Unified Definition of Entropy Production

• Unified entropy production is a framework that quantitatively measures irreversibility using path probabilities, KL divergences, and information-theoretic metrics.
• It integrates classical and quantum dynamics by applying geometric, variational, and operator methods to ensure compatibility with fluctuation theorems and the second law.
• The approach is broadly applicable to Markov processes, diffusions, quantum channels, and non-Markovian dynamics, providing consistent non-negativity across scales.

A unified definition of entropy production quantitatively characterizes the degree of irreversibility, time-reversal asymmetry, or detailed-balance violation in deterministic or stochastic dynamics across classical and quantum systems, including Markov processes, diffusions, quantum channels, mesoscopic models, and information-theoretic transformations. Rigorous unified frameworks equate entropy production to geometric, variational, information-theoretic, and operator-theoretic objects—such as Kullback–Leibler divergences on trajectory spaces, relative entropies between states or path measures, generalized gradient flows, and minimizations on information-geometric manifolds—ensuring consistency with fluctuation theorems, non-negativity, and the second law far from equilibrium.

1. Entropy Production as Kullback–Leibler Divergence on Path Space

The foundational unifying principle is that entropy production quantifies the log-likelihood ratio (or Kullback–Leibler divergence) between forward and appropriately defined reversed path measures. In both classical stochastic and quantum Markovian settings, this yields the general formula

$$\Sigma \equiv \ln\frac{P_{\rm fwd}(\omega)}{P_{\rm bwd}(\omega)}$$

where $P_{\rm fwd}$ and $P_{\rm bwd}$ are, respectively, the probability measures over system trajectories under the forward and backward dynamics. The mean entropy production is the average over forward paths: $$\langle\Sigma\rangle = D_{\mathrm{KL}}(P_{\rm fwd} \| P_{\rm bwd}) \geq 0$$ This structure directly underpins all forms of the fluctuation theorems—integral and detailed—and connects entropy production to the arrow of time and deviation from time-reversal symmetry (Kwon, 2015Cocconi et al., 2020Costa et al., 2022Yang et al., 2019).

Discrete- and continuous-time Markov process settings admit explicit forms:

• Discrete Markov chains:

$$e_{d} = \sum_{i,j} \pi_{i} P_{ij} \ln \frac{\pi_{i}P_{ij}}{\pi_{j}P_{ji}}$$

• Diffusions:

$$e_{c} = \int \rho(x) \frac{1}{2}[b(x) - b^{-}(x)]^{T}\Sigma^{\dagger}(x)[b(x) - b^{-}(x)]~dx$$

where $b^{-}$ is the time-reversed drift, $\Sigma^\dagger$ is the Moore–Penrose inverse of the diffusion tensor, and $\rho(x)$ is the stationary distribution (Costa et al., 2022).

For quantum Markov semigroups, the entropy production rate is

$$\sigma(\rho) = \lim_{\epsilon\to0}~\dfrac{1}{\epsilon} D\bigl(T^*_\epsilon(\rho) \| T^*_\epsilon{}^{\sharp}(\rho)\bigr)$$

where $T^*_\epsilon$ is the predual map and $T^*_\epsilon{}^{\sharp}$ its KMS-dual (Fagnola et al., 2012).

2. Geometric, Variational, and Projective Structures

Unified geometric approaches further identify entropy production as the squared norm of the generalized thermodynamic “force” under an appropriate metric or as a projection in information geometry. The Steepest Entropy Ascent (SEA) principle yields the metric-unifying formula: $$\sigma = (\Lambda\,|\,g^{-1}\,|\,\Lambda) \ge 0,\qquad\Lambda \equiv \nabla S - \sum_i \beta_i\nabla C_i$$ where $\nabla S$ is the entropy gradient in state space, $\beta_i$ are Lagrange multipliers enforcing conservation laws, and $g(ρ)$ is a positive-definite metric generalizing Onsager resistivity arbitrarily far from equilibrium (Beretta, 2014). The SEA formalism unifies Onsager theory, GENERIC, BGK-type kinetic models, mesoscopic Fokker–Planck models, and quantum SEA evolutions.

The information-geometric unification establishes that entropy production is the m-projection distance (minimal relative entropy) from the true path distribution to the manifold of reversible (backward) dynamics: $\sigma = \min_{q \in \mathcal{M}_R} D(p_S \| q)$ where $\mathcal{M}_R$ encodes the reversibility constraint, and $D(\cdot\|\cdot)$ is the KL divergence (Ito et al., 2018).

3. Quantum Unified Definitions and Operator Frameworks

In quantum thermodynamic and information-processing settings, unified definitions exploit relative entropy and operator statistics. For any CPTP (completely positive trace-preserving) process $\mathcal{E}$, quantum priors $\gamma$, and reference state $\tau$, the average entropy production is

$$\Sigma = D(\rho \| \gamma) - D(\mathcal{E}(\rho) \| \mathcal{E}(\gamma)) + D(\mathcal{E}(\rho)\|\tau)$$

where $D$ is Umegaki’s quantum relative entropy (Bai et al., 2024). The fully quantum entropy production operator is constructed from the Choi operator and two-time quantum states,

$$\Sigma = \log \left( \sqrt{Q_F} (Q_R^\gamma)^{-1} \sqrt{Q_F} \right)$$

with positive, Hermitian spectrum and second law $\langle\Sigma\rangle \ge 0$.

System-reservoir and explicit quantum dynamics yield a manifestly non-negative entropy production as the increase in distinguishability (relative entropy) of the full state from the product of its marginals or equilibrium reference,

$$\Delta_i S(t) = D[\rho(t) \| \rho_s(t) \otimes \rho_R^{\rm eq}] \ge 0$$

(0908.1125).

Unified functionals extend to finite quantum systems lacking a unique temperature, via “energy-matching” effective temperatures,

$$\Delta\Sigma = D\left(\rho_{SE}(\tau)\|\rho_S(\tau) \otimes \gamma_E(\beta_\tau)\right) - D\left(\rho_{SE}(0)\|\rho_S(0) \otimes \gamma_E(\beta_0)\right)$$

with decomposition into a Clausius term and a $\dot\beta$-driven correction (Nishiyama et al., 2 Feb 2026).

4. Entropy Production in Multiscale, Non-Markovian, and Generalized Kinetics

For multi-time quantum processes—including Markovian, unitary, and non-Markovian evolutions—the unified entropy production per trajectory is given by the log-ratio of joint probabilities for forward and backward Choi states, accommodating Petz recovery in non-unitary steps,

$$\Sigma(x_{n:1}) = \ln \frac{\mathcal{P}_{n:1}(x_{n:1})}{\mathcal{P}^{\rm tr}_{1:n}(x_{1:n})}$$

Non-Markovian dynamics may generate negative entropy production rates on reduced subsystems, with strict additivity and FT validity in the Markovian or closed case (Huang, 2023).

Complex non-Gaussian active matter is encompassed by a probability-flow equivalence construction whereby the entropy production rate is expressed in terms of probability currents and explicit scores, remaining non-negative and satisfying generalizations of the fluctuation theorem regardless of initial condition or statistical drive (Huang et al., 9 Apr 2025).

5. Information-Theoretic and Maximum-Entropy Unifications

Jaynes' Maximum Entropy Principle-based unification posits that for incomplete measurement or tomographically incomplete protocols, the relevant entropy production is the quantum relative entropy between the true state and the least-biased (maximum-entropy) state compatible with observed constraints,

$\Sigma = S(\rho \|\varrho_{\rm max-S})$

where $\varrho_{\rm max-S}$ is the maximum-entropy state compatible with measured observables and channel outputs (Varizi et al., 2024). This construction recovers classical, diagonal, observational, mutual-information, and standard relative-entropy-based entropy productions as specializations.

The CPM operator formalism translates all entropy productions to Radon–Nikodym derivatives of measure transformations, generalizing forms of the fluctuation and detailed fluctuation theorems, and aligning total, non-adiabatic, and housekeeping EPs under a single analytic umbrella (Yang et al., 2019).

6. Recovery of Known Theories and Implications

Unified definitions systematically recover:

• Schnakenberg’s formula for discrete Markov processes, including the case with unidirectional transitions by means of “lifted” bidirectional networks with diverging auxiliary rates (Busiello et al., 2019).
• The SEA construction for spatially and compositionally heterogeneous non-equilibrium settings, linking with Onsager’s theory and GENERIC formalism (Beretta, 2014).
• Quantum entropy production as a commutator norm in GKSL Lindblad form, vanishing precisely at quantum detailed balance (Fagnola et al., 2012).
• Pathwise fluctuation theorems and integral FT via path probabilities in both jump and diffusion settings (Kwon, 2015, Costa et al., 2022).

7. Interpretation, Limitations, and Extensions

All coherent definitions of entropy production are fundamentally measures of dynamical irreversibility: the statistical or geometric “distance” between forward and time-reversed (or model-reversed, protocol-reversed, dual, or CPM-transformed) trajectories in state or operator space. They ensure compatibility with fluctuation theorems, satisfy non-negativity (with prescribed prerequisites on initial, final, or reference data), and in the presence of certain symmetries reduce to standard forms of the second law.

Generalizations relevant to active matter, non-Markovianity, quantum coherence, and informational resource theories are tractable within these unified frameworks, though specific subtleties—such as the role of effective temperatures in small quantum environments (Nishiyama et al., 2 Feb 2026), the breakdown of marginal fluctuation relations under memory effects (Huang, 2023), or the geometrical meaning of non-additive stochastic interaction (Ito et al., 2018)—demand careful interpretation.

---

Key Unified Formulations: A Comparative Table

Framework	Unified Definition & Main Object	Scope/Features
Pathwise/Kullback–Leibler	$$\langle\Sigma\rangle = D_{\mathrm{KL}}(P_{\rm fwd}\|P_{\rm bwd})$$	Classical/quantum, Markov/non-Markov
SEA Geometric Metric	$\sigma = (\Lambda|g^{-1}|\Lambda)$	Arbitrary state spaces, constraints
Relative Entropy (Quantum)	$D(\rho\|\sigma)$, or $S(\rho\|\varrho_{\rm max-S})$	All quantum processes, MEP-based
Information geometric	$\sigma = \min_{q \in \mathcal{M}_R} D(p_S \| q)$	Classical, subsystems, projections
CPM/Operator	$$\Delta S_\nu(\omega) = \log \frac{dP}{dP^\nu}(\omega)$$	Martingale/FT, arbitrary reversals

These frameworks guarantee broad applicability, rigorous non-negativity, and compatibility with fluctuation relations, forming the modern backbone of entropy production theory in nonequilibrium thermodynamics (1409.66722412.12489Huang, 2023Costa et al., 2022Ito et al., 2018Varizi et al., 2024).