Pith. sign in

REVIEW 5 minor 21 references

An extension of a theorem of Chevet

T0 review · 0 major / 5 minor · reviewed 2026-07-13 · grok-4.5

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

Pith's one-line read A modern comparison for squares of projected soft-maxima of Gaussian vectors extends Chevet and is maximized by the regular simplex in a special case.

desk verdict Clean modern extension of Chevet's square comparison plus a natural new conjecture with a solid block-diagonal special case; ready for referees. read the letter →

arxiv 2607.09311 v1 pith:2FRKG2WY submitted 2026-07-10 math.FA math.PR

Olivier Guédon (LAMA) , Fabien Souli (LAMA) This is my paper · ORCID
classification math.FAmath.PR MSC 60G1546B0952A40
keywords GaussiancomparisonChevetinequalitysoft-maxsimplexmeanwidthorthogonalprojectioninterpolationmethodlog-sum-exp
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 modernizes Chevet's classical comparison of squared orthogonal projections of the maximum of a Gaussian vector. It proves that whenever one centered Gaussian vector has smaller pairwise distances than another, a large family of smooth functions that approximate the maximum (including the log-sum-exp soft-max) satisfy the same inequality for the square of the projection onto the orthogonal complement of the span of the coordinates. The same inequality is then used to study a new maximization problem: among Gaussian vectors with unit-variance coordinates, is the regular simplex the maximizer of these squared projected soft-maxima? The authors prove that it is, at least for a special block-diagonal family of covariances that includes all pairwise correlations equal to -1. The result sits next to the still-open simplex mean-width conjecture and supplies a new functional that the simplex is expected to maximize.

What carries the argument

An interpolation path Z_t = √(1-t)X + √t Y together with Gaussian integration by parts that yields an explicit formula for the derivative of the squared projected functional; the five sign axioms on f make the derivative non-negative, so the functional increases along the path.

What would settle it

Exhibit a pair of centered Gaussians with d_X ≤ d_Y for which the squared projected soft-max functional is larger for X than for Y, or find a C^{2} function that obeys the five axioms yet violates the comparison.

Watch

Extended reading notes

Core claim

If two centered Gaussians X and Y satisfy d_X(i,j) ≤ d_Y(i,j) for every pair and if f is a C^{2} moderate-growth function obeying five structural axioms (translation invariance by constants, non-negative first partials, non-positive mixed second partials, f ≥ max, and a sign condition that forces the mixed second derivatives to dominate the product of first derivatives), then the expected square of the projection of f onto the orthogonal complement of the coordinate span is smaller for X than for Y. The soft-max family f_eta satisfies the axioms with λ = 1/eta, recovering and extending Chevet's original statement for the true maximum.

Load-bearing premise

The five structural conditions that the test function must satisfy (especially the sign condition that ties first and second partials) are postulated rather than derived from a deeper geometric principle; they are checked by hand for log-sum-exp and a limited entropy-type family.

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.

Referee Report

0 major / 5 minor

Summary. The paper extends Chevet's comparison inequality for the squared orthogonal projections of the coordinatewise maximum of centered Gaussian vectors, using a modern interpolation-plus-differentiation argument. Theorem 1.3 states that if two centered Gaussians satisfy d_X(i,j) ≤ d_Y(i,j) for all pairs and if f is C^{2} of moderate growth obeying the five structural conditions (a) translation by constants, (b) non-negative first partials, (c) non-positive mixed second partials, (d) f ≥ max, and (e) a sign condition abla f_i abla f_j + \lambda abla^{2}_{ij}f ≤ 0, then E[(P_{H_X^ op}(f(X))+ ho)^{2}] ≤ E[(P_{H_Y^ op}(f(Y))+ ho)^{2}]. The log-sum-exp family f_eta (and a limited class of entropy-regularized suprema F_ ho) satisfy the hypotheses, recovering Chevet's original statement in the limit eta o\infty and yielding Corollary 1.4. A new conjecture (1.7) is formulated asserting that the regular-simplex Gaussian S maximises these functionals among unit-variance Gaussians; a special case (Theorem 1.8) is proved for block-diagonal covariances by reduction via Lemma 3.1 and a symmetry argument that shows the interpolating derivative is non-negative.

Significance. The result supplies a clean, fully rigorous extension of classical Slepian–Fernique–Chevet comparison inequalities to squared projections and to an explicit broader function class, obtained by elementary Gaussian integration-by-parts and path differentiation under moderate-growth hypotheses. The proofs are self-contained, avoid distribution theory, and verify the structural conditions by direct differentiation for the main examples. The partial resolution of the new simplex-type conjecture via the comparison theorem plus symmetry reductions is a concrete advance, even though a geometric interpretation of the new functionals is left open. These contributions are of clear interest in asymptotic geometric analysis and the theory of Gaussian processes.

minor comments (5)
  1. [§2.3] Page 9, line after (11): typographical error “Asssumption” (triple s).
  2. [global] Throughout: the orthogonal-projection notation P_{H^\perp_X} is inconsistently rendered (sometimes with extra spaces or missing subscripts); a uniform LaTeX macro would improve readability.
  3. [§3] In the statement of Theorem 1.8 the simplex vector S is understood to live in dimension 2n, but this is never made explicit; a one-line remark would avoid any momentary confusion with the n-dimensional definition (1).
  4. [Abstract / §1] Abstract and introduction speak of “n+1 vectors on S^{n-1}” while the body works with n-dimensional vectors; the dimension shift is standard for the simplex but could be flagged once for non-specialists.
  5. [§2.4] Proposition 2.5: the constant λ=(M+1)^{2}/eta is stated without a short verification that it is sharp or at least optimal for the given ho; a parenthetical remark would be helpful.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: comparisons follow from Gaussian integration-by-parts and an explicit interpolation path under stated structural hypotheses on f.

full rationale

The paper derives Theorem 1.3 by the classical interpolation Z_t = √(1-t)X + √t Y, obtains an explicit formula for φ'(t) via Gaussian integration-by-parts (Proposition 2.2, Lemma 2.1), and controls the sign of each term by the five listed structural conditions (a)–(e) on f. Those conditions are verified by direct differentiation for the log-sum-exp family (Lemma 2.4) and for a limited class of entropy-type suprema (Proposition 2.5); they are not defined in terms of the target functional. The special-case comparison for Conjecture 1.7 (Theorem 1.8) likewise reduces the problem, via the already-proved comparison, to a path whose derivative is shown non-negative by elementary symmetry reductions that are written out in full. No quantity is fitted to data and then re-presented as a prediction, no uniqueness theorem is imported from the authors’ prior work, and no load-bearing step collapses by construction to its own input. The derivation is therefore self-contained against external benchmarks.

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

The paper rests entirely on standard facts of Gaussian analysis (integration by parts, existence of the reproducing-kernel Hilbert space, moderate-growth differentiation under the integral) together with elementary calculus identities for the log-sum-exp family. No free parameters are fitted and no new physical or geometric entities are postulated.

assumptions (3)
  • standard math Gaussian integration-by-parts formula: E[X_i g(X)] = sum_j Σ_{ij} E[∂_j g(X)] for C^{1} moderate-growth g.
    Invoked repeatedly (Lemma 2.1, Proposition 2.2, Section 3) as the sole source of differentiation under the expectation.
  • domain assumption Moderate-growth condition on f and its derivatives up to order 2 guarantees that all expectations and path derivatives exist and can be interchanged.
    Stated at the beginning of Section 2 and used to justify the differentiation of φ(t).
  • standard math Positive-definite approximation: the comparison for singular covariances follows by continuity from the non-degenerate case.
    Explicitly invoked at the start of the proof of Theorem 1.3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An extension of a theorem of Chevet." pith.science (2026). https://pith.science/paper/2FRKG2WY

@misc{pith2026260709311,
  author       = {Pith},
  title        = {Pith review of: An extension of a theorem of Chevet},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2FRKG2WY}},
  note         = {Machine review of arXiv:2607.09311}
}
abstract

We adopt a modern approach to extend a comparison inequality of Chevet for Gaussian processes to a broader class of functions. Furthermore, we consider centered Gaussian random vectors associated with a family of $n+1$ vectors on the unit sphere $S^{n-1}$ and investigate configurations of the vertices for which these new functionals could be maximum for the regular simplex.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 1 linked inside Pith

  1. [1]

    Notes Math.Springer, Cham, 1974

    Albert Badrikian and Simone Chevet.Mesures cylindriques, espaces de Wiener et fonctions al´ eatoires gaussiennes, volume 379 ofLect. Notes Math.Springer, Cham, 1974

    1974

  2. [2]

    An extremal property of the mean width of the simplex.Math

    Franck Barthe. An extremal property of the mean width of the simplex.Math. Ann., 310(4):685– 693, 1998

    1998

  3. [3]

    An error bound in the Sudakov-Fernique inequality.Unpublished, see https://arxiv.org/pdf/math/0510424, 2005

    Sourav Chatterjee. An error bound in the Sudakov-Fernique inequality.Unpublished, see https://arxiv.org/pdf/math/0510424, 2005

    arXiv 2005

  4. [4]

    S. Chevet. Series de variables al´ eatoires Gaussiennes ` a valeurs dansE⊗ ε F. Application aux produits d’espaces de Wiener abstraits. Semin. Geom. des Espaces de Banach, Ec. polytech., Cent. Math., 1977-1978, Expose No. 19, 1-15 (1978)., 1978. AN EXTENSION OF A THEOREM OF CHEVET. 17

    1977

  5. [5]

    Processus Gaussiens et volumes mixtes.Z

    Simone Chevet. Processus Gaussiens et volumes mixtes.Z. Wahrscheinlichkeitstheorie und Verw. Gebiete, 36(1):47–65, 1976

    1976

  6. [6]

    J. M. Danskin. The theory of Max-Min and its application to weapons allocation problems. ¨Okonometrie und Unternehmensforschung. 5. Berlin-Heidelberg-New York: Springer-Verlag 126 p. with 6 fig. (1967)., 1967

    1967

  7. [7]

    Minorations des fonctions al´ eatoires gaussiennes.Ann

    Xavier Fernique. Minorations des fonctions al´ eatoires gaussiennes.Ann. Inst. Fourier, 24(2):61– 66, 1974

    1974

  8. [8]

    Majorization of Gaussian processes and geometric applications.Probab

    Yehoram Gordon. Majorization of Gaussian processes and geometric applications.Probab. The- ory Related Fields, 91(2):251–267, 1992

    1992

Show all 21 references
  1. [9]

    Kabluchko, A

    Z. Kabluchko, A. E. Litvak, and D. Zaporozhets. Mean width of regular polytopes and expected maxima of correlated Gaussian variables.Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI), 442:75–96, 2015

    2015

  2. [10]

    Analyse et prob- abilit´ es, volume 12 ofCours Sp´ ec

    Daniel Li and Herv´ e Queff´ elec.Introduction ` a l’´ etude des espaces de Banach. Analyse et prob- abilit´ es, volume 12 ofCours Sp´ ec. (Paris). Paris: Soci´ et´ e Math´ ematique de France, 2004

    2004

  3. [11]

    Daniel Li and Herv´ e Queff´ elec.Introduction to Banach spaces: analysis and probability. Volume

  4. [12]

    Translated from the French by Dani` ele Gibbons and Greg Gibbons, volume 167 ofCamb. Stud. Adv. Math.Cambridge: Cambridge University Press, 2018

    2018

  5. [13]

    Alexander E. Litvak. Around the simplex mean width conjecture. InAnalytic aspects of con- vexity, volume 25 ofSpringer INdAM Ser., pages 73–84. Springer, Cham, 2018

    2018

  6. [14]

    Marcus and Jay Rosen.Markov processes, Gaussian processes, and local times, volume 100 ofCamb

    Michael B. Marcus and Jay Rosen.Markov processes, Gaussian processes, and local times, volume 100 ofCamb. Stud. Adv. Math.Cambridge: Cambridge University Press, 2006

    2006

  7. [15]

    The discrepancy between min-max statistics of Gaussian and Gaussian-subordinated matrices.Stochastic Process

    Giovanni Peccati and Nicola Turchi. The discrepancy between min-max statistics of Gaussian and Gaussian-subordinated matrices.Stochastic Process. Appl., 158:315–341, 2023

    2023

  8. [16]

    Tyrrell Rockafellar and Roger J.-B

    R. Tyrrell Rockafellar and Roger J.-B. Wets.Variational analysis, volume 317 ofGrundlehren Math. Wiss.Berlin: Springer, 1998

    1998

  9. [17]

    A simple proof of the Gaussian correlation conjecture extended to some mul- tivariate gamma distributions.Far East J

    Thomas Royen. A simple proof of the Gaussian correlation conjecture extended to some mul- tivariate gamma distributions.Far East J. Theor. Stat., 48(2):139–145, 2014

    2014

  10. [18]

    Schechtman and M

    G. Schechtman and M. Schmuckenschl¨ ager. A concentration inequality for harmonic measures on the sphere. InGeometric aspects of functional analysis. Israel seminar (GAFA) 1992-94, pages 256–273. Basel: Birkh¨ auser, 1995

    1992

  11. [19]

    The one-sided barrier problem for Gaussian noise.Bell System Tech

    David Slepian. The one-sided barrier problem for Gaussian noise.Bell System Tech. J., 41:463– 501, 1962

    1962

  12. [20]

    V. N. Sudakov. Gaussian random processes and measures of solid angles in Hilbert space.Sov. Math., Dokl., 12:412–415, 1971

    1971

  13. [21]

    Volume I: Basic examples., volume 54 of Ergeb

    Michel Talagrand.Mean field models for spin glasses. Volume I: Basic examples., volume 54 of Ergeb. Math. Grenzgeb., 3. Folge. Berlin: Springer, 2nd revised and enlarged ed. edition, 2011. Univ Gustave Eiffel, Univ Paris Est Creteil, CNRS, LAMA UMR8050 F-77447 Marne-la-Vall´ee...

    2011

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.