Anomalous heat equation in a system connected to thermal reservoirs
Priyanka,∗ Aritra Kundu,† Abhishek Dhar,‡ and Anupam Kundu§ International Centre for Theoretical Sciences, Tata Institute of Fundamental Research, Bengaluru 560089, India (Dated: March 20, 2018) We study anomalous transport in a one-dimensional system with two conserved quantities in pres- ence of thermal baths. In this system we derive exact expressions of the temperature profile and the two point correlations in steady state as well as in the non-stationary state where the later describe the relaxation to the steady state. In contrast to the Fourier heat equation in the diffusive case, here we show that the evolution of the temperature profile is governed by a non-local anomalous heat equation. We provide numerical verifications of our results.
Transport of energy across an extended system is a — we refer to this model as the harmonic chain momen- paradigm of the vast class of non-equilibrium phenom- tum exchange (HCME) model. For the infinite HCME ena. At a macroscopic level this phenomena is often system it was shown exactly that the current autocorre- 1/2 described by the phenomenological Fourier’s law which lation has a t− decay [17]. It was also shown that, relates the energy current density j(x, t) to the gradient in contrast to∼ Eq. (1), the evolution of an initially local- of the temperature field T (x, t): j = κ ∂xT where κ ized energy perturbation satisfied a non-local fractional − 3/4 is the thermal conductivity. This law implies diffusive diffusion equation ∂te(x, t) = c( ∆) e(x, t), where energy flow across the system described by the Fourier e(x, t) is the energy perturbation− and− c is some constant heat equation [12–14, 18]. The fractional laplacian operator ( ∆)3/4 in the infinite space is defined by its Fourier spectrum:− 2 3/2 ∂τ T (x, t) = D ∂xT (x, t), (1) q whereas the same for the normal Laplacian opera- | | 2 2 3/4 tor ∆ ∂x is q . In real space ( ∆) operator is where D = κ/c (for simplicity we assume κ and the spe- non-local− ≡ [19, − 20]. − cific heat c to be independent of temperature). This While all these studies consider transport in isolated equation is widely used in experiments to understand systems, quite often the transport set up in an exper- the spreading of local energy perturbations in equilib- iment consists of an extended system connected at the rium as well as the non-equilibrium dynamics of systems two ends to heat baths at different temperatures. For connected to reservoirs. diffusive systems Eq. (1) continues to describe both non- Surprisingly, several theoretical [1–3], numerical as well equilibrium steady state (NESS) and time-dependent as experimental studies [4] suggest that in many one and properties in this set-up. It is then natural to ask: what two dimensional systems heat transfer is anomalous in would be the corresponding evolution equation for the the sense that Fourier’s law is not valid [5–7]. This temperature profile in the case of anomalous transport phenomenon is usually manifested by several interesting in the experimental set-up? A major problem that now features like divergence of thermal conductivity κ with arises follows from the fact that the fractional Laplacian system size L as κ Lα; 0 < α < 1, power-law de- is a non-local operator and so extending its definition to cay of the equilibrium∼ current-current auto-correlations, a finite domain is non-trivial. Several studies have ad- super-diffusive spreading of local energy perturbations, dressed this issue, using a phenomenological approach, nonlinear stationary temperature profiles (even for small in the context of Levy walks and Levy flights in finite temperature differences) and the presence of boundary domains [21, 22]. It is thus crucial to have examples of singularities in these profiles [1, 6–14]. specific microscopic models of systems exhibiting anoma- There is currently no general framework to describe lous transport, for which the time-evolution equation in and explain anomalous heat transport. Recently, the the- an open system set up can be derived analytically, and ory of nonlinear fluctuating hydrodynamics has been re- where one can see the non-local and fractional equation markably successful in predicting anomalous scaling of forms explicitly. This is the main aim of this Letter. dynamical correlations of conserved quantities in one- Such attempts have recently been made in [23–25] arXiv:1803.06857v1 [cond-mat.stat-mech] 19 Mar 2018 dimensional Hamiltonian systems and the corresponding where the problem of non-linear steady state tempera- slow decay of the equilibrium current-current autocorre- ture profiles and their time-evolution in the HCME model lations [2, 3, 15, 16]. This approach provides diverging was addressed. The specific model studied was a har- thermal conductivity (via Green-Kubo formula) as well monic chain of N particles where, in addition to the as Lévy scaling for the spreading of local energy perturba- Hamiltonian dynamics, the momenta of nearest neigh- tion. On the rigorous side, computations were done for borhood particles is exchanged randomly at a constant a model of harmonic chain whose Hamiltonian dynam- rate γ. The chain is attached to two Langevin baths at ics was supplemented by a stochastic part that kept the the two ends at temperatures T` and Tr. This system has conservation laws (number, energy, momentum) intact three conserved fields: the stretch r = q q (where i i+1 − i 2 qi, i = 1,...,N are the particle positions) the momen- and η2j = pj, one finds that both the above equations 2 2 tum, pi and the energy i = pi /2 + ri /2. This system can be expressed in a single equation: η˙m = ηm+1 ηm 1 1/2 − − shows anomalous current behavior j N − as well for m = 1, 2, ..., L. The system can also be interpreted as exhibits a non-linear stationary temperature∼ profile as a fluctuating interface where the algebraic volume of 2 Ti = pi /2 ss = (i/N), which was computed analyti- the interface at site m is given by ηm and the energy cally forh fixedi andT free boundary conditions — surpris- V (η) = η2/2 [13]. Hence, the stochastic exchange part in ingly the temperature-profile was different for these cases Eq. (2) can be thought of as a volume-energy conserving [23, 25]. The evolution of the non-stationary tempera- noise. We call this model as ‘harmonic chain with volume ture profile (x, τ) (where the τ = t/N 3/2 is the rescaled exchange’ (HCVE). time) to theT NESS profile was also studied [26], where by It has been shown that the HCVE model defined on eliminating the fast variables it was shown that (x, τ) an isolated infinite one dimensional lattice (i.e. λ = 0 T satisfies an energy continuity equation. From an analysis in Eq. (2) with i = , .., 1, 0, 1, .., ) exhibits super of this equation it was noted that the evolution appears diffusion of energy [18]:−∞ − ∞ to be similar to the fractional diffusion equation. How- ever, so far this has not been clearly established and in ∂te(x, t) = L [e(x, t)], − ∞ particular an explicit representation of the correspond- 1 3/4 1/4 (3) L = [( ∆) ( ∆) ], ing fractional evolution operator is not known. In this ∞ √2γ − − ∇ − Letter, we look at a simpler model of anomalous trans- port in one dimension where we derive the corresponding where the skew-fractional operator L has the Fourier ∞ fractional evolution equation for the temperature profile representation q 3/2(1 i sgn(q)) with i = √ 1 and | | − − inside a finite domain and show explicitly how this evo- sgn(q) is the Signum function. In this paper, however, lution approaches to the appropriate fractional diffusion we consider the HCVE model on a finite lattice of size L operator in the infinite domain. in open set up i.e. connected to heat baths at the two This model consists of a finite one dimensional lat- ends as described in Eq. (2). It is known that in this tice of L sites where each site carries a ‘stretch’ vari- case also, as in HCME, the stationary current scales as 1/2 able ηi, i = 1, 2, ..., L under an onsite external potential j L− [18]. 2 ∼ V (ηi) = ηi /2. The lattice is attached to two thermal Results.- We explicitly find that in the large L limit reservoirs at temperatures T` and Tr on the left and right the stationary ‘energy’ current jss is given, in the leading ends, respectively and subjected to a volume conserving order, by stochastic noise. The dynamics of this model has two r parts: (a) the usual deterministic part plus the Langevin 1 π (T` Tr) jss = − . (4) terms coming from the baths and (b) a stochastic ex- 2 γ √L change part where ηs from any two neighboring sites, In the non-stationary regime, we numerically find that chosen at random, are exchanged at some rate γ. The the temperature profile and the two- dynamics is given by Ti(t) = V (ηi(t)) point correlations C (t) = η (t)hη (t) forii = j have the i,j h i j i 6 dηi following scaling forms = V 0(ηi+1) V 0(ηi 1) dt − − p i t + δ λV 0(η ) + 2λT ζ (t) Ti(t) = , i,1 − 1 ` ` (2) T L L3/2 p (5) + δi,L λV 0(ηL) + 2λTrζr(t) 1 i j i + j t − Ci,j(t) = | − |, , , √LC √L 2L L3/2 + stochastic exchange at rate γ in the leading order for large L. The scaling functions with fixed boundary conditions (BCs) η0 = ηL+1 = 0. (y, τ) and (x, y, τ) satisfy the following equations in- Here ζ`,r(t) are mean zero and unit variance, indepen- Tside the domainC = 0 x ; 0 y 1 : dent Gaussian white noises. Note that, in contrast to D { ≤ ≤ ∞ ≤ ≤ } the HCME case, this dynamics has two conserved quan- 2 ∂y (x, y, τ) = γ∂x (x, y, τ) (6) tities: the ‘volume’ η and the energy V (η ). This model C − C i i ∂ (y, τ) = 2γ [∂ (x, y, τ)] (7) was first introduced by Bernardin and Stoltz in the closed yT − xC x=0 ∂ (y, τ) = 2∂ (0, y, τ), (8) system setup [13] where starting from the harmonic chain τ T yC with Hamiltonian given earlier, they have treated the with (x, y, 0) x = 0 and (x, y, 0) = 0. We find that positions qis and the momenta pis on the same foot- C | →∞ C ing. Note that for harmonic chain, the dynamics of the the exact solutions of these equations are given by ‘stretch’ variable ri = qi+1 qi and the momentum vari- − (y, τ) = ss(y) + r(1 y, τ) (9) able are similar: r˙i = pi+1 pi and p˙i = ri+1 ri for T T T − − − (x, y, τ) = ss(x, y) + r(x, 1 y, τ). (10) i = 1, 2, ..., N. Hence for N = L/2, defining η2j 1 = rj − C C C − 3
In the above equation, NESS part of the profiles are 1.10 =1.0 p =2.0 ss(y) = Tr + (T` Tr) 1 y, 1.05 T − − Theory r ! (11) T` Tr π x ) y
ss(x, y) = − erfc , y p ( 1.00 0.0 0.2 0.4 0.6 0.8 1.0 C − 4 γ 4γ(1 y) 0.00 ss x =1.0, =1.0
−
1.10 satisfies the following equations τ=0.1
τ=0.2 2 ∂z r(x, z, τ) = γ∂x r(x, z, τ), (16) 1.05 τ=0.3 C C ∂z r(z, τ) = 2γ [∂x r(x, z, τ)]x=0 , (17) 0.00 SS T C )
⌧ ∂τ r(z, τ) = 2∂z r(0, z, τ), (18) 1.00 -0.02 T − C y, ) ( ⌧ x =0.25
x, y, C 0.95 ( -0.06 r(x, z, τ) x = 0. The above equations are obtained
1
d ˆr function relaxes very fast over much shorter time scale T = κB ˆr, (19) [O(L)] comparedC to the evolution time scale [O(L3/2)] of dτ T the temperature field . Due to this fact, Eq. (6) and where B is an infinite order matrix with elements Bm,n T Eq. (7) do not involve the time derivative. As a result specified in [27] and κ = 1/√πγ. While it is difficult to the correlation function evolves adiabatically obeying solve this infinite order matrix equation analytically, we the (anti-)diffusion Eq. (6),C with a drive at the boundary solve it numerically by truncating it at some finite order. by the time dependent temperature field through Eq. (7). In Fig. 2, we compare the evolution from this numerical The equation for the temperature profile given in Eq. (8) solution with the same obtained from direct numerical is in the expected continuity equation. simulation of Eq. (2) and observe nice agreement. Using this solution in Eq. (12) we obtain C(x, z, τ) in Eq. (10) In the NESS the equations (6)-(8) become simpler since which we also compare with simulation results in the in- ∂ 0 as τ implying (0, y) = d. Now making τ ss set of Fig. 2 and again observe good agreement. theT variable → transformation→ ∞ zC= (1 y), the problem of From the numerical solution of Eq. (19) we, in addi- finding reduces to solving a diffusion− equation with its ss tion, observe that the eigenvalues obtained , are in gen- value atCx = 0 held fixed for all y. It is easy to show that p eral complex and are of the form λ nπ 3/2(1 isgn), the solution is given by (x, y) = d erfc(x/ 4γ(1 y)) n ss for large n > 0, while at small values≈ of | n| the eigenval-− where erfc(v) is the complementaryC error function− [27]. ues deviate from this behaviour [27]. We confirm that Now inserting this solution in Eq. (7) and solving with this is not due to the the truncation of the matrix but boundary conditions (0) = T and (1) = T , we ss ` ss r an artefact of the finiteness of the system. Note that the get the explicit expressionT Eq. (11). InT Fig. 1 we verify large n behavior of λ is similar to the Fourier spectrum the analytical results for T and numerically, where n ss ss of the non-local operator in Eq. (3) describing the we observe nice agreement. OneC can easily identify the L evolution in infinite system.∞ Hence it is interesting to see microscopic current from the equation for T˙ in Eq. (15) i if one recovers the evolution Eq. (3) in the infinite sys- as j = 2C γ(T T ) which in the steady state − i,i+1 − i+1 − i tem limit. One can follow the same calculation procedure for large provides √ . Note that the L jss = 2 ss(0, y)/ L on an infinite lattice as presented before for finite lattice term contributes− C at . Now inserting γ(Ti+1 Ti) (1/L) and arrive at an evolution equation for temperature pro- the expression− − of from Eq.O (11), one obtains the ss(0, y) file similar to Eq. (13) with only difference being that the expression for Cgiven in Eq. (4). jss lower limit of the integral on the rhs is , as the equa- We now focus on the relaxation to the NESS. It is tion is now valid in y . Now−∞ taking Fourier often convenient to separate the relaxation part as done transform on both sides−∞ it ≤ is quite≤ ∞ easy to show that the in Eq. (9) and Eq. (10) where r(z, τ) and r(x, z, τ) evolution equation in infinite space indeed reduces to the describes the approach towardsT the NESS solutionsC in skew-fractional equation in Eq. (3). A more direct and Eq. (11). It is easy to see that (x, z, τ) and (z, τ) detailed proof is given in [27]. Cr Tr 5
Conclusion: In this Letter, we have studied anoma- 112(4), 2014. lous transport in a one-dimensional system with two con- [11] Suman G. Das, Abhishek Dhar, Keiji Saito, Christian B. served quantities, in the open system setup. Starting Mendl, and Herbert Spohn. Numerical test of hydrody- from a microscopic description and acquiring knowledge namic fluctuation theory in the Fermi-Pasta-Ulam chain. Physical Review E - Statistical, Nonlinear, and Soft Mat- about scaling properties from numerical studies, we de- ter Physics, 90(1), 2014. rive exact expressions of the temperature profiles and the [12] Cédric Bernardin, Patrícia Gonçalves, Milton Jara, two point correlations in the steady state. We also study Makiko Sasada, and Marielle Simon. From Normal Dif- the evolution of these quantities towards steady state. fusion to Superdiffusion of Energy in the Evanescent Flip We explicitly show that the evolution of the temperature Noise Limit. Journal of Statistical Physics, 159(6):1327– profiles in this model is governed by a non-local operator 1368, 2015. defined inside a finite domain which correctly takes the [13] Cédric Bernardin and Gabriel Stoltz. Anomalous diffu- sion for a class of systems with two conserved quantities. previously obtained infinite system representation. We Nonlinearity, 25(4):1099–1133, 2012. provide numerical verifications of the analytical results. [14] Milton Jara, Tomasz Komorowski, and Stefano Olla. Su- Our work provides the first clear and transparent micro- perdiffusion of Energy in a Chain of Harmonic Oscillators scopic derivation of non-local heat equation describing with Noise. Communications in Mathematical Physics, anomalous transport in finite geometry and its connec- 339(2):407–453, 2015. tion to the corresponding skew-fractional equation in the [15] Herbert Spohn. Fluctuating hydrodynamics approach to infinite domain. equilibrium time correlations for anharmonic chains. Lec- ture Notes in Physics, 921:107–158, 2016. We thank C. Bernardin, C. Mejía-Monasterio, S. [16] Christian B. Mendl and Herbert Spohn. Current fluc- Olla and S. Lepri for very useful discussions. AK ac- tuations for anharmonic chains in thermal equilibrium. knowledges support from DST grant under project No. Journal of Statistical Mechanics: Theory and Experi- ECR/2017/000634. ment, 2015(3), 2015. [17] Giada Basile, Cédric Bernardin, and Stefano Olla. Mo- mentum conserving model with anomalous thermal con- ductivity in low dimensional systems. Physical Review Letters, 96(20), 2006. ∗ [email protected] [18] Cédric Bernardin, Patrícia Gonçalves, and Milton Jara. † [email protected] 3/4-Fractional Superdiffusion in a System of Harmonic ‡ [email protected] Oscillators Perturbed by a Conservative Noise. Archive § [email protected] for Rational Mechanics and Analysis, 220(2):505–542, [1] Onuttom Narayan and Sriram Ramaswamy. Anoma- 2016. lous Heat Conduction in One-Dimensional Momentum- [19] . Fractional Laplacian. Conserving Systems. Physical Review Letters, 89(20), [20] . Fractional Heat Equation. 2002. [21] G. M. Viswanathan, V. Afanasyev, Sergey V. Buldyrev, [2] Henk Van Beijeren. Exact results for anomalous trans- Shlomo Havlin, M. G.E. Da Luz, E. P. Raposo, and port in one-dimensional hamiltonian systems. Physical H. Eugene Stanley. Levy flights in random searches. Review Letters, 108(18), 2012. Physica A: Statistical Mechanics and its Applications, [3] Herbert Spohn. Nonlinear Fluctuating Hydrodynamics 282(1):1–12, 2000. for Anharmonic Chains. Journal of Statistical Physics, [22] A. Zoia, A. Rosso, and M. Kardar. Fractional Laplacian 154(5):1191–1227, feb 2014. in bounded domains. Physical Review E - Statistical, [4] Victor Lee, Chi Hsun Wu, Zong Xing Lou, Wei Li Lee, Nonlinear, and Soft Matter Physics, 76(2), 2007. and Chih Wei Chang. Divergent and Ultrahigh Ther- [23] S. Lepri, C. Mejía-Monasterio, and A. Politi. A stochastic mal Conductivity in Millimeter-Long Nanotubes. Physi- model of anomalous heat transport: Analytical solution cal Review Letters, 118(13), 2017. of the steady state. Journal of Physics A: Mathematical [5] Stefano Lepri, Roberto Livi, and Antonio Politi. Univer- and Theoretical, 42(2), 2009. sality of anomalous one-dimensional heat conductivity. [24] Luca Delfini, Stefano Lepri, Roberto Livi, and Anto- Physical Review E - Statistical Physics, Plasmas, Fluids, nio Politi. Nonequilibrium invariant measure under heat and Related Interdisciplinary Topics, 68(6), 2003. flow. Physical Review Letters, 101(12), 2008. [6] Abhishek Dhar. Heat transport in low-dimensional sys- [25] J. Cividini, A. Kundu, A. Miron, and D. Mukamel. Tem- tems. Advances in Physics, 57(5):457–537, 2008. perature profile and boundary conditions in an anoma- [7] Stefano Lepri, Roberto Livi, and Antonio Politi. Heat lous heat transport model. Journal of Statistical Mechan- transport in low dimensions: Introduction and phe- ics: Theory and Experiment, 2017(1), 2017. nomenology. Lecture Notes in Physics, 921:1–37, 2016. [26] Stefano Lepri, Carlos Mejía-Monasterio, and Antonio [8] Abhishek Dhar. Heat conduction in a one-dimensional Politi. Nonequilibrium dynamics of a stochastic model of gas of elastically colliding particles of unequal masses. anomalous heat transport. Journal of Physics A: Math- Physical Review Letters, 86(16):3554–3557, 2001. ematical and Theoretical, 43(6), 2010. [9] S. Lepri, R. Livi, and A. Politi. On the anomalous ther- [27] Supplementary Material . mal conductivity of one-dimensional lattices. Europhysics [28] Stefano Lepri and Antonio Politi. Density profiles in open Letters, 43(3):271–276, 1998. superdiffusive systems. Physical Review E - Statistical, [10] Sha Liu, Peter Hänggi, Nianbei Li, Jie Ren, and Baowen Nonlinear, and Soft Matter Physics, 83(3), 2011. Li. Anomalous heat diffusion. Physical Review Letters, 1
Supplementary Material: Anomalous heat equation in a system connected to thermal reservoirs Priyanka1, Aritra Kundu1, Abhishek Dhar1, Anupam Kundu1 1International Centre for Theoretical Sciences, Tata Institute of Fundamental Research, Bengaluru 560089, India
Abstract We here provide the details of calculations for the results presented in the main text.
DYNAMICAL OPERATORS AND DISCRETE EQUATIONS
The explicit expression of the deterministic part of the Fokker Planck (FP) equation given by operators l, b, defined as L L
L 1 X− l = (V 0(ηi+1) V 0(ηi 1))∂ηi V 0(ηL 1)∂ηL + V 0(η2)∂η L − − − − 1 i=1 2 2 = λT ∂ λ∂ V 0(η ) + λT ∂ λ∂ V 0(η ), Lb ` η1 − η1 1 r ηL − ηL L where T` and Tr are the temperatures of the reservoirs on the left and right, respectively. The stochastic part ex is given as L
L 1 ! X− = γ P (~η ) P (~η) (S1) Lex i,i+1 − i=1 where ~ηi,i+1 denote the configuration after the exchange of variable i with i + 1. The two point function defined as, 2 Ci,j = ηiηj . From the above FP equation for V (η) = η /2, the dynamical equation Ci,j = ηiηj s can easily be writtenh as i h i
C˙ ij = Ci+1,j Ci 1,j + Ci,j+1 Ci,j 1 + γ[Ci 1,j + Ci+1,j + Ci,j 1 + Ci,j+1 4Ci,j], 1 < i, j < L (S2) − − − − − − − C˙ i,i+1 = Ti+1 Ci 1,i+1 + Ci,i+2 Ti + γ[Ci 1,i+1 + Ci,i+2 2Ci,i+1], 1 < i < L (S3) − − − − − T˙i = 2[Ci,i+1 Ci 1,i] + γ[Ti+1 + Ti 1 2Ti], 1 i L (S4) − − − − ≤ ≤ The dynamical equations at the boundaries are given by
1. for i = j = 1
T˙ = 2λT + 2C 2λT + γ[T T ] (S5) 1 ` 1,2 − 1 2 − 1 2. for i = j = L
T˙L = 2λTr 2CL 1,L 2λTL + γ[TL 1 TL] (S6) − − − − − 3.i = 1 and 1 < j < L
C˙ 1,j = C2,j λC1,j + C1,j+1 C1,j 1 + γ[C1,j 1 + C1,j+1 + C2,j 3C1,j] (S7) − − − − − 4.j = L and 1 < i < L
C˙ i,L = Ci+1,L Ci 1,L Ci,L 1 λCi,L + γ[Ci 1,L + Ci+1,L + Ci,L 1 3Ci,L] (S8) − − − − − − − − 5.i = 1 and j = L
C˙ 1,L = C2,L C1,L 1 2λC1,L + γ[C2,L + C1,L 1 2C1,L] (S9) − − − − − blue,orange,green,red=1000,2000,3000,4000 2
y = k/L y = k/L
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0 -0.002 0 k k