Pith. sign in

REVIEW 2 major objections 5 minor 41 references

Relaxation times of non-reversible Markov processes

T0 review · 2 major / 5 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read Singular-value gaps of generators and two-point motions control L2-relaxation times of non-reversible Markov processes, with a collapse method that proves square-root speed-up for lifted walks.

desk verdict Solid, self-contained paper that proves the Diaconis–Miclo square-root speedup and gives a usable collapse framework for non-reversible L2 relaxation. read the letter →

arxiv 2607.10801 v1 pith:ZUD4VMXK submitted 2026-07-12 math.PR math.STstat.COstat.TH

Andreas Eberle , Francis Lörler This is my paper · ORCID
classification math.PRmath.STstat.COstat.TH MSC 60J2560J2760J6047D07
keywords singularvaluegapnon-reversiblerelaxationtimeliftcollapseliftedrandomwalktwo-pointmotionhypocoercivity
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

For non-reversible Markov processes the usual spectral gap no longer equals the L2-relaxation time, so the paper replaces it by the singular-value gap of the generator. That gap is shown to be equivalent (up to universal constants) to the relaxation time of the time-averaged transition semigroup, and the corresponding gap of the two-point motion lower-bounds the ordinary spectral gap while also controlling non-averaged relaxation from regular initial laws. To obtain concrete lower bounds on these gaps when the noise is degenerate, the authors introduce first- and second-order collapses of the generator onto a low-frequency subspace; under three abstract conditions (partial dissipativity, a gap for the second-order collapse, and a controlled interaction) they derive explicit upper bounds on the averaged relaxation time. The method is simple enough to cover continuous-time Markov chains as well as diffusions and piecewise-deterministic processes, and it yields sharp constants for lifted random walks on abelian groups (confirming a square-root speed-up conjecture), for switching flows, and for Langevin dynamics.

What carries the argument

The singular-value gap s(L) = inf ||Lf||/||f|| together with the first- and second-order collapses C1, C2 of L onto a low-frequency subspace; the three abstract conditions (A1) partial dissipativity, (B2) gap of C2, and (C2) controlled interaction via the remainder operator A turn these collapses into explicit bounds via a lifted Poincaré inequality.

What would settle it

For the lifted random walk on the discrete torus with reorientation rate gamma proportional to 1/n, compute or tightly bound the true L2-relaxation time (averaged or non-averaged) and check whether it is of order n (ballistic) rather than n^2; a substantially larger order would falsify both the collapse bounds and the claimed square-root speed-up.

Watch

Extended reading notes

Core claim

The inverse of the singular-value gap s(L) is equivalent, up to universal constants, to the L2-relaxation time of the time-averaged semigroup. Moreover the singular-value gap of the two-point motion L^(2) lower-bounds the ordinary spectral gap of L and controls non-averaged relaxation from regular initial measures. Under the three conditions (A1)+(B2)+(C2) on first- and second-order collapses one obtains matching upper bounds on these quantities.

Load-bearing premise

The three abstract conditions on partial dissipativity, the second-order collapse gap, and especially the controlled interaction between high- and low-frequency subspaces must hold with constants of the right order; they are verified case-by-case and can introduce extra dimension factors that are not always sharp.

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.

Referee Report

2 major / 5 minor

Summary. The paper develops a systematic L^{2} theory for non-reversible Markov processes based on the singular-value gap s(L) of the generator. It proves that 1/s(L) is equivalent (up to universal constants) to the relaxation time of the time-averaged semigroup (Theorem 7), that the singular-value gap of the two-point motion lower-bounds the ordinary spectral gap (Theorem 13) and controls non-averaged relaxation from regular initial laws (Theorem 14), and that an abstract two-component inequality under partial dissipativity (A1), a singular-value gap of the second-order collapse (B2), and a controlled interaction condition (C2) yields explicit upper bounds (Theorem 18 and corollaries). The framework allows non-vanishing first-order collapse and is applied to lifted random walks on abelian groups (resolving the Diaconis–Miclo square-root speed-up conjecture in the form of matching lower bounds on the gap and relaxation time), switching flows with noise, and (perturbed) Langevin dynamics, with sharp upper and lower bounds obtained by direct verification of the abstract conditions.

Significance. The work supplies a clean, broadly applicable replacement for classical spectral-gap and Poincaré methods in the non-reversible setting, together with a simpler alternative to existing hypocoercivity techniques that does not require vanishing first-order collapse. The resolution of the Diaconis–Miclo conjecture for lifted walks on abelian groups is a concrete advance of independent interest; the same machinery recovers (and slightly extends) the best known L^{2} rates for Langevin dynamics with a short proof. All main theorems are proved in full by elementary Hilbert-space arguments and explicit spectral or Bochner estimates; the variational characterizations are parameter-free and the applications reduce to direct verification of (A1)–(C2). These features make the paper a substantial contribution to quantitative Markov-process theory.

major comments (2)
  1. The abstract and introduction advertise upper bounds on non-averaged relaxation times via s(L^{(2)}). Theorems 14 and 22 deliver such bounds only after imposing Hilbert–Schmidt regularity on the initial law and after paying logarithmic factors (and, for lifted walks, an extra dimension factor). The gap between the sharp averaged bounds of Theorem 21 and the non-averaged bounds of Theorem 22 is already noted by the authors, but the limitation should be stated more prominently next to the main claims so that readers do not over-interpret the scope of the non-averaged control.
  2. Theorem 23 confirms the lower bound on gap(L) of order |V|^{-1} sqrt(gap(\Delta)) and the matching lower bound on t_rel, thereby establishing optimality of the square-root speed-up for relaxation times. The matching upper bound on gap(L) itself (without log factors) remains open with the present method. A short paragraph clarifying precisely which half of Conjecture 1.3 of Diaconis–Miclo is settled, and which half is left open, would strengthen the claim of “proof of a conjecture”.
minor comments (5)
  1. Corollary 20(ii): the final display contains “1 + 2C_{1}^{2}/\gamma^{2}”; the constant should be C_{2} (consistent with the statement of (C2)).
  2. Equation (1) and the surrounding discussion: the factor e^{2}/(e-1) is elementary but could be recorded once with a one-line proof for completeness.
  3. Section 3.1, after (16): the explicit constant 16e max(dn^{2} min(1,\gamma)/20,(1+4d)/\gamma) is useful; a brief remark that the numerical prefactors are not optimised would prevent readers from treating them as sharp.
  4. Notation: the same symbol t_rel is used for both averaged and non-averaged times (with and without overline). A consistent typographic distinction throughout (already present in places) would improve readability.
  5. References: the recent works of Xu (arXiv:2606.01683) and Huang–Li on singular-value gaps could be cross-cited more explicitly when the singular-value gap is introduced.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: pure functional-analytic derivations from definitions of singular-value gaps and collapses, with self-citations only to prior theoretical frameworks that are independently reproved or extended here.

full rationale

The paper is a self-contained mathematical development. Singular-value gap s(L) is defined variationally (Def. 2, Eq. 5) from the generator; Theorems 7, 13 and 14 then prove equivalences and bounds to relaxation times by elementary operator-norm and spectral arguments (proofs in §4.1–4.2 use only the key identity LP_t f = (P_t f – f)/t and product-semigroup properties). The two-component method (Def. 15, Thm. 17–18) introduces first- and second-order collapses C_i as the restrictions of the symmetrised forms E_i^L to a subspace H^l; the abstract inequalities (A1)+(B2)+(C2) are verified by direct computation on each concrete generator (lifted walks, switching flows, Langevin), never fitted to data. Self-citations ([17,18] on second-order lifts and space-time Poincaré) supply motivational context and terminology, but the new proofs deliberately bypass the divergence lemma of those works and supply independent estimates (e.g., Lemmas 35–37). No parameter is calibrated to a target quantity and then re-used as a “prediction”; no uniqueness theorem is imported to force the ansatz; the Diaconis–Miclo conjecture is resolved by explicit test functions and the abstract bounds, not by renaming a known result. Minor self-citation therefore does not load-bear, yielding score 1.

Assumptions & free parameters 0 free parameters · 3 assumptions · 1 invented entities

Pure functional-analytic derivation on Polish spaces with invariant probability measures. No free parameters are fitted; all constants that appear are either universal or explicit spectral quantities of the collapses. The only background axioms are standard L2 spectral theory and the definition of the generator of a Markov semigroup.

assumptions (3)
  • domain assumption The state space is Polish and the process admits a strongly continuous contraction semigroup on L2(μ) with invariant probability μ.
    Stated in the opening paragraph of Section 1; used throughout to guarantee that the generator is densely defined and that the singular-value gap is well-defined.
  • standard math D0 is a core for the restriction of L to Dom(L)∩L2_0(μ).
    Standard core assumption for unbounded operators; invoked in Definition 2 to make the variational characterisations of gap(L), λ(L) and s(L) independent of the choice of core.
  • domain assumption The orthogonal projection Π onto the low-frequency subspace Hl maps the core D0 into itself.
    Required for the collapses C1, C2 to be well-defined as operators on Hl (Definition 15); verified case-by-case in every example.
invented entities (1)
  • symmetric first- and second-order collapse (C1, C2) independent evidence
    purpose: Capture the short-time expansion of the semigroup on the low-frequency subspace so that hypocoercivity-type bounds can be proved without vanishing first-order terms.
    Defined in Definition 15; the whole two-component method rests on them. They are operators derived from L, not new physical objects, and their spectral gaps are computed explicitly in each example.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Relaxation times of non-reversible Markov processes." pith.science (2026). https://pith.science/paper/ZUD4VMXK

@misc{pith2026260710801,
  author       = {Pith},
  title        = {Pith review of: Relaxation times of non-reversible Markov processes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZUD4VMXK}},
  note         = {Machine review of arXiv:2607.10801}
}
abstract

We develop a systematic approach to quantify $L^2$-relaxation times for non-reversible Markov processes based on the singular value gap of the generator introduced by Chatterjee. The inverse of the singular value gap is equivalent to the relaxation time of the time-averaged transition semigroup. We show that, moreover, the singular value gap of the two-point motion also provides a lower bound on the usual spectral gap of the generator, and its inverse provides upper bounds on relaxation times without time averaging for sufficiently regular initial laws. We then introduce a method for deriving lower bounds on singular value gaps for Markov processes with degenerate noise that is based on the concept of a first- and second-order collapse of the generator. It follows ideas from hypocoercivity developed in a previous series of works but is simpler and more broadly applicable. In contrast to previous results, it includes settings with non-vanishing first-order collapse, and thus applies directly to Markov chains (in continuous time), but also to diffusion processes and piecewise-deterministic Markov processes. Our approach yields sharp upper and lower bounds for several classes of examples. First applications include the proof of a conjecture by Diaconis and Miclo on a square-root speed-up for lifted random walks on abelian groups, as well as bounds on relaxation times of switching flows perturbed by noise and of non-reversible diffusion processes.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 11 canonical work pages

  1. [1]

    Albritton, S

    D. Albritton, S. Armstrong, J.-C. Mourrat and M. Novack. ‘Variational methods for the kinetic Fokker-Planck equation’. In:Anal. & PDE(July 2024).doi:10. 2140/apde.2024.17.1953

    2024

  2. [2]

    Andrieu, A

    C. Andrieu, A. Durmus, N. N¨ usken and J. Roussel. ‘Hypocoercivity of piecewise deterministic Markov process-Monte Carlo’. In:Ann. Appl. Probab.31.5 (2021), pp. 2478–2517.issn: 1050-5164,2168-8737.doi:10 . 1214 / 20 - aap1653.url: https://doi.org/10.1214/20-aap1653

    work page doi:10.1214/20-aap1653 2021

  3. [3]

    Bakry, I

    D. Bakry, I. Gentil and M. Ledoux.Analysis and Geometry of Markov Diffu- sion Operators. Vol. 348. Grundlehren der mathematischen Wissenschaften. Cham: Springer, 2014.isbn: 978-3-319-00227-9.doi:10.1007/978-3-319-00227-9

    doi:10.1007/978-3-319-00227-9 2014

  4. [4]

    Bernard, M

    ´E. Bernard, M. Fathi, A. Levitt and G. Stoltz. ‘Hypocoercivity with Schur com- plements’. In:Ann. Henri Lebesgue5 (2022), pp. 523–557.doi:10.5802/ahl.129. 42

    work page doi:10.5802/ahl.129 2022

  5. [5]

    Brigati, F

    G. Brigati, F. L¨ orler and L. Wang. ‘Hypocoercivity meets lifts’. In:Kinet. Relat. Models20.0 (2026), pp. 34–55.doi:10.3934/krm.2025020

    work page doi:10.3934/krm.2025020 2026

  6. [6]

    Y. Cao, J. Lu and L. Wang. ‘On explicitL 2-convergence rate estimate for under- damped Langevin dynamics’. In:Arch. Ration. Mech. Anal.247.90 (Aug. 2023). doi:10.1007/s00205-023-01922-4

    doi:10.1007/s00205-023-01922-4 2023

  7. [7]

    Chatterjee

    S. Chatterjee. ‘Spectral gap of nonreversible Markov chains’. In:Ann. Appl. Probab. 35.4 (Aug. 2025).doi:10.1214/25-AAP2183

    work page doi:10.1214/25-aap2183 2025

  8. [8]

    F. Chen, L. Lov´ asz and I. Pak. ‘Lifting Markov Chains to Speed up Mixing’. In: Proceedings of the Thirty-First Annual ACM Symposium on Theory of Computing. STOC ’99 (May 1999), pp. 275–281.doi:10.1145/301250.301315

    doi:10.1145/301250.301315 1999

Show all 41 references
  1. [9]

    Diaconis, S

    P. Diaconis, S. Holmes and R. M. Neal. ‘Analysis of a nonreversible Markov chain sampler’. In:Ann. Appl. Probab.10.3 (2000), pp. 726–752.doi:10.1214/aoap/ 1019487508

    2000 doi

  2. [10]

    Diaconis and L

    P. Diaconis and L. Miclo. ‘On the spectral analysis of second-order Markov chains’. In:Annales de la Facult´ e des sciences de Toulouse : Math´ ematiquesSer. 6, 22.3 (2013), pp. 573–621.doi:10.5802/afst.1383

    2013 doi

  3. [11]

    Dolbeault, C

    J. Dolbeault, C. Mouhot and C. Schmeiser. ‘Hypocoercivity for kinetic equations with linear relaxation terms’. In:C. R. Acad. Sci. Paris, Ser. I347.9 (Apr. 2009), pp. 511–516.doi:10.1016/j.crma.2009.02.025

    2009 doi

  4. [12]

    Dolbeault, C

    J. Dolbeault, C. Mouhot and C. Schmeiser. ‘Hypocoercivity for linear kinetic equa- tions conserving mass’. In:Trans. Am. Math. Soc.367.6 (June 2015), pp. 3807– 3828.doi:10.1090/S0002-9947-2015-06012-7

    2015 doi

  5. [13]

    Duane, A

    S. Duane, A. D. Kennedy, B. J. Pendleton and D. Roweth. ‘Hybrid Monte Carlo’. In:Phys. Lett. B195.2 (1987), pp. 216–222.doi:10.1016/0370-2693(87)91197- X

    1987 doi

  6. [14]

    Markov processes

    A. Eberle.Lecture Notes “Markov processes”. Institute for Applied Mathematics, University of Bonn.http://wt.iam.uni-bonn.de/eberle/home/. 2025

    2025

  7. [15]

    Eberle, A

    A. Eberle, A. Guillin, L. Hahn, F. L¨ orler and M. Michel.Convergence of non- reversible Markov processes via lifting and flow Poincar´ e inequality. Mar. 2025. arXiv:2503.04238 [math.AP]

    2025 arXiv

  8. [16]

    Eberle, A

    A. Eberle, A. Guillin and R. Zimmer. ‘Couplings and quantitative contraction rates for Langevin dynamics’. In:Ann. Probab.47.4 (2019), pp. 1982–2010.doi: 10.1214/18-AOP1299

    2019 doi

  9. [17]

    Eberle and F

    A. Eberle and F. L¨ orler. ‘Non-reversible lifts of reversible diffusion processes and relaxation times’. In:Probab. Theory Relat. Fields(Aug. 2024).doi:10.1007/ s00440-024-01308-x

    2024

  10. [18]

    Eberle and F

    A. Eberle and F. L¨ orler. ‘Space-time divergence lemmas and optimal non-reversible lifts of diffusions on Riemannian manifolds with boundary’. In:J. Funct. Anal. 291.9 (2026).doi:10.1016/j.jfa.2026.111605. 43

    2026 doi

  11. [19]

    Engel and R

    K.-J. Engel and R. Nagel.One-Parameter Semigroups for Linear Evolution Equa- tions. Vol. 194. Graduate Texts in Mathematics. New York: Springer, Oct. 1999. isbn: 978-0-387-98463-6.doi:10.1007/b97696

    1999 doi

  12. [20]

    Z. Fan, B. Li and J. Lu.Sharp hypocoercive convergence estimates for under- damped Langevin dynamics via the modifiedL 2 method. 2026. arXiv:2604.10068 [math.AP]

    2026 arXiv

  13. [21]

    Y. Guo. ‘The Landau equation in a periodic box’. In:Comm. Math. Phys.231.3 (2002), pp. 391–434.doi:10.1007/s00220-002-0729-9

    2002 doi

  14. [22]

    H´ erau and F

    F. H´ erau and F. Nier. ‘Isotropic hypoellipticity and trend to equilibrium for the Fokker–Planck equation with a high-degree potential’. In:Arch. Ration. Mech. Anal.171.2 (2004), pp. 151–218

    2004

  15. [23]

    A. M. Horowitz. ‘A generalized guided Monte Carlo algorithm’. In:Phys. Lett. B 268.2 (Oct. 1991), pp. 247–252.doi:10.1016/0370-2693(91)90812-5

    1991 doi

  16. [24]

    Huang and X

    D. Huang and X. Li. ‘Bernstein-type inequalities for Markov chains and Markov processes: A simple and robust proof’. In:Bernoulli32.2 (2026), pp. 1267–1284. doi:10.3150/25-BEJ1901

    2026 doi

  17. [25]

    Hwang, S.-Y

    C.-R. Hwang, S.-Y. Hwang-Ma and S.-J. Sheu. ‘Accelerating Diffusions’. In:Ann. Appl. Prob.15.2 (May 2005), pp. 1433–1444.doi:10.1214/105051605000000025

    2005 doi

  18. [26]

    Krauth.Hamiltonian Monte Carlo vs

    W. Krauth.Hamiltonian Monte Carlo vs. event-chain Monte Carlo: an appraisal of sampling strategies beyond the diffusive regime. 2024. arXiv:2411.11690

    2024 arXiv

  19. [27]

    Leli` evre, F

    T. Leli` evre, F. Nier and G. Pavliotis. ‘Optimal Non-reversible Linear Drift for the Convergence to Equilibrium of a Diffusion’. In:J. Stat. Phys.152 (June 2013), pp. 237–274.doi:10.1007/s10955-013-0769-x

    2013 doi

  20. [28]

    D. A. Levin, Y. Peres and E. L. Wilmer.Markov Chains and Mixing Times. 2nd ed. Providence, RI: American Mathematical Society, 2017.isbn: 978-1-4704-2962-1. doi:10.1090/mbk/107

    2017 doi

  21. [29]

    R. Li, H. Zha and M. Tao. ‘Hessian-Free High-Resolution Nesterov Acceleration For Sampling’. In:Proceedings of the 39th International Conference on Machine Learning. Ed. by K. Chaudhuri et al. Vol. 162. Proceedings of Machine Learning Research. PMLR, July 2022, pp. 13125–13162

    2022

  22. [30]

    L¨ orke.Lifting Markov chains to speed up mixing

    T. L¨ orke.Lifting Markov chains to speed up mixing. Institute for Applied Math- ematics, University of Bonn. Bachelor Thesis. 2025

    2025

  23. [31]

    Lu.A sharp hypocoercive entropy decay estimate for underdamped Langevin dynamics

    J. Lu.A sharp hypocoercive entropy decay estimate for underdamped Langevin dynamics. 2026. arXiv:2605.01933 [math.AP]

    2026 arXiv

  24. [32]

    Lu and L

    J. Lu and L. Wang. ‘On explicitL 2-convergence rate estimate for piecewise de- terministic Markov processes in MCMC algorithms’. In:Ann. Appl. Probab.32.2 (Apr. 2022), pp. 1333–1361.doi:10.1214/21-AAP1710

    2022 doi

  25. [33]

    Montenegro and P

    R. Montenegro and P. Tetali.Mathematical Aspects of Mixing Times in Markov Chains. Vol. 1. 3. Now Publishers, 2006, pp. 237–354.doi:10.1561/0400000003. 44

    2006 doi

  26. [34]

    R. M. Neal.Improving Asymptotic Variance of MCMC Estimators: Non-reversible Chains are Better. July 2004. arXiv:math/0407281 [math.PR]

    2004 arXiv

  27. [35]

    Nesterov

    Y. Nesterov. ‘A method for unconstrained convex minimization problem with the rate of convergenceO(1/k 2)’. In:Dokl. Akad. Nauk. SSSR. Vol. 269. 3. 1983

    1983

  28. [36]

    V. V. Petrov.Sums of Independent Random Variables. De Gruyter, 1975.isbn: 978-3-112-57300-6.doi:10.1515/9783112573006

    1975 doi

  29. [37]

    Reed and B

    M. Reed and B. Simon.Methods of modern mathematical physics I: Functional analysis. Academic Press, 1980.isbn: 978-0-12-585050-6

    1980

  30. [38]

    D. Talay. ‘Stochastic Hamiltonian systems: exponential convergence to the invari- ant measure, and discretization by the implicit Euler scheme’. In:Markov Process. Relat. Fields8.2 (2002), pp. 163–198

    2002

  31. [39]

    Villani.Hypocoercivity

    C. Villani.Hypocoercivity. Vol. 202. Memoirs of the American Mathematical Soci- ety 950. Providence, RI: American Mathematical Society, 2009.isbn: 978-0-8218- 4498-4.doi:10.1090/S0065-9266-09-00567-5

    2009 doi

  32. [40]

    X. Wang, Y. Wang and L. Zhu.Sampling non-log-concave densities via Hessian- free high-resolution dynamics. 2026. arXiv:2601.02725 [math.PR]

    2026

  33. [41]

    Xu.Singular-value gap of nonreversible Markov processes

    R. Xu.Singular-value gap of nonreversible Markov processes. 2026. arXiv:2606. 01683 [math.PR]. 45

    2026

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.