Erdős Problem #320
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: solved (not automatically the live result).
Subject areas: number theory, unit fractions.
Read this status correctly
The status above is reproduced from GCL's immutable external-catalogue snapshot. It is not a new claim of current mathematical resolution, a source-semantic review, an independent proof check, or GCL certification. A published proof and a GCL replay are separate events.
The question in plain language
For a positive integer \(N\), consider every subset \(A\) of \(\{1,\ldots,N\}\) and form the reciprocal sum \(\sum_{n\in A}1/n\). Different subsets can produce the same rational number: for example, \(1/2=1/3+1/6\). Let \(S(N)\) be the number of distinct rational sums that actually occur, including the empty sum \(0\). Erdős Problem 320 asks for the growth of \(S(N)\) as \(N\) becomes large.
This is not a question about how many subsets there are: there are exactly \(2^N\) subsets. It asks how much arithmetic collision occurs among those subsets. The natural scale is expressed through iterated logarithms. Write \(log_1 x=log x\) and \(log_{j+1}x=log(log_jx)\) when defined. The relevant stopping index is the last depth at which all iterated logarithms down to that depth remain at least \(1\); the final value lies in the bounded interval \([1,e)\).
The exact catalogue statement is deliberately short: “Let \(S(N)\) count the number of distinct sums of the form \(\sum_{n\in A}1/n\) for \(A\subseteq\{1,\ldots,N\}\). Estimate \(S(N)\).” It does not itself prescribe a particular asymptotic normalization.
A small example
An exact enumeration with rational arithmetic gives the following initial values:
The first five values show no collision, while at \(N=6\) collisions appear. One is
The calculation was checked by constructing all \(2^N\) subsets for \(N\leq6\), summing as exact reduced fractions, and counting the resulting set; the assertions \(S(6)=52\) and the displayed equality both passed. This verifies only a finite toy case. It neither estimates the eventual order of growth nor supports any asymptotic theorem. The value \(S(6)=52\) is consistent with the exact small table discussed by Bettin, Grenié, Molteni, and Sanna, but the finite computation here is independent of a Lean replay.
What is known
The primary paper by M. N. Bleicher and P. Erdős, The number of distinct subsums of \(\sum_{i=1}^N1/i\), Mathematics of Computation 29 (1975), 29–42, proves a lower bound. Its Corollary 3, on printed page 39, gives for \(k>3\) and \(\log_{k+1}N>k+1\) a lower bound of the form
The older Bleicher–Erdős work on Egyptian-fraction denominators supplies the complementary upper-bound framework recorded on the maintained problem page. Bettin, Grenié, Molteni, and Sanna later improved the lower bound and published exact small values; the page cites their 2025 arXiv preprint. These results establish the scale up to the relevant iterated-logarithmic factors, but they should not be silently rewritten as the full asymptotic requested by the catalogue question.
The maintained Erdős Problems entry, accessed during this audit, was last edited 16 July 2026. It labels the problem solved and displays two later proof claims. Claim 55, by Ray Young, Keheng Zhu, and Yanping Luo, is marked accepted by that site and states an order-of-magnitude result. Claim 115, by Scott Duke Kominers and Joachim Neu, states a stronger full asymptotic with a positive continuous nonconstant phase function. These are important dated reports, but a website label or proof-claim marker is not GCL mathematical certification.
GCL's current state
The frozen catalogue packet is teorth/erdosproblems at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, entry number: "320", Git blob 4e60263a80af6207aca494d2e1388e44fa372889. It records informal status solved, formal status unformalized, no statement formalization, and last update 2025-08-31. The frozen packet also records the protected triage class OUTSIDE_STRICT_OPEN_TRIAGE; the catalogue metadata is evidence, not the mathematical authority.
No Problem 320 statement file was present in the issue-named FC-GDM-85F86371 provider snapshot. A check of current google-deepmind/formal-conjectures main at commit 817d8a74e6bcd505ac8407388cc12ac3cfda3d0d likewise found no FormalConjectures/ErdosProblems/320.lean. Thus there is no GCL FormalConjectures theorem, with or without sorry, to cite for this problem.
The Kominers–Neu manuscript, The Asymptotic Number of Distinct Reciprocal Subset Sums, identifies repository commit e51bf855d37c97cc66126b98144804c950f86872. Its Erdos320/Main.lean states the full theorem, but its own Assumptions.lean isolates four project-specific assumptions: a large finite certificate, two explicit prime-counting estimates, and a published finite table. The repository’s trust boundary is therefore materially different from an unconditional GCL-certified theorem.
What remains
The most useful bounded continuation is source-led: compare the exact catalogue statement, the 1975 and 1976 primary bounds, the BGMS lower bound, and the two 2026 claims line by line. First normalize the stopping convention and every quantifier on \(N\), \(k\), and \(r\). Then extract the precise theorem statement and dependency boundary from the Kominers–Neu manuscript and its pinned Lean commit, without attempting an exhaustive build. Separately record what Claim 55 proves, what its linked Lean repository covers, and which analytic inputs remain external.
Acceptance should require a reproducible statement comparison, exact source locations, and a finite check such as the \(N\leq6\) enumeration above. Falsification would be a quantifier mismatch, an unavailable cited source, or a claimed theorem whose formal statement depends on an unrecorded assumption. Stop once those comparisons are complete or once a primary source cannot be read; do not infer a proof from a catalogue status.
Sources and claim boundaries
The original work packet is MATH-PROGRAMME issue #1590, which has no prior comments at audit time. The primary 1975 paper is available as the authors’ archival PDF and as DOI 10.1090/S0025-5718-1975-0366795-4. The related 1976 paper is Denominators of Egyptian fractions II, and the later lower-bound source is arXiv:2509.10030.
The 2026 manuscript and its formal source are author-provided evidence, not a GCL adjudication. The V2 ledger alone decides accepted coverage. This article makes no mathematical certification, publication decision, priority claim, or award claim. Original SOURCE RESULT/1 citation: authenticated original RESULT/1.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1590 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
- Formal-statement coverage: No statement was confirmed in GCL's pinned formal-provider snapshot. Presence does not mean the statement or proof has been checked.
- External mathematical status: consult current problem page and its dated bibliography.