Pith. sign in

REVIEW 4 minor 21 references

A finite Stein recursion gives exact joint moments of any number of disjoint Wishart principal minors for nonnegative integer powers.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 22:31 UTC pith:SHK2SH4F

load-bearing objection Clean finite recursion that actually solves Wilks’ 1934 integer-power problem for arbitrary block number; short proof, solid numerics, one external Stein identity.

arxiv 2607.03984 v1 pith:SHK2SH4F submitted 2026-07-04 math.PR math.STstat.TH

On Wilks' problem: Exact recursive formulas via Stein's method for the joint moments of disjoint principal minors of Wishart random matrices

Robert E. Gaunt , Fr\'ed\'eric Ouimet This is my paper
classification math.PR math.STstat.TH MSC 60E0562H10
keywords joint momentsprincipal minorsStein’s methodWilks’ problemWishart distributiondegree-lowering recursionmultivariate statistics
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Wilks asked in 1934 for the joint moments of several disjoint principal minors of a Wishart matrix and called the general case extremely complicated. Those moments appear throughout classical multivariate statistics: likelihood-ratio tests, generalized correlations, MANOVA and multivariate regression. This paper solves the problem whenever the exponents are nonnegative integers and the number of blocks is finite. It applies a recently obtained Stein identity for the Wishart law to the homogeneous polynomial formed by the product of the minors. The identity produces a degree-lowering operator that can be applied a finite number of times until a constant remains; that constant is exactly the desired moment. The recursion works for any positive-definite scale matrix and any admissible shape parameter, and Monte-Carlo checks confirm that the numbers match. The result therefore supplies exact benchmarks and closed recursive algorithms where only special-function expressions for two blocks or asymptotic approximations had been available.

Core claim

For a Wishart matrix W partitioned into d diagonal blocks and any multi-index n of nonnegative integers, the joint moment E[∏ |W_rr|^{n_r}] equals the constant obtained by applying the degree-lowering operator R_{α,Σ} exactly m_n = ∑ n_r p_r times to the monomial P_n(S) = ∏ |S_rr|^{n_r}. Equivalently, the moment is (1/m_n!) times the m_n-fold application of R_{α,Σ} evaluated at any matrix.

What carries the argument

The Wishart Stein identity (3.1) and the associated degree-lowering operator R_{α,Σ}P := α tr{Σ ∇P} + 2 tr{S ∇Σ ∇P}. Euler’s homogeneous-function identity converts the Stein relation into the one-step recursion E[P(W)] = (1/m) E[(R_{α,Σ}P)(W)], which terminates after finitely many steps because P_n is a polynomial.

Load-bearing premise

The whole recursion stands or falls on the Wishart Stein identity holding for the homogeneous determinant polynomials that appear; if that identity fails for these functions, the degree-lowering step collapses.

What would settle it

Implement the recursion for a concrete low-dimensional case (e.g., d=2, small integer powers) and compare the returned constant against both the known hypergeometric closed form for two blocks and a large Monte-Carlo sample; any systematic discrepancy larger than sampling error would falsify the claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Exact numerical values of the Wilks moments become available for any number of blocks and any nonnegative integer powers without series truncation.
  • The same recursion supplies exact cumulants and finite-sample benchmarks for likelihood-ratio statistics and generalized correlation coefficients that involve products of principal minors.
  • When the scale matrix is block-diagonal the recursion factors, recovering the classical product of univariate principal-minor moments as a consistency check.
  • The algorithm is completely deterministic once the input polynomials are represented, so it can serve as a ground-truth generator for testing asymptotic approximations in multivariate analysis.

Where Pith is reading between the lines

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

  • The same Stein operator may yield recursive formulas for moments of other non-nested functionals of Wishart matrices (e.g., products of non-principal minors or of eigenvalues) once a suitable generating polynomial is identified.
  • Because the recursion terminates only for integer powers, a natural next step is to examine whether a truncated series or Padé approximant built from the same operator can approximate non-integer moments.
  • The method suggests that other classical open moment problems for matrix-variate distributions possessing a Stein identity may likewise admit finite recursive solutions when the target is polynomial.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

0 major / 4 minor

Summary. The paper solves Wilks’ 1934 problem of evaluating joint moments E[∏_r |W_rr|^{n_r}] of disjoint principal minors of a Wishart matrix W ~ W_p(α, Σ) for nonnegative integer multi-indices n and any finite number of blocks. Using the Wishart Stein identity of Bailly et al. (2026), the authors introduce a degree-lowering operator R_{α,Σ} on homogeneous polynomials and prove (Theorem 3.1) that the target moment equals the constant obtained by applying R_{α,Σ} exactly m_n = ∑ n_r p_r times to P_n(S) = ∏_r |S_rr|^{n_r}. The one-step relation follows from Euler’s identity for the symmetric gradient together with the Stein characterization; termination is immediate for polynomials. Numerical tables for d = 2, 3, 4 compare the recursive values with Monte Carlo estimates of increasing sample size and show the expected convergence.

Significance. Wilks himself described the general (non-nested) case as “extremely complicated.” Prior exact results cover only the nested/embedded setting or the two-block case (via a terminating _2F_1 of matrix argument). The present recursion supplies the first exact, finite algorithm that works for arbitrary block number, arbitrary positive block sizes, every α > p-1 and every positive-definite scale matrix, when the exponents are nonnegative integers. The derivation is short, fully algebraic once the Stein identity is granted, and free of free parameters. Independent numerical corroboration against both Monte Carlo and the known d = 2 hypergeometric formula strengthens confidence. Exact integer-power moments are useful as benchmarks for likelihood-ratio asymptotics, generalized correlation coefficients and cumulant calculations in classical multivariate analysis.

minor comments (4)
  1. The paper rests on the forward implication of Bailly et al. (2026, Cor. 3.3). A one-sentence reminder that the identity is already known to hold for all polynomials (or a pointer to the precise hypothesis in that preprint) would make the dependence fully self-contained for readers who have not yet seen the companion work.
  2. Remark 3.1 correctly notes that intermediate polynomials leave the pure-product class. A brief remark on the practical growth of the monomial basis (or a pointer to a computer-algebra implementation) would help readers who wish to recompute the tables.
  3. Tables 4.1–4.3 report relative errors that decrease with Monte Carlo sample size, as expected. Adding a single column that records the exact d = 2 hypergeometric value (already verified offline) would make the independent check visible without lengthening the text.
  4. Notation for the symmetric gradient ∇ is carefully defined, but a short parenthetical that ∇_{ij} = (1/2)∂/∂S_{ij} for i eq j is the convention used throughout would prevent occasional misreading of the second-derivative term in (3.2).

Circularity Check

1 steps flagged

No significant circularity: recursion is algebraic once the external Stein identity is granted; self-citation is present but not load-bearing for the target moments.

specific steps
  1. self citation load bearing [Section 3, statement of (3.1) and derivation of (3.8)]
    "The Wishart Stein identity used below is a consequence of the forward implication in Corollary 3.3 of Bailly et al. (2026) and is formulated as follows: if W∼W_p(α,Σ) and f is a real-valued polynomial function on S_p, then E[2 tr{(αΣ−W)∇f(W)}+4 tr{W∇Σ∇f(W)}]=0. (3.1)"

    The entire recursion rests on this identity, which is imported from a co-authored preprint rather than re-derived. Once granted, however, the remainder of the argument (Euler identity → one-step recursion → finite termination to a constant) is self-contained algebraic manipulation and does not redefine the target moments in terms of the cited result. The citation is therefore present but not circular in the strong sense of forcing the claimed prediction by construction.

full rationale

The paper's central claim (Theorem 3.1) is that E[∏ |W_rr|^{n_r}] equals the constant obtained by applying the degree-lowering operator R_{α,Σ} exactly m_n times to the homogeneous polynomial P_n. The one-step relation E[P(W)] = (1/m) E[(R P)(W)] follows from the Wishart Stein identity (3.1) plus Euler's homogeneous-function identity (3.7), both of which are elementary once (3.1) is granted. Identity (3.1) is taken from the forward implication of Bailly et al. (2026, Cor. 3.3), a co-authored preprint; this is ordinary self-citation of a characterization, not a uniqueness theorem that forbids alternatives, nor a fitted parameter renamed as a prediction. The target moments are never defined in terms of the cited works' constants, and the recursion terminates by pure degree counting for integer powers. Numerical Monte-Carlo tables and the independent d=2 hypergeometric formula of Genest et al. (2025) supply external corroboration. No step reduces the claimed moment to an input by construction. Score 1 reflects only the minor, non-load-bearing self-citation of the Stein characterization.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 1 invented entities

Pure-probability derivation. No free parameters are fitted. The load-bearing external input is the Wishart Stein identity of Bailly et al. (2026); everything else is standard multivariate analysis or elementary homogeneous-function calculus. The operator R is a definitional packaging, not a new physical entity.

axioms (4)
  • domain assumption Wishart Stein identity: for W ~ W_p(α,Σ) and polynomial f, E[2 tr{(αΣ−W)∇f(W)}+4 tr{W ∇Σ ∇f(W)}]=0 (Bailly et al. 2026, Cor. 3.3).
    Invoked at the start of §3 as equation (3.1); the entire recursion is obtained by specializing this identity to homogeneous determinant polynomials.
  • standard math Euler homogeneous-function identity tr{S ∇P(S)}=m P(S) for P homogeneous of degree m on the space of symmetric matrices.
    Proved in the paper (eqs. 3.5–3.7) from the chain rule under the symmetric-gradient convention; used to convert the Stein identity into the one-step moment recursion (3.8).
  • domain assumption Definition and basic properties of the central Wishart law W_p(α,Σ) for α>p−1 and Σ positive definite (density, Laplace transform).
    Standard background (Muirhead, Anderson); used to define the target expectation and to justify that the Stein identity applies.
  • standard math R_{α,Σ} maps the space H_m of homogeneous polynomials of degree m≥1 into H_{m−1}.
    Immediate from differentiation lowering degree and multiplication by S raising it by one; stated after (3.2) and used for termination after m_n steps.
invented entities (1)
  • Degree-lowering operator R_{α,Σ} no independent evidence
    purpose: Packages the first- and second-order terms of the Wishart Stein identity into a single map that reduces total degree by one, enabling the finite recursion.
    Defined in (3.2) purely for notational convenience; it is not an independent physical or probabilistic object and has no content beyond the already-assumed Stein identity.

pith-pipeline@v1.1.0-grok45 · 15309 in / 2951 out tokens · 27039 ms · 2026-07-11T22:31:36.969705+00:00 · methodology

0 comments
read the original abstract

In a 1934 Annals of Mathematics paper, Samuel S. Wilks posed the problem of computing joint moments of disjoint principal minors of Wishart random matrices, describing the general case as ``extremely complicated.'' These moments arise in classical multivariate statistics, including multivariate regression, analysis of variance, generalized correlation coefficients, and likelihood ratio testing. We solve Wilks' problem for nonnegative integer exponents and any finite number of disjoint principal minors using Stein's method, specifically the Wishart Stein characterization recently developed by Bailly et al. (2026). The Wishart Stein identity yields a degree-lowering recursion that terminates after finitely many steps and evaluates the desired moment exactly. Numerical experiments validate our recursive formulas against direct Monte Carlo simulations from the Wishart distribution.

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

21 extracted references · 4 canonical work pages

  1. [1]

    Anderson, T. W. , TITLE =. 2003 , PAGES =

    2003

  2. [2]

    and Gaunt, R

    Bailly, G. and Gaunt, R. E. and Ouimet, F. and Richards, D. and von Sachs, R. , TITLE =. ArXiv preprint arXiv:2606.04859 , FJOURNAL =. 2026 , DOI =

    Pith/arXiv arXiv 2026

  3. [3]

    Constantine, A. G. , TITLE =. Ann. Math. Statist. , FJOURNAL =. 1963 , NUMBER =

    1963

  4. [4]

    and Massam, H

    Drton, M. and Massam, H. and Olkin, I. , TITLE =. Ann. Statist. , FJOURNAL =. 2008 , NUMBER =

    2008

  5. [5]

    and Ouimet, F

    Genest, C. and Ouimet, F. and Richards, D. , TITLE =. Stat , FJOURNAL =. 2024 , NUMBER =

    2024

  6. [6]

    and Ouimet, F

    Genest, C. and Ouimet, F. and Richards, D. , TITLE =. Proc. Amer. Math. Soc. , FJOURNAL =. 2025 , NUMBER =

    2025

  7. [7]

    Gupta, A. K. and Nagar, D. K. , TITLE =. 2000 , PAGES =

    2000

  8. [8]

    Herz, C. S. , TITLE =. Ann. of Math. (2) , FJOURNAL =. 1955 , NUMBER =

    1955

  9. [9]

    and Kan, R

    Hillier, G. and Kan, R. M. , TITLE =. J. Quant. Econ. , FJOURNAL =. 2021 , NUMBER =

    2021

  10. [10]

    and Kan, R

    Hillier, G. and Kan, R. M. , TITLE =. Scand. J. Statist. , FJOURNAL =. 2024 , NUMBER =

    2024

  11. [11]

    and Massam, H

    Letac, G. and Massam, H. , TITLE =. Scand. J. Statist. , FJOURNAL =. 2004 , NUMBER =

    2004

  12. [12]

    Lu, I-Li and Richards, D. St. P. , TITLE =. Adv. in Appl. Math. , FJOURNAL =. 2001 , NUMBER =. doi:10.1006/aama.2001.0748 , MRNUMBER =

    work page doi:10.1006/aama.2001.0748 2001

  13. [13]

    Muirhead, R. J. , TITLE =. 1982 , PAGES =

    1982

  14. [14]

    , TITLE =

    Nabeya, S. , TITLE =. Ann. Inst. Statist. Math., Tokyo , FJOURNAL =. 1951 , NUMBER =

    1951

  15. [15]

    , TITLE =

    Nabeya, S. , TITLE =. Ann. Inst. Statist. Math., Tokyo , FJOURNAL =. 1952 , NUMBER =. doi:10.1007/BF02949786 , MRNUMBER =

    work page doi:10.1007/bf02949786 1952

  16. [16]

    , TITLE =

    Ogasawara, H. , TITLE =. J. Multivariate Anal. , FJOURNAL =. 2021 , PAGES =. doi:10.1016/j.jmva.2021.104729 , MRNUMBER =

    work page doi:10.1016/j.jmva.2021.104729 2021

  17. [17]

    , TITLE =

    Ouimet, F. , TITLE =. 2026 , NOTE =

    2026

  18. [18]

    Richards, D. St. P. , TITLE =. 2010 , ISBN =

    2010

  19. [19]

    Wilks, S. S. , TITLE =. Biometrika , FJOURNAL =. 1932 , NUMBER =

    1932

  20. [20]

    Wilks, S. S. , TITLE =. Ann. of Math. (2) , FJOURNAL =. 1934 , NUMBER =. doi:10.2307/1968435 , MRNUMBER =

    work page doi:10.2307/1968435 1934

  21. [21]

    , TITLE =

    Wishart, J. , TITLE =. Biometrika , FJOURNAL =. 1928 , NUMBER =

    1928