pith. sign in

arxiv: 2606.01683 · v1 · pith:VZYI663Unew · submitted 2026-06-01 · 🧮 math.PR · math.SP

Singular-value gap of nonreversible Markov processes

Ruochuan Xu This is my paper

Pith reviewed 2026-06-28 13:12 UTC · model grok-4.3

classification 🧮 math.PR math.SP
keywords singular-value gapnonreversible Markov processesspectral gapPoisson equationfinite-time varianceempirical averagesMCMC samplingdiffusion operators
0
0 comments X

The pith

A positive singular-value gap ensures the generator of a nonreversible Markov process is invertible on the L2 space orthogonal to constants, which solves the Poisson equation and supplies uniform upper and lower bounds on finite-time varian

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

The paper seeks to extend the spectral gap concept to nonreversible Markov processes by introducing the singular-value gap as a quantity that controls convergence of empirical averages. When this gap is positive, the generator becomes invertible away from constant functions, allowing the Poisson equation to be solved for mean-zero right-hand sides in L2. This invertibility directly yields bounds on the variance of finite-time averages that hold uniformly across all square-integrable observables. A reader would care because the result supplies concrete tools for analyzing convergence and variance in settings where reversibility fails, such as certain MCMC dynamics or diffusion processes whose spectra are hard to analyze directly.

Core claim

The singular-value gap characterizes the convergence of empirical averages for nonreversible Markov processes by supplying upper and lower bounds for finite-time variance that are uniform over L2-functions. When the gap is positive, the generator is invertible on the L2-orthogonal complement of constant functions; in particular the Poisson equation -Lf = g can be solved, which enables the variance bounds and connects the results to asymptotic variance and associated central limit theorems. The gap is also compared with the spectral gap of the reversibilized process, total-variation mixing time, and the Cheeger constant.

What carries the argument

the singular-value gap, a spectral quantity generalizing the spectral gap of reversible generators to the nonreversible setting and governing generator invertibility on mean-zero L2 functions

If this is right

  • When the singular-value gap is positive the Poisson equation -Lf = g admits a solution for every mean-zero g in L2.
  • Finite-time variance of empirical averages admits uniform upper and lower bounds that apply to every square-integrable observable.
  • The gap supplies a criterion for variance reduction when sampling observables in MCMC.
  • The gap can identify slow-mixing mechanisms that are invisible to the reversibilized spectral gap.
  • Convergence of empirical averages can be certified for diffusion operators whose spectrum is otherwise difficult to analyze.

Where Pith is reading between the lines

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

  • The uniform variance bounds may yield new concentration inequalities that apply directly to nonreversible chains without passing through reversibilization.
  • Direct computation of the gap on specific infinite-state diffusions would test whether the finite-state definition continues to control invertibility in continuous settings.
  • Comparing the gap against the Cheeger constant on the same family of processes could clarify when the singular-value approach detects mixing bottlenecks missed by conductance methods.

Load-bearing premise

The singular-value gap, as constructed for finite-state chains, extends to general nonreversible Markov processes and their generators while preserving the invertibility property and the uniform variance bounds.

What would settle it

A concrete nonreversible process or generator for which the singular-value gap is positive yet the Poisson equation -Lf = g admits no solution in the mean-zero L2 space, or for which the finite-time variance bounds fail to hold uniformly over L2-functions.

read the original abstract

We consider a generalization of the spectral gap of reversible Markov generators to nonreversible processes, following the recent work arXiv:2310.10876 on nonreversible finite-state Markov chains. Extending Chatterjee's observations, we find that this spectral quantity that we call the \textit{singular-value gap} characterizes the convergence of empirical averages, providing upper and lower bounds for finite-time variance uniformly over $L^2$-functions. A key observation is that when the singular-value gap is positive, the generator is invertible on the $L^2$-orthogonal complement of constant functions. In particular, the Poisson equation $-Lf = g$ can be solved, which enables our proof and connects our results to asymptotic variance and associated central limit theorems. We also compare the singular-value gap with the spectral gap of the reversibilized process, the mixing time in total-variation distance, and the Cheeger constant. Several examples are provided throughout the text. Among other potential applications of the singular-value gap, these examples illustrate that a positive singular-value gap can help with variance reduction for observable classes in MCMC sampling, uncover slow-mixing mechanisms, and certify convergence of empirical averages for diffusion operators with complicated spectrum.

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.

Referee Report

2 major / 2 minor

Summary. The paper extends the singular-value gap from finite-state nonreversible Markov chains (following arXiv:2310.10876 and Chatterjee) to general nonreversible Markov processes and their generators. It claims that a positive singular-value gap implies the generator L is invertible on the L²-orthogonal complement of constants, so that the Poisson equation -Lf = g is solvable for mean-zero g ∈ L²; this yields uniform upper and lower bounds on finite-time variance of empirical averages over all L² functions, with connections to asymptotic variance and CLTs. The gap is compared to the spectral gap of the reversibilized process, total-variation mixing time, and Cheeger constant, with examples illustrating applications to MCMC variance reduction, slow-mixing detection, and diffusion operators.

Significance. If the extension is rigorously justified, the uniform L² variance bounds and Poisson-equation link supply a concrete tool for nonreversible processes where classical spectral gaps are unavailable or uninformative. The examples credibly demonstrate potential uses in variance reduction and convergence certification for complicated spectra.

major comments (2)
  1. [section extending the finite-state construction (likely §3)] The central claim that a positive singular-value gap (defined via the finite-state construction) implies L is invertible on the L² complement of constants, enabling solution of -Lf = g, is load-bearing for the variance bounds. The functional-analytic conditions (closedness of L, density of the domain, and that the gap yields a uniform resolvent bound) are not automatic for unbounded generators on general state spaces; the manuscript must explicitly verify these for the diffusion examples rather than invoking the finite-state definition directly.
  2. [diffusion examples section] § on diffusion operators: the claim that the singular-value gap certifies convergence for operators with complicated spectrum requires an explicit check that the gap controls the operator norm on the complement after accounting for the domain of the generator; without this, the uniform finite-time variance bounds do not follow.
minor comments (2)
  1. [preliminaries/definition section] Clarify the precise definition of the singular-value gap for general (possibly unbounded) generators, ideally with an equation number.
  2. [comparison section] Add a short remark on how the L² setting interacts with the total-variation mixing-time comparison.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and valuable feedback on the functional-analytic aspects of our extension. We address each major comment below and have revised the manuscript accordingly to strengthen the rigor for general state spaces.

read point-by-point responses

  1. Referee: [section extending the finite-state construction (likely §3)] The central claim that a positive singular-value gap (defined via the finite-state construction) implies L is invertible on the L² complement of constants, enabling solution of -Lf = g, is load-bearing for the variance bounds. The functional-analytic conditions (closedness of L, density of the domain, and that the gap yields a uniform resolvent bound) are not automatic for unbounded generators on general state spaces; the manuscript must explicitly verify these for the diffusion examples rather than invoking the finite-state definition directly.

    Authors: We agree with the referee that these functional-analytic conditions are crucial and not automatic. In the revised version, we have expanded §3 to include a detailed verification of the closedness of the generator L, the density of its domain in L², and the derivation of a uniform resolvent bound from the singular-value gap. This is achieved by adapting the finite-state argument using the theory of strongly continuous semigroups for Markov processes. For the diffusion examples, we now provide explicit checks under the standard ellipticity and growth conditions on the coefficients, confirming that the gap indeed implies the required invertibility and resolvent estimate. revision: yes

  2. Referee: [diffusion examples section] § on diffusion operators: the claim that the singular-value gap certifies convergence for operators with complicated spectrum requires an explicit check that the gap controls the operator norm on the complement after accounting for the domain of the generator; without this, the uniform finite-time variance bounds do not follow.

    Authors: We have revised the diffusion operators section to include an explicit verification that the singular-value gap controls the operator norm on the L²-orthogonal complement of constants, taking into account the domain of the generator. Specifically, we demonstrate that the gap provides a bound on the resolvent, which directly yields the uniform finite-time variance bounds. This check is performed for the specific diffusion operators in the examples, ensuring the claims hold rigorously. revision: yes

Circularity Check

0 steps flagged

Minor self-citation to finite-state precursor; central extension remains independent

full rationale

The paper defines the singular-value gap by explicit extension of the finite-state construction in the cited arXiv:2310.10876 and then derives invertibility of L on the L2 complement plus uniform variance bounds from that definition. No equation reduces a claimed prediction or uniqueness result to a fitted parameter or to the same paper's output by construction; the cited prior work supplies the base definition while the present manuscript supplies the functional-analytic transfer to general generators. This is a standard, non-load-bearing self-citation pattern that does not force the main claims.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be extracted.

pith-pipeline@v0.9.1-grok · 5733 in / 1005 out tokens · 20185 ms · 2026-06-28T13:12:52.398897+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 10 canonical work pages

  1. [1]

    The Annals of Applied Probability , number =

    Sourav Chatterjee , title =. The Annals of Applied Probability , number =. 2025 , doi =

    2025

  2. [2]

    Klaus-Jochen Engel and Rainer Nagel , title =

  3. [3]

    Andreas Eberle , title =

  4. [4]

    2024 , eprint=

    Non-reversible lifts of reversible diffusion processes and relaxation times , author=. 2024 , eprint=

    2024

  5. [5]

    2018 , eprint=

    Concentration inequalities for Markov chains by Marton couplings and spectral methods , author=. 2018 , eprint=

    2018

  6. [6]

    Duncan, A. B. and Lelièvre, T. and Pavliotis, G. A. , year=. Variance Reduction Using Nonreversible Langevin Samplers , volume=. Journal of Statistical Physics , publisher=

  7. [7]

    Electronic Journal of Probability , publisher =

    Sunder Sethuraman and Srinivasa Varadhan , title =. Electronic Journal of Probability , publisher =. 2005 , URL =

    2005

  8. [8]

    Bounds on the L^2 Spectrum for Markov Chains and Markov Processes: A Generalization of Cheeger's Inequality , volume =

    Gregory Lawler and Alan Sokal , journal =. Bounds on the L^2 Spectrum for Markov Chains and Markov Processes: A Generalization of Cheeger's Inequality , volume =

  9. [9]

    2017 , publisher=

    Markov Chains and Mixing Times , author=. 2017 , publisher=

    2017

  10. [10]

    Irreversible Langevin samplers and variance reduction: a large deviations approach , volume=

    Rey-Bellet, Luc and Spiliopoulos, Konstantinos , year=. Irreversible Langevin samplers and variance reduction: a large deviations approach , volume=. Nonlinearity , publisher=

  11. [11]

    On the eigenfunctions of the complex Ornstein–Uhlenbeck operators , volume=

    Chen, Yong and Liu, Yong , year=. On the eigenfunctions of the complex Ornstein–Uhlenbeck operators , volume=. Kyoto Journal of Mathematics , publisher=. doi:10.1215/21562261-2693451 , number=

    work page doi:10.1215/21562261-2693451

  12. [12]

    2025 , eprint=

    Bernstein-type Inequalities for Markov Chains and Markov Processes: A Simple and Robust Proof , author=. 2025 , eprint=

    2025

  13. [13]

    Communications in Mathematical Physics , volume=

    Central limit theorem for additive functionals of reversible Markov processes and applications to simple exclusions , author=. Communications in Mathematical Physics , volume=. 1986 , publisher=

    1986

  14. [14]

    2014 , publisher=

    Analysis and Geometry of Markov Diffusion Operators , author=. 2014 , publisher=

    2014

  15. [15]

    2012 , publisher=

    Fluctuations in Markov Processes: Time Symmetry and Martingale Approximation , author=. 2012 , publisher=

    2012

  16. [16]

    ALEA, Latin American Journal of Probability and Mathematical Statistics , volume=

    Central limit theorems for additive functionals of ergodic Markov diffusions processes , author=. ALEA, Latin American Journal of Probability and Mathematical Statistics , volume=. 2012 , url =

    2012

  17. [17]

    Aldous and James A

    David J. Aldous and James A. Fill. Reversible Markov Chains and Random Walks on Graphs. 2014

    2014

  18. [18]

    and Nier, F

    Lelièvre, T. and Nier, F. and Pavliotis, G. A. , year=. Optimal Non-reversible Linear Drift for the Convergence to Equilibrium of a Diffusion , volume=. Journal of Statistical Physics , publisher=. doi:10.1007/s10955-013-0769-x , number=

    work page doi:10.1007/s10955-013-0769-x

  19. [19]

    1972 , publisher=

    Handbook of mathematical functions: with formulas, graphs, and mathematical tables , author=. 1972 , publisher=

    1972

  20. [20]

    Diaconis and L

    Accelerating Gaussian diffusions , author=. The Annals of Applied Probability , volume=. 1993 , publisher=. doi:10.1214/aoap/1177005371 , url=

    work page doi:10.1214/aoap/1177005371 1993

  21. [21]

    Neal , title =

    Persi Diaconis and Susan Holmes and Radford M. Neal , title =. The Annals of Applied Probability , number =. 2000 , doi =

    2000

  22. [22]

    The Annals of Applied Probability , number =

    Persi Diaconis and Laurent Saloff-Coste , title =. The Annals of Applied Probability , number =. 1993 , doi =

    1993

  23. [23]

    Accelerating Diffusions , volume =

    Chii-Ruey Hwang and Shu-Yin Hwang-Ma and Shuenn-Jyi Sheu , journal =. Accelerating Diffusions , volume =

  24. [24]

    Roberts and Jeffrey S

    Gareth O. Roberts and Jeffrey S. Rosenthal , title =. Probability Surveys , publisher =. 2004 , URL =

    2004

  25. [25]

    Yet Another Look at Harris' Ergodic Theorem for Markov Chains

    Hairer, Martin and Mattingly, Jonathan C. Yet Another Look at Harris' Ergodic Theorem for Markov Chains. Seminar on Stochastic Analysis, Random Fields and Applications VI. 2011

    2011

  26. [26]

    Lifting Markov chains to speed up mixing , year =

    Chen, Fang and Lov\'. Lifting Markov chains to speed up mixing , year =. doi:10.1145/301250.301315 , booktitle =

    work page doi:10.1145/301250.301315

  27. [27]

    Kontoyiannis and S

    I. Kontoyiannis and S. P. Meyn , title =. The Annals of Applied Probability , number =. 2003 , doi =

    2003

  28. [28]

    , title =

    Kontoyiannis, Ioannis and Meyn, Sean P. , title =. Probability Theory and Related Fields , year =. doi:10.1007/s00440-011-0373-4 , url =

    work page doi:10.1007/s00440-011-0373-4

  29. [29]

    Annales de la Facult\'e des sciences de Toulouse : Math\'ematiques , pages =

    Diaconis, Persi and Miclo, Laurent , title =. Annales de la Facult\'e des sciences de Toulouse : Math\'ematiques , pages =. 2013 , publisher =. doi:10.5802/afst.1383 , zbl =

    work page doi:10.5802/afst.1383 2013

  30. [30]

    Gerald Teschl , title =

  31. [31]

    Hobert and Galin L

    James P. Hobert and Galin L. Jones , title =. Statistical Science , number =. 2001 , doi =

    2001

  32. [32]

    Curvature, concentration and error estimates for Markov chain Monte Carlo , volume=

    Joulin, Aldéric and Ollivier, Yann , year=. Curvature, concentration and error estimates for Markov chain Monte Carlo , volume=. The Annals of Probability , publisher=. doi:10.1214/10-aop541 , number=

    work page doi:10.1214/10-aop541

  33. [33]

    Zeitschrift f

    On the functional central limit theorem and the law of the iterated logarithm for Markov processes , author=. Zeitschrift f. 1982 , publisher=

    1982

  34. [34]

    2014 , isbn =

    O'Donnell, Ryan , title =. 2014 , isbn =

    2014

  35. [35]

    Acta Mathematica , publisher =

    Lars H. Acta Mathematica , publisher =. 1967 , doi =

    1967

  36. [36]

    Linda Preiss Rothschild and E. M. Stein , title =. Acta Mathematica , publisher =. 1976 , doi =

    1976

  37. [37]

    I. H. Dinwoodie , title =. Journal of Mathematical Analysis and Applications , volume =. 1998 , url =

    1998

  38. [38]

    The Annals of Applied Probability , number =

    James Allen Fill , title =. The Annals of Applied Probability , number =. 1991 , doi =

    1991

  39. [39]

    Journal of Theoretical Probability , volume =

    Kavita Ramanan and Aaron Smith , title =. Journal of Theoretical Probability , volume =. 2018 , url =

    2018

  40. [40]

    ESAIM: Probability and Statistics , pages =

    Lezaud, Pascal , title =. ESAIM: Probability and Statistics , pages =. 2001 , publisher =

    2001

  41. [41]

    2002 , issn =

    Ergodicity for SDEs and approximations: locally Lipschitz vector fields and degenerate noise , journal =. 2002 , issn =. doi:https://doi.org/10.1016/S0304-4149(02)00150-3 , url =

    work page doi:10.1016/s0304-4149(02)00150-3 2002

  42. [42]

    2015 , issn =

    Variance reduction for diffusions , journal =. 2015 , issn =. doi:https://doi.org/10.1016/j.spa.2015.03.006 , url =

    work page doi:10.1016/j.spa.2015.03.006 2015

  43. [43]

    Electronic Journal of Probability , publisher =

    George Deligiannidis and Magda Peligrad and Sergey Utev , title =. Electronic Journal of Probability , publisher =. 2015 , doi =

    2015

  44. [44]

    2005 , issn =

    Martingale approximations for continuous-time and discrete-time stationary Markov processes , journal =. 2005 , issn =. doi:https://doi.org/10.1016/j.spa.2005.04.001 , url =

    work page doi:10.1016/j.spa.2005.04.001 2005

  45. [45]

    2010 , publisher=

    Dirichlet Forms and Symmetric Markov Processes , author=. 2010 , publisher=

    2010

  46. [46]

    1995 , publisher =

    Perturbation Theory for Linear Operators , author =. 1995 , publisher =

    1995

  47. [47]

    Journal of Theoretical Probability , volume=

    Nash inequalities for finite Markov chains , author=. Journal of Theoretical Probability , volume=. 1996 , publisher=

    1996