are treated in a similar way to that by @xcite . for the electron scattering ,
we assume a power - law density structure with the power - law index @xmath61 .
we also have the photospheric velocity ( @xmath62 ) and the epoch from the explosion ( @xmath63 ) as input parameters . the photospheric radius ( @xmath64 ) is defined to be the radius where the optical depth for the electron scattering is unity . by setting @xmath62 and @xmath63 ,
the normalization of the electron density is determined . for the line scattering , we use the sobolev approximation @xcite , and assume a power - law optical depth profile with the same index @xmath61 .
the parameter for the line scattering is @xmath16 , the optical depth at the photosphere .
in addition , we assume enhancement of the optical depth by a factor of @xmath65 in some region .
the parameters used in the simulations are summarized in table [ tab : param ] .
a photon packet propagating in one computational grid can have 3 possible events ; ( 1 ) escaping from the grid , ( 2 ) the electron scattering , and ( 3 ) the line scattering .
the event that actually occurs is judged by calculating the length to the 3 events .
it is simple to compute the length to the next grid @xmath66 for the given position and the direction vector of the photon packet .
the direction to the electron scattering event is computed by the randomly selected event optical depth @xmath67 .
when the optical depth reaches this value , a scattering event occurs .
thus , the distance to the electron scattering @xmath68 can be computed by @xmath69 .
when @xmath68 is shorter than @xmath66 , the electron scattering occurs if there is no contribution of line scattering .
since the line scattering is treated as a resonance , the distance to the line scattering event is @xmath70 , where @xmath71 is the comoving wavelength of the photon packet .
if @xmath72 is shortest among 3 lengths , the line scattering is taken into account .
the line scattering event actually occurs when the sum of the line scattering optical depth ( @xmath73 ) and the electron scattering optical depth in @xmath74 ( @xmath75 ) exceeds @xmath76 .
if this sum does not reach @xmath76 , then the electron scattering opacity is evaluated and added again , and the fate of the packet is the electron scattering or the escape from the grid . for illustration of this process , see figure 1 of @xcite .
when the scattering event occurs , the next direction vector of the photon packets is determined . for the electron scattering ,
this scattering angle depends on the polarization , which is discussed in the next section .
for the line scattering , the direction is determined by the isotropic probability function in the comoving frame . by the scattering event ,
the energy and the wavelength of the packet are changed . for the energy , by the energy conservation in the rest frame , @xmath77 where @xmath78 and @xmath79 are the rest - frame energy of the incoming and outgoing packets , respectively . and @xmath80 and @xmath81 are the cosines of the angles between the radial direction and incoming / outgoing propagating directions , respectively .
similarly , the change in the wavelength is given by @xmath82 where @xmath83 and @xmath84 are the rest - frame wavelength of the incoming and outgoing packet , respectively .
scattering events change the polarization properties of the photon packets . for the electron scattering
, the phase matrix can be written as follows @xcite ; @xmath85 where @xmath86 is the scattering angle on the plane of the scattering .
this matrix should be operated in the scattering frame . in general , the rotation matrix for the stokes parameters is written as follows @xcite ; @xmath87 by using these matrices , the effect on the stokes parameters is given by @xmath88 here @xmath89 and @xmath90 is stokes parameter in the rest frame before and after the scattering , respectively .
the angles @xmath91 and @xmath92 are the angles on the spherical triangle defined as in ( @xcite , see figure 1 of @xcite ) .
equation [ eq : pol_escat ] means that the the angle - dependence of the intensity of the scattered light depends the polarization properties of the incident radiation . from equation [ eq : pol_escat ] ,
the probability distribution function ( @xmath93 ) of the total intensity is @xmath94 by using this function with the rejection method as outlined in @xcite , we determine the scattering angle of the electron scattering . for the computation of polarization for the electron scattering , the code was tested with the analytic formulae by @xcite for optically thin cases , and with numerical results by @xcite for optically thick cases . for both cases , we got an excellent agreement . for the application to a sn , expanding ejecta with the steep density slope , we checked our results with those by @xcite .
we confirmed that our code gives the consistent results on the radial profile of polarization for several power - law indexes ( @xmath61 ) and the inner boundaries ( @xmath52 ) . | we present modelling of line polarization to study multi - dimensional geometry of stripped - envelope core - collapse supernovae ( sne ) .
we demonstrate that a purely axisymmetric , two - dimensional geometry can not reproduce a loop in the stokes @xmath0 diagram , i.e. , a variation of the polarization angles along the velocities associated with the absorption lines . on the contrary
, three - dimensional ( 3d ) clumpy structures naturally reproduce the loop .
the fact that the loop is commonly observed in stripped - envelope sne suggests that sn ejecta generally have a 3d structure .
we study the degree of line polarization as a function of the absorption depth for various 3d clumpy models with different clump sizes and covering factors .
comparison between the calculated and observed degree of line polarization indicates that a typical size of the clump is relatively large , @xmath1 of the photospheric radius .
such large - scale clumps are similar to those observed in the sn remnant cassiopeia a. given the small size of the observed sample , the covering factor of the clumps is only weakly constrained ( @xmath2 ) .
the presence of large - scale clumpy structure suggests that the large - scale convection or standing accretion shock instability takes place at the onset of the explosion . |
suppose @xmath0 is a closed , connected , orientable , irreducible 3-manifold such that @xmath1 is infinite .
the `` universal covering conjecture '' states that the universal covering space @xmath2 of @xmath0 must be homeomorphic to @xmath3 .
it is known that @xmath2 is an irreducible , contractible , open 3-manifold @xcite . a * whitehead manifold * is an irreducible , contractible , open 3-manifold which is not homeomorphic to @xmath3 .
the universal covering conjecture is equivalent to the statement that whitehead manifolds can not cover closed 3-manifolds .
in @xcite the author proved that `` genus one '' whitehead manifolds can not non - trivially cover other 3-manifolds , even non - compact ones .
wright @xcite extended this result to the much larger class of `` eventually end - irreducible '' whitehead manifolds , a class which includes all those whitehead manifolds which are monotone unions of cubes with a bounded number of handles .
tinsley and wright @xcite gave specific examples of whitehead manifolds which are not eventually end - irreducible and can not non - trivially cover any other 3-manifolds .
they also constructed an uncountable family of whitehead manifolds which are infinite cyclic covering spaces of other 3-manifolds and deduced from the countability of the set of homeomorphism types of closed 3-manifolds that there must be uncountably many of these which can not cover closed 3-manifolds ; however their methods did not establish which ones these were . in @xcite
the author constructed a different uncountable family of whitehead manifolds which are infinite cyclic covering spaces of other 3-manifolds and used different techniques to prove that none of them covers a closed 3-manifold .
this paper combines the methods of @xcite , @xcite , and @xcite to give a much larger class than in @xcite of specific whitehead manifolds which do not cover closed 3-manifolds but may non - trivially cover other non - compact 3-manifolds , namely the class of `` strong end sums along a locally finite tree '' of eventually end - irreducible whitehead manifolds .
in fact it is shown that whenever such a manifold covers a 3-manifold the group of covering translations must be a free group ( theorem 3.1 ) .
moreover for any countable free group there are uncountably many specific examples of orientable , irreducible open 3-manifolds whose fundamental groups are isomorphic to the given group and whose universal covering spaces belong to this class and are pairwise non - homeomorphic ( theorem 4.1 ) .
there are also uncountably many specific examples in this class which can be only infinite cyclic covering spaces of 3-manifolds and uncountably many specific examples which can not non - trivially cover any 3-manifold .
the results of @xcite use a theorem of geoghegan and mihalik @xcite which implies that whenever a whitehead manifold @xmath4 covers an orientable 3-manifold the group of covering translations must inject into the mapping class group of @xmath4 .
if @xmath4 covers a closed , orientable , irreducible 3-manifold then the group of covering translations must be finitely generated and torsion - free . in @xcite
the examples were constructed so that every finitely generated , torsion - free subgroup of their mapping class groups must have a subgroup of finite index which either has infinite abelianization or a non - trivial normal abelian subgroup .
results of waldhausen @xcite , hass - rubinstein - scott @xcite , mess @xcite , casson - jungreis @xcite , and gabai @xcite were then quoted which imply that a closed , orientable , irreducible 3-manifold with such a fundamental group must have universal covering space homeomorphic to @xmath3 .
the present paper avoids the use of the geoghegan - mihalik result and the requisite analysis of the mapping class group . for the class of whitehead manifolds under consideration results of @xcite
are used to show that the group of covering translations acts on a certain simplicial tree .
the orbit lemma of @xcite and the special ratchet lemma of @xcite are then used to prove that this action fixes no vertices , from which the result follows .
we remark that the methods by which tinsley and wright apply these lemmas in their proof of theorem 5.3 of @xcite could be adapted to prove this fact .
however , we present a different , somewhat more direct argument which is closer in spirit to wright s proof of the main theorem of @xcite .
we also give an alternative , somewhat shorter proof of the special case of the orbit lemma that we use .
the whitehead manifolds considered in @xcite , @xcite , and this paper are all `` end sums '' of whitehead manifolds ; they are obtained by gluing together a collection of whitehead manifolds in a certain way ( see the next section for the precise definition ) .
the summands in @xcite are members of a certain uncountable collection of genus one whitehead manifolds discovered by mcmillan @xcite ; the summands in @xcite are members of a different uncountable collection of genus one whitehead manifolds chosen so that the mapping class group of the end sum will have the appropriate structure as described above .
however , the main difference is not in the summands , but in how they are glued together .
the examples of @xcite are all `` strong '' end sums which have a certain `` rigidity up to isotopy '' in their construction .
the end sums in @xcite are not strong end sums ; in fact it follows from proposition 2.1 below that these manifolds can not be expressed in any way as strong end sums , even though by proposition 2.2 below their summands can be glued together in a different fashion to obtain different manifolds which are strong end sums .
thus the results of this paper apply to all the examples of @xcite but none of the examples of @xcite .
the question of which of them can not cover closed 3-manifolds ( conjecturally all of them ) is still open .
for general background on 3-manifold topology see @xcite or @xcite . we denote the manifold theoretic boundary and interior of a manifold @xmath0 by @xmath5 and @xmath6 , respectively .
we denote the topological boundary , interior , and closure of a submanifold @xmath0 of a manifold @xmath7 by @xmath8 , @xmath9 , and @xmath10 , respectively , with the subscript deleted when @xmath7 is clear from the context .
the * exterior * of @xmath0 in @xmath7 is the closure of the complement of a regular neighborhood of @xmath0 in @xmath7 .
@xmath0 is * open * if @xmath11 and no component of @xmath0 is compact . a continuous map @xmath12 of manifolds is * @xmath13-proper * if @xmath14 . it is * end - proper * if preimages of compact sets are compact .
it is * proper * if it has both these properties .
these terms are applied to a submanifold if its inclusion map has the corresponding property .
two codimension one submanifolds @xmath15 and @xmath16 of @xmath7 , each of which is either proper in @xmath7 or is a submanifold of @xmath17 , are * parallel * if some component of @xmath18 has closure homeomorphic to @xmath19 $ ] with @xmath20 , @xmath21 .
a proper codimension one submanifold of @xmath7 is * @xmath13-parallel * if it is parallel to a submanifold of @xmath17 .
an * exhaustion * @xmath22 for a connected , non - compact manifold @xmath4 is a sequence of compact , connected , codimension zero submanifolds of @xmath4 whose union is @xmath4 , such that @xmath23 , @xmath24 is either empty or a codimension zero submanifold of @xmath25 , and @xmath26 has no compact components .
a connected , non - compact 3-manifold @xmath4 is * eventually end - irreducible * if it has an exhaustion @xmath27 such that @xmath28 is incompressible in @xmath29 for all @xmath30 .
we also say that @xmath4 is * end - irreducible rel @xmath31*. @xmath4 is * eventually @xmath32-injective at @xmath33 * if there is a compact subset @xmath34 of @xmath4 such that for every compact subset @xmath35 of @xmath4 containing @xmath34 there is a compact subset @xmath36 of @xmath4 containing @xmath35 such that every loop in @xmath37 which is null - homotopic in @xmath38 is null - homotopic in @xmath39 .
we also say that @xmath4 is * @xmath32-injective at @xmath33 rel @xmath34*. it is a standard exercise to show that @xmath4 is eventually end - irreducible if and only if it is eventually @xmath32-injective at @xmath33 . note in particular
that if @xmath4 is end - irreducible rel @xmath31 , then it is @xmath32-injective at @xmath33 rel @xmath31 .
let @xmath40 be an irreducible non - compact 3-manifold such that either @xmath41 or each component of @xmath42 is a plane .
a proper plane @xmath43 in @xmath40 is * trivial * if some component of @xmath44 has closure homeomorphic to @xmath45 with @xmath46 .
@xmath40 is * @xmath47-irreducible * every proper plane in @xmath40 is trivial ( hence @xmath41 or @xmath48 ; it is * aplanar * if every proper plane in @xmath40 is either trivial or @xmath13-parallel .
a * partial plane * is a simply connected , non - compact 2-manifold with non - empty boundary .
@xmath40 is * strongly aplanar * if it is aplanar and given any proper 2-manifold @xmath49 in @xmath40 each component of which is a partial plane , there is a collar on @xmath42 which contains @xmath49 .
@xmath40 is * anannular at @xmath33 * if for every compact subset @xmath35 of @xmath40 there is a compact subset @xmath36 of @xmath40 containing @xmath35 such that @xmath50 is * anannular * , i.e. every proper , incompressible annulus in @xmath50 is @xmath13-parallel .
now suppose we are given a countable simplicial tree @xmath51 to each vertex @xmath52 of which we have associated a connected , oriented , irreducible , non - compact 3-manifold @xmath53 whose boundary is a non - empty disjoint union of planes .
suppose that to each edge @xmath54 of @xmath51 we have associated a component of @xmath55 and a component of @xmath56 , where @xmath54 has endpoints @xmath52 and @xmath57 and no boundary plane is associated to different edges .
the connected , oriented , non - compact 3-manifold @xmath4 obtained by gluing each such pair of planes by an orientation reversing homeomorphism is called the * plane sum * of the @xmath53 along @xmath51 . the image in @xmath4 of the pair of planes identified as above is denoted by @xmath58 and is called a * summing plane*. the plane sum is * degenerate * if either some summing plane is trivial or @xmath13-parallel in @xmath4 or two distinct summing planes are parallel in @xmath4 .
theorem 3.2 of @xcite gives necessary and sufficient conditions on @xmath51 and the @xmath53 for the plane sum to be non - degenerate .
for our present purposes corollary 3.3 of @xcite , which states that the plane sum is non - degenerate if no summand @xmath53 has a boundary plane @xmath58 such that @xmath59 is homeomorphic to @xmath45 , will suffice because in our case @xmath60 will be a whitehead manifold .
the plane sum is * strong * if it is non - degenerate and each summand is strongly aplanar and anannular at @xmath33 .
let @xmath4 be a non - degenerate plane sum of aplanar 3-manifolds along a locally finite tree .
let @xmath61 be a strong plane sum .
let @xmath62 and @xmath63 be the unions of the respective sets of summing planes .
suppose @xmath64 is a homeomorphism .
then @xmath65 is ambient isotopic rel @xmath25 to a homeomorphism @xmath66 such that @xmath67 .
* proof : * this is theorem 4.3 of @xcite .
@xmath68 now suppose that given @xmath51 we have associated to each vertex @xmath52 a connected , open , irreducible , oriented 3-manifold @xmath69 , and that to each edge @xmath54 we have associated an end - proper ray ( a space homeomorphic to @xmath70 ) in @xmath69 and an end - proper ray in @xmath71 , where @xmath54 has endpoints @xmath52 and @xmath57 , the rays associated to different edges are disjoint and their union is end - proper .
the exterior @xmath53 of the union of the rays contained in @xmath69 is then bounded by planes .
note that @xmath60 is homeomorphic to @xmath69 .
the plane sum @xmath4 of the @xmath53 along @xmath51 is called an * end sum * of the @xmath69 along @xmath51 .
( note that @xmath4 depends on the choice of the rays ; this dependence is investigated further in @xcite . )
a * strong end sum * is one whose associated plane sum is strong .
we conclude this section with some remarks about the existence of strong end sums . in the present context
the following is the most relevant fact ; more general results may be found in @xcite and @xcite . given a countable , locally finite tree @xmath51 , a collection @xmath72 of connected , irreducible , oriented , one - ended open 3-manifolds , and a bijection between the vertices of @xmath51 and @xmath72
, there exists a strong end sum of the @xmath69 along @xmath51 .
* proof : * this is a special case of theorem 5.1 of @xcite .
@xmath68 for later reference we briefly describe the construction of the rays required in the proof of this result .
suppose @xmath40 is a connected , orientable , irreducible , one - ended , non - compact 3-manifold whose boundary is either empty or consists of a finite set of disjoint planes .
an exhaustion @xmath73 for @xmath40 is * nice * if for all @xmath74 one has that @xmath75 is irreducible , @xmath13-irreducible , and anannular , and that for all @xmath30 one has that each component of @xmath76 has positive genus and negative euler characteristic , and if @xmath77 , one has that @xmath78 consists of a single disk in each component of @xmath42 .
one says that @xmath40 is * nice * if it has a nice exhaustion . if @xmath40 is nice , then @xmath40 is strongly aplanar and anannular at @xmath33 .
* proof : * this follows from theorem 5.3 and lemma 1.3 ( 6 ) of @xcite .
@xmath68 given @xmath69 one chooses an exhaustion @xmath27 for @xmath69 with each @xmath79 connected and of positive genus .
if @xmath80 rays are required , then for each @xmath81 one chooses a disjoint union of @xmath80 proper arcs in @xmath82 each component of which joins @xmath83 to @xmath28 .
this is done so that the endpoints match up on @xmath28 so as to give @xmath80 rays in @xmath69 .
then we obtain an exhaustion @xmath73 for @xmath53 by letting @xmath84 and for @xmath81 letting @xmath85 be the exterior in @xmath82 of the union of the arcs .
all that remains is to note that by theorem 1.1 of @xcite one can choose the arcs so that @xmath75 is irreducible , @xmath13-irreducible , and anannular . in section 4
we will give explicit constructions of examples of this type which do not rely on theorem 1.1 of @xcite .
let @xmath4 be a strong end sum of eventually end - irreducible whitehead manifolds @xmath69 along a locally finite tree @xmath51 .
if @xmath4 is a covering space of a 3-manifold @xmath0 , then there is a simplicial action of @xmath1 on @xmath51 under which no non - trivial element of @xmath1 fixes a vertex of @xmath51 .
hence 1 .
@xmath1 is a free group .
@xmath0 can not be a closed 3-manifold .
3 . if @xmath51 has countably many ends , then @xmath1 is cyclic .
if the number of ends of @xmath51 is finite and greater than two , then @xmath1 is trivial , i.e. @xmath86 . * proof : * we first show how to deduce ( 1)(4 ) from the main statement of the theorem .
( 1 ) @xmath1 has a subgroup @xmath87 of index at most two which acts on @xmath51 without inversions of the edges , hence acts freely on @xmath51 , hence is free .
it follows that @xmath1 is itself free @xcite .
( 2 ) if @xmath0 were closed then it would be a connected sum of 2-sphere bundles over @xmath88 ( * ? ? ?
* theorem 5.2 ) , hence would not be aspherical , hence its universal covering space would not be contractible .
( 3 ) if @xmath89 , then @xmath51 has uncountably many ends .
( 4 ) suppose @xmath90 is an axis for the action of @xmath1 on @xmath51 , i.e. @xmath90 is a subtree isomorphic to a triangulation of @xmath91 which is invariant under the infinite cyclic action ( see @xcite ) .
since @xmath51 has at least three ends some component of @xmath92 has non - compact closure @xmath93 , and the translates of @xmath93 yield infinitely many ends of @xmath51 .
we now prove the main statement of the theorem .
let @xmath94 be the group of covering translations .
by proposition 2.1 each @xmath95 is isotopic to a homeomorphism @xmath66 such that @xmath96 , where @xmath62 is the union of the summing planes of @xmath4 .
thus @xmath66 determines an element of @xmath97 .
we claim that this element depends only on @xmath65 .
we repeat the argument of theorem 3.2 of @xcite .
if @xmath98 were a homeomorphism isotopic to @xmath65 which determined a different automorphism , then @xmath66 and @xmath98 would send some summing plane @xmath99 to different summing planes @xmath100 and @xmath58 , hence they would be ambient isotopic in @xmath4 .
but by theorem 5 of @xcite disjoint , ambient isotopic , non - trivial , proper planes in an irreducible 3-manifold must be parallel .
this contradicts the non - degeneracy of strong end sums .
thus we have a well defined action of @xmath101 on @xmath51 .
we next state the results of @xcite and @xcite that we shall need in order to prove that no vertex is fixed by a non - trivial element of @xmath101 .
let @xmath101 be a group acting on an @xmath102-manifold @xmath4 .
one says that @xmath101 acts * without fixed points * if the only element of @xmath101 fixing a point is the identity .
@xmath101 acts * totally discontinuously * if for every compact subset @xmath103 of @xmath4 one has that @xmath104 for all but finitely many elements of @xmath101 .
( in @xcite the term properly discontinuously " is used for this property ; we follow freedman and skora s terminology @xcite in order to avoid confusion with other meanings of this term . )
let @xmath105 be the projection to the orbit space @xmath106 of the action .
then @xmath101 acts without fixed points and totally discontinuously on @xmath4 if and only if @xmath107 is a regular covering map with group of covering translations @xmath101 and @xmath106 is an @xmath102-manifold .
( see @xcite . ) in this case if @xmath4 is contractible , then @xmath101 must be torsion - free ( see e.g. @xcite or @xcite ) .
let @xmath4 be a contractible , open @xmath102-manifold , @xmath108 .
let @xmath65 be a non - trivial homeomorphism of @xmath4 onto itself such that the group @xmath109 of homeomorphisms generated by @xmath65 acts without fixed points and totally discontinuously on @xmath4 .
given compact subsets @xmath110 and @xmath111 of @xmath4 , there is a compact subset @xmath103 of @xmath4 containing @xmath110 such that every loop in @xmath112 is homotopic in @xmath113 to a loop in @xmath114 .
* proof : * except for the statement that @xmath103 contains @xmath110 this is lemma 4.1 of @xcite ; we can clearly enlarge the @xmath103 of that result to satisfy this requirement .
we now give an alternate proof for the special case in which @xmath4 is an irreducible 3-manifold .
the quotient manifold @xmath115 is an irreducible open 3-manifold having the homotopy type of a circle .
any irreducible open 3-manifold with locally free fundamental group has an exhaustion by cubes with handles ( theorem 2 of @xcite ) .
let @xmath116 be such an exhaustion for @xmath106 .
we may assume that @xmath117 is onto and @xmath118 , where @xmath105 is the covering projection .
thus @xmath119 .
now @xmath120 is a non - compact cube with handles .
there is a finite set of disjoint , proper disks in @xmath120 whose union splits @xmath120 into a compact cube with handles @xmath87 which contains @xmath121 and a 3-manifold @xmath122 whose components are non - compact cubes with handles . these splitting disks can be chosen disjoint from @xmath110 .
let @xmath123 .
suppose @xmath124 is a loop in @xmath112 .
homotop @xmath124 so that it is in general position with respect to @xmath125 .
then it meets @xmath122 in a finite set of paths @xmath126 .
since the components of @xmath122 are cubes with handles each @xmath126 can be homotoped rel @xmath127 to a path @xmath128 in @xmath125 .
this can be done so that no @xmath128 meets a splitting disk .
thus @xmath124 is homotopic in @xmath112 , and hence in @xmath113 , to a loop @xmath129 which lies in @xmath130 and hence in @xmath131 .
@xmath68 let @xmath4 be an open @xmath102-manifold and @xmath132 an open subset of @xmath4 with closure @xmath133 .
suppose @xmath132 is @xmath32-injective at @xmath33 rel @xmath34 , @xmath133 is an @xmath102-manifold , @xmath134 is proper and bicollared in @xmath4 , and each component of @xmath134 is simply connected .
let @xmath65 be a homeomorphism of @xmath4 onto itself such that each of @xmath135 and @xmath136 can be ambiently isotoped into @xmath132 .
then there is a compact subset @xmath137 of @xmath4 containing @xmath34 such that a loop in @xmath138 is null - homotopic in @xmath38 if and only if it is null - homotopic in @xmath139 for each @xmath140 . *
proof : * this is a slight variation of lemma 5.1 of @xcite which has the same proof .
@xmath68 the hypotheses of the special ratchet lemma are clearly satisfied when @xmath101 acts on @xmath51 with fixed points , i.e. some non - trivial @xmath95 is isotopic to @xmath66 such that @xmath141 for the plane summand @xmath133 associated to an end summand @xmath132
. we shall prove that @xmath132 is * @xmath32-trivial at @xmath33 * , i.e. for every compact subset @xmath90 of @xmath132 there is a compact subset @xmath142 of @xmath132 containing @xmath90 such that every loop in @xmath143 is null - homotopic in @xmath144 . by a result of c. h. edwards
@xcite and c. t. c. wall @xcite every irreducible , contractible , open 3-manifold which is @xmath32-trivial at @xmath33 must be homeomorphic to @xmath3 .
this contradicts the assumption that @xmath132 is a whitehead manifold .
so , let @xmath90 be a compact subset of @xmath132 .
now @xmath132 is @xmath32-injective at @xmath33 rel @xmath34 for some compact subset @xmath34 of @xmath132 . by the special ratchet lemma
there is a compact subset @xmath137 of @xmath4 containing @xmath34 such that a loop in @xmath138 is null - homotopic in @xmath38 if and only if it is null - homotopic in @xmath139 for all @xmath140 .
let @xmath145 $ ] be a collar on @xmath134 in @xmath133 such that @xmath146 and @xmath147 .
let @xmath148 .
then @xmath149 is a compact subset of @xmath132 which contains @xmath34 .
let @xmath150 .
since @xmath132 is @xmath32-injective at @xmath33 rel @xmath34 there is a compact subset @xmath36 of @xmath132 containing @xmath35 such that loops in @xmath151 which are null - homotopic in @xmath152 are null - homotopic in @xmath153 .
apply the orbit lemma with @xmath154 and @xmath155 to get a compact subset @xmath103 of @xmath4 containing @xmath36 such that every loop in @xmath112 is homotopic in @xmath37 to a loop in @xmath138 . by enlarging @xmath103 , if necessary
, we may assume that @xmath156 consists of cylinders @xmath157 $ ] , where @xmath158 is a disk in the component @xmath100 of @xmath134 .
there is an @xmath159 such that the collar @xmath160 $ ] misses @xmath36 .
let @xmath161 .
we claim that we may take @xmath162 .
consider a loop @xmath124 in @xmath163 .
we will show that @xmath124 is null - homotopic in @xmath144 .
first note that @xmath164 is contained in the union of the @xmath165 .
we can homotop @xmath124 in @xmath163 , if necessary , so that it misses the union of the @xmath166 $ ] , where @xmath167 is a point in the interior of @xmath158 . by pushing radially outward from @xmath166
$ ] in each @xmath168 $ ] and then off @xmath168 $ ] we obtain a homotopy of @xmath124 in @xmath163 to a loop @xmath129 in @xmath169 .
now @xmath129 is homotopic in @xmath37 to a loop @xmath170 in @xmath138 .
since @xmath4 is contractible @xmath170 is null - homotopic in @xmath4 .
since @xmath109 is totally discontinuous @xmath170 is null - homotopic in @xmath139 for some @xmath171 .
since @xmath170 lies in @xmath37 the special ratchet lemma implies that @xmath170 is null - homotopic in @xmath38 .
since @xmath172 we have that @xmath124 is null - homotopic in @xmath38 . since @xmath124 lies in @xmath152 and the components of @xmath134
are simply connected we have that @xmath124 is null - homotopic in @xmath152 .
thus @xmath124 is null - homotopic in @xmath173 , as required .
1 . given any countable free group @xmath174
there are uncountably many specific irreducible , orientable , open 3-manifolds @xmath175 such that @xmath176 , any 3-manifold @xmath0 covered by the universal covering space @xmath4 of @xmath175 must have free fundamental group , and the @xmath4 are pairwise non - homeomorphic .
if @xmath177 , then @xmath175 can be chosen so that @xmath1 must be infinite cyclic .
if @xmath174 is trivial , then @xmath178 can be chosen so that @xmath86 .
* proof : * ( 1 ) it suffices to consider the case when @xmath174 has rank two . the construction will be a generalization of that of theorem 6.1 of @xcite .
figure 1 shows a six component tangle @xmath179 in a 3-ball @xmath110 called the true lover s 6-tangle . by proposition 4.1 of @xcite
the exterior of @xmath179 is * excellent * , i.e. it is irreducible , @xmath13-irreducible , anannular , and atoroidal , contains a proper incompressible surface , and is not a 3-ball .
it follows immediately from the proof of the result cited that each of the @xmath180-tangles consisting of @xmath181 consecutive components of @xmath179 also has excellent exterior . by sliding the endpoints of the arcs of
@xmath179 one sees that the exterior of @xmath179 is homeomorphic to the exterior of the graph @xmath182 in figure 2 . by deleting the first , second , fifth , and sixth arcs
we obtain the 2-tangle @xmath183 in figure 3 , which thus has excellent exterior .
we next identify the disks which are the left and right sides of the rectangular solid @xmath110 in figures 2 and 3 to obtain a solid torus @xmath35 .
this is done so that @xmath183 becomes a simple closed curve @xmath184 and @xmath182 becomes a graph @xmath185 consisting of @xmath184 together with four disjoint arcs @xmath186 , @xmath187 , @xmath188 , @xmath189 joining @xmath184 to @xmath190 .
it follows from lemma 2.1 of @xcite that the exteriors of @xmath184 and of @xmath185 in @xmath35 are excellent .
now let @xmath36 be a regular neighborhood of @xmath184 in @xmath35 .
we construct a genus one whitehead manifold @xmath191 with exhaustion @xmath27 by using as models for @xmath192 the pair @xmath193 .
this is done so that the copies @xmath194 of the @xmath195 match up along their endpoints to give end - proper rays @xmath196 in @xmath191 .
we then let @xmath40 be @xmath191 minus the interior of a regular neighborhood @xmath7 of the union of these rays .
we choose @xmath7 so that its intersection @xmath197 with @xmath198 is a regular neighborhood of the union of the @xmath194 .
we then let @xmath85 be @xmath199 for @xmath81 and @xmath84 . since @xmath198 and @xmath75 are excellent we have that @xmath191 is an eventually end - irreducible whitehead manifold and @xmath40 is nice .
we now identify the boundary planes of @xmath40 in pairs to obtain an orientable 3-manifold @xmath175 with @xmath200 free of rank two .
the universal covering space @xmath4 of @xmath175 is then an end sum of whitehead manifolds @xmath69 each of which is homeomorphic to @xmath191 such that the plane summands @xmath53 are homeomorphic to @xmath40 .
we then apply theorem 3.1 .
if one changes the sense of the central clasp in the figures by changing the two overcrossings to undercrossings , thereby getting a new @xmath184 and @xmath185 , then the same arguments show that their exteriors in @xmath35 are excellent .
denote the old and new versions by the subscripts 0 and 1 , respectively .
embed @xmath35 in @xmath201 in a standard way so that a line segment running along the bottom front edge of @xmath110 becomes a simple closed curve @xmath202 in @xmath190 which bounds a disk in @xmath203 .
then @xmath204 and @xmath205 become the knots @xmath206 and @xmath207 in @xmath201 with normalized alexander polynomials @xmath208 and @xmath209 , respectively .
it then follows that there is no homeomorphism from the exterior of @xmath204 in @xmath35 to that of @xmath205 in @xmath35 which carries @xmath202 to a curve homologous to @xmath210 .
since if there were , then one could extend it to a homeomorphism of the exteriors in @xmath201 of these two knots .
let @xmath211 be an infinite sequence of 0 s and 1 s .
carry out the construction as before by modeling the pair @xmath192 , for @xmath74 , on @xmath212 , where @xmath213 is a regular neighborhood of @xmath214 in @xmath35 and @xmath215 .
do this so that the copy @xmath216 of @xmath202 in @xmath79 is null - homologous in @xmath217 .
( up to orientation and isotopy there is a unique such curve . )
label the various manifolds arising in the construction associated to @xmath218 by a superscript @xmath218 .
if @xmath219 is a homeomorphism , then lemma 3.3 of @xcite implies that @xmath220 can be isotoped so that for some @xmath221 and @xmath222 one has @xmath223 for all @xmath224 .
thus @xmath225 for all @xmath224 .
one could now note that this last equation generates an equivalence relation on the set @xmath226 of all such sequences and that there are uncountably many equivalence classes . in keeping with the desire to make our examples as explicit as possible , however , we prefer a more concrete approach which exhibits an explicit subset @xmath227 of @xmath226 for which the corresponding whitehead manifolds are non - homeomorphic .
we define @xmath227 and define a bijection @xmath228 as follows .
let @xmath229 .
then @xmath230 will consist of strings of consecutive 0 s which are separated by single 1 s .
the length of the @xmath231 string of 0 s is @xmath232 , where @xmath233 if @xmath234 and @xmath235 if @xmath236 .
thus @xmath237 , where the total exponent sum @xmath238 .
suppose @xmath239 is another sequence such that for some @xmath221 and @xmath222 one has @xmath225 for all @xmath240 .
locate the first 1 in this common tail .
it is followed by a string of @xmath241 0 s for some unique @xmath242 and @xmath243 .
then @xmath244 for a unique @xmath102 , and so this is the @xmath231 string of 0 s in both @xmath218 and @xmath245 .
note that @xmath246 .
suppose @xmath247 where the @xmath248 and @xmath249 correspond to the @xmath167 and @xmath250 as above .
then @xmath251 ; let @xmath252 .
if @xmath253 , then since @xmath254 has exponent sum in 3 at most @xmath255 we must have @xmath256 ; since a similar argument holds for powers of 5 we have that @xmath257 .
we inductively conclude that @xmath258 , and hence @xmath259 , for @xmath260 .
applying this argument to all @xmath261 we get that @xmath262 and @xmath263 .
thus we have uncountably many non - homeomorphic genus one whitehead manifolds @xmath264 .
we construct the corresponding @xmath265 , @xmath266 , and @xmath267 .
the @xmath268 are all homeomorphic to @xmath264 .
it then follows from proposition 2.1 that if @xmath267 and @xmath269 are homeomorphic , so are @xmath264 and @xmath270 , hence @xmath263 .
\(3 ) one can carry out the construction of @xmath40 as above with any finite number @xmath80 of boundary planes by using the true lover s @xmath271-tangle .
thus given any locally finite tree @xmath51 one can construct the corresponding strong end sum .
one can then choose @xmath51 to have the wrong number of ends or , for variety , let @xmath51 be arbitrary but choose one @xmath132 which is not homeomorphic to any of the other @xmath69 , thereby creating a fixed vertex for the action on @xmath51 . @xmath68 | this paper concerns the class of contractible open 3-manifolds which are `` locally finite strong end sums '' of eventually end - irreducible whitehead manifolds .
it is shown that whenever a 3-manifold in this class is a covering space of another 3-manifold the group of covering translations must be a free group .
it follows that such a 3-manifold can not cover a closed 3-manifold . for each countable free group a specific uncountable family of irreducible open 3-manifolds
is constructed whose fundamental groups are isomorphic to the given group and whose universal covering spaces are in this class and are pairwise non - homeomorphic . |
let @xmath9 be an indeterminate over @xmath3 .
the brauer algebra @xmath10 over @xmath11 $ ] is a unital associative @xmath11$]-algebra with generators @xmath12 @xmath13 and relations ( see @xcite ) : @xmath14 @xmath10 is a free @xmath11$]-module with rank @xmath15 . for any @xmath11$]-algebra @xmath16 with @xmath9 specialized to @xmath17
, we define @xmath18}{\mathfrak{b}}_n(x)$ ] .
this algebra was first introduced by richard brauer ( see @xcite ) when he studied how the @xmath2-tensor space @xmath19 decomposes into irreducible modules over the orthogonal group @xmath20 or the symplectic group @xmath21 , where @xmath6 is an orthogonal vector space or a symplectic vector space . in brauer s original formulation ,
the algebra @xmath10 was defined as the complex linear space with basis the set @xmath22 of all the brauer @xmath2-diagrams , graphs on @xmath23 vertices and @xmath2 edges with the property that every vertex is incident to precisely one edge .
if we arrange the vertices in two rows of @xmath2 each , the top and bottom rows , and label the vertices in each row of a @xmath2-diagram by the indices @xmath24 from left to right , then the generator @xmath25 corresponds to the @xmath2-diagram with edges connecting vertices @xmath26 ( respectively , @xmath27 ) on the top row with @xmath27 ( respectively , @xmath26 ) on bottom row , and all other edges are vertical , connecting vertex @xmath28 on the top and bottom rows for all @xmath29 .
the generator @xmath30 corresponds to the @xmath2-diagram with horizontal edges connecting vertices @xmath31 on the top and bottom rows , and all other edges are vertical , connecting vertex @xmath28 on the top and bottom rows for all @xmath29 .
the multiplication of two brauer @xmath2-diagrams is defined as follows .
we compose two diagrams @xmath32 by identifying the bottom row of vertices in the first diagram with the top row of vertices in the second diagram .
the result is a graph , with a certain number , @xmath33 , of interior loops . after removing the interior loops and the identified vertices , retaining the edges and remaining vertices
, we obtain a new brauer @xmath2-diagram @xmath34 , the composite diagram .
then we define @xmath35 . in general
, the multiplication of two elements in @xmath10 is given by the linear extension of a product defined on diagrams .
for example , let @xmath36 be the following brauer @xmath37-diagram .
let @xmath38 be the following brauer @xmath37-diagram .
then @xmath39 is equal to note that the subalgebra of @xmath10 generated by @xmath40 is isomorphic to the group algebra of the symmetric group @xmath41 over @xmath11 $ ] .
the brauer algebra as well as its quantization ( now called birman wenzl murakami algebra ) has been studied in a number of papers , e.g. , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite .
the walled brauer algebra ( which is a variant of brauer algebra , see @xcite ) is also studied in the recent preprint @xcite .
we are mainly interested in the schur
weyl duality between symplectic groups and certain specialized brauer algebras , which we now recall .
let @xmath42 be an arbitrary infinite field .
let @xmath43 .
let @xmath6 be a @xmath5-dimensional @xmath42-vector space equipped with a non - degenerate skew - symmetric bilinear form @xmath44 .
then ( see @xcite , ( * ? ? ?
* section 4 ) ) the symplectic similitude group ( respectively , the symplectic group ) relative to @xmath44 is @xmath45 @xmath46respectively , @xmath47 the symplectic similitude group and symplectic group @xmath21 act naturally on @xmath6 from the left - hand side , and hence on the @xmath2-tensor space @xmath7 .
this left action on @xmath7 is centralized by certain specialized brauer algebra , which we recall as follows .
let @xmath48}{\mathfrak{b}}_n(x)$ ] , where @xmath3 is regarded as @xmath11$]-algebra by specifying @xmath9 to @xmath49 .
let @xmath50 , where @xmath42 is regarded as @xmath3-algebra by sending each integer @xmath51 to @xmath52 .
then there is a right action of the specialized brauer algebra @xmath53 on the @xmath2-tensor space @xmath7 which commutes with the above left action of @xmath54 .
we recall the definition of this action as follows .
let @xmath55 denote the value of the usual kronecker delta .
for any @xmath56 , we set @xmath57 we fix an ordered basis @xmath58 of @xmath6 such that @xmath59 for any @xmath60 , let @xmath61 the right action of @xmath8 on @xmath7 is defined on generators by @xmath62 let @xmath63 be the following natural @xmath42-algebra homomorphisms .
let @xmath28 be a positive integer .
a composition of @xmath28 is a sequence of nonnegative integer @xmath65 with @xmath66 .
a composition @xmath65 of @xmath28 is said to be a partition if @xmath67 . in this case
, we write @xmath68 .
the conjugate of @xmath69 is defined to be a partition @xmath70 , where @xmath71 for @xmath72 . for any partition @xmath65
, we use @xmath73 to denote the largest integer @xmath74 such that @xmath75 .
( @xcite , @xcite , @xcite ) 1 ) the natural left action of @xmath54 on @xmath76 commutes with the right action of @xmath8 . moreover , if @xmath77 , then @xmath78 \2 ) if @xmath77 and @xmath79 then @xmath80 is injective , and hence an isomorphism onto @xmath81 , \3 ) if @xmath77 , then there is a decomposition of irreducible @xmath82@xmath83 bimodules @xmath84}\bigoplus_{\substack{{\lambda}\vdash n-2f\\ \ell({\lambda})\leq m}}\delta({{\lambda}})\otimes d({{\lambda}'}),\ ] ] where @xmath85 ( respectively , @xmath86 ) denotes the irreducible @xmath82-module ( respectively , the irreducible @xmath83-module ) corresponding to @xmath69 ( respectively , corresponding to @xmath87 ) . by recent work of @xcite and @xcite , part 1 ) and part 2 ) of the above lemma hold for an arbitrary infinite field .
that is , ( @xcite , @xcite ) let @xmath42 be an arbitrary infinite field .
\1 ) @xmath88 \2 ) @xmath89 , and if @xmath90 , then @xmath80 is also injective , and hence an isomorphism onto @xmath91 now there is a natural question , that is , how can one describe the kernel of the homomorphism @xmath80 .
this question is closely related to invariant theory : see @xcite . by the results in @xcite , we know that the kernel of the homomorphism @xmath80 has a rigid structure in the sense that the dimension of @xmath92 does not depend on the choice of the infinite field @xmath42 , and it is actually defined over @xmath3 .
note that in the case of schur weyl duality between general linear group and symmetric group ( @xcite , @xcite , @xcite , @xcite ) , or more generally , between the type @xmath93 quantum group and the type @xmath93 iwahori hecke algebra ( @xcite , @xcite ) , the kernel of the similar homomorphism has already been explicitly determined in @xcite in terms of kazhdan lusztig basis and in @xcite in terms of murphy basis . in this paper , we completely answer the above question by explicitly constructing an integral basis for the kernel of the homomorphism @xmath80 .
our description of @xmath92 involves a study of the permutation action of the symmetric group @xmath1 on the brauer algebra @xmath10 .
such a permutation action was previously noted in @xcite .
we construct a new integral basis for this brauer algebra , which yields integral specht filtration of this brauer algebra by right @xmath1-modules .
the kernel of @xmath80 is just one of the @xmath1-submodules appearing in this filtration . in particular
, it turns out that @xmath92 is in fact a @xmath1-submodule of @xmath8 .
the main results of this paper are presented in theorem [ thm212 ] , theorem [ thm214 ] and theorem [ thm35 ] .
it would be interesting to compare the new integral basis we obtained in this paper with the canonical basis for @xmath10 constructed in @xcite .
finally , we remark that it might be possible to give a similar description of @xmath92 also in the orthogonal case ( i.e. , the case of schur weyl duality between orthogonal group and certain specialized brauer algebra ) .
we deal with only the symplectic case in this paper because we use the main results in @xcite , where only the symplectic case is considered
. it would also be interesting to see how the description of @xmath92 we give here can be generalized to the quantized case , i.e. , the case of schur weyl duality between the quantized enveloping algebra associated to the symplectic lie algebra @xmath94 and a certain specialized birman wenzl murakami algebra ( see @xcite ) .
in this section , we shall first recall ( cf .
@xcite ) the right permutation action of the symmetric group @xmath1 on the set @xmath22 .
then we shall construct a new @xmath3-basis for the resulting right @xmath1-module , which yields filtrations of @xmath10 by right @xmath1-modules .
certain submodules occurring in this filtration will play central role in the next section . for any involution @xmath95 in the symmetric group @xmath1 , the conjugate @xmath96 of @xmath95 by @xmath97
is still an involution .
therefore , we have a right action of the symmetric group @xmath1 on the set of all the involutions in @xmath1 .
note that the set @xmath22 of all the brauer @xmath2-diagrams can be naturally identified with the set of all the involutions in @xmath1 .
hence we get ( cf .
@xcite ) a right permutation action of the symmetric group @xmath1 on the set @xmath22 of all the brauer @xmath2-diagrams .
we use @xmath98 to denote this right permutation action .
we shall adopt a new labeling of the vertices in each brauer diagram .
namely , for each brauer @xmath2-diagram @xmath99 , we shall label the vertices in the top row of @xmath99 by odd integers @xmath100 from left to right , and label the vertices in the bottom row of @xmath99 by even integers @xmath101 from left to right .
this way of labeling is more convenient when studying the permutation action from @xmath1 .
_ we shall keep this way of labeling from this section until the end of section 3 , and we shall recover our original way of labeling only in section 4 . _
let us look at an example .
suppose @xmath102 , @xmath103 is a transposition in @xmath104 .
let @xmath99 be the following brauer @xmath105-diagram .
we first identify @xmath99 with following diagram with @xmath106-vertices .
then @xmath107 can be computed in the following way . finally , @xmath107 is equal to the following brauer @xmath105-diagram .
we use @xmath108 to denote the natural identification of @xmath22 with the set of involutions in @xmath1 . for any @xmath97 and any @xmath109 , @xmath110 . for any @xmath3-algebra @xmath16
, we use @xmath111 to denote the free @xmath16-module spanned by all the brauer @xmath2-diagrams in @xmath22 .
then @xmath111 becomes a right @xmath112$]-module .
let @xmath113 .
clearly , there is a canonical isomorphism @xmath114 , which is also a right @xmath112$]-module isomorphism .
taking @xmath115 $ ] , we deduce that the brauer algebra @xmath10 becomes a right @xmath11[{\mathfrak s}_{2n}]$]-module .
similarly , the specialized brauer algebra @xmath8 becomes a right @xmath116$]-module . for any integer @xmath26 with @xmath117
, we define @xmath118 then @xmath119 is an involution on @xmath120 .
it is well - known that the subgroup @xmath121 is isomorphic to the wreath product @xmath122 of @xmath123 and @xmath124 , which is a weyl group of type @xmath125 ( c.f .
@xcite ) . [ lm21 ] for any @xmath3-algebra @xmath16
, there is a right @xmath112$]-module isomorphism @xmath126}^{r[{\mathfrak s}_{2n}]}1_{r},\ ] ] where @xmath127 denotes the rank one trivial representation of @xmath128 $ ] .
let @xmath129 be the element in @xmath22 that connects @xmath130 to @xmath131 for each integer @xmath26 with @xmath132 .
since @xmath1 acts transitively on the set of all the brauer @xmath2-diagrams , it is easy to see that the map @xmath133 which send @xmath127 to @xmath129 extends naturally to a surjective @xmath112$]-module homomorphism from @xmath134}^{r[{\mathfrak s}_{2n}]}1_{r}$ ] onto @xmath111 .
if @xmath16 is a field , then we can compare the dimensions of both modules . in that case
, we know that the surjection @xmath135 must be an injection , and hence be an isomorphism .
in general , since there are natural isomorphisms @xmath136}^{r[{\mathfrak
s}_{2n}]}1_{r}\cong r\otimes_{{\mathbb z}}\operatorname{ind}_{{\mathbb z}[{\mathbb z}_2\wr{\mathfrak s}_n]}^{{\mathbb z}[{\mathfrak s}_{2n}]}1_{{\mathbb z}},\quad { \mathfrak{b}}_{n , r}\cong r\otimes_{{\mathbb z}}{\mathfrak{b}}_{n,{\mathbb z}},\ ] ] and @xmath135 is naturally identified with @xmath137 , it suffices to show that @xmath138 is an isomorphism .
note also that the short exact sequence @xmath139}^{{\mathbb z}[{\mathfrak s}_{2n}]}1_{{\mathbb z}}\rightarrow { \mathfrak{b}}_{n,{\mathbb z}}\rightarrow 0\ ] ] splits as @xmath3-modules .
it follows that @xmath140 is canonically isomorphic to @xmath141 for any @xmath3-algebra @xmath16 .
let @xmath142 .
it is enough to show that @xmath143 . by (
* proposition 3.8 ) , we only need to show that @xmath144 for each prime number @xmath145 .
let @xmath146 , the residue field at the prime number @xmath145 .
it is clear that @xmath147 .
note that @xmath148 applying nakayama s lemma ( @xcite ) , we conclude that @xmath144 .
this completes the proof of the lemma . for any positive integer @xmath28 and any composition @xmath149 of @xmath28 , the young diagram of @xmath150
is defined to be the set @xmath151:=\{(a , b)|1\leq a\leq s , 1\leq b\leq\mu_a\}$ ] .
the elements of @xmath151 $ ] are called nodes of @xmath150 .
a @xmath150-tableau @xmath152 is defined to be a bijective map from the young diagram @xmath151 $ ] to the set @xmath153 . for each integer
@xmath51 with @xmath154 , we define @xmath155 if @xmath156 .
we denote by @xmath157 the @xmath150-tableau in which the numbers @xmath158 appear in order along successive rows .
the row stabilizer of @xmath157 , denoted by @xmath159 , is the standard young subgroup of @xmath160 corresponding to @xmath150 .
for example , if @xmath161 , then @xmath162 we define @xmath163 where @xmath164 is the length function in @xmath160 .
if @xmath150 is a partition of @xmath28 , we denote by @xmath165 the @xmath150-tableau in which the numbers @xmath158 appear in order along successive columns .
let @xmath166 be such that @xmath167 .
for example , if @xmath168 , then @xmath169 we use @xmath170 to denote the set of all the partitions of @xmath2 . for any partition @xmath150 of @xmath23
, we define the associated specht module @xmath171 to be the right ideal of the group algebra @xmath172 $ ] generated by @xmath173 .
in particular , @xmath174 is the one - dimensional trivial representation of @xmath1 , while @xmath175 is the one dimensional sign representation of @xmath1 . by (
* theorem 3.5 ) and @xcite , our @xmath171 is isomorphic to the ( dual ) specht module @xmath176 introduced in ( * ? ? ?
* section 5 ) . for any @xmath3-algebra @xmath16
, we write @xmath177 .
then @xmath178 is a complete set of pairwise non - isomorphic simple @xmath179$]-modules . for any composition @xmath180 of @xmath2 , let @xmath181 , which is a composition of @xmath23 .
we define @xmath182 .
[ lm22 ] there is a right @xmath179$]-modules isomorphism : @xmath183 this follows from lemma [ lm21 ] and ( * ? ? ?
* chapter vii , ( 2.4 ) ) .
let @xmath51 be an integer with @xmath184 .
let @xmath185 be @xmath186 pairwise distinct integers .
let @xmath187 let @xmath188 be the symmetric group on the set @xmath189 .
let @xmath190 .
let @xmath191 [ lm23 ] with the notations as above , for any @xmath192 , we have @xmath193 for any @xmath194 , it is clear that @xmath195 therefore , the lemma follows easily from a counting argument . for any non - negative even integers @xmath196 with @xmath197 ,
we define @xmath198 for any even integer @xmath28 with @xmath199 , let @xmath200 .
let @xmath180 be a composition of @xmath23 such that @xmath201 is even for each @xmath26 , we define @xmath202 [ cor26 ] let @xmath180 be a composition of @xmath23 such that @xmath201 is even for each @xmath26 . then , for any @xmath203 , we have that @xmath204 for any non - negative even integers @xmath196 with @xmath197 , the set @xmath205 is just a special case of the set @xmath206 we defined before .
therefore , by lemma [ lm23 ] , for any @xmath207 , we have @xmath208 now we note that the elements @xmath209 pairwise commute with each other .
hence the corollary follows at once .
let @xmath28 be a positive integer and @xmath150 be a composition of @xmath28 .
a @xmath150-tableau @xmath152 is called _ row standard _ if the numbers increase along rows
. we use @xmath210 to denote the set of all the row - standard @xmath150-tableaux .
suppose @xmath150 is a partition of @xmath28
. then @xmath152 is called _ column standard _ if the numbers increase down columns , and _ standard _ if it is both row and column standard . in this case , it is clear that both @xmath157 and @xmath165 are standard @xmath150-tableaux .
we use @xmath211 to denote the set of all the standard @xmath150-tableaux .
now let @xmath150 be a composition of @xmath23 .
let @xmath212 .
for any @xmath213 , let @xmath214 be such that @xmath215 .
let @xmath216 .
for any @xmath3-algebra @xmath16 , we define @xmath217 where `` @xmath218 '' is the dominance order defined in @xcite .
we write @xmath219 .
we are interested in the module @xmath220 . in the remaining part of this paper
, we shall see that this module is actually a right @xmath1-submodule of @xmath111 , and it shares many properties with the permutation module @xmath221 $ ] . in particular , it also has a specht filtration , and it is stable under base change , i.e. , @xmath222 for any @xmath3-algebra @xmath16 . for our purpose , we need to recall some results in @xcite and @xcite on the specht filtrations of permutation modules over the symmetric group @xmath1 .
let @xmath69 , @xmath150 be two partitions of @xmath23 .
a @xmath150-tableau of type @xmath69 is a map @xmath223\rightarrow\{1,2,\cdots,2n\}$ ] such that each @xmath26 appears exactly @xmath201 times .
@xmath224 is said to be semistandard if each row of @xmath224 is weakly increasing and each column of @xmath224 is strictly increasing .
let @xmath225 be the set of all the semistandard @xmath150-tableaux of type @xmath69 .
then @xmath226 only if @xmath227 . for each standard @xmath150-tableau @xmath228 ,
let @xmath229 be the tableau which is obtained from @xmath228 by replacing each entry @xmath26 in @xmath228 by @xmath230 if @xmath26 appears in row @xmath230 of @xmath231 .
then @xmath229 is a @xmath150-tableau of type @xmath69 . for each standard @xmath150-tableau @xmath152 and each semistandard @xmath150-tableau @xmath224 of type @xmath69
, we define @xmath232 then by ( * ? ? ?
* section 7 ) , the set @xmath233 form a @xmath3-basis of @xmath221 $ ] . furthermore , for any @xmath3-algebra @xmath16 , the canonical surjective homomorphism @xmath234\twoheadrightarrow x_{{\lambda}}r[{\mathfrak s}_{2n}]$ ] is an isomorphism . for each partition @xmath150 of @xmath23 and for each semistandard @xmath150-tableau @xmath224 of type @xmath69 , according to the results in ( * ? ? ?
* section 7 ) and @xcite , both the following @xmath3-submodules @xmath235 are @xmath172$]-submodules , and the quotient of @xmath236 by @xmath237 is canonically isomorphic to @xmath171 so that the image of the elements @xmath238 , where @xmath239 , forms the standard @xmath3-basis of @xmath171 . in other words
, it gives rise to the specht filtrations of @xmath221 $ ] , each semistandard @xmath150-tableau of type @xmath69 yields a factor which is isomorphic to @xmath171 so that @xmath221 $ ] has a series of factors , ordered by @xmath240 , each isomorphic to some @xmath171 , @xmath227 , the multiplicity of @xmath171 being the number of semistandard @xmath150-tableaux of type @xmath69 .
we write @xmath241 , where @xmath242 , @xmath243 for each @xmath26 , where @xmath244 means that @xmath245 repeats @xmath246 times .
let @xmath247 be the subgroup of @xmath248 consisting of all the elements @xmath249 satisfying the following condition : for any integer @xmath250 , and any integers @xmath251 with @xmath252 , and any integers @xmath196 with @xmath253 , @xmath254 let @xmath255 be a complete set of right coset representatives of @xmath247 in @xmath248 .
[ lm27 ] for any partition @xmath212 ,
let @xmath256 then @xmath257 and for any @xmath3-algebra @xmath16 , @xmath258 in @xmath111 .
the condition @xmath212 implies that for any @xmath259 , @xmath260 is an even integer .
the first statement of this lemma now follows from the following identity : @xmath261 let @xmath36 be the brauer @xmath2-diagram in which the vertex labeled by @xmath262 is connected with the vertex labeled by @xmath263 for any @xmath264 . then it is easy to see that @xmath36 appears with coefficient @xmath265 in the expression of @xmath266 as linear combinations of basis of brauer @xmath2-diagrams .
it follows that for any @xmath3-algebra @xmath16 , @xmath258 in @xmath111 , as required .
following @xcite , we define the jucys - murphy operators of @xmath172 $ ] .
@xmath267 then for each partition @xmath69 of @xmath23 , and each integer @xmath268 , we have ( by ( * ? ? ?
* ( 3.14 ) ) ) @xmath269 for each standard @xmath69-tableau @xmath152 , we define @xmath270 for each partition @xmath212 , by corollary [ cor26 ] and frobenius reciprocity , there is a surjective right @xmath172$]-module homomorphism @xmath271 from @xmath221 $ ] onto @xmath272 $ ] which extends the map @xmath273 . in particular , by lemma [ lm27 ] , @xmath274 [ prop29 ] for any partition @xmath212 , we have that @xmath275:s_{{\mathbb q}}^{{\lambda}}]=1.\ ] ] by lemma [ lm22 ] , we have @xmath276 it is well - known that each @xmath277 has a basis @xmath278 satisfying @xmath279 let @xmath69 be a fixed partition in @xmath280 .
since @xmath281\subseteq { \mathfrak{b}}_{n,{\mathbb q}}$ ] , we can write @xmath282 where @xmath283 for each @xmath152 . for each @xmath284 and each @xmath285
, we apply the operator @xmath286 on both sides of the above identity and use lemma [ lm27 ] and the above discussion .
we get that @xmath287 if and only if @xmath288 and @xmath289 . in other words ,
@xmath290 for some @xmath291 .
this implies that the projection from @xmath281 $ ] to @xmath292 is nonzero .
hence , @xmath293:s_{{\mathbb q}}^{{\lambda}}]=1 $ ] , as required . for each partition @xmath212 , by the natural surjective @xmath172$]-module homomorphism @xmath271 from @xmath221 $ ] onto @xmath272 $ ] , we know that the elements @xmath294 , where @xmath295 , spans @xmath272 $ ] as @xmath3-module . [ prop210 ] for any two partitions @xmath296 of @xmath23 , and for any @xmath297 , we have that @xmath298 . in particular , @xmath272\subseteq { \mathcal m}^{{\lambda}}$ ] .
we first prove a weak version of the claim in this proposition .
that is , for any two partitions @xmath296 of @xmath23 , and for any @xmath297 , @xmath299 we consider the dominance order @xmath300 and make induction on @xmath69 .
we start with the partition @xmath301 , which is the unique maximal partition of @xmath23 with respect to @xmath300 . in this case , @xmath302={\mathbb z}x_{(2n)}$ ] , and @xmath303={\mathbb z}x_{(2n)}$ ] , it is easy to see the claim in this proposition is true for @xmath304 .
now let @xmath305 be a partition of @xmath23 .
assume that for any partition @xmath306 of @xmath23 satisfying @xmath307 , the claim in this proposition is true .
we now prove the claim for the partition @xmath69 .
let @xmath308 be a partition of @xmath23 with @xmath226 .
we consider again the dominance order @xmath300 and make induction on @xmath150 . since @xmath309 contains a unique element @xmath310 , @xmath311 and @xmath312 so in this case the claim of this proposition is still true .
now let @xmath308 be a partition of @xmath23 with @xmath226 and @xmath313 .
assume that for any partition @xmath306 of @xmath23 satisfying @xmath314 and @xmath315 , and for any @xmath316 , @xmath299 let @xmath297 .
the homomorphism @xmath271 induces a surjective map from @xmath317 onto @xmath318 hence it also induces a surjective map @xmath319 from @xmath320 onto @xmath321 since @xmath277 is irreducible , the above map is either a zero map or an isomorphism .
if it is a zero map , then ( by induction hypothesis ) @xmath322 it remains to consider the case where @xmath319 is an isomorphism .
in particular , @xmath323 applying lemma [ lm22 ] , we know that @xmath284 . on the other hand , the homomorphism @xmath324 also induces a surjective map from @xmath325/m_{\operatorname{s}_0,\rhd}^{\mu}$ ] onto @xmath326)\bigr)/\bigl(\pi_{\mu}(m_{\operatorname{s}_0,\rhd}^{\mu})\bigr)= x_{\mu}{\mathbb z}[{\mathfrak s}_{2n}]/\bigl(\pi_{\mu}(m_{\operatorname{s}_0,\rhd}^{\mu})\bigr),\ ] ] where @xmath327 is the unique semistandard @xmath150-tableau in @xmath328 .
hence it also induces a surjective map @xmath329 from @xmath330/\bigl({\mathbb q}\otimes_{{\mathbb z}}m_{\operatorname{s}_0,\rhd}^{\mu}\bigr)\cong { \mathbb q}\otimes_{{\mathbb z}}\bigl(x_{\mu}{\mathbb z}[{\mathfrak s}_{2n}]/m_{\operatorname{s}_0,\rhd}^{\mu}\bigr)\cong s_{{\mathbb q}}^{\mu}\ ] ] onto @xmath331/\pi_{\mu}(m_{\operatorname{s}_0,\rhd}^{\mu})\bigr)\cong \bigl(x_{\mu}{\mathbb q}[{\mathfrak s}_{2n}]\bigr)/\bigl({\mathbb q}\otimes_{{\mathbb z}}\pi_{\mu}(m_{\operatorname{s}_0,\rhd}^{\mu})\bigr).\ ] ] by the specht filtration of @xmath332 , we know that @xmath277 does not occur as composition factor in @xmath333
. hence @xmath277 does not occur as composition factor in @xmath334 . by proposition [ prop29 ]
, @xmath277 occurs as composition factor with multiplicity one in @xmath335 $ ] .
therefore , @xmath335\neq{\mathbb
q}\otimes_{{\mathbb z}}\pi_{\mu}(m_{\operatorname{s}_0,\rhd}^{\mu})$ ] .
it follows that @xmath329 must be an isomorphism
. hence @xmath331/\pi_{\mu}(m_{\operatorname{s}_0,\rhd}^{\mu})\bigr)\cong s_{{\mathbb q}}^{\mu}.\ ] ] we write @xmath336 $ ] .
since @xmath277 appears only once in @xmath337 , it follows that @xmath277 must occur as composition factor in the module @xmath338 hence @xmath277 can not occur as composition factor in the module @xmath339 therefore , the image of the canonical projection @xmath340 must be contained in the image of @xmath341 .
however , by induction hypothesis , both @xmath342 and @xmath343 are contained in the @xmath344-span of @xmath345 .
it follows that @xmath346 as required .
now we begin to prove @xmath298 .
suppose that @xmath347 then ( by the @xmath3-freeness of @xmath4 ) there exist an element @xmath348 , some integers @xmath349 , and a prime divisor @xmath350 of @xmath51 , such that @xmath351 and @xmath352 we take an @xmath353 such that @xmath354 is minimal with respect to @xmath240 " .
then we take an @xmath355 such that @xmath356 and @xmath357 is maximal among the elements in the set @xmath358 .
let @xmath359 be the unique element in @xmath1 such that @xmath360 and @xmath361 .
we consider the finite field @xmath362 as a @xmath3-algebra . by
* ( 4.1 ) ) , we know that for any partitions @xmath363 of @xmath23 , and element @xmath97 , @xmath364 hence by lemma [ lm27 ] and the homomorphism @xmath271 , @xmath365 using lemma [ lm27 ] again , we get @xmath366 which is a contradiction .
this prove that @xmath298 .
[ cor211 ] for any partition @xmath212 and any @xmath3-algebra @xmath16 , @xmath220 is a right @xmath1-submodule of @xmath111 .
this follows directly from proposition [ prop210 ] .
[ thm212 ] for any partition @xmath212 and any @xmath3-algebra @xmath16 , the canonical map @xmath367 is an isomorphism , and the set @xmath368 form an @xmath16-basis of @xmath220 .
in particular , the set @xmath369 form an @xmath16-basis of @xmath111 .
we take @xmath370 , then @xmath272={\mathfrak{b}}_n$ ] .
it is well - known that @xmath371 for any @xmath3-algebra @xmath16 . applying proposition [ prop210 ] and counting the dimension
, we get that for any @xmath3-algebra @xmath16 which is a field , the set @xmath369 must form an @xmath16-basis of @xmath111 . since each basis element @xmath372 is integrally defined and @xmath373 for any @xmath3-algebra @xmath16 , it follows that the above set is still an @xmath16-basis of @xmath111 for any @xmath3-algebra @xmath16 . by the @xmath16-linear independence of the elements in this set and corollary [ cor211 ] , we also get that , for any partition @xmath374 , the set @xmath375 must form an @xmath16-basis of @xmath220 . therefore , for any @xmath3-algebra @xmath16 , the canonical map @xmath376 is an isomorphism .
_ remark 2.13 .
_ note that for any partition @xmath212 , @xmath272\subseteq{\mathcal m}^{{\lambda}}$ ] .
but @xmath272 $ ] is not necessarily equal to @xmath377 in general .
for example , one sees easily that @xmath378.\ ] ] in fact , if this is not the case , then we can write @xmath379 for some positive integers @xmath380 and some elements @xmath381 . however , if we express both sides into linear combinations of brauer @xmath105-diagrams and count the number of terms , we find that this is impossible ( as the equation @xmath382 has no solutions in @xmath3 ) .
[ thm214 ] for any partition @xmath212 and any @xmath3-algebra @xmath16 , we define @xmath383 then @xmath384 is a right @xmath112$]-submodule of @xmath385 , and there is a @xmath112$]-module isomorphism @xmath386 in particular , @xmath111 has a specht filtration .
it suffices to consider the case where @xmath387 .
we first show that @xmath388 for each @xmath284 , we use @xmath389 to denote the composition of the embedding @xmath390 and the projection @xmath391 .
suppose that @xmath392 . then @xmath389 must be an surjection .
we claim that @xmath308 .
in fact , if @xmath393 , then for any @xmath394 , @xmath395 , and @xmath396w_{\mu'}x_{\mu}w_{\mu}y_{\mu'}=0 $ ] , hence @xmath397 for any @xmath398 .
it follows that @xmath399 . therefore , @xmath400 . on the other hand , since @xmath401 $ ] , and by ( * ? ? ?
* lemma 5.7 ) , @xmath402}h_{i , j}^{\mu}\bigr ) x_{\mu}w_{\mu'}y_{\mu'}\neq 0,\ ] ] where @xmath403 is the @xmath404-hook length in @xmath151 $ ] , we get a contradiction .
therefore , @xmath392 must imply that @xmath308 . now counting the dimensions
, we deduce that @xmath405 . in a similar way
, we can prove that @xmath406 .
it follows that @xmath407 .
we now consider the natural map from @xmath221 $ ] onto @xmath408 .
since @xmath409 does not contain @xmath292 as a composition factor , it follows that ( by proposition [ prop210 ] ) the image of @xmath410 must be contained in @xmath411 .
therefore we get a surjective map from @xmath412 onto @xmath408 .
this map sends the standard basis of @xmath412 to the canonical basis of @xmath413 .
so it must be injective as well , as required .
in this section , we shall use the results obtained in section 2 and in @xcite to give an explicit and characteristic free description of the annihilator of the @xmath2-tensor space @xmath76 in the brauer algebra @xmath8 .
let @xmath42 be an arbitrary infinite field .
let @xmath43 .
let @xmath6 be a @xmath5-dimensional symplectic @xmath42-vector space .
let @xmath21 be the corresponding symplectic group , acting naturally on @xmath6 , and hence on the @xmath2-tensor space @xmath7 from the left - hand side . as we mentioned in the introduction , this left action on @xmath7 is centralized by the specialized brauer algebra @xmath50 , where @xmath42 is regarded as @xmath3-algebra by sending each integer @xmath51 to @xmath414 .
the brauer algebra @xmath53 acts on the @xmath2-tensor space @xmath7 from the right - hand side .
let @xmath80 be the natural @xmath42-algebra homomorphism @xmath415 the following is one of the main results in @xcite , which generalize earlier results in @xcite , @xcite , @xcite for the case when @xmath77 .
( ( * ? ? ? * ( 1.2 ) ) ) [ lm31 ] let @xmath42 be an arbitrary infinite field .
then @xmath416 and if @xmath90 , then @xmath80 is also injective , and hence an isomorphism . by the discussion in ( * ? ? ? * section 3 ) , @xmath7 is a tilting module over @xmath417 . by (
* ( 4.4 ) ) , the dimension of @xmath418 is independent of the choice of the infinite field @xmath42 . therefore , the dimension of @xmath419 is also independent of the choice of the infinite field @xmath42 .
that is , the dimension of the annihilator of the @xmath2-tensor space @xmath76 in the brauer algebra @xmath53 is independent of the choice of the infinite field @xmath42 .
[ lm32 ] with the notations as above , we have that @xmath420 by lemma [ lm22 ] and lemma [ lm31 ] , it suffices to consider the case where @xmath77 and to show that @xmath421 note that @xmath422 , and @xmath423 therefore , it suffices to show that @xmath424 by the well - known schur - weyl duality between the general linear group @xmath425 and the symmetric group @xmath1 on the @xmath23-tensor space @xmath426 , we know that there is a @xmath427-bimodules decomposition @xmath428 where @xmath429 denotes the irreducible weyl module with highest weight @xmath69 over @xmath430 .
here we identify @xmath69 with @xmath431 , @xmath432 are the fundamental dominant weights of @xmath425 .
it follows that @xmath433 by the branching law ( see ( * ? ? ?
* proposition 2.5.1 ) ) from @xmath425 to @xmath21 , we know that @xmath434 if @xmath435 ; and @xmath436 otherwise .
this proves that @xmath437 as required .
let @xmath196 be two integers such that @xmath438 and @xmath439 is even .
let @xmath440 let @xmath441 . if @xmath442 , we define @xmath443 to be the set of all the brauer @xmath2-diagrams @xmath99 such that for each integer @xmath444 , @xmath99 connects the vertex labeled by @xmath445 with the vertex labeled by @xmath74 for some integer @xmath446 ; and for each integer @xmath445 with @xmath447 , @xmath99 connects the vertex labeled by @xmath448 with the vertex labeled by @xmath449 ; and for each integer @xmath445 with @xmath450 , @xmath99 connects the vertex labeled by @xmath451 with the vertex labeled by @xmath452 .
if @xmath453 , we define @xmath443 to be the set of all the brauer @xmath2-diagrams @xmath99 such that for each integer @xmath454 , @xmath99 connects the vertex labeled by @xmath445 with the vertex labeled by @xmath74 for some integer @xmath455 ; and for each integer @xmath456 , @xmath99 connects the vertex labeled by @xmath448 with the vertex labeled by @xmath449 ; and for each integer @xmath457 , @xmath36 connects the vertex labeled by @xmath458 with the vertex labeled by @xmath459 . [ lm33 ]
let @xmath196 be two integers such that @xmath438 and @xmath439 is even .
suppose that @xmath460 , then @xmath461 the proof of lemma [ lm33 ] is somewhat complicated and will be given in section 4 .
given any two subsets @xmath462 with @xmath463 is even , we set @xmath464 , and @xmath465 let @xmath466 be a fixed permutation of @xmath467 let @xmath468 we define @xmath469 to be the set of all the brauer @xmath2-diagrams @xmath99 such that for each integer @xmath470 , @xmath99 connects the vertex labeled by @xmath445 with a vertex labeled by @xmath74 for some integer @xmath471 , and for each integer @xmath445 with @xmath472 , @xmath99 connects the vertex labeled by @xmath473 with the vertex labeled by @xmath474 .
note that the set @xmath443 we defined before is a special case of the set @xmath475 we defined here .
[ cor34 ] with the notations as above and suppose that @xmath476 , then we have @xmath477 let @xmath478 .
if @xmath479 , then for any brauer diagram @xmath480 , there exist at least @xmath481 bottom horizontal edges between the vertices labeled by the integers in the following set @xmath482 as a result , we deduce that there exist elements @xmath483 , @xmath484 and a brauer diagram @xmath485 , such that 1 . for any integer @xmath51 with @xmath486
, @xmath485 connects the vertex labeled by @xmath487 with the vertex labeled by @xmath186 .
2 . @xmath488 in this case , since both @xmath489 and @xmath490 are invertible , it follows directly from lemma [ lm33 ] that @xmath491 .
if @xmath492 , then for any brauer diagram @xmath480 , there exist at least @xmath493 top horizontal edges between the vertices labeled by the integers in the following set @xmath494 as a result , we deduce that there exist elements @xmath483 , @xmath484 and a brauer diagram @xmath495 , such that 1 . for any integer @xmath51 with @xmath496 , @xmath495 connects the vertex labeled by @xmath487 with the vertex labeled by @xmath186 .
@xmath497 by the same argument as before , we deduce that @xmath491 in this case .
this completes the proof of the corollary .
the following is the main result of this section , which gives an explicit and characteristic free description of the annihilator of the @xmath2-tensor space @xmath7 in the brauer algebra @xmath8 .
[ thm35 ] with the notations as in lemma [ lm31 ] and lemma [ lm32 ] , we have that @xmath498 where @xmath499 , @xmath500 is the right @xmath116$]-module associated to @xmath501 as defined in section 2 . in particular , @xmath92 is a @xmath1-submodule .
it is easy to see that for any partition @xmath284 , @xmath502 if and only if @xmath503 .
therefore , @xmath504 applying lemma [ lm31 ] and lemma [ lm32 ] , we see that to prove this theorem , it suffices to show that @xmath505 .
equivalently , it suffices to show that for any partition @xmath506 satisfying @xmath507 , and any @xmath97 , @xmath508 . by the definition of the element @xmath509 , the action @xmath510 " and the multiplication rule of brauer diagrams , we deduce that @xmath511 where @xmath512 and @xmath513 , and @xmath514 , @xmath515 such that @xmath466 is a permutation of the integers in @xmath516 .
we now apply corollary [ cor34 ] .
it follows immediately that @xmath508 as required .
this completes the proof of the theorem .
[ rem36 ] let @xmath517 be a free @xmath3-module with basis @xmath518 .
for any @xmath3-algebra @xmath16 , we define @xmath519 .
the same formulae ( see section 1 ) define an action of the algebra @xmath8 on @xmath520 , and hence an action of @xmath521 on @xmath522 .
let @xmath523 ( see ( * ? ? ?
* section 2 ) and @xcite ) be the symplectic schur algebra over @xmath16 .
if @xmath16 is a field , then @xmath523 is a quasi - hereditary algebras over @xmath16 , and @xmath522 is a tilting module over @xmath523 . applying ( * ? ? ? * ( 4.4 ) ) , we know that , for any @xmath3-algebra @xmath16 , there is a canonical isomorphism @xmath524 note that @xmath525 by the main result in @xcite , we know that the above inclusion @xmath526 " can be replaced by = " when @xmath527 is an infinite field @xmath42 .
in fact , this is always true for any @xmath3-algebra @xmath16 ( by using some localization argument in commutative algebras ) . as a consequence ,
our theorem [ thm35 ] is also always true if we replace the infinite field @xmath42 by any @xmath3-algebra @xmath16 .
we shall first fix some notations and convention . note that the element @xmath528 in lemma [ lm33 ] actually lies in @xmath4 , we can choose to work inside the brauer algebra @xmath83 in this section .
furthermore , throughout this section , we shall recover our original way of labeling of vertices in each brauer @xmath2-diagram .
that is , the vertices in each row of a brauer @xmath2-diagram will be labeled by the indices @xmath24 from left to right .
this way of labeling is more convenient when we need to express each brauer diagram in terms of the standard generators @xmath529 for @xmath530 and to consider the action of brauer diagrams on the @xmath2-tensor space @xmath7 .
let @xmath531 be an integer with @xmath532 $ ] , where @xmath533 $ ] is the largest non - negative integer not bigger than @xmath534 .
let @xmath535 , where @xmath536 and @xmath537 are considered as partitions of @xmath538 and @xmath539 respectively . in general , a bipartition of @xmath2 is a pair @xmath540 of partitions of numbers @xmath541 and @xmath542 with @xmath543
. the notions of young diagram , bitableaux , etc .
, carry over easily .
let @xmath544 be the standard @xmath306-bitableau in which the numbers @xmath24 appear in order along successive rows of the first component tableau , and then in order along successive rows of the second component tableau .
we define @xmath545 for each partition @xmath546 of @xmath539 , we denote by @xmath547 the set of all the standard @xmath546-tableaux with entries in @xmath548 .
the initial tableau @xmath231 in this case has the numbers @xmath549 in order along successive rows .
again , for each @xmath550 , let @xmath551 be the unique element in @xmath552 with @xmath215 .
let @xmath553 and @xmath554 .
then @xmath555 corresponds to the brauer @xmath2-diagram where the top horizontal edges connect @xmath556 and @xmath557 , the bottom horizontal edges connect @xmath558 and @xmath559 , for @xmath560 , and the vertical edges connects @xmath561 with @xmath562 for @xmath563 .
( ( * ? ? ?
* corollary 3.3 ) ) with the above notations , the set @xmath564 $ , $ { \sigma}\in{\mathfrak s}_{\{2f+1,\cdots , n\}}$ , $ d_1 , d_2\in{\mathfrak d}_{\nu}$,}\\ \text{where $ \nu:=((2^f ) , ( n-2f))$}\\ \end{matrix}\biggr\}.\ ] ] is a basis of brauer algebra @xmath565 , which coincides with the natural basis given by brauer @xmath2-diagrams .
[ rem41 ] intuitively , the action of the brauer @xmath2-diagram @xmath99 on @xmath576 can be thought as follows .
let @xmath577 be the set of all the horizontal edges in the top row of @xmath99 , where @xmath578 for each @xmath445 and @xmath579 .
let @xmath580 be the set of all the horizontal edges in the bottom row of @xmath99 , where @xmath581 for each @xmath445 and @xmath582 .
then for each @xmath573 , the @xmath583th position of @xmath569 is the following sum : @xmath584 we list those vertices in the top row of @xmath99 which are not connected with horizontal edges from left to right as @xmath585 .
then , for each integer @xmath586 with @xmath587 , the @xmath588th position of @xmath569 is @xmath589 .
in fact , for @xmath599 with @xmath600 , it is clear that @xmath601 , and both @xmath602 and @xmath603 are still lie in @xmath592 . now counting the number of brauer @xmath2-diagrams occurring in both sides , we prove ( [ equa41 ] ) and
hence also prove the lemma .
similarly , we fix an arbitrary element @xmath604 and define @xmath605 then @xmath606 consists of all the brauer @xmath2-diagrams whose top horizontal edges are @xmath607 the following result can be proved in the same way as lemma [ lm43 ] .
let @xmath610 , where @xmath611 for each @xmath586 . an ordered pair @xmath612 ( @xmath613 ) is called a _
symplectic pair _ in @xmath614 if @xmath615 .
two ordered pairs @xmath612 and @xmath616 are called disjoint if @xmath617 .
we define the _ symplectic length _ @xmath618 to be the maximal number of disjoint symplectic pairs @xmath612 in @xmath614 .
we now consider a special case of lemma [ lm33 ] . by the above discussion and the definition of @xmath620 , any brauer diagram @xmath621 only acts on the first @xmath622 components of any simple @xmath2-tensor @xmath623 .
therefore , to show that @xmath624 , we can assume without loss of generality that @xmath625 . note that @xmath626}\sum_{d_{2}\in{\mathfrak d}_{{\nu\!_{f}}}}\sum_{d\in\operatorname{bd}^{(f)}(n;d_2)}d\\ & = \sum_{0\leq f\leq [ n/2]}\sum_{d_{1}\in{\mathfrak d}_{{\nu\!_{f}}}}\sum_{d\in\operatorname{bd}^{(f)}(d_1;n)}d.\end{aligned}\ ] ] suppose that @xmath627 .
then there exists a simple @xmath2-tensor @xmath628 , such that @xmath629 let @xmath630 .
suppose that @xmath631 for some integer @xmath532 $ ] . by lemma [ lm43 ]
, we have that for any @xmath632 , @xmath633 therefore , we can assume without loss of generality that @xmath634 for each integer @xmath635 , @xmath636 and @xmath637 .
furthermore , if @xmath638 for some integers @xmath639 , then @xmath640 where @xmath641 denotes the transposition @xmath642 in @xmath124 , and we have used the fact that the length of @xmath641 is an odd integer . therefore , we can deduce that @xmath643 are pairwise distinct .
hence , @xmath634 for each integer @xmath635 , @xmath644 and @xmath645 . by lemma [ lm44 ] ,
it is easy to see that for any @xmath647 , @xmath648 in particular , by the same argument as before , any simple @xmath2-tensor @xmath649 with @xmath650 for some integers @xmath639 can not appear with nonzero coefficient in the expansion of the left - hand side of ( [ equa44 ] ) . therefore if a simple @xmath2-tensor @xmath651 appears with nonzero coefficient in the expansion of the left - hand side of ( [ equa44 ] ) , then @xmath652 must be pairwise distinct .
applying lemma [ lm43 ] and lemma [ lm44 ] again , we can further assume ( without loss of generality ) that there exists such a @xmath651 which appears with nonzero coefficient in the expansion of the left - hand side of ( [ equa44 ] ) , such that @xmath652 are pairwise distinct , @xmath653 , @xmath654 for each integer @xmath635 , @xmath655 , and there exists integer @xmath656 such that @xmath657 for each integer @xmath658 , and @xmath659 .
in fact , for any brauer diagram @xmath99 satisfying the above conditions 1 ) , 2 ) , 3 ) , by remark [ rem41 ] , @xmath651 does appear with nonzero coefficient in the expansion of @xmath665 , the coefficient is @xmath672 ; on the other hand , suppose that @xmath651 appears with nonzero coefficient in the expansion of @xmath665 . by our assumption on @xmath614 and @xmath673 ,
it is easy to see that the tensor factor @xmath674 with @xmath675 can only be produced through the action of @xmath676 for some @xmath677 .
this implies that @xmath666 . for each integer
@xmath586 with @xmath678 , by remark [ rem41 ] , the action of @xmath99 on @xmath576 move the vector in the @xmath679th position of @xmath576 ( i.e. , @xmath680 ) to the @xmath588th position . by our assumption on @xmath681 again , we deduce that @xmath682 .
but by the definition of @xmath683 , @xmath684 it follows that @xmath667 , and @xmath685 for any @xmath686 .
now the remaining statements of our claim follows easily from the fact that @xmath667 , our assumption on @xmath614 and @xmath673 as well as remark [ rem41 ] .
therefore , the coefficient of @xmath651 in the expansion of @xmath687 is equal to @xmath688 hence the coefficient of @xmath651 in the left - hand side of ( [ equa44 ] ) is @xmath689 a contradiction .
this completes the proof of the proposition . by the assumption that @xmath690 and the definition of @xmath443 , any brauer diagram @xmath691 only acts on the first @xmath51 components of any simple @xmath2-tensor in @xmath7
therefore , we can assume without loss of generality that @xmath692 .
also , because of proposition [ prop41 ] , we can assume that @xmath693 .
in particular , @xmath694 .
suppose that @xmath695 .
then there exists a simple @xmath2-tensor @xmath696 such that let @xmath700 .
we define @xmath701 for each @xmath702 , we define @xmath703 to be the set of all the brauer @xmath2-diagrams in @xmath704 whose rightmost @xmath531 horizontal edges in the top row are exactly those pairs in @xmath705 .
it is clear that @xmath706 therefore , @xmath707 for some @xmath702 .
henceforth , we fix such an @xmath705 .
we list the integers in the set @xmath708 as @xmath709 such that @xmath710 .
we define @xmath711 for each element @xmath712 , we define @xmath713 to be the set of all the brauer @xmath2-diagrams in @xmath703 whose vertical edges connects the vertex labeled by @xmath714 in the top row with the vertex labeled by @xmath715 in the bottom row for each integer @xmath445 with @xmath716 .
it is clear that @xmath717 now let @xmath718 be such that @xmath719 appears with nonzero coefficient in the expansion of ( [ equa47 ] ) . using the same argument in the proof of lemma [ lm44 ] , one can show that for any @xmath720 , @xmath721 as a result , we deduce that @xmath722 are pairwise distinct .
furthermore , we can assume ( if necessary , we replace @xmath576 by @xmath723 for some @xmath595 ) that @xmath724 it follows that @xmath719 appears with nonzero coefficient in the expansion of @xmath725 where @xmath726 .
we divide the remaining proof into two cases : _ case 1 .
_ @xmath727 .
then it follows from ( [ equa48 ] ) and our assumption on @xmath614 that @xmath728 , and the vectors @xmath729 lie in a subspace with dimension @xmath730 hence @xmath731 for some @xmath732 .
but ( [ equa48 ] ) implies that @xmath733 a contradiction , as required .
_ case 2 . _
@xmath734 . let @xmath735 be the symplectic subspace generated by @xmath736
. then ( [ equa48 ] ) implies that @xmath737 .
hence @xmath738 .
note that if @xmath739 for some @xmath740 , then we must have that @xmath615 for some integer @xmath74 with @xmath741 .
furthermore , in this case , if we replace the tensor factors @xmath742 in @xmath576 by @xmath743 respectively , then the coefficient of @xmath719 in ( [ equa48 ] ) changes at most one sign .
therefore , ( [ equa48 ] ) implies that there exists a simple @xmath744-tensor @xmath745 in @xmath746 such that @xmath747 appears with non - zero coefficient in @xmath748 where the above element @xmath99 is understood as element in the brauer algebra @xmath749 , acting on the @xmath744-tensor space @xmath746 .
this is impossible by proposition [ prop41 ] .
hence we complete the proof of this proposition . by the assumption that @xmath750 and the definition of @xmath443 , any brauer diagram @xmath691 only acts on the first @xmath751 components of any simple @xmath2-tensor in @xmath7 .
therefore , we can assume without loss of generality that @xmath752 .
also , because of proposition [ prop41 ] , we can assume that @xmath753 .
in particular , @xmath754 .
suppose that @xmath755 .
then there exists a simple @xmath2-tensor @xmath696 such that it follows that @xmath757 for each integer @xmath445 with @xmath758 . in particular , @xmath759 . using the same argument in the proof of lemma [ lm44 ]
, one can show that for any @xmath760 , @xmath761 as a result , we deduce that @xmath762 are pairwise different .
let @xmath763 .
we define @xmath701 for each @xmath702 , we define @xmath764 to be the set of all the brauer @xmath2-diagrams in @xmath765 whose rightmost @xmath531 horizontal edges in the bottom row are exactly those pairs in @xmath705 .
it is clear that @xmath766 therefore , @xmath767 for some @xmath702 .
henceforth , we fix such an @xmath705 .
we list the integers in the set
@xmath708 as @xmath768 such that @xmath769 .
we define @xmath770 for each element @xmath771 , we define @xmath772 to be the set of all the brauer @xmath2-diagrams in @xmath764 whose vertical edges connects the vertex labeled by @xmath714 in the bottom row with the vertex labeled by @xmath715 in the top row for each integer @xmath445 with @xmath773 .
it is clear that @xmath774 now let @xmath718 be such that @xmath719 appears with nonzero coefficient in the expansion of ( [ equa47 ] ) . applying lemma [ lm44 ] , we know that for any @xmath632 , @xmath775 as a result , we deduce that @xmath776 are pairwise distinct .
furthermore , we can assume ( if necessary , we replace @xmath719 by @xmath777 for some @xmath632 ) that @xmath778 it follows that @xmath719 appears with nonzero coefficient in the expansion of @xmath779 where @xmath780 .
we divide the remaining proof into two cases : _ case 1 . _
. then it follows from ( [ equa412 ] ) and our results on @xmath614 and @xmath782 that @xmath728 , and the vectors @xmath729 lie in a subspace with dimension @xmath783 hence @xmath731 for some @xmath732 . but
( [ equa412 ] ) implies that @xmath733 a contradiction , as required .
_ case 2 . _
@xmath784 . let @xmath735 be the symplectic subspace generated by @xmath736
. note that @xmath785 .
it follows that @xmath737 .
hence @xmath738 .
note that if @xmath739 for some @xmath740 , then we must have that @xmath615 for some integer @xmath74 with @xmath741 .
furthermore , in this case , if we replace the tensor factors @xmath742 in @xmath576 by @xmath743 respectively , then the coefficient of @xmath719 in ( [ equa412 ] ) changes at most one sign .
therefore , ( [ equa412 ] ) implies that there exists a simple @xmath744-tensor @xmath745 in @xmath746 such that @xmath786 appears with non - zero coefficient in @xmath748 where the above element @xmath99 is understood as element in the brauer algebra @xmath749 , acting on the @xmath744-tensor space @xmath746 .
this is impossible by proposition [ prop41 ] .
hence we complete the proof of this proposition .
g. benkart , m. chakrabarti , t. halverson , r. leduc , c. lee , j. stroomer , tensor product representations of general linear groups and their connections with brauer algebras , _ j. algebra _ * 166 * ( 1994 ) , 529567 . | let @xmath0 .
in this paper we study the right permutation action of the symmetric group @xmath1 on the set of all the brauer @xmath2-diagrams .
a new basis for the free @xmath3-module @xmath4 spanned by these brauer @xmath2-diagrams is constructed , which yields specht filtrations for @xmath4 . for any @xmath5-dimensional vector space @xmath6 over a field of arbitrary characteristic
, we give an explicit and characteristic free description of the annihilator of the @xmath2-tensor space @xmath7 in the brauer algebra @xmath8 .
in particular , we show that it is a @xmath1-submodule of @xmath8 . |
of the 151 nearby stars known to harbor one or more planets , 19 are well - characterized multiple - planet systems , and an additional 24 show radial velocity ( rv ) residuals indicative of additional companions @xcite .
for instance , @xcite reported additional companions around five stars , including two revealed by incomplete orbits apparent in the rv residuals ( hd 50499 and hd 217107 ) , and one as a short - period , low amplitude variation in the residuals of the fit to a long - period outer companion .
@xcite detected a 7.5 earth - mass mass companion to gj 876 in a 2-day period through analysis of the rv residuals to a 2-planet dynamical fit of the more massive , outer exoplanets .
@xcite similarly analyzed the rv residuals of 4 stars to search for neptune - mass companions .
very little is known about the frequency or nature of exoplanets with orbital distances greater than 5 au @xcite .
precise radial velocities have only reached the precision required to detect such objects within the last 10 years @xcite , which is less than the orbital period of such objects ( @xmath6 y for exoplanets orbiting solar mass stars ) .
thus , the rv curves for such planets are all necessarily incomplete , and we must obtain many more years of data before our knowledge of their orbits improves significantly . the ability to put constraints on planets with incomplete orbits , however weak , allows us to peek beyond the 5 au completeness limit inherent in the ten - year - old planet searches .
characterizing incomplete orbits also increases our sample of known multiple exoplanetary systems , which improves our understanding of the frequency of orbital resonances , the growth of multiple planets , and the mechanics of orbital migration . in this work
, we present our analysis of the rv data of @xcite in an effort to determine which of those systems have additional , low - amplitude companions .
many systems known to host one exoplanet show more distant , long - period companions with highly significant but incomplete orbits . in these systems , it can be extremely difficult to constrain the properties of the outer companion : in the case of a simple trend with no curvature , very little can be said about the nature of these companions beyond their existence , but even this informs studies of exoplanet multiplicity and the frequency of exoplanets in binary systems . in [ hip14810 ] we discuss a new multiple planet system from the n2k project , hip 14810 . in [ trends ] we describe how we have employed a false alarm probability statistic to test the significance of trends in the rv data of stars already known to host exoplanets .
we find that six stars known to host exoplanets have previously undetected trends , and thus additional companions . when the rv residuals to a single keplerian show significant curvature , one may be able to place additional constraints on the maximum @xmath7 of the additional companion . in
[ mapsection][twonew ] we present our analysis of this problem in the cases of hd 24040 b and hd 154345 b , two substellar companions new to this work with very incomplete orbits . by mapping @xmath8 space for keplerian fits ,
we show that hd 154345 b is almost certainly planetary ( @xmath9 ) , and that hd 24040 b may be planetary ( @xmath10 ) . in [ mining ]
we describe how we extended this method to the rv residuals of known planet - bearing stars which show trends .
we find that for 2 stars we can place sufficiently strong upper limits on @xmath11to suggest that the additional companions are planetary in nature .
hip 14810 is a metal - rich ( @xmath12}}= 0.23 $ ] ) g5 v , v=8.5 star which we have observed at keck observatory as part of the n2k program @xcite since nov 2005
. table [ rvhip14810 ] contains the rv data for this star .
its stellar characteristics are listed in table [ starchar ] , determined using the same lte spectral analysis used for stars in the spocs catalog @xcite .
we quickly detected a short - period , high - amplitude companion ( @xmath13 d , @xmath14 ) and a strong , @xmath15 m / s trend .
further observations revealed evidence for substantial curvature in the residuals to a planet plus trend fit .
fig.[hip14810fig ] shows the rv curve for this star decomposed into keplerian curves for the b and c components , and table [ orbitupdates ] contains the best - fit double keplerian elements .
lrc 13693.760579 & -130.8 & 1.3 + 13694.831481 & -473.6 & 1.2 + 13695.909225 & -226.9 & 1.2 + 13723.786250 & 162.6 & 1.0 + 13724.688484 & 324.9 & 1.2 + 13746.814595 & 2.4 & 1.3 + 13747.852940 & -435.82 & 0.94 + 13748.734190 & -433.3 & 1.2 + 13749.739236 & -71.4 & 1.2 + 13751.898252 & 358.3 & 1.1 + 13752.807431 & 241.05 & 0.80 + 13752.912477 & 211.6 & 1.7 + 13753.691574 & -79.8 & 1.1 + 13753.810359 & -137.6 & 1.2 + 13753.901042 & -180.2 & 1.2 + 13775.836157 & -240.01 & 0.97 + 13776.812859 & 123.4 & 1.4 + 13777.723102 & 346.7 & 1.3 + 13778.720799 & 416.0 & 1.3 + 13779.744410 & 238.4 & 1.3 + 13841.722049 & -515.7 & 1.4 + 13961.130301 & -280.9 & 1.0 + 13962.133333 & -413.4 & 1.1 + 13969.097315 & -348.3 & 1.2 + 13981.969815 & -476.5 & 1.2 + 13982.947431 & -200.9 & 1.2 + 13983.981470 & 151.5 & 1.0 + 13984.096979 & 187.3 & 1.2 + 13984.985775 & 345.7 & 1.3 + 13985.102106 & 357.6 & 1.3 a 2-planet keplerian fit yields an outer planet with @xmath16 , @xmath17 d , and eccentricity of 0.27 .
we present the orbital solutions for this two - planet system in table [ orbitupdates ] .
very long period substellar companions appear in radial velocity data first as linear trends ( constant accelerations ) , then as trends with curvature , and finally , as the duration of the observations becomes a substantial fraction of the orbital period , as recognizable portions of a keplerian velocity curve .
it is important , then , to have a statistically robust test for trends in velocity residuals . in this section
, we discuss calculating false alarm probabilities ( faps ) for such trends .
( 5.2 * ? ? ?
* ) present a detailed discussion of using false alarm probabilities ( faps ) for determining the significance of a periodic signal in an rv time series . here
, our task is similar .
we wish to test the hypothesis that a star has an additional companion with a long period , manifest only as a linear trend in the rv series .
we compare this hypothesis to the null hypothesis that the data are adequately described only by the best - fit keplerians and noise .
we first fit the data set with a keplerian model and compare the @xmath18 statistic to that of a model employing a keplerian plus a linear trend .
if this statistic improves , that is , if @xmath19 is negative , then the inclusion of the trend may be justified . to test the significance of the reduction in @xmath18
, we employ an fap test .
we first employ a bootstrap method to determine our measurement uncertainties .
we subtract the best - fit keplerian rv curve from the data and assume the null hypothesis namely that the residuals to this fit are properly characterized as noise and thus approximate the underlying probability distribution function of the noise in the measurements .
we then draw from this set of residuals ( with replacement ) a mock set of residuals with the same temporal spacing as the original set . by adding these mock residuals to the best - fit keplerian rv curve we produce a mock data set with the same temporal sampling as the original data set , but with the velocity residuals `` scrambled '' ( `` re - drawn '' might be a better term since we have drawn residuals _ with _ replacement . )
it is important in this procedure that internal errors remain associated with the scrambled residuals .
this ensures that points with error bars so large that they contribute little to the @xmath18 sum , but nonetheless lie far from the best - fit curve , do not gain significance when `` scrambled '' , inappropriately increasing @xmath18 .
we then compare @xmath20 for our mock data set to that of our genuine data . by repeating this procedure 400 times
, we produce 400 mock sets of residuals and 400 values for @xmath20 .
if the linear trend is simply an artifact of the noise , then re - drawing the residuals should not systematically improve or worsen @xmath20 .
conversely , if a linear trend is significant , then the null hypothesis , that the residuals to a keplerian are uncorrelated noise , is invalid , and re - drawing them should worsen the quality of the keplerian(s ) plus trend fit , since scrambling will remove evidence of the trend .
thus , the fraction of these sets with @xmath20 less than that of the proper , unscrambled residuals , provides a measurement of the false alarm probability that the residuals to a keplerian - only fit are correlated .
the catalog of nearby exoplanets @xcite contains 172 substellar companions with @xmath21 orbiting 148 stars within 200 pc .
since then , at least 3 more systems have been announced , including a triple - neptune @xcite , and two single - planet detections @xcite , @xcite . of these 151 systems ,
24 show significant trends in addition to the keplerian curves of the known exoplanets .
we have reanalyzed the radial velocities of @xcite to determine the significance of these trends and to find evidence for additional trends using the fap test described in [ faptrend ] .
note that we have obtained additional data for some of these systems since @xcite went to press .
we confirm here 21 of the 24 trends in @xcite to have faps below 1% ( 2 others are in systems on which we have no data to test , and the third is hd 11964 , discussed in [ lowamps ] ) .
we also confirm the trend in the 14 her system , first announced in @xcite and analyzed more thoroughly in @xcite , and [ incompletes ] .
we confirm the finding of @xcite that the trend reported in @xcite for hd 45350 b is not significant ( fap=0.6 and @xmath18 _ increases _ with the introduction of a trend ) .
we announce here the detection of statistically significant linear trends ( fap @xmath22 ) around 4 stars already known to harbor a single exoplanet : hd 83443 , gj 436 (= hip 57087 ) , hd 102117 , and hd 195019 .
gj 436 will be discussed more thoroughly in an upcoming work , ( maness et al .
, 2006 , submitted ) ) .
in one additional case , hd 168443 , we detect a radial velocity trend with fap @xmath23 in a system already known to have two exoplanets , indicating that a third , long - period companion may exist .
we present the updated orbital solutions in table [ orbitupdates ] .
hd 49674 has an fap for an additional trend of @xmath24 , which is of borderline significance when we account for the size of our sample : we should expect that around 2 of our 100 systems will prove to have faps @xmath24 purely by chance , and not because of an additional companion .
we include the fit for hd 49674 with a trend in table [ orbitupdates ] , but note here the weakness of the detection .
llrrcccllrcllrc 24040 & 17960 & 03 50 22.968 & + 17 28 34.92 & 0.65 & 7.50 & 46.5(2.2 ) & 5853(44 ) & 4.361(70 ) & 0.206(30 ) & 2.39(50 ) & 1.18 & 0.15 & 0.65 & 5.7 + 154345 & 83389 & 17 02 36.404 & + 47 04 54.77 & 0.73 & 6.76 & 18.06(18 ) & 5468(44 ) & 4.537(70 ) & -0.105(30 ) & 1.21(50 ) & 0.88 & 0.18 & -0.21 & 5.7 + & 14810 &
03 11 14.230 & + 21 05 50.49 & 0.78 & 8.52 & 52.9(4.1 ) & 5485(44 ) & 4.300(70 ) & 0.231(30 ) & 0.50(50 ) & 0.99 & 0.16 & 0.64 & 3.5 + l@ rccccccccccr hip 14810 & b & 6.6742(20 ) & 428.3(3.0 ) & 0.1470(60 ) & 158.6(2.0 ) & 13694.588(40 ) & & 3.91(55 ) & 0.0692(40 ) & 5.1 & 1.4 & 30 + hip 14810 & c & 95.2914(20 ) & 37.4(3.0 ) & 0.4088(60 ) & 354.2(2.0 ) & 13679.575(40 ) & & 0.76(12 ) & 0.407(23 ) & 3.3 & 0.77 & 21 + hd 49674 & b & 4.9437(23 ) & 13.7(2.1 ) & 0.29(15 ) & 283 & 11882.90(86 ) & 2.6(1.1 ) & 0.115(16 ) & 0.0580(33 ) & 4.4 & 0.62 & 39 + hd 83443 & b & 2.985625(60 ) & 56.4(1.4 ) & 0.008(25 ) & 24 & 11211.04(82 ) & 2.40(79 ) & 0.400(34 ) & 0.0406(23 ) & 8.2 & 0.93 & 51 + gj 436 & b & 2.643859(74 ) & 18.35(80 ) & 0.145(52 ) & 353(24 ) & 11551.72(12 ) & 1.42(35 ) & 0.0682(63 ) & 0.0278(16 ) & 4.2 & 0.93 & 60 + hd 102117 & b & 20.8079(55 ) & 11.91(77 ) & 0.106(70 ) & 283 & 10942.9(3.0 ) & -0.91(26 ) & 0.172(18 ) & 0.1532(88 ) & 3.3 & 0.83 & 45 + hd 168443 & b & 58.11289(86 ) & 475.9(1.6 ) & 0.5286(32 ) & 172.87(94 ) & 10047.387(34 ) & & 8.02(65 ) & 0.300(17 ) & 4.1 & 0.97 & 109 + hd 168443 & c & 1749.5(2.4 ) & 298.0(1.2 ) & 0.2125(15 ) & 65.07(21 ) & 10273.0(4.6 ) & & 18.1(1.5 ) & 2.91(17 ) & 4.1 & 0.97 & 109 + hd 195019 & b & 18.20163(40 ) & 272.3(1.4 ) & 0.0140(44 ) & 222(20 ) & 11015.0(1.2 ) & 1.31(51 ) & 3.70(30 ) & 0.1388(80 ) & 16 & 1.5 & 154 +
it is difficult to properly characterize the orbit of an exoplanet when the data do not span at least one complete revolution .
after one witnesses a complete orbit of the planet in a single - planet system , subsequent orbits should have exactly the same shape ( absent strong planet - planet interactions ) , and so one can interpret deviations as the effects of an additional companion . before witnessing one complete orbit , one can easily misinterpret the signature of an additional companion as it is absorbed into the orbital solution for the primary companion .
even when only one planet is present , small portions of single keplerian curves can easily mimic portions of other keplerians with very different orbital elements .
when an rv curve shows significant curvature , it may be possible to constrain the minimum mass ( @xmath11 ) and orbital period of the companion .
@xcite discussed the problem extensively , and @xcite studied the significance of non - detections in the mcdonald observatory planet search without assuming circular orbits by injecting artificial rv signals into program data to determine the strength of a just - recoverable signal .
@xcite reasonably assigned a broad range of eccentricities , @xmath26 , with an upper limit they justified by the fact that over 90% of all known exoplanets have @xmath27 @xcite .
the presence of this upper limit greatly limits the number of pathological solutions to a given rv set .
below , we explore the nature of limits on mass and period implied by a given data set and how constraining @xmath28 can improve those limits . since @xmath11 ,
not @xmath29 , is the astrophysically interesting quantity in exoplanet detection , it is useful to transform into @xmath25 , @xmath28 , @xmath11coordinates when considering constraints .
the minimum mass of a companion can be calculated from the _ mass function _
, @xmath30 , and the stellar mass , according to the relation : @xmath31 where , in the minimum mass case ( where @xmath32 ) , we set @xmath33 equal to @xmath11 .
this relation allows us to fit for the minimum mass ( which we refer to as @xmath11 for brevity ) eliminating the orbital parameter @xmath29 . using eq .
[ massfn ] , we can find the best - fit keplerian rv curve across @xmath34 space , allowing @xmath28 , @xmath35 , and @xmath36 ( the rv zero point ) to vary at many fixed values of @xmath25 and @xmath11 to map @xmath8 .
here we consider hd 24040 , a metal - rich ( @xmath12}}= 0.21 $ ] ) g0 v star at 46 pc ( stellar characteristics summarized in table [ starchar ] ) .
this star shows rv variations consistent with a planetary companion with @xmath37 y and @xmath38 ( figure [ 24040 ] ) , although longer orbital periods and minimum masses as high as @xmath39 can not be ruled out .
the rv data for hd 24040 appear in table [ rvhd24040 ] . here and
in [ incompletes ] , we use @xmath8 as a merit function and infer parameters for acceptable fits from increases of this function by 1 , 4 and 9 , which correspond to 1- , 2- , and 3-sigma confidence levels for systems with gaussian noise . because stellar jitter provides a source of pseudo - random noise which may vary on a stellar rotation timescale
, the noise in rv residuals may be non - gaussian .
thus , to the degree that the rv residuals are non - gauassian , the translation of these confidence limits into precise probabilities is not straightforward .
lrc 10838.773206 & -37.1 & 1.4 + 11043.119653 & -25.8 & 1.5 + 11072.039039 & -24.3 & 1.4 + 11073.002315 & -26.8 & 1.2 + 11170.876921 & -9.9 & 1.4 + 11411.092975 & 3.6 & 1.6 + 11550.824005 & 16.4 & 1.3 + 11551.863449 & 16.5 & 1.2 + 11793.136725 & 42.2 & 1.2 + 11899.945741 & 50.7 & 1.1 + 12516.065405 & 74.4 & 1.3 + 12575.951921 & 87.6 & 1.5 + 12854.115278 & 81.4 & 1.2 + 12856.115671 & 82.1 & 1.1 + 13071.741674 & 59.4 & 1.2 + 13072.799190 & 55.6 & 1.3 + 13196.130000 & 54.2 & 1.3 + 13207.120116 & 57.3 & 1.3 + 13208.125625 & 54.0 & 1.2 + 13241.091852 & 57.5 & 1.2 + 13302.947025 & 49.9 & 1.3 + 13339.945972 & 46.1 & 1.2 + 13368.865139 & 48.4 & 1.2 + 13426.809792 & 36.5 & 1.1 + 13696.904468 & 18.0 & 1.1 + 13982.061586 & 6.8 & 1.2 + in this case , we have enough rv information to put an upper limit on @xmath11 .
figure [ 24040_contour ] shows @xmath8 for best - fit orbits in the @xmath34 plan .
fits with @xmath25 as low as 10 y and @xmath11 as low as 5 @xmath0 are allowed .
interestingly , the data ( following the middle , @xmath40 contour ) exclude orbits with @xmath41 , providing a `` maximum minimum - mass '' . since without an assumption for the eccentricity ( which we will make below ) , we can not exclude orbits with @xmath11 as high as 30 @xmath0 , there is a chance that this companion to hd 24040 is a brown dwarf , or even stellar .
a similar case is hd 154345 , a g8 v star at 18pc ( stellar characteristics summarized in table [ starchar ] ) .
this star shows rv variations remarkably similar to those of hd 24040 ( figure [ 154345 ] ) , but with an amplitude about 6 times smaller . in this case , the maximum @xmath11 is only around 10 @xmath0 , giving us confidence that this object is likely a true exoplanet , and masses as low as 1 @xmath0 are allowed .
the rv data for hd 154345 are in table [ rvhd154345 ] .
we summarize the orbital constraints for these objects in table [ masslimits ] .
lrc 10547.110035 & 6.8 & 1.4 + 10603.955845 & 8.6 & 1.4 + 10956.015625 & 15.1 & 1.5 + 10982.963634 & 13.1 & 1.4 + 11013.868657 & 16.3 & 1.5 + 11311.065486 & 16.6 & 1.6 + 11368.789491 & 18.7 & 1.5 + 11441.713877 & 23.0 & 1.4 + 11705.917836 & 30.3 & 1.5 + 12003.078183 & 30.4 & 2.3 + 12098.916539 & 37.1 & 1.5 + 12128.797813 & 34.7 & 1.7 + 12333.173299 & 38.1 & 1.6 + 12487.860197 & 35.5 & 1.6 + 12776.985463 & 29.9 & 1.6 + 12806.951852 & 19.5 & 1.6 + 12833.801030 & 27.3 & 1.4 + 12848.772037 & 25.3 & 1.5 + 12897.776562 & 26.5 & 1.5 + 13072.046921 & 18.9 & 1.6 + 13074.077766 & 21.2 & 1.4 + 13077.128090 & 20.5 & 1.4 + 13153.943171 & 15.2 & 1.6 + 13179.992454 & 20.6 & 1.5 + 13195.819190 & 17.5 & 1.4 + 13428.162502 & 10.09 & 0.78 + 13547.914433 & 11.44 & 0.80 + 13604.829999 & 6.08 & 0.78 + 13777.155347 & 7.5 & 1.5 + 13807.077257 & 2.4 & 1.4 + 13931.955714 & 1.33 & 0.72 + 13932.913019 & 1.98 & 0.70 + cccc hd 24040 b & 10 100 & 5 20 & 5 23 + hd 68988 c & 11 60 & 11 20 & 5 7 + hd 154345 b & 7 100 & 0.8 10 & 4 25 + hd 187123 c & 10 40 & 2 5 & 5 12 + we can put more stringent constraints on these orbits by noting that ninety percent of all known exoplanets have @xmath27 @xcite . for both hd 24040 b and hd 154345 b ,
the high - period solutions all have high eccentricities ( the dashed contours in figs . [ 24040_contour ] and [ 154345_contour ] ) .
if , following @xcite , we therefore assume that @xmath27 for these objects , and that the true values of @xmath11 and @xmath25 lie within the limits of the middle ( @xmath42 ) contour , then we can constrain @xmath43 and @xmath44 y for hd 24040 b , and @xmath45 and @xmath46 y for hd 154345 b.
for exoplanetary systems in which an additional , low - amplitude signal is not well - characterized by just a linear trend for instance , where there is significant curvature ( e.g. hd 13445 or hd 68988 ) or even multiple orbits ( e.g. gj 876 ) a full , multi - planet fit is needed to properly characterize the system . in this case , we can apply an fap analysis similar to the one in
[ faptrend ] testing the @xmath47-planet hypothesis versus the null hypothesis of @xmath48 planets plus a trend plus noise , where @xmath48 is the number of previously confirmed planets .
this is a much more computationally intensive procedure than that of [ faptrend ] , since we are introducing 5 new , non - linear , highly covariant parameters ( @xmath49 , and @xmath29 ) , so we have performed only 50100 trials . in most cases
the low - amplitude signal we seek is much weaker than that of the known planet(s ) .
this means we have good initial guesses for the orbital parameters of the established exoplanets , and that those parameters are usually rather insensitive to those of the additional companion , easing the difficulty of the simultaneous 11-parameter fit ( 16-parameter for existing double systems ) . as in [ faptrend ] , we calculate the improvement in the goodness - of - fit parameter , @xmath50 with the introduction of an additional exoplanetary companion compared to a fit including only an additional trend .
we compare this reduction to that of mock data sets bootstrapped as the sum of the best - fit solution with a trend , plus noise , drawn , with replacement , from the residuals of the actual data to this fit .
we then construct an fap as the fraction of mock sets that saw a greater reduction in @xmath18 with the introduction of an additional planet than the genuine data set .
a low fap for the presence of a second planet is not tantamount to the detection of an additional exoplanet .
it is only a sign that the null hypothesis is unlikely , i.e. that the distribution of residuals is not representative of the actual noise in the system or that the presumed orbital solution from which the residuals were drawn is in error .
this would be the case if , for instance , if the residuals are correlated due to non - keplerian rv variations ( such as systematic errors or astrophysical jitter ) .
the fits discussed here are purely keplerian and not dynamical .
in particular , fits which produce unstable or unphysical orbits are allowed .
more sophisticated , newtonian fits ( e.g. * ? ? ?
* ) would better constrain the orbits of multiple planet systems .
we have identified 8 other systems in which the fap for an additional keplerian vs. a simple trend is below 2% : hd 142 , hd 13445 , hd 68988 , 23 lib (= hd 134987 ) , 14 her , @xmath51 boo (= hd 120136 ) , hd 183263 , and hd 187123 .
in addition , we have identified a ninth system , hd 114783 , which has a compelling second keplerian despite a slightly larger fap ( 6% ) .
we summarize the orbital constraints for these objects in table [ masslimits ] .
hd 142 : most of the rv data for hd 142 show a simple linear trend superimposed on the known @xmath52 m / s , 350-d orbit @xcite .
hd 142 is known to have a stellar companion ( @xmath53 ) @xcite , which could explain the trend .
the first two data points , taken in 1998 - 9 , are significantly low , producing a low fap for curvature ( @xmath23 ) .
hd 142 has = 0.52 , indicating it is a late f or early g star , suggesting it may have moderate jitter ( @xmath54 m@xmath55 , @xcite ) , so we therefore view the low fap for curvature , apparently based on only two low points , with suspicion . if the curvature is real , it is consistent with an exoplanet with period longer than the span of the observations ( @xmath56 y ) with a minimum mass of at least 4 @xmath0 .
hd 13445 (= gl 86 ) has a known planet with @xmath57 d @xcite.superimposed on that keplerian velocity curve is a velocity trend of roughly -94 m / s / yr during the past 9 years , apparently consistent with the brown dwarf companion previously reported by @xcite .
stellar or brown - dwarf companion @xcite .
there is a a hint of curvature in these residuals to the inner planet , but not enough to put meaningful constraints on this outer object beyond that fact that its period is longer than the span of the observations ( @xmath58 y ) and @xmath59 @xmath0 .
hd 68988 shows definite signs of curvature in the residuals to the 1.8 @xmath0 inner planet ( as fig . [
68988 ] shows ) .
[ 68988_contour ] shows the outer companion has @xmath60 , and the assumption of @xmath27 , using the middle contour , further restricts @xmath4 , and @xmath61 .
hd 114783 shows curvature in its residuals , and may have experienced both an rv minimum ( in 2000 ) and maximum ( in 2006 ) as the rv curve in fig . [ 114783 ] shows .
the data are only moderately inconsistent with a linear trend , however ( fap = 6% ) , indicating that the outer companion s orbit is still underconstrained .
23 lib (= hd 134987 ) shows signs of curvature in the residuals to the known inner planet .
the signal appears as a change in the level of otherwise flat residuals between 2000 and 2002 of 15 m / s ( see fig . [ 134987 ] ) .
this suggests an outer planet on a rather eccentric orbit which reached periastron in 2001 .
the small magnitude of this change in rv suggests a low - mass object , but the incomplete nature of this orbit makes us less than certain that it is due to an exoplanet .
14 her (= hd 145645 ) : this star has a known trend @xcite and has been analyzed by @xcite as a possible resonant multiple system .
the previously - known planet has @xmath62 @xmath0 and @xmath63 y , but the character of the second companion is uncertain . combining our data with the published elodie data from the geneva planet search @xcite ( fig .
[ 14her ] ) provides a good picture of the system .
the character of the orbit of the outer planet is unconstrained , and several equally acceptable but qualitatively distinct solutions exist .
one is a long period , nearly circular orbit like the one shown in fig .
10 and a mass near 2 @xmath0 .
other solutions include a 3:1 resonance with the inner planet .
the next few years of observation should break this degeneracy .
the degeneracy may also be broken by high contrast , high resolution imaging , and we suggest that such attempts be made on this interesting system . hd 183263 shows definite signs of curvature in the residuals to the known inner planet ( as fig . [ 183263 ] shows ) , but too little to constrain the mass of the distant companion . fig .
[ 183263 ] shows that there is little meaningful constraint on the orbit beyond @xmath64 y and @xmath65 .
even the assumption of @xmath27 allows for @xmath66 , so the planetary nature of the companion is very uncertain .
@xmath51 boo (= hd 120136 ) has residuals to the fit for the known inner planet which show evidence of a long - period companion which have been discussed elsewhere @xcite .
analysis of the distant companion is complicated by the lower quality of the data during the apparent periastron in 1990 .
the current best fit suggests a period greater than 15 years , but is otherwise unconstrained .
@xcite report that @xmath51 boo has a faint ( v=10.3 ) companion ( sep .
5.4 ) which may be the source of the rv residuals .
hd 187123 is known to host a 0.5 @xmath0 `` hot jupiter '' in a 3 day orbit @xcite .
observations over the subsequent 8 years have revealed a trend of -7.3 m / s in the residuals to a one planet fit @xcite . in 2001 ,
the trend began to show signs of curvature , and in 2006 a it became clear that the residuals had passed through an rv minimum ( see fig .
[ 187123 ] ) .
[ 187123_contour ] shows the @xmath8 and @xmath28 contours in @xmath34 space . in this case , the @xmath67 contour and middle @xmath8 contour provide the following constraints : @xmath68 and @xmath69 y. we now consider known single - planet systems with low faps for second planets whose best - fit solutions have periods shorter than the span of observations .
we have identified eight such systems , and we discuss them below .
five stars appear to exhibit coherent residuals ( fap @xmath70 ) to a one planet fit , but in all cases the best two - keplerian fits are not compelling ( as noted in
[ foundtrends ] , in a sample of 100 known planet - bearing stars , we expect around 2 to exhibit residuals coherent at this level purely by chance )
. these possible companions do not appear in table [ orbitupdates ] because the tentative nature of these signals do not warrant publication of a full orbital solution with errors .
hd 11964 was announced in @xcite as having a planet with a 5.5 y orbital period and a linear trend .
we find an fap for a linear trend to be 6% , suggesting that while the inner planet is real , the trend is not .
we find an fap for a second planet to be @xmath71 , and a best - fit solution finds an inner planet with @xmath72d .
this star sits 2 magnitudes above the main sequence , and the residuals to the known planet are consistent with the typical jitter for subgiants of 5.7 m / s @xcite , so this signal could represent some sort of correlated noise .
this very low amplitude signal ( @xmath73 m / s ) will thus require much more data for confirmation .
hd 177830 is already known to have a jupiter - mass object in a nearly circular , 1.12 y orbit .
this remarkable system has a low fap @xmath23 for a second planet vs. a trend .
two good two - planet solutions exist for this system : the first has @xmath74 d and @xmath75 @xmath0 , the second has @xmath76 and @xmath77 @xmath0 .
this star sits more than 3.5 magnitudes above the main sequence , and the residuals to the known planet are consistent with the typical jitter for subgiants of 5.7 m / s @xcite , so this signal could represent some sort of correlated noise .
70 vir (= hd 117176 ) is a subgiant with a massive , 116.6 d planet on an eccentric orbit ( @xmath78 ) .
the fap for a second planet is 2% , but the best - fit second planet is not persuasive : @xmath79 9.58 d , and @xmath80 m / s .
the typical internal errors for this target are 5.4 m / s , making a bona fide detection of a 7 m / s planet very difficult .
we suspect that this signal is an artifact of stellar jitter , possibly due to the advanced evolution of the star .
hd 164922 has a known planet with a 3.1 y orbital period . for this star ,
the fap for a second planet is @xmath81 .
the best fit for this second planet has @xmath82 d and @xmath83 .
the amplitude of this signal is extremely low only @xmath84 m / s making this an intriguing but marginal detection .
hd 210277 is already known to host a planet with a 1.2 y orbit .
the fap for a second planet is 2% , and the best - fit second keplerian has @xmath85 m / s signal and @xmath86 d , and a 2% fap .
the best - fit orbit has @xmath87 , which is unlikely given that nearly all known hot jupiters have @xmath88 ( although the presence of the 1.2 y , @xmath87 outer planet could be responsible , in principle , for pumping an inner planet s eccentricity . )
the extremely low amplitude of this planet makes the exoplanetary nature of this signal very uncertain .
three additional stars with low faps are of a very early spectral type ( f78 ) : hd 89744 @xcite , hd 108147 @xcite , and hd 208487 @xcite .
their low activity yields a low jitter in the estimation of @xcite , but this is likely underestimated due to poor statistics : the california and carnegie planet search has very few stars of this spectral type from which to estimate the jitter . for hd 89744 and hd 108147
we suspect that , the low fap of @xmath71 is an artifact of coherent noise , since in our judgment neither case shows a compelling evidence of a second keplerian of any period .
hd 208487 has a very low fap ( @xmath23 ) despite the modest r.m.s of the residuals to a one planet fit of 8 m / s .
we suspect that stellar jitter is the likely source of these variations .
this star was discussed by @xcite , who applied a bayesian analysis to the published rv data , concluding that a second planet was likely , having @xmath89 d and @xmath90 .
@xcite also studied the published data , and suggested a planet with @xmath91 d. we note here two plausible solutions apparent in our data .
the first , with @xmath92 d and @xmath93 , is consistent with the solution of @xcite .
we also find , however , an additional solution of equal quality with @xmath94 d ( double the period of @xcite ) and @xmath95 .
this second solution has a period uncomfortably close to that of the lunar cycle ( we often see this period in the window function of our observations due to our tendency to observe during bright time ) . for both solutions @xmath96
we reiterate that we feel that the early spectral type of this star alone can account for the observed rv residuals .
@xcite analyzed our published rv data and found a low fap for the existence of a second planet in orbit around hd 188015 .
their fap , however , is measured against a null hypothesis of a single keplerian plus noise , thus ignoring the linear trend .
we find that @xmath20 , the improvement of the goodness - of - fit parameter with the introduction of a second keplerian versus a trend to be very small in fact 60% of our mock data sets showed greater improvement .
we therefore find no motivation to hypothesize the existence of an additional , short - period planet ; the single planet and trend announced in @xcite are sufficient to explain the data .
@xcite also found a low fap for a second planet in orbit about hd 114729 , with a period of 13.8 d. using the data set from @xcite , which contains 3 recent rv measurements taken since the publication of @xcite ( their source of rv data ) , we find no such signal , and a large fap for a second planet .
we suspect our results may differ because the additional data provide for a slightly better fit to the known exoplanet , changing the character of the residuals and destroying the coherence of the spurious 13.8 d signal .
hd 150706 b , a purported 1.0@xmath0 eccentric planet at 0.8 au , was announced by the geneva extrasolar planet search team ( 2002 , washington conference scientific frontiers in research in extrasolar planets " ; udry , mayor & queloz , @xcite ) and appears in @xcite ; however , there is no refereed discovery paper giving details .
we have made eight precision velocity measurements at keck observatory from 2002 through 2006 .
these velocities show an rms scatter of 12.1 m / s , inconsistent with the reported 33 m/s semi - amplitude of hd 150706 b. the rms to a linear fit is 8 m / s , which is adequately explained by the expected jitter for a young ( 700@xmath97300 myr ) and active star like hd 150706 .
we therefore doubt the existence of a 1.0@xmath0eccentric planet orbiting hd 150706 at 0.8 au .
as noted in [ foundtrends ] , prior to this work 24 of the 150 nearby stars known to host exoplanets ( including 14 her and excluding hd 150706 ) show significant trends in their residuals and 19 host well - characterized multiple planet systems .
one of these trends is likely spurious ( hd 11964 ) , and at least 3 others may be due to stellar or brown dwarf companions ( hd 142 , hd 13445 , and @xmath51 boo ) .
we have announced here the detection of an additional 5 trends for known planet - bearing stars , 2 new single systems , and one new multiple system ( hip 14810 , which appears as a single - planet system in @xcite ) .
we have also confirmed that the previously announced trends for hd 68988 and hd 187123 are likely due to planetary - mass objects .
this brings the total number of stars with rv trends possibly due to planets to 22 , the number of known multiple - planet systems to 22 , and the number of nearby planet - bearing stars to 152 .
this means that 30% of known exoplanet systems show significant evidence of multiplicity .
considering that the mass distribution of planets increases steeply toward lower masses @xcite , our incompleteness must be considerable between 1.0 and 0.1 jupiter - masses .
thus , the actual occurrence of multiple planets among stars having one known planet must be considerably greater than 30% . from an anthropocentric perspective
, this frequency of multiplicity suggests that in some respects , the solar system is not such an aberration .
our sun has 4 giant planets , and it appears that such multiplicity is not uncommon , although circular orbits are . from a planet - hunting perspective
this result is quite welcome as well , since it means that the immediate future of rv planet searches looks bright .
as our temporal baseline expands , we will become sensitive to longer - period planets .
our search is just becoming sensitive to true jupiter analogs with 12 year orbits and 12 m / s amplitudes .
a true saturn analog would require 15 more years of observation .
as our precision improves we will become sensitive to lower - mass planets , which may be the richest domain for planets yet .
the authors would like to thank kathryn peek for obtaining the crucial 16 apr 2006 rv measurement of hip 14810 , and simon otoole and alan penny for their assistance .
the authors also thank the anonymous referee for a thorough and constructive report .
this research is based on observations obtained at the w. m. keck observatory , which is operated jointly by the university of california and the california institute of technology .
the keck observatory was made possible by the generous financial support of the w. m. keck foundation .
the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community .
we are most fortunate to have the opportunity to conduct observations from this mountain .
this research has made use of the simbad database , operated at cds , strasbourg , france , and of nasa s astrophysics data system bibliographic services , and is made possible by the generous support of sun microsystems , nasa , and the nsf . , d. a. , laughlin , g. , butler , p. , marcy , g. , johnson , j. , henry , g. , valenti , j. , vogt , s. , ammons , m. , robinson , s. , spear , g. , strader , j. , driscoll , p. , fuller , a. , johnson , t. , manrao , e. , mccarthy , c. , muoz , m. , tah , k. l. , wright , j. , ida , s. , sato , b. , toyota , e. , & minniti , d. 2005 , , 620 , 481486 , m. a. c. , & esa 1997 . , the hipparcos and tycho catalogues . astrometric and photometric star catalogues derived from the esa hipparcos space astrometry mission , publisher : noordwijk , netherlands : esa publications division , 1997 , series : esa sp series vol no : 1200 , isbn : 9290923997 ( set ) | we present four new exoplanets : hip 14810 b & c , hd 154345 b , and hd 187123 c. the two planets orbiting hip 14810 , from the n2k project , have masses of 3.9 and 0.76 @xmath0 .
we have searched the radial velocity time series of 90 known exoplanet systems and found new residual trends due to additional , long period companions .
two stars known to host one exoplanet have sufficient curvature in the residuals to a one planet fit to constrain the minimum mass of the outer companion to be substellar : hd 68988 c with @xmath1 and hd 187123 c with @xmath2 , both with @xmath3 y. we have also searched the velocity residuals of known exoplanet systems for prospective low - amplitude exoplanets and present some candidates .
we discuss techniques for constraining the mass and period of exoplanets in such cases , and for quantifying the significance of weak rv signals .
we also present two substellar companions with incomplete orbits and periods longer than 8 y : hd 24040 b and hd 154345 b with @xmath4 and @xmath5 @xmath0 , respectively . |
for over a century , the conventional way to represent many enzyme reactions has been based on the michaelis - menten kinetics @xcite ( hereafter mm model ) , in which substrate s binds reversibly with enzyme e to form an enzyme - substrate complex es , which is later transformed into product p and free enzyme e. derivation of the michaelis - menten equation is based on the following scheme @xcite : however , there is strong evidence suggesting that enzime activity depends on conformational changes in the enzyme structure @xcite .
it is therefore reasonable to assume that a conformational change takes place from the moment an es complex forms to the moment a product p is released ; such change requires a certain time for its completion .
this conformational change must then be reset before the enzyme is ready to bind substrate again , and so a recovery time must elapse once p is released .
single - enzyme studies @xcite and nuclear magnetic resonance spectroscopy studies @xcite support these notions .
a scheme that considers enzyme regeneration is presented by @xcite , , + where @xmath0 represents recovering enzyme .
conventional mm kinetics do not account in an explicit way for the times required by these conformational changes .
the aim of this paper is to present two models of enzyme catalysis , one continuous and one discrete , that take into account the times needed by enzymes to process es complexes and to reset to its original state once products has been released ; these models can be considered as an extension of conventional mm kinetics which includes delays . the main motivation behind the continuous delayed model is to establish the limitations of the mm description in non - equilibrium conditions .
in order to examine the limitations associated to the continuous nature of the first model , we consider a discrete delayed model in which the catalytic cicle of each enzyme in the system is represented by a recursive function , hereafter referred to as the enzyme map .
previous work by @xcite present an analogous description of enzyme action in which each protein molecule is represented by a clock - like automaton that binds substrate s and transforms it into product p. the time needed by a single automaton to complete this transformation and return to its original waiting state represents the enzyme turnover time @xmath1 .
the conformational changes of the enzyme ( and the different stages of the catalytic cycle ) are represented by motion along an internal ` phase ' coordinate . under the aproppriate conditions this model exhibits complex behavior such as oscillations and cluster formation ; thus another motivation is to explore the emergence of such behavior in our model . unlike other microscopic models based on stochastic simulation @xcite ,
our discrete model is completely deterministic , and each enzyme is represented by a map instead of a clock - like unit . in section 2
the continuous delayed model is presented by means of a simple reaction , and a comparison is made between numerical integration provided by mm kinetics and those produced by the delayed model .
section 3 presents the equivalent discrete model ; the results of the discrete simulation are compared with results obtained through the continuous delayed model . in the conclusions presented in section 4 we highlight the respective merits and limitations of each model .
to illustrate the effect of delays on the dynamics of enzyme catalysis we consider the simple reaction , where conversion between two substrates @xmath2 and @xmath3 is catalyzed by two unidirectional monomeric enzymes @xmath4 and @xmath5 . as mentioned in the introduction , each conformational change taking place along the catalytic cycle will require a certain time for its completion .
we then distinguish three stages in the cycle : 1 ) the enzyme is free , waiting to bind substrate ; 2 ) the enzyme has bound substrate forming an es complex ; 3 ) product p has been released and the enzyme is resetting to its initial conformation .
the processing time @xmath6 required by stage 2 is a fraction @xmath7 of the turnover time @xmath1 , while the recovery time @xmath8 associeted to stage 3 is @xmath9 .
the turnover time is the time that elapses from the moment an es complex forms until the moment the enzyme is ready to bind substrate again ; such time is given by @xmath10}\ ] ] where @xmath11 is the maximum velocity of the enzyme in @xmath12mol/(min - mg of protein ) and @xmath13 its molecular weight in daltons . when processing and recovery times are considered the reaction is described by a set of four coupled delay differential equations ( ddes ) : @xmath14}{dt}&=&\frac{1}{v_r}\left\lbrace g\!_{_{\beta}}[\mathrm{b}(t-\tau\!_{_{\beta p}})]m\!_{_{\beta 1}}(t-\tau\!_{_{\beta p}})- g\!_{_{\alpha}}[\mathrm{a}(t)]m\!_{_{\alpha 1}}(t)\right\rbrace + \ : [ \dot{\mathrm{a}}_{ext}(t ) ] \label{dde1}\\ \frac{d[\mathrm{b}]}{dt}&=&\frac{1}{v_r}\left\lbrace g\!_{_{\alpha}}[\mathrm{a}(t-\tau\!_{_{\alpha p}})]m\!_{_{\alpha 1}}(t-\tau\!_{_{\alpha p } } ) - g\!_{_{\beta}}[\mathrm{b}(t)]m\!_{_{\beta 1}}(t)\right\rbrace
\label{dde2}\\ \frac{dm\!_{_{\alpha1}}}{dt}&=&\mu_{\alpha}\times 10^{-3 } g\!_{_{\alpha}}\left\lbrace [ \mathrm{a}(t - \tau\!_{_{\alpha}})]m\!_{_{\alpha 1}}(t - \tau\!_{_{\alpha } } ) - [ \mathrm{a}(t)]m\!_{_{\alpha 1}}(t))\right\rbrace \label{dde3}\\ \frac{dm\!_{_{\beta1}}}{dt}&=&\mu_{\beta}\times 10^{-3 } g\!_{_{\beta}}\left\lbrace [ \mathrm{b}(t - \tau\!_{_{\beta}})]m\!_{_{\beta 1 } } ( t - \tau\!_{_{\beta } } ) - [ \mathrm{b}(t)]m\!_{_{\beta 1}}(t))\right\rbrace\label{dde4 } \end{aligned}\ ] ] where @xmath15 $ ] , @xmath16 $ ] are substrate concentrations in @xmath12 m ; @xmath17 , @xmath18 represent the amount of free enzyme in milligrams and @xmath19 , @xmath20 are the molecular weights of enzymes @xmath4 , @xmath5 in daltons
. the delays @xmath21 and @xmath22 are their respective turnover times , whereas @xmath23 , @xmath24 are their respective processing times .
@xmath25 is the volume in which the reacion takes place , in liters .
the rates @xmath26 and @xmath27 are given by @xmath28 ( @xmath29 ) , where @xmath30 is the michaelis constant in @xmath12 m .
@xmath31 $ ] represents the rate at which substrate a is added to the system .
equations for @xmath32 , @xmath33 , @xmath34 and @xmath35 can be easily obtained from @xmath36 $ ] , @xmath37 $ ] , @xmath38 , @xmath39 .
for instance , the term @xmath40m\!_{_{\alpha 1}}(t)$ ] is the rate at which @xmath32 appears , while @xmath41m\!_{_{\alpha 1}}(t-\tau\!_{_{\alpha p}})$ ] is the rate of dissapearance of @xmath32 .
equations ( [ dde1]-[dde4 ] ) reduce to conventional mm equations @xmath14}{dt}&=&\frac{1}{v_r}\left\lbrace v_{max_{\beta}}\frac{[\mathrm{b}(t)]}{k_{\beta}+[\mathrm{b}(t)]}m\!_{_{\beta } } -v_{max_{\alpha}}\frac{[\mathrm{a}(t)]}{k_{\alpha}+[\mathrm{a}(t)]}m\!_{_{\alpha}}\right\rbrace + [ \dot{\mathrm{a}}_{ext}](t ) \label{mm1}\\ \frac{d[\mathrm{b}]}{dt}&=&-\frac{d[\mathrm{a}]}{dt } \label{mm2},\end{aligned}\ ] ] when @xmath42 . in this case , the amounts of free enzymes @xmath17 and @xmath43 are given by @xmath44/k_{\alpha}}\label{na1}\\ m\!_{_{\beta 1}}(t ) & = & \frac{m\!_{_{\beta}}}{1+[\mathrm{b}(t)]/k_{\beta}}\label{nb1},\end{aligned}\ ] ] where @xmath45 and @xmath46 are the total amounts of enzymes @xmath4 and @xmath5 in milligrams , respectively .
notice that unlike the delayed model ( eqs .
[ dde1]-[dde4 ] ) , the mm equations ( eqs .
[ mm1]-[mm2 ] ) implicitly assume that substrates are instantaneously transformed into products and that the free enzyme amounts are instantaneously determined by substrate concentrations . on the other hand , in the delayed model the rate at which substrate a appears
is proportional to the rate at which substrate b was bound by enzyme @xmath5 at time @xmath47 .
similarly , the rate at which free enzyme ( e.g. @xmath39 ) appears at time @xmath48 equals the rate at which free enzyme dissappeared a turnover time @xmath1 earlier ( e.g. @xmath49 ) . when stationary conditions are reached in the continuous delayed model , the amounts of enzyme ( in milligrams ) in each stage are @xmath50}\tau\!_{_{\alpha}}}\\ \label{mest1 } \overline{m\!_{_{\alpha 2 } } } & = & \frac{\mu_{\alpha}\times 10^{-3}g\!_{_{\alpha}}\overline{[\mathrm{a}]}m\!_{_{\alpha}}\tau\!_{_{\alpha p}}}{1+\mu_{\alpha}\times 10^{-3}g\!_{_{\alpha}}\overline{[\mathrm{a}]}\tau\!_{_{\alpha}}}\\ \overline{m\!_{_{\alpha 3 } } } & = & \frac{\mu_{\alpha}\times 10^{-3}g\!_{_{\alpha}}\overline{[\mathrm{a}]}m\!_{_{\alpha}}\tau\!_{_{\alpha}}(1-c_{_{\alpha}})}{1+\mu_{\alpha}\times 10^{-3}g\!_{_{\alpha}}\overline{[\mathrm{a}]}\tau\!_{_{\alpha}}}\end{aligned}\ ] ] where the subscripts 1 , 2 , 3 correspond to free enzyme , occupied enzyme and recovering enzyme , respectively , and @xmath51}$ ] is the stationary value of concentration @xmath36 $ ] .
the same expressions can be used to determine @xmath52 , @xmath53 and @xmath54 by replacing @xmath51}$ ] for @xmath55}$ ] and the parameters @xmath45 , @xmath21 , @xmath19 , @xmath26 and @xmath56 for those corresponding to enzyme @xmath5 . in the continuous delayed model the stationary substrate concentration @xmath51}$ ]
corresponds to one of the solutions of a third degree polynomial when considering eqs .
( [ dde1]-[dde4 ] ) in stationary conditions .
a similar equation must be solved to find the stationary concentration @xmath55}$ ] . from the solutions of these polynomials , the stationary solution is the one satisfaying the restrictions @xmath57}<[\mathrm{s}_0]$ ] , @xmath58}<[\mathrm{s}_0]$ ] , where @xmath59 $ ] is the total substrate concentration in the system .
note that at any time @xmath59 = [ \mathrm{a}](t ) + [ \mathrm{b}](t ) + [ \alpha_2](t ) + [ \beta_2](t)$ ] , where @xmath60(t)$ ] and @xmath61(t)$ ] are the concentrations of processing enzymes .
the corresponding stationary values of the substrate concentrations @xmath51}_{mm}$ ] , @xmath55}_{mm}$ ] for the mm model ( eqs .
( [ mm1 ] ) and ( [ mm2 ] ) ) are given by the solutions of two second - degree polynomials that satisfy @xmath57}<[\mathrm{s}_0]$ ] , @xmath58}<[\mathrm{s}_0]$ ] .
note that at any time @xmath59 = [ \mathrm{a}](t ) + [ \mathrm{b}](t)$ ] . since the mm description does not consider bound substrate , the continuous delayed model and the mm model agree in the stationary state when @xmath51}_{mm}$ ] is compared to @xmath51}+\overline{[\alpha_2]}$ ]
; a similar relation applies to substrate b. in order to establish a comparison between the delayed model and mm kinetics , numerical integration of eqs .
( [ dde1]-[dde4 ] ) and ( [ mm1]-[mm2 ] ) was performed .
initial substrate concentrations are set to zero and all enzymes are initialized in their idle state .
later , substrate a is quickly added until a concentration @xmath59 $ ] is reached according to @xmath31= [ \dot{\mathrm{a}}_{ext , max } ] \exp(-(t - t_0)^2/\sigma)$ ] , where @xmath62 $ ] is the maximum rate of substrate addition reached at time @xmath63 ; the parameter @xmath64 allows to control the time span during which significant amounts of substrate are added .
hereafter we adopt @xmath65 , and @xmath66 .
unless otherwise stated , the results shown correspond to the following parameter values : the reaction volume @xmath25 is set to 10.3 attoliters , corresponding to a _ t. brucei
_ glycosome @xcite . for the sake of simplicity
the reaction is made symmetric , i.e. @xmath67 @xmath12m/(min - mg of protein ) ; the amount of enzyme is chosen to be @xmath45 = @xmath46 = 1.66 attograms , a figure equivalent to 200 molecules of an enzyme with a molecular weight of 5000 daltons .
this corresponds to an enzyme concentration @xmath68=[\beta]=32.24 $ ] @xmath12 m . for these parameter values @xmath69 sec . in the following example @xmath70
; for the dde case we use @xmath71 . figure 1a shows a comparison between the evolution of [ a ] in the mm and dde models for @xmath72 @xmath12 m and @xmath59=100 $ ] @xmath12 m .
when the stationary regime is attained , the concentration [ a ] in the dde case is below that of the mm case since in the dde model a portion of substrate is bound in the form of es complex with concentration @xmath60 $ ] .
figure 1b shows results when @xmath73 is reduced to 0.1 @xmath12 m .
the continuous thin curve correspond to the evolution of @xmath74 + [ \alpha_2]$ ] .
several new features are now apparent : a ) the difference between mm and dde free substrate concentrations widens because in the dde model the portion of substrate in the form of es complex increases as @xmath73 is reduced . and b ) oscillations are present in the dde case as a result of the history dependence of the system evolution , an usual behavior of systems described by delay differential equations @xcite .
note that a longer characteristic time is needed to reach equilibrium when @xmath73 is reduced ; this is due to the fact that a reduction of @xmath73 increases the binding rate of both substrates while the fraction of available free enzymes decreases , but due to the symmetric cyclic nature of the reaction , substrate a is replenished at a faster pace .
figure 2 presents the same comparison shown in fig .
1b , but with @xmath7 lowered to @xmath75 .
as stated above , the mm model assumes the products are released instantaneously after substrate is bound by the enzime .
therefore , as expected , for @xmath76 substrate concentrations in the dde model oscillate around those in the mm model .
figure 3 shows a comparison between the mm and dde models of the fraction of free enzyme @xmath77 and the amount of substrate a normalized to the total subtrate concentration @xmath59 $ ] for @xmath78 @xmath12 m , @xmath59=100 $ ] @xmath12 m and @xmath79 . as substrate
a is being added to the system the fraction of free enzyme @xmath77 in the dde model shows a delay with respect to the mm model .
this delay is the result of an important difference between mm kinetics and the delayed model : in the former , the amount of free enzyme is determined instantaneously by the substrate concentrations through eqs .
[ na1 ] and [ nb1 ] ; this differs from the time - delayed evolution described by eqs .
[ dde3 ] and [ dde4 ] .
@xcite proposed a discrete enzyme model in which each enzyme molecule is represented by a clock - like automaton with a phase that starts advancing once a substrate molecule binds to it , initiating a catalytic cycle whose duration represents the turnover time @xmath1 . at a certain moment along this cycle
the automaton releases a product , returning to its initial phase after a recovery time has elapsed .
enzyme reactions are thus represented in terms of a automata network , in which each automaton is coupled to the others through binding and release of intermediate products . instead of a clock - like automaton ,
the discrete model proposed here assumes that the phase @xmath80 of each single enzyme along the catalytic cycle is determined by the recursive function @xmath81 ( hereafter the enzyme map ) shown in fig .
4 . the map is a deterministic procedure to evolve the phase of the enzimes from its phase state @xmath82 at discrete time @xmath48 to @xmath83 at discrete time @xmath84 ; that is @xmath85 .
the map has a chaotic region defined in @xmath86 , which corresponds to the idle state of the enzyme ( e.g. @xmath87 for enzyme @xmath4 ) .
iterations start in this region from an initial random phase value @xmath88 ; the average number of iterations that transcur in the chaotic region is inversely proportional to @xmath89 , the width between the two intersections of the map at @xmath90 ( @xmath91 in fig .
4 ) . when an iterate @xmath82 in the chaotic region falls into the range @xmath92 the enzyme is assumed to bind a substrate molecule : the next iterate exits the chaotic region and enters the laminar region .
the laminar region is divided in two sections : the section from @xmath93 to @xmath94 corresponds to the processing of the substrate by the enzyme ( i.e. @xmath95 ) whereas the section from @xmath94 to @xmath96 corresponds to the recovery process of the enzyme ( i.e. @xmath97 ) .
once @xmath98 a product molecule is released , as indicated in fig .
4 . the region of the map between @xmath96 and @xmath99 allows to reinject in a single step the iterate into the chaotic region , where the enzyme is ready to bind a substrate again .
the enzyme map can also be used to model more complicated structures and functions such as oligomers and competitive / non - competitive inhibition ; these are properties that will be presented elsewhere . for a fixed value of @xmath100 ( e.g. @xmath101 ) , the two parameters of the map @xmath102 and @xmath7 , and the variable @xmath89 can be adjusted to obtain an equivalence with the continuous delayed model . note that the number of steps in the laminar region is @xmath103 ; then if each time step corresponds to a unit time @xmath104 , the turnover time is @xmath105 .
the time @xmath104 associated to each iteration must be small enough to produce a significant number of iterations in the laminar region ; once a given @xmath104 is chosen the length of the laminar region is set according to @xmath106 .
the parameter @xmath107 is the same one defined for the continuous delayed model .
variable @xmath89 is related to the average time @xmath108 that an enzyme remains in its idle state and is assumed proportional to the number @xmath109 of substrate molecules in the simulation volume ( i.e. @xmath110 , where @xmath111 is the enzyme rate constant ) .
the number of susbstrate molecules that corresponds to a given concentration [ a ] ( in @xmath12 m ) is @xmath11210^{-6}n_a$ ] , where @xmath113 is avogadro s number .
when [ a ] = @xmath73 the velocity of the enzyme is halved ; in this condition the time spent in the chaotic region must be equal to the time spent in the laminar region ; that is @xmath114 .
therefore , @xmath115 with @xmath116 .
finally , the number of enzymes @xmath117 that corresponds to a given mass of enzyme @xmath118 in mg is @xmath119 . by using an array of such maps it is possible to consider the effect of the discrete nature of reactants on the dynamics of the system and analize the conditions under which the continuous delayed model is no longer appropriate to describe very small systems .
figure 5 shows integration results for the delayed continuous model and its equivalent discrete counterpart . in the dde simulation
we use the same parameter values as in figs . 1 and 2 , but with @xmath120 @xmath12 m .
in the system volume of 10.3 attoliters this concentration correspond to @xmath121 substrate molecule ; on the other hand the total substrate concentration @xmath59 $ ] = 100 @xmath12 m corresponds to 620 substrate molecules .
as before , there are 200 enzyme molecules of each type with a turnover time @xmath122 s and a molecular weight of 5000 da .
therefore , in the discrete simulation we must follow the phase evolution of 400 maps , 200 for each type of enzime . if we adopt @xmath123 @xmath12s and @xmath101
, the transit in the laminar region occurs in 24000 iterates ( i.e. @xmath124 ) . as the reaction progresses the number of available free subtrate molecules ( @xmath125 and @xmath126 ) changes , and so the value of @xmath89 . for the maps representing enzime @xmath4 we have @xmath127 , where @xmath128 is the number of substrate molecules corresponding to concentration @xmath129 .
a similar relation holds for @xmath130 .
when the @xmath131 ratio is 1 the number of iterations spent in the chaotic region equals the number of iterations spent in the laminar region , so the enzyme will operate at half its maximum velocity . as can be seen in fig .
5 , agreement between both models is excellent , but there is one caveat : care must be taken so that the time step @xmath104 is small enough to guarantee that always @xmath132 ; otherwise the results provided by the discrete model might disagree from the results of the continuous delayed model . in other words , @xmath104 must be less than @xmath133 , where @xmath134 is the number of substrate molecules corresponding to the total substrate concentration @xmath59 $ ] . unlike the discrete model , the dde approach is expected to fail when the number of substrate molecules in the reaction volume is small .
as an example of this limitation , fig .
6 shows the evolution of free substrate @xmath135 in both models when @xmath59 $ ] is reduced to 32.24 @xmath12 m , a concentration equivalent to 200 substrate molecules , half the number of enzyme molecules .
the evolution of [ a ] shows how the results provided by the discrete model tend to follow those corresponding to the delayed continuous model ; however fluctuations in [ a ] can be seen as substrate molecules are bound and released by the enzymes .
moreover , synchronization between the delayed continuous model and the discrete model is lost as the reaction progresses .
this is the type of situation that can be expected in conditions of substrate starvation or when the reaction volume is reduced while maintaining @xmath59 $ ] and the amount of enzyme .
as previoulsy mentioned , in the discrete model it is possible to follow the phase evolution of every single enzyme in the system .
we show this by examining the same simple reaction used so far but with parameter values for which sustained oscillations occur ; they are @xmath136 attoliters , @xmath137 @xmath12m/(min - mg ) , @xmath138 @xmath12m/(min - mg ) , @xmath139 , @xmath140 , @xmath141 @xmath12 m , @xmath142 @xmath12 m , and @xmath59=150 $ ] @xmath12 m ( equivalent to 465 substrate molecules ) .
the number and molecular weight of the enzymes remains the same as in previous cases .
figure 7 shows the phase histograms for enzymes @xmath4 for two time moments separated by @xmath143 , illustrating how the phases of enzimes @xmath4 tend to be synchronized and how the cluster phase advances through the catalytic cycle . in the first histogram
there are some enzymes waiting to bind substrate , while in the second histogram these enzymes have already bound a substrate .
we have included processing and recovery times in enzyme kinetic models in order to account in a simple way for the conformational changes that a monomer undergoes during the catalytic cycle and its effects on the resulting dynamics . in certain cases ,
the dynamics exhibited by the continuous delayed model differs drastically when compared to that of mm kinetics .
examples found in the literature indicate that oscillations in enzyme reactions require some strong form of non - linearity : inhibition - activation mechanisms , presence of feedback loops , interaction between the elements of a network , or cooperativity between enzyme subunits @xcite .
however , the emergence of oscillations in enzyme reactions can also be promoted by the delayed dynamics . unlike the mm model , the delayed model accounts for that portion of substrate that is present in the system as es complex ; thus stationary substrate concentration values agree when the sum of free and bound substrate concentrations in the continuous delayed model are compared with free substrate concentrations in the mm model .
another important difference between mm kinetics and our model is that the fraction of free enzyme in the continuous delayed model follows a time - delayed evolution , while in the mm case such fraction is instantaneously determined by substrate concentrations .
the proposed discrete model agrees with the continuous delayed model when large enough number of substrate molecules are considered .
when this number is small the discrete model shows the expected fluctuations that are associated to the discrete nature of reactions at microscopic scales .
these effects can be important when considering reactions that take place in very small volumes such as those of organelles .
the discrete model offers the possibility of following the evolution of each single enzyme , and therefore allows for the study of situations in which self - organization phenomena arise , such as synchronization and clustering .
the authors would like to thank dr .
juan luis concepcin and dr .
luisana aviln for their valuable opinions and insight during the many fruitful discussions we had .
agmon , n. , 2000 .
conformational cycle of a single working enzyme .
b , 104 , 7830 - 7834 .
bagshaw , c.r . ,
cherny , d. , 2006 . blinking fluorophores : what do they tell us about protein dynamics ?
trans . , 34 , 5 , 979 - 982 .
benkovic s. , hammes - schiffer s. , 2003 . a perspective on enzyme catalysis .
science , 301 , 1196 - 1202 .
bennett , w. , steitz , t. , 1978 .
glucose - induced conformational change in yeast hexokinase .
pnas , 75 , 10 , 4848 - 4852 .
chickarmane v. , khodolenko , b. , sauro , m. , 2007 .
oscillatory dynamics arising from competitive inhibition and multisite phosphorilation .
bio , 244 , 68 - 76 , doi:10.1016/j.tbi.2006.05.013 eisenmesser e. , bosco , d. , akke , m. , kern , d. , 2002 .
enzyme dynamics during catalysis .
science , 295 , 5559 , 1520 - 1523 .
eisenmesser , e. , millet , o. , labeikovsky , w. , korzhnev , d. , wolf - watz , m. , bosco , a. , skalicky , j. , kay , l. , kern , d. , 2005 .
intrinsic dynamics of an enzyme underlies calalysis .
nature , 438 , 3 , 117 - 121 .
english , b. , min , w. , van oijen , a. , lee , k. , luo , g. , sun , h. , cherayil , b. , kou , s. , xie , s. , 2006 .
ever - fluctuating single enzyme molecules : michaelis - menten equation revisited .
nature , 2 , 2 , 87 - 94 .
goldbetter , a. , 1996 .
biochemical oscillations and cellular rhythms .
cambridge university press , london , pp 67 - 75 hakansson , k. , doherty , a. , shuman , s. , wigley , d. , 1997 .
x - ray crystallography reveals a large conformational change during guanyl transfer by mrna capping enzymes . cell , 89 , 545 - 553 .
haseltine , e. , rawlings , j. , 2002 .
approximate accelerated stochastic simulation of coupled fast and slow reactions for stochastic chemical kinetics .
, 115 , 1716 - 1733 .
herschlag , d. , 1988 .
the role of induced fit and conformational changes of enzymes in specificity and catalysis .
, 16 , 62 - 96 .
hess , b. , mikhailov a. , 1994 .
self - organization in living cells .
science , 264 , 223 - 224 .
huang , y. , montelione , g. , 2005 .
proteins flex to function .
nature , 438 , 36 - 37 .
kierzek , a. , 2002 .
stocks : stochastic kinetic simulations of biochemical systems with gillespie algorithm .
bioinformatics , 18 , 3 , 470 - 481 .
laub , m. , loomis , w. , 1998 .
a molecular network that produces spontaneous oscillations in excitable cells of _ dyctyostelium_. molecular biology of the cell , 9 , 3521 - 3532 .
lu , h. , xun , l. , xie , s. , 1998 .
single - molecule enzymatic dynamics .
science , 282 , 1877 - 1882 michaelis l. , menten m. , 1913 .
die kinetik der invertinwirkung .
z. 49 , 333 - 369 .
navid a. , ortoleva p. , 2004 .
simulated complex dynamics of glycolysys in the protozoan parasite _
trypanosoma brucei_. j. theor .
, 228 , 449 - 458 , doi:10.1016/j - jtbi.2004.02.007 parravano , a. , 1996 .
star formation bursts driven by delayed self - regulation .
j. , 462 , 594 - 600 .
puchalka , j. , kierzek .
, a. , 2004 . bridging the gap between stochastic and deterministic regimes in the kinetic simulations of the biochemical reaction networks .
j. , 86 , 1357 - 1372 .
rao , c. , arkin , a. , 2003 .
stochastic chemical kinetics and the quasy - steady - state assumption : application to the gillespie algorithm .
, 118 , 4999 - 5010 .
segel , e. , 1975 .
enzyme kinetics : behavior and analysis of rapid equilibrium and steady - state enzyme systems .
john wiley & sons , new york , pp .
selkov e. , 1968 .
self - oscillations in glycolysis .
j. biochem . , 4 , 79 - 86 .
stange , p. , mikhailov , a. , hess , b. , 1998 .
mutual synchronization of molecular turnover cycles in allosteric enzymes .
b , 102 , 6273 - 6289 .
stange , p. , zanette , d. , mikhailov , a. , hess , b. , 1999 .
self - organizing molecular networks .
, 79 , 233 - 247 .
walter , n. , 2006 .
michaelis - menten is dead , long live michaelis - menten .
nature , 2 , 2 , 66 - 67 .
xie , s. , lu , p. , 1999 .
single - molecule enzymology .
, 274 , 23 , 15967 - 15970 . fig .
a ) [ a ] evolution for mm and dde models when @xmath67 @xmath12m/(min - mg ) , @xmath144 @xmath12 m , @xmath59=100 $ ] @xmath12 m , @xmath145 = @xmath146 = 1.66 attograms , @xmath147 and @xmath148 attoliters .
b ) same situation with @xmath149 @xmath12 m ( see text ) .
evolution of the amount of free substrate a for dde model and the discrete model . left hand y - axis shows concentration values while right hand y - axis shows corresponding number of susbstrates .
paramater values are @xmath67 @xmath12m/(min - mg ) , @xmath151 @xmath12 m , @xmath59=100 $ ] @xmath12 m , @xmath145 = @xmath146 = 1.66 attograms , @xmath147 , @xmath152 @xmath12s and @xmath148 attoliters .
phase histograms for enzyme @xmath4 .
parameter values are @xmath137 @xmath12m(min - mg ) , @xmath138 @xmath12m/(min - mg ) , @xmath153 , @xmath154 , @xmath141 @xmath12 m , @xmath142 @xmath12 m , and @xmath59=150 $ ] @xmath12 m and @xmath136 attoliters . | a comparison is made between conventional michaelis - menten kinetics and two models that take into account the duration of the conformational changes that take place at the molecular level during the catalytic cycle of a monomer .
the models consider the time that elapses from the moment an enzyme - substrate complex forms until the moment a product molecule is released , as well as the recovery time needed to reset the conformational change that took place . in the first model
the dynamics is described by a set of delayed differential equations , instead of the ordinary differential equations associated to michaelis - menten kinetics . in the second model the delay , the discretization inherent to enzyme reactions and the stochastic binding of substrates to enzimes at the molecular level is considered .
all three models agree at equilibrium , as expected ; however , out - of - equilibrium dynamics can differ substantially .
in particular , both delayed models show oscillations at low values of the michaelis constant which are not reproduced by the michaelis - menten model .
additionally , in certain cases , the dynamics shown by the continuous delayed model differs from the dynamics of the discrete delayed model when some reactant become scarce
. , nonlinear dynamics , enzyme dynamics , enzyme reaction simulation , enzyme kinetics |
the proximity effect between a ferromagnet and a superconductor has attracted much attention recently due to its potential applications in such areas of technology as magnetoelectronics @xcite or quantum computing @xcite . the proximity effect in ferromagnet - superconductor (
f - s ) structures is also important from the point of view of the scientific interest as it allows the study of the interplay between ferromagnetism and superconductivity @xcite .
due to time reversal symmetry breaking the proximity effect in f - s structures leads to new phenomena , not observed in usual normal metal - superconductor ( n - s ) proximity systems .
those are : oscillations of the superconducting transition temperature @xcite , density of states @xcite and superconducting pairing amplitude in f - s multilayers , josephson @xmath0-junction behavior in s - f - s heterostructures @xcite , a giant mutual proximity effect in s - f systems @xcite , spin valve @xcite or spontaneous currents in f - s bilayers @xcite . for review of the literature on those and related effects see @xcite .
we have recently examined f - s proximity system and found that a spontaneously generated current flows on both sides of the interface @xcite .
such a current , which flows in the ground state of the system is a hallmark of the fulde - ferrell - larkin - ovchinnikov ( fflo ) like state , originally predicted for a bulk superconductor in a magnetic exchange field acting on spins only @xcite .
it is the purpose of the present paper to see what the effect on the fflo state will have a normal metal sandwiched between ferromagnet and superconductor ( f - n - s structure ) .
will the spontaneous current still be generated ?
if so , how will it be modified ? in the rest of the paper
we show that the presence of the normal metal has nontrivial consequences on the generation of the spontaneous currents and on the fflo state in general .
our model system is schematically shown in fig . [ fig1 ] .
it consists of a ferromagnet ( layers @xmath1 ) , a normal ( paramagnetic ) region ( layers @xmath2 ) and a semi - infinite superconductor ( layers @xmath3 ) .
the model hamiltonian is given by single orbital hubbard model @xmath4 c^+_{i\sigma } c_{j\sigma } + \frac{1}{2 } \sum_{i\sigma } u_i \hat n_{i\sigma } \hat n_{i -\sigma } , \label{hamilt}\end{aligned}\ ] ] where , in the presence of a vector potential , the nearest neighbor hopping integral is @xmath5 .
the site energy levels @xmath6 are equal to @xmath7 in the ferromagnet and @xmath8 in the normal and superconducting region , and @xmath9 is the chemical potential .
@xmath10 is the on - site electron - electron interaction , which is assumed to be negative on superconducting side and zero elsewhere , @xmath11 ( @xmath12 ) are the usual electron creation ( annihilation ) operators and @xmath13 is the electron number operator .
we study the above model in the spin - polarized hartree - fock - gorkov ( sphfg ) approximation , assuming the landau gauge for the magnetic field @xmath14 , thus @xmath15 ( see fig .
[ fig1 ] ) . in the following we assume periodicity of our model in the direction parallel to the interface and
therefore we work in @xmath16 space in the @xmath17 direction but in real space in the @xmath18 direction . as usual , self - consistency
is assured by the relations determining sc order parameter @xmath19 , current @xmath20 and the vector potential @xmath21 on each layer @xmath22 @xmath23 @xmath24 @xmath25 where @xmath26 is the @xmath27 nambu green function and @xmath28 the fermi distribution
. equation ( [ vec_pot ] ) is a lattice version of the ampere s law @xmath29 .
the above equations have to be solved self - consistently , using the method described in @xcite .
figure [ fig2 ] shows the spatial dependence of the normalized superconducting pairing amplitude @xmath30 ( left panel ) and spontaneous current @xmath31 ( right panel ) for a number of normal metal @xmath32 layers ( @xmath33 ) . as in our previous work on f - s structures @xcite
, the superconducting pairing amplitude shows clear oscillations in the ferromagnet .
the period of the oscillations is related to the ferromagnetic coherence length @xmath34 , i.e. @xmath35 . no such effect is observed in the n region .
one could expect that the inclusion of the normal metal suppresses the oscillations of @xmath36 in the f region .
interestingly , the amplitude of the @xmath36 oscillations on the ferromagnetic side remains almost unchanged .
only the phase of the oscillations can change . moreover , as for f - s bilayer @xcite , the vanishing of the pairing amplitude at @xmath37 is related to the crossing of the andreev bound states ( abs ) through the fermi energy of the system .
furthermore , in such a situation , spontaneous current is generated @xcite .
the current flowing produces a magnetic field , which splits this zero energy abs , thus lowering the total energy of the system .
the energy of this splitting is given in our model by @xmath38 , where @xmath39 is the layer averaged vector potential .
as we already mentioned , the zero energy abs is responsible for the generation of the spontaneous current .
the typical distribution of such current , flowing parallel to the interface , is shown in the right panel of fig .
the current flows in the negative @xmath17 direction on the superconducting side and in the positive direction in the whole @xmath40 region , giving a total current equal to zero , as should be in the true ground state .
interestingly , the current in the ferromagnet shows also oscillatory behavior , however , with a different period than the pairing amplitude does .
the oscillations of the current are related to the oscillations of the density of states at the fermi energy @xcite . on the other hand , in the n region ,
the distribution of the current is rather smooth , similarly to the pairing amplitude ( compare the solid and thin dotted lines in fig .
[ fig2 ] ) .
the state with spontaneous current is a true ground state , as it has lower energy than the state with no current .
this can be seen in fig .
[ fig3 ] , which shows the change in total energy of the system @xmath41 between the solution with spontaneous current and the one where the current is constrained to be zero is shown .
another important finding is that the system can be periodically switched between the states with and without spontaneous current by increasing of the thickness of the normal metal region . for @xmath42 layers and
@xmath43 the period is equal to six ( see fig . [ fig3 ] ) .
all above the results show that inclusion of the normal metal slab between the ferromagnet and superconductor has non - trivial effects on the physics of the proximity induced fflo state .
it can not be argued that the normal metal simply acts as interface transparency , as in our case the n slab is in the clean ( ballistic ) regime , and thus does not suppress the proximity effect .
however , it leads to the suppression of the oscillatory behavior of the pairing amplitude , spontaneous current and the density of states at the fermi energy .
moreover , it strongly modifies the positions of the abs , which leads to the periodical switching between the states with and without current .
interface transparency also leads to such periodical switching of the current @xcite , however , at the same time it also kills the proximity effect , suppressing the current and changing the period of the pairing amplitude oscillations .
no such effects are caused by the inclusion of the normal metal slab .
perhaps the effect of the n slab would be more similar to the effect of reduced interface transparency if the normal metal was in the dirty regime .
in conclusion we have studied the ground state properties of f - n - s proximity system . we have observed oscillatory behavior of the superconducting pairing amplitude in ferromagnet , but not in the normal region .
similarly to f - s structures , we have found spontaneously generated currents flowing in the whole f and n regions and within a distance of a few @xmath44 on superconducting side .
interestingly , the system can be switched periodically ( changing @xmath32 ) between the states with and without the spontaneous current .
all this suggests possible realization of the proximity induced fflo state in f - n - s heterostructure .
this work has been partially supported by the grant no .
pbz - min-008/p03/2003 , and by the esf network programme aqd jj .
g. e. w. bauer _
et al . _ , materials sci .
b * 84 * , 31 ( 2001 ) ; l. r. tagirov , phys . rev
. lett . * 83 * , 2058 ( 1999 ) .
g. blatter _
et al . _ ,
b * 63 * , 174511 ( 2001 )
. n. f. berk , j. r. schrieffer , phys .
* 17 * , 433 ( 1996 ) ; c. pfleiderer _ et al .
_ , nature * 412 * , 58 ( 2001 )
. h. k. wong _ et al .
_ , j. low temp
. phys . * 63 * , 307 ( 1986 ) ; j. s. jiang _
et al . _ ,
. lett . * 74 * , 314 ( 1995 ) ; th .
et al . _ ,
lett . * 77 * , 1857 ( 1996 ) .
t. kontos _
et al . _ ,
* 86 * , 304 ( 2001 ) .
a. i. buzdin _
et al . _ , jetp lett . * 35 * , 178 ( 1982 ) ; e. a. demler _ et al . _ ,
b * 55 * , 15174 ( 1997 ) ; e. vecino _ et al . _ ,
b * 64 * , 184502 ( 2001 ) ; k. halterman , o. t. valls , phys . rev .
b * 65 * , 014509 ( 2002 )
. v. v. ryazanov _ et al .
86 * , 2427 ( 2001 ) ; s. m. frolov _
et al . _ ,
b * 70 * , 144505 ( 2001 ) . v.
t. petrashov _
lett . * 83 * , 3281 ( 1999 ) .
l. r. tagirov , phys .
lett . * 83 * , 2058 ( 1999 ) .
m. krawiec _
et al . _ ,
b * 66 * , 172505 ( 2002 ) ; physica c * 387 * , 7 ( 2003 ) ; eur .
j. b * 32 * , 163 ( 2003 ) .
m. krawiec _
et al . _ ,
b * 70 * , 134519 ( 2004 ) .
b. l. gyrffy _
et al . _ , in _ physics of spin in solids : materials , methods and applications _ ,
s. halilov , kluwer academic publ .
y. a. izyumov _
et al . _ ,
45 * , 109 ( 2002 ) ; i. f. lyuksyutov , v. l. pokrovsky , adv .
phys . * 54 * , 67 ( 2005 ) .
a. i. buzdin , cond - mat/0505583 ; f. s. bergeret _
et al . _ , cond - mat/0506047 .
p. fulde , r. a. ferrell , phys . rev . *
135 * , a550 ( 1964 ) ; a. i. larkin , y. n. ovchinnikov , sov .
jetp * 20 * , 762 ( 1965 ) . | we discuss the ground state properties of the system composed of a normal metal sandwiched between ferromagnet and superconductor within a tight binding hubbard model . we have solved the spin - polarized hartree - fock - gorkov equations together with the maxwell s equation ( ampere s law ) and found a proximity induced fulde - ferrell - larkin - ovchinnikov ( fflo ) state in this system . here
we show that the inclusion of the normal metal layer in between those subsystems does not necessarily lead to the suppression of the fflo phase .
moreover , we have found that depending on the thickness of the normal metal slab the system can be switched periodically between the state with the spontaneous current flowing to that one with no current .
all these effects can be explained in terms of the andreev bound states formed in such structures . |
weakly interacting massive particles ( wimps ) are important study targets in current and future collider experiments since the wimps provide a solution to the dark matter problem . in this study , we investigate the prospects of discovering the direct pair production of wimps at the international linear collider ( ilc ) using the energy and angle spectra of the photons from initial state radiation ( isr ) @xcite .
we make the assumption that , among the particles beyond the standard model , the wimps are the only kinematically accessible at the center - of - mass ( cm ) energy of 500 gev proposed for the ilc .
similar past searches using single photons have been performed by e.g. @xcite .
prospects of such measurements have been discussed in many literature @xcite and , in particular , at the ilc @xcite which uses a cut - and - count approach .
our method uses the two - dimensional energy and angle distributions of the isr photon and thus exploits the full information of the event .
we apply this method to an example case of minimal supergravity ( msugra ) @xcite in which the neutralino is the lightest supersymmetric particle ( lsp ) .
the msugra parameters have been chosen so that the lsp becomes bino - like ; under this assumption , the dominant contribution to the neutralino pair production becomes the @xmath0-channel diagram shown in figure [ fig : tchannel ] .
-channel ( left ) and @xmath0-channel ( right ) . ] -channel ( left ) and @xmath0-channel ( right ) . ]
we perform a likelihood analysis based on the photon distributions and derive cross - section limits .
we assume an integrated luminosity of 500 fb@xmath1 at @xmath2 gev .
signal events have been generated using whizard @xcite .
a geant4-based @xcite simulation of the detector response was carried out by mokka @xcite using the international large detector ( ild ) concept @xcite .
for the background studies , we used the full standard model samples which were generated for the ild letter of intent @xcite .
signal samples were generated with two different lsp masses by varying the msugra parameter @xmath3 , resulting in lsp masses of approximately 150 and 200 gev .
the other msugra parameters have been fixed to the following values : @xmath4 gev , @xmath5 , @xmath6 and @xmath7 .
the mass spectra have been computed using softsusy @xcite ; the relevant values are summarized in table [ tab : ne1ne1a - masses - params ] .
.mass spectra of signal samples .
[ cols="<,^,^",options="header " , ]
we rely on the pandora particle flow algorithm @xcite for the reconstruction of calorimeter clusters and charged particles .
a single highly energetic photon often results into multiple clusters . in order to recover the split clusters
, we merge the clusters which lie within a cone of an opening angle of 1.5 degree around the most energetic cluster . for the event selection ,
we require that there are no charged particles in the event . in addition , we require that there are no other clusters aside from the merged cluster . the merged cluster is required to have an energy greater than 0.5 gev . for the particle identification
, we take the energy deposits in the electromagnetic calorimeter @xmath8 and the hadronic calorimeter @xmath9 to compute the energy fraction @xmath10 which is required to be greater than 0.9 . the reconstruction efficiency of signal events as a function of energy and angle
is shown in figure [ fig : ne1ne1a - eff ] .
the overall signal efficiency is 94% after the event selection .
we perform a binned likelihood analysis using the two dimensional distribution of the photon energy ( logarithmic scale ) and angle .
we choose a binning such that the number of bins is @xmath11 , with the binning widths shown in figure [ fig : binning ] .
gev ( left ) , @xmath12 gev ( center ) , and background events ( right ) .
the horizontal axis is the photon energy in logarithmic scale .
the vertical axis is the photon angle in radians , folded around @xmath13 .
the binning is shown as red lines ; the lower - left corner is cut off to make the binning consistent with the cuts in the background sample .
, title="fig : " ] gev ( left ) , @xmath12 gev ( center ) , and background events ( right ) .
the horizontal axis is the photon energy in logarithmic scale .
the vertical axis is the photon angle in radians , folded around @xmath13 .
the binning is shown as red lines ; the lower - left corner is cut off to make the binning consistent with the cuts in the background sample .
, title="fig : " ] gev ( left ) , @xmath12 gev ( center ) , and background events ( right ) .
the horizontal axis is the photon energy in logarithmic scale .
the vertical axis is the photon angle in radians , folded around @xmath13 .
the binning is shown as red lines ; the lower - left corner is cut off to make the binning consistent with the cuts in the background sample .
, title="fig : " ] the choice of @xmath14 instead of @xmath15 is motivated by the fact that the isr photon distribution diverges at low energy .
the angle is folded around @xmath16 , to exploit the symmetry of the distribution .
the bin size is adjusted to ensure the presence of sufficient signal events in each bin .
the log likelihood ratio is defined as follows : @xmath17 where @xmath18 is the observed number of events in the @xmath19-th bin , @xmath20 is the expected number of background events in the @xmath19-th bin , @xmath21 is the expected number of signal events in the @xmath19-th bin , and @xmath22 is the poisson probability density function with the expectation value of @xmath23 . the @xmath24 gev sample was used to build the template signal shape ; the same signal shape was used to build likelihood functions even for different masses in the subsequent analysis , since the true mass is not known _ a priori_.
the likelihood distributions are obtained by performing monte - carlo test experiments whereby the observed number of events in each bin is fluctuated using poissonian random numbers .
the resulting likelihood distributions for background plus signal events ( @xmath25 ) are compared with that for background events only ( @xmath26 ) to compute the probability for observing the wimp signal , as shown in figure [ fig : ne1ne1a - likelidist ] .
gev ( green ) , and background and signal with @xmath24 gev ( blue ) .
the vertical axis is the probability density for a given log - likelihood value . ] from this distribution , we calculate the expected significance for each neutralino mass .
we estimate the significance by computing the probability that the backgrounds fluctuate to deviate more than the median of the signal likelihood . in the case of @xmath12 gev ,
the estimated significance is 3.1@xmath27 ; in the case of @xmath24 gev , the estimated significance is well above 5@xmath27 .
the difference in the significance between the two cases is mostly due to the difference in the cross section . in the case of a more general wimp scenario ,
the dependence of the signal cross section on the wimp mass is expected to be considerably different . in order to take this effect into account
, we performed a scan of likelihood values by varying the wimp mass and the cross section independently .
the best signal separation power can be attained when the correct signal template function is used ; assuming this is the case , we keep our msugra signal template .
additional signal samples were generated using 15 different wimp mass points ranging from 100 gev to 240 gev by varying @xmath3 .
for this study , we use the beam polarizations of @xmath28 .
we obtain the cross section limit , defined as the cross section which gives a 50% probability of 5@xmath27 observation , as a function of the wimp mass .
integrated over the kinematically allowed region in this analysis . ] for a given mass , we adjusted the cross section until it yielded the desired probability of observing the signal .
we performed @xmath29 monte - carlo test experiments at each step of testing the cross section .
a binary search was used to find the cross section limit efficiently .
the cross section limit as a function of the wimp mass is plotted in figure [ fig : ne1ne1a - mscan ] . if the cross section is at least 14 fb , the wimp can be observed . around the mass of 200 gev , the required cross section increases since the likelihood function
is constructed using the template signal shape for @xmath24 gev .
as the wimp mass approaches 250 gev , the required cross section becomes smaller since the photon energy distribution peaks at the low energy region , thereby allowing it to be discriminated against the background shape .
note , however , that if one fixes the model and let the mass approach 250 gev , the cross section decreases faster than the cross - section limit .
thus , this result is consistent with the fact that the heavier neutralino is more difficult to observe .
we have looked at the msugra neutralino search as an example of a wimp search with single photons using the two - dimensional distribution of the isr photon energy and the angle .
we compared the likelihood distributions and calculated the probability to achieve 5@xmath27 observation . in our msugra model ,
the @xmath24 gev case can be easily observed , while the @xmath12 gev case has the significance of 3.1@xmath27 .
we obtained cross - section limits as a function of the neutralino mass .
wimps created through the @xmath0-channel with the cross section of over 15 fb can be observed with the 5@xmath27 significance at the ilc with @xmath30 gev . in principle , this method can be applied to a broader class of wimp searches with isr photons .
the authors would like to thank all members of the ilc physics subgroup @xcite for useful discussions .
this study is supported in part by the creative scientific research grant no .
18gs0202 of the japan society for promotion of science .
99 presentation : http://ilcagenda.linearcollider.org/contributiondisplay.py?contribid=218&sessionid=17&confid=4175 j. l. lopez , d. v. nanopoulos and a. zichichi , phys .
d * 55 * , 5813 ( 1997 ) [ arxiv : hep - ph/9611437 ] .
m. z. akrawy _ et al . _
[ opal collaboration ] , phys .
b * 248 * , 211 ( 1990 ) .
p. fayet , phys .
b * 117 * , 460 ( 1982 ) ; p. fayet , phys .
b * 175 * , 471 ( 1986 ) .
d. fargion , m. y. khlopov , r. v. konoplich and r. mignani , phys .
d * 54 * , 4684 ( 1996 ) .
a. birkedal , k. matchev and m. perelstein , phys .
d * 70 * , 077701 ( 2004 ) [ arxiv : hep - ph/0403004 ] .
p. konar , k. kong , k. t. matchev and m. perelstein , new j. phys .
* 11 * , 105004 ( 2009 ) [ arxiv:0902.2000 [ hep - ph ] ] . c. bartels and j. list , _ in the proceedings of 2007 international linear collider workshop ( lcws07 and ilc07 ) , hamburg , germany , 30 may - 3 jun 2007 , pp cos02 _ [ arxiv:0709.2629 [ hep - ex ] ] ; c. bartels and j. list , arxiv:0901.4890 [ hep - ex ] .
a. h. chamseddine , r. l. arnowitt and p. nath , phys .
lett . * 49 * , 970 ( 1982 ) .
e. cremmer , p. fayet and l. girardello , phys .
b * 122 * , 41 ( 1983 ) .
r. barbieri , s. ferrara and c. a. savoy , phys .
b * 119 * , 343 ( 1982 ) ; l. j. hall , j. d. lykken and s. weinberg , phys . rev .
d * 27 * , 2359 ( 1983 ) ; p.
nath , r. l. arnowitt and a. h. chamseddine , nucl .
b * 227 * , 121 ( 1983 ) .
w. kilian , t. ohl and j. reuter , arxiv:0708.4233 [ hep - ph ] ; m. moretti , t. ohl and j. reuter , arxiv : hep - ph/0102195 .
s. agostinelli _ et al . _
[ geant4 collaboration ] , nucl .
instrum .
a * 506 * , 250 ( 2003 ) .
ilc software portal , http://ilcsoft.desy.de desy 2009/87 , fermilab pub-09 - 682-e , kek report 2009 - 6 , _ international large detector - letter of intent _ , issn 0418 - 9833 , isbn 978 - 3 - 935702 - 42 - 3 . b. c. allanach , comput .
commun . * 143 * , 305 ( 2002 ) [ arxiv : hep - ph/0104145 ] .
m. a. thomson , nucl .
instrum .
a * 611 * , 25 ( 2009 ) [ arxiv:0907.3577 [ physics.ins-det ] ] .
ilc physics subgroup : http://www-jlc.kek.jp/subg/physics/ilcphys/ | we consider analysis targets at the international linear collider in which only a single photon can be observed . for such processes , we have developed a method which uses likelihood distributions using the full event information ( photon energy and angle ) .
the method was applied to a search for neutralino pair production with a photon from initial state radiation ( isr ) in the case of supergravity in which the neutralino is the lightest supersymmetric particle .
we determine the cross section required to observe the neutralino pair production with isr as a function of the neutralino mass in the range of 100 to 250 gev . |
the problem of finding the densest configurations of identical hard spheres inside a cylinder is related to many different aspects in science and engineering : similar densest structures ( e.g. zigzags and helices ) have in recent years been observed for the self - assembly of nano - spheres inside carbon nanotubes @xcite , charged particles on cylindrical surfaces @xcite , micro - spheres inside micro - channels @xcite and colloidal particles inside cylindrical channels @xcite , and also for the formation of colloidal crystal wires @xcite , the reason being that in all such cases the particles close - pack themselves like hard spheres .
relevant problems of interest include ( i ) finding a theoretical derivation of the corresponding densest packings of hard spheres and ( ii ) finding a general methodology for constructing them in practice .
for this problem , a wide range of chiral and achiral densest structures for a diameter ratio of @xmath1 cylinder diameter / sphere diameter @xmath2 $ ] have previously been found by simulated annealing @xcite and by other computational means @xcite , where @xmath3 is a critical value of @xmath4 below which the densest configurations are entirely made up of cylinder - touching spheres . in particular ,
the densest structures for @xmath5 $ ] , which have their entire structural information contained on the cylinder surface , can be described by a phyllotactic scheme @xcite that is relevant to the study of plant morphology .
$ ] : ( a ) the achiral planar zigzag at @xmath6 , ( b ) the chiral single helix at @xmath7 , ( c ) the chiral double helix at @xmath8 , and ( d ) the achiral doublets , which have two spheres located on one @xmath9 position , at @xmath10 . for each case , the required periodicity @xmath11 along @xmath12 is indicated by a solid arrow . , width=272,height=204 ] in this article , it is shown that @xmath13 the densest structures for @xmath0 $ ] can surprisingly be constructed from a simple and computationally fast sequential deposition of cylinder - touching spheres which maximizes the number of spheres per unit length everywhere along the cylinder axis ( typically less than 15 minutes of computational time for each value of @xmath4 ; much faster than simulated annealing which sometimes can take up a whole week @xcite ) . as demonstrated in fig .
[ phyllo_d_leq_2 ] for @xmath14 $ ] , the first four phases constructed from such a deposition procedure are in complete agreement with the planar zigzag , single helix , double helix and achiral doublets predicted from simulated annealing @xcite .
the findings show that all these densest packings can be derived theoretically by means of maximizing the abovementioned number density of spheres , and establish sequential deposition as a possible methodology for constructing such densest packings with the aforementioned physical systems . against the arc length @xmath15 for spheres in contact at various values of the diameter ratio @xmath4 .
the regimes @xmath16 and @xmath17 correspond to the absence and presence of a solution for @xmath18 , respectively .
as @xmath19 , the relation between @xmath20 and @xmath15 approaches @xmath21 for the 2-d packing of circular discs .
, width=251,height=188 ] this 3-d problem of packing identical spheres onto the inner surface of a cylinder is equivalent to a problem of packing objects in 2-d space , and is related to the 2-d problem of packing circular discs : in cylindrical polar coordinates @xmath22 with a radial position @xmath23 and an arc length @xmath24 , where the length quantities @xmath25 , @xmath9 and @xmath12 are all expressed in units of the sphere diameter , the equation for two such spheres in contact is given by @xmath26 which involves only two position variables ( i.e. the differences in @xmath9 and @xmath27 between the two spheres in contact ) , corresponding to a 2-d problem , and can be derived easily using trigonometric identities .
this implies the following ( fig .
[ contact_eq ] ) : ( i ) for @xmath16 , the cylinder s surface is so curved that no two spheres can be accommodated on the same @xmath9 position , i.e no solution exists for @xmath18 .
( ii ) as @xmath28 , the cylindrical surface becomes flat and the equation for spheres in contact approaches that for the 2-d packing of circular discs ; in the densest configuration , the sphere centres form a perfect triangular lattice on the phyllotactic plot @xcite of @xmath9 against @xmath12 such that every sphere is in contact with 6 other spheres .
( iii ) for any finite value of @xmath4 , the problem of finding an analytic solution to the densest configuration is complicated by the requirement of a periodicity @xmath29 along the @xmath12 axis @xcite and by the relatively complex mathematical form of eq .
( [ eq_1 ] ) . finding an analytic solution for cases of finite @xmath4 thus remains a challenging theoretical problem . to gain insights into such an analytic solution ,
a semi - analytic approach is adopted here , based on the following conjecture : for any given value of @xmath4 , if the packing of spheres is locally densest everywhere in the columnar structure , the overall structure would be at its globally densest configuration . for a sufficiently long column of spheres ,
let @xmath30 be the average number density of spheres ( number of spheres per unit length ) along @xmath9 , where @xmath31 are the indices of the spheres in ascending order of their @xmath9 positions .
this number density is related to the volume fraction , @xmath32 volume of spheres / volume of cylinder , as follows : @xmath33 thus the densest columnar structure , which by definition has the greatest volume fraction , also corresponds to the densest packing of spheres along @xmath9 .
it follows , from the abovementioned conjecture , that the densest columnar structure can be constructed from a sequential deposition of individual spheres onto the lowest possible @xmath9 positions ( determined across the entire @xmath34 range of @xmath27 ) at which they would come into contact with one or more other spheres .
this is because such an iterative process would ensure the packing of spheres along @xmath9 to be densest possible at every @xmath9 location .
a template comprised of spheres , which covers the entire @xmath34 range of @xmath27 , is needed as a base for the initial deposition of spheres . for @xmath35 $ ] ,
( [ eq_1 ] ) for spheres in contact is valid for any value of @xmath36 so that , among other possible choices of a multi - sphere template , a single sphere is already enough to make up such a template .
for @xmath37 , however , this equation is valid only for a limited range of @xmath36 , which satisfies the condition @xmath38 $ ] , so that a template comprised of more than one sphere is necessary . in general , for any given value of @xmath4 , the columnar structures of spheres generated from the abovementioned deposition procedure depend on their underlying templates , and the densest structure can be obtained by varying across a diversity of templates . against the diameter ratio @xmath4 , showing a complete overlap of data points between this deposition algorithm and simulated annealing ( data supplied by mughal @xcite ) .
the inset shows that the differences in @xmath39 between the two algorithms are of the order of @xmath40 , i.e. less than 1% of percentage differences , and thus comparable to the numerical uncertainties in the simulations .
the occasional peaks in @xmath41 are due to the relatively larger numerical uncertainties in the proximity of a structural transition .
, width=272,height=204 ] the different templates are generated through a deposition scheme in which the template structures depend only on the self - assigned relative positions @xmath42 $ ] and @xmath43 $ ] between the two initially placed spheres : after the first sphere is placed at an arbitrary position on the plane of @xmath44 coordinates , a particular value of @xmath45 is chosen for the second sphere , where the value of @xmath46 is given by the corresponding solution to eq .
( [ eq_1 ] ) .
if such a solution does not exist , the value of @xmath46 is assigned separately . along the same deposition direction of increasing or decreasing @xmath27 , which determines the chirality of a chiral structure ,
additional spheres are deposited if the existing template as made up of these two initially filled spheres does not cover the @xmath34 range of @xmath27 .
inspired by the relevant triangular - lattice - like phyllotactic patterns @xcite , the positions of such additional spheres are chosen in the spirit to maintain a continuity in the relative positions @xmath47 and @xmath48 of successively filled spheres : for @xmath49 , the angular position of the @xmath50 sphere is set to be @xmath51 , and the sphere is deposited in a @xmath9 position at which it comes into contact with one or more other spheres . if such a @xmath9 position does not exist ,
the sphere is placed at @xmath52 .
deposition of such additional spheres stops whenever the template created has covered the @xmath34 range of @xmath27 .
the iterative process of sequential deposition of spheres then starts from the angular position of the last deposited sphere in the process of template creation : in each iterative step , a sphere is deposited onto the lowest possible @xmath9 position at which it would come into contact with one or more other spheres .
starting from the angular position of the previously filled sphere , this lowest @xmath9 position is determined via a @xmath34 scan along the same deposition direction of increasing or decreasing @xmath27 as that for the process of template creation .
should there be degeneracy where this lowest @xmath9 position occurs at more than one angular position , the first such angular position encountered in the @xmath34 scan is chosen .
after a sufficiently long ( e.g. 10 times of the sphere diameter ) column of spheres has been generated , the average number density @xmath30 is evaluated via a linear fit of @xmath31 against @xmath9 , and the volume fraction is computed using eq .
( [ eq_2 ] ) .
the abovementioned procedure is repeated for a continuous range of @xmath45 from 0 to @xmath53 , and for a continuous range of @xmath46 from 0 to 1 whenever the value of @xmath46 has to be assigned separately . as such ,
a quantitative agreement in the maximum volume fraction @xmath39 between simulated annealing and this deposition algorithm has been obtained for the whole range of @xmath0 $ ] , which is demonstrated for @xmath54 $ ] in fig .
[ max_vf_vs_d ] ( the numerical resolution of @xmath4 is set to be 0.001 , except for a better resolution near @xmath55 ) .
increases from 2.040 to 2.155 : 321 symmetric at @xmath56 @xmath57 type 2 line slip , e.g. at @xmath58 @xmath57 type 3 line slip , e.g. at @xmath59 @xmath57 330 symmetric at @xmath60 .
( b ) two examples of periodicity breaking at @xmath61 and @xmath62 ; in each case , the region that does not belong to the periodic structure is highlighted with a rectangle .
, title="fig:",width=272,height=204 ] increases from 2.040 to 2.155 : 321 symmetric at @xmath56 @xmath57 type 2 line slip , e.g. at @xmath58 @xmath57 type 3 line slip , e.g. at @xmath59 @xmath57 330 symmetric at @xmath60 .
( b ) two examples of periodicity breaking at @xmath61 and @xmath62 ; in each case , the region that does not belong to the periodic structure is highlighted with a rectangle .
, title="fig:",width=272,height=102 ] and @xmath63 spheres at regular spacings of @xmath27 in each deposition step . all the metastable curves for lower - density packings
correspond to cases of @xmath64 .
the symmetric structures are indicated along with their phyllotactic notations .
, width=251,height=188 ] a complete agreement in the predicted densest structures has also been obtained for @xmath0 $ ] between the two algorithms . for @xmath65 $ ] the following sequence of densest structures
has previously been predicted from simulated annealing @xcite : ( i ) the achiral planar zigzag for @xmath66 $ ] , ( ii ) the chiral single helix for @xmath67 $ ] , ( iii ) the chiral double helix for @xmath68 , and ( iv ) the achiral doublets , which have two spheres located on one @xmath9 position , at @xmath10 . here
, the deposition algorithm yields the same sequence of densest structures for the same regimes of @xmath4 .
a gallery of the corresponding phyllotactic diagrams is shown in fig .
[ phyllo_d_leq_2 ] . in creating such diagrams ,
the boundaries of the equivalent close - packed objects in 2-d space are plotted from an equation which comes from the following transformation of eq .
( [ contact_eq ] ) : @xmath69 and @xmath70 .
this can be easily understood from the limiting case of @xmath28 : applying such a transformation to @xmath71 for the 2-d packing of circular discs correctly yields the equation for a circle of radius = @xmath72 . for @xmath73 $ ]
, the phyllotactic patterns from simulated annealing generally consist of regions with triads of interconnected spheres , which can be described in terms of a distorted triangular lattice @xcite , as well as line defects that arise due to the required periodicity @xmath11 along @xmath12 .
these are referred to as _ line - slip _
structures @xcite . at certain values of @xmath74 or equivalently @xmath4 ,
line defects are absent so that on a phyllotactic diagram every sphere appears to be in contact with 6 other spheres .
such defect - free structures , which generally correspond to the density maxima in fig .
( [ max_vf_vs_d ] ) , are described as _
symmetric _ @xcite .
in such cases , the periodicity vector is given by @xmath75 where @xmath76 and @xmath77 are any two of the three basis vectors .
the coefficients @xmath78 and @xmath79 are by definition chosen from the indices in the phyllotactic notation @xmath80 @xcite , where @xmath81 .
an example of such symmetric structures is the 220 structure in fig .
[ phyllo_d_leq_2](d ) . for this regime of @xmath4 ,
the densest structures predicted from the deposition algorithm are in complete agreement with the line - slip and symmetric structures predicted from simulated annealing .
for example , as shown in fig .
[ phyllo_d_geq_2](a ) , the deposition algorithm yields the same sequence of structures as @xmath4 increases from 2.040 to 2.155 @xcite : 321 symmetric at @xmath56 @xmath57 type 2 line slip for
@xmath82 @xmath57 type 3 line slip for @xmath83 @xmath57 330 symmetric at @xmath60 , where the different types of line - slip structures are defined in ref .
@xcite . in some cases
a periodic line - slip structure comes into existence only after a few spheres have been deposited ; two examples of such periodicity breaking are demonstrated in fig .
[ phyllo_d_geq_2](b ) .
lower - density packings , which might also form a subject of interest , can be generated from a simultaneous deposition of multiple spheres at each deposition step ( fig .
[ simul_depos ] ) . for @xmath84 , the structures with cylinder - touching spheres
become hollow so that the aforementioned agreement with the 3-d densest structures from simulated annealing no longer exists . at @xmath19 ,
the cylinder s surface becomes flat and essentially a 2-d triangular lattice of spheres is obtained .
a sequential stacking of such 2-d layers along their perpendicular direction in ab or abc sequence yields respectively the densest 3-d packings of hcp or fcc ( @xmath85 , the difference of which with the peak values in figs .
[ max_vf_vs_d ] and [ simul_depos ] clearly reflects the boundary effects imposed by the cylinder ) . for low values of @xmath4 the experimental observations of the planar
zigzag @xcite , the helical phases @xcite and the achiral doublets @xcite could be related to the following : ( i ) the particles self - assemble like hard spheres , in the sense that their shapes remain approximately spherical . (
ii ) during the process of structural formation , the particles are driven by a uni - directional force ( e.g. gravity ) which helps to maximize the number density @xmath30 . in the future a more general deposition scheme that also applies to the regime of @xmath84 and to the packing of binary hard spheres @xcite should be developed .
possibilities of applying the current algorithm for the synthesis of chiral molecules should also be explored .
a. n. khlobystov , d. a. britz , a. ardavan and g. a. d. briggs , phys .
* 92 * , 245507 ( 2004 ) .
s. srebnik and j. f. douglas , soft matter * 7 * , 6897 ( 2011 ) .
s. a. vanapalli , c. r. iacovella , k. e. sung , d. mukhija , j. m. millunchick , m. a. burns , s. c. glotzer and m. j. solomon , langmuir * 24 * , 3661 ( 2008 ) . m. a. lohr , a. m. alsayed , b. g. chen , z. zhang , r. d. kamien and a. g. yodh , phys . rev .
e * 81 * , 040401(r ) ( 2010 ) .
m. tymczenko , l. f. marsal , t. trifonov , i. rodriguez , f. ramiro - manzano , j. pallares , a. rodriguez , r. alcubilla and f. meseguer , adv .
mater * 20 * , 2315 ( 2008 ) .
m. hodak and l. a. girifalco , phys .
b * 67 * , 075419 ( 2003 ) .
g. t. pickett , m. gross and h. okuyama , phys .
lett . * 85 * , 3652 ( 2000 ) .
a. mughal , h. k. chan and d. weaire , phys .
lett . * 106 * , 115704 ( 2011 ) .
f. j. dur@xmath86n - olivencia and and m. c. gordillo , phys .
e * 79 * , 061111 ( 2009 ) .
k. koga and h. tanaka , j. chem . phys . *
124 * , 131103 ( 2006 ) .
private communication with adil mughal .
a. b. hopkins , y. jiao , f. h. stillinger and s. torquato , phys .
lett . * 107 * , 125501 ( 2011 ) h. j. h. brouwers , phys .
e * 76 * , 041304 ( 2007 ) . | the rich variety of densest columnar structures of identical hard spheres inside a cylinder can surprisingly be constructed from a simple and computationally fast sequential deposition of cylinder - touching spheres , if the cylinder - to - sphere diameter ratio is @xmath0 $ ] .
this provides a direction for theoretically deriving _ all _ these densest structures and for constructing such densest packings with nano- , micro- , colloidal or charged particles , which all self - assemble like hard spheres . |
quantum dots are semiconductor nanocrystals , 1 - 20 nm in size @xcite , that possess a discrete energy level structure .
the level structure is determined by the size , composition and shape of the quantum dot @xcite .
particular focus has been placed on ii - vi semiconductors , such as cdse @xcite . by selecting for size
, the fluorescence emission can be tailored to the application , which is attractive for use in biological labels @xcite .
in addition , the quantum dot surface can be chemically functionalized to bind to specific biomolecules . compared to other fluorescent molecules ,
quantum dots possess high brightness , low bleaching rate and absorption over a wide wavelength range @xcite .
similar to the fluorescence from other labels , quantum dot fluorescence can be driven by both single- and multiphoton processes @xcite .
fluorescent labels are conventionally excited by short wavelength light to produce fluorescence emissions .
ultraviolet ( uv ) light is often used with quantum dots due to their absorption that increases towards shorter wavelengths@xcite .
however uv is not ideal for use in living cells , due to strong absorption and scattering of uv light by surrounding biological tissue .
the use of shorter wavelengths leads to a higher potential for photochemical reactions and an increased possibility for the decomposition and subsequent release of component heavy metals , such as cadmium ( cd ) @xcite . through excitation by multiphoton processes ,
it is possible to avoid the unwanted effects from exposure to shorter wavelengths in biological applications .
two - photon excitation of quantum dots is now routinely performed , often using ti : sapphire lasers operating around 700 - 1000 nm @xcite . for imaging applications , it is often important for the excitation light to penetrate far into the sample , but the penetration depth is limited by rayleigh scattering due to the effect on beam quality . since rayleigh scattering scales as @xmath0 , long - wavelength multiphoton excitation seems promising for bioimaging applications . in a recent study , horton et al .
@xcite demonstrated in - vivo three - photon microscopy with an er - doped fiber system and red fluorescent protein labels .
they found that an optimum wavelength window due to scattering and absorption in tissue lies close to the telecom band .
quantum dots are attractive for use in such applications due to their large three - photon absorption cross - section @xcite .
we demonstrate three - photon excitation and fluorescence in commercially available cdse quantum dots using a custom built mode - locked er - doped fiber laser ( edfl ) .
such lasers , and their associated optical components , are central to telecommunications and have therefore been the subject of extensive industrial research .
these developments mean that they provide a stable , versatile and comparatively inexpensive method to generate ultrashort pulses for three - photon spectroscopy of quantum dots .
we also excited single - photon fluorescence using a 372 nm diode laser to compare one- and three - photon fluorescence emission spectra , excitation intensity dependence and bleaching rate .
previous investigations have also demonstrated two- and three - photon excitation in quantum dots @xcite .
however , to our knowledge , multiphoton excitation and absorption have not yet been investigated in quantum dots for excitation wavelengths in the telecom band .
the wavelength of expected fluorescence emissions from the sample is determined by the quantum dot energy level structure , which can be simplified to a ground state , fluorescent excited state and a continuum of absorption states @xcite . upon excitation to this continuum of states ,
a quantum dot undergoes a non - radiative decay to the excited state , followed by fluorescent decay back to the ground state@xcite .
in addition , intense excitation pulses will interact nonlinearly with the sample to generate harmonics of the excitation wavelength . in quantum dots , the third - harmonic generation ( 3hg ) process
is expected to be the most significant of these @xcite .
the process of second harmonic generation ( shg ) is negligible in comparison to 3hg due to the centrosymmetric crystal structure of cdse @xcite .
as the product of a coherent process , the harmonics are emitted in line with the excitation beam .
experimental setup . schematic for spectroscopy of quantum dots . ( a ) 372 nm light is focused onto the quantum dot sample and fluorescence is collected and coupled into a multimode fiber .
fluorescence is separated from residual excitation light using a semrock ff511-di01 dichroic mirror .
( b ) 1575 nm pulses are focused onto the quantum dot sample and fluorescence is collected off - axis to avoid excessive coupling of third harmonic ( 3hg ) light .
the majority of residual excitation and 3hg light continues to propagate on - axis past the slide and is not coupled into the multimode fiber ( mmf ) . as some scattered light is still collected , kg3 coloured glass filters are used to block any remaining ir.,width=264 ] figure [ fig1-setup ] shows the experimental setup used to probe the quantum dots .
colloidal invitrogen lumidot cdse / zns 640 core - shell quantum dots , with a nominal fluorescence wavelength of 640 nm , were deposited on a sample slide .
the toluene solvent was evaporated off , leaving a static quantum dot sample which was placed at the focus of the excitation beam .
collected fluorescence was coupled into a multimode fiber , where the spectrum was measured by a fiber - coupled spectrometer .
for the three - photon spectroscopy of the quantum dot sample , a mode - locked edfl was used to produce 1564 nm pulses at a repetition rate of 300 mhz .
116 fs pulses with a maximum average power of 186 mw were produced through er - doped fiber amplification and soliton compression .
the laser was focused to a spot with a @xmath1 @xmath2 radius , yielding an average intensity up to @xmath3 .
the soliton self - frequency shift ( ssfs ) @xcite of the amplified pulses shifted the center wavelength to 1575 nm , increasing the spectral bandwidth to 85 nm .
the pulse durations were measured by frequency resolved optical gating ( frog ) using a mesaphotonics frogscan with an ocean optics hr2000 + spectrometer , and a wavelength range of 700 - 881 nm .
the spectrum was measured using a hp 70950a optical spectrum analyser .
the resulting fluorescence emitted from the quantum dots was collected off - axis with a 35 mm focal length lens to spatially separate it from the large amount of light produced by 3hg .
however , some of the 3hg and excitation light was scattered by the sample and collected .
the fiber - coupled fluorescence was detected by an ocean optics usb650 spectrometer .
three 2 mm thick kg3 coloured - glass filters were used to block any remaining ir , with a calculated attenuation of @xmath4 db , in order to prevent damage to the detector .
the fluorescence detection spectrometer was used to capture all data for the fluorescence and 3hg spectra , and its dependence on the intensity of the excitation beam and exposure time .
measurements of the spectra were taken for different input intensities using a variable nd filter , and integration times from 5 - 30 s were used .
the relative spectrum amplitudes were inferred from gaussian fit profiles of the spectra . to check that the quantum dots behaved
as expected under uv excitation , we applied up to 3.5 mw of 372 nm light from a fiber - coupled nichia diode laser to the quantum dot sample .
basic optical isolation was employed , using a quarter waveplate and linear polariser , to ensure that the operation of the diode laser was not compromised by optical feedback from the single mode fiber .
the uv light was focused to a spot with a @xmath5 @xmath2 radius at the quantum dot sample , with a maximum intensity of @xmath6 .
the fluorescence excited by the uv laser was collected by an @xmath7 aspheric lens with a 4.02 mm focal length .
this was separated from the residual excitation light by a semrock ff511-di01 dichroic mirror , with an edge wavelength of 511 nm , and coupled into mmf .
an ocean optics qe65000 spectrometer , with a 316 - 1021 nm wavelength range and 0.22 nm / pixel resolution , was used to measure the spectrum of the uv - induced fluorescence .
measurements were taken with varying input intensities and an integration time of 10 ms .
comparison of spectra from ( a ) @xmath8=372 nm excitation and ( b ) @xmath8=1575 nm excitation , shown with the energy level structures and excitation pathways . in ( b ) , both the 640 nm fluorescence peak and a 540 nm 3hg peak are observed .
( c ) spectrum of ir excitation beam on a linear scale , where the wavelength axis aligns at a ratio of 3:1 to that of ( b ) , showing the correlation of the 3hg peak with the fundamental .
( d ) frog trace of ir pulse , where the wavelength axis refers to the second harmonic of the excitation wavelength.,width=264 ] figure [ fig2-spectra ] shows the spectra from uv and ir excitation . from a gaussian fit of the fluorescence spectra ,
the center wavelength from linear excitation is @xmath9 , where it is @xmath10 in the spectrum produced by three - photon excitation .
the fwhm linewidths are determined to be @xmath11 and @xmath12 , respectively .
the near - identical characteristics of the fluorescence spectra under uv and ir excitation indicate that the same excited energy band is reached in both cases . in the three - photon excitation spectrum ,
the secondary peak at @xmath13 corresponds with the third harmonic of the ir excitation pulse .
the peak is visible in the off - axis configuration due to scattering from the quantum dot sample . as shown in figure [ fig2-spectra ] , the third harmonic
is generated predominantly from the red - shifted soliton at the long - wavelength end of the excitation pulse spectrum .
this part of the excitation spectrum , according to frog trace measurements , corresponds to the most intense part of the excitation pulse . as a static sample of quantum dots
were used for this experiment , the time dependent spectra were also observed in each case , using an integration time of 500 ms over 10 minutes , to compare bleaching rates of the linear and three - photon excitation processes .
we observed that although a higher excitation intensity was used for the three - photon process , the emission signal of the quantum dots was bleached to 50% of the starting intensity after 566 s , compared to 103 s under linear excitation .
however the bleaching of the sample under ir excitation was not fit well by the expected double - exponential profile , suggesting that it may not follow a conventional bleaching mechanism .
further study is needed to properly assess the meaning of these preliminary observations .
fluorescence rate dependence of one- and three - photon excitation .
( a ) intensity dependent fluorescence measured under uv excitation shows a saturation intensity of @xmath14 from the saturation fit , given by the dashed gray line . excluding points above 50% of the saturation intensity , and fitting a power law @xmath15 , gives @xmath16 ( dashed blue line ) .
( b ) a power law fit of the intensity dependent fluorescence measured under three - photon excitation is shown , with an index @xmath17 ( dashed red line).,width=283 ] to verify the photon orders of the two excitation processes and search for saturation of the excitation , we measured the dependence of the fluorescence on excitation intensity , as shown in figure [ fig3-power ] .
the fluorescence spectra for both the one- and three - photon measurements , taken at varying excitation intensities , were fit with gaussian profiles .
the amplitudes of the fit profiles were compared using power law fits , as shown in figure [ fig3-power ] , in order to determine a @xmath18 power law dependence between excitation intensity @xmath19 and fluorescence counts @xmath20 , for each of the processes .
first- and third - order dependencies are expected for the one- and three - photon processes , respectively .
saturation of the quantum dots is observed under uv excitation , with saturation intensity @xmath21 , found using a fit of the form : @xmath22 by excluding data points greater than 50% of the saturation intensity found for the uv excitation data , the index of the uv intensity dependence was found to be @xmath23 .
the index for the ir induced fluorescence was @xmath24 , confirming three - photon behaviour .
third harmonic intensity was also found to exhibit cubic behaviour . as such
, the fluorescence under ir excitation could either occur via a direct three - photon mechanism or via absorption of a third harmonic photon .
this experiment does not distinguish between the two mechanisms .
our results show three - photon excitation and fluorescence in quantum dots , with the aid of ultrafast fiber lasers . with this
comes the potential to improve the effectiveness of quantum dot based biological label schemes , using low - cost hardware .
telecom - band fiber lasers provide an economical source of laser hardware that has the potential to supplant more conventional laser hardware such as ti : sapphire in many applications - of which biomicroscopy applications are no exception .
the experimental data for three - photon excitation shows that there are no significant changes or shifts in the fluorescence spectrum compared to the linear excitation case , and the third order dependence on excitation power shows that the quantum dot sample is indeed excited through a three - photon process .
the process in the study by horton et al .
@xcite could thus be replicated using quantum dots as a more favourable label agent @xcite . in future work
, we aim to address the implementation of quantum dots in biological imaging within tissue , using multiphoton processes to maximize viewing depth .
further work could improve the design of the fiber laser system used for this process and use additional types of quantum dots - including more biocompatible materials such as inp and different fluorescence wavelengths .
optimising quantum dots for longer wavelength fluorescence will significantly reduce scattering of the fluorescence signal , once again increasing viewing depth . by opting for a lower repetition rate , thus increasing the pulse energy of the fiber laser
, it will also be possible to increase the multiphoton fluorescence signal of the labels while reducing chances of cell damage due to high average power levels .
we provide the first demonstration of three - photon excitation in quantum dots using a fiber laser operating in the telecommunications band .
the experiment was conducted using a mode locked er - doped fiber laser producing 116 fs pulses with a center wavelength of 1575 nm .
we compare results from one- and three - photon excitation experiments to confirm that the three - photon spectrum is similar , and to demonstrate the third order intensity dependence of the process as an additional means of verification .
we found that the fluorescence intensity scaled with input excitation power as expected for a three - photon process .
three - photon excitation in quantum dots using ultrafast fiber lasers provides an important step toward more effective biological imaging by minimizing scattering of the excitation light and increasing the possible penetration depth .
24 ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty http://pubs.acs.org/doi/abs/10.1021/ja00072a025 [ * * , ( ) ] link:\doibase 10.1021/ja047659f [ * * , ( ) ] http://www.ncbi.nlm.nih.gov/pubmed/10062330 [ * * , ( ) ] http://www.ncbi.nlm.nih.gov/pubmed/9983472 [ * * , ( ) ] link:\doibase 10.1126/science.281.5385.2013 [ * * , ( ) ] link:\doibase 10.1038/nbt927 [ * * , ( ) ] link:\doibase 10.1126/science.1104274 [ * * , ( ) ] link:\doibase 10.1038/nmeth.1248 [ * * , ( ) ] link:\doibase 10.1038/nbt899 [ * * , ( ) ] link:\doibase 10.1117/1.2823156 [ * * , ( ) ] link:\doibase 10.1063/1.2714994 [ * * , ( ) ] link:\doibase 10.1063/1.1755420 [ * * , ( ) ] http://www.ncbi.nlm.nih.gov/pubmed/19550551 [ * * , ( ) ] link:\doibase 10.1021/nl0347334 [ * * , ( ) ] link:\doibase 10.1289/ehp.8284 [ * * , ( ) ] http://www.ncbi.nlm.nih.gov/pubmed/18040463 [ * * , ( ) ] link:\doibase 10.1038/nphoton.2012.336 [ * * , ( ) ] @noop * * , ( ) http://rmp.aps.org/abstract/rmp/v74/i4/p1283_1 [ * * , ( ) ] http://iopscience.iop.org/0953-8984/11/5/021 [ * * , ( ) ] http://books.google.com/books?hl=en&lr=&id=jdjvkrcsl8gc&pgis=1[__ ] ( , ) p. link:\doibase 10.1016/j.physb.2006.03.012 [ * * , ( ) ]
[ * * , ( ) ] http://www.opticsinfobase.org/abstract.cfm?id=8765 [ * * , ( ) ] | we demonstrate three - photon excitation in quantum dots with a mode - locked fiber laser operating in the telecommunications band .
we compare spectra and intensity dependence of fluorescence from one- and three - photon excitation of commercially available 640 nm quantum dots , using a 372 nm diode laser for one - photon excitation and 116 fs pulses from a mode - locked fiber laser with a center wavelength of 1575 nm for three - photon excitation . |
the correct form of the energy momentum tensor of an electromagnetic wave , and hence the momentum of an electromagnetic wave in a dielectric medium , has been debated for almost a century .
two different forms of the energy
momentum tensor were originally proposed by @xcite and @xcite , though more have been added in later years @xcite .
the question regularly attracts experimental interest , as on initial inspection it may appear that some of the different tensors give rise to distinct and potentially useful physical consequences .
we discuss why this is not the case , reviewing the key experiments of @xcite , @xcite , and @xcite , and theoretical work including that of @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite , the latter four authors demonstrating explicitly and directly the equivalence of a number of different energy momentum tensors .
more recent work on the subject may largely be divided into three camps .
first , there are those concerned with practical applications , focussing primarily on which expression is most useful in a given circumstance .
second , there are those extending the theoretical work into new domains
. frequently such papers are hampered by the fragmentary nature of existing literature on the topic and their conclusions may be weakened as a result .
third , there are those who are unconvinced that the controversy has been resolved , perhaps unsurprising as certain key publications remain relatively obscure .
it is clear that broad interest still exists in the subject , and from the widespread prevalence of work in the second and third categories , that the field stands to benefit greatly from a coherent and comprehensive review .
we therefore hope this paper will increase awareness that the controversy has been resolved , and that predictions regarding measurable behaviours will always be independent of the electromagnetic energy
momentum tensor chosen , provided the accompanying material tensor is also taken into account .
in particular , renewed interest in the controversy has been stimulated by the advent of optical trapping and manipulation of microparticles in fluid media @xcite .
we wish to lay to rest any concerns that predictions of observable effects in optical tweezers might depend on the division of momentum between field and medium , and to provide an introduction to the subject of the energy
momentum tensor in a dielectric medium .
we hope that by reading this paper and some of its references , those new to the field will gain a comprehensive understanding of this interesting subject .
the abraham minkowski controversy debates the correct expression for electromagnetic momentum within a material medium , and by implication the form of the four - dimensional energy momentum tensor .
before we debate the relative merits of the different expressions , we must first satisfy ourselves that it is possible to construct a well - defined energy momentum tensor .
after all , on a microscopic level the transfer of energy and momentum consists of the absorption and emission of photons as they interact with the atoms of the medium , a process which is both stochastic and quantum - mechanical in nature . fortunately , the abraham
minkowski controversy is conventionally formulated wholly within the realms of classical continuum electrodynamics @xcite , in which the fluctuations due to the inherent unpredictability of the microscopic realm become negligible .
specifically , the properties of the medium must change negligibly over unit cells which are nevertheless large in comparison with the mean volume per atom .
discontinuous inter - medium boundaries are permitted provided they are sharp on the scale of these unit cells . the transition from stochastic behaviour on the microscopic scale to the maxwell equations of the continuum , and hence by implication a domain in which both the refractive index and the energy momentum tensor are well - defined and piecewise smooth is comprehensively detailed by @xcite .
a number of authors do extend the problem into the microscopic domain @xcite .
such work must of course be evaluated on its own merits , and on its consistency with the results of experiment @xcite .
the first author to propose an expression for the energy momentum tensor of an electromagnetic wave in a dielectric medium was @xcite .
his expression corresponds to an electromagnetic wave momentum density of @xmath0 , where @xmath1 is the electric displacement field and @xmath2 is the magnetic flux density .
the total momentum of a propagating electromagnetic wave therefore increases upon entering a dielectric medium , from @xmath3 in free space , to @xmath4 , where @xmath5 is the refractive index of the medium , and we are neglecting dispersion .
an accessible derivation of the minkowski energy
momentum tensor may be found in @xcite . the electromagnetic energy
momentum tensor of minkowski was not diagonally symmetric , and this drew considerable criticism as it was held to be incompatible with the conservation of angular momentum .
@xcite therefore developed an alternative , symmetric , tensor , for which the electromagnetic wave momentum density was instead @xmath6 , with @xmath7 representing the electric field and @xmath8 the magnetic field .
the momentum of an electromagnetic wave entering a dielectric medium then falls to @xmath9 . under the abraham tensor
, a photon therefore carries less momentum within a medium than in the minkowski case .
however , both of these tensors are incomplete in isolation @xcite . momentum may also be carried by interactions occurring within the material medium , and between the medium and the electromagnetic field .
this is most obvious for the abraham tensor , as the material tensor must be considered in order to obtain the correct rate of transfer of momentum to a reflective object suspended within a dielectric medium , as in the experiments of @xcite and @xcite .
however , increasing the photon momentum from @xmath3 to @xmath4 in accordance with the minkowski tensor yields the correct result directly , and hence many authors neglect to observe that it , too , is associated with a corresponding material tensor .
any complete , thermodynamically closed system must conserve energy and momentum , and it is only by taking into account the corresponding material tensor that the minkowski formulation can be reconciled with conservation of total angular momentum @xcite .
the behaviour of a transparent object is less straightforward , with the object acquiring a fixed velocity away from the source while exposed to the electromagnetic field .
the existing literature lacks a single consistent mathematical approach capable of dealing with wholly or partially reflective and transparent objects immersed within a dielectric medium .
we address this lack in sec .
[ totememt ] .
when the force exerted by an electromagnetic wave in a dielectric medium is evaluated without recourse to the material counterpart of the minkowski tensor , then the two tensors will predict different results .
the force density predicted by the abraham tensor pair is smaller by @xmath10 which is known historically as the abraham force . here
, @xmath11 is the real instantaneous poynting vector @xmath12 , and @xmath13 and @xmath14 are the relative permittivity and permeability of the medium respectively . when the material counterpart to the minkowski electromagnetic energy
momentum tensor is also employed , as described in sec .
[ sec : atotal ] , then the minkowski tensor pair also gives rise to the smaller value . in the early part of the 20th century the importance of the material energy
momentum tensor was not fully appreciated , and vigorous debate ensued regarding the relative merits of the abraham and minkowski electromagnetic energy momentum tensors .
we shall discuss more fully how the requirement for the material counterpart to the minkowski tensor arises from conservation of total linear and angular momentum in sec .
[ proofsequiv ] .
meanwhile , it is worth noting that in some sense `` the momentum of a photon in a dielectric medium '' is an abstract concept , as it is impossible to construct an experiment to directly measure this momentum .
instead , one can only measure the total energy or momentum transferred to a detecting apparatus in a given set of circumstances .
this may include contributions both from the electromagnetic momentum and from the associated material energy
momentum tensor .
one may therefore ask whether it is possible or not to devise an experiment which distinguishes between the abraham and the minkowski tensors .
the first experiment to measure the radiation pressure due to light in a refracting medium was conducted by @xcite at the university of aberdeen .
jones sought to verify the prediction that when a mirror was immersed in a dielectric medium , the radiation pressure exerted on the mirror would be proportional to the refractive index of the medium .
later refinement of this experiment led to a paper by @xcite verifying this effect to an accuracy of @xmath15 . the experimental set - up
( shown in fig .
[ fig : jonesexpt ] ) consisted of a pair of mirrors strung on a vertical wire and tethered at each end by a gold alloy torsion fibre .
0.0005 cm ; _ e _ , brass tube 0.8 cm internal diam . ; _
f _ , magnet 0.2 cm long ; _ g _
, electromagnet ; _ h _ , 48 s.w.g .
copper wire ; _ i _ , liquid levels ; _ j _ , screwed brass ring and lead washers ; _ k _ , support for mirrors ; _
l _ , lead washer ; _ m _ , silvered glass mirror 0.2 @xmath16 0.5 @xmath16 0.02 cm ; _ n _ , brass swivel joint ; _ o _ , brass nut and washers ; _ p _
, copper / glass seal ; _
g_[_sic _ ] , glass tube ; _ r _ , cotton - wool plug ; _ s _ , triangular cast iron base ; _ v _ , rhodium - plated silver vane 0.2 @xmath16 0.5 @xmath16 0.01 cm ; _ w _ , circular glass windows 0.1 cm thick . reprinted figure with permission , from @xcite , _ the pressure of radiation in a refracting medium _ , fig . 1 , published by and (1954 ) the royal society.[fig : jonesexpt],width=321 ] the lower mirror ( vane _ v _ )
was subjected to an asymmetric illumination , giving rise to a torque about the axis of suspension and causing both lower ( _ v _ ) and upper ( _ m _ ) mirrors to rotate until this torque was neutralised by the torsion fibre . by measuring the angle of the upper mirror ,
the authors determined the magnitude of the torque , and hence the rate of momentum transfer to the lower mirror . by immersing the lower mirror in a number of liquids of varying refractive index and observing the resultant deflections , the momentum transferred
was found to scale with the refractive index , and the accuracy obtained was sufficient to show that this dependency was upon the phase refractive index ( @xmath17 ) , and not the group refractive index ( @xmath18 ) of the medium .
this finding agreed with theoretical expectations , as the behaviour examined was concerned with variations in the electric field at a point , rather than the rate of propagation through the medium of a modulation in the field . in the theoretical introduction to the 1954 paper , jones discussed whether the choice of expression for the momentum density of the electromagnetic wave was dependent either upon @xmath0 or @xmath12 , though without direct reference to the work of minkowski or abraham .
he observed that the former followed naturally from the expression for the maxwellian stresses on a body within a radiation field , whereas the latter could be suggested on relativistic grounds ( see * ? ? ?
* ; * ? ? ?
* ) but was also accompanied by a corresponding stress in the surrounding medium of @xmath19 ( in gaussian units ) , where @xmath20 and @xmath21 stand for the dielectric constant and magnetic permeability of the medium respectively , and the expression is equivalent to the abraham term ( [ eq : abrterm ] ) .
he commented that `` it does not seem possible to devise a practicable experiment to distinguish between the two cases '' .
@xcite subsequently argued that when the linear momentum flux densities due to the abraham electromagnetic tensor and the corresponding material stress tensor are combined , the result is the same as that given by the minkowski tensor .
he thus interpreted the experiment of jones and richards as providing more direct support for the minkowski formulation .
although in principle jones and richards are correct , and the abraham electromagnetic momentum is accompanied by an induced material momentum , they are wrong to identify this momentum flux with the abraham term
. we shall examine the abraham material momentum in greater detail in sec .
[ sec : atotal ] and return to this experiment in sec .
[ ex : mirror ] .
the next significant experiment conducted was that of @xcite , in response to a theoretical paper by @xcite . in this paper ,
burt and peierls argued that when a laser beam passed from air into a dielectric fluid of greater refractive index , transfer of momentum to the fluid would cause the surface to either bulge outwards , if the minkowski tensor was correct , or be depressed , if the abraham tensor was correct .
however , this conclusion was based on the mistaken assumption that the momentum of the electromagnetic wave would rapidly move ahead of any momentum associated with the material medium , allowing them to neglect the effects of the material tensor in the abraham case . in fact , the field - related material component of the momentum constitutes an excitation induced in the medium by the leading edge of the electromagnetic wave and eliminated by the trailing edge .
the excitation of the material medium therefore propagates at the same velocity as the electromagnetic wave .
ashkin and dziedzic performed this experiment , passing a laser beam vertically through a glass cell containing air and water ( fig .
[ fig : ashkinexpt ] ) .
glass fibre used to identify the surface of the water under the microscope .
a beam attenuator was placed either in position ( 1 ) or ( 2 ) depending on the power level desired within the glass cell .
reprinted figure with permission , from @xcite .
[ fig : ashkinexpt ] ] bulging or depression of the liquid surface would cause a lensing effect , altering the radial profile of the laser beam . by studying the beam profile
as it emerged from the cell , they were able to determine that the surface of the liquid was caused to bulge outwards .
in their paper they acknowledged work by @xcite , at that time still unpublished , which showed that this result does not invalidate the abraham tensor , as the predictions of the two tensors are reconciled when the material counterpart to the abraham tensor is properly taken into account . in 1968
james performed an experiment to directly measure the abraham force @xcite .
two ferrite toroids , axially aligned with one another , were connected to a piezo - electric transducer .
the toroids were subjected to time - varying electric and magnetic fields . both minkowski s and abraham s energy
momentum tensors predicted a net torque on the toroids , but with differing magnitudes due to the existence of the abraham term , eq .
( [ eq : abrterm ] ) . with the varying electric fields arranged in antiphase , the two toroids exerted torques in opposite directions upon the piezo - electric transducer , and by measuring the magnitude of this torque
, james was able to experimentally demonstrate the existence of the abraham force .
this result went largely unnoticed until the experiment of @xcite .
this team measured the torque exerted on a disc of barium titanate which was suspended on a torsion fibre in a constant axial magnetic field , and subjected to a time - varying radial electric field ( fig .
[ fig : walkerexpt ] ) . ]
the disc behaved as a torsion pendulum and its period of oscillation was measured by reflecting a laser beam off a mirror attached to the disc .
to maximise the response of the system , the time period of the electric field was synchronised with the oscillations of the disc . following an initial displacement , the oscillations of the disc settled exponentially towards a constant period , supporting the existence of the abraham force .
the experiment was also later modified to confirm the effects of a time - varying magnetic field @xcite .
as the conservation arguments leading to demonstration of the material counterpart of the minkowski tensor were not widely recognised at this time , this experiment was taken as providing support for the abraham tensor over that of minkowski @xcite .
however , @xcite and later , tangentially , @xcite have demonstrated that for a suitable combination of electromagnetic and material energy
momentum tensors the observed results are also compatible with the minkowski case , and hence these experiments can not discriminate between the two . in 1977
, jones and leslie repeated the experiment of 1954 , taking advantage of recent advances in technology .
a laser was employed as a light source instead of a tungsten lamp , allowing the refractive index of the liquid to be stated with greater precision as only one wavelength of light was involved ; a superior mirror was employed , reducing heating of the dielectric liquids and hence convection forces in the fluid ; and the torsion apparatus was restored to the neutral position by means of an external electromagnet .
the current in this electromagnet related to the radiation pressure in a highly linear manner , whereas in the previous apparatus , the authors estimate that nonlinearity in the measured deflection of the secondary mirror may have been as large as @xmath22 . through these modifications to the experiment , jones and leslie obtained final results with a standard deviation of only @xmath23 @xcite . in the accompanying theoretical paper @xcite , jones discussed whether this corresponded to a photon momentum of @xmath24 in accordance with the minkowski tensor , or @xmath25 in accordance with the abraham tensor ( where @xmath3 is the momentum of the photon in free space , and @xmath17 and @xmath26 are respectively the phase and group refractive indices ) . once again , he concluded that both were acceptable provided that in the abraham case the total momentum @xmath24 was divided into an electromagnetic component @xmath27 , and a mechanical component , which was expressed this time as @xmath28 $ ] .
numerous attempts have been made to find additional constraints by which one or another of the abraham and minkowski energy momentum tensors might be proven invalid , the earliest being the suggestion that the minkowski tensor failed to conserve angular momentum .
this led first @xcite and then @xcite to attempt to develop symmetric forms of the electromagnetic energy momentum tensor .
the einstein
laub tensor enjoyed only limited attention as it does not purport to be valid outside the rest frame of the dielectric medium .
its use indeed , even its validity in a science increasingly embracing the principle of relativity was therefore considered questionable .
it attracted considerably less interest than abraham s form , which was developed in a fully relativistic manner , working from microscopic considerations .
in contrast , @xcite claimed to demonstrate the derivation of the minkowski tensor from microscopic considerations .
however , according to @xcite he fails to justify his generalisation from the electrostatic to the dynamic case , and @xcite also deemed his argument `` not very cogent '' . as we noted in sec .
[ earlyyears ] , concerns about non - conservation of angular momentum in the minkowski electromagnetic energy momentum tensor arose from consideration of an incomplete system . when the appropriate material tensor is also taken into account , angular momentum is conserved ( sec .
[ proofsequiv ] ) .
similarly the einstein
laub tensor can be saved from the ignonimy of depending upon a specific frame of reference .
later , @xcite argued that the velocity of energy propagation , which is less than @xmath29 in a dielectric medium , should add in accordance with the relativistic velocity addition theorem .
an equivalent argument , developed by ( * ? ? ?
* ; * ? ? ?
* and 221 - 225 respectively ) , is that the energy propagation velocity should transform as a 4-vector .
this is satisfied by the electromagnetic energy flux under the minkowski tensor , but not under the abraham tensor . in the first edition of his work mller concludes `` this is a strong argument in favour of minkowski s theory '' , but in the second edition this is softened to `` which shows that minkowski s theory gives the most convenient description of the phenomenon in question '' .
this alteration perhaps reflects a change in philosophy , away from attempting to prove that one tensor is wrong and the other is right , and towards recognising that both have their place in a practical implementation of electrodynamics .
although it was ultimately dismissed by @xcite and @xcite , von laue s argument was nevertheless influential .
when pauli first wrote his encyclopedia article on relativity @xcite , he favoured the abraham tensor .
however , in the revised edition @xcite , he inserted a supplementary note reversing his position on account of von laue s argument .
in addition to rebutting the arguments of von laue and mller , skobeltsyn also proposed criteria of his own based on two thought experiments , which favoured the abraham tensor .
these are described in detail by @xcite , along with his reasons for rejecting skobeltsyn s arguments .
brevik then concludes that the second thought experiment , which deals with a capacitor free to rotate within a magnetic field , would nevertheless provide an experimental means to discriminate between the abraham and minkowski tensors .
the behaviour of the experiment depends upon the existence or otherwise of the abraham force , as does the experiment of @xcite .
the rebuttals of @xcite and @xcite discussed in sec . [ expts : walker ] are equally applicable to skobeltsyn s second experiment , and hence both skobeltsyn s arguments and brevik s interpretation of the second experiment are refuted .
@xcite also published an argument against the minkowski tensor based on non - conservation of angular momentum , similar to skobeltsyn s second argument .
this was rebutted by @xcite by invoking the material counterpart to the minkowski electromagnetic tensor .
we raise this here to discuss lai s counter - argument @xcite . in this
, he argues that his thought experiment was not intended to distinguish between what he calls the `` phenomenological '' theories derived from global conservation of momentum , but rather to determine which of abraham s and minkowski s tensors is `` physically correct '' . how this is to be interpreted is unclear .
if lai is advocating comparison of the historical forms of these tensors , then as neither incorporates effects such as electrostriction and magnetostriction , both are incomplete and demonstrably incorrect .
we can only ask which is least inaccurate in any particular situation , a conclusion of little general value . if lai permits the material tensor accompanying the abraham tensor to be amended to incorporate such effects , but does not permit the minkowski tensor to receive the same treatment , then he is crippling one of the contenders before the competition , and to conclude that one then performs better than the other is hardly surprising .
finally , if both tensors are allowed such a counterpart , then we recover the treatment of brevik , which lai dismisses as `` phenomenological '' and missing the point .
alternatively , lai may be implying that it is possible to discriminate between the minkowski and abraham tensors when both are accompanied by the appropriate material tensors .
this position , however , has been disproven by others @xcite , and hence lai s thought experiment appears to serve no useful purpose , in spite of his defence .
it is of historical note that the abraham tensor was also initially proposed without a material counterpart @xcite , and it was several decades before the role of this counterpart in determining experimental results was properly appreciated ( see sec .
[ expts : ashkin ] ) .
any treatment involving such a material counterpart is to some extent phenomenological , with the nature of the counterpart being determined or verified by observable fact @xcite .
it can be argued that only approaches which begin from universal conservation laws may be exempt from this criticism , if criticism it is , but it is these very results which lai appears to dismiss out of hand .
other arguments purportedly favouring one tensor or the other had also been proposed @xcite , but ultimately it was not to be these which decided the debate . in the 1960s and 1970s , two developments occurred which changed the face of the argument completely .
first , a number of further alternative energy
momentum tensors were proposed on various theoretical grounds @xcite .
the debate was no longer over whether to choose abraham s or minkowski s formulation , but rather over the principles upon which a valid electromagnetic energy momentum tensor should be based .
second , @xcite identified the abraham electromagnetic energy
momentum tensor with classical momentum and the minkowski tensor with crystal momentum , or pseudomomentum .
although the work was unpublished it had considerable effect , stimulating a recognition that both energy momentum tensors might have a valid role in appropriate circumstances .
although not every proposed means of discriminating between the abraham and minkowski tensors has been explicitly refuted in the literature , there is little to be gained from such a process .
inevitably , such attempts stem from inadequate exploration of the material energy
momentum tensor , and ultimately all such arguments are overturned by the direct proofs of equivalence of the differing electromagnetic energy
momentum tensors discussed in sec .
[ proofsequiv ] .
these proofs arise directly from conservation of energy , linear and angular momentum for the system as a whole , and hence it is only in those rare papers which discard these constraints that differing predictions can be made in a logically consistent manner .
towards the end of the 1960s , a number of researchers were demonstrating greater awareness of the role of the material energy momentum tensor in symmetry and in momentum conservation @xcite , and developing novel pairs of electromagnetic and material tensors with this in mind .
@xcite took this further with the formulation of their `` principle of virtual power '' , developed as a tool to test the equivalence of a number of different formulations of electromagnetic theory . by applying this principle
they were able to determine the incompleteness of both the minkowski and abraham tensors , and also of the tensors emerging from the @xcite , amperian @xcite and chu @xcite formulations of electromagnetism , and to identify the missing terms , and demonstrate that with properly constructed material counterparts , all these energy
momentum tensors are in agreement .
a cogent summary of their arguments and results may be found in a review paper by @xcite , which also covers the approach of gordon , discussed in sec .
[ sec : gordon ] .
like penfield and haus , @xcite also recognised the important role of the material energy
momentum tensor , and that conservation laws relating to energy and momentum should only be applied to closed thermodynamic systems . however , they disagree with penfield and haus on the correct form of the _ total _ energy momentum tensor to which electromagnetic and material energy momentum tensors must sum
. this disagreement does not bear directly upon the abraham
minkowski controversy , but does reflect different overall assumptions about the behaviour of matter in the presence of an electromagnetic wave .
de groot and suttorp have argued for the superiority of their expression as it is derived from considerations of the microscopic properties of the medium .
as yet , no experimental comparison of these two different total energy
momentum tensors has been performed . for the moment , however , the correct expression for the total energy
momentum tensor is unimportant . while eventually it might prove beneficial to our modelling of material behaviour to discriminate between the different propositions , for our purposes it suffices to denote the selected total energy
momentum tensor by @xmath30 .
this tensor is then constrained by conservation of linear and angular momentum such that @xmath31 and @xmath32 @xcite , where greek indices range over 0 - 3 .
a more familiar expression of the conservation of linear momentum may be @xmath33 where latin indices range over 1 - 3 , @xmath34 represents the derivative with respect to time , * g * is the momentum density , and * t * is the three - dimensional stress tensor .
the stress tensor @xmath35 is related to the energy
momentum tensor @xmath36 via @xmath37 for an electromagnetic wave in free space , * t * is called the maxwell stress tensor , but in dielectric media two different conventions exist @xcite so unless explicitly defined , this nomenclature is best avoided .
a further constraint @xmath38 corresponds to the conservation of energy , where @xmath39 is the energy density , and @xmath40 is the energy flux . for an electromagnetic wave in free space , @xmath41 where @xmath42 is the poynting vector . identifying @xmath43 as a 4-vector and writing @xmath44
we see that ( [ linmomcons ] ) is equivalent to constraint ( [ nodiv ] ) above . defining @xmath45 for symmetry , constraint ( [ symmetry ] )
requires that @xmath46 , giving @xmath47 unlike the system described by the total energy momentum tensor , the one described by the electromagnetic energy
momentum tensor need not be thermodynamically closed .
therefore , one must consider the combined system of both the electromagnetic and material energy
momentum tensors , such that @xmath48 and write @xmath49 as before , we may rewrite ( [ construct ] ) and ( [ eq : nodiv ] ) in terms of * g * and * t * : @xmath50 @xmath51 again we can define @xmath52 giving @xmath53 @xmath54 must also be subdivided , @xmath55 and from ( [ symmetry2 ] ) we see that @xmath56 but no constraint is placed upon the subdivision of @xmath57 into electromagnetic and material components . in energy
momentum tensors where the electromagnetic portion is symmetric , such as the abraham tensor , it follows that @xmath58 but this is not obligatory . for the minkowski tensor , @xmath59 , whereas @xmath60 .
similarly , @xmath61 . when the minkowski tensor is used , @xmath62 is frequently ignored as the material components of @xmath63 and @xmath35 are typically safely negligible ( see sec . [ discussion ] ) . however , because @xmath64 under the minkowski tensor , this gives the illusion of non - conservation of angular momentum via ( [ symmetry ] ) . the minkowski electromagnetic energy
momentum tensor has received much criticism for this reason , due to failure to recognise the existence of its material counterpart .
historically , neither the minkowski nor the abraham tensors were proposed in the context of a matched pair of electromagnetic and material energy
momentum tensors , and although the need for a material counterpart to the abraham tensor was proposed by @xcite , the minkowski tensor only acquired a material counterpart through the theoretical re - evaluations of the 1970s @xcite , explaining the results of @xcite and @xcite discussed in sec .
[ expts : walker ] .
we can now explain why experiments continue to be announced from time to time which purport to discriminate between the minkowski and abraham energy
momentum tensors , only to be disproven a few years later . for brevity we adopt the notation of eqs .
( [ nodiv ] ) and ( [ symmetry ] ) .
any experiment will be sensitive only to certain terms in the complete energy
momentum tensor .
an experiment involving only angular momentum will be insensitive to independently symmetric terms , and one involving only linear momentum will be insensitive to terms which are independently divergence - free . for any given experiment , we may therefore define @xmath65 as the terms in the material tensor which have no significant effect on the result , and @xmath66 as @xmath67 using @xmath66 in lieu of @xmath68 to make predictions for the behaviour of one experiment will therefore still yield the correct result , but for another experiment it may not . to illustrate this ,
consider the minkowski electromagnetic energy
momentum tensor . in general
, this tensor is used in isolation , letting @xmath69 .
the minkowski tensor is asymmetric , and hence from eqs .
( [ symmetry ] ) and ( [ symmetry2 ] ) we can conclude that @xmath65 also contains asymmetric terms . the minkowski electromagnetic energy
momentum tensor largely functions well in linear experiments such as that of jones and richards ( sec .
[ expts : jones ] above ) , but in isolation fails to correctly describe the rotary experiment of walker _
[ expts : walker ] ) .
@xcite and @xcite demonstrated that this is due to neglect of the corresponding material tensor .
a given electromagnetic and material tensor pair may be found to perform well across a broad range of experiments before failing .
this failure can arise as a result of an interaction which was not probed in the previous experiments , and is therefore described by a term which does not influence those previous experiments .
a frequent reaction is to announce that the experiment invalidates one of the electromagnetic energy
momentum tensors being considered .
however , inevitably a suitable term will be identified and added on to the corresponding material energy momentum tensor , or a new expression developed which is equivalent to the addition of such a term , and the new `` correct '' material energy
momentum tensor will be announced , rebutting the invalidation @xcite .
as the tensor pair is named by the structure of the electromagnetic member , and the necessary term may always be added to the material member , the tensor is therefore always able to be `` saved '' . as an interesting alternative
, the necessary term may be added to the electromagnetic tensor , and accompanied by a change of name .
neither the abraham nor minkowski tensors as they are commonly used incorporate the electrostrictive effect .
when the relevant term is added to the electromagnetic component of the abraham tensor , this is known as the helmholtz tensor @xcite .
regrettably , the full ramifications of the works of penfield , haus , de groot and suttorp were not immediately appreciated by the physics community , perhaps because calculating the full expressions for the material energy
momentum tensors was somewhat laborious , and perhaps because the experiment of @xcite had yet to be performed , that of @xcite was not widely publicised , and thus no convincing situation had yet been recognised for which the conventional forms of the minkowski and abraham tensors failed or disagreed .
in addition , de groot and suttorp derived a number of new tensor pairs which they felt were particularly relevant to specific situations , such as a neutral plasma , and which are documented in their book @xcite .
a more practically oriented demonstration of the equivalence of the abraham and minkowski tensors was achieved by @xcite . in response to the experiment of ashkin and dziedzic ( see sec . [ expts : ashkin ] above ) , he demonstrated that the field - related components of the material energy momentum tensor propagated along with the electromagnetic component , rather than at the speed of sound in the medium as was sometimes assumed ( for example * ? ? ? * ) , and that the optical pulse length in ashkin and dziedzic s experiment was sufficiently long for pressure within the fluid to reach equilibrium , and so purely material effects such as pressure also needed to be considered .
he then went on to show that both abraham s and minkowski s tensors predicted identical behaviour when an arbitrary radiation pulse traversed an air / fluid interface .
we note that all real material bodies are to some extent capable of deformation , and hence allow the transmission of tension and pressure in a manner analogous to the effects considered in gordon s proof . the proof
therefore readily generalises to arbitrary isotropic dielectric media for deformations small enough that dissipative effects such as hysteresis may be ignored , and has the potential for extrapolation to larger deformations and anisotropic cases .
inhomogeneous dielectric materials might then be dealt with by breaking them down into infinitesimally small homogeneous chunks . however , as pointed out by @xcite , gordon s result is only explicitly demonstrated for materials of dielectric constant sufficiently close to unity .
gordon s work was subsequently extended by @xcite with a number of interesting consequences , such as the generation of phonons by a finite width beam in an elastic solid .
recognition of the deformable nature of the dielectric medium is fundamental to gordon s approach .
we discuss the hazards associated with inappropriately treating the dielectric as a rigid body in sec .
[ anomeffects ] .
although the work of @xcite had provided the theoretical framework within which to demonstrate the formal equivalence of different choices of electromagnetic energy momentum tensor , they stopped short of explicitly demonstrating this for commonly used tensors such as those of abraham and minkowski .
this equivalence had been demonstrated by penfield and haus in 1967 , but it seems to have been little appreciated .
subsequent , far shorter demonstrations of equivalence were produced by @xcite and by @xcite , who appears unaware of the work of either penfield and haus or de groot and suttorp , as he cites later papers by @xcite as conjecturing a `` sliding '' relationship between electromagnetic and material energy momentum tensors .
the paper by mikura is particularly readable , but both are recommended to anyone seeking greater mathematical detail than is provided in this review .
@xcite also demonstrates that the behaviours predicted by the classical abraham and minkowski tensors are equivalent up to the action of the abraham force , but does not take the final step of postulating a material counterpart to the minkowski tensor .
he therefore concludes that while the minkowski tensor is not strictly correct , it is nevertheless useable under many circumstances .
a response to @xcite by @xcite supports krany s result but criticises him for not including enough references , and for confining himself to considering only the abraham and minkowski tensors .
maugin extends krany s work to consider tensors due to @xcite and @xcite in non - dissipative materials , with remarks about extension to more arbitrary cases , and the use of this approach to demonstrate the equivalence of any other energy
momentum tensor with those already considered .
@xcite then extends the work of krany and maugin to include media which may be magnetised and/or electrically polarised .
the interested reader may also have inferred the general equivalence of all choices of electromagnetic energy
momentum tensor from eqs .
( [ nodiv ] ) , ( [ symmetry ] ) and ( [ construct ] ) , as valid expressions for the total energy momentum tensor @xmath36 may only vary by a term which is symmetric and divergence free , and hence carries no momentum at all .
as mentioned in sec . [ arguments ] , an unpublished memorandum by @xcite is credited with first drawing the parallel between the minkowski electromagnetic momentum density and pseudomomentum . in doing so ,
he introduced a viewpoint in which both the abraham and minkowski expressions had a valid role . this view is further developed by @xcite , @xcite , @xcite , and @xcite , and our review would be incomplete without discussing this important recognition .
@xcite provides a thought experiment which we may use to illustrate this .
consider a perfectly transparent glass rod , initially at rest , with a perfect anti - reflection coating on each end .
a wave packet may pass either alongside the rod , or through it .
no external force acts on the combined system of rod and wave packet , so the centre of mass of the system must have constant velocity
. however , if the wave packet enters the rod , it slows down by a factor of @xmath70 .
this must therefore be accompanied by a movement of the rod in the direction of propagation of the wave packet so that the velocity of the centre of mass is preserved .
as the rod is perfectly transparent , no momentum is absorbed from the wave packet , and when the wave packet departs the rod it must leave the rod at rest . by comparing the trajectories of the centre of mass with the wave packet passing either inside or outside the glass rod , we can show that the rod experiences a finite displacement during the traversal of the wave packet , which we may interpret as resulting from the action of the lorentz force on the atomic dipoles which make up the rod .
while the rod is moving , it clearly has momentum .
the abraham approach identifies this as mechanical momentum .
however , the minkowski approach recognises that this momentum propagates with and arises from the presence of the electromagnetic wave , and hence classifies it as electromagnetic momentum .
we may well have qualms about describing the motion of a material body as electromagnetic momentum .
therefore , we may call the abraham momentum the true momentum of an electromagnetic wave , and the minkowski momentum the pseudomomentum , this being the physically useful quantity of all momentum associated with the propagation of the wave .
( nelson adopts a different terminology , in which pseudomomentum excludes electromagnetic momentum , and the total momentum associated with the propagation of a wave is termed `` wave momentum '' .
his chosen expression for electromagnetic momentum also differs , though this has no effect on the total momentum .
nelson identifies the minkowski momentum with wave momentum , less a term due to dispersion . )
this argument may appear compelling , but before the reader confidently declares the abraham momentum to be the only true electromagnetic momentum , they should recall that every expression appears compelling within the context of the arguments on which it is founded . for example
, @xcite presented an alternative chain of argument leading to an electromagnetic momentum density of @xmath71 .
this expression seems unlikely in the context of the above discussion of momentum and pseudomomentum , but in the approach taken by mansuripur it is the abraham and minkowski expressions which appear inappropriate , and this hybrid electromagnetic momentum density which seems the inevitable conclusion .
of course , it is momentum transfer which determines experimental results , and this takes place identically regardless of whether we describe it as electromagnetic or material in nature .
we therefore reprise our position stated in the abstract , that the reader may describe as electromagnetic whatever portion of the total momentum flux appears most appropriate , in the context in which they are working .
provided all momentum is ultimately accounted for , this will have no effect on their predicted experimental results .
@xcite has given the conditions under which the pseudomomentum approach will reliably describe momentum transfer to an object in a fluid dielectric medium .
the factor @xmath72 must be negligible in the vicinity of a surface surrounding the object of interest , and sufficient time must be allowed for pressure in the fluid to reach equilibrium .
in contrast the total energy
momentum tensor approach described in sec .
[ totememt ] may be employed even when the constituent variables are largely arbitrary functions of space and time , though the cost is a considerable increase in mathematical complexity .
performing calculations using the pseudomomentum density @xmath73 ( in gordon s notation and gaussian units , @xmath74 ) is directly equivalent to choosing the minkowski electromagnetic energy
momentum tensor without a material counterpart .
it therefore omits a small number of effects associated with the material counterpart tensor , including the abraham force , but these are seldom significant .
situations suited to the use of the isolated minkowski electromagnetic tensor , and hence also the pseudomomentum approach , are discussed in sec .
[ ex : mirror ] and [ discussion ] .
the obvious question now becomes , what is the total energy momentum tensor ?
can we write down a single expression that will serve our needs in all circumstances ?
unfortunately , the answer is no .
this is not a problem in electromagnetism .
the limitation is instead our understanding of material science .
we may certainly write down all contributions to the energy momentum tensor which are relevant in a particular perfectly defined ideal medium , but to do so for a generalised real medium lies beyond the scope of our current understanding .
fortunately , in many circumstances a complete understanding of the material properties of the medium is not required .
the medium may behave sufficiently like a suitable ideal , or we may be interested solely in the behaviour of the centre of mass , and we may then neglect the effects of internal tensions which merely redistribute existing force and torque densities . provided our description of the interaction of the material object with the electromagnetic wave is sufficiently accurate , details of material - material interactions may then be overlooked .
there is another important caveat .
when material stresses induced by an electromagnetic wave cause deformation of the boundary of the dielectric medium , focussing effects may then alter the subsequent path of the beam , as in the experiment of @xcite .
the results calculated for a given material conformation will then only be instantaneously true .
this effect may of course neglected if the deformation of the material medium is sufficiently small .
@xcite derives a total energy
momentum tensor for a nonviscous , compressible , nondispersive , polarisable , magnetisable , isotropic fluid .
electrostrictive effects , magnetostrictive effects , and acoustic waves are included , though in most circumstances they may be omitted for an increase in simplicity , as shown by mikura in the latter part of his paper .
we have converted from mikura s choice of gaussian units and matrix notation to si units and tensor notation .
mikura s approach may also be extended to dispersive media , as discussed in sec .
[ qualdis ] .
we will therefore distinguish between the phase and group refractive indices , denoted @xmath17 and @xmath26 respectively . the total energy
momentum tensor is given by @xmath75 where @xmath76 we have slightly condensed mikura s original scheme .
@xmath77 contains terms due to mechanical fluid flows , electrostriction , and magnetostriction .
@xmath78 contains field terms identical to those in free space .
@xmath79 contains terms relating to the polarisation of the medium , and @xmath80 and @xmath81 contain terms relating to magnetisation .
@xmath81 is separated from @xmath80 to facilitate the subsequent derivation of the abraham and minkowski electromagnetic energy
momentum tensors . in these expressions
, @xmath82 is the matter density in the local rest frame , @xmath83 is the 4-velocity of an element of the local medium , @xmath84 is the specific internal energy of non - electromagnetic nature and is a function of @xmath82 and the specific elastic entropy @xmath85 , @xmath86 and @xmath87 are related to the electric and magnetic constants of the medium by @xmath88 and @xmath89 , @xmath90 for @xmath91 and zero otherwise , @xmath92 is the total pressure including electrostrictive and magnetostrictive effects , given by @xmath93 for @xmath94 and @xmath95 is the total hydrostatic pressure in the fluid , @xmath96 @xmath97 is the usual electromagnetic field tensor @xmath98 and the polarisation and magnetisation tensors @xmath99 and @xmath100 are related to the three - dimensional polarisation and magnetisation vectors @xmath101 and @xmath102 in a given frame by @xmath103 where @xmath104 is the 3-velocity of an element of the local medium .
we define @xmath105 , @xmath106 .
@xmath101 and @xmath102 are related in turn to @xmath107 , @xmath108 , @xmath109 , and @xmath110 via @xmath111 , @xmath112 , and the relativistic constitutive relations @xmath113 for _ f _ , _ m _ , and _
p _ , the notation @xmath114 abbreviates @xmath115 , and @xmath116 denotes the dual tensor @xmath117 where @xmath118 is the levi civit symbol .
the total momentum flux and stress tensor under this total energy
momentum tensor may be obtained via @xmath119 and @xmath120 .
@xmath121\gamma^2\frac{{\mathbf{v}}}{c } + \varepsilon_0\,{\mathbf{e}}\times{\mathbf{b } } + \frac{1}{c\alpha}\,{\mathbf{p}}\times\left(\frac{{\mathbf{v}}}{c}\times{\mathbf{p}}\right ) + \frac{1}{c\beta}\,{\mathbf{m}}\times\left(\frac{{\mathbf{v}}}{c}\times{\mathbf{m}}\right ) \nonumber\\ & & - \frac{1}{c}\,{\mathbf{b}}\times\left(\frac{{\mathbf{v}}}{c}\times{\mathbf{m}}\right ) + \frac{1}{c^2}\,{\mathbf{e}}\times{\mathbf{m}}\\ { \mathbf{t}}&=&-\rho_0(c^2+\epsilon_{\mathrm{i}})\gamma^2\frac{{\mathbf{v}}\wedge { \mathbf{v}}}{c^2 } - \phi\left(\gamma^2\frac{{\mathbf{v}}\wedge{\mathbf{v}}}{c^2}+{\mathbf{i}}\right ) + \varepsilon_0{\mathbf{e}}\wedge{\mathbf{e } } + \frac{1}{\mu_0}\,{\mathbf{b}}\wedge{\mathbf{b } } - \frac{1}{2}\left(\varepsilon_0|{\mathbf{e}}|^2+\frac{1}{\mu_0}\,|{\mathbf{b}}|^2\right){\mathbf{i}}\nonumber\\ & & + \frac{1}{\alpha } \left [ { \mathbf{p}}\wedge{\mathbf{p } } + \left ( \frac{{\mathbf{v}}}{c}\times{\mathbf{p } } \right)\wedge \left ( \frac{{\mathbf{v}}}{c}\times{\mathbf{p } } \right ) -\frac{1}{2}\left(\left|\frac{{\mathbf{v}}}{c}\times{\mathbf{p}}\right|^2 - |{\mathbf{p}}|^2\right){\mathbf{i}}\right]\nonumber\\ & & + \frac{1}{\beta}\left[{\mathbf{m}}\wedge{\mathbf{m } } + \left(\frac{{\mathbf{v}}}{c}\times{\mathbf{m}}\right)\wedge \left(\frac{{\mathbf{v}}}{c}\times{\mathbf{m}}\right ) - \frac{1}{2}\left(\left|\frac{{\mathbf{v}}}{c}\times{\mathbf{m}}\right|^2 - |{\mathbf{m}}|^2\right){\mathbf{i}}\right]\\ & & - { \mathbf{b}}\wedge{\mathbf{m } } - { \mathbf{m}}\wedge{\mathbf{b } } + \frac{{\mathbf{e}}}{c } \wedge \left(\frac{{\mathbf{v}}}{c}\times{\mathbf{m}}\right ) + \left(\frac{{\mathbf{v}}}{c}\times{\mathbf{m}}\right ) \wedge \frac{{\mathbf{e}}}{c } - \left[\frac{{\mathbf{e}}}{c}\cdot\left(\frac{{\mathbf{v}}}{c}\times{\mathbf{m}}\right ) - { \mathbf{b}}\cdot{\mathbf{m}}\right]{\mathbf{i}}.\nonumber\end{aligned}\ ] ] @xmath122 denotes the outer product , @xmath123 .
note that the trace over all field - related components of the total energy momentum tensor vanishes , and it is therefore compatible with massless photons @xcite .
following mikura , we also subdivide this tensor to yield the abraham and minkowski electromagnetic and material energy
momentum tensor pairs . for the abraham tensor
, we simply write @xmath124 to obtain the minkowski tensor we define @xmath125 and then we find @xmath126 in the nonrelativistic limits , these yield the usual expressions for the electromagnetic energy momentum tensors : @xmath127 in this representation , the terms in the minkowski material counterpart giving rise to the abraham force may clearly be seen in eq .
( [ eq : minkmat ] ) .
in contrast , the field - related component of the abraham material momentum tensor arises from an implicit dependence of @xmath104 on * e * , * b * , * d * , and * h*. also note that although the minkowski material momentum density is frequently very small or zero , this does not reflect the behaviour of the particles of the dielectric medium , whose velocity is related only to the term @xmath128 .
now consider a wave packet of total momentum @xmath3 and volume @xmath129 incident upon a dielectric slab of group refractive index @xmath26 .
the velocity of the wave packet slows to @xmath130 on entering the slab , and its volume reduces to @xmath131 .
the total momentum must remain constant at @xmath3 , and hence the total momentum density increases from @xmath132 in free space to @xmath133 within the material .
however , under the abraham tensor the electromagnetic momentum density within the material is only @xmath132 and hence it must be accompanied by a material momentum of @xmath134 . identifying this with the corresponding term in ( [ eq : abrmat ] ) , and substituting for @xmath132 the free space momentum density @xmath135
, we obtain @xmath136 this expression holds for both dispersive and nondispersive media .
it is trivial to show that the same expression for @xmath104 is obtained under the minkowski tensor pair .
this result differs from the expression obtained by @xcite and @xcite , who argued that an electromagnetic wave propagating in a dielectric medium at velocity @xmath130 with momentum density @xmath137 will give rise to a momentum flux @xmath138 .
based on an observed momentum flux of @xmath139 , where @xmath132 is the momentum density in free space , they concluded the total momentum density associated with the wave within the dielectric must be given by @xmath140 .
this yields a material counterpart to be added to the abraham momentum density of @xmath141 which simplifies to @xmath142 in nondispersive media and generates the abraham force ( [ eq : abrterm ] ) .
however , this result is at odds with that of @xcite , who demonstrated that the abraham electromagnetic momentum density is correct , requiring that eq .
( [ matjones ] ) or ( [ abrterm ] ) instead be deducted from the minkowski momentum density , as seen in our minkowski material tensor ( [ eq : minkmat ] ) .
the error here is in the treatment of momentum flux .
while the expression @xmath138 will hold for a single wave propagating in isolation through the dielectric medium , it breaks down for superpositions such as the incident and reflected beams present here , and @xmath143 must be used instead .
the incident beam is associated with a medium particle velocity of @xmath104 and the reflected beam with a particle velocity of @xmath144 .
while the accompanying fields interfere , the velocities simply add , and the medium adjacent to the mirror remains stationary .
the momentum flux is therefore dependent only on field - related terms . unlike @xmath138 , @xmath143 is insensitive to the choice of abraham or minkowski momentum density . taking proper heed of the effect on the magnetic field of time - varying currents and charges induced within the medium ,
we obtain via eq .
( [ eq : boundary ] ) a flux of @xmath139 , which is in accordance with experiment . as
this result is independent of the choice of minkowski or abraham electromagnetic energy
momentum tensor , we are free to assign ( [ abrterm ] ) to the minkowski tensor in agreement with @xcite
. we also note a further advantage to our expression for momentum density : consider a wave packet of total momentum @xmath3 and unit volume entering the medium . under the total energy momentum tensor approach given here
, the volume of the wave packet falls to @xmath145 , its total momentum density rises by a factor of @xmath26 , and momentum is conserved . under jones and richards
approach the total momentum density rises by a factor of @xmath146 , giving a momentum associated with the wave packet of @xmath24 .
conservation of momentum is usually restored by positing a transfer of momentum @xmath147 to the medium boundary ( see , for example , * ? ? ?
nevertheless , in steady state experiments such as @xcite we might expect this momentum to be conducted to the mirror via pressure effects within the medium , reducing momentum transfer to the mirror at equilibrium from @xmath24 to just @xmath3 .
this reduction is not observed , and this supports our approach over that of jones and richards .
while it is always possible to return to first principles , in the manner of @xcite , and individually evaluate all the forces acting in a situation , this is frequently a far from simple task .
however , where appropriate expressions for the total energy momentum tensor are known , force and torque densities for a specific subsystem may be calculated directly .
the means by which they are calculated will depend upon whether the subsystem is spatially or qualitatively distinguished from the system as a whole .
we use the covariant generalisation of angular momentum density @xmath148 given in @xcite , where @xmath149 is the 4-vector position co - ordinate relative to the point about which torque is determined .
@xmath150 is the 3x3 antisymmetric angular momentum tensor whose three independent elements map onto the angular momentum density pseudovector @xmath151 via @xmath152 where @xmath118 is the levi civit symbol .
the elements @xmath153 represent the puzzling yet orthogonally necessary temporal components of angular momentum , and @xmath154 describes angular momentum flux , analogous to @xmath155 in the total energy momentum tensor . from ( [ eq : definem ] ) ,
( [ nodiv ] ) , and ( [ symmetry ] ) , it follows that @xmath156 which corresponds to conservation of angular momentum .
the torque density tensor will be denoted @xmath157 , and maps onto the torque pseudovector via @xmath158 .
let the system be divided into two parts , @xmath159 and @xmath160 , which are qualitatively different but occupy the same region of physical space .
we wish to calculate the momentum transfer from @xmath159 to @xmath160 .
for example , @xmath160 might be a material component , such as a dielectric medium , and @xmath159 the electromagnetic field .
we also similarly divide the linear momentum density , angular momentum tensor , and stress tensor into parts @xmath159 and @xmath160 . into @xmath161 , @xmath162 and @xmath163
we place all terms relating to the physical momentum of the material component , as it is the force and torque on the material that we wish to evaluate .
the remainder of the terms are placed into @xmath164 , @xmath165 and @xmath166 .
the force and torque densities on component @xmath160 are then evaluated using @xmath167 these are momentum balance equations , with the rate of momentum and torque transfer into system @xmath160 equalling the rate of loss from system @xmath159 . integrating over the volume of the subsystem
then gives the total force or torque exerted on @xmath160 by @xmath159 as a result of coupling between the two subsystems , for example absorption of momentum from an electromagnetic field .
@xmath168 @xmath169 and @xmath170 represent the total force and torque respectively .
the introduction of a complex refractive index allows the modelling of electromagnetic wave propagation through media demonstrating loss or gain , and it is the resulting dependency of the magnitudes of the electric and magnetic fields on position and time that gives rise to a non - zero time average for the force or torque of ( [ eq : fdens ] ) and ( [ eq : tdens ] ) . of course , by the kramers kronig relations any absorptive material medium is also necessarily dispersive .
this may be modelled by describing a separate energy momentum tensor with appropriate dielectric and magnetic constants for each mode of the electromagnetic wave , and expressing the refractive index as a complex function of the wave number , @xmath171 . for near - monochromatic waves in the steady state , dispersion
may frequently be neglected and the use of a single electromagnetic energy momentum tensor with the material counterpart tensor may suffice . now suppose we are interested in momentum flux into or out of a particular region of space .
we denote this region @xmath160 and the outside region @xmath159 . by gauss s law ,
the rate of change of momentum within region @xmath160 is dependent upon the rate of flux through the boundary .
transforming eq .
( [ eq : volforce ] ) into surface integrals , we obtain @xmath172 the normal to the boundary is taken as pointing from @xmath160 to @xmath159 .
the first integral in each equation evaluates flux from @xmath159 to @xmath160 , and only terms inbound to region @xmath160 should be retained .
similarly , the second integral evaluates flux from @xmath160 to @xmath159 and only outbound terms should be retained .
it is not required that regions @xmath159 and @xmath160 be materially identical .
evaluation of these surface integrals may frequently be simplified by judicious application of gauss s law .
note that the form of the momentum entering area @xmath160 mechanical or electromagnetic is not specified . in more general circumstances , for example momentum transfer to a solid dielectric block immersed in a dielectric fluid and exposed to an electromagnetic field
, both spatial and qualitative considerations may apply .
the block may receive momentum through its boundaries as a result of the external medium ( e.g. pressure ) and the external electromagnetic field ( e.g. reflection ) , and within its bulk as a result of coupling between matter and the electromagnetic field within the block ( e.g. absorption ) .
we denote the region containing the block region @xmath160 , and the external region @xmath159 . as in sec .
[ qualdis ] , we divide the linear momentum density , angular momentum tensor , and stress tensor within the block into material components , denoted @xmath173 , @xmath174 , and @xmath175 , and electromagnetic components , denoted @xmath176 , @xmath177 , and @xmath178 , purely dependent upon whether they relate to the outcome we wish to determine the physical momentum of the dielectric block .
we then label the region external to the block region @xmath159 , and similarly divide its stress tensor into material and other components , except that here , a term is considered material if it will transfer momentum into the material component of section @xmath160 on encountering the boundary .
the equations for the force and torque acting on the dielectric block then become @xmath179 further subdivision of field - related tensors into the different electromagnetic modes may be required for the modelling of dispersive media , as indicated in sec .
[ qualdis ] .
the labelling of terms in region @xmath159 as `` material '' should not be taken literally .
for example , if the block in region @xmath160 is perfectly reflective , then all electromagnetic waves in region @xmath159 would be classified as `` material '' as they would impart their momentum to the material component of region @xmath160 .
the designation of `` material '' is functional and context - dependent , and should not be taken as proposing a clear and unique subdivision in the manner of the abraham or minkowski approaches . at this point
we will be well served by examining some examples within the literature .
consider the reflection of light off a perfect mirror suspended within a dielectric .
we wish to evaluate the momentum transferred to the mirror , as in @xcite or @xcite .
this takes place only at the surface of the mirror , and therefore we can use the surface integral ( [ eq : boundary ] ) . for @xmath166
we take the total stress tensor within the dielectric .
we may discard the term in @xmath163 as there will be no momentum flux from the mirror to the dielectric . as noted in sec .
[ sec : atotal ] , the medium adjacent to the mirror remains stationary because the velocities induced by the incident and reflected beams sum to zero .
momentum is therefore transferred by the electromagnetic fields alone .
historically the minkowski electromagnetic energy momentum tensor in isolation has also been used here with great success .
although the assumed momentum density differs from that yielded by the total energy
momentum tensor , the simultaneous identification of the momentum flux as @xmath180 instead of @xmath143 yields the correct overall result , up to a factor of @xmath181 .
this works because the minkowski material momentum density is typically numerically very close to zero , and hence the total momentum density may be reasonably approximated by the electromagnetic momentum density @xmath73 . that the total momentum flux is given by @xmath180 then follows from gauss s law .
the result is satisfactory when the error introduced by this approximation is small .
now consider a beam of light incident upon a block of perfectly nonreflective dielectric of higher refractive index than the surrounding medium .
@xcite considers the transmission of a wave packet through a glass rod with anti - reflection coatings on both ends .
if we are interested in the behaviour of the surface , then surface terms must be taken into account using ( [ eq : fullforce ] ) , but if we are only interested in the motion of the centre of mass of the block , then we can neglect the surface forces provided the time average of the integral ( [ eq : boundary ] ) goes to zero over a period much less than the sensitivity of our apparatus . for a laser beam traversing a dielectric block , we may eliminate the field - related terms of the integral if the electromagnetic fields where the beam enters and leaves the block are identical up to a phase . at its simplest
this requires zero reflection , zero absorption , and that the two surfaces be parallel .
we may eliminate the remainder of the terms by choosing the surrounding medium to be vacuum .
the behaviour of the centre of mass of the block may then be determined using ( [ eq : volforce ] ) . for @xmath182 ,
the block accelerates away from the beam source while it is traversed by the leading edge of the beam , then continues to travel away from the source at constant velocity while the beam is turned on .
when the beam is turned off , traversal of the trailing edge restores the block to rest .
the use of ( [ eq : volforce ] ) can also readily be extended to include absorption , giving rise to an additional ongoing acceleration . the most general case to consider is a partially reflective dielectric block suspended within an arbitrary dielectric medium . in the absence of simplifying assumptions we must employ the full expression ( [ eq : fullforce ] ) , with both a surface and a volume term .
we may compare this with the work of @xcite , who calculated the forces on a dielectric using the lorentz force .
integration of the lorentz force density over the volume of the dielectric gives a volume term , whereas reflection of a portion of the incoming radiation gives rise to a surface term .
although mansuripur examined pressure and tension arising within the dielectric block as a result of the incident radiation , contributions of these effects to the net surface force , as in @xcite , are not discussed .
these forces make no contribution to the motion of the centre of mass , but will affect the behaviour of the surface if the block is treated as flexible .
mansuripur obtained an expression for the electromagnetic momentum density in a dielectric medium from the rate of transfer of momentum into the medium both as electromagnetic fields and via the lorentz force , but excluded momentum transferred to the surface via reflection , which he considered non - electromagnetic .
however , this merely represents a further scheme by which to subdivide the total energy momentum tensor , and when electromagnetic radiation passes from one medium into another , both bulk and surface forces will be present .
momentum flux through the boundary under this scheme is therefore indistinguishable from the combined electromagnetic and material tensors of either abraham or minkowski .
how mansuripur s definition of the electromagnetic momentum density extends to moving media has yet to be established . in general
, momentum transfer to the surface of a dielectric block arises both from interactions with the external environment and from pressure effects within the block itself .
this may cause regions of relative compression or rarefaction close to the surface , depending on whether the surface force is of the same or opposite sign to the lorentz force which acts on the bulk of the block . for dielectric fluids , a flow may result . when eq .
( [ eq : fullforce ] ) is used , these effects are taken into account via the material components of the total energy
momentum tensor .
such effects are seen in the experiment of @xcite discussed in sec .
[ expts : ashkin ] , where the surface of the water bulges towards the source of the laser beam , although the lorentz force within the body of the liquid acts in the opposite direction .
it is frequently possible to make simplifying assumptions .
we have already considered a perfectly reflecting mirror , for which no volume term arises , and a perfectly transmissive dielectric in vacuum , for which the surface term has no effect on the motion of the centre of mass . also , it is often possible to use the abraham or minkowski energy
momentum tensors in the traditional manner .
the abraham tensor pair are well suited to problems in highly transmissive dielectrics , such as example [ nonreflec ] , as momentum transfer to the material component may be identified with the lorentz force @xcite .
the minkowski electromagnetic energy momentum density is frequently used to calculate momentum transfer across a boundary , as discussed in sec .
[ ex : mirror ] .
it is important to note that electrostriction , magnetostriction , dispersion , and acoustic waves are rarely treated in the literature , though with some exceptions @xcite . when these effects are significant , an appropriate material energy
momentum tensor must always be used .
rigorous treatment using a lorentz force approach @xcite is also a viable option under any circumstances , provided sufficient care is taken over field / charge interactions at the surface @xcite , and may frequently be the simpler approach . where the behaviour of the surface is important , for example in the behaviour of dielectric liquids @xcite , stress- or pressure - related effects
must also be considered .
over the years , a large number of tensor - dependent effects have been proposed , by which it has been suggested that it might be possible to distinguish between the different tensor formulations .
these tend to arise either from inadequately considering the role of the material energy
momentum tensor , or from inappropriately treating the dielectric medium as a rigid body .
failure to adequately consider the role of the material energy
momentum tensor may lead to claims of experimental discrimination between the different energy momentum tensors , as in @xcite , and hypothetical thrusters based on non - conservation of momentum @xcite .
perhaps more important is the error of treating the dielectric as a rigid body , an error which recurs frequently within the literature @xcite and also leads to inappropriate predictions of unusual physical behaviour . consider a pulsed electromagnetic wave of momentum @xmath3 entering a material body of refractive index @xmath5 .
if we use the minkowski electromagnetic energy momentum tensor , on entering the body the momentum of the wave is increased to @xmath4 and hence by conservation of linear momentum , the body acquires a momentum of @xmath183 , setting it in motion towards the source of the wave .
when the pulse exits the body , the momentum of the material is restored to zero and the body returns to a stationary state . in contrast , under the abraham tensor the momentum of the pulse decreases and the body must begin moving away from the source of the wave . @xcite discussed such behaviour in the context of an einstein box a rigid box in which photons are emitted from one end and absorbed by the other , typically used as a thought experiment to demonstrate mass energy equivalency . because the body is treated as rigid and hence unable to deform
, momentum is transferred to the body as a whole , propagating across it instantaneously . in reality
, the transfer of momentum must take place at a finite velocity which can not exceed @xmath29 and will therefore either be much less than , or of comparable magnitude to , the speed of light in the medium , @xmath184 . because of this , significant compressive and expansive effects must take place , which are neglected if the body is treated as rigid . this approach also mis - handles momentum associated with the material body under the minkowski tensor . as seen in ( [ eq : minkmat ] ) , the minkowski material momentum density includes a term @xmath185 , which causes the material momentum density as a whole to be negative .
however , this term is not associated with the movement of material particles described by @xmath104 , and is therefore not related to the motion of material particles within the block . while it is necessary for the material tensor momentum density to be negative in order to offset the increase in momentum density in the minkowski electromagnetic tensor , not all of this momentum density is associated with the particles of the block .
in fact , the behaviour of these * v*-dependent terms is identical to that under the abraham tensor and the block gains positive momentum , although the material momentum density as a whole becomes negative under the minkowski scheme . when these effects are treated correctly , we recover an approach equivalent to that of @xcite , as discussed in sec . [
sec : gordon ] , and the traversal of the light pulse now generates a distortion in the boundary of the medium . when the leading edge of the pulse enters the medium , the boundary bulges outwards , as demonstrated by the experiments of @xcite , and a similar outward bulge
is produced when the leading edge exits the medium .
the traversal of the trailing edge of the pulse gives rise to a restoring impulse in each case .
meanwhile , the unopposed action of the lorentz force within the bulk of the medium gives rise to the motion of the body as a whole .
the resulting behaviour is identical under both the abraham and minkowski tensors . as discussed in sec .
[ mompseudomom ] , @xcite demonstrated that a perfectly transmitting dielectric of higher refractive index than its surroundings will advance in the direction of propagation of the wave packet , using arguments relating to the movement of the centre of mass of the system .
however he then erroneously applies the rigid body argument given above , and concludes against the minkowski electromagnetic energy momentum tensor . in reality , as we have seen in sec .
[ nonreflec ] , both choices when used correctly will yield the same lorentz forces acting on the body of the rod , accompanied by any surface effects due to pressures within the rod as per @xcite and @xcite , because regardless of whether we choose the minkowski or abraham representation , the total energy
momentum tensor remains unchanged .
we note that in a real medium , coupling between adjacent surface elements ( for example , surface tension in a liquid or lattice forces in a solid ) may cause disturbances generated by surface forces to propagate radially outwards .
thus in an absorptive medium , traversal of a radiation pulse may give rise to acoustic ( matter ) waves within the substance of the medium , but once again the predicted behaviour will be independent of the choice of energy
momentum tensors used .
@xcite showed that such photon / phonon coupling may also take place where the edges of a finite beam lie within a dielectric .
as mentioned in the introduction , current research on the electromagnetic energy
momentum tensor relates largely to practical applications , or to the extension of existing knowledge into new domains , though there are still many who appear unaware that the abraham
minkowski controversy has been resolved .
we present and discuss some of the more recent papers , to illustrate the present state of research in the field .
obviously , any theoretical result requires verification . the ability of combined electromagnetic and material energy
momentum tensors constrained by conservation of linear and angular momentum to explain historical experiments such as those discussed in this paper is well established .
however , continued testing of such theories over an increasingly broad range of domains is an essential part of the process of physics research . in this regard
, we refer to the proposal of @xcite to demonstrate the abraham force at optical frequencies , and the work of @xcite seeking violation of conservation of momentum , though his reported positive results fare poorly under independent scrutiny @xcite . extending existing classical theory ,
@xcite directly derived the force of electromagnetic radiation on a material medium from the lorentz force law , an approach which makes no pre - emptive assumptions regarding the value of the electromagnetic momentum .
he reproduced the beam edge effects first discussed by @xcite , but with the discovery that this force may be expansive , rather than constrictive , for @xmath3-polarised light , an effect which awaits experimental verification .
he also deduced that anti - reflective coatings experience a force which attempts to peel them off their substrate .
like many others , in the course of this paper he derived a unique `` correct '' expression for the electromagnetic momentum density .
there are almost as many such expressions as there are authors , and if we restrict physics to dealing only with hypotheses which may be tested by experiment , then the question of the correct expression is instead one of ontology .
however , as physical predictions are independent of the expression chosen ( see sec .
[ proofsequiv ] ) , this in no way detracts from the significance of his other findings .
other recent work includes the experiment of @xcite to determine whether , when a photon is absorbed by an atom within an atomic cloud , it should be considered as departing the medium , and therefore only transfer the momentum of @xmath186 to the atom , in accordance with @xcite , or whether it transfers the minkowski value @xmath187 , leaving behind a disturbance in the medium , as proposed by @xcite .
the latter case was determined to be correct , with significance for the construction and operation of atom interferometers .
using quantum theory , @xcite developed operators for the momentum of a photon in a dielectric medium . employing a quantisation scheme proposed by @xcite
, they arrived at a situation they recognise to be directly analogous to the abraham
minkowski controversy .
their survey of the literature included the recognition , which they attributed to @xcite , that decomposition of the total energy
momentum tensor into electromagnetic and material components is arbitrary .
however , due to the inconsistent nature of the available literature , this was stated as simply one possible position of many , rather than an obligate consequence of momentum conservation .
@xcite proceeded to develop three momentum operators abraham and minkowski momentum operators , corresponding to classical momenta calculated utilising the isolated electromagnetic energy momentum tensors of abraham and minkowski , and a canonical momentum operator , which is the generator of translations , and relates to the total energy momentum tensor .
they reanalysed the results of @xcite , and demonstrated that the experiment favours the canonical momentum operator over both the abraham and minkowski operators .
this is to be expected , as no operators were employed which related to either the corresponding material momentum flux or the maxwell stress tensor , and hence only the canonical momentum operator accounts for the total flux of momentum within the system .
their analysis is novel , in that it demonstrates that the results of jones and leslie are of sufficient accuracy to distinguish between the total energy
momentum tensor , and the electromagnetic energy momentum tensor of minkowski in isolation . by doing so
, they therefore provide additional experimental verification that the minkowski tensor , like the abraham tensor , is incomplete without a material counterpart . for any real medium
the minkowski material counterpart tensor will yield not only the abraham term , but also terms giving rise to acoustic transients @xcite and a dispersion - related coupling between the electromagnetic wave and the material medium @xcite .
this latter effect is responsible for the observed discrepancy between the canonical momentum and the minkowski electromagnetic momentum .
garrison and chiao concluded that the canonical ( total ) momentum is most applicable for interpreting experiments , and question whether the use of abraham and minkowski operators can ever be appropriate .
however , they weaken their conclusion by citing the experiment of ( * ? ? ?
* and sec .
[ expts : walker ] above ) as uniquely selecting the abraham tensor , despite having earlier cited @xcite , who indirectly disprove this assertion .
they also present the arguments of @xcite as similarly favouring the abraham tensor , though these have been refuted by @xcite , as discussed in sec .
[ arguments ] . garrison and chiao
do not examine the experiment of walker _
et al . _ or the suggestion of @xcite in their quantum - mechanical formalism , which is a pity , since they
might then have concluded that once again , it is conservation of the canonical momentum which governs the experiments behaviour .
@xcite also investigate momentum transfer from a quantum perspective , looking at the absorption of photons by charge carriers in a semi - conductor .
they found that momentum transfer can be divided into three separate components : momentum transfer to the charge carriers , given by the minkowski value @xmath188 and in agreement with previous experiment @xcite , to the bulk of the semi - conductor , given by the abraham value @xmath189 , and to the surface of the dielectric , yielding overall conservation of momentum .
a classical treatment by @xcite again results in transfer of the minkowski value of momentum to the charge carriers , with the net action on the host semiconductor maintaining global conservation of momentum .
@xcite has looked at the transfer of momentum between matter and the electromagnetic field on the quantum scale , and predicted the intriguing phenomenon of fluid flow powered by quantum vacuum fluctuations which is dependent upon the high - frequency cutoff of quantum field theory .
several have found the electromagnetic energy
momentum tensor important in extension of the casimir effect from vacuum - separated plates to those separated by a dielectric matierial .
@xcite related the importance of recognising that the minkowski tensor on its own is incomplete , and @xcite suggested that consideration of the total momentum flux density is most appropriate .
stallinga also demonstrated how a classical technique utilising an artificial `` auxiliary field '' @xmath190 may be useful in modelling dissipative systems . of interest for different reasons
is the somewhat older work of @xcite in representing the behaviour of massless photons in a material medium as that of equivalent massive particles in vacuum . due to a transformation effectively eliminating the material medium , all momentum
is contained within the electromagnetic energy
momentum tensor , and in this special case , the criteria of von laue and mller ( see sec .
[ arguments ] ) are fulfilled . in terms of the abraham
minkowski controversy , cole is in effect choosing an energy momentum tensor in which all active terms are placed within the electromagnetic tensor and the material energy
momentum tensor is always null .
more recently , @xcite have drawn attention to a related treatment of the constitutive equation of electromagnetism by @xcite , valid for a medium homogeneous and isotropic in its local rest frame .
a similar approach is also of practical relevance in modelling optical trapping . since all forces acting due to transfer of momentum from fields and media must be considered , and the trapped particle is typically suspended in a dielectric medium , a separation of the field and material components of the momentum flux that allows the surrounding medium to be ignored can simplify calculation of the optical force .
this separation requires that the surrounding medium carries none of the momentum , a condition which is adequately met by the minkowski tensor provided both dispersion and the abraham force are negligible , the fluid medium has had time to reach equilibrium , and behaviours such as electrostriction and magnetostriction , which are not supported by the electromagnetic component of this tensor , may be neglected .
the trapped particle can then be modelled as a dielectric particle of refractive index @xmath191 in free space , with the beam delivering a momentum flux density of @xmath192 .
this method is best suited to monochromatic light or nondispersive media , where @xmath5 may be treated as a constant . for a rigid particle ,
the optical forces can be calculated from the incoming and outgoing momentum fluxes @xcite . for a deformable particle , for example in an optical stretcher @xcite
, one needs to include the mechanical elastic forces that resist the field or medium forces acting to deform the particle , as well as the field or medium forces themselves . while it may be simplest to assume the minkowski momentum , the alternative tensors are nevertheless physically equivalent .
statements which are based on the interpretation of a single tensor , for example suggesting that the deforming force arises because the momentum of a photon in a medium is @xmath193 , compared to @xmath194 in vacuum , are therefore unwise .
the work of @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite effectively demonstrated that division of the total energy momentum tensor into electromagnetic and material parts is effectively arbitrary .
nevertheless , many continue to attempt to identify a superior expression for the electromagnetic energy momentum tensor .
often they are aware of the work of penfield and haus , and de groot and suttorp , and one must therefore suppose that these works , perhaps on account of complexity or lack of sufficient emphasis , are proving ineffective in conveying that such a distinction is at the very least untestable . for gordon , one may point to the limited scope to which his treatment may be applied , and for mikura , krany and maugin the problem appears to be insufficient publicity .
whatever the cause , papers continue to appear which attempt to demonstrate a uniquely privileged form for the electromagnetic energy
momentum tensor ( see , for example , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
padgett proposed an interesting experiment in the optical domain , in which a laser beam carrying orbital angular momentum traverses , and transfers angular momentum to , a glass disc .
this experiment is in many ways analogous to that of ( * ? ? ?
* and sec .
[ expts : ashkin ] above ) . in ashkin and dziedzic
s experiment , pressure effects within the liquid were responsible for restoring equivalence between the abraham and minkowski interpretations . in padgett
s experiment , a similar role will be played by shear forces within the disc .
the works cited above are far from alone in their continued fascination with the abraham
minkowski controversy .
the topic frequently continues to be a point of discussion at conferences , and the occasional physics colloquium @xcite , though awareness is gradually increasing that the distinction is of no functional significance , and hence the choice of energy
momentum tensor pair is essentially a question of personal sthetics .
the original abraham minkowski controversy , over the preferred form of the electromagnetic energy
momentum tensor in a dielectric medium , has been resolved by the recognition that division of the total energy
momentum tensor into electromagnetic and material components is arbitrary @xcite .
hence the minkowski electromagnetic energy momentum tensor , like the abraham tensor , has a ( frequently unacknowledged ) material counterpart @xcite , and the sum of these components yields the same total energy
momentum tensor as in the abraham approach . on these grounds , all choices for the electromagnetic energy
momentum tensor are equally valid and will produce the same predicted physical results , as has been demonstrated for a wide range of specific examples by @xcite , @xcite , and @xcite .
nevertheless , awareness of the resolution of the original controversy remains patchy , largely due to the fragmentary nature of the literature on the subject . in this paper
, we have reviewed the controversy from its initial formulation to the present day , incorporating discussion of key experiments believed at one time or another to have bearing on the relative truths of the abraham and minkowski models .
we have discussed the realisation that any electromagnetic energy
momentum tensor must always be accompanied by a counterpart material energy
momentum tensor , and that the division of the total energy momentum tensor into these two components is entirely arbitrary .
a consensus has yet to be reached on the ultimate form of the total energy
momentum tensor , but the limitation here arises not from classical electromagnetism but from materials science , and we believe there is scope for further work in explicitly developing the theory of dispersive media , a subject touched upon in sec . [ qualdis ] .
we believe a wider appreciation of the relationship between electromagnetic and material energy
momentum tensors could greatly benefit those working in this field , and with this colloquium we hope to assist in bringing that wider appreciation into existence .
the authors would like to thank the royal society , professor a. ashkin , the american physical society , and the national research council of canada for permission to reproduce the figures accompanying this paper .
the work of r.n.c.p was partly supported by an endeavour international postgraduate research scholarship funded by the australian government department of education , science and technology . | almost a hundred years ago , two different expressions were proposed for the energy momentum tensor of an electromagnetic wave in a dielectric .
minkowski s tensor predicted an increase in the linear momentum of the wave on entering a dielectric medium , whereas abraham s tensor predicted its decrease .
theoretical arguments were advanced in favour of both sides , and experiments proved incapable of distinguishing between the two . yet more forms were proposed , each with their advocates who considered the form that they were proposing to be the one true tensor .
this paper reviews the debate and its eventual conclusion : that no electromagnetic wave energy
momentum tensor is complete on its own .
when the appropriate accompanying energy momentum tensor for the material medium is also considered , experimental predictions of all the various proposed tensors will always be the same , and the preferred form is therefore effectively a matter of personal choice . |
the non - mesonic weak decay ( nmwd : @xmath0n@xmath4nn ) of @xmath0-hypernuclei gives us a very unique opportunity to study baryon - baryon weak interaction .
there exist several experimental observables ; life times(total decay rates ) , branching ratios(@xmath5(@xmath0p@xmath4np ) , @xmath5(@xmath0n@xmath4nn ) ) , etc .
the asymmetry parameter of decay proton from the @xmath0p@xmath4np process , , is another important observable to investigate the reaction mechanism of nmwd , because it comes from the interference between the parity - conserving and parity - violating amplitudes .
the decay angular distribution of protons from the nmwd of polarized hypernuclei , @xmath6(@xmath7 ) , is expressed as @xmath8 , where @xmath9 is the asymmetry , @xmath10 denotes the polarization of a @xmath0-hypernucleus , and @xmath7 means the emission angle of protons with respect to the polarization axis .
so far , two experiments reported the asymmetry parameters of the nmwd s of @xcite and @xcite . in kek - ps e160 ,
a large negative @xmath11 of [email protected] was observed for p - shell hypernuclei@xcite , while the @xmath11 of was measured to be [email protected] by kek - ps e278@xcite .
there is a large discrepancy between two values , which might suggest the difference of the reaction mechanisms of the nmwd s between s - shell and p - shell hypernuclei .
however , it would be too early to draw a conclusion in taking account of the statistical significance of the data and the systematic errors between two different measurements .
thus , a new measurement with improved statistics and better systematic errors had been awaited .
from year 2000 to 2002 , we performed two series of experiments to measure the nmwd of @xmath13he ( e462 ) and @xmath14c ( e508 ) in the k6 beamline at the kek 12-gev ps with the high - resolution and large - acceptance sks spectrometer .
the ( @xmath15 , k@xmath16 ) reactions at 1.05 gev/@xmath3 were used to produce highly - polarized @xmath0 hypernuclei with respect to the horizontal reaction plane .
two decay counter systems were installed symmetrically above and below the target to detect decay protons from nmwd .
each system consisted of a set of drift chambers , two sets of timing counters and a neutron counter array with six layers of plastic counters .
[ mhy_e462 ] [ mhy_e508 ] figure [ mhy ] shows excitation energy spectra of ( left ) and ( right ) in three different conditions : from top to bottom , inclusive spectra , those with a pion coincidence , and with a proton coincidence .
we succeeded to detect 52,000 events of the ground state and 67,000 events of the ground state , whose statistics are several times higher than the previous measurements .
the substitutional states of at around 10 mev excitation region , whose selection gate is shown in the figure , are open for proton emission , i.e. it goes to .
the asymmetry , @xmath9 , in eq .
( [ asymdefeq ] ) is obtained from the ratio , @xmath17 , of numbers of decay particles emitted to the parallel and anti - parallel directions with respect to that of the hypernuclear polarization .
owing to the large acceptance of sks spectrometer , we can polarize the hypernucleus upward(downward ) by selecting the ( @xmath1,k@xmath2 ) scattering direction on the left(right ) at one setup .
then , the @xmath17 is expressed as @xmath18 , where @xmath19 denotes the yield of the up ( down ) counter system at the scattering angle to the left ( right ) . in this expression , systematic errors coming from the differences of detection efficiencies and acceptances between the up and down decay counter systems are canceled out in the first order .
in fact , we found the asymmetries of pions and protons from the ( @xmath15 , p ) reaction , in which we do nt expect any asymmetries , were less than 0.3% . as shown in eq .
( [ asymdefeq ] ) , we need to know the polarization of a @xmath0 hypernucleus to obtain the asymmetry parameter of proton . in the case of , assuming the asymmetry parameter of the mesonic weak decay , , is same as that in free space([email protected] ) , we can measure the polarization of the from the pion decay asymmetry . in the case of p - shell hypernuclei , however , it is hard to estimate the polarization from the experimental data , because mesonic decay mode is suppressed due to pauli blocking , and we do nt have a good estimate of . therefore , we needed to estimate the polarization of p - shell hypernuclei with a help of a theoretical calculation in ref .
@xcite . + figure [ asympara ] shows a preliminary result of proton asymmetries in the decay of measured in different scattering angles , 6@xmath2015@xmath21 , as a function of the polarization of obtained from the mesonic decay .
the slope of the linear fit corresponds to the , and we found it to be [email protected]@xmath22 . as a systematic error , we took into account the effect of a small amount of pion contamination into protons , which tends to reduce the proton asymmetry to the negative side .
we confirmed the very small positive value of reported by ajimura @xcite with improved accuracy .
in addition , we found @xmath11 of p - shell hypernuclei obtained from a weighted average of each hypernucleus to be almost zero ( [email protected]@xmath22 ) as shown in fig .
[ asympara ] .
note that the present result is twice more accurate than the previous result . the ( p - shell ) of our result is rather consistent with that of .
it suggests that the decay mechanism of p - shell hypernuclei is similar to that of s - shell hypernuclei .
we have measured the decay asymmetry parameters of proton in the nmwd s of the polarized hypernuclei , , , and . for the s - shell hypernucleus ,
, it was found to be [email protected]@xmath22 , which is consistent to the previous value of [email protected] @xcite with smaller errors . for the p - shell hypernuclei , and , we found a similarly small value of [email protected]@xmath22 , which largely deviates from the old value of [email protected] @xcite .
the new values are different from those , -0.6@xmath24 - 0.7 , obtained in the recent theoretical calculations @xcite . we might need to reconsider the reaction mechanism of the nmwd . | we have obtained the decay asymmetry parameters in non - mesonic weak decay of polarized @xmath0-hypernuclei by measuring the proton asymmetry .
the polarized @xmath0-hypernuclei , , , and , were produced in high statistics via the ( @xmath1,k@xmath2 ) reaction at 1.05 gev/@xmath3 in the forward angles .
preliminary analysis shows that the decay asymmetry parameters are very small for these s - shell and p - shell hypernuclei . |
the r coronae borealis ( rcb ) stars are hydrogen - deficient supergiants , characterized by spectacular drops in visual brightness which occur at random intervals .
rcb stars are a very rare phenomenon with only about 35 known in the galaxy .
the atmospheres of these stars are composed mostly of helium with carbon and nitrogen enrichment .
their hydrogen depletion varies between undetectable hydrogen to severely depleted ( 5% by mass ) .
the declines in brightness are caused by the formation of clouds of carbon dust in front of the star ( see clayton 1996 for a review ) . to explain the existence of rcb stars , two scenarios have been proposed . in the first scenario
the rcb star is a merger of two white dwarfs ( wd ; iben , tutukov & yungelson 1996 ; saio & jeffery 2000,2002 ) , while in the second scenario a he - shell flash ( thermal pulse ) in a pre - wd inflates the remaining small envelope , leading to a born - again agb evolution . during this
evolution various mixing and nuclear burning processes can cause dramatic changes of the surface abundances ( iben et al .
however , no post - agb stellar evolution model can account for the predominance of helium in the atmospheres of most rcb stars ( herwig 1999,2000 ; pandey et al .
therefore , the wd - merger scenario might be better at explaining the rcb class .
unfortunately , this theoretical scenario has not been worked out in detail yet and skepticism is fueled , for example , by the long merging times of wd in binary systems ( iben et al .
some other classes of hydrogen - deficient post - agb stars show rcb - like light variations due to dust obscuration .
the born - again scenario is known to produce stars with the _ spectral and variability characteristics of rcb stars _ , which is why it was at first postulated that all rcb stars might be created this way . in this class ,
we find sakurai s object ( duerbeck & benetti 1996 ) , v605 aql ( clayton & de marco 1997 ) and fg sge ( gonzalez et al . 1998 ) , all of which were witnessed to undergo a born - again event and develop rcb spectral and light - curve characteristics .
these stars , therefore , create a strong link between the rcb and the born - again classes . a second class of stars ,
observationally linked to the rcb stars , is the wolf - rayet central stars of planetary nebula ( pn ) .
stars after van der hucht et al .
( 1981 ) , to distinguish them from massive wolf - rayet stars , or [ wc ] stars to emphasize that they are wolf - rayet star of the carbon , rather than nitrogen sequence .
] the link is established by two objects : the [ wc10 ] central star cpd56@xmath4 8032 ( de marco , barlow & storey 1997 ) , which has the same type of visual declines ( albeit shallower : @xmath51 mag ; jones et al .
1999 ) as rcb stars and the peculiar [ wc12 ] ( crowther , de marco & barlow 1998 ) v348 sgr , which is one of the most active rcb stars . the latter star , is also suspected to be a born - again star because its effective temperature could not maintain the pn s observed degree of ionization ( pollacco , hill & tudhunter 1990 ) .
the problem is therefore to understand what fraction of rcb stars share a common evolution ( wd - merger scenario or he - shell flash ) and what fraction simply shares the light - curve behavior and is grouped in the rcb evolutionary class for lack of a better explanation .
one might argue that the abundances should suffice as a discriminator .
however beside the abundances , there are other characteristics which should be taken into account when grouping these stars , because they seem to dictate different grouping rules . in this paper
, we concentrate on the above - mentioned v348 sgr , which , together with dy cen , mv sgr and hv 2671 ( in the lmc ; alcock et al .
1996 ) , constitute the class of hot rcb stars ( @xmath115000 - 20000 k ) . in both scenarios presently considered for the evolution of rcb stars ( born - again or merging wds )
, cool rcb stars should proceed at roughly constant luminosity toward the hotter region of the hr diagram .
therefore the hot rcb stars might be reasonably thought to represent the progeny of the cool ones . in this paper
, we consider the hot rcb stars in all their observable aspects , to try and place them in relation to other post - agb stars .
the abundances of the hot rcb stars , suggest that two of them , dy cen and mv sgr , should be classified as successors of the typical he - dominated cool rcb stars while v348 sgr and hv 2671 might belong to the post - agb born - again group .
however , additional characteristics do not unanimously support these sub - groups . after describing new and archival observations in sec .
[ sec : observations ] , we review the optical spectra of the four stars in sec . [
sec : spectra ] . in sec .
[ sec : observedparameters ] , we review the observed parameters of the stars , while in sec . [ sec : stell_par ] , we summarize the stellar parameters deduced from a mix of techniques . in sec .
[ sec : disk ] , we interpret the split line profiles with equatorial enhancements . in sec .
[ sec : neb_spec ] , we discuss the nebular properties . in sec . [
sec : lightcurve ] , we analyze the light - curve behavior . in sec .
[ sec : secularspectralvariabilityofthehotrcbstars ] , we discuss the spectral variability .
we finally summarize our analysis in sec .
[ sec : summary ] , compare the observations with the evolutionary models in sec.[sec : commparisonwithevolutionarymodels ] , leaving some final remarks to sec .
[ sec : theevolutionaryhistoryofthehotrcbstars ] .
a spectrum of v348 sgr was obtained on 1998 , may 14 on the 2.5 m isaac newton telescope at the observatorio del roque de los muchachos using the ids spectrograph .
the observation was taken with the 235 mm camera and the 300 l / mm grating giving a resolution of 6 .
the spectrum has been extracted , flat - field corrected , sky subtracted , and wavelength and flux calibrated ( with respect to the standard star bd+26@xmath4 2606 ) .
new spectroscopic observations were obtained of hv 2671 on 1997 , december 1 - 2 , when it was at light maximum .
the spectra were obtained with the sit / ccd spectrograph on the saao 1.9 m telescope at sutherland , south africa .
the grating used gives a reciprocal dispersion of 210 mm@xmath6 and a resolution of approximately 5 with a 250 slit , giving a useful range of about 3500 - 7500 at the angle setting used .
the spectrum has been extracted , flat - field corrected , sky - subtracted and wavelengthr- calibrated .
flux calibration is done by observing the standard star ltt 377 .
the flux calibration is not accurate because the instrument is not a spectrophotometer .
however since the observing conditions were photometric , we believe that the slope of the spectrum should be reliable .
two blue spectra ( 3900 - 4800 ) of dy cen were obtained at the anglo - australian telescope ( aat ) on 2001 , april 4 and september 9 , with the rgo spectrograph and 25 cm camera .
the 1200r grating was used in second order to give an effective resolution for 2 pixels of @xmath10.5 on the 4096x2048 eev42 detector .
the data were corrected for pixel response and wavelength calibrated using argon lamp images taken before and after the stellar exposures .
after sky subtraction , the 1-dimensional stellar spectrum was extracted .
the previously published and new visible light spectra for the sample stars are listed in tables [ tab : spectra1 ] and [ tab : spectra2 ] .
we have examined in detail those of the published spectra that could be made available to us by the authors in electronic format .
the spectra on plates listed in tables [ tab : spectra1 ] and [ tab : spectra2 ] were located in the lick plate archive and then scanned into tiff files with the kind help of tony misch and george herbig .
a cut was then made through the tiff files and a rough wavelength calibration was applied .
this method works well in providing spectra that can be easily compared in a _ qualitative _ manner to more modern spectra .
in particular we made use of the following spectra ( marked with an ` x ' in tables [ tab : spectra1 ] and [ tab : spectra2 ] ) : for hv 2671 we used a 1995 spectrum ( alcock et al . 1996 ) as well as the new spectrum described in sec.[ssec : newdata ] .
for v348 sgr we used a high resolution spectrum taken in 1987 ( leuenhagen , heber & jeffery 1994 , which we will refer to as lhj94 ) , a lower resolution spectrum taken in 1995 ( w .- r .
hamann , priv .
comm . ) , as well as the low resolution spectrum taken by us in 1998 ( sec .
[ ssec : newdata ] ) . for mv sgr
, we analyzed a high resolution spectrum taken in 1995 ( n.k .
rao , priv .
comm . ) and the two 1994 spectra of venn et al .
( 1998 ) . for dy cen
, we obtained two spectra , taken in 1989 and 1992 , respectively ( rao giridhar & lambert 1993 , giridhar rao & lambert 1996 ) .
all the analyzed spectra were taken when the stars were at maximum light .
low resolution uv spectra taken with iue , exist for v348 sgr , mv sgr and dy cen ( heck et al . 1982 ; evans et al . 1985 ; drilling & schnberner 1989 ; goldsmith et al . 1990 ; jeffery 1995 ; drilling et al . 1997 ) .
dy cen has only one long wavelength iue spectrum . in addition ,
hst fos spectra exist for v348 sgr during a decline ( two observations during a shallow minimum and one during a deep minimum ; hecht et al .
we have re - analyzed the maximum - light iue spectra for the three stars .
in figs . [ fig : atlas1 ] and [ fig : atlas2 ] , we present the optical spectroscopy of the three galactic hot rcb stars , while in fig . [
fig : spec1 ] we show the spectrum of hv 2671 compared to our low resolution spectrum of v348 sgr . figs .
[ fig : atlas1 ] and [ fig : atlas2 ] contain a mix of the spectra available , so as to obtain the largest spectral range in common to at least two stars .
in particular , for mv sgr we present the high resolution 1995 spectrum . in the spectral range 3800 - 4200 this spectrum is compared to the intermediate resolution calar alto spectrum of v348 sgr , while in the range 4200 - 5000 , it is compared with the high resolution lhj94 spectrum of v348 sgr . the dy cen spectrum in fig .
[ fig : atlas1 ] is from april 2001 . in the range 5600 - 6000
, we compare mv sgr to the lower resolution calar alto spectrum of v348 sgr ( no high resolution spectrum of v348 sgr exists in this spectral range ) and the 1992 spectrum of dy cen . finally ,
in the range 6000 - 6800 , we compare mv sgr to the high resolution v348 sgr spectrum and the 1992 spectrum of dy cen .
all stars have been shifted according to the measured radial velocities ( sec .
[ sec : observedparameters ] ) .
line identifications have been taken from leuenhagen et al .
( 1994 ) , giridhar et al .
( 1996 ) and jeffery et al .
the forbidden line spectra are discussed in sec.[sec : neb_spec ] .
the visible maximum - light spectrum of v348 sgr resembles a very cool [ wc ] central star of pn , although crowther et al .
( 1998 ) have argued that this object simply fails the [ wc ] classification in that it has no c iii @xmath75696 .
they have proposed a peculiar extreme helium star classification . however , the overall spectrum of v348 sgr at maximum light is somewhat similar to ( although cooler than ) the weak - lined [ wc11 ] central star k2 - 16 ( crowther et al .
it is also interesting that although many of the [ wc ] central stars have _ not _ been checked for variability , of those that have been monitored , only cpd-56 8032 ( a [ wc10 ] star ) has shown small rcb - like declines of about 1 magnitude ( pollacco et al .
1992 ; jones et al . 1999 ) .
the strongest emission lines in the spectrum of v348 sgr are from c ii ( pure emission , no p - cygni profiles ) and he i ( all with p - cygni profiles ) . lines of c iii are absent as are those of he ii , while some lines of c i are present .
all the balmer lines in our spectral range are present in emission with p - cygni profiles .
the spectra of hv 2671 and v348 sgr are , at 5 resolution , almost identical ( fig .
[ fig : spec1 ] ) .
the only exception is that the carbon emission spectrum looks stronger relative to the hydrogen and helium lines in v348 sgr .
this could be indicative of a larger carbon abundance in this star .
no obvious nebular lines are detected in the spectrum of hv 2671 ( see also sec . [
sec : neb_spec ] ) .
giridhar et al .
( 1996 ) discuss their two spectra of dy cen pointing out several remarkable differences . while the differences between the two spectra are discussed in depth in sec .
[ sec : secularspectralvariabilityofthehotrcbstars ] , there are some general characteristics which remain unchanged . the spectrum of dy cen is dominated by simple absorption lines of c ii , n ii , ne i and al ii and iii . at the same time , some c ii lines ( e.g. m2 and m5 ) and the he i lines , have a split absorption ( or an absorption with a reversed center ) in 1989 , while in 1992 they appear as weak split emission lines with weak p - cygni absorptions .
h@xmath8 is in emission with a p - cygni absorption .
although the spectrum of dy cen often exhibits absorption lines where the spectrum of v348 sgr has emission lines , the two spectra are much closer in appearance because more often than not they have lines in common . on the other hand , although the spectrum of mv sgr exhibits a wealth of emission lines , the strongest of them almost never coincide with strong lines in v348 sgr , giving mv sgr a completely different appearance to the other two stars .
the strongest emission lines in the spectrum of mv sgr belong , in fact , to metals ( the lines of fe i and ii , si ii , ni i , ca i and li i , are all split and have a peak separation of @xmath1100 km s@xmath6 , pandey et al .
1996 ) , with carbon lines in absorption , and helium lines in weak emission .
h@xmath8 is strongly in emission .
visible spectroscopy of rcb stars when they are below maximum light is rare .
herbig ( 1958 ) found that as v348 sgr faded , the c ii emission also faded to be replaced by a nebular spectrum including strong balmer lines and [ o ii ] @xmath73727 . by the time v348 sgr was @xmath9 mag ,
the c ii lines had disappeared .
dahari & osterbrock ( 1984 ) were able to see the c ii lines at minimum light ( @xmath10 = 18.4 mag ) , but they were very weak compared to maximum - light , and weaker than the nebular lines .
no spectra have been obtained of the other three stars when they were below maximum light . the low resolution , long wavelength iue spectra of v348 sgr , dy cen and mv sgr at maximum light are all very similar , once brightness and reddening are taken into account ( sec .
[ sec : observedparameters ] ) .
the most prominent line in that range is an emission line of mg ii at 2800
. the low resolution , short wavelength , maximum - light spectra of mv sgr and v348 sgr are also quite similar ( there is no short wavelength spectrum for dy cen ) .
the most prominent lines appear to be absorptions of c ii @xmath71334.5 ( @xmath11p-@xmath11d ) and c iv @xmath71550 ( @xmath11s-@xmath11p@xmath12 ) .
hecht et al .
( 1998 ) have published hst fos spectra of v 348 sgr , during shallow and deep minimum light .
in this section , we discuss the observed parameters of the hot rcb stars . these include the radial velocity of the star , the observed @xmath13 brightness and the reddening . for the lmc hot rcb star hv 2671
we also include the distance ( which is accurately determined ) and the resulting @xmath2 value , among the observed quantities .
for the other three stars , the distances and @xmath2 values are discussed , together with the stellar models , in sec .
[ sec : stell_par ] .
although in the text we discuss most of the literature parameters along with our own - derived values , in table [ tab : basic ] we list only the adopted values . from the lhj94 spectrum , we determine an lsr ( heliocentric ) radial velocity for v348 sgr of 142 km s@xmath6 ( 130 km s@xmath6 ) from pure emission lines ( the o ii m24 component at 4751.3 , the c ii m22 component at 5836.4 , and the only un - blended nebular line , [ n ii ] @xmath76543 ) .
when carrying out the same measurement on lines with obvious p - cygni absorptions ( e.g. , h@xmath8 ) , the resulting lsr ( heliocentric ) radial velocity is 157 km s@xmath6 ( 145 km s@xmath6 ) .
this is due to the fact the p - cygni absorption eats into the blue side of the emission lines , displacing their apparent position to the red .
there are many estimates of the reddening of v348 sgr ( see table [ tab : red ] ) . of these ,
the balmer decrement estimates of dahari & osterbrock ( 1984 ) , and houziaux ( 1968 ) are suspect because the nebular lines are contaminated by the stellar hydrogen emission .
pollacco et al .
( 1990 ) observed the nebula off the star and could therefore determine more reliable nebular fluxes .
most of the other estimates lie within 0.2 mag of the pollacco et al .
( 1990 ) estimate .
we adopt their value of @xmath14 = ( 0.45 @xmath15 0.1 ) mag .
this value nulls effectively the 2200- feature in the low resolution iue spectrum of this star . at present , the maximum light magnitude of v348 sgr is @xmath10 = 12.0 mag ( american association of variable star observers [ aavso ] ) .
alcock et al . ( 2001 ) estimate the reddening toward the bar of the lmc to be @xmath14 @xmath1 0.17 mag
. this would also be the reddening toward hv 2671 , in the absence of circumstellar reddening .
the reddening toward hv 2671 was also determined from intrinsic colors to be @xmath14 @xmath1 0.3 mag ( alcock et al .
alternatively , we can determine the reddening to hv 2671 by comparing the slope of its flux - calibrated spectrum , with that of v348 sgr , de - reddened by our adopted @xmath14 = 0.45 mag .
an almost perfect match is derived for a value of @xmath14 = 0.15 mag .
given that the rectified spectra are extremely similar ( fig .
[ fig : spec1 ] ) , this is thought to be a reasonable method .
the two estimates agree within the uncertainties involved .
we adopt the latter value for this study .
this implies that there is a minimum amount of circumstellar reddening . the brightness of hv 2671 at maximum light is @xmath10 = 16.1 mag ( alcock et al .
2001 ) . then , using the lmc distance modulus ( @xmath16 = [ [email protected] ] mag ; feast 1999 ) and assuming @xmath17 = 3.1 , the absolute @xmath13 magnitude of hv 2671 is therefore @xmath2 = 3.0 mag . from the absorption line spectrum of dy cen in 1989 ,
giridhar et al .
( 1996 ) determined the radial velocity to be ( 41@xmath154 ) km s@xmath6 from 25 c ii , 14 n ii and 20 ne i lines .
similar measurements on the 1992 spectrum returned a value of ( 29@xmath153 ) km s@xmath6 . from the nebular lines [ n ii ] @xmath7@xmath76548,83 , [ o i ] @xmath76300 , [ s ii ] @xmath7@xmath76717,31 and h@xmath8 , they determined mean values of ( 23@xmath152 ) km s@xmath6 and ( 22@xmath152 ) km s@xmath6 for 1989 and 1992 , respectively . from the emission components of 8 c ii lines , they determined radial velocity values in the range 28 - 123 km s@xmath6 in 1989 and 20 - 32 km s@xmath6 in 1992 .
giridhar et al .
( 1996 ) also used he i lines which , in 1989 , were all in absorption , returning a value for the radial velocity in the range 4 - 53 km s@xmath6 , while in 1992 were in emission with p - cygni absorption . from the emission part they determined radial velocities in the range 2 - 55 km s@xmath6 .
we adopt the radial velocity value determined from the nebular emission lines , the least affected by stellar variability , as the most representative of the systemic velocity of the star .
we should also note that the absorption lines in each of the two epochs , have very little radial velocity scatter , but there is a significant change between 1989 and 1992 .
this could be understood if dy cen is a spectroscopic binary ( as also suggested by giridhar et al .
[ 1996 ] ) .
if this fact could be corroborated , _ it would be the first confirmation of binarity in an rcb star_. the fact that the emission lines ( or lines that developed emission in 1992 ) have such a large radial velocity scatter will be amply discussed in sec .
[ sec : secularspectralvariabilityofthehotrcbstars ] .
the reddening was determined by jeffery & heber ( 1993 ) to be @xmath14 = 0.50 mag , from a stellar continuum analysis . when this value is used the slope of the iue spectrum of dy cen
aligns with that of the de - reddened spectrum of v348 sgr .
the maximum - light brightness of dy cen is @xmath10 = 13.0 mag ( aavso ) .
the radial velocity of mv sgr was measured from the 1995 spectrum . from the absorption lines of c ii @xmath74267 and he i @xmath74437 , we obtained un - corrected values of 72.5 and 60.7 km s@xmath6 , respectively . from the split emission line of fe ii @xmath75991 , we obtain 62.2 km s@xmath6 , while from the strong emission line of ca ii at 8542 we obtain 66.7 km s@xmath6 .
a mean of these values , corrected for lsr ( heliocentric ) radial velocity correspond to ( 83@xmath155 ) km s@xmath6 ( [ 95@xmath155 ] km s@xmath6 ) .
this compares favorably with the value of ( 93@xmath154 ) km s@xmath6 , determined by pandey et al .
( 1996 ) from their 1992 spectrum ( we presume their value to be corrected for the heliocentric standard of rest , although they never state that explicitly )
. this also agrees well with earlier estimates of jeffery et al .
( 1988 ; [ 91@xmath157 ] km s@xmath6 from the absorption lines , [ 81@xmath1517 ] km @xmath6 from the emission lines ) and rao , giridhar & nandy ( 1990 ; [ 95@xmath158 ] km s@xmath6 ) .
the reddening toward mv sgr was estimated by heber & schnberner ( 1981 ) from a comparison of observed ( landolt 1979 ) and predicted colors , to be @xmath14 = 0.45 mag .
drilling et al .
( 1984 ) determine @xmath14 = 0.38 mag from the strength of the 2200- bump . from comparing the de - reddened ( @xmath14 = 0.45 mag ) uv spectrum of v348 sgr to that of mv sgr we determine @xmath14 = 0.43 mag .
the maximum - light brightness of mv sgr is @xmath10 = 13.4 mag ( aavso ) .
in this section , we summarize the stellar parameters of the hot rcb stars from the literature . for all the hot rcb stars except for hv 2671 , these have been determined from modeling of the stellar continuum and lines . for hv 2671 , we derive the parameters by analogy to those of v348 sgr , but accounting for the known distance to the lmc . for this star
we have the only truly reliable estimate of @xmath2 .
the selected parameters are summarized in table [ tab : model ] .
leuenhagen & hamann ( 1994 ) modeled the uv and optical spectra of v348 sgr with their non - local thermodynamic equilibrium ( lte ) code for wr atmospheres ( hamann & koesterke 1993 ) . the stellar parameters they derive are summarized in table [ tab : model ] for their adopted distance of 5.4 kpc ( a value which can not presently be improved upon ) . beside what is listed in the table ,
they find @xmath18 = 190 km s@xmath6 using a beta velocity law with @xmath19 = 2 , and a neon abundance of 2% ( by mass ) . their hydrogen abundance of 4% ( by mass ) , might be an upper limit . the h@xmath8 line , on which they base their abundance estimate , is in fact a blend of stellar and nebular emission .
although the nebular line spectrum during maximum light is faint , we can estimate the nebular contribution to the observed h@xmath8 emission line in the following way .
we measured the equivalent width of the [ n ii ] line at 6548 , which can easily be de - blended from the neighboring stellar n ii m45 component with a double gaussian .
this is ( [email protected] ) . from pollacco
et al . ( 1990 ) we see that the nebular h@xmath8 in their spectrum is 6.2 times stronger than the [ n ii ] @xmath76548 line . from this
we can estimate the equivalent width of the nebular h@xmath8 contribution to be ( [email protected] ) .
the total equivalent width of the h@xmath8 feature in our high - resolution spectrum is ( [email protected] ) , so that the stellar contribution to this feature is predicted to be ( [email protected] ) , or less than half of the total strength .
the hydrogen abundance derived by leuenhagen & hamann ( 1994 ) is therefore a comfortable upper limit and so , we note that this star might be much more hydrogen - poor than previously thought .
code improvements ( such as the addition of line blanketing ) could easily lead to slightly different parameters , although there is no reason to believe that the stellar picture would be fundamentally changed .
a more serious unknown is the departure from spherical symmetry unaccounted for by the models ( sec .
[ sec : disk ] ) and its effects on the line intensities used to determine the stellar parameters .
v348 sgr contains absorptions as well as emission lines .
therefore , plane - parallel codes can also be used to determine its stellar parameters and abundances .
this provides us with a formidable check that fundamentally different codes such as spherical , expanding transfer codes and plane parallel static codes do indeed return parameters in the same ball park . without this check
one might fear that the different abundances obtained for rcb stars and for [ wc ] stars might in the end be an artifact . from an lte analysis of the absorption line spectrum , jeffery ( 1995 )
derived @xmath20 = 22,000 k , log g@xmath21 2.8 , and assuming the hydrogen mass fraction of leuenhagen et al .
( 1994 ) , he derived the following mass fractions : @xmath22 = 0.67 , @xmath23 = 0.28 , with log@xmath24 = 2.73 , log@xmath25 = 2.92 and log@xmath26 = 4.27 . although one could argue that the agreement is poor , we can none the less conclude that indeed the abundances of v348 sgr are closer to those of [ wc ] central stars than to the those of rcb stars including dy cen and mv sgr ( see later ) .
the temperature of hv 2671 has not been measured independently , but because of its similarity to v348 sgr , we adopt the same temperature ( 20000 k ) .
based on it and the related bolometric correction of 1.71 mag , together with the observed @xmath13 magnitude at maximum light , the reddening of @xmath14 = 0.15 mag determined in sec .
[ sec : observedparameters ] and the known distance to the lmc ( 50 kpc ; feast 1999 ) , we determine a bolometric luminosity for hv 2671 of @xmath16000 l@xmath3 .
the radius then follows from the boltzmann formula .
based again on the similarity of the spectrum at low resolution we have no indication that the abundances are not similar to those of v348 sgr .
dy cen was analyzed by jeffrey & heber ( 1993 ) . through an lte analysis of the absorption lines they derived effective temperature , gravity and abundances ( listed in table [ tab : model ] ) .
the abundances of dy cen are more similar to the mean abundances of other rcb stars than to v348 sgr .
they also derive a mass of 0.8 m@xmath3 assuming that the stellar interior consists of a co core .
this value of the mass , together with their derived value of the gravity implies a radius of the star of 12.5 r@xmath3 , and a luminosity of 20200 l@xmath3 .
they also establish an upper limit for @xmath27 of 20 km s@xmath6 .
the emission line spectrum of dy cen was examined by giridhar et al .
( 1996 ) , who reported several variations between those two epochs ( sec .
[ sec : secularspectralvariabilityofthehotrcbstars ] ) .
no analysis has ever been carried out of these emission lines . as for the distance to this object ,
rao et al . (
1993 ) quote a range of 1.4 - 4.8 kpc . at 4.8 kpc , with our own values of the observed @xmath13 magnitude and preferred reddening ( @xmath14 = 0.5 mag )
we determine @xmath2 = 2.0 mag .
this is much fainter than for the other three stars .
we therefore suggest that the distance to dy cen might be only a lower limit .
the stellar parameters of mv sgr have been derived by drilling et al .
( 1984 ) and jeffery et al .
( 1988 ) from plane - parallel lte analyzes . in the analysis of drilling et al .
( 1984 ) we find an estimate of the temperature based on a fit to uv spectrophotometry and ubv photometry .
they derive t@xmath0 = ( 16000@xmath15500 ) k and @xmath14 = 0.38 mag for a de - reddened @xmath13 magnitude of 11.9 , derived from a flux of 6.4 x 10@xmath28 erg @xmath29 .
assuming @xmath14 = 0.38 mag , this corresponds to an observed maximum - light @xmath13 magnitude of 13.1 mag .
this observed value is not from glass ( 1978 ) , as cited in drilling et al .
apparently , they used an observed value of @xmath13 = 12.70 mag from herbig ( 1964 ) .
the discrepancy between these values can not be explained .
using this temperature jeffery et al .
( 1988 ) derive the abundances from fits to the absorption lines ( table [ tab : model ] ) .
they also derive a surface gravity log(g ) = [email protected] by analogy with hd 124448 , admitting to a substantial uncertainty .
the hydrogen abundance is quoted as being small .
herbig ( 1964 ) says there is no sign of hydrogen absorption .
pollacco & hill ( 1990 ) state that there is no doubt about the hydrogen deficiency .
the emission lines of mv sgr were studied by pandey et al .
( 1996 ) , although no parameters are derived from them .
heber & schnberner ( 1981 ) derive a temperature of 19000 k , using a reddening @xmath14 = 0.46 mag , from which they derive @xmath2 = 3.9 mag and a distance of 7.8 kpc .
we have not added these values to the table since the abundance analysis is dependent on the temperature and reddening values of drilling et al .
( 1984 ) , while the other estimates ( like the distance ) are inter - related with the higher temperature determined by heber & schnberner ( 1981 ) .
it was noticeable that the double peaked line profiles seen in the c ii and he i lines of v348 sgr & dy cen and the c i , he i and h i lines of mv sgr ( pandey et al .
1996 ) closely resemble those observed in some be stars ( where they are assumed to arise in an equatorial disk ) , and therefore suggest a departure from spherical symmetry for the mass out - flows from these stars .
however , despite the similarities of the emission line profiles of be stars presented by dachs , hummel & hanuschik ( 1992 ) to those presented here we do not believe the circumstellar environment of the hot rcb stars are exactly analogous ( although we do suspect an equatorial density enhancement for the hot rcb stars ) .
current theories for the behavior of be star disks suggest that they are geometrically thin quasi keplerian excretion disks that are supported by the input of angular momentum and material at the disk / star boundary ( e.g. porter 1999 ) . however , the line profiles observed for mv sgr and v348 sgr differ in three respects from the predictions of a _
keplerian _ disk . 1 . for a keplerian disk
one would expect the weaker emission lines to be produced in a smaller region of the disk than the stronger lines and hence have a larger line width due to the greater rotational velocities of the inner regions of such a disk ( e.g. the weaker fe ii lines have a greater width than h@xmath8 for be stars disks ; dachs , hummel & hanuschik 1992 ) .
however the line widths of the weaker emission lines are systematically lower than those of stronger transitions for mv sgr ( pandey et al .
2 . the high resolution line profiles presented for mv sgr by pandey et al . ( 1996 ) show large asymmetries in the peak ratios of the double peaked line profiles . in classical
be stars this is attributed to the slow precession of a one armed density wave within the keplerian disk ( e.g. okazaki 1997 ) and the peak ratios of different lines are observed to vary in phase ( such that for a given time the blue peak is stronger than the red [ or vice versa ] for all lines ) .
the line profiles for mv sgr clearly show that this is not the case , and the asymmetries can not arise from the precession of a density wave within a keplerian disk .
the p - cygni components of the line profiles of mv sgr and v348 sgr are _ not _ observed for classical be stars . indeed , there is strong evidence that the radial component of the disk velocity is highly subsonic ( @xmath211 km s@xmath6 ) until large radii ( of the order of 100r@xmath30 ; okazaki 2001 ) , well outside the region responsible for line formation .
consequently the presence of the p - cygni absorptions at @xmath1100 km s@xmath6 in mv sgr and v348 sgr can not arise in a disk of similar structure to those of classical be stars .
4 . neither active dust nor forbidden line emission
is observed from the equatorial _
keplerian excretion _
disks found around classical be stars .
however , despite the differences between classical be stars and hot rcb stars the presence of an equatorial density enhancement of some form is attractive for a number of reasons .
polarimetric observations of r crb suggest the presence of an obscuring dusty torus ( rao & lambert 1993 , clayton et al .
1997 ) , while changes in the polarization of other rcb stars during deep declines suggest the observed light is scattered ( e.g. clayton 1996 , and references within ) - obscuration of the star by an equatorial torus with scattering of light from the polar axis into the line of sight would explain these observations .
the formation of dust in a spherical wind is also difficult because at large radii , where the temperature is low enough to allow dust to condense ( @xmath13000
k ) for mass - loss rates as low as those observed for hv 2671 & v348 sgr ( table [ tab : model ] ) , the wind density is simply too low to allow dust formation ( e.g. bjorkman 1998 ) ; a ` disk ' or ` torus ' could permit high enough densities to exist for dust to form .
additionally , the higher densities allow more efficient cooling - modeling of the energy balance of geometrically thin disks around be stars by millar & marlborough ( 1999 ) suggest that they are approximately isothermal with a temperature of @xmath10.5t@xmath31 . with additional cooling due to the high metal content of hot - rcb winds and the possibility of self shielding produced by a geometrically thick disk the temperature within such a structure may be low enough to permit dust formation .
a class of objects where the same problem of dust production in an hostile environment occurs are the b[e ] stars , stars of spectral type b with a rich emission line spectrum and pronounced ir excess due to the presence of equatorially concentrated hot dust ( the presence of forbidden line emission and hot dust distinguishes these objects from classical be stars ) .
however galactic b[e ] stars are a rather heterogeneous class due to difficulties in determining absolute luminosities and comprise both pre- and post - main sequence stars , with luminosities ranging from log(@xmath32 ) = 2.56.5 ( zorec 1998 ) . lamers et al . ( 1998 ) provide a summary of known objects and conclude that 12 galactic b[e ] stars are likely to be compact pne , while a smaller number appears to be post - agb objects - hen 3 - 1191 ( lamers et al .
1998 ) , hen3 - 847 ( miroshnichenko priv . comm . ) , mwc 922 and mwc 300 ( voors 2000 ) .
given that the hot - rcb stars share many of the same observational properties as the b[e ] stars ( indeed mv sgr meets the membership criteria listed by lamers et al . [ 1998 ] ) and other objects initially classified as b[e ] stars are now recognized to be post - agb objects , it is tempting to suggest that the hot - rcb stars share a similar circumstellar environment as the b[e ] stars .
inspection of the spectrum of the b[e ] star hd 87643 ( oudmaijer et al .
1998 ) shows that the profiles of the h i balmer series closely resemble those of the strong emission lines of c i , he i and h i in mv sgr which also show 2 absorption components ; a pronounced p - cygni absorption trough ( at @xmath1100 km s@xmath6 - the larger velocity of the blue edge of the p - cygni profile in hd 87643 reflecting a higher terminal velocity ) and the central reversal ( which appears to be slightly blue shifted in both hd 87643 and mv sgr ; pandy et al .
additionally , as with mv sgr , the weaker fe ii lines in hd 87643 are observed to be half the width of the balmer lines , while the observable forbidden lines are narrower still .
the same is also true of v348 sgr and dy cen although we note that the central reversals in these stars do not show any velocity shift , apparently indicating absorption by @xmath1stationary material .
we note that this behavior is also observed for the be shell stars which have deep central absorption troughs - it is thought that these stars represent the subset of classical be stars that are seen at high inclinations where the line of sight lies close to or in the equatorial plane and hence intersecting the circumstellar disk .
oudmaijer et al .
( 1998 ) interpret this behavior as a result of the presence of at least 2 and probably 3 different wind regimes ; a very strong high velocity wind responsible for the broad balmer lines ( and the p - cygni absorption ) , a second , lower velocity wind traced by the fe ii lines and the core of the balmer lines ( the central absorption is attributed to self absorption in a slow wind rather than doppler splitting due to rotational motion ) .
the final component , sampled by the narrow forbidden lines is attributed to a possible low density nebula ( c.f .
v348 sgr and dy cen ) .
furthermore , changes in the polarization across the h@xmath8 line and adjacent continuum clearly indicates a departure from spherical symmetry and indicate the presence of both expansion and rotational velocity components ( note that line emission from the classical be star disks shows _ _ no evidence for radial expansion ) .
oudmaijer et al .
( 1998 ) suggest that the varying wind regimes can be accommodated in a scenario where a radiation driven wind from a slowly expanding dense circumstellar disk ( responsible for the fe ii lines and central absorption in the balmer lines ) interacts with a faster polar wind ( responsible for the p - cygni absorption line components ) - given the similarities in the line profiles of the hot - rcb stars to those of hd 87643 we suggest a similar hybrid wind regime could be present in these stars as well .
the pronounced double peaked fe ii profiles and flat topped forbidden lines observed in mv sgr by pandy et al .
( 1996 ) are not found for hd 87643 by oudmaijer et al .
( 1998 ) ; however we suggest that the permitted fe ii lines may arise from a disk wind while the narrower forbidden lines may arise at relatively large radii where the distinction between the different regimes has largely disappeared .
although this seems somewhat contrived we note that flat topped optically thin line profiles typically arise in quasi - spherical outflows - therefore whatever mechanism is invoked to explain the double peaked fe ii , he i and c i lines must also allow for a quasi spherical outflow , which necessitates a change in geometry at large radii ( since the forbidden lines must arise in a low density region of the outflow ) . while the very narrow forbidden line emission in hd 87643 is absent from mv sgr , we note that it is present in both dy cen and v348 sgr suggesting the presence of a slow moving ` nebular ' component to the circumstellar environment of all three systems . in fig .
[ fig : pedestal ] we compare all the v348 sgr c ii lines which show the split ( namely @xmath74267.3 @[email protected],6582.9 and @[email protected],5891.6 ) , as well as he i @xmath75876 . from this figure
it can be seen that the components are all at similar radial velocity , separated by about ( 70@xmath155 ) km s@xmath6 .
we have fit c ii @xmath74267 and he i @xmath75876 ( the strongest , least blended lines in the sample ) with a combination of gaussians , to aid in their interpretation .
the central wavelengths of the fit components and their widths were free to vary . we started by fitting the c ii @xmath74267 profile with three gaussians , one for the broad pedestal and one each for the two emission peaks .
the fit is shown in fig .
[ fig : hei5876 ] ( a ) ( top ) while the fit parameters are listed on the left - hand - side of table [ tab : fit ] ( test 1 ) .
second ( test 2 ) , we could interpret the line morphology _ not _ as a split line but as a _ stellar _ wind emission line with a broad base and a narrow top , blended with an absorption line from material intersecting the line of sight .
we therefore fit the profile ( fig .
[ fig : hei5876 ] ( b ) ) with two emission gaussians , a broad one for the base and a narrower one for the rest of the line , after removing the central trough by snipping out the piece of spectrum . in this way ,
only the line wings were fit .
we then subtracted the full observed profile from the fit obtaining the shape of the absorption line ( fig .
[ fig : hei5876 ] ( c ) ) .
the resulting absorption line was then fit with a cloud model whose parameters we list in table [ tab : cloud ] . from this exercise
we concluded that if indeed there is a shell of c ii which is absorbing flux from the emission line , it must be approximately stationary ( the center of the line is at 3 km s@xmath6 ) with an internal velocity dispersion of 20 km s@xmath6 and a total column density of log(n/@xmath33 ) = 12.2 ( table [ tab : cloud ] ) . the same steps were carried out for the he i line at 5876 .
the only difference is that this line does not exhibit a broad pedestal .
it does however have a significant p - cygni absorption profile on its blue side . in table
[ tab : fit ] , we list the fit parameters for the he i line alongside those for the c ii fits .
the fit components which are directly comparable between the two lines are highlighted in bold .
line profile 1 , in each of the two tests , represents the broad pedestal in the c ii case and the p - cygni absorption in the case of the he i line , and is therefore not comparable between the c ii and he i fits .
we compare the two emission component positions and widths from fit test 1 . as we can see from figs .
[ fig : hei5876 ] ( a ) , the fits are excellent in both cases although the parameters vary considerably between the c ii and he i fits .
the positions of components 2 and 3 in test 1 are different in the c ii and he i lines ( 43 and 37 km s@xmath6 for c ii vs 22 and 43 km s@xmath6 for he i ) .
the split lines components do not reside at similar radial velocities which excludes a keplerian disk as their origin . on the other hand ,
if we turn to fit test 2 , we see that the position of the emission component is similar to that of the absorption trough . additionally , the number ratio of the column densities for the absorbing shells is c / he = 0.79 , reasonably close to the value of 0.45 determined by leuenhagen & hamann ( 1994 ) for the star .
this interpretation is more in line with the equatorial density enhancement more typical of b[e ] stars .
finally , we would expect the h@xmath8 profile in the spectrum of v348 sgr to also exhibit a double peak .
the reason it does not is likely to be due to the fact that the nebular h@xmath8 is filling in the trough . comparing the h@xmath8 profile to the c ii m1 component at 4745 ( fig .
[ fig : ha_comp ] ) we see that the line widths are similar , although h@xmath8 is asymmetric , with more emission in the red part than the blue , somewhat more similar to he i @xmath75876 ( except for the split ) than to the c ii profile .
additionally , the p - cygni profiles of the h@xmath8 and he i lines align , showing their origin to be similar . in order to estimate the nebular contribution to the h@xmath8 profile
, we fit the [ n ii ] line at 6548 to determine its equivalent width .
we then multiplied the profile fit by 6.2 , as explained in sec .
[ sec : stell_par ] and used it as a representation of the nebular contribution to the h@xmath8 profile . the resulting predicted profile ( h@xmath34 )
is plotted in fig .
[ fig : ha_comp ] ( b ) , together with the observed h@xmath8 profile ( h@xmath35 ) .
we then subtracted the predicted nebular profile from the total observed profile , obtaining the stellar contribution . the resulting profile ( h@xmath35 - h@xmath34 ) is also plotted in fig .
[ fig : ha_comp ] , displaced by 1.2 units , and compared to the he i @xmath75876 profile .
although the match is not perfect the peaks and trough do roughly correspond .
this hints at the fact that the stellar component of the h@xmath8 profile might be weak , split like the he i and c ii components but almost completely overwhelmed by the nebular component of the same line .
we feel that further considerations would over - interpret the data .
in this section , we review what is known about the nebulae around hot rcb stars .
unfortunately the nebular spectra at maximum light is extremely weak , so that it is impossible to carry out a full nebular analysis .
additionally , unless spatial information can be derived , it is unclear whether the forbidden lines arise far from the star , in a pn - type nebula , or whether they are formed in the outer parts of a stellar wind .
the first striking characteristic that distinguishes v348 sgr , is that it has a large ( 30 arcsec , pollacco et al .
1990 ) , ionized nebula , much like an old pn .
pollacco et al .
( 1990 ) carried out a nebular analysis of the ionized gas around v348 sgr , by observing the nebula off the star , when the star was at minimum light .
they observe lines of [ o i ] , [ o ii ] and [ o iii ] ( although [ o i ] is flagged as of possible sky origin and [ o iii ] is extremely weak ) , [ n i ] and [ n ii ] ( [ n i ] as weak as [ o iii ] ) , [ s ii ] and the balmer lines . additionally , they warn that the he i lines appear clearly extended in their spectrum , and may therefore also be of nebular origin .
the only nebular lines that could be measured in the lhj94 spectrum of v348 sgr are [ n ii ] @xmath76548 and [ o i ] @xmath7@xmath76300,6363 . their fwhm are 52 km s@xmath6 and 19 km s@xmath6 , respectively .
it is likely that the [ o i ] lines are of sky origin , while the [ n ii ] lines derive from the nebula .
if so then the nebular expansion velocity , derived from hwhm of the [ n ii ] line is 26 km s@xmath6 .
this is higher than 4.6 km s@xmath6 , the expansion velocity derived by pollacco et al .
( 1990 ) from the nebular line split .
although the line splitting in their high resolution spectra is labeled as _ possible _ , due to a low signal - to - noise ratio , their data exclude an expansion velocity as large as 26 km s@xmath6 .
other nebular lines in the lhj94 spectrum of v348 sgr include [ n ii ] @xmath76583 , blended with stellar c ii ( m2 ) .
[ o ii ] @xmath7@xmath73726,29 lie outside its spectral range , while [ s ii ] @xmath7@xmath76717,31 are blended with stellar c ii and o ii lines .
the issue of h@xmath8 has been discussed in sec .
[ sec : stell_par ] .
the nebular contribution is likely to contaminate substantially the stellar emission .
there is no doubt , however , of a stellar contribution , since a blue shifted p - cygni profile is evident in the lhj94 spectrum . considering the similarity of the spectra of v348 sgr and hv 2671
, one might wonder whether hv 2671 also has an ionized pn .
however , the lhj94 spectrum of v348 sgr shows the difficulty of detecting the pn when the star is at maximum light . detecting the pn at maximum light _ and _ low resolution might therefore be impossible
. inspection of our low resolution spectrum of v348 sgr , as seen in fig .
[ fig : spec1 ] , reveals that even the pn lines known to be present in the lhj94 spectrum , namely the [ n ii ] pair at 6548 and 6583 , are hard to identify ( the former is too weak and the latter blended with c ii lines ) .
[ o i ] @xmath366300,6363 , observed in the lhj94 spectrum , although suspected to be of sky origin , are not observed at all in the calar alto low resolution spectrum .
[ s ii ] @xmath366717,6731 are equally undetected .
the only possible detection of a nebular line in the low resolution spectrum of v348 sgr are [ o ii ] @xmath363726,3729 , but there is no evidence that the same feature is present in the spectrum of hv 2671 ( although the blue part of the spectrum suffers from a very poor signal - to - noise ratio ) . in conclusion , there is no evidence that hv 2671 has an ionized nebula although the low resolution of its spectrum should be blamed as the primary cause .
nebular emission lines of [ o i ] , [ s ii ] and [ n ii ] are detected in the spectrum of dy cen ( rao et al . 1993 ) , but are much narrower than those detected in the spectrum of mv sgr . from dy
cen s [ s ii ] lines at 6717 and 6731 rao et al .
( 1993 ) derive @xmath37 = ( 450@xmath15100 ) @xmath33 ( closer to the value for v348 sgr than that for mv sgr ) , for a temperature range @xmath38 = 500020000 k. they estimate the electron temperature to be @xmath2110000 k from the absence of [ n ii ] @xmath75755 . the fact that the lines in their spectrum are not extended leads to the conclusion that the nebula around dy cen has a diameter of less than 2 arcsec
this is considered peculiar since the nebula has low electron density and , at a distance of 4.8 kpc , should have an extent of 7 arcsec ( for @xmath39 = 20 kk and @xmath40 = 2300 l@xmath3 ) .
however , we have noted ( sec [ sec : stell_par ] ) that dy cen is likely to be further away than 4.8 kpc . in the spectrum of mv sgr , nebular lines of [ s ii ] and [ o i ] were first seen by herbig ( 1964 ) .
pandey et al .
( 1996 ) observed nebular lines such as [ n ii ] and [ o i ] with widths and shapes similar to those of the allowed lines ( i.e. @xmath1100 km s@xmath6 ) .
in addition , they used the forbidden line fluxes to derive the electron density , @xmath37 = ( 37)@xmath4110@xmath42 @xmath33 , and electron temperature , @xmath38 = 850020000 k. we confirm the detection of broad [ n ii ] and [ o i ] lines in the 1995 spectrum of mv sgr ( the [ o i ] lines at 6300 and 6363 also have a narrow sky component ) . due to their widths
, one has to deduce that they are forming possibly in a tenuous outer stellar envelope rather than in a proper pn .
the high electron density in particular confirms that the nebula around mv sgr is unlike a typical pn ( although very young pne like hb 12 or ic 4997 have densities of 10@xmath43 and 10@xmath44 @xmath33 , respectively , pottasch 1984 ) and is more likely to be the outer envelope of the star .
forbidden lines were also found in the cool rcb stars by herbig ( 1949 ) , who saw [ o ii ] @xmath73727 in the spectrum of r crb .
more recently , broad emission lines of [ o i ] , [ n ii ] and [ s ii ] were seen during the 1992 decline of v854 cen . in the 1995 - 96 decline of r crb ,
rao et al . ( 1999 ) found both narrow [ c i ] , [ o i ] , [ ca ii ] and [ fe ii ] lines and broad [ ca ii ] , [ n ii ] and [ o ii ] forbidden emission lines .
the issue of the presence and characteristics of nebulosities around rcb stars and other h - deficient stars like extreme helium stars or hydrogen - deficient carbon stars , is paramount to solving the question of their origin . beside the evidence for ionized shells around the three hot rcb stars , we know that uw cen has a neutral nebula that is about 15 arcsec across ( pollacco et al .
1991 ; clayton et al . 1999 ) .
r crb has a nebula only seen with iras , several arcminutes across ( gillett et al .
v854 cen has a 5-arcsec nebula seen in c ii @xmath71335 ( clayton & ayres 2001 ) .
no other rcb star is known to have a shell of any kind .
this has always been a major stumbling block to the theory that rcb and central stars of pn are closely related . in theory , if rcb derive from a born - again type scenario it is possible that their pn have long recombined .
rcb stars are defined by their light - curve properties , where deep declines are followed by recovery in a random fashion .
however the range of light - curve activity exhibited by rcb stars is extremely varied .
two characteristics that have emerged so far are that ( a ) rcb stars produce dust in ` puffs ' near the stellar surface and then blow them away and ( b ) that the dust production is related to the pulsation cycle of the stars ( clayton 1996 and references therein ) . on the other hand we still wonder : is the dust produced preferentially on an equatorial region ?
how can dust form near the surface of a 6000k-20000k - star ?
theoretical work has gone a long way to explaining dust formation in hostile environments ( woitke , goeres & sedlmayr 1996 ) , but problems remain . in table [ tab : mags ] we list the maximum light properties of the four hot rcb stars .
we present their light - curves from 1992 to 2000 in fig .
[ fig : light ] . in sec .
[ ssec : maximum_light ] , we discuss the long term behavior of the maximum brightness of rcb stars .
v348 sgr was discovered to be variable by woods ( 1926 ) and suggested to be an rcb star by parenago ( 1931 ) .
the light - curve behavior is summarized by herbig ( 1958 ) , hoffleit ( 1958 ) , heck et al .
( 1985 ) and the aavso .
there are some photometric data as far back as 1900 and more or less complete coverage from 1928 to the present .
v348 sgr is the second most active rcb star known ( after v854 cen ) .
it has an average time between declines of 560 days ( jurcsik 1996 ) . between 1954 and 1981 ,
v348 sgr was never at maximum light for a full year ( bateson 1982 ) .
hv 2671 has little photometric coverage before the macho era ( alcock et al .
it was discovered as a harvard variable in the early 1900 s with a variation of 1.8 mag .
it was seen in a deep decline in 1967 by kurochkin ( 1992 ) . in the macho data , shown in fig .
[ fig : light ] , which cover about 7 years , hv 2671 has two deep declines .
dy cen had four declines recorded between 1897 and 1927 but none since , although the photometric coverage has not been continuous ( hoffleit 1930 ; bateson 1978 ; pollacco & hill 1991 ; jurcsik 1996 ; aavso ) .
the royal astronomical society of new zealand ( rasnz ) and the aavso have been following dy cen since 1960 without any declines being detected ( bateson 1978 ) .
dy cen has a relatively high hydrogen abundance much like the cool rcb star v854 cen .
but unlike that very active star , dy cen seems to go against the trend which relates high decline activity with high hydrogen abundance ( jurcsik 1996 ) .
similarly , mv sgr was seen to decline at least twice ( 1930 and 1945 ) but hasnt had a large documented decline since then ( hoffleit 1959 ) . like dy cen
, the photometric coverage for mv sgr has not been continuous and there have been some reports of small declines ( jurcsik 1996 ; landolt 2001 priv .
mv sgr has been followed continuously by the aavso since 1987 without having any discernible declines .
however , at near - ir and iras wavelengths , mv sgr and dy cen show similar excesses to the much more active v348 sgr , indicating the presence of warm dust ( kilkenny & whittet 1984 ; walker 1985 ; rao & nandy 1986 ) .
so mv sgr and dy cen are likely to be producing dust but not along our line of sight to the star . the range of decline activity seen in the hot rcb stars mimics that seen in the cooler rcb stars ( clayton 1996 ) .
the historical light - curve of r crb itself , which now stretches back over two hundred years , shows that there are periods of up to 10 years where the star had no decline and other periods where it has had several declines in one year ( mattei , waagen , & foster 1991 ) .
rcb stars are true irregular variables , where the visual brightness of the star declines every time that dust forms near the stellar surface , in the line of sight to the observer .
the star returns to maximum light when the dust in the line of sight disperses .
therefore , the maximum light is thought to be representative of the intrinsic stellar brightness ( once circumstellar and interstellar reddening is taken into account ) . in this section
, we discuss _ secular _ changes in the the maximum - light brightness ( i.e. the stellar brightness _
after _ it recovers from a decline ) , and investigate to what they might be due : temperature or luminosity changes , or a change in the circumstellar reddening .
the maximum - light brightness evolution for three of the four hot rcb stars ( hv 2671 has little coverage before 1992 ) , is summarized in table [ tab : mags ] and fig . [
fig : light2 ] ( light - curves are _ not _ plotted in fig .
[ fig : light2 ] , only the maximum light estimates are plotted ) . three types of magnitudes
are plotted in fig .
[ fig : light2 ] , photographic magnitudes ( @xmath45 ) , measured by eye or with an iris photometer on blue plates , visual magnitudes ( @xmath46 ) , measured by eye through a telescope , and johnson @xmath13 photoelectric magnitudes . the transformation of @xmath45 to johnson @xmath47 magnitudes can be found in arp ( 1961 ) and pierce & jacoby ( 1995 ) .
for the three stars , the maximum - light colors are somewhat variable . for v348 sgr ,
@xmath48 varies between 0.32 and 0.40 mag ; in the case of dy cen , @xmath48 varies between 0.31 and 0.37 mag .
mv sgr varies between 0.22 and 0.28 mag ( landolt 2001 priv .
comm . ; pollacco & hill 1991 ) .
for all these values , the transformation from @xmath45 to johnson @xmath13 ( @xmath49 = 0.17 - 1.09@xmath41(@xmath48 ) ) is small . for the @xmath48 colors of the hot rcb stars
, @xmath49 will range from 0.1 to 0.2 mag .
the uncertainties in individual @xmath45 values can be seen in the scatter in fig .
[ fig : light2 ] .
also , schaeffer ( 1994 ) collected a large number of @xmath45 magnitudes by different observers for sn 1937c including measurements by hoffleit from similar plates .
these indicate that @xmath45 measurements have an uncertainty of @xmath150.2 - 0.3 mag
. there is no discernible difference between the johnson @xmath13 and @xmath46 measurements in the data plotted in fig .
[ fig : light2 ] .
a comparison of @xmath13 and @xmath46 resulted in the following transformation , @xmath50 = 0.21@xmath41(@xmath48 ) ( stanton 1999 ) .
so for @xmath48 = 0.3 mag , @xmath50 = 0.06 mag .
i.e. the correction is small .
the photographic and visual magnitudes have been transformed to v magnitudes in fig .
[ fig : light2 ] . in the first half of the 1900 s ,
v348 sgr would reach a maximum brightness of @xmath5111 mag ( hoffleit 1958 ) , but more recently its maximum brightness is v@xmath112 mag ( heck et al .
the maximum - light brightness of v348 sgr therefore faded @xmath11.0 mag from 1930 to 1957 .
it seems to have been fading more slowly since 1957 .
its maximum - light brightness also seems to have increased 0.4 mag from 1921 to 1930 , which differ by less than the admitted uncertainty .
similarly , dy cen is reported to be gradually fading ( bateson 1978 ; pollacco & hill 1991 ; rao et al . 1993 ) .
it can be seen from fig .
[ fig : light2 ] that the maximum - light brightness of dy cen faded 0.7 mag from 1975 to 1993 , remaining constant since then .
mv sgr has today a maximum - light brightness of v@xmath52 13 mag while in hoffleit s data it is @xmath5312 mag .
the maximum - light brightness of mv sgr therefore faded @xmath11.0 mag from 1941 to 1980 , but the lack of photometric coverage does not allow the beginning of the fade to be accurately determined , although the fading was well underway by 1963 .
the maximum - light brightness has been fairly constant since then .
hv 2671 has by far the least photometric coverage of the four stars . based on the published data , median m@xmath54 = 16.4 mag pre-1940 , and b = 15.5 mag in 1967 , little can be said about any possible evolution of its maximum brightness .
the fact that three out of the four known hot rcb stars seem to be fading slowly with time may indicate that stars in the hot rcb class are evolving rapidly .
the change might in fact indicate either an intrinsic fading of the star ( due to an overall fading of the bolometric luminosity or to a change of temperature ) or an increase in the circumstellar reddening .
taking first the case where the change in v brightness is due to a change in temperature , we note that for constant @xmath55 , @xmath561 , corresponds to a decrease of the bc by the same amount .
such a change would imply an increase by about 6000 k in the 10 - 20kk range .
( we use the bolometric corrections for normal supergiant stars [ allen , astrophysical quantities , 4@xmath57 edition ] .
although this is generally not appropriate for rcb stars , the fact that we are using relative values , and that we are not extracting precise quantities , make this method suitable .
additionally the bc of leuenhagen & hamann ( 1994 ) of 1.71 mag determined for the 20kk v348 sgr , is similar to the values listed in astrophysical quantities for a 17.6kk b2 supergiant of 1.58 mag ) .
an increase in the effective temperature between 11.1 and 17.6 kk would be accompanied by a change in the star
s @xmath48 by only @xmath10.14 mag .
if the change in @xmath13 by @xmath11 mag were due to reddening , @xmath58a@xmath59 = 1 translates to a @xmath58(@xmath48 ) @xmath1 0.32 mag .
therefore the color can help discriminate between a temperature and a reddening change .
the data are fragmentary but they show no color trend : dy cen was observed to have @xmath10 = 12.39 mag and @xmath48 = 0.31 mag in 1972 and @xmath10 = 12.78 mag and @xmath48 = 0.33 mag in 1987 ( rao et al . 1993 ) .
so for dy cen , the fading is inconsistent with a reddening increase , while it is more consistent with a temperature increase by @xmath16000 k. the maximum - light brightness and color of mv sgr was observed to be @xmath13 = 12.70 mag and @xmath48 = 0.26 mag in 1963 ( herbig 1964 ) and @xmath13 = 13.29 mag , @xmath48 = 0.262 mag in 2000 ( average of 8 observations , landolt 2001 priv .
its maximum - light brightness faded by @xmath60 = 0.6 mag but its @xmath48 color remained unchanged . if the fading was due to added circumstellar reddening then we would expect @xmath48 to become redder by @xmath10.2 mag .
so like dy cen , the mv sgr fading is inconsistent with a reddening increase .
v348 sgr is much more active than the other two stars and could therefore be a better candidate for a reddening increase , rather than a change in temperature .
however we would still like to argue that v348 sgr , too , has suffered an increase in temperature , rather than a change in reddening .
the evidence is circumstantial for this star , but it is strongly suggestive . in fig .
[ fig : light2 ] we compare the long term light - curves for the three hot rcb stars with those of the very active _ cool _ rcb stars r crb and ry sgr , for which we have light - curves going back over 150 years ( unfortunately there are no other rcb stars with such ample light - curve coverage ) .
the two cool rcb stars show no maximum light fading trend .
finally , we should point out that the @xmath13 brightness data is also consistent with a decrease in the bolometric luminosity .
a @xmath61 = 1.0 mag corresponds to a decrease in luminosity by a factor of 2.5 .
this can not be excluded , although this type of luminosity changes would be hard to interpret in the light of single star post - agb evolution .
here we check whether the variability in luminosity is accompanied by any type of spectral variability over the period of time for which archival data were available .
we stress that the spectral variability we are discussing in this section is not the spectral variability observed during light declines , but rather a change in the maximum - light spectrum between two different epochs . of the four hot rcb stars , two ,
v348 sgr and mv sgr have spectral data going back to the 1940s ( tables [ tab : spectra1 ] and [ tab : spectra2 ] ) .
we obtained and scanned some of the plates ( ticked in tables [ tab : spectra1 ] and [ tab : spectra2 ] ) , to search for any signs of variability .
we note that on the plate scans only significant spectral changes can be detected . a spectrum s continuum level and shape are not indicative of the true stellar continuum , so that only the lines can be considered .
we can therefore detect the presence or absence of emission lines , or a complete change of the line makeup of the spectrum .
changes in temperature , abundance or in wind strength inducing even a 50% change in our lines equivalent widths might be missed .
the 1949 , 1950 , 1954 and 1957 spectra of v348 sgr ( range @xmath13500 - 5500 ) were compared to the modern digital spectra , degraded to simulate the resolution and quality achieved by the scans of the historical spectra .
we find no significant sign of change .
however , the red spectrum ( @xmath15500 - 7000 ) from 1953 exhibits a deep absorption line at about 5850 .
this corresponds exactly with a strong emission in the degraded modern spectrum .
the emission is a blend of ne i m6 , and the red - most component of c ii m22 .
the rest of the features correspond rather well , so that a wavelength error is excluded .
it is unclear why only this feature should exhibit change .
overall we can say that no significant changes have occurred to v348 sgr between 1949 and 1995 .
the 1958 spectrum of mv sgr is , once scanned , completely useless .
however we can compare the 1963 spectrum ( in the range 4500 - 7000 ) with the 1973 spectrum and the degraded 1995 spectrum . from this comparison
it can be said that there is no clear sign of spectral variability between 1963 and 1995 .
we stress that it is impossible to compare line strengths and line ratios , but that all the lines seem to be there .
finally we should point out that the stellar temperature change suggested in sec .
[ ssec : maximum_light ] , might not produce noticeable changes in the scans of early spectra . despite the lack of obvious spectral variability over the last half century
, we found that the high resolution , high signal - to - noise digital spectra of v348 sgr and dy cen , show evidence of variability over the last decade .
we compared two spectra of v348 sgr , the lhj94 spectrum taken in 1987 and the calar alto spectrum taken in 1995 .
the different resolutions are accounted for by re - binning the higher resolution spectrum ( r = 55,000 ) to the lower resolution one ( r@xmath15000 ) . in fig .
[ fig : v348_var ] we compare the two rectified spectra in the range 58405950 .
the emission lines have increased in strength between 1987 and 1995 and there is an indication that the p - cygni profiles have shifted to bluer wavelengths ( e.g. , he i @xmath75876 ) . at this resolution
it is unclear whether the line splits have changed between the two epochs .
if confirmed by a third epoch of observations , these changes could be interpreted as an increase in wind density and velocity .
we compared the two maximum - light spectra of hv 2671 after degrading the slightly higher resolution of the 1995 spectrum to match the 1997 spectrum .
the 1995 spectrum was taken in non - photometric conditions and its overall flux level had to be multiplied by a factor of 20 to match the continuum level of the 1997 spectrum . within the limitation of the resolution as well as the low signal - to - noise ratio of the 1995 spectrum , we can say that no significant variation occurred in the intervening years , although the details of some of the emission line profiles are different . with the current very limited data - set
it is not clear whether this is a real effect .
dy cen was observed in 1989 and 1992 by giridhar et al .
( 1996 ) and in 2001 ( april and september ) by us .
unfortunately the two data - sets do not have an overlapping spectral range .
a comparison of the 1989 and 1992 spectra led giridhar et al .
( 1996 ) to conclude that 1 ) the mean radial velocity of the absorption lines had changed between 42 and 29 km s@xmath6 ; 2 ) some c ii lines ( e.g. @[email protected],5537.6 [ m10 ] , @[email protected],6098.5,6102.6 [ m24 ] ) had blue - shifted absorption in 1989 , and red - shifted absorption in 1992 ; 3 ) h@xmath8 acquired a blue - shifted absorption between 1989 and 1992 ; and 4 ) he i @xmath75876 acquired a blue - shifted absorption and a split emission line in 1992 , while in 1989 it had only a split absorption .
we have re - examined these spectra and confirm the behavior observed by giridhar et al .
( 1996 ) . in fig .
[ fig : v348_dycen_comp ] we show four lines which changed .
the 1992 spectrum of dy cen shows a stronger h@xmath8 with a weak p - cygni absorption , not unlike the spectrum of v348 sgr , while the 1989 spectrum shows only a slightly asymmetric emission .
the he i line at 5876 shows the emergence of a blue - shifted absorption , while the core of the line is rising into a split emission ( much like the profile observed in the spectrum of v348 sgr , plotted alongside for comparison , thin solid line ) .
last , the c ii m2 doublet , which is in pure absorption in the 1989 spectrum ( note that the emission peak is the nebular [ n ii ] line at 6583 ) , is also acquiring blue - shifted absorption in 1992 , while the absorption troughs are partly filled . in fig .
[ fig : ha_dycen ] we show h@xmath8 and the two c ii components of the m2 doublet , in their respective radial - velocity - corrected velocity space . here
, we can see how the newly formed p - cygni profiles match for the three lines , while the still weak emission components appear to have a split appearance . in the 2001 spectra ,
the c ii line @xmath74267 is a weak but very broad ( 590 km s@xmath6 ; fig .
[ fig:2001obs ] ) emission .
although this spectral range can not be compared directly to the red spectrum of giridhar et al .
( 1996 ) , the presence of a broad emission line such as the one observed is _ per se _ an indication of wind activity .
the line variability observed ( the disappearance of the narrow emission component in the september spectrum [ fig .
[ fig:2001obs ] ] ) , might be a short term variability superimposed on a more general behavior . once again , a third epoch of observations , overlapping in spectral range with early spectra , needs to be taken to determine whether the wind of dy cen is truly strengthening . mv sgr was observed on several occasions ( table [ tab : spectra1 ] and [ tab : spectra2 ] ) .
no variations were seen .
the maximum - light spectra all look very similar . in conclusion , two out of the four known hot rcb stars
might show signs of increased wind density , although more observations are needed to confirm this trend .
moreover , we must beware of the fact that additional , short term variability might be present ( like in the case of dy cen ) . if these changes were attributed to increased wind mass - loss , we could propose that the _ luminosity _ has increased since it is by far the largest cause of changes in the wind efficiency . on the other hand
, this would be very surprising in view of the post - agb nature of these stars .
we might therefore have to invoke an increasing temperature , and opacity ( i.e. , a higher metal - to - hydrogen ratio , brought about by mass - loss ) . without a third epoch of observations
it is speculative to make any further conclusions .
before attempting an interpretation of the evolutionary status of the hot rcb stars we summarize the salient points that characterize the hot rcb stars .
a summary is also presented in tabular form in table [ tab : summary ] . 1 .
v348 sgr and hv 2671 have almost identical spectra at low resolution , exhibiting only emission lines .
dy cen has mostly absorption lines in 1989 while in 1992 it has developed a few more emission lines .
mv sgr has strong emission lines that belong mostly to heavy elements , with only weak carbon and nitrogen emission and absorption . solely on the basis of the spectrum one would group v348 sgr and hv 2671 together , leaving dy cen and mv sgr in two separate categories .
2 . the stellar temperatures are in the range 15 - 20 kk .
this , and the light variability , are the only truly common characteristics of the hot rcb stars .
the abundances span a wide range .
v348 sgr , and presumably hv 2671 , are very carbon - rich , grouping closely with the members of the [ wc ] central stars of pn ( e.g. leuenhagen et al .
1996 ; de marco & crowther 1998,1999 ; de marco et al . 2002 ) , dy cen and mv sgr are composed mainly of helium , more like normal rcb stars ( asplund et al .
the emission lines of v348 sgr have increased in intensity between 1987 and 1996 , and the p - cygni profiles have become deeper and bluer .
dy cen has suffered a similar fate between 1989 and 1992 where some of its absorption lines have turned into emission and developed a weak p - cygni absorption . on the other hand mv sgr
does not appear to have changed .
for hv 2671 we do not have enough data to draw any conclusions .
this behavior , seen in v348 sgr and dy cen , could be consistent with strengthening of the wind and hence an increase in mass - loss , but a third epoch with overlapping spectral range is necessary in order to validate this conclusion .
the absorption lines of dy cen have changed in position between 1989 and 1992 .
this behavior is highly suggestive of binarity ( as first suggested by giridhar et al .
4 . v348 sgr and dy cen ( in 1992 ) exhibit split emission lines of he i and c ii ( though h@xmath8 , were it not contaminated by nebular emission , would also show the same split ) .
some of these emission lines also exhibit p - cygni absorption .
mv sgr also exhibits split emission lines , although mostly of heavy and s - process elements .
split lines are consistent with an equatorial density enhancement like those encountered in b[e ] stars .
5 . three hot rcb stars exhibit forbidden emission lines . for the fourth ,
hv 2671 , we have insufficient data .
the forbidden lines in v348 sgr and dy cen point to low electron densities ( @xmath1400 @xmath33 ) , and relatively low expansion velocities ( 10 - 20 km s@xmath6 ) , while the forbidden lines in the spectrum of mv sgr point to high velocities ( @xmath1100 km s@xmath6 ) and high densities ( @xmath110@xmath62 @xmath33 ) .
additionally , v348 sgr has a visible nebulosity about @xmath130 arcsec ( pollacco et al .
1990 ) in diameter . no resolved nebulosity exists for the other stars .
ionized nebulae around relatively cool stars might be consistent with the possibility that these stars might have been hotter in the past .
the maximum - light @xmath13 brightness of three hot rcb has declined in the last 70 - 100 years .
we favor a temperature increase of @xmath15000 k , to interpret this fading . for dy cen and mv sgr ,
this is confirmed by the lack of a color change , which would be expected in case of a reddening increase .
no color data exist for v348 sgr .
all four stars have been classified as rcb stars on the basis of their large , random declines in brightness . at present , v348 sgr and hv 2671 are very active , while mv sgr and dy cen have had no declines in 50 years and are among the least active rcb stars known .
ir excess data indicates , however , that new dust is present around all four stars .
in this section , we compare the properties of the four hot rcb stars , summarized in sec.[sec : summary ] , with the latest evolutionary models ( herwig et al . 1997 ; herwig 2000,2001 ; bloecker 2001 ) .
we will exclude hv 2671 from this discussion in view of the scant information we have about this star .
however it is likely that this object groups together with v348 sgr .
* v348 sgr . * clearly , v348 sgr distinguishes itself from the other two objects because of its high carbon abundance . because of its prominent emission line spectrum , it has been also been classified as a [ wc12 ] wolf - rayet central star of pn .
however , it does not fit seamlessly into the late type wolf - rayet star class ( see the properties listed by koesterke [ 2001 ] ) .
v348 sgr has a large , extended pn in contrast to most other stars of this group which have rather young ( compact and high density ) pne ( e.g cpd-56@xmath48032 and he2 - 113 [ de marco et al .
1997 ] , m4 - 18 [ de marco & crowther 1998 ] ) , although it is not dissimilar from the large tenuous pn of the [ wc11 ] star k2 - 16 .
no doubt , the [ wc ] class remains enigmatic , but the very fact that it is an heterogeneous one ( possibly comprising several undiscovered born - again stars ) , means that it can accommodate v348 sgr among its members .
thus , from the properties of the pn , v348 sgr might in fact be the only late - type [ wc ] star which owes its h - deficiency to a born - again evolutionary history .
however , v348 sgr displays a suspiciously small oxygen abundance .
if this star had initially formed from a sun - like composition , then its initial oxygen abundance would have been between 0.5 and 1@xmath63 by mass .
herwig ( 2000 ) presented agb models which include additional mixing processes resulting in abundances consistent with the oxygen mass fractions observed in the majority of the [ wc]-like and pg1159 stars .
basically , additional dredge - up induced by convective overshoot leads to a higher oxygen abundance in the agb intershell .
in the course of the mixing ( and possibly burning processes ) associated with the born - again evolution , the intershell material , which contains significant amounts of carbon , come to the surface .
if this mechanism is responsible for the hydrogen - deficiency and the huge carbon abundance of v348 sgr ( @xmath64 ) , then we would expect a significant amount of oxygen , much higher then the observed upper limit of 1% .
[ wc]-type ( early and late ) and pg1159 stars have oxygen abundances in the range 5 - 15% , consisten with evolutionary models .
very low oxygen abundances are however not unheard of .
one other [ wc]-late type star ( iras21282 + 5050 ) and one pg1159 star ( ngc7094 ) have similarly low oxygen abundances .
therefore , although the old nebula and the h - deficiency support a born - again evolutionary origin for v348 sgr , the very low oxygen abundance casts serious doubts on this interpretation .
unfortunately , the alternative , a wd - merger scenario , does not help us either . while it would be possible to accommodate a very low oxygen abundance ,
this scenario could not account for the large carbon abundance .
* this star has characteristics in common with both v348 sgr and mv sgr .
the nebular properties are similar to those of v348 sgr ( a low pn expansion velocity and a low pn electron density , generally compatible with a born - again evolution ) .
in addition this star shares with v348 sgr signs of increasing wind activity .
however , the stellar chemistry is - as is the case in mv sgr - dominated by helium like the cool rcb stars
. _ helium predominance is irreproducible by any current single star agb or post - agb evolutionary calculations_. another difficulty is the high carbon abundance ( 15 times solar ) . even if it were possible to uncover the helium - rich layer left over from h - shell burning , evolutionary calculations predict nitrogen , not carbon , as the dominant element after helium .
in addition , although the understanding of the internal mixing processes has improved , uncertainties remain .
dy cen is clearly the star in the hot rcb class , which is most likely to be a progeny of the majority group of cool rcb stars .
* due to the predominance of helium in the stellar atmosphere of mv sgr , we might conclude that this star is too a typical descendant of the cool rcb class .
it shares with the other two stars the decreasing visual luminosity , which we interpret as increasing stellar temperature , indicating a leftward motion in the hr diagram , consistent with a post - agb star increasing in temperature .
however , the cno elements are dominated by nitrogen ( after helium ) as expected for the result of hydrogen burning .
the interpretation that we do , in fact , see the helium - layer , resulting from hydrogen - shell burning , is consistent with the absence of any hydrogen .
however , it is unclear how hydrogen was eliminated to uncover the helium layer : the absence of hydrogen can not result from hydrogen - ingestion during a very late thermal pulse ( vltp ) : during a vltp , hydrogen is ingested into the @xmath65c - rich layers of the intershell region .
there the @xmath65c abundance is such that proton - capture transformation of @xmath65c into @xmath66c dominates , and only a little @xmath67n , if any , can form .
mv sgr has a high expansion velocity , high density , hydrogen - rich nebula , inconsistent with the born - again scenario unless deeper imaging detect a low - density extended pn around it , and/or the central nebulosity proves to be hydrogen - deficient .
this star requires an evolutionary scenario capable of removing the remaining envelope completely , but not through dredge - up , which would leave some trace of hydrogen in the photosphere , or through hydrogen - ingestion mechanism associated with the vltp .
current stellar evolutionary models can not explain this star .
the stellar abundances seems to point to at least three classes for the four hot rcb stars ! yet , other characteristics , though never entirely homogeneous , seem to draw these stars together .
one question is therefore raised , whether a post - agb star , with , say , rcb - type abundances can change significantly its photospheric abundances to become , say , a [ wc]-like type star . such change , happening not during the agb , but in the post - agb phase , is not in agreement with evolutionary calculations .
however , in view of the strangeness of these stellar classes , it can not be excluded a priori .
the suggestion that the spectrum of dy cen is developing stronger emission lines , and becoming similar to that of the chemically very different v348 sgr , make this possibility appealing .
further spectroscopy , to confirm a long - term change which is not either periodic or stochastic shorter term variability , is therefore eagerly sought .
if the hot rcb stars turn out to be the progeny of the cooler ones , we might also wonder about the possibility that the hot rcb stars constitute a link between the cool rcb stars and the chemically similar extreme helium stars .
this was extensively discussed by pandey et al .
they conclude that it is not straightforward to establish a link between the two classes , on the basis of different heavy element abundances . again
, if we could open the possibility of significant abundance changes , there might be a new open evolutionary channel _ out _ of the rcb class .
o(he ) stars , which are hot h - deficient post agb stars ( four are known , of which two are associated with a pn ; werner 2001 ) have been proposed as descendants of the rcb stars .
however they might be too hydrogen - rich ( lotr4 has 33% hydrogen by mass ; rauch et al .
1996,1998 ) to make definitive conclusions .
last , we have shown how all the hot rcb stars for which we have high resolution data , show split lines .
we have interpreted these lines as an equatorial bulge , similar to what is interpreted to be around b[e ] stars , and _ not _ a keplerian disk , which might have pointed to binarity .
the origin of this bulge is not understood , but it might be a key in the explanation of the dust - making activity of these stars , since it might present a denser , more shielded environment where dust could condense . understanding the spectral variability properties of the split lines is paramount in characterizing the evolution of the equatorial enhancement , as well as how it relates to the pulsation variability observed in other rcb stars .
regarding the possibility that rcb stars derive from the merger of a helium and a co wd , we note that dy cen is likely to be a binary .
although further observations are sought , this fact alone would contradict a merger scenario , which is more likely to result in a single star ( jefferey & saio 2002 ) .
we thank simon jeffery for providing the high resolution spectrum of v348 sgr .
wolf - rainer hamann is thanked for providing the unpublished calar alto spectrum of v348 sgr , while kameswara rao , sunetra giridhar are thanked for the unpublished mv sgr spectrum and for both the dy cen spectra used in this analysis .
george jacoby is thanked for helpful conversations . thanks to tony misch and george herbig for help in obtaining and scanning the plate spectra .
thanks to janet mattei and the aavso for the photometric data .
od gratefully acknowledges financial support from the asimov fellowship program .
fh would like to thank d.a .
vandenberg for support through his operating grant from the natural science and engineering research council of canada .
alcock , c. et al .
1996 , apj , * 470 * , 583 alcock , c. et al .
2001 , apj , 554 , 298 arp , h. 1961 , apj , 133 , 869 asplund , m. , lambert , d. l. , kipper , t. , pollacco , d. , & shetrone , m. d. 1999 , a&a , 343 , 507 asplund , m. , gustafsson , b. , lambert , d.l . , & rao , n.k .
2000 , a&a , 353 , 287 bateson , f.m . 1978 ,
rasnz pvss , 6 , 39 bateson , f.m .
1982 , rasnz pvss , 10 , 1 benetti , s. , duerbeck , h.w .
, seitter , w.c .
, harrison , t. , & hoff , w. 1996 , iauc no .
6325 bjorkman , j. e. , 1998 , in _ dusty b[e ] stars _ , eds a.m. hubert & c. jaschek , p.189 clayton , g.c .
1996 , pasp , * 108 * , 225 clayton , g.c . , bjorkman , k.s . ,
nordsieck , k.h . ,
zellner , n.e.b .
, sculte - ladbeck , r.e .
1997 , apj , 476 , 870 clayton , g.c . , & de marco , o. 1997 , aj , 114 , 2679 clayton , g.c . , kerber , f. , gordon , k.d . ,
lawson , w.a , wolff , m.j .
, pollacco , d.l . , & furlan , e. 1999 , apj , 517 , l143 clayton , g.c . , & ayres , t.r .
2001 , apj , 560 , 986 crowther , p.a . , de marco , o. , & barlow , m.j .
1998 , mnras , 296 , 367 dachs , j. , hummel , w. , & hanuschik , r. w. 1992 , a&as , 95 , 437 dahari , o. , & osterbrock , d. e. 1984 .
apj , 277 , 648 de marco , o. , barlow , m.j . , & storey , p.j . 1997 , mnras , 292 , 86 de marco , o. , & crowther , p.a . 1998 , mnras , 296 , 419 de marco , o. , & crowther , p.a . 1999 , mnras , 306 , 931 drilling , j.s . , schnberner , d. , heber , u. , & lynas - gray , a.e . 1984 , apj , 278 , 224 drilling , j. s. , hecht , j. h. , clayton , g. c. , mattei , j. a. , landolt , a. u. , & whitney , b. a. 1997 , apj , 476 , 865 drilling , j.s . , & hill , p.w . 1986 , in _ hydrogen deficient stars & related objects _ , eds . k. hunger , d. schnberner , n.k .
rao ( dordrecht , reidel ) , p. 499
drilling , j.s . , & schnberner , d. 1989 , , 343 , l45 duerbeck , h.w . , & benetti , s. 1996 , apj , 468 , l111 evans , a. , whittet , d.c.b . ,
davies , j.k . , kilkenny , d. , & bode , m.f .
1985 , , 217 , 767 feast , m.w .
1999 , in _ new views of the magellanic clouds _ , iau symp .
190 , eds .
chu , n.b .
suntzeff , j.e .
hesser , and d.a .
bohlender , p. 542 fich , m. , blitz , l , & stark , a.a .
1989 , apj , 342 , 272 gillett , f. c. , backman , d. e. , beichman , c. , & neugebauer , g. 1986 , apj , 310 , 842 giridhar , s. , rao , n.k . , & lambert , d.l .
1996 , a.s.p .
ser . , 96 , 171 glass , i.s .
1978 , mnras , 185 , 23 goldsmith , m.j . ,
evans , a. , albinson , j.s . , & bode , m.f .
1990 , mnras , 245 , 119 gonzalez , g. , lambert , d.l . ,
wallerstein , g. , rao , n.k . , smith , v.v . , & mccarthy , j.k
. 1998 , apjs , 114 , 133 grevesse , n . , noels , a. , & sauval , a.j .
1996 , asp conf .
ser . , 99 , 117 hamann , w. r. , & koesterke , l. 1993 , in planetary nebulae , iau symp .
155 , eds .
r. weinberger and a. acker , ( kluwer : dordrecht ) , p. 87
harrison , t. e. 1996 , pasp , 108 , 1112 heber , u. , heck , a. , houziaux , l. , manfroid , j. , & schnberner , d. 1984 , proc .
4@xmath57 european iue conf . , esa sp-218 , p. 367
heber , u. , & schnberner , d. 1981 , a&a , 102 , 73 hecht , j.h . ,
clayton , g.c . ,
drilling , j.s . , and
jeffery , c.s .
1998 , apj , 501 , 813 heck , a. , houziaux , l. , cassatella , a. , di serego alighieri , s. , & macchetto , f. 1982 , proc .
3@xmath68 rd european iue conf .
heck , a. , houziaux , l. , manfroid , j. , jones , d.h.p .
, & andrews , p.j . 1985 , astron .
suppl . , 61 , 375 herbig , g.h .
1949 , apj , 110 , 143 herbig , g. h. 1958 , apj , 127 , 312 herbig , g. h. 1964 , apj , 140 , 1317 herbig , g. h. 1975 , apj , 199 , 702 herwig , f. 1999 , a&a , 349 , l5 herwig , f. 2000 , a&a , 360 , 952 herwig , f. 2001 , apj , 554 , l71 hoffleit , d. 1930 , harvard bull . , 874 , 1 hoffleit , d. 1958 , aj , 63 , 78 hoffleit , d. 1959 , aj , 64 , 241 houziaux , l. 1968 , bull .
, 19 , 265 houziaux , l. , bouchet , p. , heck , a. , & manfroid , j. 1987 , qjras , 28 , 231 hummel , w. 1994 , a&a , 289 , 458 iben , i. , kaler , j. b. , truran , j. w. , & renzini , a.1983 , apj , 264 , 605 iben , i 1995 , phys .
reports , 250 , 2 iben , i. jr . , tutukov , a. v. , & yungelson , l. r. 1996 , apj , 456 , 750 jacoby , g.h . &
ford , h.c .
1983 , apj , 266 , 298 jeffery , c.s . , heber , u. , hill , p.w . , & pollacco , d. 1988 , mnras , 231 , 175 jeffery , c.s . , &
heber , u. 1993 , a&a , 270 , 167 jeffery , c.s .
1995 , a&a , 299 , 135 jones , a. , lawson , w. , de marco , o. , kilkenny , d. , van wyk , f. , roberts , g. 1999 , observatory , 119 , 76 jurcsik , j 1996 , acta astr .
, 46 , 325 kerber , koeppen , j. , roth , m. , & trager , s. c. 1999a , a&a , 344 , l79 .
kerber , f. blommaert , j. a. d. l. , groenewegen , m. a. t. , kimeswenger , s. , kaufl , h. u. , & asplund , m. 1999b , a&a , 350 , l27 .
kilkenny , d. , and whittet , d.c.b .
1984 , mnras , 208 , 25 kilkenny , d. , coulson , i. m. , laing , j. d. , jones , j. s. , & engelbrecht , c. 1985 , saao circ .
, 9 , 87 kurochkin , n.e .
1992 , sov .
lett . , 18 , 410 lamers , h.j.g.l.m . , zickgraff , f .- j . , de winter , d. , houziaux , l. , zorec , j. 1998 , a&a , 340 , 117 landolt , a.r
1979 , iauc no .
3419 leuenhagen , u. , heber , u. , & jeffery , c.s .
1994 , a&as , 103 , 445 ( lhj94 ) leuenhagen , u. & hamann , w. r. 1994 , a&a , 283 , 567 leuenhagen , u. , hamann , w.r . ,
& jeffery , c.s .
1996 , a&a , 312 , 167 leuenhagen , u. & hamann , w. r. 1998 , a&a , 330 , 265 lundmark , k. 1921 , pasp , 33 , 314 mattei , j. a. , waagen , e. o. , & foster , g. 1991 , aavso monograph no .
4 nakano , s. , et al .
1996 , iauc no .
6322 neckel , th . , & klare , g. 1980 , a&as , 42 , 251 okazaki , a.t .
1997 , a&a , 318 , 548 okazaki , a.t .
2001 , pasj , 53 , 119 oudmaijer , r.d .
, proga , d. , drew , j.e . , de winter , d. 1998 , mnras , 300 , 1700 pandey , g. , rao , n.k . , & lambert , d.l .
1996 , mnras , 282 , 889 pandey , g. , rao , n.k . , lambert , d.l . ,
jeffery , c.s . , & asplund , m. , 1996 , mnras , 324 , 937 parenago , p.p .
1931 , variable stars , ussr , 3 , 113 pierce , m.j . , & jacoby , g.h .
1995 , aj , 110 , 2885 pollacco , d. l. , & hill , p. w. 1991 , mnras , 248 , 572 pollacco , d.l . , hill , p.w . ,
houziaux , l. , and manfroid , j. 1991 , mnras , 248 , 1 pollacco , d. l. , kilkenny , d. , marang , f. , van wyk , f. , & roberts , g. 1992 , mnras , 256 , 669 .
pollacco , d. l. , tadhunter , c. n. , and hill , p. w. 1990 , mnras , 245 , 204 porter , j.m .
1999 , a&a , 348 , 512 pottasch , s.r . ,
1984 , planetary nebulae , ( reidel , dordrecht ) rao , n. k. , and lambert , d. l. 1993 , aj , 105 , 1915 rao , n. k. , et al . 1999 , mnras , 310 , 717 rao , n. k. , & nandy , k. 1986 , mnras , 222 , 357 rao , n. k. , giridhar , s. , & nandy , k. 1990 , a&a , 234 , 410 rao , n.k .
, giridhar , s. , & lambert , d.l .
1993 , a&a , 280 , 201 rauch , t. , kppern , i. , & werner , k. , 1996 , a&a , 310 , 6131 rauch , t. , dreizler , s. , wolff , b. , 1998 , a&a , 338 , 651 saio , h. , & jeffery , c.s . , 2000 ,
mnras , 313 , 671 saio , h. , & jeffery , c.s . , 2002 , mnras , in press schaeffer , b.e . 1994 , apj , 426 , 493 seitter , w. c. 1985 , mit .
a.g . , 63 , 181 sherwood , v.e
1975 , in _ hydrogen deficient stars & related objects _ , eds . k. hunger , d. schnberner , n.k .
rao ( dordrecht , reidel ) , p. 147
stanton , r.h .
1999 , jaavso , 27 , 97 van der hucht , k. , conti , p.s . ,
lundstrom , i. , stenholm , b. 1981 , space sci .
, 28 , 277 vassiliadis , e. , & wood , p.r .
1994 , apjs , 92 , 125 venn , k.a . ,
smartt , s.j . ,
lennon , d.j .
, & dufton , p.l .
1998 , a&a , 334 , 987 voors , r.h.m .
2000 , phd thesis walker , h. j. 1985 , a&a , 152 , 58 werner , k. 2001 , apss , 275 , 27 woitke , p. , goeres , a. , & sedlmayr , e. 1996 , a&a , 313 , 217 woods , i.e. 1926 , bull .
harvard coll .
woods , i.e. 1928 , bull .
harvard coll .
855 , 22 zorec , j. 1998 , in _ dusty b[e ] stars _ , eds a.m. hubert & c. jaschek , p. 27
lllllll & date & telescope&wavelength & resolution@xmath69 & source & + v348 sgr@xmath70 & & & & & & + x&1949 july - august & lick 36 `` refractor & 4000 - 5000 & 130 mm@xmath6&herbig 1958 & + x&1950 april & lick 36 '' refractor & 4000 - 5000 & 130 mm@xmath6&herbig 1958 & + & 1951 july & lick crossley 36 `` & 3500 - 5000 & 430 mm@xmath6&herbig 1958 & + x&1953 may & lick crossley 36 '' & 3500 - 5000 & 430 mm@xmath6&herbig 1958 & + & 1953 may & lick 36 `` refractor & 4000 - 5000 & 130 mm@xmath6&herbig 1958 & + & 1953 august & lick 36 '' refractor & red & 200 mm@xmath6&herbig 1958 & + x&1954 june & lick crossley 36 `` & 3500 - 5000 & 430 mm@xmath6&herbig 1958 & + & 1955 july & lick crossley 36 '' & 3500 - 5000 & 430 mm@xmath6&herbig 1958 & + & 1956 june & lick crossley 36 `` & 3500 - 5000 & 430 mm@xmath6&herbig 1958 & + x&1957 may - june & lick 36 '' refractor & 4000 - 5000 & 130 mm@xmath6&herbig 1958 & + & 1958 june & palomar 200 & 3800 - 4800 & 18 mm@xmath6&houziaux 1968 & + & & & 5000 - 6800 & 27 mm@xmath6&houziaux 1968 & + & 1981 july - november&lick 120 `` & 3500 - 8500 & 550 & dahari & osterbrock 1984 & + & 1982 april & ctio 4 m & 5000 - 6700 & 25 mm@xmath6 & dahari & osterbrock 1984 & + & 1982 may - june & lick 120 '' & 3500 - 7100 & 550 , 1100 & dahari & osterbrock 1984 & + x&1987 april & eso 3.6 m & 3900 - 4800 & 55000 & lhj94 & + & & & & & jeffery 1995 & + & 1987 july & aat 3.9 m & 3000 - 7400 & 1100 , 3000 & pollacco et al .
1990 & + x&1995 july & calar alto 3.5 m & 3780 - 6200 & 5000 & hamann priv .
+ x&1998 may & int 2.5 m & 3500 - 6800 & 1000 & this paper & + hv 2671@xmath70 & & & & & & + x & 1995 november&saao 74``&3500 - 5200&1400&alcock et al .
1996 & + x & 1997 december&saao 74''&3500 - 7500&1100&this paper & + cccccccc star & @xmath13 & @xmath14 & distance&@xmath2 & helio .
@xmath71 & declines + & ( mag ) & ( mag ) & ( kpc ) & ( mag ) & ( km s@xmath6 ) & + v348 sgr & 12.0@xmath69 & 0.45@xmath72 & & & 130@xmath155 & very active + hv 2671 & 16.1@xmath70 & 0.15 & 50@xmath73 & 3.0 & 259@xmath1531@xmath70 & active + dy cen & 13.0@xmath69 & 0.5@xmath74 & & & 22@xmath152@xmath75 & inactive + mv sgr & 13.4@xmath69 & 0.43 & & & 95@xmath155 & inactive + lll method & @xmath14 & reference + ubv&0.9&houziaux 1968 + nearby stars&0.4&neckel & klare 1980 + ubv&0.59&heck et al . 1982
+ balmer line decrement&1.4 - 1.5&dahari & osterbrock 1984 + c ii cog&0.9 &
+ 2175 bump&0.44&heber et al .
1984 + ubv&0.66 & + balmer line decrement&[email protected]&houziaux et al .
1987 + balmer line decrement&[email protected]&pollacco et al .
1990 + stellar continuum&0.5&leuenhagen & hamann 1994 + stellar continuum&[email protected]&jeffery 1995 + ccccccccccc parameter & hv 2671 & v348 sgr@xmath69 & dy cen@xmath73 & mv sgr & rcb@xmath76 & [ wc]@xmath77 & sun@xmath78 + @xmath79 ( kpc ) & 50.0 & 5.4 & & & & & + @xmath14 ( mag ) & 0.15 & 0.5 & 0.5 & 0.38@xmath80 & & & + @xmath10 ( mag ) & 16.1 & 12.37 & 12.7 & 12.0@xmath80 & & & + @xmath2 ( mag ) & 3.0 & 2.84 & & & & & + @xmath40 ( l@xmath3 ) & 6025 & 5130 & & & & & + @xmath81 ( mag ) & 1.71 & 1.71 & & & & & + @xmath20 ( kk ) & 20.0 & 20.0 & 19.5 & 16.0@xmath80 & & & + @xmath82 ( r@xmath3 ) & 3.7 & 6.0 & & & & & + @xmath83 ( m@xmath3 yr@xmath6 ) & & 6.5 & & & & & + @xmath84/cm
s@xmath85 ) & & & 2.15 & [email protected]@xmath86 & & & + @xmath87(% ) & & 4@xmath70 & 2.4 & @xmath10@xmath88 & 0 - 5 & @xmath2110 & 74.4 + @xmath22(% ) & & 40 & 94.7 & 99.85@xmath86 & 98.6 & 40 - 50 & 25.0 + @xmath23(% ) & & 55 & 2.9 & 0.05@xmath86 & 0.7 & 40 - 50 & 0.20 + @xmath24(% ) & & 0.5 - 1@xmath72 & & 0.09@xmath86 & 0.4 & 0 - 1 & 0.09 + @xmath25(% ) & & @xmath890.01 & & & 0.2 & 5 - 10 & 0.80 + @xmath26(% ) & & @xmath210.1@xmath72 & & 0.01@xmath86 & 0.02 & & 0.09 + code & & non - lte sp@xmath74 & lte pp@xmath75 & lte pp@xmath75 & non - lte pp@xmath75 & non - lte sp@xmath74& + llll star & year & max brightness & reference + v348 sgr & 1900 - 54 & @xmath45=11.0 & 1 + & 1954 - 81 & @xmath46=11.6 - 11.8 & 2 + & 1973 & @xmath13=12.1 & 3 + & 1981 & @xmath13=12.3 & 3 + & 1990 & @xmath13=12.0 & 4 + & 1995 & @xmath13=11.9 & 4 + & 1998 & @xmath13=11.9 & 4 + & 1960 - 2001 & @xmath46=12.0 & 5 + dy cen & pre-1930 & @xmath45=12.0 & 6 + & 1960 - 1977 & @xmath46=12.4 - 12.7 & 7 + & 1972 & @xmath13=12.4 & 8 + & 1982 - 3 & @xmath13=12.5 & 9 + & 1987 & @xmath13=12.7 & 10 + & 1987 & @xmath13=12.8 & 11 + & 1990 & @xmath13=12.8 & 4 + & 1982 - 2001 & @xmath46=12.7 - 13.0 & 5 + mv sgr & pre-1928 & @xmath45=12.1 & 12 + & 1924 - 49 & @xmath45 = 12.0 & 13 + & 1963 & @xmath13=12.70 & 14 + & 1985 & @xmath13=13.34 & 15 + & 1980 - 2000 & @xmath13=13.0 - 13.4 & 4 + & 1987 - 2001 & @xmath46 = 13.4 & 5 + cccccccc & & & + test & line profile & center & fwhm & ew & center & fwhm & ew + & & ( km s@xmath6 ) & ( km s@xmath6 ) & ( ) & ( km s@xmath6 ) & ( km s@xmath6 ) & ( ) + & 1 & 15 & 384 & 1.7 & 101 & 60 & 0.24 + 1 & 2 & * 43 * & * 47 * & * 0.41 * & * 22 * & * 46 * & * 0.61 * + & 3 & * 37 * & * 74 * & * 1.0 * & * 43 * & * 49 * & * 1.2 * + 2 & 1 & 26 & 551 & 1.8 & 96 & 49 & 0.2 + & 2 & * 11 * & * 124 * & * 2.2 * & * 19 * & * 85 * & * 2.6 * + llll & v348 sgr & dy cen & mv sgr + @xmath93(pn ) & low & low & high + @xmath94(pn ) & low & low & high + split stellar lines & yes & yes & yes + stellar & [ wc]-like & he+c , no & h - burning ashes by + abundances & & sign of h- & cn - cycle , emission + & & burning & of heavy elements + spectral & strengthening emission & strengthening emission & none + variability & lines = increasing wind ? & lines = increasing wind ? & + long - term maximum & decreasing & decreasing & decreasing + light variability & = increasing t@xmath0 & = increasing t@xmath0 & = increasing t@xmath0 + decline activity & very high & low & low + | we investigate the evolutionary status of four stars : v348 sgr , dy cen and mv sgr in the galaxy and hv 2671 in the lmc .
these stars have in common random deep declines in visual brightness which are characteristic for r coronae borealis ( rcb ) stars .
rcb stars are typically cool , hydrogen deficient supergiants .
the four stars studied in this paper are hotter ( t@xmath0 = 15 - 20 kk ) than the majority of rcb stars ( t@xmath0 = 5000 - 7000 k ) . although these are commonly grouped together as the _ hot rcb stars
_ they do not necessarily share a common evolutionary history .
we present new observational data and an extensive collection of archival and previously - published data which is reassessed to ensure internal consistency .
we find temporal variations of various properties on different time scales which will eventually help us to uncover the evolutionary history of these objects .
dy cen and mv sgr have typical rcb helium abundances which excludes any currently known post - agb evolutionary models .
moreover , their carbon and nitrogen abundances present us with further problems for their interpretation .
v348 sgr and hv 2671 are in general agreement with a born - again post - agb evolution and their abundances are similar to wolf - rayet central stars of pn .
the three galactic stars in the sample have circumstellar nebulae which produce forbidden line radiation ( for hv 2671 we have no information ) . v348
sgr and dy cen have low density , low expansion velocity nebulae ( resolved in the case of v348 sgr ) , while mv sgr has a higher density , higher expansion velocity nebula .
all three stars on the other hand have split emission lines which indicate the presence of an equatorial bulge but _ not _ of a keplerian disk . in addition , the historical light - curves for the three galactic hot rcb stars , show evidence for a significant fading in their maximum - light brightnesses of @xmath11 mag over the last 70 yr . from this we deduce that their effective temperature increased by a few thousand degrees . if v348 sgr is a born - again star , as we presume , this means that the star is returning from the born - again agb phase to the central star of pn phase .
spectroscopically , no dramatic change is observed over the last 50 years for v348 sgr and mv sgr .
however , there is some evidence that the winds of v348 sgr and dy cen have increased in strength in the last decade .
hv 2671 , located in the lmc has not been analyzed in detail , but at 5- resolution is almost identical to v348 sgr . through
the bolometric correction derived for v348 sgr and the known distance , we can estimate the absolute v magnitude of hv 2671 ( @xmath2 = 3.0 mag ) and its bolometric luminosity ( @xmath16000 l@xmath3 ) . |
cosmic strings @xcite are linear topological defects which arise in a range of relativistic quantum field theories ( for reviews see e.g @xcite ) .
good analogs of cosmic strings are vortex lines in superfluids and superconductors .
line defects in crystals can be viewed as another analog system .
cosmic strings form lines of trapped energy density , and this energy density can curve space - time and have important effects in cosmology @xcite .
cosmic strings are predicted to form in many particle physics models beyond the standard model . in particular , they are predicted to form at the end of inflation in many inflationary models , e.g. supergravity models @xcite and brane inflation models @xcite .
cosmic strings may also survive as cosmic superstrings in alternatives to inflation such as string gas cosmology " @xcite .
the key point for cosmology is that in any field theory model which admits cosmic string solutions , a network of strings inevitably forms at some point during the early universe @xcite , and it persists to the present time .
hence , the detection of cosmic strings would give us information about particle physics at very high energy scales .
since cosmic strings are relativistic objects , a straight string is described by one number , namely its mass per unit length @xmath0 which also equals its tension , or equivalently by the dimensionless number @xmath1 , where @xmath2 is newton s gravitational constant ( we are using units in which the speed of light is @xmath3 ) . in simple quantum field theory models
the tension is related to the energy scale @xmath4 at which the strings are formed via @xmath5 ( see the following section for a more precise discussion ) .
the cosmological signatures of strings are thus more substantial for larger values of @xmath0 which implies larger values of the energy scale @xmath4 .
hence , searching for cosmological signatures of strings is a tool to probe particle physics beyond the standard model at the highest energy scales ( as opposed to accelerator experiments like the lhc which probe new physics at low energy scales ) .
in fact , current limits on cosmic strings already @xcite provide a constraint g < 1.5 10 ^ -7 which rules out certain grand unified particle physics models with very high scale symmetry breaking .
this limit comes from the observational upper bound on the contribution of cosmic strings to the angular power spectrum of cosmic microwave background ( cmb ) anisotropies obtained by combining the results of the wmap satellite @xcite with those of the south pole telescope @xcite ( see also @xcite for a comparable limit obtained by combining results from wmap and from the atacama cosmology telescope @xcite , and @xcite for earlier limits ) . given the constraints on particle physics models which can be derived from current observations , it is of great interest to try to improve the observational upper bounds on the cosmic string tension since this will allow us to constrain high energy scale particle physics models more strongly than what is possible today .
cosmic strings can also produce many good things for cosmology in addition to contributing to cosmological structure formation .
cosmic strings may play a role in baryogenesis ( see e.g. @xcite ) .
certain types of strings can provide a mechanism for the production @xcite of seed magnetic fields which are coherent on galactic scales cusps on cosmic string loops can also yield a contribution to ultra - high - energy cosmic rays @xcite .
finally , cosmic string loops may assist in the assembly of the large mass concentrations required to seed super - massive black holes . for all of the above reasons it would thus be wonderful to have evidence for the existence of cosmic strings in nature .
the search for cosmic strings is therefore of great interest independent of whether the search in fact finds signals of cosmic strings .
if it does , then we will have discovered something completely new in the universe .
if it does not , then we will have derived tighter constraints on particle physics at very high energy scales . in this talk i will review recent work on signatures of cosmic strings in new observational windows . up to the present time ,
the tightest and most robust constraints on the cosmic string tension have come from analyses of cmb temperature maps .
here , i will focus on the signatures of strings in cmb polarization maps and 21 cm redshift surveys , two emerging windows to explore the cosmos .
the main points to take away from this talk are the following .
firstly , cosmic strings lead to nonlinearities already at very high redshifts .
hence , the signatures of cosmic strings are more pronounced at higher than at lower redshifts where they are masked by the nonlinearities produced by the gaussian density fluctuations which must be present and which dominate the total power spectrum of cosmological perturbations .
secondly , cosmic strings lead to perturbations which are highly non - gaussian and which predict specific geometrical patterns in position space . by computing a power spectrum , information about these patterns
hence , tighter limits on the cosmic string tension can be obtained if we analyze the data in position space .
thirdly , 21 cm redshift surveys appear to be an ideal window to search for cosmic string signatures @xcite .
the outline of this talk is as follows .
we first present a brief review of the basics of cosmic strings . in section 3
we introduce the two main mechanisms which will play a role in determining the cosmic string signals in observations , namely the kaiser - stebbins @xcite ( see also @xcite ) lensing effect and the cosmic string wake @xcite .
we also briefly review the well - known resulting signal of strings in cmb temperature maps .
the key sections of this talk are sections 4 and 5 . in the first
, we discuss the signal of a long straight cosmic string in cmb polarization maps , and in the second we turn to the signal in 21 cm redshift surveys .
in a class of relativistic quantum field theories , cosmic strings form after a phase transition in the early universe during which an internal symmetry in field space is spontaneously broken .
let us consider a simple toy model involving a complex scalar field @xmath6 with potential v ( ) = ( ||^2 - ^2 ) ^2 where @xmath4 is the vacuum expectation value of the modulus of @xmath6 and @xmath7 is a coupling constant . to determine whether a particular field theory admits cosmic string solutions or not , the key concept is that of the _ vacuum manifold _ @xmath8 , the set of field values which minimizes the potential . in the above example
@xmath8 is homotopically equivalent to the circle @xmath9 . in thermal equilibrium ,
the potential obtains finite temperature corrections @xcite ( see e.g. @xcite for a review ) .
specifically , there is an extra contribution v_t ( ) ~ t^2 ||^2 , where @xmath10 is the temperature . hence , there is a critical temperature @xmath11 above which the lowest potential energy state is @xmath12 and the field symmetry ( rotation in the complex field plane ) is unbroken .
thus , in the early universe the average value of @xmath6 at each point in space will be @xmath12 , but as the universe cools below the temperature @xmath11 this state becomes unstable and at each point in space @xmath6 will want to roll down the potential to take on a value in @xmath8 .
the key point is @xcite that by causality there can be no correlation between the field values in @xmath8 which are taken on at points in space which are out of causal contact , i.e. which are further apart than the hubble distance @xmath13 .
hence , there is a probability of order @xmath14 that the field values in @xmath8 evaluated for a loop @xmath15 in space of radius @xmath13 will form an incontractible loop in @xmath8 .
this implies that there must be a point in space on any disk bounded by @xmath15 where @xmath12 . around these points there is trapped potential and spatial gradient energy .
energy minimization arguments make it obvious that the distinguished points on different disks ( with the same boundary ) form part of a line .
this line is the cosmic string .
it is a line of points with @xmath12 surrounded by a tube of trapped energy .
the width @xmath16 of the tube is determined by minimizing the sum of potential and spatial gradient energy , leading to the result w = ^-1/2 ^-1 such that the resulting mass per unit length @xmath0 is independent of the coupling constant : = ^2 ( up to numerical factors independent of @xmath7 ) . to recap
, causality tells us that at the symmetry breaking phase transition at time @xmath17 a network of cosmic strings will form in any field theory model which admits cosmic string solutions .
cosmic strings are closed - they can not have ends .
hence , they are either string loops or infinite in length .
note that causality in fact ensures that ( in an infinite space ) strings of infinite length will be formed . a way to picture the network of strings is as a random walk with typical curvature radius @xmath18 which is bounded from above by the hubble length @xmath13 . the same causality argument which is used to predict the formation of strings
can be used to show that the network of strings will survive at all times @xmath19 , including the present time .
hence , cosmic strings formed in the very early universe will lead to signatures which can be searched for in current cosmological observations .
we will work in terms of a one - scale model @xcite of the cosmic string distribution which is given by the correlation length @xmath18 which describes both the mean curvature radius and the separation of the long " strings ( long " meaning with length greater than @xmath13 ) .
kibble s causality argument tells us that @xmath20 .
dynamical arguments show that @xmath18 can not be much smaller than @xmath13 .
the argument is as follows : cosmic strings are relativistic objects .
if the curvature radius is smaller than @xmath13 , then the strings will move with relativistic speed .
hence , intersections of strings will occur .
it can be shown @xcite that unless the relative speed of two intersecting string segments is extremely close to the speed of light , then the string segments will not cross , but they will intersect and exchange ends . in this way , the long strings can form string loops , which in turn oscillate and decay by emitting gravitational radiation .
this process is described by a boltzmann equation ( see e.g. @xcite ) from which it follows that if @xmath21 then @xmath22 will increase , and that there will be a dynamical fixed point _ scaling solution _ with ( t ) ~t .
the approach to a scaling solution has been confirmed in a series of numerical simulations @xcite .
the process of long string intercommutation leaves behind a distribution of cosmic string loops which also achieves a scale - invariant form .
we will not discuss this issue since we will focus on the cosmological signatures of the long strings .
since cosmic strings carry energy , they can gravitate and produce cosmological signals .
this was first pointed out in @xcite . in the early 1980s cosmic strings
were studied as an alternative to cosmological inflation @xcite .
cosmic string formation produces entropy fluctuations on super - hubble scales which then seed a curvature fluctuation which grows on super - hubble scales .
such fluctuations are called active " .
they are also incoherent " meaning that there are no phase correlations in phase space between different fourier modes @xcite .
hence , although cosmic strings predict a spectrum of density fluctuations which is scale - invariant and hence an angular power spectrum of cmb anisotropies which is also scale - invariant on large angular scales @xcite , the acoustic oscillations which are characteristic of models like inflation where a scale - invariant spectrum of adiabatic fluctuations are generated whose amplitude remains constant on super - hubble scales are absent @xcite .
once the acoustic oscillations were discovered by various experiments such as the boomerang experiment @xcite , interest in cosmic strings collapsed .
however , given the realization that cosmic strings are produced after inflation in many particle physics models , interest in searching for observational signatures of cosmic strings as a supplementary ( not main ) source of structure in the universe has increased .
the search for strings in the cosmos is emerging as a promising way to probe physics at energy scales much larger than those that can ever be reached in terrestrial accelerators , as already mentioned above .
independent of whether strings are in fact discovered or the bounds on the existence of strings are improved we will have learned a lot about particle physics .
in this talk i will focus on gravitational effects of cosmic strings .
these are based on two main actors " , firstly the _ kaiser - stebbins _ lensing effect @xcite , and secondly the existence of _ string wakes _ @xcite .
the string lensing effect is based on the fact that a long straight cosmic string with equal tension and energy per unit length leads to a conical structure of space perpendicular to the string @xcite .
unwrapping the cone onto a plane leads to a deficit angle " = 8 g .
if we now consider a cosmic string with transverse velocity @xmath23 and associated relativistic gamma factor @xmath24 , then if we look at the cmb in direction of the string ( see fig .
1 ) , then photons passing on the two sides of the string are observed with a relative doppler shift = 8 ( v ) v g .
this leads to a line discontinuity in the cmb sky , the linear direction corresponding to the projection of the tangent vector of the string onto our last light cone @xcite . in a direction transverse to the direction of the string ,
then photons passing on different sides of the string are measured with a relative doppler shift .
this is the kaiser - stebbins effect.,height=302 ] given a scaling distribution of strings , each string which intersects our past light cone will lead to a line discontinuity in the cmb sky . as shown in @xcite , for strings formed at a phase transition at time @xmath25 in the early universe ( as opposed to idealized strings being present at all times ) , the deficit angle at time @xmath13 has finite depth @xmath26 , namely d(t ) ( t - t_c ) t for @xmath27 .
hence , the signature of an individual string in the sky corresponds to a rectangle in the sky with the above value of @xmath28 .
the dimensions of the rectangle are the angles given by the comoving distances corresponding to length @xmath29 and depth @xmath13 , where @xmath30 is a constant of order @xmath14 .
we will be making use of a toy model for the ( long ) cosmic string scaling first introduced in @xcite and widely used since then in cosmic string research .
we replace the infinite string network at time @xmath13 by a set of straight string segments of length @xmath29 whose center of mass positions , tangent vectors and velocity vectors are randomly distributed and uncorrelated .
these string segments live for one hubble expansion time ( the average time between string intercommutations ) . in the subsequent hubble time steps
there are new string segments with larger length whose centers , tangent and velocity vectors are assumed to be uncorrelated with those of the string segments in the previous hubble time step . as a reflection of the scaling of the string network , we consider a fixed number @xmath31 per string segments in each hubble volume .
the number @xmath31 can be determined by comparing with numerical simulations .
causality tells us that @xmath32 .
numerical simulations give values of the order @xmath33 @xcite .
string segments in each hubble time step contribute to the total kaiser - stebbins effect .
string segments in the first hubble time step after recombination are the most numerous .
the size in the sky of the corresponding temperature patches is about one degree .
strings from later hubble time steps are less numerous but larger ( see fig .
patch of the sky at @xmath34 resolution ( the specifications are chosen to be comparable to those of the spt and act telescopes - in fact both of these telescopes map a larger fraction of the sky ) in a model in which the fluctuations are given by a scaling distribution of cosmic strings . the color coding indicates the amplitude of the temperature anisotropy.,height=529 ] to identify the signal of cosmic strings in cmb temperature maps , good angular resolution is more important than full sky coverage .
hence , experiments such as spt and act will yield better limits than even the planck satellite .
the distinctive edges in cmb temperature maps produced by strings are washed out in the angular power spectrum .
hence , tighter limits can be set on the string tension by analyzing the cmb maps in position space using edge detection algorithms rather than by simply computing a power spectrum .
specifically , by using a numerical implementation of the canny edge detection algorithm @xcite it appears that a bound one order of magnitude stronger than the current bound might be achievable by using data from the spt telescope @xcite ( see also @xcite for initial work on applying the canny algorithm to cmb maps ) .
the projected bound is g 2 10 ^ -8 .
the second main actor " in the story presented here is the cosmic string wake @xcite , consider a long straight string segment moving through the uniform matter distribution of the early universe . from the point of view of a point behind the moving string
, it appears as though matter acquires a velocity perturbation v = 4 v ( v ) g from above and below towards the plane behind the moving string .
this in turn leads to a wedge - shaped overdensity ( twice the background density ) behind the string .
this is the _
wake_. working again in the context of the toy model of @xcite , each string segment in each hubble expansion time generated a wake .
consider a string at time @xmath35 .
the physical dimension of the induced wake at the time @xmath35 will be c_1 t_i v ( v ) t_i 4 v ( v ) g t_i , where the first factor is the size along the tangent vector of the string , the second factor is the depth ( in direction opposite to the string motion ) , and the third factor is the mean width . at the leading edge ( the instantaneous location of the string ) , the width of the wake is zero , whereas at the trailing edge ( the initial position of the string segment ) the width is twice the mean width @xmath16 .
is the string velocity @xmath23 , and @xmath36 is its related gamma factor.,height=170 ] once formed , a wake will grow in thickness via gravitational accretion .
this process can be studied using the zeldovich approximation @xcite .
the idea is to consider a shell of matter which is located initially ( at the time @xmath35 when the wake is laid down ) at a physical height @xmath37 above the center of the wake .
the quantity @xmath38 is the initial comoving height . as a consequence of the gravitational pull of the matter overdensity in the wake , a comoving displacement @xmath39 gradually builds up ( where @xmath40 ) . the physical height at time @xmath41 then can be written as h(q , t ) = a(t ) ( q - ) , where @xmath42 is the cosmological scale factor . the time evolution of the height is then determined via newtonian gravity = - , where @xmath43 is the newtonian gravitational potential which is determined via the poisson equation in terms of the mass overdensity .
we then calculate the value @xmath44 ( which we call @xmath45 ) for which the shell is turning around " at time @xmath13 , i.e. = 0 . after turnaround
, the shell will virialize at a physical height which is half the turnaround height this virialized region forms the wake .
the result of a straightforward calculation shows that ( in agreement with what follows from linear perturbation theory ) , the comoving height of the wake grows linearly in the scale factor , i.e. q_nl(t , t_i ) = v ( v ) g ( z(t_i ) + 1)^-1/2 t_0 , where the expression on the right hand side is the ratio of scale factors multiplying the initial comoving width of the wake ( modulo a factor of order @xmath14 ) .
note that @xmath46 is the cosmological redshift . in the context of cosmic strings
this analysis was originally done in @xcite ( accretion onto loops ) and @xcite ( accretion onto wakes ) .
since the turnaround height itself is half the height the shell would have if it were simply to expand with the hubble flow , the resulting overdensity in the wake is a factor of @xmath47 .
note that for accretion onto a string loop , the resulting overdensity is @xmath48 since there is contraction in all three spatial directions ( see e.g. @xcite ) .
let us end this section with a couple of comments .
first of all , the planar dimensions of the wake will retain constant comoving size .
secondly , there is an important difference between the lensing signal due to string segments and the wake signal .
since the string segment only lives for one hubble expansion time , only strings whose finite time world sheet intersects the past light cone lead to an observable signal . on the other hand ,
wakes persist even after the string segment which has seeded them has decayed .
hence , all string segments within the past light cone lead to observable wake signals . in the following ,
it is signals due to wakes which will be discussed .
when primordial cmb quadrupole radiation scatters off of a gas cloud , the residual free electrons in the cloud lead to polarization .
wakes are regions of enhanced density , and hence also of enhanced free electron density .
photons emitted at the time of recombination acquire extra polarization when they pass through a string wake .
it is this signal which we study here @xcite .
there are two polarization modes - e and b modes .
a gaussian random field of density fluctuations leads to pure e - mode polarization .
in contrast , cosmic string wakes lead to a statistically equal distribution of e and b - mode polarization .
the reason for the generation of a b - mode component is that there is a distinguished vector given by the normal vector to the wake .
this vector is independent of the cmb quadrupole vector .
the amplitude @xmath49 of the polarization signal depends on the column density of free electrons which the cmb photons encounter when they cross the wake . if @xmath13 is the time when the photons are passing through the wake , the column density is proportional to the residual ionization fraction @xmath50 at time @xmath13 , the number density of baryons and the width of the wake .
the value of @xmath49 is then determined by multiplying the result with the scattering cross section @xmath51 and the amplitude @xmath52 of the cmb quadrupole .
the result is ( see @xcite for details ) : [ result ] & & ( ) ^1/2 _ t f(t ) g v ( v ) + & & _ b _ c(t_0 ) m_p^-1 t_0 ( z(t ) + 1 ) ^2 ( z(t_i ) + 1 ) ^1/2 , where @xmath35 is the time when the wake is laid down , @xmath53 is the fraction of the total energy density which is in baryons , and @xmath54 is the critical energy density ( energy density of a spatially flat universe ) at the present time @xmath55 .
note that the induced polarization increases as the wake formation time decreases .
this is because early wakes have had time to accrete more matter and are thicker than late wakes .
the value of @xmath49 also increases as @xmath46 increases .
this is because the density in a wake is larger at earlier times than later ones . inserting the value of the constants appearing above and evaluating the result in units of the characteristic values
@xmath56 we obtain [ value ] ~ f(t ) g v ( v ) _ b ( ) ^2 ( ) ^3 10 ^ 7 .
the residual ionization fraction drops off after recombination to a value of between @xmath57 and @xmath58 ( see e.g. @xcite ) , but increases again at the time of reionization to a value of order unity .
however , as can be seen from ( [ value ] ) , the predicted cmb polarization has an amplitude which is suppressed by either @xmath50 for times @xmath13 between recombination and reionization , or by the square of @xmath59 for times after reionization . in spite of the small amplitude of the signal
, the string - induced polarization may still be detectable because of the specific geometry of the signal in position space : a cosmic string wake leads to a rectangular region in the sky with extra polarization with almost uniform polarization axis and with an amplitude which is increasing monotonically from the leading edge of the wake ( where the string is located ) to the trailing edge ( where the string was at the time of wake formation ) , and because statistically an equal amount of e - mode and b - mode polarization is generated . in this context ,
i wish to comment on the statement which is often heard that `` b - mode polarization is the holy grail of inflation . ''
what is usually meant by this statement is that b - mode polarization is predicted in inflationary models from the spectrum of gravitational waves which inflation generates .
thus , it is claimed that the detection of b - mode polarization will be a signal of the gravitational waves which inflation predicts .
however , as we have shown above , cosmic strings predict direct b - mode polarization independent of gravitational waves .
thus , it is not correct to interpret a potential detection of b - mode polarization as being due to gravitational waves .
secondly , there are sources of gravitational waves which generically lead to a higher amplitude of such waves than what is obtained in the simplest single field slow - roll inflation models .
specifically , the cosmic string loops which are inevitably produced via the dynamics of the scaling network of long strings will oscillate and decay by emitting gravitational waves .
thus , a detection of gravitational waves ( via b - mode polarization or other means ) is more likely to be a signal of something different from inflation .
there is , in fact , an interesting twist to the story : if b - mode polarization is discovered and shown to be due to gravitational waves , and a blue spectrum ( more power on shorter wavelengths ) is measured , then this would rule out these gravitational waves as being due to inflation . instead , it would be the confirmation of an effect first predicted in the context of superstring theory , namely the spectrum induced in the string gas alternative to inflationary cosmology @xcite .
in conclusion , b - mode polarization can be viewed as the `` holy grail '' of early universe cosmology , but not as the `` holy grail '' of inflation .
the 21 cm redshift survey technique is emerging as a promising tool to probe cosmology , in particular the high redshift universe , the universe during the dark ages " ( i.e. before the onset of star formation ) .
the physics is the following : after the time of recombination but before reionization the baryonic matter in the universe is mostly in the form of neutral hydrogen .
neutral hydrogen has a hyperfine transition line at a frequency corresponding for 21 cm .
if we consider the primordial cmb radiation passing through a gas cloud , then the spectrum at the rest frame frequency of 21 cm is changed by excitation or de - excitation of the hyperfine transition .
the primordial 21 cm photons can be absorbed by the gas cloud . in turn
, a hot gas cloud will emit 21 cm photons . whether the net effect is an absorption or emission effect depends on the temperature of the gas cloud .
any gas cloud which intersects our past light cone at some time between recombination and the present time will yield a 21 cm signal .
if the time of intersection corresponds to a redshift @xmath46 , then the emission / absorption signal will be seen at the redshifted wavelength ( t ) = ( z(t ) + 1 ) _ 0 , where @xmath60 is 21 cm .
hence , 21 cm redshift surveys provide a means for mapping the distribution of baryonic matter as a function of redshift , including at times before the onset of star formation .
cosmic string wakes are nonlinear density perturbations present at arbitrarily early times with a distinctive geometric pattern in position space .
thus , as realized in @xcite , ( see also @xcite for early work on cosmic strings and 21 cm redshift surveys ) , cosmic strings predict striking signals , in particular at redshifts before reionization .
let us begin this section with some basic equations required to compute the 21 cm signal ( see @xcite for a comprehehsive review of 21 cm cosmology ) . let us consider primordial cmb photons passing through a gas cloud at redshift @xmath46 .
then the brightness of the 21 cm radiation emerging from the gas cloud is [ basic1 ] t_b ( ) = t_s ( 1 - e^- _ ) + t _ ( ) e^- _ , where @xmath61 is the spin temperature " of the hydrogen atoms in the gas cloud , and @xmath62 is the temperature of the cmb photons before entering the gas cloud .
the quantity @xmath63 is the optical depth at the frequency being considered .
the second term in ( [ basic1 ] ) describes the absorption of the primordial cmb photon due to excitation of the hydrogen atoms , the first term gives the contribution due to de - excitation of the hydrogen atoms .
the spin temperature introduced above determines the excitation level of the hydrogen atoms .
it is related to the gas temperature in the wake and to the cmb temperature via a collision coefficient @xmath64 : [ basic2 ] t_s = t _ . the relative brightness temperature measured by the observer today is [ basic3 ] t_b ( ) = , where the denominator is due to the redshifting of the temperatures between the time when the gas cloud intersects our past light cone and the present time .
let us now turn to the application to cosmic string wakes .
5 presents a sketch of 21 cm photons from a string wake .
as indicated , there will be a frequency dispersion caused by the different times the photons are emitted .
its magnitude is proportional to the width of the wake and is given by [ freqdisp ] = 2 sin ( ) ( ) h w , where @xmath65 is the hubble expansion rate , @xmath16 is the wake width , and @xmath66 is the angle between the perpendicular to the plane of the wake and the line of sight towards us . the optical depth for 21 cm photons passing through the wake is ( in units where @xmath67 ) [ optical ] _ = ( ) ( ) , where @xmath68 is the column number density of hydrogen atoms , @xmath69 is the spontaneous emission coefficient of the 21 cm transition , and @xmath70 is the line profile [ line ] ( ) = for [ _ 10 - , _ 10 + ] . note that the width of the string wake ( and hence the dependence on @xmath71 cancels out between the column density and the line profile .
this will lead to the relative brightness temperature signal being independent of the string tension . to determine the relative brightness temperature
, we also need to know the spin temperature of the hydrogen gas inside the wake .
this is in turn determined ( [ basic2 ] ) from the kinetic temperature of the wake .
assuming that the kinetic temperature is obtained via thermalization from the kinetic energy acquired during the zeldovich collapse of the wake leads to the result @xcite [ waketemp ] t_k [ 20 k ] ( g ) _
6 ^ 2 ( v ( v))^2 , where @xmath72 is the value of @xmath1 in units of @xmath73 , @xmath74 is the redshift of wake formation and @xmath75 is the redshift at which the wake intersects the past light cone .
inserting this result ( [ waketemp ] ) into the expression for the 21 cm brightness temperature ( [ basic1 ] ) yields the following relative brightness temperature t_b ( ) = [ 0.07 k ] ( 1 - ) ( 1 + z)^1/2 whose amplitude is of the order t_b ( ) ~200 mk for , z + 1 = 30 ( we have taken the redshift of emission from the wake to be larger than the redshift of rionization so that we do not have to bother about nonlinearities from the gaussian fluctuations ) . the amplitude of the cosmic string - induced 21 cm signal is thus very large . whether the signal is in emission or absorption depends on which of @xmath76 or @xmath62 is larger .
in the former case ( low redshifts ) we have an emission signal , in the latter case it is a signal in absorption . is a decreasing function of time , whereas the wake temperature increases for a fixed wake . ]
the transition between emission and absorption takes place when ( g ) _ 6 ^ 2 0.1 ( v ( v))^-2 . for values of @xmath1 which are comparable or lower than the current upper bound , the signal is thus more likely to be in absorption rather than in emission .
wakes formed after the time of recombination @xmath77 intersect the past light cone over a large area in the sky ( comparable or greater to a square degree ) .
each wake will produce a signal in the form of a thin wedge in three dimensional redshift survey space .
the leading edge of the wedge ( where the cosmic string segment is located at the final time ) corresponds to a later time and hence to a frequency which has undergone less frequency redshift .
thus , the two `` large '' dimensions of the wedge are not exactly perpendicular to the frequency axis .
the geometry is illustrated in fig .
the string segment `` lives '' until the time @xmath78 .
its initial position is @xmath79 , its final position @xmath80 .
the string wake extends in comoving coordinates from @xmath79 to @xmath80 .
its thickness at @xmath79 vanishes and is maximal at @xmath80 .
the time @xmath55 is the present time .
our past light cone ( indicated by the line labelled @xmath81 ) intersects the string wake . for the configuration shown , the past light cone intersects the leading ( thin ) edge of the wake earlier than the trailing edge .
hence , 21 cm photons from the leading edge are redshifted more than those from the trailing edge .
this gives rise to a characteristic wedge of extra 21 cm absorption / emission due to the string wake in 21 cm redshift maps .
this wedge is sketched on the left of the figure , a sketch in which the horizontal axis is the same comoving spatial coordinate as in the left sketch , and the vertical axis is the detected 21 cm frequency.,height=170 ] whereas the amplitude of the brightness temperature signal at a fixed point inside the wedge does not depend to a first approximation on @xmath1 , the width of the wedge does depend on it .
the relative width is given by [ freqwidth ] & = & g v_s _ s ( z_i + 1 ) ^1/2 ( z(t ) + 1 ) ^-1/2 + & & 3 10 ^ -5 ( g ) _ 6 v ( v ) , using @xmath82 and @xmath83 in the second line .
hence , an instrument with good frequency resolution is required to be able to measure the cosmic string wake signal at full strength .
if the frequency resolution is worse than the above value , then the wake signal is still detectable , but with a reduced effective brightness .
the analysis presented up to this point has assumed that the initial thermal temperature @xmath84 of the gas is negligible compared to the temperature acquired during infall . for small values of @xmath1
this assumption will fail . in this case the wake will not experience shocks .
the initial thermal velocities will dominate and will lead to a wider but more diffuse wake .
the width in the presence of initial gas temperature compared to the width @xmath85 for @xmath86 is given by @xcite w(t)|_t_k
< t_g = w(t ) |_t_g = 0 .
since the hydrogen column density remains unchanged the temperature at a fixed point in the sky will decrease since the frequency dispersion increases . in figure 7
we show the brightness temperature excess as a function of @xmath1 for various values of the formation redshift . for large values of @xmath1 ,
the relative brightness temperature is positive . for smaller values
, it is an absorption effect .
the kink point along each curve corresponds to the lowest value of @xmath1 for which there is shock - heating .
the ( brown ) almost horizontal lines on the graph correspond to the pixel by pixel sensitivity of an experiment such as the square kilometer array ( ska ) , with a pixel size optimized as a function of @xmath1 ( see @xcite ) .
we see that the string signal is larger than the predicted noise level for values of @xmath1 substantially smaller than the kink value . )
correspond to @xmath87 and @xmath88 .
the horizontal axis is the value of @xmath1 in units of @xmath73 .
the two brown almost horizontal lines indicate the expected thermal noise per pixel of an experiment such as the ska ( the @xmath1 dependence of the horizontal lines comes from the choice of the optimal pixel size as a function of @xmath1 ( see @xcite).,height=170 ] since the signal of a string wake has a very special pattern in position space , it is possible to search for this signal even if the relative brightness temperature at a fixed point in the sky is smaller than the pixel noise , in the same way that the line discontinuities caused by string segments in cmb temperature maps can be picked out for values of @xmath1 where the pixel by pixel signal is hidden in the noise . as in the case of cmb temperature maps
, it is important to use position space algorithms to analyze the maps . for an attempt to use minkowski functionals to pick out string signals in 21 cm redshift maps see @xcite .
naturally , it is also possible to compute the power spectrum of the cosmic string wake signal in 21 cm redshift maps . the angular power spectrum at a fixed value of the frequency has been computed in @xcite .
it is also possible to compute the 21 cm signal of a cosmic string loop .
since the overdensity in the region which accretes around the string loop is @xmath48 for a loop compared to @xmath47 for a wake ( contraction occurs in all three dimensions ) , the induced brightness temperature is in fact even larger than for a string wake @xcite .
however , there are no special patterns which allow the string loop signal to be teased apart from background point sources . hence
, we consider it to be more promising to search for the signals of string wakes .
i have discussed the search for cosmic string signals in new observational windows .
there is good motivation for this work : detecting a cosmic string in cosmology would in itself be a great discovery which might lead to the solution of some outstanding problems in cosmology such as the origin of primordial magnetic fields which are coherent on galactic scales . maybe more importantly , searching for cosmic strings in the sky is a way to probe particle physics beyond the standard model at energy scales which can never be reached in terrestrial accelerators .
cosmic strings contribute to structure formation and hence leave imprints in the structure in the universe .
however , the contribution of strings to the total power of cosmological perturbations is already bounded from above to be less than about @xmath89 at the present time .
the bulk of the power comes from almost gaussian primordial fluctuations ( here called gaussian noise " ) such as those which could be generated by inflation or its alternatives .
the first main point to take away from this talk is that cosmic strings produce nonlinearities already at very high redshifts ( whereas the gaussian noise does not ) .
hence , the signatures of cosmic strings will be more pronounced at high redshifts .
hence , cmb anisotropy maps and 21 cm redshift surveys are ideal windows to probe for strings .
the second main point is that cosmic strings produce fluctuations which are non - gaussian .
more specifically , string wakes induce signals with specific geometrical patterns in position space .
hence , better limits on cosmic strings can be achieved when analyzing the data in position space ( e.g. with edge detection algorithms ) rather than via power spectra computations .
there are other windows to probe cosmic strings .
one is via gravitational waves .
oscillating cosmic string loops emit gravitational radiation @xcite .
a scaling network on strings will produce a scale - invariant spectrum of gravitational waves ( see e.g. @xcite ) .
cusps on cosmic strings may produce more distinctive signals @xcite .
however , it is expected that back - reaction effects ( see e.g. @xcite ) could greatly reduce the gravitational wave signal of cusps , and hence limits on @xmath1 derived from cusp signals should be taken with lots of grains of doubt . another window to probe cosmic strings is via high redshift galaxy surveys
. the relative contribution of strings to correlation functions of nonlinear mass concentrations in the universe increases as the redshift grows .
hence , it is interesting to look for signals of cosmic strings in the high redshift galaxy distribution .
quite recently , the contribution of cosmic string loops to galaxy formation at high redshifts has been studied @xcite .
corresponding work on the effects of cosmic string wakes is ongoing at mcgill .
i wish to thank prof .
pauchy hwang for the invitation to lecture at cospa2012 and for his wonderful hospitality .
i wish to thank all of my collaborators on the recent cosmic string work , and in particular rebecca danos for permission to use various figures drawn for @xcite , and oscar hernandez for permission to use figure 7 which is taken from @xcite .
this research has been supported in part by an nserc discovery grant and by funds from the canada research chair program .
a. vilenkin , `` cosmic strings and domain walls , '' phys .
rept . *
121 * , 263 ( 1985 ) ; + a. vilenkin and e.p.s .
shellard , _ cosmic strings and other topological defects _ ( cambridge univ . press , cambridge , 1994 )
; + m. b. hindmarsh and t. w. b. kibble , `` cosmic strings , '' rept . prog .
phys . * 58 * , 477 ( 1995 ) [ arxiv : hep - ph/9411342 ] ; + r. h. brandenberger , `` topological defects and structure formation , '' int .
j. mod .
phys . a * 9 * , 2117 ( 1994 ) [ arxiv : astro - ph/9310041 ] .
y. b. zeldovich , `` cosmological fluctuations produced near a singularity , '' mon . not .
soc . * 192 * , 663 ( 1980 ) ; + a. vilenkin , `` cosmological density fluctuations produced by vacuum strings , '' phys .
lett . * 46 * , 1169 ( 1981 ) [ erratum - ibid . *
46 * , 1496 ( 1981 ) ] .
r. jeannerot , `` a supersymmetric so(10 ) model with inflation and cosmic strings , '' phys . rev .
d * 53 * , 5426 ( 1996 ) [ arxiv : hep - ph/9509365 ] ; + r. jeannerot , j. rocher and m. sakellariadou , `` how generic is cosmic string formation in susy guts , '' phys .
d * 68 * , 103514 ( 2003 ) [ arxiv : hep - ph/0308134 ] .
s. sarangi and s. h. h. tye , `` cosmic string production towards the end of brane inflation , '' phys .
b * 536 * , 185 ( 2002 ) [ arxiv : hep - th/0204074 ] .
r. h. brandenberger and c. vafa , `` superstrings in the early universe , '' nucl
. phys .
b * 316 * , 391 ( 1989 ) . ; + a. nayeri , r. h. brandenberger and c. vafa , `` producing a scale - invariant spectrum of perturbations in a hagedorn phase of string cosmology , '' phys .
lett . * 97 * , 021302 ( 2006 ) [ arxiv : hep - th/0511140 ] ; + r. h. brandenberger , a. nayeri , s. p. patil and c. vafa , `` string gas cosmology and structure formation , '' int .
j. mod .
a * 22 * , 3621 ( 2007 ) [ hep - th/0608121 ] ; + r. h. brandenberger , `` string gas cosmology , '' arxiv:0808.0746 [ hep - th ] . t. w. b. kibble , `` some implications of a cosmological phase transition , '' phys .
* 67 * , 183 ( 1980 ) ; + t. w. b. kibble , `` phase transitions in the early universe , '' acta phys .
b * 13 * , 723 ( 1982 ) . c. dvorkin , m. wyman and w. hu , `` cosmic string constraints from wmap and the south pole telescope , '' phys .
d * 84 * , 123519 ( 2011 ) [ arxiv:1109.4947 [ astro-ph.co ] ] . c. l. bennett
_ et al . _ , `` first year wilkinson microwave anisotropy probe ( wmap ) observations : preliminary maps and basic results , '' astrophys .
j. suppl .
* 148 * , 1 ( 2003 ) [ arxiv : astro - ph/0302207 ] . j. e. ruhl _ et al . _
[ the spt collaboration ] , `` the south pole telescope , '' proc .
spie int .
soc . opt .
5498 * , 11 ( 2004 ) [ arxiv : astro - ph/0411122 ] .
j. urrestilla , n. bevis , m. hindmarsh and m. kunz , `` cosmic string parameter constraints and model analysis using small scale cosmic microwave background data , '' jcap * 1112 * , 021 ( 2011 ) [ arxiv:1108.2730 [ astro-ph.co ] ] .
a. kosowsky [ the act collaboration ] , `` the atacama cosmology telescope project : a progress report , '' new astron .
rev . * 50 * , 969 ( 2006 ) [ arxiv : astro - ph/0608549 ] .
l. pogosian , s. h. h. tye , i. wasserman and m. wyman , `` observational constraints on cosmic string production during brane inflation , '' phys .
d * 68 * , 023506 ( 2003 ) [ erratum - ibid .
d * 73 * , 089904 ( 2006 ) ] [ arxiv : hep - th/0304188 ] ; + m. wyman , l. pogosian and i. wasserman , `` bounds on cosmic strings from wmap and sdss , '' phys .
d * 72 * , 023513 ( 2005 ) [ erratum - ibid .
d * 73 * , 089905 ( 2006 ) ] [ arxiv : astro - ph/0503364 ] ; + a. a. fraisse , `` limits on defects formation and hybrid inflationary models with three - year wmap observations , '' jcap * 0703 * , 008 ( 2007 ) [ arxiv : astro - ph/0603589 ] ; + u. seljak , a. slosar and p. mcdonald , `` cosmological parameters from combining the lyman - alpha forest with cmb , galaxy clustering and sn constraints , '' jcap * 0610 * , 014 ( 2006 ) [ arxiv : astro - ph/0604335 ] ; + r. a. battye , b. garbrecht and a. moss , `` constraints on supersymmetric models of hybrid inflation , '' jcap * 0609 * , 007 ( 2006 ) [ arxiv : astro - ph/0607339 ] ; + r. a. battye , b. garbrecht , a. moss and h. stoica , `` constraints on brane inflation and cosmic strings , '' jcap * 0801 * , 020 ( 2008 ) [ arxiv:0710.1541 [ astro - ph ] ] ; + n. bevis , m. hindmarsh , m. kunz and j. urrestilla , `` cmb power spectrum contribution from cosmic strings using field - evolution simulations of the abelian higgs model , '' phys .
d * 75 * , 065015 ( 2007 ) [ arxiv : astro - ph/0605018 ] ; + n. bevis , m. hindmarsh , m. kunz and j. urrestilla , `` fitting cmb data with cosmic strings and inflation , '' phys .
lett . *
100 * , 021301 ( 2008 ) [ astro - ph/0702223 [ astro - ph ] ] ; + r. battye and a. moss , `` updated constraints on the cosmic string tension , '' phys .
d * 82 * , 023521 ( 2010 ) [ arxiv:1005.0479 [ astro-ph.co ] ] .
r. h. brandenberger , a. -c .
davis and m. hindmarsh , `` baryogenesis from collapsing topological defects , '' phys .
b * 263 * , 239 ( 1991 ) ; + r. h. brandenberger , a. -c .
davis and m. trodden , `` cosmic strings and electroweak baryogenesis , '' phys .
b * 335 * , 123 ( 1994 ) [ hep - ph/9403215 ] ; + r. h. brandenberger , a. -c .
davis , t. prokopec and m. trodden , `` local and nonlocal defect mediated electroweak baryogenesis , '' phys .
d * 53 * , 4257 ( 1996 ) [ hep - ph/9409281 ] .
r. h. brandenberger and x. -m .
zhang , `` anomalous global strings and primordial magnetic fields , '' phys .
d * 59 * , 081301 ( 1999 ) [ hep - ph/9808306 ] .
j. h. macgibbon and r. h. brandenberger , `` gamma - ray signatures from ordinary cosmic strings , '' phys .
d * 47 * , 2283 ( 1993 ) [ astro - ph/9206003 ] ; + p. bhattacharjee , c. t. hill and d. n. schramm , `` grand unified theories , topological defects and ultrahigh - energy cosmic rays , '' phys .
lett . * 69 * , 567 ( 1992 ) ; + p. bhattacharjee and n. c. rana , `` ultrahigh - energy particle flux from cosmic strings , '' phys .
b * 246 * , 365 ( 1990 ) ; + j. h. macgibbon and r. h. brandenberger , `` high - energy neutrino flux from ordinary cosmic strings , '' nucl .
b * 331 * , 153 ( 1990 ) .
r. brandenberger , y. -f .
cai , w. xue and x. -m .
zhang , `` cosmic ray positrons from cosmic strings , '' arxiv:0901.3474 [ hep - ph ] .
r. h. brandenberger , r. j. danos , o. f. hernandez and g. p. holder , `` the 21 cm signature of cosmic string wakes , '' jcap * 1012 * , 028 ( 2010 ) [ arxiv:1006.2514 [ astro-ph.co ] ] . n. kaiser and a. stebbins , `` microwave anisotropy due to cosmic strings , '' nature * 310 * , 391 ( 1984 ) . j. r. gott , iii , `` gravitational lensing effects of vacuum strings : exact solutions , '' astrophys .
j. * 288 * , 422 ( 1985 ) . j. silk and a. vilenkin , `` cosmic strings and galaxy formation , '' phys .
lett . * 53 * , 1700 ( 1984 ) ; + m. rees , baryon concentrations in string wakes at @xmath90 : implications for galaxy formation and large - scale structure , " mon . not .
. soc . * 222 * , 27p ( 1986 ) ; + t. vachaspati , `` cosmic strings and the large - scale structure of the universe , '' phys .
lett . * 57 * , 1655 ( 1986 ) ; + a. stebbins , s. veeraraghavan , r. h. brandenberger , j. silk and n. turok , `` cosmic string wakes , '' astrophys .
j. * 322 * , 1 ( 1987 ) . l. dolan and r. jackiw , `` symmetry behavior at finite temperature , '' phys .
d * 9 * , 3320 ( 1974 ) .
d. a. kirzhnits and a. d. linde , `` macroscopic consequences of the weinberg model , '' phys .
b * 42 * , 471 ( 1972 ) .
r. h. brandenberger , `` quantum field theory methods and inflationary universe models , '' rev .
phys . * 57 * , 1 ( 1985 ) .
n. turok and r. h. brandenberger , `` cosmic strings and the formation of galaxies and clusters of galaxies , '' phys .
d * 33 * , 2175 ( 1986 ) ; + h. sato , `` galaxy formation by cosmic strings , '' prog .
phys . * 75 * , 1342 ( 1986 ) ; + a. stebbins , `` cosmic strings and cold matter '' , ap .
j. ( lett . ) * 303 * , l21 ( 1986 ) .
e. p. s. shellard , `` cosmic string interactions , '' nucl .
b * 283 * , 624 ( 1987 ) .
a. albrecht and n. turok , `` evolution of cosmic strings , '' phys .
lett . * 54 * , 1868 ( 1985 ) ; + d. p. bennett and f. r. bouchet , `` evidence for a scaling solution in cosmic string evolution , '' phys .
lett . * 60 * , 257 ( 1988 ) ; + b. allen and e. p. s. shellard , `` cosmic string evolution : a numerical simulation , '' phys .
* 64 * , 119 ( 1990 ) ; + c. ringeval , m. sakellariadou and f. bouchet , `` cosmological evolution of cosmic string loops , '' jcap * 0702 * , 023 ( 2007 ) [ arxiv : astro - ph/0511646 ] ; + v. vanchurin , k. d. olum and a. vilenkin , `` scaling of cosmic string loops , '' phys .
d * 74 * , 063527 ( 2006 ) [ arxiv : gr - qc/0511159 ] ; + j. j. blanco - pillado , k. d. olum and b. shlaer , `` large parallel cosmic string simulations : new results on loop production , '' phys .
d * 83 * , 083514 ( 2011 ) [ arxiv:1101.5173 [ astro-ph.co ] ] .
j. magueijo , a. albrecht , d. coulson and p. ferreira , `` doppler peaks from active perturbations , '' phys .
lett . * 76 * , 2617 ( 1996 ) [ arxiv : astro - ph/9511042 ] ; +
u. l. pen , u. seljak and n. turok , `` power spectra in global defect theories of cosmic structure formation , '' phys .
lett . * 79 * , 1611 ( 1997 ) [ arxiv : astro - ph/9704165 ] ; + l. perivolaropoulos , `` spectral analysis of microwave background perturbations induced by cosmic strings , '' astrophys . j. * 451 * , 429 ( 1995 ) [ arxiv : astro - ph/9402024 ] . j. h. traschen , n. turok and r. h. brandenberger , `` microwave anisotropies from cosmic strings , '' phys .
d * 34 * , 919 ( 1986 ) ; + r. h. brandenberger and n. turok , `` fluctuations from cosmic strings and the microwave background , '' phys . rev .
d * 33 * , 2182 ( 1986 ) .
d. mauskopf _ et al . _
[ boomerang collaboration ] , `` measurement of a peak in the cosmic microwave background power spectrum from the north american test flight of boomerang , '' astrophys . j. * 536 * , l59 ( 2000 ) [ arxiv : astro - ph/9911444 ] . a. vilenkin , `` gravitational field of vacuum domain walls and strings , '' phys . rev .
d * 23 * , 852 ( 1981 ) ; + r. gregory , `` gravitational stability of local strings , '' phys .
lett . * 59 * , 740 ( 1987 ) .
j. c. r. magueijo , `` inborn metric of cosmic strings , '' phys .
d * 46 * , 1368 ( 1992 ) .
l. perivolaropoulos , `` cobe versus cosmic strings : an analytical model , '' phys .
b * 298 * , 305 ( 1993 ) [ arxiv : hep - ph/9208247 ] ; + l. perivolaropoulos , `` statistics of microwave fluctuations induced by topological defects , '' phys .
d * 48 * , 1530 ( 1993 ) [ arxiv : hep - ph/9212228 ] .
j. canny , `` a computational approach to edge detection '' , ieee trans .
pattern analysis and machine intelligence * 8 * , 679 ( 1986 ) .
r. j. danos and r. h. brandenberger , `` canny algorithm , cosmic strings and the cosmic microwave background , '' int .
j. mod .
d * 19 * , 183 ( 2010 ) [ arxiv:0811.2004 [ astro - ph ] ] .
s. amsel , j. berger and r. h. brandenberger , `` detecting cosmic strings in the cmb with the canny algorithm , '' jcap * 0804 * , 015 ( 2008 ) [ arxiv:0709.0982 [ astro - ph ] ] .
a. stewart and r. brandenberger , `` edge detection , cosmic strings and the south pole telescope , '' jcap * 0902 * , 009 ( 2009 ) [ arxiv:0809.0865 [ astro - ph ] ] .
y. .b .
zeldovich , `` gravitational instability : an approximate theory for large density perturbations , '' astron .
astrophys . * 5 * , 84
a. sornborger , r. h. brandenberger , b. fryxell and k. olson , `` the structure of cosmic string wakes , '' astrophys . j. * 482 * , 22 ( 1997 ) [ astro - ph/9608020 ] .
r. h. brandenberger , n. kaiser and n. turok , `` dissipationless clustering of neutrinos around a cosmic string loop , '' phys .
d * 36 * , 2242 ( 1987 ) ; + r. h. brandenberger , n. kaiser , d. n. schramm and n. turok , `` galaxy and structure formation with hot dark matter and cosmic strings , '' phys .
lett . * 59 * , 2371 ( 1987 ) .
r. h. brandenberger , l. perivolaropoulos and a. stebbins , `` cosmic strings , hot dark matter and the large scale structure of the universe , '' int .
j. mod .
phys . a * 5 * , 1633 ( 1990 ) ; + l. perivolaropoulos , r. h. brandenberger and a. stebbins , `` dissipationless clustering of neutrinos in cosmic string induced wakes , '' phys . rev .
d * 41 * , 1764 ( 1990 ) .
m. pagano and r. brandenberger , `` the 21 cm signature of a cosmic string loop , '' jcap * 1205 * , 014 ( 2012 ) [ arxiv:1201.5695 [ astro-ph.co ] ] .
r. j. danos , r. h. brandenberger and g. holder , `` a signature of cosmic strings wakes in the cmb polarization , '' phys .
d * 82 * , 023513 ( 2010 ) [ arxiv:1003.0905 [ astro-ph.co ] ] .
o. dore , g. holder , m. alvarez , i. t. iliev , g. mellema , u. -l .
pen and p. r. shapiro , `` the signature of patchy reionization in the polarization anisotropy of the cmb , '' phys .
d * 76 * , 043002 ( 2007 ) [ astro - ph/0701784 [ astro - ph ] ] .
r. h. brandenberger , `` is the spectrum of gravitational waves the holy grail of inflation ?
, '' arxiv:1104.3581 [ astro-ph.co ] .
r. h. brandenberger , a. nayeri , s. p. patil and c. vafa , `` tensor modes from a primordial hagedorn phase of string cosmology , '' phys .
lett . * 98 * , 231302 ( 2007 ) [ hep - th/0604126 ] .
r. khatri and b. d. wandelt , `` cosmic ( super)string constraints from 21 cm radiation , '' phys .
lett . * 100 * , 091302 ( 2008 ) [ arxiv:0801.4406 [ astro - ph ] ] ; + a. berndsen , l. pogosian and m. wyman , `` correlations between 21 cm radiation and the cmb from active sources , '' arxiv:1003.2214 [ astro-ph.co ] .
s. furlanetto , s. p. oh and f. briggs , `` cosmology at low frequencies : the 21 cm transition and the high - redshift universe , '' phys .
rept . *
433 * , 181 ( 2006 ) [ arxiv : astro - ph/0608032 ] .
o. f. hernandez and r. h. brandenberger , `` the 21 cm signature of shock heated and diffuse cosmic string wakes , '' jcap * 1207 * , 032 ( 2012 ) [ arxiv:1203.2307 [ astro-ph.co ] ] .
e. mcdonough and r. h. brandenberger , `` searching for signatures of cosmic string wakes in 21 cm redshift surveys using minkowski functionals , '' arxiv:1109.2627 [ astro-ph.co ]
. o. f. hernandez , y. wang , r. brandenberger and j. fong , `` angular 21 cm power spectrum of a scaling distribution of cosmic string wakes , '' jcap * 1108 * , 014 ( 2011 ) [ arxiv:1104.3337 [ astro-ph.co ] ] .
t. vachaspati and a. vilenkin , `` gravitational radiation from cosmic strings , '' phys .
d * 31 * , 3052 ( 1985 ) ; + r. l. davis , nucleosynthesis problems for string models of galaxy formation " , phys .
* b 161 * , 285 ( 1985 ) .
r. h. brandenberger , a. albrecht and n. turok , `` gravitational radiation from cosmic strings and the microwave background , '' nucl .
b * 277 * , 605 ( 1986 ) .
t. damour and a. vilenkin , `` gravitational wave bursts from cosmic strings , '' phys .
* 85 * , 3761 ( 2000 ) [ gr - qc/0004075 ] .
r. h. brandenberger , `` on the decay of cosmic string loops , '' nucl .
b * 293 * , 812 ( 1987 ) .
b. shlaer , a. vilenkin and a. loeb , `` early structure formation from cosmic string loops , '' jcap * 1205 * , 026 ( 2012 ) [ arxiv:1202.1346 [ astro-ph.co ] ] . | cosmic strings are predicted in many models beyond the standard model of particle physics . in models which admit strings , a network of strings
will inevitably be formed in a phase transition in the early universe and will persist to the present time .
strings leave behind distinctive features in cosmology . searching for these signatures in new observational windows
provides a way to constrain particle physics at the high energy scale and is thus complementary to searches for new physics at the low energy end , for example at the lhc .
specifically , i will discuss signatures of cosmic strings in cosmic microwave background polarization maps and in 21 cm redshift surveys . |
the industrial demand for high - assurance software opens a window of opportunity for mathematically based development methods , able to design complex systems at ever - increasing levels of reliability and security .
this paper s contribution is placed at a specific corner of the broad landscape of formal methods for software development : that of _ algebraic specification _
@xcite , a family of methods which , having played a pioneering role , constitutes at present a large and mature body of knowledge and active research .
such methods have a double origin .
on the one hand they can be traced back to early work on data abstraction and modular decomposition of programs @xcite . on the other hand , to research on semantics of program specifications building on results from algebraic logic and model theory . especially relevant in this respect is the original work of the so - called adj group @xcite whose initial algebra semantics was the first , full formal approach to software development put forward . this double origin ,
temporally located around mid seventies , is not surprising : _ compositionality _ is both a basic requirement in program development and a major asset in algebraic semantics .
the whole area flourished rapidly from the outset : not only different approaches to semantics ( final , observational , loose ) emerged , but also the initial tie to many - sorted equational logic was soon extended , first to conditional - equational logic , and later to order - sorted , partial and full first - order among other variants .
the emergence of the first effective algebraic specification languages obj @xcite and clear @xcite overlaps another major development : the introduction of institutions by j. goguen and r. burstall @xcite
. institution theory , which develops model theory independent of the underlying logical system , made possible to decouple specification methodology from the particularities of whatever semantics one may consider more suitable to a specific problem @xcite .
moreover , although for a long time the impact of these methods in industry has been limited , a successful effort has been made in the last 15 years towards convergence on generic frameworks with suitable tool support .
the compass and , later , the cofi initiative @xcite , which lead to the development of casl @xcite , are relevant milestones in this process . besides casl , cafeobj @xcite and maude @xcite
are currently used in several industrial applications and tool development . actually , research in such methods , either at a foundational or methodological level , found applications in new , unsuspected areas for example , in documenting service interfaces @xcite , characterising contracts in contract - based programming @xcite or test generation for software composition @xcite .
for the working software engineer , a software component is documented by an _ interface _ , which provides a language through which it interacts with its environment , and a _ specification _ of the intended meaning of the services provided .
this specification is implemented by a concrete piece of software respecting the specified semantics .
algebraic specification methods build on the observation that these somehow vague concepts from software engineering can be framed rigorously in terms of well - known mathematical notions .
thus , an _ interface _ corresponds to a _ signature _
, i.e. , a set of names for the relevant types , called _ sorts _ , and a family of service or operation names , classified by their arity and input - output sorts .
a signature generates a formal language , giving a rigorous meaning to what we have called before the component s interaction language .
once fixed a signature , a _ specification _ describes a _ class of models _ for that signature , and an _ implementation _ identifies a specific model within such a class . if functions provide suitable abstractions of the services offered by a software component , this analogy can be made even more concrete by identifying _
interfaces _ with _ algebraic signatures _ , ( denotations of ) _ specifications _ with _ classes of algebras _ , and _ implementations _ with specific _
algebras_. the analogy extends to the entire software development process along which components are _ refined _ by incrementally adding detail and reducing under - specification .
formally , this is a process of structural transformations witnessed by _ signature morphisms _ , which map functionally sorts and operations from a signature to another respecting the sort translation of functional types . in such a context
this paper raises and discusses the following question : _ can more flexible notions of refinement emerge from replacing signature morphisms by some weaker notion of transformation ? _ the quest for _ weaker notions of transformation _ lead us to a different setting , that of algebraic logic @xcite .
the key conceptual tool is that of a _ deductive system _ , i.e. , a formal language generated by a _ signature _ , and a _ consequence relation_. interrelating such systems , through maps connecting logical properties , has been studied from early in the last century .
such maps were called _ translations _ and investigated as part of an ambitious programme addressing tools to handle the multiplicity of logics . as a result
, several intuitive notions of translation are scattered in the literature .
many logicians tailored the notion , for their own purposes , to relate specific logics and to obtain specific results . in general , however , a translation is regarded as a map between sets of formulas of different logics such that the image of a theorem is still a theorem .
they were used originally to clarify the relationship between classical and constructive logics .
our starting point is the observation that specifications describe classes of models and those can be naturally associated to deductive systems .
then , translations that both reflect and preserve consequence relations seem interesting candidates to witness weaker forms of refinement . in this paper
we will single out a specific sort of translations based on _ multifunctions _
, , functions mapping an element to a set of elements .
such translations are called _ interpretations _ and constitute a central tool in the study of equivalent algebraic semantics ( see , , @xcite ) .
a paradigmatic example is the interpretation of the _ classical propositional calculus _ into the _ equational theory of boolean algebras _ ( cf .
* example 4.1.2 ) ) .
this paper explores interpretations between the deductive systems corresponding to classes of models of specifications as possible witnesses of refinement steps .
the notion seems able to capture a number of transformations which are difficult to deal with in classical terms .
examples include data encapsulation and the decomposition of operations into atomic transactions .
it also seems promising in the context of new , emerging computing paradigms which entail the need for more flexible approaches to what counts as a valid transformation along the development process ( see , for example , @xcite ) . in this context
, the contribution of this paper , which combines and extends previous results by the authors reported in @xcite and @xcite , is twofold . on the one hand it puts forward a detailed characterisation of refinement witnessed by interpretations , referred in the sequel as _ refinement by interpretation _ , and exemplifies its potentialities in a number of small , yet illustrative examples . on the other hand
, it renders the whole approach at a sufficiently abstract setting to be applicable over _ arbitrary _ , technically _
@xmath0-dimensional _ , deductive systems .
the dimension fixes the kind of relationship between terms one is interested in .
dimension 2 , for example , encompasses equations , regarded as instances of a binary predicate asserting , for example , term equality , bisimilarity , or observational equivalence .
similarly , a unary predicate asserting the validity of a formula is enough to represent a proposition , leading to 1-dimensional deductive systems . refinement by interpretation in a general , @xmath0-dimensional setting provides a suitable context to deal simultaneously with deductive systems arising from classes of models presented in different logics , for example , as a set of equations , propositions or modal formulas . once stated the paper s contributions , it is important to clearly delimit its scope .
first of all it should be stressed that the focus of this paper is not placed on _ specifications _ , understood as syntactic entities which describe in a structured , modular way classes of models , but rather on the _ classes of models _ themselves , which constitute their denotations .
deductive systems , the basic tool in our approach , correspond to such classes .
this means that the whole area of _ specification structuring _
@xcite is , for the moment , left out .
our approach is not concerned with the fact that specifications describing the relevant classes of models are _ flat _ ,
i.e. given by a finite set of sentences , or _ structured _ , _
i.e. _ , built by systematic application of a number of operators , such as _ union _ , _ translation _ or _ hiding _ , all of them well characterised in the literature and implemented in a number of computer - supported modelling tools .
this does not deny the fundamental importance of specification structuring .
research on this topic started with the introduction of clear @xcite , by the end of the seventies , and its role can not be underestimated .
actually , the recursive definition of structured specifications provides basic modular procedures for software composition and architecture .
moreover structuring operations allows one to go beyond the specification power of simple , unstructured specifications @xcite . clearly , the approach proposed in this paper can be tuned to specification refinement in a strict sense . in a recent publication
@xcite we showed how _ refinement by interpretation _ can be lifted to the level of structured specifications with the usual operators mentioned above .
we believe , however , that by focussing on _ classes of models _ this piece of research acquires a broader scope of application and is worth on its own . in particular , it pays off when dealing with requirements that can not be properly formalised in a specification ( for example , the property that a controller has a finite number of states ) .
note this does not entail any loss of expressivity : for each specification , one may recursively compute its denotation ( a signature and a class of models ) and work directly with them .
on the other hand , the discussion on which operators should be considered in a specification calculus is still active .
for example , very recently , reference @xcite introduced two new operators for specification composition in order to deal with non - protecting importation modes .
this further justifies the relevance of a semantic approach as proposed here .
another concept to make precise is _ refinement_. the word
is taken here in the broad sense of a transformation mapping an abstract to a more concrete class of models .
as such classes are represented by deductive systems , a refinement will map a deductive system into another , while preserving the consequence relation .
this is in line with the usual meaning of refinement : all requirements stated at the original level are still valid after refinement .
moreover , it will be shown in the paper that _
refinement by interpretation _ between classes of models boils down to the standard notion of refinement as inclusion of classes of models whenever the witnessing interpretation is simply an identity . finally , a note on the expressivity of deductive systems .
actually , deductive systems can play a double role representing both logics , on top of which all the specification machinery can be developed , and classes of models as discussed above .
clearly , any institution induces , for each signature , a deductive system through its satisfaction relation , and conversely , deductive systems may be viewed as special cases of institutions as discussed in @xcite .
for example , a deductive system may represent the class of boolean algebras ; but the latter can also be specified as a theory in a suitable institution .
another example is provided by modal logics which can be regarded as both a deductive system or a theory in the first - order ( fol ) institution through the standard translation .
this provides a uniform view of seemingly different settings , enabling us to discuss how essentially the same conceptual tool , that of _ logical interpretation _ , can be used to interrelate logics and refine classes of models both regarded as deductive systems .
reciprocally , deductive system can be endowed with an algebraic semantics , as discussed in section [ sc : algsem ] .
note that the quest for such a uniform representation of logics and logical theories pops up in other contexts , namely on the design of logical frameworks .
a prime example is provided by the logics - as - theories approach proposed by f. rabe in @xcite , resorting to a type theoretical framework , and further developed in the context of the latin project @xcite .
section [ sc : pre ] introduces _
@xmath0-dimensional deductive systems _ and their semantics following @xcite .
this paves the way to the formulation of refinement by interpretation in a general setting in sections [ sc : gen ] and [ sc : rbi2 ] . before that , however , in section [ sc : rbi1 ] , the approach is instantiated for the case of algebraic specifications over the institution of horn clause logic .
this is a popular framework for algebraic specifications which not only deserves attention on its own , but also provides a simpler setting to build up intuitions .
finally , section [ sc : conc ] concludes and suggests some problems deserving further attention .
_ specifications of complex systems resort to different logics , and even to their combination .
consequently a characterisation of _ refinement by interpretation _ needs to be orthogonal to whatever logic is used in specifications .
this is achieved through the notion of @xmath0-_dimensional deductive systems _ ,
of which the equational case is just an instance for @xmath1 .
this section reviews such systems and their semantics , following @xcite , to provide the background for the sections to follow . _ roughly speaking , a deductive system is a general mathematical tool to reason about formulas in a language generated by a signature .
formally , it is defined as a pair @xmath2 , where @xmath3 is a signature and @xmath4 is a substitution - invariant consequence relation between sets of formulas and individual formulas .
clearly , any standard sentential logical system , defined in the usual way by a set of axioms and a set of inference rules ( for instance , classical and intuitionist propositional calculus , referred to in the sequel as cpc and ipc , respectively ) , is a deductive system .
first order logic can also be formulated as a deductive system @xcite , which shows how broad the concept is .
the formal notion of a deductive system , in this abstract perspective , was originally considered by ukasiewicz and tarski @xcite and intensely studied , from an algebraic point of view , by many logicians .
this gave rise to a new , extremely relevant area of mathematics , that of _ abstract algebraic logic _
@xcite .
although in some literature on algebraic logic this substitution - invariant consequence relation has been called a _ logic _ ( cf .
@xcite ) , we adopt along the paper the designation of _ deductive system _ used by blok and pigozzi @xcite .
this terminology allows us to distinguish this concept from the habitual meaning logic has in computer science , typically understood as an abstract framework to express specifications and often abstracted as an _ institution _
@xcite or a @xmath5-institution @xcite .
as mentioned above , translation maps were introduced in the early 20th century as a means to interrelate deductive systems .
they were first used to understand the relationship between classical and constructive logics .
the well - known gdel translation of classical logic into intuitionistic logic has inspired disperse works on comparing different logics by means of translations .
illustrative examples include the works of kolmogorov @xcite , glivenko @xcite , and gdel @xcite involving classical , intuitionist , and modal logics . to the best of our knowledge
the first general definition of translation between deductive systems is due to prawitz and malmns @xcite .
more recently , wjcicki @xcite presented a systematic study of translations between logics , focussing on inter - relations between sentential logics .
and the quest goes on ( cf .
at the turn of the century , silva , dottaviano and sette @xcite proposed a general definition of translation between logics as maps preserving consequence relations .
then , feitosa and dottaviano studied intensively the subclass of translations that preserve and reflect consequence relations and coined the name _ conservative translation _
@xcite .
conservative translations which are able to relate a formula to a set of formulas , and are therefore defined as _ multifunctions _ , are called _ interpretations_. those which commute with substitutions were originally used in abstract algebraic logic to define a very important class of deductive systems referred to as _ algebraisable _
in particular , they abstract the strict relationship between classic propositional logic and the class of boolean algebras . a deductive system is said to be _
algebraisable _ whenever there exists a class @xmath6 of algebras such that the consequence relation induced by @xmath6 is equivalent to the consequence in the deductive system .
such an equivalence was originally defined by means of two mutually inverse interpretations . since then , this link between logic and universal algebra has been successfully explored . in particular , for an algebraisable @xmath7 , properties of @xmath7 can be related to algebraic properties of its equivalent algebraic semantics .
this kind of results , of which many examples exist , are often called _
bridge theorems_. in order to broaden the spectrum of application of deductive systems , blok and pigozzi et al .
@xcite introduced consequence relations over @xmath0-tuples of formulas , for @xmath0 a non - zero natural number .
_ @xmath0-deductive systems _ , the result of this generalisation , are the higher dimensional version of the well known sentential logics .
their theory provides a unified treatment for several such as the ones corresponding to assertional , equational , and inequational logics .
this generalisation also allows to regard interpretations witnessing algebraisability as a special kind of translations between @xmath0-deductive systems . an _ equation _ , represented in this paper by a formal expression @xmath8 ,
can be regarded as a pair of terms ( or formulas ) @xmath9 .
this , in turn , is an instance of a binary predicate standing for the equality of two terms .
similarly , a _
unary predicate _ asserting the validity of a formula is enough to represent a proposition .
the first observation leads to what will be characterised in the sequel as a @xmath10-dimensional deductive system , of which the equational case is a particular instance .
the second corresponds , roughly speaking , to sentential logics in a quite broad sense ( to include , for example , first - order predicate logic when suitably formalised ) . in general , adding a @xmath0-ary predicate to a strict universal horn theory without equality , gives rise to a representation of a @xmath0-dimensional , thus providing a suitable context to deal simultaneously with different specification logics .
we go even a step further considering @xmath0-deductive systems over _ many sorted _ languages , because , in general , software systems manipulate several sorts of data .
almost all notions can be formulated in this broader setting as discussed later .
note there are other generalisations that allow the reuse of arguments and tools from abstract algebraic logic in computer science contexts .
hidden logics , introduced by pigozzi and martins in @xcite ( see also @xcite and @xcite ) are a prime example .
they have been efficiently used to develop specification and verification methodologies for object oriented software systems .
examples include the boolean logics , , 1-dimensional multi - sorted logics with bool as the only visible sort , and equality - test operations for some of the hidden sorts in place of equality predicates .
the syntactic support for @xmath0-dimensional deductive systems is that of a @xmath0-term .
let @xmath11 be a signature and @xmath12 a @xmath13-sorted set of variables .
@xmath0-term _ of sort @xmath14 over signature @xmath3 is a sequence of @xmath0 @xmath3-terms , all of the same sort @xmath14 , @xmath15 , abbreviated to @xmath16 whenever references to sorts can be omitted .
@xmath0-variable of sort @xmath14 _ is a sequence of @xmath0 variables all of the same sort @xmath14 .
@xmath17 is the sorted set of all @xmath0-terms over @xmath3 with variables in @xmath18 , i.e. , @xmath19 where @xmath20 is the set of all terms over @xmath3 , of sort @xmath14 , with variables in @xmath18 . whenever each @xmath21 , for each sort @xmath14 in @xmath3 , is non empty
, their union acts as the carrier of the @xmath3-_term algebra _ freely generated from @xmath18 , which we denote by @xmath22 .
a _ substitution on _
@xmath22 is just an endomorphism over @xmath22 .
let us fix some notation and terminology : if @xmath23 is a @xmath0-term over @xmath3 , @xmath24 is a @xmath3-algebra , and @xmath25 , we denote by @xmath26 the value @xmath16 takes in @xmath24 when variables @xmath27 are instantiated respectively by @xmath28 .
more precisely , if @xmath29 then @xmath30 , where @xmath31 is any homomorphism from @xmath22 to @xmath24 such that @xmath32 for all @xmath33 .
let @xmath34 be an arbitrary but fixed family of countably infinite disjoint sets @xmath35 of variables of sort @xmath36 .
following a typical procedure in similar contexts ( , @xcite ) , we assume in the sequel @xmath37 fixed for each set of sorts @xmath13 and large enough to contain all variables needed
. symbols of variables are obviously disjoint of any other symbol in the signature . as usual in sentential logic frameworks
, we will refer to formulas ( @xmath0-formulas ) as synonymous to terms ( @xmath0-terms respectively ) .
accordingly , we will denote @xmath38 by @xmath39 .
moreover , for each nonzero natural number @xmath0 , given a sorted signature @xmath3 , a _
@xmath0-formula _ of sort @xmath14 over @xmath3 is any element of @xmath40 .
the set of all @xmath0-formulas will be denoted by @xmath41 . also note that an @xmath13-sorted subset @xmath42 of @xmath0-formulas is identified with the unsorted set @xmath43 , which allows writing @xmath44 to mean that @xmath45 , for some sort @xmath14 .
a set @xmath46 is said to be _ globally finite _ when @xmath47 is a finite set for each sort @xmath14 of @xmath3 , equal to @xmath48 except for a finite number of them , , @xmath43 is finite . in this
setting , a @xmath0-dimensional deductive system is defined as a substitution - invariant consequence relation on the set of @xmath0-formulas .
the following definition generalises the one due to w. blok and d. pigozzi @xcite for the one - sorted case .
[ df : klogic ] a _ @xmath0-dimensional deductive system _ is a pair @xmath49 , where @xmath3 is a sorted signature and @xmath50 is a relation such that , for all @xmath51 , the following conditions hold : 1 .
@xmath52 for each @xmath53 ; 2 .
if @xmath54 , and @xmath55 for each @xmath53 , then @xmath56 ; 3 . if @xmath54 , then @xmath57 for every substitution @xmath58 .
a @xmath0- is _ specifiable _ if @xmath59 is _ compact _ ( or _ finitary _ in the terminology of abstract algebraic logic ) , , if , whenever @xmath54 , there exists a globally finite subset @xmath60 of @xmath42 such that @xmath56 . the relation @xmath59 , abbreviated to @xmath4 whenever @xmath61 is clear from the context , is called _ the consequence relation of _ @xmath61 .
it is easy to see that , for any @xmath51 , @xmath62 and @xmath63 imply @xmath64 .
every consequence relation @xmath4 has a natural extension to a relation between sets of @xmath0-formulas , also denoted by @xmath4 , defined by @xmath65 if @xmath66 for each @xmath67 .
finally , the relation of _ interderivability _ between sorted sets is defined by @xmath68 if @xmath69 and @xmath70 .
we abbreviate @xmath71 and @xmath72 by @xmath73 and @xmath74 , respectively .
let @xmath61 be a ( not necessarily specifiable ) @xmath0- .
thm _ of @xmath61 is a @xmath0-formula @xmath16 such that @xmath75 , , @xmath76 .
the set of all theorems is denoted by @xmath77 .
an _ inference rule _ is a pair @xmath78 where @xmath79 a globally finite set of @xmath0-formulas and @xmath16 a @xmath0-formula , usually represented as @xmath80 a rule such as ( [ sequent ] ) is said to be a _
derivable rule _ of @xmath61 if @xmath81 .
a set of @xmath0-formulas @xmath82 closed under the consequence relation , , such that @xmath83 implies @xmath84 , is called a _ theory _ of @xmath61 .
the set of all theories is denoted by @xmath85 ; it forms a complete lattice under set - theoretic inclusion , which is algebraic if @xmath61 is specifiable .
given any set of @xmath0-formulas @xmath42 , the set of all consequences of @xmath42 , in symbols @xmath86 , is the smallest theory that contains @xmath42 .
it is easy to see that @xmath87 .
often , a specifiable @xmath0- is presented in the so - called hilbert style , , by a set of axioms ( @xmath0-formulas ) and inference rules .
we say that a @xmath0-formula @xmath88 is _ directly derivable _ from a set @xmath42 of @xmath0-formulas by a rule such as ( [ sequent ] ) if there is a substitution @xmath89 such that @xmath90 and @xmath91 . given a set @xmath92 of @xmath0-formulas and a set @xmath93 of inference rules
, we say that @xmath88 is _ derivable _ from @xmath42 by @xmath92 and @xmath93 , in symbols @xmath94 , if there is a _ proof _ , , a finite sequence of @xmath0-formulas , @xmath95 such that @xmath96 , and for each @xmath33 , one of the following conditions holds : 1 .
@xmath97 , 2 .
@xmath98 is a substitution instance of a @xmath0-formula in @xmath92 , 3 .
@xmath98 is directly derivable from @xmath99 by one of the inference rules in @xmath93 .
it is clear that @xmath100 is a specifiable @xmath0- .
moreover , a @xmath0-deductive system @xmath61 is specifiable iff there exist possibly infinite sets @xmath92 and @xmath93 , of axioms and inference rules respectively , such that , for any @xmath0-formulas @xmath88 and any set @xmath42 of @xmath0-formulas , @xmath101 iff @xmath102 ( see @xcite for the one sorted case ) .
this justifies that all the examples of specifiable deductive systems introduced in this paper are presented by a set of axioms and inference rules .
if a deductive system @xmath61 is equal to @xmath100 , for some sets @xmath92 and @xmath93 with @xmath103 , @xmath61 is said to be _
finitely axiomatisable_. a @xmath0- @xmath104 is an _ extension _ of the @xmath0- @xmath105 if @xmath106 whenever @xmath107 for all @xmath108 ( , @xmath109 ) .
a @xmath0- @xmath110 is _ an extension by axioms and rules of _ a specifiable @xmath0- @xmath61 if it can be axiomatised by adding axioms and inference rules to the axioms and rules of some axiomatisation of @xmath61 .
typical examples of @xmath0-deductive systems are the ones induced by algebraic specifications ( see section [ sc : rbi1 ] ) . given a signature @xmath3 , they are defined over pairs of @xmath3-terms @xmath9 , representing equations @xmath111 , and have therefore dimension @xmath1 . similarly , @xmath3-conditional equations can be taken as pairs @xmath112 where @xmath42 is a globally finite subset of @xmath113 and @xmath114 . as a particular case ,
an equation @xmath8 is a conditional equation without premisses , @xmath115 . in general
, a conditional equation @xmath116 will be written as @xmath117 . in the sequel
we will often use @xmath118 , instead of @xmath113 , for the set of all equations over @xmath37 , and , similarly , @xmath119 , for the set of all @xmath3-conditional equations over @xmath37 .
let @xmath120 and @xmath24 be a @xmath3-algebra .
we write @xmath121 if , for every homomorphism @xmath122 , @xmath123 for @xmath124 , @xmath125 is abbreviated to @xmath126 .
an equation @xmath127 is an _ identity _ of @xmath24 if @xmath128 .
similarly , a conditional equation @xmath129 is a _
quasi_-_identity _ of @xmath24 if @xmath130 , which is simply written as @xmath131 when clear from the context .
these definitions extend , as expected , to classes of algebras .
given a class of @xmath3-algebras @xmath6 , the ( _ semantic _ ) _ equational _ _ consequence _ _ relation _
@xmath132 determined by @xmath6 is defined by @xmath133 in this case @xmath127 is said to be a @xmath6-_consequence _ of @xmath42 .
when clear from the context , we simply write @xmath134 , where @xmath129 , for @xmath135 .
both @xmath136 and @xmath132 extend to sets of equations @xmath137 : @xmath138 iff @xmath139 for each @xmath140 , and respectively for @xmath132 .
for a set @xmath141 of quasi - equations , adopting the notational convention explained above and rather standard in universal algebra , @xmath142 stands for @xmath131 for each @xmath143 ( analogously for a class @xmath6 of @xmath3-algebras ) .
the equational consequence relation @xmath132 satisfies the conditions of definition [ df : klogic ] .
hence it constitutes an example of a 2- ( perhaps the most important one ! ) often simply denoted by @xmath6 .
a class @xmath6 of @xmath3-algebras is axiomatised by a set @xmath141 of conditional equations if @xmath144 .
it can be proved that , if @xmath6 is a class of @xmath3-algebras axiomatised by a set @xmath141 of conditional equations , then the relation @xmath145 is specifiable ( see @xcite for the one - sorted case ) . in this case
it can be defined in the hilbert style , taking the set of @xmath3-equations in @xmath141 , together with reflexivity , as the set of axioms , and the set of @xmath3-conditional equations in @xmath141 , along with symmetry , transitivity and congruence rules , as inference rules .
any specifiable equational deductive system over @xmath3 is the natural extension ( by axioms and rules ) of the ( 2-dimensional ) free deductive system over @xmath3 denoted by @xmath146 and defined in figure [ fig:3 ] .
note that the consequence relation associated to @xmath147 coincides with @xmath148 , where @xmath149 is the class of all @xmath3-algebras .
recall the discussion of the double role played by deductive systems in subsection [ sc : scope ] . actually , it is well known that any deductive system can be seen as a @xmath5-institution @xcite , and taking care of some extra technicalities , as an institution . however some special finitary ones are , in a more natural way , viewed as theories of institutions , _
i.e. _ , as sets of sentences equipped with extra features the deductive apparatus they induce .
for instance , given a quasivariety of algebras , we can define a ( 2-dimensional ) deductive system over the set of equations by using the identities and quasidentities that define the quasivariety , together with the axioms of reflexivity and the inference rules of symmetry , transitivity and congruence rules .
this deductive system can naturally be seen as a theory of the institution of horn clause logic .
as discussed in @xcite , a semantics for arbitrary @xmath0- needs to go beyond the usual algebraic structures , resorting to algebras endowed with a set of @xmath0-tuples .
formally , [ df : kstruct ] a _ @xmath0-structure _ over a signature @xmath150 is a pair @xmath151 where @xmath24 is a @xmath3-algebra and @xmath152 is a sorted set @xmath153 such that @xmath154 in this definition , the sorted set @xmath152 of designated elements of @xmath24 , can be regarded as a set of truth values on @xmath24 : a formula holds if its interpretation is one of these elements .
this is why @xmath152 is called a _
filter _ : for a deductive system representing the constructive propositional calculus on a boolean algebra the notion boils down to the familiar , boolean filter .
let @xmath155 be a @xmath0-structure .
a @xmath0-formula @xmath156 is said to be a _ semantic consequence _ in @xmath157 of a set of @xmath0-formulas @xmath42 , in symbols @xmath158 , if , for every assignment @xmath159 , @xmath160 whenever @xmath161 for every @xmath162 , where @xmath163 is the component @xmath14 of the sorted set @xmath152 .
notice that the same notation is used for the assignment and its natural extension to formulas .
a @xmath0-formula @xmath164 is _ valid _ in @xmath157 , and conversely @xmath157 is a _ model _ of @xmath16 , if @xmath165 .
a rule such as ( [ sequent ] ) is a _
validity _ , or a _ valid rule _ , of @xmath157 , and conversely @xmath157 is a _ model _ of the rule , if @xmath166 .
a formula @xmath16 is a _ semantic consequence _ of a set of @xmath0-formulas @xmath42 for an arbitrary class @xmath167 of @xmath0-structures over @xmath3 , in symbols @xmath168 , if @xmath169 for each @xmath170 .
it can be proved that @xmath171 is always a @xmath0- , even if not always specifiable .
similarly , a @xmath0-formula or a rule is a _ validity of _ @xmath167 if it is a validity of each member of @xmath167 .
a @xmath0-structure @xmath157 is a _ model _ of a @xmath0- @xmath61 if @xmath54 always implies @xmath169 , if every consequence of @xmath61 is a semantic consequence of @xmath157 .
the special models whose underlying algebra is the formula algebra , _
i.e. _ , models of the form @xmath172 , for @xmath82 a theory , are called _ lindenbaum - tarski models_. the class of all models of @xmath61 is denoted by @xmath173 .
if @xmath61 is a specifiable @xmath0- , then @xmath157 is a model of @xmath61 iff every axiom and rule of inference is a validity of @xmath157 . in the equational case
the semantics based on @xmath10-structures boils down to the traditional algebraic semantics .
more precisely , given a quasi - equational class @xmath6 of @xmath3-algebras ( _ i.e. _ axiomatised by a set of quasi - equations over @xmath3 ) , the algebra of any model of the 2-deductive system induced by @xmath6 with the identity as its filter belongs to @xmath6 . a class of @xmath0-structures @xmath167 is a _ @xmath0-structure semantics _ of @xmath61 if @xmath174 .
the class of all models of @xmath61 forms a @xmath0-structure semantics of @xmath61 .
this fact is expressed in a specific completeness theorem @xcite , stating that , for any @xmath0- @xmath61 , @xmath54 iff @xmath175 , for every @xmath176 .
_ this section introduces , exemplifies and discusses the concept of _ refinement by interpretation _ the core contribution of the paper framed in the specific , but popular , setting of algebraic specification over the institution of horn clause logic . _
this section introduces refinement by logical interpretation for the specific case of algebraic specifications . as usual
( see , @xcite ) , an _ algebraic specification _
@xmath177 is considered here as a structured specification over the institution of horn clause logic ( * hcl * ) @xcite .
each @xmath177 denotes a pair @xmath178 where @xmath3 is a signature and @xmath179 is a class of @xmath3-algebras .
the class @xmath179 of @xmath3-algebras is called the _ model class of _ @xmath177 , and each @xmath3-algebra in @xmath179 a _ model of _ @xmath177 .
if @xmath180 is a conditional equation @xmath181 ( respectively , an equation @xmath182 ) , we write @xmath183 for @xmath184 ( respectively , @xmath185 ) . both cases
extend , as expected , to sets of conditional equations ( respectively , equations ) . when an algebraic specification @xmath177 is _ flat _ , or _
basic _ , its model class @xmath179 of algebras is axiomatised by a set @xmath141 of conditional equations . in this case
@xmath177 is the pair @xmath186 .
if the definition is restricted to formulas over a specific set of @xmath3-variables @xmath187 , @xmath177 is said @xmath18-_flat_. when @xmath141 is a set of equations the flat specification @xmath188 is called a ( flat ) _ equational specification_. recall that a class @xmath6 of @xmath3-algebras axiomatisable by a set of equations is called a _
variety_. a variety can also be characterised as a nonempty class @xmath6 of @xmath3-algebras closed under homomorphic images , subalgebras and direct products ( cf .
@xcite , @xcite ) .
this famous result , due to birkhoff , turns out to be very useful to show that a given algebraic specification is not an equational specification . in the sequel , for simplicity , when clear from the context
, we refer to algebraic specifications simply as specifications . in this context _
stepwise refinement _
@xcite of specifications refers to the process through which a complex design is produced by incrementally adding details and reducing under - specification .
this proceeds step - by - step until the class of models becomes restricted to such an extent that a program can be easily manufactured .
formally , given a specification @xmath177 , the implementation process builds a correct realisation from a concrete enough class of @xmath3-models @xmath6 such that @xmath6 is a subset of the class of denotations of @xmath177 . during this process , the specification is enriched according to specific design decisions , iteratively approaching the intended meaning for the final program .
starting from an initial abstract specification @xmath189 , refinement builds a chain of specifications @xmath190 where , for @xmath191 , @xmath192 represents reverse inclusion of the respective classes of models .
transitivity of inclusion assures that @xmath193 . from @xmath194
onwards each element in this chain is the result of a _ refinement step_. in order to deal with more complex requirements along the implementation process , for example to enable the possibility of renaming , adding or grouping together different signature components , refinement steps @xmath195 are traditionally taken up to _ signature morphisms_. recall that a _ signature morphism _ from @xmath196 to @xmath197 is a pair @xmath198 , where @xmath199 and @xmath200 is a @xmath201-family of functions respecting the sorts of operation names in @xmath202 , _
i.e. _ , @xmath203 ( where for @xmath204 given a signature morphism @xmath205 and a @xmath206-algebra @xmath24 , let @xmath207 denote the _ reduct _ of @xmath24 along @xmath58 , , for any @xmath208 , @xmath209 , and for all @xmath210 , @xmath211 .
the notation @xmath212 and @xmath213 refer , respectively , to the carrier of sort @xmath14 and the interpretation of operation symbol @xmath214 in the algebra @xmath24 . in this context
we say that @xmath215 is a refinement of @xmath177 witnessed by @xmath58 , or simply a @xmath58-refinement , when @xmath216 , where @xmath217 .
having fixed the notation and the basic concepts we may now jump to the kernel of this section : representing classes of specification models as 2-deductive systems and introduce interpretations as possible witnesses to the refinement steps . as clarified in the introduction ( subsection [ sc
: scope ] ) , our approach is based on the representation of specifications model classes as deductive systems .
our focus is essentially semantic , _
i.e. _ , on what specifications denote , and , therefore , the whole theory of refinement by interpretation discussed here is independent from any specification structuring mechanism
. actually , any specification @xmath177 denotes a class of algebras its _ model class _ , @xmath218 .
this in turn induces a 2- according to the procedure sketched in subsection [ sc : eqqq ] .
its consequence relation is @xmath219 .
the possibility of this consequence relation being non finitary , for example if arising from non flat specifications , is also covered in this approach . as we will see
, it turns out that @xmath0-deductive systems are an efficient universal tool to develop , in this way , a theory about all classes of models over a fixed signature .
a most important particular case is that of _ flat _ specifications : each of them can be identified with the deductive system it induces .
this observation will be assumed throughout the text and in the examples .
moreover , using interpretations between the induced deductive systems , one may go from one model class to another . in a sense to become clearer below
, we say that a specification @xmath215 _ refines _ another specification @xmath177 _ by interpretation _
@xmath220 , if @xmath220 is an interpretation of @xmath177 such that @xmath221 for any formula @xmath180 . before jumping to the technical definition ,
consider the following example which , although elementary , may help to build up some intuitions for this notion of refinement .
[ firstex ] the introduction of a two - element boolean algebra together with equality tests for each sort allows the software engineer to formulate arbitrary universal first - order axioms as equations in the boolean sort .
the importance of this construction comes from the fact that , although many of the most natural specification conditions take the form of universal first - order sentences , only equational and conditional - equational axioms are guaranteed to possess an initial model .
this motivation , as well as equality - test algebras in general , are extensively discussed in @xcite .
consider , thus , two flat specifications @xmath222 and @xmath223 .
the former has a signature @xmath3 which declares a sort @xmath14 and a function @xmath224 .
@xmath222 has an empty set of axioms ; thus , the corresponding class of models consists of all algebras over its signature . on the other hand ,
specification @xmath223 is depicted in figure [ fig : unica ] .
it is not difficult to see , by induction on the structure of proofs , that the translation [ cols="^,^,^,^ " , ] we may now incorporate in the approach proposed in this paper the important tool given in lemma [ th : nsc ] : let @xmath225 be a self - translation between two @xmath226-deductive systems @xmath61 and @xmath110 over the signature @xmath3 .
then , if @xmath110 has the conjunction property , @xmath220 is an interpretation iff its associated function is a conservative translation .
first we prove that @xmath227 : since for any @xmath228 there is a @xmath229 such that @xmath230 , we have , by the conjunction property of @xmath110 , that @xmath231 and , by @xmath232 of definition [ df : klogic ] , that @xmath233 .
analogously , since for each @xmath229 there is a @xmath234 such @xmath235 , we have , by @xmath232 of definition [ df : klogic ] , that @xmath236 .
hence , for any interpretation @xmath220 and for all @xmath237 , @xmath238 , which implies that @xmath239 is a conservative translation . in a similar way ,
if @xmath239 is a conservative translation , @xmath220 is an interpretation .
the connection to conservative translations turns out to be very useful in practice . the following theorem , which builds on results in @xcite , provides a sufficient condition for a translation to be an interpretation .
[ th : nsc ] let @xmath220 be a @xmath240-translation from @xmath3 to @xmath206 , @xmath61 a @xmath0- over @xmath3 and @xmath110 a @xmath241- over @xmath206 .
suppose that @xmath220 is functional and injective .
if @xmath242 , for every set of formulas @xmath42 , then @xmath220 interprets @xmath61 in @xmath110 . from the inclusion @xmath243 we have that @xmath244 .
suppose now that @xmath245 . hence there is a @xmath246 such that @xmath247 . since @xmath220 is injective @xmath248 , and so , @xmath249 , _
i.e. _ , @xmath54 .
the approach to refinement proposed in this paper , in particular when specialised to 2-dimension deductive systems , should also be related to the extensive work of maibaum , sadler and veloso in the 70 s and the 80 s , as documented , for example , in @xcite .
the authors resort to interpretations between theories and conservative extensions to define a syntactic notion of refinement according to which a specification @xmath215 refines a specification @xmath177 if there is an interpretation of @xmath215 into a conservative extension of @xmath177 .
it is shown that these refinements can be vertically composed , therefore entailing stepwise development .
this notion is , however , somehow restrictive since it requires all maps to be conservative , whereas in program development it is usually enough to guarantee that requirements are preserved by the underlying translation .
moreover , in their approach , the interpretation edge of a refinement diagram needs to satisfy extra properties . as related work one should also mention @xcite where interpretations between theories are studied in the abstract framework of @xmath5-institutions
the first reference is a generalisation of the work of maibaum and his collaborators , whereas the second one generalises the way algebraic semantics on sentential logics is dealt with in abstract algebraic logic to the abstract setting of @xmath5-institutions .
similar developments could arise by considering institutions and their ( co-)morphisms @xcite .
the work of meseguer @xcite on _ general logics _ , in which a theory of interpretations between logical systems is developed , should also be mentioned .
our own approach to refinement by interpretation can be placed between these general works and the original contribution of maibaum .
actually , on the one hand , we deal with general @xmath0-deductive systems therefore subsuming all frameworks above which are based on equational or first order logic ( , on specific instances of @xmath0-deductive systems ) . on the other hand , however , our results are formulated in terms of a concrete and intuitive notion of a deductive system ; their scaling to an abstract , institutional level is still to be done .
the paper introduced a new notion of refinement and started the development of a corresponding theory of _ refinement by interpretation_. the results obtained and their applications seem promising , in the sense that a number of useful transformations of ( classes of models of ) specifications are captured as refinement steps . in order to clarify the scope of our results we should point out that the development in section 3 can be straightforwardly generalised as to apply to any horn fragment of a structural logic @xmath250 ( _ i.e. _ , a logic whose axioms have the form @xmath251 , where @xmath252 is a subset of the atomic formulas of the logic , with @xmath253 possibly empty ) .
all one has to do is to represent such a fragment by the natural equivalent deductive system taking the atomic formulas of @xmath250 as its set of formulas and a presentation given by the axiomatisation of @xmath250 .
the generalisation made along sections [ sc : gen ] and [ sc : rbi2 ] turns it relevant to the specification meta - level , , whenever an implementation step requires a change in the underlying logic .
this often arises in formal software development with the need for accommodating new requirements ( as in example [ ex : ordbams ] ) or when a particular theorem prover , embodying a specific logic , is to be used for design validation ( as in example [ ex : tool ] ) .
our most recent work @xcite is another generalisation effort aiming at reframing this notion of refinement in a categorical setting based on a characterisation of abstract logics as coalgebras for the closure system contravariant functor @xcite upon the category @xmath254 of sets and functions . to conclude we would like to remark again the semantic perspective from which this work was developed , as extensively discussed in the introduction
this entails the need for further research on how refinement by interpretation , which is entirely based on properties of arbitrary deductive systems , can be smoothly combined with concrete specification structuring operations .
preliminary work on this topic is reported in @xcite in which the emphasis is shifted to specifications .
current work in this direction includes the development of a refinement calculus of structured specifications over a @xmath5-institution . as a general remark we would like to stress again that the approach developed in this paper can be applied to any notion of algebraic specification based on any fixed set of specification structuring combinators , further justifying the relevance of the semantic perspective adopted in this paper .
a practical limitation of this approach of reducing the specifications to their class of models is that often such classes have infinitely many models .
however in some cases subclasses possessing the same theory can be considered instead .
for instance , if the class is a finitely generated quasivariety @xmath6 @xcite , and consequently the associated 2-deductive system is finitary with a presentation given by the axiomatisation of the quasivariety , we can replace @xmath6 by the set of its generators which induces the same deductive system .
an example is the class of boolean algebras which are generated ( actually , as a variety ) by the two - element boolean algebra . from an application point of view
, this semantic approach seems to have its own potentialities which we would like to recall .
actually , in a number of cases it is relevant , and even mandatory , to start the implementation procedure from a set of models that does not come from a structured specification
. this can be the case when reusing designs ( a recurrent strategy in engineering ) or even to express meta - requirements that can not be easily accommodated within the classic refinement procedure .
focussing on _ classes of models _ , on the other hand , makes possible to deal with requirements that can not be properly formalised in a specification .
note this does not entail any loss of expressivity .
the approach proposed in this paper can be tuned to specification refinement in a strict sense : for each specification , one may recursively compute its denotation ( a signature and a class of models ) and work directly with it . in general
, we believe that this approach has a real application potential , namely to deal with specifications spanning through different specification logics . particularly deserving to be considered , but still requiring further investigation , are observational logic @xcite , hidden logic @xcite and behavioural logic @xcite . in all of these cases
the satisfaction of requirements is discussed up to some particular satisfaction relation and their verification is checked with respect to relations obtained by replacing strict equality by its underlying notion of satisfaction . in this context , a semantics based on @xmath0-structures paves the way to the unification of all of these approaches . actually ,
in all of them , models consist of algebras whose @xmath0-structures are of the form @xmath255 , where @xmath256 captures the particular satisfaction relation in each formalism . in particular , the strict models of a ( classical ) algebraic specification @xmath177 consist of algebras @xmath24 whose @xmath0-data structure @xmath257 is a model of @xmath258 .
naturally , most of the models of software specifications are not admissible choices as implementations .
therefore , the choice of adequate filters along the implementation process becomes a crucial , although not trivial task .
this should be driven by the system nature ( for example , adopting observational equality to deal with objects with encapsulated data ) .
a similar concern is , moreover , shared by other general approaches to formal development , as , for example , @xcite in the context of behavioral logic .
a lot of other questions remain to be answered .
one such topic , as mentioned above , concerns _ horizontal composition _ of refinements by interpretation ; _ vertical _ composition raising no special problems as shown in theorems [ th : ver1 ] and [ th : ver2 ] . to illustrate the kind of results we are investigating suppose , for example , that @xmath220 interprets @xmath177 in @xmath215 .
the challenge would be to prove that @xmath220 also interprets an enrichment of @xmath177 by axioms in an appropriate sub - specification of @xmath215 .
a closely related issue is the extension of this approach to the level of ( structured ) specifications .
we believe that this can be captured in a somehow standard way , which will be most relevant in studying the interplay between horizontal ( _ i.e. _ , architectural ) and vertical ( _ i.e. _ , implementation driven ) levels of specification composition .
for example , the _ union _ of two specifications will correspond to the union of the corresponding consequence relations .
actually , a structured specification also defines a class of models and therefore induces a deductive system .
another topic to explore is the equivalence of algebraic specifications up to logical interpretation . as a starting point
, it would be worth to explore the relation @xmath259 defined as follows : @xmath260 if there are interpretations @xmath220 and @xmath261 such that @xmath262 and @xmath263 .
it is not difficult to see that @xmath183 implies @xmath264 and @xmath265 implies @xmath266 .
more challenging seems to be a stronger equivalence , studied in the context of equivalence between deductive systems @xcite , which requires interpretations to be _ mutually _ inverse .
last but not least , framing refinement by interpretation in the context of recent works on heterogeneous specification , raises interesting questions and opens the opportunity for computer - based support . actually , classical translations between logics ( , modal into first - order or the latter into equational logic ) are at the basis of hets @xcite , the heterogeneous specification framework . to go further in this direction
entails the need to regard interpretations from an institutional point of view @xcite , as some sort of comorphisms , and develop on top of it a calculus of refinements by interpretation .
the authors express their gratitude to the anonymous referees who raised a number of pertinent questions entailing a more precise characterisation of the paper s contributions and a clarification of their scope .
this work was funded by erdf - european regional development fund through the compete programme ( operational programme for competitiveness ) and by national funds through the fct ( portuguese foundation for science and technology ) within project ` fcomp-01 - 0124-feder-028923 ` ( nasoni ) and the project ` pest - c / mat / ui4106/2011 ` with compete number ` fcomp-01 - 0124-feder-022690 ` ( cidma - ua ) .
the first author also acknowledges the financial assistance by the projects getfun , reference ` fp7-people-2012-irses ` , and nociones de completud , reference ` ffi2009 - 09345 ` ( micinn - spain ) .
a. madeira was supported by the fct within the project ` norte-01 - 0124-feder-000060 ` .
r. m. burstall and j. a. goguen .
the semantics of clear , a specification language . in d.
bjrner , editor , _ abstract software specifications ( 1979 copenhagen winter school , january 22 - february 2 , 1979 ) _ , volume 86 of _ lecture notes in computer science _ ,
pages 292332 .
springer , 1980 .
m. bidoit and r. hennicker .
an algebraic semantics for contract - based software components . in j.
meseguer and g. rosu , editors , _ algebraic methodology and software technology ( amast 2008 - urbana , il , usa , july 28 - 31 , 2008 _ , volume 5140 of _ lecture notes in computer science _ , pages 216231 .
springer , 2008 .
m. codescu , f. horozal , m. kohlhase , t. mossakowski , and f. rabe .
project abstract : logic atlas and integrator ( latin ) . in james
davenport , william farmer , josef urban , and florian rabe , editors , _ intelligent computer mathematics _ ,
volume 6824 of _ lecture notes in computer science _ , pages 289291 .
springer berlin / heidelberg , 2011 .
b. dolle and w. dosch .
transforming functional signatures of algebraic specifications into object - oriented class signatures . in _
apsec 05 : proceedings of the 12th asia - pacific software engineering conference _ , pages 323332 .
ieee computer society , 2005 .
j. fiadeiro and t. s. maibaum . generalising interpretations between theories in the context of ( @xmath5- ) institutions . in _ proceedings of the first imperial college department of computing workshop on theory and formal methods _ , pages 126147 , london , uk , 1993 .
springer - verlag .
j. fiadeiro and a. sernadas . structuring theories on consequence . in d.
sanella and a. tarlecki , editors , _ recent trends in data type specification .
specification of abstract data types ( papers from the fifth workshop on specification of abstract datac types , gullane , 1987 ) _ , volume 332 of _ lecture notes in computer science_. springer - verlag , berlin , 1988 .
k. gdel .
an interpretation of the intuitionistic proposicional calculus ( 1933 ) . in s.
feferman et al , editor , _ collected works of kurt gdel ( vol .
i ) _ , pages 301303 .
oxford : oxford university press , 1986 .
j. goguen , j. thatcher , and e. wagner . an initial algebra approach to the specification , correctness and implementation of abstract data types . in r.
yeh , editor , _ current trends in programming methodology _
, pages 80149 .
prentice - hall international , 1978 .
j. goguen , t. winkler , j. meseguer , k. futatsugi , and j .-
jouannaud . introducing obj . in j.
goguen and g. malcolm , editors , _ software engineering with obj : algebraic specification in practice_. cambridge university press , 1996 . j. meseguer .
general logics . in h.
d. ebbinghaus , j. fernandez - prida , m. garrido , d. lascar , and m. rodriguez - artalejo , editors , _ logic colloqium87 _ , studies in logic and the foundations of mathematics ( volume 129 ) , pages 275330 .
elsevier , 1989 .
m. a. martins , a. madeira , and l. s. barbosa .
refinement via interpretation . in d.
v. hung and p. krishnan , editors , _ seventh ieee international conference on software engineering and formal methods ( sefm 2009 , hanoi , vietnam , 23 - 27
november 2009 ) _ , pages 250259 .
ieee computer society , 2009 .
t. mossakowski , c. maeder , and k. lttich .
the heterogeneous tool set , hets . in o.
grumberg and m. huth , editors , _ tools and algorithms for the construction and analysis of systems ( tacas 2007 - braga , portugal , march 24 - april 1 , 2007 ) _ , volume 4424 of _ lecture notes in computer science _ ,
pages 519522 .
springer , 2007 .
t. s. maibaum , m. r. sadler , and p. a. veloso .
logical specification and implementation . in j.
mathai and r. k. shyamasundar , editors , _ foundations of software technology and theoretical computer science ( fsttcs , bangalore , india , dec.13 - 15 , 1984 ) _ , volume 4424 of _ lecture notes in computer science _ , pages 1330 , london , uk , 1984 .
springer .
t. s. e. maibaum , paulo a. s. veloso , and m. r. sadler . a theory of abstract data types for program development : bridging the gap ? in hartmut ehrig , christiane floyd , maurice nivat , and
james w. thatcher , editors , _ mathematical foundations of software development ( tapsoft , berlin , germany , march 25 - 29 , 1985 ) _ , volume 186 of _ lecture notes in computer science _ , pages 214230 .
springer , 1985 .
c. j. rodrigues , m. a. martins , a. madeira , and l. s. barbosa .
refinement by interpretation in @xmath5-institutions . in john derrick ,
eerke a. boiten , and steve reeves , editors , _
proceedings 15th international refinement workshop _ ,
volume 55 of _ eptcs _ , pages 5364 , 2011 .
d. sannella .
the common framework initiative for algebraic specification and development of software : recent progress . in m.
cerioli and g. reggio , editors , _ recent trends in algebraic development techniques ( revised selected papers of wadt 2001,genova , italy , april 1 - 3 , 2001 ) _ , volume 2267 of _ lecture notes in computer science _ , pages 328344 .
springer , 2001 . j. da silva , i. m. dottaviano , and a. m. sette . in _ models , algebras , and
proofs : selected papers of the x latin american symposium on mathematical logic , ( bogot , 1995 ) _ , pages 435448 .
notes pure appl . math .
( 203 ) , 1999 .
d. sannella and a. tarlecki .
horizontal composability revisited . in k. futatsugi , j .-
jouannaud , and j. meseguer , editors , _ algebra , meaning , and computation , essays dedicated to joseph a. goguen on the occasion of his 65th birthday _ , volume 4060 of _ lecture notes in computer science _ , pages 296316 .
springer , 2006 .
a. tarlecki . moving between logical systems . in magne
haveraaen , olaf owe , and ole - johan dahl , editors , _ recent trends in data type specification ( selected papers of 11th workshop on specification of abstract data types joint with the 8th compass workshop , oslo , norway , september 19 - 23 , 1995 ) _ , volume 1130 of _ lecture notes in computer science _ , pages 478502 .
springer , 1996 .
b. yu , l. kong , y. zhang , and h. zhu . testing java components based on algebraic specifications . in _ first international conference on software testing , verification , and validation , icst 2008 _ , pages 190199 .
ieee computer society , 2008 . | stepwise refinement of algebraic specifications is a well known formal methodology for program development . however , traditional notions of refinement based on signature morphisms are often too rigid to capture a number of relevant transformations in the context of software design , reuse , and adaptation .
this paper proposes a new approach to refinement in which signature morphisms are replaced by _
logical interpretations _ as a means to witness refinements .
the approach is first presented in the context of equational logic , and later generalised to deductive systems of arbitrary dimension .
this allows , for example , refining sentential into equational specifications and the latter into modal ones . |
the hartree - fock ( hf ) approximation ignores the correlation between electrons , but gives roughly 99% of the total electronic energy @xcite .
moreover , it is often accurate for the prediction of molecular structure @xcite , computationally cheap and can be applied to large systems , especially within local ( linear - scaling ) strategies @xcite . to reduce the computational cost still further ,
various numerical techniques have been developed including , for example , density fitting ( or resolution of the identity ) @xcite , pseudospectral and cholesky decomposition @xcite , dual basis methods @xcite , and both attenuation @xcite and resolution @xcite of the coulomb operator .
unfortunately , the part of the energy which the hf approximation ignores can have important chemical effects and this is particularly true when bonds are formed and/or broken . consequently
, realistic model chemistries require a satisfactory treatment of electronic correlation .
the concept of electron correlation was introduced by wigner @xcite and defined as @xmath5 by lwdin @xcite , where @xmath6 is the exact non - relativistic energy .
feynman refered to @xmath0 as the `` stupidity energy '' @xcite because of the difficulties associated with its characterization in large systems . even though it is a formidable challenge to determine the correlation energy accurately , even in simple systems
, recent heroic calculations on the helium atom @xcite have demonstrated how near - exact results can be found .
indeed , this elementary chemical system has been compared to the number @xmath7 by charles schwartz @xcite : _
`` for thousands of years mathematicians have enjoyed competing with one other to compute ever more digits of the number @xmath7 . among modern physicists ,
a close analogy is computation of the ground state energy of the helium atom , begun 75 years ago by e. a. hylleraas . '' _
although @xmath0 in the helium atom is now known very accurately , certain correlation effects remain incompletely understood and , for example , even the coulomb hole @xcite itself is more subtle than one might imagine .
the primary effect of correlation is to decrease the likelihood of finding the two electrons close together and increase the probability of their being far apart .
however , accurate calculations have revealed the existence of a secondary coulomb hole , implying that correlation also brings distant electrons closer together @xcite .
the same observation has been made in the h@xmath8 molecule by per _
_ @xcite and it appears that secondary ( or long - range ) coulomb holes may be ubiquitous in two - electron systems @xcite . in order to get benchmark results for the development of intracule functional theory ( ift ) @xcite , we have recently initiated an exhaustive study of two - electron systems @xcite . in the present frontier article , we review our recent progress in the determination of the correlation energy in various high - density two - electron systems : the helium - like ions ( sec .
[ subsec : helium ] ) , two electrons on the surface of a sphere ( sec .
[ subsec : spherium ] ) , the hooke s law atom ( sec .
[ subsec : hookium ] ) , and two electrons trapped in a spherical box ( sec .
[ subsec : ballium ] ) .
it is reasonable to ask whether an understanding of the high - density regime is relevant to normal chemical systems but it turns out that most of the high - density behaviour of electrons is surprisingly similar to that at typical atomic and molecular electron densities .
much can be learned about the languid waltz of a pair of electrons in a covalent bond from their frenetic jig in the high - density limit .
moreover , it has led to an understanding of key systems , such as the uniform electron gas @xcite , which form the cornerstone of the popular local density approximation in solid - state physics @xcite .
we also show ( sec .
[ sec : ldl ] ) that , in the large - dimension limit , the high - density correlation energy of two electrons is given by a simple universal rule which is independent of the external confining potential . just as one learns about interacting systems by studying non - interacting ones and then introducing the interaction perturbatively , one can understand our three - dimensional world by studying high - dimensional analogues and introducing dimension - reduction perturbatively .
in this study , we confine our attention to the @xmath9 ground states of two - electron systems .
this allows us to ignore the spin coordinates and focus on the spatial part of the wave function .
atomic units are used throughout .
the basic coordinates of a two - electron system.,scaledwidth=20.0% ]
for two electrons confined in a spherically - symmetric external potential @xmath2 , the hamiltonian is @xmath10 where the first two terms represent the kinetic energy of the electrons and @xmath11 is the coulomb operator ( fig .
[ fig : scheme ] ) . after a suitable scaling of the coordinates and energy @xcite
, the hamiltonian can be recast as @xmath12 where @xmath13 measures the confinement strength .
equation is well poised for a perturbation treatment in which the zeroth- and first - order hamiltonians are @xmath14 and the one - electron hamiltonian is given by @xmath15 the zeroth - order wave function satisfies the eigenequation @xmath16 and the zeroth- and first - order energies are @xmath17 following hylleraas @xcite , we can use perturbation theory to expand both the exact @xcite and hartree - fock ( hf ) @xcite energies as series in @xmath18 , yielding @xmath19 and @xmath20 where @xmath3 is the dimensionality of the space .
it is straightforward to show that @xmath21 and therefore , in the high - density ( large-@xmath13 ) limit , we find @xmath22 \notag \\ & = e^{(2)}(d ,
v ) - e_{\rm hf}^{(2)}(d , v).\end{aligned}\ ] ] [ cols="<,^,^,^,^,^,^",options="header " , ] coefficients of the exact ( dashed ) , hf ( dotted ) and correlation ( solid ) energies with respect to @xmath23 , for @xmath24 ( eqs . , and ) .,scaledwidth=45.0% ] expanding and in terms of @xmath13 and @xmath3 yields @xmath25 thus showing that both @xmath26 and @xmath27 depend on the leading power @xmath23 of the external potential but not on @xmath28 . subtracting these energies yields @xmath29 and this completes the proof that , in the high - density limit , the leading coefficient @xmath30 of the large-@xmath3 expansion of the correlation energy is universal , _
i.e. _ it does not depend on the external potential @xmath2 .
the result is related to the cusp condition @xcite @xmath31 which arises from the cancellation of the coulomb operator singularity by the @xmath3-dependent angular part of the kinetic operator @xcite .
the @xmath32 and @xmath33 coefficients can be found by considering the langmuir vibrations of the electrons around their equilibrium positions @xcite .
the general expressions depend on @xmath34 and @xmath35 , but are not reported here .
however , for @xmath24 , which includes many of the most common external potentials , we find @xmath36 showing that @xmath37 , unlike @xmath30 , is potential - dependent .
numerical values of @xmath37 are reported in table [ tab : ec ] for various systems , and the components of the correlation energy are shown graphically in fig .
[ fig : proof - fig ] .
in this paper , we have reviewed our recent progress in the determination of the high - density correlation energy for four two - electron systems : the helium atom ( he ) , the hooke s law atom ( ho ) , two electrons confined on the surface of a sphere ( sp ) , and two electrons trapped in a ball ( ba ) . in the large-@xmath13 limit for @xmath4 , we have found @xmath38 these striking similarities can be rationalized by treating the dimensionality @xmath3 of space as a system parameter , and we have proved that , as @xmath3 grows , all such correlation energies exhibit the same universal behaviour @xmath39 in a @xmath3-dimensional space .
this is true _ irrespective _ of the nature of the external potential that confines the electrons .
we thank yves bernard , joshua hollett and andrew gilbert for kind support and many stimulating discussions .
thanks the nci national facility for a generous grant of supercomputer time and the australian research council ( grants dp0771978 and dp0984806 ) for funding .
j. d. morgan iii , the dimensional dependence of rates of convergence of rayleigh - ritz variational calculations on atoms and molecules , dimensional scaling in chemical physics , kluwer academic publishers , dordrecht , 1993 , p. 336 . | we review our recent progress in the determination of the high - density correlation energy @xmath0 in two - electron systems .
several two - electron systems are considered , such as the well known helium - like ions ( helium ) , and the hooke s law atom ( hookium ) .
we also present results regarding two electrons on the surface of a sphere ( spherium ) , and two electrons trapped in a spherical box ( ballium ) .
we also show that , in the large - dimension limit , the high - density correlation energy of two opposite - spin electrons interacting _ via _ a coulomb potential is given by @xmath1 for any radial external potential @xmath2 , where @xmath3 is the dimensionality of the space .
this result explains the similarity of @xmath0 in the previous two - electron systems for @xmath4 . |
the movement of a light - emitting object relative to an observer causes relativistic beaming of the light ( aka doppler boosting ) since the specific intensity is not a relativistic invariant . when the object s radial velocity ( rv ) is periodically modulated , as is the case in binary systems , a periodic sinusoidal variation in the observed flux is induced .
this effect is one of three that can be seen in light curves of binary systems , not necessarily eclipsing .
the other two are the ellipsoidal effect , when a member of the binary system is tidally affected by the gravitational pull of its companion , and the reflection effect , when light originating from one component is reflected by the other .
the three effects were discussed in the context of planet - host stars by @xcite , and main sequence stellar binaries by @xcite . both studies addressed binary systems with orbital periods much longer than a day , finding that the required photometric accuracy necessitates space - based observations .
observational detections of the beaming effect are rare .
@xcite noted that the beaming effect is likely to be seen in their ground - based data of a sdb star + wd binary system with a period of @xmath2=2.28 hr , although the effect s amplitude was not measured and the light curve is dominated by the ellipsoidal effect .
@xcite measured the beaming and ellipsoidal effects in a _
kepler _ light curve of an a - type star + low - mass wd binary system with @xmath2=5.19 d @xcite , and
in fact used these measurements to resolve the puzzling nature of the system ( see also * ? ? ?
@xcite identified the beaming and ellipsoidal effects in _ corot _ data of a brown dwarf host star ( corot-3 ; @xmath2=4.26 d ) , and @xcite observed all three effects with _ kepler _ in a 9.69 hr sdb + wd binary .
we show here that for high mass ratio detached , non - interacting double white dwarf ( wd ) binaries the relative beaming amplitude is above the 10@xmath3 level , while the ellipsoidal and reflection amplitudes are smaller by an order of magnitude or more .
therefore , these systems represent an excellent opportunity to measure the beaming effect using ground - based telescopes . following the discovery of the first eclipsing detached double wd binary nltt 11748 by @xcite we immediately noticed it is an excellent candidate for ground - based observations of the beaming effect .
the observed rv amplitude is large ( @xmath4270 km s@xmath5 ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) , while the secondary is @xmath430 times fainter than the primary @xcite .
in addition , the orbital period is short enough to be covered within a single night .
nltt 11748 and similar systems where both the ( photometric ) beaming amplitude and the ( spectroscopic ) rv amplitude can be measured are a valuable tool for testing the relation between these two quantities , which is currently approximated using theoretical assumptions ( e.g. , * ? ? ?
* ; * ? ? ?
we discuss the beaming effect in non - interacting double wd systems in .
our observations and data analysis are described in .
we present the results in where we also test them and compare our findings to theoretical expectations . in
we summarize our results and give our conclusions .
the relatively large beaming amplitude in non - interacting double wd binaries originates from the unique wd mass - radius , or mass - luminosity relation . in those binary systems
the photometric primary ( hereafter , the primary ) is a helium core wd , a few times smaller in mass but larger in radius and luminosity than the photometric secondary ( hereafter , the secondary ) , a c / o core wd ( see also table 1 in @xcite ) . as shown by @xcite the observed beaming effect is the weighted difference between the individual beaming effects of the two binary components , shifted by a phase of 0.5 .
however , the larger mass and smaller luminosity of the c / o wd make its own beaming effect much smaller , and even negligible relative to that of the he wd .
this can be seen by assuming both stars radiate as blackbodies and rewriting equation ( 6 ) of @xcite as : @xmath6 where @xmath7 is the radial velocity amplitude of the primary , @xmath8 the speed of light , @xmath9 ( @xmath10 ) is the mass of the primary ( secondary ) and @xmath11 ( @xmath12 ) is the flux from the primary ( secondary ) .
@xmath13 relates to the spectral index @xmath14 as : @xmath15 where @xmath16 and @xmath17 ( @xmath18 ) is the effective temperature of the primary ( secondary ) . for a high mass ratio non - interacting double wd binary the term @xmath19 in eq .
( [ eq : beamamp1 ] ) right hand side becomes of order unity .
this follows from the mass ratio and the flux ratio of the two stars , and also from the fact that @xmath20 is itself close to unity for the expected effective temperatures of the two stars . therefore eq .
( [ eq : beamamp1 ] ) reduces to : @xmath21 or in physical parameters : @xmath22 where @xmath23 the orbital inclination angle .
the approximation neglects the contribution of the secondary and assumes a blackbody spectrum and a circular orbit .
since @xmath24 has a value of a few in the optical for the typical temperatures of he wds ( e.g. , * ? ? ?
* ) , the equation above shows that the beaming amplitude is expected to be at the few 10@xmath3 level , and therefore detectable from the ground .
we note that this is not the accuracy required from a single measurement but the overall relative sinusoidal variability amplitude along the entire orbit .
[ fig : beamamp2 ] shows the expected sdss-@xmath0 beaming amplitude versus orbital period , for nltt 11748-like systems ( solid line ) .
also plotted ( dashed line ) is the expected amplitude for a system with an inclination angle of @xmath25 degrees .
the figure shows that systems similar to nltt 11748 , but with longer periods and/or lower inclinations are also expected to show a beaming amplitude above the 10@xmath3 level .
we observed nltt 11748 for 3.54.0 hr during each of the three nights of 2010 february 7 , 9 & 12 ut , obtaining data over all orbital phases .
most orbital phases were observed during two different nights .
our observations were done at the lcogt 2.0 m faulkes telescope north ( ftn ) , located at the haleakala observatory , hawaii .
we used the spectral instruments 600 series camera and the sdss-@xmath0 filter .
the camera has a 4k @xmath26 4k back - illuminated fairchild imaging ccd with a @xmath27 field of view ( fov ) .
we used the default @xmath28 pixel binning mode , with an effective pixel scale of @xmath29 .
the telescope pointing was adjusted to have the guiding camera fov include a suitable guide star .
our observations were done while defocusing the telescope , spreading the stellar psf over more pixels and increasing the overall open shutter time relative to the camera dead time , mostly ccd readout time .
the exposure time was 120 s , resulting in a median cycle time of 143 s. data taken during each night was processed the same way .
preliminary reduction , including bias and flat - field corrections was done using standard iraf routines .
we applied aperture photometry using iraf / phot by using a few aperture sizes and eventually choosing a radius of 10 pixels ( @xmath30 ) , resulting in the smallest scatter in the final light curves of the target and similar brightness stars , within 1 mag from that of the target .
the target s light curve was calibrated by dividing it by a comparison light curve a weighted average light curve of four comparison stars showing no variability , whether inter- or intra - night .
this calibration included propagation of the errors of both the target and comparison light curves . out of the total of 286 exposures
seven were taken completely or partially during the 3 min eclipses @xcite , and were ignored in our analysis .
two additional clear outliers in the target s light curve were visually verified to result from a cosmic ray hitting within the photometric aperture .
therefore , a total of 277 exposures were used in our analysis .
the error estimates of the calibrated light curve were directly based on the results of iraf / phot .
we rescaled these errors to better reflect the true noise in the data , by fitting a 3rd degree polynomial to the flux ratio light curves of each night , and inflating the error bars to make the reduced @xmath31 equal one .
the derived scaling factors were 1.83 , 1.37 & 1.53 for the three nights , feb . 7 , 9 & 12 , respectively .
we note that this polynomial fit was done only for the purpose of rescaling the error bars .
the target s airmass varied from @xmath41 to @xmath42 during our observations , which is expected to affect the resulting light curve , depending on the difference in color between the target and comparison light curves .
we removed these trends by taking the data from all three nights together and fitting a 2nd degree polynomial to the target s relative flux versus airmass , and dividing by the fit .
this step resulted in the final light curve , shown in fig .
[ fig : lcnight ] . in order to look for the beaming effect in the final light curve we simultaneously fitted for the three known effects ( beaming , reflection and ellipsoidal ) . in our modeling
we took the beaming effect to be a _ sine _
modulation at the orbital period .
the reflection effect has the same period but is lagging in phase by @xmath32 , hence it is a _ cosine _
modulation at the orbital period .
the ellipsoidal effect has half the orbital period but the same phase as the reflection effect , so it is a cosine modulation at the first harmonic of the period .
we used the ephemeris of @xcite and fitted the phase folded light curve with a 5-parameter model @xcite , including a sine and cosine terms for the orbital period and its first harmonic : @xmath33 where @xmath34 , time subtracted the mid primary eclipse time .
the ephemeris of @xcite is based on observations done during 1.5 months prior to the observations presented here .
this resulted in a negligible uncertainty in the phase during our observations , smaller than 20 seconds , or 10@xmath3 of the period , justifying keeping the phase fixed in our fit .
results for the fitted coefficients are listed in the middle column of .
the fitted beaming relative amplitude is @xmath35 , showing that the beaming signal is detected with a significance of @xmath36 .
the other three fitted coefficients were found to be consistent with zero .
the @xmath31 statistics of the fitted model was 283.9 , and the reduced @xmath31 was @xmath37 for the 272 degrees of freedom , indicating a good estimate of the errors . fig .
[ fig : lccomb ] presents the phase folded light curve and the fitted model is overplotted .
we have tested our results in order to verify they are not a statistical fluctuation , and that the beaming signal was not accidently injected into the data by our analysis . assuming the light curve is non - variable ( equivalent to fitting a constant value equal to the weighted average ) results in @xmath38 for 276 degrees of freedom , and @xmath39 , clearly rejecting the no variability hypothesis . a bootstrap analysis , done by randomly permuting the light curve points in time showed that the probability of the measured beaming signal being a random one is smaller than 10@xmath40 .
both these tests show , in addition to the amplitude s high s / n , that the detected variability has a high statistical significance
. it could be that the way we accounted for the airmass induced variation has injected the sine signal into the data .
we tested this in two ways .
first , we changed the polynomial degree used in fitting the relative flux versus airmass . using a degree of 1 gave a 0.5@xmath41 increase in the fitted beaming amplitude and increasing the polynomial degree up to 6 decreased the amplitude by @xmath42 .
in all these fits the other three fitted amplitudes were found to be consistent with zero .
second , we examined the light curves of similar color stars in the field , with a @xmath43 term within 0.1 mag from that of nltt 11748 . for those stars
@xmath44 was similar at the 0.01 level to that of a non - variable model , indicating they show no significant variability , and that the way we accounted for the airmass variation during our observations did not induce the detected variability .
another possibility is that the detected signal originated from a variability in one of the four comparison stars .
we checked this in two ways .
first , we repeated our analysis four times while excluding one of the four comparison stars , using only the other three .
this changed the fitted coefficients by less than 1@xmath41 and resulted in similar @xmath31 values .
second , we applied the same analysis for each comparison star , using only the other three as comparisons .
as in the case of the similar color stars described above , for all four comparison stars this resulted in a similar @xmath31 when fitting the 5-parameter model and a non - variable model , and the fitted beaming amplitude was consistent with zero .
these two tests show that the variability identified in the target did not originate from the comparison stars .
we repeated our analysis three times using data from only two of the three nights .
we were able to detect the beaming amplitude in each of the three analyzes , although with a lower significance , but still higher than 4@xmath41 . in each of those three
analyzes the other three coefficients were found to be consistent with zero . finally , we fitted the data also with a simpler , 2-parameter model , including a zero point and only the sine component corresponding to the beaming amplitude ( @xmath45 ) .
the result was the same as in the original 5-parameter model , and is listed on the second column from the right in . for estimating the expected amplitude of the beaming effect in our data we first calculated it as a function of wavelength assuming a blackbody spectrum , using eq .
( [ eq : beamamp2 ] ) .
this function is plotted as a solid black line in fig .
[ fig : beamamp ] .
we then weighted it by the sdss-@xmath0 filter transmission curve and the ccd qe . the final estimates for each band are marked as filled circles in fig .
[ fig : beamamp ] where the error bars due to the uncertainties in @xmath17 and @xmath7 are comparable to the size of the markers . fig .
[ fig : beamamp ] also justifies our choice to observe in sdss-@xmath0 , as the beaming amplitude increases towards the blue , but observations in sdss-@xmath46 are difficult at ftn .
the expected sdss-@xmath0 amplitude is ( @xmath47 , which is within 1@xmath41 from the measured amplitude of ( @xmath48 , marked by a diamond in fig .
[ fig : beamamp ] .
therefore , the measured amplitude for the beaming effect is consistent with the expected value assuming a blackbody spectrum .
another way to estimate the expected beaming amplitude , in a more direct manner and without the need to assume a blackbody spectrum is to integrate over a measured spectrum of the system using eq . 2 of @xcite .
for that end we used the ntt / efosc2 spectrum obtained by @xcite , which covers the sdss-@xmath0 band .
this resulted in an expected amplitude of @xmath49 , less than 2@xmath41 away from the measured amplitude .
however , we caution that the uncertainty on this result is difficult to estimate .
this low resolution spectrum was derived from two 20 min exposures , and was affected by the doppler shift variation during the exposure , and by the rv difference between the two exposures , which could have been at the @xmath5050 km s@xmath5 level , depending on the exact orbital phases they were taken at .
we estimate the amplitude of the ellipsoidal effect following @xcite to be about @xmath51 , where @xmath52 is the orbital semi - major axis , @xmath53 the radius of the primary , and we have ignored an order to unity coefficient ( see also * ? ? ? * ; * ? ? ?
this expression is @xmath54 for nltt 11748 .
although it is only a rough estimate and a more accurate one can be derived for this system , we use it only to show that the ellipsoidal effect is too small to be detected in our data . for the reflection effect we use @xmath55 as an upper limit , which is @xmath56 for nltt 11748 .
therefore , both the ellipsoidal and reflection effects are not expected to be detected from the ground , in agreement with our results .
we have presented a ground - based measurement , carried out with a 2.0 m telescope , of the relativistic beaming effect in the high mass ratio wd binary nltt 11748 @xcite .
we showed that the measured amplitude , of @xmath1 , is within 1@xmath41 from an approximated theoretical estimate , and within 2@xmath41 from a more direct estimate using a low resolution spectrum of the system .
the ellipsoidal and reflection effects were not detected in our data , in agreement with theoretical expectations . to our knowledge
this is the first ground - based measurement of the relativistic beaming effect .
our measurement is a first step in testing the relation between the beaming and rv amplitudes in systems similar to nltt 11748 , and it suggests it can be approximated by assuming a blackbody spectrum . although we were not able to obtain a high precision measurement of the beaming amplitude , it is identified at a high significance , of more than 7@xmath41 .
although short period detached he + c / o wd binaries are rare , their discovery rate is increasing .
five such systems were recently discovered @xcite by rv monitoring , and are expected to show beaming amplitudes of @xmath57 in the optical .
all five have orbital periods of only a few hours , shorter than that of nltt 11748 , so they can be used to further test the relation between the rv amplitude and the beaming amplitude , and also the wavelength dependency of the latter , shown in fig .
[ fig : beamamp ] for nltt 11748 .
the evolution of a low - mass he wd requires it to have a binary companion ( e.g. , * ? ? ?
therefore , once a he wd is identified the beaming effect can be used to look for a binary companion photometrically , in a very similar way to spectroscopic monitoring . since in this case
the light curve becomes analogous to the rv curve , high precision photometry can in principle be used to model the orbit and measure the mass ratio of the two components .
the authors warmly thank marten van kerkwijk and shai zucker for helpful discussions . this paper uses observations obtained with facilities of the las cumbres observatory global telescope .
dlk was supported by nasa through hubble fellowship grant # 01207.01-a awarded by the stsci which is operated by aura , inc .
, for nasa , under contract nas 5 - 26555 .
this work was supported by the national science foundation under grants phy 05 - 51164 and ast 07 - 07633 . expected beaming amplitude according to eq .
( [ eq : beamamp3 ] ) , for an nltt 11748-like system as a function of orbital period in the sdss-@xmath0 band .
the plot was produced using the masses of the two wds in nltt 11748 and the effective temperature of the primary .
the solid line corresponds to a completely edge - on system , close to nltt 11748 inclination of @xmath66 degrees , while the dashed line corresponds to a similar system with @xmath23=60 degrees . the black filled circle marks the expected nltt 11748 position .
the plot clearly shows that similar systems to nltt 11748 , with a longer period and a lower inclination are also expected to show a beaming amplitude above the 10@xmath3 level , marked with a horizontal dotted line . ]
nightly light curves of nltt 11748 , observed at ftn on the nights of ( from left to right ) 2010 february 7 , 9 & 12 ut .
the overplotted solid line is the fitted model .
although the scatter in the data is relatively large , the small variability can be seen in the data . ]
nltt 11748 light curve phase folded with the known ephemeris .
phase zero is time of primary eclipse .
data from different nights are marked by different colors ( top light curve ) .
the bottom light curve was generated by binning the points in phase , using 0.04 phase bins , and was shifted downward for clarity .
the binned light curve is shown for visualization only , and was not used in our analysis .
overplotted solid line ( black ) is the fitted 5-parameter model , constructed from the coefficients listed in . ]
amplitude of the expected relativistic beaming effect for nltt 11748 as a function of wavelength ( black solid line ) , assuming a blackbody spectrum .
the filled circles are the calculated amplitudes in each filter , accounting for the ccd qe and filter transmission curves .
the plotted colored curves account for both the ccd qe and filter transmission .
the sdss-@xmath0 beaming amplitude measured here is marked by a diamond , along with error bars ( green ) . ] | we report on the first ground - based measurement of the relativistic beaming effect ( aka doppler boosting ) .
we observed the beaming effect in the detached , non - interacting eclipsing double white dwarf ( wd ) binary nltt 11748 .
our observations were motivated by the system s high mass ratio and low luminosity ratio , leading to a large beaming - induced variability amplitude at the orbital period of 5.6 hr .
we observed the system during 3 nights at the 2.0 m faulkes telescope north with the sdss-@xmath0 filter , and fitted the data simultaneously for the beaming , ellipsoidal and reflection effects .
our fitted relative beaming amplitude is @xmath1 , consistent with the expected amplitude from a blackbody spectrum given the photometric primary radial velocity amplitude and effective temperature .
this result is a first step in testing the relation between the photometric beaming amplitude and the spectroscopic radial velocity amplitude in nltt 11748 and similar systems .
we did not identify any variability due to the ellipsoidal or reflection effects , consistent with their expected undetectable amplitude for this system .
low - mass , helium - core wds are expected to reside in binary systems where in some of those systems the binary companion is a faint c / o wd and the two stars are detached and non - interacting , as in the case of nltt 11748 .
the beaming effect can be used to search for the faint binary companion in those systems using wide - band photometry . |
core collapse supernovae with hydrogen lines near maximum light and a pronounced plateau in their visible band light curve that remain within @xmath0 mag of maximum brightness for an extended period are called type iip supernovae .
the plateau duration is often 60 - 100 rest - frame days and is followed by an exponential tail powered by radioactive decay at late times .
type iip supernovae constitute about 67% of all core collapse supernovae in a volume limited sample ( @xmath1 mpc ) @xcite .
their characteristic optical light curves are attributed to the hydrogen envelope of the progenitor remaining largely intact before the core collapse .
several lines of evidence suggest that these stars were red supergiants when they exploded .
type ii supernovae ( as well as the type ib / c supernovae ) are associated with recent star forming regions of the spiral galaxies which suggest that their progenitor stars are relatively short lived and therefore explode from massive stars ( @xmath2 ) which have faster burning of their nuclear fuels at generally higher central temperatures compared to less massive stars .
their plateau brightness and duration combined with their expansion velocities suggest a pre - supernova radius of typically red supergiant dimensions ( e.g. @xmath3 ) .
while there are more massive red supergiants in the local group of galaxies , no high - mass red supergiant progenitors above @xmath4 have been found for iip supernovae in direct detection efforts in volume - limited search for supernova progenitor stars @xcite .
this _ red supergiant problem _ has led to a viewpoint that these more massive stars could have core masses high enough to form black holes and supernovae from them could be too faint to have been detected . at the same time a minimum stellar mass for a sn iip to arise from is about @xmath5 , consistent with the upper limit to white dwarf progenitor masses .
a census of type iip progenitors and the partial loss of their hydrogen envelopes could potentially be important for neutron star vs black hole formation and their number distribution in the galaxy . in a type iip supernova , the expanding ejecta interacts with the slow wind of the red supergiant even though the star retained most of its hydrogen envelope intact .
this interaction generates a hot region , bounded by the forward and reverse shocks , which may emit thermal x - rays .
these shocks may also accelerate charged particles and synchrotron radiation from relativistic electrons can lead to radio emission that is strong enough to be detected at extragalactic distance scales .
the measurable radio and x - ray properties give complementary information about the regions shocked by the forward and reverse shocks and can constrain the physical properties of the interaction region and the progenitor star . because type iip supernovae have a plateau phase of high optical luminosity , energetic electrons near the forward shock find a dense photon environment of seed photons which may be compton boosted to x - ray energies .
the resultant compton cooling can affect the population of the electrons which also emit at radio wavelengths and show suppressed radio fluxes above a characteristic cooling break until the end of the plateau phase where the optical photon density undergoes a rapid decline @xcite . using sn 2004dj as a prototype of its class
we would like to address the questions : 1 .
what is the origin of x - rays detected by the chandra from young type iip supernovae ? 2 .
how fast was the progenitor losing mass via winds in the final phase before explosion ?
how efficiently does the supernova accelerate cosmic ray electrons in its forward shock ?
4 . what is the extent of turbulent magnetic field amplification in the post - shock material ? 5 . are synchrotron and inverse compton losses important in explaining the radio lightcurves ?
sn 2004dj , discovered by k. itagaki ( see * ? ? ?
* ) on 31.76 july 2004 ( ut ) in the spiral galaxy ngc 2403 , was the closest @xcite normal type iip sn @xcite observed to date .
it was discovered @xmath6 days @xcite after its progenitor s core collapse and had a peak magnitude of 11.2 mag in the v - band @xcite .
analysis of its pre - explosion images indicated that the initial main sequence mass of the progenitor of sn 2004dj was about @xmath7 @xcite ; see however @xcite where the progenitor had a main - sequence mass of @xmath8 .
sn 2004dj occurred at a position coincident with sandage star number 96 ( hereafter s96 , and resolved by hst to be a young star cluster ) in ngc 2403 . to date , only seven supernovae of type iip are known x - ray emitters .
only four of them are known radio emitters : supernovae 1999em , 2002hh , 2004dj and 2004et .
x - rays from circumstellar interaction has been studied in detail for type iip supernovae including sn 1999em @xcite , and 2004et @xcite .
because of its close range , sn 2004dj was easily detectable in the x - ray and radio .
the chandra x - ray observatory had targeted it multiple times early after its explosion .
the earliest detection was reported by @xcite .
the x - ray and radio luminosities are both sensitive to the mass loss rate and the initial mass of the progenitor star . in this paper
we demonstrate that for sn 2004dj the determination of the inverse compton powerlaw spectrum from the forward shock , the characteristics of the x - ray line emitting thermal plasma from the reverse shock and the radio light curve peak , uniquely determines some properties of the explosion and the progenitor star .
these include the pre - supernova mass loss rate @xmath9 , the fraction of post - shock energy density which goes into relativistic electrons @xmath10 magnetic fields @xmath11 . the time dependent nature of the thermal and non - thermal x - ray fluxes also bear signatures of the circumstellar density profile as the forward shock encounters more and more circumstellar matter and the ejecta profile as the reverse shock ploughs into the expanding ejecta .
we also see the correlation of the inverse compton flux with the optical light curve .
as the nearest ( @xmath12 mpc ) supernova since sn 1987a , sn 2004dj enjoyed extensive multi - wavelength coverage .
the details of the observations in radio , optical and x - rays , used in this work , are given below .
sn 2004dj was observed by a target of opportunity program ( pi : walter lewin , cycle : 5 , obsids : 4627 - 4630 ) using the chandra x - ray observatory on four occasions : 2004 august 9 ( see fig [ spec1 ] ) , august 23 , october 3 and december 22 ( see fig [ spec4 ] ) .
the acis - s chips were used on all occasions , without any grating , for 50 ks each .
the supernova was detected in the first of these observations @xcite .
we analyze all four observations in this work for the first time .
see table [ obstab ] for details of observations .
.observation summary of sn 2004dj with chandra [ cols="<,^,>,^ " , ] [ obstab ] before spectral analysis , the data from separate epochs were processed separately but in a similar fashion , following the prescribed threads from the chandra science center using ciao 4.4 with caldb 4.4.8 . the level 2
events were filtered in energy to reject all counts below 0.3 kev and above 10 kev .
the resulting events were mapped and the supernova easily identified .
the source region was marked off and a light curve was generated from the rest of the counts .
flare times were identified in this light curve , as time - ranges where the count rate rose above 3 times the rms .
this resulted in a good time interval table which was used to further filter out the events .
the spectrum , response and background files were then generated for these good events . in order to obtain the highest available spectral resolution
, the data were not binned ; unbinned data were analyzed in the next steps .
sn 2004dj has been observed extensively in radio bands from aug 2004 until may 2007 with various telescopes including the very large array ( vla ) , giant metrewave radio telescope ( gmrt ) and multi - element radio linked interferometer network ( merlin ) .
merlin observed sn 2004dj extensively in the 5 ghz band covering the period from 2004 august 5 to december 2 .
these observations are reported in @xcite .
the gmrt observations started on 2004 aug 12 and continued until 2007 may 22 .
the observations were made in the 1420 , 610 , 325 and 235 mhz bands .
the first two gmrt observations in 1420 mhz band are reported in @xcite .
the vla followed sn 2004dj extensively starting from aug 1 , 2004 until may 28 , 2007 .
the observations were made in all vla bands starting from 1.4 ghz band upto 44 ghz band .
the first 8.4 ghz observation was reported by @xcite .
@xcite discussed these published radio observations of sn 2004dj to interpret the physical properties of sn 2004dj and its interaction with the circumstellar matter .
they established free free absorption to be the radio absorption mechanism based on the published radio observations .
a comprehensive paper including all the radio observations and their detailed interpretation is underway ( chandra et al . , 2012 , under preparation . )
supernova ejecta hits the pre - explosion wind at a velocity much larger than its characteristic sound speed .
this creates a strong forward shock moving into the circumstellar matter arising from the mass loss from the progenitor @xcite . in case of a type iip supernova ,
the material behind this shock is too hot ( @xmath13 kev ) and tenuous to contribute significantly to the x - ray flux in the chandra bands ( @xmath14 kev ) through its thermal emission .
however , the radio emission from type iip supernovae is said to arise from the non - thermal electrons accelerated in this region @xcite . some optical photons from a supernova may also be inverse compton scattered off energetic electrons in this region into the x - rays . scattering from a non - thermal population of electrons with a power law index @xmath15 generates a spectral index of @xmath16 in the optically thin radio synchrotron .
so the inverse compton spectrum may be modeled as a power law with a photon index of @xmath17 . in the frame of the ejecta
however , the circumstellar matter impinges supersonically on the ejecta . according to @xcite this drives an inward ( in mass coordinates ) propagating wave which reheats the ejecta , which would have otherwise been adiabatically cooled by its rapid expansion , to around @xmath18 kev .
a supernova typically ejects more matter at lower velocity . as the reverse shock moves into
slower moving ejecta , its heats up more gas which can contribute to the x - ray flux .
the plasma in this region can be modeled as a collisionally - ionized diffuse hot gas .
kev . ]
we imported the extracted spectra into xspec 12.7.1 for further analysis .
the data from all epochs were jointly fitted with the combination of a simple photon power law using _ powerlaw _ and collisionally - ionized diffuse gas using _ apec _ models @xcite , passed through a photo - electric absorbing column using _
wabs_. the column density for wabs , plasma temperature for apec and photon index for powerlaw were shared between all epochs . the emission measure for apec and normalization for powerlaw
were determined separately for each epoch from the joint analysis .
as , the fits were performed on unbinned data , each bin would have too few photons for the @xmath19 statistic to be useful .
we therefore used the cstat statistic based on @xcite . in order to evaluate the goodness of fit , we simulated 10000 spectra based on the best fit model and found out that only 25% of these simulations had cstat less than that for the data .
we therefore conclude that this model provides a good fit to the data .
the best fit absorption column density is @xmath20 .
see table [ obstab ] for model fluxes on each date .
a self - similar solution for the interaction of the fast moving ejecta and the nearly static ( negligibly moving ) circumstellar matter was found by @xcite .
this can be used to generate a complete description of the forward and reverse - shocked material .
following @xcite we consider supernova ejecta described by @xmath21 where the density falls due to free expansion and higher velocities have less matter , described by the power law index @xmath22 .
this ejecta interacts with the circumstellar matter , described by a power law profile @xmath23 . for a steady wind ( @xmath24 )
, we have @xmath25 where @xmath26 is the mass loss rate and @xmath27 is the wind velocity of the progenitor . here
@xmath28 @xcite .
shock jump conditions for a fast shock dictate that the density behind the forward shock is given by @xmath29 . in the thin shell approximation , this can be further related to the density behind the reverse shock as @xmath30 according to @xcite @xmath22 is expected to be 12 for a red supergiant progenitor . using this value
simplifies the above expression to @xmath31 .
therefore the reverse - shocked material is much denser than the circumstellar matter .
so it can cool efficiently via radiative processes and contribute to the thermal x - ray flux .
the velocity of a shock determines the energy imparted to particles crossing it , and hence the post - shock plasma temperatures .
furthermore , the forward and reverse shock velocities are connected by jump conditions once the ejecta and circumstellar matter profile are specified .
so , the post - shock temperature of ejecta which has crossed the reverse shock can be expressed in terms of the blastwave velocity .
since the former is an observable , the latter can be derived from it .
give the temperature of the reverse - shocked material @xmath32 , as @xmath33 where @xmath34 is the mean atomic weight per particle , @xmath22 is the power law index of the ejecta profile , @xmath35 is the power law index of the circumstellar density profile and @xmath36 is the forward shock velocity in units of @xmath37 km s@xmath38 .
@xmath34 is 0.61 in the solar - like outermost zone of the progenitor .
@xmath35 is 2 for a circumstellar matter set up by a steady wind , and we again use @xmath39 . with these assumptions , the forward shock velocity can be expressed as @xmath40 since , the best fit temperature is @xmath41 kev ( see fig [ t_ind ] ) , the implied velocity is @xmath42 km s@xmath38 .
this is consistent with the fastest - moving ejecta in a typical type iip supernova .
therefore , the observed thermal component of the x - ray spectrum can be explained as emission from reverse - shocked material , provided this is the blastwave velocity .
@xcite have proposed diagnostics for circumstellar interaction in type iip supernovae using the detection of high - velocity absorption features in h@xmath43 and he i lines during the photospheric stage .
the highest velocity h@xmath43 absorption feature observed in the optical spectra of sn 2004dj is at 8200 km s@xmath38 between 64 @xcite and 102 days @xcite with respect to the explosion date of june 28 proposed by @xcite .
this overlaps with the epoch of chandra observations and provides a lower limit to the allowed forward shock velocity , so the velocity that we obtain is consistent with constraints from optical spectroscopy .
the reverse - shocked matter contributes bulk of the thermal x - rays . in order to find the circumstellar density it is necessary to estimate the amount of reverse - shocked material in terms of the forward - shocked material .
using equations [ rw ] and [ r0 ] and again assuming @xmath39 , we get @xmath44 similarly , following we can express the total mass of reverse - shocked material as @xmath45 where @xmath46 is the blastwave radius .
multiplying these two results , we get , @xmath47 these have to be converted to number densities to get the emission measure which determines the thermal x - ray flux . for a solar - like composition appropriate for the outermost shells of the supernova , @xmath48 . using these , we have the emission measure for the reverse - shocked material as @xmath49 we estimate the shock radius from the velocity determined in the previous sub - section as @xmath50 , with the time calculated from the explosion date of 2004 june @xmath51 @xcite or jd 2,453,168 . as all the other quantities are known , the circumstellar density can now be determined from the emission measure .
the apec fit to the thermal part of the x - ray spectrum of sn 2004dj gives , @xmath52 ^ 2}\int n_e n_h dv = ( 7.5\pm2.7)\times10^{-6}\;,\ ] ] for the first epoch ( see fig [ norms1 ] ) .
ngc 2403 is nearby and has negligible redshift .
we adopt a distance to ngc 2403 of 3.06 mpc based on cepheids from @xcite . eliminating the emission measure between the last two equations , we have
, @xmath53 therefore the mass loss rate of the progenitor was @xmath54 this gives us @xmath55 ; however @xcite define a non - dimensional @xmath56 . for sn 2004dj
, we therefore get @xmath57 , which can now be compared with estimates for circumstellar density from the radio emission .
electrons are believed to be accelerated to relativistic velocities in the forward shock .
these are in turn , responsible for the radio emission from supernovae .
the simplest model by @xcite , is to assume that a fraction @xmath58 or @xmath59 of the thermal energy is used to accelerate electrons and amplify magnetic fields respectively . as a result
the radio emission from a supernova is dependent upon these fractions and does not measure the circumstellar density directly .
instead , according to @xcite , the radio emission measures @xmath60 where @xmath61 is the peak flux at peak frequency @xmath62 at @xmath63 days from the explosion and @xmath64 .
the equipartition factor is defined as @xmath65 . using the @xmath66 ghz , radio light curve from @xcite for sn 2004dj
, we have @xmath67 . the same relativistic electrons which generate the radio spectrum via synchrotron emission may also contribute to the x - ray flux by inverse compton scattering the optical photons from the supernova s peak bolometric luminosity of around @xmath68 .
this may explain the dominant non - thermal component seen in the x - ray spectrum at early times ( see fig [ spec1 ] ) .
for an electron index @xmath69 this is expected to generate a powerlaw with photon index 2 , consistent with the observations ( see fig [ t_ind ] ) .
following @xcite the normalization of the inverse compton flux at 1 kev can be written as @xmath70 where @xmath71 is the minimum lorentz factor of the relativistic electrons .
the normalization of the inverse compton flux obtained for the first epoch ( @xmath72 ) of chandra observations , is seen to be @xmath73 ( see fig [ norms1 ] ) . substituting this in the left hand side and using a @xmath67
as found in this work , @xmath74 as implied by the temperature of the reverse - shocked plasma , @xmath75 from @xcite , we get @xmath76 where @xmath77 if the electron spectrum extends all the way down to those at rest . however
, if one only considers relativistic electrons with @xmath78 as used by @xcite for sn 2002ap , it implies @xmath79 .
we prefer this latter ( more realistic ) assumption as the bulk of the accelerated electrons can not be at rest or very low velocities .
so we shall use this latter value in further calculations .
now that we know @xmath80 from radio synchrotron , @xmath81 from thermal x - rays and @xmath43 from inverse compton , we can use the definition of @xmath80 to get @xmath82 which in turn implies @xmath83 hence we have determined both @xmath58 and @xmath59 using a combination of x - ray , radio and optical data . function . note that the data do not rule out a constant mass loss rate model , but suggest one with variable mass loss . ]
the first things that are apparent from the chandra observations of sn 2004dj is that the total x - ray flux softens and falls with time .
these gross features have already been observed in another type iip supernova , sn 2004et by @xcite .
however , the quality of the sn 2004dj spectra allows a quantitative explanation for the first time .
while both the thermal and non - thermal components fall eventually , their proportion changes with time . at early times , the spectra is dominated by the inverse compton flux and is therefore harder ( see fig [ spec1 ] ) .
as the bolometric light curve of the supernova decays , the source of seed photons to be scattered turns off .
so at late times , the spectra is dominated by emission from the reverse - shocked thermal plasma and is therefore softer ( see fig [ spec4 ] ) .
since the post shock temperature in a self similar blastwave does not change much , as the shock velocity is slowly varying , the variation in the thermal x - ray flux would come from the variation in the emission measure .
we have already derived in equation [ em ] that the emission measure scales as @xmath84 .
therefore , if the radius scales approximately as @xmath50 , then we have @xmath85 for a constant pre - explosion mass loss rate . we comparing this to the temporal variation of the normalization of the apec flux from sn 2004dj ( see fig [ t_lc ] ) .
while a constant mass loss rate can not be ruled out , there is a hint of variation up to 50% from a constant mass loss .
the time variation in the inverse compton flux comes from the expansion if the blastwave and the dimming of the supernova s supply of seed photon . @xcite
have shown that the normalization of the inverse compton flux scales as @xmath86 thus the inverse compton flux is expected to behave as @xmath87 in the plateau phase where @xmath88 is nearly constant for a type iip supernova .
thereafter , as the supernova luminosity falls off , the inverse compton flux is expected to fall off from this behavior .
this prediction is borne out by the observations as the first two epochs ( see fig [ ic_lc ] ) which lie in the plateau phase can be fitted by a @xmath87 function , while the late time flux normalization falls off significantly below this curve . function . the time interval is measured similarly as in fig [ t_lc ] . note that the first two epochs , for which @xmath88 is nearly constant , lie on the line .
also note that one the @xmath88 decreases due to the termination of the plateau phase beyond 100 days , the inverse compton flux also falls off . ]
our results have wide ranging implications for the progenitors of type iip supernovae , conditions of the plasma in the supernova shock waves , particle acceleration and magnetic field amplification schemes .
these are discussed briefly below . according to @xcite
the progenitor of sn 2004dj had a mass of @xmath89 . in this mass range
the progenitor dies as a red supergiant and @xcite find theoretical mass loss rates in the range @xmath90 for such progenitors .
however , we find pre - explosion mass loss rate for the progenitor of sn 2004dj to be @xmath91 . this can be reconciled if the wind velocity is @xmath92 , which is unlikely as the wind velocity of a typical red supergiant is around @xmath93 .
however , these rates have been calculated for stars of solar metallicity .
@xcite suggest that the metallicity of the sn 2004dj site is @xmath94}=8.4 $ ] . assuming a solar value of 8.7 , this implies a host metallicity of @xmath95 .
using @xmath96 following would bring down the theoretical values closer to the observed one .
stars in the mass - range 11 - 19 @xmath97 were evolved using mesa @xcite for three different metallicities z=0.4 , 0.50 and 0.75 @xmath98 . the mass loss scheme that was used
is described as the _ dutch _ scheme in the mesa code ; see sec 6.6 mesa code paper by @xcite .
this scheme switches on a rgb wind depending upon the burning stage .
the scheme takes care of the surface temperature changes in the different stages of stellar evolution and adjusts itself accordingly .
the mass loss in the rgb branch , which is relevant for the final stages of type iip progenitors , follows .
the stars were evolved till the central density had reached @xmath99
g @xmath100 .
the calculated mass loss was the average value over the last century of evolution .
this is the mass which the blastwave sweeps up during first few years of its evolution .
metallicity , and theoretical line ( dot - dashed cyan ) from ( with @xmath101 ) plotted for comparison .
error bars are @xmath102 confidence intervals .
note the disagreement between the observed properties of the sn 2004dj progenitor and the theoretical models . ] in fig [ mmdot ] we compare the progenitor mass of sn 2004dj as determined by @xcite and mass loss rate determined in this work , with the expectation from stellar evolution and standard mass loss rate prescriptions .
there is an obvious disagreement between the observed mass loss rate and theoretical predictions .
such a disagreement was also noted by @xcite who determined mass loss rates for the sn 1999em and sn 2004dj using absorption lines from optical spectroscopy .
the analysis of x - ray spectra to yield parameters for the underlying explosion models of young supernovae and snrs is significantly complicated by the possibility that their x - ray emitting plasmas may not be ionization equilibrium .
the x - ray emission comes from an impulsively heated ( by the blastwave or the reverse shock ) uniform and homogeneous gas which is initially cold and neutral . for young extragalactic supernovae , which are point sources contained within the telescope beam - size , the x - ray spectra are spatially
averaged over the whole object . in the case of non - equilibrium plasma
the spectra depend upon the shock velocity @xmath103 , the ionization timescale @xmath104 ( the product of the supernova s age @xmath105 and the post - shock electron number density @xmath106 ) and the extent of electron heating at the shock front ( which determine how different the electron and ion temperatures @xmath107 are ) .
spectra consisting of lines whose strengths depend upon the heavy elements abundances and their relative fractions . both x - ray continuum and
lines arise from electrons undergoing bremsstrahlung or electron impact excitation of ions . @xcite
have emphasized that for snrs ionization equilibrium models are often not adequate , and the constant - temperature , single - ionization timescale nei model is better than an equilibrium model .
young supernovae have shorter lifespans till their present age but may be expanding through denser circumstellar media and so these issues of adequacy of nei vs ie models may or may not be relevant . because substantial x - ray data from chandra exists in multiple epochs for this target , we investigate the relevant parameter space to see whether nei models would constrain emission model parameters sufficiently well to require their use ( as opposed to ie models ) .
the nei model is an xspec provides a non - equilibrium ionization collisional plasma model which uses aped @xcite to calculate the resulting spectra .
it assumes a constant temperature and single ionization timescale parameter . in nei models the ionic states of each abundant element ( e.g. , c , n , o , ne , mg , si , s , ca , fe , ni ) are solved for through time - dependent ionization equations .
apec @xcite , which is used throughout this work , on the other hand does an ionization equilibrium calculation for x - ray emission spectra of collisionally - ionized diffuse gas calculated using the atomdb code v2.0.1 ( http://www.atomdb.org/ ) that includes a set of relevant ions and their lines and perform a population calculations .
the atomic data and line transitions information are at a more comprehensive state than other models like mekal and has a wavelength coverage from ir to x - ray . however , when the resulting grid of feasible values for @xmath104 is sampled it was noted that the data does not constrain it at all . in case of sn 2004dj , the post reverse shock density at the first observation epoch
is @xmath108 , and the age is @xmath109 s. this will give a @xmath110 , which should be enough time for the plasma to come to ionization equilibrium .
this may not be strictly true for the material behind the forward shock , however it would be too hot and too low density to contribute significantly to the chandra spectrum .
therefore it is safe to make the assumption of ionization equilibrium while modeling the chandra x - ray spectrum of sn 2004dj , although higher resolution data such as will be available with astro - h might reveal nei features .
radio or x - ray light curves of supernovae are often explained using varying amounts of accelerated electrons and amplified magnetic fields .
however it is difficult to ascertain the relative contribution of each . by modeling the observed x - ray spectrum as a combination of thermal and inverse compton flux
, we have been able to determine the equipartition factor for sn 2004dj .
our result shows that the plasma is away from equipartition , with electrons having more of the total energy than magnetic fields .
since @xmath58 and @xmath59 have also been found individually , theories of particle acceleration and magnetic field amplification can now be tested against them .
this is an unique opportunity .
since the electrons and magnetic fields are responsible for producing the synchrotron spectra of supernovae , it is important to look breaks in the radio spectrum of a young supernova .
@xcite associate such a break in sn 1993j ( not a type iip , but a type iib supernova ) with the phenomenon of synchrotron aging of radiating electrons . the magnetic field in the shocked region
can be calculated from the break in the spectrum , independent of the equipartition assumption between energy density of relativistic particles and magnetic energy density .
the synchrotron cooling break appears at a certain frequency when electrons radiating above that frequency have lost a significant portion of their energy to synchrotron emission .
demanding that the synchrotron loss time scale is equal to the age of the supernova , we can use equation 5 of @xcite to express the frequency of the synchrotron cooling break as @xmath111 which has the same temporal evolution as derived in @xcite . for the epochs at which it was observed in the radio
this is much higher than the frequencies accessible with the evla , therefore synchrotron cooling will be unimportant for studying the radio light curves of sn 2004dj . however for future studies of young type iip supernovae , these frequencies may become detectable with mm wave observatories such as the sma or alma .
another important mechanism for electron cooling is inverse compton losses against the low energy photons from the supernova photosphere . due to this process
high energy electrons may lose bulk of their energy in boosting the seed photons to x - rays , thereby contributing to the x - ray flux , as discussed in this paper for sn 2004dj .
demanding that the inverse compton loss time scale is equal to the age of the supernova , we can use equation 6 of @xcite to express the frequency of the inverse compton cooling break as @xmath112 for the determined relevant values of the parameters determined in this work , the @xmath113 is within the frequency coverage of the vla .
therefore , any interpretation of the radio light curve of sn 2004dj should account for inverse compton losses and expect to see such a cooling break .
in this work we have used sn 2004dj as a prototype for type iip supernovae to answer some fundamental questions about their circumstellar interaction . 1 .
what is the origin of x - rays detected by the chandra from young type iip supernovae ?
the x - rays detected by chandra arise from a combination of thermal and non - thermal processes .
they are dominated at early times by optical supernova photons inverse compton scattered by relativistic electrons at the forward shock which falls off as @xmath114 . at late times
it is dominated by thermal emission from reverse shock heated plasma which only falls as @xmath87 .
2 . how fast was the progenitor losing mass via winds in the final phase before explosion ?
the progenitor was losing mass at a rate of @xmath91 .
this is less than the value expected from mass loss prescriptions in the literature for the putative @xmath115 progenitor .
3 . how efficiently does the supernova accelerate cosmic ray electrons in its forward shock ? around a third ( @xmath116 ) of the energy thermalized by the collision of the ejecta with the circumstellar matter is used in accelerating electrons to relativistic velocities .
what is the extent of turbulent magnetic field amplification in the post - shock material ? around a tenth ( @xmath117 ) of the thermal energy available is used in the turbulent amplification of magnetic fields .
5 . are synchrotron and inverse compton losses important in explaining the radio lightcurves ? in the case of sn
2004dj synchrotron losses are unimportant while inverse compton losses will have to be taken into account while explaining its radio lightcurve . | type iip ( plateau ) supernovae are the most commonly observed variety of core collapse events .
they have been detected in a wide range of wavelengths from radio , through optical to x - rays .
the standard picture of a type iip supernova has the blastwave interacting with the progenitor s circumstellar matter to produce a hot region bounded by a forward and a reverse shock .
this region is thought to be responsible for most of the x - ray and radio emission from these objects .
yet the origin of x - rays from these supernovae is not well understood quantitatively .
the relative contributions of particle acceleration and magnetic field amplification in generating the x - ray and radio emission need to be determined . in this work
we analyze archival chandra observations of sn 2004dj , the nearest supernova since sn 1987a , along with published radio and optical information .
we determine the pre - explosion mass loss rate , blastwave velocity , electron acceleration and magnetic field amplification efficiencies .
we find that a greater fraction of the thermal energy goes into accelerating electrons than into amplifying magnetic fields .
we conclude that the x - ray emission arises out of a combination of inverse compton scattering by non - thermal electrons accelerated in the forward shock and thermal emission from supernova ejecta heated by the reverse shock . |
w 39 is a large region in the inner galaxy ( @xmath6 , @xmath7 ) first identified in the @xcite 1.4 ghz radio continuum survey and more recently cataloged as a large infrared `` bubble '' by both professional astronomers ( n24 ; * ? ? ?
* ) and citizen scientists ( mwp1g18908 - 003146 ; * ? ? ?
it is a significant high - mass star - forming region and a possible example of sequential or triggered star formation based on the hierarchical bubble structure seen in the infrared , but , owing to its highly obscured location in the inner galaxy , it has been poorly studied .
recent high - resolution ( sub - arcminute ) surveys at radio and infrared wavelengths have provided an opportunity to examine this region in detail for the first time .
this paper presents an analysis of the region itself along with an investigation of the influence of the region on star formation in its vicinity . in the next section
the key parameters of the various data sets used in the study are summarized .
the determination of the radial velocity and distance to the region is described in [ sec : rvd ] .
an analysis of w 39 and three associated compact regions follows in [ sec : ghii ] . in [ sec : yso ] the young stellar object ( yso ) distribution surrounding w 39 is explored , and our conclusions are presented in
[ sec : conc ] .
at cm radio wavelengths we use vla galactic plane survey ( vgps ; @xcite ) 21-cm line and continuum data .
the vgps maps have an angular resolution of @xmath8 arcmin , and the line spectra have a velocity resolution of 1.56 km s@xmath2 .
the maps include single - dish short - spacing data , making them ideal for obtaining flux density measurements of large angular scale ( @xmath9 arcmin ) objects like w 39 .
to trace molecular gas we use @xmath10co ( j@xmath111@xmath120 ) observations from the boston university / five college radio astronomy observatory galactic ring survey ( bu / fcrao grs ; @xcite ) .
these data have an angular resolution of 46 arcsec and 0.21 km s@xmath2 velocity resolution .
we also make use of 1.1 mm continuum data from the bolocam galactic plane survey ( bgps ; @xcite ) . at infrared wavelengths we primarily use data from the glimpse @xcite _ spitzer _ legacy survey .
glimpse provides 1.7 to 2.0 arcsec resolution images in four bands centered at 3.6 , 4.5 , 5.8 , and 8.0 @xmath13 m along with a point source catalog that contains near- and mid - infrared photometric data .
in addition , we use the 6 arcsec resolution images at 24 @xmath13 m from the mipsgal _ spitzer _
legacy survey @xcite . for near - infrared photometry of point sources we use data from the two - micron all sky survey ( 2mass ; @xcite ) and the united kingdom infrared telescope ( ukirt ) infrared deep sky surveys - galactic plane survey ( ukidss - gps ; @xcite ) .
in figure [ fig : w39cont ] we show the vgps 1420 mhz continuum map of the area around w 39 along with the position and velocity of cm - wavelength observations of the radio recombination lines ( rrl ) h85@xmath14 , h87@xmath14 , and h88@xmath14 @xcite .
w 39 has a bubble morphology best seen in the infrared . at radio wavelengths the brightest emission ( @xmath15 k ) from w 39 is concentrated along a u - shaped ridge opening to lower galactic longitudes
. fainter emission ( @xmath16 30 40 k ) is found within the bubble and also tracing the bubble edges at lower galactic longitude . there is a local minimum in the radio emission around @xmath17 that is a likely location for the exciting ob star(s ) .
three of the rrl observations that lie on the strong ridge of emission have very similar velocities , and we adopt the average of these values , @xmath18 km s@xmath2 , as the radial velocity of w 39 .
the rrl velocity of g18.95 - 0.02 is significantly different from the main ridge and , as we discuss in [ sec : irmorph ] , this region is likely not directly associated with w 39 .
the nature of the rrl emission associated with the arc - shaped structure g18.64 - 0.29 is more speculative .
both @xcite and @xcite classify this object as a supernova remnant ( snr ) based on multi - frequency radio observations , and absorption studies by @xcite place the object at a distance of @xmath19 kpc .
rrl emission is not expected from snrs so it is likely that the observed rrl emission , with a difference in velocity of only @xmath19 km s@xmath2 from the main w 39 ridge , is originating from the edge of the w 39 bubble .
absorption spectra , obtained from the vgps , were used to determine if all three of the rrl sources are located at the same near or far kinematic distance . a radio continuum source in the inner galaxy at the near distance
can only show absorption up to its @xmath20 while a source at the far distance can show absorption at velocities beyond @xmath20 up to the tangent velocity , which at @xmath21 is 110 to 140 km s@xmath2 depending on the kinematic model used ( see [ sec : kindis ] ) . for each of the rrl sources `` on '' spectra
were obtained by averaging spectra where the 1420 mhz continuum brightness temperature was above a threshold of 100 k for g19.07 - 0.28 , 100 k for g18.88 - 0.49 , and 55 k for g19.04 - 0.43 .
`` off '' spectra were then calculated by taking the average spectrum of a nearby ( 12 arcmin offset ) similarly sized region .
the resulting on - off absorption spectra are shown in figure [ fig : hispec ] .
the uncertainty in an absorption spectrum @xmath22 can be quantified by taking the maximum of either the standard deviation in channels where no absorption is expected to occur ( e.g. , at negative @xmath20 in this case ) or by constructing `` off - off '' spectra that measure the intrinsic variability of emission structure .
we found @xmath23 k in each case .
looking at figure [ fig : hispec ] we see that the two brightest rrl sources are clearly at the near kinematic distance because of the lack of absorption at velocities @xmath24 km s@xmath2 .
the ridge rrl source has a much poorer quality spectrum , but there is no evidence for absorption beyond @xmath25 km s@xmath2 . we conclude that all of the emission associated with the three rrl observations is at the near kinematic distance . in figure [ fig : ir8_24 ]
we show an 8 and 24 @xmath13 m image of the w 39 region using data from _
glimpse and mipsgal .
the infrared bubble structure , best seen at 8 @xmath13 m and identified by @xcite and @xcite , is indicated by an ellipse . as expected for a thermal radio source w 39
is very bright in the infrared .
of particular interest though is the layered structure of the infrared emission ; moving from the interior of the u - shaped structure to the exterior , we first see 24 @xmath13 m emission followed by a layer of 8 @xmath13 m emission .
this is exactly what would be expected for a region being heated from one side by ionizing radiation . within the region
polycyclic aromatic hydrocarbons ( pah ) are strongly depleted allowing 24.0 @xmath13 m emission from small dust grains to dominate while in the photodissociation region ( pdr ) surrounding the region pahs can survive resulting in strong 8 @xmath13 m emission ( e.g. , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
the continuity of this layered infrared morphology is consistant with w 39 being a single large region .
we note that this layered infrared morphology does not extend to the region around @xmath26 . in figure
[ fig : co ] we show grs maps of @xmath10co emission between @xmath27 km s@xmath2 integrated in 20 km s@xmath2 increments between 0 and + 40 km s@xmath2 , 10 km s@xmath2 increments between 40 and + 70 km s@xmath2 , and a single 60 km s@xmath2 interval between + 70 and + 130 km s@xmath2 .
the g18.95 - 0.01 region is morphologically very similar to co emission seen in the grs at @xmath28 km s@xmath2 consistent with the lower rrl velocity observed by @xcite .
based on this we do not consider this region to be part of w 39 .
the co maps also show that most of the molecular gas along the line of sight to w 39 is found between + 60 and + 70 km s@xmath2 .
the ( near ) kinematic distance to w 39 was calculated using @xmath18 km s@xmath2 and three different galactic rotation models : 1 ) the r@xmath29= 8 kpc hydrodynamical model of @xcite , 2 ) the r@xmath29=8.5 kpc , v@xmath29=220 km s@xmath2 model of @xcite , and 3 ) a version of the @xcite model scaled to r@xmath29=8.0 kpc .
the resulting distances are 4.5 kpc , 4.7 kpc , and 4.4 kpc respectively . in each case
the error in the calculated distance caused by the uncertainty in @xmath20 is only @xmath30 kpc , much less than the systematic uncertainty associated with the choice of rotation model .
we adopt the average ( standard deviation ) of these values as the distance ( uncertainty ) to w 39 : @xmath31 kpc . at this distance , w 39 is probably part of the scutum - centaurus arm of our galaxy @xcite .
the flux density at 1420 mhz of w 39 , @xmath32 jy , was measured from vgps data using imview @xcite to integrate the emission within a user - defined area roughly following the @xmath33 k contour , but omitting emission at @xmath34 and subtracting the contribution from compact and point sources found within the contour . the main contribution to
the uncertainty comes from the definition of the background level which varies depending on the number and location of the points used to define it and the order of the two - dimensional polynomial fit used . the observed radio continuum flux density ( @xmath35 ) of an region at a distance ( @xmath36 )
can be related to the ionizing radiation luminosity ( @xmath37 ) of the star(s ) located within the region by : @xmath38 where @xmath39 is the fraction of the ionizing radiation from the star(s ) that is intercepted by the region ( e.g. , @xmath40 for an ionization bounded region , @xmath41 for a `` blister '' region ) , and we used @xmath42 ghz and @xmath43 k to calculate the constant @xcite . as pointed out by @xcite equation [ eqn : logq ] does not depend on the details of the density structure of the region . for @xmath44
pc and @xmath45 jy we find @xmath46 , equivalent to the output of two o3 v stars ( or 10 o7 v stars ) using the calibration of @xcite .
varying distance and @xmath35 within the uncertainties , and allowing @xmath47 to range between 7500 k and 10,000 k , we find the full range of @xmath48 . given the morphology of the region it is very likely that @xmath49 and w 39 is a giant region ( ghiir ; * ? ? ?
* ) being powered by numerous ob stars .
the formulas given in @xcite , and derived in detail in @xcite , can be used to estimate the average electron density of the region and the total mass of ionized gas given @xmath35 , @xmath36 , @xmath47 , and the angular size of the region .
an appropriate size scale for w 39 is given by the separation between the two sides of the region , @xmath50 , or 30 pc at 4.5 kpc .
this is slightly smaller than the value of @xmath51 reported in @xcite , but that value included emission at higher galactic latitude that we do not associate with the region .
we find values of @xmath52 @xmath53 and @xmath54 m@xmath29 .
unlike the @xmath55 calculation these values do depend strongly on the details of the region s density structure and should only be considered order - of - magnitude estimates .
the radio source g19.07 - 0.28 is the brightest of the compact radio sources found on the periphery of w 39 with @xmath56 jy . using the same formulas and calibrations from the previous subsection we find @xmath57 equivalent to a single o8 v star .
the ionized gas mass is @xmath58 m@xmath29 and the average electron density is @xmath59 @xmath53 .
this radio source is also the infrared source iras 18239 - 1228 and has four well - defined iras flux density measurements . using the technique of @xcite we calculate the far - infrared luminosity @xmath60 which is the luminosity of an o7 - o8 v star .
the higher resolution glimpse 8.0 @xmath13 m image of the region is shown in figure [ fig : bubble ] . at this wavelength
the source is clearly a bubble structure with a radius of @xmath61 ( @xmath62 pc ) .
ionized gas , traced by the 1420 mhz radio emission , and warm dust , traced by 24 @xmath13 m emission , fill the central region .
the compact bubble morphology of the region and the agreement in the stellar spectral type determined by the radio and infrared flux support the idea that the region is ionization bounded ( @xmath63 ) .
the radio source g18.88 - 0.49 is the second brightest of the compact radio sources found on the periphery of w 39 with @xmath64 jy . from this
we find @xmath65 equivalent to a single o8 v star .
the ionized gas mass is @xmath66 m@xmath29 and the average electron density is @xmath59 @xmath53 . unlike g19.07 - 0.28
there is no iras point source associated with this region .
the higher resolution glimpse 8.0 @xmath13 m image of the region is shown in figure [ fig : west ] .
although the morphology is more complex than g19.07 - 0.28 , the presence of bright - rimmed structures at the edge of the radio continuum emission along with the coincident 24 @xmath13 m emission suggest that we are also looking at a dusty , ionization - bounded , bubble structure similar in size to g19.07 - 0.28 .
the remaining radio source on the periphery of w 39 is g18.94 - 0.43 .
we measured @xmath67 jy , which corresponds to @xmath68 , equivalent to a single o9 v star .
the ionized gas mass is @xmath69 m@xmath29 , and the average electron density is @xmath70 @xmath53 . as with g18.88 - 0.49 there is no iras point source associated with the region .
the high resolution image of the region shown in figure [ fig : west ] reveals a complex morphology that is difficult to interpret in detail . as we would expect for an region there appears to be bright - rimmed structures along the low - latitude portion of the region and a minimum in the 8 @xmath13 m emission coincident with the radio continuum maximum .
the radio continuum view of the w 39 region is of an evolved massive hii region ( powered by multiple mid- to early - type o stars ) with multiple areas of less intense massive star formation ( i.e. , the three regions powered by single late o stars ) occurring along the periphery of the region .
one possibility is that these secondary regions of star formation have been triggered by the expansion of the w 39 region .
while proving this idea is difficult we can at least investigate the relevant timescales for the most clearly defined secondary region , g19.07 - 0.28 , to see if this scenario is at all plausible .
we can obtain a rough estimate of the age of g19.07 - 0.28 using the classic description of the evolution of an region @xcite .
the o star first creates a ionized gas region about the size of the strmgren sphere on an essentially instantaneous timescale of @xmath71 yr .
the overpressurized region then expands from its initial radius , @xmath72 to the observed radius @xmath73 at time @xmath74 as , @xmath75 where @xmath76 is the isothermal sound speed in the ionized gas ( @xmath77 km s@xmath2 ) . to estimate the initial density of the cloud in which the region formed we examined the nearby 1.1 mm source from the bgps , g019.077 - 00.287 .
the peak brightness of this source is 2.3 jy beam@xmath2 corresponding to a column density of @xmath78 @xmath79 using equation a.27 in @xcite .
the deconvolved radius of the source is @xmath80 ( @xmath81 pc ) . adopting a path length of 4 pc we obtain a density @xmath82 @xmath53 .
in such an environment an o8 v star would rapidly create an region of size @xmath83 pc . using this value and the observed size of the region @xmath84 pc in equation [ eqn : h2evo ] we find @xmath85 years . while this is clearly a rough estimate it is important to note that this timescale is approximately one order of magnitude smaller than the timescales appropriate for the evolution of the much larger w 39 region .
as pointed out by @xcite , an approximate lower limit to the age of an evolved region can be obtained by the sound crossing time , @xmath86 years in the case of w 39 .
the lack of non - thermal radio emission from snrs gives a consistent upper limit to the age of @xmath87 years based on the main sequence lifetime of early o stars @xcite .
we conclude that the probable timescales associated with the expansion of w 39 ( @xmath88 years ) and the likely ages of the secondary regions of massive star formation ( @xmath89 years ) are consistent with a scenario of triggered star formation .
in order to determine if the only recent , and thus likely triggered , star formation activity surrounding w 39 is restricted to the three compact regions discussed previously , we used the _ spitzer _
glimpse catalog to identify yso candidates ( ysos hereafter ) in the w 39 field within @xmath90 , @xmath91 .
the high latitude cutoff was chosen to avoid including the prominent infrared dark clouds ( irdcs ) found at @xmath34 which are not associated with w 39 .
a total of 10657 sources with valid magnitudes in all four irac bands were identified .
figure [ fig : ccd ] is a color - color ( cc ) diagram showing all of these sources along with the yso classification criteria from @xcite .
we identify 57 class i ysos , 129 class ii ysos and 57 class i / ii ysos ( [ 3.6]-[4.5 ] @xmath92 and [ 5.8]-[8.0]@xmath93 ) .
class i ysos are objects with a substantial circumstellar envelope , class ii ysos have most of the circumstellar material residing in a disk , and class i / ii objects are most likely class ii or stellar objects with poor [ 8.0 ] photometry causing them to scatter into this region of the cc diagram . in the canonical picture
, a class i yso will evolve to a class ii yso with each stage of evolution lasting for @xmath94 years and @xmath95 years respectively @xcite . in figure
[ fig : cmd ] the sample is plotted in a [ 3.6]-[4.5 ] vs. [ 3.6 ] color - magnitude ( cm ) diagram .
we see there is a bright cutoff at [ 3.6 ] @xmath96 due to saturation and a faint , color - dependent , sensitivity cutoff at [ 3.6 ] @xmath97 . included in the cm diagram are two lines showing the position of main - sequence stars ( b0 v to m5 v ) and giant stars ( g5 iii to m0 iii ) at a distance of 4.5 kpc . the crosses are t - tauri stars from @xcite shifted from the distance of taurus - auriga to w 39 .
the diamonds are intermediate - mass herbig aebe stars from @xcite and @xcite again shifted to the distance of w 39 . for the herbig aebe sample we have used l and m band magnitudes as rough proxies for [ 3.6 ] and [ 4.5 ] .
finally , the asterisks represent agb stars from the _ spitzer_/irac study of galactic agb stars by reiter , marengo , & fazio ( in preparation ) .
the cm diagram illustrates that our sample is primarily sensitive to intermediate - mass ysos at a distance of 4.5 kpc .
the cm diagram also provides some insights into likely contaminants in our sample .
we see that the majority of background main - sequence stars are too faint to be detected while foreground main - sequence stars with moderate amounts of reddening will be a source of contamination for the least - red class ii ysos . the major contaminant , for both class i and class ii ysos , are reddened background agb stars . unfortunately , since we are looking toward the inner galaxy , background stellar distances approaching 20 to 30 kpc are possible meaning even the faintest , reddest , sources in our sample could be matched by a distant , heavily reddened agb . in figure
[ fig : pos ] we show the spatial distribution of the various yso classes .
it is clear that star formation activity is not uniformly distributed along the periphery of the w 39 bubble with the most striking feature being the lack of ysos at the high and low longitude ends of the bubble around @xmath98 and @xmath99 respectively .
this is consistent with the lack of molecular gas seen at these locations in the @xmath10co images shown in figure [ fig : co ] .
there is also a swath of ysos located within the infrared bubble that roughly follows the @xmath10co distribution seen in the lower left panel of figure [ fig : co ] .
almost all of the @xmath77 class i sources found within the bubble are associated with an irdc that runs roughly from @xmath100 to @xmath101 at @xmath102 .
if associated with w 39 , these sources could represent recent star formation occurring in the near or far wall of the bubble , but it is also possible that the irdc is not directly related to w 39 .
we tried to measure the @xmath20 of the irdc , but were unable to clearly identify any @xmath10co features that were unambiguously associated with the irdc .
higher resolution radio / mm line observations of this irdc in tracers of cold , dense gas would help to clarify its relation to w 39 .
qualitatively it appears from figure [ fig : pos ] that most of the star - formation activity located on the periphery of w 39 is concentrated around the compact regions . in order examine this quantitatively
we divided our search area into 32 roughly @xmath103 areas and calculated the percentage of sources in each area that have a yso classification ( see figure [ fig : red ] ) .
the `` background '' yso percentage , using the median of the eight high and low longitude columns , is 1.8% with an interquartile range of 0.85% .
the ysos within the bubble show up as a slight excess over the background ( @xmath104% ) but the main outliers are the areas around g18.88 - 0.49 and g19.07 - 0.28 with yso percentages of 6.6% and 6.0% respectively .
we investigated the yso population around g18.88 - 0.49 ( 19 ysos ; g18 hereafter ) and g19.07 - 0.28 ( 17 ysos ; g19 hereafter ) in more detail ( the areas are shown in figure [ fig : pos ] ) . for each yso , jhk photometry
was obtained either from 2mass or ukidss , and 24 @xmath13 m photometry was obtained from mipsgal images if possible .
ukidss photometry was used if an unambiguous match to the irac source could be made using a @xmath105 search radius , otherwise the associated 2mass photometry given in the glimpse catalog was used .
aperture photometry on the mipsgal images was obtained for 15 of the ysos using a 6 arcsec radius aperture , a user - defined aperture correction of 1.67 , and a zero - magnitude flux density of 7.17 jy @xcite .
we have adopted a conservative uncertainty of 10% for the [ 24 ] photometry due to image background variations and uncertainty in the aperture correction estimate .
the yso sample photometry is summarized in table [ tbl : ysophot ] .
when available the 24 @xmath13 m photometry provides a very clear discriminant between class i and class ii ysos . at or longward of 24 @xmath13 m a class ii yso s spectral energy distribution ( sed ) is typically flat or falling while a class i yso s sed will typically be rising . in figure
[ fig : ccd13_4 ] we plot the sample of ysos with 24 @xmath13 m photometry on a [ 3.6]-[5.8 ] vs. [ 8.0]-[24 ] cc diagram along with the yso classification criteria from @xcite . as can be seen from comparing columns two and three of table [ tbl : ysoclass ] the irac - only and irac & mips yso classifications are typically in agreement .
two of the ysos are classified in the @xcite scheme as possible `` hot excess '' sources , which are either very heavily embedded class ii ysos or class i ysos undergoing very rapid accretion . to confirm that our ysos are most likely intermediate mass objects (
i.e. , @xmath106 m@xmath29 ) we obtained yso model fits using the sed fitting program of @xcite for those sources with 24 @xmath13 m photometry .
following @xcite we use apertures of 2 for ukidss , 3 for 2mass and glimpse , and 7 for mipsgal and set a minimum flux uncertainty of 10% to prevent any one data point from dominating the fit .
we allowed the distance to vary between 4.3 and 4.7 kpc and @xmath107 to vary between 5 and 50 magnitudes . in table
[ tbl : modfit ] we show the details of the best - fitting yso model as well as the yso stage , where stage i has @xmath108 yr@xmath2 , stage ii has @xmath109 yr@xmath2 and @xmath110 and stage iii has @xmath111 yr@xmath2 and @xmath112 ( @xmath113 is the envelope accretion rate in @xmath114 yr@xmath2 , @xmath115 and @xmath116 are the stellar and disk masses respectively in @xmath114 ) , and the stages typically correspond one - to - one with the observationally defined class i , ii , and iii .
the sed fitter typically returned multiple models that were equally good fits statistically . as outlined in @xcite there are techniques that can be used to obtain a weighted average model for the yso based on this redundancy . in this paper , since we are using the sed fitter primarily to compare with cc diagram classifications and as a rough check on the yso mass , we simply report the details of the single best - fitting model .
we find that our best - fit models are all in the intermediate - mass range @xmath106 m@xmath29 .
a comparison between the sed fitting results and the results from the @xcite cc diagram is interesting .
sources g18 - 1 , g18 - 2 , g19 - 3 , and g19 - 7 all lie well within the class i region of the cc diagram and are also identified as stage i objects by the sed fitter .
all but one of the ysos with [ 3.6]-[5.8 ] between 1.5 and 2.5 , including the two `` hot excess '' sources , are classified as reddened stage ii objects by the sed fitter .
the exception is the reddest source , g19 - 17 which is classified as stage i. for the sources with [ 3.6]-[5.8]@xmath117 agreement between the cc diagram and the sed fitter is mixed ; the sed fitter classifies g19 - 6 as stage ii , but classifies g18 - 9 as stage iii and the very red g19 - 8 yso as stage i.
this study represents the first study of w 39 at high resolution at infrared and radio wavelengths .
the main conclusions of our study are : \1 ) w 39 is a giant region powered by numerous ob stars with a likely total ionizing luminosity @xmath118 , roughly equivalent to a cluster of over 30 o8 v stars .
future work will focus on identifying the ob stars powering w 39 , using techniques described in @xcite , as the first step toward obtaining a non - kinematic distance estimate to the region .
\2 ) three compact regions are located on the periphery of w 39 .
these regions are each likely powered by a single late o - type star ( o7 - o9 v ) .
two of the regions have a distinct dust - filled bubble morphology seen in infrared 8.0 and 24 @xmath13 m images .
\3 ) simple kinematic models suggest that the myr timescales associated with the expansion of w 39 are about an order of magnitude larger than the evolutionary timescales associated with the smaller regions .
this is consistent with the scenario where the formation of the smaller regions is triggered by the expansion of w 39 .
\4 ) recent star formation surrounding w 39 is not restricted just to the massive stars powering the the small regions .
we used infrared data to identify a number of young / heavily - embedded intermediate - mass ysos not directly associated with the regions .
\5 ) star formation activity is not distributed uniformly around the periphery of w 39 , rather it is concentrated in the two regions surrounding the small regions and potentially in an irdc that may be associated with the front or rear wall of a bubble structure .
future studies of the w 39 region will explore the stellar content of the compact regions in more detail , and will examine the potential relationship of the irdc with w 39 .
this work is partially supported through nasa grant adap - nnx10ad55 g .
the authors would like to thank massimo marengo for sharing the _ spitzer _ agb data and chip kobulnicky for his helpful comments and suggestions regarding yso identification .
llcccccccccc g18 - 1 & g018.8008 - 00.4714 & 18.621(0.073 ) & 16.716(0.028 ) & 15.152(0.018 ) & 13.256(0.134 ) & 10.939(0.044 ) & 9.132(0.050 ) & 7.711(0.053 ) & 3.11(0.1 ) + g18 - 2 & g018.8885 - 00.4746 & 17.940(0.04 ) & 16.828(0.035 ) & 14.651(0.013 ) & 12.963(0.083 ) & 10.408(0.199 ) & 9.327(0.068 ) & 9.402(0.115 ) & 2.73(0.1 ) + g18 - 3 & g018.8709 - 00.4210 & 17.628(0.03 ) & 15.308(0.009 ) & 13.312(0.004 ) & 10.994(0.071 ) & 10.034(0.068 ) & 9.229(0.064 ) & 8.482(0.082 ) & 5.57(0.1 ) + g18 - 4 & g018.8240 - 00.4871 & & & 14.615(0.011 ) & 11.293(0.049 ) & 9.951(0.054 ) & 9.175(0.047 ) & 8.480(0.028 ) & 5.98(0.1 ) + g18 - 5 & g018.8632 - 00.4093 & & & 13.420(0.004 ) & 10.911(0.058 ) & 9.817(0.054 ) & 8.828(0.043 ) & 7.951(0.052 ) & 5.11(0.1 ) + g18 - 6 & g018.8344 - 00.4897 & & & & 13.080(0.146 ) & 11.157(0.090 ) & 9.890(0.101 ) & 9.392(0.179 ) & + g18 - 7 & g018.9653 - 00.4008 & & 15.743(0.013 ) & 13.357(0.004 ) & 12.016(0.187 ) & 11.761(0.094 ) & 11.244(0.120 ) & 10.646(0.149 ) & + g18 - 8 & g018.8509 - 00.4104 & 14.680(0.041 ) & 13.260(0.035 ) & 12.293(0.036 ) & 10.645(0.048 ) & 10.104(0.061 ) & 9.594(0.055 ) & 8.950(0.085 ) & + g18 - 9 & g018.8623 - 00.4220 & 18.260(0.06 ) & 14.051(0.079 ) & 11.900(0.034 ) & 10.366(0.053 ) & 10.160(0.044 ) & 9.469(0.062 ) & 8.657(0.118 ) & 5.06(0.1 ) + g18 - 10 & g018.8601 - 00.4122 & & & & 11.964(0.066 ) & 11.417(0.094 ) & 10.627(0.089 ) & 9.892(0.137 ) & + g18 - 11 & g018.8369 - 00.5183 & & 16.546(0.024 ) & 13.760(0.056 ) & 11.882(0.081 ) & 11.451(0.070 ) & 10.882(0.072 ) & 10.440(0.154 ) & + g18 - 12 & g018.8510 - 00.4460 & 17.477(0.026 ) & 13.779(0.002 ) & 11.874(0.001 ) & 10.557(0.036 ) & 10.268(0.065 ) & 9.792(0.070 ) & 9.337(0.165 ) & + g18 - 13 & g018.8793 - 00.4835 & & 17.544(0.067 ) & 15.021(0.018 ) & 13.201(0.121 ) & 13.101(0.154 ) & 10.617(0.226 ) & 9.372(0.305 ) & + g18 - 14 & g018.8879 - 00.4633 & 14.138(0.038 ) & 12.501(0.028 ) & 11.897(0.025 ) & 11.399(0.083 ) & 11.408(0.054 ) & 11.528(0.339 ) & 9.581(0.274 ) & + g18 - 15 & g018.9445 - 00.4404 & 16.538(0.012 ) & 14.426(0.004 ) & 13.203(0.004 ) & 12.248(0.131 ) & 12.187(0.124 ) & 11.402(0.237 ) & 10.203(0.199 ) & + g18 - 16 & g018.8869 - 00.4096 & 18.539(0.07 ) & 15.156(0.008 ) & 13.589(0.005 ) & 12.613(0.072 ) & 12.436(0.099 ) & 12.080(0.197 ) & 10.820(0.159 ) & + g18 - 17 & g018.8263 - 00.4613 & & 17.650(0.07 ) & 14.761(0.01 ) & 12.656(0.117 ) & 12.484(0.169 ) & 10.700(0.139 ) & 9.203(0.223 ) & + g18 - 18 & g018.8016 - 00.4756 & 13.295(0.033 ) & 12.535(0.033 ) & 12.282(0.039 ) & 12.113(0.079 ) & 12.062(0.065 ) & 11.586(0.334 ) & 8.812(0.250 ) & + g18 - 19 & g018.9573 - 00.4718 & & 12.775(0.046 ) & 11.934(0.040 ) & 11.152(0.135 ) & 11.092(0.123 ) & 10.313(0.164 ) & 8.804(0.350 ) & + g19 - 1 & g019.0697 - 00.2032 & & & 14.885(0.015 ) & 12.787(0.135 ) & 12.278(0.155 ) & 10.854(0.300 ) & 9.324(0.141 ) & + g19 - 2 & g019.0365 - 00.2851 & & & 15.916(0.042 ) & 12.937(0.118 ) & 12.195(0.111 ) & 11.342(0.126 ) & 10.236(0.102 ) & + g19 - 3 & g019.0136 - 00.2416 & & & & 13.536(0.088 ) & 11.447(0.084 ) & 10.191(0.048 ) & 9.191(0.038 ) & 6.11(0.1 ) + g19 - 4 & g019.1116 - 00.3521 & 16.085(0.008 ) & 12.961(0.001 ) & 10.335(0.001 ) & 7.614(0.030 ) & 6.704(0.051 ) & 5.856(0.029 ) & 5.093(0.028 ) & 2.05(0.1 ) + g19 - 5 & g019.0774 - 00.3147 & & 15.291(0.009 ) & 11.974(0.001 ) & 8.202(0.052 ) & 7.078(0.059 ) & 6.159(0.035 ) & 5.557(0.025 ) & 3.21(0.1 ) + g19 - 6 & g019.1253 - 00.2782 & 13.528(0.001 ) & 12.207(0.001 ) & 10.870(0.001 ) & 9.380(0.039 ) & 8.908(0.047 ) & 8.396(0.038 ) & 7.057(0.025 ) & 4.09(0.1 ) + g19 - 7 & g018.9996 - 00.2385 & & 13.627(0.036 ) & 11.308(0.046 ) & 8.863(0.090 ) & 7.253(0.102 ) & 6.280(0.061 ) & 5.708(0.173 ) & 1.56(0.1 ) + g19 - 8 & g019.0391 - 00.2217 & 15.994(0.007 ) & 13.901(0.003 ) & 12.665(0.002 ) & 11.909(0.061 ) & 11.638(0.080 ) & 11.460(0.219 ) & 10.402(0.088 ) & 4.25(0.1 ) + g19 - 9 & g019.0530 - 00.2148 & 15.734(0.006 ) & 13.040(0.001 ) & 10.950(0.001 ) & 9.115(0.039 ) & 8.367(0.095 ) & 7.731(0.043 ) & 6.798(0.036 ) & 2.94(0.1 ) + g19 - 10 & g019.0103 - 00.2320 & 17.523(0.027 ) & 14.303(0.004 ) & 12.552(0.002 ) & 11.372(0.056 ) & 11.070(0.073 ) & 10.586(0.114 ) & 10.047(0.150 ) & + g19 - 11 & g019.1071 - 00.3067 & & 14.602(0.005 ) & 12.529(0.002 ) & 11.250(0.087 ) & 10.965(0.098 ) & 10.803(0.094 ) & 10.327(0.075 ) & + g19 - 12 & g019.0941 - 00.2894 & 16.036(0.008 ) & 13.720(0.002 ) & 12.085(0.002 ) & 10.382(0.034 ) & 9.651(0.045 ) & 9.012(0.045 ) & 8.423(0.076 ) & + g19 - 13 & g019.0228 - 00.2800 & & 14.449(0.004 ) & 12.779(0.003 ) & 11.545(0.064 ) & 11.302(0.090 ) & 11.188(0.129 ) & 10.597(0.249 ) & + g19 - 14 & g019.1150 - 00.3511 & 14.682(0.003 ) & 11.561(0.001 ) & 9.568(0.001 ) & 7.371(0.044 ) & 6.846(0.041 ) & 6.030(0.031 ) & 5.449(0.026 ) & 2.5(0.3 ) + g19 - 15 & g019.0488 - 00.2932 & 13.602(0.001 ) & 12.428(0.001 ) & 11.863(0.001 ) & 11.381(0.068 ) & 11.312(0.089 ) & 10.571(0.125 ) & 9.346(0.140 ) & + g19 - 16 & g019.0806 - 00.2861 & 17.782(0.04 ) & 15.623(0.01 ) & 14.229(0.009 ) & 11.043(0.121 ) & 11.270(0.184 ) & 8.043(0.113 ) & 6.651(0.286 ) & + g19 - 17 & g019.0612 - 00.2453 & 15.077(0.004 ) & 13.922(0.003 ) & 13.219(0.004 ) & 12.274(0.083 ) & 11.964(0.123 ) & 10.714(0.139 ) & 9.128(0.195 ) & 3.78(0.1 ) + lccc g18 - 1 & i & i & i + g18 - 2 & i & i & i + g18 - 3 & i & i & ii + g18 - 4 & i & i / he & ii + g18 - 5 & i & i & ii + g18 - 6 & i & & + g18 - 7 & ii & & + g18 - 8 & ii & & + g18 - 9 & ii & ii & iii + g18 - 10 & ii & & + g18 - 11 & ii & & + g18 - 12 & ii & & + g18 - 13 & i / ii & & + g18 - 14 & i / ii & & + g18 - 15 & i / ii & & + g18 - 16 & i / ii & & + g18 - 17 & i / ii & & + g18 - 18 & i / ii & & + g18 - 19 & i / ii & & + g19 - 1 & i & & + g19 - 2 & i & & + g19 - 3 & i & i & i + g19 - 4 & i & i & ii + g19 - 5 & i & he & ii + g19 - 6 & i & ii & ii + g19 - 7 & i & i & i
+ g19 - 8 & ii & ii & i + g19 - 9 & ii & i / ii & ii + g19 - 10 & ii & & + g19 - 11 & ii & & + g19 - 12 & ii & & + g19 - 13 & ii & & + g19 - 14 & ii & i / ii & ii + g19 - 15 & i / ii & & + g19 - 16 & i / ii & & + g19 - 17 & i / ii & i / ii & i + lccccccc g18 - 1 & 3008766 & 18.19 & 14.52 & 3.16 & @xmath119 & @xmath120 & i + g18 - 2 & 3009923 & 49.46 & 12.28 & 7.72 & @xmath121 & @xmath122 & i + g18 - 3 & 3002489 & 81.37 & 9.75 & 4.15 & 0 & @xmath123 & ii + g18 - 4 & 3012453 & 81.37 & 45.36 & 5.77 & 0 & @xmath124 & ii + g18 - 5 & 3014480 & 81.37 & 8.89 & 4.39 & 0 & @xmath125 & ii + g18 - 9 & 3003037 & 49.46 & 22.31 & 6.90 & 0 & @xmath126 & iii + g19 - 3 & 3012136 & 18.19 & 13.20 & 3.41 & @xmath127 & @xmath128 & i + g19 - 4 & 3002683 & 81.37 & 12.81 & 4.37 & 0 & @xmath129 & ii + g19 - 5 & 3014289 & 75.52 & 28.43 & 7.65 & 0 & @xmath130 & ii + g19 - 6 & 3018962 & 18.19 & 5.11 & 4.55 & 0 & @xmath131 & ii + g19 - 7 & 3010932 & 18.19 & 14.35 & 7.51 & @xmath132 & @xmath133 & i + g19 - 8 & 3008529 & 41.41 & 6.19 & 2.66 & @xmath134 & @xmath135 & i + g19 - 9 & 3015777 & 41.41 & 16.34 & 4.93 & @xmath136 & @xmath137 & ii + g19 - 14 & 3017501 & 75.52 & 16.34 & 8.84 & 0 & @xmath138 & ii + g19 - 17 & 3013269 & 18.19 & 5.34 & 2.14 & @xmath139 & @xmath140 & i + | we present the first detailed study of the large , @xmath0 pc diameter , inner - galaxy region w 39 .
radio recombination line observations combined with absorption spectra and galactic rotation models show that the region lies at @xmath1 km s@xmath2 corresponding to a near kinematic distance of 4.5@xmath3 kpc .
analysis of radio continuum emission shows that the region is being powered by a cluster of ob stars with a combined hydrogen - ionizing luminosity of @xmath4 , and that there are three compact regions located on the periphery of w 39 , each with @xmath5 ( single o7 - o9 v star equivalent ) . in the infrared
, w 39 has a hierarchical bubble morphology , and is a likely site of sequential star formation involving massive stars .
kinematic models of the expansion of w 39 yield timescales of order myr consistent with a scenario where the formation of the smaller regions has been triggered by the expansion of w 39 . using _ spitzer _
glimpse and mipsgal data we show that star - formation activity is not distributed uniformly around the periphery of w 39 but is concentrated in two areas that include the compact regions as well as a number of intermediate - mass class i and class ii ysos . |
on general grounds one expects that the two fundamental theories of contemporary physics , the theory of general relativity and the quantum theory in the form of the standard model of particle physics , can be unified at the planck energy scale .
this unification requires to quantize gravity , which leads to very fundamental difficulties .
one is related with the energy of the fundamental vacumm state .
another one is the status of lorentz invariance : the fuzzy nature of space time in quantum gravity may lead to violations of this fundamental symmetry . for the last two decades theoretical studies and experimental searches of lorentz invariance violation ( liv )
have received a lot of attention ( see e.g. the reviews by * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
possible consequences of liv are energy and helicity dependent photon propagation velocities .
the energy dependence can be constrained by recording the arrival times of photons of different energies emitted by distant objects at approximately the same time @xcite , e.g. during a gamma - ray burst @xcite or a flare of an active galactic nucleus @xcite . on the other hand ,
the helicity dependence can be constrained by measuring how the polarization direction of an x - ray beam of cosmological origin changes as function of energy @xcite .
recently , an upper limit on the helicity dependence of photon propagation has been set using integral / spi observations of the polarization of the crab nebulae @xcite . in this paper
, we derive much stronger constraints from a polarization measurement of a gamma - ray bursts ( grb ) , a source at cosmological distance . to date only a few polarization measurements are available for grbs .
@xcite reported a high degree of polarization , @xmath0 , for grb 021206 .
however , successive reanalysis of the same data set could not confirm this claim , reporting a degree of polarization compatible with zero @xcite .
@xcite reported a strong polarization signal in grb 930131 ( @xmath1 ) and grb 960924 ( @xmath2 ) , but this result could not be statistically constrained .
grb 041219a was detected by the integral burst alert system ( ibas ; @xcite ) , and is the longest and brightest grb localized by integral @xcite to date @xcite .
@xcite , using the data of the integral spectrometer ( spi ; * ? ? ?
* ) , reported a high degree of polarization of the prompt emission ( @xmath3 ) for the brightest part of this grb .
also , @xcite have performed a similar measurement , using this time the integral / ibis telescope , reporting variable polarization properties of the burst all along its duration . in this paper
, we reuse these data in order to measure the polarization in two energy bands and check for a shift in the polarization angle as a possible effect of lorentz invariance violation . in section [ sec : level1 ] , we will present the integral / ibis observations of polarization during the burst , and the recent measure of the burst distance we obtained thanks to far infrared observations of its host galaxy . in section [ sec : level2 ] , we will derive constraints on liv from these observations and we will conclude in section [ sec : level3 ] .
thanks to its two position sensitive detector layers isgri @xcite ( made of cdte crystals and sensitive in the 151000 kev energy band ) , and picsit @xcite ( made of csi bars and sensitive in the 200 kev10 mev energy band ) , ibis can be used as a compton polarimeter @xcite .
the concept behind a compton polarimeter is the polarization dependency of the differential cross section for compton scattering @xmath4 where @xmath5 is the classical electron radius , @xmath6 the energy of the incident photon , @xmath7 the energy of the scattered photon , @xmath8 the scattering angle , and @xmath9 the azimuthal angle relative to the polarization direction .
linearly polarized photons scatter preferentially perpendicularly to the incident polarization vector .
hence by examining the scatter angle distribution of the detected photons @xmath10 , \label{eq : azimuth}\end{aligned}\ ] ] one can derive the polarization angle @xmath11 and the polarization fraction @xmath12 , where @xmath13 is the amplitude expected for a 100% polarized source , derived by montecarlo simulations of the instrument . to measure the polarization
, we followed the same procedure described in @xcite that allowed to successfully detect a polarized signal from the crab nebula . one important difference , anyway , is that the crab measurement was integrated over several observation periods and a long time ( @xmath14ms ) , while this measurement here integrates over only a few seconds .
so we have checked if instrumental azimuthal variations , potentially washed out by the long crab observation , could be important in the grb case .
to do so , we have simulated ten grb - like observations using data from the integral payload ground calibrations campaign .
this campaign , made with the full flight model of the telescope , was done with standard radioactive non polarized sources .
we use in this work the calibrations made with a @xmath15sn source with a peak energy of 392 kev , the closest to the energy bands we are considering here . to simulate grbs
, we divide the data set in ten subsets containing each a similar number of compton events as in the observations we show in figure [ fig : evol ] . in all these subsets
, we found systematics at a level up to @xmath16 maximum , with a polarization angle around 40@xmath17 .
we made another tests considering spurious events , as described in @xcite , and again we found a systematic azimuthal variation of less than @xmath18 , with a polarization angle of @xmath19 .
both value are far from the ones we derived below for the source and give confidence that our results are not due to instrumental azimuthal variations , even if such variations effectively exist . to derive the gamma - ray burst flux as a function of @xmath9 , the compton photons were divided in 6 bins of 30@xmath17 as a function of the azimuthal scattering angle . to improve the signal - to - noise ratio in each bin , we took advantage of the @xmath20-symmetry of the differential cross section ( see eq .
[ eq : azimuth ] ) , i.e. the first bin contains the photons with @xmath21 and @xmath22 , etc .
the derived detector images were then deconvolved to obtain sky images , where the flux of the source in each bin is measured by fitting the instrumental psf to the source peak .
we finally fitted the polarigrams to eq .
[ eq : azimuth ] using a least squares technique ( see fig . [
fig : pola ] ) in order to derive @xmath23 and @xmath24 , and the errors on the parameters are dominated by the statistics of the data points .
we computed the scatter angle distribution into two energy band in order to detect a possible polarimetric angle shift with energy , reminiscent of a possible liv effect .
the two bands were chosen to be [ 200 - 250 kev ] and [ 250 - 325 kev ] where the source has merely the same signal to noise ratio . in figure
[ fig : evol ] , we show the measured evolution of the polarimetric angle shift between these two energy bands , along the burst duration .
these shifts are all consistent with zero with a mean value of @xmath25 . also , as an exemple , we show in figure [ fig : pola ] and figure [ fig : contour1 ] , the portion of the grb light curve where the polarimetric signal was strong , that is starting at 01:47:02 u.t . until 01:47:12 u.t .
( p9 interval in @xcite ) .
a modulated signal is seen in the two energy bands , corresponding to @xmath26 for the first band , and @xmath27 for the second ones ( see fig.[fig : contour1 ] for error contours ) . to evaluate the goodness of our fits
, we computed the chance probability ( see eq . 2 in @xcite ) that our polarigrams are due to an unpolarized signal , and reported these values in fig .
[ fig : pola ] .
the corresponding polarimetric angles were @xmath28 deg . and @xmath29 deg
, that is consistent at the @xmath30 level ( see figure [ fig : contour1 ] )
. propagation of errors gives an upper limit of @xmath31 at the @xmath32 level , for a possible phase shift .
evolution during the burst duration of the polarimetric angle shift measured between the [ 200 - 250 kev ] and in the [ 250 - 325 kev ] energy range .
the mean value , @xmath25 , is consistent with zero.,width=340 ] scatter angle distribution of grb041219a in the 200 - 250 kev and in the 250 - 325 kev energy band during the p9 time interval ( see text ) .
these distributions give the source count rate by azimuthal angle of the compton scattering , and are consistent with a highly polarized signal . the chance probability of a non - polarized signal
is reported in each panel .
the polarization angles derived from these distributions are consistent within 68@xmath33 ( see text).,width=340 ] @xcite performed deep infra - red imaging of the grb region using the wircam instrument at the 3.6 m canadian french hawaiian telescope ( cfht ) at mauna kea .
thanks to multi - band ( @xmath34 ) imaging they were able to identify the host galaxy of grb 041219a , and to give an lower limit to its photomeric redshift of @xmath35 , at the @xmath32 confidence level .
this implies a luminosity distance of 85 mpc , assuming standard cosmological parameters ( @xmath36 , @xmath37 , and @xmath38 km / s / mpc ) .
on general grounds , lorentz violating operators of dimension @xmath39 modify the standard dispersion relations @xmath40 by terms of the order of @xmath41 where @xmath42 is the reduced planck scale ( @xmath43 gev ) , used as a reference scale since liv is expected to arise in the quantum regime of gravity . in order to account for the severe limits on liv , one therefore usually only considers operators of dimension greater or equal to 5 which provide corrections which are tamed by at least one inverse plank scale .
if we restrict our attention to pure electrodynamics , there is a single term of dimension 5 which gives corrections of order @xmath44 and is compatible with gauge invariance and rotational symmetry @xcite : @xmath45 where @xmath46 is a 4-vector that characterizes the preferred frame and @xmath47 .
the uniqueness of this term makes the analysis somewhat model - independent ( see however below ) .
the light dispersion relation is given by ( @xmath48 and @xmath49 ) : @xmath50 where the sign of the cubic term is determined by the chirality ( or circular polarization ) of the photons , which leads to a rotation of the polarization during the propagation of linearly polarized photons .
this effect is known as vacuum birefringence .
since we have the approximative relation : @xmath51 the direction of polarization rotates during propagation along a distance @xmath52 by an angle : @xmath53 for grb041219a , if we set @xmath54 , derived from the measures we made along the burst duration , and the lower limit luminosity distance reported above of @xmath55 mpc = @xmath56 cm , corresponding to z=0.02 , we get an upper limit on the vacuum birefringence effect : @xmath57 although the dimension 5 operator ( [ dim5 ] ) is unique , it is physicaly relevant only if there does not appear operators of lower dimensions . in the general case however , radiative corrections induced by dimension 5 operators lead to dimension 3 operators through quadratic divergences which contribute an extra factor @xmath58 , where @xmath59 is an ultraviolet cut - off of order @xmath42 . barring extreme fine tuning , one needs a symmetry argument to cancel such terms .
supersymmetry appears to be the only symmetry which cancels such contributions @xcite .
it is true that supersymmetry is broken in nature but , in the case where it is softly broken , quadratic divergences contribute a factor @xmath60 , where @xmath61 , the scale of supersymmetry breaking , is of the order of a tev@xmath62 , thus providing an extra factor @xmath63 compared to untamed dimension 3 operators .
unfortunately , the dimension 5 operator of ( [ dim5 ] ) is not compatible with supersymmetry @xcite .
we therefore have to resort in this case to dimension 6 operators .
we thus assume that the light dispersion relation is given by : @xmath64 we can derive then the approximate relation : @xmath65 which implies that the direction of polarization rotates during propagation of : @xmath66 again , for grb041219a , if we set @xmath54 and the luminosity distance @xmath55 mpc = @xmath56 cm , corresponding to z=0.02 , we get an upper limit on the vacuum birefringence effect : @xmath67 this is still too large an upper bound to be really constraining since one expects couplings at most of order one .
using a recent determination of the distance of grb041219a @xcite for which a high degree of polarization is observed @xcite , we were able to increase by 4 orders of magnitude the existing constraint on lorentz invariance violations , arising from birefringence
. turned into a constraint on the coupling @xmath68 of dimension 5 lorentz violating interactions , that is of corrections of order @xmath69 to the dispersion relations , this gives the very stringent constraint @xmath70 .
most presumably , this means that such operators are vanishing , which might point towards a symmetry such as supersymmetry in action . in that case , the pressure is on the next corrections of order @xmath71 corresponding to operators of dimension 6 .
we showed that , although astrophysical constraints are not yet really constraining , they are getting closer to the relevant regime ( @xmath68 of order @xmath72 or smaller ) .
our result can be compared to the limits derived using the possible energy dependence of the group velocity of photons in distant grbs derived with fermi @xcite .
however , written in the same way as we did in this work , this limit was @xmath73 that is much less constraining .
jacobson , t. , liberati , s. , mattingly , d. annals of physics , 321 , 150 ( 2006 ) liberati , s. , & maccione , l. , annual review of nuclear and particle science , 59 , 245 ( 2009 ) mattingly , d. , living reviews in relativity , 8 , 5 ( 2005 ) amelino - camelia , g. , ellis , j. r. , mavromatos , n. e. , nanopoulos , d. v. , and sarkar , s. , 393 , 763 ( 1998 ) abdo , a. a. , ackermann , m. , ajello , m. , et al .
, , 462 , 331 ( 2009 ) aharonian , f. , akhperjanian , a. g. , barres de almeida , u. , et al . , phys .
lett . , 101 ( 2008 )
gambini , r. , pullin , j. , phys . rev .
d 59 , 124021 ( 1999 ) maccione l. , liberati s. , celotti a. , kirk j.g .
, and ubertini p. , physical review d , 78 , 103003 , ( 2008 ) coburn , w. , & boggs , s.e , , 423 , 415 ( 2003 ) rutledge , r.e .
, & fox , d.e . ,
mnras , 350 , 1288 ( 2004 ) wigger , c. , et al . , apj , 613 , 1088 ( 2004 ) willis , d.r . ,
barlow , e.j . , bird , a.j . , et al . ,
a&a , 439 , 245 ( 2005 ) mereghetti , s. , et al . , a&a , 411 , l291 ( 2003 ) winkler , c. , courvoisier , t.j .- l . , di cocco g. , et al . ,
a&a , 411 , l1 ( 2003 ) vianello , g. , gtz , d. , & mereghetti , s. , a&a , 495 , 1005 , ( 2009 ) mcglynn , s. , clark , d.j . ,
dean , a.j .
, et al . , a&a , 466 , 895 ( 2007 ) vedrenne , g. , roques , j .-
, schnfelder , v. , et al . ,
a&a , 411 , l63 ( 2003 ) gtz , d .. , et al . , apj , 695 , l208 ( 2009 ) lebrun , f. , leray , j.p . , lavocat , p. , et al . ,
a&a , 411 , l141 ( 2003 ) labanti , c. , di cocco , g. , ferro , g. , et al . ,
a&a , 411 , l149 ( 2003 ) lei , f. , et al . ,
space science reviews , 82 , 309 ( 1997 ) forot , m. , et al . , apj , 688 , l29 ( 2008 ) gtz , d. , et al . , mnras , in press , ( 2011 ) , arxiv : astro - ph/1103.3663 myers , r.c . , and pospelov , m. , phys .
lett . , 90 , 211601 ( 2003 )
nibbelink , s.g . , and pospelov , m. , phys . rev .
94 , 081601 ( 2005 ) | one of the experimental tests of lorentz invariance violation is to measure the helicity dependence of the propagation velocity of photons originating in distant cosmological obejcts . using a recent determination of the distance of the gamma - ray burst grb 041219a , for which a high degree of polarization is observed in the prompt emission , we are able to improve by 4 orders of magnitude the existing constraint on lorentz invariance violation , arising from the phenomenon of vacuum birefringence . |
in recent years , topologically nontrivial properties of band theory in electronic systems have become the driving force in condensed matter physics @xcite . although electronic band theory has been studied over the past decades , the topological properties associated with it have only been appreciated recently .
the haldane model @xcite was the first example of quantum anomalous hall effect ( zero magnetic field quantum hall effect ) in electronic systems that originates completely from the topology of the energy bands . for many decades ,
this model was deemed as a toy model due to lack of experimental evidence .
quite recently , the experimental realization of haldane model @xcite has been reported in ultracold fermionic atoms in a periodically modulated optical honeycomb lattice @xcite .
this observation has motivated numerous theoretical studies of the effects of interactions in the so - called haldane - hubbard model @xcite .
on the other hand , topologically nontrivial energy bands occur frequently in ordered quantum spin magnets , however , their topological properties are rarely studied in quantum magnetism .
the study of topological spin excitations ( magnons ) recently began with the work of katsura - nagaosa - lee @xcite on the kagome and pyrochlore heisenberg ferromagnets with a nearest - neighbour ( nn ) dzyaloshinsky - moriya ( dm ) interaction @xcite .
in addition to breaking the inversion symmetry of the lattice , the dm interaction plays the role of spin - orbit coupling as in electronic systems . as a result ,
the bosons acquire a phase ( magnetic flux ) while hopping on the lattice and the system exhibits similar topological properties as in electronic systems .
however , in bosonic systems there is no fermi energy or fermi level , and the idea of completely filled band does not apply . in this respect , the chern number characterizing the nontrivial topology of the energy bands is , in fact , independent of the statistical nature of the particles .
it merely predicts counter - propagating edge state modes in the vicinity of the bulk gap as a result of the bulk - edge correspondence .
this leads to magnon edge state modes in ordered quantum magnets .
interestingly , the magnon edge state modes carry a transverse heat ( spin ) current upon the application of a longitudinal temperature gradient .
as magnons are uncharged particles , there is no lorentz force , the dm interaction plays the role of a magnetic field by altering the propagation of the magnon in the system , thus leads to thermal hall effect dubbed _ magnon hall effect_. it was discovered experimentally by onose _
et al_. , @xcite in the ferromagnetic insulator lu@xmath1v@xmath1o@xmath2 with a 3d pyrochlore lattice structure .
recently , thermal hall effect has been observed experimentally on the 2d kagome magnet cu(1 - 3 , bdc ) @xcite . besides , _ phonon hall effect _ had been observed previously @xcite in a completely different scenario .
mathematically , _
magnon hall effect _ is manifested as a result of the nontrivial topology of magnon bulk band encoded in the berry curvature @xmath3 , which acts a magnetic field , where @xmath4 is a vector potential .
this result was derived by matsumoto and murakami @xcite and relates the transverse thermal conductivity @xmath5 directly to the berry curvature of the magnon bulk bands reminiscent of hall conductivity in electronic systems @xcite .
it simply shows that the dm interaction can appear in any form provided the system exhibits a nontrivial topology in the magnon bulk bands .
it was also shown that due to the boson population of the bands at high and low temperatures , thermal conductivity changes sign as function of temperature or magnetic field on the kagome and pyrochlore lattices @xcite .
in addition , spin nernst and torque effects have been recently proposed theoretically in these systems @xcite .
the heisenberg ferromagnet on the honeycomb lattice is known to exhibit dirac points in the magnon energy bands @xcite . in this regard ,
we have proposed and studied a topological magnon insulator on the honeycomb lattice @xcite .
we observed many distinctive features on the honeycomb lattice .
firstly , the dm interaction appears as a next - nearest - neighbour ( nnn ) coupling transversing through the triangular plaquettes of the nnn sites on the honeycomb lattice as opposed to the kagome and pyrocholre lattices .
it generates a spin chirality on the triangular plaquettes of the nnn sites given by @xmath6 on sublattice @xmath7 and @xmath8 on sublattice @xmath9 , where @xmath10 is the spin at site @xmath11 . in the spin wave bosonic representation ,
the resulting hamiltonian @xcite is analogous to the haldane model in electronic systems @xcite , which is known to possess a nontrivial insulating phase .
secondly , thermal hall conductivity @xmath12 does not change sign as a function of temperature or magnetic field and the low temperature dependence follows a @xmath13 law @xcite .
recently , spin nernst effect has been studied in this system @xcite . in this paper
, we show that the topological properties of magnons are not different from those of hardcore boson systems .
this is due to a one - to - one correspondence between bosonic systems and quantum magnetic systems .
we study a haldane - like hardcore bosons on the honeycomb lattice , which possesses a superfluid phase and a mott phase .
this model has a quantum monte carlo sign problem which hinders an explicit numerical simulation using this method .
however , it can be studied by other available numerical schemes .
we utilize the magnetic spin analogue of this model by mapping it to a quantum xy model with a dm interaction .
this enables us to study the magnetic excitations ( magnons ) of the quantum system .
we observe nontrivial topological properties of this model by studying the magnon energy bands , edge states , berry curvature , thermal and spin nernst conductivities of the corresponding quantum spin model of the hardcore bosons .
the correspondence between hardcore bosons and quantum spin systems suggests that the associated hardcore boson model exhibits the same properties in the boson language , because its excitations have to be magnons as well .
these results are important as they can be studied in cold atom experiments .
in fact , the phase transitions between superfluid phase and mott phase have been reported experimentally in ultracold bosonic atom experiments in optical lattices @xcite .
hence , it is possible to realize magnon topological properties by using cold atoms trapped in honeycomb optical lattices as has been reported in fermionic systems @xcite .
as mentioned above , the shortcoming of most topological bosonic models is that quantum monte carlo ( qmc ) simulation is not applicable due to a deliberating sign problem . in this paper , we study one of these models in which qmc suffers a sign problem . for hardcore bosons on the honeycomb lattice without any topological effects ,
qmc has been utilized effectively in the study of the ground state phase diagrams @xcite . in this system
there is a u(1 ) invariance , hence the bosonic excitations ( magnons ) always exhibit a gapless ( goldstone ) mode at the @xmath14 points ( @xmath15 ) and the two magnon bands exhibit dirac points at the corners of the brillouin zone @xmath16 .
such 2d systems with dirac nodes have no topological effects in the excitation spectrum .
the topological properties of these systems have received no attention mostly for the reason mentioned above .
fortunately , the hardcore boson hamiltonian has a one - to - one correspondence to a spin-@xmath0 quantum xxz or xy model @xcite .
hence , the topological effects in this model can be studied in terms of the magnetic excitations of the quantum spin model . as in many electronic systems ,
topological effects arise when a gap opens at @xmath17 .
there are many different ways to open a gap at @xmath17 .
the simplest way is to add a staggered on - site potential , as has been shown recently @xcite . in the magnetic spin language , this corresponds to a staggered magnetic field .
an alternative way is to add a haldane - like imaginary hopping interaction along the nnn sites , this would correspond to a dm interaction in the spin language @xcite .
the resulting hamiltonian of a gapped hardcore boson model on the honeycomb lattice can be written as @xmath18 where @xmath19 and @xmath20 are the bosonic creation and annihilation operators respectively ; @xmath21 ; @xmath22 depending on the hopping directions of the bosons as shown in fig .
[ unit ] , and @xmath23 is the phase of the hopping amplitude .
the bosonic operators obey the algebra @xmath24=0 $ ] for @xmath25 and @xmath26 . )
denote the direction of the imaginary hopping amplitude , which corresponds to the chiral dm interaction in spin variables .
it generates a chirality on the triangular plaquettes .
the coordinates are @xmath27 ; @xmath28 , and @xmath29 .
, width=192 ] here , @xmath30 denotes nn hopping , @xmath31 is the chemical potential , and @xmath32 is the nnn hopping with imaginary phase . the model in [ hardcore ] has been recently studied in a different context @xcite . in that study , the authors did not investigate the mapping to quantum spin systems , but focused mainly on the haldane - hubbard hamiltonian with a hubbard on - site interacting potential .
however , topological effects should also arise in noninteracting hardcore bosons [ hardcore ] .
in contrast to haldane model in fermionic systems @xcite , the hardcore bosons [ hardcore ] can be mapped to a spin-@xmath0 quantum spin system @xcite . in the spin language , eq .
[ hardcore ] maps to @xmath33 where @xmath34 ; @xmath35 , with @xmath36 , and @xmath37 .
the new parameters are related to the hardcore boson variables by @xmath38 , @xmath39 , @xmath40 .
figure [ unit ] shows the direction of the fictitious magnetic flux generated by the dm interaction , which corresponds to the imaginary hopping amplitude with @xmath41 for @xmath42 on sublattice @xmath7 plus cyclic permutations , and @xmath43 on sublattice @xmath9 for @xmath42 etc . as mentioned above , this alternating dm interaction has the effect on magnons as the chirality operators @xmath6 on sublattice @xmath7 and @xmath8 on sublattice @xmath9 , where @xmath10 is the spin at site @xmath11 .
it is important to note that the spin hamiltonian [ hh1 ] is unfrustrated in all the parameter regimes .
it simply describes a ferromagnetic insulator .
hence , eqs .
[ hardcore ] and [ hh1 ] describe an ordered system and the magnon excitations of the spin system correspond to excitations of the hard - core bosons .
it is well - known that topologically nontrivial properties of magnon excitations such as magnon edge states , thermal hall effect , and spin nernst effect , can be understood entirely by semiclassical approximations @xcite .
spin wave theory provides the simplest way to study the topological effects in ordered quantum magnetic systems . in this section
, we apply this formalism to the present model .
the mean field phase diagram provides the possible phases in the system . in the mean field approximation ,
the spins can be approximated as classical vectors parameterized by a unit vector : @xmath44 .
there are two sublattices on the honeycomb lattice as shown in fig .
the classical energy with @xmath45 ( coplanar spins ) is given by @xmath46 where @xmath47 is the number of unit cells , @xmath48 is the critical field , and @xmath49 is the coordination number of the lattice . at the mean - field level , the dm interaction does not contribute to the classical energy .
the filling factor is given by @xmath50 .
in this model , there are only two possible phases a canted ferromagnet or superfluid at small magnetic field and a fully polarized ferromagnet or mott phase , with the spins eqnarrayed along the @xmath51-axis at large magnetic field . in both ferromagnets @xmath52 . the filling factor for the mott phases are @xmath53 ( empty ) and @xmath54 ( full ) for @xmath55 respectively .
the phase boundary is obtained by minimizing eq .
[ cla ] yielding @xmath56 . in ordered quantum magnets , it is customary to study magnetic excitations by linear spin - wave theory . in the spin wave expansion ,
we first rotate the coordinate axes so that the @xmath51-axis coincides with the local direction of the classical polarization .
@xmath57 then , we express the operators in terms of the linearized holstein primakoff ( hp ) transformation , @xmath58 . the bosonic spin wave hamiltonian becomes @xmath59\nonumber\\&- v_d\sum_{\la \la ij\ra\ra}(e^{i\nu_{ij}\phi } c^\dagger_i c_{j}+h.c.)+v_0\sum_i c_{i}^\dagger c_{i } , \label{hp3}\end{aligned}\ ] ] where @xmath60 , @xmath61 , @xmath62 .
at zero magnetic field , the spins are aligned along the @xmath63-axis and @xmath64 .
hence , the @xmath51-component of the dm interaction does not contribute in the linear spin wave expansion , as it is not along the quantization axis . in this case
, one should take @xmath65 . as one can discern from eq .
[ hp3 ] , the xy model does not have a simple analogue to haldane model @xcite as the heisenberg ferromagnets @xcite .
this is due to additional off - diagonal term with coefficient @xmath66 .
fourier transforming eq .
[ hp3 ] we obtain h= _ ^_()_+ , where @xmath67 , and @xmath68 . the momentum space hamiltonian is given by @xmath69\nonumber\\&- \tilde v_{2}\sigma_x(\tau_+\gamma_\bo + h.c . ) , \label{hhhm}\end{aligned}\ ] ] where @xmath70 and @xmath71 .
we have introduced two pauli matrices @xmath72 and @xmath73 , where @xmath74 , while @xmath75 and @xmath76 are identity @xmath77 matrices in each space .
the structure factors are _ * k*=2 _ _ ; _ * k*= _ e^i*k*_;where @xmath78 and @xmath79 are the nn and nnn vectors shown in fig .
[ unit ] respectively .
the spin wave hamiltonian in eq .
[ hhhm ] is hermitian but it is not diagonal .
it is diagonalized by a matrix @xmath80 via the transformation @xmath81 , which satisfies the relation ^ ( ) = ( ) ; ^= , with @xmath82 .
the matrix @xmath83 contains the bogoliubov operators @xmath84 and @xmath85 $ ] are the eigenvalues .
this transformation leads to a non - hermitian bogoliubov hamiltonian @xmath86 , which is given by @xmath87 where @xmath88 the eigenvalues of the non - hermitian bogoliubov hamiltonian give the magnon energy bands . for @xmath89 , the matrices @xmath90 and @xmath91 are constants of motion and can be diagonalized separately . in this case
the hamiltonian corresponds to the conventional xy model ( or hardcore bosons ) and the energy bands are gapless at @xmath17 . for @xmath92
this symmetry is broken and dynamics are introduced into the system .
the positive eigenvalues are given by @xmath93 ^ 2}\rb^2-\lb \tilde v_{2}|\gamma_\bo|\rb^2 } , \label{xym}\end{aligned}\ ] ] where @xmath94 labels the top and the bottom bands of the positive energies .
magnon berry curvatures of the hardcore bosons for the top and bottom bands of fig .
[ band ] . ]
although the dm interaction does not contribute to the classical energy , the magnon excitations depend on it , thus a chiral superfluid is formed in the magnon bands .
the two magnon bulk bands in chiral superfluid phase are shown in fig .
[ band ] . at @xmath17 , @xmath95 and @xmath96 , hence a gap of magnitude @xmath97
is generated in the magnon bulk bands .
as mentioned above , the topological properties of this system are manifested once a gap opens at @xmath17 in the magnon bulk bands . from the bulk - edge correspondence , we expect magnon edge states in the vicinity of the bulk gap .
indeed , we observe such edge states as shown in fig .
due to the mapping from spin variables to bosonic operators and verse versa , the hardcore boson model [ hardcore ] also exhibits bosonic edge states .
an important consequence of gapped systems is the non - vanishing of the berry curvature associated with the topology of the energy bands .
the berry curvature associated with the gapped hard - core bosons is given by @xmath98 , \label{bb}\ ] ] where @xmath99 are the positive eigenvectors .
it is more feasible to calculate the berry curvature numerically .
hence , eq . [ bb ] can be written as @xmath100}{\lb\epsilon_{\bo\lambda}-\epsilon_{\bo\lambda^\prime}\rb^2 } , \label{chern2}\end{aligned}\ ] ] where @xmath101 defines the velocity operators .
figure [ berry ] shows the berry curvatures for the top and the bottom bands .
it is obvious that the dominant contributions come from the states near @xmath102 and @xmath103 . in the low - energy limit
, we expand the hamiltonian near @xmath17 .
we obtain @xmath104\nonumber\\ & -i\sg_y \bar v_{2}(\xi \tau_x k_x+ \tau_y k_y)+\xi m{\bold i}_\sg\tau_z , \label{hald}\end{aligned}\ ] ] where @xmath105 for states at @xmath17 and @xmath106 . at the fully polarized mott states , @xmath107 or @xmath108 , we have @xmath109 and @xmath110
. then , the model decouples into two bogoliubov hamiltonians . in this case , eq .
[ hald ] is similar to the haldane model @xcite .
this is a direct consequence of the hard - core boson hamiltonian [ hardcore ] . in the preceding sections
, we have explicitly demonstrated the nontrivial topological properties of the gapped honeycomb hardcore bosons .
as mentioned in the introduction , the magnon edge states carry a heat or spin current upon the application of a temperature gradient .
this is , in fact , due to the nontrivial topological properties of the berry curvature .
these edge states lead to two important phenomena thermal hall effect@xcite and spin nernst effect .
@xcite they are characterized by two conductivities given by@xcite @xmath111 where @xmath112 is the spin nernst conductivity and @xmath5 is the thermal hall conductivity ; @xmath113=[e^{{\epsilon_\lambda(\bold k)}/k_bt}-1]^{-1}$ ] is the bose function and the @xmath114 functions are @xmath115 ; @xmath116 and @xmath117 is a polylogarithm .
the plots of @xmath112 and @xmath5 as functions of temperature for varying magnetic field are shown in fig .
the spin nernst conductivity vanishes at zero temperature as there is no thermal excitations . in stark contrast to the heisenberg ferromagnet @xcite
, it increases with increasing magnetic field as the system progresses from chiral superfluid phase to mott phase and approaches a constant value at high temperature .
the thermal hall conductivity on the other hand is negative and never changes sign . at small magnetic field
the system is in the chiral superfluid phase . as the magnetic field increases
, @xmath5 decreases as the system transits to the mott insulator phase .
as mentioned above , the hardcore boson model eq .
[ hardcore ] should exhibit the same properties due to the mapping from spin variables to bosonic operators .
the schwinger boson mean - field theory@xcite provides another semiclassical approach to study the magnon excitations in quantum magnets .
this method has been employed to study the magnon bulk bands of su(2 ) heisenberg ferromagnets with dm interactions.@xcite in this representation , the spin operators are mapped to bosons using the following transformation @xmath118 subject to the constraint @xmath119 where @xmath120 or @xmath121 .
the schwinger boson method is built upon the symmetry of the hamiltonian .
since the xy model is not su(2)-invariant , the bond operators are different .
they are given by@xcite @xmath122 where @xmath123 .
the first two bond operators represent ferromagnetic and antiferromagnetic correlations and the last two represent easy axis and xy ferromagnetic correlations respectively
. however , these bond operators are not independent .
they are related by the identities : @xmath124 where @xmath125 represents normal ordering , _
i.e. _ , the creation operators placed to the left of the annihilation operators .
the schwinger boson hamiltonian of eq .
[ hh1 ] is given by @xmath126\label{schh}\nonumber\\&- \frac{d}{2}\sum_{\la\la ij\ra\ra}\big[e^{i\nu_{ij}\phi}:f_{ij,\up}^\dg f_{ij,\dw:}+h.c .
( \lambda-\sg \frac{h}{2 } ) c_{i,\sg}^\dg c_{i,\sg},\end{aligned}\ ] ] where the constant terms have been dropped , @xmath127 , and @xmath128 is a lagrange multiplier .
next , we rotation the coordinate by @xmath129 about the @xmath130-axis , which changes the bond operator @xmath131 , @xmath132 the quartic operators in eq .
[ schh ] can be brought to quadratic form by performing mean field hartree - fock decoupling @xmath133 using the fact that @xmath134 , we write @xmath135 , where @xmath136 are real variables .
the resulting mean field hamiltonian is given by @xmath137\nonumber\\&- \frac{d}{2}\sum_{\la\la ij\ra\ra,\sg}\big[e^{i\sg \nu_{ij}\phi } q_{1,-\sg}c_{i,\sg}^\dg c_{j,\sg}+h.c . \big]\nonumber\\
& -\frac{d}{2}\sum_{\la\la ij\ra\ra,\sg}\big[q_{2,-\sg}\sg c_{i,\sg}^\dg c_{j,\sg}+h.c .
\big]\nonumber\\&+\sum_{i,\sg } ( \lambda-\sg \frac{h}{2 } ) c_{i,\sg}^\dg c_{i,\sg } , \label{sch}\end{aligned}\ ] ] the momentum space hamiltonian can be written in the same form as the hp case with summation over spins and the bogoliubov hamiltonian is given by @xmath138\nonumber\\&- \tilde v_{2}i\sigma_y(\tau_+\gamma_\bo + h.c . ) , \label{schw}\end{aligned}\ ] ] where @xmath139 , @xmath140 , @xmath141 , and @xmath142 . to determine the @xmath143 and the @xmath144 one needs to diagonalize eq .
[ schw ] using the bogoliubov quasiparticles and solve a self - consistent equation .
we do not perform this procedure here as it is not necessary since it gives quantitatively the same result as the conventional spin wave theory .
in fact , experimental results reported on the kagome ferromagnet @xcite suggest that the holstein primakoff transformation is a better predictor than the schwinger boson representation . the purpose of this section is to provide an alternative representation of the spin wave hamiltonian .
the spinful haldane - hubbard model has recently garnered much attention.@xcite the hamiltonian is governed by @xmath145 we restrict the operators to obey only the boson commutation relation @xmath146=\delta_{ij}\delta_{\sg\sg^\prime}$ ] .
the phase diagram of this model for fermionic systems have been discussed in ref .
@xcite . for bosons ,
it has not been studied extensively in the topological context .
considering only the flux enclosed by the unit cell in the big triangular plaquettes of the nnn sites , in the strong coupling limit @xmath147 ; @xmath148 , the spinful haldane - hubbard model maps to @xcite@xmath149 the coupling constants are given by @xmath150 , @xmath151 , @xmath152 , and we have set @xmath37 .
the last term sums over the triangular plaquettes on the nnn sites .
it corresponds exactly to the dm interaction as discussed above . in this model ,
the interactions are ferromagnetic .
quite remarkably , eq . [ h ] is exactly the model studied in ref .
@xcite these results show a strong relationship between bosonic systems and quantum spin systems .
the main result of this paper is that the topological properties of hardcore bosons and that of magnons in quantum magnets are not independent . this is because topological effects manifest in the excitations of the corresponding system and the hardcore bosons correspond to quantum magnetic systems . in the spin language ,
we have shown that the haldane - hardcore bosons exhibit interesting nontrivial topological properties .
this suggests that the bosonic excitations of hardcore bosons ( which have to be magnons ) possess the same topological properties .
although qmc is not applicable in the present model , other numerical schemes should provide further insights into these topological effects .
we believe that these nontrivial topological effects should also manifest in frustrated systems by including a next - nearest - neighbour antiferromagnetic interaction .
we also showed that in the strong coupling limit of the spinful haldane - hubbard model on the honeycomb lattice , the system maps to heisenberg ferromagnet whose topological magnon excitations have been studied recently .
though the noninteracting hardcore boson model captures topologically nontrivial properties , in most cases of physical interest interactions are not negligible .
the simplest interaction is the repulsive nn interaction given by @xcite @xmath153 where @xmath154 . on the honeycomb lattice
, such interaction does not introduce any frustration in the system .
it is easy to show that in the spin variables , @xmath155 does not contribute to the gap opening at @xmath17 , hence its effects are negligible in the topological considerations .
furthermore , we claim that these topological properties should also be observed in frustrated systems @xcite . for the xy model on the honeycomb lattice ,
frustration is induced by an antiferromagnetic nnn coupling @xcite .
the system also possesses ordered states in the phase diagram , and the excitations of these ordered states should exhibit similar topological properties when a nontrivial gap is introduced . however , the explicit analysis of this frustrated model is beyond the purview of this paper .
the correspondence between magnetic spins systems and hardcore bosonic systems suggests experimental procedures in cold atoms on the honeycomb optical lattices to search for these interesting topological properties in bosonic and magnetic systems .
the author would like to thank african institute for mathematical sciences ( aims ) .
research at perimeter institute is supported by the government of canada through industry canada and by the province of ontario through the ministry of research and innovation .
99 m. z. hasan and c. l. kane , rev .
phys . * 82 * , 3045 ( 2010 ) .
qi and s .- c .
zhang , rev .
mod . phys . * 83 * , 1057 ( 2011 ) .
r. yu _ et al .
_ , science , * 329 * , 61 ( 2010 )
. y. l. chen _
_ , science * 329 * , 659 ( 2010 ) .
c.l . kane and e.j .
mele , phys .
lett . * 95 * , 146802 ( 2005 ) ; _ ibid _ phys .
lett . * 95 * , 226801 ( 2005 ) .
moore and l. balents , phys .
b * 75 * , 121306 ( 2007 ) .
l. fu , c.l .
kane , and e.j .
mele , phys .
* 98 * , 106803 ( 2007 ) .
et al . _ , nature phys . * 5 * , 438 , ( 2009 ) .
f. d. m. haldane , , 2015 ( 1988 ) .
_ , nature * 515 * , 237 ( 2014 ) .
c. hickey , p. rath , a. paramekanti , , 134414 ( 2015 ) .
et al . _ , , 094502 ( 2015 ) .
t. i. vanhala _
_ , arxiv:1512.08804 . v. s. arun , r. sohal , c. hickey , and a. paramekanti , , 115110 ( 2016 ) . c. strohm , g. l. j. a. rikken , and p. wyder , , 155901 ( 2005 ) .
l. sheng , d. n. sheng , and c. s. ting , , 155901 ( 2006 ) .
y. kagan , l. a. maksimov , , 145902 ( 2008 ) .
et al_. , , 225901 ( 2010 ) .
l. zhang , j. ren , j .- s .
wang , and b. li , , 225901 ( 2010 ) .
l. zhang , j. ren , j .- s .
wang , and b. li , j. phys
. condens .
matter * 23 * , 305402 ( 2011 ) .
r. matsumoto and s. murakami , , 197202 ( 2011 ) ; , 184406 ( 2011 ) .
d. j. thouless , m. kohmoto , m. p. nightingale , and m. den nijs , phys . rev . lett .
* 49 * , 405 ( 1982 ) ; m. kohmoto , annals of physics * 160 * , 343 ( 1985 ) .
m. p. a. fisher _ et al .
_ , , 546 ( 1989 ) .
d. jaksch , _ et al .
_ , , 3108 ( 1998 ) . c. orzel , _ et al
_ , science * 291 * , 2386 ( 2001 ) .
m. greiner _
et al . _ ,
nature , * 415 * , 39 ( 2002 ) .
duan , e. demler , and m. d. lukin , , 090402 ( 2003 ) . c. becker _
et al . _ , new journal of physics , * 12 * , 065025 ( 2010 ) .
stefan wessel , , 174301 ( 2007 ) .
j. y gan _ et al .
_ , , 214509 ( 2007 ) . t. matsubara and h. matsuda , prog . theor
* 16 * , 569 ( 1956 ) .
h. guo , y. niu , s. chen , s. feng , , 121401(r ) ( 2016 ) .
s. banerjee _
_ , arxiv:1511.05282 .
arovas and assa auerbach , , 316 ( 1988 ) .
a. a. burkov and a. h. macdonald , , 115320 ( 2002 ) .
c. n. varney _
et al . , _ , 115125 ( 2010 ) .
a. banerjee _
et al . , _ , 047208 ( 2008 ) ; j .-
et al . , _ , 037202 ( 2015 ) ; k. roychowdhury _ et al . , _ 075141 ( 2015 ) .
c. n. varney _
et al . _ , , 077201 ( 2011 ) ; c. n. varney
_ et al . _
; new j. phys .
* 14 * , 115028 ( 2012 ) ; j. carrasquilla _
et al . _ , 241109(r ) ( 2013 ) ; z. zhu , d. a. huse , and s. r. white , , 257201 ( 2013 ) . | this paper presents a connection between the topological properties of hardcore bosons and that of magnons in quantum spin magnets .
we utilize the haldane - like hardcore bosons on the honeycomb lattice as an example .
we show that this system maps to a spin-@xmath0 quantum xy model with a next - nearest - neighbour dzyaloshinsky - moriya interaction .
we obtain the magnon excitations of the quantum spin model and compute the edge states , berry curvature , thermal and spin nernst conductivities . due to the mapping from spin variables to bosons ,
the hardcore bosons possess the same nontrivial topological properties as those in quantum spin system .
these results are important in the study of magnetic excitations in quantum magnets and they are also useful for understanding the control of ultracold bosonic quantum gases in honeycomb optical lattices , which is experimentally accessible . |
chern simons ( cs ) theories are certainly some of the most interesting field theories in 2 + 1 dimensions . among the interesting phenomena one sees in these theories
are : the quantum hall effect , landau levels , non - trivial topological structures , vortices , and anyons . for a review of these phenomena
see @xcite .
some of these ( 2 + 1)-dimensional phenomena have been observed experimentally in condensed matter systems .
cs theories are also often described as topological theories @xcite since for most gauge groups the cs coupling must obey a quantization condition for the theory to remain gauge invariant . within this rich literature on cs theories
there is also considerable work on sym cs theories .
these theories have their own remarkable properties .
it has been shown that there is a finite anomaly that shifts the cs coupling @xcite , and it has been conjectured by witten @xcite that this theory spontaneously breaks supersymmetry for some values of the cs coupling
. there are other related reasons for interest in cs theories .
witten @xcite has conjectured that string field theory is essentially a non - commutative cs theory .
recently this led to a conjecture by susskind @xcite that relates string theory to the fractional quantum hall effect .
since these theories are interesting from so many different points of view , it is certainly useful to numerically simulate them .
the method we will use is sdlcq ( supersymmetric discrete light - cone quantization ) .
this is a numerical method that can be used to solve any theory with enough supersymmetry to be finite .
the central point of this method is that using dlcq we can construct a finite dimensional representation of the superalgebra @xcite . from this representation of the superalgebra
, we construct a finite - dimensional hamiltonian which we diagonalize numerically .
we repeat the process for larger and larger representations and extrapolate the solution to the continuum .
we have already solved standard ( 2 + 1)-dimensional supersymmetric yang mills ( sym ) theories by this method @xcite , so it is clear that sym - cs theories are within our reach . in this paper we will start with sym cs theory in 2 + 1 dimensions and dimensionally
reduce it to two dimensions by requiring all of the fields to be independent of the transverse coordinate .
this is an interesting ( 1 + 1)-dimensional supersymmetric theory in its own right .
it is a good starting point for our numerical simulations because we can solve this problem using two completely independent codes , one a mathematica code and the other a c++ code .
when we move on to 2 + 1 dimensions in future work , the ( 1 + 1)-dimensional results will be related to the ( 2 + 1)-dimensional results in a nontrivial way .
we found this to be the case in our previous work on ( 2 + 1)-dimensional sym theories , and it will most probably be the case here as well .
this will provide us an important check on our code for solving the ( 2 + 1)-dimensional problem .
it is important to develop this chain of numerical checks , since there are no analytical or other numerical solutions to check our results against .
many of the most interesting aspects of cs theory will be lost by this reduction , most notably the quantization of the cs coupling .
however , one particularly interesting property that will be preserved is the fact that the cs term simulates a mass for the theory .
it is well known that supersymmetric abelian cs theory is simply the theory of a free massive fermion and a free massive boson . in the non - abelian theory
additional interactions are introduced , but we expect to also see this mass effect .
this is particularly interesting here because dimensionally reduced @xmath0 sym is a very stringy theory .
the low - mass states are dominated by fock states with many constituents , and as the size of the superalgebraic representation is increased , states with lower masses and more constituents appear .
the connection between string theory and supersymmetric gauge theory leads one to expect this type of behavior ; however , these gauge theories are not very qcd - like .
ultimately one might like to make a connection with the low - mass spectrum observed in nature .
we expect to find states in sym
cs theory that are much more qcd - like , because of the effective mass for the constituents introduced by the cs term .
we will see that some of the low - mass states will be dominated by valence - like fock states that have only a few constituents . as we go to larger and larger representation of the superalgebra
, we will resolve these states better and better , and any new states that appear will be heavier .
we will also find highly mixed states without a valence structure .
finally , we find that at strong coupling the low - energy spectrum is dominated by states that are a reflection of the bps states of the sym theory in 1 + 1 dimensions @xcite and are therefore independent of the coupling @xmath1 .
> from the wave functions of the bound states we will be able to find the structure functions of both the valence and sea partons . some of the highly mixed states will have a double - humped structure function .
some of these structure functions are similar to those conjectured in various phenomenological calculations ; they are the result of the solution of a nontrivial gauge theory . in sec .
[ sec : supercs ] we give a general discussion of supersymmetric light - cone - quantized sym
cs theory in light - cone gauge .
we then give the dimensionally reduced discrete formulation of the supercharges and discuss the other symmetries of the theory . in sec .
[ sec : results ] we discuss our numerical method and some of the new wrinkles that appear in cs theory .
we also present and discuss in this section the spectrum of bound states as well as a variety of properties of these states including their structure functions .
we will see that it is a very qcd - like theory as opposed to pure sym theory , which is very stringy . finally , in sec .
[ sec : summary ] , we summarize the results and discuss the prospects and challenges for calculating in the full ( 2 + 1)-dimensional theory .
we will consider @xmath0 supersymmetric cs theory in 2 + 1 dimensions as the starting point of our discussion . while we will reduce this theory to a ( 1 + 1)-dimensional theory for numerical simulation here
, we eventually plan to simulate the full theory , and it is therefore useful to present a detailed light - cone formulation in light - cone gauge .
the lagrangian of this theory is @xmath2 where @xmath3 is the cs coupling and @xmath4 the two components of the spinor @xmath5 are in the adjoint representation of @xmath6 or @xmath7 .
we will work in the large-@xmath8 limit .
the field strength and the covariant derivative are @xmath9\ , , \quad \quad d_{\mu}=\partial_{\mu}+ig[a_{\mu},\quad]\,.\ ] ] the supersymmetric variations of the fields are @xmath10 where and @xmath11 but is more natural for light - cone quantization . in this representation the spinor term of the cs lagrangian enters eq .
( [ eq : cslagrangian ] ) with a plus sign . ] @xmath12 this leads to the supercurrent @xmath13 in the usual manner via @xmath14 light - cone coordinates in 2 + 1 dimensions are @xmath15 where @xmath16 is the light - cone time and @xmath17 .
the totally anti - symmetric tensor is defined by @xmath18 .
the variations of the three parts of the lagrangian in eq .
( [ lagrangian ] ) determine the ( ` chiral ' ) components @xmath19 of the supercharge via eq .
( [ lq ] ) to be @xmath20 explicitly they are @xmath21\right)\ , , \nonumber \\
q^+&=&-i 2^{5/4}\int d^2x\ , \psi\left(\partial^+ a^2 -\partial^2 a^+ + ig[a^+,a^2]\right)\,.\end{aligned}\ ] ] one can convince oneself by calculating the energy - momentum tensor @xmath22 that the supercharge fulfills the supersymmetry algebra @xmath23 in order to express the supercharge in terms of the physical degrees of freedom , we have to use equations of motion , some of which are constraint equations .
the equations of motion for the gauge fields are @xmath24 where @xmath25 for @xmath26 this is a constraint for @xmath27 , @xmath28 in light - cone gauge , @xmath29 , this reduces to @xmath30\,.\ ] ] the equation of motion for the fermion is @xmath31 expressing everything in terms of @xmath32 and @xmath33 leads to the equations of motion @xmath34 the second of which is a constraint equation .
the constraint equations are used to eliminate @xmath33 and @xmath27 .
we now reduce the theory dimensionally to two dimensions by setting @xmath35 and @xmath36 for all fields .
this yields , from eq .
( [ supercharges ] ) and the constraints , @xmath37 + 2\psi\psi-\frac{\kappa}{g } \partial_-\phi\right)\frac{1}{\partial_-}\psi\,.\ ] ] the mode expansions in two dimensions are @xmath38\ , , \nonumber\\ % \psi_{ij}(0,x^- ) & = & \frac{1}{2\sqrt{\pi}}\int_0^\infty dk^+\left[b_{ij}(k^+)e^{-{\rm i}k^+x^-}+ b^\dagger_{ji}(k^+)e^{{\rm i}k^+x^-}\right]\,.\end{aligned}\ ] ] to discretize the theory we impose periodic boundary conditions on the boson and fermion fields alike , and obtain an expansion of the fields with discrete momentum modes .
thus the discrete version of the cs part of the supercharge is @xmath39 where @xmath40 is a rescaled cs coupling and @xmath41 and @xmath42 are rescaled discrete field operators @xmath43 the ordinary supersymmetric part of the supercharge is listed elsewhere @xcite .
it is important to note that the supercharge for @xmath0 sym in 2 + 1 dimensions has a contribution of the form @xmath44 of course , the light - cone energy is @xmath45 , so @xmath46 behaves like a mass , and here we see that @xmath47 appears in a very similar way to @xmath46 and therefore behaves in many ways like a mass .
when comparing the two contributions to the supercharge , we see that we have a relative @xmath48 between them .
thus the usual eigenvalue problem @xmath49 has to be solved by using fully complex methods .
we retain the @xmath50-symmetry , which is associated with the orientation of the large-@xmath8 string of partons in a state @xcite . in a ( 1 + 1)-dimensional model
this orientation parity is usually referred as a @xmath51 symmetry , and we will follow that here .
it gives a sign when the color indices are permuted @xmath52 we will use this symmetry to reduce the hamiltonian matrix size and hence the numerical effort .
all of our states will be labeled by the @xmath51 sector in which they appear .
we will not attempt to label the states by their normal parity ; in the light - cone this is only an approximate symmetry .
such a labeling could be done in an approximate way , as was shown by hornbostel @xcite , and might be useful for comparison purposes if at some point there are results from lattice simulations of the present theory .
we convert the mass eigenvalue problem @xmath53 to a matrix eigenvalue problem by introducing a basis where @xmath54 is diagonal . in sdlcq
this is done by first discretizing the supercharge @xmath55 and then constructing @xmath56 from the square of the supercharge : @xmath57 . to discretize the supercharge
, we introduce discrete longitudinal momenta @xmath58 as fractions @xmath59 of the total longitudinal momentum @xmath54 , where @xmath60 is an integer that determines the resolution of the discretization and is known in dlcq as the harmonic resolution @xcite . because light - cone longitudinal momenta are always positive , @xmath60 and each @xmath61 are positive integers ; the number of constituents is then bounded by @xmath60 .
the integrals in @xmath55 are approximated by a trapezoidal form .
the continuum limit is then recovered by taking the limit @xmath62 . in constructing the discrete approximation we drop the longitudinal zero - momentum mode .
for some discussion of dynamical and constrained zero modes , see the review @xcite and previous work @xcite .
inclusion of these modes would be ideal , but the techniques required to include them in a numerical calculation have proved to be difficult to develop , particularly because of nonlinearities . for dlcq calculations that can be compared with exact solutions , the exclusion of zero modes does not affect the massive spectrum @xcite . in scalar theories
it has been known for some time that constrained zero modes can give rise to dynamical symmetry breaking @xcite , and work continues on the role of zero modes and near zero modes in these theories @xcite . to obtain the spectrum of the cs theory
we solve the complex eigenvalue problem , eq .
( [ evp ] ) . for the numerical evaluation
we can exploit the structure of the supercharge @xmath63 where @xmath41 and @xmath42 are real matrices .
the hamiltonian has thus an easy decomposition into a real and imaginary part in the bosonic sector @xmath64 and the fermionic sector @xmath65 our earliest sdlcq calculations were done using a code written in mathematica and performed on a pc .
this code was then rewritten in c++ and substantially revised .
it runs on a linux workstation with 2 gb of ram and can handle as many as 2,000,000 fock states .
the present calculation was done in both codes as a check .
we limit the calculation to resolution @xmath66 because it seems sufficient here and because the present c++ code , which was primarily written for ( 2 + 1)-dimensional models , would require some non - trivial modifications to run at higher longitudinal resolutions .
the low - energy spectrum , with @xmath67 , is fit to @xmath68 .
these fits are shown in fig .
[ fig1 ] for some of the low - energy states . in principle
, higher energy states can also be found , but at these couplings the states just above these are difficult to disentangle because of level crossings . in the @xmath51 even sector
we find the continuum masses to be ( in units of @xmath69 ) @xmath70 , 18.33 , 27.46 , and 43.20 , whereas in the @xmath51 odd sector we have @xmath71 , 29.13 , 32.83 , 39.52 , and 47.40 .
[ cols="^,^ " , ]
we have presented an analysis of dimensionally reduced supersymmetric ym theory with a cs term using the numerical method sdlcq , which exactly preserves the supersymmetry of this theory .
we constructed finite dimensional representations of the superalgebra and from them a finite dimensional hamiltonian which we solved numerically .
as we go to higher dimensional representations , the solutions converge rapidly .
> from these solutions we extracted the properties of the bound states of this theory .
we found that the bound states are very different from the bound states of @xmath0 sym in 1 + 1 dimensions .
the bound states of sym are characterized by their stringy nature , that is a set of states where the states with the most partons have the lowest energy .
the cs theory is generally very qcd - like .
the states with more partons tend to be more massive , and many of the low - mass states have a valence - like structure .
these states have a dominant component of the wave function with a particular number and type of parton .
we have found the spectrum of these states and studied it as a function of the cs coupling .
we found that , as expected , the cs coupling behaves as an effective mass .
as it increases we see that the masses of the bound states generally increase , and the average number of partons of the low - mass bound states tends to decrease .
also , the sym theory has an interesting set of massless bps states .
these states are reflected in the cs theory as a set of states whose masses are approximately independent of @xmath1 and equal to the square of the sum of the cs masses of the partons in the bound state .
since these bps states are also present in the ( 2 + 1)-dimensional sym theory , we expect to see their reflection in the ( 2 + 1)-dimensional cs theory .
we have investigated the structure functions for these bound states and found that they behave in very interesting ways .
the distribution for the states with a valence structure are peaked near the inverse of the number of valence partons , and the sea distribution appears to peak at small @xmath72 in most if not all cases .
we also see strongly mixed states with interesting double - humped structure functions .
there have been interesting conjectures about structure functions of this type in qcd @xcite , and it is interesting that we actually find such a structure function in the solution of an actual gauge theory . in summary ,
the sdlcq solutions of dimensionally reduced supersymmetric cs theory are very interesting , maybe more interesting , at least with respect to their qcd - like structure , than are the solutions for pure supersymmetric sym theory .
clearly this provides a strong base and encouragement to move on to 2 + 1 dimensions .
g. v. dunne , arxiv : hep - th/9902115 .
for a nice discussion of the topological properties in the context of an abelian theory , see a. polychronakos , ann .
phys . * 203 * , 231 ( 1990 ) .
kao , k. lee , and t. lee , phys .
b * 373 * , 94 ( 1996 ) , arxiv : hep - th/9506170 .
e. witten , in _ the many faces of the superworld _ , edited by m.a .
shifman , ( world scientific , singapore , 2000 ) , p. 156
, arxiv : hep - th/9903005 .
e. witten , nucl .
phys . * b268 * , 253 ( 1986 ) . l. susskind , arxiv : hep - th/0101029 .
y. matsumura , n. sakai , and t. sakai , phys .
d * 52 * , 2446 ( 1995 ) .
j. r. hiller , s. pinsky and u. trittmann , phys .
d * 64 * , 105027 ( 2001 ) , arxiv : hep - th/0106193 . j. r. hiller , s. pinsky and u. trittmann , phys .
d * 63 * , 105017 ( 2001 ) , arxiv : hep - th/0101120 .
f. antonuccio , o. lunin , s. pinsky , and s. tsujimaru , phys .
d * 60 * , 115006 ( 1999 ) , arxiv : hep - th/9811254 . o. lunin and s. pinsky , phys .
d * 63 * , 045019 ( 2001 ) , arxiv : hep - th/0005282 . s. pinsky and u. trittmann , phys .
d * 62 * , 087701 ( 2000 ) , arxiv : hep - th/0005055 . j. r. hiller , o. lunin , s. pinsky and u. trittmann , phys . lett .
b * 482 * , 409 ( 2000 ) , arxiv : hep - th/0003249 .
p. haney , j. r. hiller , o. lunin , s. pinsky and u. trittmann , phys .
d * 62 * , 075002 ( 2000 ) , arxiv : hep - th/9911243 . o. lunin and s. pinsky , aip conf .
* 494 * , 140 ( 1999 ) , arxiv : hep - th/9910222 .
f. antonuccio , o. lunin and s. pinsky , phys .
d * 59 * , 085001 ( 1999 ) , arxiv : hep - th/9811083 . f. antonuccio , h. c. pauli , s. pinsky and s. tsujimaru , phys . rev .
d * 58 * , 125006 ( 1998 ) , arxiv : hep - th/9808120 .
f. antonuccio , o. lunin , s. pinsky , h. c. pauli and s. tsujimaru , phys .
d * 58 * , 105024 ( 1998 ) , arxiv : hep - th/9806133 .
f. antonuccio , o. lunin and s. pinsky , phys .
d * 58 * , 085009 ( 1998 ) , arxiv : hep - th/9803170
. f. antonuccio , o. lunin and s. s. pinsky , phys .
b * 429 * , 327 ( 1998 ) , arxiv : hep - th/9803027 .
d. kutasov , nucl .
b * 414 * , 33 ( 1994 ) .
k. hornbostel , ph.d .
thesis , slac report no . 333 , 1988 .
pauli and s.j .
brodsky , phys .
d * 32 * , 1993 ( 1985 ) ; * 32 * , 2001 ( 1985 ) .
brodsky , h .- c .
pauli , and s.s .
pinsky , phys .
rep . * 301 * , 299 ( 1998 ) , arxiv : hep - ph/9705477 .
rozowsky and c.b .
85 * , 1614 ( 2000 ) , arxiv : hep - th/0003301 .
u. trittmann , nucl .
b * 587 * , 311 ( 2000 ) , arxiv : hep - th/0005075 ; arxiv : hep - th/0110058 .
chernyak and a.r .
zhitnitsky , phys . rep . * 112 * , 173 ( 1984 ) . | a supersymmetric formulation of a three - dimensional sym chern simons theory using light - cone quantization is presented , and the supercharges are calculated in light - cone gauge . the theory is dimensionally reduced by requiring all fields to be independent of the transverse dimension .
the result is a non - trivial two - dimensional supersymmetric theory with an adjoint scalar and an adjoint fermion .
we perform a numerical simulation of this sym chern simons theory in 1 + 1 dimensions using sdlcq ( supersymmetric discrete light - cone quantization ) .
we find that the character of the bound states of this theory is very different from previously considered two - dimensional supersymmetric gauge theories .
the low - energy bound states of this theory are very `` qcd - like . ''
the wave functions of some of the low mass states have a striking valence structure .
we present the valence and sea parton structure functions of these states .
in addition , we identify bps - like states which are almost independent of the coupling .
their masses are proportional to their parton number in the large - coupling limit . |
it is of great importance to study the anthropogenic caesium radioisotopes @xmath6 and @xmath0 , since their production and emission rates are much higher than other radioisotopes from nuclear fissions at nuclear power plants ( npp ) and they have rather low movility in soils .
thus , @xmath5cs are widely used as tracer radionuclides for monitoring the npp - related environmental radioactivity . the @xmath5cs nucleus ( @xmath7y )
is a long - lived beta emitter decaying a @xmath8 branching ratio to the metastable state ( @xmath9 m ) of @xmath5ba@xmath10 at 661.7kev .
while the @xmath4cs nucleus is also a beta emitter , its decay time ( @xmath11y ) is much shorter and emits many gamma - rays correlated with the beta electron . due to these different life - times ,
it is possible to date the anthropogenic radioactivity with the ratio of @xmath4cs to @xmath5cs concentrations in soil measured .
however , since the @xmath4cs has much shorter half life - time compared to the @xmath6 , the dating range depends solely on a minimum detectable amount ( mda ) of activity for the short - lived @xmath4cs radioisotope .
moreover , it is well known that the @xmath6 contamination arises mainly from three different sources : atmospheric nuclear weapon tests ( nwt ) in the period of the late 1950s and early 1960s , in particular , in the nothern hemisphere ( unscear , 2000 ) , the chernobyl accident taken place in may 1986 , and discharges from nuclear power plants ( npps ) . on the contrary ,
the @xmath0 does not come from the nwt , since it is induced only by the neutrons impinged on the fission radionuclide @xmath12 , which rarely occurs in the course of a nuclear explosion due to a lack of the reaction time .
however , the @xmath0 can be produced in reactors at npps .
thus , knowing the ratio of @xmath4cs to @xmath5cs concentrations in soil may determine even their origin .
note that the fallout ratio of the chernobyl explosion is known to be about @xmath13 in 1986 ( thomas and martin , 1986 ) . in order to determine the ultra - low level caesium concentrations quantitatively in soil ,
it is essential to understand the natural background originated mainly from uranium- , thorium- , actinum - series nuclides , and the potassium @xmath1k .
the @xmath14-ray peaks of the @xmath0 , in particular , those at 604.7kev and 795.6kev are overlapped with a great deal of @xmath14-ray peaks from natural radionuclides . moreover
, strong peaks such as a 1460kev line of the @xmath1k at high energies produce the compton - continuum background that covers the wide range of the gamma energies for caesium decays .
thus , we need to get rid of the natural background as much as possible .
one of the best ways known to eliminate the @xmath15 is to use the ammonium molybdophosphase ( amp ) precipitation method .
amp is an inorganic compound that selects the caesium exclusively and has high adsorption capacity ( suss and pfrepper , 1981 ) .
this method has been widely applied to sea - water samples in order to enrich the cs concentration . however , this radiochemical extraction for a soil sample should be emphasized by its effective background suppression , thereby significantly lowering the minimum detection limit . in the present work ,
we want to investigate the minimum detection limit to the measurement of the @xmath0 in surface soils in mountain areas near the gori npp in korea , aiming at determining the ratio of the @xmath0 to @xmath6 . in korea ,
eighteen nuclear reactors are being operated at four different npp sites .
one of the npp sites is located in gori , a small town in the south - east coast of korea , where four pressurized - water reactors are in operation .
the first gori reactor has been operated first time in korea since 1978 .
the environmental radioactivity in the vicinity of the gori npp has been monitored last years .
the samples were taken from soils , rain water , sea water , surface water , milk , seaweed , egg , etc .
recently , high concentrations of the @xmath5cs were first reported in the soil samples from the mountain areas near the gori npp in the range of @xmath16 bq / kg - dry , while the @xmath5cs concentrations in surface soils of korea are known to be ranged in @xmath17 bq / kg - dry with a mean value of @xmath18 bq / kg - dry ( kim et al . , 1998 ) .
taking into account the fact that the chernobyl accident took place more than thirty years ago , the @xmath4cs concentration was reduced by three orders of magnitude , while a half of the original @xmath5cs concentration is survived ( pourcelot et al . , 2003 ) .
if the ratio of @xmath4cs to @xmath5cs is higher than the average ratio to be mapped out for the distant area from the gori power plant , the origin should be investigated thoroughly . in this regard ,
it is of great significance to conduct the methodological study for the measurement of the ultra - low level concentration of @xmath4cs in its own right .
the present work is organized as follows : in section 2 , we briefly explain the process of the experiment . in section 3 ,
we discuss the detection limits to the @xmath0 concentration . in the final section ,
we summarize and draw conclusions of this work .
soil samples were collected from 96 sites at top areas of four different mountains near the gori npp : mt daleum ( @xmath19 , @xmath20 ) , mt ilgwang ( @xmath21 , @xmath20 ) , mt samgak ( @xmath22 , @xmath23 ) , and mt daewoon ( @xmath24 , @xmath23 ) .
all these mountains are located within 7 - 12 km distance from the gori npp .
uncultivated soil was sampled from the @xmath25 layer .
each sample was taken in close proximity to give approximately @xmath26 to @xmath27 kg .
since we are mostly interested in measuring both @xmath0 and @xmath6 , we have selected the samples with the highest @xmath6 concentration : @xmath28 bq / kg - dry . from that sample , we took two samples for the measurement and prepared them with different methods : sample a was air - dried at 65@xmath29c for 10 hours and sieved with 2 mm mesh size .
it was then filled in a 450ml marinelli beaker . on the other hand
, sample b was heated at 450@xmath29c for 10 hours .
one liter of 37% hcl solvent was added to sample b , and the solution was stirred with heat for 4 hours .
soil contents in the sample solution were sieved through two layers of the normal filter(5c , 125mm@xmath30 ) and the glass - fiber filter(gf / c , 110mm@xmath30 ) in the buhner filtering system .
unfiltered sediment was washed out with hot water .
for precipitation with ammonium - molybdophosphate(amp ) , the sample was stirred with the 80 g amp solution at ph2 for 10 hours .
the sample was then sieved with a membrane filter of 0.45@xmath31 m pore size , and after cleansing the sample bottle with 1% hcl solution , the cleansing solution was put to the filter to sieve remaining contents . after the amp precipitation , the sample was dried with infrared light , and put into a standard 55ml marinelli beaker for @xmath14-ray measurement .
we have measured the @xmath5cs concentration in the soil samples by directly measuring the 661.7kev , 604.7kev , and 795.8kev gamma rays with an hpge detector .
the hpge @xmath14-ray detector has a 30% relative efficiency at 1330kev of @xmath32co , and the schematic setup of the detector and the shielding system is shown in fig .
[ fig : setup ] .
the hpge detector was coupled to standard nim electronics and a pc with gamma - vision@xmath33 spectroscopy software .
the system was shielded with a uniform lead castle of 10 cm in thickness .
certified reference materials ( crm ) of mixed gamma - ray sources from kriss ( korea research institute of standards and science ) were employed for the efficiency calibration of the system .
the crm provides with 13 gamma - ray peaks from 10 radioactive elements : @xmath34am ( 59.5kev ) , @xmath35cd ( 88.0kev ) , @xmath36co ( 122.1kev , 136.5kev ) , @xmath37ce ( 165.86kev ) , @xmath38cr ( 320.08kev ) , @xmath39sn ( 391.70kev ) , @xmath40sr ( 514.00kev ) , @xmath5cs ( 661.66kev ) , @xmath41y ( 898.04kev , 1836.05kev ) , and @xmath32co ( 1173.23kev , 1332.49kev ) .
figure [ fig : eff ] represents efficiency curves of the hpge detector , fitted with the following empirical logarithmic polynomials : @xmath42 where @xmath43 denotes the gamma - ray energy .
the @xmath44 are the fitting parameters of the polynomial .
the efficiency for the 450ml sample is lower than that for the 45ml amp sample in the low - energy region , because of a larger self - absorption of gamma - rays in the sample volume .
cs and @xmath5cs.,width=302,height=302 ] figure [ fig : amp ] depicts the gamma - ray spectrum of the amp sample overlaid with that of the dry - soil sample .
the chemical separation shows a significant improvement in a signal - to - background ratio .
the background is reduced dramatically by almost a order of magnitude due to the amp method .
the upward arrows in fig .
[ fig : amp ] designate the positions of gamma - ray energies from @xmath4cs and @xmath5cs .
we are now in a position to discuss the mda of activity of samples . the criterion for limits of detectability is usually given by the mda of activity .
the most widely used definition of the mda was first defined by curie ( curie , 1968 ) as follows : @xmath45 where @xmath46 denotes the number of background events measured during the time @xmath47 .
the @xmath48 represents the detection efficiency as defined in eq.([eq:1 ] ) , and @xmath49 stands for a branching fraction for @xmath4cs emitting the respective gamma - ray .
the minimum detectable activity depends on the square root of the number of background events which mainly come from three different reasons : the sample itself , compton continuum , and natural radioactivity .
note that it actually depends on the inverse square root of the time , since the background events also increase as a function of time . in fig .
[ fig : spectrum ] , we draw the gamma - ray spectrum of the dry - soil sample measured at a normal laboratory without the amp precipitation taken into account .
the peak at 609 kev corresponds to that of the gamma - ray from @xmath50 . because of the background
, it is very difficult to identify the spectrum of the sample , in particular , in the higer region of the gamma - ray energy .
cs can not be resolved from the strong 794.7 kev gamma - ray peak from @xmath51ac.,width=302,height=226 ] in order to reduce the background of the sample , we have brought the sample to an underground laboratory located in yangyang , a small town in the east coast of korea .
the underground laboratory is specialized to search for dark matter for which is it essential to develop low - background measurement .
it utilizes the space in a tunnel of yangyang pumped storage power plant located deep under a mountain .
figure [ fig : amp ] shows the gamma - ray spectrum of the amp - precipitated sample measured at the underground laboratory in yangyang . as compared with fig .
[ fig : spectrum ] , it is shown that the background is drastically reduced . in fig .
[ fig : amp2 ] , we present the full energy - range gamma - ray energy spectrum of the amp - precipitated sample measured at the underground laboratory . as mentioned before ,
the amp precipitation method has greatly reduced the mda by almost one order of magnitude .
it filters selectively out cs isotopes from the sample .
moreover , the compton continuum background from @xmath1k is also shown to be noticeably reduced . however , the 795 kev peak from @xmath51ac is interfered with the 795.6kev from @xmath4cs .
therefore , we found it difficult to resolve the peak from @xmath4cs even with very low background .
the 604.7 kev peak from @xmath4cs can be well resolved from underlying background .
the 609.3 kev peak from @xmath52bi appears in the vicinity of the 604.7 kev peak , but a good energy resolution of the hpge detector helps the two peaks resolved clearly .
it therefore should be noted that the 604.7 peak should be searched for in the @xmath4cs survey .
it is then necessary to suppress background contributions underneath the 604.7 kev peak .
the region near the peak is between the 661.7 kev @xmath5cs peak and its compton edge .
the yield in the region is thus mainly due to multiple compton scattering of the 661.7 kev @xmath5cs gamma ray in a hpge crystal .
a possible experimental approach is a gamma - ray detection with a compton suppressed system so that it may be possible to suppress such multiple compton scattering events and to lower the mda significantly .
figure [ fig : simple ] summarizes how to resolve peaks of @xmath0 and @xmath6 from the background .
table [ tab:1 ] lists the values of the mda from the present work , with and without the amp precipitation method considered .
[ tab:1 ] .the values of the mda for the @xmath4cs concentration : the unit is given in bq / kg - dry .
[ cols="<,^,^",options="header " , ]
in the present work , we investigated the caesium concentrations in soil samples taken from mountain areas in the vicinity of the gori nuclear power plant , emphasizing the minimum detection limit to the measurement of @xmath0 concentration .
we carried out the sampling by taking @xmath53 cm layer of surface soil at the top areas of four different mountains within the @xmath54 km distance of the gori nuclear power plant .
we have used the ammonium molybdophosphase ( amp ) precipitation method to get rid of the @xmath1k existing in natural radioactivity , which reduces the minimum detectable activity ( mda ) of activity about ten times smaller than those without the amp precipitation method . even though we were able to reduce the mda for the @xmath0 from @xmath55 to @xmath2 by one order of magnitude smaller , using the ammonium - molybdophosphate precipitation method in order to lower the background mainly arising from the natural radioacnuclide @xmath15 , it is not possible to determine the @xmath0 concentration quantitatively . while it is of great difficulty to track down the origin of the @xmath6 that are found in higher concentrations in soil samples near the gori nuclear power plant ,
the results imply that the @xmath6 found in the samples are at least 20 years old . taking into account the fact that the chernobyl accident took place 22 years ago from now
, one can assume that the present @xmath6 concentrations may have come from the chernobyl explosion .
since it is known that the fallout ratio of the chernobyl accident is about @xmath13 in 1986 ( thomas and martin , 1986 ) , the present ratio should be dropped to be about @xmath56 , considering that the half life - times of the @xmath0 and @xmath6 are , respectively , @xmath57y and @xmath58y .
in fact , the fallout due to the chernobyl accident did not affect korea much , since the distribution of @xmath6 is similar to those of other countries ( c.s .
et al . _ , 1998 ) .
thus , we can infer from it that the present concentration of the @xmath0 should be less than @xmath59 bq / kg - dry which is definitely lower than the minimum detection limit reached in the present work . in order to reduce the minimum detectable amount of @xmath0 activity
further , we need to get rid of the compton continuum , using the compton suppresion method , which is now under investigation .
the present work is supported by inha university research grant ( inha-37453 ) .
aoyama , m. _ et al . ,
low level @xmath6 measurements in deep seawater samples . app .
rad . and isot .
53 , 159 - 162 .
djingova , r. and kuleff , i. , 2002 .
concentration of caesium-137 , cobalt-60 and potassium-40 in some wild and edible plants around the nuclear power plant in bulgaria .
j. environ .
radioactivity 59 , 61 - 73 .
tranter , t.j .
_ et al . , _ 2002 .
evaluation of ammonium molybdophosphate - polyacrylonitrile ( amp - pan ) as a cesium selective sorbent for the removal of @xmath6 from acidic nuclear waste solutions . adv .
6 , 107 - 121 .
yoshida , s. and muramatsu , y. 1998 .
concentrations of alkali and alkaline earth elements in mushrooms and plants collected in a japanese pine forest , and their relationship with @xmath6 .
j. environ .
radioactivity 41 , 183 - 205 . | we investigate the caesium concentrations in soils in mountain areas near gori nuclear power plant in korea , focusing on the measurement limits to the @xmath0 . in order to lower the minimum detectable amount ( mda ) of activity for the @xmath0
, we have used the ammonium molybdophosphase ( amp ) precipitation method to get rid of the @xmath1k existing in natural radioactivity , which reduces the mda of activity about ten times smaller than those without the amp precipitation method .
the mda results for the @xmath0 were found to be in the range between @xmath2 and @xmath3 bq / kg - dry weight . in order to diminish the background ,
we also have measured a part of the soil samples in yangyang , a small town in the east coast of korea .
however , it turns out that in order to detect the @xmath0 in the samples the mda should be reduced to the level of mbq / kg - dry weight .
@xmath4cs , @xmath5cs , ammonium molybdophosphate ( amp ) precipitation method , minimum detectable amount ( mda ) |
the celebrated theorem of cheeger and mller establishes the equality of reidemeister and analytic torsion on an odd - dimensional closed manifold equipped with a flat euclidean bundle .
this was originally conjectured by ray - singer @xcite , proven by cheeger and mller @xcite and subsequently extended by mller @xcite and bismut - zhang @xcite .
the importance and usefulness of the theorem stems from the fact that the reidemeister torsion , or r - torsion , is a combinatorial invariant of simplicial complexes while the analytic torsion is a smooth invariant defined via the spectrum of the hodge laplacian .
this connection is behind many applications in topology , number theory , and mathematical physics .
one particularly interesting aspect of this theorem is that it allows us to use analysis to study the size of the torsion in cohomology .
for example , if @xmath0 is a complex of free abelian groups and @xmath1 then the reidemeister torsion ( after some canonical choices , see ( * ? ? ?
* example 1.3 ) ) is given by @xmath2 this relationship has been recently exploited to study the growth of torsion in group homology by studying the analytic torsion of locally symmetric spaces @xcite . + since locally symmetric spaces often have a natural compactification to a stratified space , it is natural to look for an analogue of the the cheeger - mller theorem in this context .
on such a space , the natural cohomology to consider is the intersection cohomology of goresky - macpherson @xcite . in 1987
, dar @xcite introduced the intersection @xmath3-torsion on stratified spaces , an analogue of the @xmath3-torsion defined in terms of intersection cohomology and a choice of perversity .
dar also proposed that the intersection @xmath3-torsion should be related to the analytic torsion of some appropriately chosen incomplete iterated edge metric adapted to the singularities of the stratified spaces .
there are many recent advances on this question @xcite , but still no relation obtained even in the simplest case where the stratified space has only isolated conical singularities . in @xcite
, we proposed instead to relate the intersection @xmath3-torsion on a stratified space of depth one to the analytic torsion of a fibred cusp metric using the geometric microlocal analysis methods of melrose .
specifically , let @xmath4 be the interior of a manifold with boundary @xmath5 and assume that the boundary participates in a fiber bundle of closed manifolds @xmath6 let @xmath7 be a boundary defining function for @xmath8 that is , a smooth function on @xmath5 that vanishes precisely at @xmath9 and with non - vanishing differential there .
a metric @xmath10 on @xmath4 is a fibered cusp metric , or @xmath11-metric , if it is asymptotically of the form @xmath12 where @xmath13 is a submersion metric on @xmath14 we say @xmath10 is an ` even ' @xmath11-metric if @xmath15 and @xmath16 are functions of @xmath17 see @xcite for more details .
let @xmath18 be a vector bundle with flat connection @xmath19 and bundle metric @xmath20 not necessarily compatible with @xmath21 but smooth all the way down to @xmath22 in fact we assume that @xmath23 is _ even _ , meaning that it extends smoothly to the double of @xmath4 across @xmath22 the analytic torsion of @xmath24 is defined in @xcite following @xcite by means of the renormalized trace of the heat kernel , since the usual operator trace is not defined .
we prove that this analytic torsion can be expressed in terms of the reidemeister torsion of @xmath5 relative to @xmath25 and in terms of the reidemeister torsion @xmath26 of dar @xcite of the singular space @xmath27 obtained from @xmath5 by ` coning - off ' the fibers of the fiber bundle @xmath28 [ thm : dcheegermuller ] let @xmath29 be an odd - dimensional riemannian manifold with fibered cusp ends and an even @xmath11-metric .
let @xmath30 be a unimodular representation meaning that @xmath31 endow the associated flat vector bundle @xmath32 with an even bundle metric @xmath23 and assume that it is ` strongly acyclic at infinity ' in that @xmath33 then the logarithm of the analytic torsion satisfies @xmath34 where @xmath35 is the mapping cylinder of @xmath36 and @xmath37 is an orthonormal basis of @xmath38-harmonic forms with respect to the metrics @xmath10 and @xmath23 inducing a basis for @xmath39 .
the strong acyclicity condition at infinity , which is satisfied in many situations , greatly simplifies the analysis , since then there are no non - zero small eigenvalues forming under the degeneration , the spectrum of the hodge laplacian of the fibered cusp metric is discrete and the corresponding heat kernel is trace class . in the present paper
, we specialize to cusp metrics , which are fibered cusps for which @xmath40 is a point , and we replace the strong acyclicity condition at infinity on @xmath41 with a much weaker ` witt condition ' for which small eigenvalues do appear . with this weaker condition
the hodge laplacian typically has continuous spectrum ( albeit bounded away from @xmath42 ) and the heat kernel is no longer trace class .
+ we can describe exactly how we will manage this extension by recalling the proof of theorem [ thm : dcheegermuller ] in the case where @xmath40 is a point .
let @xmath43 be a smooth closed manifold obtained by doubling @xmath5 across @xmath44 and @xmath45 a flat bundle over @xmath46 we consider a family of metrics @xmath47 that in a tubular neighborhood of @xmath48 has the form @xmath49 this can be visualized as stretching the manifold @xmath43 in the direction normal to the hypersurface @xmath48 until it has two infinite cusp ends in place of the hypersurface .
in fact , while for @xmath50 the metrics @xmath51 are smooth riemannian metrics on @xmath52 as @xmath53 the metric degenerates along @xmath48 and becomes a cusp metric on @xmath54 . in the limit
, the de rham operator @xmath55 associated to @xmath51 has two model operators .
on @xmath56 we obtain @xmath57 the de rham operator of the limiting cusp metric .
the other model operator relates the two sides of @xmath58 and is actually on @xmath59 it is the de rham operator @xmath60 of a metric with cylindrical ends , but twisted by the ` vertical cohomology bundle ' , @xmath61 in @xcite we carried out a careful analysis of the spectrum of @xmath62 as @xmath63 by describing the precise asymptotics of the schwartz kernels of the resolvent and heat kernel .
in particular we proved that there are finitely many eigenvalues of @xmath64 that converge to zero as @xmath65 we call these the _ small eigenvalues _ and denote the product of the non - zero small eigenvalues by @xmath66 ( and the square of this product by @xmath67 ) .
if @xmath68 is polyhomogeneous in @xmath69 the metric @xmath51 is of ` product - type ' and the flat bundle is witt in that @xmath70 then the determinant of the laplacian satisfies @xmath71 the acyclicity condition at infinity that we imposed in the fibered cusp setting implies that the last two terms in do not contribute to the analytic torsion . to remove the acyclicity condition in the cusp setting and
compute the limit of analytic torsion as @xmath53 , we will establish that @xmath72 is polyhomogeneous in @xmath73 and compute its finite part as @xmath53 ( corollary [ poly.1 ] ) , and we will compute the determinant of the model operator @xmath74 ( [ sec : atdb ] , especially ) .
another consequence of removing the acyclicity condition is that the analytic torsion should be thought of not as a number , but as a function that assigns a number to each basis of the cohomology @xmath75 so we will also compute the behavior of a basis of harmonic forms as @xmath53 ( ) .
these pieces together determine the limit of analytic torsion as @xmath65 in theorem [ rt.10 ] we determine how the reidemeister torsion of @xmath43 is related to the reidemeister torsion of @xmath76 all together we establish the following theorem in corollary [ ff.20 ] .
[ thm : introhccheegermuller ] let @xmath77 be an odd dimensional riemannian manifold with @xmath78 an even cusp metric
. let @xmath79 be a flat witt bundle with unimodular holonomy endowed with a bundle metric @xmath23 that extends smoothly to the double of @xmath4 across @xmath22 let @xmath80 and @xmath81 be bases of @xmath82 and @xmath83 respectively , consisting of harmonic forms orthonormal with respect to @xmath23 , @xmath78 and @xmath15 .
the canonical identification gives a basis @xmath84 for @xmath85 using these bases to define the corresponding @xmath3-torsions , we have the following formula : @xmath86 where @xmath87 is the cone over @xmath48 .
if @xmath45 is in fact a flat euclidean vector bundle , then by poincar duality the formula simplifies to @xmath88 there are many representations that satisfy the hypotheses in this theorem .
for instance , if @xmath89 , one can just take the trivial representation . on the other hand , notice that if we require , as in dar s original definition , that the flat vector bundle @xmath45 be defined on the stratified space @xmath27 , then @xmath45 is automatically trivial on @xmath48 and the witt condition can not be achieved when @xmath90 .
however , as we describe in @xcite , the intersection @xmath3-torsion can also be defined for flat vector bundles defined only on the smooth stratum of a stratified space of depth one . with this extra degree of freedom
, one can give many natural examples of unimodular representations satisfying the hypotheses of our theorem , see for instance @xcite for examples on hyperbolic @xmath91-manifolds with cusps .
however , in this case and more generally on odd dimensional hyberbolic manifolds with cusps , notice that as pointed out in ( * ? ? ?
* remark 1.2 ) , if the holonomy representation of @xmath45 is orthogonal or more generally if it becomes a direct sum of irreducible representations when restricted to each connected component of @xmath48 , then by ( * ? ? ?
* remarks 3.5 ( 3 ) of chapter vii ) , the witt condition is satisfied if and only if the strong acyclicity condition at infinity @xmath92 is satisfied , so that theorem [ thm : introhccheegermuller ] reduces to theorem [ thm : dcheegermuller ] in this case .
+ on hyperbolic manifolds with cusps , analytic torsion was first studied by park @xcite , who proved that a relation discovered by fried @xcite between analytic torsion and ruelle zeta functions continues to hold on noncompact hyperbolic spaces . combined with our result
, this gives a description of ruelle zeta functions in terms of intersection @xmath3-torsion .
recently there has been also an impressive sequence of papers by mller and pfaff @xcite , see also @xcite , in which the selberg trace formula is used to great effect in analyzing analytic torsion .
the methods in these papers are closely tied to the algebraic structure of locally symmetric spaces .
calegari and venkatesh @xcite study relationships between the torsion in the homology of arithmetic groups , for certain incommensurable groups , by a careful study of noncompact arithmetic three - manifolds with cusps .
they define a reidemeister torsion as the ` regulator ' of the homology groups divided by the size of torsion in the first homology groups , thus extending the r - torsion of compact arithmetic three - manifolds .
they define the analytic torsion by a renormalized trace of the heat kernel . in (
* theorem 6.8.3 ) they prove a relative cheeger - mller theorem for trivial coefficients comparing a ratio of reidemeister torsions to a ratio of analytic torsions , for two manifolds with isometric cusp structure .
their proof requires a careful study of small eigenvalues , which in their context refers to near - zero eigenvalues on a truncated hyperbolic manifold .
we have heard from venkatesh that one can in principle deduce a formula for the ratio of reidemeister torsion and analytic torsion from the proof of their relative cheeger - mller theorem .
a theorem close to ours in the hyperbolic setting is an interesting cheeger - mller theorem due to pfaff @xcite .
this theorem applies to noncompact hyperbolic manifolds with cusps @xmath4 of odd dimension @xmath93 and flat vector bundles @xmath45 induced by the irreducible representations of @xmath94 or @xmath95 that are not invariant with respect to the cartan involution .
pfaff uses constructions of harder @xcite to define a canonical reidemeister torsion @xmath96 ( similar to that used in @xcite ) .
let @xmath97 be a neighborhood of the cusps .
pfaff uses the renormalized trace of melrose to define analytic torsion and is then able to compute the difference @xmath98 in terms of the rank of @xmath99 the betti numbers and volume of @xmath100 and some weights associated to the holonomy representation of @xmath101 notice that in this setting , the witt condition is never satisfied .
moreover , the bundle metric used by pfaff does not extend to a bundle metric on the double .
finally we mention a preprint of boris vertman @xcite . in his paper
, vertman investigates the cheeger - mller theorem for flat unitary bundles over odd - dimensional manifolds with product - type cusps satisfying the witt condition .
his approach , by gluing methods as in @xcite , is completely different to ours .
+ for hyperbolic surfaces , cusp formation corresponds to converging to the boundary of teichmller space and so has been the subject of much study . for example
, seeley and singer @xcite studied the @xmath102 operator as a cusp is formed .
we can apply our analysis to study this situation as well . in propositions [ lem :
burgersmalleigen ] and [ lem : wolpertdet ] we recover results of wolpert and burger @xcite on the asymptotics of small eigenvalues and the determinant .
+ the paper is organized as follows .
section [ sec : cuspmets ] recalls our conventions for cusp metrics and analytic torsion . in
[ sec : horop ] we analyze the model operator @xmath60 on @xmath103 and compute its contribution to the asymptotics of analytic torsion .
section [ sec : small ] is devoted to the study of the small eigenvalues , including their polyhomogeneity in @xmath73 , and culminates in the computation of the corresponding determinant .
this section also includes an analysis of the asymptotics of an appropriately chosen basis of harmonic forms .
these results are collected in [ sec : at ] and yield the asymptotics of analytic torsion under degeneration to a manifold with cusp ends . in section [ sec : rt ] we show how the r - torsion of the closed manifold @xmath43 relates to the r - torsion of @xmath104 the manifold with cusp ends .
finally , in [ sec : cheegermuller ] , we combine this study with our analysis of analytic torsion to obtain our cheeger - mller theorem . in the last section , [ wolpert ] ,
we specialize to dimension two and explain the relevance of our results to families of hyperbolic metrics approaching the boundary of teichmller space . +
* acknowledgements . *
p. a. was supported by nsf grant dms-1104533 and simons foundation grant # 317883 .
f. r. was supported by a canada research chair , nserc and frqnt .
d. s. was supported by a crm postdoctoral fellowship and by nsf emsw21-rtg 1045119 .
the authors are happy to acknowledge useful conversations with steven boyer , dan burghelea , nathan dunfield , rafe mazzeo , richard melrose , werner mller and jonathan pfaff .
they are also happy to acknowledge the hospitality of the birs workshop geometric scattering theory and applications " held in november 2014 .
in this section , we recall the definition of cusp metrics and a very useful replacement for the tangent bundle that is adapted to the geometry .
we also recall the definition of analytic torsion on closed manifolds and manifolds with ends asymptotic to cusps .
+ on a closed riemannian manifold @xmath105 of dimension @xmath93 , the heat kernel of any laplace - type operator satisfies @xmath106 with @xmath107 hence its zeta function @xmath108 extends from a holomorphic function on @xmath109 to a meromorphic function on all of @xmath110 which has at worst simple poles and is regular at the origin .
if @xmath32 is a flat vector bundle endowed with a fiber metric @xmath23 ( not necessarily compatible with the flat connection ) , and @xmath111 is the hodge laplacian on @xmath45-valued differential forms of degree @xmath112 then @xmath113 is the logarithm of the analytic torsion of @xmath114 if @xmath45 is acyclic and its holonomy is unimodular , that is if @xmath115 then the analytic torsion is independent of the choice of metrics @xmath116 for a general flat bundle with unimodular holonomy , we choose a basis @xmath117 of each @xmath118 and let @xmath119 be an orthonormal basis of harmonic representatives with respect to the metrics @xmath120 then we define @xmath121^{(-1)^q } \right),\ ] ] where @xmath122 = |{\operatorname{det}}w^q|$ ] with @xmath123 the matrix such that @xmath124 it is this quantity that is independent of the choice of metrics . let @xmath125 be a smooth manifold with boundary @xmath126 let @xmath7 be a smooth , non - negative function on @xmath125 that vanishes precisely on @xmath48 and such that @xmath127 does not vanish anywhere on @xmath126 we call such a function a ` boundary defining function ' for @xmath128 or ` bdf ' for short .
we fix a choice of bdf , and our constructions will depend ( mildly ) on this choice .
let us single out a subset of the vector fields on @xmath129 @xmath130 and point out that there is a vector bundle over @xmath125 whose space of sections is @xmath131 we denote this bundle @xmath132 and refer to it as the ` @xmath133-tangent bundle ' of @xmath134 ( the @xmath133 more generally denotes a fibration on the boundary of @xmath135 in our present context the fibration is @xmath136 ) the @xmath133-tangent bundle is isomorphic to the usual tangent bundle of @xmath129 but not in a canonical way .
the dual bundle @xmath137 is called the ` @xmath133-cotangent bundle ' of @xmath134 note that @xmath138 is a section of @xmath139 that is non - degenerate at @xmath140 we can use @xmath7 to rescale the @xmath133-tangent bundle at @xmath48 ( see ( * ? ? ?
* chapter 8) ) , and we refer to the bundle @xmath141 as the @xmath142-tangent bundle or ` hyperbolic cusp tangent bundle ' .
its dual bundle @xmath143 is the @xmath142-cotangent bundle of @xmath129 and we point out that the one form @xmath144 as a section of @xmath145 is non - degenerate at @xmath126 similarly if @xmath146 is a local coordinate on @xmath48 then @xmath147 as a local section of @xmath145 is _ non - vanishing _ at @xmath148
an @xmath142-metric is a bundle metric on the @xmath142-tangent bundle .
the simplest @xmath142-metrics are those that in some collar neighborhood of @xmath48 of the form @xmath149_x\times z$ ] take the form @xmath150 with @xmath15 a metric on @xmath48 independent of @xmath151 we refer to such metrics as * product - type * @xmath142-metrics .
an @xmath142-metric @xmath152 is * product - type to order @xmath153 * if there is a product - type metric @xmath154 such that @xmath155 where @xmath156 denotes the bundle of symmetric bilinear forms on @xmath157 in this paper our results will hold for @xmath142-metrics that are product - type to order @xmath158 + the heat kernel of a laplace - type operator associated to an @xmath142-metric is not as well - behaved as the corresponding object on a closed manifold ( @xcite , @xcite ) .
first , the heat kernel is possibly not trace class .
fortunately it is well - behaved enough that we can make sense of its renormalized trace @xmath159 moreover , from @xcite and the appendix of @xcite , the asymptotics of the renormalized trace of the heat kernel are more complicated as @xmath160 @xmath161 furthermore , one does not always have exponential convergence of @xmath162 to @xmath163 as @xmath164 we will deal with these differences by adding appropriate additional assumptions .
let us say that * a flat bundle @xmath45 is witt * if , upon restricting to @xmath128 we have @xmath165 where @xmath166 if @xmath167 is a hodge laplacian associated to a witt bundle , then we know from ( * ? ? ?
* proposition 7.3 ) that @xmath168 if @xmath152 is product - type to order two and @xmath93 is odd , then @xmath169 again , if @xmath167 is a hodge laplacian associated to a witt bundle , the zeta function @xmath170 is a holomorphic function on @xmath109 that extends to a meromorphic function on @xmath171 with at worst double poles , but regular at the origin .
thus for flat witt bundles we may define analytic torsion for a cusp manifold just as for a closed manifold . if @xmath93 is even , there may be a pole at the origin , but we may still define analytic torsion by taking as a replacement for @xmath172 the coefficient of @xmath173 in the laurent series expansion of @xmath174 at the origin .
we say that a closed riemannian manifold @xmath105 with a two - sided hypersurface @xmath48 is undergoing * cusp degeneration * if the metric is degenerating from a smooth metric to a cusp metric on @xmath175 we will carry out these degenerations in a controlled fashion by studying ` cusp surgery metrics ' .
let us start by performing the ` radial blow - up ' of @xmath176 in @xmath177_{{\varepsilon}}.$ ] recall that this is a smooth manifold with corners , @xmath178_{{\varepsilon } } , z \times \ { 0 \}],\ ] ] obtained by replacing @xmath179 with its inward pointing spherical normal bundle ( see @xcite ) .
figure [ fig : singlespace ] represents the space @xmath180 there is a natural map , known as the blown - down map , @xmath181_{\varepsilon},\ ] ] obtained by collapsing the new boundary hypersurface of @xmath182 back to @xmath183 the manifold @xmath182 has three boundary hypersurfaces
. one , @xmath184 , will not be relevant to our studies and will be cheerfully ignored .
the other two are @xmath185 known as the surgery boundary and denoted @xmath186 and @xmath187 where the @xmath93 in the subscript recalls that this is where most of @xmath188 ended up . given any blow - down map , the `
interior lift of a set ' is equal to the closure of the lift of that set minus the set being blown - up ; thus @xmath189 is the interior lift of @xmath190 which we denote @xmath191 there is a natural choice of boundary defining function for @xmath186 which we fix once and for all : @xmath192 when there is no possibility of confusion , we will denote this simply as @xmath193 the interior of @xmath194 can be identified with the normal bundle to @xmath48 in @xmath43 ; @xmath194 corresponds to its fiberwise compactification .
the normal bundle to @xmath48 is trivial by assumption , and so we have @xmath195.\ ] ] ( of course any closed interval would serve , but our usual choice of coordinates will correspond to @xmath196,$ ] so we use this interval throughout . ) we endow @xmath194 with a trivial fibration @xmath197.\ ] ] analogously to the @xmath142-tangent bundle , we will define a ` cusp surgery tangent bundle ' or @xmath198-bundle .
first let @xmath199_{{\varepsilon}}$ ] be the composition of @xmath200 with the obvious projection and define @xmath201 next let @xmath202 and define @xmath203 so that @xmath204 is its space of sections . finally , let @xmath205 and let @xmath206 denote the dual bundle .
the one - forms @xmath207 where @xmath146 denotes a coordinate along @xmath128 lift from the interior of @xmath182 to a spanning set of sections of @xmath208 note that , as sections of @xmath209 these do not degenerate at @xmath210 a cusp surgery metric is a bundle metric on @xmath211 we say that * an @xmath198-metric is of product type * if there is a tubular neighborhood @xmath212_x \times z \subseteq m$ ] around @xmath48 in which the metric takes the form @xmath213 where @xmath15 is a metric on @xmath48 that is independent of both @xmath7 and @xmath214 we say that * an @xmath198-metric @xmath51 is of product type to order @xmath153 * if @xmath215 for some product type metric @xmath216 where @xmath217 denotes the bundle of symmetric two - tensors on @xmath208 + let @xmath218 be a flat vector bundle endowed with a bundle metric @xmath20 not necessarily compatible with the flat connection , and let @xmath219 be the corresponding de rham operator .
we will consider this as an operator on the bundle @xmath220 one of the advantages of using the @xmath198-cotangent bundle , as opposed to the usual cotangent bundle of @xmath221 is that the leading order behavior of @xmath222 will be described by tractable model operators , discussed below .
we are interested in the action of @xmath222 as an unbounded operator on @xmath223 the natural @xmath38 space associated to an @xmath198-metric @xmath51 and the bundle metric on @xmath101 however , for some constructions it will be easier to work with @xmath224 where @xmath166 thus our main object of interest is the operator @xmath225 acting as an unbounded operator on @xmath226 + if @xmath51 is of product - type to order two , then we have simple expressions for the model operators of @xmath227 first let us write @xmath228 this splitting distinguishes between forms with a @xmath127 and forms without a @xmath127 . with respect to this splitting , a direct computation tells us that @xmath229 is given near @xmath230 by @xmath231 up to higher order terms in @xmath232 as @xmath198-differential operators . here
@xmath233 is the number operator on @xmath48 that multiplies a differential form by its degree .
the first model operator , known as the * vertical operator * , is @xmath234 where @xmath235 is a section of @xmath236 on @xmath194 and @xmath237 is any smooth extension of @xmath235 to @xmath182 .
its null space forms a vector bundle over @xmath194 which is just the space of scaled harmonic forms on @xmath48 , thought of as a trivial vector bundle over @xmath196 $ ] and then pulled - back along @xmath238 we will denote this bundle by @xmath239 the second model operator , known as the * horizontal operator * , is defined by @xmath240 where @xmath241 denotes the projection onto @xmath242 , @xmath235 is a section of @xmath242 and @xmath237 is any choice of extension off @xmath230 . in terms of , the operator @xmath60 is given by @xmath243 where @xmath241 denotes the projection onto @xmath48-harmonic forms . in projective coordinates near @xmath186 @xmath244 in which @xmath245 we let @xmath246 so that @xmath247 as an operator acting on @xmath248 where the restriction of @xmath249 to @xmath230 is well defined as a section of @xmath250 . thus , @xmath60 is a @xmath251-operator in the sense of melrose @xcite .
if @xmath45 is a witt bundle , then @xmath60 is fredholm ( * ? ? ?
* lemma 2.1 ) .
finally , @xmath229 induces an operator on @xmath252 this face is the manifold with boundary @xmath253 $ ] and @xmath254 is the twisted de rham operator corresponding to the @xmath142-metric @xmath255 and the flat bundle @xmath256
let @xmath43 be a closed manifold of dimension @xmath257 @xmath48 a two - sided hypersurface with fixed boundary defining function @xmath7 and @xmath51 a cusp surgery metric , product - type to second order .
if @xmath218 is a flat vector bundle of witt type , then we have seen that there is a model @xmath251-operator @xmath258 acting on @xmath259 in this section we study this operator and its contribution to the asymptotics of analytic torsion .
+ first let us compute its null space , and that of its square .
note that if @xmath260 then the projections of @xmath261 and @xmath262 onto the spaces of forms of fixed _ vertical _ degree @xmath263 ( that is , having degree @xmath263 in @xmath48 ) , which we denote @xmath264 and @xmath265 respectively , are also in the null space of @xmath266 now since @xmath267 we see that @xmath268 and that @xmath269 thus we have found @xmath270 we are interested in @xmath60 as an unbounded operator on @xmath271 i.e. , with the measure @xmath272 on @xmath59 with respect to this measure , @xmath273 is in @xmath38 iff @xmath274 and hence the @xmath38 kernel of @xmath60 is @xmath275 next consider @xmath276 and note that , for @xmath277 @xmath278 notice that as @xmath279 , @xmath280 is of order @xmath281 for @xmath282 , while for @xmath283 , @xmath284 , so @xmath285 is never in @xmath38 with respect to the density @xmath286 hence the null space of @xmath74 acting on smooth sections of @xmath287 is @xmath288 since the function @xmath289 is never in @xmath290 , @xmath291 which could also have been deduced from the formal self - adjointness of @xmath60 .
let us emphasize in particular that @xmath74 has no @xmath38-kernel on forms of _ total _ degree ( i.e. , degree in @xmath272 plus degree in @xmath48 ) equal to zero or @xmath292 let @xmath293 where @xmath294 .
each of these is a laplace - type operator on @xmath295 and we denote the corresponding zeta function by @xmath296 from ( * ? ?
* theorem 10.2 ) , we know that the contribution of the horizontal operator @xmath60 to the asymptotics of analytic torsion is through @xmath297 in this subsection we will compute this contribution .
+ from the previous subsection we see that the heat kernel of @xmath60 satisfies @xmath298 we introduce the abbreviation @xmath299 and note that @xmath300 where @xmath301 with the convention that @xmath302 so we can write @xmath303 and hence is equal to @xmath304.\ ] ] it thus suffices to compute the determinant of @xmath305 on @xmath103 ( endowed with the metric @xmath306 and bdf @xmath307 ) , @xmath308 our strategy will be to compute the variation in @xmath309 of the renormalized trace ( see ) and use this to compute the determinant ( see ) . once we have computed these one - dimensional determinants , we return to in
. + let us start with the two cases we can compute directly .
[ lem:2cases ] when @xmath310 we have @xmath311 when @xmath312 we have @xmath313 in the coordinate @xmath314 , @xmath315 hence @xmath316 is the euclidean laplacian and @xmath317 is the euclidean laplacian plus one . in the former case
the restriction to the diagonal is @xmath318 and in the latter @xmath319 as these are independent of @xmath320 the renormalized trace is computed by multiplying these expressions by the renormalized volume .
the renormalized volume is , for our choice of measure @xmath272 and of boundary defining function @xmath321 @xmath322 this proves that @xmath323 .
it is easy to see that the renormalized mellin transform over @xmath324 of a power of @xmath325 is equal to zero ; indeed , let us define , for any function @xmath326 with an asymptotic expansion in @xmath325 as @xmath327 and @xmath328 as @xmath329 @xmath330 each of these extends to a meromorphic function on @xmath171 which we denote by the same symbol , and the renormalized mellin transform of @xmath261 is @xmath331 if @xmath332 , then @xmath333 and @xmath334 and so @xmath335 hence we see that the zeta function of @xmath316 is identically zero . for @xmath336
we have , for any @xmath337 @xmath338 hence @xmath339 next let us compute the variation of the renormalized trace for arbitrary @xmath340 first note that @xmath341\ ] ] where @xmath41 denotes @xmath342 similarly @xmath343 , \quad { \partial}_a ( -p(-a)p(a ) ) = -\alpha p(-a)p(a ) + 2p(-a)\alpha p(a ) - p(-a)p(a)\alpha.\ ] ] next , by duhamel s formula , we have @xmath344 where @xmath345 } \ ; d\tau .
\end{gathered}\ ] ] note that since the _ renormalized _ trace does not vanish on commutators , @xmath346 is not automatically zero .
let us focus first on @xmath347 we can rewrite it as @xmath348 } - 2 { \left [ p(a ) , e^{tp(-a)p(a)}p(-a)\alpha \right ] } \right).\end{gathered}\ ] ] in turn let us write this as @xmath349 where @xmath350 consists of the summands involving commutators . from the uniqueness of the solution to the heat equation , we note that @xmath351 and so we can write @xmath352 as @xmath353 which we can rewrite as @xmath354 where @xmath355 thus we have @xmath356 where @xmath346 and @xmath350 involve renormalized traces of commutators . + to address these terms
we will make use of appropriate trace defect formul .
we will use the same conventions as in @xcite regarding mellin transform and indicial operators , namely : @xmath357}_{{\partial}}. \end{gathered}\ ] ] the latter means that @xmath358 acts on a section @xmath235 over the boundary by choosing an extension @xmath359 off of the boundary , applying @xmath360 to @xmath361 and then restricting back to the boundary .
the result is independent of the choice of extension .
[ lem : renormcommut ] on any manifold with boundary @xmath43 with a fixed choice of bdf @xmath362 * ( * ? ? ?
* lemma 5.10 ) if @xmath363 is a @xmath251-pseudodifferential operator and @xmath364 is a smoothing @xmath251-pseudodifferential operator , then @xmath365 ) = \frac i{2\pi}\int_{{\mathbb{r } } } { \operatorname{tr}}_{{\partial}}{\left ( { \partial}_{\lambda}i(a,\lambda ) i(b,\lambda ) \right)}\ ; d\lambda,\ ] ] where the trace in the integrand is the trace @xmath366 * with @xmath363 and @xmath364 as above , @xmath367 ) = -\frac1{4\pi } \int_{{\mathbb{r } } } { \operatorname{tr}}_{{\partial } } ( { \partial}_{\lambda}^2i(a,\lambda)i(b,\lambda ) ) \ ; d\lambda.\ ] ] following @xcite ( cf .
@xcite ) , let us use riesz renormalization to define the renormalized trace , @xmath368 for any operator @xmath364 such that @xmath369 is trace - class for large enough @xmath370 we have @xmath371 ) = \operatorname*{\operatorname{fp}}_{z=0 } { \operatorname{tr}}(\varrho^z[a , b ] ) = \operatorname*{\operatorname{fp}}_{z=0 } { \operatorname{tr}}([\varrho^z , a]b ) = \operatorname*{\operatorname{fp}}_{z=0 } { \operatorname{tr}}(z\varrho^z { \widetilde{a}}(z ) b ) = \operatorname*{\operatorname{res}}_{z=0 } { \operatorname{tr}}(\varrho^z { \widetilde{a}}(z ) b)\ ] ] where @xmath372 note that this is a holomorphic function of @xmath146 and has indicial operator @xmath373 with taylor expansion at @xmath374 given by @xmath375 in particular , we point out that @xmath376,\lambda).\ ] ] to compute the residue @xmath377 note that @xmath378 so only the term with @xmath379 contributes to the residue at @xmath380 hence @xmath381 extends to a meromorphic function with a simple pole at @xmath374 and residue equal to @xmath382 as required .
now replace @xmath364 with @xmath383 .
as before , since @xmath384 only the terms with @xmath379 contribute to the residue .
furthermore , for small @xmath146 we have @xmath385 and so @xmath386 we see that to compute the residue we will only need ( minus one times ) the @xmath387 term in @xmath388 since the @xmath387 term in the expansion of @xmath389 is @xmath390 we find @xmath391 to compute the indicial operator of @xmath392 let us recall that our bdf is @xmath393 so that @xmath394 and hence @xmath395 it follows that @xmath396 and @xmath397 at both ends of @xmath59 we can use this observation and lemma [ lem : renormcommut ] to compute @xmath346 and @xmath398 indeed , note from and the fact that @xmath399 that @xmath400p(a ) + p(-a)[p(a),\log\varrho ] ) , \lambda ) \\ = -(\tfrac1i { \partial}_{\lambda}i(p(-a),\lambda))i(p(a),\lambda ) + p(-a)(\tfrac1i{\partial}_{\lambda}i(p(a),\lambda ) ) = 2a \end{gathered}\ ] ] ( the same at both ends of @xmath103 ) .
hence @xmath401 } \\ = -2 \int_0^t \frac i{2\pi}\int_{{\mathbb{r}}}{\partial}_{\lambda}{\left ( e^{-\tau(a^2+\lambda^2)}2a \right)}e^{-(t-\tau)(a^2+\lambda^2)}\ ; d\lambda d\tau \end{gathered}\ ] ] where we multiply by two in applying the trace defect formula since we have the same contribution from each end of @xmath103 .
since the integrand is odd in @xmath402 , we see that @xmath403 .
next for @xmath350 we have ( again multiplying by two to take into account both ends of @xmath103 ) @xmath404 } - 2 { \left [ p(a ) , e^{tp(-a)p(a)}p(-a)\alpha \right ] } \right ) \\ = \frac t{2\pi } \int_{{\mathbb{r } } } { \left ( -({\partial}_{\lambda}^2i(p(-a)p(a),\lambda))e^{-t(a^2+\lambda^2 ) } + 2 ( { \partial}_{\lambda}^2i(p(a),\lambda))e^{-t(a^2+\lambda^2)}i(p(-a),\lambda ) \right ) } \ ; d\lambda \\ = \frac t{\pi } \int_{{\mathbb{r } } } e^{-t(a^2+\lambda^2 ) } \ ; d\lambda = \sqrt{\frac t\pi}e^{-ta^2},\end{gathered}\ ] ] and so altogether @xmath405 @xmath406 + [ lem : det1 ] when @xmath407 we have @xmath408 note that since @xmath409 is an odd function of @xmath309 for each fixed @xmath410 as are its derivatives in @xmath410 we have @xmath411 since @xmath412 integrating from @xmath42 to @xmath309 yields @xmath413 hence from lemma [ lem:2cases ] we have @xmath414 consider this integral as a function of @xmath415 writing it alternately as @xmath416 we see that it is @xmath417 as @xmath418 and @xmath419 as @xmath420 it follows that the integral @xmath421 exists for all @xmath173 with real part in @xmath422 for these @xmath173 we can use fubini s theorem to find @xmath423 we also know that @xmath424 so altogether it follows that the zeta function of @xmath425 is equal to @xmath426 and hence @xmath427 next , using the notation of , we point out that @xmath428 since this interchange is justified in the particular regions of @xmath429 where these functions are holomorphic .
hence , for @xmath430 writing @xmath431 , we have @xmath432 we now examine @xmath433 and @xmath434 in a neighborhood of @xmath435 for @xmath433 , note that @xmath436 and hence , @xmath437 for @xmath438 we start by integrating by parts to find @xmath439 indeed , the integration by parts is justified for each of @xmath440 and @xmath441 in the region where it is holomorphic , and the resulting boundary terms cancel out when we add together the meromorphically continued @xmath440 and @xmath442 as above , @xmath443 extends meromorphically from @xmath444 to the complex plane with simple poles at a subset of @xmath445 for @xmath430 @xmath446 extends meromorphically from @xmath447 to the complex plane with a single , simple pole at @xmath448 and residue @xmath449 where @xmath450 is the orthogonal projection onto the @xmath38-null space of @xmath451 we will need a more explicit formula for this residue .
we have shown that @xmath452 let us write @xmath453 so , for @xmath454 the schwartz kernel of the projection onto @xmath455 is given by @xmath456 and hence @xmath457 thus , for @xmath430 the residue equals @xmath458 which determines the behavior of @xmath434 near @xmath435 indeed , since @xmath459 and @xmath460 we have @xmath461 thus altogether we have @xmath462 and , interchanging @xmath463 and @xmath464 , this shows that @xmath465 hence there exist constants @xmath466 such that @xmath467 note that @xmath468 , so from lemmas [ lem:2cases ] and [ lem : det1 ] we have @xmath469 which shows that @xmath470 and @xmath471 thus altogether we have shown that @xmath472 @xmath406 + finally , let us compute the contribution to the analytic torsion we have been looking for . using the computation of the determinant , we have for all @xmath473 @xmath474 \\ = -q{\left ( \log c_{|v/2-q| } - { \operatorname{sign}}(2q - v ) \log
|v-2q| \right ) } + ( q+1 ) { \left ( \log c_{|v/2-q| } - { \operatorname{sign}}(v-2q ) \log
|v-2q| \right ) } \\ = \log c_{|v/2-q| } + ( 2q+1 ) { \operatorname{sign}}(2q - v ) \log |v-2q|.\end{gathered}\ ] ] substituting this into , we find @xmath475 , d_b , { \mathcal{h}}^*(z;f))= \\
\frac12\sum_{\substack{0\leq q\leq v \\ q\neq v/2 } } ( -1)^qb_q { \left ( \log c_{|v/2-q| } + ( 2q+1 ) { \operatorname{sign}}(2q - v ) \log |v-2q| \right)}.\end{gathered}\ ] ] if we assume that the metric @xmath23 is compatible with the flat connection on @xmath45 ( which implies orthogonal holonomy ) , then we can use poincar duality to rewrite as a sum over @xmath476 : @xmath477
let @xmath51 be an @xmath198-metric , product - type to order two , and @xmath478 a flat vector bundle with bundle metric @xmath20 not necessarily compatible with the flat connection .
in @xcite we showed that , as long as @xmath479 is witt , there exists @xmath480 and @xmath481 such that @xmath482 and such that every eigenvalue in @xmath483 for sufficiently small @xmath73 converges to zero as @xmath65 we call the eigenvalues of @xmath229 in @xmath483 the * small eigenvalues * and the sum of their eigenspaces the * small eigenforms * , @xmath484 note that the space of harmonic forms @xmath485 is a subspace of @xmath486 .
write @xmath487 , @xmath488 , so that @xmath489 and @xmath490 , where the @xmath73 subscripts are used to remind us that all of the operators except @xmath11 depend on @xmath73 .
note that the small eigenvalues of @xmath229 and of @xmath222 are the same , and the eigenforms differ by a factor of @xmath491 , so we may usually speak without ambiguity . as we will see in corollary [ poly.1 ] below , the product of the positive eigenvalues in a given degree is polyhomogeneous in @xmath73 .
the quantity @xmath492 is thus well - defined , where @xmath111 denotes the action of @xmath493 on forms of degree @xmath494 in this section we will compute the contribution to the analytic torsion coming from these small eigenvalues , which by ( * ? ? ?
* theorem 10.2 ) is given by @xmath495 from @xcite , the dimension of the space of small eigenforms is equal to @xmath496 we can express these dimensions directly in terms of the topology of @xmath52 @xmath128 and @xmath497.$ ] let us write @xmath498 for the singular space obtained from @xmath499 by coning off @xmath500 , which is two copies of @xmath48 , so that we have @xmath501 it follows from work of hausel , hunsicker , mazzeo @xcite and lemma 8.20 of @xcite that @xmath502 where @xmath503 denotes the @xmath504 @xmath38-cohomology group of @xmath505 with coefficients in @xmath101 on the other hand , from the computation of the @xmath38-null space of the horizontal model operator in [ sec : nulldb ] , we see that for @xmath506 @xmath507 for positive @xmath69 a subspace of dimension @xmath508 of these eigenforms will correspond to the eigenvalue zero , and the rest will correspond to positive small eigenvalues .
this suggests that , to understand the small eigenforms corresponding to non - zero small eigenvalues , we look for a long exact sequence linking @xmath509 @xmath510 and @xmath511 observe first that @xmath43 is homeomorphic to the union of @xmath499 with @xmath512 , with an overlap region homotopic to @xmath513 .
the associated mayer - vietoris sequence is @xmath514 where the map @xmath515 is given by @xmath516 and @xmath517 is the natural inclusion .
note that @xmath515 is injective when restricted to @xmath518 . by using the identification
@xmath519 & \mapsto & \lambda_1-\lambda_2 , \end{array}\ ] ] we obtain from a new long exact sequence @xmath520 with @xmath521 .
moreover , for @xmath522 , we can replace @xmath523 with @xmath524 replacing singular cohomology with hodge cohomology ( notationally by replacing @xmath525 with @xmath526 ) endows these spaces with inner products , so for @xmath522 , we have short exact sequences @xmath527 and hence the dimension of the space of eigenforms corresponding to positive small eigenvalues is equal to @xmath528 this in turn suggests that we relate the vector spaces @xmath529 to the space of small eigenforms with positive eigenvalue , and that we relate their respective orthocomplements , which we denote @xmath530 and @xmath531 , to the space of harmonic forms on @xmath43 . indeed ,
as the notation suggests , we will see in the next subsection that these vector spaces are restrictions of the subspaces of small eigenforms of @xmath493 with positive eigenvalue , or with zero eigenvalue , to the two boundary faces of @xmath182 . finally ,
from , we deduce the decomposition @xmath532 to obtain a similar decomposition for @xmath533 it is convenient to use poincar duality , which exchanges @xmath45 with the dual flat bundle @xmath534 to which the above reasoning applies equally well .
so let us consider the dual of the long exact sequence for @xmath535 notice that this sequence could have alternately been obtained by looking at the long exact sequence associated to the pair @xmath536 and using the thom isomorphism @xmath537 along with the identification @xmath538 for @xmath539 this means that the long exact sequence dual to for @xmath534 can be obtained by looking at the long exact sequence associated to the pair @xmath540 : @xmath541 & { \mathrm{h}}^q_c(m_0;f ) \ar[r ] & { \mathrm{h}}^q(m;f ) \ar[r ] & { \mathrm{h}}^q(z;f ) \ar[r ] & { \mathrm{h}}^{q+1}_c(m_0;f ) \ar[r ] & \cdots . } \label{se.3b}\ ] ] by using the identification @xmath542 for @xmath533 , we get @xmath541 { \mathrm{h}}^q_{(2)}(m_0;f ) \ar[r]^-{{\widehat{i}}_q } & { \mathrm{h}}^q(m;f ) \ar[r]^-{{\widehat{{\partial}}}_q } & { \mathrm{h}}^q(z;f ) \ar[r]^-{{\widehat{j}}_q } & { \mathrm{h}}^{q+1}_{(2)}(m_0;f ) \ar[r ] & \cdots } \label{se.3c}\ ] ] where @xmath543 @xmath544 and @xmath545 are poincar duals of the maps @xmath546 @xmath547 and @xmath548 in for @xmath535 using the hodge @xmath549-operators of @xmath20 @xmath15 and @xmath550 we can define for @xmath551 the spaces of harmonic forms @xmath552 so that for @xmath551 we have the decompositions @xmath553 this last decomposition is of course related to the long exact sequence via the natural identifications @xmath554
if @xmath93 is even , then we also have to discuss the case @xmath555 where we have the decompositions @xmath556 indeed , using the natural identification @xmath557,$ ] this can be checked directly using the commutative diagram @xmath558 & { \mathrm{h}}^{\frac{m}2}_c(m_0;f ) \ar[r]\ar[d ] & { \mathrm{h}}^{\frac{m}2}(m;f ) \ar[r]^-{{\widehat{{\partial}}}_{\frac{m}2}}\ar[d ] & { \mathrm{h}}^{\frac{m}2}(z;f ) \\ { \mathrm{h}}^{\frac{m}2 - 1}(z;f ) \ar[r]^-{{\partial}_{\frac{m}2 - 1 } } & { \mathrm{h}}^{\frac{m}2}(m;f ) \ar[r ] & { \mathrm{h}}^{\frac{m}2}(m_0;f ) \ar[r ] & { \mathrm{h}}^{\frac{m}2}(z;f ) , } \ ] ] where the top and bottom rows come from the long exact sequences and .
alternatively , we could proceed more analytically and deduce the decompositions from theorem [ se.11 ] below . as discussed in the previous subsection
, the surgery long exact sequences and suggest that the vector spaces @xmath559 and @xmath560 correspond to restrictions to the boundary faces of @xmath182 of the eigenforms corresponding to positive small eigenvalues . in this subsection
, we will both see that this is indeed the case and compute the rate at which these eigenvalues approach zero as @xmath63 .
we first make the following simple observation .
if @xmath561 is a positive small eigenvalue with eigenform @xmath562 @xmath563 then @xmath564 and @xmath565 are also eigenforms with the same small eigenvalue @xmath566 furthermore , at least one of these two eigenforms is non - zero .
[ se.7 ] the first statement is immediate since both @xmath567 and @xmath568 commute with @xmath229 . to see that @xmath569 and @xmath565 can not be both zero
, it suffices to notice that for @xmath570 @xmath571 the key step for the computation of the decay rates of the small eigenvalues is the following lemma .
suppose @xmath572 is a section of @xmath573 , with @xmath574 , such that * @xmath575 and @xmath576 ; * @xmath572 is polyhomogeneous on @xmath182 ; * @xmath577 and @xmath578 is such that @xmath579 ; * @xmath580 ; * @xmath581)\ne 0,$ ] where @xmath582\in { \mathrm{h}}^q_{(2)}(m_0 ; f)$ ] is the cohomology class associated to @xmath583 then @xmath584)||_{l^2}^2{\varepsilon}^{m-1 - 2q } + o({\varepsilon}^{m-1 - 2q}),\ ] ] where @xmath585 is the constant from and @xmath586 is understood as a map into the harmonic forms on @xmath48 ( identified with the cohomology by the de rham isomorphism ) . moreover , the @xmath587-form @xmath588 is well - defined and * @xmath589 and @xmath590 ; * @xmath591 is polyhomogeneous on @xmath182 ; * @xmath592 and @xmath593 is such that @xmath594 ; * @xmath595 ; * @xmath596 is a non - zero multiple of @xmath581 ) \in { \mathcal{h}}^q_{+}(z;f)$ ] .
in particular , if @xmath572 is an eigenform associated to a small eigenvalue @xmath597 then @xmath598)||_{l^2}^2{\varepsilon}^{m-1 - 2q } + o({\varepsilon}^{m-1 - 2q}),\ ] ] and @xmath591 is also an eigenform for the small eigenvalue @xmath566 [ se.8 ] since @xmath599)$ ] is always 0 by the witt condition , the theorem statement is empty for @xmath600 , so we may assume that @xmath601 with this understood , let @xmath602 be the restriction of @xmath572 to @xmath603 . since @xmath604 , we know from @xcite that @xmath602 is in the @xmath38-kernel of @xmath605 this is helpful in describing the expansion of @xmath602 near @xmath606 which has two components , which we write as @xmath607 and @xmath608 near @xmath609 using the splitting into tangential and normal parts , we see from that @xmath610 indeed , clearly , the coefficients @xmath611 and @xmath612 must be in @xmath613 and @xmath614 for @xmath602 to be in the kernel of @xmath615
. then the powers @xmath153 and @xmath616 are obtained by solving the equation @xmath617 which gives @xmath618 @xmath619 .
since @xmath620 and @xmath602 is in @xmath621 this means @xmath622 , so that in fact we have that @xmath623 now , since @xmath624=\widetilde{u}_+-\widetilde{u}_-$ ] with @xmath625 and @xmath624\neq 0,$ ] we must have @xmath626 so at least one must be nonzero .
therefore the expansion of @xmath572 at @xmath230 must have a nonzero term of order @xmath627 moreover , since @xmath626 the coefficient of this term can not be of the form and hence can not be in the @xmath38-kernel of @xmath628 . a priori
, @xmath572 could have lower order terms at @xmath230 .
notice however that @xmath572 can not have terms of order less than @xmath629 at @xmath230 that are in the kernel of @xmath630 indeed , if there were such a term of order @xmath631 with @xmath632 or with @xmath633 and @xmath634 the coefficient @xmath635 of this term would be in the kernel of @xmath636 so of the form @xmath637 let @xmath638 ; it is a boundary defining function for @xmath639 since @xmath640 near @xmath641 @xmath642 would be of order @xmath643 since @xmath572 is bounded , this forces @xmath644 .
recall that the exterior differential @xmath11 does not depend on @xmath69 and that @xmath645 therefore , to understand @xmath646 we must examine the first term in the expansion of @xmath572 at @xmath603 ( resp .
@xmath230 ) which is not in the kernel of @xmath647 ( resp .
@xmath648 ) ; the discussion above indicates that such a term exists .
let us denote it by @xmath649 and suppose for contradiction that @xmath649 occurs at order @xmath650 with either @xmath651 or @xmath633 and @xmath634 and with @xmath652 minimal .
then consider @xmath653 ; it is polyhomogeneous on @xmath182 with a nontrivial term of order either @xmath654 or @xmath631 ( the loss of an order happens if and only if if the term is at @xmath230 and is not a section of @xmath655 ) .
let @xmath656 be @xmath657 divided by its leading - order coefficient in @xmath73 , so that @xmath656 has expansions at @xmath230 and @xmath603 with restriction @xmath658 to @xmath603 and/or restriction @xmath659 to @xmath230 being nontrivial . since @xmath660 @xmath661 is also in the range of @xmath662 and so @xmath658 and @xmath659 are in @xmath663 and @xmath664 respectively .
in particular , we immediately see that @xmath659 must be a section of @xmath665 suppose that @xmath649 is at @xmath230 but is not a section of the @xmath666 then the hodge decomposition on @xmath48 would imply that @xmath659 is not a section of @xmath667 which is a contradiction .
so @xmath649 must either be at @xmath603 or a section of @xmath655 at @xmath668 next suppose it is at @xmath603 at order @xmath652 ; then by the hodge decomposition on @xmath669 the coefficient of @xmath649 at @xmath603 can not be in @xmath670 so must have a term of order zero or smaller at @xmath671 , which would contradict our assumption that @xmath572 is bounded with @xmath672 .
thus @xmath673 , @xmath674 , and @xmath649 occurs at @xmath230 and is a section of @xmath655 .
furthermore , the same logic as in the previous paragraph tells us that no term at @xmath230 within one order of @xmath649 , inclusive , is not a section of @xmath655 ; if it were , @xmath659 would have a contribution from @xmath567 applied to that term and would not be a section of @xmath655 either .
we conclude that @xmath649 occurs at @xmath230 , is equal to @xmath675 , and that @xmath676 . since @xmath677 and @xmath678
, we must have that @xmath679 but is not contained in @xmath680 . by examining ,
we see that there must be a nontrivial term of order @xmath681 in the expansion of the first component of @xmath635 at the boundary of @xmath668 since @xmath682 , this means that @xmath683 is of order @xmath684 at @xmath685 .
however , unless @xmath686 and @xmath687 if @xmath688 this would contradict the assumption that @xmath572 is bounded at @xmath639 and since there is in fact a nontrivial term of order @xmath689 at @xmath690 we must in fact have @xmath691 and @xmath692 thus let @xmath693 be the coefficient of the term of order @xmath694 of @xmath572 at @xmath668 by the argument above , @xmath695 must be an element of the form . to be consistent with
, this means that @xmath696 with @xmath697 and where @xmath698 now a simple computation using the explicit form of @xmath648 shows that the leading order term @xmath659 is precisely the vector with zero in the first factor and @xmath699 in the second factor .
squaring and then integrating with respect to b - surgery densities , this finally gives @xmath700 from there , the properties of @xmath591 are easily obtained . the only slightly tricky part is to show that @xmath591 is polyhomogeneous .
however , we may write @xmath701 since @xmath702 is bounded and has limit a positive constant as @xmath73 approaches zero , it follows from a direct series expansion construction that its inverse is also polyhomogeneous in @xmath73 with limit a positive constant as @xmath73 approaches zero . since the product of a polyhomogeneous function of @xmath73 and a polyhomogeneous function on @xmath182 is polyhomogeneous on @xmath221 the result follows .
we can now consider the projections .
the following theorem shows that their leading - order behavior at @xmath603 and @xmath230 is what we expect . in the statement of the result
, we will tacilty make the following natural identification following from , @xmath703 recall from @xcite that for some index family @xmath704 , @xmath705 is the projection onto eigenforms of @xmath229 corresponding to small eigenvalues , where @xmath706 and where @xmath707 is the space of operators whose kernels are smooth and polyhomogeneous on @xmath708 with index family @xmath709
. then let @xmath710 be the subspace of @xmath711 consisting of forms of pure degree @xmath712 .
we have : * @xmath713 where @xmath714 are projections onto the harmonic forms and onto the eigenforms associated to positive small eigenvalues respectively . * each of the projections @xmath715 and @xmath716 can itself be written as a sum of two projections in @xmath717 @xmath718 where @xmath719 restricts to the projection onto @xmath720 on @xmath603 and restricts to zero on @xmath690 while @xmath721 vanishes on @xmath603 and on @xmath230 restricts to the projection onto @xmath722 for @xmath723 and onto @xmath724 for @xmath725 similarly , @xmath726 restricts to the projection onto @xmath727 on @xmath603 and vanishes on @xmath690 while @xmath728 vanishes on @xmath603 and on @xmath230 restricts to the projection onto @xmath729 for @xmath730 and onto @xmath731 for @xmath725 furthermore , the image of each of these projections admits a basis of polyhomogeneous forms on @xmath180 * for @xmath732 @xmath733 and the maps @xmath734 are isomorphisms .
* for @xmath735 @xmath736 and the maps @xmath737 are isomorphisms . * for @xmath738
when @xmath93 is even , @xmath739 [ se.11 ] in degree @xmath740 @xmath741 is just the projection onto @xmath742 , which is @xmath743 , where @xmath744 is the space of flat sections of @xmath101 since @xmath744 does not depend on @xmath69 we clearly see that @xmath745 and that its range admits a basis of polyhomogeneous sections of @xmath45 on @xmath180 since @xmath746 is trivial in degree zero , @xmath747 and @xmath748 thus , we can take @xmath749 ; this projection is polyhomogeneous on @xmath750 and its image restricts to the image of the projection onto @xmath751 at @xmath603 and to @xmath42 at @xmath668 we claim the image has a basis which is polyhomogeneous on @xmath182 and restricts to a basis of harmonic forms corresponding to @xmath751 at @xmath603 and to 0 at @xmath668 indeed , the argument is standard and proceeds as follows .
take a basis of harmonic forms corresponding to @xmath751 at @xmath752 ; each is polyhomogeneous on @xmath641 so we can extend each basis element to a polyhomogeneous form on @xmath180 then applying the projection @xmath753 to each element yields a polyhomogeneous basis for @xmath73 small .
( this argument to go from polyhomogeneity of the projection to polyhomogeneity of a basis works for any of the projections involved in this proof , as all are polyhomogeneous on @xmath750 and have restrictions at @xmath754 and @xmath755 to projections which have polyhomogeneous bases on @xmath603 and @xmath230 respectively . ) by lemma [ se.8 ] , we see that for @xmath572 and @xmath591 in the range of @xmath756 @xmath757 since @xmath229 and @xmath711 commute , this means that @xmath758 from here , we now proceed inductively to prove the rest of the theorem for @xmath522 .
suppose it is true for degree @xmath759 ; we must show it is true for degree @xmath712 .
first , we take @xmath760 to be the projection onto the range of @xmath761 .
we claim that it has all the required properties .
indeed , it is clear that @xmath762 . if @xmath763 is a polyhomogeneous basis of @xmath764 then by lemma [ se.8 ] , @xmath765 is a polyhomogeneous basis of @xmath766 .
next we construct @xmath767 if @xmath768 then @xmath769 if not , let @xmath770 be a non - zero element and let @xmath771 be the corresponding element in @xmath772 let @xmath773 be a cutoff function taking values between @xmath42 and @xmath774 , with @xmath775 for @xmath776 and @xmath777 for @xmath778 where @xmath480 is chosen small enough , and consider @xmath779 . by direct computation , @xmath780 @xmath781 and @xmath565 is polyhomogeneous and bounded on @xmath221 vanishing at both @xmath603 and @xmath668 first set @xmath782 since @xmath783 we still have that @xmath784 by the properties of @xmath785 we also have that @xmath786 and @xmath787 by construction , @xmath788 which implies by the inductive step that @xmath789 however , by the hodge decomposition @xmath790 is orthogonal to the harmonic forms , and since @xmath791 the inductive hypothesis implies that @xmath790 is also orthogonal to the image of @xmath792 therefore @xmath793 finally , note that @xmath794 is also polyhomogeneous and bounded on @xmath221 vanishing to positive order at both @xmath603 and @xmath668 therefore @xmath795 is polyhomogeneous on @xmath221 and implies that it is in @xmath796 and thus bounded , and in fact vanishing at both @xmath603 and @xmath668 the same is therefore true for @xmath797 now set @xmath798 by the hodge decomposition , @xmath799 is a pure form of degree @xmath712 with @xmath800 and @xmath801 by the resolvent construction and mapping properties of surgery operators from @xcite , @xmath799 is polyhomogeneous on @xmath221 bounded , and vanishes to positive order at @xmath603 and @xmath668 finally , set @xmath802 ; we have that @xmath803 so @xmath656 is harmonic .
since @xmath804 @xmath656 has the same restrictions at @xmath603 and @xmath230 as @xmath646 namely 0 and @xmath805 thus , taking a basis @xmath806 of @xmath807 we can find harmonic forms @xmath808 on @xmath221 polyhomogeneous and bounded on @xmath221 such that @xmath809 and with @xmath810\in { \mathrm{h}}^q(m ; f)$ ] a positive multiple of @xmath811 $ ] for @xmath812 we can thus define @xmath721 to be the projection on the span of @xmath813 to construct @xmath814 we can proceed in a similar fashion .
if @xmath815 then we can just pick @xmath816 otherwise , take a class @xmath817 and choose a class @xmath818 such that @xmath819 represent @xmath820 by a smooth form @xmath821 of degree @xmath712 on @xmath43 ; without loss of generality , we can assume that in a tubular neighborhood @xmath822 @xmath821 is of the form @xmath823 with @xmath824 independent of @xmath151 in particular , if @xmath825 this means by the witt condition that @xmath826 so that @xmath827 if instead @xmath828 then the norm of @xmath829 is uniformly bounded as @xmath73 approaches zero .
thus , for any @xmath732 we have that the form @xmath830 on @xmath182 is in @xmath831 since @xmath832 it is such that @xmath833 moreover , we have that @xmath834 while @xmath835 represents the class @xmath836 consider then @xmath837 then we still have that @xmath780 @xmath838 and @xmath839 represents the class @xmath840 we also have that @xmath841 so the form @xmath842 is a well - defined @xmath712-form with @xmath801 as in the construction of @xmath843 the form @xmath844 is harmonic with @xmath845 and with @xmath846 the harmonic representative of the class @xmath840 thus , starting with with a basis @xmath847 of @xmath848 we can construct harmonic forms @xmath849 such that @xmath850 @xmath851 represents @xmath852 and @xmath853 then we define @xmath814 to be the projection on the range of @xmath854 finally , we set @xmath855 as before , it is polyhomogeneous with a basis which is polyhomogeneous on @xmath182 . since we understand the restrictions of every space on the right - hand side , we conclude that it has the appropriate restrictions at @xmath603 and @xmath230 .
let @xmath856 , then consider @xmath857 , which we claim is equal to zero .
suppose not .
then it is certainly in the image of @xmath858 , and as it is in the image of @xmath568 , it must be in the image of one of the @xmath859 projections . by duality and the fact that @xmath860 , it is orthogonal to everything in the image of @xmath861 , so it must be in the image of @xmath862 . however , @xmath863 ; therefore , by the inductive hypothesis , @xmath864 .
hence @xmath865 as required .
then lemma [ se.8 ] immediately gives the required estimates for @xmath866 .
this completes the inductive step , and the proof for @xmath522 . for @xmath867
, we obtain the results by applying poincar duality and the corresponding results for @xmath535 finally , if @xmath93 is even , then , applying lemma [ se.8 ] as in the case @xmath868 to @xmath869 , we obtain the part of @xmath870 projecting on @xmath871 using the hodge @xmath549-operator and the corresponding result for @xmath534 we get the other part of @xmath872 since each eigenspace of positive small eigenvalues is even dimensional and formed of pairs of eigenfunctions by lemma [ se.7 ] , we see from the statement of the theorem when @xmath873 that we get all of @xmath874 so @xmath875 in every degree , the product of positive small eigenvalues is polyhomogeneous in @xmath50 . furthermore , for @xmath876 the product of all positive small eigenvalues of @xmath493 in degree @xmath712 is asymptotic to @xmath877 as @xmath878 where the subscript @xmath879 for a map @xmath11 denotes the restriction @xmath880 of @xmath11 to the orthogonal complement of its kernel with respect to the @xmath38-inner product .
we use the convention that @xmath881 when @xmath882 is the zero map .
for @xmath883 the product is asymptotic to @xmath884 as @xmath878 where @xmath885 .
if @xmath93 is even and @xmath555 then the product is asymptotic to @xmath886 as @xmath878 where @xmath887 .
[ poly.1 ] by theorem [ se.11 ] , the product of positive small eigenvalues in degree @xmath712 is polyhomogeneous in @xmath73 since it is given by @xmath888 in the asymptotic behavior of this product for @xmath730 , the first term comes from the eigenvalues corresponding to @xmath889 and is computed using . here
, we are relying on the fact that the map @xmath890 maps an orthonormal basis by harmonic representatives of @xmath891 to an orthogonal basis of harmonic representatives of @xmath892 indeed , if @xmath602 and @xmath893 are orthogonal harmonic representatives of classes in @xmath891 and @xmath572 and @xmath591 are extensions in the range of @xmath662 then similarly to , we have that @xmath894 from which we see that @xmath895 and @xmath896 are orthogonal as claimed .
the second term comes from the small eigenvalues corresponding to @xmath897 which by lemma [ se.7 ] are the same as those corresponding to @xmath898 this proves the first statement for @xmath45 and the same argument also proves it for @xmath535 the second statement in the theorem follows immediately by poincar duality . for the last statement , it suffices to notice that from part ( v ) of theorem [ se.11 ] , the positive small eigenvalues in degree @xmath899 are the same as the small eigenvalues coming from @xmath869 and @xmath900 we can now compute .
first let us define @xmath901 and note that , with @xmath902 in degree @xmath712 , taking the finite part in @xmath73 gives @xmath903 thus we have @xmath904 if we assume that the metric @xmath23 is compatible with the flat connection on @xmath45 ( which implies orthogonal holonomy ) , then we can use poincar duality to rewrite as a sum over @xmath905 @xmath906 the second consequence of the analysis in @xcite that we will use is about the asymptotics of harmonic forms as @xmath907 recall from
[ subsec : at ] that when @xmath93 is odd , the analytically defined metric invariant quantity is @xmath908^{(-1)^q } \right)\ ] ] where @xmath909 is a fixed basis of @xmath910 @xmath119 is an orthonormal basis of harmonic representatives with respect to the metrics @xmath51 and @xmath23 , and where @xmath122 = |{\operatorname{det}}w^q|$ ] with @xmath123 the matrix such that @xmath124 in this section we compute the asymptotic expansion of @xmath911^{(-1)^q } \right)$ ] when @xmath93 is odd .
we are interested in the coefficient of @xmath912 as terms dependent on @xmath73 will cancel out with those in the expansion of @xmath913 + to compute this contribution , we will make a specific choice for the basis @xmath840 namely , we let @xmath914 and @xmath915 be bases of orthonormal harmonic representatives for @xmath499 and for @xmath48 with respect to the metrics @xmath916 and @xmath15 respectively and the metric @xmath917 by an orthogonal transformation we can also assume without loss of generality that they are compatible with the decompositions @xmath918 then take @xmath919 to be the subset of @xmath920 which is a basis compatible with the canonical decomposition @xmath921 similarly , for @xmath735 we take @xmath919 to be the subset of @xmath922 compatible with the canonical decomposition @xmath923 with these choices , the constant term in the asymptotic expansion of @xmath924 $ ] comes from @xmath925 if @xmath730 and from @xmath915 otherwise . to be precise , let @xmath41 be a harmonic form on @xmath48 with @xmath926\in { \mathcal{h}}^{q-1}_h(z;f)$ ] and @xmath927 and let @xmath928 be a harmonic form poincar dual to @xmath929 so that @xmath930 if @xmath931 the element @xmath932\in { \mathrm{h}}^q(m;f)$ ] corresponding to the form @xmath41 in the decomposition can be represented by a form @xmath933 with support in a tubular neighborhood @xmath934 of @xmath48 in @xmath43 such that @xmath935 on the other hand , by theorem [ se.11 ] , the harmonic form @xmath119 with respect to @xmath51 of @xmath38-norm equal to @xmath774 representing a positive multiple of the class as @xmath932 $ ] in @xmath118 is asymptotically of the form @xmath936 in a neighborhood of @xmath194 .
thus , from , we see that asymptotically as @xmath73 tends to zero , @xmath932\sim \gamma_{{\varepsilon } } [ \omega_{{\varepsilon}}]$ ] with @xmath937 this implies that for @xmath931 @xmath938 = -\dim { \mathcal{h}}^{q-1}_h(z;f )
\left(\frac12 \log c_{\frac{m-1}2-(q-1 ) } + ( q-1-\frac{m-1}2)\log{\varepsilon}\right)+ o(1 ) \label{ff.12}\ ] ] as @xmath73 tends to zero .
when @xmath939 the form @xmath933 representing the cohomology class @xmath932 $ ] corresponding to the form @xmath940 under the decomposition is such that its restriction to @xmath48 is @xmath926,$ ] in other words , @xmath941 another application of theorem [ se.11 ] shows that the harmonic form @xmath119 with respect @xmath942 of @xmath38-norm equal to @xmath774 representing a positive multiple of the class as @xmath932 $ ] in @xmath118 is asymptotically of the form @xmath943 as @xmath73 tends to zero . from , we thus see that @xmath944\sim \frac{\sqrt{c_{q-\frac{m-1}2 } } } { { \varepsilon}^{q-\frac{m-1}2 } } [ \omega_{{\varepsilon}}]\ ] ] as @xmath73 tends to zero . taking the logarithm , we obtain that for @xmath735 @xmath938= \dim { \mathcal{h}}^{q}_h(z;f)\left ( \frac{1}2\log c_{q-\frac{m-1}2 } + ( \frac{m-1}2-q)\log{\varepsilon}\right)+ o(1 ) .
\label{ff.15}\ ] ] combining and , we see that when @xmath93 is odd , @xmath945^{(-1)^q } ] = -\frac12 \sum_{\substack{0\leq q \leq m \\ q \neq v/2 } } ( -1)^q \dim { \mathcal{h}}^{q}_h(z;f ) \log c_{|v/2-q|}.\ ] ] if we assume that the metric @xmath23 is compatible with the flat connection on @xmath45 ( which implies orthogonal holonomy ) , then we can use poincar duality to rewrite as a sum over @xmath476 : @xmath946
let @xmath43 be a closed manifold with a two - sided hypersurface @xmath126 we endow @xmath182 with an @xmath198 metric @xmath51 and a flat bundle @xmath218 with bundle metric @xmath20 not necessarily compatible with the flat connection , and in this section we determine the limit as @xmath63 of analytic torsion .
+ in ( * ? ? ?
* theorem 10.2 ) , we have computed the constant term in the expansion as @xmath63 of the logarithm of analytic torsion : @xmath947,g_0,f , g_f ) \\ + { \operatorname{lat}}([-\pi/2,\pi/2 ] , d_b , { \mathcal{h}}^*(z;f ) ) - \tfrac12\sum ( -1)^q q \log\tau_{{\mathrm{small}}}(\delta_q ) . \label{finfor.1}\end{gathered}\ ] ] we have now computed the last two terms , which leads to the following theorem . [ thm : mainat ] let @xmath43 be an odd - dimensional closed manifold , @xmath48 a two - sided hypersurface in @xmath52 and @xmath218 a flat vector bundle endowed with a bundle metric @xmath23 that is even at @xmath948 not necessarily compatible with the flat connection .
let @xmath51 be an @xmath198 metric that is product - type to order two , and let @xmath949 and @xmath950 be orthonormal bases of harmonic representatives with respect to the metrics @xmath916 and @xmath15 respectively , as well as @xmath917 then let @xmath951 be the corresponding basis of @xmath952 induced by @xmath953 , @xmath81 and the decompositions and
. then the analytic torsion satisfies @xmath954,\mu_0,f ) \\ + \frac12\sum_{\substack{0\leq
q\leq v \\ q\neq v/2 } } ( -1)^q { \left [ 2 \log |{\operatorname{det}}(j_q)_\perp| + \dim{\mathcal{h}}^q(z;f ) { \left ( ( 2q+1 ) { \operatorname{sign}}(2q - v ) \log |v-2q| \right ) } \right]}.\end{gathered}\ ] ] if @xmath23 is compatible with the flat connection on @xmath45 ( which implies orthonormal holonomy ) , then we have @xmath955,\mu_0,f)\\ + \sum_{q=0}^{\frac{v}2 - 1 } ( -1)^q\big [ 2\log |{\operatorname{det}}((j_q)_{\perp})|+\dim{\mathcal{h}}^q(z;f)\left ( ( v-2q)\log(v-2q)\right ) \big ] .\end{gathered}\ ] ] the term @xmath956 , d_b , { \mathcal{h}}^*(z;f))$ ] is computed in to be @xmath957 the term @xmath958 is computed in and is equal to @xmath959 and the contribution from the harmonic basis is computed in to be @xmath960^{(-1)^q } ] = \frac12 \sum_{\substack{0\leq q \leq m \\ q \neq v/2 } } ( -1)^q { \left ( - \dim { \mathrm{h}}^{q}_h(z;f ) \log c_{|v/2-q| } \right)}.\ ] ] adding these together and using the fact that @xmath961 we see that @xmath962,\mu_0,f ) \\ +
\frac12\sum_{\substack{0\leq q\leq v \\ q\neq v/2 } } ( -1)^q { \left ( 2 \log |{\operatorname{det}}(j_q)_\perp| + ( 2q+1 ) b_q { \operatorname{sign}}(2q - v ) \log |v-2q| \right)}.\end{gathered}\ ] ] if @xmath23 is compatible with the flat connection on @xmath99 then we can rewrite this as @xmath962,\mu_0,f ) \\
+ \sum_{q
< v/2 } ( -1)^q { \left ( 2 \log |{\operatorname{det}}(j_q)_\perp| + ( v-2q ) b_q\log |v-2q| \right)}.\end{gathered}\ ] ] when @xmath93 is even and @xmath23 is a metric compatible with the flat connection , we can compute the analytic torsion of the manifold with cusp directly from .
the analytic torsion of a manifold with cusp @xmath963 with link @xmath48 , of even dimension @xmath93 , equipped with a flat euclidean vector bundle @xmath45 satisfying the witt condition , is given by @xmath964 [ even.2 ] applying to the case where @xmath965 is the disjoint union of two copies of @xmath966 and using , and the fact that @xmath967 , we see that @xmath968,g_0,f)= m \sum_{q<\frac{m-1}2 } ( -1)^q \dim{\mathcal{h}}^q(z;f)\log(m-1 - 2q),\ ] ] from which the result follows .
we assume as in the previous section that @xmath969 is a flat vector bundle such that @xmath970 ( the witt condition ) and with holonomy inducing a unimodular representation @xmath971 .
we also let @xmath23 be a choice of bundle metric for @xmath45 .
moreover , we will now assume that @xmath93 is odd . to study the change of the @xmath3-torsion under a pinching surgery
, we will make use of the long exact sequence . as a complex
, we will denote this long exact sequence by @xmath972 recall that @xmath973 is the singular space associated to @xmath641 where @xmath87 is the disjoint union of the cones of each connected components of @xmath126 to relate the @xmath3-torsion of @xmath43 with an appropriate intersection @xmath3-torsion on @xmath974 we will need another exact sequence , namely the mayer - vietoris sequence obtained by writing @xmath975 as the union of @xmath603 with @xmath976 the pseudomanifold @xmath975 has a natural stratification of depth one , with singular stratum given by a disjoint union of points .
let @xmath977 be a choice of triangulation on @xmath975 compatible with this stratification and the decomposition .
recall from @xcite that we can then use @xmath977 and its first barycentric subdivision @xmath978 to define the complex of cochains @xmath979 .
this complex has natural restrictions to @xmath980 and @xmath981 so there is an induced mayer - vietoris short exact sequence of finite dimensional complexes @xmath982 & { \mathcal{r}}^*_{{\overline{m}}}({\widehat{m}};f)\ar[r ] & c^*_{t'}(m_0;f ) \oplus { \mathcal{r}}^*_{{\overline{m}}}({\mathcal{c}}z \sqcup { \mathcal{c}}z;f ) \ar[r ] & c^*_{t'}(z \sqcup z;f)\ar[r ] & 0 . }
\label{rt.1a}\ ] ] here , @xmath983 , where @xmath984 is seen as a @xmath985-module via the representation @xmath986 given by the holonomy of @xmath99 @xmath987 is the lift of @xmath978 to the universal cover @xmath988 of @xmath603 , and @xmath989 is the group of cochains associated to the triangulation @xmath990 similarly , we have that @xmath991,\ ] ] where @xmath992 labels the connected components of @xmath126 by ( * ? ? ?
* proposition 8.14 ) , the short exact sequence induces a mayer - vietoris long exact sequence involving intersection cohomology @xmath993 & { \operatorname{ih}}^q_{{\overline{m}}}({\widehat{m}};f)\ar[r]^-{i_q ' } & { \mathrm{h}}^q(m_0;f ) \oplus { \operatorname{ih}}^q_{{\overline{m}}}({\mathcal{c}}z \sqcup { \mathcal{c}}z;f ) \ar[r]^-{j_q ' } & { \mathrm{h}}^q(z \sqcup z;f)\ar[r]^-{{\partial}_q ' } & \cdots , } \label{rt.2}\ ] ] where @xmath994 and @xmath995 we will denote the long exact sequence by @xmath996 given any bases for @xmath510 @xmath997 @xmath998 , @xmath518 , and @xmath523 , we define corresponding bases for @xmath999 , and @xmath1000 via the direct sum . using these bases to define the @xmath3-torsions of @xmath52 @xmath974 @xmath641 @xmath87 , @xmath1001 @xmath128 @xmath1002 @xmath1003 and @xmath1004 , we have @xmath1005 [ rt.3 ] by the formula of milnor @xcite , we have that @xmath1006 and @xmath1007 combining these two relations gives the result .
note that the direct sum assumption is used to write , for example , @xmath1008 .
we now make a particular choice of bases for these spaces that allows a direct comparison with and also makes some of the terms in more explicit , in particular @xmath1009 recall from section [ sec : harmbases ] the decompositions : @xmath1010 @xmath1011 @xmath1012 we use the same bases as in section [ sec : harmbases ] for @xmath1013 , @xmath518 , and @xmath118 ; namely , orthonormal bases @xmath914 , @xmath1014 compatible with the decompositions and with @xmath919 the basis induced from @xmath953 , @xmath81 and the decompositions and .
for @xmath732 we use the canonical identification @xmath1015 to get a corresponding basis for @xmath1016 similarly , for @xmath1017 the canonical identification @xmath1018 gives a corresponding basis for @xmath1019 we also need to make a choice of basis of @xmath1020 we will take the one induced by our choice of basis for @xmath518 and the canonical identification @xmath1021 it remains to make a choice of basis for @xmath523 when @xmath735 but at the moment we can at least make a partial computation of @xmath1009 with the choice of bases made above , the contribution to @xmath1022 coming from cohomology classes of degree @xmath730 is given by @xmath1023 [ rt.5 ] first we compute the contribution to @xmath1024 . to do this , it suffices to notice that the restriction of @xmath1025 to the second factor induces the identity map @xmath1026 with respect to our choice of bases , while @xmath1027 composed with the projection on the first factor gives the canonical identification @xmath1028 ( for @xmath730 ) used to choose our basis for @xmath523 . as for @xmath1029 , the decompositions and are such that in the long exact sequence @xmath1030 @xmath1031 with our choice of bases , this means that the contribution to @xmath1029 coming from cohomology classes of degree @xmath730 is given by @xmath1032 ( recall that @xmath1033 is the restriction of a map @xmath11 to @xmath1034 ) .
since @xmath1035 and @xmath1036 are the identity maps on @xmath531 and @xmath530 for our choices of bases , the result follows .
in degree @xmath735 we will take advantage of some cancellations occurring between @xmath1029 and @xmath1024 to compute @xmath1022 directly .
first , notice that for @xmath735 the long exact seqence @xmath1004 corresponds to the relative long exact sequence associated to the pair @xmath1037 @xmath993 & { \mathrm{h}}^q_c(m_0;f)\ar[r]^-{i_q ' } & { \mathrm{h}}^q(m_0;f ) \ar[r]^-{j_q ' } & { \mathrm{h}}^q(z \sqcup z;f)\ar[r]^-{{\partial}_q ' } & \cdots , } \quad q\ge \frac{m+1}2 , \label{rt.7}\ ] ] under the canonical identification @xmath1038 this leads to the following commutative diagram between the long exact sequences @xmath1003 and @xmath1004 when @xmath1039 @xmath993 & { \mathrm{h}}^q_c(m_0;f)\ar[r]^{i_q'}\ar[d]^{{\widehat{i}}_q } & { \mathrm{h}}^q(m_0;f ) \ar[r]^-{j_q'}\ar[d]^{{\operatorname{id } } } & { \mathrm{h}}^q(z \sqcup z;f)\ar[r]^{{\partial}_q'}\ar[d]^{\beta_q } & { \mathrm{h}}^{q+1}_c(m_0;f)\ar[r]\ar[d]^{\alpha_{q+1 } } & \cdots \\
\cdots \ar[r ] & { \mathrm{h}}^q(m;f ) \ar[r]^{i_q } & { \mathrm{h}}^q(m_0;f ) \ar[r]^{j_q } & { \mathrm{h}}^q(z;f ) \ar[r]^-{{\partial}_q } & h^{q+1}(m;f ) \ar[r ] & \cdots , } \label{rt.8}\ ] ] where @xmath1040 defined in is the standard push - forward map and the map @xmath1041 is given by @xmath1042 and the canonical identification @xmath1043 this definition suggests that we take a different orthonormal basis of harmonic forms on @xmath1044 namely , if @xmath1045 is our chosen basis for @xmath1046 then we take the basis @xmath1047 for @xmath1048 since this change of basis is orthogonal , it has no effect on the torsion of @xmath513 , and in particular still holds if we compute the torsion of @xmath513 with respect to this new basis .
we can now make the following simple observations . for @xmath1049
the following assertions hold : 1 . @xmath1050 ; 2 .
@xmath1051 ; 3 .
if @xmath1052 is an orthonormal basis of @xmath1053 then @xmath1054 is a basis of @xmath1055 this basis is , however , not necessarily orthonormal ; 4 .
@xmath1056 ; 5 .
the composition @xmath1057 ^ -{i_q } & { \mathrm{h}}^q(m_0;f ) \ar[r]^-{j_q ' } & { \mathrm{h}}^q(z;f)\oplus { \mathrm{h}}^q(z;f ) \ar[r]^-{{\operatorname{pr}}_d } & { \mathrm{h}}^q(z;f ) \ar[r ] & { \mathcal{h}}^q_h(z;f ) } \ ] ] is the identity map , where @xmath1058 is the map @xmath1059 in particular , the map @xmath1060 is injective when restricted to @xmath1061 [ des.1 ] the proof of ( 1 ) and ( 2 ) follows from the identification of @xmath542 when @xmath1062 for ( 3 ) , it follows by noticing that , still under the identification @xmath1063 we have that @xmath1064 where @xmath1065 is the diagonal inclusion . for ( 4 ) , this is by exactness of the bottom sequence in , since @xmath1066 finally , ( 5 ) follows from and the definition of the map @xmath1067 therefore , removing the span of @xmath1068 in @xmath1069 @xmath1070 in @xmath1071 @xmath530 in @xmath118 and @xmath1072 in @xmath509 we obtain from the following commutative diagram of long exact sequences : @xmath993 & { \mathrm{h}}^q_h(m_0;f)\ar[r]\ar[d]^{{\operatorname{id } } } & { \mathrm{h}}^q(m_0;f)/i_q({\mathcal{h}}^q_h(z;f ) ) \ar[r]\ar[d]^{{\operatorname{id } } } & { \mathrm{h}}^q(z;f)\ar[r]\ar[d]^{{\operatorname{id } } } & \cdots
\\ \cdots \ar[r ] & { \mathrm{h}}^q_h(m_0;f)\ar[r ] & { \mathrm{h}}^q(m_0;f)/i_q({\mathcal{h}}^q_h(z;f ) ) \ar[r ] & { \mathrm{h}}^q(z;f)\ar[r ] & \cdots . } \label{rt.9}\ ] ] in this diagram , the contribution of the top row to @xmath1073 is cancelled by the contribution from the bottom row .
therefore , in degree @xmath1017 the only contributions to @xmath1073 come from * @xmath1074 when restricted to the span of @xmath1075 in @xmath1076 * @xmath1077 and * @xmath1078 to reach this conclusion , we have tacitly assumed that we have chosen a basis of @xmath523 for @xmath1049 which includes a basis of @xmath1079 we can go one step further and choose this basis so that the corresponding basis of @xmath1080 is the image under @xmath1060 of the chosen basis on @xmath1081 with this choice , @xmath1060 does not contribute to @xmath1073 and we are left with the contributions of @xmath1074 and @xmath1082 we are now ready to state the refinement of theorem [ rt.3 ] .
let @xmath914 and @xmath915 be bases of @xmath1013 and @xmath518 , orthonormal with respect to the metrics @xmath916 and @xmath15 respectively , and compatible with the decompositions .
let @xmath919 be a basis of @xmath118 compatible with and and choose the basis for @xmath998 induced by .
using these bases to define the corresponding @xmath3-torsions , we have the relation @xmath1083 [ rt.10 ] by theorem [ rt.3 ] , it remains therefore to compute @xmath1084 by lemma [ rt.5 ] and the discussion above , it remains to compute the contributions coming from @xmath1085 and @xmath1074 in degree @xmath1062 first , notice that with the choice of basis we have made for @xmath509 @xmath1085 is almost an isometry ; more precisely , @xmath1086 is an isometry . the contribution of @xmath1025 to @xmath1073
is therefore given by @xmath1087 on the other hand , identifying the span of @xmath1075 isometrically with @xmath1053 we see that the map @xmath1088 corresponds to @xmath1089 hence to the adjoint of @xmath1090 where @xmath534 is the flat vector bundle dual of @xmath45 .
thus , the contribution of @xmath1091 to @xmath1073 is given by @xmath1092 with the convention again that @xmath1093 for @xmath533 .
combining and with lemma [ rt.5 ] and using the fact that @xmath1094 gives the result .
if @xmath45 is a euclidean flat vector bundle then @xmath1095 by ( * ? ?
* proposition 1.19 ) , so using poincar duality , the formula simplifies to @xmath1096 even though we have made a specific choice of basis for @xmath523 to prove theorem [ rt.10 ] , the final result is independent of such a choice .
let @xmath105 be a closed riemannian manifold with a flat bundle @xmath32 corresponding to a unimodular representation @xmath1097 .
the cheeger - mller theorem @xcite gives an equality between the analytic torsion and the r - torsion for every choice of basis @xmath951 of the cohomology groups : @xmath1098 we have analyzed the behavior of both sides of this equation under analytic cusp surgery - the left - hand side in theorem [ thm : mainat ] and the right - hand side in theorem [ rt.10 ] . in this section we use these results to conclude a cheeger - mller theorem for witt representations on manifolds with cusps .
let @xmath43 be an odd - dimensional manifold , @xmath1099 a two sided hypersurface , @xmath51 a cusp surgery metric which is product - type to order two , and @xmath478 a flat witt vector bundle with holonomy inducing a unimodular representation @xmath971 .
let @xmath23 be a choice of bundle metric of @xmath45 and let @xmath914 and @xmath915 be bases of @xmath1100 and @xmath518 , consisting of harmonic forms which are orthonormal with respect to the metrics @xmath23 , @xmath916 and @xmath15 .
choose the basis @xmath84 for @xmath998 induced by .
using these bases to define the corresponding @xmath3-torsions and analytic torsion , we have the following formula : @xmath1101 , \mu_{m_0},f ) = \log { \left ( \frac{i\tau^{{\overline{m}}}({\widehat{m}}_0 , \mu_{m_0},f)\tau(z;f)}{i\tau^{{\overline{m}}}({\mathcal{c}}z,\mu_{{\mathcal{c}}z},f)^2 } \right ) } -\sum_{q>\frac{m-1}2 } ( -1)^q \frac{\dim h^q(z;f)}{2 } \log 2\\ - \sum_{\substack{0\leq q\leq m-1 \\ q\neq \frac{m-1}2 } } ( -1)^q \frac{\dim { \mathrm{h}}^q(z;f)}{2}\left [ |m-1 - 2q|\log|m-1 - 2q| \right ] .
\end{gathered}\ ] ] [ ff.19 ] assume ( by an orthogonal change of basis , if necessary ) that @xmath953 and @xmath81 respect the decompositions and , then pick the basis @xmath951 for @xmath118 induced by and .
the theorem then follows immediately from theorems [ thm : mainat ] and [ rt.10 ] together with the cheeger - mller theorem on @xmath43 . in particular , applying this result to the case where @xmath965 is the disjoint union of two copies of @xmath966 a manifold with cusp with link @xmath48 , where @xmath916 is product - type to order two , we obtain the following : let @xmath79 be a flat witt bundle with unimodular holonomy endowed with a bundle metric @xmath23 that extends smoothly to the double of @xmath4 across @xmath22 let @xmath1102 and @xmath81 be bases of @xmath1103 and @xmath518 respectively , consisting of harmonic forms orthonormal with respect to @xmath23 , @xmath916 and @xmath15 .
the canonical identification gives a basis @xmath84 for @xmath1104 using these bases to define the corresponding @xmath3-torsions and analytic torsion , we have the following formula : @xmath1105.\end{gathered}\ ] ] [ ff.20 ] note that @xmath45 extends to a flat witt vector bundle on the double @xmath1106 and the hypothesis on @xmath23 means that it comes from a bundle metric on @xmath43 as well , so the previous theorem does indeed apply .
in addition to analyzing the analytic torsion , our results can also be used to analyze the behavior of families of hyperbolic metrics on surfaces which approach the boundary of teichmller space , giving a new perspective on results of wolpert and of burger @xcite . in particular
, we analyze the so - called ` plumbing construction ' .
first we describe this construction : let @xmath1107 be a hyperbolic surface with nodes @xmath1108 @xmath1109 @xmath1110 , where each node represents a pair of cusps at punctures @xmath1111 and @xmath1112 of @xmath1113 let @xmath1114 @xmath1115 @xmath1116 be neighborhoods of @xmath1111 and @xmath1112 respectively .
suppose without loss of generality that each @xmath1117 is a disk of radius @xmath1118 for some fixed @xmath1119 and that we have local coordinates @xmath1120 and @xmath1121 with @xmath1122 , @xmath1123 and such that the hyperbolic metric @xmath916 takes the forms @xmath1124 in terms of these coordinates .
then , for a sufficiently small choice of @xmath1125 there exists an open set @xmath1126 disjoint from each @xmath1117 and a set of beltrami differentials @xmath1127
@xmath1128 which are supported in @xmath1126 and which span the tangent space of the boundary of teichmller space at @xmath1129 for @xmath173 sufficiently close to the origin in @xmath1130 we can solve the beltrami equation for @xmath1131 and obtain a family of hyperbolic surfaces ( with cusps ) which we denote @xmath1132 the hyperbolic metrics @xmath1133 on these surfaces are smooth in @xmath173 and are conformal to the hyperbolic metric @xmath916 on @xmath1107 in each @xmath1117 @xcite .
we now introduce a degeneration parameter @xmath1134 describing the opening of the nodes .
for each @xmath1135 sufficiently close to the origin , and each @xmath1136,$ ] we remove a pair of disks @xmath1137 of radius @xmath1138 around the @xmath992th node and identify the annuli @xmath1139 and @xmath1140 in complex coordinates @xmath1141 by @xmath1142 this produces new riemann surface @xmath1143 spanning a neighborhood of @xmath1107 in teichmller space @xcite .
each surface @xmath1143 can be equipped with a unique hyperbolic metric @xmath1144 in the conformal class specified by the complex structure . to describe the behavior of @xmath1144 as @xmath1145 , notice that the local model describing the opening of the node , the degeneration fixture @xmath1146 is a complex manifold fibering over the disk @xmath1147 with fiber above @xmath820 naturally identified with the annulus @xmath1148 for @xmath1149 . on each fibre ,
the unique complete hyperbolic metric in the conformal class specified by the complex structure is given by @xmath1150 at @xmath1151 this model degenerates to give two cusps as in .
in fact , making the change of variables @xmath1152 , @xmath1153 , we obtain @xmath1154 which is precisely the degeneration model considered in the present paper . for each sufficiently small @xmath1135 , we can use this model and construct an approximate hyperbolic metric @xmath1155 by gluing .
more precisely , let @xmath1156 be a metric which on each @xmath1157 is given by the metric @xmath1158 .
then let @xmath1159 be a cutoff function on @xmath1107 which is zero within a distance @xmath1160 of each node and identically 1 outside a distance @xmath1161 of each node , and whose gradient has support in a union of annuli with inner radius @xmath1160 and outer radius @xmath1161 about each node .
finally , as in @xcite , define a new metric @xmath1155 on @xmath1107 by @xmath1162 this family of metrics is smooth in @xmath1163 away from the nodes @xcite .
the metrics @xmath1155 are _ not _ necessarily exactly hyperbolic , as their curvature may not be identically @xmath1164 on the annuli where @xmath1165 may be nonzero .
however , their curvature may be computed directly ; call it @xmath1166 then we must have @xmath1167 where @xmath1168 is the solution of the prescribed - curvature equation @xmath1169 where @xmath1170 is the laplacian with non - negative spectrum corresponding to the metric @xmath1155 . as shown in @xcite ,
the metric @xmath1155 is a good approximation of @xmath1144 in the sense that @xmath1171 see also @xcite for an expansion at @xmath1172 . in @xcite ,
wolpert also obtains estimates for the small eigenvalues and for the determinants of the laplacian of the metrics @xmath1144 ; see also @xcite for a sharpening of the eigenvalue asymptotics . since the metrics @xmath1155 are exactly of the form in a fixed neighborhood of the nodes , our work may be used to recover and extend these results in many cases . indeed ,
for any fixed @xmath1173 , we may directly apply our results to @xmath1174 so using together with the prescribed curvature equation and its solution @xmath1175 , we can derive corresponding results for @xmath1176 first we analyze the behavior of the small eigenvalues of @xmath1177 the curvature @xmath1178 and the corresponding solution of the prescribed curvature equation are analyzed carefully in @xcite . from (
* section 3 ) , we conclude that @xmath1179 as @xmath1180 uniformly in @xmath1181 ( in fact , @xmath1182 ) ) . as a consequence
, it is proved in ( * ? ? ?
* section 4 ) that for any @xmath1183 there exists an @xmath1184 such that if @xmath1185 then for all sufficiently small @xmath173 and all points on the surface , @xmath1186 by the well - known result of dodziuk ( * ? ? ?
3.3 ) , the quotients of the nonzero eigenvales of the laplacians for @xmath1187 and @xmath1188 therefore approach 1 as @xmath1180 which allows us to apply our analysis of the small eigenvalues in section [ sec : small ] to conclude : [ lem : burgersmalleigen ] as @xmath73 goes to zero , the positive small eigenvalues @xmath1189 of @xmath1190 satisfy @xmath1191 where @xmath1192 can be computed explicitly for each small eigenvalue using the methods of section [ sec : small ] . as a particular case ,
consider the situation with @xmath1193 from the long exact sequence , there is one small eigenvalue if the manifold @xmath1194 becomes disconnected in the limit and zero if it does not .
if it becomes disconnected and the volumes of the two connected components of the limit are @xmath1195 and @xmath1196 we conclude from lemma [ se.8 ] that the leading asymptotic of the single small eigenvalue is @xmath1197 this agrees with the result of burger @xcite , who computed these eigenvalue asymptotics , to the same accuracy and with specific values of @xmath1198 using methods involving a comparison with the graph laplacian
. we can also analyze the determinant of the laplacian @xmath1199 observe that it is easy to show , by applying the maximum principle to the prescribed curvature equation , that whenever @xmath73 is small enough so that @xmath1200 then there is a constant @xmath1201 such that @xmath1202 by substituting these bounds into the prescribed curvature equation , we also get a @xmath1203 bound for @xmath1204 therefore , for sufficiently small @xmath69 there is a universal constant @xmath1201 such that @xmath1205 we may now apply the polyakov formula in the form from @xcite : @xmath1206 using integration by parts and the bounds we have just proven , it is straightforward to show that for sufficiently small @xmath69 there is a universal constant @xmath1201 ( possibly different from the one above ) such that @xmath1207 we now claim : since @xmath1209 is a family of cusp surgery metrics , the log determinant in question has a polyhomogeneous expansion as @xmath907 we need to investigate all the divergent terms in this expansion , so we need to be specific about the leading orders . recall from (
7 ) that the polyhomogeneous expansion of the log determinant is obtained by a pair of renormalized pushforwards : first an integration in the spatial variables to get the renormalized trace , then an integration in time to get the log determinant . from (
* ? ? ? * eq . ( 7.18 ) ) , the first integral shows that the renormalized trace is an element of @xmath1210 that is , it has order @xmath1211 at @xmath1212 , @xmath1211 at @xmath1213 , and order @xmath1214 at @xmath1215 . at @xmath1213 ,
the term of order @xmath1211 comes from @xmath1216 and the term of order @xmath1164 comes from @xmath1217 .
moreover , as explained in ( * ? ? ?
* lemma 7.10 ) , the terms in the expansion at @xmath1216 are lifts of polyhomogeneous expansions on @xmath149_{{\varepsilon}}\times { \mathbb{r}}^+_{\tau},$ ] smooth and non - singular in @xmath69 and so after the second push - forward ( the integration in time ) , the singular expansion in @xmath73 will come only from @xmath1218 in particular the leading term will not have order @xmath1219 but rather @xmath1220 it follows that , since the second push - forward sends both @xmath1213 and @xmath1215 to @xmath1221 the log determinant has index set equal to @xmath1222 we conclude that there exist constants @xmath1223 @xmath1224 , @xmath1225 , and @xmath1226 such that , as @xmath1180 @xmath1227 if @xmath1228 or if @xmath1229 and @xmath1230 et cetera , we are done .
however , from the definition of the determinant and the renormalized push - forward theorem , we know that @xmath1231 where @xmath1232 is the coefficient of the @xmath1233 term in the expansion of the renormalized trace at @xmath1234 from ( * ? ? ?
7 ) , @xmath1235 therefore , restricting to the diagonal and integrating yields @xmath1236 this integral may be evaluated and gives @xmath1237 which is negative .
this completes the calculation . combining this lemma with , we obtain a leading - order asymptotic formula for @xmath1238 .
in particular , we see that @xmath1239 which agrees with the result of wolpert ( * ? ? ?
* theorem 5.3 ) when we use the description of the determinant in terms of the selberg zeta function .
jean - michel bismut , xiaonan ma , and weiping zhang , _ asymptotic torsion and toeplitz operators _ , available online at http://www.math.u-psud.fr/~bismut/web_page_of_jean-michel_bismut/selected_publications_and_videos_files/bismutmazhangglob.pdf , 2011 .
colin guillarmou and david sher , _ low energy resolvent for the hodge laplacian : applications to riesz transform , sobolev estimates and analytic torsion _ , to appear , int .
res . not .
; available online at arxiv:1310.4694 , 2013 .
g. harder , _ on the cohomology of discrete arithmetically defined groups _ , discrete subgroups of lie groups and applications to moduli ( internat .
bombay , 1973 ) , oxford univ . press , bombay , 1975 , pp .
129160 . ,
_ the asymptotics of the ray - singer analytic torsion for hyperbolic @xmath91-manifolds _ , metric and dinferential geometry . the jeff cheeger anniversary volume , progress in math .
297 , birkhuser , 2012 , pp .
317352 .
jonathan pfaff , _ selberg zeta functions on odd - dimensional hyperbolic manifolds of finite volume _ , to appear in journal fr die reine und angewandte mathematik , available online at arxiv:1205.1754 , 2012 .
, _ families of riemann surfaces and weil - petersson geometry _ , cbms regional conference series in mathematics , vol .
113 , published for the conference board of the mathematical sciences , washington , dc ; by the american mathematical society , providence , ri , 2010 . | we establish a cheeger - mller theorem for unimodular representations satisfying a witt condition on a noncompact manifold with cusps .
this class of spaces includes all non - compact hyperbolic spaces of finite volume , but we do not assume that the metric has constant curvature nor that the link of the cusp is a torus .
we use renormalized traces in the sense of melrose to define the analytic torsion and we relate it to the intersection r - torsion of dar of the natural compactification to a stratified space .
our proof relies on our recent work on the behavior of the hodge laplacian spectrum on a closed manifold undergoing degeneration to a manifold with fibered cusps . |
qcd matter with @xmath1 flavors , @xmath2 colors and zero baryon chemical potential undergoes two finite temperature phase transitions , the chiral transition and the deconfinement .
the chiral transition is associated with the spontaneous breaking of the chiral symmetry , @xmath3 , below the critical temperature ( @xmath4 ) for massless quarks .
the order parameter for this transition is the chiral condensate . on the other hand
the deconfinement transition is associated with the spontaneous symmetry breaking , @xmath5 , above the critical temperature @xmath6 for infinitely heavy quarks .
the order parameter for this transition is the polyakov loop expectation value . for finite non - zero quark masses
both the chiral symmetry and the @xmath0 symmetry are explicitly broken .
nevertheless these two transitions show up as crossover , second or first order transitions depending on the values of the quark masses .
the chiral transition depends on the number of quark flavors and the deconfinement transition depends on the number of colors .
since the chiral symmetry and the @xmath0 symmetry seem nothing to do with each other one would expect that these two transitions occur independently .
however lattice qcd calculations have shown that both these transitions occur simultaneously , @xmath7 [ 1 - 5 ] .
furthermore strong correlation between the chiral condensate and the polyakov loop is observed around the phase transition point@xcite .
this is clear evidence that there is interaction between these two order parameters .
so studying the interaction between the chiral condensate and the polyakov loop is fundamental to understanding the interplay between the chiral transition and deconfinement .
there are several studies on the possible causes of the simultaneous occurrence of the chiral transition and the deconfinement transition . mixing between the gauge and matter field operators
has been suggested to explain the simultaneous chiral and deconfinement transitions @xcite .
some other studies consider that for small quark masses the chiral transition drives the deconfinement transition @xcite .
though most of the studies are concerned with why @xmath7 only a few have considered the effect of the interaction on the phase transition itself @xcite .
it seems natural to expect that if the interaction between the two order parameters can result in the simultaneous occurrence of the two transitions then the interaction may also have important effect on the phase transition .
one of the most interesting cases to study for the possible effects of this interaction is the 2-flavor chiral transition .
lattice qcd calculations have not yet been able to settle on the order of this phase transition .
lattice calculations by different groups do not agree on the order of this phase transition .
some lattice groups find the transition is second order @xcite and other groups find the transition is first order @xcite .
conventionally this transition is believed to be second order and in the universality class of @xmath8 heisenberg magnet @xcite .
but the effect of interaction between the chiral order parameter ( @xmath9 ) and the polyakov loop ( @xmath10 ) may change the behavior of this transition .
so in the present work we investigate the effect of interaction between @xmath9 and @xmath10 on the 2-flavor chiral transition within an effective lagrangian .
previously , the effect of interaction between the chiral order parameter and the polyakov loop has been studied in the renormalization group approach @xcite .
the main difference between the present work and previous studies is that we consider the explicit breaking of the @xmath0 symmetry .
explicit breaking of the @xmath0 symmetry can be introduced in the effective lagrangian by terms such as @xmath11 , @xmath12 .
previous studies @xcite have considered interaction term , such as @xmath13 , which respects both the chiral and the @xmath0 symmetry .
however , in the chiral limit , the interaction terms need not respect the @xmath0 symmetry .
so terms such as @xmath12 should be considered . as we discuss later such a interaction term
is always present if the explicit @xmath0 symmetry breaking is large , for example in the chiral limit .
for simplicity we consider @xmath14 color qcd .
we expect that different @xmath2 will not qualitatively change the physics we are discussing here . for @xmath15
the chiral order parameter @xmath9 is a four component vector field whereas for @xmath14 the polyakov loop @xmath10 is a real scalar field . in this work we basically study the effect of the three terms , @xmath16 , @xmath17 and @xmath18 in the effective lagrangian .
our main result is that the strong explicit breaking of the @xmath0 symmetry can make the chiral transition first order .
we show that the chiral transition can be first order even at the mean field level .
we also carry out numerical monte carlo simulations which confirm the first order phase transition for large enough explicit @xmath0 symmetry breaking .
we mention here that the @xmath15 chiral transition can be first order from the interaction term @xmath19 without @xmath0 symmetry breaking @xcite .
however lattice qcd results indicate that effect of the @xmath0 symmetric interaction term is small . a possible first order chiral transition can result more likely from the explicit @xmath0 symmetry breaking as we will argue later . conventionally explicit symmetry breaking weakens a phase transition .
but our results suggest that for a system with two order parameter fields explicit symmetry breaking can make the transition stronger .
it is interesting to note that the chiral order parameter @xmath9 does not couple to gauge fields directly .
the gauge fields seem to affect the chiral phase transition indirectly , through the polyakov loop .
we mention here that the effect of explicit symmetry breaking discussed here should not be restricted to the @xmath0 symmetry .
we expect that the explicit breaking of chiral symmetry may have some effect on the deconfinement transition in the large quark mass region .
we mention here that interaction between the chiral order parameter and the diquark fields is considered to study the chiral / color - superconducting transition at low temperature and high density @xcite .
this paper is organized as follows . in section - ii
we describe the effective lagrangian and discuss the effect of the interaction between @xmath9 and @xmath10 on the chiral transtition .
we describe our numerical monte carlo calculations and results in section - iii .
the discussions and conclusions are presented in section - iv .
we consider the following lagrangian @xcite in 3 dimension for the @xmath9 and the @xmath10 fields , @xmath20 some of the parameters of this reduced 3d lagrangian depend explicitly on temperature .
this lagrangian is invariant under @xmath8 rotation of the @xmath9 field .
the last two terms of @xmath21 break the @xmath22 symmetry ( @xmath23 ) explicitly .
the interactions between the chiral order parameter and the polyakov loop are taken care by 5th and 6th terms in @xmath21 .
the signs of the couplings @xmath24 and @xmath25 decides the correlation between the fluctuations of the @xmath9 and the @xmath10 fields .
for example when @xmath26 , the thermal average of the correlation between the @xmath9 and @xmath10 fluctuations , @xmath27 , is positive .
when @xmath28 , @xmath27 is negative .
such `` anti - correlation '' is seen between the fluctuations of the the chiral condensate and the polyakov loop in the results of lattice qcd calculations@xcite .
the correlation between the variations of the two order parameters with respect to temperature is of the same sign as the correlation between the fluctuations .
note that in the above lagrangian when @xmath29 the @xmath9 field acts like an ordering field for the @xmath10 field . in the chiral symmetric phase the chiral order parameter is small and
the polyakov loop expectation value is large .
the large expectation value of @xmath10 , in the chiral symmetric phase , can result only from the last term in @xmath21 ( with @xmath30 ) because @xmath31 is small . 0.2 cm as we have mentioned before previous studies @xcite have considered only the interaction term @xmath32 . when the coupling parameters @xmath33 , the chiral transition and the deconfinement transition do not always occur simultaneously . for
large value of the coupling @xmath24 the transitions can occur simultaneously and are of first order when coefficients of @xmath34 and @xmath35 are negative in eq.1 @xcite .
a large positive @xmath24 would increase the critical temperature for the deconfinement transition as the coefficent of @xmath35 term is negative for high temperatures .
lattice results on the other hand show that inclusion of dynamical quarks decrease the deconfinement transition temperature .
so in qcd the coupling @xmath24 should be small .
for small @xmath24 both the transitions are second order and do not occur simultaneously .
0.2 cm the coupling parameters @xmath25 and @xmath36 represent the strength of the @xmath0 explicit breaking .
so they should increase with decreasing quark masses as the @xmath0 breaking becomes severe .
this can be seen explicitly in the large quark mass region @xcite . to see the effect of the explicit @xmath0 breaking
let us consider @xmath37 and @xmath29 .
at the mean field level one can consider the temperature variation of the parameters @xmath38 and @xmath39 .
for simplicity we fix @xmath40 , @xmath41 and vary the @xmath38 parameter . to find the @xmath38 dependence of @xmath10 and @xmath9 expectation values one has to solve the following coupled equations , @xmath42 we have checked numerically that for large enough @xmath25 these equations give two solutions which correspond to a degenerate minima of the effective potential @xmath21 at some particular value of @xmath38 .
it may seem surprising that the potential @xmath21 has degenerate minima even though there is no cubic term for @xmath9 and @xmath10 in it .
however because of the coupling @xmath25 these two fields are mixed . though the mixing anle varies as @xmath38 is varied . with variation of @xmath38 the minimum of @xmath21 , in the @xmath43 plane , moves in directions other than the @xmath31 or @xmath10 axes . to understand how the minimum of @xmath21 behaves one should express @xmath21 in terms of variables which are the linear combinations of @xmath31 and @xmath10
this would invariably lead to cubic terms of the new fields in the effective potential . for some choice of parameters
the cubic term may then be important to cause degenerate minima of @xmath21 .
even if the explicit symmetry breaking is not strong enough at mean field level fluctuations at higher order can make the transition first order . at one loop
the fluctuations of the @xmath9 field will contribute to a non zero 3-point function of the @xmath10 field .
this 3-point function can be calculated perturbatively . in the high temperature approximation
the 3-point function is given by , @xmath44 assuming zero momentum to all the external @xmath45lines .
@xmath46 here is the mass of the @xmath9 field fluctuations .
given a suitable choice of parameters in the effective lagrangian the three point function can be significant . the consequence of this is a @xmath47 term in the potential @xmath21 with temperature dependent @xmath48 .
this term can cause discontinous change in @xmath10 as temperature varies .
when @xmath10 field changes discontinuously the @xmath9 field will also changes discontinuously because of the coupling term @xmath49 .
if the coefficient of the @xmath50 term changes from @xmath51 to @xmath52 due to discontinuous change in @xmath10 then @xmath9 will change discontinuously from zero to non - zero . the @xmath53 term can also come from other types of explicit symmetry breaking coupling terms but we think @xmath49 is the simplest term in our model .
+ the situation with @xmath54 and @xmath55 is same as the one discussed above . when the explicit symmetry breaking parameter @xmath36 is large , @xmath22 symmetry of the @xmath10 field is lost and the @xmath10 field always has non - zero expectation value @xmath56 . in order to study the fluctuations one must expand the potential @xmath21 around @xmath56 , @xmath57
this gives rise to a term like @xmath58 coming from the expansion of @xmath59 around @xmath60 .
now the term @xmath58 is similar to the one discussed above .
so there can be first order transition when @xmath61 like in the case when @xmath62 .
we observed that at the mean field the transition becomes second order for large quartic couplings @xmath63 .
now in the following section we discuss the numerical monte carlo simulations of the effective lagrangian ( eq.1 ) and the results .
these calculations include higher order as well as non perturbative fluctuations .
vs @xmath68,title="fig : " ] [ mf ] -0.3 cm vs @xmath68,title="fig : " ] [ pt ] -0.3 cm -0.3 cm since we take @xmath25 to be positive we see @xmath87 and @xmath94 increase or decrease simultaneously .
the choice of the values for the parameters is such that the variation of @xmath87 and @xmath94 are similar in magnitude . for @xmath95 increase in @xmath87 should lead to decrease in @xmath94 .
so in this case the hysteresis loop for @xmath87 will look somewhat similar to fig.1 while the hysteresis loop for @xmath94 will be more or less inverted about the y - axis .
0.2 cm for the second set of parameters we choose @xmath96 , @xmath97 , @xmath98 , @xmath99 and @xmath100 .
the hysteresis loop of @xmath87 and @xmath94 are observed by again varying the parameter @xmath68 .
the choice of @xmath69 and @xmath36 is such that the expectation value of @xmath10 is non - zero and positive .
the sign of @xmath24 assures that increase in @xmath87 leads to decrease in @xmath94 and vice - versa as the parameter @xmath68 is varied .
the hysteresis curves of the two order parameters are shown in fig.3 and fig.4 .
the values of @xmath101 in our calculations are chosen so that we can see first order phase transition clearly and the variation of @xmath87 and @xmath94 are of order o(1 ) across the transition point .
note that with suitable choice of @xmath69 and @xmath102 one change the average of the polyakov loop across the transition point .
the choice @xmath24 was such that the chiral transition turned out to be second order when the coupling @xmath36 was set to zero .
we also did simulations on a 4d lattice .
the results in this case are very similar to the 3d simulations . -1.3 cm vs @xmath68,title="fig : " ] [ mf ] -0.3 cm -0.3 cm vs @xmath68,title="fig : " ] [ pt ] -0.3 cm -0.3
cm the results in fig.1 - 4 show strong first order phase transition for the @xmath9 and @xmath10 fields . by fixing the coupling parameters @xmath24 , @xmath25 and @xmath36
we observed that the strength of the transition depends on the values of the quartic couplings @xmath63 and @xmath102 .
the transition becomes weaker with increase in any of the quartic couplings @xmath101 .
however even for larger quartic couplings a suitable choice of the parameters @xmath24 , @xmath25 and @xmath36 made the transition strong first order .
0.2 cm we also did calculations with small explicit symmetry breaking for the @xmath9 field by considering a linear @xmath9 term in the lagrangian . in this case
we found that the hysteresis loops of both @xmath87 and @xmath94 shrinking in size with increase in the coefficient of linear @xmath9 term in the lagrangian .
these results suggest that the transition becomes weaker when the chiral symmetry is explicitly broken .
using a simple effective lagrangian , which captures the chiral symmetry and @xmath0 symmetry of qcd , we have investigated the effect of the explicit symmetry breaking on the chiral phase transition . since we consider the chiral limit the @xmath22 symmetry of the polyakov loop is explicitly broken .
as we incorporate the explicit @xmath22 breaking interaction terms in the effective lagrangian we find the chiral transition becomes first order .
we observed that the first order transition becomes weaker when a small explicit symmetry breaking is considered for the chiral order parameter @xmath9 .
the results of our calculations show that two flavor chiral transition is of first order for large enough @xmath0 explicit breaking .
as we have mentioned before at present some lattice studies suggest that the transition is first order @xcite and some other studies show the transition is second order @xcite
. these conflicting results may be because the quark masses studied are not small enough or the lattices used are not big enough . in the previous study , for @xmath15 and @xmath103 @xcite , the second order
chiral transition becomes first order when coupled to the polyakov loop .
this is because the deconfinement transition is first order for @xmath103 with a cubic term @xmath104 in the effective potential with exact @xmath105 symmetry .
however when the quark masses are finite and decrease the deconfinement transition becomes weaker .
this can be understood due to explicit breaking of @xmath105 symmetry .
already in the large quark mass region the deconfinement transition becomes crossover which implies the the explicit symmetry breaking dominates over the effects of the above @xmath105 symmetric cubic term . for smaller quark masses
the explicit @xmath105 symmetry breaking likely grows and expected to be maximal in the chiral limit .
so for smaller quark masses , i.e in the chiral limit one should rather consider the effect of the explicit breaking of the @xmath105 symmetry .
the effects of explicit symmetry breaking discussed in this work should not be restricted to the explicit breaking of the @xmath0 symmetry . for 2-flavor and 2-color qcd
the deconfinement transition in the large quark mass region may have the effects coming from the explicit breaking of the chiral symmetry .
it may be possible that this effect change the phase transition behavior of the deconfinement transition in the heavy quark mass region .
we have benefited greatly from discussions with t. hatsuda and a. m. srivastava .
we would like to thank g. baym , s. datta , m. laine , t. matsuura and m. ohtani for comments .
this work is supported by the jsps postdoctoral fellowship for foreign researchers . | we study the effects of the interaction between the chiral condensate and the polyakov loop on the chiral transition within an effective lagrangian .
we find that the effects of the interaction change the order of the phase transition when the explicit breaking of the @xmath0 symmetry of the polyakov loop is large .
our results suggest that the chiral transition in 2-flavor qcd may be first order . |
observations of both cool and hot intergalactic gas make it clear that heavy elements are by no means confined to galaxies .
larson ( 1974 ) predicted that late in the galaxy formation process , galactic winds would carry heavy elements out into intergalactic space .
soon afterwards x - ray spectra of intracluster gas confirmed that the gas was quite metal - rich . in recent years observations of absorption lines in qso spectra
have elucidated the impact of supernovae on the diffuser gas that is found in the field , at least at high redshift .
these observations imply that , from a redshift @xmath1 onwards , there does not seem to be a parcel of gas that has not felt the impact of spernovae , probably mediated by galactic winds . while the ubiquity of the products of supernovae is beyond doubt , many aspects of the interaction of supernovae and intergalactic gas are highly uncertain .
outstanding questions include : ( i ) in what types of galaxy were intergalactic metals manufactured ?
( ii ) which type of supernova ( core - collapse or ia ) has been most important , and at what epoch ? ( iii )
how important have supernovae been for the entropy budget of intergalactic gas ?
most of the luminosity in the universe comes from spiral galaxies like the milky way .
do such galaxies blow sn - driven winds ? studies of the local ism show that a significant fraction of the pressure in the ism comes from cosmic - rays and the magnetic field .
supernovae and fast stellar winds are the main energy sources of the ism , so the dynamical importance of cosmic rays and magnetic fields suggests that a significant fraction of the kinetic energy which these objects pump into the ism is chanelled into cosmic rays and magnetic fields .
hence , assuming there to be @xmath2 supernova per century , the cosmic - ray reservoir is being energized at a rate @xmath3 .
how does this power , which could build up the observed interstellar pressure within a disk @xmath4 in radius and @xmath5 thick within @xmath6 , manifest itself ?
the local cosmic - ray energy spectrum is such that the energy is overwhelmingly contained in only mildly relativistic particles .
such particles have enormous lifetimes because they are too fast to suffer much coulomb scattering and too slow to produce significant synchrotron radiation which explains why the non - thermal radio luminosity of galaxies like the milky way is @xmath7 ( condon 1992 ) .
hence , the energy imparted by spernovae can not be radiated either directly or by transfer to thermal particles , and is not simply building up within the disk .
so it must drive expansion of the cosmic - ray plasma : the milky way must be blowing a wind that is working on the local igm at a rate @xmath8 .
what volume within the local group would this wind have filled to the present epoch ? there are two ways of answering this question .
if the wind were expanding into a vacuum , it would expand at @xmath9 and extend to @xmath10 in a hubble time .
this gives us an upper limit on the radius of influence of an @xmath11 galaxy .
a lower limit comes from assuming that the extragalactic baryons that are required by primordial nucleosynthesis theory were once distributed in galactic halos like dark matter , and the sn - driven wind raised its temperature by of order the virial temperature .
we assume that @xmath12 and @xmath13 , so with baryons following dark matter in a singular - isothermal halo of circular speed @xmath14 , the baryonic mass interior to radius @xmath15 is given by @xmath16 . equating the energy required to heat this material by the virial temperature to the time - integral of the sn - power , we find , so
@xmath17 is independent of @xmath14 . ]
@xmath18 we see that even the present supernova rate within the milky way would impact the igm out to near the mid point to m31 , and in reality the sn rate has almost certainly been substantially larger in the past .
recently , tripp , savage & jenkins ( 2000 ) have suggested that o@xmath19 absorption in quasar spectra points to a major reservoir of baryons in warm ( @xmath20k ) gas in galaxy groups . as pen ( 1999 ) points out , the thermal energy of this gas can not derive from gravity alone : without non - gravitational heating it would be so clumpy that its soft x - ray emission would violate the constraint imposed by the unresolved component of the x - ray background .
it is tempting to conclude that galactic winds provide the required heat source ( see below ) .
the most compelling evidence for the impact of supernovae comes from rich clusters of galaxies .
arnaud et al ( 1992 ) show that for these objects gas mass and @xmath21-band luminosity are highly correlated @xmath22 . at least in part
the steepness of this correlation will arise for two reasons .
first , more luminous clusters tend to have higher frations of early - type galaxies , which have lower @xmath23 per unit stellar mass , @xmath24 .
second , lower - luminosity clusters are less likely to attract infall , and more likely to suffer outflow of gas renzini ( 1997 ) finds that systems with temperature in excess of @xmath25 do not lose gas .
arnaud et al .
note that @xmath26 is as tightly correlated with the luminosity from e and s0 galaxies as with total luminosity : @xmath27 and argue that the exponent in this relation can be taken to be unity , so that with @xmath28 for early type galaxies set to @xmath29 , @xmath30 from the fact that @xmath31 it is clear that much of the igm has never been in a galaxy .
while the bulk of the gas may be primordial , the heavy elements in it are most certainly not . with @xmath29 the estimate of arnaud et al .
( 1992 ) becomes @xmath32 while renzini ( 1997 ) finds that for clusters hotter than @xmath33 @xmath34 independent of cluster temperature .
again adopting @xmath29 , renzini s value of the fe abundance becomes @xmath35 the igm - abundances of several @xmath36-elements , o , ne , mg , and especially si , have been determined for many clusters ( mushotzky et al .
when meteoric rather than solar - photospheric abundances are used as a point of reference , the igm proves to be only mildly @xmath36-enhanced by a factor @xmath37 relative to solar ( brighenti & mathews , 1999 ) . what does nucleosynthesis theory have to say about these mass fractions ?
the picture is confused by uncertainties in the yields of fe and to a certain extent si , from core - collapse sne , and the rates ( current and past ) of type ia sne . a key input into nucleosynthetic theory is the initial mass function ( imf ) .
i shall assume that for @xmath38 this is of salpeter s form , since there is now significant evidence that the imf is universal ( which is surprising ) and lies near salpeter s form at larger masses .
in particular , recent work based on hipparcos parallaxes ( binney , dehnen & bertelli , 2000 ) does not support the contention of scalo ( 1986 ) that the local imf is steeper than salpeter at @xmath38 , while kennicut , tamblyn & congdon ( 1994 ) argue that measurements of h@xmath36 fluxes from external disks rule out a steep imf . for a salpeter imf extending between @xmath39 and @xmath40
we expect 0.0068 core - collapse sne per m@xmath41 of star formation , and @xmath42 of the initial stellar mass would by now have been returned to the ism .
the average core - collapse sn is expected to inject @xmath43 of si and @xmath44 of fe into the ism ( gibson , loewenstein & mushotzky , 1997 ) , so from core - collapse sne we expect @xmath45 of si and @xmath46 of fe , where @xmath47 is the mass in stars before the onset of mass loss .
these masses fall short of those observed [ eqs ( [ mfea ] and ( [ mfer ] ) ] by a factor of order 2 .
the calculation ignores the existence of intergalactic stars ( mendez et al . , 1996 ) , but it also ignores the retention of heavy elements within galaxies . in practice these two omissions will roughly cancel , and a more elaborate calculation would leave us significantly short of heavy elements .
the resolution of this shortfall is controversial .
some argue for a flatter than salpeter imf ( david , 1997 ; gibson et al . , 1997 ) .
others argue that the missing heavies were produced by type ia supernovae ( renzini et al . , 1993 ;
ishimaru & arimoto , 1997 ) after all more than half of the fe in the solar neighbourhood is thought to come from type ia sne .
a type ia sn injects @xmath48 of si and @xmath49 of fe into the ism ( thielemann , nomoto & hashimoto , 1993 ) , so these objects inject four times as much fe as si , whereas core - collapse sne inject at least as much si as fe .
consequently , the observed ratio of si to fe masses in the igm ( @xmath50 ) constrains the importance of type ia sne .
the big problem with imposing this constraint at the present time is uncertainty in the ratio of mean yields of si and fe from core - collapse suprnovae : if as , is perfectly plausible , core - collapse supernovae produce significantly less fe than si , one will need a good number of type ia sne to bring the overall ratio @xmath51 down to @xmath50 as observed .
the current rate of type ia sne can in principle be determined in two ways : ( i ) direct observation ( with still depressingly uncertain results ) , and ( ii ) from models of isolated elliptical galaxies , which require the heating rate to decline towards the current epoch in such a way that galaxies of appropriate luminosity and ambient igm pressure are now suffering cooling catastrophes in increasing numbers ( loewenstein & mathews , 1987 ; ciotti et al . , 1991 ; binney & tabor , 1995 ) . the rates in times past , which will surely differ from the present one , are well - nigh impossible to determine or predict .
if rich clusters were fashioned by gravity alone , they should evolve self - similarly .
kaiser ( 1991 ) pointed out that it is then almost inevitable that there should be more x - ray luminous clusters in the past than now .
this conclusion follows because the luminosity of an individual cluster scales as the product @xmath52 of mass times density , and with @xmath53 the usual spectral index of the primoridial fluctuations , @xmath54 for @xmath55 , while @xmath56 , so the characteristic luminosity will be roughly independent of @xmath57 .
the number density of freshly collapsed objects will , on the other hand , be much higher at high @xmath57 .
hence there should have been many more luminous clusters in the past , in contradiction with observation .
the obvious way of suppressing the x - ray luminosities of early clusters is to have them form from preheated gas , which will either not be trapped by early potential wells , or will be trapped at lower density than in the unheated case .
kaiser estimates the entropy boost required by observing that the gas density at the core of a rich cluster is about @xmath58 times the current mean density , or the mean density at @xmath59 .
so we could get the gas onto the required adiabat by heating it to current cluster temperatures @xmath60k at @xmath59 .
alternatively , we could heat it to @xmath61 ^ 210^{7.5}\,$]k at redshift @xmath62 . pen
s lower limit on the entropy of gas in groups that was discussed above implies a similar energy budget .
the current maximum density of the gas is only an order - of - magnitude greater than the current cosmic mean density , so two order of magnitude smaller than the maximum density of cluster gas , and the temperature would be at @xmath63k an order - of magnitude lower than the temperature of cluster gas .
so the gas would be on a slightly higher adiabat .
on the other hand , it could be put onto that adiabat rather later and therefore at a smaller cost in energy .
do sne provide the requisite energy ?
the energy release per cosmic baryon is @xmath64 where @xmath65 is the number of baryons per unit mass , @xmath66 is the energy and @xmath67 is the mass of iron produced by a single sn . since core - collapse sne produce the least fe per unit energy , we will maximize the energy release per fe nucleus synthesized if we assume that type ia sne are unimportant . using equations ( [ mgas ] ) and ( [ mfer ] ) to evaluate equation ( [ esn ] ) under this assumption , and then equating the result to the thermal energy per baryon , @xmath68 , of a plasma of molecular weight @xmath69 , we find the plasma temperature to be @xmath70 this is clearly an upper limit on the achievable temperature because it is assumes core - collapse sne to be dominant while ignoring radiative losses , which are liable to be significant for many core - collapse sne , because they tend to go off in dense gas clouds .
my judgment is that the enegy budget is too tight , even if we assume that the gas is heated as late as @xmath71 , but others may disagree .
agn are strong candidates for providing the energy if sne can not do the job ( begelman , this volume ) .
as one moves down the mass scale through virial temperatures below @xmath72 , the ratio of gas mass to stellar mass declines ( renzini , 1997 ) .
one interpretation of this result is ( binney , 1980 ; whitmore , gilmore & jones , 1993 ) that intracluster gas is the material from which disks are formed in the field , and that its early enrichment to @xmath73 by early - type galaxies is the counterpart of the pre - enrichment of the galactic disk by the bulge , which ostriker & thuan ( 1975 ) argued is the correct solution of the g - dwarf problem . whereas in rich clusters
late - infalling , high angular - momentum gas was shocked to the virial temperature before it could form an accretion disk around a spheroid , in systems like the local group disks did form , and the floor metallicity , @xmath73 , mirrors the current metallicity of intracluster gas . in this picture , the declining fraction of mass in the igm as one moves to poorer and poorer clusters simply reflects the fact that a larger and larger fraction of the original igm has settled into disks . the argument i have given that star - forming @xmath11 galaxies blow powerful winds suggests that all @xmath11 galaxies , cluster and field alike , lose most of their heavy elements into deep space .
if this is so , conventional galaxy evolution models require substantial revision
. moreover , in this picture overwhelming quantities of metal - enriched baryons are stored in deep space .
more than one tentative line of argument now suggests that this is indeed the case . within a few years
we will probably know for sure . | an energetic argument implies that a galaxy like the milky way is blowing a powerful wind that carries away most of the heavy elements currently synthesized and has impacted the igm out to at least @xmath0 .
rich clusters of galaxies appear to be closed systems in which most heavy elements are ejected from galaxies .
more supernovae are required than the yield of core - collapse sne from a salpeter imf .
x - ray observations imply that the igm in groups and clusters as been strongly preheated .
sne probably can not supply the required energy , which must come from agn . |
the measurement of the temperature dependence of the rate constant is one of the important tools of chemical kinetics in determining mechanisms of chemical reactions.@xcite significant deviations from a simple exponential behavior can be evidence of tunneling and of other nuclear quantum effects.@xcite these effects are particularly strong for hydrogen transfer reactions with a high activation barrier or at low temperatures .
recently , however , quantum effects have been observed also in many enzymatic reactions at physiological temperatures.@xcite it therefore becomes more and more important to have accurate theoretical methods for computing the temperature dependence of the rate constant.@xcite probably the oldest yet still the best known expression for the thermal rate constant @xmath6 at temperature @xmath7 is the empirical arrhenius law,@xcite@xmath8 here @xmath9 is the boltzmann constant , @xmath10 the activation energy , and the temperature dependence is purely exponential .
an improvement over the arrhenius law was provided by the transition state theory ( tst),@xcite in which @xmath11 where @xmath12 is the planck s constant , @xmath13 and @xmath14 are the partition functions of the transition state and the reactants , respectively , and @xmath15 is the barrier height for the reaction . here
the temperature dependence includes a fractional power @xmath16 in addition to the exponential .
nevertheless , both arrhenius law and tst are basically purely classical , so they can not take into account tunneling and other nuclear quantum effects .
although the simplest quantum effects _ can _ be taken into account within the tst in an _ ad hoc _ fashion , by replacing the partition functions by their quantum analogs for the simple harmonic oscillator ( which takes into account the zero - point energy , the wigner tunneling correction,@xcite and approximate quantization of the vibrational motion ) , a more systematic approach requires quantum treatment of the nuclear motion .
this is of course extremely difficult , and therefore various approximate yet accurate methods have been developed .
these include , e.g. , the semiclassical methods@xcite or the so - called quantum transition state theories.@xcite in this paper , we evaluate the temperature dependence of the rate constant starting from the quantum instanton ( qi ) approximation.@xcite this quantum transition state theory has been shown to describe correctly not only all of the above - mentioned quantum effects , but also corner - cutting , coupling of vibrational and rotational motions , multiple tunneling paths , etc . as we evaluate the temperature dependence of the rate constant directly , we can speed up the qi calculation significantly by avoiding the tedious umbrella sampling necessary for computing the rate constant itself . furthermore ,
if it is only the temperature dependence of the rate constant that is needed , we can increase the accuracy of the qi approximation for the rate constant by canceling some small remaining systematic errors .
this can be useful , e.g. , if we know @xmath17 at a temperature @xmath18 very accurately and would like to know @xmath6 at other temperatures .
the temperature dependence of the rate constant is computed via a thermodynamic integration @xcite with respect to the inverse temperature .
a similar thermodynamic integration in the framework of the qi model was used by ceotto and miller to compute the rate constant for several one and two - dimensional systems using a discrete variable representation.@xcite below , we develop a general thermodynamic integration procedure based on the path integral implementation , which is suitable for many - dimensional systems .
the method is tested on two simple systems for which exact quantum dynamical calculations are feasible : the eckart barrier and the full nine - dimensional h @xmath0h@xmath1 h@xmath2 h reaction .
while the former system is the simplest one - dimensional model of a bimolecular reaction , the latter is the simplest bimolecular chemical reaction with an energy barrier . as such ,
it has been widely studied and attained the status of a benchmark reaction.@xcite despite the apparent simplicity , this reaction remains a challenging test for new approximations .
this is due to the presence of strong quantum effects , as the lightest atoms are involved in both bond breaking and bond formation .
this reaction was investigated not only for the temperature dependence of its rate constant , but also for the kinetic isotope effect,@xcite the presence of the geometric phase effect,@xcite etc .
the remainder of this paper is organized as follows : section ii describes the methodology , i.e. , the qi approximation , the thermodynamic integration with respect to the inverse temperature , the path integral formalism , and the relevant estimators . computational details and ,
in particular , the analysis of various errors are presented in sec .
section iv contains the results .
first , it is shown how the low and high temperature limits are obtained .
the numerical results for the eckart barrier and the h@xmath19h@xmath20 reaction are then presented .
the results are compared with the arrhenius law , tst , tst with the wigner tunneling correction , and the exact quantum calculation .
finally , we discuss how the efficiency is related to the dependence of the statistical error on the number of imaginary time slices in the path integral , and how this error , in turn , depends on the system under study .
section v concludes the paper .
the qi approximation for thermal rate constants was introduced in ref . .
the most direct derivation @xcite starts from the exact miller - schwartz - tromp formula for the rate constant,@xcite@xmath21 where @xmath22 is the symmetrized flux - flux correlation function,@xmath23 with hamiltonian operator @xmath24 , inverse temperature @xmath25 , time @xmath26 , and @xmath27 the flux operator through the dividing surface @xmath28 .
a stationary - phase approximation applied to eq .
( [ miller_schwartz_tromp ] ) yields the qi approximation for the rate constant,@xcite @xmath29 where @xmath30 is a specific type of energy variance,@xcite
@xmath31 ^{1/2}. \label{dh}\ ] ] the delta - delta correlation function used above is defined as @xmath32 where the generalized delta function operator is given by @xmath33 \equiv\delta\left [ \xi\left ( \mathbf{\hat{r}}\right ) -\xi_{\gamma}\right ] m^{-1/2}\left\vert \nabla\xi\right\vert \label{delta_op}\ ] ] and @xmath34 is the reaction coordinate such that @xmath35 at the transition state .
similarly , @xmath36 defines the position of the dividing surface @xmath37 .
we have used mass - scaled coordinates in which all degrees of freedom have the same mass @xmath38 . in practice , the exact delta function constraint is approximated by a gaussian constraint corresponding to a harmonic constraint potential @xmath39,@xcite@xmath40 & \approx \sqrt{\frac{\beta}{2\pi \sigma^2 } } e^{-\beta v_{\text{constr}}(\mathbf{r})},\label{gauss}\\ v_{\text{constr}}(\mathbf{r } ) & = \frac{1}{2}\left ( \frac{\xi(\mathbf{r})-\xi_{\gamma}}{\sigma}\right ) ^{2}. \label{v_constr}\ ] ] the accuracy of the qi approximation has been already verified in numerous applications.@xcite the main shortcoming of the qi method is the neglect of recrossing which is however , neglected in any quantum or classical transition state theories .
the recrossing effects on the quantum instanton rate constant have been quantified for several collinear reactions by ceotto and miller.@xcite fortunately , the recrossing effects become generally less important in higher dimensions . the goal of this paper is to compute the temperature dependence of the rate constant , i.e. , the ratio @xmath41 . within the qi approximation ,
this ratio is given by @xmath42 where we multiplied and divided the numerator and denominator by @xmath43 .
in this expression , quantities @xmath30 and @xmath44 can be computed directly by the metropolis monte - carlo procedure because they are thermodynamic averages . on the other hand , the ratios @xmath45 and @xmath46
can not be computed this way since they involve ratios of quantities at different temperatures .
these ratios can , however , be calculated by the method of _ thermodynamic integration _
@xcite with respect to the inverse temperature @xmath47,@xmath48 , \label{ti_qr}\\ \frac{c_{\text{dd}}\left ( \beta_{2},0\right ) } { c_{\text{dd}}\left ( \beta _ { 1},0\right ) } & = \exp\left [ -\int_{\beta_{1}}^{\beta_{2}}e^{\ddag}(\beta)d\beta\right ] , \label{ti_cdd}\ ] ] where @xmath49 and @xmath50 are the thermal energies of the reactants and of the transition state , respectively .
a similar thermodynamic integration was used within a discrete variable representation of the qi approximation to compute the rate constant for several collinear triatomic reactions.@xcite unlike @xmath51 and @xmath52 , the energies are normalized quantities because they can be written as logarithmic derivatives:@xmath53 hence they can be computed directly by a monte carlo procedure .
quantum thermodynamic effects can be treated rigorously using the imaginary time path integral ( pi).@xcite let @xmath54 be the number of degrees of freedom ( @xmath55 for the eckart barrier and @xmath56 for the h@xmath57 potential ) and @xmath58 the number of imaginary time slices in the pi . the pi representations of the partition function@xcite and the delta - delta correlation function@xcite are@xmath59 , \label{pi_qr}\\ c_{\text{dd}}^{p}\left(\beta , 0\right ) & = c\int d\mathbf{r}^{\left ( 1\right ) } \cdots\int d\mathbf{r}^{\left ( p\right ) } \delta\left [ \xi_{a}\left ( \mathbf{r}^{\left ( 0\right ) } \right ) \right ]
\nonumber \\ & \times \delta\left [ \xi_{b}\left ( \mathbf{r}^{\left ( p/2\right ) } \right ) \right ] \exp\left [ -\beta \phi\left ( \left\ { \mathbf{r}^{\left ( s\right ) } \right\ } \right ) \right ] , \label{pi_cdd}\ ] ] where @xmath60^{dp/2}$ ] , @xmath61 is a @xmath54-dimensional vector representing the @xmath62th time slice , and the effective potential @xmath63 is given by@xmath64 for @xmath65 , the above expressions reproduce classical statistical mechanics , while exact quantum statistics is reached in the limit @xmath66 . in practice , there are two main strategies for evaluating thermodynamic averages using the pi : the pi molecular dynamics ( pimd)@xcite or pi monte carlo ( pimc).@xcite we use the pimc procedure together with the metropolis algorithm .
the basic idea is to sample the pi configuration space according to an appropriate weight @xmath67 , which is , e.g. , for @xmath52 given by @xmath68 \delta\left [ \xi_{b}\left ( \mathbf{r}^{\left ( p/2\right ) } \right ) \right ] \exp\left [ -\beta\phi\left ( \left\ { \mathbf{r}^{\left ( s\right ) } \right\ } \right ) \right ] , \label{rho_ts}\ ] ] and then , at each sampled configuration , to evaluate the so - called estimator @xmath69 of the relevant physical quantity @xmath70 .
the final estimate of @xmath70 is given by the average @xmath71 along the pimc trajectory .
using the pi representation of @xmath51 and @xmath52 , one can obtain estimators for all quantities needed in eq .
( [ qi_tdep ] ) , i.e. , @xmath72 , @xmath73 , and the logarithmic derivatives of @xmath74 , @xmath52 ( i.e. , the energies @xmath49 , @xmath50 ) .
those for @xmath72 and @xmath73 are listed in ref . .
the simplest estimator for the energy @xmath49 , the so - called barker or thermodynamic estimator ( te),@xcite can be derived directly from eq .
( [ er_logder ] ) and the pi expression ( [ pi_qr ] ) , giving@xmath75 as observed by herman _
et al._,@xcite the te can have a large statistical error , which can be avoided with the so - called virial ( ve)@xcite or centroid virial ( cve),@xcite estimators . invoking the virial theorem ,
the kinetic energy in these two estimators is replaced by an expression involving the gradient of the potential energy.@xcite this is convenient in the pimd implementations since the gradient is already available . in pimc simulations , however , only the potential is needed for the random walk , and in order to avoid computing the gradients , alternative approaches have been proposed .
one can , e.g. , employ the centroid thermodynamic estimator@xcite or more generally , use a procedure based on rescaling coordinates@xcite in which the gradients of the potential are replaced by a single derivative that can be evaluated by finite difference.@xcite variants of the latter approach have been applied successfully to compute thermal energies and heat capacities,@xcite kinetic isotope effects@xcite , equilibrium isotope effects@xcite , or the derivatives of the flux - flux correlation function@xcite needed in the generalized qi model.@xcite the ve for @xmath49 can be derived most directly by the change of coordinates @xmath76 in the pi ( [ pi_qr ] ) , yielding@xmath77}{d\delta\beta } \right\ } \nonumber \\ & = \frac{1}{p } \sum_{s=1}^{p } \left\ { v(\mathbf{r}^{(s ) } ) + \frac{dv[(1+q)^{1/2}\mathbf{r}^{(s)}]}{dq } \right\ } , \label{ev}\ ] ] where @xmath78 is a small dimensionless parameter and the @xmath78-derivative is evaluated by finite difference at @xmath79 .
similarly , the cve can be obtained by the change of variables @xmath80 , where one first subtracts the so - called centroid coordinate @xmath81 .
the resulting estimator is@xmath82}{dq } \right\}. \nonumber \ ] ] in the case of constrained simulations near the transition state , the constrained weight function ( [ rho_ts ] ) can be approximated by using the gaussian approximation of the delta function from eqs .
( [ gauss])-([v_constr ] ) .
besides a prefactor , this amounts to adding a constraint potential@xmath83 to the effective potential @xmath63 .
assuming that @xmath84 is independent of temperature and following a derivation similar to that for estimators of @xmath49 , one obtains the thermodynamic , virial , and centroid virial estimators for @xmath85@xmath86 } { dq},\label{e_ts_ve_rig}\\ e_{\text{cve}}^{\ddag , p } & = e_{r,\text{cve}}^{p}-\frac{1}{\beta}+\phi_{\text{constr}}(\{\mathbf{r}^{(s)}\ } ) \nonumber \\ & + \frac{d\phi_{\text{constr}}[\{\mathbf{r}^{(c)}+(1+q)^{1/2}(\mathbf{r}^{(s)}-\mathbf{r}^{(c)})\}]}{dq}. \label{e_ts_cve_rig}\ ] ] although the above estimators converge to the exact results , we found that the statistical errors can be decreased slightly by employing an alternative set of estimators , derived using an exact relation@xmath87 which is valid for a harmonic constraint potential for any value of @xmath58 .
the new estimators are@xmath88}{dq } , \label{e_ts_ve_adh}\\ e_{\text{cve}}^{\ddag , p } & = e_{r,\text{cve}}^{p}+ \frac{d\phi _ { \text{constr}}[\{\mathbf{r}^{(c)}+(1+q)^{1/2}(\mathbf{r}^{(s)}-\mathbf{r}^{(c)})\}]}{dq}. \label{e_ts_cve_adh}\ ] ] estimators ( [ e_ts_te_adh])-([e_ts_cve_adh ] ) are in a way more intuitive than estimators ( [ e_ts_te_rig])-([e_ts_cve_rig ] ) : in the limit of a sharp constraint , the constrained energy should be independent of the type of constraint . it should be stressed that the last terms in the ve and cve in eqs .
( [ e_ts_ve_rig ] ) , ( [ e_ts_cve_rig ] ) , ( [ e_ts_ve_adh ] ) , and ( [ e_ts_cve_adh ] ) are important ; without them the agreement among the te , ve , and cve is lost . in other words ,
an intuitive guess such as @xmath89 would not give a correct answer for the constrained energy .
finally , we also tested a constraint potential that is proportional to temperature , i.e.,@xmath90 where @xmath91 is a harmonic potential independent of temperature .
as a result , the constraint ( [ gauss ] ) itself is actually independent of temperature .
following again a derivation similar to that for estimators of @xmath49 , one obtains another set of the te , ve , and cve for @xmath85 that look exactly like the estimators ( [ e_ts_te_adh])-([e_ts_cve_adh ] ) for @xmath84 independent of temperature .
the only difference is that the random walk is done with a different constraint .
all calculations were performed with a pimc code implemented in fortran 90 .
sampling of the configurational space in the pimc simulation was done using three types of moves .
staging algorithm @xcite was employed to move all unconstrained beads .
constrained beads , i.e. , beads @xmath92 and @xmath93 which feel the constraint potential @xmath84 , were sampled with the free particle single slice algorithm.@xcite finally , whole chain moves@xcite were used to speed up sampling of the potential energy surface .
the gaussian constraint potential must be strong enough in order to exert the constraining effect on the system . when this condition was satisfied , the converged results were independent of the constraint .
however , the statistical root mean square error ( rmse ) of the transition state energy @xmath50 increases with the strength of the constraint because sampling of the configuration space becomes more difficult
. therefore the selected strength of the constraint should take into account these two effects .
we have used @xmath94 a.u . in both systems .
all quantities needed in the ratio ( [ qi_tdep ] ) were evaluated using the above mentioned estimators .
the thermodynamic integrations ( [ ti_qr ] ) and ( [ ti_cdd ] ) were evaluated with the simpson rule using 25 values of @xmath47 between @xmath95 and @xmath96 with the reference temperature @xmath97 and the minimum temperature @xmath98 .
the number of beads was chosen to be inversely proportional to the temperature , with the maximal number of beads ( used for @xmath98 ) being @xmath99 for the eckart barrier and @xmath100 for the h@xmath19h@xmath20 reaction .
the error of the final result consists of four main error contributions : a ) the statistical error due to the monte carlo simulation , b ) the error due to the discretization of the ti , c ) the error due to the discretization of the pi ( i.e. , the finite @xmath58 error ) , and d ) the actual error of the qi approximation .
we have carefully separated these four contributions and attempted to make the first three contributions small in comparison with the error of the qi . in more complicated systems , this may not be possible and especially the final statistical error may be comparable to or larger than the error of the qi . because the exponentiation of the ti is quite sensitive to various errors , a detailed analysis of errors
was carried out for the ratio @xmath101 , i.e. , over the largest temperature range , where the first three types of errors are the greatest .
the ti was evaluated by four different numerical methods , namely the trapezoidal , simpson , simpson @xmath102 , and boole methods.@xcite comparing the analytical bounds on the _ discretization errors _ of the ti integrals using a numerical estimate of a higher order derivative,@xcite one can conclude that both the simpson and simpson 3/8 methods were much better than the trapezoidal rule and that the boole method did not provide any further improvement . specifically , for the eckart barrier the error of the ratio due to the discretization of the ti was @xmath103 , @xmath104 , @xmath105 , or @xmath106 for the trapezoidal , simpson , simpson @xmath102 , or boole methods , respectively . for the h+h@xmath20 reaction , the discretization error of the final ratio was @xmath107 , @xmath108 , @xmath109 , or @xmath108 , in the same order .
it should be emphasized that these error estimates are very conservative , as the actual difference between the final ratios based on different methods was an order of magnitude smaller than what one would expect from the error estimates .
the final results displayed in the plots used the simpson method .
the _ statistical _ rmses were estimated with the block averaging method using a variable block size @xcite to remove correlation of the pimc data .
the statistical error of the ti was evaluated using an appropriate formula for each integration method and assuming that the statistical errors of energies at different temperatures were uncorrelated .
as expected , the statistical error did not depend much on the integration method , and was always close to the statistical error for the simpson method .
the statistical error of the final ratio was @xmath108 for the eckart barrier and @xmath110 for the h+h@xmath20 reaction .
the _ finite _ @xmath58 _ error _ , i.e. , the error due to the discretization of the feynman pi , was obtained by repeating calculations of all quantities at all temperatures with twice smaller numbers of beads ( @xmath111 ) and then extrapolating each quantity to @xmath66 , assuming @xmath112 convergence .
we emphasize that we used the extrapolated results only for estimating the finite @xmath58 error of the computed ratio and _ not _ for estimating the ratio itself , which could be dangerous .
the finite @xmath58 error of the ratio was @xmath113 for the eckart barrier and @xmath114 for the h+h@xmath20 reaction .
[ we note that for h+h@xmath20 one of the temperatures ( @xmath115 , a temperature in the vicinity of which a sharp bend in the @xmath116 dependence occurs ) required a five times longer simulation to reduce the ti discretization errors .
this was because a small statistical error had a huge effect on the estimate of the fourth derivative and hence on the analytical estimate of the discretization error . ] to sum up , in both systems the ti discretization error was negligible to the statistical and finite @xmath58 errors , which , in turn , were small in comparison to the error of the qi approximation .
at high temperatures , the rate constant is expected to behave classically and follow the arrhenius law or the more accurate tst result . whereas the arrhenius law ( [ k_arrh ] ) predicts the rate constant ratio to be a simple exponential function of the inverse temperature , @xmath117 tst ( [ k_tst ] ) gives the ratio of rate constants @xmath118 in particular , assuming the partition functions @xmath119 and @xmath51 to be separable into products of classical rotational and vibrational partition functions , the temperature dependence ( [ tdep_tst ] ) of tst rate constant includes an additional fractional power law besides the exponential dependence in the arrhenius law ( [ tdep_arrh ] ) . at somewhat lower temperatures , when quantum effects start to play a role , the basic tst expression ( [ tdep_tst ] ) can be improved in several ways : first , classical partition functions @xmath51 and @xmath120 can be replaced by their exact quantum analogs for a harmonic potential .
second , quantum tunneling can be included approximately via the wigner tunneling correction.@xcite this method corrects the rate constant with a multiplicative factor@xmath121 where @xmath122 is the imaginary frequency of the asymmetric stretch along the reaction coordinate .
the correction can be derived by treating the motion through the transition state as a vibration on an upside down potential and expanding the quantum partition function to second order in @xmath47 .
although an improvement over tst , the wigner tunneling correction can not describe multidimensional tunneling . a simple model of an activated chemical reaction is provided by the eckart barrier , a one - dimensional system described by the potential @xmath123 ^{-2}.\ ] ] we use standard parameter values @xmath124 ,
@xmath125 , mass @xmath126 and reaction coordinate @xmath127 .
the exact quantum rate constant @xmath128 for this reaction can be obtained by integrating the exact quantum mechanical cumulative reaction probability , which is known analytically.@xcite figure [ fig : eckart ] ( a ) compares the qi results with the exact qm results , tst ( which is equal to the arrhenius law here ) , and the tst including the wigner tunneling correction .
the reference temperature is @xmath129 and the plot shows ratios for temperatures down to @xmath130 . since classical recrossing does not occur for the eckart barrier , all tsts should converge to the correct quantum results at high temperatures .
the figure confirms that this is indeed the case : note that all curves are tangent at the high temperature limit . at low temperatures ,
one reaches the quantum regime where tunneling is important and consequently the arrhenius plot of the exact qm result has a large curvature . while tst has a huge error , the qi approximation agrees very well with the qm result .
note that the wigner tunneling correction improves over the tst and captures the tunneling effect partially but still fails to recover the curvature of the exact result . .
] figure [ fig : eckart ] ( b ) shows the relative error of the rate constant ratio for the different methods . whereas both tst and tst with the wigner tunneling correction deteriorate rapidly with decreasing temperature , the qi method has an error below @xmath131 for all temperatures above @xmath132 .
the qi approximation has a significant error ( @xmath133 ) only at very low temperatures , below @xmath134 .
however , this error was well understood already in the original paper by miller et al . @xcite and can be remedied by considering two separate dividing surfaces at very low temperatures .
( here we have used a single dividing surface at all temperatures for simplicity . ) the temperature dependence of the reactant and transition state energies is shown in fig .
[ fig : eckart_ti ] . while both curves are quite smooth , small discretization errors in the integrals can have large effects on the exponentiated result . by a detailed error analysis described in sec .
iii , we found that the simpson method was sufficient for the ti over the whole temperature range .
note that the ve for @xmath135 gives zero , but can be easily corrected with an analytical correction @xmath136 .
the three different estimators for the constrained energy @xmath50 at @xmath137 are compared in figs .
[ fig : eckart_evsp_adhoc ] and [ fig : eckart_evsp_rigorous ] .
panel ( a ) of fig .
[ fig : eckart_evsp_adhoc ] , which uses the simpler estimators ( [ e_ts_te_adh])-([e_ts_cve_adh ] ) , shows that the te , ve , and cve agree for all examined values of @xmath58 and , in particular , converge to the same value for @xmath66 .
the three estimators , however , differ in their statistical convergence .
unlike for the unconstrained result , where the cve is the optimal estimator , for the constrained energy , the optimal estimator is the ve .
this can be clearly seen in fig .
[ fig : eckart_evsp_adhoc ] ( b ) which shows the rmses of the different estimators for different values of @xmath58 . while the rmse of the te increases with @xmath58 , the rmses of the ve and cve remain approximately constant as a function of @xmath58 , with the ve having a much smaller statistical error .
assuming that the desired convergence is achieved for @xmath138 , the speedup factor achieved by using the ve compared to the te and cve is approximately @xmath139 and @xmath140 respectively .
it is clear from the figure that both the speedup factor and the best estimator depend on @xmath58 and hence on the temperature . .
the constraint potential is independent of temperature and estimators ( [ e_ts_te_adh])-([e_ts_cve_adh ] ) are used . ]
figure [ fig : eckart_evsp_rigorous ] shows the same results , but computed with the estimators ( [ e_ts_te_rig])-([e_ts_cve_rig ] ) .
the statistical errors are very similar , although for the ve slightly larger than those in fig .
[ fig : eckart_evsp_adhoc ] .
the temperature dependence of the rate constant of the @xmath141 reaction was studied on the boothroyd - keogh - martin - peterson ( bkmp2 ) reactive potential energy surface.@xcite the classical transition state of this system has a collinear configuration with equal bond lengths @xmath142 . a suitable reaction coordinate is therefore given by the difference of the bond lengths , @xmath143 figure [ fig : h3 ] ( a ) shows the temperature dependence of the rate constant in the range from @xmath144 to @xmath4 .
the exact qm results are from ref . .
at high temperatures the tst curve is tangent to the exact qm curve , but at low temperatures , there is a significant discrepancy even for the tst with the wigner tunneling correction . on the other hand
, the qi approximation agrees very well with the exact qm result all the way to @xmath144 .
the relative error of the rate constant ratio is shown in fig . [
fig : h3 ] ( b ) which confirms that the error of the qi approach is within @xmath5 in the full temperature range whereas all other approximations have huge errors already for temperatures as high as @xmath145 .
h@xmath146h@xmath147h reaction .
( a ) temperature dependence of the rate constant .
( b ) temperature dependence of the relative error of the ratio @xmath148 . ] in case of the @xmath141 reaction , the ve had to be corrected for both @xmath135 and @xmath50 calculations .
the reason is that the virial theorem only holds for bound systems .
the transition state of the @xmath149 system can translate freely as a whole and the three translational degrees of freedom yield a correction of @xmath150 to the ve . in the reactant region ,
both the @xmath151 atom and @xmath152 molecule can move freely and the six translational degrees of freedom give a correction of @xmath153 to the ve .
the temperature dependence of the reactant and transition state energies is shown in fig .
[ fig : h3_ti ] . while both curves are quite smooth
, small discretization errors in the integrals can have large effects on the exponentiated result . by an error analysis described in sec .
iii , we found that the simpson method was sufficient for the ti over the whole temperature range .
h@xmath146h@xmath147h reaction .
temperature dependence of the reactant ( a ) and transition state ( b ) energies . ] figures [ fig : h3_evsp_adhoc ] and [ fig : h3_evsp_vconstr_propto_t ] show how the constraint energy @xmath50 and the rmse of @xmath50 depend on @xmath58 for @xmath137 .
panel ( a ) of fig .
[ fig : h3_evsp_adhoc ] , which uses the estimators ( [ e_ts_te_adh])-([e_ts_cve_adh ] ) and a constraint potential independent of temperature , shows again that the te , corrected ve , and cve give approximately the same results for all values of @xmath58 and , within a statistical error , converge to the same limiting value for @xmath66 . panel ( b ) of fig .
[ fig : h3_evsp_adhoc ] shows that while the statistical error of the cve is approximately constant as a function of @xmath58 , the rmses of the te and ve grow with @xmath58 .
however , in this case , the results are quite well converged for @xmath154 and at this point the rmse of the cve is still larger than the rmse of the te , although it is already smaller than the rmse of the ve .
while for lower temperatures where larger values of @xmath58 are needed , cve would eventually become the optimal estimator , it is not so for @xmath137 .
the growth of the rmse of the ve with @xmath58 is due to the fact that unlike for the eckart barrier , the transition state of the h@xmath57 system can move freely as a whole .
h@xmath146h@xmath147h reaction at @xmath137 .
the constraint potential is independent of temperature and estimators ( [ e_ts_te_adh])-([e_ts_cve_adh ] ) are used . ] finally , fig .
[ fig : h3_evsp_vconstr_propto_t ] shows analogous results , still using estimators ( [ e_ts_te_adh])-([e_ts_cve_adh ] ) , but obtained with a constraint potential ( [ v_constr_propto_t ] ) proportional to @xmath7 ( chosen such that the two types of constraints coincide for @xmath137 ) .
as expected , the statistical errors of the te , ve , and cve are similar to those in fig . [
fig : h3_evsp_adhoc ] [ obtained with the constraint potential ( [ v_constr ] ) independent of @xmath7 ] .
h@xmath156h@xmath147h reaction at @xmath137 .
the constraint potential is proportional to temperature and estimators ( [ e_ts_te_adh])-([e_ts_cve_adh ] ) are used . ]
a general method for the direct evaluation of the temperature dependence of the quantum rate constant was presented .
the main advantage of this method is the increased efficiency : evaluating the temperature dependence directly , without computing the rate constant at any given temperature , allows us to avoid a tedious umbrella sampling procedure .
besides efficiency , the direct calculation of the temperature dependence of the rate constant can also improve the accuracy : our ratios @xmath157 for both the eckart barrier and the @xmath158 reaction have somewhat smaller relative errors than the errors obtained for the absolute qi rate constants in previous studies of these systems.@xcite the smaller relative error in the ratio of rate constants is due to a favorable cancellation of various systematic errors , such as the systematic error of about @xmath159 of the qi model at high temperatures ( that can also be removed by an ad hoc correction of @xmath72)@xcite and small recrossing effects in the @xmath141 reaction , also at high temperatures .
it is noteworthy that for both reactions , the rmses of transition state energies depend on the strength of the constraint . weakening the constraint
facilitates sampling of the configuration space and the error of the cve decreases , approaching the well - known unconstrained situation where the cve is typically the optimal estimator .
however , at the same time the constraint must be strong enough to exert the constraining effect and describe the situation near the transition state properly . as a result ,
`` ranking '' of the estimators is not universal but can change with the potential used as a constraint and is in general different from the ranking for unconstrained simulations .
the dependence of the error of the ve on @xmath58 is best understood in terms of the ring polymer interpretation@xcite of the discretized pi : the quantum thermodynamics of the original system can be interpreted as the classical thermodynamics of the ring polymer .
the constrained pi simulation for the one - dimensional eckart barrier is completely bound , resulting in the rmse independent of @xmath58 . in the full - dimensional hydrogen exchange reaction , on the other hand , even the constrained simulation allows the system to move as a whole .
this is exactly where the ve is known to have a rmse increasing with @xmath58 .
the cve estimator is usually the optimal estimator in unconstrained systems with some translational ( i.e. , free - particle ) degrees of freedom . in a system in which only two slices are bound ( in our case ,
slices @xmath160 and @xmath58 ) , the symmetry between different slices is lost and so is , to some extent , the advantage of subtracting the centroid .
this explains why the rmse of the ve can sometimes be smaller than the rmse of the cve for all values of @xmath58 , which we observed in figs .
[ fig : eckart_evsp_adhoc ] and [ fig : eckart_evsp_rigorous ] . to sum up , while in generic systems at very low temperatures , the cve is expected to be the optimal estimator for energy , at finite temperatures in constrained simulations , the ve or even the te can have the smallest rmse .
the results obtained in this paper can serve as a guide for choosing the best estimator for a given system .
however , since the additional cost of evaluating all three estimators is negligible in comparison to the cost of the pimc random walk or pimd simulation , we recommend computing all three estimators , evaluating their rmses , and using the one with the smallest rmse in a given situation .
this research was supported by the swiss nsf ( grant no . @xmath161 ) and by the epfl .
authors thank tom zimmermann for helpful discussions . | a general method for the direct evaluation of the temperature dependence of the quantum - mechanical reaction rate constant in many - dimensional systems is described .
the method is based on the quantum instanton approximation for the rate constant , thermodynamic integration with respect to the inverse temperature , and the path integral monte carlo evaluation .
it can describe deviations from the arrhenius law due to the coupling of rotations and vibrations , zero - point energy , tunneling , corner - cutting , and other nuclear quantum effects .
the method is tested on the eckart barrier and the full - dimensional h @xmath0h@xmath1 h@xmath2 h reaction . in the temperature range from @xmath3 to @xmath4 ,
the error of the present method remains within @xmath5 despite the very large deviations from the arrhenius law .
the direct approach makes the calculations much more efficient , and the efficiency is increased even further ( by up to two orders of magnitude in the studied reactions ) by using optimal estimators for reactant and transition state thermal energies .
which of the estimators is optimal , however , depends on the system and the strength of constraint in a constrained simulation . |
in 1936 birkhoff and von neumann @xcite have shown the existence of a propositional calculus as fundamental ingredient of quantum mechanics ( qm ) , which could be written using only the outputs of measures .
it does not assume any set of numbers or even a particular vectorial space , but contains the essentials of qm such as uncertainty relations and complementary properties .
of course , the authors showed that there are three different realizations for this propositional calculus , corresponding to the real or complex numbers or still quaternions .
octonions and higher dimensional extensions of the complex numbers are discarded , since they can not have a conservation law for the probability current @xcite .
we can ask : which of these three realizations of the general qm of birkhoff and von neumann is present in nature ?
here it is implicit the hypothesis that the set of numbers of a given theory reflects part of the physical information about the system . while the differences between the real and complex qm are relatively simple and well known @xcite , the quaternionic version has many new and rich characteristics . therefore , it sounds strange that such possibility is not much explored , but there are very good reasons for this .
first , the problem of writing a quaternionic schrdinger equation is not trivial since it involves the explicit use of imaginary unit .
second , the representation of composite systems by a direct product is more difficult due to the noncommutativity of the quaternionic valued wave functions . here , we implement a quaternionic version of schwinger s measurement algebra and build the dynamics based on the action principle . in each step
, the analogy with the usual qm is used as inspiration , but the peculiarities emerging from the quaternionic noncommutativity are always emphasized .
the theory constructed by this means is quite distinct from adler s approach @xcite , having similarities with the work of finkelstein , jauch , schiminovich and speiser , @xcite .
the classical theory of physical measurements is based on the concept that the interaction between the system under observation and the measurement apparatus can be done arbitrarily small or , at least , precisely compensated , in such way to specify an idealized measurement which does not disturb any other property of the system .
however , the experiment had demonstrated that the interaction can not be done arbitrarily small neither the disturb produced can be precisely compensated since it is uncontrollable and unpredictable .
the fact that the interaction can not be arbitrarily small is expressed by the finite size of the planck constant , while the uncontrollable character of the interaction is given by the uncertainty principle .
therefore , the measurement of a given property can produce a significant change in the value of another previously measured property , and then there is no sense in speaking about an microscopic system with definite values for all its attributes .
this is in contradiction with the classical representation of physical quantities by numbers
. the laws of a microscopic physical system must then be expressed in a non - classical mathematical language constituting a symbolic expression of the properties of microscopic measurements . in what follows
, we will develop the general lines of such mathematical structure discussing about simplified physical systems where any physical quantity @xmath0 can have only a finite number of different values @xmath1 ... .
the most simple measurement consider an ensemble of similar independent systems which is divided by the apparatus of measurement in sub - ensembles distinguished by the defined values of the physical quantity under measurement .
let us denote @xmath2 the selective measurement accepting any system having value @xmath3 for the property @xmath0 and rejecting any other .
the addition of such symbols is defined as implying a less specific measure , resulting in a sub - ensemble associated with any value under the sum , none of them being distinguished of the others by the measurement .
the multiplication of measurement symbols implies the sequence of measurements reading from right to left . from the physical meaning of such operations
, we learn that addition is commutative and associative while multiplication is only associative . using @xmath4 and @xmath5 to represent respectively the measures which accept and reject all systems , the properties of the elementary selective measurement
are given by @xmath6 from the meaning of the measurements represented by @xmath4 and @xmath5 we directly read the following algebraic properties : @xmath7 what justifies the adopted notation .
the algebraic properties of @xmath4 , @xmath8 and @xmath2 are consistent provided that the multiplication be distributive , @xmath9 all laws of multiplication for measurement symbols given above can be combined in a single expression , @xmath10 with the introduction of the symbol @xmath11 known as _ _
kronecker s _ _ _ delta_. from these definitions one sees that the measurement symbols belong to a noncommutative ring @xcite .
two quantities @xmath12 and @xmath13 are compatible when the measurement of one of them does not destroy the knowledgement of a previous measurement of the other . the selective measures @xmath14 and @xmath15 , taken in this order , produce an ensemble where it is possible , simultaneously , to attribute the values @xmath16 to @xmath12 and @xmath17 to @xmath13 . the symbol for such composite measurement is @xmath18 from such definition it is easy to see that the compatibility is an equivalence relation .
a complete set @xmath0 of compatible quantities @xmath19 means that any pair of such properties is compatible and there is no other compatible quantity outside the set , except the functions constructed from the set @xmath0 . in fact , @xmath0 is an _
equivalence class_. the measurement symbol @xmath20 describes a complete measurement where the selected systems have definite values for the maximum number of possible attributes .
any tentative for determining the value of another independent physical quantity will produce uncontrollable changes on the previously measured values .
therefore , the optimum information about a given system is achieved making a complete selective measurement .
the systems accepted by the complete selective measurement @xmath2 are known to being in the state @xmath3 .
the symbolic properties for the complete measures are the same as for the elementary selective measurements , i.e. , ( [ medida1 ] ) , ( [ medida2 ] ) and ( medida3 ) .
a more general kind of measure incorporates a change on the state of the system .
the symbol @xmath21 represents a complete selective measurement which accepts systems in the @xmath16 state and let out systems in the state @xmath3 .
the measurement process @xmath2 is the special case when no change on the state occurs,@xmath22 the properties of successive measurements of this specie are given by @xmath23 since if @xmath24 the second stage of the apparatus does not select any system emerging from the first one , and if @xmath25 all systems coming from the first stage are accepted by the second , being the composite measurement a selection of systems in the state @xmath26 and letting it out in the state @xmath16 .
observe that if we interchange both stages , then @xmath27 what is not the same as ( [ medidaprod ] ) .
therefore , we realize that the multiplication of complete measurements symbols is noncommutative .
the physical quantities belonging to a complete set do not exhaust the totality of physical attributes in a system .
one can form others complete sets @xmath28 ... , which are mutually incompatible and , for each choice of non - interfering physical characteristics , there is a set of selective measurements concerning to systems in the appropriate states @xmath29 ... the most general selective measurement links two complete sets of incompatible properties .
let @xmath30 be the measurement process rejecting all systems which are not in the state @xmath31 and allowing to emerge only systems in the state @xmath3 .
the composite measurement @xmath32 will select systems in the state @xmath33 and let them in the state @xmath3 , so it should be proportional to the selective measurement @xmath34 .
the examples considered until now include the passing of all or none system through both stages , as realized by the symbols @xmath4 and @xmath5 .
notwithstanding , in general we can just admit that measures of the property @xmath35 upon a system in the state @xmath36 , which belongs to a complete set incompatible with @xmath37 , will furnishes an statistical distribution of all possible results .
so , only a fraction of the systems emerging from the first stage is accepted by the second one .
we can express this by the general multiplication law : @xmath38 where @xmath39 is a number characterizing the statistical relationship between the states @xmath31 and @xmath36 .
in particular , @xmath40 where @xmath41 means that @xmath3 and @xmath42 are defined sets of values for the complete set @xmath0 . since that the numbers @xmath43 link the states @xmath3 and @xmath31 they are called _
transformation function_. the measurement symbols @xmath44 equipped with addition and multiplication as defined above and together with the scalar ring @xmath45 form an algebra , which we call the _ measurement algebra_. observe that nothing was said about the particular set of numbers @xmath39 to be adopted .
in fact , as matter for mathematical and physical meaning consistency , it is enough that @xmath45 belongs to an scalar ring .
of course , the order in which the scalars @xmath43appear in the product ( [ prodgeral ] ) is very important , since it reflects on the ring multiplication law , allowing the definition of different measurement algebras .
therefore , the most general form to indicate the multiplication rule for measurement symbols is @xmath46 since it does not make any reference to the order of the scalar on the product .
however , we will maintain the scalars on a preferable central position on the product .
our main interest here is to suppose that the scalars are quaternions and investigate what are the physical implications of such assumption .
the reason to take the scalars on a central multiplicative position comes from the recognition that measurement symbols are in fact projectors on the several possible states of two different complete sets of observables . to reinforce such character ,
we adopt the notation@xmath47 then , the most general way in which a measurement symbol can appear together an scalar is@xmath48 being @xmath49 any element of the ring under which the measurement algebra is defined . as stated before , we will assume that the numbers @xmath49 are _ quaternions _ , defined by @xmath50 when @xmath51 we simply denote @xmath52 .
this notation is useful because it maintains separated in an explicitly way the two parts of the measurement symbol corresponding to the physical hilbert space of states @xmath53 and its dual @xmath54 . in the language of second quantization , this notation directly alludes to the annihilation ( right ) and creation ( left ) processes of particles or field fluctuations involved in a measurement act .
it is important to stand out that since the products of vector _ by _ scalars are defined over a noncommutative ring , these products have sense only a _ definite _ order , which we take as right for the kets @xmath55 and left for the bras @xmath56 , where @xmath57 is the hilbert space of eigenstates of a given complete set of observables .
quaternions are a particular realization of a _ clifford algebra _
lounesto , so a even more general theory can be recognized .
the fundamental transformation law for the measurement symbols is essentially unaffected by the specific choice of the scalar ring . actually ,
using the notation of the previous section , measurement symbols of one kind can be transformed in symbols of another kind : @xmath58 carefully preserving the composition of products , one can interpret this relation as a double mapping of vectors @xmath59 and covectors @xmath60 on the linear combinations @xmath61 and @xmath62 respectively .
therefore , the composition law for transformation functions in a quaternionic ring is @xmath63 from which we obtain the completeness relations@xmath64 however , since quaternions do not commute , the preservation of the number of degrees of freedom imply that@xmath65 except for systems with only one degree of freedom , this does not mean that @xmath66 for any pair of quaternionic transformation functions .
then , the relation ( [ comutq ] ) implies a restriction , but its interpretations is not easy .
one of the most important actions over the measurement algebra is the _ trace functional _ , which associates each element of the algebra to one scalar . since here
the scalar ring is noncommutative , there are three kinds of trace functional called respectively _ left , right _ and _ central _
trace:@xmath67@xmath68@xmath69 in the standard complex case , the trace functional is related to the statistical interpretation of quantum mechanics . here
we have a more complicated situation since none of the above trace functionals has an invariant law of transformation .
nevertheless , the multiplication law is invariant under the following mapping : [ transfsemeq ] @xmath70 where quaternions @xmath71 , @xmath72 are not null . therefore , the transformation function @xmath43 can not itself have a direct physical interpretation , and shall configure in a combination invariant under ( [ transfsemeq ] ) .
the appropriate basement for the statistical interpretation of the transformation function can be inferred from a sequence of elementary measurement , @xmath73 , which differs from @xmath74 only by virtue of the disturbance caused by the intermediary measurement of the attribute @xmath0 .
only a fraction of the systems selected by the initial measurement of @xmath37 is transmitted through the complete set .
hence , we obtain the following symbolic statement : @xmath75 where the number @xmath76 should be invariant under ( [ transfsemeq ] ) .
it means that@xmath77 now , if one considers a measurement of the property @xmath0 which does not distinguish between two states , one arrives on the additivity of @xmath78@xmath79so , taking a measurement of @xmath0 unable to select any of such states , one obtains @xmath80what implies : @xmath81such properties characterize @xmath82 as a _ probability measure _ @xcite of observing the state @xmath3 in a measurement made over a system known to be in the state @xmath31
. however , probability measures are positive real numbers , then we must to impose a restriction on the the numbers which figure in the measurement algebra .
until now , all we have made can be applied equally to quaternions or complex numbers .
in fact , no physical information was used to select the nature of such numbers , being only necessary they form a scalar ring in order to obtain an algebra from the elementary selective measurements .
therefore , any field , as @xmath83 or @xmath84 , for instance , is candidate to figure as scalars in this construction of the quantum theory , but also a ring which is not a field , as quaternions or octonions , could be used .
the extension of quantum mechanics that we want to do here is to get quaternions as the scalar ring used to construct the measurement algebra .
so , the probability measure @xmath82 must satisfy @xmath85 . besides , the arbitrary reading convention in the multiplicative law implies that such probability shall be _
symmetric_. the simplest way to accomplish all these properties is to demand @xmath86 and @xmath87 to be a conjugated pair . of course , in such case one obtains@xmath88@xmath89 on the other hand,@xmath90 this let us with the following statements:@xmath91 again , the simplest way to solve this system is taking@xmath92 with this choice one is able to recover all the properties of the probability measure @xmath82 . using an exponential representation for @xmath71
we see that the first condition above can be written in the form @xmath93 where @xmath94 ^{1/2 } \\
e_{\lambda } & = \frac{\lambda_{a}^{1}e_{1}+\lambda_{a}^{2}e_{2}+\lambda _ { a}^{3}e_{3}}{\left [ \left ( \lambda_{a}^{1}\right ) ^{2}+\left ( \lambda_{a}^{2}\right ) ^{2}+\left ( \lambda_{a}^{3}\right ) ^{2}\right ] ^{1/2 } } \\ \varphi\left ( a\right ) & = \arctan\left ( \frac{\lambda_{a}^{0}}{\left\vert \lambda_{a}\right\vert } \right ) \quad\varphi\left ( a\right ) \in\left [ 0,\pi% \right]\end{aligned}\ ] ] the choice for the signal in @xmath0 is arbitrary and no physical effect can be distinguished by one particular choice .
therefore we will take the positive signal .
since @xmath71 is a unitary arbitrary number its phase @xmath95 can be an arbitrary real number .
thus , besides the problems concerning about the definition of the trace functional one is still able to construct an statistical interpretation for the quaternionic quantum mechanics .
in fact , such result indicates that the roots for the statistical interpretation are in the propositional calculus of birkhoff and von neumann @xcite , and not in the particular system of numbers adopted to construct the theory .
another very important piece for the construction of the statistical interpretation was the automorphism @xmath96 for the scalar ring @xmath97 .
but , physically , what means such identification ?
we know that the elements of the scalar ring represent logical relations between the possible physical states of the system under consideration .
clearly , it is even possible to say when two of such relation are the same thing for states taken in distinct physical systems without departing the traditional concepts of pure logic , i.e. , without using the concepts of structured networks introduced by birkhoff and von neumann birkhoff - neumann .
however , this defines such numbers modulo automorphisms @xcite . in the case of a quantum theory with only real numbers
this is sufficient to determine completely such numbers @xcite . in the complex case
, it still stands an ambiguity , which is manifested under the existence of a conjugated algebra . in quaternionic quantum mechanics such
ambiguousness is infinitely bigger .
this requires the introduction of more structure elements in the theory . in the following
we will delimitate what are suchs structures .
other important aspect of the probabilistic interpretation for ( probtrans ) is the symmetry @xmath98 remember the arbitrary convention for reading the measurement symbols and their products : the order of the events is read from right to left . but any equation involving the measurement symbols is equally valid if interpreted in the opposite sense and none physical result can depend of what is the convention adopted . introducing the right - handed interpretation , @xmath99
acquire the same meaning of @xmath100 in the left - handed convention .
we can conclude that the probability connecting the states @xmath3 and @xmath31 in a given sequence must be constructed symmetrically from @xmath43 and @xmath101 .
of course , this is the reason why @xmath102 should be symmetric .
the introduction of the opposite convention for the measurement symbols will be called the _ adjoint _ operation and will denoted by @xmath103 . therefore , @xmath104 and @xmath105 in particular , @xmath106 what means that @xmath2 is a self - adjoint operator . for the product of measurements symbols we have @xmath107
the meaning of addition is not changed by the adjoint operation what permits to extend these properties for all element in the measurement algebra : @xmath108 where @xmath109 .
taking infinitesimal variations of the two fundamental properties of the transformations functions , we find@xmath110 & = \delta\left\langle a|c\right\rangle \label{diffunctrans2 } \\
\delta\overline{\left\langle a|b\right\rangle } & = \delta\left\langle b|a\right\rangle \notag\end{aligned}\ ] ] in the ordinary complex case @xcite the numbers @xmath111 are interpreted as representing the matrix elements of an infinitesimal operator,@xmath112 where the constant @xmath113 was chosen in order to assure that the operator @xmath114 is self - adjoint . here
, it is an open question what constant should be chosen since actually we have _ tree _ imaginary unities .
the most general case is let the imaginary unity to be an _ operator _
@xmath115 where @xmath116 can be considered as a particular case for @xmath84 .
let it be so , defining@xmath117 where @xmath115 is a quaternionic valued operator that we will be fixed later under the requirement of @xmath118 be a self - adjoint operator . using this definition it is easy to see that the additivity and the skewsymmetry in ordering infinitesimal operators are the same as in the complex case @xcite,@xmath119@xmath120 on the other hand,@xmath121 what let us to the operatorial identity,@xmath122 if we impose@xmath123@xmath124 = \hat{0 } \label{superselecao}\ ] ] we obtain:@xmath125 this identity can be interpreted as a generalization of the complex conjugation over @xmath84 , and shows that the operator @xmath115behaves like an imaginary unit .
the condition ( [ autoadjq ] ) assures the reality of the spectrum associated to infinitesimal operators . the condition ( [ superselecao ] ) can be satisfied in several ways : 1 . demanding that all infinitesimal operator commutes with the imaginary unity ; 2 . letting the imaginary unity to commute with any operator ; 3 .
claiming that an infinitesimal operator commutes with any other operator . in the standard quantum mechanics schwinger
choose the last option schwinger , which was subsequently extended to more general variations by several authors @xcite .
here , we can see no reason to discard the other two options .
in fact , in their work on quaternionic quantum theory , finkelstein , jauch schiminovich and speiser @xcite have adopted a particular case of the second option in the list above interpreting it as a superselection rule . ] . for while
, we will require that at least one of the three conditions above is satisfied , i.e. , we will work directly assuming only the general statement ( [ superselecao ] ) . with these choices
, unitary infinitesimal operators can be expressed as@xmath126 and infinitesimal variations of operators are induced by the commutator with the generator@xmath127 = \left [ \hat{g},\hat { x}\right ] \label{comvarind2}\ ] ] these are all ingredients necessary to describe completely the one particle physical states .
we will not approach here the problem of representing composite systems , but it is clear that such extension is possible .
now we are ready to analyse the _ dynamic _ characteristics which are changed by the use of quaternions .
the quantum dynamics for the system will be obtained from the schwinger action principle @xcite here expressed as@xmath128@xmath129 _ { t_{1}}^{t_{2}}+\int_{t_{1}}^{t_{2}}dt\frac { \vec{% \delta}\hat{l}}{\vec{\delta}\mathbf{\hat{q}}}\cdot\left ( \delta \mathbf{\hat{% q}-}\widehat{\mathbf{\dot{q}}}\delta t\right ) = \hat{g}_{2}-\hat{g}_{1}\]]@xmath130 the hamiltonian @xmath131 and lagrangian @xmath132 operators are self - adjoints .
schwinger action principle is the quantum counterpart of the classical weiss principle @xcite , which can be considered the most general variational principle for classical fields .
schwinger principle has been successfully applied in minkowiski @xcite , curved @xcite or torsioned spaces @xcite , as well as to describe quantum gauge transformations bfield and many other problems . here
, we will apply the action principle to extract dinamic and kinematic information from a canonical formulation for quaternionic quantum mechanics .
the canonical ( anti)commutation relations can be obtained from the action using the canonical infinitesimal generator,@xmath133from which we extract the following set of functional relationships:@xmath134 \hat{p}% _ { r}\delta \hat{q}^{r}-\hat{\iota}\left [ \hat{q}^{s},\hat{p}_{r}\right ] _ { \mp } \delta \hat{q}^{r}\mp \hat{\iota}\hat{p}_{r}\left [ \hat{q}^{s},\delta \hat{q}^{r}\right ] _ { \mp } \\
\hat{0 } & = -\left [ \hat{p}_{s},\hat{\iota}\right ] \hat{p}_{r}\delta \hat{q}% ^{r}-\hat{\iota}\left [ \hat{p}_{s},\hat{p}_{r}\right ] _ { \mp } \delta \hat{q}% ^{r}\mp \hat{\iota}\hat{p}_{r}\left [ \hat{p}_{s},\delta \hat{q}^{r}\right ] _ { \mp } \\
\hat{0 } & = \left [ \hat{q}^{s},\hat{\iota}\right ] \delta \hat{p}_{r}\hat{q}% ^{r}+\hat{\iota}\left [ \hat{q}^{s},\delta \hat{p}_{r}\right ] _
{ \mp } \hat{q}% ^{r}\pm \hat{\iota}\delta \hat{p}_{r}\left [ \hat{q}^{s},\hat{q}^{r}\right ] _ { \mp } \\
\delta \hat{p}_{s } & = \left [ \hat{p}_{s},\hat{\iota}\right ] \delta \hat{p}_{r}% \hat{q}^{r}+\hat{\iota}\left [ \hat{p}_{s},\delta \hat{p}_{r}\right ] _ { \mp } % \hat{q}^{r}\pm \hat{\iota}\delta \hat{p}_{r}\left [ \hat{p}_{s},\hat{q}^{r}% \right ] _ { \mp }
\end{aligned}\]]this gives a system of equations between the canonical variables and their variations whose formal solution is unknown .
one possible solution is to choose infinitesimal variations in order that @xmath135 _ { \mp } & = \left [ \hat{p}% _ { s},\delta \hat{q}^{r}\right ] _ { \mp } = \hat{0 } \\ \left [ \hat{q}^{s},\delta \hat{p}_{r}\right ] _ { \mp } & = \left [ \hat{p}% _ { s},\delta \hat{p}_{r}\right ] _
{ \mp } = \hat{0}\end{aligned}\ ] ] however , terms involving the ( anti)commutator of @xmath115 still remain , which could imply in deviationsfrom the canonical commutation relations .
that is why in @xcite is adopted the superselection rule@xmath136 = \left [ \hat{p}_{s},\hat{\iota } \right ] = \hat{0 } \label{superselcpq}\ ] ] which conduct to@xmath137 _ { \mp } & = \hat{0 } \\ \left [ \hat{q}^{s},\hat{q}^{r}\right ] _ { \mp } & = \hat{0 } \\ -\hat{\iota}\left [ \hat{q}^{s},\hat{p}_{r}\right ] _ { \mp } & = \delta _ { \,r}^{s}\end{aligned}\ ] ] to obtain an expression closer to the complex case , let us to suppose that the anti - hermitean operator @xmath115 is also
_ unitary_. by this way,@xmath137 _ { \mp } & = \hat{0 } \\ \left [ \hat{q}^{s},\hat{q}^{r}\right ] _ { \mp } & = \hat{0 } \\
\left [ \hat{q}^{s},\hat{p}_{r}\right ] _ { \mp } & = \hat{\iota}\delta_{\,r}^{s}\end{aligned}\ ] ]
this means that to obtain the standard form of the heisenberg algebra for the canonical variables @xmath138 and @xmath139 one shall to demand _ both _ conditions 2 and 3 from section [ condicoes ] .
the equation of motion for operators can also be obtained from the variational principle doing variations only in the temporal parameter,@xmath140 + % \frac{\vec{\partial}\hat{a}}{\vec{\partial}t}\ ] ] assuming that a quaternionic harmonic oscillator is described by the following lagrangian operator from the operator in this section .
we maintain it only over the imaginary unity in order to reinforce that here it is an _
operator_.]@xmath141 taking functional variations of this operator , we find@xmath142 \right ) = \\ & = \frac{1}{2}\left ( \frac{d\left ( \delta q^{\dagger}\dot{q}+\dot { q}% ^{\dagger}\delta q\right ) } { dt}-\left [ \delta q^{\dagger}\left ( \ddot { q}% + \omega^{2}q\right ) + \left ( \ddot{q}^{\dagger}+\omega^{2}q^{\dagger } \right ) \delta q\right ] \right)\end{aligned}\ ] ] therefore , the infinitesimal generator for the functional variations in the fundamental operator is@xmath143 whose induced variations are@xmath144 + \left [ q^{\beta},\hat{\iota}% \dot{q}_{\alpha}^{\dagger}\delta q^{\alpha}\right ] \right ) = \\ & = -\frac{1}{2}\left ( \hat{\iota}\delta q^{\alpha\dagger}\left [ q^{\beta } , % \dot{q}_{\alpha}\right ] + \hat{\iota}\left [ q^{\beta},\dot{q}_{\alpha } ^{\dagger}\right ] \delta q^{\alpha}\right)\end{aligned}\]]@xmath145 + \left [ q^{\beta},\dot{q}_{\alpha } ^{\dagger}\right ] \delta q^{\alpha}\right ) \\ \delta q^{\beta\dagger } & = -\hat{\iota}\left ( \delta q^{\alpha\dagger } \left [ q^{\beta\dagger},\dot{q}_{\alpha}\right ] + \left [ q^{\beta\dagger } , \dot{q}% _ { \alpha}^{\dagger}\right ] \delta q^{\alpha}\right ) \\ \delta\dot{q}^{\beta } & = \hat{\iota}\left ( \delta\dot{q}_{\alpha}\left [ \dot{% q}^{\beta},q^{\alpha\dagger}\right ] + \left [ \dot{q}^{\beta},q^{\alpha } % \right ] \delta\dot{q}_{\alpha}^{\dagger}\right ) \\ \delta\dot{q}^{\beta\dagger } & = \hat{\iota}\left ( \delta\dot{q}_{\alpha } % \left [ \dot{q}^{\beta\dagger},q^{\alpha\dagger}\right ] + \left [ \dot { q}% ^{\beta\dagger},q^{\alpha}\right ] \delta\dot{q}_{\alpha}^{\dagger}\right)\end{aligned}\ ] ] assuming that the operators @xmath146 , @xmath147 , @xmath148 and @xmath149 are kinematically independent , we have the canonical commutation relations,@xmath150 & = \left [ q^{\beta},\dot{q}_{\alpha}\right ] = 0 \\ \left [ q^{\beta\dagger},\dot{q}_{\alpha}\right ] & = \left [ q^{\beta},\dot{q}% _ { \alpha}^{\dagger}\right ] = \hat{\iota}\delta_{\,\alpha}^{\beta}\end{aligned}\ ] ]
taking variations only over the final state in a given transition,@xmath151 we have @xmath152 now , let us identify the description @xmath3 as the generalized coordinates , i.e. , the description where the operators @xmath153 are diagonal , and the state @xmath154 as an arbitrary state @xmath155 . from the commutation relations deduced before we have @xmath156 but ,
@xmath157 then,@xmath158 inserting a completeness relation for the coordinate eigenstates , we find@xmath159 the first equation gives the representation of the momentum operator in the coordinate representation assuming that the spectrum of @xmath115 is know , while the second is the _ _ schrdinger equation__. if , by hypothesis , the operator @xmath115 has always the same value in any point of the coordinate space and at any instant of time , then [ eqschqconst ] @xmath160 where @xmath161 is the expected value of @xmath115 .
of course , this last hypothesis is contained in the statement 2 of the section [ condicoes ] and it imply that the _ operator _ @xmath115 is actually a constant imaginary pure quaternion .
the schwinger measurement algebra formulation for quantum kinematics is a powerful tool to disconnect the physical contents in quantum measurements from the mathematical requirements of consistence . at same time
, it provides a natural way to achieve generalizations of standard quantum mechanics and provide a clear view of the price paid for such generalizations . in particular ,
besides we have found difficulties to construct a linear functional relating operators and values in the quaternionic ring , it was still possible to achieve a well defined statistical interpretation for quaternionic quantum mechanics .
the essential elements for such construction are the noncompatibility of successive measurements , providing the fundamental law of multiplication for measurement symbols , and the automorphism @xmath96 of the scalar ring . in principle
, any theory with these basic characteristics can also have an statistical interpretation .
notwithstanding , for an appropriate interpretation , some additional properties are required for the probability measure @xmath82 , such as the conservation of their associated current in a closed system .
in fact , it is the essential feature that adler used to prove the non - extensivity of quantum mechanics for octonions or higher dimension hypercomplex numbers @xcite .
it must be stressed that there are several problems which are not investigated above , such as the effect of the superselection rules ( superselcpq ) over representations of the canonical variables @xmath139and @xmath138 , or the physical effects of the new quaternionic degrees of freedom .
although we have not treated composite systems ( i.e. , many particle systems ) it is possible to advance some characteristics which should originate from the quaternionic noncommutativity .
it is well known that in the classical physics there are no phase relations to be considered among subsystems of a bigger system ( non - interacting particles ) if we sum or multiply ( by cartesian product ) their phase spaces .
in complex quantum mechanics there are phase relations between states which are important if we sum their state spaces , but they are not important for the product of such spaces ( understood as a tensorial product ) . in quaternionic quantum mechanics
these phase relations should be important whatever one is dealing with sum or product of spaces , since the phase factor now is a quaternion .
this new feature of quaternionic quantum mechanics can be interpreted in terms of a complementarity argument . in classical physics
does exist complementarity relations because all measurements can , in principle , have an infinity precision . in the real and complex quantum mechanics
there are complementarity relations among physical properties of the same system , but not between properties of different non - interacting systems . in quaternionic quantum mechanics
there are complementarity between some properties for any pair of physical systems or subsystems .
this is because the phase factor @xmath162 can not be additively composed when multiplying quaternions , as specified in the section [ estatistica ] . therefore , there is no reasonable way to form composite systems in order to have all observables associated in one system to be compatible with all the observables in any other systems or , in other words , to commute with all the others observables in different systems .
actually , one can expect this to be greater difficulty to describe many particle systems in quaternionic quantum theory it is important to observe that besides the notion of a quaternionic hilbert space has been a little vague here it is possible to develop the concepts the geometry of states , as done by schwinger @xcite , for the quaternionic ring .
the idea and properties of such vectorial space emerge naturally in the geometry of states .
this was not done here simply by matters of space and convenience since that we were interested not only on the kinematical side , but also in the dynamic aspects of the quaternionic theory . for those interested in the spectral theory of quaternionic hilbert spaces
is interesting to check @xcite where the main theorems and ideas are introduced with a pedagogical explanation of how to perform the calculations in a vectorial space of scalars in @xmath97 . with respect to quartenionic quantum mechanics of a single particle
one can observe that the points where the operator @xmath115 appears are essentially the same where the planck constant @xmath163 should be .
of course , using a different system of units , one realizes that the operator @xmath164 takes the place of the combination @xmath165 accordingly the analogy applied here . by this way
, the introduction of operators which fail to commute with @xmath115 can be understood as to promote the planck constant to a new dynamic variable , being interesting to investigate the fluctuations in the quantum of action in such case . on the other hand , the superselection rule expressed by the second condition in the section [ condicoes ] together with the hypothesis made in the final of the section [ iconst ]
classical _ meaning to @xmath115 excluding the interference between their different states .
this is equivalent to freezethe actual value of the imaginary unity operator suppressing this new possibilities .
therefore , we find a natural extension of the equations ( eqschqconst ) admitting that the operator @xmath115 actually is a new _ fundamental field _ , i.e. , a new dynamic variable which depends from the space - time point where it is observed .
this idea was partially developed in @xcite where it is proposed a quaternionic general covariance principle , which means a theory for the parallel transport of quaternions over a manifold , and a field equation for the operator @xmath115 .
one of the most surprising results of this theory is that the field equations obtained are very similar to the electromagnetic ones but with _ three _ fundamental vectorial bosons , one neutral and massless and two others massive and charged .
so , quaternionic quantum mechanics could be considered one of the first attempts to construct an unified theory for the electroweak interactions ( 1963 ) and perhaps could model at least a sector of the complete electroweak interactions .
p. a. moran , _ an introduction to probability theory , _ claredon press , 1968 ; b. v. gnedenko , _ the theory of probability , _ chelsea publishing company , 1937 ; a. n. kolmogorov , _ foundations of the theory of probability , _ chelsea publishing company , 1950 .
j. s. schwinger , _ quantum kinematics and dynamics _ , addison wesley , 1991 ; _ pnas _ * 45 * ( 1959 ) , 1542 ; _ pnas _ * 46 * ( 1960 ) , 257 ; _ pnas _ * 46 * ( 1960 ) , 570 ; _ pnas _ * 46 * ( 1960 ) , 883 ; _ pnas _ * 46 * ( 1960 ) , 1401 ; _ pnas _ * 47 * ( 1961 ) , 1075 .
j. s. schwinger , _ phys .
* 82 * ( 1951 ) , 914 ; _ phys .
* 91 * ( 1953 ) , 713 ; _ phys .
_ * 91 * ( 1953 ) , 728 ; _ phys
* 92 * ( 1953 ) , 1283 ; _ phys .
rev . _ * 93 * ( 1954 ) , 615 ; _ phys .
* 94 * ( 1954 ) , 1362 . | a quaternionic version of quantum mechanics is constructed using the schwinger formulation based on measurements and a variational principle .
commutation relations and evolution equations are provided , and the results are compared with other formulations . |
in @xcite a new approach was proposed for finding the low - energy eigenstates of very large or infinite - dimensional quantum hamiltonians .
this proposal combines both diagonalization and monte carlo methods , each being used to solve a portion of the problem for which the technique is most efficient .
the first part of the proposal is to diagonalize the hamiltonian restricted to a subspace containing the most important basis vectors for each low energy eigenstate .
this may be accomplished either through variational techniques or an _ ab initio _ method such as quasi - sparse eigenvector ( qse ) diagonalization .
the second step is to include the contribution of the remaining basis vectors by monte carlo sampling .
the use of diagonalization allows one to consider systems with fermion sign oscillations and extract information about wavefunctions and excited states .
the use of monte carlo provides tools to handle the exponential increase in the number of basis states for large volume systems .
the first half of this proposal was discussed in @xcite . an adaptive diagonalization algorithm known as the quasi - sparse eigenvector method
was introduced to find the most important basis vectors for each low energy eigenstate .
this technique is especially valuable when little is known about the low energy states .
it is also the only method available which can handle non - orthogonal bases , non - hermitian hamiltonians , infinite dimensional systems , and which can find several low energy states with like quantum numbers simultaneously . in this paper
we discuss the second half of the diagonalization / monte carlo scheme .
we introduce several new monte carlo techniques which we call stochastic error correction ( sec ) .
there are two general varieties of stochastic error correction , methods based on a series expansion and those which are not .
the series method starts with an eigenvector of the hamiltonian restricted to some starting subspace and then includes the contribution of the remaining basis states as terms in an ordered expansion .
the idea is to form a perturbative expansion centered around a good non - perturbative starting point . as an example of a non - series method we discuss a technique called the stochastic lanczos method .
this method again starts with eigenvectors of a hamiltonian submatrix . using these as starting vectors
, we define krylov vectors , @xmath0 , similar to standard lanczos diagonalization .
the new ingredient is that matrix elements between krylov vectors , @xmath1 are computed using matrix diffusion monte carlo .
since the method does not rely on a series expansion , it has the advantage that the starting vectors need not be close to the exact eigenvectors . one can generate a large class of stochastic error correction methods based on other non - series algorithms , various ways of resumming the series expansion , or combinations of the two techniques . in this introductory paper we concentrate on describing the basic principles and implementation of the series and non - series approaches .
we also present three test problems which demonstrate the potential of the new approach for a range of different problems . in the first example we determine the low energy spectrum of @xmath2 theory using qse diagonalization and first order corrections using the series method . in the second example we find the low energy spectrum of compact @xmath3 in @xmath4 dimensions using the stochastic lanczos method . in the last example we find the ground state of the @xmath4 dimensional hubbard model using qse diagonalization and first order series stochastic error correction . in each case
we compare with published results in the literature .
we conclude with a summary and some general comments on the new computational scheme .
let @xmath5 be the eigenvectors of a hamiltonian @xmath6 restricted to some subspace @xmath7 .
let @xmath8 be the remaining basis vectors in the full space not contained in @xmath7 .
we can represent @xmath6 as @xmath9{cccccc}% \lambda_{1 } & 0 & \cdots & \left\langle 1\right| h\left| a_{1}\right\rangle & \left\langle 1\right| h\left| a_{2}\right\rangle & \cdots\\ 0 & \lambda_{2 } & \cdots & \left\langle 2\right| h\left| a_{1}\right\rangle & \left\langle 2\right| h\left| a_{2}\right\rangle & \cdots\\ \vdots & \vdots & \ddots & \vdots & \vdots & \cdots\\ \left\langle a_{1}\right| h\left| 1\right\rangle & \left\langle a_{1}\right| h\left| 2\right\rangle & \cdots & e\cdot\lambda_{a_{1 } } & \left\langle a_{1}\right| h\left| a_{2}\right\rangle & \cdots\\ \left\langle a_{2}\right| h\left| 1\right\rangle & \left\langle a_{2}\right| h\left| 2\right\rangle & \cdots & \left\langle a_{2}\right|
h\left| a_{1}\right\rangle & e\cdot\lambda_{a_{2 } } & \cdots\\ \vdots & \vdots & \vdots & \vdots & \vdots & \ddots \end{array } \right ] .\ ] ] we have used dirac s bra - ket notation to represent the terms of the matrix . in cases where the basis is non - orthogonal or the hamiltonian is non - hermitian , the precise meaning of terms such as @xmath10 is the action of the dual vector to @xmath11 upon the vector @xmath12 .
we have written the diagonal terms for the basis vectors @xmath13 with an explicit factor @xmath14 for reasons to be explained shortly .
let us assume that @xmath15 is close to some exact eigenvector of @xmath6 which we denote as @xmath16 .
more concretely we assume that the components of @xmath17 outside @xmath7 are small enough so that we can expand in inverse powers of the introduced parameter @xmath18 in order to simply the expansion we choose to shift the diagonal entries so that @xmath19 the series method of stochastic error correction is based on the @xmath20 expansion , @xmath21{c}% 1\\ c_{2}^{\prime}e^{-1}+c_{2}^{\prime\prime}e^{-2}+\cdots\\ \vdots\\ c_{a_{1}}^{\prime}e^{-1}+c_{a_{1}}^{\prime\prime}e^{-2}+\cdots\\ c_{a_{2}}^{\prime}e^{-1}+c_{a_{2}}^{\prime\prime}e^{-2}+\cdots\\ \vdots \end{array } \right ] , \label{unnorm}%\]]@xmath22 it is convenient to choose the normalization of the eigenvector such that the @xmath15 component remains 1 .
the convergence of the expansion is controlled by the proximity of @xmath15 to @xmath16 .
if @xmath15 is not at all close to @xmath16 then it will be necessary to use a non - series method such as the stochastic lanczos method discussed in the next section . at first order in @xmath20
we find @xmath23@xmath24@xmath25 at second order the contributions are @xmath26 @xmath27 @xmath28 \tfrac{1}{\lambda_{m}}\left [ \sum_{k}% \tfrac{\left\langle m\right| h\left| a_{k}\right\rangle \left\langle a_{k}\right| h\left| 1\right\rangle } { \lambda_{a_{k}}}\right ] .\nonumber\end{aligned}\ ] ] these contributions can be calculated by straightforward monte carlo sampling .
all that is required is an efficient way of generating random basis vectors @xmath29 with known probability rates .
let @xmath30 denote the probability of selecting @xmath31 on a given trial .
if for example we are calculating the first order correction to the eigenvalue , then we have@xmath32
we now consider a method called stochastic lanczos which does not require the starting vectors to be close to exact eigenvectors of @xmath33 this is essential if the eigenvectors of @xmath6 are not quasi - sparse and require extremely large numbers of basis states to represent accurately .
let @xmath34 be the full hilbert space for our system . as in the previous section
let @xmath7 be the subspace over which we have diagonalized @xmath6 exactly .
let @xmath35 be the projection operator for @xmath7 and let @xmath36 and @xmath37 be the eigenvalues and eigenvectors of @xmath6 restricted to @xmath7 so that @xmath38 let @xmath39 be an auxiliary subspace , one which contains @xmath7 but excludes very high - energy states .
let @xmath40 be the projection operator for @xmath39 .
we will choose @xmath39 such that @xmath41 is bounded above .
let@xmath42 be a real constant which is greater than the midpoint of the minimum and maximum eigenvalues of @xmath41 . as @xmath43 the operator @xmath44 ^{n}$ ] maps any given state in @xmath39 to the corresponding lowest - energy eigenvector of @xmath41 with non - zero overlap
the stochastic lanczos method uses the operators @xmath45 ^{n}$ ] to approximate the low - energy eigenvalues and eigenvectors of @xmath41 .
the goal is to diagonalize @xmath6 in a subspace spanned by vectors @xmath46 ^{d}\left| j\right\rangle , \label{slbasis}%\ ] ] for several values of @xmath47 and @xmath48 .
this requires calculating @xmath49 and @xmath50 .
if our hamiltonian matrix is hermitian , both of these terms can be written in the general form @xmath51 ^{n}\left| j\right\rangle .\ ] ] therefore it suffices to determine the matrix @xmath52 ^{n}p_{s}.\ ] ] for non - orthogonal bases and non - hermitian hamiltonians , the only change is that we use vectors @xmath53 ^{d}\left| j\right\rangle\ ] ] to generate approximate right eigenvectors of @xmath6 and vectors in the dual space @xmath54 ^{d}%\ ] ] to produce approximate left eigenvectors .
adding and subtracting @xmath55 , we can rewrite @xmath56 ^{n}p_{s}.\ ] ] @xmath57 can now be evaluated recursively as @xmath58 where @xmath59 ^{n}p_{s}.\ ] ] the components of @xmath60 are computed by matrix diffusion monte carlo .
one could also directly evaluate the components of @xmath57 .
however the calculation for @xmath60 eliminates the need to sample the matrix @xmath55 , which is already known .
any general matrix product @xmath61 is a sum of degree @xmath62 monomials , @xmath63 _ { jk}=\sum_{i_{1},\cdots i_{n-1}% } m_{ji_{1}}^{(1)}m_{i_{1}i_{2}}^{(2)}\cdots m_{i_{n-1}k}^{(n)}. \label{sum}%\ ] ] we can interpret ( [ sum ] ) as a sum over paths through the set of basis vectors of @xmath64@xmath65 with an associated weight @xmath66 . the components of @xmath60 are sampled using ensembles of random walkers .
we refer the interested reader to @xcite for a review of methods in diffusion monte carlo .
we end the section with a discussion of the fermion sign problem .
the sign problem is a general issue for any monte carlo calculation . for a system with sign oscillations
the evaluation of a euclidean - time green s function involves sums @xmath67 with the property that @xmath68 where @xmath34 is the volume , @xmath69 is the euclidean time , and @xmath70 is a positive constant
. we will refer to this term as the cancellation ratio .
the exponential dependence on @xmath34 and @xmath69 makes computations difficult even for small systems .
the sign problem will affect the calculation of @xmath60 in the stochastic lanczos algorithm and terms in the series method discussed in the previous section .
the effect however is different from the sign problem in typical monte carlo green s function calculations .
stochastic error correction is a calculation of eigenvalues and eigenvectors rather than a sampling of the partition function or the time evolution of a given initial state .
therefore the quantity of interest is not @xmath71 but the action of @xmath6 or @xmath72 on approximate eigenvectors of @xmath6 .
due to homogeneity in @xmath6 the explicit volume dependence does not appear in the cancellation ratio .
instead we find @xmath73 where @xmath74 is a positive constant .
the sign problem will return if the starting point of the sec calculation is very poor and it becomes necessary to use @xmath62 such that @xmath75 is large .
however in many cases @xmath75 can be kept small even for large @xmath62 since the most important part of the hamiltonian , @xmath76 is diagonalized exactly . in short
the sign problem is less severe because stochastic error correction uses the result of subspace diagonalization as its starting point .
the first example we consider is @xmath77 theory in @xmath4 dimensions near the @xmath78 symmetry restoration phase transition .
we will use qse diagonalization and the series version of stochastic error correction to probe the low energy spectrum of the theory on both sides of the phase transition .
in @xcite magruder demonstrated the existence of a phase transition in @xmath2 by extending chang s duality argument for @xmath79 .
the statement of the main result is as follows .
consider the two lagrange densities@xmath80 the counterterm @xmath81 is defined so that in the @xmath82 system the @xmath83 self - energy graphs vanish at zero - momentum up to two - loop order . by shifting the field@xmath84 we note that the same counterterm @xmath85 ( same mass dependence but @xmath86 replaced by @xmath87 ) is also sufficient to renormalize @xmath88 . by equating @xmath82 and @xmath88
we obtain a duality constraint between the two theories .
one feature of this constraint is that the @xmath89 limit of @xmath88 is mapped to the @xmath90 limit of @xmath82 .
therefore @xmath88 , whose reflection symmetry @xmath78 is broken at small @xmath91 , must eventually reach the symmetric phase for sufficiently large coupling .
limit of @xmath82 is mapped to the @xmath89 limit of @xmath88 and so there is no analogous argument for a phase transition in @xmath82 .
numerical calculations indicate that there is no phase transition for @xmath82 @xcite@xcite . ]
the @xmath88 phase transition was studied using quasi - sparse eigenvector diagonalization with stochastic error correction .
quantities such as the critical coupling , critical exponents , and the low lying energy spectrum were studied and , where possible , compared with monte carlo results .
a full discussion methods and results are presented in @xcite .
we will very briefly summarize some of the results below .
the two spatial dimensions of our system are taken to be a periodic box of size @xmath92 by @xmath92 .
we will use the modal field formalism to describe the hamiltonian for the theory . in the following we
let the vectors @xmath93 represent ordered integer pairs @xmath94 such that @xmath95 .
the parameter @xmath96 corresponds with a momentum cutoff scale of @xmath97 .
the modal field hamiltonian has the form @xmath98
q_{-\vec{n}}q_{\vec{n}}\\ & + \tfrac{1}{(2l)^{2}}\tfrac{g}{4!}\sum_{\vec{n}_{1}+\vec{n}_{2}+\vec{n}% _ { 3}+\vec{n}_{4}=0}q_{\vec{n}_{1}}q_{\vec{n}_{2}}q_{\vec{n}_{3}}q_{\vec{n}% _ { 4}}\nonumber\end{aligned}\ ] ] where @xmath99 and@xmath100 in figure 1 we have plotted the lowest energy eigenstates in the rest frame for @xmath101 and @xmath102 ( in units where @xmath103 ) .
this choice of parameters corresponds with @xmath104 different momentum modes and a momentum cutoff scale of @xmath105 .
the states shown in figure 1 are the three lowest eigenstates in the even and odd @xmath78 symmetry sectors , and the energies are measured relative to the ground state energy . in our calculation
qse diagonalization was used keeping @xmath106 fock states , and the stochastic error correction was computed to first - order using the series method .
error bars shown include statistical error and an estimate of the contribution from higher order corrections .
we see clear evidence of a second - order phase transition near @xmath107 .
we have labelled the energies of the states according to their physical interpretation in the symmetric phase .
@xmath108 is the energy for the one - particle state , @xmath109 is for the two(three)-particle threshold , and @xmath110 is for the first state above the two(three)-particle threshold . at finite volume
these energies are continuous functions of the coupling @xmath91 .
one feature which was also observed in @xmath79 @xcite is the crossing of energy levels due to the double degeneracy of states in the broken symmetry phase .
@xmath111 is connected to a one - particle state in the broken phase while @xmath112 is connected to a two - particle state
. the levels @xmath113 and @xmath114 therefore cross near the critical point .
another interesting phenomenon is the appearance of a bound state in the broken symmetry phase . in
both the odd and even symmetry sectors we can measure the ratio of the two - particle to one - particle energies:@xmath115{|l|l|l|}\hline $ \frac{g}{4!}$ & $ e_{2}^{\prime}/e_{2}$ & $ e_{3}^{\prime}/e_{3}$\\\hline $ 0.2 $ & $ 2.01(4)$ & $ 1.98(4)$\\\hline $ 0.3 $ & $ 2.01(4)$ & $ 2.05(4)$\\\hline $ 0.4 $ & $ 1.95(4)$ & $ 1.96(4)$\\\hline $ 0.5 $ & $ 1.87(4)$ & $ 1.87(4)$\\\hline $ 0.6 $ & $ 1.86(4)$ & $ 1.82(4)$\\\hline
\end{tabular } } % \ ] ] these results are consistent with the binding energies reported in @xcite and @xcite , which indicate a ratio of @xmath116 near the critical point .
compact u(1 ) lattice gauge theory in @xmath4 dimensions is a simple but phenomenologically interesting gauge model .
it is asymptotically free and in the usual continuum limit describes massless non - interacting photons . on the other hand
if the continuum limit is reached by rescaling the mass gap to remain constant , one instead finds a confining theory of massive bosons @xcite .
the hamiltonian has the form @xmath117 in ( [ hamil ] ) @xmath118 are link gauge fields , @xmath119 is the sum of the links circuiting a plaquette,@xmath120 and@xmath121 is the strong coupling parameter , which tends to infinity as the lattice spacing @xmath122 goes to @xmath123 .
we follow the notation of @xcite in which an overall constant factor of @xmath124 multiplying the right - hand side of ( [ hamil ] ) is suppressed .
the energy levels we measure are therefore in units of @xmath124 .
the diagonalization of lattice gauge hamiltonians is constrained by the requirements of gauge invariance . to preserve gauge invariance
it is most convenient to use a basis which diagonalizes the electric field part of the hamiltonian@xmath125 \bigotimes_{l^{\prime}}\left| n_{l^{\prime}}\right\rangle \cdots=\left [ \sum_{l}n_{l}^{2}\right ] \bigotimes_{l^{\prime}}\left| n_{l^{\prime}% } \right\rangle .\ ] ] as our next example of stochastic error correction we will address the @xmath126 lattice system at @xmath127 using this electric field basis . in @xcite it was noted that this poses a challenge to standard diagonalization techniques . even on the small @xmath126 lattice a surprisingly large number of states , about @xmath128 , are needed to accurately describe the low energy spectrum at @xmath127 .
this problem can be circumvented by modifying the basis states to incorporate more of the physics of the ground state .
for example one can introduce a disordered background of magnetic flux as suggested in @xcite , and that approach is followed in an ongoing project @xcite .
however we would like to directly address the problem described in @xcite and show how the stochastic lanczos method handles the proliferation of large numbers of basis states in the original electric field basis .
we will choose our starting subspace @xmath7 to include all basis states @xmath129 which satisfy@xmath130 and which can be reached from the strong coupling vacuum by at most two transitions via the plaquette operators @xmath131 .
we take the auxiliary space @xmath39 to be the subspace spanned by basis vectors@xmath132 using matrix diffusion monte carlo , we diagonalize the subspace formed by the states@xmath46 ^{d}\left| j\right\rangle .\ ] ] @xmath133 are the eigenvectors of the hamiltonian restricted to the original subspace @xmath7 . in our calculations we use @xmath134 and @xmath135 for cutoff values , @xmath136 in table 2
we show the results for the ground state energy @xmath137 for the different cutoff values @xmath138 and the extrapolated value at @xmath139 .
the errors shown are estimated statistical errors . for comparison
we show the results of @xcite obtained using green s function monte carlo ( gfmc ) .
@xmath140{|l|l|l|l|l|l|}\hline & $ l_{\max}^{2}=24 $ & $ l_{\max}^{2}=28 $ & $ l_{\max}^{2}=32 $ & $ l_{\max}% ^{2}=\infty$ & gfmc\\\hline $ e_{0}$ & $ -7.394(3)$ & $ -7.430(3)$ & $ -7.438(3)$ & $ -7.442(4)$ & $ -7.4432(5)$\\\hline
\end{tabular } } % \ ] ] in table 3 we show the masses for the lightest six particles in the system extrapolated to the limit @xmath141 .
we have labelled the particles according to their spin @xmath142 and sign under conjugation @xmath143 we also include results from @xcite for the lowest antisymmetric and symmetric glueballs.@xmath144{|l|l|l|}\hline $ j^{c}$ & mass & gfmc\\\hline $ \left| 0^{-}\right\rangle $ & $ 3.03(2)$ & $ 3.01(6)$\\\hline $ \left| 0^{+}\right\rangle $ & $ 4.03(3)$ & $ 4.05(8)$\\\hline $ \left| 2^{-}\right\rangle $ & $ 6.8(1)$ & \\\hline $ \left| 2^{+}\right\rangle $ & $ 6.8(1)$ & \\\hline $ \left| 0^{+}\right\rangle $ & $ 7.0(2)$ & \\\hline $ \left| 0^{-}\right\rangle $ & $ 7.1(2)$ & \\\hline
\end{tabular } } % \ ] ] the results we find appear in agreement with @xcite . unlike most monte carlo algorithms , the sec method is able to find excited states with the same quantum numbers as lower lying states .
this was also evident in the @xmath2 example where we could track many different states crossing the phase transition .
the reason for this advantage goes back to the design of stochastic error correction as a monte carlo improvement of a diagonalization scheme . for the @xmath3 example one can reliably find the eigenvalues and eigenvectors for the first twenty or so states in the low energy spectrum
the last example we consider is the two - dimensional hubbard model defined by the hamiltonian @xmath145 the summation @xmath146 is over nearest neighbor pairs .
@xmath147(@xmath148 ) is the creation(annihilation ) operator for a spin @xmath149 electron at site @xmath150 .
@xmath151 is the hopping parameter , and @xmath152 controls the on - site coulomb repulsion .
the model has attracted considerable attention in recent years due to its possible connection to @xmath47-wave pairing and stripe correlations in high-@xmath153 cuprate superconductors . in spite of its simple form , the computational difficulties associated with finding the ground state of the model are substantial even for small systems .
fermion sign problems render monte carlo simulations ineffective for @xmath152 positive and away from half - filling , and the collective effect of very large numbers of basis fock states make most diagonalization approaches very difficult .
a brief overview of the history and literature pertaining to numerical aspects of the hubbard model can be found in @xcite . in terms of momentum space variables ,
the hubbard hamiltonian on an @xmath154 periodic lattice has the form@xmath155 \\ & + \tfrac{u}{n^{2}}\sum_{\substack{p_{x}-q_{x}+r_{x}-s_{x}% = 0\operatorname{mod}n\\p_{y}-q_{y}+r_{y}-s_{y}=0\operatorname{mod}n}% } c_{p_{x},p_{y}}^{\uparrow\dagger}c_{q_{x,}q_{y}}^{\uparrow}c_{r_{x},r_{y}% } ^{\downarrow\dagger}c_{s_{x},s_{y}}^{\downarrow}.\nonumber\end{aligned}\ ] ] as a test of our methods , we use qse diagonalization with stochastic error correction to find the ground state energy of the @xmath126 hubbard model with @xmath156 electrons per spin .
the corresponding hilbert space has about @xmath157 dimensions . for the qse diagonalization we use momentum fock states which diagonalize the quadratic part of the hamiltonian .
the hamiltonian is invariant under the symmetry group generated by reflections about the @xmath158 and @xmath159 axes , interchanges between @xmath158 and @xmath160 and interchanges between @xmath161 and @xmath162 .
we find it convenient to work with symmetrized fock states .
we will compute stochastic error corrections to first order using the series method . in table 4
we present results for the ground state energy .
we encountered no trouble with the sign problem , and in fact one can easily see that each term in the first order series expression ( [ firs ] ) is negative definite .
the energies are measured relative to the energy of the fermi sea at @xmath163 .
the errors reported are statistical errors associated with the first order sec calculation . where available
, we compare with the results presented in @xcite , which we label as exact , projector quantum monte - carlo ( pqmc ) , and stochastic diagonalization ( sd ) .
stochastic diagonalization is a subspace diagonalization technique similar to qse but one which uses a different method for selecting the subspace and is based on a variational principle @xcite .
although the precise number of basis states used in the sd calculations is not listed , we infer from numbers reported for a modified @xmath126 hubbard system that roughly @xmath164 states were used . @xmath165{|l|l|l|l|l|l|l|}\hline coupling & states & qse & qse+sec & exact & sd & pqmc\\\hline $ u=2t$ & $ % \begin{array } [ c]{c}%
100\\ 500\\ 1000 \end{array } $ & $ % \begin{array } [ c]{c}% -.4797\\ -.4945\\ -.5006 \end{array } $ & $ % \begin{array } [ c]{c}% -.50147(5)\\ -.50181(3)\\ -.50198(1 ) \end{array } $ & $ -.50194 $ & $ -.5010 $ & $ -.44(5)$\\\hline $ u=4t$ & $ % \begin{array } [ c]{c}% 100\\ 500\\ 1000 \end{array } $ & $ % \begin{array } [ c]{c}% -1.620\\ -1.748\\ -1.800 \end{array } $ & $ % \begin{array } [ c]{c}% -1.8113(4)\\ -1.8242(3)\\ -1.8302(1 ) \end{array } $ & $ -1.8309 $ & $ -1.829 $ & $ -1.8(2)$\\\hline $ u=5t$ & $ % \begin{array } [ c]{c}% 500\\ 1000\\ 2000 \end{array } $ & $ % \begin{array } [ c]{c}% -2.558\\ -2.651\\ -2.685 \end{array } $ & $ % \begin{array } [ c]{c}% -2.7073(4)\\ -2.7208(2)\\ -2.7231(1 ) \end{array } $ & $ -2.7245 $ & $ -2.723 $ & $ -2.9(3)$\\\hline \end{tabular } } % \ ] ] apparently qse diagonalization with sec handles the @xmath126 system quite well with relatively few states .
much larger systems are being studied using both higher series corrections and stochastic lanczos techniques @xcite .
in this paper we presented two versions of stochastic error correction , the series method and the stochastic lanczos method .
the series method starts with eigenvectors of the hamiltonian restricted to some optimized subspace and includes the contribution of the remaining basis states as an ordered expansion .
the stochastic lanczos method starts with eigenvectors of a hamiltonian submatrix and constructs matrix elements of krylov vectors using matrix diffusion monte carlo .
this method has the advantage that the starting vectors need not be close to exact eigenvectors .
we presented three different examples which demonstrate the potential of the new approach for strongly coupled scalar , gauge , and fermionic theories .
in the first example we calculated the low energy spectrum of @xmath2 using the series method , and in the second example we found the spectrum of compact @xmath3 in @xmath4 dimensions using the stochastic lanczos method . in both examples we found agreement with results from the literature .
we also found that unlike typical monte carlo results , the sec method is able to find the eigenvalues and eigenvectors for excited states with the same quantum numbers as lower lying states .
this advantage is due to its design as a monte carlo improvement of a diagonalization scheme . in the last example we found the ground state of the @xmath4 dimensional hubbard model using qse diagonalization and first order series stochastic error correction . in this calculation
we encountered no fermion sign problem and found that our methods yielded very accurate results with far less effort than existing techniques .
we believe that the methods we have presented hold considerable potential for studying a wide range of non - perturbative quantum systems and answering questions difficult to address using other methods . | we propose a method for eliminating the truncation error associated with any subspace diagonalization calculation .
the new method , called stochastic error correction , uses monte carlo sampling to compute the contribution of the remaining basis vectors not included in the initial diagonalization .
the method is part of a new approach to computational quantum physics which combines both diagonalization and monte carlo techniques . |
associated with string theory and higher dimensional universes , there has been a renewed interest in higher dimensional solutions to einstein field equations , among the most interesting being black rings @xcite , kk bubbles as well as black strings .
one simple way to lift the 4d schwarzschild black hole to a 5d black string is to add an extra flat dimension . in other words
it is possible to uniformly extend the 4d black hole with @xmath0 horizon into the fifth dimension producing a hypercylindrical black hole @xmath1 @xcite . a striking fact related to black strings is their generic instability and final fate and whether these end up as black holes or different objects is still not well understood @xcite , @xcite . in this sense
it is interesting testing more general black string solutions both static and non - static .
some generalizations are obtained by applying a boost to the string solution , dotating the string with the momentum along the fifth coordinate @xcite . in this work
we derive and report a generalized black string solution with mass , arbitrary tension and an additional parameter related to deformation from spherical symmetry . to generate a generalized black string solution we take as seed the zipoy - vorhees ( zv ) family of solutions and then lift it to 5d by the procedure introduced in @xcite,@xcite .
particular cases are the static black string with arbitrary tension of c. h. lee @xcite and the gregory - laflamme black string @xcite . trapped surfaces and horizons , as well as the mass and string tension parameters , are analyzed and discussed .
we briefly explain the idea of the generating technique ( see details in @xcite ) .
let us consider the 5d vacuum einstein action , @xmath2 where the hatted quantities mean the fifth dimensional version of the scalar curvature @xmath3 , the metric tensor @xmath4 and the gravitational newton constant @xmath5 and @xmath6 is the fifth coordinate .
the equations of motion derived of the variation of action ( [ s5d ] ) coincide with the einstein - scalar coupled equations in four dimensions . for a static line element given in weyl coordinates , @xmath7 and a minimaly coupled scalar field @xmath8 , these field equations are given by @xmath9 let us consider the zv solution @xcite in weyl coordinates ( for a discussion on zipoy - voorhees solution see @xcite ) , with the line element ( [ a1 ] ) and the metric functions given by : @xmath10^ { \frac{q}{2 } } , \nonumber\\ \sigma&=&\frac{1}{2}\ln\left[\frac{(r_{+}+r_{-})^{2}-4m^{2}}{4r_{+}r_{- } } \right]^{\frac{q^{2}}{4 } } , \nonumber\\ r_{\pm } ^{2}&=&\rho^{2}+(z \pm m)^{2}. \label{a1.1}\end{aligned}\ ] ] zv solution is characterized by two free parameters @xmath11 and @xmath12 related to mass and spherical - symmetry deformation , respectively .
for @xmath13 we recover schwarzschild solution while if @xmath14 it represents the darmois solution @xcite .
adding the scalar field in order to use the generating algorithm of @xcite,@xcite @xmath15,\ ] ] where the scalar field is a solution of laplace s equation @xmath16 , we obtain the spacetime with a scalar field @xmath8 , expressed as @xmath17 with @xmath18.\ ] ] the above solution is an exact solution of the einstein - scalar equations ( [ field_eqs ] ) .
using now the method described in @xcite , and considering the metric ( [ a3 ] ) as seed , we lift the solution to five dimensions : @xmath19 explicitly @xmath20 where @xmath21 transforming the metric ( [ a6 ] ) to the schwarzschild - like coordinates @xmath22 , with @xmath23 we obtain : @xmath24 where @xmath25 and @xmath26 .
this solution is a generalized 5d black string characterized by three free parameters : @xmath27 , @xmath12 and @xmath11 .
note that to recover transverse spherical symmetry , we must choose @xmath28 in ( [ a8 ] ) .
doing so the term with @xmath29 does not appear in @xmath30 and @xmath31 , obtaining an element proportional to @xmath32 : @xmath33 the case @xmath34 and @xmath13 corresponds to the black string @xcite , @xmath35 in the case @xmath36 and @xmath37 solution ( [ a8 ] ) becomes @xmath38 -dt^{2},\ ] ] having in square brackets the schwarzschild line element with the euclidean signature . by analytic continuation of the 2-sphere in ( [ kk ] ) to a 2-de sitter making @xmath39 and @xmath40 we get the expression of the 5d kk - bubble of nothing @xcite @xmath41 + d \psi^{2}.\ ] ] another interesting particular case is the black string with arbitrary tension @xcite .
performing in ( [ sph ] ) the coordinate transformation @xmath42 the metric is transformed into : @xmath43 + d^{\frac{4}{\sqrt{3}}a}d \omega^{2},\ ] ] where @xmath44 .
parametrizing with @xmath45 related to @xmath46 by @xmath47 the metric ( [ a22 ] ) is written as @xmath48 in this parametrization the solution has only two free parameters @xmath49 and we identify it as the black string with arbitrary tension previously derived in @xcite and studied in @xcite .
notice that we have given a much simpler derivation of the solution .
rod structure of the 4d zipoy - vorhees solution , width=325,height=151 ] rod structure of the 5d black string generated from zipoy - vorhees and inserting a finite rod with scalar charge @xmath50 in the interval @xmath51 in the fifth dimension .
the bold dotted line along @xmath52 is a rod with negative mass density ( conical singularity).,width=325,height=151 ] it is illustrative to sketch briefly how the generation method works in terms of the rod - structure @xcite , @xcite . the zv seed solution , eqs .
( [ a1])-([a1.1 ] ) , is characterized by the rod - structure shown in fig .
it differs from schwarzschild only in the mass density along the killing directions : for schwarzschild the density in the interval @xmath53 $ ] along the @xmath54 direction is 1 , while for zv the density is @xmath55 . in the intervals
@xmath56 $ ] and @xmath57 $ ] along @xmath52 the density is also @xmath55 ( @xmath13 for schwarzschild ) .
introducing the scalar field @xmath8 and then lifting to 5d in terms of the rod structure amounts to introduce into the fifth dimension a rod of density @xmath58 .
the rod structure of the 5d black string , eqs .
( [ a6])-([a6.1 ] ) , is ( shown in fig .
[ fig2 ] ) the following : there is a density @xmath58 in the interval @xmath53 $ ] along the @xmath59 direction .
a conical singularity along the @xmath52 direction is shown as the negative density @xmath60 in the interval @xmath53 $ ] ( bold dotted line ) ; also there is a positive density of @xmath55 in the intervals @xmath61 $ ] and @xmath57 $ ] along the @xmath52 direction .
the density is @xmath62 in the interval @xmath53 $ ] along the @xmath54 direction ; to have a physical horizon , if one wants to keep calling the system a black string , the density along the timelike direction should be positive , i.e. @xmath63 . the invariant that defines trapped or marginally trapped surfaces for the generated solution ( [ a8 ] ) is analyzed in what follows . trapped surfaces @xmath64 and located horizons corresponding to the spacetime ( [ a8 ] ) , are determined by the scalar @xmath65 ( see @xcite ) . to calculate it ,
we fix the coordinates @xmath66 and denote the local coordinates on the surface @xmath64 by @xmath67 .
the function @xmath68 , gives the canonical volume element of the surfaces @xmath64 .
introducing @xmath69 , @xmath70 and defining @xmath71 the invariant @xmath65 for the obtained solution amounts to @xmath72 for the solution of our interest , @xmath73\ ] ] and therefore , @xmath65 is obtained as @xmath74[r - m(1-\cos\theta)]\}^{a^{2}+ \frac{q^{2}}{4}-3}[{\tilde v}_{,r}]^{2 } } { r^{a^{2}+\frac{q^{2}}{4}+\frac{a}{\sqrt{3}}+\frac{q}{2}+1 } ( r-2m)^{a^{2}+\frac{q^{2}}{4}-\frac{a}{\sqrt{3}}-\frac{q}{2}+1}}\ ] ] with @xmath75m^2r- ( 1+a^2 + \frac{q^{2}}{4}+q ) \sin^2{\theta}m^3 \nonumber\\ & = & ( r - m)[2r(r-2m)+m^{2}\sin^{2}\theta(a^{2}+ \frac{q^{2}}{4}+1)]-mq[r^{2}-2mr+m^{2}\sin^{2}\theta],\end{aligned}\ ] ] let us analyze the marginally trapped surfaces defined when @xmath76 as well as the singularities if @xmath65 diverges . to determine those values of @xmath65 cancelations in the denominator coming from @xmath77 should be taken into account .
assuming that @xmath78 and @xmath79 , then there exists a singularity at @xmath80 since the exponent of @xmath81 in the denominator of ( [ a12 ] ) is always positive , @xmath82 .
also there are two marginally trapped surfaces at @xmath83 , nevertheless these are wrapped or hidden by the naked singularity located at @xmath84 . depending on the values of @xmath27 and @xmath12 , in the product of @xmath85 and @xmath86 may be cancelations that define horizons that coincide with the naked singularity @xmath84 .
another kind of marginally trapped surfaces that overlap the singularity in some regions , arise from the real root @xmath87 of the third degree polynomial @xmath85 .
these surfaces when @xmath88 do not cover completely the singularity due to the dependence on @xmath89 .
then the situation when @xmath88 is that for some ranges of @xmath89 the singularity is hidden but coming from other direction the singularity is naked , i.e. the singularity is partially naked ( or partially dressed ) . for the black string with arbitrary tension ( @xmath90 ) ,
@xmath65 amounts to @xmath91 the solution is singular at @xmath84 unless @xmath13 ( @xmath92 ) , case in which it possesses a horizon at @xmath84 ( @xmath93 ) . in any other case ( @xmath94 ) there is a naked singularity at @xmath84 that hides horizons at @xmath95 .
the @xmath65 diverges at @xmath80 showing the singularity there . in the case @xmath34 and @xmath13 that we recover the black string solution @xcite , it possesses a horizon at @xmath84 as the expression for @xmath65 shows , @xmath96 in this case the naked singularity disappears and instead a horizon is located at @xmath84 . with respect to the analysis of the singularities we have attempted to draw penrose diagram to dilucidate the global structure of the solutions .
however , due to the reduced symmetry of the line element ( [ a8 ] ) the metric function @xmath30 does not depend only on @xmath81 but also on @xmath89 .
even if one drops the @xmath89 dependence , by fixing the angle , the integration to be performed involve hypergeometric functions , and the analysis by this method does not produce a clear picture . on the other hand , the information one obtains by studying the scalar @xmath65 , is essentially the same that can be obtained from the kretschmann scalar . to characterize the 5d black string we need to determine the mass and tension of the source using the asymptotic expansion of the metric functions @xmath97 and @xmath98 of ( [ a8 ] ) , @xmath99 comparing with the asymptotic form of metric around a stationary matter source as given in @xcite , @xcite : @xmath100
the mass and string tension of the 5d black string ( [ a8 ] ) are identified then as ( we are considering @xmath101 ) @xmath102 therefore when ( [ a1 ] ) is lifted to 5d , by means of the scalar field @xmath103 , the string tension changes in @xmath104 , that can modify the tension and even make it vanish by calibrating @xmath105 , however , from the rod - structure we learned that @xmath106 in order to have a horizon . in the case @xmath34 and @xmath13 we recover the black string solution @xcite , with @xmath107 . the case @xmath108 and @xmath37 corresponds to a black string with @xmath109 .
in this paper by equiping first the 4d zipoy - vorhees solution with a scalar field , we lifted it to five dimensions ; the so generated 5d solution is a black string characterized by three free parameters : mass , tension and deformation parameter .
there is a physical spacetime singularity at @xmath80 .
moreover , the horizons do not cover completely the naked singularity at @xmath84 , except in the case that @xmath84 is the horizon itself .
therefore , in 5d the cosmic censorship conjecture does not hold .
the static black string with arbitrary tension is a particular case when there is transversal spherical symmetry .
the stability of the 5d generalized black string deserves further analysis , particularly to acquaint how the deformation parameter @xmath27 can affect the stability of the solution , since it modifies the string tension .
we have seen that the three parameter generalizations of the black string solutions are singular .
this may well be an indicator that the black strings are unstable in the sense of @xcite , @xcite .
a.f . acknowledges the support of the basque government grant gico7/51-it-221 - 07 and the spanish science ministry grant fis2007 - 61800 .
n. b. would like to thank the colleagues of upv / ehu for warm hospitality .
l. a. lpez acknowledges conacyt - mxico for a ph .
d. grant . partial support of conacyt - mexico project 49182-f is also acknowledged . | starting with a zipoy - voorhees line element we construct and study the three parameter family of solutions describing a deformed black string with arbitrary tension . |
the calculation of steady - state transport properties of open quantum systems such as quantum dots is a challenging and unsolved problem .
perturbative methods @xcite may be used to study the weak correlation regime , but they fail to provide a reliable description of the competition between kondo- and coulomb - blockade physics in strongly interacting dots @xcite . to avoid these limitations of conventional perturbation theory ,
various non - perturbative numerical approaches have been developed .
time - dependent density - matrix renormalization group ( tdmrg ) calculations @xcite and real - time monte carlo ( rt - mc ) approaches @xcite try to compute the relaxation into the interacting steady state after some switching of parameters , such as voltage bias or interaction . while the short - time transients can be very accurately captured with these methods @xcite , the approach to the steady - state may occur on rather long , in the worst case exponentially large times scales . due to finite - size effects in the tdmrg and an exponentially growing sign problem with increasing time in rt - mc , the access to long times
is severely limited in both approaches .
furthermore , the tdmrg is performed for a finite , closed system ; whether a relaxation to a reasonable approximation of the interacting steady - state is guaranteed for some intermediate time scale much smaller than poincar s recurrence time is not obvious .
this latter problems may be avoided by numerical renormalization group ( nrg ) @xcite and functional renormalization group ( frg ) calculations @xcite , which attempt a direct description of the non - equilibrium steady state .
however , the former introduces an artificial discretization and truncation of the spectrum of the hamiltonian , which can lead to artifacts in the time evolution .
the frg , on the other hand , is again perturbative in nature , and experience up to now shows that it works best in the extreme non - equilibrium limit @xcite .
none of the methods developed so far is able to provide a complete and reliable description of the model in all parameter regimes .
more importantly , the most interesting regime , where all relevant energy scales voltage , temperature , magnetic field etc .
are of the same order as the relevant low - energy scale of the model , is usually the one which is not accessible .
therefore , the development of new or improved simulation approaches is a worthwhile and important task .
recently , a new and rather unconventional approach to calculate the steady - state transport through interacting quantum dots or similar structures was proposed by han and heary @xcite .
their formalism , which is based on hershfield s density operator @xcite , maps the non - equilibrium steady - state of the interacting model onto an infinite set of auxiliary equilibrium systems , each characterized by some complex voltage .
the appealing feature of this approach is that powerful methods exist for the numerical solution of equilibrium models .
the complexification of the voltage bias , however , introduces a formidable new problem in the form of an analytical continuation in the voltage on top of the already challenging analytical continuation from matsubara frequencies to real frequencies . in ref .
@xcite this double analytical continuation was performed using a phenomenological formula based on general structures of the self - energy found in second order perturbation theory .
the purpose of this study is to explore to what extent an unbiased numerical implementation of the method by han and heary is feasible .
we will address two issues : ( i ) the use of recently developed , accurate continuous - time quantum monte - carlo ( ct - qmc ) algorithms to simulate quantum impurity models as solvers for the effective equilibrium impurity problems with complex voltage bias ; and ( ii ) the analytical continuation of matsubara frequency data via some maximum entropy method . in particular , we will compare the performance of the weak - coupling @xcite and hybridization expansion @xcite algorithms and propose a kernel for the maximum entropy ( me ) procedure which is compatible with the analytical properties of the green function .
the paper is organized as follows .
section [ han ] describes the imaginary - time approach to steady state transport by han and heary . a brief introduction to the ct - qmc for equilibrium problems and their suitability for models with complex voltage bias follows in section [ qmc ] .
section [ maxent ] is devoted to the issue of analytical continuation in the voltage and frequency domain and presents some results for equilibrium and non - equilibrium situations .
we will finish the paper with a conclusion and outlook in section [ conclusion ] .
we briefly review the imaginary - time formulation of steady - state transport through an interacting quantum dot proposed by han and heary @xcite , which is based on the work of hershfield @xcite .
we consider a spin - degenerate , single - level quantum dot attached to two non - interacting fermionic leads .
this system can be described by the single - impurity anderson model with hamiltonian ( @xmath0 ) @xmath1 where @xmath2 and @xmath3 label the left and right reservoirs , respectively .
the index @xmath4 denotes the wave - vector of the lead states and @xmath5 the spin quantum number .
a gate voltage @xmath6 may be applied to shift the dot energy level position relative to the particle - hole symmetric configuration @xmath7 . to keep things simple , we assume a @xmath4-independent hybridization @xmath8 and
consider the wide - band limit for the dispersion of the leads .
we then end up with a bare level broadening @xmath9 , @xmath10 , where @xmath11 denotes the density of states of the leads at the fermi energy . in the case of non - equilibrium steady - state transport
, the leads are supposed to be unaffected by the current flowing through the dot and characterized by free fermion correlators @xmath12 with @xmath13 the fermi distribution function for inverse temperature @xmath14 and @xmath15 the value of the chemical potential for lead @xmath16 . we restrict ourselves to the case where the inverse temperatures of the left and right lead are the same , @xmath17 , and symmetrically applied voltage bias , @xmath18 .
the bias voltage is denoted by @xmath19 . in ref .
, hershfield introduced a hermitian operator @xmath20 by means of which the non - equilibrium , steady - state expectation value of a local observable @xmath21 may be written as @xmath22 the above expectation value is of the form @xmath23 , and hence resembles the equilibrium expression . under certain assumptions involving the non - trivial exchange of limiting procedures ,
the operator @xmath20 can be expressed as @xmath24 where the scattering states @xmath25 are related to the bare conduction states @xmath26 by the second - quantized lippmann - schwinger equation @xcite @xmath27 the liouvillians are defined as @xmath28 $ ] and @xmath29 $ ] , with @xmath30 the hybridization part of the hamiltonian .
the @xmath31 " denotes the operators after @xmath32 , and the fraction in eq .
( [ lseq ] ) denotes the corresponding geometric series in @xmath33 , i.e. a series of iterated commutators with @xmath34 . for @xmath35
it is impossible to calculate an explicit expression for the @xmath20-operator . more importantly ,
although @xmath36 looks like an effective hamiltonian for the system , it can not be used to define a consistent description of imaginary - time and real - time dynamics .
the real - time dynamics is always controlled by @xmath34 alone , but @xmath34 and @xmath36 will in general have a different spectrum . therefore , the analytically continued imaginary - time dynamics does not reproduce the real - time dynamics . since @xmath36 does not yield the correct real - time dynamics , han and heary @xcite introduce an additional trick .
starting with a fully established non - interacting steady - state ensemble at time @xmath37 , the fully interacting steady state is formally reached by propagating the system to @xmath38 . in a path integral representation the expectation value for an observable @xmath21 becomes @xmath39 a(\{\psi_{\alpha k \sigma}^\dagger(0 ) , \psi_{\alpha k \sigma}^{\phantom{\dagger}}(0)\ } ) { \ensuremath{\mathrm{e}^{{\ensuremath{\mathrm{i}}}\int_0^\infty l(t){\ensuremath{\,\mathrm{d}t\ , } } } } } \right\rangle_0.\ ] ] here , the average @xmath40 is performed using eq . with @xmath41 and @xmath42 , where @xmath43 can be explicitly constructed using non - interacting scattering states .
it was argued in ref . that the time evolution via @xmath34 maps the non - interacting scattering states to the interacting ones and the lagrangian for the real - time evolution reads @xmath44 aiming at a description which yields @xmath45 as real - time evolution operator for @xmath46 and @xmath47 as imaginary - time evolution operator for @xmath48 , the lagrangian is reexpressed with respect to the spectrum of @xmath36 , @xmath49 .
statistical expectation values take a form analogous to equilibrium expectation values , with a uniform fermi level @xmath50 .
due to the discrepancy between @xmath34 and @xmath36 , the real - time lagrangian transforms to @xmath51 , so the effective fermi levels of left and right leads have different time evolution rates .
these rates can be factored out as time - dependent phase factors of the grassmann fields by introducing new field variables @xmath52 .
the extra time evolution rate is generated by @xmath53 acting on the phase factor , and thus @xmath54 describes the correct time evolution .
to obtain a matsubara - like theory , the fields @xmath55 are now wick rotated , @xmath56 . however , under the replacement @xmath57 the exponential factor becomes @xmath58 , which means that it diverges as @xmath59 and decays as @xmath60 . to circumvent this problem ,
han and heary introduce a second analytic continuation to ensure matsubara s periodic boundary conditions and thereby obtain a well - defined effective equilibrium system .
this is achieved by complexifying the voltage occurring in the extra time evolution rate according to @xmath61 , @xmath62 . for the particular choice @xmath63 the matsubara boundary conditions
are conserved @xcite .
the final result of these manipulations is that both the lagrangian and the fields now have their time evolution with respect to the effective equilibrium hamiltonian @xmath64 . in a perturbative expansion around the non - interacting limit , one may then switch to the interaction picture with respect to the non - interacting effective hamiltonian @xmath65 .
as before , @xmath43 is hershfield s boundary condition operator for the corresponding fully established non - interacting steady state , for which an explicit expression can be given . we may now proceed along the usual lines and integrate out the conduction electron degrees of freedom to obtain an effective action @xmath66 for the electrons on the dot .
as we are by construction in the stationary state , the bare dot green s function @xmath67 appearing in the quadratic term in the action ( [ action ] ) depends on the time difference only .
we therefore may perform a fourier transform to fermionic matsubara frequencies and find the form @xcite @xmath68 with @xmath69 , @xmath70 , and @xmath71 .
the desired green s function for the stationary state of the interacting system is finally obtained by solving the quantum impurity problem for each @xmath72 , @xmath73 , performing the analytical continuation @xmath74 and evaluating the resulting expression at the physical voltage @xmath75 .
although the preceding discussion seems to be based on simple manipulations of the functional integral , one has to show formally the equivalence of the complexified auxiliary equilibrium time - evolution based on the action ( [ action ] ) and the actual physical time evolution with respect to @xmath34 as given by ( [ eq : true_t_evolution ] ) after the analytical continuation @xmath76 in the former .
up to now such a formal proof is still lacking , only an argument based on the inspection of the contributions to perturbation expansion has been put forward @xcite .
it is therefore interesting to see if an unbiased numerical implementation of this formalism is possible and produces physically meaningful results .
in order to compute the self - energy from action ( [ action ] ) as a function of matsubara frequency we employ continuous - time monte carlo ( ct - qmc ) solvers . the continuous - time monte carlo technique in the weak - coupling @xcite and hybridization expansion @xcite formulation has been discussed in considerable detail in the literature and we will present here merely a short summary of the formalism . the idea is to expand the partition function @xmath77 $ ] into a series of diagrams , and to sample ( collections of ) these diagrams by a monte carlo procedure .
we split the hamiltonian @xmath34 of the impurity model into two parts , @xmath78 and @xmath79 , and employ an interaction representation in which the time evolution of operators is given by @xmath78 : @xmath80 . in this interaction representation
, the partition function can be expressed as a time ordered exponential , which is then expanded into powers of @xmath81 , @xmath82 = \sum_{n=0}^\infty \int_0^\beta d\tau_1\cdots \int_{\tau_{n-1}}^\beta d\tau_n { \ensuremath{\mathrm{tr}\,}}\big [ e^{-(\beta-\tau_n)h_1}(-h_2 ) \cdots e^{-(\tau_2-\tau_1)h_1}(-h_2)e^{-\tau_1h_1}\big ] .
\label{z_interaction_picture}\ ] ] equation ( [ z_interaction_picture ] ) represents the partition function as a sum over monte carlo configurations @xmath83 ; @xmath84 , @xmath85 , @xmath86 , @xmath87 with weight @xmath88d\tau^n . \label{weight0}\ ] ] two types of expansions have been considered . in the weak - coupling approach
@xcite the partition function is expanded into powers of the interaction , @xmath89 , while the time evolution between operators is given by the quadratic part of the hamiltonian , @xmath90 .
the monte carlo configuration becomes a collection of interaction vertices on the imaginary time interval and the weight ( [ weight0 ] ) evaluates to @xmath91d\tau^n . \label{weight}\ ] ] here @xmath92 is an @xmath93 matrix whose elements are noninteracting green functions evaluated at all time intervals defined by the vertex positions .
note that in the case of half filling of interest here , only even perturbation orders appear in the expansion .
away from half - filling , odd perturbation orders become relevant and ising - type auxiliary fields must be introduced to avoid or reduce the sign problem .
we will in this paper employ the continuous - time auxiliary field algorithm described in ref .
@xcite , which for models with density - density interactions and an appropriate choice of parameters is equivalent to the weak - coupling algorithm @xcite .
the alternative approach is the hybridization expansion @xcite where the partition function is expanded in powers of the hybridization term , @xmath94 while the time evolution between operators is given by the impurity plus bath part of the hamiltonian .
this time evolution no longer couples the impurity and the bath .
it therefore becomes possible to integrate out the bath degrees of freedom analytically to obtain @xmath95 \nonumber\\ & & \times \det m^{-1}(\{\tau_1 , \alpha_1\},\ldots,\{\tau_{n},\alpha_n\ } ; \{\tau_1',\alpha_1'\},\ldots,\{\tau_{n}',\alpha_n'\ } ) ( d\tau)^{2n}. \label{weight_strong}\end{aligned}\ ] ] the configurations @xmath96 are now collections of @xmath97 time arguments @xmath98 corresponding to annihilation operators with flavor indices @xmath99 and @xmath97 time arguments @xmath100 corresponding to creation operators with flavor indices @xmath101 .
the element @xmath102 of the matrix @xmath103 is given by the hybridization function @xmath104 , which is defined in terms of the hybridization parameters @xmath105 and the bath energy levels @xmath106 @xcite .
in a model with density - density interactions only , one can separate the operators according to flavors , which leads to the so - called segment representation @xcite .
this segment representation allows a simple and efficient evaluation of the trace over the impurity states in eq .
( [ weight_strong ] ) .
the implementation of the weak - coupling ct - qmc for the action is straightforward .
the non - interacting green s function is being fourier - transformed and the resulting @xmath107 inserted into eq . .
the implementation of the hybridization approach is more subtle , as except in the equilibrium limit @xmath108 , @xmath109 the hybridization function @xmath104 which appears in the action lacks a physical meaning , because it is not directly related to the hopping amplitudes @xmath110 in the physical hamiltonian .
however , the hybridization function is implicitly defined by rewriting the effective action as @xcite @xmath111 , with @xmath112 and @xmath113 .
consequently , the hybridization function can be constructed from as @xmath114 after straightforward algebraic manipulation , we obtain @xmath115 + \sum\limits_{\alpha=\pm 1 } [ { \ensuremath{\mathrm{i}}}\omega_n - { \varepsilon}_d + \alpha ( { \ensuremath{\mathrm{i}}}\varphi_m-\phi ) ] \frac{{\ensuremath{\mathrm{i}}}\gamma}{2 } { \ensuremath{\mathrm{sgn}}}(\omega_n -\frac{\alpha}{2}\varphi_m ) - \left(\frac{{\ensuremath{\mathrm{i}}}\varphi_m - \phi}{2}\right)^2 } { { \ensuremath{\mathrm{i}}}\omega_n - { \varepsilon}_d + \frac{{\ensuremath{\mathrm{i}}}\gamma}{2 } \sum\limits_{\alpha = \pm 1 } { \ensuremath{\mathrm{sgn}}}(\omega_n - \frac{\alpha}{2 } \varphi_m ) } \;\;.\ ] ] note that the expression @xmath116 emerges from imposing the wide - band limit for the leads .
the hybridization approach is only able to cope with finite bands , because in the limit of infinitely wide bands of constant dos , the expansion order diverges .
the @xmath117-function must therefore be replaced by a sufficiently well - behaved function corresponding to a finite bandwidth and thus decaying rapidly enough for large frequencies @xmath118 .
the high - frequency behavior of expression is given by @xmath119 which means that the numerical evaluation of eq .
requires some care .
conventionally , one regularizes the sum by analytically evaluating the dangerous parts and then numerically calculating the difference between the full function and the problematic parts , i.e.@xmath120{\ensuremath{\mathrm{e}^{-{\ensuremath{\mathrm{i}}}\omega_n\tau}}}\;\;.\ ] ] the leading order high - frequency tail @xmath121 results in a constant shift @xmath122 in @xmath123 , @xmath124 .
the first term @xmath125 in the high - frequency expansion yields @xmath126 and diverges for @xmath127 and @xmath128 .
these divergences are a direct consequence of the wide - band limit , i.e. we need to regularize them in order to be able to use the hybridization expansion algorithm .
this regularization is introduced by cutting the divergences with a sufficiently large cutoff parameter @xmath129 , i.e. we use @xmath130 in practice , the value @xmath131 was used .
the contribution @xmath132 is fourier transformed easily by accumulating the series numerically .
note that the term @xmath133 has , besides the additional oscillations from the cosine modulation in eq . , the same structure as in the plain equilibrium anderson model , where @xmath134 .
we will therefore illustrate the properties of the quantity @xmath135 in the following section . .
the upper panel shows the non - trivial contribution @xmath136 , eq . to the hybridization function @xmath123
, the lower panel shows the imaginary - time green s function @xmath137 . raising @xmath138 leads to increasingly oscillating imaginary - time green s functions and hybridization functions .
the oscillations need to be resolved well by the qmc solver in order to guarantee an unbiased solution . as implied by eq .
a strong negative shift @xmath122 occurs in the hybridization function when sweeping through the region @xmath139 .
the imaginary parts @xmath140 and @xmath141 are small and also show oscillations .
, scaledwidth=49.0% ] typical input data for both , the weak - coupling and the strong - coupling approach , are shown in fig .
[ fig : imagtimedata ] . with increasing
imaginary voltage @xmath138 , oscillations with @xmath142 nodes occur in both , the imaginary - time green s function and the hybridization function .
moreover , the shift grows quadratically , introducing a strong shift of the hybridization function towards negative values . the strongly oscillatory behavior for large @xmath138 makes a correspondingly fine resolution of the imaginary - time interval necessary .
in a standard hirsch - fye algorithm @xcite , the interval @xmath143 has to be represented by a comparatively small and fixed number of equidistant mesh points , i.e. these oscillations can not be adequately resolved .
this limitation does not apply to ct - qmc , and it is hence the method of choice to access also large @xmath138 .
in contrast to the equilibrium case , complex sampling weights @xmath144 are obtained in both the weak - coupling and strong coupling formulation . as usual , one uses the modulus @xmath145 of the weight to determine the acceptance probability , while the phase @xmath146 has to be treated as additional observable .
usually , such an approach leads to a sign problem and severely limits the applicability of the monte - carlo simulations .
therefore , we must anticipate a generalized sign problem , i.e. @xmath147 exponentially or worse .
the situation is especially problematic for the hybridization expansion due to the additional shift towards negative values . indeed , as illustrated in fig .
[ fig : phase ] the sign problem becomes increasingly severe with increasing imaginary voltage @xmath138 , limiting this algorithm to small @xmath138 . from fig .
[ fig : phase ] it also becomes clear that the sign problem in the weak - coupling ct - qmc simulations is much milder and this approach allows us to simulate impurity models with large @xmath138 . and @xmath148 ( eqs . and ) of the weak - coupling ( solid lines ) and the strong - coupling ( dashed lines ) solver , respectively , as a function of the imaginary voltage . on a logarithmic scale
, the average phase decays faster than linearly for the strong - coupling approach when @xmath138 is increased .
no strong dependence on @xmath138 is found for the weak - coupling algorithm .
, scaledwidth=49.0% ] to demonstrate the quality of the imaginary - time data which can be obtained with the weak - coupling ct - qmc method , we show in fig .
( [ fig : selfenergy ] ) the imaginary part of the matsubara axis self - energy computed for @xmath149 , @xmath150 , @xmath151 and @xmath152 ( @xmath153 ) , @xmath154 ( @xmath155 ) , @xmath156 ( @xmath157 ) , and @xmath158 ( @xmath159 ) .
the equilibrium kondo temperature for this parameter set is @xmath160 , i.e. we are reasonably deep in the kondo regime of the anderson model .
moreover , the values for @xmath161 and @xmath162 are such that @xmath163 and @xmath164 , i.e. precisely in the parameter region which is hard or impossible to access for other methods . even for large complex voltage
the accuracy of the numerical data is very good ( error bars on the order of the line width ) for both small and large matsubara frequencies .
in contrast to the results presented in ref . , which are based on discrete - time hirsch - fye simulations , no discontinuities
are observed for @xmath165 in the ct - qmc data .
we note , however , that a recent preprint @xcite reports a trick by use of which this issue could be resolved within the discrete - time formalism . , @xmath166 , @xmath151 and @xmath167 .
the equilibrium kondo scale here is @xmath168 .
we easily obtain high - quality data for all values @xmath155 , @xmath169 and @xmath170 of the complexified voltage , even in this most challenging parameter regime @xmath171 , @xmath164 and @xmath163 .
each @xmath142-value was run on a single intel xeon(r ) e5345 cpu for approx .
24 hours , so the data were obtained with relatively moderate computational effort .
, scaledwidth=49.0% ]
as noted in ref .
@xcite , at finite interaction , branch cuts occur for @xmath172 ( @xmath173 odd ) in the complexified green s function @xmath174 . introducing the complex vector variable @xmath175 we hence assume the green s function to be holomorphic as a function of two complex variables in domains @xmath176 , where for @xmath177 @xmath178 are the cones emerging from the branch cut condition for positive ( @xmath179 ) or negative ( @xmath180 ) imaginary voltages ( see illustration in fig .
[ fig : cones ] ) . note that domains like @xmath181 are well - known objects in the theory of functions of several complex variables and are called tubular cone domains . for a good introduction
see , e. g. , ref .
@xcite . in ref .
@xcite this structure is described by the cauchy representation @xmath182 for the corresponding self - energy .
however , eq . is only approximate , because the @xmath72-dependence of the functions @xmath183 is not taken into account .
such a non - trivial dependence appears as a result of higher - order corrections in @xmath184 .
let us start by discussing the analytically continued bare green s function @xmath185 with @xmath186 . .
branch cuts occur for @xmath187 , with @xmath188 ( solid lines ) , for @xmath189 , but also at @xmath190 ( dash - dotted lines ) , @xmath191 ( dashed lines ) , @xmath192 ( dotted lines ) , and so on , for @xmath193 . concentrating on the retarded sector of the green s function , @xmath194
, we introduce the cones @xmath195 bounded by the branch cuts with imaginary - part ratios @xmath196 and @xmath197 . adding the real subspaces @xmath198 , the tubular cones @xmath199
are obtained as domains of holomorphy . , scaledwidth=49.0% ] the corresponding geometric structure of the complex space is depicted in fig .
[ fig : cones ] , the branch cuts given by the black lines @xmath200 .
note that the green s function does not vanish for all directions within a given @xmath181 as @xmath201 .
[ textstelle : g_bounded ] on the other hand , @xmath202 is at least bounded , and we assume that nonzero interactions do not alter this fundamental property .
[ sec : shift_argument ] one can thus always find a constant @xmath203 such that the imaginary part of the function @xmath204 is positive .
integral representations of the form @xmath205 which are valid for the class of holomorphic functions with non - negative imaginary part also hold for @xmath206 , since @xmath207 is also a function with non - negative imaginary part .
this class of functions on tubular cone domains was extensively studied by mathematicians . in ref .
@xcite , vladimirov finds a generalization of herglotz - nevanlinna representations @xcite to such domains .
see appendix [ sec : appa ] for details .
the validity of the imaginary - voltage formalism is presently based on the assumption of asymptotic convergence of the perturbation series in @xmath184 .
thus , the influence of the branch cut between @xmath208 and @xmath181 is expected to become negligible as @xmath209 , i.e. all branch cuts with @xmath210 can be ignored .
the maximal value @xmath211 may for example be estimated from the expansion order histogram of the weak - coupling qmc simulation , since a given branch cut with index @xmath212 is only established by diagrammatic contributions with order larger than a certain value @xmath97 , which is roughly proportional to @xmath213 . as stated in ref . @xcite we are required to first take the limit @xmath214 and then @xmath215 . in our language , the spectral function is given by @xmath216 since branch cuts with index @xmath217 vanish we choose the domain @xmath218 with @xmath219 and @xmath220 for the analytic continuation of the interacting green s function .
this choice of domain is illustrated in fig .
[ fig : cont2dgeometry ] . in practice ,
the critical branch cut is yet chosen arbitrarily but to be small , see section [ maxent ] .
the branch cut @xmath221 ( dotted line ) is negligible , while the critical branch cuts @xmath222 are not .
the green s function is therefore holomorphic in the cone domain @xmath223 bounded by the ratios @xmath224 , with @xmath225 given by eq . .
investigating the green s function at the edge of this domain is compatible with the limiting procedure of taking @xmath226 and then @xmath227 for the spectral function @xmath228 .
this is indicated by the bold dash - dotted arrow . using the integral representation ,
a most likely limit of the green s function at the edge , @xmath229 , @xmath230 , will be inferred from the qmc data @xmath231 in the domain using a maximum entropy method ( sec .
[ maxent ] ) .
the spatial locations of the qmc data points in the domain are symbolized by the crosses . in the case of strong interaction , for small matsubara frequencies
we are limited to small values of @xmath138 .
, scaledwidth=49.0% ] as shown in appendix [ sec : appa ] the poisson kernel representation resulting from vladimirov s theorem is @xmath232 with @xmath233 where @xmath234 and @xmath235 are the real and imaginary parts of @xmath236 .
the numerical analytic continuation of imaginary - time quantum monte carlo data is a highly ill - posed problem . even if the finite set of qmc data did not contain
any stochastic noise there would exist an infinite - dimensional manifold of solutions to the integral equation associated with the continuation , i.e. the spectral representation @xmath237 ( \omega_n ) \label{eq : kernel1d}\ ] ] for the conventional continuation problem .
hence , a regularization procedure picking a most probable " solution is required .
typically , this is approached with a maximum entropy method ( mem ) , a rigorous framework rooted in bayesian logic which can be understood as an automatic ockham s razor , in the sense of being maximally noncommittal with regard to missing information " @xcite . the spectral function @xmath228 is interpreted as a probability distribution .
a default model @xmath238 is introduced as a - priori information about the solution @xmath228 .
additional information , given by the measured imaginary - time data @xmath239 , is inferred through the kernel @xmath240 $ ] in . if there is no additional information the procedure will pick @xmath241 , in bryan s mem algorithm @xcite . in practice ,
a functional @xmath242 = \chi^2[a ] - \alpha s[a],\quad \alpha > 0\ ] ] is minimized in the space of candidate solutions for a given hyper - parameter @xmath16 .
the qmc data must be gaussian distributed , such that the likelihood penalty @xmath243 $ ] is given by @xmath244 = \frac{1}{2}\sum_{\rho,\eta=1}^n ( \bar g_\rho - k_\text{eq}[a]_\rho)c^{-1}_{\rho \eta } ( \bar g_\eta - k_\text{eq}[a]_\eta ) , \label{eq : chisquared}\ ] ] where @xmath245 are the measured mean real or imaginary parts of the imaginary - frequency green s function @xmath246 , and @xmath247 are the elements of the inverse covariance matrix .
the default model @xmath238 is invoked through the entropy @xmath248 = \int { \ensuremath{\,\mathrm{d}{\varepsilon}\ , } } \left [ a({\varepsilon } ) - d({\varepsilon } ) - a({\varepsilon } ) \log \frac{a({\varepsilon})}{d({\varepsilon } ) } \right ] .
\label{eq : entropy}\ ] ] for a detailed theoretical justification of this choice for the entropy see ref .
@xcite .
the easiest way of fixing the regularization parameter @xmath16 is to employ the condition @xmath249 ( historic mem ) .
it is , however , more reasonable to calculate a posterior probability distribution @xmath250 .
setting @xmath16 to the maximum of the posterior probability distribution is called classic mem .
marginalizing @xmath16 by choosing @xmath250 as weights for @xmath21 when integrating over @xmath16 is empirically found to be most suitable and is also most justified from the theoretical point of view ( bryan s mem ) . in order to adapt the above procedure to the double analytic continuation problem
, a non - negative quantity has to be found which 1 .
uniquely represents any possible function in the data range of interest
say @xmath223 in order to define a @xmath251 for inference ; 2 . easily allows calculating the non - equilibrium spectral function @xmath228 .
we choose @xmath252 as such a representation , since due to the kramers - kronig relations and the validity of the representation , @xmath253 yields a unique and simple representation of all possible functions @xmath254 .
the non - equilibrium spectral function is easily accessible , since @xmath255 . in the case of zero interaction , @xmath256
it is easy to verify that @xmath257 is a positive function with @xmath258 if one constrains the @xmath259-integration to an arbitrary finite interval of length @xmath260 .
this fact and the fact that @xmath261 do not imply @xmath262 in general .
we however assume @xmath263 and expect to obtain revealing signatures within the mem , in case the real @xmath253 is not positive definite for a given data set .
note that even in the presence of regions where @xmath264 , a mem can be implemented , by identifying the nodes of @xmath253 , as in the case of bosonic spectral functions .
in general , positivity may be enforced by adding a positive real constant @xmath265 to the spectral function and adding a corresponding term to the image .
as particular example for this procedure , we quote here the case of the nambu off - diagonal green s function @xmath266 , where the positivity is enforced as @xmath267 @xcite .
we hence choose as a kernel function for the @xmath251 functional and only take data in @xmath223 into account .
the entropy expression is adopted for a two - dimensional default model @xmath268 . first note
that since the input data for the poisson kernel are obtained from statistically independent qmc simulations , the covariance @xmath269 in the @xmath251 functional has a block - diagonal shape @xmath270 the submatrices @xmath271 are covariances for the subset of data @xmath272 at a fixed @xmath138 , estimated from the output of the corresponding equilibrium qmc simulation .
our implementation of the maximum entropy method is based on bryan s standard algorithm introduced in ref .
a singular value decomposition ( svd ) of the kernel @xmath273 is performed , with @xmath110 , @xmath274 orthogonal , and the singular values @xmath275 @xmath276 .
many important quantities may be reduced to the @xmath277-dimensional singular space @xmath278 .
most notably , the @xmath279-dimensional optimization problem given by @xmath280\stackrel{!}{=}\text{min } \label{eq : optimprob}\ ] ] may be solved within the singular space using levenberg - marquardt iterations . as @xmath277 is comparably small after truncating the singular space with respect to the floating point precision of the singular values @xmath281 ( typically , @xmath282 ) ,
the algorithm is still sufficiently efficient , even though a two - dimensional frequency grid is required for the numerical resolution of @xmath253 , and hence @xmath283 .
the algorithm enables us to calculate several important data qualifiers and posterior probabilities and therefore to classify both input data quality and candidate solutions .
the posterior @xmath284 with @xmath285\ , { \ensuremath{\mathrm{e}^{-\chi^2/2}}}$ ] , @xmath286 , and the jeffreys prior @xcite @xmath287 , is calculated using a gaussian approximation for @xmath288 , centered around the solution @xmath289 of eq . .
the usual procedures and strategies for data qualification and improvement of results as described in @xcite are adopted : assuming a flat prior @xmath290 , the posterior for the default model @xmath291 is computed easily .
@xmath292 serves as evidence for the quality of prior information when comparing within sets of default models for given qmc data .
whereas a posterior probability for the domain parameter @xmath293 for given data and given default model , @xmath294 , would be a sensible extension to the algorithm , we have not derived it yet .
useful ingredients might be found in the literature on blind deconvolution in signal processing , see @xcite . in our implementation , a small enough @xmath293 is chosen a priori . picking appropriate data sets with well - estimated covariance from the qmc output
is also a non - trivial part of the problem .
a good check is to determine the most probable mock error rescaling @xmath5 where the covariance @xmath269 is formally substituted by @xmath295 . if the most probable @xmath5 ( merit " ) , i.e. the solution of @xmath296 deviates from 1 by more than a few tens of percent , the input data are rejected @xcite .
@xmath297 is the @xmath251 value of the classic mem solution , the number of data points @xmath298 , and the number of good " data points @xmath299 with @xmath300 the eigenvalues of @xmath301_{\tilde a_\text{classic}}.\ ] ] in practice , a maximal matsubara frequency @xmath302 compatible with the error rescaling merit was determined , and all data @xmath303 in @xmath223 , with @xmath304 were used for inference .
presumably , better data selection strategies do exist .
for example , using independent measurements for @xmath305 and taking them into account by using a schwarz representation ( see appendix [ sec : appa ] ) could yield better results .
furthermore , the largest matsubara frequency index @xmath302 could be determined for each @xmath138 individually .
the latter appears to be necessary for non - equilibrium data . for the truncation of singular values ,
a threshold @xmath306 was used , @xmath307 for an @xmath308 by @xmath298 kernel matrix .
while for the conventional wick rotation @xmath309 was sufficient , @xmath310 had to be chosen in our case in order to take all relevant search directions in the @xmath253 space into account .
quadruple precision floating point arithmetic was found to be unnecessary . for discretizing the @xmath311 function , logarithmic meshes for the @xmath259 and
@xmath312 variables were used .
although @xmath313 does not decay for all directions as @xmath314 , choosing a finite mesh and truncating the integrals was not found to be critical . as a test case we consider the equilibrium limit @xmath108 .
the data for @xmath152 can be analytically continued with the standard wick rotation , using eq . and
the standard mem .
figure [ fig : spectra_equilibrium ] compares this 1d spectral function to the result based on the 2d data set @xmath272 and continued using the domain @xmath218 and the kernel function defined in eq . .
as default models for high temperatures we use lorentzians with variable width @xmath315 .
they read @xmath316 for the 1d continuation and @xmath317 for the 2d continuation , with @xmath318 .
an annealing procedure in the temperature was used for both , the 1d and 2d data for invoking adequate prior information , i.e. we used the @xmath253 solution of the next higher temperature as default model , starting with the lorentzian at the highest temperature . )
obtained using the conventional wick rotation ( 1d ) and the unconventional two - variable continuation ( 2d ) , with @xmath319 .
the domain parameter for 2d continuation is @xmath320 .
subfigure ( d ) shows the posterior probabilities @xmath321 of the default models as a function of @xmath315 ( eqs . and ) .
, scaledwidth=49.0% ] this default model selection procedure appears not to have any strict bayesian justification , however the physical argument is freezing out the high - frequency degrees of freedom and using present data for inferring low - energy details of the spectrum step by step @xcite .
a similar idea plays the key role in several modern renormalization group techniques .
note that gaussian default models are not well - suited for our data , since the high - frequency tail in the wide - band limit is lorentzian .
this manifests itself quantitively in the following way : for the gaussian default models we tested all had @xmath292 one order of magnitude lower than the lorentzian ones . for both , the gaussian and the lorentzian
, we can expect the quantity @xmath322 to be @xmath323 , due to the overall broadening introduced by a finite interaction @xmath184 . indeed , for the parameters @xmath319 , @xmath108 shown in fig .
[ fig : spectra_equilibrium](d ) , the ( unnormalized ) posterior probabilities @xmath321 and @xmath292 as a function of the parameter @xmath315 are peaked at @xmath324 for both , the 1d and 2d continuation procedures , respectively .
these probabilities were calculated for @xmath325 .
the most probable @xmath315 was chosen as default model .
however , a strong dependence of the results on @xmath326 was not observed .
the spectral functions shown in fig . [ fig : spectra_equilibrium ] were obtained for @xmath327 , @xmath328 , and @xmath169 .
we chose @xmath320 for the 2d domain , using @xmath329 , @xmath169 , @xmath330 for @xmath327 , @xmath328 , and @xmath169 , respectively .
note that due to the simple data selection strategy described in the previous section we only took into account data points with @xmath331 . using a global @xmath302 , the estimate for the covariance submatrix @xmath332 in eq
. eventually becomes singular , even though @xmath271 with @xmath333 and @xmath334 could still be estimated for a limited set of matsubara frequencies .
we expect that using such additional , well - estimated @xmath271 might lead to more structured spectral functions . in practice , however , the merit @xmath5 must yet be viewed as a rather crude measure of the quality of the covariance estimate .
so for the purpose of both simplicity and reproducibility we used the stronger restriction .
the kondo temperature for @xmath335 is @xmath336 , i.e.we can expect first signatures of strong coupling physics like hubbard bands and a temperature dependent quasi - particle peak of reduced width in the spectra . indeed both the 1d and 2d mem reproduces these features .
more importantly , the overall shape of the spectra obtained agrees for all temperatures shown in fig .
[ fig : spectra_equilibrium](a - c ) .
the results depend only slightly on the choice of @xmath315 for relevant values of @xmath292 .
although the spectra inferred from the 2d procedure using our current implementation appear to be less structured , the overall shape seems to be reconstructed quite well . for more serious calculations , the detailed high - frequency behavior ( and behavior for large @xmath337 )
should be introduced with a more sophisticated default model , e.g. based on perturbation theory . figure [ fig : defaultatilde ] shows the lorentzian default model we used at the highest temperature in the annealing procedure , @xmath338 . at the lowest temperature @xmath339 , the representation shown in fig .
[ fig : inferredatilde ] was obtained .
the equilibrium spectral function shown in fig .
[ fig : spectra_equilibrium](c ) is given by the cut @xmath340 .
other values of @xmath341 do not have any physical meaning . for first annealing step in equilibrium.,scaledwidth=49.0% ] inferred from the qmc data at the lowest temperature , @xmath339 , for the equilibrium test case shown in fig .
[ fig : spectra_equilibrium].,scaledwidth=49.0% ] note that certain structures appear in the inferred @xmath342 which vary as the domain parameter @xmath293 is changed : they occur for @xmath343 .
we interpret them as resulting from the properties of the kernel function discussed in sec .
[ sec : kernelstructure ] in combination with the mem principle of only incorporating changes which are strongly supported by data .
also , at larger distance from the origin , discretization errors from the discretization of the double integral are most dominant for this most structured region of the kernel . at finite bias ,
the qualitative structure of the inferred representation remains unchanged .
the rule of thumb @xmath344 appeared to be a good choice for preparing the equilibrium qmc data for inference . for @xmath345
a first interesting observation is that at sufficiently low temperatures @xmath302 seems to be considerably smaller than @xmath346 .
in fact , the simple data selection strategy yielding @xmath302 does not appear to produce a sufficiently informative data set to obtain quantitative agreement with for example rt - mc calculations @xcite .
we observed this problem for @xmath347 and the interaction strengths @xmath348 and @xmath349 and several values of the bias voltage @xmath161 . on the other hand , by picking an @xmath302 for each @xmath138 separately , we found larger sets of admissible input data , which tend to show a good agreement with rt - mc data for the current - voltage characteristics . while the procedure is yet somewhat arbitrary , the following criteria were used to restrict the choices of data sets producing convergent mem solutions : * ensure an error rescaling @xmath350 ; * discard strongly oscillating solutions and solutions with obvious artifacts around @xmath351 ; * discard solutions which strongly violate the physical sum rule @xmath352 . in many cases ,
too small values @xmath353 were obtained .
note that the mem as we implemented it only has prior information about the value of the truncated double integral @xmath354 , because two - dimensional probability densities are considered when the entropy expression is straightforwardly generalized with respect to @xmath355 ; * use as many data points as possible , starting with small @xmath118 , to maximize the amount of accessible information .
note that the domain parameter @xmath293 was , again , chosen somewhat arbitrarily : for @xmath348 we only investigated @xmath356 , for @xmath349 we picked @xmath357 , with @xmath358 .
the dependence of the results on the particular choice of @xmath293 was not studied systematically yet , but work along these lines is under way and the results will be presented elsewhere . the usual annealing procedure with temperatures @xmath338 , @xmath327 , @xmath347 , where for @xmath338 the lorentzian default models with @xmath359 ( @xmath348 ) and @xmath360 ( @xmath349 ) were found to be most suitable based on the posterior @xmath361 .
the current @xmath362 was computed using meir and wingreen s equation @xcite @xmath363 a(\omega ) , \label{eq : meirwingreen}\ ] ] with @xmath364 .
our experience up to now indicates that for too small sets of qmc data the method systematically underestimates the current , because bryan s algorithm by convention does not incorporate any changes to @xmath365 in case the data do not provide sufficient evidence for such modifications . as a result ,
the current is too small , because in the vicinity of @xmath351 the less structured default model obtained from the next higher temperature ( initially the broad lorentzian ) is much flatter than the true solution , which features a sharp abrikosov - suhl resonance in the relevant frequency range .
hence , the spectral function obtained from the mem has less spectral weight in the integration window in eq . than the true @xmath228 . .
, scaledwidth=49.0% ] due to this trend of underestimation , in fig .
[ fig : vgl_rtqmc ] we compare the largest values of the current compatible with the above - listed restrictions to data obtained using a recently developed rt - mc approach @xcite .
a generally good agreement is obtained . however , the data selection procedure is still too arbitrary to consider these results unbiased .
error bars are not available . if we only considered a fixed set of data @xmath366 , the covariance @xmath367 would be estimated easily @xcite . however , due to large off - diagonal terms , attempting to estimate an error bar for @xmath362 is rather cumbersome .
the @xmath368 run did not converge to a solution meeting our criteria for @xmath348 .
spectra resulting from the procedure described above are shown in fig .
[ fig : spectranoneq ] .
these are the spectral functions used to compute the current in fig .
[ fig : vgl_rtqmc ] .
while oscillations appear , presumably due to the neglected error of the covariance estimate @xcite , it is evident that the overall spectral weight at small @xmath369 is larger for @xmath348 than for @xmath349 when @xmath370 .
this is consistent with the expectation that the quasi - particle resonance for @xmath349 is already suppressed , because @xmath371 , whereas @xmath372 for @xmath348 . in fig .
[ fig : spectrapth ] , we show a comparison of the spectral functions for @xmath373 and @xmath347 to the result obtained from fourth - order perturbation theory @xcite . based on the results presented in ref .
, we expect that fourth - order perturbation theory is quite accurate at this interaction strength and temperature .
besides the unphysical oscillations in the mem result and a bias towards the high - temperature default model , especially for larger voltage biases , the agreement between the spectral functions , in particular the qualitative distribution of the spectral weight , seems satisfactory .
used for the computation of the current shown in fig .
[ fig : vgl_rtqmc ] . , scaledwidth=49.0% ] table [ tab : dataselu4 ] shows the norm @xmath374 for the functions presented in the figure .
.norms of the spectral functions shown in fig .
[ fig : spectranoneq ] . [ cols="^,^,^",options="header " , ] we did not consider larger values of @xmath142 , although at least @xmath375 yields further relevant information about @xmath253 . for a test case the spectra did not show dramatic qualitative changes as additional values at larger @xmath138 were included , as long as the error scaling merit remained @xmath350 .
however , the level of arbitrariness in the data selection would have been even larger , because of the corresponding additional @xmath302 parameters .
obtaining reliable spectral functions at finite bias will obviously require more effort and we will briefly comment on possible avenues for this effort in the conclusion .
we finish with some remarks about the structure of the kernel function and its role in the continuation problem . in the language of bayesian inference
the kernel function defines the information channel through which evidence about the shape of the representation function @xmath313 and thus also the physical spectral function @xmath228 is extracted from the monte carlo data . for the information provided by a single data point ,
this channel results in vague ( strong ) evidence for changes in a given compact region @xmath376 , see eq .
, depending on whether the subset of column vectors @xmath377 of @xmath184 spanning @xmath378 is associated with small ( large ) singular values @xmath281 , and a small ( large ) overlap of the column vectors @xmath379 of @xmath110 with the data point .
for this reason , very small singular values yield irrelevant components of the channel and are therefore projected out in bryan s algorithm by introducing the threshold @xmath306 , eq . .
we can neither perform the svd analytically , nor can we analytically take into account structural changes which occur when rotating the basis of @xmath380 to the eigenbasis of the covariance matrix @xmath269 in order to consider statistically independent data .
we can however consider values of the kernel in @xmath381 for a given data point , assuming it to be uncorrelated with other data points so that it may be investigated separately . within our qmc implementation
, experience shows that correlations between matsubara frequencies @xmath382 , @xmath383 are monotonically decreasing as a function of distance @xmath384 , though very slowly .
let us first consider a single uncorrelated imaginary part of a green s function at matsubara frequency @xmath118 in the standard wick rotation problem .
the spectral function @xmath385 is inferred through the lorentzian - shaped kernel , @xmath386(\omega_n ) = - \frac{\omega_n}{{\varepsilon}^2 + \omega_n^2}.\ ] ] for all @xmath118 the kernel is centered around @xmath387 and higher frequencies are associated to larger values of the kernel as the width given by @xmath118 is increased . as compared to @xmath388 the values of the kernel at large frequencies are still small
. we can therefore expect large singular values and thus relevant components of the kernel to be associated with small frequencies only .
this is in agreement with the well - known observation that high - frequency information about the spectral function is better put into the default model as prior knowledge and a good resolution is obtained for the fortunately most interesting low - frequency region . in the case of our two - dimensional continuation
the situation is quite similar . for given data
@xmath303 the poisson kernel in eq .
is @xmath389 it is the product of two lorentzians . in analogy to the argument given above one may expect the best resolution for data @xmath390 with @xmath391 this does not depend on the physical voltage @xmath161 , except that the critical branch cut index @xmath392 appears to be decreasing as a function of @xmath161 .
this can be estimated from the expansion order histogram for the example shown in fig .
[ fig : exporderhist ] . , which suppresses the odd perturbation orders @xcite .
the results are for @xmath393 , @xmath394 , and @xmath395 for indicated voltages @xmath161 .
the average order decreases as @xmath161 is increased .
a similar behavior is obtained for different values of @xmath138 and @xmath184 .
, scaledwidth=49.0% ] consequently , the domain parameter @xmath396 could presumably be raised as @xmath161 be increased .
however , in the limit of very large voltages , especially the low - frequency region of the physical spectrum @xmath397 is not expected to be in the best resolvable region .
thus , the approach based on a representation of data in @xmath223 appears to be limited to relatively small voltages . note that , since @xmath398 is the most interesting parameter regime , this is presumably no serious drawback
however , identifying subtle details in the range @xmath399 may require more care than the case @xmath351 for the standard wick rotation .
fixing the @xmath312 and @xmath259 variables in the kernel and analyzing the dependence as a function of the data coordinates @xmath138 , @xmath118 we similarly find that large values of the kernel are found in the vicinity of the domain boundary , i.e. for @xmath400 pairs close to the cone boundary , @xmath401 , with @xmath138 , @xmath118 not being too large .
hence , data close to the boundary provide the most relevant information .
this appears to explain the importance of an @xmath142-dependent @xmath302 in our computation of non - equilibrium spectra .
the imaginary time formulation for steady state transport in strongly correlated quantum impurity systems proposed by han and heary is based on the solution of a family of quantum impurity models subject to complex voltages , and a subsequent double analytical continuation with respect to frequency and voltage .
a main purpose of the study presented in this paper was to investigate to what extent an unbiased , numerical implementation of this approach is feasible and whether or not it yields physically plausible results . to solve the impurity problem we employed two recently developed continuous - time impurity solvers .
the hybridization expansion approach was found to be unsuitable in the case of large complex voltages , due to a serious sign problem resulting from the shift of the hybridization function to negative values .
the weak - coupling approach , on the other hand , works well for small and large @xmath138 .
even though the non - interacting green s function @xmath137 becomes complex and oscillating , the resulting sign problem is mild , enabling us to obtain highly accurate , unbiased imaginary - frequency data for all relevant complex voltages .
this part of the problem can be considered as solved , leaving us with the double analytical continuation problem . a main result of this work is the derivation of an analytical expression of the kernel ( eq . ( [ eq : poissonkernelepsdomain ] ) ) for the analytical continuation procedure .
this kernel is consistent with the analytical structure ( branch cuts ) of the theory and maps a function of two variables , @xmath402 , to the interacting green s function in a tubular cone domain of the complex voltage and frequency space .
the physical spectral function for a dot under voltage bias @xmath161 is obtained as @xmath403 .
we have implemented and tested an analytical continuation procedure based on the maximum entropy method and our proposed kernel .
we want to emphasize that both the data selection procedure and the estimate of the covariance entering into the maximum entropy employed for @xmath345 are at this point still rather rudimentary and leave room for improvement .
our results for the non - equilibrium case should therefore be viewed as preliminary and illustrate the presently most plausible spectral functions and currents which can be obtained using our current implementation .
nevertheless , taking into account the obvious challenges inherent in a double analytical continuation procedure , we find physically reasonable spectral functions for the interacting equilibrium model and , to a lesser extent , also under finite bias .
a comparison of the spectral functions with fourth - order perturbation theory shows that the approach is able to reproduce the correct trends , albeit the strong oscillations resulting from the maximum entropy approach render a detailed comparison meaningless . on the other hand ,
the current calculated using these spectral functions is in fair agreement with recent results from a real - time monte - carlo approach .
we hope that further improvements in data selection strategies , a better understanding of the precise behavior of the green s function across the branch cuts , improved default model functions and , very importantly , the inclusion of the sum rules into the maximum entropy algorithm will eventually enable us to obtain more accurate results and turn the combination of monte - carlo and double analytical continuation into a reliable tool for the study of steady - state properties of quantum impurity systems using han and heary s formalism .
we acknowledge useful conversations with jong han , sebastian fuchs , emanuel gull , and kurt schnhammer .
further acknowledges the hospitality of the center for computation and technology ( cct ) at louisiana state university and the financial support by the german academic exchange service ( daad ) through the ppp exchange program .
p.w . acknowledges support from snf grant pp002 - 118866 .
m.j . acknowledges nsf grant dmr-0706379 .
based on the argument given in section [ sec : shift_argument ] we restrict ourselves to the class of functions with positive imaginary part in the domain @xmath218 , typically denoted as @xmath404 in the mathematical literature . for a good overview of the concepts and terminologies used in the mathematical context see ref .
@xcite and the first volume of ref .
vladimirov found the following generalization of herglotz - nevanlinna representations to several complex variables @xcite .
it is essentially @xcite the 1 .
the poisson integral @xmath408 $ ] is pluriharmonic in @xmath409 ; 2 . the function @xmath410 , @xmath411 ,
is represented by the poisson formula @xmath412(\underline z ) + ( \underline a , \underline y),\ ] ] for some @xmath413 , where @xmath414 is the dual cone of @xmath269 ; 3 . for all @xmath415 , under the assumption that @xmath269 is regular , the schwarz representation @xmath416 holds , with @xmath417.@xmath418 let us introduce the relevant mathematical terminology .
a cone @xmath419 with vertex at zero is defined @xcite by the property that @xmath420 .
its dual cone @xmath421 . here ,
@xmath408(\underline z ) = \int_{\mathbb{r}^m}\mathrm{d}^mx\ , \mu(x ) \mathcal{p}_c(\underline z - \underline x)$ ] with the poisson kernel @xmath422 and the cauchy kernel @xmath423 we will not explicitly use the schwarz kernel @xmath424 , the reader may find it in ref .
a holomorphic mapping is said to be biholomorphic iff it is one - to - one .
two domains @xmath425 are biholomorphically equivalent iff a biholomorphic mapping @xmath426 exists . for the concept of pluriharmonicity see introductory volumes of ref . @xcite . in the case of
@xmath223 we rewrite eq . as @xmath427 hence , the dual cone @xmath428 \}. \end{split}\ ] ] evaluating the integrals @xmath429 in yields @xmath430 eq
. follows immediately from the definition . in order to prove the validity of the representation based on vladimirov s theorem , we first determine @xmath431 due to the boundedness of the green s function .
now we need to show that the poisson integral @xmath432 $ ] with respect to the measure @xmath433 is pluriharmonic for all functions @xmath434 .
note that for the @xmath142-dimensional octant @xmath435 is pluriharmonic for all functions @xmath436 .
fortunately , as we restrict ourselves to @xmath437 in our application , all tubular cone domains are known to be biholomorphically equivalent they are simply connected through linear transformations . to see the advantage more explicitly
, we introduce the biholomorphism @xmath438 given by the linear operation @xmath439 obviously , @xmath440 we explicitly show that the kernel representation for a function @xmath441 may also be derived by applying the corresponding poisson kernel @xmath442 for the tubular octant to the corresponding function @xmath443 and transforming back to @xmath444 .
since the representation for @xmath445 is valid , we will have shown explicitly that is valid for all @xmath446 .
for this purpose it suffices to show that @xmath447 because then @xmath448 and therefore @xmath449 .
we introduced the integration variables @xmath234 and @xmath450 of the poisson integrals @xmath432 $ ] , @xmath451 and @xmath452 $ ] , @xmath453 , respectively . since @xmath454 , transforming @xmath455 in @xmath456 then yields . with a similar procedure as for @xmath457 it is straightforward to show that @xmath458 to finish the argument we verify that eq .
holds by inserting @xmath459 representations for any tubular cone domains in @xmath460 are similarly related due to the biholomorphic equivalence . in particular , valid representations for @xmath181
are obtained easily .
for example , the poisson kernel with respect to @xmath461 reads @xmath462 and could in principle be used for an enhanced continuation procedure invoking data from all sectors of the complex space .
99 s. hershfield , j. h. davies , and j. w. wilkins , phys .
lett . * 67 * , 3720 ( 1991 ) .
s. hershfield , j. h. davies , and j. w. wilkins , phys .
b * 46 * , 7046 ( 1991 ) . t. fujii and k. ueda , phys . rev .
b * 68 * , 155310 ( 2003 ) .
p. werner , t. oka , m. eckstein , and a. j. millis , phys .
b * 81 * , 035108 ( 2010 ) . f. heidrich - meisner , a. e. feiguin , e. dagotto , phys . rev .
b * 79 * , 235336 ( 2009 ) p. schmitteckert , phys . rev .
b * 70 * , 121302 ( 2004 ) l. mhlbacher and e. rabani , phys . rev .
lett . * 100 * , 176403 ( 2008 ) .
s. weiss , j. eckel , m. thorwart , and r. egger , phys .
b * 77 * , 195316 ( 2008 ) .
p. werner , t. oka , and a. j. millis , phys .
b * 79 * , 035320 ( 2009 ) .
m. schiro and m. fabrizio , phys .
b * 79 * , 153302 ( 2009 )
. t. l. schmidt , p. werner , l. mhlbacher , and a. komnik , phys .
b * 78 * , 235110 ( 2008 ) . f. b. anders , phys .
* 101 * , 066804 ( 2008 ) .
a. rosch , j. paaske , j. kroha , and p. wlfle , j. phys .
. jpn . * 74 * , 118 ( 2005 ) .
s. g. jakobs , v. meden , h. schoeller , phys .
* 99 * , 150603 ( 2007 ) .
r. gezzi , th .
pruschke , and v. meden , phys .
b * 75 * , 045324 ( 2007 ) .
pruschke , r. gezzi , a. dirks , nato science series b : electron transport in nanosystems , 249 ( 2009 ) .
h. schoeller , and f. reininghaus , phys . rev .
b * 80 * , 045117 ( 2009 ) .
pruschke , a. dirks , r. gezzi , physica b * 404 * , 3141 ( 2009 ) .
j. e. han and r. j. heary , phys .
* 99 * , 236808 ( 2007 ) .
s. hershfield , phys .
lett . * 70 * , 2134 ( 1993 ) .
a. n. rubtsov , v. v. savkin and a. i. lichtenstein , phys .
b * 72 * , 035122 ( 2005 ) . p. werner , a. comanac , l. de medici , m. troyer and a. j. millis , phys .
lett . * 97 * , 076405 ( 2006 ) .
j. e. han , phys .
b * 75 * , 125122 ( 2007 ) .
e. gull , p. werner , o. parcollet and m. troyer , europhys .
lett . * 82 * 57003 ( 2008 ) .
k. mikelsons , a. macridin , and m. jarrell , phys .
e * 79 * , 057701 ( 2009 ) .
p. werner and a. j. millis , phys .
b * 74 * , 155107 ( 2006 ) .
j. e. hirsch and r. m. fye , phys .
lett . * 56 * , 2521 ( 1986 ) j. e. han , arxiv:1001.4989 a .- p .
jauho , n. s. wingreen , and y. meir , phys .
b * 50 * , 5528 ( 1994 ) v. s. vladimirov , methods of the theory of functions of several complex variables " , m.i.t . press ( 1966 )
g. m. khenkin , a. g. vitushkin ( eds . ) , encyclopedia of mathematical sciences : several complex variables ii , 179 ff . ( 1994 ) .
v. s. vladimirov , sov .
* 19 * , 254 ( 1978 ) .
r. nevanlinna , eindeutige analytische funktionen " , berlin ( 1936 ) .
e. t. jaynes , phys . rev . *
106 * , 620630 ( 1957 ) m. jarrell , j. e. gubernatis , physics reports * 269 * , 133 ( 1996 ) . n. wu , the maximum entropy method " , 162 ( 1997 ) .
r. k. bryan , eur .
biophys . j. * 18 * , 165 ( 1990 ) .
h. jeffreys , proceedings of the royal society of london , series a * 186 * , no .
1007 , 453 ( 1946 ) w. v. d. linden , r. preuss , and w. hanke , j. phys . : condens .
matter * 8 * , 3881 ( 1996 ) m. pinchas , and b. z. bobrovskya , signal processing * 86 * , issue 10 , 2913 ( 2006 ) v. s. vladimirov , mat .
( n.s . ) * 79(1 ) * , 128152 ( 1969 ) ; mathematics of the ussr - sbornik * 8(1 ) * , 125 ( 1969 ) .
m. jarrell , a. macridin , k. mikelsons , and d.g.s.p .
doluweera , in _ lectures on the physics of strongly correlated systems xii _ , aip conference proc .
* 1014 * , a. avella and f. mancini
( eds . ) , 34 ( 2008 ) | recently han and heary proposed an approach to steady - state quantum transport through mesoscopic structures , which maps the non - equilibrium problem onto a family of auxiliary quantum impurity systems subject to imaginary voltages .
we employ continuous - time quantum monte - carlo solvers to calculate accurate imaginary time data for the auxiliary models .
the spectral function is obtained from a maximum entropy analytical continuation in both matsubara frequency and complexified voltage . to enable the analytical continuation
we construct a kernel which is compatible with the analytical structure of the theory .
while it remains a formidable task to extract reliable spectral functions from this unbiased procedure , particularly for large voltages , our results indicate that the method in principle yields results in agreement with those obtained by other methods . |
data stream mining has been the subject of extensive research over the last decade or so
. one of the major issues with data stream mining is dealing with concept drift that causes models built by classifiers to degrade in accuracy over a period of time . while data steam environments require that models are updated to reflect current concepts , the capture and storage of recurrent concepts allows a classifier to use an older version of the model that provides a better fit with newly arriving data in place of the current model .
this approach removes the need to explicitly re - learn the model , thus improving both accuracy and computational cost .
a number of methods have been proposed that deal with the capture and exploitation of recurring concepts @xcite , @xcite , @xcite , @xcite and @xcite .
although achieving higher accuracy as expected during phases of concept recurrence in the stream , a major issue with existing approaches is the setting of user defined parameters to determine whether a current concept matches with one from the past .
such parameters are difficult to set , particularly due to the drifting nature of real world data streams .
our approach avoids this problem by applying the discrete fourier transform ( dft ) as a meta learner .
the dft , when applied on a concept ( decision tree model ) results in a spectral representation that captures the classification power of the original models .
one very attractive property of the fourier representation of decision tree is that most of the energy and classification power is contained within the low order coefficients @xcite .
the implication of this is that that when a concept c recurs as concept c * with relatively small differences caused by noise or concept drift , then such differences are likely to manifest in the high order coefficients of spectra s and s * ( derived from c and c * respectively ) , thus increasing the likelihood of c * being recognized as a recurrence of c. the dft , apart from its use in meta learning , has a number of other desirable properties that make it attractive for mining high speed data streams .
this includes the ability to classify directly from the spectra generated , thus eliminating the need for expensive traversal of a tree structure .
our experimental results in section 5 clearly show the accuracy , processing speed and memory advantages of applying the dft as opposed to the meta learning approach proposed by gama and kosina in @xcite .
the rest of the paper is as follows . in section 2
we review work done in the area of capturing recurrences .
we describe the basics of applying the dft to decision trees in section 3 . in section 4
we discuss a novel approach to optimizing the computation of the fourier spectrum from a decision tree .
our experimental results are presented in section 5 and we conclude the paper in section 6 where we draw conclusions on the research and discuss some directions for future research .
while a vast literature on concept drift detection exists @xcite only a small body of work exists so far on exploitation of recurrent concepts .
the methods that exist fall into two broad categories .
firstly , methods that store past concepts as models and then use a meta learning mechanism to find the best match when a concept drift is triggered @xcite , @xcite . secondly , methods that store past concepts as an ensemble of classifiers .
lazarescu in @xcite proposes an evidence forgetting mechanism for data instances based on a multiple window approach and a prediction module to adapt classifiers based on an estimation of the future rate of change .
whenever the difference between the observed and estimated rates of change are above a user defined threshold a classifier that best represents the current concept is stored in a repository .
experimentation on the stagger dataset showed that the proposed approach outperformed the flora method on classification accuracy with re - emergence of previous concepts in the stream .
ramamurthy and bhatnagar @xcite use an ensemble approach based on a set of classifiers in a global set g. an ensemble of classifiers is built dynamically from a collection of classifiers in g if none of the existing individual classifiers are able to meet a minimum accuracy threshold based on a user defined acceptance factor .
whenever the ensemble accuracy falls below the accuracy threshold , then the global set g is updated with a new classifier trained on the current chunk of data .
another ensemble based approach by katakis et al .
is proposed in @xcite .
a mapping function is applied on data stream instances to form conceptual vectors which are then grouped together into a set of clusters .
a classifier is incrementally built on each cluster and an ensemble is formed based on the set of classifiers .
experimentation on the usenet dataset showed that the ensemble approach produced better accuracy than a simple incremental version of the naive bayes classifier .
gomes et al @xcite used a two layer approach with the first layer consisting of a set of classifiers trained on the current concept while the second contains classifiers created from past concepts .
a concept drift detector is used to flag changes in concept and when a warning state is triggered incoming data instances are buffered in a window to prepare a new classifier .
if the number of instances in the warning window is below a user defined threshold then the classifier in layer 1 is used instead of re - using classifiers in layer 2 .
one major issue with this method is validity of the assumption that explicit contextual information is available in the data stream .
gama and kosina also proposed a two layered system in @xcite designed for delayed labeling , similar in some respects to the gomes et al .
@xcite approach . in their approach
gama and kosina pair a base classifier in the first layer with a referee in second layer .
referees learn regions of feature space which its corresponding base classifier predicts accurately and is thus able to express a level of confidence on its base classifier with respect to a newly generated concept .
the base classifier which receives the highest confidence score is selected , provided that it is above a user defined hit ratio parameter ; if not , a new classifier is learned .
the discrete fourier transform ( dft ) has a vast area of application in very diverse domains such as time series analysis , signal processing , image processing and so on .
it turns out as park @xcite and kargupta @xcite show that the dft is very effective in terms of classification when applied on a decision tree model .
kargupta and park in @xcite explored the use of the dft in a distributed environment but did not explore its usage in a data stream environment as this research sets out to do .
r0.2 kargupta and park in @xcite showed that the fourier spectrum consisting of a set of fourier coefficients fully captures a decision tree in algebraic form , meaning that the fourier representation preserves the same classification power as the original decision tree .
a decision tree can be represented in compact algebraic form by applying the dft to the paths of the tree .
we illustrate the process by considering a binary tree for simplicity but in practice the dft can be applied to non binary trees as well @xcite . for trees with a total of d binary valued features the @xmath0 fourier coefficient @xmath1 is given by : @xmath2 where @xmath3 is the classification outcome of path vector x and @xmath4 , the fourier basis function is given by : @xmath5 figure 1 shows a simple example with 3 binary valued features @xmath6 , @xmath7 and @xmath8 , out of which only @xmath6 and @xmath8 are actually used in the classification . as shown in @xcite only coefficients for paths that are defined by attributes that actually appear in the tree need to be computed as all other coefficients are guaranteed to be zero in value .
thus any coefficient of the form @xmath9 will be zero since attribute @xmath7 does not appear in the tree . with the wild card operator * in place we can use equations ( 1 ) and ( 2 ) to calculate non zero coefficients .
thus for example we can compute : @xmath10 and so on .
in addition to the properties discussed above , the fourier spectrum of a given decision tree has two very useful properties that make it attractive as a tree compression technique @xcite : 1 .
all coefficients corresponding to partitions not defined in the tree are zero
the magnitudes of the fourier coefficients decrease exponentially with their order , where the order is taken as the number of defining attributes in the partition . taken together these properties mean that the spectrum of a decision tree can be approximated by computing only a small number of low order coefficients , thus reducing storage overhead . with a suitable thresholding scheme in place ,
the fourier spectrum consisting of the set of low order coefficients is thus an ideal mechanism for capturing past concepts .
furthermore , classification of unlabeled data instances can be done directly in the fourier domain as it is well known that the inverse of the dft defined in expression ( 3 ) can be used to recover the classification vector , instead of the use of a tree traversal which can be expensive in the case of deep trees .
expression 3 uses the complex conjugate @xmath11 function for the inverse operation in place of the original basis function of @xmath4 .
@xmath12 due to thresholding and loss of some high order coefficient values the classification value @xmath3 for a given data instance x may need to be rounded to the nearest integer in order to assign the class value . for example , with binary classes a value for f is rounded up to 1 if it is in the range @xmath13 and rounded down to 0 in the range @xmath14 .
we first present the basic algorithm used in section 5.1 and then go on to discuss an optimization that we used for energy thresholding in section 5.2 .
we use cbdt @xcite as the base algorithm which maintains a forest of trees .
this forest of trees is dynamic in the sense that it can adapt to changing concepts at drift detection points .
we thus define the memory consumed by this forest as _
active_. we integrate the basic cbdt algorithm with the adwin @xcite drift detector to signal concept drift . at the first concept drift point the best performing tree ( in terms of accuracy ) is identified and the dft is applied after energy thresholding after which the resulting spectrum is stored in the repository for future use if the current concept recurs .
the spectra stored in the repository are fixed in nature as the intention is to capture past concepts . at each subsequent drift point a winner model is chosen by polling both the active memory and the repository . if the winner emerges from the active memory , two checks are made before the dft is applied .
first of all , we check whether the difference in accuracy between the winner tree in active memory ( t ) and the best performing model in the repository is greater than a tie threshold @xmath15 .
if this check is passed then the dft is applied to t and a further check is made to ensure that its fourier representation is not already in the repository .
if the winner model at a drift point emerges from an already existing spectrum in the repository then no fourier conversion is applied on any of the trees in active memory .
whichever model is chosen as the winner it is applied to classify all unlabeled data instances until a new winner emerges at a subsequent drift point .
the least performing model m having the lowest weighted accuracy function is deleted if the repository has no room for new models.the weighted accuracy of m is defined by : @xmath16 , where @xmath17 is the number of times that m was declared a winner since it was inserted into the repository . [
fct [ cbdt ] read an instance i from the data stream + + if best model m is in repository call classify to classify i + append 0 to adwin s window for m if classification is correct , else append 1 + + identify best performing model f in repository + + apply dft on model m to produce f * using energy threshold _ @xmath18 _ + + insert f * into repository + identify best performing model m by polling active memory and repository + goto 1 [ h ! ] classify [ classifyusingfourierspectrum ] + + if @xmath3 is greater than 0.5 , return class1 , otherwise class2 in order to avoid unnecessary computation of higher order coefficients which yield increasingly low returns on classification accuracy , energy threshold is highly desirable . to threshold on energy a subset s of the ( lower order ) coefficients needs to be determined such that @xmath19 , where @xmath20 denotes the total energy across the spectrum and @xmath21 is the desired energy threshold value . in our optimized thresholding ,
we first compute the cumulative energy @xmath22 at order i given by : @xmath23 .
given an order i , an upper bound estimate for the cumulative energy across the rest of the spectrum is given by : @xmath24 , as the exponential decay property ensures that the energy at each of the orders @xmath25 , @xmath26 , @xmath27 , @xmath28 is less than energy @xmath29 at order i , where d is number of attributes in the dataset .
thus a lower bound estimate for the fraction of the cumulative energy @xmath30 at order i to the total energy across all orders can then be expressed as : @xmath31 where @xmath29 is actual ( computed ) energy at order i. the lower bound estimate allows the specification of a threshold @xmath21 based on the energy captured by a given order i which is more meaningful to set rather than an arbitrary threshold .
the scheme expressed by equation ( [ equ : estimatedcumulativeenergy ] ) enables the thresholding process to be done algorithmically . if the cumulative energy @xmath32 , then we can guarantee that the actual energy captured is at least @xmath21 , since @xmath30 is a lower bound estimate . on the other hand
if @xmath33 , then @xmath34 can be expressed as : @xmath35 thus equation ( [ euqation2 ] ) enables the cumulative fraction to be easily updated incrementally for the next higher order ( i+1 ) by simply computing the energy at that order while exploiting the exponential decay property of fourier spectrum .
the thresholding method guarantees that no early termination will take place .
this is because @xmath30 is a lower bound estimate and hence the order that it returns will never be less than the true order that captures a given fraction @xmath21 of the total actual energy in the spectrum .
this section elaborates on our empirical study involving the following learning systems : cbdt , fct ( fourier concept trees ) and metact .
the fct incorporates the fourier compressed trees in a repository in addition to the forest of trees that standard cbdt maintains .
we implement gama s meta learning approach with cbdt as the base learner , namely metact .
the main focus of the study is to assess the extent to which recurrences are recognized using old models preserved in classifier pools .
all experimentation was done with the following parameter values : * hoeffding tree parameters the desired probability of choosing the correct split attribute=0.99 , tie threshold=0.01 , growth check interval=32 * tree forest parameters maximum node count=5000 , maximum number of fourier trees=50 , accuracy tie threshold @xmath15=0.01 * adwin parameters drift significance value=0.01 , warning significance value=0.3 ( metact only ) all experiments were done on the same software with c # .net runtime and hardware with intel i5 cpu and 8 gb ram , clearning the memory in each run to have a fair comparison .
we experimented with data generated from 3 data generators commonly used in drift detection and recurrent concept mining , namely sea concept @xcite , rbf and rotating hyperplane generators .
in addition we used 2 real - world datasets , _ spam _ and the _ nsw electricity _ which have also been commonly used in previous research .
for the synthetic datasets , each of the 4 concepts spanned 5,000 instances and reappeared 25 times in a data set , yielding a total of 500,000 instances with 100 true concept drift points . in order to challenge the concept recognition process , we added a 10% noise level for all synthetic data sets to ensure that concepts recur in similar , but not exact form .
we used moa @xcite as the tool to generate these datasets . 1 .
* sea : * the concepts are defined by the function @xmath36 .
we ordered the concepts as concept1 , concept2 , concept3 and concept4 generated using threshold values 8,7,9 and 9.5 respectively on the first data segment of size 20,000 .
we generated 96 recurrences of a modified form of these concepts by using different seed values in moa for each sequence of recurrence .
thus , our version of this dataset differed from the one used by gama and kosina @xcite .
who simply used 3 concepts with the third being an exact copy of the first . 2 .
* rbf : * the number of centroids parameter was adjusted to generate different concepts for the rbf dataset .
concept1 , concept2 , concept3 and concept4 were produced with the number of centroids set to 5,15 , 25 and 35 respectively .
similar to the sea dataset , the seed parameter helped in producing similar concepts for a given centroid count value .
this dataset had 10 attributes .
3 . * rotating hyperplane : * the number of drifting attributes was adjusted to 2,4,6 , and 8 in a 10 dimensional data set to create the four concepts .
the concept ordering , generation of similar concepts and concatenation were exactly the same as in the other data sets mentioned above . 1 .
_ spam data set : _ the spam dataset was used in it original form which encapsulates an evolution of spam messages .
there are 9,324 instances and 499 informative attributes , which was different from the one version used by gama that had 850 attributes .
electricity data set : _ nsw electricity dataset is also used in its original form .
there are two classes _ up _ and _ down _ that indicate the change of price with respect to the moving average of the prices in last 24 hours . in our preliminary experiments
, we found optimal values for the two parameters , _ delay _ in receiving labels for the instances in short term memory , and _ hit percentage threshold value _ as 200 and 80% , respectively .
the latter parameter reflects the estimated similarity of the current concept with one from the past and thus controls the degree of usage of classifiers from the pool .
our focus in this series of experiments was to assess the models in terms of accuracy , memory consumption and processing times .
none of the previous studies reported in the recurrent concept mining literature undertook a comparative study against other approaches and so we believe our empirical study to be the first such effort .
furthermore , all of the previous studies focused exclusively on accuracy without tracking memory and execution time overheads and so this study would also be the first of its kind .
a delay period of 200 was used with all three approaches in order to perform a fair comparison .
figure 2 clearly shows that overall , fct significantly outperforms its two rivals with respect to classification accuracy .
the major reason for fct s superior performance was its ability to re - use previous classifiers as shown in the segment 20k-25k on the rbf dataset where the concept is similar to concept1 that occurred in interval 1 - 5k .
this is in contrast to metact which was unable to recognize the recurrence of concept1 .
a similar situation occurs in the interval 25k-35k where the concept is similar to the previously occurring concepts , which are concept2 and 3 .
as expected cbdt , operating on its own without support for concept recurrence had a relatively flat trajectory throughout the stream segment .
a similar trend to the rbf dataset was observed in rotating hyperplane and sea datasets as well .
it can be clearly seen that fct was successful in reusing the models learned before on data segments from 20k to 25k and from 30k to 35k . though a preserved model was reused on the data segment from 25k to 30k ( corresponding to concept3 ) , the accuracy was not as high as in the above two segments . on the segment from 35k to 40k ,
concept recurrence was not picked up by either fct or metact resulting in a new classifier being used .
we omit the figure for the sea dataset due to space constraints . in summary , fct outperformed metact over 90 recurring concepts whereas metact did better in 6 occurrences , thus maintaining the same trend as with the other 2 synthetic datasets that we experimented with .
the next experiment was on the nsw electricity data set .
figure [ accuracycurves ] shows that overall , fct was the winner here as well , outperforming metact at 25 segments out of 35 that we tracked . sudden fall in accuracy of metact is occational but due to incorrect selection of winner which was a decision stump .
our experimentation on accuracy has revealed , especially in the case of fct , the key role that concept capture and re - use has played in improving accuracy .
the question is , what price has to be paid in terms of memory overhead in storing these recurrent concepts ?
table1 clearly shows that the fourier transformed trees consume a small fraction of the memory used by the pool of trees kept in fct s active memory , despite the fact that collectively these models outperform their decision tree counterparts at a greater number of points in the stream progression .
r0.7 comparison of overall memory consumption across fct and metact is complicated by the fact that the latter tended to have immature trees in its classifier pool that under fits concepts . despite this
, table 1 reveals that fct s memory consumption is competitive to that of metact . the only case where metact had a substantially lower consumption was with the spam dataset with a lower overhead for active memory .
fct and metact have two very contrasting methods of classification .
the former routes each unlabeled instance to a single tree , which is the best performing tree selected at the last concept drift point .
in contrast metact classifies by routing an unlabeled instance to all referees to obtain their assessment of their corresponding models and in general will have more processing overhead on a per instance basis .
however , fct has potentially more overhead at concept drift points if the winner tree is one that is selected from the active forest as this tree needs to be converted into its fourier representation .
thus it is interesting to contrast the run time performances of the two approaches .
[ tab : overheadcomparison ] .average memory consumption ( in kbs ) and processing speed instances per second ) comparison [ cols= " < , < , < , < , < ,
< , < " , ] table 1 shows that in general fct has a higher processing speed ( measured in instances processed per second ) ; the only exception was with the electricity dataset where metact was faster .
the electricity data contains a relatively larger number of drift points in comparison to the other datasets and this in turn required a greater number of dft operations to be performed , thus slowing down the processing . in our future research
we will investigate methods of optimizing the dft process . finally , we close this section with two general observations on fct which hold across all experiments reported above .
firstly , we note that the discrete fourier transform ( dft ) , as expected , was able to capture the _ essence _ of a concept to the extent that when it reappeared in a modified form in the presence of noise it was still recognizable and was able to classify it accurately . secondly , not only was the dft robust to noise , it actually performed better than the original decision trees at concept recurrence points due to its better generalization capability .
having established the superiority of fct we were interested in exploring the sensitivity of fct s accuracy on two key factors .
r0.5 fct s energy threshold parameter controls the extent to which it captures recurring contexts .
we ran experiments with all datasets we experimented with and tracked accuracy across four different thresholds : 95% , 80% , 40% and 20% .
the trends observed for all datasets were very similar and hence we display results for the sea concepts dataset due to space constraints .
figure [ energyaccuracy ] clearly shows that very little difference in accuracy exists between the trajectories for 40% and 95% , showing the resilience of the dft in capturing the classification power of concepts at low energy levels such as 40% .
thus the low order fourier coefficients that survive the 40% threshold hold almost the same classification power of spectra at the 80% or 90% levels which contain more coefficients .
such higher energy spectra would represent larger decision trees in which some of the decision nodes would be split into leaf nodes , thus enabling them to reach a slightly higher level of accuracy .
in section 5.4 we observed that fct outperformed metact by recognizing concepts from the past even though the concepts did not recur exactly in their original form due partly to noise and partly due to different data instances being produced as a result of re - seeding of the concept generation functions . in this experiment we explicitly test the resilience of fct to noise level by subjecting it to three different levels of noise - 10% , 20% and 30% . for reasons of completeness we also included metact in he experimentation to aid in the interpretation of results .
figure 4 reveals three interesting pieces of information .
firstly , fct is still able to recognize recurring concepts at the 20% noise level even though the models it re - uses do not have quite the same classification power ( when compared to the 10% noise level ) on the current concept due to data instances being corrupted by a relatively higher level of noise .
secondly , fct s concept recurrence recognition is essentially disabled at the 30% noise level as shown by its flat trajectory , thus essentially performing at the level of the base cbdt system .
it is able to avoid drops in accuracy on account of the forest of trees that is maintained and is able to switch quickly and seamlessly from one tree to another when concepts change occurs .
thirdly , although metact is not the focus of this experiment we see that metact s ability to recognize recurring concepts is disabled at the 20% level , showing once gain the resilience of the dft to noise . at the 30% level
its accuracy drops quite sharply at certain points in the stream.this is due to the fact that it learns a single new classifier and relies on it to classify instances in the current concept .
in contrast , fct exploits the entire forest of trees and switches from one tree to another tree in its active forest in response to drift .
[ accuracynoise ]
in this research we proposed a novel mechanism for mining data streams by capturing and exploiting recurring concepts .
our experimentation showed that the discrete fourier transform when applied on decision trees captures concepts very effectively , both in terms of information content and conciseness .
the fourier transformed trees were robust to noise and were thus able to recognize concepts that reappeared in modified form , thus contributing significantly to improving accuracy .
overall our proposed approach significantly outperformed the meta learning approach by gama and kosina @xcite in terms of classification accuracy while being competitive in terms of memory and processing speed .
we were able to optimize the derivation of the fourier spectrum by an efficient thresholding process but there is further scope for optimization in the computation of low order coefficients in streams exhibiting frequent drifts , as our experimentation with the nsw electricity dataset reveals .
our future work will concentrate on two areas .
firstly we plan to investigate the use of multi - threading on a parallel processor platform to optimize the dft operation .
allocating the dft process to a thread while another thread processes the incoming stream will greatly speed up processing for fct as the two processes are independent of each other and can be executed in parallel .
secondly , the computation of the fourier basis function that requires a vector dot product computation can be optimized by using patterns in the two vectors involved .
[ dawak ] alippi , c. , boracchi , g. , & roveri , m. ( 2013 ) . just - in - time
classifiers for recurrent concepts .
ieee transactions on neural networks and learning systems , 24(4 ) , 620 - 634 .
doi:10.1109/tnnls.2013.2239309 bifet , a. , & gavald , r. ( 2007 ) .
learning from time - changing data with adaptive windowing symposium conducted at the meeting of the 2007 siam international conference on data mining ( sdm07 ) , minneapolis , minnesota .
hoeglinger , s. , pears , r. , & koh , y. ( 2009 ) .
cbdt : a concept based approach to data stream mining . in t. theeramunkong , b. kijsirikul , n. cercone , & t .- b .
ho ( eds . ) , advances in knowledge discovery and data mining ( vol . 5476 , pp .
1006 - 1012 ) : springer berlin / heidelberg.doi:10.1007/978-3-642-01307-2_107 kargupta , h. , & park , b .- h .
( 2004 ) . a fourier spectrum - based approach to represent decision trees for mining data streams in mobile environments .
ieee trans . on knowl . and data eng .
, 16(2 ) , 216 - 229 .
doi:10.1109/tkde.2004.1269599 ramamurthy , s. , & bhatnagar , r. ( 2007 ) .
tracking recurrent concept drift in streaming data using ensemble classifiers .
symposium conducted at the meeting of the sixth international conference on machine learning and applications , 2007 .
street , w. n. , & kim , y. ( 2001 . a streaming ensemble algorithm ( sea ) for large - scale classification .
presented at the meeting of the proceedings of the seventh acm sigkdd international conference on knowledge discovery and data mining , san francisco , california . | in this research we address the problem of capturing recurring concepts in a data stream environment .
recurrence capture enables the re - use of previously learned classifiers without the need for re - learning while providing for better accuracy during the concept recurrence interval .
we capture concepts by applying the discrete fourier transform ( dft ) to decision tree classifiers to obtain highly compressed versions of the trees at concept drift points in the stream and store such trees in a repository for future use .
our empirical results on real world and synthetic data exhibiting varying degrees of recurrence show that the fourier compressed trees are more robust to noise and are able to capture recurring concepts with higher precision than a meta learning approach that chooses to re - use classifiers in their originally occurring form . |
we study markov chain monte carlo ( mcmc ) methods for bayesian inference of phylogeny .
we begin by presenting the relevant background material by defining phylogenetic trees , evolutionary models ( in section [ sec : evolutionary - models ] ) , and the associated mcmc methods ( in section [ sec : spr ] ) .
we refer the interested reader to semple and steel @xcite for a more comprehensive introduction to the mathematics of phylogeny . finally , we present our results and discuss related work in section [ sec : our - results ] .
a phylogenetic tree is an unrooted tree @xmath0 on @xmath1 leaves ( called taxa , corresponding to @xmath1 species ) where internal vertices have degree three .
let @xmath2 denote the edges of @xmath0 and @xmath3 denote the vertices . in the phylogenetic reconstruction problem
, we observe a collection of labelings of the leaves of @xmath0 from a set @xmath4 , and our goal is to infer the tree @xmath0 from which they were generated from . for example , if @xmath5 then we are given ( aligned ) dna sequences for @xmath1 species , and we are trying to determine the tree describing the evolutionary history of the present - day species .
the labelings on the leaves of @xmath0 are the projection of labelings on all vertices of @xmath0 , and these labelings of @xmath6 are generated in the following manner .
there is a stochastic process along edges of @xmath0 ( e.g. , modeling the evolutionary process of dna substitutions ) which is defined by a continuous - time markov chain .
thus , for each edge @xmath7 there is a @xmath8 rate matrix @xmath9 and a time @xmath10 , which is called the branch length of @xmath11 . in this paper , as is typical in the phylogenetic setting , we assume there is a single rate matrix @xmath9 that is common to all edges . the rate matrix is assumed to be reversible with respect to some distribution @xmath12 on @xmath4 .
hence , fix @xmath12 as the stationary vector for @xmath9 ( i.e. , @xmath13 ) .
( the matrix @xmath9 is usually scaled so that we expect one `` substitution '' ( i.e. , change ) per unit of time . )
the rate matrix @xmath9 defines a continuous time markov chain , and together with @xmath14 defines a transition matrix on edge @xmath11 : @xmath15 the matrix @xmath16 is a stochastic matrix of size @xmath17 , and thus defines a discrete - time markov chain , which is time - reversible with stationary distribution @xmath12 , i.e. , @xmath18 , and @xmath19 ( for all @xmath20 ) .
the simplest four - state ( i.e. , @xmath21 ) evolutionary model has a single parameter for the off - diagonal entries of the rate matrix @xmath9 ; this model is known as the jukes - cantor model .
the most general reversible four - state model is the gtr ( general time reversible ) model . for @xmath22 (
often studied for mathematical interest ) , the model is binary and the rate matrix has a single parameter ; this model is known as the cfn ( cavender - farris - neyman ) model .
see felsenstein @xcite or yang @xcite for an introduction to these evolutionary models .
given @xmath0 , the rate matrix @xmath9 and the branch lengths @xmath23 , we then define the following distribution on labelings of the vertices of @xmath0 .
let @xmath24 for @xmath25 .
we first orient the edges of @xmath0 away from an arbitrarily chosen root @xmath26 of the tree .
( we can choose the root arbitrarily since each @xmath16 is reversible with respect to @xmath12 . ) then , the probability of a labeling @xmath27 is @xmath28 the distribution @xmath29 can be generated in an equivalent algorithmic manner .
choose @xmath30 from @xmath12 .
then for each edge @xmath31 , given an assignment for exactly one of the endpoints , say @xmath32 , choose @xmath33 from the distribution defined by the row of @xmath16 corresponding to the label @xmath32 .
let @xmath34 be the marginal distribution of @xmath29 on the labelings of the leaves of @xmath0 ( thus @xmath34 is a distribution on @xmath35 where @xmath1 is the number of leaves of @xmath0 ) .
fix @xmath36 with parameters @xmath37 and @xmath38 as the generating tree .
the goal of phylogeny reconstruction is to reconstruct @xmath36 ( and possibly @xmath37 and @xmath38 ) from @xmath39 ( more precisely , from independent samples from @xmath39 ) .
let @xmath40 denote a set of rate matrices with non - zero entries where @xmath41 .
the set @xmath40 is the set of possible rate matrices .
the set can be arbitrary , usually it is determined by the model considered ( e.g. , for the jukes - cantor model @xmath40 would contain rate matrices whose off - diagonal entries are the same ) .
one often assumes that the rate matrix @xmath37 is known . in this case
we would set @xmath42 . on the other hand
, our results also apply if one sets @xmath40 to be the set of all rate matrices with non - zero entries .
we consider the likelihood of a tree @xmath0 as , the maximum over rate matrices @xmath43 and over assignments of non - zero branch lengths @xmath44 to the edges of @xmath0 , of the probability that the tree @xmath45 generated @xmath46 .
more formally , the maximum expected log - likelihood of tree @xmath0 for distribution @xmath47 is defined by @xmath48 where @xmath49 for a set of characters @xmath50 where @xmath51 , define the log - likelihood of a tree @xmath0 as @xmath52 our goal is to sample from the distribution on the set of phylogenetic trees with @xmath1 leaves where the weight of a tree is @xmath53 . in section [ sec : posterior ]
we will look at the straightforward extension to the setting where we are given a prior on trees and parameters @xmath54 , and our goal is to sample from the posterior distribution .
we analyze a markov chain using transitions made by subtree pruning - and - regrafting ( spr ) .
spr transitions are a natural combinatorial transition , which is also popular in practice . in section [ remarks ]
we discuss several other well - studied choices for the transitions .
here we consider trees weighted by their maximum likelihood . in section [ sec : posterior ] we discuss how the markov chain definition and our main result extends to sampling the posterior distribution .
an spr transition from a tree t works by choosing an ( internal or terminal ) edge @xmath55 .
if @xmath11 is an internal edge we consider one of the two subtrees in @xmath56 , either the subtree @xmath57 containing @xmath58 or the subtree @xmath59 containing @xmath60 .
let @xmath57 denote the selected subtree .
if @xmath11 is a terminal edge , let @xmath57 be the endpoint of @xmath11 that is a leaf .
let @xmath61 denote the tree formed by removing @xmath57 from @xmath0 , in particular , we remove @xmath57 and edge @xmath11 from @xmath0 and `` smooth away '' the vertex @xmath60 ( that is , contract one of the two adjacent edges ) .
we then choose an edge @xmath62 in @xmath61 and we attach @xmath63 onto @xmath62 by adding a new intermediate vertex along @xmath62
. see figure [ fig : spr ] for an illustration .
let @xmath64 denote the tree resulting from the above transition .
is marked by an arrow .
the subtree containing @xmath65 and @xmath66 is pruned , and then regrafted at the edge @xmath62 marked by a starred arrow .
the resulting tree is illustrated.,height=192 ] we analyze the following markov chain which chooses a random subtree @xmath63 to prune , and then chooses an edge to regraft @xmath63 along based on the maximum likelihood of the resulting tree .
this markov chain is an analogous to heat bath chains studied in statistical physics ( as opposed to metropolis chains ) , e.g. , see @xcite , thus we refer to the below chain as the heat bath spr markov chain .
here is the formal definition of the transitions @xmath67 of the heat bath spr markov chain . from a tree @xmath68 at time @xmath69 we proceed as follows . 1 . choose a random subtree @xmath63 of @xmath68 , by choosing a random edge @xmath11 and then choosing one of the two subtrees hanging off of @xmath11
let @xmath61 denote the tree formed by deleting @xmath63 and @xmath11 from @xmath0 .
2 . for each edge @xmath62 of tree @xmath61 , let @xmath70 where @xmath71 is the tree formed by pruning @xmath63 from @xmath0 and regrafting @xmath63 onto edge @xmath62 . let @xmath72 be the distribution on edges of @xmath61 where @xmath73 and @xmath74 .
sample an edge @xmath62 from the distribution @xmath72 on edges of @xmath61 .
graft @xmath63 onto edge @xmath62 and move to this new tree , i.e. , set @xmath75 .
we now verify that the above markov chain is ergodic and reversible with respect to the distribution @xmath12 on trees where @xmath76 , and thus @xmath12 is also the unique stationary distribution .
let @xmath77 and @xmath78 be neighboring states of the markov chain .
let @xmath63 be the tree that is pruned and regrafted to obtain @xmath78 from @xmath77 .
note that the same tree @xmath63 can be pruned and regrafted to obtain @xmath77 from @xmath78 .
the transition probability from @xmath77 to @xmath78 is the probability of choosing @xmath63 in step @xmath79 times @xmath80 .
similarly the transition probability from @xmath78 to @xmath77 is the probability of choosing @xmath63 in step @xmath79 times @xmath81
( note that @xmath82 is the same in both cases since pruning @xmath63 results in the same tree @xmath61 ) .
the detailed balance condition is satisfied for @xmath76 and hence it is the unique stationary distribution .
let @xmath83 denote the ( total ) variation distance between a pair of probability distributions @xmath46 and @xmath84 defined on the same finite , discrete space , and let @xmath85 denote the @xmath69-step distribution of the markov chain from initial state @xmath86 .
the mixing time @xmath87 is defined as @xmath88 which is the time to reach variation distance @xmath89 of the stationary distribution from the worst initial state .
note , it is straightforward to then `` boost '' so that for any @xmath90 , after @xmath91 steps we are within variation distance @xmath92 of @xmath12 , from the worst initial state ( see aldous @xcite ) .
mcmc algorithms are an important tool for phylogenetic reconstruction .
mrbayes @xcite is a popular program that relies on mcmc methods for bayesian inference of phylogeny .
mrbayes uses a sophisticated variant of mcmc known as metropolis - coupled mcmc @xcite . for statistical inference problems , such as phylogenetic reconstruction
, it is often easy to design appropriate mcmc algorithms , such as the above markov chain we defined using spr transitions , which converge in the limit over time to the desired posterior distribution .
however , the computational efficiency of these methods rely on their _ fast _ convergence to the posterior distribution .
since theoretical results are typically lacking , heuristic methods are used to measure convergence to the desired distribution . hence , there are often no rigorous guarantees on the scientific computations which rely on the random samples produced by the mcmc methods .
our goal is to provide some theoretical understanding of settings where mcmc methods for phylogenetic reconstruction are provably fast and hence yield accurate results , and settings where the mcmc methods are slow and consequently the samples may be misleading .
there are several works with computational experiments on the convergence rates of mcmc algorithms for phylogenetic reconstruction , e.g. , see the recent works @xcite .
there is relatively little theoretical work .
diaconis and holmes @xcite proved fast convergence of a markov chain to the uniform distribution over phylogenetic trees . recently , several works have shown examples of heterogeneous data where mcmc algorithms are provably slow to converge .
mossel and vigoda @xcite proved slow convergence for a class of examples with data arising from a uniform mixture of a pair of 5-taxa trees ( with different topologies ) .
tefankovi and vigoda @xcite proved slow convergence for a class of mixture examples from a pair of 5-taxa trees that share the same topology but differ in their branch lengths . in these slow mixing results ,
the convergence time is exponential in the number of characters ( i.e. , sequence length ) . in this paper
we show fast convergence for data from a homogenous source of closely related species . in particular , for data generated from a single tree ( of any size )
when all the branch lengths are sufficiently short , we prove fast convergence .
the requirement of sufficiently short branches is for our proof technique , but it is important to note that the slow mixing results mentioned earlier @xcite require , in an analogous manner , sufficiently short branches . if one searches for the tree with the maximum likelihood ( or maximum a posteriori probability ) our methods show that in our setting of very short branch lengths , the space of trees ( connected by the spr moves ) has no local maxima and hence one can find the optimal tree using hill climbing . for simplicity , we present our results here for the case where the weight of a tree is the maximum likelihood of generating the given data @xmath93 where the maximum is over a rate matrix @xmath9 ( common to all edges ) and a set of branch lengths @xmath44 . this
is closely related to the posterior distribution when the priors are uniformly distributed .
our results extend to @xmath94-regular priors , which are priors that are lower bounded by some @xmath90 , see section [ sec : posterior ] for a discussion on the extension of our results to sampling the posterior distribution .
we are interested in the mixing time @xmath87 , defined as the number of steps until the chain is within variation distance @xmath89 of the stationary distribution .
we prove that the heat bath spr markov chain converges quickly to its stationary distribution when the data is generated from a tree @xmath36 where all of the branch lengths are sufficiently small , and there are sufficiently many samples generated from @xmath36 . here
is the formal statement of our main result .
[ thm : main ] consider any reversible @xmath95-state model , any phylogenetic tree @xmath36 on @xmath1 taxa and any rate matrix @xmath37 with no zero entries .
for all @xmath96 , there exists @xmath97 , such that for all @xmath98 and any choice of branch lengths @xmath99 for @xmath100 , there exists @xmath101 where the following holds . for a data set with @xmath102 characters , each chosen independently from the distribution @xmath39 , then , with probability @xmath103 over the data generated , the heat bath spr markov chain has mixing time @xmath104 .
since there is considerable quantification in the above theorem we will take a moment to dissect it at a high - level .
first off , the requirement that @xmath102 comes from needing the data to look very much like the generating distribution @xmath105 .
therefore , how much data we need depends on several quantities , such as the minimum probability of a configuration arising in @xmath47 , which depends on the minimum branch length and the minimum rate in @xmath37 .
hence , @xmath106 depends on @xmath107 and @xmath37 , and is exponential in @xmath1 .
a somewhat related question has been studied by steel and szkely @xcite on how large @xmath108 needs to be so that the maximum likelihood tree is the generating tree @xmath36 . in their results
one also needs @xmath108 to be exponential in @xmath1 .
our proof uses the fact that in the setting of theorem [ thm : main ] , where the branch lengths are sufficiently short , the leading term of the maximum likelihood function is actually maximum parsimony .
such a result is well - known in the mathematical phylogeny community , and was first observed by felsenstein @xcite .
we require a more detailed statement of such a result , which we present in lemma [ lem : opt - weights ] in section [ sec : likelihood ] .
our main technical contribution is a combinatorial proof that , in the setting of theorem [ thm : main ] , spr moves can be used in a greedy manner to quickly find the maximum parsimony tree .
this result is presented in section [ sec : a - analysis ] .
finally , in section [ sec : rapid - mixing ] we show how theorem [ thm : main ] follows in a straightforward manner from these combinatorial results . in section [ sec : posterior ]
we discuss how theorem [ thm : main ] extends to bayesian inference .
we make some concluding remarks in section [ remarks ] .
to prove theorem [ thm : main ] we will analyze , for every tree @xmath0 , the maximum expected log - likelihood @xmath109 where @xmath105 ( recall that @xmath109 is the maximum expected log - likelihood of @xmath0 maximized over all rate matrices @xmath9 and all edge lengths @xmath44 , see ) . to analyze @xmath109 we will consider the dominant terms of the likelihood function .
we will show that @xmath110 where @xmath111 is the entropy of the stationary distribution of @xmath37 and thus is the same for every @xmath0 . by taking @xmath112
sufficiently small the last term @xmath113 can be ignored .
therefore , the dominant term is @xmath114
. we will prove that the function @xmath115 decreases with each optimal spr move .
hence , since @xmath116 is negative , we then have that @xmath36 has the highest maximum expected log - likelihood , and as the markov chain gets closer to @xmath36 the maximum expected log - likelihood will increase .
theorem [ thm : main ] will then follow in a straightforward manner .
let @xmath0 be a tree with leaves @xmath117 .
let @xmath118 be a partition of the leaves into two parts @xmath119 .
note that we only consider the partition of the leaves without any regard for the internal vertices .
let @xmath120 denote the size of the cut ( i.e. , a subset of edges ) of minimum size that disconnects @xmath121 from @xmath122 .
tuffley and steel @xcite showed that the quantity @xmath120 is the parsimony score of the character corresponding to @xmath118 ( see also semple and steel ( * ? ? ?
* proposition 5.1.6 ) ) , where the character corresponding to @xmath123 assigns all leaves in @xmath121 some @xmath124 and assigns all leaves in @xmath122 some @xmath125 .
for an edge @xmath7 , the removal of @xmath11 splits @xmath0 into two components .
this induces a partition of the leaves of @xmath0 into two parts .
we will call this partition @xmath126 .
the following is the main technical tool for our results .
the lemma describes the high - order terms of the likelihood function as @xmath127 . throughout this paper
, the asymptotic notations @xmath128 and @xmath129 are parameterized by @xmath130 .
roughly speaking , the lemma shows there is a function @xmath115 which plays a leading role in the maximum likelihood . in words , @xmath115 looks at each partition @xmath118 of leaves in the generating tree @xmath36 realized by cutting a single edge @xmath62 .
it then considers the minimum number of edges in @xmath0 to realize this partition @xmath118 times the branch length of @xmath62 in the generating tree . as mentioned in the introduction , there are earlier results which show that the leading term of the likelihood function is the parsimony score , e.g. , felsenstein @xcite , and in the below lemma , the function @xmath131 is the leading term of the expected parsimony score .
we require a more detailed statement than we found in the literature .
[ lem : opt - weights ] let @xmath132 be trees with leaves @xmath117 and @xmath37 be a rate matrix reversible with respect to @xmath133 .
assume that the matrix @xmath37 is normalized ( that is , @xmath134 ) and that @xmath37 has no zero entries .
let @xmath36 have branch lengths @xmath135 , where @xmath136 $ ] , for all @xmath100 , where @xmath96 .
let @xmath137 for @xmath105 the following holds : @xmath138 where @xmath139 is the entropy of @xmath133 , and the constant in the @xmath140 is independent of the choice of the @xmath141 ( but does depend on @xmath142 and @xmath143 ) . as a consequence of the above lemma , to analyze the expected log - likelihood on the tree space , when @xmath112 is sufficiently small , we simply have to consider the function @xmath144 .
in the next subsection we will investigate the combinatorial properties of @xmath131 . before presenting the proof of lemma [ lem : opt - weights ] we give a brief outline of its proof .
let @xmath47 denote the probability distribution defined by @xmath37 and @xmath36 on assignments of labels from @xmath4 to the leaves . in this generating distribution @xmath47 ,
to prove we only need to consider two types of assignments .
the first type are constant assignments where no substitutions occur and thus all leaves receive the same label @xmath145 , these are denoted as @xmath146 .
the second type are assignments obtained by a substitution along just one edge @xmath62 . in this case
, the cut obtained by deleting edge @xmath62 plays an important role . by deleting @xmath62 from @xmath36 ,
the leaves are partitioned into two sets @xmath121 and @xmath122 , denoted as @xmath147 .
if a substitution only occurs along edge @xmath62 , then the leaves in @xmath121 will receive the same label @xmath145 , and the leaves in @xmath122 will receive another label @xmath148 .
we denote such an assignment by @xmath149 .
any other type of assignment requires at least two substitutions , and hence has probability at most @xmath150 , which is dominated by the @xmath151 term of . for any tree @xmath0 , to prove that @xmath109 is lower bounded by the right - hand side of , we compute the expected log - likelihood of @xmath47 for the rate matrix @xmath152 and the set of branch lengths @xmath44 where @xmath153 for every edge @xmath11 .
for each edge @xmath62 and its corresponding assignment @xmath149 , the quantity @xmath154 is the minimum number of edges which require a substitution to obtain the assignment @xmath149 on @xmath0 .
hence , the quantity @xmath144 plays an important role when we sum over all edges @xmath62 of @xmath36 .
in particular , by a calculation ( as detailed in below ) , the expected log - likelihood @xmath155 , for this set of branch lengths @xmath44 , is @xmath156 .
since @xmath157 , this implies the lower bound of . to obtain the upper bound of we consider three cases : when the rate matrix @xmath9 has a stationary distribution different from @xmath37 , when there is an edge @xmath11 where @xmath14 is long ( namely @xmath158 ) , and when all edges are short . in the first case of different stationary distributions , by considering the constant assignments it will be easy to establish that there is a difference in the first term of the right - hand side of . when there is a long edge , then the constant assignments are too unlikely to occur . finally ,
if all edges are shorter than @xmath159 , then , by calculation , we show that the expected log - likelihood is at most the right - hand side of .
we now present the formal proof of lemma [ lem : opt - weights ] .
we first make some observations about the distribution @xmath47 .
let @xmath160 denote the transition matrix for @xmath37 , as defined in .
note , for any @xmath11 , any @xmath20 where @xmath161 , we have @xmath162 for any @xmath145 we have @xmath163 for @xmath145 , let @xmath164 denote the constant assignment @xmath165 for all leaves @xmath60 .
note to achieve @xmath146 in @xmath47 we assign label @xmath166 to the root and then we have no substitutions , or we have at least two edges with substitutions .
thus , @xmath167 for an edge @xmath100 and @xmath20 where @xmath161 , let @xmath168 denote the assignment of label @xmath166 to all leaves in one of the partitions of @xmath169 and label @xmath170 to all leaves in the other partition of @xmath169 . in this case
we have : @xmath171 ( to see why is correct , w.l.o.g . , assume that the root is a leaf in the first partition of @xmath169 , and hence to achieve @xmath172 we need to label the root by @xmath166 and have a substitution on @xmath11 , or at least two edges with substitutions . )
now we compute @xmath173 where @xmath153 for each edge @xmath11 of @xmath0 and @xmath152 .
again we will make some observations about @xmath174 . by the same reasoning as we used for
, we obtain @xmath175 we can obtain assignment @xmath176 on @xmath0 using a substitution on @xmath177 edges , and we can not obtain this assignment with fewer substitutions .
hence , @xmath178 therefore , @xmath179 in order to compute the high - order terms of @xmath173 we do not need to consider labelings other than @xmath146 and @xmath176 ( the other labelings have probability @xmath150 in @xmath47 ) . combining , , , and we obtain @xmath180 where in the last inequality we used the fact that @xmath37 is normalized .
this proves the lower bound in .
it remains to prove the upper bound in .
we will show that no rate matrix and no assignment of branch lengths can do better than the bound established in .
let @xmath9 be a rate matrix with stationary distribution @xmath12 .
if @xmath181 then we bound @xmath182 as follows .
first , note that the terms in the sum are negative and hence to obtain an upper bound we will only consider the constant assignments .
second , the probability of constant assignment @xmath146 in @xmath47 is @xmath183 and similarly @xmath184 .
thus @xmath185 where @xmath186 is the k - l divergence of @xmath12 from @xmath133 .
since , by the gibbs inequality , the kl - divergence is positive when @xmath187 , we have established the upper bound in for the case @xmath181 .
now we assume @xmath188 .
let @xmath44 be an assignment of branch lengths to @xmath0 .
let @xmath189 .
suppose that there exists an edge @xmath190 with branch length @xmath191 .
we are going to show that such a @xmath44 has a tiny log - likelihood because of the constant leaf labelings ( i.e. , @xmath192 ) . by , we have @xmath193 , where @xmath194 .
hence , @xmath195 thus @xmath196 as @xmath127 , is smaller than the right - hand side of and we are done . we are now left with the case in which all edges @xmath190 have branch lengths @xmath197 . since we can generate the leaf labelings starting from any vertex , then by starting at a leaf , we see that : @xmath198 moreover , for @xmath100 , to generate @xmath176 , we need to have substitutions across all edges in a cut that realizes @xmath169 .
since the edges are short this happens with probability @xmath199 where @xmath200 is the size of the cut .
since there are at most @xmath201 such cuts and each has size at least @xmath154 , we have that : @xmath202 hence , @xmath203 in light of lemma [ lem : opt - weights ] , we need to analyze how @xmath115 changes with spr moves . by taking @xmath108 sufficiently large , for each subtree @xmath63 , we will only need to analyze the effect of the optimal spr move for @xmath63 ( optimal in terms of minimizing @xmath204 ) . the quantity @xmath115 looks at cuts obtained by single edges of @xmath36 .
for a tree @xmath0 , we classify the edges of @xmath36 as good or bad if their corresponding cut in @xmath36 is realizable in @xmath0 by cutting a single edge .
more precisely , let @xmath205 be the set of good edges for @xmath0 .
let @xmath206 .
the following lemma says that for every tree @xmath207 obtained from @xmath0 by an spr move using @xmath63 , if @xmath207 has more bad edges than @xmath0 , then this was not the optimal spr move using @xmath63 .
namely , there is a tree @xmath61 which is also obtained from @xmath0 by an spr move using @xmath63 , and @xmath61 is such that @xmath208 .
( more precisely , each term in @xmath204 is less than or equal to the corresponding term in @xmath209 and there is a term in @xmath204 which is strictly smaller than the corresponding term in @xmath209 ) .
our proof has some similarity to those of bruen and bryant @xcite which connect the parsimony score of a character to the minimum number of spr transitions needed to obtain the character .
[ lem : t ] for every generating tree @xmath36 and all trees @xmath210 where @xmath0 and @xmath207 differ by a prune - and - regraft of a subtree @xmath63 and such that there exists @xmath211 the following holds .
there exists a tree @xmath61 which differs from @xmath0 by a prune - and - regraft of @xmath63 and such that @xmath212 for every partition @xmath118 realized by single edges in @xmath36 and @xmath213 for partition @xmath118 realized by @xmath214 in @xmath36 .
suppose an edge @xmath215 is good for @xmath0 and is bad for @xmath207 .
let @xmath216 be the partition of the leaves induced by @xmath214 in @xmath36 .
thus , in @xmath0 , there is an edge @xmath217 which partitions the leaves into @xmath218 and @xmath219 .
see figure [ fig : setup ] for an illustration of the setup .
is good for @xmath0.,height=115 ] let @xmath220 denote the subtree `` hanging off '' of @xmath221 in @xmath0 .
more precisely , after deleting @xmath222 from @xmath0 , let @xmath220 be the subtree containing @xmath221 .
let @xmath218 denote the leaves in @xmath220 .
similarly , let @xmath219 denote the leaves and @xmath223 denote the subtree hanging off of @xmath224 .
let @xmath60 denote the root of the subtree @xmath63 .
first we claim that @xmath225 .
suppose @xmath226 and without loss of generality suppose @xmath227 .
see figure [ fig : f - in - s ] for an illustration of this case . , this scenario can not occur.,height=144 ] thus we must be grafting @xmath63 into an edge of @xmath228 .
after such a move , the edge @xmath222 still separates @xmath218 and @xmath219 , and thus @xmath214 is still good . therefore , @xmath225 . from now on , we assume , without loss of generality , that @xmath229 where @xmath230
. see figure [ fig : tt2 ] .
, this must be the scenario.,height=124 ] we construct the tree @xmath61 by taking @xmath0 , pruning @xmath63 and then regrafting @xmath63 along edge @xmath222 , see figure [ fig : t ] .
.,height=144 ] note that @xmath207 is obtained from @xmath0 by regrafting @xmath63 onto an edge in @xmath223 ( otherwise @xmath214 would be good for @xmath207 ) , see figure [ fig : tww ] .
, @xmath63 is regrafted into @xmath223.,height=163 ] the following claim says that the tree @xmath61 satisfies the conclusion of the lemma .
[ claim : t ] for every partition @xmath123 of leaves realized by edges of @xmath36 , it holds that : @xmath231 moreover , for the partition @xmath232 ( corresponding to @xmath214 ) , we have that : @xmath233 the proof of the claim proceeds by constructing a cut in @xmath61 realizing @xmath119 by a small modification of a cut in @xmath207 realizing @xmath119 .
assuming the claim , the proof of the lemma is now complete .
we now prove the above claim .
we continue using the setup and notation from the proof of lemma [ lem : t ] in section [ sec : a - analysis ] .
recall , the claim says that for every partition @xmath123 of leaves realized by edges of @xmath36 that @xmath234 and for the partition @xmath232 , @xmath235 .
first we argue that @xmath235 .
note that @xmath236 since @xmath214 is bad for @xmath207 . on the other hand , @xmath237 since cutting @xmath238 separates @xmath218 and @xmath219 .
now we just need to argue that @xmath234 .
let @xmath239 be an edge in @xmath36 .
let @xmath123 be the corresponding partition in @xmath36 .
note that if @xmath240 is in the subtree with leaves @xmath218 then @xmath241 on the other hand , if @xmath240 is in the subtree with leaves @xmath219 then @xmath242 consider a minimum cut @xmath243 that realizes @xmath119 in @xmath207 and amongst these minimum cuts is the one with the fewest number of edges in subtrees @xmath244 and @xmath63 .
we claim that @xmath221 is reachable from a leaf of @xmath244 in @xmath245 and that @xmath60 is reachable from a leaf of @xmath63 in @xmath245 .
suppose that @xmath221 is not reachable from a leaf of @xmath244 .
let @xmath246 be the edge in @xmath247 closest to @xmath221 .
we claim that @xmath248 realizes @xmath119 .
if there was a pair of leaves in @xmath244 each in different @xmath249 that are connected in @xmath250 then by the choice of @xmath246 one of those leaves would be connected to @xmath221 in @xmath245 , a contradiction with the assumption that @xmath221 is not reachable from a leaf of @xmath244 in @xmath245 .
thus @xmath121 and @xmath122 are still separated in @xmath244 in @xmath250 ; @xmath121 and @xmath122 are still separated by @xmath251 in @xmath252 since @xmath253 in this subtree ; and @xmath254 ensures that pairs across @xmath222 are separated .
note that @xmath255 and @xmath251 has fewer edges in @xmath244 and @xmath63 , a contradiction with the choice of @xmath66 .
thus @xmath221 is reachable from some leaf of @xmath244 in @xmath245 .
the argument for @xmath63 and @xmath60 is the same . since a leaf of @xmath63 is reachable from @xmath60 in @xmath245 , then in other words a ( non - empty ) subset of @xmath121 and/or @xmath122 are reachable from @xmath60 .
moreover , since @xmath66 realizes the partition @xmath119 , then a subset of only one of the @xmath249 is reachable from @xmath60 in @xmath245 ; we will say @xmath60 is of type @xmath249 to signify the @xmath249 reachable from @xmath60 .
analogously , we say @xmath256 is of type @xmath249 for the set reachable from @xmath221
. if @xmath60 and @xmath221 are of the same type @xmath249 , let @xmath257 we claim @xmath251 realizes @xmath119 in @xmath61 . to see this ,
note that if a path ( between a pair of leaves ) exists in @xmath61 and does not exist in @xmath207 then it must include @xmath258 which is the new vertex in @xmath61 where @xmath63 is regrafted , see figure [ fig : t ] for an illustration .
now we argue that such a path can not connect a leaf in @xmath121 with a leaf in @xmath122 in @xmath259 .
note that only a subset of @xmath249 is reachable from @xmath258 in @xmath259 , since @xmath258 can reach the same set of vertices ( outside of @xmath63 ) in @xmath259 as @xmath221 does in @xmath245 , and only a subset of @xmath249 is reachable from @xmath60 in @xmath260 .
finally , since @xmath261 we have that @xmath212 which completes the proof in this case .
now suppose @xmath221 is of type @xmath121 and @xmath60 is of type @xmath122 .
this means a leaf of @xmath63 is in @xmath122 and since @xmath262 , it is also in @xmath218 and we are in case , thus @xmath263 .
note @xmath66 has to separate @xmath221 from @xmath219 by some set of edges @xmath264 .
let @xmath265 .
the new pairs of leaves that are connected in @xmath259 ( but not in @xmath245 ) are either both from @xmath219 and hence @xmath122 , or are connected by a path that exists in @xmath61 and does not exist in @xmath207 . as in the previous case ,
if a path ( between a pair of leaves ) exists in @xmath61 and does not exist in @xmath207 then it must include @xmath258 which is the new vertex in @xmath61 where @xmath63 is regrafted . note that @xmath258 is disconnected from @xmath244 in @xmath259 ( since @xmath254 ) .
the leaves of @xmath223 are from @xmath122 and the leaves of @xmath63 reachable from @xmath60 in @xmath260 are also from @xmath122 .
therefore the new paths do not connect leaves of @xmath121 and @xmath122 .
this completes this case since @xmath255 .
finally , suppose @xmath221 is of type @xmath122 and @xmath60 is of type @xmath121 . in this case
a leaf of @xmath244 is in @xmath122 and is also in @xmath218 and therefore we are again in case , thus @xmath263 .
note @xmath66 has to separate @xmath60 from @xmath219 by some set of edges @xmath264 .
let @xmath266 .
once again , the new pairs of leaves that are connected in @xmath259 ( but not in @xmath245 ) are either both from @xmath219 and hence @xmath122 , or are connected by a path that exists in @xmath61 and does not exist in @xmath207 .
note that @xmath258 is disconnected from the leaves of @xmath63 in @xmath259 .
the leaves of @xmath223 are from @xmath122 and the leaves of @xmath244 reachable from @xmath221 are also from @xmath122
. therefore the new paths in @xmath259 do not connect leaves of @xmath121 and @xmath122 .
this completes this case since @xmath255 .
this completes the proof of the claim . using lemma [ lem : t ]
, we will prove that for every subtree @xmath63 the optimal spr move using @xmath63 does not increase the number of bad edges , and there is a subtree @xmath63 where the optimal spr move using @xmath63 decreases the number of bad edges .
it will then be straightforward to prove rapid mixing by analyzing the time until the number of bad edges is zero , and hence we have reached @xmath36 . [
lem : increase ] for all trees @xmath36 , every choice of parameters @xmath267 , for all trees @xmath268 the following holds , where @xmath269 is defined in . 1 . for any subtree @xmath63 of @xmath0
the following holds .
let @xmath270 be any tree which minimizes @xmath271 amongst the spr neighbors of @xmath0 which differ by a prune - and - regraft of @xmath63 .
then , @xmath272 2 .
there exists a subtree @xmath63 of @xmath0 where the following holds .
let @xmath270 be any tree which minimizes @xmath271 amongst the spr neighbors of @xmath0 which differ by a prune - and - regraft of @xmath63 .
then , @xmath273 part 1 of lemma [ lem : increase ] follows immediately from lemma [ lem : t ] . to prove part 2 we choose a particular ` minimal ' subtree @xmath63 . roughly speaking
, we consider the bad edge @xmath214 that is closest to the leaves in @xmath36 , and take the subtree @xmath63 hanging off of @xmath214 .
if is violated then there exists @xmath274 , and hence by lemma [ lem : t ] , there exists @xmath61 ( which differs from @xmath0 and @xmath270 by a prune - and - regraft of @xmath63 ) such that no cuts increased in size and the cut corresponding to @xmath214 is smaller .
therefore , @xmath275 , contradicting the choice of @xmath270 .
therefore , part 1 holds .
we now prove part 2 .
we first claim that there is an spr move that decreases the number of bad edges .
[ claim : fewer - bad ] for every tree @xmath0 , there is an spr move resulting in a tree @xmath61 where @xmath276 now we argue that part 2 of the lemma follows from the above claim and part 1 .
we then go back to prove the claim .
consider a subtree @xmath63 of @xmath0 .
let @xmath277 denote those trees obtainable from @xmath0 by a prune - and - regraft of @xmath63 .
note , for any @xmath278 , we have that @xmath279 , since when we prune @xmath63 from @xmath0 and @xmath61 we have the same subtree remaining .
let @xmath61 denote the neighboring tree from claim [ claim : fewer - bad ] with fewer bad edges , and let @xmath63 denote the subtree where @xmath278 .
let @xmath280 denote the tree in @xmath277 which minimizes @xmath281 .
as noted above we must have that @xmath279 .
thus , @xmath280 is also the neighbor of @xmath61 that minimizes @xmath281 .
therefore , we can apply part 1 of the lemma for tree @xmath61 and subtree @xmath63 , and conclude that @xmath282 .
combined with we then have that : @xmath283 which proves part 2 .
we now prove claim [ claim : fewer - bad ] .
let @xmath214 in @xmath36 be an edge in @xmath284 that is `` closest '' to the leaves in the following precise sense .
say @xmath214 joins subtrees @xmath285 and @xmath286 in @xmath36 where the number of vertices in @xmath285 is at most the number of vertices in @xmath286 .
then we say the distance of @xmath214 to the leaves is the number of vertices of @xmath285 .
note , by the choice of @xmath214 , @xmath285 contains no bad edges for @xmath0 .
first , note that @xmath285 must contain at least two leaves because , in any tree , any single leaf can be separated from the rest of the leaves by deleting one edge ( which would contradict that @xmath214 is bad ) .
let @xmath287 and @xmath288 denote the two subtrees of @xmath285 hanging from the root of @xmath285 in @xmath36 . both @xmath287 and @xmath288 must exist since @xmath285 contains at least two leaves .
let @xmath218 and @xmath219 denote the leaves in @xmath287 and @xmath288 , respectively .
since @xmath214 is the closest bad edge to the leaves , there is a subtree @xmath220 in @xmath0 whose leaves are @xmath218 , and also a subtree @xmath223 whose leaves are @xmath219 .
moreover , by induction , @xmath289 and @xmath290 . in @xmath0 , by pruning @xmath223 and then regrafting along the edge incident to @xmath220 we obtain a copy of @xmath285 in @xmath0 .
see figure [ fig : claim ] for an illustration .
let @xmath61 be the tree resulting from this spr move .
note , @xmath214 is now a good edge in @xmath61 . with fewer bad edges.,height=144 ]
it remains to argue that other edges of @xmath36 did not change from good for @xmath0 to bad for @xmath61 .
note , edges in @xmath287 and @xmath288 remain good in @xmath61 since they are realizable in @xmath220 and @xmath223 , respectively .
consider an edge @xmath62 of @xmath36 where @xmath291 .
let @xmath119 be the partition of the leaves realized by @xmath62 in @xmath36 .
note that @xmath216 are in the same partition , since tree @xmath285 is not cut by @xmath62 .
let @xmath292 be the edge in @xmath0 that realizes @xmath119 .
after pruning - and - regrafting @xmath223 ( to form @xmath61 ) , @xmath292 still realizes the partition @xmath119 since @xmath218 and @xmath219 are in the same partition .
hence , @xmath62 is still good for @xmath61 .
therefore , @xmath293 , which completes the proof of the claim . finally , we prove that when the number of bad edges increases then @xmath115 also increases by a significant amount . as a consequence , in our analysis of the markov chain , by taking @xmath112
sufficiently small , we can focus on how a transition changes @xmath115 and hence on the change in the number of bad edges .
[ lem : gap ] for any trees @xmath0 and @xmath61 which differ by one spr move , if @xmath294 then @xmath295 let @xmath63 be the subtree used to move between @xmath0 and @xmath61 .
let @xmath277 denote those trees obtainable from @xmath0 by a prune - and - regraft of @xmath63 .
note @xmath296 .
consider @xmath270 which minimizes @xmath271 amongst the spr neighbors of @xmath0 which differ by a prune - and - regraft of @xmath63 . since @xmath297 then @xmath270 is also the neighbor of @xmath61 that minimizes @xmath271 .
fix @xmath298 . by part 1 of lemma [ lem :
increase ] , @xmath299 the proof of our main theorem now follows from a straightforward argument .
we show that the heat bath spr markov chain behaves like a local search algorithm , and then a simple coupling argument gives the mixing result .
let @xmath300 denote the space of phylogenetic trees on @xmath1 taxa .
for a tree @xmath301 , and subtree @xmath63 of @xmath0 , let @xmath277 denote those trees obtainable from @xmath0 by pruning - and - regrafting @xmath63 .
let @xmath66 be the constant in the @xmath140 notation of for the chosen @xmath107 and @xmath302 . by choosing @xmath303 (
note that @xmath303 is an upper bound on @xmath112 ) sufficiently small then for every tree @xmath0 , in , the @xmath304 is smaller than @xmath305 and therefore : @xmath306 for some @xmath307 . fix a tree @xmath268 and a subtree @xmath63 of @xmath0 . by lemma [ lem : gap ] and , for every @xmath278 where @xmath308 we have : @xmath309 for a character @xmath310 , let @xmath311 . a straightforward application of hoeffding s inequality @xcite and a union bound over @xmath310 implies , for all @xmath90 : @xmath312 let @xmath313 denote a lower - bound on the off - diagonal entries in the rate matrix . for @xmath303 sufficiently small , every labeling of the leaves has probability at least @xmath314 ; this follows from the fact that for every edge , every transition has probability @xmath315 , see for a precise statement .
hence , by choosing @xmath303 sufficiently small ( relative to @xmath316 and the constant in the error term of ) , then for all @xmath310 , @xmath317 .
let @xmath318 then , for @xmath319 , @xmath320 with probability @xmath321 for @xmath108 sufficiently large .
the probability of moving from @xmath0 to @xmath61 after choosing @xmath63 in step 1 is at most : @xmath322 for @xmath108 sufficiently large . therefore , with probability @xmath323 , the chain will move from @xmath0 to some @xmath270 ( where @xmath270 is a tree that can be obtained from @xmath0 by an spr move and such that it minimizes @xmath281 ) , and thus by part 1 of lemma [ lem : increase ] the number of bad edges will not increase . moreover ,
if we choose the subtree @xmath63 satisfying part 2 of lemma [ lem : increase ] then the number of bad edges will decrease .
hence , with probability @xmath324 the number of bad edges decreases by at least one . in expectation , after @xmath325 steps of the chain the number of bad edges will be zero , in which case we have reached @xmath36 . by markov s inequality , with probability @xmath326 after @xmath327 steps we reach @xmath36 .
once we reach @xmath36 the probability of moving to a different tree within @xmath327 steps is at most @xmath328 .
hence the claimed mixing time follows by an elementary coupling argument ( c.f .
, @xcite for an introduction to the coupling technique ) since from any pair of initial trees , both chains ( run independently ) reach @xmath36 at time @xmath327 with probability @xmath329 .
the goal is often to randomly sample from the posterior distribution over trees . to do this , we consider a markov chain whose stationary distribution is the posterior distribution and analyze the chain s mixing time , which is a measure of the convergence time of the chain to its stationary distribution .
let @xmath330 denote a prior density where @xmath331 our results extend to priors that are lower bounded by some @xmath90 as in mossel and vigoda @xcite . in particular , for all trees @xmath0 and all branch lengths @xmath44 where @xmath332 for all edges @xmath11 , we require @xmath333 .
we refer to these priors as @xmath334-regular priors .
applying bayes law we get the posterior distribution : @xmath335 where @xmath336 each tree @xmath0 then has a posterior weight @xmath337 finally , the posterior distribution @xmath46 on trees is defined as @xmath338 . to sample from the posterior distribution ,
the markov chain is defined as in section [ sec : spr ] except that in step 2 the weight @xmath339 is now set as @xmath340 defined in .
this ensures that the markov chain is reversible with respect to the posterior distribution , and hence this is the unique stationary distribution .
theorem [ thm : main ] then extends to hold for any priors which are @xmath341-regular the proof easily extends to this case in the following manner .
in particular , we need to modify the statement of lemma [ lem : opt - weights ] so that , for any tree @xmath0 , is achieved for @xmath152 and for every set of branch lengths @xmath44 where @xmath342 for all edges @xmath11
. then we can use the same proof as lemma 21 in mossel and vigoda @xcite to get an analog of to hold for the posterior weights defined in in place of the maximum likelihood function @xmath343 , and the remainder of the proof of theorem [ thm : main ] remains the same .
* nni transitions : * in a nni transition , an internal edge @xmath11 is chosen . since internal vertices have degree three ,
there are four subtrees hanging off of @xmath11 .
there are three possible ways of attaching these four subtrees to @xmath11 , and an nni transition moves to one of these rearrangements .
there are trees @xmath0 ( different from the generating tree @xmath36 ) where no nni neighbor ( strictly ) improves @xmath115 ; moreover , there are cases where there is also no improvement in the next term of .
we are uncertain as to whether theorem [ thm : main ] holds for a markov chain based on nni transitions
. it would be especially intriguing if there are cases where chains based on nni transitions are slow to converge ( so - called torpidly mixing ) , whereas a chain based on spr transitions is provably fast to converge ( rapidly mixing ) .
* possible future work : * there are now several works with proofs of convergence of mcmc algorithms for phylogenetic reconstruction in certain settings rapid mixing results in this paper and torpid mixing results in mossel and vigoda @xcite and tefankovi and vigoda @xcite ) .
all of these results require that the branch lengths are sufficiently small so that only the dominant terms of the likelihood function need to be considered .
a natural avenue for extending this paper , is to allow arbitrary branch lengths on the terminal edges .
* rapid or torpid mixing for general pure distributions : * the most tantalizing question to the authors is whether there exists a pure distribution ( i.e. , a single generating tree as in the setting of this paper ) where markov chains based on all of the natural transitions ( e.g. , nni , spr and tbr transitions ) are slow to converge to the stationary distribution ( in other words , they are torpidly mixing ) .
we expect simulations can be quite useful for finding such a bad example , if one exists ; in fact , our previous work @xcite on this topic was inspired by some intriguing findings from some simple simulations .
we are thankful to the anonymous referees for many useful comments on the manuscript .
b. a. berg .
introduction to markov chain monte carlo simulations and their statistical analysis . in _
markov chain monte carlo : innovations and applications _ , eds . , w. s. kendall , f. liang , and j .- s .
wang , world scientific publishing co. , singapore .
pages 1 - 52 , 2005 . | this paper studies a markov chain for phylogenetic reconstruction which uses a popular transition between tree topologies known as subtree pruning - and - regrafting ( spr ) .
we analyze the markov chain in the simpler setting that the generating tree consists of very short edge lengths , short enough so that each sample from the generating tree ( or character in phylogenetic terminology ) is likely to have only one mutation , and that there enough samples so that the data looks like the generating distribution .
we prove in this setting that the markov chain is rapidly mixing , i.e. , it quickly converges to its stationary distribution , which is the posterior distribution over tree topologies .
our proofs use that the leading term of the maximum likelihood function of a tree @xmath0 is the maximum parsimony score , which is the size of the minimum cut in @xmath0 needed to realize single edge cuts of the generating tree .
our main contribution is a combinatorial proof that in our simplified setting , spr moves are guaranteed to converge quickly to the maximum parsimony tree .
our results are in contrast to recent works showing examples with heterogeneous data ( namely , the data is generated from a mixture distribution ) where many natural markov chains are exponentially slow to converge to the stationary distribution . |
recent developments in the machine learning community have brought innovative ideas and methods to solve problems in data mining and predictive analytics with important applications in the business world such as credit risk , fraud detection , survey direct response , customer segmentation , etc .. among such methodologies ensemble learning has been an active research area within the last two decades , where it is believed that combining the predictions of different models or base models , e.g. , by weighted averaging , can bring an improvement in the prediction error in comparison to the error provided by the best individual model or base model .
alternative ways that we will be using throughout the paper to refer to base models are base learner , level-0 learner or level-0 models .
the setting of the approximation for regression we are considering is in the context of supervised learning : given an historical dataset @xmath0 with @xmath1 training samples @xmath2 , @xmath3 , where @xmath4 , and @xmath5 is a stochastic error , the task is to approximate the unknown function @xmath6 by @xmath7 , so we fit a model @xmath8 to the training data by minimizing the square error @xmath9 interested readers can consult @xcite and @xcite for a deeper discussion about this topic .
the aim of this work is to approximate the unknown function @xmath10 using a linear combination of base learners , or learning algorithms .
so , instead of a single estimator @xmath7 we have a collection of them ; @xmath11 .
we then learn each individual @xmath12 model separately using the error function in equation ( 1 ) on the training samples .
once this is achieved , the outputs of the learned models are combined resulting in an approximating ensemble function @xmath13 : @xmath14 the above function may be used to make predictions of @xmath15 for a new instance @xmath16 .
there are different ways to determine the @xmath17 weight vector . in section 2.3
we introduce the way we are going to calculate it in this paper . the expected error of @xmath18 , or @xmath19
@xmath20 @xmath21 @xmath22 , is defined as @xmath23 $ ] .
the @xmath24 can be used to measure the generalization error of a predictor on new instances , and can be decomposed as the sum of two prediction errors @xcite , the variance , @xmath25 , or spread of @xmath18 around its average prediction , @xmath26 $ ] , and the square of the bias , @xmath27 , or amount by which the average prediction of @xmath18 differs from @xmath28 . @xmath29 + [ \bar{f}(\boldsymbol{x}_0 ) - f^*(\boldsymbol{x}_0 ) ] ^2 $ ] this is known as the bias - variance dilemma .
the variance term can be further decomposed for an ensemble as follows @xcite : @xmath30 = e[{(\tilde{f}(\boldsymbol{x}_0)-e(\tilde{f}(\boldsymbol{x}_0 ) ) ) } ^2 ] = \\
\sum_{j=1}^{j}\alpha_j^2(e[f_j^2(\boldsymbol{x}_0)]-e^2[f_j(\boldsymbol{x}_0 ) ] ) + \\ 2\sum_{j
< k}\alpha_j\alpha_k(e[f_j(\boldsymbol{x}_0)f_k(\boldsymbol{x}_0 ) ] - e[f_j(\boldsymbol{x}_0)]e[f_k(\boldsymbol{x}_0 ) ] ) \end{split}\ ] ] the expectation operator is taken with respect to the historical , or training dataset @xmath0 .
the first sum contains the lower limit of the ensemble variance , which is the weighted mean of the variance of ensemble members .
the second sum contains the cross terms of the ensemble members , which disappears if the models are completely uncorrelated @xcite . according to the principle of the above bias - variance trade - off , an ensemble consisting of diverse models with much disagreement is more likely to have a good generalization performance @xcite .
so , how to generate diverse models is very important .
for example , for neural networks , @xcite suggest to initialize different weights for each neural network models , training neural networks with different training subsets , etc .. in our work , the main driver to achieve that purpose will be to inject a diversity mechanism into the cross - validation procedure , which will generate new partitions on which corresponding ensembles will be trained .
we will then select a partition , or ensemble where the correlation between their base components satisfies a minimum correlation criterion , which is intended to have an effect in second sum of equation ( 2 ) to lower the variance .
the aim is to establish a trade - off between bias and variance when training any predictor or classifier with good generalization , i.e. , with good performance when applying it on new instances , not present in the training mechanism .
it is generally accepted that combining different classifiers can reduce the variance while not affecting the bias @xcite , @xcite .
it should be mentioned briefly that another approach for improving the generalization error is based on statistical learning theory @xcite , which has its origins with @xcite in @xmath31 @xmath32 @xmath33 or pac learning .
the goal of pac learning is to understand how large a dataset needs to be in order to give good generalization , we will not use this approach in our work . according to @xcite ,
most of the research focuses on one specific approach to build the ensemble , e.g. , sub - sampling from the training set or manipulating the induction algorithm , and further investigation is necessary to achieve gains by combining several approaches .
this observation of using different ideas along with the potential gain we can achieve in prediction accuracy by a joint design of ensemble generation , and integration steps , have motivated us to investigate the performance of standard stacking approaches , and develop new extensions to stacking for regression . in this work we present methodology that we have developed within the framework of ensemble learning through stacked generalization for regression . the main idea behind stacking
is to generate a meta - training set from the model predictions on validation sets with the aim to integrate them through a learning mechanism , e.g. by weighted average .
we propose to use a set of simple base learners , and consider the selection of the best base learner as a benchmark to compare the performance of the stacking approaches .
our empirical evaluation shows that the oracle composed of best base learners performs better than the best of the standard stacking approaches used in this work . in this paper
we will refer to any of the standard stacking approaches as level-1 learner , level-1 ensemble , or base ensemble as well .
we propose a couple of extensions to the standard stacking approach .
the first extension considers combining standard stackings into an ensemble of level-1 learners , using a two - step level-2 ensemble learning , i.e. , we are making the ensemble grow . the two - step level-2 ensemble learning scheme introduced here
provides a way to measure the diversity of the base ensembles , or level-1 learners through computing pairwise correlations between its components , which will be exploited in a second extension with the aim to improve the prediction accuracy of the two - step level-2 ensemble . the second extension to standard stacking is composed of two parts .
the initial part consists of an extra diversity mechanism injected in the data generation which creates different cross - validation partitions from current cross - validation partition , on which new level-2 ensembles are trained .
the final part , it is an algorithm based on computed level-1 learner correlation criterion , and ranking based - criterion , applied after two - step level-2 ensemble learning , which selects the most appropriate ensemble of standard stacking approaches , and/or corresponding data partition on which to make the final prediction . we show in this work that the latter extension performs better than the best of the standard stacking approaches , or level-1 ensembles , and it is as good as the best base model by cross - validation .
we also show that such extension performs better than two of state - of - the - art ensemble learning algorithms , and it is as good as a third state - of - the - art method . in section 2 , we summarize the state - of - the - art , the stacking framework , and survey some recent approaches of stacking regression . in section 3
, we introduce a set of base models , and level-1 learning which consists of standard stacking approaches , each composed of different combinations of base models .
in section 4 , we present our first extension to standard stacking , level-2 learning , a mixture of ensembles of level-1 learners .
we introduce two different learning schemes we have developed : two - step ensemble learning , and all - at - once ensemble learning .
the former has much richer structure than the latter , which will be exploited in the algorithm presented later in section 5 . in section 5
, we introduce the second extension to stacking , which initially consists of building a diversity mechanism to generate different @xmath34 partitions by perturbing the current @xmath34 partition ; and finally a partition selection is done , using either a computed level-1 learner correlation - based criterion , or a ranking - based criterion within a max - min rule - based heuristic algorithm . at the end
, we present such algorithm for systematic ensemble learning based on generating the partitions in an orderly way . in section 6
, the experimental setup is described for the comparison of the different ensemble approaches , the oracle of databases , and three state of the art ensemble learning algorithms .
section 7 presents and discusses the experimental results , and section 8 concludes .
our results match in error performance the ones obtained with the oracle of databases .
in addition to that our results performed better than two of the state - of - the - art ensemble algorithms for regression , and it also matches the ones obtained with a third state - of - the - art ensemble method for regression .
we review some background information about ensemble learning in section 2.1 , followed by the framework of cross - validation in section 2.2 , and in section 2.3 , we present stacking introduced by @xcite .
we then describe in section 2.4 recent stacking approaches of incremental learning , model integration , and we review as well some recent studies for regression , less referred in literature .
typically , ensemble learning consists of three phases , the first is ensemble generation which generates a set of base models .
second is ensemble pruning where the ensemble is pruned by eliminating some of the base models generated earlier .
finally , in the integration step a strategy is used to obtain the prediction for new cases .
most of the work done in ensemble generation is about generating a single learning algorithm known as homogeneous learning . to ensure a level of diversity amongst the base models ,
one of the following strategies are used @xcite : varying the learning parameters of the learning algorithm , varying the training data employed such as bagging @xcite , and boosting @xcite , both use re - sampling techniques to obtain different training sets for each of the classifiers .
varying the feature set employed , such as random subspace @xcite , which is a feature selection algorithm , where each base learner corresponds to a different subset of attributes from the feature space . in randomize outputs
approach , rather than present different regressors with different samples of input data , each regressor is presented with the same training data , but with output values for each instance perturbed by a randomization process @xcite , @xcite .
another approach , known as heterogeneous learning , is to generate models using different learning algorithms where each model is different from each other .
the conventional approach where each base model has been built with a different learning algorithm but the same training data is called @xmath35 , whereas a @xmath36-@xmath35 heterogeneous ensemble set where the base models are built from more than one learning algorithm , but each base model is not required to be built with a distinct learning algorithm @xcite .
the base models in this last approach can be generated from homogeneous and/or heterogeneous learning . in pruning step
the objective is to reduce the size of the ensemble with the aim to improve the prediction accuracy @xcite .
it has been shown that given a set of @xmath1 base models it is possible that an ensemble performs better if only uses a subset of such models instead of all of them .
@xcite reduces the size using a measure of diversity and accuracy , where the diversity is measured with the positive correlation of base models in their prediction errors .
they used homogeneous learning for stacked regression .
the integration approach can be performed in two ways , either by combining the predictions of different models , or selecting one of the base models to make the final prediction @xcite .
integration can be classified as @xmath37 or @xmath38 .
it is static when for any new instance the combination of predictions is done globally in the same way , based on the whole instance set ; whereas for different new instances the dynamic approach combines the predictions based on different localized regions of the instance space . for the static case
the simplest way to combine is to average the predictions of the base models , which is referred to as the unweighted averaging @xcite . among the weighted averaging techniques we have one based on the performance of individual ensemble members @xcite , or in the case of artificial neural networks @xmath39 to penalize correlations @xcite , and @xcite where the base models produced are negatively correlated .
the simplest static selection approach is to choose the model with minimum cross - validation overall training error referred to as selection by cross - validation @xcite , o also as select best @xcite .
it is very common to use these methods to benchmark ensemble approaches .
cross - validation @xmath40 is a resampling technique often used in data mining for model selection and estimation of the prediction error in regression and classification problems . in order to measure the performance of one model the historical set @xmath0
is split in two parts , the training @xmath41 and the validation @xmath42 .
the training part @xmath41 is used to fit a model which is then used to predict the data on the validation set @xmath42 .
because we know in advance the actual values for the predicted variable in the validation set @xmath42 , then a performance measure such as the @xmath24 , defined below lines for cross - validation , is used to estimate the accuracy achieved by the model .
we should point out that estimating the prediction error must be done on data points on @xmath42 which are outside of the data @xmath41 that was used for model estimation to prevent from having _
overfitting_. overfitting the training data happens when the model fits so well the training data that results in poor generalization when it is used to predict new instances not present in the fitting phase .
the general @xmath43-fold cross - validation procedure is as follows : \1 .
randomly split the data in @xmath43 almost equal parts @xmath44 .
\2 . for each @xmath45th
fold define @xmath46 and @xmath47 to be the validation and training sets for the @xmath45th fold of a @xmath43-fold cross - validation .
invoke the learning algorithm on the training set @xmath48 to estimate a model @xmath49 and calculate its prediction error on the @xmath46 fold .
repeat the above procedure for @xmath50 .
let @xmath51 be the @xmath52-th observation in @xmath46 and be @xmath53 the prediction of the estimated model on @xmath51 .
the cross - validation estimate of mean square error is : @xmath54 we will be using the cross - validated mean square error normalized by the sample variance for model performance , which is defined as @xmath55 stacking is the abbreviation used to refer to stacked generalization @xcite , it has been used for regression tasks @xcite , and classification purposes @xcite .
further development has been done by @xcite . for more recent extensions of stacking see @xcite . in stacking the general idea
is to have level-0 models which consist of a set of @xmath56 different predictors whose task is to generate predictions based on different partitions of a dataset .
those @xmath56 level-0 models are the learned models on a training dataset @xmath57 corresponding to @xmath56 learning algorithms . given a dataset @xmath58 , where @xmath59 is the actual value and
@xmath51 is a vector representing the attribute values of the @xmath52th instance , randomly split the data into @xmath43 almost equal parts to get @xmath46 and @xmath48 the validation and training sets for the @xmath45th fold of a @xmath43-fold cross - validation .
these datasets constitute the @xmath60 @xmath61 .
it should be noted that the validation subsets @xmath46 , @xmath62 are mutually exclusive and that their union gives the whole dataset @xmath57 . for each instance @xmath51 in @xmath46 , the validation set for the @xmath45th cross - validation fold ,
let @xmath63 denote @xmath64-th model fitted on training subset @xmath48 and @xmath65 denote the prediction of the estimated model @xmath66 on @xmath51 . at the end of the @xmath43-fold @xmath34 procedure ,
the data set assembled from the outputs of the @xmath56 models along the @xmath43 validation subsets is @xmath67 the @xmath68 @xmath61 consists of @xmath69 where + @xmath70 .
then , using some learning algorithm that we call the @xmath68 @xmath71 to derive from these level-1 data a model @xmath7 for @xmath72 for prediction of new data @xmath73 .
the @xmath68 @xmath74 is @xmath75 the @xmath17 vector is chosen by minimizing @xmath76 this prediction is expected to be more accurate than , e.g. the one we get with the best level-1 model selected from cross - validation .
@xcite used as level-0 learners either linear regression with different number of variables , or different collections of shrinkage parameters for ridge regression .
then , he used the level-0 models outputs , level-1 data , as input for the level-1 generalizer to minimize the least squares criterion under the constraints @xmath77 , @xmath78 , with the additional constraint @xmath79 .
breiman concluded that both were better than the corresponding single best model as selected by cross - validation . stacking improved subset selection more than it improved ridge regression though , which breiman suggested was due to a greater instability of subset selection .
he also stacked together subset and ridge regressions , which results in a predictor more accurate than either ridge , or variable selection .
the key enabling concept in all ensemble based systems is diversity @xcite .
little research , however , has been devoted to constructing ensembles with different base classifiers @xcite ; heterogeneous ensembles are obtained when more than one learning algorithm is used . in @xcite , they use heterogeneous base classifiers as base learners , and propose two extensions for stacking in classification problems .
one extension is based on an extended set of meta - level features , and the other uses multi - response model trees to learn the meta - level .
they showed , that the latter performs better than existing methods of stacking approaches , and selecting the best classifier from the ensemble by cross - validation .
in heterogeneous learning the generated base models are expected to be diverse .
however , the problem is the lack of control on the diversity of the ensemble components during the generation phase @xcite .
one approach , commonly followed in the literature , combines two approaches : using different induction algorithms , e.g. , heterogeneous models , mixed with the use of different parameter sets , e.g. different neighbourhood sizes , see @xcite and @xcite . another way to overcome
the lack of control in diversity is through ensemble pruning ; @xcite embedded the ensemble integration phase in the ensemble selection one .
@xcite extended the technique of stacked regression to prune an ensemble of base learners by taking into account the accuracy , and diversity .
the diversity is measured by the positive correlation of base learners in their prediction errors .
the above examples show that an important part of the problems at the integration phase can be solved by a joint design of the generation , pruning , when appropriate , and integration phases @xcite . in this work
we propose a set of base learners diverse and accurate , and combine them using stacking into what we already called , standard stacking approach , or level-1 learning .
we generate different standard stackings based on different combination of base models .
we also combine such standard stackings into an ensemble of level-1 learners through a level-2 ensemble learning giving an extra level of smoothing in the predictions , which is a way to control the diversity of such ensembles .
there has been considerably less attention given to the area of homogeneity in the area of stacked regression problems @xcite .
for example , among the best state - of - the art stacking for regression methods is wmetacomb @xcite , which integrates two different integration methods using homogeneous learning .
the method uses a weighted average to combine stacked regression , and a dynamic stacking method , @xmath80 ; the weights are determined based on the error performance of both methods .
tests conducted on 30 regression data sets resulted , on one side , that wmetacomb never looses against stacked regression , where it wins 5 , and draws the remaining 25 ; on the other side , it looses 3 against @xmath81 , where it wins 13 and draws 14 .
a state of the art method using heterogeneous learning for stacking regression is cocktail ensemble learning @xcite .
they used an error - ambiguity decomposition to analyse the optimal linear combination of two ensembles , and extended it to multiple ensembles via pairwise combinations .
they claim that such hybrid ensemble is superior to the base ensemble , simple averaging , and selecting the best ensemble .
it should be noted that our initial work based on standard stacking differs from the literature in two aspects .
first , we used a particular set of level-0 models , that we will introduce in section 3.1 , and second , in section 3.2 we introduce three standard stacking approaches each of them composed of a different combination of base learners .
the @xmath82 learning algorithms we consider in this work are three versions of linear regression , and a fourth learning algorithm which can be seen as a type of non - linear regression .
let @xmath11 be the set of @xmath56 learned base learners of corresponding @xmath56 different learning algorithms trained on a dataset @xmath57 .
next , we will define the prediction models that we use as first level-0 learners .
it is worth to mention that the methodology developed in this work can work with any number @xmath56 of different base learners .
for the learned level-0 learner of constant regression @xmath83 , or naive regression , we just consider the simple average value of the dependent variable in the training set as the predictor , i.e. , @xmath84 .
second , when we formulate a learned linear regression model @xmath85 on a training dataset , we have in total @xmath86 terms , corresponding to @xmath87 single terms plus the constant term , i.e. , @xmath88 , , where @xmath73 and @xmath89 are both vectors .
third , we formulate a learned quadratic polynomial regression model @xmath90 on a training dataset , where we include @xmath87 single terms corresponding to @xmath87 independent variables , @xmath87 quadratic terms corresponding to the squared independent variables , as well as @xmath91 cross terms plus one term for the constant term . let @xmath92 be the matrix representation of the quadratic polynomial , where @xmath73 and @xmath93 are both vectors , and @xmath94 is a symmetric matrix .
we are going to introduce a fourth learned level-0 learner trained on a training dataset known as radial basis function networks @xmath95 . in next section
, we will denote @xmath96 same as @xmath97 , i.e. @xmath98 . in the same way we define @xmath99 as @xmath100 , @xmath101 as @xmath102 , and @xmath103 as @xmath104 .
the radial basis function approach is a very active research discipline particularly in neural networks and penalized splines @xcite .
we will be considering radial basis function networks as one type of radial basis model , which can be considered as an statistical model for non - parametric regression , to be introduced shortly .
_ radial basis function networks _ given the set of @xmath1 training points @xmath105 where @xmath106 and @xmath107 . in order to obtain a radial basis function approximation of the set @xmath108 , a model @xmath7 of the form @xmath109 where @xmath110 and , @xmath111 is the radial distance between @xmath112 and @xmath51 , is fitted to the data by minimizing the error function in equation ( 1 ) . in the above model
the points @xmath51 are considered as the centers of the radial basis functions . in real applications to localize the radial basis functions on the input vector @xmath51
can be very expensive , so usually a reduced number of centers @xmath113 where @xmath114 are estimated first . in our work
an optimal number of centers is calculated by splitting the historical data in training and validation parts .
this will give us the number of radial basis functions of the model .
other parameter that is estimated in this part is the spread parameter @xmath115 which play a role in the gaussian and multi - quadratic models .
we use @xmath45-means clustering for estimation of centers and @xmath115 s .
once we estimate these set of parameters , we will estimate the weights @xmath116 s assuming a fixed number of radial basis functions , and a constant spread for each function . in this way
the estimation of weights is linear and can be estimated by standard least squares optimization of the error function as in equation ( 1 ) .
this is an advantage of this kind of models @xcite where we are avoiding to optimize together all these parameters , which would require a non - linear optimization of the weights where the error function is non - convex , and with many local minima @xcite .
another advantage to do it this way is that the interpretation of the model , from the statistical point of view , is very simple as a linear combination of non - linear transformation of the input variables .
nevertheless , this two parameter learning approach is a very active research subject .
there are six basis functions , which are recognized as having useful properties for @xmath117 models @xcite : inverse multi - quadratic , gaussian , multi - quadratic , thin plate spline , cubic and linear .
for example , the gaussian is defined as @xmath118 for @xmath119 and @xmath120 , and is considered a local function in the sense that @xmath121 as @xmath122 .
the same property is exhibited by inverse multi - quadratic function .
on the other hand , the last four are global radial basis functions , which all have the property @xmath123 as @xmath122 , e.g. this is valid for thin plate spline , which is defined as @xmath124
. for formulations of the other functions please see @xcite . among the group of local radial basis functions , we are interested on gaussian radial basis function as the main replacement of the quadratic regression model because , we think , it will provide a better non - linear behaviour than the quadratic regression model .
we introduce and define in section 3.2.1 , the standard stacking approaches proposed in this work based on different combinations of level-0 learners .
the basic methodology is described next : 1 .
we estimate first learned constant regression @xmath96 , linear regression @xmath99 , and quadratic regression @xmath101 on the training sets of a @xmath43-fold @xmath34 .
the first one is just the average value taken on the dependent variable values .
the last two are calculated from ridge regression by minimizing the penalized sum of squares using tikhonov regularization .
it should be mention that the we used a fixed value for the regularization parameter in the ridge regression estimate , we do not optimize such regularization parameter using cross - validation .
we then compute the predictions of each based model on the validation folds of a @xmath43-fold @xmath34 .
+ then , we define a level-1 @xmath71 , or learning ensemble @xmath125 as a linear combination of base learners in set @xmath126 : @xmath127 where we estimate the weight vector @xmath17 by @xmath43-fold cross - validation @xmath40 based on a stacked generalization procedure to combine the individual model predictions + for the determination of the weights , we followed what we found in literature , which can be made by optimizing last equation in section 2.3 , which we reproduce here for convenience .
optimal weight learning from @xmath34 .
* given a dataset @xmath58 do @xmath43-fold @xmath34 .
+ @xmath128 + @xmath129 + in the above equation , the @xmath68 @xmath61 consists of @xmath69 where + @xmath70 , @xmath130 , so @xmath131 are predictions of level-0 learners on @xmath51 , for @xmath132 .
, we will use a level-1 generalizer , or ensemble learner for predictions of a `` new '' data @xmath73 on the testing dataset .
@xmath133 + where the optimal @xmath134 weight vectors of models @xmath135 are estimated based on @xmath57 the whole historical data , and the weight vector @xmath17 is estimated using equation ( 3 ) for a given @xmath34 .
we call the learning , in equation ( 3 ) and equation ( 4 ) , to determine the weight vector @xmath17 , one - step learning , or @xmath136 learning , and denote the level-1 generalizer @xmath137 in equation ( 4 ) as @xmath138 for the set of base models in @xmath139}. so , the approximation @xmath140 for a new @xmath73 based on optimal weight vector @xmath141 for a given @xmath34 is @xmath142 as a result of the fourth learned base learner @xmath143 , @xmath117 , we consider two additional standard approaches @xmath144 , and @xmath145 , such that for a prediction of a new observation @xmath112 based on corresponding optimal weights @xmath146 and @xmath147 we have @xmath148 @xmath149 so , the ensembles @xmath150 , and @xmath151 are linear combinations of base learners in the sets of @xmath152 } , and @xmath153 } , respectively . for the generation of the above ensemble learners we are keeping @xmath96 in each of possible combination of the the generated ensembles .
we have found empirical evidence that for a few databases the @xmath96 model has a higher weight than any of the other models , giving the corresponding ensembles better prediction error than any of the individual models .
therefore , all possible combinations of the remaining three level-0 learners taken two at a time makes a total of three level-1 learners : @xmath154 , @xmath150 and @xmath151 by keeping present @xmath96 in each ensemble .
the motivation behind the level-2 ensemble learning we are proposing is to see whether combining , or mixing a set of level-1 learners using stacked generalization , we can improve the prediction error in comparison to the level-1 learning , particularly against the best level-1 learner among the standard stacking approaches . in the next two sections ,
we propose , as our first extension to standard stacking , two different level-2 learning approaches based on stacked generalization : the first approach , presented in section 4.1 , is based on two - step optimization , and the other , which is presented in section 4.2 , we call it all - at - once optimization .
it is worth to mention that , the iterative nature of two - step learning will be advantageous over the all - at - once optimization to improve prediction error as we will discuss later at the end of this section .
we will introduce our scheme of level-2 learning through iterative optimization based on optimal weights . given the new predictions by the level-1 learners , @xmath155 , @xmath156 and @xmath157
, we can ask ourselves how to combine them ? we suggest to apply the same @xmath158fold @xmath34 procedure on this new set of level-1 learners , instead of the ensemble composed of @xmath159 , @xmath160 , @xmath161 and @xmath162 , to determine the corresponding set of weights @xmath163 , @xmath164 , and @xmath165 , for @xmath154 , @xmath150 and @xmath151 respectively .
the optimal weight vector @xmath166 is obtained through an optimization process , level-2 @xmath71 , resulting in corresponding set of optimal weights @xmath167 , @xmath168 and @xmath169 .
once we get the optimal weight vector @xmath166 , we can now combine the predictions of trained level-1 learners @xmath170 , @xmath171 and @xmath172 optimally as @xmath173 to get the approximation of level-2 learner .
it should be pointed out that if we expand , e.g. , the equation for @xmath174 according to @xmath175 @xmath176 @xmath177 we get the following expression for @xmath174 in terms of the models @xmath96 , @xmath99 , @xmath101 and @xmath103 : @xmath178 where @xmath179
@xmath180 @xmath181 @xmath182 analysing any @xmath183 s weight , e.g. @xmath184 corresponding to model @xmath99 , we can observe that it is additively combining the @xmath17 s weights of model @xmath99 , @xmath185 and @xmath186 , which are present in ensembles @xmath154 and @xmath150 from first step of learning , through the extra weights @xmath163 and @xmath164 that were gotten in the second step of learning for such ensembles .
therefore , the set of weight vectors @xmath17 and @xmath187 were obtained running two separated but iterative optimizations which were called first step and second step of learning respectively .
we will denote this two - step optimization using the @xmath187 s weights as follows : * given a dataset @xmath58 do @xmath43-fold @xmath34 .
+ @xmath188 + @xmath189 @xmath190 in the above equation , the @xmath191 - 2 data consists of @xmath69 where + @xmath70 , @xmath192 , and @xmath193 on @xmath51 are predictions of level-1 learners for @xmath132 .
we call this two - step learning , or @xmath194 learning , and denote the approximation , or level-2 ensemble learner based on weight vector @xmath187 for predictions of a `` new '' data @xmath112 on the testing dataset as @xmath195 where the optimal @xmath196 weight vectors of models @xmath197 are estimated using an optimization similar to equation ( 3 ) for a given @xmath34 , where we replace @xmath198 . on the other hand ,
the weight vector @xmath187 is estimated using equation ( 5 ) for a given @xmath34 .
we can as well run both optimizations at once where we optimize simultaneously both weight vectors @xmath17 and @xmath187 which now will be denoted as @xmath199 and @xmath200 respectively to emphasize that the optimized values are different from the iterative optimization .
therefore , given @xmath201 @xmath202 @xmath203 @xmath204 we can formulate such optimization via stacked generalization as follows : @xmath205 @xmath206 @xmath207 @xmath208 @xmath209 @xmath210 @xmath211 where @xmath212 , @xmath213 and @xmath214 are the set of weights for the set of level-0 learners @xmath215 in the learning ensemble @xmath216 .
@xmath217 , @xmath218 and @xmath219 are the set of weights for the set of level-0 learners @xmath220 in the learning ensemble @xmath221 .
@xmath222 , @xmath223 and @xmath224 are the set of weights of the level-0 learners @xmath225 in the learning ensemble @xmath226 . and @xmath227 , @xmath228 and @xmath229 are the set of weights for the level-1 learners @xmath154 , @xmath150 and @xmath151 in learning ensemble @xmath230 .
this optimization , as we already said , optimize all parameters at once . , and we call it @xmath231 learning , and denote the approximation , or level-2 ensemble learner based on weight vector @xmath183 for predictions of a `` new '' data @xmath112 on the testing dataset as @xmath232 where the optimal @xmath134 weight vectors of models @xmath233 are estimated based on @xmath57 the whole historical data
. there is one more question that we should ask ourselves at this point .
is the two - step learning better than all - at - once learning ?
this question could be answered in two ways .
first by noting that according to the equivalence that we just showed for both methodologies , in terms of the corresponding gamma weights , we would expect to get very close results in terms of the prediction error for any database considered .
second , we found empirical evidence by running both methodologies across the database considered where we actually got very close results in terms of their corresponding error performances .
however , the iterative nature of the two - step ensemble learning and its richness in structure in terms of of the pairwise correlation between each pair of level-1 ensemble predictions in the mixture , will be exploited and bring some insight about how to make two - step ensemble learning better than all - at - once optimization to improve error performance , as we show in detail in next section , particularly section 5.2 .
in stacked generalization the components of the ensemble , i.e. the level-0 learners , must be as diverse as possible , meaning that they provide predictions which are different from each other , otherwise if all of them provide the same predictions there will not be any improvements when we combine them . in section 5.1 , we present in detail our second extension to standard stacking approach .
as a first part of such extension , we propose to inject a diversity mechanism by perturbing the original @xmath34 partition to generate new @xmath34 partitions on which we can train new generated level-2 learners using stacked generalization , and two - step ensemble learning .
particularly , in section 5.1.1 , we introduce a combinatorial mechanism to generate all possible @xmath34 partitions forming an exhaustive search space , as well as a systematic way to generate @xmath34 partitions to shorten significantly the huge combinatorial search space given by the exhaustive search . in section 5.1.2 ,
we formulate two - step ensemble learning based on a new generated partition . in section 5.2 , we introduce , as a final part of the second extension , methodology we have developed for this work , that allows us to quantify the quality of a partition , and select one on which to make the final prediction . in section 5.2.1
, we exploit the structure of two - step level-2 ensemble learning to measure its diversity in terms of its base ensembles , or level-1 learners ; introducing the concept of computed level-1 ensemble correlation in their prediction errors
. this will provide us with relevant diversity information about the quality of the different @xmath34 partitions . in section 5.2.2
, we introduce a decision matrix to support the selection of a partition . in section 5.2.3
, we present a correlation - based criterion , and a partition ranking - based criterion to help us make a more informed decision by ranking such partitions . in section 5.2.4
, we introduce heuristic rule - based algorithm @xmath41 , that uses the above two criteria , to select the best @xmath34 partition that contains a minimum correlation rank , with the aim to improve the prediction accuracy of the level-2 ensemble corresponding to the original @xmath34 . in our initial @xmath43-fold @xmath34
was obtained by removing one of the @xmath43 subsets , which results in a set of partitions , that will be represented by @xmath234 . in order to generate different subsets of learning datasets
we now take out two subsets instead of only one fold out of @xmath43 .
the total number of possibly different partitions sets that can be generated this way out of a @xmath43-fold @xmath34 partition is prohibitively large , with a total of @xmath235 new partitions in addition to the original @xmath234 partition .
so , we will generate only @xmath236 new different @xmath34 partitions in an orderly way .
it will provide us with a systematic mechanism to explore a reduced set of alternatives , and to test the algorithms proposed in this paper in a computationally efficient way .
for some @xmath237 in @xmath238 , the partition set for a new systematic generated partition will be denoted as @xmath239 .
please see section 6.1 for details of the implementation of the new partitions .
given the partition set of corresponding generated partition @xmath239 , we have the following formulation of two - step ensemble learning : * given a dataset @xmath58 do @xmath43-fold @xmath239 .
+ @xmath240 + @xmath241 @xmath190 we denote the level-2 ensemble learner based on weight vector @xmath242 , for predictions of a `` new '' data @xmath112 on hold - out dataset , given @xmath239 as @xmath243 equation ( 7 ) and equation ( 8) , are equivalent to equation ( 5 ) and equation ( 6 ) respectively . in equation
( 8) we denote with an index @xmath237 the dependency on a given @xmath239 partition of level-1 learners @xmath244 , with @xmath132 , and the level-2 learner @xmath245 as well .
the reason of this is that , the @xmath246 weight vectors of level-1 learners present in equation ( 7 ) is estimated using an optimization similar to equation ( 3 ) , but in the context of a given @xmath247 partition , and the weight vector @xmath187 in equation ( 8) is estimated as well for a given @xmath239 by solving equation ( 7 ) . in section 6.1 , we explain in a little bit more detail the two - step ensemble learning . in section 5.2.1 , we will introduce some concepts to help us define a level-1 ensemble correlation - based criterion , which will provide relevant information about the quality of a cross - validation partition . in section 5.2.2 ,
we introduce a decision matrix which will facilitate the way we make the selection of an appropriate partition . in section 5.2.3
, we present a correlation - based criterion , and a partition ranking - based criterion that can be used to rank partitions . in section 5.2.4 ,
both criteria are employed by algorithm @xmath41 , which is a max - min rule - based systematic learning algorithm , to select best @xmath34 partition , that contains a minimum correlation rank .
next , we will introduce some notation that we will use in the description of the selection criteria in the next two sections . in the framework of two - step level-2 learning represented by equation ( 8) ,
let us look at the level-2 data which are input for the optimization process in equation ( 7 ) , i.e. , the actual values @xmath59 along with predictions @xmath248 , for @xmath249 .
the actual values @xmath59 along with the above predictions collected from each of the @xmath43 validation folds of @xmath247 are used to define the error matrix as @xmath250 where each element in @xmath251 is defined as @xmath252 @xmath253 and @xmath254 with @xmath255 from the error matrix @xmath251 , the sample variance - covariance matrix between columns of size @xmath256 : @xmath257 is calculated as well as its sample correlation matrix between columns @xmath258 . in our prediction problem matrix @xmath259 is symmetric , and the diagonal elements are equal to one , as it is the correlation between the predictions of level-1 learners @xmath260 and @xmath261 .
we are interested in the @xmath262 elements above the diagonal of such matrix , and we put them in a row - wise fashion into the following row vector @xmath263 : @xmath264 in the last equation we are representing correlation vector @xmath265 by an alternative vector @xmath266 containing @xmath267 criteria , or elements , where @xmath268 .
each criterion @xmath269 applied on an alternative @xmath266 and represented for element @xmath270 in vector @xmath266 with @xmath271 , measures the performance of alternative @xmath266 on criterion @xmath269 , and corresponds to an error correlation function between a pair of ensemble , or level-1 learner predictors , @xmath260 and @xmath261 with @xmath272 in vector @xmath263 .
so , when we talk about partition @xmath239 we will be referring to alternative @xmath266 , and vice - versa .
it should be emphasized that it is here where we exploit the structure of two - step level-2 ensemble algorithm as opposed to all - at - once level-2 learning , which lacks such structure by construction . at the end of the first step of two - step level-2 learning , or level-1 learning , and for a given @xmath43-fold @xmath247 partition
, we can compute the error correlations @xmath273 , before the optimization in equation ( 7 ) , between the prediction errors of level-1 learner approximations @xmath274 and @xmath275 where @xmath276 , and @xmath46 with @xmath277 are validation folds .
this correlation information provides relevant information about the quality of the @xmath43-fold @xmath247 partition as we can see in vector @xmath263 from equation ( 9 ) , and it is the basis for algorithm @xmath41 that we present in the next sections . in this section
we build a decision matrix which will facilitate the way we make the selection of an appropriate level-2 learner .
the matrix will have as rows alternatives corresponding to the different generated @xmath34 partitions . in the general case ,
if we have @xmath278 different alternatives , @xmath266 with @xmath279 , which are associated to partitions @xmath239 with @xmath280 , and @xmath281 criteria @xmath282 , @xmath283 we can build the following decision matrix @xmath284 : @xmath285 where @xmath270 is the performance of alternative @xmath266 on criteria @xmath269 , for @xmath271 .
we can determine the minimum correlation set @xmath286 as the line vector of minimums column - wise .
we will refer alternatively to the decision matrix @xmath284 as the set of alternatives , i.e. , we will also use the notation : @xmath287 . from now on ,
when using a set @xmath284 of alternatives we will be making reference to the systematic generation of alternatives , @xmath266 with @xmath288 , where @xmath113 , the number of new alternatives , is equal to the number of @xmath43 of folds by design of systematic learning .
, i.e. , @xmath236 .
it should be emphasized that each alternative @xmath266 in @xmath284 is related to a corresponding @xmath43-fold @xmath239 partition , as vector @xmath266 is equal to vector @xmath265 according to equation ( 9 ) . on the other hand , in equation ( 8) , level-2 learning , we make reference to a level-2 generalizer , or learner @xmath245 for a @xmath239 , for some @xmath237 in @xmath289 .
we use @xmath245 for prediction of a new instance @xmath73 . in section 5.2.3
, we present in detail two rule - based criteria used for selection of a @xmath247 partition .
these two rules are employed within algorithm @xmath41 in section 5.2.4 to choose one alternative @xmath266 for some @xmath237 in @xmath290 .
once we have such alternative we can use its corresponding level-2 learner @xmath245 , to make predictions of new instances @xmath73 .
one of such rules is a correlation based criterion between predictions of level-1 learners , while the other rule is based on a rank - based criterion of such correlations .
in @xcite the authors use an uncorrelation maximization algorithm , which is based on a correlation based criterion , to remove ensemble learners based on a threshold @xmath291 with the objective of determining the best subset of ensemble learners .
their algorithm is based on the principle of model diversity , where the correlation between the predictors should be as small as possible @xcite . in this section
we will introduce a computed level-1 ensemble correlation criterion , and a rank based criterion , not to prune an ensemble , but to handle additional , and diverse @xmath34 partitions , rank them and select one .
the selected partition has to contain a minimum correlation rank as we will explain in algorithm @xmath41 .
one of its advantages is that they do not depend on extra parameter values like @xmath292 .
the basis of the correlation criterion to select a particular @xmath247 partition , is to compute for a two - step level-2 ensemble , @xmath293 , or corresponding partition @xmath239 , the error correlation between its level-1 learner predictions @xmath294 at the end of level-1 learning step , for @xmath295 , represented by alternative @xmath266 in @xmath284 .
the correlations , measured at this extra level of smoothing in the predictions , gives a way to control the diversity of its base ensembles , or level-1 learner components , and therefore be able to quantify the quality of a partition using a correlation based criterion , and a correlation rank based criterion .
in next section we introduce algorithm @xmath41 , where we select the best alternative by grouping alternatives into different segments .
for the first segment of alternatives we use a median based selection rule , @xmath296 rule , and in case we are not able to do a selection due to potential outliers in such alternative , then we go to next segment where we use @xmath297 rule for such selection instead .
if we determine that there are potential outliers in such alternatives , then we go to next segment and apply @xmath297 rule again , until we eventually select one with no potential outliers , or if we ran out of alternatives then we keep the original alternative @xmath298 corresponding to original partition @xmath299 .
we will describe some of the inputs required for algorithm @xmath41 .
first , @xmath286 is the set of minimum correlations collected across all @xmath300-th criteria , @xmath269 in @xmath284 ; @xmath286 is then used for @xmath296 rule to make a selection , as we will describe in next section .
algorithm @xmath41 also requires , a set @xmath301 of rank vectors of alternatives , and a set @xmath302 of scores of alternatives , which we define below , and are used by the @xmath297 rule in algorithm @xmath41 to make a selection .
these inputs are generated before hand by the average ranking method , which we describe next briefly .
the average ranking method @xmath303 @xcite selects one criterion , @xmath269 of @xmath284 , which contains a set of alternative s performances : @xmath304 , for @xmath305 .
then , it computes a set of rankings on them to get @xmath306 , where @xmath307 , returns the rank of @xmath308 in a set of alternatives @xmath309 , @xmath310 , @xmath311 according to the @xmath269 criterion .
rank value @xmath312 means the best criterion value , and rank value @xmath278 means the worst criterion value .
it repeats this procedure for criterion @xmath269 , with @xmath313 .
the ranking for an alternative @xmath308 is then given by the rank vector @xmath314 once the rank vector @xmath315 is computed for each alternative @xmath308 in @xmath284 , its score is calculated by adding the ranks of @xmath308 for each criterion .
@xmath316 let @xmath301 be the set of rank vectors : @xmath317 and @xmath302 the set of alternative scores @xmath318 next , we give the specification of algorithm @xmath41 in terms of its inputs , and outputs : @xmath319 \leftarrow s(\textit{m+1},q,\boldsymbol{g},g^{min},\boldsymbol{t},\boldsymbol{ar } ) \ ] ] so , algorithm @xmath41 returns best alternative .
then , we can predict with corresponding best level-2 learner @xmath320 for a new instance @xmath73 : @xmath321 . in a high level description , the main objective of algorithm
@xmath41 is to consider alternatives in @xmath284 containing a minimum rank in @xmath301 , and select one of them by using one of two selection rules . in order to do
that we proceed at different segments , we first find all rank vectors of alternatives in @xmath301 with @xmath322 , or @xmath323 , in at least one of its coordinate elements . using a selection rule at @xmath324 , @xmath296 selection rule ,
an alternative is selected as long as such alternative is not disregarded when applying two rules used to detect potential outliers in it . otherwise , all alternatives with @xmath322 are eliminated from @xmath301 , and we then proceed to next segment , @xmath325 , by finding all alternatives with a minimum rank , @xmath326 , and put them in updated set @xmath301 .
a second selection rule , @xmath327 is used at this segment to select an alternative with minimum score as long as such alternative is not disregarded if outliers are not detected in it .
otherwise , the disregarded alternative is eliminated from set @xmath301 , and repeat the same steps at same @xmath328 ; applying @xmath327 rule on remaining alternatives in @xmath301 .
if we run out of elements in @xmath301 , then we go to next segment by increasing @xmath328 by 1 , finding all alternatives with next minimum rank , and applying @xmath327 rule until either an alternative is selected with no potential outliers at @xmath328 , or either we reach the maximum number of segments @xmath113 , or there is not any alternative left in @xmath301 . in any of these two last cases is true then we keep original partition for final prediction .
it is worth to mention that at @xmath323 we applied @xmath296 rule as opposed to had applied @xmath297 , because the former rule is more robust than the latter rule based on empirical evidence through the databases considered in this work .
in addition to that , we believe that at @xmath324 we have an extreme case where the rank vectors have at least one of its element with minimum rank , @xmath322 across all rank vector available , and the @xmath296 rule , takes into account extra information , where it selects from @xmath284 a compromised alternative containing the median of the minimum correlation values , which corresponds to one of the rank vectors with @xmath322 . on the other hand , at the upper segments , @xmath329 , we do not have that extreme case any more , and there is not need to use the more specialized @xmath296 rule that takes into account correlation information explicitly .
next , we describe in more detail algorithm @xmath41 , focusing more in the selection rules , and two rules designed to help us detect an alternative with potential outliers .
first rule , checks if an alternative contains two rank values with a minimum value at @xmath328 , the second rule detects if such alternative has a maximum rank value equal to @xmath278 .
+ in the first part , labelled @xmath330 @xmath331 , lines 1 - 14 , where @xmath323 , the objective is to apply @xmath296 rule on @xmath284 in order to get an alternative @xmath332 , or center alternative , containing the median value in set @xmath286 , with no potential outliers in it .
we mark criterion @xmath333 , which contains such minimum median value , where @xmath334 . in line 7 and
line 8 , we apply @xmath335 and @xmath336 rules on @xmath337 to help us find minimum and maximum rank values in it , respectively .
those values will help us to determine whether such alternative has any potential outliers .
on one hand , @xmath335 rule finds a criterion @xmath338 in @xmath339 with minimum rank : @xmath340 . on the other hand ,
@xmath336 rule finds a criterion @xmath341 in @xmath337 with maximum rank : @xmath342 . in line 10 , we check if the center rank vector has two minimum rank values , i.e. , @xmath343 , or if its maximum rank value @xmath342 is equal to @xmath278 . if any or both of these conditions are true then we go to second part of the algorithm , as we have found an alternative with potential outliers
. otherwise , if both conditions are false then we select @xmath332 as @xmath344 , and go to line 36 , and return alternative @xmath344 , whose corresponding two - step level-2 ensemble , @xmath320 , will be used to make predictions of new instances @xmath73 .
+ in the second part , line 15 to line 35 , labelled @xmath345 @xmath346 @xmath331 , we first go to next segment by increasing @xmath328 by 1 , and collect all rank vectors from @xmath301 with @xmath326 into @xmath347 . the objective is to find a best alternative @xmath348 with minimum @xmath349 score , or corresponding best rank vector @xmath350 from @xmath347 , that has minimum rank , i.e. , @xmath351 , with no potential outliers in it .
so , we apply again the same @xmath335 and @xmath336 rules , but now on @xmath350 . in line 26 , we check whether the best rank vector has two minimum rank values at @xmath328 , or if its maximum rank value is equal to @xmath278 . if any or both of these conditions are true then we go on in second part of the algorithm .
otherwise , if both conditions are false then we select @xmath332 as @xmath344 , and go to line 36 , and return alternative @xmath344 , whose corresponding two - step level-2 ensemble , @xmath320 , will be used to make predictions of new instances @xmath73 .
@xmath352 , } \textit{q } \text { } \text{[number of criteria ] } $ ] @xmath353 + @xmath354 @xmath355 [ find a minimum correlation alternative based on median rule ] [ find criterion @xmath338 with minimum rank , @xmath356 , in center rank vector excluding rank of @xmath357 from it ] [ find @xmath341 with maximum rank ] [ unmark alternative that has potential outliers ] [ update @xmath284 and @xmath301 eliminating alternatives with @xmath358 + + @xmath359 @xmath360 @xmath361 [ find best rank vector ] .
[ find criterion @xmath338 with minimum rank , @xmath362 , in best rank vector excluding rank of @xmath363 ] [ find @xmath341 with maximum rank ] [ unmark alternative that has potential outliers ] @xmath359 [ return best alternative , whose level-2 learner @xmath320 will be used for prediction ] +
in our initial @xmath43-fold @xmath34 we partitioned the original historical dataset @xmath57 in the following disjoint sub - samples @xmath46 with @xmath62 .
then , we built the learning or training subsets @xmath48 by where @xmath364 , and we denoted @xmath365 as the validation folds . the above results in a set of partitions @xmath366 , where @xmath48 are the training subsets , and @xmath46 the validation folds , with @xmath277 .
making an abuse of notation , such set will be represented as : @xmath367 in order to generate different subsets of learning datasets we now take out two folds instead of only one fold .
the total number of possibly different partitions sets that can be generated this way out of a @xmath43-fold @xmath34 partition is prohibitively large , with a total of @xmath235 new partitions in addition to the original @xmath234 partition .
the method proposed next generates only @xmath236 new different @xmath34 partitions in an orderly way
. it will provide us with a systematic mechanism to explore a reduced set of alternatives , and to test the algorithms proposed in this paper in a computationally efficient way .
it should be noted that for the systematic procedure the number of partitions @xmath113 coincides with @xmath43 , the number of folds , i.e. , @xmath236 .
for some @xmath237 in @xmath238 , and e.g. , @xmath368 , the partition set for a new systematic generated @xmath239 partition is @xmath369 where @xmath370 are the training subsets , and @xmath46 the validation folds , for @xmath371 .
let @xmath372 denote @xmath64-th level-1 learner estimated on training subset @xmath370
. given the partition @xmath239 , we have the formulation of two - step ensemble learning given by equation ( 7 ) .
we reproduce below equation ( 8) for convenience , which represents the level-2 ensemble learner based on weight vector @xmath242 , for predictions of a `` new '' data @xmath112 on hold - out dataset , given @xmath239 as @xmath243 in order to summarize the method of two - step level-2 learning from top to bottom , let us emphasized that the weight vector @xmath373 for @xmath245 or , level-2 learner , present in equation ( 8) is estimated based on training datasets of partition @xmath239 using @xmath191 - 2 data as input to equation ( 7 ) : predictions of level-1 learners @xmath374 along with actual values @xmath59 , where @xmath276 , and @xmath46 for @xmath277 are validation folds .. the optimal weight vector @xmath196 , for @xmath132 , corresponding to a parameter for each level-1 learner approximation , @xmath274 with @xmath132 in equation ( 7 ) , was estimated previously by solving an optimization problem similar to equation ( 3 ) , but in the context of a @xmath43-fold @xmath239 partition , using level-1 data as input to such optimization : predictions of level-0 learners @xmath375 , along with actual values @xmath59 , where @xmath276 , and @xmath46 for @xmath277 are validation folds .. for @xmath376 the index @xmath377 takes values in set @xmath378 corresponding to base learners in @xmath154
. on the other hand , for @xmath379 , index @xmath380 takes values across set @xmath381 corresponding to base learners in @xmath150 , and for @xmath382 , index @xmath383 takes values in set @xmath384 corresponding to base learners in @xmath151 . in order to evaluate empirically max - min systematic two - step level-2 ensemble learning , @xmath385
, we consider four simple base models already introduced in section 3.1 : constant regression , @xmath96 , linear regression , @xmath99 , quadratic regression , @xmath101 and radial basis networks , @xmath103 , and take the best model in terms of error performance for each database to build an oracle , which we called @xmath386 .
we also consider three standard stacking approaches , or level-1 learners , @xmath154 , a linear combination of base learners in set @xmath387 , @xmath150 , a linear combination of base learners in set @xmath388 , and @xmath151 , a linear combination of base learners in @xmath389 , and compare @xmath386 versus the best performer , @xmath390 .
we are also be comparing @xmath385 versus @xmath390 , and against @xmath386 .
the outputs , or predictions of our base models @xmath96 , @xmath99 , @xmath101 , and @xmath103 are represented in the next tables , and figures , by the acronyms @xmath97 , @xmath391 , @xmath392 and @xmath117 respectively .
these acronyms represent common simple models , which were already introduced in section 3.1 .
so , @xmath154 is a combination of @xmath393 , @xmath150 is a combination of @xmath394 , and @xmath151 is a combination of @xmath395 .
in addition to that , we also compared our results versus @xmath396 package @xcite implemented in @xmath397 , a statistical programming software tool @xcite .
we also use two regression tree ensemble algorithms which are provided in @xmath398 @xcite @xcite , a data mining software tool .
we use the @xmath399 @xcite interface to call these two regression tree algorithms from @xmath397 .
one is known as @xmath400 and the other @xmath401@xmath402@xmath400 @xcite , @xcite .
we use the @xmath398 default parameters for @xmath400 , and @xmath401@xmath402@xmath400 , and the @xmath403 default parameters for the generation of their corresponding predictions .
the empirical study uses thirty databases which are popular in the literature , most of them from the uci repository of machine learning databases and domain theories @xcite . for each data
set , we split it randomly in two parts , where the first part has @xmath404 percent of the observations used for training , and the remaining @xmath405 percent of the observations is used for testing .
we conduct a @xmath43-fold @xmath34 partition with @xmath368 on the training part to fit the models in our approach , and measure the performance on the testing part . for the base models of the oracle @xmath386 , and the three state - of - the art ensemble algorithms against we are comparing our results ,
we pass the whole training part to each method to fit its parameters , and then we measure their performance on the test part for comparison purposes .
then , a non - parametric two - sided wilcoxon signed rank test @xcite is applied , as suggested in @xcite , to compare two approaches over all data sets with a significance level of 0.05 .
the null hypothesis is that the average difference between both approaches is zero .
the @xmath406-value is compared against the @xmath407 confidence level to determine the null hypothesis must be accepted or rejected .
alternatively , we compute the @xmath408 confidence interval around a point estimate to decide whether the zero average difference falls within the interval to accept or reject the null hypothesis .
we present in section 7.1 the results regarding standard stacking level-1 learning , and in section 7.2 the second extension to standard stacking approach : systematic level-2 learning . while in section 7.3
, we compare systematic level-2 ensemble learning against three of the state of the art ensemble learning algorithms for regression . in this section , we present different results , among them the comparison of each of the level-1 learners versus an oracle of databases composed of the best level-0 learner for each database .
in addition to that , we select the best performer among the level-1 learners versus the database oracle . in table 1 , please see appendix a for all table results , we present the @xmath409 for the thirty databases considered in this work .
if we look at table 1 , in the first column we have the database names .
next , we have split such table in two parts , in the first half we have the @xmath410 for the level-0 learners : @xmath411 , @xmath412 , @xmath413 and @xmath117 . on the other hand , in the second half of the table
, we have the @xmath410 for the level-1 ensemble @xmath154 , linear combination of base learners in @xmath393 , followed by the other two level-1 ensembles @xmath150 , linear combination of base learners in @xmath394 , and @xmath151 , linear combination of base learners in @xmath395 .
the ensemble weights are determined using the optimal way as we described in section 3.2 .
finally , in the last two columns are the number of records , and the total number of variables included in each database , independent variables plus dependent variable .
it should be mentioned that we will build a database oracle based on the best level-0 model performer in each database , which we will compare later , in terms of error performance , against the best level-1 learner , as we will describe shortly .
we will first focus our analysis on the level-0 learners , first half of the table .
we can observe from table 1 that the prediction performance of the level-0 learner @xmath391 , as measured by the @xmath409 , is the best , indicated as underlined bold error , in ten out of thirty databases among the four models , @xmath97 , @xmath391 , @xmath392 and @xmath117 .
the model @xmath413 is the best in thirteen out of the thirty databases . while the model @xmath117 is the best in seven of such databases . here ,
we build @xmath386 the oracle which consists of the best level-0 model performer in terms of the @xmath409 for each database .
the average error performances across all databases , in terms of @xmath410 , are @xmath414 , @xmath415 , @xmath416 , @xmath417 for @xmath97 , @xmath391 , @xmath392 and @xmath117 level-0 learners respectively . among the level-0 learners
the @xmath391 model has the best error performance with a @xmath415 @xmath409 across all databases . in the second part of table 1 ,
the level-1 learner @xmath154 is the best , among the level-1 learners , in ten out of thirty databases , while @xmath150 is the best in seven of those databases , and @xmath151 is the best in eleven databases .
in addition to that , there are two ties , where for @xmath418 and @xmath419 , the @xmath410 of @xmath154 are tied with the corresponding errors for @xmath151 and @xmath150 respectively . in relation to
the level-1 learners their average error performances across all databases , are @xmath420 , @xmath421 , and @xmath422 for @xmath154 , @xmath150 and @xmath151 respectively .
then , the best level-0 learner performer is @xmath154 with a @xmath409 of @xmath420 .
next , we will compare the performance results between the database oracle and each of the level-1 learners to determine which level-1 learner performs better in average versus the oracle . in fig . 1
, we visualize the difference of the @xmath409 s between the oracle , @xmath386 , or best performer among the models @xmath97 , @xmath391 , @xmath392 and @xmath117 on the testing part , and the ensemble @xmath154 for each of the thirty databases considered .
the differences are sorted by its absolute values . a positive difference , right part of the figure , means that accuracy is in favour of @xmath154 , while a negative difference means that the accuracy is against it .
we will use this convention for all figures where we present the differences in performance between models or ensembles . by looking at the positive values in fig . 1
, we see that the ensemble @xmath154 has better error performance than the oracle in the following databases : @xmath423 , @xmath424 , @xmath425 , @xmath426 , @xmath427 , @xmath428 , @xmath429 , @xmath430 , @xmath431 , @xmath432 , @xmath433 , @xmath434 and @xmath435 .
it is worth to mention that the best performer in each database is not know in advance , as we are using the hold - out sample as testing dataset .
the result of two - sided non - parametric wilcoxon signed - rank test to evaluate the performance of the oracle @xmath386 , and the ensemble @xmath154 gives a @xmath406-value of @xmath436 .
we reject @xmath437 , the null hypothesis that there was not significant difference in accuracy , in terms of @xmath409 , between the database oracle @xmath386 , and the ensemble @xmath154 , i.e. @xmath437 stated that the average difference between their corresponding @xmath409s is zero .
the @xmath408 confidence interval is @xmath438 , with a point estimate of the average difference is @xmath439 , which lead us to conclude a better performance of @xmath386 vs @xmath154 as the zero average difference is to the right of such confidence interval .
1 * normalized mean square error difference + results of \{@xmath386 } vs \{@xmath154 } on the other hand , for the ensembles @xmath150 and @xmath151 we do not present the result figures of each of them versus the oracle @xmath386 , or best performer among the models @xmath97 , @xmath391 , @xmath392 and @xmath117 on the testing part across the thirty databases considered .
however , the result of two - sided non - parametric wilcoxon signed - rank tests to evaluate the performance of the oracle @xmath386 , and the ensemble @xmath150 gives a @xmath406-value of @xmath440 .
the @xmath408 confidence interval is @xmath441 with a point estimate of the average difference of @xmath442 .
we reject the null hypothesis ho , and therefore @xmath386 outperforms @xmath150 in prediction accuracy .
the result of two - sided non - parametric wilcoxon signed - rank tests to evaluate the performance of the oracle @xmath386 , and the ensemble @xmath151 gives a @xmath406-value of @xmath443 .
the @xmath408 confidence interval is @xmath444 , with a point estimate of the average difference of @xmath445 .
we reject ho , and @xmath386 outperforms @xmath151 in prediction accuracy , as the average difference between both models being different than zero is significant . on the other hand , the level-1 learner which performed better , in terms of @xmath406-value , against the database oracle
@xmath386 is @xmath154 , so we select it as the best level-1 learner , and it will be denoted as @xmath390 from now on . in section 7.2.1 , we first present the results of applying the algorithm @xmath41 from section 5.2.4 to select the best max - min rule - based level-2 learner , or corresponding partition with the aim of improving the predictions from the original level-2 learner . in section 7.2.2 ,
we compare the performance results of this best level-2 learner against the best level-1 learner , then in section 7.2.3 , we compare the performance of best level-2 learner against the oracle of best models . in order to improve the two - step level-2 learning ,
we have introduced the methodology of systematic two - step level-2 learning for the mixing of level-1 learners
. we will illustrate algorithm @xmath41 from section 5.2.4 by using it to select the @xmath34 partition , and applying it on each of the thirty databases considered .
we will present the corresponding @xmath409 s of systematic two - step level-2 learning @xmath446 , for @xmath447 , with @xmath368 . in table 2
, we show the corresponding results , where in the first column is as usual the database name , followed by the next six columns corresponding to the normalized mean square errors @xmath448 of the ensembles @xmath449 , @xmath450 , @xmath451 , @xmath452 , @xmath453 and @xmath454 .
the next column @xmath331 refers to the selection rule that was employed to select the @xmath34 partition , among @xmath234 , @xmath455 , @xmath456 , @xmath457 , @xmath458 and @xmath459 , on which we are going to make the final prediction , partition @xmath460 . for such selection
we used the two max - min rule - based based criteria described in algorithm @xmath41 .
the selected @xmath460 is indicated as underlined error in table 2 for each of the databases considered .
if we look again at table 2 , where we have the original two - step level-2 learner , @xmath461 , versus systematic two - step level-2 learners , @xmath462 , we see that we have improved , indicated as bold errors , in fifteen databases : @xmath463 , @xmath464 , @xmath465 , @xmath466 , @xmath467 , @xmath468 , @xmath430 , @xmath469 .
@xmath470 , _ autompg _ , @xmath431 , @xmath424 , @xmath432 , @xmath434 and @xmath419 .
the corresponding selected best max - min level-2 learners for such databases by the algorithm @xmath41 are @xmath471 , @xmath472 , @xmath471,@xmath473 , @xmath473 , @xmath472 , @xmath474 , @xmath472 , @xmath472 , @xmath473 , @xmath473 , @xmath472 , @xmath473 , @xmath475 and @xmath472 , respectively . on the other hand , for nine databases the algorithm @xmath41 selected @xmath174 , which is exactly the original level-2 learner , @xmath174 . in the remaining six databases we lost a bit of accuracy where , e.g. , in @xmath476 , the algorithm selected max - min level-2 learner @xmath477 with an error of @xmath478 , the other five databases are @xmath479 , @xmath480 , @xmath481 , @xmath482 and @xmath483 . in table 3 , we present the @xmath409 s of @xmath484 corresponding to the best max - min level-2 learners selected by algorithm @xmath41 , which are represented in table 2 as underlined errors .
now , we compare them against the corresponding @xmath409 s of @xmath485 , the best level-1 learner . in order to mark in table 3 the ensembles with best error performance between the best level-1 learning and systematic level-2 learning for each database , we highlight theirs corresponding @xmath409 s in bold . as we can see from table 3 ,
the @xmath409 s of @xmath486 are the best , indicated as bold errors under its corresponding column , against the corresponding @xmath409 of @xmath390 , in nineteen out of thirty databases .
the best level-1 learner @xmath390 is the best in eight databases . while for three databases the error performance is the same between them . in fig .
2 , we are visualizing the @xmath409 s differences of max - min systematic level-2 learning , and level-1 learning for the thirty databases considered .
we can clearly see that the best max - min level-2 learner @xmath320 is better than the best level-1 learner @xmath390 in the following nineteen databases : @xmath487 , @xmath483 , @xmath479 , @xmath430 , @xmath469 , @xmath432 , @xmath464 , @xmath434 , @xmath488 , @xmath470 , @xmath463 , @xmath481 , @xmath480 , @xmath482 , @xmath431 , @xmath424 , @xmath465 , @xmath435 , and _
solarflares_. the result of @xmath408 two - sided non - parametric wilcoxon signed - rank test to evaluate the performance of the best level-1 ensemble , @xmath390 , and the ensemble @xmath320 gives a test statistics @xmath489 with @xmath406-value of @xmath490 .
so , we do not reject ho .
the @xmath408 confidence interval is @xmath491 , and the point estimate of the average difference is @xmath492 .
however , the @xmath493 two - sided non - parametric wilcoxon signed - rank test is significant to reject @xmath437 with a @xmath406-value of @xmath490 .
the @xmath493 confidence interval is ( 0.00083 , 0.020 ) with a point estimate of @xmath492 , where the zero average difference corresponding to the null hypothesis is outside and to the left of such interval .
therefore , the ensemble @xmath320 outperforms @xmath390 at a @xmath493 significance level .
2 * normalized mean square error difference + results of \{@xmath390 } vs \{@xmath486 } based on the above results , table 3 and fig .
2 , we see that max - min systematic level-2 learning with @xmath320 is a step in the right direction being in average more accurate and robust than the best level-1 learning @xmath390 .
therefore , we are achieving our goal of improving the performance error of best level-1 ensemble , @xmath390 , through the two - step level-2 learning and systematization proposed in this work . in fig .
3 , we present the @xmath409 s difference between the best performer , or database oracle , among the models @xmath97 , @xmath391 , @xmath392 and @xmath117 on the testing part , and the best max - min level-2 learner @xmath320 for each of the thirty databases considered . by looking at the positive value part of such figure , we see that @xmath320 has better error performance than the database oracle , @xmath386 , in eleven out of thirty databases : @xmath424 , @xmath494 , @xmath432 , @xmath466 , @xmath434 , @xmath482 , @xmath429 , @xmath431 , @xmath464 , @xmath435 and @xmath427 . the result of @xmath408 two - sided non - parametric wilcoxon signed - rank test to evaluate the performance of the oracle @xmath495 , and the ensemble @xmath320 gives a @xmath406-value of @xmath496 .
so , we do not reject @xmath437 , as there is not significant accuracy difference between the oracle @xmath386 , and the two - step systematic ensemble learning @xmath320 .
the @xmath408 confidence interval is @xmath497 , and the point estimate of the average difference is @xmath498 . *
3 * normalized mean square error difference + results of \{@xmath386 } vs \{@xmath486 } for the three state - of - the art ensemble algorithms , @xmath396 , @xmath400 , and @xmath499 , we pass to each of them the same training and test datasets we used in our methods , to fit its parameters , and to measure their performance for comparison purposes versus our results in this paper . the above three methods are well regarded state - of - the - art ensemble learning algorithms for regression problems .
the two - step level-2 systematic ensemble learning presented in this paper constitutes a new state - of - the - art ensemble learning algorithm , as we have tested it against each of the three ensemble learning algorithms just mentioned , and we have gotten excellent results .
our ensemble learning algorithm performed better than @xmath396 and @xmath500 in terms of error performance , and it is as good as @xmath501 . in this section we present results that support our above claims .
the @xmath396 package @xcite , @xmath502 , is available in the @xmath397 statistical software @xcite , @xmath503 .
@xmath396 solves the penalized residual sum of square , that can perform lasso penalized regression , ridge regression , and the elastic net .
we use the version 3.01 of @xmath397 software , and version 1.9 - 3 of @xmath396 package to get the results presented in this paper .
the lasso penalty encourages sparsity , so this penalty performs a sort of variable selection as some of the coefficients are shrunken to zero , while the ridge penalty encourages highly correlated variables to have similar coefficients .
the elastic net is a linear combination of the two penalties through an @xmath136 parameter , with @xmath504 .
note that elastic net becomes the lasso when @xmath505 , and the ridge regression when @xmath506 .
it has also a tuning parameter @xmath507 , which is a complexity parameter . in our experiments
, we used the default parameter for @xmath136 , which is @xmath505 . for the @xmath508 parameter we estimated it using cross - validation with its default parameter , and we take the minimum of @xmath508 values as an estimate . in fig .
4 , we present the @xmath409 s difference on the testing part between the @xmath396 method , and the best max - min level-2 learner @xmath320 for each of the thirty databases considered . * fig .
4 * normalized mean square error difference + results of \{@xmath403 } vs \{@xmath486 } the result of @xmath408 two - sided non - parametric wilcoxon signed - rank test to evaluate the performance of @xmath396 , and the ensemble @xmath509 gives a @xmath406-value of @xmath510 .
so , we reject @xmath437 in favour of @xmath320 , as there is a significant accuracy difference between @xmath396 , and the two - step systematic ensemble learning @xmath320 .
the @xmath408 confidence interval is @xmath511 , and the point estimate of the average difference is @xmath512 .
we use a regression tree algorithm @xmath500 @xcite , which is based on an earlier @xmath513 regression tree @xcite .
@xmath400 algorithm is available in @xmath398 data mining software @xcite @xcite , @xmath514 .
although we install @xmath398 version 3.6.1 , we will only need to interact with it from @xmath397 .
we use the @xmath399 @xcite interface version 0.4 - 13 to call this regression tree algorithm from @xmath397 .
we use the @xmath399 default parameters for @xmath500 for the generation of their corresponding predictions .
@xmath500 is a regression tree algorithm , which produces a regression tree such that each leaf node consists of a linear model for combining the numerical features .
@xmath500 uses a tree pruning mechanism , which is a sort of variable selection , and @xmath500 can perform non - linear regression with the partitions provided by the internal nodes and is thus more powerful than linear regression . in fig . 5 , we present the @xmath409 s difference on the testing part between @xmath400 method , and the best max - min level-2 learner @xmath509 for each of the thirty databases considered . * fig . 5
* normalized mean square error difference + results of \{@xmath400 } vs \{@xmath486 } the result of @xmath408 two - sided non - parametric wilcoxon signed - rank test to evaluate the performance of @xmath500 , and the ensemble @xmath509 gives a @xmath406-value of @xmath515 .
so , we reject @xmath437 in favour of @xmath509 , as there is significant accuracy difference between @xmath400 , and the two - step systematic ensemble learning @xmath509 .
the @xmath408 confidence interval is @xmath516 , and the point estimate of the average difference is @xmath517 .
in this part , we consider multiple @xmath500 trees combined by the popular bootstrap aggregation ( bagging ) method ( breiman 1996 ) , which is available in @xmath398 , and is called @xmath518 @xcite .
we use again the @xmath399 @xcite interface to call this regression tree algorithm from @xmath397 .
we use the @xmath399 default parameters for @xmath518 for the generation of their corresponding predictions . in fig .
6 , we present the @xmath409 s difference on the testing part between @xmath499 method , and the best max - min level-2 learner @xmath509 for each of the thirty databases considered .
the result of @xmath408 two - sided non - parametric wilcoxon signed - rank test to evaluate the performance of @xmath400 , and the ensemble @xmath509 gives a @xmath406-value of @xmath519 .
so , we do not reject @xmath437 , as there is not significant accuracy difference between @xmath518 , and the two - step systematic ensemble learning @xmath509 .
the @xmath408 confidence interval is @xmath520 , and the point estimate of the average difference is @xmath521 . *
6 * normalized mean square error difference + results of \{@xmath518 } vs \{@xmath486 } it should be emphasized that the iterative nature of two - step level-2 learning and its richness in structure in terms of correlation between each pair of ensembles was exploited in section 5.1 , through the generation of @xmath34 partitions , and in section 5.2 , to select a partition on which to make the final prediction .
a difference to what is developed for @xmath403 @xcite , and the default parameter @xmath522 used in this work , where the diversity comes by using lasso penalized regression as variable selection mechanism , @xmath400 @xcite whose diversity comes by pruning a regression tree , and @xmath401@xmath402@xmath400 @xcite , which in addition to pruning a regression tree , creates more diversity by generating different training sub - samples through bagging ; the diversity in our approach , which uses combinations of standard stacking approaches , comes through the generation of different partitions on which such ensembles are trained .
in addition to that , our approach controls the diversity of the different level-2 ensembles generated , or corresponding partitions , through a computed correlation criterion built on top of two - step ensemble learning , which provides an extra level of smoothing for the base ensembles in each partition .
we have empirically tested three standard stacking approaches , or level-1 ensembles , each composed of a combination of simple based models , for heterogeneous stacking for regression .
we have shown that the best of such ensembles performed not as good as the best base model from the ensemble by cross - validation .
we have proposed two extensions to standard stacking approach .
the first is to create an ensemble composed of different combinations of such standard approaches , through a two - step level-2 ensemble learning .
we mentioned that the richness of structure in such two - step level-2 ensemble learning provided the basis to compute the base ensemble pairwise correlations , which will help us on improving the prediction accuracy in the second extension .
the second extension was built to systematically generate different partitions from the current partition , and correspondingly two - step level-2 ensembles ; that along a partition correlation - based criterion , or a partition ranking - based criterion within an heuristic algorithm allowed us to select the best 2-level ensemble , and show that it performed better than the best of the standard stacking approaches , and is as good as the best of the base models from the standard stacking approaches by cross - validation .
we also compared our results of best systematic level-2 ensemble learning versus three state of the art ensemble learning algorithms for regression , and found that our results performed better than two of those ensemble methods for regression , and it is as good as the third of such algorithms in terms of error performance .
as a future research to do , it is worth to mention , that the systematic way to generate @xmath34 partitions introduced in section 5.1.1 will allow us to rank them , and reduce the size of the original @xmath34 partition by eliminating the least relevant fold , and apply iteratively max - min rule - based algorithm @xmath41 .
we are currently investigating different stopping criterion that could prevent us from iterating the algorithm one more time , if after the current iteration it could detect that we may not get better prediction accuracy going into the next iteration after pruning the current partition .
* table 1 * normalized mean square errors + level-0 learners : \{@xmath97 } , \{@xmath391},\{@xmath392 } , \{@xmath117 } and level-1 learners : \{@xmath154 } , \{@xmath150 } , \{@xmath151 } * table 2 * normalized mean square errors + systematic generation of training sub - samples + first iteration of algorithm + results for \{@xmath461 } , \{@xmath473 } , \{@xmath475 } , \{@xmath472 } , \{@xmath474 } and \{@xmath471 } peter j bentley and jonathan p wakefield . finding acceptable solutions in the pareto - optimal range using multiobjective genetic algorithms . in _ soft computing in engineering design and manufacturing _ , pages 231240 .
springer , 1998 .
r. caruana , a. niculescu - mizil , g. crew , and a. ksikes .
ensemble selection from libraries of models . in _ proceedings of the twenty - first international conference on machine learning _
, page 18 .
acm , 2004 .
jerome friedman , trevor hastie , and robert tibshirani .
regularization paths for generalized linear models via coordinate descent . _ journal of statistical software _
, 330 ( 1):0 122 , 2010 . url http://www.jstatsoft.org
/ v33/i01/. mp perrone and ln cooper . when networks disagree : ensemble methods for hybrid neural networks .
r. j. mammone ( ed . )
neural networks for speech and imageprocessing_. new york : chapman and hall , 1993 .
lean yu , kin keung lai , shouyang wang , and wei huang .
a bias - variance - complexity trade - off framework for complex system modeling . in _ computational science and its applications - iccsa 2006 _ , pages 518527 .
springer , 2006 . | the motivation of this work is to improve the performance of standard stacking approaches or ensembles , which are composed of simple , heterogeneous base models , through the integration of the generation and selection stages for regression problems .
we propose two extensions to the standard stacking approach . in the first extension
we combine a set of standard stacking approaches into an ensemble of ensembles using a two - step ensemble learning in the regression setting .
the second extension consists of two parts . in the initial part
a diversity mechanism is injected into the original training data set , systematically generating different training subsets or partitions , and corresponding ensembles of ensembles . in the final part after measuring the quality of the different partitions or ensembles , a max - min rule - based selection algorithm is used to select the most appropriate ensemble / partition on which to make the final prediction .
we show , based on experiments over a broad range of data sets , that the second extension performs better than the best of the standard stacking approaches , and is as good as the oracle of databases , which has the best base model selected by cross - validation for each data set .
in addition to that , the second extension performs better than two state - of - the - art ensemble methods for regression , and it is as good as a third state - of - the - art ensemble method .
* keywords : * ensemble learning , stacked generalization , systematic cross - validation , ensemble selection , regression , max - min based rules . |
we study a sample of 43 grbs selected for the high - quality of their time - resolved spectra obtained with the batse spectroscopy detectors ( sensitive in the energy range @xmath0 kev ) .
the time over which each spectrum was accumulated was varied so that the signal - to - noise ratio was greater than 15 ( in the hard x - ray energy band ) .
these data provide excellent temporal resolution : in many cases we obtain more than 10 spectra per burst with accumulation times as short as 256 ms .
we fitted each grb time - resolved spectrum with two models : ( 1 ) the band model [ 1 ] , and ( 2 ) the optically thin synchrotron shock model ( otssm ) @xcite . the ( purely phenomenological ) 4-parameter band model @xcite consists of two power - law components ( of spectral indexes @xmath1 and @xmath2 ) joined smoothly by an exponential roll - over near a break energy @xmath3 .
@xmath4 \left(\frac{e}{{\rm 100 \ , kev } } \right)^\beta \ ; \ ; \ ; { \rm for } \;\ ; e \geq \left ( \alpha - \beta \right ) e_{0 } \end{aligned}\ ] ] we used the ( three - parameter ) otssm of refs .
we performed an independent spectral fitting for the band and otssm models for each of the time - resolved spectra of all grbs of our sample . for each grb
we obtain 4 ( 3 ) best fit parameters as a function of time representing the complete spectral evolution .
we find grb spectral evolutions of two types : ( 1 ) a tracking behaviour " , with spectral parameters in approximate one - to - one correspondence with the changing energy flux , and ( 2 ) a hard - to - soft evolution " , with spectral parameters evolving independently of the energy flux ( see , e.g. , ref .
@xcite ) . fig.[gp-20-fig01 ] shows the distribution for _ all _ collected time - resolved spectra of the low - energy spectral index @xmath1 .
a few bursts show values @xmath5 2/3 typically during the initial - rising part of their most intense pulses .
the high energy spectral index @xmath2 is less constrained , and in some cases varies substantially over consecutive spectra within the same burst .
the @xmath2 distribution ( fig.[gp-20-fig02 ] left panel ) is peaked near 2 for the band model representation , and is broader for the otssm fits .
break energies @xmath3 are typically well below 500 kev .
interestingly , we find that the otssm provides a very good representation of time - resolved spectral data .
fig.[gp-20-fig02 ] ( right panel ) show the cumulative distribution of the reduced @xmath6 for the band and otssm models .
we studied 43 grbs from the batse spectral archive selected by their large signal - to - noise ratios .
we collected information for a total of 1046 spectra .
our results indicate that the otssm is quite successful in describing the majority of grb spectra .
fig.[gp-20-fig03 ] shows the spectral evolution of the remarkable grb 990123 demonstrating the validity of the otssm for very intense bursts .
however , violations of the simple otssm are apparent in about @xmath7 ( at @xmath8 level ) of our time resolved spectra . these violations
( typically with a low - energy index @xmath9 ) always occur at the beginning of major grb pulses ( as in fig.[gp-20-fig04 ] ) .
the otssm was derived @xcite for idealized plasma and hydrodynamic conditions that are most likely valid far from the central source .
several plasma and dynamic conditions ( probably involving emission sites close to a central object ) may produce the apparent suppression of soft photons at the beginning of some grb pulses . band , d. , matteson , j. , ford , l. , _ et .
al _ ; _ apj _ * 413 * , 281 - 292 ( 1993 ) .
ford , l.a . , et al .
; apj * 439 * , 307 ( 1995 ) .
tavani , m. ; _ apj _ * 466 * , 768 - 778 ( 1996 ) .
tavani , m. , ghirlanda g. , band d. ; in preparation ( 2000 ) . | we present the main results of a study of time - resolved spectra of 43 intense grbs detected by batse .
we considered the 4-parameter band model and the optically thin synchrotron shock model ( otssm ) .
we find that the large majority of time - resolved spectra of grbs are in remarkable agreement with the otssm .
however , about 15 % of _ initial grb pulses _ show an apparent low - energy photon suppression .
this phenomenon indicates that complex radiative conditions modifying optically thin emission may occur during the initial phases of some grbs . |
automatic object recognition and image classification are important and challenging tasks .
this paper is inspired by the remarkable recent work of poggio , serre , and their colleagues @xcite , on rapid object categorization using a feedforward architecture closely modeled on the human visual system .
the main directions it departs from that work are twofold .
first , trading - off biological accuracy for computational efficiency , our results exploit more engineering - motivated mathematical tools such as wavelet and grouplet transforms @xcite , allowing faster computation and limiting ad - hoc parameters .
second , the approach is generalized by adding a degree of _ feedback _ ( another known component of human perception ) , yielding significant performance and robustness improvement in multiple - object scenes . in experiments ,
the resulting scale- and translation - invariant algorithm achieves or exceeds state - of - the - art performance in object recognition , but also in texture and satellite image classification , and in language identification .
as in @xcite , the algorithm is hierarchical . in addition , motivated in part by the relative uniformity of cortical anatomy @xcite , the two layers of the hierarchy are made to be computationally similar , as shown in fig .
[ fig : algo ] .
layer one performs a wavelet transform @xcite in the @xmath0 unit followed by a local maximum operation in the @xmath1 unit .
the transform in the @xmath2 unit in layer two is similar to the grouplet transform @xcite , and is followed by a global maximum operation in the @xmath3 unit .
+ @xmath4 * wavelet transform*. the frequency and orientation tuning of cells in visual cortex v1 can be interpreted as performing a wavelet transform of the retinal image @xcite .
let us denote @xmath5 a gray - level image of size @xmath6 .
a translation - invariant wavelet transform is performed on the ] where @xmath7 denotes the orientation ( horizontal , vertical , diagonal ) , @xmath8 is a wavelet function and @xmath9 are the wavelet coefficients .
scale invariance is achieved by a normalization @xmath10 where @xmath11 is the image energy within the support of the wavelet @xmath12 .
one can verify that @xmath13 where @xmath14 and @xmath15 are the coefficients of @xmath5 and of its @xmath16-time zoomed version @xmath17 .
the normalization also makes the recognition invariant to global linear illumination change .
+ @xmath18 * local maximum * limited translation invariance is achieved at this stage by keeping the local maximum of @xmath14 coefficients in a subsampling procedure : @xmath19 the maximum being taken at each scale @xmath20 and orientation @xmath21 within a spatial neighborhood of size proportional to @xmath22 .
the resulting @xmath23 map at scale @xmath20 and orientation @xmath21 is thus of size @xmath24 .
@xmath25 * grouplet - like transform*. cells in visual cortex v2 and v4 have larger receptive fields comparing to those in v1 and are tuned to geometrically more complex stimuli such as contours and corners @xcite .
the geometrical grouplets recently proposed by mallat @xcite imitate this mechanism by grouping and re - transforming the wavelet coefficients .
the procedure in @xmath2 is similar to the grouplet transform . instead of grouping the wavelet coefficients with a multi - scale geometrically adaptive association field and then re - transforming them with haar - like functions as in @xcite ,
responses of @xmath2 are obtained via inner products between @xmath23 coefficients and sliding patch functions of different sizes : @xmath26 where @xmath27 of support size @xmath28 are patch functions that group the 3 wavelet orientations in a square of size @xmath29 .
while the grouplet functions are adaptively chosen to fit the geometry in the image @xcite , the patch functions @xmath27 , @xmath30 are learned with a simple random sampling as in @xcite : each patch is extracted at a random scale and a random position from the @xmath23 coefficients of a randomly selected training image , the rationale being that patterns that appear with high probability are likely to be learned .
+ @xmath31 * global maximum*. a global maximum operation in space and in scale is applied on @xmath32 and the resulting @xmath33 coefficients @xmath34 are thus invariant to image translation and scale change . + * classification * the classification uses @xmath33 coefficients as features and thus inherits the translation and scale invariance .
while various classifiers such as svms can be used , a simple but robust nearest neighbor classifier will be applied in the experiments .
structures that appear with a high probability are likely to be learned as patch functions through random sampling .
however , they are not necessarily salient and neither are the resulting @xmath33 features .
this suggests active selection of the learned patches .
for example , lowe and mutch have constructed sparse patches by retaining one salient direction at each position @xcite .
a simple patch selection is proposed here by sorting the variances of the @xmath33 coefficients of the _ training _ images .
a small @xmath35 variance implies that the corresponding patch @xmath27 is not salient .
[ fig : c2:variance]-a plots the variance of the @xmath33 coefficients of the motorcycle and the background images in the caltech5 database ( see fig .
[ fig : caltech5 ] ) , the @xmath32 patches being learned from the same images . out of the 1000 patches , 200 salient ones whose resulting @xmath33 have non - negligible variance are selected .
other patches usually correspond to nonsalient structures such as a common background and are therefore excluded .
[ fig : c2:variance]-b and c show that after patch selection the 200 @xmath33 coefficients are mainly positioned around the object , as opposed to the 1000 @xmath33 coefficients spreading over all the image prior to patch selection .
the recognition using these salient patches is not only more robust but also 5 times faster .
[ cols="^,^,^ " , ]
inspired by the biologically motivated work of @xcite , we have described a wavelet - based algorithm which can compete with the state - of - the - art methods for fast and robust object recognition , texture and satellite image classification , language recognition and sound classification .
a feedback procedure has been introduced to improve recognition performance in multiple - object scenes .
potential applications also include video archiving ( semantic video analysis ) , video surveillance , high - throughput drug development , texture retrieval , and robotic learning by imitation . to further improve and extend the algorithm , a key aspect will be a more refined use of feedback between different levels .
such feedback will naturally involve stability and convergence questions , which will in turn both guide the design of the algorithm and shape its performance .
in addition , contrary to the nervous system , the algorithm need not be constrained by information transmission delays between different levels .
preliminary ideas in this direction are briefly discussed in the appendix .
+ * appendix *
the first step towards introducing a dynamic systems perspective aimed at further development of feedback mechanisms is simply to rewrite the algorithm in terms of differential equations , which puts it in a form more suitable to subsequent analysis of stability and convergence . with spike amplitude equal to @xmath46 @xmath47
the function @xmath48 is discussed later in this section and in a companion paper .
the dynamics of @xmath49 can be modified in turn so that states corresponding to each object appear in sequence according to the state @xmath50 where , componentwise , @xmath53 where @xmath54 is active and and @xmath55 otherwise .
note that @xmath50 smoothly transitions between @xmath56 and @xmath46 according to the attended object .
the positive gain @xmath57 is chosen such that @xmath58 , where @xmath59 is the spike duration , itself a fraction of the interspike period .
techniques for globally stable spike - based clustering are described in a companion paper , based on modified fitzhugh - nagumo neural oscillators @xcite , similar to @xcite , @xmath60\nonumber\end{aligned}\ ] ] where @xmath61 is the membrane potential of the oscillator , @xmath62 is an internal state variable representing gate voltage , @xmath63 represents the external current input , and @xmath64 , @xmath65 and @xmath66 are strictly positive constants . using a diagonal metric transformation @xmath67
, one easily shows , similarly to @xcite , that one of the most immediate additional feedback mechanisms to be explored is that of generalized diffusive connections ( @xcite , section 3.1.2 ) . in a feedback hierarchy , these correspond to achieving consensus between multiple processes of different dimensions .
similarly to @xcite , composite variables for dynamic tracking can be used at every level , based on both top - down an bottom - up information .
this allows one to implicitly introduce time - derivatives of signals in the differential equations , without having to measure or compute these terms explicitly .
99 k. chen and d.l .
wang , `` a dynamically coupled neural oscillator network for image segmentation . '' , _ neural networks _ ,
15 , 423 - 439 .
r. fergus and p. perona and a. zisserman , `` object class recognition by unsupervised scale - invariant learning '' , _ cvrp _ , vol.2 , pp.264 - 271 , 2003 .
r. fitzhugh , `` impulses and physiological states in theoretical models of nerve membrane '' , _ biophysical journal _ , vol.1 , pp.445 - 466 , 1961 .
j. hawkins , s.blakeslee , _ on intelligence _ , times books , 2004 .
r.m.haralick , k.shanmugam , i.dinstein , `` textural features for image classification '' , _ ieee trans . on sys man cy , smc-3 , ( 6 ) : 610 - 621 _ , 1973 .
k. kim , k. jung , s. park , and h. kim , `` support vector machines for texture classification '' , _ ieee trans .
pami _ , vol.24 , no.11 , pp.1542 - 1550 , 2002 . s. lazebnik , c. schmid and j. ponce , `` a sparse texture representation using local affine regions '' , _ ieee transactions on pattern analysis and machine intelligence _ , vol .
27 , no . 8 , pp .
1265 - 1278 , 2005 .
x. liu and d. wang , `` texture classification using spectral histograms '' , _ ieee trans .
pami _ , vol.12 , pp.661 - 670 , 2003 .
w. lohmiller , and j.j.e .
slotine , `` global convergence rates of nonlinear diffusion for time - varying images '' , _ scale - space theories in computer vision , lecture notes in computer science _ , vol.1682 , springer verlag ( 1999 ) .
s. lu and c. tan , `` script and language identification in noisy and degraded document images '' , _ ieee trans .
pami _ , vol.30 , no.1 pp.14 - 24 , 2008 .
b.luo , j-f.aujol , y.gousseau , s.ladjal , `` indexing of satellite images with different resolutions by wavelet features '' , _ ieee trans image proc _ , accepted , 2008 .
s. mallat , `` geometrical grouplets '' , _ acha _ , to appear , 2008 .
s. mallat , _ a wavelet tour of signal processing _ , academic press , 2nd edition , 1999 .
v. mountcastle , `` an organizing principle for cerebral function : the unit model and the distributed system '' , the mindful brain , mit press , 1978 .
j. nagumo . and s. arimoto , and s. yoshizawa , `` an active pulse transmission line simulating nerve axon , '' _ proceedings of the ire _ , 50(10 ) , pp.2061 - 2070,1962 .
pham and j.j.e .
slotine , `` stable concurrent synchronization in dynamic system networks , '' _ neural networks _ , 20(1 ) , 2007
. t. randen and j. husoy , `` filtering for texture classification : a comparative study '' , _ ieee trans on image proc _ ,
vol.21 , no.4 , pp.291 - 310 , 1999 . j. mutch and d. lowe , `` multiclass object recognition with sparse , localized features '' , _ cvpr 06 _ , pp.11 - 18 .
rao and d.h .
ballard , `` predictive coding in the visual cortex : a functional interpretation of some extra - classical receptive - field effects '' , _ nature neuroscience _ , 2 , 1 , 79 , 1999 . t. serre , l. wolf , s. bileschi , m. riesenhuber and t. poggio , `` robust object recognition with cortex - like mechanisms '' , _ ieee trans .
pami _ , vol.29 , no.3 , pp.411 - 426 , 2007 .
von melchner , l. , pallas , s.l . and
sur , m , `` visual behavior mediated by retinal projections directed to the auditory pathway '' , _ nature _ , 404 , 2000 .
d. walther , t. serre , t. poggio and c. koch , `` modeling feature sharing between object detection and top - down attention '' , vss , may 2005 .
w. wang and j.j.e .
slotine , `` on partial contraction analysis for coupled nonlinear oscillators , '' _ biological cybernetics _ , 92(1 ) , 2005 .
m. weber and m. welling and p. perona , `` unsupervised learning of models for recognition '' , _ eccv _
, pp.18 - 32 , 2000 . | we investigate a biologically motivated approach to fast visual classification , directly inspired by the recent work @xcite . specifically , trading - off biological accuracy for computational efficiency , we explore using wavelet and grouplet - like transforms to parallel the tuning of visual cortex v1 and v2 cells , alternated with max operations to achieve scale and translation invariance .
a feature selection procedure is applied during learning to accelerate recognition .
we introduce a simple attention - like feedback mechanism , significantly improving recognition and robustness in multiple - object scenes . in experiments ,
the proposed algorithm achieves or exceeds state - of - the - art success rate on object recognition , texture and satellite image classification , language identification and sound classification . |
the lyman alpha ( ly@xmath1 ) forest " ( @xcite ; @xcite ) of spectral features caused by hi absorption along the line of sight to a quasar probes the state of the intergalactic medium over a wide range of physical conditions . during the past few years
, high - precision observations made using the hires spectrograph ( @xcite ) on the 10 m keck telescope have quantified the statistics of these low column density absorbers to unprecedented accuracy ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ) . during the same time span ,
cosmological simulations that incorporate gas dynamics , radiative cooling , and photoionization have been able to reproduce many of the observed properties of quasar absorption spectra ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) .
the rapid progress on theoretical and observational fronts has led to the emergence of a new paradigm for the origin of the high - redshift ( @xmath12 ) ly@xmath1 forest , in which most ly@xmath1 forest lines are produced by regions of low to moderate overdensity in hierarchically collapsing structures that are not in dynamical or thermal equilibrium .
ly@xmath1 lines of lower column density generally arise in gas of lower physical density , which has a lower neutral hydrogen fraction because of the reduced recombination rate . in this paper , we use a matched comparison between a cosmological hydrodynamic simulation and a keck hires spectrum of the quasar q1422 + 231 to constrain the metal abundance of this low density gas . the recent detection of metal lines associated with ly@xmath1 forest absorbers having column densities @xmath13 ( @xcite ; @xcite ; songaila & cowie 1996 , hereafter @xcite ) has provided a new avenue for investigating the ionization state and enrichment history of the high - redshift intergalactic medium ( igm ) .
@xcite showed that 75% of ly@xmath1 absorbers with @xmath14 have associated civ absorption . using simple photoionization models they estimate that the mean metallicity of these absorbers is between [ c / h]@xmath15 and @xmath16 .
studies that use cosmological simulations to model the density and temperature of the absorbing gas obtain a better match to the data for a mean metallicity of [ c / h]@xmath17 with around one dex of scatter , assuming either a power - law ionizing background ( @xcite ) or a reprocessed quasar ionizing background ( hellsten 1997 , hereafter @xcite ) .
the question of how metals came to reside in these intermediate - density ly@xmath1 absorbers remains unanswered .
since ly@xmath1 absorbers with @xmath18 are optically thin and associated with density peaks of overdensity @xmath19 , it is unlikely that they contain star - forming regions that produce _ in situ _ enrichment .
thus some transport mechanism must be invoked to explain the presence of metals in these regions .
one possibility is that the metals are ejected from nearby , forming galaxies , by some combination of supernova blowout ( @xcite ) and tidal stripping ( @xcite ; @xcite ) .
since the enriching proto - galaxies are likely to form more efficiently in high density environments , these scenarios predict a significant correlation between metallicity and density ( see , e.g. , figure 6 of gnedin & ostriker 1997 ) .
alternatively , the igm may have been enriched at a very early epoch by more ubiquitous population iii objects ( e.g. , @xcite ) , in which case all ly@xmath1 absorbers might be expected to have roughly the same metallicity . because civ lines probe only a small range of hi column densities ( and hence physical densities ) with adequate statistics , it is difficult to distinguish between these enrichment models using only civ data .
metal abundances can be studied at higher densities in lyman limit systems ( @xmath20 ) and damped ly@xmath1 systems ( @xmath21 ) , but these are objects where _ in situ _ enrichment is likely and where ( especially for lyman limit systems ) uncertain radiative transfer effects complicate the inference of metal abundances from line strengths . in this paper , we attempt to extend metallicity constraints to the low column density ( @xmath4 ) , and hence low physical density , ly@xmath1 forest .
carbon is difficult to detect in this regime because the low density reduces @xmath22 and ionizes more civ to cv .
however , @xcite have used composite spectra to attempt to detect civ in such systems , and we briefly discuss their results in [ sec : disc ] . in this paper
, we focus instead on ovi , which is expected to be the one detectable metal absorption feature tracing ly@xmath1 absorbers with @xmath4 because of its high ionization state and large oscillator strength ( hellsten 1998 , hereafter @xcite ) .
the difficulty with this approach , and the reason that it has not been previously attempted , is that the ovi absorption features lie embedded within the ly@xmath1 forest , which is quite crowded at these redshifts .
we overcome this problem by using a line identification scheme specifically designed to select candidate ovi features and by using artificial spectra extracted from a realistic cosmological simulation to calibrate the efficiency of ovi detection and the contamination from narrow ly@xmath1 lines . with these procedures
, we can test whether the low density regions of the ly@xmath1 forest are consistent with a uniform metallicity extrapolated from the civ data at higher densities .
the low density igm is highly photoionized by the metagalactic ultraviolet ( uv ) background . for our standard spectral shape ,
we assume that the uv background is produced by quasar emission reprocessed by ly@xmath1 forest absorption ( haardt & madau 1996 , hereafter @xcite ) .
we find that if the igm metallicity is uniform at [ c / h]@xmath17 , the uv background has the spectral shape given by @xcite , and oxygen has the factor of three overabundance ( relative to solar ) predicted by type ii supernova enrichment models , then ovi should be readily detectable in the spectrum of q1422 + 231 .
however , our detection algorithm finds very few candidate ovi lines in the spectrum .
the absence of detectable ovi features has several possible interpretations : ( 1 ) oxygen is not overabundant relative to carbon in the low density igm , ( 2 ) the metallicity of low density regions as traced by @xmath4 ly@xmath1 absorption systems is lower than the metallicity of intermediate - density regions traced by higher column density absorption , or ( 3 ) there are many fewer high - energy photons capable of photoionizing ov to ovi than are predicted by the @xcite ionizing background . to distinguish between these interpretations , we apply a second algorithm , the optical depth ratio technique of songaila ( 1998 , hereafter @xcite ) designed to detect ovi in regions where significant civ absorption is found .
since we have independently determined [ c / h ] in these regions , we can use this technique to discriminate between different ionization conditions and abundance patterns . by applying this technique to the spectrum of q1422 + 231 and calibrating the results using artificial spectra , we find that : ( 1 ) a significant metallicity gradient ( declining metallicity with declining density ) must exist regardless of whether helium has mostly reionized or not , ( 2 ) if helium has reionized by @xmath23 , our results are consistent with [ o / c ] @xmath24 , and ( 3 ) if the epoch of helium reionization does not begin until @xmath25 , a highly implausible oxygen overabundance of [ o / c ] @xmath26 is required .
combining the results from these two analysis techniques , our favored scenario for the low density igm at @xmath27 is one in which more than half of the volume of the universe has helium predominantly reionized by @xmath23 , oxygen is overabundant relative to carbon by a factor @xmath28 , and spatial regions with overdensities @xmath29 have an average metallicity _ at least _ a factor of 3 higher than regions near the mean baryonic density .
section [ sec : modeling ] describes our cosmological simulation , the q1422 + 231 data , and our procedure for constructing artificial absorption spectra .
section [ sec : ovisearch ] discusses previous ovi searches and reviews @xcite s argument that ovi should be the most effective tracer of metallicity in ly@xmath1 absorbers with @xmath30 .
section [ sec : civ ] examines civ absorption at intermediate column densities , repeating the general arguments of @xcite and rauch et al .
( 1997a ) but with a much closer match between the theoretical and observational analysis procedures .
section [ sec : search ] is the heart of the paper .
it describes our algorithm for identifying candidate ovi lines , presents the results of the ovi searches in the real and artificial spectra , and discusses the properties of candidate ovi absorbers in q1422 + 231 and the simulations .
figures [ fig : auto5][fig : ovisel ] demonstrate the paper s central results .
section [ sec : assumptions ] discusses the impact of varying the assumptions of our standard theoretical model .
section [ sec : pixmet ] describes the optical depth ratio technique for quantifying ovi absorption , shows the results from this technique applied to q1422 + 231 and artificial spectra , and discusses these results in conjunction with the results from section [ sec : search ] .
section [ sec : disc ] summarizes our conclusions and discusses them in light of other recent observational and theoretical developments .
we perform a cosmological hydrodynamic simulation of a @xmath5-dominated cold dark matter ( lcdm ) model with @xmath31 , @xmath32 , @xmath33 , @xmath34 , inflationary spectral index @xmath35 , and rms fluctuation amplitude @xmath36 .
these values yield a cobe - normalized lcdm model that is in agreement with most current observational constraints ( @xcite ) .
we chose the lcdm model because it reproduces the characteristics of the ly@xmath1 forest of q1422 + 231 most closely ( dav , in preparation ) , although the differences from other popular cosmologies are slight .
we choose the value of the baryon density @xmath37 to match recent observations of the deuterium abundance that give @xmath38 ( @xcite ; @xcite ) . matching the observed mean transmission in the ly@xmath1 forest using reasonable estimates for the intensity of the metagalactic photoionizing background
also requires a baryon abundance of this order ( @xcite ; @xcite ; @xcite ) .
the simulations were performed using ptreesph ( @xcite ) , a version of the treesph code ( @xcite ; @xcite ) implemented on massively parallel supercomputers .
we use @xmath39 dark matter particles and @xmath39 gas particles in a periodic cube 11.111@xmath40 comoving mpc on a side and a gravitational softening length of 3@xmath40 comoving kpc ( equivalent plummer softening ) .
the masses of the dark matter and gas particles are @xmath41 and @xmath42 , respectively .
the initial fluctuations are a gaussian random field with the power spectrum computed using the hu & sugiyama ( 1996 ) formulation of the cdm transfer function .
we include radiative cooling for primordial composition gas and photoionization and heating from a spatially uniform uv background .
we adopt the uv spectral shape and intensity ( versus redshift ) from @xcite , with the exception that the intrinsic quasar spectrum has been softened from a power law index of @xmath43 to @xmath44 , as is now favored by those authors ( madau , private communication ) ; we shall still refer to this as the @xcite spectrum .
we compute the relative abundances of ionic species assuming equilibrium between ionization and recombination ( see @xcite ) ; we will examine possible effects of this assumption in [ sec : hot ] .
we include a prescription for star formation , but for the results contained in this paper , this has no significant effect ( @xcite ) .
we ran from a starting redshift of @xmath45 to @xmath46 .
this simulation took roughly 3400 node - hours on 16 processors of the cray t3e at the pittsburgh supercomputing center .
in this paper we will be comparing our simulations to only one quasar spectrum , that of q1422 + 231 ( shown in the upper left panel of figure [ fig : qspec ] ) , since that is the only one currently available to us .
it was generously provided to us by a. songaila and l. cowie , and their observations of q1422 using keck s hires spectrograph are described more fully in @xcite . here
we briefly note that the quasar emission redshift is @xmath0 , and the spectral resolution ( as provided to us ) is 0.06 , corresponding to roughly @xmath47 per pixel from 3900 to 7200 .
the signal - to - noise ratio is roughly @xmath48 per pixel in the region of ly@xmath49 and ovi absorption ( @xmath50 ) , @xmath51 per pixel in the pure ly@xmath1 forest region , and over 100 per pixel in the metal line region redwards of the ly@xmath1 emission peak . due to a lyman limit system at @xmath52 ,
the spectrum of q1422 effectively truncates around 4000 .
thus @xmath5340004800 represents the usable region for our ovi search , corresponding to @xmath54 , with poorer sensitivity towards lower redshifts because of the dropping intrinsic flux of the quasar and the degrading response of hires towards the blue . to model q1422 as closely as possible , we have developed a new method for generating artificial quasar spectra from a cosmological simulation .
our approach produces continuous line - of - sight artificial spectra between two desired wavelengths , analogous to real quasar spectra , rather than a large number of small , disjoint spectra at a single redshift as has been typically done in other studies ( e.g. , @xcite ) .
the advantages of these _ continuous artificial spectra _ are that they automatically include the redshift evolution intrinsic to any given model and that the similarity between the artificial spectra and the real quasar spectrum allows the identical analysis routines to be applied to both .
this reduces the algorithm - dependent variations of continuum fitting , noise estimation , and most importantly for this work , metal line identification .
generating continuous spectra requires that particle information be output at frequent intervals during a simulation .
since we can not output all the simulation data every few timesteps owing to disk space and i / o time constraints , we output only the particle information required to generate the desired spectra .
we choose the frequency of output data so that the redshift interval between outputs exactly equals the length of the box in redshift space , given by @xmath55 l,\ ] ] where @xmath56 is the hubble constant at redshift @xmath57 , @xmath58 is the hubble constant today , @xmath59 is the simulation size in comoving @xmath40mpc , and @xmath60 is the speed of light . at each output timestep during the simulation run , we choose six random lines of sight from within the volume .
we output information ( @xmath61 ) for all gas particles whose sph smoothing volume ( i.e. , the region of space a gas particle represents ) intersects one of these lines of sight . at each output , a new set of six lines of sight is randomly chosen .
we begin outputting spectra continuously at @xmath62 ; roughly 100 outputs are required to @xmath46 .
we use this particle information together with the assumed ionizing background interpolated to the output redshift to generate optical depths along these lines of sight for the following ions : hi , hei , heii , civ , cii , siiv , nv , and ovi .
we use cloudy 90 ( @xcite ) to compute lookup tables for the fraction of gas in each ionization state at a given density and temperature for our adopted @xmath63 , accounting for both photoionization and collisional ionization , and we use routines extracted from the tipsy package ( @xcite ) to compute smoothed gas properties along each line of sight ; see @xcite for more details .
we assume that the gas is optically thin , which is appropriate for the ly@xmath1 forest regime .
we compute the profiles of optical depth versus observed wavelength ( or , equivalently , versus redshift ) individually for each output box , then join the contiguous output boxes to produce continuous spectra for redshifts below @xmath62 along each of the six lines of sight .
when producing these continuous spectra , we have to pay special attention to the regions where the spectra from contiguous output boxes are joined , because the physical properties there are discontinuous . since the simulation volume is periodic , we can arbitrarily shift the particle positions , and hence the computed optical depths , in a periodic sense within a single output box .
we shift the optical depths so that the point of lowest hi absorption lies at the ends of the spectra at any given output .
thus we join the spectra in regions of very low optical depth , which for our purposes are uninteresting ; they also arise from voids that have relatively smooth and homogeneous physical properties .
we smooth these transition regions over a small interval of five pixels out of a total of 1000 output pixels ( where one pixel in our artificial spectrum corresponds to @xmath64 at these redshifts ) . with a list of optical depths versus redshift for various ions
, we are now equipped to produce artificial quasar spectra .
we select a quasar spectrum to simulate ( in our case q1422 ) , specifying its redshift , the wavelength interval over which we wish to produce a spectrum , its resolution in wavelength space , and a spatially uniform metallicity .
we assume an abundance pattern typical of low - metallicity stars and hii regions , presumably enriched by type ii supernovae : @xmath65 [ fe / h ] = [ c / h ] , [ n / c ] @xmath66 , [ si / c ] @xmath67 , and [ o / c ] @xmath68 ; we discuss these ratios further in [ sec : o / c ] . in each wavelength bin
, we sum the contribution to the optical depth from all ions with redshifts lower than the quasar redshift .
we consider the entire lyman series up to ly-30 , and both doublet components for the metal species .
this procedure yields a list of `` observed '' optical depths versus wavelength at the specified spectral resolution .
we now add noise and fit a continuum to each artificial spectrum .
although we know the true location of the continuum in advance , it is important to perform a continuum fit to maintain equivalence between the procedures for analyzing observed and artificial spectra .
we have developed automated routines for estimating the signal - to - noise ratio on a pixel - by - pixel basis and fitting a continuum to raw quasar data .
we estimate the noise by first identifying small regions of saturation and measuring the average detector noise , and then finding regions of continuum " to estimate the average noise from photon statistics , within 30 segments of the spectrum . in the dense ly@xmath1
forest it is difficult to find sizable regions where the flux is near the continuum , so typically the shot noise is somewhat overestimated because it includes a contribution from fluctuating low level absorption .
we have not corrected for this effect . nevertheless , our noise estimates agree reasonably well with those of @xcite .
we estimate the continuum by fitting smoothed second - order polynomials to the spectral peaks in 30 intervals .
while this may appear crude , no method can recover the true continuum because even in the artificial spectra the flux rarely approaches the true continuum level for @xmath27 .
thus we strive only to implement a method that can be applied identically to the real and artificial spectra , is similar to methods traditionally used in observational analyses , and uses a low order fitting function in order to minimize `` overfitting '' of true absorption fluctuations . because our ovi search below is based on identification of narrow isolated absorption features , the details of continuum fitting are not critical for our current purposes .
we estimate the continuum and noise characteristics of q1422 using the above algorithms . to incorporate noise into the artificial spectra
, we take the detector noise and shot noise as estimated in the observed spectra and add it directly ( with a gaussian random distribution ) to the intensity of the artificial spectra , giving the artificial spectra very similar noise characteristics to the observed spectra .
we continuum fit the artificial spectra , and the resulting data are output as a single artificial quasar spectrum
. the intensity of the metagalactic photoionizing background ( @xmath63 ) is not tightly constrained by current observations .
we use the intensity predicted by @xcite when evolving the simulation , but because @xmath63 does not significantly affect the dynamics in the optically thin , low - to - moderate overdensity regions , we can vary the intensity of @xmath63 _ a posteriori _ to desired levels and obtain nearly identical results to having redone the entire simulation with that @xmath63 ( @xcite ) . to constrain @xmath63 , we note that the mean opacity @xmath69 of the ly@xmath1 forest depends on the parameter combination @xmath70 , since the hi optical depths are proportional to this combination ( for fixed gas temperature ) . after obtaining a noise - added , continuum - fitted artificial spectrum ,
we measure its mean transmission , then use that information to adjust @xmath63 , and repeat the above procedure to produce a new quasar spectrum .
we typically iterate several times until the mean transmission converges to within 1% of the observed mean transmission for q1422 ; this corresponds to a determination of @xmath63 to better than 2% ( given our adopted @xmath37 ) .
we perform this process individually for each artificial quasar spectrum . for our lcdm model ,
the average factor by which the @xcite intensity must be increased is @xmath71 , where the variance is computed over the six artificial spectra . as a side note ,
the implied value of @xmath37 for consistency with the @xcite intensity itself is @xmath72 for this model .
once we have an intensity of @xmath63 for each artificial spectrum , we now recalculate the metal species optical depths using this new intensity , because unlike hi absorption , metal absorption does not scale in a trivial way with @xmath63 .
once we calculate the new optical depths , we again fit a continuum and add noise in the manner described above .
therefore , each artificial spectrum has the correct hi mean transmission , and the metal absorption is computed assuming a spatially uniform metallicity and ionizing background with the @xcite shape and an intensity constrained by the mean transmission of q1422 . in figure
[ fig : qspec ] ( left panels ) we show the spectrum of q1422 ( reproduced from @xcite ) and an example of one of our six artificial spectra for q1422 with metallicity @xmath73 , with the corresponding continuum fit and noise level ( multiplied by four for visibility ) in the ly@xmath1 forest region .
the differences between these two spectra are due to : ( 1 ) emission features in the quasar not present in the artificial spectra , ( 2 ) the downward slope of the observed continuum in the ly@xmath1 forest of q1422 caused by the intrinsically varying spectral intensity of the quasar and the declining response of the hires spectrograph towards the blue , and ( 3 ) the intrinsic fluctuations in the continuum level of q1422 .
these features in q1422 are subsequently normalized out by fitting a continuum level to the peaks , as shown in the upper right panel of figure [ fig : qspec ] ; the identical continuum fitting routine is applied to the artificial spectrum for consistency . after this procedure
, the artificial spectrum contains all the significant characteristics of the q1422 spectrum , as shown in the lower right panel of figure [ fig : qspec ] .
we also fit a continuum to the spectra redwards of the ly@xmath1 peak ; this procedure is much more robust since there are plenty of regions of virtually unabsorbed continuum .
there are some features of real quasar spectra that are not reproduced in these continuous artificial spectra , although we expect that these differences will not significantly affect our results .
first , correlations of lines across the boundaries where the segmented spectra have been joined are not properly reproduced ; thus caution must be taken when using these spectra to measure line - of - sight correlation functions . in this paper
, we are counting and measuring only isolated lines , so this problem does not arise .
second , metal line absorption from systems at low redshift is not properly taken into account , since our artificial spectra stop at @xmath46 . in practice ,
the influence of interloping metal lines on the mean transmission in the ly@xmath1 forest region is extremely small , so our determination of @xmath63 is virtually unaffected .
third , real spectra have intrinsic small - scale variations in their noise level , e.g. , owing to a fluctuating response across an echelle order .
the noise level of the artificial spectra is constant over a @xmath74 scale , so these small - scale fluctuations are not reproduced .
we do not expect this to have a significant effect on our conclusions , but in future work we will model some of these effects in greater detail .
previous searches for ovi absorption at high redshifts have had very limited success .
the only claimed detection of an ovi line at @xmath75 is by kirkman & tytler ( 1997 ; hereafter @xcite ) .
they found a single narrow feature in q1422 which they identified as ovi primarily on the basis of associated civ absorption .
the doublet companion was subsumed by an hi line .
the method of searching for ovi associated with civ is taken from low - redshift studies ( @xcite ; @xcite ) .
however , given a universe with a homogeneous metallicity and an ionizing background shape similar to that predicted by @xcite , this strategy is not the optimal way to find ovi .
the reason for this is illustrated by the line observability index ( lox ) plot in figure [ fig : lox ] .
the lox is described more fully in @xcite ; here we outline the relevant features .
the lox is calculated using the density - temperature relation taken from a balance between cosmic expansion and photoionization heating , which predicts @xmath76 , and the peak density - column density relation computed from artificial spectra , @xmath77 using these relations , a given hi column density can be associated with a particular ionization condition , from which the fraction in any ionization state can be calculated using cloudy 90 ( @xcite ) . combined with an assumed ( uniform ) metallicity , here taken to be @xmath78 solar for civ and @xmath79 solar for ovi , this analysis predicts a rest equivalent width @xmath80 for a typical absorption line of a given species . by the appropriate choice of additive constants ( see @xcite ) , the lox is given by @xmath81 where the equation is exact in the limit of weak lines .
figure [ fig : lox ] shows the lox of ovi and civ given the @xcite spectrum ( solid curves ) .
the horizontal dot - dashed line corresponds roughly to the detectability limit in the ly@xmath1 forest of q1422 , while the horizontal dotted line reflects the detectability limit redwards of the ly@xmath1 emission peak .
the dashed curves will be addressed in [ sec : cut ] .
this figure is similar to figures 35 of @xcite except that the lox is now calculated for the lcdm model used in this paper , which has a higher baryon fraction .
the impact of the higher @xmath37 is minimal , as @xcite predicted it would be .
as stated in @xcite , given the assumed metallicity and ionization state , civ will be observable in lines with hi column densities @xmath82 , while ovi will be observable to significantly lower hi column densities , down to almost @xmath83 ( if the ovi line is not subsumed by an hi line ) . since the column density distribution @xmath84 ( @xcite ; @xcite ) , there are many more hi lines at lower column densities .
this means that in any given quasar spectrum , if the metallicity is uniform across all hi column densities , photoionized ovi should be associated with hi lines that often have no associated civ absorption .
we demonstrate this point more quantitatively in
[ sec : nhist ] below .
@xcite claimed that the ovi detected in their @xmath52 lyman limit system in q1422 must be collisionally ionized , in part by arguing that few high - energy photons capable of ionizing oxygen to ovi would exist far from quasars .
they argued that while @xcite assumed uniform emissivity of quasars , in reality quasars are discrete sources , and far from quasars there should be further softening of the radiation field due to reprocessing by ambient heii .
they calculated the mean free path of heii - ionizing ( 4 ry ) photons to be roughly @xmath85mpc , and from this value obtained a volume filling factor of only 2% . while valid for heii , the extension of this argument to ovi is less certain
. photons capable of ionizing ov to ovi require energies greater than 114 ev @xmath86 8.4 ry . since the cross section of absorption scales as @xmath87 , the mean free path of an ov ionizing photon
is @xmath88 times that of an heii ionizing photon . given this additional factor of about 1000 in volume , a universe with the observed distribution of quasars and a @xcite ionizing background
will effectively be transparent to ov - ionizing photons .
this does not imply that the ovi absorption system detected by @xcite is necessarily photoionized it is associated with a high column density hi absorber that might well be embedded in a collapsed gas halo hot enough to cause collisional ionization . nevertheless , while our assumption of a uniform ionizing background may be an oversimplification
, we do not expect that its spectrum should have _ systematically _ fewer ov - ionizing photons than predicted by @xcite , provided that most heii has been reionized to heiii by the redshift in question .
in @xcite we analyzed civ artificial spectra in a manner similar to @xcite s procedure , allowing a direct comparison with @xcite s results .
however , the artificial spectra used were of the segment spectra " variety , , many small intervals of spectra at particular redshifts , and the detailed modeling of q1422 was not as sophisticated as we do here . in this section ,
we describe a reanalysis of civ absorption in q1422 , alongside an analysis of civ in our continuous artificial spectra .
the main additional ingredient is that we now identify and fit civ and hi absorption features in the artificial spectra and q1422 using the same routines
. nevertheless , the basic results are similar to those of @xcite . the automated voigt profile fitter autovp ( @xcite ) has been extended to identify and fit metal line doublets lying outside the ly@xmath1 forest , such as civ .
once a spectrum has been fitted with ly@xmath1 lines , the identification scheme searches for spectral features in the ly@xmath1 forest region where the intensity drops below some threshold , taken to be 0.7 . within the wavelength interval where the intensity is below that threshold , autovp searches for associated civ absorption in the two regions corresponding to the doublet positions . if an absorption feature is due to civ , it should be present in both regions , with the strong component region ( 1548 ) having twice the optical depth of the weak component region ( 1551 ) because of the difference in oscillator strengths . to identify any absorption that could possibly be civ , autovp doubles the optical depths of each pixel in the weak component region , superimposes them on the strong line region with an appropriate shift in wavelength , then takes the smaller of the two optical depths ( strong component or double the weak component ) as its measure of civ absorption .
since civ is the dominant species of absorption redwards of the ly@xmath1 emission peak , the implicit assumption that all absorption identified in this way is civ turns out to be quite good .
we then fit this remaining absorption with civ voigt profiles , and sum the column densities within the region .
the associated hi column density is found by summing all column densities of hi lines within the associated spectral region . since the ly@xmath1 feature is typically saturated and therefore yields an inaccurate estimate of the hi column density , autovp searches down the lyman series until it finds the first non - saturated lyman series region , or alternatively the last observable lyman series region before the blue cutoff in the spectrum ; this is the hi line which is fit . wherever possible ,
the unsaturated regions of other lyman series lines ( except ly@xmath1 ) are converted to optical depths , and used to constrain the optical depth at the associated position in the fitted lyman series line region ; this minimizes the contribution due to coincident absorption from other lyman series lines .
after this constraint has been applied , the lyman series region is fit with voigt profiles to obtain the hi column density .
often each region has several hi and civ lines ; we do not attempt to associate individual civ lines with hi lines , as varying ionization conditions and bulk motions can make this identification very confusing and produce unphysical results .
our procedure is similar to that of @xcite , although they fit all lyman series lines simultaneously to obtain the hi column density . in any case , we apply the exact same detection and measurement algorithm to q1422 and the artificial spectra , so whatever biases are introduced will be similar in both data sets .
figure [ fig : civratio ] shows our reanalysis of civ lines in q1422 and an identical analysis of our six artificial q1422 spectra assuming a uniform metallicity of [ c / h ] @xmath89 .
the identified civ systems in q1422 correspond reasonably well with @xcite s systems ; the main differences arise from the estimate of @xmath90 .
we conclude that our procedure is not markedly different in practice from that of @xcite .
the scatter in the civ / hi ratio at a given column density in the simulated spectra arises from spatial variations in the density and temperature and from variations introduced by line identification and fitting .
the ionizing background and metallicity are assumed to be spatially uniform .
the assumed metallicity of [ c / h ] @xmath89 yields a mean civ / hi ratio in good agreement with q1422 , as we show in figure [ fig : metfrac ] .
rauch et al .
( 1997a ) and hdhwk found that a 1 dex scatter in metallicity was required for simulations to reproduce the observed scatter in civ / hi . our more closely matched comparison here implies a smaller intrinsic scatter of about 0.5 dex ; the reduction arises mainly because of additional `` observational '' scatter introduced by the line fitting algorithm .
we determine the metallicity required to match the mean value of civ / hi by varying the assumed metallicity and then identifying and fitting metal lines in the artificial spectra .
figure [ fig : metfrac ] ( solid curve ) shows the mean log column density of observed civ lines ( @xmath91 ) as a function of metallicity .
we find @xmath91 to be the most robust statistic for this comparison .
similar results are obtained by using the fraction of hi lines with detectable civ or the median value of civ / hi .
the solid horizontal line is @xmath91 for q1422 s civ systems , identified by the same method that was applied to the artificial spectra .
the crossover is seen to occur right around [ c / h ] = @xmath92 ( slightly higher if the median civ / hi is used ) .
the errorbars indicate the variance of @xmath93 over the six artificial spectra ; they do not include the intrinsic scatter in civ / hi .
the dotted curve will be discussed in [ sec : hot ] and the dashed curve in [ sec : cut ] . from this analysis
, we confirm the results of rauch et al .
( 1997a ) and @xcite that a mean metallicity of [ c / h ] @xmath3 with a small amount of spatial scatter , together with an ionizing background of the @xcite shape , reproduces the observed distribution of civ / hi ratios in q1422 quite well .
note that the cosmological model and @xmath37 value used for our simulations are substantially different from those used by @xcite , demonstrating the overall lack of sensitivity of the mean metallicity determination to these assumptions .
we now turn our attention to the question of whether this metallicity distribution is consistent with observations of ovi absorption .
coincident hi absorption complicates the detection of ovi(1032,1037 ) lines in quasar spectra . for q1422 ,
ovi is detectable only below about 4800 , lying in a region with a high density of ly@xmath1 ( 1216 ) and ly@xmath49 ( 1026 ) absorbers .
ovi lines are expected to be narrower than hi lines because thermal broadening is a factor of four lower . however , as one goes to lower density structures , broadening by residual hubble flow across the absorbing region becomes progressively more important ( @xcite ) , making ovi lines broader than naively predicted by thermal broadening .
nevertheless , we expect that the distribution of ovi line widths will have a tail to smaller @xmath94-parameter ( i.e. , gaussian velocity width ) than hi line widths .
since our goal is to distinguish ovi absorption from hi absorption rather than to identify all possible ovi features , we focus on this tail of narrow ovi lines in our detection algorithm . figure [ fig : auto5 ] shows a comparison between a roughly 100 segment in one of our continuous artificial spectra versus the spectrum of q1422 .
the top panel shows an artificial spectrum in the same wavelength range without any metals added ( i.e. , only lyman series absorption ) .
the second panel shows the absorption caused purely by ovi , including both doublet lines .
the third panel shows the artificial spectrum with a metallicity of [ c / h]@xmath89 and [ o / c ] @xmath68 , essentially corresponding to the optical depth sum of the first and second panels , since no other ions contribute significantly in this wavelength range
. the bottom panel shows q1422 between 4510 and 4605 .
the noisy lines represent the data ; the smooth curves will be explained below . from the third panel of figure [ fig : auto5 ]
it is apparent that some narrow features from ovi do survive despite the blanketing by hi .
we tried a variety of algorithms to quantify the presence of these additional narrow features .
the standard technique of decomposing the full spectrum into a superposition of voigt profiles ( which are effectively just gaussian optical depth profiles in this low column density regime ) proved ineffective .
many of the hi features are intrinsically asymmetric , and in voigt profile decomposition narrow components are often introduced to fit the wings of these asymmetric lines .
it is therefore more useful to focus on _ isolated _ narrow lines . as an alternative to voigt profile decomposition
, we developed a much simpler algorithm that searches for spectral features that dip by more than @xmath95 and then return to their original level all within a very small velocity interval , @xmath96 ( 45 pixels ) .
the @xmath95 criterion eliminates noise spikes , while the velocity range criterion is tuned to optimally select out the tail of narrow ovi lines that we would like to identify .
we call this procedure our _ narrow line detection algorithm_. after identifying the narrow feature , we fit it using a voigt profile to obtain its column density and @xmath94-parameter . the identification and fits are shown as the smooth curves in figure [ fig : auto5 ] .
as can be seen there , this algorithm effectively picks out many of the narrow ovi features that survive hi blanketing .
occasionally , narrow lines are detected in the spectrum even without any metals ( top panel of figure [ fig : auto5 ] ) .
these narrow lines arise primarily at velocity caustics , which produce hi lines whose widths are close to those expected from thermal broadening
. figure [ fig : auto5 ] shows , however , that these false detections are significantly fewer in number than the true ovi lines for an oxygen abundance of [ o / h]@xmath97 .
figure [ fig : auto5 ] suggests that narrow lines are seen much more rarely in q1422 ( bottom panel ) than in an artificial spectrum with @xmath73 .
these qualitative impressions are borne out by the statistics we present in the next section .
we apply the narrow line detection algorithm to q1422 between 4025 and 4745 .
we apply the same algorithm over the same interval to the six artificial q1422 spectra , first with no metals , then with a series of uniform metallicity values ranging from @xmath98 to @xmath99 . at each metallicity
we count the number of narrow lines detected .
for q1422 , we detect 11 narrow lines , while in the zero - metallicity case we detect an average of around seven per artificial spectrum .
figure [ fig : ovi ] shows the number of narrow features detected in the artificial spectra as a function of metallicity , quantitatively confirming the inference from figure [ fig : auto5 ] that a metallicity of [ c / h]@xmath89 with a relative overabundance [ o / c ] @xmath68 is strongly inconsistent with the paucity of narrow features in the spectrum of q1422 . at @xmath73 , the expected number of narrow line detections is @xmath100 ( where the @xmath101 error bars reflect the variance over the six artificial spectra ) , which is inconsistent with the q1422 data at @xmath102 level .
we can strengthen this conclusion with an improvement to our narrow line detection algorithm .
the narrow lines we identify as candidate ovi lines should have a doublet companion if they are indeed ovi . while hi blanketing usually prevents actual identification of the doublet companion feature , one can at least require that there be enough absorption at the doublet position to obscure ( or represent ) the companion line . since we do not know which member of the doublet to assign to any given narrow line , the stronger line at 1032 or the weaker line at 1037
, we must examine both doublet wavelength positions ; if neither contains sufficient absorption , we can exclude that line from our list of ovi candidates .
we can apply a similar technique to associated hi absorption . from the lox plot of figure
[ fig : lox ] , one can see that ovi lines are not expected to be detectable for systems with @xmath103 .
below these densities , the amount of oxygen is small and much of it has been ionized to ovii .
this @xmath90 cutoff is for @xmath73 , and it scales with metallicity . using this metallicity - dependent relation , we can examine the spectrum at the two relevant positions to see if there is sufficient absorption to allow a ly@xmath1 line of the appropriate hi column density . if not
, we can exclude that line .
the hi selection criterion removes more potential ovi detections than the doublet companion selection criterion . applying these two additional constraints
, we obtain the results plotted in figure [ fig : ovisel ] .
the number of candidate ovi lines in q1422 has dropped from 11 to 5 at @xmath73 .
the number of lines excluded now depends on the assumed metallicity , even for q1422 , since the associated hi line strength required to exclude a narrow line as ovi depends on the metallicity . a slightly smaller number of lines ( @xmath104 on average ) are excluded from the no - metallicity artificial spectra ( when narrow lines are excluded by requiring @xmath105 , as at @xmath73 ) , making q1422 compatible with having no narrow lines due to ovi
. a slightly larger number of lines are excluded in the spectra with metals . at @xmath73 , however , this larger number still represents a much smaller fraction of the total number of narrow lines detected ; the number of expected ovi detections is now @xmath106 , inconsistent with q1422 by @xmath107 .
the number of narrow lines in the artificial spectra is consistent with that in q1422 only for metallicities @xmath108 . in
[ sec : o / c ] , [ sec : hot ] , and [ sec : cut ] we examine the sensitivity of this result to the various assumptions we have made in our modeling . before that , we examine some physical properties of the narrow lines identified by our algorithm .
we first focus on the artificial spectra with @xmath73 .
even though this model is inconsistent with the q1422 data , it is helpful to examine the properties of its ovi absorbers and associated hi and civ lines in order to reinforce the points made earlier in [ sec : ovisearch ] . from the lox plot of figure
[ fig : lox ] , we claimed that ovi should trace low column density ( and hence low density ) systems . in order to quantify this
, we associate hi lines with each narrow feature in our artificial spectra .
since we can not tell directly from an artificial spectrum whether a feature is the stronger or weaker ovi component , we sum the column densities of nearby hi lines within @xmath109 of the two appropriate wavelengths , then use the greater of the two column densities .
owing to the high density of hi lines with @xmath110 , this is an inexact procedure , as it is difficult to associate a single hi feature with a given ovi line ( even more so than in the case of civ ) .
nevertheless , summing all the lines within @xmath109 ( the typical width of hi features ) provides a reasonable estimate of the hi column density .
we also apply a similar procedure to civ absorbers identified in our artificial spectra .
we plot the histogram of associated hi column densities for ovi ( narrow ) and civ lines in figure [ fig : nhist ] .
ovi lines trace hi with column densities down to @xmath111 ( associated hi lines identified below this column density are probably spurious ) with a maximum sensitivity for systems around @xmath112 .
civ traces systems with hi column densities down to @xmath113 , with a maximum sensitivity around @xmath114 .
thus , as we stated earlier , ovi lines trace lower density regions , and the typical " hi column density traced by ovi absorbers is around @xmath112 .
the gas density in @xmath115 systems is close to the mean cosmic baryon density .
also as we stated earlier , the majority of lines with detectable ovi absorption would have no detectable associated civ absorption . in figure
[ fig : bvsn ] we plot the @xmath94-parameter of ovi - selected narrow lines ( i.e. , those tallied in figure [ fig : ovisel ] ) against the column density ( inferred by assuming that they are hi lines ) . in the upper left panel
we show all the narrow lines in the six artificial spectra with @xmath73 . in the lower left panel
we show the narrow lines identified in the artificial spectra with no metals .
the lower right panel shows the extra " metal lines , i.e. those in the @xmath73 spectra but not in the zero - metallicity artificial spectra .
finally , the upper right panel shows the ovi - selected narrow lines in q1422 .
this figure shows that our narrow line detection algorithm effectively picks out the tail of narrow lines that arise owing to the presence of ovi .
the second panel of figure [ fig : auto5 ] ( the artificial spectrum with only ovi absorption ) shows that many ovi features detectable by eye do not pass our narrow line selection criteria .
still more ovi features are subsumed by hi absorption .
the ovi lines that we do identify tend to be both strong and narrow .
there are many more narrow ovi lines in the @xmath73 artificial spectra than seen in the no - metallicity spectra .
if one further considers only lines with @xmath116 ( dotted line in figure [ fig : bvsn ] ) , there are no lines in the zero metallicity case , while there are many such lines in the @xmath73 case .
the simulation tests therefore imply that any narrow line detected by this algorithm having @xmath117 has a good chance of being a true ovi line , if not an intervening metal line from a low redshift absorption system .
we now examine in detail the four narrow lines found in q1422 , shown in the upper right panel of figure [ fig : bvsn ] .
there is a fifth narrow line , but it has @xmath118 , and it appears to be due to a single cold pixel in q1422 , so we discard it .
two of the remaining lines have @xmath117 , making them strong candidates for ovi , while one line ( the circled point in figure [ fig : bvsn ] ) is the ovi system identified by @xcite . for each of these four systems ,
the doublet position is subsumed by an hi line , so no direct confirmation of ovi absorption is possible .
figures [ fig : plotseg](a - d ) show the four narrow systems identified in q1422 , along with the associated ly@xmath1 , civ1548 , and civ1551 spectral regions . for the various regions
, the solid line corresponds to the assumption that the narrow feature is ovi1032 , while the dashed line assumes ovi1037 absorption .
the other ions have been offset in flux for ease of viewing .
the first system , at 4276.5 , has a @xmath94-parameter of only @xmath119 .
if it is ovi1032 , it has an associated ly@xmath1 line with @xmath120 ( confirmed by the ly@xmath49 component , not shown ) , but no detectable civ absorption ( giving @xmath121 at a @xmath122 level ) .
since it is a very weak line that is just above our @xmath95 detection threshold , there is a possibility that it is an abnormally large noise fluctuation
. it would be interesting to confirm this feature using another spectrum of q1422 + 231 , @xcite . if it is an ovi line , it has @xmath123 , and we can determine the metallicity and oxygen overabundance assuming an @xcite spectrum , the density - column density relation from equation ( [ eqn : rho - nhi ] ) , and the solar abundance values from anders & grevesse ( 1989 ) .
the ionization fractions are then 5.4% for civ , 35% for ovi , and @xmath124 for hi , giving [ o / h ] @xmath89 and [ c / h ] @xmath125 .
thus this system , if ovi , would have a metallicity of @xmath126 , with an overabundance of [ o / c ] @xmath127 .
these values are quite consistent with the metallicity distribution inferred from the statistics of narrow lines
. the second system ( 4278.9 ) has @xmath128 .
with only a modest ly@xmath1 line nearby ( corresponding to this system being ovi1037 ) , and no civ absorption , it seems quite likely that this system is a narrow hi line .
the third system ( 4517.2 ) is a deep , narrow line with @xmath129 .
however , there is a limited amount of corresponding ly@xmath1 absorption , and no civ absorption .
for such a large ovi line ( @xmath130 ) , it is virtually inconceivable that there would be so little ly@xmath1 and no civ absorption .
this contradiction favors an alternative explanation : this is an intervening metal line from a low redshift absorption system .
the fourth and final system is the one presented in @xcite and discussed in [ sec : ovisearch ] .
the associated civ absorption provides compelling evidence that this is indeed an ovi line .
we measure a @xmath94-parameter of @xmath131 , as opposed to @xmath132 from @xcite . since we do not remove known ly@xmath49 absorption before fitting , the @xcite value is probably more accurate for the ovi feature itself .
all in all , of the four lines , only one is confirmed to be ovi ( the @xcite system ) , with another having some likelihood of being ovi .
this detailed analysis emphasizes the paucity of detectable ovi absorption in q1422 .
throughout this paper we have assumed an overabundance of [ o / c ] @xmath68 compared with solar abundance ratios . had we assumed solar abundance ratios , we would have favored the opposite conclusion , that a uniform metallicity of [ c / h ] @xmath89 is ( within uncertainties ) consistent with the ovi data .
this possibility therefore merits further discussion .
the theoretical motivation for oxygen overabundance is the assumption that early star formation produced enrichment patterns characteristic of type ii supernova yields , which have enhanced representation of @xmath1-process elements .
high redshift enrichment almost certainly relies on type ii supernovae ; in a blowout model , both energetic and time constraints require type ii supernovae ( @xcite ) , while in a tidal stripping model ( @xcite ; @xcite ) , there is insufficient time to produce a large number of type i supernovae .
direct evidence for type ii enrichment patterns in the early universe comes from local halo stars , which show an overabundance of oxygen relative to carbon of 35 ( @xcite ) , and there is evidence that the very earliest stars are even more overabundant in @xmath1-process elements ( @xcite )
. there is also mounting evidence for type ii enrichment patterns at high redshift in forming galaxies ( @xcite ) and in damped ly@xmath1 absorbers ( @xcite ) .
thus an overabundance of [ o / c ] @xmath133 for the igm at redshifts
@xmath27 seems a reasonable , and perhaps even somewhat conservative , assumption .
we have also assumed a uniform metallicity . from the civ data ,
one sees that at least some spatial scatter in metallicity is likely .
such a scatter applied to the simulations would serve to _ strengthen _ our conclusions , because it would tend to increase the number of ovi systems in the strong - absorption tail of the distribution , and these stronger lines are more likely to be detected despite coincident hi absorption . for spatial variations in the ionizing background ,
the situation is more complex , but a similar argument is still probably valid as long as there is no systematic softening of the ionizing background above 114 ev .
our simulation assumed ionization equilibrium ( balance between recombination and ionization rates ) at all times when computing ionic abundances .
departures from equilibrium abundances can , under some circumstances , allow substantial heating of the igm during the reionization epoch ( @xcite ) .
the low density gas that produces the ly@xmath1 forest loses energy mainly from adiabatic expansion , so the cooling timescale is comparable to the hubble time .
the relation between gas temperature and density in the igm therefore depends on the assumed reionization epoch and associated heat injection ( @xcite ) .
our equilibrium abundance approach yields the temperature - density relation that arises for minimal heating or a high reionization redshift .
the higher temperatures that arise in plausible alternative reionization scenarios could in principle have a significant effect on our ovi modeling because the ionization fractions are sensitive to temperature .
one example of the effect of different reionization assumptions appears in figure 7 of rauch et al .
( 1997b ) , which shows the difference in the density - temperature relation between a simulation run assuming ionization equilibrium ( @xcite ) and a simulation in which reionization occurred at @xmath134 and the heat input was tracked in a non - equilibrium manner ( @xcite ) .
the latter simulation shows hotter temperatures in the low - density regions , with the discrepancy growing towards lower densities , as expected .
while we are currently unable to perform non - equilibrium simulations of reionization , we can add the difference in the temperatures _ post facto _ to estimate the potential effect of the additional thermal energy .
thermal pressure gradients are small compared to gravitational forces in low density regions , so the dynamics of the gas should be unaffected by higher temperatures , and the _ post facto _ change of the temperature - density relation therefore accounts accurately for the full effect of reionization heating . in support of this claim ,
we note that the distributions of the gas density for the two simulations considered by rauch et al .
( 1997b ) are very similar ( see their figure 6 ) .
the difference in temperatures in the two rauch et al .
( 1997b ) simulations can be approximated as @xmath135}\;\;\;\ ; { \rm for } \;\rho/\bar\rho < 10,\ ] ] where @xmath136 is the density and @xmath137 is the mean baryonic density . there is no difference for overdensities greater than ten , as the cooling time there is sufficiently short to equilibrate the system . by adding this heat to each particle of our simulation
before we construct our artificial spectra , we obtain a set of reionization heated " spectra . for these spectra ,
the factor by which the @xcite photoionizing background spectrum must be multiplied in order to match the mean transmission of q1422 is @xmath138 .
we perform the same civ analysis on the reionization heated spectra to determine the mean metallicity that should be inferred from the observed civ abundance in this hotter igm model .
we find that reionization heating has a negligible effect ( see the dotted line in figure [ fig : metfrac ] ) because most of the civ absorption arises from systems with overdensities greater than ten , translating to a column density @xmath139 ( cf .
[ [ eqn : rho - nhi ] ] ; also see figure [ fig : nhist ] ) . for our ovi analysis ,
the results are somewhat more sensitive to reionization heating because ovi arises in lower density regions .
if the igm is hotter , more oxygen is ionized to ovii .
however , this is countered by the fact that to match the mean transmission of q1422 in the ly@xmath1 forest region , a hotter igm requires a lower ionizing background intensity .
a hotter igm also means that the tail of ovi lines below a given @xmath94-parameter is smaller , reducing the efficiency of our narrow line search procedure for detecting ovi and therefore increasing the ovi abundance inferred from a given number of detections .
this last effect turns out to be minimal ; the dominant effects are the first two , with the first being somewhat more important .
the results are shown in figure [ fig : ovihot ] ( solid line ) , indicating that the maximum metallicity allowed for agreement with the ovi data is around @xmath140 instead of @xmath141 .
so , in the final analysis , the effect of reionization heating does not significantly change our overall conclusion that a uniform metallicity of @xmath73 is inconsistent with the ovi data .
a more dramatic effect arises if the ionizing background contains many fewer photons capable of ionizing oxygen to ovi .
such a scenario might arise if , for example , helium reionization has not occurred by @xmath25 .
we consider an idealized model of such a scenario in this section . to test the sensitivity of our analysis to the softness of the assumed ionizing background , we generate a new @xmath63 with the shape from @xcite , except that the intensity is reduced by a factor of ten above 4 ry ; we refer to this as the _ cut _ @xmath63 .
we generate a new set of artificial spectra using this cut @xmath63 ; the mean transmission is virtually identical to that of the original spectra because these high energy photons do not significantly affect the hi opacity . with the cut @xmath63
, civ is much more common , and we find the mean metallicity for agreement with civ data to be [ c / h ] @xmath142 , as shown by the dashed line in figure [ fig : metfrac ] .
when we apply the same narrow line detection procedure to spectra generated with the cut @xmath63 , we obtain the results presented in figure [ fig : ovicut ] .
this shows that a metallicity of @xmath143 is in agreement with the ovi data .
the cut @xmath63 scenario therefore implies a mean metallicity in the ly@xmath1 forest of around @xmath144 . with the assumed [ o / c ]
@xmath145 , this scenario is consistent with a uniform metallicity ; however , we shall show in [ sec : pixmet ] that a cut @xmath63 also requires a much higher oxygen overabundance , thus a metallicity gradient is still required .
there is some evidence that helium reionization occurs around @xmath146 , from a sudden drop in the siiv / civ ratio near that redshift ( @xcite ; @xcite ) .
this evidence is controversial , as other studies have not detected such a drop ( @xcite ) , though @xcite shows that it appears with several different methods of analysis .
if many more ovi lines were detected in quasar spectra at @xmath147 , this would provide an interesting argument in favor of the helium reionization scenario .
while q1422 is poorly suited for this investigation , we hope that in the future we can examine other quasar spectra over a wider range of redshifts .
figure [ fig : metfrac ] implies that an increase in ovi due to a change in the ionizing background shape should be accompanied by an increase in civ / hi ratios .
one piece of evidence arguing that helium _ has _ been reionized by this epoch is the measurement of the heii opacity at @xmath148 by @xcite . from the strength of a heii discontinuity about 5.3 from quasar q0302 - 003 , they find @xmath149 ( @xmath150 errors ) .
the @xcite spectrum we use yields @xmath151 at @xmath152 , while the cut @xmath63 spectrum yields @xmath153 , well outside the quoted uncertainty of the observed optical depth .
however , the observation of patchy heii absorption at @xmath154 by @xcite suggests that heii reionization does occur near this redshift , and that the @xcite measurement in the immediate vicinity of q0302 might provide a poor estimate of the overall mean heii absorption .
a more secure determination of @xmath155 at @xmath27 may have to await the deployment of the cosmic origins spectrograph on the hubble space telescope .
our narrow line detection results are consistent with either the metallicity being lower in lower density regions or the ionizing background being significantly truncated above 4 ry ( assuming a reasonable overabundance of oxygen relative to carbon ) . in order to distinguish between these two scenarios ,
we apply a different algorithm for detecting ovi that was developed by @xcite . in this algorithm ,
no metal line identification is done ; rather , the presence of metals is assessed statistically by a pixel - by - pixel search for excess optical depth at the expected position of associated absorption . in this section
we describe this _ optical depth ratio _ technique . using this technique , we search for ovi absorption associated with all regions of space showing significant civ absorption . for all pixels redwards of the quasar s ly@xmath1 emission peak , we first identify pixels that have @xmath156 . from these
we select pixels consistent with absorption arising from the weak doublet component of civ ( 1551 ) . specifically
, we only accept a pixel if the optical depth at the strong doublet component s position ( 1548 ) is at least double the optical depth at the weak component s position ( with some allowance for noise ) .
the final subset of pixels represents all regions along the line of sight to the quasar showing significant civ absorption . for q1422 ,
376 out of a total of roughly 25,000 pixels redwards of the ly@xmath1 emission peak are selected as having significant civ absorption .
for each of the selected pixels , we find the optical depth at the associated ovi position and calculate the apparent column density ratio " : @xmath157 where @xmath158 is the optical depth , @xmath159 is the rest wavelength , and @xmath160 is the oscillator strength for each ion .
@xmath161 is not the true ovi column density because there is a significant contribution from coincident hi absorption .
although the lox plot of figure [ fig : lox ] shows that the detectability of civ and ovi peaks at different hi column densities , it also shows that the weaker ( and therefore more common ) civ lines , associated with @xmath162 , should have significant amounts of photoionized ovi .
this is seen when we apply the optical depth ratio technique to artificial spectra with oxygen ( [ o / h ] @xmath97 ) and without oxygen , the latter being used to assess the contribution from coincident hi absorption .
the results are shown in figure [ fig : pixmet ] , as a histogram of @xmath163 ( this is similar to figures 20e - g from @xcite , except that @xcite uses randomly selected regions of the observed spectrum instead of a zero - metallicity model spectrum as the `` control '' sample ) .
there is a significant excess of ovi seen in the [ o / h ] @xmath97 spectra ( solid line ) as compared to the spectra without no oxygen ( dotted line ) ; this is mostly due to photoionized ovi associated with civ absorbers , as the covering fraction of collisionally ionized ovi is quite small .
we also apply the technique to q1422 , and find a significant presence of ovi , consistent with @xcite . the q1422 results ( dashed histogram in figure [ fig : pixmet ] )
are seen to be in better agreement with the @xmath73 spectra than with the zero - metallicity spectra .
we will quantify these results further in the next section .
our conclusion from optical depth ratios that [ o / h ] @xmath8 may seem inconsistent with our earlier conclusion from narrow lines that [ o / h ] @xmath164 , but it is not , _ provided _ that the igm metallicity is lower at lower gas densities . with the narrow line detection algorithm
, we are examining _ all _ regions of space for ovi absorption , so the measure is weighted towards the low - density regions that occupy most of the volume .
conversely , the optical depth ratio technique examines regions of space selected to have significant civ absorption , which are typically higher density regions .
in fact , from [ sec : civ ] we _ know _ the metallicity in these regions is @xmath165 in the full hm spectrum case , or @xmath144 in the case of our cut @xmath63 spectrum ( cf . [ sec : cut ] ) .
thus , we can use the detection of ovi in these regions as a probe of the shape of the ionizing background and the assumed oxygen overabundance , as we will show in the next section . for regions that have detectable civ absorption , the optical depth ratio technique is significantly more sensitive to the presence of ovi than the narrow line detection algorithm .
because most ovi positions are contaminated by hi absorption , the opportunity to detect a narrow ovi line associated with civ is very rare .
the optical depth ratio technique can still achieve a statistical detection of ovi in such cases because it examines only the precise locations of expected ovi absorption and compares their mean optical depth to an ovi - free control sample .
we attempted to quantify the presence of ovi associated with ly@xmath1 absorption of a given optical depth , but were unsuccessful .
the optical depth ratio technique is much less reliable when used with hi rather than civ , since the typical line widths of hi features are significantly different from those of civ or ovi ; it is only because ovi and civ have similar thermal broadening ( and identical bulk flow broadening because we are examing the same physical region ) that optical depths at associated wavelengths can be meaningfully compared .
we now apply the optical depth ratio technique to artificial spectra with varying metallicity . for the artificial spectra , the assumed [ c / h ]
is held fixed at the value that produces agreement with the civ data ( @xmath92 for the @xcite spectrum and @xmath166 for the cut spectrum ) , and we vary only [ o / h ] ( or equivalently , the oxygen overabundance ) .
we plot the median value of @xmath167 , since the median provides the most robust characteristic quantity in this highly skewed distribution .
the median @xmath168 for q1422 , and @xmath169 for the artificial spectra without oxygen .
these values are shown as the horizontal dashed and dotted lines in figure [ fig : zpix ] , respectively .
the varying [ o / h ] case is shown by the solid line , with the error bars as usual computed over the six artificial spectra ( the error bars on the other artificial spectra measurements are comparable ) .
agreement with q1422 is seen to occur at [ o / h ] @xmath170 , and given that the metallicity of these regions is [ c / h ] @xmath3 , this implies [ o / c ] @xmath24 .
thus for the case of the hm spectrum , we have independently determined that our assumed oxygen overabundance is consistent with q1422 .
we now examine the cut @xmath63 spectrum case , as shown by the dot - dashed line in figure [ fig : zpix ] . with a softer spectrum ,
ovi is much less abundant and civ is more abundant , meaning that a significantly higher overabundance of oxygen is required to reach agreement with the q1422 data for ovi associated with civ regions .
agreement is reached with q1422 s median @xmath171 for [ o / h ] @xmath172 , which when combined with [ c / h ] @xmath142 for the cut spectrum implies [ o / c ] @xmath173 .
a factor of @xmath174 oxygen overabundance seems fairly implausible , as no set of objects has ever been observed to have @xmath1-process overabundances approaching this level .
we conclude that the regions showing civ absorption are in much better agreement with a @xcite spectrum with no softening above 4 ry , and therefore that most helium is doubly ionized by redshift @xmath23 .
this scenario is also consistent with an oxygen overabundance of [ o / c ] @xmath24 .
free we can assess the presence of a gradient of metallicity with density , free from assumptions about the oxygen overabundance , by comparing [ o / h ] from the narrow associated with civ regions with the limits on [ o / h ] in lower density gas from the narrow line detection algorithm . for the @xcite spectrum case , civ regions have [ o / h ] @xmath8 , while the narrow line search [ o / h ] @xmath175 , implying at least a factor of 3 decrease in metallicity at lower densities . for the cut @xmath63 spectrum ,
civ regions have [ o / h ] @xmath172 , whereas the narrow line search yields [ o / h ] @xmath125 , implying an even _ greater _ metallicity gradient .
thus regardless of the assumed ionizing background , a metallicity gradient is required to match the observed amount of ovi from both detection methods , barring the unlikely scenarios that the metagalactic ionizing flux is significantly _ harder _ than the @xcite spectrum , or the oxygen overabundance is significantly lower in the lower density regions . as mentioned before
, @xcite detected a significant jump in siiv / civ ratio above @xmath146 , that when combined with the siiv / civ ratio above @xmath146 , which , when combined with photoionization models , is consistent with an ionizing spectrum heavily truncated ( by a factor @xmath174 ) above 4 ry .
this inference is in direct contradiction with the results of the previous section .
one way to reconcile these results may be to invoke an igm that has a patchy ionization structure .
this type of structure is implied by the heii measurements of @xcite , as stated in [ sec : cut ] , and has been suggested by @xcite to explain the significant presence of ovi detected at @xmath27 . in this section
we consider a simplistic , `` 50/50 '' model of such a patchy igm to determine the implications for our ovi detections .
consider a universe where half the volume has already reionized helium ( and hence has an @xcite spectrum ) , and half the volume has a significantly truncated ionizing flux above 4 ry ( as in our cut @xmath63 spectrum ) .
civ absorption then implies a mean metallicity of [ c / h ] @xmath176 .
this value can be determined from figure [ fig : metfrac ] , as this is the metallicity where the solid ( @xcite spectrum ) and dashed ( cut spectrum ) curves are roughly equally above and below the observed @xmath177 of q1422 .
the additional scatter introduced by the varying ionization conditions implies that there can be virtually no intrinsic spatial scatter in the metallicity of civ absorbers .
similarly , bisecting the solid and dot - dashed curves in figure [ fig : zpix ] implies agreement with the q1422 optical depth ratio results at [ o / h]@xmath178 . taken together ,
the [ c / h ] and [ o / h ] constraints imply [ o / c ] @xmath179 in the 50/50 model .
this overabundance is perhaps marginally plausible if one invokes enrichment almost entirely from high - mass stars yielding a high fraction of @xmath1-process elements .
the narrow line detection results become statistically consistent with the q1422 data only for [ o / h ] @xmath175 ( metallicity @xmath180 ) , as seen by bisecting the solid and dot - dashed curves in figure [ fig : ovicut ] .
thus , a metallicity gradient is still required in the 50/50 model , as the optical depth ratio technique ( tracing civ regions only ) yields [ o / h ] higher by @xmath181 dex than the narrow line detection technique ( tracing mostly lower density regions ) . while the 50/50 patchy igm model is marginally consistent with the ovi data , it remains to be seen whether such a scenario is consistent with the observations of siiv / civ by @xcite .
currently , with only one quasar spectrum available to us , we do not have sufficient statistics to investigate siiv in detail ; we plan to conduct these investigations when more data become available , preferably at @xmath147 as well as at @xmath27 .
we present a systematic search for ovi absorption in the spectrum of q1422 + 231 ( @xmath0 ) , using a narrow line detection algorithm proven effective at identifying ovi absorption in artificial spectra , and an optical depth ratio technique introduced by @xcite .
the first technique traces ovi predominantly in systems with @xmath182 , whereas the second technique traces only ovi associated with civ absorption , in @xmath183 systems . by comparing q1422 and artificial spectra having varying metallicities
, we determine that 1 .
[ o / h ] must be lower in lower density regions , for either an @xcite ionizing background or an ionizing background significantly truncated above 4 ry .
if [ o / c ] is constant in systems up to @xmath184 , our results imply that regions traced by @xmath30 systems ( corresponding to gas at roughly the mean baryonic density ) have a mean metallicity lower by at least a factor of 3 compared to regions traced by @xmath162 systems ( corresponding to a baryonic overdensity of @xmath29 ) .
2 . more than half the universe must have helium reionized by @xmath25 .
if helium has completely reionized by @xmath23 ( the highest redshift probed by the q1422 data ) , then our analysis implies [ o / c ] @xmath24 , in good agreement with overabundance measurements of type ii supernovae enriched systems .
if a significant portion of the universe has not reionized helium by @xmath25 and therefore has a softer ionizing background spectrum , then the required oxygen overabundance is higher . for example , if half of the volume has not reionized helium , then [ o / c ] @xmath179 , already greater than the observed overabundance of any class of type ii supernovae enriched systems . if the spectrum were soft throughout the universe at @xmath27 then an implausibly high overabundance , [ o / c ] @xmath173 , would be required .
these conclusions are in good agreement with the recent study of @xcite , who used composite spectra to investigate civ absorption in systems with @xmath185 and found that the metallicity of these absorbers must be [ c / h ] @xmath186 .
cosmological simulations show that @xmath187 roughly corresponds to the dividing line between overdense and underdense regions of the universe ( though the value of @xmath90 that marks this division depends on redshift and , to a lesser extent , on cosmological parameters ) .
our results therefore imply that mildly overdense regions such as filaments and sheets have been enriched , while underdense regions are virtually chemically pristine .
simulations that self - consistently enrich the igm by tracking metal production and transport find that a strong metallicity gradient is predicted between the mildly overdense and underdense regions ( see figure 3 in @xcite ) ; this predicted gradient is in good agreement with the @xcite data and with the scenario we present above .
recent measurements of a jump in the siiv / civ ratio around @xmath25 ( @xcite ) may be difficult to reconcile with conclusion ( 2 ) above .
while we have yet to conduct a systematic comparison of siiv in observed and artificial spectra , primarily because of the small numbers of siiv systems detectable in our one available quasar spectrum , we expect our results will be in agreement with @xcite , who argues that such a jump requires a much softer ionizing background at @xmath27 . one way to reconcile these results may be to invoke patchy helium reionization at that epoch , as suggested by @xcite ; such a model may be tested in greater detail by searching for an anti - correlation between ovi and siiv detections .
if helium reionization occurs around @xmath25 , our narrow line algorithm should yield many more ovi detections at redshifts @xmath188 .
such searches are difficult because of the poor blue sensitivity of the hires spectrograph ( and complete loss of sensitivity at @xmath189 ) , but quasar spectra do exist that could provide constraints down to @xmath190 . the presence of a substantial number of ovi lines in this regime would strongly favor the late helium reionization scenario ; there is already some weak evidence that ovi is more abundant at @xmath147 ( @xcite ) .
if ovi features continue to be virtually undetectable down to @xmath190 , this would be compelling evidence against the late helium reionization scenario , since the heii absorption measurements of davidsen , kriss , & zheng ( 1996 ) imply that helium has been reionized by this redshift .
we hope to work with observers to attempt this search in the near future . in a broader context ,
our work illustrates the power of combining cosmological hydrodynamic simulations of structure formation with high - quality quasar spectra to infer the ionization state and the enrichment history of the high - redshift igm .
future observations and simulations promise a wealth of information , which , when combined , will help us to better understand the evolution of the igm and its connection to early star formation and the epoch of primeval galaxies .
we are very grateful to antoinette songaila and len cowie for making their keck quasar spectrum of q1422 + 231 available to us .
we thank eric linder and antoinette songaila for helpful discussions .
we also thank francesco haardt and piero madau for providing us with their latest ionizing background in electronic form .
uh acknowledges support from a postdoctoral research grant from the danish natural sciences research council . this work was supported in part by the psc , ncsa and sdsc supercomputing centers , by nasa theory grants nagw-2422 ,
nagw-2523 , nag5 - 2882 , nag5 - 3111 , and nag5 - 3820 , by nasa ltsa grant nag5 - 3525 , by nasa hpcc / ess grant nag 5 - 2213 , and by the nsf under grants ats90 - 18256 and asc 93 - 18185 .
finally , we acknowledge the essential contribution of s. vogt and the hires team in building the instrument that made this kind of investigation possible .
anders , e. & grevesse , n. 1989 , geochim .
acta , 53 , 197 boksenberg , a. 1998 , to appear in structure and evolution of the igm from qso absorption line systems , proc .
13th iap colloquium , eds .
p. petitjean and s. charlot , nouvelles frontieres , paris , astro - ph/9710258 burles , s. & tytler , d. 1996 , , 460 , 584 burles , s. & tytler , d. 1998a , , in press , astro - ph/9712108 burles , s. & tytler , d. 1998b , , submitted , astro - ph/9712109 cen , r. , miralda - escud , j. , ostriker , j.p .
, & rauch m. 1994 , , 427 , l9 cowie , l. l. , songaila , a. , kim , t .- s . &
hu , e. m. 1995 , , 109 , 1522 fontana , a. , cowie , l. l. , & songaila , a. 1998 , nature , submitted dav , r. 1998 , to appear in structure and evolution of the igm from qso absorption line systems , proc .
13th iap colloquium , eds .
p. petitjean and s. charlot , nouvelles frontieres , paris dav , r. , hernquist , l. , weinberg , d.h . & katz , n. 1997 , , 477 , 21 dav , r. , dubinski , j. & hernquist , l. 1997 , newast , 2 , 277 davidsen , a. f. , kriss , g. a. , & zheng , w. 1996 , nature , 380 , 47 edvardsson , b. , andersen , j. , gustafsson , b. , lambert , d. l. , nissen , p. e. & tomkin , j. 1993 , , 275 , 101 ferland , g.j .
1996 , univ . of kentucky ,
dept . of astronomy , internal report gnedin , n. y. & ostriker , j. p. 1997
, , 486 , 581 gnedin , n. y. 1998 , , accepted , astro - ph/9709224 haardt , f. & madau , p. 1996
, , 461 , 20 ( @xcite ) haehnelt , m. g. , steinmetz , m. & rauch m. 1996 , , 465 , l65 haiman , z. & loeb , a. 1997 , , 483 , 21 hellsten , u. , dav , r. , hernquist , l. , weinberg , d.h . & katz , n. 1997 , , 487 , 482 ( @xcite ) hellsten , u. , hernquist , l. , katz , n. & weinberg , d.h .
1998 , , in press , astro - ph/9708090 ( @xcite ) hernquist , l. & katz , n. 1989 , , 70 , 419 hernquist , l. , katz , n. , weinberg , d.h . , & miralda - escud , j. 1996 , , 457 , l51 hogan , c. j. , anderson , s. f. & rugers , m. h. 1997 , , 113 , 1495 hu , e.m . ,
kim , t.s . , cowie , l.l . , songaila , a. , & rauch , m. 1995 , 110 , 1526 hui , l. , & gnedin , n. 1997 , , 292 , 27 hu , w. , & sugiyama , n. 1996 , , 471 , 542 katz , n. & quinn , t. 1995 , tipsy manual katz , n. , weinberg d.h . , & hernquist , l. 1996 , , 105 , 19 kim , t.s . , hu , e.m . ,
cowie , l.l . , & songaila , a. 1997 , accepted , astro - ph/9704184 kirkman , d. & tytler , d. 1997 , , 489 , l123 ( @xcite ) kirkman , d. & tytler , d. 1998 , , accepted , astro - ph/9701209 lu , l. 1991 , , 379 , 99 lu , l. , sargent , w. l. w. , barlow , t. a. , churchill , c. w. , & vogt , s. s. 1996a , , 107 , 475 lu , l. , sargent , w. l. w. , womble , d. s. , & takada - hidai , m. 1996b , , 472 , 509 lu , l. , sargent , w.l.w . ,
barlow , t.a .
, & rauch , m. 1998 , , submitted , astro - ph/9802189 lynds , r. 1971 , , 164 , l73 mcwilliam , a. 1997 , , 35 , 503 miralda - escud j. , & rees , m. j. 1994 , , 266 , 343 miralda - escud , j. , cen , r. , ostriker , j.p . , & rauch , m. 1996 , , 471 , 582 miralda - escud , j. & rees , m. 1997 , , 478 , l57 pettini , m. , lipman , k. & hunstead , r. w. 1995 , , 451 , 100 rauch , m. , haehnelt , m. g. , & steinmetz , m. 1997a , , 481 , 601 rauch , m. , miralda - escud , j. , sargent , w.l.w . ,
barlow , t.a.,hernquist , l. , weinberg d.h . ,
katz , n. , cen , r. , ostriker , j.p .
1997b , , 489 , 7 reimers , d. , khler , s. , wisotzki , l. , groote , d. , rodriguez - pascual , p. & wamsteker , w. 1997 , , 327 , 890 sargent , w.l.w . ,
young , p.j . , boksenberg , a. & tytler , d. 1980 , , 42 , 41 songaila , a. & cowie , l. l. 1996 , , 112 , 335 ( @xcite ) songaila , a. 1998 , , accepted , astroph/9803010 ( @xcite ) tytler , d. , fan , x.m . ,
burles , s. , cotrell , l. , davis , c. , kirkman , d. & zuo , l. 1995 , in qso absorption lines , ed . g. meylan ( springer - verlag ) , 289 vogt , s. s. , et al . 1994 , spie , 2198 , 326 weinberg , d. h. , hernquist , l. , & katz , n. 1996 , , 477 , 8 weinberg , d. h. , miralda - escud , j. , hernquist , l. & katz , n. 1997 , , 490 , 564 weinberg , d.h .
, hernquist , l. , katz , n. , croft , r. & miralda - escude , j. 1998 , in proc . of the 13th iap colloquium , structure and evolution of the igm from qso absorption line systems , eds .
p. petitjean & s. charlot , ( paris : nouvelles frontires ) , astro - ph/9709303 zhang , y. , anninos , p. & norman , m.l .
1995 , , 453 , l57 | we present a systematic search for ovi(1032,1037 ) absorption in a keck hires spectrum of the @xmath0 quasar q1422 + 231 , with the goal of constraining the metallicity and ionization state of the low density intergalactic medium ( igm ) .
comparison of civ absorption measurements to models of the ly@xmath1 forest based on cosmological simulations shows that absorbers with @xmath2 have a mean carbon abundance [ c / h ] @xmath3 , assuming a metagalactic photoionizing background with the spectral shape predicted by haardt & madau ( 1996 , hm ) . in these models , lower column density absorption arises in lower density gas where most civ is photoionized to cv .
therefore , ovi should be the most sensitive tracer of metallicity in ly@xmath1 absorbers with @xmath4 .
ovi lines lie at wavelengths heavily contaminated by lyman series absorption , so we interpret the search results by comparing to carefully constructed , mock q1422 spectra drawn from a hydrodynamic simulation of a @xmath5-dominated cold dark matter model . a search for deep , narrow absorption features yields only a few candidate ovi lines in the spectrum of q1422 .
hi absorption blankets the position of the doublet companion line in each case , and the total number of narrow lines is statistically consistent with that in zero - metallicity artificial spectra .
artificial spectra generated with the hm background and [ o / h ] @xmath6 predict too many narrow lines and are statistically inconsistent with the data .
we also search for ovi associated with civ systems , using the optical depth ratio technique of songaila ( 1998 ) . with this method
we _ do _ find significant ovi absorption ; matching the data requires [ o / c ] @xmath7 and corresponding [ o / h ] @xmath8 .
taken together , the narrow line and optical depth ratio results imply that ( a ) the metallicity in the low density regions of the igm is at least a factor of three below that in the overdense regions where civ absorption is detectable , and ( b ) oxygen is overabundant in the civ regions , consistent with the predictions of type ii supernova enrichment models and the observed abundance pattern in old halo stars . the photoionizing background spectrum would be truncated above 4 ry in regions that have not undergone helium reionization ( heii@xmath9heiii ) , and in this case matching the q1422 data requires lower [ c / h ] but higher [ o / h ] . taking [ o / c]@xmath10 as the maximum plausible overabundance of oxygen ,
we conclude that helium must have been reionized through at least 50% of the volume from @xmath11 . |
after the discovery of fractional quantum hall effect,@xcite , we realized that new kind of orders beyond landau s symmetry breaking paradigm is possible .
this new kind order is called topological order @xcite for gapped states and quantum order@xcite for general states .
the new orders reflect patterns of long range entanglements in the ground state . gapped @xmath0 spin liquids have the simplest kind of topological order
@xmath0 topological order.@xcite those topological ordered states may appear in frustrated spin systems or dimmer models.@xcite physically , the topological orders can be ( partially ) characterized by robust ground - state degeneracy.@xcite the low energy effective theory for those @xmath0 topologically ordered states is a @xmath0 gauge theory .
topological order is a property of a many - body ground state that is robust against any perturbations , even those perturbations that break all the symmetries . in this paper , we like to study the interplay between topological order and symmetry .
we like to find out how to characterize topological ordered states that also have certain symmetries .
recently , it was found that for spin liquids with all the lattice symmetries ( such as lattice translation and rotation symmetry ) , there can be hundreds different @xmath0 topological orders.@xcite we will call those topological orders symmetric topological orders .
it is shown that the different symmetric @xmath0 topological orders can be characterized by different project symmetry groups ( psg ) .
so those symmetric topological orders are good examples to study the relation between topological order and symmetry . here
, we would like to study the low energy effective theories for those different @xmath0 topological orders and ask how different symmetric @xmath0 topological orders are reflected in low energy effective theories .
we find that all different @xmath0 topological orders can be described by the same effective mutual @xmath1 chern - simons ( cs ) theories.@xcite the lattice symmetry is realized projectively in the effective mutual cs theories .
it turns out that different symmetric @xmath0 topological orders have different projective realizations of the lattice symmetries . to confirm our results , the projective construction ( the slave - particle theory ) @xcite
is used to calculate the ground - state degeneracies , the ground - state quantum numbers , and the psgs of quasi - particles .
those results agree with those obtained from the effective mutual cs theories .
furthermore , we also used the effective mutual cs theories to study gapless edge states for some @xmath0 topologically ordered states .
the key to understand topological orders is to construct states that can have long range quantum entanglements .
the projective construction introduced in the study of high @xmath2 superconductors is a powerful way to construct such states.@xcite in this section , we will briefly review the projective construction of @xmath0 topologically ordered states .
a spin-1/2 model can be viewed as a hard - core - boson model , if we identify @xmath3 state as a zero - boson state @xmath4 and @xmath5 state as a one - boson state @xmath6 . in the follow
we will use the boson - picture to describe our model .
we first introduce a `` mean - field '' fermion hamiltonian:@xcite @xmath7 where @xmath8 .
we will use @xmath9 and @xmath10 to denote the @xmath11 complex matrices whose elements are @xmath12 and @xmath13 .
let @xmath14 be the ground state of the above free fermion hamiltonian ( the lowest energy state obtained by filling all the negative energy levels ) .
then a many - boson wave function can be obtained through @xmath15 where @xmath16 is the number of lattice sites , @xmath17 and @xmath18 , @xmath19 , @xmath20 , label the location of bosons ( up - spins ) . here
, we have assumed that there are @xmath21 up - spins and @xmath22 down - spins .
we may view @xmath23 as variational parameters and the physical spin wave function @xmath24 as a trial wave function .
the trial ground state of a spin hamiltonian can be obtained by minimizing the average energy @xmath25 .
first let us consider the following spin hamiltonian @xmath26 where @xmath27 are the pauli matrices and @xmath28 labels the site of a square lattice .
we find that if we choose the variational parameters to be @xmath29 then the spin wave function minimize the average energy .
in fact the wave function is the exact ground state of hamiltonian @xmath30 .
@xcite it was found that all the excitations above the ground state are gapped and the ground state contains a non - trivial topological order described by a @xmath0 effective gauge theory .
we will call such a state z2e state . ref .
introduced another many - spin state on square lattice which is described by @xmath31 and @xmath32 .
however , it is not clear what kind of spin hamiltonian gives rise to the spin state described by the above variational parameters . despite of this
, some physical properties of the spin state were obtained under the assumptions that the state is stable for a certain local spin hamiltonian.@xcite again , all excitations above the spin state have finite energy gaps .
the spin state is a spin liquid with no spin order .
but it contains a non - trivial topological order described by an effective @xmath0 gauge theory .
so we will call such a spin state z2a state .
naively , one may expect the z2a and the z2e states to be the same state since both have @xmath0 gauge theory as their low energy effective theory . in the following
, we will show that they are different quantum states with different topological orders .
line and the @xmath33 line get an additional minus sign . ]
one way to study a topological order is to study its ground state degeneracy on a torus .
naively , we expect the z2a and the z2e state to have 4 degenerate ground states , as implied by the effective @xmath0 gauge theory .
the argument goes as the following .
first , we note that the physical boson wave function @xmath34 is invariant under the following @xmath35 gauge transformations@xcite @xmath36 where @xmath37 .
so the average energy @xmath38 satisfies @xmath39 next we assume @xmath40 give rise to a ( variational ) ground state of a hamiltonian .
we would like to show that the following four ansatz @xmath41 produce four degenerate ground states . here
@xmath43 and @xmath44 have values @xmath45 or @xmath46 .
@xmath47 if the link @xmath48 crosses the @xmath49 line ( see fig . [ xyline ] ) and @xmath50 otherwise . similarly , @xmath51 if the link @xmath48 crosses the @xmath33 line and @xmath52 otherwise .
physically , the degenerate states arise from adding @xmath53 flux through the two holes of the torus .
the values of @xmath42 reflect the presence or the absence of the @xmath53 flux in the two holes .
we note that @xmath54 represents the ground state .
we also note that @xmath55 with different @xmath56 and @xmath57 are _ locally _ gauge equivalent .
this is because , on an infinite system , the change , say , @xmath58 can be generated by an @xmath35 gauge transformation @xmath59 , where @xmath60 , and @xmath61 if @xmath62 and @xmath63 if @xmath64 . as a result , @xmath65 . on the other hand , on a torus , @xmath66 with different @xmath56 and @xmath57
are not gauge equivalent in the global sense .
there is no @xmath35 gauge transformation defined on the torus that connects those ansatz .
so the four ansatz give rise to four different degenerate states .
this is how we obtain the four - fold ground state degeneracy for the @xmath0 states .
however , the above argument is valid only for even by even lattice . for odd by odd lattice ,
the argument breaks down . to understand the failure of the above argument , let us construct the mean - field ground state more carefully .
let us start with a simple case of the z2a state .
for the ansatz , the `` mean - field '' hamiltonian in momentum space becomes @xmath67 where @xmath68 and @xmath69 here @xmath70 and @xmath71 are diagonalized quasiparticles operators @xmath72 where @xmath73 and @xmath74 are the functions of @xmath75 and @xmath76 the mean - field ground state is obtained by filling all the negative levels and is given by @xmath77 where the state @xmath78 is defined through @xmath79 .
( note that all the particles @xmath70 has positive energy and all the particles @xmath71 has negative energy . )
since @xmath80 is linear combination of @xmath81 and @xmath82 and there are @xmath83 different @xmath84-levels , the mean - field state @xmath85 contains @xmath83 number of fermions . here
@xmath86 are sizes of the lattice in the @xmath87- and @xmath33-directions .
clearly , when both @xmath88 and @xmath89 are odd , @xmath85 contains an odd number of fermions .
such a mean - field state does not correspond to any physical spin state since the corresponding spin wave function vanishes .
( note that is a projection to the subspace with 0 or 2 fermions per site . ) to get a non - zero physical spin wave function we need to start with a mean - field state with one extra fermion in the empty @xmath90-band ( or a hole in the filled @xmath91-band ) .
but by choosing different states for the extra fermion ( or the hole ) , we can obtain many different spin wave functions which are nearly degenerate .
so when both @xmath88 and @xmath89 are odd , the excitations in the z2a state are gapless , or we may say that the z2a state has infinite degeneracy .
physically , the z2a state on odd by odd lattice always contains a unpaired spinon .
the different states of the unpaired spinon gives rise to the infinite degeneracy .
when one of @xmath86 is even , the mean - field state @xmath92 gives rise to a non - zero physical spin state .
there is no unpaired spinon and the excitations are gaped .
each ansatz @xmath93 produces a single physical spin state and the z2a state has four - fold degeneracy on a torus with an even number of lattice sites .
because the spin hamiltonian is translation invariant , the ground states carry definite crystal momentum .
to calculate the crystal momentum , we note that in the @xmath94 sector described by the ansatz @xmath95 , the fermion wave function satisfies the periodic boundary condition .
so @xmath96 are quantized as @xmath97 where @xmath98 are integers . and the spin state produced by the ansatz @xmath95 has the following crystal momentum : @xmath99 we would like to point out that the above crystal momentum is actually the crystal momentum of the mean - field state .
however , the even - fermion - per - site projection commutes with the translation operator , and thus the crystal momentum is unchanged by projection . when @xmath56 and/or @xmath57 are equal to 1 , the fermion wave function is antiperiodic in the @xmath33- and/or @xmath87-directions . in the case , @xmath100 and/or @xmath101 are quantized as @xmath102 and/or @xmath103 .
the crystal momentum of the spin state produce by the ansatz @xmath93 can be calculated in the similar fashion .
for example in the @xmath104 sector , the crystal momentum is given by @xmath105 the results are summarized in the table [ z2acm ] .
.crystal momenta @xmath106 of the four ground states , @xmath107= ( 0,0 ) , ( 0,1 ) , ( 1,0 ) , ( 1,1 ) , of the z2a spin liquid on three different lattices , @xmath108 ( even , even ) , ( even , odd ) , ( odd , even ) .
[ cols="^,^,^,^,^",options="header " , ]
in the above sections we have calculated the topological properties for the z2a and the z2e states . due to their different topological properties , we find that the two states have different topological orders .
then an important issue is to find the low energy effective theories that describe the two different topological orders .
we find that a mutual @xmath110 cs theory with different projective realizations of the lattice symmetry can describe the two kind of topological orders .
we reach the conclusion by comparing the topological properties of the mutual @xmath109 cs theory with those of the z2a and the z2e states .
all the topological properties , including topological degeneracy , quantum numbers , and edge states , agree , indicating the equivalence between the @xmath0 topological states on lattice and the mutual @xmath109 cs theory .
first we introduce the lagrangian for the mutual @xmath109 cs theory : @xmath111 where @xmath112 is the gauge field strength for gauge field @xmath113 and @xmath114 is the gauge field strength for gauge field @xmath115 .
the excitations are described by the currents which are defined as @xmath116 and @xmath117 .
the gauge charges of @xmath118 and @xmath119 are quantized as integers . from the equation motions for @xmath120 and @xmath121
@xmath122 we find that a charge for gauge field @xmath123 induces flux of gauge field @xmath124 . as a result ,
the charge for gauge field @xmath125 and the charge for gauge field @xmath124 have a semionic mutual statistics .
that is , moving an @xmath123-charge around an @xmath126-charge generates a phase @xmath53 .
this catches the key topological property for the @xmath0 spin liquid .
it is well known that the @xmath0 spin liquid states contain @xmath0 vortex and @xmath0 charge excitations .
and the @xmath127 vortex and the @xmath0 charge have semionic mutual statistics between them .
so we will propose that the mutual chern - simons theory [ mcsl ] describes a @xmath0 gauge theory .
the @xmath123-charge can be identified as the @xmath0 charge and the @xmath124-charge as the @xmath0 vortex .
furthermore , the energy gap for both of the gauge fields comes from the mutual cs term @xmath128 the mutual @xmath109 cs theory describes a gapped topological state .
this also agrees with the @xmath129 topological states where all excitations are gapped . however , we have two kinds of @xmath0 topological orders z2a and z2e
. how can the two different @xmath0 topological orders be described by the same @xmath109 cs theory ? in the following
we will show that two different @xmath0 topological orders are described by the same @xmath109 cs theory but with different realizations of the lattice symmetry . to obtain two different realizations of lattice symmetry
, we note that @xmath0 vortices for the exactly soluble model ( the z2e state ) live on the even plaquettes .
the vortices on the odd plaquettes are actually the @xmath0 charge.@xcite so under a translation by one lattice spacing , a @xmath0 vortex is changed into a @xmath0 charge !
so in the mutual @xmath109 cs theory that describes the z2e state , @xmath124 and @xmath130 must exchange under the translation by one lattice spacing . also , the z2a state contains @xmath131 flux through each square .
this @xmath131 flux also affects how @xmath118 is transformed under translation . to see this ,
let us consider two wilson loop operators @xmath132 and @xmath133 along two loops @xmath134 and @xmath135 .
both loops wrap around the torus in @xmath33-direction .
however the loop @xmath135 is displaced from the loop @xmath134 by one lattice constant in the @xmath87-direction . in the following , we will assume the lattice constant is @xmath136 .
due to the @xmath131 flux through each square , we see that @xmath137 , where @xmath89 is the length of the torus in the @xmath33-direction .
so under a translation by one lattice constant in the @xmath87-direction , @xmath138 must change to @xmath139 , to account for the change in the wilson loop .
the above discussion motivates us to define two types of mutual @xmath109 cs theories which have different realizations of translation symmetries .
let @xmath140 and @xmath141 be the translations by one lattice spacing in the @xmath87 and @xmath33 directions respectively .
the first type of the mutual @xmath109 cs theory is denoted as z2a type which describes the z2a state .
the @xmath131 flux makes the gauge fields transform non - trivially under translations : @xmath142 since the translation @xmath143 ( @xmath144 ) may shift @xmath145 ( @xmath146 by @xmath53 , this reproduces the different patterns of crystal momenta of the degenerate ground states on different lattices . the other type of the mutual cs theory is denoted as z2e type that describes the z2e state .
it has no flux .
however , the gauge fields still transform non - trivially under translations : @xmath147 @xmath148 and @xmath149 will exchange under a translation operation by one lattice spacing . in the next a few sections
, we will calculate the topological properties of the above two types of mutual cs theory .
first , we calculate the topological degeneracy for the ground states . in the temporal gauge , @xmath150 and on an even - by - even lattice ,
the fluctuations @xmath148 and @xmath149 are periodic .
we can expand them as @xmath151 where @xmath152 @xmath153 where @xmath98 are integers .
@xmath154 and @xmath155 are the gauge fields with non - zero momentum and @xmath156 and @xmath157 are the zero modes with zero momentum for the gauge fields @xmath130 and @xmath158 . because the existence of the mass gap , the degree freedoms for gauge fields with non - zero momentum @xmath159and @xmath160 have nothing to do with the low energy physics .
it is the degree freedoms of zero momentum @xmath156 and @xmath157 that determine the low energy physics .
the effective lagrangian eq.([mcs ] ) determines the dynamics of @xmath161 and @xmath162 which corresponds to two particles on a plane with a finite magnetic field .
@xmath163 are the coordinates of the first particle , and @xmath164 are the coordinates of the second particle .
thus we map the original mutual @xmath109 cs theory to a quantum mechanics model of two particles ( see appendix ) .
the energy spectrum for the quantum mechanics model can be solved easily .
the lowest energy levels for above model reveal the topological characters for the ground states .
the degeneracy for @xmath163 degrees of freedom and the degeneracy for @xmath165 degrees of freedom are given as @xmath166 and @xmath167 .
for both the z2a type and the z2e type cs theories , there exist four degenerate ground states @xmath168 however , the above result only applies to even - by - even lattice .
for other cases , even - by - odd , odd - by - even and odd - by - odd , the situations are changed .
we will discuss those more complicated cases in appendix .. we find that for the z2a type mutual cs theory , the ground state degeneracy remain to be @xmath169 for even - by - odd and odd - by - even lattices . for the z2e type mutual cs theory ,
the ground state degeneracy becomes @xmath170 for even - by - odd , odd - by - even , and odd - by - odd lattices .
one way to understand the later result is to note that if @xmath88 is odd then one gauge field will turn into the other one as we go around the lattice along @xmath87-direction .
thus the gauge fields have a twisted boundary condition : @xmath171 this twisted boundary condition means that @xmath115 and @xmath172 can be viewed as a single gauge field on a lattice whose size is doubled in the @xmath87-direction .
there are only two zero modes in the mode expansion . as a result
the ground - state degeneracy on even - by - odd , odd - by - even and odd - by - odd is reduced to @xmath170 .
we can also use the cs theories to calculate the crystal momenta of the ground states ( see appendix ) .
the results agree with those in tables [ z2acm ] and [ z2ecm ] .
we can also use the mutual @xmath109 cs theories to study edge excitations .
first , let us consider the exact soluble model ( [ hexct ] ) on a finite @xmath83 lattice with a periodic boundary condition only along @xmath33-direction .
the lattice has two edges along @xmath33 -direction located at @xmath173 and @xmath174 .
such a lattice model can be obtained from the periodic lattice model ( [ hexct ] ) by setting @xmath175 for a column of plaquettes .
the resulting model is still exactly soluble .
we find that the ground states have @xmath176-fold degeneracy which arise from @xmath177 on the column of plaquettes with @xmath175 .
those degenerate states can be viewed as gapless edge excitations on the two boundaries .
since there are @xmath178 edge sites , we find that there are @xmath179 edge states per edge site , indicating that the gapless edge states are described by majorana fermions . indeed , the gapless edge excitations can be mapped to a majorana fermion system exactly . to obtain the gapless edge states from the mutual cs theories , we introduce @xmath180 and rewrite the mutual @xmath1 cs effective theory
as @xmath181 the charges of @xmath123 and @xmath124 are quantized as integers . converting the @xmath123 and @xmath124 charges to the @xmath182 and @xmath183 charges , we find that the @xmath182 and @xmath183 charges are still quantized as integers .
however , @xmath184 charge for the @xmath182 and @xmath185 field is also allowed .
the mutual cs theory ( [ cspm ] ) has one right - moving and one left - moving branches of edge excitations .
the two branches of the edge excitations are described by the following 1d fermion theory @xcite @xmath186 at low energies , where @xmath187 represent terms that are consistent with the underlying symmetries of the lattice model .
@xmath188 carries a unit of @xmath189 charge and @xmath190 a unit of @xmath191 charge .
we note that the @xmath125 and @xmath124 charges , as the @xmath0 charge and the @xmath0 vortex , are conserved only mod 2 .
so @xmath187 may contain terms that change @xmath192 charge by @xmath193 and @xmath194 .
thus , the following terms @xmath195 are allowed in the low energy effective lagrangian .
the additional terms will open an energy gap for the edge excitations and one may conclude that the z2e state in the exactly soluble model ( [ hexct ] ) has no gapless edge excitations in general .
however , the above conclusion is not quite correct .
we see that although the presence of the edge breaks the translation symmetry in the @xmath87-direction , the finite system still has the translation symmetry in the @xmath33-direction . under the translation in the @xmath33-direction by lattice spacing , @xmath123 and @xmath126 is exchanged , or @xmath196 @xmath197 are changed into @xmath198 @xmath199 .
so the translation in the @xmath33-direction changes the sign of the @xmath191 charge and hence changes @xmath200 to @xmath201 . as a result
, only the following term @xmath202 can be added to the edge effective lagrangian , which do not break the translation symmetry along the edge .
introducing majorana fermions @xmath203 we can rewrite the edge effective lagrangian as @xmath204 the @xmath205 term gaps a pair of majorana fermions and leave the other pair gapless .
so the z2e state has right - moving and left - moving gapless edge excitations described by majorana fermions , provided that the edge is in the @xmath87- or @xmath33-direction .
the presence of the translation symmetry in the @xmath87- or @xmath33-direction is crucial for the existence of the gapless edge excitations for the z2e type mutual @xmath109cs theory and the exact soluble model . for the z2a state ,
although the low energy effective theory has the same form as the exactly soluble model , the translation does not induce the exchange between @xmath123 and @xmath124 . as a result , in general , there are no gapless edge excitations for the z2a type mutual @xmath109 cs theory and the z2a state .
in this paper , two kinds of @xmath0 topological ordered states for frustrated spin systems , z2a state and z2e state , are studied . using the slave - particle theory
, we calculate their ground - state degeneracy , their ground - state quantum numbers , their gapless edge state , and the projective symmetry group of their quasi - particles .
we propose a mutual @xmath110 chern - simons theory with two different realizations of lattice symmetry as the effective field theories that describe the two types of topological orders .
we show that the effective theories produce the same low energy physics , including the degeneracy of the ground state , the quantum number for the ground state and the edge states .
it turns out that the different @xmath0 topological orders are reflected in different realizations of the lattice symmetry in the same effective mutual chern - simons theory .
we like to mention that the z2a phase appears to be an example of `` weak symmetry breaking in dimension 2 '' , while the z2e phase appears to be an example of `` weak symmetry breaking in dimension 1 '' discussed in ref . .
so these two phases are examples of the two basic ways that lattice symmetries and topological structure can be entangled .
this research is supported by nsf grant no .
dmr-0706078 , nfsc no . 10228408 , and nfsc no .
we have used ansatz @xmath206 to describe the four degenerate ground states for the z2e state . here
@xmath42 . @xmath207
have values @xmath45 or @xmath46 , with @xmath208 if the link @xmath48 crosses the @xmath87 or @xmath33 line ( see fig .
[ xyline ] ) and @xmath209 otherwise .
it is pointed out that the above result of four degenerate ground states is right only for the z2e state on an even - by - even lattice . on other kinds of lattice ( even by odd , odd by even and odd by odd ) , there exist only two different ground states .
the other two states are projected out since the mean - field ground states contain odd numbers of fermions .
let s calculate the topological degeneracy for the z2e state on different lattices in detail .
it was pointed out that the total number of the @xmath210 fermions on @xmath211 and @xmath212 is always even if @xmath213 . to determine if the mean - field ground state contains even or odd number of @xmath210 fermions , we will only pay attention to the occupation on the following four momentum points : @xmath214 , @xmath215 , @xmath216 , @xmath217 .
firstly , we discuss the topological degeneracy for z2e state on an even by even lattice . for the ground state described by @xmath94 ,
the energy levels for both @xmath218 and @xmath219 have positive energies at @xmath220 ( see eq .
[ h2 ] ) . thus the @xmath214 level is not occupied .
we also see from eq .
[ h2 ] that , at @xmath221 , @xmath218 has a positive energy @xmath219 has a negative energy .
thus the @xmath222 level is occupied by a @xmath223 particle .
similarly , we find that the @xmath224 level is occupied by a @xmath225 particle , the @xmath226 level is occupied by a @xmath225 particle and a @xmath227 particle .
therefore , four particles occupy the points @xmath228 , @xmath229 , @xmath216 , @xmath217 .
because the meanfield ground state @xmath230 has even number particles , it survives the even - particle - per - site projection .
also , the total contribution to the crystal momentum from the @xmath231 levels is zero .
thus the total crystal momentum is determined by the particles that occupy the @xmath228 , @xmath232 , @xmath216 , @xmath217 levels .
we find that the total crystal momentum of the above state is @xmath233 . for the ground states described by @xmath234 @xmath235 , and @xmath104 , non of the high - symmetry points @xmath228 ,
@xmath232 , @xmath216 , @xmath217 exist .
thus the ground states have even number particles , so they are all permitted under the even - particle - per - site projection .
the total crystal momenta of the above states are all zero
. therefore , there are four degenerate ground states on even - by - even lattice .
one carries crystal momentum @xmath236 and other three carry crystal momentum @xmath228 .
this corresponds to the first column of table [ z2ecm ] .
secondly , we discuss the topological degeneracy for z2e state on an even by odd lattice . for the ground state described by @xmath94
, the @xmath237 level is not occupied ; the @xmath238 level is occupied by one @xmath225 particle , as before .
the points @xmath215 and @xmath217 do not exist . as a result
, only one particle occupies the high - symmetry points .
because the ground state @xmath239 has odd number particles , it is forbidden by the even - particle - per - site projection .
for the ground state described by @xmath235 , the points @xmath228 , @xmath229 , @xmath216 , @xmath217 do not exist .
thus the ground state has even number particles , so it is permitted by the projection .
such a state carries a @xmath228 crystal momentum . for the ground state described by @xmath240
, the @xmath241 level is occupied by a @xmath223 particle , and the @xmath242 level is occupied by a @xmath225 and a @xmath223 particles .
the @xmath216 and @xmath228 points do not exist . as a result
, three particles occupy the high - symmetry points .
the state is forbidden by the projection .
for the ground state noted by @xmath104 , the points @xmath243 , @xmath215 , @xmath216 , @xmath217 do not exist .
because the ground state @xmath244 has even number particles , it is also permitted by the projection .
such a state also carries a @xmath228 crystal momentum .
therefore there are two degenerate ground states on an even by odd lattice .
similarly topological degeneracy for z2e state on an odd by is also two .
all those states carry a @xmath228 crystal momentum .
this corresponds to the second and third columns of table [ z2ecm ] .
last , let us discuss the topological degeneracy for z2e state on an odd by odd lattice .
for the ground state described by @xmath94 , the @xmath220 level is not occupied .
the points @xmath216 @xmath215 and @xmath217 do not exist . as a result , no particle occupies the high - symmetry points .
the ground state @xmath245 has even number particles which is permitted by the projection . for the ground state described by @xmath240
, the @xmath238 level is occupied by a @xmath225 particle .
the points @xmath228 @xmath215 and @xmath246 do not exist . as a result
, one particle occupies the high - symmetry points .
because the ground state @xmath247 has odd number particles , it is not permitted by the projection . for the ground state described by @xmath235 , the @xmath221 level is occupied by a @xmath223 particle .
the points @xmath228 @xmath248 and @xmath246 do not exist . as a result
, one particle occupies the high - symmetry points .
because the ground state @xmath247 has odd number particles , it is not permitted by the projection . for the ground state described by @xmath104 , at the @xmath242 level
is occupied by a @xmath218 and a @xmath219 particle .
the points @xmath249 , @xmath215 , and @xmath228 do not exist . as a result ,
two particle occupies the high - symmetry points .
the ground state @xmath250 has even number particles , so the state is permitted by the projection . in conclusion ,
z2e state has four - fold degeneracy on an even by even lattice and two - fold degeneracy on an even by odd lattice , odd by even lattice or odd by odd lattice .
the crystal momenta of those ground states are given by the table [ z2ecm ] .
to calculate the topological properties for the ground states of the mutual @xmath109 cs theories , one needs to quantize the gauge fields .
we will choose the temporal gauge @xmath251 .
in the temporal gauge , the physical degrees of freedom are described by @xmath252 and @xmath253 .
after the mode expansion , the effective lagrangian can be written as where @xmath154 and @xmath255 represent the terms that certain only the @xmath256 modes .
the masses are given as @xmath257 @xmath258 and @xmath259 @xmath260 because the existence of the mass gap , the degree freedoms for gauge fields with non - zero momentum @xmath261and @xmath262 have nothing to do with the low energy physics .
so we will concentrate on the dynamics of @xmath263 and @xmath264 . @xmath265 using the conjugate momentum we write down the following effective hamiltonian to describe the low energy physics of the mutual @xmath110 cs theory @xmath268 by choosing different landau gauges , the above can rewritten as @xmath269 or @xmath270 the low energy properties of the @xmath1 chern - simons theory is described by the above hamiltonian . [
[ topological - degeneracy - and - crystal - momenta - for - the - z2e - type - mutual - times - cs - theory ] ] topological degeneracy and crystal momenta for the z2e type mutual @xmath109 cs theory ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ we note that @xmath271 and @xmath272 are related by a @xmath273 gauge transformation .
thus @xmath274 and @xmath275 are also related by a @xmath273 gauge transformation , which implies that @xmath274 and @xmath275 should be viewed as the same point . similarly each of the three pairs @xmath276 and @xmath277
, @xmath278 and @xmath279 , @xmath280 and @xmath281 , also should be viewed as the same point .
thus the above hamiltonian describes two particles , each moves on a @xmath282 torus .
each particle also see @xmath283 flux through the torus .
the first particle is described by @xmath163 .
since there are two units of flux through the torus , the ground states for the first particle has a degeneracy @xmath284 .
similarly , the ground states for the second particle also has a degeneracy @xmath167 . and
the wave - functions @xmath286 for the four ground states with degenerate energy are given as @xmath287 @xmath288 @xmath289and @xmath290 , @xmath291 \exp \left [ -\frac 1{4\pi } \theta _ y^2\right ] , \\ \psi _ 2 & \simeq & e^{-\imth \theta _ x}\exp \left [ -\frac 1{4\pi } \left ( \theta _ y-\pi \right ) ^2\right ] \exp \left [ -\frac 1{4\pi } \theta _ y^2\right ] , \nonumber \\ \psi _ 3 & \simeq & e^{-\imth \theta _ x}\exp \left [ -\frac 1{4\pi } \theta _ y^2\right ] \exp \left [ -\frac 1{4\pi } \left ( \theta _ y-\pi \right ) ^2\right ] , \nonumber
\\ \psi _ 4 & \simeq & e^{-\imth \theta
_ x}e^{-\imth \theta _
\left [ -\frac 1{4\pi } \left ( \theta _ y-\pi \right ) ^2 -\frac 1{4\pi } \left ( \theta _ y-\pi \right ) ^2\right ] .
\nonumber\end{aligned}\ ] ] now let s calculate the crystal momentum for the four - fold degenerate ground states . for the z2e type mutual @xmath109 cs theory ,
the translation operations @xmath292 are known as @xmath293 thus we have the translation operation for its zero modes @xmath294 and @xmath295 @xmath296 under the translation operators , we have so @xmath299 and @xmath289 can not be the eigenstates for the ground state . instead
, the eigenstates for the ground state are given as @xmath300 and @xmath301 and @xmath302 as a result , on an even - by - even lattice , the crystal momentum of the e type mutual @xmath109 cs theory is @xmath303 for the ground states @xmath287 @xmath304 , @xmath305 and @xmath306 for the ground state @xmath307 for other cases , on an even - by - odd , odd - by - even or odd - by - odd lattice , the situations are changed .
because for odd number rows along -axis or -axis , one gauge field @xmath115 ( @xmath172 ) will turn into the other one @xmath172 ( @xmath115 ) .
for example , on a @xmath308 even - by - odd lattice ( @xmath88 is an even number and @xmath89 is an odd number@xmath309 , under such a twisted boundary condition for odd number @xmath89 , one has @xmath310 the quantization for gauge fields in eq.[modeaa ] can not be applied to the gauge fields under a twisted boundary condition .
now after putting the mutual @xmath109 cs theory on a @xmath311 even - by - even lattice , we have a periodic boundary condition , @xmath312 in the temporal gauge , @xmath150 and on such even - by - even lattice , we can expand the fluctuations for the gauge fields as @xmath313 where @xmath152 @xmath314 where @xmath98 are integers . @xmath154 and @xmath155 are the gauge fields with non - zero momentum and @xmath156 and @xmath157 are the zero modes with zero momentum for the gauge fields @xmath130 and @xmath158 .
@xmath315 and @xmath316 ( @xmath317 and @xmath318 ) are not independent and have constraints to obey the original twisted boundary condition in eq.[t ] , we must have @xmath319 to calculate the topological degeneracy , we map the original mutual @xmath109 cs theory on even - by - odd lattice to two - particle quantum mechanics model on a torus in a magnetic field @xmath320 . in the
landau gauge , the effective hamiltonian of the two - particle quantum mechanics model is given as @xmath321 where @xmath322 @xmath323 and @xmath324 @xmath325 however , because of the constraint in eq.[c ] , the two particles @xmath326 and @xmath327 are bound into a single particle ! as a result , there are two degenerate ground states instead of four . in addition , we can write down the wave - functions for the two ground states in the landau gauge with topological degeneracy : for the wave - function @xmath287 @xmath328 and the wave - function @xmath299 , @xmath329 now let s calculate the crystal momentum for the two - fold degenerate ground states .
the ground states are invariant under the translation operations @xmath330 then the crystal momentum @xmath106 is @xmath228 for the e type mutual @xmath109 cs theory on an even - by - odd lattice . furthermore , using the same method , we calculated the topological degeneracies and the crystal momenta for the ground states of the z2e type mutual @xmath109 cs theory on an odd - by - even or odd - by - odd lattices .
the results are similar to those on an even - by - odd lattice : the ground states have two - fold degeneracy@xmath331and @xmath303 . [ [ topological - degeneracy - and - crystal - momenta - for - the - z2a - type - mutual - times - cs - theory ] ] topological degeneracy and crystal momenta for the z2a type mutual @xmath109 cs theory ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ in this part , we will calculate the topological degeneracy and crystal momenta for z2a type mutual @xmath109 cs theory . the effective hamiltonian to describe the low energy physics of the z2a
type the mutual @xmath109 cs theory can be written in the landau gauge as @xmath332 it is noted that there exists the heisenberg algebra for zero modes of the gauge fields .
the magnetic
translation operators @xmath333 and @xmath334 consist of the heisenberg algebra @xmath335 because the hamiltonian is invariant for the operations @xmath336 and @xmath337 @xmath338 the ground states are the eigenstates of @xmath336 and @xmath339 so one can draw a conclusion from the heisenberg algebra that the ground states have two - degeneracy for @xmath340 . on the other hand , for @xmath341 one can do the same calculation .
so the ground states have two - degeneracy for @xmath342 which is also characterized by the eigenstates of @xmath343 and @xmath344 as a result , for the z2a type mutual @xmath109 cs theory , the ground states have four - fold degeneracy : @xmath168 now let s calculate the crystal momentum for the four - fold degenerate ground states . for the z2a type mutual @xmath109 chern - simons theory ,
the translation operations for the gauge fields are given by eq .
( [ aatrans ] ) .
the translation operations for zero modes of the gauge fields are given as ( [ aatrans ] ) @xmath349 and @xmath350 as a result , the real ground states can be labeled by the eigenvalues of @xmath351 ( or @xmath352 @xmath353 @xmath354 ) which are @xmath46 and @xmath355 .
we denote the four ground states with topological degeneracy as @xmath287 @xmath288 @xmath289 and @xmath356 @xmath357 firstly , on an even - by - even lattice , the translation operations for its zero modes lead to trivial results @xmath358 and @xmath359 from them , we have @xmath360 then the crystal momentum @xmath106 of the four - fold degenerate ground states @xmath361 is @xmath362 secondly on an odd by even lattice ( @xmath88 is odd number and @xmath89 is even number ) , the translation operations are given as @xmath363 and @xmath364 now the translation operator @xmath141 turns into the magnetictranslation operator @xmath365 @xmath366 under the translation operations on the wave functions in eq.[wave ] , we have @xmath367 and @xmath368 using the same method , we can obtain that the crystal momentum of the two ground states @xmath369 and @xmath299 is @xmath362 the crystal momentum of the other two ground states @xmath289 and @xmath290 is @xmath370 thirdly , on an even - by - odd lattice ( @xmath88 is even number and @xmath89 is odd number ) , the translation operations for its zero modes lead to non - trivial results @xmath371 and @xmath372 then the translation operator @xmath140 turns into the magnetictranslation operator @xmath373 @xmath374 from them , we have @xmath375 and @xmath376 the crystal momentum of two ground states @xmath369 and @xmath377 is @xmath362 the crystal momentum of the other two ground states latexmath:[$ fourthly for @xmath88 and @xmath89 are all odd numbers ( on an odd - by - odd lattice ) , the translation operations become @xmath379 and @xmath380 then the translation operators @xmath140 and @xmath141 turn into the magnetic translation operator @xmath381 and @xmath382 @xmath383 now @xmath140 and @xmath141 must obey the heisenberg algebra for @xmath384 and @xmath336 @xmath385 on the other hand , the translation symmetry of the system leads to the commutation relationship between @xmath140 and @xmath141 @xmath386 the only solve to the eq.[an ] and eq.[com ] is @xmath387 .
that is , there do nt exist the four degenerate ground states at all .
we can see that for the real ground states , the @xmath123 and @xmath124 charges for the excitations can not be zero on an odd by odd lattice .
so the non zero background charge leads to an infinity degeneracy on odd by odd lattice for the z2a type mutual @xmath109 cs theory . | we study several different @xmath0 topological ordered states in frustrated spin systems .
the effective theories for those different @xmath0 topological orders all have the same form a @xmath0 gauge theory which can also be written as a mutual @xmath1 chern - simons theory .
however , we find that the different @xmath0 topological orders are reflected in different projective realizations of lattice symmetry in the same effective mutual chern - simons theory .
this result is obtained by comparing the ground - state degeneracy , the ground - state quantum numbers , the gapless edge state , and the projective symmetry group of quasi - particles calculated from the slave - particle theory and from the effective mutual chern - simons theories .
our study reveals intricate relations between topological order and symmetry .
keywords : topological order , mutual chern - simons theory , spin liquid |
we consider the random forced burgers equation @xmath1 that describes weak 1d acoustic perturbations in the reference frame moving with the sound velocity @xcite .
the external force @xmath2 in this frame is generally short - correlated in time , so let us assume that @xmath3 then the statistics of @xmath2 can be thought gaussian and therefore is completely characterized by ( [ phipc ] ) .
we are interested in turbulence with a large value of reynolds number @xmath4 , where @xmath5 is the characteristic scale of the stirring force correlator @xmath6 .
this problem was intensively studied during the last years @xcite .
the main feature of burgers turbulence is the formation of shock waves with large negative velocity gradient inside and small viscous width of the front .
the positive velocity gradients are decreased by the dynamics of burgers equation due to self - advection of velocity . on the contrary the increasing of negative gradients
could be stopped only by viscosity .
the motion of shock waves leads to a strong intermittency , the pdf of velocity gradients @xmath7 is strongly non - gaussian .
the one way to describe the intermittency is to study rare events with large fluctuations of velocity , that give the main contribution to the high momenta @xmath8 or to the pdf tails .
the right tail ( positive large @xmath9 ) of pdf @xmath10 was first found by feigelman @xcite for the problem of charge density wave in an impurity potential .
later it was recovered using operator product expansion @xcite ( see also @xcite ) , instanton calculus @xcite , minimizers approach @xcite and mapping closure @xcite .
the left tail in the inviscid limit seems to be algebraic , probably @xmath11 .
the large negative gradients exist practically only inside the shock waves .
the maximal value of gradient is proportional to the square of the velocity jump on the shock wave : @xmath12 .
then roughly the tail of the shock wave amplitude pdf has the form @xmath13 .
the analysis of the instanton structure predicts the value @xmath14 @xcite .
this prediction is consistent with the assumption , that the tails of @xmath15 should not depend on the viscousity @xmath16 .
we are interested in the statistics of large values of gradients @xmath17 .
the velocity field configurations @xmath18 that make a contribution to the probability @xmath19 of the equality @xmath20 have the gradient greater or equal to @xmath21 somewhere .
the probability @xmath19 decays very fast while @xmath21 increases , i.e. the contribution of events with gradient greater than @xmath21 somewhere is highly suppressed .
then one believes that only some specific field configurations @xmath22 ( `` optimal fluctuations '' @xcite or instantons ) make contribution to @xmath19 at large @xmath23 . under this assumption to calculate @xmath19
one should find this optimal field configuration @xmath22 and estimate the probability of its realization .
all instantons of this type are posed at the far tail of the statistical weight of averaging @xmath24 $ ] . indeed , to produce large fluctuation of @xmath25 the stirring force @xmath2
also should be large , and the probability of such fluctuation @xmath2 is low .
the weight @xmath26 $ ] may not contain a large parameter , but it should have fast tails , e.g. exponential ones .
then the concurrence between statistical weight and the value of calculated quantity makes the contributing realizations of @xmath27 rather determined .
this approach was introduced by lifshitz @xcite .
later it was applyed to determine high order correlation functions in field theory @xcite and in the systems of hydrodynamic type : simultaneous ( see , e.g. @xcite ) and non - simultaneous @xcite ones .
the paper is organized as follows . in sec .
[ spa ] we derive the equations for the instanton .
[ nc ] is devoted to the detailed description of our scheme of numerical calculations . in sec .
[ vi ] we discuss the numerical results and describe the behavior of the solution of instanton equations at large times .
the velocity gradients pdf @xmath19 can be written as the path integral @xmath28 where the effective action @xmath29 has the form @xcite @xmath30 the integration over @xmath31 gives rise to @xmath32 , and the factor @xmath33 was chosen for our convenience .
note that if retarded regularization of the path integral ( [ pdf ] ) is used then @xmath34 and we have no normalizing @xmath25-dependent denominators in ( [ pdf ] ) .
one can find some analogies between the appearance of the second field @xmath35 and technique that was developed by keldysh @xcite for nonequilibrium dynamics description .
we are interested in the tails of pdf @xmath19 , i.e. the parameter @xmath21 in the integral ( [ pdf ] ) is large .
the asymptotics of @xmath19 at large @xmath36 is determined by the saddle - point configuration of fields @xmath18 , @xmath37 ( and also parameter @xmath31 ) , near which the variation of the integrand is equal to zero @xcite .
the saddle - point configuration ( sometimes called classical trajectory or instanton ) is governed by the following equations @xmath38 where @xmath39 is the convolution @xmath40 the solution should satisfy boundary conditions @xmath41 the value of @xmath31 is tuned in such a way that the condition @xmath20 holds .
the quantity @xmath31 is a lagrange multiplier for finding the extremum of @xmath29 with the condition @xmath42 .
the equation for @xmath35 should be solved moving back in time because of the signs at @xmath43 and @xmath44 in the instanton equation ( [ ie2 ] ) .
the convolution @xmath45 is the optimal configuration of external force @xmath2 that produces large negative gradient .
in what follows we will measure the length in @xmath5 units , i.e. we set @xmath46 . rescaling the time @xmath47 and fields @xmath25
, @xmath35 one can exclude the parameter @xmath16 from the instanton equations : @xmath48 at @xmath49 one has @xmath50 .
the only parameter in the instanton equations is @xmath51 .
note that the steady - state kink solution of burgers equation with the negative gradient @xmath21 is @xmath52 thus the physical meaning of @xmath53 is the square of the ratio of pumping scale @xmath46 and the kink width @xmath54 . the effective action @xmath55 at the instanton that gives the right exponent : @xmath56 is equal to @xmath57 the freedom of rescaling the fields @xmath25 , @xmath35 and the time @xmath47 with appropriate change of @xmath16 gives us the following relation : @xmath58 with the function @xmath59 to be determined .
one can prove by straitforward calculation the following relation between functions @xmath60 and @xmath59 : @xmath61 the relations of such sort are well - known in classical mechanics ; here @xmath62 and @xmath31 are conjugate variables , and saddle - point configuration is the trajectory of extremal action .
the instanton equations ( [ ie1],[ie2 ] ) are hamiltonian : @xmath63 the hamiltonian @xmath64 is the integral of motion , i.e. @xmath65 . since both @xmath66 and @xmath67 tend to zero at @xmath68 we have @xmath69 . from the instanton equations and the condition
@xmath70 we get @xmath71 the last term is due to viscousity . at the right tail
it is unimportant , and we have @xmath72 , i.e. @xmath73 . at the viscous left tail
its contribution to the action is of the same order as other terms . if @xmath74 is a powerlike function : @xmath75 , then one has @xmath76 .
the high momenta can be calculated by the instanton method in a following way .
because @xmath77 is a narrow function for large @xmath78 , and only narrow velocity interval , which position depends on @xmath78 , contributes to @xmath79 .
the position of this interval is exactly the saddle - point in the integral @xmath80 ( see ( [ lnps ] ) ) , that satisfies the equation @xmath81 combining it with @xmath82 we again get ( [ fsa ] ) . to get the instanton equations for the average @xmath79 one should only substitute @xmath31 in ( [ ie2 ] ) for @xmath83 .
then the instanton equations become the same as in @xcite .
one also should consider fluctuations near the instanton as a background .
the way how the fluctuations can be taken into account is unknown yet but their influence to @xmath74 due to their phase volume is small in comparison with @xmath84 while @xmath85 . at smaller gradients
the fluctuations essentially change the answer and we have the algebraic tail @xcite .
the preliminary calculations that were made in @xmath86 , @xmath87 variables have shown that the width of the instanton equations solution grows with @xmath88 and is proportional to @xmath89 , while its amplitude is proportional to @xmath90 . to avoid the necessity of treating simultaneously narrow structure at small @xmath87 and wide one at large @xmath87 we used the following variables : @xmath91 where @xmath92 is some constant of the order of unity . the instanton equations in these variables
take the form @xmath93 where @xmath94 .
the boundary conditions for @xmath95 , @xmath96 are analogous to ( [ bc ] ) .
let us describe now the general structure of the numerical scheme that finds the solution of our boundary problem .
the diffusion terms @xmath97 , @xmath98 in instanton equations ( [ iext1],[iext2 ] ) have opposite signs .
if one considers these equations as two linked cauchy problems , then the natural direction of time in ( [ iext1 ] ) is positive , while in ( [ iext2 ] ) the direction is negative .
assume that at a given value of @xmath31 the approximate solution @xmath99 is known .
let us try to make it closer to the true solution of the problem .
for this purpose let us solve the cauchy problem for ( [ iext2 ] ) starting from @xmath100 and moving up to large enough @xmath101 . then using @xmath96 that we have got in the previous step we solve the cauchy problem for ( [ iext1 ] ) moving from @xmath102 up to @xmath103 . as a result
we get the new values @xmath104 .
further we will use the sign @xmath105 for the mapping @xmath106 .
the stationary point @xmath95 of the mapping @xmath105 and the corresponding function @xmath107 are the desired solution of ( [ iext1],[iext2 ] ) .
the numerical experiments have shown that iterations @xmath108 , \quad { \tilde u}^{(0 ) } \equiv 0 \label{si}\ ] ] converge if @xmath109 . while @xmath110 the simple iterations ( [ si ] ) are divergent .
curve 1 at the fig .
[ fig1](a ) shows how the value of the gradient @xmath111 depends on the number of iteration for @xmath112 .
it can be seen that the stationary point of @xmath105 is unstable , but the mapping @xmath113 \big]$ ] has two stable stationary points .
the stability properties of the iterations are determined by the spectrum of the linearization @xmath114 of @xmath105 near a stationary point : @xmath115 = { \tilde u } + { \hat k } v + \dots\ ] ] the iteration process is convergent if the modulus of all eigenvalues of the linear part @xmath114 is less than 1 .
the period doubling indicates that while @xmath31 passes through @xmath116 one of @xmath117 s eigenvalues passes through @xmath118 @xcite .
let us denote this eigenvalue as @xmath119 .
this knowledge allows us to construct the new mapping with stable stationary point that coincides with one of the mapping @xmath105 .
let us proceed the following iterations : @xmath120 \equiv c\,f \big [ { \tilde u}^{(i)}\big ] + ( 1-c ) { \tilde u}^{(i ) } , \label{fc}\ ] ] where @xmath121 is some constant .
it is easy to check that the stationary points of the mappings @xmath105 and @xmath122 do coincide .
the linear part of @xmath122 is equal to @xmath123 , its eigenvalue corresponding to unstable @xmath119 is @xmath124 .
if we take the value of @xmath121 inside the interval @xmath125 , then @xmath126 , and the iterations ( [ fc ] ) are convergent .
since we do nt know @xmath119 a priori , the value of @xmath121 that provides the convergence of iterations , was determined from experiment .
[ fig1](a ) illustrates the influence of the decreasing of @xmath121 on the dependence of @xmath127 vs. number of iteration . in the final version of the computer
code the value of @xmath121 was changed in an adaptive way : each time the value of @xmath128 was decreased after an iteration , @xmath121 was multiplied by @xmath129 .
one can compare from the fig .
[ fig1](b ) the iterations run for @xmath130 , @xmath131 and for adaptive decreasing of @xmath121 , all three for @xmath112 .
the initial and final values of @xmath121 in the case of adaptive change were equal to @xmath132 and @xmath133 , respectively .
it is clear that for the two last cases the iterations converge to the same solution . as a function of number of iteration . at fig .
[ fig1](a ) the dots corresponding to one run are joined by a line for convenience .
the curves are : ( a ) simple iterations according to ( [ si ] ) ( curve 1 ) and according to ( [ fc ] ) with @xmath130 ( curve 2 ) ; ( b ) @xmath130 ( curve 1 , its thickness is determined by the amplitude of serrated oscillations ) , @xmath134 ( curve 2 ) , and adaptive decreasing of @xmath121 during the run , the initial value is @xmath130 , @xmath121 was multiplied by @xmath129 four times ( curve 3 ) ; @xmath112.,width=288 ] the solution of cauchy problem is found numerically using the method of finite differencies .
the grid covers the rectangular domain @xmath135 , @xmath136 . in numerical calculations some of the boundary conditions ( [ bc ] ) that in principle are posed at the infinity
were considered as they are posed at ( large enough ) @xmath137 and @xmath138 .
typical values used : @xmath139 and @xmath140 .
the grid had uniform mesh intervals in @xmath141 , typical number of grid sites along @xmath141 axe was eqial to @xmath142 .
the first calculations had shown that the solution changes rapidly in the vicinity of @xmath103 , while at large @xmath143 it varies slowly . because of this we used nonuniform grid for variable @xmath143 . the time step was smaller inside the interval @xmath144 , typical value used : @xmath145 .
the number of grid sites inside this interval varied from @xmath146 to @xmath147 .
the computer power we had limited the number of all time steps in the grid by @xmath148 . during all calculations we used @xmath149 . the equation ( [ iext2 ] ) should be solved backward in time .
the source term in right hand side of the equation ( [ iext2 ] ) provides us the initial condition @xmath151 .
this means that at small times the field @xmath96 is localized in a very narrow interval centered at @xmath152 .
such initial condition can not be accurately discretized , so we used another way to represent @xmath96 at small times . for small @xmath143 the field @xmath96
is very narrow . at its support
we can approximate the velocity @xmath95 by a linear profile : @xmath153 .
the evolution of @xmath96 in such a velocity field is described by the derivative of gaussian contour : @xmath154 we use such a representation for @xmath96 for @xmath155 .
the value of @xmath156 is chosen in such a way , that the velocity field @xmath95 is still linear at the width of @xmath96 . from the other hand , @xmath96 at @xmath157 already becomes wide in comparison with mesh interval @xmath158 .
typical value of @xmath156 used : @xmath159 . for times @xmath160 the solution is found by fully implicit scheme @xmath161 where @xmath162 , @xmath163 are equal to @xmath164 the function @xmath165 is expressed via velocity field : @xmath166 .
here @xmath167 , @xmath168 mesh intervals , and @xmath169 , @xmath170 site coordinates .
the numerical scheme used is monotonous and stable , it is of the first order of accuracy in @xmath171 and of the second order in @xmath158 @xcite . at this stage
we use the initial condition @xmath173 .
the viscousity , source and self - advection terms are treated by splitting technique @xcite . at each time step it
is first calculated the change of @xmath95 due to the source , then to the viscousity , and at the last to the nonlinearity . from calculated already grid layer @xmath174 the next layer @xmath175
is found in a following order : first , the equation @xmath176 , is solved : @xmath177 the convolution @xmath178 is calculated as the result of inverse fast fourier transform ( fft ) acting on the product of @xmath179 s and @xmath96 s fft images .
the external force correlator during all calculations was equal to @xmath180 .
the numbers @xmath181 that are found in such a way _ are not _ a final solution for the layer @xmath182 , since only the source term has been taken into account yet .
we use them as an input at the next step , we will denote them as @xmath183 ( note that they _ do not _ coincide with @xmath183 in ( [ rsou ] ) ) . next , the viscousity and linear part of advection are taken into account , according to the equation @xmath184 .
the fully implicit scheme was used analogously to ( [ rsp ] ) : @xmath185 here @xmath186 .
again , the numbers @xmath181 do not form a final solution , and we send them to the next step with a @xmath187 notation .
the nonlinear part of the equation @xmath188 was solved by explicit conservative scheme @xcite : @xmath189 that finally gives us the next layer of the velocity field @xmath182 .
this scheme is of the first order of accuracy in @xmath190 and of the second one in @xmath191 .
in this section we represent the results of our calculations , that show the structure of the instanton and its change with @xmath192 .
the minimal value of @xmath31 at which the reliable results in numerics were obtained is @xmath112 . for @xmath193 ( a ) , @xmath194 ( b ) , and @xmath112 ( c ) .
the values of levels can be calculated from the two given levels according to arithmetic progression law.,width=288 ] the general features of the instanton structure change with @xmath31 can be obtained from fig .
[ fig2 ] that shows the level curves of @xmath195 for three values of @xmath31 . since @xmath172 and @xmath150 are the odd functions of @xmath141 , we draw only the region where @xmath196 .
the calculations were done in a rectangular @xmath197 , @xmath198 , whose dimensions are a bit larger than it is shown at the figure
. one can see that the instanton life - time and the maximal value of @xmath199 rapidly increase with the growth of @xmath192 .
the growth leads to the deformation of level curves near @xmath103 because of the influence of nonlinearity , that is weak at @xmath200 ( fig .
[ fig2](a ) ) and very strong at @xmath201 ( fig .
[ fig2](c ) ) . detailed analysis of the instanton solution based on the results of numerical calculations allows us to distinguish five different regimes in the instanton time evolution .
below we discuss them consequently from @xmath202 to @xmath203 . at @xmath103 ; @xmath112.,width=288 ] the first regime consist in the viscous smearing of the field @xmath35 up to the scale of the kink width @xmath204 ( see ( [ tanh ] ) , fig .
[ fig3 ] ) .
since the viscousity plays the crucial role in this regime , we will also use dimensional variables . at @xmath205
we have @xmath206 , and the width of the kink in the velocity profile is equal to @xmath207 ( see ( [ tanh ] ) ) .
since @xmath35 is very narrow , the viscousity dominates the evolution .
the width of @xmath35 obeys the diffusion law and equals to @xmath208 .
these two widths become comparable at time @xmath209 , or @xmath210 .
this means that during the whole time of smearing of @xmath35 by viscousity the width of the kink is of the order of @xmath211 .
indeed , if the shape of the velocity profile deviates from steady - state kink solution ( [ tanh ] ) , then the change of the kink width during this time period would be of the order of @xmath212 . at this regime the source term @xmath213 in instanton equation ( [ ie1 ] ) is unimportant .
when the width of @xmath35 becomes of the order of @xmath214 the rate of expansion of @xmath35 due to the velocity gradient becomes comparable with the rate due to viscousity ( such a balance determines the width of the kink ) .
the next ( second ) regime was exhaustively studied in @xcite .
it consists in dilation of fields @xmath66 , @xmath67 up to the pump scale @xmath46 .
the fields are advected by velocity @xmath66 , and considering evolution back in time they are expanded by it since @xmath215 .
the time needed for the expansion is equal to @xmath216 . during the 3rd5th regimes
the width of @xmath66 and @xmath67 field is much greater than @xmath46 .
then it is naturally to substitute @xmath6 by @xmath217 .
the instanton equations take the following simple form : @xmath218 where @xmath219 , @xmath220 .
while moving to large negative time numerical solution has a tendency to fall in @xmath221 ( see fig .
[ fig4 ] , curves @xmath222 and @xmath223 ) .
then the equations for @xmath66 , @xmath224 reduce to the burgers equations with the evolution back in time .
( solid curves ) and @xmath225 for three values of @xmath143 ; @xmath112.,width=288 ] such a substitution is not always possible . during such an evolution
the shock waves occure ( see , e.g. , the level curves at fig .
[ fig2](c ) near @xmath226 , fig .
[ fig4 ] , curves 1 and 2 ) . for transition to equations ( [ equv ] ) to be valid
the width of these shock waves should be larger than @xmath46 .
otherwise the substitution of @xmath227 by @xmath228 is not valid .
the width of the shock wave in dimensionless variables is greater than @xmath229 only if its height is smaller than @xmath229 .
however , right after the 2nd regime when the width of @xmath66 and @xmath67 fields becomes greater than the pumping force correlation length @xmath46 the amplitude of the velocity field @xmath66 is of the order of @xmath230 .
the amplitude of @xmath66 becames small only at very large times ( it is shown below that such a crossover happens at @xmath231 ) .
it means that there is an intermediate regime that goes after the 2nd one , where the substitution of @xmath227 by @xmath228 is inapplicable . in this ( third ) regime
the fields @xmath66 and @xmath67 are smooth functions in the interval wider than @xmath229 . at the ends of this interval they contain shock waves
the value of @xmath66 and @xmath67 rapidly goes to zero ( as it is shown schematically in fig .
[ fig5 ] ) .
we will use the word `` shock '' for these structures at the ends of the interval , while for narrow structure in the velocity field @xmath66 near @xmath232 , @xmath49 we will use the word `` kink '' .
now we will consider the structure of the shocks in detail .
let us denote the height of the shocks in @xmath233 and @xmath234 fields by @xmath235 and @xmath236 , respectively .
their position we will denote as @xmath237 ( @xmath238 , shocks are posed at @xmath239 ) . and @xmath67 for the 3rd regime , @xmath240.,width=288 ] .,width=288 ] in the third regime the width of the shocks in the field @xmath67 is determined by the competition of squeezing them by the velocity @xmath66 and spreading by viscousity ( see fig .
[ fig6 ] ) .
since we are in a strongly nonlinear regime the viscousity is weak , and the width of @xmath67 s shocks is very small .
the shocks in @xmath67 are stationed at the center of @xmath66 s shocks , i.e. in almost linear velocity profile with the gradient @xmath241 .
the width of @xmath67 s shocks can be estimated as @xmath242 .
then the good approximation for @xmath234 near the shock @xmath243 is @xmath244 , where @xmath245 is step - function .
during the evolution forward in time the shocks in @xmath66 do not break down because of source term @xmath213 .
the source prevents the destructive effect of advection term @xmath246 .
the shocks in @xmath67 should carry the @xmath66 s shocks of the height @xmath247 , i.e. of strength @xmath248 .
thus we should have @xmath249 in this regime .
now we show it more carefully .
let us write the instanton equation ( [ ie1 ] ) in the reference frame of the shock near the point @xmath250 ( see fig . [ fig6 ] ) .
we have two contributions to time derivative @xmath251 : from the growth of @xmath247 in time ( of order @xmath252 ) and from the motion of the shock ( of order @xmath253 ) . neglecting the first one we write the following equation for the velocity @xmath233 : @xmath254 where the new function @xmath255 is determined by the equation @xmath256 with the condition @xmath257 with @xmath258 .
integrating this equation once we obtain @xmath259 since @xmath260 we have @xmath261 .
the next step consists in finding the solution of ( [ ie1],[ie2 ] ) between the shocks posed at @xmath262 considering the fields @xmath66 , @xmath67 as smooth ones and using the boundary condition @xmath263 the instanton equations ( [ ie1],[ie2 ] ) take the form
@xmath264 here we approximate the shocks as jump discontinuities , and the condition ( [ bcup ] ) relates the heights of the jumps ( see fig .
[ fig5 ] ) . here
the diffusion terms and the term @xmath213 are omitted .
one can check that they are negligible since the characteristic @xmath86-scale of the solution is large enough .
the equations ( [ equp ] ) can be integrated by characteristics ( or lagrangian trajectories ) .
the velocity of the shocks is equal to @xmath265 , i.e. all the trajectories disappear at the shocks ( if we consider the evolution back in time ) .
the value of @xmath66 ( or @xmath67 ) is conserved in time if we follow the lagrangian trajectory .
this means that the relation @xmath266 holds everywhere between the shocks .
due to self - advection the velocity field @xmath66 becames more and more linear as a function of @xmath86 while @xmath88 increases .
this happens at the border between the 2nd and the 3rd regimes . in the third regime
we can take @xmath267 between the shocks .
the field @xmath67 is equal to @xmath268 .
the velocity @xmath66 simply squeezes or expands the field @xmath67 without changing its shape . since @xmath269 , the field @xmath67 should have the same scaling as @xmath270 , i.e. @xmath271 . the concave form of @xmath67 ( as one can see @xmath112 is not yet good enough for a clear picture ) is shown in fig . [ fig7 ] .
let us determine the time dependence of @xmath272 .
we have @xmath273 solving this equation we get @xmath274 . since @xmath275 at @xmath276 , we get @xmath277 .
the shocks heights are equal to @xmath278 , @xmath279 .
the width of the shocks in @xmath67 field is of the order of @xmath280 .
the shocks in velocity field @xmath66 have the width of the order of @xmath229 , since @xmath66 is pumped by @xmath213 .
( solid curve ) and @xmath281 ( dashed curve ) profiles at @xmath282 .
@xmath112.,width=288 ] at large time @xmath283 the height @xmath247 ( and consequently the width of the shocks in @xmath67 ) becames of the order of @xmath229 .
this indicates the end of the third regime and the beginning of the fourth .
going further back in time we finally enter the domain of validity of the equations ( [ equv ] ) .
the solution falls into @xmath221 .
again , the solution has two shocks between which it is a smooth function .
the shock position satisfies @xmath284 , between the shocks @xmath285 holds .
this regime exactly corresponds to self - similar solution @xmath286 of inviscid burgers equation @xmath287 ( here @xmath288 is a parameter ) . at the far tail of the instanton due to viscous dissipation the solution @xmath221 transformes to the derivative of gaussian contour the self - similar solution of the diffusion equation ( see fig .
[ fig4 ] , @xmath289 ) . during the fifth regime
the advection term @xmath246 becomes irrelevant . in @xmath141
, @xmath143 variables the solution tends to @xmath290 , that was observed in numerical calculations . during the 4th and the 5th regimes
the amplitude of the velocity field @xmath66 is less than unity .
it means that the amplitude of @xmath66 is lower than the order of the typical statistical fluctuations , and the saddle - point approximation is meaningless there .
the typical events that demonstrate large negative gradients start from some velocity configuration @xmath291 with the amplitude of the order of unity and are governed by the 3rd regime first .
the action on these events and their further evolution almost do not depend on the initial velocity field @xmath291 , and the dependence of @xmath19 on @xmath21 remains unaltered by the averaging over all possible configurations @xmath291 .
we considered the 4th and the 5th regimes since they are the parts of the whole solution of our nonlinear boundary problem .
let us now run the whole evolution forward in time . at the beginning ( 5th regime )
the field @xmath66 is pumped by very wide @xmath67 .
the pumping force is proportional to @xmath292 . during the 4th regime
the source is localised at shocks in @xmath67 that leads to a formation of shocks in @xmath66 .
the @xmath66 s shocks want to break down because of self - advection , but the source term @xmath293 keeps them going . when the growing height of @xmath67 becomes larger than unity the shocks in @xmath67 become narrow . the balance between the terms @xmath246 and @xmath294 in ( [ ie1 ] ) changes a little , that results in the change of the form of @xmath66 s shocks it is determined by the shape of @xmath6 now .
the distance between the @xmath67 s shocks decreases in time and eventually it becomes comparable with unity . after
this @xmath67 becomes even more narrow and the efficiency of the source term begins to fall down .
the self - advection of the velocity destroyes the shocks and leads to a formation of the kink at @xmath232 , while @xmath67 transformes to @xmath295 . the kink shape at @xmath112 is shown in fig .
schematically time evolution of the instanton is illustrated in fig .
[ fig8 ] . one can present the action @xmath59 in the form of @xmath296 like in expression ( [ sextr ] ) . for @xmath112 the action density @xmath297 that was obtained from numerical calculations is shown in fig .
[ fig9 ] . while @xmath298 the convolution @xmath39 is localized at shocks , so @xmath299 .
the maximum of @xmath297 is posed at @xmath300 .
further increasing @xmath87 leads to the decreasing of the density @xmath297 because @xmath301 becomes more and more narrow without an adequate growth of its amplitude .
this region of small time @xmath302 was studied in @xcite .
it was shown that the contribution @xmath303 to the extremal action from this interval is of order @xmath304 , and the main contribution to the action from it comes from the region @xmath300 the border between the 2nd and the 3rd regimes . exactly these two regimes determine the optimal configuration of noise providing the event with large negative gradient .
the contribution of region of time @xmath305 ( 3rd regime ) to the extremal action @xmath59 can be estimated as @xmath306 note that the value of @xmath307 is again cumulated from the region @xmath300 .
the crucial point is that the contribution to the action @xmath59 from the tail of the instanton ( or large time @xmath308 ) is finite , i.e. the integral ( [ sd ] ) converges ( the addition to the action from interval @xmath309 is negligible ) . also this contribution is not dominant , i.e. it is not much greater than the contribution of the order @xmath304 from small times ( @xmath302 ) .
it means that in our case the instanton is localized enough in time .
its long - time dynamics does not destroy the fact that it is the main fluctuation determining the statistics of large negative gradients . as a function of `` time '' @xmath143 .
@xmath112.,width=288 ] ( see ( [ fsa ] ) ) as a function of gradient @xmath62.,width=288 ] at the fig .
[ fig10 ] the function @xmath310 that was obtained from numerical calculations is shown .
we used different grid parameters during calculations for instanton structure and for this figure .
here we used @xmath311 with boundary condition @xmath312 .
this boundary condition was used as initial condition for @xmath95 during iterations .
it turned out , that e.g. for @xmath313 we get the following values of @xmath314 and @xmath62 with different value of @xmath138 : [ cols="^,^,^",options="header " , ] although , when calculating with @xmath315 , the condition @xmath316 holds within 15% ( as one can see from fig . [ fig4 ] )
we prefer to use the grid shorter in time ( @xmath311 ) to have smaller time step .
the value of @xmath62 strongly depends on time step .
it was observed in numerical experiment .
such sensitivity is characteristic of burgers equation also .
although the calculations with small @xmath317 give us worse accuracy at the tail of the instanton , the smallness of the time step allows us accurately describe the main part of the instanton where the nonlinearity level is high .
one can see the cubic asymptotics @xmath318 at @xmath319 .
the instanton structure for @xmath320 that was described in @xcite was confirmed by our numerical calculations . the case @xmath321 corresponding to the pdf s left tail is more complicated .
the function @xmath314 has minimal value at @xmath322 . at further decrease of @xmath62
it starts to grow and finally tends to the value @xmath323 . in this case
the coefficient @xmath324 is small .
we have examined the remote left tail of the velocity gradients pdf @xmath7 in burgers forced turbulence . the possibility of direct numerical solving of instanton equations by iterations is demonstrated .
numerical calculations and the analysis of the instanton behavior at the time large compared with its lifetime @xmath325 with the solution at small time from @xcite show that @xmath0 .
we are grateful to g.e .
falkovich , a.v .
fouxon , i.v .
kolokolov , v.v .
lebedev and e.v .
podivilov for useful discussions .
this work was partially supported by russian foundation for basic research ( gr .
98 - 02 - 17814 ) , by intas ( m.s . , gr .
96 - 0457 ) within the program of international center for fundamental physics in moscow , by the grants of minerva foundation , germany and mitchell research fund ( m.s . ) .
99 e - mail : [email protected] e - mail : [email protected] j.m .
burgers , _ the nonlinear diffusion equation _ ( d. reidel , dordrecht , 1974 ) .
a.m. polyakov , phys . rev .
e * 52 * , 6183 ( 1995 )
. v. gurarie and a. migdal , phys
e * 54 * , 4908 ( 1996 ) .
e. balkovsky , g. falkovich , i. kolokolov , and v. lebedev , phys .
. lett . * 78 * , 1452 ( 1997 ) .
boldyrev , phys . rev .
e * 55 * , 6907 ( 1997 ) .
w. e , k. khanin , a. mazel , and ya .
sinai , phys .
* 78 * , 1904 ( 1997 ) .
t. gotoh and r.h .
kraichnan , phys .
fluids * 10 * , 2859 ( 1998 ) . m.v .
feigelman , zh .
. fiz . * 79 * , 1095 ( 1980 ) [ sov .
jetp * 52 * , 555 ( 1980 ) ] .
r. kraichnan , phys .
fluids * 11 * , 3738 ( 1999 ) ; w. e and e. vanden eijnden , phys .
fluids * 12 * , 149 ( 2000 ) .
lifshitz , usp .
nauk * 83 * , 617 ( 1964 ) [ sov .
uspehi * 7 * , 549 ( 1965 ) ] .
lipatov , zh .
. fiz . * 72 * , 411 ( 1977 ) [ sov .
jetp * 45 * , 216 ( 1977 ) ] .
g. falkovich , i. kolokolov , v. lebedev , and a. migdal , phys .
e * 54 * , 4896 ( 1996 ) .
m. chertkov , phys .
e * 55 * , 2722 ( 1997 ) ; e. balkovsky and v. lebedev , phys .
e * 58 * , 5776 ( 1998 ) .
kolokolov , pisma v zh .
teor . fiz . *
71 * , 20 ( 2000 ) [ jetp letters * 71 * , 12 ( 2000 ) ] .
martin , e.d .
siggia , and h.a . rose , phys .
rev . a * 8 * , 423 ( 1973 ) ; c. de dominicis , j. de physique * 37 * , colloque c-1 , 247 ( 1976 ) [ in french ] ; h. janssen , z. phys .
b * 23 * , 377 ( 1976 ) ; c. de dominicis and l. peliti , phys . rev .
b * 18 * , 353 ( 1978 ) .
keldysh , zh .
* 47 * , 1515 ( 1964 ) [ sov .
jetp * 20 * , 1018 ( 1965 ) ] .
arnold , _ geometrical methods in the theory of ordinary differential equations _
( springer , new york , 1988 ) . a.a .
samarskii , a.v .
gulin , _ numerical methods _
( nauka , moscow , 1989 ) [ in russian ] .
r. temam , _ navier - stokes equations .
theory and numerical analysis . _
( north - holland publishing company , amsterdam new york oxford , 1979 ) .
fletcher , _ computational galerkin methods _
( springer - verlag , new york berlin
tokio , 1984 ) . | we consider 1d burgers equation driven by large - scale white - in - time random force .
the tails of the velocity gradients probability distribution function ( pdf ) are analyzed by saddle - point approximation in the path integral describing the velocity statistics .
the structure of the saddle - point ( instanton ) , that is velocity field configuration realizing the maximum of probability , is studied numerically in details .
the numerical results allow us to find analytical solution for the long - time part of the instanton .
its careful analysis confirms the result of [ phys .
. lett . * 78 * , 1452 ( 1997 ) ] based on short - time estimations that the left tail of pdf has the form @xmath0 .
pacs numbers : 47.27.gs , 05.10.-a 2 |
dynamics on complex networks , such as the diffusion of information in a social networks , are commonly modelled as markov processes .
an advantage of this approach is that for every ( static ) network with positive edge - weights we can define a corresponding markov process by interpreting the network as the state space of a random walker , and assigning the state - transition probabilities according to the link weights .
this direct correspondence between the state space of the markov process and the network enables us to examine the interplay between structure and dynamics from two sides . on the one hand
, one can assess how the topological properties of a network influence the dynamical process .
on the other hand , this coupling between topology and dynamics allows us to explore the structure of a network by means of a dynamical process .
specifically , for a linear markov process the impact of the network structure on the dynamics will be mediated by the spectral properties of the matrix governing the time - evolution of the process , e.g. the adjacency matrix or the laplacian @xcite .
reversely , spectral properties can be used to uncover salient structural properties of a network , such as modular organisation @xcite .
while simple markov models have been very successful in modelling dynamics of complex systems and found many applications , they have one obvious disadvantage . in this class of models ,
the future state of the system only depends on its current state and does not account for its history . in a diffusion process , for instance
, the next position of a random walker only depends on the currently occupied node and its outgoing links , but not on any of the previously visited nodes
. however , as it has been emphasised recently , for a broad range of networked systems , flows tend to exhibit a temporal path dependence @xcite .
think of human mobility : the places a person is likely to visit next , often depend strongly on where the person came from .
for instance , a person coming to work from home is likely to return home afterwards @xcite .
other examples of processes with temporal memory include web traffic , journal citation flows and email cascades .
such processes therefore can not be reproduced accurately by simple markov models
. however , the impact of this temporal correlations can often be well - approximated already by second - order markov ( @xmath0 ) models @xcite .
importantly , the transition probabilities to define these models can be obtained empirically by measuring pathways of interaction cascades , rendering such an approach suitable for applications like information spreading or human mobility . in the standard markovian network model ( @xmath1 ) ,
the elementary states are identified with the nodes of the original network ( see figure [ fig1 ] ) . in the @xmath0 model ,
elementary states correspond to sequences of two nodes , and thus can be identified with _ directed edges _ in the original network ( see figure [ fig1 ] ) .
therefore we may think of a @xmath2 model alternatively as a random walk between directed edges of the original network .
the state space of the @xmath0 model defines a new network describing the observed dynamics , which we call here the @xmath0 or memory network .
the structural properties of the memory network can now be studied by many tools of network science , allowing us to uncover interesting patterns of flow in an @xmath0 model .
( right side , top panel ) , the state space is isomorphic to the physical network : every node corresponds to one state , every link indicates a transition between those states . in a second order markov model
the state space is structured like a _ directed line - graph _ of the original network .
the states in this @xmath2 network can be identified with the directed edges in the original network and are connected if it is possible to traverse from one edge to another edge in the original directed network .
note that , when projecting back these @xmath2 dynamics onto the physical network , the probability to move from one node to another will thus appear non - markovian . ]
the purpose of this paper is to highlight how the formalism of higher order markov models provides a simple means to extend network theoretic tools to take into account important dynamical properties as encoded in the @xmath0 representation .
indeed many network theoretic tools may be understood as the outcome of a matrix iteration or eigenvector computation @xcite , which can be naturally associated to a dynamical process .
in this paper we focus on the problem of community detection , though higher - order models can be equally applied to other problems , including the assessment of node centralities @xcite or the speedup ( or slowdown ) of spreading processes @xcite . while ref .
@xcite discussed how the map equation can account for higher - order flows in community detection , here our main example will be the markov stability @xcite formalism , as it incorporates many commonly used community detection measures as special cases , notably , the concept of modularity @xcite , diverse potts models @xcite , and spectral clustering @xcite .
we thereby highlight that all these classical algorithms can indeed be naturally generalised to account for memory effects by using @xmath2 networks , showing that these representations provide a general tool for the analysis of a dynamics occuring on a latent network structure .
moreover , as spectral clustering can be shown to be a special case of the markov stability formalism for long times , we can draw further connections to recently proposed techniques of graph partitioning based on non - backtracking random walks @xcite , and discuss how these can be understood from the point of view of second order markov processes . we remark that as clustering an @xmath0 network is equivalent to finding a partition of the edges of the original system , it naturally leads to the detection of overlapping communities in the same way as the analysis of line graphs @xcite .
this can be a highly desirable feature , especially when analysing social networks , which tend to be organised in overlapping social circles @xcite .
the remainder of this article is organised as follows .
initially , we review the markov stability formalism for community detection , a general framework to detect flow - based communities in complex networks , and particularly emphasise its properties in the case of directed networks @xcite . we show how this quality function can be naturally generalised for the analysis of @xmath0 networks , which we illustrate by studying a flight network of the united states from the perspective of second order markov models . in this context
, we also discuss the possibility to generate realistic pathway data using models of second - order markov processes , even if only time aggregated network information is available .
subsequently , we propose a mechanism to extract pathway statistics to build second - order models from event - based , temporal network data , which do not contain pathway statistics _ a priori_. using computer - generated data and time - resolved interactions records of school - children ,
we demonstrate how we can uncover communities which capture important flow - constraints imposed by the temporal activation patterns of the links .
finally , we draw connections between the @xmath2 network representation and non - backtracking random walks to uncover communities in sparse networks .
let us consider a continuous - time random walk on a network with @xmath3 nodes , governed by the following kolmogorov forward equation : @xmath4 here @xmath5 denotes the probability of a walker to be present on node @xmath6 at time @xmath7 ; @xmath8 is the weight of the link from node @xmath9 to node @xmath6 , i.e. , @xmath10 is the ( weighted ) adjacency matrix ; finally , @xmath11 is the weighted out - degree of node @xmath6 .
clearly this dynamics is driven by the normalised laplacian matrix @xmath12 , defined as @xmath13 .
the perron - froebenius theorem guarantees that a unique stationary solution exist for the above dynamics if the network is strongly connected . for an undirected network ,
this stationary solution is @xmath14 , where @xmath15 is the node degree and @xmath16 is the total weight of the undirected links . for a general directed network
the stationary solution is given by the dominant ( left ) eigenvector of @xmath12 , which depends on the global network properties and can not be expressed by a simple analytical formula .
[ [ ergodic ] ] making the dynamics ergodic ( when it is not ) .
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + note that for the continuous - time random walk defined above , a strongly connected graph implies an unique stationary distribution and an ergodic dynamics .
these criteria are a common requirement for dynamics - based methods for network analysis @xcite .
however , in a majority of real - world systems the system contains not just one but several strongly connected components ( sccs ) , whose sizes typically have different orders of magnitude .
a standard solution is to address this issue by either neglecting some nodes and focusing only on the largest scc only , or by making the system strongly connected through random teleportations @xcite .
the first solution has the advantage of not distorting the dynamics , while the second allows for the analysis of the whole system .
when analyzing real - world systems in the following , we will for these reasons use the first solution if the vast majority of nodes belongs to the largest scc , while we will opt for the second solution if the system comprises a large number of disconnected components .
the structure of a network has a strong effect on the dynamics of a diffusing random walker : in unstructured ( random ) networks , the diffusion process will evolve almost isotropically and the walker will quickly reach its stationary distribution .
however , if the network contains structure , a walker can get trapped inside a group of nodes for a time far longer than expected .
such groups of nodes thus constitute flow - retaining , _ dynamical communities _ in the network .
hence a random walk dynamics can effectively be used to define a quality function for a network partition based on the persistence of the diffusion inside the groups . by optimising such a quality function ,
we can therefore search for a modular partition of the network .
important examples for such an approach include the map equation @xcite and the markov stability framework @xcite , which we consider in this paper .
note that , as highlighted in refs .
@xcite this notion of community is markedly different to the commonly considered structural viewpoint of communities , in that we are interested in flow - retaining , rather than densely ( homogeneously ) connected substructures .
the markov stability of a partition @xmath17 is defined as : @xmath18 where @xmath19 is the probability for a walker to be in community @xmath20 initially and at time @xmath7 , for a system at stationarity .
note that as all information about the initial position is lost after infinite time , @xmath21 also describes the probability of two independent walkers to be in @xmath20 . for the process , the markov stability quality function can be written explicitly as : @xmath22,\ ] ] where @xmath23 denotes the matrix exponential .
intuitively , with increasing time the walker will be able to explore larger and larger parts of the graph .
the larger the time , the larger the modules that will be found in general . by ` zooming ' through different markov times one can thus reveal adequate scales of robust community structure , which manifest themselves as plateaux in time where the same partitions are found consistently @xcite . in the limit @xmath24
, it can be shown by eigenvalue decomposition that @xmath25 is maximised by a bi - partition in accordance with the normalised fiedler eigenvector , a classic method in spectral clustering @xcite .
interestingly , the expression for markov stability can be rewritten as the newman - girvan modularity @xcite of a _ different _ network , whose adjacency matrix is given by @xmath26.\ ] ] here @xmath27 is the flow of probability from @xmath9 to @xmath6 after a time interval @xmath7 at stationarity . in this new network @xmath28 , the weight of a connection between two nodes is thus modulated according to its importance for the diffusion dynamics .
this rewriting of markov stability as the modularity of a different network opens the possibility to use any modularity based algorithm for the optimization of stability , too . for the results in the present manuscript ,
we have made use of the louvain algorithm @xcite , a fast greedy optimisation heuristic that has been shown to have a good performance in the literature . in practice
, it can be more efficient to consider ( [ rt ] ) in the limit of small times @xmath29 . performing a taylor expansion and keeping only linear terms in @xmath7 , results in the linearised markov stability : @xmath30.\ ] ] note that for an _ undirected _ network , we have @xmath31 and @xmath32 , and one recovers the modularity of the original network for @xmath33 as well as the potts - model heuristic @xcite for other markov times ( see @xcite for details ) .
however , for _ directed _ networks is not equivalent to the directed modularity of the original network , but is given by : @xmath34.\ ] ] interestingly , similar adjusted adjacency ( or laplacian ) matrices have been proposed as means for community detection in the literature based on different reasoning @xcite .
note that , as with all applications of unsupervised algorithms , care is needed when interpreting the results .
for instance , the optimisation of markov stability for a fixed markov time inherits the limitations of modularity maximisation , such as a resolution limit @xcite , and a high degeneracy of optimial solutions @xcite .
nevertheless , this class of method has been applied successfully in numerous applications @xcite .
indeed , some of these problems can be circumvented by carefully sweeping through different resolution parameters @xcite and finding partitions that are robust over a range of parameters @xcite .
most importantly , however , the notion of community inherent to the methods presented here is based on flows @xcite .
this is in contrast to methods like stochastic blockmodels ( sbm ) , see e.g. ref .
@xcite , which employ a generative model for the whole network and aim at finding patterns of pairwise connections in the system , though very recent work aims to incorporate dynamical aspects within the sbm framework @xcite .
let us now show how we can incorporate second - order markov models into the markov stability framework . as higher - order markov models provide more faithful representations of the dynamics observed in real world systems , this enables us to capture more of the real flow constraints in the uncovered communities .
for simplicity let us initially consider undirected networks composed of @xmath3 nodes and @xmath16 links .
the dynamics of a second order markov process are encoded by the transition matrix @xmath35 , which describes the probability that a walker moves from node @xmath6 to node @xmath36 if it came from @xmath9 in the previous step . by definition ,
this transition matrix is normalised such that @xmath37 . as sequences of two vertices in our original network correspond to the nodes in the @xmath2 network ,
the entries in @xmath38 describe precisely the transitions from @xmath0-node @xmath39 to @xmath0-node @xmath40 .
the ( second - order ) @xmath2 process on the original network is thus equivalent to a first - order markov process , albeit on a different network : namely , the @xmath0 network composed of @xmath41 nodes . as each undirected link of the original network
can be traversed in two distinct directions ( from @xmath9 to @xmath6 , and vice versa ) it accounts for 2 nodes in the @xmath0 network , @xmath39 and @xmath42 .
this implies that the @xmath2 network is _ directed _ even if the original network is undirected . to see this ,
observe that if @xmath43 , there can not be a link between @xmath40 and @xmath39 , even if a transition between @xmath39 and @xmath40 exists .
henceforth we use greek letters to denote @xmath0 nodes , and latin characters to denote the nodes in the original @xmath1 network . similarly to ( [ ctrw ] )
, we can define a continuous - time random walk on the @xmath0 network as : @xmath44 where @xmath45 is the probability of finding a walker on @xmath0 node @xmath46 at time @xmath7 .
likewise the stationary distribution @xmath47 of this process is given by the left eigenvector of the corresponding laplacian , associated with eigenvalue @xmath48 : @xmath49 = 0.\end{aligned}\ ] ] note that we can also observe how a simple @xmath50 markov dynamics evolves from the point of view of an @xmath2 network , i.e. , we can view the @xmath50 dynamics from the point of view of transitions between the ( directed ) edges of the graph .
as this is equivalent to lifting the @xmath50 dynamics into the larger @xmath2 state - space , we will denote this representation as a @xmath51 network . in this case
it can be shown easily that @xmath52 , where @xmath53 is the weight of edge @xmath46 and @xmath54 @xcite .
the transition probability to any out - neighbour of @xmath46 on the @xmath55 network is simply @xmath56 here @xmath57 is the set of out - neighbours of @xmath46 , and @xmath58 is the out - strength of @xmath46 . note that , since the original network has been undirected we have @xmath59 , where @xmath6 is the endpoint of memory node ( the directed edge ) @xmath60 . while second order transition matrices @xmath38 can be directly generated from temporal pathway data , in some cases only information about the aggregated ( first - order ) network might be available .
however , in such cases we can still create @xmath2 networks using a simple memory model as described in the following , calibrated by the pathway statistics of similar datasets for which this temporal information is obtainable .
for instance , the temporal information pathways of one email dataset may be used to fit model - parameters to generate a second - order model for different email communication network , where only aggregated information is obtainable .
the key idea underpinning the model is to weight different types of transitions between two directed edges .
as illustrated in figure [ figr ] , we define three different types of transitions @xcite . 1 . a _ return step _ , in which a walker coming from @xmath39 jumps to @xmath42 . in other words : a walker coming from node @xmath9 to @xmath6 returns to node @xmath9 .
2 . a _ triangular step _ , in which a walker coming from @xmath39 moves to edge @xmath40 , where @xmath43 is a neighbor of @xmath9 .
an _ exploratory step _ , in which the walker moves from @xmath39 to an edge @xmath61 , whose endpoint @xmath62 is neither @xmath9 , nor any of the neighbors of @xmath9 .
network using a second - order transition model .
* from each directed edge in the original network a walker can in principle perform three types of moves , which get weighted according to the parameters @xmath63 : a _ return step _ ( @xmath64 ) , a _ triangular step _ ( @xmath65 ) and an _ exploratory step _ ( @xmath66 ) .
the picture shows these different options for a generic network edge . ] to account for their relative importance we assign positive weights @xmath64 , @xmath65 and @xmath66 to the different types of transition as follows .
let us denote the adjacency matrix of the directed line graph associated with our network by @xmath67 .
then we can decompose @xmath67 into three matrices @xmath68 ( return links ) , @xmath69 ( triangular ) , and @xmath70 ( exploratory ) each containing only the links of the respective type . to obtain a weighted memory model we now define the weighted @xmath71 adjacency matrix : @xmath72 from which we can compute the associated second order transition matrix @xmath73 simply by normalising each row to sum to @xmath74 .
as can be easily verified , this definition implies that when we project the resulting walk onto the node space , we obtain a ( first order ) markov process when @xmath75 . we further remark that if the dynamics is ergodic on a graph for any set of parameters , it will remain ergodic for any value of @xmath64 , @xmath65 and @xmath66 , provided each parameter is strictly positive .
let us now demonstrate how second - order markov dynamics can help to uncover the organisation of systems where pathway data are available , and compare their outcome with the community structure obtained by first - order dynamics and by simplified models of second - order markov dynamics , as defined in section [ sec : mem_models ] . to do so ,
we consider a flight network in the united states .
the data used to construct the network consists of individual flight trajectories of people navigating between different airports in the us . from these trajectories
one can directly obtain a @xmath50 or @xmath2 network as discussed above and in ref .
the main purpose of our analysis here is to illustrate the differences that may arise in community detection when considering the same mobility data from different modelling perspectives .
we analysed the modular structure of this system for the following three scenarios : i ) a first order markov network ( @xmath76 ) , viewed from the perspectives of the edges ; ii ) a second order markov network ( @xmath0 ) where the transition probabilities are directly obtained from empirical data ; iii ) a second order markov network ( @xmath77 ) , where transitions are approximated by a simple second - order transition matrix @xmath78 ( see text ) , whose parameters have been fitted from the data . in each case , optimising the linearised stability for these different markov models yields a clustering of the edges , which can be interpreted in terms of an overlapping community structure at the airport level . to compare the different scenarios we concentrate on the results obtained for markov time @xmath33 ( see fig . [ airports ] ) .
let us also note that , for each network , only few hardly active airports do not belong to the largest scc .
for this reason , as argued in section [ ergodic ] , we restrict the scope to nodes in this scc . to be more precise ,
we restrict the scope to the intersection of the largest sccs of the three processes , in order to ensure that stability is well - defined for each of them . in order to compare the communities associated to each markov process
, we first calculate the normalised variation of information @xcite between the three different partitions of @xmath0 nodes obtained by optimising stability at @xmath33 . by construction ,
smaller values of variation of information indicate more similar community structures .
we obtain a variation of information of @xmath79 between @xmath0 and @xmath76 , @xmath80 between @xmath0 and @xmath77 , and @xmath81 between @xmath76 and @xmath77 .
these results confirm that the clusters obtained from @xmath77 provide an intermediate solution between @xmath0 and @xmath76 , with the advantage of requiring far fewer parameters than in the @xmath0 case .
after having found the edge - communities in each scenario , each node can be characterised by the set of group labels of its incident edges . in order to quantify the apparent difference between the covers
, we measure for each node the entropy of its associated group labels .
@xmath82 where @xmath83 is its fraction of edges assigned to community @xmath84 .
a value @xmath85 indicates that the node is associated to a single community , while higher values indicate a more diverse participation .
the level of overlap of an edge partition can now be characterised by the distribution of entropy values on the set of nodes .
we observe in figure [ airports ] that analyzing the @xmath0 network results in more overlapping communities than the expanded first - order model , while the @xmath77 network is characterized by intermediate participation values .
similarly as observed for the map equation @xcite , accounting for memory via second order dynamics therefore uncovers communities with a stronger overlap , in agreement with empirical observations that higher - order dynamics tends to constrain flows within these modules .
node or link activity in networked systems often exhibits non - trivial temporal patterns , such as heterogeneous inter - event times and correlations between activations times of neighbouring edges @xcite .
different network representations of such temporal data can be adopted , each associated to a different notion of temporal community .
a first , ` naive ' approach is to neglect the edge - timings and work with a static network representation , where the weight of each edge corresponds to an aggregation of the activity over the whole observed time interval . in this case ,
community detection aims at grouping nodes which have the aggregated edge - weight concentrated inside modules .
alternatively , one can represent the system by a set of ( coupled ) adjacency matrices @xmath86 , where each @xmath86 represents the system in a short - time frame of the whole observation period . by applying community detection to this representation one
tries to uncover meaningful structures in each time window , while enforcing some continuity between different time intervals @xcite .
this approach accordingly aims at tracking the evolution of communities in the course of time . as we now discuss , a third notion of community is naturally associated to the representation of dynamics on temporal networks as a second - order markov process . in situations
when the activations of neighbouring edges present temporal correlations , the dynamics of a random walker is expected to be poorly reproduced by a first - order markov process @xcite . however , by finding a second - order representation of the temporal data and using the methodology introduced above , one can capture the observed temporal constraints on flow , including causal paths and a high levels of synchrony between edge activity .
the mapping from a time series into a second - order markov model is realized as follows .
we consider a system described by an ordered set of adjacency matrices @xmath86 defining the connections between nodes at time @xmath87 $ ] , where @xmath38 is the number of observations
. the static representation of the temporal network is provided by the adjacency matrix @xmath88 .
each directed edge with in @xmath89 defines a @xmath0 node . the connection strength between @xmath0 nodes
is now obtained by simulating pathways of a random walk process on the temporal network as follows .
a walker is initially randomly assigned to a node . at every time
step the walker waits until at least one edge is available for transport . if an edge becomes available
, the walker leaves the node with probability @xmath90 and remains on its current node with probability @xmath91 .
if there are multiple possible transitions , the walker takes each edge with a probability proportional to its weight .
this process is repeated multiple times for the observed interval @xmath92 $ ] in order to generate sufficiently many trajectories . from these trajectories
, one can simply construct a @xmath0 network by evaluating the transition frequencies between different edge pairs .
we remark here that several other procedures could be defined to generate @xmath0 networks , too .
for instance , we could account for the duration of the intervals between time steps , or define the walker s leaving probability to be proportional to the number of available contacts . note that in the above construction , when @xmath93 , the walker always takes the first available edge , and one recovers the dynamics studied in e.g. ref .
the use of non - zero values of @xmath94 introduces a source of randomness in the dynamics , which can prevent spurious effects such as an overly strong tendency for backtracking as observed in @xcite .
however , with increasing values of @xmath94 , the original ordering of events becomes less important , and the impact of the exact timings is expected to be diluted @xcite .
let us now show that a @xmath0 representation helps at uncovering hidden structures in temporal networks .
as a first illustrative example , we consider a computer - generated benchmark of a social network defined as follows .
the underlying structure is a complete graph , where the ( directed ) social interactions are divided into @xmath95 different , non - overlapping types , e.g. , work relations , friends , etc .
[ ex_graph ] a shows such a network with @xmath96 nodes and @xmath97 edge types .
we assume that different types of relationships dominate at different times , as often observed in empirical data @xcite .
therefore , at each time , all edges of one single group are simultaneously active .
initially , a randomly chosen group is active . at each step ,
the currently active group remains active with probability @xmath98 ; with probability @xmath99 , a new active group is randomly selected among all groups ( including the currently active one ) . for the parameters considered here , the walker thus has three outgoing edges available for transport at each step ( see figure [ ex_graph ] ) .
it is important to note that for the generation of trajectories on this temporal network , a change of the probability for the walker to stay at its node @xmath94 is effectively equivalent to a change in @xmath98 . for the sake of simplicity
, we therefore set @xmath100 here .
clearly , the trajectories of the random walker and the corresponding structure of the @xmath0 network are affected by the value of @xmath98 .
when @xmath101 , a new active group is randomly selected at each step of the random walker . in this limit , the average dynamics is equivalent to a first - order random walk on a fully connected network , and thus no underlying structure can be found .
increasing @xmath98 implies that the random walker remains for longer time periods inside one group of edges before leaving , and it is thus possible to uncover the group structure .
the extreme case @xmath102 is again trivial : the initial group remains active for all times .
this case is therefore not considered in the following .
the results of our analysis are summarised in fig .
[ ex_graph ] , where we compare the uncovered communities with the planted solution into groups by means of normalised variation of information ( vi ) .
our observations are twofold .
first , for values of @xmath7 slightly larger than @xmath33 , we find the planted partition with high accuracy .
second , one observes that the underlying communities can be successfully found when the value of @xmath98 is sufficiently large , with a clear transition around @xmath103 ( see inset in figure [ ex_graph ] ) . to demonstrate the utility of our approach with a real - world example ,
we have considered an empirical dataset from the sociopatterns project @xcite .
the data , described in detail in @xcite , consists of time - dependent face - to - face contacts , captured by wireless wearable sensors , between children and teachers in a primary school . in total , the dataset is made of 77,602 contact events between 242 individuals during two consecutive days . as the children belong to 10 different classes , we expect to find strong structural communities associated to these classes in the dynamic contact network . in the following analysis ,
the pathways were generated with a probability for the walker to stay at his node at each step set to @xmath104 , but similar results were obtained when varying this parameter . by scanning through markov times we can find several persisting partitions .
as validation we first identified the markov time @xmath105 around which the algorithm robustly detects @xmath106 communities , and verified that the communities overlap with the partition into classes of the students .
one should note here that the partition into classes can in fact be uncovered even from the aggregated , weighted network associated to the data @xcite .
the advantage of our approach becomes apparent when looking at further stable partitions found for smaller markov times .
for instance , we find another robust split of the @xmath0 network at markov time @xmath107 into @xmath108 communities .
for this partition most communities are still identified with classes , as can be seen in fig .
[ spstand ] . however , there are also additional modules in which students from different classes are mixed .
these modules display activity patterns which are highly localized in time and therefore are bound to get lost when averaging over the temporal dimension in the data .
interestingly , these latter communities correspond in fact to interactions taking place between students during lunch breaks .
notably , these interaction are not detectable from the aggregate network or when using a memoryless representation of the dynamics ( e.g. , the @xmath51 network ) . by using the approach outlined here , we can thus not only detect structural communities , in which actors maintain a higher amount of interactions between each other throughout time , but also temporally localized communities , which are associated to groups that interact with each other strongly over certain time - periods .
[ splunch ]
recently , there has been an increased interest in spectral methods based on the so - called non - backtracking matrix of a network . while standard spectral algorithms for graph partitioning tend to fail when applied to very sparse networks , krzakala _ et al .
_ @xcite demonstrated how using the non - backtracking matrix one can design spectral algorithms which behave optimally right until the theoretical limit of detectability of a stochastic block model .
as the name suggests , the non - backtracking matrix is intimately related to the non - backtracking random walk on a network .
the non - backtracking random walk is a diffusion process in which a particle behaves just like a simple random walker on the network , albeit with one additional constraint : after arriving at a node the walker is not allowed to immediately return to the node from which she originated ( she can not ` backtrack ' ) . interestingly , this non - backtracking random walk can be simply phrased in terms of a memory dynamics . following the notation of section [ sec : mem_models ] the non - backtracking matrix
is defined on the set of @xmath0-nodes and may be simply written as @xmath109 clearly this matrix describes the line graph for a memory walker with parameters set to @xmath110 ( see section [ sec : mem_models ] ) .
the associated transition matrix @xmath111 of the non - backtracking random walk is simply obtained through normalisation : @xmath112 where @xmath113 is the weighted out - degree of @xmath0-node @xmath46 .
as this is just a diagonally scaled version ( by the degree ) of the non - backtracking operator , we may employ the flow based transition matrix @xmath114 for spectral partitioning of the nodes @xcite .
in fact , @xmath114 is just one particular instance of a whole continuum of possible transition matrices , each with varying amounts of memory .
we can thus assess the effect the introduced memory has on the clustering by varying the parameters @xmath63 and performing a spectral clustering analysis .
nodes with a planted bipartition .
the average degree of the network vs. the spectral clustering performance for classifying the _ nodes _ as measured by the normalised mutual information ( nmi ) .
note that the theoretical detectability limit is given by @xmath115 ( red dashed line ) * ( a ) * clustering performance based on the adjacency matrix @xmath10 , the non - backtracking matrix @xmath116 , and the normalised laplacian matrix @xmath117 . *
( b ) * clustering performance based on different @xmath118 transition matrices with different parameter settings . ] like in @xcite , we consider the problem of spectral clustering here in the simplest possible setup , which is the detection of a bipartition of equally sized groups in a large , sparse network .
the networks we consider here consist of @xmath119 nodes , divided into two groups according to a simple stochastic block model @xcite .
we denote the average degree of each node by @xmath84 , the average degree to nodes inside its group by @xmath120 , and the average degree to nodes outside its own group by @xmath121 .
note that these quantities are coupled by the simple relation @xmath122 .
similar to @xcite , while keeping the ratio @xmath123 fixed , we vary the average degree @xmath84 and observe the results we obtain from different operators via spectral clustering . following @xcite , each node is grouped using the following protocol .
we first compute the second dominant ( left ) eigenvector of the respective matrix operator .
second , for each we sum over the eigenvector components corresponding to all of its incoming links .
the nodes are grouped according to the sign of this sum : if is is positive the node is assigned to group 1 if not to the second group . in case
the respective matrix operator is directly defined on the node space , the nodes are simply partitioned according to the signs of the eigenvector .
note that in contrast to the problems considered above we here look at the problem of community detection from the perspective of a hard clustering of the @xmath124 rather than the edges and moreover assume that there is a fixed , known number of non - overlapping node - communities .
we thus do not expect that using second - order transition matrices will necessarily improve the performance ( for the non - overlapping node - clustering ) , indeed accounting for memory can also be detrimental for this purpose as we will see in the following . however , our main purpose here was not to design an improved node clustering algorithm but to better understand the connection between spectral clustering and memory matrices . in figure
[ fig : spectral]a , we initially perform a spectral clustering analysis the non - backtracking matrix @xmath116 @xcite , the adjacency matrix @xmath10 , the normalised laplacian matrix @xmath125 as has been observed by krzakala et al .
, the clustering based on the backtracking matrix @xmath116 performs clearly better than the one based on the adjacency or laplacian matrix , in particular when the graph becomes very sparse ( see figure [ fig : spectral]a ) .
figure [ fig : spectral]b shows the results for the spectral clustering based on the memory walk matrices @xmath126 with various parameter settings @xmath63 .
there appears to be a clear trend : the more weight there is assigned to exploratory steps , the easier it appears to decide on the node groupings .
spectral clustering based on the non - backtracking second - order transition matrix @xmath127 performs almost as well as the clustering based on @xmath116 for very sparse networks . for networks with a higher average degree ( @xmath128 ) ,
the results are even slightly better .
as the second eigenvector of @xmath38 , which is used for the clustering , describes the approach of a memory dynamics towards stationarity it should not be too surprising that the results deteriorate if more memory is introduced : as very sparse graphs have a tendril like structure , additional memory can lead to a localisation of the dynamics in parts of the graph which correspond to strong local bottlenecks . stated differently , the slowest time - scale of the diffusion may not correlated with moving from one planted ( node ) community to the other , but local obstacles become more important making the second eigenvector a bad predictor of the bi - partition .
this is in particular interesting as for most memory dynamics reported @xcite , the return flow appears to be significantly large , which would imply that the dynamical constraints on the flow are much more pronounced on a local level than one may expect from the perspective of an aggregated network .
this observation appears to be aligned with the fact that many real - world systems tend to be composed of overlapping communities @xcite .
network - based models have been extremely successful for the analysis of complex systems , and have led to the discovery of structural features with profound impacts on its dynamics @xcite . however , to accurately capture the complexity of real - world interacting systems , simple network models are often not sufficient . for this reason ,
different approaches have been proposed recently , which increase the model dimension and enrich the network paradigm .
one approach that has gained prominence in the literature , is to account for the different types of interactions present in the system by means of multiplex or multilayer networks @xcite .
our approach developed in the work presented here proceeds in a similar spirit : by providing a larger state space representation in the _ temporal _ dimension , we aim to obtain new insights about the dynamical processes taking place in the system . despite their similarities
( see ref .
@xcite ) , multiplex networks and higher - order markov models differ , as the former model tends to emphasise differences in the connections of the system , while the second emphasises pathways . in this paper
, we have focused on the applicability of second - order markov dynamics in the context of flow - based community detection .
second - order markov dynamics model real - world temporal dynamics more accurately , so the found communities are expected to better capture the actual flow constraints in the system . in particular ,
we have shown how the markov stability framework , and thus techniques like potts models and spectral clustering , can be based on a second - order markov dynamics , which is equivalent to considering transitions between directed edges in the original network .
as clustering of second order models yields a partition of the edges , one can readily interpret the results as an overlapping clustering of the nodes in the original system , which is a beneficial feature of the approach .
some practical concerns are associated with moving to a second - order model .
naturally , there is an increased computational cost , as the number of nodes in the second - order model is equivalent to the number of directed edges in the original network .
however , as most networks are sparse , this cost tends to be outweighed by the benefits gained from a higher - order dynamical representation .
another practical issue is that pathway statistics , needed to construct a second - order transition matrix directly from data , may not always be immediately accessible . to make the toolkit of second order models available in such scenarios , we have presented two different strategies
first , we have analysed a simple model able to generate realistic second - order dynamics , which can be calibrated from similar datasets .
second , we have demonstrated how to represent temporal network data as a second order network .
we have tested these two strategies by focusing on a flight network in the united states , and a ( temporal ) social interaction network in a school environment , respectively . in both cases , the second order dynamics , even approximated , allowed us to extract temporal patterns in the data that would have been missed by an aggregated first - order model .
finally , we have highlighted the relationship between second - order markov models and the recently introduced spectral clustering formalism based on the non - backtracking matrix .
interestingly , the non - backtracking matrix corresponds to a scaled version of the transition matrix of a non - backtracking random walk , which is a special case of the second order dynamics discussed in this manuscript .
as we have demonstrated , this connection opens up the possibility to investigate further second order markov processes and their relationships to spectral clustering .
investigating these issues in more detail will be the subject of future work .
we acknowledge financial support from f.r.s - fnrs , arc and from the cost action td1210 knowescape .
this paper presents research results of the belgian network dysco ( dynamical systems , control , and optimismtion ) , funded by the interuniversity attraction poles programme , initiated by the belgian state , science policy office . | complex systems made of interacting elements are commonly abstracted as networks , in which nodes are associated with dynamic state variables , whose evolution is driven by interactions mediated by the edges .
markov processes have been the prevailing paradigm to model such a network - based dynamics , for instance in the form of random walks or other types of diffusions . despite the success of this modelling perspective for numerous applications
, it represents an over - simplification of several real - world systems .
importantly , simple markov models lack memory in their dynamics , an assumption often not realistic in practice . here , we explore possibilities to enrich the system description by means of second - order markov models , exploiting empirical pathway information .
we focus on the problem of community detection and show that standard network algorithms can be generalized in order to extract novel temporal information about the system under investigation .
we also apply our methodology to temporal networks , where we can uncover communities shaped by the temporal correlations in the system .
finally , we discuss relations of the framework of second order markov processes and the recently proposed formalism of using non - backtracking matrices for community detection . |
black holes have been the leading candidate to power the central engines in agn for over three decades ( lynden - bell 1969 ) , but direct evidence for their presence has been elusive . in nearby galaxies ,
spatially resolved kinematics have provided strong evidence for the ubiquity of nuclear black holes , with dynamical black hole detections reported for 37 galaxies ( kormendy & gebhardt 2001 ) .
such observations are available only for a handful of the nearest agn ( harms et al .
1994 , miyoshi et al .
1995 , greenhill et al .
1996 ) .
black hole mass , along with mass accretion rate , is a fundamental property of agn . via the eddington limit , a maximum luminosity for the idealized case of spherical accretion ( @xmath1 ergs s@xmath2 ) , the black hole mass sets an approximate upper limit to agn energetics .
it is also the integral of the accretion history of the agn .
however , direct kinematic observations of the black hole mass are limited by finite spatial resolution ( a typical agn at redshift 2 would require nano - arcsecond resolution to probe the sphere of influence of the black hole ) , not to mention that scattered light from the bright central source dilutes any kinematic signal from orbiting material . for these reasons ,
various less direct methods for estimating black hole mass have been devised .
one set of methods ( 2.1,2.2 ) assumes the broad - line region ( blr ) is gravitationally bound by the central black hole potential , so that the black hole mass can be estimated from the orbital radius and the doppler velocity .
the reverberation mapping technique utilizes the time lag between continum and emission lines to derive the distance of the blr from the black hole ( blandford & mckee 1982 , peterson 1993 ) .
about three dozen agn black hole masses have been measured using this technique .
a less costly alternative is to infer the blr size from the optical or ultraviolet luminosity ( mclure & dunlop 2001 , vestergaard 2002 ) , with which it is correlated , at least over a limited range of luminosities ( kaspi et al .
2000 ) . a different approach to estimating black hole mass is to exploit the correlation , seen in nearby normal galaxies , between black hole mass and stellar velocity dispersion , @xmath0 ( ferrarese & merritt 2000 , gebhardt et al .
if agn host galaxies are similar to non - active galaxies , this correlation should hold also for them . since
stellar velocity dispersion measurements are still difficult for higher redshift agn , the stellar velocity dispersion can possibly be inferred from effective radius and central surface brightness assuming agn host galaxies occupy the same fundamental plane as ordinary ellipticals ( odowd et al .
2002 ) .
some previous studies have found a tight relation between mass and luminosity in agn ( dibai 1981 , wandel & yahil 1985 , padovani & rafanelli 1988 , koratkar & gaskell 1991 , kaspi et al .
2000 ) ; however , the scatter is large when the black hole masses are restricted to the most reliable estimates ( from reverberation mapping ) .
one might have expected a correlation between agn black hole mass and luminosity since the eddington luminosity is proportional to black hole mass , but if there is a range in accretion rates and/or efficiencies , the relation will be weaker . in this paper
, we collect and compare all agn black hole mass estimates from the literature , and we make new black hole mass estimates from stellar velocity dispersions ( 2 ) .
we calculate bolometric luminosities for these same agn to investigate their mass luminosity relation , and look for trends of eddington ratio with luminosity ( 3 ) .
table [ t_sum ] summarizes the number of black hole mass estimates from the various methods .
we use @xmath3 km s@xmath2 and @xmath4 throughout this paper .
very few black hole masses in agn have been measured with spatially resolved dynamics near the central black hole which is the preferred method for estimating black hole mass in nearby ( inactive ) galaxies .
the two cases in which this has been done with maser kinematics ( ngc 1068 and ngc 4258 ) are listed in table 2 .
remaining black hole masses are determined with less direct methods . assuming that broad - line clouds are virialized , for which there has been increasing evidence ( krolik et al .
1991 , wandel et al . 1999 ,
krolik 2001 ) , the black hole mass can be estimated : @xmath5 the virial assumption may not be correct , however ; radiation pressure and/or magnetic fields may contribute significantly to the dynamics ( krolik 2001 ) , and outflows or winds could cause the observed line widths to exceed those induced by the black hole potential alone . in these cases the black hole mass calculated from eq .
[ virial ] would be overestimated . in reverberation mapping
, the blr size is estimated from the time lag between the ionizing continuum and the broad - line strength ( peterson 1993 ) . to date
, 36 agn black hole masses have been measured from combining reverberation - mapped blr sizes with broad - line velocities ( wandel et al . 1999 ,
ho 1999 , kaspi et al .
2000 , onken & peterson 2002 ) .
these are listed in table [ t_rev ] , along with the redshifts , bolometric luminosities , and published agn types .
contributing to the uncertainty in the black hole mass estimation are the blr orbits and velocities assumed .
the broad line velocity can be determined from the observed spectra , either as the mean of the fwhm derived from each line or as the fwhm from the root mean square ( rms ) spectrum ( peterson et al . 1988 ) .
kaspi et al .
( 2000 ) showed that the two velocity estimates are similar ; however , the difference between the two gives black hole mass uncertainties as large as a factor of ten ( figure [ rev ] ) .
assumptions about the orbital shape and inclination of the broad - line clouds introduce additional uncertainties .
an isotropic distribution with random inclinations is often assumed for the broad - line clouds , in which case velocity is derived from equation ( 2 ) with @xmath6 ( netzer 1990 ) : @xmath7 however , the random orbits assumption may not be valid for quasars .
mclure & dunlop ( 2001 ) reproduced the fhwm distribution of seyferts and quasars with two disc components , and determined that the average relationship between observed fwhm and actual orbital velocity corresponds to @xmath8 .
thus for the same agn , the black hole mass estimates in mclure & dunlop ( 2001 ) are factor of 3 larger than those of kaspi et al .
considering orbital shape alone , the full range of uncertainty in mass appears to be 2 orders of magnitude , from @xmath9 to @xmath10 200 ( krolik 2001 ) . in figure 1
we compare 34 reverberation - mapped black hole masses calculated for two different estimates of the broad - line velocities ( kaspi et al .
the derived black hole masses for a given object differ by less than an order of magnitude , making reverberation mapping one of the more robust techniques for estimating agn black hole masses .
it is however resource intensive , time consuming , and not applicable to most agn ( those without broad lines ) .
consequently , relatively few agn black hole masses have been well estimated .
since reverberation mapping is a laborious process , alternative ways of deriving the blr size are of interest .
several authors have noted that @xmath11 ( where known from reverberation mapping ) appears to correlate with uv / optical luminosity ( koratkar & gaskell 1991 , kaspi et al .
1996 , wandel et al . 1999 ,
kaspi et al .
the proportionality has been reported as @xmath12 ( wandel et al .
1999 ) , which corresponds to a constant ionization parameter , but in the most recent studies appears to be @xmath13 ( kaspi et al . 2000 , vestergaard et al . 2002
mclure & jarvis 2002 ) . using this relation and assuming random isotropic orbits ( @xmath14 in eq .
[ width ] ) , we obtain : @xmath15 there is large scatter in the @xmath11
@xmath16 correlation ( e.g. , figure 7 of kaspi et al .
2000 ) , and it has been established only over a limited range of luminosities , hence it yields correspondingly uncertain black hole masses .
we list these values in table [ t_opt ] , along with the redshift , bolometric luminosity , and agn type , and in figure [ opt_rev ] we compare them to all available reverberation mapping estimates .
the differences range up to an order of magnitude , with an rms difference of 0.50 in the log of the ratio . if optical luminosity is well correlated with bolometric luminosity , the fitted correlation of kaspi et al .
( 2000 ) leads to a precise relation between black hole mass and bolometric luminosity ( something we would like to investigate rather than assume )
. the eddington ratio ( i.e. , the ratio of bolometric luminosity to eddington luminosity ) would then depend on bolometric luminosity to the 0.3 power .
although there are some concerns , black hole mass estimates with this method remain important given the difficulty of more accurate estimates and the relatively small number of agn for which any black hole mass estimates have been made . thus , we collected all such black hole mass estimates available in the literature ( 26 from mclure & dunlop 2001 , 3 from laor 2001 , 80 from gu et al .
2001 , 30 from oshlack , webster & whiting 2002 ) , re - computed using eq .
[ lopt ] for consistency with our cosmology . in nearby galaxies
there is apparently a close connection between the central black hole and the bulge kinematics .
specifically , black hole mass ( determined from spatially resolved kinematics ) correlates well with stellar velocity dispersion , as @xmath17 ( gebhardt et al .
2000a ) or @xmath18 ( ferrarese & merritt 2000 ) . from the collective analysis by tremaine et al . ( 2002 ) : @xmath19 agn host galaxies appear to be very much like normal galaxies . this is particularly well established for radio - loud agn , whose host galaxies follow the usual kormendy relation ( taylor et al . 1996
; mclure et al . 1999 ; urry et al .
2000 ; bettoni et al .
present data on host galaxies are in accord with the `` grand unification '' hypothesis , suggested on other grounds , that agn are simply a transient phase of normal galaxies ( cavaliere & padovani 1989 ) .
therefore it is reasonable to expect that the same @xmath20@xmath0 correlation should be present in agn host galaxies , in which case we can use eq .
[ sig ] to infer black hole mass .
gebhardt et al .
( 2000b ) and ferrarese et al .
( 2001 ) estimated black hole masses in this way for a few seyfert galaxies ( 7 and 6 respectively ) , and found good agreement with reverberation mapping values .
an increasing number of agn have published measurements of stellar velocity dispersion .
black hole masses calculated from @xmath0 have been published for 21 seyferts ( wu & han 2001 ) and 12 bl lac objects ( falomo et al .
2002 ; barth , ho & sargent 2002 ) ; we rescaled these to our cosmology as needed . from the literature we collected velocity dispersions for an additional 108 agn ( 36 seyfert galaxies and 72 radio galaxies ) , and calculated their black hole masses according to eq .
all 141 black hole masses are presented in table [ t_sig ] .
for 14 seyfert galaxies both velocity dispersions and reverberation - mapped blr sizes are available . in figure [ sig_rev ]
we compare the two associated black hole mass estimates .
they agree relatively well , with scatter much less than an order of magnitude .
stellar velocity dispersions are not extensively known for agn host galaxies , nor are they easy to measure , particularly at higher redshift . however , by the same `` grand unification '' of host galaxies with normal galaxies , we can infer the velocity dispersions ( albeit with additional scatter ) from the morphological parameters of the bulge : @xmath21 , the effective radius , and @xmath22 , the surface brightness at that radius .
these have been very well measured for more than 100 agn using the excellent spatial resolution of the hubble space telescope ( hst ) , which yields more robust results than observations in typical ground - based seeing .
thus , at least for radio - loud agn , black hole mass can be derived from @xmath23 and @xmath24 ( odowd et al . 2002 ) .
if sufficiently accurate , this would be an extremely valuable method since the required imaging data are much easier to obtain than @xmath0 , and such a method could be applied widely and at higher redshift than the direct method . using this method
, we estimate 59 new black hole masses for 45 bl lac objects , 10 radio galaxies , and 4 radio - quiet agn , all of which have host galaxies detected with hst .
surface brightnesses and effective radii from urry et al .
( 2000 ) and dunlop et al .
( 2002 ) are used to derive stellar velocity dispersion via fundamental plane relation of jorgensen et al .
( 1996 ) : @xmath25 here , @xmath26 for cosmological correction to @xmath27=75 km s@xmath2 .
black hole masses are then estimated using eq .
morphological parameters and derived black hole masses are given in table [ t_fp ] .
bolometric luminosity is not straightforward to derive for most of these objects because of beaming and obscuration . to test the accuracy of this fundamental plane method for estimating black hole mass
, we considered 72 radio galaxies for which all three parameters of the fundamental plane are measured ( bettoni et al .
values in arcsec rather than kpc ( barth et al . 2002 ) . ] figure [ fp_sig ] shows the comparison of black hole masses derived indirectly from @xmath22 and @xmath21 with those derived directly from @xmath0 .
( this is in effect an unusual projection of the fundamental plane . )
points are coded to highlight the homogeneous data of bettoni et al .
( filled circles ) , which are more tightly correlated than the additional heterogeneous data ( open squares and crosses ) collected by them .
the six most extreme outliers are marked with crosses .
the mean black hole masses determined by the two methods agree to within 10% , while the rms scatter is a factor of 4 or so ( slightly higher for the heterogeneous data than for the homogeneous data ) .
although the fundamental plane method introduces additional scatter compared to direct measurement of stellar velocity dispersion , estimating black hole masses in this way is so far one of the few ways to infer agn black hole mass for high redshift agn ( perhaps the only method for agn that lack broad emission lines ) .
of course , the underlying assumption of `` grand unification '' of agn and galaxies remains untested , particularly at high redshift .
bolometric luminosity of agn is sometimes approximated from optical luminosity , since integration of the spectral energy distribution ( sed ) , which spans many decades in wavelength , is usually hampered by lack of wavelength coverage and by variability .
here we are able in many cases to determine bolometric luminosity by integrating all available flux points in the sed .
this is particularly important given the role of optical luminosity in deriving some black hole masses , otherwise correlations between @xmath20 and @xmath28 can be induced . for 234 of the 377 agn for which black hole mass has been estimated in the tables
, we were able to determine bolometric luminosity .
the other 143 objects are radio galaxies and bl lac objects , for which obscuration and beaming are significant . for 82 of the 234 ,
there are numerous published fluxes from ultraviolet to far - infrared wavelengths , which we collected using the ned database .
multiple observations for the same band were simply averaged , and the galactic extinction law ( cardelli , clayton & mathis 1989 ) was used to correct for dust ( with @xmath29 also taken from ned ) .
we then integrated these seds directly to get the bolometric luminosity .
for the remaining agn , mostly quasars at relatively high redshift , sufficient flux points were unavailable . in 152 cases , including most of the luminous quasars , we obtained the bolometric luminosity by fitting the average sed for that agn type to the available flux points .
average seds are from various sources : radio - loud and radio - quiet quasar seds are from elvis et al .
( 1994 ) ; seyfert 1 seds are from mas - hesse et al .
( 1994 ) ; and seyfert 2 seds are from schmitt et al .
optical flux was corrected for galactic extinction using individual reddening values from ned .
we note that the bolometric luminosities are roughly 10 times the optical luminosity ( precisely , in the case of sed fitting for quasars , and within a factor of 5 - 6 in the case of direct integration of the seds ) .
bolometric luminosities for a total of 234 agn are given in tables [ t_kin ] , [ t_rev ] , [ t_opt ] and [ t_sig ] .
the associated black hole masses were estimated as follows : 2 from maser kinematics , 36 from broad - line widths plus reverberation mapping , 139 from broad - line widths plus the @xmath30 - @xmath11 relation , and 57 from the @xmath20 - @xmath0 relation . in order to check our bolometric luminosity measurements ,
we compare them with previous estimates by padovani & rafanelli ( 1988 ) , who integrated available optical to far - infrared fluxes for 58 seyfert galaxies and quasars .
twenty - six agn in the padovani & rafanelli sample have bolometric luminosities estimated here ; we rescaled the former values to @xmath3 km s@xmath2 simply by multiplying by 4/9 ( @xmath31 km s@xmath2 , @xmath32 in their calculation ) .
the comparison is shown in figure [ lbol ] .
the two estimations agree well although the padovani & rafanelli values may be systematically lower due to the more limited spectral range in their calculation .
we now compare bolometric luminosity with black hole mass .
figure 6a includes only the 36 reverberation - mapped quasars and seyfert galaxies , and figure 6b includes the 57 seyfert galaxies for which black hole mass was estimated from observed stellar velocity dispersion .
there is large scatter and little correlation between bolometric luminosity and black hole mass .
for a given black hole mass , the bolometric luminosity ranges over more than two orders of magnitude .
figure 6c shows the mass luminosity plot for agn with black hole masses that were derived from optical luminosity and broad - line velocity ( mclure & dunlop 2001 , laor 2001 , gu et al .
2001 , oshlack et al .
2002 ) . even here there not much more of a correlation , although one will appear if optical and bolometric luminosities are well correlated .
that is , since black hole masses for these agn were derived from @xmath30 , the slope indicated by the solid line is implied if @xmath28 is proportional to @xmath30 .
figure [ ml_all ] shows the mass luminosity relation for all 234 agn . even more clearly than in figure 6
, there is hardly any trend of luminosity with black hole mass . for a given agn black hole mass , the bolometric luminosity ranges over at least two , and as much as four , orders of magnitude .
the eddington ratio must span a similarly large range .
the eddington ratio does define an approximate ( but not hard ) upper limit to the distribution of luminosities ; that is , points are missing from the upper left region above the dotted line , in fact previously noted by mcleod , rieke & storrie - lombardi ( 1999 ) .
the lack of points in the lower right , however , is a selection effect : this part of the diagram gets filled in simply by including lower luminosity agn , continuously down to galaxies . among the low - luminosity objects with large black holes
are the radio galaxies and bl lac objects for which we do not have good estimates of bolometric luminosity ( cf .
odowd et al .
2002 ) ; the box indicates the approximate region they occupy , calculated from the observed luminosities of bl lacs using the family of seds from fossati et al .
( 1998 ) and correcting for beaming factors in the range 3 - 10 ( dondi & ghisellini 1995 ) .
agn lore has it that the eddington ratio is 0.1 - 1 for high - luminosity sources and an order of magnitude or more smaller for low - luminosity sources .
our sample of agn spans 5 decades in bolometric luminosity so we should be very sensitive to any such trends .
in figure [ l_er ] we plot eddington ratio versus bolometric luminosity ( top panel ) . at most luminosities , the eddington ratio spans two decades or so , except at the very highest luminosity .
there appears to be a deficit of high luminosity objects with low eddington ratios ( i.e. , with black holes in the range @xmath33 ) .
however , these include some of the radio sources for which we do not have good bolometric luminosities ( see table [ t_fp ] ) .
furthermore , if more massive black holes are rare ( i.e. , there is a steep mass function ) , they would on average be found at high redshift , yet low - luminosity radio sources at high redshift are excluded from flux - limited samples .
there is also a deficiency of points in the upper left corner of the plot ; these would be agn with luminosities of @xmath34 ergs s@xmath2 and black hole masses less than @xmath35 @xmath36 .
( note that low - luminosity agn may be more difficult to detect because of dilution by host galaxy light . )
thus there is no immediate evidence of any real trend in eddington ratio with luminosity .
we also plot eddington ratio versus black hole mass ( bottom panel ) .
again , there are no clear trends that can not be explained by sample selection effects .
for example , objects with luminosities below @xmath37 ergs s@xmath2 are not called seyfert galaxies or quasars and thus do not appear in this diagram .
( one could add them , and they would fill in the lower left corner of the plot . ) agn with luminosities greater than @xmath38 ergs s@xmath2 are rare and thus probably too distant , on average , to have black hole mass estimates .
with such a heterogeneous sample , we hesitate to make any strong statements , but certainly we see only very weak trends or correlations , and those are quite plausibly induced by sample selection effects .
we can see this by plotting the eddington ratio versus redshift ( figure [ z_er ] ) .
again there is little if any trend .
high eddington ratio objects ( @xmath39 ) are perhaps missing at low redshift , but this can be explained as a volume effect ( i.e. , given the steep luminosity function of agn , one has to survey a large volume to find a relatively rare high - luminosity agn ) .
more obviously , low eddington ratio objects @xmath39 ) are absent at high redshift , and this is partly a flux limit issue , since low - luminosity agn fall out of samples at high redshift .
thus any trends that do appear to the eye in this plot are explained by obvious selection effects .
finally , we look at radio luminosity versus black hole mass ( fig .
[ l5ghz ] , top panel ) since previous reports have suggested there is a correlation between the two ( mclure et al .
1999 ; lacy et al .
2001 ) , although more recent investigations have not found such a correlation ( ho 2002 ; oshlack et al 2002 ) .
again , there is little evidence of a correlation , particularly given the missing low - luminosity sources like bl lacs that do appear to have high black hole masses ( and thus should help fill in the lower right corner of the plot ) .
very low - luminosity agn ( @xmath40 w / hz ) with massive black holes may be missing , though this is hard to quantify given the missing bl lacs and radio galaxies . to further investigate this point , we consider radio - loudness .
there have been suggestions that black hole mass is a factor in radio loudness , such that @xmath41 ( @xmath42 ) requires @xmath43 ( laor 2000 ) . in figure [ l5ghz ]
we plot radio loudness versus black hole mass for the same objects ( bottom panel ) .
the radio - loud agn have a very broad distribution of masses , so there clearly is no threshold effect . in the radio - quiet regime ( @xmath44 ) , the distribution of masses is narrower , with no black holes masses greater than @xmath43 .
we note that almost all of the high - mass black holes are estimated from the optical luminosity method ; that these occur in radio - loud agn , therefore , could be explained if an appreciable fraction of the optical luminosity is beamed .
if instead the absence of high - mass radio - quiet agn is real , this would be a very significant distinction between the radio - quiet and radio - loud agn .
however , given the heterogeneous sample discussed here , the absence of evidence of these objects is not evidence of their absence , and more work will be required on this point .
we estimated and/or collected from the literature black hole masses for 377 agn , obtained with various methods .
these span a range of nearly 4 orders of magnitude , from @xmath45 to @xmath46 .
direct comparisons suggest that reverberation mapping and stellar velocity dispersion give reliable black hole mass estimates within factors of a few while using optical luminosity to infer broad - line size or using the fundamental plane to infer velocity dispersion leads to somewhat larger uncertainties . in the case of virial estimates ( reverberation mapping , optical luminosity , or other ) , additional uncertainties enter through the unknown orbits and the possible non - virial motions of the line - emitting gas .
we estimated bolometric luminosities for most of the agn , apart from those affected strongly by beaming or by obscuration of the nuclear emission . comparing bolometric luminosity to black hole mass for 234 agn , we find little or no correlation .
gaps in coverage of the @xmath28@xmath20 plane are due at least in part to high - mass , low - luminosity objects like the bl lac objects and radio galaxies for which we have no good bolometric luminosity estimates . for a given black hole mass , bolometric luminosities range over as many as four orders of magnitude .
the eddington ratios span nearly as large a range , 23 orders of magnitude at most luminosities .
these are much larger than any uncertainties in the estimates of either black hole mass or luminosity .
there are no strong trends of eddington ratio with luminosity , contrary to long - held preconceptions .
the absence of low eddington ratios at high redshifts ( high luminosities ) can be explained at least in part by selection effects in flux - limited surveys wherein highly sub - eddington agn disappear progressively at higher redshifts .
we also do not confirm previously reported trends of radio luminosity with black hole mass , and while our results indicate a modest dependence of radio loudness on black hole mass , selection effects may exaggerate or even produce this trend . on the whole , black hole mass seems to have remarkably little to do with the appearance of active nuclei , either their luminosities or radio power .
of course , the present sample includes a randomly selected mix of agn , with black hole masses estimated in different ways , by different people , from different data sets .
there may be real trends dependent on other variables not taken into account here ( e.g. , agn type ) .
it is obviously of interest to apply the more robust black hole mass estimation methods reverberation mapping and stellar velocity dispersion to a large sample of agn , at as high a redshift as possible , although these methods will probably not work for the typical agn at @xmath473 . in practice ,
such a study would start with measurements of stellar velocity dispersions at @xmath48 , which require 4- to 10-m class telescopes .
we thank matthew odowd for suggesting the fundamental plane method of estimating black hole masses and for helpful discussions .
we thank aaron barth for his careful reading of the manuscript , and meredith hughes for help with the research .
support for proposals 5938 , 5939 , 6363 , and 7893 was provided by nasa through grants from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc .
, under nasa contract nas 5 - 26555 .
barth , a. , ho , l. , & sargent , w. l. w. 2002 , apj , submitted bettoni , d. , falomo , r. , fasano , g. , govoni , f. , salvo , m. , & scarpa , r. 2001 , a&a , 380 , 471 blandford , r. d. , & mckee , c. f. 1982 , apj , 255 , 419 cardelli , j. a. , clayton , g. c. , & mathis , j. 1989 , apj , 345 , 245 cavaliere , a. , & padovani , p. 1989 , apj , 340 , l5 dibai , e. a. 1981 , soviet astr . , 24 , 389 di nella , h. , garcia , a. m. , garnier , r. , & paturel , g. 1995 , a&as , 113 , 151 dondi , l. , & ghisellini , g. 1995 , mnras , 273 , 583 dunlop , j. s. , mclure , r. j. , kukula , m. j. , baum , s. a. , odea , c. p. , & hughes , d. h. 2002 , mnras , submitted ( astro - ph/0108397 ) elvis , m. , et al .
1994 , apjs , 95 , 1 falomo , r. , kotilainen , j. k. , & treves , a. 2002 , apj , 569 , l35 ferrarese , l. , & merritt , d. 2000 , apj , 539 , l9 ferrarese , l. , et al .
2001 , apj , 555 , l79 fossati , g. , maraschi , l. , celotti , a. , comastri , a. & ghisellini , g. 1998 , mnras , 299 , 433 gebhardt , k. , et al .
2000a , apj , 539 , l13 gebhardt , k. , et al . 2000b , apj , 543 , l5 greenhill , l. j. , et al .
1996 , apj , 481 , l23 gu , m. , cao , x. , & jiang , d. r. 2001 , mnras , 327 , 1111 harms , r. j. et al .
1994 , apj , 435 , l35 ho , l. c. 1999 , in observational evidence for black holes in the universe , ed . s. k. chakrabarti ( dordrecht : kluwer ) , 157 ho , l. c. 2002 , apj , 564 , 120 jorgensen , i. , franx , m. , & kjargaard , p. 1996 , mnras , 280 , 167 kaspi , s. , smith , p. s. , maoz , d. , netzer , h. , & jannuzi , b. t. 1996 , apj , 471 , l75 kaspi , s. , et al .
2000 , apj , 533 , 631 koratkar , a. p. , & c. m. gaskell 1991 , apj , 370 , l61 krolik , j. h. , et al .
1991 , apj , 371 , 541 krolik , j. h. 2001 , apj , 551 , 72 kormendy , j. , & gebhardt , k. 2001 , in the 20th texas symposium on relativistic astrophysics , ed . h. martel & j. c. whleer , in press ( astro - ph/0105230 ) lacy , m. , laurent - meuleisen , s. a. , ridgway , s. e. , becker , r. h. , & white , r. l. 2001 , apj , 551 , l17 laor , a. 2000 , apj , 543 , l111 laor , a. 2001 , apj , 553 , 677 lynden - bell , d. 1969 , nature , 223 , 690 mcleod , k. k. , rieke , g. h. , & storrie - lombardi , l. j. 1999 apj , 511 , l67 mclure , r. j. , kukula , m. j. , dunlop , j. s. , baum , s. a. , & odea , c. p. 1999 , mnras , 308 , 377 mclure , r. j. , & dunlop , j. s. 2001 , mnras , 327 , 199 mclure , r. j. , & jarvis , m. j. 2002 , mnras , submitted ( astroph 0204473 ) mas - hesse , j. m. , et al .
1995 a&a , 298 , 22 miyoshi , m. , et al .
1995 , nature , 373 , 127 nelson , c. & whittle , m. 1995 , apjs , 99 , 67 netzer , h. 1990 , in active galactic nuclei : 1990 sass - fee lectures , ed .
courvoisier & m. mayor ( berlin : springer ) , 137 schmitt , h. , et al .
1997 , aj , 114 , 592 odowd , m. , c. m. , urry , & r. scarpa 2001 , apj , submitted oliva , e. , origlia , l. , kotilainen , j.k .
, moorwood , a.f.m .
1995 , a&a 301 , 55 oliva , e. , origlia , l. , maiolino , r. , moorwood , a. f. m. 1999 a&a 350 , 9 onken , c. a. & peterson , b. a. 2002 , apj , acce2002 , apj , in press oshlack , a. , webster , r. , & whiting , m. 2002 , pasp , in press padovani , p. , & rafanelli ,
, a&a 205 , 53 peterson , b. m. 1988 , pasp , 100 , 18 peterson , b. m. 1993 , pasp , 105 , 247 taylor , g. l. , dunlop , j. s. , hughes , d. h. , & robson , e. i. 1996 , mnras , 283 , 930 tremaine , s. , et al .
2002 , apj , in press ( astro - ph/0203468 ) urry , c. m. , et al .
2000 , apj , 532 , 816 vestergaard , m. 2002 , apj , 571 , 733 wandel , a. , & yahil , a. 1985 , apj , 295 , l1 wandel , a. , peterson , b. m. , & makkan , m. a. 1999 , apj , 526 , 579 wu , x .- b . , &
han , j. l. 2001 , a&a , 380 , 31 lcr spatially resolved kinematics & 2 & greenhill et al .
1996 , miyoshi et al .
1995 + reverberation mapping & 36 & ho 1999 , kaspi et al .
2000 , onken & peterson 2002 + @xmath49@xmath11 relation & 139 & mclure & dunlop 2001 , laor 2001 , gu et al .
2001 , oshlack et al .
2002 + @xmath20@xmath0 relation & 33 & wu & han 2001 , barth et al .
2002 , falomo et al .
2002 + & 108 & this work + fundamental plane & 59 & this work + [ t_sum ] lcrrrrrr 3c 120 & 0.033 & 45.34&i & 7.42 & 1 & sy1 + 3c 390.3 & 0.056 & 44.88&i & 8.55 & 1 & sy1 + akn 120 & 0.032 & 44.91&i & 8.27 & 1 & sy1 + f 9 & 0.047 & 45.23&f & 7.91 & 1 & sy1 + ic 4329a & 0.016 & 44.78&i & 6.77 & 1 & sy1 + mrk 79 & 0.022 & 44.57&i & 7.86 & 1 & sy1 + mrk 110 & 0.035 & 44.71&f & 6.82 & 1 & sy1 + mrk 335 & 0.026 & 44.69&i & 6.69 & 1 & sy1 + mrk 509 & 0.034 & 45.03&i & 7.86 & 1 & sy1 + mrk 590 & 0.026 & 44.63&i & 7.20 & 1 & sy1 + mrk 817 & 0.032 & 44.99&i & 7.60 & 1 & sy1 + ngc 3227 & 0.004 & 43.86&i & 7.64 & 1 & sy1 + ngc 3516 & 0.009 & 44.29&i & 7.36 & 3 & sy1 + ngc 3783 & 0.010 & 44.41&i & 6.94 & 2 & sy1 + ngc 4051 & 0.002 & 43.56&i & 6.13 & 1 & sy1 + ngc 4151 & 0.003 & 43.73&i & 7.13 & 1 & sy1 + ngc 4593 & 0.009 & 44.09&i & 6.91 & 3 & sy1 + ngc 5548 & 0.017 & 44.83&i & 8.03 & 1 & sy1 + ngc 7469 & 0.016 & 45.28&i & 6.84 & 1 & sy1 + pg 0026 + 129 & 0.142 & 45.39&i & 7.58 & 1 & rqq + pg 0052 + 251 & 0.155 & 45.93&f & 8.41 & 1 & rqq + pg 0804 + 761 & 0.100 & 45.93&f & 8.24 & 1 & rqq + pg 0844 + 349 & 0.064 & 45.36&f & 7.38 & 1 & rqq + pg 0953 + 414 & 0.239 & 46.16&f & 8.24 & 1 & rqq + pg 1211 + 143 & 0.085 & 45.81&f & 7.49 & 1 & rqq + pg 1229 + 204 & 0.064 & 45.01&i & 8.56 & 1 & rqq + pg 1307 + 085 & 0.155 & 45.83&f & 7.90 & 1 & rqq + pg 1351 + 640 & 0.087 & 45.50&i & 8.48 & 1 & rqq + pg 1411 + 442 & 0.089 & 45.58&f & 7.57 & 1 & rqq + pg 1426 + 015 & 0.086 & 45.19&i & 7.92 & 1 & rqq + pg 1613 + 658 & 0.129 & 45.66&i & 8.62 & 1 & rqq + pg 1617 + 175 & 0.114 & 45.52&f & 7.88 & 1 & rqq + pg 1700 + 518 & 0.292 & 46.56&f & 8.31 & 1 & rqq + pg 2130 + 099 & 0.061 & 45.47&i & 7.74 & 1 & rqq + pg 1226 + 023 & 0.158 & 47.35&i & 7.22 & 1 & rlq + pg 1704 + 608 & 0.371 & 46.33&i & 8.23 & 1 & rlq + [ t_rev ] lcrrrrr mrk 841 & 0.036 & 45.84&i & 8.10 & 1 & sy1 + ngc 4253 & 0.013 & 44.40&i & 6.54 & 1 & sy1 + ngc 6814 & 0.005 & 43.92&i & 7.28 & 1 & sy1 + 0054 + 144 & 0.171 & 45.47&f & 8.90 & 2 & rqq + 0157 + 001 & 0.164 & 45.62&f & 7.70 & 2 & rqq + 0204 + 292 & 0.109 & 45.05&f & 6.67 & 2 & rqq + 0205 + 024 & 0.155 & 45.45&f & 7.86 & 2 & rqq + 0244 + 194 & 0.176 & 45.51&f & 8.03 & 2 & rqq + 0923 + 201 & 0.190 & 46.22&f & 8.94 & 2 & rqq + 1012 + 008 & 0.185 & 45.51&f & 7.79 & 2 & rqq + 1029 - 140 & 0.086 & 46.03&f & 9.08 & 2 & rqq + 1116 + 215 & 0.177 & 46.02&f & 8.21 & 2 & rqq + 1202 + 281 & 0.165 & 45.39&f & 8.29 & 2 & rqq + 1309 + 355 & 0.184 & 45.63&f & 8.00 & 2 & rqq + 1402 + 261 & 0.164 & 45.13&f & 7.29 & 2 & rqq + 1444 + 407 & 0.267 & 45.93&f & 8.06 & 2 & rqq + 1635 + 119 & 0.146 & 45.13&f & 8.10 & 2 & rqq + 0022 - 297 & 0.406 & 44.98&f & 7.91 & 3 & rlq + 0024 + 348 & 0.333 & 45.31&f & 6.37 & 3 & rlq + 0056 - 001 & 0.717 & 46.54&f & 8.71 & 3 & rlq + 0110 + 495 & 0.395 & 45.78&f & 8.34 & 3 & rlq + 0114 + 074 & 0.343 & 44.02&f & 6.80 & 4 & rlq + 0119 + 041 & 0.637 & 45.57&f & 8.38 & 3 & rlq + 0133 + 207 & 0.425 & 45.83&f & 9.52 & 3 & rlq + 0133 + 476 & 0.859 & 46.69&f & 8.73 & 3 & rlq + 0134 + 329 & 0.367 & 46.44&f & 8.74 & 3 & rlq + 0135 - 247 & 0.831 & 46.64&f & 9.13 & 3 & rlq + 0137 + 012 & 0.258 & 45.22&f & 8.57 & 2 & rlq + 0153 - 410 & 0.226 & 44.74&f & 7.56 & 4 & rlq + 0159 - 117 & 0.669 & 46.84&f & 9.27 & 3 & rlq + 0210 + 860 & 0.186 & 44.92&f & 6.54 & 3 & rlq + 0221 + 067 & 0.510 & 44.94&f & 7.29 & 4 & rlq + 0237 - 233 & 2.224 & 47.72&f & 8.52 & 3 & rlq + 0327 - 241 & 0.888 & 46.01&f & 8.60 & 4 & rlq + 0336 - 019 & 0.852 & 46.32&f & 8.98 & 3 & rlq + 0403 - 132 & 0.571 & 46.47&f & 9.07 & 3 & rlq + 0405 - 123 & 0.574 & 47.40&f & 9.47 & 3 & rlq + 0420 - 014 & 0.915 & 47.00&f & 9.03 & 3 & rlq + 0437 + 785 & 0.454 & 46.15&f & 8.79 & 3 & rlq + 0444 + 634 & 0.781 & 46.12&f & 8.53 & 3 & rlq + 0454 - 810 & 0.444 & 45.32&f & 8.13 & 3 & rlq + 0454 + 066 & 0.405 & 45.12&f & 7.42 & 4 & rlq + 0502 + 049 & 0.954 & 46.36&f & 8.88 & 4 & rlq + 0514 - 459 & 0.194 & 45.36&f & 7.55 & 3 & rlq + 0518 + 165 & 0.759 & 46.34&f & 8.53 & 3 & rlq + 0538 + 498 & 0.545 & 46.43&f & 9.58 & 3 & rlq + 0602 - 319 & 0.452 & 45.69&f & 9.02 & 3 & rlq + 0607 - 157 & 0.324 & 46.30&f & 8.68 & 3 & rlq + 0637 - 752 & 0.654 & 47.16&f & 9.41 & 3 & rlq + 0646 + 600 & 0.455 & 45.58&f & 8.74 & 3 & rlq + 0723 + 679 & 0.846 & 46.41&f & 8.67 & 3 & rlq + 0736 + 017 & 0.191 & 45.97&f & 8.00 & 2 & rlq + 0738 + 313 & 0.631 & 46.94&f & 9.40 & 3 & rlq + 0809 + 483 & 0.871 & 46.54&f & 7.96 & 3 & rlq + 0838 + 133 & 0.684 & 46.23&f & 8.52 & 3 & rlq + 0906 + 430 & 0.668 & 45.99&f & 7.90 & 3 & rlq + 0912 + 029 & 0.427 & 45.26&f & 7.72 & 4 & rlq + 0921 - 213 & 0.052 & 44.63&f & 8.14 & 4 & rlq + 0923 + 392 & 0.698 & 46.26&f & 9.28 & 3 & rlq + 0925 - 203 & 0.348 & 46.35&f & 8.46 & 4 & rlq + 0953 + 254 & 0.712 & 46.59&f & 9.00 & 3 & rlq + 0954 + 556 & 0.901 & 46.54&f & 8.07 & 3 & rlq + 1004 + 130 & 0.240 & 46.21&f & 9.10 & 2 & rlq + 1007 + 417 & 0.612 & 46.71&f & 8.79 & 3 & rlq + 1016 - 311 & 0.794 & 46.63&f & 8.89 & 4 & rlq + 1020 - 103 & 0.197 & 44.87&f & 8.36 & 2 & rlq + 1034 - 293 & 0.312 & 46.20&f & 8.75 & 3 & rlq + 1036 - 154 & 0.525 & 44.55&f & 7.80 & 4 & rlq + 1045 - 188 & 0.595 & 45.80&f & 6.83 & 3 & rlq + 1100 + 772 & 0.311 & 46.49&f & 9.31 & 3 & rlq + 1101 - 325 & 0.355 & 46.33&f & 8.61 & 4 & rlq + 1106 + 023 & 0.157 & 44.97&f & 7.50 & 4 & rlq + 1107 - 187 & 0.497 & 44.25&f & 6.90 & 4 & rlq + 1111 + 408 & 0.734 & 46.26&f & 9.82 & 3 & rlq + 1128 - 047 & 0.266 & 44.08&f & 6.72 & 4 & rlq + 1136 - 135 & 0.554 & 46.78&f & 8.78 & 3 & rlq + 1137 + 660 & 0.656 & 46.85&f & 9.36 & 3 & rlq + 1150 + 497 & 0.334 & 45.98&f & 8.73 & 3 & rlq + 1151 - 348 & 0.258 & 45.56&f & 9.02 & 3 & rlq + 1200 - 051 & 0.381 & 46.41&f & 8.41 & 4 & rlq + 1202 - 262 & 0.789 & 45.81&f & 9.00 & 3 & rlq + 1217 + 023 & 0.240 & 45.83&f & 8.41 & 2 & rlq + 1237 - 101 & 0.751 & 46.63&f & 9.28 & 4 & rlq + 1244 - 255 & 0.633 & 46.48&f & 9.04 & 3 & rlq + 1250 + 568 & 0.321 & 45.61&f & 8.42 & 3 & rlq + 1253 - 055 & 0.536 & 46.10&f & 8.43 & 3 & rlq + 1254 - 333 & 0.190 & 45.52&f & 8.83 & 4 & rlq + 1302 - 102 & 0.286 & 45.86&f & 8.30 & 2 & rlq + 1352 - 104 & 0.332 & 45.81&f & 8.15 & 4 & rlq + 1354 + 195 & 0.720 & 47.11&f & 9.44 & 3 & rlq + 1355 - 416 & 0.313 & 46.48&f & 9.73 & 3 & rlq + 1359 - 281 & 0.803 & 46.19&f & 8.07 & 4 & rlq + 1450 - 338 & 0.368 & 43.94&f & 6.46 & 4 & rlq + 1451 - 375 & 0.314 & 46.16&f & 8.82 & 3 & rlq + 1458 + 718 & 0.905 & 46.93&f & 8.98 & 3 & rlq + 1509 + 022 & 0.219 & 44.54&f & 7.99 & 4 & rlq + 1510 - 089 & 0.361 & 46.38&f & 8.65 & 3 & rlq + 1545 + 210 & 0.266 & 45.86&f & 8.93 & 2 & rlq + 1546 + 027 & 0.412 & 46.00&f & 8.72 & 3 & rlq + 1555 - 140 & 0.097 & 44.94&f & 7.25 & 4 & rlq + 1611 + 343 & 1.401 & 46.99&f & 9.57 & 3 & rlq + 1634 + 628 & 0.988 & 45.47&f & 7.28 & 3 & rlq + 1637 + 574 & 0.750 & 46.68&f & 9.18 & 3 & rlq + 1641 + 399 & 0.594 & 46.89&f & 9.42 & 3 & rlq + 1642 + 690 & 0.751 & 45.78&f & 7.76 & 3 & rlq + 1656 + 053 & 0.879 & 47.21&f & 9.62 & 3 & rlq + 1706 + 006 & 0.449 & 44.01&f & 6.63 & 4 & rlq + 1721 + 343 & 0.206 & 45.63&f & 8.04 & 3 & rlq + 1725 + 044 & 0.293 & 46.07&f & 8.07 & 3 & rlq + 1726 + 455 & 0.714 & 45.85&f & 8.22 & 3 & rlq + 1828 + 487 & 0.691 & 46.78&f & 9.85 & 3 & rlq + 1849 + 670 & 0.657 & 46.23&f & 9.14 & 3 & rlq + 1856 + 737 & 0.460 & 46.21&f & 8.89 & 3 & rlq + 1928 + 738 & 0.302 & 46.68&f & 8.91 & 3 & rlq + 1945 + 725 & 0.303 & 45.54&f & 6.48 & 3 & rlq + 1954 - 388 & 0.626 & 46.31&f & 8.63 & 4 & rlq + 2004 - 447 & 0.240 & 45.32&f & 7.48 & 4 & rlq + 2043 + 749 & 0.104 & 46.23&f & 9.62 & 3 & rlq + 2059 + 034 & 1.012 & 46.84&f & 9.13 & 4 & rlq + 2111 + 801 & 0.524 & 45.83&f & 8.73 & 3 & rlq + 2120 + 099 & 0.932 & 45.75&f & 8.19 & 4 & rlq + 2128 - 123 & 0.501 & 46.76&f & 9.61 & 3 & rlq + 2135 - 147 & 0.200 & 46.17&f & 8.94 & 2 & rlq + 2141 + 175 & 0.213 & 46.23&f & 8.74 & 2 & rlq + 2143 - 156 & 0.698 & 46.65&f & 7.68 & 4 & rlq + 2155 - 152 & 0.672 & 45.67&f & 7.59 & 3 & rlq + 2201 + 315 & 0.298 & 46.62&f & 8.87 & 3 & rlq + 2216 - 038 & 0.901 & 47.17&f & 9.24 & 3 & rlq + 2218 + 395 & 0.655 & 46.11&f & 7.14 & 3 & rlq + 2247 + 140 & 0.237 & 45.47&f & 7.59 & 2 & rlq + 2251 + 158 & 0.859 & 47.27&f & 9.17 & 3 & rlq + 2255 - 282 & 0.926 & 46.96&f & 9.16 & 3 & rlq + 2311 + 469 & 0.741 & 46.55&f & 9.30 & 3 & rlq + 2329 - 415 & 0.671 & 46.22&f & 8.93 & 4 & rlq + 2342 + 821 & 0.735 & 45.56&f & 7.31 & 3 & rlq + 2344 + 092 & 0.673 & 47.07&f & 9.31 & 3 & rlq + 2345 - 167 & 0.576 & 45.92&f & 8.72 & 3 & rlq + 2349 - 014 & 0.173 & 45.94&f & 8.78 & 2 & rlq + 2355 - 082 & 0.210 & 45.01&f & 8.39 & 2 & rlq + [ t_opt ] lcrrrrrr ngc 1566 & 0.005 & 100.&n & 6.92 & 44.45&i&sy1 + ngc 2841 & 0.002 & 209.&n & 8.21 & 43.67&i&sy1 + ngc 3982 & 0.004 & 62.&n & 6.09 & 43.54&i&sy1 + ngc 3998 & 0.003 & 319.&n & 8.95 & 43.54&i&sy1 + mrk 10 & 0.029 & 137.&n & 7.47 & 44.61&i&sy1 + ugc 3223 & 0.016 & 106.&n & 7.02 & 44.27&i&sy1 + ngc 513 & 0.002 & 152.&n & 7.65 & 42.52&i&sy2 + ngc 788 & 0.014 & 140.&n & 7.51 & 44.33&i&sy2 + ngc 1052 & 0.005 & 207.&n & 8.19 & 43.84&i&sy2 + ngc 1275 & 0.018 & 248.&n & 8.51 & 45.04&i&sy2 + ngc 1320 & 0.009 & 116.&n & 7.18 & 44.02&i&sy2 + ngc 1358 & 0.013 & 173.&n & 7.88 & 44.37&i&sy2 + ngc 1386 & 0.003 & 120.&n & 7.24 & 43.38&i&sy2 + ngc 1667 & 0.015 & 173.&n & 7.88 & 44.69&i&sy2 + ngc 2110 & 0.008 & 220.&n & 8.30 & 44.10&i&sy2 + ngc 2273 & 0.006 & 124.&n & 7.30 & 44.05&i&sy2 + ngc 2992 & 0.008 & 158.&n & 7.72 & 43.92&i&sy2 + ngc 3185 & 0.004 & 61.&n & 6.06 & 43.08&i&sy2 + ngc 3362 & 0.028 & 92.&n & 6.77 & 44.27&i&sy2 + ngc 3786 & 0.009 & 142.&n & 7.53 & 43.47&i&sy2 + ngc 4117 & 0.003 & 95.&n & 6.83 & 43.64&f&sy2 + ngc 4339 & 0.004 & 132.&n & 7.40 & 43.38&i&sy2 + ngc 5194 & 0.002 & 102.&n & 6.95 & 43.79&i&sy2 + ngc 5252 & 0.023 & 190.&n & 8.04 & 45.39&f&sy2 + ngc 5273 & 0.004 & 79.&n & 6.51 & 43.03&i&sy2 + ngc 5347 & 0.008 & 93.&n & 6.79 & 43.81&i&sy2 + ngc 5427 & 0.009 & 74.&n & 6.39 & 44.12&i&sy2 + ngc 5929 & 0.008 & 121.&n & 7.25 & 43.04&i&sy2 + ngc 5953 & 0.007 & 101.&n & 6.94 & 44.05&i&sy2 + ngc 6104 & 0.028 & 148.&n & 7.60 & 43.60&i&sy2 + ngc 7213 & 0.006 & 185.&n & 7.99 & 44.30&i&sy2 + ngc 7319 & 0.023 & 130.&n & 7.38 & 44.19&i&sy2 + ngc 7603 & 0.030 & 194.&n & 8.08 & 44.66&i&sy2 + ngc 7672 & 0.013 & 98.&n & 6.88 & 43.86&i&sy2 + ngc 7682 & 0.017 & 123.&n & 7.28 & 43.93&i&sy2 + ngc 7743 & 0.006 & 83.&n & 6.59 & 43.60&i&sy2 + mrk 1 & 0.016 & 115.&n & 7.16 & 44.20&i&sy2 + mrk 3 & 0.014 & 269.&n & 8.65 & 44.54&i&sy2 + mrk 78 & 0.037 & 172.&n & 7.87 & 44.59&i&sy2 + mrk 270 & 0.010 & 148.&n & 7.60 & 43.37&i&sy2 + mrk 348 & 0.015 & 118.&n & 7.21 & 44.27&i&sy2 + mrk 533 & 0.029 & 144.&n & 7.56 & 45.15&i&sy2 + mrk 573 & 0.017 & 123.&n & 7.28 & 44.44&i&sy2 + mrk 622 & 0.023 & 100.&n & 6.92 & 44.52&i&sy2 + mrk 686 & 0.014 & 144.&n & 7.56 & 44.11&i&sy2 + mrk 917 & 0.024 & 149.&n & 7.62 & 44.75&i&sy2 + mrk 1018 & 0.042 & 195.&n & 8.09 & 44.39&i&sy2 + mrk 1040 & 0.017 & 151.&n & 7.64 & 44.53&i&sy2 + mrk 1066 & 0.012 & 105.&n & 7.01 & 44.55&i&sy2 + mrk 1157 & 0.015 & 95.&n & 6.83 & 44.27&i&sy2 + akn 79 & 0.018 & 143.&n & 7.54 & 45.24&f&sy2 + akn 347 & 0.023 & 186.&n & 8.00 & 44.84&f&sy2 + ic 5063 & 0.011 & 160.&n & 7.74 & 44.53&i&sy2 + ii zw55 & 0.025 & 212.&n & 8.23 & 44.54&f&sy2 + f 341 & 0.016 & 114.&n & 7.15 & 44.13&i&sy2 + ugc 3995 & 0.016 & 155.&n & 7.69 & 44.39&i&sy2 + ugc 6100 & 0.029 & 156.&n & 7.70 & 44.48&i&sy2 + 1es 1959 + 65 & 0.048 & 195.&f & 8.09&-&&bll + mrk 180 & 0.045 & 209.&ba & 8.21&-&&bll + mrk 421 & 0.031 & 219.&ba & 8.29&-&&bll + mrk 501 & 0.034 & 372.&ba & 9.21&-&&bll + i zw 187 & 0.055 & 171.&ba & 7.86&-&&bll + 3c 371 & 0.051 & 249.&ba & 8.51&-&&bll + 1514 - 241 & 0.049 & 196.&ba & 8.10&-&&bll + 0521 - 365 & 0.055 & 269.&ba & 8.65&-&&bll + 0548 - 322 & 0.069 & 202.&ba & 8.15&-&&bll + 0706 + 591 & 0.125 & 216.&ba & 8.26&-&&bll + 2201 + 044 & 0.027 & 197.&ba & 8.10&-&&bll + 2344 + 514 & 0.044 & 294.&ba & 8.80&-&&bll + 3c 29 & 0.045 & 208.&b & 8.20&-&&rg + 3c 31 & 0.017 & 248.&b & 8.50&-&&rg + 3c 33 & 0.059 & 230.&b & 8.38&-&&rg + 3c 40 & 0.018 & 171.&b & 7.86&-&&rg + 3c 62 & 0.148 & 273.&b & 8.67&-&&rg + 3c 76.1 & 0.032 & 200.&b & 8.13&-&&rg + 3c 78 & 0.029 & 261.&b & 8.60&-&&rg + 3c 84 & 0.017 & 246.&b & 8.49&-&&rg + 3c 88 & 0.030 & 189.&b & 8.03&-&&rg + 3c 89 & 0.139 & 250.&b & 8.52&-&&rg + 3c 98 & 0.031 & 173.&b & 7.88&-&&rg + 3c 120 & 0.033 & 200.&b & 8.13&-&&rg + 3c 192 & 0.060 & 192.&b & 8.06&-&&rg + 3c 196.1 & 0.198 & 210.&b & 8.21&-&&rg + 3c 223 & 0.137 & 202.&b & 8.15&-&&rg + 3c 293 & 0.045 & 185.&b & 7.99&-&&rg + 3c 305 & 0.041 & 178.&b & 7.92&-&&rg + 3c 338 & 0.030 & 290.&b & 8.78&-&&rg + 3c 388 & 0.091 & 365.&b & 9.18&-&&rg + 3c 444 & 0.153 & 155.&b &
7.68&-&&rg + 3c 449 & 0.017 & 224.&b & 8.33&-&&rg + gin 116 & 0.033 & 285.&b & 8.75&-&&rg + ngc 315 & 0.017 & 311.&b & 8.90&-&&rg + ngc 507 & 0.017 & 329.&b & 9.00&-&&rg + ngc 708 & 0.016 & 241.&b & 8.46&-&&rg + ngc 741 & 0.018 & 280.&b & 8.72&-&&rg + ngc 4839 & 0.023 & 244.&b & 8.48&-&&rg + ngc 4869 & 0.023 & 199.&b & 8.12&-&&rg + ngc 4874 & 0.024 & 266.&b & 8.63&-&&rg + ngc 6086 & 0.032 & 322.&b & 8.96&-&&rg + ngc 6137 & 0.031 & 295.&b & 8.81&-&&rg + ngc 7626 & 0.025 & 324.&b & 8.97&-&&rg + 0039 - 095 & 0.000 & 280.&b & 8.72&-&&rg + 0053 - 015 & 0.038 & 297.&b & 8.82&-&&rg + 0053 - 016 & 0.043 & 249.&b & 8.51&-&&rg + 0055 - 016 & 0.045 & 302.&b & 8.85&-&&rg + 0110 + 152 & 0.044 & 196.&b & 8.09&-&&rg + 0112 - 000 & 0.045 & 252.&b & 8.53&-&&rg + 0112 + 084 & 0.000 & 365.&b & 9.18&-&&rg + 0147 + 360 & 0.018 & 242.&b & 8.46&-&&rg + 0131 - 360 & 0.030 & 251.&b & 8.53&-&&rg + 0257 - 398 & 0.066 & 219.&b & 8.29&-&&rg + 0306 + 237 & 0.000 & 249.&b & 8.51&-&&rg + 0312 - 343 & 0.067 & 257.&b & 8.57&-&&rg + 0325 + 024 & 0.030 & 219.&b & 8.29&-&&rg + 0431 - 133 & 0.033 & 269.&b & 8.65&-&&rg + 0431 - 134 & 0.035 & 222.&b & 8.31&-&&rg + 0449 - 175 & 0.031 & 158.&b & 7.72&-&&rg + 0546 - 329 & 0.037 & 389.&b & 9.29&-&&rg + 0548 - 317 & 0.034 & 123.&b & 7.28&-&&rg + 0634 - 206 & 0.056 & 195.&b & 8.09&-&&rg + 0718 - 340 & 0.029 & 331.&b & 9.01&-&&rg + 0915 - 118 & 0.054 & 275.&b & 8.69&-&&rg + 0940 - 304 & 0.038 & 389.&b & 9.29&-&&rg + 1043 - 290 & 0.060 & 229.&b & 8.37&-&&rg + 1107 - 372 & 0.010 & 295.&b & 8.81&-&&rg + 1123 - 351 & 0.032 & 447.&b & 9.53&-&&rg + 1258 - 321 & 0.015 & 263.&b & 8.61&-&&rg + 1333 - 337 & 0.013 & 288.&b & 8.77&-&&rg + 1400 - 337 & 0.014 & 309.&b & 8.89&-&&rg + 1404 - 267 & 0.022 & 295.&b & 8.81&-&&rg + 1510 + 076 & 0.053 & 336.&b & 9.03&-&&rg + 1514 + 072 & 0.035 & 269.&b & 8.65&-&&rg + 1520 + 087 & 0.034 & 220.&b & 8.29&-&&rg + 1521 - 300 & 0.020 & 166.&b & 7.80&-&&rg + 1602 + 178 & 0.041 & 213.&b & 8.24&-&&rg + 1610 + 296 & 0.032 & 322.&b & 8.96&-&&rg + 2236 - 176 & 0.070 & 245.&b & 8.49&-&&rg + 2322 + 143 & 0.045 & 204.&b & 8.17&-&&rg + 2322 - 122 & 0.082 & 224.&b & 8.33&-&&rg + 2333 - 327 & 0.052 & 269.&b & 8.65&-&&rg + 2335 + 267 & 0.030 & 345.&b & 9.08&-&&rg + [ t_sig ] lcrrrrrr 0122 + 090 & 0.339 & 20.64 & 4.13&1 & 298 . & 8.82&bll + 0145 + 138 & 0.124 & 20.91 & 3.43&1 & 237 . &
8.42&bll + 0158 + 001 & 0.229 & 21.88 & 5.87&1 & 194 . & 8.08&bll + 0229 + 200 & 0.139 & 21.07 & 6.97&1 & 378 . & 9.24&bll + 0257 + 342 & 0.247 & 21.28 & 5.68&1 & 270 . &
8.66&bll + 0317 + 183 & 0.190 & 22.56 & 8.82&1 & 181 . & 7.95&bll + 0331 - 362 & 0.308 & 22.09 & 11.54&1 & 285 . & 8.75&bll + 0347 - 121 & 0.188 & 20.63 & 3.37&1 & 270 . & 8.65&bll + 0350 - 371 & 0.165 & 20.77 & 4.16&1 & 296 . & 8.82&bll + 0414 + 009 & 0.287 & 22.78 & 16.78&1 & 256 . & 8.56&bll + 0419 + 194 & 0.512 & 19.71 & 1.91&1 & 263 . & 8.61&bll + 0506 - 039 & 0.304 & 21.21 & 5.91&1 & 285 . & 8.75&bll + 0525 + 713 & 0.249 & 21.10 & 6.46&1 & 334 . & 9.03&bll + 0607 + 710 & 0.267 & 21.76 & 8.19&1 & 269 . & 8.65&bll + 0737 + 744 & 0.315 & 21.41 & 7.92&1 & 318 . & 8.94&bll + 0922 + 749 & 0.638 & 19.79 & 4.40&1 & 467 . & 9.61&bll + 0927 + 500 & 0.188 & 21.55 & 5.39&1 & 225 . & 8.34&bll + 0958 + 210 & 0.344 & 20.13 & 3.25&1 & 334 . & 9.03&bll + 1104 + 384 & 0.031 & 19.50 & 2.25&1 & 413 . & 9.39&bll + 1133 + 161 & 0.460 & 21.75 & 7.09&1 & 223 . & 8.32&bll + 1136 + 704 & 0.045 & 20.05 & 2.50&1 & 320 . & 8.95&bll + 1207 + 394 & 0.615 & 20.73 & 6.14&1 & 348 . & 9.10&bll + 1212 + 078 & 0.136 & 21.35 & 7.17&1 & 327 . & 8.99&bll + 1215 + 303 & 0.130 & 23.31 & 16.98&1 & 199 . & 8.12&bll + 1218 + 304 & 0.182 & 21.64 & 6.84&1 & 259 . & 8.58&bll + 1221 + 245 & 0.218 & 21.39 & 3.73&1 & 182 . & 7.97&bll + 1229 + 643 & 0.164 & 20.42 & 4.87&1 & 417 . & 9.41&bll + 1248 - 296 & 0.370 & 20.57 & 4.53&1 & 331 . & 9.01&bll + 1255 + 244 & 0.141 & 21.36 & 5.42&1 & 259 . & 8.58&bll + 1407 + 595 & 0.495 & 21.01 & 8.26&1 & 391 . & 9.30&bll + 1418 + 546 & 0.152 & 21.51 & 8.39&1 & 334 . & 9.03&bll + 1426 + 428 & 0.129 & 20.62 & 4.55&1 & 354 . & 9.13&bll + 1440 + 122 & 0.162 & 22.21 & 9.41&1 & 238 . & 8.44&bll + 1534 + 014 & 0.312 & 21.47 & 7.50&1 & 294 . & 8.80&bll + 1704 + 604 & 0.280 & 20.30 & 2.99&1 & 289 . & 8.77&bll + 1728 + 502 & 0.055 & 21.08 & 3.06&1 & 200 . & 8.13&bll + 1757 + 703 & 0.407 & 20.51 & 3.67&1 & 285 . & 8.75&bll + 1807 + 698 & 0.051 & 18.60 & 1.90&1 & 618 . & 10.10&bll + 1853 + 671 & 0.212 & 21.37 & 4.40&1 & 211 . & 8.23&bll + 2005 - 489 & 0.071 & 21.30 & 6.89&1 & 335 . & 9.03&bll + 2143 + 070 & 0.237 & 21.68 & 6.64&1 & 241 .
& 8.46&bll + 2200 + 420 & 0.069 & 21.80 & 5.71&1 & 212 . &
8.23&bll + 2254 + 074 & 0.190 & 22.48 & 13.29&1 & 264 . & 8.62&bll + 2326 + 174 & 0.213 & 21.13 & 5.29&1 & 284 . & 8.74&bll + 2356 - 309 & 0.165 & 21.08 & 4.52&1 & 262 . &
8.60&bll + 0230 - 027 & 0.239 & 21.80 & 5.13&2 & 182 . & 7.97&rg + 0307 + 169 & 0.256 & 21.40 & 6.27&2 & 271 . & 8.66&rg + 0345 + 337 & 0.244 & 23.30 & 8.73&2 & 112 . & 7.12&rg + 0917 + 459 & 0.174 & 23.00 & 14.60&2 & 209 . & 8.21&rg + 0958 + 291 & 0.185 & 22.00 & 5.67&2 & 178 . &
7.93&rg + 1215 - 033 & 0.184 & 22.00 & 5.67&2 & 179 . & 7.93&rg + 1215 + 013 & 0.118 & 21.00 & 3.13&2 & 209 . &
8.20&rg + 1330 + 022 & 0.215 & 22.90 & 10.47&2 & 167 . & 7.82&rg + 1342 - 016 & 0.167 & 22.90 & 15.53&2 & 234 . & 8.41&rg + 2141 + 279 & 0.215 & 23.50 & 16.53&2 & 168 . &
7.82&rg + 0257 + 024 & 0.115 & 21.70 & 7.80&2 & 285 . & 8.75&rqq + 1549 + 203 & 0.250 & 22.20 & 3.33&2 & 100 . &
6.92&rqq + 2215 - 037 & 0.241 & 21.40 & 4.47&2 & 208 . & 8.20&rqq + 2344 + 184 & 0.138 & 23.80 & 11.67&2 & 109 . & 7.07&rqq + [ t_fp ] | black hole mass , along with mass accretion rate , is a fundamental property of active galactic nuclei .
black hole mass sets an approximate upper limit to agn energetics via the eddington limit .
we collect and compare all agn black hole mass estimates from the literature ; these 177 masses are mostly based on the virial assumption for the broad emission lines , with the broad - line region size determined from either reverberation mapping or optical luminosity .
we introduce 200 additional black hole mass estimates based on properties of the host galaxy bulges , using either the observed stellar velocity dispersion or using the fundamental plane relation to infer @xmath0 ; these methods assume that agn hosts are normal galaxies .
we compare 36 cases for which black hole mass has been generated by different methods and find , for individual objects , a scatter as high as a couple of orders of magnitude .
the less direct the method , the larger the discrepancy with other estimates , probably due to the large scatter in the underlying correlations assumed .
using published fluxes , we calculate bolometric luminosities for 234 agns and investigate the relation between black hole mass and luminosity .
in contrast to other studies , we find no significant correlation of black hole mass with luminosity , other than those induced by circular reasoning in the estimation of black hole mass .
the eddington limit defines an approximate upper envelope to the distribution of luminosities , but the lower envelope depends entirely on the sample of agn included . for any given black hole mass
, there is a range in eddington ratio of up to three orders of magnitude . |
cosmic ray particles with energy in excess of @xmath1 ev have been detected during the last thirty years by several independent experiments , such as agasa ( takeda et al .
1998 ; takeda et al . 1999
; hayashida et al . 1994 ) , fly s eye ( bird et al .
1993 , 1994 , 1995 ) , haverah park ( lawrence , reid and watson 1991 ) , yakutsk ( efimov 1991 ) , volcano ranch ( linsley 1963 ) and more recently by the high resolution fly s eye experiment ( kieda et al .
these events represent now more than ever a big challenge for our understanding of particle physics and astrophysics . while hystorically the first reactions to the detection of these particles were related to the already difficult problem of accelerating particles to the highest observed energies , it became soon clear that the existence of cosmic rays having energy larger than @xmath2 ev [ the so - called ultra - high energy cosmic rays ( uhecr ) ] was more than that , and indeed represented a much more serious challenge to known physics .
soon after the discovery of the cosmic microwave background radiation ( cmbr ) , zatsepin and kuzmin ( 1966 ) and independently greisen ( 1966 ) recognized that the propagation of a proton in the cmbr bath had to be limited to short distances due to photopion production .
if the sources of uhecrs are distributed homogeneously in the sky , this immediately implies that the flux of uhecrs above @xmath2 ev should be strongly suppressed .
this is the so - called greisen - zatsepin - kuzmin ( gzk ) cutoff .
similar arguments apply to nuclei .
this puzzling situation inspired on one side a proliferation of models of the generation of particles with sufficiently high energy , and on the other side it fueled interest in the study of the propagation of uhecrs [ see for instance lee , olinto and sigl 1995 ; lemoine , sigl , olinto and schramm 1997 and bhattacharjee and sigl 2000 , and references therein ] for different compositions and for realistic models for the distributions of the sources and for the intergalactic magnetic field , which still remains only contrained by upper limits , generally based on measurements of the faraday rotation of light coming from distant quasars ( kronberg 1994 ; blasi , burles and olinto 1998 ) .
these limits are at the level of @xmath3 gauss , although larger values are allowed in large scale structures . the angular deflection of ultra high energy protons in such fields would be comparable with or smaller than the angular resolution of current experiments , so that in principle it should be possible to do astronomy using uhecrs as probes .
several efforts have been put into the search for candidate nearby sources in the direction of arrival of uhecrs , but with no result ( see for instance elbert and sommers 1995 ) .
recent analysis of the distribution of arrival directions of uhecrs ( takeda et al .
1999 ; uchihori et al .
2000 ) , in search for a possible large scale anisotropy also gave negative results : with the present statistics , the observed distribution appears to be consistent with isotropy , but indications have been found of small scales anisotropies , at a few degrees level . if confirmed , this finding will hopefully provide hints about the sources of uhecrs .
the paper is structured as follows : in section 2 we give a short outline of the observational situation ; in section 3 we present a critical view of the gzk cutoff , its meaning and its potential power in limiting wide classes of models . in section 4
we introduce the astrophysics and particle physics inspired models ; in section 5 we discuss topological defects variants of td models , while in section 6 we discuss the models of relic quasistable massive particles .
we conclude in section 7 .
the cosmic ray spectrum is measured from fractions of gev to a ( current ) maximum energy of @xmath4 ev .
the spectrum above a few gev and up to @xmath5 ev ( the knee ) is measured to be a power law with slope @xmath6 , while at higher energies and up to @xmath7 ev ( the ankle ) the spectrum has a different slope , of @xmath8 . at energy
larger than @xmath9 ev a flattening seems to be present .
there are currently 92 events above @xmath10 ev , 47 of which have been detected by the agasa experiment .
the information available on the composition of cosmic rays at the highest energies is quite poor .
a study of the shower development was possible only for the fly s eye event ( bird et al .
1995 ) and disfavors a photon primary ( halzen et al .
a reliable analysis of the composition is however possible only on statistical basis , because of the large fluctuations between one shower and the other at fixed type of primary particle .
the fly s eye collaboration reports of a predominantly heavy composition at @xmath11 ev , with a smooth transition to light composition at @xmath7 ev ( see talk by alessandro , these proceedings ) .
this trend was not confirmed by agasa ( hayashida et al .
1994 ; yoshida and dai 1998 ) . in ( takeda et al .
1999 ) the directions of arrival of the agasa events above @xmath10 ev were studied in detail : no appreciable departure from isotropy was found , with the exception of a few small scale anisotropies in the form of doublets and triplets of events within an angular scale comparable with the angular resolution of the experiments ( @xmath12 for agasa ) .
this analysis was repeated in ( uchihori et al .
2000 ) for the whole sample of events above @xmath10 ev , and a total of 12 doublets and 3 triplets were found within @xmath13 angular scales .
the attempt to associate these multiplets with different types of astrophysical sources possibly clustered in the local supercluster did not give positive result ( see stanev , these proceedings ) .
although the existence of uhecrs is experimentally well established , it represents a big challenge from the theoretical point of view , because of a combination of puzzles related to the production and to the propagation of these particles .
we will summarize the different parts of this puzzle in the following .
the first and most important part of the mystery is that the universe should be dark at energies in excess of a few @xmath9 ev due to photopion production of uhecrs on the photons of the cmbr .
this interaction has a typical pathlength @xmath14 mpc ( corresponding to a travel time of @xmath15 years ) at ultra high energies ( see stanev , these proceedings ) . if the sources of uhecrs are distributed nearly homogeneously in the universe , the photopion production results in an observed spectrum which has a pronounced cutoff that starts at @xmath16 ev .
this is the so - called gzk cutoff ( greisen 1966 ; zatsepin and kuzmin 1966 ) .
it is worth spending a few more words on the meaning of the cutoff , since it is so crucial in defining the problem of uhecrs .
particles generated at distances closer than @xmath17 can reach the detector and be uhecr events .
the pathlength @xmath18 becomes gradually larger at lower energies , so that particles with these energies can come from correspondingly larger distances .
it is clear that the problem of uhecrs exists because , for a homogeneous distribution of sources there are not enough nearby sources to provide the observed fluxes . in fig .
1 we show the results of agasa observations at energies @xmath19 ev and the theoretical prediction for a homogeneous distribution of sources and an injection spectrum @xmath20 ( dashed line ) and @xmath21 ( dash - dotted line ) .
two comments are in order on this figure : 1 ) the predicted spectra may present a recovery above some energy , due to the flux contributed by the nearby sources ; 2 ) the energy position of the cutoff increases for a locally overdense distribution of sources . only if the universe presents local overdensities by a factor @xmath22 it is possible to reconcile the expected spectra with the observed spectra and fluxes ( berezinsky et al .
1990 ) , provided suitable sources are found in the local universe . these overdensities are not observed ( blanton , blasi and olinto 2000 ) . at this point
the second part of the mystery enters : upper limits on the intergalactic magnetic field based on the faraday rotation measurements are at the level of @xmath23 gauss ( kronberg 1994 ; blasi et al .
1999 ) , so that typical deflection angles of protons are forced to be within a few degrees .
searches for sources of uhecrs have been carried out , but no plausible candidate was found within @xmath13 of the direction of arrival of the events ( elbert and sommers 1995 ) .
if the candidate particle is an iron nucleus then the deflection angles are likely to become larger and it becomes correspondingly harder to look for sources . however , if the deflection angle becomes too large , then the regime of propagation becomes closer to diffusive and this forces the distance to the source to be even smaller , making the situation more problematic .
a possible exception to this is if the iron nuclei are accelerated locally in the galaxy and deflected and isotropized in an extended halo . in this case
no gzk cutoff is expected .
the problems mentioned above are all related to the propagation of uhecrs .
however there is another problem , that hystorically was the first to be studied , and is related to the mechanism able to produce such particles .
we dedicate the next section to this topic .
astrophysical models for the production of uhecrs are generally based on an acceleration mechanism to be applied to a class of scenarios of astrophysical relevance .
we do not discuss here any of the acceleration models , but we outline the general features and problems they encounter
. an exhaustive discussion of most of the models which are currently under investigation can be found in ( olinto 2000 ) and some of them are discussed by stanev ( these proceedings ) .
a very broad classification of these models is that between galactic and extragalactic ones . in general ,
galactic models require heavy composition , in order to emulate the isotropic distribution of arrival directions . in most of the extragalactic models that have been proposed so far ,
the highest energies are reached only for the extreme values of the parameters involved [ see ( norman , melrose and achterberg 1995 ) ] .
strong constraints on astrophysical models come from the spectrum they generate ( after accounting for propagation ) .
in fact , as long as the acceleration models are associated with sources with a homogeneous distribution in the sky , the problem of uhecrs remains , independently on the maximum energy .
the case of a single source accidentally present in the nearby universe ( for instance in the local supercluster ) was investigated by berezinsky , grigorieva and dogiel 1990 ; blasi and olinto 1999 ; sigl , lemoine and biermann 1999 ) .
even with relatively strong fields of @xmath24 gauss , the anisotropy is too large compared with observations ( see however ( ahn et al . 1999 ) for an alternative model involving m87 ) .
the difficulty in acceleration to the highest energies ( norman , melrose and achterberg 1995 ) , the presence of the gzk cutoff in most of the cases and the lack of counterparts in the arrival directions fueled the interest in a new class of models that could avoid these problems .
these are _ particle physics inspired models _ , in which uhecrs are generated as a result of the decay of very massive particles ( from here the name of top - down models ) .
the problem of reaching the maximum energies is , in these models , solved _ by construction_. the spectra of the particles resulting from the decay are determined in principle by qcd , the channel of reactions being : @xmath25 , where @xmath26 are quarks that hadronize , generating mainly pions and a small fraction of protons and neutrons .
the spectra are generally very flat ( roughly @xmath27 , although the realistic calculations do not give power law spectra ) , which represents one of the peculiar features of td models . at the production ,
most of the ultra - high energy particles are gamma rays , but propagation effects can change the ratio of gamma rays to protons .
the gamma rays generated at distances larger than the absorption length produce a cascade at low energies ( mev - gev ) which represents a powerful tool to contrain td models .
there are basically two ways of generating the very massive particles and make them decay at the present time : 1 ) trapping them inside topological defects ; 2 ) making them quasi - stable ( lifetime larger than the present age of the universe ) in the early universe .
we discuss these two possibilities separately in the next two sections .
topological defects are naturally formed at phase transitions and their existence has been proven by direct observations in several experiments on liquid crystals and ferromagnetic materials .
similar symmetry breakings at particle physics level are responsible for the formation of cosmic topological defects [ for a review see ( vilenkin and shellard 1994 ) ] .
the fact that topological defects can generate uhecrs was first proposed in the pioneering work of hill , schramm and walker ( 1987 ) .
the general idea is that the stability of the defect can be locally broken by different types of proceses ( see below ) : this results in the false vacuum , trapped within the defect , to fall into the real vacuum ( outside universe ) , so that the gauge bosons of the field trapped in the defect acquire a mass @xmath28 . at this point
, the very massive and unstable particles rapidly decay producing high energy particles .
several topological defects have been studied in the literature : ordinary strings ( bhattacharjee and rana 1990 ) , superconducting strings ( hill , schramm and walker 1987 ) , bound states of magnetic monopoles ( hill 1983 ; bhattacharjee and sigl 1995 ) , networks of monopoles and strings ( berezinsky , martin and vilenkin 1997 ) , necklaces ( berezinsky and vilenkin 1997 ) and vortons ( masperi and silva 1998 ) . only strings ,
necklaces and monopolonia will be considered here , while a more extended discussion can be found in more detailed reviews ( bhattacharjee and sigl 2000 ; berezinsky , blasi and vilenkin 1998 ) .
strings can generate uhecrs if there are configurations in which microscopic or macroscopic portions of strings annihilate . in the contact regions the phase of the field trapped in the string becomes undetermined and
the vacuum expectation value becomes non zero .
if @xmath29 is the symmetry breaking scale at which the string formed , it is easy to see that during intercommutation of strings or self - intersection , only one ( or a few ) x - particles are generated .
it was shown ( shellard 1987 ; gill and kibble 1994 ) that self - intersection events provide a flux of uhecrs which is much smaller than required . the same conclusion holds for intercommutation between strings .
the efficiency of the process can be enhanced by multiple loop fragmentation : as a nonintersecting closed loop oscillates and radiates its energy away , the loop configuration gradually changes .
after the loop has lost a substantial part of its energy , it becomes likely to self - intersect and fragment into smaller and smaller loops , until the typical size of a loop becomes comparable with the string width @xmath30 . at this point
the energy is radiated in the form of x - particles .
although the process of loop fragmentation is not well known , some analytical approximations ( berezinsky , blasi and vilenkin 1998 ) show that appreciable uhecr fluxes imply utterly large gamma ray cascade fluxes .
battacharjee and sigl ( 2000 ) argued however that there might be models of loop fragmentation in which this result is mitigated .
another way of liberating x - particles is through cusp annihilation ( brandenberger 1987 ) .
cusps can be produced along a string loop ( turok 1984 ) or due to kinks propagating in opposite directions on a long string ( mohazzab and brandenberger 1993 ) .
although during the cusp annihilation a macroscopic fraction of the string length can be transformed into x - particles , the corresponding uhecr flux is far too low ( bhattacharjee 1989;gill and kibble 1994 ) .
the idea that long strings lose energy mainly through formation of closed loops was recently challenged by vincent , antunes and hindmarsh ( 1998 ) .
their simulations seem to show that the string can produce x - particles directly and that this process dominates over the generation of closed loops .
this new picture was recently questioned by moore and shellard ( 1998 ) . even if the results of vincent et al .
( 1998 ) are correct however , they can not solve the problem of uhecrs ( berezinsky , blasi and vilenkin 1998 ) : in fact the typical separation between two segments of a long string is comparable with the hubble scale , so that uhecrs would be completely absorbed .
if by accident a string is close to us ( within a few tens mpc ) then the uhecr events would appear to come from a filamentary region of space , implying a large anisotropy which is not observed .
even if the uhecr particles do not reach us the gamma ray cascade due to absorption of uhe gamma rays produced at large distances imposes limits on the efficiency of direct production of x - particles by strings .
the role of monopolonia ( bound states of monopoles and antimonopoles ) for uhecrs was first pointed out by hill ( 1983 ) and schramm and hill ( 1983 ) .
when the monopolonium is produced it is in a very excited state , and later decays to the ground state , where the typical distance between the monopole and the antimonopole is smaller than the size of the inner ( quantum ) stable orbit . at this point
the monopole and the antimonopole annihilate and generate x - particles .
this process was studied in detail by bhattacharjee and sigl ( 1995 ) .
however , recently blanco - pillado and olum ( 1999 ) have found that plasma friction on the monopoles result in a short lifetime for the monopolonium , making it useless for uhecrs . as an alternative ,
a similar system was proposed by berezinsky , martin and vilenkin ( 1997 ) , where the symmetry breakings @xmath31 ( @xmath32 ) results first in the formation of monopoles and then of strings connecting them and forming an infinite network .
the shrinking of strings during their evolution causes the distance between monopoles to decrease . during this stage ,
monopoles accelerate and therefore radiate very high energy gluons , that generate hadrons through fragmentation .
the fluxes of uhecrs that result from this process are negligible ( berezinsky , blasi and vilenkin 1998 ) .
an interesting special case of monopole - string networks is realized when the following chain of symmetry breakings occurs : @xmath33 . in this case
each monopole gets attached to two strings , to form a necklace ( berezinsky and vilenkin 1997 ) .
the critical parameter that defines the dynamics of this network is the ratio @xmath34 where @xmath35 is the monopole mass and @xmath36 is the typical separation between monopoles ( e.g. the length of a string segment ) .
berezinsky and vilenkin ( 1997 ) proposed that there might be cases in which the system evolves toward a state where @xmath37 , although only numerical simulations can confirm this point . in this case , the distance between the monopoles decreases and in the end the monopoles annihilate , with the production of x - particles and their decay to uhecrs .
the rate of generation of x - particles is easily found to be @xmath38 .
the quantity @xmath39 is upper limited by the cascade radiation , given by @xmath40 ( @xmath41 ) .
the typical distance from the earth at which the monopole - antimonopole annihilations occur is comparable with the typical separation between necklaces , @xmath42 kpc .
clearly , necklaces provide an example in which the typical separation between defects is smaller than the pathlength of gamma rays and protons at ultra - high energies .
hence necklaces behave like a homogeneous distribution of sources , so that the proton component has the usual gzk cutoff .
this component dominates the uhecr flux up to @xmath1 ev , while at higher energies gamma rays take over .
the fluxes obtained in ( berezinsky , blasi and vilenkin 1998 ) are reported in fig .
2 , where the susy - qcd fragmentation functions ( berezinsky and kachelriess 1998 ) were used .
the dashed lines are for @xmath43 gev , the dotted lines for @xmath44 gev and the solid lines for @xmath45 gev .
the two curves for gamma rays refer to two different assumptions about the radio background at low frequencies ( protheroe and biermann 1996 ) .
super heavy particles with very long lifetime can be produced in the early universe and generate uhecrs at present .
the existence of particles satisfying these requirements was studied recently by berezinsky , kachelriess and vilenkin ( 1997 ) ; kuzmin and rubakov ( 1997 ) , chung , kolb and riotto ( 1998 ) and kuzmin and tkachev ( 1998 ) and reviewed by kuzmin and tkachev ( 1998a ) . in order to keep the same symbolism used in previous sections
, we will call these particles x - particles .
x - particles can be produced in the early universe through different mechanisms .
the simplest of them is the _ gravitational production _ :
particles are produced naturally in a time variable gravitational field or indeed in a generic time variable classical field . in the gravitational case
no additional coupling is required ( all particles interact gravitationally ) .
if the time variable field is the inflaton field @xmath46 , a direct coupling of the x - particles to @xmath46 is needed .
the gravitational production of particles was first proposed by zeldovich and starobinsky ( 1972 ) .
it does not require any additional assumption neither on the x - particles nor on cosmology .
in particular inflation is not required a priori , and indeed it reduces the effect .
it can be shown that at time t , gravitational production can only generate x - particles with mass @xmath47 , where @xmath48 is the hubble constant and @xmath49 is the inflaton mass .
chung , kolb and riotto ( 1998 ) and kuzmin and tkachev ( 1998 ) demostrated the impressive result that the fraction of the critical mass contributed by x - particles with @xmath50 gev produced gravitationally is @xmath51 , with no additional assumption ! in other words ,
cold dark matter can naturally be explained in terms of x - particles in this range of masses .
if the x - particles are directly coupled to the inflaton field , they can be effectively generated during preheating ( kofman , linde and starobinsky 1994 ; felder , kofman and linde 1998 ) .
alternative mechanisms for the production of x - particles are based on non - equilibrium thermal generation during the preheating stage ( berezinsky , kachelriess and vilenkin 1997 ) . as mentioned in the beginning of this section , in order for x - particles to be useful dark matter candidates and generate uhecrs they need to be long lived .
the gravitational coupling by itself induces a lifetime much shorter than the age of the universe for the range of masses which we are interested in .
therefore , in order to have long lifetimes , additional symmetries must be postulated : for instance discrete gauge symmetries can protect x - particles from decay , while being very weakly broken , perhaps by instanton effects ( kuzmin and rubakov 1998 ) .
these effects can allow decay times larger than the age of the universe , as shown by hamaguchi , nomura and yanagida ( 1998 ) .
the slow decay of x - particles produces uhecrs .
the interesting feature of this model is that x - particles cluster in the galactic halo , as cold dark matter ( berezinsky , blasi and vilenkin 1998 ) . [
if monopolonia survived they would also cluster in the halo ] .
hence uhecrs are expected to be produced locally , with no absorption , and as a consequence the observed spectra are nearly identical to the emission spectra , and therefore gamma rays dominate .
the very flat spectra and the gamma ray composition are two of the signatures
. the calculations of the expected fluxes have been performed by berezinsky , blasi and vilenkin ( 1998 ) , birkel and sarkar ( 1998 ) and blasi ( 1999 ) . in figure 3
we report the results found in ( blasi 1999 ) .
the two solid lines are obtained for susy - qcd fragmentation functions and the dashed lines are for the qcd fragmentation function in the mlla approximation , with @xmath43 gev ( solid lines ) and @xmath52 gev ( thin lines ) .
the strongest signature of the model is a slight anisotropy due to the asymmetric position of the sun in the galaxy ( dubovsky and tinyakov 1998 ; berezinsky , blasi and vilenkin 1998 ) .
more recently a detailed evaluation of the amplitude and phase of the first harmonic has been carried out by berezinsky and mikhailov ( 1999 ) and medina tanco and watson ( 1999 ) .
the two papers agree that the present data is consistent with the anisotropy expected in the model of x - particles in the halo .
in fact , as discussed in section 2 , observations at present do not suggest any appreciable deviation from isotropy , with the exception of a few degree scale anisotropies showing up in the form of doublets and triplets of events within an angular scale comparable to the resolution of the experiments .
uchihori et al .
( 2000 ) investigated the total of 92 events above @xmath10 ev collected by volcano ranch , haverah park , yakutsk and akeno and found 12 doublets and 3 triplets within @xmath53 . in td models
the presence of these multiplets of events is usually not well accomodated because of the homogeneous distribution of the topological defects or of the superheavy particles .
however , it was shown by blasi and sheth ( 2000 ) that in the latter the presence of the multiplets is actually naturally explained , taking into account the dark matter distribution in the galactic halo . the spatial distribution of cold dark matter has been studied in detail in n - body simulations ( ghigna et al . 1999 ; moore et al . 1999 ; tormen , diaferio and syer 1998 ) and a few elements seem to be well established : 1 ) the density of dark matter has a behaviour @xmath54 with @xmath55 in the central part of galaxies ; 2 ) at large radii the radial profile is @xmath56 ; 3 ) in addition to the smooth component , simulations show very clearly the existence of a clumped component .
berezinsky ( 2000 ) suggested that the multiplets might be correlated with the clumps of dark matter .
blasi and sheth ( 2000 ) carried out a numerical simulation of the arrival directions of uhecr events from a realistic distribution of dark matter in the halo ( including the clumps ) and derived the probabilities to detect the observed numbers of doublets and triplets ( or more ) .
they adopted a navarro - frenk - white ( nfw : navarro , frenk and white 1996 ) smooth distribution of dark matter and used the results of the n - body simulations for the small scale structure , assuming each clump being represented by an isothermal sphere .
the numbers of doublets and triplets found by generating a large number of halo configurations and the corresponding pattern of arrival directions of uhecrs are comparable with the observed ones and in excess of the number expected from an isotropic distribution of arrival directions .
surprisingly , the main reason for the increase in the predicted number of doublets and triplets with respect to isotropy is that the smooth component has an nfw profile .
the effect of the clumps , apart from increasing slightly the numbers of multiplets , is to enhance the probability of having more than the average number of doublets and triplets .
this result can be understood by accounting for the shape of the dark matter profile , increasing toward the central part of the galaxy .
predictions for future full sky experiments were proposed .
the problem of the existence of uhecrs is all but solved .
astrophysical models are being narrowed down by several constraints : most of the extragalactic sources would suffer a gzk cutoff , since the large overdensities needed locally to avoid it do not seem to be present . moreover it is difficult to reach the highest energies ( norman , melrose and achterberg 1995 ) .
galactic accelerators usually require an iron dominated composition in order to account for the isotropy of arrival directions .
further studies are needed to quantify the expected and observed anisotropies in these models , since the magnetic field strength and extension in the halo are poorly known .
top - down models naturally provide the highest energies , and at least some of them can describe quite well the observed sprectal shape above @xmath2 ev . as positive examples we considered in some detail the cases of necklaces and relic super massive particles in the halo of the galaxy . in the former model , the composition of uhecrs should be dominated by protons up to @xmath1 ev , and by photons at higher energies . in the latter , uhecrs are gamma rays and a peculiar pattern of large and small scale anisotropies should be visible in future observations . in all top - down models
there is at least an appreciable fraction of uhecrs in the form of gamma rays .
experiments like hires ( corbato et al .
1992 ) , currently operating , the auger project ( cronin 1992 ) , the proposed telescope array ( teshima et al .
1992 ) and the owl - airwatch satellite ( streitmatter 1997 ) will nail down the answers to the dark points .
an unambiguous determination of the composition will be fundamental : heavy composition would rule out all top - down models and open a window of opportunity for galactic scenarios . a gamma ray dominated composition would instead be a smoking gun for top - down models
. the measurement of the anisotropy on different scales will also be a crucial step : galactic models all predict some degree of anisotropy toward the galactic disk or center . a peculiar pattern of anisotropy is also predicted by super heavy particles clustered as dark matter in the galactic halo .
* aknowledgments * this work was supported by the doe and the nasa grant nag 5 - 7092 at fermilab .
ahn , e.j . , biermann , p.l . , medina - tanco , g. , stanev , t. : 1999 , preprint astro - ph/9911123 .
streitmatter , r.e . : 1997 proc . of _ workshop on observing giant cosmic air showers from @xmath57 ev particles from space _ , eds .
krizmanic , j.f . ,
ormes , j.f . , and streitmatter , r.e .
( aip conference proceedings 433 , 1997 ) . | while several arguments can be proposed against the existence of particles with energy in excess of @xmath0 ev in the cosmic ray spectrum , these particles are actually observed and their origin seeks for an explanation . after a description of the problems encountered in explaining these ultra - high energy cosmic rays ( uhecrs ) in the context of astrophysical sources , we will review the so - called _ top - down _ ( td ) models , in which uhecrs are the result of the decay of very massive unstable particles , possibly created in the early universe .
particular emphasis will be given to the signatures of the td models , likely to be accessible to upcoming experiments like auger .
= -1.2truecm = -2.5truecm = cmbx10 = cmr10 = cmti10 = cmbx10 scaled1 = cmr10 scaled1 = cmti10 scaled1 = cmbx9 = cmr9 = cmti9 = cmbx8 = cmr8 = cmti8 = cmr7 = cmti7 * opening the ultra high energy cosmic ray window * * from the top * pasquale blasi nasa / fermilab astrophysics center fermi national accelerator laboratory , box 500 , batavia , il 60510 - 0500 |
the variety of qpo modes from old accreting neutron stars in low mass x - ray binaries ( lmxrbs ) ( see van der klis 1995 ) has been recently enriched by the discovery of kilohertz oscillations in light curves obtained with the rossi x - ray timing explorer ( rxte ) .
these are the first qpos that involve timescales comparable to the dynamical timescales in the vicinity of a collapsed star .
there are currently eleven lmxrbs from which qpos have been detected between frequencies of @xmath5 and @xmath6 khz .
most of these are type i x - ray bursters with persistent luminosities in the @xmath7 ergs s@xmath8 range , that according to the classification of hasinger and van der klis ( 1989 ) either belong to ( 4u 1728 - 34 , 4u 1608 - 52 , 4u 1636 - 536 , 4u 1735 - 444 and 4u 1820 - 30 ) or are suspected members of ( 4u 0614 + 091 and ks 1731 - 260 ) the atoll " group .
sco x-1 , gx 5 - 1 and gx 17 + 2 are instead high luminosity z "- type lmxrbs .
a pair of khz qpo peaks is often present in the power spectra .
this is the case for the persistent emission of the five atoll sources , at least on occasions and/or over some range of count rates . in 4u 1728 - 34 and 4u 0614 + 091 the centroid frequency of the qpos
is positively correlated with the source count rate across the factor of @xmath9 qpo frequency variations ( from @xmath10 to @xmath11 hz ) observed from both sources . when a pair of peaks is present , the difference @xmath12 between their centroid frequencies @xmath13 and @xmath14 is consistent with being constant . in 4u 1728 - 34
@xmath12 remains around @xmath15 hz ( within a few percent ) as the frequency of the khz qpos changes by @xmath16% .
a nearly coherent signal at @xmath17 hz has been detected during a number of x - ray bursts ( strohmayer 1996a ) . in the persistent emission of 4u 0614 + 091
the separation @xmath12 is consistent with @xmath18 hz ( within @xmath19% ) , across a @xmath20% variation of the qpo frequency and a @xmath21 d interval between two rxte observations . during part of an observation
, a third narrower peak has been marginally detected at a frequency of @xmath22 hz , consistent with the separation of the pair of khz qpo peaks ( ford 1997 ) . in the persistent emission of ks 1731 - 260 ,
two khz qpo peaks have been detected around @xmath23 and @xmath24 hz .
the frequency difference ( @xmath25 hz ) between these qpos is consistent with half the frequency of the nearly coherent @xmath26 hz signal observed in a type i burst from this source ( smith 1997 ; wijnands & van der klis 1997 ) .
khz qpos have also been discovered in 4u 1608 - 52 ( berger 1996 ) and 4u 1735 - 444 ( wijnands 1996 ) .
a qpo peak with an intensity - dependent frequency from @xmath27 to @xmath28 hz has been detected in 4u 1820 - 30 .
close to the upper end of this range a second qpo peak at @xmath29 hz is present ( smale , zhang & white 1997 ) . in the persistent emission of 4u 1636 - 536 a pair of qpo peaks around 840 - 920 hz and 1150 - 1200
hz have been detected , the frequency separation of which is constant around @xmath30 hz . a nearly coherent signal at @xmath31 hz modulates the x - ray flux of this source during type i bursts ( zhang 1996 , wijnands 1997 ) ; note that half this frequency is compatible with the separation @xmath12 .
periodic oscillations at 589 hz have also been observed during type i bursts from an unknown burster in the direction of the galactic centre , perhaps mxb 1743 - 29 ( strohmayer 1996b ) .
gx 5 - 1 , gx 17 + 2 and sco x-1 , currently the only high - luminosity z "- type lmxrbs in the group , display khz qpos ( @xmath32 khz ) with a centroid frequency which is positively correlated with the inferred mass accretion rate . in gx 5 - 1
a pair of khz qpo peaks has been detected over part of the horizontal branch ( hb ) , which mantain a constant separation of @xmath33 hz .
the higher frequency qpo peak of gx 5 - 1 increases from @xmath34 to @xmath23 hz as the source moves from the left to the right end of the hb ( van der klis 1996a ) . in gx 17 + 2 a khz qpo peak is observed to drift from @xmath35 to @xmath36 hz qpo as the source moves down the upper part of the so - called normal branch .
a higher frequency peak separated by @xmath37 hz is observed close to the lower end of this range , together with an additional qpo peak around @xmath38 hz , interpreted as hb oscillations ( van der klis 1997a ) . in sco x-1 , @xmath39 khz qpos have been detected in the normal and flaring branches ; their frequency correlates well both with the inferred variations of the mass accretion rate and the frequency variations of the @xmath40 hz qpo that occur along the two branches .
a pair of khz qpo peaks has been detected on occasions .
however their separation decreased from @xmath41 to @xmath42 hz as the higher frequency peak moved from @xmath43 to @xmath44 hz .
correspondingly additional qpos at @xmath45 hz were present , which are intepreted as hb qpos ( van der klis 1996b , 1997b ) .
modelling of khz qpos has focussed on the presence of peaks at @xmath13 and @xmath14 , the constancy of their separation @xmath12 and the coherence of the signal ( either during type i bursts or in the persistent emission ) at frequency @xmath12 ( or @xmath46 ) ( white & zhang 1997 ) ] .
the application of beat frequency concepts to these signals is natural . in the magnetospheric beat frequency model , mbfm , originally developed to interpret the 5 - 50 hz qpos from high luminosity z "- type lmxrbs ( alpar & shaham 1985 ; lamb 1985 ) , matter inhomogeneities in the accretion disk cause accretion to be modulated at the beat frequency between the keplerian frequency , @xmath47 , of the innermost region of the accretion disk ( @xmath48 ) at the magnetospheric boundary , and the neutron star spin frequency , @xmath49 . as @xmath48 decreases with increasing accretion rates , @xmath47 increases and so does the qpo beat frequency @xmath50 .
of the various qpo modes with frequencies @xmath51 hz in z "- type lmxrbs , only hb qpos are currently interpreted in terms of the mbfm ( for a review see lamb 1991 ) .
the basic physical picture of the mbfm has been confirmed mainly through the study of qpos in a few accreting x - ray pulsars ( exo 2030 + 375 and a 0535 + 26 in particular ) , where , unlike z "- type lmxrbs , the neutron star spin frequency @xmath49 is ( by definition ) clearly detected , and independent estimates of the neutron star magnetic field exist ( angelini 1989 ; finger 1996 ) . besides the signals at @xmath49 and @xmath52 , an additional qpo signal at @xmath47
is expected in the mbfm from inhomogeneities at the inner disk boundary . for self - luminous blobs ,
the amplitude of this signal should be higher for smaller values of @xmath48 , due to the larger local gravitational energy and keplerian velocities ( causing a larger doppler boosting around the orbit ) .
similarly , a small value of @xmath48 increases the range of inclination angles for which a signal at @xmath47 can be caused by the neutron star occulting the blobs at @xmath48 or conversely , by the blobs at @xmath48 obscuring part of the radiation produced closer to the neutron star surface . in the mbfm for khz qpos from lmxrbs ,
the lower frequency qpos are interpreted as the beat frequency ( @xmath53 ) and the higher frequency qpos as the keplerian frequency ( @xmath54 ) at the magnetospheric boundary .
the nearly coherent signal , occasionally detected at a frequency compatible with the separation @xmath12 of the qpo pair ( or with @xmath46 ) , provides the first direct evidence of the neutron star spin frequency @xmath49 in lmxrbs containing an old neutron star ( strohmayer 1996a ; ford 1997 ) .
inferred magnetospheric radii @xmath55 km , ( here @xmath56 is the neutron star mass and @xmath57 hz ) correspond to a couple of neutron star radii at the most . the neutron star spin frequency of several hundred hz and the inferred magnetic field strength of @xmath58 g agree with spin up by accretion torques , and scenarios for the formation of millisecond pulsars .
miller ( 1996 ) proposed an alternative beat frequency model based on the idea that the magnetic field of the neutron star is so low that it does not affect the dynamics of the innermost disk regions and , yet , high enough to give rise to beamed radiation at the neutron star surface .
we do not discuss the details of this model , but notice that here too the qpo signal at the highest frequency @xmath14 corresponds to the innermost keplerian frequency of the disk , which in this model ends at the sonic radius where the radial velocity becomes supersonic ( as opposed to the magnetospheric radius ) , and the lower frequency @xmath13 is the beat frequency between the keplerian frequency and the neutron star spin frequency , exactly like in the mbfm .
the parameters inferred from beat frequency models for the khz qpo from lmxrbs are such that frame dragging , one of the general relativistic effects of the
gravitomagnetic " field associated with rotating bodies , is substantial and can lead to observable phenomena . in particular the orbital plane of a test particle ,
if not coaligned with the equatorial plane , must precess in a prograde fashion around the angular momentum axis of the rotating object . in the weak field limit the nodal precession frequency , first calculated by lense & thirring ( 1918 )
( see also wilkins 1972 ) , is given by @xmath59 , where @xmath60 and @xmath61 are the mass and specific angular momentum of the rotating object and @xmath62 is the radius of the test particle ( circular ) orbit . by using @xmath63 , and @xmath64
this can be written as @xmath65 where @xmath66 g @xmath67 is the moment of inertia of the neutron star and @xmath68 hz its spin frequency . by using the values of @xmath49 and
@xmath47 inferred from the application of beat frequency models to the khz qpos from lmxrbs , it is apparent that lense thirring
frequencies of a few tens of hz are expected in the innermost disk region which produces the qpo signal at @xmath47 .
note also that in this regime , higher order terms in @xmath69 would provide a correction to eq .
( 1 ) at a level of less than one percent . since sources with a pair of khz qpo peaks allow identification of both @xmath47 and @xmath49 , the only parameter in eq .
( 1 ) that is not immediately identified from observations is @xmath70 .
the theoretical uncertainty in its determination depends on the choice of the equation of state ( eos ) and the star mass : from models of rotating neutron stars ( friedman , ipser and parker 1986 ; cook , shapiro & teukolsky 1994 ) , it can be seen that @xmath70 varies in the range @xmath71 for any mass and eos .
we notice that @xmath72 in eq .
( 1 ) is thus independent of any other uncertainty related to lmxrbs , such as source distance , mass accretion rate or efficiency of conversion of gravitational energy into radiative energy .
the neutron star spin frequencies discussed here lead to a stellar oblateness which , in turn , causes the appearance of a quadrupole term in the gravitational potential .
it is well known that the quadrupole term causes a precession of orbits tilted away from the star equatorial plane , with the precession frequency @xmath73 ( as opposed to @xmath74 for the lense - thirring precession ) . for the small radii of interest to us , it is important to check that this classical precession rate is still smaller than the lense thirring rate , to make sure that the effect we predict is a genuine relativistic one .
the classical precession frequency @xmath75 is given by ( kahn and woltjer 1959 ) @xmath76 where @xmath77 is the tilt angle off the equatorial plane , and @xmath78 is the coefficient of the quadrupole term in an expansion of the gravitational potential in spherical harmonics , @xmath79 , with @xmath80 the second order legendre polynomial .
outside the star , we have @xmath81 , where the coefficient @xmath82 is given , in terms of the star moment of inertia , by @xmath83 , where @xmath84 is the star rotation axis . for rotationally flattened stars , far from breakup , @xmath85 . for rotationally flattened stars , @xmath86 , and both @xmath87 and @xmath88
are increased by rotation . in the existing literature
we have been able to find only the numerical values of @xmath89 for non - rotating neutron star models and the corresponding rotation increased values , @xmath90 ( friedman , ipser and parker 1986 , cook , shapiro and teukolsky 1994 ) .
using the rotating polytropic models calculated by james ( 1964 ) , we derived the approximate relation @xmath91 . assuming that this holds also for neutron star models , we have @xmath92 .
this is probably a fairly accurate approximation because the rotation frequencies we are interested in , @xmath93 , correspond to still slowly rotating neutron star models : in fact , perusal of the tabulated models in the references given above shows that the ratio @xmath94 of rotational to binding energy for these @xmath49 is @xmath95 , while models nearly at breakup have @xmath96 , for all eos .
we find , defining @xmath97 , where @xmath98 , that the parameter @xmath99 is roughly independent of spin frequency in the slow rotation regime of interest here , and it depends , albeit weakly , upon model mass and eos ( range of @xmath100 ) . eventually , we find @xmath101 note that the negative sign reflects the fact that classical precession is retrograde for @xmath102 .
comparison of this with equation [ 1 ] shows that , even at these small radii , lense thirring precession is faster .
the total precession frequency @xmath103 is thus somewhat decreased by oblateness induced precession .
modulation of the x - ray flux at the precession frequency of the inner disk boundary is expected to occur through the same geometrical effects producing the modulation at the keplerian frequency .
in fact , the precessional motion of the blobs generating the signal at @xmath47 would give rise to a signal at the precession frequency , both for self - luminous blobs ( the projected velocity along the line of sight and , therefore , the doppler boosting will be modulated at the precession frequency ) and occulting , or occulted , blobs ( the duration of the occultations wiil depend on the phase of the precessional motion ) .
moreover we note that , unlike the signal at @xmath47 , an x - ray modulation at the precession frequency can in principle be generated , even in the absence of blobs , by the changing aspect of the innermost disk ring .
this would modulate the ring emission as seen from the earth or could partially occult the emitting regions close to the neutron star surface .
lense thirring precession of viscous accretion disks is generally neglected because , following the pioneering work by bardeen and patterson ( 1975 ) , many authors found that viscosity leads the disk to lie in the equatorial plane of the rotating object .
however , this conclusion has recently been challenged by ( some of ) the same authors : even if the innermost disk regions are confined to the equatorial plane of the neutron star , pringle ( 1996 ) and maloney , begelman and pringle ( 1996 ) have shown that radiation pressure can lead to corrugation of the disk and its lifting off the neutron star equatorial plane .
pringle ( 1996 ) has shown that even initially planar disks are unstable to warping , so that tilting should be generic in centrally illuminated sources .
precessing disks have also been recently revisited by papaloizou ( 1997 ) .
it is possible that other effects , such as parametric amplification ( binney 1978 ) of the z oscillations of diamagnetic blobs by a dipolar magnetic field tilted with respect to the star spin axis , may intervene ( vietri and stella , in preparation ) .
the power spectra of 4u 1728 - 34 given in fig .
1a clearly show a broad peak centered around a frequency of @xmath16 , 26 and 35 hz when the higher frequency qpo peak is at a frequency of @xmath104 , 980 and 1100 hz , respectively .
the width of the @xmath105 hz peak is of @xmath106 hz .
the power spectrum of 4u 0614 + 091 ( see fig .
1b ) and ks 1731 - 260 ( see fig .
1c ) also shows a similarly broad peak centered around @xmath107 hz and @xmath108 hz , when @xmath104 hz and @xmath109 hz , respectively .
we propose that these tens of hz qpo peaks arise from the precession of the innermost disk regions , where the signals at the keplerian and beat frequencies also originate .
the thick bars in fig .
1 represent the precession frequency calculated from the sum of the lense - thirring and classical frequencies ( eqs . 1 and 3 ) by using @xmath110 ( see above ) , and @xmath111 , 323 and 262 hz in the case of 4u 1728 - 34 , 4u 0614 + 091 and ks 1731 - 260 , respectively . in all three cases
, we adopted the same neutron star model with eos l , @xmath112 , @xmath113 and @xmath114 ( friedman , ipser & parker 1986 ) .
lense - thirring precession dominates , with the classical term providing a @xmath16% correction .
the fact that the ( absolute ) width of the tens of hz qpo peaks is comparable to that of the khz qpo peaks indicates that lifetime broadening dominates the coherence in both cases .
both @xmath47 and @xmath49 are independently determined in these three sources .
thus the range of precession frequencies allowed by the model is determined _ only _ by the neutron star parameters @xmath115 and , to a lesser extent , @xmath116 and @xmath99 .
the intervals marked by the filled squares and thick bars in fig . 1
represent the whole range of precession frequencies that has been obtained by using virtually all neutron star models given by friedman , ipser & parker ( 1986 ) and cook , shapiro & teukolsky ( 1994 ) ( we excluded models with @xmath117 and supramassive " models ) .
we regard the very good agreement with the frequency of the corresponding power spectrum peaks as a strong indication in favor of our relativistic plus classical precession interpretation .
we note that for all three sources the qpo peaks are very close to the high - frequency end of the intervals corresponding to the relatively stiff eos l and @xmath118 .
as the frequency of the khz qpos from 4u 1728 - 34 varied , the best fit precession frequency remained in good agreement with the observed peak , showing that the power law dependence ( eq . [ 1 ] ) is probably in the correct range .
the detection of khz qpos in the hb of gx 5 - 1 is especially important for the modelling of qpos in z "- type lmxrbs .
if one retains the standard interpretation in which the @xmath119 hz qpos in the hb arise from the mbfm , then a magnetosphere extending up to @xmath120 neutron star radii is required ( implying a magnetic field of @xmath121 g ) which argues against any interpretation of the khz qpos involving keplerian orbits at @xmath106 km . if on the contrary , a keplerian beat frequency model is adopted for the khz qpos , then the @xmath119 hz qpos are unlikely to arise from the mbfm .
we regard the presence of a pair of khz qpos and the constancy of their frequency difference as strong evidence in favor of the latter interpretation . in this context
, it appears natural to interpret also the hb qpos of gx 5 - 1 in terms of our precession model . for @xmath122 hz and @xmath123 hz , the maximum precession frequency ( again the one from eos
l and @xmath124 ) is @xmath125 hz . similarly in the case of gx 17 + 2 , we obtain a maximum precession frequency of @xmath107 hz , when the source is in the upper normal branch with @xmath126 hz and @xmath127 hz , and hb qpos at @xmath38 hz are simultaneously present .
these precession frequencies are clearly a factor of 22.5 smaller than those of the hb qpos .
one possibility is that the neutron stars of z "- type lmxrbs have @xmath128 , _
i.e. _ a somewhat higher value than friedman , ipser & parker ( 1986 ) and cook , shapiro & teukolsky ( 1994 ) calculate .
alternatively the hb qpos might arise from the second harmonics of the precession frequency , due to some yet unknown reason .
we note that in the case of gx 5 - 1 the dependence of the hb qpo frequency on @xmath14 is in the same range as the dependence of @xmath129 on @xmath47 in our model . in summary , we propose that the peaks observed in the power spectra of several atoll " lmxrbs at a frequency of tens of @xmath130 ( and perhaps also the hb qpos of z "- type lmxrbs ) are due to precession of the innermost disk region , which is dominated by the lense - thirring effect .
the precession frequency depends upon the keplerian frequency at the innermost disk radius , and the neutron star spin frequency , both of which are independently determined through the application of beat frequency models to the khz qpos from these sources .
the best fit is obtained for a stiff eos and @xmath124 , a value supported also from the interpretation of the highest observed values of @xmath14 in terms of the keplerian frequency of the marginally stable orbit ( kaaret , ford & chen 1997 ; zhang , strohmayer & swank 1997 ) .
alpar , a. , & shaham , j. 1985 , nature 316 , 239 angelini , l. , stella , l. & parmar , a.n .
1989 , apj 346 , 906 bardeen , j.m . , patterson , j.a .
, 1975 , , 195 , l65 berger , m. , , 1996 , apj , 469 , l13 binney , j.j . , 1978 , , 183 , 779 cook , g.b . ,
shapiro , s.l .
& teukolsky , s. 1994 , apj , 424 , 823 finger , m. h. , wilson , r. b. & harmon , b. a. 1996 , apj , 459 , 288 ford , e. , , 1997 , apj , 475 , l123 friedman , j.l , ipser , j.r . & parker , l. 1986 , apj , 304 , 115 ghosh , p. , & lamb , f.k .
1979 , apj , 234 , 296 ghosh , p. , & lamb , f.k .
1992 , in _ x - ray binaries and recycled pulsars _ , ed . e.p.j .
van den heuvel and s.a .
rappaport ( dordrecht : kluwer ) , p.487 hasinger , g. , & van der klis , m. 1989 , a&a 225 , 79 james , r. a. 1964 , apj , 140 , 552 kaaret , p. , ford , e.g. & chen , k. 1997 , apjl , 480 , l27 kahn , f.d . & woltjer l. 1959 , apj , 130 , 705 lamb , f.k . , 1991 , in _ neutron stars : theory and observation _ , ed .
j. ventura and d. pines , ( dordrecht : kluwer ) , p.445 lamb , f.k . , shibazaki , n. , alpar , a. , & shaham , j. 1985 , nature 317 , 681 lense , j. , & thirring , h. 1918 , physik . z. 19 , 156 maloney , p.r . , begelman , m , c . ,
pringle , j.e .
, 1996 , , 472 , 582 miller , m.c . , lamb , f.k . , & psaltis , d. 1997 , apj , submitted papaloizou , j.c.b .
, larwood , j.d . , nelson , r.p .
, terquem , c. , 1997 , to appear in
proceedings of the eara workshop on accretion disks " , lecture notes in physics .
pringle , j.e . , 1992 , , 258 , 811 pringle , j.e . , 1996 , , 281 , 357 smale , a.p .
, zhang , w. , white , n.e . , 1997 ,
apjl , 483 , l119 smith , d.a . , morgan , e.h . &
bradt , h. 1997 , apj , 479 , l137 strohmayer , t. , , 1996a , apj 469 , l9 strohmayer , t. , lee , u. , & jahoda , k. 1996b , iau circ .
6484 van der klis , m. 1995 , in _ x - ray binaries _ , eds .
w. h. g. lewin , j. van paradijs & e. p. j. van den heuvel ( cambridge university press ) , p. 252
van der klis , m. , , 1996a , iau circ .
6511 van der klis , m. , , 1996b , apj , 469 , l1 van der klis , m. , , 1997a , iau circ .
6565 van der klis , m. , wijnands , r.a.d . ,
horne , k. & chen , w. 1997b , apjl , 481 , l97 white , n.e . &
zhang , w. 1997 , apjl , submitted wijnands , r.a.d . , , 1996 ,
6447 wijnands , r.a.d . , , 1997 , apjl , 479 , l141 wijnands , r.a.d . & van der klis , m. 1997 , apjl , 482 , l65 wilkins , d.c .
1972 , phys .
d 5 , 814 zhang , w. , lapidus , i. , white , n.e . , & titarchuk , l. 1996 , apj , 469 , l17 zhang , w. , strohmayer , t.e . & swank , j.h .
1997 , apjl , 482 , l167 | we show in this letter that relativistic dragging of inertial frames around fast rotating collapsed stars is substantial and can give rise to observable effects .
we apply this to the khz quasi periodic oscillations ( qpos ) sources , low mass x - ray binaries ( lmxrbs ) containing an accreting neutron star . within the beat frequency model , both the keplerian frequency of the innermost region of the accretion disk ( @xmath0 khz ) and the neutron star spin frequency ( @xmath1 khz )
are directly observed . from these the lense - thirring precession frequency ( tens of hz ) of the same material in the innermost disk regions which gives rise to the khz qpos is determined within a factor of @xmath2 , depending on the neutron star equation of state . the classical contribution from neutron star oblateness decreases the precession frequency slightly .
the broad peaks at frequencies @xmath3 hz in the power spectra of the atoll "- sources 4u 1728 - 34 , 4u 0614 + 091 and ks 1731 - 260 and their variations with the higher khz qpo frequency are well matched by lense - thirring precession of material in the innermost disk region .
we also suggest that the @xmath4 hz horizontal branch qpos of gx 5 - 1 and gx 17 + 2 ( and likewise other z "- type low mass x - ray binaries ) arise from the same mechanism . -1.2 cm 24.0 cm 17.cm -.5 cm |
among the deepest challenges in glass physics is understanding the temperature dependence of the stokes - einstein ratio in the neighborhood of the glass temperature @xmath0 @xcite .
an apparently related phenomenon is the stretched - exponential relaxation of density fluctuations and other correlations .
both phenomena have been cited as evidence for dynamic heterogeneities @xcite , which supposedly provide rapid diffusion paths relevant to the first case , and a variety of localized , relaxation environments in the second @xcite .
here , i offer a counterproposal that these phenomena can be explained more systematically by the multi - species shear - transformation - zone ( stz ) theory that bouchbinder and i @xcite invoked to understand the broad range of time scales observed in viscoelastic experiments .
a short account of this proposal has appeared in @xcite .
the present version of the stz theory is an extension of the flow - defect theories of turnbull , cohen , spaepen , argon and others @xcite in which localized clusters of molecules undergo irreversible rearrangements in response to applied shear stresses . in more recent publications @xcite , my coworkers and
i have argued that these flow defects , i.e. the stz s , must be dynamic entities .
they are thermally activated , structural fluctuations that appear and disappear on time scales of the order of the @xmath1 relaxation time @xmath2 . in the presence of a shear stress ,
they are the places where irreversible molecular rearrangements occur .
the stz s also may be associated with self diffusion of a tagged molecule . at most times ,
such a molecule remains in its cage , surrounded by its neighbors .
it participates in a rearrangement involving these neighbors only when a thermally activated structural fluctuation a shear transformation
occurs at its location . when that happens , the tagged molecule jumps some distance , perhaps an intermolecular spacing @xmath3 , in some direction , while one or more of its neighbors jumps in the other direction .
a sequence of such jumps becomes a diffusive trajectory . in almost all of the earlier stz literature , for example in @xcite ,
the main focus has been on nonlinear phenomena .
we have computed the quasistationary responses of amorphous materials to strong , slow , driving forces , and have argued that the configurational states of such systems must necessarily be described by the effective temperatures of their structural degrees of freedom @xcite . here , in contrast , i focus on well equilibrated glassy systems , near their glass temperatures , subject only to infinitesimally weak perturbations .
newtonian linear viscosity , self diffusion , and time dependent relaxation of small density fluctuations as seen in scattering experiments , all fall into this category . in this linear limit
, the effective temperature is the same as the ambient temperature , and the noisy fluctuations that activate the creation and annihilation of stz s are purely thermal .
therefore , this paper and its immediate predecessors @xcite probe different aspects of glass physics than usually have been discussed in connection with stz theory . the most important way in which these papers differ from the earlier ones is in their introduction of multiple species of stz s . in @xcite ,
bouchbinder and i argued that the stz s in well equilibrated , low - temperature systems predictably occur with a broad range of internal transition rates , and that the resulting multi - species theory accurately accounts for experimentally observed , frequency - dependent , viscoelastic response functions @xcite . in sec.[stz ] of this paper , i briefly summarize the linearized equations of motion for this class of stz theories .
then , in sec.[stokes - einstein ] , i show that this multi - species theory predicts a strongly temperature dependent viscosity that , even in the absence of enhanced diffusion , is consistent with the observed stokes - einstein violations .
the extension of this theory to self - diffusion is discussed in sec.[self - diffusion ] .
there , i show that the presence of slowly responding stz s in the multi - species theory requires the existence of something like cascades of correlated hopping events in order that the theory be consistent with observations of normal fickian diffusion near the glass temperature .
i also show that the distribution of stz transition rates predicts stretched - exponential decay of density fluctuations in apparent agreement with observations .
i conclude in sec.[conclusions ] with remarks about ways in which these theoretical proposals need to be tested experimentally .
the stz theory for amorphous systems in thermodynamic equilibrium near their glass temperatures is concisely summarized by the master equation for the number densities @xmath4 of stz s oriented parallel or antiparallel to the shear stress @xmath5 : @xmath6.\end{aligned}\ ] ] here , times are measured in microscopic units , perhaps molecular vibration periods of the order of picoseconds .
the @xmath7 are the rates of forward and backward stz transitions .
the symbol @xmath8 denotes the internal transition rate , defined for this purpose to be @xmath9 . writing eq.([ndot ] ) with this explicit @xmath8 dependence implies that a range of different species of stz s , with different values of @xmath8 , will be relevant to the phenomena of interest . in the following discussion ,
i denote the probability distribution for the @xmath8 s by @xmath10 .
the second pair of terms on the right - hand side of eq.([ndot ] ) are the rates of stz creation and annihilation .
these are thermally activated processes , expressed in the form of a detailed - balance relation in which the equilibrium stz density is @xmath11 where @xmath12 is the temperature in energy units .
@xmath13 is the stz formation energy which , to a first approximation , is the energy required to create the amorphous analog of a vacancy - interstitial pair .
@xmath14 is the average volume per molecule , proportional to the cube of the average molecular spacing @xmath3 .
the attempt frequency @xmath15 is best understood as a dimensionless , thermal noise strength .
it is a super - arrhenius function whose strong @xmath16 dependence governs the equilibrium glass transition . because time is measured in molecular units , @xmath15 must be of the order of unity at large @xmath16 , where the attempt frequency is the molecular vibration frequency . as @xmath16 decreases through the glass temperature @xmath17 ,
the rate of thermally activated structural rearrangements slows dramatically , and @xmath15 falls rapidly toward zero . the structural relaxation rate , @xmath18 ,
is conventionally identified as the @xmath1 relaxation rate @xmath19 . for present purposes ,
i take @xmath15 to be an observable but not necessarily predictable quantity .
see @xcite for my best attempt to date to compute it from first principles .
all of the phenomena to be considered here are technically `` slow '' in the sense that their characteristic time scales are much longer than the times on which the @xmath4 in eq.([ndot ] ) relax to quasistationary values .
note that this was not true for the oscillatory viscoelasticity discussed in @xcite , where the broad range of driving frequencies included those that are fast as well as slow on intrinsic stz time scales . here
, we are interested only in linear steady - state viscous flow , or in diffusional relaxation over times of the order of or longer than @xmath2 ; therefore , all of the information that we need can be obtained by setting @xmath20 on the left - hand side of eq.([ndot ] ) and solving for the @xmath4 . in this quasistationary approximation , to first order in the stress @xmath5 , the steady - state plastic strain rate @xmath21 has the form @xmath22\cr \\&= & \epsilon_0\,\int_0^{\infty } d\nu\,\tilde p(\nu)\,{1\over \tau(\nu)}\,{\cal t}(s),\end{aligned}\ ] ] where @xmath23 and , to first order in @xmath5 , @xmath24 equation([gammadot ] ) comes directly from eq.([ndot ] ) with no additional assumptions .
its structure implies that @xmath25 is the event rate for stz s of species @xmath8 . as explained in appendix [ appendix ]
, @xmath25 is the frequency at which an stz is created in one of its internal states and annihilated in the other state , having made an arbitrary ( odd ) number of transitions between those states during its lifetime .
there are several aspects of these equations that need emphasis .
first , they assume that @xmath26 , the volume of the plastic core of a shear transformation , is independent of @xmath8 .
second , the definition of @xmath2 introduced earlier and again in eq.([tau ] ) , is not tied directly to the temperature dependence of the viscosity @xmath27 .
that is , @xmath27 is not always simply proportional to @xmath28 .
third , @xmath29 is the stress - induced bias between forward and backward transitions .
the final expression in eq.([calt ] ) is the classic einstein formula , deduced from the assumption that the underlying transition is a stress - enhanced , thermally activated process . in @xcite , this result was shown to follow from the second law of thermodynamics , with @xmath16 more generally being the thermodynamically defined effective temperature @xmath30 . a simple application of the preceding analysis to rapidly sheared systems , where the transverse diffusion constant is @xmath31 , confirms that @xmath30 is the same as the effective temperature determined by a fluctuation - dissipation relation , at least within the stz theory . ( see @xcite and references cited there . ) throughout the present paper
, thermodynamic equilibrium implies that @xmath32 .
in @xcite , bouchbinder and i derived a formula for the distribution @xmath10 starting with the assumption that the zero - stress transition rate has the form @xmath33 where @xmath34 is an activation energy .
@xmath35 , in analogy to @xmath36 in eq.([ndot ] ) , is an attempt frequency that , in principle , can depend on both @xmath34 and @xmath16 . continuing this analogy
, we argued that @xmath35 must be unity for small @xmath34 and large @xmath16 ( fast transitions ) , but becomes small in the opposite limit . because @xmath34 is measured _ downward
_ from some reference energy , its probability distribution has the form @xmath37 where @xmath38 is a positive , @xmath16-independent constant that is less than unity .
then , for small @xmath34 ( large @xmath8 ) , @xmath39 this distribution diverges at small @xmath8 and must be cut off there . to choose that cutoff , say at @xmath40
, we argued that the internal transition rates can not be slower than the rate of spontaneous structural rearrangements .
therefore , we chose @xmath41 , which , according to eq.([tau ] ) , is several orders of magnitude smaller than @xmath36 for systems near their glass temperatures .
the corresponding distribution over barrier heights , @xmath42 , as given in eq.([pdelta ] ) with a sharp cutoff at large @xmath34 , is roughly consistent with that found experimentally by argon and kuo @xcite .
using these constraints , bouchbinder and i wrote a three - parameter @xmath8 distribution in the form @xmath43},\ ] ] where @xmath44 is a normalization constant .
we used @xmath45 in accord with viscoelastic data for various materials near their glass temperatures @xcite .
the exponent @xmath46 , which determines the sharpness of the cutoff at @xmath47 , is not sharply determined by experiment . for most purposes ,
we have used @xmath48 .
there was no need to specify the quantity @xmath35 in eq.([nudef0 ] ) ; its principal role is to provide a physical mechanism for producing very small values of @xmath8 without requiring unphysically large values of @xmath34 .
these first - principles estimates of @xmath10 and @xmath47 , with no additional fitting parameters , are accurately confirmed by the experimental data .
a central feature of the present analysis is that , as implied by eq.([tau ] ) and discussed in the appendix , there are two different limiting populations of stz s at low temperatures .
there are `` fast '' stz s , with @xmath49 , for which @xmath50 .
these stz s may make multiple , back and forth transitions between their two states during their lifetimes .
there is also a substantial population of `` slow '' stz s with @xmath51 and @xmath52 , which make at most one transition before disappearing .
both populations are important for determining the inverse viscosity : @xmath53 equation ( [ eta ] ) makes it clear that the distribution @xmath10 plays a crucial role in determining the viscosity . at high temperatures ,
@xmath47 is large , of the order of unity , and @xmath10 must be peaked at a comparably large value of @xmath8 . as a result , eq.([eta ] ) becomes a conventional formula in which @xmath54 . on the other hand , at temperatures near or below @xmath55 , the integration in eq.([eta ] ) is dominated by the slow stz s with @xmath56 , and the conventional formula for @xmath27 is enhanced by a factor that can be shown to be roughly proportional to @xmath57 .
suppose , for the moment , that diffusion in this system is entirely normal , with a diffusion constant given simply by @xmath58 .
then eq.([eta ] ) predicts that the stokes - einstein ratio is @xmath59^{-1 } \equiv c_{se}(\theta),\ ] ] where @xmath60 . since @xmath2 has cancelled out of eq.([se ] ) , and @xmath10 is a function only of @xmath61 ( apart from a cutoff at large @xmath8 when necessary ) , the only temperature dependence of @xmath62 occurs via the ratio @xmath63 . in @xcite , we estimated that @xmath64 .
that estimate has been used in evaluating eq.([se ] ) to plot the graphs of @xmath65 as functions of @xmath66 shown in fig .
[ cse(t ) ] . the balance between the contributions of slow and fast stz s in eq.([se ] )
is determined largely by the exponent @xmath38 in eq.([pnu ] ) , thus @xmath65 is shown in the figure for five different values of @xmath38 .
this theory does not ( yet ) have enough microscopic physical content to make an accurate connection between low and high - temperature behaviors , or even to make material - specific predictions of the shape of @xmath10 or the length scale @xmath67 .
if , as argued in the preceding paragraph , @xmath10 becomes peaked at large values of @xmath8 for high temperatures , then @xmath68 in that limit .
thus , the values of @xmath69 in fig .
[ cse(t ) ] imply stokes - einstein violations of roughly the magnitude seen experimentally , arising entirely from the anomalously large viscosity predicted by eq.([eta ] ) .
turn now to the statistical problem of self diffusion . here , instead of averaging over probabilities of single events as above , we must consider sequences of multiple hopping events .
the statistical weight assigned to any such sequence depends on how far the molecule moves and how long it takes to get there .
technically , the molecule executes a `` continuous - time random walk '' ( ctrw ) @xcite .
note that the model of stz - enabled diffusion steps contains no information about dynamics on sub - molecular space or time scales . in effect , it is coarse grained on the scale of the molecular spacing , and it `` knows '' nothing about fast vibrational motions other than that they carry the thermal noise that activates barrier - crossing transitions .
therefore , this model can not describe @xmath70 relaxation or any details of how a molecule moves within the cage formed by its neighbors , or of how it escapes from that cage .
the basic ingredient of the ctrw analysis is the one - step probability that , after waiting a time @xmath71 , an stz fluctuates into existence at the position of the tagged molecule , and the molecule jumps a distance @xmath72 .
( there is no loss of generality in projecting the diffusion mechanism onto one dimension . ) for an stz with an internal rate @xmath8 , this normalized one - step probability distribution is @xmath73 where @xmath74 is the probability of a jump of length @xmath72 . as indicated , @xmath74 may depend on @xmath8 . the crucial assumption here
is that the time @xmath75 appearing in this formula is the same as the quantity defined in eq.([tau ] ) . in both cases
, @xmath25 is the average rate at which stz s appear , make one net internal transformation ( perhaps after multiple back - and - forth transformations ) , and then disappear .
this elementary fluctuation mechanism should be common to both plastic deformation and self diffusion .
the probability of any random walk consisting of a sequence of such steps is a product of space - time convolutions of factors @xmath76 .
if the stz events are uncorrelated a major assumption then each factor can be averaged independently over @xmath8 .
the walk ends after a final time interval @xmath71 within which the molecule does not make a further jump ; thus the complete path probability contains one factor @xmath77 again included in the convolution integrals and averaged over @xmath8 .
these space and time convolutions are converted into products by computing fourier and laplace transforms , in terms of which the multi - step walk probabilities can be summed to all orders . the self - intermediate scattering function @xmath78 , i.e. the @xmath79th fourier component of the diffusion profile , is given by an inverse laplace transform : @xmath80 where @xmath81 is the laplace transform of the averaged @xmath82 , and @xmath83 is the analogous transformation of @xmath76 .
a crucial ingredient in these formulas is the jump - length distribution @xmath84 .
if , according to the discussion following eq.([calt ] ) , each stz event moves the tagged molecule a mean - square distance @xmath85 , then it would be natural to choose a gaussian distribution @xmath86 independent of @xmath8 . with this choice , however , the slow stz s play the role of deep traps @xcite , effectively shutting down long - time diffusion . to see this ,
compute the mean - square displacement @xmath87 , which , for a @xmath8-independent @xmath88 , becomes @xmath89_{k=0}= a^2\,\int { dw\over 2\,\pi\,i}\ , { e^{w\,t/\tau_{\alpha}}\over w^2\,\tilde k(w)}.\ ] ] if @xmath10 cuts off sharply at small @xmath8 , then @xmath90 is analytic at @xmath91 , and the long - time behavior is obtained by integrating around the pole there .
the result is that @xmath92 , with @xmath93 .
therefore , ordinary diffusion is suppressed by a factor @xmath94 . for @xmath95 in eq.([pnu ] ) , @xmath96 diverges , and @xmath97 vanishes .
this result is manifestly inconsistent with the experimental data reported in @xcite , where fickian diffusion was observed near the glass temperature .
given the @xmath10 determined by the frequency - dependent viscoelasticity @xcite , this discrepancy between diffusion theory and experiment appears to falsify the postulate of uncorrelated short jumps within the context of the present theory .
it seems , instead , to indicate that the cascades of correlated stz `` flips '' seen in low - temperature numerical simulations ( see @xcite and earlier papers cited there ) are relevant to laboratory systems at temperatures near @xmath0 .
a systematic generalization of the ctrw analysis to include strongly correlated sequences of jumps is beyond the scope of this investigation ; but it is plausible that such sequences might appear approximately in the form of anomalously long jumps initiated by individual slow events . unlike the fast stz s ,
the slow ones are stable only in near - equilibrium glass formers at low temperatures , and they only occasionally undergo shear transformations during their lifetimes . their small values of @xmath8 imply that their internal energy barriers are high ; therefore , when they do make transitions , they release substantial amounts of energy that could trigger subsequent events .
suppose , as a first guess , that a slow event sends a tagged molecule on a random walk that lasts for a time @xmath75 , during which the jump rate is @xmath19 .
the mean - square displacement during this walk would be @xmath98 , which is equal to @xmath85 for fast stz s , but becomes substantially larger for slow ones .
the corresponding jump - length distribution is @xmath99 this choice of @xmath84 , if correct , would solve the problem of overly slow diffusion . when eq.([fhat ] ) is used in eq.([fhat - w ] ) , the quantity @xmath100 in the denominator of eq.([zsquared ] ) cancels out , and @xmath101 at all times .
this means that diffusion in this approximation is normal in the sense that @xmath58 , and that the assumption leading to the expression for the stokes - einstein ratio in eq.([se ] ) would be correct .
finally , consider the self - intermediate scattering function @xmath78 given by eqs.([fhat - w]-[jdef ] ) . to evaluate @xmath78 numerically ,
close the contour of integration in eq.([fhat - w ] ) around the branch cut on the negative real @xmath102 axis .
some results are shown in fig.[ser-1 ] .
the parameters correspond roughly to a metallic glass near @xmath0 : @xmath103 , @xmath104 , and @xmath48 .
the wavenumbers are @xmath105 , from left to right .
all four curves exhibit stretched - exponential relaxation of the form @xmath106 $ ] , with @xmath107 , in the same order of decreasing @xmath79 .
several features of these results are notable .
first , there is a fickian limit at small @xmath79 , where @xmath108 approaches unity , and @xmath109 as shown by the dashed line in fig.[ser-1 ] for @xmath110 .
second , there is a non - fickian limit at large @xmath79 , where the curves begin to lie on top of each other , and @xmath108 approaches @xmath111 .
this limiting behavior occurs because any realistic choice of the factor @xmath84 vanishes at large @xmath79 , leaving the initial , @xmath79-independent function @xmath100 in the numerator of eq.([fhat - w ] ) as the only contribution to the relaxation function .
note that @xmath112 has the form of a local relaxation function averaged over a distribution of environments with different relaxation times @xmath75 .
the transition between large-@xmath79 and small-@xmath79 behavior is temperature dependent _ via _ the quantity @xmath113 in eq.([fhat - w ] ) , which produces a @xmath16 dependence here in much the way it did for the viscosity in eq.([eta ] ) . at fixed @xmath79 , according to these equations , the system changes from stretched - exponential to fickian as the temperature increases .
it is not a coincidence that , in the large-@xmath79 limit given in eq.([kdef2 ] ) and at low temperatures , @xmath114 ; but @xmath115 is not even an exact result for all choices of @xmath38 and @xmath46 , nor is it exact in the limit of long times .
in summary , the multi - species stz theory , with the transition - rate distribution @xmath10 derived directly from statistical principles , and with essentially no arbitrary parameters , has been convincingly successful in predicting frequency - dependent viscoelastic responses of glassy materials . as argued here in sec.[stokes - einstein ] , this theory predicts that the steady - state , linear viscosity is strongly temperature dependent in the neighborhood of the glass temperature .
that prediction , in turn , suggests that observed anomalies in the temperature dependence of the stokes - einstein ratio are due to large viscosities , and not necessarily to the emergence of rapid diffusion paths at low temperatures .
the multi - species theory also predicts stretched - exponential decay of density fluctuations , and further predicts temperature and wave - number dependences of the stretched - exponential function that appear to agree with observations .
the outstanding uncertainty in this theoretical picture is that , in its simplest version , it predicts that the slow stz s act like traps and shut down fickian diffusion at low temperatures , in contradiction to experimental measurements @xcite . to resolve this discrepancy ,
i have suggested in sec.[self - diffusion ] that the correlated cascades of stz - like events observed in low - temperature molecular - dynamics simulations @xcite need somehow to be incorporated into the diffusion analysis .
this is done here , in an admittedly _ ad hoc _ way , by choosing a @xmath8-dependent jump size distribution @xmath84 in eq.([fhat ] ) .
it is not clear whether eq.([fhat ] ) is a realistic approximation , or whether a more accurate description of correlated diffusion steps might modify the predicted stokes - einstein ratio .
the analysis presented here calls for a unified investigation of stokes - einstein violations , stretched - exponential relaxation , and frequency dependent viscoelastic response functions , all in comparable glass - forming materials near their glass temperatures . checking the consistency of the various formulas shown here would test the theory as a whole and its interpretation of diverse dynamic phenomena . for example , it ought to be possible to measure viscoelastic responses ( or internal - friction functions ) using the same materials , under the same conditions , that are used for measuring @xmath78 in the large-@xmath79 limit shown in eq.([kdef2 ] ) .
then , it should be possible either to deduce consistent approximations for @xmath10 , or else to falsify the predicted consistency .
similarly , it would be useful to make the latter kinds of measurements for the same systems in which fickian diffusion is observed , and thus to test the approximation for @xmath84 in eq.([fhat ] ) .
detailed studies of the @xmath79 and @xmath16 dependences of @xmath78 , again on the same systems for which other measurements are made , might further resolve the outstanding issues .
to interpret the event rate @xmath25 defined in eq.([tau ] ) , it is useful to perform a continuous - time , random - walk analysis similar to that used for diffusion in sec.[self - diffusion ] , but now restricted to transitions back and forth between the internal @xmath116 states of a single stz .
consider an elementary process in which an stz is created in a @xmath117 state and then , after a sequence of back and forth transitions , is annihilated in a @xmath116 state .
the stz has made a single forward or backward step during its lifetime .
the rates for these transitions are contained in the equations of motion , eq.([ndot ] ) , for the densities of @xmath116 stz s , @xmath4 . according to these equations
, @xmath36 is the rate at which stz s ( of either sign ) are annihilated by thermal fluctuations , and @xmath7 is the rate at which @xmath117 stz s make transitions to @xmath116 stz s .
now , use these rate factors to compute the probability of a @xmath117 stz either being annihilated or making a transition to its @xmath116 state at time @xmath118 , if it formed at @xmath119 .
this probability is the normalized waiting - time distribution @xmath120 \,e^{-\bigl[\rho + r(\pm s)\bigr]\,t}.\ ] ] the probability that this stz has _ not _ disappeared or changed its state at time @xmath71 is @xmath121\,t}.\ ] ] now compute the probability per unit time for a sequence of such transitions .
suppose , for example , that the stz starts in its @xmath122 state and makes one transition to its @xmath123 state before being annihilated at time @xmath124 .
the probability per unit time for this particular sequence of transitions is @xmath125 where @xmath126 is the formation rate . for any such sequence of transitions ,
the rate is a convolution of functions @xmath127 , multiplied by factors @xmath7 , and integrated over intermediate times @xmath128 , etc .. for simplicity , assume that @xmath5 is not a function of time .
then we can perform all of these integrations by using the laplace transforms @xmath129 so that the laplace transform of the rate for any sequence of transitions is just a product of factors @xmath130 and @xmath7 .
the sum of all sequences in which the stz flips forward and backward , any number of times , but always returns to its initial orientation , is @xmath131^{-1}.\ ] ] thus , the laplace transform for the elementary @xmath132 or @xmath133 process is @xmath134 to compute the jump rate relevant to stress - free , stz - induced diffusion , set @xmath135 in this formula and integrate over @xmath124 by setting @xmath136 .
the result is @xmath137 where , as usual , @xmath138 .
denote the total event rate for a given @xmath8 by @xmath139 which recovers eq.([tau ] ) in the context of diffusion .
note what is happening here .
if @xmath49 , then the stz makes multiple back and forth transitions during its lifetime , and @xmath140 . at the opposite extreme ,
where the internal transitions are very slow so that @xmath51 , then @xmath141 . in other words
, if the diffusing particle encounters a very slow stz , it most likely will not make a jump at all .
the stress - driven deformation rate , say @xmath142 , is the difference between the contributions from sequences with one net forward transition , starting from a @xmath122 state , and sequences with one net backward transition , starting from a @xmath123 state : @xmath143 where @xmath144.\ ] ] in the limit of small stress @xmath5 , this result recovers eq.([gammadot ] ) and the viscosity calculation leading to eq.([eta ] ) .
i thank eran bouchbinder , michael cates , mark ediger , and takeshi egami for helpful discussions during the course of this investigation .
this work was supported in part by the division of materials science and engineering , office of basic energy sciences , department of energy , de - ac05 - 00or-22725 , through a subcontract from oak ridge national laboratory . | the shear - transformation - zone ( stz ) theory has been remarkably successful in accounting for broadly peaked , frequency - dependent , viscoelastic responses of amorphous systems near their glass temperatures @xmath0 .
this success is based on the theory s first - principles prediction of a wide range of internal stz transition rates .
here , i show that the stz rate - distribution causes the newtonian viscosity to be strongly temperature dependent ; and i propose that it is this temperature dependence , rather than any heterogeneity - induced enhancement of diffusion , that is responsible for stokes - einstein violations near @xmath0 .
i also show that stretched - exponential relaxation of density fluctuations emerges naturally from the same distribution of stz transition rates that predicts the viscoelastic behavior . to be consistent with observations of fickian diffusion near @xmath0 , however , an stz - based diffusion theory somehow must include the cascades of correlated displacement events that are seen in low - temperature numerical simulations . |
accurate determinations of metal abundances in planetary nebulae ( pne ) are important for understand the chemical evolution of galaxies and the nucleosynthesis and mixing processes in low- and intermediate - mass stars . a long - standing problem in nebular abundance studies
has been that the heavy - element abundances derived from optical recombination lines ( orls ) are systematically higher than those derived from collisionally excited lines ( cels ; see liu 2001 , 2003 for recent reviews ) . in extreme cases ,
the discrepancies exceed a factor of 10 . given that cel abundances are sensitive to electron temperature and density , knowledge of temperature and density variations is an essential ingredient in understanding the abundance discrepancy problem .
information on temperature and density variations can be obtained by comparing results derived from plasma diagnostics which behave differently as temperature and density vary .
traditionally , nebular electron temperatures and densities are derived from cel ratios .
however , for pne including high - density regions ( @xmath12e@xmath13@xmath14 ) , such as ic 4997 @xcite , mz 3 @xcite and m 2 - 24 @xcite , cels are strongly suppressed by collisional de - excitation , and so depending on their critical densities , they may become unusable in probing such high - density regions . on the other hand ,
hydrogen recombination lines have very high critical densities ( @xmath15@xmath14 ) .
consequently , the hydrogen recombination spectrum provides a powerful tool to probe high - density regions .
comparison of electron temperatures and densities derived from the hydrogen recombination spectrum ( lines plus continuum ) with those derived from cels should enable us to quantify temperature and density variations in nebulae . by measuring the balmer discontinuity , @xcite determined electron temperatures in 14 pne and found that they are systematically lower than those derived from the [ ] nebular - to - auroral forbidden line ratio .
this can be explained by the presence of large temperature fluctuations since the [ ] @xmath16 auroral line of higher excitation energy tends to be emitted in higher temperature regions than the [ ] @xmath17 nebular lines of lower excitation energy , as suggested by @xcite .
recent reviews on temperature fluctuations of gaseous nebulae are given by @xcite and references therein .
however , no photoionization models yield such large temperature fluctuations suggested by observations . moreover , temperature fluctuations are unable to explain the large disparity between the orl and cel abundances observed in some pne @xcite . instead
, it is found that the hypothesis that pne may contain clumps of different temperature , density and chemical composition , embedded in diffuse ` normal ' material , can explain many of the observed patterns @xcite .
peimbert s concept of temperature fluctuations alone may not be a realistic description of the physical conditions in a nebula , although it has proven to be a convenient and useful way to characterize the complexities of real nebulae .
it is expected that there are also variations of electron density inside a pn .
@xcite pointed out that electron densities derived from various plasma diagnostics should correlate positively with the critical densities of the diagnostic lines involved if there are large density variations within a nebula .
@xcite found that electron densities derived from the [ ] @xmath9 ratios are generally lower than those derived from the optical [ ] and [ ] doublet ratios , suggesting the presence of high density regions in pne , where the 52- and 88-@xmath18 lines are suppressed by collisional de - excitation due to their relatively low critical densities .
@xcite showed that the discrepancies between @xmath0([]@xmath3 ) and @xmath0(bal ) can be explained by the presence of condensations in nebulae and gave an analytical method of estimating the filling factor of condensations .
our detailed studies of the peculiar pne mz 3 and m 2 - 24 suggest that the presence of high - density regions in some pne can significantly affect abundance determinations @xcite .
however , density inhomogeneities alone can not explain the large disparity between the orl and cel abundances either , unless the condensations are also h - deficient @xcite .
@xcite found that the discrepancy between the orl and cel abundances correlates with the difference between the temperatures derived from the [ ] collisional lines and from the balmer jump of hydrogen recombination spectrum , thus lending strong support to the argument that some extremely cold h - deficient knots may co - exist with normal nebular material in many pne .
however , it is difficult to explain the origin and evolution of such h - deficient knots with the current theories of element production in intermediate - mass stars . in this paper
, we present determinations of electron temperatures and densities for a large sample of pne based on the analysis of nebular hydrogen recombination spectrum .
section 2 describes the sample . in section 3 , we describe a new approach to deriving electron temperature and density simultaneously from the observed nebular hydrogen recombination spectrum ( lines plus continuum ) and present the results . in section 4 , we compare results derived from a variety of temperature and density indicators and discuss the significance of variations in electron temperature and density in pne .
a discussion of he / h abundance determinations is also presented in this section . in section 5 , we present electron temperatures and densities derived from the paschen discontinuity and decrements .
a summary then follows in section 6 .
48 galactic pne ( 37 disc pne and 11 bulge pne ) are analyzed in the current work .
the spectra were obtained in a series of observing runs using telescopes at eso la silla and la palma .
14 objects were observed using the isis double long - slit spectrograph mounted on the 4.2 m william herschel telescope ( wht ) at la palma observatory in 1996 july august and in 1997 august .
a tek @xmath19 @xmath20 charge coupled device ( ccd ) was used , which yielded a spatial sampling of 0.3576 arcsec per pixel projected on the sky . in 1996 ,
a 600gmm@xmath21 and a 316gmm@xmath21 grating were used for the blue and red arm , respectively . in 1997 , they were replaced by a 1200gmm@xmath21 grating .
23 objects were observed with the eso 1.52 m telescope using the boller & chivens ( b&c ) long - slit spectrograph .
the spectra were obtained during runs in 1995 july , 1996 july , 1997 february and 2001 june . in 1995
the detector was a ford @xmath22 @xmath23 chip , which was superseded in 1996 by a thinned uv - enhanced loral @xmath22 @xmath23 chip of much improved quantum efficiency .
the b&c spectrograph has a useful slit length of about 3.5arcmin . in order to reduce the ccd read - out noise ,
the ccd was binned by a factor of two along the slit direction , yielding a spatial sampling of 1.63 arcsec per pixel projected on the sky .
a slit width of 2arcsec was used throughout the eso 1.52 m runs .
11 objects were observed using the intermediate dispersion spectrograph ( ids ) on the 2.5 m isaac newton telescope ( int ) at the observatorio del roque de los muchachos , on la palma , spain in 2001 august . a slit 4arcmin long and 1arcsec wide
was used for nebular observations .
the r1200b grating was used to cover wavelengths from 35005000 at a spectral resolution of 0.95 , determined from the _ fwhm _ of the calibration arc lines , while the r300v grating , together with a gg385 order sorting filter , was used to cover wavelengths from 38007700 at a resolution of 4.0 _
fwhm_. all the spectra were reduced using midas or iraf following the standard procedure .
the spectra were bias - subtracted , flat - fielded and cosmic - rays removed , and then wavelength calibrated using exposures of a calibration arc lamp .
the spectra were flux - calibrated by observing spectrophotometric standard stars .
1-d spectra were obtained by integrating along the slit . using the galactic reddening law of @xcite , the logarithmic extinction at h@xmath24 , @xmath25 , was derived from the observed balmer decrements , the h@xmath26/h@xmath24 , h@xmath27/h@xmath24 and h@xmath28/h@xmath24 ratios .
a mean value for each nebula was used to deredden the spectra .
the object names and the extinction constants are given in table [ tbl-1 ] .
amongst the pne in our sample , far - infrared spectra for 21 of them have been obtained using the long wavelength spectrometer ( lws ) on board the infrared space observatory ( _ iso _ ) .
fluxes of far - ir fine structure lines derived from these spectrum have been published by @xcite except for three objects , hu 1 - 2 , ngc 6210 and ngc 6818 . for the latter three pne
, we measured far - ir line fluxes using spectra retrieved from the _ iso _ data archive . in order to normalize the @xmath29 line
fluxes to h@xmath24 , the absolute @xmath30 fluxes tabulated in @xcite were used .
in order to calculate theoretical nebular spectra , accurate determinations of the he@xmath31/h@xmath31 and he@xmath32/h@xmath31 abundance ratios are necessary ( see section 3 ) .
helium ionic abundances were derived from and recombination lines , using @xmath33 which have only a weak dependence on the adopted electron temperature and density .
he@xmath31/h@xmath31 abundance ratios were calculated from the strong line @xmath34 under the assumption of case a. the effective recombination coefficient @xmath35 is derived from the formulae presented by @xcite .
he@xmath32/h@xmath31 abundance ratios were calculated from the @xmath36 line , using the effective recombination coefficient @xmath37 of @xcite .
the results are presented in table [ tbl-1 ] . for the purpose of comparison ,
various forbidden line ratios are also used to determine electron temperatures and densities . for densities ,
results derived from the following diagnostic ratios were used , [ ] @xmath38m/@xmath39 m , [ ] @xmath5 , [ ] @xmath40 , [ ] @xmath41 and [ ] @xmath42 .
electron densities from the [ ] 52-@xmath43m/88-@xmath43 m ratio are taken from @xcite except for ngc 6210 and ngc 6818 , for which we obtain the values of @xmath44 and 2.86 ( @xmath14 ) , respectively , using our own measured line fluxes .
densities deduced from the optical [ ] @xmath5 , [ ] @xmath40 , [ ] @xmath41 and [ ] @xmath42 doublet ratios are taken from wang et al .
( in preparation ) . a detailed discussion of these density indicators is given in their paper .
electron temperatures derived from the [ ] @xmath2 and [ ] @xmath1 ratios together with their uncertainties are presented in table [ tbl-1 ] . for the former ,
values were derived assuming electron densities derived from the [ ] 52-@xmath43m/88-@xmath43 m line ratio . for the latter ,
the average densities derived from the [ ] @xmath41 and [ ] @xmath42 doublet line ratios were adopted , given that cl@xmath31 and ar@xmath32 have ionization potentials of 23.8 ev and 40.7 ev , respectively , comparable to the value of 35.1 ev for o@xmath31 , and therefore anticipating that the [ ] lines arise from similar ionized regions as [ ] and [ ] lines . as the density increases , the resultant temperatures become increasingly sensitive to the adopted density . when the density reaches a value higher than @xmath45 @xmath14 , the [ ] @xmath2 and [ ] @xmath1 ratios are no longer useful temperature diagnostics .
for example , high density regions have been detected in m 2 - 24 and ic 4997 @xcite .
for both pne , temperatures deduced from the [ ] @xmath1 ratio have large uncertainties .
the recombination spectrum consists of continuum and line emission .
the continuous spectrum has a jump at the limit of the balmer series at 3646 .
high - order balmer lines ( @xmath46 , @xmath47 ) converge at the balmer discontinuity and merge with the continuum jump at shorter wavelength .
the size of the balmer jump is a function of electron temperature . on the other hand ,
intensities of high - order balmer lines ( @xmath46 , @xmath48 ) relative to h@xmath24 are sensitive to electron density .
therefore , the hydrogen recombination spectrum near the balmer jump region provides valuable temperature and density diagnostics .
our method and results of the electron temperature and density determinations based on this technique are presented below .
several processes contribute to the continuum emission from a nebula : free - free and free - bound emission and the two - photon decay .
apart from hydrogen , the second most abundant element , helium also contributes significantly to the first two processes .
in addition , emission from the central star and local dust can in principle contaminate the observed continuum spectrum . in the following
, a detailed calculation including all the processes is presented .
the free - free ( or bremsstrahlung ) emission , which is emitted when an free electron interacts but does not combine with a positively charged ion such as h@xmath31 , he@xmath31 and he@xmath32 , dominates the nebular continuum in the infrared and radio region .
the emission coefficient of this process is given by @xcite , @xmath49,\ ] ] where @xmath50 and @xmath51 are ion and electron densities respectively , @xmath52 is the nuclear charge , @xmath53 is the electron temperature , @xmath54 denotes the rydberg energy and @xmath55 is the free - free gaunt factor . here
@xmath55 is computed using the method outlined by @xcite . in the optical region ,
free - bound emission is the main contributor to the continuum . according to @xcite ,
the emission coefficient of continuous emission due to recombinations of h@xmath31 , he@xmath31 or he@xmath32 with electrons is @xmath56,\ ] ] where @xmath57 is the ionization potential of the ( @xmath58,@xmath59 ) state , @xmath60 is the lowest state that can contribute to the free - bound emission at the given frequency @xmath61 and @xmath62 is the photoionization cross - section for the state , which is computed using the method described by @xcite .
another important source of continuous emission is two - photon decay of the 2@xmath63 level of the hydrogen atom ( two - photon emission from he@xmath31 and he@xmath64 is negligible for nebulae of cosmic compositions ; brown @xmath65 mathews , 1970 ) .
single photon transition from the 2@xmath63 level to the ground 1@xmath63 is forbidden .
radiative decay of the 2@xmath63 level to the ground can only occur via two photon emission , in which the electron jumps to a virtual level , which can lie in anywhere between @xmath66 and @xmath67 levels , and then to the ground state , emitting two photons in the process .
as a result , continuum emission longward of ly@xmath26 will occur .
the emission coefficient of the two - photon process is given by @xmath68,\ ] ] where @xmath69 and @xmath70 is the probability per second for emitting a photon in the interval d@xmath71 . approximately , the function @xmath72 is given by the following expression @xmath73^{0.8}\}+0.88[y(1-y)]^{1.53}[4y(1-y)]^{0.8}\ } { \rm s}^{-1}\ ] ] @xcite .
the transition probability for the two - photon decay is @xmath74 s@xmath21 .
the equilibrium population of the 2@xmath63 state is given by @xmath75 @xcite , where @xmath76 is the proton density , @xmath77 is the effective recombination coefficient for populating 2@xmath63 , and @xmath78 and @xmath79 are the collisional de - excitation transition rates of 2@xmath63 to 2@xmath80 states by electrons and protons , respectively .
note that the collisional transition from 2@xmath80 to 2@xmath63 has been ignored in equation ( 6 ) . under case a conditions , this is a good approximation .
under case b conditions , however , the population of the 2@xmath80 state is comparable to that of the metastable 2@xmath63 level since ly@xmath26 is trapped between the 2@xmath80 and 1@xmath63 states .
consequently , there is considerable collisional excitation from 2@xmath80 to 2@xmath63 , which enhances the two - photon emission .
therefore , the emission coefficient of two - photon process given by equation ( 4 ) can be significantly underestimated .
however , this hardly affects our resultant electron temperatures and densities since an additional continuum ( see below ) , is added to fit our observed spectrum in order to simulate direct and scattered stellar light . for a limited wavelength region , such an approach is sufficient to account for any excess emission from the two - photon process .
finally , direct or scattered light from the central star can contaminate the observed continuum level . for our observations ,
the slit position was normally chosen to sample the brightest parts of the nebulae , sometimes passing through the central star .
some spectra were derived by scanning across the nebular surface .
it is quite difficult to evaluate the level and spectral energy distribution of the contaminating stellar continuum .
as central stars of pne are relatively hot ( @xmath8120,000k ) , in our calculation , a rayleigh - jeans approximation for a black body ( @xmath82 ) is adopted to simulate the stellar continuum .
this stellar component also partially compensates for the underestimated two - photon continuum , as mentioned above . for simplicity
, we shall call this term ` stellar continuum ' hereafter .
continuum emission from dust grains peaks in the infrared and is negligible in the optical region .
thus it has been neglected in our calculation .
apart from continuum emission , recombination line emission from bound - bound transitions of h@xmath64 , he@xmath64 and he@xmath31 is also included .
the emission coefficient of a recombination line is given by @xmath83,\ ] ] where @xmath84 is the effective recombination coefficient of the line .
for and recombination lines , emission coefficients are determined as in @xcite . here
the considered maximum principle quantum number of is @xmath85 . for recombination lines ,
emission coefficients are taken from @xcite and @xcite .
all line profiles are assumed gaussian with a _ fwhm _
@xmath86 where @xmath87 is the _ fwhm _ caused by thermal broadening , given by @xmath88 where @xmath89 is the mass of the atom , and @xmath90 is the _ fwhm _ caused by other broadening effects , such as nebular expansion , turbulence , seeing conditions and instrument . for a typical
pn observed here , @xmath91 .
line emission coefficients are thus converted into emission coefficients per unit wavelength by @xmath92.\ ] ] just longward of the balmer jump , high - order balmer lines converge and merge into a continuum - like feature .
the slope of the feature is sensitive to electron density , and thus provide an excellent density diagnostic .
the region is contaminated by the @xmath933664 line of multiplet v 1 .
however , the weak recombination line has a relatively narrow line width compared to the profile of the continuum - like feature and consequently , can not cause any difficulty in our analysis .
summing up contributions from all the processes discussed above , the flux at a given wavelength is given by @xmath94~[{\rm erg~cm}^{-3}~{\rm s}^{-1}~{\rm \aa}^{-1}].\ ] ] equation ( 11 ) shows that , in order to synthesize the nebular spectrum near the balmer jump wavelength region , the required input parameters include nebular electron temperature and density , the magnitude of contaminating stellar continuum , he@xmath31/h@xmath31 and he@xmath32/h@xmath31 ionic abundance ratios and @xmath95 . in general , decreasing ( increasing ) of the nebular electron temperature will steepen ( flatten ) the general shape of the theoretical continuum and increase ( decrease ) the balmer discontinuities ; decreasing ( increasing ) of electron density will steepen ( flatten ) the continuous feature redwards of 3646 and decrease ( increase ) the intensities of high - order balmer lines relative to a low - order line , such as h@xmath24 , as shown in fig . [ trend ] .
accordingly , matching the dereddened observed spectrum to those calculated from equation ( 11 ) by varying the electron temperature and density , we can determine the electron temperature and density simultaneously provided other parameters , such as he@xmath31/h@xmath31 and he@xmath32/h@xmath31 abundance ratios and @xmath95 are known .
electron temperatures and densities for 48 pne are derived by fitting the synthetic to the observed spectra . for synthesis , the required he@xmath31/h@xmath31 and he@xmath32/h@xmath31 abundance ratios
have been observationally determined , and @xmath95 was derived by fitting the line profiles of strong recombination lines deconvolved with the thermal broadening , as listed in table [ tbl-1 ] .
three free parameters , nebular electron temperature and density and the fraction of the scattered rayleigh - jeans stellar continuum , are optimized simultaneously , using least chi square method to match the observed spectra . for the optimization ,
the fitted data points were specified explicitly by eye , within the wavelength range from 35503850 , but excluding regions of strong ( forbidden ) lines .
the results are partially shown in fig . [ all ] together with the residuals of the fits . in table
[ tbl-2 ] , we report the derived electron temperatures and densities along with their estimated errors . the balmer jump and decrement of the recombination spectrum have previously been used to determine nebular electron temperatures and densities in four pne , m 2 - 24 @xcite , ngc 6153 @xcite , m 1 - 42 and m 2 - 36 @xcite . for the four pne ,
the results derived here are in good agreement with the earlier results within the errors .
the errors caused by uncertainties in the input parameters he@xmath31/h@xmath31 and he@xmath32/h@xmath31 abundance ratios are negligible due to the small measurement errors of strong and lines and their small contribution to the continuum emission compared to hydrogen ( @xmath96 ) .
errors in @xmath95 can also be ignored as this quantity is well determined by fitting strong hydrogen lines .
the complicated velocity fields within a nebula can cause different lines to have different deviations from their laboratory wavelengths , and thus can affect the quality of line - fitting made under the simple assumption of a uniform velocity for all ions and lines .
line - fitting can also be affected by blends of unknown lines and deviations of line profiles from gaussian .
inspection of fig .
[ all ] shows that the observed intensities of the 2s@xmath97s@xmath98p@xmath97p@xmath99 @xmath100 lines are generally lower than its predicted values .
@xcite attributed this to the destruction of lyman photons by photoionization of h@xmath64 or by absorption of dust grains .
the wavelength region near the balmer jump also contains a few strong bowen fluorescence lines which are not considered in our simulations , thus giving large residuals at the corresponding wavelengths .
however , these effects do not contribute to errors in the derived electron temperatures and densities since these wavelengths have been excluded in the fitting . the observed spectra plotted in fig .
[ all ] , were de - reddened using extinction constants listed in table [ tbl-1 ] .
all spectra were normalized to h 11 @xmath101 in order to minimize errors in derived temperatures and densities caused by uncertainties in the reddening correction and flux calibration , and taking the advantage of the small wavelength difference between the balmer jump and h 11 .
the stellar continuum may have a spectral energy distribution that deviates from the rayleigh - jeans approximation adopted here .
however , this continuum is expected to be smooth and vary only slightly over such a short wavelength range of fitting . therefore , we have also ignored any possible errors caused by the fitting of the stellar continuum .
errors in the derived temperatures and densities are dominated by two effects .
noise is an important source of error . for most pne
, we found that uncertainties due to this factor are about 5@xmath102 in the derived temperatures and 0.1 dex in the densities .
the other major error source originates from the extensive wings of the strong [ ] @xmath103 3726 , 3729 lines , which can affect the shape of the continuum near the balmer jump . the effect is larger for low - excitation pne , such as ngc 40 and ngc 6720 .
the extensive wings of the [ ] lines can cause the continuum redwards of the balmer jump to be overestimated . as a result
, the temperatures can be overestimated .
the errors can be reduced by including data points at longer wavelengths in the fitting , where the extensive wings of the [ ] lines have less impact . however , increasing the wavelength range of the fitting would increase uncertainties in the reddening correction and in fitting the stellar continuum . to evaluate errors due to this effect , each spectrum was fitted twice for two wavelength ranges with one extending further into the red than the other .
the extensive wings of the [ ] lines can also decrease the shape of the continuum - like feature redwards of the balmer jump , leading to overestimated densities . for most
pne , this effect is negligible given the large wavelength difference ( about 70 ) between the balmer jump and the moderate strength of [ ] lines .
however , the effect can be significant for those with extremely strong [ ] lines .
on the other hand , the extensive wings of the [ ] lines have much less effect on the integrated intensities of high - order balmer lines ( @xmath46 , @xmath104 ) which can be determined using a line profile fitting technique .
comparison of densities derived from the continuum - like feature redwards of the balmer jump with those deduced from the integrated intensities of high - order balmer lines ( e.g. liu et al .
2001b ) allows us to estimate the errors in the derived densities caused by the extensive wings of the [ ] lines .
the final errors given in table [ tbl-2 ] were based on estimates of contributions from the two dominating sources of uncertainty discussed above .
electron densities deduced from the balmer decrements , and from the [ ] @xmath5 , [ ] @xmath40 , [ ] @xmath41 and [ ] @xmath42 doublet ratios are compared in fig .
[ necorr ] .
[ necorr ] shows that for most pne , the hydrogen recombination spectrum yields higher densities compared to forbidden lines diagnostics .
there is a trend such that the discrepancy becomes larger at higher densities .
two extreme cases are ic 4997 and m 2 - 24 . for ic 3568 , [ ] @xmath41 yields an abnormally low density .
the cause of it is unclear but is probably due to measurement error . in fig .
[ necorrir ] we compare electron densities derived from the [ ] far - infrared fine structure line ratio and from the hydrogen recombination spectrum .
evidently , densities derived from the [ ] @xmath105 ratio are significantly lower than those derived from the balmer decrements .
in fact , it has been found that the far - ir lines yield systematic lower densities , lower than even those deduced from optical forbidden line density - diagnostics @xcite .
a linear fit shows that the density discrepancy increases with the electron density .
o@xmath64 , s@xmath64 , cl@xmath106 , o@xmath106 and ar@xmath32 have ionization potentials of 13.6 , 10.4 , 23.8 , 35.1 and 40.7 ev , respectively , and therefore embrace a wide range of regions of differing states of ionization . on the other hand ,
the hydrogen recombination spectrum arises from the entire ionized nebula .
thus the discrepancies between densities derived from recombination spectrum and those deduced from forbidden line ratios are not likely to be caused by ionization stratifications .
we notice that the [ ] , [ ] , [ ] , [ ] and [ ] density - diagnostic lines have critical densities of @xmath107 @xmath14 .
therefore , if the nebulae contain regions of higher densities , these forbidden lines will be suppressed in these regions by collisional de - excitation .
in contrast , the hydrogen recombination spectrum is practically unaffected by collisional de - excitation . as a result , in such cases the hydrogen recombination spectrum will yield higher average densities compared to the [ ] , [ ] , [ ] , [ ] and [ ] lines ratios .
the same argument has been developed by @xcite to explain the systematically lower densities derived from the [ ] fine - structure line ratio compare to those deduced from optical forbidden line ratios .
the [ ] lines have higher critical densities than the [ ] , [ ] , [ ] lines .
thus , the effects are weaker for [ ] doublet ratio than for the other cases . given that the [ ] far - infrared fine - structure lines have the lowest critical densities amongst those density - diagnostic forbidden lines ( @xmath108([]@xmath109 @xmath110 ) , the effects are largest for the [ ] @xmath105 ratio .
this can be seen clearly in fig .
[ necorr ] and fig .
[ necorrir ] .
a nebular model including these high - density condensations and their effects on temperature determinations will be discussed in section 4.3 .
[ tecorr ] and fig .
[ tecorrir ] show that electron temperatures derived from the [ ] @xmath1 and @xmath111 ratios [ hereafter , @xmath0([]@xmath112 and @xmath0([]@xmath113 are systematically higher than those deduced from the hydrogen recombination spectrum [ hereafter @xmath0(bal ) ] , although there are a few exceptions where @xmath0(bal ) is higher than @xmath0([]@xmath3 ) .
the most extreme case is he 2 - 118 , for which @xmath0(bal ) is 5000 k higher than @xmath0([]@xmath3 ) . in our sample , the four pne with the largest temperatures discrepancies ,
hf 2 - 2 , m 1 - 42 , m 2 - 36 and ngc 6153 , were previously studied in detail by @xcite , @xcite and @xcite .
the discrepancy between @xmath0([]@xmath3 ) and @xmath0(bal ) was first discovered by @xcite and was attributed to temperature fluctuations within the nebula @xcite .
he characterized the temperature structure of a nebula by the average temperature @xmath114 and a mean square temperature fluctuation @xmath10 as follows : @xmath115 and @xmath116 where @xmath50 is the ion density for an observed emission line . assuming that all the oxygen is twice ionized and that there are no density and composition variations in the nebula , @xmath114 and @xmath10 can be derived from two of the three expressions , @xmath117_{\rm na})=t_0[1+\frac{1}{2}(\frac{9.13\times10 ^ 4}{t_0}-3)t^2],\ ] ] @xmath117_{\rm fn})=t_0[1+\frac{1}{2}(\frac{2.92\times10 ^ 4}{t_0}-3)t^2],\ ] ] and @xmath118 @xcite . in fig .
[ tecorr ] and fig . [ tecorrir ] , lines showing @xmath0([]@xmath3 ) as a function of @xmath119 for the case of @xmath120 , 0.02 , 0.06 and 1.00 are over plotted . fig .
[ tecorr ] shows that most nebulae have @xmath121 .
excluding a few extreme nebulae having @xmath122 , we obtain a mean value of @xmath123 .
the value is in excellent agreement with previous result @xmath124 derived by @xcite although their analysis contained larger measurement errors .
however , even excluding the extreme nebulae , a similar comparison between @xmath0([]@xmath4 ) and @xmath119 yields an average @xmath125 .
as we shall show in the following subsection , it is likely that @xmath0([]@xmath4 ) have been overestimated because of density variations in the nebula , resulting in higher @xmath126 when comparing @xmath0([]@xmath4 ) and @xmath119 .
another interesting point revealed by fig .
[ tecorr ] is that @xmath119 covers a wider range of values than @xmath0([]@xmath3 ) .
except a few extreme cases , @xmath0([]@xmath3 ) falls between 8000 and 14000 k , whereas @xmath119 varies from 900 to 19000 k. @xcite has argued that temperature fluctuations of gaseous nebulae can cause lower @xmath119 compared to @xmath0([]@xmath3 ) . however , even @xmath123 is beyond any photoionization model prediction .
another explanation of higher @xmath119 compared to @xmath0([]@xmath3 ) is that there are some condensations with @xmath127 @xmath14 in pne , which then suppress the @xmath0 ( [ ] ) @xmath1034959,5007 relative to the @xmath934363 , leading to an artificially higher @xmath0([]@xmath3 ) @xcite .
but no such high density condensations have been found in pne having extremely large temperature discrepancies @xcite .
recently , @xcite presented an empirical nebular model containing two components , each with its own temperature , density and chemical composition to account for the temperature discrepancy .
this seems to be the most plausible explanation at present .
for pne having @xmath128([]@xmath3 ) , as in he 2 - 118 , the electron temperature might conceivably increase in low - ionization regions owing to heating by shock waves in the outer regions . as a result , the average temperature of the entire nebula [ @xmath0(bal ) ] appears higher than that of high - ionization regions [ @xmath0([]@xmath3 ) ] . in 4.1 , we found that electron densities derived from the recombination spectrum are systematically higher than those derived from optical forbidden lines ratios , and confirm the earlier result of @xcite that densities derived from the [ ] far - ir fine - structure lines are the lowest of all .
these results show that nebular electron densities derived from various diagnostics are correlated with the critical density of the diagnostic lines , and suggest that dense clumps are generally present in pne . in such cases ,
the [ ] @xmath129 ratio is no longer a good temperature indicator given the relatively low critical densities of the [ ] @xmath130- and 88-@xmath131 lines ( 3500 and 1500 @xmath14 , respectively ) .
when nebulae contain condensations with densities in the range between 10@xmath132 and 10@xmath133 @xmath14 , the [ ] @xmath130- and 88-@xmath131 far - infrared lines will be heavily suppressed by collisional de - excitation , leading to overestimates of @xmath0([]@xmath4 ) . on the other hand ,
condensations of densities in this range will not affect @xmath0([]@xmath3 ) given that the [ ] @xmath134 and @xmath16 lines have higher critical densities of @xmath135 @xmath14 and @xmath136 @xmath14 , respectively ( cf . fig [ tecorrir ] ) .
another piece of evidence pointing to the presence of dense clumps comes from fig .
[ oiiite ] where we compare @xmath0([]@xmath3 ) and @xmath0([]@xmath4 ) .
if nebulae have temperature fluctuations but are homogeneous in density , then @xmath0([]@xmath4 ) is expected to be always lower than @xmath0([]@xmath3 ) @xcite . however , fig .
[ oiiite ] shows about two - third of the objects have higher @xmath0([]@xmath4 ) compared to @xmath0([]@xmath3 ) , suggesting that density variations play an important role in @xmath0([]@xmath4 ) determinations .
@xcite showed that electron temperatures derived from the [ ] @xmath137 ratio may be overestimated if nebulae contain ionized condensations of electron densities higher than 10@xmath133 @xmath14 .
however , except for a few extreme pne , such as ic 4997 @xcite , mz 3 @xcite and m 2 - 24 @xcite there is no evidence that nebulae contain a substantial amount of ionized gas in condensations with densities in excess of 10@xmath133 @xmath14 , as the presence of such condensations should have been revealed by the recombination spectrum ( liu et al . 2000 , 2001a , 2001b ; fig .
[ necorr ] ) .
therefore , we consider @xmath138 @xmath14 as a safe upper limit for the density of these condensations . in order to estimate the filling factor of the condensations
, we use an analytical method originally developed by @xcite for explaining the discrepancy of @xmath0([]@xmath3 ) and @xmath119 . assuming that some high - density regions ( @xmath127 @xmath14 ) are present , they deduced nebular - to - auroral [ ] line ratios ( see their eq .
following the same deduction of their eq .
2.3 but assuming that the electron density of these condensations is about @xmath139 @xmath14 instead of @xmath45 @xmath14 , we obtain the infrared - to - auroral line ratio @xmath140 where @xmath141 is line emission coefficient , @xmath142 , f and n refer to ir fine - structure line and nebular line , l and h refer to the low- and high - density regions , @xmath143 is the density contrast and @xmath144 is the filling factor . in the following calculation , we assume that : 1 ) density variations are the only cause of the lower @xmath0([]@xmath4 ) compared to @xmath0([]@xmath3 ) and @xmath145 ; 2 ) the nebular gas has a homogeneous electron temperature , i.e. , @xmath0([]@xmath3 ) or @xmath145 ; 3 ) the condensations have densities lower than the critical density of the [ ] auroral line but higher than the critical densities of the [ ] infrared lines , whereas low - density regions have a density that is lower than the critical densities of the [ ] auroral and infrared lines . setting @xmath146_{\rm fn}})=i_{52\micron+88\micron}/i_{4959}$ ]
, we have @xmath147 . for a simple assumption @xmath148 @xmath14 and @xmath149 @xmath14 ,
the ratio @xmath150 is about 0.04 , whereas @xmath151 is practically independent of the density when the densities are lower than the critical density of the nebular line and thus @xmath152 .
therefore , @xmath153 is given by @xmath154_{\rm fn}})/f(t_{\rm bal})-1}{0.04-f(t_{[{\rm o}~{\rm iii}]_{\rm fn}})/f(t_{\rm bal})}.\ ] ] in fig .
[ tecorrir ] , the lines showing the variation of @xmath0([]@xmath4 ) as a function of @xmath0(bal ) for @xmath155 , 0.1 , 0.5 , 1.0 and 2.0 are plotted .
we obtain an average value of @xmath156 .
however , it should be borne in mind that the @xmath153 derived by eq . ( 18 ) is a upper limit since temperature variations are actually present and contribute partly to the measured temperature discrepancies .
@xmath153 can also be estimated by comparing @xmath0([]@xmath3 ) with @xmath0([]@xmath4 ) , as shown in fig .
[ oiiite ] . assuming that @xmath0([]@xmath3 ) is constant across the entire nebula , @xmath153 is likewise given by @xmath154_{\rm fn}})/f(t_{[{\rm o}~{\rm iii}]_{\rm na}})-1}{0.04-f(t_{[{\rm o}~{\rm iii}]_{\rm fn}})/f(t_{[{\rm o}~{\rm iii}]_{\rm na}})}.\ ] ] in fig .
[ oiiite ] , the lines showing the variation of @xmath0([]@xmath4 ) as a function of @xmath0(bal ) for @xmath155 , 0.1 , 0.5 , 1.0 and 2.0 are also plotted .
an average value of @xmath157 is obtained .
it is noteworthy that the presence of temperature variations will increase @xmath0([]@xmath158 ) with respect to @xmath0([]@xmath4 ) , in contrast to density variations .
therefore , @xmath153 derived from eq .
( 19 ) is a lower limit .
the upper- and lower - limit of @xmath153 are listed in table .
[ tbl-3b ] . for a few pne
having a lower limit @xmath159 , the effect of temperature variations on the temperature determination is larger than that of density variations , causing higher @xmath0([]@xmath112 compared to @xmath0([]@xmath4 ) .
for all the pne , the mean value of @xmath153 is in the range from 0.4 to 1.0 .
thus , we suggest that @xmath160 may be a representative value that can be used for a typical nebula . adopting the density contrast @xmath161 ,
a representative value of the condensations filling factor @xmath162 is about @xmath163 .
the value is similar to that found by @xcite . adopting @xmath160 for all the pne
, we obtain new @xmath0([]@xmath4 ) , for which the contribution of the clumps has been deduced using equation ( 17 ) . comparing the new @xmath0([]@xmath4 ) and @xmath0(bal ) , we re - evaluate the average value @xmath164 , which is in good agreement with the value yielded by the comparison of @xmath0([]@xmath3 ) and @xmath0(bal ) ( see section 4.2 ) . in fig .
[ crene ] , the differences between temperatures derived from the [ ] forbidden lines and from the balmer jump are plotted against electron densities derived from the hydrogen recombination spectrum , the [ ] @xmath5 , [ ] @xmath40 , [ ] @xmath41 [ ] @xmath42 and the [ ] @xmath165m/@xmath166 m ratios , respectively . in all cases , the differences between the two temperature are anti - correlated with nebular densities derived from various density indicators . a least - squares fit yields
the following relation : @xmath167 and a linear correlation coefficient of 0.51 , where @xmath168([])@xmath169 and @xmath170 is that derived from hydrogen recombination spectrum , which represents the average density of the entire nebula .
@xcite found that the discrepancies of o@xmath32/h@xmath106 abundance ratios derived from orls and from cels are greater for larger , lower - surface brightness pne than for compact , dense ones . on the other hand
, @xcite found that the abundance discrepancies are greater for those with larger temperature discrepancies .
our results are therefore consistent with these earlier findings .
@xcite found that , @xmath171 where @xmath172 .
from the two relations given in equations ( 20 ) and ( 21 ) , we find that the ratio of abundances derived from orls and from cels has a power - law dependence on density , @xmath173 where @xmath174 . in general ,
electron densities of pne are expected to decline monotonically as nebulae expand .
@xcite found an approximate relation between nebular mean electron density and nebular size , @xmath175 , where r is the nebular radius .
consequently , it is suggested that the temperature discrepancy is related to the evolutionary state of pne ; more expended pne show larger temperature discrepancies . on the other hand ,
the actual evolution in a given nebula may depend on many factors .
furthermore , the electron density measured for a given nebula may depend on the diagnostic used .
for example , electron densities of low - ionization regions , as probed by the [ ] @xmath5 and [ ] @xmath40 line ratios , can be enhanced by interactions of fast stellar winds with earlier slow winds . therefore , there are slight difference in deriving the relation of @xmath176 and @xmath170 from different density diagnostics , as shown in fig .
[ crene ] .
we also use electron density derived from these forbidden line ratios to estimate the @xmath26-value in equation ( 22 ) .
the results show that the value is within the range of 0.3 and 0.9 . to explain the relation shown in equation ( 20 ) and ( 22 )
, we apply a chemically inhomogeneous two - component model presented by @xcite . in their hypothesis
, the nebula consists of two components , h - deficient clumps and diffuse material with ` normal ' abundances ( @xmath177solar ) .
these h - deficient clumps are expected to have a low electron temperature due to enhanced cooling caused by high metallicity and pressure equilibrium between the two components , so that only recombination lines are emitted . as a result ,
orls yield overestimated heavy element abundances . in a given pn ,
the contribution of the two components to the balmer temperature and abundance determinations depends on their densities , i.e. , the contribution becomes weaker as density declines .
if the clumps have neutral cores , as found in abell 30 @xcite and ngc 7293 @xcite , they are expected to be quite stable during the evolution of the pne . as a result , the relative contribution of the low - density and high - temperature material to the average temperature and abundance determinations decrease as nebulae expand .
consequently , the discrepancies in temperature and abundance determinations become larger as the nebulae expend . a long - standing problem in abundance determinations for pne
has been that the heavy element abundances derived from orls are systematically higher than those derived from cels .
the ratio of abundances derived from the two types of emission lines is found to vary from target to target .
@xcite analysed a pn sample and found the orl / cel discrepancy factors for these heavy element abundances span a range from about 2 to as large as 20 . for the most extreme
known pne hf 2 - 2 , the orl / cel abundance ratio reaches a record value of 84 @xcite .
another problem in nebular astrophysics is that @xmath0([]@xmath3 ) are systematically higher than @xmath119 , as we have discussed in 4.2 .
the two problems are found to be related .
a positive correlation between the ratio of o@xmath32/h@xmath31 derived from orls and from [ ] cels and the temperature discrepancy [ @xmath176=@xmath0([]@xmath3)-@xmath119 ] has been shown by @xcite .
the explanations for the dichotomy in abundance and temperature determinations have focussed on temperature and density variations and chemical inhomogeneities .
however , it has been shown that the temperature and density variations alone can be ruled out by current observations @xcite . at present ,
the most plausible explanation is that there exist some cold h - deficient clumps in nebula @xcite where only recombination lines are emitted . to fit the observed pattern of ngc 6153 ,
@xcite have presented several empirical models incorporating h - deficient clumps .
@xcite also constructed photoionization models of ngc 6153 and m 1 - 42 , incorporating h - deficient clumps , which reproduce the observed integrated line ratio satisfactorily . in all the models ,
the h - deficient clumps contain only about @xmath179 of the total mass , suggesting that the h - deficient components contain only a tiny amount of nebular gas .
nevertheless , they can seriously enhance the intensities of orls , causing overestimated orl abundances .
therefore , the he / h abundance ratio could be overestimated owing to the existence of the h - deficient clumps can lead to since all the observable strong and lines are recombination lines .
some h - deficient components have been found in some pne , such as abell 30 and abell 78 @xcite .
@xcite analysed optical spectra of the h - deficient knots in abell 30 and found the he / h abundance ratios of the knots are more than ten .
these pne with extremely large abundance and temperature discrepancies , such as ngc 6153 , m 1 - 42 and hf 2 - 2 , have been found to have high he / h abundances . @xcite
defined the pne having high he / h abundance ( @xmath180 ) as type - i pne . according to our suggestion , these pne with high he / h abundances might result from contamination from h - deficient clumps
. quantitative analysis of overestimated he / h abundances in pne is quite difficult . for this purpose , as mentioned by @xcite , one has to rely on the measurement of @xmath9310830 line which is strongly enhanced by collisional excitation from the 2s@xmath181s metastable level . if the he / h abundance is enhanced by h - deficient clumps , one expects that there is a positive correlation between the he / h abundance and the orl / cel abundance ratio of heavy element . for most pne , however , heavy element orls are quite weak and thus yield large uncertainties in the resultant abundance .
instead , we can study the relation between the he / h abundance and @xmath176 since the orl / cel abundance ratio has been found to be positively correlated with @xmath176 by @xcite . fig .
[ hete ] plots he / h versus @xmath176 .
the measurement error of the he / h abundance ratio is negligible . in the figure , filled and open circles represent the galactic disc and bulge pne respectively and the solid line is a linear fit represented by the relation : @xmath182 which yields a linear correlation coefficient of 0.50 . although there is considerable scatter , there is a trend of increasing he / h abundance with increasing difference between the temperature from [ ] forbidden lines and balmer discontinuity for both the galactic disc and bulge pne .
therefore , it seems plausible that the he / h abundances in pne have been generally overestimated due to the existence of cold h - deficient clumps .
[ hete ] also shows that the most of pne have enhancement of helium with respect to the sun , which is often ascribed to the second and third dredge - up .
however , the normal model of dredge - up can not explain some extreme cases , such as he 2 - 111 which has a he / h abundance of 0.219 @xcite .
for these cases , the more likely reason is that pollution of the h - deficient clumps causes an overestimate of the he / h abundance ratio . on the other hand
, we can not completely reject the possibility that helium in nebulae has been partly enhanced by the products of nucleosynthesis of the central star .
that may be the main cause of the scatter in the derived he / h abundances .
the presence of hydrogen - deficient clumps may be further proof .
the orl / cel abundance discrepancies are also found in regions @xcite . and
the existence of some high density clumps in regions has been indicated by previous studies ( e.g. odell et al .
therefore , the h - deficient material might also be present in regions , leading to overestimated orl abundances .
if this is the case , the primordial helium abundance , which is based on observations of some metal poor extragalactic regions , could have been overestimated .
the hydrogen recombination spectrum near the paschen jump can also provides a diagnostic of electron temperature and density [ hereafter , @xmath183 and @xmath184 .
unfortunately , in our sample , the spectra covering the paschen jump are available for only four pne , ngc 7027 , ngc 6153 , m 1 - 42 and ngc 7009 .
using the same method described in section 3 , we match the theoretical with the observed spectra at 8250 .
given that the observed spectra near the paschen discontinuity may include a contribution from a dust emission tail , we added a constant continuum for the match .
the results are shown in fig .
all the spectra are normalized to p 20 in order to decrease the uncertainty of reddening correction .
table [ tbl-4 ] lists the derived temperatures and densities .
one advantage of using the spectrum near the paschen jump is that the resultant temperature and density are not affected by the continuum of the hot central star and two - photon emission .
however , the emission from local dust may be stronger in this band .
in addition , the estimation of the continuum bluewards the paschen jump is affected by the telluric absorption ( h@xmath185o and o@xmath185 ) . and
the absorption feature might also cause a error in fitting the continuum longward of the paschen jump of some pne , such as m 1 - 42 . as a result , the temperatures and densities derived from the hydrogen recombination spectra at the paschen jump are less reliable than those derived from those at the balmer jump .
table [ tbl-4 ] shows that the @xmath186 are consistent with the @xmath187 within the errors .
however , the @xmath183 are found to be 4000 , 1000 , 500 and 1200k lower than the @xmath119 for ngc 7027 , ngc 6153 , m 1 - 42 and ngc 7009 , respectively .
this may be due to systematic errors in the reddening correction or flux calibration . as an alternative , it remains possible that the discrepancy is caused by temperature variations .
further investigations including a larger data sample of pne is needed to clarify the point .
in this work , we presented a method to derive nebular electron temperature and density simultaneously by matching the theoretical with the observed hydrogen recombination spectra . using the method , we derived electron temperatures and densities of 48 galactic pne .
the temperatures and densities derived by this method are compared with those derived from other indicators , enabling us to study electron temperature and density variations in pne .
our main findings are as follows .
the temperatures derived from the nebular - auroral line ratios of [ ] are systematically higher than those derived from hydrogen recombination spectra , suggesting that temperature variations are generally present in nebulae .
we obtained the rms temperature fluctuation parameter @xmath188 as a representative value .
the larger discrepancy in temperatures derived from [ ] @xmath129 line ratios and from hydrogen recombination spectra , however , can not be attributed simply to the temperature variations .
the presence of clumps with @xmath138 @xmath14 can account well for the discrepancy . a comparison between electron temperatures derived from [ ] @xmath129 line ratios and those derived from hydrogen recombination spectra and [ ] @xmath1 line ratios suggests that the filling factor of the condensations has a typical value of @xmath163 .
the discrepancies in temperatures derived from [ ] @xmath1 line ratios and from hydrogen recombination spectra are related to the evolution of pne .
the orl / cel abundance ratio is found to have a power - law dependence on electron density ( @xmath189 , @xmath174 ) .
we attribute this to different evolutionary scenario of two components in pne , i.e. expanding diffuse nebular material and stable cold h - deficient clumps .
the contribution of the clumps to the temperature and abundance determinations increases as the nebulae expand .
he / h abundance has a positive correlation with the temperature discrepancy , suggesting that he / h abundances might be overestimated due to the existence of h - deficient clumps .
if this is the case , traditional up - dredge theory should be modified .
the primordial helium abundance determination will be affected if there are significant condensations in regions .
further investigations are needed .
the densities derived from the spectrum near the paschen jump are in good agreement with those derived form that near the balmer jump . however , the former yield generally lower temperatures than the latter .
the reason remain unknown .
we conclude that the existence of h - deficient clumps in pne is the most plausible explanation for the origin of temperature and density variations and the orl / cel abundance discrepancies .
however , it is a challenge to clarify the origin and evolution of these clumps . in further work
, we will study the statistical relationships between the temperature and abundance discrepancies and other physical conditions of pne such as radio flux , morphology , chemical abundances and the nature of the central star .
the authors are grateful to w. wang for his help with part of the data reduction .
we also thank j .- s chen and s .- g .
luo for their help with the preparation of this paper .
constructive suggestions and comments by the referee s. m. viegas were appreciated .
this work was supported in part by beijing astrophysics center ( bac ) .
lcccccc hu 1 - 2 & 0.512 & 0.046 & 0.079 & @xmath190 & & 144 + ic 3568 & 0.261 & 0.095 & 0.001 & @xmath191 & @xmath192 & 132 + ngc 7027 & 1.380 & 0.058 & 0.041 & @xmath193 & @xmath194&82 + ngc 7009 & 0.130 & 0.093 & 0.011 & @xmath195 & @xmath196 & 132 + ngc 40 & 0.700 & 0.062 & 0.000 & @xmath197 & @xmath198 & 132 + ngc 6210 & 0.055 & 0.099 & 0.001 & @xmath199 & @xmath200 & 148 + ngc 6826 & 0.064 & 0.099 & 0.000 & @xmath201 & @xmath202 & 132 + ngc 6572 & 0.400 & 0.110 & 0.000 & @xmath197 & @xmath203 & 132 + ngc 6741 & 1.100 & 0.081 & 0.031 & @xmath204 & @xmath205 & 132 + ngc 6790 & 1.100 & 0.081 & 0.003 & @xmath206 & & 132 + ngc 6884 & 1.050 & 0.081 & 0.016 & @xmath207 & @xmath208 & 132 + ngc 7662 & 0.180 & 0.061 & 0.038 & @xmath209 & @xmath210 & 156 + ngc 6543 & 0.100 & 0.117 & 0.000 & @xmath211 & @xmath212 & 140 + m 2 - 24 & 0.800 & 0.099 & 0.005 & @xmath213 & & 156 + cn 2 - 1 & 1.063 & 0.104 & 0.004 & @xmath214 & & 132 + ngc 6720 & 0.200 & 0.094 & 0.017 & @xmath215 & @xmath216 & 132 + ngc 6620 & 0.760 & 0.099 & 0.021 & @xmath217 & & 132 + ngc 6567 & 0.670 & 0.105 & 0.007 & @xmath218 & @xmath219 & 132 + h 1 - 35 & 1.360 & 0.115 & 0.001 & @xmath220 & & 132 + h 1 - 50 & 0.700 & 0.100 & 0.011 & @xmath221 & & 132 + he 2 - 118 & 0.158 & 0.065 & 0.000 & @xmath222 & & 148 + m 1 - 20 & 1.200 & 0.094 & 0.000 & @xmath223 & & 123 + m 3 - 21 & 0.570 & 0.100 & 0.007 & @xmath224 & & 140 + m 3 - 32 & 0.579 & 0.120 & 0.013 & @xmath225 & & 140 + ic 1297 & 0.220 & 0.079 & 0.042 & @xmath226 & & 132 + ic 4634 & 0.454 & 0.097 & 0.000 & @xmath227 & @xmath228 & 140 + ic 4776 & 0.265 & 0.088 & 0.000 & @xmath229 & & 136 + ic 4997 & 0.438 & 0.139 & 0.000 & @xmath230 & & 127 + ngc 5873 & 0.004 & 0.063 & 0.040 & @xmath231 & & 127 + ngc 5882 & 0.380 & 0.113 & 0.003 & & & 127 + ngc 6302 & 1.460 & 0.105 & 0.067 & @xmath232 & @xmath233 & 132 + ngc 6818 & 0.400 & 0.050 & 0.051 & @xmath204 & @xmath234 & 148 + ngc 6153 & 1.300 & 0.123 & 0.011 & @xmath235 & @xmath236 & 148 + m 2 - 36 & 0.270 & 0.133 & 0.003 & @xmath237 & @xmath238 & 148 + m 1 - 42 & 0.700 & 0.139 & 0.009 & @xmath239 & @xmath240 & 140 + ngc 6778 & 0.890 & 0.155 & 0.008 & @xmath241 & & 140 + hf 2 - 2 & 0.200 & 0.131 & 0.002 & @xmath242 & & 132 + dddm 1 & 0.137 & 0.060 & 0.000 & @xmath243 & & 99 + hu 2 - 1 & 0.777 & 0.097 & 0.000 & @xmath244 & & 99 + ic 2003 & 0.347 & 0.054 & 0.049 & @xmath245 & & 99 + ic 5217 & 0.501 & 0.088 & 0.008 & @xmath246 & & 115 + me 2 - 2 & 0.343 & 0.138 & 0.000 & @xmath247 & & 90 + ngc 6803 & 0.869 & 0.112 & 0.004 & @xmath248 & & 107 + ngc 6807 & 0.642 & 0.084 & 0.000 & @xmath249 & & 107 + ngc 6833 & 0.000 & 0.078 & 0.000 & @xmath250 & & 99 + ngc 6879 & 0.401 & 0.091 & 0.003 & @xmath251 & & 99 + ngc 6891 & 0.287 & 0.094 & 0.000 & @xmath252 & & 115 + vy 1 - 1 & 0.421 & 0.094 & 0.000 & @xmath253 & & 99 + lcc hu 1 - 2 & @xmath254 & @xmath255 + ic 3568 & @xmath256 & @xmath257 + ngc 7027 & @xmath258 & @xmath259 + ngc 7009 & @xmath260 & @xmath261 + ngc 40 & @xmath262 & @xmath263 + ngc 6210 & @xmath264 & @xmath261 + ngc 6826 & @xmath265 & @xmath266 + ngc 6572 & @xmath267 & @xmath268 + ngc 6741 & @xmath269 & @xmath270 + ngc 6790 & @xmath271 & @xmath272 + ngc 6884 & @xmath273 & @xmath261 + ngc 7662 & @xmath274 & @xmath275 + ngc 6543 & @xmath276 & @xmath257 + m 2 - 24 & @xmath277 & @xmath278 + cn 2 - 1 & @xmath279 & @xmath280 + ngc 6720 & @xmath281 & @xmath282 + ngc 6620 & @xmath283 & @xmath284 + ngc 6567 & @xmath285 & @xmath286 + h 1 - 35 & @xmath287 & @xmath280 + h 1 - 50 & @xmath288 & @xmath289 + he 2 - 118 & @xmath290 & @xmath291 + m 1 - 20 & @xmath292 & @xmath293 + m 3 - 21 & @xmath294 & @xmath295 + m 3 - 32 & @xmath296 & @xmath270 + ic 1297 & @xmath297 & @xmath298 + ic 4634 & @xmath299 & @xmath300 + ic 4776 & @xmath301 & @xmath280 + ic 4997 & @xmath302 & @xmath303 + ngc 5873 & @xmath304 & @xmath305 + ngc 5882 & @xmath306 & @xmath298 + ngc 6302 & @xmath307 & @xmath308 + ngc 6818 & @xmath309 & @xmath310 + ngc 6153 & @xmath311 & @xmath312 + m 2 - 36 & @xmath311 & @xmath313 + m 1 - 42 & @xmath314 & @xmath315 + ngc 6778 & @xmath316 & @xmath317 + hf 2 - 2 & @xmath318 & @xmath319 + dddm 1 & @xmath320 & @xmath321 + hu 2 - 1 & @xmath322 & @xmath323 + ic 2003 & @xmath324 & @xmath325 + ic 5217 & @xmath326 & @xmath325 + me 2 - 2 & @xmath327 & @xmath328 + ngc 6803 & @xmath329 & @xmath330 + ngc 6807 & @xmath331 & @xmath332 + ngc 6833 & @xmath333 & @xmath334 + ngc 6879 & @xmath335 & @xmath336 + ngc 6891 & @xmath337 & @xmath338 + vy 1 - 1 & @xmath337 & @xmath325 + lccccc ic 3568 & 0.527 & 0.254 & 0.315 & 1.174 & 0.253 + ngc 7027 & 0.281 & 0.169 & 0.254 & 0.708 & 0.532 + ngc 7009 & 1.210 & 0.720 & 0.433 & 0.730 & 0 .
+ ngc 40 & 1.345 & 0.433 & 0.374 & 2.408 & 0 .
+ ngc 6210 & 0.641 & 0.392 & 0.469 & 0.681 & 0.208 + ngc 6826 & 0.641 & 0.374 & 0.520 & 0.768 & 0.415 + ngc 6572 & 0.401 & 0.151 & 0.374 & 1.857 & 1.639 + ngc 7662 & 0.271 & 0.238 & 0.228 & 0.145 & 0 . + ngc 6543 & 1.505 & 0.688 & 0.872 & 1.300 & 0.282 + ngc 6720 & 0.836 & 0.303 & 0.374 & 1.981 & 0.246 + ic 4634 & 0.688 & 0.258 & 0.411 & 1.867 &
0.634 + ngc 6818 & 0.258 & 0.171 & 0.254 & 0.543 & 0.516
+ ngc 6153 & 2.543 & 0.714 & 0.254 & 2.987 & 0 .
+ m 2 - 36 & 2.543 & 2.056 & 0.714 & 0.249 & 0 .
+ m 1 - 42 & 2.543 & 1.422 & 0.543 & 0.850 & 0 .
+ average & & & & 1.0 & 0.4 + | a method is presented to derive electron temperatures and densities of planetary nebulae ( pne ) simultaneously , using the observed hydrogen recombination spectrum , which includes continuum and line emission . by matching theoretical spectra to observed spectra around the balmer jump at about 3646
, we determine electron temperatures and densities for 48 galactic pne .
the electron temperatures based on this method [ hereafter @xmath0(bal ) ] are found to be systematically lower than those derived from [ ] @xmath1 and [ ] @xmath2 ratios [ hereafter , @xmath0([]@xmath3 ) and @xmath0([]@xmath4 ) ] . and
the electron densities based on this method are found to be systematically higher than those derived from [ ] @xmath5 , [ ] @xmath6 , [ ] @xmath7 , [ ] @xmath8 and [ ] @xmath9 ratios .
these results suggest that temperature and density fluctuations are generally present within nebulae .
the comparison of @xmath0([]@xmath3 ) and @xmath0(bal ) suggests that the fractional mean - square temperature variation ( @xmath10 ) has a representative value of 0.031 .
a majority of temperatures derived from the [ ] @xmath2 ratio are found to be higher than @xmath0([]@xmath3 ) , which is attributed to the existence of dense clumps in nebulae those [ ] ir fine - structure lines are suppressed by collisional de - excitation in the clumps . by comparing @xmath0([]@xmath4 ) , @xmath0([]@xmath3 ) and @xmath0(bal ) and assuming a simple two - density - component model , we find that the filling factor of dense clumps has a representative value of 7@xmath11 . the discrepancies between @xmath0([]@xmath3 ) and @xmath0(bal ) are found to be anti - correlated with electron densities derived from various density indicators ; high - density nebulae have the smallest temperature discrepancies .
this suggests that temperature discrepancy is related to nebular evolution .
in addition , he / h abundances of pne are found to be positively correlated with the difference between @xmath0([]@xmath3 ) and @xmath0(bal ) , suggesting that he / h abundances might have been overestimated generally because of the possible existence of h - deficient knots .
electron temperatures and densities deduced from spectra around the paschen jump regions at 8250 are also obtained for four pne , ngc 7027 , ngc 6153 , m 1 - 42 and ngc 7009 .
electron densities derived from spectra around the paschen jump regions are in good agreement with the corresponding values derived from spectra around the balmer jump , whereas temperatures deduced from the spectra around the paschen jump are found to be lower than the corresponding values derived from spectra around the balmer jump for all the four cases .
the reason remains unclear . |
the past decade has seen the abundant use of the sphere decoding ( sd ) algorithm @xcite as a tool for facilitating near maximum likelihood ( ml ) decoding over the coherent delay - limited ( or quasi - static ) multiple - input multiple - output ( mimo ) channel .
the sd algorithm allows for efficient optimal or near optimal decoding of a large number of high rate space - time codes that map constituent constellation symbols linearly in space and time @xcite .
as the algorithm s computational cost depends on the fading channel , it is generally known that in implementing the sd algorithm , one can tradeoff computational complexity for error performance by selectively choosing when to decode and when not to .
equivalently , in the presence of constraints on the computational reserves that may be allocated to decoding , the algorithm is faced with the prospect of encountering channel realizations that force it to violate its run - time constraints , thus having to declare decoding outages that inevitably reduce reliability .
this naturally raises the intriguing question of how large computational reserves are actually required for near ml performance .
while this question is hard to answer in general , or even ask in a rigorously meaningful way , we show herein that by following @xcite and considering the decoding of sequences of codes in the high signal - to - noise ratio ( snr ) limit , not only can the question be made rigorous : it also admits surprisingly simple explicit and general answers .
drawing from the diversity multiplexing tradeoff ( dmt ) setting which has already been successfully applied to concisely describe the high snr diversity exponent in the reliability analysis of reduced complexity decoders @xcite , we introduce the _ sd complexity exponent _ as a measure of complexity of the sd algorithm .
the sd complexity exponent characterizes the decoding complexity in the high snr limit under the assumption that the code - rate scales with snr in order to provide a given multiplexing gain .
this approach naturally takes into account the dependency of the sd complexity on the codeword density and the codebook size , as well as the snr and the fading characteristics of the wireless channel .
similar to previous work on the dmt relating the code rate and probability of decoding error , it is seen that also the complexity , although hard to characterize at any finite snr , has mathematically tractable characterizations in the high snr asymptote . these characterizations in turn yield valuable insights into the behavior of the algorithm .
the sd algorithm is equivalent to a branch - and - bound search @xcite over a regular tree and like most other works on sd complexity @xcite we view the number of visited nodes @xmath0 as the complexity of the algorithm . to identify an appropriate scale of interest for complexity at high snr ,
it is useful to note that in order to achieve a multiplexing gain of @xmath1 the code must in the high snr limit have rate of denotes the base-2 logarithm and @xmath2 is the standard landau notation where @xmath3 implies @xmath4 .
similarly , @xmath5 implies that @xmath6 for @xmath7 . ]
@xmath8 bits per channel use where @xmath9 denotes the snr .
consequently , the cardinality of the codebook is @xmath10 where @xmath11 is the codeword length and where @xmath12 denotes equality in the snr exponent ( cf .
@xcite and section [ sec : notation ] ) . in the worst case , as will be shown later , the sphere decoder is in essence forced to perform a complete search over the entire codebook , and its complexity is in this case @xmath13 .
however , although feasible , this event is also highly improbable .
thus , in order to quantify the probability that the sphere decoder visits a certain ( large ) number of nodes , we introduce a ( complexity ) rate - function @xmath14 over @xmath15 given implicitly by @xmath16 where @xmath0 is the complexity of the sd algorithm . in short , @xmath14 captures the decay - rate of the probability that the complexity exceeds a given snr dependent threshold @xmath17 , or equivalently , that the algorithm visits a specific sizable subset of the codebook .
this decay - rate should be contrasted with the minimum probability of decoding error , which vanishes in the high snr limit as @xmath18 where @xmath19 is the diversity gain of the code under maximum likelihood ( ml ) decoding . we can
thus judiciously argue that for any @xmath20 such that @xmath21 , the probability that the complexity exceeds @xmath17 is at high snr insignificant in comparison to the overall probability of error of the decoder .
in other words : for @xmath20 such that @xmath22 , imposing a run - time limit of @xmath17 on the complexity of the algorithm and declaring a decoding outage whenever this limit is not met would cause a vanishing degradation in terms of the overall error probability at high snr .
this motivates us to deviate from the traditional worst - case complexity measure that fails to meaningfully describe the effective complexity , and to define the sd complexity exponent @xmath23 as the infimum of all @xmath20 for which @xmath21 .
in essence , @xmath24 represents the minimum computational reserves required for achieving dmt optimal performance using the sd algorithm .
precise definitions of @xmath23 , and a rigorous treatment of the notion of a vanishing degradation in the overall error probability , is given in section [ sec : decoding - complexity ] and by theorem [ thrm : vanishingnew ] .
the main topic of this work will then be to give closed form expressions , and bounds , for the sd complexity exponent @xmath23 when decoding different classes of full rate linear codes , to be described later , including the codes proposed in @xcite .
most other works on sphere decoding complexity consider uncoded ( spatially multiplexed ) systems and asymptotic results in terms of the signal space dimension , see , e.g. , @xcite .
our work is instead more related to the analysis in @xcite , which considers the complexity tail distribution for a fixed signal space dimension .
however , unlike @xcite we also incorporate the space - time codes into the analysis , as well as the snr scalings of the rates of these codes mandated by the dmt . in parallel with our work
the work in @xcite provides an analysis of the complexity tail distribution for unconstrained lattice sequential decoders , in the presence of dmt optimal random lattice codes .
a fundamental difference with our work and @xcite is that @xcite considers unconstrained lattice decoding , whereas we explicitly take into account the constellation boundary in the decoder .
another difference is that the lattice codes considered in our work can be explicitly constructed and can have arbitrary dmt performance , unlike the random codes in @xcite which are non - explicit and which are restricted to being dmt optimal .
we also take our analysis one step further by coupling the complexity tail distribution to the dmt performance of the code in order to obtain the sd complexity exponent . regarding the ultimate complexity limits on dmt optimal decoding
, we have previously established that lattice reduction ( lr)-aided linear decoders are sufficient for achieving the entire dmt tradeoff at a worst - case complexity of @xmath25 , i.e. , corresponding to a complexity exponent of @xmath26 @xcite . this is lower than the sd complexity exponent that we will present in what follows .
however one notable difference is that the statements made herein are fundamentally stronger in terms of error probability as they not only imply full diversity but also a vanishing snr gap to the ml decoder ( see theorem [ thrm : vanishingnew ] for details ) .
such a result was not established for the decoders in @xcite .
the general definition of the sd complexity exponent is given in definition [ def : complexityexponent ] and theorem [ thrm : vanishingnew ] then describes how sphere decoding and the time - out policies to be employed can guarantee a gap to ml that vanishes with increasing snr .
however , before proceeding with the statement of these results , we first consider the code - channel system , describe the basic workings of the sd algorithm , and handle different pertinent aspects that are necessary for the exposition that follows . following the definition of the sd complexity exponent , theorem [ thrm :
upper - bound ] gives , in the form of an optimization problem , a general upper bound @xmath27 on the sd complexity exponent @xmath23 when decoding any full rate code with multiplexing gain @xmath1 and diversity @xmath19 .
an explicit closed form expression for @xmath27 is then given in theorem [ thrm : upper - bounddmtopt ] for all dmt optimal full rate codes .
the bound @xmath27 is already useful in itself in that it establishes that the sd complexity exponent is much lower than the worst - case snr exponent @xmath28 associated with a full search of the codebook .
however , in the interest of also establishing the tightness of this bound , lemma [ lm : lower - bound ] provides easy - to - check sufficient conditions on the generator matrix of the code lattice , that guarantee the tightness of @xmath27 in the most general setting . building on this , theorem [ thrm : perm - tightness - fullrate ] establishes that , given any full rate design of arbitrary dmt performance , there is always at least one non - random sd column ordering @xcite for which @xmath29 , i.e. , for which the exact @xmath23 can be explicitly calculated from the result of theorem [ thrm : upper - bound ] .
theorem [ thrm : layered ] goes one step further and establishes the exact sd complexity exponent , given any _ threaded _ code design and the natural column ordering , to be @xmath29 while theorem [ thrm : layereddmt ] provides an explicit expression for @xmath23 for any dmt optimal threaded code design .
surprisingly , this simple expression ( see fig . [ fig : exponents ] ) can also serve as an upper bound on @xmath23 for any full - rate code , irrespective of the fading statistics . finally , and along a different path ,
theorem [ thrm : nvd ] establishes @xmath23 for any @xmath30 approximately universal code @xcite , irrespective of its specific structure , thus identifying the exact @xmath23 even for possibly undiscovered code structural designs , as long as these designs are approximately universal , i.e. , as long as they achieve dmt optimality for all fading statistics .
some general discussions of these results are then provided in section [ sec : discussions ] .
[ c]multiplexing gain @xmath1 [ c]complexity exponent @xmath23 [ c]@xmath31 [ c]@xmath32 [ c]@xmath33 [ c]@xmath34 [ c]@xmath35 for decoding threaded minimum delay dmt optimal codes with @xmath36 for @xmath37 .
the sd complexity exponent is illustrated by the bold lines .
the same exponent also serves as an upper bound to the sd complexity exponent when decoding any minimum delay dmt optimal full rate linear dispersive code .
the thin lines show the quadratic function given by @xmath38 which provides the exact complexity exponent at integer multiplexing gains .
[ fig : exponents],title="fig:",width=313 ] the sd complexity exponent for decoding the class of full rate threaded dmt optimal codes with minimum delay , i.e. , for which @xmath39 where @xmath40 denotes the number of transmit antennas , is shown shown in fig .
[ fig : exponents ] is given by in theorem [ thrm : layereddmt ] .
] in fig .
[ fig : exponents ] for @xmath41 .
the results shown in the figure apply to codes such as those presented in @xcite . before proving the aforementioned results
, it is worth commenting on the somewhat counterintuitive result suggested by the sd complexity exponent in fig .
[ fig : exponents ] .
namely , that while @xmath23 initially increases as a function of the multiplexing gain @xmath1 , it then decreases as @xmath1 approaches its maximum value @xmath40 .
the initial increase can easily be explained by the fact that the cardinality ( and density ) of the codebook @xmath42 increases as a function of @xmath1 .
however , the decrease at high multiplexing gains can be understood in light of the coupling of the complexity with the overall probability of error : in short , at high multiplexing gains the error probability is also higher and this implies that the decoder may time out for a larger set of problem instances without seriously affecting the overall performance , leading to an overall reduction in decoding complexity .
pushing the code - decoder pair towards the maximal data rate does therefore not imply that the decoding complexity is maximized .
this effect is discussed further in section [ sec : outages ] , in terms of information theoretic outages .
we let @xmath43 , @xmath44 , and @xmath45 , denote the set of integer , real and complex numbers respectively and @xmath46 and @xmath47 the set of @xmath48-vectors and @xmath49-matrices over @xmath50 .
vectors are denoted by lower - case bold letters @xmath51 , and matrices are denoted by upper - case bold letters @xmath52 .
we use @xmath53 and @xmath54 to denote the transpose and hermitian ( conjugate ) transpose of vectors and matrices , and @xmath55 to denote the matrix to vector operation whereby the columns of the argument are stacked on top of each other
. we use @xmath56 to denote the @xmath57 identity matrix , and use @xmath58 to denote the zero vector or matrix where the dimensions are given by the context in which @xmath58 is used . deviating slightly from standard usage , we refer to a tall matrix @xmath59 where @xmath60 as unitary if @xmath61 . for @xmath62
we use @xmath63 to respectively denote the real and imaginary parts of @xmath64 .
for @xmath65 we use @xmath66 to denote the floor operation defined as the largest integer smaller than or equal to @xmath64 , @xmath67 to denote the ceil operation defined as the smallest integer larger than or equal to @xmath64 , and we let @xmath68 . we let @xmath69 denote the ordered ( structurally ) non - zero singular values @xcite of a matrix @xmath70 where @xmath60 . we will on occasion use @xmath71 to denote the largest singular value when the dimension of @xmath52 is not explicit .
we make no notational difference between random variables and their realizations .
we will use the notation @xmath72 to label the stochastic event within the brackets .
finally , in order to simplify notation we will make use of the @xmath12 ( and @xmath73 , @xmath74 , @xmath75 , @xmath76 ) notation for equalities ( and inequalities ) in the snr exponent , cf .
specifically , we write @xmath77 and @xmath78 to denote @xmath79 and @xmath80 when @xmath81 and @xmath82 simultaneously hold .
the definition of @xmath75 and @xmath76 follows after replacing @xmath83 by @xmath84 and @xmath85 by @xmath86 in the above expressions .
we consider the standard block rayleigh fading @xmath87 quasi - static point - to - point mimo channel model with coherence - time @xmath11 given by @xmath88 where @xmath89 , where @xmath90 , and where @xmath91 denote the transmitted space - time block codeword , the block of received signals , and additive spatially and temporally white gaussian noise .
the channel gains @xmath92 are assumed to be i.i.d .
circularly symmetric complex gaussian where the extension to arbitrary fading distributions is discussed . ]
( i.e. , rayleigh fading ) and constant over the duration of the transmission ( i.e. , quasi - static rayleigh fading ) . we shall assume throughout that @xmath93 .
the transmitted codewords @xmath94 are assumed to be drawn uniformly from a codebook @xmath42 and we assume that @xmath95 so that the parameter @xmath9 takes on the interpretation of an average snr . note also here that one use of is viewed as @xmath11 uses of the wireless channel in the definition of the data - rate .
we shall herein restrict our attention to full rate ( complex ) linear dispersive codes @xcite of the form @xmath96 where @xmath97 are constellation symbols drawn from a finite alphabet @xmath98 , where @xmath99 is a set of linearly independent _ dispersion matrices _ , and where @xmath100 is a parameter regulating the transmit power .
the notion of full rate implies that each codeword @xmath89 carries @xmath101 constellation symbols .
we will further make the additional assumption that @xmath98 belongs in the class of qam - like alphabets of the form @xmath102 \ , \}\ ] ] where @xmath103 is a parameter regulating the size of the constellation .
a detailed exposition of the mathematical machinery that allows for this relaxation can be found in ( * ? ? ?
* section iii ) ]
. we will use @xmath104 to denote the extended ( infinite ) constellation obtained by letting @xmath105 in , and note that @xmath104 is nothing but the set of gaussian integers .
qam constellations will in general also include a translation and scaling of the underlying lattice @xmath104 .
however , as including such a translation would not affect the results obtained herein , and as the scaling can easily be included in the dispersion matrices @xmath106 or @xmath100 , we omit these variations and concentrate on in the interest of notational simplicity . the channel model in may also be equivalently expressed in a vectorized form according to @xmath107 where @xmath108 , where @xmath109 , where @xmath110 , and where @xmath111 denotes the kronecker product @xcite .
we shall mainly work with rather than directly . in the vectorized form the codewords @xmath112
are given by @xmath113 for @xmath114 , and where the full rank matrix @xmath115 is referred to as the _ generator matrix _ of the code .
the linear dispersive codes form a subset of the lattice codes @xcite as the codewords constitute a subset of the ( complex ) lattice @xmath116 .
the parameters @xmath100 and @xmath117 are , as noted , chosen in order to satisfy given transmit power and rate constraints .
in particular , in order to ensure a multiplexing gain of @xmath118 or equivalently a rate of @xmath8 , it must hold that @xmath119 which by and implies that @xmath120 .
the code structure described above includes the codes proposed in @xcite , as well as the qam - based codes of @xcite , as special cases .
finally , we will throughout , with slight abuse of terminology but still in line with @xcite , use the term _ code _ when referring to the whole _ family of codes _ that is generated by a single generator matrix @xmath121 for different multiplexing gains and snrs , and trust that no confusion should follow by this usage .
the coherent ml decoder for is well known to be @xmath122 the resulting diversity gain of the code , under ml decoding , is correspondingly given by ( cf .
@xcite ) @xmath123 where the notation @xmath19 accentuates the dependence of the diversity on the multiplexing gain @xmath1 .
one of the main features of the linear dispersive codes , as was noted in the introduction , is that their lattice structure allows for efficient optimal and near optimal
solutions to using the sphere decoding algorithm . using the linearity of the map from @xmath124 to @xmath109 we obtain ( cf . )
@xmath125 where the code - channel generator matrix @xmath126 is given by @xmath127 we can thus , instead of solving directly , equivalently obtain an estimate of @xmath124 through @xmath128 where is an optimization problem suitable for the sphere decoder , and then easily recover @xmath129 from @xmath130 .
the sphere decoding algorithm solves by a branch - and - bound search on a regular tree .
detailed descriptions of the algorithm are found in @xcite and the semi - tutorial papers @xcite , and most implementation issues will not be repeated herein .
however , in order to make our results precise and to introduce notation we need to review some or the key ideas as they apply to . to this end
, note that by the rotational invariance of the euclidean norm it follows that is equivalent to @xmath131 where @xmath132 is the thin qr - decomposition of @xmath126 ( i.e. , @xmath133 is unitary and @xmath134 is upper triangular ) and where @xmath135 .
the sphere decoder solves by enumerating symbol vectors @xmath136 within a given sphere of radius @xmath137 , i.e. , @xmath138 that satisfy @xmath139 if is satisfied for at least one @xmath140 , then also the ml solution must satisfy as the ml solution yields the minimum metric in .
the set of vectors that satisfy is found by recursively considering partial symbol vectors @xmath141 for @xmath142 .
specifically , if @xmath143 is the vector containing the last @xmath144 components of @xmath138 , a necessary condition for to be satisfied is given by @xmath145 where @xmath146 denotes the last @xmath144 components of @xmath147 , and where @xmath148 denotes the @xmath149 lower right corner of @xmath150 .
this follows due to the upper triangularity of @xmath150 .
any set of vectors @xmath114 with common last @xmath144 components that fail to satisfy may be excluded from the set of ml candidate vectors . enumerating all partial symbol vectors that satisfy , beginning with @xmath151 , extending these to @xmath152 and so on , yields a recursive procedure for enumerating all @xmath114 that satisfy .
the enumeration of partial symbol vectors @xmath153 is equivalent to the traversal of a regular tree with @xmath154 layers
one per symbol @xmath155 where @xmath155 is the @xmath144th component of @xmath124 and @xmath156 children per node @xcite .
there is a one - to - one correspondence between the nodes at layer @xmath144 ( the layers are enumerated with the root node corresponding to @xmath157 ) and the partial vectors @xmath153 .
we say that a node is visited by the sphere decoder if and only if the corresponding partial vector @xmath153 satisfies , i.e. , there is a bijection between the visited nodes at layer @xmath144 and the set @xmath158 due to this relation we will in what follows not make the distinction between nodes and partial symbol vectors and simply refer to @xmath159 as nodes at layer @xmath144 when discussing the search .
the total number of visited nodes ( in all layers of the tree ) is given by @xmath160 where @xmath161 is the number of visited nodes at layer @xmath144 of the search tree .
the total number of visited nodes is commonly taken as a measure of the sphere decoder complexity ( see @xcite ) and this will also be done in what follows .
note however that as the total number of flops required for evaluating the bound in may be upper and lower bounded by constants that are independent of @xmath9 @xcite our results relating to the sd complexity exponent would not change if we instead considered @xmath0 to be the number of flops spent by the decoder .
the description of the sphere decoder is not complete without specifying how the search radius is selected . in the interest of obtaining the sd complexity exponent
, we may argue that any reasonable choice of a fixed ( non - random ) search radius should satisfy @xmath162 to see this , it is sufficient to note that the metric in satisfies @xmath163 for the transmitted vector @xmath124 . thus , if @xmath164 the transmitted symbol vector is excluded from the search , resulting in a decoding error . by considering a radius that grows slowly with snr ,
say @xmath165 , it can be shown that @xmath166 for @xmath167 , i.e. , by selecting @xmath168 the probability of excluding the transmitted vector will ( for increasing @xmath9 ) vanish faster than the probability of error and cause vanishing degradation of the overall probability of error . at the same time
, if the radius does not tend to infinity with increasing @xmath9 , it will follow that @xmath169 is bounded away from zero .
this implies a non - vanishing probability of error and a resulting diversity gain of zero , which is clearly undesirable .
thus , as the complexity exponent is not affected by the particular choice of @xmath170 , we shall unless otherwise stated in the following for simplicity assume that @xmath165 , with @xmath168 in order to ensure vanishing degradation to the overall probability of error .
this said , the derived sd complexity exponent would be the same if we considered adaptive radius updates as used in the schnorr - euchner ( se ) implementation @xcite .
this may be shown by following the argument in @xcite , and we give a proof of this statement in the present setting in appendix [ app : radius - updates ] . the sphere decoder complexity , or equivalently the number of visited nodes @xmath0 , is as stated a random variable with a distribution that depends on a number of parameters , e.g. , the system dimensions @xmath171 ,
@xmath40 and @xmath11 , the snr @xmath9 , the multiplexing gain @xmath1 , the generator matrix @xmath121 , and the search radius @xmath172 .
this is well known and follows by the randomness of the bound in .
naturally this randomness must be considered when properly analyzing the sphere decoder complexity , unless one resorts to a worst - case analysis .
however , we argue that the worst - case analysis is unnecessarily pessimistic . in order to illustrate one of the key problems with focusing on the worst - case complexity
consider the event that @xmath173 and @xmath174 . in this case
it is easily seen that and are always satisfied . as a consequence
, the complexity of the sphere decoder would be equal to @xmath175 where we have used that @xmath176 to obtain the size of @xmath177 in .
the worst - case complexity is therefore comparable to that of a full search over @xmath42 as @xmath178 .
however , there is also no point in decoding when @xmath173 as all codewords would yield the same ml metric which in turn implies a high probability of error .
essentially the same argument , for opting out of decoding , can be made whenever the mimo channel is in information theoretic outage @xcite . in this case
it follows by fano s inequality that the probability of decoding error will be bounded away from zero .
in fact , for a code with a diversity gain of @xmath19 any set of channel matrices @xmath179 for which @xmath180 , may be neglected by the decoder with vanishing degradation of the overall probability of error .
however , rather than identifying and excluding a set of bad channel matrices directly , a more pragmatic approach is to impose a run - time constraint on the decoder and ensure that this constraint is such that the probability of it being violated is insignificant in relation to the probability of error .
this leads to the following measure of the decoding complexity , which we will use throughout .
[ def : complexityexponent ] let @xmath181 where @xmath0 is the number of nodes visited by the sphere decoder ( cf . ) .
the _ sd complexity exponent _ is then given by @xmath182 where @xmath19 ( cf . ) is the diversity gain of the code at multiplexing gain @xmath1 . in order to illustrate the operational significance of @xmath23
, we recall that in addition to the instances where the ml decoder makes an incorrect decision , a time - limited sphere decoder can additionally make decoding errors when the search radius is selected too small , i.e. , when @xmath183 ( cf . ) , or when the run - time limit of @xmath17 becomes active , i.e. , when @xmath184 .
these extra errors cause a gap to ml performance which can be quantified as @xmath185 describing the ratio between the probability of error of the time - limited sphere decoder and the ml decoder .
with respect to @xmath23 we then have the following . [ thrm : vanishingnew ]
a sphere decoder with a computational constraint activated at @xmath17 flops , allows for a vanishing gap to ml performance for all @xmath186 , i.e. , @xmath187 the above simply states that for any @xmath188 it is possible to design a decoder based on the sd algorithm that achieves a vanishing snr gap to the ml decoder , at a worst - case complexity of @xmath17 . to see this
apply the union bound to get @xmath189 where the second and third term tend to zero with increasing @xmath9 as the numerator tends to zero at a faster rate than the denominator cf . , .
this immediately translates to a vanishing snr gap to the ml decoder at high snr . in short , the probabilities of the events that the search space is empty or that the complexity of the run - time - unconstrained sphere decoder exceeds @xmath17 are insignificant in comparison to the probability of ml decoding errors .
furthermore , one can not in general time - limit the sphere decoder to @xmath17 for some @xmath190 and expect an arbitrary small gap to ml performance
. specifically , one can show ( cf . )
that @xmath191 for any @xmath190 , and as a result , it follows one can not time - limit the decoder to @xmath17 for any @xmath190 .
the same statement naturally holds whenever @xmath14 is strictly decreasing in @xmath20 .
] for @xmath190 that @xmath192 implying that any attempt to significantly reduce the complexity below @xmath24 will be at the expense of the vanishing snr gap to ml decoding .
we proceed to establish upper and lower bounds on the sd complexity exponent , in essence through the application of a principle ( dating back to gauss ) which states that the number of integer lattice points within a ( large ) set is well approximated by the volume of the set @xcite .
thus , in order to approximate the number of nodes at layer @xmath144 of the search tree , i.e. , the size of @xmath193 defined in , we are primarily concerned with the volume of @xmath194+\sqrt{-1}[-\eta,\eta])^k \cap \eset_k$ ] where @xmath195 is the elliptical set given by ( cf . ) @xmath196 the use of the volume principle for assessing the sphere decoder complexity was previously used in @xcite although its prior use in the communications literature is limited to the case of lattice decoding ( i.e. , where the constellation boundary constraint imposed by @xmath197 $ ] is ignored by the decoder ) .
herein , we have to take the constellation boundary into account to obtain tight bounds on the sd complexity exponent .
the upper and lower bounds presented in this section are essentially obtained in three main steps : 1 ) the volume principle is used to obtain an expression for the number @xmath198 of visited nodes at layer @xmath144 in terms of the singular values of @xmath199 ; 2 ) the singular values of @xmath199 for @xmath142 are related to the singular values of the channel matrix @xmath200 ; and 3 ) the theory of large deviations is used similarly to @xcite to identify random events likely to cause an atypically large decoding complexity . establishing the upper bound on @xmath23 turns out to be easier mathematically .
the reason for this is primarily in the second step where the interlacing property of singular values of sub - matrices @xcite can be used to lower bound the singular values of @xmath199 by the singular values of @xmath200 , to yield results that are universally applicable for any full rank generator matrix @xmath121 , cf .
theorem [ thrm : upper - bound ] .
although the interlacing property gives both upper and lower bounds on the singular values of @xmath199 , the upper bounds are unfortunately not sufficient for establishing tight lower bounds on @xmath23 .
we are therefore forced to develop tighter bounds that depend on some technical assumptions on @xmath121 , cf .
lemma [ lm : lower - bound ] .
while these conditions are , at least in principle , easily verified for any given code design , they are generally hard to verify for arbitrary classes of codes . nevertheless , for some important classes discussed in section [ sec : tightness ] we are able to conclude that the upper bound on @xmath23 is tight , thereby establishing @xmath23 exactly .
the derivation of the upper bound on @xmath23 is given in the following sub - sections , while the derivation of the lower bound , which is similar in spirit to the upper bound but complicated by some technical details , is primarily given in appendix [ app : lower - bound ] , and discussed in section [ sec : tightness ] .
as noted , we begin by establishing bounds on @xmath201 in terms of the singular values of the matrix @xmath199 in , the sphere radius @xmath172 and the constellation size @xmath117 . to this end , consider the following lemma , which corresponds to rigorous applications of the volume principle discussed above .
the proof of the lemma is given in appendix [ app : upper - bound ] .
[ lm : upper - bound ] let @xmath202 be the ellipsoidal set given by @xmath203 where @xmath204 and @xmath205 .
let @xmath206 be the hypercube given by @xmath207 then , the number of integer points contained in the intersection of @xmath208 and @xmath209 is upper bounded as latexmath:[\[\label{eq : lm - upper } and the number of integer points contained in @xmath208 is lower bounded by @xmath211 , @xmath212 denote the singular values of @xmath213 .
although lemma [ lm : upper - bound ] is phrased in terms of real valued quantities , it is easily applied to complex valued sets by considering each complex dimension as two real valued dimensions . in particular , the expression in is equivalent to @xmath214 where @xmath215 and @xmath216 by noting that if @xmath217 is the singular value decomposition ( svd ) @xcite of @xmath199 , then @xmath218 is an svd of @xmath219 , it follows that the singular values of @xmath219 are the same as those of @xmath199 albeit with a multiplicity of @xmath220 .
thus , applying in lemma [ lm : upper - bound ] to @xmath221 ( cf .
) yields an upper bound on the number of nodes visited at layer @xmath144 , which is given as @xmath222 ^ 2\ ] ] where @xmath223 , @xmath224 denote the singular values of @xmath199 . here
, in essence , the additive @xmath225 term accounts for edge effects in the volume approximation , the first term in the minimum accounts for the size of the search sphere , and the second term in the minimum accounts for the constellation boundary .
the lower bound in will be used later in order to assess the tightness of the upper bound on @xmath23 developed next .
the reason for providing a lower bound on @xmath226 and not @xmath227 is that we can not a - priori rule out that @xmath228 in is such that @xmath229 , a case which if not ruled out would lead to the trivial lower bound @xmath230 . the interlacing theorem of singular values of sub - matrices ( cf .
7.3.9 ) and ( * ? ?
* corollary 3.1.3 ) ) states that the singular values of @xmath199 are bounded by the singular values of @xmath150 according @xmath231 where @xmath232 and @xmath233 denotes the @xmath234th singular values of @xmath199 and @xmath150 respectively . as @xmath235 where @xmath236 has a set of orthogonal columns that span the range of @xmath126 it follows that @xmath237 .
further , by the definition of @xmath126 in we have that @xmath238 where @xmath239 due to the assumption that @xmath121 is full rank .
the singular values of @xmath240 are the same as those of the channel matrix @xmath200 in , albeit with a multiplicity of @xmath11 , i.e. , @xmath241 where @xmath242 this can be seen by noting that if @xmath243 is the svd of @xmath200 , then @xmath244 is an svd of @xmath240 ( albeit with a non - standard ordering of the singular values ) .
alternatively , on can apply ( * ? ? ?
* theorem 4.2.12 ) to the eigenvalues of @xmath245 . combining the above
yields a lower bound on the singular values of @xmath199 in terms of the singular values of the channel matrix @xmath200 according to @xmath246 and an upper bound on the number of nodes visited by the sphere decoder at layer @xmath144 according to @xmath247 ^ 2 .\ ] ] in order to bound the probability that the right hand side of is atypically large in the high snr regime , it is useful to consider the snr dependent parameterization of the singular values ( or eigenvalues ) of @xmath248 introduced in @xcite , i.e. , snr dependent random variables @xmath249 , for @xmath250 , defined by @xmath251 note that by this definition @xmath252 . the variables , @xmath253 for @xmath250 are referred to as the _ singularity levels _ of @xmath200 as they give an indication of how close to singular the channel @xmath200 is in relation to the inverse snr @xmath254 .
as @xmath255 , @xmath256 and @xmath257 it holds that @xmath258 ^ 2 \operatorname*{\overset{.}{\leq}}\rho^{\nu_{i}}\ ] ] where @xmath259 by it follows that @xmath260 however , as the snr exponent on the right hand side of is non decreasing in @xmath144 it must for the total number of visited nodes @xmath0 hold that @xmath261 or for any given @xmath262 hold that @xmath263 provided @xmath9 is sufficiently large .
consider now the set ( cf . and ) @xmath264 where @xmath265 . as holds ( asymptotically ) for any @xmath262 , and since @xmath266 implies that @xmath267 for any @xmath268 by and , it follows that @xmath269 equivalently ( cf . ) @xmath270 for @xmath268 .
the value of the bound in is that the right hand side is readily computed using large deviation theory @xcite . a sequence of random vectors @xmath271 parameterized by @xmath9
is said to satisfy the _
large deviation principle _
@xcite with _ rate function _
@xmath272 , @xmath273 if for every open set @xmath274 it holds that @xmath275 and if for every closed set @xmath276 it holds that @xmath277 although not stated formally , one of the central results of @xcite is that the sequence of random variables given by @xmath278 ( cf . )
satisfies the large deviation principle with rate function ( see the proof of theorem 4 in @xcite ) @xmath279 if @xmath280 and @xmath281 otherwise .
this observation was key in establishing the dmt in @xcite . by combining with , and noting that @xmath282 is a closed set
, it follows that @xmath283 for any @xmath268 .
as @xmath284 for all @xmath285 it follows that @xmath286 is non - decreasing and it can additionally be verified that @xmath286 is left - continuous , i.e. , @xmath287 which implies that @xmath288 from it follows that the complexity exponent @xmath23 is upper bounded by @xmath27 where @xmath289 and where the last equality follows as @xmath290 is non - decreasing .
further , by the left - continuity of @xmath290 it follows that the supremum on the right is attained , i.e. , the supremum can be replaced by a maximum .
note however that the condition that @xmath291 is satisfied if and only if there exist an @xmath292 such that @xmath293 .
thus , @xmath27 in could alternatively be obtained as the solution to a constrained maximization problem according to [ eq : opt1 ] @xmath294 where follows from the constraint @xmath292 , and where and follows from @xmath293 .
it is straightforward to show that the optimal @xmath20 in must be such that is satisfied with equality . by further noting that the sum in contains only @xmath40 distinct terms , each with multiplicity @xmath11
, it can be seen that @xmath295 where we have also used the full rate assumption that @xmath101 .
we summarize the above in the following theorem .
[ thrm : upper - bound ] the sd complexity exponent @xmath23 of decoding any full rate linear dispersive code with multiplexing gain @xmath1 and diversity @xmath19 is upper bounded as @xmath296 where [ eq : optimization ] @xmath297 the upper bound given by theorem [ thrm : upper - bound ] can naturally be computed given explicit values for the multiplexing gain @xmath1 and diversity gain does not assume a diversity optimal code . ] @xmath19 .
however , it is also possible in some cases to give general solutions as a function of @xmath1 when the dmt curve @xmath19 of the code is known explicitly . in particular
, dmt optimal codes such as those presented in @xcite have a diversity gain of @xmath298 at any integer multiplexing gain @xmath299 @xcite . in this case
it is straightforward to verify that an is constant in @xmath64 for @xmath300 and @xmath301 . ]
optimal @xmath302 in is given by @xmath303 and @xmath304 to see this , note that the objective function in is symmetric with respect to permutations of the set of @xmath249 for @xmath250 .
as the sum in the diversity constraint places more weigh on @xmath305 than on @xmath249 , for @xmath306 , it is optimal to increase @xmath307 until the term in containing @xmath307 saturates ( i.e. , when @xmath308 ) , then to increase @xmath309 etcetera , until the constraint is satisfied with equality .
this yields the aforementioned solution .
note also that @xmath310 are the same singularity levels that give the typical outages in @xcite , ( cf .
section [ sec : outages ] ) . inserting the optimal solution into yields
@xmath311 which is a remarkably simple upper bound on the sd complexity exponent of decoding any dmt optimal code at an integer multiplexing gain @xmath299 . for a dmt optimal code at a possibly non - integer value of @xmath1 , let @xmath144 be the integer such that @xmath312 , i.e. , @xmath313 .
the optimal solution is in this case given by @xmath314 @xmath315 and @xmath316 substituting the above solution back into yields @xmath317 we summarize the above in the following theorem .
[ thrm : upper - bounddmtopt ] the sd complexity exponent @xmath23 of decoding any dmt optimal full rate linear dispersive code with integer multiplexing gain @xmath299 is upper bounded as @xmath318 for general @xmath1 where @xmath319 the sd complexity exponent @xmath23 is upper bounded as @xmath296 where @xmath320 the function @xmath27 in is a piecewise linear function in @xmath1 , although slightly more involved than the set of straight lines describing the optimal dmt @xmath19 . for @xmath39
the function in coincides with the curve for @xmath23 shown in fig .
[ fig : exponents ] .
we now turn to specific cases where we can exactly establish the sd complexity exponent @xmath23 by establishing that the upper bound @xmath296 is in fact tight . to this end , we begin with the following lemma , which provides a sufficient condition for @xmath29 , i.e. , for the tightness of the upper bound .
[ lm : lower - bound ] let @xmath321 be the matrix consisting of the first @xmath322 columns of the generator matrix @xmath323 . if there exists , for @xmath324 , unitary matrices @xmath325 such that @xmath326 then @xmath29 for all @xmath327 $ ] , where @xmath27 is given by in theorem [ thrm : upper - bound ] .
the proof of lemma [ lm : lower - bound ] is similar in spirit to the proof of theorem [ thrm : upper - bound ] , although riddled with technical details , and therefore relegated to appendix [ app : lower - bound ] .
in essence , the condition posed in implies that there are certain orientations of the right singular vectors of the channel @xmath200 ( in relation to the code generated by @xmath121 ) for which the lower bound in is sufficiently tight .
details are provided in appendix [ app : lower - bound ] , and some additional discussions of and the general applicability of the lemma can be found in section [ sec : discussions ] . however , we first apply lemma [ lm : lower - bound ] to find @xmath23 in some very important special cases . to this end , it is useful to first note that permuting the columns of @xmath121 , i.e. , replacing @xmath121 with @xmath328 where @xmath329 is a permutation matrix , does not change the code @xmath42 . instead , the effect such a permutation would have is that it would change the order in which the symbols in @xmath124 are enumerated by the sphere decoder described in section [ sec : sphere - decoder ] ( cf .
* section iv ) ) . in the present context
, the first @xmath330 columns of @xmath328 , i.e. , @xmath331_{|p}$ ] , may differ from those of @xmath121 .
thus , we see that depends not only on the code itself , but also on the order in which the constituent symbols @xmath332 are enumerated by the sphere decoder ( cf .
@xcite where the topic of column ordering is discussed in detail ) . in the context of lemma [ lm :
lower - bound ] , it can be seen that as @xmath333 has rank @xmath330 for any unitary @xmath334 due to the full rank assumption on @xmath121 .
one can therefore select @xmath330 linearly independent columns , or equivalently , find a permutation matrix @xmath335 such that @xmath336_{|p}$ ] has full rank . using a similar argument
, we can recursively construct a ( single ) @xmath335 for which there are @xmath334 for @xmath324 satisfying @xmath337_{|p } \big ) = pt\ ] ] by constructing @xmath338 from @xmath334 by removing a column , selecting the appropriate columns from @xmath121 , and starting the recursion with an arbitrary @xmath339 . interpreting the above in light of theorem [ thrm : upper - bound ] and lemma [ lm : lower - bound ]
we can thus establish that for any full rate linear dispersive code design , @xmath27 as defined in theorem [ thrm : upper - bound ] and given in theorem [ thrm : upper - bounddmtopt ] for dmt optimal codes , is the tightest upper bound on the sd complexity exponent that can possibly hold under arbitrary column orderings .
this is formalized in the following .
[ thrm : perm - tightness - fullrate ] given any full rate linear dispersive code achieving diversity @xmath19 , there is always at least one column ordering for which @xmath29 .
however , while theorem [ thrm : perm - tightness - fullrate ] is useful in the sense that it tells us that one could not improve upon the tightness of @xmath27 without introducing further assumptions regarding the particular code design considered , it is obviously not of practical interest to use the worst possible column ordering . therefore , we turn our attention to the important class of threaded codes @xcite for which we will show that the _ natural _ column ordering @xmath340 implies @xmath29 .
the class of threaded code designs is of particular interest , as it includes full rate codes that perform very well in a variety of settings .
the threaded algebraic space - time ( tast ) codes @xcite , codes constructed from cyclic division algebras ( cdas ) @xcite , and specifically modified cda codes @xcite that were shown ( cf .
@xcite ) to achieve the entire dmt , are prime examples .
the cda based threaded designs are also the only currently known explicit constructions capable of achieving the dmt for all values of @xmath40 and simultaneously over all @xmath327 $ ] .
all these codes have a common threaded structure .
specifically an @xmath57 threaded code is built from @xmath48 component codes mapped cyclically in threads ( or layers ) to the codewords @xmath94 .
for example , in the special case of @xmath341 , the thread structure is given by @xmath342 where the numbers @xmath343 indicate the thread to which a particular entry of @xmath94 belongs .
in general , symbol @xmath344 in thread @xmath345 is mapped to @xmath346_{j , k}$ ] where @xmath347 and where @xmath348 denotes the modulo @xmath48 operation .
for example , in the case of perfect codes
@xcite which also employ a threaded structure , the code follows from @xmath349 where @xmath350 are full rank diagonal matrices incorporating a properly chosen thread - separating scalar @xmath351 , where @xmath352 is a ( unitary ) full rank generator matrix for the component code of each thread , @xmath353 are the constellation symbols of thread @xmath345 , and where @xmath354 denotes the matrix to vector operation obtained by stacking the elements of @xmath94 according to their thread ( cf . the column based stacking of the @xmath355 operation ) .
regarding the cost of decoding by such codes , we note that the corresponding generator matrix @xmath356 is related to @xmath357 through a permutation of the rows ( cf . ) in such a way that the @xmath358th block @xmath359 of @xmath121 contains exactly one non zero row which it self is one of the rows of @xmath360 .
consequently , @xmath361 has rank @xmath362 and contains exactly @xmath362 non - zero rows .
this holds for any @xmath48 and @xmath363 .
now , let @xmath364 be a unitary matrix with the property that any @xmath362 rows of @xmath334 are linearly independent .
such matrices can clearly be constructed , and an example is the matrix that contains the first @xmath362 discrete fourier transform ( dft ) vectors of length @xmath48 .
let @xmath365 be the matrix containing only the non - zero rows of @xmath366 , and let @xmath367 be the full rank matrix consisting of the rows of @xmath334 matching the non - zero rows of @xmath368 .
it follows that @xmath369 is full rank as both @xmath370 and @xmath371 are full rank and square matrices .
note also that the same argument can be made regardless of the ordering of the threads , and for any other code with a threaded structure , provided the symbols in @xmath124 are grouped into layers as in .
this is stated in the following .
[ thrm : layered ] the sd complexity exponent , given any threaded code with @xmath372 that is decoded with the natural column ordering or under any other threat - wise grouping , is @xmath29 where @xmath27 is given in theorem [ thrm : upper - bound ] . consequently directly from theorems [ thrm : upper - bounddmtopt ] and [ thrm : layered ]
, we have the following result for dmt optimal threaded codes .
[ thrm : layereddmt ] sphere decoding with thread - wise grouping of any dmt optimal threaded code with @xmath372 , achieves dmt optimality with a sd complexity exponent of @xmath373 which , for integer values of @xmath299 , simplifies to @xmath374 we briefly note that as expected the complexity increases with increasing @xmath375 for any fixed @xmath1 which is quite natural as the size of the codebook @xmath42 and the signal space dimension increase .
one can however also note that @xmath23 is independent of the number of receive antennas @xmath171 ( provided @xmath93 ) .
this is specific to the dmt optimal behavior and threaded structure of the codes , and may be explained by the fact that even though the channel quality generally improves by adding receive antennas - thus generally reducing complexity - the same improvement also occurs in the error probability performance of the code , and these two effects cancel each other in the sd complexity exponent .
we here go one step further and identify a class of codes for which we can state , without limitations on the actual code structure , that @xmath29 for any column ordering . in particular , we establish this for the class of all @xmath30 approximately universal codes , i.e. , all minimum delay codes that can achieve dmt optimality over the @xmath377 channel irrespective of fading statistics @xcite .
this is accomplished , albeit only for the specific case of @xmath378 , by proving that follows from the so called non - vanishing determinant ( nvd ) condition @xcite which is well known to be a necessary and sufficient condition for approximate universality .
we consequently have the following . [ thrm : nvd ] any @xmath376 full rate approximately universal linear dispersive code , irrespective of its structure , introduces a sd complexity exponent of @xmath379 as the nvd property does not depend on the ordering of the columns of @xmath121 , it also follows that the conclusion of theorem [ thrm : nvd ] holds irrespective of the column ordering .
[ cc]0 [ cc]1/2 [ cc]1 [ cc]3/2 [ cc]0 [ cc]1 [ cc]2 [ cc]3 [ cc]4 [ lb]@xmath380 [ lb]@xmath381 [ lb]@xmath382 [ bc]@xmath383 [ cc]@xmath384 [ cc]@xmath385 we recall that the claim of theorem [ thrm : upper - bound ] may be expressed in terms of the function ( cf . ) @xmath386 which provides a conditional , asymptotic , upper bound on the sphere decoding complexity according to @xmath387 in terms of the multiplexing gain @xmath1 and the singularity level @xmath302 .
the final upper bound @xmath27 in is then given as the worst - case @xmath383 over all singularity levels that occur with a probability that is larger than or equal to the probability of error , given asymptotically by the diversity gain @xmath19 of the code .
the characterization of the dmt in @xcite relies on the asymptotic probability of outages at high snr , i.e. , the probability that the i.i.d .
rayleigh fading awgn channel given by @xmath388 with a power constraint @xmath389 can not support an asymptotic data - rate of @xmath8 . as was shown in @xcite ,
this occurs when the singularity levels belong to the outage set @xmath390 , and the diversity of the outage event is given by the most likely set of singularity levels that satisfy this condition , i.e. , @xmath391 where @xmath392 is given in .
if we restrict attention to the set of singularity levels whose probability of occurring does not vanish exponentially fast , i.e. , for which @xmath393 or equivalently @xmath394 , we can for dmt optimal codes make an interesting connection between the decoding complexity and information theoretic outages . in particular , as @xmath391 it follows that @xmath395 if and only if @xmath396 .
we can thus , for dmt optimal codes , equivalently express according to [ eq : outage - interpretation ] @xmath397 which may be interpreted as the worst - case complexity ( bound ) over all channels that are not in outage .
this significantly strengthens the connection between channel and decoding outages touched upon in section [ sec : decoding - complexity ] .
the concept is illustrated for @xmath398 in fig .
[ fig : outages ] where @xmath383 is plotted as a function of @xmath399 over @xmath400 . in this case ,
@xmath401 . note also here that we know by theorem [ thrm : nvd ] that @xmath29 .
singularity levels that are in the outage region @xmath402 are shown in a darker shade .
it can be seen that increasing the multiplexing gain @xmath1 increases the codeword density and codebook size and consequently broadens the set of singularity levels that can potentially lead to higher complexity . however , increasing the multiplexing gain also reduces the set of channels that support the data - rate , thus limiting the set of singularity levels for which the decoder needs to be applied , leading to an overall reduction in the sd complexity exponent as @xmath1 approaches its maximum value .
further , the connection to information theoretic outages allows for an intuitive explanation of the result of theorem [ thrm : layereddmt ] , and in particular . to this end , it is illustrative to consider a heuristic argument involving low rank channel matrices @xmath200 . as noted in @xcite , the typical outages at integer multiplexing gains @xmath299 are caused by channels that are close to the set of rank @xmath144 matrices , i.e. , that have @xmath403 small singular values .
if we for the purpose of illustration assume that @xmath200 has rank @xmath144 , it follows that @xmath240 for @xmath404 , and @xmath126 , has rank @xmath405 , and consequently a null - space of dimension @xmath406 .
this implies that the @xmath407 lower right block of @xmath150 is columns of @xmath150 are linearly independent .
in fact , the rigorous treatment of this technical detail is largely responsible for much of the difficulty in establishing the lower bounds on @xmath23 .
in particular , condition in lemma [ lm : lower - bound ] guarantees that this happens with sufficiently high probability , while lemma [ lm : perturbation ] provides a perturbation analysis that allows us to extent this intuitive reasoning to channels that are close to the set of rank deficient matrices .
] identically equal to zero , and the sphere decoder pruning criteria become totally ineffective up to and including layer @xmath406 .
as the size of @xmath177 is @xmath408 for @xmath299 , we see that the number of nodes searched at layer @xmath406 of the sd search tree is @xmath409 ( cf . ) . in order to ensure close to optimal performance
, the sphere decoder must be able to decode for channels where @xmath403 singular values are close to zero .
however , channels with even more singular values close to zero occur with a probability that is small in relation to the outage probability or the probability of ml decoder error , and can thus be safely ignored by the decoder .
it is perfectly conceivable that the sphere decoding complexity may rise under specific codes and under specific fading statistics that tend to regularly introduce channel instances that are difficult to decode for .
a natural question is then whether one can bound the complexity , irrespective of the code and of the statistical characterization of the channel . considering theorem [ thrm : upper - bound ] , and the proof of this theorem , we can see that the i.i.d .
rayleigh fading assumption only enters through the rate function @xmath392 .
consequently , we may directly restate theorem [ thrm : upper - bound ] for other fading distributions after updating and with the appropriate rate - function @xmath392 . some relevant examples of rate - functions for other fading distributions are given in @xcite .
further , regardless of which @xmath392 applies , the upper bound @xmath27 in theorem [ thrm : upper - bound ] is non - decreasing in @xmath19 and maximized when @xmath19 corresponds to the outage exponent . in this case
@xmath27 is , again , given by , which does not explicitly depend the fading distribution other than through the assumption that @xmath410 vanishes exponentially fast ; an assumption that holds for all reasonable distributions .
this implies that sd complexity exponent is universally upper bounded as ( cf . theorem [ thrm : upper - bounddmtopt ] ) @xmath411 for any full - rate code and statistical characterization of the channel .
this is also clearly the tightest upper bound that can hold for all ( full - rate ) codes and fading statistics . in @xcite a family of dmt optimal @xmath378 space - time codes called fast decodable codes
@xcite were constructed . the sd complexity exponent ( and also its upper bound )
provides an interesting approach for comparing the complexity of decoding regular codes and fast decodable codes . before doing so
it should be noted that these fast decodable codes are not , strictly speaking , of the form in as the real and imaginary part of each constituent symbol is dispersed separately .
nevertheless , the fast decodable codes may be decoded by an equivalent _ real valued _ sphere decoder that performs a search over a tree with @xmath412 layers and @xmath413 branches per node , and we can compare the reported worst - case complexity of this real valued sphere decoder to the complexity of the complex valued sphere decoder considered herein .
the fast decodable codes have the appealing property that the upper right @xmath414 block of the real valued @xmath415 ( cf . )
is always a diagonal matrix , regardless of the particular realization of @xmath200 .
while the regular real valued sphere decoder for a @xmath376 full rate code would perform a ( bounded ) search over the entire tree , it is sufficient for the fast decodable codes to ( without loss of optimality ) perform a search over only the @xmath416 first layers , and extend each node at layer @xmath416 to a valid codeword through a faster , linear , ml decoding .
this simplified version of the real valued sphere decoder can be viewed as a search over a regular tree where each node has @xmath413 children up to layer @xmath416 , but only one child per node for the @xmath416 remaining layers .
consequently , the worst - case number of nodes visited by the simplified sphere decoder is @xmath417 as opposed to @xmath418 for the regular real valued sphere decoder , cf .
thus , fast decodability implies a reduction by a factor of 2 in the worst - case snr exponent , which is significant at high snr .
[ bc]@xmath1 [ bc]@xmath419 [ bc]@xmath23 [ bc]snr exponent [ c]multiplexing gain @xmath1 , with the ( worst - case ) snr exponent @xmath420 of the regular sphere decoder , and with the maximal snr exponent @xmath421 of the simplified sphere decoder.,title="fig:",width=313 ] however , this worst - case snr exponent of the simplified sd algorithm should be viewed in light of the sd complexity exponent induced by any @xmath376 approximately universal code as given by theorem [ thrm : nvd ] , i.e. , @xmath422 .
the worst - case snr exponent of the regular sphere decoder and of the simplified sphere decoder , and the sd complexity exponent @xmath23 are shown in fig .
[ fig : fastdec ] . for multiplexing gains
lower than or equal to @xmath423 , the sd complexity exponent and the worst - case snr exponent of the simplified sphere decoder actually coincide , while the sd complexity exponent is strictly lower for multiplexing gains higher than one .
the interpretation is that a run - time constrained sphere decoder will yield asymptotic ml performance for _ any _ @xmath376 approximately universal code at a complexity that is comparable to the reported worst - case complexity of the fast decodable codes at low multiplexing gains , and significantly better at high multiplexing gains .
thus , in a sense , all approximately universal codes are fast decodable at high snr .
however , we hasten to add that the fast decodable structure can naturally still be desirable in many cases of practical interest .
finally , we discuss the application of lemma [ lm : lower - bound ] to codes not considered herein . to this end , note that for any given generator matrix @xmath121 of some code not covered by section [ sec : tightness ] , it should be clear that if holds then lemma [ lm : lower - bound ] could be used to establish a tight lower bound on @xmath23 .
this said , we also wish to caution the reader that only represents a sufficient condition for @xmath29 .
it does not necessarily follow that @xmath424 if is not true .
in other words , the question of if there are code designs that improve upon the bound @xmath27 is not answered in the positive by finding code designs for which does not hold . as for testing it
should also be noted that one does not have to restrict the search for @xmath334 to the set of unitary matrices .
any full rank matrix @xmath425 can be factored , e.g. , by the qr decomposition , as @xmath426 where @xmath427 has rank @xmath362 and where @xmath334 is unitary .
hence , as @xmath428 is full rank , it follows that @xmath429 is rank deficient if and only if fails to hold .
hence , the statement of lemma [ lm : lower - bound ] could be phrased in terms of the existence of any @xmath334 , not necessarily unitary .
further , let @xmath430 where @xmath431 denotes the determinant , and note that @xmath432 is a polynomial in the elements of @xmath52 .
it can thus be seen that if @xmath433 for some @xmath425 , i.e. , @xmath432 is not the zero polynomial , it follows that the set of @xmath52 for which @xmath434 has zero lebesgue measure .
it is then a straightforward extension to show that holds either never or almost always with respect to the set of unitary matrices @xmath334 over the stiefel manifold ( i.e. , the set of all unitary @xmath435 matrices ) endowed with the haar ( uniform ) measure .
this suggests a rather interesting conceptual method for verifying .
given a specific generator matrix one could at least in theory test the condition of lemma [ lm : lower - bound ] by selecting @xmath334 ( or @xmath52 ) uniformly at random , and the condition of lemma [ lm : lower - bound ] would be proven with probability one if true .
however , finite precision computations will limit the practical applicability of such an approach , although symbolic computations could potentially be a way to test a specific code design .
the work addressed the open question of identifying the computational cost of near - ml sphere decoding . in the high - snr high - rate regime ,
the introduced sd complexity exponent asymptotically described this cost , concisely revealing the cost s natural dependencies to the codeword density , the codebook size , as well as to the snr , dimensions and fading characteristics of the wireless channel .
this exponent currently sets the bar with respect to the computational reserves required for decoding with arbitrarily close to ml performance , and the clear challenge is now to identify transceivers with a lesser complexity exponent that can still guarantee a vanishing ml gap .
the simplicity of the provided guarantees can offer insight into designing robust encoders , decoders , and time - out policies , as well as guidelines for network planning in settings where rate , reliability , and computational complexity are principal intertwined concerns .
such guarantees can apply towards substantial savings in energy , processing power , and hardware .
[ c]@xmath436 [ c]@xmath437 [ c]@xmath438 [ c]@xmath439 [ c]@xmath208 [ c]@xmath209 in the case of @xmath31 .
the lemma provides an upper bound on the number of integer points within the shaded area , corresponding to the intersection of the ellipsoid and the constellation boundary.,title="fig:",width=313 ] in the following , we provide a proof of lemma [ lm : upper - bound ] , starting with the upper bound in and then establishing the lower bound in . to this end , note that the length of the @xmath234th semi - axis of @xmath208 , denoted @xmath440 , is given by @xmath441 let @xmath436 be the smallest orthotope ( box ) , aligned with and containing @xmath208 , i.e. , @xmath436 is an orthotope with side lengths @xmath442 ( see fig . [
fig : lattice ] ) .
let @xmath437 be a hypercube with side - length @xmath443 , centered at the origin and aligned with @xmath436 ( see fig .
[ fig : lattice ] ) .
as the diagonal of @xmath209 is @xmath443 it follows that @xmath444 , regardless of the orientation of @xmath437 .
let @xmath438 be given by @xmath445 and note that @xmath446 as @xmath447 and @xmath444 .
as @xmath436 and @xmath437 are aligned , it follows that @xmath438 is also an orthotope .
let @xmath448 denote the side - lengths of @xmath438 and note that @xmath449 .
the mean value theorem @xcite states that for any convex body ( or set ) @xmath450 it holds that @xmath451 where @xmath452^n\ ] ] denotes the unit cube in @xmath453 , i.e. , the volume of the set equals the average number of integer points in the set when perturbed by a uniform random perturbation over the unit cube @xcite .
this statement is a rigorous version of the intuitive notion that the number of integer points in @xmath454 may be approximated by its volume , and it can be used to obtain a non - random upper bound on the number of integer points in @xmath455 . to this end
, let @xmath439 be the orthotope , aligned with and centered around @xmath438 , with side lengths @xmath456 ( see fig . [
fig : lattice ] ) . by construction
, it follows that @xmath457 for _ any _ @xmath458 .
it therefore follows by that @xmath459 \ , , \ ] ] where @xmath460 as @xmath446 the upper bound in follows . in order to establish the lower bound in we
may redefine @xmath436 to be the orthotope , aligned with the semiaxes of @xmath208 and with the same center , having side - lengths @xmath461 for @xmath212 .
now , by construction @xmath462 which implies that @xmath463 .
let @xmath437 be another orthotope , aligned with @xmath436 and with side - lengths @xmath464 .
it follows that @xmath465 for any @xmath458 . by reasoning similar to what is used in the proof of the upper bound ( cf . )
it follows that @xmath466 which establish the lower bound in .
let @xmath467 be an optimal point of and let @xmath468 be the largest integer for which ( cf . ) @xmath469 note that we can without loss of generality assume that @xmath470 as otherwise @xmath471 ( cf . ) and @xmath29 would be a trivial statement .
it follows that @xmath472 for @xmath473 and we may also without loss of generality assume that @xmath474 for @xmath250 as the objective in does not increase in @xmath253 beyond @xmath475 .
the goal will be to show that layer @xmath476 of the sphere decoder contains close to @xmath477 nodes with a probability that is large with respect to the probability of decoding error @xmath478 . to this end ,
let @xmath479 be the singular value decomposition of @xmath200 , where @xmath480 and where @xmath61 .
let @xmath481 denote the last @xmath482 columns of @xmath483 ( corresponding to the @xmath362 largest singular values ) and let @xmath265 be the random vector of singularity levels given by .
now , consider the set of conditions ( or events ) given by [ eq : conditions ] @xmath484 for some given ( small ) @xmath262 , @xmath485 for some given @xmath486 , @xmath487 and @xmath488 note also that by choosing @xmath489 sufficiently small , we may without loss of generality assume that @xmath490 implies that @xmath491 for all @xmath250 .
the following proof is structured as follows : first , in sections [ sec : boundary - condition ] and [ sec : eigenvalue - bound ] , it is established that represent sufficient conditions for the number of nodes @xmath492 visited in layer @xmath493 to be close to @xmath477 .
then , in section [ sec : probability ] , it is established that the set of conditions in are simultaneously satisfied with a probability that is large with respect to the probability of error .
we begin by proving that , given , the constellation boundary may be ignored , i.e. , that @xmath177 may be replaced by @xmath104 in without changing the result , thus making the lower bound in lemma [ lm : upper - bound ] applicable . to this end , let @xmath494 be an arbitrary point in the @xmath144-dimensional infinite constellation ( i.e. , the gaussian integer lattice ) and assume that @xmath143 satisfies the sphere constraint at layer @xmath144 , i.e.@xmath495 note that @xmath496 , where @xmath153 denotes the last @xmath144 components of the transmitted symbol vector @xmath114 and where @xmath497 denotes the last @xmath144 components of @xmath498 .
it follows that @xmath499 which implies that @xmath500 and @xmath501 by the interlacing property of singular values ( cf . )
it further follows that @xmath502 where we recall that @xmath503 is the power scaling and @xmath504 , and where the last inequality is implied by and @xmath505 .
as @xmath255 and @xmath506 by it follows that @xmath507 by , , @xmath508 and since @xmath509 , it follows that @xmath510 given that @xmath9 is sufficiently large .
this implies that @xmath141 .
thus , any integer point that satisfies the sphere constraint must also belong to the constellation , and we can proceed using to lower bound the complexity .
we proceed to provide bounds on the singular values of @xmath511 in order to lower bound the number of nodes visited in layer @xmath493 . however , as stated previously the interlacing theorem is , unlike in the derivation of the upper bound , not sufficient for our purposes . instead , we consider the following lemma , proven in appendix [ app : perturbation ] .
[ lm : perturbation ] let @xmath70 , @xmath60 be an arbitrary matrix and @xmath512 be the qr decomposition of @xmath52 .
partition @xmath52 , @xmath236 and @xmath150 according to @xmath513 where @xmath514 and @xmath515 .
then , assuming that @xmath516 for @xmath224 , it holds that @xmath517 \sigma_{i}(\bfa ) \ , .\ ] ] applied to the effective channel matrix @xmath126 it follows that @xmath518 \sigma_{i}(\bfm)\ ] ] where @xmath519 contains the first @xmath330 columns ( @xmath520 ) of @xmath126 , assuming that @xmath521 as will be shown for @xmath522 later . in order to lower bound @xmath523 note
that @xmath524 where @xmath525 denotes the first @xmath330 columns of @xmath121 and @xmath526 as @xmath527 where @xmath334 denotes the matrix containing the @xmath362 singular vectors corresponding to the @xmath362 largest singular values , and where @xmath528 denotes that @xmath529 is positive semi - definite , it follows that @xmath530 considering the smallest singular value of @xmath531 yields @xmath532 and concequently @xmath533 where the first inequality follows by and the last inequality follows by together with @xmath534 and as @xmath486 is fixed ( independent of @xmath9 ) .
further , @xmath535 where @xmath536 , and it follows from that @xmath537 for @xmath522 . as @xmath538 for @xmath473 , it follows by comparing and that @xmath539 for @xmath522 , given that @xmath489 is sufficiently small and that @xmath9 is sufficiently large , making lemma [ lm : perturbation ] applicable for @xmath493 . for the maximal singular value of @xmath126 we have ( cf . )
@xmath540 where the last inequality follows as @xmath541 .
combined with it follows that @xmath542 \operatorname*{\overset{.}{\leq}}\rho^{\frac{1}{2}\delta } \
, , \ ] ] and from and that @xmath543 for @xmath522 . consequently ( cf . ) , @xmath544 given that @xmath489 is sufficiently small .
further , as @xmath545 for @xmath224 where @xmath546 by the condition for @xmath468 in , it follows that the lower bound in tends to infinity with increasing @xmath9 provided that @xmath489 is small , and we may conclude that @xmath547 ^ 2 \operatorname*{\overset{.}{\geq}}\rho^{\frac{rt}{\kappa}+\alpha_{\iota_{t}(i)}^\star - 1 - 3\delta } > 0\ ] ] where the last inequality holds , again , provided @xmath489 is small . combining in lemma [ lm : upper - bound ] and , and making the real valued expansion as we did for theorem [ thrm : upper - bound ] , yields a lower bound on the number of nodes visited by the sphere decoder in layer @xmath493 given by @xmath548 where @xmath549 by noting that @xmath550 for @xmath473 by the assumption that @xmath551 and the definition of @xmath468 , it follows that @xmath552 for @xmath473 . further , as @xmath553 for @xmath554 , also by the definition of @xmath468 , the right hand side of is equal to @xmath555 for @xmath554 .
thus , it follows that @xmath556 where the last equality follows due to the optimality of @xmath557 in , and we then obtain from that @xmath558 given that @xmath9 is sufficiently large and that @xmath262 is small .
however , as @xmath262 can be chosen arbitrarily small it is concluded that represents sufficient conditions under which the number of nodes visited is arbitrarily close to the upper bound of @xmath477 given by theorem [ thrm : upper - bound ] .
we now turn to the probability that the conditions imposed by are simultaneously satisfied .
the events in are independent and @xmath559 follows by the i.i.d . rayleigh assumption on @xmath200 , which make the singular values and singular vectors of @xmath248 independent @xcite . ] .
as imply @xmath560
it follows that @xmath561 given that @xmath9 is sufficiently large .
the assumption made in lemma [ lm : lower - bound ] , i.e. , condition , guarantees that @xmath562 for some @xmath481 .
however , by the continuity of singular values @xcite it follows for sufficiently small @xmath486 ( cf . ) that @xmath563 , which implies @xmath564 as @xmath559 is independent of @xmath9 .
the same is true for @xmath565 , i.e. , @xmath566 .
it may also be shown that @xmath567 converges to a strictly positive limit in which case the the subset of the constellation defined by @xmath568 contains an asymptotically deterministic and strictly positive fraction of the full constellation , cf . the proof of lemma 1 in @xcite .
when @xmath569 the statement that @xmath27 is tight is trivial as @xmath570 .
] and that therefore @xmath571 .
it follows that @xmath572 the probability of @xmath573 may again be assessed by using large deviation techniques as in @xcite . in particular , it is noted that the condition imposed by @xmath573 ( cf . ) specifies an open set of admissible @xmath302 . applying and yields @xmath574 where the second inequality follows from and the feasibility of @xmath557 .
thus , @xmath575 by the definition of the sd complexity exponent @xmath23 ( cf . )
it follows by that @xmath576 .
as the bound holds for arbitrarily small @xmath262 , it follows that @xmath29 , establishing the tightness of and lemma [ thrm : upper - bound ] .
the derivations above make the assumption that the search radius @xmath172 is a non - random function of @xmath9 that satisfies @xmath255 .
it is thus natural to ask if the sd complexity exponent could potentially be improved by choosing @xmath172 adaptively based on the problem data @xmath200 and @xmath577 , as is done when using , e.g. , the schnorr - euchner sd algorithm implementation @xcite .
however , we will show here that it can not , and therefore that the assumption of a non - adaptive radius is made without loss of generality .
the argument is similar to the one in @xcite . to this end
, we note that even if @xmath172 is adaptively chosen , it can not be chosen smaller than the distance to the closest codeword , i.e. , the ( square root of the ) minimum metric in , because otherwise no codeword will be chosen by the search .
as was already argued in section [ sec : search - radius ] , the distance to the _ transmitted _ codeword is @xmath578 and @xmath579 can be made arbitrarily close to one by appropriately choosing @xmath580 .
consequently , whenever the transmitted codeword @xmath124 is the ml decision , i.e. , yields the minimum metric in , we could use @xmath581 where @xmath582 as an ( arbitrarily likely ) probabilistic lower bound on an adaptive search radius @xmath172 in the proof of the lower bounds on the complexity ( e.g. , in ) and obtain the same snr exponents in these bounds .
thus , to complete the argument we must only rule out the possibility that the probability that @xmath124 yields the minimum metric in under the conditions of @xmath583 imposed in is small , as the lower bound is derived explicitly under @xmath583 . to this end , assume that it is false , i.e. , that @xmath584 for any ( fixed and snr independent ) @xmath580 . in this case
we could lower bound the error probability of the ml decoder according to ( cf . )
@xmath585 which would violate the definition of @xmath19 as the diversity order of the ml decoder .
consequently , at sufficiently high snr it must hold that @xmath586 for any @xmath580 .
we can thus choose @xmath580 such that , under @xmath583 , it follows that @xmath587 and @xmath588 with arbitrary high probability , implying that @xmath589 where @xmath590 is the minimum metric in . in other words
, there is some @xmath580 for which @xmath591 completing the proof of lemma [ lm : lower - bound ] with @xmath582 in place of @xmath172 ( as @xmath592 throughout the search ) proves that @xmath29 under also if we allow for sd implementations that adaptively choose and update the search radius @xmath172 .
finally , it should be noted that what is shown here is not that adaptively choosing the search radius can not reduce complexity
it does but only that this reduction is not significant enough to reduce the complexity exponent .
consider the matrix , @xmath593 , given by @xmath594 where @xmath595 , and where @xmath483 and @xmath596 denote the right and left singular vector of @xmath52 respectively .
partition @xmath597 according to @xmath598 where @xmath599 and @xmath600 . by the nature of the qr decomposition
, it holds that @xmath601 where @xmath602 denotes the projection onto the orthogonal complement of the range of @xmath603 ( i.e. the nullspace of @xmath604 ) . additionally , let @xmath605 as @xmath606 is a unitary matrix it follows that @xmath607 in what follows
, we will consider the singular values of @xmath608 in order to establish the lemma . to this end , we will make use of two results due to weyl and stewart . for a modern proof of theorem [ thrm : weyl ] , see e.g. ( * ? ? ?
* corollary 7.3.8 ) .
the statement in theorem [ thrm : stewart ] follows by combining ( * ? ? ?
* theorem 2.3 ) and ( * ? ? ? * theorem 2.4 ) . in the following
, @xmath609 denotes the spectral matrix norm .
[ thrm : weyl ] for arbitrary @xmath610 it holds that @xmath611 for @xmath610 such that @xmath612 , @xmath613 where @xmath614 denotes the moore - penrose pseudo inverse @xcite . by noting that @xmath615 for @xmath616 and using the assumption that @xmath617 it follows by theorem [ thrm : weyl ] that @xmath618 implying that @xmath619 is full rank .
as @xmath620 is directly implied by @xmath617 it follows that @xmath621 which makes theorem [ thrm : stewart ] applicable to @xmath622 . as @xmath623 it follows that @xmath624 where we used theorem [ thrm : stewart ] and the fact that @xmath625 and @xmath626 @xcite . by noting that @xmath627 , that @xmath628 , that @xmath629 and that @xmath630 ( cf . )
, it follows that @xmath631 \sigma_{i}(\bfa ) \geq \| \bfp - \underline{\bfp } \|.\ ] ] by again applying theorem [ thrm : weyl ]
to it follows that @xmath632 .
note however that @xmath633 where @xmath634 .
as @xmath635 and @xmath636 it follows that @xmath637 and @xmath638 .
thus , @xmath639 establishing the lemma .
let @xmath640 be the un - normalized extended codebook corresponding to the un - normalized lattice points @xmath641 , i.e. , where @xmath642 .
a space - time code is said to satisfy the non - vanishing determinant ( nvd ) condition if @xcite @xmath643 i.e. , if there are no non - zero un - normalized ( difference ) codewords with arbitrarily small determinants .
the proof of the theorem draws from the well known fact that the nvd condition is a necessary condition for achieving approximate universality , and divides the problem into a few ( exhaustive ) cases where either the condition in lemma [ lm : lower - bound ] is shown to hold , or the nvd property is shown to be violated , thus eliminating the possibility of nvd codes that would violate the rank condition in lemma [ lm : lower - bound ] . to this end , consider a partitioning of the @xmath414 generator matrix according to @xmath644 where @xmath645 .
first of all , let us note that the case where @xmath646 is trivially satisfied as @xmath121 is full rank , i.e. , the matrix @xmath647 is full rank for any unitary @xmath648
. we can thus restrict attention to the case of @xmath649 and consider the rank of @xmath650 where @xmath651 . in the cases where
the nvd property is shown to not hold , it is sufficient to consider non - zero ( unnormalized ) codewords of the form @xmath652 where @xmath653 , and @xmath654 , and where @xmath655 .
the ( un - normalized ) codewords in matrix form are in this case given by @xmath656 $ ] and we have @xmath657 $ ] .
all codewords discussed in what follows are assumed to have this structure .
we will now consider different cases depending on the rank of @xmath658 and @xmath659 .
however , as it is straightforward to see that @xmath660 has zero determinant ( for any @xmath661 ) if either @xmath662 or @xmath663 , the cases that need consideration are those when the rank of both @xmath658 and @xmath659 is equal to one ( case a ) , when the rank of both @xmath658 and @xmath659 is equal to two ( case b ) , and when the rank or either @xmath658 or @xmath659 is equal to one and the rank of the other is equal to two ( case c ) . consider the case where the rank of both @xmath658 and @xmath659 is one , i.e. , where @xmath664 and @xmath665 .
if @xmath666 and @xmath667 are not linearly dependent , the condition in is satisfied for any @xmath668 such that @xmath669 and @xmath670 , and we can thus restrict attention to the case where @xmath666 and @xmath667 are linearly dependent . here , we may without loss of generality assume that @xmath671 by absorbing any complex scalars into @xmath672 and @xmath673 .
note however that given any @xmath580 we can always find a point @xmath674 , where @xmath675 such that is the projection of the two dimensional gaussian integer lattice onto a one dimensional space , one can also view the statement as an application of dirichlet s box principle , cf .
@xcite . in general , the problem of finding non - zero integer vectors that are approximately orthogonal to a given vector is known as approximate integer relations ( irs ) , and is related to simultaneous diophantine approximations @xcite . ]
for any such @xmath661 it follows that ( cf .
* theorem 7.3.10 ) ) @xmath677 i.e. , the maximal singular value of @xmath660 can be made arbitrarily small .
however , this violates the assumed nvd property of the code as a small maximal singular value implies a small determinant , and concludes case a. when @xmath658 and @xmath659 are full rank we can always find a vector @xmath668 such that @xmath678 and @xmath679 are linearly independent ( thus satisfying the condition of lemma [ lm : lower - bound ] ) unless @xmath658 and @xmath659 are linearly dependent , i.e. , when @xmath680 for some @xmath62 .
however , in this case we have that @xmath681 for any @xmath661 which implies that the columns of @xmath660 are linearly dependent , and the rank of @xmath660 is zero .
this concludes case b. in this case we may assume that @xmath664 has rank one and @xmath659 has rank two ( the opposite case is handled equivalently ) . here , as both the set of @xmath668 for which @xmath682 and where @xmath679 is linearly dependent of @xmath683 have zero measure , we may pick @xmath668 such that @xmath684 and such that @xmath679 is linearly independent of @xmath683 , thus satisfying the conditions of lemma [ lm : lower - bound ] .
this concludes the proof of theorem [ thrm : nvd ] .
a. d. murugan , h. e. gamal , m. o. damen , and g. caire , `` a unified framework for tree search decoding : rediscovering the sequential decoder , '' _ ieee trans .
inform . theory _
52 , no . 3 , pp .
933 953 , mar .
2006 .
j. jaldn and p. elia , `` dmt optimality of lr - aided linear decoders for a general class of channels , lattice designs , and system models , '' _ ieee trans .
inform . theory _
56 , no .
47654780 , oct . 2010 .
d. seethaler , j. jaldn , c. studer , and h. blcskei , `` tail behavior of sphere - decoding complexity in random lattices , '' in _ proc .
ieee international symposium on information theory , isit _ , june 2009 , pp .
729 733 . h. yao and g. w. wornell , `` structured space - time block codes with optimal diversity - multiplexing tradeoff and minimum delay , '' in _ proc .
ieee global telecommunications conference , globecom _
, san francisco , ca , dec .
belfiore , g. rekaya , and e. viterbo , `` the golden code : a 2 x 2 full - rate space - time code with non - vanishing determinants , '' _ ieee trans .
inform . theory _
51 , no . 4 , pp .
14321436 , apr .
2005 .
p. elia , k. r. kumar , s. a. pawar , p. vijay kumar , and h .- f .
lu , `` explicit space - time codes achieving the diversity - multiplexing gain tradeoff , '' _ ieee trans . inform . theory _
52 , no . 9 , pp . 38693884 , sept . 2006 .
w. abediseid and m. damen , `` lattice sequential decoder for coded mimo channel : performance and complexity analysis , '' jan .
2011 , submitted to the ieee trans .
inform . theory .
available as arxiv:1101.0339v1 .
h. el gamal , g. caire , and m. o. damen , `` lattice coding and decoding achieve the optimal diversity - multiplexing tradeoff of mimo channels , '' _ ieee trans .
inform . theory _
50 , no . 6 , pp .
968985 , june 2004 .
b. a. sethuraman , b. sundar rajan , and v. shashidhar , `` full - diversity , high - rate , space - time block codes from division algebras , '' _ ieee trans .
49 , no .
10 , pp . 25962616 , oct .
2003 .
o. tirkkonen and r. kashaev , `` combined information and performance optimization of linear mimo modulations , '' in _ proc .
ieee international symposium on information theory , isit _ , lausanne , switzerland , june 2002 .
j. paredes , a. b. gershman , and m. gharavi - alkhansari , `` a 2x2 space - time code with non - vanishing determinant and fast maximum likelihood decoding , '' in _ proc .
ieee international conference on acoustics , speech , and signal processing , icassp _ , honolulu , hawaii , apr .
2007 , pp . 877880 .
m. samuel and m. p. fitz , `` reducing the detection complexity by using @xmath685 multi - strata space - time codes , '' in _ proc .
ieee international symposium on information theory , isit _ , nice , france , june 2007 . | in the setting of quasi - static multiple - input multiple - output ( mimo ) channels , we consider the high signal - to - noise ratio ( snr ) asymptotic complexity required by the sphere decoding ( sd ) algorithm for decoding a large class of full rate linear space - time codes . with sd complexity having random fluctuations induced by the random channel , noise and codeword realizations ,
the introduced _ sd complexity exponent _ manages to concisely describe the computational reserves required by the sd algorithm to achieve arbitrarily close to optimal decoding performance .
bounds and exact expressions for the sd complexity exponent are obtained for the decoding of large families of codes with arbitrary performance characteristics . for the particular example of decoding
the recently introduced threaded cyclic division algebra ( cda ) based codes the only currently known explicit designs that are uniformly optimal with respect to the diversity multiplexing tradeoff ( dmt ) the sd complexity exponent is shown to take a particularly concise form as a non - monotonic function of the multiplexing gain . to date
, the sd complexity exponent also describes the minimum known complexity of any decoder that can provably achieve a gap to maximum likelihood ( ml ) performance which vanishes in the high snr limit .
diversity - multiplexing tradeoff , sphere decoding , complexity , space - time codes , large deviations . |
the concept of groupoid is central in non commutative geometry .
groupoids generalize the concepts of spaces , groups and equivalence relations .
it is clear nowadays that groupoids are natural substitutes of singular spaces .
many people have contributed to realizing this idea .
we can find for instance a groupoid - like treatment in dixmier s works on transformation groups , @xcite , or in brown - green - rieffel s work on orbit classification of relations , @xcite .
in foliation theory , several models for the leaf space of a foliation were realized using groupoids , mainly by people like haefliger ( @xcite ) and wilkelnkemper ( @xcite ) , for mention some of them .
there is also the case of orbifolds , these can be seen indeed as tale groupoids , ( see for example moerdijk s paper @xcite ) .
there are also some particular groupoid models for manifolds with corners and conic manifoldss worked by people like monthubert @xcite , debord - lescure - nistor ( @xcite ) and aastrup - melo - monthubert - schrohe ( @xcite ) for example .
the way we treat singular spaces in non commutative geometry is by associating to them algebras . in the case
when the singular space is represented by a lie groupoid , we can , for instance , consider the convolution algebra of differentiable functions with compact support over the groupoid ( see connes or paterson s books @xcite and @xcite ) .
this last algebra plays the role of the algebra of smooth functions over the singular space represented by the groupoid . from the convolution algebra
it is also possible to construct a @xmath1-algebra , @xmath2 , that plays , in some sense , the role of the algebra of continuous functions over the singular space .
the idea of associating algebras in this sense can be traced back in works of dixmier ( @xcite ) for transformation groups , connes ( @xcite ) for foliations and renault ( @xcite ) for locally compact groupoids , for mention some of them .
using methods of noncommutative geometry , we would like to get invariants of this algebras , and hence , of the spaces they represent .
for that , connes showed that many groupoids and algebras associated to them appeared as
non commutative analogues ` of spaces to which many tools of geometry ( and topology ) such as k - theory and characteristic classes could be applied ( @xcite , @xcite ) . one classical way to obtain invariants in classical geometry ( topology ) , is through the index theory in the sense of atiyah - singer . in the lie groupoid case
, there is a pseudodifferential calculus , developed by connes ( @xcite ) , monthubert - pierrot ( @xcite ) and nistor - weinstein - xu ( @xcite ) in general .
some interesting particular cases were treated in the groupoid - spirit by melrose ( @xcite ) , moroianu ( @xcite ) and others ( see @xcite ) .
let @xmath3 be a lie groupoid , there is an analytic index morphism , ( see @xcite ) , @xmath4 where @xmath5 is the lie algebroid of @xmath6 .
the `` @xmath1-index '' @xmath7 is a homotopy invariant of the @xmath6-pseudodifferential elliptic operators and has proved to be very useful in very different situations , ( see @xcite , @xcite , @xcite for example ) .
one way to define the above index map is using the connes ' tangent groupoid associated to @xmath6 as explained by hilsum and skandalis in @xcite or by monthubert and pierrot in @xcite .
the tangent groupoid is a lie groupoid @xmath8\ ] ] with @xmath9 $ ] and the groupoid structure is given by the groupoid structure of @xmath5 at @xmath10 and by the groupoid structure of @xmath6 for @xmath11 .
one of the main features about the tangent groupoid is that its @xmath12algebra @xmath13 is a continuous field of @xmath12algebras over the closed interval @xmath0 $ ] , with associated fiber algebras @xmath14 at @xmath10 , and @xmath2 for @xmath11 .
in fact , it gives a @xmath12algebraic quantization of the poisson manifold @xmath15 ( in the sense of @xcite ) , and this is the main point why it allows to define the index morphism as a sort of deformation .
thus , the tangent groupoid construction has been very useful in index theory ( @xcite , @xcite , @xcite ) but also for other purposes ( @xcite , @xcite ) . now , to understand the purpose of the present work ,
let us first say that the indices ( in the sense of atiyah - singer - connes ) have not necessarily to be considered as elements in @xmath16 .
indeed , it is possible to consider indices in @xmath17 .
the @xmath18-indices are more refined but they have several inconvenients ( see alain connes book section @xmath19 for a discussion on this matter ) , nevertheless this kind of indices have the great advantage that one can apply to them the existent tools ( such as pairings with cyclic cocycles or chern - connes character ) in order to obtain numerical invariants . in this work
we begin a study of more refined indices .
in particular we are looking for indices between the @xmath18 and the @xmath1-levels ; trying to keep the advantages of both approaches ( see @xcite for a more complete discussion ) . in the case of lie groupoids
this refinement could mean forget for a moment the powerful tools of the theory of @xmath1-algebras and instead , working in a purely algebraic and geometric level .
in the present article , we construct an algebra of @xmath20 functions over @xmath21 , denoted by @xmath22 .
this algebra is also a field of algebras over the closed interval @xmath0 $ ] , with associated fiber algebras , @xmath23 , at @xmath10 , and @xmath24 for @xmath11 , where @xmath23 is the schwartz algebra of the lie algebroid .
furthermore , we will have @xmath25 as inclusions of algebras .
let us explain in some words why we define an algebra over the tangent groupoid such that in zero it is schwartz : the `` schwartz algebras '' have in general the good @xmath26theory groups .
for example , we are interested in the symbols of @xmath6-pdo and more precisely in their homotopy classes in @xmath27-theory , that is , we are interested in the group @xmath28 . here
it would not be enough to take the @xmath26theory of @xmath29 ( see the example in @xcite p.142 ) , however it is enough to consider the schwartz algebra @xmath30 .
indeed , the fourier transform shows that this last algebra is stable under holomorphic calculus on @xmath14 and so it has the `` good '' @xmath27-theory , meaning that @xmath31 .
none of the inclusions in ( [ incl ] ) is stable under holomorphic calculus , but that is precisely what we wanted because our algebra @xmath22 have the remarkable property that its evaluation at zero is stable under holomorphic calculus while its evaluation at one ( for example ) is not .
the algebra @xmath22 is , as vector space , a particular case of a more general construction that we do for deformation to the normal cone manifolds from which the tangent groupoid is a special case ( see @xcite and @xcite ) .
a deformation to the normal cone manifold ( dnc for simplify ) is a manifold associated to an injective immersion @xmath32 that is considered as a sort of blow up in differential geometry .
the construction of a dnc manifold has very nice functorial properties ( section [ dnc ] ) which we exploit to achieve our construction .
we think that our construction could be used also for other purposes , for example , it seems that it could help to give more understanding in quantization theory ( see again @xcite ) .
the article is organized as follows .
in the second section we recall the basic facts about lie groupoids .
we explain very briefly how to define the convolution algebra @xmath24 . in the third section
we explain the `` deformation to the normal cone '' construction associated to an injective immersion . even if this could be considered as classical material , we do it in some detail since we will use in the sequel very explicit descriptions that we could not find elsewhere .
we also review some functorial properties associated to these deformations .
a particular case of this construction is the tangent groupoid associated to a lie groupoid . in the fourth section we start by constructing a vector space @xmath33 for any deformation to the normal cone manifold @xmath34 ; this space already exhibits the characteristic of being a field of vector spaces over the closed interval @xmath0 $ ] , such that in zero we have a schwartz space while out of zero we have @xmath35 .
we then define the algebra @xmath22 , the main result is precisely that the product is well defined .
the last section is devoted to motivate the construction of our algebra by explaining in a few words some further developements that will immediately follow from this work .
all the results of the present work are part of the author s phd thesis .
let us recall what a groupoid is : a @xmath36 consists of the following data : two sets @xmath6 and @xmath37 , and maps * @xmath38 called the source and target map respectively , * @xmath39 called the product map ( where @xmath40 ) , * @xmath41 the unit map and * @xmath42 the inverse map such that , if we note @xmath43 , @xmath44 and @xmath45 , we have * @xmath46 , @xmath47 when this is possible . * @xmath48 and @xmath49 , @xmath50 with @xmath51 and @xmath52 . * @xmath53 and @xmath54 , @xmath55 . * @xmath56 and @xmath57 .
generally , we denote a groupoid by @xmath3 where the parallel arrows are the source and target maps and the other maps are given .
now , a lie groupoid is a groupoid in which every set and map appearing in the last definition is @xmath20 ( possibly with borders ) , and the source and target maps are submersions . for @xmath58 subsets of @xmath37
we use the notation @xmath59 for the subset @xmath60 . all along this paper
, @xmath3 is going to be a lie groupoid .
we recall how to define an algebra structure in @xmath61 using smooth haar systems .
a @xmath62 over a lie groupoid consists of a family of measures @xmath63 in @xmath64 for each @xmath65 such that , * for @xmath66 we have the following compatibility condition : @xmath67 * for each @xmath68 the map @xmath69 belongs to @xmath70 a lie groupoid always posses a smooth haar system .
in fact , if we fix a smooth ( positive ) section of the 1-density bundle associated to the lie algebroid we obtain a smooth haar system in a canonical way .
the advantage of using 1-densities is that the measures are locally equivalent to the lebesgue measure .
we suppose for the rest of the paper a given smooth haar system given by 1-densities ( for complete details see @xcite ) .
we can now define a convolution product on @xmath61 : let @xmath71 , we set @xmath72 this gives a well defined associative product .
there is a way to avoid the haar system when one works with lie groupoids , using half densities ( see connes book @xcite ) .
let @xmath73 be a @xmath20 manifold and @xmath74 be a @xmath20 submanifold .
we denote by @xmath75 the normal bundle to @xmath76 in @xmath73 , @xmath77 , @xmath78 .
we define the following set @xmath79 . \end{aligned}\ ] ] the purpose of this section is to recall how to define a @xmath20-structure with boundary in @xmath34 .
this is more or less classical , for example it was extensively used in @xcite .
here we are only going to do a sketch .
let us first consider the case where @xmath80 and @xmath81 ( where we identify canonically @xmath82 ) .
we denote by @xmath83 and by @xmath84 for @xmath85 as above . in this case
we clearly have that @xmath86 $ ] ( as a set ) .
consider the bijection @xmath87 \rightarrow { \mathscr{d}}_{p}^{n}\end{aligned}\ ] ] given by @xmath88 which inverse is given explicitly by @xmath89 we can consider the @xmath20-structure with border on @xmath84 induced by this bijection . in the general case .
let @xmath90 be a local chart in @xmath73 and suppose it is an @xmath76-slice , so that it satisfies * @xmath91 * if @xmath92 , @xmath93 ( we note @xmath94 ) with this notation we have that @xmath95 is an open subset .
we may define a function @xmath96 in the following way : for @xmath97 we have @xmath98 .
if we write @xmath99 , then @xmath100 is a diffeomorphism , where @xmath101 .
we set @xmath102 and @xmath103 for @xmath11 . here
@xmath104 is the normal component of the derivate @xmath105 for @xmath106 .
it is clear that @xmath107 is also a bijection ( in particular it induces a @xmath108 structure with border over @xmath109 ) .
let us define , with the same notations as above , the following set @xmath110 : ( x , t\cdot \xi)\in u \}.\ ] ] which is an open subset of @xmath111 $ ] and thus a @xmath20 manifold ( with border ) .
it is immediate that @xmath112 is diffeomorphic to @xmath113 through the restriction of @xmath114 , used in ( [ psi ] ) .
now we consider an atlas @xmath115 of @xmath73 consisting of @xmath116slices .
it is clear that @xmath117 and if we take @xmath118 defined as the composition @xmath119 then we obtain the following result . [ atlas ] @xmath120 is a @xmath20 atlas with border over @xmath34 .
in fact the proposition can be proved directly from the following elementary lemma [ ellema2 ] let @xmath121 a @xmath20 diffeomorphism where @xmath122 and @xmath123 are open subsets .
we write @xmath124 and we suppose that @xmath125 .
then the function @xmath126 defined by @xmath127 is a @xmath20 map . since the result will hold if and
only if it is true in each coordinate , it is enough to prove that if we have @xmath128 a @xmath108 map with @xmath129 , then the map @xmath130 given by @xmath131 is a @xmath108 map .
for that , we write @xmath132 with @xmath133 a @xmath108 map such that @xmath134
. then @xmath135 from which we immediately get the result .
let @xmath74 be as above .
the set @xmath34 provided with the @xmath108 structure with border induced by the atlas described in the last proposition is called @xmath136 @xmath74 .
we will often write dnc instead of deformation to the normal cone .
following the same steps , it is possible to define a deformation to the normal cone associated to an injective immersion @xmath32 .
[ exdcn ] let us mention some basic examples of dcn manifolds @xmath34 : * consider the case when @xmath137 .
we have that @xmath138 $ ] with the usual @xmath20 structure on @xmath139 $ ] .
we used this fact implicitly for cover @xmath34 as in ( [ atlasdcn ] ) . * consider the case when @xmath74 is an open subset .
then we do not have any deformation at zero and we immediately see by definition that @xmath34 is just the open subset of @xmath140 $ ] consisting in the union of @xmath141 $ ] and @xmath139 $ ] .
the most important feature about the dnc construction is that it is in some sense functorial .
more explicitly , let @xmath142 and @xmath143 be @xmath20-couples as above and let @xmath144 be a couple morphism , i.e. , a @xmath20 map @xmath145 , with @xmath146 .
we define @xmath147 by the following formulas : + @xmath148 and + @xmath149 for @xmath11 , + where @xmath150 is by definition the map @xmath151 induced by @xmath152 .
we have the following proposition , which is also an immediate consequence of the lemma above .
the map @xmath153 is @xmath20 .
if we consider the category @xmath154 of @xmath20 pairs given by a @xmath20 manifold and a @xmath20 submanifold and pair morphisms as above , we can reformulate the proposition and say that we have a functor @xmath155 where @xmath156 denote the category of @xmath20 manifolds with border .
let @xmath3 be a lie groupoid .
@xmath157 associated to @xmath6 is the groupoid that has @xmath158 as the set of arrows and @xmath159 $ ] as the units , with : * @xmath160 and @xmath161 at @xmath10 .
* @xmath162 and @xmath163 at @xmath164 . * the product is given by @xmath165 and @xmath166 if @xmath167 and if @xmath168 .
* the unit map @xmath169 is given by @xmath170 and @xmath171 for @xmath11 .
we denote @xmath172 .
as we have seen above @xmath173 can be considered as a @xmath20 manifold with border . as a consequence of the functoriality of the dnc construction
we can show that the tangent groupoid is in fact a lie groupoid .
indeed , it is easy to check that if we identify in a canonical way @xmath174 with @xmath175 , then @xmath176 where we are considering the following pair morphisms : @xmath177 finally , if @xmath178 is a smooth haar system on @xmath6 , then , setting * @xmath179 at @xmath180 and * @xmath181 at @xmath182 for @xmath11 , where @xmath183 , one obtains a smooth haar system for the tangent groupoid ( details may be found in @xcite ) .
[ exgt ] we finish this section with some interesting examples of groupoids and their tangent groupoids .
let @xmath185 be a lie group considered as a lie groupoid , @xmath186 .
in this case the normal bundle to the inclusion @xmath187 is of course identified with the lie algebra of the group .
hence , the tangent groupoid is a deformation of the group in its lie algebra : @xmath188 $ ] .
let @xmath190 be a smooth vector bundle over a @xmath20 manifold @xmath76 ( connexe ) .
we can consider the lie groupoid @xmath191 induced by the vector structure of the fibers , @xmath77 , @xmath192 and the composition is given by the vector sum @xmath193 . in this case
the normal vector bundle associated to the zero section can be identified to @xmath194 itself . hence ,
as a set the tangent groupoid is @xmath195 $ ] but the @xmath20-structure at zero is given locally as in ( [ psi ] ) .
let @xmath73 a @xmath20-manifold .
we can consider the product groupoid @xmath197 .
the tangent groupoid in this case takes the following form @xmath198 $ ] .
this is called the tangent groupoid to @xmath73 and it was introduced by connes for giving a very conceptual proof of the atiyah - singer index theorem ( see @xcite and @xcite ) .
in this section we will show how to construct an algebra for the tangent groupoid which consist of @xmath20 functions that satisfy a rapid decay condition at zero while out of zero they satisfy a compact support condition .
this algebra is the main construction in this work . our algebra for
the tangent groupoid will be a particular case of a construction associated to any deformation to the normal cone .
we start by defining a space for dncs associated to open subsets of @xmath199 .
[ ladef ] let @xmath200 and @xmath201 an open subset , and let @xmath202 .
* let @xmath203 $ ] be a compact subset .
we say that @xmath27 is a conic compact subset of @xmath204 $ ] relative to @xmath205 if @xmath206 * let @xmath207 .
we say that @xmath208 has compact conic support @xmath27 , if there exists a conic compact @xmath27 of @xmath204 $ ] relative to @xmath205 such that if @xmath11 and @xmath209 then @xmath210 . *
we denote by @xmath211 the set of functions @xmath212 that have compact conic support and that satisfy the following condition : * * @xmath213 @xmath214 , @xmath215 and @xmath216 it exists @xmath217 such that @xmath218 now , the spaces @xmath211 are invariant under diffeomorphisms .
more precisely if @xmath121 is a @xmath20 diffeomorphism as in lemma [ ellema2 ] then we can prove the next result .
[ gtilde ] let @xmath219 , then @xmath220 .
the first observation is that @xmath221 , thanks to lemma [ ellema2 ] .
let us check that it has compact conic support .
for that , let @xmath222 $ ] the conic compact support of @xmath223 .
we let @xmath224})\subset u\times [ 0,1],\ ] ] which is a conic compact subset of @xmath225 $ ] relative to @xmath205 , and it is immediate by definition that @xmath226 if @xmath11 and @xmath227 , that is , @xmath228 has compact conic support @xmath27 .
we now check the rapid decay property @xmath229 : for simplify the proof we first introduce some useful notation .
writing @xmath230 as in the lemma [ ellema2 ] , we denote @xmath231 and @xmath232 .
we denote also @xmath233 and @xmath234 where @xmath235 is also as above , @xmath77 , @xmath236 in particular by definition we have @xmath237 .
we also write @xmath238 and @xmath239 . hence
, what we would like is to find bounds for expressions of the following type @xmath240 for arbitrary @xmath241 and @xmath242 .
a simple calculation shows that the derivates @xmath243 are of the following form @xmath244 where @xmath245 is a finite sum of products of the form @xmath246 we are only interested in see what happens in the set @xmath247 since out of this set we have that @xmath223 and all its derivates vanish ( @xmath248 iff @xmath249 ) . for a point @xmath250
we have that @xmath251 is in a compact set and then it follows that the expressions @xmath252 are bounded in @xmath253 . for the expressions @xmath254 , we proceed first by developing as in lemma [ ellema2 ] , that is , @xmath255 now , since we are only considering points in @xmath253 , it is immediate that we can find constants @xmath256 such that @xmath257 in the same way ( remember @xmath258 is a diffeomorphism ) we can have constants @xmath259 such that @xmath260 putting all together , and using the property @xmath229 for @xmath223 , we get bounds @xmath261 such that @xmath262 and this concludes the proof .
we can resume the last invariance result as follows : if @xmath263 is a @xmath20 pair diffeomorphic to @xmath264 with @xmath265 , an open subset of a vector space @xmath194 , and @xmath266 , then @xmath267 is well defined and does not depend on the pair diffeomorphism . with the last compatibility result in hand we are ready to give the main definition in this work .
[ src ] let @xmath268 . *
we say that @xmath223 has compact conic support @xmath27 , if there exists a compact subset @xmath269 $ ] with @xmath270 ( conic compact relative to @xmath76 ) such that if @xmath11 and @xmath271 then @xmath272 .
* we say that @xmath223 is rapidly decaying at zero if for every @xmath90 @xmath76-slice chart and for every @xmath273)$ ] , the map @xmath274 given by @xmath275 is in @xmath211 , where @xmath276 is the projection @xmath277 $ ] given by @xmath278 , and @xmath279 for @xmath11 .
finally , we denote by @xmath33 the set of functions @xmath280 that are rapidly decaying at zero with compact conic support .
* by definition of @xmath33 we see that @xmath281 is contained as a vector subspace . *
it is clear that @xmath282)$ ] can be considered as a subspace of @xmath33 by extending by zero the functions at @xmath75 .
following the lines of the last remark we are going to precise a possible decomposition of our space @xmath33 that will be very useful in the sequel .
let @xmath283 a family of @xmath116slices covering @xmath76 .
consider the open cover of @xmath284 $ ] consisting in @xmath285,\phi_{\alpha } ) \}_{\alpha \in \delta}$ ] union with @xmath139 $ ] .
we can take a partition of the unity subordinated to the last cover , @xmath286 that is , we have the following properties : * @xmath287 * @xmath288 $ ] and @xmath289 $ ] .
* @xmath290 let @xmath291 , we denote @xmath292 and @xmath293 } \cdot ( \lambda \circ p ) \in { c^{\infty}}(m\times ( 0,1]),\ ] ] then we obtain the following decomposition : @xmath294 now , since @xmath208 is conic compactly supported we can suppose , without lost of generality , that * @xmath295)$ ] , and * that @xmath296 is compactly supported in @xmath297 $ ] . what we conclude of all this
, is that we can decompose our space @xmath33 as follows @xmath298).\end{aligned}\ ] ] as we mentioned in the introduction , we want to see the space @xmath33 as a field of vector spaces over the interval @xmath0 $ ] , where at zero we talked about schwartz spaces . in our case
we are interested in schwartz functions on the vector bundle @xmath75 .
let us first recall the notion of the schwartz space associated to a vector bundle .
let @xmath299 be a smooth vector bundle over a @xmath20 manifold @xmath76 .
we define the schwartz space @xmath300 as the set of @xmath20 functions @xmath301 such that @xmath223 is a schwartz function at each fiber ( uniformly ) and @xmath223 has compact support in the direction of @xmath76 , @xmath77 , if there exists a compact subset @xmath302 such that @xmath303 for @xmath304 .
the vector space @xmath300 is an associative algebra with the product given as follows : for @xmath305 , we put @xmath306 where @xmath307 is a smooth haar system of the lie groupoid @xmath191 .
a classical fourier argument can be applied to show that the last algebra is isomorphic to @xmath308 ( punctual product ) . in particular
this implies that @xmath309 .
in the case we are interested , we have a couple @xmath142 and a vector bundle associated to it , that is , the normal bundle over @xmath76 , @xmath75 .
the reason why we gave the last definition is because we get evaluation linear maps @xmath310 and @xmath311 for @xmath164 .
consequently , we have that the vector space @xmath33 is a field of vector spaces over the closed interval @xmath0 $ ] , which fibers spaces are : @xmath312 at @xmath10 and @xmath35 for @xmath11 .
let us finish this subsection by giving the examples of spaces @xmath33 corresponding to the dcn manifolds seen at [ exdcn ] above .
* for @xmath137 , we have that @xmath313)$ ] . * for @xmath74 an open subset we have that @xmath314 where @xmath315 $ ] is the open subset consisting of the union of @xmath316 $ ] and @xmath139 $ ] . in this section
we define an algebra structure on @xmath22 .
we start by defining a function @xmath317 by the following formulas : for @xmath318 , we let @xmath319 and @xmath320 if we canonically identify @xmath174 with @xmath175 , the map above is nothing else that the integration along the fibers of @xmath321 . we have the following proposition : [ elproducto ] @xmath322 is a well defined linear map . the interesting part of the proposition is that the map is well defined since it will evidently be linear . let us suppose for the moment that the last proposition is true . under this assumption , we will define the product in @xmath22 .
[ conv ] let @xmath323 , we define a function @xmath324 in @xmath21 by @xmath325 and @xmath326 for @xmath11 .
we can enounce our main result .
@xmath327 defines an associative product on @xmath22 .
remember we are assuming for the moment the proposition [ elproducto ] .
let @xmath328 .
we let @xmath329 the function in @xmath175 defined by : @xmath330 and @xmath331 for @xmath11 . now , from the leibnitz formula for the derivate of a product it is immediate that @xmath332 . finally , by definition we have
that @xmath333 hence , thanks to proposition [ elproducto ] , @xmath324 is a well defined element in @xmath22 . for the associativity of the product , let us remark that when one restrict the product to @xmath334 , this coincides with the product classically considered on @xmath334 ( which is associative , see for example @xcite ) .
the associativity for @xmath22 is proved exactly in the same way that for @xmath334 .
we have then to prove proposition [ elproducto ] .
we are going to start locally .
let @xmath335 be an open set and @xmath336 .
let @xmath337 the canonical projection @xmath338 .
we set @xmath339 , then @xmath340 is also an open subset , @xmath341 and @xmath342 .
we denote also by @xmath343 the restriction @xmath344 .
we have as in lemma [ ellema2 ] a @xmath20 map @xmath345 , which in this case is explicitly written by @xmath346 we define @xmath347 as follows @xmath348 let us prove the following lemma .
[ ptilde ] @xmath349 is well defined .
the first observation is that the integral in the definition of @xmath350 is always well defined .
indeed , we deduce it from the next two points : * for @xmath10 , @xmath351 . * for @xmath11 , @xmath352 .
once we can derivate under the integral symbol , we obtain that @xmath353 .
then , we just have to show that @xmath354 verifies the two conditions of the definition [ ladef ] .
for the first , if @xmath355 $ ] is the compact conic support of @xmath258 , then it is enough to put @xmath356})(k)\ ] ] in order to obtain a conic compact subset of @xmath357 $ ] relative to @xmath205 and to check that @xmath358 is the compact conic support of @xmath354 .
let us now verify the condition @xmath229 .
let @xmath214 , @xmath215 and @xmath359 .
we want to find @xmath360 such that @xmath361 for @xmath362 and @xmath363 we have by hypothesis that it exists @xmath364 such that @xmath365 then , we also have that @xmath366 @xmath367 with @xmath368 we can now give the proof of the proposition [ elproducto ] .
let us first fix some notation .
we suppose dim @xmath369 and dim @xmath370 , in particular this implies that dim @xmath371 .
let @xmath90 and @xmath372 be @xmath37-slices in @xmath373 and @xmath6 respectively such that the following diagram commutes @xmath374^m \ar[d]_{\phi } & { \mathscr{u } } ' \ar[d]^{\phi ' } \\ u \ar[r]_p & u ' , } \ ] ] where @xmath375 is the canonical projection ( as above ) and @xmath376 .
this is possible since @xmath377 is a surjective submersion .
now , we apply the dnc construction to the diagram above to obtain , thanks to the functoriality of the construction , the following commutative diagram @xmath378^{{\mathscr{d}}(m ) } \ar[d]_{\tilde{\phi } } & { \mathscr{d}}_{{\mathscr{v}}'}^{{\mathscr{u } } ' } \ar[d]^{\tilde{\phi ' } } \\
{ \mathscr{d}}_{v}^{u } \ar[r]_{{\mathscr{d}}(p ) } & { \mathscr{d}}_{v'}^{u ' } \\ \omega_{\phi } \ar[u]^{\psi_{\phi } } \ar[r]_{\tilde{p } } & \omega_{\phi ' } \ar[u]_{\psi_{\phi ' } } , } \ ] ] where @xmath379 and @xmath380 are as in section [ dnc ] .
let @xmath381 , we define @xmath382 then , from the last commutative diagram , we get that @xmath383 hence , thanks to [ ptilde ] , we can conclude that we have a well defined linear map @xmath384 we now use the proposition [ gtilde ] to write @xmath385 and so for @xmath386 , we see that @xmath387 we use again the last commutative diagram to see that @xmath388 we then have a well defined linear map @xmath389 to pass to the global case we only have to use the decomposition of @xmath390 and of @xmath391 as in ( [ decomposicion ] ) , and of course the invariance under diffeomorphisms ( proposition [ gtilde ] ) .
let us recall that we have well defined evaluation morphisms as in ( [ e0 ] ) and ( [ e1 ] ) . in the case of a the tangent groupoid
they are by definition morphisms of algebras .
hence , the algebra @xmath22 is field of algebras over the closed interval @xmath0 $ ] , with associated fiber algebras , @xmath23 , at @xmath10 , and @xmath24 for @xmath11 .
it is very interesting to see what this means in the examples given in [ exgt ] .
let @xmath3 be a lie groupoid . in index theory for lie groupoids
the tangent groupoid has been used to define the analytic index associated to the group , as a morphism @xmath392 ( see @xcite ) or as a @xmath393-element in @xmath394 ( see @xcite ) , this can be done because one has the following short exact sequence of @xmath1-algebras @xmath395 ) \longrightarrow c_{r}^ { * } ( { \mathscr{g}}^t ) \stackrel{e_0}{\longrightarrow } c_0(a^*{\mathscr{g } } ) \longrightarrow 0,\end{aligned}\ ] ] and because of the fact that the @xmath27-groups of the algebra @xmath396)$ ] vanish ( homotopy invariance ) .
the index defined at the @xmath1-level has proven to be very useful ( see for example @xcite ) but extracting numerical invariants from it , with the existent tools , is very difficult . in non commutative geometry , and also in classical geometry , the tools for obtain more explicit invariants are more developed for the smooth objects ; in our case this means the convolution algebra @xmath24 , where we can for example apply chern - weil - connes theory . hence , in some way , the indices defined in @xmath397 are more refined objects and for some cases it would be preferable to work with them .
unfortunately this indices are not good enough , since for example they are not homotopy invariants ; in @xcite alain connes discusses this and also other reasons why it is not enough to keep with the @xmath18-indices .
the main reason to construct the algebra @xmath398 is that it gives an intermediate way between the @xmath18-level and the @xmath1-level and will allow us in @xcite to define another analytic index morphism associated to the groupoid , with the advantage that this index will take values in a group that allows to do pairings with cyclic cocycles and in general to apply chern - connes theory to it .
the way we are going to define our index is by obtaining first a short exact sequence analogue to ( [ se * ] ) , that is , a sequence of the following kind @xmath399 the problem here will be that we do not dispose of the advantages of the @xmath26theory for @xmath1-algebras , since the algebras we are considering are not of this type ( we do not have for example homotopy invariance ) .
7 dixmier , j. les @xmath1-algbres et leurs reprsentations . _ cahiers scientifiques , fasc .
xxix gauthier - villars & cie _ , paris ( 1964 ) .
haefliger , a. groupodes dholonomie et classifiants .
_ astrisque _ transversal structure of foliations ( 1982 ) , no .
116 , 7097 .
hilsum , m. and skandalis , g. morphismes @xmath27-orients despaces de feuilles et fonctorialit en thorie de kasparov ( daprs une conjecture da.connes ) _
ann.sci.cole norm.sup._ ( 4 ) 20 ( 1987 ) , no . 3 , 325 - 390 . | we construct an algebra of smooth functions over the tangent groupoid associated to any lie groupoid .
this algebra is a field of algebras over the closed interval @xmath0 $ ] which fiber at zero is the algebra of schwartz functions over the lie algebroid , whereas any fiber out of zero is the convolution algebra of the initial groupoid .
our motivation comes from index theory for lie groupoids .
in fact , our construction gives an intermediate algebra between the enveloping @xmath1-algebra and the convolution algebra of compactly supported functions of the tangent groupoid ; and it will allows us , in a further work , to define other analytic index morphisms as a sort of deformations . primary 58 - 06 , 19 - 06 ; secondary 58h15 , 19k56 .
lie groupoids , tangent groupoid , k - theory , index theory . |
trapped ions provide a promising candidate to pursue quantum information processing .
successful implementations of basic quantum protocols as well as the creation of entangled states support this view @xcite . a promising route to scalability of ion - trap based quantum information processing was proposed in which segmented trap electrodes allow splitting , shuttling , and recombination of ion crystals @xcite .
to overcome the difficulties of assembling three - dimensional trap structures and simplify fabrication , several groups are developing planar ( surface ) ion traps where all electrodes lie within one plane and can be constructed using standard microfabrication methods @xcite . in particular , recently the nist group successfully transported ions through junctions in such planar traps @xcite . despite the recent progress
, there are still a number of unresolved difficulties with planar traps . in the effort to develop this technology , one strives for miniaturization in order to easily achieve the high trap frequencies required for fast splitting of ion crystals @xcite , and to increase coupling rates in approaches where trapped ions are coupled to @xcite or via solid state elements @xcite .
thus , planar traps with ion - surface separations of less than 100 @xmath0 m are pursued .
however , the observed ion heating rates are more than 3 orders of magnitude higher than those expected from johnson noise considerations @xcite , imposing a major obstacle in developing this approach .
this so - called `` anomalous '' heating in ion traps is usually discussed in terms of the `` patch potential '' model @xcite .
the essential features of this model are that the electric field noise responsible for heating of the ion motion arises from a large number of randomly fluctuating sources and that the electric field of individual sources scales with source - ion separation as an electrical dipole field .
these two features reproduce the observed @xmath1 scaling of the electric field noise with the separation , @xmath2 , between the ion and trap electrodes . besides the scaling with distance
, the noise spectral density of anomalous heating is believed to scale as @xmath3 with frequency @xcite and experiments indicated increase of noise on particular traps with time , possibly due to contamination of the trap electrodes @xcite .
more recently it has been found that the noise is drastically reduced at low temperatures @xcite , and seems to be related to thermally activated processes @xcite , further supporting the @xmath3 nature .
this phenomenology suggests a connection with noise in various other systems . electric field noise is a factor in nanomechanics @xcite , single spin detection @xcite , and measurement of weak forces ( e.g. gravitation ) @xcite .
moreover charge noise limits the performance of nanoelectronic and quantum electronic devices , such as single electron transistors @xcite , josephson qubits @xcite , superconducting coplanar resonators @xcite , and quantum dots @xcite . in these systems
decoherence is believed to be caused by tunneling two level systems ( tls ) which possess a dipole moment @xcite .
such tls are common in amorphous dielectrics and are known to contribute to dielectric losses and attenuation of phonons @xcite .
the latter mechanism has also been suggested to be relevant in dissipation encountered in efforts to laser cool dielectric mechanical resonators @xcite . in a different direction , in measurements of non - contact friction using metalized atomic force microscope cantilevers close to metal surfaces , dissipation
9 to 11 orders of magnitude higher than expected for clean metal surfaces is observed @xcite , albeit in a different frequency and distance regime than those accessible with ion traps .
this effect is suggested to be due to impurity atoms adsorbed on the metal surfaces , and it is suggested that less than one atomic monolayer of adsorbate is sufficient to produce this effect @xcite . in addition , the initially high level of dissipation close to metal surfaces has been found to further increase by up to two orders of magnitude upon deposition of dielectrics on a gold surface@xcite .
finally , magnetic traps for neutral atoms also suffer from electromagnetic field noise @xcite . in this case ,
magnetic fields in the radio or microwave frequency range couple to the atomic spins and result in atom loss by causing spin flips .
contrary to the situations discussed above , this mechanism is well understood and accounted for by johnson noise on the metallic trap electrodes @xcite . here
we study anomalous heating in ion traps .
we describe a simple method for fabrication of planar traps and measure their heating rates using the doppler recooling method @xcite . taking advantage of the segmented trap geometry
, we measure heating rates at various positions above the trap and find strong time and position dependence , which indicates a major role of surface contamination in excess heating rates .
the heating rates in the ion - loading region of the trap have increased by more than one order of magnitude over the time span of a few months .
in contrast , the heating rates sufficiently far from the loading region remain low throughout our study . finally , we present a phenomenological model for anomalous heating in ion - traps that matches with the above general context .
this is done by adapting the patch potential model assuming a distribution of oscillating microscopic dipoles on the trap electrodes , with properties such that the @xmath3 nature of noise in ion traps is reproduced . using this model we estimate the density of electrical dipole sources on a trap electrode surface that would give rise to heating rates reported in the literature . in this context
, we discuss the possible relevance of impurity atoms adsorbed on trap electrodes and tunneling tls to anomalous heating in ion traps .
our model also explains why no `` anomalous '' heating effects are observed in magnetic atom traps .
we trap @xmath4ca@xmath5ions in an asymmetric surface electrode trap with trapping height of approximately 240 @xmath0 m , a schematic of which is shown in fig . [
fig : trap](a ) .
a resistive oven produces a beam of neutral calcium atoms propagating parallel to the trap plane and along the axis of the trap .
great care has been taken to minimize direct exposure of trap surfaces to the atomic beam : a stainless steel block , elevated by 200 @xmath0 m above the plane of the trap , shields the trap from the calcium produced by the oven .
measurements of fluorescence of neutral calcium produced by the oven beam reveal undetectably low fluorescence level at a height of 100 @xmath0 m above the trap plane .
the ca atoms are photo - ionized in the trapping region with a two - photon process from the ground state to the continuum via the 4@xmath6p@xmath7 state using 250 mw/@xmath8 of laser light at 422 nm ( first stage ) and 750 mw/@xmath8 at 375 nm ( second stage ) .
both laser beams are focused to waists of approximately 100 @xmath0 m .
the trap is driven at frequency @xmath915 mhz .
this drive is amplified by a radio frequency amplifier to @xmath10mw and stepped up via a helical resonator in a quarter - wave configuration to a voltage of amplitude typically around 100 v. this is fed to the rf electrodes of the trap , with capacitance at the rf feedthrough of approximately 10 pf .
dc electrodes placed on either side of the rf electrodes are used to move the ion along the axial direction and to compensate micromotion @xcite .
dc voltages used for trapping are between -10 v and 15 v , resulting in trap frequencies of @xmath11 mhz in the horizontal radial , vertical radial , and axial directions respectively .
when the doppler cooling lasers are switched off , ions typically remain in the trap between 5 s and 30 s depending on whether a high or low heating rate region of the trap is used .
a diode laser stabilized at 794 nm is frequency doubled using a ring cavity to produce a wavelength of 397 nm , used for doppler cooling and detection of the ions .
a second diode laser at 866 nm acts as a repump .
both lasers are frequency stabilized to within 100 khz with independent cavities using the pound - drever - hall method .
the frequency of the lasers can be varied by changing the cavity lengths with piezoelectric elements and the 397 nm laser frequency can also be adjusted without affecting the cavity using an acousto optic modulator in a double - pass configuration .
the detection laser at 397 nm is intensity stabilized at the trap to a value of 38@xmath12 .
the intensity of the 866 nm repump laser is adjusted at approximately 115@xmath12 .
ion fluorescence is collected by a microscope objective with na=0.27 .
it is simultaneously detected by a photomultiplier tube and a ccd camera , using a 9:1 beam splitter .
this configuration yields on the photomulitplier a maximum of 110 kcounts / s , which is reduced to 50 kcounts / sec after the cooling laser powers have been adjusted .
for our heating rate measurements , we use a planar trap with asymmetric radio frequency electrode .
the ratio between the widths of the two rf rails is 2:1 ( see fig . [
fig : trap ] ) .
this geometry gives rise to a tilted radio frequency quadrupole with tilt angle with respect to the vertical ( @xmath13 ) direction of approximately 25@xmath14 , which allows efficient laser cooling of both radial modes . in this configuration standard paul trap operation conditions
can be achieved by aligning the static potential quadrupole with the radio frequency quadrupole .
then ion motion along each of the trap axes is described by the usual mathieu equations @xcite .
if the static and radio frequency quadrupoles are not tilted by the same amount , the result is coupling of the mathieu equations for ion motion in different directions which could modify trap stability @xcite .
this complication will arise in planar traps with symmetric radio frequency rails , since in this case tilting the trap axes with respect to the vertical direction will always result in a tilt of the static quadrupole with respect to the radio frequency quadrupole and consequently coupled equations of motion for the ions .
the trap used in this work was designed with the goal of having several independent trapping regions on the same device .
thus it has 10 pairs of dc electrode segments on the two outer sides of the radio frequency electrodes ( see fig . [
fig : trap ] ) .
this gives enough freedom to trap in different regions of the trap and to transport ions along the trap axis .
a central dc segment is located directly underneath the ion , facilitating micromotion compensation in the vertical direction @xcite .
the central segment is 250 @xmath0 m wide , while the dc segments on the sides of the radio frequency electrodes are 1000 @xmath0 m by 400 @xmath0 m in the @xmath15 and @xmath16 directions respectively .
the widths of the radio frequency electrodes are 400 @xmath0 m and 200 @xmath0 m respectively .
we fabricated this design with a minimum dimension ( gaps between trap electrodes ) of 10@xmath0 m , and electrode thickness of approximately 5@xmath0 m .
the trap electrodes are deposited on a sapphire substrate which has rms surface roughness less than 1 nm . the initial step is evaporation of a 5 nm thick titanium adhesion layer , followed by evaporation of a 100 nm thick gold layer .
evaporation of both is done uniformly on the substrate .
photolithography is carried out using a negative photoresist ( micro resist technology , ma - n 440 ) , suitable for deposition of resist with thickness of several micrometers . after the development step ( micro resist technology ma - d 332s ) , the photoresist is removed from the parts of the substrate where trap electrodes will be deposited and only the regions where the sapphire substrate will be exposed are covered by photoresist of approximately 4.5@xmath0 m thickness . the evaporated gold acts as a seed layer during electroplating using a commercial gold electroplating solution ( metalor , ecf 60 ) at a temperature of 50@xmath17c and current density of 1.4ma/@xmath8 .
the resulting deposition rate is approximately 75nm / min and the process is carried on for 70minutes .
the resulting gold layer has thickness varying between 4.4@xmath0 m and 6.3@xmath0 m on different areas of a circular 50 mm diameter wafer .
the surface rms roughness of the gold film as measured with a surface profilometer is approximately 17 nm . after electroplating ,
the photoresist is removed and the seed layer is etched away using diluted `` aqua regia '' , with hcl : hno@xmath18:h@xmath19o in a 3:1:7 ratio . in this process etching of the seed layer in the regions of the gaps between trap electrodes
will proceed at a rate considerably slower than in the rest of the trap , due to insufficient wetting of high aspect ratio troughs by the etching solution . to improve the etching rate of the seed layer , we use a surfactant ( triton x-100 ) .
finally , after removal of the gold seed layer , a longer titanium - specific etching step to remove the remains of the titanium adhesion layer is performed in a diluted hydrochloric acid solution ( 32% hcl with addition of triton - x 100 , heated to 50@xmath17c ) .
the final etching steps do not influence the surface roughness of the trap electrodes . before installation in vacuum
all traps are cleaned in a sequence of acetone / isopropyl alcohol / deionized water baths , and blow dried in a nitrogen gas stream . using the above procedure , we have fabricated traps of various sizes . the ion trapping height in these varies between 500@xmath0 m and 125@xmath0 m .
we successfully trapped single ions in three of those traps , the first with ion height of 500@xmath0 m above the trap surface , and the other two with ion height of approximately 240@xmath0 m .
for what follows , we focus on one of the latter , which was extensively studied .
prior to the heating rate measurements , this trap was baked in vacuum at 150@xmath14c for two weeks , to reach a base pressure of @xmath20 mbar .
the heating rates of the trap have been measured using the doppler recooling method in which the motional energy of the ion is determined by taking advantage of the increased doppler shift for an ion with increased kinetic energy @xcite .
this allows one to estimate the ion energy by comparing the instantaneous ion fluorescence to the steady state fluorescence level of a doppler cooled ion . to evaluate the change in fluorescence
, doppler cooling of the ion is turned off for time @xmath21 during which time the ions heat up .
the change in the ion fluorescence is monitored as the ion cools down after turning the doppler cooling on .
this change in fluorescence level is used to determine the energy that the ion acquired during @xmath21 , and thus deduce the ion heating rate .
application of the doppler recooling method as described in the original proposal is practical for situations where the ion mode which is doppler cooled does not experience significant micromotion , for example to cooling of the axial mode of a linear paul trap . while this geometry is frequently sufficient , it is not applicable whenever a single doppler cooling beam is available , as in situations where planar traps are part of a complex setup granting limited optical access to the ions . in these cases
micromotion contribution to doppler broadening can be significant and has to be taken into account in determining ion energy from changes in fluorescence .
following the procedure discussed in @xcite we numerically find that for our experimental parameters micromotion contributes significantly to doppler broadening , and can lead to an overestimation of the ion energy by a factor between 2 and 4 if the method is applied without accounting for micromotion .
nevertheless , the numerical integration required to obtain and fit fluorescence recooling curves to experimental data is computationally intense , which makes it practically infeasible . in the case of @xmath4ca@xmath5ions ,
further complications to application of the doppler recooling method arise due to the electronic level scheme which is more comlicated than that of a simple two - level atom . in order to estimate the ion energy , the linewidth of the resonance and the saturation parameter(s )
are required . for two level systems
this is straightforward , however for multi - level dynamics the situation is more complex . in particular , dark - resonances and the strength of the repump light field(s ) have significant influence on the line shape @xcite .
one possible resolution is the detailed modeling of the dynamics and careful measurement of all relevant laser parameters such as detunings and intensities . nevertheless , due to the complexity of the situation , instead of modelling the recooling dynamics in the presence of micromotion and multiple atomic levels we use here a simple workaround by calibrating the energy scale with well - defined electric field noise .
apart from ease of implementation a further advantage is that one can separate out the effect of different energies of the three motional modes on the fluorescence signal by choosing particular noise frequency bands and/or excitation geometries , i.e. the electrodes to which the noise is applied .
thus , one can study or avoid implicit assumptions while extracting a single heating rate for the three motional modes . for these experiments ,
the intensity of the detection laser at 397 nm is actively stabilized at a value of 38@xmath12 .
the frequency of the detection laser is detuned to 5mhz on the red side of the resonance of @xmath22
. the intensity of the 866 nm repump laser is adjusted at 115@xmath12 , and its frequency is adjusted to the @xmath23 transition by maximizing the fluorescence level . to measure recooling curves , we turn off doppler cooling and allow the ion to heat by switching off the repump laser at 866 nm for the variable time @xmath21 . during a time interval @xmath24
the ion gets pumped into the @xmath25 level and will not be cooled by the doppler cooling laser .
after the repump laser is turned on , ion fluorescence is acquired using time bins of @xmath26 .
the procedure is repeated typically 1000 times and the fluorescence curves are averaged to improve statistics .
the inset in fig.[fig : pvso ] shows a typical recooling curve obtained using this procedure . to extract the ion energy from the recooling curves
, we use the analysis procedure of @xcite and extract a quantity @xmath27 , to which we will refer as the scaled energy .
this would correspond to the ion energy in the case of a two - level ion experiencing no micromotion , but in our case it is found to be merely proportional to the ion energy .
the latter is evident in the linear increase of @xmath27 with heating time @xmath21 , shown in fig .
[ fig : pvso ] . to calibrate the scaled energy , we heat the ion with externally supplied white electric field noise provided by a signal generator that has been calibrated against a spectrum analyzer .
the noise is white to a very large degree , with spectral density at the ion secular frequencies varying by less than 10 @xmath28 .
we vary the noise spectral density between @xmath29 and @xmath30 .
the noise is applied to one of the trap electrodes with almost equal projection of the electric field on all the three modes of ion motion and the corresponding electric field noise at the ion position is changed between @xmath31 and @xmath32 .
the calibration is performed by determining the heating rate @xmath33 from the recooling curves and comparing it to the expected ion heating rate @xmath34 , at various levels of applied external noise .
the latter is calculated using @xcite @xmath35 where @xmath21 is the time the ion is allowed to heat before measuring its energy , @xmath36 is the secular frequency of the @xmath37 mode of the ion @xmath38 is the power spectral density of the electric field at that frequency . in this expression heating of the secular sidebands of the micromotion at @xmath39
is omitted , since it is expected to be weaker by a factor @xmath40 which is significantly below the accuracy of this method .
heating rates are determined for different values of the externally applied noise .
the results of such measurements are shown in fig .
[ fig : mesevsinde ] , where we plot @xmath33 versus @xmath34 .
we find linear dependence on time scales about one hour , but the slopes between different trends taken several hours apart vary by up to a factor of roughly 2 .
this is most likely due to drifts in the experimental parameters that we do not stabilize accurately , namely the repump laser detuning and intensity . from linear fits to these curves
we determine @xmath27 to be a factor of @xmath41 higher than the actual energy acquired by the ion . as a result , the uncertainty in our determination of heating rates of our trap is ultimately determined by the uncertainty in the calibration of @xmath27 vs @xmath42 .
heating rate measurements were performed at different positions along the axis of the trap , as shown in fig .
[ fig : alongtheaxislog ] .
the measured results have been normalized to phonons of 1 mhz frequency .
the single filled data point at axial position of 2250 @xmath43 was measured one and a half months after the installation of the trap in vacuum and the open data points were obtained later , after an abrupt increase in heating rates of our trap .
this occurred approximately after two months of trap operation .
it coincided with a failure of the ion pump used to maintain the ultra - high vacuum and was followed by an increase in the stray electrostatic fields at the ion positions @xcite . during the pump failure
elevated pressures of @xmath44 mbar were reached in the vacuum chamber , while before and after the pressure was in the low @xmath45 range .
the heating rate appears to have a maximum around the region used for ion loading in the trap .
this is the range of axial positions between roughly 1900 @xmath0 m and 2700 @xmath0 m , which was used extensively as an ion - loading zone .
far from the loading region the heating rate is close to the values measured in the `` pristine '' trap @xmath46 months after trap installation in vacuum .
the heating rates near the loading zone of the trap increased from @xmath47 phonons / ms to more than @xmath48 phonons / ms in the first few months of trap operation . far from this zone
, the heating rates maintained their lower value .
likewise , uncooled ion lifetimes decreased by one order of magnitude in the parts of the trap where increased heating rates were observed .
the change in heating rates coincided with an abrupt increase in stray static fields in the same region of the trap @xcite and it is very likely that both of these effects are related to extensive use of the region around axial coordinate of 2250 @xmath0 m as a loading zone .
a number of effects can be responsible for the change in trap behavior . from the geometry of our vacuum apparatus
, we expect that possible ti contamination from the titanium sublimation pump is rather homogeneous .
similarly , surface contamination with atomic ca from the oven is not expected to peak at the loading zone , located near the center of the trap , but rather to fall off monotonically moving towards the oven direction as a result of the atomic beam flux screening , discussed in sec .
[ sec : setup ] .
one possible meachanism of surface contamination is bombardment of the trap electrodes by electrons created during trap loading or by @xmath4ca@xmath5ions which are created by photoionization outside the trapping volume .
these can impinge on the gold surface with energies up to 100 ev when accelerated by the rf field .
@xmath4ca@xmath5ions with energies of a few ev can get physisorbed on the trap electrodes and later form chemical compounds , while ions at higher energies can sputter material from the trap surface .
in addition , it is possible that the laser light used for ion creation and detection locally alters the chemical composition .
as the mechanism leading to heating of trapped ions is not fully understood , it is useful to summarize heating rates in different traps constructed from different electrode materials and used to trap different ion species . since the noise spectral density , @xmath49 , causing anomalous heating
is believed to scale as @xmath3 with frequency@xcite , it is convenient to regularize experimental results by deriving @xmath50 @xcite where @xmath51 is the frequency of the ion mode which is being measured .
this is straightforwardly done for heating rate measurements performed using the sideband spectroscopy method , where the heating rate of a specific secular mode can be measured . in order to regularize results measured using the doppler recooling method one needs to take into account that the energy of all three secular modes is redistributed and finally removed from the ion , as discussed in @xcite .
thus , a properly averaged `` effective '' frequency has to be used . by energy conservation considerations , we find that the effective frequency is @xmath52 . in our measurements ,
the axial frequency is significantly smaller than the radial ones and @xmath53 turns out to be very close to the axial frequency . using this approach ,
we have compiled a graph of representative data found in the literature @xcite , see fig .
[ fig : heat - rate - summary ] . in the graph
we specify ion species used in the measurements and trap electrode material , as well as temperature of the trap , if other than room temperature .
we include some of the measurements performed in cooled trap apparatuses , but focus on the room temperature results where measurements are abundant .
we observe that the majority of traps measured at the same temperature fall around a general trend following a @xmath54 law ( corresponding to the diagonal of the figure ) .
the traps with the lowest heating rates have been a molybdenum ring trap @xcite used to trap @xmath55 and a gold - on - quartz surface trap @xcite used to trap @xmath56 .
nevertheless , the presence of both gold and molybdenum traps with noise spectral densities well within the average implies that more than material - specific the source of the noise is specific to trap preparation and cleaning procedures .
as we discuss in what follows , less than one monolayer of adsorbed impurity atoms , as well as a thin dielectric layer formed on the trap surface can be responsible for the observed heating rates .
summary of representative noise spectral densities for different traps found in the literature , plotted versus ion - trap distance @xmath2 .
the ion species and trap electrode material are specified .
temperature is specified when different than room temperature .
references are organized in chronological order of appearance .
the trap used in this work fell well within the average trend initially ( lower point , corresponding to heating rate measured @xmath57 months after installing the trap in vacuum ) , but still falls within the limits of the trend even after deterioration of trap performance .
the diagonal in the figure corresponds to a @xmath54 trend .
data are from references @xcite . ]
we now proceed to derive a phenomenological model of ion heating that can provide physical information to help identify the possible sources of anomalous heating in ion traps .
the model is required to reproduce the correct scaling of noise spectral density with ion - trap distance and with frequency .
thus we consider a distribution of fluctuating dipoles on the trap electrode surface to be the source of anomalous heating . in many situations
@xmath3 noise is attributed to systems with a debye - type relaxation , and a distribution of relaxation rates @xcite .
accordingly , we consider a typical dipole with dipole moment @xmath0 , relaxation rate @xmath58 , leading to an autocorrelation function @xmath59 the corresponding dipole spectral density is @xmath60 to recover the @xmath3 noise spectrum , we assume a distribution of relaxation rates scaling as @xmath61 , where the normalization @xmath62 depends only logarithmically on the limiting relaxation rates . in the context of thermally
activated dipole relaxation @xmath63 corresponds to the range of energy barriers through which relaxation can occurr .
the dipole moment spectral density is now obtained @xmath64 fluctuating dipoles on the trap electrodes produce fluctuating electric fields of order @xmath65 at the ion position , a distance @xmath66 away .
this relation will not be exact , since a fluctuating dipole near a conducting surface will get either enhanced or screened depending on its orientation @xcite , and the exact mechanism of screening of charges by metal surfaces at separations of the order of interatomic distances is unclear @xcite , but it will suffice for our current estimates . moreover , fluctuating electric dipoles will produce negligible magnetic field noise .
the reason is that in the near - field of dipole radiation the electromagnetic field is predominantly electric .
the near - field condition is always satisfied in current experiments , so this mechanism will cause electric field noise and virtually no magnetic field noise . to estimate the electric field noise spectral density , we perform statistical averaging which results in scaling as @xmath67 with the number , @xmath68 , of dipoles . the noise spectral density from a planar trap with surface density @xmath69 of dipoles
is then given by the integral over the trap surface @xmath70 the rudimentary result of eq .
[ eq : spectral - density - final ] allows us to estimate the product @xmath71 of the density of dipoles on the trap surface with the squared dipole moment , based on the measured noise spectral density , secular frequency , and ion - trap distance .
we use this result to evaluate different possible sources of anomalous heating due to contaminated trap electrode surfaces . by assuming a typical dipole moment of 1 d and value of normalization constant @xmath72 as reasonable estimates , the value of the average trend of heating rate data in fig .
[ fig : heat - rate - summary ] leads to @xmath73 .
this value corresponds to typical surface densities of atoms on solid surfaces .
first we consider impurity atoms adsorbed on the trap electrodes , following the analysis of volokitin _ et .
_ @xcite . in this case
, a metal surface with small coverage by adsorbed molecules can give rise to dissipation and noise several orders of magnitude higher than those expected from a clean metal surface .
to put this in the context of our model , we consider adsorbed atoms on the metal surface which form polar bonds and can switch between different configurations giving rise to fluctuating dipoles . as a typical dipole moment
we take the value for cs adsorbed on cu(100 ) as considered by the above authors in @xcite i.e. @xmath74 4 d. using eq .
[ eq : spectral - density - final ] , we obtain a coverage fraction @xmath75 for a disordered gold surface , which is a very reasonable result . while in the case of tunneling tls , which we consider next , the typical ranges of activation energies and relaxation times are somewhat known and well understood , such details are in general unclear for surface adsorbates .
nevertheless , this mechanism requires less than one monolayer of adsorbate and can , as a result , be very common unless specific care is taken to target and remove material - specific adsorbates from metal surfaces .
we also consider tunneling tls in an amorphous dielectric film on the trap electrodes as the fluctuating dipoles .
the thickness of the dielectric layer can be estimated as @xmath76 , where @xmath77 is the density of tls in the dielectric .
while this quantity is relatively well determined at temperatures in the few k range @xcite , the situation at higher temperatures is unclear . it has been argued that the tls density of states increases linearly with temperature @xcite , and according to that estimate one obtains at room temperature a density @xmath78 of tls with energies less than the estimated upper limit of @xmath79 @xcite . with the above value of @xmath80 , this would correspond to a dielectric of thickness @xmath81 .
while this is a high value , it is only intended as an order of magnitude estimate , given the crudeness of the approximations used .
it is possible that oxides on metal surfaces , residues of fabrication , gases adsorbed on the trap electrodes , and material deposited on trap surfaces while the trap is in operation form dielectric layers in the nanometer range .
we reported a simple fabrication method for segmented planar ion traps , suitable for ion transport and for operation with a tilt of the main trap axes with respect to the direction perpendicular to the trap plane .
we described a modification of the doppler recooling method for heating rate measurements and used the method to measure heating rates of one gold - on - sapphire planar trap .
the heating rates were found to deteriorate with time , and to be position dependent with a maximum around the loading zone of the trap , suggesting surface contamination as a major source of anomalous heating .
we presented a phenomenological model that can explain anomalous heating in the ion traps and used it to evaluate different possible sources of anomalous heating encountered in other systems .
we found that both adsorbed impurity atoms close to a density of one monolayer and tunneling tls in amorphous dielectrics on the trap electrodes can be responsible for the observed heating .
careful surface characterization and cleaning procedures of ion traps are needed to have a more definite conclusion on the sources of and solutions to anomalous heating .
the authors would like to thank john martinis for useful discussions and d. frank ogletree for pointing out important work on non - contact friction .
this work was supported by the austrian ministry of sciences with a start grant .
n. daniilidis was supported in part by the european union with a marie curie fellowship .
s. narayanan was partially supported by the laboratory directed research and development program of lawrence berkeley national laboratory under u.s .
department of energy contract no .
de - ac02 - 05ch11231 , and st .
schulz was supported by the alexander von humboldt foundation with a feodor lynen fellowship .
s seidelin , j chiaverini , r reichle , j j bollinger , d leibfried , j britton , j h wesenberg , r b blakestad , r j epstein , d b hume , w m itano , j d jost , c langer , r ozeri , n shiga , and d j wineland .
, 96:253003 , 2006 .
j britton , d leibfried , j beall , r b blakestad , j j bollinger , j chiaverini , r j epstein , j d jost , d kielpinski , c langer , r ozeri , r reichle , s seidelin , n shiga , j h wesenberg , and d j wineland . .
, 2006 .
d r leibrandt , j labaziewicz , r j clark , i l chuang , r epstein , c ospelkaus , j wesenberg , j bollinger , d leibfried , d wineland , d stick , j sterk , c monroe , c .- s .
pai , y low , r frahm , and r e slusher . . , 2009 .
r j epstein , s seidelin , d leibfried , j h wesenberg , j j bollinger , j m amini , r b blakestad , j britton , j p home , w m itano , j d jost , e knill , c langer , r ozeri , n shiga , and d j wineland . .
, 76:33411 , 2007 . | we report heating rate measurements in a microfabricated gold - on - sapphire surface electrode ion trap with trapping height of approximately 240 @xmath0 m . using the doppler recooling method
, we characterize the trap heating rates over an extended region of the trap .
the noise spectral density of the trap falls in the range of noise spectra reported in ion traps at room temperature .
we find that during the first months of operation the heating rates increase by approximately one order of magnitude .
the increase in heating rates is largest in the ion loading region of the trap , providing a strong hint that surface contamination plays a major role for excessive heating rates .
we discuss data found in the literature and possible relation of anomalous heating to sources of noise and dissipation in other systems , namely impurity atoms adsorbed on metal surfaces and amorphous dielectrics . |
artificial kagome spin ice is a well - known geometrically frustrated two - dimensional magnetic system.@xcite it consists of elongated ferromagnetic nanomagnets placed on the nodes of a kagome lattice , or equivalently on the bonds of the hexagonal lattice ( see fig . ) , and coupled via their dipolar magnetic fields .
most of the experimental studies on artificial spin ice have been performed on static systems@xcite or systems with slow magnetization dynamics@xcite using imaging techniques such as magnetic force microscopy ( mfm ) and x - ray photoemission electron microscopy ( peem ) . in recent years , however , thermally active artificial spin systems have gained a considerable interest@xcite and provided a successful route to reach the low - energy magnetic states of artificial spin ice.@xcite additionally , analysis of the thermal behavior of nanomagnetic systems is particularly important for understanding the limitations of future spintronic devices.@xcite however , as the dynamics of artificial spin systems gets faster , observations using traditional microscopy techniques become limited by their temporal resolution ( approximately 1 s for peem ) , so that the magnetic correlations of the systems with rapidly fluctuating moments can not be probed .
therefore , complementary techniques are needed to study magnetic correlations in systems with faster fluctuation timescales in order to gain information about the magnetic phases over a broad temperature range .
this is particularly of interest for thermally - induced magnetization dynamics,@xcite order - disorder transitions@xcite and spin - wave excitations in magnonic crystals.@xcite here we show that soft x - ray resonant magnetic scattering ( sxrms ) is a highly sensitive momentum - resolved technique for studying magnetic correlations in mesoscopic systems with fast magnetization dynamics .
field - driven studies of athermal systems using this technique have already been reported , with measurements of the x - ray magnetic circular dichroism ( xmcd ) signals at the bragg peaks.@xcite such measurements can directly give information about the ordering in the sample or the net magnetic moment .
however , systems without long - range order , but rather short - range correlations , are characterized by _ diffuse _ scattering with a relatively weak xmcd signal . in the present work ,
we focus our attention on a highly dynamic regime of thermally - active artificial kagome spin ice .
the magnetic correlations in this regime have not yet been explored due to difficulties in capturing the weak magnetic diffuse signal , which we overcome by masking out the bragg peaks . comparing experimental scattering patterns with the patterns calculated from monte carlo simulations , we observe the emergence of quasi - pinch points in the kagome ice i phase of artificial kagome spin ice , and
explain their relation to the pinch - point singularities in both in - field@xcite and zero - field spin ice pyrochlores.@xcite while we currently can not access the kagome ice ii phase experimentally , we develop a theoretical understanding of how genuine pinch points in the kagome ice ii phase emerge smoothly from precursors or quasi - pinch points in the structure factor of the kagome ice i phase by virtue of breaking the @xmath0 symmetry associated with magnetic charge order.@xcite these sharp pinch points in the magnetic structure factor@xcite are characteristic features of the magnetic coulomb phase .
our experimental data resolving x - ray magnetic diffuse scattering in highly dynamic artificial kagome spin ice confirms the existence of kagome ice i phase correlations and thus provides convincing evidence for the picture of a classical spin - liquid that thermally samples the large manifold of near degenerate spin - ice states .
our 2@xmath12 mm@xmath2 arrays of nanomagnets , that are superparamagnetic at room temperature , were produced using electron beam lithography .
the nanomagnets have an elongated shape with a length of 62 nm and a width of 24 nm , and are placed on the nodes of a two - dimensional kagome lattice with lattice vectors of length 170 nm ( see inset of fig . ) .
a 70 nm - thick polymethylmethacrylate ( pmma ) layer was spin - coated on a si ( 100 ) substrate .
the patterns were exposed in the resist using a vistec ebpg electron beam writer operated at 100 kev accelerating voltage .
after development , a 5 nm - thick permalloy ( ni@xmath3fe@xmath4 ) film was deposited by thermal evaporation and capped with about 3 nm of al to prevent oxidation . the remaining resist with unwanted metallic material
was removed in acetone by ultrasound - assisted lift - off .
plane group .
@xmath5 , @xmath6 and @xmath7 are the structural lattice unit vectors .
thin gray lines represent mirror planes and serve as a guide for the eye .
the experimental scattering plane is parallel to the crystallographic direction @xmath8121@xmath9 indicated by the black arrow .
there are three nanomagnets per unit cell , indicated by the three nanomagnets shaded in black.,title="fig:",width=264 ] [ fig : figure1 ] to model the behavior of artificial kagome spin ice , the monte - carlo method was used .
we have generated magnetic moment configurations for a kagome lattice with ising degrees of freedom by thermal monte - carlo sampling of an array of 24@xmath124 unit cells using the implementation of m@xmath10ller and moessner.@xcite each nanomagnet is modeled as a needle dipole with uniform magnetic moment density @xmath11 along the length @xmath12 of the nanomagnet,@xcite and we use the value of @xmath13 corresponding to our sample geometry throughout this paper , where @xmath14 is the distance between the vertices of the kagome lattice or the bond length of the parent hexagonal lattice .
the interaction between such needle dipoles is equal to the magnetostatic energy of pairs of magnetic charges @xmath15 at the tips of each dipole , and the value of the ising spin defines which end hosts the positive charge .
this gives rise to an effective description of magnetic configurations in the system in terms of the total magnetic charge @xmath16 at a vertex , i.e. the sum of charges at the three dipole tips closest to it ( a detailed picture of the energetics in the system can be obtained in a multipole expansion of the available charge configurations).@xcite using this representation , the paramagnetic phase is fully disordered allowing all possible random configurations with @xmath17 or @xmath18 at each vertex . in the kagome ice i phase , the ice rule ( `` two in , one out '' or vice versa at each vertex ) enforces @xmath17 , with an exponentially suppressed population of @xmath18 excitations upon lowering the temperature .
the kagome ice ii phase is fully ordered in terms of charges but still carries a macroscopic degeneracy in terms of spin configurations , with a finite entropy per magnetic moment.@xcite finally , the ( six - fold degenerate ) ground state configuration forms a crystal of ` loops ' of spins with long range order ( lro ) in both charge and moment ordering . for more details about these phases
see ref . .
the sampled configurations of the magnetic moments from the kagome ice i and paramagnetic phases are then used to numerically simulate the sxrms patterns that are compared with the experimental results .
sxrms experiments were performed using the resoxs chamber@xcite at the sim beamline@xcite of the swiss light source , paul scherrer institut .
experimental scattering patterns were measured at an x - ray energy of 708 ev corresponding to the fe l@xmath19 edge and at 690 ev ( below the edge ) to verify that the scattering signal at resonance is of magnetic origin . the off - specular reflection geometry ( fig . )
ensures sensitivity to the in - plane magnetic moments of the sample .
the angle of incidence @xmath20 is kept constant throughout the experiment .
the diffraction patterns were acquired for four seconds with a princeton instrument pme charge - coupled device ( ccd ) camera with 1340@xmath11300 pixels ( 20@xmath120 @xmath21m@xmath2 pixel size ) . using the ccd camera ,
an extended fraction of reciprocal space can be recorded simultaneously,@xcite which is ideally suited for capturing diffuse scattering .
however , the ccd detectors used for soft x - rays have a lower dynamic range compared to the cmos hybrid photon counting detectors developed for hard x - rays that are used for detecting structural diffuse scattering.@xcite as a result of the low dynamic range , it is much harder to capture diffuse scattering in the soft x - ray scattering experiments , since the diffuse intensity is usually three orders of magnitude lower than the bragg peak intensity .
in addition , the magnetic scattering contribution is usually about two to three orders of magnitude lower than the charge scattering contribution .
this results in five to six orders of magnitude difference in intensity between the magnetic diffuse and the structural bragg scattering , which is beyond the dynamic range of the standard ccd detectors . to separate the two
, we therefore placed a custom - made arc - shaped aluminum mask in front of the ccd detector to block the high intensity bragg peaks and the specular reflection .
thus features in diffuse magnetic scattering with intensities @xmath22 of the bragg peak intensities could be resolved .
this simple technique opens the possibility to investigate magnetic diffuse scattering , and the associated magnetic correlations , that might otherwise be overlooked in such experiments . in order to numerically simulate the two - dimensional sxrms patterns , we make use of kinematic scattering theory as described in refs . and .
for more details about the numerical implementation of this theory see ref . and , in particular , our previous work ref .
. to directly compare the simulated pattern with the experiment , only the magnetic contribution to the scattering signal was calculated .
we assume that the nanomagnets are homogeneous permalloy particles supporting a single magnetic domain , so we obtain three distinct form factors , one for magnets on each of the three sublattices in the unit cell of the kagome lattice . improving on the method in ref .
, in this work we also include the scattering geometry and the reflectivity of the sample , which is important to reproduce the shape of the scattering pattern obtained on the 2d detector and the intensity decay in the @xmath23 direction .
taking this into account , we consider a plane wave scattered by the sample and collected at each detector pixel with a momentum transfer @xmath24 . the simulated scattering pattern is then corrected for the fresnel reflectivity from a flat interface , approximated as @xmath25 at high angles,@xcite where @xmath26 is the wavevector transfer in the scattering plane and @xmath27 is the wavevector transfer at the critical angle .
synchrotron x - rays .
@xmath28 , @xmath29 are the incident x - ray wavevector and angle , and @xmath30 , @xmath31 are the final wavevector and angle . @xmath32 and @xmath33 are the momentum transfers along the z and x directions , respectively . the orientation of the nanomagnets relative to the scattering plane is shown schematically in the inset.,title="fig:",width=325 ] [ fig : figure2 ] = 160 k is indicated with an arrow .
the standard errors are smaller than the data markers.,title="fig:",width=283 ] [ fig : figure3 ]
in order to investigate the magnetic correlations in the dynamic regime , our samples are designed to achieve an appropriate balance between the magnetic anisotropy energy , the energy associated with the dipolar interaction between the nanomagnets and the thermal energy of the system .
the interplay between these energies results in a blocking temperature @xmath34 . tuning the lateral size and thickness of the nanomagnets , as well as the distance between them , we move the blocking temperature to sufficiently low values to ensure fast fluctuations of the magnetic moments , while at the same time keeping the interaction energy between the nanomagnets high enough to preserve collective magnetic behavior .
the blocking temperature @xmath34 of the nanomagnets was estimated from zero field - cooled and field - cooled ( zfc / fc ) magnetization measurements performed with a superconducting quantum interface device ( squid ) magnetometer , as shown in fig . .
a characteristic peak is found at the blocking temperature @xmath34 that represents a crossover from static to dynamic behavior and is associated with an average energy barrier of the magnetic switching process.@xcite for our samples the blocking temperature is around 160 k. in order to probe magnetic correlations in our highly dynamic artificial kagome spin ice , diffuse scattering patterns were measured at several temperatures above the blocking point .
eleven scattering patterns were taken at each temperature to check that the scattering pattern did not change in time and therefore confirm that the sample is thermally equilibrated . to single out the magnetic scattering part ( figs . and ) , we took the difference between on - resonance and off - resonance patterns at x - ray energies of 708 ev and 690 ev , respectively .
the off - resonance patterns feature no magnetic diffuse scattering and provide a background intensity .
it should be noted here that the arcs of small peaks in the experimental patterns in figs . and arise due to contamination from higher - order harmonics of the undulator .
the dark blue broad stripes are the shadows of the aluminum mask , which is used to cover the structural bragg peaks .
the specular reflection is covered by the mask and is estimated to be at @xmath35 and @xmath36 . [ fig : figure4 ] at the lower temperature of 200 k , the scattering has bands of intensity oriented along @xmath23 , indicating the presence of magnetic correlations ( fig . ) .
the stripes become broader near @xmath35 , forming elongated diamond - shaped features that appear to touch at @xmath37 ( the touching point is obscured by the mask ) . as we will argue below
, this feature is related to the formation of quasi - pinch points " in the structure factor . since the temperature of the system is above the blocking temperature @xmath34 = 160 k , the system is dynamic and the magnetic correlations arise from the ice - rule constraints .
this constrained collective motion is characteristic of classical spin ices and spin liquids.@xcite at the higher temperature of 280 k , the scattering signal becomes uniform indicating that the thermal energy at this temperature is sufficient to break the ice - rule short range correlations of artificial kagome spin ice and it is in a disordered state ( fig . ) .
it should be noted that the intensities of the experimental patterns are slightly asymmetric with higher intensity at negative @xmath38 .
this is due to a small misalignment of the sample from the @xmath8121@xmath9 crystallographic direction .
to understand the observed diffuse magnetic scattering patterns , we match our experimental data to predictions based on monte carlo simulations . given a set of spin configurations from a monte - carlo run
, scattering patterns were obtained using kinematic scattering theory,@xcite as described in the methods section [ sec : methods ] .
the calculated scattering patterns corresponding to experimental results are shown in figs . and .
experimental data at 200 k visually matches the calculated scattering patterns in the kagome ice i phase .
hence , the scattering pattern reflects ice - like correlations resulting from the ice rule `` two moments in one moment out '' or vice versa and vertex charges of @xmath39 are energetically enforced .
the experimental data taken at 280 k matches the scattering patterns calculated from paramagnetic phase configurations . at this temperature , the simulations indicate that the ice rule is frequently violated and vertices with a total charge of @xmath40 are common . at both temperatures ,
we have used an overall prefactor to scale the intensity of numerically simulated patterns from monte carlo configurations to match the experimental data .
also , an overall broadening of the experimental data compared with the numerically simulated patterns can be attributed to the limited resolution of the instrument and small variations in the shape of the nanomagnets which produce an additional structural and magnetic diffuse background .
given the characteristic differences in the scattering within these high- and low - temperature regimes , we can conclude that , from the magnetic diffuse signal obtained with soft x - ray resonant magnetic scattering , we can distinguish magnetic correlations associated with different phases in an artificial spin system .
bulk 3d spin ice realizes a cooperative paramagnet where spins fluctuate in a correlated manner between low energy states.@xcite dipolar correlations are present despite the absence of long - range order and are one of the most interesting features of spin ice systems.@xcite in reciprocal space , their characteristic hallmarks are sharp singularities in the spin structure factor known as bow ties or pinch points .
they were initially found in ferroelectrics@xcite and more recently , their presence was predicted for many magnetic systems,@xcite and they have been observed in both in - field@xcite and zero - field spin ice pyrochlores.@xcite pinch points reflect the presence of a local divergence free condition of atomic spins , when considering the lattice flux where the local ice rule is obeyed . in the pyrochlore spin ice
, the ice rule refers to the local energy minimization condition of spins on the corners of a single tetrahedron with two spins pointing towards and two spins pointing away from the center.@xcite in order to determine whether pinch points may in principle be observed for artificial kagome spin ice , we have calculated the spin structure factor .
this is the most familiar quantity and is independent of the particular experimental setup in a neutron or x - ray scattering experiment . as an experimentally accessible signature
, we have also calculated the magnetic diffuse scattering in a transmission geometry from our monte carlo simulations .
results for these are shown in fig . .
for the patterns in the transmission geometry , that we use also to assess the width of the pinch points , we assume an incident angle of 30@xmath41 and calculate the forward scattering of x - rays .
only the magnetic scattering contribution was calculated . in figs .
, we show data for the paramagnetic and kagome ice i phases using the same set of monte carlo configurations that we used for comparison with the experimental results shown in fig . .
in addition , we also show simulation data for the kagome ice ii phase in figs . and .
in this phase an additional magnetic bragg contribution appears at the pinch point positions of ( 01@xmath42 ) , ( 10@xmath42 ) , ( 0@xmath421 ) and ( @xmath4201 ) fig . .
however it is absent at the ( 1@xmath420 ) and ( @xmath4210 ) positions , which makes the latter ideal for the measurements of the pinch point intensity profiles , see insets in figs . , and .
we also note that the kagome ice ii phase can be distinguished from the kagome ice i phase by its partial magnetic order , as also shown in ref . .
evidence of this can be best seen by the magnetic bragg peaks appearing in kagome ice ii phase at , for example , ( 11@xmath43 ) or ( @xmath44@xmath44@xmath45 ) and equivalent positions [ see fig . ] .
it should also be noted that , for the scattering patterns of fig . , we have used the crystallographic conventions for the hexagonal planar group @xmath46@xcite with the unit cell defined in fig . .
to describe this planar hexagonal lattice , we employ @xmath47 indices , with the @xmath48 index being redundant . for the native kagome unit cell ,
the positions of the pinch points are described by @xmath47 indices that have @xmath49 values . in previous works
@xcite that considered the three - dimensional pyrochlore unit cell , the pinch point positions are described by @xmath50 @xmath51 indices .
[ fig : figure5 ] to show the evolution of the correlations across the different phases , let us first consider the structure factor . upon lowering the temperature from the paramagnetic phase , where the scattering signal is uniform fig .
, characteristic regions of high- and low - intensity scattering start to emerge and form quasi - pinch points in the kagome ice i phase ( fig . ) .
these subsequently evolve into sharp , singular features in the kagome ice ii phase ( fig . ) .
an analogous evolution of the of the pinch points is seen in the transmission geometry calculations [ see figs . , , and insets ] .
finally , in the reflection geometry of our experiments , we now recognize that the previously described touching of elongated diamond - shaped features at regular points in reciprocal space ( figs . and ) , should also evolve into sharper and sharper features that correspond to the appearance of the pinch points in fig . .
as already noted , the magnetic pinch points emerge at the positions of the structural bragg peaks ( not shown ) , some of which we have indicated in figs . and , for example at ( 01@xmath42 ) , ( 10@xmath42 ) , ( 1@xmath420 ) , ( 0@xmath421 ) , ( @xmath4201 ) and ( @xmath4210 ) at the edges of the first brillouin zone .
we now explain these observations in more detail .
the emergence of pinch points requires the presence of a local conservation law for lattice fluxes,@xcite which is one of the defining features of a coulomb phase .
in particular , the observation of pinch points in neutron scattering patterns has provided a direct experimental evidence for a field - induced and zero - field coulomb phase in pyrochlore magnets .
@xcite however , the trivalent vertices of the planar kagome lattice can not inherently realize conservation of flux , and only allow configurations with a local lattice divergence of absolute value @xmath52 .
nevertheless , the charge - ordered kagome ice ii phase can be shown to be a coulomb phase , due to the fact that it spontaneously breaks the @xmath0 sublattice symmetry due to the formation of magnetic charge order.@xcite in this symmetry broken phase , the allowed spin configurations are identical to the constraints found on the kagome planes of the three - dimensional pyrochlore spin ice with a magnetic field along the [ 111 ] direction that was experimentally investigated in refs . .
in this 3d situation , the sublattice symmetry of the ( buckled ) kagome planes of the pyrochlore lattice is inherently broken , as one sublattice features out - of - plane spins pointing along [ 111 ] direction , while the out - of - plane spins on the other sublattice point towards the negative [ 111 ] direction . in the presence of a field along the [ 111 ] axis , the flux along that direction is fixed and vertices in the kagome planes are forced to have alternating minority spins ( pointing towards or away from the center of a tetrahedron ) @xcite , which amounts to an explicit breaking of the sublattice symmetry .
the local flux conservation is fulfilled for the low - energy configurations in this three - dimensional setting and , by extension , the same is true for the kagome ice ii phase .
however , this 3d picture requires the global sublattice symmetry to be explicitly broken , and it would be desirable to develop a more local picture that can accommodate the spontaneous symmetry breaking of the ice i to ice ii transition .
indeed , such a viewpoint can be developed on the basis of a dimer mapping.@xcite let us consider the mapping from oriented hard - core dimers to a divergence - free configuration of fluxes on the links of the lattice by huse et al.@xcite .
indeed , due to the @xmath0 sublattice symmetry breaking of magnetic charge order , the kagome ice ii phase allows a unique mapping of spin configurations to oriented hard - core dimers on the kagome lattice:@xcite at each vertex , there is precisely one minority charge , which connects to the minority charge of one adjacent vertex , and uniquely identifies the corresponding link as carrying a dimer oriented from the negative minority charge to the positive one , while the other two links are empty.@xcite following huse et al.,@xcite we can then define the lattice flux carried by a link in terms of the occupation number @xmath53 of dimers on the link @xmath54 as @xmath55 using the fact that the total monopole charge @xmath56 has long - range order in the kagome ice ii phase and thus sets the orientation of the fluxes for the respective sublattices .
note that this lattice flux has twice the magnitude on the links carrying dimers compared with the flux on the empty ones .
therefore , one can picture a pair of fluxes entering one vertex , being carried to an adjacent vertex along the dimer , and re - emerging as two separate fluxes from there .
clearly , such `` flux dimers '' satisfy flux conservation .
thus , as in the case of the square and cubic lattices of huse et al.,@xcite the lattice divergence vanishes , and the field can be cast into the habitual language of a lattice gauge theory and forms a coulomb phase . as the kagome ice ii phase is a genuine coulomb phase with a local conservation law , its structure factor is predicted to feature sharp singularities at the pinch points.@xcite this is shown by the pinch point intensity profile in the inset of fig . .
to proceed with our discussion , we note that spin configurations can be uniquely mapped to a dimer covering only in the monopole charge ordered kagome ice ii phase .
by contrast , the kagome ice i phase has no long range charge order and thus must violate the divergence - free condition .
consequently , it does not allow for uniquely defined flux dimers as in eq . .
nevertheless , there will be charge ordered clusters in the kagome ice i phase so that , while a global dimer mapping does not exist , it is still possible to locally identify dimers within each charge - ordered cluster . as the structure factor is expressed in reciprocal space , and can be thought of as a spatial average over the system , the overall structure encoded in the local conservation law of the low - temperature ordered phase
is therefore also observed in the kagome ice i phase .
additionally , the larger number of vertices that are not covered by flux dimers gives rise to the broadening of the pinch - points .
indeed , a broadening of the pinch points can be seen in our calculations of the structure factor and of the resonant magnetic scattering in a transmission geometry for the kagome ice i phase ( fig . ) . comparing the calculations from the different phases in fig .
, we note that the quasi - pinch - points in the scattering profiles of the kagome ice i phase are very similar to the fully developed pinch points of kagome ice ii phase .
however , following the arguments above , no sharp singularity can exist at the center of these quasi - pinch points .
finally in the paramagnetic phase there is a random distribution of @xmath39 and @xmath40 charges .
therefore a smoothly varying diffuse signal is established , as can be seen in figs . and .
it should be noted that , in general , the experimental detection of the pinch points in a transmission geometry is extremely challenging : due to the p6 mm symmetry of the lattice and the absence of systematic extinctions , structural bragg peaks are positioned at the same @xmath57 values as the pinch points , thus overlapping with them .
furthermore , a soft x - ray detector with the required dynamic range to capture both strong bragg scattering and a weaker magnetic diffuse scattering is not commonly available . despite these difficulties , our results show that it is possible to qualitatively distinguish the different phases of artificial kagome spin ice and capture features that are directly related to pinch points in the reflection geometry employed in our set - up . given these insights , we conclude that our experimental data provides the first evidence of the emergence of pinch - point scattering in artificial kagome spin ice .
these correlations are highly distinct from those found in the high - temperature paramagnetic phase , and present direct evidence that the ice - rules are obeyed , reflecting the spin dynamics that minimize the dominant nearest neighbor interaction terms of the needle dipole hamiltonian on the kagome lattice .
we have demonstrated that the measurement of diffuse scattering with resonant synchrotron x - rays provides a highly sensitive method for the investigation of short - range correlations in nanomagnet systems . in particular , we have shown that the phases of artificial kagome spin ice can be distinguished and comparison with monte carlo simulations confirms the realization of kagome ice i phase magnetic correlations at high moment fluctuation rates .
this combination of a highly dynamic system with strong short - range correlations is characteristic of a classical spin liquid ( or a cooperative paramagnet ) .
although the kagome ice i phase is not a genuine coulomb phase , since the divergence - free condition does not extend throughout the lattice , the features of quasi - pinch points already bear similarities to the fully - developed pinch points with sharp singularities associated with the kagome ice ii coulomb phase .
we have used a mapping to oriented dimer coverings to argue that such quasi - pinch points should indeed be expected above the phase transition to the charge ordered kagome ice ii phase .
we conclude that , like in the bulk pyrochlore spin ices , diffuse scattering gives a unique signature of ice - like behavior in artificial spin ices .
this technique also promises to cast light onto a number of other problems in nanoscale magnetic systems , for example to understand the ordering processes during magnetic self - assembly of nanoparticles@xcite or the formation of skyrmion lattices.@xcite this highly sensitive method will therefore be an important tool for the discovery of novel physics in two - dimensional systems beyond artificial spin ice .
we are grateful to v. guzenko , a. weber , t. neiger , m. horisberger and e. deckardt for their support in the sample preparation , n. s. bingham for his help with experiments , p. schifferle for technical support at the x11ma beamline , and t. fennell and r. moessner for helpful discussions .
part of this work was carried out at the x11ma beamline of the swiss light source , paul scherrer institute , villigen , switzerland .
the first soft x - ray magnetic diffuse scattering patterns were taken at the sextants beamline of synchrotron soleil , saclay , france under the proposal 20120875 .
the research leading to these results has received funding from the swiss national science foundation , the european community s seventh framework programme ( fp7/2007 - 2013 ) grant no .
290605 ( cofund psi - fellow ) and the royal society grant no .
uf120157 .
48ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ]
+ 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _
_ , vol .
( ) p. @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _
( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ ( , ) pp . @noop _ _ , vol .
( , ) pp . @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) | the study of magnetic correlations in dipolar - coupled nanomagnet systems with synchrotron scattering provides a means to uncover emergent phenomena and exotic phases , in particular in systems with thermally active magnetic moments . from the diffuse signal of soft resonant magnetic scattering
, we have measured magnetic correlations in a highly dynamic artificial kagome spin ice with sub-70 nm permalloy nanomagnets . on comparing experimental scattering patterns with monte carlo simulations based on a needle - dipole model ,
we conclude that kagome ice i phase correlations exist in our experimental system even in the presence of moment fluctuations , which is analogous to bulk spin ice and spin liquid behavior .
in addition , we describe the emergence of quasi - pinch points in the magnetic diffuse scattering in the kagome ice i phase .
these quasi - pinch points bear similarities to the fully developed pinch points with singularities of a magnetic coulomb phase , and continually evolve into the latter on lowering the temperature . the possibility to measure
magnetic diffuse scattering with soft x - rays opens the way to study magnetic correlations in a variety of nanomagnetic systems . |
liquid crystals are characterized by large thermal fluctuations in their local orientational order arising from collective alignment of the long axis of their constituent molecules @xcite . due to such soft anisotropy
, liquid crystals tend to respond easily to external forces .
confining geometries such as thin films , on which most applications of liquid crystals are based , change the fluctuation spectrum .
this can cause not only structural changes @xcite but also leads to fluctuation - induced effective forces between the substrates @xcite , also known as thermodynamic casimir effect . in correlated fluids such as liquid crystals this casimir force exhibits a universal power - law decay as a function of the separation between the substrates @xcite .
however , this behavior is modified in the presence of other characteristic scales in the system @xcite . in the case that the substrates are laterally modulated , discrete lateral modes of thermal fluctuations are also excited . under such conditions , in addition to the forces acting perpendicularly to the substrates , effective lateral forces arise @xcite with potentially interesting technological applications .
we study the influence of anchoring conditions , which vary periodically in one lateral direction , on the fluctuations of a uniformly ordered nematic liquid crystal .
obviously , the periodicity @xmath0 of the substrate pattern gives rise to an oscillatory behavior for the lateral force as a function of the lateral shift @xmath1 between the substrates .
for small inhomogeneities the lateral force is proportional to @xmath2 .
( the analysis of nonperiodic patterns would provide an understanding of nematic phases exposed to chemically disordered substrates . )
the present study actually extends our previous work @xcite where we considered the case in which only one of two confining substrates exhibits a chemical pattern so that there are no lateral forces . here ,
in addition to the fluctuation - induced lateral forces , we calculate the lateral force between the patterned substrates across the vacuum , i.e. , the background van der waals force acting parallel to the substrates .
this background force is generated by the necessary chemical modulations providing the laterally varying anchoring strengths . in sec .
[ ii ] our model and the theoretical formalism are specified . in sec .
[ iii ] the fluctuation - induced lateral force is obtained . in sec .
[ iv ] we calculate the lateral van der waals force between the patterned substrates .
the results for the fluctuation - induced normal force are presented in sec .
[ v ] and finally sec .
[ vi ] summarizes our results .
we consider a nematic liquid crystal confined by two flat but chemically patterned substrates at a separation @xmath3 . the patterns on the both substrates consist of the same periodic stripes of anchoring energies per area @xmath4 and @xmath5 along the @xmath6 direction but shifted relative to each other by the length @xmath1 ( see fig . 1 )
the substrates are translationally invariant in the y direction .
the stripes are considered to vary with respect to the strength of homeotropic anchoring so that the mean orientation of the director @xmath7 is spatially homogeneous but the thermal fluctuations vary laterally giving rise to effective lateral forces . and @xmath5 with the widths @xmath8 and @xmath9 , respectively . the wavelength of the periodicity is denoted as @xmath10 and the lateral shift between the origins of the patterns on the top and the bottom substrate is denoted by @xmath1 .
anchoring at both boundaries is homeotropic everywhere so that the thermal average of the director field @xmath11 is spatially homogeneous . ]
based on the bulk structural frank free energy @xcite given by @xmath12=\frac{1}{2}\int_{v}{\rm d}^3x\big[k_1(\mbox{\boldmath $ \nabla$}\cdot{\bf n})^2 + k_2({\bf n}\cdot \mbox{\boldmath $ \nabla$}\times{\bf n})^2 \nonumber\\+k_3({\bf n}\times \mbox{\boldmath $ \nabla$}\times{\bf n})^2\big ] , \label{eq0}\end{aligned}\ ] ] where @xmath13 is the nematic volume , @xmath14 , @xmath15 , and @xmath16 are the splay , the twist , and the bend elastic constants , respectively , the free energy of gaussian fluctuations in the one - constant approximation reads @xmath17=\sum_{i=1}^{2}{k\over 2}\int_{v } { \rm d}^3x\left[\nabla \nu_{i}({\bf x},z)\right]^{2 } , \label{eqe}\ ] ] where @xmath18 , @xmath19 , is either of the two independent components of the fluctuating part @xmath20 of the director @xmath7 , @xmath21 is the effective elastic constant , and @xmath22 are the lateral components of the cartesian coordinates @xmath23 . to describe the interaction of the liquid crystal and the substrates ,
we employ the rapini - papoular surface free energy given by @xmath24=-{1\over 2}\int_{s } { \rm d}^2x \,w^{z=0}({\bf x})\,({\bf n}\cdot \hat{{\bf z}})^2-\nonumber\\{1\over 2}\int_{s } { \rm d}^2x \,w^{z = d}({\bf x})\,({\bf n}\cdot \hat{{\bf z}})^2\ = f_s^{z=0}+f_s^{z = d } \label{eq3}\end{aligned}\ ] ] where @xmath25 is the anchoring energy per area @xmath26 and @xmath27 is the unit vector in z - direction . here
$ ] at the lower substrate located at @xmath29 and @xmath30 $ ] at the upper substrate located at @xmath31 where @xmath32 and @xmath33 describe the stripe modulations at the lower and the upper substrates , respectively , @xmath34 is the heaviside step function , and @xmath0 is the periodicity .
the stripes have the same width @xmath35 and are separated by sharp chemical steps . the functions @xmath36 and @xmath37 equal one at the regions characterized by @xmath4 and zero elsewhere at the lower and the upper substrates , respectively .
thus the surface free energy of the gaussian fluctuations given by @xmath38=f_{s}[\nu_{1}]+f_{s}[\nu_{2}]$ ] reads @xmath39={1\over 2}\big[w_a\int_{s}{\rm d}^2x\ , \big[\nu ( { \bf x},z=0)\big]^2 a(x ) \nonumber\\ + w_b\int_{s}{\rm d}^2x\ , \big[\nu ( { \bf x},z=0)\big]^2 [ 1-a(x)]\big]\ , \label{eq3}\end{aligned}\ ] ] at the lower substrate and @xmath40={1\over 2}\big[w_a\int_{s}{\rm d}^2x\ , \big[\nu ( { \bf x},z = d)\big]^2 b(x ) \nonumber\\ + w_b\int_{s}{\rm d}^2x\ , \big[\nu ( { \bf x},z = d)\big]^2 [ 1-b(x)]\big]\ , \label{eq3}\end{aligned}\ ] ] at the upper substrate . minimization of total free energy @xmath41 + f_{\rm s}^{z=0}[\nu_1,\nu_2]+f_{\rm s}^{z = d}[\nu_1,\nu_2]$ ] leads to two boundary conditions : @xmath42=0 , \ ; z=0,\label{eq : bca}\end{aligned}\ ] ] @xmath43=0 , \ ; z = d,\label{eq : bcb}\end{aligned}\ ] ] where @xmath44 is either @xmath45 or @xmath46 .
the normalized [ see , c.f .
, after eq .
( [ eq9 ] ) ] partition function @xmath47 of the fluctuating fields @xmath48 , @xmath49 , subject to the boundary conditions given by eqs .
( [ eq : bca ] ) and ( [ eq : bcb ] ) can be calculated within the path integral approach ( see ref .
@xcite and references therein ) .
the boundary conditions act as constraints which can be implemented by delta functions .
they , in turn , can be written as integral representations by introducing two auxiliary fields localized at @xmath29 and @xmath31 , respectively @xcite .
after performing the corresponding gaussian integrals over @xmath18 , @xmath49 , the path integral reduces to a gaussian functional integral over the auxiliary fields with a matrix kernel @xmath50 , so that obtaining the result for @xmath47 reduces to calculating @xmath51 . for the geometry considered here ,
the matrix @xmath50 is found to have the following matrix elements @xmath52 : @xmath53 \big[1+{\lambda_b-\lambda_a\over \lambda_a}a(x')\big]\nonumber\\ & + & { \lambda_b ( \lambda_b-\lambda_a)\over \lambda_a } [ a(x)-a(x')]\partial_z-\lambda^2_b\partial^2_z\big\ } g({\bf x}-{\bf x'},z - z{'})\big |_{z = z{'}=0},\nonumber\\ & m_{12}&({\bf x},{\bf x}{'})= \big[1+{{\lambda_b-\lambda_a}\over \lambda_a}[a(x')+b(x ) ] -2\lambda_b\partial_{z'}-{{\lambda_{b}(\lambda_b-\lambda_a)}\over \lambda_a}[a(x')+b(x)]\partial_{z'}\nonumber\\ & + & \big({{\lambda_b-\lambda_a}\over \lambda_a}\big)^{2}a(x ' ) b(x)+\lambda_b^{2}\partial^{2}_{z'})\big ] g({\bf x}-{\bf x'},z - z{'})\big |_{z = d , z{'}=0},\nonumber\\ & m_{21}&({\bf x},{\bf x}{'})= \big[1+{{\lambda_b-\lambda_a}\over \lambda_a}[a(x)+b(x ' ) ] -2\lambda_b\partial_{z}-{{\lambda_{b}(\lambda_b-\lambda_a)}\over \lambda_a}[a(x)+b(x')]\partial_{z}\nonumber\\ & + & \big({{\lambda_b-\lambda_a}\over \lambda_a}\big)^{2}a(x ) b(x')+\lambda_b^{2}\partial^{2}_{z})\big ] g({\bf x}-{\bf x'},z - z{'})\big |_{z{'}=d , z=0},\nonumber\\ & m_{22}&({\bf x},{\bf x}{'})=\big\{\big[1+{\lambda_b-\lambda_a\over \lambda_a}b(x)\big ] \big[1+{\lambda_b-\lambda_a\over \lambda_a}b(x')\big]\nonumber\\ & + & { \lambda_b ( \lambda_b-\lambda_a)\over \lambda_a } [ b(x')-b(x)]\partial_z-\lambda^2_b\partial^2_z\big\ } g({\bf x}-{\bf x'},z - z{'})\big |_{z = z{'}=d},\nonumber\\ \label{eq9}\end{aligned}\ ] ] where @xmath54 is the so - called extrapolation length and @xmath55 is the two - point correlation function of the scalar field @xmath18 in the bulk with its statistical weight given by @xmath56 where the normalizing factor @xmath57 is the bulk partition function and @xmath58 is the thermal energy . in terms of the partition function @xmath47 ,
the free energy is given by @xmath59 .
we note that normalizing @xmath47 by @xmath57 amounts to subtracting the bulk free energy , so that the free energy @xmath60 includes only the surface free energy and the finite - size contribution .
the surface free energy depends neither on @xmath3 nor on @xmath1 , so the fluctuation - induced force @xmath61 reads @xmath62 where @xmath63 is either @xmath64 or @xmath65 corresponding to lateral or normal displacements giving rise to lateral or normal forces , respectively .
the matrix kernel @xmath50 is a functional of the patterning function @xmath36 on the substrates and therefore calculation of the inverse of @xmath50 is nontrivial .
however , in systems with in - plane symmetries one may proceed by a lateral fourier transformation with respect to the lateral coordinates @xmath22 . in ref .
@xcite , it is shown how the electrodynamic casimir force can be calculated for a periodically modulated substrate ( see also refs .
@xcite ) . in this reference ,
the lateral periodicity is used to transform the matrix @xmath50 to a block - diagonal form in fourier space @xmath66 in which @xmath67 .
similarly , also here the matrix elements of the block @xmath68 with @xmath69 , @xmath70 , are given by @xmath71 for @xmath72 , where @xmath73 is the lateral extension of the system in the @xmath6 direction and the @xmath74 are @xmath75 matrices providing the following decomposition of the matrix @xmath50 : @xmath76 however , it is interesting to note that the matrix @xmath77 can also be represented , using a more direct derivation than in ref .
@xcite , in a form in which @xmath50 is diagonal @xcite . in view of the discrete lateral periodicity along the @xmath6 direction , it is suitable to express @xmath6 and @xmath78 as @xmath79 with @xmath80 and @xmath81 , @xmath82 .
since @xmath36 is periodic with wavelength @xmath0 it follows that @xmath83 and @xmath84 depends on @xmath85 and @xmath86 only via the difference @xmath87 .
this property and translational invariance along the @xmath88 direction imply the fourier decomposition @xmath89 where @xmath90 is diagonal .
furthermore with @xmath81 , @xmath91 one can form @xmath92 as expected , the different representations of @xmath50 in terms of the matrices @xmath93 or @xmath94 ( @xmath93 is the fourier transform of @xmath95 and @xmath94 is the fourier transform of @xmath96 ) do not change the final result for the force [ eq . ( [ force ] ) ] which is given by @xcite @xmath97 here @xmath98 denotes the partial trace with respect to the indices @xmath99 of the infinite - dimensional matrix @xmath94 ( or @xmath93 ) and we have taken into account the contribution of both fluctuating components of the director field . in the following we continue with the block - diagonal form of the matrix kernel @xmath50 since in this representation the patterning functions @xmath100 and @xmath101 are somehow simpler than their counterparts in the diagonal form of @xmath50 .
accordingly , the matrices @xmath102 [ eq . ( [ eq2 ] ) ] are given by @xmath103 \phi_{m}^{ba}(d)&\phi_{m}^{bb}(0 ) \end{array } \right ) + \delta_{m,0}\left ( \begin{array}{cc } { z_{a}^2\over 2p}-{\lambda_b^{2}p\over 2 } & { ( 1-\lambda_{b}p)^{2}\over 2p}e^{-pd } \\[3 mm ] \ { ( 1-\lambda_{b}p)^{2}\over 2p}e^{-pd } & { z_{b}^2\over 2p}-{\lambda_b^{2}p\over 2 } \end{array } \right ) \nonumber\\ & + & { \lambda_b-\lambda_a\over 2\lambda_a}\delta_{m,0 } \left ( \begin{array}{cc } 0 & \big[a_0\big({z_{b}^c\over p}-\lambda_b\big)+b_0\big({z_{a}^c\over p}-\lambda_b\big)\big]e^{-pd } \\[3 mm ] \big[a_0\big({z_{b}^c\over p}-\lambda_b\big)+b_0\big({z_{a}^c\over p}-\lambda_b\big)\big]e^{-pd}&0 \end{array } \right ) \end{aligned}\ ] ] for @xmath104 even , and @xmath105
b_m\big({z_{a}\over p}-\lambda_b\big)e^{-pd}+a_m\big({z_{b}\over p_m}-\lambda_b\big)e^{-p_{m}d } & ~~ b_{m}z_{b}\big ( { 1\over p}+{1\over p_m}\big ) \end{array } \right ) \ ] ] for @xmath104 odd , with @xmath106 , @xmath107 , @xmath108 , @xmath109 , @xmath110 , @xmath111 , @xmath112 , and @xmath113 where the prime at the summation sign indicates that in the sum the terms with even @xmath114 are excluded .
in the limit @xmath115 we find that the contributions from the elements @xmath116 [ eqs .
( [ eqb ] ) and ( [ eq2 ] ) ] to the force decrease rapidly with increasing absolute values of @xmath117 , so that the expression for the force [ eq . ( [ tforce ] ) ] converges already at small orders of @xmath50 with @xmath118 .
taking into account only the elements @xmath116 [ eq . ( [ eqb ] ) ] for @xmath119 , the asymptotic behavior of the fluctuation - induced lateral force @xmath120 in the limit @xmath121 is given by @xmath122 with the dimensionless function @xmath123 the lateral force oscillates as function of @xmath1 reflecting the underlying lateral periodic pattern and its magnitude decays exponentially as function of @xmath124 , because @xmath125 . for arbitrary values of @xmath3
, we evaluate the force in eq .
( [ tforce ] ) numerically .
although the matrix @xmath94 [ eq . ( [ eqb ] ) ] is infinite - dimensional , the value of the force saturates at some finite values for @xmath99 .
our numerical results for the fluctuation - induced lateral force as function of the shift @xmath1 for arbitrary strength of the contrast @xmath126 are shown in fig .
the lateral force acts against the increase of the lateral displacement @xmath1 in the interval @xmath127 by being a restoring force and acts favorably with the increase of @xmath1 in the interval @xmath128 $ ] by being a pulling force .
therefore the force is antisymmetric with respect to @xmath129 . upon approaching the maximum misalignment , i.e. , @xmath129 , the restoring force vanishes .
this implies that the interaction free energy @xmath130 has its maximum at @xmath129 where the opposing parts of the substrates face each other and attains its minimum at @xmath131 .
the force is maximal at @xmath132 and @xmath133 where @xmath134 exhibits its strongest dependence on @xmath1 ( fig .
[ fig00 ] ) . in fig .
[ fig3d ] we show the decay of @xmath135 as function of @xmath3 .
asymptotically , @xmath136 vanishes as @xmath137 for @xmath138 , @xmath139 , and @xmath132 .
we note that due to the assumption @xmath140 interchanging @xmath141 and @xmath142 leaves the system unchanged [ fig .
[ fig0 ] ] .
thus the force must be identical for @xmath143 . while in eqs .
( [ osci ] ) and ( [ osci1 ] ) this symmetry is explicitly valid up to the second order in @xmath144 ) , the numerical results respect this symmetry fully .
this provides a very useful check of the numerical calculations . in units of @xmath145 between two periodically patterned substrates at distance @xmath3 as function of the shift @xmath1 in units of the periodicity @xmath0 for @xmath146 , and @xmath147 ( see fig .
[ fig0 ] ) . the anchoring on the stripes in terms of the extrapolation lengths
is taken to be @xmath138 and @xmath139 .
@xmath148 is antisymmetric around @xmath129 .
there is no restoring force if the misalignment is maximal , i.e. , at @xmath129 .
@xmath149 means that the plates are pulled back towards the preferred alignment at @xmath131 ; for @xmath150 the plates are pulled forward towards preferred alignment at @xmath151 . ] in units of @xmath152 between two periodically patterned substrates at distance @xmath3 as function of the shift @xmath1 in units of the periodicity @xmath0 for @xmath153 , and @xmath147 and @xmath138 , @xmath139 . for all values of @xmath124
the inflection points of the potential are at @xmath132 and @xmath133 . ] in units of @xmath145 between two periodically patterned substrates as function of the film thickness @xmath3 in units of the periodicity @xmath0 .
the anchoring on the stripes in terms of the extrapolation lengths is taken to be @xmath138 , @xmath139 , and the lateral shift between the patterns is @xmath132 .
asymptotically , @xmath154 for @xmath132 vanishes as @xmath155 ( dashed line ) .
the variation of @xmath135 for small values of @xmath124 is shown in the inset for @xmath138 ( circles ) and @xmath156 ( triangles ) .
the other system parameters remain the same . as expected the absolute value of the amplitude of the lateral force increases upon increasing the contrast @xmath157 . ]
endowing the substrates with the envisaged stripe patterns requires corresponding chemical patterns which in turn involve at least two different species providing the chemical contrast .
these species do not only interact ( differently ) with the nematic liquid crystal in the vicinity of the substrate , giving rise to two different extrapolation lengths @xmath141 and @xmath142 , but also interact across the liquid crystal with each other via dispersion forces .
the later interaction provides a lateral force as well which adds to the fluctuation - induced lateral force .
note that such a lateral force due to direct interactions is the same if the two patterned substrates are separated by vacuum or by a nematic liquid crystal , as long as the _ mean _ nematic order is not affected by the stripe patterns . in order to estimate these direct interactions between the patterned substrates , we consider each substrate to be covered by a monolayer
whose chemical composition varies periodically , alternating between a- and b - particles .
we neglect non - additivity aspects of the dispersion forces and consider pairwise interactions between the patterned monolayers at @xmath29 and @xmath31 . since we consider @xmath3 to be large compared with the diameters of the @xmath158 and @xmath94 particles
, we can disregard that particles forming the monolayers occupy discrete lattice sites . as pair potentials between the two species we take lennard - jones potentials @xmath159
, ~ i , j = a , b\ ] ] where a(b)-particles give rise to the extrapolation length @xmath160 . since in the present context @xmath161 , for the lateral force only the attractive part of the pair potentials matters
this leads to the following expression for the van der waals potential energy between the two monolayers : @xmath162 ^ 3 } \label{23}\ ] ] with @xmath163\nonumber\\ + e_{bb}[1-a(x_1)][1-b(x_2;\delta)]\end{aligned}\ ] ] and @xmath164 where @xmath165 is the areal number density of @xmath166-particles in the monolayer forming the stripe @xmath160 . carrying out the integration over @xmath167 and @xmath168 in eq .
( [ 23 ] ) , one obtains @xmath169^{5/2 } } , \label{pot}\end{aligned}\ ] ] where @xmath73 is the lateral extension of the system both in the @xmath6 and @xmath88 directions . from this the lateral van der waals force @xmath170 can be calculated : @xmath171 with the scaling function @xmath172^{7/2 } } \label{eq28}\end{aligned}\ ] ] where @xmath173 .
the force and the potential as function of @xmath174 are shown in fig .
[ vdw ] . the comparison between figs . [ fig3 ] and [ vdw ] ( a ) reveals that the lateral van der waals force is practically constant over a wide range of shift values and varies steeply around the positions of maximum and minimum misalignment while the fluctuation - induced force varies more smoothly across all shift values . apart from that , the qualitative features are the same for both forces .
( [ eq27 ] ) and ( [ eq28 ] ) ] of the lateral force induced by direct van der waals interactions between the chemically patterned monolayers covering the two substrates at distance @xmath3 as function of the shift @xmath1 in units of the periodicity @xmath0 of the pattern for @xmath175 and @xmath176 .
( b ) corresponding lateral potential @xmath177.$ ] ] in order to estimate the lateral van der waals force we assume that the particles are closely packed within the monolayer @xcite , so that the areal number density is @xmath178^{-2}/\sqrt{3}$ ] where @xmath179 ( @xmath180 ) is the radius of the @xmath158 ( @xmath94 ) particles forming a triangular lattice .
according to table i in ref .
@xcite typical values are @xmath181 and @xmath182 .
for @xmath183 one has @xmath184 .
according to fig .
[ vdw ] ( a ) for @xmath185 this implies @xmath186 for the contribution @xmath187 to the actual lateral van der waals force @xmath188 proportional to @xmath173 .
thus for a suitably chosen constant @xmath189 , without compromising the goal of achieving @xmath190 ( as used in our calculations ) , the lateral van der waals force can be quite smaller than @xmath191 per area @xmath192 from fig .
[ fig3 ] one finds , for the same system parameters @xmath0 and @xmath3 considered above and at room temperature , for the fluctuation - induced lateral force @xmath193 .
thus the background lateral van der waals force tends to be stronger than the nematic fluctuation induced force . however , for a suitably chosen chemical contrast of the particles forming the chemical stripes , it appears to be possible to determine the fluctuation - induced lateral force by measuring the shear force once with and once without the nematic liquid between the patterned substrates .
the ratio of the two forces for @xmath175 , @xmath138 , @xmath139 , and @xmath194 as function of @xmath1 is shown in fig .
[ fig56 ] .
it appears that the fluctuation - induced lateral force becomes more prominent around @xmath195 . from figs .
[ fig3 ] and [ vdw ] ( a ) one notices that both @xmath196 and @xmath197 vanish linearly at @xmath198 [ as @xmath199 and @xmath200 at @xmath129 , respectively , for @xmath175 , @xmath201 , @xmath183 , and @xmath202 and since the slope of @xmath203 is much larger than the corresponding slope of @xmath135 , the ratio @xmath204 at @xmath205 is small . from fig .
[ fig56 ] one should not draw the conclusion that the lateral fluctuation induced force is at most half a percent of the corresponding lateral van der waals background force .
this ratio is inversely proportional to @xmath206 .
figure [ fig56 ] corresponds to a parameter choice for which @xmath189 is estimated by an individual @xmath207 and not by the actual contrast expressed by e , which vanishes for @xmath208 .
accordingly , for suitable choices of @xmath158 and @xmath94 , @xmath189 can be significantly smaller than the @xmath207 used in fig .
[ fig56 ] which then leads to a significantly larger ratio . and the lateral van der waals force @xmath188 as function of the shift @xmath1 in units of the periodicity @xmath0 for @xmath175 , @xmath138 , @xmath139 , @xmath209 , and at @xmath210 .
@xmath135 is more prominent around @xmath132 and @xmath133 . ]
the effect of a periodic anchoring at one substrate , with the second substrate being homogeneous , on the fluctuation - induced normal force was studied in ref .
it turned out that for the description of the normal force the single patterned substrate can be replaced by a uniform substrate with an effective anchoring strength , i.e. , the force is given by the force found between two uniform substrates characterized by their effective anchoring . depending on the model parameters ,
the normal force is either repulsive or attractive corresponding to an effective similar - dissimilar or an effective similar - similar boundary condition , respectively . in the present case of two patterned substrates ,
we have calculated the normal force numerically [ eq . ( [ tforce ] ) ] .
figure [ fig1 ] shows the dependence of the normal force @xmath211 on @xmath3 . in units of @xmath145 between two periodically patterned substrates as function of the film thickness @xmath3 in units of the periodicity @xmath0 . the anchoring on the stripes in terms of the extrapolation lengths
is taken to be @xmath139 and @xmath138 ( full line ) , @xmath212 ( dashed line ) . for @xmath213 as shown here the anchoring is weak but finite @xcite at both substrates and the force is attractive .
as expected the absolute value of the amplitude of the force decreases upon decreasing the extrapolation length .
the force depends very weakly on the shift @xmath1 . here
@xmath174 is set to @xmath214 . ] for @xmath215 as shown here , the force is attractive and decays monotonically as function of @xmath3 . in this regime ,
the anchoring is weak at both substrates so that the effect of the periodicity is not visible .
we note that in this case the fluctuation - induced normal force [ fig . [ fig1 ] ] is about @xmath216 times larger than the fluctuation - induced lateral force [ fig .
[ fig3d ] ] for @xmath217 .
figure [ fig8 ] , however , shows the behavior of @xmath218 as function of @xmath1 in the regime of strong but finite anchoring @xmath219 . in this case
the normal force is oscillatory and attractive , and its magnitude is comparable with the fluctuation - induced lateral force [ fig .
[ fig9 ] ] . in units of @xmath145 between two periodically patterned substrates at distance @xmath220 as function of the shift @xmath1 in units of the periodicity @xmath0 . the anchoring on the stripes in terms of the extrapolation lengths
is taken to be @xmath221 and @xmath222 ( triangles ) , @xmath223 ( circles ) . for @xmath224 as shown here the anchoring is strong but finite at both substrates and the force is attractive .
as expected the absolute value of the amplitude of the force decreases upon increasing the extrapolation length @xcite .
the force oscillates as function of the shift @xmath1 and for complete misalignment , i.e. , @xmath129 the attraction is weakest . ]
divided by fluctuation - induced lateral force @xmath148 as function of the contrast @xmath126 in units of the periodicity @xmath0 for @xmath221 , @xmath220 , corresponding to strong but finite anchoring , and @xmath132 .
@xmath211 and @xmath225 are of comparable size . ]
we have calculated the fluctuation - induced forces acting on two substrates chemically modulated with period @xmath0 and confining a nematic film of thickness @xmath3 ( fig .
[ fig0 ] ) .
the substrates are characterized by homeotropic anchoring with alternating extrapolation lengths @xmath141 and @xmath142 .
we have studied the shear force as function of the lateral shift @xmath1 between the patterns on the substrates and of their separation @xmath3 . for @xmath226 and @xmath227 ,
the lateral force sinusoidally oscillates as a function of @xmath174 and decays exponentially with @xmath124 [ eq .
( [ osci ] ) and fig .
[ fig3d ] ] . for stronger contrasts , the lateral force and its corresponding potential
have been evaluated numerically [ figs .
[ fig3 ] and [ fig00 ] ] .
it turns out that for a suitably chosen chemical contrast the fluctuation - induced lateral force is comparable [ fig .
[ fig56 ] ] with the background lateral van der waals force between the corresponding monolayers on the top and bottom substrates forming the chemical heterogeneity [ fig .
[ vdw ] ] . in order to complete the picture of the forces in the presence of the patterned substrates we have also calculated numerically the fluctuation - induced normal forces [ figs .
[ fig1 ] and [ fig8 ] ] and found them to be comparable with the fluctuation - induced lateral force in the case of strong anchoring [ fig .
[ fig9 ] ] .
patterning at small length scales gives rise to rich interfacial phenomena .
surface modulations open the possibility of controlling the morphology of wetting films and generate structural phase transitions which are central to the behavior of structural forces induced by distortions of the liquid crystal order parameter . in such cases , in addition to the fluctuation - induced forces , the substrates are subject to liquid - crystalline _ elastic _ forces @xcite .
such elastic forces scale with k [ eq . ( [ eqe ] ) ] and are for large @xmath3 larger than the fluctuation - induced forces which scale with @xmath228 .
since for small @xmath3 the elastic forces scale as @xmath229 @xcite but the fluctuation - induced force as @xmath230 @xcite , the latter can , however , even dominate for small @xmath3 .
mean field _ director contributions are completely eliminated for such model parameters and boundary conditions for which the director structure is uniform . in ref .
@xcite it has been demonstrated that this uniformity can indeed occur for suitable combinations of model parameters , even in cases of competing planar and homeotropic anchoring conditions , which have not been considered here .
capillary forces due to capillary condensation and the formation of bridge phases @xcite are other sources for the structural forces in the vicinity of the nematic - isotropic phase transition .
if the fluid is confined to very narrow slits , patterning may give rise to capillary bridges of different liquid crystalline order .
under such conditions , it would be interesting to study the stress under the shear strains by shifting the lateral substrate structures out of phase @xcite which might give rise to rather strong lateral forces .
lateral forces can technologically be used to align the parallel substrate structures .
for instance , for those ranges of the model parameters for which the liquid crystalline lateral forces are significant , the liquid crystal can be filled into the slit pore to align the substrate structure and then be removed .
p. g. de gennes and j. prost , _ the physics of liquid crystals _ ( clarendon , oxford , 1993 ) .
s. kondrat , a. poniewierski , and l. harnau , eur .
j. e * 10 * , 163 ( 2003 ) .
a. poniewierski and s. kondrat , j. mol
. liq . * 112 * , 61 ( 2004 ) .
a. ajdari , l. peliti , and j. prost , phys .
lett . * 66 * , 1481 ( 1991 ) .
p. zihrel , r. podgornik , and s. umer , chem .
. lett . * 295 * , 99 ( 1998 ) .
p. ziherl , f. karimi pour haddadan , r. podgornik , and s. umer , phys .
e * 61 * , 5361 ( 2000 ) .
f. karimi pour haddadan , d. w. allender , and s. umer , phys .
e * 64 * , 061701 ( 2001 ) .
p. ziherl and i. muevi , liq . cryst . * 28 * , 1057 ( 2001 ) . t. emig , a. hanke , r. golestanian , and m. kardar , phys
. rev . a * 67 * , 022114 ( 2003 ) .
r. bscher and t. emig , preprint , cond - mat/0412766 ( 2004 ) .
f. karimi , f. schlesener , and s. dietrich , phys .
e * 70 * , 041701 ( 2004 ) .
h. li and m. kardar , phys .
lett . * 67 * , 3275 ( 1991 ) .
r. golestanian and m. kardar , phys .
a * 58 * , 1713 ( 1998 ) .
t. emig , europhys .
62 * , 466 ( 2003 ) .
r. bscher and t. emig , phys .
a * 69 * , 062101 ( 2004 ) .
a. hanke , private communication .
p. m. chaikin and t. c. lubensky , _ principles of condensed matter physics _ ( cambridge university press , cambridge , 1995 ) .
t. getta and s. dietrich , phys .
e * 47 * , 1856 ( 1993 ) .
s. kondrat , a. poniewierski , and l. harnau , liq . cryst . * 32 * , 95 ( 2005 ) .
k. koevar , a. bortnik , i. muevic , and s. umer , phys .
. lett . * 86 * , 5914 , ( 2001 ) .
d. andrienko , p. patrcio , and o. i. vinogradova , j. chem . phys . * 121 * , 4414 ( 2004 ) . h. stark , j - i .
fukuda , and h. yokoyama , phys .
* 92 * , 205502 ( 2004 ) .
h. bock and m. schoen , j. phys : condens .
matter * 12 * , 1545 ( 2000 ) . | we consider a nematic liquid crystal confined by two parallel flat substrates whose anchoring conditions vary periodically in one lateral direction . within the gaussian approximation
, we study the effective forces between the patterned substrates induced by the thermal fluctuations of the nematic director . the shear force oscillates as function of the lateral shift between the patterns on the lower and the upper substrates .
we compare the strength of this fluctuation - induced lateral force with the lateral van der waals force arising from chemically structured adsorbed monolayers .
the fluctuation - induced force in normal direction is either repulsive or attractive , depending on the model parameters .
[ present address : ] institute for studies in theoretical physics and mathematics ( ipm ) , school of physics , po box 19395 - 5531 , tehran , iran |
the modeling and interpretation of the spectral energy distribution ( sed ) of stellar populations via stellar population synthesis ( sps ) is a fundamental tool to understand galaxy properties and their evolution .
yet we are far from a complete comprehension and a reliable modeling of some stellar evolutionary phases which strongly affect the energy output of stellar populations . among them , the so - called thermally pulsing - asymptotic giant branch ( tp - agb for short ) phase has been the focus of debate among modelists for several years ( e.g. ( * ? ? ?
* maraston 2005 ) ; ( * ? ? ?
* bruzual 2007 ) ; ( * ? ? ?
* marigo 2008 ) ) .
tp - agb stars are believed to be major contributors to the nir flux of stellar populations with age between 0.5 and 1.5 gyr
. however , quantitative predictions of the lifetimes , luminosities and emission continuum shapes and spectral features of these stars and of their influence on the integrated spectra of even `` simple '' stellar populations , are still very uncertain and strongly model - dependent .
different treatments of tp - agb stars ( i.e. prescriptions to model their lifetimes and luminosities , different stellar spectral libraries ) in popular sps codes can yield significant differences , in particular concerning _
i ) _ the flux ratios in different nir bands and in nir vs optical bands and _ ii ) _ the presence ( or absence ) of sharp spectral features due to molecules in the atmospheres of those stars .
the age and spectral ranges affected by tp - agb stars are of particular interest for two main applications . _
a ) _ at low redshift , the nir is often used as tracer of stellar mass in galaxies , both unresolved ( e.g. ( * ? ? ?
* bell 2003 ) ) and resolved ones ( e.g. ( * ? ? ? * zibetti , charlot & rix , 2009 ) ) , because of the ( supposed ) lower sensitivity to the young stellar components that often dominate the visible part of the spectrum and because of lower dust extinction . however , different treatments of tp - agb stars in sps models result in systematic differences in global stellar mass estimates of several tens of percent and in substantially different estimates of the stellar mass density contrast of structures like spiral arms within galaxies ( see zibetti , charlot & rix , 2009 ) .
_ b ) _ at @xmath1 , the bulk of stellar populations has mean ages in the range 0.51.5 gyr that is ( allegedly ) mostly affected by tp - agb stars . as demonstrated by , e.g. , (
* maraston et al . ( 2006 ) ) and ( * ? ? ?
* kannappan & gawiser ( 2007 ) ) , uncertainties on tp - agb models severely affect stellar mass and age determinations for galaxies at those redshifts ( see also ( * ? ? ?
* conroy , gunn & white , 2009 ) ) .
we directly test the impact of tp - agb stars on the integrated sed of galaxies by obtaining spectra in h and k bands , which we combine with available optical spectroscopy and photometry .
such an sed can be confronted with sps predictions and used to constrain model templates for galaxies at low and high redshifts . in order to maximize the impact of tp - agb stars , we target `` post - starburst '' galaxies , whose star formation an intense burst 0.5 - 1.5 gyr back in the past , followed by negligible activity
. they can be relatively easily selected from the large sdss spectral database by requiring no emission in h@xmath3 or in [ oii ] and equivalent width of h@xmath4 in absorption ( following ( * ? ? ?
* goto 2005 ) ) .
in addition we require light - weighted mean stellar age between 0.5 and 1.5 gyr as determined from optical spectral indexes ( little sensitive to tp - agb modeling ) according to the bayesian method of ( * ? ? ?
* gallazzi ( 2005 ) ) .
as the most prominent nir features of tp - agb stars are located at 1.41 and 1.77@xmath5 m and , thus , fall in the atmospheric gaps between j and h , and h and k bands for @xmath0 galaxies , we selected galaxies at @xmath60.2 , so that those two features move to the middle of the h and k atmospheric windows , respectively .
the resulting sample of 16 galaxies provides a good coverage of the age - metallicity domain that is most strongly affected by tp - agb stars .
nir spectra in h and k bands have been obtained with isaac at the eso - vlt during p86 and assembled with the sdss spectra to provide flux - calibrated seds ranging from @xmath73000 to @xmath721000 with a rest - frame resolution of 3 in the optical and 40 in the nir and typical s / n@xmath830 per resolution element ( see examples in fig .
[ fig : obs9pan ] ) .
spectra have been rescaled to match the broad - band `` total '' fluxes derived from sdss , ukidss and isaac imaging in matched apertures corresponding to the sdss petrosian aperture .
the h- and k - band spectra of all 16 galaxies in the sample appear smooth and featureless .
in particular we do not detect either of the two sharp features at 1.41@xmath5 m and 1.77@xmath5 m ( c@xmath9 ) rest - frame , in clear contrast with the maraston ( 2005 ) models predicting extremely strong and abrupt flux drops of 10 to 30% at those wavelengths .
our observations are the first ones to prove this _ directly _ thanks to the nir spectroscopy and the choice of a suitable redshift range .
nine examples out of our 16 observed galaxies are reported in fig .
[ fig : obs9pan ] to demonstrate the lack of such nir spectral features . spanning the full range of metallicities over the age range where tp - agb stars are expected to peak ,
there is no sign of sharp nir spectral features . over the last few years a general consensus has grown among sps modelists about the need for a boosted nir flux due to tp - agb stars with respect to earlier generation models ( e.g. bc03 ) , although with strong differences concerning its exact amount .
* kriek ( 2010 ) ) have recently challenged this , by showing that the composite ( medium - band nir ) sed of 62 post - starburst galaxies at @xmath10 is fully consistent with bc03 and inconsistent with maraston ( 2005 ) .
although very careful analysis is needed to compare our new observations with different models ( in prep . ) , preliminary analysis shows that the majority of our spectra is extremely well reproduced by bc03 simple stellar populations ( ssp ) with no need for further tweaking / fitting or inclusion of dust .
this suggests that predictions from early models might not require very substantial correction in terms of nir output .
0.8 gyr ( top ) , to @xmath71 gyr ( middle ) to @xmath71.4 gyr ( bottom ) . in each row
light - weighted mean metallicity increases from left to right , from sub - solar to super - solar values .
the spectra in each panel are cross - calibrated based on sdss , ukidss and isaac photometry normalized to the flux at 0.5@xmath5m(rest ) and shown as a function of rest - frame wavelength .
the vertical dashed lines mark 1.41 and 1.77@xmath5 m , where sharp spectral features are expected based on the maraston ( 2005 ) models ( cf .
their fig .
14 , 15 and 18 ) . however neither is observed .
each panel is labeled with the light - weighted stellar age and metallicity derived using the bayesian method of gallazzi et al .
( 2005).,scaledwidth=75.0% ] | we present new spectro - photometric nir observations of 16 post - starburst galaxies especially designed to test for the presence of strong carbon features of thermally pulsing agb ( tp - agb ) stars , as predicted by recent models of stellar population synthesis .
selection based on clear spectroscopic optical features indicating the strong predominance of stellar populations with ages between 0.5 and 1.5 gyr and redshift around 0.2 allows us to probe the spectral region that is most affected by the carbon features of tp - agb stars ( unaccessible from the ground for @xmath0 galaxies ) in the evolutionary phase when their impact on the ir luminosity is maximum . nevertheless , none of the observed galaxies display such features .
moreover the nir fluxes relative to optical are consistent with those predicted by the original bruzual & charlot ( 2003 ) models , where the impact of tp - agb stars is much lower than has been recently advocated . |
self - regulation plays an important role in biological and technical systems .
homeostatically regulated steady states are a precondition to life , examples being the concentration of blood glucose controlled by insulin @xcite and glucagon , the ph value of blood @xcite and the body temperature @xcite , which are all autoregulated in order to maintain stable conditions .
further examples are the concentration of ions , proteins and transmitters in the brain , their respective levels are all self regulated @xcite . furthermore , homeostasis is implemented and can be found in technical systems , for example in microrobotic swarms @xcite .
adaption typically introduces a slow time scale into the dynamical system @xcite , a process also denoted meta learning , a central notion in the context of neuromodulation @xcite and emotional control @xcite .
the resulting dynamical system then has both fast and slow variables and critical transitions in the form of tipping processes may occur @xcite .
classical homeostasis involves the regulation of a scalar quantity , like the body temperature .
more complex forms of homeostasis are however also important in the realm of life . for example , an animal may want to achieve a certain time averaged distribution of behaviors , like foraging , resting and engaging socially , over the course of several days .
this kind of adaptive behavior has been termed polyhomeostasis @xcite .
it occurs when a dynamical system tries to achieve , via the continuous adaption of slow variables , a given target distribution for the time - averaged activity of a subset of fast variables .
polyhomeostatically adapting systems are typically slow - fast dynamical systems and their dynamical behavior can tip spontaneously from one state into another . for a network of rate - encoding neurons tipping transitions from laminar flow to intermittent chaotic bursts of activities
have been observed @xcite .
tipping transitions can occur both in adaptive and in driven systems .
potential tipping scenarios are currently discussed intensively in the context of climate research @xcite , they may be related to a slow driving of external parameters @xcite , to noisy input inducing a stochastic escape from a local attractor @xcite , or through a dynamical effect when the rate of change of a control parameter reaches a certain threshold @xcite . here
we study the phenomenon of self - organized tipping for a polyhomeostatic adapting system driven by a steady - state stochastic input .
we examine a previously proposed model @xcite for regulating the firing rate distribution of individual neurons based on information - theoretical principles .
this type of model has been studied previously for the case of discrete time systems and unimodal target firing rate distributions @xcite . here
we examine the case of continuous time and bimodal target distribution functions , corresponding to sparse coding .
for bimodal firing rate distributions the neural activity tends to switch in between states close to minimal and maximal activity .
similar bimodal activity states are observed also in many other systems , e.g. dynamical gene regulation networks @xcite .
we find that bimodal target distributions may lead to self - organized bistability within a certain range of parameters .
we consider a single leaky integrator neuron with noisy input and a sigmoidal transfer function having two degrees of freedom . to achieve a special behavior here the temporal output distribution of the firing rate
we use polyhomeostasis to change the intrinsic parameters which are directly influencing the mapping of the membrane potential to the firing rate in order to obtain a specific output distribution .
we derive these parameter changing rules using stochastic adaption and show that two degrees of freedom already result in a good behavior approximation , for most of the parameters studied . for bimodal adaption target distributions
we observe self - organized and quasi periodic stochastic tipping in between two quasi - stationary attractors resulting from competing adaption gradients .
, see eq .
( [ eq_g ] ) , for thresholds @xmath0 ( red lines ) and @xmath1 ( green lines ) and various gains @xmath2 : 1 ( dotted ) , 3 ( dashed ) , 9 ( solid ) . ]
biological neurons integrate incoming signals and emit an axon potential , a spike , whenever the membrane potential has reached a certain threshold .
the membrane potential then returns , after a short refractory period , rapidly to its resting value .
this behavior can be captured using spiking integrate and fire neural models @xcite . in many circumstances
the firing rate , the number of spikes per unit time , is important and rate encoding neural models can be used @xcite .
here we consider a single rate - encoding leaky integrator driven by white noise @xmath3 , @xmath4 where @xmath5 is the membrane potential and @xmath6 the relaxation rate .
the firing rate @xmath7 $ ] is a nonlinear function of the membrane potential @xmath8 , which we have selected as @xmath9 where @xmath10 is the gain and @xmath11 is the threshold .
the polynomial transfer function ( [ eq_g ] ) maps the membrane potential @xmath12 $ ] to the normalized firing rate @xmath13 $ ] which approaches zero and unity for small and large membrane potentials respectively , compare fig .
[ fig_transferfunction ] .
the slope of @xmath14 is @xmath15 at the threshold @xmath11 .
usually the intrinsic parameters of the transfer function ( [ eq_g ] ) , @xmath2 and @xmath11 are taken as given by some a priori considerations . here
we will consider them to be slow variables , @xmath16 and @xmath17 , adapting slowly such that a target dynamical behavior is approached on the average for the firing rate @xmath18 .
the stochastic driving @xmath19 $ ] in ( [ eq_model ] ) is simulated through white noise plateaus : the values are generated according to a uniform probability distribution ( white noise ) , but they remain constant for short time intervals on the order of unity .
the membrane potential averages the input driving noise , due to the leak rate @xmath20 in ( [ eq_model ] ) , its distribution function @xmath21 having a mean @xmath22 and variance @xmath23 .
the firing - rate statistics is given by @xmath24 where the length @xmath25 of the sliding observation window is substantially larger than the relaxation rate @xmath26 .
the firing - rate distribution @xmath27 is an important quantity characterizing the information processing capability of biological and artificial neurons .
no information is encoded for a constant firing rate , the next value is always exactly the same as before , so no new information is transferred .
one may assume that a certain distribution @xmath28 of firing rates may constitute an optimal working regime .
possible functional dependencies for @xmath28 can be derived by information - theoretical considerations , e.g. maximizing information entropy , as discussed further below . considering a given target firing - rate distribution @xmath28 ,
the closeness of the actual firing - rate distribution @xmath29 is measured by the kullback - leibler divergence ( kl - divergence0 @xmath30 , their relative entropy , @xcite : @xmath31 the kullback - leibler divergence is positive definite and vanishes only when the two distribution coincide .
the kl - divergence is generically not symmetric but becomes symmetric in the limiting case of similar distributions @xmath32 and @xmath33 , becoming equivalent in this limit to the @xmath34 test @xcite .
our aim is now to rewrite ( [ eq_kl ] ) as an integral over the membrane potential @xmath35 , using @xmath36 where @xmath21 is the membrane potential distribution . using @xmath37 and eqs .
( [ eq_kl ] ) and ( [ eq_x_y ] ) , we obtain @xmath38 \equiv \int \ ! \mathrm
dx\ , \rho ( x)\ , \frac{\partial d}{\partial \theta } \label{eq_d_kl_theta}\ ] ] for the derivative of the kullback - leibler divergence with respect to the intrinsic parameters @xmath39 of the transfer function @xmath14 , see ( [ eq_g ] ) .
we consider now the case that the system does not dispose of prior information about the distribution of input stimuli and the thereby resulting distribution of membrane potential @xmath21 .
the best strategy to minimize the kullback - leibler is then to minimize the individual terms of the integral ( [ eq_d_kl_theta ] ) through the stochastic adaption rules @xcite @xmath40 for the intrinsic parameters of the transfer function @xmath14 , where the @xmath41 are appropriate small adaption rates .
, see ( [ eq_q_gaussian ] ) , with some selected parameters @xmath42 and @xmath43 .
the target firing - rated distributions is bimodal for @xmath44 . ] in order to evaluate ( [ eq_deltatheta ] ) , respectively eq .
( [ eq_d_kl_theta ] ) , we need to specify the target firing rate distribution @xmath28 . for this purpose
we use information - theoretical considerations . given a continuous probability distribution function @xmath33 its shannon entropy @xmath45 can be defined as @xmath46 among all the real - valued distributions with specified mean @xmath47 and standard deviation @xmath48 the gaussian distribution @xcite @xmath49 has maximal information entropy , with @xmath50 and @xmath51 , which is easily obtained using variational calculus , @xmath52~,\ ] ] where @xmath53 and @xmath54 are the respective lagrange parameters . in fig .
[ fig_targetdistributions ] examples for @xmath28 are illustrated for several values of @xmath55 and @xmath56 .
the support of the target firing rates is compact , @xmath13 $ ] , and both negative and positive @xmath55 and @xmath56 can be considered .
the normalization factor @xmath57 cancels out in ( [ eq_d_kl_theta ] ) , since only ratios are involved . for positive @xmath44 and @xmath58 one obtains bimodal target distributions .
this is an interesting case , since sparse coding , which is realized when only a minority of neurons of a given network is active , and a majority is inactive @xcite , is characterized by a skewed bimodal distribution .
with @xmath59 , @xmath60 , compare fig .
[ fig_targetdistributions ] . plotted
are the membrane potential @xmath35 ( solid blue line , upper panel ) , the threshold @xmath11 ( dashed red line , upper panel ) , the gain @xmath2 ( solid green line , middle panel ) and the firing rate @xmath61 ( solid black line , lower panel ) .
@xmath62 , @xmath63 , @xmath64 . ] from ( [ eq_q_gaussian ] ) and ( [ eq_g ] ) we find the relations @xmath65 and @xmath66 which we can use to evaluate the stochastic adaption rules ( [ eq_deltatheta ] ) as @xmath67 \right ] \label{eq_delta_a}\ ] ] and @xmath68~. \label{eq_delta_b}\ ] ] these two adaption rules will lead to an adaption of the time - averaged firing rate distribution @xmath29 towards the target distribution @xmath69 whenever the adaption time - scales @xmath70 are substantially larger than the time constants of the neural dynamics , which in turn are determined by the time scale of the incoming stimuli and by the leak - rate @xmath20 in ( [ eq_model ] ) .
the transfer function @xmath14 contains only two free parameters , the gain @xmath2 and the threshold @xmath11 .
perfect adaption @xmath71 , for all @xmath13 $ ] can hence not be expected .
the system tries to minimize the kullback - leibler divergence by adapting the available degrees of freedom , which are just two in our case .
( bars ) vs. achieved distribution @xmath32 ( points ) for different distributions . @xmath55 and @xmath56
are given in each diagram .
@xmath62 , @xmath72 , @xmath63 , @xmath73 , @xmath74 $ ] . ]
the equations ( [ eq_model ] ) , ( [ eq_delta_a ] ) and ( [ eq_delta_b ] ) form a set of first order differential equations with respect to time .
we solve them numerically using the euler method with one evaluation per time step .
the random white noise is generated through a pseudo - random number generator with a uniform distribution .
the values for the leak @xmath20 , the time step @xmath75 and the learning rates @xmath76 and @xmath77 are shown in the corresponding figures .
we performed a series of simulations with the aim to study two issues .
polyhomeostatic adaption had been studied previously for the case of discrete time systems @xcite , here we examine the case of continuous time .
the case of a bimodal target distribution is , in addition , highly interesting , as it confronts the system with a dilemma .
the transfer function @xmath14 , compare fig .
[ fig_transferfunction ] , is strictly monotonic . the distribution of the membrane potential @xmath21 is hence mono - modal .
there is no easy way for the adapting neuron to achieve , as a steady state time - average , a bimodal output firing rate distribution @xmath29 .
the question then is whether the system will find a way out of this dilemma through spontaneous behavioral changes . for most simulations we used ,
if not stated otherwise , @xmath78 for the leak rate and @xmath79 for the integration time step .
a typical time series is given in fig .
[ fig_timeseries ] .
note that the adaption of the intrinsic parameters @xmath2 and @xmath11 takes place on a slower time scale than the one of the primary dynamic variables , @xmath35 and @xmath61 , as typical for a slow - fast dynamical system . applying moderate to small learning rates
@xmath80 the neuron s firing rate @xmath61 approximates various types target distributions @xmath33 quite well . in fig .
[ fig_targetvsachieveddistributions ] the achieved and the respective target firing rated distributions are compared .
the respective relative entropies are well minimized and presented in table [ tab : relativentropiesvarioustargetdistributions ] . strictly speaking the stochastic adaption rules ( [ eq_delta_a ] ) and ( [ eq_delta_b ] ) are equivalent to approximating the firing - rate statistics ( [ eq_def_p ] ) , which is a time - averaged quantity , towards the target distribution function @xmath28 only in the limit of very small adaption rates , @xmath76 and @xmath77 . small but finite values for the adaption rates , as used in our simulations , correspond to to a trailing averaging procedure over a limited time interval , and the value of kullback - leibler divergence achieved hence depend weakly on the actual values used for the learning rates . for very high learning rates , @xmath81 ,
the threshold @xmath11 follows the membrane potential @xmath35 nearly instantaneously , both variables become highly correlated . therefore the firing rate distribution @xmath32 can not approximate the target distribution @xmath33 any more , in fact the resulting kullback - leibler divergence is then very high .
the tipping in dynamic behavior as a function of adaption rate amplitude is typical for a rate - induced tipping transition @xcite .
llll @xmath42 & @xmath43 & shape & @xmath82 0 & 0 & uniform & 0.043 -10 & 0 & left dominant & 0.034 + 10 & 0 & right dominant & 0.028 -10 & + 10 & left / right dominant & 0.018 + 20 & -20 & hill & 0.076 -20 & + 20 & left / right , symmetric & 0.175 -20 & + 19 & left / right , left skewed & 0.244 -20 & + 18.5 & left / right , left skewed & 0.283 and the threshold @xmath83 of the transfer function for various target distributions ( @xmath55 and @xmath56 given in the legend ) . the respective target and
achieved firing rate distributions are given in fig .
[ fig_targetvsachieveddistributions ] . ] due to the sigmoidal shape of the transfer function , several target distributions lead to specific fingerprints in the gain - threshold phase diagram which we present in fig . [ fig_gainthresholdphasediagramvarioustargetdistributions ] .
the threshold , for example , for a left ( right ) dominant target distribution is high ( low ) and is therefore sensitive to the mean @xmath50 of @xmath28 .
small gains @xmath2 result in quite flat transfer functions @xmath14 , compare fig .
[ fig_transferfunction ] , mapping the membrane potentials to similar firing rates @xmath61 .
high gains @xmath2 discriminate , relative to the threshold @xmath11 , on the other side between high and low membrane potentials .
the gain is therefore smaller for hill shaped and flat target distributions , as compared to the left and right dominant target distributions ( e.g. @xmath84 , @xmath85 ) for which intermediate values are suppressed .
left ( right ) dominant target distributions ( compare table [ tab : relativentropiesvarioustargetdistributions ] ) correspond directly to high ( low ) transfer function thresholds .
uniform , hill and other not unilateral dominant target distributions lead to intermediate transfer function thresholds with a wide variety of the transfer function gains . for symmetrical target distributions from hill shaped to diametrical shaped
there is a transition from low to high gains . given in the legend ) .
@xmath62 , @xmath64 , @xmath86 , @xmath87 . ] while the left or right dominant target distributions are easily approximated due to the sigmoidal shape of the transfer function , the bimodal left and right dominant target distributions puts the system in dilemma : since intermediate values are to be suppressed the transfer function gain @xmath2 can not be too small . because of this there exists at least two quasi - stationary fixed points , one for the left , one for the right part of the distribution . for zero or small learnings rates @xmath88
the system is trapped in a single local fixed point . only the left or right part of the target distribution is then approximated , the kullback - leibler divergence is not well minimized .
l|lllllll @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 & @xmath95 & @xmath96 @xmath82 & 0.306 & 0.295 & 0.293 & 0.289 & 0.283 & 0.154 & 0.109 increasing the learning rates @xmath97 allows the system to escape stochastically from the respective local fixed points : the transfer function threshold @xmath11 conquers the local gradient and moves to the other fixed point and back ( compare fig . [ fig_stochastic_escape ] ) . in the long - term observation
the system therefore approximates the left and the right part of the target distribution and hence minimizes the relative entropy , compare table [ tab : relativentropies ] .
these tipping transitions between the two quasi - stationary fixed points are illustrated in fig .
[ fig_timeseriesstochasticescape ] , which shows a typical time series for a skewed target distribution .
note that there are two fixed points for the gain and threshold and a direct correspondence to the periods of high and low firing rates @xmath18 .
very low learning rates @xmath76 , @xmath77 , lead to deep and big basins of attraction for the respective fixed points , while on the other hand high learning rates result in the closely following of the threshold to the membrane potential which prohibits reaching the target distribution .
this mechanism is reminiscent to the case of langevin dynamics in a double - well potential @xcite , where a stochastically driven particle may switch forth and back between two local minima @xcite .
the switching time is controlled for the double - well problem by the kramer s escape rate , which depends exponentially on the potential barrier height .
it is difficult to formulate a quantitative mapping to the double - well problem , the local attractors visible in figs .
[ fig_stochastic_escape ] and [ fig_timeseriesstochasticescape ] , and the effective barriers in between them , are self - organized structures .
note that the strength @xmath98 of the noise term ( [ eq_model ] ) is constant and influences the transition rate only weakly , due to the continuous adaption of the transfer function , via ( [ eq_delta_a ] ) and ( [ eq_delta_b ] ) , to the average strength of the stochastic driving .
, transfer function threshold @xmath11 ( dashed ) , transfer function gain @xmath2 and firing rate @xmath61 .
@xmath62 , @xmath99 , @xmath73 , @xmath86 , @xmath87 . ]
we showed that polyhomeostatic adaption of continuous - time leaky integrator leads to desired firing rate distributions .
we also run further simulations using white noise and gaussian noise input and replace the transfer function by other qualitatively different ( but still sigmoidal ) functions , see appendix .
it turns out that the polyhomeostatic adaption as well as the self - organized stochastic escape are quite robust principles .
however , the quality of the approximation ( as seen by visual overlapping ) and the value of the kullback - leibler divergence depend on the learnings rates , but also on the input distribution and the input s strength .
the stochastic tipping as a function of adaption rates has a close relation to the phenomenon of stochastic escape .
the strength of the driving input noise is constant , but its influence is averaged out for very low adaption rates .
stochastic escape from one local attractor to another is not possible .
the stochasticity of the input becomes relevant for intermediate values of adaption rates and stochastic transitions between the two quasistationary attractors are most frequent .
finally , for very large adaption rates , the system tips into another dynamical state , tracking the stochastic input signal nearly instantaneously .
this sequence of behaviors is self organized and reached from any initial state .
10 l. plum , b.f .
belgardt , j.c .
brning , the journal of clinical investigation * 116 * , 17611766 ( 2006 ) .
schaefer , blood ph and pco2 homeostatis in chronic respiratory acidosis related to the use of amine and other buffers , annals of the new york academy of sciences * 92 * , 401413 ( 1961 ) .
m. tresguerres , s.k .
parks , e. salazar , l.r .
levin , g.g .
goss , j. buck , bicarbonate - sensing soluble adenylyl cyclase is an essential sensor for acid / base homeostasis proceedings of the national academy of sciences * 107 * , 442447 ( 2010 ) .
n. charkoudian , skin blood flow in adult human thermoregulation : how it works , when it does not , and why , mayo clinic proceedings * 78 * , 603612 ( 2003 ) . e. marder , and j.m
. goaillard , variability , compensation and homeostasis in neuron and network function , nature reviews neuroscience * 7 * , 563 ( 2006 ) .
s. kernbach , o. kernbach , collective energy homeostasis in a large - scale microrobotic swarm , robotics and autonomous systems * 59 * , 10901101 ( 2011 ) . c. gros , _ complex and adaptive dynamical systems , a primer _ , springer ( 2008 ) ; second edition 2010 .
k. doya , metalearning and neuromodulation , neural networks * 15 * , 495506 ( 2002 ) . c. gros , cognition and emotion : perspectives of a closing gap cognitive computation * 2 * , 78 ( 2010 ) . c. kuehn , a mathematical framework for critical transitions : bifurcations , fast - slow systems and stochastic dynamics , physica d : nonlinear phenomena * 240 * , 10201035 ( 2011 ) . d. markovic , c. gros , self - organized chaos through polyhomeostatic optimization , physical review letters * 105 * , 068702 ( 2010 ) . d. markovic , c. gros , intrinsic adaption in autonomous recurrent neural networks , neural computation * 24 * , 523 ( 2012 ) .
p. ashwin , s. wieczorek , r. vitolo , p. cox , tipping points in open systems : bifurcation , noise - induced and rate - dependent examples in the climate system phil .
a * 370 * 1166 - 1184 ( 2012 ) .
lenton , h. held , e. kriegler , j.w .
hall , w. lucht , s. rahmstorf , h.j .
schellnhuber , tipping elements in the earth climate system , proceedings of the national academy of sciences * 105 * , 1786 ( 2008 ) . s.m .
baer , t. erneux , j. rinzel , the slow passage through a hopf bifurcation : delay , memory effects , and resonance , siam journal on applied mathematics * 49 * , 5571 ( 1989 ) .
l. gammaitoni , p. hnggi , p. jung , f. marchesoni , stochastic resonance , reviews of modern physics * 70*,223 ( 1998 ) .
mcdonnell , d. abbott , what is stochastic resonance ? definitions , misconceptions , debates , and its relevance to biology , plos computational biology * 5 * , e1000348 ( 2009 ) .
j. triesch , a gradient rule for the plasticity of a neuron s intrinsic excitability , in _ proceedings of icann 2005 _ , w. duch et al .
( eds . ) , lncs * 3696 * , 65 ( 2005 ) .
m. stemmler , c. koch , how voltage - dependent conductances can adapt to maximize the information encoded by neuronal firing rate , nature neuroscience * 2 * , 521 ( 1999 ) .
davidson , d.h .
erwin , gene regulatory networks and the evolution of animal body plans .
science * 311 * , 796 ( 2006 ) .
burkitt , a review of the integrate - and - fire neuron model : i. homogeneous synaptic input , biological cybernetics * 95 * , 119 ( 2006 ) .
a. borst , f.e .
theunissen , information theory and neural coding , nature neuroscience * 2 * , 947958 ( 1999 ) .
olshausen , and d.j . field , sparse coding of sensory inputs , current opinion in neurobiology * 14 * , 481487 ( 2004 ) .
p. hanggi , escape from a metastable state , journal of statistical physics * 42 * , 105 ( 1986 ) .
the polyhomeostatic adaption of the system is not changing qualitatively by replacing the transfer function @xmath100 .
instead it turns out that the system is robust against changing the transfer function as long as it remains sigmoidal .
we also applied a transfer function @xmath101 with a polynomial decay to @xmath102 , which limits the membrane potential @xmath103 to be non - negative .
it turns out that the shape of the target distribution @xmath33 is also well approximated using this transfer function .
also stochastic escape from one fixed point to another and back can be observed as for some target distributions two fixed points are necessary .
, see eq .
( [ eq_g_g0 ] ) , for thresholds @xmath0 ( red lines ) and @xmath1 ( green lines ) and various gains @xmath2 : 1/3 ( dotted ) , 3 ( dashed ) , 9 ( solid ) .
no inflection point is present for exponents @xmath104 . ]
the transfer function has an inflection point for exponents @xmath105 ; it is absent for @xmath104 , compare fig .
[ fig_transferfunction_g0 ] .
the transfer function @xmath100 behaves as @xmath106 the slope is @xmath15 which approaches zero and unity for small and large membrane potentials respectively . from eq .
( [ eq_g_g0 ] ) we find the relations @xmath107 @xmath108~,\ ] ] which we can use to evaluate the stochastic adaption rules ( [ eq_deltatheta ] ) as @xmath109 \right ] \label{eq_delta_a_g0}\ ] ] and @xmath110\ , \big [ 1 - 2 y + \left ( \lambda _ 1 + 2 \lambda _ 2 y \right ) \left ( 1 - y \right ) y \big ] \right]~. \label{eq_delta_b_g0}\ ] ] llll @xmath42 & @xmath43 & shape & @xmath82 0 & 0 & uniform & 0.060131 -10 & 0 & left dominant & 0.069351 + 10 & 0 & right dominant & 0.114578 -10 & + 10 & left / right dominant & 0.051811 + 20 & -20 & hill & 0.148098 -20 & + 20 & left / right , symmetric & 0.189217 -20 & + 19 & left / right , left skewed & 0.063934 -20 & + 18.5 & left / right , left skewed & 0.261215 applying this transfer function @xmath100 it turns out that the target distribution is well approximated also in this case , even though the membrane potential is restricted to non - negative numbers
[ tab : relativentropiesvarioustargetdistributions_g0 ] lists the well minimized kullback - leibler divergences for several target distributions .
we conclude that the stochastic adaption rules are therefore generic and qualitatively independent on the concrete realization of the transfer function
. however , quantitatively the resulting relative entropies depend on the choice of the transfer function which also influences the optimal adaption rates @xmath76 and @xmath77 . and @xmath111 given in the legend ) .
@xmath112 , @xmath113 , @xmath86 , @xmath114 .
right : time series : membrane potential @xmath35 , transfer function threshold @xmath11 ( dashed ) , transfer function gain @xmath2 and firing rate @xmath61 .
@xmath62 , @xmath63 , @xmath113 , @xmath86 , @xmath114.,title="fig : " ] and @xmath111 given in the legend ) .
@xmath112 , @xmath113 , @xmath86 , @xmath114 .
right : time series : membrane potential @xmath35 , transfer function threshold @xmath11 ( dashed ) , transfer function gain @xmath2 and firing rate @xmath61 .
@xmath62 , @xmath63 , @xmath113 , @xmath86 , @xmath114.,title="fig : " ] for the non - symmetric convex target distribution ( @xmath86 , @xmath114 ) there are two fixed points .
since the target distribution can not be well approximated by only one fixed point the system escapes stochastically from one to the other and back with a certain period , compare fig .
[ fig_stochastic_escape_appendix ] . for small learning rates
@xmath115 the system in trapped in only one fixed point .
the relative entropy therefore is not well minimized . for intermediate learning rates
@xmath116 the perturbation is high enough to stochastically escape from that fixed point and approach another one .
[ fig_stochastic_escape_appendix ] shows a typical time series for this tipping .
this has also an effect on the relative entropy which therefore is even smaller than without stochastic escape ( see tab . [
tab : relativentropies_g0 ] ) . for high learning rates
@xmath117 the system s behavior changes : the transfer function is close to a heaviside step function and the threshold follows the membrane potential quickly . in that state
the achieved distribution is not close to the target distribution , therefore the relative entropy is not minimized anymore ( see tab .
[ tab : relativentropies_g0 ] ) . | polyhomeostatic adaption occurs when evolving systems try to achieve a target distribution function for certain dynamical parameters , a generalization of the notion of homeostasis . here
we consider a single rate encoding leaky integrator neuron model driven by white noise , adapting slowly its internal parameters , the threshold and the gain , in order to achieve a given target distribution for its time - average firing rate .
for the case of sparse encoding , when the target firing - rated distribution is bimodal , we observe the occurrence of spontaneous quasi - periodic adaptive oscillations resulting from fast transition between two quasi - stationary attractors .
we interpret this behavior as self - organized stochastic tipping , with noise driving the escape from the quasi - stationary attractors . |
there are new possibilities for strong cp problem solution , which allow the existence of axions with a large mass ( 1 mev ) , while their interaction with ordinary particles remain at the level of the invisible axions .
the models rely on the hypothesis of mirror particles @xcite and susy at the tev scale @xcite .
the existence of these heavy axions is not forbidden by the laboratory experiments or astrophysical data .
this article describes the experimental search for 5.5 mev solar axions , which can be produced by @xmath4 reaction .
axion flux should be proportional to the @xmath5-neutrino flux , which has been estimated with high accuracy @xcite .
the searches have been performed with the use of bismuth orthogermanate @xmath6 ( bgo ) scintillation and bolometric detectors .
the solar axions are supposed to interact with atoms via the reaction of axioelectric effect @xmath7 .
this kind of interaction is governed by @xmath8-constant and the cross section depends on the charge as @xmath9 . from this point of view
the bgo detector is a very suitable target , because of the high @xmath10 of bismuth nucleus .
the high energy solar axions and axions from a nuclear reactor have been looked by the borexino @xcite , the cast @xcite and the texono @xcite collaborations .
this paper is based on the results of obtained with bgo scintillation @xcite and bgo bolometer detectors @xcite .
the 5.5 mev axion production probability ratio @xmath11 depends only on the isovector axion - nucleon coupling constant @xmath12 ( see @xcite and refs therein ) : @xmath13 ^ 2\left(\frac{p_a}{p_\gamma}\right)^3 = 0.54(g_{an}^{3})^2 \left(\frac{p_a}{p_\gamma}\right)^3.\ ] ] where @xmath14 and @xmath15 are , respectively , the photon and axion momenta and @xmath16 is isovector nuclear magnetic momenta . at the earth s
surface the axion flux is : @xmath17 where @xmath18 is the @xmath5-neutrino flux .
the cross section for the a.e .
effect was calculated in @xcite .
bgo crystal with mass 2.46 kg contains 1.65 kg of bi .
the crystal was shaped as a cylinder , 76 mm in diameter and 76 mm in height .
the detector signal was measured by an r1307 photoelectron multiplier , which had an optical contact with a crystal end surface .
the external @xmath19 activity was suppressed using a passive shield that consisted of layers of lead and bismuth ( @xmath20 ) with the total thickness @xmath21 110 @xmath22 .
the setup was located on the earth s surface . in order to suppress the cosmic - ray background we used an active veto , which consisted of five @xmath23 cm plastic scintillators .
the measurements were performed over 29.8 days .
the energy spectrum of the bgo detector in the range of ( 011 ) mev is shown in fig .
1 . in inset
the calibration spectrum measured with pu - be neutron source is shown . in the spectrum
, one can identify two pronounced peaks at 1.460 mev and 2.614 mev ; these are due to the natural radioactivity of the @xmath24 and of @xmath25 from the @xmath26 family .
the positions and intensities of these peaks were used for monitoring of time stability of the detector .
four cubic ( @xmath27 @xmath28 ) bgo bolometers , containing 1.65 kg of bi , were arranged in a four - plex module , one single plane set - up .
the scintillation light was monitored with an auxiliary bolometer made of high - purity germanium @xcite .
the detector was installed in the @xmath29 dilution refrigerator in the underground laboratory of l.n.g.s . and operated at a temperature of few mk .
the four crystals were housed in a highly pure copper structure , the same described in @xcite .
neutron transmutation doped ( ntd ) germanium thermistor was coupled to each bolometer , ntd acts as a thermometer : recording the temperature rises produced by particle interaction and producing voltage pulses proportional to the energy deposition .
details on electronics and on the cryogenic set - up can be found in @xcite .
the detector was operated for a total live time of 151.7 days .
the different light yield of interacting particles was used to discriminate @xmath30events from @xmath31 ones .
this allowed to strongly increase the sensitivity , due to rejection of all @xmath30events in the region of interest .
the total statistics in the range of ( 0.3 - 4 ) mev for @xmath31 events are in shown in fig.[muratova_valentina_fig2.pdf ] .
the most intense gamma lines are produced by @xmath32 decays . in first approximation ,
the energy resolution of large mass bolometric detector is independent of the energy - the fwhm is @xmath33 kev at 2614 kev ( @xmath34 ) and @xmath35 kev at 570 kev ( @xmath32 ) .
the expected number of axioelectric absorption events are : @xmath36 where @xmath37 is the axioelectric effect cross section ; @xmath38 is the axion flux ( [ fluxa ] ) ; @xmath39 is the number of bi atoms ; @xmath40 is the measurement time ; and @xmath41 is the detection efficiency for 5.5 mev electrons .
the axion flux @xmath38 is proportional to the constant @xmath42 , and the cross section @xmath37 is proportional to the constant @xmath43 . as a result ,
the @xmath44 value depends on the product @xmath45 .
the experimentally found condition @xmath46 imposes some constraints on the range of possible @xmath47 and @xmath48 values . the range of excluded @xmath49 values is shown in fig . 3 , at @xmath50
the limits @xmath51 @xmath52 were obtained for bgo scintillating and bgo bolometer detectors , correspondingly .
these constraints are completely model - independent and valid for any pseudoscalar particle with coupling @xmath53 less than @xmath54 . for hadronic axion model with concrete relation between @xmath1 and
@xmath48 one can obtain a constraint on the @xmath8 constant , depending on the axion mass ( fig .
3 ) . for @xmath48
= 1 mev , this limit corresponds to @xmath55 at @xmath56 .
figure 3 also shows the constraints which were obtained in the borexino experiment for 478 kev @xmath57 solar axions @xcite and in the texono reactor experiment for 2.2 mev axions produced in the @xmath58 reaction @xcite .
borexino coll .
reported new more stringent limits on @xmath8 coupling for 5.5 mev solar axions @xcite .
unlike our work , these limits on @xmath8 coupling were obtained in assumption that the axion interacts with electron through the compton conversion process . in the model of the mirror axion
@xcite an allowed parameter window is found within the p - q scale @xmath59 gev and the axion mass @xmath60 1 mev . the limit ( [ limgaegan ] )
may be represented as a limit on the value @xmath61 by taking the following relations into account : @xmath62 and @xmath63 . for axion masses
about 1 mev , the limit is @xmath64 , which is close to the lower bound of mirror axion window .
our results set constraints on the parameter space of the cp - odd higgs @xmath65 , which arise in the next - to - minimal supersymmetric standard model due to the spontaneous breaking of approximate symmetries such as pq - symmetry .
the corresponding exclusion region can be obtained from fig.[muratova_valentina_fig3.pdf ] using the conversion @xmath66 where @xmath67 is the coupling of the cp - odd higgs to fermions and @xmath68 is the gauge coupling .
the limit ( [ limgaegan ] ) translates into @xmath69 mev for @xmath70 mev , which is compatible with the limits obtained in reactor experiments exploring compton conversion .
z. berezhiani et al . , phys .
b500 , 286 ( 2001 ) .
l.j . hall and t.
watari , phys .
d70 , 115001 ( 2004 ) .
g. bellini et al . , ( borexino coll . ) nature , 512 , 7515 , 383 ( 2014 ) g. bellini et al . , ( borexino coll . )
epj , c54 , 61 ( 2008 ) .
g. bellini et al . , ( borexino coll . ) , phys . rev .
d 85 , 092003 ( 2012 ) .
s. andriamonje et al . , ( cast coll . )
jcap 1003 , 032 , ( 2010 ) .
arxiv:0904.2103 h.m .
chang et al . , ( texono coll . ) phys . rev .
d75 , 052004 ( 2007 ) .
a.v . derbin et al . , europ . phys .
j. c73 , 2490 ( 2013 ) .
arxiv:1306.4574 a.v .
derbin et al . , europ .
j. c74 , 3035 ( 2014 ) .
arxiv:1405.3782 a.r .
zhitnitskii and yu.i .
skovpen , yad .
fiz . , 29b , 995 ( 1979 ) . l. cardani ,
s.di domizio , l. gironi , jinst 7 , p10022 ( 2012 ) .
beeman et al . , jinst 8 , p05021 ( 2013 ) .
f. alessandria et al . , astropart .
35 , 839849 ( 2012 ) .
s. pirro et al . , nucl .
methods a 444 , 331 ( 2000 ) . c. arnaboldi et al .
methods a 559 , 826 ( 2006 ) . | a search for axioelectric absorption of 5.5 mev solar axions produced in the @xmath0 reaction has been performed with a bgo detectors .
a model - independent limit on the product of axion - nucleon @xmath1 and axion - electron @xmath2 coupling constants has been obtained : @xmath3 for 90% c.l .. |
first data from the large hadron collider ( lhc ) on the nuclear - size dependence of jet - medium interactions in pb+pb collisions at @xmath16 atev @xcite showed that the jet - medium coupling at the lhc is weaker than expected @xcite from fixed coupling extrapolations from @xmath17 atev data at the relativistic heavy ion collider ( rhic ) @xcite . from the factor of @xmath18 increase of global multiplicity per unit rapidity , @xmath19 in central pb+pb reactions at the lhc
relative to rhic , substantially more suppression of high-@xmath20 pions was predicted than observed . in this paper
we estimate the reduction of the jet - medium coupling implied by the new data and test the consistency for a wide variety of jet - energy loss models .
an open question from studies of the nuclear modification factor of jets at rhic is whether jet - medium interactions in dense deconfined quark - gluon plasma ( qgp)-matter can be better described in terms of weakly - coupled perturbative qcd ( pqcd ) tomography or novel strongly - coupled gravity - dual anti - de - sitter / conformal field theory ( ads / cft ) string - model holography @xcite .
the first data from lhc provide stringent new tests of jet - medium interaction models for a higher qgp - density range and for an order of magnitude higher transverse momentum range .
the doubling of the qgp density at lhc relative to rhic extends the temperature range by approximately 30% beyond the range explored at rhic . in order to extract either tomographic or holographic information from the jet - quenching observable systematics ,
it is important to specify the initial jet flux at the jet - production points and the initial geometry of the qgp medium .
two competing models , the glauber @xcite model and the color glass condensate ( cgc ) @xcite , have been used so far describe the initial qgp medium geometry and to predict the impact parameter @xmath6 and beam energy @xmath21 dependence of those distributions .
the glauber model is based on an eikonal wood - saxon nuclear geometry and assumes incoherent superposition of proton - proton collisions while cgc models are based on non - linear , small - x saturation effects @xcite .
here , we calculate jet - quenching observables using both initial geometries as a measure of current systematic theoretical errors associated with uncertainties of the initial qgp geometries as a function of impact parameter . in this paper , we use a simplified analytic `` polytrope '' jet - energy - loss model @xcite that can interpolate between a wide class of weakly and strongly - coupled models of jet - medium interaction in high - energy nuclear collisions : @xmath22\;. \label{genericeloss}\end{aligned}\ ] ] the energy loss per unit length is characterized by the three exponents ( @xmath23 ) that determine the jet - energy dependence @xmath24 , the path - length dependence @xmath25 , and the local temperature - power dependence @xmath26 .
this polytrope model reproduces the results of detailed djordjevic - gyulassy - levai - vitev ( dglv ) opacity series calculations remarkably well @xcite but has the advantage of allowing quick estimates of the variations of predictions to even large deformations of dynamical assumptions . the energy loss per unit length , @xmath27 depends on the local proper time @xmath28 in a frame where the jet rapidity @xmath29 vanishes .
the jet path is thus perpendicular to the beam axis in that frame and can be assumed to be a straight eikonal line @xmath30 from the production point @xmath31 in direction of the jet azimuthal angle @xmath32 relative to the reaction plane @xmath33 .
bjorken longitudinal expansion @xcite is taken into account by @xmath34 until a freeze - out isotherm is reached specified by @xmath35=t_f=100 $ ] mev .
as in refs .
@xcite , we assume that the energy loss depends monotonically on the co - moving local entropy density .
we consider an initial qgp - formation time of @xmath36 fm / c @xcite .
the dimensionless effective jet - medium coupling @xmath10 is interpreted as proportional to @xmath37 in the case of radiative energy - loss tomography while it is interpreted to be proportional to @xmath38 in terms of the thooft coupling in gravity dual string holography .
we note that the considered monotonic power - law dependence of @xmath39 on the temperature ( or entropy ) field in eq .
( [ genericeloss ] ) is a dynamical assumption consistent with pqcd tomography as well as ads / cft holography models in the literature .
however , there also exist non - monotonic models , e.g. see liao and shuryak @xcite involving e.g. magnetic monopole enhanced @xmath39 near the critical qcd transition point of @xmath40 mev .
the scope of the present paper is limited to monotonic models as in eq .
( [ genericeloss ] ) to avoid extra model assumptions and parameters .
we also note that the @xmath41 polytrope model describes approximately both the pqcd and the ads / cft falling string cases @xcite .
an @xmath42-energy dependence is numerically similar to the logarithmic @xmath43 dependence predicted by fixed coupling pqcd - energy loss in the range @xmath44 relevant at lhc energies .
this power law is also predicted to be the lower bound of the power @xmath45 in the falling - string scenario in an ads / cft conformal holography .
variants of the polytrope model with path dependencies varying from @xmath46 to @xmath47 have been considered in refs .
recent work @xcite on the falling string energy - loss in conformal ads geometry has identified important corrections to the original works @xcite that may reduce the effective @xmath47 path - length power - law dependence assumed in @xcite to @xmath48 similar to the predicted radiative energy loss in pqcd . while such a falling string scenario essentially leads to the same polytrope exponents as radiative pqcd - energy loss , the jet - medium coupling strength differs significantly in the two cases . as we also show below , eq .
( [ genericeloss ] ) reduces to the survival probability model referred to as `` ads / cft''-like in refs .
@xcite in the limit @xmath49 . for @xmath50 ,
on the other hand , the polytrope model reduces to the heavy - quark string - drag energy - loss of conformal ads holography @xcite . as we emphasize below , polytropes with @xmath14 lead in fact to a generalized class of `` geometric optics '' or `` survival probability '' models of nuclear jet modifications .
we will demonstrate that the recent lhc data on @xmath51 data rules out this class of dynamical models . in this work ,
we further generalize ref .
@xcite by using the full pqcd jet - production @xmath20-spectra for the initial quark and gluon jets avoiding the local spectral index approximation .
in addition , we convolute the quenched jet spectra with kkp pion fragmentation functions @xcite that have been tested well against rhic @xmath52 spectra in ref .
@xcite .
the magnitude of the cube of the initial temperature profile is assumed to scale with the observed rapidity density @xmath53 .
the initial transverse coordinate distribution , @xmath54 , of the qgp is modelled according to a glauber and a higher eccentricity cgc - like elliptically deformed geometry [ see eqs.([fkln1 ] ) to ( [ fkln2 ] ) below ] . and the @xmath55 at rhic energies after fragmentation for an energy loss with @xmath46 ( left panel ) as well as an energy loss with @xmath47 ( right panel ) .
glauber initial conditions are displayed by the red dashed lines and dcgc1.2 initial conditions by the blue solid lines . those results are obtained for @xmath56 .
the grey dotted line represents an energy - loss for dcgc1.2 initial conditions considering just binary collisions ( `` jia '' dcgc1.2 ) as well as @xmath14 , similar to the one in refs .
the dark red dashed - dotted line displays an energy loss with glauber initial conditions and @xmath57 .
all calculations assume an initialization time of @xmath36 fm .
the black dashed - dotted line is the result for cgc initial conditions from ref.@xcite with @xmath58 fm .
data are taken from ref .
at the partonic level , the nuclear modification factor @xmath0 is the ratio of the jet spectrum for jets penetrating a qgp produced in a+a collisions to the initial jet spectrum predicted by pqcd without final state interactions : @xmath59}{dyd\phi dp_0 ^ 2}{\big/ } \frac{dn^{jet}_{vac}(p_f)}{dyd\phi dp_0 ^ 2}\,.\end{aligned}\ ] ] denoting the invariant jet distribution by @xmath60 , @xmath61 the nuclear modification factor for a quark ( q ) or gluon ( g ) jet with a final momentum @xmath62 , produced at a transverse coordinate @xmath31 and propagating in direction @xmath32 is , from eq .
( [ genericeloss ] ) , @xmath63}{g_0(p_f ) } \frac{dp^2_0}{dp^2_f}\ , .
\label{raaqg}\end{aligned}\ ] ] at @xmath64 centrality as a function of @xmath20 at rhic energies .
the grey dotted line shows an energy loss for dcgc1.2 initial conditions considering just binary collisions ( `` jia '' dcgc1.2 ) and ( @xmath65 ) , while the red dashed and dark red dashed - dotted lines dispay glauber initial conditions for ( @xmath66 ) and ( @xmath67 ) , respectively . the initialization time is chosen to be @xmath36 fm .
the data are taken from ref .
] the polytrope model introduced in eq .
( [ genericeloss ] ) is convenient because the initial jet - parton momentum , @xmath68 depends on the final quenched parton momentum @xmath62 analytically as @xcite @xmath69d\tau \right]^\frac{1}{1-a}\hspace*{-2ex } , \label{finmom}\end{aligned}\ ] ] where @xmath70 for gluon(quark ) jets .
non - monotonic density - dependent scenarios as in ref .
@xcite can be simulated by introducing an additional local temperature - dependent function @xmath71 inside the path integral .
however , as noted in the introduction , we limit our applications to monotonic temperature dependencies of the energy loss per unit length given by @xmath72 .
note that if @xmath10 is dimensionless and no additional dimensionful scales influence the energy loss , the temperature exponent depends on @xmath73 .
any deviations from the dimensional constraint @xmath74 found by fitting jet systematics can be used to help identify the existence of other relevant dimensionful scales .
non - conformal physics that may modify energy loss near the cross - over temperature @xmath75 or running coupling effects depending on @xmath76 could lead to violations of eq.([sumrule ] ) . in case of heavy - quark drag holography , ( @xmath77 ) ,
the sum rule is violated because of occurrence of an extra @xmath78 factor of dimension @xmath79 @xcite .
the limit with @xmath14 is of special interest and used frequently in the literature because it leads to a pure exponential dependence of @xmath68 : @xmath80 with a _ jet - energy independent _ effective opacity @xmath81 this corresponds to a generalized `` geometric optics '' limit with @xmath82 if no other relevant scales are involved . in this particular case ,
the nuclear modification factor at the parton level then reduces to @xmath83 approximating the initial jet spectrum by a simple power law , @xmath84 , where @xmath85 is the jet spectral index , the nuclear modification factor reduces to the simplest possible _ energy independent _ `` jet survival probability '' in azimuthal direction @xmath32 given by @xmath86 since at both rhic and lhc the spectral indices of quarks and gluons differ and vary significantly with the jet energy , we must use eq .
( [ a1lim ] ) rather than the naive constraint of eq .
( [ a1limlim ] ) to explore the consequences of @xmath14 energy loss models . for the general @xmath87 case , we use eqs .
( [ raaqg ] ) and ( [ finmom ] ) to compute the parton level nuclear modification factor .
the final pion nuclear modification reads @xmath88 with @xmath89 .
the parton level @xmath90 is averaged over the initial jet production geometry in a given centrality class .
we use the kkp pion fragmentation functions @xcite that have been successfully tested on the @xmath91 spectra at rhic and lhc .
figure [ fig1 ] shows the partonic level @xmath0 and @xmath92 as a function of @xmath20 for gluons ( red dashed - dotted lines ) and quarks ( blue dashed lines ) prior to fragmentation , as well as for pions ( black solid lines ) after fragmentation for a ( @xmath93 polytrope energy loss and glauber initial conditions .
the upper panel displays rhic and the lower panel lhc energies where the coupling @xmath10 was reduced to fit the central 0 - 5% @xmath94 reference point at @xmath95 gev . the purpose of this figure is simply to demonstrate the remarkable similarity of quenched pion and quenched quark jets at both rhic and lhc in spite of the much higher number of initial gluon jets especially at lhc . as can be seen from fig .
[ fig1 ] , the pion nuclear modification factor is completely dominated by quark quenching and fragmentation at both rhic and lhc energies due to the fact that the gluon - energy loss is enhanced by a factor of @xmath96 . the dominance of the quenched quark fragmentation reduces the sensitivity to the fragmentation function uncertainties because quark fragmentation functions are the best constrained experimentally
. from the calculated @xmath97 one obtains the centrality dependent @xmath98 for a given @xmath99 by averaging over the azimuthal angles @xmath32 @xcite .
then , the @xmath100 is computed via @xmath101 we fix the value of the jet - medium coupling @xmath10 at rhic energies in each case by fitting to the same one reference point with @xmath102 in central 0 - 5% collisions . and the @xmath103 at lhc energies after fragmentation , considering either the same coupling as for rhic energies ( left panel ) , or a reduced coupling ( right panel ) .
here we compare glauber initial conditions ( red dashed lines ) and dcgc1.2 initial conditions ( blue solid lines ) for an ( @xmath66 ) energy - loss with a scenario for dcgc1.2 initial conditions considering just binary collisions ( `` jia '' dcgc1.2 ) and ( @xmath65 ) ( grey dotted lines ) .
data are taken from refs .
@xcite . ]
glauber participant and binary collision initial condition geometries are computed using the monte carlo model introduced in ref .
to simulate cgc - like higher eccentricity initial conditions , we simply deform the glauber initial geometry via the rescaling : @xmath104 where the scaling factors are determined by fitting tabulated glauber and mc - kln second moments as a function of @xmath6 : @xmath105 here , @xmath106 denotes the geometric average at a given @xmath6 .
assuming that the ratios of eccentricities ( @xmath107 ) and the root - mean - square ( @xmath108 ) for glauber and cgc initial conditions can be expressed via @xmath109 and @xmath110 , we found that deformations with @xmath111 and @xmath112 reproduce the numerical mc - kln tables of jia very well @xcite . in the following
, we will refer to the deformed cgc geometry corresponding to @xmath113 as `` dcgc1.2 '' .
please note that the glauber geometry applied corresponds to 86% participant plus 14% binary fraction .
given @xmath114 , the deformed cgc geometry has mean in - plane and out - of-(reaction)-plane moments of @xmath115 the energy - density field transforms under this elliptic deformation as @xmath116 we checked that the temperature profiles and eccentricities of this azimuthally deformed glauber model coincide with the ones of the fkln model of drescher et .
@xcite .
in addition , we will also consider a qgp geometry that corresponds to deformed glauber initial conditions that are based on pure binary collisions which leads to a larger eccentricity .
we will refer to that enhanced eccentricity geometry as `` jia '' dcgc1.2 as this geometry , together with @xmath14 , reproduces the phenix @xmath92-data reviewed below . since we showed in ref .
@xcite that event - by - event geometrical fluctuations lead to very similar @xmath0 and @xmath92 as event - averaged geometries , we consider only the event - averaged geometries in this paper .
figure [ fig2 ] shows the centrality dependence of @xmath117 ( lower panels ) and @xmath55 ( upper panels ) for pions with @xmath118 gev at rhic energies considering a polytrope energy - loss scenario with @xmath46 ( left panel ) and @xmath47 ( right panel ) for three @xmath119 models .
red dashed lines correspond to glauber initial conditions for a pqcd - like scenario with @xmath120 as shown in the left panel , while a hybrid quadratic path - length dependence with @xmath121 is displayed in the right panels .
note that there is virtually no dependence on the @xmath122 exponent and for both scenarios , the results for the elliptic flow fall below the phenix data .
the same polytropes for the deformed @xmath113 dcgc1.2 cgc - like geometry are shown in solid blue .
the enhanced eccentricity increases the predicted @xmath123 but the solid blue curves remain well below the @xmath92-data .
therefore , neither geometries nor linear ( @xmath46 ) or quadratic @xmath124 path dependences account for the phenix data considering our default pqcd - like @xmath125 polytrope case .
the long dashed - dotted black curves represents to the results of the `` survival probability '' model used in refs .
that model fits remarkably well both the centrality dependence of @xmath0 and @xmath92 simultaneously at the reference of @xmath118 gev .
this model , corresponds to the doubly special limit [ eq . ( [ a1limlim ] ) ] of the more general `` geometrical optics '' @xmath126 case given by eq .
( [ a1lim ] )
. however , since the spectral index is not constant as a function of @xmath20 at rhic , this double limit is not consistent .
therefore , we also show the dotted grey lines in fig .
[ fig2 ] using the full spectral shape with eq .
( [ raapi ] ) and corresponding to @xmath127 . applying the high eccentricity `` jia '' dcgc1.2 geometry of refs .
@xcite , we find that eq .
( [ raapi ] ) results in @xmath123 close to jia s curve and the data .
the results for @xmath56 are comparable to those shown in ref .
@xcite where , however , the full convolution over the parton spectra and pion fragmentation functions was not performed .
we note that the fkln model used in ref .
@xcite appeared to reproduce the data for @xmath47 , but we discovered that the numerics of the fkln code used was not stable to small variations of parameter settings .
this is the reason we abandoned the fkln calculations in this work and chose instead to compare glauber to deformed glauber dcgc1.2 geometries .
our conclusion with new deformed geometry is that @xmath56 polytropes underestimate @xmath92 at rhic independent of path - length dependence .
if the simple geometric optics model with @xmath128 is the correct explanation of @xmath92-data , then the @xmath51 should be approximately independent of @xmath20 as can be seen from eq .
( [ raapi ] ) . in fig .
[ fig3 ] this prediction is compared to rhic data .
we note that the curve for @xmath49 deceases slightly due to using the full spectral shape as well as the convolution over the pion fragmentation functions .
current rhic data are indeed consistent with an approximately @xmath20 independence in the @xmath129 gev window , but the large systematic errors beyond @xmath130 gev prevent a stringent test of this flatness . as seen from the figure , @xmath131 energy exponents are consistent within the error bars . future higher statistics measurements at rhic in the range @xmath132 gev are obviously needed to differentiate between the energy - loss models .
having constrained the coupling of different polytrope models at rhic , we turn to the predictions of these models for lhc conditions in figs .
[ fig4 ] and [ fig5 ] . here
, the lhc data are taken from refs .
@xcite . in eq .
( [ raapi ] ) we use the parton spectra predicted by pqcd as well as the doubling of the density between rhic and lhc .
we scale the temperature field used in eq .
( [ finmom ] ) by a factor @xmath133 relative to rhic .
[ fig4](b ) is one of the main results showing that all three rhic constrained polytrope models shown in fig .
[ fig3 ] overpredict the jet suppression at lhc for most central collisions at the lhc - reference point of @xmath134 gev .
thus , we find in fig .
[ fig4](b ) that our extrapolation from rhic to lhc conditions assuming the same jet - medium coupling as the density increases by a factor of @xmath135 from rhic to lhc underpredicts the observed @xmath136 at all centralities and that this discrepancy is robust to substantial variations of the @xmath87 exponents of the energy - loss model assumed .
this results extends the robustness of the evidence for a reduced jet - medium coupling at lhc found in refs .
@xcite to a much broader class of models .
the main result found in fig .
[ fig2 ] was that the measured @xmath137 and @xmath138 at rhic at @xmath118 gev appears to be best described with a polytrope model for @xmath49 .
the predicted @xmath20-flatness of the nuclear modification factor shown in fig .
[ fig3 ] is also consistent with the measured data .
however , we by comparing with lhc data in fig . [ fig5 ]
we learn that the @xmath127 dynamics is completely falsified by the rising @xmath51 at rhic energies which is independent of the magnitude of the jet - medium coupling .
5b , in which the jet - medium couplings of the three different polytropes are reduced to fit the @xmath130 gev reference point at most central collisions , clearly shows that , within the current experimental uncertainties , the rapid rise of the @xmath51 at the lhc rules out any energy - loss model with an exponent @xmath11 .
in fact , the data appear to slightly favour the @xmath57 case . as a further consistency test
we compare the centrality dependence of the @xmath139 for the polytrope model with @xmath120 in fig .
[ fig6 ] @xcite to alice data .
the observed agreement between data and model calculations is in fact to robust to changes between glauber and deformed dcgc1.2 geometries for initial conditions . since our results support the conclusion that the qgp created at lhc appears more transparent than expected based on fixed - coupling extrapolations from rhic @xcite , it is important to try to quantify the magnitude of the reduction needed to obtain the agreement shown in fig .
[ fig4](d ) .
the ratio of the coupling needed to reproduce the lhc reference point to the one that fits the rhic data provides a useful measure of the degree of weakening of the effective jet - medium coupling as the qgp density doubles from rhic to lhc . in pqcd
radiative energy loss , @xmath140 and the strong coupling thus scales from rhic to lhc as @xmath141 where @xmath142 . inserting the values used in fig .
[ fig4 ] , which are summarized in table [ tablekappa1 ] for the @xmath46 energy loss shown as well as a @xmath47 energy loss not shown here in detail , we find that @xmath143 for an energy - loss scenario of @xmath56 , indicating a plausible moderate reduction of the pqcd coupling due to slow running ( creeping ) above the deconfinement temperature ( see also ref .
@xcite ) , while for the `` jia '' optical limit @xmath144 and for @xmath145 @xmath146 . remarkably , by comparing the values for @xmath56 in table [ tablekappa1 ] with table [ tablekappa2 ] , we find that the ratio of lhc to rhic effective couplings is insensitive to the assumed @xmath147 in the range 0 - 1 fm / c .
@xmath21 & glauber & dcgc1.2 & glauber & glauber & `` jia '' + & a=1/3 & a=1/3 & a=1/3 & a=0 & a=1 + & z=1 & z=1 & z=2 & z=1 & z=2 + 0.20 & 0.93 & 1.09 & 0.55 & 3.30 & 0.057 + 2.76 & 0.66 & 0.66 & 0.33 & 2.72 & 0.017 + lhc / rhic & 0.71 & 0.61 & 0.60 & 0.82 & 0.33 + on the other hand , in the falling - string scenario @xcite , the effective jet - medium coupling @xmath148 is related to the square root of the thooft coupling @xmath149 .
gravity - dual descriptions require that @xmath150 . for heavy - quark quenching
, it was found in ref .
@xcite that large @xmath151 provides a reasonably good fit to the rhic data as well as to the bulk @xmath92 elliptic flow . in the falling string scenario , @xmath152 and @xmath153
are then related via @xmath154 from the holographic point of view , tables [ tablekappa1 ] and [ tablekappa2 ] imply that @xmath152 must be reduced by a rather large factor of @xmath155 relative to rhic for the @xmath56 scenario and an even larger factor of @xmath156 for the optical `` jia '' limit .
this result implies a rather strong breaking of conformal symmetry over a narrow temperature interval .
it is not yet clear if current non - conformal holographic models are consistent with such a strong variation ( see , for example , refs .
@xcite ) .
@xmath21 & glauber & dcgc1.2 & glauber + & z=1 & z=1 & z=2 + 0.20 & 0.60 & 0.58 & 0.44 + 2.76 & 0.45 & 0.43 & 0.26 + lhc / rhic & 0.75 & 0.74 & 0.59 + figs .
[ fig2 ] and [ fig4 ] already showed that there is a remarkably insensitivity of the elliptic flow to the initial conditions and the strength of the coupling constant [ compare e.g. fig .
[ fig4](a ) with fig . [
fig4](c ) ] . fig .
[ fig7 ] demonstrates that elliptic flow measurements at lhc , which recently are in remarkable agreement between atlas , alice , and cms @xcite will not allow for a disentangling of glauber vs. dcgc1.2 initial conditions .
preliminary data from cms up to @xmath157 gev @xcite ( right panel of fig .
[ fig7 ] ) show that the the elliptic flow for central collisions indeed stays flat for high momenta while @xmath158 for more peripheral collisions decreases slowly approaching the value for central collisions .
using a generic , polytrope power - law model of energy , path - length , and monotonic density dependence characterized by the three exponents @xmath87 that can interpolate between a wide class of weakly and strongly - coupled jet - medium interactions , we investigated those interactions as well as the jet - medium coupling at both rhic and lhc energies for glauber and cgc - like , deformed glauber ( dcgc1.2 ) initial conditions .
we found that at rhic energies the measured data for the nuclear modification factor and the elliptic flow as a function of centrality are best described by the scenario with @xmath49 that also describes the flatness of the @xmath159 .
however , this energy - loss model is completely falsified at lhc energies because it does not describe the rising of the @xmath160 . in fact ,
the rapid rise of the nuclear modification factor rules out any energy - loss model with an energy exponent @xmath11 and slightly favours the @xmath57 case . moreover , we find that an extrapolation from rhic to lhc energies leads to an underprediction of the nuclear modification factor @xcite , independently of the @xmath87 exponents assumed , emphasizing the robust evidence for a reduced jet - medium coupling at the lhc .
we showed that in terms of pqcd a moderate reduction of the effective jet - medium coupling with @xmath161 relative to rhic ( @xmath162 ) allows to describe both the nuclear modification factor and the elliptic flow as a function of centrality and as a function of @xmath20 .
thus , the lhc data are compatible with @xmath15 pqcd - like energy - loss models where the jet - medium coupling is reduced by approximately 10% between rhic and lhc . in terms of strongly - coupled holographic models ,
however , our lhc fit requires a much larger reduction of the effective thooft coupling from @xmath151 by a factor of @xmath163 to @xmath164 .
this suggests that stronger non - conformal effects must be considered for a holographic phenomenology of the lhc .
we found that both glauber and cgc - like , deformed glauber ( dcgc1.2 ) initial conditions describe the centrality dependence of the nuclear modification factor @xmath165 and the elliptic flow @xmath158 well for rhic and lhc energies which does not allow us to disentangle the initial conditions using those two observables .
as demonstrated in fig .
[ fig1 ] , the nuclear modification factor as well as the elliptic flow at both rhic and lhc energies for @xmath166 gev are dominated by quark jet - quenching and fragmentation because gluon jets are strongly quenched and fragmentation leads to pions with a smaller fractional momentum .
this underlines the conclusion of ref .
@xcite that single - hadron jet - flavour tomography observables are mainly sensitive to quark rather than gluon jet - medium interactions .
the reduced jet - medium coupling quantified in this work therefore primarily refers to the apparent weakening of quark - jet interactions in a qgp when the density approximately doubles from rhic to lhc .
it remains a challenge to identify jet observables more sensitive to gluon jet - quenching to test color casimir scaling currently assumed in both weakly - coupled pqcd tomography and strongly - coupled string holography .
di - hadron and jet - shape observables could help to probe gluon versus quark jet - medium interactions in the future .
m.g . and b.b .
acknowledge support from doe under grant no .
de - fg02 - 93er40764 .
is supported by the alexander von humboldt foundation via the feodor lynen program .
the authors thank g. torrieri , j. jia , a. buzzatti , a. ficnar , w. horowitz , j. liao , m. mia , j. noronha , j. harris , g. roland , b. cole , and w. zajc for extensive and fruitful discussions . | recent data on the nuclear modification factor @xmath0 of jet fragments in @xmath1 atev pb+pb collisions at the large hadron collider ( lhc ) indicate that the jet - medium coupling in a quark - gluon plasma ( qgp ) is reduced at lhc energies and not compatible with the coupling deduced from data at the relativistic hadron collider ( rhic ) .
we estimate the reduction factor from a combined fit to the available data on @xmath2 and the elliptic flow @xmath3 at @xmath4 atev over a transverse momentum range @xmath5 gev and a broad impact parameter , @xmath6 , range .
we use a simple analytic `` polytrope '' model ( @xmath7 ) to investigate the dynamical jet - energy loss model dependence . varying @xmath8 interpolates between weakly - coupled and strongly - coupled models of jet - energy dependence while @xmath9 covers a wide range of possible jet - path dependencies from elastic and radiative to holographic string mechanisms .
our fit to lhc data indicates an approximate 40% reduction of the coupling @xmath10 from rhic to lhc and excludes energy - loss models characterized by a jet - energy exponent with @xmath11 . in particular , the rapid rise of @xmath0 with @xmath12 gev combined with the slow variation of the asymptotic @xmath13 at the lhc rules out popular exponential geometric optics models ( @xmath14 ) .
the lhc data are compatible with @xmath15 pqcd - like energy - loss models where the jet - medium coupling is reduced by approximately 10% between rhic and lhc . |
let @xmath5 be separable hilbert spaces and let @xmath6 be a banach space such that @xmath7 is a gelfand triple with continuous and dense embeddings .
we are interested in stochastic evolution equations of the form @xmath8 where @xmath9 is a @xmath10-valued cylindrical wiener process on a probability space @xmath11 with a normal filtration @xmath12 and the maps @xmath13\times v\to v',\qquad b:\omega\times[0,t]\times h\to l_2(u;h)\ ] ] are @xmath12-progressively measurable and satisfy 1 .
[ eq : mu2 ] monotonicity : there exists @xmath14 such that for all @xmath15 , @xmath16 $ ] @xmath17 2 .
[ eq : mu0 ] hemicontinuity : for all @xmath18 , @xmath19 and @xmath16,$ ] the mapping @xmath20 is continuous .
[ eq : mu ] coercivity : there exist @xmath21 , @xmath22 , @xmath23 such that for all @xmath24 , @xmath16 $ ] @xmath25 4 .
[ eq : mu2 ] growth of @xmath26 : there exists @xmath27 such that for all @xmath28 , @xmath16 $ ] @xmath29 5 .
growth of @xmath30 : there exists @xmath31 and @xmath32-adapted @xmath33 such that for all @xmath34 , @xmath16 $ ] @xmath35 the literature devoted to the study of these equations is quite extensive .
the question of existence of a unique ( variational ) solution to equations of the form is well - understood : first results were established in @xcite , for an overview in the above stated generality and further references we refer the reader to @xcite .
existence of a strong solution under various assumptions appeared in @xcite and numerical approximations were studied in @xcite . in the case of linear operator @xmath26 which generates a strongly continuous semigroup , more is known concerning regularity and maximal regularity ( see e.g. @xcite ) . naturally , the time regularity of a solution to is limited by the regularity of the driving wiener process @xmath9 . in particular ,
since the trajectories of @xmath9 are only @xmath36-hlder continuous for @xmath37 , it can be seen from the integral formulation of that the trajectories of @xmath38 are @xmath36-hlder continuous as functions taking values in @xmath39 .
this can be improved if some additional regularity in space of the solution is known , that is , the equation is satisfied in a stronger sense . in this note , we are particularly interested in situations where such additional space regularity is either not available or limited .
this is typically the case when * @xmath40 is linear but the noise is not smooth enough : if @xmath41 is a variational solution to then the standard assumption is @xmath42 .
is the space of weak@xmath43-measurable mappings @xmath44 such that @xmath45 * @xmath40 is nonlinear as for instance the @xmath4-laplacian @xmath46 or a more general nonlinear operator with @xmath4-growth and , in addition , the noise represents the same difficulty as in ( i ) . in order to formulate our main result
we need several additional assumptions upon the operator @xmath26 and the initial datum @xmath47 . on the one hand ,
we introduce a notion of @xmath48-monotonicity which represents a stronger version of the monotonicity assumption on @xmath26 , on the other hand , we suppose certain regularity in time of @xmath26 as well as regularity of the initial condition .
to be more precise , we assume * [ eq : gmon ] @xmath48-monotonicity : there exists a bounded ( possibly nonlinear ) mapping @xmath49 and @xmath50 such that for all @xmath51 , @xmath16 $ ] @xmath52 * time regularity of @xmath26 : there exists @xmath53 such that for all @xmath54 , @xmath55 $ ] @xmath56 * regularity of @xmath47 : @xmath57 a.s . for all @xmath16 $ ] and there exists @xmath58 such that @xmath59 note that it can be readily checked that the operators @xmath26 in the above mentioned examples ( i ) and ( ii ) are @xmath48-monotone . indeed ,
if @xmath40 is linear and symmetric negative definite we can choose @xmath60 and , as was shown in @xcite , the @xmath4-laplacian is covered via @xmath61 which is the natural quantity to establish its regularity properties .
finally we have all in hand to state our result .
[ eq : time ] assume that _
( h1)-(h8 ) _
hold true .
if @xmath38 is a solution to , in particular @xmath62 then @xmath63 if one drops the assumption ( h8 ) then holds locally in time away from 0 .
[ rem : new ] the statement of theorem [ eq : time ] continues to hold if we replace _
( h6 ) _ with the following assumption : * [ eq : gmon ]
modified @xmath48-monotonicity : there exists a separable hilbert space @xmath64 ( generally different from @xmath65 ) and a bounded mapping @xmath66 and @xmath67 such that for all @xmath51 , @xmath16 $ ] @xmath68 in this case we have to replace @xmath65 by @xmath64 in and .
let us now explain what are the main motivations for such a result .
first , it turns out that is the natural time regularity that allows to establish the optimal rates of convergence for numerical schemes based on time discretization ( or a space - time discretization provided a suitable space regularity can be proved as well ) .
indeed , with this time regularity at hand , a finite element based space - time discretization of stochastic @xmath4-laplace type systems will be studied in @xcite .
a similar strategy can be directly applied to establish rates of convergence for time discretization of more general monotone spdes satisfying ( among others ) the key @xmath48-monotonicity assumption .
second , if @xmath26 is a linear infinitesimal generator of a strongly continuous semigroup @xmath69 on @xmath65 then the ( mild ) solution to with @xmath70 is given by the stochastic convolution @xmath71 and our result gives @xmath72 .
recall that the space @xmath73 here is the borderline case regarding regularity for the stochastic convolution , namely , @xmath74 may not even have a pathwise continuous version whereas for @xmath75 has @xmath36-hlder continuous trajectories for @xmath76 ( see ( * ? ? ?
* theorem 5.16 , subsection 5.4.2 ) ) .
consequently , the borderline case is typically not covered by known methods such as factorization @xcite or stochastic maximal regularity ( see ( * ? ? ?
* theorem 1.1 , theorem 1.2 ) ) and theorem [ eq : time ] provides an additional information based on a rather simple argument . a complete proof will be given in @xcite .
it is based on a new version of the it formula which applies to time differences and yields the following : let @xmath77 and @xmath78 $ ] then it holds true a.s .
@xmath79 here @xmath80 denotes the backward it stochastic differential and @xmath81 the quadratic variation process .
the appearance of the backward it stochastic integral comes from the fact that the it formula is applied to the time difference @xmath82 . indeed ,
if @xmath83 denotes the martingale part of @xmath41 , then for every fixed @xmath84 the process @xmath85 is a ( forward ) local martingale with respect to the forward filtration given by @xmath86 , @xmath87 whereas for every fixed @xmath88 $ ] the process @xmath89 is a ( backward ) local martingale with respect to the backward filtration given by @xmath90 , @xmath91.$ ] as the next step , we substitute for @xmath92 and @xmath93 in , take expectation and apply hypotheses ( h5)-(h8 ) . finally we obtain that @xmath94 which implies the required regularity .
in this section we present some concrete examples of problems which are covered by our result .
let us assume that @xmath95 is linear dissipative and symmetric infinitesimal generator of a strongly continuous semigroup on @xmath65 .
then the square root @xmath96 is well - defined and setting @xmath97 ( equipped with the graph norm ) we obtain , for all @xmath51 , that @xmath98 thus the hypothesis ( h6 ) holds true with @xmath99 and theorem [ eq : time ] applies .
let @xmath100 be a bounded lipschitz domain and let @xmath101 . we suppose that @xmath102 satisfies ( h1 ) and ( h5 ) .
we are interested in the system @xmath103 where @xmath104 is a general nonlinear operator with @xmath4-growth , i.e. @xmath105 for all @xmath106 with some constants @xmath107 , @xmath108 and @xmath109 .
then the assumptions ( h1)-(h4 ) are satisfied with @xmath110 and , in addition , it is well known from the deterministic setting ( and was already discussed in @xcite in the stochastic setting ) that an important role for this system is played by the function @xmath111 it is used in regularity theory @xcite and also for the numerical approximation @xcite .
the essential property of @xmath112 can be characterized by the inequality @xmath113 for some positive constants @xmath114 depending only on @xmath4 ( see for instance @xcite ) .
consequently , for all @xmath115 , @xmath116 and therefore ( h6 ) is satisfied and theorem [ eq : time ] yields @xmath117 note that in case of the heat equation ( i.e. @xmath118 ) the operator @xmath119 is the identity . in continuum mechanics ,
the motion of a homogeneous incompressible fluid is described by its velocity field @xmath120 and its pressure function @xmath121 .
if the flow is slow motion can be described via the system @xmath122 where @xmath123 and @xmath124 satisfy the hypotheses of subsection [ subsec : plapla ] and @xmath125 is the symmetric gradient of the velocity field @xmath126 . in comparison to the navier
stokes system the convective term @xmath127 on the right - hand - side of the momentum equation @xmath128 is neglected ( see @xcite for the corresponding navier stokes system for power - law fluids and further references ) . in the following functional analytical setting @xmath129 where @xmath130 the pressure function does not appear .
similarly to the @xmath4-laplace system we set @xmath131 and obtain , for all @xmath132 , @xmath133 where @xmath134 .
corollary [ rem : new ] applies and we gain @xmath135 [ m ] e. acerbi , n. fusco ( 1989 ) : regularity for minimizers of nonquadratic functionals : the case 1<p<2 . j. math .
1 , 115135 . j. w. barrett , w.b .
liu ( 1993 ) : finite element approximation of the p - laplacian .
61(204 ) , 523537 .
d. breit ( 2015 ) : regularity theory for nonlinear systems of spdes .
146 , 329349 .
d. breit ( 2015 ) : existence theory for stochastic power law fluids .
fluid mech .
17 , 295326 .
d. breit , m. hofmanov , s. loisel , g.j .
lord : space - time approximation of stochastic @xmath4-laplace type systems , in preparation .
z. brzeniak , on stochastic convolution in banach spaces and applications , stoch .
61 ( 3 - 4 ) ( 1997 ) 245295 .
g. da prato , j. zabczyk ( 1992 ) : stochastic equations in infinite dimensions , encyclopedia math .
44 , cambridge university press , cambridge .
l. diening and f. ettwein ( 2008 ) : _ fractional estimates for non - differentiable elliptic systems with general growth .
_ forum mathematicum 20 , no . 3 , 523556 . l. diening and m. ruika , _ interpolation operators in orlicz sobolev spaces _ , num .
( 2007 ) , no . 1 , 107129 . b. gess , strong solutions for stochastic partial differential equations of gradient type , j. funct .
, 263 ( 8) ( 2012 ) , 2355 - 2383 .
i. gyngy , a. millet ( 2005 ) : on discretization schemes for stochastic evolution equations .
potential analysis 23 , 99134 .
i. gyngy , a. millet ( 2009 ) : rate of convergence of space - time discretization for stochastic evolution equations .
potential analysis 30 , 2964 .
m. hofmanov ( 2013 ) : strong solutions of semilinear stochastic partial differential equations .
nodea nonlinear differential equations appl .
20 ( 3 ) , 757778 .
n. v. krylov , b. l. rozovskii ( 1979 ) : stochastic evolution equations .
itogi nauki i tekhniki .
14 , viniti , moscow , 71146 ; english transl . j. sov .
, 16 ( 4 ) ( 1981 ) , 12331277 .
e. pardoux ( 1975 ) : equations aux drives partielles stochastiques non linaires monotones .
etude de solutions fortes de type it , ph.d .
thesis , universit paris sud . | we report on a time regularity result for stochastic evolutionary pdes with monotone coefficients .
if the diffusion coefficient is bounded in time without additional space regularity we obtain a fractional sobolev type time regularity of order up to @xmath0 for a certain functional @xmath1 of the solution .
namely , @xmath2 in the case of the heat equation and @xmath3 for the @xmath4-laplacian .
the motivation is twofold . on the one hand
, it turns out that this is the natural time regularity result that allows to establish the optimal rates of convergence for numerical schemes based on a time discretization . on the other hand , in the linear case
, i.e. where the solution is given by a stochastic convolution , our result complements the known stochastic maximal space - time regularity results for the borderline case not covered by other methods . |
lenticular galaxies have been introduced by edwin ( * ? ? ? * hubble ( 1936 ) ) as an intermediate type between ellipticals and spirals : they have large - scale stellar disks as spirals but lack blue spiral arms and hii - regions , and they look smooth and red as ellipticals . now lenticulars are thought to be ( trans-)formed from spirals by removing gas and quenching star formation in their disks ; also dynamical heating is required to make the s0 disks stable against spiral wave perturbations .
there are some evidences that this transformation might take place at the redshift of @xmath1 , within dense environments , groups or clusters , where and when suddenly the dominance of blue ( spiral ? ) galaxies is replaced by s0 dominance ( e.g. , ( * ? ? ?
* fasano et al .
2000 ) , ( * ? ? ?
* wilman et al . 2009 ) ) .
but if this scenario is valid , the star formation in the outer disks of nearby s0s should proceed only 4 gyr ago , and we must see intermediate - age stellar populations there . with deep long - slit spectroscopy of nearby s0 galaxies ,
i study the stellar population properties along the radius beyond several scalelengths of their large - scale disks to check the validity of this paradigm .
the sample consists of nearby lenticular galaxies for which deep long - slit spectra have been obtained at the russian 6 m telescope for the last five years in the frames of several observational programs .
the main part of the sample are edge - on lenticular galaxies selected for kinematical study by natalia sotnikova which have been observed in the frame of her observational proposal ; i use here these data to derive lick indices .
four s0 galaxies seen moderately inclined represent a part of our sample of nearby early - type disk galaxies group members whose central parts have been studied earlier with the multi - pupil fiber spectrograph of the 6 m telescope ( ( * ? ? ? * silchenko 2006 ) ) .
the galaxies are homogeneously distributed over the luminosities , their blue absolute magnitudes being spread from 19 to 21 , and over the environment densities .
we have one galaxy ( ngc 4570 ) in the virgo cluster where the intracluster medium influence is inavoidable , and one galaxy ( ngc 4111 ) in the ursa major cluster where x - ray gas is not detected . among group galaxies , ngc 524 and ngc 5353 are the central group galaxies embedded into x - ray haloes , ngc 5308 is a member galaxy in the x - ray bright group , and ngc 502 and ic 1541 though being members of rich groups , lie outside their x - ray halo ( ( * ? ? ?
* osmond & ponman 2004 ) ; also we have checked the archive asca images ) .
ngc 3414 is a central galaxy in the rich group undetected in x - ray .
ngc 2732 is a host of a few faint satellites .
ngc 1029 , ngc 2549 , and ngc 7332 are in triplets . by using the ned environment searcher ,
we have no found any galaxies within 300 kpc off ngc 1032 and ngc 1184 so we take them as isolated field galaxies .
the long - slit spectral observations have been made with the focal reducer scorpio ( ( * ? ? ? * afanasiev & moiseev 2005 ) ) installed at the prime focus of the russian 6 m telescope ( at the special astrophysical observatory of the russian academy of sciences ) .
we exposed the rather narrow spectral range rich by absorpion lines , 48005500 , which is quite suitable for the study of stellar kinematics and stellar population properties .
the slit width was one arcsecond , the spectral resolution about 2 .
the ccd 2k@xmath22k and 2k@xmath24k were used as detectors , and the scale along the slit was 0.36 arcsec per pixel .
the slit length is about 6 arcminutes so at the edges of the slit we can take the sky background to subtract from the galaxy spectra .
inhomogeneties of optics transparency and spectral resolution along the slit are checked with the twilight exposures .
the lick index system was calibrated by observing the standard lick stars ( ( * ? ? ?
* worthey et al . 1994 ) ) with the same instrumental setup as the galaxies were observed .
we calculated the lick indices h - beta , mgb , fe5270 , and fe5335 along the slit up to several scalelengths of exponential disks . to estimate the radial variations of the ssp - equivalent ages , metallicities and magnesium - to - iron ratios of the stellar populations
, i confronted the measured lick indices to the models of old stellar populations by ( * ? ? ?
* thomas et al . ( 2003 ) ) . , and of the large - scale disks for the sample lenticulars .
the dashed line is the line of equality ; the filled histograms attached to the axes characterize the distributions of the bulge and disk ages.,width=326 ]
i have measured the lick indices in 15 s0s along the radius beyond 24 exponential scalelengths of their disks , and have estimated the ages and abundances of the stellar populations for the bulges and for the disks .
the bulges have solar metallicities or higher , and the disks have solar metallicity or lower .
the disks demonstrate very high magnesium - to - iron ratios of their stellar populations so they can not be descendants of spiral galaxies with prolonged star formation .
the ssp - equivalent ages of the disks are mostly old
60% of the sample demonstrate the ages older than 10 gyr , and almost always ( except one case ) larger than the ssp - equivalent ages of their corresponding bulges ( fig .
[ fig1 ] ) .
the parameters of the stellar populations of the disks do not show any correlation with the disk luminosities or masses .
the only correlations found are those with the photometric disk scaleheight taken from the decomposition results by ( * ? ? ?
* mosenkov et al . ( 2010 ) ) .
the whole scaleheight range is from 0.3 kpc ( found in the disks with the ages of 23 gyr and [ mg / fe]@xmath3 ) to 0.60.9 kpc ( found mostly in the disks with the ages @xmath4 gyr and [ mg / fe]@xmath5 ) .
all the old disks of our sample galaxies are so thick disks , while the only two young outer disks , those of ngc 4111 and ngc 7332 , are certainly thin disks .
the disk of ngc 2732 , with its ssp - equivalent age of 8 gyr and the scaleheight of 0.5 kpc , is halfway between thin and thick disks .
many arguments evidence for s0s and spiral galaxies being relatives , but this time the thing looks like s0s are progenitors of spirals , opposite to what is thought before .
indeed , if we compare stars of the thick disk of our own galaxy with the thick outer stellar disks of s0s studied here we will see full resemblance : the ages @xmath4 gyr , [ mg / fe]@xmath6 , the total metallicity [ z / h ] is between 0.0 and 0.7 ( such parameters of the thick disk of our galaxy are found by e.g. ( * ? ? ?
* bernkopf & fuhrmann 2006 ) or by ( * ? ? ?
* schuster et al . 2006 ) ) .
so if one provides fresh cold gas accretion into the disks of these s0s , after several gyrs of star formation we would get a typical spiral galaxy , with the thick old stellar disk and thin younger stellar disk .
i propose now the following evolutionary sequence .
all disk galaxies were s0s immediately after their birth ; later , at @xmath7 , some of them acquired cold gas accretion sources these became spirals , and some of them failed to find such sources those remained lenticulars .
in large cluster - size and group - size dark haloes , there are little chance to find external sources of cold gas accretion , due to hot gas effect , so in nearby clusters the dominant disk - galaxy population is s0s .
the open question remains what can be these sources of cold - gas prolonged accretion
they may be cosmologically motivated filaments ( ( * ? ? ?
* dekel & birnboim 2006 ) ) or rich systems of irregular - type dwarf satellites which have been merging with the host one after another . | by studying the stellar population properties along the radius in 15 nearby s0 galaxies , i have found that the outer stellar disks are mostly old , with the ssp - equivalent ages of 815 gyr , being often older than the bulges .
this fact puts into doubt a currently accepted paradigm that s0 galaxies have formed at @xmath0 by quenching star formation in spiral galaxies . |
the origin and nature of ultra - high energy cosmic rays ( uhecrs ) remains unsolved in contemporary astroparticle physics .
possible indications of sources or excesses in the arrival direction distribution of uhecrs have been claimed by ground - based experiments , though not fully confirmed @xcite . in order to be identified from earth , extremely powerful sources capable of accelerating cosmic rays up to ultra - high energies must be within a limited range of distances set by the greisen - zatseptin - kuzmin ( gzk ) effect @xcite .
since the distribution of matter within the gzk range is inhomogeneous and anisotropic , one would expect uhecr arrival directions to exhibit a corresponding anisotropy . to identify the sources of uhecrs , measurements of the energy spectrum and arrival directions with high statistics are essential .
this is rather challenging because of the extremely low flux of a few per km@xmath0 per century at extreme energies such as @xmath1 ev .
the observational exposure is , therefore , a critical factor .
jem - euso ( the extreme universe space observatory on - board the japanese experiment module ) @xcite on the international space station ( iss ) @xcite is an innovative space mission .
looking down the earth from space , it utilizes the atmosphere as a detector of cosmic ray air showers with the aim of significantly increasing the exposure to uhecrs compared to the largest ground - based air shower arrays presently in operation @xcite .
the jem - euso telescope will be accommodated on the exposed facility ( ef ) on the jem module _
@xcite of the iss .
the scientific objectives include charged particle astronomy and astrophysics , as well as , other exploratory objectives @xcite such as the detection of extreme energy gamma rays and neutrinos .
the jem - euso telescope exploits the fluorescence light that is emitted during the development of the extensive air shower ( eas ) , initiated by a primary cosmic ray particle in the atmosphere to estimate the particle s energy and arrival direction .
this is an established technique that has been employed by several ground - based uhecr observatories @xcite , but never in space - based observations .
the estimation of the exposure of a space - based experiment such as jem - euso requires accounting for : a ) the characteristics of the eas development in the atmosphere as observed from space , b ) the properties of the telescope , including its orbit and field of view ( fov ) , c ) the various sources of steady background like night - glow and moonlight , d ) the overall optical transmission properties of the atmosphere , in particular the possible presence of clouds , and e ) the effect of anthropogenic light , atmospheric flashes such as transient luminous events ( tles ) and meteors . items a ) and b ) are the principal factors determining the threshold in energy and maximum aperture of the telescope .
item c ) limits the observational duty cycle of the mission .
items d ) and e ) affect the instantaneous aperture of the telescope . in the following sections , all of these aspects will be reviewed .
the outline of this article is as follows : in section [ apparatus ] , we summarize the key aspects of the jem - euso mission . in sections [ background ] and [ sec :
clouds ] , we discuss the estimation of the observational duty cycle , local light effects and the role of clouds .
section [ trigger ] is devoted to the trigger architecture and eas simulation .
computation of the aperture for both clear and cloudy conditions will be described in section [ sec : aperture ] . the exposure in the nadir observation and its uniformity
will then be derived in section [ exp - exp ] .
the paper concludes with a discussion of the results and a summary in sections [ discussion ] and [ summary ] .
the jem - euso telescope @xcite consists of four principal parts : the photon collecting optics @xcite , the focal surface ( fs ) detector @xcite , the electronics @xcite , and the mechanical structure @xcite .
the main parameters of jem - euso telescope are summarized in table [ tab1 ] .
the telescope optics consists of three double - sided curved circular fresnel lenses with 2.65 m maximum diameter .
the minimum diameter of the lenses is 1.9 m owing to cuts on opposite sides .
this shape is referred to as ` side - cut ' and is required to satisfy constraints of the h - iib transfer vehicle ( htv ) _ kounotori _ @xcite which will transport the jem - euso telescope to the iss .
the uv photons are focused onto the fs which consists of 137 photo - detector modules ( pdms ) .
each pdm comprises of a @xmath2 set of elementary cells ( ecs ) .
each ec is formed by a @xmath3 array of multi - anode photomultiplier tubes ( mapmts ) hamamatsu photonics k.k .
r11265 - 03-m64 with @xmath4 (= 64 ) pixels .
each pixel has a spatial resolution of @xmath5 .
the fs detector converts photons into electrical pulses with @xmath62 ns width , which are counted by the electronics during a gate time unit ( gtu ) of 2.5 @xmath7s .
the imaging part of the telescope is an extremely fast , highly pixelized , large - aperture , and wide - fov digital camera .
it is sensitive to near ultraviolet ( uv ) wavelength band between about 300 and 430 nm with single photon counting capability .
the telescope records the spatial and temporal profile of the uv light emitted as an eas develops in the atmosphere .
[ cols="<,>,<",options="header " , ] of table [ table_single ] with the matrix of cloud occurrence of table [ tab2 ] .
filled circles represent the case where we require @xmath8 for optically thick clouds . the cloud efficiency @xmath9 is defined as the average of the values expressed by the filled circles .
the error bar on the left shows the uncertainty on the points ( @xmath63% ) . , width=287 ] in table [ table_single ] , the average values @xmath10 are summarized for the different test clouds and showers simulated with energies above @xmath11 ev and an assumed differential flux of @xmath12 . in case of optically thick clouds with @xmath13 ,
the trigger efficiency depends on @xmath14 .
high altitude clouds in particular absorb eas signals emitted beneath the cloud and result in a significant lowering of the trigger efficiency . at middle altitudes , @xmath15 @xmath16 km , clouds only influence eass of small zenith angles , which develop at lower altitudes . in the presence of high clouds with @xmath17 , the signal from an eas below cloud level
is only attenuated by a factor of @xmath18 and the effect on the trigger efficiency is limited . if @xmath14 is well below the altitudes where eas develop , the clouds do not attenuate the signals . as a result of the different cuts ,
@xmath19 slowly increases with energy . the results in table [ table_single ]
are then weighted with the relative cloud occurrence of table [ tab2 ] to estimate the average effect as a function of energy .
the results are shown in figure [ simp_fig2 ] .
for cloudy cases , the average of all triggered events is shown by triangles and it tends to increase with energy .
the results obtained by applying the selection @xmath8 for @xmath20 cases is indicated by filled circles .
this cut requires that for optically thick clouds , the eas maximum is located above the cloud - top .
this ensures that fitting the eas profile will not introduce significant distortion of the reconstructed eas profile . with the above cut
, the fraction of selected events over the ones triggering in clear - sky conditions ( the reference case ) is almost constant at higher energies .
this is because a certain fraction of clouds with @xmath21 exist at higher altitudes . from table
[ tab2 ] , for example , clouds with @xmath22 km account for @xmath620% of cloud coverage .
therefore , a part of eas develops below such clouds . as
this value is nearly constant as a function of energy , we define it as the ` cloud efficiency ' @xmath9 and it accounts for @xmath23 of the trigger eass above @xmath24 ev .
this is due to the fact that the average @xmath25 dependence is dominated by the zenith angle .
the energy plays a smaller role .
the value @xmath26 is an important factor for estimating the effective exposure of the mission .
currently , a detailed study on the reconstruction of the events passing such a trigger selection as well as those occurring in a clear atmosphere is in progress .
it should be emphasized that the main telescope of jem - euso will be operated along with am system @xcite .
from the above results , the exposure per year of operation for events that trigger jem - euso , defined as the ` annual exposure ' is evaluated as a function of energy : @xmath27)}. \quad\ ] ] in this estimation , we use @xmath28 and @xmath29 , respectively .
the operational inefficiencies related to iss ( rockets docking on iss , lid operation , detector maintenance or aging , etc . ) as well as quality cuts on reconstruction are not taken into account yet , and will be addressed in future . therefore , the present results constitute an upper limit on the effective exposure of the instrument for the assumed conditions . : a ) entire observation area ( filled circles ) ; b ) cut on distance @xmath30 km and on zenith angle @xmath31 ( open squares ) .
, width=287 ] in figure [ fig224 ] , the annual exposure as a function of energy is shown .
the filled circles indicate the geometrical aperture for the entire observation area .
the open squares include a zenith angle cut of @xmath31 and a cut on distance of @xmath30 km .
these two exposures correspond to the highest and lowest aperture curves of figure [ fig2 ] .
the jem - euso annual exposure for the full sample of data is expected to be @xmath6 9 times larger than that of the pierre auger observatory with the corresponding annual exposure of about 7000 km@xmath0 sr yr , at energies around @xmath32 ev . because of the steeply rising aperture at lower energies , the subsets of data with reduced and flat exposure will be used to cross - check with measurements by other ground - based experiments down to @xmath33 ev .
it is important to underline that the cuts shown in figure [ fig224 ] are extreme in order to obtain at @xmath34 ev an annual exposure comparable to that of auger , which means acquiring a statistically similar data sample . as shown in figure [ fig2 ] for the apertures , less stringent geometrical cuts will lead to different exposure curves located in between the two lines shown in figure [ fig224 ] with a flat plateau starting gradually at higher energies . in this way , it will be possible to have a comparison of uhecr fluxes for one entire decade in energy using the data acquired without geometrical cuts .
it is worthwhile remembering here that the aperture and exposure have been derived with specific assumptions on the detector properties , background level , shower development in atmosphere , etc .
all the systematic uncertainties that would increase or decrease the collected light at telescope level , either for the eas or for the background , would be responsible to shift the energy scale of the aperture and exposure curves by the square root of the systematic uncertainty .
on the other hand , the scaling factor would be linear in case it involves only the eas propagation ( i.e. hadronic interaction model , fluorescence yield ) . in the previous analysis , a constant background level of @xmath35 photons m@xmath36 ns@xmath37 sr@xmath37 was assumed .
however , the background is variable with time . to take into account the effective background variation ,
the exposure over the time when @xmath38 given as a function of @xmath39 is approximated by the following relation : @xmath40 figure [ bg - threshold ] shows the exposure as a function of @xmath39 for various maximum allowed background levels obtained by convolving the trigger probability at a specific fixed background level with the fraction of time during which such background level occurs according to the estimation by equations ( [ eqn : obsduty ] ) and ( [ eqn : moonshine ] ) . as described in section [ trigger ] , the trigger system is capable of dynamically adjusting the thresholds to cope with variable background intensity .
the trigger efficiency curve scales in energy approximately in a proportional way to @xmath41 because it depends on the poissonian fluctuations of the average background level .
the exposure obtained with a fixed @xmath42 of 500 photons m@xmath36 ns@xmath37 sr@xmath37 is essentially equivalent to the one obtained from equation ( [ eqn : conv ] ) when the integration of @xmath42 is extended up to 1600 photons m@xmath36 ns@xmath37 sr@xmath37 .
it is possible to observe that at higher energies there is still some margin of gain if a higher level of background is accepted ( see i.e. the curve at less than 6300 photons m@xmath36 ns@xmath37 sr@xmath37 ) .
this is particularly useful to explore the extreme energy ranges where the flux is rapidly decreasing with energy . in any case , all the conclusions obtained in this paper are derived assuming only the standard condition of 500 photons m@xmath36 ns@xmath37 sr@xmath37 constant background level . unlike ground - based observatories , the global iss orbit and better sensitivities for eas with large zenith angles
allows observation of the entire celestial sphere .
the exposure distribution is practically flat in right ascension .
apart from possible local or seasonal deviation from the global average of cloud coverage and of background level , the relationship between the expected overall exposure and declination can be analytically calculated as a function of only @xmath43 , knowing the observable nighttime at a given latitude .
( dashed line ) and @xmath44 ( dotted line ) . the vertical axis indicates the deviation from the uniform distribution.,width=287 ] .
figure [ fig3 ] shows the expected distribution of observed exposure as a function of declination on the celestial sphere with different zenith angle cuts for all zenith angles ( solid line ) , @xmath45 ( dashed line ) and @xmath44 ( dotted line ) . the vertical axis indicates the deviation from the uniform distribution
. for the case of @xmath44 , minor excesses and deficits arise in very limited regions near celestial poles and equator , respectively .
this is because the iss has a slightly longer residence time at high latitudes .
jem - euso can achieve in general a nearly constant exposure for the full range of zenith angles for which the arrival direction analysis will be performed .
this is one of the advantageous features of space - based observation , as only small correction factors are needed .
the convolution of observational duty cycle @xmath46 @xmath620% , cloud efficiency @xmath47 @xmath672% and effect of local light @xmath48 @xmath610% gives an overall conversion factor from geometric aperture to exposure of about @xmath613% with slight variations depending on the cuts applied to the different terms .
this is due to the particular features of observation from space : a ) the possibility of operating in the presence of some diffuse moonlight up to 68% of the time , and b ) the possibility of measuring in a substantial fraction of the solid angle acceptance even in presence of low- or middle- altitude clouds .
as an example , the distribution of cloud occurrence around the location of auger has been compared to the iss case by using tovs data though with statistical errors @xmath49 .
results show that , despite a significantly higher fraction of clear atmosphere at auger site ( @xmath50 ) compared to the world average ( day and night ) spanned by the iss ( @xmath51 ) , when the fraction of time in which the cloud - top is located below 3.2 km is added to the clear atmosphere , both auger and iss orbit give similar results ( @xmath52 ) . for space - based observation
this is the factor that should be compared with the clear atmosphere fraction of a ground - based observatory .
our results are in agreement with those obtained by @xcite for the observational duty cycle .
our result is slightly higher than what has been reported in @xcite for the owl observatory mainly due to the fact that owl will fly at higher altitudes ( @xmath6 1000 km ) , reducing the duty factor for full darkness by about 15%20% due to the shorter nighttime .
moreover , in our estimation we have taken into account the possibility of accepting some moonlight as long as the total diffuse light is less than 1500 photons m@xmath36 ns@xmath37 sr@xmath37 .
concerning clouds , our efficiency is a factor of @xmath53 times larger than @xcite essentially due to the fact that the cut on @xmath25 is applied only to optically thick clouds .
this is motivated by the fact that thin clouds might distort the shower profile but not affect the arrival direction of the primary particle and therefore may be used to perform anisotropy analyses . even though it is not considered in the present analysis , it is interesting to observe that optically thick clouds ( @xmath54 ) , located at low altitudes , may play another positive role , which is blocking the anthropogenic light and , therefore , allowing eas measurement also in regions typically polluted by city light .
our results do not support the conclusions in @xcite , where it was claimed that the sensitivity of space - based fluorescence detectors is of unacceptably small level . in that approach ,
only cloud - free scenes are considered and we know from section [ sec : clouds ] that they account for only 1/3 of the time . moreover ,
very strict conditions were applied on the extension of the cloud - free area . as an example , no correlation on the altitude of the possibly cloudy pixel and location of the track was applied .
this tends to reject very inclined events with much longer paths in atmosphere even though all the detectable part of the track is located above clouds .
it is clear that these constraints severely reduce the overall efficiency .
the authors commented that less rigorous constraints could increase the cloud - free efficiency by even a factor of 3 .
finally , as previously mentioned , simulations have proved @xcite the feasibility of reconstructing eas with reasonable uncertainty in presence of clouds .
it is important to stress that the am system will have an important role in monitoring the atmospheric conditions in which eas develop , together with information from satellites , ground - based observations and meteorological models .
we are , therefore , confident that it will be possible to characterize the atmospheric conditions in which each eass has been detected , and , account for these conditions in the data analyses . finally , it has to be mentioned that most of the above problems are related to the estimation of the energy and @xmath55 of the eas . the estimation of the exposure is more easily assessed and does not depend strongly on the level of accuracy of the eas reconstructed parameters .
the present results indicate that jem - euso has the potential to reach an annual exposure of nearly one order of magnitude higher than auger at energies around 10@xmath56 ev .
however , a final assessment of the annual exposure requires an evaluation of event selection efficiency to ensure the quality of the reconstructed eass in terms of energy , angular resolution , and @xmath55 .
this will be described in detail in a forthcoming paper .
preliminary results ( see @xcite ) , assuming clear atmosphere indicate that this condition is satisfied by most of the events .
it is possible that the full data sample will be subdivided in sub - samples of data peculiar for each analysis . as an example ,
those events reconstructed in thin cloud conditions are still most probably usable for the anisotropy analyses , granted that a lower limit on the energy of the event is assigned with high confidence . moreover
, the possibility of defining a sample of events detected in ` golden conditions ' such as inclined eass in clear - atmosphere will guarantee a cross - check of the reliability of wider samples of events
. it will be possible to significantly increase the exposure by tilting the telescope . in the tilt mode ,
the observation area is scaled by @xmath57 as a function of titling angle @xmath58 of the optical axis from the nadir .
this will increase the sample at the highest energies and help to compensate the reduction of the observation area in case of periods of lower orbiting altitude .
as an example , in case of @xmath59 km , tilting the instrument by @xmath60 would give an observation area similar to the case of @xmath61 km in nadir mode .
at the same time , the observation in nadir mode would extend by @xmath630% the lowest energies where the measurement would be feasible .
the analyses in such ` quasi - nadir mode ' in which the optical axis is tilted by @xmath62 , can be easily assimilated to the nadir one . in case of even larger tilting angles ( @xmath63 ) ,
a dedicated study is necessary to evaluate the performance .
this will be addressed in future .
we wish also to point out that jem - euso has considerably improved with respect to the original extreme universe space observatory @xcite that successfully completed phase - a study within esa .
the main improvements can be ascribed to the baseline optics of the jem - euso telescope @xcite ( with @xmath61.5 better focusing capability ) , to the fs detector @xcite ( @xmath61.6 higher detection efficiency ) , to the better geometrical layout of pdms on the fs that maximizes the filling factor @xcite , and to the improved performance of the electronics @xcite , which allow more complex trigger algorithms @xcite .
the most important factors which determine the annual exposure of jem - euso mission have been reviewed .
the analytical calculations indicate that the operational duty cycle of jem - euso , or the fraction of time in which the eas measurement is not hampered by the brightness of the atmosphere , is of the order of @xmath46 @xmath620% .
the local light such as city light , atmospheric flashes and auroras will reduce the effective instantaneous observational area to @xmath64 of the geometrical area .
the role of clouds has been thoroughly investigated and the cloud efficiency , defined as the ratio of the effective average aperture to the geometrical aperture , is found to be @xmath47 @xmath672% .
all the above factors give an overall conversion factor from geometric aperture to exposure of about @xmath613% .
simulations show that jem - euso can reach almost full efficiency at energies around @xmath65 ev for a restricted subset of events , and for the full aperture at energies @xmath66 ev .
the expected annual exposure of jem - euso around 10@xmath56 ev is equivalent to about 9 years exposure of auger .
this value has to be presented as the potential of jem - euso in nadir mode .
a study of the selection efficiency of events due to quality cuts on reconstruction in clear and cloudy conditions will be performed in future to refine these results .
this work was partially supported by basic science interdisciplinary research projects of riken and jsps kakenhi grant ( 22340063 , 23340081 , and 24244042 ) , by the italian ministry of foreign affairs , general direction for the cultural promotion and cooperation , by the helmholtz alliance for astroparticle physics hap funded by the initiative and networking fund of the helmholtz association , germany , and by slovak academy of sciences mvts jem - euso as well as vega grant agency project 2/0081/10 .
the spanish consortium involved in the jem - euso space mission is funded by micinn under projects aya2009 - 06037-e / esp , aya - esp 2010 - 19082 , aya2011 - 29489-c03 - 01 , aya2012 - 39115-c03 - 01 , csd2009 - 00064 ( consolider multidark ) and by comunidad de madrid ( cam ) under project s2009/esp-1496 . | we evaluate the exposure during nadir observations with jem - euso , the extreme universe space observatory , on - board the japanese experiment module of the international space station . designed as a mission to explore the extreme energy universe from space
, jem - euso will monitor the earth s nighttime atmosphere to record the ultraviolet light from tracks generated by extensive air showers initiated by ultra - high energy cosmic rays . in the present work ,
we discuss the particularities of space - based observation and we compute the annual exposure in nadir observation .
the results are based on studies of the expected trigger aperture and observational duty cycle , as well as , on the investigations of the effects of clouds and different types of background light .
we show that the annual exposure is about one order of magnitude higher than those of the presently operating ground - based observatories . |
game theory provides mathematical concepts and tools for modeling and analyzing interactive scenarios . in _
noncooperative _ game theory , the possible actions of the players are represented explicitly , together with payoffs that the players want to maximize for themselves .
basic models are the _ extensive _ form represented by a game tree with possible imperfect information represented by information sets , and the _ strategic _ ( or `` normal '' ) form that lists the players strategies , which they choose independently , together with a table of the players payoffs for each strategy profile .
the central concept for noncooperative games is the _ nash equilibrium _ which prescribes a strategy for each player that is optimal when the other players keep their prescribed strategies fixed .
every finite game has an equilibrium when players are allowed to _ mix _ ( randomize ) their actions ( nash , 1951 ) .
a game may have more than one nash equilibrium .
finding one or all nash equilibria of a game is often a laborious task . in this paper , we describe the _ game theory explorer _ ( gte ) software that allows to create games in extensive or strategic form , and to compute their nash equilibria .
gte is primarily intended for the applied game theorist who is not an expert in equilibrium computation , for example an experimental economist who designs experiments to test if subjects play equilibrium strategies .
the user can easily vary the parameters of the game - theoretic model , and study more complex games , because their equilibrium analysis is quickly provided by the algorithm .
the analysis of a game with general mathematical parameters is also aided by knowing its equilibria for specific numerical values of those parameters .
as our exposition will demonstrate , gte can also be used for more theoretical research in game theory , for example on strategic stability ( see section [ s - strategic ] ) .
the ease of creating , displaying , and analyzing games suggests gte also as an _ educational tool _ for game theory .
computing equilibria is a main research topic of the authors , and in later sections we will explain some of our results on finding equilibria of two - player games in strategic form ( avis et al . , 2010 ) and extensive form based on the `` sequence form '' ( von stengel , 1996 ) .
scientific algorithms are often implemented as prototopes to show that they work , as done , for example , by audet et al.(2001 ) , or von stengel , van den elzen , and talman ( 2002 ) . however , providing a robust user interface to create games is much more involved , and necessary to make such algorithms useful for a wider research community . in particular ,
the drawing of game trees should be done with a friendly _ graphical user interface _ ( gui ) where the game tree can be created and seen on the screen , and be stored , retrieved , and changed in an intuitive manner .
this is one of the purposes of the gte software presented in this article .
an existing suite of software for game - theoretic analysis is _ gambit _ ( mckelvey , mclennan , and turocy , 2010 ) .
gambit has been developed over the course of nearly 25 years and presents a library of solution algorithms , formats for storing games , ways to program the creation of games with the help of the python programming language , and a gui for creating game trees .
it is open - source software that is free to use and that can be extended by anyone .
given the mature state of gambit and the joint research interests and close contacts with its developers , it is clear that any improvements offered by gte should eventually be integrated into gambit .
other existing game solvers are gameplan ( langlois , 2006 ) and xgame ( belhaiza , mve , and audet , 2010 ) . the main difference of gte to gambit is the provided access to the software and the user interface . in terms of access
, gambit needs to be downloaded and installed ; it is offered on the main personal computing platforms windows , linux or mac .
getting the program to run may require some patience and technical experience with software installions , which may present a `` barrier to entry '' for its use .
in contrast , gte is started in a _ web browser _ via the web address http://www.gametheoryexplorer.org .
all interaction with the software is via the browser interface .
the created games and their output can be saved as files by the user on their local computer .
this avoids the technical hurdles of installing software on the user side , and simplifies updating the software .
the graphical display of game trees in gte is user - friendly and can be customized , such as growing the tree in the vertical or horizontal direction .
gte can even be used just as a drawing tool for games , which can be exported as pictures to file formats for use in papers or presentations . providing application software via the web has the following disadvantages compared to installed software .
first , a higher complexity of the program for the required communication over the internet ; however , manifold standard solutions are freely available .
second , limited control over the user s local computing resources for security reasons .
this is an issue because equilibrium computation for larger games is computationally intensive .
for that reason , this computation takes place on the central public server rather than the user s client computer ; we will explain the technical issues in section [ s - software ] .
third , in our case , currently very limited use of the existing gambit software .
we envisage a loosely coupled integration , in particular with gambit s game solvers .
gte is very much under development in this respect .
we describe gte , first from the perspective of the user , with an example in section [ s - example ] , and the general creation of extensive and strategic - form games in sections [ s - extensive ] and [ s - strategic ] .
we explain the main algorithms for finding all equilibria for games in strategic form in section [ s - algostrat ] , including issues for handling larger games where one has to restrict oneself to finding sample equilibria ( which , in particular , does not allow to decide if the game has a unique equilibrium ) . for extensive games , the computation of behavior strategies , which are exponentially less complex than mixed strategies ,
is outlined in section [ s - sequenceform ] .
the software architecture and the communication between server and client computers is discussed , with as little technical jargon as possible , in section [ s - software ] . in conclusion ,
we mention the difficulties of incentives and funding for implementing user - friendly scientific software , and call for contributions of volunteers .
in this section we describe a simple @xmath0 game , due to bagwell ( 1995 ) , together with its `` commitment '' or `` stackelberg '' variant which has more strategies .
this game is created and analyzed very simply with gte , where we demonstrate the computed equilibria .
our detailed analysis may also serve as a tutorial introduction to the basics of noncooperative game theory . for
a textbook on game theory see osborne ( 2004 ) .
a basic model of a noncooperative game is a game in _
strategic form_. each player has a finite list of strategies .
players choose simultaneously a strategy each , which defines a strategy _ profile _ , with a given _ payoff _ to each player .
the strategic form is the table of payoffs for all strategy profiles . for two players ,
the strategies of player 1 are the @xmath1 rows and those of player 2 the @xmath2 columns of a table , which in each cell has a payoff pair .
the two @xmath3 matrices of payoffs to player 1 and 2 define such a two - player game in strategic form , which is also called a bimatrix game .
bimatrix game .
bottom left : best response payoffs surrounded by boxes ( currently not part of gte output ) .
right : output of the computed equilibria of this game.,title="fig : " ] 8ex bimatrix game .
bottom left : best response payoffs surrounded by boxes ( currently not part of gte output ) .
right : output of the computed equilibria of this game.,title="fig : " ] .... strategic form : 2 x 2 payoff player 1 l r t 5 3 b 6 4 2 x 2 payoff player 2 l r t 2 1 b 3 4 ee = extreme equilibrium , ep = expected payoffs rational : ee 1 p1 : ( 1 ) 0 1 ep= 4 p2 : ( 1 ) 0 1 ep= 4 decimal : ee 1 p1 : ( 1 ) 0 1.0 ep= 4.0 p2 : ( 1 ) 0 1.0 ep= 4.0 connected component 1 : { 1 } x { 1 } .... fig . [ f1 ] shows a @xmath0 game with the payoff to player 1 shown in the bottom left corner and the payoff to player 2 in the top right corner of each cell . in gte , such a game is entered by giving the two payoff matrices ( fig . [ fstratinput ] shows the input screen for a larger example ) , with a graphical display as shown at the top left of fig .
[ f1 ] . at the bottom left the same table is shown with a box around each payoff that is maximal against the respective strategy of the other player ( these boxes are currently not part of gte output ) .
this shows that the bottom strategy @xmath4 of player 1 is the only _ best response _ ( payoff - maximizing strategy ) against both columns @xmath5 and @xmath6 of player 2 . for player 2
, the best responses are @xmath5 against @xmath7 and @xmath6 against @xmath4 . in this game ,
strategy @xmath7 is strictly dominated by @xmath4 and can therefore be disregarded because it will never be played by player 1 .
the best response @xmath6 against @xmath4 then defines the strategy profile @xmath8 as the unique nash equilibrium of the game , that is , a pair of mutual best responses .
the text output at the right of fig .
[ f1 ] , which pops up in a window after clicking a button that starts the equilibrium computation , shows the two payoff matrices and the nash equilibria .
the equilibrium strategies are shown as vectors of probabilities , both in rational output as exact fractions of integers and in decimal . in this case
there is only one equilibrium , which is the pair of vectors @xmath9 and @xmath9 that describe the probabilities that player 1 plays his pure strategies @xmath7 and @xmath4 and player 2 her strategies @xmath5 and @xmath6 , respectively .
this pair of strategies also defines the unique equilibrium component shown as ` { 1 } x { 1 } ` , an output which is of more interest in the example of fig .
[ f3 ] below . the game in fig .
[ f1 ] can also be represented in _ extensive form _ , as shown in fig .
[ f2 ] which is automatically generated by gte .
an extensive game is a game tree with nodes as game states and outgoing edges as moves chosen by the player to move at that node . _
information sets _ , due to kuhn ( 1953 ) , describe a player s lack of information about the current game state and have the same outgoing moves at each node . here
, player 2 is not informed about the move of player 1 , in accordance with the players simultaneous choice of moves in the strategic form . as an extensive game with an information set for player 2 who is uninformed about the move of player 1 . ] and [ f2 ] where player 2 is informed about the first move of player 1 .
right : strategic form of this game as generated by gte ( except for the boxes around the best - response payoffs).,title="fig : " ] 3 the game in fig .
[ f1 ] is an example due to bagwell ( 1995 ) .
it is a simple version of a `` cournot '' game . in the corresponding `` stackelberg '' or _
commitment _ game , player 1 is a leader who commits to his strategy , about which player 2 , as a follower , is informed .
the game tree , and thus the game , is changed by becoming a game of perfect information where each information set is a singleton . to change the game in this way
the information set is dissolved , which is a simple operation in gte .
the new game tree is shown on the left in fig .
then the moves of player 2 , who can react to the choice of player 1 , get new names , here at the right node @xmath10 and @xmath11 instead of the original moves @xmath5 and @xmath6 at the original information set that remain the moves at the left node of player 2 .
a game tree with perfect information can be solved by `` backward induction '' which defines a _ subgame perfect equilibrium _ or spe , which is indeed a nash equilibrium . here
, player 2 s optimal moves are @xmath5 and @xmath11 , which defines her strategy @xmath12 . given these choices of player 2 , the optimal move of player 1 is @xmath7 .
this defines the spe @xmath13 . in general , a strategy in an extensive game specifies a move at each information set of the player , so player 2 has the four strategies @xmath14 , @xmath12 , @xmath15 , @xmath16 listed as columns in the strategic form on the right in fig .
[ f3 ] , which is generated by gte . because player 1 has only one information set given by the singleton that contains the root node of the tree , his strategies are just the moves @xmath7 and @xmath17 .
the spe @xmath13 is one of the cells in the strategic form with the two best - response payoffs @xmath18 and @xmath19 for the two players . in a game tree with perfect information and different payoffs at each terminal node ,
backward induction defines a unique spe .
however , the game has in general additional nash equilibria that are not subgame perfect . in this example
, the strategy pair @xmath20 is also an equilibrium , which can be seen from the strategic form . here
player 2 chooses the right move @xmath6 and @xmath11 at both information sets , and the best response of player 1 is then @xmath4 , with payoffs @xmath21 and @xmath21 to the two players .
this is an equilibrium because neither player can unilaterally improve his or her payoff , under the crucial assumption of equilibrium that the strategy of the other player stays fixed : when player 1 chooses @xmath7 instead of @xmath4 and player 2 plays @xmath16 , then player 1 receives a payoff of @xmath22 rather than @xmath21 and therefore prefers to stay with @xmath4 . in turn , @xmath11 is an optimal move when player 1 chooses @xmath4 .
player 2 can not improve her payoff by changing from @xmath6 to @xmath5 because that part of the game tree is not reached due to the move @xmath4 by player 1 .
this equilibrium is in effect the equilibrium @xmath8 in the original simultaneous game in fig .
[ f1 ] translated to the commitment game where player 2 , even though she can now react to the move of player 1 , always chooses the equivalent of the original move @xmath6 . however , this equilibrium is not subgame perfect because it prescribes the suboptimal move @xmath6 in the subgame that starts with the left node of player 2 .
in addition to these two pure - strategy equilibria , the game in fig .
[ f3 ] has additional equilibria where player 2 makes a random choice at the node that is unreached due to the move of player 1 .
because the node is unreached , any choice of player 2 is optimal because it has no effect on her payoffs , but that random choice must not change the preference of player 1 for his move in order to keep the equilibrium property .
move @xmath4 of player 1 , followed necessarily by move @xmath11 of player 2 , is optimal for player 1 as long as his expected payoff for @xmath7 is at most @xmath21 , which is the case whenever player 2 makes move @xmath6 with probability at least @xmath23 .
the two extreme cases are the pure strategy equilibrium @xmath20 already discussed and the mixed strategy equilibrium where player 1 chooses @xmath4 and player 2 mixes between @xmath12 and @xmath16 with probability @xmath23 each .
this is represented by the probability vector @xmath24 for the four strategies of player 2 .
similarly , an equilibrium that has the same outcome with payoffs @xmath18 and @xmath19 as the spe but a suboptimal random choice between @xmath10 and @xmath11 of player 2 is @xmath25 where player mixes between @xmath14 and @xmath12 with probability @xmath23 each ; @xmath23 is the largest probability that player 2 can assign to @xmath10 so that @xmath7 stays a best response of player 1 .
the preceding analysis is straightforward and simple , but a complete list of all equilibria is nevertheless of interest .
this is provided by gte with the following output : .... strategic form : 2 x 4 payoff player 1 la lb ra rb t 5 5 3 3 b 6 4 6 4 2 x 4 payoff player 2 la lb ra rb t 2 2 1 1 b 3 4 3 4 ee = extreme equilibrium , ep = expected payoffs rational : ee 1 p1 : ( 1 ) 0 1 ep= 4 p2 : ( 1 ) 0 1/2 0 1/2 ep= 4 ee 2 p1 : ( 1 ) 0 1 ep= 4 p2 : ( 2 ) 0 0 0 1 ep= 4 ee 3 p1 : ( 2 ) 1 0 ep= 5 p2 : ( 3 ) 0 1 0 0 ep= 2 ee 4 p1 : ( 2 ) 1 0 ep= 5 p2 : ( 4 ) 1/2 1/2 0 0 ep= 2 decimal : ee 1 p1 : ( 1 ) 0 1.0 ep= 4.0 p2 : ( 1 ) 0 0.5 0 0.5 ep= 4.0 ee 2 p1 : ( 1 ) 0 1.0 ep= 4.0 p2 : ( 2 ) 0 0 0 1.0 ep= 4.0 ee 3 p1 : ( 2 ) 1.0 0 ep= 5.0 p2 : ( 3 ) 0 1.0 0 0 ep= 2.0 ee 4 p1 : ( 2 ) 1.0 0 ep= 5.0 p2 :
( 4 ) 0.5 0.5 0 0 ep= 2.0 connected component 1 : { 1 } x { 1 , 2 } connected component 2 : { 2 } x { 3 , 4 } .... this output gives the four `` extreme '' equilibria described above , in both rational and decimal description .
each equilibrium strategy of a player is preceded by an identifying number in parentheses such as ` ( 1 ) ` and ` ( 2 ) ` for the two strategies of player 1 , each of which appears in two equilibria .
all four equilibrium strategies of player 2 are distinct , marked with ` ( 1 ) ` to ` ( 4 ) ` .
the connected components listed at the end of the output show how these extreme equilibria can be arbitrarily combined : the first connected component ` { 1 } x { 1 , 2 } ` says that strategy ` ( 1 ) ` of player 1 , which is @xmath9 ( the pure strategy @xmath4 ) , together with _ any convex combination _ of strategies ` ( 1 ) ` and ` ( 2 ) ` of player 2 , which are the strategies @xmath24 and @xmath26 ( the latter being the pure strategy @xmath16 ) , defines an equilibrium .
this and the other connected component ` { 2 } x { 3 , 4 } ` describe the full set of nash equilibria of the game . in principle
, gte provides such a complete description of all equilibria for any two - player game , except that the computation time , which is in general exponential in the size of the game , may be prohibitively long for larger games .
in this section , we describe the construction in gte of an extensive game which corresponds to the game of the previous section but where the first player s commitment is imperfectly observed , as studied by bagwell ( 2005 ) .
the game is created in five stages : drawing the raw tree , assigning players , combining nodes into information sets , defining moves and chance probabilities , and setting payoffs .
graphical and other more permanent settings , such as the orientation of the game tree , can also be changed . after explaining the gui operations , we consider the interesting equilibrium structure of the constructed game .
4 -5ex the first stage of creating a new extensive game defines the tree structure , beginning from a simple tree that has a root node with two children , as shown in fig .
[ ftree ] .
clicking on a leaf ( terminal node ) creates two children for that node , and clicking on a nonterminal node creates an additional child .
the next stage is to select players , by selecting a `` player assignment '' button for each player , where the name of the player can also be changed , for example to `` alice '' and `` bob '' instead of the defaults `` 1 '' and `` 2 '' ( which we have not done ) .
clicking on a nonterminal node then asigns the player ( originally unassigned with a black node ) , for example player 2 for the four nonterminal nodes closest to the leaves in fig .
[ fplayers ] . assigning a chance node , represented by a black square
, defines per default uniform probabilities on the outgoing edges , which can be changed .
all nonterminal nodes have to be assigned a player before the next stage can be reached .
it is possible to go back to a previous stage at any time , here to alter the tree structure by adding or deleting nodes .
the next stage is to create information sets by clicking on two nodes ( or information sets ) of the same player , which are then merged into a single information set .
it is necessary that the respective nodes have the same number of children because these will be the moves at the newly created information set .
when an information set is created , the program adjusts , as far as possible , the levels of nodes in the tree so that all nodes in the information set are at the same level and the information set appears horizontally ( if the game tree grows in the vertical direction , otherwise vertically as in fig .
[ fsettings ] ) .
in addition , crossings between edges and information sets are minimized .
[ finfosets ] shows the resulting game tree , which at this stage has its final shape , except for the definition of moves and payoffs .
[ fheadline ] shows a `` headline bar '' that indicates the current stage of the creation of the game tree , which defines either the tree , the players , the information sets , the moves , or the payoffs .
the location of the headline bar relative to the entire browser window can be seen in fig .
[ fsettings ] . in fig .
[ fheadline ] , the current stage is that of creating information sets . underneath the stage indicator
are buttons that define the _ mode _ of operation for the computer mouse .
the default mode at the `` information set '' stage is that of merging two nodes ( more generally the information set that they are currently contained in ) into a larger information set .
the two other mode buttons at this stage are to dissolve an information set into singletons , or ( indicated by the scissors ) to cut it into two smaller sets .
the default mode for the earlier `` tree '' stage is to add children to a node ( indicated by the @xmath27 sign ) .
the alternative mode button ( with the @xmath28 sign ) is to delete a node and all its descendants . at the `` players '' stage
, the mode buttons correspond to the player ( including chance ) to be assigned to the node when clicking on the node .
[ fpayoffs ] shows the game tree after the completed stages of assigning moves ( and chance probabilities ) and payoffs , which work as follows . when the `` moves '' stage is chosen in the headline bar , all outgoing edges at an information set get unique pre - assigned names from a list by traversing the tree in `` breadth - first '' manner . per default
these are the upper - case letters in alphabetical order for player 1 and the lower - case letters for player 2 .
there are two ways to change these default move names : first , by clicking on a move label , which allows to enter an alternative name via the keyboard ; this is not restricted to a single letter and need not be unique . on the left in fig .
[ fmoves ] , this is shown for the chance probability of the right chance move which is changed from @xmath23 to @xmath29 ( which could also be entered as @xmath30 ) . the remaining probability for the other chance move
is automatically set to @xmath31 so that the probabilities sum to one .
a second , quick way to change all move names is the mode button below the `` players '' stage in the headline bar where all move names for a player are entered at once , or altered from the displayed current list ; this is shown in the right picture of fig .
[ fmoves ] for player 2 whose default moves @xmath32 are changed to @xmath33 . 1 at the final `` payoffs '' stage , the leaves of the tree
get payoffs to the two players , which are at first consecutively given as @xmath34 in order to have a unique payoff to identify each leaf . as shown in fig .
[ feditpayoffs ] , a player s payoffs can then be changed at once by replacing them with the intended payoffs for the game , where the numbering helps to identify the leaves .
payoffs can also be changed individually .
in addition , payoffs can be generated randomly .
a zero - sum option can be chosen which automatically sets the other player s payoff to the negative of the payoff to the current player .
the game tree is stored as its logical structure .
its graphical layout is generated automatically .
its parameters can be changed , such as the orientation of the game tree , which can grow vertically or horizontally in either direction , the default being top - down .
[ fsettings ] shows a change of settings so that the game tree grows from left to right .
other parameters such as the colors used for the players , line thickness , fonts , and other dimensions can also be changed .
[ fsettings ] also shows the general layout of the graphical interface in the web browser .
the top left offers file manipulation functions such as starting a new game , storing and loading the current game ( together with its settings ) , and exporting it to a picture format ( .png ) and a scalable graphics format ( .fig ) that can be further manipulated with the ` xfig ` drawing program and converted to .pdf or .eps files for inclusion in documents . on the top right various solution algorithms
can be selected and started , and at the very right zoom operations and change of settings can be selected . .
] the game in fig .
[ fpayoffs ] has an interesting equilibrium structure .
it is due to bagwell ( 1995 ) and represents the commitment game shown earlier in fig .
[ f3 ] , but where the commitment of player 1 is imperfectly observed due to some transmission noise : namely , with a small probability ( here @xmath31 ) , the second player observes the commitment incorrectly as the opposite move , which is represented by the chance moves and the two information sets of player 2 .
the resulting strategic form is shown in fig .
[ fimpsf ] . finding all equilibria with gte
gives the following output , with three equilibria that are isolated , each in a separate component .
.... strategic form : 2 x 4 payoff player 1 la lb ra rb t 5 249/50 151/50 3 b 6 201/50 299/50 4 2 x 4 payoff player 2 la lb ra rb t 2 199/100 101/100 1 b
3 399/100 301/100 4 ee = extreme equilibrium , ep = expected payoffs rational : ee 1 p1 : ( 1 ) 1/100 99/100 ep= 393/98 p2 : ( 1 ) 0 25/49 0 24/49 ep= 397/100 ee 2 p1 : ( 2 ) 0 1 ep= 4 p2 : ( 2 ) 0 0 0 1 ep= 4 ee 3 p1 : ( 3 ) 99/100 1/100 ep= 489/98 p2 : ( 3 ) 24/49 25/49 0 0 ep= 201/100 decimal : ee 1 p1 : ( 1 ) 0.01 0.99 ep= 4.0102 p2 : ( 1 ) 0 0.5102 0 0.4898 ep= 3.97
ee 2 p1 : ( 2 ) 0 1.0 ep= 4.0 p2 : ( 2 ) 0 0 0 1.0 ep= 4.0 ee 3 p1 : ( 3 ) 0.99 0.01 ep= 4.9898 p2 : ( 3 ) 0.4898 0.5102 0 0 ep= 2.01 connected component 1 : { 1 } x { 1 } connected component 2 : { 2 } x { 2 } connected component 3 : { 3 } x { 3 } .... in particular , the original spe @xmath13 of the commitment game with perfect observation and payoffs @xmath18 and @xmath19 has disappeared . because the game does not have any nontrivial subgames ( which are subtrees where each player knows that they are in the subgame ) , the concepts of subgame perfection and backward induction no longer apply in the game with imperfectly observable commitment in fig .
[ fpayoffs ] .
the reason why @xmath13 is no longer an equilibrium is the following .
because player 1 commits to move @xmath7 with certainty , player 2 should choose move @xmath5 when seeing @xmath7 , because this gives her a higher payoff than @xmath6 .
however , when player 2 sees move @xmath4 of player 1 , it must be due to an error in the observation and player 2 would therefore also choose @xmath10 rather than @xmath11 ; in short , @xmath14 and not @xmath12 is a best response to @xmath7 .
however , @xmath35 is not an equilibrium either because against @xmath12 player 1 would choose @xmath4 ; in a sense , player 1 would exploit the fact that player 2 interprets @xmath4 as an erroneous signal .
this , however , seems to imply that player 1 has lost his commitment power due to the noise in the observation .
a bagwell ( 1995 ) pointed out , and as is shown in the above list of equilibria , this loss of commitment power only applies when the players are restricted to use pure strategies .
there is in fact a mixed equilibrium , listed as the component ` { 3 } x { 3 } ` , which has payoffs @xmath36 and @xmath37 to the two players that are close to the `` stackelberg '' payoffs @xmath18 and @xmath19 when no noise is present . here
player 1 himself adds a small amount of noise to the commitment and plays @xmath7 and @xmath4 with probabilities @xmath29 and @xmath31 . in turn ,
player 2 mixes between @xmath14 and @xmath12 with probabilities @xmath38 and @xmath39 .
that is , player 2 chooses @xmath5 with certainty and is indifferent between @xmath10 and @xmath11 because when she sees move @xmath4 , this signal may equally likely be received due to the chance move or due to player 1 s randomization .
the mixture between @xmath10 and @xmath11 is such that player 1 in turn is indifferent between @xmath7 and @xmath4 .
the game also has the pure strategy equilibrium @xmath20 as before , listed as ` { 2 } x { 2 } ` with payoffs @xmath21 and @xmath21 , and another mixed equilibrium ` { 1 } x { 1 } ` with similar payoffs . with gte , the game in fig .
[ fpayoffs ] is created in a few minutes , and the equilibria are computed instantly .
the game does not allow abstract parameters , such as @xmath40 for the error probability as in the analysis by bagwell ( 1995 ) , which is here chosen as @xmath31 .
however , as a quick way to test a typical case of this game - theoretic model , gte is a valuable tool .
in this section we give examples of strategic - form games analyzed with gte .
these also demonstrate the use of gte as a research tool for game theory , for questions on the possible number of equilibria , or the description of equilibrium components in the context of strategic stability . game in strategic form . ]
[ fstratinput ] shows the input of a game in strategic form , currently implemented for two players only . for each player one needs to specify the number of strategies , which are the rows for player 1 and columns for player 2 , and a payoff matrix .
names for the strategies are generated automatically as upper case letters for player 1 and lower case letters for player 2 , and can be changed .
if the strategic form has been generated from the extensive form , then the strategies are shown as tuples of moves , one for each information set .
one can choose a _ zero - sum _ input mode where the payoffs to player 2 are automatically the negative of the payoffs to player 1 .
similarly , one can input a _
game where the square payoff matrices to player 1 and 2 are a matrix @xmath41 and its transpose @xmath42 ; the game is symmetric because it does not change when the players are exchanged . in both cases ,
only the payoff matrix of player 1 is entered .
payoffs can be entered as integers , fractions , or with a decimal point where the display can be switched between fractions and decimals ; internally they are all stored as fractions .
the `` align and update '' button aligns the entries in columns and updates the second player s payoffs for zero - sum and symmetric games .
the input of bimatrix games and the computation of their equilibria has the functionality of the popular webpage of savani ( 2005 ) which has been used tens of thousands of times .
the extra feature in gte is the graphical output as in fig .
[ fimpsf ] , for example , which is accessed by the `` matrix layout '' tab shown at the top of fig .
[ fstratinput ] . the game in fig .
[ fstratinput ] has 75 equilibria , listed as follows ; the output displays each equilibrium in a single line , which we have here broken into two lines , one per player , to fit the page . ....
ee 1 p1 : ( 1 ) 1/30 1/6 3/10 3/10 1/6
1/30 ep= 3/2 p2 : ( 1 ) 1/6 1/30 3/10
3/10 1/30 1/6 ep= 3/2 ee 2 p1 : ( 2 ) 0 0 1/33 5/33 4/11 5/11 ep= 24/11 p2 : ( 2 ) 0 0 5/33 1/33 5/11 4/11 ep= 24/11 ee 3 p1 : ( 3 ) 1/128 0 0 7/64 47/128 33/64 ep= 12/7 p2 : ( 3 ) 0 0 5/21 13/189 59/189 8/21 ep= 297/128 ... ee 74 p1 : ( 74 ) 0 0 13/32 0 0 19/32 ep= 450/103 p2 : ( 74 ) 33/103 70/103 0 0 0 0 ep= 477/16 ee 75 p1 : ( 75 ) 0 0 0 0 1 0 ep= 270 p2 : ( 75 ) 1 0 0 0 0 0 ep= 270 connected component 1 : { 1 } x { 1 } ... connected component 75 : { 75 } x { 75 } .... this game is the smallest known example that refutes a conjecture by quint and shubik ( 1995 ) that an @xmath43 game has at most @xmath44 equilibrium components , here for @xmath45 .
it has been constructed by von stengel ( 1999 ) using methods from polytope theory , with the specific small integers in fig .
[ fstratinput ] described by savani and von stengel ( 2004 , p. 25 ) . studying games with large numbers of equilibria is obviously greatly aided by computational tools ; for another example see von stengel ( 2012 ) .
game which has two equilibrium components , shown on the right .
bottom left : the same game but with _ perturbed _ payoffs so that only component 1 remains ( see bottom right ) .
, title="fig : " ] 4ex game which has two equilibrium components , shown on the right .
bottom left : the same game but with _ perturbed _ payoffs so that only component 1 remains ( see bottom right ) .
, title="fig : " ] .... ee 1 p1 : ( 1 ) 1/3 1/3 1/3 ep= -1/3 p2 : ( 1 ) 1/3 1/3 1/3 ep= -1/3 ee 2 p1 : ( 2 ) 1 0 0 ep= 0 p2 : ( 2 ) 0 1 0 ep= 0 ee 3 p1 : ( 2 ) 1 0 0 ep= 0 p2 : ( 3 ) 0 0 1 ep= 0 ee 4 p1 : ( 3 ) 0 1 0 ep= 0 p2 : ( 3 ) 0 0 1 ep= 0 ee 5 p1 : ( 3 ) 0 1 0 ep= 0 p2 : ( 4 ) 1 0 0 ep= 0 ee 6 p1 : ( 4 ) 0 0 1 ep= 0 p2 : ( 4 ) 1 0 0 ep= 0 ee 7 p1 : ( 4 ) 0 0 1 ep= 0 p2 : ( 2 ) 0 1 0 ep= 0 connected component 1 : { 1 } x { 1 } connected component 2 : { 2 , 4 } x { 2 } { 3 , 4 } x { 4 } { 4 } x { 2 , 4 } { 2 , 3 } x { 3 } { 3 } x { 3 , 4 } { 2 } x { 2 , 3 } ---------------------------------------------------------------- ee 1 p1 : ( 1 ) 1/3 1/3 1/3 ep= -3/10 p2 : ( 1 ) 1/3 1/3 1/3 ep= -3/10 connected component 1 : { 1 } x { 1 } .... fig .
[ fring ] shows at the top left a symmetric `` anti - coordination '' game where the only nonzero payoffs are @xmath46 to both players on the diagonal .
because no payoff is positive , any cell with payoff zero to both players is an equilibrium , and these equilibria are connected by line segments to form a `` ring '' that defines the topologically connected component 2 as in the output shown on the right in fig .
[ fring ] .
the game also has an isolated completely mixed equilibrium shown as component 1 .
interestingly , only component 1 is _ strategically stable _ in the sense that there is always an equilibrium nearby when the payoffs are slightly perturbed . in the game shown at the bottom
left , the payoffs are perturbed so that the only equilibrium that remains is the completely mixed equilibrium in component 1 ; the perturbations from zero to @xmath47 can be changed to independent arbitrarily small positive reals with the same effect .
game which has only a `` ring '' of equilibria as its sole stable component , shown on the right . ] .... ee 1 p1 : ( 1 ) 0 1 0 ep= 0 p2 : ( 1 ) 0 1 0 ep= 0 ee 2 p1 : ( 1 ) 0 1 0 ep= 0 p2 : ( 2 ) 0 0 1 ep= 0 ee 3 p1 : ( 2 ) 1 0 0 ep= 0 p2 : ( 3 ) 1 0 0 ep= 0 ee 4 p1 : ( 2 ) 1 0 0 ep= 0 p2 : ( 2 ) 0 0 1 ep= 0 ee 5 p1 : ( 3 ) 0 0 1 ep= 0 p2 : ( 3 ) 1 0 0 ep= 0 ee 6 p1 : ( 3 ) 0 0 1 ep= 0 p2 : ( 1 ) 0 1 0 ep= 0 connected component 1 : { 1 , 3 } x { 1 } { 2 , 3 } x { 3 } { 3 } x { 1 , 3 } { 1 , 2 } x { 2 } { 2 } x { 2 , 3 } { 1 } x { 1 , 2 } .... the concept of strategic stability is due to kohlberg and mertens ( 1986 ) .
[ fkm ] shows a game that is also symmetric like that at the top left of fig .
[ fring ] ( the symmetry is easy to see due to the staggered payoffs in the lower left and upper right of each cell ) .
it is strategically equivalent ( by subtracting constants from columns of the row player s payoffs and from rows of the column player s payoffs ) to the game of kohlberg and mertens ( 1986 , p. 1034 ) and has a `` ring '' consisting of line segments as its only equilibrium component . as kohlberg and mertens
have shown , any point on that ring can be chosen so that a suitably perturbed game has its only equilibrium near that point , which can be verified by experimenting with numeric perturbations of the game . correspondingly , the entire ring defines the minimal stable component of the game .
game with two equilibrium components , shown on the right . in component 1 ,
player 1 plays @xmath48 and player 2 plays so that player 1 gets at most payoff @xmath49 , shown by the hexagon at the bottom right ( see also hauk and hurkens , 2002 , fig . 4 ) .
component 2 is the pure - strategy equilibrium @xmath50 .
, title="fig : " ] 2ex .... ee 1 p1 : ( 1 ) 0 0 0 1 ep= 0 p2 : ( 1 ) 31/282 5/141 241/282 ep= 0 ee 2 p1 : ( 1 ) 0 0 0 1 ep= 0 p2 : ( 2 ) 0 3/4 1/4 ep= 0 ee 3 p1 : ( 1 ) 0 0 0 1 ep= 0 p2 : ( 3 ) 0 1 0 ep= 0 ee 4 p1 : ( 1 ) 0 0 0 1 ep= 0 p2 : ( 4 ) 3/4 1/4 0 ep= 0 ee 5 p1 : ( 1 ) 0 0 0 1 ep= 0 p2 : ( 5 ) 1/2 0 1/2 ep= 0 ee 6 p1 : ( 1 ) 0 0 0 1 ep= 0 p2 : ( 6 ) 11/16 0 5/16 ep= 0 ee 7 p1 : ( 2 ) 1 0 0 0 ep= 5 p2 : ( 7 ) 1 0 0 ep= 4 connected component 1 : { 1 } x { 1 , 2 , 3 , 4 , 5 , 6 } connected component 2 : { 2 } x { 7 } .... -16ex game with two equilibrium components , shown on the right . in component 1 ,
player 1 plays @xmath48 and player 2 plays so that player 1 gets at most payoff @xmath49 , shown by the hexagon at the bottom right ( see also hauk and hurkens , 2002 , fig . 4 ) .
component 2 is the pure - strategy equilibrium @xmath50 . ]
[ fhaukhurkens ] shows another game related to strategic stability , due to hauk and hurkens ( 2002 , p. 74 ) .
this game corresponds to an extensive game where player 1 can first choose between an `` outside option '' @xmath48 with constant payoff zero to both players or else play in a @xmath51 simultaneous game .
the game has two equilibrium components , the pure equilibrium @xmath50 and the outside option component where player 2 randomizes in such a way that player 1 gets at most payoff zero and therefore will not deviate from @xmath48 .
interestingly , the larger component in this game has the property that it has _
zero but is nevertheless strategically stable , that is , any perturbed game has equilibria near that component , as shown by hauk and hurkens ( 2002 ) .
the larger component in the game in fig .
[ fring ] has also index zero and is not stable , which is what one would normally expect .
the index is a certain integer defined for an equilibrium component , with the important property that the sum over all equilibrium indices is one , and that for generic perturbations each equilibrium has index @xmath52 or @xmath46 , which is always @xmath52 for a pure - strategy equilibrium ( see shapley , 1974 ) .
a symbolic computation of the index of an equilibrium component for a bimatrix game is described by balthasar ( 2009 , chapter 2 ) .
its implementation is planned as a future additional feature of gte .
all currently implemented algorithms of gte apply only to two - player games .
a simple next stage is to allow three or more players at least for the purpose of drawing games and generating their strategic form .
algorithms for finding all equilibria of a game with any number of players using polynomial algebra are described by datta ( 2010 ) ; some of these require computer algebra packages , so far not part of gte .
in this section , we describe the algorithm used in gte that finds all equilibria of a two - player game in strategic form .
this algorithm searches the vertices of certain polyhedra defined by the payoff matrices , which is practical for up to about 20 strategies per player . for larger games ,
one can normally still find in reasonable time one equilibrium , or possibly several equilibria with varying starting points , using the path - following algorithms by lemke and howson ( 1964 ) and lemke ( 1965 ) mentioned at the end of this section .
we assume some familiarity with _ pivoting _ as used in the simplex algorithm for linear programming , which in the present context is explained further in avis et al .
( 2010 ) .
the _ best response condition _ describes , in a general finite game in strategic form , when a profile of mixed strategies is a nash equilibrium .
if player 1 , say , has @xmath1 pure strategies , then his set @xmath53 of mixed strategies is the set of probability vectors @xmath54 so that @xmath55 and @xmath56 .
geometrically , @xmath53 is a _ simplex _ which is the convex hull of the @xmath1 unit vectors in @xmath57 .
if @xmath58 is part of a nash equilibrium , then @xmath58 has to be a best response against the strategies of the other players .
the best response condition states that this is the case if and only if every
_ pure strategy _
@xmath59 so that @xmath60 is such a best response ( the condition is easy to see and was used by nash , 1951 ) .
this is a finite condition , which requires to compute player 1 s expected payoff for each pure strategy @xmath59 , and to check if the strategies @xmath59 in the _ support _ @xmath61 of @xmath58 give indeed maximal payoff .
these maximal payoffs have to be _ equal _ , and the resulting equations typically determine the probabilities of the mixed strategies of the other players . strategic form , and its equilibrium components .
, title="fig : " ] 2 .... ee 1 p1 : ( 1 ) 1 0 ep= 1 p2 : ( 1 ) 1 0 ep= 3 ee 2 p1 : ( 1 ) 1 0 ep= 1 p2 : ( 2 ) 1/2 1/2 ep= 3 ee 3 p1 : ( 2 ) 0 1 ep= 2 p2 : ( 3 ) 0 1 ep= 2 connected component 1 : { 1 } x { 1 , 2 } connected component 2 : { 2 } x { 3 } .... for two players , these constraints define `` best response polyhedra '' that simplify the search for equilibria .
we illustrate this geometric apprach with an example ; for a general exposition and detailed references see von stengel ( 2002 ; 2007 ) .
[ fthreat ] shows a simple game tree which has the spe @xmath8 , and another equilibrium component where player 2 `` threatens '' to choose @xmath5 with probability at least @xmath23 and player 1 chooses @xmath7 . as a function of the mixed strategy of the other player . ]
[ fupper ] shows on the left the mixed strategy @xmath62 of player 2 , which is defined by the probability that she chooses move @xmath6 , say , together with the expected payoff to player 1 for his two pure strategies @xmath7 and @xmath4 .
these expected payoffs are linear functions of @xmath62 ( because the game has only two players ; for more than two players , the expected payoff for a pure strategy of player 1 is a product of mixed strategy probabilities of the other players , which is no longer linear ) .
the right picture shows the expected payoffs for the pure strategies @xmath5 and @xmath6 of player 2 as a function of the mixed strategy of player 1 .
these pictures show when a pure strategy is a _
best response _ against a mixed strategy of the other player : @xmath7 is a best response of player 1 when prob@xmath63 , and @xmath4 is a best response when prob@xmath64 .
strategy @xmath6 of player 2 is always a best response , and @xmath5 is a best response when prob@xmath65 . and @xmath66 in ( [ pq ] ) that define the `` upper envelope '' of the payoffs in fig .
[ fupper ] , with additional circled facet labels for unplayed strategies , and equilibrium strategy pairs . ] in fig .
[ fupperlabel ] , the same pictures are shown more abstractly where @xmath67 and @xmath68 are the expected payoffs to player 1 and 2 , respectively , which are required to be at least as large as the expected payoff for every pure strategy . in a general @xmath3 bimatrix game , let @xmath69 and @xmath70 be the payoffs to player 1 and 2 , respectively , for each pure strategy pair @xmath71 .
analogous to @xmath53 , let @xmath72 be the mixed strategy simplex of player 2 given by @xmath73 .
then @xmath66 and @xmath74 are the _ best response polyhedra _ @xmath75 in fig .
[ fupperlabel ] , @xmath74 is shown on the left and refers to the set of mixed strategies @xmath62 of player 2 together with the best responses of player 1 and corresponding payoff @xmath67 to player 1 .
consider one of the inequalities @xmath76 in the definition of @xmath74 for some pure strategy @xmath59 of player 1 , which in the example is either @xmath7 or @xmath4 .
when this inequality is _ tight _ , that is , holds as equality , then @xmath59 is clearly a best response to @xmath62 .
this tight inequality defines a _ face _ ( intersection with a valid hyperplane ) of the polyhedron , and is often a facet ( a face of maximum dimension ) , as here for both strategies @xmath7 and @xmath4 ; however , for the polyhedron @xmath66 on the right in fig .
[ fupperlabel ] the tight inequality @xmath77 when @xmath78 is the pure strategy @xmath5 of player 2 does not define a facet because it only holds for the single point where prob@xmath65 and @xmath79 , that is , for @xmath80 .
we _ label _ each point @xmath81 of @xmath74 with the strategy @xmath59 of player 1 whenever @xmath59 is a best response to @xmath62 , that is , when @xmath82 .
in addition , @xmath81 is labeled with strategy @xmath78 of player 2 if @xmath83 .
the labels for these _ unplayed _ pure strategies are shown with circles around them in fig .
[ fupperlabel ] .
so any point @xmath84 where prob@xmath85 ( the left edge of @xmath74 ) is labeled with @xmath6 , and any point @xmath86 where prob@xmath87 and hence prob@xmath88 ( the right edge of @xmath74 ) is labeled with @xmath5 .
similarly , in the right picture the left and right edges of @xmath66 have label @xmath4 and @xmath7 , respectively .
the bottom edge of @xmath66 is labeled with the pure strategy @xmath6 , and the left vertex @xmath89 has label @xmath5 in addition to the labels @xmath4 and @xmath6 . with this labeling ,
an equilibrium @xmath90 of the game with payoffs @xmath67 and @xmath68 to player 1 and 2 is given by a pair @xmath91 in @xmath92 that is _ completely labeled _ , that is , each pure strategy , here @xmath93 , of either player appears as a label of @xmath94 or @xmath81 .
otherwise , a _ missing label _ represents a pure strategy that has positive probability but is not a best response , which does not hold in an equilibrium because it contradicts the best response condition . in fig .
[ fupperlabel ] , one equilibrium @xmath90 , indicated by the small disks to mark the two points @xmath94 and @xmath81 , is given by @xmath95 with @xmath96 , which is the pure strategy pair @xmath8 . here
the point @xmath81 of @xmath74 has labels @xmath4 and @xmath5 and the point @xmath94 of @xmath66 has labels @xmath6 and @xmath7 , so these two points together have all labels @xmath93 . a second equilibrium component , indicated by the rectangles , is given by any point @xmath81 on the entire edge of @xmath74 that has label @xmath7 and @xmath89 in @xmath66 which has the three labels @xmath97 .
these are also the two equilibrium components in the gte output in fig .
[ fthreat ] .
the edge of @xmath74 with label @xmath7 is the convex hull of the two vertices @xmath98 with labels @xmath7 and @xmath6 , and @xmath99 with labels @xmath7 and @xmath4 .
each of these vertices of @xmath74 , together with the vertex @xmath100 of @xmath66 , forms a completely labeled _
vertex pair_. for these two vertex pairs of @xmath92 , one label ( @xmath6 or @xmath4 ) appears twice and represents a best - response pure strategy that is played with probability zero , which is allowed in equilibrium .
the _ extreme equilibria _
computed by gte are in fact all vertex pairs of @xmath92 that are completely labeled , which are then further processed so as to detect the equilibrium components .
we give an outline of how these computations work in gte , described in detail and compared with other approaches by avis et al .
( 2010 ) .
_ vertex enumeration _ , that is , listing all vertices of a polyhedron defined by linear inequalities , is a well - studied problem where we use the _ lexicographic reverse search _ method _ lrs _ by avis ( 2000 ; 2006 ) .
this method reverses the steps of the simplex algorithm for linear programming for a certain deterministic pivoting rule .
it starts with a vertex @xmath101 of the polyhedron and a linear objective function that is maximized at that vertex . the simplex algorithm for maximizing this linear function computes from every vertex a path of _ pivoting steps _ to @xmath101 . with a deterministic pivoting rule ,
that path is unique . in lrs
, the pivoting rule chooses as entering variable the variable with the least index ( i.e. , smallest subscript ) that improves the objective function , and the leaving variable via a lexicographic rule .
the unique paths of simplex steps from the vertices to @xmath101 define a tree with root @xmath101 .
the lrs algorithm explores this tree by traversing the tree edges in the reverse direction using a depth - first search . in lrs , as in the simplex algorithm ,
the vertices of @xmath92 are represented by _ basic feasible solutions _ to the inequalities in ( [ pq ] ) when they are represented in equality form via slack variables . here , these are vectors @xmath102 in @xmath103 and @xmath6 in @xmath57 so that the constraints that define @xmath66 and @xmath74 are written as @xmath104 for @xmath105 , @xmath106 , and @xmath107 and @xmath108 so that @xmath109 and @xmath110 .
a feasible solution @xmath111 to ( [ rs ] ) defines a point in @xmath92 and vice versa .
a tight inequality corresponds to a slack variable that is zero .
in fact , the labels for a strategy pair @xmath112 , where @xmath67 and @xmath68 are chosen minimally so that ( [ rs ] ) holds , which determines @xmath6 and @xmath102 , are the pure strategies @xmath59
so that @xmath113 or @xmath114 for player 1 and @xmath83 or @xmath115 for player 2 . the equilibrium condition of being completely labeled is equivalent to the _ complementarity _ condition @xmath116 in a basic feasible solution to ( [ rs ] ) , the labels correspond to the indices @xmath59 or @xmath78 of the nonbasic variables , and to basic variables that happen to have value zero .
basic solutions where basic variables have value zero are called _ degenerate _ , and games where this happens are also called degenerate .
it is easy to see ( see von stengel , 2002 ) that this corresponds to a mixed strategy of a player that has more best responses than the size of its support ; in terms of labels , this defines a point of @xmath66 with more than @xmath1 labels or a point of @xmath74 with more than @xmath2 labels .
an example is the degenerate game in fig .
[ fthreat ] where the pure strategy @xmath7 ( which is a mixed strategy @xmath117 with a support of size one ) has two best responses , which is the point @xmath100 of @xmath66 in fig .
[ fupperlabel ] with three labels @xmath118 . for extensive games ,
the strategic form is typically degenerate , as the examples in fig .
[ f3 ] and [ fthreat ] demonstrate . in lrs ,
all arithmetic computations are done in _ arbitrary precision arithmetic _
( rather than floating point arithmetic with possible rounding errors ) in order to recognize with certainty that a basic variable is zero .
in addition , lrs uses the economical _ integer pivoting _ method where basic feasible solutions are represented with integers rather than rational numbers ; see , for example , von stengel ( 2007 , section 3.5 ) .
after enumerating all vertices of @xmath66 and @xmath74 with lrs , those pairs that fulfill the complementarity condition ( [ compl ] ) are the extreme equilibria of the game .
avis et al .
( 2010 ) also describe an improved method _
lrsnash _ , which is used in gte , where only the vertices of one polytope , say @xmath66 , are generated , and the set @xmath17 of labels _ missing _ from each such vertex @xmath94 is used to identify the face @xmath119 of the other polytope that has exactly the labels in @xmath17 .
if this face is not empty , then each of its vertices @xmath81 defines an extreme equilibrium @xmath90 .
an alternative method to enumerate all extreme equilibria is the eee method of audet et al .
( 2001 ) which performs a depth - first search that chooses one tight inequality for either player in each search step . in avis
et al . ( 2010 ) , eee is implemented with exact integer arithmetic and compared with _
lrsnash _ , and performs better for larger games from size @xmath120 onwards .
given the list of extreme equilibria , the set of all equilibria is completely described as follows ( for details see avis et al . , 2010 ) .
consider the bipartite graph @xmath121 with edges @xmath90 that are the extreme equilibria .
each connected component @xmath122 of this graph defines a topologically connected component of equilibria , as it is output by gte . for each @xmath122 , identify the maximal bipartite `` cliques '' in @xmath121 of the form @xmath123 , that is , every pair @xmath90 in @xmath124 is an extreme equilibrium .
then any convex combination of @xmath125 paired with any convex combination of @xmath126 is also a nash equilibrium , and every equilibrium of the game can be represented in this way .
there may be several such cliques that define a component @xmath122 , as in fig .
[ fring ] and [ fkm ] above . in gte , the maximal cliques @xmath124 are computed with an extension of the fast clique enumeration algorithm by bron and kerbosch ( 1973 ) . for a two - player game in strategic form
, gte computes by default all nash equilibria , as just described .
however , the state - of - the - art algorithms _ lrsnash _ and eee have exponential running time due the typically exponential growth of the number of vertices of @xmath92 .
these equilibrium enumeration algorithms therefore take prohibitively long for games with more than 25 , certainly 30 pure strategies for each player . in addition
, it is np - hard to decide if a bimatrix game has more than one nash equilibrium ( gilboa and zemel , 1989 ; conitzer and sandholm , 2008 ) , so one can not expect to answer that question quickly for larger games . in practice
, a finite game may represent a discretization of an infinite game , where it may be useful to know its equilibria when each player has several hundred strategies . in that case
, it is still possible to use methods that find one equilibrium , such as the classic algorithm by lemke and howson ( 1964 ) .
this is a path - following algorithm that follows a sequence of pivoting steps , similar to the simplex algorithm for linear programming ( for expositions see shapley , 1974 , or von stengel , 2002 ; 2007 ) . like the simplex algorithm
, it can take exponentially many steps on certain worst - case instances ( savani and von stengel , 2006 ) . however , these seem to be rare in practice , and the algorithm is typically fast for random games .
the lemke
howson algorithm has a free parameter , which is a pure strategy of a player that together with the best response to that strategy defines a `` starting point '' of the algorithm in the space of mixed strategies @xmath127 .
varying this starting point over all pure strategies may lead to different equilibria ( however , if that is not the case , the game may still have equilibria that are elusive to the algorithm ) .
gte uses a related but different method that is implemented using the algorithm by lemke ( 1965 ) , which allows for starting points that can be arbitrary mixed strategy pairs @xmath128 ( von stengel , van den elzen , and talman , 2002 ) .
in addition , the computation can be interpreted as the `` tracing procedure '' by harsanyi and selten ( 1988 ) , where players strategies are best responses to a weighted combination of their _ prior _ @xmath128 and the actual strategies that they are playing .
initially , the players only react to the prior , whose weight is then decreased ; equilibrium is reached when the weight of the prior becomes zero .
if starting from sufficiently many random priors gives always the same equilibrium , this does not prove that this equilibrium is unique , but may be considered as sufficient reason to suggest this as a plausible way to play the game .
the computational experiments of von stengel , van den elzen , and talman ( 2002 ) show that for random games , the method finds an equilibrium in a small number of pivoting steps , and that typically many equilibria are found by varying the prior by choosing it uniformly at random from the strategy simplices . for a systematic investigation of this method in gte
, it would be useful to generate larger games as discretizations of games defined by arithmetic formulas for payoff functions ; this is a future programming project where a closer integration with gambit is envisaged .
the standard approach to finding equilibria of an extensive game is to convert it to strategic form , and to apply the corresponding algorithms .
however , the number of pure strategies grows typically _ exponentially _ in the size of the game tree .
in contrast , the _ sequence form _ is a strategic description that has the same size as the game tree , which is used in gte .
we sketch the main ideas here ; for more details see von stengel ( 1996 ; 2002 ) or von stengel , van den elzen , and talman ( 2002 ) .
[ fef5 ] shows an extensive game where player 1 has four moves at the root and two moves at his second information set .
player 2 has perfect information and four singleton information sets with two moves each .
a pure strategy of a player defines a move for every information set , so player 1 has 8 and player 2 has 16 pure strategies . some of these pure strategies can be identified because they describe the same `` plan '' of playing the game .
a _ reduced _ pure strategy does not specify a move at an information set that is unreached due to an earlier own move of the player .
for example , after move @xmath10 player 2 does not need to specify move @xmath129 or @xmath130 because the information set with that move can not be reached .
we represent such an unspecified move by an asterisk `` '' at the place where the move would be listed , so @xmath131 is a reduced strategy where @xmath132 stands for either @xmath129 or @xmath130 , as in the strategy @xmath133 where no reduction is possible .
the reduced strategic form is shown in fig .
[ fsf5 ] , which is a @xmath134 game .
each cell shows the payoffs that result when the two strategies meet . .
in a reduced strategy , any move at an information set that can not be reached due to an earlier own move is left unspecified and replaced by `` '' . ] in this game , a _ mixed strategy _ requires four independent probabilities for player 1 ( the fifth is then determined because probabilities sum to one ) and eleven for player 2 .
in general , the number of reduced strategies is _ exponential _ in the size of the game tree , so there is a large number of mixed strategy probabilities to compute . .
all empty cells have zero entries .
the rows are played with a probability distribution , whereas the columns are played with weights @xmath135 subject to the equations @xmath136 . ]
the _ sequence form _ is a compact strategic description of the game that has the _ same _ size as the game tree . instead of strategies
, it uses _ sequences _ of moves of a given player along a path from the root to a leaf .
the payoff matrices are sparse and contain payoffs for those pairs of sequences of the two players that lead to a leaf , as shown in fig .
[ fseq5 ] for the game in fig .
this table is evidently more compact , and vastly more so for larger games . in this game ,
sequences and reduced strategies coincide for player 1 , because that player does not get any information about the moves of the other player . for player 2 ,
her sequences @xmath137 are played with probabilities @xmath135 that are not distributions over the set of all sequences , but are subject to three separate equations @xmath138 , @xmath139 , @xmath140 .
these equations express the revealed information to player 2 following the respective moves @xmath41 , @xmath4 , and @xmath122 of player 1 , and can be derived systematically from the structure of the information sets ( with one equation per information set , where we have substituted some equations such as @xmath141 for the unused nonterminal sequence @xmath11 ; also , player 2 s empty sequence @xmath142 leads here to the rightmost leaf and has constant probability @xmath143 ) .
the twelve reduced strategies of player 2 are in effect the combinations of one of the sequences @xmath144 combined with one of @xmath145 and one of @xmath146 , respectively . in the sequence form , these sequences are randomized _ independently_.
this randomization translates to a _ behavior strategy _
, where the player randomizes locally over his moves at each information set rather than globally over his pure strategies as in a mixed strategy .
the underlying assumption about the information sets is that of _ perfect recall _ which says that a player does not forget what he knew or did earlier . using the sequence form implies the theorem of kuhn ( 1953 ) that a player with perfect recall can replace a mixed strategy by an equivalent behavior strategy . in gte
, the sequence form is implemented with the path - following algorithm of lemke ( 1965 ) mentioned at the end of the previous section , as described by von stengel , van den elzen , and talman ( 2002 ) , which starts from an arbitrary `` prior '' as a starting vector and finds one equilibrium .
it can be applied to relatively large extensive games where the strategic form is hopelessly large to be used . for smaller extensive games , an enumeration of all nash equilibria based on the sequence form
is implemented as described by huang ( 2011 , chapter 2 ) ; for a related approach see audet , belhaiza , and hansen ( 2009 ) .
the number of independent probabilities in the sequence form is at most that number in the reduced strategic form .
for example , player 2 in fig .
[ fseq5 ] has four such independent probabilities , like @xmath147 , as opposed to eleven for her reduced strategies ; player 1 has four independent probabilities for both sequences and reduced strategies .
huang s approach only uses these independent probabilities and thus keeps a low dimension of the suitably defined best - response polyhedra , which is important for the employed vertex enumeration algorithms that have exponential running time .
in this final section , we describe the architecture of the gte software on the server and client computers , and its development .
the gte program is open - source and part of the gambit project , with the goal of further integration with the existing gambit modules .
the development is time intensive and relies on the efforts of volunteers , where we strongly invite any interested programmer to contribute .
the gte software is accessed by a web browser . behind its web address
runs a _ server _
computer , which delivers a webpage and the game - drawing software of gte to the user s _ client _ computer .
the user can then create games , and store them on his client computer .
the _ game solvers _ for equilibrium computation reside on the server .
they are invoked by sending a description of the game from the client to the server over the internet .
when the algorithm has terminated , the server sends the text with the computed equilibria back to the client where they are displayed .
the client program with the graphical user interface ( gui ) is written in a variant of the javascript programming language called actionscript which is displayed with the `` flash player '' .
flash is a common software that runs as a `` plug - in '' on most web browsers .
we chose flash for gte in 2010 because of the predictability and speed of its graphical display .
however , flash does not work on the ipad and iphone mobile devices ( it does work on regular apple computers ) , with the more universally accepted html5 standard as its suggested replacement . in the future
, we plan to replace the actionscript programs by regular javascript along with streamlining and further improving the gui part of gte . for security reasons
, the client program requires active permission of the user when writing or reading files on the client computer . when storing or loading a game as a local file , this is anyhow initiated by the user and therefore causes no additional delay
we have designed a special xml format for games in extensive or strategic form , where xml stands for `` extensible markup language '' similar to the html language for web pages .
the tree is described by its logical structure .
for example , the leftmost leaf in fig .
[ fthreat ] reached by move @xmath7 is encoded as follows : -.5em ....
< outcome move="t " > < payoff player="1">1</payoff > < payoff player="2">3</payoff > < /outcome > .... in addition , the file contains parameters for graphical settings such as the orientation of the game tree .
the files can be converted to and from the file format used by gambit .
the xml description is also used to encode the game for sending it to the server for running a game solver .
communication with the server requires identifying the user when there are different simultaneous client sessions , and the use of the relevant internet protocols .
these are standard methods used for typical internet transactions , which we use as freely available routines in the java programming language in which we have written the server programs . the first version of gte was written on a publically provided server ( the google app engine ) .
for security reasons , all programs on that platform have to be written in java , and are restricted in their ability to access programs written in other programming languages , called `` native '' code .
in particular , we could not use the c program lrs for vertex enumeration , which was therefore rewritten in java .
however , this requires to duplicate in java any improvement of lrs such as the _ lrsnash _ program mentioned in section [ s - algostrat ] , which is inefficient and prone to errors .
in addition , it is difficult to incorporate other game solvers .
for that reason , the web server for gte is now provided on a university computer , and uses the original lrs code of avis ( 2006 ) .
gte allows to compute all nash equilibria of a two - player game , but the running time increases exponentially with the size of the game .
this means that there is a relatively narrow range of input size ( currently about 15 to 20 strategies per player ) where the efficiency of the implementation ( such as a factor of ten in running time ) matters , and beyond which the computation takes an impractically long time . for small games , the game solvers would run satisfactorily on the client using javascript , although at present there are no good arbitrary precision arithmetic routines for this programming language , apart from having to implement the algorithms separately . for games of `` medium '' size ( below the `` exponential barrier '' ) ,
it is best to use a state - of - the - art algorithm on the server , which works equally well for small games .
however , one problem is that computations of , say , one hour also use the computational resources of the server , which degrade its performance , and which will be slow if there are multiple such computations at the same time . here
, we provide the option of installing the server on any local computer , which then also works independently of an internet connection .
this requires downloading and compiling the gte code , together with the necessary components of a java compiler , a lightweight server program called `` jetty '' to communicate locally with the web browser , and the ( free ) software for compiling the actionscript part of gte into an executable flash program .
this requires time and patience , probably more so than an installation of gambit , but offers the same interface as that known from the web , so that the user may find the investment worth it .
all gte software is open source and free to use and alter under the gnu general public licence ( which requires derived software to be free as well ) .
the software repository is at https://github.com/gambitproject/gte/wiki ( including documentation ) , as part of the gambit project which has long been open - source .
the version control system and collaborative tool `` git '' stores the current and all previous versions of the project .
git efficiently stores snapshots of the entire project with all modules and documentation , which can be branched off and re - merged in decentralized development branches .
the github repository for storing projects developed with git is free to use for open - source software , and offers a convenient web interface for administration and documentation .
open - source software is the appropriate way to develop an academic project such as gte which has limited commercial use .
academic software should be publically available for making computations reproducible , which is increasingly recognized as a need for validating scientific results that rely on computation ( barnes , 2010 ) . on the algorithm side ,
one of the next steps is to integrate other existing game solvers .
foremost among these are the algorithms already implemented in gambit , which should be easily incorporated via their existing file formats for input and output .
these include ( see mckelvey , mclennan , and turocy , 2010 ) algorithms for finding nash equilibria of more than two players with polynomial systems of equations , or via iterated polymatrix approximation ( govindan and wilson , 2004 ) , or simplicial subdivision ( van der laan , talman , and van der heyden , 1987 ) , and the recent implementation of `` action - graph games '' ( jiang , leyton - brown , and bhat , 2011 ) . moreover , larger games should not be created manually but automatically .
one focus of the current development of gambit is to provide better facilities for the automatic generation of games using the python programming language .
so far , the focus of gte development has been to design a robust and attractive graphical `` front end '' for creating and manipulating games .
such an implementation does not count as scholarly research , so it is difficult to fund it directly with a research grant , or to let a phd student invest a lot of time in it . as a topic for an msc thesis or undergraduate programming project ,
usually too much time is needed to become familiar with the existing code or the necessary background in game theory .
the gambit project and gte were sponsored by studentships from the `` google summer of code '' in 2011 and 2012 . here , the most successful projects for gte were those where the students could bring in their technical expertise on web graphics , for example , to complement our background in game - theoretic algorithms .
as with open - source software in general , the development of gte relies on the efforts , competence , and enthusiasm of volunteers , where we were lucky to get help from excellent students listed in the acknowledgments .
we invite and encourage any interested programmer to contribute to this project .
3ex**acknowledgments * * we are indebted to mark egesdal , alfonso gmez - jordana and martin prause for their invaluable contributions as programmers of gte .
mark egesdal designed the main program structure and coined the name `` game theory explorer '' , with financial support from a sticerd research grant at the london school of economics in 2010 .
alfonso gmez - jordana and martin prause were funded by the google summer of code in 2011 and 2012 for the open - source gambit project , and continue as contributing volunteers .
karen bletzer wrote conversion programs between gte and gambit file formats .
wan huang implemented the enumeration of nash equilibria based on the sequence form .
we thank theodore l. turocy for inspiring discussions and for his support of gte as part of gambit .
david avis has written the _
code for computing all nash equilibria of a two - player game .
all contributions and financial support are gratefully acknowledged .
= -2emby2em = .25ex plus.1ex minus.1ex audet , c. , s. belhaiza , and p. hansen ( 2009 ) , a new sequence form approach for the enumeration of all extreme nash equilibria for extensive form games .
international game theory review 11 , 437451 .
avis , d. ( 2000 ) , lrs : a revised implementation of the reverse search vertex enumeration algorithm . in : polytopes combinatorics and computation , eds . g. kalai and g. ziegler , dmv seminar band 29 , birkhuser , basel , pp .
177198 .
kuhn , h. w. ( 1953 ) , extensive games and the problem of information . in : contributions
to the theory of games ii , eds . h. w. kuhn and a. w. tucker , annals of mathematics studies 28 , princeton univ . press , princeton , pp . 193216 .
van der laan , g. , a. j. j. talman , and l. van der heyden ( 1987 ) , simplicial variable dimension algorithms for solving the nonlinear complementarity problem on a product of unit simplices using a general labelling
. mathematics of operations research 12 , 377397 . | this paper presents the `` game theory explorer '' software tool to create and analyze games as models of strategic interaction .
a game in extensive or strategic form is created and nicely displayed with a graphical user interface in a web browser .
state - of - the - art algorithms then compute all nash equilibria of the game after a mouseclick . in tutorial fashion , we present how the program is used , and the ideas behind its main algorithms .
we report on experiences with the architecture of the software and its development as an open - source project .
* keywords * game theory , nash equilibrium , scientific software |
Subsets and Splits
Data Mining Abstracts
This query searches for and retrieves 2000 documents containing the phrase 'data mining' in the abstract, providing limited but specific document samples.