Minimum Entropy Production State (MEPS)

• Minimum Entropy Production State (MEPS) is defined as the state or protocol that minimizes entropy production under specified nonequilibrium constraints, with applications ranging from linear irreversible thermodynamics to stochastic systems.
• It employs a variational framework that uses network formulations, path-integral methods, and optimal transport metrics to determine the least-dissipative evolution of a system.
• MEPS offers practical insights across domains such as chemical kinetics, active matter, quantum control, and plasma physics, while emphasizing the dependency on modeling assumptions and fixed resource constraints.

Minimum Entropy Production State (MEPS) denotes a state, steady distribution, or finite-time protocol that minimizes entropy production subject to the constraints of a specified nonequilibrium model. In classical linear irreversible thermodynamics, MEPS is the steady state selected under fixed thermodynamic constraints near equilibrium. In contemporary stochastic thermodynamics, the same term has been extended to discrete-state Markov jump processes, optimal-transport formulations, anisotropic Langevin systems, coupled information-thermodynamic games, active matter, and information-geometric quantum evolutions. The common structure is variational: entropy production is treated as the quantity minimized over admissible currents, forces, controls, or probability paths, but the precise meaning of “state,” the admissible constraints, and even the validity of minimization depend strongly on regime and modeling assumptions (Mohite et al., 2 Nov 2025).

1. Classical meaning and network formulation

In the classical Onsager–Prigogine setting, the entropy production rate is a bilinear form of fluxes and thermodynamic forces, $\sigma = \sum_i J_i X_i$, and the minimum entropy production principle applies in the linear, near-equilibrium, time-independent regime under fixed constraints. In that regime, the realized steady state minimizes total entropy production among admissible steady states, or equivalently minimizes the corresponding quadratic dissipation functional when Onsager reciprocity holds (Niven et al., 2014).

A network-theoretic reformulation identifies the macroscopic constraints explicitly. On a connected finite graph with antisymmetric edge currents $j_e$ and local linear constitutive law $a_e = \ell_e j_e$, the entropy production rate is

$$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$

Steady states satisfy Kirchhoff’s current law, $\partial j^* = 0$, so the steady current lies in the cycle space. Schnakenberg’s cycle affinities

$A^\alpha = \sum_{e\in E} c_e^\alpha a_e$

act as the macroscopic observables that prevent relaxation to equilibrium, and the steady macroscopic currents obey

$A^\alpha = \sum_\beta L^{\alpha\beta} J_\beta.$

Under fixed cycle affinities, the unique entropy-production minimizer is steady and has

$J = L^{-1} A, \qquad \sigma^* = A^\top L^{-1} A.$

In this formulation, MEPS is the steady current configuration determined by the fixed Schnakenberg observables (Polettini, 2011).

For continuous-time master equations, the same logic specializes to

$$j_{ij} = p_j w_{ji} - p_i w_{ij}, \qquad a_{ij} = \ln\frac{w_{ij} p_j}{w_{ji} p_i},$$

with

$\sigma = \frac{1}{2}\sum_{i,j} j_{ij} a_{ij}.$

Near detailed balance, the invariant distribution is a local minimum of $j_e$0 among nearby distributions compatible with the fixed cycle affinities encoded in the rates, thereby recovering a precise Markovian version of Prigogine’s principle (Polettini, 2011).

2. Far-from-equilibrium variational formulation

A far-from-equilibrium formulation of MEPS for discrete-state Markov jump processes is provided by an exact path-integral representation of the transition probability measure in terms of currents $j_e$1, traffics $j_e$2, and effective affinities $j_e$3. For each bidirectional transition $j_e$4, the current and traffic are

$j_e$5

with mobility–affinity parametrization

$j_e$6

and local detailed balance

$j_e$7

The path probability measure is

$j_e$8

with Doi–Peliti action density

$j_e$9

Minimization over $a_e = \ell_e j_e$0 gives

$a_e = \ell_e j_e$1

and the effective non-quadratic dissipation functional

$a_e = \ell_e j_e$2

When the transition affinities are known, $a_e = \ell_e j_e$3 coincides with the Schnakenberg entropy-production integrand $a_e = \ell_e j_e$4 (Mohite et al., 2 Nov 2025).

This leads to a min–max principle,

$a_e = \ell_e j_e$5

which is presented as a Minimum Action Principle for the entropy production rate. Within this framework, MEPS has two forms. A steady-state MEPS is the most-likely saddle-path solution minimizing $a_e = \ell_e j_e$6 under the imposed constraints. A time-dependent MEPS is the protocol minimizing the total entropy production between fixed endpoints over a fixed duration, subject to local detailed balance and admissible dynamics (Mohite et al., 2 Nov 2025).

The same work shows that the exact far-from-equilibrium dissipation functional is non-quadratic, while its close-to-equilibrium limit recovers the Onsager–Machlup form:

$a_e = \ell_e j_e$7

Thus, the classical quadratic MEPS emerges as the small-affinity limit of a broader non-quadratic variational structure (Mohite et al., 2 Nov 2025).

3. Geometric structure, thermodynamic length, and transport metrics

In the far-from-equilibrium Markov setting, thermodynamic length appears directly in current space. If

$a_e = \ell_e j_e$8

then the finite-time entropy production satisfies

$a_e = \ell_e j_e$9

For controlled non-conservative affinities $$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$0, the driving metric is

$$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$1

and the slow-driving minimum is obtained by linear interpolation in the arc-length coordinate $$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$2:

$$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$3

The exact finite-time optimum is

$$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$4

with total entropy production

$$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$5

A central conclusion is that discontinuous endpoint jumps are generic, model-independent features of finite-time optimal driving and reduce entropy production (Mohite et al., 2 Nov 2025).

A related but differently constrained Markovian formulation considers continuous-time Markov jump processes at fixed activity. There, without additional constraints, a given time evolution can be realized at arbitrarily small entropy production at the expense of diverging activity. At fixed activity, however, the entropy-production minimizer is realized by conservative forces, and the minimum is governed by the graph-based Wasserstein distance $$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$6:

$$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$7

where $$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$8 is the prescribed total activity. In this setting, MEPS is the conservative, detailed-balance dynamics minimizing entropy production under fixed activity and endpoint constraints (Dechant, 2021).

For overdamped Langevin dynamics in anisotropic temperature fields, the geometric picture is altered by the metric tensor $$\sigma = \sum_{e\in E} j_e a_e = \sum_{e\in E} \ell_e j_e^2.$$9. Under full actuation, minimal entropy production is again an optimal-transport action,

$\partial j^* = 0$0

so finite-time minimal entropy production is expressed by a weighted Wasserstein geometry. Under conservative-only actuation, anisotropy produces an intrinsic housekeeping contribution associated with circulating currents, and entropy production may remain strictly positive even when the probability density is unchanged. In that sense, a steady MEPS in anisotropic environments need not be a zero-dissipation state (Miangolarra et al., 2023).

4. Domain-specific generalizations

In information thermodynamics, MEPS has been formulated for a bipartite Markov jump system controlled by two players, each minimizing its own partial entropy production plus a penalty for failing to achieve a target transition. In the linear irreversible thermodynamics regime, the Nash equilibrium yields a trade-off:

$\partial j^* = 0$1

with $\partial j^* = 0$2. The total entropy production is minimized when the penalty is equally shared,

$\partial j^* = 0$3

Here, MEPS is not a single subsystem optimum but the equal-penalty-sharing Nash equilibrium for the coupled task (Fujimoto et al., 2021).

In magnetically confined plasmas, MEPS appears in a hybrid construction with maximum entropy. The local equilibrium or reference state is fixed by imposing the minimum entropy production theorem on a subset of Prigogine-type fluctuations $\partial j^* = 0$4, for which entropy production is quadratic, together with MaxEnt under scale-invariant restrictions on the remaining variables. For the energy variable $\partial j^* = 0$5, the resulting reference distribution is the Gamma law

$\partial j^* = 0$6

while the full reference distribution multiplies this factor by Gaussian terms in the Prigogine variables. In this literature, MEPS is therefore a local-equilibrium reference state rather than merely a current configuration (Sonnino et al., 2013).

In chemical reaction networks, MEPS is tied to thermodynamic equilibrium rather than to arbitrary nonequilibrium steady states. The entropy production rate is proportional to the square of the reaction velocity only around equilibrium, whereas around a generic nonequilibrium steady state it takes the form

$\partial j^* = 0$7

The vanishing of the constant and linear terms is therefore a diagnostic for equilibrium-like minimum-entropy-production behavior. This suggests that the quadratic relation $\partial j^* = 0$8 is a practical marker for the domain of validity of MEPS in chemical kinetics (Banerjee et al., 2013).

For active microswimmers with internal dissipation, MEPS is defined by an exact lower bound on total dissipation, including both external and internal contributions:

$\partial j^* = 0$9

Equality holds when the active flow is a linear superposition of two passive Stokes problems. In this context, MEPS is a hydrodynamic minimum-dissipation state determined by drag matrices and propulsion constraints, not by Prigogine’s near-equilibrium theorem (Daddi-Moussa-Ider et al., 2023).

In information-geometric quantum control, the term is path-based. For pure-state evolutions generated by time-dependent $A^\alpha = \sum_{e\in E} c_e^\alpha a_e$0 Hamiltonians, the Fisher–Rao metric on probability paths defines a geodesic problem, and the minimum-action path between fixed endpoints minimizes total entropy production. MEPS therefore denotes the family of states lying on the geodesic rather than a stationary fixed point (Cafaro et al., 2021).

5. Bounds, saturation properties, and constructive examples

The far-from-equilibrium action principle yields a unified set of bounds. For the total inflow into a state $A^\alpha = \sum_{e\in E} c_e^\alpha a_e$1,

$A^\alpha = \sum_{e\in E} c_e^\alpha a_e$2

which is a non-quadratic speed limit. For vorticity-like observables in state space, the non-quadratic thermodynamic–kinetic uncertainty relation is

$A^\alpha = \sum_{e\in E} c_e^\alpha a_e$3

The same framework also gives the large-deviation rate functional

$A^\alpha = \sum_{e\in E} c_e^\alpha a_e$4

and fluctuation relations

$A^\alpha = \sum_{e\in E} c_e^\alpha a_e$5

Geodesic protocols are singled out because they saturate the speed-limit and thermodynamic–kinetic uncertainty bounds and therefore constitute time-dependent MEPS in the strongest variational sense (Mohite et al., 2 Nov 2025).

Worked examples in the same framework include a two-state model with rates

$A^\alpha = \sum_{e\in E} c_e^\alpha a_e$6

and a three-state driven ring. In both cases the slow-driving geodesic minimizes entropy production through linear interpolation in $A^\alpha = \sum_{e\in E} c_e^\alpha a_e$7, while the exact finite-time optimum exhibits endpoint jumps and obeys $A^\alpha = \sum_{e\in E} c_e^\alpha a_e$8 (Mohite et al., 2 Nov 2025).

A complementary constructive theory for stationary Markov chains with prescribed thermodynamic currents uses cycle decomposition. On a periodic ring of length $A^\alpha = \sum_{e\in E} c_e^\alpha a_e$9 with cycle current $A^\alpha = \sum_\beta L^{\alpha\beta} J_\beta.$0, the minimal entropy production is

$A^\alpha = \sum_\beta L^{\alpha\beta} J_\beta.$1

and the minimizing dynamics is the uniform chain obtained from the optimal cycle weights. More general multi-cycle networks admit explicit minimizing edge weights

$A^\alpha = \sum_\beta L^{\alpha\beta} J_\beta.$2

with $A^\alpha = \sum_\beta L^{\alpha\beta} J_\beta.$3 fixed by normalization. In this literature, MEPS is the stationary Markov dynamics that realizes prescribed thermodynamic currents with the least possible entropy production (Andrieux, 2023).

6. Scope, controversies, and limitations

MEPS is not a universal selection principle. In nonlinear self-organized transport, whether the realized state is a minimum- or maximum-entropy-production state depends on the ensemble of control and on the topology of the transport reorganization. In the boundary-layer plasma models of zonal flow and Bénard convection, series versus parallel organization and force-driven versus flux-driven operation lead respectively to MinEP or MaxEP, and the operative state is selected by a tangent-line condition on a pleated generalized potential rather than by entropy production alone (Kawazura et al., 2011).

Far from equilibrium, even for continuous-time Markov chains, real nonequilibrium steady states generally violate both MINEP and MAXEP. A variational definition of MEPS can still be given by minimizing

$A^\alpha = \sum_\beta L^{\alpha\beta} J_\beta.$4

over normalized distributions, which yields the condition

$A^\alpha = \sum_\beta L^{\alpha\beta} J_\beta.$5

However, numerical sampling shows that the physical steady state need not coincide with this minimizer, even though for large interconnected continuous-time Markov chains the steady-state entropy production tends empirically to approach the minimum as system size increases (Ray et al., 14 Jul 2025).

Constraint dependence is equally central. For Markov jump processes, entropy production can be made arbitrarily small if activity is allowed to diverge, so a meaningful MEPS requires fixing activity, currents, affinities, endpoints, or an equivalent resource budget. With fixed activity, the minimizer is conservative and detailed-balance-like; without that constraint, “minimum entropy production” becomes degenerate (Dechant, 2021).

The main far-from-equilibrium action formulation also has explicit assumptions: Markovianity, ergodicity, local detailed balance, positivity and finiteness of traffics, and admissible controls bounded so that $A^\alpha = \sum_\beta L^{\alpha\beta} J_\beta.$6 remains finite. Non-Markovian or non-local-detailed-balance dynamics fall outside the derivation, and coarse-graining affects inferred entropy production and the tightness of the associated bounds (Mohite et al., 2 Nov 2025).

Taken together, these results suggest that MEPS is best understood not as a universal law but as a family of exact or asymptotic variational statements whose content depends on the space of admissible processes. In near-equilibrium thermodynamics it reproduces Prigogine’s steady-state theorem; in far-from-equilibrium stochastic models it becomes a non-quadratic action principle; and in anisotropic, strategic, active, and quantum settings it remains meaningful only after the relevant control, geometry, and dissipation constraints have been specified (Polettini, 2011).