betaFeedback

WorksPhysics and Astronomy

The normal approximation to macroscopic dynamical fluctuations

Abstract

Under local detailed balance, the Lagrangian of macroscopic dynamical fluctuations has a canonical structure: its time-antisymmetric part is the entropy flux, and its time-symmetric part is a Legendre pair of kinetic and dissipation functions. We show what survives of that structure for normal (Gaussian) fluctuations around a nonequilibrium steady state. The quadratic Lagrangian keeps the canonical form, with the Hessian of the dissipation function as noise matrix. However, the force on the fluctuations differs from the linearized thermodynamic force by a term proportional to the driving and to the state dependence of the dynamical activity, so that local detailed balance is violated at the Gaussian level. Nevertheless, the Gaussian dynamics is gradient flow for the quadratic part of the nonequilibrium entropy, around an emergent nondissipative current. Finally, we show which parts of the Lagrangian can be reconstructed from measured Gaussian fluctuations.

Introduction

Macroscopic equations describe typical behavior; fluctuations around them carry the information about how that behavior comes about. Since Onsager [1, 2], the linear (Gaussian) processes around equilibrium have been the main bridge between fluctuations and response: their drift is fixed by the thermodynamic force, their noise by the Onsager matrix, and time-reversal symmetry produces reciprocity and the fluctuation–dissipation theorem.

Away from equilibrium, the large deviations of macroscopic trajectories are known to have a canonical structure as soon as local detailed balance holds [3]. That is very similar to and inspired by nonlinear extensions of the relation between fluctuations and gradient flow. using the (global) detailed balance condition [1, 4, 5, 6]. The time-antisymmetric part of the path-space Lagrangian can now be identified with the (irreversible) entropy flux in terms of a thermodynamic force. Everything else is encoded in a Legendre pair of time-symmetric functions, the kinetic function ψ\psi and the dissipation function ψ∗\psi^{*}. The zero-cost path gives the macroscopic relaxation equation toward nonequilibrium, and its structure extends those of gradient flows and GENERIC.

However in practice, what is most often measured or simulated are the small, normal fluctuations around the macroscopic evolution: linear-noise approximations of chemical kinetics [7, 8], fluctuating hydrodynamics and Gaussian current statistics. The present paper asks what of the canonical structure survives at that level, and what the Gaussian level can tell about the full structure as an inverse problem. The main results are:

  1. The quadratic Lagrangian of normal fluctuations around a nonequilibrium steady state keeps the canonical form (Sec. IV). Its noise matrix Γ\Gamma is the Hessian of the dissipation function ψ∗\psi^{*} at the stationary force; its drift defines a linear force F(v)\mathcal{F}(v) on fluctuations.

  1. That force is not the linearized thermodynamic force. The difference is first order in the stationary driving, coming from the state dependence of the dynamical activity. Local detailed balance is thus violated at the Gaussian level (§ 5). It is restored with respect to the quadratic part of the nonequilibrium entropy, together with what we can call an emergent nondissipative current.

  1. The associated Ornstein–Uhlenbeck dynamics for states and integrated currents (Sec. VI) allows one to reconstruct the quadratic Lagrangian from data. Current noise measures the frenetic Hessian at the operating point, information that the relaxation equation alone does not contain.

  1. Linear response is expressed through the Gaussian correlations (Sec. VIII C). Violations of the fluctuation–dissipation and Green–Kubo relations are governed by the nondissipative current. Details are postponed to a future publication.

Two families of examples run through the text: chemical reaction networks with one free species, including the Schlögl model, and driven diffusion on a ring (Sec. III). Section VII gives connections to and distinctions from previous works. Section VIII discusses multistability and normal fluctuations around limit cycles. Detailed computations are in the Appendices.

Canonical structure of dynamical fluctuations

We recall the main results of [3, 9]. Let zz denote the macroscopic state (a density profile, a vector of concentrations, …\ldots) and jj a current, related by

z˙=Dj(1)\dot{z}=\mathcal{D}j \tag*{(1)}

with D\mathcal{D} a linear operator (minus a divergence, a stoichiometric matrix, the identity). For a system with large parameter NN (number of particles, volume), we assume that the paths of states and currents satisfy a large deviation principle,

Prob⁡[(z(s),j(s))0≤s≤t]≍e−NI0(z(0))exp⁡[−N∫0tds L(j(s),z(s))],(2)\operatorname{Prob}\left[(z(s),j(s))_{0\leq s\leq t}\right]\asymp e^{-NI_0(z(0))}\exp\left[-N\int_0^t\mathrm{d}s\,\mathcal{L}(j(s),z(s))\right], \tag*{(2)}

with L(j,z)≥0\mathcal{L}(j,z)\geq0 convex in jj.

Spaces and pairings. At each state zz (element of a differentiable manifold) there are two pairs of dual spaces: the tangent space (velocities z˙\dot{z}) with its dual, the cotangent space; and the space of currents jj with its dual, the space of forces ff. The adjoint D†\mathcal{D}^{\dagger} maps cotangent vectors to forces, (a,Dj)=(D†a,j)(a,\mathcal{D}j)=(\mathcal{D}^{\dagger}a,j). In a basis, currents and forces have components jbj_b, fbf_b, where bb labels the transport channels: the reactions of a network, the bonds of a graph, or positions xx in a continuum (where sums become integrals); tangent vectors have components aca_c, with cc labelling species or positions. The entropy flux is the pairing (F,j)=∑bFbjb(F,j)=\sum_b F_bj_b. We set kB=1k_B=1, so entropies are dimensionless. Round brackets (⋅,⋅)(\cdot,\cdot) always denote such pairings (except when delimiting the argument of a function such as L\mathcal{L}); angular brackets ⟨⋅⟩\langle\cdot\rangle denote expectations.

Local detailed balance. There are a thermodynamic force F(z)F(z) and a Hamiltonian flow jH(z)j^H(z) with (F,jH)=0(F,j^H)=0 such that for all z˙=Dj\dot{z}=\mathcal{D}j,

L(jH(z)−j,z)−L(jH(z)+j,z)=(j,F(z)).(3)\mathcal{L}(j^H(z)-j,z)-\mathcal{L}(j^H(z)+j,z)=(j,F(z)). \tag*{(3)}

Remark: The force may contain a gradient part D†dS\mathcal{D}^{\dagger}\mathrm{d}S and a non-gradient (rotational, cycle) part GG,

F(z)=D†dS(z)+G(z),(4)F(z)=\mathcal{D}^{\dagger}\mathrm{d}S(z)+G(z), \tag*{(4)}

with SS an equilibrium entropy; then G≠0G\ne0 is the nonequilibrium driving.

Canonical form. The time-symmetric part of L\mathcal{L} is free but given by a Legendre pair (ψ,ψ∗)(\psi,\psi^*) of convex functions, even in their first argument and vanishing at zero [3]:

L(j,z)=ψ(j−JH(z),z)+ψ∗(F(z)2,z)−(j−JH(z),F(z)2).(5)\mathcal{L}(j,z)=\psi\left(j-J^{H}(z),z\right)+\psi^*\left(\frac{F(z)}{2},z\right)-\left(j-J^{H}(z),\frac{F(z)}{2}\right). \tag*{(5)}

The pair may also depend on the driving GG; we suppress that dependence in the notation. We call ψ∗\psi^* the dissipation function: it determines how forces are turned into currents.

Relaxation equation. The zero-cost path L(jz,z)=0\mathcal{L}(j_z,z)=0 gives the typical current and the macroscopic evolution,

jz=JH(z)+∂fψ∗(F(z)2,z),z˙=Djz.(6)j_z=J^{H}(z)+\partial_f\psi^*\left(\frac{F(z)}{2},z\right),\qquad\dot{z}=\mathcal{D}j_z. \tag*{(6)}

Equilibrium corresponds to G=0G=0; the entropy SS then increases along (II.6). Away from equilibrium there is another nonequilibrium entropy S(z)S(z), the static fluctuation functional of the stationary state (aka quasipotential), which is a Lyapunov function [3].

Assumptions. We assume that (II.6) has a stable stationary point zsz^s (Sec. VIII relaxes this), with stationary current js=jzsj^s=j_{z^s}, Djs=0\mathcal{D}j^s=0, and that L\mathcal{L} is twice continuously differentiable near (js,zs)(j^s,z^s). For simplicity of notation we take JH=0J^{H}=0 from now on, as in both running examples; Remark IV.2 explains the changes for JH≠0J^{H}\ne0.

Running examples

Chemical networks with one free species

Consider a well-mixed volume NN with one free species ZZ (concentration zz) and two reactions i=1,2i=1,2, each producing one ZZ in its forward direction, so that D=(1,1)\mathcal{D}=(1,1). All other species are chemostatted. With mass-action rates r±i(z)r_{\pm i}(z),

Fi(z)=log⁡ri(z)r−i(z),(7)F_i(z)=\log\frac{r_i(z)}{r_{-i}(z)}, \tag*{(7)}
ψ∗(f,z)=∑i=1,22ai(z)(cosh⁡fi−1),ai=rir−i,(8)\psi^*(f,z)=\sum_{i=1,2}2a_i(z)\left(\cosh f_i-1\right),\qquad a_i=\sqrt{r_i r_{-i}}, \tag*{(8)}

so that ji=2aisinh⁡(Fi/2)=ri−r−ij_i=2a_i\sinh(F_i/2)=r_i-r_{-i} and z˙=j1+j2\dot{z}=j_1+j_2. The prefactor aia_i is the frenetic part of the channel; the corresponding traffic is

ti=ri+r−i=2aicosh⁡(Fi/2).(9)t_i=r_i+r_{-i}=2a_i\cosh(F_i/2). \tag*{(9)}

The cycle (1,−1)T∈ker⁡D(1,-1)^{\mathsf{T}}\in\ker\mathcal{D} (reaction 1 forward, reaction 2 backward) has affinity F1−F2F_1-F_2. At a stationary state, J=j1s=−j2sJ=j_1^s=-j_2^s is the stationary cycle current.

Catalytic network. As a first example [10], take

A1+Z ⇌k+1k−1 B1+2Z,A2 ⇌k+2k−2 B2+Z,(10)A_1+Z\ \underset{k_{-1}}{\stackrel{k_{+1}}{\rightleftharpoons}}\ B_1+2Z,\qquad A_2\ \underset{k_{-2}}{\stackrel{k_{+2}}{\rightleftharpoons}}\ B_2+Z, \tag*{(10)}

with chemostat concentrations α1,2,β1,2\alpha_{1,2},\beta_{1,2}. In units where k+1α1=k+2α2=1k_{+1}\alpha_1=k_{+2}\alpha_2=1, and with

κ1=k−1β1,κ2=k−2β2,\kappa_1=k_{-1}\beta_1,\qquad\kappa_2=k_{-2}\beta_2,
r1=z,r−1=κ1z2,r2=1,r−2=κ2z.(11)r_1=z,\qquad r_{-1}=\kappa_1z^2,\qquad r_2=1,\qquad r_{-2}=\kappa_2z. \tag*{(11)}

The cycle affinity F1−F2=log⁡(κ2/κ1)F_1-F_2=\log(\kappa_2/\kappa_1) does not depend on zz; the network is in equilibrium if and only if κ1=κ2\kappa_1=\kappa_2. The stationary concentration is

zs=1−κ2+Δ2κ1,Δ=(1−κ2)2+4κ1,(12)z^s=\frac{1-\kappa_2+\Delta}{2\kappa_1},\qquad\Delta=\sqrt{(1-\kappa_2)^2+4\kappa_1}, \tag*{(12)}

with cycle current J=κ2zs−1J=\kappa_2z^s-1.

Schlögl model. The second example is Schlögl’s model [11],

A+2Z ⇌k1k−1 3Z,B ⇌k2k−2 Z,(13)A+2Z\ \underset{k_{-1}}{\stackrel{k_1}{\rightleftharpoons}}\ 3Z,\qquad B\ \underset{k_{-2}}{\stackrel{k_2}{\rightleftharpoons}}\ Z, \tag*{(13)}

with r1=k1αz2r_1=k_1\alpha z^2, r−1=k−1z3r_{-1}=k_{-1}z^3, r2=k2βr_2=k_2\beta, r−2=k−2zr_{-2}=k_{-2}z. The cycle converts A⇌BA\rightleftharpoons B, with affinity log⁡(k1k−2α/k−1k2β)\log(k_1k_{-2}\alpha/k_{-1}k_2\beta). The cubic rate equation is bistable in a range of chemostat concentrations, which requires a nonzero cycle affinity (Sec. VIII A).

Driven diffusion on a ring

Let ρ(x)\rho(x), x∈[0,L)x \in[0,L) with periodic boundary conditions, be a density with local entropy ∫s(ρ(x)) dx\int s(\rho(x))\,\mathrm{d}x (ss concave), driven by a constant field EE along the ring. Then

D=−∂x,D†=∂x,F(ρ)=∂xs′(ρ)+E,(14)\mathcal{D}=-\partial_x,\qquad\mathcal{D}^{\dagger}=\partial_x,\qquad F(\rho)=\partial_xs'(\rho)+E, \tag*{(14)}
ψ∗(f,ρ)=∫χ(ρ(x))f(x)2 dx,(15)\psi^*(f,\rho)=\int\chi(\rho(x))f(x)^2\,\mathrm{d}x, \tag*{(15)}

with mobility χ(ρ)≥0\chi(\rho)\geq0. The current is j=χ(ρ)E−D(ρ)∂xρj=\chi(\rho)E-D(\rho)\partial_x\rho with diffusivity D=−χs′′D=-\chi s'' (Einstein relation). For the weakly asymmetric exclusion process, χ=ρ(1−ρ)\chi=\rho(1-\rho), s=−ρlog⁡ρ−(1−ρ)log⁡(1−ρ)s=-\rho\log\rho-(1-\rho)\log(1-\rho) and D=1D=1, and the hydrodynamic equation is the viscous Burgers equation. The uniform profile ρˉ\bar{\rho} is stationary, with current js=χ(ρˉ)Ej^s=\chi(\bar{\rho})E around the ring. The non-gradient part of the force is the field: ∫E dx=EL≠0\int E\,\mathrm{d}x=EL\ne0.

Such examples, weakly driven or boundary-driven diffusions are the main models of the macroscopic fluctuation theory developed by [12, 13, 14], as limits of stochastic interacting particle systems.

A nonlocal mobility (kernel) χ(x,y;ρ)\chi(x,y;\rho) and nonlocal entropy s(ρ)s(\rho) fit the same framework. The formulas below then extend with χ\chi read as an operator and s(ρ)s(\rho) as a general functional of ρ\rho.

The quadratic Lagrangian

Scaling and definition

Normal fluctuations live on the scale N−1/2N^{-1/2}. We fix a stable stationary point (zs,js)(z^s,j^s) of (II.6). Writing

z=zs+vN,j=js+ιN,(16)z=z^s+\frac{v}{\sqrt{N}},\qquad j=j^s+\frac{\iota}{\sqrt{N}}, \tag*{(16)}

the rescaled state fluctuation vv and current fluctuation ι\iota satisfy v˙=Dι\dot{v}=\mathcal{D}\iota, and their path probabilities are governed by the quadratic Lagrangian

L2(ι,v)=lim⁡N→∞NL(js+ιN,zs+vN),(17)L_2(\iota,v)=\lim_{N\to\infty}N\mathcal{L}\left(j^s+\frac{\iota}{\sqrt{N}},z^s+\frac{v}{\sqrt{N}}\right), \tag*{(17)}

in the sense that, as N→∞N\to\infty,

Prob⁡[(v(s),ι(s))0≤s≤t]∝exp⁡[−∫0tds L2(ι(s),v(s))].(18)\operatorname{Prob}\left[(v(s),\iota(s))_{0\leq s\leq t}\right]\propto\exp\left[-\int_0^t\mathrm{d}s\,L_2(\iota(s),v(s))\right]. \tag*{(18)}

The passage from the large deviation principle (II.2) to this central-limit statement is not automatic; it holds under smoothness of the cumulant generating functions [15], and is a theorem for density-dependent Markov jump processes [8] and their linear-noise approximation [7]. We assume it throughout.

Main formula

Introduce the stationary force Fs=F(zs)F^{s}=F(z^{s}) and three linear objects, all evaluated at the operation point (Fs,zs)(F^{s},z^{s}):

Γ=∂f2ψ∗(Fs/2,zs),(noise matrix)(19)\Gamma=\partial_{f}^{2}\psi^{*}(F^{s}/2,z^{s}), \qquad\text{(noise matrix)} \tag*{(19)}
Θv=(∂z∂fψ∗(Fs/2,zs),v),(frenetic sensitivity)(20)\Theta v=(\partial_{z}\partial_{f}\psi^{*}(F^{s}/2,z^{s}),v), \qquad\text{(frenetic sensitivity)} \tag*{(20)}
Λv=ddεjs+εv∣ε=0=12F′v+Θv,(current response)(21)\Lambda v=\left.\frac{\mathrm{d}}{\mathrm{d}\varepsilon}j^{s+\varepsilon v}\right|_{\varepsilon=0}=\frac{1}{2}F'v+\Theta v, \qquad\text{(current response)} \tag*{(21)}

where F′=∂zF(zs)F'=\partial_{z}F(z^{s}). Γ\Gamma is symmetric and strictly positive definite (by the assumed stability) Hessian of the dissipation function on the space of forces, evaluated at steady thermodynamic force F(zs)/2F(z^{s})/2. The outer round brackets in (IV.5) denote the pairing of the tangent space and its dual, where linearization allows us to identify state perturbations vv with tangent vectors at the stationary state zsz^{s}. Λ\Lambda maps state fluctuations to current fluctuations. Define the force acting on state fluctuations

F(v)=2Γ−1Λv=F′v+2Γ−1Θv.(22)\mathcal{F}(v)=2\Gamma^{-1}\Lambda v=F'v+2\Gamma^{-1}\Theta v. \tag*{(22)}

Proposition IV.1. The quadratic Lagrangian is

L2(ι,v)=12(ι−Λv,Γ−1(ι−Λv)),(23)L_{2}(\iota,v)=\frac{1}{2}(\iota-\Lambda v,\Gamma^{-1}(\iota-\Lambda v)), \tag*{(23)}

and has the canonical form

L2(ι,v)=ψ2(ι)+ψ2∗(F(v)2)−(ι,F(v)2),(24)L_{2}(\iota,v)=\psi_{2}(\iota)+\psi_{2}^{*}\left(\frac{\mathcal{F}(v)}{2}\right)-\left(\iota,\frac{\mathcal{F}(v)}{2}\right), \tag*{(24)}

with the quadratic Legendre pair

ψ2(ι)=12(ι,Γ−1ι),ψ2∗(f)=12(f,Γf).(25)\psi_{2}(\iota)=\frac{1}{2}(\iota,\Gamma^{-1}\iota), \qquad\psi_{2}^{*}(f)=\frac{1}{2}(f,\Gamma f). \tag*{(25)}

The proof (Appendix A) uses only that L(jz,z)=0\mathcal{L}(j_{z},z)=0 and ∂jL(jz,z)=0\partial_{j}\mathcal{L}(j_{z},z)=0 for all zz, together with ∂j2L(jz,z)=∂j2ψ=(∂f2ψ∗)−1\partial_{j}^{2}\mathcal{L}(j_{z},z)=\partial_{j}^{2}\psi=(\partial_{f}^{2}\psi^{*})^{-1} by Legendre duality.

The force F(v)\mathcal{F}(v) in (IV.7) has two parts. The first, F′v\mathcal{F}'v, is the variation of the thermodynamic force. The second, 2Γ−1Θv2\Gamma^{-1}\Theta v, comes from the sensitivity of the dissipation function on the state. Since ψ∗\psi^* is even in ff, ∂fψ∗(0,z)=0\partial_f\psi^*(0,z)=0 for all zz, we have that

Fs=0⟹Θ=0,F(v)=F′v.(26)F^s=0 \quad\Longrightarrow\quad\Theta=0,\qquad\mathcal{F}(v)=\mathcal{F}'v. \tag*{(26)}

Therefore, the second part of F\mathcal{F} is a genuinely nonequilibrium term.

Two classes of dissipation functions. (i) For quadratic ψ∗(f,z)=(f,χ(z)f)\psi^*(f,z)=(f,\chi(z)f), one has Jz=χFJ_z=\chi F, Γ=2χ\Gamma=2\chi and F(v)=F′v+χ−1(∂zχ,v)Fs\mathcal{F}(v)=\mathcal{F}'v+\chi^{-1}(\partial_z\chi,v)F^s. (ii) For channel-wise products ψ∗(f,z)=∑iai(z)φ(fi)\psi^*(f,z)=\sum_i a_i(z)\varphi(f_i), as in (III.2), with φ\varphi even and convex,

Fi(v)=Fi′v+2(∂zlog⁡ai,v)φ′(Fis/2)φ′′(Fis/2).(27)\mathcal{F}_i(v)=\mathcal{F}'_iv+2(\partial_z\log a_i,v)\frac{\varphi'(F_i^s/2)}{\varphi''(F_i^s/2)}. \tag*{(27)}

which for φ(f)=2(cosh⁡f−1)\varphi(f)=2(\cosh f-1) gives the factor tanh⁡(Fi/2)=ji/ti\tanh(F_i/2)=j_i/t_i, the ratio of current to traffic [16, 17].

Examples

One-species networks. See Example III A. Here Γ=diag⁡(t1,t2)\Gamma=\operatorname{diag}(t_1,t_2) and Λ=(λ1,λ2)T\Lambda=(\lambda_1,\lambda_2)^{\mathsf{T}} with λi=ri′−r−i′\lambda_i=r'_i-r'_{-i} (derivatives at zsz^s). Then Fi=2λi/ti\mathcal{F}_i=2\lambda_i/t_i per unit vv, and by (IV.12),

Fi=Fi′+2tanh⁡Fis2(log⁡ai)′=Fi′+(ri′ri+r−i′r−i)jiti.(28)\mathcal{F}_i=\mathcal{F}'_i+2\tanh\frac{F_i^s}{2}(\log a_i)'=\mathcal{F}'_i+\left(\frac{r'_i}{r_i}+\frac{r'_{-i}}{r_{-i}}\right)\frac{j_i}{t_i}. \tag*{(28)}

For the catalytic network, F1′=F2′=−1/zs\mathcal{F}'_1=\mathcal{F}'_2=-1/z^s, (log⁡a1)′=3/(2zs)(\log a_1)'=3/(2z^s) and (log⁡a2)′=1/(2zs)(\log a_2)'=1/(2z^s), so that

F1=−1+3J/t1zs,F2=−1−J/t2zs(29)\mathcal{F}_1=\frac{-1+3J/t_1}{z^s},\qquad\mathcal{F}_2=\frac{-1-J/t_2}{z^s} \tag*{(29)}

in terms of the cycle current JJ.

Ring. See Example III B. At the uniform state, Γ=2χ(ρˉ)\Gamma=2\chi(\bar{\rho}) and

Λv=cv−D∂xv,c=χ′(ρˉ)E,(30)\Lambda v=cv-D\partial_xv,\qquad c=\chi'(\bar{\rho})E, \tag*{(30)}

with D=−χs′′D=-\chi s'' at ρˉ\bar{\rho}, and

F(v)=s′′(ρˉ)∂xv+cχ(ρˉ)v.(31)\mathcal{F}(v)=s''(\bar{\rho})\partial_xv+\frac{c}{\chi(\bar{\rho})}v. \tag*{(31)}

The first term is F′v\mathcal{F}'v; the second is the “correction.” For exclusion, c=E(1−2ρˉ)c = E(1 - 2\bar{\rho}) is the drift velocity in the linearized dynamics.

Remark IV.2. With a Hamiltonian flow JH(z)J^{H}(z), Λ\Lambda and consequently F(v)\mathcal{F}(v) remain unchanged, and both formulas (IV.8) as well as (IV.9) hold with ι\iota replaced by ι−(∂zJH,v)\iota- (\partial_z J^{H},v).

Local detailed balance at the Gaussian level

Formal structure and its mismatch

Reversing the current fluctuation at fixed vv in (IV.9) gives

L2(−ι,v)−L2(ι,v)=(ι,F(v)).(32)L_2(-\iota,v)-L_2(\iota,v)=(\iota,\mathcal{F}(v)). \tag*{(32)}

The quadratic Lagrangian therefore has the form of local detailed balance, with F(v)\mathcal{F}(v) in the role of the force. Local detailed balance proper would require that to be the entropy flux of the fluctuation. However, by (IV.7),

F(v)−F′v=2Γ−1Θv,(33)\mathcal{F}(v)-\mathcal{F}'v=2\Gamma^{-1}\Theta v, \tag*{(33)}

which vanishes only at equilibrium. It also vanishes if the dissipation function does not depend on the state at the operating point, Θ=0\Theta=0. In conclusion, the antisymmetric part of L2L_2 need not be the physical entropy flux.

The origin of the mismatch is the time reversal itself. Physical time reversal maps js+ι/Nj^s+\iota/\sqrt{N} to −js−ι/N-j^s-\iota/\sqrt{N}. When js≠0j^s\ne0, the reversed path lies outside the Gaussian window:

L(−js,zs)−L(js,zs)=(js,F(zs))>0\mathcal{L}(-j^s,z^s)-\mathcal{L}(j^s,z^s)=(j^s,\mathcal{F}(z^s))>0

is the stationary entropy production, and its probability is exponentially small in NN. What remains at the Gaussian level is the reversal ι↦−ι\iota\mapsto-\iota around the stationary current, which is not a symmetry of the physics. The antisymmetric part (V.1) then also picks up how the current responds kinetically to the state, through Θ\Theta.

Gaussian entropy and nondissipative current

Local detailed balance can be restored at the Gaussian level, but with respect to new objects. Let C=⟨vvT⟩C=\langle vv^{\mathsf{T}}\rangle be the stationary covariance of vv (Sec. VI), which solves the Lyapunov equation (for derivation see Appendix B)

KC+CK†+B=0,K=DΛ,B=DΓD†.(34)KC + CK^{\dagger} + B = 0,\qquad K = \mathcal{D}\Lambda,\qquad B = \mathcal{D}\Gamma\mathcal{D}^{\dagger}. \tag*{(34)}

Define the Gaussian entropy S2(v)=−12(v,C−1v)S_2(v) = -\frac{1}{2}(v,C^{-1}v) and

F0(v)=D†∇S2(v)=−D†C−1v,Jnd(v)=Λv−12ΓF0(v).(35)\mathcal{F}_0(v) = \mathcal{D}^{\dagger}\nabla S_2(v) = -\mathcal{D}^{\dagger}C^{-1}v,\qquad J^{\mathrm{nd}}(v) = \Lambda v - \frac{1}{2}\Gamma\mathcal{F}_0(v). \tag*{(35)}

Proposition V.1. (a) L2(ι,v)=ψ2(ι−Jnd(v))+ψ2∗(F0(v)/2)−(ι−Jnd(v),F0(v)/2)L_2(\iota,v) = \psi_2(\iota-J^{\mathrm{nd}}(v))+\psi_2^*(\mathcal{F}_0(v)/2)-(\iota-J^{\mathrm{nd}}(v),\mathcal{F}_0(v)/2), so that

L2(Jnd(v)−ι,v)−L2(Jnd(v)+ι,v)=(ι,F0(v)).(36)L_2(J^{\mathrm{nd}}(v)-\iota,v)-L_2(J^{\mathrm{nd}}(v)+\iota,v)=(\iota,\mathcal{F}_0(v)). \tag*{(36)}

(b) JndJ^{\mathrm{nd}} is nondissipative: (Jnd(v),F0(v))=0(J^{\mathrm{nd}}(v),\mathcal{F}_0(v))=0 for all vv.

(c) DJnd(v)=−QC−1v\mathcal{D}J^{\mathrm{nd}}(v)=-QC^{-1}v with Q=−KC−12BQ=-KC-\frac{1}{2}B antisymmetric. In particular S2S_2 is conserved by the flow DJnd\mathcal{D}J^{\mathrm{nd}}.

The proof is in Appendix C. The Gaussian dynamics is thus the gradient flow of S2S_2 with mobility Γ/2\Gamma/2, plus a nondissipative current JndJ^{\mathrm{nd}}, exactly the structure (II.5) with a Hamiltonian-like part. In other words, the zero-cost flow for L2L_2 is part of the GENERIC family [5, 6]. However, from the validity of the normal approximation we must have C−1=−S′′(zs)C^{-1}=-S''(z^s) (shown in Proposition D.1, when SS is twice differentiable at zsz^s). Hence, S2S_2 (defined under (V.3)) is the quadratic part of the nonequilibrium entropy SS of [3], not of the equilibrium entropy SS. Proof is in Appendix D.

Comparing with physical local detailed balance, two violations appear, both vanishing at equilibrium:

  1. the Gaussian force F0(v)\mathcal{F}_0(v) differs from F′vF'v whenever the stationary covariance differs from the equilibrium one;

  1. the emergent current JndJ^{\mathrm{nd}} is nonzero. Its part outside ker⁡D\operatorname{ker}\mathcal{D} moves the state while conserving S2S_2 (a circulation Q≠0Q\ne0); its part in ker⁡D\operatorname{ker}\mathcal{D} is a cycle current driven by the state fluctuation.

At equilibrium G=0G=0, JH=0J^{H}=0: K=12BS′′K=\frac{1}{2}BS'', C−1=−S′′C^{-1}=-S'', F0(v)=F′v\mathcal{F}_0(v)=F'v and Jnd=0J^{\mathrm{nd}}=0 and the standard Onsager (linear) processes are recovered.

Examples

One-species networks. See Example III A. The state is one-dimensional, so Q=0Q=0, and we can solve (V.3),

K=λ1+λ2=12(t1F1+t2F2),C=−t1+t22K.(37)K=\lambda_1+\lambda_2=\frac{1}{2}(t_1\mathcal{F}_1+t_2\mathcal{F}_2),\qquad C=-\frac{t_1+t_2}{2K}. \tag*{(37)}

See also the end of Appendix D. The emergent current is a pure cycle current,

Jnd(v)=ωv(1−1),ω=λ1t2−λ2t1t1+t2=t1t22(t1+t2)(F1−F2).(38)J^{\mathrm{nd}}(v)=\omega v\begin{pmatrix}1\\-1\end{pmatrix},\qquad\omega=\frac{\lambda_1t_2-\lambda_2t_1}{t_1+t_2}=\frac{t_1t_2}{2(t_1+t_2)}(\mathcal{F}_1-\mathcal{F}_2). \tag*{(38)}

The physical force has no fluctuating cycle component: F1−F2\mathcal{F}_1-\mathcal{F}_2 is fixed by the chemostats, so F1′=F2′\mathcal{F}'_1=\mathcal{F}'_2. The Gaussian force does: F1−F2\mathcal{F}_1-\mathcal{F}_2 is, by (IV.13), proportional to the stationary cycle current. For the catalytic network,

F1−F2=Jzs(3t1+1t2),ω=J(t1+3t2)4zs(1+zs).(39)\mathcal{F}_1-\mathcal{F}_2=\frac{J}{z^s}\left(\frac{3}{t_1}+\frac{1}{t_2}\right),\qquad\omega=\frac{J(t_1+3t_2)}{4z^s(1+z^s)}. \tag*{(39)}

and, using t1+t2=2(1+zs)t_1+t_2=2(1+z^s) and K=−ΔK=-\Delta,

C=1+zsΔ,Czs=1+zszsΔ.(40)C=\frac{1+z^s}{\Delta},\qquad\frac{C}{z^s}=\frac{1+z^s}{z^s\Delta}. \tag*{(40)}

At equilibrium (κ1=κ2=κ\kappa_1=\kappa_2=\kappa, zs=1/κz^s=1/\kappa, Δ=1+κ\Delta=1+\kappa) the Fano factor (variance of the number of molecules divided by its mean) is C/zsC/z^s and equals one, which is the Poisson value you would predict from the equilibrium expression of the entropy S(z)=−zlog⁡(z/zeq)+zS(z)=-z\log(z/z^{\mathrm{eq}})+z.

Ring. See Example III B. On mean-zero fluctuations, K=D∂x2−c∂xK=D\partial_x^2-c\partial_x and B=−2χ∂x2B=-2\chi\partial_x^2, and the Lyapunov equation is solved by the local-equilibrium covariance C=−1/s′′(ρˉ)C=-1/s^{\prime\prime}(\bar{\rho}) times the identity (Appendix E). Hence F0(v)=F′v\mathcal{F}_0(v)=\mathcal{F}'v: the Gaussian entropy is the local-equilibrium one, as for the product invariant measure of exclusion. The “generalized local detailed balance” (V.5) holds with current,

Jnd(v)=cv,Q=−cs′′(ρˉ)∂x,(41)J^{\mathrm{nd}}(v)=cv,\qquad Q=-\frac{c}{s^{\prime\prime}(\bar{\rho})}\partial_x, \tag*{(41)}

which transports fluctuations at the drift velocity c=χ′(ρˉ)Ec=\chi'(\bar{\rho})E without changing S2S_2. For exclusion at half filling, c=0c=0: there, normal fluctuations cannot be told apart from equilibrium ones.

Intuitive meaning

At equilibrium a fluctuation relaxes because the thermodynamic force pulls it back, and its reversal costs exactly the entropy it produces. Around a driven steady state, a fluctuation also changes how active the transport channels are, and the stationary current through those channels then changes even at fixed force. Seen from the fluctuation, that kinetic effect looks like an extra force, which is the term 2Γ−1Θv2\Gamma^{-1}\Theta v. It is proportional both to how strongly the activity depends on the state and to how hard the system is driven. In the chemical example, a fluctuation in zz changes the activity of the two reactions unequally and so drives a cycle current, even though the cycle affinity itself is fixed. On the ring, a density fluctuation changes the mobility and so the field-driven current, and the fluctuation is carried along. The Gaussian description absorbs all such effects into its own entropy and its own nondissipative current.

Normal dynamics

We come to the second point of discussion about the normal approximation, the Gaussian dynamics itself.

Langevin equations

The Gaussian path measure (IV.3) with Lagrangian (IV.8) is that of the linear stochastic dynamics

ι(t)=Λv(t)+η(t),v˙(t)=Dι(t),⟨η(t)η(t′)T⟩=Γδ(t−t′)(42)\iota(t) = \Lambda v(t) + \eta(t), \qquad\dot{v}(t) = \mathcal{D}\iota(t), \qquad\langle\eta(t)\eta(t')^{\mathsf T} \rangle= \Gamma\delta(t-t') \tag*{(42)}

with white noise η\eta on the space of currents. For the state alone,

v˙=Kv+ξ,K=DΛ,⟨ξ(t)ξ(t′)T⟩=Bδ(t−t′),B=DΓDT.(43)\dot{v} = Kv + \xi, \qquad K = \mathcal{D}\Lambda, \qquad\langle\xi(t)\xi(t')^{\mathsf T} \rangle= B\delta(t-t'), \qquad B = \mathcal{D}\Gamma\mathcal{D}^{\mathsf T}. \tag*{(43)}

The integrated current fluctuation I(t)=∫0tι(s) dsI(t) = \int_0^t \iota(s)\,\mathrm{d}s satisfies DI(t)=v(t)−v(0)\mathcal{D}I(t) = v(t)-v(0) and

I˙=KII+Λv(0)+η,KI=ΛD,(44)\dot{I} = K_I I + \Lambda v(0) + \eta, \qquad K_I = \Lambda\mathcal{D}, \tag*{(44)}

with the intertwining DKI=KD\mathcal{D}K_I = K\mathcal{D}. The noise matrix Γ\Gamma is the same for all channels whose currents are measured: the Hessian has become the strength of the current noise in (VI.1).

When D\mathcal{D} has a nontrivial co-kernel (conservation laws, as for the ring), KK has a kernel and the fluctuations are restricted to the subspace compatible with the conservation laws, on which KK is assumed stable. All inverses below are taken on that subspace.

Stationary correlations

The stationary covariance CC solves (V.3), and ⟨v(t)v(0)T⟩=eKtC\langle v(t)v(0)^{\mathsf{T}}\rangle= e^{Kt}C for t≥0t \ge0 (Onsager regression). For current fluctuations, for t>0t > 0,

⟨ι(t)ι(0)T⟩=ΛeKt(CΛ†+DΓ),(45)\langle\iota(t)\iota(0)^{\mathsf{T}}\rangle= \Lambda e^{Kt}(C\Lambda^{\dagger}+\mathcal{D}\Gamma), \tag*{(45)}

plus the white-noise part Γδ(t)\Gamma\delta(t). The integrated current has a diverging variance, linear in time, along ker⁡D\ker\mathcal{D}.

Proposition VI.1. The long-time covariance rate of the integrated current is

Σ=lim⁡t→∞1t⟨I(t)I(t)T⟩=ΠΓΠ†,Π=1−ΛK−1D.(46)\Sigma= \lim_{t\to\infty}\frac{1}{t}\langle I(t)I(t)^{\mathsf{T}}\rangle= \Pi\Gamma\Pi^{\dagger}, \qquad\Pi= 1-\Lambda K^{-1}\mathcal{D}. \tag*{(46)}

Π\Pi is the (oblique) projector onto the cycle space ker⁡D\ker\mathcal{D} along the image space im⁡Λ\operatorname{im}\Lambda of state-induced currents. The limit etKI→Πe^{tK_I}\to\Pi as t→∞t\to\infty for KIK_I in (VI.3). Equivalently, Σ=∫−∞∞⟨ι(t)ι(0)T⟩ dt\Sigma=\int_{-\infty}^{\infty}\langle\iota(t)\iota(0)^{\mathsf{T}}\rangle\,\mathrm{d}t.

The proof is in Appendix B. The long-time current fluctuations are therefore the current noise projected onto cycles, along the directions in which currents follow the state.

Note: At equilibrium, the image space im⁡Λ\operatorname{im}\Lambda is the Γ−1\Gamma^{-1}-orthogonal complement of ker⁡D\ker\mathcal{D}, so that Π\Pi is the orthogonal projector in the noise metric (x,Γ−1y)(x,\Gamma^{-1}y). Out of equilibrium, Θ\Theta tilts the image space im⁡Λ\operatorname{im}\Lambda and Π\Pi becomes oblique. This obliqueness is what breaks the Green–Kubo relation by adding a frenetic contribution; more on that in Sec. VIII C.

One-species networks. See Example III A. ker⁡D\ker\mathcal{D} is spanned by (1,−1)(1,-1). Writing I=Ic(1,−1)+…I=I_c(1,-1)+\ldots, the cycle current has variance rate

Σc=λ12t1+λ22t2K2=t1t2(t1F12+t2F22)(t1F1+t2F2)2.(47)\Sigma_c=\frac{\lambda_1^2t_1+\lambda_2^2t_2}{K^2}=\frac{t_1t_2\left(t_1\mathcal{F}_1^2+t_2\mathcal{F}_2^2\right)}{\left(t_1\mathcal{F}_1+t_2\mathcal{F}_2\right)^2}. \tag*{(47)}

At equilibrium, F1=F2\mathcal{F}_1=\mathcal{F}_2 and Σc=t1t2/(t1+t2)\Sigma_c=t_1t_2/(t_1+t_2): the two reactions act as resistors in series.

Ring. See Example III B. The spatially averaged current Iˉ=L−1∫I dx\bar{I}=L^{-1}\int I\,dx obeys Iˉ˙=ηˉ\dot{\bar{I}}=\bar{\eta}, because ∫Λv dx=c∫v dx=0\int\Lambda v\,dx=c\int v\,dx=0 by mass conservation. Hence ΣIˉ=2χ(ρˉ)/L\Sigma_{\bar{I}}=2\chi(\bar{\rho})/L, unchanged by the drive at the Gaussian level.

Gaussian reconstruction

The quadratic Lagrangian is fully identified from stationary time series of states and integrated currents. Starting from the Langevin equations (VI.1) and (VI.3) we find:

Γ=lim⁡Δt→0⟨ΔI ΔIT⟩Δt,Λ=lim⁡Δt→0⟨ΔI v(t)T⟩ΔtC−1(48)\Gamma=\lim_{\Delta t\to0}\frac{\langle\Delta I\,\Delta I^{\mathsf T}\rangle}{\Delta t},\qquad \Lambda=\lim_{\Delta t\to0}\frac{\langle\Delta I\,v(t)^{\mathsf T}\rangle}{\Delta t}C^{-1} \tag*{(48)}

with ΔI=I(t+Δt)−I(t)\Delta I=I(t+\Delta t)-I(t); that is, the noise matrix is the quadratic variation of the currents, and Λ\Lambda is the regression of current increments on the state. From states alone, one obtains only K=DΛK=D\Lambda, as ddt⟨v(t)v(0)T⟩∣0+C−1\left.\frac{d}{dt}\langle v(t)v(0)^{\mathsf T}\rangle\right|_{0+}C^{-1}, and B=DΓD†B=D\Gamma D^{\dagger} from (V.3): the cycle components of Γ\Gamma and Λ\Lambda remain invisible. From currents alone, short times give Γ\Gamma, long times give Σ\Sigma, and the regression of current correlations gives KIK_I, as in (VI.3).

That identifies part of the original Lagrangian. The noise matrix is the Hessian ∂f2ψ∗\partial_f^2\psi^* at the stationary force, while the relaxation equation only gives the gradient ∂fψ∗\partial_f\psi^* at that force. If F′F' is known from thermodynamics, Θ=Λ−12ΓF′\Theta=\Lambda-\frac{1}{2}\Gamma F' gives the mixed derivative ∂z∂fψ∗\partial_z\partial_f\psi^*.

For chemical networks that is enough to recover everything locally: the mean current gives ri−r−ir_i-r_{-i} and the current noise gives ti=ri+r−it_i=r_i+r_{-i}, hence both rates, hence aia_i, FiF_i and L\mathcal{L} at zsz^s. Varying a control parameter, such as a chemostat concentration, moves the stationary state; repeating the measurement along the way reconstructs the rate functions, and hence L\mathcal{L}, over the range of states that can be made stationary, provided the dependence of the rates on the control parameter is known, as it is for mass-action kinetics.

In general, if the driving can be varied at fixed state, integrating d2ds2ψ∗(sF/2,z)=14(F,Γ(s)F)\frac{d^2}{ds^2}\psi^*(sF/2,z)=\frac{1}{4}(F,\Gamma(s)F) along the ray recovers ψ∗(⋅,z)\psi^*(\cdot,z) there.

Connections with earlier Gaussian theories and fluctuating hydrodynamics

The algebra of the Gaussian level is standard: a linear stochastic dynamics, its stationary covariance from a Lyapunov equation, and a split of the drift into a symmetric and an antisymmetric part. What the canonical structure adds is the extension of the Langevin dynamics by including the current variables, and the physical identification of all objects: the current noise as the Hessian of the dissipation function at the operating point, and the linear force derived from (the Hessian of) the nonequilibrium entropy in combination with an emergent current from the state dependence of the dynamical activity.

In the following, we compare our results with previous studies and make the connections explicit.

Gaussian fluctuations around steady states

Lax [18] treated Gaussian fluctuations around nonequilibrium steady states as linear processes whose stationary covariance CC solves a Lyapunov equation, our (V.3), and stressed that CC is in general not the inverse Hessian of the equilibrium entropy. Keizer [19] built on this a statistical thermodynamics of nonequilibrium processes. Two of his ingredients reappear here. First, in his canonical theory of elementary processes the noise intensity of each process is fixed by the sum of its forward and backward rates. That is exactly our noise matrix for a channel-wise dissipation function: for ψ∗(f,z)=∑i2ai(z)(cosh⁡fi−1)\psi^*(f,z) = \sum_i 2a_i(z)(\cosh f_i - 1), the Hessian at the operating force is Γii=2aicosh⁡(Fi/2)=ti=ri+r−i\Gamma_{ii} = 2a_i\cosh(F_i/2) = t_i = r_i + r_{-i}, the traffic. Keizer’s postulate is thus the statement that the noise matrix is ∂f2ψ∗\partial_f^2\psi^* at the stationary force, not at zero force. Second, Keizer proposed a nonequilibrium entropy whose second derivative at the steady state is −C−1-C^{-1}, and suggested measuring it through fluctuations. Proposition D.1 derives that identification from large deviations: −C−1-C^{-1} is the Hessian of the nonequilibrium entropy SS, the quasipotential of the stationary distribution. Keizer’s entropy is therefore our Gaussian entropy S2S_2, and it is the quadratic part of a globally defined object.

Of course, knowing S2S_2 does not determine the Gaussian dynamics: by Proposition V.1 the drift is the gradient flow of S2S_2 plus the emergent current JndJ^{\mathrm{nd}}, and it is JndJ^{\mathrm{nd}} that will control the violations of fluctuation–dissipation and Green–Kubo relations. Moreover, working with currents rather than states, we keep track of cycle components of Γ\Gamma and JndJ^{\mathrm{nd}} that a description in terms of states alone, as in [18, 19], cannot see. In the one-species networks the state is one-dimensional and the entire nonequilibrium signature at the Gaussian level is such a cycle current.

Nonequilibrium potentials

For a diffusion z˙=b(z)+2/N σ(z)ξ\dot{z}=b(z)+\sqrt{2/N}\,\sigma(z)\xi with diffusion matrix B=σσTB=\sigma\sigma^{\mathsf{T}}, Graham [20] and Graham and Tél [21, 22] defined the nonequilibrium potential Φ\Phi as the quasipotential of the stationary distribution, Pst≍e−NΦP^{\mathrm{st}}\asymp e^{-N\Phi}. It solves the Hamilton–Jacobi equation (∇Φ,b)+(∇Φ,B∇Φ)=0(\nabla\Phi,b)+(\nabla\Phi,B\nabla\Phi)=0, which is equivalent to the decomposition

b=−B∇Φ+r,(∇Φ,r)=0,(49)b=-B\nabla\Phi+r,\qquad(\nabla\Phi,r)=0, \tag*{(49)}

of the drift into a part driven by the potential and a remainder orthogonal to its gradient. The remainder conserves Φ\Phi, and the time-reversed (adjoint) dynamics has drift −B∇Φ−r-B\nabla\Phi-r.

Proposition V.1(c) is the Gaussian instance of (VII.1). With Φ2=−S2=12(v,C−1v)\Phi_2=-\mathcal{S}_2=\frac{1}{2}(v,C^{-1}v) and B=DΓD†B=\mathcal{D}\Gamma\mathcal{D}^{\dagger}, the state drift splits as

Kv=−12BC−1v+DJnd(v),DJnd(v)=−QC−1v,Q†=−Q,(50)Kv=-\frac{1}{2}BC^{-1}v+\mathcal{D}J^{\mathrm{nd}}(v),\qquad\mathcal{D}J^{\mathrm{nd}}(v)=-QC^{-1}v,\qquad Q^{\dagger}=-Q, \tag*{(50)}

and DJnd\mathcal{D}J^{\mathrm{nd}} is orthogonal to ∇Φ2=C−1v\nabla\Phi_2=C^{-1}v. Furthermore, here, the decomposition is lifted from states to currents, Eq. (V.5), where it takes the form of a generalized local detailed balance with respect to S2\mathcal{S}_2.

Kwon, Ao and Thouless [23] showed that any linear stochastic dynamics x˙=−Fx+ζ\dot{x}=-Fx+\zeta, ⟨ζζT⟩=2D δ\langle\zeta\zeta^{\mathsf{T}}\rangle=2D\,\delta, can be rewritten exactly as

(S+T)x˙=−Ux+ξ,S=ST≥0,T=−TT,U=UT,(51)(S+T)\dot{x}=-Ux+\xi,\qquad S=S^{\mathsf{T}}\geq0,\qquad T=-T^{\mathsf{T}},\qquad U=U^{\mathsf{T}}, \tag*{(51)}

with ⟨ξξT⟩=2S δ\langle\xi\xi^{\mathsf{T}}\rangle=2S\,\delta and stationary distribution ∝e−12(x,Ux)\propto e^{-\frac{1}{2}(x,Ux)}. The construction solves the Lyapunov equation: U=C−1U=C^{-1}, and (S+T)−1=D+Q(S+T)^{-1}=D+Q with QQ antisymmetric, so that F=(D+Q)UF=(D+Q)U. In our notation, D=12BD=\frac{1}{2}B and F=−KF=-K, and the identity is our

K=−(12B+Q)C−1,Q=−KC−12B=−Q†,(52)K=-\left(\frac{1}{2}B+Q\right)C^{-1},\qquad Q=-KC-\frac{1}{2}B=-Q^{\dagger}, \tag*{(52)}

the same algebra as Proposition V.1(c). The decomposition (VII.3) is thus a mathematical fact about linear processes, and our Proposition V.1 is in that sense its restatement.

The physics enters through three identifications that [23] does not make: (i) the potential U=C−1U=C^{-1} is the Hessian of the nonequilibrium entropy of the underlying nonlinear dynamics (Proposition D.1); (ii) the symmetric part DD is 12DΓD†\frac{1}{2}D\Gamma D^{\dagger}, the projection of the Hessian at the operating point; and (iii) the antisymmetric part QQ has a definite origin, the frenetic sensitivity Θ\Theta in the force (IV.7), so that it is of first order in the stationary driving and proportional to the state dependence of the activity. In particular, Q≠0Q\ne0 is a measurable signature of driving with state-dependent kinetics, not an arbitrary gauge choice.

Fluctuating hydrodynamics and macroscopic fluctuation theory

Fluctuating hydrodynamics adds noise to the hydrodynamic equations, following Landau and Lifshitz [24]: random stress and heat-flux terms are added to the Navier–Stokes equations, with covariance fixed by fluctuation–dissipation. This theory was extended and rigorously proven by [12, 13, 14] for large class of weakly driven or boundary-driven diffusions as limits of stochastic interacting particle systems.

In our framework, fluctuating hydrodynamics is the special case of a quadratic dissipation function, ψ∗(f,ρ)=∫χ(ρ)f2 dx\psi^{*}(f,\rho)=\int\chi(\rho)f^{2}\,\mathrm{d}x, which is exactly Example III B, Eq. (III.9). The quadratic Lagrangian L2L_{2} is then linearized fluctuating hydrodynamics, with Γ=2χ(ρˉ)\Gamma=2\chi(\bar{\rho}), and macroscopic fluctuation theory (MFT) is the large-deviation level of the same theory. MFT contains the decomposition that our Gaussian results linearize. With VV the quasipotential, the hydrodynamic current splits as

J(ρ)=−χ(ρ) ∂xδVδρ+A(ρ),∫∂xδVδρA(ρ) dx=0,(53)J(\rho)=-\chi(\rho)\,\partial_x\frac{\delta V}{\delta\rho}+A(\rho),\qquad\int\partial_x\frac{\delta V}{\delta\rho}A(\rho)\,\mathrm{d}x=0, \tag*{(53)}

which is equivalent to the Hamilton–Jacobi equation for VV, and the adjoint hydrodynamics, which governs the optimal fluctuation paths, reverses the first part and keeps AA [12].

Linearizing (VII.5) around ρˉ\bar{\rho} gives Proposition V.1: VV becomes −S2-S_{2}, the force −∂xδV/δρ-\partial_x\delta V/\delta\rho becomes F0\mathcal{F}_{0}, and AA becomes JndJ^{\mathrm{nd}}. For the ring, AA linearizes to the transport current Jnd=cvJ^{\mathrm{nd}}=cv with c=χ′(ρˉ)Ec=\chi^{\prime}(\bar{\rho})E, and the Gaussian entropy is the local-equilibrium one, in agreement with the product invariant measure of the weakly asymmetric exclusion process on a ring.

The comparison also shows what is specific to the present setting. In fluctuating hydrodynamics and MFT the noise is fixed by the near-equilibrium mobility. Our general result, that the noise matrix is the Hessian ∂f2ψ∗\partial_f^2\psi^* at the operating force, has no counterpart there.

Outlook

Multistability: the Schlögl model

When the relaxation equation has several stable stationary points, as the Schlögl model (III.7) does for a range of parameters, the normal approximation applies around each of them separately, on times short compared with the switching time, which grows exponentially in NN with the barrier in the nonequilibrium entropy SS. All formulas of Secs. IV–VI for one-species networks apply at each stable point, with

λ1=2k1αzs−3k−1(zs)2,λ2=−k−2.(54)\lambda_1 = 2k_1\alpha z^s - 3k_{-1}(z^s)^2,\qquad\lambda_2=-k_{-2}. \tag*{(54)}

Bistability requires a nonzero cycle affinity, since with detailed balance the rate equation has a unique stable point [25]. Consequently F1≠F2\mathcal{F}_1 \ne\mathcal{F}_2 at each of the two stable points, and both carry an emergent cycle current ϖ\varpi. As a stable point approaches a saddle-node bifurcation, K=λ1+λ2→0K=\lambda_1+\lambda_2\to0, so CC and Σc\Sigma_c diverge as 1/∣K∣1/\lvert K\rvert and 1/K21/K^2. The normal approximation then breaks down: fluctuations become non-Gaussian, on a larger scale in NN.

Limit cycles

Write the relaxation equation (II.6) as z˙=X(z)\dot{z}=X(z), with X(z)=DjzX(z)=\mathcal{D}j_z, and suppose it has a stable periodic orbit γ(t)\gamma(t) of period TT, as in the Brusselator [26]. Limit cycles need at least two state variables, so they do not occur for one-species networks. Writing

z=γ(t)+vN,j=jγ(t)+ιN,(55)z=\gamma(t)+\frac{v}{\sqrt{N}},\qquad j=j_{\gamma(t)}+\frac{\iota}{\sqrt{N}}, \tag*{(55)}

the derivation of Appendix A, with the stationary point replaced by the point γ(t)\gamma(t) on the orbit, gives a time-periodic quadratic Lagrangian

L2(ι,v;t)=12(ι−Λ(t)v, Γ(t)−1(ι−Λ(t)v)).(56)L_2(\iota,v;t)=\frac{1}{2}\left(\iota-\Lambda(t)v,\ \Gamma(t)^{-1}(\iota-\Lambda(t)v)\right). \tag*{(56)}

with Γ(t)=∂f2ψ∗(F(γ(t))/2,γ(t))\Gamma(t) = \partial_f^2\psi^*(F(\gamma(t))/2,\gamma(t)) and Λ(t)v=ddϵjγ(t)+ϵv∣ϵ=0\Lambda(t)v = \left.\frac{d}{d\epsilon}j_{\gamma(t)+\epsilon v}\right|_{\epsilon=0}. The associated Gaussian dynamics is that of (VI.1), now with periodic coefficients:

ι=Λ(t)v+η,v˙=K(t)v+Dη,⟨η(t)η(t′)T⟩=Γ(t)δ(t−t′),(57)\iota= \Lambda(t)v+\eta,\qquad\dot{v}=K(t)v+\mathcal{D}\eta,\qquad\langle\eta(t)\eta(t')^\mathsf{T}\rangle=\Gamma(t)\delta(t-t'), \tag*{(57)}

with K(t)=DΛ(t)K(t)=\mathcal{D}\Lambda(t). The state noise Dη\mathcal{D}\eta has covariance B(t)=DΓ(t)D†B(t)=\mathcal{D}\Gamma(t)\mathcal{D}^{\dagger}, as in (VI.2).

It is possible to record the time along the limit cycle, called the asymptotic phase ϕ\phi: each zz in the basin of attraction of the orbit has an asymptotic phase ϕ(z)∈R/TZ\phi(z)\in\mathbb{R}/T\mathbb{Z}: the solution starting at zz converges to γ(t+ϕ(z))\gamma(t+\phi(z)). Hence ϕ(γ(s))=s\phi(\gamma(s))=s and ϕ(z(t))=ϕ(z(0))+t\phi(z(t))=\phi(z(0))+t along solutions. The level sets of ϕ\phi are the isochrons. To first order, ϕ(γ(t)+v/N)=t+θ/N\phi(\gamma(t)+v/\sqrt{N})=t+\theta/\sqrt{N} defining the phase fluctuation θ\theta. It turns out that θ\theta accumulates noise without any restoring force: it is a time-changed Brownian motion. For further details, we refer to [27]. Bounds on the number of coherent oscillations in terms of the cycle affinities have been conjectured [28].

Local detailed balance is violated here in an essential way. On a limit cycle, the reversed orbit γ(−t)\gamma(-t) does not solve the relaxation equation, and by (II.3) its cost per period is NσTN\sigma_T with

σT=∮(γ˙(t),F(γ(t))) dt>0,(58)\sigma_T=\oint\left(\dot{\gamma}(t),F(\gamma(t))\right)\,dt>0, \tag*{(58)}

the entropy produced per period. Time reversal thus maps the Gaussian window around γ\gamma into a window around an exponentially improbable path, and no reversal symmetry, not even a modified one, acts on the phase.

Linear response

In Section VI all expectations ⟨⋅⟩\langle\cdot\rangle are taken in the Gaussian dynamics (VI.1), that is, in the normal approximation. The variables vv and ι\iota are the rescaled fluctuations of (IV.1), their stationary law is N(0,C)\mathcal{N}(0,C) with CC from (V.3), and the time correlations are ⟨v(t)v(s)T⟩=eK(t−s)C\langle v(t)v(s)^\mathsf{T}\rangle=e^{K(t-s)}C and (VI.4). They are the limits as N↑∞N\uparrow\infty of NN times the connected (cumulant) correlations of the state and current of the original process, at fixed times. Higher connected correlations (third, fourth,...) are smaller still, of order N−2N^{-2} and beyond. At leading order only the covariance survives, so the process is Gaussian. The correlations are those within a single well of a stable stationary point.

At the macroscopic level the linear response is the same as the Gaussian one. Consider a perturbation F→F+hF \to F+h at fixed dissipation function, i.e., ψ∗\psi^* unchanged. The typical current becomes jzh=∂zψ∗(12(F(z)+h),z)j_z^h = \partial_z\psi^*\left(\frac{1}{2}(F(z)+h),z\right), and expanding around zsz^s to first order in δz=z−zs\delta z = z-z^s and hh,

jzh=js+Λδz+12Γh,δz˙=Kδz+12DΓh,(59)j_z^h = j^s + \Lambda\delta z + \frac{1}{2}\Gamma h,\qquad\dot{\delta z} = K\delta z + \frac{1}{2}D\Gamma h, \tag*{(59)}

where the hh-term is the Hessian Γ=∂ψ2ψ∗(Fs/2,zs)\Gamma= \partial_\psi^2\psi^*(F^s/2,z^s) acting on h/2h/2. On the other hand, in the Gaussian dynamics, ι=Λv+12Γh+η\iota= \Lambda v + \frac{1}{2}\Gamma h + \eta and ⟨v˙⟩=K⟨v⟩+12DΓh\langle\dot{v}\rangle= K\langle v\rangle+ \frac{1}{2}D\Gamma h. The latter coincides with (VIII.6) upon identifying δz=⟨v⟩/N\delta z = \langle v\rangle/\sqrt{N} and taking the perturbation of order N−1/2N^{-1/2} at the Gaussian level, h→h/Nh \to h/\sqrt{N}, so that the shifted state remains in the Gaussian window. Hence, starting from the steady state,

δz(t)=∫−∞teK(t−s)12DΓh(s) ds,δj(t)=Λδz(t)+12Γh(t),(60)\delta z(t) = \int_{-\infty}^{t} e^{K(t-s)}\frac{1}{2}D\Gamma h(s)\,\mathrm{d}s,\qquad\delta j(t) = \Lambda\delta z(t) + \frac{1}{2}\Gamma h(t), \tag*{(60)}

and for constant hh the stationarity condition Dδj=0D\delta j = 0 shifts the state by δvˉ=−12K−1DΓh\delta\bar{v} = -\frac{1}{2}K^{-1}D\Gamma h, giving δj=12ΠΓh\delta j = \frac{1}{2}\Pi\Gamma h.

The results on linear response will be part of subsequent work.

Ballistic regime

Our scaling, with drift and noise both of order NN in the action as for density-dependent jump processes [8], is the diffusive or reactive one. For strongly asymmetric lattice gases and for integrable systems the relevant scaling is Euler (ballistic), x∼t∼Lx \sim t \sim L. The hydrodynamics is then hyperbolic, for instance the inviscid Burgers equation ∂tρ+∂x[ρ(1−ρ)]=0\partial_t\rho+ \partial_x[\rho(1-\rho)] = 0 for the totally asymmetric exclusion process, or generalized hydrodynamics for integrable models, with dissipation only at shocks.

Fluctuation theories exist at that scale, but none with the canonical structure of Sec. II. In ballistic macroscopic fluctuation theory [29] there is no dynamical noise at leading order: fluctuations come from the randomness of the initial state and are transported deterministically by the Euler flow, and the action is the initial-state large-deviation functional pushed along characteristics. For the totally asymmetric exclusion process, the large deviations of Jensen and Varadhan at Euler scale [30, 31] have a rate function given by the entropy production of non-entropic weak solutions, so that the only time-antisymmetric cost sits at shocks. Beyond the Gaussian window, asymmetric systems show Kardar–Parisi–Zhang fluctuations with t1/3t^{1/3} scaling [32], so a Gaussian theory can only describe short times or weak asymmetry; on the ring this is the window in which our drift velocity cc transports fluctuations without distorting them.

In the language of Sec. II, the ballistic limit is the one in which the Hamiltonian current JHJ^{H}, the Euler current that conserves the entropy of smooth solutions, is of order one, while the force FF and the noise Γ\Gamma are suppressed by 1/L1/L. The canonical structure then degenerates: the zero-cost flow is purely Hamiltonian, the dynamical Legendre pair (ψ,ψ∗)(\psi,\psi^{*}) disappears at leading order, and local detailed balance reduces to the time-reversal invariance of the Euler flow, supplemented by entropy production at shocks. There is therefore no canonical formalism for dynamical fluctuations at the ballistic scale in the present sense. Ballistic MFT takes its place, and the diffusive corrections, at Navier–Stokes order, are where FF and Γ\Gamma re-enter as weak dissipation on top of an Euler JHJ^{H}. That “weakly dissipative” setting, with JHJ^{H} the Euler current and the canonical pair at order 1/L1/L, is a natural extension of the present work.

Conclusion

Most often measured or simulated around a stable nonequilibrium steady state are the normal fluctuations. In this paper we asked 1) what survives of local detailed balance and of the canonical structure of macroscopic dynamical fluctuations, 2) whether and how the normal fluctuations and perhaps the larger dynamical fluctuations as well, can be reconstructed.

The quadratic Lagrangian of the fluctuations (v,ι)(v,\iota) keeps the canonical form with a quadratic dissipation function and a linear force F(v)\mathcal{F}(v) acting on the state fluctuation. The noise matrix is the Hessian of ψ∗\psi^{*} at the actual stationary force Fs/2.F\mathcal{F}^{s}/2. \mathcal{F} is not the linearized thermodynamic force Fv′\mathcal{F}'_{v}: it contains an additional term expressing the sensitivity of the dynamical activity to the state, proportional to the stationary driving. It vanishes when the dissipation function does not depend on the state. As a result, normal fluctuations around steady nonequilibrium are Onsager processes with respect to a modified entropy plus a nondissipative flow, that is of GENERIC-type. Secondly, the Gaussian path measure is that of a linear (Ornstein–Uhlenbeck) dynamics, for the state and for the integrated current. The quadratic Lagrangian can be identified from stationary time series.

We mention that for linear response, the Gaussian description reproduces the macroscopic response, and the emergent nondissipative current is expected to control the violation of fluctuation–dissipation and Green–Kubo relations.

Acknowledgements

KM is supported by Charles University, project GA UK No. 414225.

References

Derivation of the quadratic Lagrangian L2L_2

Write L(j,z)\mathcal{L}(j,z) in the form (II.5) with jH=0j^H=0. For every zz, jzj_z minimizes L(⋅,z)\mathcal{L}(\cdot,z) with L(jz,z)=0\mathcal{L}(j_z,z)=0, so ∂jL(jz,z)=0\partial_j\mathcal{L}(j_z,z)=0. By Legendre duality, ∂jψ(jz,z)=F(z)/2\partial_j\psi(j_z,z)=F(z)/2 and

∂j2L(jz,z)=∂j2ψ(jz,z)=[∂j2ψ∗(F(z)/2,z)]−1.(61)\partial_j^2\mathcal{L}(j_z,z)=\partial_j^2\psi(j_z,z)=\left[\partial_j^2\psi^*\left(F(z)/2,z\right)\right]^{-1}. \tag*{(61)}

Taylor’s theorem in jj at fixed zz gives L(jz+δ,z)=12(δ,∂j2L(jz,z)δ)+O(∣δ∣3)\mathcal{L}(j_z+\delta,z)=\frac{1}{2}(\delta,\partial_j^2\mathcal{L}(j_z,z)\delta)+O(|\delta|^3). With (IV.1), jz=j∗+Λv/N+O(1/N)j_z=j^*+\Lambda v/\sqrt{N}+O(1/N), so δ=(ι−Λv)/N+O(1/N)\delta=(\iota-\Lambda v)/\sqrt{N}+O(1/N), and ∂j2L(jz,z)=Γ−1+O(N−1/2)\partial_j^2\mathcal{L}(j_z,z)=\Gamma^{-1}+O(N^{-1/2}). The derivative of jz=∂fψ∗(F(z)/2,z)j_z=\partial_f\psi^*(F(z)/2,z) in the direction vv is Λv=12ΓF′v+Θv\Lambda v=\frac{1}{2}\Gamma F'v+\Theta v, which is (IV.6). Multiplying by NN and letting N→∞N\to\infty gives (IV.8). Expanding (IV.8) and using F(v)/2=Γ−1ΛvF(v)/2=\Gamma^{-1}\Lambda v gives (IV.9).

Gaussian process

From (VI.2), v(t)=eKtv(0)+∫0teK(t−s)Dη(s) dsv(t)=e^{Kt}v(0)+\int_0^t e^{K(t-s)}D\eta(s)\,\mathrm{d}s, which gives ⟨v(t)v(0)T⟩=eKtC\langle v(t)v(0)^\mathsf{T}\rangle=e^{Kt}C. Taking the equal-time covariance and imposing stationarity yields the Lyapunov equation (V.3) (see, e.g., [33]). For t>0t>0, ⟨η(0)v(t)T⟩=ΓD†eKTt\langle\eta(0)v(t)^\mathsf{T}\rangle=\Gamma D^\dagger e^{K^\mathsf{T}t}, which gives (VI.4).

Proof of Proposition VI.1. Integrating (VI.2) gives ∫0tv ds=K−1(v(t)−v(0))−K−1D∫0tη ds\int_0^t v\,\mathrm{d}s=K^{-1}(v(t)-v(0))-K^{-1}D\int_0^t\eta\,\mathrm{d}s. Hence

I(t)=ΛK−1(v(t)−v(0))+Π∫0tη(s) ds.(62)I(t)=\Lambda K^{-1}(v(t)-v(0))+\Pi\int_0^t\eta(s)\,\mathrm{d}s. \tag*{(62)}

This can be read as a decomposition to the space of cycle currents and state-induced currents. The first term is bounded in time, so Σ=ΠΓΠ†\Sigma=\Pi\Gamma\Pi^\dagger.

Properties of Π\Pi: a) DΠ=D−KK−1D=0D\Pi=D-KK^{-1}D=0, so im⁡Π⊂ker⁡D\operatorname{im}\Pi\subset\ker D, b) Πx=x\Pi x=x for x∈ker⁡Dx\in\ker D, c) ΠΛ=Λ−ΛK−1K=0\Pi\Lambda=\Lambda-\Lambda K^{-1}K=0. Thus Π\Pi is the projector onto ker⁡D\ker D along im⁡Λ\operatorname{im}\Lambda. Moreover etKI=1+Λ(∫0tesK ds)D→1−ΛK−1De^{tK_I}=1+\Lambda\left(\int_0^t e^{sK}\,\mathrm{d}s\right)D\to1-\Lambda K^{-1}D.

The equivalence with ∫⟨ι(t)ι(0)T⟩dt=Γ−ΛK−1(CΛ†+DΓ)−(ΛC+ΓD†)K−†Λ†\int\langle\iota(t)\iota(0)^\mathsf{T}\rangle\mathrm{d}t=\Gamma-\Lambda K^{-1}(C\Lambda^\dagger+D\Gamma)-(\Lambda C+\Gamma D^\dagger)K^{-\dagger}\Lambda^\dagger follows from expanding ΠΓΠ†\Pi\Gamma\Pi^\dagger and using K−1C+CK−†=−K−1BK−†K^{-1}C+CK^{-\dagger}=-K^{-1}BK^{-\dagger}, which is (V.3).

Proof of Proposition V.1

Since Λv=Jnd(v)+12ΓF0(v)\Lambda v=J^{\mathrm{nd}}(v)+\frac{1}{2}\Gamma F_0(v), expanding (IV.8) around JndJ^{\mathrm{nd}} gives (a) exactly as in Appendix A. For (c), define Q=−KC−12BQ=-KC-\frac{1}{2}B. The Lyapunov equation (V.3) gives Q+Q†=−(KC+CK†)−B=0Q + Q^\dagger= -(KC + CK^\dagger) - B = 0. Then

DJnd(v)=Kv+12DΓD†C−1v=(K+12BC−1)v=−QC−1v.(63)\mathcal{D}J^{\mathrm{nd}}(v) = Kv + \frac{1}{2}\mathcal{D}\Gamma\mathcal{D}^{\dagger}C^{-1}v = \left(K + \frac{1}{2}B C^{-1}\right)v = -Q C^{-1}v. \tag*{(63)}

For (b), (Jnd,F0)=(DJnd,∇S2)=(QC−1v,C−1v)=0(J^{\mathrm{nd}},\mathcal{F}_0) = (\mathcal{D}J^{\mathrm{nd}},\nabla S_2) = (Q C^{-1}v,C^{-1}v) = 0 by antisymmetry of QQ. A part of JndJ^{\mathrm{nd}} in ker⁡D\ker\mathcal{D} is orthogonal to F0∈im⁡D†\mathcal{F}_0 \in\operatorname{im}\mathcal{D}^{\dagger} automatically.

At equilibrium, Λ=12ΓD†S′′\Lambda= \frac{1}{2}\Gamma\mathcal{D}^{\dagger}S^{\prime\prime} and C−1=−S′′C^{-1} = -S^{\prime\prime} solve (V.3), so Jnd=0J^{\mathrm{nd}} = 0.

The Hessian of the nonequilibrium entropy

Hamiltonian on force space. For a force ff (an element of the dual of the space of currents at zz) define

H(z,f)=sup⁡j{(f,j)−L(j,z)},(64)\mathcal{H}(z,f) = \sup_j \{(f,j) - \mathcal{L}(j,z)\}, \tag*{(64)}

the Legendre transform of L\mathcal{L} in jj. It is the cumulant generating function of the current at fixed state. With the canonical form (II.5), the substitution k=j−JH(z)k = j - J^{H}(z) and Legendre duality ψ∗∗=ψ∗\psi^{**} = \psi^* give

H(z,f)=(f,JH(z))+ψ∗(12F(z)+f,z)−ψ∗(12F(z),z).(65)\mathcal{H}(z,f) = (f,J^{H}(z)) + \psi^*\left(\frac{1}{2}F(z)+f,z\right) - \psi^*\left(\frac{1}{2}F(z),z\right). \tag*{(65)}

For JH=0J^{H}=0, local detailed balance (II.3) is equivalent to the symmetry

H(z,f)=H(z,−F(z)−f),(66)\mathcal{H}(z,f) = \mathcal{H}(z,-F(z)-f), \tag*{(66)}

since L(−j,z)=L(j,z)+(j,F(z))\mathcal{L}(-j,z) = \mathcal{L}(j,z) + (j,F(z)) turns sup⁡j{(−F−f,j)−L(j,z)}\sup_j\{(-F-f,j)-\mathcal{L}(j,z)\} into H(z,f)\mathcal{H}(z,f) after j→−jj \to-j. The same follows directly from (D.2) and the evenness of ψ∗\psi^*.

Hamilton–Jacobi equation. The quasipotential V=−SV=-S is the rate function of the stationary measure, Pst(z)≍e−NV(z)P^{\mathrm{st}}(z) \asymp e^{-NV(z)},

V(z)=inf⁡{∫−∞0L(j,z) ds:z˙=Dj, z(−∞)=zs, z(0)=z}.(67)V(z) = \inf\left\{\int_{-\infty}^{0}\mathcal{L}(j,z)\,\mathrm{d}s : \dot{z}=\mathcal{D}j,\ z(-\infty)=z^s,\ z(0)=z\right\}. \tag*{(67)}

Wherever VV is differentiable,

H(z,D†dV(z))=0,V(zs)=0.(68)\mathcal{H}(z,\mathcal{D}^{\dagger}\mathrm{d}V(z)) = 0,\qquad V(z^s)=0. \tag*{(68)}

Proposition D.1. Let V=−SV=-S be C2C^2 near zsz^s, solve (D.5), and have a strict local minimum at zsz^s, M:=V′′(zs)>0M:=V''(z^s)>0. Let K=DΛK=\mathcal{D}\Lambda be stable and B=DΓD†B=\mathcal{D}\Gamma\mathcal{D}^{\dagger} be positive definite (both on the subspace compatible with the conservation laws, if any). Then

M=C−1,that isC−1=−S′′(zs),(69)M=C^{-1},\qquad\text{that is}\qquad C^{-1}=-S''(z^s), \tag*{(69)}

where CC is the unique solution of the Lyapunov equation (V.3).

Proof. Put z=zs+xz=z^s+x. Since V(zs)=0V(z^s)=0 and zsz^s is a minimum, ∂V(z)=Mx+O(∣x∣2)\partial V(z)=Mx+O(|x|^2), hence f=D†∂V=D†Mx+O(∣x∣2)f=\mathcal{D}^{\dagger}\partial V=\mathcal{D}^{\dagger}Mx+O(|x|^2). Expanding (D.2) in ff at fixed zz, using that ∂fψ∗(F/2,z)+JH(z)=jz\partial_f\psi^*(F/2,z)+J^H(z)=j_z and ∂f2ψ∗(F/2,z)=Γ(z)\partial_f^2\psi^*(F/2,z)=\Gamma(z),

H(z,f)=(f,jz)+12(f,Γ(z)f)+O(∣f∣3).(70)\mathcal{H}(z,f)=(f,j_z)+\frac{1}{2}(f,\Gamma(z)f)+O(|f|^3). \tag*{(70)}

Now Djs=0\mathcal{D}j^s=0 and jz=js+Λx+O(∣x∣2)j_z=j^s+\Lambda x+O(|x|^2), so

(D†Mx,jz)=(Mx,Djz)=(Mx,Kx)+O(∣x∣3).(71)(\mathcal{D}^{\dagger}Mx,j_z)=(Mx,\mathcal{D}j_z)=(Mx,Kx)+O(|x|^3). \tag*{(71)}

Inserting f=D†Mxf=\mathcal{D}^{\dagger}Mx gives, to second order in xx,

0=H(z,D†∂V)=(Mx,Kx)+12(Mx,BMx)+O(∣x∣3).(72)0=\mathcal{H}(z,\mathcal{D}^{\dagger}\partial V)=(Mx,Kx)+\frac{1}{2}(Mx,BMx)+O(|x|^3). \tag*{(72)}

A quadratic form vanishes for all xx if and only if its symmetric part does. Using M=M†M=M^{\dagger} we get

MK+K†M+MBM=0.(73)MK+K^{\dagger}M+MBM=0. \tag*{(73)}

Since M>0M>0 is invertible, multiplying (D.10) from the left and from the right by M−1M^{-1} gives KM−1+M−1K†+B=0KM^{-1}+M^{-1}K^{\dagger}+B=0, i.e. M−1M^{-1} solves (V.3). For stable KK the Lyapunov equation has a unique solution, so M−1=CM^{-1}=C.

For Example III A, D=(1,1)\mathcal{D}=(1,1), V′(z)=log⁡(w−(z)/w+(z))V'(z)=\log(w_-(z)/w_+(z)) with w±w_{\pm} the sums of the forward and reverse rates, so V′′(zs)=−K/wV''(z^s)=-K/w with w=w+(zs)=w−(zs)=12(t1+t2)w=w_+(z^s)=w_-(z^s)=\frac{1}{2}(t_1+t_2). That equals C−1C^{-1} of (V.6). For the Schlögl example with (k1α,k−1,k−2,k2β)=(6,1,11,6)(k_1\alpha,k_{-1},k_{-2},k_2\beta)=(6,1,11,6), the rate equation is z˙=−(z−1)(z−2)(z−3)\dot{z}=-(z-1)(z-2)(z-3), with stable points z=1z=1 and z=3z=3. At z=1z=1: K=−2K=-2, w=12w=12, and V′′(1)=1/6=C−1V''(1)=1/6=C^{-1}.

Driven diffusion on a ring

In Fourier modes eikxe^{\mathrm{i}kx}, k=2πn/Lk = 2\pi n/L, n≠0n \ne0: K(k)=−Dk2−ickK(k) = -Dk^{2} - \mathrm{i}ck and B(k)=2χk2B(k) = 2\chi k^{2}, all at ρˉ\bar{\rho}.

The Lyapunov equation 2Re⁡K(k) C(k)+B(k)=02\operatorname{Re}K(k)\,C(k) + B(k) = 0 gives C(k)=χ/D=−1/s′′C(k) = \chi/D = -1/s^{\prime\prime}, independent of kk. Thus F0(v)=∂x(s′′v)=Fv′F_{0}(v) = \partial_{x}(s^{\prime\prime}v) = F'_{v}, and Jnd(v)=Λv+χ ∂x(v/C)=cv−D∂xv+D∂xv=cvJ^{\mathrm{nd}}(v) = \Lambda v + \chi\,\partial_{x}(v/C) = cv - D\partial_{x}v + D\partial_{x}v = cv.

Each Fourier mode decays at rate Dk2Dk^{2} and rotates at frequency ckck; the rotation is the circulation QQ.

References

  1. [1]A. Mielke, D. R. M. Renger, and M. A. Peletier. A Generalization of Onsager’s Reciprocity Relations to Gradient Flows with Nonlinear Mobility. J. Non-Equilib. Thermodyn., 41(2):141–149, 2016.
  2. [2]L. Onsager. Reciprocal Relations in Irreversible Processes. I. Phys. Rev., 37(4):405–426, 1931.DOI
  3. [3]C. Maes and K. Netočný. Relaxation to nonequilibrium. J. Non-Equilib. Thermodyn., 51(3):279–291, 2026.
  4. [4]A. Mielke, M. A. Peletier, and D. R. M. Renger. On the Relation between Gradient Flows and the Large-Deviation Principle, with Applications to Markov Chains and Diffusion. Potential Analysis, 41(4):1293–1327, 2014.DOI
  5. [5]R. Kraaij, A. Lazarescu, C. Maes, and M. Peletier. Deriving GENERIC from a Generalized Fluctuation Symmetry. J. Stat. Phys., 170(3):492–508, 2017.DOI
  6. [6]R. C. Kraaij, A. Lazarescu, C. Maes, and M. Peletier. Fluctuation symmetry leads to GENERIC equations with non-quadratic dissipation. Stoch. Proc. Appl., 130(1):139–170, 2020.DOI
  7. [7]N. G. van Kampen. Stochastic Processes in Physics and Chemistry. Elsevier, Amsterdam, 3rd edition, 2007.
  8. [8]T. G. Kurtz. Limit theorems for sequences of jump Markov processes approximating ordinary differential processes. J. Appl. Probab., 8:344–356, 1971.DOI
  9. [9]C. Maes and K. Netočný. Canonical structure of dynamical fluctuations in mesoscopic nonequilibrium steady states. EPL, 82(3):30003, 2008.
  10. [10]G. Falasco and M. Esposito. Macroscopic stochastic thermodynamics. Rev. Mod. Phys., 97:015002, 2025.DOI
  11. [11]F. Schlögl. Chemical reaction models for non-equilibrium phase transitions. Z. Phys., 253(2):147–161, 1972.
  12. [12]L. Bertini, A. De Sole, D. Gabrielli, G. Jona-Lasinio, and C. Landim. Macroscopic fluctuation theory. Rev. Mod. Phys., 87:593–636, 2015.DOI
  13. [13]L. Bertini, A. De Sole, D. Gabrielli, G. Jona-Lasinio, and C. Landim. Fluctuations in Stationary Nonequilibrium States of Irreversible Processes. Phys. Rev. Lett., 87:040601, 2001.pubmed.ncbi.nlm.nih.gov/11461605
  14. [14]L. Bertini, A. De Sole, D. Gabrielli, G. Jona-Lasinio, and C. Landim. Macroscopic fluctuation theory for stationary non equilibrium states. J. Stat. Phys., 107(3/4):635–675, 2002.
  15. [15]W. Bryc. A remark on the connection between the large deviation principle and the central limit theorem. Statist. Probab. Lett., 18:253–256, 1993.DOI
  16. [16]C. Maes, K. Netočný, and B. Wynants. Steady state statistics of driven diffusions. Physica A, 387(12):2675–2689, 2008.DOI
  17. [17]C. Maes, K. Netočný, and B. Wynants. On and beyond entropy production: the case of Markov jump processes. Markov Proc. Relat. Fields, 14:445–464, 2008.arxiv.org/abs/0709.4327
  18. [18]M. Lax. Fluctuations from the nonequilibrium steady state. Rev. Mod. Phys., 32:25–64, 1960.DOI
  19. [19]J. Keizer. Statistical Thermodynamics of Nonequilibrium Processes. Springer, New York, 1987.DOI
  20. [20]R. Graham. Covariant formulation of non-equilibrium statistical thermodynamics. Z. Phys. B, 26:397–405, 1977.DOI
  21. [21]R. Graham and T. Tél. Existence of a potential for dissipative dynamical systems. Phys. Rev. Lett., 52:9–12, 1984.DOI
  22. [22]R. Graham and T. Tél. Nonequilibrium potential for coexisting attractors. Phys. Rev. A, 33:1322–1337, 1986.pubmed.ncbi.nlm.nih.gov/9896751
  23. [23]C. Kwon, P. Ao, and D. J. Thouless. Structure of stochastic dynamics near fixed points. Proc. Natl. Acad. Sci. USA, 102:13029–13033, 2005.DOI
  24. [24]L. D. Landau and E. M. Lifshitz. Fluid Mechanics. Pergamon, Oxford, 2nd edition, 1987. Chap. XVII.
  25. [25]G. Nicolis and I. Prigogine. Self-Organization in Nonequilibrium Systems. Wiley, New York, 1977.
  26. [26]I. Prigogine and R. Lefever. Symmetry breaking instabilities in dissipative systems. II. J. Chem. Phys., 48:1695–1700, 1968.DOI
  27. [27]P. Gaspard. Correlation time of mesoscopic chemical clocks. J. Chem. Phys., 117:8905–8916, 2002.DOI
  28. [28]A. C. Barato and U. Seifert. Coherence of biochemical oscillations is bounded by driving force and network topology. Phys. Rev. E, 95:062409, 2017.DOI
  29. [29]B. Doyon, G. Perfetto, T. Sasamoto, and T. Yoshimura. Ballistic macroscopic fluctuation theory. SciPost Phys., 15:136, 2023.DOI
  30. [30]L. H. Jensen. Large deviations of the asymmetric simple exclusion process in one dimension. PhD thesis, New York University, 2000.
  31. [31]S. R. S. Varadhan. Large deviations for the asymmetric simple exclusion process. In Tadahisa Funaki and Hirofumi Osada, editors, Stochastic Analysis on Large Scale Interacting Systems, volume 39 of Advanced Studies in Pure Mathematics, pages 1–27. Mathematical Society of Japan, Tokyo, 2004.
  32. [32]B. Derrida and J. L. Lebowitz. Exact large deviation function in the asymmetric exclusion process. Phys. Rev. Lett., 80:209–213, 1998.DOI
  33. [33]C. W. Gardiner. Handbook of stochastic methods for physics, chemistry, and the natural sciences. Springer, 1985.

Paper details

vlnc:W_z5y558hz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents