Work

Fixed-size subset sums: sharp support, Rényi entropy and tail bounds

Year: 2026
Type: preprint
Abstract: Unrefereed research candidate. Proof of the Mogilnykh–Vorob’ev–Valyuzhenich conjecture, the exact Rényi minimum for sampling without replacement, and the hypergeometric tail problem from n = 4k. The question. Pick k of n given real numbers at random, without replacement, and add them. How concentrated can that random sum be? The paper answers three classical forms of this question exactly: how often the sum can be exactly zero, how small its entropy can be, and how rarely it can be nonnegative. Results. (1) Zero sums. If the numbers add up to zero and are not all zero, at least 2·C(n−2, k−1) of the k-subsets have a nonzero sum whenever k ≥ 5 and n ≥ 2k+1, and only the pattern (1, −1, 0, …, 0), up to scaling and reordering, attains this. This is the smallest support of an eigenfunction in the first nontrivial eigenspace of the Johnson graph, and proves Conjecture 1 of Mogilnykh, Vorob’ev and Valyuzhenich (2018), who had checked it by computer for n ≤ 600. A formula with all equality cases is given for every n and k, including the case (1, 1, −1, −1, 0) at n = 5, k = 2 that earlier classifications missed (Theorem 1.1). (2) Entropy. The minimum Rényi entropy of the sum is found for every finite order, answering a question of Ai, Kuželka and Pelekis (2026): one outlier is optimal except at half-sampling (n = 2k), where two outliers win above an explicit order αk (α2 = 2, about 1.386k for large k). So the universal Bernoulli lower bound holds exactly for orders up to 2 (Theorems 1.2 and 1.3, Corollary 1.4). (3) Tails. For two-valued populations, the probability that the sum is at least its mean is at least k/n whenever n ≥ 4k, with a single outlier the unique minimiser for n > 4k (Theorem 1.5); Ai and Pelekis (2026) had n ≥ 8k. The proof uses a 2026 lemma of Ai, Karagiannis, Kuželka and Pelekis. Consequently the Aydinian–Blinovsky conjecture (that two-valued populations are the worst case) would imply the Manickam–Miklós–Singhi conjecture for all n ≥ 4k. This does not prove the Manickam–Miklós–Singhi conjecture, and below 4k a single outlier is not always the worst case. (4) Exact analogues over Abelian groups, where elements of order two change the answer at half-sampling (Theorem 1.6), and a small-ball inequality for sampling without replacement; a known incidence-rank result is re-derived. Limits. Real numbers only for the zero-sum theorem (it fails over Z2, Z3 and Z5); first Johnson eigenspace only; the tail problem below n = 4k is open. Finite computer checks corroborate but do not replace the written proofs. Contents. The paper (LaTeX source and PDF); five standard-library Python scripts of exact finite checks (each runs in seconds and exits nonzero on any failure, also under python -O) with their recorded outputs; a claim ledger and prior-art note; the bibliography from registry records; a machine-readable index; a research-gate record; a rating-free review summary; a replay receipt and SHA-256 sums. Assurance. Complete written proofs for every result. Internal AI-assisted review only (five reviewer roles, an adversarial review and an audit by a second AI model, clean replays, a re-review and a proof audit): not peer review, independent reproduction or formal verification. Creator: Anonymous. Text and data CC0-1.0; code MIT.
Author: Anonymous
Language: English
Open Access status: green
OpenAlex ID: W7233565533
Cites: 17
Cited by:
Related to: