Linear-time rows for quadratic e-value discovery matrices
Year: 2026
Type: preprint
Abstract: Unrefereed research candidate. In one sentence: this paper answers, for the pairwise-product merger U2, a question of Vovk and Wang by computing each row of an e-value discovery matrix in time linear in the number of hypotheses, so that the whole matrix takes time proportional to its size. Context. An e-value is a nonnegative statistic whose expectation is at most one when a null hypothesis is true, so a large value is evidence against that hypothesis. When many hypotheses are tested at once, Vovk and Wang summarise the evidence in a discovery matrix: the entry in row r and column j says how strongly the data support “more than j of the r hypotheses with the largest e-values are false”, simultaneously for every r and j. With independent or sequential e-values, merging by the mean of pairwise products (U2) gives much tighter matrices than the arithmetic mean, but each entry is a minimum over exponentially many subsets and the best published algorithm needed O(K2) operations per row for K hypotheses. Vovk and Wang (Canadian Journal of Statistics, 2024, Sections 7 and 9) asked whether O(K) is possible. Results. (1) For every fixed nonnegative combination F = c + aU1 + bU2 of a constant, the arithmetic mean and the pairwise-product mean, one discovery row, with a witness subset for each entry, takes O(K) arithmetic operations and the complete matrix O(K2). For pure U2, and for mixtures evaluated with the source's singleton convention, the values are exactly the source's discovery matrix. (2) The key step is a central-insertion inequality: if inserting observations into a gap of an ordered sample does not increase the pairwise-product mean, it does not increase the arithmetic mean either. With a second lemma on adding a common maximum, it makes an implicit array of candidate subsets totally monotone, so the SMAWK matrix-searching algorithm of Aggarwal, Klawe, Moran, Shor and Wilber finds all its row minima in linear time. The array is neither Monge nor row-unimodal. (3) An exact rational implementation agrees with complete subset enumeration; a complete 10,000 × 10,000 U2 discovery matrix takes about two minutes in plain Python, with 3.77 × 108 entry evaluations against 1.67 × 1011 for the previous algorithm. (4) For higher-degree means Ud, d ≥ 3, a single chord inequality would settle every fixed degree for positive e-values; the paper reduces it to insertions of copies of the two values next to the gap, proves the single-insertion case and gives exact computational evidence, but no proof. Limits. Exact in rational arithmetic; the bounds count arithmetic operations, not bit operations. Floating-point results are approximations, not certified. The statistical reading needs the source's assumptions: independent or sequential e-values and mixture weights fixed in advance with a + b + c ≤ 1. Degrees d ≥ 3 remain open. Contents. The paper (LaTeX source and PDF); a standard-library Python implementation with exact and floating-point modes; scripts that check it against complete subset enumeration (with a negative control), check every worked example, the higher-degree identities and evidence, the operation counts and the full 10,000 × 10,000 example, with their recorded outputs; a dated prior-art record; the bibliography from registry records; a machine-readable index; a research-gate record; a replay receipt and SHA-256 sums. Assurance. Written proofs with exact computational checks; the checks test the code and its conventions and do not replace the proofs. Internal AI-assisted review only (five reviewer roles, an adversarial review and an audit of new proofs by a second AI model, clean replays, one revision round, a re-review and a proof audit): not peer review, independent reproduction or formal verification. The question and the discovery-matrix framework are due to Vovk and Wang; SMAWK is due to Aggarwal, Klawe, Moran, Shor and Wilber. Creator: Anonymous. Text and data CC0-1.0; code MIT.
Author: Anonymous
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7233600950
