Cutoff for non-negatively curved diffusions

Justin Salez

REVIEW 1 major objections 4 minor 1 cited by

Cutoff for non-negatively curved diffusions

T0 review · 1 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Cutoff holds for every non-negatively curved diffusion under the product condition.

desk verdict Elegant theorem, but the proof leans on a reverse Pinsker inequality that is false as stated, so the main argument does not go through. read the letter →

arxiv 2501.01304 v2 pith:U7OKWOJJ submitted 2025-01-02 math.PR

Justin Salez This is my paper · ORCID

classification math.PR MSC 60J6060J2558J65

keywords cutoffphenomenonmixingtimespectralgapBakry–ÉmerycurvaturevarentropyLangevindiffusionproductconditionprocesses

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 resolves the cutoff phenomenon for a broad class of Markov diffusions: any diffusion with non-negative Bakry–Émery curvature, such as a Langevin diffusion in a convex potential, undergoes a sharp transition to equilibrium (cutoff) as soon as its spectral gap times its mixing time diverges. The author proves a universal bound on the width of the mixing window, valid for any deterministic starting point, and derives cutoff from the product condition alone, without model-specific analysis. The proof is elementary and rests on a new differential inequality linking the entropy of the process to its varentropy.

What carries the argument

The engine of the proof is a differential inequality between the entropy and the varentropy of the process: $\frac{d}{dt} \mathrm{Ent}(X_t) \leq - \frac{\mathrm{Varent}(X_t)}{2t}$. It is derived from the local Poincaré inequality $\mathrm{Var}[g(X_t)] \leq 2t\, \mathbb{E}[\Gamma g(X_t)]$, which follows from non-negative curvature, together with the chain rule for the carré du champ. The reverse Pinsker inequality $\mathrm{Ent}(X) \leq 1 + \frac{\sqrt{\mathrm{Varent}(X)}}{1 - \mathrm{tv}(X)}$ converts this into an integrable differential inequality for the entropy after the mixing time, whose integration yields the window bound. The positive curvature case replaces the local Poincaré inequality with a time-uniform version.

What would settle it

Compute, for a concrete non-negatively curved diffusion such as the Ornstein–Uhlenbeck process in high dimension, the quantity $\frac{d}{dt} \mathrm{Ent}(X_t) + \frac{\mathrm{Varent}(X_t)}{2t}$ along the trajectory; if it is ever positive, Lemma 3 is false, and the bound in Theorem 1 collapses.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that for any non-negatively curved diffusion starting from a deterministic point, the width of the mixing window satisfies $w_{\mathrm{mix}}(\varepsilon) \leq \frac{3}{\lambda \varepsilon^3} + 3 \sqrt{\frac{t_{\mathrm{mix}}(1-\varepsilon)}{\lambda \varepsilon^3}}$ for every $\varepsilon \in (0,1/2)$, where $\lambda$ is the spectral gap. From this estimate, cutoff follows whenever the product condition $\lambda\, t_{\mathrm{mix}}(\varepsilon) \to \infty$ holds: the ratio $t_{\mathrm{mix}}(1-\varepsilon)/t_{\mathrm{mix}}(\varepsilon)$ tends to $1$ for every $\varepsilon \in (0,1)$. Under the stronger positive curvature condition $\mathrm{CD}(\kappa,\infty)$, the sharper bound $w_{\mathrm{mix}}(\varepsilon) \leq \frac{3}{\kappa \varepsilon^2}$ holds, and cutoff follows when $\kappa\, t_{\mathrm{mix}}(\varepsilon) \to \infty$. The result applies on Euclidean spaces and on weighted Riemannian manifolds, and extends to worst-case mixing times by taking a maximum over initial states.

Load-bearing premise

The proof rests on the reverse Pinsker inequality (Lemma 1), borrowed from reference [19] and not proved in the paper, which is what turns the decay of varentropy into a usable differential inequality for the entropy; if that inequality failed for these diffusions, the window bound would not follow.

Editorial extensions

If this is right

Any sequence of Langevin diffusions in convex potentials with $\lambda\, t_{\mathrm{mix}}(\varepsilon) \to \infty$ exhibits cutoff, removing the need for model-specific mixing-time analyses.
On compact manifolds, cutoff for non-negatively curved diffusions is completely characterized by the product condition, since the condition is necessary there.
The bound extends to worst-case mixing times by maximizing over initial states, so the cutoff criterion holds uniformly over compact state spaces.
Under positive curvature, the mixing window is at most $\frac{3}{\kappa \varepsilon^2}$, giving a sharper quantitative statement whenever a positive curvature lower bound is available.

Reading between the lines

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

The entropy–varentropy differential inequality is likely to hold for a wider class of Markov processes with appropriate curvature bounds, potentially yielding cutoff criteria for discrete chains that satisfy a comparable local Poincaré inequality.
Because the bound depends only on the spectral gap, it suggests that within this class the mixing window is controlled by a single scalar quantity, masking any finer geometric structure.
A natural testable extension is to sharpen the constant in Theorem 1 for specific models, such as the Ornstein–Uhlenbeck process, where the exact window may be computed and compared to the bound.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

The short version: Theorem 1 is probably true, but the proof given here is not valid. Lemma 1, the reverse Pinsker inequality borrowed from your JEMS paper, is false in the form stated, and the proof uses it exactly where it fails.

Here is the counterexample. Take any probability space with a set A of mass δ < 1/e, and set f = 1/δ on A, 0 elsewhere. Then Ent = log(1/δ), Varent = 0, and tv = 1−δ. Inequality (10) would give log(1/δ) ≤ 1, which is false. The worry that this density is not strictly positive does not save the lemma: add a tiny background η on A^c and renormalize. The entropy goes to log(1/δ), the varentropy goes to 0, and the total variation goes to 1−δ, so the inequality still fails for small η. This is not an edge case; the proof of Theorem 1 applies Lemma 1 at times where tv = 1−ε with ε < 1/2, which is precisely the regime of the counterexample.

What is genuinely new and good: Lemma 3, the differential inequality d/dt Ent ≤ −Varent/(2t), is original and looks correct under the CD(0,∞) assumption. The strategy of turning varentropy decay into a window estimate is attractive, and the paper is very clearly written. The use of curvature to get local Poincare (12) is standard. Lemma 2 is cited without proof, but that is minor and it is a known result.

The problem is that Lemma 1 is the bridge that converts Lemma 3 into the differential inequality (13). With Lemma 1 false, the inequality d/dt Ent ≤ −(ε Ent − 1)^2/(2t) is not justified. The proof of the cutoff criterion collapses at this point. My guess is that a corrected reverse Pinsker inequality, for instance Ent ≤ log(1/(1−tv)) + sqrt(Varent)/(1−tv), would restore something close to the stated window bound, and the theorem may be salvageable. But as written, the argument is invalid.

This is a serious flaw, not a cosmetic one. I would recommend rejecting the current version, but not because the idea is bad. The paper deserves the referee process precisely because the result is important and the repair may be straightforward. I would send it back with a clear request to fix the reverse Pinsker step.

Referee Report

1 major / 4 minor

Summary. The paper proves a universal bound on the mixing-window width of Markov diffusions with non-negative Bakry-Émery curvature: from any deterministic starting point, wmix(ε) ≤ 3/(λ ε³) + 3√(tmix(1−ε)/(λ ε³)), where λ is the spectral gap. From this it derives a cutoff criterion under the product condition λ tmix(ε) → ∞, covering Riemannian Langevin diffusions and general CD(0,∞) Markov processes. The proof introduces a differential inequality between varentropy and entropy, Lemma 3, and combines it with a reverse Pinsker-type inequality and a mixing-time estimate in terms of entropy; integrating and optimizing the resulting differential inequality yields the window bound. A positive-curvature variant gives the sharper bound wmix(ε) ≤ 3/(κ ε²).

Significance. If the technical flaw described below is corrected, this is a major result: it confirms the product-condition conjecture for the entire CD(0,∞) diffusion class, covers deterministic initial conditions and worst-case starting sets, gives a complete characterization of cutoff on compact manifolds, and unifies several model-specific cutoff proofs. The varentropy differential inequality in Lemma 3 is simple and appears to be new; the integration and optimization steps are transparent and check out. The theorem is parameter-free and the constants are explicit, with no fitted or adjustable quantities. The main obstacle to acceptance is not the scope or importance of the result but the validity of Lemma 1 as stated.

major comments (1)

[Section 2, Lemma 1 (Eq. (10))]

minor comments (4)

[Section 2, Lemma 3]
[Section 2, Lemma 2]
[Section 2, paragraph after Lemma 2]
[Section 1, Corollary 1]

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Theorem 1 is derived from a new differential inequality plus independently-proved prior lemmas.

full rationale

The proof of Theorem 1 is not circular. The only external ingredients are Lemma 1 (a reverse Pinsker inequality with varentropy) and Lemma 2 (a spectral-gap mixing-time bound), both explicitly stated as borrowed from the author's prior paper [19]. Those are general proven mathematical statements whose assumptions do not include the cutoff conclusion, and they are not fitted to any data or normalized so as to force the result. The genuinely new ingredient, Lemma 3, is proved in the text from the CD(0,∞) sub-commutation property, the chain rule, and standard calculus; it yields the differential inequality (13), which is then integrated and optimized over a free time parameter. No fitted quantity is renamed as a prediction, and no equation reduces to its own input by construction. The presence of self-citations is therefore not circular: the cited lemmas are independent support in the sense of being checkable general results rather than assumptions tailored to this theorem.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters are fit; λ, κ and t mix are inputs. All auxiliary inequalities are cited from prior published work or classical Bakry-Emery theory. The paper introduces no new objects beyond the differential inequality in Lemma 3.

assumptions (5)

standard math Reverse Pinsker inequality: Ent(X) ≤ 1 + sqrt(Varent(X))/(1 - tv(X)) for any random variable with density.
Lemma 1, borrowed from [19, Lemma 8]; unproved in this text, used to convert varentropy decay into Eq. (13).
standard math Mixing-time estimate: tmix(ε) ≤ (1 + Ent(X0))/(λ ε) for reversible processes, with propagated version (11).
Lemma 2, borrowed from [19, Lemma 7]; used to bound tmix(ε) in terms of entropy at time t.
standard math Local Poincare inequality: under CD(0,∞), Var[g(X_t)] ≤ 2t E[Γ g(X_t)] for all t≥0.
Eq. (12) in the proof of Lemma 3; the paper calls it classical from sub-commutation, no proof given.
domain assumption Regularity: f_t and log f_t are in Dom(L) with chain rule and curvature condition applicable.
Section 1 definition of non-negatively curved diffusions; needed to apply the local Poincare and chain rule to g=log f_t.
standard math Total variation distance to equilibrium is non-increasing over time for reversible Markov processes.
Used in proof of Theorem 1 to claim tv(X_t) ≤ 1-ε for all t ≥ t0; not stated explicitly but standard.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Cutoff for non-negatively curved diffusions." pith.science (2026). https://pith.science/paper/U7OKWOJJ

@misc{pith2026250101304,
  author       = {Pith},
  title        = {Pith review of: Cutoff for non-negatively curved diffusions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U7OKWOJJ}},
  note         = {Machine review of arXiv:2501.01304}
}

read the original abstract

We resolve the long-standing problem of elucidating the cutoff phenomenon for a vast and important class of Markov processes, namely Markov diffusions with non-negative Bakry-\'Emery curvature. More precisely, we prove that any sequence of non-negatively curved diffusions exhibits cutoff in total variation as soon as the product condition is satisfied. Our result holds in Euclidean spaces as well as on Riemannian manifolds, and for arbitrary non-random initial conditions. It vastly simplifies, unifies and generalizes a number of isolated works that have established cutoff through a delicate and model-dependent analysis of mixing times. The proof is elementary: we exploit a new simple differential relation between varentropy and entropy to produce a quantitative bound on the width of the mixing window.

Figures

Figures reproduced from arXiv: 2501.01304 by the authors.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

A universal cutoff phenomenon for mean-field exchange models
math.PR 2025-06 conditional novelty 8.0 of 10

Mean-field exchange models with any symmetric, non-degenerate redistribution variable exhibit a Wasserstein cutoff at time n log n/(2h) with Gaussian profile set by the size-biased log-variable's first two moments.

Reference graph

Works this paper leans on

21 extracted references · 11 canonical work pages · cited by 1 Pith paper

[1]

Random walks on ﬁnite groups and rapidly mixing Marko v chains

David Aldous. Random walks on ﬁnite groups and rapidly mixing Marko v chains. In Seminar on probability, XVII , volume 986 of Lecture Notes in Math. , pages 243–297. Springer, Berlin, 1983

1983
[2]

Shuﬄing cards and stopping times

David Aldous and Persi Diaconis. Shuﬄing cards and stopping times . American Math- ematical Monthly, pages 333–348, 1986

1986
[3]

Bakry and Michel ´Emery

D. Bakry and Michel ´Emery. Diﬀusions hypercontractives. In S´ eminaire de probabilit´ es, XIX, 1983/84 , volume 1123 of Lecture Notes in Math. , pages 177–206. Springer, Berlin, 1985

1983
[4]

Analysis and geometry of Markov diﬀusion operators , volume 348 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]

Dominique Bakry, Ivan Gentil, and Michel Ledoux. Analysis and geometry of Markov diﬀusion operators , volume 348 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences] . Springer, Cham, 2014

2014
[5]

Abrupt convergence for a family of Ornstein -Uhlenbeck processes

Gerardo Barrera. Abrupt convergence for a family of Ornstein -Uhlenbeck processes. Braz. J. Probab. Stat. , 32(1):188–199, 2018

work page 2018
[6]

Charac terization of cutoﬀ for reversible Markov chains

Riddhipratim Basu, Jonathan Hermon, and Yuval Peres. Charac terization of cutoﬀ for reversible Markov chains. Ann. Probab., 45(3):1448–1487, 2017

work page 2017
[7]

Universal cutoﬀ for Dyson Ornstein Uhlenbeck process

Jeanne Boursier, Djalil Chafa ¨ ı, and Cyril Labb´ e. Universal cutoﬀ for Dyson Ornstein Uhlenbeck process. Probab. Theory Related Fields , 185(1-2):449–512, 2023

work page 2023
[8]

On cutoﬀ via rigidity for high dimension al curved diﬀu- sions, 2024

Djalil Chafa ¨ ı and Max Fathi. On cutoﬀ via rigidity for high dimension al curved diﬀu- sions, 2024. 10

work page 2024

Show all 21 references

[9]

The sample size required in importance sampling

Sourav Chatterjee and Persi Diaconis. The sample size required in importance sampling. Ann. Appl. Probab. , 28(2):1099–1135, 2018

2018
[10]

The cutoﬀ phenomen on for ergodic Markov processes

Guan-Yu Chen and Laurent Saloﬀ-Coste. The cutoﬀ phenomen on for ergodic Markov processes. Electron. J. Probab. , 13:no. 3, 26–78, 2008

2008
[11]

Cover and Joy A

Thomas M. Cover and Joy A. Thomas. Elements of information theory . Wiley- Interscience [John Wiley & Sons], Hoboken, NJ, second edition, 2006

2006
[12]

The cutoﬀ phenomenon in ﬁnite Markov chains

Persi Diaconis. The cutoﬀ phenomenon in ﬁnite Markov chains. Proc. Nat. Acad. Sci. U.S.A., 93(4):1659–1664, 1996

1996
[13]

Total variation cu toﬀ in birth-and-death chains

Jian Ding, Eyal Lubetzky, and Yuval Peres. Total variation cu toﬀ in birth-and-death chains. Probab. Theory Related Fields , 146(1-2):61–85, 2010

2010
[14]

Concentra- tion of information on discrete groups, 2024

Jonathan Hermon, Xiangying Huang, Francesco Pedrotti, and Justin Salez. Concentra- tion of information on discrete groups, 2024

2024
[15]

Optimal lossless compre ssion: Source varen- tropy and dispersion

Ioannis Kontoyiannis and Sergio Verdu. Optimal lossless compre ssion: Source varen- tropy and dispersion. pages 1739–1743, 07 2013

2013
[16]

Coupling from the past

David A. Levin and Yuval Peres. Markov chains and mixing times . American Mathemat- ical Society, Providence, RI, 2017. Second edition of [ MR2466937 ], With contributions by Elizabeth L. Wilmer, With a chapter on “Coupling from the past” by J ames G. Propp and David B. Wilson

2017
[17]

Aim research workshop on sharp thresholds for mixing times

Y Peres. Aim research workshop on sharp thresholds for mixing times. 2004

2004
[18]

Universality of cutoﬀ for exclusion with reservoirs

Justin Salez. Universality of cutoﬀ for exclusion with reservoirs . Ann. Probab. , 51(2):478–494, 2023

2023
[19]

Cutoﬀ for non-negatively curved Markov chains

Justin Salez. Cutoﬀ for non-negatively curved Markov chains. J. Eur. Math. Soc. (JEMS), 26(11):4375–4392, 2024

2024
[20]

The varentropy criterion is sharp on expanders

Justin Salez. The varentropy criterion is sharp on expanders. Ann. H. Lebesgue , 7:239– 250, 2024

2024
[21]

Saloﬀ-Coste

L. Saloﬀ-Coste. Precise estimates on the rate at which certain diﬀusions tend to equi- librium. Math. Z. , 217(4):641–677, 1994. 11

1994

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

Random walks on ﬁnite groups and rapidly mixing Marko v chains

David Aldous. Random walks on ﬁnite groups and rapidly mixing Marko v chains. In Seminar on probability, XVII , volume 986 of Lecture Notes in Math. , pages 243–297. Springer, Berlin, 1983

1983

[2] [2]

Shuﬄing cards and stopping times

David Aldous and Persi Diaconis. Shuﬄing cards and stopping times . American Math- ematical Monthly, pages 333–348, 1986

1986

[3] [3]

Bakry and Michel ´Emery

D. Bakry and Michel ´Emery. Diﬀusions hypercontractives. In S´ eminaire de probabilit´ es, XIX, 1983/84 , volume 1123 of Lecture Notes in Math. , pages 177–206. Springer, Berlin, 1985

1983

[4] [4]

Analysis and geometry of Markov diﬀusion operators , volume 348 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]

Dominique Bakry, Ivan Gentil, and Michel Ledoux. Analysis and geometry of Markov diﬀusion operators , volume 348 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences] . Springer, Cham, 2014

2014

[5] [5]

Abrupt convergence for a family of Ornstein -Uhlenbeck processes

Gerardo Barrera. Abrupt convergence for a family of Ornstein -Uhlenbeck processes. Braz. J. Probab. Stat. , 32(1):188–199, 2018

work page 2018

[6] [6]

Charac terization of cutoﬀ for reversible Markov chains

Riddhipratim Basu, Jonathan Hermon, and Yuval Peres. Charac terization of cutoﬀ for reversible Markov chains. Ann. Probab., 45(3):1448–1487, 2017

work page 2017

[7] [7]

Universal cutoﬀ for Dyson Ornstein Uhlenbeck process

Jeanne Boursier, Djalil Chafa ¨ ı, and Cyril Labb´ e. Universal cutoﬀ for Dyson Ornstein Uhlenbeck process. Probab. Theory Related Fields , 185(1-2):449–512, 2023

work page 2023

[8] [8]

On cutoﬀ via rigidity for high dimension al curved diﬀu- sions, 2024

Djalil Chafa ¨ ı and Max Fathi. On cutoﬀ via rigidity for high dimension al curved diﬀu- sions, 2024. 10

work page 2024