On the exponential convergence to equilibrium for ultrafast diffusion equations

Xinhang Tong; Yi C. Huang

REVIEW 2 major objections 5 minor 8 references

On the exponential convergence to equilibrium for ultrafast diffusion equations

T0 review · 2 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read The paper establishes exponential convergence to equilibrium for ultrafast diffusion equations in R^n, covering Gaussian weights in any dimension.

desk verdict Poincaré-based proof extends exponential convergence to Gaussian weights in R^n, but Theorem 6's rate is off by a γ^{2(r+1)} factor; qualitative result survives with corrected constant. read the letter →

arxiv 2509.07382 v1 pith:R6KZUQDN submitted 2025-09-09 math.AP math.CA

Yi C. Huang , Xinhang Tong This is my paper · ORCID

classification math.APmath.CA MSC 35B4035K5535A23

keywords ultrafastdiffusionexponentialconvergencePoincaréinequalityGaussianmeasureshigherdimensionsentropydissipationlog-concaveweightslong-timebehavior

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 establishes exponential convergence to equilibrium for the ultrafast diffusion equation in R^n, when the equilibrium is a Gaussian or, more generally, any measure satisfying a Poincaré inequality. The proof is built on one algebraic identity and one analytic inequality: the free-energy gap is rewritten as an exact weighted L2 distance of u=f/m to 1, and the Poincaré inequality then controls that distance by the entropy-dissipation rate. This avoids the optimal-transport machinery used in the one-dimensional result [FI25], which excluded Gaussian weights. If correct, it settles the open higher-dimensional Gaussian case and provides an explicit exponential rate constant depending only on the Poincaré constant, the band bounds c and C, and the normalization of the equilibrium.

What carries the argument

The carrying object is the entropy-dissipation estimate (2), Fρ[f]-Fρ[m] ≤ K Iρ[f], where Fρ[f]=∫ρ/f^r dx is the free energy, m=γρ^{1/(r+1)} is the normalized equilibrium, and Iρ[f] is the dissipation rate -d/dt Fρ[f] = r²∫u|∇(u^{-(r+1)})|² m dx. Three elementary ingredients make the estimate work: the Taylor identity θ^{-(r+2)} that converts the free-energy gap into a weighted L2 distance; the Poincaré inequality, applied to the zero-mean function u-1, which converts that L2 distance into the gradient L2 energy; and the pointwise inequality u ≤ C, which converts the gradient energy into the dissipation. Gronwall's lemma then converts the differential form of (2) into the exponential L2 deca

What would settle it

For a fixed Gaussian ρ, take a sequence of band functions u=f/m in [c,C] with sharp transitions between c and C and evaluate the ratio (Fρ[f]-Fρ[m])/Iρ[f]. If this ratio is unbounded, estimate (2) fails and with it the paper's mechanism; if it stays bounded, compute the supremum and compare with the claimed K. A direct numerical simulation of (1) in one dimension with Gaussian ρ would then confirm or refute the predicted e^{-t/K} decay.

Watch

Extended reading notes

Core claim

The central claim is the functional inequality Fρ[f]-Fρ[m] ≤ K Iρ[f] holding for every f in the band c m ≤ f ≤ C m, with K = C_P C^{2r+3} / (2r(r+1) γ^{r+1} c^{r+2}). The proof is a three-step chain. Taylor expansion of x^{-r} around 1 turns the free-energy gap into (r(r+1)/(2γ^{r+1}))∫θ^{-(r+2)}|u-1|² m dx, with θ between u and 1, so the gap is equivalent to the squared L2(m) distance of u to 1. Applying the Poincaré inequality to u-1, whose mean under m is zero, bounds that distance by ∫|∇u|² m dx. A pointwise comparison then bounds ∫|∇u|² m dx by the dissipation Iρ[f] using u ≤ C. Gronwall's lemma turns the resulting differential inequality into exponential decay of ∫|f/m-1|² m dx with ra

Load-bearing premise

The proof collapses if the solution leaves the fixed band c m ≤ f ≤ C m—all constants in the rate depend on keeping u=f/m between c and C for all time—or if the equilibrium fails the Poincaré inequality.

Editorial extensions

If this is right

Every solution of (1) with initial data in P^{c,C} converges to the equilibrium m in L2(m) with explicit exponential rate e^{-t/K}, with K given by the displayed constant.
The result covers Gaussian equilibria in any dimension, answering the open problems left by [FI25].
The compactly supported case recovers [IPS19, Theorem 1.4] as a special case.
The same proof applies to any equilibrium m with finite Poincaré constant; log-concave and some non-log-concave measures qualify by the cited results.
Because the argument uses only the Poincaré inequality and pointwise band bounds, the rate constant is explicit and the method does not rely on one-dimensional structure.

Reading between the lines

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

The proof's only dimension-dependent input is the Poincaré constant C_P; for Gaussian measures this constant is known explicitly, so the exponential rate could be made fully explicit without further estimates.
The persistence of the band c ≤ f/m ≤ C is handled by an appendix sketch that adapts the one-dimensional argument through compact balls. A fully self-contained proof of this persistence in R^n would be a natural companion, and any new equilibrium measure would need that step verified.
The same free-energy-gap-plus-Poincaré template may apply to other diffusion equations whose entropy gap is quadratic in f/m; the ultrafast structure is used only through the explicit form of the dissipation Iρ[f].
The constant K grows like c^{-(r+2)} and C^{2r+3}, so the uniform exponential rate deteriorates as the allowed band widens; the theorem is sharp in the band parameter but does not claim a uniform rate over all initial data.

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

Here's my take. The paper delivers on its promise: a short, Poincaré-based proof of exponential convergence for weighted ultrafast diffusion, bypassing the optimal transport machinery of Fathi–Iacobelli. That gives the first result for Gaussian weights in higher dimensions and answers an open problem. The proof of Theorem 5 is a clean Taylor-expansion-plus-Poincaré argument, and Lemma 1's two-sided bounds are standard. The dependence on Bobkov's theorem and the existing well-posedness theory is explicit and appropriate.

The main problem is a factor error in the entropy dissipation. The paper defines Iρ as -d/dt Fρ and writes Iρ = r^2 ∫ u |∇(u^{-(r+1)})|^2 m dx. But from Fρ = γ^{-(r+1)} ∫ u^{-r} m, direct computation gives -dF/dt = r^2 γ^{-2(r+1)} ∫ u |∇(u^{-(r+1)})|^2 m dx. So the integral they call Iρ is γ^{2(r+1)} times the true dissipation. This is not a minor notational slip: the K in Theorem 5 is used in Theorem 6's exponential rate e^{-t/K}. With the corrected identity, the rate becomes e^{-t/(K γ^{2(r+1)})}, slower by that factor. The qualitative theorem survives because the same Gronwall argument works with the corrected constant, but Theorem 6 as stated is wrong.

The second soft spot is Appendix A. The well-posedness in P^{c,C} on R^n is a condensed version of the FI25 argument using a diagonal compactness argument and results from IPS19. It's plausible, but the pointwise bounds c ≤ u ≤ C for all t are load-bearing — both Lemma 1 and Theorem 5 use them — so this appendix needs close checking. I'd want a referee to verify that the maximum principle and compactness argument really carry over. I don't see an obvious break, but it's not a full proof as written.

Overall, the core idea is solid and useful. The factor error is fixable in a revision, and the well-posedness gap is a matter of detail rather than concept. I'd send it to peer review and ask for a corrected rate and a fleshed-out appendix. Once those land, it's a nice addition to the entropy-methods literature.

Referee Report

2 major / 5 minor

Summary. The paper studies the ultrafast diffusion equation ∂t f = -r div(f ∇(ρ/f^{r+1})) on R^n with r>1 and probability weight ρ. It defines the equilibrium density m = γρ^{1/(r+1)} and considers solutions starting in the class P^{c,C} = {cm ≤ f ≤ Cm}. The main results are: (1) a two-sided bound (Lemma 1) relating the functional difference Fρ[f]-Fρ[m] to the L^2(m) distance between u=f/m and 1; (2) a dissipation estimate (Theorem 5) of the form Fρ[f]-Fρ[m] ≤ K Iρ[f] via the Poincaré inequality; and (3) exponential convergence to equilibrium (Theorem 6) with rate e^{-t/K}. The appendix sketches well-posedness and persistence in P^{c,C} by approximating with problems on balls and passing to the limit. The paper claims to extend the one-dimensional, Gaussian-excluded results of Fathi–Iacobelli to Gaussian measures in higher dimensions.

Significance. If correct, the paper would provide a strikingly simple proof of exponential convergence for ultrafast diffusion with log-concave weights, in particular Gaussian measures in any dimension. The method is transparent: a pointwise Taylor estimate plus a Poincaré inequality, with no fitted parameters. This would be a meaningful advance over the optimal-transport arguments in [FI25], which are restricted to one-dimensional non-Gaussian weights. However, the exact rate stated in Theorem 6 is incorrect owing to a factor of γ^{2(r+1)} in the entropy-production identity, and the persistence proof in Appendix A is only a sketch. The qualitative conclusion (exponential convergence) is very likely salvageable with a corrected constant, but the paper as written needs revision.

major comments (2)

[Section 1, Eq. (2) and definition of Iρ] The identity Iρ[f] = -d/dt Fρ[f] is off by a factor γ^{2(r+1)}. Direct computation from m = γρ^{1/(r+1)} gives Fρ[f] = γ^{-(r+1)}∫ u^{-r} m dx and ρ f^{-(r+1)} = γ^{-(r+1)} u^{-(r+1)}. Therefore -d/dt Fρ[f] = r^2 ∫ f |∇(ρ f^{-(r+1)})|^2 dx = r^2 γ^{-2(r+1)} ∫ u |∇(u^{-(r+1)})|^2 m dx. Thus the quantity called Iρ in Theorem 5 is γ^{2(r+1)} times the actual entropy production. The proof of Theorem 5 establishes Fρ[f]-Fρ[m] ≤ K Iρ[f] = K γ^{2(r+1)}(-d/dt Fρ[f]). Consequently the decay rate e^{-t/K} in Theorem 6 is too fast; the correct rate is e^{-t/(Kγ^{2(r+1)})} (or equivalently K should be multiplied by γ^{2(r+1)}). The qualitative exponential convergence survives after this correction, but the quantitative statement as written is incorrect.
[Appendix A (persistence in P^{c,C})] Remark 2 and the proofs of Lemma 1 and Theorem 5 rely on the solution f(t) remaining in P^{c,C} for all t. The appendix is only a sketch: it constructs approximating problems on B(0,k), cites [IPS19] for existence and compactness, and then states that 'by a diagonal argument' the limit f ∈ P^{c,C} solves (1). This is not a complete proof. The diagonal limit is local, and the uniform bounds c ≤ f/m ≤ C on all of R^n do not follow without additional control at infinity. Moreover, the truncated densities mk have discontinuities at ∂B(0,k), and the constants ak,bk require uniform estimates as k→∞. Since the constants k1,k2, and K in the main theorems all depend on c and C, the central claim is conditional on this persistence. The authors should either supply a rigorous proof of the persistence or state it as an explicit assumption backed by a precise reference.

minor comments (5)

[Appendix A] The definition of Vk is written with set-builder notation '{ akV, in B(0,k), +∞, otherwise }'; this is presumably meant as a piecewise definition and should be typeset accordingly.
[Abstract and Introduction] The phrase 'Gaussian-excluded one-dimensional weights' is awkward; it should be clarified whether [FI25] excludes Gaussian weights or the method excludes them.
[Title] The title contains a typographical artifact: 'ULTRAF AST' should be 'ULTRAFAST'.
[Proof of Lemma 1] In Equation (3), the Taylor expansion uses θ between min{1,u} and max{1,u}; it would be clearer to state θ = θ(x) depends on x. This is harmless.
[Theorem 6] The constant in the exponential should be updated once the Iρ factor is fixed; also the notation 'Fρ[f0] - Fρ[m] = ∫ ρ/f0^r dx - 1/γ^{r+1}' is correct but the reader must remember Fρ[m]=1/γ^{r+1}; consider writing it explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the proof is self-contained given standard external ingredients (Poincaré inequality, well-posedness from [IPS19]/[FI25]); the possible gamma-factor issue is a correctness concern, not circularity.

full rationale

The paper's derivation chain is: set u=f/m; Lemma 1 Taylor-expands Fρ[f]-Fρ[m] and bounds it by ∫|u-1|^2 m dx using only the pointwise bound c≤u≤C; Theorem 5 applies the Poincaré inequality to u-1 (which has mean zero) and uses the elementary pointwise inequality |∇u|^2 ≤ C^{2r+3}(r+1)^{-2} u|∇(u^{-(r+1)})|^2, both explicitly stated; Theorem 6 then integrates the resulting differential inequality. There are no fitted parameters, no data subset used for calibration, and no step presupposes the exponential decay being proved. The imported ingredients are external: the Poincaré inequality for log-concave measures (Bobkov) and well-posedness/persistence of the class P^{c,C} from [IPS19] and [FI25], neither authored by the present authors and neither containing the target Gaussian exponential-convergence result. Appendix A is a standard compactness/localization argument: it invokes [IPS19] for existence, maximum principle, compactness, and L1-contractivity, not for the main theorem. The skeptical reader's gamma-factor concern about the displayed entropy-dissipation identity Iρ is a correctness issue: direct computation from (1) gives -d/dt Fρ = r^2 γ^{-2(r+1)}∫ u|∇(u^{-(r+1)})|^2 m dx, so the constant K in Theorem 5 may be off by γ^{2(r+1)}. But that is an algebraic error, not circularity: the paper defines Iρ as -d/dt Fρ, and the proof does not define the target decay rate in terms of itself or fit any parameter to the conclusion. Therefore the circularity burden is essentially zero.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters fitted to data and no invented entities. The central claim rests on two external ingredients: the Poincaré inequality for m (a theorem for log-concave measures) and the well-posedness and persistence of bounds from [IPS19] and [FI25], which is summarized rather than fully proved. All other steps are elementary calculus.

assumptions (3)

domain assumption The equilibrium measure m satisfies a Poincaré inequality with constant C_P.
Used in the proof of Theorem 5. For log-concave measures (Gaussians) this is the theorem of Bobkov [Bob99], cited in Remark 4; it is an external input, not derived.
domain assumption Solutions f(t) to (1) exist and remain in P^{c,C} for all t >= 0, preserving the pointwise bounds c <= f/m <= C.
Invoked in Remark 2 and Appendix A; established by truncating to balls, applying [IPS19, Theorem 1.2], a maximum principle [IPS19, Corollary 3.8], and a diagonal argument. This is load-bearing for all constants, and the proof is only sketched.
standard math Taylor expansion with integral remainder for x^{-r} around 1.
Used in Lemma 1 to rewrite u^{-r}-1 as r(r+1)/2 theta^{-(r+2)}(u-1)^2; standard.

how reviews work

0 comments

Cite this review

Pith. "Pith review of On the exponential convergence to equilibrium for ultrafast diffusion equations." pith.science (2026). https://pith.science/paper/R6KZUQDN

@misc{pith2026250907382,
  author       = {Pith},
  title        = {Pith review of: On the exponential convergence to equilibrium for ultrafast diffusion equations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R6KZUQDN}},
  note         = {Machine review of arXiv:2509.07382}
}

abstract

We propose a simple proof of the exponential convergence to equilibrium for ultrafast diffusion equations in $\mathbb{R}^n$. Our approach, based on the direct use of Poincar\'e inequality, gets rid of the optimal transport arguments used in \cite{fathi2025} which are valid for Gaussian-excluded one-dimensional weights. This simplification allows us to extend their results to Gaussian measures in higher dimensions.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

8 extracted references · 8 canonical work pages

[1]

Invariances in variance estimates

Franck Barthe and Dario Cordero-Erausquin. Invariances in variance estimates. Proceedings of the London Mathematical Society , 106(1):33--64, 2013

work page 2013
[2]

Weighted fast diffusion equations (Part I): Sharp asymptotic rates without symmetry and symmetry breaking in Caffarelli-Kohn-Nirenberg inequalities

Matteo Bonforte, Jean Dolbeault, Matteo Muratori, and Bruno Nazaret. Weighted fast diffusion equations (Part I): Sharp asymptotic rates without symmetry and symmetry breaking in Caffarelli-Kohn-Nirenberg inequalities . Kinetic and Related Models , 10(1):33--59, 2017

work page 2017
[3]

Weighted fast diffusion equations (Part II): Sharp asymptotic rates of convergence in relative error by entropy methods

Matteo Bonforte, Jean Dolbeault, Matteo Muratori, and Bruno Nazaret. Weighted fast diffusion equations (Part II): Sharp asymptotic rates of convergence in relative error by entropy methods. Kinetic and Related Models , 10(1):61--91, 2017

work page 2017
[4]

Intertwinings, second-order brascamp--lieb inequalities and spectral estimates

Michel Bonnefont and Ald \'e ric Joulin. Intertwinings, second-order brascamp--lieb inequalities and spectral estimates. Studia Mathematica , 260:285--316, 2021

work page 2021
[5]

Sergei G. Bobkov. Isoperimetric and analytic inequalities for log-concave probability measures. The Annals of Probability , 27(4):1903--1921, 1999

work page 1903
[6]

Exponential convergence for ultrafast diffusion equations with log-concave weights

Max Fathi and Mikaela Iacobelli. Exponential convergence for ultrafast diffusion equations with log-concave weights. arXiv:2507.13060 , 2025

work page Pith review arXiv 2025
[7]

Asymptotic analysis for a very fast diffusion equation arising from the 1D quantization problem

Mikaela Iacobelli. Asymptotic analysis for a very fast diffusion equation arising from the 1D quantization problem. Discrete and Continuous Dynamical Systems , 39(9):4929--4943, 2019

work page 2019
[8]

Patacchini, and Filippo Santambrogio

Mikaela Iacobelli, Francesco S. Patacchini, and Filippo Santambrogio. Weighted ultrafast diffusion equations: From well-posedness to long-time behaviour. Archive for Rational Mechanics and Analysis , 232(3):1165--1206, 2019

work page 2019

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

Invariances in variance estimates

Franck Barthe and Dario Cordero-Erausquin. Invariances in variance estimates. Proceedings of the London Mathematical Society , 106(1):33--64, 2013

work page 2013

[2] [2]

Weighted fast diffusion equations (Part I): Sharp asymptotic rates without symmetry and symmetry breaking in Caffarelli-Kohn-Nirenberg inequalities

Matteo Bonforte, Jean Dolbeault, Matteo Muratori, and Bruno Nazaret. Weighted fast diffusion equations (Part I): Sharp asymptotic rates without symmetry and symmetry breaking in Caffarelli-Kohn-Nirenberg inequalities . Kinetic and Related Models , 10(1):33--59, 2017

work page 2017

[3] [3]

Weighted fast diffusion equations (Part II): Sharp asymptotic rates of convergence in relative error by entropy methods

Matteo Bonforte, Jean Dolbeault, Matteo Muratori, and Bruno Nazaret. Weighted fast diffusion equations (Part II): Sharp asymptotic rates of convergence in relative error by entropy methods. Kinetic and Related Models , 10(1):61--91, 2017

work page 2017

[4] [4]

Intertwinings, second-order brascamp--lieb inequalities and spectral estimates

Michel Bonnefont and Ald \'e ric Joulin. Intertwinings, second-order brascamp--lieb inequalities and spectral estimates. Studia Mathematica , 260:285--316, 2021

work page 2021

[5] [5]

Sergei G. Bobkov. Isoperimetric and analytic inequalities for log-concave probability measures. The Annals of Probability , 27(4):1903--1921, 1999

work page 1903

[6] [6]

Exponential convergence for ultrafast diffusion equations with log-concave weights

Max Fathi and Mikaela Iacobelli. Exponential convergence for ultrafast diffusion equations with log-concave weights. arXiv:2507.13060 , 2025

work page Pith review arXiv 2025

[7] [7]

Asymptotic analysis for a very fast diffusion equation arising from the 1D quantization problem

Mikaela Iacobelli. Asymptotic analysis for a very fast diffusion equation arising from the 1D quantization problem. Discrete and Continuous Dynamical Systems , 39(9):4929--4943, 2019

work page 2019

[8] [8]

Patacchini, and Filippo Santambrogio

Mikaela Iacobelli, Francesco S. Patacchini, and Filippo Santambrogio. Weighted ultrafast diffusion equations: From well-posedness to long-time behaviour. Archive for Rational Mechanics and Analysis , 232(3):1165--1206, 2019

work page 2019