Information-Geometric Signatures of Nonconservative Driving Preprint: APS/123-QED

Andrea Auconi, Sosuke Ito · 2026

Read Information-Geometric Signatures of Nonconservative Driving Preprint: APS/123-QED by Andrea Auconi, Sosuke Ito (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-03757.txt.

Introduction

Andrea Auconi and Sosuke Ito propose an information-geometric signature that detects when a system is driven by nonconservative forces, breaking detailed balance. Detailed balance means every microscopic transition is matched by its reverse at equilibrium; its violation signals a nonequilibrium steady state. Their tools are the Kullback-Leibler divergence and the Fisher information.

For Markov jump processes in detailed balance, the acceleration of the divergence near equilibrium equals twice the Fisher information with respect to time. Relaxation toward a nonequilibrium steady state generally breaks that relation, and the discrepancy, called the relaxation gap, lower-bounds the steady-state entropy production rate. The bound is especially tight for simple cyclic networks, and analogous results hold for Fokker-Planck dynamics.

Catalog record

CREATORAndrea Auconi, Sosuke Ito
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.03757

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Information-Geometric Signatures of Nonconservative Driving Preprint: APS/123-QED Andrea Auconi, Sosuke Ito Abstract We propose an information-geometric signature of nonconservative driving that detects violations of detailed balance using the Kullback–Leibler divergence and the Fisher information. For Markov jump processes satisfying detailed balance, we show that, near equilibrium, the acceleration of the Kullback–Leibler divergence relative to the equilibrium state is given by twice the Fisher information with respect to time. In contrast, for relaxation toward a nonequilibrium steady state, this relation is generally violated even near the steady state. We refer to the resulting discrepancy as the relaxation gap and derive a lower bound on the steady-state entropy production rate in terms of this gap. We demonstrate that this bound is particularly tight for networks with simple cyclic topologies. Finally, we show that analogous relations and bounds hold for Fokker–Planck dynamics. Determining whether stochastic dynamics satisfies detailed balance is fundamental for characterizing nonequilibrium behavior. When detailed balance is broken, nonconservative driving can sustain probability currents even in the nonequilibrium steady state (NESS), leading to a positive entropy production rate that quantifies dissipation 37 ; 38 . Systems without detailed balance have been studied in terms of oscillatory behavior 37 ; 5 ; 32 ; 45 ; 47 , relaxation speed 13 ; 7 ; 44 ; 42 ; 19 ; 24 ; 40 ; 26 ; 25 , response 35 ; 29 ; 34 ; 2 , and stability 16 ; 36 ; 15 ; 10 ; 37 ; 27 ; 22 ; 43 ; 3 . In particular, in Markov-chain Monte Carlo methods, dynamics without detailed balance are known to accelerate convergence to a target distribution 7 ; 44 ; 42 ; 19 . Thermodynamic trade-off relations 26 inspired by thermodynamic uncertainty relations 4 ; 18 have also motivated studies on the connection between relaxation speed and steady-state dissipation. Previous studies have often investigated the effects of detailed balance violation by fixing the steady-state distribution and comparing conservative dynamics satisfying detailed balance with nonconservative dynamics violating it 7 . While these approaches clarify how nonequilibrium currents modify relaxation, they do not directly provide a criterion for detecting detailed balance violation from the relaxation behavior of a single system. Here, we develop such a criterion from an information-geometric perspective 1 ; 9 ; 41 ; 23 . Information geometry characterizes dynamical changes in probability distributions through quantities such as the intrinsic speed 21 ; 20 ; 31 , defined as the square root of the Fisher information with respect to time, and the Kullback–Leibler divergence 36 ; 1 relative to the stationary distribution. Although these quantities have been used to analyze the stability of NESSs 22 and formulate thermodynamic trade-off relations near NESSs 3 , how they are constrained by detailed balance during relaxation remains unclear. We show that, for Markov jump processes satisfying detailed balance, the acceleration of the Kullback–Leibler divergence relative to the equilibrium state is given by twice the squared intrinsic speed near equilibrium. In contrast, for relaxation toward the NESS, this relation is generally violated, and we refer to the resulting discrepancy as the relaxation gap. We further derive, near the NESS, a bound on the relaxation gap in terms of the NESS entropy production rate, linking this information-geometric signature of nonconservative driving to the dissipation required to maintain the NESS. Our results apply to both discrete-state Markov jump processes and continuous-state Fokker–Planck dynamics. We also show that the bound can be fully saturated in networks with simple cyclic topologies, confirming its tightness. Setup. Consider a system with n allowed microstates or nodes, and denote by \bm{p}\equiv\left[p_{1}(t),...,p_{n}(t)\right] the probability vector in this discrete space. We consider its time evolution as a Markov jump process (MJP) 46 , d_{t}p_{i}=\sum_{j}M_{ij}p_{j}, (1) which we assume to be ergodic 46 , where M_{ij}\geq 0 is the constant transition rate from node j to node i(\neq j) , -M_{ii}\equiv\sum_{j(\neq i)}M_{ji}>0 is the escape rate of node i , and all sums over j run from 1 to n . In matrix notation, this is written as d_{t}\bm{p}=M\bm{p} . We assume transitions to be locally reversible, meaning M_{ji}>0 if M_{ij}>0 . The anti-symmetric net current on the edge j\rightarrow i is defined as J_{ij}\equiv M_{ij}p_{j}-M_{ji}p_{i} , and the dynamical activity 28 is defined as A_{ij}\equiv M_{ij}p_{j}+M_{ji}p_{i} , which leads to the expression d_{t}p_{i}=(1/2)\sum_{j}(A_{ij}+J_{ij}) . Let us also define the thermodynamic force, F_{ij}\equiv\ln[(M_{ij}p_{j})/(M_{ji}p_{i})] , and note the relation A_{ij}=J_{ij}\coth(F_{ij}/2) valid for A_{ij}>0 ( i\neq j ). Denote by \bm{p}^{*} the NESS distribution, which is defined by d_{t}\bm{p}|_{\bm{p}=\bm{p}^{*}}=\bm{0} , and whose uniqueness is guaranteed by the ergodicity of the dynamics through the Perron-Frobenius theorem 46 . Explicitly, from Eq. ( 1 ), the steady state condition is written as \sum_{j}M_{ij}p^{*}_{j}=0 . The NESS entropy production rate 37 ; 38 is defined as \sigma^{*}\equiv\frac{1}{2}\sum_{i,j}J^{*}_{ij}F^{*}_{ij}, (2) where the superscript ∗ denotes quantities evaluated at the steady state (e.g., J_{ij}^{*}=J_{ij}|_{\bm{p}=\bm{p}^{*}} and F_{ij}^{*}=F_{ij}|_{\bm{p}=\bm{p}^{*}} ) and therefore constant in time. The detailed balance condition, \forall(i,j) , J^{*}_{ij}=0 , corresponds to the equilibrium case \sigma^{*}=0 . Let us further define averages with respect to the steady-state vector as \langle f\rangle\equiv\sum_{i}p^{*}_{i}f_{i} for any test vector \bm{f}=[f_{1},\dots,f_{n}] . Small perturbations. Let us define the perturbation vector \bm{\phi}=[\phi_{1},\dots,\phi_{n}] around the steady state as p_{i}=p^{*}_{i}e^{\phi_{i}}, (3) with normalization \langle e^{\phi}\rangle=1 . Assume this perturbation to be small, meaning \max_{i}|\phi_{i}|\ll 1 , so that its dynamics [Eq. ( 1 )] can be linearized to d_{t}\phi_{i}=\frac{1}{2p^{*}_{i}}\sum_{j}\left(A^{*}_{ij}+J^{*}_{ij}\right)\phi_{j}, (4) where we used \sum_{j}M_{ij}p^{*}_{j}=0 , see the Supplementary Materials (SM) for details of the calculations. This expression explicitly decomposes the linearized dynamics into equilibrium and nonequilibrium parts. The Kullback-Leibler (KL) divergence 1 of a distribution \bm{p} relative to another distribution \bm{q} is defined as D[\bm{p}||\bm{q}]\equiv\sum_{i}p_{i}\ln\left({p_{i}}/{q_{i}}\right) . This quantity D[\bm{p}||\bm{p}^{*}] is zero only when \bm{p}=\bm{p}^{*} . In the weak-perturbation limit \max_{i}|\phi_{i}|\rightarrow 0 , considered hereafter and denoted by \bm{p}\simeq\bm{p}^{*} , we obtain D[\bm{p}||\bm{p}^{*}]=-\langle\phi\rangle=\frac{1}{2}\langle\phi^{2}\rangle\geq 0, (5) where the normalization implies \langle\phi\rangle=-\langle\phi^{2}\rangle/2+\mathcal{O}(\langle\phi^{3}\rangle) . Let us study the stability of the NESS from the time evolution of the divergence, d_{t}D[\bm{p}||\bm{p}^{*}]=\frac{1}{2}\sum_{i,j}A^{*}_{ij}\phi_{i}\phi_{j}\leq 0, (6) where we used the symmetry J^{*}_{ij}=-J^{*}_{ji} , and the inequality follows from the steady-state dynamical activity A^{*} being the negative of a weighted graph Laplacian matrix which is known to be negative-semidefinite 8 . Indeed, it is symmetric A^{*}_{ij}=A^{*}_{ji} , and the steady-state condition implies that the sum over any row or column is zero, \sum_{i}A^{*}_{ij}=\sum_{j}A^{*}_{ij}=0 . Eq. ( 6 ) means that, by evaluating the dynamics at second order around the steady state [Eq. ( 5 )], the Glansdorff-Prigogine criterion for stability 15 ; 37 ; 27 ; 22 is satisfied. This establishes D[\bm{p}||\bm{p}^{*}] as a valid Lyapunov function whose monotonic decrease governs the approach to the steady state 36 ; 10 . Furthermore, the nonnegative quantity -d_{t}D[\bm{p}||\bm{p}^{*}](\geq 0) is known as the Hatano-Sasa excess (or nonadiabatic) entropy production rate 17 ; 14 ; 11 . Intrinsic speed and relaxation. We first consider the intrinsic speed of the dynamics v_{\rm info}(\bm{p}) . The Fisher information metric 1 with respect to time defines the squared intrinsic speed 21 ; 20 , \left(v_{\rm info}(\bm{p})\right)^{2}\equiv\sum_{i}\frac{(d_{t}p_{i})^{2}}{p_{i}}=\sum_{i}p_{i}(d_{t}\ln p_{i})^{2}\geq 0. (7) This intrinsic speed is also given by (v_{\rm info}(\bm{p}(t)))^{2}=d_{\tau}^{2}D[\bm{p}(t+\tau)||\bm{p}(t)]|_{\tau=0} . The expression near the steady state \bm{p}\simeq\bm{p}^{*} is given by \left(v_{\rm info}(\bm{p})\right)^{2}=\langle(d_{t}\phi)^{2}\rangle. (8) Figure 1: (a) Schematic

Read the complete text