Exponential Ergodicity in Relative Entropy and $L^2$-Wasserstein Distance for non-equilibrium partially dissipative Kinetic SDEs

Eva Kopfer; Panpan Ren; Pierre Monmarch\'e; Xing Huang

REVIEW 3 major objections 5 minor 2 cited by

Exponential Ergodicity in Relative Entropy and $L^2$-Wasserstein Distance for non-equilibrium partially dissipative Kinetic SDEs

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Kinetic SDEs converge exponentially in entropy even when the invariant measure has no explicit formula.

desk verdict Plausible and important result, but the proof of Lemma 3.3 has a wrong constant and the core functional inequalities are imported from a companion preprint. read the letter →

arxiv 2507.02518 v2 pith:E2XK64PS submitted 2025-07-03 math.PR

Xing Huang , Eva Kopfer , Pierre Monmarché , Panpan Ren This is my paper · ORCID

classification math.PR MSC 60H1060K3582C22

keywords hypercontractivityexponentialergodicityrelativeentropyWassersteindistancekineticSDEsnon-equilibriumsteadystateMcKean-Vlasovmean-fieldinteractingparticlesystem

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proves that a broad class of kinetic stochastic differential equations with partially dissipative drift relax to their steady state exponentially fast in relative entropy and in $L^2$-Wasserstein distance, without assuming the forces derive from a potential or that the invariant measure is explicit. The setting is genuinely non-equilibrium: the force field is only assumed to satisfy a joint dissipativity condition outside a compact set, and the noise can be degenerate in the position coordinate. The result closes a gap that had remained open for partially dissipative non-equilibrium kinetic SDEs, where only $L^2$ or total-variation rates were previously available. The same mechanism yields exponential entropy and Wasserstein convergence for McKean-Vlasov equations and mean-field particle systems under small nonlinear perturbations, with rates uniform in the number of particles.

What carries the argument

The argument is carried by a four-step chain. First, a hypocoercive estimate (built on a time-dependent matrix $G_t$ that couples position and velocity gradients) converts a Poincaré inequality for $\bar\mu$ into $L^2$-exponential ergodicity of the semigroup. Second, hyperboundedness of the semigroup from $L^2(\bar\mu)$ to $L^4(\bar\mu)$ is imported from the companion analysis of log-Sobolev inequalities for decoupled SDEs; combining $L^2$-ergodicity with hyperboundedness upgrades the semigroup to genuine hypercontractivity, and by duality the adjoint is hypercontractive from $L^{4/3}$ to $L^2$, which is the exact input for exponential decay of relative entropy. Third, the log-Sobolev inequality for $\bar\mu$ implies a transport inequality controlling $W_2$ by relative entropy. Fourth, a log-Harnack inequality bounds the entropy after one time unit by the $W_2$ distance of the initial laws, so the Wasserstein decay follows from the entropy decay. The same machinery is rerun with constants uniform in the interaction strength and in $N$ for the mean-field particle system.

What would settle it

Take $d=1$, $\sigma=1$, and $b(x,y)=-x-\gamma y+\varepsilon\sin x$ with small $\varepsilon$, which satisfies (B). Compute the spectral gap of the density evolution operator and simulate the relative entropy of the solution law against the invariant measure; a zero gap or subexponential entropy decay for any such admissible $\varepsilon>0$ would contradict Theorem 2.3.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is Theorem 2.3: under assumption (B) (Lipschitz drift plus a two-component dissipativity condition outside a compact) and invertibility of $\sigma\sigma^*$, the kinetic SDE (2.3) has a unique invariant probability measure $\bar\mu$, and the semigroup satisfies hypercontractivity of its adjoint, which yields constants $c,\tilde c,\lambda>0$ such that $\operatorname{Ent}(\mathcal L(X_t^{\mu_0},Y_t^{\mu_0})|\bar\mu)\le c e^{-2\lambda t}\operatorname{Ent}(\mu_0|\bar\mu)$ and $W_2(\mathcal L(X_t^{\mu_0},Y_t^{\mu_0}),\bar\mu)\le \tilde c e^{-\lambda t}W_2(\mu_0,\bar\mu)$. Thus the process forgets its initial law exponentially in information-theoretic and transport metrics, even though $\bar\mu$ has no explicit density and the drift is not of gradient form.

Load-bearing premise

The load-bearing premise is that the invariant measure satisfies a strong isoperimetric inequality (a log-Sobolev inequality over all phase-space directions) and hyperboundedness of the semigroup, with constants controlled by the same drift and noise parameters. The paper imports these facts from the companion analysis [25] rather than proving them here; if they fail for some drift satisfying (B), the entropy and Wasserstein conclusions of Theorems 2.3, 2.6, and 2.7 collapse.

Editorial extensions

If this is right

Non-equilibrium kinetic SDEs with non-gradient forces and unknown invariant density inherit the same exponential relative-entropy and Wasserstein rates as equilibrium Langevin dynamics, under only the partial dissipativity condition (B).
For McKean-Vlasov kinetic SDEs with small interaction coefficient $K_I$, there is a unique stationary solution, and $\bar\mu$ attracts all initial laws at explicit exponential rates in both $W_2$ and relative entropy.
For the $N$-particle mean-field system, a uniform-in-$N$ exponential contraction holds, up to an additive empirical-measure error of order $N^{-1/2}$ for $d<2$, $N^{-1/2}\log(1+N)$ for $d=2$, and $N^{-2/d}$ for $d>2$.
Because relative entropy domination is strictly stronger than $L^2$ dominance, these results upgrade the known $L^2$ hypocoercivity for non-equilibrium kinetic processes to information-theoretic convergence.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The 'hyperboundedness plus $L^2$-ergodicity implies hypercontractivity' transfer used here is not tied to the kinetic structure; the same lemma should yield entropy decay for other degenerate diffusion semigroups whose invariant measures satisfy weaker functional inequalities.
A quantitative version of Theorem 2.3 would give non-asymptotic, dimension-scalable bounds for underdamped Langevin samplers with non-gradient or data-dependent forces, a setting where explicit invariant densities are typically absent.
The additive empirical-measure term in the particle-system estimate matches the known Wasserstein rate of empirical measures, suggesting that the uniform-in-$N$ entropy and $W_2$ rates cannot be improved in general without changing the metric or adding smoothness assumptions.
For the example $b(x,y)=-x-\gamma y+F(x,y)$ with small $|\nabla F|$, the constants $\lambda$ and $c$ should be computable explicitly from the proof; comparing them with direct numerical entropy decay would show how sharp the route through log-Sobolev and hypercontractivity is.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

First, the thing to know: the paper has a plausible and important main theorem, but the proof as written has a concrete error in Lemma 3.3, and the theorem leans on a companion paper for the hardest functional inequalities. Both are fixable, but a referee should not let the current version through.

The actual contribution is real. Theorem 2.3 gives exponential decay in relative entropy and W2 for kinetic SDEs with partially dissipative, non-gradient drift and no explicit invariant density. That combination appears to be new, and the strategy is clean: import L2-ergodicity from [37] and hyperboundedness/LSI from [25], then use the known hypercontractivity-to-entropy step [51, 4]. The extension to McKean-Vlasov and to N-particle systems with uniform rates is useful and follows the same pattern.

The soft spots, in order of severity.

First, Lemma 3.3: the constant is wrong. The proof defines epsilon = delta1/M + 1/2, and then needs epsilon M <= delta1 to make the last term in the dissipation bound nonpositive. With the displayed value, epsilon M - delta1 = M/2 > 0, so the Gronwall argument doesn't close. This is very likely a typo — any epsilon small enough, say delta1/(2M), makes the argument work — but as written the lemma is false. A referee should ask for the corrected constant.

Second, the dependency on [25]. The proof of Theorem 2.3 imports the LSI and hyperboundedness for the unknown invariant measure from [25, Thms 3.4, 3.5]. This is not circular, but it is a substantial outsource: the existence of a finite LSI constant for a non-equilibrium steady state is the hardest part of the problem. The present paper should at least spell out why the hypotheses of [25] match Assumption (B) and why the constants are finite. Without that, a skeptical reader cannot verify the foundation.

Third, the tensorization step in Theorem 2.7, equation (3.30), is only sketched and the exponent bookkeeping is hard to follow. It may be correct, but the induction should be written out.

Overall, the central claim is plausible and important; the flaws are repairable. This deserves a serious referee, with the request that Lemma 3.3 be fixed and the [25] dependency made explicit. If I were in the area, I'd cite it after the revision.

Referee Report

3 major / 5 minor

Summary. The paper studies kinetic SDEs of the form dX_t=Y_t dt, dY_t=b(X_t,Y_t)dt+σ dW_t under a partially dissipative condition (B), with no gradient-structure assumption on the drift and no explicit invariant density. The main results are Theorem 2.3, giving exponential ergodicity in relative entropy and in the L2-Wasserstein distance for the classical kinetic SDE, and Theorems 2.6 and 2.7, extending this to McKean-Vlasov SDEs and to mean-field interacting particle systems with rates uniform in the number of particles under a small interaction constant. The proof strategy is to obtain L2-exponential ergodicity by hypocoercivity, combine it with hyperboundedness and a full-gradient log-Sobolev inequality imported from the companion paper [25] to obtain hypercontractivity, and then convert hypercontractivity into entropy decay; Wasserstein decay is derived from Talagrand's inequality and a log-Harnack inequality.

Significance. If the proof can be completed, the paper would fill a real gap in the literature: previous exponential entropy-decay results for kinetic SDEs either required an explicit Gibbs invariant measure or a uniformly dissipative drift, whereas Theorem 2.3 covers genuinely non-equilibrium stationary states without an explicit density. The hypercontractivity route from L2-ergodicity plus hyperboundedness is attractive, and the McKean-Vlasov and mean-field extensions with constants independent of the number of particles are valuable. The paper is clearly organized and the statements are natural. However, the current version is conditional on functional inequalities that are not proved here, and the main technical lemma contains an inconsistency that breaks the proof as written. The significance is therefore real but conditional on nontrivial repairs.

major comments (3)

[§3.1, Lemma 3.3] The choice of ε is incompatible with the claimed dissipation estimate. The proof defines ε = δ1/M + 1/2, where M = 2(2K_b+1)^2 + (2K_b+1) and δ1 is the minimum eigenvalue of σσ*. Later it obtains the bound ⟨R_t z,z⟩ ≤ −ε α(t)^2 |x|^2/2 + (εM − δ1)|y|^2. Since εM = δ1 + M/2, the coefficient (εM − δ1) is strictly positive, so the conclusion dN_t/dt ≤ −ε α(t)^2 μbar(|∇f_t|^2)/2 does not follow. This is a load-bearing error: Lemma 3.3 is the L2-exponential ergodicity input for Theorem 2.3, and the same lemma is used for the McKean-Vlasov and particle-system results. The argument can likely be repaired by taking ε = δ1/(2M) or an analogous small multiple of δ1/M, but as written the Gronwall dissipation proof fails.
[§3.1, Proof of Theorem 2.3, Step 1] Hyperboundedness (3.2) and the full-gradient log-Sobolev inequality (3.3) are quoted from the companion paper [25, Theorems 3.4 and 3.5] without any verification that assumption (B) together with invertibility of σσ* implies the hypotheses of those theorems. These inequalities are not technical accessories: (3.3) provides the Poincaré inequality needed by Lemma 3.3, and (3.2) together with L2-ergodicity is the only route to hypercontractivity. Since μbar is an unknown non-equilibrium density, the existence of a finite constant C1 in (3.3) is a strong functional inequality that the present manuscript does not establish. The authors should either prove (3.2)–(3.3) under the stated assumptions or explicitly include them as additional hypotheses; as it stands, Theorem 2.3, and consequently Theorems 2.6 and 2.7, are conditional on an external result whose hypotheses are not checked.
[§3.3, Step 6 and conclusion] In the derivation of (3.42), a positive term appears to be dropped. Starting from (3.41), one has A ≤ α B + c̄ R_d(N) with A = W2((P^N_t̂)*νN, μbar^⊗N)^2 and B = W2(νN, μbar^⊗N)^2. After replacing B by 2 W2(νN, μbar_N)^2 + 2 c̄ R_d(N)/(1−α), the extra term c̄ R_d(N) from (3.41) must be kept; the displayed constant in (3.42) is therefore too small and the inequality as stated can fail, for example when νN = μbar_N and the W2 term vanishes. The proof can be repaired by retaining the missing term and enlarging constants, but the computation needs correction. In the same passage, the final sentence says 'we obtain (2.15)' where the entropy bound (2.16) is meant.

minor comments (5)

[§3.1, Lemma 3.3] The sentence 'By an approximation technique, we can assume that b ∈ C^1 with ∥∇b∥∞ < K_b' is not elaborated; since b is only assumed Lipschitz, a brief justification or a reference for the approximation step would improve rigor.
[§3.2, Lemma 3.4] In the display following equation (3.14), the right-hand side '2|z̃| + 2|θ^μ_1(z)|^2' appears to be missing the square on the first term; it should be of the form C(1+|z̃|^2+|z|^2).
[References] Reference [22] contains a formatting error: the Springer proceedings entry is immediately followed by an unrelated journal citation 'J. Differential Equations 413(2024), 632–661', which seems to belong to a different reference.
[Introduction] In the passage 'high-dimensionl algorithms', 'dimensionl' should be 'dimensional'.
[§3.1, Step 2 of Theorem 2.3] The derivation of (2.4) from truncation uses 'Fatou’s lemma and dominated convergence theorem'; monotone convergence for |f_n| ↑ |f| would be more direct and avoids the need for a separate dominated-convergence justification.

Circularity Check

0 steps flagged · score 2.0 of 10

No construction-level circularity: entropy/W2 decay is a genuine consequence of imported LSI and hyperboundedness; the companion-paper dependence is a missing-support risk, not a circular reduction.

full rationale

The central claims are not equivalent by construction to their inputs. Theorem 2.3's proof combines the L2 contraction obtained in Lemma 3.3, which is proved in this paper by a hypocoercivity computation conditional on a Poincaré inequality, with hyperboundedness (3.2) and the full-gradient log-Sobolev inequality (3.3) imported from the companion paper [25]. The passage from these ingredients to exponential entropy decay and W2 decay uses genuine implications: Lemma 3.1 turns hyperboundedness plus L2-ergodicity into hypercontractivity and entropy decay, and the Talagrand and log-Harnack steps convert entropy decay into Wasserstein decay. These are standard results applied in a new setting, not a renamed input or a fitted parameter presented as a prediction. Theorems 2.6 and 2.7 similarly import existence, uniform LSI, and hyperboundedness from [25] and log-Harnack from [26], but then genuinely propagate these bounds through the McKean-Vlasov fixed-point argument and the synchronous-coupling particle estimate. The real concern is external correctness rather than circularity: [25] is a self-citation by three of the four authors, and the present paper does not verify that the hypotheses of [25, Theorems 3.4 and 3.5] are exactly (B) plus invertibility of sigma sigma*, nor does it control the constants C0, C1 there. If those functional inequalities fail under (B) or have uncontrolled constants, Theorems 2.3, 2.6, and 2.7 lose their foundation. That is a missing-support risk, not evidence that the claimed conclusions are already assumed. Accordingly, the score is 2 rather than 0 because the proof leans on self-cited companion work at a load-bearing point, but there is no construction-level circularity.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No free parameters are fitted and no new entities are postulated. The proofs depend on imported functional inequalities (log-Sobolev, hyperboundedness) from [25], log-Harnack inequalities from [46] and [26], heat kernel estimates from [12], and a hypocoercivity computation adapted from [37]. The most fragile import is the full-gradient log-Sobolev inequality for the non-explicit invariant measure, which is assumed as a black box.

assumptions (4)

domain assumption Under (B), the invariant measure mu_bar satisfies the full-gradient log-Sobolev inequality (3.3) and the hyperboundedness (3.2) with constants depending on the data in (B).
Imported from the companion paper [25, Theorems 3.4 and 3.5], not proved here; all three main theorems rest on it.
domain assumption The log-Harnack inequality P1 log f (x) <= log P1 f (y) + c1|x-y|^2 holds for the kinetic semigroup and its McKean-Vlasov and particle variants.
Cited to [46] and [26]; used to convert W2 contraction to entropy bounds.
domain assumption The hypocoercivity computation of Lemma 3.3 is valid, including the smoothing approximation that b can be taken C^1 with ||nabla b||_infinity < Kb.
The proof follows [37, Theorem 2] but contains a wrong constant (epsilon = delta1/M + 1/2) that breaks the dissipation estimate as written, and the approximation argument is not detailed.
domain assumption Tensorization of hypercontractivity for the adjoint semigroup (P*_{t1})^{otimes N} holds as in (3.30).
Proved by induction in the paper, but the displayed inequalities appear to contain exponent errors; the standard result is known for Markov semigroups.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Exponential Ergodicity in Relative Entropy and $L^2$-Wasserstein Distance for non-equilibrium partially dissipative Kinetic SDEs." pith.science (2026). https://pith.science/paper/E2XK64PS

@misc{pith2026250702518,
  author       = {Pith},
  title        = {Pith review of: Exponential Ergodicity in Relative Entropy and $L^2$-Wasserstein Distance for non-equilibrium partially dissipative Kinetic SDEs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E2XK64PS}},
  note         = {Machine review of arXiv:2507.02518}
}

abstract

In this paper, we derive exponential ergodicity in relative entropy for general kinetic SDEs under a partially dissipative condition. It covers non-equilibrium situations where the forces are not of gradient type and the invariant measure does not have an explicit density, extending previous results set in the equilibrium case. The key argument is to establish the hypercontractivity of the associated semigroup, which follows from its hyperboundedness and its $L^2$-exponential ergodicity. Moreover, we obtain exponential ergodicity in the $L^2$-Wasserstein distance by combining Talagrand's inequality with a log-Harnack inequality. These results are further extended to the McKean-Vlasov setting and to the associated mean-field interacting particle systems, with convergence rates that are uniform in the number of particles in the latter case, under small nonlinear perturbations.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Log-Sobolev inequalities for boundary-driven anharmonic chains
math-ph 2026-07 conditional novelty 7.0 of 10

A weakly anharmonic boundary-driven oscillator chain obeys a dimension-free full-gradient logarithmic Sobolev inequality and an O(N^3)-time boundary space-time logarithmic Sobolev inequality.
Local exponential stability of mean-field Langevin descent-ascent and associated particle system
cs.LG 2026-02 conditional novelty 7.0 of 10

Starting close to the mixed Nash equilibrium, mean-field Langevin descent-ascent converges exponentially fast in Wasserstein distance.

Reference graph

Works this paper leans on

53 extracted references · 49 canonical work pages · cited by 2 Pith papers

[37]

L2 geometric ergodicity for the kinetic Langevin process with non-equilibrium steady states

P. Monmarch´ e,L2 geometric ergodicity for the kinetic Langevin process with non- equilibrium steady states, arXiv:2501.18004

work page Pith review arXiv
[25]

Huang, E

X. Huang, E. Kopfer, P. Ren, Log-Sobolev Inequality for Decoupled and McKean- Vlasov SDEs and Application on Exponential Ergodicity, arXiv:2501.16092v3

work page arXiv
[1]

Albritton, S

D. Albritton, S. Armstrong, J.-C. Mourrat, M. Novack, Variational methods for the kinetic fokker–planck equation, Anal. PDE 17(2024), 1953–2010

work page 2024
[2]

Bakry, M

D. Bakry, M. ´Emery, Diffusions hypercontractives, S´ eminaire de Probabilit´ es XIX 1983/84: Proceedings. Berlin, Heidelberg: Springer Berlin Heidelberg, 2006. 177– 206

work page 1983
[3]

Bakry, I

D. Bakry, I. Gentil, M. Ledoux, Analysis and geometry of Markov diffusion operators, volume 348 of Grundlehren Math. Wiss. Cham: Springer, 2014

work page 2014
[4]

Bao, F.-Y

J. Bao, F.-Y. Wang, C. Yuan, Hypercontractivity for functional stochastic differential equations, Stochastic Process. Appl. 125(2015), 3636–3656

work page 2015
[5]

Baudoin, Bakry–´ emery meet villani, J

F. Baudoin, Bakry–´ emery meet villani, J. Funct. Anal. 273(2017), 2275–2291

work page 2017
[6]

Bernard, M

´E. Bernard, M. Fathi, A. Levitt, G. Stoltz, Hypocoercivity with schur complements, Ann. H. Lebesgue 5(2022), 523–557. 23

work page 2022

Show all 53 references

[7]

Bonetto, J

F. Bonetto, J. L. Lebowitz, L. Rey-Bellet, Fourier’s law: a challenge to theorists, Imperial College Press , London, 2000, 128–150

2000
[8]

Brigati, F

G. Brigati, F. L¨ orler, L. Wang, Hypocoercivity meets lifts, arXiv:2412.10890

arXiv
[9]

Camrud, A

E. Camrud, A. Durmus, P. Monmarch´ e, G. Stoltz, Second order quantitative bounds for unadjusted generalized Hamiltonian Monte Carlo, arXiv:2306.09513

arXiv
[10]

Y. Cao, J. Lu, L. Wang, On explicit L2-convergence rate estimate for underdamped langevin dynamics, Arch. Ration. Mech. Anal. 247(2023), Paper No. 90, 34 pp

2023
[11]

Cesbron, M

L. Cesbron, M. Herda, On a Vlasov-Fokker-Planck equation for stored electron beams. J. Differential Equations 404(2024), 316–353

2024
[12]

Chaudru de Raynal, S

P.-E. Chaudru de Raynal, S. Menozzi, A. Pesce, X. Zhang, Heat kernel and gradient estimates for kinetic SDEs with low regularity coefficients,Bull. Sci. Math. 183(2023), Paper No. 103229, 56 pp

2023
[13]

F. Chen, Y. Lin, Z. Ren, S. Wang. Uniform-in-time propagation of chaos for kinetic mean field langevin dynamics, Electron. J. Probab., 29(2024), 1–43

2024
[14]

F. Chen, Z. Ren, S. Wang, Uniform-in-time propagation of chaos for mean field Langevin dynamics, arXiv:2212.03050
[15]

Chizat, Mean-field langevin dynamics: Exponential convergence and annealing, Transactions on Machine Learning Research, 2022

L. Chizat, Mean-field langevin dynamics: Exponential convergence and annealing, Transactions on Machine Learning Research, 2022

2022
[16]

Dietert, L2-hypocoercivity for non-equilibrium kinetic equations, arXiv:2310.13456

H. Dietert, L2-hypocoercivity for non-equilibrium kinetic equations, arXiv:2310.13456

arXiv
[17]

Dolbeault, C

J. Dolbeault, C. Mouhot, C. Schmeiser, Hypocoercivity for kinetic equations with linear relaxation terms, C. R. Math. Acad. Sci. Paris 347(2009), 511–516

2009
[18]

A Erdogdu, R

M. A Erdogdu, R. Hosseinzadeh, On the convergence of langevin monte carlo: The interplay between tail growth and smoothness. In Conference on Learning Theory , pages 1776–1822. PMLR, 2021

2021
[19]

Fournier, A

N. Fournier, A. Guillin, On the rate of convergence in Wasserstein distance of the empirical measure, Probab. Theory Related Fields 162(2015), 707–738

2015
[20]

Evans, Gary P

Denis J. Evans, Gary P. Morriss, Statistical mechanics of nonequilibrium liquids, Academic Press, 1990

1990
[21]

Q. Fu, A. Wilson, Mean-field underdamped langevin dynamics and its space-time discretization, arXiv:2312.16360. 24

arXiv
[22]

Giacomin, S

G. Giacomin, S. Olla, E. Saada, H. Spohn, G. Stoltz, Stochastic dynamics out of equilibrium, Springer Proceedings in Mathematics & Statistics , vol. 282, Springer Nature Switzerland AG 2019. J. Differential Equations 413(2024), 632–661

2024
[23]

Guillin, W

A. Guillin, W. Liu, L. Wu, C. Zhang, Uniform Poincar´ e and logarithmic Sobolev inequalities for mean field particle systems, Ann. Appl. Probab. 32(2022), 1590–1614

2022
[24]

Guillin, P

A. Guillin, P. Monmarch´ e, Uniform Long-Time and Propagation of Chaos Estimates for Mean Field Kinetic Particles in Non-convex Landscapes,J. Stat. Phys. 185(2021), Paper No. 15, 20 pp

2021
[26]

Huang, F.-Y

X. Huang, F.-Y. Wang, Regularities and exponential ergodicity in entropy for SDEs driven by distribution dependent noise, Bernoulli 30(2024), 3303–3323

2024
[27]

Iacobucci, S

A. Iacobucci, S. Olla, G. Stoltz, Thermo-mechanical transport in rotor chains. J. Stat. Phys. 183(2021), no. 2, Paper No. 26, 34 pp

2021
[28]

Iacobucci, S

A. Iacobucci, S. Olla, G. Stoltz, Convergence rates for nonequilibrium Langevin dynamics, Ann. Math. Qu´ e.43(2019),73–98

2019
[29]

Lelievre, G

T. Lelievre, G. Stoltz, Partial differential equations and stochastic methods in molec- ular dynamics, Acta Numerica, 25(2016),681–880

2016
[30]

Letizia, S

V. Letizia, S. Olla, Nonequilibrium isothermal transformations in a temperature gradient from a microscopic dynamics, Ann. Probab. 45(2017), 3987–4018

2017
[31]

Y. Lu, P. Monmarch´ e, Convergence of time-averaged mean field gradient descent dynamics for continuous multi-player zero-sum games, arXiv:2505.07642

arXiv
[32]

Y.-A. Ma, N. S. Chatterji, X. Cheng, N. Flammarion, P. L. Bartlett, M. I. Jordan, Is there an analog of Nesterov acceleration for gradient-based MCMC? Bernoulli, 27(2021),1942–1992

2021
[33]

Marini, L

E. Marini, L. Andreis, F. Collet, M. Formentin, Noise-induced periodicity in a frus- trated network of interacting diffusions, NoDEA Nonlinear Differential Equations Appl. 30(2023), Paper No. 34, 35 pp

2023
[34]

Mattingly, A.M

J.C. Mattingly, A.M. Stuart, D.J. Higham, Ergodicity for SDEs and approxima- tions: locally Lipschitz vector fields and degenerate noise, Stochastic Process. Appl. 101(2002),185–232. 25

2002
[35]

Menegaki, Quantitative Rates of Convergence to Non-equilibrium Steady State for a Weakly Anharmonic Chain of Oscillators, J

A. Menegaki, Quantitative Rates of Convergence to Non-equilibrium Steady State for a Weakly Anharmonic Chain of Oscillators, J. Stat. Phys. , 181(2020),53–94

2020
[36]

Monmarch´ e, Almost sure contraction for diffusions on Rd

P. Monmarch´ e, Almost sure contraction for diffusions on Rd. Application to gener- alised Langevin diffusions, Stochastic Process. Appl. 161(2023), 316–349

2023
[38]

Monmarch´ e, An entropic approach for Hamiltonian Monte Carlo: the idealized case, Ann

P. Monmarch´ e, An entropic approach for Hamiltonian Monte Carlo: the idealized case, Ann. Appl. Probab. 34(2024), 2243–2293

2024
[39]

Monmarch´ e, A note on a vlasov-fokker-planck equation with non-symmetric in- teraction, to appear in Ann

P. Monmarch´ e, A note on a vlasov-fokker-planck equation with non-symmetric in- teraction, to appear in Ann. Fac. Sci. Toulouse Math. 2025

2025
[40]

Monmarch´ e, M

P. Monmarch´ e, M. Ramil, Overdamped limit at stationarity for non-equilibrium Langevin diffusions, Electron. Commun. Probab. 27(2022), Paper No. 3, 8 pp

2022
[41]

Monmarch´ e, J

P. Monmarch´ e, J. Reygner, Local convergence rates for Wasserstein gradient flows and McKean-Vlasov equations with multiple stationary solutions, arXiv:2404.15725

arXiv
[42]

Monmarch´ e, K

P. Monmarch´ e, K. Schuh, Non-asymptotic entropic bounds for non-linear kinetic langevin sampler with second-order splitting scheme, arXiv:2412.03560

arXiv
[43]

Monmarch´ e, S

P. Monmarch´ e, S. Wang, Logarithmic sobolev inequalities for non-equilibrium steady states, Potential Anal.(2025). https://doi.org/10.1007/s11118-025-10211-6

2025 doi
[44]

Nitanda, D

A. Nitanda, D. Wu,T. Suzuki, Convex analysis of the mean field Langevin dynamics, In International Conference on Artificial Intelligence and Statistics, pages 9741–9757. PMLR, 2022

2022
[45]

F. Otto, C. Villani, Generalization of an inequality by Talagrand and links with the logarithmic Sobolev inequality, J. Funct. Anal. 173(2000), 361–400

2000
[46]

Ren, F.-Y

P. Ren, F.-Y. Wang, Exponential convergence in entropy and Wasserstein for McKean-Vlasov SDEs, Nonlinear Anal. 206(2021), Paper No. 112259, 20 pp

2021
[47]

Suzuki, D

T. Suzuki, D. Wu, A. Nitanda, Convergence of mean-field Langevin dynamics: time- space discretization, stochastic gradient, and variance reduction, In Proceedings of the 37th International Conference on Neural Information Processing Systems , NIPS ’23, Red Hook, NY, USA, 2024....

2024
[48]

Talay, Stochastic Hamiltonian systems: exponential convergence to the invariant measure, and discretization by the implicit Euler scheme, Markov Process

D. Talay, Stochastic Hamiltonian systems: exponential convergence to the invariant measure, and discretization by the implicit Euler scheme, Markov Process. Related Fields, 8(2002),163–198. 26

2002
[49]

Vempala, A

S. Vempala, A. Wibisono, Rapid convergence of the unadjusted Langevin algorithm: Isoperimetry suffices, In H. Wallach, H. Larochelle, A. Beygelzimer, F. d Alch´ e-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 32. Curran Associa...

2019
[50]

Villani, Hypocoercivity Mem

C. Villani, Hypocoercivity Mem. Amer. Math. Soc. 202(2009), iv+141 pp

2009
[51]

Wang, Hypercontractivity and applications for stochastic Hamiltonian systems, J

F.-Y. Wang, Hypercontractivity and applications for stochastic Hamiltonian systems, J. Funct. Anal. 272(2017), 5360–5383

2017
[52]

Wang, Harnack inequalities on manifolds with boundary and applications, J

F.-Y. Wang, Harnack inequalities on manifolds with boundary and applications, J. Math. Pures Appl. , 94(2010), 304–321

2010
[53]

G. Wang, L. Chizat, Open problem: Convergence of single-timescale mean-field langevin descent-ascent for two-player zero-sum games, InThe Thirty Seventh Annual Conference on Learning Theory, pages 5345–5350. PMLR, 2024. 27

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [37]

L2 geometric ergodicity for the kinetic Langevin process with non-equilibrium steady states

P. Monmarch´ e,L2 geometric ergodicity for the kinetic Langevin process with non- equilibrium steady states, arXiv:2501.18004

work page Pith review arXiv

[2] [25]

Huang, E

X. Huang, E. Kopfer, P. Ren, Log-Sobolev Inequality for Decoupled and McKean- Vlasov SDEs and Application on Exponential Ergodicity, arXiv:2501.16092v3

work page arXiv

[3] [1]

Albritton, S

D. Albritton, S. Armstrong, J.-C. Mourrat, M. Novack, Variational methods for the kinetic fokker–planck equation, Anal. PDE 17(2024), 1953–2010

work page 2024

[4] [2]

Bakry, M

D. Bakry, M. ´Emery, Diffusions hypercontractives, S´ eminaire de Probabilit´ es XIX 1983/84: Proceedings. Berlin, Heidelberg: Springer Berlin Heidelberg, 2006. 177– 206

work page 1983

[5] [3]

Bakry, I

D. Bakry, I. Gentil, M. Ledoux, Analysis and geometry of Markov diffusion operators, volume 348 of Grundlehren Math. Wiss. Cham: Springer, 2014

work page 2014

[6] [4]

Bao, F.-Y

J. Bao, F.-Y. Wang, C. Yuan, Hypercontractivity for functional stochastic differential equations, Stochastic Process. Appl. 125(2015), 3636–3656

work page 2015

[7] [5]

Baudoin, Bakry–´ emery meet villani, J

F. Baudoin, Bakry–´ emery meet villani, J. Funct. Anal. 273(2017), 2275–2291

work page 2017

[8] [6]

Bernard, M

´E. Bernard, M. Fathi, A. Levitt, G. Stoltz, Hypocoercivity with schur complements, Ann. H. Lebesgue 5(2022), 523–557. 23

work page 2022