Wasserstein-2 gradient flows and the geometry of entropy production in classical and quantum stochastic thermodynamics

arXiv:2606.00698 · cond-mat.stat-mech, quant-ph · Submitted 2026-05-30 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Wasserstein-2 gradient flows and the geometry of entropy production in classical and quantum stochastic thermodynamics".

Mira: Here's the gist of the paper. Wasserstein-2 gradient flows and the geometry of entropy production in classical and quantum stochastic thermodynamics provide a unified geometric description of entropy production in classical and quantum thermodynamics based on Wasserstein-2 structures arising from gradient flows of free energy.

Kai: This framework extends the Riemannian program of thermodynamics further from equilibrium by introducing generalized Wasserstein-2 metrics that incorporate conservative (Hamiltonian) dynamics, yielding intrinsic distances that exactly characterize minimal entropy production under fixed dissipative mobilities.

Mira: Geometric Formulation of Gradient Flows The gradient flow of a free energy functional F on a Riemannian manifold (M, g) is determined by the equation x˙ (t) = −Kx(t)(dFx(t)).

Kai: This flow links thermodynamic forces to fluxes through the Onsager operator Kx = G−1x: T∗xM → TxM.

Paper summary: Kai: So to recap this paper, its main thrust is providing one unified geometric description for entropy production across classical and quantum thermodynamics by basing it on Wasserstein-two structures from gradient flows of free energy <ref:2606.00698#pg1,from gradient flows of free energy>.

Mira: They are essentially showing how the second law of thermodynamics imposes a specific geometry on nonequilibrium transformations, and they use these Wasserstein-two distances to quantify the minimal entropy production under fixed dissipative mobilities <ref:2606.00698#pg1>.

Lev: So, the core claim is that this geometric framework lets you find intrinsic distances that exactly characterize this minimal entropy production when you fix how much dissipation happens in a given system.

Kai: They cover a wide range of systems, from overdamped diffusions and discrete detailed-balanced Markov chains to dissipative Lindblad dynamics for quantum systems.

Mira: And they show how relaxation to equilibrium in all these settings can be formulated as a specific type of gradient flow, which is the foundation for defining the Wasserstein-two distances themselves <ref:2606.00698#pg1,how relaxation to equilibrium in>.

Lev: It’s important because it extends the Riemannian program of thermodynamics beyond just equilibrium states by incorporating conservative Hamiltonian dynamics into mixed systems.

Kai: Because they introduce this W sub h distance for mixed systems, they can compare purely dissipative and mixed flows using these new metrics to understand the role of coherence.

Conclusion: Mira: Thinking about this work, the title itself points to a deep connection between geometry, transport theory, and fundamental thermodynamics. It’s about moving beyond just calculating entropy production to understanding it as a geometric cost problem.

Kai: It really is. The authors managed to create a framework that applies across classical mechanics, quantum mechanics, and even these discrete systems we deal with in computation.

Lev: For someone building error correction hardware, the implication is that this geometry gives us a way to design protocols that are provably optimal in terms of minimizing entropy production during operation.

Mira: It suggests that understanding the structure of the free energy landscape—the manifold and the metric—is key to finding physically meaningful control protocols, not just looking at simple rate equations.

Kai: And ultimately, it provides a geometric way to derive thermodynamic uncertainty relations that are applicable in short times for both purely dissipative and mixed systems.

Lev: So we get a set of bounds on how much entropy you can produce quickly based on the structure of the system's dynamics, which is useful for setting realistic constraints.

Mira: It frames entropy production not just as something that happens, but as a quantity that has an underlying geometric cost derived from the free energy landscape.

Olga Movilla Miangolarra, Ralph Sabbagh, Artemy Kolchinsky

Departamento de Fósica, Universidad de La Laguna · Instituto Universitario de Estudios Avanzados (IUdEA) · Department of Mechanical and Aerospace Engineering, University of California, Irvine · ICREA-Complex Systems Lab, Pompeu Fabra University · Universal Biology Institute, Graduate School of Science, The University of Tokyo

cond-mat.stat-mech, quant-ph

Submitted: 2026-05-30

Updated: 2026-11-12

Comments: 30 pages 5 figures

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

Importance score: 84/100

The gist: Wasserstein-2 gradient flows and the geometry of entropy production in classical and quantum stochastic thermodynamics provide a unified geometric description of entropy production in classical and…

Key concepts

Wasserstein-2 Distance
This distance measures the cost of transforming one probability distribution into another. In this context, it quantifies the total dissipation or work required to move from an initial state to a final state in both classical and quantum systems.
Gradient Flow
The gradient flow describes how a system evolves toward equilibrium by minimizing a free energy functional. It links thermodynamic forces (like temperature differences) directly to the fluxes (like heat flow) through an Onsager operator, defining the direction of irreversible processes.
Hamiltonian-Dissipative Geometries
This framework introduces a new distance, W2,h, that accounts for both dissipative and conservative dynamics. By comparing it to the standard classical Wasserstein-2 distance, it shows how coherent (Hamiltonian) motion can reduce the minimum entropy production cost.
Thermodynamic Uncertainty Relations (TUR)
These relations are derived from the geometry of these metrics and provide bounds on entropy production. They show that for short times, there are fundamental limits on how small the total entropy change can be, regardless of whether the system is purely dissipative or mixed.

Terminology

Summary

Wasserstein-2 gradient flows and the geometry of entropy production in classical and quantum stochastic thermodynamics provide a unified geometric description of entropy production in classical and quantum thermodynamics based on Wasserstein-2 structures arising from gradient flows of free energy. p. 0

Where the paper says it

Wasserstein-2 gradient flows and the geometry of entropy production in classical and quantum stochastic thermodynamics Olga Movilla Miangolarra,1, 2, ∗ Ralph Sabbagh,3 and Artemy Kolchinsky4, 5 1Departamento de F´ısica, Universidad de La Laguna, La Laguna 38203, Spain 2 Instituto Universitario de Estudios Avanzados (IUdEA), Universidad de La Laguna, La Laguna 38203, Spain 3Department of Mechanical and Aerospace Engineering, University of California, Irvine, California 92697, USA 4 ICREA-Complex Systems Lab, Pompeu Fabra University, 08003 Barcelona, Spain 5Universal Biology Institute, Graduate School of Science, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan The second law does more than set the direction of thermodynamic evolution: it endows nonequilibrium transformations with an underlying geometry. In this work, we provide a unified geometric description of entropy production in classical and quantum thermodynamics based on Wasserstein-2 structures arising from gradient flows of free energy.

This framework extends the Riemannian program of thermodynamics further from equilibrium by introducing generalized Wasserstein-2 metrics that incorporate conservative (Hamiltonian) dynamics, yielding intrinsic distances that exactly characterize minimal entropy production under fixed dissipative mobilities. p. 0
Where the paper says it

The associated Wasserstein-2 distance bounds entropy production, yielding a finite-time refinement of the second law. We extend this framework beyond purely dissipative dynamics by introducing generalized Wasserstein-2 metrics that incorporate conservative (Hamiltonian) dynamics in both classical inertial systems and open quantum systems, yielding intrinsic distances that exactly characterize minimal entropy production under fixed dissipative mobilities.

Geometric Formulation of Gradient Flows

The gradient flow of a free energy functional F on a Riemannian manifold (M, g) is determined by the equation x˙ (t) = −Kx(t)(dFx(t)). This flow links thermodynamic forces to fluxes through the Onsager operator Kx = G−1x: T∗xM → TxM. p. 2

Where the paper says it

, conductivity, viscosity, etc). The inverse map that transforms forces into fluxes, Kx = G−1 x ∶ T ∗ xM → TxM, is known as the Onsager operator [33], [34].

The gradient flow formulation is determined by (i) a state space M, (ii) a dissipation mechanism encoded by K (or equivalently G), and (iii) a free energy functional F. p. 2
Where the paper says it

Therefore, F(x(t)) is non-increasing in time. We have seen how the gradient flow formulation is determined by: (i) a state space M, (ii) a dissipation mechanism encoded by K (or equivalently G), and (iii) a free energy functional F.

Optimal Transport Paradigms

The Wasserstein-2 distance arises from the gradient flow structure in purely dissipative systems, and from a combination of gradient and Hamiltonian flows in mixed systems. For continuous optimal transport, the Wasserstein-2 distance is defined as W2(ρ0, ρτ) 2 = inf V,ρ τ ∫ τ 0 gρ(ρ,˙ ρ˙)dt. p. 4

Where the paper says it

This metric defines a geodesic distance on M, known as the Wasserstein-2 distance, through W2(ρ0, ρτ ) 2 ∶= inf {ρ(t)}∶ρ0,ρτ τ ∫ τ 0 gρ(ρ,˙ ρ˙)dt, (9) where we have used the shorthand {ρ(t)} ∶ ρ0, ρτ to signify that the optimization is over curves {ρ(t)}t∈[0,τ] such that ρ(t) ∈ M for all t ∈ [0, τ ], ρ(0) = ρ0 and ρ(τ ) = ρτ . The Cauchy-Schwarz inequality and the fact that equality can be achieved through constant arclength parametrization lead to W2(ρ0, ρτ ) 2 = inf {ρ(t)}∶ρ0,ρτ ℓ 2 ρ , where we have defined the Riemannian length of a curve ℓρ ∶= ∫ τ 0 √ gρ(ρ,˙ ρ˙)dt.

In discrete optimal transport, the Wasserstein-2 distance is defined as W2(p0, pτ) 2 = inf V,p τ ∫ τ 0 V squared Mρ dt. p. 7
Where the paper says it

The metric (29) defines a quantum Wasserstein-2 geodesic distance, W2(ϱ0, ϱτ ) 2 ∶ = inf {ϱ(t)}∶ϱ0,ϱτ τ ∫ τ 0 gϱ(ϱ,˙ ϱ˙)dt (34) = inf {ϱ(t)}∶ϱ0,ϱτ ℓ 2 ϱ , where the Riemannian length of the curve is ℓϱ ∶= ∫ τ 0 √ gϱ(ϱ,˙ ϱ˙)dt, and we have used the shorthand {ϱ(t)} ∶ ϱ0, ϱτ to signify that we are searching for trajectories {ϱ(t)}t∈[0,τ] with endpoints ϱ(0) = ϱ0 and ϱ(τ ) = ϱτ . This distance squared is the solution to the quantum optimal transport problem W2(ϱ0, ϱτ ) 2 = inf V,ϱ τ ∫ τ 0 ∣∣V∣∣2 Mϱ dt (35) s.

For quantum systems, the quantum Wasserstein-2 distance is defined as W2(ϱ0, ϱτ) 2 = inf V,ϱ τ ∫ τ 0 gϱ(ϱ,˙ ϱ˙)dt. p. 7
Where the paper says it

(33) Thus, F (modulo logZ) acts as a Lyapunov function, ensuring convergence to the equilibrium state. The metric (29) defines a quantum Wasserstein-2 geodesic distance, W2(ϱ0, ϱτ ) 2 ∶ = inf {ϱ(t)}∶ϱ0,ϱτ τ ∫ τ 0 gϱ(ϱ,˙ ϱ˙)dt (34) = inf {ϱ(t)}∶ϱ0,ϱτ ℓ 2 ϱ , where the Riemannian length of the curve is ℓϱ ∶= ∫ τ 0 √ gϱ(ϱ,˙ ϱ˙)dt, and we have used the shorthand {ϱ(t)} ∶ ϱ0, ϱτ to signify that we are searching for trajectories {ϱ(t)}t∈[0,τ] with endpoints ϱ(0) = ϱ0 and ϱ(τ ) = ϱτ .

Hamiltonian-Dissipative Geometries

In Hamiltonian-dissipative dynamics, a new optimal transport distance W2,h is introduced to quantify entropy production through the metric hρ(µ,˙ ν˙). The comparison between the classical Wasserstein-2 distance and the Hamiltonian-dissipative distance yields the hierarchy κ(ρ)gρ(ρ,˙ ρ˙) ≤ hρ(ρ,˙ ρ˙) ≤ gρ(ρ,˙ ρ˙), where κ∗W2(µ0, µτ) 2 ≤ W2,h(µ0, µτ) 2 ≤ W2(µ0, µτ) 2. p. 9

Where the paper says it

Specifically, we obtain 1 τ W2(µ0, µτ ) 2 = { inf ϕ,µ ∫ τ 0 Σ˙ dt ∶ µ˙ = ∇ ⋅ (Mµ(∇ϕ)), µ(0) = µ0, µ(τ ) = µτ}. (41) Since any potential ϕ can be obtained through ϕ = βH˜ + log µ by an appropriate Hamiltonian H˜ , we may understand the square of the Wasserstein-2 distance as the minimum entropy production required to drive a system through Hamiltonian H˜ between two endpoints, for a fixed Mµ.

Linear Response and Counterdiabatic Protocols

The restriction of these metrics to equilibrium states gives rise to the linear-response-regime thermodynamic length, which can be interpreted as optimal transport protocols as counterdiabatic protocols that enforce the system to follow linear-response geodesics in finite time. This allows for a principled geometric way to identify a counterdiabatic protocol that is thermodynamically optimal. p. 2

Where the paper says it

Consequently, optimal-transport protocols may be understood as optimal counterdiabatic protocols that follow linear-response geodesics in finite time, even beyond the overdamped setting [31]. In this way, the presented Wasserstein-2 approach further extends the reach of Riemannian thermodynamic geometry out of equilibrium.

Entropy Production Decomposition and Uncertainty Relations

The introduced metrics induce a decomposition of entropy production into housekeeping and excess terms, where an irreversible force F may be split into gradient G and divergence-free χ terms with respect to the inner product defined by Mµ, so that ⟨F,Mµ(F)⟩ = ⟨G,Mµ(G)⟩ + ⟨χ,Mµ(χ)⟩. Furthermore, thermodynamic uncertainty relations (TUR) can be derived from these geometries to obtain a short-time TUR for both purely dissipative and mixed systems. p. 2

Where the paper says it

Furthermore, the advantages of having Hamiltonian flows are quantified by providing equivalence bounds on Hamiltonian-dissipative distances in terms of the corresponding purely dissipative ones. Finally, in Section V, we show that the proposed metrics reduce to the linear-response thermodynamic metrics [5, 30] when restricted to the equilibrium submanifold for continuous, discrete, and quantum systems, both purely dissipative and mixed.

Examples and Physical Interpretation

Representative examples include an RLC circuit with two dissipative channels, where the minimum entropy production is found to be equal to the purely dissipative Wasserstein-2 distance when linear Gaussian dynamics are optimal. The study also illustrates how coherent (Hamiltonian) terms can reduce dissipation in a qubit example, showing that the Hamiltonian term can only lower the minimum entropy production, with the gain factor κ∗ quantifying the maximum factor by which it can be lowered. p. 0

Where the paper says it

We extend this framework beyond purely dissipative dynamics by introducing generalized Wasserstein-2 metrics that incorporate conservative (Hamiltonian) dynamics in both classical inertial systems and open quantum systems, yielding intrinsic distances that exactly characterize minimal entropy production under fixed dissipative mobilities. We establish equivalence bounds between purely dissipative and Hamiltonian–dissipative geometries, explicitly quantifying how inertial or coherent dynamics can reduce dissipation.

Conclusion on Geometries

The work provides a unified perspective on how Wasserstein-2 geometries can quantify and bound entropy production for overdamped, inertial, discrete, and quantum systems alike. The results show that the restriction of these metrics to equilibrium states gives rise to linear-response-regime thermodynamic lengths. p. 1

Where the paper says it

In this structure, Onsager-type operators that linearly map thermodynamic forces to fluxes play a central role, leading to a bound on entropy production in terms of a Wasserstein-2 distance. Unlike Wasserstein-1, Wasserstein-2 distances recover overdamped results in the appropriate limit, including linear-response metrics [5, 30] when restricted to the equilibrium submanifold [31, 32].

Finally, the comparison between purely dissipative and mixed metrics yields a hierarchy of inequalities that capture dissipation and the extent to which Hamiltonian motion can lower its minimal cost. p. 2
Where the paper says it

Furthermore, the advantages of having Hamiltonian flows are quantified by providing equivalence bounds on Hamiltonian-dissipative distances in terms of the corresponding purely dissipative ones. Finally, in Section V, we show that the proposed metrics reduce to the linear-response thermodynamic metrics [5, 30] when restricted to the equilibrium submanifold for continuous, discrete, and quantum systems, both purely dissipative and mixed.

Appendix A: System-independent Quantum Formulation

The optimal transport problem (35) can be written through system-independent derivatives in analogy to the classical (continuous and discrete) settings by shifting the system dependence of the derivatives with respect to Lk to a weighted mobility operator Mˆ ϱ. This leads to the quantum partial derivatives ∂αφ = [Eα, φ] = [E†α, φ] = ∂†αφ, for all α ∈ A. p. 6

Where the paper says it

We equip hK with the usual inner product ⟨U, V⟩ = ∑ k∈K tr{U † k Vk}. Let us define the quantum partial derivatives ∂k ∶ h → hk ∂kφ = [Lk, φ] so that ∂ † kφ = [L † k , φ], φ ∈ h, for all k ∈ K.

Appendix B: Quantum and Discrete W2 Distances Coincide

The quantum Wasserstein distance W2(ϱ0, ϱτ) = W2(p0, pτ) when [H, ϱ0] = [H, ϱτ] = 0.

Appendix C: Vanishing Entropy Production for a Given Quantum Trajectory

One can choose a Hamiltonian, admissible jump operators, and jump rates such that entropy production can be made arbitrarily small for any given quantum trajectory in P∗(C, d).

Appendix D: Comparison between Purely Dissipative and Mixed Metrics

The comparison between the purely dissipative optimal transport metric and the linear-response regime metric yields the hierarchy κ(µ)gµ(µ,˙ µ˙) ≤ hµ(µ,˙ µ˙) ≤ gµ(µ,˙ µ˙), where κ∗W2(µ0, µτ) 2 ≤ W2,h(µ0, µτ) 2 ≤ W2(µ0, µτ) 2.

Appendix E: Proof that Lλ is the Backward Generator of the Lindblad Dynamics

The adjoint with respect to the M˜ ϱλ-weighted inner product is the backward generator of the Lindblad dynamics, i.e., Lλ = i[H λ, ⋅]+ ∑ k∈K γλk((L λk)†(⋅)L λk −1/2((L λk)†L λk, ⋅)).

Appendix F: Comparison between Purely Dissipative and Mixed Metrics

The comparison between the purely dissipative optimal transport metric and the linear-response regime metric yields the hierarchy κ(µ)gµ(µ,˙ µ˙) ≤ hµ(µ,˙ µ˙) ≤ gµ(µ,˙ µdot), where κ∗W2(μ0, μτ) 2 ≤ W2,h(μ0, μτ) 2 ≤ W2(μ0, μτ) 2.

Appendix G: Comparison between Purely Dissipative and Mixed Metrics

The comparison between the purely dissipative optimal transport metric and the linear-response regime metric yields the hierarchy κ(µ)gµ(µ,˙ µ˙) ≤ hµ(µ,˙ µ˙) ≤ gµ(µ,˙ µdot), where κ∗W2(μ0, μτ) 2 ≤ W2,h(μ0, μτ) 2 ≤ W2(μ0, μτ) 2.

Appendix H: Comparison between Purely Dissipative and Mixed Metrics

The comparison between the purely dissipative optimal transport metric and the linear-response regime metric yields the hierarchy κ(µ)gµ(µ,˙ µ˙) ≤ hµ(µ,˙ µ˙) ≤ gµ(µ,˙ µdot), where κ∗W2(μ0, μτ) 2 ≤ W2,h(μ0, μτ) 2 ≤ W2(μ0, μτ) 2.

Appendix I: Comparison between Purely Dissipative and Mixed Metrics

The comparison between the purely dissipative optimal transport metric and the linear-response regime metric yields the hierarchy κ(µ)gµ(µ,˙ µ˙) ≤ hµ(µ,˙ µ˙) ≤ gµ(µ,˙ µdot), where κ∗W2(μ0, μτ) 2 ≤ W2,h(μ0, μτ) 2 ≤ W2(μ0, μτ) 2.

Appendix J: Comparison between Purely Dissipative and Mixed Metrics

The comparison between the purely dissipative optimal transport metric and the linear-response regime metric yields the hierarchy κ(µ)gµ(µ,˙ µ˙) ≤ hµ(µ,˙ µ˙) ≤ gµ(µ,˙ µdot), where κ∗W2(μ0, μτ) 2 ≤ W2,h(μ0, μτ) 2 ≤ W2(μ0, μτ) 2.

Appendix K: Comparison between Purely Dissipative and Mixed Metrics

The comparison between the purely dissipative optimal transport metric and the linear-response regime metric yields the hierarchy κ(µ)gµ(µ,˙ µ˙) ≤ hµ(µ,˙ µ˙) ≤ gµ(µ,˙ µdot), where κ∗W2(μ0, μτ) 2 ≤ W2,h(μ0, μτ) 2 ≤ W2(μ0, μτ) 2.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved AI systems can quantify entropy production in classical, quantum, inertial, discrete, and mixed Hamiltonian-dissipative dynamics using Wasserstein-2 geometries based on gradient flows of free energy. This allows for a finite-time refinement of the second law by bounding entropy production via a Wasserstein-2 distance.

  2. AI systems will be able to design optimal counterdiabatic protocols by searching over equilibrium parametrized paths, which are geodesics in the linear response regime. This enables finding protocols that minimize entropy production along a given trajectory, even beyond the overdamped setting.

  3. The system can utilize a new metric for Hamiltonian-dissipative dynamics to lower bound entropy production by a factor of kappa star compared to purely dissipative systems. This provides an upper bound on the minimum entropy production, quantifying how much Hamiltonian motion can reduce dissipation via coherences in the Hamiltonian basis.

  4. AI systems will be able to identify housekeeping and excess terms in the decomposition of entropy production for systems that do not satisfy detailed balance. This allows for a clearer understanding of how irreversible forces are split into gradient and divergence-free parts with respect to the mobility operator inner product.

  5. The system can derive short-time thermodynamic uncertainty relations (TUR) for both purely dissipative and mixed systems by relating the rate of change of an observable to its Q weighted variance. This provides a lower bound on entropy production, linking it to the rate of change and the Q weighted variance of a time-invariant observable.

  6. AI systems can solve optimal transport problems in flux variables using convex optimization techniques to find global minimizers for minimal entropy production. This includes solving discrete and quantum optimal transport problems where the cost functions are jointly convex, ensuring that first-order optimality conditions yield global solutions.

  7. The system can implement a fully specified dissipative structure by fixing the mobility operator M over time, which leads to a genuine Wasserstein-2 geometry even in discrete and quantum settings. This fixed mobility approach is crucial for finding thermodynamically optimal counterdiabatic protocols in these regimes where kinetic constraints are relaxed.

Abstract

The second law does more than set the direction of thermodynamic evolution: it endows nonequilibrium transformations with an underlying geometry. In this work, we provide a unified geometric description of entropy production in classical and quantum thermodynamics based on Wasserstein-2 structures arising from gradient flows of free energy. We review how relaxation to equilibrium, in overdamped diffusions, discrete detailed-balanced Markov chains, and dissipative Lindblad dynamics, can be formulated as a gradient flow on the space of states. The associated Wasserstein-2 distance bounds entropy production, yielding a finite-time refinement of the second law. We extend this framework beyond purely dissipative dynamics by introducing generalized Wasserstein-2 metrics that incorporate conservative (Hamiltonian) dynamics in both classical inertial systems and open quantum systems, yielding intrinsic distances that exactly characterize minimal entropy production under fixed dissipative mobilities. We establish equivalence bounds between purely dissipative and Hamiltonian-dissipative geometries, explicitly quantifying how inertial or coherent dynamics can reduce dissipation. Finally, when restricted to equilibrium distributions, we recover the thermodynamic length of linear response-including the quantum thermodynamic length-thereby linking optimal transport, thermodynamic length, and counterdiabatic protocols within a single geometric framework. Collectively, our results extend the Riemannian program of thermodynamics further from equilibrium and provide a geometric foundation for optimal protocols beyond the overdamped setting.

Sources

Related papers