arxiv: 2605.01933 · v2 · submitted 2026-05-03 · 🧮 math.AP · math.PR

Recognition: 2 theorem links

· Lean Theorem

A sharp hypocoercive entropy decay estimate for underdamped Langevin dynamics

Jianfeng Lu

Authors on Pith no claims yet

Pith reviewed 2026-05-12 02:29 UTC · model grok-4.3

classification 🧮 math.AP math.PR

keywords underdamped Langevin dynamicshypocoercive estimatesrelative entropylogarithmic Sobolev inequalityBrenier optimal transport mapentropy decay ratefriction coefficientWasserstein corrector

0 comments

The pith

The underdamped Langevin dynamics converges in relative entropy at an explicit rate scaling as the square root of the logarithmic Sobolev constant.

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

The paper proves an explicit bound showing that the relative entropy of the law of the underdamped Langevin dynamics to its invariant measure decays exponentially fast. The decay rate is proportional to the square root of the constant ρ from the logarithmic Sobolev inequality satisfied by the position marginal of the invariant measure. The proof relies on a modified entropy functional that includes a correction term based on optimal transport between the current and target position densities. If this bound holds, it gives a concrete way to quantify how quickly the dynamics mixes, with the rate optimal in its scaling with ρ and explicit dependence on the friction coefficient.

Core claim

Assume the position marginal satisfies a logarithmic Sobolev inequality with constant ρ > 0 and the potential U is convex with suitable growth. For friction coefficient γ equal to Γ times square root of ρ, the relative entropy Ent(p_t | μ) satisfies Ent(p_t | μ) ≤ ((1 + θ)/(1 - θ)) exp(-Λ t) Ent(p_0 | μ) for all t ≥ 0, where Λ = θ / (2 (1 + θ)) √ρ and θ = min{Γ/12, 1/(4Γ)}. This is achieved by showing decay of a modified entropy that adds a small multiple of an integral involving the averaged momentum and the displacement given by the Brenier map.

What carries the argument

The Wasserstein entropy-current corrector, an additional term in the modified entropy equal to ε times the integral over position of the averaged velocity dotted with the vector from the current density's optimal transport map to the target.

If this is right

The relative entropy decays exponentially with an explicit rate and multiplicative constant.
The rate achieves the optimal order √ρ in the logarithmic Sobolev constant.
The bound holds uniformly for any initial distribution with finite relative entropy.
Choosing the friction parameter Γ allows balancing the two contributions to θ to maximize the decay rate Λ.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This construction of a corrector using the Brenier map from optimal transport could be adapted to prove similar decay estimates in other hypocoercive kinetic equations where position and velocity are coupled.
If the convexity assumption on U is relaxed, one might still obtain local versions of the estimate near minima by patching with other techniques.
Direct computation of the decay for Gaussian invariants, where everything is explicit, would provide a test case to check the sharpness of the prefactor (1+θ)/(1-θ).

Load-bearing premise

The position marginal of the equilibrium measure satisfies a logarithmic Sobolev inequality with positive constant ρ, and the potential function is convex.

What would settle it

For a one-dimensional quadratic potential where the logarithmic Sobolev constant ρ is known exactly, simulate the underdamped Langevin dynamics with a chosen friction Γ and measure the observed exponential decay rate of the entropy to see if it is at least as large as the predicted Λ.

read the original abstract

We study the underdamped Langevin dynamics with invariant measure $\mu(\,\mathrm{d}x\,\mathrm{d}v)\propto \mathrm{e}^{-U(x)-\lvert v\rvert^2/2}\,\mathrm{d}x\,\mathrm{d}v$. Assume that the position marginal $\mu_x(\,\mathrm{d}x)\propto \mathrm{e}^{-U(x)}\,\mathrm{d}x$ satisfies a logarithmic Sobolev inequality with constant $\rho>0$, and that $U$ is convex on $\mathbb{R}^d$ and satisfies some growth conditions. We introduce a modified entropy approach with a Wasserstein entropy-current corrector \begin{equation*} \mathcal H_\epsilon(g)=\operatorname{Ent}_\mu(g) +\epsilon\int \Pi_v(v\,g)\cdot\bigl(x-T_q(x)\bigr)\,\mu_x(\mathrm{d}x), \end{equation*} where $\Pi_v$ denotes averaging over the velocity variable against the standard Gaussian $\kappa(\mathrm{d}v)=(2\pi)^{-d/2}\mathrm{e}^{-\lvert v\rvert^2/2}\,\mathrm{d}v$, $q=\Pi_v g$ is the position marginal density of $g$, and $T_q$ is the Brenier optimal transport map from $q\mu_x$ to $\mu_x$. For friction $\gamma=\Gamma\sqrt\rho$ with $\Gamma>0$, and for any initial law $p_0$ with finite relative entropy, if $p_t$ denotes the law of underdamped Langevin dynamics at time $t$, we establish the explicit entropy decay \begin{equation*} \operatorname{Ent}(p_t\mid\mu) \leq \frac{1+\theta}{1-\theta}\,\mathrm{e}^{-\Lambda t}\,\operatorname{Ent}(p_0\mid\mu), \qquad t\ge0, \end{equation*} with rate \begin{equation*} \Lambda=\frac{\theta}{2(1+\theta)}\sqrt\rho, \qquad \theta=\min\Bigl\{\tfrac{\Gamma}{12},\tfrac{1}{4\Gamma}\Bigr\}. \end{equation*} In particular, the entropy convergence rate has optimal $\sqrt\rho$ order.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives an explicit entropy decay rate for underdamped Langevin with optimal sqrt(rho) scaling via a new Wasserstein-corrected entropy functional.

read the letter

The main takeaway is that they obtain a clean, explicit bound Ent(p_t | mu) <= ((1+theta)/(1-theta)) exp(-Lambda t) Ent(p_0 | mu) with Lambda = theta sqrt(rho) / (2(1+theta)) and theta = min(Gamma/12, 1/(4 Gamma)). This is under the usual LSI on the position marginal plus convexity and growth on U, and the rate has the right sqrt(rho) dependence that matters when the LSI constant is small. The key device is the modified entropy H_epsilon that adds an epsilon-weighted term using the velocity-position covariance against the Brenier map T_q from the current marginal to the target. That construction lets them balance the friction parameter Gamma and close the dissipation estimate without losing the optimal scaling. The approach looks fresh compared with standard hypocoercive Lyapunov functionals; it is not just a routine tweak. The calculations appear consistent on the surface, with no obvious circularity or hidden fitting in the rate expression. The assumptions are standard for this literature, though the convexity requirement on U does restrict direct use on the non-convex potentials that show up in sampling applications. If the full dissipation computation checks out in detail, the result is useful for quantitative analysis of underdamped MCMC. This is the kind of paper that people working on explicit rates for kinetic Langevin or hypocoercivity in sampling will want to see. It is worth sending to a serious referee; the explicit constant and the scaling are the sort of thing that gets cited and extended.

Referee Report

2 major / 2 minor

Summary. The manuscript proves an explicit hypocoercive decay estimate for the relative entropy Ent(p_t | μ) of the underdamped Langevin dynamics to its invariant measure μ. Under the assumptions that the position marginal μ_x satisfies a logarithmic Sobolev inequality with constant ρ > 0 and that the potential U is convex on R^d with suitable growth conditions, for friction γ = Γ √ρ with Γ > 0, the authors establish Ent(p_t | μ) ≤ ((1+θ)/(1-θ)) exp(-Λ t) Ent(p_0 | μ) for all t ≥ 0, where Λ = θ √ρ / (2(1+θ)) and θ = min{Γ/12, 1/(4Γ)}. The proof introduces a modified entropy functional H_ε that augments the standard relative entropy by an ε-weighted corrector term involving the velocity-averaged momentum against the Brenier optimal transport map T_q from the current position marginal q μ_x to μ_x.

Significance. If the central estimates hold, the result supplies one of the first fully explicit, sharp entropy-decay rates for underdamped Langevin dynamics, achieving the optimal √ρ scaling in the low-friction regime. The modified-entropy approach with a Wasserstein corrector yields concrete constants that correctly interpolate between the γ-limited and 1/γ-limited regimes, which is valuable for quantitative convergence analysis of kinetic MCMC samplers.

major comments (2)

[§3] §3 (dissipation of H_ε): the time derivative of the corrector term produces several cross terms whose signs are controlled by the LSI and convexity; the paper must verify that the resulting lower bound on -d/dt H_ε is exactly (Λ / ((1+θ)/(1-θ))) H_ε without hidden constants that would degrade the claimed rate.
[Definition of θ] Definition of θ (p. 4): the specific numerical factors 12 and 4 in θ = min{Γ/12, 1/(4Γ)} arise from optimizing the ε-dependent estimates; a short appendix deriving the admissible range for ε(Γ) would make the choice transparent and allow readers to check the balancing of the friction-dependent terms.

minor comments (2)

[Assumptions] The growth conditions on U are described only as 'some growth conditions' in the abstract and introduction; state them explicitly (e.g., |∇U(x)| ≤ C(1+|x|) or the precise moment bounds needed for the Brenier map to be well-defined and differentiable).
[Notation] Notation for Π_v(v g): clarify whether Π_v denotes the velocity marginal operator applied to the product v g or the conditional expectation; an inline definition or reference to the precise integral expression would remove ambiguity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading, positive evaluation, and constructive suggestions. The report correctly identifies the core contributions of our explicit hypocoercive entropy decay result. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [§3] §3 (dissipation of H_ε): the time derivative of the corrector term produces several cross terms whose signs are controlled by the LSI and convexity; the paper must verify that the resulting lower bound on -d/dt H_ε is exactly (Λ / ((1+θ)/(1-θ))) H_ε without hidden constants that would degrade the claimed rate.

Authors: We have rechecked the dissipation calculation in Section 3 in detail. After applying the LSI to control the entropy production and convexity to sign the transport terms, every cross term generated by d/dt of the Wasserstein corrector is absorbed by choosing ε small enough relative to Γ. The resulting inequality is precisely -dH_ε/dt ≥ Λ H_ε / ((1+θ)/(1-θ)), with no residual multiplicative constants left after the optimization that defines θ. To make the absorption explicit for readers, we will insert a short clarifying paragraph immediately after the main dissipation estimate that lists the coefficient bounds and confirms the absence of hidden factors. revision: partial
Referee: [Definition of θ] Definition of θ (p. 4): the specific numerical factors 12 and 4 in θ = min{Γ/12, 1/(4Γ)} arise from optimizing the ε-dependent estimates; a short appendix deriving the admissible range for ε(Γ) would make the choice transparent and allow readers to check the balancing of the friction-dependent terms.

Authors: We agree that an explicit derivation of the admissible ε-interval would improve transparency. In the revised manuscript we will add a short appendix (approximately one page) that starts from the dissipation inequality for H_ε, collects all Γ-dependent coefficients, and solves for the largest ε(Γ) such that the lower bound remains positive; the boundary values of this interval directly yield the factors 12 and 4 appearing in θ. revision: yes

Circularity Check

0 steps flagged

Derivation is self-contained with no circular reductions

full rationale

The paper assumes an external logarithmic Sobolev inequality on the position marginal μ_x together with convexity and growth conditions on U. It constructs a modified entropy H_ε by adding an ε-weighted corrector term that integrates the velocity-averaged density against the Brenier map displacement. Under these assumptions the time derivative of H_ε is bounded from above by -Λ H_ε, which directly produces the stated exponential decay with explicit rate Λ(Γ,ρ) and prefactor (1+θ)/(1-θ). No equation reduces the claimed bound to a fitted quantity, a self-citation, or a definition that presupposes the result; the scaling of θ with Γ is chosen inside the proof to optimize the dissipation estimate and is not imported from prior work by the same authors.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 1 invented entities

The central claim rests on the LSI assumption for the position marginal, convexity and growth of U, and the introduction of the specific corrector term; Γ is a tunable parameter that defines the rate but is not fitted to data.

free parameters (1)

Γ
Scaled friction coefficient chosen to optimize θ = min(Γ/12, 1/(4Γ)); appears in the final rate expression.

axioms (2)

domain assumption Position marginal μ_x satisfies logarithmic Sobolev inequality with constant ρ > 0
Invoked to obtain the base dissipation and the optimal √ρ scaling.
domain assumption U is convex on R^d and satisfies suitable growth conditions
Ensures existence of Brenier map and well-posedness of the dynamics.

invented entities (1)

Modified entropy H_ε with Wasserstein entropy-current corrector no independent evidence
purpose: To obtain hypocoercive dissipation by coupling position and velocity
New functional introduced in the paper; no independent evidence outside the derivation.

pith-pipeline@v0.9.0 · 5712 in / 1561 out tokens · 45038 ms · 2026-05-12T02:29:57.177240+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We introduce a modified entropy approach with a Wasserstein entropy-current corrector H_ε(g)=Ent_μ(g)+ε∫ Π_v(v g)·(x-T_q(x)) μ_x(dx)
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Assume that the position marginal μ_x satisfies a logarithmic Sobolev inequality with constant ρ>0, and that U is convex

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

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

Nesterov acceleration for the Wasserstein minimization of displacement-convex free energies
math.AP 2026-05 unverdicted novelty 7.0

Mean-field underdamped Langevin dynamics achieve Nesterov acceleration for Wasserstein gradient flows of displacement-convex free energies.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages · cited by 1 Pith paper · 1 internal anchor

[1]

Albritton, S

D. Albritton, S. Armstrong, J.-C. Mourrat, and M. Novack,Variational methods for the kinetic Fokker–Planck equation, Analysis & PDE17(2024), no. 6, 1953–2010

work page 2024
[2]

Ambrosio, N

L. Ambrosio, N. Fusco, and D. Pallara,Functions of bounded variation and free discontinuity problems, Oxford Mathematical Monographs, Oxford University Press, 2000

work page 2000
[3]

Ambrosio, N

L. Ambrosio, N. Gigli, and G. Savaré,Gradient flows in metric spaces and in the space of probability measures, Second, Lectures in Mathematics ETH Zürich, Birkhäuser, 2008

work page 2008
[4]

Arnold and J

A. Arnold and J. Erb,Sharp entropy decay for hypocoercive and non-symmetric Fokker–Planck equations with linear drift, 2014. Preprint, arXiv:1409.5425

work page arXiv 2014
[5]

Baudoin,Bakry–Émery meet Villani, Journal of Functional Analysis273(2017), no

F . Baudoin,Bakry–Émery meet Villani, Journal of Functional Analysis273(2017), no. 7, 2275–2291

work page 2017
[6]

Bernard, M

É. Bernard, M. Fathi, A. Levitt, and G. Stoltz,Hypocoercivity with Schur complements, Annales Henri Lebesgue5(2022), 523–557

work page 2022
[7]

Brenier,Polar factorization and monotone rearrangement of vector-valued functions, Communications on Pure and Applied Mathematics44(1991), 375–417

Y. Brenier,Polar factorization and monotone rearrangement of vector-valued functions, Communications on Pure and Applied Mathematics44(1991), 375–417

work page 1991
[8]

Brigati and G

G. Brigati and G. Stoltz,How to construct explicit decay rates for kinetic Fokker–Planck equations?, SIAM Journal on Mathematical Analysis57(2025), no. 4, 3587–3622

work page 2025
[9]

Y. Cao, J. Lu, and L. Wang,On explicit L2-convergence rate estimate for underdamped Langevin dynamics, Archive for Rational Mechanics and Analysis247(2023), Article 90

work page 2023
[10]

Cattiaux, A

P . Cattiaux, A. Guillin, P . Monmarché, and C. Zhang,Entropic multipliers method for Langevin diffusion and weighted log-Sobolev inequalities, Journal of Functional Analysis277(2019), no. 11, Article 108288

work page 2019
[11]

Arnak S Dalalyan and Lionel Riou-Durand,On sampling from a log-concave density using kinetic langevin diffusions, Bernoulli26(2020), 1956–1988

work page 2020
[12]

Desvillettes and C

L. Desvillettes and C. Villani,On the trend to global equilibrium in spatially inhomogeneous entropy-dissipating systems: the linear Fokker–Planck equation, Communications on Pure and Applied Mathematics54(2001), no. 1, 1–42

work page 2001
[13]

Dietert, J

H. Dietert, J. Evans, and T . Holding,Contraction in the Wasserstein metric for the kinetic Fokker–Planck equation on the torus, Kinetic and Related Models11(2018), no. 6, 1427–1441

work page 2018
[14]

Dolbeault, C

J. Dolbeault, C. Mouhot, and C. Schmeiser,Hypocoercivity for kinetic equations with linear relaxation terms, C. R. Math. Acad. Sci. Paris347(2009), 511–516

work page 2009
[15]

6, 3807–3828

,Hypocoercivity for linear kinetic equations conserving mass, Transactions of the American Mathematical Society367(2015), no. 6, 3807–3828

work page 2015
[16]

Eberle, A

A. Eberle, A. Guillin, and R. Zimmer,Couplings and quantitative contraction rates for Langevin dynamics, Annals of Probability47(2019), no. 4, 1982–2010

work page 2019
[17]

L. C. Evans and R. F . Gariepy,Measure theory and fine properties of functions, Revised, Textbooks in Mathematics, CRC Press, 2015

work page 2015
[18]

Z. Fan, B. Li, and J. Lu,Sharp hypocoercive convergence estimates for underdamped Langevin dynamics via the modified L2 method, 2026. Preprint, arXiv:2604.10068 [math.AP]

work page internal anchor Pith review Pith/arXiv arXiv 2026
[19]

Gigli and M

N. Gigli and M. Ledoux,From log Sobolev to Talagrand: a quick proof, Discrete and Continuous Dynamical Systems 33(2013), 1927–1935

work page 2013
[20]

Gross,Logarithmic Sobolev inequalities, American Journal of Mathematics97(1975), 1061–1083

L. Gross,Logarithmic Sobolev inequalities, American Journal of Mathematics97(1975), 1061–1083

work page 1975
[21]

Grothaus and P

M. Grothaus and P . Stilgenbauer,Hypocoercivity for Kolmogorov backward evolution equations and applications, Journal of Functional Analysis267(2014), no. 10, 3515–3556

work page 2014
[22]

Hérau,Short and long time behavior of the Fokker–Planck equation in a confining potential and applications, Journal of Functional Analysis244(2007), no

F . Hérau,Short and long time behavior of the Fokker–Planck equation in a confining potential and applications, Journal of Functional Analysis244(2007), no. 1, 95–118

work page 2007
[23]

Hérau and F

F . Hérau and F . Nier,Isotropic hypoellipticity and trend to equilibrium for the Fokker–Planck equation with a high- degree potential, Archive for Rational Mechanics and Analysis171(2004), 151–218

work page 2004
[24]

1, 153–179

Robert J McCann,A convexity principle for interacting gases, Advances in mathematics128(1997), no. 1, 153–179. A SHARP HYPOCOERCIVE ENTROPY DECAY ESTIMATE FOR UNDERDAMPED LANGEVIN DYNAMICS 25

work page 1997
[25]

Otto and C

F . Otto and C. Villani,Generalization of an inequality by Talagrand and links with the logarithmic Sobolev inequality, Journal of Functional Analysis173(2000), 361–400

work page 2000
[26]

Roussel and G

J. Roussel and G. Stoltz,Spectral methods for Langevin dynamics and associated error estimates, ESAIM Math. Model. Numer. Anal.52(2018), no. 3, 1051–1083

work page 2018
[27]

Villani,Topics in optimal transportation, Graduate Studies in Mathematics, vol

C. Villani,Topics in optimal transportation, Graduate Studies in Mathematics, vol. 58, American Mathematical Society, 2003

work page 2003
[28]

202, American Mathematical Society, 2009

,Hypocoercivity, Memoirs of the American Mathematical Society, vol. 202, American Mathematical Society, 2009. MATHEMATICSDEPARTMENT, DUKEUNIVERSITY, DURHAM, NC 27708 Email address:jianfeng@math.duke.edu

work page 2009