Microscopic energy flows in disordered Ising spin systems
aa r X i v : . [ c ond - m a t . s t a t - m ec h ] J un Microscopic energy flows in disordered Ising spinsystems
E. Agliari , , , M. Casartelli , and A. Vezzani , Dipartimento di Fisica, Universit`a di Parma, Viale G.P. Usberti n.7/A (ParcoArea delle Scienze), 43100 - Parma - ITALY INFN, gruppo collegato di Parma, Viale G.P. Usberti n.7/A (Parco Area delleScienze), 43100 - Parma - ITALY Albert-Ludwigs-Universit¨at Freiburg, Hermann-Herder-Str. 3, 79104 -Freiburg - GERMANY S3, National Research Center, CNR-INFM, via Campi 213/a, 41100 - Modena-ITALY
Abstract.
An efficient microcanonical dynamics has been recently introducedfor Ising spin models embedded in a generic connected graph even in the presenceof disorder i.e. with the spin couplings chosen from a random distribution. Sucha dynamics allows a coherent definition of local temperatures also when openboundaries are coupled to thermostats, imposing an energy flow. Within thisframework, here we introduce a consistent definition for local energy currents andwe study their dependence on the disorder. In the linear response regime, when theglobal gradient between thermostats is small, we also define local conductivitiesfollowing a Fourier dicretized picture. Then, we work out a linearized “mean-field approximation”, where local conductivities are supposed to depend on localcouplings and temperatures only. We compare the approximated currents withthe exact results of the nonlinear system, showing the reliability range of themean-field approach, which proves very good at high temperatures and not soefficient in the critical region. In the numerical studies we focus on the disorderedcylinder but our results could be extended to an arbitrary, disordered spin modelon a generic discrete structures.keywords: Transport processes (Theory), Heat conduction, Disordered systems(Theory)
1. Introduction
The transport properties featured by systems in stationary states, far from equilibriumis of both theoretical and practical interest: On the one hand, there exist non trivialproblems (e.g. validity bounds of the Fourier description at microscopic scale, influenceof spatial and topological inhomogeneities featured by the substrate, etc.), which stilllack an exhaustive solution; on the other hand, from nanoscales to biological matter,the emergence of new materials poses a challenging number of practical problems(e.g. fluctuations, clusterizations, correlations, role of geometrical irregularities, etc.),where the previous, theoretical approaches get an applicative relevance [1, 2, 3, 4].Discrete models, from simple or interactive random walks to classical spin models,play an important role for all such questions [5, 6, 7, 8, 9, 10, 11, 12]. In particular,here we deal with an Ising system coupled with thermostats imposing an energy flow,and we study its behavior at microscopic scales. icroscopic energy flows in disordered Ising spin systems J ij >
0, have been considered in [6]. The cylindrical geometry of the lattice suggestedto average observables (temperature, magnetic energy, etc.) not only in time but alsoalong the “columns”, i.e. periodic rings at equal distance from the thermostats, andhence orthogonal to the flux flowing along “rows”. This procedure clearly acceleratesthe numerical convergence to well stabilized stationary values. In this frame, thevalidity of the Fourier picture for the energy transport is ensured only on the average,by verifying that the average energy flux through a column is proportional to theaverage temperature difference between the sides of the column. However it is clearthat, due to quenched disorder, the local temperature fluctuates even within a singlecolumn. A natural question is then whether such spatial fluctuations influence thelocal transport properties.In this perspective, a basic problem, constituting a non trivial task in itself, isa self-consistent definition of microscopic currents able to account for the substrateinhomogeneity (here generically referring to topological disorder and/or non-constantcouplings). In the present paper we give a consistent solution to this problem which,remarkably, applies to arbitrary, connected supports. The microscopic link currentsoriginating in each microscopic move can be defined indeed independently of thesubjacent geometry.Afterwards, resuming the case of the disordered ferromagnetic cylinder, this verygeneral definition is used to check the validity of the Fourier law at microscopic scale.First, we verify to this end that, locally, the currents depend on the disorder in anon trivial way. More precisely, local currents and temperatures depend not only onthe relative coupling J ij , but also on the whole configuration of magnetic couplings.Second, we show that, notwithstanding such a complex behavior, a local conductivitycan be defined on each link. In fact, in a linear response regime (i.e. for small enoughtemperature differences ∆ T at the borders), the ratio between the local currents andthe local temperature gradients are independent of ∆ T , providing a good definitionfor the local conductivities K ij . Clearly, each K ij also depends in a non trivial wayon the system parameters, its value being indeed determined not only by the localcoupling and temperature but also by the actual realization of the global disorder.Despite such a complex situation, a simple correlation between local conductivitiesand local couplings is present, since on the average the conductivities increase with J ij . Hence, we introduced an approximated linear fitting K MF ij = AJ ij + B , where K MF ij are the approximated “mean field” conductivities and the constants A and B depend on the system parameters only, i.e. they are independent of the disorderrealization. Within this approach the conduction properties (i.e. the currents) canbe simply evaluated by solving the discretized linear Fourier equations with suitableborder conditions. The approximated currents obtained in this way may be comparedwith the exact results of the fully non linear system. We evidence that the approachworks quite well at large enough temperatures, while near T c (the critical temperature icroscopic energy flows in disordered Ising spin systems
2. Microcanonical Dynamics
The very novelty of the dynamics introduced in [6] consists in assigning to the links,beside the usual magnetic energy E mij due to the spin configuration, a “kinetic” energy E kij , i.e. a non-negative definite quantity whose variation can compensate the positiveor negative gaps of magnetic energy determined by the spin flips of the adjacent nodes.This is clearly different from the Creutz microcanonical procedure [14, 15], where therequired energy compensation is extracted or assigned from a bounded amount ofenergy lying on the nodes themselves: in the latter case, in fact, both the energyboundness and the connectivity of the nodes could entail many limitations, rangingfrom geometrical or topological constraints to the non ergodicity of the system at lowenergy density.Precisely, an elementary move consists in the following:(i) starting from a random distribution of link energies, extract randomly a link ( i, j )and one of the possible four spin configurations for it;(ii) evaluate the variation ∆ E m of the magnetic energy due to this choice, checkingthe whole neighborhood of the link. Obviously, if couplings are ferromagnetic,being J nk ∈ [1 − ε , ε ] for any link ( n, k ) and ε >
0, then ∆ E m is a real number,and if ε = 0 then ∆ E m is an integer whose value depends on the connectivity ofthe k and n nodes;(iii) if ∆ E m ≤
0, accept the choice and increase the link kinetic energy E kij of ∆ E m ;(iv) if ∆ E m >
0, accept the choice and decrease the link kinetic energy of ∆ E m onlyif the link energy remains non negative.The unit time step will be a series of N moves, where N is the number of links.Since a link is defined only by the adjacent nodes, and nothing in the rule above refersto a definite structure (e.g. a lattice), the move is implementable on every non-orientedconnected graph. This ensures the great generality of the dynamics.Starting from the observation that magnetic and kinetic energies behave asnon correlated observables, in [6] many points have been supported by theoreticalarguments and tested numerically, for both homogeneous and disordered links. Inparticular, at equilibrium the Boltzmann distribution is recovered and the system isergodic at all temperatures. This opens the possibility to extend the definition oftemperature T ij as a local observable, in fact a link observable, which recovers the icroscopic energy flows in disordered Ising spin systems h E kij i . This holds also in non-equilibrium, stationary states,i.e. states forced by thermostats at different temperature.As for thermostats, their definition requires in general the presence of contactborders, which for the cylinder are the first and last columns. In this case, everythermostat consists in a number of additional columns (2 are enough) regulated atevery step by the usual Metropolis equilibrium dynamics with the wanted inversetemperature β .
3. Definition of Local Currents
In order to calculate local conductivities also in the presence of disorder (due to eithertopological and coupling inhomogeneity) it is necessary to define a current for eachlink. We now introduce the scheme through which we are able to consistently assigna current to any arbitrary link in the structure considered. Such a scheme can beapplied to a generic structure, as envisaged in Fig. 1.In our dynamics energies are naturally assigned to each links; however in orderto define link currents it is useful to assign one half of the link energy to each of itsrelevant nodes. Let us consider the link ( i, j ) connecting i and j ; Let V i and V j theirneighborhoods, z i = | V i | and z j = | V j | their respective coordinations. Then, k ∈ V i and h ∈ V j are the neighbouring site labels for i and j , respectively. We fix a directionfor currents on adjacent links: currents on links connected through i are incomingwhile those connected through j are outgoing.Now, given a spin flip involving the link under consideration, namely either the i -th spin, or the j -th spin or both, for adjacent links as well as for ( i, j ) we have apossible energy variation denoted with ∆ E ik , ∆ E jh and ∆ E ij respectively. For links( i, k ) and ( h, j ) half of such variations contributes to the current on the link itself sinceit represents the energy flow outgoing and incoming from the sites k and h . Thereforewe have I ki = − ∆ E ki , I jh = + ∆ E jh , where different signs derives from the flow direction we have chosen (see Fig. 1).As for I ij and the energy variation ∆ E ij , they satisfy the following X k ∈ V i I ki − I ij = X k ∈ V i ∆ E ki E ij , (1) − X h ∈ V j I jh + I ij = X h ∈ V j ∆ E jh E ij . (2)In the above equations, the left-hand side represents the currents arriving anddeparting from i (and j ) while the right-hand side is the consequent energy variation.Hence, we get ∆ E ij = − P k ∈ V i ∆ E ki − P h ∈ V j ∆ E hj , as consistent with energyconservation, and I ij = 12 − X k ∈ V i ∆ E ki + X h ∈ V j ∆ E jh . (3)This scheme works for any arbitrary topology and, of course, even in the presenceof a disordered distribution of couplings: It only requires the knowledge of the localenergy variations consequent to any spin-flip. icroscopic energy flows in disordered Ising spin systems I k i I ij jik k I k i I k i I k i k k I jh I jh I jh h h h Figure 1.
Schematic representation of currents for an arbitrary link belongingto an arbitrary graph. In this example the link considered is the one labelled as( i, j ) with z i = 4 while z j = 3. The arrows indicates the direction attributed tothe energy flow on each link.
4. Numerical results
In the following we report and discuss the results obtained by means of Monte Carlosimulations performed on squared cylinders of N = L × L sites, for different realizationsof disorder (encoded by the N × N matrix J ). The first and last columns of L sites areopen, in contact with thermostats at temperatures T and T (to fix ideas T < T ),and different choices of temperatures are considered. Since we are interested in localquantities, we especially focus on small sizes, which allow fast thermalization thoughdisplaying the relevant features of the non-equilibrium behavior (see [6]).First of all, let us consider local currents and local temperatures. For a singlerealization of the disorder in the window (1 − ǫ, ǫ ), currents I ij may be calculatedaccording to the scheme described in the previous section; one can also measure thelocal temperatures T ij which, for randomly distributed couplings equal the relevantaverage kinetic energy h E kij i [6]. Then, from such local temperatures T ij , we canestimate a temperature T i to associate to each node, namely T i = 1 | V i | X j ∈ V i T ij , (4)Therefore, the local temperature gradient among nodes i and j is naturally given by∆ T ij = T i − T j .Numerical data for the average currents h I ij i and ∆ T ij , as a function of thepertaining coupling strength J ij , are shown in Fig. 2; similar results are obtained fordifferent realizations J . We notice that, being ∆ T = T − T the global difference oftemperature, local gradients are distributed around the expected value ∆ T /L , withlarge spread especially for low temperatures T , T and a slight correlation with the icroscopic energy flows in disordered Ising spin systems J ij I i j J ij ∆ T = 5 . T = 4 . T = 3 . T = 2 . T = 1 . Figure 2.
Local currents I ij (left panel) and local temperature gradients ∆ T ij (right panel) versus the corresponding coupling J ij for a cylindrical lattice withlinear size L = 10, ∆ T = 0 . T , as shown by the legend.The coupling pattern J is the same for all sets of data points depicted. pertaining couplings, that is, large interaction strengths J ij correspond to smallergradients ∆ T ij . Local currents display a larger degree of correlation: large interactionstrength J ij correspond to larger magnitudes for currents h I ij i . We also notice thatdifferent temperatures for thermostats give rise to similar, though shifted, distributionsof data points. Other realizations of the same disorder give, of course, different pointdistributions, but the linear interpolation and the value of the fluctuations prove to bevery robust, so that the definition of interpolating currents ¯ I ij is reliable at every fixed ǫ and T , T . It is also noteworthy that, for a fixed ∆ T , currents are not monotonic in T : referring to Fig. 2 (left panel), their magnitude is maximum at T = 3, i.e. aroundthe critical temperature expected for the (disordered) two-dimensional Ising model[6]. Moreover, when ∆ T ≪
1, the linear response theory holds: Both the local currents h I ij i and the local temperature gradients ∆ T ij are proportional to the global differenceof temperature ∆ T , as corroborated by the collapse of data points in Fig. 3, where thevalues relevant to different temperatures are compatible with the numerical error. Inthis perspective, given local currents and local gradients, we can introduce the localconductivities according to K ij = h I ij i T i − T j , (5)which are well defined quantities describing the microscopic conduction of the system.We expect that a similar definition works as well for generic topologies, at least in theregime of small gradients.From energy conservation and equation (5) we obtain the local Fourier equation icroscopic energy flows in disordered Ising spin systems J ij I i j / ∆ T ∆ T = 0 . T = 0 . T = 0 . Figure 3.
Local currents I ij divided by the global temperature difference ∆ T as a function of the pertaining coupling J ij . As explained by the legend differentgradients, namely ∆ T = 0 .
05, ∆ T = 0 . T = 0 . T = 5, the coupling pattern J and L = 10 are kept fixed. Notice that theaverage difference between data pertaining to ∆ T = 0 .
05 and ∆ T = 0 .
1, and datapertaining to ∆ T = 0 .
1, ∆ T = 0 . . .
4, therefore belowthe error ≈ characterized by link dependent conductivities X j ∈ V i K ij ( T i − T j ) = − ∂ h E i i ∂t , (6)where h E i i = 1 / P j ∈ V i h E ij i is the total energy relevant to site i ; the continuumnotation is used for convenience, with the usual warning about the meaning ofderivatives in these discrete-time systems (see for instance [16]). In general terms,the expression in Eq. 6 describes a system where an external field, or gradient, alongone axis and a fluctuating local field, or disorder, have been applied; the former makesthe temperature increase by a constant amount per row of nodes, while the lattergives rise to currents non-trivially depending on the whole environment. Indeed, thesame equation is also used in the context of random resistor networks [17], where thevoltage and the conductance play the role of the temperature and of the conductivity,respectively.We remark that conductivities K ij depend on system parameters in a verycomplex way, indeed their values is determined by the temperature T , by the degreeof disorder ǫ and by the whole coupling pattern J . Moreover, we verified that the icroscopic energy flows in disordered Ising spin systems J ij K i j ǫ = 0 . ǫ = 0 . ǫ = 0 . Figure 4.
Local conductivities K ij versus pertaining local couplings J ij for asquare cylinder with linear size L = 10 and thermostats at temperatures T = 4 . T = 5 .
0, respectively; different degrees of disorder ǫ = 0 . ǫ = 0 . ǫ = 0 .
2, have been considered and represented with different symbols, as shownin legend. Linear curves represent the best fits, whose angular coefficients areapproximately 358, 279 and 226 respectively. dependences on the three arguments are intrinsically interplaying, namely that given K ij = f ( J , T, ǫ ), factorizations like f ( J , T, ǫ ) = f ( J ) · f ( T, ǫ ) are ruled out.As shown in Fig. 4, local conductivities are correlated on the average with thelocal couplings, and such a correlation gets stronger (the fitting curve has larger slope)for smaller values of ǫ . The “ordered system limit”, i.e. ǫ →
0, is in a sense singular,since the distributions of the J ij shrinks in a single point.Now, from such local conductivities it is possible to derive an estimate for theconductivity K ( T, J ) expected for a system at a temperature T and in the presenceof disorder J , by assuming ∆ T ≪ K ( T, J ) = 1 N X i X j ∈ V i K ij . (7)In Fig. 5 we compare such measures realized at different temperatures with a“mesoscopic” measure of K based on the heat flow passing from one layer to thenext one in a similar cylinder [6]. The very good agreement between the two estimatesprovides a further confirmation about the consistency of our definition of local currentsand conductivities. icroscopic energy flows in disordered Ising spin systems T K Figure 5.
Comparison between average local conductivities measured accordingto Eq. 7 ( • ) and according to a “mesoscopic” measure of conductivity (continuousline) [6], respectively, as a function of temperature.
5. Mean Field Approach
Within the linear response approach, local temperatures can be evaluated by solvingthe Fourier equation (6) imposing the stationarity of local energy at every node: X j ∈ V i K ij ( T i − T j ) = 0 . (8)Suitable boundary conditions should be chosen forcing the temperatures at the bordersto be fixed at T and T + ∆ T respectively. Then currents can be evaluated fromtemperatures as h I ij i = K ij ( T i − T j ). Clearly, Eq. 8 is useless for practical calculations,as K ij has to be obtained from the (numerical) solution of the whole spin dynamics.Moreover, we have already evidenced that K ij ’s depend on the whole configuration oflocal couplings J ij in a non trivial way. However Fig. 4 suggests that a simple “meanfield” approximation should be possible imposing K ij to be dependent only on the localcoupling J ij , i.e. the mean-field, local conductivities are defined as K MF ij = AJ ij + B ,where A and B are the fitting parameters used in Fig. 4 depending on T and ǫ only. Clearly, this approach is much simpler since, once A and B are known, themean-field, local conductivities can be inferred for any realization of the disorder.Then, using the linear equations (8), one can obtain mean-field, local temperatures P j ∈ V i K MFij ( T MFi − T MF j ) = 0 and currents I MFij = K MF ij ( T MFi − T MFj ). icroscopic energy flows in disordered Ising spin systems φ ij and φ MF ij are defined as φ ij = h I ij i − ¯ I ij , (9) φ MF ij = I MF ij − ¯ I ij , (10)we can quantify the correlation between the real values h I ij i of currents, i.e. thoseobtained from numerical simulations, and the estimate values I MF ij , i.e. those obtainedfrom the mean-filed approach, by means of the correlation coefficient C = E ( φ ij · φ MF ij ) − E φ ij E φ MF ij q(cid:2) E ( φ MF hn ) − ( E φ MF ij ) (cid:3) , (11)where averages E are obtained summing over the whole set of links and ¯ I ij is thevalue of the local current obtained with the linear fit of Figure 2. Notice that C ranges from − C = 0 means no correlation.We measured the quantity C finding strictly positive values for all the temperaturesconsidered. The positivity of correlation evidences that I MF ij approximates currents h I ij i better than ¯ I ij . This remarkable property is easily explained since the mean fieldfield approach not only takes into accounts the correlations between local couplings,currents and conductivities evidenced in Figs. 2-4, but also takes into account the localconservation of energies encoded in Eq. 8 and representing one of the basic featuresof the microscopic spin dynamics. In particular, on the contrary of ¯ I ij , currents I MFij are conserved at every node. It is also worth noting that larger values of C , andtherefore a better efficiency of the the mean-field approximation, are found for largetemperatures. For instance, at T = 5 we get C = 0 .
63; as the temperature is loweredthe correlation decreases displaying a possible minimum around T c . Critical effectsapart, the mean-field approach seems to provide good estimates especially for largetemperatures: indeed for T approximately larger than 3 . C > . T = 500, C is close to 0 .
94. We also underline that C = 1 means thatlocal conductivities are purely local quantities, independent of the neighborhood.We remark that the links crossed by large currents may be sharply identifiedwithin this approximation, which, for example, captures the link evidenced in Fig. 6:although its intermediate coupling value, it carries a large current. In other terms,our approximation is able to locate the regions characterized by high currents which,in realistic realizations, could lead to a failure of the link itself.
6. Conclusions and Perspectives
In this work we addressed the general problem of a spin model on arbitrary discretestructures with a microcanonical dynamics, focusing on the conduction properties atmicroscopic scales. First, we introduced a consistent definition of local currents. Thenwe applied such general definition in the case of disordered ferromagnetic Ising modelon a cylindrical structure, where boundaries are coupled with thermostats at differentfixed temperatures. We highlighted that the local microscopic currents depend nontrivially on the whole distribution of quenched couplings and a consistent definitionof local conductivity has been introduced, at least in regime of linear response, i.e.when the temperature gradient ∆ T is small. icroscopic energy flows in disordered Ising spin systems J ij I i j I MF ij I ij ¯ I ij T C Figure 6.
Main figure: ¯ I (line), I MF ( (cid:3) ) and I nk ( • ) versus the related couplingstrength for T = 4 . T = 5; analogous results are found also for othertemperatures. Inset: temperature dependence for the correlation parameter C . In spite of the aforementioned dependence of local quantities (currents,temperatures and conductivities) on the whole disorder arrangement, numerical resultsevidenced a simple correlation between local conductivities and the pertaining localcouplings. These correlations suggested the development of an approximated mean-field like approach, where local conductivities depend on local couplings only, andthe conduction properties (i.e. the currents) can be easily evaluated by solvingthe discretized linear Fourier equations with suitable border conditions. Suchapproximation is especially effective at large temperatures. With respect to previouswork on the cylindrical spin model, the strict requirement on the smallness of ∆ T isdue precisely to the fact that in the present case we focus on the microscopic aspects.This pattern of results naturally indicates preferential lines of futuredevelopments. First of all, the definition of currents holds on a general geometricalsubstrate. In a complex topological structure, some aspects of the framework we haveworked out are expected to be still valid, in particular the presence of a linear responseregime and the definition of local conductivities whose values may depend, however,in a non trivial way on the topology of the underlying structure.Moreover, an important point to investigate concerns the extent of correlationlength, as a function of temperature, degree of disorder and topology. In other terms,it would be interesting to determine the length of the radius such that the externalpattern constitutes a practically uniform background, without any influence on thelocal currents and conductivities. Also the response of local conductivity to small icroscopic energy flows in disordered Ising spin systems T ≫
1, analytical and numerical estimates say that this radiuspossibly reduces to one link, which means that the conductivity is a local property.We also infer the interest of a more precise characterization of the behaviour ofthe system both at low temperatures and for T ≈ T c , where the linear approximationhas some difficulties, possibly not only of numerical nature. Finally, the whole set ofresults clearly indicates that we have an extremely flexible calculation engine readyfor more complex realizations of disorder, including the case of null and negativecouplings, i.e. diluted and spin-glass systems. [1] Lebon G, Jou D and Casas-V´azquez J, 2008 Understanding Non-Equilibrium Thermodynamics:Foundations, Applications, Frontiers , Springer-Verlag, Berlin (Germany)[2] Van Vliet CM, 2008
Equilibrium and Non-equilibrium Statistical Mechanics , World ScientificPublishing Co., London (UK)[3] Chowdhury D, Nishinari K, Santen L, Schadschneider A, 2010
Stochastic Transport in ComplexSystems: From Molecules to Vehicles , Elsevier Science[4] Astarita G, Ocone R, 2002
Special Topics in Transport Phenomena , Kindle Edition.[5] Lepri S, Livi R and Politi A, 2003
Phys. Rep. , 1[6] Agliari E, Casartelli M and Vezzani A, 2009
J.Stat. Mech.
P07041[7] Frehland E, 1980
Biophys. Chem. , , 63.[8] Gouyet J-F, Plapp M, Dieterich W and Maass P, 2003 Adv. Phys. , , 523.[9] Agliari E, Casartelli M and Vezzani A, 2007 Eur. Phys. J. B , 499[10] Agliari E, Casartelli M and Vezzani A, 2008 Eur. Phys. J. B , 257[11] Harris R and Grant M, 1988 Phys. Rev. B , 9323[12] Lecomte V, R´acz Z and van Wijland F, 2005 J. Stat. Mech.
P02008[13] Agliari E, Casartelli M and Vivo E, submitted[14] Creutz M, 1983
Phys. Rev. Lett. , 1411[15] Creutz M, 1986 Annals of Physics , 62[16] Casartelli M, Macellari N and Vezzani A, 2007
Eur. Phys. J. B , 149[17] Kirkpatrick S, 1973 Rev. Mod. Phys.45