Local Detailed Balance: Theory & Applications

• Local detailed balance is a fundamental condition that equates transition rate ratios with entropy changes, ensuring thermodynamic consistency in nonequilibrium systems.
• It generalizes global detailed balance by applying a local, pairwise requirement on each elementary transition, suitable for systems with multiple reservoirs and driven dynamics.
• LDB underpins the construction of mesoscopic models, the derivation of fluctuation relations, and computational sampling techniques in stochastic thermodynamics.

Local detailed balance (LDB) is a core constraint on stochastic transition rates or path probabilities, ensuring that entropy production and thermodynamic consistency are maintained at the level of elementary system transitions, even under nonequilibrium stationary conditions. While the classical (global) detailed balance condition implies reversibility for Markovian dynamics invariant under equilibrium, LDB prescribes only a local (pairwise) requirement on each elementary jump, generalizing the structure to nonequilibrium settings with multiple reservoirs, nonconservative forces, or driven dynamics. This principle underpins the construction of mesoscopic models consistent with microscopic thermodynamic laws, the derivation of fluctuation relations, and the design of efficient Markov processes for sampling nonsmooth or discrete target distributions.

1. Mathematical Formulation and Core Concepts

Local detailed balance formalizes the correspondence between the ratio of forward and backward transition probabilities for each elementary event and the corresponding entropy change in the environment. In a Markov jump process with transition rates $k(x,y)$, LDB asserts: $$\log \frac{k(x,y)}{k(y,x)} = \frac{\Delta S(x\to y)}{k_B}$$ where $\Delta S(x\to y)$ is the entropy increment exported to the reservoirs or environment per transition $x\to y$ (Maes, 2020).

For diffusive or jump processes in contact with a heat bath at temperature $T$, the pathwise ratio of probability densities for a forward trajectory $\omega$ and its time-reversed path $\theta\omega$ satisfies: $$k_B\,\log \frac{P[\omega]}{P[\theta\omega]} = \Delta S(\omega)$$ where $\Delta S(\omega)$ denotes the total entropy flow into the environment along $\omega$ (Maes, 2020, Falasco et al., 2021).

In settings with multiple thermal or chemical reservoirs at different intensities, each jump is associated to a specific reservoir, and the corresponding LDB reads: $$\frac{W(C'\leftarrow C)}{W(C\leftarrow C')} = \exp[\Delta S_{\mathrm{exch}}(C\to C')]$$ with $\Delta S_{\mathrm{exch}}$ the exchange entropy, often explicitly $\beta\,\Delta E$ for energy exchange at inverse temperature $\beta$ (Bauer et al., 2014).

LDB is a local (pairwise) constraint; no global symmetry (reversibility) of the generator is required. This distinguishes it from global detailed balance, which demands invariance of the entire process under time-reversal, i.e.,

$\pi(x) k(x,y) = \pi(y) k(y,x)\quad\forall\,x,y$

for stationary measure $\pi$.

2. Physical Foundations and Microscopic Derivation

LDB emerges generically from coarse-graining and weak-coupling limits of microscopic deterministic or Hamiltonian dynamics with a reservoir structure. Microscopically, assume a deterministic map $T$ on the product state space of a system and reservoirs, preserving energy and obeying microreversibility (Bauer et al., 2014). When the system-reservoir coupling is "star-shaped" (only one reservoir involved per elementary event), coarse-graining yields a Markov process over mesostates (system configuration plus reservoir energy).

For sufficiently large reservoirs acting as equilibrium baths, microcanonical detailed balance for the exact microscopic dynamics reduces to LDB for the system alone: $$\frac{W(C'\leftarrow C)}{W(C\leftarrow C')} = \exp\left[ \beta_{a} (\mathcal{E}(C) - \mathcal{E}(C')) \right]$$ for jumps exchanging energy only with reservoir $a$, temperature $T_a=1/\beta_a$. More general exchanges (volume, particles, etc.) enter additively in $\Delta S_{\mathrm{exch}}$ (Bauer et al., 2014).

In the quantum context, LDB is realized when the Lindblad generator admits a stationary reference state (not necessarily the Gibbs state of the bare Hamiltonian) and the jump operators and bath couplings satisfy commutation relations encoding microreversibility with respect to the equilibrium structure imposed by the baths (Barra et al., 2017).

3. LDB in Markov Processes, Coarse-Graining, and Fluctuation Relations

LDB plays an organizing role in the construction and analysis of Markovian models, both for physical and computational systems.

Markov Jump and Diffusion Processes: In both jump and diffusive frameworks, LDB ensures thermodynamic consistency and provides a pathwise generator for entropy production rates: $$S(\omega) = \sum_{\text{jumps } \tau} \log \frac{k(x_{\tau^-},x_\tau)}{k(x_\tau,x_{\tau^-})}$$ with the steady-state rate

$$\sigma = k_B \sum_{x,y} \rho(x) k(x,y) \log \frac{k(x,y)}{k(y,x)}$$

(Maes, 2020).

Fluctuation Relations: LDB is equivalent to central nonequilibrium fluctuation theorems. Once LDB holds, the detailed and integral fluctuation relations follow: $P_t(S = s) / P_t(S = -s) = e^{s / k_B}$

$\langle e^{-S(\omega)/k_B}\rangle = 1$

where $S(\omega)$ is the entropy production along trajectory $\omega$ (Maes, 2020). Breakdown of LDB (e.g., due to hidden currents in coarse-grained states) directly leads to violations of these relations, rendering LDB both necessary and "only half of the story" for full fluctuation-response theory (Maes, 2020, Piephoff et al., 13 Nov 2025).

Coarse-graining: LDB typically persists under adiabatic elimination of fast degrees of freedom in weakly driven systems, provided clear time-scale separation and dissipative (time-antisymmetric) driving dominate. At mesoscopic and macroscopic scales, transition rates $k_{i \to j}$ between basins or macrostates satisfy: $$\log \frac{k_{ji}}{k_{ij}} = -\frac{\mathcal{F}_j - \mathcal{F}_i}{T} + \frac{W_{ji}}{T}$$ with $\mathcal{F}_i$ free energy and $W_{ji}$ nonconservative work along the transition path (Falasco et al., 2021). Violations can occur in the absence of strict separation or if hidden nonequilibrium currents persist in eliminated variables (Hartich et al., 2021, Piephoff et al., 13 Nov 2025).

4. LDB in Stochastic Thermodynamics and Nonequilibrium Steady States

Entropy production: The precise encoding of entropy flow assures that every trajectory segment or jump carries a quantifiable thermodynamic cost or gain, and allows operational assignment of heat and work at the level of stochastic transitions (Maes, 2020, Bauer et al., 2014, Piephoff et al., 13 Nov 2025).

Driven systems and nonequilibrium steady states: For processes with external forces, non-identical reservoirs, or active matter driven by internal degrees of freedom, LDB generalizes to include all relevant contributions: $$\frac{k(x, y)}{k(y, x)} = \exp\left[\beta (\Delta E_{xy} - F \cdot d_{xy})\right]$$ where $F$ is a generalized affinity (e.g., chemical, mechanical) and $d_{xy}$ the conjugate displacement (Hartich et al., 2021, Khodabandehlou et al., 2024). In active matter, LDB is restored by embedding the dynamics in a two-temperature or two-bath scenario, specifying entropy contributions from both the ambient and "hot spot" reservoirs (Khodabandehlou et al., 2024).

Coarse-grained violations and Milestoning: Standard Markovian coarse-graining alone is insufficient to guarantee LDB except at infinite time-scale separation (Hartich et al., 2021). Alternatives such as Milestoning—where transitions are defined based on splitting probabilities between properly chosen boundaries—restore exact LDB and eliminate ambiguities associated with rare recrossings or internal memory effects (Hartich et al., 2021).

5. LDB in Model Construction and Computational Methods

LDB underlies both physical model building and computational schemes for sampling or optimization.

Monte Carlo and Jump Process Sampling: The construction of locally-balanced Markov jump processes (LBMJP) employs LDB to guarantee preservation of any specified (possibly discrete or nonsmooth) target distribution $\pi$. Rather than imposing full global reversibility, the log-ratio condition

$$\log r(x, y) - \log r(y, x) = \log \pi(y) - \log \pi(x)$$

is enforced, with balancing functions $g(t)$ controlling proposal acceptance and ergodicity (Livingstone et al., 17 Apr 2025). This framework enables $\pi$-exact sampling even for discontinuous or degenerate $\pi$, bypassing the smoothness requirements of overdamped Langevin diffusions.

Discretization of Stochastic PDEs: For spatially or temporally discretized stochastic conservation laws, LDB (specifically, the adjoint relationship between drift and noise) is maintained by matching the discrete operators (gradient, divergence, Laplacian) and ensuring the fluctuation-dissipation relation holds at the discrete level. This guarantees the unique steady state is the Gibbs measure at any grid resolution (Banerjee et al., 2017).

Chemical Reaction Networks: In mass-action reaction networks, LDB is defined via zero-order and first-order conditions on the local ratios of forward-to-backward fluxes for each reaction, with implications for the existence of a global stochastic potential $U(x)$. The existence of $U$ facilitates the analysis of metastable transitions and large deviations (Jia et al., 2019).

6. Quantum and Many-Body Generalizations

In quantum settings, quantum local detailed balance generalizes the notion of LDB to Lindblad dynamics, requiring compatibility between system-bath interaction operators and the existence of an effective equilibrium state (often not the Gibbs state of the system Hamiltonian) (Barra et al., 2017). At the many-body quantum level, LDB arises for the dynamics of "slow, coarse" observables, justified via the eigenstate thermalization hypothesis and typicality. The ratio of transition rates between macrostates then reflects their entropy difference: $R_{x,y}/R_{y,x} = \exp[S_B(x) - S_B(y)]$ where $S_B(x) = \log V_x$ is the Boltzmann entropy of macrostate $x$ (Strasberg et al., 2022).

7. Limitations, Violations, and "Completeness"

While LDB ensures the correct antisymmetric (irreversible) structure of transitions and is essential for thermodynamic consistency and entropy accounting, it does not specify the symmetric (frenetic, activity) part of the dynamics, which affects kinetic responses, stationary distributions away from equilibrium, and transport coefficients. LDB alone is necessary but not sufficient for full fluctuation-response theory and macroscopic stochastic thermodynamics (Maes, 2020).

Failures of LDB at the mesoscopic level, as when hidden non-equilibrated degrees of freedom persist after elimination, manifest as systematic deviations in predicted entropy production or fluctuation theorems (Piephoff et al., 13 Nov 2025, Hartich et al., 2021). Restoring LDB requires careful construction of transition statistics (e.g., Milestoning), inclusion of all relevant environmental couplings, or more elaborate hierarchical modeling. The completeness of thermodynamic description thus relies on supplementing LDB with precise dynamical and structural information appropriate to the system scale and observable of interest.