Pith. sign in

REVIEW 7 minor 26 references

Mixing of Glauber Dynamics on High Overlap Gibbs Measures

T0 review · 0 major / 7 minor · reviewed 2026-07-10 · grok-4.5

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

Pith's one-line read For any temperature of the Sherrington–Kirkpatrick model, a large enough constant external field makes Glauber dynamics mix in polynomial time.

desk verdict Solid qualitative resolution of poly-time Glauber mixing for SK at every fixed β under n-independent field strength; the high-overlap correlation control is the real technical step. read the letter →

arxiv 2607.06813 v1 pith:GTJWDEXL submitted 2026-07-07 math.PR cond-mat.dis-nncs.DSmath.STstat.TH

Afonso S. Bandeira , Ahmed El Alaoui , Almut Rödder This is my paper · ORCID
classification math.PRcond-mat.dis-nncs.DSmath.STstat.TH MSC 60J1082B4482C20
keywords GlauberdynamicsSherrington–KirkpatrickmodelspectralgapstochasticlocalizationhighoverlapquadraticGibbsmeasuresexternalfieldmixingtime
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

The paper shows that quadratic Gibbs measures on the hypercube mix rapidly under Glauber dynamics once the external field is large enough to force high overlap between independent samples. High overlap reduces control of the full correlation matrix to control of covariances on small disagreement sets; those sets are then handled by a uniform bound on the operator norms of small principal submatrices of the interaction matrix. Stochastic localization with pinnings turns the uniform correlation bound into a spectral-gap lower bound, hence polynomial mixing. Applied to the Sherrington–Kirkpatrick model, the result yields: for every fixed inverse temperature β there exists a field strength θ that depends on β but not on system size n such that, with high probability over the GOE interaction matrix, lazy Glauber dynamics mixes in polynomial time. The same argument covers related models such as Z2-synchronization with side information.

What carries the argument

The high-overlap correlation bound (Lemma 2.2): after conditioning on a single-spin disagreement, two independent replicas disagree on at most a δ-fraction of coordinates with high probability; the spectral norm of the correlation matrix is then controlled by the covariances of the zero-field measures on those small disagreement sets, which are bounded by the small-submatrix hypothesis.

What would settle it

Fix any β and compute (or rigorously bound) the smallest θ such that the spectral gap of SK Glauber dynamics remains at least n^{-C} with high probability; if that θ must grow with n, or if for every constant θ the gap decays super-polynomially on a positive-density set of GOE matrices, the claimed n-independent threshold fails.

Watch

Extended reading notes

Core claim

If every principal submatrix of size at most δn of the interaction matrix J has operator norm at most 1/5, then there exists a field strength θ (depending on ∥J∥ and δ but independent of n) such that the quadratic Gibbs measure with external field θ1 mixes under lazy Glauber dynamics in time polynomial in n. For the Sherrington–Kirkpatrick model this hypothesis holds with high probability for any fixed β once δ is chosen small enough, so a constant external field of size depending only on β already guarantees polynomial mixing.

Load-bearing premise

The argument needs both a uniform bound on the norms of all small principal submatrices of the interaction and a quantitative guarantee that a large constant field forces nearly all spins to align even after any pinning of size up to (1-δ)n.

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

0 major / 7 minor

Summary. The paper proves that quadratic Gibbs measures on the hypercube with interaction matrix J satisfying a uniform small-principal-submatrix bound max_{|S|≤δn} ∥J_{S,S}∥ ≤ 1/5 mix under lazy Glauber dynamics in polynomial time once the external field strength θ is large enough (depending on ∥J∥ and δ but not on n). The argument controls correlation matrices of all pinned measures via a high-overlap reduction (Lemmas 2.1–2.2): when two replicas disagree on only a small set, the correlation norm is bounded by covariances of zero-field measures on those small sets; a large field forces the required overlap even after single-spin conditioning and worst-case pinnings (Lemma 2.3, Corollary 2.4, Proposition 3.2). Stochastic localization with pinnings (Lemma 3.1) then yields a polynomial spectral-gap lower bound. As the main application, for the Sherrington–Kirkpatrick model at any fixed finite β the GOE matrix satisfies the submatrix hypothesis with high probability for sufficiently small δ(β), so there exists θ=θ(β) independent of n for which Glauber mixes in poly(n) time w.h.p. (Corollary 1.2). A sketched extension covers Z₂-synchronization with side information.

Significance. The result settles a qualitative long-standing question for the SK model: for every inverse temperature β there is a constant external field (independent of system size) that restores polynomial-time Glauber mixing, even deep in the low-temperature regime. Prior spectral-radius criteria only reached β ≲ 0.3, while algorithmic stochastic localization reached the full RS regime β<1 only in Wasserstein distance or for modified samplers. The paper does not reach the Almeida–Thouless line, which it correctly flags as open, but the qualitative statement is new and the method—high-overlap control of pinned correlations plus stochastic localization—is clean, self-contained, and reusable. The GOE small-submatrix verification and the pinning-induced field bound are standard and carefully checked; the reduction of correlations to small-disagreement covariances is the main technical contribution and appears free of circularity.

minor comments (7)
  1. Throughout the introduction and Section 4 the extracted text contains numerous missing spaces (e.g., “Ontheotherhand”, “replicasymmetrycanhold”, “F urther Applications”, “highoverlap”). These should be corrected in the source before publication.
  2. Theorem 1.1 and the choice (9) give θ ≳ ∥J∥/δ. For the SK corollary it would help the reader to record the resulting dependence θ(β) (roughly O(β^{3}) under the δ ∼ 1/β^{2} of §3.3), even if constants are not optimized.
  3. Lemma 2.2: the probability hypothesis is stated with ε_m/m while the final additive error is ε_m; the proof correctly sets ε_m = 2m exp(-c_{3}m), but a one-line remark after (10) would make the bookkeeping transparent.
  4. In the proof of Theorem 1.1 the covariance bound (11) cites BB19/CE25 with the formula 1/(1-2∥2J_{S,S}∥). A parenthetical reminder that ∥J_{S,S}∥≤1/5 implies 2∥2J∥=4/5<1 (so the high-temperature hypothesis applies) would remove any momentary doubt.
  5. Corollary 4.1 is only sketched. Either supply a short appendix with the analogous C_{1}, C_{2} estimates or clearly label it as a corollary whose proof is identical to that of Corollary 1.2 up to the indicated changes.
  6. Notation: sdiam(J) is defined in §1.3 but never used; either drop it or employ it when invoking the EKZ22 spectral condition for the fully-pinned regime.
  7. The AI-use paragraph is unusually candid; if the journal style guide permits it, keep it, but move any non-scientific remarks out of the main text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the mixing bound is derived from independently verified small-submatrix norms and quantitative high-overlap under large field, without reducing the claim to a fit or self-citation by construction.

full rationale

The central claim (Theorem 1.1 / Corollary 1.2) is obtained by (i) verifying the hypothesis max_{|S|≤δn} ∥J_{S,S}∥ ≤ 1/5 for GOE via Gaussian concentration + union bound (Section 3.3), (ii) proving a spin-alignment lemma that a field θ ≳ ∥J∥/δ forces high replica overlap even after single-spin conditioning and worst-case pinnings of size (1-δ)n (Lemma 2.3, Corollary 2.4, Proposition 3.2), (iii) reducing the correlation-matrix operator norm to small-disagreement covariances via the high-overlap replica identity (Lemma 2.2), and (iv) feeding the resulting uniform κ_k bound into the black-box stochastic-localization spectral-gap estimate of CE25 (Lemma 3.1). None of these steps is definitional of the target mixing time, none fits a free parameter to data and re-labels it a prediction, and the load-bearing external tools (EKZ22 high-temperature mixing, BB19 covariance bound, CE25 localization) are cited as independent results whose hypotheses are checked inside the paper. Self-citations (e.g., BDL+26 open problems, EAMS22 context) are non-load-bearing. The argument is therefore self-contained against its stated assumptions.

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

The paper rests on standard spectral and concentration facts for GOE matrices, the stochastic-localization spectral-gap machinery of CE25, and elementary large-deviation estimates for quadratic Gibbs measures. No free parameters are fitted to data; θ is an existential constant whose size is controlled by explicit (if non-optimal) inequalities. No new physical entities are postulated.

assumptions (3)
  • standard math Operator-norm concentration of GOE principal submatrices: E[λ_max(√n W_{S,S})] ≤ 2√|S| and sub-Gaussian tails, used via union bound to verify the hypothesis of Theorem 1.1 for eta W.
    Invoked in the proof of Corollary 1.2; classical random-matrix fact.
  • domain assumption Stochastic localization with pinnings yields a spectral-gap lower bound once all correlation matrices of measures with at least T_6 free spins are uniformly bounded (Lemma 3.1, taken from CE25).
    Black-box tool from the cited localization literature; the paper verifies its hypotheses rather than re-proving the framework.
  • domain assumption High-temperature spectral-gap bound for quadratic Gibbs measures whose interaction has operator norm <1/2 (used when fewer than δn spins remain free).
    Cited from EKZ22/CE25; applied only on the small free-set regime where the paper's submatrix hypothesis supplies the norm bound.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mixing of Glauber Dynamics on High Overlap Gibbs Measures." pith.science (2026). https://pith.science/paper/GTJWDEXL

@misc{pith2026260706813,
  author       = {Pith},
  title        = {Pith review of: Mixing of Glauber Dynamics on High Overlap Gibbs Measures},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GTJWDEXL}},
  note         = {Machine review of arXiv:2607.06813}
}
abstract

We show fast mixing of Glauber dynamics for certain quadratic Gibbs measures with large external fields. The main ingredient is an overlap condition that allows us to control correlation matrices uniformly over all pinnings, by controlling norms of small submatrices of the interaction matrix. Using stochastic localization, we then obtain a lower bound on the spectral gap and, consequently, polynomial-time mixing of Glauber dynamics. As a direct application, we consider the Sherrington-Kirkpatrick model, whose interaction matrix is a scaled GOE matrix. For this model, we show that for any fixed finite inverse temperature $\beta$, there exists a strength of external field $\theta$, not depending on the size of the system, for which Glauber dynamics mixes in polynomial time (with high probability on the draw of the interaction matrix).

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

  1. [1]

    Contextual Stochastic Block Models , url =

    Deshpande, Yash and Sen, Subhabrata and Montanari, Andrea and Mossel, Elchanan , booktitle =. Contextual Stochastic Block Models , url =

    work page

  2. [2]

    Synchronization over

    Cucuringu, Mihai , journal =. Synchronization over. 2015 , doi =

    work page 2015

  3. [3]

    Applied and Computational Harmonic Analysis , volume =

    Angular Synchronization by Eigenvectors and Semidefinite Programming , author =. Applied and Computational Harmonic Analysis , volume =. 2011 , doi =

    work page 2011

  4. [4]

    2014 , doi =

    Decoding Binary Node Labels from Censored Edge Measurements: Phase Transition and Efficient Recovery , author =. 2014 , doi =

    work page 2014

  5. [5]

    2020 IEEE International Symposium on Information Theory (ISIT) , pages =

    Jean Barbier and Galen Reeves , title =. 2020 IEEE International Symposium on Information Theory (ISIT) , pages =. 2020 , month = jun, publisher =. doi:10.1109/ISIT44484.2020.9173970 , url =

    work page doi:10.1109/isit44484.2020.9173970 2020

  6. [6]

    Electronic Communications in Probability , volume=

    Bounds on the covariance matrix of the Sherrington--Kirkpatrick model , author=. Electronic Communications in Probability , volume=. 2024 , publisher=

    work page 2024

  7. [7]

    The two point function of the SK model without external field at high temperature , author=. Probab. Math. Phys , volume=

    work page

  8. [8]

    Duke Mathematical Journal , number =

    Yuansi Chen and Ronen Eldan , title =. Duke Mathematical Journal , number =. 2025 , doi =

    work page 2025

Show all 26 references
  1. [9]

    SIAM Journal on Computing , volume=

    Spectral independence in high-dimensional expanders and applications to the hardcore model , author=. SIAM Journal on Computing , volume=. 2021 , publisher=

    2021

  2. [10]

    Anari, Nima and Koehler, Frederic and Vuong, Thuy-Duong , TITLE =. S. [2024] 2024 , ISBN =. doi:10.1145/3618260.3649622 , URL =

    2024 doi

  3. [11]

    Bauerschmidt, Roland and Bodineau, Thierry , TITLE =. J. Funct. Anal. , FJOURNAL =. 2019 , NUMBER =. doi:10.1016/j.jfa.2019.01.007 , URL =

    2019 doi

  4. [12]

    Journal of Physics A: Mathematical and General , volume=

    Stability of the Sherrington-Kirkpatrick solution of a spin glass model , author=. Journal of Physics A: Mathematical and General , volume=

  5. [13]

    arXiv preprint arXiv:2604.11921 , year=

    Replica symmetry up to the de Almeida-Thouless line in the Sherrington-Kirkpatrick model , author=. arXiv preprint arXiv:2604.11921 , year=

    arXiv

  6. [14]

    Communications in Mathematical Physics , volume=

    An iterative construction of solutions of the TAP equations for the Sherrington--Kirkpatrick model , author=. Communications in Mathematical Physics , volume=. 2014 , publisher=

    2014

  7. [15]

    Probability theory and related fields , volume=

    A spectral condition for spectral gap: fast mixing in high-temperature Ising models , author=. Probability theory and related fields , volume=. 2022 , publisher=

    2022

  8. [16]

    Communications in Mathematical Physics , volume=

    Correlation inequalities for Ising spin lattices , author=. Communications in Mathematical Physics , volume=. 1975 , publisher=

    1975

  9. [17]

    and Peres, Yuval , TITLE =

    Levin, David A. and Peres, Yuval , TITLE =. 2017 , PAGES =. doi:10.1090/mbk/107 , URL =

    2017 doi

  10. [18]

    Preprint available online: https://people

    Topics in Mathematics of Data Science , author=. Preprint available online: https://people. math. ethz. ch/ abandeira//BandeiraSingerStrohmer-MDS-draft. pdf , year=

  11. [19]

    arXiv preprint arXiv:2511.22621 , year=

    Exponentially Slow Mixing of the Low Temperature SK Model , author=. arXiv preprint arXiv:2511.22621 , year=

  12. [20]

    arXiv preprint arXiv:2304.11449 , year=

    Posterior sampling in high dimension via diffusion processes , author=. arXiv preprint arXiv:2304.11449 , year=

    arXiv

  13. [21]

    2024 IEEE 65th Annual Symposium on Foundations of Computer Science (FOCS) , pages=

    Locally stationary distributions: A framework for analyzing slow-mixing markov chains , author=. 2024 IEEE 65th Annual Symposium on Foundations of Computer Science (FOCS) , pages=. 2024 , organization=

    2024

  14. [22]

    The Annals of Statistics , volume=

    Local convexity of the tap free energy and amp convergence for z 2-synchronization , author=. The Annals of Statistics , volume=. 2023 , publisher=

    2023

  15. [23]

    2022 IEEE 63rd Annual Symposium on Foundations of Computer Science (FOCS) , pages=

    Sampling from the Sherrington-Kirkpatrick Gibbs measure via algorithmic stochastic localization , author=. 2022 IEEE 63rd Annual Symposium on Foundations of Computer Science (FOCS) , pages=. 2022 , organization=

    2022

  16. [24]

    The Annals of Probability , volume=

    Sudakov--Fernique post-AMP, and a new proof of the local convexity of the TAP free energy , author=. The Annals of Probability , volume=. 2024 , publisher=

    2024

  17. [25]

    arXiv preprint arXiv:2603.29571 , year=

    Randomstrasse101: Open Problems of 2025 , author=. arXiv preprint arXiv:2603.29571 , year=

    2025

  18. [26]

    arXiv preprint arXiv:2605.03718 , year=

    Potential Hessian Ascent III: Sampling the Sherrington--Kirkpatrick Model at Beta< 1/2 , author=. arXiv preprint arXiv:2605.03718 , year=

    arXiv

Pith tools

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