Pith. sign in

REVIEW 4 minor 2 cited by

A simple random-weight average of nonnegative data is a valid finite-sample p-value for the claim that every mean is at most one.

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

2026-07-10 07:48 UTC pith:ZUO2F4GV

load-bearing objection Clean full proof of Gaffke’s 2005 conjecture: K is a valid finite-sample p-value for simultaneous mean bounds on independent nonnegative r.v.s.

arxiv 2607.08415 v1 pith:ZUO2F4GV submitted 2026-07-09 math.ST math.PRstat.TH

An Exact Distribution-Free Test for Means of Nonnegative Random Variables

Nikos Vlassis , Philip S. Thomas This is my paper
classification math.ST math.PRstat.TH MSC 62G1062G15
keywords distribution-free p-valuenonnegative random variablesDirichlet weightsfinite-sample validitytwo-point systemschain measuresexponential transfer
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.

The paper settles a 2005 conjecture by proving that a particular statistic is a valid one-sided p-value for the hypothesis that every mean of a collection of independent nonnegative random variables is at most one. The statistic is formed by adjoining a zero, drawing uniform Dirichlet weights, and computing the probability that the weighted average stays at most one. Because the variables need not be identically distributed, continuous, or otherwise restricted, the result supplies a distribution-free exact test that works for every sample size. A sympathetic reader cares because many practical problems (risk, reliability, bounded observations) reduce to one-sided mean statements under nonnegativity, yet classical asymptotic or bootstrap methods lose finite-sample control once the distributions become heterogeneous. The proof first reduces the problem to two-point mean-one laws, then builds a dominating chain measure by successive insertion of each variable, using an exponential-transfer identity and a log-concavity argument to guarantee that every mass transport is upward.

Core claim

Whenever independent nonnegative random variables satisfy EXi ≤ 1 for every i, the random variable K(X) = P{∑ xi Di ≤ 1} (D ~ Dir(1,…,1) independent of X) obeys P{K(X) ≤ α} ≤ α for every α ∈ [0,1]. Consequently K(X) is a finite-sample, distribution-free p-value for the simultaneous null that all means are at most one.

What carries the argument

The local insertion lemma (Lemma 6) together with the exponential-transfer identity (Lemma 5): each new two-point variable is inserted into a maximal chain so that the hybrid measure never decreases the expectation of any increasing payoff, with the required nonnegativity of transfer coefficients supplied by a Stein identity for exponential shifts and a likelihood-ratio inequality that follows from log-concavity of the partial sums.

Load-bearing premise

The densities of the successive partial sums that appear along the chain must remain log-concave so that every mass-transport coefficient stays nonnegative after the variables have been sorted by their low values.

What would settle it

Exhibit any finite collection of independent nonnegative random variables with all means ≤ 1 for which the empirical frequency of K(X) ≤ α exceeds α by a statistically clear margin, for some fixed α (for example by exhaustive enumeration of a small two-point system whose parameters violate the claimed ordering of heta j).

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

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 proves Gaffke's 2005 conjecture: for independent nonnegative random variables X1,...,Xn (not necessarily identically distributed) with EXi≤1, the statistic K(X)=P{∑ xi Di≤1} with D~Dir(1,...,1) independent of X satisfies P{K(X)≤α}≤α for every α∈[0,1]. Thus K(X) is a finite-sample, distribution-free p-value for the simultaneous one-sided mean null. The argument proceeds by reduction: first to mean-one two-point marginals (via a mixture representation, Lemma 7), then by induction on a hybrid chain measure that dominates the product law on increasing payoffs (Proposition 3), with the local insertion step controlled by an exact Stein-type identity and a total-positivity comparison for exponential shifts (Lemmas 5–6). The general case follows by rescaling.

Significance. The result settles a long-standing conjecture and supplies a genuinely distribution-free, finite-sample p-value under minimal assumptions (independence and nonnegativity only). The construction is explicit, the reduction to two-point systems is clean, and the analytic core (exponential-transfer identity plus log-concavity of the partial sums Gj) is standard once the sorting of the low values is imposed. The paper therefore adds a usable exact test to the nonparametric toolkit for nonnegative means, with clear connections to the earlier confidence-bound work of Learned-Miller and Thomas. Strengths include the fully self-contained derivation, the absence of free parameters, and the transparent inductive structure that turns a local mass-transport comparison into global domination.

minor comments (4)
  1. The footnote on AI assistance is unusual for a pure-mathematics paper; if the journal style requires disclosure it should be moved to an acknowledgments paragraph rather than left as a numbered footnote on the first page.
  2. In the definition of the hybrid measure μ k (display (7)), a short parenthetical reminder that the chain measure u Ck lives on the power set of [k] while π>k lives on the power set of {k+1,...,n} would make the product construction immediately transparent to a reader who has not yet internalized the notation.
  3. Lemma 5 invokes the preservation of total positivity under the exponential translation kernel and cites Karlin (1968). A one-sentence pointer to the precise statement (e.g., the TP2 property of the kernel) would help readers who are not specialists in total positivity.
  4. The sentinel construction Gk (display (23)) is elegant but appears abruptly; a brief sentence explaining why the terminal edge must flip the E0 term would improve readability of the induction.

Circularity Check

0 steps flagged

No circularity: the validity of K is derived from first principles via chain domination and an exponential-transfer identity, not assumed or fitted.

full rationale

The paper proves Gaffke's 2005 conjecture that K(X) is a valid finite-sample p-value under independent nonnegative variables with means ≤1. The derivation is self-contained: (i) reduce to mean-one two-point systems by a mixture representation (Lemma 7) and rescaling; (ii) encode outcomes by high sets and dominate the product measure π by a chain measure ν_C via inductive insertion (Proposition 3); (iii) choose each insertion position by a local lemma (Lemma 6) whose nonnegativity of transfer coefficients η_j rests on a Stein-type identity and a likelihood-ratio inequality for log-concave densities of partial exponential sums (Lemma 5). Log-concavity is elementary (independent scaled exponentials, closed under convolution) and is proved inside the paper; the total-positivity citation (Karlin 1968) is classical external mathematics, not a self-citation. Gaffke (2005) and Learned-Miller–Thomas (2020) are cited only for historical context and a related confidence-bound result; neither supplies an unproved black-box step inside the induction. No quantity is fitted to data and then re-presented as a prediction, no uniqueness theorem is imported from the authors, and the target inequality P{K≤α}≤α is never assumed. Consequently the circularity score is zero.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 1 invented entities

The paper is a pure existence/validity proof. It imports only standard probabilistic facts (Dirichlet representation via exponentials, preservation of log-concavity under convolution, total positivity of the exponential kernel) and the modeling assumptions of independence and nonnegativity. No numerical parameters are fitted; the sole 'invented' objects are the auxiliary chain measures used as proof devices.

axioms (4)
  • standard math Independent unit exponential random variables admit the Dirichlet representation Di=Ei/∑ Er used to define K.
    Classical fact invoked in equation (1) and throughout Sections 2–3.
  • standard math Convolution of log-concave densities remains log-concave; the exponential translation kernel preserves total positivity (Karlin 1968).
    Used to obtain the likelihood-ratio inequality (22) that yields heta+≥ heta- in Lemma 5.
  • domain assumption The Xi are mutually independent and almost surely nonnegative.
    Stated in the theorem hypothesis and essential for the product structure of π and for the mixture representation in Lemma 7.
  • standard math Every mean-one law on [0,∞) is a mixture of mean-one two-point (or degenerate) laws (Lemma 7).
    Proved in the paper by an explicit integral construction; once established it reduces the general case to the two-point case.
invented entities (1)
  • Chain measure u C and hybrid measure μ k no independent evidence
    purpose: Dominating measures that turn the product law π into a totally ordered chain along which the rejection set is a terminal segment whose mass is controlled by K itself.
    Purely auxiliary constructions introduced in Section 2; they have no independent empirical content outside the proof.

pith-pipeline@v1.1.0-grok45 · 14397 in / 2667 out tokens · 31542 ms · 2026-07-10T07:48:01.884199+00:00 · methodology

0 comments
read the original abstract

Let $X=(X_1,\ldots,X_n)$ be independent nonnegative random variables, not necessarily identically distributed. Let $D=(D_0,D_1,\ldots,D_n)\sim\operatorname{Dir}(1,\ldots,1)$ be independent of $X$, and define $K(x)=\mathbb{P}\{\sum_{i=1}^n x_iD_i\le1\}$. We prove that, for every $n\ge1$, whenever $\mathbb{E} X_i\le1$ for every $i$, $\mathbb{P}\{K(X)\le\alpha\}\le\alpha$ for all $0\le\alpha\le1$. Thus $K(X)$ is a finite-sample, distribution-free $p$-value for testing the null hypothesis $\mathbb{E}X_i \le 1$ for all $i$. This proves a conjecture of Gaffke (2005).

discussion (0)

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

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. On Feige's conjecture

    math.PR 2026-07 accept novelty 7.0

    For independent nonnegative mean-one random variables, P(sum < n+1) is at least (n/(n+1))^n ≥ 1/e, proving Feige's conjecture with a matching extremal example.

  2. On the Order-Conditional Optimality of Gaffke's Bound

    math.ST 2026-07 conditional novelty 6.0

    Gaffke's bound is Buehler-optimal within the class of lower confidence bounds that induce its own sample ordering, for the maximum marginal mean of independent nonnegative variables.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages · cited by 2 Pith papers · 1 internal anchor

  1. [1]

    N. Gaffke. Three test statistics for a nonparametric one-sided hypothesis on the mean of a nonnegative variable. Mathematical Methods of Statistics, 14(4):451--467, 2005

    work page 2005

  2. [2]

    S. Karlin. Total Positivity, Volume I. Stanford University Press, 1968

    work page 1968

  3. [3]

    A New Confidence Interval for the Mean of a Bounded Random Variable

    E. Learned-Miller and P. S. Thomas. A new confidence interval for the mean of a bounded random variable. arXiv:1905.06208v2, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 1905